]>
Commit | Line | Data |
---|---|---|
064af421 | 1 | /* |
78c8df12 | 2 | * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc. |
064af421 | 3 | * |
a14bc59f BP |
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: | |
064af421 | 7 | * |
a14bc59f BP |
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. | |
064af421 BP |
15 | */ |
16 | ||
17 | #include <config.h> | |
18 | #include "classifier.h" | |
38c449e0 | 19 | #include "classifier-private.h" |
064af421 BP |
20 | #include <errno.h> |
21 | #include <netinet/in.h> | |
844dff32 | 22 | #include "byte-order.h" |
68d1c8c3 | 23 | #include "dynamic-string.h" |
07b37e8f | 24 | #include "odp-util.h" |
d8ae4d67 | 25 | #include "ofp-util.h" |
13751fd8 | 26 | #include "packets.h" |
52054c15 | 27 | #include "util.h" |
13751fd8 JR |
28 | #include "vlog.h" |
29 | ||
30 | VLOG_DEFINE_THIS_MODULE(classifier); | |
064af421 | 31 | |
69d6040e JR |
32 | struct trie_ctx; |
33 | ||
34 | /* Ports trie depends on both ports sharing the same ovs_be32. */ | |
35 | #define TP_PORTS_OFS32 (offsetof(struct flow, tp_src) / 4) | |
36 | BUILD_ASSERT_DECL(TP_PORTS_OFS32 == offsetof(struct flow, tp_dst) / 4); | |
cabd4c43 | 37 | |
627fb667 JR |
38 | static struct cls_match * |
39 | cls_match_alloc(struct cls_rule *rule) | |
40 | { | |
3016f3e4 JR |
41 | int count = count_1bits(rule->match.flow.map); |
42 | ||
43 | struct cls_match *cls_match | |
44 | = xmalloc(sizeof *cls_match - sizeof cls_match->flow.inline_values | |
45 | + MINIFLOW_VALUES_SIZE(count)); | |
627fb667 | 46 | |
f80028fe JR |
47 | *CONST_CAST(const struct cls_rule **, &cls_match->cls_rule) = rule; |
48 | *CONST_CAST(int *, &cls_match->priority) = rule->priority; | |
49 | miniflow_clone_inline(CONST_CAST(struct miniflow *, &cls_match->flow), | |
50 | &rule->match.flow, count); | |
627fb667 JR |
51 | rule->cls_match = cls_match; |
52 | ||
53 | return cls_match; | |
54 | } | |
cabd4c43 | 55 | |
e48eccd1 | 56 | static struct cls_subtable *find_subtable(const struct classifier *cls, |
dfea28b3 | 57 | const struct minimask *); |
e48eccd1 | 58 | static struct cls_subtable *insert_subtable(struct classifier *cls, |
e65413ab JR |
59 | const struct minimask *) |
60 | OVS_REQUIRES(cls->mutex); | |
e48eccd1 | 61 | static void destroy_subtable(struct classifier *cls, struct cls_subtable *) |
e65413ab | 62 | OVS_REQUIRES(cls->mutex); |
b5d97350 | 63 | |
dfea28b3 JR |
64 | static const struct cls_match *find_match_wc(const struct cls_subtable *, |
65 | const struct flow *, | |
66 | struct trie_ctx *, | |
67 | unsigned int n_tries, | |
68 | struct flow_wildcards *); | |
69 | static struct cls_match *find_equal(const struct cls_subtable *, | |
627fb667 | 70 | const struct miniflow *, uint32_t hash); |
b5d97350 | 71 | |
dfea28b3 JR |
72 | static inline const struct cls_match * |
73 | next_rule_in_list__(const struct cls_match *rule) | |
74 | { | |
75 | const struct cls_match *next = NULL; | |
76 | next = OBJECT_CONTAINING(rculist_next(&rule->list), next, list); | |
77 | return next; | |
78 | } | |
79 | ||
80 | static inline const struct cls_match * | |
81 | next_rule_in_list(const struct cls_match *rule) | |
82 | { | |
83 | const struct cls_match *next = next_rule_in_list__(rule); | |
84 | return next->priority < rule->priority ? next : NULL; | |
85 | } | |
86 | ||
c501b427 | 87 | static inline struct cls_match * |
dfea28b3 | 88 | next_rule_in_list_protected__(struct cls_match *rule) |
c501b427 JR |
89 | { |
90 | struct cls_match *next = NULL; | |
dfea28b3 | 91 | next = OBJECT_CONTAINING(rculist_next_protected(&rule->list), next, list); |
c501b427 JR |
92 | return next; |
93 | } | |
94 | ||
95 | static inline struct cls_match * | |
dfea28b3 | 96 | next_rule_in_list_protected(struct cls_match *rule) |
c501b427 | 97 | { |
dfea28b3 | 98 | struct cls_match *next = next_rule_in_list_protected__(rule); |
c501b427 JR |
99 | return next->priority < rule->priority ? next : NULL; |
100 | } | |
101 | ||
e65413ab JR |
102 | /* Iterates RULE over HEAD and all of the cls_rules on HEAD->list. |
103 | * Classifier's mutex must be held while iterating, as the list is | |
104 | * protoceted by it. */ | |
b5d97350 BP |
105 | #define FOR_EACH_RULE_IN_LIST(RULE, HEAD) \ |
106 | for ((RULE) = (HEAD); (RULE) != NULL; (RULE) = next_rule_in_list(RULE)) | |
dfea28b3 JR |
107 | #define FOR_EACH_RULE_IN_LIST_PROTECTED(RULE, HEAD) \ |
108 | for ((RULE) = (HEAD); (RULE) != NULL; \ | |
109 | (RULE) = next_rule_in_list_protected(RULE)) | |
13751fd8 JR |
110 | |
111 | static unsigned int minimask_get_prefix_len(const struct minimask *, | |
112 | const struct mf_field *); | |
e48eccd1 | 113 | static void trie_init(struct classifier *cls, int trie_idx, |
e65413ab JR |
114 | const struct mf_field *) |
115 | OVS_REQUIRES(cls->mutex); | |
13751fd8 | 116 | static unsigned int trie_lookup(const struct cls_trie *, const struct flow *, |
c0bfb650 | 117 | union mf_value *plens); |
f358a2cb | 118 | static unsigned int trie_lookup_value(const rcu_trie_ptr *, |
c0bfb650 JR |
119 | const ovs_be32 value[], ovs_be32 plens[], |
120 | unsigned int value_bits); | |
f358a2cb | 121 | static void trie_destroy(rcu_trie_ptr *); |
13751fd8 | 122 | static void trie_insert(struct cls_trie *, const struct cls_rule *, int mlen); |
f358a2cb | 123 | static void trie_insert_prefix(rcu_trie_ptr *, const ovs_be32 *prefix, |
69d6040e | 124 | int mlen); |
13751fd8 | 125 | static void trie_remove(struct cls_trie *, const struct cls_rule *, int mlen); |
f358a2cb | 126 | static void trie_remove_prefix(rcu_trie_ptr *, const ovs_be32 *prefix, |
69d6040e | 127 | int mlen); |
13751fd8 | 128 | static void mask_set_prefix_bits(struct flow_wildcards *, uint8_t be32ofs, |
c30cfa6b | 129 | unsigned int n_bits); |
13751fd8 | 130 | static bool mask_prefix_bits_set(const struct flow_wildcards *, |
c30cfa6b | 131 | uint8_t be32ofs, unsigned int n_bits); |
81a76618 BP |
132 | \f |
133 | /* cls_rule. */ | |
b5d97350 | 134 | |
81a76618 | 135 | /* Initializes 'rule' to match packets specified by 'match' at the given |
5cb7a798 BP |
136 | * 'priority'. 'match' must satisfy the invariant described in the comment at |
137 | * the definition of struct match. | |
66642cb4 | 138 | * |
48d28ac1 BP |
139 | * The caller must eventually destroy 'rule' with cls_rule_destroy(). |
140 | * | |
eb391b76 BP |
141 | * Clients should not use priority INT_MIN. (OpenFlow uses priorities between |
142 | * 0 and UINT16_MAX, inclusive.) */ | |
47284b1f | 143 | void |
eb391b76 | 144 | cls_rule_init(struct cls_rule *rule, const struct match *match, int priority) |
47284b1f | 145 | { |
5cb7a798 BP |
146 | minimatch_init(&rule->match, match); |
147 | rule->priority = priority; | |
627fb667 | 148 | rule->cls_match = NULL; |
5cb7a798 BP |
149 | } |
150 | ||
151 | /* Same as cls_rule_init() for initialization from a "struct minimatch". */ | |
152 | void | |
153 | cls_rule_init_from_minimatch(struct cls_rule *rule, | |
eb391b76 | 154 | const struct minimatch *match, int priority) |
5cb7a798 BP |
155 | { |
156 | minimatch_clone(&rule->match, match); | |
81a76618 | 157 | rule->priority = priority; |
627fb667 | 158 | rule->cls_match = NULL; |
685a51a5 JP |
159 | } |
160 | ||
48d28ac1 BP |
161 | /* Initializes 'dst' as a copy of 'src'. |
162 | * | |
b2c1f00b | 163 | * The caller must eventually destroy 'dst' with cls_rule_destroy(). */ |
48d28ac1 BP |
164 | void |
165 | cls_rule_clone(struct cls_rule *dst, const struct cls_rule *src) | |
166 | { | |
5cb7a798 BP |
167 | minimatch_clone(&dst->match, &src->match); |
168 | dst->priority = src->priority; | |
627fb667 | 169 | dst->cls_match = NULL; |
48d28ac1 BP |
170 | } |
171 | ||
b2c1f00b BP |
172 | /* Initializes 'dst' with the data in 'src', destroying 'src'. |
173 | * | |
174 | * The caller must eventually destroy 'dst' with cls_rule_destroy(). */ | |
175 | void | |
176 | cls_rule_move(struct cls_rule *dst, struct cls_rule *src) | |
177 | { | |
178 | minimatch_move(&dst->match, &src->match); | |
179 | dst->priority = src->priority; | |
627fb667 | 180 | dst->cls_match = NULL; |
b2c1f00b BP |
181 | } |
182 | ||
48d28ac1 BP |
183 | /* Frees memory referenced by 'rule'. Doesn't free 'rule' itself (it's |
184 | * normally embedded into a larger structure). | |
185 | * | |
186 | * ('rule' must not currently be in a classifier.) */ | |
187 | void | |
5cb7a798 | 188 | cls_rule_destroy(struct cls_rule *rule) |
48d28ac1 | 189 | { |
627fb667 | 190 | ovs_assert(!rule->cls_match); |
5cb7a798 | 191 | minimatch_destroy(&rule->match); |
48d28ac1 BP |
192 | } |
193 | ||
81a76618 BP |
194 | /* Returns true if 'a' and 'b' match the same packets at the same priority, |
195 | * false if they differ in some way. */ | |
193eb874 BP |
196 | bool |
197 | cls_rule_equal(const struct cls_rule *a, const struct cls_rule *b) | |
198 | { | |
5cb7a798 | 199 | return a->priority == b->priority && minimatch_equal(&a->match, &b->match); |
193eb874 BP |
200 | } |
201 | ||
81a76618 | 202 | /* Returns a hash value for 'rule', folding in 'basis'. */ |
57452fdc BP |
203 | uint32_t |
204 | cls_rule_hash(const struct cls_rule *rule, uint32_t basis) | |
205 | { | |
5cb7a798 | 206 | return minimatch_hash(&rule->match, hash_int(rule->priority, basis)); |
73f33563 BP |
207 | } |
208 | ||
81a76618 | 209 | /* Appends a string describing 'rule' to 's'. */ |
07b37e8f BP |
210 | void |
211 | cls_rule_format(const struct cls_rule *rule, struct ds *s) | |
212 | { | |
5cb7a798 | 213 | minimatch_format(&rule->match, s, rule->priority); |
064af421 | 214 | } |
3ca1de08 BP |
215 | |
216 | /* Returns true if 'rule' matches every packet, false otherwise. */ | |
217 | bool | |
218 | cls_rule_is_catchall(const struct cls_rule *rule) | |
219 | { | |
5cb7a798 | 220 | return minimask_is_catchall(&rule->match.mask); |
3ca1de08 | 221 | } |
064af421 BP |
222 | \f |
223 | /* Initializes 'cls' as a classifier that initially contains no classification | |
224 | * rules. */ | |
225 | void | |
e48eccd1 JR |
226 | classifier_init(struct classifier *cls, const uint8_t *flow_segments) |
227 | OVS_EXCLUDED(cls->mutex) | |
064af421 | 228 | { |
e65413ab | 229 | ovs_mutex_init(&cls->mutex); |
e65413ab | 230 | ovs_mutex_lock(&cls->mutex); |
064af421 | 231 | cls->n_rules = 0; |
f2c21402 | 232 | cmap_init(&cls->subtables_map); |
fe7cfa5c | 233 | pvector_init(&cls->subtables); |
f2c21402 | 234 | cmap_init(&cls->partitions); |
476f36e8 JR |
235 | cls->n_flow_segments = 0; |
236 | if (flow_segments) { | |
237 | while (cls->n_flow_segments < CLS_MAX_INDICES | |
238 | && *flow_segments < FLOW_U32S) { | |
239 | cls->flow_segments[cls->n_flow_segments++] = *flow_segments++; | |
240 | } | |
241 | } | |
13751fd8 | 242 | cls->n_tries = 0; |
e65413ab JR |
243 | for (int i = 0; i < CLS_MAX_TRIES; i++) { |
244 | trie_init(cls, i, NULL); | |
245 | } | |
246 | ovs_mutex_unlock(&cls->mutex); | |
064af421 BP |
247 | } |
248 | ||
249 | /* Destroys 'cls'. Rules within 'cls', if any, are not freed; this is the | |
afae68b1 JR |
250 | * caller's responsibility. |
251 | * May only be called after all the readers have been terminated. */ | |
064af421 | 252 | void |
e48eccd1 JR |
253 | classifier_destroy(struct classifier *cls) |
254 | OVS_EXCLUDED(cls->mutex) | |
064af421 | 255 | { |
e48eccd1 | 256 | if (cls) { |
78c8df12 BP |
257 | struct cls_partition *partition; |
258 | struct cls_subtable *subtable; | |
13751fd8 JR |
259 | int i; |
260 | ||
e65413ab | 261 | ovs_mutex_lock(&cls->mutex); |
13751fd8 | 262 | for (i = 0; i < cls->n_tries; i++) { |
f358a2cb | 263 | trie_destroy(&cls->tries[i].root); |
13751fd8 | 264 | } |
064af421 | 265 | |
6bc3bb82 | 266 | CMAP_FOR_EACH (subtable, cmap_node, &cls->subtables_map) { |
03868246 | 267 | destroy_subtable(cls, subtable); |
064af421 | 268 | } |
f2c21402 | 269 | cmap_destroy(&cls->subtables_map); |
c906cedf | 270 | |
6bc3bb82 | 271 | CMAP_FOR_EACH (partition, cmap_node, &cls->partitions) { |
f2c21402 | 272 | ovsrcu_postpone(free, partition); |
c906cedf | 273 | } |
f2c21402 | 274 | cmap_destroy(&cls->partitions); |
cabd4c43 | 275 | |
fe7cfa5c | 276 | pvector_destroy(&cls->subtables); |
e65413ab JR |
277 | ovs_mutex_unlock(&cls->mutex); |
278 | ovs_mutex_destroy(&cls->mutex); | |
064af421 BP |
279 | } |
280 | } | |
281 | ||
13751fd8 | 282 | /* Set the fields for which prefix lookup should be performed. */ |
f358a2cb | 283 | bool |
e48eccd1 | 284 | classifier_set_prefix_fields(struct classifier *cls, |
13751fd8 JR |
285 | const enum mf_field_id *trie_fields, |
286 | unsigned int n_fields) | |
e48eccd1 | 287 | OVS_EXCLUDED(cls->mutex) |
13751fd8 | 288 | { |
f358a2cb | 289 | const struct mf_field * new_fields[CLS_MAX_TRIES]; |
abadfcb0 | 290 | struct mf_bitmap fields = MF_BITMAP_INITIALIZER; |
f358a2cb JR |
291 | int i, n_tries = 0; |
292 | bool changed = false; | |
13751fd8 | 293 | |
e65413ab | 294 | ovs_mutex_lock(&cls->mutex); |
f358a2cb | 295 | for (i = 0; i < n_fields && n_tries < CLS_MAX_TRIES; i++) { |
13751fd8 JR |
296 | const struct mf_field *field = mf_from_id(trie_fields[i]); |
297 | if (field->flow_be32ofs < 0 || field->n_bits % 32) { | |
298 | /* Incompatible field. This is the only place where we | |
299 | * enforce these requirements, but the rest of the trie code | |
300 | * depends on the flow_be32ofs to be non-negative and the | |
301 | * field length to be a multiple of 32 bits. */ | |
302 | continue; | |
303 | } | |
304 | ||
abadfcb0 | 305 | if (bitmap_is_set(fields.bm, trie_fields[i])) { |
13751fd8 JR |
306 | /* Duplicate field, there is no need to build more than |
307 | * one index for any one field. */ | |
308 | continue; | |
309 | } | |
abadfcb0 | 310 | bitmap_set1(fields.bm, trie_fields[i]); |
13751fd8 | 311 | |
f358a2cb JR |
312 | new_fields[n_tries] = NULL; |
313 | if (n_tries >= cls->n_tries || field != cls->tries[n_tries].field) { | |
314 | new_fields[n_tries] = field; | |
315 | changed = true; | |
316 | } | |
317 | n_tries++; | |
318 | } | |
319 | ||
320 | if (changed || n_tries < cls->n_tries) { | |
321 | struct cls_subtable *subtable; | |
322 | ||
323 | /* Trie configuration needs to change. Disable trie lookups | |
324 | * for the tries that are changing and wait all the current readers | |
325 | * with the old configuration to be done. */ | |
326 | changed = false; | |
327 | CMAP_FOR_EACH (subtable, cmap_node, &cls->subtables_map) { | |
328 | for (i = 0; i < cls->n_tries; i++) { | |
329 | if ((i < n_tries && new_fields[i]) || i >= n_tries) { | |
330 | if (subtable->trie_plen[i]) { | |
331 | subtable->trie_plen[i] = 0; | |
332 | changed = true; | |
333 | } | |
334 | } | |
335 | } | |
336 | } | |
337 | /* Synchronize if any readers were using tries. The readers may | |
338 | * temporarily function without the trie lookup based optimizations. */ | |
339 | if (changed) { | |
340 | /* ovsrcu_synchronize() functions as a memory barrier, so it does | |
341 | * not matter that subtable->trie_plen is not atomic. */ | |
342 | ovsrcu_synchronize(); | |
13751fd8 | 343 | } |
13751fd8 | 344 | |
f358a2cb JR |
345 | /* Now set up the tries. */ |
346 | for (i = 0; i < n_tries; i++) { | |
347 | if (new_fields[i]) { | |
348 | trie_init(cls, i, new_fields[i]); | |
349 | } | |
350 | } | |
351 | /* Destroy the rest, if any. */ | |
352 | for (; i < cls->n_tries; i++) { | |
353 | trie_init(cls, i, NULL); | |
354 | } | |
355 | ||
356 | cls->n_tries = n_tries; | |
357 | ovs_mutex_unlock(&cls->mutex); | |
358 | return true; | |
13751fd8 | 359 | } |
f358a2cb | 360 | |
e65413ab | 361 | ovs_mutex_unlock(&cls->mutex); |
f358a2cb | 362 | return false; /* No change. */ |
13751fd8 JR |
363 | } |
364 | ||
365 | static void | |
e48eccd1 | 366 | trie_init(struct classifier *cls, int trie_idx, const struct mf_field *field) |
e65413ab | 367 | OVS_REQUIRES(cls->mutex) |
13751fd8 JR |
368 | { |
369 | struct cls_trie *trie = &cls->tries[trie_idx]; | |
370 | struct cls_subtable *subtable; | |
371 | ||
372 | if (trie_idx < cls->n_tries) { | |
f358a2cb JR |
373 | trie_destroy(&trie->root); |
374 | } else { | |
375 | ovsrcu_set_hidden(&trie->root, NULL); | |
13751fd8 | 376 | } |
13751fd8 JR |
377 | trie->field = field; |
378 | ||
f358a2cb | 379 | /* Add existing rules to the new trie. */ |
f2c21402 | 380 | CMAP_FOR_EACH (subtable, cmap_node, &cls->subtables_map) { |
13751fd8 JR |
381 | unsigned int plen; |
382 | ||
383 | plen = field ? minimask_get_prefix_len(&subtable->mask, field) : 0; | |
13751fd8 | 384 | if (plen) { |
627fb667 | 385 | struct cls_match *head; |
13751fd8 | 386 | |
f2c21402 | 387 | CMAP_FOR_EACH (head, cmap_node, &subtable->rules) { |
627fb667 | 388 | struct cls_match *match; |
13751fd8 | 389 | |
dfea28b3 | 390 | FOR_EACH_RULE_IN_LIST_PROTECTED (match, head) { |
627fb667 | 391 | trie_insert(trie, match->cls_rule, plen); |
13751fd8 JR |
392 | } |
393 | } | |
394 | } | |
f358a2cb JR |
395 | /* Initialize subtable's prefix length on this field. This will |
396 | * allow readers to use the trie. */ | |
397 | atomic_thread_fence(memory_order_release); | |
398 | subtable->trie_plen[trie_idx] = plen; | |
13751fd8 JR |
399 | } |
400 | } | |
401 | ||
5f0476ce JR |
402 | /* Returns true if 'cls' contains no classification rules, false otherwise. |
403 | * Checking the cmap requires no locking. */ | |
064af421 BP |
404 | bool |
405 | classifier_is_empty(const struct classifier *cls) | |
406 | { | |
e48eccd1 | 407 | return cmap_is_empty(&cls->subtables_map); |
064af421 BP |
408 | } |
409 | ||
dbda2960 | 410 | /* Returns the number of rules in 'cls'. */ |
064af421 BP |
411 | int |
412 | classifier_count(const struct classifier *cls) | |
afae68b1 | 413 | OVS_NO_THREAD_SAFETY_ANALYSIS |
064af421 | 414 | { |
afae68b1 JR |
415 | /* n_rules is an int, so in the presence of concurrent writers this will |
416 | * return either the old or a new value. */ | |
e48eccd1 | 417 | return cls->n_rules; |
064af421 BP |
418 | } |
419 | ||
c906cedf BP |
420 | static uint32_t |
421 | hash_metadata(ovs_be64 metadata_) | |
422 | { | |
423 | uint64_t metadata = (OVS_FORCE uint64_t) metadata_; | |
965607c8 | 424 | return hash_uint64(metadata); |
c906cedf BP |
425 | } |
426 | ||
427 | static struct cls_partition * | |
e48eccd1 | 428 | find_partition(const struct classifier *cls, ovs_be64 metadata, uint32_t hash) |
c906cedf BP |
429 | { |
430 | struct cls_partition *partition; | |
431 | ||
f2c21402 | 432 | CMAP_FOR_EACH_WITH_HASH (partition, cmap_node, hash, &cls->partitions) { |
c906cedf BP |
433 | if (partition->metadata == metadata) { |
434 | return partition; | |
435 | } | |
436 | } | |
437 | ||
438 | return NULL; | |
439 | } | |
440 | ||
441 | static struct cls_partition * | |
e48eccd1 | 442 | create_partition(struct classifier *cls, struct cls_subtable *subtable, |
c906cedf | 443 | ovs_be64 metadata) |
e65413ab | 444 | OVS_REQUIRES(cls->mutex) |
c906cedf BP |
445 | { |
446 | uint32_t hash = hash_metadata(metadata); | |
447 | struct cls_partition *partition = find_partition(cls, metadata, hash); | |
448 | if (!partition) { | |
449 | partition = xmalloc(sizeof *partition); | |
450 | partition->metadata = metadata; | |
451 | partition->tags = 0; | |
183126a1 | 452 | tag_tracker_init(&partition->tracker); |
f2c21402 | 453 | cmap_insert(&cls->partitions, &partition->cmap_node, hash); |
c906cedf | 454 | } |
03868246 | 455 | tag_tracker_add(&partition->tracker, &partition->tags, subtable->tag); |
c906cedf BP |
456 | return partition; |
457 | } | |
458 | ||
69d6040e JR |
459 | static inline ovs_be32 minimatch_get_ports(const struct minimatch *match) |
460 | { | |
461 | /* Could optimize to use the same map if needed for fast path. */ | |
462 | return MINIFLOW_GET_BE32(&match->flow, tp_src) | |
463 | & MINIFLOW_GET_BE32(&match->mask.masks, tp_src); | |
464 | } | |
465 | ||
b5d97350 BP |
466 | /* Inserts 'rule' into 'cls'. Until 'rule' is removed from 'cls', the caller |
467 | * must not modify or free it. | |
064af421 BP |
468 | * |
469 | * If 'cls' already contains an identical rule (including wildcards, values of | |
470 | * fixed fields, and priority), replaces the old rule by 'rule' and returns the | |
471 | * rule that was replaced. The caller takes ownership of the returned rule and | |
48d28ac1 BP |
472 | * is thus responsible for destroying it with cls_rule_destroy(), freeing the |
473 | * memory block in which it resides, etc., as necessary. | |
064af421 BP |
474 | * |
475 | * Returns NULL if 'cls' does not contain a rule with an identical key, after | |
476 | * inserting the new rule. In this case, no rules are displaced by the new | |
477 | * rule, even rules that cannot have any effect because the new rule matches a | |
886af6ea JR |
478 | * superset of their flows and has higher priority. |
479 | */ | |
dfea28b3 | 480 | const struct cls_rule * |
e48eccd1 JR |
481 | classifier_replace(struct classifier *cls, struct cls_rule *rule) |
482 | OVS_EXCLUDED(cls->mutex) | |
064af421 | 483 | { |
886af6ea JR |
484 | struct cls_match *new = cls_match_alloc(rule); |
485 | struct cls_match *old_rule = NULL; | |
03868246 | 486 | struct cls_subtable *subtable; |
dfea28b3 | 487 | const struct cls_rule *old_cls_rule = NULL; |
886af6ea JR |
488 | uint32_t ihash[CLS_MAX_INDICES]; |
489 | uint8_t prev_be32ofs = 0; | |
490 | struct cls_match *head; | |
491 | uint32_t basis; | |
492 | uint32_t hash; | |
493 | int i; | |
b5d97350 | 494 | |
e65413ab | 495 | ovs_mutex_lock(&cls->mutex); |
03868246 JR |
496 | subtable = find_subtable(cls, &rule->match.mask); |
497 | if (!subtable) { | |
498 | subtable = insert_subtable(cls, &rule->match.mask); | |
b5d97350 BP |
499 | } |
500 | ||
886af6ea JR |
501 | /* Add new node to segment indices. |
502 | * | |
503 | * Readers may find the rule in the indices before the rule is visible in | |
504 | * the subtables 'rules' map. This may result in us losing the opportunity | |
505 | * to quit lookups earlier, resulting in sub-optimal wildcarding. This | |
506 | * will be fixed later by revalidation (always scheduled after flow table | |
507 | * changes). */ | |
508 | basis = 0; | |
509 | for (i = 0; i < subtable->n_indices; i++) { | |
510 | ihash[i] = minimatch_hash_range(&rule->match, prev_be32ofs, | |
511 | subtable->index_ofs[i], &basis); | |
512 | cmap_insert(&subtable->indices[i], &new->index_nodes[i], ihash[i]); | |
513 | prev_be32ofs = subtable->index_ofs[i]; | |
514 | } | |
515 | hash = minimatch_hash_range(&rule->match, prev_be32ofs, FLOW_U32S, &basis); | |
516 | head = find_equal(subtable, &rule->match.flow, hash); | |
517 | if (!head) { | |
518 | cmap_insert(&subtable->rules, &new->cmap_node, hash); | |
519 | rculist_init(&new->list); | |
520 | } else { | |
521 | /* Scan the list for the insertion point that will keep the list in | |
522 | * order of decreasing priority. */ | |
523 | struct cls_match *iter; | |
524 | ||
525 | FOR_EACH_RULE_IN_LIST_PROTECTED (iter, head) { | |
526 | if (new->priority >= iter->priority) { | |
527 | if (iter == head) { | |
528 | /* 'new' is the new highest-priority flow in the list. */ | |
529 | cmap_replace(&subtable->rules, &iter->cmap_node, | |
530 | &new->cmap_node, hash); | |
531 | } | |
532 | ||
533 | if (new->priority == iter->priority) { | |
534 | rculist_replace(&new->list, &iter->list); | |
535 | old_rule = iter; | |
536 | } else { | |
537 | rculist_insert(&iter->list, &new->list); | |
538 | } | |
539 | goto out; | |
540 | } | |
541 | } | |
542 | ||
543 | /* Insert 'new' at the end of the list. */ | |
544 | rculist_push_back(&head->list, &new->list); | |
545 | out:; | |
546 | } | |
547 | ||
b5d97350 | 548 | if (!old_rule) { |
e65413ab | 549 | old_cls_rule = NULL; |
886af6ea | 550 | subtable->n_rules++; |
13751fd8 | 551 | |
886af6ea JR |
552 | /* Rule was added, not replaced. Update 'subtable's 'max_priority' |
553 | * and 'max_count', if necessary. | |
554 | * | |
555 | * The rule was already inserted, but concurrent readers may not see | |
556 | * the rule yet as the subtables vector is not updated yet. | |
557 | * This will have to be fixed by revalidation later. */ | |
558 | if (subtable->n_rules == 1) { | |
559 | subtable->max_priority = new->priority; | |
560 | subtable->max_count = 1; | |
561 | pvector_insert(&cls->subtables, subtable, new->priority); | |
562 | } else if (subtable->max_priority == new->priority) { | |
563 | ++subtable->max_count; | |
564 | } else if (new->priority > subtable->max_priority) { | |
565 | subtable->max_priority = new->priority; | |
566 | subtable->max_count = 1; | |
567 | pvector_change_priority(&cls->subtables, subtable, new->priority); | |
568 | } | |
569 | ||
570 | /* Add rule to partitions. | |
571 | * | |
572 | * Concurrent readers might miss seeing the rule until this update, | |
573 | * which might require being fixed up by revalidation later. */ | |
627fb667 | 574 | rule->cls_match->partition = NULL; |
c906cedf BP |
575 | if (minimask_get_metadata_mask(&rule->match.mask) == OVS_BE64_MAX) { |
576 | ovs_be64 metadata = miniflow_get_metadata(&rule->match.flow); | |
627fb667 JR |
577 | rule->cls_match->partition = create_partition(cls, subtable, |
578 | metadata); | |
c906cedf BP |
579 | } |
580 | ||
064af421 | 581 | cls->n_rules++; |
13751fd8 | 582 | |
886af6ea JR |
583 | /* Add rule to tries. |
584 | * | |
585 | * Concurrent readers might miss seeing the rule until this update, | |
586 | * which might require being fixed up by revalidation later. */ | |
e65413ab | 587 | for (int i = 0; i < cls->n_tries; i++) { |
13751fd8 JR |
588 | if (subtable->trie_plen[i]) { |
589 | trie_insert(&cls->tries[i], rule, subtable->trie_plen[i]); | |
590 | } | |
591 | } | |
69d6040e | 592 | |
886af6ea | 593 | /* Add rule to ports trie. */ |
69d6040e JR |
594 | if (subtable->ports_mask_len) { |
595 | /* We mask the value to be inserted to always have the wildcarded | |
596 | * bits in known (zero) state, so we can include them in comparison | |
597 | * and they will always match (== their original value does not | |
598 | * matter). */ | |
599 | ovs_be32 masked_ports = minimatch_get_ports(&rule->match); | |
600 | ||
601 | trie_insert_prefix(&subtable->ports_trie, &masked_ports, | |
602 | subtable->ports_mask_len); | |
603 | } | |
c906cedf | 604 | } else { |
e65413ab | 605 | old_cls_rule = old_rule->cls_rule; |
886af6ea JR |
606 | |
607 | /* Remove old node from indices. | |
608 | * | |
609 | * The effect of this late removal of a duplicate rule on concurrent | |
610 | * readers is similar to that of adding the new rule to the segment | |
611 | * indices early (at the beginning of this function.) */ | |
612 | for (i = 0; i < subtable->n_indices; i++) { | |
613 | cmap_remove(&subtable->indices[i], &old_rule->index_nodes[i], | |
614 | ihash[i]); | |
615 | } | |
616 | ||
627fb667 | 617 | rule->cls_match->partition = old_rule->partition; |
dfea28b3 | 618 | CONST_CAST(struct cls_rule *, old_cls_rule)->cls_match = NULL; |
f2c21402 JR |
619 | |
620 | /* 'old_rule' contains a cmap_node, which may not be freed | |
621 | * immediately. */ | |
622 | ovsrcu_postpone(free, old_rule); | |
064af421 | 623 | } |
886af6ea | 624 | |
e65413ab JR |
625 | ovs_mutex_unlock(&cls->mutex); |
626 | return old_cls_rule; | |
064af421 BP |
627 | } |
628 | ||
08944c1d BP |
629 | /* Inserts 'rule' into 'cls'. Until 'rule' is removed from 'cls', the caller |
630 | * must not modify or free it. | |
631 | * | |
632 | * 'cls' must not contain an identical rule (including wildcards, values of | |
633 | * fixed fields, and priority). Use classifier_find_rule_exactly() to find | |
634 | * such a rule. */ | |
635 | void | |
636 | classifier_insert(struct classifier *cls, struct cls_rule *rule) | |
637 | { | |
dfea28b3 | 638 | const struct cls_rule *displaced_rule = classifier_replace(cls, rule); |
cb22974d | 639 | ovs_assert(!displaced_rule); |
08944c1d BP |
640 | } |
641 | ||
48d28ac1 BP |
642 | /* Removes 'rule' from 'cls'. It is the caller's responsibility to destroy |
643 | * 'rule' with cls_rule_destroy(), freeing the memory block in which 'rule' | |
747f140a JR |
644 | * resides, etc., as necessary. |
645 | * | |
646 | * Does nothing if 'rule' has been already removed, or was never inserted. | |
647 | * | |
648 | * Returns the removed rule, or NULL, if it was already removed. | |
649 | */ | |
dfea28b3 JR |
650 | const struct cls_rule * |
651 | classifier_remove(struct classifier *cls, const struct cls_rule *rule) | |
e48eccd1 | 652 | OVS_EXCLUDED(cls->mutex) |
064af421 | 653 | { |
c906cedf | 654 | struct cls_partition *partition; |
747f140a | 655 | struct cls_match *cls_match; |
627fb667 | 656 | struct cls_match *head; |
03868246 | 657 | struct cls_subtable *subtable; |
476f36e8 | 658 | int i; |
f2c21402 JR |
659 | uint32_t basis = 0, hash, ihash[CLS_MAX_INDICES]; |
660 | uint8_t prev_be32ofs = 0; | |
064af421 | 661 | |
e65413ab | 662 | ovs_mutex_lock(&cls->mutex); |
747f140a JR |
663 | cls_match = rule->cls_match; |
664 | if (!cls_match) { | |
665 | rule = NULL; | |
666 | goto unlock; /* Already removed. */ | |
667 | } | |
668 | ||
03868246 | 669 | subtable = find_subtable(cls, &rule->match.mask); |
627fb667 JR |
670 | ovs_assert(subtable); |
671 | ||
69d6040e JR |
672 | if (subtable->ports_mask_len) { |
673 | ovs_be32 masked_ports = minimatch_get_ports(&rule->match); | |
674 | ||
675 | trie_remove_prefix(&subtable->ports_trie, | |
676 | &masked_ports, subtable->ports_mask_len); | |
677 | } | |
13751fd8 JR |
678 | for (i = 0; i < cls->n_tries; i++) { |
679 | if (subtable->trie_plen[i]) { | |
680 | trie_remove(&cls->tries[i], rule, subtable->trie_plen[i]); | |
681 | } | |
682 | } | |
683 | ||
476f36e8 JR |
684 | /* Remove rule node from indices. */ |
685 | for (i = 0; i < subtable->n_indices; i++) { | |
f2c21402 JR |
686 | ihash[i] = minimatch_hash_range(&rule->match, prev_be32ofs, |
687 | subtable->index_ofs[i], &basis); | |
688 | cmap_remove(&subtable->indices[i], &cls_match->index_nodes[i], | |
689 | ihash[i]); | |
690 | prev_be32ofs = subtable->index_ofs[i]; | |
476f36e8 | 691 | } |
f2c21402 | 692 | hash = minimatch_hash_range(&rule->match, prev_be32ofs, FLOW_U32S, &basis); |
476f36e8 | 693 | |
f2c21402 | 694 | head = find_equal(subtable, &rule->match.flow, hash); |
627fb667 | 695 | if (head != cls_match) { |
c501b427 JR |
696 | rculist_remove(&cls_match->list); |
697 | } else if (rculist_is_empty(&cls_match->list)) { | |
f2c21402 | 698 | cmap_remove(&subtable->rules, &cls_match->cmap_node, hash); |
b5d97350 | 699 | } else { |
dfea28b3 | 700 | struct cls_match *next = next_rule_in_list_protected(cls_match); |
064af421 | 701 | |
c501b427 | 702 | rculist_remove(&cls_match->list); |
f2c21402 JR |
703 | cmap_replace(&subtable->rules, &cls_match->cmap_node, |
704 | &next->cmap_node, hash); | |
b5d97350 | 705 | } |
064af421 | 706 | |
627fb667 | 707 | partition = cls_match->partition; |
183126a1 BP |
708 | if (partition) { |
709 | tag_tracker_subtract(&partition->tracker, &partition->tags, | |
03868246 | 710 | subtable->tag); |
183126a1 | 711 | if (!partition->tags) { |
f2c21402 JR |
712 | cmap_remove(&cls->partitions, &partition->cmap_node, |
713 | hash_metadata(partition->metadata)); | |
714 | ovsrcu_postpone(free, partition); | |
183126a1 | 715 | } |
c906cedf BP |
716 | } |
717 | ||
03868246 JR |
718 | if (--subtable->n_rules == 0) { |
719 | destroy_subtable(cls, subtable); | |
fe7cfa5c JR |
720 | } else if (subtable->max_priority == cls_match->priority |
721 | && --subtable->max_count == 0) { | |
722 | /* Find the new 'max_priority' and 'max_count'. */ | |
723 | struct cls_match *head; | |
eb391b76 | 724 | int max_priority = INT_MIN; |
fe7cfa5c | 725 | |
f2c21402 | 726 | CMAP_FOR_EACH (head, cmap_node, &subtable->rules) { |
fe7cfa5c JR |
727 | if (head->priority > max_priority) { |
728 | max_priority = head->priority; | |
729 | subtable->max_count = 1; | |
730 | } else if (head->priority == max_priority) { | |
731 | ++subtable->max_count; | |
732 | } | |
733 | } | |
734 | subtable->max_priority = max_priority; | |
735 | pvector_change_priority(&cls->subtables, subtable, max_priority); | |
4d935a6b | 736 | } |
13751fd8 | 737 | |
b5d97350 | 738 | cls->n_rules--; |
627fb667 | 739 | |
f2c21402 | 740 | ovsrcu_postpone(free, cls_match); |
dfea28b3 | 741 | CONST_CAST(struct cls_rule *, rule)->cls_match = NULL; |
747f140a | 742 | unlock: |
e65413ab | 743 | ovs_mutex_unlock(&cls->mutex); |
747f140a JR |
744 | |
745 | return rule; | |
064af421 BP |
746 | } |
747 | ||
13751fd8 | 748 | /* Prefix tree context. Valid when 'lookup_done' is true. Can skip all |
c0bfb650 JR |
749 | * subtables which have a prefix match on the trie field, but whose prefix |
750 | * length is not indicated in 'match_plens'. For example, a subtable that | |
751 | * has a 8-bit trie field prefix match can be skipped if | |
752 | * !be_get_bit_at(&match_plens, 8 - 1). If skipped, 'maskbits' prefix bits | |
753 | * must be unwildcarded to make datapath flow only match packets it should. */ | |
13751fd8 JR |
754 | struct trie_ctx { |
755 | const struct cls_trie *trie; | |
756 | bool lookup_done; /* Status of the lookup. */ | |
757 | uint8_t be32ofs; /* U32 offset of the field in question. */ | |
13751fd8 | 758 | unsigned int maskbits; /* Prefix length needed to avoid false matches. */ |
c0bfb650 JR |
759 | union mf_value match_plens; /* Bitmask of prefix lengths with possible |
760 | * matches. */ | |
13751fd8 JR |
761 | }; |
762 | ||
763 | static void | |
764 | trie_ctx_init(struct trie_ctx *ctx, const struct cls_trie *trie) | |
765 | { | |
766 | ctx->trie = trie; | |
767 | ctx->be32ofs = trie->field->flow_be32ofs; | |
768 | ctx->lookup_done = false; | |
769 | } | |
770 | ||
48c3de13 BP |
771 | /* Finds and returns the highest-priority rule in 'cls' that matches 'flow'. |
772 | * Returns a null pointer if no rules in 'cls' match 'flow'. If multiple rules | |
74f74083 EJ |
773 | * of equal priority match 'flow', returns one arbitrarily. |
774 | * | |
775 | * If a rule is found and 'wc' is non-null, bitwise-OR's 'wc' with the | |
776 | * set of bits that were significant in the lookup. At some point | |
777 | * earlier, 'wc' should have been initialized (e.g., by | |
778 | * flow_wildcards_init_catchall()). */ | |
dfea28b3 | 779 | const struct cls_rule * |
e48eccd1 | 780 | classifier_lookup(const struct classifier *cls, const struct flow *flow, |
74f74083 | 781 | struct flow_wildcards *wc) |
48c3de13 | 782 | { |
c906cedf | 783 | const struct cls_partition *partition; |
c906cedf | 784 | tag_type tags; |
eb391b76 | 785 | int best_priority = INT_MIN; |
fe7cfa5c JR |
786 | const struct cls_match *best; |
787 | struct trie_ctx trie_ctx[CLS_MAX_TRIES]; | |
788 | struct cls_subtable *subtable; | |
c906cedf | 789 | |
f358a2cb JR |
790 | /* Synchronize for cls->n_tries and subtable->trie_plen. They can change |
791 | * when table configuration changes, which happens typically only on | |
792 | * startup. */ | |
793 | atomic_thread_fence(memory_order_acquire); | |
794 | ||
03868246 JR |
795 | /* Determine 'tags' such that, if 'subtable->tag' doesn't intersect them, |
796 | * then 'flow' cannot possibly match in 'subtable': | |
c906cedf BP |
797 | * |
798 | * - If flow->metadata maps to a given 'partition', then we can use | |
799 | * 'tags' for 'partition->tags'. | |
800 | * | |
801 | * - If flow->metadata has no partition, then no rule in 'cls' has an | |
802 | * exact-match for flow->metadata. That means that we don't need to | |
03868246 | 803 | * search any subtable that includes flow->metadata in its mask. |
c906cedf | 804 | * |
03868246 | 805 | * In either case, we always need to search any cls_subtables that do not |
c906cedf | 806 | * include flow->metadata in its mask. One way to do that would be to |
03868246 JR |
807 | * check the "cls_subtable"s explicitly for that, but that would require an |
808 | * extra branch per subtable. Instead, we mark such a cls_subtable's | |
809 | * 'tags' as TAG_ALL and make sure that 'tags' is never empty. This means | |
810 | * that 'tags' always intersects such a cls_subtable's 'tags', so we don't | |
811 | * need a special case. | |
c906cedf | 812 | */ |
f2c21402 | 813 | partition = (cmap_is_empty(&cls->partitions) |
c906cedf BP |
814 | ? NULL |
815 | : find_partition(cls, flow->metadata, | |
816 | hash_metadata(flow->metadata))); | |
817 | tags = partition ? partition->tags : TAG_ARBITRARY; | |
48c3de13 | 818 | |
ff8241db | 819 | /* Initialize trie contexts for find_match_wc(). */ |
fe7cfa5c | 820 | for (int i = 0; i < cls->n_tries; i++) { |
13751fd8 JR |
821 | trie_ctx_init(&trie_ctx[i], &cls->tries[i]); |
822 | } | |
ec988646 | 823 | |
b5d97350 | 824 | best = NULL; |
fe7cfa5c JR |
825 | PVECTOR_FOR_EACH_PRIORITY(subtable, best_priority, 2, |
826 | sizeof(struct cls_subtable), &cls->subtables) { | |
dfea28b3 | 827 | const struct cls_match *rule; |
c906cedf | 828 | |
fe7cfa5c | 829 | if (!tag_intersects(tags, subtable->tag)) { |
c906cedf BP |
830 | continue; |
831 | } | |
74f74083 | 832 | |
fe7cfa5c | 833 | rule = find_match_wc(subtable, flow, trie_ctx, cls->n_tries, wc); |
eb391b76 BP |
834 | if (rule && rule->priority > best_priority) { |
835 | best_priority = rule->priority; | |
1f3c5efc | 836 | best = rule; |
b5d97350 | 837 | } |
48c3de13 | 838 | } |
13751fd8 | 839 | |
627fb667 | 840 | return best ? best->cls_rule : NULL; |
48c3de13 BP |
841 | } |
842 | ||
b5d97350 BP |
843 | /* Finds and returns a rule in 'cls' with exactly the same priority and |
844 | * matching criteria as 'target'. Returns a null pointer if 'cls' doesn't | |
c084ce1d | 845 | * contain an exact match. */ |
dfea28b3 | 846 | const struct cls_rule * |
e48eccd1 | 847 | classifier_find_rule_exactly(const struct classifier *cls, |
76ecc721 | 848 | const struct cls_rule *target) |
064af421 | 849 | { |
dfea28b3 JR |
850 | const struct cls_match *head, *rule; |
851 | const struct cls_subtable *subtable; | |
064af421 | 852 | |
03868246 | 853 | subtable = find_subtable(cls, &target->match.mask); |
0722ee5c | 854 | if (!subtable) { |
98abae4a | 855 | return NULL; |
4d935a6b JR |
856 | } |
857 | ||
03868246 | 858 | head = find_equal(subtable, &target->match.flow, |
5cb7a798 BP |
859 | miniflow_hash_in_minimask(&target->match.flow, |
860 | &target->match.mask, 0)); | |
98abae4a JR |
861 | if (!head) { |
862 | return NULL; | |
863 | } | |
b5d97350 BP |
864 | FOR_EACH_RULE_IN_LIST (rule, head) { |
865 | if (target->priority >= rule->priority) { | |
627fb667 | 866 | return target->priority == rule->priority ? rule->cls_rule : NULL; |
064af421 BP |
867 | } |
868 | } | |
869 | return NULL; | |
870 | } | |
871 | ||
81a76618 BP |
872 | /* Finds and returns a rule in 'cls' with priority 'priority' and exactly the |
873 | * same matching criteria as 'target'. Returns a null pointer if 'cls' doesn't | |
874 | * contain an exact match. */ | |
dfea28b3 | 875 | const struct cls_rule * |
81a76618 | 876 | classifier_find_match_exactly(const struct classifier *cls, |
eb391b76 | 877 | const struct match *target, int priority) |
81a76618 | 878 | { |
dfea28b3 | 879 | const struct cls_rule *retval; |
81a76618 BP |
880 | struct cls_rule cr; |
881 | ||
882 | cls_rule_init(&cr, target, priority); | |
883 | retval = classifier_find_rule_exactly(cls, &cr); | |
48d28ac1 | 884 | cls_rule_destroy(&cr); |
81a76618 BP |
885 | |
886 | return retval; | |
887 | } | |
888 | ||
faa50f40 BP |
889 | /* Checks if 'target' would overlap any other rule in 'cls'. Two rules are |
890 | * considered to overlap if both rules have the same priority and a packet | |
891 | * could match both. */ | |
49bdc010 | 892 | bool |
e48eccd1 | 893 | classifier_rule_overlaps(const struct classifier *cls, |
faa50f40 | 894 | const struct cls_rule *target) |
e48eccd1 | 895 | OVS_EXCLUDED(cls->mutex) |
49bdc010 | 896 | { |
03868246 | 897 | struct cls_subtable *subtable; |
49bdc010 | 898 | |
e65413ab | 899 | ovs_mutex_lock(&cls->mutex); |
03868246 | 900 | /* Iterate subtables in the descending max priority order. */ |
eb391b76 | 901 | PVECTOR_FOR_EACH_PRIORITY (subtable, target->priority - 1, 2, |
fe7cfa5c | 902 | sizeof(struct cls_subtable), &cls->subtables) { |
5cb7a798 BP |
903 | uint32_t storage[FLOW_U32S]; |
904 | struct minimask mask; | |
627fb667 | 905 | struct cls_match *head; |
49bdc010 | 906 | |
03868246 | 907 | minimask_combine(&mask, &target->match.mask, &subtable->mask, storage); |
f2c21402 | 908 | CMAP_FOR_EACH (head, cmap_node, &subtable->rules) { |
627fb667 | 909 | struct cls_match *rule; |
49bdc010 | 910 | |
dfea28b3 | 911 | FOR_EACH_RULE_IN_LIST_PROTECTED (rule, head) { |
4d935a6b JR |
912 | if (rule->priority < target->priority) { |
913 | break; /* Rules in descending priority order. */ | |
914 | } | |
faa50f40 | 915 | if (rule->priority == target->priority |
5cb7a798 | 916 | && miniflow_equal_in_minimask(&target->match.flow, |
3016f3e4 | 917 | &rule->flow, &mask)) { |
e65413ab | 918 | ovs_mutex_unlock(&cls->mutex); |
49bdc010 JP |
919 | return true; |
920 | } | |
921 | } | |
922 | } | |
923 | } | |
924 | ||
e65413ab | 925 | ovs_mutex_unlock(&cls->mutex); |
49bdc010 JP |
926 | return false; |
927 | } | |
6ceeaa92 BP |
928 | |
929 | /* Returns true if 'rule' exactly matches 'criteria' or if 'rule' is more | |
930 | * specific than 'criteria'. That is, 'rule' matches 'criteria' and this | |
931 | * function returns true if, for every field: | |
932 | * | |
933 | * - 'criteria' and 'rule' specify the same (non-wildcarded) value for the | |
934 | * field, or | |
935 | * | |
936 | * - 'criteria' wildcards the field, | |
937 | * | |
938 | * Conversely, 'rule' does not match 'criteria' and this function returns false | |
939 | * if, for at least one field: | |
940 | * | |
941 | * - 'criteria' and 'rule' specify different values for the field, or | |
942 | * | |
943 | * - 'criteria' specifies a value for the field but 'rule' wildcards it. | |
944 | * | |
945 | * Equivalently, the truth table for whether a field matches is: | |
946 | * | |
947 | * rule | |
948 | * | |
949 | * c wildcard exact | |
950 | * r +---------+---------+ | |
951 | * i wild | yes | yes | | |
952 | * t card | | | | |
953 | * e +---------+---------+ | |
954 | * r exact | no |if values| | |
955 | * i | |are equal| | |
956 | * a +---------+---------+ | |
957 | * | |
958 | * This is the matching rule used by OpenFlow 1.0 non-strict OFPT_FLOW_MOD | |
959 | * commands and by OpenFlow 1.0 aggregate and flow stats. | |
960 | * | |
81a76618 | 961 | * Ignores rule->priority. */ |
6ceeaa92 BP |
962 | bool |
963 | cls_rule_is_loose_match(const struct cls_rule *rule, | |
5cb7a798 | 964 | const struct minimatch *criteria) |
6ceeaa92 | 965 | { |
5cb7a798 BP |
966 | return (!minimask_has_extra(&rule->match.mask, &criteria->mask) |
967 | && miniflow_equal_in_minimask(&rule->match.flow, &criteria->flow, | |
968 | &criteria->mask)); | |
6ceeaa92 | 969 | } |
b5d97350 | 970 | \f |
5ecc9d81 BP |
971 | /* Iteration. */ |
972 | ||
973 | static bool | |
627fb667 | 974 | rule_matches(const struct cls_match *rule, const struct cls_rule *target) |
5ecc9d81 BP |
975 | { |
976 | return (!target | |
3016f3e4 | 977 | || miniflow_equal_in_minimask(&rule->flow, |
5cb7a798 BP |
978 | &target->match.flow, |
979 | &target->match.mask)); | |
5ecc9d81 BP |
980 | } |
981 | ||
dfea28b3 | 982 | static const struct cls_match * |
03868246 | 983 | search_subtable(const struct cls_subtable *subtable, |
f2c21402 | 984 | struct cls_cursor *cursor) |
5ecc9d81 | 985 | { |
f2c21402 JR |
986 | if (!cursor->target |
987 | || !minimask_has_extra(&subtable->mask, &cursor->target->match.mask)) { | |
dfea28b3 | 988 | const struct cls_match *rule; |
5ecc9d81 | 989 | |
f2c21402 JR |
990 | CMAP_CURSOR_FOR_EACH (rule, cmap_node, &cursor->rules, |
991 | &subtable->rules) { | |
992 | if (rule_matches(rule, cursor->target)) { | |
5ecc9d81 BP |
993 | return rule; |
994 | } | |
995 | } | |
996 | } | |
997 | return NULL; | |
998 | } | |
999 | ||
5f0476ce JR |
1000 | /* Initializes 'cursor' for iterating through rules in 'cls', and returns the |
1001 | * first matching cls_rule via '*pnode', or NULL if there are no matches. | |
5ecc9d81 | 1002 | * |
6ceeaa92 | 1003 | * - If 'target' is null, the cursor will visit every rule in 'cls'. |
5ecc9d81 | 1004 | * |
6ceeaa92 BP |
1005 | * - If 'target' is nonnull, the cursor will visit each 'rule' in 'cls' |
1006 | * such that cls_rule_is_loose_match(rule, target) returns true. | |
5ecc9d81 | 1007 | * |
6ceeaa92 | 1008 | * Ignores target->priority. */ |
78c8df12 BP |
1009 | struct cls_cursor cls_cursor_start(const struct classifier *cls, |
1010 | const struct cls_rule *target, | |
1011 | bool safe) | |
5f0476ce | 1012 | OVS_NO_THREAD_SAFETY_ANALYSIS |
5ecc9d81 | 1013 | { |
5f0476ce | 1014 | struct cls_cursor cursor; |
03868246 | 1015 | struct cls_subtable *subtable; |
5ecc9d81 | 1016 | |
5f0476ce | 1017 | cursor.safe = safe; |
e48eccd1 | 1018 | cursor.cls = cls; |
5f0476ce | 1019 | cursor.target = target && !cls_rule_is_catchall(target) ? target : NULL; |
78c8df12 | 1020 | cursor.rule = NULL; |
5f0476ce JR |
1021 | |
1022 | /* Find first rule. */ | |
e65413ab | 1023 | ovs_mutex_lock(&cursor.cls->mutex); |
5f0476ce | 1024 | CMAP_CURSOR_FOR_EACH (subtable, cmap_node, &cursor.subtables, |
e65413ab | 1025 | &cursor.cls->subtables_map) { |
dfea28b3 | 1026 | const struct cls_match *rule = search_subtable(subtable, &cursor); |
f2c21402 | 1027 | |
5ecc9d81 | 1028 | if (rule) { |
5f0476ce | 1029 | cursor.subtable = subtable; |
78c8df12 | 1030 | cursor.rule = rule->cls_rule; |
5f0476ce | 1031 | break; |
5ecc9d81 BP |
1032 | } |
1033 | } | |
1034 | ||
5f0476ce | 1035 | /* Leave locked if requested and have a rule. */ |
78c8df12 | 1036 | if (safe || !cursor.rule) { |
e65413ab | 1037 | ovs_mutex_unlock(&cursor.cls->mutex); |
5f0476ce JR |
1038 | } |
1039 | return cursor; | |
1040 | } | |
1041 | ||
dfea28b3 | 1042 | static const struct cls_rule * |
1caa1561 | 1043 | cls_cursor_next(struct cls_cursor *cursor) |
5f0476ce | 1044 | OVS_NO_THREAD_SAFETY_ANALYSIS |
5ecc9d81 | 1045 | { |
dfea28b3 | 1046 | const struct cls_match *rule = cursor->rule->cls_match; |
03868246 | 1047 | const struct cls_subtable *subtable; |
dfea28b3 | 1048 | const struct cls_match *next; |
5ecc9d81 | 1049 | |
955f579d BP |
1050 | next = next_rule_in_list__(rule); |
1051 | if (next->priority < rule->priority) { | |
1caa1561 | 1052 | return next->cls_rule; |
5ecc9d81 BP |
1053 | } |
1054 | ||
955f579d | 1055 | /* 'next' is the head of the list, that is, the rule that is included in |
f2c21402 | 1056 | * the subtable's map. (This is important when the classifier contains |
03868246 | 1057 | * rules that differ only in priority.) */ |
955f579d | 1058 | rule = next; |
f2c21402 | 1059 | CMAP_CURSOR_FOR_EACH_CONTINUE (rule, cmap_node, &cursor->rules) { |
5ecc9d81 | 1060 | if (rule_matches(rule, cursor->target)) { |
1caa1561 | 1061 | return rule->cls_rule; |
5ecc9d81 BP |
1062 | } |
1063 | } | |
1064 | ||
03868246 | 1065 | subtable = cursor->subtable; |
f2c21402 JR |
1066 | CMAP_CURSOR_FOR_EACH_CONTINUE (subtable, cmap_node, &cursor->subtables) { |
1067 | rule = search_subtable(subtable, cursor); | |
5ecc9d81 | 1068 | if (rule) { |
03868246 | 1069 | cursor->subtable = subtable; |
1caa1561 | 1070 | return rule->cls_rule; |
5ecc9d81 BP |
1071 | } |
1072 | } | |
1073 | ||
1caa1561 BP |
1074 | return NULL; |
1075 | } | |
1076 | ||
1077 | /* Sets 'cursor->rule' to the next matching cls_rule in 'cursor''s iteration, | |
1078 | * or to null if all matching rules have been visited. */ | |
1079 | void | |
1080 | cls_cursor_advance(struct cls_cursor *cursor) | |
1081 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
1082 | { | |
1083 | if (cursor->safe) { | |
1084 | ovs_mutex_lock(&cursor->cls->mutex); | |
1085 | } | |
1086 | cursor->rule = cls_cursor_next(cursor); | |
1087 | if (cursor->safe || !cursor->rule) { | |
1088 | ovs_mutex_unlock(&cursor->cls->mutex); | |
1089 | } | |
5ecc9d81 BP |
1090 | } |
1091 | \f | |
03868246 | 1092 | static struct cls_subtable * |
e48eccd1 | 1093 | find_subtable(const struct classifier *cls, const struct minimask *mask) |
b5d97350 | 1094 | { |
03868246 | 1095 | struct cls_subtable *subtable; |
064af421 | 1096 | |
f2c21402 | 1097 | CMAP_FOR_EACH_WITH_HASH (subtable, cmap_node, minimask_hash(mask, 0), |
5a87054c | 1098 | &cls->subtables_map) { |
03868246 JR |
1099 | if (minimask_equal(mask, &subtable->mask)) { |
1100 | return subtable; | |
064af421 BP |
1101 | } |
1102 | } | |
b5d97350 | 1103 | return NULL; |
064af421 | 1104 | } |
064af421 | 1105 | |
e65413ab | 1106 | /* The new subtable will be visible to the readers only after this. */ |
03868246 | 1107 | static struct cls_subtable * |
e48eccd1 | 1108 | insert_subtable(struct classifier *cls, const struct minimask *mask) |
e65413ab | 1109 | OVS_REQUIRES(cls->mutex) |
b5d97350 | 1110 | { |
c906cedf | 1111 | uint32_t hash = minimask_hash(mask, 0); |
03868246 | 1112 | struct cls_subtable *subtable; |
476f36e8 JR |
1113 | int i, index = 0; |
1114 | struct flow_wildcards old, new; | |
1115 | uint8_t prev; | |
3016f3e4 | 1116 | int count = count_1bits(mask->masks.map); |
064af421 | 1117 | |
3016f3e4 JR |
1118 | subtable = xzalloc(sizeof *subtable - sizeof mask->masks.inline_values |
1119 | + MINIFLOW_VALUES_SIZE(count)); | |
f2c21402 | 1120 | cmap_init(&subtable->rules); |
f80028fe JR |
1121 | miniflow_clone_inline(CONST_CAST(struct miniflow *, &subtable->mask.masks), |
1122 | &mask->masks, count); | |
476f36e8 JR |
1123 | |
1124 | /* Init indices for segmented lookup, if any. */ | |
1125 | flow_wildcards_init_catchall(&new); | |
1126 | old = new; | |
1127 | prev = 0; | |
1128 | for (i = 0; i < cls->n_flow_segments; i++) { | |
1129 | flow_wildcards_fold_minimask_range(&new, mask, prev, | |
1130 | cls->flow_segments[i]); | |
1131 | /* Add an index if it adds mask bits. */ | |
1132 | if (!flow_wildcards_equal(&new, &old)) { | |
f2c21402 | 1133 | cmap_init(&subtable->indices[index]); |
f80028fe JR |
1134 | *CONST_CAST(uint8_t *, &subtable->index_ofs[index]) |
1135 | = cls->flow_segments[i]; | |
476f36e8 JR |
1136 | index++; |
1137 | old = new; | |
1138 | } | |
1139 | prev = cls->flow_segments[i]; | |
1140 | } | |
1141 | /* Check if the rest of the subtable's mask adds any bits, | |
1142 | * and remove the last index if it doesn't. */ | |
1143 | if (index > 0) { | |
1144 | flow_wildcards_fold_minimask_range(&new, mask, prev, FLOW_U32S); | |
1145 | if (flow_wildcards_equal(&new, &old)) { | |
1146 | --index; | |
f80028fe | 1147 | *CONST_CAST(uint8_t *, &subtable->index_ofs[index]) = 0; |
f2c21402 | 1148 | cmap_destroy(&subtable->indices[index]); |
476f36e8 JR |
1149 | } |
1150 | } | |
f80028fe | 1151 | *CONST_CAST(uint8_t *, &subtable->n_indices) = index; |
476f36e8 | 1152 | |
f80028fe JR |
1153 | *CONST_CAST(tag_type *, &subtable->tag) = |
1154 | (minimask_get_metadata_mask(mask) == OVS_BE64_MAX | |
1155 | ? tag_create_deterministic(hash) | |
1156 | : TAG_ALL); | |
064af421 | 1157 | |
13751fd8 JR |
1158 | for (i = 0; i < cls->n_tries; i++) { |
1159 | subtable->trie_plen[i] = minimask_get_prefix_len(mask, | |
1160 | cls->tries[i].field); | |
1161 | } | |
1162 | ||
69d6040e | 1163 | /* Ports trie. */ |
f358a2cb | 1164 | ovsrcu_set_hidden(&subtable->ports_trie, NULL); |
f80028fe | 1165 | *CONST_CAST(int *, &subtable->ports_mask_len) |
69d6040e JR |
1166 | = 32 - ctz32(ntohl(MINIFLOW_GET_BE32(&mask->masks, tp_src))); |
1167 | ||
f2c21402 | 1168 | cmap_insert(&cls->subtables_map, &subtable->cmap_node, hash); |
ec988646 | 1169 | |
03868246 | 1170 | return subtable; |
064af421 BP |
1171 | } |
1172 | ||
01c0f83a | 1173 | /* RCU readers may still access the subtable before it is actually freed. */ |
b5d97350 | 1174 | static void |
e48eccd1 | 1175 | destroy_subtable(struct classifier *cls, struct cls_subtable *subtable) |
e65413ab | 1176 | OVS_REQUIRES(cls->mutex) |
b5d97350 | 1177 | { |
476f36e8 JR |
1178 | int i; |
1179 | ||
fe7cfa5c | 1180 | pvector_remove(&cls->subtables, subtable); |
01c0f83a JR |
1181 | cmap_remove(&cls->subtables_map, &subtable->cmap_node, |
1182 | minimask_hash(&subtable->mask, 0)); | |
1183 | ||
1184 | ovs_assert(ovsrcu_get_protected(struct trie_node *, &subtable->ports_trie) | |
1185 | == NULL); | |
1186 | ovs_assert(cmap_is_empty(&subtable->rules)); | |
69d6040e | 1187 | |
476f36e8 | 1188 | for (i = 0; i < subtable->n_indices; i++) { |
f2c21402 | 1189 | cmap_destroy(&subtable->indices[i]); |
476f36e8 | 1190 | } |
f2c21402 | 1191 | cmap_destroy(&subtable->rules); |
fe7cfa5c | 1192 | ovsrcu_postpone(free, subtable); |
4aacd02d BP |
1193 | } |
1194 | ||
13751fd8 JR |
1195 | struct range { |
1196 | uint8_t start; | |
1197 | uint8_t end; | |
1198 | }; | |
1199 | ||
c0bfb650 JR |
1200 | static unsigned int be_get_bit_at(const ovs_be32 value[], unsigned int ofs); |
1201 | ||
13751fd8 JR |
1202 | /* Return 'true' if can skip rest of the subtable based on the prefix trie |
1203 | * lookup results. */ | |
1204 | static inline bool | |
1205 | check_tries(struct trie_ctx trie_ctx[CLS_MAX_TRIES], unsigned int n_tries, | |
1206 | const unsigned int field_plen[CLS_MAX_TRIES], | |
1207 | const struct range ofs, const struct flow *flow, | |
1208 | struct flow_wildcards *wc) | |
1209 | { | |
1210 | int j; | |
1211 | ||
1212 | /* Check if we could avoid fully unwildcarding the next level of | |
1213 | * fields using the prefix tries. The trie checks are done only as | |
1214 | * needed to avoid folding in additional bits to the wildcards mask. */ | |
1215 | for (j = 0; j < n_tries; j++) { | |
1216 | /* Is the trie field relevant for this subtable? */ | |
1217 | if (field_plen[j]) { | |
1218 | struct trie_ctx *ctx = &trie_ctx[j]; | |
1219 | uint8_t be32ofs = ctx->be32ofs; | |
1220 | ||
1221 | /* Is the trie field within the current range of fields? */ | |
1222 | if (be32ofs >= ofs.start && be32ofs < ofs.end) { | |
1223 | /* On-demand trie lookup. */ | |
1224 | if (!ctx->lookup_done) { | |
c0bfb650 JR |
1225 | memset(&ctx->match_plens, 0, sizeof ctx->match_plens); |
1226 | ctx->maskbits = trie_lookup(ctx->trie, flow, | |
1227 | &ctx->match_plens); | |
13751fd8 JR |
1228 | ctx->lookup_done = true; |
1229 | } | |
1230 | /* Possible to skip the rest of the subtable if subtable's | |
c0bfb650 JR |
1231 | * prefix on the field is not included in the lookup result. */ |
1232 | if (!be_get_bit_at(&ctx->match_plens.be32, field_plen[j] - 1)) { | |
1817dcea JR |
1233 | /* We want the trie lookup to never result in unwildcarding |
1234 | * any bits that would not be unwildcarded otherwise. | |
1235 | * Since the trie is shared by the whole classifier, it is | |
1236 | * possible that the 'maskbits' contain bits that are | |
1237 | * irrelevant for the partition relevant for the current | |
1238 | * packet. Hence the checks below. */ | |
13751fd8 | 1239 | |
13751fd8 | 1240 | /* Check that the trie result will not unwildcard more bits |
1817dcea | 1241 | * than this subtable would otherwise. */ |
13751fd8 JR |
1242 | if (ctx->maskbits <= field_plen[j]) { |
1243 | /* Unwildcard the bits and skip the rest. */ | |
1244 | mask_set_prefix_bits(wc, be32ofs, ctx->maskbits); | |
1245 | /* Note: Prerequisite already unwildcarded, as the only | |
1246 | * prerequisite of the supported trie lookup fields is | |
1817dcea JR |
1247 | * the ethertype, which is always unwildcarded. */ |
1248 | return true; | |
1249 | } | |
1250 | /* Can skip if the field is already unwildcarded. */ | |
1251 | if (mask_prefix_bits_set(wc, be32ofs, ctx->maskbits)) { | |
13751fd8 JR |
1252 | return true; |
1253 | } | |
1254 | } | |
1255 | } | |
1256 | } | |
1257 | } | |
1258 | return false; | |
1259 | } | |
1260 | ||
3016f3e4 JR |
1261 | /* Returns true if 'target' satisifies 'flow'/'mask', that is, if each bit |
1262 | * for which 'flow', for which 'mask' has a bit set, specifies a particular | |
1263 | * value has the correct value in 'target'. | |
1264 | * | |
1265 | * This function is equivalent to miniflow_equal_flow_in_minimask(flow, | |
a64759f0 JR |
1266 | * target, mask) but this is faster because of the invariant that |
1267 | * flow->map and mask->masks.map are the same, and that this version | |
1268 | * takes the 'wc'. */ | |
3016f3e4 JR |
1269 | static inline bool |
1270 | miniflow_and_mask_matches_flow(const struct miniflow *flow, | |
1271 | const struct minimask *mask, | |
e9319757 | 1272 | const struct flow *target) |
3016f3e4 JR |
1273 | { |
1274 | const uint32_t *flowp = miniflow_get_u32_values(flow); | |
1275 | const uint32_t *maskp = miniflow_get_u32_values(&mask->masks); | |
a64759f0 | 1276 | uint32_t idx; |
3016f3e4 | 1277 | |
a64759f0 JR |
1278 | MAP_FOR_EACH_INDEX(idx, mask->masks.map) { |
1279 | uint32_t diff = (*flowp++ ^ flow_u32_value(target, idx)) & *maskp++; | |
1280 | ||
1281 | if (diff) { | |
3016f3e4 JR |
1282 | return false; |
1283 | } | |
1284 | } | |
1285 | ||
1286 | return true; | |
1287 | } | |
1288 | ||
dfea28b3 | 1289 | static inline const struct cls_match * |
476f36e8 JR |
1290 | find_match(const struct cls_subtable *subtable, const struct flow *flow, |
1291 | uint32_t hash) | |
b5d97350 | 1292 | { |
dfea28b3 | 1293 | const struct cls_match *rule; |
b5d97350 | 1294 | |
f2c21402 | 1295 | CMAP_FOR_EACH_WITH_HASH (rule, cmap_node, hash, &subtable->rules) { |
3016f3e4 | 1296 | if (miniflow_and_mask_matches_flow(&rule->flow, &subtable->mask, |
e9319757 | 1297 | flow)) { |
b5d97350 | 1298 | return rule; |
064af421 BP |
1299 | } |
1300 | } | |
c23740be | 1301 | |
064af421 BP |
1302 | return NULL; |
1303 | } | |
1304 | ||
e9319757 JR |
1305 | /* Returns true if 'target' satisifies 'flow'/'mask', that is, if each bit |
1306 | * for which 'flow', for which 'mask' has a bit set, specifies a particular | |
1307 | * value has the correct value in 'target'. | |
1308 | * | |
1309 | * This function is equivalent to miniflow_and_mask_matches_flow() but this | |
1310 | * version fills in the mask bits in 'wc'. */ | |
1311 | static inline bool | |
1312 | miniflow_and_mask_matches_flow_wc(const struct miniflow *flow, | |
1313 | const struct minimask *mask, | |
1314 | const struct flow *target, | |
1315 | struct flow_wildcards *wc) | |
1316 | { | |
1317 | const uint32_t *flowp = miniflow_get_u32_values(flow); | |
1318 | const uint32_t *maskp = miniflow_get_u32_values(&mask->masks); | |
1319 | uint32_t idx; | |
1320 | ||
1321 | MAP_FOR_EACH_INDEX(idx, mask->masks.map) { | |
1322 | uint32_t mask = *maskp++; | |
1323 | uint32_t diff = (*flowp++ ^ flow_u32_value(target, idx)) & mask; | |
1324 | ||
1325 | if (diff) { | |
1326 | /* Only unwildcard if none of the differing bits is already | |
1327 | * exact-matched. */ | |
1328 | if (!(flow_u32_value(&wc->masks, idx) & diff)) { | |
1329 | /* Keep one bit of the difference. */ | |
1330 | *flow_u32_lvalue(&wc->masks, idx) |= rightmost_1bit(diff); | |
1331 | } | |
1332 | return false; | |
1333 | } | |
1334 | /* Fill in the bits that were looked at. */ | |
1335 | *flow_u32_lvalue(&wc->masks, idx) |= mask; | |
1336 | } | |
1337 | ||
1338 | return true; | |
1339 | } | |
1340 | ||
386cb9f7 JR |
1341 | /* Unwildcard the fields looked up so far, if any. */ |
1342 | static void | |
1343 | fill_range_wc(const struct cls_subtable *subtable, struct flow_wildcards *wc, | |
1344 | uint8_t to) | |
1345 | { | |
1346 | if (to) { | |
1347 | flow_wildcards_fold_minimask_range(wc, &subtable->mask, 0, to); | |
1348 | } | |
1349 | } | |
1350 | ||
dfea28b3 | 1351 | static const struct cls_match * |
476f36e8 | 1352 | find_match_wc(const struct cls_subtable *subtable, const struct flow *flow, |
13751fd8 JR |
1353 | struct trie_ctx trie_ctx[CLS_MAX_TRIES], unsigned int n_tries, |
1354 | struct flow_wildcards *wc) | |
476f36e8 JR |
1355 | { |
1356 | uint32_t basis = 0, hash; | |
dfea28b3 | 1357 | const struct cls_match *rule = NULL; |
476f36e8 | 1358 | int i; |
13751fd8 | 1359 | struct range ofs; |
476f36e8 | 1360 | |
ec988646 | 1361 | if (OVS_UNLIKELY(!wc)) { |
476f36e8 JR |
1362 | return find_match(subtable, flow, |
1363 | flow_hash_in_minimask(flow, &subtable->mask, 0)); | |
1364 | } | |
1365 | ||
13751fd8 | 1366 | ofs.start = 0; |
476f36e8 JR |
1367 | /* Try to finish early by checking fields in segments. */ |
1368 | for (i = 0; i < subtable->n_indices; i++) { | |
55847abe | 1369 | const struct cmap_node *inode; |
f2c21402 | 1370 | |
13751fd8 | 1371 | ofs.end = subtable->index_ofs[i]; |
476f36e8 | 1372 | |
13751fd8 JR |
1373 | if (check_tries(trie_ctx, n_tries, subtable->trie_plen, ofs, flow, |
1374 | wc)) { | |
386cb9f7 JR |
1375 | /* 'wc' bits for the trie field set, now unwildcard the preceding |
1376 | * bits used so far. */ | |
1377 | fill_range_wc(subtable, wc, ofs.start); | |
1378 | return NULL; | |
13751fd8 JR |
1379 | } |
1380 | hash = flow_hash_in_minimask_range(flow, &subtable->mask, ofs.start, | |
1381 | ofs.end, &basis); | |
f2c21402 | 1382 | inode = cmap_find(&subtable->indices[i], hash); |
476f36e8 | 1383 | if (!inode) { |
386cb9f7 JR |
1384 | /* No match, can stop immediately, but must fold in the bits |
1385 | * used in lookup so far. */ | |
1386 | fill_range_wc(subtable, wc, ofs.end); | |
1387 | return NULL; | |
476f36e8 JR |
1388 | } |
1389 | ||
1390 | /* If we have narrowed down to a single rule already, check whether | |
a64759f0 | 1391 | * that rule matches. Either way, we're done. |
476f36e8 JR |
1392 | * |
1393 | * (Rare) hash collisions may cause us to miss the opportunity for this | |
1394 | * optimization. */ | |
f2c21402 | 1395 | if (!cmap_node_next(inode)) { |
476f36e8 | 1396 | ASSIGN_CONTAINER(rule, inode - i, index_nodes); |
e9319757 JR |
1397 | if (miniflow_and_mask_matches_flow_wc(&rule->flow, &subtable->mask, |
1398 | flow, wc)) { | |
1399 | return rule; | |
476f36e8 | 1400 | } |
e9319757 | 1401 | return NULL; |
476f36e8 | 1402 | } |
386cb9f7 | 1403 | ofs.start = ofs.end; |
476f36e8 | 1404 | } |
13751fd8 JR |
1405 | ofs.end = FLOW_U32S; |
1406 | /* Trie check for the final range. */ | |
1407 | if (check_tries(trie_ctx, n_tries, subtable->trie_plen, ofs, flow, wc)) { | |
386cb9f7 JR |
1408 | fill_range_wc(subtable, wc, ofs.start); |
1409 | return NULL; | |
13751fd8 | 1410 | } |
a64759f0 JR |
1411 | hash = flow_hash_in_minimask_range(flow, &subtable->mask, ofs.start, |
1412 | ofs.end, &basis); | |
1413 | rule = find_match(subtable, flow, hash); | |
69d6040e JR |
1414 | if (!rule && subtable->ports_mask_len) { |
1415 | /* Ports are always part of the final range, if any. | |
1416 | * No match was found for the ports. Use the ports trie to figure out | |
1417 | * which ports bits to unwildcard. */ | |
1418 | unsigned int mbits; | |
c0bfb650 | 1419 | ovs_be32 value, plens, mask; |
69d6040e JR |
1420 | |
1421 | mask = MINIFLOW_GET_BE32(&subtable->mask.masks, tp_src); | |
1422 | value = ((OVS_FORCE ovs_be32 *)flow)[TP_PORTS_OFS32] & mask; | |
c0bfb650 | 1423 | mbits = trie_lookup_value(&subtable->ports_trie, &value, &plens, 32); |
69d6040e JR |
1424 | |
1425 | ((OVS_FORCE ovs_be32 *)&wc->masks)[TP_PORTS_OFS32] |= | |
86f35fb5 | 1426 | mask & be32_prefix_mask(mbits); |
69d6040e | 1427 | |
386cb9f7 JR |
1428 | /* Unwildcard all bits in the mask upto the ports, as they were used |
1429 | * to determine there is no match. */ | |
1430 | fill_range_wc(subtable, wc, TP_PORTS_OFS32); | |
1431 | return NULL; | |
69d6040e | 1432 | } |
e9319757 | 1433 | |
13751fd8 | 1434 | /* Must unwildcard all the fields, as they were looked at. */ |
476f36e8 JR |
1435 | flow_wildcards_fold_minimask(wc, &subtable->mask); |
1436 | return rule; | |
1437 | } | |
1438 | ||
627fb667 | 1439 | static struct cls_match * |
dfea28b3 | 1440 | find_equal(const struct cls_subtable *subtable, const struct miniflow *flow, |
03868246 | 1441 | uint32_t hash) |
064af421 | 1442 | { |
627fb667 | 1443 | struct cls_match *head; |
064af421 | 1444 | |
f2c21402 | 1445 | CMAP_FOR_EACH_WITH_HASH (head, cmap_node, hash, &subtable->rules) { |
3016f3e4 | 1446 | if (miniflow_equal(&head->flow, flow)) { |
b5d97350 | 1447 | return head; |
064af421 BP |
1448 | } |
1449 | } | |
1450 | return NULL; | |
1451 | } | |
13751fd8 JR |
1452 | \f |
1453 | /* A longest-prefix match tree. */ | |
13751fd8 JR |
1454 | |
1455 | /* Return at least 'plen' bits of the 'prefix', starting at bit offset 'ofs'. | |
1456 | * Prefixes are in the network byte order, and the offset 0 corresponds to | |
1457 | * the most significant bit of the first byte. The offset can be read as | |
1458 | * "how many bits to skip from the start of the prefix starting at 'pr'". */ | |
1459 | static uint32_t | |
1460 | raw_get_prefix(const ovs_be32 pr[], unsigned int ofs, unsigned int plen) | |
1461 | { | |
1462 | uint32_t prefix; | |
1463 | ||
1464 | pr += ofs / 32; /* Where to start. */ | |
1465 | ofs %= 32; /* How many bits to skip at 'pr'. */ | |
1466 | ||
1467 | prefix = ntohl(*pr) << ofs; /* Get the first 32 - ofs bits. */ | |
1468 | if (plen > 32 - ofs) { /* Need more than we have already? */ | |
1469 | prefix |= ntohl(*++pr) >> (32 - ofs); | |
1470 | } | |
1471 | /* Return with possible unwanted bits at the end. */ | |
1472 | return prefix; | |
1473 | } | |
1474 | ||
1475 | /* Return min(TRIE_PREFIX_BITS, plen) bits of the 'prefix', starting at bit | |
1476 | * offset 'ofs'. Prefixes are in the network byte order, and the offset 0 | |
1477 | * corresponds to the most significant bit of the first byte. The offset can | |
1478 | * be read as "how many bits to skip from the start of the prefix starting at | |
1479 | * 'pr'". */ | |
1480 | static uint32_t | |
1481 | trie_get_prefix(const ovs_be32 pr[], unsigned int ofs, unsigned int plen) | |
1482 | { | |
1483 | if (!plen) { | |
1484 | return 0; | |
1485 | } | |
1486 | if (plen > TRIE_PREFIX_BITS) { | |
1487 | plen = TRIE_PREFIX_BITS; /* Get at most TRIE_PREFIX_BITS. */ | |
1488 | } | |
1489 | /* Return with unwanted bits cleared. */ | |
1490 | return raw_get_prefix(pr, ofs, plen) & ~0u << (32 - plen); | |
1491 | } | |
1492 | ||
c30cfa6b | 1493 | /* Return the number of equal bits in 'n_bits' of 'prefix's MSBs and a 'value' |
13751fd8 JR |
1494 | * starting at "MSB 0"-based offset 'ofs'. */ |
1495 | static unsigned int | |
c30cfa6b | 1496 | prefix_equal_bits(uint32_t prefix, unsigned int n_bits, const ovs_be32 value[], |
13751fd8 JR |
1497 | unsigned int ofs) |
1498 | { | |
c30cfa6b | 1499 | uint64_t diff = prefix ^ raw_get_prefix(value, ofs, n_bits); |
13751fd8 | 1500 | /* Set the bit after the relevant bits to limit the result. */ |
c30cfa6b | 1501 | return raw_clz64(diff << 32 | UINT64_C(1) << (63 - n_bits)); |
13751fd8 JR |
1502 | } |
1503 | ||
1504 | /* Return the number of equal bits in 'node' prefix and a 'prefix' of length | |
1505 | * 'plen', starting at "MSB 0"-based offset 'ofs'. */ | |
1506 | static unsigned int | |
1507 | trie_prefix_equal_bits(const struct trie_node *node, const ovs_be32 prefix[], | |
1508 | unsigned int ofs, unsigned int plen) | |
1509 | { | |
c30cfa6b | 1510 | return prefix_equal_bits(node->prefix, MIN(node->n_bits, plen - ofs), |
13751fd8 JR |
1511 | prefix, ofs); |
1512 | } | |
1513 | ||
1514 | /* Return the bit at ("MSB 0"-based) offset 'ofs' as an int. 'ofs' can | |
1515 | * be greater than 31. */ | |
1516 | static unsigned int | |
1517 | be_get_bit_at(const ovs_be32 value[], unsigned int ofs) | |
1518 | { | |
1519 | return (((const uint8_t *)value)[ofs / 8] >> (7 - ofs % 8)) & 1u; | |
1520 | } | |
1521 | ||
1522 | /* Return the bit at ("MSB 0"-based) offset 'ofs' as an int. 'ofs' must | |
1523 | * be between 0 and 31, inclusive. */ | |
1524 | static unsigned int | |
1525 | get_bit_at(const uint32_t prefix, unsigned int ofs) | |
1526 | { | |
1527 | return (prefix >> (31 - ofs)) & 1u; | |
1528 | } | |
1529 | ||
1530 | /* Create new branch. */ | |
1531 | static struct trie_node * | |
1532 | trie_branch_create(const ovs_be32 *prefix, unsigned int ofs, unsigned int plen, | |
1533 | unsigned int n_rules) | |
1534 | { | |
1535 | struct trie_node *node = xmalloc(sizeof *node); | |
1536 | ||
1537 | node->prefix = trie_get_prefix(prefix, ofs, plen); | |
1538 | ||
1539 | if (plen <= TRIE_PREFIX_BITS) { | |
c30cfa6b | 1540 | node->n_bits = plen; |
f358a2cb JR |
1541 | ovsrcu_set_hidden(&node->edges[0], NULL); |
1542 | ovsrcu_set_hidden(&node->edges[1], NULL); | |
13751fd8 JR |
1543 | node->n_rules = n_rules; |
1544 | } else { /* Need intermediate nodes. */ | |
1545 | struct trie_node *subnode = trie_branch_create(prefix, | |
1546 | ofs + TRIE_PREFIX_BITS, | |
1547 | plen - TRIE_PREFIX_BITS, | |
1548 | n_rules); | |
1549 | int bit = get_bit_at(subnode->prefix, 0); | |
c30cfa6b | 1550 | node->n_bits = TRIE_PREFIX_BITS; |
f358a2cb JR |
1551 | ovsrcu_set_hidden(&node->edges[bit], subnode); |
1552 | ovsrcu_set_hidden(&node->edges[!bit], NULL); | |
13751fd8 JR |
1553 | node->n_rules = 0; |
1554 | } | |
1555 | return node; | |
1556 | } | |
1557 | ||
1558 | static void | |
f358a2cb | 1559 | trie_node_destroy(const struct trie_node *node) |
13751fd8 | 1560 | { |
f358a2cb JR |
1561 | ovsrcu_postpone(free, CONST_CAST(struct trie_node *, node)); |
1562 | } | |
1563 | ||
1564 | /* Copy a trie node for modification and postpone delete the old one. */ | |
1565 | static struct trie_node * | |
1566 | trie_node_rcu_realloc(const struct trie_node *node) | |
1567 | { | |
1568 | struct trie_node *new_node = xmalloc(sizeof *node); | |
1569 | ||
1570 | *new_node = *node; | |
1571 | trie_node_destroy(node); | |
1572 | ||
1573 | return new_node; | |
13751fd8 JR |
1574 | } |
1575 | ||
e48eccd1 | 1576 | /* May only be called while holding the classifier mutex. */ |
13751fd8 | 1577 | static void |
f358a2cb | 1578 | trie_destroy(rcu_trie_ptr *trie) |
13751fd8 | 1579 | { |
f358a2cb JR |
1580 | struct trie_node *node = ovsrcu_get_protected(struct trie_node *, trie); |
1581 | ||
13751fd8 | 1582 | if (node) { |
f358a2cb JR |
1583 | ovsrcu_set_hidden(trie, NULL); |
1584 | trie_destroy(&node->edges[0]); | |
1585 | trie_destroy(&node->edges[1]); | |
1586 | trie_node_destroy(node); | |
13751fd8 JR |
1587 | } |
1588 | } | |
1589 | ||
1590 | static bool | |
1591 | trie_is_leaf(const struct trie_node *trie) | |
1592 | { | |
f358a2cb JR |
1593 | /* No children? */ |
1594 | return !ovsrcu_get(struct trie_node *, &trie->edges[0]) | |
1595 | && !ovsrcu_get(struct trie_node *, &trie->edges[1]); | |
13751fd8 JR |
1596 | } |
1597 | ||
1598 | static void | |
1599 | mask_set_prefix_bits(struct flow_wildcards *wc, uint8_t be32ofs, | |
c30cfa6b | 1600 | unsigned int n_bits) |
13751fd8 JR |
1601 | { |
1602 | ovs_be32 *mask = &((ovs_be32 *)&wc->masks)[be32ofs]; | |
1603 | unsigned int i; | |
1604 | ||
c30cfa6b | 1605 | for (i = 0; i < n_bits / 32; i++) { |
13751fd8 JR |
1606 | mask[i] = OVS_BE32_MAX; |
1607 | } | |
c30cfa6b JR |
1608 | if (n_bits % 32) { |
1609 | mask[i] |= htonl(~0u << (32 - n_bits % 32)); | |
13751fd8 JR |
1610 | } |
1611 | } | |
1612 | ||
1613 | static bool | |
1614 | mask_prefix_bits_set(const struct flow_wildcards *wc, uint8_t be32ofs, | |
c30cfa6b | 1615 | unsigned int n_bits) |
13751fd8 JR |
1616 | { |
1617 | ovs_be32 *mask = &((ovs_be32 *)&wc->masks)[be32ofs]; | |
1618 | unsigned int i; | |
1619 | ovs_be32 zeroes = 0; | |
1620 | ||
c30cfa6b | 1621 | for (i = 0; i < n_bits / 32; i++) { |
13751fd8 JR |
1622 | zeroes |= ~mask[i]; |
1623 | } | |
c30cfa6b JR |
1624 | if (n_bits % 32) { |
1625 | zeroes |= ~mask[i] & htonl(~0u << (32 - n_bits % 32)); | |
13751fd8 JR |
1626 | } |
1627 | ||
c30cfa6b | 1628 | return !zeroes; /* All 'n_bits' bits set. */ |
13751fd8 JR |
1629 | } |
1630 | ||
f358a2cb | 1631 | static rcu_trie_ptr * |
13751fd8 JR |
1632 | trie_next_edge(struct trie_node *node, const ovs_be32 value[], |
1633 | unsigned int ofs) | |
1634 | { | |
1635 | return node->edges + be_get_bit_at(value, ofs); | |
1636 | } | |
1637 | ||
1638 | static const struct trie_node * | |
1639 | trie_next_node(const struct trie_node *node, const ovs_be32 value[], | |
1640 | unsigned int ofs) | |
1641 | { | |
f358a2cb JR |
1642 | return ovsrcu_get(struct trie_node *, |
1643 | &node->edges[be_get_bit_at(value, ofs)]); | |
13751fd8 JR |
1644 | } |
1645 | ||
c0bfb650 JR |
1646 | /* Set the bit at ("MSB 0"-based) offset 'ofs'. 'ofs' can be greater than 31. |
1647 | */ | |
1648 | static void | |
1649 | be_set_bit_at(ovs_be32 value[], unsigned int ofs) | |
1650 | { | |
1651 | ((uint8_t *)value)[ofs / 8] |= 1u << (7 - ofs % 8); | |
1652 | } | |
1653 | ||
1654 | /* Returns the number of bits in the prefix mask necessary to determine a | |
1655 | * mismatch, in case there are longer prefixes in the tree below the one that | |
1656 | * matched. | |
1657 | * '*plens' will have a bit set for each prefix length that may have matching | |
1658 | * rules. The caller is responsible for clearing the '*plens' prior to | |
1659 | * calling this. | |
13751fd8 JR |
1660 | */ |
1661 | static unsigned int | |
f358a2cb | 1662 | trie_lookup_value(const rcu_trie_ptr *trie, const ovs_be32 value[], |
c0bfb650 | 1663 | ovs_be32 plens[], unsigned int n_bits) |
13751fd8 | 1664 | { |
13751fd8 | 1665 | const struct trie_node *prev = NULL; |
c0bfb650 JR |
1666 | const struct trie_node *node = ovsrcu_get(struct trie_node *, trie); |
1667 | unsigned int match_len = 0; /* Number of matching bits. */ | |
13751fd8 | 1668 | |
27ce650f | 1669 | for (; node; prev = node, node = trie_next_node(node, value, match_len)) { |
13751fd8 JR |
1670 | unsigned int eqbits; |
1671 | /* Check if this edge can be followed. */ | |
27ce650f JR |
1672 | eqbits = prefix_equal_bits(node->prefix, node->n_bits, value, |
1673 | match_len); | |
1674 | match_len += eqbits; | |
c30cfa6b | 1675 | if (eqbits < node->n_bits) { /* Mismatch, nothing more to be found. */ |
27ce650f | 1676 | /* Bit at offset 'match_len' differed. */ |
c0bfb650 | 1677 | return match_len + 1; /* Includes the first mismatching bit. */ |
13751fd8 JR |
1678 | } |
1679 | /* Full match, check if rules exist at this prefix length. */ | |
1680 | if (node->n_rules > 0) { | |
c0bfb650 | 1681 | be_set_bit_at(plens, match_len - 1); |
13751fd8 | 1682 | } |
27ce650f | 1683 | if (match_len >= n_bits) { |
c0bfb650 | 1684 | return n_bits; /* Full prefix. */ |
f0e5aa11 | 1685 | } |
13751fd8 | 1686 | } |
c0bfb650 JR |
1687 | /* node == NULL. Full match so far, but we tried to follow an |
1688 | * non-existing branch. Need to exclude the other branch if it exists | |
1689 | * (it does not if we were called on an empty trie or 'prev' is a leaf | |
1690 | * node). */ | |
1691 | return !prev || trie_is_leaf(prev) ? match_len : match_len + 1; | |
13751fd8 JR |
1692 | } |
1693 | ||
1694 | static unsigned int | |
1695 | trie_lookup(const struct cls_trie *trie, const struct flow *flow, | |
c0bfb650 | 1696 | union mf_value *plens) |
13751fd8 JR |
1697 | { |
1698 | const struct mf_field *mf = trie->field; | |
1699 | ||
1700 | /* Check that current flow matches the prerequisites for the trie | |
1701 | * field. Some match fields are used for multiple purposes, so we | |
1702 | * must check that the trie is relevant for this flow. */ | |
1703 | if (mf_are_prereqs_ok(mf, flow)) { | |
f358a2cb | 1704 | return trie_lookup_value(&trie->root, |
13751fd8 | 1705 | &((ovs_be32 *)flow)[mf->flow_be32ofs], |
c0bfb650 | 1706 | &plens->be32, mf->n_bits); |
13751fd8 | 1707 | } |
c0bfb650 JR |
1708 | memset(plens, 0xff, sizeof *plens); /* All prefixes, no skipping. */ |
1709 | return 0; /* Value not used in this case. */ | |
13751fd8 JR |
1710 | } |
1711 | ||
1712 | /* Returns the length of a prefix match mask for the field 'mf' in 'minimask'. | |
1713 | * Returns the u32 offset to the miniflow data in '*miniflow_index', if | |
1714 | * 'miniflow_index' is not NULL. */ | |
1715 | static unsigned int | |
1716 | minimask_get_prefix_len(const struct minimask *minimask, | |
1717 | const struct mf_field *mf) | |
1718 | { | |
c30cfa6b | 1719 | unsigned int n_bits = 0, mask_tz = 0; /* Non-zero when end of mask seen. */ |
13751fd8 JR |
1720 | uint8_t u32_ofs = mf->flow_be32ofs; |
1721 | uint8_t u32_end = u32_ofs + mf->n_bytes / 4; | |
1722 | ||
1723 | for (; u32_ofs < u32_end; ++u32_ofs) { | |
1724 | uint32_t mask; | |
1725 | mask = ntohl((OVS_FORCE ovs_be32)minimask_get(minimask, u32_ofs)); | |
1726 | ||
1727 | /* Validate mask, count the mask length. */ | |
1728 | if (mask_tz) { | |
1729 | if (mask) { | |
1730 | return 0; /* No bits allowed after mask ended. */ | |
1731 | } | |
1732 | } else { | |
1733 | if (~mask & (~mask + 1)) { | |
1734 | return 0; /* Mask not contiguous. */ | |
1735 | } | |
1736 | mask_tz = ctz32(mask); | |
c30cfa6b | 1737 | n_bits += 32 - mask_tz; |
13751fd8 JR |
1738 | } |
1739 | } | |
1740 | ||
c30cfa6b | 1741 | return n_bits; |
13751fd8 JR |
1742 | } |
1743 | ||
1744 | /* | |
1745 | * This is called only when mask prefix is known to be CIDR and non-zero. | |
1746 | * Relies on the fact that the flow and mask have the same map, and since | |
1747 | * the mask is CIDR, the storage for the flow field exists even if it | |
1748 | * happened to be zeros. | |
1749 | */ | |
1750 | static const ovs_be32 * | |
1751 | minimatch_get_prefix(const struct minimatch *match, const struct mf_field *mf) | |
1752 | { | |
27bbe15d | 1753 | return miniflow_get_be32_values(&match->flow) + |
13751fd8 JR |
1754 | count_1bits(match->flow.map & ((UINT64_C(1) << mf->flow_be32ofs) - 1)); |
1755 | } | |
1756 | ||
1757 | /* Insert rule in to the prefix tree. | |
1758 | * 'mlen' must be the (non-zero) CIDR prefix length of the 'trie->field' mask | |
1759 | * in 'rule'. */ | |
1760 | static void | |
1761 | trie_insert(struct cls_trie *trie, const struct cls_rule *rule, int mlen) | |
1762 | { | |
69d6040e JR |
1763 | trie_insert_prefix(&trie->root, |
1764 | minimatch_get_prefix(&rule->match, trie->field), mlen); | |
1765 | } | |
1766 | ||
1767 | static void | |
f358a2cb | 1768 | trie_insert_prefix(rcu_trie_ptr *edge, const ovs_be32 *prefix, int mlen) |
69d6040e | 1769 | { |
13751fd8 | 1770 | struct trie_node *node; |
13751fd8 JR |
1771 | int ofs = 0; |
1772 | ||
1773 | /* Walk the tree. */ | |
f358a2cb | 1774 | for (; (node = ovsrcu_get_protected(struct trie_node *, edge)); |
13751fd8 JR |
1775 | edge = trie_next_edge(node, prefix, ofs)) { |
1776 | unsigned int eqbits = trie_prefix_equal_bits(node, prefix, ofs, mlen); | |
1777 | ofs += eqbits; | |
c30cfa6b | 1778 | if (eqbits < node->n_bits) { |
13751fd8 JR |
1779 | /* Mismatch, new node needs to be inserted above. */ |
1780 | int old_branch = get_bit_at(node->prefix, eqbits); | |
f358a2cb | 1781 | struct trie_node *new_parent; |
13751fd8 | 1782 | |
f358a2cb JR |
1783 | new_parent = trie_branch_create(prefix, ofs - eqbits, eqbits, |
1784 | ofs == mlen ? 1 : 0); | |
1785 | /* Copy the node to modify it. */ | |
1786 | node = trie_node_rcu_realloc(node); | |
1787 | /* Adjust the new node for its new position in the tree. */ | |
13751fd8 | 1788 | node->prefix <<= eqbits; |
c30cfa6b | 1789 | node->n_bits -= eqbits; |
f358a2cb | 1790 | ovsrcu_set_hidden(&new_parent->edges[old_branch], node); |
13751fd8 JR |
1791 | |
1792 | /* Check if need a new branch for the new rule. */ | |
1793 | if (ofs < mlen) { | |
f358a2cb JR |
1794 | ovsrcu_set_hidden(&new_parent->edges[!old_branch], |
1795 | trie_branch_create(prefix, ofs, mlen - ofs, | |
1796 | 1)); | |
13751fd8 | 1797 | } |
f358a2cb | 1798 | ovsrcu_set(edge, new_parent); /* Publish changes. */ |
13751fd8 JR |
1799 | return; |
1800 | } | |
1801 | /* Full match so far. */ | |
1802 | ||
1803 | if (ofs == mlen) { | |
1804 | /* Full match at the current node, rule needs to be added here. */ | |
1805 | node->n_rules++; | |
1806 | return; | |
1807 | } | |
1808 | } | |
1809 | /* Must insert a new tree branch for the new rule. */ | |
f358a2cb | 1810 | ovsrcu_set(edge, trie_branch_create(prefix, ofs, mlen - ofs, 1)); |
13751fd8 JR |
1811 | } |
1812 | ||
1813 | /* 'mlen' must be the (non-zero) CIDR prefix length of the 'trie->field' mask | |
1814 | * in 'rule'. */ | |
1815 | static void | |
1816 | trie_remove(struct cls_trie *trie, const struct cls_rule *rule, int mlen) | |
1817 | { | |
69d6040e JR |
1818 | trie_remove_prefix(&trie->root, |
1819 | minimatch_get_prefix(&rule->match, trie->field), mlen); | |
1820 | } | |
1821 | ||
1822 | /* 'mlen' must be the (non-zero) CIDR prefix length of the 'trie->field' mask | |
1823 | * in 'rule'. */ | |
1824 | static void | |
f358a2cb | 1825 | trie_remove_prefix(rcu_trie_ptr *root, const ovs_be32 *prefix, int mlen) |
69d6040e | 1826 | { |
13751fd8 | 1827 | struct trie_node *node; |
f358a2cb | 1828 | rcu_trie_ptr *edges[sizeof(union mf_value) * 8]; |
13751fd8 JR |
1829 | int depth = 0, ofs = 0; |
1830 | ||
1831 | /* Walk the tree. */ | |
69d6040e | 1832 | for (edges[0] = root; |
f358a2cb | 1833 | (node = ovsrcu_get_protected(struct trie_node *, edges[depth])); |
13751fd8 JR |
1834 | edges[++depth] = trie_next_edge(node, prefix, ofs)) { |
1835 | unsigned int eqbits = trie_prefix_equal_bits(node, prefix, ofs, mlen); | |
69d6040e | 1836 | |
c30cfa6b | 1837 | if (eqbits < node->n_bits) { |
13751fd8 JR |
1838 | /* Mismatch, nothing to be removed. This should never happen, as |
1839 | * only rules in the classifier are ever removed. */ | |
1840 | break; /* Log a warning. */ | |
1841 | } | |
1842 | /* Full match so far. */ | |
1843 | ofs += eqbits; | |
1844 | ||
1845 | if (ofs == mlen) { | |
1846 | /* Full prefix match at the current node, remove rule here. */ | |
1847 | if (!node->n_rules) { | |
1848 | break; /* Log a warning. */ | |
1849 | } | |
1850 | node->n_rules--; | |
1851 | ||
1852 | /* Check if can prune the tree. */ | |
f358a2cb JR |
1853 | while (!node->n_rules) { |
1854 | struct trie_node *next, | |
1855 | *edge0 = ovsrcu_get_protected(struct trie_node *, | |
1856 | &node->edges[0]), | |
1857 | *edge1 = ovsrcu_get_protected(struct trie_node *, | |
1858 | &node->edges[1]); | |
1859 | ||
1860 | if (edge0 && edge1) { | |
1861 | break; /* A branching point, cannot prune. */ | |
1862 | } | |
1863 | ||
1864 | /* Else have at most one child node, remove this node. */ | |
1865 | next = edge0 ? edge0 : edge1; | |
13751fd8 JR |
1866 | |
1867 | if (next) { | |
c30cfa6b | 1868 | if (node->n_bits + next->n_bits > TRIE_PREFIX_BITS) { |
13751fd8 JR |
1869 | break; /* Cannot combine. */ |
1870 | } | |
f358a2cb JR |
1871 | next = trie_node_rcu_realloc(next); /* Modify. */ |
1872 | ||
13751fd8 | 1873 | /* Combine node with next. */ |
c30cfa6b JR |
1874 | next->prefix = node->prefix | next->prefix >> node->n_bits; |
1875 | next->n_bits += node->n_bits; | |
13751fd8 | 1876 | } |
13751fd8 | 1877 | /* Update the parent's edge. */ |
f358a2cb JR |
1878 | ovsrcu_set(edges[depth], next); /* Publish changes. */ |
1879 | trie_node_destroy(node); | |
1880 | ||
13751fd8 JR |
1881 | if (next || !depth) { |
1882 | /* Branch not pruned or at root, nothing more to do. */ | |
1883 | break; | |
1884 | } | |
f358a2cb JR |
1885 | node = ovsrcu_get_protected(struct trie_node *, |
1886 | edges[--depth]); | |
13751fd8 JR |
1887 | } |
1888 | return; | |
1889 | } | |
1890 | } | |
1891 | /* Cannot go deeper. This should never happen, since only rules | |
1892 | * that actually exist in the classifier are ever removed. */ | |
1893 | VLOG_WARN("Trying to remove non-existing rule from a prefix trie."); | |
1894 | } |