]>
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); |
e48eccd1 | 63 | static struct cls_match *insert_rule(struct classifier *cls, |
e65413ab JR |
64 | struct cls_subtable *, struct cls_rule *) |
65 | OVS_REQUIRES(cls->mutex); | |
b5d97350 | 66 | |
dfea28b3 JR |
67 | static const struct cls_match *find_match_wc(const struct cls_subtable *, |
68 | const struct flow *, | |
69 | struct trie_ctx *, | |
70 | unsigned int n_tries, | |
71 | struct flow_wildcards *); | |
72 | static struct cls_match *find_equal(const struct cls_subtable *, | |
627fb667 | 73 | const struct miniflow *, uint32_t hash); |
b5d97350 | 74 | |
dfea28b3 JR |
75 | static inline const struct cls_match * |
76 | next_rule_in_list__(const struct cls_match *rule) | |
77 | { | |
78 | const struct cls_match *next = NULL; | |
79 | next = OBJECT_CONTAINING(rculist_next(&rule->list), next, list); | |
80 | return next; | |
81 | } | |
82 | ||
83 | static inline const struct cls_match * | |
84 | next_rule_in_list(const struct cls_match *rule) | |
85 | { | |
86 | const struct cls_match *next = next_rule_in_list__(rule); | |
87 | return next->priority < rule->priority ? next : NULL; | |
88 | } | |
89 | ||
c501b427 | 90 | static inline struct cls_match * |
dfea28b3 | 91 | next_rule_in_list_protected__(struct cls_match *rule) |
c501b427 JR |
92 | { |
93 | struct cls_match *next = NULL; | |
dfea28b3 | 94 | next = OBJECT_CONTAINING(rculist_next_protected(&rule->list), next, list); |
c501b427 JR |
95 | return next; |
96 | } | |
97 | ||
98 | static inline struct cls_match * | |
dfea28b3 | 99 | next_rule_in_list_protected(struct cls_match *rule) |
c501b427 | 100 | { |
dfea28b3 | 101 | struct cls_match *next = next_rule_in_list_protected__(rule); |
c501b427 JR |
102 | return next->priority < rule->priority ? next : NULL; |
103 | } | |
104 | ||
e65413ab JR |
105 | /* Iterates RULE over HEAD and all of the cls_rules on HEAD->list. |
106 | * Classifier's mutex must be held while iterating, as the list is | |
107 | * protoceted by it. */ | |
b5d97350 BP |
108 | #define FOR_EACH_RULE_IN_LIST(RULE, HEAD) \ |
109 | for ((RULE) = (HEAD); (RULE) != NULL; (RULE) = next_rule_in_list(RULE)) | |
dfea28b3 JR |
110 | #define FOR_EACH_RULE_IN_LIST_PROTECTED(RULE, HEAD) \ |
111 | for ((RULE) = (HEAD); (RULE) != NULL; \ | |
112 | (RULE) = next_rule_in_list_protected(RULE)) | |
13751fd8 JR |
113 | |
114 | static unsigned int minimask_get_prefix_len(const struct minimask *, | |
115 | const struct mf_field *); | |
e48eccd1 | 116 | static void trie_init(struct classifier *cls, int trie_idx, |
e65413ab JR |
117 | const struct mf_field *) |
118 | OVS_REQUIRES(cls->mutex); | |
13751fd8 | 119 | static unsigned int trie_lookup(const struct cls_trie *, const struct flow *, |
c0bfb650 | 120 | union mf_value *plens); |
f358a2cb | 121 | static unsigned int trie_lookup_value(const rcu_trie_ptr *, |
c0bfb650 JR |
122 | const ovs_be32 value[], ovs_be32 plens[], |
123 | unsigned int value_bits); | |
f358a2cb | 124 | static void trie_destroy(rcu_trie_ptr *); |
13751fd8 | 125 | static void trie_insert(struct cls_trie *, const struct cls_rule *, int mlen); |
f358a2cb | 126 | static void trie_insert_prefix(rcu_trie_ptr *, const ovs_be32 *prefix, |
69d6040e | 127 | int mlen); |
13751fd8 | 128 | static void trie_remove(struct cls_trie *, const struct cls_rule *, int mlen); |
f358a2cb | 129 | static void trie_remove_prefix(rcu_trie_ptr *, const ovs_be32 *prefix, |
69d6040e | 130 | int mlen); |
13751fd8 | 131 | static void mask_set_prefix_bits(struct flow_wildcards *, uint8_t be32ofs, |
c30cfa6b | 132 | unsigned int n_bits); |
13751fd8 | 133 | static bool mask_prefix_bits_set(const struct flow_wildcards *, |
c30cfa6b | 134 | uint8_t be32ofs, unsigned int n_bits); |
81a76618 BP |
135 | \f |
136 | /* cls_rule. */ | |
b5d97350 | 137 | |
81a76618 | 138 | /* Initializes 'rule' to match packets specified by 'match' at the given |
5cb7a798 BP |
139 | * 'priority'. 'match' must satisfy the invariant described in the comment at |
140 | * the definition of struct match. | |
66642cb4 | 141 | * |
48d28ac1 BP |
142 | * The caller must eventually destroy 'rule' with cls_rule_destroy(). |
143 | * | |
eb391b76 BP |
144 | * Clients should not use priority INT_MIN. (OpenFlow uses priorities between |
145 | * 0 and UINT16_MAX, inclusive.) */ | |
47284b1f | 146 | void |
eb391b76 | 147 | cls_rule_init(struct cls_rule *rule, const struct match *match, int priority) |
47284b1f | 148 | { |
5cb7a798 BP |
149 | minimatch_init(&rule->match, match); |
150 | rule->priority = priority; | |
627fb667 | 151 | rule->cls_match = NULL; |
5cb7a798 BP |
152 | } |
153 | ||
154 | /* Same as cls_rule_init() for initialization from a "struct minimatch". */ | |
155 | void | |
156 | cls_rule_init_from_minimatch(struct cls_rule *rule, | |
eb391b76 | 157 | const struct minimatch *match, int priority) |
5cb7a798 BP |
158 | { |
159 | minimatch_clone(&rule->match, match); | |
81a76618 | 160 | rule->priority = priority; |
627fb667 | 161 | rule->cls_match = NULL; |
685a51a5 JP |
162 | } |
163 | ||
48d28ac1 BP |
164 | /* Initializes 'dst' as a copy of 'src'. |
165 | * | |
b2c1f00b | 166 | * The caller must eventually destroy 'dst' with cls_rule_destroy(). */ |
48d28ac1 BP |
167 | void |
168 | cls_rule_clone(struct cls_rule *dst, const struct cls_rule *src) | |
169 | { | |
5cb7a798 BP |
170 | minimatch_clone(&dst->match, &src->match); |
171 | dst->priority = src->priority; | |
627fb667 | 172 | dst->cls_match = NULL; |
48d28ac1 BP |
173 | } |
174 | ||
b2c1f00b BP |
175 | /* Initializes 'dst' with the data in 'src', destroying 'src'. |
176 | * | |
177 | * The caller must eventually destroy 'dst' with cls_rule_destroy(). */ | |
178 | void | |
179 | cls_rule_move(struct cls_rule *dst, struct cls_rule *src) | |
180 | { | |
181 | minimatch_move(&dst->match, &src->match); | |
182 | dst->priority = src->priority; | |
627fb667 | 183 | dst->cls_match = NULL; |
b2c1f00b BP |
184 | } |
185 | ||
48d28ac1 BP |
186 | /* Frees memory referenced by 'rule'. Doesn't free 'rule' itself (it's |
187 | * normally embedded into a larger structure). | |
188 | * | |
189 | * ('rule' must not currently be in a classifier.) */ | |
190 | void | |
5cb7a798 | 191 | cls_rule_destroy(struct cls_rule *rule) |
48d28ac1 | 192 | { |
627fb667 | 193 | ovs_assert(!rule->cls_match); |
5cb7a798 | 194 | minimatch_destroy(&rule->match); |
48d28ac1 BP |
195 | } |
196 | ||
81a76618 BP |
197 | /* Returns true if 'a' and 'b' match the same packets at the same priority, |
198 | * false if they differ in some way. */ | |
193eb874 BP |
199 | bool |
200 | cls_rule_equal(const struct cls_rule *a, const struct cls_rule *b) | |
201 | { | |
5cb7a798 | 202 | return a->priority == b->priority && minimatch_equal(&a->match, &b->match); |
193eb874 BP |
203 | } |
204 | ||
81a76618 | 205 | /* Returns a hash value for 'rule', folding in 'basis'. */ |
57452fdc BP |
206 | uint32_t |
207 | cls_rule_hash(const struct cls_rule *rule, uint32_t basis) | |
208 | { | |
5cb7a798 | 209 | return minimatch_hash(&rule->match, hash_int(rule->priority, basis)); |
73f33563 BP |
210 | } |
211 | ||
81a76618 | 212 | /* Appends a string describing 'rule' to 's'. */ |
07b37e8f BP |
213 | void |
214 | cls_rule_format(const struct cls_rule *rule, struct ds *s) | |
215 | { | |
5cb7a798 | 216 | minimatch_format(&rule->match, s, rule->priority); |
064af421 | 217 | } |
3ca1de08 BP |
218 | |
219 | /* Returns true if 'rule' matches every packet, false otherwise. */ | |
220 | bool | |
221 | cls_rule_is_catchall(const struct cls_rule *rule) | |
222 | { | |
5cb7a798 | 223 | return minimask_is_catchall(&rule->match.mask); |
3ca1de08 | 224 | } |
064af421 BP |
225 | \f |
226 | /* Initializes 'cls' as a classifier that initially contains no classification | |
227 | * rules. */ | |
228 | void | |
e48eccd1 JR |
229 | classifier_init(struct classifier *cls, const uint8_t *flow_segments) |
230 | OVS_EXCLUDED(cls->mutex) | |
064af421 | 231 | { |
e65413ab | 232 | ovs_mutex_init(&cls->mutex); |
e65413ab | 233 | ovs_mutex_lock(&cls->mutex); |
064af421 | 234 | cls->n_rules = 0; |
f2c21402 | 235 | cmap_init(&cls->subtables_map); |
fe7cfa5c | 236 | pvector_init(&cls->subtables); |
f2c21402 | 237 | cmap_init(&cls->partitions); |
476f36e8 JR |
238 | cls->n_flow_segments = 0; |
239 | if (flow_segments) { | |
240 | while (cls->n_flow_segments < CLS_MAX_INDICES | |
241 | && *flow_segments < FLOW_U32S) { | |
242 | cls->flow_segments[cls->n_flow_segments++] = *flow_segments++; | |
243 | } | |
244 | } | |
13751fd8 | 245 | cls->n_tries = 0; |
e65413ab JR |
246 | for (int i = 0; i < CLS_MAX_TRIES; i++) { |
247 | trie_init(cls, i, NULL); | |
248 | } | |
249 | ovs_mutex_unlock(&cls->mutex); | |
064af421 BP |
250 | } |
251 | ||
252 | /* Destroys 'cls'. Rules within 'cls', if any, are not freed; this is the | |
afae68b1 JR |
253 | * caller's responsibility. |
254 | * May only be called after all the readers have been terminated. */ | |
064af421 | 255 | void |
e48eccd1 JR |
256 | classifier_destroy(struct classifier *cls) |
257 | OVS_EXCLUDED(cls->mutex) | |
064af421 | 258 | { |
e48eccd1 | 259 | if (cls) { |
78c8df12 BP |
260 | struct cls_partition *partition; |
261 | struct cls_subtable *subtable; | |
13751fd8 JR |
262 | int i; |
263 | ||
e65413ab | 264 | ovs_mutex_lock(&cls->mutex); |
13751fd8 | 265 | for (i = 0; i < cls->n_tries; i++) { |
f358a2cb | 266 | trie_destroy(&cls->tries[i].root); |
13751fd8 | 267 | } |
064af421 | 268 | |
6bc3bb82 | 269 | CMAP_FOR_EACH (subtable, cmap_node, &cls->subtables_map) { |
03868246 | 270 | destroy_subtable(cls, subtable); |
064af421 | 271 | } |
f2c21402 | 272 | cmap_destroy(&cls->subtables_map); |
c906cedf | 273 | |
6bc3bb82 | 274 | CMAP_FOR_EACH (partition, cmap_node, &cls->partitions) { |
f2c21402 | 275 | ovsrcu_postpone(free, partition); |
c906cedf | 276 | } |
f2c21402 | 277 | cmap_destroy(&cls->partitions); |
cabd4c43 | 278 | |
fe7cfa5c | 279 | pvector_destroy(&cls->subtables); |
e65413ab JR |
280 | ovs_mutex_unlock(&cls->mutex); |
281 | ovs_mutex_destroy(&cls->mutex); | |
064af421 BP |
282 | } |
283 | } | |
284 | ||
13751fd8 | 285 | /* Set the fields for which prefix lookup should be performed. */ |
f358a2cb | 286 | bool |
e48eccd1 | 287 | classifier_set_prefix_fields(struct classifier *cls, |
13751fd8 JR |
288 | const enum mf_field_id *trie_fields, |
289 | unsigned int n_fields) | |
e48eccd1 | 290 | OVS_EXCLUDED(cls->mutex) |
13751fd8 | 291 | { |
f358a2cb | 292 | const struct mf_field * new_fields[CLS_MAX_TRIES]; |
abadfcb0 | 293 | struct mf_bitmap fields = MF_BITMAP_INITIALIZER; |
f358a2cb JR |
294 | int i, n_tries = 0; |
295 | bool changed = false; | |
13751fd8 | 296 | |
e65413ab | 297 | ovs_mutex_lock(&cls->mutex); |
f358a2cb | 298 | for (i = 0; i < n_fields && n_tries < CLS_MAX_TRIES; i++) { |
13751fd8 JR |
299 | const struct mf_field *field = mf_from_id(trie_fields[i]); |
300 | if (field->flow_be32ofs < 0 || field->n_bits % 32) { | |
301 | /* Incompatible field. This is the only place where we | |
302 | * enforce these requirements, but the rest of the trie code | |
303 | * depends on the flow_be32ofs to be non-negative and the | |
304 | * field length to be a multiple of 32 bits. */ | |
305 | continue; | |
306 | } | |
307 | ||
abadfcb0 | 308 | if (bitmap_is_set(fields.bm, trie_fields[i])) { |
13751fd8 JR |
309 | /* Duplicate field, there is no need to build more than |
310 | * one index for any one field. */ | |
311 | continue; | |
312 | } | |
abadfcb0 | 313 | bitmap_set1(fields.bm, trie_fields[i]); |
13751fd8 | 314 | |
f358a2cb JR |
315 | new_fields[n_tries] = NULL; |
316 | if (n_tries >= cls->n_tries || field != cls->tries[n_tries].field) { | |
317 | new_fields[n_tries] = field; | |
318 | changed = true; | |
319 | } | |
320 | n_tries++; | |
321 | } | |
322 | ||
323 | if (changed || n_tries < cls->n_tries) { | |
324 | struct cls_subtable *subtable; | |
325 | ||
326 | /* Trie configuration needs to change. Disable trie lookups | |
327 | * for the tries that are changing and wait all the current readers | |
328 | * with the old configuration to be done. */ | |
329 | changed = false; | |
330 | CMAP_FOR_EACH (subtable, cmap_node, &cls->subtables_map) { | |
331 | for (i = 0; i < cls->n_tries; i++) { | |
332 | if ((i < n_tries && new_fields[i]) || i >= n_tries) { | |
333 | if (subtable->trie_plen[i]) { | |
334 | subtable->trie_plen[i] = 0; | |
335 | changed = true; | |
336 | } | |
337 | } | |
338 | } | |
339 | } | |
340 | /* Synchronize if any readers were using tries. The readers may | |
341 | * temporarily function without the trie lookup based optimizations. */ | |
342 | if (changed) { | |
343 | /* ovsrcu_synchronize() functions as a memory barrier, so it does | |
344 | * not matter that subtable->trie_plen is not atomic. */ | |
345 | ovsrcu_synchronize(); | |
13751fd8 | 346 | } |
13751fd8 | 347 | |
f358a2cb JR |
348 | /* Now set up the tries. */ |
349 | for (i = 0; i < n_tries; i++) { | |
350 | if (new_fields[i]) { | |
351 | trie_init(cls, i, new_fields[i]); | |
352 | } | |
353 | } | |
354 | /* Destroy the rest, if any. */ | |
355 | for (; i < cls->n_tries; i++) { | |
356 | trie_init(cls, i, NULL); | |
357 | } | |
358 | ||
359 | cls->n_tries = n_tries; | |
360 | ovs_mutex_unlock(&cls->mutex); | |
361 | return true; | |
13751fd8 | 362 | } |
f358a2cb | 363 | |
e65413ab | 364 | ovs_mutex_unlock(&cls->mutex); |
f358a2cb | 365 | return false; /* No change. */ |
13751fd8 JR |
366 | } |
367 | ||
368 | static void | |
e48eccd1 | 369 | trie_init(struct classifier *cls, int trie_idx, const struct mf_field *field) |
e65413ab | 370 | OVS_REQUIRES(cls->mutex) |
13751fd8 JR |
371 | { |
372 | struct cls_trie *trie = &cls->tries[trie_idx]; | |
373 | struct cls_subtable *subtable; | |
374 | ||
375 | if (trie_idx < cls->n_tries) { | |
f358a2cb JR |
376 | trie_destroy(&trie->root); |
377 | } else { | |
378 | ovsrcu_set_hidden(&trie->root, NULL); | |
13751fd8 | 379 | } |
13751fd8 JR |
380 | trie->field = field; |
381 | ||
f358a2cb | 382 | /* Add existing rules to the new trie. */ |
f2c21402 | 383 | CMAP_FOR_EACH (subtable, cmap_node, &cls->subtables_map) { |
13751fd8 JR |
384 | unsigned int plen; |
385 | ||
386 | plen = field ? minimask_get_prefix_len(&subtable->mask, field) : 0; | |
13751fd8 | 387 | if (plen) { |
627fb667 | 388 | struct cls_match *head; |
13751fd8 | 389 | |
f2c21402 | 390 | CMAP_FOR_EACH (head, cmap_node, &subtable->rules) { |
627fb667 | 391 | struct cls_match *match; |
13751fd8 | 392 | |
dfea28b3 | 393 | FOR_EACH_RULE_IN_LIST_PROTECTED (match, head) { |
627fb667 | 394 | trie_insert(trie, match->cls_rule, plen); |
13751fd8 JR |
395 | } |
396 | } | |
397 | } | |
f358a2cb JR |
398 | /* Initialize subtable's prefix length on this field. This will |
399 | * allow readers to use the trie. */ | |
400 | atomic_thread_fence(memory_order_release); | |
401 | subtable->trie_plen[trie_idx] = plen; | |
13751fd8 JR |
402 | } |
403 | } | |
404 | ||
5f0476ce JR |
405 | /* Returns true if 'cls' contains no classification rules, false otherwise. |
406 | * Checking the cmap requires no locking. */ | |
064af421 BP |
407 | bool |
408 | classifier_is_empty(const struct classifier *cls) | |
409 | { | |
e48eccd1 | 410 | return cmap_is_empty(&cls->subtables_map); |
064af421 BP |
411 | } |
412 | ||
dbda2960 | 413 | /* Returns the number of rules in 'cls'. */ |
064af421 BP |
414 | int |
415 | classifier_count(const struct classifier *cls) | |
afae68b1 | 416 | OVS_NO_THREAD_SAFETY_ANALYSIS |
064af421 | 417 | { |
afae68b1 JR |
418 | /* n_rules is an int, so in the presence of concurrent writers this will |
419 | * return either the old or a new value. */ | |
e48eccd1 | 420 | return cls->n_rules; |
064af421 BP |
421 | } |
422 | ||
c906cedf BP |
423 | static uint32_t |
424 | hash_metadata(ovs_be64 metadata_) | |
425 | { | |
426 | uint64_t metadata = (OVS_FORCE uint64_t) metadata_; | |
965607c8 | 427 | return hash_uint64(metadata); |
c906cedf BP |
428 | } |
429 | ||
430 | static struct cls_partition * | |
e48eccd1 | 431 | find_partition(const struct classifier *cls, ovs_be64 metadata, uint32_t hash) |
c906cedf BP |
432 | { |
433 | struct cls_partition *partition; | |
434 | ||
f2c21402 | 435 | CMAP_FOR_EACH_WITH_HASH (partition, cmap_node, hash, &cls->partitions) { |
c906cedf BP |
436 | if (partition->metadata == metadata) { |
437 | return partition; | |
438 | } | |
439 | } | |
440 | ||
441 | return NULL; | |
442 | } | |
443 | ||
444 | static struct cls_partition * | |
e48eccd1 | 445 | create_partition(struct classifier *cls, struct cls_subtable *subtable, |
c906cedf | 446 | ovs_be64 metadata) |
e65413ab | 447 | OVS_REQUIRES(cls->mutex) |
c906cedf BP |
448 | { |
449 | uint32_t hash = hash_metadata(metadata); | |
450 | struct cls_partition *partition = find_partition(cls, metadata, hash); | |
451 | if (!partition) { | |
452 | partition = xmalloc(sizeof *partition); | |
453 | partition->metadata = metadata; | |
454 | partition->tags = 0; | |
183126a1 | 455 | tag_tracker_init(&partition->tracker); |
f2c21402 | 456 | cmap_insert(&cls->partitions, &partition->cmap_node, hash); |
c906cedf | 457 | } |
03868246 | 458 | tag_tracker_add(&partition->tracker, &partition->tags, subtable->tag); |
c906cedf BP |
459 | return partition; |
460 | } | |
461 | ||
69d6040e JR |
462 | static inline ovs_be32 minimatch_get_ports(const struct minimatch *match) |
463 | { | |
464 | /* Could optimize to use the same map if needed for fast path. */ | |
465 | return MINIFLOW_GET_BE32(&match->flow, tp_src) | |
466 | & MINIFLOW_GET_BE32(&match->mask.masks, tp_src); | |
467 | } | |
468 | ||
b5d97350 BP |
469 | /* Inserts 'rule' into 'cls'. Until 'rule' is removed from 'cls', the caller |
470 | * must not modify or free it. | |
064af421 BP |
471 | * |
472 | * If 'cls' already contains an identical rule (including wildcards, values of | |
473 | * fixed fields, and priority), replaces the old rule by 'rule' and returns the | |
474 | * rule that was replaced. The caller takes ownership of the returned rule and | |
48d28ac1 BP |
475 | * is thus responsible for destroying it with cls_rule_destroy(), freeing the |
476 | * memory block in which it resides, etc., as necessary. | |
064af421 BP |
477 | * |
478 | * Returns NULL if 'cls' does not contain a rule with an identical key, after | |
479 | * inserting the new rule. In this case, no rules are displaced by the new | |
480 | * rule, even rules that cannot have any effect because the new rule matches a | |
481 | * superset of their flows and has higher priority. */ | |
dfea28b3 | 482 | const struct cls_rule * |
e48eccd1 JR |
483 | classifier_replace(struct classifier *cls, struct cls_rule *rule) |
484 | OVS_EXCLUDED(cls->mutex) | |
064af421 | 485 | { |
627fb667 | 486 | struct cls_match *old_rule; |
03868246 | 487 | struct cls_subtable *subtable; |
dfea28b3 | 488 | const struct cls_rule *old_cls_rule = NULL; |
b5d97350 | 489 | |
e65413ab | 490 | ovs_mutex_lock(&cls->mutex); |
03868246 JR |
491 | subtable = find_subtable(cls, &rule->match.mask); |
492 | if (!subtable) { | |
493 | subtable = insert_subtable(cls, &rule->match.mask); | |
b5d97350 BP |
494 | } |
495 | ||
03868246 | 496 | old_rule = insert_rule(cls, subtable, rule); |
b5d97350 | 497 | if (!old_rule) { |
e65413ab | 498 | old_cls_rule = NULL; |
13751fd8 | 499 | |
627fb667 | 500 | rule->cls_match->partition = NULL; |
c906cedf BP |
501 | if (minimask_get_metadata_mask(&rule->match.mask) == OVS_BE64_MAX) { |
502 | ovs_be64 metadata = miniflow_get_metadata(&rule->match.flow); | |
627fb667 JR |
503 | rule->cls_match->partition = create_partition(cls, subtable, |
504 | metadata); | |
c906cedf BP |
505 | } |
506 | ||
064af421 | 507 | cls->n_rules++; |
13751fd8 | 508 | |
e65413ab | 509 | for (int i = 0; i < cls->n_tries; i++) { |
13751fd8 JR |
510 | if (subtable->trie_plen[i]) { |
511 | trie_insert(&cls->tries[i], rule, subtable->trie_plen[i]); | |
512 | } | |
513 | } | |
69d6040e JR |
514 | |
515 | /* Ports trie. */ | |
516 | if (subtable->ports_mask_len) { | |
517 | /* We mask the value to be inserted to always have the wildcarded | |
518 | * bits in known (zero) state, so we can include them in comparison | |
519 | * and they will always match (== their original value does not | |
520 | * matter). */ | |
521 | ovs_be32 masked_ports = minimatch_get_ports(&rule->match); | |
522 | ||
523 | trie_insert_prefix(&subtable->ports_trie, &masked_ports, | |
524 | subtable->ports_mask_len); | |
525 | } | |
c906cedf | 526 | } else { |
e65413ab | 527 | old_cls_rule = old_rule->cls_rule; |
627fb667 | 528 | rule->cls_match->partition = old_rule->partition; |
dfea28b3 | 529 | CONST_CAST(struct cls_rule *, old_cls_rule)->cls_match = NULL; |
f2c21402 JR |
530 | |
531 | /* 'old_rule' contains a cmap_node, which may not be freed | |
532 | * immediately. */ | |
533 | ovsrcu_postpone(free, old_rule); | |
064af421 | 534 | } |
e65413ab JR |
535 | ovs_mutex_unlock(&cls->mutex); |
536 | return old_cls_rule; | |
064af421 BP |
537 | } |
538 | ||
08944c1d BP |
539 | /* Inserts 'rule' into 'cls'. Until 'rule' is removed from 'cls', the caller |
540 | * must not modify or free it. | |
541 | * | |
542 | * 'cls' must not contain an identical rule (including wildcards, values of | |
543 | * fixed fields, and priority). Use classifier_find_rule_exactly() to find | |
544 | * such a rule. */ | |
545 | void | |
546 | classifier_insert(struct classifier *cls, struct cls_rule *rule) | |
547 | { | |
dfea28b3 | 548 | const struct cls_rule *displaced_rule = classifier_replace(cls, rule); |
cb22974d | 549 | ovs_assert(!displaced_rule); |
08944c1d BP |
550 | } |
551 | ||
48d28ac1 BP |
552 | /* Removes 'rule' from 'cls'. It is the caller's responsibility to destroy |
553 | * 'rule' with cls_rule_destroy(), freeing the memory block in which 'rule' | |
747f140a JR |
554 | * resides, etc., as necessary. |
555 | * | |
556 | * Does nothing if 'rule' has been already removed, or was never inserted. | |
557 | * | |
558 | * Returns the removed rule, or NULL, if it was already removed. | |
559 | */ | |
dfea28b3 JR |
560 | const struct cls_rule * |
561 | classifier_remove(struct classifier *cls, const struct cls_rule *rule) | |
e48eccd1 | 562 | OVS_EXCLUDED(cls->mutex) |
064af421 | 563 | { |
c906cedf | 564 | struct cls_partition *partition; |
747f140a | 565 | struct cls_match *cls_match; |
627fb667 | 566 | struct cls_match *head; |
03868246 | 567 | struct cls_subtable *subtable; |
476f36e8 | 568 | int i; |
f2c21402 JR |
569 | uint32_t basis = 0, hash, ihash[CLS_MAX_INDICES]; |
570 | uint8_t prev_be32ofs = 0; | |
064af421 | 571 | |
e65413ab | 572 | ovs_mutex_lock(&cls->mutex); |
747f140a JR |
573 | cls_match = rule->cls_match; |
574 | if (!cls_match) { | |
575 | rule = NULL; | |
576 | goto unlock; /* Already removed. */ | |
577 | } | |
578 | ||
03868246 | 579 | subtable = find_subtable(cls, &rule->match.mask); |
627fb667 JR |
580 | ovs_assert(subtable); |
581 | ||
69d6040e JR |
582 | if (subtable->ports_mask_len) { |
583 | ovs_be32 masked_ports = minimatch_get_ports(&rule->match); | |
584 | ||
585 | trie_remove_prefix(&subtable->ports_trie, | |
586 | &masked_ports, subtable->ports_mask_len); | |
587 | } | |
13751fd8 JR |
588 | for (i = 0; i < cls->n_tries; i++) { |
589 | if (subtable->trie_plen[i]) { | |
590 | trie_remove(&cls->tries[i], rule, subtable->trie_plen[i]); | |
591 | } | |
592 | } | |
593 | ||
476f36e8 JR |
594 | /* Remove rule node from indices. */ |
595 | for (i = 0; i < subtable->n_indices; i++) { | |
f2c21402 JR |
596 | ihash[i] = minimatch_hash_range(&rule->match, prev_be32ofs, |
597 | subtable->index_ofs[i], &basis); | |
598 | cmap_remove(&subtable->indices[i], &cls_match->index_nodes[i], | |
599 | ihash[i]); | |
600 | prev_be32ofs = subtable->index_ofs[i]; | |
476f36e8 | 601 | } |
f2c21402 | 602 | hash = minimatch_hash_range(&rule->match, prev_be32ofs, FLOW_U32S, &basis); |
476f36e8 | 603 | |
f2c21402 | 604 | head = find_equal(subtable, &rule->match.flow, hash); |
627fb667 | 605 | if (head != cls_match) { |
c501b427 JR |
606 | rculist_remove(&cls_match->list); |
607 | } else if (rculist_is_empty(&cls_match->list)) { | |
f2c21402 | 608 | cmap_remove(&subtable->rules, &cls_match->cmap_node, hash); |
b5d97350 | 609 | } else { |
dfea28b3 | 610 | struct cls_match *next = next_rule_in_list_protected(cls_match); |
064af421 | 611 | |
c501b427 | 612 | rculist_remove(&cls_match->list); |
f2c21402 JR |
613 | cmap_replace(&subtable->rules, &cls_match->cmap_node, |
614 | &next->cmap_node, hash); | |
b5d97350 | 615 | } |
064af421 | 616 | |
627fb667 | 617 | partition = cls_match->partition; |
183126a1 BP |
618 | if (partition) { |
619 | tag_tracker_subtract(&partition->tracker, &partition->tags, | |
03868246 | 620 | subtable->tag); |
183126a1 | 621 | if (!partition->tags) { |
f2c21402 JR |
622 | cmap_remove(&cls->partitions, &partition->cmap_node, |
623 | hash_metadata(partition->metadata)); | |
624 | ovsrcu_postpone(free, partition); | |
183126a1 | 625 | } |
c906cedf BP |
626 | } |
627 | ||
03868246 JR |
628 | if (--subtable->n_rules == 0) { |
629 | destroy_subtable(cls, subtable); | |
fe7cfa5c JR |
630 | } else if (subtable->max_priority == cls_match->priority |
631 | && --subtable->max_count == 0) { | |
632 | /* Find the new 'max_priority' and 'max_count'. */ | |
633 | struct cls_match *head; | |
eb391b76 | 634 | int max_priority = INT_MIN; |
fe7cfa5c | 635 | |
f2c21402 | 636 | CMAP_FOR_EACH (head, cmap_node, &subtable->rules) { |
fe7cfa5c JR |
637 | if (head->priority > max_priority) { |
638 | max_priority = head->priority; | |
639 | subtable->max_count = 1; | |
640 | } else if (head->priority == max_priority) { | |
641 | ++subtable->max_count; | |
642 | } | |
643 | } | |
644 | subtable->max_priority = max_priority; | |
645 | pvector_change_priority(&cls->subtables, subtable, max_priority); | |
4d935a6b | 646 | } |
13751fd8 | 647 | |
b5d97350 | 648 | cls->n_rules--; |
627fb667 | 649 | |
f2c21402 | 650 | ovsrcu_postpone(free, cls_match); |
dfea28b3 | 651 | CONST_CAST(struct cls_rule *, rule)->cls_match = NULL; |
747f140a | 652 | unlock: |
e65413ab | 653 | ovs_mutex_unlock(&cls->mutex); |
747f140a JR |
654 | |
655 | return rule; | |
064af421 BP |
656 | } |
657 | ||
13751fd8 | 658 | /* Prefix tree context. Valid when 'lookup_done' is true. Can skip all |
c0bfb650 JR |
659 | * subtables which have a prefix match on the trie field, but whose prefix |
660 | * length is not indicated in 'match_plens'. For example, a subtable that | |
661 | * has a 8-bit trie field prefix match can be skipped if | |
662 | * !be_get_bit_at(&match_plens, 8 - 1). If skipped, 'maskbits' prefix bits | |
663 | * must be unwildcarded to make datapath flow only match packets it should. */ | |
13751fd8 JR |
664 | struct trie_ctx { |
665 | const struct cls_trie *trie; | |
666 | bool lookup_done; /* Status of the lookup. */ | |
667 | uint8_t be32ofs; /* U32 offset of the field in question. */ | |
13751fd8 | 668 | unsigned int maskbits; /* Prefix length needed to avoid false matches. */ |
c0bfb650 JR |
669 | union mf_value match_plens; /* Bitmask of prefix lengths with possible |
670 | * matches. */ | |
13751fd8 JR |
671 | }; |
672 | ||
673 | static void | |
674 | trie_ctx_init(struct trie_ctx *ctx, const struct cls_trie *trie) | |
675 | { | |
676 | ctx->trie = trie; | |
677 | ctx->be32ofs = trie->field->flow_be32ofs; | |
678 | ctx->lookup_done = false; | |
679 | } | |
680 | ||
48c3de13 BP |
681 | /* Finds and returns the highest-priority rule in 'cls' that matches 'flow'. |
682 | * Returns a null pointer if no rules in 'cls' match 'flow'. If multiple rules | |
74f74083 EJ |
683 | * of equal priority match 'flow', returns one arbitrarily. |
684 | * | |
685 | * If a rule is found and 'wc' is non-null, bitwise-OR's 'wc' with the | |
686 | * set of bits that were significant in the lookup. At some point | |
687 | * earlier, 'wc' should have been initialized (e.g., by | |
688 | * flow_wildcards_init_catchall()). */ | |
dfea28b3 | 689 | const struct cls_rule * |
e48eccd1 | 690 | classifier_lookup(const struct classifier *cls, const struct flow *flow, |
74f74083 | 691 | struct flow_wildcards *wc) |
48c3de13 | 692 | { |
c906cedf | 693 | const struct cls_partition *partition; |
c906cedf | 694 | tag_type tags; |
eb391b76 | 695 | int best_priority = INT_MIN; |
fe7cfa5c JR |
696 | const struct cls_match *best; |
697 | struct trie_ctx trie_ctx[CLS_MAX_TRIES]; | |
698 | struct cls_subtable *subtable; | |
c906cedf | 699 | |
f358a2cb JR |
700 | /* Synchronize for cls->n_tries and subtable->trie_plen. They can change |
701 | * when table configuration changes, which happens typically only on | |
702 | * startup. */ | |
703 | atomic_thread_fence(memory_order_acquire); | |
704 | ||
03868246 JR |
705 | /* Determine 'tags' such that, if 'subtable->tag' doesn't intersect them, |
706 | * then 'flow' cannot possibly match in 'subtable': | |
c906cedf BP |
707 | * |
708 | * - If flow->metadata maps to a given 'partition', then we can use | |
709 | * 'tags' for 'partition->tags'. | |
710 | * | |
711 | * - If flow->metadata has no partition, then no rule in 'cls' has an | |
712 | * exact-match for flow->metadata. That means that we don't need to | |
03868246 | 713 | * search any subtable that includes flow->metadata in its mask. |
c906cedf | 714 | * |
03868246 | 715 | * In either case, we always need to search any cls_subtables that do not |
c906cedf | 716 | * include flow->metadata in its mask. One way to do that would be to |
03868246 JR |
717 | * check the "cls_subtable"s explicitly for that, but that would require an |
718 | * extra branch per subtable. Instead, we mark such a cls_subtable's | |
719 | * 'tags' as TAG_ALL and make sure that 'tags' is never empty. This means | |
720 | * that 'tags' always intersects such a cls_subtable's 'tags', so we don't | |
721 | * need a special case. | |
c906cedf | 722 | */ |
f2c21402 | 723 | partition = (cmap_is_empty(&cls->partitions) |
c906cedf BP |
724 | ? NULL |
725 | : find_partition(cls, flow->metadata, | |
726 | hash_metadata(flow->metadata))); | |
727 | tags = partition ? partition->tags : TAG_ARBITRARY; | |
48c3de13 | 728 | |
ff8241db | 729 | /* Initialize trie contexts for find_match_wc(). */ |
fe7cfa5c | 730 | for (int i = 0; i < cls->n_tries; i++) { |
13751fd8 JR |
731 | trie_ctx_init(&trie_ctx[i], &cls->tries[i]); |
732 | } | |
ec988646 | 733 | |
b5d97350 | 734 | best = NULL; |
fe7cfa5c JR |
735 | PVECTOR_FOR_EACH_PRIORITY(subtable, best_priority, 2, |
736 | sizeof(struct cls_subtable), &cls->subtables) { | |
dfea28b3 | 737 | const struct cls_match *rule; |
c906cedf | 738 | |
fe7cfa5c | 739 | if (!tag_intersects(tags, subtable->tag)) { |
c906cedf BP |
740 | continue; |
741 | } | |
74f74083 | 742 | |
fe7cfa5c | 743 | rule = find_match_wc(subtable, flow, trie_ctx, cls->n_tries, wc); |
eb391b76 BP |
744 | if (rule && rule->priority > best_priority) { |
745 | best_priority = rule->priority; | |
1f3c5efc | 746 | best = rule; |
b5d97350 | 747 | } |
48c3de13 | 748 | } |
13751fd8 | 749 | |
627fb667 | 750 | return best ? best->cls_rule : NULL; |
48c3de13 BP |
751 | } |
752 | ||
b5d97350 BP |
753 | /* Finds and returns a rule in 'cls' with exactly the same priority and |
754 | * matching criteria as 'target'. Returns a null pointer if 'cls' doesn't | |
c084ce1d | 755 | * contain an exact match. */ |
dfea28b3 | 756 | const struct cls_rule * |
e48eccd1 | 757 | classifier_find_rule_exactly(const struct classifier *cls, |
76ecc721 | 758 | const struct cls_rule *target) |
064af421 | 759 | { |
dfea28b3 JR |
760 | const struct cls_match *head, *rule; |
761 | const struct cls_subtable *subtable; | |
064af421 | 762 | |
03868246 | 763 | subtable = find_subtable(cls, &target->match.mask); |
0722ee5c | 764 | if (!subtable) { |
98abae4a | 765 | return NULL; |
4d935a6b JR |
766 | } |
767 | ||
03868246 | 768 | head = find_equal(subtable, &target->match.flow, |
5cb7a798 BP |
769 | miniflow_hash_in_minimask(&target->match.flow, |
770 | &target->match.mask, 0)); | |
98abae4a JR |
771 | if (!head) { |
772 | return NULL; | |
773 | } | |
b5d97350 BP |
774 | FOR_EACH_RULE_IN_LIST (rule, head) { |
775 | if (target->priority >= rule->priority) { | |
627fb667 | 776 | return target->priority == rule->priority ? rule->cls_rule : NULL; |
064af421 BP |
777 | } |
778 | } | |
779 | return NULL; | |
780 | } | |
781 | ||
81a76618 BP |
782 | /* Finds and returns a rule in 'cls' with priority 'priority' and exactly the |
783 | * same matching criteria as 'target'. Returns a null pointer if 'cls' doesn't | |
784 | * contain an exact match. */ | |
dfea28b3 | 785 | const struct cls_rule * |
81a76618 | 786 | classifier_find_match_exactly(const struct classifier *cls, |
eb391b76 | 787 | const struct match *target, int priority) |
81a76618 | 788 | { |
dfea28b3 | 789 | const struct cls_rule *retval; |
81a76618 BP |
790 | struct cls_rule cr; |
791 | ||
792 | cls_rule_init(&cr, target, priority); | |
793 | retval = classifier_find_rule_exactly(cls, &cr); | |
48d28ac1 | 794 | cls_rule_destroy(&cr); |
81a76618 BP |
795 | |
796 | return retval; | |
797 | } | |
798 | ||
faa50f40 BP |
799 | /* Checks if 'target' would overlap any other rule in 'cls'. Two rules are |
800 | * considered to overlap if both rules have the same priority and a packet | |
801 | * could match both. */ | |
49bdc010 | 802 | bool |
e48eccd1 | 803 | classifier_rule_overlaps(const struct classifier *cls, |
faa50f40 | 804 | const struct cls_rule *target) |
e48eccd1 | 805 | OVS_EXCLUDED(cls->mutex) |
49bdc010 | 806 | { |
03868246 | 807 | struct cls_subtable *subtable; |
49bdc010 | 808 | |
e65413ab | 809 | ovs_mutex_lock(&cls->mutex); |
03868246 | 810 | /* Iterate subtables in the descending max priority order. */ |
eb391b76 | 811 | PVECTOR_FOR_EACH_PRIORITY (subtable, target->priority - 1, 2, |
fe7cfa5c | 812 | sizeof(struct cls_subtable), &cls->subtables) { |
5cb7a798 BP |
813 | uint32_t storage[FLOW_U32S]; |
814 | struct minimask mask; | |
627fb667 | 815 | struct cls_match *head; |
49bdc010 | 816 | |
03868246 | 817 | minimask_combine(&mask, &target->match.mask, &subtable->mask, storage); |
f2c21402 | 818 | CMAP_FOR_EACH (head, cmap_node, &subtable->rules) { |
627fb667 | 819 | struct cls_match *rule; |
49bdc010 | 820 | |
dfea28b3 | 821 | FOR_EACH_RULE_IN_LIST_PROTECTED (rule, head) { |
4d935a6b JR |
822 | if (rule->priority < target->priority) { |
823 | break; /* Rules in descending priority order. */ | |
824 | } | |
faa50f40 | 825 | if (rule->priority == target->priority |
5cb7a798 | 826 | && miniflow_equal_in_minimask(&target->match.flow, |
3016f3e4 | 827 | &rule->flow, &mask)) { |
e65413ab | 828 | ovs_mutex_unlock(&cls->mutex); |
49bdc010 JP |
829 | return true; |
830 | } | |
831 | } | |
832 | } | |
833 | } | |
834 | ||
e65413ab | 835 | ovs_mutex_unlock(&cls->mutex); |
49bdc010 JP |
836 | return false; |
837 | } | |
6ceeaa92 BP |
838 | |
839 | /* Returns true if 'rule' exactly matches 'criteria' or if 'rule' is more | |
840 | * specific than 'criteria'. That is, 'rule' matches 'criteria' and this | |
841 | * function returns true if, for every field: | |
842 | * | |
843 | * - 'criteria' and 'rule' specify the same (non-wildcarded) value for the | |
844 | * field, or | |
845 | * | |
846 | * - 'criteria' wildcards the field, | |
847 | * | |
848 | * Conversely, 'rule' does not match 'criteria' and this function returns false | |
849 | * if, for at least one field: | |
850 | * | |
851 | * - 'criteria' and 'rule' specify different values for the field, or | |
852 | * | |
853 | * - 'criteria' specifies a value for the field but 'rule' wildcards it. | |
854 | * | |
855 | * Equivalently, the truth table for whether a field matches is: | |
856 | * | |
857 | * rule | |
858 | * | |
859 | * c wildcard exact | |
860 | * r +---------+---------+ | |
861 | * i wild | yes | yes | | |
862 | * t card | | | | |
863 | * e +---------+---------+ | |
864 | * r exact | no |if values| | |
865 | * i | |are equal| | |
866 | * a +---------+---------+ | |
867 | * | |
868 | * This is the matching rule used by OpenFlow 1.0 non-strict OFPT_FLOW_MOD | |
869 | * commands and by OpenFlow 1.0 aggregate and flow stats. | |
870 | * | |
81a76618 | 871 | * Ignores rule->priority. */ |
6ceeaa92 BP |
872 | bool |
873 | cls_rule_is_loose_match(const struct cls_rule *rule, | |
5cb7a798 | 874 | const struct minimatch *criteria) |
6ceeaa92 | 875 | { |
5cb7a798 BP |
876 | return (!minimask_has_extra(&rule->match.mask, &criteria->mask) |
877 | && miniflow_equal_in_minimask(&rule->match.flow, &criteria->flow, | |
878 | &criteria->mask)); | |
6ceeaa92 | 879 | } |
b5d97350 | 880 | \f |
5ecc9d81 BP |
881 | /* Iteration. */ |
882 | ||
883 | static bool | |
627fb667 | 884 | rule_matches(const struct cls_match *rule, const struct cls_rule *target) |
5ecc9d81 BP |
885 | { |
886 | return (!target | |
3016f3e4 | 887 | || miniflow_equal_in_minimask(&rule->flow, |
5cb7a798 BP |
888 | &target->match.flow, |
889 | &target->match.mask)); | |
5ecc9d81 BP |
890 | } |
891 | ||
dfea28b3 | 892 | static const struct cls_match * |
03868246 | 893 | search_subtable(const struct cls_subtable *subtable, |
f2c21402 | 894 | struct cls_cursor *cursor) |
5ecc9d81 | 895 | { |
f2c21402 JR |
896 | if (!cursor->target |
897 | || !minimask_has_extra(&subtable->mask, &cursor->target->match.mask)) { | |
dfea28b3 | 898 | const struct cls_match *rule; |
5ecc9d81 | 899 | |
f2c21402 JR |
900 | CMAP_CURSOR_FOR_EACH (rule, cmap_node, &cursor->rules, |
901 | &subtable->rules) { | |
902 | if (rule_matches(rule, cursor->target)) { | |
5ecc9d81 BP |
903 | return rule; |
904 | } | |
905 | } | |
906 | } | |
907 | return NULL; | |
908 | } | |
909 | ||
5f0476ce JR |
910 | /* Initializes 'cursor' for iterating through rules in 'cls', and returns the |
911 | * first matching cls_rule via '*pnode', or NULL if there are no matches. | |
5ecc9d81 | 912 | * |
6ceeaa92 | 913 | * - If 'target' is null, the cursor will visit every rule in 'cls'. |
5ecc9d81 | 914 | * |
6ceeaa92 BP |
915 | * - If 'target' is nonnull, the cursor will visit each 'rule' in 'cls' |
916 | * such that cls_rule_is_loose_match(rule, target) returns true. | |
5ecc9d81 | 917 | * |
6ceeaa92 | 918 | * Ignores target->priority. */ |
78c8df12 BP |
919 | struct cls_cursor cls_cursor_start(const struct classifier *cls, |
920 | const struct cls_rule *target, | |
921 | bool safe) | |
5f0476ce | 922 | OVS_NO_THREAD_SAFETY_ANALYSIS |
5ecc9d81 | 923 | { |
5f0476ce | 924 | struct cls_cursor cursor; |
03868246 | 925 | struct cls_subtable *subtable; |
5ecc9d81 | 926 | |
5f0476ce | 927 | cursor.safe = safe; |
e48eccd1 | 928 | cursor.cls = cls; |
5f0476ce | 929 | cursor.target = target && !cls_rule_is_catchall(target) ? target : NULL; |
78c8df12 | 930 | cursor.rule = NULL; |
5f0476ce JR |
931 | |
932 | /* Find first rule. */ | |
e65413ab | 933 | ovs_mutex_lock(&cursor.cls->mutex); |
5f0476ce | 934 | CMAP_CURSOR_FOR_EACH (subtable, cmap_node, &cursor.subtables, |
e65413ab | 935 | &cursor.cls->subtables_map) { |
dfea28b3 | 936 | const struct cls_match *rule = search_subtable(subtable, &cursor); |
f2c21402 | 937 | |
5ecc9d81 | 938 | if (rule) { |
5f0476ce | 939 | cursor.subtable = subtable; |
78c8df12 | 940 | cursor.rule = rule->cls_rule; |
5f0476ce | 941 | break; |
5ecc9d81 BP |
942 | } |
943 | } | |
944 | ||
5f0476ce | 945 | /* Leave locked if requested and have a rule. */ |
78c8df12 | 946 | if (safe || !cursor.rule) { |
e65413ab | 947 | ovs_mutex_unlock(&cursor.cls->mutex); |
5f0476ce JR |
948 | } |
949 | return cursor; | |
950 | } | |
951 | ||
dfea28b3 | 952 | static const struct cls_rule * |
1caa1561 | 953 | cls_cursor_next(struct cls_cursor *cursor) |
5f0476ce | 954 | OVS_NO_THREAD_SAFETY_ANALYSIS |
5ecc9d81 | 955 | { |
dfea28b3 | 956 | const struct cls_match *rule = cursor->rule->cls_match; |
03868246 | 957 | const struct cls_subtable *subtable; |
dfea28b3 | 958 | const struct cls_match *next; |
5ecc9d81 | 959 | |
955f579d BP |
960 | next = next_rule_in_list__(rule); |
961 | if (next->priority < rule->priority) { | |
1caa1561 | 962 | return next->cls_rule; |
5ecc9d81 BP |
963 | } |
964 | ||
955f579d | 965 | /* 'next' is the head of the list, that is, the rule that is included in |
f2c21402 | 966 | * the subtable's map. (This is important when the classifier contains |
03868246 | 967 | * rules that differ only in priority.) */ |
955f579d | 968 | rule = next; |
f2c21402 | 969 | CMAP_CURSOR_FOR_EACH_CONTINUE (rule, cmap_node, &cursor->rules) { |
5ecc9d81 | 970 | if (rule_matches(rule, cursor->target)) { |
1caa1561 | 971 | return rule->cls_rule; |
5ecc9d81 BP |
972 | } |
973 | } | |
974 | ||
03868246 | 975 | subtable = cursor->subtable; |
f2c21402 JR |
976 | CMAP_CURSOR_FOR_EACH_CONTINUE (subtable, cmap_node, &cursor->subtables) { |
977 | rule = search_subtable(subtable, cursor); | |
5ecc9d81 | 978 | if (rule) { |
03868246 | 979 | cursor->subtable = subtable; |
1caa1561 | 980 | return rule->cls_rule; |
5ecc9d81 BP |
981 | } |
982 | } | |
983 | ||
1caa1561 BP |
984 | return NULL; |
985 | } | |
986 | ||
987 | /* Sets 'cursor->rule' to the next matching cls_rule in 'cursor''s iteration, | |
988 | * or to null if all matching rules have been visited. */ | |
989 | void | |
990 | cls_cursor_advance(struct cls_cursor *cursor) | |
991 | OVS_NO_THREAD_SAFETY_ANALYSIS | |
992 | { | |
993 | if (cursor->safe) { | |
994 | ovs_mutex_lock(&cursor->cls->mutex); | |
995 | } | |
996 | cursor->rule = cls_cursor_next(cursor); | |
997 | if (cursor->safe || !cursor->rule) { | |
998 | ovs_mutex_unlock(&cursor->cls->mutex); | |
999 | } | |
5ecc9d81 BP |
1000 | } |
1001 | \f | |
03868246 | 1002 | static struct cls_subtable * |
e48eccd1 | 1003 | find_subtable(const struct classifier *cls, const struct minimask *mask) |
b5d97350 | 1004 | { |
03868246 | 1005 | struct cls_subtable *subtable; |
064af421 | 1006 | |
f2c21402 | 1007 | CMAP_FOR_EACH_WITH_HASH (subtable, cmap_node, minimask_hash(mask, 0), |
5a87054c | 1008 | &cls->subtables_map) { |
03868246 JR |
1009 | if (minimask_equal(mask, &subtable->mask)) { |
1010 | return subtable; | |
064af421 BP |
1011 | } |
1012 | } | |
b5d97350 | 1013 | return NULL; |
064af421 | 1014 | } |
064af421 | 1015 | |
e65413ab | 1016 | /* The new subtable will be visible to the readers only after this. */ |
03868246 | 1017 | static struct cls_subtable * |
e48eccd1 | 1018 | insert_subtable(struct classifier *cls, const struct minimask *mask) |
e65413ab | 1019 | OVS_REQUIRES(cls->mutex) |
b5d97350 | 1020 | { |
c906cedf | 1021 | uint32_t hash = minimask_hash(mask, 0); |
03868246 | 1022 | struct cls_subtable *subtable; |
476f36e8 JR |
1023 | int i, index = 0; |
1024 | struct flow_wildcards old, new; | |
1025 | uint8_t prev; | |
3016f3e4 | 1026 | int count = count_1bits(mask->masks.map); |
064af421 | 1027 | |
3016f3e4 JR |
1028 | subtable = xzalloc(sizeof *subtable - sizeof mask->masks.inline_values |
1029 | + MINIFLOW_VALUES_SIZE(count)); | |
f2c21402 | 1030 | cmap_init(&subtable->rules); |
f80028fe JR |
1031 | miniflow_clone_inline(CONST_CAST(struct miniflow *, &subtable->mask.masks), |
1032 | &mask->masks, count); | |
476f36e8 JR |
1033 | |
1034 | /* Init indices for segmented lookup, if any. */ | |
1035 | flow_wildcards_init_catchall(&new); | |
1036 | old = new; | |
1037 | prev = 0; | |
1038 | for (i = 0; i < cls->n_flow_segments; i++) { | |
1039 | flow_wildcards_fold_minimask_range(&new, mask, prev, | |
1040 | cls->flow_segments[i]); | |
1041 | /* Add an index if it adds mask bits. */ | |
1042 | if (!flow_wildcards_equal(&new, &old)) { | |
f2c21402 | 1043 | cmap_init(&subtable->indices[index]); |
f80028fe JR |
1044 | *CONST_CAST(uint8_t *, &subtable->index_ofs[index]) |
1045 | = cls->flow_segments[i]; | |
476f36e8 JR |
1046 | index++; |
1047 | old = new; | |
1048 | } | |
1049 | prev = cls->flow_segments[i]; | |
1050 | } | |
1051 | /* Check if the rest of the subtable's mask adds any bits, | |
1052 | * and remove the last index if it doesn't. */ | |
1053 | if (index > 0) { | |
1054 | flow_wildcards_fold_minimask_range(&new, mask, prev, FLOW_U32S); | |
1055 | if (flow_wildcards_equal(&new, &old)) { | |
1056 | --index; | |
f80028fe | 1057 | *CONST_CAST(uint8_t *, &subtable->index_ofs[index]) = 0; |
f2c21402 | 1058 | cmap_destroy(&subtable->indices[index]); |
476f36e8 JR |
1059 | } |
1060 | } | |
f80028fe | 1061 | *CONST_CAST(uint8_t *, &subtable->n_indices) = index; |
476f36e8 | 1062 | |
f80028fe JR |
1063 | *CONST_CAST(tag_type *, &subtable->tag) = |
1064 | (minimask_get_metadata_mask(mask) == OVS_BE64_MAX | |
1065 | ? tag_create_deterministic(hash) | |
1066 | : TAG_ALL); | |
064af421 | 1067 | |
13751fd8 JR |
1068 | for (i = 0; i < cls->n_tries; i++) { |
1069 | subtable->trie_plen[i] = minimask_get_prefix_len(mask, | |
1070 | cls->tries[i].field); | |
1071 | } | |
1072 | ||
69d6040e | 1073 | /* Ports trie. */ |
f358a2cb | 1074 | ovsrcu_set_hidden(&subtable->ports_trie, NULL); |
f80028fe | 1075 | *CONST_CAST(int *, &subtable->ports_mask_len) |
69d6040e JR |
1076 | = 32 - ctz32(ntohl(MINIFLOW_GET_BE32(&mask->masks, tp_src))); |
1077 | ||
f2c21402 | 1078 | cmap_insert(&cls->subtables_map, &subtable->cmap_node, hash); |
ec988646 | 1079 | |
03868246 | 1080 | return subtable; |
064af421 BP |
1081 | } |
1082 | ||
01c0f83a | 1083 | /* RCU readers may still access the subtable before it is actually freed. */ |
b5d97350 | 1084 | static void |
e48eccd1 | 1085 | destroy_subtable(struct classifier *cls, struct cls_subtable *subtable) |
e65413ab | 1086 | OVS_REQUIRES(cls->mutex) |
b5d97350 | 1087 | { |
476f36e8 JR |
1088 | int i; |
1089 | ||
fe7cfa5c | 1090 | pvector_remove(&cls->subtables, subtable); |
01c0f83a JR |
1091 | cmap_remove(&cls->subtables_map, &subtable->cmap_node, |
1092 | minimask_hash(&subtable->mask, 0)); | |
1093 | ||
1094 | ovs_assert(ovsrcu_get_protected(struct trie_node *, &subtable->ports_trie) | |
1095 | == NULL); | |
1096 | ovs_assert(cmap_is_empty(&subtable->rules)); | |
69d6040e | 1097 | |
476f36e8 | 1098 | for (i = 0; i < subtable->n_indices; i++) { |
f2c21402 | 1099 | cmap_destroy(&subtable->indices[i]); |
476f36e8 | 1100 | } |
f2c21402 | 1101 | cmap_destroy(&subtable->rules); |
fe7cfa5c | 1102 | ovsrcu_postpone(free, subtable); |
4aacd02d BP |
1103 | } |
1104 | ||
13751fd8 JR |
1105 | struct range { |
1106 | uint8_t start; | |
1107 | uint8_t end; | |
1108 | }; | |
1109 | ||
c0bfb650 JR |
1110 | static unsigned int be_get_bit_at(const ovs_be32 value[], unsigned int ofs); |
1111 | ||
13751fd8 JR |
1112 | /* Return 'true' if can skip rest of the subtable based on the prefix trie |
1113 | * lookup results. */ | |
1114 | static inline bool | |
1115 | check_tries(struct trie_ctx trie_ctx[CLS_MAX_TRIES], unsigned int n_tries, | |
1116 | const unsigned int field_plen[CLS_MAX_TRIES], | |
1117 | const struct range ofs, const struct flow *flow, | |
1118 | struct flow_wildcards *wc) | |
1119 | { | |
1120 | int j; | |
1121 | ||
1122 | /* Check if we could avoid fully unwildcarding the next level of | |
1123 | * fields using the prefix tries. The trie checks are done only as | |
1124 | * needed to avoid folding in additional bits to the wildcards mask. */ | |
1125 | for (j = 0; j < n_tries; j++) { | |
1126 | /* Is the trie field relevant for this subtable? */ | |
1127 | if (field_plen[j]) { | |
1128 | struct trie_ctx *ctx = &trie_ctx[j]; | |
1129 | uint8_t be32ofs = ctx->be32ofs; | |
1130 | ||
1131 | /* Is the trie field within the current range of fields? */ | |
1132 | if (be32ofs >= ofs.start && be32ofs < ofs.end) { | |
1133 | /* On-demand trie lookup. */ | |
1134 | if (!ctx->lookup_done) { | |
c0bfb650 JR |
1135 | memset(&ctx->match_plens, 0, sizeof ctx->match_plens); |
1136 | ctx->maskbits = trie_lookup(ctx->trie, flow, | |
1137 | &ctx->match_plens); | |
13751fd8 JR |
1138 | ctx->lookup_done = true; |
1139 | } | |
1140 | /* Possible to skip the rest of the subtable if subtable's | |
c0bfb650 JR |
1141 | * prefix on the field is not included in the lookup result. */ |
1142 | if (!be_get_bit_at(&ctx->match_plens.be32, field_plen[j] - 1)) { | |
1817dcea JR |
1143 | /* We want the trie lookup to never result in unwildcarding |
1144 | * any bits that would not be unwildcarded otherwise. | |
1145 | * Since the trie is shared by the whole classifier, it is | |
1146 | * possible that the 'maskbits' contain bits that are | |
1147 | * irrelevant for the partition relevant for the current | |
1148 | * packet. Hence the checks below. */ | |
13751fd8 | 1149 | |
13751fd8 | 1150 | /* Check that the trie result will not unwildcard more bits |
1817dcea | 1151 | * than this subtable would otherwise. */ |
13751fd8 JR |
1152 | if (ctx->maskbits <= field_plen[j]) { |
1153 | /* Unwildcard the bits and skip the rest. */ | |
1154 | mask_set_prefix_bits(wc, be32ofs, ctx->maskbits); | |
1155 | /* Note: Prerequisite already unwildcarded, as the only | |
1156 | * prerequisite of the supported trie lookup fields is | |
1817dcea JR |
1157 | * the ethertype, which is always unwildcarded. */ |
1158 | return true; | |
1159 | } | |
1160 | /* Can skip if the field is already unwildcarded. */ | |
1161 | if (mask_prefix_bits_set(wc, be32ofs, ctx->maskbits)) { | |
13751fd8 JR |
1162 | return true; |
1163 | } | |
1164 | } | |
1165 | } | |
1166 | } | |
1167 | } | |
1168 | return false; | |
1169 | } | |
1170 | ||
3016f3e4 JR |
1171 | /* Returns true if 'target' satisifies 'flow'/'mask', that is, if each bit |
1172 | * for which 'flow', for which 'mask' has a bit set, specifies a particular | |
1173 | * value has the correct value in 'target'. | |
1174 | * | |
1175 | * This function is equivalent to miniflow_equal_flow_in_minimask(flow, | |
a64759f0 JR |
1176 | * target, mask) but this is faster because of the invariant that |
1177 | * flow->map and mask->masks.map are the same, and that this version | |
1178 | * takes the 'wc'. */ | |
3016f3e4 JR |
1179 | static inline bool |
1180 | miniflow_and_mask_matches_flow(const struct miniflow *flow, | |
1181 | const struct minimask *mask, | |
e9319757 | 1182 | const struct flow *target) |
3016f3e4 JR |
1183 | { |
1184 | const uint32_t *flowp = miniflow_get_u32_values(flow); | |
1185 | const uint32_t *maskp = miniflow_get_u32_values(&mask->masks); | |
a64759f0 | 1186 | uint32_t idx; |
3016f3e4 | 1187 | |
a64759f0 JR |
1188 | MAP_FOR_EACH_INDEX(idx, mask->masks.map) { |
1189 | uint32_t diff = (*flowp++ ^ flow_u32_value(target, idx)) & *maskp++; | |
1190 | ||
1191 | if (diff) { | |
3016f3e4 JR |
1192 | return false; |
1193 | } | |
1194 | } | |
1195 | ||
1196 | return true; | |
1197 | } | |
1198 | ||
dfea28b3 | 1199 | static inline const struct cls_match * |
476f36e8 JR |
1200 | find_match(const struct cls_subtable *subtable, const struct flow *flow, |
1201 | uint32_t hash) | |
b5d97350 | 1202 | { |
dfea28b3 | 1203 | const struct cls_match *rule; |
b5d97350 | 1204 | |
f2c21402 | 1205 | CMAP_FOR_EACH_WITH_HASH (rule, cmap_node, hash, &subtable->rules) { |
3016f3e4 | 1206 | if (miniflow_and_mask_matches_flow(&rule->flow, &subtable->mask, |
e9319757 | 1207 | flow)) { |
b5d97350 | 1208 | return rule; |
064af421 BP |
1209 | } |
1210 | } | |
c23740be | 1211 | |
064af421 BP |
1212 | return NULL; |
1213 | } | |
1214 | ||
e9319757 JR |
1215 | /* Returns true if 'target' satisifies 'flow'/'mask', that is, if each bit |
1216 | * for which 'flow', for which 'mask' has a bit set, specifies a particular | |
1217 | * value has the correct value in 'target'. | |
1218 | * | |
1219 | * This function is equivalent to miniflow_and_mask_matches_flow() but this | |
1220 | * version fills in the mask bits in 'wc'. */ | |
1221 | static inline bool | |
1222 | miniflow_and_mask_matches_flow_wc(const struct miniflow *flow, | |
1223 | const struct minimask *mask, | |
1224 | const struct flow *target, | |
1225 | struct flow_wildcards *wc) | |
1226 | { | |
1227 | const uint32_t *flowp = miniflow_get_u32_values(flow); | |
1228 | const uint32_t *maskp = miniflow_get_u32_values(&mask->masks); | |
1229 | uint32_t idx; | |
1230 | ||
1231 | MAP_FOR_EACH_INDEX(idx, mask->masks.map) { | |
1232 | uint32_t mask = *maskp++; | |
1233 | uint32_t diff = (*flowp++ ^ flow_u32_value(target, idx)) & mask; | |
1234 | ||
1235 | if (diff) { | |
1236 | /* Only unwildcard if none of the differing bits is already | |
1237 | * exact-matched. */ | |
1238 | if (!(flow_u32_value(&wc->masks, idx) & diff)) { | |
1239 | /* Keep one bit of the difference. */ | |
1240 | *flow_u32_lvalue(&wc->masks, idx) |= rightmost_1bit(diff); | |
1241 | } | |
1242 | return false; | |
1243 | } | |
1244 | /* Fill in the bits that were looked at. */ | |
1245 | *flow_u32_lvalue(&wc->masks, idx) |= mask; | |
1246 | } | |
1247 | ||
1248 | return true; | |
1249 | } | |
1250 | ||
386cb9f7 JR |
1251 | /* Unwildcard the fields looked up so far, if any. */ |
1252 | static void | |
1253 | fill_range_wc(const struct cls_subtable *subtable, struct flow_wildcards *wc, | |
1254 | uint8_t to) | |
1255 | { | |
1256 | if (to) { | |
1257 | flow_wildcards_fold_minimask_range(wc, &subtable->mask, 0, to); | |
1258 | } | |
1259 | } | |
1260 | ||
dfea28b3 | 1261 | static const struct cls_match * |
476f36e8 | 1262 | find_match_wc(const struct cls_subtable *subtable, const struct flow *flow, |
13751fd8 JR |
1263 | struct trie_ctx trie_ctx[CLS_MAX_TRIES], unsigned int n_tries, |
1264 | struct flow_wildcards *wc) | |
476f36e8 JR |
1265 | { |
1266 | uint32_t basis = 0, hash; | |
dfea28b3 | 1267 | const struct cls_match *rule = NULL; |
476f36e8 | 1268 | int i; |
13751fd8 | 1269 | struct range ofs; |
476f36e8 | 1270 | |
ec988646 | 1271 | if (OVS_UNLIKELY(!wc)) { |
476f36e8 JR |
1272 | return find_match(subtable, flow, |
1273 | flow_hash_in_minimask(flow, &subtable->mask, 0)); | |
1274 | } | |
1275 | ||
13751fd8 | 1276 | ofs.start = 0; |
476f36e8 JR |
1277 | /* Try to finish early by checking fields in segments. */ |
1278 | for (i = 0; i < subtable->n_indices; i++) { | |
55847abe | 1279 | const struct cmap_node *inode; |
f2c21402 | 1280 | |
13751fd8 | 1281 | ofs.end = subtable->index_ofs[i]; |
476f36e8 | 1282 | |
13751fd8 JR |
1283 | if (check_tries(trie_ctx, n_tries, subtable->trie_plen, ofs, flow, |
1284 | wc)) { | |
386cb9f7 JR |
1285 | /* 'wc' bits for the trie field set, now unwildcard the preceding |
1286 | * bits used so far. */ | |
1287 | fill_range_wc(subtable, wc, ofs.start); | |
1288 | return NULL; | |
13751fd8 JR |
1289 | } |
1290 | hash = flow_hash_in_minimask_range(flow, &subtable->mask, ofs.start, | |
1291 | ofs.end, &basis); | |
f2c21402 | 1292 | inode = cmap_find(&subtable->indices[i], hash); |
476f36e8 | 1293 | if (!inode) { |
386cb9f7 JR |
1294 | /* No match, can stop immediately, but must fold in the bits |
1295 | * used in lookup so far. */ | |
1296 | fill_range_wc(subtable, wc, ofs.end); | |
1297 | return NULL; | |
476f36e8 JR |
1298 | } |
1299 | ||
1300 | /* If we have narrowed down to a single rule already, check whether | |
a64759f0 | 1301 | * that rule matches. Either way, we're done. |
476f36e8 JR |
1302 | * |
1303 | * (Rare) hash collisions may cause us to miss the opportunity for this | |
1304 | * optimization. */ | |
f2c21402 | 1305 | if (!cmap_node_next(inode)) { |
476f36e8 | 1306 | ASSIGN_CONTAINER(rule, inode - i, index_nodes); |
e9319757 JR |
1307 | if (miniflow_and_mask_matches_flow_wc(&rule->flow, &subtable->mask, |
1308 | flow, wc)) { | |
1309 | return rule; | |
476f36e8 | 1310 | } |
e9319757 | 1311 | return NULL; |
476f36e8 | 1312 | } |
386cb9f7 | 1313 | ofs.start = ofs.end; |
476f36e8 | 1314 | } |
13751fd8 JR |
1315 | ofs.end = FLOW_U32S; |
1316 | /* Trie check for the final range. */ | |
1317 | if (check_tries(trie_ctx, n_tries, subtable->trie_plen, ofs, flow, wc)) { | |
386cb9f7 JR |
1318 | fill_range_wc(subtable, wc, ofs.start); |
1319 | return NULL; | |
13751fd8 | 1320 | } |
a64759f0 JR |
1321 | hash = flow_hash_in_minimask_range(flow, &subtable->mask, ofs.start, |
1322 | ofs.end, &basis); | |
1323 | rule = find_match(subtable, flow, hash); | |
69d6040e JR |
1324 | if (!rule && subtable->ports_mask_len) { |
1325 | /* Ports are always part of the final range, if any. | |
1326 | * No match was found for the ports. Use the ports trie to figure out | |
1327 | * which ports bits to unwildcard. */ | |
1328 | unsigned int mbits; | |
c0bfb650 | 1329 | ovs_be32 value, plens, mask; |
69d6040e JR |
1330 | |
1331 | mask = MINIFLOW_GET_BE32(&subtable->mask.masks, tp_src); | |
1332 | value = ((OVS_FORCE ovs_be32 *)flow)[TP_PORTS_OFS32] & mask; | |
c0bfb650 | 1333 | mbits = trie_lookup_value(&subtable->ports_trie, &value, &plens, 32); |
69d6040e JR |
1334 | |
1335 | ((OVS_FORCE ovs_be32 *)&wc->masks)[TP_PORTS_OFS32] |= | |
86f35fb5 | 1336 | mask & be32_prefix_mask(mbits); |
69d6040e | 1337 | |
386cb9f7 JR |
1338 | /* Unwildcard all bits in the mask upto the ports, as they were used |
1339 | * to determine there is no match. */ | |
1340 | fill_range_wc(subtable, wc, TP_PORTS_OFS32); | |
1341 | return NULL; | |
69d6040e | 1342 | } |
e9319757 | 1343 | |
13751fd8 | 1344 | /* Must unwildcard all the fields, as they were looked at. */ |
476f36e8 JR |
1345 | flow_wildcards_fold_minimask(wc, &subtable->mask); |
1346 | return rule; | |
1347 | } | |
1348 | ||
627fb667 | 1349 | static struct cls_match * |
dfea28b3 | 1350 | find_equal(const struct cls_subtable *subtable, const struct miniflow *flow, |
03868246 | 1351 | uint32_t hash) |
064af421 | 1352 | { |
627fb667 | 1353 | struct cls_match *head; |
064af421 | 1354 | |
f2c21402 | 1355 | CMAP_FOR_EACH_WITH_HASH (head, cmap_node, hash, &subtable->rules) { |
3016f3e4 | 1356 | if (miniflow_equal(&head->flow, flow)) { |
b5d97350 | 1357 | return head; |
064af421 BP |
1358 | } |
1359 | } | |
1360 | return NULL; | |
1361 | } | |
1362 | ||
afae68b1 JR |
1363 | /* |
1364 | * As the readers are operating concurrently with the modifications, a | |
1365 | * concurrent reader may or may not see the new rule, depending on how | |
1366 | * the concurrent events overlap with each other. This is no | |
1367 | * different from the former locked behavior, but there the visibility | |
1368 | * of the new rule only depended on the timing of the locking | |
1369 | * functions. | |
1370 | * | |
1371 | * The new rule is first added to the segment indices, so the readers | |
1372 | * may find the rule in the indices before the rule is visible in the | |
1373 | * subtables 'rules' map. This may result in us losing the | |
1374 | * opportunity to quit lookups earlier, resulting in sub-optimal | |
1375 | * wildcarding. This will be fixed by forthcoming revalidation always | |
1376 | * scheduled after flow table changes. | |
1377 | * | |
1378 | * Similar behavior may happen due to us removing the overlapping rule | |
1379 | * (if any) from the indices only after the new rule has been added. | |
1380 | * | |
1381 | * The subtable's max priority is updated only after the rule is | |
1382 | * inserted, so the concurrent readers may not see the rule, as the | |
1383 | * updated priority ordered subtable list will only be visible after | |
1384 | * the subtable's max priority is updated. | |
1385 | * | |
1386 | * Similarly, the classifier's partitions for new rules are updated by | |
1387 | * the caller after this function, so the readers may keep skipping | |
1388 | * the subtable until they see the updated partitions. | |
1389 | */ | |
627fb667 | 1390 | static struct cls_match * |
e48eccd1 | 1391 | insert_rule(struct classifier *cls, struct cls_subtable *subtable, |
f2c21402 | 1392 | struct cls_rule *new_rule) |
e65413ab | 1393 | OVS_REQUIRES(cls->mutex) |
064af421 | 1394 | { |
627fb667 | 1395 | struct cls_match *old = NULL; |
f2c21402 JR |
1396 | struct cls_match *new = cls_match_alloc(new_rule); |
1397 | struct cls_match *head; | |
476f36e8 | 1398 | int i; |
f2c21402 | 1399 | uint32_t basis = 0, hash, ihash[CLS_MAX_INDICES]; |
13751fd8 | 1400 | uint8_t prev_be32ofs = 0; |
476f36e8 JR |
1401 | |
1402 | /* Add new node to segment indices. */ | |
1403 | for (i = 0; i < subtable->n_indices; i++) { | |
f2c21402 JR |
1404 | ihash[i] = minimatch_hash_range(&new_rule->match, prev_be32ofs, |
1405 | subtable->index_ofs[i], &basis); | |
1406 | cmap_insert(&subtable->indices[i], &new->index_nodes[i], ihash[i]); | |
13751fd8 | 1407 | prev_be32ofs = subtable->index_ofs[i]; |
476f36e8 | 1408 | } |
f2c21402 JR |
1409 | hash = minimatch_hash_range(&new_rule->match, prev_be32ofs, FLOW_U32S, |
1410 | &basis); | |
1411 | head = find_equal(subtable, &new_rule->match.flow, hash); | |
b5d97350 | 1412 | if (!head) { |
f2c21402 | 1413 | cmap_insert(&subtable->rules, &new->cmap_node, hash); |
c501b427 | 1414 | rculist_init(&new->list); |
4aacd02d | 1415 | goto out; |
b5d97350 BP |
1416 | } else { |
1417 | /* Scan the list for the insertion point that will keep the list in | |
1418 | * order of decreasing priority. */ | |
627fb667 | 1419 | struct cls_match *rule; |
476f36e8 | 1420 | |
dfea28b3 | 1421 | FOR_EACH_RULE_IN_LIST_PROTECTED (rule, head) { |
f2c21402 | 1422 | if (new->priority >= rule->priority) { |
b5d97350 | 1423 | if (rule == head) { |
f2c21402 JR |
1424 | /* 'new' is the new highest-priority flow in the list. */ |
1425 | cmap_replace(&subtable->rules, &rule->cmap_node, | |
1426 | &new->cmap_node, hash); | |
b5d97350 | 1427 | } |
064af421 | 1428 | |
f2c21402 | 1429 | if (new->priority == rule->priority) { |
c501b427 | 1430 | rculist_replace(&new->list, &rule->list); |
4aacd02d | 1431 | old = rule; |
b5d97350 | 1432 | } else { |
c501b427 | 1433 | rculist_insert(&rule->list, &new->list); |
b5d97350 | 1434 | } |
fe7cfa5c | 1435 | goto out; |
b5d97350 BP |
1436 | } |
1437 | } | |
064af421 | 1438 | |
b5d97350 | 1439 | /* Insert 'new' at the end of the list. */ |
c501b427 | 1440 | rculist_push_back(&head->list, &new->list); |
064af421 | 1441 | } |
4aacd02d BP |
1442 | |
1443 | out: | |
1444 | if (!old) { | |
fe7cfa5c JR |
1445 | subtable->n_rules++; |
1446 | ||
1447 | /* Rule was added, not replaced. Update 'subtable's 'max_priority' | |
1448 | * and 'max_count', if necessary. */ | |
1449 | if (subtable->n_rules == 1) { | |
f2c21402 | 1450 | subtable->max_priority = new->priority; |
fe7cfa5c | 1451 | subtable->max_count = 1; |
f2c21402 JR |
1452 | pvector_insert(&cls->subtables, subtable, new->priority); |
1453 | } else if (subtable->max_priority == new->priority) { | |
fe7cfa5c | 1454 | ++subtable->max_count; |
f2c21402 JR |
1455 | } else if (new->priority > subtable->max_priority) { |
1456 | subtable->max_priority = new->priority; | |
fe7cfa5c | 1457 | subtable->max_count = 1; |
f2c21402 | 1458 | pvector_change_priority(&cls->subtables, subtable, new->priority); |
fe7cfa5c | 1459 | } |
476f36e8 JR |
1460 | } else { |
1461 | /* Remove old node from indices. */ | |
1462 | for (i = 0; i < subtable->n_indices; i++) { | |
f2c21402 | 1463 | cmap_remove(&subtable->indices[i], &old->index_nodes[i], ihash[i]); |
476f36e8 | 1464 | } |
4aacd02d BP |
1465 | } |
1466 | return old; | |
064af421 | 1467 | } |
13751fd8 JR |
1468 | \f |
1469 | /* A longest-prefix match tree. */ | |
13751fd8 JR |
1470 | |
1471 | /* Return at least 'plen' bits of the 'prefix', starting at bit offset 'ofs'. | |
1472 | * Prefixes are in the network byte order, and the offset 0 corresponds to | |
1473 | * the most significant bit of the first byte. The offset can be read as | |
1474 | * "how many bits to skip from the start of the prefix starting at 'pr'". */ | |
1475 | static uint32_t | |
1476 | raw_get_prefix(const ovs_be32 pr[], unsigned int ofs, unsigned int plen) | |
1477 | { | |
1478 | uint32_t prefix; | |
1479 | ||
1480 | pr += ofs / 32; /* Where to start. */ | |
1481 | ofs %= 32; /* How many bits to skip at 'pr'. */ | |
1482 | ||
1483 | prefix = ntohl(*pr) << ofs; /* Get the first 32 - ofs bits. */ | |
1484 | if (plen > 32 - ofs) { /* Need more than we have already? */ | |
1485 | prefix |= ntohl(*++pr) >> (32 - ofs); | |
1486 | } | |
1487 | /* Return with possible unwanted bits at the end. */ | |
1488 | return prefix; | |
1489 | } | |
1490 | ||
1491 | /* Return min(TRIE_PREFIX_BITS, plen) bits of the 'prefix', starting at bit | |
1492 | * offset 'ofs'. Prefixes are in the network byte order, and the offset 0 | |
1493 | * corresponds to the most significant bit of the first byte. The offset can | |
1494 | * be read as "how many bits to skip from the start of the prefix starting at | |
1495 | * 'pr'". */ | |
1496 | static uint32_t | |
1497 | trie_get_prefix(const ovs_be32 pr[], unsigned int ofs, unsigned int plen) | |
1498 | { | |
1499 | if (!plen) { | |
1500 | return 0; | |
1501 | } | |
1502 | if (plen > TRIE_PREFIX_BITS) { | |
1503 | plen = TRIE_PREFIX_BITS; /* Get at most TRIE_PREFIX_BITS. */ | |
1504 | } | |
1505 | /* Return with unwanted bits cleared. */ | |
1506 | return raw_get_prefix(pr, ofs, plen) & ~0u << (32 - plen); | |
1507 | } | |
1508 | ||
c30cfa6b | 1509 | /* Return the number of equal bits in 'n_bits' of 'prefix's MSBs and a 'value' |
13751fd8 JR |
1510 | * starting at "MSB 0"-based offset 'ofs'. */ |
1511 | static unsigned int | |
c30cfa6b | 1512 | prefix_equal_bits(uint32_t prefix, unsigned int n_bits, const ovs_be32 value[], |
13751fd8 JR |
1513 | unsigned int ofs) |
1514 | { | |
c30cfa6b | 1515 | uint64_t diff = prefix ^ raw_get_prefix(value, ofs, n_bits); |
13751fd8 | 1516 | /* Set the bit after the relevant bits to limit the result. */ |
c30cfa6b | 1517 | return raw_clz64(diff << 32 | UINT64_C(1) << (63 - n_bits)); |
13751fd8 JR |
1518 | } |
1519 | ||
1520 | /* Return the number of equal bits in 'node' prefix and a 'prefix' of length | |
1521 | * 'plen', starting at "MSB 0"-based offset 'ofs'. */ | |
1522 | static unsigned int | |
1523 | trie_prefix_equal_bits(const struct trie_node *node, const ovs_be32 prefix[], | |
1524 | unsigned int ofs, unsigned int plen) | |
1525 | { | |
c30cfa6b | 1526 | return prefix_equal_bits(node->prefix, MIN(node->n_bits, plen - ofs), |
13751fd8 JR |
1527 | prefix, ofs); |
1528 | } | |
1529 | ||
1530 | /* Return the bit at ("MSB 0"-based) offset 'ofs' as an int. 'ofs' can | |
1531 | * be greater than 31. */ | |
1532 | static unsigned int | |
1533 | be_get_bit_at(const ovs_be32 value[], unsigned int ofs) | |
1534 | { | |
1535 | return (((const uint8_t *)value)[ofs / 8] >> (7 - ofs % 8)) & 1u; | |
1536 | } | |
1537 | ||
1538 | /* Return the bit at ("MSB 0"-based) offset 'ofs' as an int. 'ofs' must | |
1539 | * be between 0 and 31, inclusive. */ | |
1540 | static unsigned int | |
1541 | get_bit_at(const uint32_t prefix, unsigned int ofs) | |
1542 | { | |
1543 | return (prefix >> (31 - ofs)) & 1u; | |
1544 | } | |
1545 | ||
1546 | /* Create new branch. */ | |
1547 | static struct trie_node * | |
1548 | trie_branch_create(const ovs_be32 *prefix, unsigned int ofs, unsigned int plen, | |
1549 | unsigned int n_rules) | |
1550 | { | |
1551 | struct trie_node *node = xmalloc(sizeof *node); | |
1552 | ||
1553 | node->prefix = trie_get_prefix(prefix, ofs, plen); | |
1554 | ||
1555 | if (plen <= TRIE_PREFIX_BITS) { | |
c30cfa6b | 1556 | node->n_bits = plen; |
f358a2cb JR |
1557 | ovsrcu_set_hidden(&node->edges[0], NULL); |
1558 | ovsrcu_set_hidden(&node->edges[1], NULL); | |
13751fd8 JR |
1559 | node->n_rules = n_rules; |
1560 | } else { /* Need intermediate nodes. */ | |
1561 | struct trie_node *subnode = trie_branch_create(prefix, | |
1562 | ofs + TRIE_PREFIX_BITS, | |
1563 | plen - TRIE_PREFIX_BITS, | |
1564 | n_rules); | |
1565 | int bit = get_bit_at(subnode->prefix, 0); | |
c30cfa6b | 1566 | node->n_bits = TRIE_PREFIX_BITS; |
f358a2cb JR |
1567 | ovsrcu_set_hidden(&node->edges[bit], subnode); |
1568 | ovsrcu_set_hidden(&node->edges[!bit], NULL); | |
13751fd8 JR |
1569 | node->n_rules = 0; |
1570 | } | |
1571 | return node; | |
1572 | } | |
1573 | ||
1574 | static void | |
f358a2cb | 1575 | trie_node_destroy(const struct trie_node *node) |
13751fd8 | 1576 | { |
f358a2cb JR |
1577 | ovsrcu_postpone(free, CONST_CAST(struct trie_node *, node)); |
1578 | } | |
1579 | ||
1580 | /* Copy a trie node for modification and postpone delete the old one. */ | |
1581 | static struct trie_node * | |
1582 | trie_node_rcu_realloc(const struct trie_node *node) | |
1583 | { | |
1584 | struct trie_node *new_node = xmalloc(sizeof *node); | |
1585 | ||
1586 | *new_node = *node; | |
1587 | trie_node_destroy(node); | |
1588 | ||
1589 | return new_node; | |
13751fd8 JR |
1590 | } |
1591 | ||
e48eccd1 | 1592 | /* May only be called while holding the classifier mutex. */ |
13751fd8 | 1593 | static void |
f358a2cb | 1594 | trie_destroy(rcu_trie_ptr *trie) |
13751fd8 | 1595 | { |
f358a2cb JR |
1596 | struct trie_node *node = ovsrcu_get_protected(struct trie_node *, trie); |
1597 | ||
13751fd8 | 1598 | if (node) { |
f358a2cb JR |
1599 | ovsrcu_set_hidden(trie, NULL); |
1600 | trie_destroy(&node->edges[0]); | |
1601 | trie_destroy(&node->edges[1]); | |
1602 | trie_node_destroy(node); | |
13751fd8 JR |
1603 | } |
1604 | } | |
1605 | ||
1606 | static bool | |
1607 | trie_is_leaf(const struct trie_node *trie) | |
1608 | { | |
f358a2cb JR |
1609 | /* No children? */ |
1610 | return !ovsrcu_get(struct trie_node *, &trie->edges[0]) | |
1611 | && !ovsrcu_get(struct trie_node *, &trie->edges[1]); | |
13751fd8 JR |
1612 | } |
1613 | ||
1614 | static void | |
1615 | mask_set_prefix_bits(struct flow_wildcards *wc, uint8_t be32ofs, | |
c30cfa6b | 1616 | unsigned int n_bits) |
13751fd8 JR |
1617 | { |
1618 | ovs_be32 *mask = &((ovs_be32 *)&wc->masks)[be32ofs]; | |
1619 | unsigned int i; | |
1620 | ||
c30cfa6b | 1621 | for (i = 0; i < n_bits / 32; i++) { |
13751fd8 JR |
1622 | mask[i] = OVS_BE32_MAX; |
1623 | } | |
c30cfa6b JR |
1624 | if (n_bits % 32) { |
1625 | mask[i] |= htonl(~0u << (32 - n_bits % 32)); | |
13751fd8 JR |
1626 | } |
1627 | } | |
1628 | ||
1629 | static bool | |
1630 | mask_prefix_bits_set(const struct flow_wildcards *wc, uint8_t be32ofs, | |
c30cfa6b | 1631 | unsigned int n_bits) |
13751fd8 JR |
1632 | { |
1633 | ovs_be32 *mask = &((ovs_be32 *)&wc->masks)[be32ofs]; | |
1634 | unsigned int i; | |
1635 | ovs_be32 zeroes = 0; | |
1636 | ||
c30cfa6b | 1637 | for (i = 0; i < n_bits / 32; i++) { |
13751fd8 JR |
1638 | zeroes |= ~mask[i]; |
1639 | } | |
c30cfa6b JR |
1640 | if (n_bits % 32) { |
1641 | zeroes |= ~mask[i] & htonl(~0u << (32 - n_bits % 32)); | |
13751fd8 JR |
1642 | } |
1643 | ||
c30cfa6b | 1644 | return !zeroes; /* All 'n_bits' bits set. */ |
13751fd8 JR |
1645 | } |
1646 | ||
f358a2cb | 1647 | static rcu_trie_ptr * |
13751fd8 JR |
1648 | trie_next_edge(struct trie_node *node, const ovs_be32 value[], |
1649 | unsigned int ofs) | |
1650 | { | |
1651 | return node->edges + be_get_bit_at(value, ofs); | |
1652 | } | |
1653 | ||
1654 | static const struct trie_node * | |
1655 | trie_next_node(const struct trie_node *node, const ovs_be32 value[], | |
1656 | unsigned int ofs) | |
1657 | { | |
f358a2cb JR |
1658 | return ovsrcu_get(struct trie_node *, |
1659 | &node->edges[be_get_bit_at(value, ofs)]); | |
13751fd8 JR |
1660 | } |
1661 | ||
c0bfb650 JR |
1662 | /* Set the bit at ("MSB 0"-based) offset 'ofs'. 'ofs' can be greater than 31. |
1663 | */ | |
1664 | static void | |
1665 | be_set_bit_at(ovs_be32 value[], unsigned int ofs) | |
1666 | { | |
1667 | ((uint8_t *)value)[ofs / 8] |= 1u << (7 - ofs % 8); | |
1668 | } | |
1669 | ||
1670 | /* Returns the number of bits in the prefix mask necessary to determine a | |
1671 | * mismatch, in case there are longer prefixes in the tree below the one that | |
1672 | * matched. | |
1673 | * '*plens' will have a bit set for each prefix length that may have matching | |
1674 | * rules. The caller is responsible for clearing the '*plens' prior to | |
1675 | * calling this. | |
13751fd8 JR |
1676 | */ |
1677 | static unsigned int | |
f358a2cb | 1678 | trie_lookup_value(const rcu_trie_ptr *trie, const ovs_be32 value[], |
c0bfb650 | 1679 | ovs_be32 plens[], unsigned int n_bits) |
13751fd8 | 1680 | { |
13751fd8 | 1681 | const struct trie_node *prev = NULL; |
c0bfb650 JR |
1682 | const struct trie_node *node = ovsrcu_get(struct trie_node *, trie); |
1683 | unsigned int match_len = 0; /* Number of matching bits. */ | |
13751fd8 | 1684 | |
27ce650f | 1685 | for (; node; prev = node, node = trie_next_node(node, value, match_len)) { |
13751fd8 JR |
1686 | unsigned int eqbits; |
1687 | /* Check if this edge can be followed. */ | |
27ce650f JR |
1688 | eqbits = prefix_equal_bits(node->prefix, node->n_bits, value, |
1689 | match_len); | |
1690 | match_len += eqbits; | |
c30cfa6b | 1691 | if (eqbits < node->n_bits) { /* Mismatch, nothing more to be found. */ |
27ce650f | 1692 | /* Bit at offset 'match_len' differed. */ |
c0bfb650 | 1693 | return match_len + 1; /* Includes the first mismatching bit. */ |
13751fd8 JR |
1694 | } |
1695 | /* Full match, check if rules exist at this prefix length. */ | |
1696 | if (node->n_rules > 0) { | |
c0bfb650 | 1697 | be_set_bit_at(plens, match_len - 1); |
13751fd8 | 1698 | } |
27ce650f | 1699 | if (match_len >= n_bits) { |
c0bfb650 | 1700 | return n_bits; /* Full prefix. */ |
f0e5aa11 | 1701 | } |
13751fd8 | 1702 | } |
c0bfb650 JR |
1703 | /* node == NULL. Full match so far, but we tried to follow an |
1704 | * non-existing branch. Need to exclude the other branch if it exists | |
1705 | * (it does not if we were called on an empty trie or 'prev' is a leaf | |
1706 | * node). */ | |
1707 | return !prev || trie_is_leaf(prev) ? match_len : match_len + 1; | |
13751fd8 JR |
1708 | } |
1709 | ||
1710 | static unsigned int | |
1711 | trie_lookup(const struct cls_trie *trie, const struct flow *flow, | |
c0bfb650 | 1712 | union mf_value *plens) |
13751fd8 JR |
1713 | { |
1714 | const struct mf_field *mf = trie->field; | |
1715 | ||
1716 | /* Check that current flow matches the prerequisites for the trie | |
1717 | * field. Some match fields are used for multiple purposes, so we | |
1718 | * must check that the trie is relevant for this flow. */ | |
1719 | if (mf_are_prereqs_ok(mf, flow)) { | |
f358a2cb | 1720 | return trie_lookup_value(&trie->root, |
13751fd8 | 1721 | &((ovs_be32 *)flow)[mf->flow_be32ofs], |
c0bfb650 | 1722 | &plens->be32, mf->n_bits); |
13751fd8 | 1723 | } |
c0bfb650 JR |
1724 | memset(plens, 0xff, sizeof *plens); /* All prefixes, no skipping. */ |
1725 | return 0; /* Value not used in this case. */ | |
13751fd8 JR |
1726 | } |
1727 | ||
1728 | /* Returns the length of a prefix match mask for the field 'mf' in 'minimask'. | |
1729 | * Returns the u32 offset to the miniflow data in '*miniflow_index', if | |
1730 | * 'miniflow_index' is not NULL. */ | |
1731 | static unsigned int | |
1732 | minimask_get_prefix_len(const struct minimask *minimask, | |
1733 | const struct mf_field *mf) | |
1734 | { | |
c30cfa6b | 1735 | unsigned int n_bits = 0, mask_tz = 0; /* Non-zero when end of mask seen. */ |
13751fd8 JR |
1736 | uint8_t u32_ofs = mf->flow_be32ofs; |
1737 | uint8_t u32_end = u32_ofs + mf->n_bytes / 4; | |
1738 | ||
1739 | for (; u32_ofs < u32_end; ++u32_ofs) { | |
1740 | uint32_t mask; | |
1741 | mask = ntohl((OVS_FORCE ovs_be32)minimask_get(minimask, u32_ofs)); | |
1742 | ||
1743 | /* Validate mask, count the mask length. */ | |
1744 | if (mask_tz) { | |
1745 | if (mask) { | |
1746 | return 0; /* No bits allowed after mask ended. */ | |
1747 | } | |
1748 | } else { | |
1749 | if (~mask & (~mask + 1)) { | |
1750 | return 0; /* Mask not contiguous. */ | |
1751 | } | |
1752 | mask_tz = ctz32(mask); | |
c30cfa6b | 1753 | n_bits += 32 - mask_tz; |
13751fd8 JR |
1754 | } |
1755 | } | |
1756 | ||
c30cfa6b | 1757 | return n_bits; |
13751fd8 JR |
1758 | } |
1759 | ||
1760 | /* | |
1761 | * This is called only when mask prefix is known to be CIDR and non-zero. | |
1762 | * Relies on the fact that the flow and mask have the same map, and since | |
1763 | * the mask is CIDR, the storage for the flow field exists even if it | |
1764 | * happened to be zeros. | |
1765 | */ | |
1766 | static const ovs_be32 * | |
1767 | minimatch_get_prefix(const struct minimatch *match, const struct mf_field *mf) | |
1768 | { | |
27bbe15d | 1769 | return miniflow_get_be32_values(&match->flow) + |
13751fd8 JR |
1770 | count_1bits(match->flow.map & ((UINT64_C(1) << mf->flow_be32ofs) - 1)); |
1771 | } | |
1772 | ||
1773 | /* Insert rule in to the prefix tree. | |
1774 | * 'mlen' must be the (non-zero) CIDR prefix length of the 'trie->field' mask | |
1775 | * in 'rule'. */ | |
1776 | static void | |
1777 | trie_insert(struct cls_trie *trie, const struct cls_rule *rule, int mlen) | |
1778 | { | |
69d6040e JR |
1779 | trie_insert_prefix(&trie->root, |
1780 | minimatch_get_prefix(&rule->match, trie->field), mlen); | |
1781 | } | |
1782 | ||
1783 | static void | |
f358a2cb | 1784 | trie_insert_prefix(rcu_trie_ptr *edge, const ovs_be32 *prefix, int mlen) |
69d6040e | 1785 | { |
13751fd8 | 1786 | struct trie_node *node; |
13751fd8 JR |
1787 | int ofs = 0; |
1788 | ||
1789 | /* Walk the tree. */ | |
f358a2cb | 1790 | for (; (node = ovsrcu_get_protected(struct trie_node *, edge)); |
13751fd8 JR |
1791 | edge = trie_next_edge(node, prefix, ofs)) { |
1792 | unsigned int eqbits = trie_prefix_equal_bits(node, prefix, ofs, mlen); | |
1793 | ofs += eqbits; | |
c30cfa6b | 1794 | if (eqbits < node->n_bits) { |
13751fd8 JR |
1795 | /* Mismatch, new node needs to be inserted above. */ |
1796 | int old_branch = get_bit_at(node->prefix, eqbits); | |
f358a2cb | 1797 | struct trie_node *new_parent; |
13751fd8 | 1798 | |
f358a2cb JR |
1799 | new_parent = trie_branch_create(prefix, ofs - eqbits, eqbits, |
1800 | ofs == mlen ? 1 : 0); | |
1801 | /* Copy the node to modify it. */ | |
1802 | node = trie_node_rcu_realloc(node); | |
1803 | /* Adjust the new node for its new position in the tree. */ | |
13751fd8 | 1804 | node->prefix <<= eqbits; |
c30cfa6b | 1805 | node->n_bits -= eqbits; |
f358a2cb | 1806 | ovsrcu_set_hidden(&new_parent->edges[old_branch], node); |
13751fd8 JR |
1807 | |
1808 | /* Check if need a new branch for the new rule. */ | |
1809 | if (ofs < mlen) { | |
f358a2cb JR |
1810 | ovsrcu_set_hidden(&new_parent->edges[!old_branch], |
1811 | trie_branch_create(prefix, ofs, mlen - ofs, | |
1812 | 1)); | |
13751fd8 | 1813 | } |
f358a2cb | 1814 | ovsrcu_set(edge, new_parent); /* Publish changes. */ |
13751fd8 JR |
1815 | return; |
1816 | } | |
1817 | /* Full match so far. */ | |
1818 | ||
1819 | if (ofs == mlen) { | |
1820 | /* Full match at the current node, rule needs to be added here. */ | |
1821 | node->n_rules++; | |
1822 | return; | |
1823 | } | |
1824 | } | |
1825 | /* Must insert a new tree branch for the new rule. */ | |
f358a2cb | 1826 | ovsrcu_set(edge, trie_branch_create(prefix, ofs, mlen - ofs, 1)); |
13751fd8 JR |
1827 | } |
1828 | ||
1829 | /* 'mlen' must be the (non-zero) CIDR prefix length of the 'trie->field' mask | |
1830 | * in 'rule'. */ | |
1831 | static void | |
1832 | trie_remove(struct cls_trie *trie, const struct cls_rule *rule, int mlen) | |
1833 | { | |
69d6040e JR |
1834 | trie_remove_prefix(&trie->root, |
1835 | minimatch_get_prefix(&rule->match, trie->field), mlen); | |
1836 | } | |
1837 | ||
1838 | /* 'mlen' must be the (non-zero) CIDR prefix length of the 'trie->field' mask | |
1839 | * in 'rule'. */ | |
1840 | static void | |
f358a2cb | 1841 | trie_remove_prefix(rcu_trie_ptr *root, const ovs_be32 *prefix, int mlen) |
69d6040e | 1842 | { |
13751fd8 | 1843 | struct trie_node *node; |
f358a2cb | 1844 | rcu_trie_ptr *edges[sizeof(union mf_value) * 8]; |
13751fd8 JR |
1845 | int depth = 0, ofs = 0; |
1846 | ||
1847 | /* Walk the tree. */ | |
69d6040e | 1848 | for (edges[0] = root; |
f358a2cb | 1849 | (node = ovsrcu_get_protected(struct trie_node *, edges[depth])); |
13751fd8 JR |
1850 | edges[++depth] = trie_next_edge(node, prefix, ofs)) { |
1851 | unsigned int eqbits = trie_prefix_equal_bits(node, prefix, ofs, mlen); | |
69d6040e | 1852 | |
c30cfa6b | 1853 | if (eqbits < node->n_bits) { |
13751fd8 JR |
1854 | /* Mismatch, nothing to be removed. This should never happen, as |
1855 | * only rules in the classifier are ever removed. */ | |
1856 | break; /* Log a warning. */ | |
1857 | } | |
1858 | /* Full match so far. */ | |
1859 | ofs += eqbits; | |
1860 | ||
1861 | if (ofs == mlen) { | |
1862 | /* Full prefix match at the current node, remove rule here. */ | |
1863 | if (!node->n_rules) { | |
1864 | break; /* Log a warning. */ | |
1865 | } | |
1866 | node->n_rules--; | |
1867 | ||
1868 | /* Check if can prune the tree. */ | |
f358a2cb JR |
1869 | while (!node->n_rules) { |
1870 | struct trie_node *next, | |
1871 | *edge0 = ovsrcu_get_protected(struct trie_node *, | |
1872 | &node->edges[0]), | |
1873 | *edge1 = ovsrcu_get_protected(struct trie_node *, | |
1874 | &node->edges[1]); | |
1875 | ||
1876 | if (edge0 && edge1) { | |
1877 | break; /* A branching point, cannot prune. */ | |
1878 | } | |
1879 | ||
1880 | /* Else have at most one child node, remove this node. */ | |
1881 | next = edge0 ? edge0 : edge1; | |
13751fd8 JR |
1882 | |
1883 | if (next) { | |
c30cfa6b | 1884 | if (node->n_bits + next->n_bits > TRIE_PREFIX_BITS) { |
13751fd8 JR |
1885 | break; /* Cannot combine. */ |
1886 | } | |
f358a2cb JR |
1887 | next = trie_node_rcu_realloc(next); /* Modify. */ |
1888 | ||
13751fd8 | 1889 | /* Combine node with next. */ |
c30cfa6b JR |
1890 | next->prefix = node->prefix | next->prefix >> node->n_bits; |
1891 | next->n_bits += node->n_bits; | |
13751fd8 | 1892 | } |
13751fd8 | 1893 | /* Update the parent's edge. */ |
f358a2cb JR |
1894 | ovsrcu_set(edges[depth], next); /* Publish changes. */ |
1895 | trie_node_destroy(node); | |
1896 | ||
13751fd8 JR |
1897 | if (next || !depth) { |
1898 | /* Branch not pruned or at root, nothing more to do. */ | |
1899 | break; | |
1900 | } | |
f358a2cb JR |
1901 | node = ovsrcu_get_protected(struct trie_node *, |
1902 | edges[--depth]); | |
13751fd8 JR |
1903 | } |
1904 | return; | |
1905 | } | |
1906 | } | |
1907 | /* Cannot go deeper. This should never happen, since only rules | |
1908 | * that actually exist in the classifier are ever removed. */ | |
1909 | VLOG_WARN("Trying to remove non-existing rule from a prefix trie."); | |
1910 | } |