]>
Commit | Line | Data |
---|---|---|
9558d2a5 JG |
1 | /* |
2 | * Copyright (c) 2015 Nicira, Inc. | |
3 | * | |
4 | * Licensed under the Apache License, Version 2.0 (the "License"); | |
5 | * you may not use this file except in compliance with the License. | |
6 | * You may obtain a copy of the License at: | |
7 | * | |
8 | * http://www.apache.org/licenses/LICENSE-2.0 | |
9 | * | |
10 | * Unless required by applicable law or agreed to in writing, software | |
11 | * distributed under the License is distributed on an "AS IS" BASIS, | |
12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |
13 | * See the License for the specific language governing permissions and | |
14 | * limitations under the License. | |
15 | */ | |
16 | ||
17 | #include <config.h> | |
18 | #include <errno.h> | |
19 | #include <stdbool.h> | |
20 | ||
21 | #include "bitmap.h" | |
22 | #include "compiler.h" | |
ee89ea7b | 23 | #include "openvswitch/hmap.h" |
e29747e4 | 24 | #include "openvswitch/match.h" |
9558d2a5 JG |
25 | #include "nx-match.h" |
26 | #include "odp-netlink.h" | |
0d71302e | 27 | #include "openvswitch/ofp-match.h" |
9558d2a5 JG |
28 | #include "ovs-rcu.h" |
29 | #include "packets.h" | |
30 | #include "tun-metadata.h" | |
ee89ea7b | 31 | #include "util.h" |
9558d2a5 JG |
32 | |
33 | struct tun_meta_entry { | |
34 | struct hmap_node node; /* In struct tun_table's key_hmap. */ | |
9558d2a5 | 35 | struct tun_metadata_loc loc; |
69a72315 | 36 | uint32_t key; /* (class << 8) | type. */ |
9558d2a5 JG |
37 | bool valid; /* True if allocated to a class and type. */ |
38 | }; | |
39 | ||
4e548ad9 | 40 | /* Maps from TLV option class+type to positions in a struct tun_metadata's |
9558d2a5 JG |
41 | * 'opts' array. */ |
42 | struct tun_table { | |
43 | /* TUN_METADATA<i> is stored in element <i>. */ | |
44 | struct tun_meta_entry entries[TUN_METADATA_NUM_OPTS]; | |
45 | ||
46 | /* Each bit represents 4 bytes of space, 0-bits are free space. */ | |
47 | unsigned long alloc_map[BITMAP_N_LONGS(TUN_METADATA_TOT_OPT_SIZE / 4)]; | |
48 | ||
49 | /* The valid elements in entries[], indexed by class+type. */ | |
50 | struct hmap key_hmap; | |
51 | }; | |
52 | BUILD_ASSERT_DECL(TUN_METADATA_TOT_OPT_SIZE % 4 == 0); | |
53 | ||
9558d2a5 JG |
54 | static enum ofperr tun_metadata_add_entry(struct tun_table *map, uint8_t idx, |
55 | uint16_t opt_class, uint8_t type, | |
8d8ab6c2 JG |
56 | uint8_t len); |
57 | static void tun_metadata_del_entry(struct tun_table *map, uint8_t idx); | |
9558d2a5 | 58 | static void memcpy_to_metadata(struct tun_metadata *dst, const void *src, |
2234a7db JG |
59 | const struct tun_metadata_loc *, |
60 | unsigned int idx); | |
9558d2a5 JG |
61 | static void memcpy_from_metadata(void *dst, const struct tun_metadata *src, |
62 | const struct tun_metadata_loc *); | |
63 | ||
64 | static uint32_t | |
65 | tun_meta_key(ovs_be16 class, uint8_t type) | |
66 | { | |
67 | return (OVS_FORCE uint16_t)class << 8 | type; | |
68 | } | |
69 | ||
70 | static ovs_be16 | |
71 | tun_key_class(uint32_t key) | |
72 | { | |
73 | return (OVS_FORCE ovs_be16)(key >> 8); | |
74 | } | |
75 | ||
76 | static uint8_t | |
77 | tun_key_type(uint32_t key) | |
78 | { | |
79 | return key & 0xff; | |
80 | } | |
81 | ||
82 | /* Returns a newly allocated tun_table. If 'old_map' is nonnull then the new | |
83 | * tun_table is a deep copy of the old one. */ | |
8d8ab6c2 JG |
84 | struct tun_table * |
85 | tun_metadata_alloc(const struct tun_table *old_map) | |
9558d2a5 JG |
86 | { |
87 | struct tun_table *new_map; | |
88 | ||
89 | new_map = xzalloc(sizeof *new_map); | |
90 | ||
91 | if (old_map) { | |
92 | struct tun_meta_entry *entry; | |
93 | ||
94 | *new_map = *old_map; | |
95 | hmap_init(&new_map->key_hmap); | |
96 | ||
97 | HMAP_FOR_EACH (entry, node, &old_map->key_hmap) { | |
98 | struct tun_meta_entry *new_entry; | |
99 | struct tun_metadata_loc_chain *chain; | |
100 | ||
101 | new_entry = &new_map->entries[entry - old_map->entries]; | |
102 | hmap_insert(&new_map->key_hmap, &new_entry->node, entry->node.hash); | |
103 | ||
104 | chain = &new_entry->loc.c; | |
105 | while (chain->next) { | |
106 | chain->next = xmemdup(chain->next, sizeof *chain->next); | |
107 | chain = chain->next; | |
108 | } | |
109 | } | |
110 | } else { | |
111 | hmap_init(&new_map->key_hmap); | |
112 | } | |
113 | ||
114 | return new_map; | |
115 | } | |
116 | ||
117 | /* Frees 'map' and all the memory it owns. */ | |
8d8ab6c2 JG |
118 | void |
119 | tun_metadata_free(struct tun_table *map) | |
9558d2a5 JG |
120 | { |
121 | struct tun_meta_entry *entry; | |
122 | ||
123 | if (!map) { | |
124 | return; | |
125 | } | |
126 | ||
127 | HMAP_FOR_EACH (entry, node, &map->key_hmap) { | |
128 | tun_metadata_del_entry(map, entry - map->entries); | |
129 | } | |
130 | ||
070767fc | 131 | hmap_destroy(&map->key_hmap); |
9558d2a5 JG |
132 | free(map); |
133 | } | |
134 | ||
9558d2a5 | 135 | void |
8d8ab6c2 | 136 | tun_metadata_postpone_free(struct tun_table *tab) |
9558d2a5 | 137 | { |
8d8ab6c2 | 138 | ovsrcu_postpone(tun_metadata_free, tab); |
9558d2a5 JG |
139 | } |
140 | ||
141 | enum ofperr | |
8d8ab6c2 JG |
142 | tun_metadata_table_mod(struct ofputil_tlv_table_mod *ttm, |
143 | const struct tun_table *old_tab, | |
144 | struct tun_table **new_tab) | |
9558d2a5 | 145 | { |
4e548ad9 | 146 | struct ofputil_tlv_map *ofp_map; |
9558d2a5 JG |
147 | enum ofperr err = 0; |
148 | ||
4e548ad9 ML |
149 | switch (ttm->command) { |
150 | case NXTTMC_ADD: | |
8d8ab6c2 | 151 | *new_tab = tun_metadata_alloc(old_tab); |
9558d2a5 | 152 | |
4e548ad9 | 153 | LIST_FOR_EACH (ofp_map, list_node, &ttm->mappings) { |
8d8ab6c2 | 154 | err = tun_metadata_add_entry(*new_tab, ofp_map->index, |
9558d2a5 JG |
155 | ofp_map->option_class, |
156 | ofp_map->option_type, | |
157 | ofp_map->option_len); | |
158 | if (err) { | |
8d8ab6c2 | 159 | tun_metadata_free(*new_tab); |
dea2dd0f | 160 | *new_tab = NULL; |
8d8ab6c2 | 161 | return err; |
9558d2a5 JG |
162 | } |
163 | } | |
164 | break; | |
165 | ||
4e548ad9 | 166 | case NXTTMC_DELETE: |
8d8ab6c2 | 167 | *new_tab = tun_metadata_alloc(old_tab); |
9558d2a5 | 168 | |
4e548ad9 | 169 | LIST_FOR_EACH (ofp_map, list_node, &ttm->mappings) { |
8d8ab6c2 | 170 | tun_metadata_del_entry(*new_tab, ofp_map->index); |
9558d2a5 JG |
171 | } |
172 | break; | |
173 | ||
4e548ad9 | 174 | case NXTTMC_CLEAR: |
8d8ab6c2 | 175 | *new_tab = tun_metadata_alloc(NULL); |
9558d2a5 JG |
176 | break; |
177 | ||
178 | default: | |
179 | OVS_NOT_REACHED(); | |
180 | } | |
181 | ||
8d8ab6c2 | 182 | return 0; |
9558d2a5 JG |
183 | } |
184 | ||
185 | void | |
8d8ab6c2 JG |
186 | tun_metadata_table_request(const struct tun_table *tun_table, |
187 | struct ofputil_tlv_table_reply *ttr) | |
9558d2a5 | 188 | { |
9558d2a5 JG |
189 | int i; |
190 | ||
4e548ad9 ML |
191 | ttr->max_option_space = TUN_METADATA_TOT_OPT_SIZE; |
192 | ttr->max_fields = TUN_METADATA_NUM_OPTS; | |
417e7e66 | 193 | ovs_list_init(&ttr->mappings); |
9558d2a5 JG |
194 | |
195 | for (i = 0; i < TUN_METADATA_NUM_OPTS; i++) { | |
8d8ab6c2 | 196 | const struct tun_meta_entry *entry = &tun_table->entries[i]; |
4e548ad9 | 197 | struct ofputil_tlv_map *map; |
9558d2a5 JG |
198 | |
199 | if (!entry->valid) { | |
200 | continue; | |
201 | } | |
202 | ||
203 | map = xmalloc(sizeof *map); | |
204 | map->option_class = ntohs(tun_key_class(entry->key)); | |
205 | map->option_type = tun_key_type(entry->key); | |
206 | map->option_len = entry->loc.len; | |
207 | map->index = i; | |
208 | ||
417e7e66 | 209 | ovs_list_push_back(&ttr->mappings, &map->list_node); |
9558d2a5 JG |
210 | } |
211 | } | |
212 | ||
6728d578 | 213 | /* Copies the value of field 'mf' from 'tnl' (which must be in non-UDPIF format) * into 'value'. |
9558d2a5 JG |
214 | * |
215 | * 'mf' must be an MFF_TUN_METADATA* field. | |
216 | * | |
8d8ab6c2 JG |
217 | * This uses the tunnel metadata mapping table created by tun_metadata_alloc(). |
218 | * If no such table has been created or if 'mf' hasn't been allocated in it yet, | |
219 | * this just zeros 'value'. */ | |
9558d2a5 | 220 | void |
6728d578 | 221 | tun_metadata_read(const struct flow_tnl *tnl, |
9558d2a5 JG |
222 | const struct mf_field *mf, union mf_value *value) |
223 | { | |
8d8ab6c2 | 224 | const struct tun_table *map = tnl->metadata.tab; |
9558d2a5 | 225 | unsigned int idx = mf->id - MFF_TUN_METADATA0; |
8d8ab6c2 | 226 | const struct tun_metadata_loc *loc; |
9558d2a5 JG |
227 | |
228 | if (!map) { | |
229 | memset(value->tun_metadata, 0, mf->n_bytes); | |
230 | return; | |
231 | } | |
232 | ||
233 | loc = &map->entries[idx].loc; | |
234 | ||
235 | memset(value->tun_metadata, 0, mf->n_bytes - loc->len); | |
236 | memcpy_from_metadata(value->tun_metadata + mf->n_bytes - loc->len, | |
6728d578 | 237 | &tnl->metadata, loc); |
9558d2a5 JG |
238 | } |
239 | ||
6728d578 | 240 | /* Copies 'value' into field 'mf' in 'tnl' (in non-UDPIF format). |
9558d2a5 JG |
241 | * |
242 | * 'mf' must be an MFF_TUN_METADATA* field. | |
243 | * | |
8d8ab6c2 JG |
244 | * This uses the tunnel metadata mapping table created by tun_metadata_alloc(). |
245 | * If no such table has been created or if 'mf' hasn't been allocated in it yet, | |
246 | * this function does nothing. */ | |
9558d2a5 | 247 | void |
6728d578 | 248 | tun_metadata_write(struct flow_tnl *tnl, |
9558d2a5 JG |
249 | const struct mf_field *mf, const union mf_value *value) |
250 | { | |
8d8ab6c2 | 251 | const struct tun_table *map = tnl->metadata.tab; |
9558d2a5 | 252 | unsigned int idx = mf->id - MFF_TUN_METADATA0; |
8d8ab6c2 | 253 | const struct tun_metadata_loc *loc; |
9558d2a5 JG |
254 | |
255 | if (!map || !map->entries[idx].valid) { | |
256 | return; | |
257 | } | |
258 | ||
259 | loc = &map->entries[idx].loc; | |
6728d578 | 260 | memcpy_to_metadata(&tnl->metadata, |
2234a7db | 261 | value->tun_metadata + mf->n_bytes - loc->len, loc, idx); |
9558d2a5 JG |
262 | } |
263 | ||
81f71381 YHW |
264 | /* Deletes field 'mf' in 'tnl' (in non-UDPIF format). |
265 | * 'mf' must be an MFF_TUN_METADATA* field. | |
266 | */ | |
267 | void | |
268 | tun_metadata_delete(struct flow_tnl *tnl, const struct mf_field *mf) | |
269 | { | |
270 | unsigned int idx; | |
271 | ||
272 | if (tnl->flags & FLOW_TNL_F_UDPIF) { | |
273 | return; | |
274 | } | |
275 | ||
276 | idx = mf->id - MFF_TUN_METADATA0; | |
277 | ovs_assert(idx < TUN_METADATA_NUM_OPTS); | |
278 | ULLONG_SET0(tnl->metadata.present.map, idx); | |
279 | } | |
280 | ||
9558d2a5 | 281 | static const struct tun_metadata_loc * |
8d8ab6c2 | 282 | metadata_loc_from_match(const struct tun_table *map, struct match *match, |
4f7b100c JG |
283 | const char *name, unsigned int idx, |
284 | unsigned int field_len, bool masked, char **err_str) | |
9558d2a5 JG |
285 | { |
286 | ovs_assert(idx < TUN_METADATA_NUM_OPTS); | |
287 | ||
4f7b100c JG |
288 | if (err_str) { |
289 | *err_str = NULL; | |
290 | } | |
291 | ||
9558d2a5 JG |
292 | if (map) { |
293 | if (map->entries[idx].valid) { | |
294 | return &map->entries[idx].loc; | |
295 | } else { | |
296 | return NULL; | |
297 | } | |
298 | } | |
299 | ||
4f7b100c JG |
300 | if (match->tun_md.alloc_offset + field_len > TUN_METADATA_TOT_OPT_SIZE) { |
301 | if (err_str) { | |
302 | *err_str = xasprintf("field %s exceeds maximum size for tunnel " | |
303 | "metadata (used %d, max %d)", name, | |
304 | match->tun_md.alloc_offset + field_len, | |
305 | TUN_METADATA_TOT_OPT_SIZE); | |
306 | } | |
307 | ||
308 | return NULL; | |
309 | } | |
310 | ||
311 | if (ULLONG_GET(match->wc.masks.tunnel.metadata.present.map, idx)) { | |
312 | if (err_str) { | |
313 | *err_str = xasprintf("field %s set multiple times", name); | |
314 | } | |
315 | ||
9558d2a5 JG |
316 | return NULL; |
317 | } | |
318 | ||
1cb20095 JG |
319 | match->tun_md.entry[idx].loc.len = field_len; |
320 | match->tun_md.entry[idx].loc.c.offset = match->tun_md.alloc_offset; | |
321 | match->tun_md.entry[idx].loc.c.len = field_len; | |
322 | match->tun_md.entry[idx].loc.c.next = NULL; | |
323 | match->tun_md.entry[idx].masked = masked; | |
9558d2a5 JG |
324 | match->tun_md.alloc_offset += field_len; |
325 | match->tun_md.valid = true; | |
326 | ||
1cb20095 | 327 | return &match->tun_md.entry[idx].loc; |
9558d2a5 JG |
328 | } |
329 | ||
330 | /* Makes 'match' match 'value'/'mask' on field 'mf'. | |
331 | * | |
6728d578 | 332 | * 'mf' must be an MFF_TUN_METADATA* field. 'match' must be in non-UDPIF format. |
9558d2a5 | 333 | * |
8d8ab6c2 JG |
334 | * If there is a tunnel metadata mapping table associated with the switch, |
335 | * this function is effective only if there is already a mapping for 'mf'. | |
336 | * Otherwise, the metadata mapping table integrated into 'match' is used, | |
337 | * adding 'mf' to its mapping table if it isn't already mapped (and if there | |
338 | * is room). If 'mf' isn't or can't be mapped, this function returns without | |
339 | * modifying 'match'. | |
9558d2a5 JG |
340 | * |
341 | * 'value' may be NULL; if so, then 'mf' is made to match on an all-zeros | |
342 | * value. | |
343 | * | |
344 | * 'mask' may be NULL; if so, then 'mf' is made exact-match. | |
4f7b100c JG |
345 | * |
346 | * If non-NULL, 'err_str' returns a malloc'ed string describing any errors | |
347 | * with the request or NULL if there is no error. The caller is reponsible | |
348 | * for freeing the string. | |
9558d2a5 JG |
349 | */ |
350 | void | |
351 | tun_metadata_set_match(const struct mf_field *mf, const union mf_value *value, | |
4f7b100c JG |
352 | const union mf_value *mask, struct match *match, |
353 | char **err_str) | |
9558d2a5 | 354 | { |
8d8ab6c2 | 355 | const struct tun_table *map = match->flow.tunnel.metadata.tab; |
9558d2a5 JG |
356 | const struct tun_metadata_loc *loc; |
357 | unsigned int idx = mf->id - MFF_TUN_METADATA0; | |
358 | unsigned int field_len; | |
1cb20095 | 359 | bool is_masked; |
9558d2a5 JG |
360 | unsigned int data_offset; |
361 | union mf_value data; | |
362 | ||
1cb20095 | 363 | field_len = mf_field_len(mf, value, mask, &is_masked); |
4f7b100c JG |
364 | loc = metadata_loc_from_match(map, match, mf->name, idx, field_len, |
365 | is_masked, err_str); | |
9558d2a5 JG |
366 | if (!loc) { |
367 | return; | |
368 | } | |
369 | ||
370 | data_offset = mf->n_bytes - loc->len; | |
371 | ||
372 | if (!value) { | |
373 | memset(data.tun_metadata, 0, loc->len); | |
374 | } else if (!mask) { | |
375 | memcpy(data.tun_metadata, value->tun_metadata + data_offset, loc->len); | |
376 | } else { | |
377 | int i; | |
378 | for (i = 0; i < loc->len; i++) { | |
379 | data.tun_metadata[i] = value->tun_metadata[data_offset + i] & | |
380 | mask->tun_metadata[data_offset + i]; | |
381 | } | |
382 | } | |
2234a7db JG |
383 | memcpy_to_metadata(&match->flow.tunnel.metadata, data.tun_metadata, |
384 | loc, idx); | |
9558d2a5 JG |
385 | |
386 | if (!value) { | |
387 | memset(data.tun_metadata, 0, loc->len); | |
388 | } else if (!mask) { | |
389 | memset(data.tun_metadata, 0xff, loc->len); | |
390 | } else { | |
391 | memcpy(data.tun_metadata, mask->tun_metadata + data_offset, loc->len); | |
392 | } | |
2234a7db JG |
393 | memcpy_to_metadata(&match->wc.masks.tunnel.metadata, data.tun_metadata, |
394 | loc, idx); | |
9558d2a5 JG |
395 | } |
396 | ||
8d8ab6c2 JG |
397 | /* Copies all MFF_TUN_METADATA* fields from 'tnl' to 'flow_metadata'. This |
398 | * is called during action translation and therefore 'tnl' must be in | |
399 | * non-udpif format. */ | |
9558d2a5 | 400 | void |
6728d578 | 401 | tun_metadata_get_fmd(const struct flow_tnl *tnl, struct match *flow_metadata) |
9558d2a5 | 402 | { |
9558d2a5 JG |
403 | int i; |
404 | ||
8d8ab6c2 | 405 | ULLONG_FOR_EACH_1 (i, tnl->metadata.present.map) { |
9558d2a5 | 406 | union mf_value opts; |
8d8ab6c2 | 407 | const struct tun_metadata_loc *old_loc = &tnl->metadata.tab->entries[i].loc; |
9558d2a5 JG |
408 | const struct tun_metadata_loc *new_loc; |
409 | ||
4f7b100c JG |
410 | new_loc = metadata_loc_from_match(NULL, flow_metadata, NULL, i, |
411 | old_loc->len, false, NULL); | |
9558d2a5 | 412 | |
8d8ab6c2 | 413 | memcpy_from_metadata(opts.tun_metadata, &tnl->metadata, old_loc); |
9558d2a5 | 414 | memcpy_to_metadata(&flow_metadata->flow.tunnel.metadata, |
2234a7db | 415 | opts.tun_metadata, new_loc, i); |
9558d2a5 JG |
416 | |
417 | memset(opts.tun_metadata, 0xff, old_loc->len); | |
418 | memcpy_to_metadata(&flow_metadata->wc.masks.tunnel.metadata, | |
2234a7db | 419 | opts.tun_metadata, new_loc, i); |
9558d2a5 JG |
420 | } |
421 | } | |
422 | ||
423 | static uint32_t | |
424 | tun_meta_hash(uint32_t key) | |
425 | { | |
426 | return hash_int(key, 0); | |
427 | } | |
428 | ||
429 | static struct tun_meta_entry * | |
430 | tun_meta_find_key(const struct hmap *hmap, uint32_t key) | |
431 | { | |
432 | struct tun_meta_entry *entry; | |
433 | ||
434 | HMAP_FOR_EACH_IN_BUCKET (entry, node, tun_meta_hash(key), hmap) { | |
435 | if (entry->key == key) { | |
436 | return entry; | |
437 | } | |
438 | } | |
439 | return NULL; | |
440 | } | |
441 | ||
442 | static void | |
443 | memcpy_to_metadata(struct tun_metadata *dst, const void *src, | |
2234a7db | 444 | const struct tun_metadata_loc *loc, unsigned int idx) |
9558d2a5 JG |
445 | { |
446 | const struct tun_metadata_loc_chain *chain = &loc->c; | |
447 | int addr = 0; | |
448 | ||
449 | while (chain) { | |
abdf31d1 | 450 | memcpy(dst->opts.u8 + chain->offset, (uint8_t *)src + addr, |
9558d2a5 JG |
451 | chain->len); |
452 | addr += chain->len; | |
453 | chain = chain->next; | |
454 | } | |
2234a7db JG |
455 | |
456 | ULLONG_SET1(dst->present.map, idx); | |
9558d2a5 JG |
457 | } |
458 | ||
459 | static void | |
460 | memcpy_from_metadata(void *dst, const struct tun_metadata *src, | |
461 | const struct tun_metadata_loc *loc) | |
462 | { | |
463 | const struct tun_metadata_loc_chain *chain = &loc->c; | |
464 | int addr = 0; | |
465 | ||
466 | while (chain) { | |
abdf31d1 | 467 | memcpy((uint8_t *)dst + addr, src->opts.u8 + chain->offset, |
9558d2a5 JG |
468 | chain->len); |
469 | addr += chain->len; | |
470 | chain = chain->next; | |
471 | } | |
472 | } | |
473 | ||
474 | static int | |
475 | tun_metadata_alloc_chain(struct tun_table *map, uint8_t len, | |
476 | struct tun_metadata_loc_chain *loc) | |
9558d2a5 JG |
477 | { |
478 | int alloc_len = len / 4; | |
479 | int scan_start = 0; | |
480 | int scan_end = TUN_METADATA_TOT_OPT_SIZE / 4; | |
481 | int pos_start, pos_end, pos_len; | |
482 | int best_start = 0, best_len = 0; | |
483 | ||
484 | while (true) { | |
485 | pos_start = bitmap_scan(map->alloc_map, 0, scan_start, scan_end); | |
486 | if (pos_start == scan_end) { | |
487 | break; | |
488 | } | |
489 | ||
490 | pos_end = bitmap_scan(map->alloc_map, 1, pos_start, | |
491 | MIN(pos_start + alloc_len, scan_end)); | |
492 | pos_len = pos_end - pos_start; | |
493 | if (pos_len == alloc_len) { | |
494 | goto found; | |
495 | } | |
496 | ||
497 | if (pos_len > best_len) { | |
498 | best_start = pos_start; | |
499 | best_len = pos_len; | |
500 | } | |
501 | scan_start = pos_end + 1; | |
502 | } | |
503 | ||
504 | if (best_len == 0) { | |
505 | return ENOSPC; | |
506 | } | |
507 | ||
508 | pos_start = best_start; | |
509 | pos_len = best_len; | |
510 | ||
511 | found: | |
512 | bitmap_set_multiple(map->alloc_map, pos_start, pos_len, 1); | |
513 | loc->offset = pos_start * 4; | |
514 | loc->len = pos_len * 4; | |
515 | ||
516 | return 0; | |
517 | } | |
518 | ||
519 | static enum ofperr | |
520 | tun_metadata_add_entry(struct tun_table *map, uint8_t idx, uint16_t opt_class, | |
8d8ab6c2 | 521 | uint8_t type, uint8_t len) |
9558d2a5 JG |
522 | { |
523 | struct tun_meta_entry *entry; | |
524 | struct tun_metadata_loc_chain *cur_chain, *prev_chain; | |
525 | ||
526 | ovs_assert(idx < TUN_METADATA_NUM_OPTS); | |
527 | ||
528 | entry = &map->entries[idx]; | |
529 | if (entry->valid) { | |
4e548ad9 | 530 | return OFPERR_NXTTMFC_ALREADY_MAPPED; |
9558d2a5 JG |
531 | } |
532 | ||
533 | entry->key = tun_meta_key(htons(opt_class), type); | |
534 | if (tun_meta_find_key(&map->key_hmap, entry->key)) { | |
4e548ad9 | 535 | return OFPERR_NXTTMFC_DUP_ENTRY; |
9558d2a5 JG |
536 | } |
537 | ||
538 | entry->valid = true; | |
539 | hmap_insert(&map->key_hmap, &entry->node, | |
540 | tun_meta_hash(entry->key)); | |
541 | ||
542 | entry->loc.len = len; | |
543 | cur_chain = &entry->loc.c; | |
544 | memset(cur_chain, 0, sizeof *cur_chain); | |
545 | prev_chain = NULL; | |
546 | ||
547 | while (len) { | |
548 | int err; | |
549 | ||
550 | if (!cur_chain) { | |
551 | cur_chain = xzalloc(sizeof *cur_chain); | |
482553cc | 552 | prev_chain->next = cur_chain; |
9558d2a5 JG |
553 | } |
554 | ||
555 | err = tun_metadata_alloc_chain(map, len, cur_chain); | |
556 | if (err) { | |
557 | tun_metadata_del_entry(map, idx); | |
4e548ad9 | 558 | return OFPERR_NXTTMFC_TABLE_FULL; |
9558d2a5 JG |
559 | } |
560 | ||
561 | len -= cur_chain->len; | |
562 | ||
9558d2a5 JG |
563 | prev_chain = cur_chain; |
564 | cur_chain = NULL; | |
565 | } | |
566 | ||
567 | return 0; | |
568 | } | |
569 | ||
570 | static void | |
571 | tun_metadata_del_entry(struct tun_table *map, uint8_t idx) | |
9558d2a5 JG |
572 | { |
573 | struct tun_meta_entry *entry; | |
574 | struct tun_metadata_loc_chain *chain; | |
575 | ||
576 | if (idx >= TUN_METADATA_NUM_OPTS) { | |
577 | return; | |
578 | } | |
579 | ||
580 | entry = &map->entries[idx]; | |
581 | if (!entry->valid) { | |
582 | return; | |
583 | } | |
584 | ||
585 | chain = &entry->loc.c; | |
586 | while (chain) { | |
587 | struct tun_metadata_loc_chain *next = chain->next; | |
588 | ||
589 | bitmap_set_multiple(map->alloc_map, chain->offset / 4, | |
590 | chain->len / 4, 0); | |
591 | if (chain != &entry->loc.c) { | |
592 | free(chain); | |
593 | } | |
594 | chain = next; | |
595 | } | |
596 | ||
597 | entry->valid = false; | |
598 | hmap_remove(&map->key_hmap, &entry->node); | |
599 | memset(&entry->loc, 0, sizeof entry->loc); | |
600 | } | |
601 | ||
8d8ab6c2 JG |
602 | /* Converts from Geneve netlink attributes in 'attr' to tunnel metadata |
603 | * in 'tun'. In reality, there is very little conversion done since we are | |
604 | * just copying over the tunnel options in the form that they were received | |
605 | * on the wire. By always using UDPIF format, this allows us to process the | |
606 | * flow key without any knowledge of the mapping table. We can do the | |
607 | * conversion later if necessary. */ | |
608 | void | |
609 | tun_metadata_from_geneve_nlattr(const struct nlattr *attr, bool is_mask, | |
610 | struct flow_tnl *tun) | |
9558d2a5 | 611 | { |
8d8ab6c2 JG |
612 | int attr_len = nl_attr_get_size(attr); |
613 | ||
614 | memcpy(tun->metadata.opts.gnv, nl_attr_get(attr), attr_len); | |
615 | tun->flags |= FLOW_TNL_F_UDPIF; | |
6728d578 JG |
616 | |
617 | if (!is_mask) { | |
8d8ab6c2 | 618 | tun->metadata.present.len = attr_len; |
6728d578 | 619 | } else { |
8d8ab6c2 JG |
620 | /* We need to exact match on the length so we don't |
621 | * accidentally match on sets of options that are the same | |
622 | * at the beginning but with additional options after. */ | |
623 | tun->metadata.present.len = 0xff; | |
6728d578 | 624 | } |
8d8ab6c2 | 625 | } |
6728d578 | 626 | |
8d8ab6c2 JG |
627 | /* Converts from the flat Geneve options representation extracted directly |
628 | * from the tunnel header to the representation that maps options to | |
629 | * pre-allocated locations. The original version (in UDPIF form) is passed | |
630 | * in 'src' and the translated form in stored in 'dst'. To handle masks, the | |
631 | * flow must also be passed in through 'flow' (in the original, raw form). */ | |
632 | int | |
633 | tun_metadata_from_geneve_udpif(const struct tun_table *tun_tab, | |
634 | const struct flow_tnl *flow, | |
635 | const struct flow_tnl *src, | |
636 | struct flow_tnl *dst) | |
637 | { | |
638 | const struct geneve_opt *opt = src->metadata.opts.gnv; | |
639 | const struct geneve_opt *flow_opt = flow->metadata.opts.gnv; | |
640 | int opts_len = flow->metadata.present.len; | |
641 | ||
642 | dst->metadata.tab = tun_tab; | |
643 | dst->flags = src->flags & ~FLOW_TNL_F_UDPIF; | |
644 | dst->metadata.present.map = 0; | |
9558d2a5 | 645 | |
9558d2a5 JG |
646 | while (opts_len > 0) { |
647 | int len; | |
648 | struct tun_meta_entry *entry; | |
649 | ||
650 | if (opts_len < sizeof(*opt)) { | |
651 | return EINVAL; | |
652 | } | |
653 | ||
654 | len = sizeof(*opt) + flow_opt->length * 4; | |
655 | if (len > opts_len) { | |
656 | return EINVAL; | |
657 | } | |
658 | ||
8d8ab6c2 | 659 | entry = tun_meta_find_key(&tun_tab->key_hmap, |
9558d2a5 JG |
660 | tun_meta_key(flow_opt->opt_class, |
661 | flow_opt->type)); | |
662 | if (entry) { | |
663 | if (entry->loc.len == flow_opt->length * 4) { | |
8d8ab6c2 JG |
664 | memcpy_to_metadata(&dst->metadata, opt + 1, &entry->loc, |
665 | entry - tun_tab->entries); | |
9558d2a5 JG |
666 | } else { |
667 | return EINVAL; | |
668 | } | |
669 | } else if (flow_opt->type & GENEVE_CRIT_OPT_TYPE) { | |
670 | return EINVAL; | |
671 | } | |
672 | ||
673 | opt = opt + len / sizeof(*opt); | |
674 | flow_opt = flow_opt + len / sizeof(*opt); | |
675 | opts_len -= len; | |
676 | } | |
677 | ||
678 | return 0; | |
679 | } | |
680 | ||
5bb08b0e JG |
681 | static void |
682 | tun_metadata_to_geneve__(const struct tun_metadata *flow, struct ofpbuf *b, | |
683 | bool *crit_opt) | |
684 | { | |
5bb08b0e JG |
685 | int i; |
686 | ||
5bb08b0e | 687 | *crit_opt = false; |
9558d2a5 | 688 | |
6728d578 | 689 | ULLONG_FOR_EACH_1 (i, flow->present.map) { |
8d8ab6c2 | 690 | const struct tun_meta_entry *entry = &flow->tab->entries[i]; |
9558d2a5 JG |
691 | struct geneve_opt *opt; |
692 | ||
693 | opt = ofpbuf_put_uninit(b, sizeof *opt + entry->loc.len); | |
694 | ||
695 | opt->opt_class = tun_key_class(entry->key); | |
696 | opt->type = tun_key_type(entry->key); | |
697 | opt->length = entry->loc.len / 4; | |
698 | opt->r1 = 0; | |
699 | opt->r2 = 0; | |
700 | opt->r3 = 0; | |
701 | ||
702 | memcpy_from_metadata(opt + 1, flow, &entry->loc); | |
5bb08b0e | 703 | *crit_opt |= !!(opt->type & GENEVE_CRIT_OPT_TYPE); |
9558d2a5 | 704 | } |
5bb08b0e JG |
705 | } |
706 | ||
6728d578 JG |
707 | static void |
708 | tun_metadata_to_geneve_nlattr_flow(const struct flow_tnl *flow, | |
5bb08b0e JG |
709 | struct ofpbuf *b) |
710 | { | |
711 | size_t nlattr_offset; | |
712 | bool crit_opt; | |
713 | ||
6728d578 | 714 | if (!flow->metadata.present.map) { |
5bb08b0e JG |
715 | return; |
716 | } | |
717 | ||
718 | /* For all intents and purposes, the Geneve options are nested | |
719 | * attributes even if this doesn't show up directly to netlink. It's | |
720 | * similar enough that we can use the same mechanism. */ | |
721 | nlattr_offset = nl_msg_start_nested(b, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS); | |
722 | ||
6728d578 | 723 | tun_metadata_to_geneve__(&flow->metadata, b, &crit_opt); |
9558d2a5 JG |
724 | |
725 | nl_msg_end_nested(b, nlattr_offset); | |
726 | } | |
727 | ||
6728d578 JG |
728 | /* Converts from processed tunnel metadata information (in non-udpif |
729 | * format) in 'flow' to a stream of Geneve options suitable for | |
730 | * transmission in 'opts'. Additionally returns whether there were | |
731 | * any critical options in 'crit_opt' as well as the total length of | |
732 | * data. */ | |
5bb08b0e | 733 | int |
6728d578 | 734 | tun_metadata_to_geneve_header(const struct flow_tnl *flow, |
5bb08b0e JG |
735 | struct geneve_opt *opts, bool *crit_opt) |
736 | { | |
737 | struct ofpbuf b; | |
738 | ||
4e548ad9 | 739 | ofpbuf_use_stack(&b, opts, TLV_TOT_OPT_SIZE); |
6728d578 | 740 | tun_metadata_to_geneve__(&flow->metadata, &b, crit_opt); |
5bb08b0e JG |
741 | |
742 | return b.size; | |
743 | } | |
744 | ||
6728d578 JG |
745 | static void |
746 | tun_metadata_to_geneve_mask__(const struct tun_metadata *flow, | |
747 | const struct tun_metadata *mask, | |
748 | struct geneve_opt *opt, int opts_len) | |
9558d2a5 | 749 | { |
9558d2a5 JG |
750 | /* All of these options have already been validated, so no need |
751 | * for sanity checking. */ | |
9558d2a5 JG |
752 | while (opts_len > 0) { |
753 | struct tun_meta_entry *entry; | |
754 | int len = sizeof(*opt) + opt->length * 4; | |
755 | ||
8d8ab6c2 | 756 | entry = tun_meta_find_key(&flow->tab->key_hmap, |
9558d2a5 JG |
757 | tun_meta_key(opt->opt_class, opt->type)); |
758 | if (entry) { | |
759 | memcpy_from_metadata(opt + 1, mask, &entry->loc); | |
760 | } else { | |
761 | memset(opt + 1, 0, opt->length * 4); | |
762 | } | |
763 | ||
764 | opt->opt_class = htons(0xffff); | |
765 | opt->type = 0xff; | |
766 | opt->length = 0x1f; | |
767 | opt->r1 = 0; | |
768 | opt->r2 = 0; | |
769 | opt->r3 = 0; | |
770 | ||
771 | opt = opt + len / sizeof(*opt); | |
772 | opts_len -= len; | |
773 | } | |
774 | } | |
775 | ||
6728d578 JG |
776 | static void |
777 | tun_metadata_to_geneve_nlattr_mask(const struct ofpbuf *key, | |
778 | const struct flow_tnl *mask, | |
779 | const struct flow_tnl *flow, | |
780 | struct ofpbuf *b) | |
781 | { | |
8d8ab6c2 | 782 | const struct nlattr *tnl_key, *geneve_key; |
6728d578 JG |
783 | struct nlattr *geneve_mask; |
784 | struct geneve_opt *opt; | |
785 | int opts_len; | |
786 | ||
787 | if (!key) { | |
788 | return; | |
789 | } | |
790 | ||
8d8ab6c2 JG |
791 | tnl_key = nl_attr_find__(key->data, key->size, OVS_KEY_ATTR_TUNNEL); |
792 | if (!tnl_key) { | |
793 | return; | |
794 | } | |
795 | ||
796 | geneve_key = nl_attr_find_nested(tnl_key, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS); | |
6728d578 JG |
797 | if (!geneve_key) { |
798 | return; | |
799 | } | |
800 | ||
801 | geneve_mask = ofpbuf_tail(b); | |
802 | nl_msg_put(b, geneve_key, geneve_key->nla_len); | |
803 | ||
804 | opt = CONST_CAST(struct geneve_opt *, nl_attr_get(geneve_mask)); | |
805 | opts_len = nl_attr_get_size(geneve_mask); | |
806 | ||
807 | tun_metadata_to_geneve_mask__(&flow->metadata, &mask->metadata, | |
808 | opt, opts_len); | |
809 | } | |
810 | ||
811 | /* Convert from the tunnel metadata in 'tun' to netlink attributes stored | |
812 | * in 'b'. Either UDPIF or non-UDPIF input forms are accepted. | |
813 | * | |
814 | * To assist with parsing, it is necessary to also pass in the tunnel metadata | |
815 | * from the flow in 'flow' as well in the original netlink form of the flow in | |
816 | * 'key'. */ | |
817 | void | |
818 | tun_metadata_to_geneve_nlattr(const struct flow_tnl *tun, | |
819 | const struct flow_tnl *flow, | |
820 | const struct ofpbuf *key, | |
821 | struct ofpbuf *b) | |
822 | { | |
823 | bool is_mask = tun != flow; | |
824 | ||
825 | if (!(flow->flags & FLOW_TNL_F_UDPIF)) { | |
826 | if (!is_mask) { | |
827 | tun_metadata_to_geneve_nlattr_flow(tun, b); | |
828 | } else { | |
829 | tun_metadata_to_geneve_nlattr_mask(key, tun, flow, b); | |
830 | } | |
831 | } else if (flow->metadata.present.len || is_mask) { | |
832 | nl_msg_put_unspec(b, OVS_TUNNEL_KEY_ATTR_GENEVE_OPTS, | |
833 | tun->metadata.opts.gnv, | |
834 | flow->metadata.present.len); | |
835 | } | |
836 | } | |
837 | ||
838 | /* Converts 'mask_src' (in non-UDPIF format) to a series of masked options in | |
839 | * 'dst'. 'flow_src' (also in non-UDPIF format) and the original set of | |
840 | * options 'flow_src_opt'/'opts_len' are needed as a guide to interpret the | |
841 | * mask data. */ | |
842 | void | |
843 | tun_metadata_to_geneve_udpif_mask(const struct flow_tnl *flow_src, | |
844 | const struct flow_tnl *mask_src, | |
845 | const struct geneve_opt *flow_src_opt, | |
846 | int opts_len, struct geneve_opt *dst) | |
847 | { | |
6728d578 JG |
848 | memcpy(dst, flow_src_opt, opts_len); |
849 | tun_metadata_to_geneve_mask__(&flow_src->metadata, | |
850 | &mask_src->metadata, dst, opts_len); | |
851 | } | |
852 | ||
9558d2a5 | 853 | static const struct tun_metadata_loc * |
8d8ab6c2 JG |
854 | metadata_loc_from_match_read(const struct tun_table *map, |
855 | const struct match *match, unsigned int idx, | |
856 | const struct flow_tnl *mask, bool *is_masked) | |
9558d2a5 | 857 | { |
1cb20095 JG |
858 | union mf_value mask_opts; |
859 | ||
9558d2a5 | 860 | if (match->tun_md.valid) { |
1cb20095 JG |
861 | *is_masked = match->tun_md.entry[idx].masked; |
862 | return &match->tun_md.entry[idx].loc; | |
9558d2a5 JG |
863 | } |
864 | ||
1cb20095 JG |
865 | memcpy_from_metadata(mask_opts.tun_metadata, &mask->metadata, |
866 | &map->entries[idx].loc); | |
867 | ||
868 | *is_masked = map->entries[idx].loc.len == 0 || | |
869 | !is_all_ones(mask_opts.tun_metadata, | |
870 | map->entries[idx].loc.len); | |
9558d2a5 JG |
871 | return &map->entries[idx].loc; |
872 | } | |
873 | ||
8d8ab6c2 JG |
874 | /* Generates NXM formatted matches in 'b' based on the contents of 'match'. |
875 | * 'match' must be in non-udpif format. */ | |
9558d2a5 JG |
876 | void |
877 | tun_metadata_to_nx_match(struct ofpbuf *b, enum ofp_version oxm, | |
878 | const struct match *match) | |
879 | { | |
9558d2a5 JG |
880 | int i; |
881 | ||
8d8ab6c2 | 882 | ULLONG_FOR_EACH_1 (i, match->wc.masks.tunnel.metadata.present.map) { |
9558d2a5 | 883 | const struct tun_metadata_loc *loc; |
1cb20095 | 884 | bool is_masked; |
9558d2a5 JG |
885 | union mf_value opts; |
886 | union mf_value mask_opts; | |
887 | ||
8d8ab6c2 JG |
888 | loc = metadata_loc_from_match_read(match->flow.tunnel.metadata.tab, |
889 | match, i, &match->wc.masks.tunnel, | |
890 | &is_masked); | |
891 | memcpy_from_metadata(opts.tun_metadata, &match->flow.tunnel.metadata, | |
892 | loc); | |
893 | memcpy_from_metadata(mask_opts.tun_metadata, | |
894 | &match->wc.masks.tunnel.metadata, loc); | |
be7ac2f3 BP |
895 | nxm_put_entry_raw(b, MFF_TUN_METADATA0 + i, oxm, opts.tun_metadata, |
896 | is_masked ? mask_opts.tun_metadata : NULL, loc->len); | |
9558d2a5 JG |
897 | } |
898 | } | |
899 | ||
8d8ab6c2 JG |
900 | /* Formatted matches in 's' based on the contents of 'match'. 'match' must be |
901 | * in non-udpif format. */ | |
9558d2a5 JG |
902 | void |
903 | tun_metadata_match_format(struct ds *s, const struct match *match) | |
904 | { | |
8d8ab6c2 | 905 | int i; |
9558d2a5 | 906 | |
8d8ab6c2 JG |
907 | if (match->flow.tunnel.flags & FLOW_TNL_F_UDPIF || |
908 | (!match->flow.tunnel.metadata.tab && !match->tun_md.valid)) { | |
6728d578 JG |
909 | return; |
910 | } | |
911 | ||
8d8ab6c2 | 912 | ULLONG_FOR_EACH_1 (i, match->wc.masks.tunnel.metadata.present.map) { |
9558d2a5 | 913 | const struct tun_metadata_loc *loc; |
1cb20095 JG |
914 | bool is_masked; |
915 | union mf_value opts, mask_opts; | |
9558d2a5 | 916 | |
8d8ab6c2 JG |
917 | loc = metadata_loc_from_match_read(match->flow.tunnel.metadata.tab, |
918 | match, i, &match->wc.masks.tunnel, | |
919 | &is_masked); | |
9558d2a5 | 920 | |
1cb20095 | 921 | ds_put_format(s, "tun_metadata%u", i); |
8d8ab6c2 JG |
922 | memcpy_from_metadata(mask_opts.tun_metadata, |
923 | &match->wc.masks.tunnel.metadata, loc); | |
1cb20095 | 924 | |
8d8ab6c2 | 925 | if (!ULLONG_GET(match->flow.tunnel.metadata.present.map, i)) { |
1cb20095 JG |
926 | /* Indicate that we are matching on the field being not present. */ |
927 | ds_put_cstr(s, "=NP"); | |
928 | } else if (!(is_masked && | |
929 | is_all_zeros(mask_opts.tun_metadata, loc->len))) { | |
930 | ds_put_char(s, '='); | |
9558d2a5 | 931 | |
8d8ab6c2 JG |
932 | memcpy_from_metadata(opts.tun_metadata, |
933 | &match->flow.tunnel.metadata, loc); | |
9558d2a5 | 934 | ds_put_hex(s, opts.tun_metadata, loc->len); |
1cb20095 JG |
935 | |
936 | if (!is_all_ones(mask_opts.tun_metadata, loc->len)) { | |
937 | ds_put_char(s, '/'); | |
938 | ds_put_hex(s, mask_opts.tun_metadata, loc->len); | |
939 | } | |
9558d2a5 JG |
940 | } |
941 | ds_put_char(s, ','); | |
942 | } | |
943 | } | |
1f7270d7 BP |
944 | |
945 | struct tun_metadata_allocation * | |
946 | tun_metadata_allocation_clone(const struct tun_metadata_allocation *src) | |
947 | { | |
948 | return src && src->valid ? xmemdup(src, sizeof *src) : NULL; | |
949 | } | |
950 | ||
951 | void | |
952 | tun_metadata_allocation_copy(struct tun_metadata_allocation *dst, | |
953 | const struct tun_metadata_allocation *src) | |
954 | { | |
955 | if (src && src->valid) { | |
956 | *dst = *src; | |
957 | } else { | |
958 | memset(dst, 0, sizeof *dst); | |
959 | } | |
960 | } |