| 1 | |
| 2 | |
| 3 | * Copyright (C) 2012 Red Hat, Inc.
|
| 4 | *
|
| 5 | * This file is released under the GPL.
|
| 6 |
|
| 7 |
|
| 8 | #include "dm-array.h"
|
| 9 | #include "dm-space-map.h"
|
| 10 | #include "dm-transaction-manager.h"
|
| 11 |
|
| 12 | #include <linux/export.h>
|
| 13 | #include <linux/device-mapper.h>
|
| 14 |
|
| 15 | #define DM_MSG_PREFIX "array"
|
| 16 |
|
| 17 |
|
| 18 |
|
| 19 | |
| 20 | * The array is implemented as a fully populated btree, which points to
|
| 21 | * blocks that contain the packed values. This is more space efficient
|
| 22 | * than just using a btree since we don't store 1 key per value.
|
| 23 |
|
| 24 | struct array_block {
|
| 25 | __le32 csum;
|
| 26 | __le32 max_entries;
|
| 27 | __le32 nr_entries;
|
| 28 | __le32 value_size;
|
| 29 | __le64 blocknr;
|
| 30 | } __packed;
|
| 31 |
|
| 32 |
|
| 33 |
|
| 34 | |
| 35 | * Validator methods. As usual we calculate a checksum, and also write the
|
| 36 | * block location into the header (paranoia about ssds remapping areas by
|
| 37 | * mistake).
|
| 38 |
|
| 39 | #define CSUM_XOR 595846735
|
| 40 |
|
| 41 | static void array_block_prepare_for_write(struct dm_block_validator *v,
|
| 42 | struct dm_block *b,
|
| 43 | size_t size_of_block)
|
| 44 | {
|
| 45 | struct array_block *bh_le = dm_block_data(b);
|
| 46 |
|
| 47 | bh_le->blocknr = cpu_to_le64(dm_block_location(b));
|
| 48 | bh_le->csum = cpu_to_le32(dm_bm_checksum(&bh_le->max_entries,
|
| 49 | size_of_block - sizeof(__le32),
|
| 50 | CSUM_XOR));
|
| 51 | }
|
| 52 |
|
| 53 | static int array_block_check(struct dm_block_validator *v,
|
| 54 | struct dm_block *b,
|
| 55 | size_t size_of_block)
|
| 56 | {
|
| 57 | struct array_block *bh_le = dm_block_data(b);
|
| 58 | __le32 csum_disk;
|
| 59 |
|
| 60 | if (dm_block_location(b) != le64_to_cpu(bh_le->blocknr)) {
|
| 61 | DMERR_LIMIT("%s failed: blocknr %llu != wanted %llu", __func__,
|
| 62 | (unsigned long long) le64_to_cpu(bh_le->blocknr),
|
| 63 | (unsigned long long) dm_block_location(b));
|
| 64 | return -ENOTBLK;
|
| 65 | }
|
| 66 |
|
| 67 | csum_disk = cpu_to_le32(dm_bm_checksum(&bh_le->max_entries,
|
| 68 | size_of_block - sizeof(__le32),
|
| 69 | CSUM_XOR));
|
| 70 | if (csum_disk != bh_le->csum) {
|
| 71 | DMERR_LIMIT("%s failed: csum %u != wanted %u", __func__,
|
| 72 | (unsigned int) le32_to_cpu(csum_disk),
|
| 73 | (unsigned int) le32_to_cpu(bh_le->csum));
|
| 74 | return -EILSEQ;
|
| 75 | }
|
| 76 |
|
| 77 | return 0;
|
| 78 | }
|
| 79 |
|
| 80 | static struct dm_block_validator array_validator = {
|
| 81 | .name = "array",
|
| 82 | .prepare_for_write = array_block_prepare_for_write,
|
| 83 | .check = array_block_check
|
| 84 | };
|
| 85 |
|
| 86 |
|
| 87 |
|
| 88 | |
| 89 | * Functions for manipulating the array blocks.
|
| 90 |
|
| 91 |
|
| 92 | |
| 93 | * Returns a pointer to a value within an array block.
|
| 94 | *
|
| 95 | * index - The index into _this_ specific block.
|
| 96 |
|
| 97 | static void *element_at(struct dm_array_info *info, struct array_block *ab,
|
| 98 | unsigned int index)
|
| 99 | {
|
| 100 | unsigned char *entry = (unsigned char *) (ab + 1);
|
| 101 |
|
| 102 | entry += index * info->value_type.size;
|
| 103 |
|
| 104 | return entry;
|
| 105 | }
|
| 106 |
|
| 107 | |
| 108 | * Utility function that calls one of the value_type methods on every value
|
| 109 | * in an array block.
|
| 110 |
|
| 111 | static void on_entries(struct dm_array_info *info, struct array_block *ab,
|
| 112 | void (*fn)(void *, const void *, unsigned int))
|
| 113 | {
|
| 114 | unsigned int nr_entries = le32_to_cpu(ab->nr_entries);
|
| 115 |
|
| 116 | fn(info->value_type.context, element_at(info, ab, 0), nr_entries);
|
| 117 | }
|
| 118 |
|
| 119 | |
| 120 | * Increment every value in an array block.
|
| 121 |
|
| 122 | static void inc_ablock_entries(struct dm_array_info *info, struct array_block *ab)
|
| 123 | {
|
| 124 | struct dm_btree_value_type *vt = &info->value_type;
|
| 125 |
|
| 126 | if (vt->inc)
|
| 127 | on_entries(info, ab, vt->inc);
|
| 128 | }
|
| 129 |
|
| 130 | |
| 131 | * Decrement every value in an array block.
|
| 132 |
|
| 133 | static void dec_ablock_entries(struct dm_array_info *info, struct array_block *ab)
|
| 134 | {
|
| 135 | struct dm_btree_value_type *vt = &info->value_type;
|
| 136 |
|
| 137 | if (vt->dec)
|
| 138 | on_entries(info, ab, vt->dec);
|
| 139 | }
|
| 140 |
|
| 141 | |
| 142 | * Each array block can hold this many values.
|
| 143 |
|
| 144 | static uint32_t calc_max_entries(size_t value_size, size_t size_of_block)
|
| 145 | {
|
| 146 | return (size_of_block - sizeof(struct array_block)) / value_size;
|
| 147 | }
|
| 148 |
|
| 149 | |
| 150 | * Allocate a new array block. The caller will need to unlock block.
|
| 151 |
|
| 152 | static int alloc_ablock(struct dm_array_info *info, size_t size_of_block,
|
| 153 | uint32_t max_entries,
|
| 154 | struct dm_block **block, struct array_block **ab)
|
| 155 | {
|
| 156 | int r;
|
| 157 |
|
| 158 | r = dm_tm_new_block(info->btree_info.tm, &array_validator, block);
|
| 159 | if (r)
|
| 160 | return r;
|
| 161 |
|
| 162 | (*ab) = dm_block_data(*block);
|
| 163 | (*ab)->max_entries = cpu_to_le32(max_entries);
|
| 164 | (*ab)->nr_entries = cpu_to_le32(0);
|
| 165 | (*ab)->value_size = cpu_to_le32(info->value_type.size);
|
| 166 |
|
| 167 | return 0;
|
| 168 | }
|
| 169 |
|
| 170 | |
| 171 | * Pad an array block out with a particular value. Every instance will
|
| 172 | * cause an increment of the value_type. new_nr must always be more than
|
| 173 | * the current number of entries.
|
| 174 |
|
| 175 | static void fill_ablock(struct dm_array_info *info, struct array_block *ab,
|
| 176 | const void *value, unsigned int new_nr)
|
| 177 | {
|
| 178 | uint32_t nr_entries, delta, i;
|
| 179 | struct dm_btree_value_type *vt = &info->value_type;
|
| 180 |
|
| 181 | BUG_ON(new_nr > le32_to_cpu(ab->max_entries));
|
| 182 | BUG_ON(new_nr < le32_to_cpu(ab->nr_entries));
|
| 183 |
|
| 184 | nr_entries = le32_to_cpu(ab->nr_entries);
|
| 185 | delta = new_nr - nr_entries;
|
| 186 | if (vt->inc)
|
| 187 | vt->inc(vt->context, value, delta);
|
| 188 | for (i = nr_entries; i < new_nr; i++)
|
| 189 | memcpy(element_at(info, ab, i), value, vt->size);
|
| 190 | ab->nr_entries = cpu_to_le32(new_nr);
|
| 191 | }
|
| 192 |
|
| 193 | |
| 194 | * Remove some entries from the back of an array block. Every value
|
| 195 | * removed will be decremented. new_nr must be <= the current number of
|
| 196 | * entries.
|
| 197 |
|
| 198 | static void trim_ablock(struct dm_array_info *info, struct array_block *ab,
|
| 199 | unsigned int new_nr)
|
| 200 | {
|
| 201 | uint32_t nr_entries, delta;
|
| 202 | struct dm_btree_value_type *vt = &info->value_type;
|
| 203 |
|
| 204 | BUG_ON(new_nr > le32_to_cpu(ab->max_entries));
|
| 205 | BUG_ON(new_nr > le32_to_cpu(ab->nr_entries));
|
| 206 |
|
| 207 | nr_entries = le32_to_cpu(ab->nr_entries);
|
| 208 | delta = nr_entries - new_nr;
|
| 209 | if (vt->dec)
|
| 210 | vt->dec(vt->context, element_at(info, ab, new_nr - 1), delta);
|
| 211 | ab->nr_entries = cpu_to_le32(new_nr);
|
| 212 | }
|
| 213 |
|
| 214 | |
| 215 | * Read locks a block, and coerces it to an array block. The caller must
|
| 216 | * unlock 'block' when finished.
|
| 217 |
|
| 218 | static int get_ablock(struct dm_array_info *info, dm_block_t b,
|
| 219 | struct dm_block **block, struct array_block **ab)
|
| 220 | {
|
| 221 | int r;
|
| 222 |
|
| 223 | r = dm_tm_read_lock(info->btree_info.tm, b, &array_validator, block);
|
| 224 | if (r)
|
| 225 | return r;
|
| 226 |
|
| 227 | *ab = dm_block_data(*block);
|
| 228 | return 0;
|
| 229 | }
|
| 230 |
|
| 231 | |
| 232 | * Unlocks an array block.
|
| 233 |
|
| 234 | static void unlock_ablock(struct dm_array_info *info, struct dm_block *block)
|
| 235 | {
|
| 236 | dm_tm_unlock(info->btree_info.tm, block);
|
| 237 | }
|
| 238 |
|
| 239 |
|
| 240 |
|
| 241 | |
| 242 | * Btree manipulation.
|
| 243 |
|
| 244 |
|
| 245 | |
| 246 | * Looks up an array block in the btree, and then read locks it.
|
| 247 | *
|
| 248 | * index is the index of the index of the array_block, (ie. the array index
|
| 249 | * / max_entries).
|
| 250 |
|
| 251 | static int lookup_ablock(struct dm_array_info *info, dm_block_t root,
|
| 252 | unsigned int index, struct dm_block **block,
|
| 253 | struct array_block **ab)
|
| 254 | {
|
| 255 | int r;
|
| 256 | uint64_t key = index;
|
| 257 | __le64 block_le;
|
| 258 |
|
| 259 | r = dm_btree_lookup(&info->btree_info, root, &key, &block_le);
|
| 260 | if (r)
|
| 261 | return r;
|
| 262 |
|
| 263 | return get_ablock(info, le64_to_cpu(block_le), block, ab);
|
| 264 | }
|
| 265 |
|
| 266 | |
| 267 | * Insert an array block into the btree. The block is _not_ unlocked.
|
| 268 |
|
| 269 | static int insert_ablock(struct dm_array_info *info, uint64_t index,
|
| 270 | struct dm_block *block, dm_block_t *root)
|
| 271 | {
|
| 272 | __le64 block_le = cpu_to_le64(dm_block_location(block));
|
| 273 |
|
| 274 | __dm_bless_for_disk(block_le);
|
| 275 | return dm_btree_insert(&info->btree_info, *root, &index, &block_le, root);
|
| 276 | }
|
| 277 |
|
| 278 |
|
| 279 |
|
| 280 | static int __shadow_ablock(struct dm_array_info *info, dm_block_t b,
|
| 281 | struct dm_block **block, struct array_block **ab)
|
| 282 | {
|
| 283 | int inc;
|
| 284 | int r = dm_tm_shadow_block(info->btree_info.tm, b,
|
| 285 | &array_validator, block, &inc);
|
| 286 | if (r)
|
| 287 | return r;
|
| 288 |
|
| 289 | *ab = dm_block_data(*block);
|
| 290 | if (inc)
|
| 291 | inc_ablock_entries(info, *ab);
|
| 292 |
|
| 293 | return 0;
|
| 294 | }
|
| 295 |
|
| 296 | |
| 297 | * The shadow op will often be a noop. Only insert if it really
|
| 298 | * copied data.
|
| 299 |
|
| 300 | static int __reinsert_ablock(struct dm_array_info *info, unsigned int index,
|
| 301 | struct dm_block *block, dm_block_t b,
|
| 302 | dm_block_t *root)
|
| 303 | {
|
| 304 | int r = 0;
|
| 305 |
|
| 306 | if (dm_block_location(block) != b) {
|
| 307 | |
| 308 | * dm_tm_shadow_block will have already decremented the old
|
| 309 | * block, but it is still referenced by the btree. We
|
| 310 | * increment to stop the insert decrementing it below zero
|
| 311 | * when overwriting the old value.
|
| 312 |
|
| 313 | dm_tm_inc(info->btree_info.tm, b);
|
| 314 | r = insert_ablock(info, index, block, root);
|
| 315 | }
|
| 316 |
|
| 317 | return r;
|
| 318 | }
|
| 319 |
|
| 320 | |
| 321 | * Looks up an array block in the btree. Then shadows it, and updates the
|
| 322 | * btree to point to this new shadow. 'root' is an input/output parameter
|
| 323 | * for both the current root block, and the new one.
|
| 324 |
|
| 325 | static int shadow_ablock(struct dm_array_info *info, dm_block_t *root,
|
| 326 | unsigned int index, struct dm_block **block,
|
| 327 | struct array_block **ab)
|
| 328 | {
|
| 329 | int r;
|
| 330 | uint64_t key = index;
|
| 331 | dm_block_t b;
|
| 332 | __le64 block_le;
|
| 333 |
|
| 334 | r = dm_btree_lookup(&info->btree_info, *root, &key, &block_le);
|
| 335 | if (r)
|
| 336 | return r;
|
| 337 | b = le64_to_cpu(block_le);
|
| 338 |
|
| 339 | r = __shadow_ablock(info, b, block, ab);
|
| 340 | if (r)
|
| 341 | return r;
|
| 342 |
|
| 343 | return __reinsert_ablock(info, index, *block, b, root);
|
| 344 | }
|
| 345 |
|
| 346 | |
| 347 | * Allocate an new array block, and fill it with some values.
|
| 348 |
|
| 349 | static int insert_new_ablock(struct dm_array_info *info, size_t size_of_block,
|
| 350 | uint32_t max_entries,
|
| 351 | unsigned int block_index, uint32_t nr,
|
| 352 | const void *value, dm_block_t *root)
|
| 353 | {
|
| 354 | int r;
|
| 355 | struct dm_block *block;
|
| 356 | struct array_block *ab;
|
| 357 |
|
| 358 | r = alloc_ablock(info, size_of_block, max_entries, &block, &ab);
|
| 359 | if (r)
|
| 360 | return r;
|
| 361 |
|
| 362 | fill_ablock(info, ab, value, nr);
|
| 363 | r = insert_ablock(info, block_index, block, root);
|
| 364 | unlock_ablock(info, block);
|
| 365 |
|
| 366 | return r;
|
| 367 | }
|
| 368 |
|
| 369 | static int insert_full_ablocks(struct dm_array_info *info, size_t size_of_block,
|
| 370 | unsigned int begin_block, unsigned int end_block,
|
| 371 | unsigned int max_entries, const void *value,
|
| 372 | dm_block_t *root)
|
| 373 | {
|
| 374 | int r = 0;
|
| 375 |
|
| 376 | for (; !r && begin_block != end_block; begin_block++)
|
| 377 | r = insert_new_ablock(info, size_of_block, max_entries, begin_block, max_entries, value, root);
|
| 378 |
|
| 379 | return r;
|
| 380 | }
|
| 381 |
|
| 382 | |
| 383 | * There are a bunch of functions involved with resizing an array. This
|
| 384 | * structure holds information that commonly needed by them. Purely here
|
| 385 | * to reduce parameter count.
|
| 386 |
|
| 387 | struct resize {
|
| 388 | |
| 389 | * Describes the array.
|
| 390 |
|
| 391 | struct dm_array_info *info;
|
| 392 |
|
| 393 | |
| 394 | * The current root of the array. This gets updated.
|
| 395 |
|
| 396 | dm_block_t root;
|
| 397 |
|
| 398 | |
| 399 | * Metadata block size. Used to calculate the nr entries in an
|
| 400 | * array block.
|
| 401 |
|
| 402 | size_t size_of_block;
|
| 403 |
|
| 404 | |
| 405 | * Maximum nr entries in an array block.
|
| 406 |
|
| 407 | unsigned int max_entries;
|
| 408 |
|
| 409 | |
| 410 | * nr of completely full blocks in the array.
|
| 411 | *
|
| 412 | * 'old' refers to before the resize, 'new' after.
|
| 413 |
|
| 414 | unsigned int old_nr_full_blocks, new_nr_full_blocks;
|
| 415 |
|
| 416 | |
| 417 | * Number of entries in the final block. 0 iff only full blocks in
|
| 418 | * the array.
|
| 419 |
|
| 420 | unsigned int old_nr_entries_in_last_block, new_nr_entries_in_last_block;
|
| 421 |
|
| 422 | |
| 423 | * The default value used when growing the array.
|
| 424 |
|
| 425 | const void *value;
|
| 426 | };
|
| 427 |
|
| 428 | |
| 429 | * Removes a consecutive set of array blocks from the btree. The values
|
| 430 | * in block are decremented as a side effect of the btree remove.
|
| 431 | *
|
| 432 | * begin_index - the index of the first array block to remove.
|
| 433 | * end_index - the one-past-the-end value. ie. this block is not removed.
|
| 434 |
|
| 435 | static int drop_blocks(struct resize *resize, unsigned int begin_index,
|
| 436 | unsigned int end_index)
|
| 437 | {
|
| 438 | int r;
|
| 439 |
|
| 440 | while (begin_index != end_index) {
|
| 441 | uint64_t key = begin_index++;
|
| 442 |
|
| 443 | r = dm_btree_remove(&resize->info->btree_info, resize->root,
|
| 444 | &key, &resize->root);
|
| 445 | if (r)
|
| 446 | return r;
|
| 447 | }
|
| 448 |
|
| 449 | return 0;
|
| 450 | }
|
| 451 |
|
| 452 | |
| 453 | * Calculates how many blocks are needed for the array.
|
| 454 |
|
| 455 | static unsigned int total_nr_blocks_needed(unsigned int nr_full_blocks,
|
| 456 | unsigned int nr_entries_in_last_block)
|
| 457 | {
|
| 458 | return nr_full_blocks + (nr_entries_in_last_block ? 1 : 0);
|
| 459 | }
|
| 460 |
|
| 461 | |
| 462 | * Shrink an array.
|
| 463 |
|
| 464 | static int shrink(struct resize *resize)
|
| 465 | {
|
| 466 | int r;
|
| 467 | unsigned int begin, end;
|
| 468 | struct dm_block *block;
|
| 469 | struct array_block *ab;
|
| 470 |
|
| 471 | |
| 472 | * Lose some blocks from the back?
|
| 473 |
|
| 474 | if (resize->new_nr_full_blocks < resize->old_nr_full_blocks) {
|
| 475 | begin = total_nr_blocks_needed(resize->new_nr_full_blocks,
|
| 476 | resize->new_nr_entries_in_last_block);
|
| 477 | end = total_nr_blocks_needed(resize->old_nr_full_blocks,
|
| 478 | resize->old_nr_entries_in_last_block);
|
| 479 |
|
| 480 | r = drop_blocks(resize, begin, end);
|
| 481 | if (r)
|
| 482 | return r;
|
| 483 | }
|
| 484 |
|
| 485 | |
| 486 | * Trim the new tail block
|
| 487 |
|
| 488 | if (resize->new_nr_entries_in_last_block) {
|
| 489 | r = shadow_ablock(resize->info, &resize->root,
|
| 490 | resize->new_nr_full_blocks, &block, &ab);
|
| 491 | if (r)
|
| 492 | return r;
|
| 493 |
|
| 494 | trim_ablock(resize->info, ab, resize->new_nr_entries_in_last_block);
|
| 495 | unlock_ablock(resize->info, block);
|
| 496 | }
|
| 497 |
|
| 498 | return 0;
|
| 499 | }
|
| 500 |
|
| 501 | |
| 502 | * Grow an array.
|
| 503 |
|
| 504 | static int grow_extend_tail_block(struct resize *resize, uint32_t new_nr_entries)
|
| 505 | {
|
| 506 | int r;
|
| 507 | struct dm_block *block;
|
| 508 | struct array_block *ab;
|
| 509 |
|
| 510 | r = shadow_ablock(resize->info, &resize->root,
|
| 511 | resize->old_nr_full_blocks, &block, &ab);
|
| 512 | if (r)
|
| 513 | return r;
|
| 514 |
|
| 515 | fill_ablock(resize->info, ab, resize->value, new_nr_entries);
|
| 516 | unlock_ablock(resize->info, block);
|
| 517 |
|
| 518 | return r;
|
| 519 | }
|
| 520 |
|
| 521 | static int grow_add_tail_block(struct resize *resize)
|
| 522 | {
|
| 523 | return insert_new_ablock(resize->info, resize->size_of_block,
|
| 524 | resize->max_entries,
|
| 525 | resize->new_nr_full_blocks,
|
| 526 | resize->new_nr_entries_in_last_block,
|
| 527 | resize->value, &resize->root);
|
| 528 | }
|
| 529 |
|
| 530 | static int grow_needs_more_blocks(struct resize *resize)
|
| 531 | {
|
| 532 | int r;
|
| 533 | unsigned int old_nr_blocks = resize->old_nr_full_blocks;
|
| 534 |
|
| 535 | if (resize->old_nr_entries_in_last_block > 0) {
|
| 536 | old_nr_blocks++;
|
| 537 |
|
| 538 | r = grow_extend_tail_block(resize, resize->max_entries);
|
| 539 | if (r)
|
| 540 | return r;
|
| 541 | }
|
| 542 |
|
| 543 | r = insert_full_ablocks(resize->info, resize->size_of_block,
|
| 544 | old_nr_blocks,
|
| 545 | resize->new_nr_full_blocks,
|
| 546 | resize->max_entries, resize->value,
|
| 547 | &resize->root);
|
| 548 | if (r)
|
| 549 | return r;
|
| 550 |
|
| 551 | if (resize->new_nr_entries_in_last_block)
|
| 552 | r = grow_add_tail_block(resize);
|
| 553 |
|
| 554 | return r;
|
| 555 | }
|
| 556 |
|
| 557 | static int grow(struct resize *resize)
|
| 558 | {
|
| 559 | if (resize->new_nr_full_blocks > resize->old_nr_full_blocks)
|
| 560 | return grow_needs_more_blocks(resize);
|
| 561 |
|
| 562 | else if (resize->old_nr_entries_in_last_block)
|
| 563 | return grow_extend_tail_block(resize, resize->new_nr_entries_in_last_block);
|
| 564 |
|
| 565 | else
|
| 566 | return grow_add_tail_block(resize);
|
| 567 | }
|
| 568 |
|
| 569 |
|
| 570 |
|
| 571 | |
| 572 | * These are the value_type functions for the btree elements, which point
|
| 573 | * to array blocks.
|
| 574 |
|
| 575 | static void block_inc(void *context, const void *value, unsigned int count)
|
| 576 | {
|
| 577 | const __le64 *block_le = value;
|
| 578 | struct dm_array_info *info = context;
|
| 579 | unsigned int i;
|
| 580 |
|
| 581 | for (i = 0; i < count; i++, block_le++)
|
| 582 | dm_tm_inc(info->btree_info.tm, le64_to_cpu(*block_le));
|
| 583 | }
|
| 584 |
|
| 585 | static void __block_dec(void *context, const void *value)
|
| 586 | {
|
| 587 | int r;
|
| 588 | uint64_t b;
|
| 589 | __le64 block_le;
|
| 590 | uint32_t ref_count;
|
| 591 | struct dm_block *block;
|
| 592 | struct array_block *ab;
|
| 593 | struct dm_array_info *info = context;
|
| 594 |
|
| 595 | memcpy(&block_le, value, sizeof(block_le));
|
| 596 | b = le64_to_cpu(block_le);
|
| 597 |
|
| 598 | r = dm_tm_ref(info->btree_info.tm, b, &ref_count);
|
| 599 | if (r) {
|
| 600 | DMERR_LIMIT("couldn't get reference count for block %llu",
|
| 601 | (unsigned long long) b);
|
| 602 | return;
|
| 603 | }
|
| 604 |
|
| 605 | if (ref_count == 1) {
|
| 606 | |
| 607 | * We're about to drop the last reference to this ablock.
|
| 608 | * So we need to decrement the ref count of the contents.
|
| 609 |
|
| 610 | r = get_ablock(info, b, &block, &ab);
|
| 611 | if (r) {
|
| 612 | DMERR_LIMIT("couldn't get array block %llu",
|
| 613 | (unsigned long long) b);
|
| 614 | return;
|
| 615 | }
|
| 616 |
|
| 617 | dec_ablock_entries(info, ab);
|
| 618 | unlock_ablock(info, block);
|
| 619 | }
|
| 620 |
|
| 621 | dm_tm_dec(info->btree_info.tm, b);
|
| 622 | }
|
| 623 |
|
| 624 | static void block_dec(void *context, const void *value, unsigned int count)
|
| 625 | {
|
| 626 | unsigned int i;
|
| 627 |
|
| 628 | for (i = 0; i < count; i++, value += sizeof(__le64))
|
| 629 | __block_dec(context, value);
|
| 630 | }
|
| 631 |
|
| 632 | static int block_equal(void *context, const void *value1, const void *value2)
|
| 633 | {
|
| 634 | return !memcmp(value1, value2, sizeof(__le64));
|
| 635 | }
|
| 636 |
|
| 637 |
|
| 638 |
|
| 639 | void dm_array_info_init(struct dm_array_info *info,
|
| 640 | struct dm_transaction_manager *tm,
|
| 641 | struct dm_btree_value_type *vt)
|
| 642 | {
|
| 643 | struct dm_btree_value_type *bvt = &info->btree_info.value_type;
|
| 644 |
|
| 645 | memcpy(&info->value_type, vt, sizeof(info->value_type));
|
| 646 | info->btree_info.tm = tm;
|
| 647 | info->btree_info.levels = 1;
|
| 648 |
|
| 649 | bvt->context = info;
|
| 650 | bvt->size = sizeof(__le64);
|
| 651 | bvt->inc = block_inc;
|
| 652 | bvt->dec = block_dec;
|
| 653 | bvt->equal = block_equal;
|
| 654 | }
|
| 655 | EXPORT_SYMBOL_GPL(dm_array_info_init);
|
| 656 |
|
| 657 | int dm_array_empty(struct dm_array_info *info, dm_block_t *root)
|
| 658 | {
|
| 659 | return dm_btree_empty(&info->btree_info, root);
|
| 660 | }
|
| 661 | EXPORT_SYMBOL_GPL(dm_array_empty);
|
| 662 |
|
| 663 | static int array_resize(struct dm_array_info *info, dm_block_t root,
|
| 664 | uint32_t old_size, uint32_t new_size,
|
| 665 | const void *value, dm_block_t *new_root)
|
| 666 | {
|
| 667 | int r;
|
| 668 | struct resize resize;
|
| 669 |
|
| 670 | if (old_size == new_size) {
|
| 671 | *new_root = root;
|
| 672 | return 0;
|
| 673 | }
|
| 674 |
|
| 675 | resize.info = info;
|
| 676 | resize.root = root;
|
| 677 | resize.size_of_block = dm_bm_block_size(dm_tm_get_bm(info->btree_info.tm));
|
| 678 | resize.max_entries = calc_max_entries(info->value_type.size,
|
| 679 | resize.size_of_block);
|
| 680 |
|
| 681 | resize.old_nr_full_blocks = old_size / resize.max_entries;
|
| 682 | resize.old_nr_entries_in_last_block = old_size % resize.max_entries;
|
| 683 | resize.new_nr_full_blocks = new_size / resize.max_entries;
|
| 684 | resize.new_nr_entries_in_last_block = new_size % resize.max_entries;
|
| 685 | resize.value = value;
|
| 686 |
|
| 687 | r = ((new_size > old_size) ? grow : shrink)(&resize);
|
| 688 | if (r)
|
| 689 | return r;
|
| 690 |
|
| 691 | *new_root = resize.root;
|
| 692 | return 0;
|
| 693 | }
|
| 694 |
|
| 695 | int dm_array_resize(struct dm_array_info *info, dm_block_t root,
|
| 696 | uint32_t old_size, uint32_t new_size,
|
| 697 | const void *value, dm_block_t *new_root)
|
| 698 | __dm_written_to_disk(value)
|
| 699 | {
|
| 700 | int r = array_resize(info, root, old_size, new_size, value, new_root);
|
| 701 |
|
| 702 | __dm_unbless_for_disk(value);
|
| 703 | return r;
|
| 704 | }
|
| 705 | EXPORT_SYMBOL_GPL(dm_array_resize);
|
| 706 |
|
| 707 | static int populate_ablock_with_values(struct dm_array_info *info, struct array_block *ab,
|
| 708 | value_fn fn, void *context,
|
| 709 | unsigned int base, unsigned int new_nr)
|
| 710 | {
|
| 711 | int r;
|
| 712 | unsigned int i;
|
| 713 | struct dm_btree_value_type *vt = &info->value_type;
|
| 714 |
|
| 715 | BUG_ON(le32_to_cpu(ab->nr_entries));
|
| 716 | BUG_ON(new_nr > le32_to_cpu(ab->max_entries));
|
| 717 |
|
| 718 | for (i = 0; i < new_nr; i++) {
|
| 719 | r = fn(base + i, element_at(info, ab, i), context);
|
| 720 | if (r)
|
| 721 | return r;
|
| 722 |
|
| 723 | if (vt->inc)
|
| 724 | vt->inc(vt->context, element_at(info, ab, i), 1);
|
| 725 | }
|
| 726 |
|
| 727 | ab->nr_entries = cpu_to_le32(new_nr);
|
| 728 | return 0;
|
| 729 | }
|
| 730 |
|
| 731 | int dm_array_new(struct dm_array_info *info, dm_block_t *root,
|
| 732 | uint32_t size, value_fn fn, void *context)
|
| 733 | {
|
| 734 | int r;
|
| 735 | struct dm_block *block;
|
| 736 | struct array_block *ab;
|
| 737 | unsigned int block_index, end_block, size_of_block, max_entries;
|
| 738 |
|
| 739 | r = dm_array_empty(info, root);
|
| 740 | if (r)
|
| 741 | return r;
|
| 742 |
|
| 743 | size_of_block = dm_bm_block_size(dm_tm_get_bm(info->btree_info.tm));
|
| 744 | max_entries = calc_max_entries(info->value_type.size, size_of_block);
|
| 745 | end_block = dm_div_up(size, max_entries);
|
| 746 |
|
| 747 | for (block_index = 0; block_index != end_block; block_index++) {
|
| 748 | r = alloc_ablock(info, size_of_block, max_entries, &block, &ab);
|
| 749 | if (r)
|
| 750 | break;
|
| 751 |
|
| 752 | r = populate_ablock_with_values(info, ab, fn, context,
|
| 753 | block_index * max_entries,
|
| 754 | min(max_entries, size));
|
| 755 | if (r) {
|
| 756 | unlock_ablock(info, block);
|
| 757 | break;
|
| 758 | }
|
| 759 |
|
| 760 | r = insert_ablock(info, block_index, block, root);
|
| 761 | unlock_ablock(info, block);
|
| 762 | if (r)
|
| 763 | break;
|
| 764 |
|
| 765 | size -= max_entries;
|
| 766 | }
|
| 767 |
|
| 768 | return r;
|
| 769 | }
|
| 770 | EXPORT_SYMBOL_GPL(dm_array_new);
|
| 771 |
|
| 772 | int dm_array_del(struct dm_array_info *info, dm_block_t root)
|
| 773 | {
|
| 774 | return dm_btree_del(&info->btree_info, root);
|
| 775 | }
|
| 776 | EXPORT_SYMBOL_GPL(dm_array_del);
|
| 777 |
|
| 778 | int dm_array_get_value(struct dm_array_info *info, dm_block_t root,
|
| 779 | uint32_t index, void *value_le)
|
| 780 | {
|
| 781 | int r;
|
| 782 | struct dm_block *block;
|
| 783 | struct array_block *ab;
|
| 784 | size_t size_of_block;
|
| 785 | unsigned int entry, max_entries;
|
| 786 |
|
| 787 | size_of_block = dm_bm_block_size(dm_tm_get_bm(info->btree_info.tm));
|
| 788 | max_entries = calc_max_entries(info->value_type.size, size_of_block);
|
| 789 |
|
| 790 | r = lookup_ablock(info, root, index / max_entries, &block, &ab);
|
| 791 | if (r)
|
| 792 | return r;
|
| 793 |
|
| 794 | entry = index % max_entries;
|
| 795 | if (entry >= le32_to_cpu(ab->nr_entries))
|
| 796 | r = -ENODATA;
|
| 797 | else
|
| 798 | memcpy(value_le, element_at(info, ab, entry),
|
| 799 | info->value_type.size);
|
| 800 |
|
| 801 | unlock_ablock(info, block);
|
| 802 | return r;
|
| 803 | }
|
| 804 | EXPORT_SYMBOL_GPL(dm_array_get_value);
|
| 805 |
|
| 806 | static int array_set_value(struct dm_array_info *info, dm_block_t root,
|
| 807 | uint32_t index, const void *value, dm_block_t *new_root)
|
| 808 | {
|
| 809 | int r;
|
| 810 | struct dm_block *block;
|
| 811 | struct array_block *ab;
|
| 812 | size_t size_of_block;
|
| 813 | unsigned int max_entries;
|
| 814 | unsigned int entry;
|
| 815 | void *old_value;
|
| 816 | struct dm_btree_value_type *vt = &info->value_type;
|
| 817 |
|
| 818 | size_of_block = dm_bm_block_size(dm_tm_get_bm(info->btree_info.tm));
|
| 819 | max_entries = calc_max_entries(info->value_type.size, size_of_block);
|
| 820 |
|
| 821 | r = shadow_ablock(info, &root, index / max_entries, &block, &ab);
|
| 822 | if (r)
|
| 823 | return r;
|
| 824 | *new_root = root;
|
| 825 |
|
| 826 | entry = index % max_entries;
|
| 827 | if (entry >= le32_to_cpu(ab->nr_entries)) {
|
| 828 | r = -ENODATA;
|
| 829 | goto out;
|
| 830 | }
|
| 831 |
|
| 832 | old_value = element_at(info, ab, entry);
|
| 833 | if (vt->dec &&
|
| 834 | (!vt->equal || !vt->equal(vt->context, old_value, value))) {
|
| 835 | vt->dec(vt->context, old_value, 1);
|
| 836 | if (vt->inc)
|
| 837 | vt->inc(vt->context, value, 1);
|
| 838 | }
|
| 839 |
|
| 840 | memcpy(old_value, value, info->value_type.size);
|
| 841 |
|
| 842 | out:
|
| 843 | unlock_ablock(info, block);
|
| 844 | return r;
|
| 845 | }
|
| 846 |
|
| 847 | int dm_array_set_value(struct dm_array_info *info, dm_block_t root,
|
| 848 | uint32_t index, const void *value, dm_block_t *new_root)
|
| 849 | __dm_written_to_disk(value)
|
| 850 | {
|
| 851 | int r;
|
| 852 |
|
| 853 | r = array_set_value(info, root, index, value, new_root);
|
| 854 | __dm_unbless_for_disk(value);
|
| 855 | return r;
|
| 856 | }
|
| 857 | EXPORT_SYMBOL_GPL(dm_array_set_value);
|
| 858 |
|
| 859 | struct walk_info {
|
| 860 | struct dm_array_info *info;
|
| 861 | int (*fn)(void *context, uint64_t key, void *leaf);
|
| 862 | void *context;
|
| 863 | };
|
| 864 |
|
| 865 | static int walk_ablock(void *context, uint64_t *keys, void *leaf)
|
| 866 | {
|
| 867 | struct walk_info *wi = context;
|
| 868 |
|
| 869 | int r;
|
| 870 | unsigned int i;
|
| 871 | __le64 block_le;
|
| 872 | unsigned int nr_entries, max_entries;
|
| 873 | struct dm_block *block;
|
| 874 | struct array_block *ab;
|
| 875 |
|
| 876 | memcpy(&block_le, leaf, sizeof(block_le));
|
| 877 | r = get_ablock(wi->info, le64_to_cpu(block_le), &block, &ab);
|
| 878 | if (r)
|
| 879 | return r;
|
| 880 |
|
| 881 | max_entries = le32_to_cpu(ab->max_entries);
|
| 882 | nr_entries = le32_to_cpu(ab->nr_entries);
|
| 883 | for (i = 0; i < nr_entries; i++) {
|
| 884 | r = wi->fn(wi->context, keys[0] * max_entries + i,
|
| 885 | element_at(wi->info, ab, i));
|
| 886 |
|
| 887 | if (r)
|
| 888 | break;
|
| 889 | }
|
| 890 |
|
| 891 | unlock_ablock(wi->info, block);
|
| 892 | return r;
|
| 893 | }
|
| 894 |
|
| 895 | int dm_array_walk(struct dm_array_info *info, dm_block_t root,
|
| 896 | int (*fn)(void *, uint64_t key, void *leaf),
|
| 897 | void *context)
|
| 898 | {
|
| 899 | struct walk_info wi;
|
| 900 |
|
| 901 | wi.info = info;
|
| 902 | wi.fn = fn;
|
| 903 | wi.context = context;
|
| 904 |
|
| 905 | return dm_btree_walk(&info->btree_info, root, walk_ablock, &wi);
|
| 906 | }
|
| 907 | EXPORT_SYMBOL_GPL(dm_array_walk);
|
| 908 |
|
| 909 |
|
| 910 |
|
| 911 | static int load_ablock(struct dm_array_cursor *c)
|
| 912 | {
|
| 913 | int r;
|
| 914 | __le64 value_le;
|
| 915 | uint64_t key;
|
| 916 |
|
| 917 | if (c->block)
|
| 918 | unlock_ablock(c->info, c->block);
|
| 919 |
|
| 920 | c->block = NULL;
|
| 921 | c->ab = NULL;
|
| 922 | c->index = 0;
|
| 923 |
|
| 924 | r = dm_btree_cursor_get_value(&c->cursor, &key, &value_le);
|
| 925 | if (r) {
|
| 926 | DMERR("dm_btree_cursor_get_value failed");
|
| 927 | dm_btree_cursor_end(&c->cursor);
|
| 928 |
|
| 929 | } else {
|
| 930 | r = get_ablock(c->info, le64_to_cpu(value_le), &c->block, &c->ab);
|
| 931 | if (r) {
|
| 932 | DMERR("get_ablock failed");
|
| 933 | dm_btree_cursor_end(&c->cursor);
|
| 934 | }
|
| 935 | }
|
| 936 |
|
| 937 | return r;
|
| 938 | }
|
| 939 |
|
| 940 | int dm_array_cursor_begin(struct dm_array_info *info, dm_block_t root,
|
| 941 | struct dm_array_cursor *c)
|
| 942 | {
|
| 943 | int r;
|
| 944 |
|
| 945 | memset(c, 0, sizeof(*c));
|
| 946 | c->info = info;
|
| 947 | r = dm_btree_cursor_begin(&info->btree_info, root, true, &c->cursor);
|
| 948 | if (r) {
|
| 949 | DMERR("couldn't create btree cursor");
|
| 950 | return r;
|
| 951 | }
|
| 952 |
|
| 953 | return load_ablock(c);
|
| 954 | }
|
| 955 | EXPORT_SYMBOL_GPL(dm_array_cursor_begin);
|
| 956 |
|
| 957 | void dm_array_cursor_end(struct dm_array_cursor *c)
|
| 958 | {
|
| 959 | if (c->block) {
|
| 960 | unlock_ablock(c->info, c->block);
|
| 961 | dm_btree_cursor_end(&c->cursor);
|
| 962 | }
|
| 963 | }
|
| 964 | EXPORT_SYMBOL_GPL(dm_array_cursor_end);
|
| 965 |
|
| 966 | int dm_array_cursor_next(struct dm_array_cursor *c)
|
| 967 | {
|
| 968 | int r;
|
| 969 |
|
| 970 | if (!c->block)
|
| 971 | return -ENODATA;
|
| 972 |
|
| 973 | c->index++;
|
| 974 |
|
| 975 | if (c->index >= le32_to_cpu(c->ab->nr_entries)) {
|
| 976 | r = dm_btree_cursor_next(&c->cursor);
|
| 977 | if (r)
|
| 978 | return r;
|
| 979 |
|
| 980 | r = load_ablock(c);
|
| 981 | if (r)
|
| 982 | return r;
|
| 983 | }
|
| 984 |
|
| 985 | return 0;
|
| 986 | }
|
| 987 | EXPORT_SYMBOL_GPL(dm_array_cursor_next);
|
| 988 |
|
| 989 | int dm_array_cursor_skip(struct dm_array_cursor *c, uint32_t count)
|
| 990 | {
|
| 991 | int r;
|
| 992 |
|
| 993 | do {
|
| 994 | uint32_t remaining = le32_to_cpu(c->ab->nr_entries) - c->index;
|
| 995 |
|
| 996 | if (count < remaining) {
|
| 997 | c->index += count;
|
| 998 | return 0;
|
| 999 | }
|
| 1000 |
|
| 1001 | count -= remaining;
|
| 1002 | r = dm_array_cursor_next(c);
|
| 1003 |
|
| 1004 | } while (!r);
|
| 1005 |
|
| 1006 | return r;
|
| 1007 | }
|
| 1008 | EXPORT_SYMBOL_GPL(dm_array_cursor_skip);
|
| 1009 |
|
| 1010 | void dm_array_cursor_get_value(struct dm_array_cursor *c, void **value_le)
|
| 1011 | {
|
| 1012 | *value_le = element_at(c->info, c->ab, c->index);
|
| 1013 | }
|
| 1014 | EXPORT_SYMBOL_GPL(dm_array_cursor_get_value);
|
| 1015 |
|
| 1016 |
|