| 1 | /* |
| 2 | * Copyright (C) 2007 Oracle. All rights reserved. |
| 3 | * |
| 4 | * This program is free software; you can redistribute it and/or |
| 5 | * modify it under the terms of the GNU General Public |
| 6 | * License v2 as published by the Free Software Foundation. |
| 7 | * |
| 8 | * This program is distributed in the hope that it will be useful, |
| 9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 11 | * General Public License for more details. |
| 12 | * |
| 13 | * You should have received a copy of the GNU General Public |
| 14 | * License along with this program; if not, write to the |
| 15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
| 16 | * Boston, MA 021110-1307, USA. |
| 17 | */ |
| 18 | |
| 19 | #include <linux/sched.h> |
| 20 | #include "ctree.h" |
| 21 | #include "disk-io.h" |
| 22 | #include "print-tree.h" |
| 23 | #include "transaction.h" |
| 24 | |
| 25 | #define BLOCK_GROUP_DATA EXTENT_WRITEBACK |
| 26 | #define BLOCK_GROUP_METADATA EXTENT_UPTODATE |
| 27 | #define BLOCK_GROUP_DIRTY EXTENT_DIRTY |
| 28 | |
| 29 | static int finish_current_insert(struct btrfs_trans_handle *trans, struct |
| 30 | btrfs_root *extent_root); |
| 31 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
| 32 | btrfs_root *extent_root); |
| 33 | |
| 34 | static int cache_block_group(struct btrfs_root *root, |
| 35 | struct btrfs_block_group_cache *block_group) |
| 36 | { |
| 37 | struct btrfs_path *path; |
| 38 | int ret; |
| 39 | struct btrfs_key key; |
| 40 | struct extent_buffer *leaf; |
| 41 | struct extent_map_tree *free_space_cache; |
| 42 | int slot; |
| 43 | u64 last = 0; |
| 44 | u64 hole_size; |
| 45 | u64 first_free; |
| 46 | int found = 0; |
| 47 | |
| 48 | if (!block_group) |
| 49 | return 0; |
| 50 | |
| 51 | root = root->fs_info->extent_root; |
| 52 | free_space_cache = &root->fs_info->free_space_cache; |
| 53 | |
| 54 | if (block_group->cached) |
| 55 | return 0; |
| 56 | |
| 57 | path = btrfs_alloc_path(); |
| 58 | if (!path) |
| 59 | return -ENOMEM; |
| 60 | |
| 61 | path->reada = 2; |
| 62 | first_free = block_group->key.objectid; |
| 63 | key.objectid = block_group->key.objectid; |
| 64 | key.offset = 0; |
| 65 | |
| 66 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
| 67 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
| 68 | |
| 69 | if (ret < 0) |
| 70 | return ret; |
| 71 | |
| 72 | if (ret && path->slots[0] > 0) |
| 73 | path->slots[0]--; |
| 74 | |
| 75 | while(1) { |
| 76 | leaf = path->nodes[0]; |
| 77 | slot = path->slots[0]; |
| 78 | if (slot >= btrfs_header_nritems(leaf)) { |
| 79 | ret = btrfs_next_leaf(root, path); |
| 80 | if (ret < 0) |
| 81 | goto err; |
| 82 | if (ret == 0) { |
| 83 | continue; |
| 84 | } else { |
| 85 | break; |
| 86 | } |
| 87 | } |
| 88 | |
| 89 | btrfs_item_key_to_cpu(leaf, &key, slot); |
| 90 | if (key.objectid < block_group->key.objectid) { |
| 91 | if (key.objectid + key.offset > first_free) |
| 92 | first_free = key.objectid + key.offset; |
| 93 | goto next; |
| 94 | } |
| 95 | |
| 96 | if (key.objectid >= block_group->key.objectid + |
| 97 | block_group->key.offset) { |
| 98 | break; |
| 99 | } |
| 100 | |
| 101 | if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) { |
| 102 | if (!found) { |
| 103 | last = first_free; |
| 104 | found = 1; |
| 105 | } |
| 106 | if (key.objectid > last) { |
| 107 | hole_size = key.objectid - last; |
| 108 | set_extent_dirty(free_space_cache, last, |
| 109 | last + hole_size - 1, |
| 110 | GFP_NOFS); |
| 111 | } |
| 112 | last = key.objectid + key.offset; |
| 113 | } |
| 114 | next: |
| 115 | path->slots[0]++; |
| 116 | } |
| 117 | |
| 118 | if (!found) |
| 119 | last = first_free; |
| 120 | if (block_group->key.objectid + |
| 121 | block_group->key.offset > last) { |
| 122 | hole_size = block_group->key.objectid + |
| 123 | block_group->key.offset - last; |
| 124 | set_extent_dirty(free_space_cache, last, |
| 125 | last + hole_size - 1, GFP_NOFS); |
| 126 | } |
| 127 | block_group->cached = 1; |
| 128 | err: |
| 129 | btrfs_free_path(path); |
| 130 | return 0; |
| 131 | } |
| 132 | |
| 133 | struct btrfs_block_group_cache *btrfs_lookup_block_group(struct |
| 134 | btrfs_fs_info *info, |
| 135 | u64 bytenr) |
| 136 | { |
| 137 | struct extent_map_tree *block_group_cache; |
| 138 | struct btrfs_block_group_cache *block_group = NULL; |
| 139 | u64 ptr; |
| 140 | u64 start; |
| 141 | u64 end; |
| 142 | int ret; |
| 143 | |
| 144 | block_group_cache = &info->block_group_cache; |
| 145 | ret = find_first_extent_bit(block_group_cache, |
| 146 | bytenr, &start, &end, |
| 147 | BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA); |
| 148 | if (ret) { |
| 149 | return NULL; |
| 150 | } |
| 151 | ret = get_state_private(block_group_cache, start, &ptr); |
| 152 | if (ret) |
| 153 | return NULL; |
| 154 | |
| 155 | block_group = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
| 156 | if (block_group->key.objectid <= bytenr && bytenr < |
| 157 | block_group->key.objectid + block_group->key.offset) |
| 158 | return block_group; |
| 159 | return NULL; |
| 160 | } |
| 161 | static u64 find_search_start(struct btrfs_root *root, |
| 162 | struct btrfs_block_group_cache **cache_ret, |
| 163 | u64 search_start, int num, |
| 164 | int data, int full_scan) |
| 165 | { |
| 166 | int ret; |
| 167 | struct btrfs_block_group_cache *cache = *cache_ret; |
| 168 | u64 last; |
| 169 | u64 start = 0; |
| 170 | u64 end = 0; |
| 171 | u64 cache_miss = 0; |
| 172 | int wrapped = 0; |
| 173 | |
| 174 | if (!cache) { |
| 175 | cache = btrfs_lookup_block_group(root->fs_info, search_start); |
| 176 | if (!cache) |
| 177 | return search_start; |
| 178 | } |
| 179 | again: |
| 180 | ret = cache_block_group(root, cache); |
| 181 | if (ret) |
| 182 | goto out; |
| 183 | |
| 184 | last = max(search_start, cache->key.objectid); |
| 185 | |
| 186 | while(1) { |
| 187 | ret = find_first_extent_bit(&root->fs_info->free_space_cache, |
| 188 | last, &start, &end, EXTENT_DIRTY); |
| 189 | if (ret) { |
| 190 | if (!cache_miss) |
| 191 | cache_miss = last; |
| 192 | goto new_group; |
| 193 | } |
| 194 | |
| 195 | start = max(last, start); |
| 196 | last = end + 1; |
| 197 | if (last - start < num) { |
| 198 | if (last == cache->key.objectid + cache->key.offset) |
| 199 | cache_miss = start; |
| 200 | continue; |
| 201 | } |
| 202 | if (data != BTRFS_BLOCK_GROUP_MIXED && |
| 203 | start + num > cache->key.objectid + cache->key.offset) |
| 204 | goto new_group; |
| 205 | return start; |
| 206 | } |
| 207 | out: |
| 208 | return search_start; |
| 209 | |
| 210 | new_group: |
| 211 | last = cache->key.objectid + cache->key.offset; |
| 212 | wrapped: |
| 213 | cache = btrfs_lookup_block_group(root->fs_info, last); |
| 214 | if (!cache) { |
| 215 | no_cache: |
| 216 | if (!wrapped) { |
| 217 | wrapped = 1; |
| 218 | last = search_start; |
| 219 | data = BTRFS_BLOCK_GROUP_MIXED; |
| 220 | goto wrapped; |
| 221 | } |
| 222 | return search_start; |
| 223 | } |
| 224 | if (cache_miss && !cache->cached) { |
| 225 | cache_block_group(root, cache); |
| 226 | last = cache_miss; |
| 227 | cache = btrfs_lookup_block_group(root->fs_info, last); |
| 228 | } |
| 229 | if (!full_scan) |
| 230 | cache = btrfs_find_block_group(root, cache, last, data, 0); |
| 231 | if (!cache) |
| 232 | goto no_cache; |
| 233 | *cache_ret = cache; |
| 234 | cache_miss = 0; |
| 235 | goto again; |
| 236 | } |
| 237 | |
| 238 | static u64 div_factor(u64 num, int factor) |
| 239 | { |
| 240 | if (factor == 10) |
| 241 | return num; |
| 242 | num *= factor; |
| 243 | do_div(num, 10); |
| 244 | return num; |
| 245 | } |
| 246 | |
| 247 | struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root, |
| 248 | struct btrfs_block_group_cache |
| 249 | *hint, u64 search_start, |
| 250 | int data, int owner) |
| 251 | { |
| 252 | struct btrfs_block_group_cache *cache; |
| 253 | struct extent_map_tree *block_group_cache; |
| 254 | struct btrfs_block_group_cache *found_group = NULL; |
| 255 | struct btrfs_fs_info *info = root->fs_info; |
| 256 | u64 used; |
| 257 | u64 last = 0; |
| 258 | u64 hint_last; |
| 259 | u64 start; |
| 260 | u64 end; |
| 261 | u64 free_check; |
| 262 | u64 ptr; |
| 263 | int bit; |
| 264 | int ret; |
| 265 | int full_search = 0; |
| 266 | int factor = 8; |
| 267 | int data_swap = 0; |
| 268 | |
| 269 | block_group_cache = &info->block_group_cache; |
| 270 | |
| 271 | if (!owner) |
| 272 | factor = 8; |
| 273 | |
| 274 | if (data == BTRFS_BLOCK_GROUP_MIXED) { |
| 275 | bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA; |
| 276 | factor = 10; |
| 277 | } else if (data) |
| 278 | bit = BLOCK_GROUP_DATA; |
| 279 | else |
| 280 | bit = BLOCK_GROUP_METADATA; |
| 281 | |
| 282 | if (search_start) { |
| 283 | struct btrfs_block_group_cache *shint; |
| 284 | shint = btrfs_lookup_block_group(info, search_start); |
| 285 | if (shint && (shint->data == data || |
| 286 | shint->data == BTRFS_BLOCK_GROUP_MIXED)) { |
| 287 | used = btrfs_block_group_used(&shint->item); |
| 288 | if (used + shint->pinned < |
| 289 | div_factor(shint->key.offset, factor)) { |
| 290 | return shint; |
| 291 | } |
| 292 | } |
| 293 | } |
| 294 | if (hint && (hint->data == data || |
| 295 | hint->data == BTRFS_BLOCK_GROUP_MIXED)) { |
| 296 | used = btrfs_block_group_used(&hint->item); |
| 297 | if (used + hint->pinned < |
| 298 | div_factor(hint->key.offset, factor)) { |
| 299 | return hint; |
| 300 | } |
| 301 | last = hint->key.objectid + hint->key.offset; |
| 302 | hint_last = last; |
| 303 | } else { |
| 304 | if (hint) |
| 305 | hint_last = max(hint->key.objectid, search_start); |
| 306 | else |
| 307 | hint_last = search_start; |
| 308 | |
| 309 | last = hint_last; |
| 310 | } |
| 311 | again: |
| 312 | while(1) { |
| 313 | ret = find_first_extent_bit(block_group_cache, last, |
| 314 | &start, &end, bit); |
| 315 | if (ret) |
| 316 | break; |
| 317 | |
| 318 | ret = get_state_private(block_group_cache, start, &ptr); |
| 319 | if (ret) |
| 320 | break; |
| 321 | |
| 322 | cache = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
| 323 | last = cache->key.objectid + cache->key.offset; |
| 324 | used = btrfs_block_group_used(&cache->item); |
| 325 | |
| 326 | if (full_search) |
| 327 | free_check = cache->key.offset; |
| 328 | else |
| 329 | free_check = div_factor(cache->key.offset, factor); |
| 330 | if (used + cache->pinned < free_check) { |
| 331 | found_group = cache; |
| 332 | goto found; |
| 333 | } |
| 334 | cond_resched(); |
| 335 | } |
| 336 | if (!full_search) { |
| 337 | last = search_start; |
| 338 | full_search = 1; |
| 339 | goto again; |
| 340 | } |
| 341 | if (!data_swap) { |
| 342 | data_swap = 1; |
| 343 | bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA; |
| 344 | last = search_start; |
| 345 | goto again; |
| 346 | } |
| 347 | found: |
| 348 | return found_group; |
| 349 | } |
| 350 | |
| 351 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
| 352 | struct btrfs_root *root, |
| 353 | u64 bytenr, u64 num_bytes) |
| 354 | { |
| 355 | struct btrfs_path *path; |
| 356 | int ret; |
| 357 | struct btrfs_key key; |
| 358 | struct extent_buffer *l; |
| 359 | struct btrfs_extent_item *item; |
| 360 | u32 refs; |
| 361 | |
| 362 | WARN_ON(num_bytes < root->sectorsize); |
| 363 | path = btrfs_alloc_path(); |
| 364 | if (!path) |
| 365 | return -ENOMEM; |
| 366 | |
| 367 | key.objectid = bytenr; |
| 368 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
| 369 | key.offset = num_bytes; |
| 370 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
| 371 | 0, 1); |
| 372 | if (ret < 0) |
| 373 | return ret; |
| 374 | if (ret != 0) { |
| 375 | BUG(); |
| 376 | } |
| 377 | BUG_ON(ret != 0); |
| 378 | l = path->nodes[0]; |
| 379 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
| 380 | refs = btrfs_extent_refs(l, item); |
| 381 | btrfs_set_extent_refs(l, item, refs + 1); |
| 382 | btrfs_mark_buffer_dirty(path->nodes[0]); |
| 383 | |
| 384 | btrfs_release_path(root->fs_info->extent_root, path); |
| 385 | btrfs_free_path(path); |
| 386 | finish_current_insert(trans, root->fs_info->extent_root); |
| 387 | del_pending_extents(trans, root->fs_info->extent_root); |
| 388 | return 0; |
| 389 | } |
| 390 | |
| 391 | int btrfs_extent_post_op(struct btrfs_trans_handle *trans, |
| 392 | struct btrfs_root *root) |
| 393 | { |
| 394 | finish_current_insert(trans, root->fs_info->extent_root); |
| 395 | del_pending_extents(trans, root->fs_info->extent_root); |
| 396 | return 0; |
| 397 | } |
| 398 | |
| 399 | static int lookup_extent_ref(struct btrfs_trans_handle *trans, |
| 400 | struct btrfs_root *root, u64 bytenr, |
| 401 | u64 num_bytes, u32 *refs) |
| 402 | { |
| 403 | struct btrfs_path *path; |
| 404 | int ret; |
| 405 | struct btrfs_key key; |
| 406 | struct extent_buffer *l; |
| 407 | struct btrfs_extent_item *item; |
| 408 | |
| 409 | WARN_ON(num_bytes < root->sectorsize); |
| 410 | path = btrfs_alloc_path(); |
| 411 | key.objectid = bytenr; |
| 412 | key.offset = num_bytes; |
| 413 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
| 414 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
| 415 | 0, 0); |
| 416 | if (ret < 0) |
| 417 | goto out; |
| 418 | if (ret != 0) { |
| 419 | btrfs_print_leaf(root, path->nodes[0]); |
| 420 | printk("failed to find block number %Lu\n", bytenr); |
| 421 | BUG(); |
| 422 | } |
| 423 | l = path->nodes[0]; |
| 424 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
| 425 | *refs = btrfs_extent_refs(l, item); |
| 426 | out: |
| 427 | btrfs_free_path(path); |
| 428 | return 0; |
| 429 | } |
| 430 | |
| 431 | int btrfs_inc_root_ref(struct btrfs_trans_handle *trans, |
| 432 | struct btrfs_root *root) |
| 433 | { |
| 434 | return btrfs_inc_extent_ref(trans, root, root->node->start, |
| 435 | root->node->len); |
| 436 | } |
| 437 | |
| 438 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
| 439 | struct extent_buffer *buf) |
| 440 | { |
| 441 | u64 bytenr; |
| 442 | u32 nritems; |
| 443 | struct btrfs_key key; |
| 444 | struct btrfs_file_extent_item *fi; |
| 445 | int i; |
| 446 | int level; |
| 447 | int ret; |
| 448 | int faili; |
| 449 | int err; |
| 450 | |
| 451 | if (!root->ref_cows) |
| 452 | return 0; |
| 453 | |
| 454 | level = btrfs_header_level(buf); |
| 455 | nritems = btrfs_header_nritems(buf); |
| 456 | for (i = 0; i < nritems; i++) { |
| 457 | if (level == 0) { |
| 458 | u64 disk_bytenr; |
| 459 | btrfs_item_key_to_cpu(buf, &key, i); |
| 460 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
| 461 | continue; |
| 462 | fi = btrfs_item_ptr(buf, i, |
| 463 | struct btrfs_file_extent_item); |
| 464 | if (btrfs_file_extent_type(buf, fi) == |
| 465 | BTRFS_FILE_EXTENT_INLINE) |
| 466 | continue; |
| 467 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
| 468 | if (disk_bytenr == 0) |
| 469 | continue; |
| 470 | ret = btrfs_inc_extent_ref(trans, root, disk_bytenr, |
| 471 | btrfs_file_extent_disk_num_bytes(buf, fi)); |
| 472 | if (ret) { |
| 473 | faili = i; |
| 474 | goto fail; |
| 475 | } |
| 476 | } else { |
| 477 | bytenr = btrfs_node_blockptr(buf, i); |
| 478 | ret = btrfs_inc_extent_ref(trans, root, bytenr, |
| 479 | btrfs_level_size(root, level - 1)); |
| 480 | if (ret) { |
| 481 | faili = i; |
| 482 | goto fail; |
| 483 | } |
| 484 | } |
| 485 | } |
| 486 | return 0; |
| 487 | fail: |
| 488 | WARN_ON(1); |
| 489 | for (i =0; i < faili; i++) { |
| 490 | if (level == 0) { |
| 491 | u64 disk_bytenr; |
| 492 | btrfs_item_key_to_cpu(buf, &key, i); |
| 493 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
| 494 | continue; |
| 495 | fi = btrfs_item_ptr(buf, i, |
| 496 | struct btrfs_file_extent_item); |
| 497 | if (btrfs_file_extent_type(buf, fi) == |
| 498 | BTRFS_FILE_EXTENT_INLINE) |
| 499 | continue; |
| 500 | disk_bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
| 501 | if (disk_bytenr == 0) |
| 502 | continue; |
| 503 | err = btrfs_free_extent(trans, root, disk_bytenr, |
| 504 | btrfs_file_extent_disk_num_bytes(buf, |
| 505 | fi), 0); |
| 506 | BUG_ON(err); |
| 507 | } else { |
| 508 | bytenr = btrfs_node_blockptr(buf, i); |
| 509 | err = btrfs_free_extent(trans, root, bytenr, |
| 510 | btrfs_level_size(root, level - 1), 0); |
| 511 | BUG_ON(err); |
| 512 | } |
| 513 | } |
| 514 | return ret; |
| 515 | } |
| 516 | |
| 517 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
| 518 | struct btrfs_root *root, |
| 519 | struct btrfs_path *path, |
| 520 | struct btrfs_block_group_cache *cache) |
| 521 | { |
| 522 | int ret; |
| 523 | int pending_ret; |
| 524 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
| 525 | unsigned long bi; |
| 526 | struct extent_buffer *leaf; |
| 527 | |
| 528 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
| 529 | if (ret < 0) |
| 530 | goto fail; |
| 531 | BUG_ON(ret); |
| 532 | |
| 533 | leaf = path->nodes[0]; |
| 534 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); |
| 535 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); |
| 536 | btrfs_mark_buffer_dirty(leaf); |
| 537 | btrfs_release_path(extent_root, path); |
| 538 | fail: |
| 539 | finish_current_insert(trans, extent_root); |
| 540 | pending_ret = del_pending_extents(trans, extent_root); |
| 541 | if (ret) |
| 542 | return ret; |
| 543 | if (pending_ret) |
| 544 | return pending_ret; |
| 545 | return 0; |
| 546 | |
| 547 | } |
| 548 | |
| 549 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
| 550 | struct btrfs_root *root) |
| 551 | { |
| 552 | struct extent_map_tree *block_group_cache; |
| 553 | struct btrfs_block_group_cache *cache; |
| 554 | int ret; |
| 555 | int err = 0; |
| 556 | int werr = 0; |
| 557 | struct btrfs_path *path; |
| 558 | u64 last = 0; |
| 559 | u64 start; |
| 560 | u64 end; |
| 561 | u64 ptr; |
| 562 | |
| 563 | block_group_cache = &root->fs_info->block_group_cache; |
| 564 | path = btrfs_alloc_path(); |
| 565 | if (!path) |
| 566 | return -ENOMEM; |
| 567 | |
| 568 | while(1) { |
| 569 | ret = find_first_extent_bit(block_group_cache, last, |
| 570 | &start, &end, BLOCK_GROUP_DIRTY); |
| 571 | if (ret) |
| 572 | break; |
| 573 | |
| 574 | last = end + 1; |
| 575 | ret = get_state_private(block_group_cache, start, &ptr); |
| 576 | if (ret) |
| 577 | break; |
| 578 | |
| 579 | cache = (struct btrfs_block_group_cache *)(unsigned long)ptr; |
| 580 | err = write_one_cache_group(trans, root, |
| 581 | path, cache); |
| 582 | /* |
| 583 | * if we fail to write the cache group, we want |
| 584 | * to keep it marked dirty in hopes that a later |
| 585 | * write will work |
| 586 | */ |
| 587 | if (err) { |
| 588 | werr = err; |
| 589 | continue; |
| 590 | } |
| 591 | clear_extent_bits(block_group_cache, start, end, |
| 592 | BLOCK_GROUP_DIRTY, GFP_NOFS); |
| 593 | } |
| 594 | btrfs_free_path(path); |
| 595 | return werr; |
| 596 | } |
| 597 | |
| 598 | static int update_block_group(struct btrfs_trans_handle *trans, |
| 599 | struct btrfs_root *root, |
| 600 | u64 bytenr, u64 num_bytes, int alloc, |
| 601 | int mark_free, int data) |
| 602 | { |
| 603 | struct btrfs_block_group_cache *cache; |
| 604 | struct btrfs_fs_info *info = root->fs_info; |
| 605 | u64 total = num_bytes; |
| 606 | u64 old_val; |
| 607 | u64 byte_in_group; |
| 608 | u64 start; |
| 609 | u64 end; |
| 610 | |
| 611 | while(total) { |
| 612 | cache = btrfs_lookup_block_group(info, bytenr); |
| 613 | if (!cache) { |
| 614 | return -1; |
| 615 | } |
| 616 | byte_in_group = bytenr - cache->key.objectid; |
| 617 | WARN_ON(byte_in_group > cache->key.offset); |
| 618 | start = cache->key.objectid; |
| 619 | end = start + cache->key.offset - 1; |
| 620 | set_extent_bits(&info->block_group_cache, start, end, |
| 621 | BLOCK_GROUP_DIRTY, GFP_NOFS); |
| 622 | |
| 623 | old_val = btrfs_block_group_used(&cache->item); |
| 624 | num_bytes = min(total, cache->key.offset - byte_in_group); |
| 625 | if (alloc) { |
| 626 | if (cache->data != data && |
| 627 | old_val < (cache->key.offset >> 1)) { |
| 628 | int bit_to_clear; |
| 629 | int bit_to_set; |
| 630 | cache->data = data; |
| 631 | if (data) { |
| 632 | bit_to_clear = BLOCK_GROUP_METADATA; |
| 633 | bit_to_set = BLOCK_GROUP_DATA; |
| 634 | cache->item.flags &= |
| 635 | ~BTRFS_BLOCK_GROUP_MIXED; |
| 636 | cache->item.flags |= |
| 637 | BTRFS_BLOCK_GROUP_DATA; |
| 638 | } else { |
| 639 | bit_to_clear = BLOCK_GROUP_DATA; |
| 640 | bit_to_set = BLOCK_GROUP_METADATA; |
| 641 | cache->item.flags &= |
| 642 | ~BTRFS_BLOCK_GROUP_MIXED; |
| 643 | cache->item.flags &= |
| 644 | ~BTRFS_BLOCK_GROUP_DATA; |
| 645 | } |
| 646 | clear_extent_bits(&info->block_group_cache, |
| 647 | start, end, bit_to_clear, |
| 648 | GFP_NOFS); |
| 649 | set_extent_bits(&info->block_group_cache, |
| 650 | start, end, bit_to_set, |
| 651 | GFP_NOFS); |
| 652 | } else if (cache->data != data && |
| 653 | cache->data != BTRFS_BLOCK_GROUP_MIXED) { |
| 654 | cache->data = BTRFS_BLOCK_GROUP_MIXED; |
| 655 | set_extent_bits(&info->block_group_cache, |
| 656 | start, end, |
| 657 | BLOCK_GROUP_DATA | |
| 658 | BLOCK_GROUP_METADATA, |
| 659 | GFP_NOFS); |
| 660 | } |
| 661 | old_val += num_bytes; |
| 662 | } else { |
| 663 | old_val -= num_bytes; |
| 664 | if (mark_free) { |
| 665 | set_extent_dirty(&info->free_space_cache, |
| 666 | bytenr, bytenr + num_bytes - 1, |
| 667 | GFP_NOFS); |
| 668 | } |
| 669 | } |
| 670 | btrfs_set_block_group_used(&cache->item, old_val); |
| 671 | total -= num_bytes; |
| 672 | bytenr += num_bytes; |
| 673 | } |
| 674 | return 0; |
| 675 | } |
| 676 | static int update_pinned_extents(struct btrfs_root *root, |
| 677 | u64 bytenr, u64 num, int pin) |
| 678 | { |
| 679 | u64 len; |
| 680 | struct btrfs_block_group_cache *cache; |
| 681 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 682 | |
| 683 | if (pin) { |
| 684 | set_extent_dirty(&fs_info->pinned_extents, |
| 685 | bytenr, bytenr + num - 1, GFP_NOFS); |
| 686 | } else { |
| 687 | clear_extent_dirty(&fs_info->pinned_extents, |
| 688 | bytenr, bytenr + num - 1, GFP_NOFS); |
| 689 | } |
| 690 | while (num > 0) { |
| 691 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
| 692 | WARN_ON(!cache); |
| 693 | len = min(num, cache->key.offset - |
| 694 | (bytenr - cache->key.objectid)); |
| 695 | if (pin) { |
| 696 | cache->pinned += len; |
| 697 | fs_info->total_pinned += len; |
| 698 | } else { |
| 699 | cache->pinned -= len; |
| 700 | fs_info->total_pinned -= len; |
| 701 | } |
| 702 | bytenr += len; |
| 703 | num -= len; |
| 704 | } |
| 705 | return 0; |
| 706 | } |
| 707 | |
| 708 | int btrfs_copy_pinned(struct btrfs_root *root, struct extent_map_tree *copy) |
| 709 | { |
| 710 | u64 last = 0; |
| 711 | u64 start; |
| 712 | u64 end; |
| 713 | struct extent_map_tree *pinned_extents = &root->fs_info->pinned_extents; |
| 714 | int ret; |
| 715 | |
| 716 | while(1) { |
| 717 | ret = find_first_extent_bit(pinned_extents, last, |
| 718 | &start, &end, EXTENT_DIRTY); |
| 719 | if (ret) |
| 720 | break; |
| 721 | set_extent_dirty(copy, start, end, GFP_NOFS); |
| 722 | last = end + 1; |
| 723 | } |
| 724 | return 0; |
| 725 | } |
| 726 | |
| 727 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, |
| 728 | struct btrfs_root *root, |
| 729 | struct extent_map_tree *unpin) |
| 730 | { |
| 731 | u64 start; |
| 732 | u64 end; |
| 733 | int ret; |
| 734 | struct extent_map_tree *free_space_cache; |
| 735 | free_space_cache = &root->fs_info->free_space_cache; |
| 736 | |
| 737 | while(1) { |
| 738 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
| 739 | EXTENT_DIRTY); |
| 740 | if (ret) |
| 741 | break; |
| 742 | update_pinned_extents(root, start, end + 1 - start, 0); |
| 743 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
| 744 | set_extent_dirty(free_space_cache, start, end, GFP_NOFS); |
| 745 | } |
| 746 | return 0; |
| 747 | } |
| 748 | |
| 749 | static int finish_current_insert(struct btrfs_trans_handle *trans, struct |
| 750 | btrfs_root *extent_root) |
| 751 | { |
| 752 | struct btrfs_key ins; |
| 753 | struct btrfs_extent_item extent_item; |
| 754 | int ret; |
| 755 | int err = 0; |
| 756 | u64 start; |
| 757 | u64 end; |
| 758 | struct btrfs_fs_info *info = extent_root->fs_info; |
| 759 | |
| 760 | btrfs_set_stack_extent_refs(&extent_item, 1); |
| 761 | btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY); |
| 762 | btrfs_set_stack_extent_owner(&extent_item, |
| 763 | extent_root->root_key.objectid); |
| 764 | |
| 765 | while(1) { |
| 766 | ret = find_first_extent_bit(&info->extent_ins, 0, &start, |
| 767 | &end, EXTENT_LOCKED); |
| 768 | if (ret) |
| 769 | break; |
| 770 | |
| 771 | ins.objectid = start; |
| 772 | ins.offset = end + 1 - start; |
| 773 | err = btrfs_insert_item(trans, extent_root, &ins, |
| 774 | &extent_item, sizeof(extent_item)); |
| 775 | clear_extent_bits(&info->extent_ins, start, end, EXTENT_LOCKED, |
| 776 | GFP_NOFS); |
| 777 | } |
| 778 | return 0; |
| 779 | } |
| 780 | |
| 781 | static int pin_down_bytes(struct btrfs_root *root, u64 bytenr, u32 num_bytes, |
| 782 | int pending) |
| 783 | { |
| 784 | int err = 0; |
| 785 | struct extent_buffer *buf; |
| 786 | |
| 787 | if (!pending) { |
| 788 | buf = btrfs_find_tree_block(root, bytenr, num_bytes); |
| 789 | if (buf) { |
| 790 | if (btrfs_buffer_uptodate(buf)) { |
| 791 | u64 transid = |
| 792 | root->fs_info->running_transaction->transid; |
| 793 | if (btrfs_header_generation(buf) == transid) { |
| 794 | free_extent_buffer(buf); |
| 795 | return 1; |
| 796 | } |
| 797 | } |
| 798 | free_extent_buffer(buf); |
| 799 | } |
| 800 | update_pinned_extents(root, bytenr, num_bytes, 1); |
| 801 | } else { |
| 802 | set_extent_bits(&root->fs_info->pending_del, |
| 803 | bytenr, bytenr + num_bytes - 1, |
| 804 | EXTENT_LOCKED, GFP_NOFS); |
| 805 | } |
| 806 | BUG_ON(err < 0); |
| 807 | return 0; |
| 808 | } |
| 809 | |
| 810 | /* |
| 811 | * remove an extent from the root, returns 0 on success |
| 812 | */ |
| 813 | static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root |
| 814 | *root, u64 bytenr, u64 num_bytes, int pin, |
| 815 | int mark_free) |
| 816 | { |
| 817 | struct btrfs_path *path; |
| 818 | struct btrfs_key key; |
| 819 | struct btrfs_fs_info *info = root->fs_info; |
| 820 | struct btrfs_root *extent_root = info->extent_root; |
| 821 | struct extent_buffer *leaf; |
| 822 | int ret; |
| 823 | struct btrfs_extent_item *ei; |
| 824 | u32 refs; |
| 825 | |
| 826 | key.objectid = bytenr; |
| 827 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
| 828 | key.offset = num_bytes; |
| 829 | |
| 830 | path = btrfs_alloc_path(); |
| 831 | if (!path) |
| 832 | return -ENOMEM; |
| 833 | |
| 834 | ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); |
| 835 | if (ret < 0) |
| 836 | return ret; |
| 837 | BUG_ON(ret); |
| 838 | |
| 839 | leaf = path->nodes[0]; |
| 840 | ei = btrfs_item_ptr(leaf, path->slots[0], |
| 841 | struct btrfs_extent_item); |
| 842 | refs = btrfs_extent_refs(leaf, ei); |
| 843 | BUG_ON(refs == 0); |
| 844 | refs -= 1; |
| 845 | btrfs_set_extent_refs(leaf, ei, refs); |
| 846 | btrfs_mark_buffer_dirty(leaf); |
| 847 | |
| 848 | if (refs == 0) { |
| 849 | u64 super_used; |
| 850 | u64 root_used; |
| 851 | |
| 852 | if (pin) { |
| 853 | ret = pin_down_bytes(root, bytenr, num_bytes, 0); |
| 854 | if (ret > 0) |
| 855 | mark_free = 1; |
| 856 | BUG_ON(ret < 0); |
| 857 | } |
| 858 | |
| 859 | /* block accounting for super block */ |
| 860 | super_used = btrfs_super_bytes_used(&info->super_copy); |
| 861 | btrfs_set_super_bytes_used(&info->super_copy, |
| 862 | super_used - num_bytes); |
| 863 | |
| 864 | /* block accounting for root item */ |
| 865 | root_used = btrfs_root_used(&root->root_item); |
| 866 | btrfs_set_root_used(&root->root_item, |
| 867 | root_used - num_bytes); |
| 868 | |
| 869 | ret = btrfs_del_item(trans, extent_root, path); |
| 870 | if (ret) { |
| 871 | return ret; |
| 872 | } |
| 873 | ret = update_block_group(trans, root, bytenr, num_bytes, 0, |
| 874 | mark_free, 0); |
| 875 | BUG_ON(ret); |
| 876 | } |
| 877 | btrfs_free_path(path); |
| 878 | finish_current_insert(trans, extent_root); |
| 879 | return ret; |
| 880 | } |
| 881 | |
| 882 | /* |
| 883 | * find all the blocks marked as pending in the radix tree and remove |
| 884 | * them from the extent map |
| 885 | */ |
| 886 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
| 887 | btrfs_root *extent_root) |
| 888 | { |
| 889 | int ret; |
| 890 | int err = 0; |
| 891 | u64 start; |
| 892 | u64 end; |
| 893 | struct extent_map_tree *pending_del; |
| 894 | struct extent_map_tree *pinned_extents; |
| 895 | |
| 896 | pending_del = &extent_root->fs_info->pending_del; |
| 897 | pinned_extents = &extent_root->fs_info->pinned_extents; |
| 898 | |
| 899 | while(1) { |
| 900 | ret = find_first_extent_bit(pending_del, 0, &start, &end, |
| 901 | EXTENT_LOCKED); |
| 902 | if (ret) |
| 903 | break; |
| 904 | update_pinned_extents(extent_root, start, end + 1 - start, 1); |
| 905 | clear_extent_bits(pending_del, start, end, EXTENT_LOCKED, |
| 906 | GFP_NOFS); |
| 907 | ret = __free_extent(trans, extent_root, |
| 908 | start, end + 1 - start, 0, 0); |
| 909 | if (ret) |
| 910 | err = ret; |
| 911 | } |
| 912 | return err; |
| 913 | } |
| 914 | |
| 915 | /* |
| 916 | * remove an extent from the root, returns 0 on success |
| 917 | */ |
| 918 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root |
| 919 | *root, u64 bytenr, u64 num_bytes, int pin) |
| 920 | { |
| 921 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
| 922 | int pending_ret; |
| 923 | int ret; |
| 924 | |
| 925 | WARN_ON(num_bytes < root->sectorsize); |
| 926 | if (root == extent_root) { |
| 927 | pin_down_bytes(root, bytenr, num_bytes, 1); |
| 928 | return 0; |
| 929 | } |
| 930 | ret = __free_extent(trans, root, bytenr, num_bytes, pin, pin == 0); |
| 931 | pending_ret = del_pending_extents(trans, root->fs_info->extent_root); |
| 932 | return ret ? ret : pending_ret; |
| 933 | } |
| 934 | |
| 935 | static u64 stripe_align(struct btrfs_root *root, u64 val) |
| 936 | { |
| 937 | u64 mask = ((u64)root->stripesize - 1); |
| 938 | u64 ret = (val + mask) & ~mask; |
| 939 | return ret; |
| 940 | } |
| 941 | |
| 942 | /* |
| 943 | * walks the btree of allocated extents and find a hole of a given size. |
| 944 | * The key ins is changed to record the hole: |
| 945 | * ins->objectid == block start |
| 946 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
| 947 | * ins->offset == number of blocks |
| 948 | * Any available blocks before search_start are skipped. |
| 949 | */ |
| 950 | static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root |
| 951 | *orig_root, u64 num_bytes, u64 empty_size, |
| 952 | u64 search_start, u64 search_end, u64 hint_byte, |
| 953 | struct btrfs_key *ins, u64 exclude_start, |
| 954 | u64 exclude_nr, int data) |
| 955 | { |
| 956 | struct btrfs_path *path; |
| 957 | struct btrfs_key key; |
| 958 | u64 hole_size = 0; |
| 959 | u64 aligned; |
| 960 | int ret; |
| 961 | int slot = 0; |
| 962 | u64 last_byte = 0; |
| 963 | u64 orig_search_start = search_start; |
| 964 | int start_found; |
| 965 | struct extent_buffer *l; |
| 966 | struct btrfs_root * root = orig_root->fs_info->extent_root; |
| 967 | struct btrfs_fs_info *info = root->fs_info; |
| 968 | u64 total_needed = num_bytes; |
| 969 | int level; |
| 970 | struct btrfs_block_group_cache *block_group; |
| 971 | int full_scan = 0; |
| 972 | int wrapped = 0; |
| 973 | u64 cached_start; |
| 974 | |
| 975 | WARN_ON(num_bytes < root->sectorsize); |
| 976 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
| 977 | |
| 978 | level = btrfs_header_level(root->node); |
| 979 | |
| 980 | if (num_bytes >= 32 * 1024 * 1024 && hint_byte) { |
| 981 | data = BTRFS_BLOCK_GROUP_MIXED; |
| 982 | } |
| 983 | |
| 984 | if (search_end == (u64)-1) |
| 985 | search_end = btrfs_super_total_bytes(&info->super_copy); |
| 986 | if (hint_byte) { |
| 987 | block_group = btrfs_lookup_block_group(info, hint_byte); |
| 988 | block_group = btrfs_find_block_group(root, block_group, |
| 989 | hint_byte, data, 1); |
| 990 | } else { |
| 991 | block_group = btrfs_find_block_group(root, |
| 992 | trans->block_group, 0, |
| 993 | data, 1); |
| 994 | } |
| 995 | |
| 996 | total_needed += empty_size; |
| 997 | path = btrfs_alloc_path(); |
| 998 | check_failed: |
| 999 | search_start = find_search_start(root, &block_group, search_start, |
| 1000 | total_needed, data, full_scan); |
| 1001 | search_start = stripe_align(root, search_start); |
| 1002 | cached_start = search_start; |
| 1003 | btrfs_init_path(path); |
| 1004 | ins->objectid = search_start; |
| 1005 | ins->offset = 0; |
| 1006 | start_found = 0; |
| 1007 | path->reada = 2; |
| 1008 | |
| 1009 | ret = btrfs_search_slot(trans, root, ins, path, 0, 0); |
| 1010 | if (ret < 0) |
| 1011 | goto error; |
| 1012 | |
| 1013 | if (path->slots[0] > 0) { |
| 1014 | path->slots[0]--; |
| 1015 | } |
| 1016 | |
| 1017 | l = path->nodes[0]; |
| 1018 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); |
| 1019 | |
| 1020 | /* |
| 1021 | * a rare case, go back one key if we hit a block group item |
| 1022 | * instead of an extent item |
| 1023 | */ |
| 1024 | if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY && |
| 1025 | key.objectid + key.offset >= search_start) { |
| 1026 | ins->objectid = key.objectid; |
| 1027 | ins->offset = key.offset - 1; |
| 1028 | btrfs_release_path(root, path); |
| 1029 | ret = btrfs_search_slot(trans, root, ins, path, 0, 0); |
| 1030 | if (ret < 0) |
| 1031 | goto error; |
| 1032 | |
| 1033 | if (path->slots[0] > 0) { |
| 1034 | path->slots[0]--; |
| 1035 | } |
| 1036 | } |
| 1037 | |
| 1038 | while (1) { |
| 1039 | l = path->nodes[0]; |
| 1040 | slot = path->slots[0]; |
| 1041 | if (slot >= btrfs_header_nritems(l)) { |
| 1042 | ret = btrfs_next_leaf(root, path); |
| 1043 | if (ret == 0) |
| 1044 | continue; |
| 1045 | if (ret < 0) |
| 1046 | goto error; |
| 1047 | |
| 1048 | search_start = max(search_start, |
| 1049 | block_group->key.objectid); |
| 1050 | if (!start_found) { |
| 1051 | aligned = stripe_align(root, search_start); |
| 1052 | ins->objectid = aligned; |
| 1053 | if (aligned >= search_end) { |
| 1054 | ret = -ENOSPC; |
| 1055 | goto error; |
| 1056 | } |
| 1057 | ins->offset = search_end - aligned; |
| 1058 | start_found = 1; |
| 1059 | goto check_pending; |
| 1060 | } |
| 1061 | ins->objectid = stripe_align(root, |
| 1062 | last_byte > search_start ? |
| 1063 | last_byte : search_start); |
| 1064 | if (search_end <= ins->objectid) { |
| 1065 | ret = -ENOSPC; |
| 1066 | goto error; |
| 1067 | } |
| 1068 | ins->offset = search_end - ins->objectid; |
| 1069 | BUG_ON(ins->objectid >= search_end); |
| 1070 | goto check_pending; |
| 1071 | } |
| 1072 | btrfs_item_key_to_cpu(l, &key, slot); |
| 1073 | |
| 1074 | if (key.objectid >= search_start && key.objectid > last_byte && |
| 1075 | start_found) { |
| 1076 | if (last_byte < search_start) |
| 1077 | last_byte = search_start; |
| 1078 | aligned = stripe_align(root, last_byte); |
| 1079 | hole_size = key.objectid - aligned; |
| 1080 | if (key.objectid > aligned && hole_size >= num_bytes) { |
| 1081 | ins->objectid = aligned; |
| 1082 | ins->offset = hole_size; |
| 1083 | goto check_pending; |
| 1084 | } |
| 1085 | } |
| 1086 | if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) { |
| 1087 | if (!start_found) { |
| 1088 | last_byte = key.objectid; |
| 1089 | start_found = 1; |
| 1090 | } |
| 1091 | goto next; |
| 1092 | } |
| 1093 | |
| 1094 | |
| 1095 | start_found = 1; |
| 1096 | last_byte = key.objectid + key.offset; |
| 1097 | |
| 1098 | if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED && |
| 1099 | last_byte >= block_group->key.objectid + |
| 1100 | block_group->key.offset) { |
| 1101 | btrfs_release_path(root, path); |
| 1102 | search_start = block_group->key.objectid + |
| 1103 | block_group->key.offset; |
| 1104 | goto new_group; |
| 1105 | } |
| 1106 | next: |
| 1107 | path->slots[0]++; |
| 1108 | cond_resched(); |
| 1109 | } |
| 1110 | check_pending: |
| 1111 | /* we have to make sure we didn't find an extent that has already |
| 1112 | * been allocated by the map tree or the original allocation |
| 1113 | */ |
| 1114 | btrfs_release_path(root, path); |
| 1115 | BUG_ON(ins->objectid < search_start); |
| 1116 | |
| 1117 | if (ins->objectid + num_bytes >= search_end) |
| 1118 | goto enospc; |
| 1119 | if (!full_scan && data != BTRFS_BLOCK_GROUP_MIXED && |
| 1120 | ins->objectid + num_bytes > block_group-> |
| 1121 | key.objectid + block_group->key.offset) { |
| 1122 | search_start = block_group->key.objectid + |
| 1123 | block_group->key.offset; |
| 1124 | goto new_group; |
| 1125 | } |
| 1126 | if (test_range_bit(&info->extent_ins, ins->objectid, |
| 1127 | ins->objectid + num_bytes -1, EXTENT_LOCKED, 0)) { |
| 1128 | search_start = ins->objectid + num_bytes; |
| 1129 | goto new_group; |
| 1130 | } |
| 1131 | if (test_range_bit(&info->pinned_extents, ins->objectid, |
| 1132 | ins->objectid + num_bytes -1, EXTENT_DIRTY, 0)) { |
| 1133 | search_start = ins->objectid + num_bytes; |
| 1134 | goto new_group; |
| 1135 | } |
| 1136 | if (exclude_nr > 0 && (ins->objectid + num_bytes > exclude_start && |
| 1137 | ins->objectid < exclude_start + exclude_nr)) { |
| 1138 | search_start = exclude_start + exclude_nr; |
| 1139 | goto new_group; |
| 1140 | } |
| 1141 | if (!data) { |
| 1142 | block_group = btrfs_lookup_block_group(info, ins->objectid); |
| 1143 | if (block_group) |
| 1144 | trans->block_group = block_group; |
| 1145 | } |
| 1146 | ins->offset = num_bytes; |
| 1147 | btrfs_free_path(path); |
| 1148 | return 0; |
| 1149 | |
| 1150 | new_group: |
| 1151 | if (search_start + num_bytes >= search_end) { |
| 1152 | enospc: |
| 1153 | search_start = orig_search_start; |
| 1154 | if (full_scan) { |
| 1155 | ret = -ENOSPC; |
| 1156 | goto error; |
| 1157 | } |
| 1158 | if (wrapped) { |
| 1159 | if (!full_scan) |
| 1160 | total_needed -= empty_size; |
| 1161 | full_scan = 1; |
| 1162 | } else |
| 1163 | wrapped = 1; |
| 1164 | } |
| 1165 | block_group = btrfs_lookup_block_group(info, search_start); |
| 1166 | cond_resched(); |
| 1167 | if (!full_scan) |
| 1168 | block_group = btrfs_find_block_group(root, block_group, |
| 1169 | search_start, data, 0); |
| 1170 | goto check_failed; |
| 1171 | |
| 1172 | error: |
| 1173 | btrfs_release_path(root, path); |
| 1174 | btrfs_free_path(path); |
| 1175 | return ret; |
| 1176 | } |
| 1177 | /* |
| 1178 | * finds a free extent and does all the dirty work required for allocation |
| 1179 | * returns the key for the extent through ins, and a tree buffer for |
| 1180 | * the first block of the extent through buf. |
| 1181 | * |
| 1182 | * returns 0 if everything worked, non-zero otherwise. |
| 1183 | */ |
| 1184 | int btrfs_alloc_extent(struct btrfs_trans_handle *trans, |
| 1185 | struct btrfs_root *root, u64 owner, |
| 1186 | u64 num_bytes, u64 empty_size, u64 hint_byte, |
| 1187 | u64 search_end, struct btrfs_key *ins, int data) |
| 1188 | { |
| 1189 | int ret; |
| 1190 | int pending_ret; |
| 1191 | u64 super_used, root_used; |
| 1192 | u64 search_start = 0; |
| 1193 | struct btrfs_fs_info *info = root->fs_info; |
| 1194 | struct btrfs_root *extent_root = info->extent_root; |
| 1195 | struct btrfs_extent_item extent_item; |
| 1196 | |
| 1197 | btrfs_set_stack_extent_refs(&extent_item, 1); |
| 1198 | btrfs_set_stack_extent_owner(&extent_item, owner); |
| 1199 | |
| 1200 | WARN_ON(num_bytes < root->sectorsize); |
| 1201 | ret = find_free_extent(trans, root, num_bytes, empty_size, |
| 1202 | search_start, search_end, hint_byte, ins, |
| 1203 | trans->alloc_exclude_start, |
| 1204 | trans->alloc_exclude_nr, data); |
| 1205 | BUG_ON(ret); |
| 1206 | if (ret) |
| 1207 | return ret; |
| 1208 | |
| 1209 | /* block accounting for super block */ |
| 1210 | super_used = btrfs_super_bytes_used(&info->super_copy); |
| 1211 | btrfs_set_super_bytes_used(&info->super_copy, super_used + num_bytes); |
| 1212 | |
| 1213 | /* block accounting for root item */ |
| 1214 | root_used = btrfs_root_used(&root->root_item); |
| 1215 | btrfs_set_root_used(&root->root_item, root_used + num_bytes); |
| 1216 | |
| 1217 | clear_extent_dirty(&root->fs_info->free_space_cache, |
| 1218 | ins->objectid, ins->objectid + ins->offset - 1, |
| 1219 | GFP_NOFS); |
| 1220 | |
| 1221 | if (root == extent_root) { |
| 1222 | set_extent_bits(&root->fs_info->extent_ins, ins->objectid, |
| 1223 | ins->objectid + ins->offset - 1, |
| 1224 | EXTENT_LOCKED, GFP_NOFS); |
| 1225 | WARN_ON(data == 1); |
| 1226 | goto update_block; |
| 1227 | } |
| 1228 | |
| 1229 | WARN_ON(trans->alloc_exclude_nr); |
| 1230 | trans->alloc_exclude_start = ins->objectid; |
| 1231 | trans->alloc_exclude_nr = ins->offset; |
| 1232 | ret = btrfs_insert_item(trans, extent_root, ins, &extent_item, |
| 1233 | sizeof(extent_item)); |
| 1234 | |
| 1235 | trans->alloc_exclude_start = 0; |
| 1236 | trans->alloc_exclude_nr = 0; |
| 1237 | |
| 1238 | BUG_ON(ret); |
| 1239 | finish_current_insert(trans, extent_root); |
| 1240 | pending_ret = del_pending_extents(trans, extent_root); |
| 1241 | |
| 1242 | if (ret) { |
| 1243 | return ret; |
| 1244 | } |
| 1245 | if (pending_ret) { |
| 1246 | return pending_ret; |
| 1247 | } |
| 1248 | |
| 1249 | update_block: |
| 1250 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0, |
| 1251 | data); |
| 1252 | BUG_ON(ret); |
| 1253 | return 0; |
| 1254 | } |
| 1255 | |
| 1256 | /* |
| 1257 | * helper function to allocate a block for a given tree |
| 1258 | * returns the tree buffer or NULL. |
| 1259 | */ |
| 1260 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
| 1261 | struct btrfs_root *root, |
| 1262 | u32 blocksize, u64 hint, |
| 1263 | u64 empty_size) |
| 1264 | { |
| 1265 | struct btrfs_key ins; |
| 1266 | int ret; |
| 1267 | struct extent_buffer *buf; |
| 1268 | |
| 1269 | ret = btrfs_alloc_extent(trans, root, root->root_key.objectid, |
| 1270 | blocksize, empty_size, hint, |
| 1271 | (u64)-1, &ins, 0); |
| 1272 | if (ret) { |
| 1273 | BUG_ON(ret > 0); |
| 1274 | return ERR_PTR(ret); |
| 1275 | } |
| 1276 | buf = btrfs_find_create_tree_block(root, ins.objectid, blocksize); |
| 1277 | if (!buf) { |
| 1278 | btrfs_free_extent(trans, root, ins.objectid, blocksize, 0); |
| 1279 | return ERR_PTR(-ENOMEM); |
| 1280 | } |
| 1281 | btrfs_set_buffer_uptodate(buf); |
| 1282 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
| 1283 | buf->start + buf->len - 1, GFP_NOFS); |
| 1284 | set_extent_bits(&BTRFS_I(root->fs_info->btree_inode)->extent_tree, |
| 1285 | buf->start, buf->start + buf->len - 1, |
| 1286 | EXTENT_CSUM, GFP_NOFS); |
| 1287 | buf->flags |= EXTENT_CSUM; |
| 1288 | btrfs_set_buffer_defrag(buf); |
| 1289 | trans->blocks_used++; |
| 1290 | return buf; |
| 1291 | } |
| 1292 | |
| 1293 | static int drop_leaf_ref(struct btrfs_trans_handle *trans, |
| 1294 | struct btrfs_root *root, struct extent_buffer *leaf) |
| 1295 | { |
| 1296 | struct btrfs_key key; |
| 1297 | struct btrfs_file_extent_item *fi; |
| 1298 | int i; |
| 1299 | int nritems; |
| 1300 | int ret; |
| 1301 | |
| 1302 | BUG_ON(!btrfs_is_leaf(leaf)); |
| 1303 | nritems = btrfs_header_nritems(leaf); |
| 1304 | for (i = 0; i < nritems; i++) { |
| 1305 | u64 disk_bytenr; |
| 1306 | |
| 1307 | btrfs_item_key_to_cpu(leaf, &key, i); |
| 1308 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
| 1309 | continue; |
| 1310 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); |
| 1311 | if (btrfs_file_extent_type(leaf, fi) == |
| 1312 | BTRFS_FILE_EXTENT_INLINE) |
| 1313 | continue; |
| 1314 | /* |
| 1315 | * FIXME make sure to insert a trans record that |
| 1316 | * repeats the snapshot del on crash |
| 1317 | */ |
| 1318 | disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); |
| 1319 | if (disk_bytenr == 0) |
| 1320 | continue; |
| 1321 | ret = btrfs_free_extent(trans, root, disk_bytenr, |
| 1322 | btrfs_file_extent_disk_num_bytes(leaf, fi), 0); |
| 1323 | BUG_ON(ret); |
| 1324 | } |
| 1325 | return 0; |
| 1326 | } |
| 1327 | |
| 1328 | static void reada_walk_down(struct btrfs_root *root, |
| 1329 | struct extent_buffer *node) |
| 1330 | { |
| 1331 | int i; |
| 1332 | u32 nritems; |
| 1333 | u64 bytenr; |
| 1334 | int ret; |
| 1335 | u32 refs; |
| 1336 | int level; |
| 1337 | u32 blocksize; |
| 1338 | |
| 1339 | nritems = btrfs_header_nritems(node); |
| 1340 | level = btrfs_header_level(node); |
| 1341 | for (i = 0; i < nritems; i++) { |
| 1342 | bytenr = btrfs_node_blockptr(node, i); |
| 1343 | blocksize = btrfs_level_size(root, level - 1); |
| 1344 | ret = lookup_extent_ref(NULL, root, bytenr, blocksize, &refs); |
| 1345 | BUG_ON(ret); |
| 1346 | if (refs != 1) |
| 1347 | continue; |
| 1348 | mutex_unlock(&root->fs_info->fs_mutex); |
| 1349 | ret = readahead_tree_block(root, bytenr, blocksize); |
| 1350 | cond_resched(); |
| 1351 | mutex_lock(&root->fs_info->fs_mutex); |
| 1352 | if (ret) |
| 1353 | break; |
| 1354 | } |
| 1355 | } |
| 1356 | |
| 1357 | /* |
| 1358 | * helper function for drop_snapshot, this walks down the tree dropping ref |
| 1359 | * counts as it goes. |
| 1360 | */ |
| 1361 | static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root |
| 1362 | *root, struct btrfs_path *path, int *level) |
| 1363 | { |
| 1364 | struct extent_buffer *next; |
| 1365 | struct extent_buffer *cur; |
| 1366 | u64 bytenr; |
| 1367 | u32 blocksize; |
| 1368 | int ret; |
| 1369 | u32 refs; |
| 1370 | |
| 1371 | WARN_ON(*level < 0); |
| 1372 | WARN_ON(*level >= BTRFS_MAX_LEVEL); |
| 1373 | ret = lookup_extent_ref(trans, root, |
| 1374 | path->nodes[*level]->start, |
| 1375 | path->nodes[*level]->len, &refs); |
| 1376 | BUG_ON(ret); |
| 1377 | if (refs > 1) |
| 1378 | goto out; |
| 1379 | |
| 1380 | /* |
| 1381 | * walk down to the last node level and free all the leaves |
| 1382 | */ |
| 1383 | while(*level >= 0) { |
| 1384 | WARN_ON(*level < 0); |
| 1385 | WARN_ON(*level >= BTRFS_MAX_LEVEL); |
| 1386 | cur = path->nodes[*level]; |
| 1387 | |
| 1388 | if (*level > 0 && path->slots[*level] == 0) |
| 1389 | reada_walk_down(root, cur); |
| 1390 | |
| 1391 | if (btrfs_header_level(cur) != *level) |
| 1392 | WARN_ON(1); |
| 1393 | |
| 1394 | if (path->slots[*level] >= |
| 1395 | btrfs_header_nritems(cur)) |
| 1396 | break; |
| 1397 | if (*level == 0) { |
| 1398 | ret = drop_leaf_ref(trans, root, cur); |
| 1399 | BUG_ON(ret); |
| 1400 | break; |
| 1401 | } |
| 1402 | bytenr = btrfs_node_blockptr(cur, path->slots[*level]); |
| 1403 | blocksize = btrfs_level_size(root, *level - 1); |
| 1404 | ret = lookup_extent_ref(trans, root, bytenr, blocksize, &refs); |
| 1405 | BUG_ON(ret); |
| 1406 | if (refs != 1) { |
| 1407 | path->slots[*level]++; |
| 1408 | ret = btrfs_free_extent(trans, root, bytenr, |
| 1409 | blocksize, 1); |
| 1410 | BUG_ON(ret); |
| 1411 | continue; |
| 1412 | } |
| 1413 | next = btrfs_find_tree_block(root, bytenr, blocksize); |
| 1414 | if (!next || !btrfs_buffer_uptodate(next)) { |
| 1415 | free_extent_buffer(next); |
| 1416 | mutex_unlock(&root->fs_info->fs_mutex); |
| 1417 | next = read_tree_block(root, bytenr, blocksize); |
| 1418 | mutex_lock(&root->fs_info->fs_mutex); |
| 1419 | |
| 1420 | /* we dropped the lock, check one more time */ |
| 1421 | ret = lookup_extent_ref(trans, root, bytenr, |
| 1422 | blocksize, &refs); |
| 1423 | BUG_ON(ret); |
| 1424 | if (refs != 1) { |
| 1425 | path->slots[*level]++; |
| 1426 | free_extent_buffer(next); |
| 1427 | ret = btrfs_free_extent(trans, root, |
| 1428 | bytenr, blocksize, 1); |
| 1429 | BUG_ON(ret); |
| 1430 | continue; |
| 1431 | } |
| 1432 | } |
| 1433 | WARN_ON(*level <= 0); |
| 1434 | if (path->nodes[*level-1]) |
| 1435 | free_extent_buffer(path->nodes[*level-1]); |
| 1436 | path->nodes[*level-1] = next; |
| 1437 | *level = btrfs_header_level(next); |
| 1438 | path->slots[*level] = 0; |
| 1439 | } |
| 1440 | out: |
| 1441 | WARN_ON(*level < 0); |
| 1442 | WARN_ON(*level >= BTRFS_MAX_LEVEL); |
| 1443 | ret = btrfs_free_extent(trans, root, path->nodes[*level]->start, |
| 1444 | path->nodes[*level]->len, 1); |
| 1445 | free_extent_buffer(path->nodes[*level]); |
| 1446 | path->nodes[*level] = NULL; |
| 1447 | *level += 1; |
| 1448 | BUG_ON(ret); |
| 1449 | return 0; |
| 1450 | } |
| 1451 | |
| 1452 | /* |
| 1453 | * helper for dropping snapshots. This walks back up the tree in the path |
| 1454 | * to find the first node higher up where we haven't yet gone through |
| 1455 | * all the slots |
| 1456 | */ |
| 1457 | static int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root |
| 1458 | *root, struct btrfs_path *path, int *level) |
| 1459 | { |
| 1460 | int i; |
| 1461 | int slot; |
| 1462 | int ret; |
| 1463 | struct btrfs_root_item *root_item = &root->root_item; |
| 1464 | |
| 1465 | for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
| 1466 | slot = path->slots[i]; |
| 1467 | if (slot < btrfs_header_nritems(path->nodes[i]) - 1) { |
| 1468 | struct extent_buffer *node; |
| 1469 | struct btrfs_disk_key disk_key; |
| 1470 | node = path->nodes[i]; |
| 1471 | path->slots[i]++; |
| 1472 | *level = i; |
| 1473 | WARN_ON(*level == 0); |
| 1474 | btrfs_node_key(node, &disk_key, path->slots[i]); |
| 1475 | memcpy(&root_item->drop_progress, |
| 1476 | &disk_key, sizeof(disk_key)); |
| 1477 | root_item->drop_level = i; |
| 1478 | return 0; |
| 1479 | } else { |
| 1480 | ret = btrfs_free_extent(trans, root, |
| 1481 | path->nodes[*level]->start, |
| 1482 | path->nodes[*level]->len, 1); |
| 1483 | BUG_ON(ret); |
| 1484 | free_extent_buffer(path->nodes[*level]); |
| 1485 | path->nodes[*level] = NULL; |
| 1486 | *level = i + 1; |
| 1487 | } |
| 1488 | } |
| 1489 | return 1; |
| 1490 | } |
| 1491 | |
| 1492 | /* |
| 1493 | * drop the reference count on the tree rooted at 'snap'. This traverses |
| 1494 | * the tree freeing any blocks that have a ref count of zero after being |
| 1495 | * decremented. |
| 1496 | */ |
| 1497 | int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root |
| 1498 | *root) |
| 1499 | { |
| 1500 | int ret = 0; |
| 1501 | int wret; |
| 1502 | int level; |
| 1503 | struct btrfs_path *path; |
| 1504 | int i; |
| 1505 | int orig_level; |
| 1506 | struct btrfs_root_item *root_item = &root->root_item; |
| 1507 | |
| 1508 | path = btrfs_alloc_path(); |
| 1509 | BUG_ON(!path); |
| 1510 | |
| 1511 | level = btrfs_header_level(root->node); |
| 1512 | orig_level = level; |
| 1513 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
| 1514 | path->nodes[level] = root->node; |
| 1515 | extent_buffer_get(root->node); |
| 1516 | path->slots[level] = 0; |
| 1517 | } else { |
| 1518 | struct btrfs_key key; |
| 1519 | struct btrfs_disk_key found_key; |
| 1520 | struct extent_buffer *node; |
| 1521 | |
| 1522 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
| 1523 | level = root_item->drop_level; |
| 1524 | path->lowest_level = level; |
| 1525 | wret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
| 1526 | if (wret < 0) { |
| 1527 | ret = wret; |
| 1528 | goto out; |
| 1529 | } |
| 1530 | node = path->nodes[level]; |
| 1531 | btrfs_node_key(node, &found_key, path->slots[level]); |
| 1532 | WARN_ON(memcmp(&found_key, &root_item->drop_progress, |
| 1533 | sizeof(found_key))); |
| 1534 | } |
| 1535 | while(1) { |
| 1536 | wret = walk_down_tree(trans, root, path, &level); |
| 1537 | if (wret > 0) |
| 1538 | break; |
| 1539 | if (wret < 0) |
| 1540 | ret = wret; |
| 1541 | |
| 1542 | wret = walk_up_tree(trans, root, path, &level); |
| 1543 | if (wret > 0) |
| 1544 | break; |
| 1545 | if (wret < 0) |
| 1546 | ret = wret; |
| 1547 | ret = -EAGAIN; |
| 1548 | break; |
| 1549 | } |
| 1550 | for (i = 0; i <= orig_level; i++) { |
| 1551 | if (path->nodes[i]) { |
| 1552 | free_extent_buffer(path->nodes[i]); |
| 1553 | path->nodes[i] = NULL; |
| 1554 | } |
| 1555 | } |
| 1556 | out: |
| 1557 | btrfs_free_path(path); |
| 1558 | return ret; |
| 1559 | } |
| 1560 | |
| 1561 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
| 1562 | { |
| 1563 | u64 start; |
| 1564 | u64 end; |
| 1565 | u64 ptr; |
| 1566 | int ret; |
| 1567 | while(1) { |
| 1568 | ret = find_first_extent_bit(&info->block_group_cache, 0, |
| 1569 | &start, &end, (unsigned int)-1); |
| 1570 | if (ret) |
| 1571 | break; |
| 1572 | ret = get_state_private(&info->block_group_cache, start, &ptr); |
| 1573 | if (!ret) |
| 1574 | kfree((void *)(unsigned long)ptr); |
| 1575 | clear_extent_bits(&info->block_group_cache, start, |
| 1576 | end, (unsigned int)-1, GFP_NOFS); |
| 1577 | } |
| 1578 | while(1) { |
| 1579 | ret = find_first_extent_bit(&info->free_space_cache, 0, |
| 1580 | &start, &end, EXTENT_DIRTY); |
| 1581 | if (ret) |
| 1582 | break; |
| 1583 | clear_extent_dirty(&info->free_space_cache, start, |
| 1584 | end, GFP_NOFS); |
| 1585 | } |
| 1586 | return 0; |
| 1587 | } |
| 1588 | |
| 1589 | int btrfs_read_block_groups(struct btrfs_root *root) |
| 1590 | { |
| 1591 | struct btrfs_path *path; |
| 1592 | int ret; |
| 1593 | int err = 0; |
| 1594 | int bit; |
| 1595 | struct btrfs_block_group_cache *cache; |
| 1596 | struct btrfs_fs_info *info = root->fs_info; |
| 1597 | struct extent_map_tree *block_group_cache; |
| 1598 | struct btrfs_key key; |
| 1599 | struct btrfs_key found_key; |
| 1600 | struct extent_buffer *leaf; |
| 1601 | |
| 1602 | block_group_cache = &info->block_group_cache; |
| 1603 | |
| 1604 | root = info->extent_root; |
| 1605 | key.objectid = 0; |
| 1606 | key.offset = BTRFS_BLOCK_GROUP_SIZE; |
| 1607 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
| 1608 | |
| 1609 | path = btrfs_alloc_path(); |
| 1610 | if (!path) |
| 1611 | return -ENOMEM; |
| 1612 | |
| 1613 | while(1) { |
| 1614 | ret = btrfs_search_slot(NULL, info->extent_root, |
| 1615 | &key, path, 0, 0); |
| 1616 | if (ret != 0) { |
| 1617 | err = ret; |
| 1618 | break; |
| 1619 | } |
| 1620 | leaf = path->nodes[0]; |
| 1621 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
| 1622 | cache = kmalloc(sizeof(*cache), GFP_NOFS); |
| 1623 | if (!cache) { |
| 1624 | err = -1; |
| 1625 | break; |
| 1626 | } |
| 1627 | |
| 1628 | read_extent_buffer(leaf, &cache->item, |
| 1629 | btrfs_item_ptr_offset(leaf, path->slots[0]), |
| 1630 | sizeof(cache->item)); |
| 1631 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
| 1632 | cache->cached = 0; |
| 1633 | cache->pinned = 0; |
| 1634 | key.objectid = found_key.objectid + found_key.offset; |
| 1635 | btrfs_release_path(root, path); |
| 1636 | |
| 1637 | if (cache->item.flags & BTRFS_BLOCK_GROUP_MIXED) { |
| 1638 | bit = BLOCK_GROUP_DATA | BLOCK_GROUP_METADATA; |
| 1639 | cache->data = BTRFS_BLOCK_GROUP_MIXED; |
| 1640 | } else if (cache->item.flags & BTRFS_BLOCK_GROUP_DATA) { |
| 1641 | bit = BLOCK_GROUP_DATA; |
| 1642 | cache->data = BTRFS_BLOCK_GROUP_DATA; |
| 1643 | } else { |
| 1644 | bit = BLOCK_GROUP_METADATA; |
| 1645 | cache->data = 0; |
| 1646 | } |
| 1647 | |
| 1648 | /* use EXTENT_LOCKED to prevent merging */ |
| 1649 | set_extent_bits(block_group_cache, found_key.objectid, |
| 1650 | found_key.objectid + found_key.offset - 1, |
| 1651 | bit | EXTENT_LOCKED, GFP_NOFS); |
| 1652 | set_state_private(block_group_cache, found_key.objectid, |
| 1653 | (unsigned long)cache); |
| 1654 | |
| 1655 | if (key.objectid >= |
| 1656 | btrfs_super_total_bytes(&info->super_copy)) |
| 1657 | break; |
| 1658 | } |
| 1659 | |
| 1660 | btrfs_free_path(path); |
| 1661 | return 0; |
| 1662 | } |