| 1 | #include <linux/module.h> |
| 2 | #include "ctree.h" |
| 3 | #include "disk-io.h" |
| 4 | #include "print-tree.h" |
| 5 | #include "transaction.h" |
| 6 | |
| 7 | static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root |
| 8 | *orig_root, u64 num_blocks, u64 search_start, u64 |
| 9 | search_end, struct btrfs_key *ins); |
| 10 | static int finish_current_insert(struct btrfs_trans_handle *trans, struct |
| 11 | btrfs_root *extent_root); |
| 12 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
| 13 | btrfs_root *extent_root); |
| 14 | |
| 15 | struct btrfs_block_group_cache *btrfs_find_block_group(struct btrfs_root *root, |
| 16 | struct btrfs_block_group_cache |
| 17 | *hint, int data) |
| 18 | { |
| 19 | struct btrfs_block_group_cache *cache[8]; |
| 20 | struct btrfs_block_group_cache *found_group = NULL; |
| 21 | struct btrfs_fs_info *info = root->fs_info; |
| 22 | u64 used; |
| 23 | u64 last = 0; |
| 24 | u64 hint_last; |
| 25 | int i; |
| 26 | int ret; |
| 27 | int full_search = 0; |
| 28 | if (hint) { |
| 29 | used = btrfs_block_group_used(&hint->item); |
| 30 | if (used < (hint->key.offset * 2) / 3) { |
| 31 | return hint; |
| 32 | } |
| 33 | radix_tree_tag_clear(&info->block_group_radix, |
| 34 | hint->key.objectid + hint->key.offset - 1, |
| 35 | BTRFS_BLOCK_GROUP_AVAIL); |
| 36 | last = hint->key.objectid + hint->key.offset; |
| 37 | hint_last = last; |
| 38 | } else { |
| 39 | hint_last = 0; |
| 40 | last = 0; |
| 41 | } |
| 42 | while(1) { |
| 43 | ret = radix_tree_gang_lookup_tag(&info->block_group_radix, |
| 44 | (void **)cache, |
| 45 | last, ARRAY_SIZE(cache), |
| 46 | BTRFS_BLOCK_GROUP_AVAIL); |
| 47 | if (!ret) |
| 48 | break; |
| 49 | for (i = 0; i < ret; i++) { |
| 50 | used = btrfs_block_group_used(&cache[i]->item); |
| 51 | if (used < (cache[i]->key.offset * 2) / 3) { |
| 52 | info->block_group_cache = cache[i]; |
| 53 | found_group = cache[i]; |
| 54 | goto found; |
| 55 | } |
| 56 | radix_tree_tag_clear(&info->block_group_radix, |
| 57 | cache[i]->key.objectid + |
| 58 | cache[i]->key.offset - 1, |
| 59 | BTRFS_BLOCK_GROUP_AVAIL); |
| 60 | last = cache[i]->key.objectid + |
| 61 | cache[i]->key.offset; |
| 62 | } |
| 63 | } |
| 64 | last = hint_last; |
| 65 | again: |
| 66 | while(1) { |
| 67 | ret = radix_tree_gang_lookup(&info->block_group_radix, |
| 68 | (void **)cache, |
| 69 | last, ARRAY_SIZE(cache)); |
| 70 | if (!ret) |
| 71 | break; |
| 72 | for (i = 0; i < ret; i++) { |
| 73 | used = btrfs_block_group_used(&cache[i]->item); |
| 74 | if (used < cache[i]->key.offset) { |
| 75 | info->block_group_cache = cache[i]; |
| 76 | found_group = cache[i]; |
| 77 | goto found; |
| 78 | } |
| 79 | radix_tree_tag_clear(&info->block_group_radix, |
| 80 | cache[i]->key.objectid + |
| 81 | cache[i]->key.offset - 1, |
| 82 | BTRFS_BLOCK_GROUP_AVAIL); |
| 83 | last = cache[i]->key.objectid + |
| 84 | cache[i]->key.offset; |
| 85 | } |
| 86 | } |
| 87 | info->block_group_cache = NULL; |
| 88 | if (!full_search) { |
| 89 | last = 0; |
| 90 | full_search = 1; |
| 91 | goto again; |
| 92 | } |
| 93 | found: |
| 94 | if (!found_group) { |
| 95 | ret = radix_tree_gang_lookup(&info->block_group_radix, |
| 96 | (void **)&found_group, 0, 1); |
| 97 | BUG_ON(ret != 1); |
| 98 | } |
| 99 | return found_group; |
| 100 | } |
| 101 | |
| 102 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
| 103 | struct btrfs_root *root, |
| 104 | u64 blocknr, u64 num_blocks) |
| 105 | { |
| 106 | struct btrfs_path *path; |
| 107 | int ret; |
| 108 | struct btrfs_key key; |
| 109 | struct btrfs_leaf *l; |
| 110 | struct btrfs_extent_item *item; |
| 111 | struct btrfs_key ins; |
| 112 | u32 refs; |
| 113 | |
| 114 | find_free_extent(trans, root->fs_info->extent_root, 0, 0, (u64)-1, |
| 115 | &ins); |
| 116 | path = btrfs_alloc_path(); |
| 117 | BUG_ON(!path); |
| 118 | btrfs_init_path(path); |
| 119 | key.objectid = blocknr; |
| 120 | key.flags = 0; |
| 121 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
| 122 | key.offset = num_blocks; |
| 123 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
| 124 | 0, 1); |
| 125 | if (ret != 0) { |
| 126 | printk("can't find block %Lu %Lu\n", blocknr, num_blocks); |
| 127 | BUG(); |
| 128 | } |
| 129 | BUG_ON(ret != 0); |
| 130 | l = btrfs_buffer_leaf(path->nodes[0]); |
| 131 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
| 132 | refs = btrfs_extent_refs(item); |
| 133 | btrfs_set_extent_refs(item, refs + 1); |
| 134 | btrfs_mark_buffer_dirty(path->nodes[0]); |
| 135 | |
| 136 | btrfs_release_path(root->fs_info->extent_root, path); |
| 137 | btrfs_free_path(path); |
| 138 | finish_current_insert(trans, root->fs_info->extent_root); |
| 139 | del_pending_extents(trans, root->fs_info->extent_root); |
| 140 | return 0; |
| 141 | } |
| 142 | |
| 143 | static int lookup_extent_ref(struct btrfs_trans_handle *trans, |
| 144 | struct btrfs_root *root, u64 blocknr, |
| 145 | u64 num_blocks, u32 *refs) |
| 146 | { |
| 147 | struct btrfs_path *path; |
| 148 | int ret; |
| 149 | struct btrfs_key key; |
| 150 | struct btrfs_leaf *l; |
| 151 | struct btrfs_extent_item *item; |
| 152 | |
| 153 | path = btrfs_alloc_path(); |
| 154 | btrfs_init_path(path); |
| 155 | key.objectid = blocknr; |
| 156 | key.offset = num_blocks; |
| 157 | key.flags = 0; |
| 158 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
| 159 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, |
| 160 | 0, 0); |
| 161 | if (ret != 0) |
| 162 | BUG(); |
| 163 | l = btrfs_buffer_leaf(path->nodes[0]); |
| 164 | item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); |
| 165 | *refs = btrfs_extent_refs(item); |
| 166 | btrfs_release_path(root->fs_info->extent_root, path); |
| 167 | btrfs_free_path(path); |
| 168 | return 0; |
| 169 | } |
| 170 | |
| 171 | int btrfs_inc_root_ref(struct btrfs_trans_handle *trans, |
| 172 | struct btrfs_root *root) |
| 173 | { |
| 174 | return btrfs_inc_extent_ref(trans, root, bh_blocknr(root->node), 1); |
| 175 | } |
| 176 | |
| 177 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
| 178 | struct buffer_head *buf) |
| 179 | { |
| 180 | u64 blocknr; |
| 181 | struct btrfs_node *buf_node; |
| 182 | struct btrfs_leaf *buf_leaf; |
| 183 | struct btrfs_disk_key *key; |
| 184 | struct btrfs_file_extent_item *fi; |
| 185 | int i; |
| 186 | int leaf; |
| 187 | int ret; |
| 188 | |
| 189 | if (!root->ref_cows) |
| 190 | return 0; |
| 191 | buf_node = btrfs_buffer_node(buf); |
| 192 | leaf = btrfs_is_leaf(buf_node); |
| 193 | buf_leaf = btrfs_buffer_leaf(buf); |
| 194 | for (i = 0; i < btrfs_header_nritems(&buf_node->header); i++) { |
| 195 | if (leaf) { |
| 196 | key = &buf_leaf->items[i].key; |
| 197 | if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY) |
| 198 | continue; |
| 199 | fi = btrfs_item_ptr(buf_leaf, i, |
| 200 | struct btrfs_file_extent_item); |
| 201 | if (btrfs_file_extent_type(fi) == |
| 202 | BTRFS_FILE_EXTENT_INLINE) |
| 203 | continue; |
| 204 | ret = btrfs_inc_extent_ref(trans, root, |
| 205 | btrfs_file_extent_disk_blocknr(fi), |
| 206 | btrfs_file_extent_disk_num_blocks(fi)); |
| 207 | BUG_ON(ret); |
| 208 | } else { |
| 209 | blocknr = btrfs_node_blockptr(buf_node, i); |
| 210 | ret = btrfs_inc_extent_ref(trans, root, blocknr, 1); |
| 211 | BUG_ON(ret); |
| 212 | } |
| 213 | } |
| 214 | return 0; |
| 215 | } |
| 216 | |
| 217 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
| 218 | struct btrfs_root *root, |
| 219 | struct btrfs_path *path, |
| 220 | struct btrfs_block_group_cache *cache) |
| 221 | { |
| 222 | int ret; |
| 223 | int pending_ret; |
| 224 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
| 225 | struct btrfs_block_group_item *bi; |
| 226 | struct btrfs_key ins; |
| 227 | |
| 228 | find_free_extent(trans, extent_root, 0, 0, (u64)-1, &ins); |
| 229 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
| 230 | BUG_ON(ret); |
| 231 | bi = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0], |
| 232 | struct btrfs_block_group_item); |
| 233 | memcpy(bi, &cache->item, sizeof(*bi)); |
| 234 | mark_buffer_dirty(path->nodes[0]); |
| 235 | btrfs_release_path(extent_root, path); |
| 236 | |
| 237 | finish_current_insert(trans, extent_root); |
| 238 | pending_ret = del_pending_extents(trans, extent_root); |
| 239 | if (ret) |
| 240 | return ret; |
| 241 | if (pending_ret) |
| 242 | return pending_ret; |
| 243 | return 0; |
| 244 | |
| 245 | } |
| 246 | |
| 247 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
| 248 | struct btrfs_root *root) |
| 249 | { |
| 250 | struct btrfs_block_group_cache *cache[8]; |
| 251 | int ret; |
| 252 | int err = 0; |
| 253 | int werr = 0; |
| 254 | struct radix_tree_root *radix = &root->fs_info->block_group_radix; |
| 255 | int i; |
| 256 | struct btrfs_path *path; |
| 257 | |
| 258 | path = btrfs_alloc_path(); |
| 259 | if (!path) |
| 260 | return -ENOMEM; |
| 261 | |
| 262 | while(1) { |
| 263 | ret = radix_tree_gang_lookup_tag(radix, (void **)cache, |
| 264 | 0, ARRAY_SIZE(cache), |
| 265 | BTRFS_BLOCK_GROUP_DIRTY); |
| 266 | if (!ret) |
| 267 | break; |
| 268 | for (i = 0; i < ret; i++) { |
| 269 | radix_tree_tag_clear(radix, cache[i]->key.objectid + |
| 270 | cache[i]->key.offset - 1, |
| 271 | BTRFS_BLOCK_GROUP_DIRTY); |
| 272 | err = write_one_cache_group(trans, root, |
| 273 | path, cache[i]); |
| 274 | if (err) |
| 275 | werr = err; |
| 276 | cache[i]->last_alloc = cache[i]->first_free; |
| 277 | } |
| 278 | } |
| 279 | btrfs_free_path(path); |
| 280 | return werr; |
| 281 | } |
| 282 | |
| 283 | static int update_block_group(struct btrfs_trans_handle *trans, |
| 284 | struct btrfs_root *root, |
| 285 | u64 blocknr, u64 num, int alloc) |
| 286 | { |
| 287 | struct btrfs_block_group_cache *cache; |
| 288 | struct btrfs_fs_info *info = root->fs_info; |
| 289 | u64 total = num; |
| 290 | u64 old_val; |
| 291 | u64 block_in_group; |
| 292 | int ret; |
| 293 | while(total) { |
| 294 | ret = radix_tree_gang_lookup(&info->block_group_radix, |
| 295 | (void **)&cache, blocknr, 1); |
| 296 | if (!ret) { |
| 297 | printk(KERN_CRIT "blocknr %Lu lookup failed\n", |
| 298 | blocknr); |
| 299 | return -1; |
| 300 | } |
| 301 | block_in_group = blocknr - cache->key.objectid; |
| 302 | WARN_ON(block_in_group > cache->key.offset); |
| 303 | radix_tree_tag_set(&info->block_group_radix, |
| 304 | cache->key.objectid + cache->key.offset - 1, |
| 305 | BTRFS_BLOCK_GROUP_DIRTY); |
| 306 | |
| 307 | old_val = btrfs_block_group_used(&cache->item); |
| 308 | num = min(total, cache->key.offset - block_in_group); |
| 309 | total -= num; |
| 310 | blocknr += num; |
| 311 | if (alloc) { |
| 312 | old_val += num; |
| 313 | if (blocknr > cache->last_alloc) |
| 314 | cache->last_alloc = blocknr; |
| 315 | } else { |
| 316 | old_val -= num; |
| 317 | if (blocknr < cache->first_free) |
| 318 | cache->first_free = blocknr; |
| 319 | } |
| 320 | btrfs_set_block_group_used(&cache->item, old_val); |
| 321 | } |
| 322 | return 0; |
| 323 | } |
| 324 | |
| 325 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, struct |
| 326 | btrfs_root *root) |
| 327 | { |
| 328 | unsigned long gang[8]; |
| 329 | u64 first = 0; |
| 330 | int ret; |
| 331 | int i; |
| 332 | struct radix_tree_root *pinned_radix = &root->fs_info->pinned_radix; |
| 333 | |
| 334 | while(1) { |
| 335 | ret = find_first_radix_bit(pinned_radix, gang, |
| 336 | ARRAY_SIZE(gang)); |
| 337 | if (!ret) |
| 338 | break; |
| 339 | if (!first) |
| 340 | first = gang[0]; |
| 341 | for (i = 0; i < ret; i++) { |
| 342 | clear_radix_bit(pinned_radix, gang[i]); |
| 343 | } |
| 344 | } |
| 345 | return 0; |
| 346 | } |
| 347 | |
| 348 | static int finish_current_insert(struct btrfs_trans_handle *trans, struct |
| 349 | btrfs_root *extent_root) |
| 350 | { |
| 351 | struct btrfs_key ins; |
| 352 | struct btrfs_extent_item extent_item; |
| 353 | int i; |
| 354 | int ret; |
| 355 | u64 super_blocks_used; |
| 356 | struct btrfs_fs_info *info = extent_root->fs_info; |
| 357 | |
| 358 | btrfs_set_extent_refs(&extent_item, 1); |
| 359 | ins.offset = 1; |
| 360 | ins.flags = 0; |
| 361 | btrfs_set_key_type(&ins, BTRFS_EXTENT_ITEM_KEY); |
| 362 | btrfs_set_extent_owner(&extent_item, extent_root->root_key.objectid); |
| 363 | |
| 364 | for (i = 0; i < extent_root->fs_info->extent_tree_insert_nr; i++) { |
| 365 | ins.objectid = extent_root->fs_info->extent_tree_insert[i]; |
| 366 | super_blocks_used = btrfs_super_blocks_used(info->disk_super); |
| 367 | btrfs_set_super_blocks_used(info->disk_super, |
| 368 | super_blocks_used + 1); |
| 369 | ret = btrfs_insert_item(trans, extent_root, &ins, &extent_item, |
| 370 | sizeof(extent_item)); |
| 371 | BUG_ON(ret); |
| 372 | } |
| 373 | extent_root->fs_info->extent_tree_insert_nr = 0; |
| 374 | extent_root->fs_info->extent_tree_prealloc_nr = 0; |
| 375 | return 0; |
| 376 | } |
| 377 | |
| 378 | static int pin_down_block(struct btrfs_root *root, u64 blocknr, int pending) |
| 379 | { |
| 380 | int err; |
| 381 | struct btrfs_header *header; |
| 382 | struct buffer_head *bh; |
| 383 | |
| 384 | if (!pending) { |
| 385 | bh = btrfs_find_tree_block(root, blocknr); |
| 386 | if (bh) { |
| 387 | if (buffer_uptodate(bh)) { |
| 388 | u64 transid = |
| 389 | root->fs_info->running_transaction->transid; |
| 390 | header = btrfs_buffer_header(bh); |
| 391 | if (btrfs_header_generation(header) == |
| 392 | transid) { |
| 393 | btrfs_block_release(root, bh); |
| 394 | return 0; |
| 395 | } |
| 396 | } |
| 397 | btrfs_block_release(root, bh); |
| 398 | } |
| 399 | err = set_radix_bit(&root->fs_info->pinned_radix, blocknr); |
| 400 | } else { |
| 401 | err = set_radix_bit(&root->fs_info->pending_del_radix, blocknr); |
| 402 | } |
| 403 | BUG_ON(err); |
| 404 | return 0; |
| 405 | } |
| 406 | |
| 407 | /* |
| 408 | * remove an extent from the root, returns 0 on success |
| 409 | */ |
| 410 | static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root |
| 411 | *root, u64 blocknr, u64 num_blocks, int pin) |
| 412 | { |
| 413 | struct btrfs_path *path; |
| 414 | struct btrfs_key key; |
| 415 | struct btrfs_fs_info *info = root->fs_info; |
| 416 | struct btrfs_root *extent_root = info->extent_root; |
| 417 | int ret; |
| 418 | struct btrfs_extent_item *ei; |
| 419 | struct btrfs_key ins; |
| 420 | u32 refs; |
| 421 | |
| 422 | key.objectid = blocknr; |
| 423 | key.flags = 0; |
| 424 | btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); |
| 425 | key.offset = num_blocks; |
| 426 | |
| 427 | find_free_extent(trans, root, 0, 0, (u64)-1, &ins); |
| 428 | path = btrfs_alloc_path(); |
| 429 | BUG_ON(!path); |
| 430 | btrfs_init_path(path); |
| 431 | |
| 432 | ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); |
| 433 | if (ret) { |
| 434 | printk("failed to find %Lu\n", key.objectid); |
| 435 | btrfs_print_tree(extent_root, extent_root->node); |
| 436 | printk("failed to find %Lu\n", key.objectid); |
| 437 | BUG(); |
| 438 | } |
| 439 | ei = btrfs_item_ptr(btrfs_buffer_leaf(path->nodes[0]), path->slots[0], |
| 440 | struct btrfs_extent_item); |
| 441 | BUG_ON(ei->refs == 0); |
| 442 | refs = btrfs_extent_refs(ei) - 1; |
| 443 | btrfs_set_extent_refs(ei, refs); |
| 444 | btrfs_mark_buffer_dirty(path->nodes[0]); |
| 445 | if (refs == 0) { |
| 446 | u64 super_blocks_used; |
| 447 | |
| 448 | if (pin) { |
| 449 | ret = pin_down_block(root, blocknr, 0); |
| 450 | BUG_ON(ret); |
| 451 | } |
| 452 | |
| 453 | super_blocks_used = btrfs_super_blocks_used(info->disk_super); |
| 454 | btrfs_set_super_blocks_used(info->disk_super, |
| 455 | super_blocks_used - num_blocks); |
| 456 | ret = btrfs_del_item(trans, extent_root, path); |
| 457 | if (ret) |
| 458 | BUG(); |
| 459 | ret = update_block_group(trans, root, blocknr, num_blocks, 0); |
| 460 | BUG_ON(ret); |
| 461 | } |
| 462 | btrfs_release_path(extent_root, path); |
| 463 | btrfs_free_path(path); |
| 464 | finish_current_insert(trans, extent_root); |
| 465 | return ret; |
| 466 | } |
| 467 | |
| 468 | /* |
| 469 | * find all the blocks marked as pending in the radix tree and remove |
| 470 | * them from the extent map |
| 471 | */ |
| 472 | static int del_pending_extents(struct btrfs_trans_handle *trans, struct |
| 473 | btrfs_root *extent_root) |
| 474 | { |
| 475 | int ret; |
| 476 | int wret; |
| 477 | int err = 0; |
| 478 | unsigned long gang[4]; |
| 479 | int i; |
| 480 | struct radix_tree_root *pending_radix; |
| 481 | struct radix_tree_root *pinned_radix; |
| 482 | |
| 483 | pending_radix = &extent_root->fs_info->pending_del_radix; |
| 484 | pinned_radix = &extent_root->fs_info->pinned_radix; |
| 485 | |
| 486 | while(1) { |
| 487 | ret = find_first_radix_bit(pending_radix, gang, |
| 488 | ARRAY_SIZE(gang)); |
| 489 | if (!ret) |
| 490 | break; |
| 491 | for (i = 0; i < ret; i++) { |
| 492 | wret = set_radix_bit(pinned_radix, gang[i]); |
| 493 | BUG_ON(wret); |
| 494 | wret = clear_radix_bit(pending_radix, gang[i]); |
| 495 | BUG_ON(wret); |
| 496 | wret = __free_extent(trans, extent_root, |
| 497 | gang[i], 1, 0); |
| 498 | if (wret) |
| 499 | err = wret; |
| 500 | } |
| 501 | } |
| 502 | return err; |
| 503 | } |
| 504 | |
| 505 | /* |
| 506 | * remove an extent from the root, returns 0 on success |
| 507 | */ |
| 508 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root |
| 509 | *root, u64 blocknr, u64 num_blocks, int pin) |
| 510 | { |
| 511 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
| 512 | int pending_ret; |
| 513 | int ret; |
| 514 | |
| 515 | if (root == extent_root) { |
| 516 | pin_down_block(root, blocknr, 1); |
| 517 | return 0; |
| 518 | } |
| 519 | ret = __free_extent(trans, root, blocknr, num_blocks, pin); |
| 520 | pending_ret = del_pending_extents(trans, root->fs_info->extent_root); |
| 521 | return ret ? ret : pending_ret; |
| 522 | } |
| 523 | |
| 524 | /* |
| 525 | * walks the btree of allocated extents and find a hole of a given size. |
| 526 | * The key ins is changed to record the hole: |
| 527 | * ins->objectid == block start |
| 528 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
| 529 | * ins->offset == number of blocks |
| 530 | * Any available blocks before search_start are skipped. |
| 531 | */ |
| 532 | static int find_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root |
| 533 | *orig_root, u64 num_blocks, u64 search_start, u64 |
| 534 | search_end, struct btrfs_key *ins) |
| 535 | { |
| 536 | struct btrfs_path *path; |
| 537 | struct btrfs_key key; |
| 538 | int ret; |
| 539 | u64 hole_size = 0; |
| 540 | int slot = 0; |
| 541 | u64 last_block = 0; |
| 542 | u64 test_block; |
| 543 | int start_found; |
| 544 | struct btrfs_leaf *l; |
| 545 | struct btrfs_root * root = orig_root->fs_info->extent_root; |
| 546 | struct btrfs_fs_info *info = root->fs_info; |
| 547 | int total_needed = num_blocks; |
| 548 | int total_found = 0; |
| 549 | int fill_prealloc = 0; |
| 550 | int level; |
| 551 | int update_block_group = 0; |
| 552 | struct btrfs_block_group_cache *hint_block_group; |
| 553 | |
| 554 | path = btrfs_alloc_path(); |
| 555 | ins->flags = 0; |
| 556 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
| 557 | |
| 558 | level = btrfs_header_level(btrfs_buffer_header(root->node)); |
| 559 | /* find search start here */ |
| 560 | if (0 && search_start && num_blocks) { |
| 561 | u64 used; |
| 562 | ret = radix_tree_gang_lookup(&info->block_group_radix, |
| 563 | (void **)&hint_block_group, |
| 564 | search_start, 1); |
| 565 | if (ret) { |
| 566 | used = btrfs_block_group_used(&hint_block_group->item); |
| 567 | if (used > (hint_block_group->key.offset * 9) / 10) |
| 568 | search_start = 0; |
| 569 | else if (search_start < hint_block_group->last_alloc) |
| 570 | search_start = hint_block_group->last_alloc; |
| 571 | } else { |
| 572 | search_start = 0; |
| 573 | } |
| 574 | } |
| 575 | if (num_blocks == 0) { |
| 576 | fill_prealloc = 1; |
| 577 | num_blocks = 1; |
| 578 | total_needed = (min(level + 1, BTRFS_MAX_LEVEL) + 2) * 3; |
| 579 | } |
| 580 | if (1 || !search_start) { |
| 581 | trans->block_group = btrfs_find_block_group(root, |
| 582 | trans->block_group, |
| 583 | 0); |
| 584 | if (trans->block_group->last_alloc > search_start) |
| 585 | search_start = trans->block_group->last_alloc; |
| 586 | update_block_group = 1; |
| 587 | } |
| 588 | check_failed: |
| 589 | btrfs_init_path(path); |
| 590 | ins->objectid = search_start; |
| 591 | ins->offset = 0; |
| 592 | start_found = 0; |
| 593 | ret = btrfs_search_slot(trans, root, ins, path, 0, 0); |
| 594 | if (ret < 0) |
| 595 | goto error; |
| 596 | |
| 597 | if (path->slots[0] > 0) |
| 598 | path->slots[0]--; |
| 599 | |
| 600 | while (1) { |
| 601 | l = btrfs_buffer_leaf(path->nodes[0]); |
| 602 | slot = path->slots[0]; |
| 603 | if (slot >= btrfs_header_nritems(&l->header)) { |
| 604 | if (fill_prealloc) { |
| 605 | info->extent_tree_prealloc_nr = 0; |
| 606 | total_found = 0; |
| 607 | } |
| 608 | ret = btrfs_next_leaf(root, path); |
| 609 | if (ret == 0) |
| 610 | continue; |
| 611 | if (ret < 0) |
| 612 | goto error; |
| 613 | if (!start_found) { |
| 614 | ins->objectid = search_start; |
| 615 | ins->offset = (u64)-1 - search_start; |
| 616 | start_found = 1; |
| 617 | goto check_pending; |
| 618 | } |
| 619 | ins->objectid = last_block > search_start ? |
| 620 | last_block : search_start; |
| 621 | ins->offset = (u64)-1 - ins->objectid; |
| 622 | goto check_pending; |
| 623 | } |
| 624 | btrfs_disk_key_to_cpu(&key, &l->items[slot].key); |
| 625 | if (btrfs_key_type(&key) != BTRFS_EXTENT_ITEM_KEY) |
| 626 | goto next; |
| 627 | if (key.objectid >= search_start) { |
| 628 | if (start_found) { |
| 629 | if (last_block < search_start) |
| 630 | last_block = search_start; |
| 631 | hole_size = key.objectid - last_block; |
| 632 | if (hole_size >= num_blocks) { |
| 633 | ins->objectid = last_block; |
| 634 | ins->offset = hole_size; |
| 635 | goto check_pending; |
| 636 | } |
| 637 | } |
| 638 | } |
| 639 | start_found = 1; |
| 640 | last_block = key.objectid + key.offset; |
| 641 | next: |
| 642 | path->slots[0]++; |
| 643 | } |
| 644 | // FIXME -ENOSPC |
| 645 | check_pending: |
| 646 | /* we have to make sure we didn't find an extent that has already |
| 647 | * been allocated by the map tree or the original allocation |
| 648 | */ |
| 649 | btrfs_release_path(root, path); |
| 650 | BUG_ON(ins->objectid < search_start); |
| 651 | if (ins->objectid >= btrfs_super_total_blocks(info->disk_super)) { |
| 652 | if (search_start == 0) |
| 653 | return -ENOSPC; |
| 654 | search_start = 0; |
| 655 | goto check_failed; |
| 656 | } |
| 657 | for (test_block = ins->objectid; |
| 658 | test_block < ins->objectid + num_blocks; test_block++) { |
| 659 | if (test_radix_bit(&info->pinned_radix, test_block)) { |
| 660 | search_start = test_block + 1; |
| 661 | goto check_failed; |
| 662 | } |
| 663 | } |
| 664 | if (!fill_prealloc && info->extent_tree_insert_nr) { |
| 665 | u64 last = |
| 666 | info->extent_tree_insert[info->extent_tree_insert_nr - 1]; |
| 667 | if (ins->objectid + num_blocks > |
| 668 | info->extent_tree_insert[0] && |
| 669 | ins->objectid <= last) { |
| 670 | search_start = last + 1; |
| 671 | WARN_ON(1); |
| 672 | goto check_failed; |
| 673 | } |
| 674 | } |
| 675 | if (!fill_prealloc && info->extent_tree_prealloc_nr) { |
| 676 | u64 first = |
| 677 | info->extent_tree_prealloc[info->extent_tree_prealloc_nr - 1]; |
| 678 | if (ins->objectid + num_blocks > first && |
| 679 | ins->objectid <= info->extent_tree_prealloc[0]) { |
| 680 | search_start = info->extent_tree_prealloc[0] + 1; |
| 681 | WARN_ON(1); |
| 682 | goto check_failed; |
| 683 | } |
| 684 | } |
| 685 | if (fill_prealloc) { |
| 686 | int nr; |
| 687 | test_block = ins->objectid; |
| 688 | while(test_block < ins->objectid + ins->offset && |
| 689 | total_found < total_needed) { |
| 690 | nr = total_needed - total_found - 1; |
| 691 | BUG_ON(nr < 0); |
| 692 | info->extent_tree_prealloc[nr] = test_block; |
| 693 | total_found++; |
| 694 | test_block++; |
| 695 | } |
| 696 | if (total_found < total_needed) { |
| 697 | search_start = test_block; |
| 698 | goto check_failed; |
| 699 | } |
| 700 | info->extent_tree_prealloc_nr = total_found; |
| 701 | } |
| 702 | if (update_block_group) { |
| 703 | ret = radix_tree_gang_lookup(&info->block_group_radix, |
| 704 | (void **)&trans->block_group, |
| 705 | ins->objectid, 1); |
| 706 | if (ret) { |
| 707 | trans->block_group->last_alloc = ins->objectid; |
| 708 | } |
| 709 | } |
| 710 | ins->offset = num_blocks; |
| 711 | btrfs_free_path(path); |
| 712 | return 0; |
| 713 | error: |
| 714 | btrfs_release_path(root, path); |
| 715 | btrfs_free_path(path); |
| 716 | return ret; |
| 717 | } |
| 718 | /* |
| 719 | * finds a free extent and does all the dirty work required for allocation |
| 720 | * returns the key for the extent through ins, and a tree buffer for |
| 721 | * the first block of the extent through buf. |
| 722 | * |
| 723 | * returns 0 if everything worked, non-zero otherwise. |
| 724 | */ |
| 725 | int btrfs_alloc_extent(struct btrfs_trans_handle *trans, |
| 726 | struct btrfs_root *root, u64 owner, |
| 727 | u64 num_blocks, u64 search_start, |
| 728 | u64 search_end, struct btrfs_key *ins) |
| 729 | { |
| 730 | int ret; |
| 731 | int pending_ret; |
| 732 | u64 super_blocks_used; |
| 733 | struct btrfs_fs_info *info = root->fs_info; |
| 734 | struct btrfs_root *extent_root = info->extent_root; |
| 735 | struct btrfs_extent_item extent_item; |
| 736 | struct btrfs_key prealloc_key; |
| 737 | |
| 738 | btrfs_set_extent_refs(&extent_item, 1); |
| 739 | btrfs_set_extent_owner(&extent_item, owner); |
| 740 | |
| 741 | if (root == extent_root) { |
| 742 | int nr; |
| 743 | BUG_ON(info->extent_tree_prealloc_nr == 0); |
| 744 | BUG_ON(num_blocks != 1); |
| 745 | ins->offset = 1; |
| 746 | info->extent_tree_prealloc_nr--; |
| 747 | nr = info->extent_tree_prealloc_nr; |
| 748 | ins->objectid = info->extent_tree_prealloc[nr]; |
| 749 | info->extent_tree_insert[info->extent_tree_insert_nr++] = |
| 750 | ins->objectid; |
| 751 | ret = update_block_group(trans, root, |
| 752 | ins->objectid, ins->offset, 1); |
| 753 | BUG_ON(ret); |
| 754 | return 0; |
| 755 | } |
| 756 | /* do the real allocation */ |
| 757 | ret = find_free_extent(trans, root, num_blocks, search_start, |
| 758 | search_end, ins); |
| 759 | if (ret) |
| 760 | return ret; |
| 761 | |
| 762 | /* then do prealloc for the extent tree */ |
| 763 | ret = find_free_extent(trans, root, 0, ins->objectid + ins->offset, |
| 764 | search_end, &prealloc_key); |
| 765 | if (ret) |
| 766 | return ret; |
| 767 | |
| 768 | super_blocks_used = btrfs_super_blocks_used(info->disk_super); |
| 769 | btrfs_set_super_blocks_used(info->disk_super, super_blocks_used + |
| 770 | num_blocks); |
| 771 | ret = btrfs_insert_item(trans, extent_root, ins, &extent_item, |
| 772 | sizeof(extent_item)); |
| 773 | |
| 774 | finish_current_insert(trans, extent_root); |
| 775 | pending_ret = del_pending_extents(trans, extent_root); |
| 776 | if (ret) |
| 777 | return ret; |
| 778 | if (pending_ret) |
| 779 | return pending_ret; |
| 780 | ret = update_block_group(trans, root, ins->objectid, ins->offset, 1); |
| 781 | return 0; |
| 782 | } |
| 783 | |
| 784 | /* |
| 785 | * helper function to allocate a block for a given tree |
| 786 | * returns the tree buffer or NULL. |
| 787 | */ |
| 788 | struct buffer_head *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
| 789 | struct btrfs_root *root, u64 hint) |
| 790 | { |
| 791 | struct btrfs_key ins; |
| 792 | int ret; |
| 793 | struct buffer_head *buf; |
| 794 | |
| 795 | ret = btrfs_alloc_extent(trans, root, root->root_key.objectid, |
| 796 | 1, hint, (unsigned long)-1, &ins); |
| 797 | if (ret) { |
| 798 | BUG(); |
| 799 | return NULL; |
| 800 | } |
| 801 | BUG_ON(ret); |
| 802 | buf = btrfs_find_create_tree_block(root, ins.objectid); |
| 803 | set_buffer_uptodate(buf); |
| 804 | set_buffer_checked(buf); |
| 805 | set_radix_bit(&trans->transaction->dirty_pages, buf->b_page->index); |
| 806 | return buf; |
| 807 | } |
| 808 | |
| 809 | static int drop_leaf_ref(struct btrfs_trans_handle *trans, |
| 810 | struct btrfs_root *root, struct buffer_head *cur) |
| 811 | { |
| 812 | struct btrfs_disk_key *key; |
| 813 | struct btrfs_leaf *leaf; |
| 814 | struct btrfs_file_extent_item *fi; |
| 815 | int i; |
| 816 | int nritems; |
| 817 | int ret; |
| 818 | |
| 819 | BUG_ON(!btrfs_is_leaf(btrfs_buffer_node(cur))); |
| 820 | leaf = btrfs_buffer_leaf(cur); |
| 821 | nritems = btrfs_header_nritems(&leaf->header); |
| 822 | for (i = 0; i < nritems; i++) { |
| 823 | key = &leaf->items[i].key; |
| 824 | if (btrfs_disk_key_type(key) != BTRFS_EXTENT_DATA_KEY) |
| 825 | continue; |
| 826 | fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); |
| 827 | if (btrfs_file_extent_type(fi) == BTRFS_FILE_EXTENT_INLINE) |
| 828 | continue; |
| 829 | /* |
| 830 | * FIXME make sure to insert a trans record that |
| 831 | * repeats the snapshot del on crash |
| 832 | */ |
| 833 | ret = btrfs_free_extent(trans, root, |
| 834 | btrfs_file_extent_disk_blocknr(fi), |
| 835 | btrfs_file_extent_disk_num_blocks(fi), |
| 836 | 0); |
| 837 | BUG_ON(ret); |
| 838 | } |
| 839 | return 0; |
| 840 | } |
| 841 | |
| 842 | /* |
| 843 | * helper function for drop_snapshot, this walks down the tree dropping ref |
| 844 | * counts as it goes. |
| 845 | */ |
| 846 | static int walk_down_tree(struct btrfs_trans_handle *trans, struct btrfs_root |
| 847 | *root, struct btrfs_path *path, int *level) |
| 848 | { |
| 849 | struct buffer_head *next; |
| 850 | struct buffer_head *cur; |
| 851 | u64 blocknr; |
| 852 | int ret; |
| 853 | u32 refs; |
| 854 | |
| 855 | WARN_ON(*level < 0); |
| 856 | WARN_ON(*level >= BTRFS_MAX_LEVEL); |
| 857 | ret = lookup_extent_ref(trans, root, bh_blocknr(path->nodes[*level]), |
| 858 | 1, &refs); |
| 859 | BUG_ON(ret); |
| 860 | if (refs > 1) |
| 861 | goto out; |
| 862 | /* |
| 863 | * walk down to the last node level and free all the leaves |
| 864 | */ |
| 865 | while(*level >= 0) { |
| 866 | WARN_ON(*level < 0); |
| 867 | WARN_ON(*level >= BTRFS_MAX_LEVEL); |
| 868 | cur = path->nodes[*level]; |
| 869 | if (btrfs_header_level(btrfs_buffer_header(cur)) != *level) |
| 870 | WARN_ON(1); |
| 871 | if (path->slots[*level] >= |
| 872 | btrfs_header_nritems(btrfs_buffer_header(cur))) |
| 873 | break; |
| 874 | if (*level == 0) { |
| 875 | ret = drop_leaf_ref(trans, root, cur); |
| 876 | BUG_ON(ret); |
| 877 | break; |
| 878 | } |
| 879 | blocknr = btrfs_node_blockptr(btrfs_buffer_node(cur), |
| 880 | path->slots[*level]); |
| 881 | ret = lookup_extent_ref(trans, root, blocknr, 1, &refs); |
| 882 | BUG_ON(ret); |
| 883 | if (refs != 1) { |
| 884 | path->slots[*level]++; |
| 885 | ret = btrfs_free_extent(trans, root, blocknr, 1, 1); |
| 886 | BUG_ON(ret); |
| 887 | continue; |
| 888 | } |
| 889 | next = read_tree_block(root, blocknr); |
| 890 | WARN_ON(*level <= 0); |
| 891 | if (path->nodes[*level-1]) |
| 892 | btrfs_block_release(root, path->nodes[*level-1]); |
| 893 | path->nodes[*level-1] = next; |
| 894 | *level = btrfs_header_level(btrfs_buffer_header(next)); |
| 895 | path->slots[*level] = 0; |
| 896 | } |
| 897 | out: |
| 898 | WARN_ON(*level < 0); |
| 899 | WARN_ON(*level >= BTRFS_MAX_LEVEL); |
| 900 | ret = btrfs_free_extent(trans, root, |
| 901 | bh_blocknr(path->nodes[*level]), 1, 1); |
| 902 | btrfs_block_release(root, path->nodes[*level]); |
| 903 | path->nodes[*level] = NULL; |
| 904 | *level += 1; |
| 905 | BUG_ON(ret); |
| 906 | return 0; |
| 907 | } |
| 908 | |
| 909 | /* |
| 910 | * helper for dropping snapshots. This walks back up the tree in the path |
| 911 | * to find the first node higher up where we haven't yet gone through |
| 912 | * all the slots |
| 913 | */ |
| 914 | static int walk_up_tree(struct btrfs_trans_handle *trans, struct btrfs_root |
| 915 | *root, struct btrfs_path *path, int *level) |
| 916 | { |
| 917 | int i; |
| 918 | int slot; |
| 919 | int ret; |
| 920 | for(i = *level; i < BTRFS_MAX_LEVEL - 1 && path->nodes[i]; i++) { |
| 921 | slot = path->slots[i]; |
| 922 | if (slot < btrfs_header_nritems( |
| 923 | btrfs_buffer_header(path->nodes[i])) - 1) { |
| 924 | path->slots[i]++; |
| 925 | *level = i; |
| 926 | return 0; |
| 927 | } else { |
| 928 | ret = btrfs_free_extent(trans, root, |
| 929 | bh_blocknr(path->nodes[*level]), |
| 930 | 1, 1); |
| 931 | BUG_ON(ret); |
| 932 | btrfs_block_release(root, path->nodes[*level]); |
| 933 | path->nodes[*level] = NULL; |
| 934 | *level = i + 1; |
| 935 | } |
| 936 | } |
| 937 | return 1; |
| 938 | } |
| 939 | |
| 940 | /* |
| 941 | * drop the reference count on the tree rooted at 'snap'. This traverses |
| 942 | * the tree freeing any blocks that have a ref count of zero after being |
| 943 | * decremented. |
| 944 | */ |
| 945 | int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root |
| 946 | *root, struct buffer_head *snap) |
| 947 | { |
| 948 | int ret = 0; |
| 949 | int wret; |
| 950 | int level; |
| 951 | struct btrfs_path *path; |
| 952 | int i; |
| 953 | int orig_level; |
| 954 | |
| 955 | path = btrfs_alloc_path(); |
| 956 | BUG_ON(!path); |
| 957 | btrfs_init_path(path); |
| 958 | |
| 959 | level = btrfs_header_level(btrfs_buffer_header(snap)); |
| 960 | orig_level = level; |
| 961 | path->nodes[level] = snap; |
| 962 | path->slots[level] = 0; |
| 963 | while(1) { |
| 964 | wret = walk_down_tree(trans, root, path, &level); |
| 965 | if (wret > 0) |
| 966 | break; |
| 967 | if (wret < 0) |
| 968 | ret = wret; |
| 969 | |
| 970 | wret = walk_up_tree(trans, root, path, &level); |
| 971 | if (wret > 0) |
| 972 | break; |
| 973 | if (wret < 0) |
| 974 | ret = wret; |
| 975 | btrfs_btree_balance_dirty(root); |
| 976 | } |
| 977 | for (i = 0; i <= orig_level; i++) { |
| 978 | if (path->nodes[i]) { |
| 979 | btrfs_block_release(root, path->nodes[i]); |
| 980 | } |
| 981 | } |
| 982 | btrfs_free_path(path); |
| 983 | return ret; |
| 984 | } |
| 985 | |
| 986 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
| 987 | { |
| 988 | int ret; |
| 989 | struct btrfs_block_group_cache *cache[8]; |
| 990 | int i; |
| 991 | |
| 992 | while(1) { |
| 993 | ret = radix_tree_gang_lookup(&info->block_group_radix, |
| 994 | (void **)cache, 0, |
| 995 | ARRAY_SIZE(cache)); |
| 996 | if (!ret) |
| 997 | break; |
| 998 | for (i = 0; i < ret; i++) { |
| 999 | radix_tree_delete(&info->block_group_radix, |
| 1000 | cache[i]->key.objectid + |
| 1001 | cache[i]->key.offset - 1); |
| 1002 | kfree(cache[i]); |
| 1003 | } |
| 1004 | } |
| 1005 | return 0; |
| 1006 | } |
| 1007 | |
| 1008 | int btrfs_read_block_groups(struct btrfs_root *root) |
| 1009 | { |
| 1010 | struct btrfs_path *path; |
| 1011 | int ret; |
| 1012 | int err = 0; |
| 1013 | struct btrfs_block_group_item *bi; |
| 1014 | struct btrfs_block_group_cache *cache; |
| 1015 | struct btrfs_key key; |
| 1016 | struct btrfs_key found_key; |
| 1017 | struct btrfs_leaf *leaf; |
| 1018 | u64 group_size_blocks = BTRFS_BLOCK_GROUP_SIZE / root->blocksize; |
| 1019 | u64 used; |
| 1020 | |
| 1021 | root = root->fs_info->extent_root; |
| 1022 | key.objectid = 0; |
| 1023 | key.offset = group_size_blocks; |
| 1024 | key.flags = 0; |
| 1025 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
| 1026 | |
| 1027 | path = btrfs_alloc_path(); |
| 1028 | if (!path) |
| 1029 | return -ENOMEM; |
| 1030 | |
| 1031 | while(1) { |
| 1032 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, |
| 1033 | &key, path, 0, 0); |
| 1034 | if (ret != 0) { |
| 1035 | err = ret; |
| 1036 | break; |
| 1037 | } |
| 1038 | leaf = btrfs_buffer_leaf(path->nodes[0]); |
| 1039 | btrfs_disk_key_to_cpu(&found_key, |
| 1040 | &leaf->items[path->slots[0]].key); |
| 1041 | cache = kmalloc(sizeof(*cache), GFP_NOFS); |
| 1042 | if (!cache) { |
| 1043 | err = -1; |
| 1044 | break; |
| 1045 | } |
| 1046 | bi = btrfs_item_ptr(leaf, path->slots[0], |
| 1047 | struct btrfs_block_group_item); |
| 1048 | memcpy(&cache->item, bi, sizeof(*bi)); |
| 1049 | memcpy(&cache->key, &found_key, sizeof(found_key)); |
| 1050 | cache->last_alloc = cache->key.objectid; |
| 1051 | cache->first_free = cache->key.objectid; |
| 1052 | key.objectid = found_key.objectid + found_key.offset; |
| 1053 | btrfs_release_path(root, path); |
| 1054 | ret = radix_tree_insert(&root->fs_info->block_group_radix, |
| 1055 | found_key.objectid + |
| 1056 | found_key.offset - 1, |
| 1057 | (void *)cache); |
| 1058 | BUG_ON(ret); |
| 1059 | used = btrfs_block_group_used(bi); |
| 1060 | if (used < (key.offset * 2) / 3) { |
| 1061 | radix_tree_tag_set(&root->fs_info->block_group_radix, |
| 1062 | found_key.objectid + |
| 1063 | found_key.offset - 1, |
| 1064 | BTRFS_BLOCK_GROUP_AVAIL); |
| 1065 | } |
| 1066 | if (key.objectid >= |
| 1067 | btrfs_super_total_blocks(root->fs_info->disk_super)) |
| 1068 | break; |
| 1069 | } |
| 1070 | |
| 1071 | btrfs_free_path(path); |
| 1072 | return 0; |
| 1073 | } |