| 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 | #include <linux/sched.h> |
| 19 | #include <linux/pagemap.h> |
| 20 | #include <linux/writeback.h> |
| 21 | #include <linux/blkdev.h> |
| 22 | #include <linux/sort.h> |
| 23 | #include <linux/rcupdate.h> |
| 24 | #include <linux/kthread.h> |
| 25 | #include <linux/slab.h> |
| 26 | #include <linux/ratelimit.h> |
| 27 | #include <linux/percpu_counter.h> |
| 28 | #include "hash.h" |
| 29 | #include "tree-log.h" |
| 30 | #include "disk-io.h" |
| 31 | #include "print-tree.h" |
| 32 | #include "volumes.h" |
| 33 | #include "raid56.h" |
| 34 | #include "locking.h" |
| 35 | #include "free-space-cache.h" |
| 36 | #include "math.h" |
| 37 | #include "sysfs.h" |
| 38 | #include "qgroup.h" |
| 39 | |
| 40 | #undef SCRAMBLE_DELAYED_REFS |
| 41 | |
| 42 | /* |
| 43 | * control flags for do_chunk_alloc's force field |
| 44 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
| 45 | * if we really need one. |
| 46 | * |
| 47 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
| 48 | * if we have very few chunks already allocated. This is |
| 49 | * used as part of the clustering code to help make sure |
| 50 | * we have a good pool of storage to cluster in, without |
| 51 | * filling the FS with empty chunks |
| 52 | * |
| 53 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
| 54 | * |
| 55 | */ |
| 56 | enum { |
| 57 | CHUNK_ALLOC_NO_FORCE = 0, |
| 58 | CHUNK_ALLOC_LIMITED = 1, |
| 59 | CHUNK_ALLOC_FORCE = 2, |
| 60 | }; |
| 61 | |
| 62 | /* |
| 63 | * Control how reservations are dealt with. |
| 64 | * |
| 65 | * RESERVE_FREE - freeing a reservation. |
| 66 | * RESERVE_ALLOC - allocating space and we need to update bytes_may_use for |
| 67 | * ENOSPC accounting |
| 68 | * RESERVE_ALLOC_NO_ACCOUNT - allocating space and we should not update |
| 69 | * bytes_may_use as the ENOSPC accounting is done elsewhere |
| 70 | */ |
| 71 | enum { |
| 72 | RESERVE_FREE = 0, |
| 73 | RESERVE_ALLOC = 1, |
| 74 | RESERVE_ALLOC_NO_ACCOUNT = 2, |
| 75 | }; |
| 76 | |
| 77 | static int update_block_group(struct btrfs_root *root, |
| 78 | u64 bytenr, u64 num_bytes, int alloc); |
| 79 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
| 80 | struct btrfs_root *root, |
| 81 | u64 bytenr, u64 num_bytes, u64 parent, |
| 82 | u64 root_objectid, u64 owner_objectid, |
| 83 | u64 owner_offset, int refs_to_drop, |
| 84 | struct btrfs_delayed_extent_op *extra_op, |
| 85 | int no_quota); |
| 86 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
| 87 | struct extent_buffer *leaf, |
| 88 | struct btrfs_extent_item *ei); |
| 89 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
| 90 | struct btrfs_root *root, |
| 91 | u64 parent, u64 root_objectid, |
| 92 | u64 flags, u64 owner, u64 offset, |
| 93 | struct btrfs_key *ins, int ref_mod); |
| 94 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
| 95 | struct btrfs_root *root, |
| 96 | u64 parent, u64 root_objectid, |
| 97 | u64 flags, struct btrfs_disk_key *key, |
| 98 | int level, struct btrfs_key *ins, |
| 99 | int no_quota); |
| 100 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
| 101 | struct btrfs_root *extent_root, u64 flags, |
| 102 | int force); |
| 103 | static int find_next_key(struct btrfs_path *path, int level, |
| 104 | struct btrfs_key *key); |
| 105 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
| 106 | int dump_block_groups); |
| 107 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
| 108 | u64 num_bytes, int reserve); |
| 109 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
| 110 | u64 num_bytes); |
| 111 | int btrfs_pin_extent(struct btrfs_root *root, |
| 112 | u64 bytenr, u64 num_bytes, int reserved); |
| 113 | |
| 114 | static noinline int |
| 115 | block_group_cache_done(struct btrfs_block_group_cache *cache) |
| 116 | { |
| 117 | smp_mb(); |
| 118 | return cache->cached == BTRFS_CACHE_FINISHED || |
| 119 | cache->cached == BTRFS_CACHE_ERROR; |
| 120 | } |
| 121 | |
| 122 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
| 123 | { |
| 124 | return (cache->flags & bits) == bits; |
| 125 | } |
| 126 | |
| 127 | static void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
| 128 | { |
| 129 | atomic_inc(&cache->count); |
| 130 | } |
| 131 | |
| 132 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) |
| 133 | { |
| 134 | if (atomic_dec_and_test(&cache->count)) { |
| 135 | WARN_ON(cache->pinned > 0); |
| 136 | WARN_ON(cache->reserved > 0); |
| 137 | kfree(cache->free_space_ctl); |
| 138 | kfree(cache); |
| 139 | } |
| 140 | } |
| 141 | |
| 142 | /* |
| 143 | * this adds the block group to the fs_info rb tree for the block group |
| 144 | * cache |
| 145 | */ |
| 146 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
| 147 | struct btrfs_block_group_cache *block_group) |
| 148 | { |
| 149 | struct rb_node **p; |
| 150 | struct rb_node *parent = NULL; |
| 151 | struct btrfs_block_group_cache *cache; |
| 152 | |
| 153 | spin_lock(&info->block_group_cache_lock); |
| 154 | p = &info->block_group_cache_tree.rb_node; |
| 155 | |
| 156 | while (*p) { |
| 157 | parent = *p; |
| 158 | cache = rb_entry(parent, struct btrfs_block_group_cache, |
| 159 | cache_node); |
| 160 | if (block_group->key.objectid < cache->key.objectid) { |
| 161 | p = &(*p)->rb_left; |
| 162 | } else if (block_group->key.objectid > cache->key.objectid) { |
| 163 | p = &(*p)->rb_right; |
| 164 | } else { |
| 165 | spin_unlock(&info->block_group_cache_lock); |
| 166 | return -EEXIST; |
| 167 | } |
| 168 | } |
| 169 | |
| 170 | rb_link_node(&block_group->cache_node, parent, p); |
| 171 | rb_insert_color(&block_group->cache_node, |
| 172 | &info->block_group_cache_tree); |
| 173 | |
| 174 | if (info->first_logical_byte > block_group->key.objectid) |
| 175 | info->first_logical_byte = block_group->key.objectid; |
| 176 | |
| 177 | spin_unlock(&info->block_group_cache_lock); |
| 178 | |
| 179 | return 0; |
| 180 | } |
| 181 | |
| 182 | /* |
| 183 | * This will return the block group at or after bytenr if contains is 0, else |
| 184 | * it will return the block group that contains the bytenr |
| 185 | */ |
| 186 | static struct btrfs_block_group_cache * |
| 187 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, |
| 188 | int contains) |
| 189 | { |
| 190 | struct btrfs_block_group_cache *cache, *ret = NULL; |
| 191 | struct rb_node *n; |
| 192 | u64 end, start; |
| 193 | |
| 194 | spin_lock(&info->block_group_cache_lock); |
| 195 | n = info->block_group_cache_tree.rb_node; |
| 196 | |
| 197 | while (n) { |
| 198 | cache = rb_entry(n, struct btrfs_block_group_cache, |
| 199 | cache_node); |
| 200 | end = cache->key.objectid + cache->key.offset - 1; |
| 201 | start = cache->key.objectid; |
| 202 | |
| 203 | if (bytenr < start) { |
| 204 | if (!contains && (!ret || start < ret->key.objectid)) |
| 205 | ret = cache; |
| 206 | n = n->rb_left; |
| 207 | } else if (bytenr > start) { |
| 208 | if (contains && bytenr <= end) { |
| 209 | ret = cache; |
| 210 | break; |
| 211 | } |
| 212 | n = n->rb_right; |
| 213 | } else { |
| 214 | ret = cache; |
| 215 | break; |
| 216 | } |
| 217 | } |
| 218 | if (ret) { |
| 219 | btrfs_get_block_group(ret); |
| 220 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
| 221 | info->first_logical_byte = ret->key.objectid; |
| 222 | } |
| 223 | spin_unlock(&info->block_group_cache_lock); |
| 224 | |
| 225 | return ret; |
| 226 | } |
| 227 | |
| 228 | static int add_excluded_extent(struct btrfs_root *root, |
| 229 | u64 start, u64 num_bytes) |
| 230 | { |
| 231 | u64 end = start + num_bytes - 1; |
| 232 | set_extent_bits(&root->fs_info->freed_extents[0], |
| 233 | start, end, EXTENT_UPTODATE, GFP_NOFS); |
| 234 | set_extent_bits(&root->fs_info->freed_extents[1], |
| 235 | start, end, EXTENT_UPTODATE, GFP_NOFS); |
| 236 | return 0; |
| 237 | } |
| 238 | |
| 239 | static void free_excluded_extents(struct btrfs_root *root, |
| 240 | struct btrfs_block_group_cache *cache) |
| 241 | { |
| 242 | u64 start, end; |
| 243 | |
| 244 | start = cache->key.objectid; |
| 245 | end = start + cache->key.offset - 1; |
| 246 | |
| 247 | clear_extent_bits(&root->fs_info->freed_extents[0], |
| 248 | start, end, EXTENT_UPTODATE, GFP_NOFS); |
| 249 | clear_extent_bits(&root->fs_info->freed_extents[1], |
| 250 | start, end, EXTENT_UPTODATE, GFP_NOFS); |
| 251 | } |
| 252 | |
| 253 | static int exclude_super_stripes(struct btrfs_root *root, |
| 254 | struct btrfs_block_group_cache *cache) |
| 255 | { |
| 256 | u64 bytenr; |
| 257 | u64 *logical; |
| 258 | int stripe_len; |
| 259 | int i, nr, ret; |
| 260 | |
| 261 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
| 262 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; |
| 263 | cache->bytes_super += stripe_len; |
| 264 | ret = add_excluded_extent(root, cache->key.objectid, |
| 265 | stripe_len); |
| 266 | if (ret) |
| 267 | return ret; |
| 268 | } |
| 269 | |
| 270 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
| 271 | bytenr = btrfs_sb_offset(i); |
| 272 | ret = btrfs_rmap_block(&root->fs_info->mapping_tree, |
| 273 | cache->key.objectid, bytenr, |
| 274 | 0, &logical, &nr, &stripe_len); |
| 275 | if (ret) |
| 276 | return ret; |
| 277 | |
| 278 | while (nr--) { |
| 279 | u64 start, len; |
| 280 | |
| 281 | if (logical[nr] > cache->key.objectid + |
| 282 | cache->key.offset) |
| 283 | continue; |
| 284 | |
| 285 | if (logical[nr] + stripe_len <= cache->key.objectid) |
| 286 | continue; |
| 287 | |
| 288 | start = logical[nr]; |
| 289 | if (start < cache->key.objectid) { |
| 290 | start = cache->key.objectid; |
| 291 | len = (logical[nr] + stripe_len) - start; |
| 292 | } else { |
| 293 | len = min_t(u64, stripe_len, |
| 294 | cache->key.objectid + |
| 295 | cache->key.offset - start); |
| 296 | } |
| 297 | |
| 298 | cache->bytes_super += len; |
| 299 | ret = add_excluded_extent(root, start, len); |
| 300 | if (ret) { |
| 301 | kfree(logical); |
| 302 | return ret; |
| 303 | } |
| 304 | } |
| 305 | |
| 306 | kfree(logical); |
| 307 | } |
| 308 | return 0; |
| 309 | } |
| 310 | |
| 311 | static struct btrfs_caching_control * |
| 312 | get_caching_control(struct btrfs_block_group_cache *cache) |
| 313 | { |
| 314 | struct btrfs_caching_control *ctl; |
| 315 | |
| 316 | spin_lock(&cache->lock); |
| 317 | if (cache->cached != BTRFS_CACHE_STARTED) { |
| 318 | spin_unlock(&cache->lock); |
| 319 | return NULL; |
| 320 | } |
| 321 | |
| 322 | /* We're loading it the fast way, so we don't have a caching_ctl. */ |
| 323 | if (!cache->caching_ctl) { |
| 324 | spin_unlock(&cache->lock); |
| 325 | return NULL; |
| 326 | } |
| 327 | |
| 328 | ctl = cache->caching_ctl; |
| 329 | atomic_inc(&ctl->count); |
| 330 | spin_unlock(&cache->lock); |
| 331 | return ctl; |
| 332 | } |
| 333 | |
| 334 | static void put_caching_control(struct btrfs_caching_control *ctl) |
| 335 | { |
| 336 | if (atomic_dec_and_test(&ctl->count)) |
| 337 | kfree(ctl); |
| 338 | } |
| 339 | |
| 340 | /* |
| 341 | * this is only called by cache_block_group, since we could have freed extents |
| 342 | * we need to check the pinned_extents for any extents that can't be used yet |
| 343 | * since their free space will be released as soon as the transaction commits. |
| 344 | */ |
| 345 | static u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
| 346 | struct btrfs_fs_info *info, u64 start, u64 end) |
| 347 | { |
| 348 | u64 extent_start, extent_end, size, total_added = 0; |
| 349 | int ret; |
| 350 | |
| 351 | while (start < end) { |
| 352 | ret = find_first_extent_bit(info->pinned_extents, start, |
| 353 | &extent_start, &extent_end, |
| 354 | EXTENT_DIRTY | EXTENT_UPTODATE, |
| 355 | NULL); |
| 356 | if (ret) |
| 357 | break; |
| 358 | |
| 359 | if (extent_start <= start) { |
| 360 | start = extent_end + 1; |
| 361 | } else if (extent_start > start && extent_start < end) { |
| 362 | size = extent_start - start; |
| 363 | total_added += size; |
| 364 | ret = btrfs_add_free_space(block_group, start, |
| 365 | size); |
| 366 | BUG_ON(ret); /* -ENOMEM or logic error */ |
| 367 | start = extent_end + 1; |
| 368 | } else { |
| 369 | break; |
| 370 | } |
| 371 | } |
| 372 | |
| 373 | if (start < end) { |
| 374 | size = end - start; |
| 375 | total_added += size; |
| 376 | ret = btrfs_add_free_space(block_group, start, size); |
| 377 | BUG_ON(ret); /* -ENOMEM or logic error */ |
| 378 | } |
| 379 | |
| 380 | return total_added; |
| 381 | } |
| 382 | |
| 383 | static noinline void caching_thread(struct btrfs_work *work) |
| 384 | { |
| 385 | struct btrfs_block_group_cache *block_group; |
| 386 | struct btrfs_fs_info *fs_info; |
| 387 | struct btrfs_caching_control *caching_ctl; |
| 388 | struct btrfs_root *extent_root; |
| 389 | struct btrfs_path *path; |
| 390 | struct extent_buffer *leaf; |
| 391 | struct btrfs_key key; |
| 392 | u64 total_found = 0; |
| 393 | u64 last = 0; |
| 394 | u32 nritems; |
| 395 | int ret = -ENOMEM; |
| 396 | |
| 397 | caching_ctl = container_of(work, struct btrfs_caching_control, work); |
| 398 | block_group = caching_ctl->block_group; |
| 399 | fs_info = block_group->fs_info; |
| 400 | extent_root = fs_info->extent_root; |
| 401 | |
| 402 | path = btrfs_alloc_path(); |
| 403 | if (!path) |
| 404 | goto out; |
| 405 | |
| 406 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
| 407 | |
| 408 | /* |
| 409 | * We don't want to deadlock with somebody trying to allocate a new |
| 410 | * extent for the extent root while also trying to search the extent |
| 411 | * root to add free space. So we skip locking and search the commit |
| 412 | * root, since its read-only |
| 413 | */ |
| 414 | path->skip_locking = 1; |
| 415 | path->search_commit_root = 1; |
| 416 | path->reada = 1; |
| 417 | |
| 418 | key.objectid = last; |
| 419 | key.offset = 0; |
| 420 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 421 | again: |
| 422 | mutex_lock(&caching_ctl->mutex); |
| 423 | /* need to make sure the commit_root doesn't disappear */ |
| 424 | down_read(&fs_info->commit_root_sem); |
| 425 | |
| 426 | next: |
| 427 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
| 428 | if (ret < 0) |
| 429 | goto err; |
| 430 | |
| 431 | leaf = path->nodes[0]; |
| 432 | nritems = btrfs_header_nritems(leaf); |
| 433 | |
| 434 | while (1) { |
| 435 | if (btrfs_fs_closing(fs_info) > 1) { |
| 436 | last = (u64)-1; |
| 437 | break; |
| 438 | } |
| 439 | |
| 440 | if (path->slots[0] < nritems) { |
| 441 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
| 442 | } else { |
| 443 | ret = find_next_key(path, 0, &key); |
| 444 | if (ret) |
| 445 | break; |
| 446 | |
| 447 | if (need_resched() || |
| 448 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
| 449 | caching_ctl->progress = last; |
| 450 | btrfs_release_path(path); |
| 451 | up_read(&fs_info->commit_root_sem); |
| 452 | mutex_unlock(&caching_ctl->mutex); |
| 453 | cond_resched(); |
| 454 | goto again; |
| 455 | } |
| 456 | |
| 457 | ret = btrfs_next_leaf(extent_root, path); |
| 458 | if (ret < 0) |
| 459 | goto err; |
| 460 | if (ret) |
| 461 | break; |
| 462 | leaf = path->nodes[0]; |
| 463 | nritems = btrfs_header_nritems(leaf); |
| 464 | continue; |
| 465 | } |
| 466 | |
| 467 | if (key.objectid < last) { |
| 468 | key.objectid = last; |
| 469 | key.offset = 0; |
| 470 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 471 | |
| 472 | caching_ctl->progress = last; |
| 473 | btrfs_release_path(path); |
| 474 | goto next; |
| 475 | } |
| 476 | |
| 477 | if (key.objectid < block_group->key.objectid) { |
| 478 | path->slots[0]++; |
| 479 | continue; |
| 480 | } |
| 481 | |
| 482 | if (key.objectid >= block_group->key.objectid + |
| 483 | block_group->key.offset) |
| 484 | break; |
| 485 | |
| 486 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
| 487 | key.type == BTRFS_METADATA_ITEM_KEY) { |
| 488 | total_found += add_new_free_space(block_group, |
| 489 | fs_info, last, |
| 490 | key.objectid); |
| 491 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
| 492 | last = key.objectid + |
| 493 | fs_info->tree_root->leafsize; |
| 494 | else |
| 495 | last = key.objectid + key.offset; |
| 496 | |
| 497 | if (total_found > (1024 * 1024 * 2)) { |
| 498 | total_found = 0; |
| 499 | wake_up(&caching_ctl->wait); |
| 500 | } |
| 501 | } |
| 502 | path->slots[0]++; |
| 503 | } |
| 504 | ret = 0; |
| 505 | |
| 506 | total_found += add_new_free_space(block_group, fs_info, last, |
| 507 | block_group->key.objectid + |
| 508 | block_group->key.offset); |
| 509 | caching_ctl->progress = (u64)-1; |
| 510 | |
| 511 | spin_lock(&block_group->lock); |
| 512 | block_group->caching_ctl = NULL; |
| 513 | block_group->cached = BTRFS_CACHE_FINISHED; |
| 514 | spin_unlock(&block_group->lock); |
| 515 | |
| 516 | err: |
| 517 | btrfs_free_path(path); |
| 518 | up_read(&fs_info->commit_root_sem); |
| 519 | |
| 520 | free_excluded_extents(extent_root, block_group); |
| 521 | |
| 522 | mutex_unlock(&caching_ctl->mutex); |
| 523 | out: |
| 524 | if (ret) { |
| 525 | spin_lock(&block_group->lock); |
| 526 | block_group->caching_ctl = NULL; |
| 527 | block_group->cached = BTRFS_CACHE_ERROR; |
| 528 | spin_unlock(&block_group->lock); |
| 529 | } |
| 530 | wake_up(&caching_ctl->wait); |
| 531 | |
| 532 | put_caching_control(caching_ctl); |
| 533 | btrfs_put_block_group(block_group); |
| 534 | } |
| 535 | |
| 536 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
| 537 | int load_cache_only) |
| 538 | { |
| 539 | DEFINE_WAIT(wait); |
| 540 | struct btrfs_fs_info *fs_info = cache->fs_info; |
| 541 | struct btrfs_caching_control *caching_ctl; |
| 542 | int ret = 0; |
| 543 | |
| 544 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
| 545 | if (!caching_ctl) |
| 546 | return -ENOMEM; |
| 547 | |
| 548 | INIT_LIST_HEAD(&caching_ctl->list); |
| 549 | mutex_init(&caching_ctl->mutex); |
| 550 | init_waitqueue_head(&caching_ctl->wait); |
| 551 | caching_ctl->block_group = cache; |
| 552 | caching_ctl->progress = cache->key.objectid; |
| 553 | atomic_set(&caching_ctl->count, 1); |
| 554 | btrfs_init_work(&caching_ctl->work, caching_thread, NULL, NULL); |
| 555 | |
| 556 | spin_lock(&cache->lock); |
| 557 | /* |
| 558 | * This should be a rare occasion, but this could happen I think in the |
| 559 | * case where one thread starts to load the space cache info, and then |
| 560 | * some other thread starts a transaction commit which tries to do an |
| 561 | * allocation while the other thread is still loading the space cache |
| 562 | * info. The previous loop should have kept us from choosing this block |
| 563 | * group, but if we've moved to the state where we will wait on caching |
| 564 | * block groups we need to first check if we're doing a fast load here, |
| 565 | * so we can wait for it to finish, otherwise we could end up allocating |
| 566 | * from a block group who's cache gets evicted for one reason or |
| 567 | * another. |
| 568 | */ |
| 569 | while (cache->cached == BTRFS_CACHE_FAST) { |
| 570 | struct btrfs_caching_control *ctl; |
| 571 | |
| 572 | ctl = cache->caching_ctl; |
| 573 | atomic_inc(&ctl->count); |
| 574 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); |
| 575 | spin_unlock(&cache->lock); |
| 576 | |
| 577 | schedule(); |
| 578 | |
| 579 | finish_wait(&ctl->wait, &wait); |
| 580 | put_caching_control(ctl); |
| 581 | spin_lock(&cache->lock); |
| 582 | } |
| 583 | |
| 584 | if (cache->cached != BTRFS_CACHE_NO) { |
| 585 | spin_unlock(&cache->lock); |
| 586 | kfree(caching_ctl); |
| 587 | return 0; |
| 588 | } |
| 589 | WARN_ON(cache->caching_ctl); |
| 590 | cache->caching_ctl = caching_ctl; |
| 591 | cache->cached = BTRFS_CACHE_FAST; |
| 592 | spin_unlock(&cache->lock); |
| 593 | |
| 594 | if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) { |
| 595 | ret = load_free_space_cache(fs_info, cache); |
| 596 | |
| 597 | spin_lock(&cache->lock); |
| 598 | if (ret == 1) { |
| 599 | cache->caching_ctl = NULL; |
| 600 | cache->cached = BTRFS_CACHE_FINISHED; |
| 601 | cache->last_byte_to_unpin = (u64)-1; |
| 602 | } else { |
| 603 | if (load_cache_only) { |
| 604 | cache->caching_ctl = NULL; |
| 605 | cache->cached = BTRFS_CACHE_NO; |
| 606 | } else { |
| 607 | cache->cached = BTRFS_CACHE_STARTED; |
| 608 | } |
| 609 | } |
| 610 | spin_unlock(&cache->lock); |
| 611 | wake_up(&caching_ctl->wait); |
| 612 | if (ret == 1) { |
| 613 | put_caching_control(caching_ctl); |
| 614 | free_excluded_extents(fs_info->extent_root, cache); |
| 615 | return 0; |
| 616 | } |
| 617 | } else { |
| 618 | /* |
| 619 | * We are not going to do the fast caching, set cached to the |
| 620 | * appropriate value and wakeup any waiters. |
| 621 | */ |
| 622 | spin_lock(&cache->lock); |
| 623 | if (load_cache_only) { |
| 624 | cache->caching_ctl = NULL; |
| 625 | cache->cached = BTRFS_CACHE_NO; |
| 626 | } else { |
| 627 | cache->cached = BTRFS_CACHE_STARTED; |
| 628 | } |
| 629 | spin_unlock(&cache->lock); |
| 630 | wake_up(&caching_ctl->wait); |
| 631 | } |
| 632 | |
| 633 | if (load_cache_only) { |
| 634 | put_caching_control(caching_ctl); |
| 635 | return 0; |
| 636 | } |
| 637 | |
| 638 | down_write(&fs_info->commit_root_sem); |
| 639 | atomic_inc(&caching_ctl->count); |
| 640 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
| 641 | up_write(&fs_info->commit_root_sem); |
| 642 | |
| 643 | btrfs_get_block_group(cache); |
| 644 | |
| 645 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
| 646 | |
| 647 | return ret; |
| 648 | } |
| 649 | |
| 650 | /* |
| 651 | * return the block group that starts at or after bytenr |
| 652 | */ |
| 653 | static struct btrfs_block_group_cache * |
| 654 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) |
| 655 | { |
| 656 | struct btrfs_block_group_cache *cache; |
| 657 | |
| 658 | cache = block_group_cache_tree_search(info, bytenr, 0); |
| 659 | |
| 660 | return cache; |
| 661 | } |
| 662 | |
| 663 | /* |
| 664 | * return the block group that contains the given bytenr |
| 665 | */ |
| 666 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
| 667 | struct btrfs_fs_info *info, |
| 668 | u64 bytenr) |
| 669 | { |
| 670 | struct btrfs_block_group_cache *cache; |
| 671 | |
| 672 | cache = block_group_cache_tree_search(info, bytenr, 1); |
| 673 | |
| 674 | return cache; |
| 675 | } |
| 676 | |
| 677 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
| 678 | u64 flags) |
| 679 | { |
| 680 | struct list_head *head = &info->space_info; |
| 681 | struct btrfs_space_info *found; |
| 682 | |
| 683 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
| 684 | |
| 685 | rcu_read_lock(); |
| 686 | list_for_each_entry_rcu(found, head, list) { |
| 687 | if (found->flags & flags) { |
| 688 | rcu_read_unlock(); |
| 689 | return found; |
| 690 | } |
| 691 | } |
| 692 | rcu_read_unlock(); |
| 693 | return NULL; |
| 694 | } |
| 695 | |
| 696 | /* |
| 697 | * after adding space to the filesystem, we need to clear the full flags |
| 698 | * on all the space infos. |
| 699 | */ |
| 700 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) |
| 701 | { |
| 702 | struct list_head *head = &info->space_info; |
| 703 | struct btrfs_space_info *found; |
| 704 | |
| 705 | rcu_read_lock(); |
| 706 | list_for_each_entry_rcu(found, head, list) |
| 707 | found->full = 0; |
| 708 | rcu_read_unlock(); |
| 709 | } |
| 710 | |
| 711 | /* simple helper to search for an existing extent at a given offset */ |
| 712 | int btrfs_lookup_extent(struct btrfs_root *root, u64 start, u64 len) |
| 713 | { |
| 714 | int ret; |
| 715 | struct btrfs_key key; |
| 716 | struct btrfs_path *path; |
| 717 | |
| 718 | path = btrfs_alloc_path(); |
| 719 | if (!path) |
| 720 | return -ENOMEM; |
| 721 | |
| 722 | key.objectid = start; |
| 723 | key.offset = len; |
| 724 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 725 | ret = btrfs_search_slot(NULL, root->fs_info->extent_root, &key, path, |
| 726 | 0, 0); |
| 727 | if (ret > 0) { |
| 728 | btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]); |
| 729 | if (key.objectid == start && |
| 730 | key.type == BTRFS_METADATA_ITEM_KEY) |
| 731 | ret = 0; |
| 732 | } |
| 733 | btrfs_free_path(path); |
| 734 | return ret; |
| 735 | } |
| 736 | |
| 737 | /* |
| 738 | * helper function to lookup reference count and flags of a tree block. |
| 739 | * |
| 740 | * the head node for delayed ref is used to store the sum of all the |
| 741 | * reference count modifications queued up in the rbtree. the head |
| 742 | * node may also store the extent flags to set. This way you can check |
| 743 | * to see what the reference count and extent flags would be if all of |
| 744 | * the delayed refs are not processed. |
| 745 | */ |
| 746 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, |
| 747 | struct btrfs_root *root, u64 bytenr, |
| 748 | u64 offset, int metadata, u64 *refs, u64 *flags) |
| 749 | { |
| 750 | struct btrfs_delayed_ref_head *head; |
| 751 | struct btrfs_delayed_ref_root *delayed_refs; |
| 752 | struct btrfs_path *path; |
| 753 | struct btrfs_extent_item *ei; |
| 754 | struct extent_buffer *leaf; |
| 755 | struct btrfs_key key; |
| 756 | u32 item_size; |
| 757 | u64 num_refs; |
| 758 | u64 extent_flags; |
| 759 | int ret; |
| 760 | |
| 761 | /* |
| 762 | * If we don't have skinny metadata, don't bother doing anything |
| 763 | * different |
| 764 | */ |
| 765 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) { |
| 766 | offset = root->leafsize; |
| 767 | metadata = 0; |
| 768 | } |
| 769 | |
| 770 | path = btrfs_alloc_path(); |
| 771 | if (!path) |
| 772 | return -ENOMEM; |
| 773 | |
| 774 | if (!trans) { |
| 775 | path->skip_locking = 1; |
| 776 | path->search_commit_root = 1; |
| 777 | } |
| 778 | |
| 779 | search_again: |
| 780 | key.objectid = bytenr; |
| 781 | key.offset = offset; |
| 782 | if (metadata) |
| 783 | key.type = BTRFS_METADATA_ITEM_KEY; |
| 784 | else |
| 785 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 786 | |
| 787 | again: |
| 788 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, |
| 789 | &key, path, 0, 0); |
| 790 | if (ret < 0) |
| 791 | goto out_free; |
| 792 | |
| 793 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
| 794 | if (path->slots[0]) { |
| 795 | path->slots[0]--; |
| 796 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
| 797 | path->slots[0]); |
| 798 | if (key.objectid == bytenr && |
| 799 | key.type == BTRFS_EXTENT_ITEM_KEY && |
| 800 | key.offset == root->leafsize) |
| 801 | ret = 0; |
| 802 | } |
| 803 | if (ret) { |
| 804 | key.objectid = bytenr; |
| 805 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 806 | key.offset = root->leafsize; |
| 807 | btrfs_release_path(path); |
| 808 | goto again; |
| 809 | } |
| 810 | } |
| 811 | |
| 812 | if (ret == 0) { |
| 813 | leaf = path->nodes[0]; |
| 814 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); |
| 815 | if (item_size >= sizeof(*ei)) { |
| 816 | ei = btrfs_item_ptr(leaf, path->slots[0], |
| 817 | struct btrfs_extent_item); |
| 818 | num_refs = btrfs_extent_refs(leaf, ei); |
| 819 | extent_flags = btrfs_extent_flags(leaf, ei); |
| 820 | } else { |
| 821 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 822 | struct btrfs_extent_item_v0 *ei0; |
| 823 | BUG_ON(item_size != sizeof(*ei0)); |
| 824 | ei0 = btrfs_item_ptr(leaf, path->slots[0], |
| 825 | struct btrfs_extent_item_v0); |
| 826 | num_refs = btrfs_extent_refs_v0(leaf, ei0); |
| 827 | /* FIXME: this isn't correct for data */ |
| 828 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
| 829 | #else |
| 830 | BUG(); |
| 831 | #endif |
| 832 | } |
| 833 | BUG_ON(num_refs == 0); |
| 834 | } else { |
| 835 | num_refs = 0; |
| 836 | extent_flags = 0; |
| 837 | ret = 0; |
| 838 | } |
| 839 | |
| 840 | if (!trans) |
| 841 | goto out; |
| 842 | |
| 843 | delayed_refs = &trans->transaction->delayed_refs; |
| 844 | spin_lock(&delayed_refs->lock); |
| 845 | head = btrfs_find_delayed_ref_head(trans, bytenr); |
| 846 | if (head) { |
| 847 | if (!mutex_trylock(&head->mutex)) { |
| 848 | atomic_inc(&head->node.refs); |
| 849 | spin_unlock(&delayed_refs->lock); |
| 850 | |
| 851 | btrfs_release_path(path); |
| 852 | |
| 853 | /* |
| 854 | * Mutex was contended, block until it's released and try |
| 855 | * again |
| 856 | */ |
| 857 | mutex_lock(&head->mutex); |
| 858 | mutex_unlock(&head->mutex); |
| 859 | btrfs_put_delayed_ref(&head->node); |
| 860 | goto search_again; |
| 861 | } |
| 862 | spin_lock(&head->lock); |
| 863 | if (head->extent_op && head->extent_op->update_flags) |
| 864 | extent_flags |= head->extent_op->flags_to_set; |
| 865 | else |
| 866 | BUG_ON(num_refs == 0); |
| 867 | |
| 868 | num_refs += head->node.ref_mod; |
| 869 | spin_unlock(&head->lock); |
| 870 | mutex_unlock(&head->mutex); |
| 871 | } |
| 872 | spin_unlock(&delayed_refs->lock); |
| 873 | out: |
| 874 | WARN_ON(num_refs == 0); |
| 875 | if (refs) |
| 876 | *refs = num_refs; |
| 877 | if (flags) |
| 878 | *flags = extent_flags; |
| 879 | out_free: |
| 880 | btrfs_free_path(path); |
| 881 | return ret; |
| 882 | } |
| 883 | |
| 884 | /* |
| 885 | * Back reference rules. Back refs have three main goals: |
| 886 | * |
| 887 | * 1) differentiate between all holders of references to an extent so that |
| 888 | * when a reference is dropped we can make sure it was a valid reference |
| 889 | * before freeing the extent. |
| 890 | * |
| 891 | * 2) Provide enough information to quickly find the holders of an extent |
| 892 | * if we notice a given block is corrupted or bad. |
| 893 | * |
| 894 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool |
| 895 | * maintenance. This is actually the same as #2, but with a slightly |
| 896 | * different use case. |
| 897 | * |
| 898 | * There are two kinds of back refs. The implicit back refs is optimized |
| 899 | * for pointers in non-shared tree blocks. For a given pointer in a block, |
| 900 | * back refs of this kind provide information about the block's owner tree |
| 901 | * and the pointer's key. These information allow us to find the block by |
| 902 | * b-tree searching. The full back refs is for pointers in tree blocks not |
| 903 | * referenced by their owner trees. The location of tree block is recorded |
| 904 | * in the back refs. Actually the full back refs is generic, and can be |
| 905 | * used in all cases the implicit back refs is used. The major shortcoming |
| 906 | * of the full back refs is its overhead. Every time a tree block gets |
| 907 | * COWed, we have to update back refs entry for all pointers in it. |
| 908 | * |
| 909 | * For a newly allocated tree block, we use implicit back refs for |
| 910 | * pointers in it. This means most tree related operations only involve |
| 911 | * implicit back refs. For a tree block created in old transaction, the |
| 912 | * only way to drop a reference to it is COW it. So we can detect the |
| 913 | * event that tree block loses its owner tree's reference and do the |
| 914 | * back refs conversion. |
| 915 | * |
| 916 | * When a tree block is COW'd through a tree, there are four cases: |
| 917 | * |
| 918 | * The reference count of the block is one and the tree is the block's |
| 919 | * owner tree. Nothing to do in this case. |
| 920 | * |
| 921 | * The reference count of the block is one and the tree is not the |
| 922 | * block's owner tree. In this case, full back refs is used for pointers |
| 923 | * in the block. Remove these full back refs, add implicit back refs for |
| 924 | * every pointers in the new block. |
| 925 | * |
| 926 | * The reference count of the block is greater than one and the tree is |
| 927 | * the block's owner tree. In this case, implicit back refs is used for |
| 928 | * pointers in the block. Add full back refs for every pointers in the |
| 929 | * block, increase lower level extents' reference counts. The original |
| 930 | * implicit back refs are entailed to the new block. |
| 931 | * |
| 932 | * The reference count of the block is greater than one and the tree is |
| 933 | * not the block's owner tree. Add implicit back refs for every pointer in |
| 934 | * the new block, increase lower level extents' reference count. |
| 935 | * |
| 936 | * Back Reference Key composing: |
| 937 | * |
| 938 | * The key objectid corresponds to the first byte in the extent, |
| 939 | * The key type is used to differentiate between types of back refs. |
| 940 | * There are different meanings of the key offset for different types |
| 941 | * of back refs. |
| 942 | * |
| 943 | * File extents can be referenced by: |
| 944 | * |
| 945 | * - multiple snapshots, subvolumes, or different generations in one subvol |
| 946 | * - different files inside a single subvolume |
| 947 | * - different offsets inside a file (bookend extents in file.c) |
| 948 | * |
| 949 | * The extent ref structure for the implicit back refs has fields for: |
| 950 | * |
| 951 | * - Objectid of the subvolume root |
| 952 | * - objectid of the file holding the reference |
| 953 | * - original offset in the file |
| 954 | * - how many bookend extents |
| 955 | * |
| 956 | * The key offset for the implicit back refs is hash of the first |
| 957 | * three fields. |
| 958 | * |
| 959 | * The extent ref structure for the full back refs has field for: |
| 960 | * |
| 961 | * - number of pointers in the tree leaf |
| 962 | * |
| 963 | * The key offset for the implicit back refs is the first byte of |
| 964 | * the tree leaf |
| 965 | * |
| 966 | * When a file extent is allocated, The implicit back refs is used. |
| 967 | * the fields are filled in: |
| 968 | * |
| 969 | * (root_key.objectid, inode objectid, offset in file, 1) |
| 970 | * |
| 971 | * When a file extent is removed file truncation, we find the |
| 972 | * corresponding implicit back refs and check the following fields: |
| 973 | * |
| 974 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
| 975 | * |
| 976 | * Btree extents can be referenced by: |
| 977 | * |
| 978 | * - Different subvolumes |
| 979 | * |
| 980 | * Both the implicit back refs and the full back refs for tree blocks |
| 981 | * only consist of key. The key offset for the implicit back refs is |
| 982 | * objectid of block's owner tree. The key offset for the full back refs |
| 983 | * is the first byte of parent block. |
| 984 | * |
| 985 | * When implicit back refs is used, information about the lowest key and |
| 986 | * level of the tree block are required. These information are stored in |
| 987 | * tree block info structure. |
| 988 | */ |
| 989 | |
| 990 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 991 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, |
| 992 | struct btrfs_root *root, |
| 993 | struct btrfs_path *path, |
| 994 | u64 owner, u32 extra_size) |
| 995 | { |
| 996 | struct btrfs_extent_item *item; |
| 997 | struct btrfs_extent_item_v0 *ei0; |
| 998 | struct btrfs_extent_ref_v0 *ref0; |
| 999 | struct btrfs_tree_block_info *bi; |
| 1000 | struct extent_buffer *leaf; |
| 1001 | struct btrfs_key key; |
| 1002 | struct btrfs_key found_key; |
| 1003 | u32 new_size = sizeof(*item); |
| 1004 | u64 refs; |
| 1005 | int ret; |
| 1006 | |
| 1007 | leaf = path->nodes[0]; |
| 1008 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); |
| 1009 | |
| 1010 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
| 1011 | ei0 = btrfs_item_ptr(leaf, path->slots[0], |
| 1012 | struct btrfs_extent_item_v0); |
| 1013 | refs = btrfs_extent_refs_v0(leaf, ei0); |
| 1014 | |
| 1015 | if (owner == (u64)-1) { |
| 1016 | while (1) { |
| 1017 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { |
| 1018 | ret = btrfs_next_leaf(root, path); |
| 1019 | if (ret < 0) |
| 1020 | return ret; |
| 1021 | BUG_ON(ret > 0); /* Corruption */ |
| 1022 | leaf = path->nodes[0]; |
| 1023 | } |
| 1024 | btrfs_item_key_to_cpu(leaf, &found_key, |
| 1025 | path->slots[0]); |
| 1026 | BUG_ON(key.objectid != found_key.objectid); |
| 1027 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { |
| 1028 | path->slots[0]++; |
| 1029 | continue; |
| 1030 | } |
| 1031 | ref0 = btrfs_item_ptr(leaf, path->slots[0], |
| 1032 | struct btrfs_extent_ref_v0); |
| 1033 | owner = btrfs_ref_objectid_v0(leaf, ref0); |
| 1034 | break; |
| 1035 | } |
| 1036 | } |
| 1037 | btrfs_release_path(path); |
| 1038 | |
| 1039 | if (owner < BTRFS_FIRST_FREE_OBJECTID) |
| 1040 | new_size += sizeof(*bi); |
| 1041 | |
| 1042 | new_size -= sizeof(*ei0); |
| 1043 | ret = btrfs_search_slot(trans, root, &key, path, |
| 1044 | new_size + extra_size, 1); |
| 1045 | if (ret < 0) |
| 1046 | return ret; |
| 1047 | BUG_ON(ret); /* Corruption */ |
| 1048 | |
| 1049 | btrfs_extend_item(root, path, new_size); |
| 1050 | |
| 1051 | leaf = path->nodes[0]; |
| 1052 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
| 1053 | btrfs_set_extent_refs(leaf, item, refs); |
| 1054 | /* FIXME: get real generation */ |
| 1055 | btrfs_set_extent_generation(leaf, item, 0); |
| 1056 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
| 1057 | btrfs_set_extent_flags(leaf, item, |
| 1058 | BTRFS_EXTENT_FLAG_TREE_BLOCK | |
| 1059 | BTRFS_BLOCK_FLAG_FULL_BACKREF); |
| 1060 | bi = (struct btrfs_tree_block_info *)(item + 1); |
| 1061 | /* FIXME: get first key of the block */ |
| 1062 | memset_extent_buffer(leaf, 0, (unsigned long)bi, sizeof(*bi)); |
| 1063 | btrfs_set_tree_block_level(leaf, bi, (int)owner); |
| 1064 | } else { |
| 1065 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); |
| 1066 | } |
| 1067 | btrfs_mark_buffer_dirty(leaf); |
| 1068 | return 0; |
| 1069 | } |
| 1070 | #endif |
| 1071 | |
| 1072 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
| 1073 | { |
| 1074 | u32 high_crc = ~(u32)0; |
| 1075 | u32 low_crc = ~(u32)0; |
| 1076 | __le64 lenum; |
| 1077 | |
| 1078 | lenum = cpu_to_le64(root_objectid); |
| 1079 | high_crc = btrfs_crc32c(high_crc, &lenum, sizeof(lenum)); |
| 1080 | lenum = cpu_to_le64(owner); |
| 1081 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
| 1082 | lenum = cpu_to_le64(offset); |
| 1083 | low_crc = btrfs_crc32c(low_crc, &lenum, sizeof(lenum)); |
| 1084 | |
| 1085 | return ((u64)high_crc << 31) ^ (u64)low_crc; |
| 1086 | } |
| 1087 | |
| 1088 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, |
| 1089 | struct btrfs_extent_data_ref *ref) |
| 1090 | { |
| 1091 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), |
| 1092 | btrfs_extent_data_ref_objectid(leaf, ref), |
| 1093 | btrfs_extent_data_ref_offset(leaf, ref)); |
| 1094 | } |
| 1095 | |
| 1096 | static int match_extent_data_ref(struct extent_buffer *leaf, |
| 1097 | struct btrfs_extent_data_ref *ref, |
| 1098 | u64 root_objectid, u64 owner, u64 offset) |
| 1099 | { |
| 1100 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || |
| 1101 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || |
| 1102 | btrfs_extent_data_ref_offset(leaf, ref) != offset) |
| 1103 | return 0; |
| 1104 | return 1; |
| 1105 | } |
| 1106 | |
| 1107 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, |
| 1108 | struct btrfs_root *root, |
| 1109 | struct btrfs_path *path, |
| 1110 | u64 bytenr, u64 parent, |
| 1111 | u64 root_objectid, |
| 1112 | u64 owner, u64 offset) |
| 1113 | { |
| 1114 | struct btrfs_key key; |
| 1115 | struct btrfs_extent_data_ref *ref; |
| 1116 | struct extent_buffer *leaf; |
| 1117 | u32 nritems; |
| 1118 | int ret; |
| 1119 | int recow; |
| 1120 | int err = -ENOENT; |
| 1121 | |
| 1122 | key.objectid = bytenr; |
| 1123 | if (parent) { |
| 1124 | key.type = BTRFS_SHARED_DATA_REF_KEY; |
| 1125 | key.offset = parent; |
| 1126 | } else { |
| 1127 | key.type = BTRFS_EXTENT_DATA_REF_KEY; |
| 1128 | key.offset = hash_extent_data_ref(root_objectid, |
| 1129 | owner, offset); |
| 1130 | } |
| 1131 | again: |
| 1132 | recow = 0; |
| 1133 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
| 1134 | if (ret < 0) { |
| 1135 | err = ret; |
| 1136 | goto fail; |
| 1137 | } |
| 1138 | |
| 1139 | if (parent) { |
| 1140 | if (!ret) |
| 1141 | return 0; |
| 1142 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 1143 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
| 1144 | btrfs_release_path(path); |
| 1145 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
| 1146 | if (ret < 0) { |
| 1147 | err = ret; |
| 1148 | goto fail; |
| 1149 | } |
| 1150 | if (!ret) |
| 1151 | return 0; |
| 1152 | #endif |
| 1153 | goto fail; |
| 1154 | } |
| 1155 | |
| 1156 | leaf = path->nodes[0]; |
| 1157 | nritems = btrfs_header_nritems(leaf); |
| 1158 | while (1) { |
| 1159 | if (path->slots[0] >= nritems) { |
| 1160 | ret = btrfs_next_leaf(root, path); |
| 1161 | if (ret < 0) |
| 1162 | err = ret; |
| 1163 | if (ret) |
| 1164 | goto fail; |
| 1165 | |
| 1166 | leaf = path->nodes[0]; |
| 1167 | nritems = btrfs_header_nritems(leaf); |
| 1168 | recow = 1; |
| 1169 | } |
| 1170 | |
| 1171 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
| 1172 | if (key.objectid != bytenr || |
| 1173 | key.type != BTRFS_EXTENT_DATA_REF_KEY) |
| 1174 | goto fail; |
| 1175 | |
| 1176 | ref = btrfs_item_ptr(leaf, path->slots[0], |
| 1177 | struct btrfs_extent_data_ref); |
| 1178 | |
| 1179 | if (match_extent_data_ref(leaf, ref, root_objectid, |
| 1180 | owner, offset)) { |
| 1181 | if (recow) { |
| 1182 | btrfs_release_path(path); |
| 1183 | goto again; |
| 1184 | } |
| 1185 | err = 0; |
| 1186 | break; |
| 1187 | } |
| 1188 | path->slots[0]++; |
| 1189 | } |
| 1190 | fail: |
| 1191 | return err; |
| 1192 | } |
| 1193 | |
| 1194 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
| 1195 | struct btrfs_root *root, |
| 1196 | struct btrfs_path *path, |
| 1197 | u64 bytenr, u64 parent, |
| 1198 | u64 root_objectid, u64 owner, |
| 1199 | u64 offset, int refs_to_add) |
| 1200 | { |
| 1201 | struct btrfs_key key; |
| 1202 | struct extent_buffer *leaf; |
| 1203 | u32 size; |
| 1204 | u32 num_refs; |
| 1205 | int ret; |
| 1206 | |
| 1207 | key.objectid = bytenr; |
| 1208 | if (parent) { |
| 1209 | key.type = BTRFS_SHARED_DATA_REF_KEY; |
| 1210 | key.offset = parent; |
| 1211 | size = sizeof(struct btrfs_shared_data_ref); |
| 1212 | } else { |
| 1213 | key.type = BTRFS_EXTENT_DATA_REF_KEY; |
| 1214 | key.offset = hash_extent_data_ref(root_objectid, |
| 1215 | owner, offset); |
| 1216 | size = sizeof(struct btrfs_extent_data_ref); |
| 1217 | } |
| 1218 | |
| 1219 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
| 1220 | if (ret && ret != -EEXIST) |
| 1221 | goto fail; |
| 1222 | |
| 1223 | leaf = path->nodes[0]; |
| 1224 | if (parent) { |
| 1225 | struct btrfs_shared_data_ref *ref; |
| 1226 | ref = btrfs_item_ptr(leaf, path->slots[0], |
| 1227 | struct btrfs_shared_data_ref); |
| 1228 | if (ret == 0) { |
| 1229 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); |
| 1230 | } else { |
| 1231 | num_refs = btrfs_shared_data_ref_count(leaf, ref); |
| 1232 | num_refs += refs_to_add; |
| 1233 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); |
| 1234 | } |
| 1235 | } else { |
| 1236 | struct btrfs_extent_data_ref *ref; |
| 1237 | while (ret == -EEXIST) { |
| 1238 | ref = btrfs_item_ptr(leaf, path->slots[0], |
| 1239 | struct btrfs_extent_data_ref); |
| 1240 | if (match_extent_data_ref(leaf, ref, root_objectid, |
| 1241 | owner, offset)) |
| 1242 | break; |
| 1243 | btrfs_release_path(path); |
| 1244 | key.offset++; |
| 1245 | ret = btrfs_insert_empty_item(trans, root, path, &key, |
| 1246 | size); |
| 1247 | if (ret && ret != -EEXIST) |
| 1248 | goto fail; |
| 1249 | |
| 1250 | leaf = path->nodes[0]; |
| 1251 | } |
| 1252 | ref = btrfs_item_ptr(leaf, path->slots[0], |
| 1253 | struct btrfs_extent_data_ref); |
| 1254 | if (ret == 0) { |
| 1255 | btrfs_set_extent_data_ref_root(leaf, ref, |
| 1256 | root_objectid); |
| 1257 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); |
| 1258 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); |
| 1259 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); |
| 1260 | } else { |
| 1261 | num_refs = btrfs_extent_data_ref_count(leaf, ref); |
| 1262 | num_refs += refs_to_add; |
| 1263 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); |
| 1264 | } |
| 1265 | } |
| 1266 | btrfs_mark_buffer_dirty(leaf); |
| 1267 | ret = 0; |
| 1268 | fail: |
| 1269 | btrfs_release_path(path); |
| 1270 | return ret; |
| 1271 | } |
| 1272 | |
| 1273 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
| 1274 | struct btrfs_root *root, |
| 1275 | struct btrfs_path *path, |
| 1276 | int refs_to_drop, int *last_ref) |
| 1277 | { |
| 1278 | struct btrfs_key key; |
| 1279 | struct btrfs_extent_data_ref *ref1 = NULL; |
| 1280 | struct btrfs_shared_data_ref *ref2 = NULL; |
| 1281 | struct extent_buffer *leaf; |
| 1282 | u32 num_refs = 0; |
| 1283 | int ret = 0; |
| 1284 | |
| 1285 | leaf = path->nodes[0]; |
| 1286 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
| 1287 | |
| 1288 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { |
| 1289 | ref1 = btrfs_item_ptr(leaf, path->slots[0], |
| 1290 | struct btrfs_extent_data_ref); |
| 1291 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); |
| 1292 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { |
| 1293 | ref2 = btrfs_item_ptr(leaf, path->slots[0], |
| 1294 | struct btrfs_shared_data_ref); |
| 1295 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); |
| 1296 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 1297 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { |
| 1298 | struct btrfs_extent_ref_v0 *ref0; |
| 1299 | ref0 = btrfs_item_ptr(leaf, path->slots[0], |
| 1300 | struct btrfs_extent_ref_v0); |
| 1301 | num_refs = btrfs_ref_count_v0(leaf, ref0); |
| 1302 | #endif |
| 1303 | } else { |
| 1304 | BUG(); |
| 1305 | } |
| 1306 | |
| 1307 | BUG_ON(num_refs < refs_to_drop); |
| 1308 | num_refs -= refs_to_drop; |
| 1309 | |
| 1310 | if (num_refs == 0) { |
| 1311 | ret = btrfs_del_item(trans, root, path); |
| 1312 | *last_ref = 1; |
| 1313 | } else { |
| 1314 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
| 1315 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); |
| 1316 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) |
| 1317 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); |
| 1318 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 1319 | else { |
| 1320 | struct btrfs_extent_ref_v0 *ref0; |
| 1321 | ref0 = btrfs_item_ptr(leaf, path->slots[0], |
| 1322 | struct btrfs_extent_ref_v0); |
| 1323 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); |
| 1324 | } |
| 1325 | #endif |
| 1326 | btrfs_mark_buffer_dirty(leaf); |
| 1327 | } |
| 1328 | return ret; |
| 1329 | } |
| 1330 | |
| 1331 | static noinline u32 extent_data_ref_count(struct btrfs_root *root, |
| 1332 | struct btrfs_path *path, |
| 1333 | struct btrfs_extent_inline_ref *iref) |
| 1334 | { |
| 1335 | struct btrfs_key key; |
| 1336 | struct extent_buffer *leaf; |
| 1337 | struct btrfs_extent_data_ref *ref1; |
| 1338 | struct btrfs_shared_data_ref *ref2; |
| 1339 | u32 num_refs = 0; |
| 1340 | |
| 1341 | leaf = path->nodes[0]; |
| 1342 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
| 1343 | if (iref) { |
| 1344 | if (btrfs_extent_inline_ref_type(leaf, iref) == |
| 1345 | BTRFS_EXTENT_DATA_REF_KEY) { |
| 1346 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
| 1347 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); |
| 1348 | } else { |
| 1349 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); |
| 1350 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); |
| 1351 | } |
| 1352 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { |
| 1353 | ref1 = btrfs_item_ptr(leaf, path->slots[0], |
| 1354 | struct btrfs_extent_data_ref); |
| 1355 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); |
| 1356 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { |
| 1357 | ref2 = btrfs_item_ptr(leaf, path->slots[0], |
| 1358 | struct btrfs_shared_data_ref); |
| 1359 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); |
| 1360 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 1361 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { |
| 1362 | struct btrfs_extent_ref_v0 *ref0; |
| 1363 | ref0 = btrfs_item_ptr(leaf, path->slots[0], |
| 1364 | struct btrfs_extent_ref_v0); |
| 1365 | num_refs = btrfs_ref_count_v0(leaf, ref0); |
| 1366 | #endif |
| 1367 | } else { |
| 1368 | WARN_ON(1); |
| 1369 | } |
| 1370 | return num_refs; |
| 1371 | } |
| 1372 | |
| 1373 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
| 1374 | struct btrfs_root *root, |
| 1375 | struct btrfs_path *path, |
| 1376 | u64 bytenr, u64 parent, |
| 1377 | u64 root_objectid) |
| 1378 | { |
| 1379 | struct btrfs_key key; |
| 1380 | int ret; |
| 1381 | |
| 1382 | key.objectid = bytenr; |
| 1383 | if (parent) { |
| 1384 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; |
| 1385 | key.offset = parent; |
| 1386 | } else { |
| 1387 | key.type = BTRFS_TREE_BLOCK_REF_KEY; |
| 1388 | key.offset = root_objectid; |
| 1389 | } |
| 1390 | |
| 1391 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
| 1392 | if (ret > 0) |
| 1393 | ret = -ENOENT; |
| 1394 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 1395 | if (ret == -ENOENT && parent) { |
| 1396 | btrfs_release_path(path); |
| 1397 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
| 1398 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
| 1399 | if (ret > 0) |
| 1400 | ret = -ENOENT; |
| 1401 | } |
| 1402 | #endif |
| 1403 | return ret; |
| 1404 | } |
| 1405 | |
| 1406 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
| 1407 | struct btrfs_root *root, |
| 1408 | struct btrfs_path *path, |
| 1409 | u64 bytenr, u64 parent, |
| 1410 | u64 root_objectid) |
| 1411 | { |
| 1412 | struct btrfs_key key; |
| 1413 | int ret; |
| 1414 | |
| 1415 | key.objectid = bytenr; |
| 1416 | if (parent) { |
| 1417 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; |
| 1418 | key.offset = parent; |
| 1419 | } else { |
| 1420 | key.type = BTRFS_TREE_BLOCK_REF_KEY; |
| 1421 | key.offset = root_objectid; |
| 1422 | } |
| 1423 | |
| 1424 | ret = btrfs_insert_empty_item(trans, root, path, &key, 0); |
| 1425 | btrfs_release_path(path); |
| 1426 | return ret; |
| 1427 | } |
| 1428 | |
| 1429 | static inline int extent_ref_type(u64 parent, u64 owner) |
| 1430 | { |
| 1431 | int type; |
| 1432 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
| 1433 | if (parent > 0) |
| 1434 | type = BTRFS_SHARED_BLOCK_REF_KEY; |
| 1435 | else |
| 1436 | type = BTRFS_TREE_BLOCK_REF_KEY; |
| 1437 | } else { |
| 1438 | if (parent > 0) |
| 1439 | type = BTRFS_SHARED_DATA_REF_KEY; |
| 1440 | else |
| 1441 | type = BTRFS_EXTENT_DATA_REF_KEY; |
| 1442 | } |
| 1443 | return type; |
| 1444 | } |
| 1445 | |
| 1446 | static int find_next_key(struct btrfs_path *path, int level, |
| 1447 | struct btrfs_key *key) |
| 1448 | |
| 1449 | { |
| 1450 | for (; level < BTRFS_MAX_LEVEL; level++) { |
| 1451 | if (!path->nodes[level]) |
| 1452 | break; |
| 1453 | if (path->slots[level] + 1 >= |
| 1454 | btrfs_header_nritems(path->nodes[level])) |
| 1455 | continue; |
| 1456 | if (level == 0) |
| 1457 | btrfs_item_key_to_cpu(path->nodes[level], key, |
| 1458 | path->slots[level] + 1); |
| 1459 | else |
| 1460 | btrfs_node_key_to_cpu(path->nodes[level], key, |
| 1461 | path->slots[level] + 1); |
| 1462 | return 0; |
| 1463 | } |
| 1464 | return 1; |
| 1465 | } |
| 1466 | |
| 1467 | /* |
| 1468 | * look for inline back ref. if back ref is found, *ref_ret is set |
| 1469 | * to the address of inline back ref, and 0 is returned. |
| 1470 | * |
| 1471 | * if back ref isn't found, *ref_ret is set to the address where it |
| 1472 | * should be inserted, and -ENOENT is returned. |
| 1473 | * |
| 1474 | * if insert is true and there are too many inline back refs, the path |
| 1475 | * points to the extent item, and -EAGAIN is returned. |
| 1476 | * |
| 1477 | * NOTE: inline back refs are ordered in the same way that back ref |
| 1478 | * items in the tree are ordered. |
| 1479 | */ |
| 1480 | static noinline_for_stack |
| 1481 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, |
| 1482 | struct btrfs_root *root, |
| 1483 | struct btrfs_path *path, |
| 1484 | struct btrfs_extent_inline_ref **ref_ret, |
| 1485 | u64 bytenr, u64 num_bytes, |
| 1486 | u64 parent, u64 root_objectid, |
| 1487 | u64 owner, u64 offset, int insert) |
| 1488 | { |
| 1489 | struct btrfs_key key; |
| 1490 | struct extent_buffer *leaf; |
| 1491 | struct btrfs_extent_item *ei; |
| 1492 | struct btrfs_extent_inline_ref *iref; |
| 1493 | u64 flags; |
| 1494 | u64 item_size; |
| 1495 | unsigned long ptr; |
| 1496 | unsigned long end; |
| 1497 | int extra_size; |
| 1498 | int type; |
| 1499 | int want; |
| 1500 | int ret; |
| 1501 | int err = 0; |
| 1502 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
| 1503 | SKINNY_METADATA); |
| 1504 | |
| 1505 | key.objectid = bytenr; |
| 1506 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 1507 | key.offset = num_bytes; |
| 1508 | |
| 1509 | want = extent_ref_type(parent, owner); |
| 1510 | if (insert) { |
| 1511 | extra_size = btrfs_extent_inline_ref_size(want); |
| 1512 | path->keep_locks = 1; |
| 1513 | } else |
| 1514 | extra_size = -1; |
| 1515 | |
| 1516 | /* |
| 1517 | * Owner is our parent level, so we can just add one to get the level |
| 1518 | * for the block we are interested in. |
| 1519 | */ |
| 1520 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { |
| 1521 | key.type = BTRFS_METADATA_ITEM_KEY; |
| 1522 | key.offset = owner; |
| 1523 | } |
| 1524 | |
| 1525 | again: |
| 1526 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
| 1527 | if (ret < 0) { |
| 1528 | err = ret; |
| 1529 | goto out; |
| 1530 | } |
| 1531 | |
| 1532 | /* |
| 1533 | * We may be a newly converted file system which still has the old fat |
| 1534 | * extent entries for metadata, so try and see if we have one of those. |
| 1535 | */ |
| 1536 | if (ret > 0 && skinny_metadata) { |
| 1537 | skinny_metadata = false; |
| 1538 | if (path->slots[0]) { |
| 1539 | path->slots[0]--; |
| 1540 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
| 1541 | path->slots[0]); |
| 1542 | if (key.objectid == bytenr && |
| 1543 | key.type == BTRFS_EXTENT_ITEM_KEY && |
| 1544 | key.offset == num_bytes) |
| 1545 | ret = 0; |
| 1546 | } |
| 1547 | if (ret) { |
| 1548 | key.objectid = bytenr; |
| 1549 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 1550 | key.offset = num_bytes; |
| 1551 | btrfs_release_path(path); |
| 1552 | goto again; |
| 1553 | } |
| 1554 | } |
| 1555 | |
| 1556 | if (ret && !insert) { |
| 1557 | err = -ENOENT; |
| 1558 | goto out; |
| 1559 | } else if (WARN_ON(ret)) { |
| 1560 | err = -EIO; |
| 1561 | goto out; |
| 1562 | } |
| 1563 | |
| 1564 | leaf = path->nodes[0]; |
| 1565 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); |
| 1566 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 1567 | if (item_size < sizeof(*ei)) { |
| 1568 | if (!insert) { |
| 1569 | err = -ENOENT; |
| 1570 | goto out; |
| 1571 | } |
| 1572 | ret = convert_extent_item_v0(trans, root, path, owner, |
| 1573 | extra_size); |
| 1574 | if (ret < 0) { |
| 1575 | err = ret; |
| 1576 | goto out; |
| 1577 | } |
| 1578 | leaf = path->nodes[0]; |
| 1579 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); |
| 1580 | } |
| 1581 | #endif |
| 1582 | BUG_ON(item_size < sizeof(*ei)); |
| 1583 | |
| 1584 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
| 1585 | flags = btrfs_extent_flags(leaf, ei); |
| 1586 | |
| 1587 | ptr = (unsigned long)(ei + 1); |
| 1588 | end = (unsigned long)ei + item_size; |
| 1589 | |
| 1590 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
| 1591 | ptr += sizeof(struct btrfs_tree_block_info); |
| 1592 | BUG_ON(ptr > end); |
| 1593 | } |
| 1594 | |
| 1595 | err = -ENOENT; |
| 1596 | while (1) { |
| 1597 | if (ptr >= end) { |
| 1598 | WARN_ON(ptr > end); |
| 1599 | break; |
| 1600 | } |
| 1601 | iref = (struct btrfs_extent_inline_ref *)ptr; |
| 1602 | type = btrfs_extent_inline_ref_type(leaf, iref); |
| 1603 | if (want < type) |
| 1604 | break; |
| 1605 | if (want > type) { |
| 1606 | ptr += btrfs_extent_inline_ref_size(type); |
| 1607 | continue; |
| 1608 | } |
| 1609 | |
| 1610 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { |
| 1611 | struct btrfs_extent_data_ref *dref; |
| 1612 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); |
| 1613 | if (match_extent_data_ref(leaf, dref, root_objectid, |
| 1614 | owner, offset)) { |
| 1615 | err = 0; |
| 1616 | break; |
| 1617 | } |
| 1618 | if (hash_extent_data_ref_item(leaf, dref) < |
| 1619 | hash_extent_data_ref(root_objectid, owner, offset)) |
| 1620 | break; |
| 1621 | } else { |
| 1622 | u64 ref_offset; |
| 1623 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); |
| 1624 | if (parent > 0) { |
| 1625 | if (parent == ref_offset) { |
| 1626 | err = 0; |
| 1627 | break; |
| 1628 | } |
| 1629 | if (ref_offset < parent) |
| 1630 | break; |
| 1631 | } else { |
| 1632 | if (root_objectid == ref_offset) { |
| 1633 | err = 0; |
| 1634 | break; |
| 1635 | } |
| 1636 | if (ref_offset < root_objectid) |
| 1637 | break; |
| 1638 | } |
| 1639 | } |
| 1640 | ptr += btrfs_extent_inline_ref_size(type); |
| 1641 | } |
| 1642 | if (err == -ENOENT && insert) { |
| 1643 | if (item_size + extra_size >= |
| 1644 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { |
| 1645 | err = -EAGAIN; |
| 1646 | goto out; |
| 1647 | } |
| 1648 | /* |
| 1649 | * To add new inline back ref, we have to make sure |
| 1650 | * there is no corresponding back ref item. |
| 1651 | * For simplicity, we just do not add new inline back |
| 1652 | * ref if there is any kind of item for this block |
| 1653 | */ |
| 1654 | if (find_next_key(path, 0, &key) == 0 && |
| 1655 | key.objectid == bytenr && |
| 1656 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
| 1657 | err = -EAGAIN; |
| 1658 | goto out; |
| 1659 | } |
| 1660 | } |
| 1661 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; |
| 1662 | out: |
| 1663 | if (insert) { |
| 1664 | path->keep_locks = 0; |
| 1665 | btrfs_unlock_up_safe(path, 1); |
| 1666 | } |
| 1667 | return err; |
| 1668 | } |
| 1669 | |
| 1670 | /* |
| 1671 | * helper to add new inline back ref |
| 1672 | */ |
| 1673 | static noinline_for_stack |
| 1674 | void setup_inline_extent_backref(struct btrfs_root *root, |
| 1675 | struct btrfs_path *path, |
| 1676 | struct btrfs_extent_inline_ref *iref, |
| 1677 | u64 parent, u64 root_objectid, |
| 1678 | u64 owner, u64 offset, int refs_to_add, |
| 1679 | struct btrfs_delayed_extent_op *extent_op) |
| 1680 | { |
| 1681 | struct extent_buffer *leaf; |
| 1682 | struct btrfs_extent_item *ei; |
| 1683 | unsigned long ptr; |
| 1684 | unsigned long end; |
| 1685 | unsigned long item_offset; |
| 1686 | u64 refs; |
| 1687 | int size; |
| 1688 | int type; |
| 1689 | |
| 1690 | leaf = path->nodes[0]; |
| 1691 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
| 1692 | item_offset = (unsigned long)iref - (unsigned long)ei; |
| 1693 | |
| 1694 | type = extent_ref_type(parent, owner); |
| 1695 | size = btrfs_extent_inline_ref_size(type); |
| 1696 | |
| 1697 | btrfs_extend_item(root, path, size); |
| 1698 | |
| 1699 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
| 1700 | refs = btrfs_extent_refs(leaf, ei); |
| 1701 | refs += refs_to_add; |
| 1702 | btrfs_set_extent_refs(leaf, ei, refs); |
| 1703 | if (extent_op) |
| 1704 | __run_delayed_extent_op(extent_op, leaf, ei); |
| 1705 | |
| 1706 | ptr = (unsigned long)ei + item_offset; |
| 1707 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); |
| 1708 | if (ptr < end - size) |
| 1709 | memmove_extent_buffer(leaf, ptr + size, ptr, |
| 1710 | end - size - ptr); |
| 1711 | |
| 1712 | iref = (struct btrfs_extent_inline_ref *)ptr; |
| 1713 | btrfs_set_extent_inline_ref_type(leaf, iref, type); |
| 1714 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { |
| 1715 | struct btrfs_extent_data_ref *dref; |
| 1716 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); |
| 1717 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); |
| 1718 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); |
| 1719 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); |
| 1720 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); |
| 1721 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { |
| 1722 | struct btrfs_shared_data_ref *sref; |
| 1723 | sref = (struct btrfs_shared_data_ref *)(iref + 1); |
| 1724 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); |
| 1725 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); |
| 1726 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
| 1727 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); |
| 1728 | } else { |
| 1729 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); |
| 1730 | } |
| 1731 | btrfs_mark_buffer_dirty(leaf); |
| 1732 | } |
| 1733 | |
| 1734 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, |
| 1735 | struct btrfs_root *root, |
| 1736 | struct btrfs_path *path, |
| 1737 | struct btrfs_extent_inline_ref **ref_ret, |
| 1738 | u64 bytenr, u64 num_bytes, u64 parent, |
| 1739 | u64 root_objectid, u64 owner, u64 offset) |
| 1740 | { |
| 1741 | int ret; |
| 1742 | |
| 1743 | ret = lookup_inline_extent_backref(trans, root, path, ref_ret, |
| 1744 | bytenr, num_bytes, parent, |
| 1745 | root_objectid, owner, offset, 0); |
| 1746 | if (ret != -ENOENT) |
| 1747 | return ret; |
| 1748 | |
| 1749 | btrfs_release_path(path); |
| 1750 | *ref_ret = NULL; |
| 1751 | |
| 1752 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
| 1753 | ret = lookup_tree_block_ref(trans, root, path, bytenr, parent, |
| 1754 | root_objectid); |
| 1755 | } else { |
| 1756 | ret = lookup_extent_data_ref(trans, root, path, bytenr, parent, |
| 1757 | root_objectid, owner, offset); |
| 1758 | } |
| 1759 | return ret; |
| 1760 | } |
| 1761 | |
| 1762 | /* |
| 1763 | * helper to update/remove inline back ref |
| 1764 | */ |
| 1765 | static noinline_for_stack |
| 1766 | void update_inline_extent_backref(struct btrfs_root *root, |
| 1767 | struct btrfs_path *path, |
| 1768 | struct btrfs_extent_inline_ref *iref, |
| 1769 | int refs_to_mod, |
| 1770 | struct btrfs_delayed_extent_op *extent_op, |
| 1771 | int *last_ref) |
| 1772 | { |
| 1773 | struct extent_buffer *leaf; |
| 1774 | struct btrfs_extent_item *ei; |
| 1775 | struct btrfs_extent_data_ref *dref = NULL; |
| 1776 | struct btrfs_shared_data_ref *sref = NULL; |
| 1777 | unsigned long ptr; |
| 1778 | unsigned long end; |
| 1779 | u32 item_size; |
| 1780 | int size; |
| 1781 | int type; |
| 1782 | u64 refs; |
| 1783 | |
| 1784 | leaf = path->nodes[0]; |
| 1785 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
| 1786 | refs = btrfs_extent_refs(leaf, ei); |
| 1787 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); |
| 1788 | refs += refs_to_mod; |
| 1789 | btrfs_set_extent_refs(leaf, ei, refs); |
| 1790 | if (extent_op) |
| 1791 | __run_delayed_extent_op(extent_op, leaf, ei); |
| 1792 | |
| 1793 | type = btrfs_extent_inline_ref_type(leaf, iref); |
| 1794 | |
| 1795 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { |
| 1796 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); |
| 1797 | refs = btrfs_extent_data_ref_count(leaf, dref); |
| 1798 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { |
| 1799 | sref = (struct btrfs_shared_data_ref *)(iref + 1); |
| 1800 | refs = btrfs_shared_data_ref_count(leaf, sref); |
| 1801 | } else { |
| 1802 | refs = 1; |
| 1803 | BUG_ON(refs_to_mod != -1); |
| 1804 | } |
| 1805 | |
| 1806 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
| 1807 | refs += refs_to_mod; |
| 1808 | |
| 1809 | if (refs > 0) { |
| 1810 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
| 1811 | btrfs_set_extent_data_ref_count(leaf, dref, refs); |
| 1812 | else |
| 1813 | btrfs_set_shared_data_ref_count(leaf, sref, refs); |
| 1814 | } else { |
| 1815 | *last_ref = 1; |
| 1816 | size = btrfs_extent_inline_ref_size(type); |
| 1817 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); |
| 1818 | ptr = (unsigned long)iref; |
| 1819 | end = (unsigned long)ei + item_size; |
| 1820 | if (ptr + size < end) |
| 1821 | memmove_extent_buffer(leaf, ptr, ptr + size, |
| 1822 | end - ptr - size); |
| 1823 | item_size -= size; |
| 1824 | btrfs_truncate_item(root, path, item_size, 1); |
| 1825 | } |
| 1826 | btrfs_mark_buffer_dirty(leaf); |
| 1827 | } |
| 1828 | |
| 1829 | static noinline_for_stack |
| 1830 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, |
| 1831 | struct btrfs_root *root, |
| 1832 | struct btrfs_path *path, |
| 1833 | u64 bytenr, u64 num_bytes, u64 parent, |
| 1834 | u64 root_objectid, u64 owner, |
| 1835 | u64 offset, int refs_to_add, |
| 1836 | struct btrfs_delayed_extent_op *extent_op) |
| 1837 | { |
| 1838 | struct btrfs_extent_inline_ref *iref; |
| 1839 | int ret; |
| 1840 | |
| 1841 | ret = lookup_inline_extent_backref(trans, root, path, &iref, |
| 1842 | bytenr, num_bytes, parent, |
| 1843 | root_objectid, owner, offset, 1); |
| 1844 | if (ret == 0) { |
| 1845 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); |
| 1846 | update_inline_extent_backref(root, path, iref, |
| 1847 | refs_to_add, extent_op, NULL); |
| 1848 | } else if (ret == -ENOENT) { |
| 1849 | setup_inline_extent_backref(root, path, iref, parent, |
| 1850 | root_objectid, owner, offset, |
| 1851 | refs_to_add, extent_op); |
| 1852 | ret = 0; |
| 1853 | } |
| 1854 | return ret; |
| 1855 | } |
| 1856 | |
| 1857 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
| 1858 | struct btrfs_root *root, |
| 1859 | struct btrfs_path *path, |
| 1860 | u64 bytenr, u64 parent, u64 root_objectid, |
| 1861 | u64 owner, u64 offset, int refs_to_add) |
| 1862 | { |
| 1863 | int ret; |
| 1864 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
| 1865 | BUG_ON(refs_to_add != 1); |
| 1866 | ret = insert_tree_block_ref(trans, root, path, bytenr, |
| 1867 | parent, root_objectid); |
| 1868 | } else { |
| 1869 | ret = insert_extent_data_ref(trans, root, path, bytenr, |
| 1870 | parent, root_objectid, |
| 1871 | owner, offset, refs_to_add); |
| 1872 | } |
| 1873 | return ret; |
| 1874 | } |
| 1875 | |
| 1876 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
| 1877 | struct btrfs_root *root, |
| 1878 | struct btrfs_path *path, |
| 1879 | struct btrfs_extent_inline_ref *iref, |
| 1880 | int refs_to_drop, int is_data, int *last_ref) |
| 1881 | { |
| 1882 | int ret = 0; |
| 1883 | |
| 1884 | BUG_ON(!is_data && refs_to_drop != 1); |
| 1885 | if (iref) { |
| 1886 | update_inline_extent_backref(root, path, iref, |
| 1887 | -refs_to_drop, NULL, last_ref); |
| 1888 | } else if (is_data) { |
| 1889 | ret = remove_extent_data_ref(trans, root, path, refs_to_drop, |
| 1890 | last_ref); |
| 1891 | } else { |
| 1892 | *last_ref = 1; |
| 1893 | ret = btrfs_del_item(trans, root, path); |
| 1894 | } |
| 1895 | return ret; |
| 1896 | } |
| 1897 | |
| 1898 | static int btrfs_issue_discard(struct block_device *bdev, |
| 1899 | u64 start, u64 len) |
| 1900 | { |
| 1901 | return blkdev_issue_discard(bdev, start >> 9, len >> 9, GFP_NOFS, 0); |
| 1902 | } |
| 1903 | |
| 1904 | static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, |
| 1905 | u64 num_bytes, u64 *actual_bytes) |
| 1906 | { |
| 1907 | int ret; |
| 1908 | u64 discarded_bytes = 0; |
| 1909 | struct btrfs_bio *bbio = NULL; |
| 1910 | |
| 1911 | |
| 1912 | /* Tell the block device(s) that the sectors can be discarded */ |
| 1913 | ret = btrfs_map_block(root->fs_info, REQ_DISCARD, |
| 1914 | bytenr, &num_bytes, &bbio, 0); |
| 1915 | /* Error condition is -ENOMEM */ |
| 1916 | if (!ret) { |
| 1917 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
| 1918 | int i; |
| 1919 | |
| 1920 | |
| 1921 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
| 1922 | if (!stripe->dev->can_discard) |
| 1923 | continue; |
| 1924 | |
| 1925 | ret = btrfs_issue_discard(stripe->dev->bdev, |
| 1926 | stripe->physical, |
| 1927 | stripe->length); |
| 1928 | if (!ret) |
| 1929 | discarded_bytes += stripe->length; |
| 1930 | else if (ret != -EOPNOTSUPP) |
| 1931 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
| 1932 | |
| 1933 | /* |
| 1934 | * Just in case we get back EOPNOTSUPP for some reason, |
| 1935 | * just ignore the return value so we don't screw up |
| 1936 | * people calling discard_extent. |
| 1937 | */ |
| 1938 | ret = 0; |
| 1939 | } |
| 1940 | kfree(bbio); |
| 1941 | } |
| 1942 | |
| 1943 | if (actual_bytes) |
| 1944 | *actual_bytes = discarded_bytes; |
| 1945 | |
| 1946 | |
| 1947 | if (ret == -EOPNOTSUPP) |
| 1948 | ret = 0; |
| 1949 | return ret; |
| 1950 | } |
| 1951 | |
| 1952 | /* Can return -ENOMEM */ |
| 1953 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
| 1954 | struct btrfs_root *root, |
| 1955 | u64 bytenr, u64 num_bytes, u64 parent, |
| 1956 | u64 root_objectid, u64 owner, u64 offset, |
| 1957 | int no_quota) |
| 1958 | { |
| 1959 | int ret; |
| 1960 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 1961 | |
| 1962 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
| 1963 | root_objectid == BTRFS_TREE_LOG_OBJECTID); |
| 1964 | |
| 1965 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
| 1966 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
| 1967 | num_bytes, |
| 1968 | parent, root_objectid, (int)owner, |
| 1969 | BTRFS_ADD_DELAYED_REF, NULL, no_quota); |
| 1970 | } else { |
| 1971 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
| 1972 | num_bytes, |
| 1973 | parent, root_objectid, owner, offset, |
| 1974 | BTRFS_ADD_DELAYED_REF, NULL, no_quota); |
| 1975 | } |
| 1976 | return ret; |
| 1977 | } |
| 1978 | |
| 1979 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
| 1980 | struct btrfs_root *root, |
| 1981 | u64 bytenr, u64 num_bytes, |
| 1982 | u64 parent, u64 root_objectid, |
| 1983 | u64 owner, u64 offset, int refs_to_add, |
| 1984 | int no_quota, |
| 1985 | struct btrfs_delayed_extent_op *extent_op) |
| 1986 | { |
| 1987 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 1988 | struct btrfs_path *path; |
| 1989 | struct extent_buffer *leaf; |
| 1990 | struct btrfs_extent_item *item; |
| 1991 | struct btrfs_key key; |
| 1992 | u64 refs; |
| 1993 | int ret; |
| 1994 | enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_ADD_EXCL; |
| 1995 | |
| 1996 | path = btrfs_alloc_path(); |
| 1997 | if (!path) |
| 1998 | return -ENOMEM; |
| 1999 | |
| 2000 | if (!is_fstree(root_objectid) || !root->fs_info->quota_enabled) |
| 2001 | no_quota = 1; |
| 2002 | |
| 2003 | path->reada = 1; |
| 2004 | path->leave_spinning = 1; |
| 2005 | /* this will setup the path even if it fails to insert the back ref */ |
| 2006 | ret = insert_inline_extent_backref(trans, fs_info->extent_root, path, |
| 2007 | bytenr, num_bytes, parent, |
| 2008 | root_objectid, owner, offset, |
| 2009 | refs_to_add, extent_op); |
| 2010 | if ((ret < 0 && ret != -EAGAIN) || (!ret && no_quota)) |
| 2011 | goto out; |
| 2012 | /* |
| 2013 | * Ok we were able to insert an inline extent and it appears to be a new |
| 2014 | * reference, deal with the qgroup accounting. |
| 2015 | */ |
| 2016 | if (!ret && !no_quota) { |
| 2017 | ASSERT(root->fs_info->quota_enabled); |
| 2018 | leaf = path->nodes[0]; |
| 2019 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
| 2020 | item = btrfs_item_ptr(leaf, path->slots[0], |
| 2021 | struct btrfs_extent_item); |
| 2022 | if (btrfs_extent_refs(leaf, item) > (u64)refs_to_add) |
| 2023 | type = BTRFS_QGROUP_OPER_ADD_SHARED; |
| 2024 | btrfs_release_path(path); |
| 2025 | |
| 2026 | ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid, |
| 2027 | bytenr, num_bytes, type, 0); |
| 2028 | goto out; |
| 2029 | } |
| 2030 | |
| 2031 | /* |
| 2032 | * Ok we had -EAGAIN which means we didn't have space to insert and |
| 2033 | * inline extent ref, so just update the reference count and add a |
| 2034 | * normal backref. |
| 2035 | */ |
| 2036 | leaf = path->nodes[0]; |
| 2037 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
| 2038 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
| 2039 | refs = btrfs_extent_refs(leaf, item); |
| 2040 | if (refs) |
| 2041 | type = BTRFS_QGROUP_OPER_ADD_SHARED; |
| 2042 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); |
| 2043 | if (extent_op) |
| 2044 | __run_delayed_extent_op(extent_op, leaf, item); |
| 2045 | |
| 2046 | btrfs_mark_buffer_dirty(leaf); |
| 2047 | btrfs_release_path(path); |
| 2048 | |
| 2049 | if (!no_quota) { |
| 2050 | ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid, |
| 2051 | bytenr, num_bytes, type, 0); |
| 2052 | if (ret) |
| 2053 | goto out; |
| 2054 | } |
| 2055 | |
| 2056 | path->reada = 1; |
| 2057 | path->leave_spinning = 1; |
| 2058 | /* now insert the actual backref */ |
| 2059 | ret = insert_extent_backref(trans, root->fs_info->extent_root, |
| 2060 | path, bytenr, parent, root_objectid, |
| 2061 | owner, offset, refs_to_add); |
| 2062 | if (ret) |
| 2063 | btrfs_abort_transaction(trans, root, ret); |
| 2064 | out: |
| 2065 | btrfs_free_path(path); |
| 2066 | return ret; |
| 2067 | } |
| 2068 | |
| 2069 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
| 2070 | struct btrfs_root *root, |
| 2071 | struct btrfs_delayed_ref_node *node, |
| 2072 | struct btrfs_delayed_extent_op *extent_op, |
| 2073 | int insert_reserved) |
| 2074 | { |
| 2075 | int ret = 0; |
| 2076 | struct btrfs_delayed_data_ref *ref; |
| 2077 | struct btrfs_key ins; |
| 2078 | u64 parent = 0; |
| 2079 | u64 ref_root = 0; |
| 2080 | u64 flags = 0; |
| 2081 | |
| 2082 | ins.objectid = node->bytenr; |
| 2083 | ins.offset = node->num_bytes; |
| 2084 | ins.type = BTRFS_EXTENT_ITEM_KEY; |
| 2085 | |
| 2086 | ref = btrfs_delayed_node_to_data_ref(node); |
| 2087 | trace_run_delayed_data_ref(node, ref, node->action); |
| 2088 | |
| 2089 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
| 2090 | parent = ref->parent; |
| 2091 | ref_root = ref->root; |
| 2092 | |
| 2093 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
| 2094 | if (extent_op) |
| 2095 | flags |= extent_op->flags_to_set; |
| 2096 | ret = alloc_reserved_file_extent(trans, root, |
| 2097 | parent, ref_root, flags, |
| 2098 | ref->objectid, ref->offset, |
| 2099 | &ins, node->ref_mod); |
| 2100 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
| 2101 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, |
| 2102 | node->num_bytes, parent, |
| 2103 | ref_root, ref->objectid, |
| 2104 | ref->offset, node->ref_mod, |
| 2105 | node->no_quota, extent_op); |
| 2106 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
| 2107 | ret = __btrfs_free_extent(trans, root, node->bytenr, |
| 2108 | node->num_bytes, parent, |
| 2109 | ref_root, ref->objectid, |
| 2110 | ref->offset, node->ref_mod, |
| 2111 | extent_op, node->no_quota); |
| 2112 | } else { |
| 2113 | BUG(); |
| 2114 | } |
| 2115 | return ret; |
| 2116 | } |
| 2117 | |
| 2118 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
| 2119 | struct extent_buffer *leaf, |
| 2120 | struct btrfs_extent_item *ei) |
| 2121 | { |
| 2122 | u64 flags = btrfs_extent_flags(leaf, ei); |
| 2123 | if (extent_op->update_flags) { |
| 2124 | flags |= extent_op->flags_to_set; |
| 2125 | btrfs_set_extent_flags(leaf, ei, flags); |
| 2126 | } |
| 2127 | |
| 2128 | if (extent_op->update_key) { |
| 2129 | struct btrfs_tree_block_info *bi; |
| 2130 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); |
| 2131 | bi = (struct btrfs_tree_block_info *)(ei + 1); |
| 2132 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); |
| 2133 | } |
| 2134 | } |
| 2135 | |
| 2136 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, |
| 2137 | struct btrfs_root *root, |
| 2138 | struct btrfs_delayed_ref_node *node, |
| 2139 | struct btrfs_delayed_extent_op *extent_op) |
| 2140 | { |
| 2141 | struct btrfs_key key; |
| 2142 | struct btrfs_path *path; |
| 2143 | struct btrfs_extent_item *ei; |
| 2144 | struct extent_buffer *leaf; |
| 2145 | u32 item_size; |
| 2146 | int ret; |
| 2147 | int err = 0; |
| 2148 | int metadata = !extent_op->is_data; |
| 2149 | |
| 2150 | if (trans->aborted) |
| 2151 | return 0; |
| 2152 | |
| 2153 | if (metadata && !btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) |
| 2154 | metadata = 0; |
| 2155 | |
| 2156 | path = btrfs_alloc_path(); |
| 2157 | if (!path) |
| 2158 | return -ENOMEM; |
| 2159 | |
| 2160 | key.objectid = node->bytenr; |
| 2161 | |
| 2162 | if (metadata) { |
| 2163 | key.type = BTRFS_METADATA_ITEM_KEY; |
| 2164 | key.offset = extent_op->level; |
| 2165 | } else { |
| 2166 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 2167 | key.offset = node->num_bytes; |
| 2168 | } |
| 2169 | |
| 2170 | again: |
| 2171 | path->reada = 1; |
| 2172 | path->leave_spinning = 1; |
| 2173 | ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, |
| 2174 | path, 0, 1); |
| 2175 | if (ret < 0) { |
| 2176 | err = ret; |
| 2177 | goto out; |
| 2178 | } |
| 2179 | if (ret > 0) { |
| 2180 | if (metadata) { |
| 2181 | if (path->slots[0] > 0) { |
| 2182 | path->slots[0]--; |
| 2183 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
| 2184 | path->slots[0]); |
| 2185 | if (key.objectid == node->bytenr && |
| 2186 | key.type == BTRFS_EXTENT_ITEM_KEY && |
| 2187 | key.offset == node->num_bytes) |
| 2188 | ret = 0; |
| 2189 | } |
| 2190 | if (ret > 0) { |
| 2191 | btrfs_release_path(path); |
| 2192 | metadata = 0; |
| 2193 | |
| 2194 | key.objectid = node->bytenr; |
| 2195 | key.offset = node->num_bytes; |
| 2196 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 2197 | goto again; |
| 2198 | } |
| 2199 | } else { |
| 2200 | err = -EIO; |
| 2201 | goto out; |
| 2202 | } |
| 2203 | } |
| 2204 | |
| 2205 | leaf = path->nodes[0]; |
| 2206 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); |
| 2207 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 2208 | if (item_size < sizeof(*ei)) { |
| 2209 | ret = convert_extent_item_v0(trans, root->fs_info->extent_root, |
| 2210 | path, (u64)-1, 0); |
| 2211 | if (ret < 0) { |
| 2212 | err = ret; |
| 2213 | goto out; |
| 2214 | } |
| 2215 | leaf = path->nodes[0]; |
| 2216 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); |
| 2217 | } |
| 2218 | #endif |
| 2219 | BUG_ON(item_size < sizeof(*ei)); |
| 2220 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
| 2221 | __run_delayed_extent_op(extent_op, leaf, ei); |
| 2222 | |
| 2223 | btrfs_mark_buffer_dirty(leaf); |
| 2224 | out: |
| 2225 | btrfs_free_path(path); |
| 2226 | return err; |
| 2227 | } |
| 2228 | |
| 2229 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
| 2230 | struct btrfs_root *root, |
| 2231 | struct btrfs_delayed_ref_node *node, |
| 2232 | struct btrfs_delayed_extent_op *extent_op, |
| 2233 | int insert_reserved) |
| 2234 | { |
| 2235 | int ret = 0; |
| 2236 | struct btrfs_delayed_tree_ref *ref; |
| 2237 | struct btrfs_key ins; |
| 2238 | u64 parent = 0; |
| 2239 | u64 ref_root = 0; |
| 2240 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
| 2241 | SKINNY_METADATA); |
| 2242 | |
| 2243 | ref = btrfs_delayed_node_to_tree_ref(node); |
| 2244 | trace_run_delayed_tree_ref(node, ref, node->action); |
| 2245 | |
| 2246 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
| 2247 | parent = ref->parent; |
| 2248 | ref_root = ref->root; |
| 2249 | |
| 2250 | ins.objectid = node->bytenr; |
| 2251 | if (skinny_metadata) { |
| 2252 | ins.offset = ref->level; |
| 2253 | ins.type = BTRFS_METADATA_ITEM_KEY; |
| 2254 | } else { |
| 2255 | ins.offset = node->num_bytes; |
| 2256 | ins.type = BTRFS_EXTENT_ITEM_KEY; |
| 2257 | } |
| 2258 | |
| 2259 | BUG_ON(node->ref_mod != 1); |
| 2260 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
| 2261 | BUG_ON(!extent_op || !extent_op->update_flags); |
| 2262 | ret = alloc_reserved_tree_block(trans, root, |
| 2263 | parent, ref_root, |
| 2264 | extent_op->flags_to_set, |
| 2265 | &extent_op->key, |
| 2266 | ref->level, &ins, |
| 2267 | node->no_quota); |
| 2268 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
| 2269 | ret = __btrfs_inc_extent_ref(trans, root, node->bytenr, |
| 2270 | node->num_bytes, parent, ref_root, |
| 2271 | ref->level, 0, 1, node->no_quota, |
| 2272 | extent_op); |
| 2273 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
| 2274 | ret = __btrfs_free_extent(trans, root, node->bytenr, |
| 2275 | node->num_bytes, parent, ref_root, |
| 2276 | ref->level, 0, 1, extent_op, |
| 2277 | node->no_quota); |
| 2278 | } else { |
| 2279 | BUG(); |
| 2280 | } |
| 2281 | return ret; |
| 2282 | } |
| 2283 | |
| 2284 | /* helper function to actually process a single delayed ref entry */ |
| 2285 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
| 2286 | struct btrfs_root *root, |
| 2287 | struct btrfs_delayed_ref_node *node, |
| 2288 | struct btrfs_delayed_extent_op *extent_op, |
| 2289 | int insert_reserved) |
| 2290 | { |
| 2291 | int ret = 0; |
| 2292 | |
| 2293 | if (trans->aborted) { |
| 2294 | if (insert_reserved) |
| 2295 | btrfs_pin_extent(root, node->bytenr, |
| 2296 | node->num_bytes, 1); |
| 2297 | return 0; |
| 2298 | } |
| 2299 | |
| 2300 | if (btrfs_delayed_ref_is_head(node)) { |
| 2301 | struct btrfs_delayed_ref_head *head; |
| 2302 | /* |
| 2303 | * we've hit the end of the chain and we were supposed |
| 2304 | * to insert this extent into the tree. But, it got |
| 2305 | * deleted before we ever needed to insert it, so all |
| 2306 | * we have to do is clean up the accounting |
| 2307 | */ |
| 2308 | BUG_ON(extent_op); |
| 2309 | head = btrfs_delayed_node_to_head(node); |
| 2310 | trace_run_delayed_ref_head(node, head, node->action); |
| 2311 | |
| 2312 | if (insert_reserved) { |
| 2313 | btrfs_pin_extent(root, node->bytenr, |
| 2314 | node->num_bytes, 1); |
| 2315 | if (head->is_data) { |
| 2316 | ret = btrfs_del_csums(trans, root, |
| 2317 | node->bytenr, |
| 2318 | node->num_bytes); |
| 2319 | } |
| 2320 | } |
| 2321 | return ret; |
| 2322 | } |
| 2323 | |
| 2324 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
| 2325 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
| 2326 | ret = run_delayed_tree_ref(trans, root, node, extent_op, |
| 2327 | insert_reserved); |
| 2328 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || |
| 2329 | node->type == BTRFS_SHARED_DATA_REF_KEY) |
| 2330 | ret = run_delayed_data_ref(trans, root, node, extent_op, |
| 2331 | insert_reserved); |
| 2332 | else |
| 2333 | BUG(); |
| 2334 | return ret; |
| 2335 | } |
| 2336 | |
| 2337 | static noinline struct btrfs_delayed_ref_node * |
| 2338 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
| 2339 | { |
| 2340 | struct rb_node *node; |
| 2341 | struct btrfs_delayed_ref_node *ref, *last = NULL;; |
| 2342 | |
| 2343 | /* |
| 2344 | * select delayed ref of type BTRFS_ADD_DELAYED_REF first. |
| 2345 | * this prevents ref count from going down to zero when |
| 2346 | * there still are pending delayed ref. |
| 2347 | */ |
| 2348 | node = rb_first(&head->ref_root); |
| 2349 | while (node) { |
| 2350 | ref = rb_entry(node, struct btrfs_delayed_ref_node, |
| 2351 | rb_node); |
| 2352 | if (ref->action == BTRFS_ADD_DELAYED_REF) |
| 2353 | return ref; |
| 2354 | else if (last == NULL) |
| 2355 | last = ref; |
| 2356 | node = rb_next(node); |
| 2357 | } |
| 2358 | return last; |
| 2359 | } |
| 2360 | |
| 2361 | /* |
| 2362 | * Returns 0 on success or if called with an already aborted transaction. |
| 2363 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. |
| 2364 | */ |
| 2365 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
| 2366 | struct btrfs_root *root, |
| 2367 | unsigned long nr) |
| 2368 | { |
| 2369 | struct btrfs_delayed_ref_root *delayed_refs; |
| 2370 | struct btrfs_delayed_ref_node *ref; |
| 2371 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
| 2372 | struct btrfs_delayed_extent_op *extent_op; |
| 2373 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 2374 | ktime_t start = ktime_get(); |
| 2375 | int ret; |
| 2376 | unsigned long count = 0; |
| 2377 | unsigned long actual_count = 0; |
| 2378 | int must_insert_reserved = 0; |
| 2379 | |
| 2380 | delayed_refs = &trans->transaction->delayed_refs; |
| 2381 | while (1) { |
| 2382 | if (!locked_ref) { |
| 2383 | if (count >= nr) |
| 2384 | break; |
| 2385 | |
| 2386 | spin_lock(&delayed_refs->lock); |
| 2387 | locked_ref = btrfs_select_ref_head(trans); |
| 2388 | if (!locked_ref) { |
| 2389 | spin_unlock(&delayed_refs->lock); |
| 2390 | break; |
| 2391 | } |
| 2392 | |
| 2393 | /* grab the lock that says we are going to process |
| 2394 | * all the refs for this head */ |
| 2395 | ret = btrfs_delayed_ref_lock(trans, locked_ref); |
| 2396 | spin_unlock(&delayed_refs->lock); |
| 2397 | /* |
| 2398 | * we may have dropped the spin lock to get the head |
| 2399 | * mutex lock, and that might have given someone else |
| 2400 | * time to free the head. If that's true, it has been |
| 2401 | * removed from our list and we can move on. |
| 2402 | */ |
| 2403 | if (ret == -EAGAIN) { |
| 2404 | locked_ref = NULL; |
| 2405 | count++; |
| 2406 | continue; |
| 2407 | } |
| 2408 | } |
| 2409 | |
| 2410 | /* |
| 2411 | * We need to try and merge add/drops of the same ref since we |
| 2412 | * can run into issues with relocate dropping the implicit ref |
| 2413 | * and then it being added back again before the drop can |
| 2414 | * finish. If we merged anything we need to re-loop so we can |
| 2415 | * get a good ref. |
| 2416 | */ |
| 2417 | spin_lock(&locked_ref->lock); |
| 2418 | btrfs_merge_delayed_refs(trans, fs_info, delayed_refs, |
| 2419 | locked_ref); |
| 2420 | |
| 2421 | /* |
| 2422 | * locked_ref is the head node, so we have to go one |
| 2423 | * node back for any delayed ref updates |
| 2424 | */ |
| 2425 | ref = select_delayed_ref(locked_ref); |
| 2426 | |
| 2427 | if (ref && ref->seq && |
| 2428 | btrfs_check_delayed_seq(fs_info, delayed_refs, ref->seq)) { |
| 2429 | spin_unlock(&locked_ref->lock); |
| 2430 | btrfs_delayed_ref_unlock(locked_ref); |
| 2431 | spin_lock(&delayed_refs->lock); |
| 2432 | locked_ref->processing = 0; |
| 2433 | delayed_refs->num_heads_ready++; |
| 2434 | spin_unlock(&delayed_refs->lock); |
| 2435 | locked_ref = NULL; |
| 2436 | cond_resched(); |
| 2437 | count++; |
| 2438 | continue; |
| 2439 | } |
| 2440 | |
| 2441 | /* |
| 2442 | * record the must insert reserved flag before we |
| 2443 | * drop the spin lock. |
| 2444 | */ |
| 2445 | must_insert_reserved = locked_ref->must_insert_reserved; |
| 2446 | locked_ref->must_insert_reserved = 0; |
| 2447 | |
| 2448 | extent_op = locked_ref->extent_op; |
| 2449 | locked_ref->extent_op = NULL; |
| 2450 | |
| 2451 | if (!ref) { |
| 2452 | |
| 2453 | |
| 2454 | /* All delayed refs have been processed, Go ahead |
| 2455 | * and send the head node to run_one_delayed_ref, |
| 2456 | * so that any accounting fixes can happen |
| 2457 | */ |
| 2458 | ref = &locked_ref->node; |
| 2459 | |
| 2460 | if (extent_op && must_insert_reserved) { |
| 2461 | btrfs_free_delayed_extent_op(extent_op); |
| 2462 | extent_op = NULL; |
| 2463 | } |
| 2464 | |
| 2465 | if (extent_op) { |
| 2466 | spin_unlock(&locked_ref->lock); |
| 2467 | ret = run_delayed_extent_op(trans, root, |
| 2468 | ref, extent_op); |
| 2469 | btrfs_free_delayed_extent_op(extent_op); |
| 2470 | |
| 2471 | if (ret) { |
| 2472 | /* |
| 2473 | * Need to reset must_insert_reserved if |
| 2474 | * there was an error so the abort stuff |
| 2475 | * can cleanup the reserved space |
| 2476 | * properly. |
| 2477 | */ |
| 2478 | if (must_insert_reserved) |
| 2479 | locked_ref->must_insert_reserved = 1; |
| 2480 | locked_ref->processing = 0; |
| 2481 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); |
| 2482 | btrfs_delayed_ref_unlock(locked_ref); |
| 2483 | return ret; |
| 2484 | } |
| 2485 | continue; |
| 2486 | } |
| 2487 | |
| 2488 | /* |
| 2489 | * Need to drop our head ref lock and re-aqcuire the |
| 2490 | * delayed ref lock and then re-check to make sure |
| 2491 | * nobody got added. |
| 2492 | */ |
| 2493 | spin_unlock(&locked_ref->lock); |
| 2494 | spin_lock(&delayed_refs->lock); |
| 2495 | spin_lock(&locked_ref->lock); |
| 2496 | if (rb_first(&locked_ref->ref_root) || |
| 2497 | locked_ref->extent_op) { |
| 2498 | spin_unlock(&locked_ref->lock); |
| 2499 | spin_unlock(&delayed_refs->lock); |
| 2500 | continue; |
| 2501 | } |
| 2502 | ref->in_tree = 0; |
| 2503 | delayed_refs->num_heads--; |
| 2504 | rb_erase(&locked_ref->href_node, |
| 2505 | &delayed_refs->href_root); |
| 2506 | spin_unlock(&delayed_refs->lock); |
| 2507 | } else { |
| 2508 | actual_count++; |
| 2509 | ref->in_tree = 0; |
| 2510 | rb_erase(&ref->rb_node, &locked_ref->ref_root); |
| 2511 | } |
| 2512 | atomic_dec(&delayed_refs->num_entries); |
| 2513 | |
| 2514 | if (!btrfs_delayed_ref_is_head(ref)) { |
| 2515 | /* |
| 2516 | * when we play the delayed ref, also correct the |
| 2517 | * ref_mod on head |
| 2518 | */ |
| 2519 | switch (ref->action) { |
| 2520 | case BTRFS_ADD_DELAYED_REF: |
| 2521 | case BTRFS_ADD_DELAYED_EXTENT: |
| 2522 | locked_ref->node.ref_mod -= ref->ref_mod; |
| 2523 | break; |
| 2524 | case BTRFS_DROP_DELAYED_REF: |
| 2525 | locked_ref->node.ref_mod += ref->ref_mod; |
| 2526 | break; |
| 2527 | default: |
| 2528 | WARN_ON(1); |
| 2529 | } |
| 2530 | } |
| 2531 | spin_unlock(&locked_ref->lock); |
| 2532 | |
| 2533 | ret = run_one_delayed_ref(trans, root, ref, extent_op, |
| 2534 | must_insert_reserved); |
| 2535 | |
| 2536 | btrfs_free_delayed_extent_op(extent_op); |
| 2537 | if (ret) { |
| 2538 | locked_ref->processing = 0; |
| 2539 | btrfs_delayed_ref_unlock(locked_ref); |
| 2540 | btrfs_put_delayed_ref(ref); |
| 2541 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", ret); |
| 2542 | return ret; |
| 2543 | } |
| 2544 | |
| 2545 | /* |
| 2546 | * If this node is a head, that means all the refs in this head |
| 2547 | * have been dealt with, and we will pick the next head to deal |
| 2548 | * with, so we must unlock the head and drop it from the cluster |
| 2549 | * list before we release it. |
| 2550 | */ |
| 2551 | if (btrfs_delayed_ref_is_head(ref)) { |
| 2552 | btrfs_delayed_ref_unlock(locked_ref); |
| 2553 | locked_ref = NULL; |
| 2554 | } |
| 2555 | btrfs_put_delayed_ref(ref); |
| 2556 | count++; |
| 2557 | cond_resched(); |
| 2558 | } |
| 2559 | |
| 2560 | /* |
| 2561 | * We don't want to include ref heads since we can have empty ref heads |
| 2562 | * and those will drastically skew our runtime down since we just do |
| 2563 | * accounting, no actual extent tree updates. |
| 2564 | */ |
| 2565 | if (actual_count > 0) { |
| 2566 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); |
| 2567 | u64 avg; |
| 2568 | |
| 2569 | /* |
| 2570 | * We weigh the current average higher than our current runtime |
| 2571 | * to avoid large swings in the average. |
| 2572 | */ |
| 2573 | spin_lock(&delayed_refs->lock); |
| 2574 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; |
| 2575 | avg = div64_u64(avg, 4); |
| 2576 | fs_info->avg_delayed_ref_runtime = avg; |
| 2577 | spin_unlock(&delayed_refs->lock); |
| 2578 | } |
| 2579 | return 0; |
| 2580 | } |
| 2581 | |
| 2582 | #ifdef SCRAMBLE_DELAYED_REFS |
| 2583 | /* |
| 2584 | * Normally delayed refs get processed in ascending bytenr order. This |
| 2585 | * correlates in most cases to the order added. To expose dependencies on this |
| 2586 | * order, we start to process the tree in the middle instead of the beginning |
| 2587 | */ |
| 2588 | static u64 find_middle(struct rb_root *root) |
| 2589 | { |
| 2590 | struct rb_node *n = root->rb_node; |
| 2591 | struct btrfs_delayed_ref_node *entry; |
| 2592 | int alt = 1; |
| 2593 | u64 middle; |
| 2594 | u64 first = 0, last = 0; |
| 2595 | |
| 2596 | n = rb_first(root); |
| 2597 | if (n) { |
| 2598 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); |
| 2599 | first = entry->bytenr; |
| 2600 | } |
| 2601 | n = rb_last(root); |
| 2602 | if (n) { |
| 2603 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); |
| 2604 | last = entry->bytenr; |
| 2605 | } |
| 2606 | n = root->rb_node; |
| 2607 | |
| 2608 | while (n) { |
| 2609 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); |
| 2610 | WARN_ON(!entry->in_tree); |
| 2611 | |
| 2612 | middle = entry->bytenr; |
| 2613 | |
| 2614 | if (alt) |
| 2615 | n = n->rb_left; |
| 2616 | else |
| 2617 | n = n->rb_right; |
| 2618 | |
| 2619 | alt = 1 - alt; |
| 2620 | } |
| 2621 | return middle; |
| 2622 | } |
| 2623 | #endif |
| 2624 | |
| 2625 | static inline u64 heads_to_leaves(struct btrfs_root *root, u64 heads) |
| 2626 | { |
| 2627 | u64 num_bytes; |
| 2628 | |
| 2629 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + |
| 2630 | sizeof(struct btrfs_extent_inline_ref)); |
| 2631 | if (!btrfs_fs_incompat(root->fs_info, SKINNY_METADATA)) |
| 2632 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
| 2633 | |
| 2634 | /* |
| 2635 | * We don't ever fill up leaves all the way so multiply by 2 just to be |
| 2636 | * closer to what we're really going to want to ouse. |
| 2637 | */ |
| 2638 | return div64_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(root)); |
| 2639 | } |
| 2640 | |
| 2641 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
| 2642 | struct btrfs_root *root) |
| 2643 | { |
| 2644 | struct btrfs_block_rsv *global_rsv; |
| 2645 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; |
| 2646 | u64 num_bytes; |
| 2647 | int ret = 0; |
| 2648 | |
| 2649 | num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
| 2650 | num_heads = heads_to_leaves(root, num_heads); |
| 2651 | if (num_heads > 1) |
| 2652 | num_bytes += (num_heads - 1) * root->leafsize; |
| 2653 | num_bytes <<= 1; |
| 2654 | global_rsv = &root->fs_info->global_block_rsv; |
| 2655 | |
| 2656 | /* |
| 2657 | * If we can't allocate any more chunks lets make sure we have _lots_ of |
| 2658 | * wiggle room since running delayed refs can create more delayed refs. |
| 2659 | */ |
| 2660 | if (global_rsv->space_info->full) |
| 2661 | num_bytes <<= 1; |
| 2662 | |
| 2663 | spin_lock(&global_rsv->lock); |
| 2664 | if (global_rsv->reserved <= num_bytes) |
| 2665 | ret = 1; |
| 2666 | spin_unlock(&global_rsv->lock); |
| 2667 | return ret; |
| 2668 | } |
| 2669 | |
| 2670 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
| 2671 | struct btrfs_root *root) |
| 2672 | { |
| 2673 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 2674 | u64 num_entries = |
| 2675 | atomic_read(&trans->transaction->delayed_refs.num_entries); |
| 2676 | u64 avg_runtime; |
| 2677 | |
| 2678 | smp_mb(); |
| 2679 | avg_runtime = fs_info->avg_delayed_ref_runtime; |
| 2680 | if (num_entries * avg_runtime >= NSEC_PER_SEC) |
| 2681 | return 1; |
| 2682 | |
| 2683 | return btrfs_check_space_for_delayed_refs(trans, root); |
| 2684 | } |
| 2685 | |
| 2686 | /* |
| 2687 | * this starts processing the delayed reference count updates and |
| 2688 | * extent insertions we have queued up so far. count can be |
| 2689 | * 0, which means to process everything in the tree at the start |
| 2690 | * of the run (but not newly added entries), or it can be some target |
| 2691 | * number you'd like to process. |
| 2692 | * |
| 2693 | * Returns 0 on success or if called with an aborted transaction |
| 2694 | * Returns <0 on error and aborts the transaction |
| 2695 | */ |
| 2696 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
| 2697 | struct btrfs_root *root, unsigned long count) |
| 2698 | { |
| 2699 | struct rb_node *node; |
| 2700 | struct btrfs_delayed_ref_root *delayed_refs; |
| 2701 | struct btrfs_delayed_ref_head *head; |
| 2702 | int ret; |
| 2703 | int run_all = count == (unsigned long)-1; |
| 2704 | int run_most = 0; |
| 2705 | |
| 2706 | /* We'll clean this up in btrfs_cleanup_transaction */ |
| 2707 | if (trans->aborted) |
| 2708 | return 0; |
| 2709 | |
| 2710 | if (root == root->fs_info->extent_root) |
| 2711 | root = root->fs_info->tree_root; |
| 2712 | |
| 2713 | delayed_refs = &trans->transaction->delayed_refs; |
| 2714 | if (count == 0) { |
| 2715 | count = atomic_read(&delayed_refs->num_entries) * 2; |
| 2716 | run_most = 1; |
| 2717 | } |
| 2718 | |
| 2719 | again: |
| 2720 | #ifdef SCRAMBLE_DELAYED_REFS |
| 2721 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); |
| 2722 | #endif |
| 2723 | ret = __btrfs_run_delayed_refs(trans, root, count); |
| 2724 | if (ret < 0) { |
| 2725 | btrfs_abort_transaction(trans, root, ret); |
| 2726 | return ret; |
| 2727 | } |
| 2728 | |
| 2729 | if (run_all) { |
| 2730 | if (!list_empty(&trans->new_bgs)) |
| 2731 | btrfs_create_pending_block_groups(trans, root); |
| 2732 | |
| 2733 | spin_lock(&delayed_refs->lock); |
| 2734 | node = rb_first(&delayed_refs->href_root); |
| 2735 | if (!node) { |
| 2736 | spin_unlock(&delayed_refs->lock); |
| 2737 | goto out; |
| 2738 | } |
| 2739 | count = (unsigned long)-1; |
| 2740 | |
| 2741 | while (node) { |
| 2742 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
| 2743 | href_node); |
| 2744 | if (btrfs_delayed_ref_is_head(&head->node)) { |
| 2745 | struct btrfs_delayed_ref_node *ref; |
| 2746 | |
| 2747 | ref = &head->node; |
| 2748 | atomic_inc(&ref->refs); |
| 2749 | |
| 2750 | spin_unlock(&delayed_refs->lock); |
| 2751 | /* |
| 2752 | * Mutex was contended, block until it's |
| 2753 | * released and try again |
| 2754 | */ |
| 2755 | mutex_lock(&head->mutex); |
| 2756 | mutex_unlock(&head->mutex); |
| 2757 | |
| 2758 | btrfs_put_delayed_ref(ref); |
| 2759 | cond_resched(); |
| 2760 | goto again; |
| 2761 | } else { |
| 2762 | WARN_ON(1); |
| 2763 | } |
| 2764 | node = rb_next(node); |
| 2765 | } |
| 2766 | spin_unlock(&delayed_refs->lock); |
| 2767 | cond_resched(); |
| 2768 | goto again; |
| 2769 | } |
| 2770 | out: |
| 2771 | ret = btrfs_delayed_qgroup_accounting(trans, root->fs_info); |
| 2772 | if (ret) |
| 2773 | return ret; |
| 2774 | assert_qgroups_uptodate(trans); |
| 2775 | return 0; |
| 2776 | } |
| 2777 | |
| 2778 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
| 2779 | struct btrfs_root *root, |
| 2780 | u64 bytenr, u64 num_bytes, u64 flags, |
| 2781 | int level, int is_data) |
| 2782 | { |
| 2783 | struct btrfs_delayed_extent_op *extent_op; |
| 2784 | int ret; |
| 2785 | |
| 2786 | extent_op = btrfs_alloc_delayed_extent_op(); |
| 2787 | if (!extent_op) |
| 2788 | return -ENOMEM; |
| 2789 | |
| 2790 | extent_op->flags_to_set = flags; |
| 2791 | extent_op->update_flags = 1; |
| 2792 | extent_op->update_key = 0; |
| 2793 | extent_op->is_data = is_data ? 1 : 0; |
| 2794 | extent_op->level = level; |
| 2795 | |
| 2796 | ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr, |
| 2797 | num_bytes, extent_op); |
| 2798 | if (ret) |
| 2799 | btrfs_free_delayed_extent_op(extent_op); |
| 2800 | return ret; |
| 2801 | } |
| 2802 | |
| 2803 | static noinline int check_delayed_ref(struct btrfs_trans_handle *trans, |
| 2804 | struct btrfs_root *root, |
| 2805 | struct btrfs_path *path, |
| 2806 | u64 objectid, u64 offset, u64 bytenr) |
| 2807 | { |
| 2808 | struct btrfs_delayed_ref_head *head; |
| 2809 | struct btrfs_delayed_ref_node *ref; |
| 2810 | struct btrfs_delayed_data_ref *data_ref; |
| 2811 | struct btrfs_delayed_ref_root *delayed_refs; |
| 2812 | struct rb_node *node; |
| 2813 | int ret = 0; |
| 2814 | |
| 2815 | delayed_refs = &trans->transaction->delayed_refs; |
| 2816 | spin_lock(&delayed_refs->lock); |
| 2817 | head = btrfs_find_delayed_ref_head(trans, bytenr); |
| 2818 | if (!head) { |
| 2819 | spin_unlock(&delayed_refs->lock); |
| 2820 | return 0; |
| 2821 | } |
| 2822 | |
| 2823 | if (!mutex_trylock(&head->mutex)) { |
| 2824 | atomic_inc(&head->node.refs); |
| 2825 | spin_unlock(&delayed_refs->lock); |
| 2826 | |
| 2827 | btrfs_release_path(path); |
| 2828 | |
| 2829 | /* |
| 2830 | * Mutex was contended, block until it's released and let |
| 2831 | * caller try again |
| 2832 | */ |
| 2833 | mutex_lock(&head->mutex); |
| 2834 | mutex_unlock(&head->mutex); |
| 2835 | btrfs_put_delayed_ref(&head->node); |
| 2836 | return -EAGAIN; |
| 2837 | } |
| 2838 | spin_unlock(&delayed_refs->lock); |
| 2839 | |
| 2840 | spin_lock(&head->lock); |
| 2841 | node = rb_first(&head->ref_root); |
| 2842 | while (node) { |
| 2843 | ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); |
| 2844 | node = rb_next(node); |
| 2845 | |
| 2846 | /* If it's a shared ref we know a cross reference exists */ |
| 2847 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { |
| 2848 | ret = 1; |
| 2849 | break; |
| 2850 | } |
| 2851 | |
| 2852 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
| 2853 | |
| 2854 | /* |
| 2855 | * If our ref doesn't match the one we're currently looking at |
| 2856 | * then we have a cross reference. |
| 2857 | */ |
| 2858 | if (data_ref->root != root->root_key.objectid || |
| 2859 | data_ref->objectid != objectid || |
| 2860 | data_ref->offset != offset) { |
| 2861 | ret = 1; |
| 2862 | break; |
| 2863 | } |
| 2864 | } |
| 2865 | spin_unlock(&head->lock); |
| 2866 | mutex_unlock(&head->mutex); |
| 2867 | return ret; |
| 2868 | } |
| 2869 | |
| 2870 | static noinline int check_committed_ref(struct btrfs_trans_handle *trans, |
| 2871 | struct btrfs_root *root, |
| 2872 | struct btrfs_path *path, |
| 2873 | u64 objectid, u64 offset, u64 bytenr) |
| 2874 | { |
| 2875 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
| 2876 | struct extent_buffer *leaf; |
| 2877 | struct btrfs_extent_data_ref *ref; |
| 2878 | struct btrfs_extent_inline_ref *iref; |
| 2879 | struct btrfs_extent_item *ei; |
| 2880 | struct btrfs_key key; |
| 2881 | u32 item_size; |
| 2882 | int ret; |
| 2883 | |
| 2884 | key.objectid = bytenr; |
| 2885 | key.offset = (u64)-1; |
| 2886 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 2887 | |
| 2888 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
| 2889 | if (ret < 0) |
| 2890 | goto out; |
| 2891 | BUG_ON(ret == 0); /* Corruption */ |
| 2892 | |
| 2893 | ret = -ENOENT; |
| 2894 | if (path->slots[0] == 0) |
| 2895 | goto out; |
| 2896 | |
| 2897 | path->slots[0]--; |
| 2898 | leaf = path->nodes[0]; |
| 2899 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
| 2900 | |
| 2901 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
| 2902 | goto out; |
| 2903 | |
| 2904 | ret = 1; |
| 2905 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); |
| 2906 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 2907 | if (item_size < sizeof(*ei)) { |
| 2908 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); |
| 2909 | goto out; |
| 2910 | } |
| 2911 | #endif |
| 2912 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
| 2913 | |
| 2914 | if (item_size != sizeof(*ei) + |
| 2915 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) |
| 2916 | goto out; |
| 2917 | |
| 2918 | if (btrfs_extent_generation(leaf, ei) <= |
| 2919 | btrfs_root_last_snapshot(&root->root_item)) |
| 2920 | goto out; |
| 2921 | |
| 2922 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); |
| 2923 | if (btrfs_extent_inline_ref_type(leaf, iref) != |
| 2924 | BTRFS_EXTENT_DATA_REF_KEY) |
| 2925 | goto out; |
| 2926 | |
| 2927 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); |
| 2928 | if (btrfs_extent_refs(leaf, ei) != |
| 2929 | btrfs_extent_data_ref_count(leaf, ref) || |
| 2930 | btrfs_extent_data_ref_root(leaf, ref) != |
| 2931 | root->root_key.objectid || |
| 2932 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || |
| 2933 | btrfs_extent_data_ref_offset(leaf, ref) != offset) |
| 2934 | goto out; |
| 2935 | |
| 2936 | ret = 0; |
| 2937 | out: |
| 2938 | return ret; |
| 2939 | } |
| 2940 | |
| 2941 | int btrfs_cross_ref_exist(struct btrfs_trans_handle *trans, |
| 2942 | struct btrfs_root *root, |
| 2943 | u64 objectid, u64 offset, u64 bytenr) |
| 2944 | { |
| 2945 | struct btrfs_path *path; |
| 2946 | int ret; |
| 2947 | int ret2; |
| 2948 | |
| 2949 | path = btrfs_alloc_path(); |
| 2950 | if (!path) |
| 2951 | return -ENOENT; |
| 2952 | |
| 2953 | do { |
| 2954 | ret = check_committed_ref(trans, root, path, objectid, |
| 2955 | offset, bytenr); |
| 2956 | if (ret && ret != -ENOENT) |
| 2957 | goto out; |
| 2958 | |
| 2959 | ret2 = check_delayed_ref(trans, root, path, objectid, |
| 2960 | offset, bytenr); |
| 2961 | } while (ret2 == -EAGAIN); |
| 2962 | |
| 2963 | if (ret2 && ret2 != -ENOENT) { |
| 2964 | ret = ret2; |
| 2965 | goto out; |
| 2966 | } |
| 2967 | |
| 2968 | if (ret != -ENOENT || ret2 != -ENOENT) |
| 2969 | ret = 0; |
| 2970 | out: |
| 2971 | btrfs_free_path(path); |
| 2972 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
| 2973 | WARN_ON(ret > 0); |
| 2974 | return ret; |
| 2975 | } |
| 2976 | |
| 2977 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
| 2978 | struct btrfs_root *root, |
| 2979 | struct extent_buffer *buf, |
| 2980 | int full_backref, int inc, int no_quota) |
| 2981 | { |
| 2982 | u64 bytenr; |
| 2983 | u64 num_bytes; |
| 2984 | u64 parent; |
| 2985 | u64 ref_root; |
| 2986 | u32 nritems; |
| 2987 | struct btrfs_key key; |
| 2988 | struct btrfs_file_extent_item *fi; |
| 2989 | int i; |
| 2990 | int level; |
| 2991 | int ret = 0; |
| 2992 | int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, |
| 2993 | u64, u64, u64, u64, u64, u64, int); |
| 2994 | |
| 2995 | ref_root = btrfs_header_owner(buf); |
| 2996 | nritems = btrfs_header_nritems(buf); |
| 2997 | level = btrfs_header_level(buf); |
| 2998 | |
| 2999 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
| 3000 | return 0; |
| 3001 | |
| 3002 | if (inc) |
| 3003 | process_func = btrfs_inc_extent_ref; |
| 3004 | else |
| 3005 | process_func = btrfs_free_extent; |
| 3006 | |
| 3007 | if (full_backref) |
| 3008 | parent = buf->start; |
| 3009 | else |
| 3010 | parent = 0; |
| 3011 | |
| 3012 | for (i = 0; i < nritems; i++) { |
| 3013 | if (level == 0) { |
| 3014 | btrfs_item_key_to_cpu(buf, &key, i); |
| 3015 | if (btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY) |
| 3016 | continue; |
| 3017 | fi = btrfs_item_ptr(buf, i, |
| 3018 | struct btrfs_file_extent_item); |
| 3019 | if (btrfs_file_extent_type(buf, fi) == |
| 3020 | BTRFS_FILE_EXTENT_INLINE) |
| 3021 | continue; |
| 3022 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
| 3023 | if (bytenr == 0) |
| 3024 | continue; |
| 3025 | |
| 3026 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); |
| 3027 | key.offset -= btrfs_file_extent_offset(buf, fi); |
| 3028 | ret = process_func(trans, root, bytenr, num_bytes, |
| 3029 | parent, ref_root, key.objectid, |
| 3030 | key.offset, no_quota); |
| 3031 | if (ret) |
| 3032 | goto fail; |
| 3033 | } else { |
| 3034 | bytenr = btrfs_node_blockptr(buf, i); |
| 3035 | num_bytes = btrfs_level_size(root, level - 1); |
| 3036 | ret = process_func(trans, root, bytenr, num_bytes, |
| 3037 | parent, ref_root, level - 1, 0, |
| 3038 | no_quota); |
| 3039 | if (ret) |
| 3040 | goto fail; |
| 3041 | } |
| 3042 | } |
| 3043 | return 0; |
| 3044 | fail: |
| 3045 | return ret; |
| 3046 | } |
| 3047 | |
| 3048 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
| 3049 | struct extent_buffer *buf, int full_backref, int no_quota) |
| 3050 | { |
| 3051 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1, no_quota); |
| 3052 | } |
| 3053 | |
| 3054 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
| 3055 | struct extent_buffer *buf, int full_backref, int no_quota) |
| 3056 | { |
| 3057 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0, no_quota); |
| 3058 | } |
| 3059 | |
| 3060 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
| 3061 | struct btrfs_root *root, |
| 3062 | struct btrfs_path *path, |
| 3063 | struct btrfs_block_group_cache *cache) |
| 3064 | { |
| 3065 | int ret; |
| 3066 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
| 3067 | unsigned long bi; |
| 3068 | struct extent_buffer *leaf; |
| 3069 | |
| 3070 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
| 3071 | if (ret < 0) |
| 3072 | goto fail; |
| 3073 | BUG_ON(ret); /* Corruption */ |
| 3074 | |
| 3075 | leaf = path->nodes[0]; |
| 3076 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); |
| 3077 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); |
| 3078 | btrfs_mark_buffer_dirty(leaf); |
| 3079 | btrfs_release_path(path); |
| 3080 | fail: |
| 3081 | if (ret) { |
| 3082 | btrfs_abort_transaction(trans, root, ret); |
| 3083 | return ret; |
| 3084 | } |
| 3085 | return 0; |
| 3086 | |
| 3087 | } |
| 3088 | |
| 3089 | static struct btrfs_block_group_cache * |
| 3090 | next_block_group(struct btrfs_root *root, |
| 3091 | struct btrfs_block_group_cache *cache) |
| 3092 | { |
| 3093 | struct rb_node *node; |
| 3094 | spin_lock(&root->fs_info->block_group_cache_lock); |
| 3095 | node = rb_next(&cache->cache_node); |
| 3096 | btrfs_put_block_group(cache); |
| 3097 | if (node) { |
| 3098 | cache = rb_entry(node, struct btrfs_block_group_cache, |
| 3099 | cache_node); |
| 3100 | btrfs_get_block_group(cache); |
| 3101 | } else |
| 3102 | cache = NULL; |
| 3103 | spin_unlock(&root->fs_info->block_group_cache_lock); |
| 3104 | return cache; |
| 3105 | } |
| 3106 | |
| 3107 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
| 3108 | struct btrfs_trans_handle *trans, |
| 3109 | struct btrfs_path *path) |
| 3110 | { |
| 3111 | struct btrfs_root *root = block_group->fs_info->tree_root; |
| 3112 | struct inode *inode = NULL; |
| 3113 | u64 alloc_hint = 0; |
| 3114 | int dcs = BTRFS_DC_ERROR; |
| 3115 | int num_pages = 0; |
| 3116 | int retries = 0; |
| 3117 | int ret = 0; |
| 3118 | |
| 3119 | /* |
| 3120 | * If this block group is smaller than 100 megs don't bother caching the |
| 3121 | * block group. |
| 3122 | */ |
| 3123 | if (block_group->key.offset < (100 * 1024 * 1024)) { |
| 3124 | spin_lock(&block_group->lock); |
| 3125 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; |
| 3126 | spin_unlock(&block_group->lock); |
| 3127 | return 0; |
| 3128 | } |
| 3129 | |
| 3130 | again: |
| 3131 | inode = lookup_free_space_inode(root, block_group, path); |
| 3132 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
| 3133 | ret = PTR_ERR(inode); |
| 3134 | btrfs_release_path(path); |
| 3135 | goto out; |
| 3136 | } |
| 3137 | |
| 3138 | if (IS_ERR(inode)) { |
| 3139 | BUG_ON(retries); |
| 3140 | retries++; |
| 3141 | |
| 3142 | if (block_group->ro) |
| 3143 | goto out_free; |
| 3144 | |
| 3145 | ret = create_free_space_inode(root, trans, block_group, path); |
| 3146 | if (ret) |
| 3147 | goto out_free; |
| 3148 | goto again; |
| 3149 | } |
| 3150 | |
| 3151 | /* We've already setup this transaction, go ahead and exit */ |
| 3152 | if (block_group->cache_generation == trans->transid && |
| 3153 | i_size_read(inode)) { |
| 3154 | dcs = BTRFS_DC_SETUP; |
| 3155 | goto out_put; |
| 3156 | } |
| 3157 | |
| 3158 | /* |
| 3159 | * We want to set the generation to 0, that way if anything goes wrong |
| 3160 | * from here on out we know not to trust this cache when we load up next |
| 3161 | * time. |
| 3162 | */ |
| 3163 | BTRFS_I(inode)->generation = 0; |
| 3164 | ret = btrfs_update_inode(trans, root, inode); |
| 3165 | WARN_ON(ret); |
| 3166 | |
| 3167 | if (i_size_read(inode) > 0) { |
| 3168 | ret = btrfs_check_trunc_cache_free_space(root, |
| 3169 | &root->fs_info->global_block_rsv); |
| 3170 | if (ret) |
| 3171 | goto out_put; |
| 3172 | |
| 3173 | ret = btrfs_truncate_free_space_cache(root, trans, inode); |
| 3174 | if (ret) |
| 3175 | goto out_put; |
| 3176 | } |
| 3177 | |
| 3178 | spin_lock(&block_group->lock); |
| 3179 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
| 3180 | !btrfs_test_opt(root, SPACE_CACHE)) { |
| 3181 | /* |
| 3182 | * don't bother trying to write stuff out _if_ |
| 3183 | * a) we're not cached, |
| 3184 | * b) we're with nospace_cache mount option. |
| 3185 | */ |
| 3186 | dcs = BTRFS_DC_WRITTEN; |
| 3187 | spin_unlock(&block_group->lock); |
| 3188 | goto out_put; |
| 3189 | } |
| 3190 | spin_unlock(&block_group->lock); |
| 3191 | |
| 3192 | /* |
| 3193 | * Try to preallocate enough space based on how big the block group is. |
| 3194 | * Keep in mind this has to include any pinned space which could end up |
| 3195 | * taking up quite a bit since it's not folded into the other space |
| 3196 | * cache. |
| 3197 | */ |
| 3198 | num_pages = (int)div64_u64(block_group->key.offset, 256 * 1024 * 1024); |
| 3199 | if (!num_pages) |
| 3200 | num_pages = 1; |
| 3201 | |
| 3202 | num_pages *= 16; |
| 3203 | num_pages *= PAGE_CACHE_SIZE; |
| 3204 | |
| 3205 | ret = btrfs_check_data_free_space(inode, num_pages); |
| 3206 | if (ret) |
| 3207 | goto out_put; |
| 3208 | |
| 3209 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, |
| 3210 | num_pages, num_pages, |
| 3211 | &alloc_hint); |
| 3212 | if (!ret) |
| 3213 | dcs = BTRFS_DC_SETUP; |
| 3214 | btrfs_free_reserved_data_space(inode, num_pages); |
| 3215 | |
| 3216 | out_put: |
| 3217 | iput(inode); |
| 3218 | out_free: |
| 3219 | btrfs_release_path(path); |
| 3220 | out: |
| 3221 | spin_lock(&block_group->lock); |
| 3222 | if (!ret && dcs == BTRFS_DC_SETUP) |
| 3223 | block_group->cache_generation = trans->transid; |
| 3224 | block_group->disk_cache_state = dcs; |
| 3225 | spin_unlock(&block_group->lock); |
| 3226 | |
| 3227 | return ret; |
| 3228 | } |
| 3229 | |
| 3230 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
| 3231 | struct btrfs_root *root) |
| 3232 | { |
| 3233 | struct btrfs_block_group_cache *cache; |
| 3234 | int err = 0; |
| 3235 | struct btrfs_path *path; |
| 3236 | u64 last = 0; |
| 3237 | |
| 3238 | path = btrfs_alloc_path(); |
| 3239 | if (!path) |
| 3240 | return -ENOMEM; |
| 3241 | |
| 3242 | again: |
| 3243 | while (1) { |
| 3244 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
| 3245 | while (cache) { |
| 3246 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) |
| 3247 | break; |
| 3248 | cache = next_block_group(root, cache); |
| 3249 | } |
| 3250 | if (!cache) { |
| 3251 | if (last == 0) |
| 3252 | break; |
| 3253 | last = 0; |
| 3254 | continue; |
| 3255 | } |
| 3256 | err = cache_save_setup(cache, trans, path); |
| 3257 | last = cache->key.objectid + cache->key.offset; |
| 3258 | btrfs_put_block_group(cache); |
| 3259 | } |
| 3260 | |
| 3261 | while (1) { |
| 3262 | if (last == 0) { |
| 3263 | err = btrfs_run_delayed_refs(trans, root, |
| 3264 | (unsigned long)-1); |
| 3265 | if (err) /* File system offline */ |
| 3266 | goto out; |
| 3267 | } |
| 3268 | |
| 3269 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
| 3270 | while (cache) { |
| 3271 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) { |
| 3272 | btrfs_put_block_group(cache); |
| 3273 | goto again; |
| 3274 | } |
| 3275 | |
| 3276 | if (cache->dirty) |
| 3277 | break; |
| 3278 | cache = next_block_group(root, cache); |
| 3279 | } |
| 3280 | if (!cache) { |
| 3281 | if (last == 0) |
| 3282 | break; |
| 3283 | last = 0; |
| 3284 | continue; |
| 3285 | } |
| 3286 | |
| 3287 | if (cache->disk_cache_state == BTRFS_DC_SETUP) |
| 3288 | cache->disk_cache_state = BTRFS_DC_NEED_WRITE; |
| 3289 | cache->dirty = 0; |
| 3290 | last = cache->key.objectid + cache->key.offset; |
| 3291 | |
| 3292 | err = write_one_cache_group(trans, root, path, cache); |
| 3293 | btrfs_put_block_group(cache); |
| 3294 | if (err) /* File system offline */ |
| 3295 | goto out; |
| 3296 | } |
| 3297 | |
| 3298 | while (1) { |
| 3299 | /* |
| 3300 | * I don't think this is needed since we're just marking our |
| 3301 | * preallocated extent as written, but just in case it can't |
| 3302 | * hurt. |
| 3303 | */ |
| 3304 | if (last == 0) { |
| 3305 | err = btrfs_run_delayed_refs(trans, root, |
| 3306 | (unsigned long)-1); |
| 3307 | if (err) /* File system offline */ |
| 3308 | goto out; |
| 3309 | } |
| 3310 | |
| 3311 | cache = btrfs_lookup_first_block_group(root->fs_info, last); |
| 3312 | while (cache) { |
| 3313 | /* |
| 3314 | * Really this shouldn't happen, but it could if we |
| 3315 | * couldn't write the entire preallocated extent and |
| 3316 | * splitting the extent resulted in a new block. |
| 3317 | */ |
| 3318 | if (cache->dirty) { |
| 3319 | btrfs_put_block_group(cache); |
| 3320 | goto again; |
| 3321 | } |
| 3322 | if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE) |
| 3323 | break; |
| 3324 | cache = next_block_group(root, cache); |
| 3325 | } |
| 3326 | if (!cache) { |
| 3327 | if (last == 0) |
| 3328 | break; |
| 3329 | last = 0; |
| 3330 | continue; |
| 3331 | } |
| 3332 | |
| 3333 | err = btrfs_write_out_cache(root, trans, cache, path); |
| 3334 | |
| 3335 | /* |
| 3336 | * If we didn't have an error then the cache state is still |
| 3337 | * NEED_WRITE, so we can set it to WRITTEN. |
| 3338 | */ |
| 3339 | if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE) |
| 3340 | cache->disk_cache_state = BTRFS_DC_WRITTEN; |
| 3341 | last = cache->key.objectid + cache->key.offset; |
| 3342 | btrfs_put_block_group(cache); |
| 3343 | } |
| 3344 | out: |
| 3345 | |
| 3346 | btrfs_free_path(path); |
| 3347 | return err; |
| 3348 | } |
| 3349 | |
| 3350 | int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr) |
| 3351 | { |
| 3352 | struct btrfs_block_group_cache *block_group; |
| 3353 | int readonly = 0; |
| 3354 | |
| 3355 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
| 3356 | if (!block_group || block_group->ro) |
| 3357 | readonly = 1; |
| 3358 | if (block_group) |
| 3359 | btrfs_put_block_group(block_group); |
| 3360 | return readonly; |
| 3361 | } |
| 3362 | |
| 3363 | static const char *alloc_name(u64 flags) |
| 3364 | { |
| 3365 | switch (flags) { |
| 3366 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: |
| 3367 | return "mixed"; |
| 3368 | case BTRFS_BLOCK_GROUP_METADATA: |
| 3369 | return "metadata"; |
| 3370 | case BTRFS_BLOCK_GROUP_DATA: |
| 3371 | return "data"; |
| 3372 | case BTRFS_BLOCK_GROUP_SYSTEM: |
| 3373 | return "system"; |
| 3374 | default: |
| 3375 | WARN_ON(1); |
| 3376 | return "invalid-combination"; |
| 3377 | }; |
| 3378 | } |
| 3379 | |
| 3380 | static int update_space_info(struct btrfs_fs_info *info, u64 flags, |
| 3381 | u64 total_bytes, u64 bytes_used, |
| 3382 | struct btrfs_space_info **space_info) |
| 3383 | { |
| 3384 | struct btrfs_space_info *found; |
| 3385 | int i; |
| 3386 | int factor; |
| 3387 | int ret; |
| 3388 | |
| 3389 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | |
| 3390 | BTRFS_BLOCK_GROUP_RAID10)) |
| 3391 | factor = 2; |
| 3392 | else |
| 3393 | factor = 1; |
| 3394 | |
| 3395 | found = __find_space_info(info, flags); |
| 3396 | if (found) { |
| 3397 | spin_lock(&found->lock); |
| 3398 | found->total_bytes += total_bytes; |
| 3399 | found->disk_total += total_bytes * factor; |
| 3400 | found->bytes_used += bytes_used; |
| 3401 | found->disk_used += bytes_used * factor; |
| 3402 | found->full = 0; |
| 3403 | spin_unlock(&found->lock); |
| 3404 | *space_info = found; |
| 3405 | return 0; |
| 3406 | } |
| 3407 | found = kzalloc(sizeof(*found), GFP_NOFS); |
| 3408 | if (!found) |
| 3409 | return -ENOMEM; |
| 3410 | |
| 3411 | ret = percpu_counter_init(&found->total_bytes_pinned, 0); |
| 3412 | if (ret) { |
| 3413 | kfree(found); |
| 3414 | return ret; |
| 3415 | } |
| 3416 | |
| 3417 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
| 3418 | INIT_LIST_HEAD(&found->block_groups[i]); |
| 3419 | kobject_init(&found->block_group_kobjs[i], &btrfs_raid_ktype); |
| 3420 | } |
| 3421 | init_rwsem(&found->groups_sem); |
| 3422 | spin_lock_init(&found->lock); |
| 3423 | found->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
| 3424 | found->total_bytes = total_bytes; |
| 3425 | found->disk_total = total_bytes * factor; |
| 3426 | found->bytes_used = bytes_used; |
| 3427 | found->disk_used = bytes_used * factor; |
| 3428 | found->bytes_pinned = 0; |
| 3429 | found->bytes_reserved = 0; |
| 3430 | found->bytes_readonly = 0; |
| 3431 | found->bytes_may_use = 0; |
| 3432 | found->full = 0; |
| 3433 | found->force_alloc = CHUNK_ALLOC_NO_FORCE; |
| 3434 | found->chunk_alloc = 0; |
| 3435 | found->flush = 0; |
| 3436 | init_waitqueue_head(&found->wait); |
| 3437 | |
| 3438 | ret = kobject_init_and_add(&found->kobj, &space_info_ktype, |
| 3439 | info->space_info_kobj, "%s", |
| 3440 | alloc_name(found->flags)); |
| 3441 | if (ret) { |
| 3442 | kfree(found); |
| 3443 | return ret; |
| 3444 | } |
| 3445 | |
| 3446 | *space_info = found; |
| 3447 | list_add_rcu(&found->list, &info->space_info); |
| 3448 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
| 3449 | info->data_sinfo = found; |
| 3450 | |
| 3451 | return ret; |
| 3452 | } |
| 3453 | |
| 3454 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
| 3455 | { |
| 3456 | u64 extra_flags = chunk_to_extended(flags) & |
| 3457 | BTRFS_EXTENDED_PROFILE_MASK; |
| 3458 | |
| 3459 | write_seqlock(&fs_info->profiles_lock); |
| 3460 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
| 3461 | fs_info->avail_data_alloc_bits |= extra_flags; |
| 3462 | if (flags & BTRFS_BLOCK_GROUP_METADATA) |
| 3463 | fs_info->avail_metadata_alloc_bits |= extra_flags; |
| 3464 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
| 3465 | fs_info->avail_system_alloc_bits |= extra_flags; |
| 3466 | write_sequnlock(&fs_info->profiles_lock); |
| 3467 | } |
| 3468 | |
| 3469 | /* |
| 3470 | * returns target flags in extended format or 0 if restripe for this |
| 3471 | * chunk_type is not in progress |
| 3472 | * |
| 3473 | * should be called with either volume_mutex or balance_lock held |
| 3474 | */ |
| 3475 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) |
| 3476 | { |
| 3477 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
| 3478 | u64 target = 0; |
| 3479 | |
| 3480 | if (!bctl) |
| 3481 | return 0; |
| 3482 | |
| 3483 | if (flags & BTRFS_BLOCK_GROUP_DATA && |
| 3484 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
| 3485 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; |
| 3486 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && |
| 3487 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
| 3488 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; |
| 3489 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && |
| 3490 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
| 3491 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; |
| 3492 | } |
| 3493 | |
| 3494 | return target; |
| 3495 | } |
| 3496 | |
| 3497 | /* |
| 3498 | * @flags: available profiles in extended format (see ctree.h) |
| 3499 | * |
| 3500 | * Returns reduced profile in chunk format. If profile changing is in |
| 3501 | * progress (either running or paused) picks the target profile (if it's |
| 3502 | * already available), otherwise falls back to plain reducing. |
| 3503 | */ |
| 3504 | static u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) |
| 3505 | { |
| 3506 | /* |
| 3507 | * we add in the count of missing devices because we want |
| 3508 | * to make sure that any RAID levels on a degraded FS |
| 3509 | * continue to be honored. |
| 3510 | */ |
| 3511 | u64 num_devices = root->fs_info->fs_devices->rw_devices + |
| 3512 | root->fs_info->fs_devices->missing_devices; |
| 3513 | u64 target; |
| 3514 | u64 tmp; |
| 3515 | |
| 3516 | /* |
| 3517 | * see if restripe for this chunk_type is in progress, if so |
| 3518 | * try to reduce to the target profile |
| 3519 | */ |
| 3520 | spin_lock(&root->fs_info->balance_lock); |
| 3521 | target = get_restripe_target(root->fs_info, flags); |
| 3522 | if (target) { |
| 3523 | /* pick target profile only if it's already available */ |
| 3524 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { |
| 3525 | spin_unlock(&root->fs_info->balance_lock); |
| 3526 | return extended_to_chunk(target); |
| 3527 | } |
| 3528 | } |
| 3529 | spin_unlock(&root->fs_info->balance_lock); |
| 3530 | |
| 3531 | /* First, mask out the RAID levels which aren't possible */ |
| 3532 | if (num_devices == 1) |
| 3533 | flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0 | |
| 3534 | BTRFS_BLOCK_GROUP_RAID5); |
| 3535 | if (num_devices < 3) |
| 3536 | flags &= ~BTRFS_BLOCK_GROUP_RAID6; |
| 3537 | if (num_devices < 4) |
| 3538 | flags &= ~BTRFS_BLOCK_GROUP_RAID10; |
| 3539 | |
| 3540 | tmp = flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 | |
| 3541 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID5 | |
| 3542 | BTRFS_BLOCK_GROUP_RAID6 | BTRFS_BLOCK_GROUP_RAID10); |
| 3543 | flags &= ~tmp; |
| 3544 | |
| 3545 | if (tmp & BTRFS_BLOCK_GROUP_RAID6) |
| 3546 | tmp = BTRFS_BLOCK_GROUP_RAID6; |
| 3547 | else if (tmp & BTRFS_BLOCK_GROUP_RAID5) |
| 3548 | tmp = BTRFS_BLOCK_GROUP_RAID5; |
| 3549 | else if (tmp & BTRFS_BLOCK_GROUP_RAID10) |
| 3550 | tmp = BTRFS_BLOCK_GROUP_RAID10; |
| 3551 | else if (tmp & BTRFS_BLOCK_GROUP_RAID1) |
| 3552 | tmp = BTRFS_BLOCK_GROUP_RAID1; |
| 3553 | else if (tmp & BTRFS_BLOCK_GROUP_RAID0) |
| 3554 | tmp = BTRFS_BLOCK_GROUP_RAID0; |
| 3555 | |
| 3556 | return extended_to_chunk(flags | tmp); |
| 3557 | } |
| 3558 | |
| 3559 | static u64 get_alloc_profile(struct btrfs_root *root, u64 orig_flags) |
| 3560 | { |
| 3561 | unsigned seq; |
| 3562 | u64 flags; |
| 3563 | |
| 3564 | do { |
| 3565 | flags = orig_flags; |
| 3566 | seq = read_seqbegin(&root->fs_info->profiles_lock); |
| 3567 | |
| 3568 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
| 3569 | flags |= root->fs_info->avail_data_alloc_bits; |
| 3570 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
| 3571 | flags |= root->fs_info->avail_system_alloc_bits; |
| 3572 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
| 3573 | flags |= root->fs_info->avail_metadata_alloc_bits; |
| 3574 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); |
| 3575 | |
| 3576 | return btrfs_reduce_alloc_profile(root, flags); |
| 3577 | } |
| 3578 | |
| 3579 | u64 btrfs_get_alloc_profile(struct btrfs_root *root, int data) |
| 3580 | { |
| 3581 | u64 flags; |
| 3582 | u64 ret; |
| 3583 | |
| 3584 | if (data) |
| 3585 | flags = BTRFS_BLOCK_GROUP_DATA; |
| 3586 | else if (root == root->fs_info->chunk_root) |
| 3587 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
| 3588 | else |
| 3589 | flags = BTRFS_BLOCK_GROUP_METADATA; |
| 3590 | |
| 3591 | ret = get_alloc_profile(root, flags); |
| 3592 | return ret; |
| 3593 | } |
| 3594 | |
| 3595 | /* |
| 3596 | * This will check the space that the inode allocates from to make sure we have |
| 3597 | * enough space for bytes. |
| 3598 | */ |
| 3599 | int btrfs_check_data_free_space(struct inode *inode, u64 bytes) |
| 3600 | { |
| 3601 | struct btrfs_space_info *data_sinfo; |
| 3602 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 3603 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 3604 | u64 used; |
| 3605 | int ret = 0, committed = 0, alloc_chunk = 1; |
| 3606 | |
| 3607 | /* make sure bytes are sectorsize aligned */ |
| 3608 | bytes = ALIGN(bytes, root->sectorsize); |
| 3609 | |
| 3610 | if (btrfs_is_free_space_inode(inode)) { |
| 3611 | committed = 1; |
| 3612 | ASSERT(current->journal_info); |
| 3613 | } |
| 3614 | |
| 3615 | data_sinfo = fs_info->data_sinfo; |
| 3616 | if (!data_sinfo) |
| 3617 | goto alloc; |
| 3618 | |
| 3619 | again: |
| 3620 | /* make sure we have enough space to handle the data first */ |
| 3621 | spin_lock(&data_sinfo->lock); |
| 3622 | used = data_sinfo->bytes_used + data_sinfo->bytes_reserved + |
| 3623 | data_sinfo->bytes_pinned + data_sinfo->bytes_readonly + |
| 3624 | data_sinfo->bytes_may_use; |
| 3625 | |
| 3626 | if (used + bytes > data_sinfo->total_bytes) { |
| 3627 | struct btrfs_trans_handle *trans; |
| 3628 | |
| 3629 | /* |
| 3630 | * if we don't have enough free bytes in this space then we need |
| 3631 | * to alloc a new chunk. |
| 3632 | */ |
| 3633 | if (!data_sinfo->full && alloc_chunk) { |
| 3634 | u64 alloc_target; |
| 3635 | |
| 3636 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
| 3637 | spin_unlock(&data_sinfo->lock); |
| 3638 | alloc: |
| 3639 | alloc_target = btrfs_get_alloc_profile(root, 1); |
| 3640 | /* |
| 3641 | * It is ugly that we don't call nolock join |
| 3642 | * transaction for the free space inode case here. |
| 3643 | * But it is safe because we only do the data space |
| 3644 | * reservation for the free space cache in the |
| 3645 | * transaction context, the common join transaction |
| 3646 | * just increase the counter of the current transaction |
| 3647 | * handler, doesn't try to acquire the trans_lock of |
| 3648 | * the fs. |
| 3649 | */ |
| 3650 | trans = btrfs_join_transaction(root); |
| 3651 | if (IS_ERR(trans)) |
| 3652 | return PTR_ERR(trans); |
| 3653 | |
| 3654 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
| 3655 | alloc_target, |
| 3656 | CHUNK_ALLOC_NO_FORCE); |
| 3657 | btrfs_end_transaction(trans, root); |
| 3658 | if (ret < 0) { |
| 3659 | if (ret != -ENOSPC) |
| 3660 | return ret; |
| 3661 | else |
| 3662 | goto commit_trans; |
| 3663 | } |
| 3664 | |
| 3665 | if (!data_sinfo) |
| 3666 | data_sinfo = fs_info->data_sinfo; |
| 3667 | |
| 3668 | goto again; |
| 3669 | } |
| 3670 | |
| 3671 | /* |
| 3672 | * If we don't have enough pinned space to deal with this |
| 3673 | * allocation don't bother committing the transaction. |
| 3674 | */ |
| 3675 | if (percpu_counter_compare(&data_sinfo->total_bytes_pinned, |
| 3676 | bytes) < 0) |
| 3677 | committed = 1; |
| 3678 | spin_unlock(&data_sinfo->lock); |
| 3679 | |
| 3680 | /* commit the current transaction and try again */ |
| 3681 | commit_trans: |
| 3682 | if (!committed && |
| 3683 | !atomic_read(&root->fs_info->open_ioctl_trans)) { |
| 3684 | committed = 1; |
| 3685 | |
| 3686 | trans = btrfs_join_transaction(root); |
| 3687 | if (IS_ERR(trans)) |
| 3688 | return PTR_ERR(trans); |
| 3689 | ret = btrfs_commit_transaction(trans, root); |
| 3690 | if (ret) |
| 3691 | return ret; |
| 3692 | goto again; |
| 3693 | } |
| 3694 | |
| 3695 | trace_btrfs_space_reservation(root->fs_info, |
| 3696 | "space_info:enospc", |
| 3697 | data_sinfo->flags, bytes, 1); |
| 3698 | return -ENOSPC; |
| 3699 | } |
| 3700 | data_sinfo->bytes_may_use += bytes; |
| 3701 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
| 3702 | data_sinfo->flags, bytes, 1); |
| 3703 | spin_unlock(&data_sinfo->lock); |
| 3704 | |
| 3705 | return 0; |
| 3706 | } |
| 3707 | |
| 3708 | /* |
| 3709 | * Called if we need to clear a data reservation for this inode. |
| 3710 | */ |
| 3711 | void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes) |
| 3712 | { |
| 3713 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 3714 | struct btrfs_space_info *data_sinfo; |
| 3715 | |
| 3716 | /* make sure bytes are sectorsize aligned */ |
| 3717 | bytes = ALIGN(bytes, root->sectorsize); |
| 3718 | |
| 3719 | data_sinfo = root->fs_info->data_sinfo; |
| 3720 | spin_lock(&data_sinfo->lock); |
| 3721 | WARN_ON(data_sinfo->bytes_may_use < bytes); |
| 3722 | data_sinfo->bytes_may_use -= bytes; |
| 3723 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
| 3724 | data_sinfo->flags, bytes, 0); |
| 3725 | spin_unlock(&data_sinfo->lock); |
| 3726 | } |
| 3727 | |
| 3728 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
| 3729 | { |
| 3730 | struct list_head *head = &info->space_info; |
| 3731 | struct btrfs_space_info *found; |
| 3732 | |
| 3733 | rcu_read_lock(); |
| 3734 | list_for_each_entry_rcu(found, head, list) { |
| 3735 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) |
| 3736 | found->force_alloc = CHUNK_ALLOC_FORCE; |
| 3737 | } |
| 3738 | rcu_read_unlock(); |
| 3739 | } |
| 3740 | |
| 3741 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
| 3742 | { |
| 3743 | return (global->size << 1); |
| 3744 | } |
| 3745 | |
| 3746 | static int should_alloc_chunk(struct btrfs_root *root, |
| 3747 | struct btrfs_space_info *sinfo, int force) |
| 3748 | { |
| 3749 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
| 3750 | u64 num_bytes = sinfo->total_bytes - sinfo->bytes_readonly; |
| 3751 | u64 num_allocated = sinfo->bytes_used + sinfo->bytes_reserved; |
| 3752 | u64 thresh; |
| 3753 | |
| 3754 | if (force == CHUNK_ALLOC_FORCE) |
| 3755 | return 1; |
| 3756 | |
| 3757 | /* |
| 3758 | * We need to take into account the global rsv because for all intents |
| 3759 | * and purposes it's used space. Don't worry about locking the |
| 3760 | * global_rsv, it doesn't change except when the transaction commits. |
| 3761 | */ |
| 3762 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
| 3763 | num_allocated += calc_global_rsv_need_space(global_rsv); |
| 3764 | |
| 3765 | /* |
| 3766 | * in limited mode, we want to have some free space up to |
| 3767 | * about 1% of the FS size. |
| 3768 | */ |
| 3769 | if (force == CHUNK_ALLOC_LIMITED) { |
| 3770 | thresh = btrfs_super_total_bytes(root->fs_info->super_copy); |
| 3771 | thresh = max_t(u64, 64 * 1024 * 1024, |
| 3772 | div_factor_fine(thresh, 1)); |
| 3773 | |
| 3774 | if (num_bytes - num_allocated < thresh) |
| 3775 | return 1; |
| 3776 | } |
| 3777 | |
| 3778 | if (num_allocated + 2 * 1024 * 1024 < div_factor(num_bytes, 8)) |
| 3779 | return 0; |
| 3780 | return 1; |
| 3781 | } |
| 3782 | |
| 3783 | static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type) |
| 3784 | { |
| 3785 | u64 num_dev; |
| 3786 | |
| 3787 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
| 3788 | BTRFS_BLOCK_GROUP_RAID0 | |
| 3789 | BTRFS_BLOCK_GROUP_RAID5 | |
| 3790 | BTRFS_BLOCK_GROUP_RAID6)) |
| 3791 | num_dev = root->fs_info->fs_devices->rw_devices; |
| 3792 | else if (type & BTRFS_BLOCK_GROUP_RAID1) |
| 3793 | num_dev = 2; |
| 3794 | else |
| 3795 | num_dev = 1; /* DUP or single */ |
| 3796 | |
| 3797 | /* metadata for updaing devices and chunk tree */ |
| 3798 | return btrfs_calc_trans_metadata_size(root, num_dev + 1); |
| 3799 | } |
| 3800 | |
| 3801 | static void check_system_chunk(struct btrfs_trans_handle *trans, |
| 3802 | struct btrfs_root *root, u64 type) |
| 3803 | { |
| 3804 | struct btrfs_space_info *info; |
| 3805 | u64 left; |
| 3806 | u64 thresh; |
| 3807 | |
| 3808 | info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
| 3809 | spin_lock(&info->lock); |
| 3810 | left = info->total_bytes - info->bytes_used - info->bytes_pinned - |
| 3811 | info->bytes_reserved - info->bytes_readonly; |
| 3812 | spin_unlock(&info->lock); |
| 3813 | |
| 3814 | thresh = get_system_chunk_thresh(root, type); |
| 3815 | if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) { |
| 3816 | btrfs_info(root->fs_info, "left=%llu, need=%llu, flags=%llu", |
| 3817 | left, thresh, type); |
| 3818 | dump_space_info(info, 0, 0); |
| 3819 | } |
| 3820 | |
| 3821 | if (left < thresh) { |
| 3822 | u64 flags; |
| 3823 | |
| 3824 | flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0); |
| 3825 | btrfs_alloc_chunk(trans, root, flags); |
| 3826 | } |
| 3827 | } |
| 3828 | |
| 3829 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
| 3830 | struct btrfs_root *extent_root, u64 flags, int force) |
| 3831 | { |
| 3832 | struct btrfs_space_info *space_info; |
| 3833 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
| 3834 | int wait_for_alloc = 0; |
| 3835 | int ret = 0; |
| 3836 | |
| 3837 | /* Don't re-enter if we're already allocating a chunk */ |
| 3838 | if (trans->allocating_chunk) |
| 3839 | return -ENOSPC; |
| 3840 | |
| 3841 | space_info = __find_space_info(extent_root->fs_info, flags); |
| 3842 | if (!space_info) { |
| 3843 | ret = update_space_info(extent_root->fs_info, flags, |
| 3844 | 0, 0, &space_info); |
| 3845 | BUG_ON(ret); /* -ENOMEM */ |
| 3846 | } |
| 3847 | BUG_ON(!space_info); /* Logic error */ |
| 3848 | |
| 3849 | again: |
| 3850 | spin_lock(&space_info->lock); |
| 3851 | if (force < space_info->force_alloc) |
| 3852 | force = space_info->force_alloc; |
| 3853 | if (space_info->full) { |
| 3854 | if (should_alloc_chunk(extent_root, space_info, force)) |
| 3855 | ret = -ENOSPC; |
| 3856 | else |
| 3857 | ret = 0; |
| 3858 | spin_unlock(&space_info->lock); |
| 3859 | return ret; |
| 3860 | } |
| 3861 | |
| 3862 | if (!should_alloc_chunk(extent_root, space_info, force)) { |
| 3863 | spin_unlock(&space_info->lock); |
| 3864 | return 0; |
| 3865 | } else if (space_info->chunk_alloc) { |
| 3866 | wait_for_alloc = 1; |
| 3867 | } else { |
| 3868 | space_info->chunk_alloc = 1; |
| 3869 | } |
| 3870 | |
| 3871 | spin_unlock(&space_info->lock); |
| 3872 | |
| 3873 | mutex_lock(&fs_info->chunk_mutex); |
| 3874 | |
| 3875 | /* |
| 3876 | * The chunk_mutex is held throughout the entirety of a chunk |
| 3877 | * allocation, so once we've acquired the chunk_mutex we know that the |
| 3878 | * other guy is done and we need to recheck and see if we should |
| 3879 | * allocate. |
| 3880 | */ |
| 3881 | if (wait_for_alloc) { |
| 3882 | mutex_unlock(&fs_info->chunk_mutex); |
| 3883 | wait_for_alloc = 0; |
| 3884 | goto again; |
| 3885 | } |
| 3886 | |
| 3887 | trans->allocating_chunk = true; |
| 3888 | |
| 3889 | /* |
| 3890 | * If we have mixed data/metadata chunks we want to make sure we keep |
| 3891 | * allocating mixed chunks instead of individual chunks. |
| 3892 | */ |
| 3893 | if (btrfs_mixed_space_info(space_info)) |
| 3894 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); |
| 3895 | |
| 3896 | /* |
| 3897 | * if we're doing a data chunk, go ahead and make sure that |
| 3898 | * we keep a reasonable number of metadata chunks allocated in the |
| 3899 | * FS as well. |
| 3900 | */ |
| 3901 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
| 3902 | fs_info->data_chunk_allocations++; |
| 3903 | if (!(fs_info->data_chunk_allocations % |
| 3904 | fs_info->metadata_ratio)) |
| 3905 | force_metadata_allocation(fs_info); |
| 3906 | } |
| 3907 | |
| 3908 | /* |
| 3909 | * Check if we have enough space in SYSTEM chunk because we may need |
| 3910 | * to update devices. |
| 3911 | */ |
| 3912 | check_system_chunk(trans, extent_root, flags); |
| 3913 | |
| 3914 | ret = btrfs_alloc_chunk(trans, extent_root, flags); |
| 3915 | trans->allocating_chunk = false; |
| 3916 | |
| 3917 | spin_lock(&space_info->lock); |
| 3918 | if (ret < 0 && ret != -ENOSPC) |
| 3919 | goto out; |
| 3920 | if (ret) |
| 3921 | space_info->full = 1; |
| 3922 | else |
| 3923 | ret = 1; |
| 3924 | |
| 3925 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
| 3926 | out: |
| 3927 | space_info->chunk_alloc = 0; |
| 3928 | spin_unlock(&space_info->lock); |
| 3929 | mutex_unlock(&fs_info->chunk_mutex); |
| 3930 | return ret; |
| 3931 | } |
| 3932 | |
| 3933 | static int can_overcommit(struct btrfs_root *root, |
| 3934 | struct btrfs_space_info *space_info, u64 bytes, |
| 3935 | enum btrfs_reserve_flush_enum flush) |
| 3936 | { |
| 3937 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
| 3938 | u64 profile = btrfs_get_alloc_profile(root, 0); |
| 3939 | u64 space_size; |
| 3940 | u64 avail; |
| 3941 | u64 used; |
| 3942 | |
| 3943 | used = space_info->bytes_used + space_info->bytes_reserved + |
| 3944 | space_info->bytes_pinned + space_info->bytes_readonly; |
| 3945 | |
| 3946 | /* |
| 3947 | * We only want to allow over committing if we have lots of actual space |
| 3948 | * free, but if we don't have enough space to handle the global reserve |
| 3949 | * space then we could end up having a real enospc problem when trying |
| 3950 | * to allocate a chunk or some other such important allocation. |
| 3951 | */ |
| 3952 | spin_lock(&global_rsv->lock); |
| 3953 | space_size = calc_global_rsv_need_space(global_rsv); |
| 3954 | spin_unlock(&global_rsv->lock); |
| 3955 | if (used + space_size >= space_info->total_bytes) |
| 3956 | return 0; |
| 3957 | |
| 3958 | used += space_info->bytes_may_use; |
| 3959 | |
| 3960 | spin_lock(&root->fs_info->free_chunk_lock); |
| 3961 | avail = root->fs_info->free_chunk_space; |
| 3962 | spin_unlock(&root->fs_info->free_chunk_lock); |
| 3963 | |
| 3964 | /* |
| 3965 | * If we have dup, raid1 or raid10 then only half of the free |
| 3966 | * space is actually useable. For raid56, the space info used |
| 3967 | * doesn't include the parity drive, so we don't have to |
| 3968 | * change the math |
| 3969 | */ |
| 3970 | if (profile & (BTRFS_BLOCK_GROUP_DUP | |
| 3971 | BTRFS_BLOCK_GROUP_RAID1 | |
| 3972 | BTRFS_BLOCK_GROUP_RAID10)) |
| 3973 | avail >>= 1; |
| 3974 | |
| 3975 | /* |
| 3976 | * If we aren't flushing all things, let us overcommit up to |
| 3977 | * 1/2th of the space. If we can flush, don't let us overcommit |
| 3978 | * too much, let it overcommit up to 1/8 of the space. |
| 3979 | */ |
| 3980 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
| 3981 | avail >>= 3; |
| 3982 | else |
| 3983 | avail >>= 1; |
| 3984 | |
| 3985 | if (used + bytes < space_info->total_bytes + avail) |
| 3986 | return 1; |
| 3987 | return 0; |
| 3988 | } |
| 3989 | |
| 3990 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_root *root, |
| 3991 | unsigned long nr_pages, int nr_items) |
| 3992 | { |
| 3993 | struct super_block *sb = root->fs_info->sb; |
| 3994 | |
| 3995 | if (down_read_trylock(&sb->s_umount)) { |
| 3996 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); |
| 3997 | up_read(&sb->s_umount); |
| 3998 | } else { |
| 3999 | /* |
| 4000 | * We needn't worry the filesystem going from r/w to r/o though |
| 4001 | * we don't acquire ->s_umount mutex, because the filesystem |
| 4002 | * should guarantee the delalloc inodes list be empty after |
| 4003 | * the filesystem is readonly(all dirty pages are written to |
| 4004 | * the disk). |
| 4005 | */ |
| 4006 | btrfs_start_delalloc_roots(root->fs_info, 0, nr_items); |
| 4007 | if (!current->journal_info) |
| 4008 | btrfs_wait_ordered_roots(root->fs_info, nr_items); |
| 4009 | } |
| 4010 | } |
| 4011 | |
| 4012 | static inline int calc_reclaim_items_nr(struct btrfs_root *root, u64 to_reclaim) |
| 4013 | { |
| 4014 | u64 bytes; |
| 4015 | int nr; |
| 4016 | |
| 4017 | bytes = btrfs_calc_trans_metadata_size(root, 1); |
| 4018 | nr = (int)div64_u64(to_reclaim, bytes); |
| 4019 | if (!nr) |
| 4020 | nr = 1; |
| 4021 | return nr; |
| 4022 | } |
| 4023 | |
| 4024 | #define EXTENT_SIZE_PER_ITEM (256 * 1024) |
| 4025 | |
| 4026 | /* |
| 4027 | * shrink metadata reservation for delalloc |
| 4028 | */ |
| 4029 | static void shrink_delalloc(struct btrfs_root *root, u64 to_reclaim, u64 orig, |
| 4030 | bool wait_ordered) |
| 4031 | { |
| 4032 | struct btrfs_block_rsv *block_rsv; |
| 4033 | struct btrfs_space_info *space_info; |
| 4034 | struct btrfs_trans_handle *trans; |
| 4035 | u64 delalloc_bytes; |
| 4036 | u64 max_reclaim; |
| 4037 | long time_left; |
| 4038 | unsigned long nr_pages; |
| 4039 | int loops; |
| 4040 | int items; |
| 4041 | enum btrfs_reserve_flush_enum flush; |
| 4042 | |
| 4043 | /* Calc the number of the pages we need flush for space reservation */ |
| 4044 | items = calc_reclaim_items_nr(root, to_reclaim); |
| 4045 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; |
| 4046 | |
| 4047 | trans = (struct btrfs_trans_handle *)current->journal_info; |
| 4048 | block_rsv = &root->fs_info->delalloc_block_rsv; |
| 4049 | space_info = block_rsv->space_info; |
| 4050 | |
| 4051 | delalloc_bytes = percpu_counter_sum_positive( |
| 4052 | &root->fs_info->delalloc_bytes); |
| 4053 | if (delalloc_bytes == 0) { |
| 4054 | if (trans) |
| 4055 | return; |
| 4056 | if (wait_ordered) |
| 4057 | btrfs_wait_ordered_roots(root->fs_info, items); |
| 4058 | return; |
| 4059 | } |
| 4060 | |
| 4061 | loops = 0; |
| 4062 | while (delalloc_bytes && loops < 3) { |
| 4063 | max_reclaim = min(delalloc_bytes, to_reclaim); |
| 4064 | nr_pages = max_reclaim >> PAGE_CACHE_SHIFT; |
| 4065 | btrfs_writeback_inodes_sb_nr(root, nr_pages, items); |
| 4066 | /* |
| 4067 | * We need to wait for the async pages to actually start before |
| 4068 | * we do anything. |
| 4069 | */ |
| 4070 | max_reclaim = atomic_read(&root->fs_info->async_delalloc_pages); |
| 4071 | if (!max_reclaim) |
| 4072 | goto skip_async; |
| 4073 | |
| 4074 | if (max_reclaim <= nr_pages) |
| 4075 | max_reclaim = 0; |
| 4076 | else |
| 4077 | max_reclaim -= nr_pages; |
| 4078 | |
| 4079 | wait_event(root->fs_info->async_submit_wait, |
| 4080 | atomic_read(&root->fs_info->async_delalloc_pages) <= |
| 4081 | (int)max_reclaim); |
| 4082 | skip_async: |
| 4083 | if (!trans) |
| 4084 | flush = BTRFS_RESERVE_FLUSH_ALL; |
| 4085 | else |
| 4086 | flush = BTRFS_RESERVE_NO_FLUSH; |
| 4087 | spin_lock(&space_info->lock); |
| 4088 | if (can_overcommit(root, space_info, orig, flush)) { |
| 4089 | spin_unlock(&space_info->lock); |
| 4090 | break; |
| 4091 | } |
| 4092 | spin_unlock(&space_info->lock); |
| 4093 | |
| 4094 | loops++; |
| 4095 | if (wait_ordered && !trans) { |
| 4096 | btrfs_wait_ordered_roots(root->fs_info, items); |
| 4097 | } else { |
| 4098 | time_left = schedule_timeout_killable(1); |
| 4099 | if (time_left) |
| 4100 | break; |
| 4101 | } |
| 4102 | delalloc_bytes = percpu_counter_sum_positive( |
| 4103 | &root->fs_info->delalloc_bytes); |
| 4104 | } |
| 4105 | } |
| 4106 | |
| 4107 | /** |
| 4108 | * maybe_commit_transaction - possibly commit the transaction if its ok to |
| 4109 | * @root - the root we're allocating for |
| 4110 | * @bytes - the number of bytes we want to reserve |
| 4111 | * @force - force the commit |
| 4112 | * |
| 4113 | * This will check to make sure that committing the transaction will actually |
| 4114 | * get us somewhere and then commit the transaction if it does. Otherwise it |
| 4115 | * will return -ENOSPC. |
| 4116 | */ |
| 4117 | static int may_commit_transaction(struct btrfs_root *root, |
| 4118 | struct btrfs_space_info *space_info, |
| 4119 | u64 bytes, int force) |
| 4120 | { |
| 4121 | struct btrfs_block_rsv *delayed_rsv = &root->fs_info->delayed_block_rsv; |
| 4122 | struct btrfs_trans_handle *trans; |
| 4123 | |
| 4124 | trans = (struct btrfs_trans_handle *)current->journal_info; |
| 4125 | if (trans) |
| 4126 | return -EAGAIN; |
| 4127 | |
| 4128 | if (force) |
| 4129 | goto commit; |
| 4130 | |
| 4131 | /* See if there is enough pinned space to make this reservation */ |
| 4132 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
| 4133 | bytes) >= 0) |
| 4134 | goto commit; |
| 4135 | |
| 4136 | /* |
| 4137 | * See if there is some space in the delayed insertion reservation for |
| 4138 | * this reservation. |
| 4139 | */ |
| 4140 | if (space_info != delayed_rsv->space_info) |
| 4141 | return -ENOSPC; |
| 4142 | |
| 4143 | spin_lock(&delayed_rsv->lock); |
| 4144 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
| 4145 | bytes - delayed_rsv->size) >= 0) { |
| 4146 | spin_unlock(&delayed_rsv->lock); |
| 4147 | return -ENOSPC; |
| 4148 | } |
| 4149 | spin_unlock(&delayed_rsv->lock); |
| 4150 | |
| 4151 | commit: |
| 4152 | trans = btrfs_join_transaction(root); |
| 4153 | if (IS_ERR(trans)) |
| 4154 | return -ENOSPC; |
| 4155 | |
| 4156 | return btrfs_commit_transaction(trans, root); |
| 4157 | } |
| 4158 | |
| 4159 | enum flush_state { |
| 4160 | FLUSH_DELAYED_ITEMS_NR = 1, |
| 4161 | FLUSH_DELAYED_ITEMS = 2, |
| 4162 | FLUSH_DELALLOC = 3, |
| 4163 | FLUSH_DELALLOC_WAIT = 4, |
| 4164 | ALLOC_CHUNK = 5, |
| 4165 | COMMIT_TRANS = 6, |
| 4166 | }; |
| 4167 | |
| 4168 | static int flush_space(struct btrfs_root *root, |
| 4169 | struct btrfs_space_info *space_info, u64 num_bytes, |
| 4170 | u64 orig_bytes, int state) |
| 4171 | { |
| 4172 | struct btrfs_trans_handle *trans; |
| 4173 | int nr; |
| 4174 | int ret = 0; |
| 4175 | |
| 4176 | switch (state) { |
| 4177 | case FLUSH_DELAYED_ITEMS_NR: |
| 4178 | case FLUSH_DELAYED_ITEMS: |
| 4179 | if (state == FLUSH_DELAYED_ITEMS_NR) |
| 4180 | nr = calc_reclaim_items_nr(root, num_bytes) * 2; |
| 4181 | else |
| 4182 | nr = -1; |
| 4183 | |
| 4184 | trans = btrfs_join_transaction(root); |
| 4185 | if (IS_ERR(trans)) { |
| 4186 | ret = PTR_ERR(trans); |
| 4187 | break; |
| 4188 | } |
| 4189 | ret = btrfs_run_delayed_items_nr(trans, root, nr); |
| 4190 | btrfs_end_transaction(trans, root); |
| 4191 | break; |
| 4192 | case FLUSH_DELALLOC: |
| 4193 | case FLUSH_DELALLOC_WAIT: |
| 4194 | shrink_delalloc(root, num_bytes * 2, orig_bytes, |
| 4195 | state == FLUSH_DELALLOC_WAIT); |
| 4196 | break; |
| 4197 | case ALLOC_CHUNK: |
| 4198 | trans = btrfs_join_transaction(root); |
| 4199 | if (IS_ERR(trans)) { |
| 4200 | ret = PTR_ERR(trans); |
| 4201 | break; |
| 4202 | } |
| 4203 | ret = do_chunk_alloc(trans, root->fs_info->extent_root, |
| 4204 | btrfs_get_alloc_profile(root, 0), |
| 4205 | CHUNK_ALLOC_NO_FORCE); |
| 4206 | btrfs_end_transaction(trans, root); |
| 4207 | if (ret == -ENOSPC) |
| 4208 | ret = 0; |
| 4209 | break; |
| 4210 | case COMMIT_TRANS: |
| 4211 | ret = may_commit_transaction(root, space_info, orig_bytes, 0); |
| 4212 | break; |
| 4213 | default: |
| 4214 | ret = -ENOSPC; |
| 4215 | break; |
| 4216 | } |
| 4217 | |
| 4218 | return ret; |
| 4219 | } |
| 4220 | |
| 4221 | static inline u64 |
| 4222 | btrfs_calc_reclaim_metadata_size(struct btrfs_root *root, |
| 4223 | struct btrfs_space_info *space_info) |
| 4224 | { |
| 4225 | u64 used; |
| 4226 | u64 expected; |
| 4227 | u64 to_reclaim; |
| 4228 | |
| 4229 | to_reclaim = min_t(u64, num_online_cpus() * 1024 * 1024, |
| 4230 | 16 * 1024 * 1024); |
| 4231 | spin_lock(&space_info->lock); |
| 4232 | if (can_overcommit(root, space_info, to_reclaim, |
| 4233 | BTRFS_RESERVE_FLUSH_ALL)) { |
| 4234 | to_reclaim = 0; |
| 4235 | goto out; |
| 4236 | } |
| 4237 | |
| 4238 | used = space_info->bytes_used + space_info->bytes_reserved + |
| 4239 | space_info->bytes_pinned + space_info->bytes_readonly + |
| 4240 | space_info->bytes_may_use; |
| 4241 | if (can_overcommit(root, space_info, 1024 * 1024, |
| 4242 | BTRFS_RESERVE_FLUSH_ALL)) |
| 4243 | expected = div_factor_fine(space_info->total_bytes, 95); |
| 4244 | else |
| 4245 | expected = div_factor_fine(space_info->total_bytes, 90); |
| 4246 | |
| 4247 | if (used > expected) |
| 4248 | to_reclaim = used - expected; |
| 4249 | else |
| 4250 | to_reclaim = 0; |
| 4251 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + |
| 4252 | space_info->bytes_reserved); |
| 4253 | out: |
| 4254 | spin_unlock(&space_info->lock); |
| 4255 | |
| 4256 | return to_reclaim; |
| 4257 | } |
| 4258 | |
| 4259 | static inline int need_do_async_reclaim(struct btrfs_space_info *space_info, |
| 4260 | struct btrfs_fs_info *fs_info, u64 used) |
| 4261 | { |
| 4262 | return (used >= div_factor_fine(space_info->total_bytes, 98) && |
| 4263 | !btrfs_fs_closing(fs_info) && |
| 4264 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); |
| 4265 | } |
| 4266 | |
| 4267 | static int btrfs_need_do_async_reclaim(struct btrfs_space_info *space_info, |
| 4268 | struct btrfs_fs_info *fs_info) |
| 4269 | { |
| 4270 | u64 used; |
| 4271 | |
| 4272 | spin_lock(&space_info->lock); |
| 4273 | used = space_info->bytes_used + space_info->bytes_reserved + |
| 4274 | space_info->bytes_pinned + space_info->bytes_readonly + |
| 4275 | space_info->bytes_may_use; |
| 4276 | if (need_do_async_reclaim(space_info, fs_info, used)) { |
| 4277 | spin_unlock(&space_info->lock); |
| 4278 | return 1; |
| 4279 | } |
| 4280 | spin_unlock(&space_info->lock); |
| 4281 | |
| 4282 | return 0; |
| 4283 | } |
| 4284 | |
| 4285 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
| 4286 | { |
| 4287 | struct btrfs_fs_info *fs_info; |
| 4288 | struct btrfs_space_info *space_info; |
| 4289 | u64 to_reclaim; |
| 4290 | int flush_state; |
| 4291 | |
| 4292 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); |
| 4293 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
| 4294 | |
| 4295 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info->fs_root, |
| 4296 | space_info); |
| 4297 | if (!to_reclaim) |
| 4298 | return; |
| 4299 | |
| 4300 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
| 4301 | do { |
| 4302 | flush_space(fs_info->fs_root, space_info, to_reclaim, |
| 4303 | to_reclaim, flush_state); |
| 4304 | flush_state++; |
| 4305 | if (!btrfs_need_do_async_reclaim(space_info, fs_info)) |
| 4306 | return; |
| 4307 | } while (flush_state <= COMMIT_TRANS); |
| 4308 | |
| 4309 | if (btrfs_need_do_async_reclaim(space_info, fs_info)) |
| 4310 | queue_work(system_unbound_wq, work); |
| 4311 | } |
| 4312 | |
| 4313 | void btrfs_init_async_reclaim_work(struct work_struct *work) |
| 4314 | { |
| 4315 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); |
| 4316 | } |
| 4317 | |
| 4318 | /** |
| 4319 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space |
| 4320 | * @root - the root we're allocating for |
| 4321 | * @block_rsv - the block_rsv we're allocating for |
| 4322 | * @orig_bytes - the number of bytes we want |
| 4323 | * @flush - whether or not we can flush to make our reservation |
| 4324 | * |
| 4325 | * This will reserve orgi_bytes number of bytes from the space info associated |
| 4326 | * with the block_rsv. If there is not enough space it will make an attempt to |
| 4327 | * flush out space to make room. It will do this by flushing delalloc if |
| 4328 | * possible or committing the transaction. If flush is 0 then no attempts to |
| 4329 | * regain reservations will be made and this will fail if there is not enough |
| 4330 | * space already. |
| 4331 | */ |
| 4332 | static int reserve_metadata_bytes(struct btrfs_root *root, |
| 4333 | struct btrfs_block_rsv *block_rsv, |
| 4334 | u64 orig_bytes, |
| 4335 | enum btrfs_reserve_flush_enum flush) |
| 4336 | { |
| 4337 | struct btrfs_space_info *space_info = block_rsv->space_info; |
| 4338 | u64 used; |
| 4339 | u64 num_bytes = orig_bytes; |
| 4340 | int flush_state = FLUSH_DELAYED_ITEMS_NR; |
| 4341 | int ret = 0; |
| 4342 | bool flushing = false; |
| 4343 | |
| 4344 | again: |
| 4345 | ret = 0; |
| 4346 | spin_lock(&space_info->lock); |
| 4347 | /* |
| 4348 | * We only want to wait if somebody other than us is flushing and we |
| 4349 | * are actually allowed to flush all things. |
| 4350 | */ |
| 4351 | while (flush == BTRFS_RESERVE_FLUSH_ALL && !flushing && |
| 4352 | space_info->flush) { |
| 4353 | spin_unlock(&space_info->lock); |
| 4354 | /* |
| 4355 | * If we have a trans handle we can't wait because the flusher |
| 4356 | * may have to commit the transaction, which would mean we would |
| 4357 | * deadlock since we are waiting for the flusher to finish, but |
| 4358 | * hold the current transaction open. |
| 4359 | */ |
| 4360 | if (current->journal_info) |
| 4361 | return -EAGAIN; |
| 4362 | ret = wait_event_killable(space_info->wait, !space_info->flush); |
| 4363 | /* Must have been killed, return */ |
| 4364 | if (ret) |
| 4365 | return -EINTR; |
| 4366 | |
| 4367 | spin_lock(&space_info->lock); |
| 4368 | } |
| 4369 | |
| 4370 | ret = -ENOSPC; |
| 4371 | used = space_info->bytes_used + space_info->bytes_reserved + |
| 4372 | space_info->bytes_pinned + space_info->bytes_readonly + |
| 4373 | space_info->bytes_may_use; |
| 4374 | |
| 4375 | /* |
| 4376 | * The idea here is that we've not already over-reserved the block group |
| 4377 | * then we can go ahead and save our reservation first and then start |
| 4378 | * flushing if we need to. Otherwise if we've already overcommitted |
| 4379 | * lets start flushing stuff first and then come back and try to make |
| 4380 | * our reservation. |
| 4381 | */ |
| 4382 | if (used <= space_info->total_bytes) { |
| 4383 | if (used + orig_bytes <= space_info->total_bytes) { |
| 4384 | space_info->bytes_may_use += orig_bytes; |
| 4385 | trace_btrfs_space_reservation(root->fs_info, |
| 4386 | "space_info", space_info->flags, orig_bytes, 1); |
| 4387 | ret = 0; |
| 4388 | } else { |
| 4389 | /* |
| 4390 | * Ok set num_bytes to orig_bytes since we aren't |
| 4391 | * overocmmitted, this way we only try and reclaim what |
| 4392 | * we need. |
| 4393 | */ |
| 4394 | num_bytes = orig_bytes; |
| 4395 | } |
| 4396 | } else { |
| 4397 | /* |
| 4398 | * Ok we're over committed, set num_bytes to the overcommitted |
| 4399 | * amount plus the amount of bytes that we need for this |
| 4400 | * reservation. |
| 4401 | */ |
| 4402 | num_bytes = used - space_info->total_bytes + |
| 4403 | (orig_bytes * 2); |
| 4404 | } |
| 4405 | |
| 4406 | if (ret && can_overcommit(root, space_info, orig_bytes, flush)) { |
| 4407 | space_info->bytes_may_use += orig_bytes; |
| 4408 | trace_btrfs_space_reservation(root->fs_info, "space_info", |
| 4409 | space_info->flags, orig_bytes, |
| 4410 | 1); |
| 4411 | ret = 0; |
| 4412 | } |
| 4413 | |
| 4414 | /* |
| 4415 | * Couldn't make our reservation, save our place so while we're trying |
| 4416 | * to reclaim space we can actually use it instead of somebody else |
| 4417 | * stealing it from us. |
| 4418 | * |
| 4419 | * We make the other tasks wait for the flush only when we can flush |
| 4420 | * all things. |
| 4421 | */ |
| 4422 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
| 4423 | flushing = true; |
| 4424 | space_info->flush = 1; |
| 4425 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
| 4426 | used += orig_bytes; |
| 4427 | if (need_do_async_reclaim(space_info, root->fs_info, used) && |
| 4428 | !work_busy(&root->fs_info->async_reclaim_work)) |
| 4429 | queue_work(system_unbound_wq, |
| 4430 | &root->fs_info->async_reclaim_work); |
| 4431 | } |
| 4432 | spin_unlock(&space_info->lock); |
| 4433 | |
| 4434 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
| 4435 | goto out; |
| 4436 | |
| 4437 | ret = flush_space(root, space_info, num_bytes, orig_bytes, |
| 4438 | flush_state); |
| 4439 | flush_state++; |
| 4440 | |
| 4441 | /* |
| 4442 | * If we are FLUSH_LIMIT, we can not flush delalloc, or the deadlock |
| 4443 | * would happen. So skip delalloc flush. |
| 4444 | */ |
| 4445 | if (flush == BTRFS_RESERVE_FLUSH_LIMIT && |
| 4446 | (flush_state == FLUSH_DELALLOC || |
| 4447 | flush_state == FLUSH_DELALLOC_WAIT)) |
| 4448 | flush_state = ALLOC_CHUNK; |
| 4449 | |
| 4450 | if (!ret) |
| 4451 | goto again; |
| 4452 | else if (flush == BTRFS_RESERVE_FLUSH_LIMIT && |
| 4453 | flush_state < COMMIT_TRANS) |
| 4454 | goto again; |
| 4455 | else if (flush == BTRFS_RESERVE_FLUSH_ALL && |
| 4456 | flush_state <= COMMIT_TRANS) |
| 4457 | goto again; |
| 4458 | |
| 4459 | out: |
| 4460 | if (ret == -ENOSPC && |
| 4461 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { |
| 4462 | struct btrfs_block_rsv *global_rsv = |
| 4463 | &root->fs_info->global_block_rsv; |
| 4464 | |
| 4465 | if (block_rsv != global_rsv && |
| 4466 | !block_rsv_use_bytes(global_rsv, orig_bytes)) |
| 4467 | ret = 0; |
| 4468 | } |
| 4469 | if (ret == -ENOSPC) |
| 4470 | trace_btrfs_space_reservation(root->fs_info, |
| 4471 | "space_info:enospc", |
| 4472 | space_info->flags, orig_bytes, 1); |
| 4473 | if (flushing) { |
| 4474 | spin_lock(&space_info->lock); |
| 4475 | space_info->flush = 0; |
| 4476 | wake_up_all(&space_info->wait); |
| 4477 | spin_unlock(&space_info->lock); |
| 4478 | } |
| 4479 | return ret; |
| 4480 | } |
| 4481 | |
| 4482 | static struct btrfs_block_rsv *get_block_rsv( |
| 4483 | const struct btrfs_trans_handle *trans, |
| 4484 | const struct btrfs_root *root) |
| 4485 | { |
| 4486 | struct btrfs_block_rsv *block_rsv = NULL; |
| 4487 | |
| 4488 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) |
| 4489 | block_rsv = trans->block_rsv; |
| 4490 | |
| 4491 | if (root == root->fs_info->csum_root && trans->adding_csums) |
| 4492 | block_rsv = trans->block_rsv; |
| 4493 | |
| 4494 | if (root == root->fs_info->uuid_root) |
| 4495 | block_rsv = trans->block_rsv; |
| 4496 | |
| 4497 | if (!block_rsv) |
| 4498 | block_rsv = root->block_rsv; |
| 4499 | |
| 4500 | if (!block_rsv) |
| 4501 | block_rsv = &root->fs_info->empty_block_rsv; |
| 4502 | |
| 4503 | return block_rsv; |
| 4504 | } |
| 4505 | |
| 4506 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
| 4507 | u64 num_bytes) |
| 4508 | { |
| 4509 | int ret = -ENOSPC; |
| 4510 | spin_lock(&block_rsv->lock); |
| 4511 | if (block_rsv->reserved >= num_bytes) { |
| 4512 | block_rsv->reserved -= num_bytes; |
| 4513 | if (block_rsv->reserved < block_rsv->size) |
| 4514 | block_rsv->full = 0; |
| 4515 | ret = 0; |
| 4516 | } |
| 4517 | spin_unlock(&block_rsv->lock); |
| 4518 | return ret; |
| 4519 | } |
| 4520 | |
| 4521 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, |
| 4522 | u64 num_bytes, int update_size) |
| 4523 | { |
| 4524 | spin_lock(&block_rsv->lock); |
| 4525 | block_rsv->reserved += num_bytes; |
| 4526 | if (update_size) |
| 4527 | block_rsv->size += num_bytes; |
| 4528 | else if (block_rsv->reserved >= block_rsv->size) |
| 4529 | block_rsv->full = 1; |
| 4530 | spin_unlock(&block_rsv->lock); |
| 4531 | } |
| 4532 | |
| 4533 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
| 4534 | struct btrfs_block_rsv *dest, u64 num_bytes, |
| 4535 | int min_factor) |
| 4536 | { |
| 4537 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
| 4538 | u64 min_bytes; |
| 4539 | |
| 4540 | if (global_rsv->space_info != dest->space_info) |
| 4541 | return -ENOSPC; |
| 4542 | |
| 4543 | spin_lock(&global_rsv->lock); |
| 4544 | min_bytes = div_factor(global_rsv->size, min_factor); |
| 4545 | if (global_rsv->reserved < min_bytes + num_bytes) { |
| 4546 | spin_unlock(&global_rsv->lock); |
| 4547 | return -ENOSPC; |
| 4548 | } |
| 4549 | global_rsv->reserved -= num_bytes; |
| 4550 | if (global_rsv->reserved < global_rsv->size) |
| 4551 | global_rsv->full = 0; |
| 4552 | spin_unlock(&global_rsv->lock); |
| 4553 | |
| 4554 | block_rsv_add_bytes(dest, num_bytes, 1); |
| 4555 | return 0; |
| 4556 | } |
| 4557 | |
| 4558 | static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
| 4559 | struct btrfs_block_rsv *block_rsv, |
| 4560 | struct btrfs_block_rsv *dest, u64 num_bytes) |
| 4561 | { |
| 4562 | struct btrfs_space_info *space_info = block_rsv->space_info; |
| 4563 | |
| 4564 | spin_lock(&block_rsv->lock); |
| 4565 | if (num_bytes == (u64)-1) |
| 4566 | num_bytes = block_rsv->size; |
| 4567 | block_rsv->size -= num_bytes; |
| 4568 | if (block_rsv->reserved >= block_rsv->size) { |
| 4569 | num_bytes = block_rsv->reserved - block_rsv->size; |
| 4570 | block_rsv->reserved = block_rsv->size; |
| 4571 | block_rsv->full = 1; |
| 4572 | } else { |
| 4573 | num_bytes = 0; |
| 4574 | } |
| 4575 | spin_unlock(&block_rsv->lock); |
| 4576 | |
| 4577 | if (num_bytes > 0) { |
| 4578 | if (dest) { |
| 4579 | spin_lock(&dest->lock); |
| 4580 | if (!dest->full) { |
| 4581 | u64 bytes_to_add; |
| 4582 | |
| 4583 | bytes_to_add = dest->size - dest->reserved; |
| 4584 | bytes_to_add = min(num_bytes, bytes_to_add); |
| 4585 | dest->reserved += bytes_to_add; |
| 4586 | if (dest->reserved >= dest->size) |
| 4587 | dest->full = 1; |
| 4588 | num_bytes -= bytes_to_add; |
| 4589 | } |
| 4590 | spin_unlock(&dest->lock); |
| 4591 | } |
| 4592 | if (num_bytes) { |
| 4593 | spin_lock(&space_info->lock); |
| 4594 | space_info->bytes_may_use -= num_bytes; |
| 4595 | trace_btrfs_space_reservation(fs_info, "space_info", |
| 4596 | space_info->flags, num_bytes, 0); |
| 4597 | spin_unlock(&space_info->lock); |
| 4598 | } |
| 4599 | } |
| 4600 | } |
| 4601 | |
| 4602 | static int block_rsv_migrate_bytes(struct btrfs_block_rsv *src, |
| 4603 | struct btrfs_block_rsv *dst, u64 num_bytes) |
| 4604 | { |
| 4605 | int ret; |
| 4606 | |
| 4607 | ret = block_rsv_use_bytes(src, num_bytes); |
| 4608 | if (ret) |
| 4609 | return ret; |
| 4610 | |
| 4611 | block_rsv_add_bytes(dst, num_bytes, 1); |
| 4612 | return 0; |
| 4613 | } |
| 4614 | |
| 4615 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
| 4616 | { |
| 4617 | memset(rsv, 0, sizeof(*rsv)); |
| 4618 | spin_lock_init(&rsv->lock); |
| 4619 | rsv->type = type; |
| 4620 | } |
| 4621 | |
| 4622 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_root *root, |
| 4623 | unsigned short type) |
| 4624 | { |
| 4625 | struct btrfs_block_rsv *block_rsv; |
| 4626 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 4627 | |
| 4628 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
| 4629 | if (!block_rsv) |
| 4630 | return NULL; |
| 4631 | |
| 4632 | btrfs_init_block_rsv(block_rsv, type); |
| 4633 | block_rsv->space_info = __find_space_info(fs_info, |
| 4634 | BTRFS_BLOCK_GROUP_METADATA); |
| 4635 | return block_rsv; |
| 4636 | } |
| 4637 | |
| 4638 | void btrfs_free_block_rsv(struct btrfs_root *root, |
| 4639 | struct btrfs_block_rsv *rsv) |
| 4640 | { |
| 4641 | if (!rsv) |
| 4642 | return; |
| 4643 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
| 4644 | kfree(rsv); |
| 4645 | } |
| 4646 | |
| 4647 | int btrfs_block_rsv_add(struct btrfs_root *root, |
| 4648 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, |
| 4649 | enum btrfs_reserve_flush_enum flush) |
| 4650 | { |
| 4651 | int ret; |
| 4652 | |
| 4653 | if (num_bytes == 0) |
| 4654 | return 0; |
| 4655 | |
| 4656 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
| 4657 | if (!ret) { |
| 4658 | block_rsv_add_bytes(block_rsv, num_bytes, 1); |
| 4659 | return 0; |
| 4660 | } |
| 4661 | |
| 4662 | return ret; |
| 4663 | } |
| 4664 | |
| 4665 | int btrfs_block_rsv_check(struct btrfs_root *root, |
| 4666 | struct btrfs_block_rsv *block_rsv, int min_factor) |
| 4667 | { |
| 4668 | u64 num_bytes = 0; |
| 4669 | int ret = -ENOSPC; |
| 4670 | |
| 4671 | if (!block_rsv) |
| 4672 | return 0; |
| 4673 | |
| 4674 | spin_lock(&block_rsv->lock); |
| 4675 | num_bytes = div_factor(block_rsv->size, min_factor); |
| 4676 | if (block_rsv->reserved >= num_bytes) |
| 4677 | ret = 0; |
| 4678 | spin_unlock(&block_rsv->lock); |
| 4679 | |
| 4680 | return ret; |
| 4681 | } |
| 4682 | |
| 4683 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
| 4684 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, |
| 4685 | enum btrfs_reserve_flush_enum flush) |
| 4686 | { |
| 4687 | u64 num_bytes = 0; |
| 4688 | int ret = -ENOSPC; |
| 4689 | |
| 4690 | if (!block_rsv) |
| 4691 | return 0; |
| 4692 | |
| 4693 | spin_lock(&block_rsv->lock); |
| 4694 | num_bytes = min_reserved; |
| 4695 | if (block_rsv->reserved >= num_bytes) |
| 4696 | ret = 0; |
| 4697 | else |
| 4698 | num_bytes -= block_rsv->reserved; |
| 4699 | spin_unlock(&block_rsv->lock); |
| 4700 | |
| 4701 | if (!ret) |
| 4702 | return 0; |
| 4703 | |
| 4704 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
| 4705 | if (!ret) { |
| 4706 | block_rsv_add_bytes(block_rsv, num_bytes, 0); |
| 4707 | return 0; |
| 4708 | } |
| 4709 | |
| 4710 | return ret; |
| 4711 | } |
| 4712 | |
| 4713 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src_rsv, |
| 4714 | struct btrfs_block_rsv *dst_rsv, |
| 4715 | u64 num_bytes) |
| 4716 | { |
| 4717 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
| 4718 | } |
| 4719 | |
| 4720 | void btrfs_block_rsv_release(struct btrfs_root *root, |
| 4721 | struct btrfs_block_rsv *block_rsv, |
| 4722 | u64 num_bytes) |
| 4723 | { |
| 4724 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
| 4725 | if (global_rsv == block_rsv || |
| 4726 | block_rsv->space_info != global_rsv->space_info) |
| 4727 | global_rsv = NULL; |
| 4728 | block_rsv_release_bytes(root->fs_info, block_rsv, global_rsv, |
| 4729 | num_bytes); |
| 4730 | } |
| 4731 | |
| 4732 | /* |
| 4733 | * helper to calculate size of global block reservation. |
| 4734 | * the desired value is sum of space used by extent tree, |
| 4735 | * checksum tree and root tree |
| 4736 | */ |
| 4737 | static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info) |
| 4738 | { |
| 4739 | struct btrfs_space_info *sinfo; |
| 4740 | u64 num_bytes; |
| 4741 | u64 meta_used; |
| 4742 | u64 data_used; |
| 4743 | int csum_size = btrfs_super_csum_size(fs_info->super_copy); |
| 4744 | |
| 4745 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_DATA); |
| 4746 | spin_lock(&sinfo->lock); |
| 4747 | data_used = sinfo->bytes_used; |
| 4748 | spin_unlock(&sinfo->lock); |
| 4749 | |
| 4750 | sinfo = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
| 4751 | spin_lock(&sinfo->lock); |
| 4752 | if (sinfo->flags & BTRFS_BLOCK_GROUP_DATA) |
| 4753 | data_used = 0; |
| 4754 | meta_used = sinfo->bytes_used; |
| 4755 | spin_unlock(&sinfo->lock); |
| 4756 | |
| 4757 | num_bytes = (data_used >> fs_info->sb->s_blocksize_bits) * |
| 4758 | csum_size * 2; |
| 4759 | num_bytes += div64_u64(data_used + meta_used, 50); |
| 4760 | |
| 4761 | if (num_bytes * 3 > meta_used) |
| 4762 | num_bytes = div64_u64(meta_used, 3); |
| 4763 | |
| 4764 | return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10); |
| 4765 | } |
| 4766 | |
| 4767 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
| 4768 | { |
| 4769 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; |
| 4770 | struct btrfs_space_info *sinfo = block_rsv->space_info; |
| 4771 | u64 num_bytes; |
| 4772 | |
| 4773 | num_bytes = calc_global_metadata_size(fs_info); |
| 4774 | |
| 4775 | spin_lock(&sinfo->lock); |
| 4776 | spin_lock(&block_rsv->lock); |
| 4777 | |
| 4778 | block_rsv->size = min_t(u64, num_bytes, 512 * 1024 * 1024); |
| 4779 | |
| 4780 | num_bytes = sinfo->bytes_used + sinfo->bytes_pinned + |
| 4781 | sinfo->bytes_reserved + sinfo->bytes_readonly + |
| 4782 | sinfo->bytes_may_use; |
| 4783 | |
| 4784 | if (sinfo->total_bytes > num_bytes) { |
| 4785 | num_bytes = sinfo->total_bytes - num_bytes; |
| 4786 | block_rsv->reserved += num_bytes; |
| 4787 | sinfo->bytes_may_use += num_bytes; |
| 4788 | trace_btrfs_space_reservation(fs_info, "space_info", |
| 4789 | sinfo->flags, num_bytes, 1); |
| 4790 | } |
| 4791 | |
| 4792 | if (block_rsv->reserved >= block_rsv->size) { |
| 4793 | num_bytes = block_rsv->reserved - block_rsv->size; |
| 4794 | sinfo->bytes_may_use -= num_bytes; |
| 4795 | trace_btrfs_space_reservation(fs_info, "space_info", |
| 4796 | sinfo->flags, num_bytes, 0); |
| 4797 | block_rsv->reserved = block_rsv->size; |
| 4798 | block_rsv->full = 1; |
| 4799 | } |
| 4800 | |
| 4801 | spin_unlock(&block_rsv->lock); |
| 4802 | spin_unlock(&sinfo->lock); |
| 4803 | } |
| 4804 | |
| 4805 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
| 4806 | { |
| 4807 | struct btrfs_space_info *space_info; |
| 4808 | |
| 4809 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
| 4810 | fs_info->chunk_block_rsv.space_info = space_info; |
| 4811 | |
| 4812 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
| 4813 | fs_info->global_block_rsv.space_info = space_info; |
| 4814 | fs_info->delalloc_block_rsv.space_info = space_info; |
| 4815 | fs_info->trans_block_rsv.space_info = space_info; |
| 4816 | fs_info->empty_block_rsv.space_info = space_info; |
| 4817 | fs_info->delayed_block_rsv.space_info = space_info; |
| 4818 | |
| 4819 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
| 4820 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; |
| 4821 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; |
| 4822 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; |
| 4823 | if (fs_info->quota_root) |
| 4824 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; |
| 4825 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
| 4826 | |
| 4827 | update_global_block_rsv(fs_info); |
| 4828 | } |
| 4829 | |
| 4830 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
| 4831 | { |
| 4832 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
| 4833 | (u64)-1); |
| 4834 | WARN_ON(fs_info->delalloc_block_rsv.size > 0); |
| 4835 | WARN_ON(fs_info->delalloc_block_rsv.reserved > 0); |
| 4836 | WARN_ON(fs_info->trans_block_rsv.size > 0); |
| 4837 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); |
| 4838 | WARN_ON(fs_info->chunk_block_rsv.size > 0); |
| 4839 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); |
| 4840 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
| 4841 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); |
| 4842 | } |
| 4843 | |
| 4844 | void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans, |
| 4845 | struct btrfs_root *root) |
| 4846 | { |
| 4847 | if (!trans->block_rsv) |
| 4848 | return; |
| 4849 | |
| 4850 | if (!trans->bytes_reserved) |
| 4851 | return; |
| 4852 | |
| 4853 | trace_btrfs_space_reservation(root->fs_info, "transaction", |
| 4854 | trans->transid, trans->bytes_reserved, 0); |
| 4855 | btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved); |
| 4856 | trans->bytes_reserved = 0; |
| 4857 | } |
| 4858 | |
| 4859 | /* Can only return 0 or -ENOSPC */ |
| 4860 | int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans, |
| 4861 | struct inode *inode) |
| 4862 | { |
| 4863 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 4864 | struct btrfs_block_rsv *src_rsv = get_block_rsv(trans, root); |
| 4865 | struct btrfs_block_rsv *dst_rsv = root->orphan_block_rsv; |
| 4866 | |
| 4867 | /* |
| 4868 | * We need to hold space in order to delete our orphan item once we've |
| 4869 | * added it, so this takes the reservation so we can release it later |
| 4870 | * when we are truly done with the orphan item. |
| 4871 | */ |
| 4872 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
| 4873 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
| 4874 | btrfs_ino(inode), num_bytes, 1); |
| 4875 | return block_rsv_migrate_bytes(src_rsv, dst_rsv, num_bytes); |
| 4876 | } |
| 4877 | |
| 4878 | void btrfs_orphan_release_metadata(struct inode *inode) |
| 4879 | { |
| 4880 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 4881 | u64 num_bytes = btrfs_calc_trans_metadata_size(root, 1); |
| 4882 | trace_btrfs_space_reservation(root->fs_info, "orphan", |
| 4883 | btrfs_ino(inode), num_bytes, 0); |
| 4884 | btrfs_block_rsv_release(root, root->orphan_block_rsv, num_bytes); |
| 4885 | } |
| 4886 | |
| 4887 | /* |
| 4888 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation |
| 4889 | * root: the root of the parent directory |
| 4890 | * rsv: block reservation |
| 4891 | * items: the number of items that we need do reservation |
| 4892 | * qgroup_reserved: used to return the reserved size in qgroup |
| 4893 | * |
| 4894 | * This function is used to reserve the space for snapshot/subvolume |
| 4895 | * creation and deletion. Those operations are different with the |
| 4896 | * common file/directory operations, they change two fs/file trees |
| 4897 | * and root tree, the number of items that the qgroup reserves is |
| 4898 | * different with the free space reservation. So we can not use |
| 4899 | * the space reseravtion mechanism in start_transaction(). |
| 4900 | */ |
| 4901 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, |
| 4902 | struct btrfs_block_rsv *rsv, |
| 4903 | int items, |
| 4904 | u64 *qgroup_reserved, |
| 4905 | bool use_global_rsv) |
| 4906 | { |
| 4907 | u64 num_bytes; |
| 4908 | int ret; |
| 4909 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
| 4910 | |
| 4911 | if (root->fs_info->quota_enabled) { |
| 4912 | /* One for parent inode, two for dir entries */ |
| 4913 | num_bytes = 3 * root->leafsize; |
| 4914 | ret = btrfs_qgroup_reserve(root, num_bytes); |
| 4915 | if (ret) |
| 4916 | return ret; |
| 4917 | } else { |
| 4918 | num_bytes = 0; |
| 4919 | } |
| 4920 | |
| 4921 | *qgroup_reserved = num_bytes; |
| 4922 | |
| 4923 | num_bytes = btrfs_calc_trans_metadata_size(root, items); |
| 4924 | rsv->space_info = __find_space_info(root->fs_info, |
| 4925 | BTRFS_BLOCK_GROUP_METADATA); |
| 4926 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, |
| 4927 | BTRFS_RESERVE_FLUSH_ALL); |
| 4928 | |
| 4929 | if (ret == -ENOSPC && use_global_rsv) |
| 4930 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes); |
| 4931 | |
| 4932 | if (ret) { |
| 4933 | if (*qgroup_reserved) |
| 4934 | btrfs_qgroup_free(root, *qgroup_reserved); |
| 4935 | } |
| 4936 | |
| 4937 | return ret; |
| 4938 | } |
| 4939 | |
| 4940 | void btrfs_subvolume_release_metadata(struct btrfs_root *root, |
| 4941 | struct btrfs_block_rsv *rsv, |
| 4942 | u64 qgroup_reserved) |
| 4943 | { |
| 4944 | btrfs_block_rsv_release(root, rsv, (u64)-1); |
| 4945 | if (qgroup_reserved) |
| 4946 | btrfs_qgroup_free(root, qgroup_reserved); |
| 4947 | } |
| 4948 | |
| 4949 | /** |
| 4950 | * drop_outstanding_extent - drop an outstanding extent |
| 4951 | * @inode: the inode we're dropping the extent for |
| 4952 | * |
| 4953 | * This is called when we are freeing up an outstanding extent, either called |
| 4954 | * after an error or after an extent is written. This will return the number of |
| 4955 | * reserved extents that need to be freed. This must be called with |
| 4956 | * BTRFS_I(inode)->lock held. |
| 4957 | */ |
| 4958 | static unsigned drop_outstanding_extent(struct inode *inode) |
| 4959 | { |
| 4960 | unsigned drop_inode_space = 0; |
| 4961 | unsigned dropped_extents = 0; |
| 4962 | |
| 4963 | BUG_ON(!BTRFS_I(inode)->outstanding_extents); |
| 4964 | BTRFS_I(inode)->outstanding_extents--; |
| 4965 | |
| 4966 | if (BTRFS_I(inode)->outstanding_extents == 0 && |
| 4967 | test_and_clear_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
| 4968 | &BTRFS_I(inode)->runtime_flags)) |
| 4969 | drop_inode_space = 1; |
| 4970 | |
| 4971 | /* |
| 4972 | * If we have more or the same amount of outsanding extents than we have |
| 4973 | * reserved then we need to leave the reserved extents count alone. |
| 4974 | */ |
| 4975 | if (BTRFS_I(inode)->outstanding_extents >= |
| 4976 | BTRFS_I(inode)->reserved_extents) |
| 4977 | return drop_inode_space; |
| 4978 | |
| 4979 | dropped_extents = BTRFS_I(inode)->reserved_extents - |
| 4980 | BTRFS_I(inode)->outstanding_extents; |
| 4981 | BTRFS_I(inode)->reserved_extents -= dropped_extents; |
| 4982 | return dropped_extents + drop_inode_space; |
| 4983 | } |
| 4984 | |
| 4985 | /** |
| 4986 | * calc_csum_metadata_size - return the amount of metada space that must be |
| 4987 | * reserved/free'd for the given bytes. |
| 4988 | * @inode: the inode we're manipulating |
| 4989 | * @num_bytes: the number of bytes in question |
| 4990 | * @reserve: 1 if we are reserving space, 0 if we are freeing space |
| 4991 | * |
| 4992 | * This adjusts the number of csum_bytes in the inode and then returns the |
| 4993 | * correct amount of metadata that must either be reserved or freed. We |
| 4994 | * calculate how many checksums we can fit into one leaf and then divide the |
| 4995 | * number of bytes that will need to be checksumed by this value to figure out |
| 4996 | * how many checksums will be required. If we are adding bytes then the number |
| 4997 | * may go up and we will return the number of additional bytes that must be |
| 4998 | * reserved. If it is going down we will return the number of bytes that must |
| 4999 | * be freed. |
| 5000 | * |
| 5001 | * This must be called with BTRFS_I(inode)->lock held. |
| 5002 | */ |
| 5003 | static u64 calc_csum_metadata_size(struct inode *inode, u64 num_bytes, |
| 5004 | int reserve) |
| 5005 | { |
| 5006 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 5007 | u64 csum_size; |
| 5008 | int num_csums_per_leaf; |
| 5009 | int num_csums; |
| 5010 | int old_csums; |
| 5011 | |
| 5012 | if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM && |
| 5013 | BTRFS_I(inode)->csum_bytes == 0) |
| 5014 | return 0; |
| 5015 | |
| 5016 | old_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); |
| 5017 | if (reserve) |
| 5018 | BTRFS_I(inode)->csum_bytes += num_bytes; |
| 5019 | else |
| 5020 | BTRFS_I(inode)->csum_bytes -= num_bytes; |
| 5021 | csum_size = BTRFS_LEAF_DATA_SIZE(root) - sizeof(struct btrfs_item); |
| 5022 | num_csums_per_leaf = (int)div64_u64(csum_size, |
| 5023 | sizeof(struct btrfs_csum_item) + |
| 5024 | sizeof(struct btrfs_disk_key)); |
| 5025 | num_csums = (int)div64_u64(BTRFS_I(inode)->csum_bytes, root->sectorsize); |
| 5026 | num_csums = num_csums + num_csums_per_leaf - 1; |
| 5027 | num_csums = num_csums / num_csums_per_leaf; |
| 5028 | |
| 5029 | old_csums = old_csums + num_csums_per_leaf - 1; |
| 5030 | old_csums = old_csums / num_csums_per_leaf; |
| 5031 | |
| 5032 | /* No change, no need to reserve more */ |
| 5033 | if (old_csums == num_csums) |
| 5034 | return 0; |
| 5035 | |
| 5036 | if (reserve) |
| 5037 | return btrfs_calc_trans_metadata_size(root, |
| 5038 | num_csums - old_csums); |
| 5039 | |
| 5040 | return btrfs_calc_trans_metadata_size(root, old_csums - num_csums); |
| 5041 | } |
| 5042 | |
| 5043 | int btrfs_delalloc_reserve_metadata(struct inode *inode, u64 num_bytes) |
| 5044 | { |
| 5045 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 5046 | struct btrfs_block_rsv *block_rsv = &root->fs_info->delalloc_block_rsv; |
| 5047 | u64 to_reserve = 0; |
| 5048 | u64 csum_bytes; |
| 5049 | unsigned nr_extents = 0; |
| 5050 | int extra_reserve = 0; |
| 5051 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
| 5052 | int ret = 0; |
| 5053 | bool delalloc_lock = true; |
| 5054 | u64 to_free = 0; |
| 5055 | unsigned dropped; |
| 5056 | |
| 5057 | /* If we are a free space inode we need to not flush since we will be in |
| 5058 | * the middle of a transaction commit. We also don't need the delalloc |
| 5059 | * mutex since we won't race with anybody. We need this mostly to make |
| 5060 | * lockdep shut its filthy mouth. |
| 5061 | */ |
| 5062 | if (btrfs_is_free_space_inode(inode)) { |
| 5063 | flush = BTRFS_RESERVE_NO_FLUSH; |
| 5064 | delalloc_lock = false; |
| 5065 | } |
| 5066 | |
| 5067 | if (flush != BTRFS_RESERVE_NO_FLUSH && |
| 5068 | btrfs_transaction_in_commit(root->fs_info)) |
| 5069 | schedule_timeout(1); |
| 5070 | |
| 5071 | if (delalloc_lock) |
| 5072 | mutex_lock(&BTRFS_I(inode)->delalloc_mutex); |
| 5073 | |
| 5074 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
| 5075 | |
| 5076 | spin_lock(&BTRFS_I(inode)->lock); |
| 5077 | BTRFS_I(inode)->outstanding_extents++; |
| 5078 | |
| 5079 | if (BTRFS_I(inode)->outstanding_extents > |
| 5080 | BTRFS_I(inode)->reserved_extents) |
| 5081 | nr_extents = BTRFS_I(inode)->outstanding_extents - |
| 5082 | BTRFS_I(inode)->reserved_extents; |
| 5083 | |
| 5084 | /* |
| 5085 | * Add an item to reserve for updating the inode when we complete the |
| 5086 | * delalloc io. |
| 5087 | */ |
| 5088 | if (!test_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
| 5089 | &BTRFS_I(inode)->runtime_flags)) { |
| 5090 | nr_extents++; |
| 5091 | extra_reserve = 1; |
| 5092 | } |
| 5093 | |
| 5094 | to_reserve = btrfs_calc_trans_metadata_size(root, nr_extents); |
| 5095 | to_reserve += calc_csum_metadata_size(inode, num_bytes, 1); |
| 5096 | csum_bytes = BTRFS_I(inode)->csum_bytes; |
| 5097 | spin_unlock(&BTRFS_I(inode)->lock); |
| 5098 | |
| 5099 | if (root->fs_info->quota_enabled) { |
| 5100 | ret = btrfs_qgroup_reserve(root, num_bytes + |
| 5101 | nr_extents * root->leafsize); |
| 5102 | if (ret) |
| 5103 | goto out_fail; |
| 5104 | } |
| 5105 | |
| 5106 | ret = reserve_metadata_bytes(root, block_rsv, to_reserve, flush); |
| 5107 | if (unlikely(ret)) { |
| 5108 | if (root->fs_info->quota_enabled) |
| 5109 | btrfs_qgroup_free(root, num_bytes + |
| 5110 | nr_extents * root->leafsize); |
| 5111 | goto out_fail; |
| 5112 | } |
| 5113 | |
| 5114 | spin_lock(&BTRFS_I(inode)->lock); |
| 5115 | if (extra_reserve) { |
| 5116 | set_bit(BTRFS_INODE_DELALLOC_META_RESERVED, |
| 5117 | &BTRFS_I(inode)->runtime_flags); |
| 5118 | nr_extents--; |
| 5119 | } |
| 5120 | BTRFS_I(inode)->reserved_extents += nr_extents; |
| 5121 | spin_unlock(&BTRFS_I(inode)->lock); |
| 5122 | |
| 5123 | if (delalloc_lock) |
| 5124 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); |
| 5125 | |
| 5126 | if (to_reserve) |
| 5127 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
| 5128 | btrfs_ino(inode), to_reserve, 1); |
| 5129 | block_rsv_add_bytes(block_rsv, to_reserve, 1); |
| 5130 | |
| 5131 | return 0; |
| 5132 | |
| 5133 | out_fail: |
| 5134 | spin_lock(&BTRFS_I(inode)->lock); |
| 5135 | dropped = drop_outstanding_extent(inode); |
| 5136 | /* |
| 5137 | * If the inodes csum_bytes is the same as the original |
| 5138 | * csum_bytes then we know we haven't raced with any free()ers |
| 5139 | * so we can just reduce our inodes csum bytes and carry on. |
| 5140 | */ |
| 5141 | if (BTRFS_I(inode)->csum_bytes == csum_bytes) { |
| 5142 | calc_csum_metadata_size(inode, num_bytes, 0); |
| 5143 | } else { |
| 5144 | u64 orig_csum_bytes = BTRFS_I(inode)->csum_bytes; |
| 5145 | u64 bytes; |
| 5146 | |
| 5147 | /* |
| 5148 | * This is tricky, but first we need to figure out how much we |
| 5149 | * free'd from any free-ers that occured during this |
| 5150 | * reservation, so we reset ->csum_bytes to the csum_bytes |
| 5151 | * before we dropped our lock, and then call the free for the |
| 5152 | * number of bytes that were freed while we were trying our |
| 5153 | * reservation. |
| 5154 | */ |
| 5155 | bytes = csum_bytes - BTRFS_I(inode)->csum_bytes; |
| 5156 | BTRFS_I(inode)->csum_bytes = csum_bytes; |
| 5157 | to_free = calc_csum_metadata_size(inode, bytes, 0); |
| 5158 | |
| 5159 | |
| 5160 | /* |
| 5161 | * Now we need to see how much we would have freed had we not |
| 5162 | * been making this reservation and our ->csum_bytes were not |
| 5163 | * artificially inflated. |
| 5164 | */ |
| 5165 | BTRFS_I(inode)->csum_bytes = csum_bytes - num_bytes; |
| 5166 | bytes = csum_bytes - orig_csum_bytes; |
| 5167 | bytes = calc_csum_metadata_size(inode, bytes, 0); |
| 5168 | |
| 5169 | /* |
| 5170 | * Now reset ->csum_bytes to what it should be. If bytes is |
| 5171 | * more than to_free then we would have free'd more space had we |
| 5172 | * not had an artificially high ->csum_bytes, so we need to free |
| 5173 | * the remainder. If bytes is the same or less then we don't |
| 5174 | * need to do anything, the other free-ers did the correct |
| 5175 | * thing. |
| 5176 | */ |
| 5177 | BTRFS_I(inode)->csum_bytes = orig_csum_bytes - num_bytes; |
| 5178 | if (bytes > to_free) |
| 5179 | to_free = bytes - to_free; |
| 5180 | else |
| 5181 | to_free = 0; |
| 5182 | } |
| 5183 | spin_unlock(&BTRFS_I(inode)->lock); |
| 5184 | if (dropped) |
| 5185 | to_free += btrfs_calc_trans_metadata_size(root, dropped); |
| 5186 | |
| 5187 | if (to_free) { |
| 5188 | btrfs_block_rsv_release(root, block_rsv, to_free); |
| 5189 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
| 5190 | btrfs_ino(inode), to_free, 0); |
| 5191 | } |
| 5192 | if (delalloc_lock) |
| 5193 | mutex_unlock(&BTRFS_I(inode)->delalloc_mutex); |
| 5194 | return ret; |
| 5195 | } |
| 5196 | |
| 5197 | /** |
| 5198 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode |
| 5199 | * @inode: the inode to release the reservation for |
| 5200 | * @num_bytes: the number of bytes we're releasing |
| 5201 | * |
| 5202 | * This will release the metadata reservation for an inode. This can be called |
| 5203 | * once we complete IO for a given set of bytes to release their metadata |
| 5204 | * reservations. |
| 5205 | */ |
| 5206 | void btrfs_delalloc_release_metadata(struct inode *inode, u64 num_bytes) |
| 5207 | { |
| 5208 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 5209 | u64 to_free = 0; |
| 5210 | unsigned dropped; |
| 5211 | |
| 5212 | num_bytes = ALIGN(num_bytes, root->sectorsize); |
| 5213 | spin_lock(&BTRFS_I(inode)->lock); |
| 5214 | dropped = drop_outstanding_extent(inode); |
| 5215 | |
| 5216 | if (num_bytes) |
| 5217 | to_free = calc_csum_metadata_size(inode, num_bytes, 0); |
| 5218 | spin_unlock(&BTRFS_I(inode)->lock); |
| 5219 | if (dropped > 0) |
| 5220 | to_free += btrfs_calc_trans_metadata_size(root, dropped); |
| 5221 | |
| 5222 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
| 5223 | btrfs_ino(inode), to_free, 0); |
| 5224 | if (root->fs_info->quota_enabled) { |
| 5225 | btrfs_qgroup_free(root, num_bytes + |
| 5226 | dropped * root->leafsize); |
| 5227 | } |
| 5228 | |
| 5229 | btrfs_block_rsv_release(root, &root->fs_info->delalloc_block_rsv, |
| 5230 | to_free); |
| 5231 | } |
| 5232 | |
| 5233 | /** |
| 5234 | * btrfs_delalloc_reserve_space - reserve data and metadata space for delalloc |
| 5235 | * @inode: inode we're writing to |
| 5236 | * @num_bytes: the number of bytes we want to allocate |
| 5237 | * |
| 5238 | * This will do the following things |
| 5239 | * |
| 5240 | * o reserve space in the data space info for num_bytes |
| 5241 | * o reserve space in the metadata space info based on number of outstanding |
| 5242 | * extents and how much csums will be needed |
| 5243 | * o add to the inodes ->delalloc_bytes |
| 5244 | * o add it to the fs_info's delalloc inodes list. |
| 5245 | * |
| 5246 | * This will return 0 for success and -ENOSPC if there is no space left. |
| 5247 | */ |
| 5248 | int btrfs_delalloc_reserve_space(struct inode *inode, u64 num_bytes) |
| 5249 | { |
| 5250 | int ret; |
| 5251 | |
| 5252 | ret = btrfs_check_data_free_space(inode, num_bytes); |
| 5253 | if (ret) |
| 5254 | return ret; |
| 5255 | |
| 5256 | ret = btrfs_delalloc_reserve_metadata(inode, num_bytes); |
| 5257 | if (ret) { |
| 5258 | btrfs_free_reserved_data_space(inode, num_bytes); |
| 5259 | return ret; |
| 5260 | } |
| 5261 | |
| 5262 | return 0; |
| 5263 | } |
| 5264 | |
| 5265 | /** |
| 5266 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
| 5267 | * @inode: inode we're releasing space for |
| 5268 | * @num_bytes: the number of bytes we want to free up |
| 5269 | * |
| 5270 | * This must be matched with a call to btrfs_delalloc_reserve_space. This is |
| 5271 | * called in the case that we don't need the metadata AND data reservations |
| 5272 | * anymore. So if there is an error or we insert an inline extent. |
| 5273 | * |
| 5274 | * This function will release the metadata space that was not used and will |
| 5275 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes |
| 5276 | * list if there are no delalloc bytes left. |
| 5277 | */ |
| 5278 | void btrfs_delalloc_release_space(struct inode *inode, u64 num_bytes) |
| 5279 | { |
| 5280 | btrfs_delalloc_release_metadata(inode, num_bytes); |
| 5281 | btrfs_free_reserved_data_space(inode, num_bytes); |
| 5282 | } |
| 5283 | |
| 5284 | static int update_block_group(struct btrfs_root *root, |
| 5285 | u64 bytenr, u64 num_bytes, int alloc) |
| 5286 | { |
| 5287 | struct btrfs_block_group_cache *cache = NULL; |
| 5288 | struct btrfs_fs_info *info = root->fs_info; |
| 5289 | u64 total = num_bytes; |
| 5290 | u64 old_val; |
| 5291 | u64 byte_in_group; |
| 5292 | int factor; |
| 5293 | |
| 5294 | /* block accounting for super block */ |
| 5295 | spin_lock(&info->delalloc_root_lock); |
| 5296 | old_val = btrfs_super_bytes_used(info->super_copy); |
| 5297 | if (alloc) |
| 5298 | old_val += num_bytes; |
| 5299 | else |
| 5300 | old_val -= num_bytes; |
| 5301 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
| 5302 | spin_unlock(&info->delalloc_root_lock); |
| 5303 | |
| 5304 | while (total) { |
| 5305 | cache = btrfs_lookup_block_group(info, bytenr); |
| 5306 | if (!cache) |
| 5307 | return -ENOENT; |
| 5308 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
| 5309 | BTRFS_BLOCK_GROUP_RAID1 | |
| 5310 | BTRFS_BLOCK_GROUP_RAID10)) |
| 5311 | factor = 2; |
| 5312 | else |
| 5313 | factor = 1; |
| 5314 | /* |
| 5315 | * If this block group has free space cache written out, we |
| 5316 | * need to make sure to load it if we are removing space. This |
| 5317 | * is because we need the unpinning stage to actually add the |
| 5318 | * space back to the block group, otherwise we will leak space. |
| 5319 | */ |
| 5320 | if (!alloc && cache->cached == BTRFS_CACHE_NO) |
| 5321 | cache_block_group(cache, 1); |
| 5322 | |
| 5323 | byte_in_group = bytenr - cache->key.objectid; |
| 5324 | WARN_ON(byte_in_group > cache->key.offset); |
| 5325 | |
| 5326 | spin_lock(&cache->space_info->lock); |
| 5327 | spin_lock(&cache->lock); |
| 5328 | |
| 5329 | if (btrfs_test_opt(root, SPACE_CACHE) && |
| 5330 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
| 5331 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
| 5332 | |
| 5333 | cache->dirty = 1; |
| 5334 | old_val = btrfs_block_group_used(&cache->item); |
| 5335 | num_bytes = min(total, cache->key.offset - byte_in_group); |
| 5336 | if (alloc) { |
| 5337 | old_val += num_bytes; |
| 5338 | btrfs_set_block_group_used(&cache->item, old_val); |
| 5339 | cache->reserved -= num_bytes; |
| 5340 | cache->space_info->bytes_reserved -= num_bytes; |
| 5341 | cache->space_info->bytes_used += num_bytes; |
| 5342 | cache->space_info->disk_used += num_bytes * factor; |
| 5343 | spin_unlock(&cache->lock); |
| 5344 | spin_unlock(&cache->space_info->lock); |
| 5345 | } else { |
| 5346 | old_val -= num_bytes; |
| 5347 | btrfs_set_block_group_used(&cache->item, old_val); |
| 5348 | cache->pinned += num_bytes; |
| 5349 | cache->space_info->bytes_pinned += num_bytes; |
| 5350 | cache->space_info->bytes_used -= num_bytes; |
| 5351 | cache->space_info->disk_used -= num_bytes * factor; |
| 5352 | spin_unlock(&cache->lock); |
| 5353 | spin_unlock(&cache->space_info->lock); |
| 5354 | |
| 5355 | set_extent_dirty(info->pinned_extents, |
| 5356 | bytenr, bytenr + num_bytes - 1, |
| 5357 | GFP_NOFS | __GFP_NOFAIL); |
| 5358 | } |
| 5359 | btrfs_put_block_group(cache); |
| 5360 | total -= num_bytes; |
| 5361 | bytenr += num_bytes; |
| 5362 | } |
| 5363 | return 0; |
| 5364 | } |
| 5365 | |
| 5366 | static u64 first_logical_byte(struct btrfs_root *root, u64 search_start) |
| 5367 | { |
| 5368 | struct btrfs_block_group_cache *cache; |
| 5369 | u64 bytenr; |
| 5370 | |
| 5371 | spin_lock(&root->fs_info->block_group_cache_lock); |
| 5372 | bytenr = root->fs_info->first_logical_byte; |
| 5373 | spin_unlock(&root->fs_info->block_group_cache_lock); |
| 5374 | |
| 5375 | if (bytenr < (u64)-1) |
| 5376 | return bytenr; |
| 5377 | |
| 5378 | cache = btrfs_lookup_first_block_group(root->fs_info, search_start); |
| 5379 | if (!cache) |
| 5380 | return 0; |
| 5381 | |
| 5382 | bytenr = cache->key.objectid; |
| 5383 | btrfs_put_block_group(cache); |
| 5384 | |
| 5385 | return bytenr; |
| 5386 | } |
| 5387 | |
| 5388 | static int pin_down_extent(struct btrfs_root *root, |
| 5389 | struct btrfs_block_group_cache *cache, |
| 5390 | u64 bytenr, u64 num_bytes, int reserved) |
| 5391 | { |
| 5392 | spin_lock(&cache->space_info->lock); |
| 5393 | spin_lock(&cache->lock); |
| 5394 | cache->pinned += num_bytes; |
| 5395 | cache->space_info->bytes_pinned += num_bytes; |
| 5396 | if (reserved) { |
| 5397 | cache->reserved -= num_bytes; |
| 5398 | cache->space_info->bytes_reserved -= num_bytes; |
| 5399 | } |
| 5400 | spin_unlock(&cache->lock); |
| 5401 | spin_unlock(&cache->space_info->lock); |
| 5402 | |
| 5403 | set_extent_dirty(root->fs_info->pinned_extents, bytenr, |
| 5404 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
| 5405 | if (reserved) |
| 5406 | trace_btrfs_reserved_extent_free(root, bytenr, num_bytes); |
| 5407 | return 0; |
| 5408 | } |
| 5409 | |
| 5410 | /* |
| 5411 | * this function must be called within transaction |
| 5412 | */ |
| 5413 | int btrfs_pin_extent(struct btrfs_root *root, |
| 5414 | u64 bytenr, u64 num_bytes, int reserved) |
| 5415 | { |
| 5416 | struct btrfs_block_group_cache *cache; |
| 5417 | |
| 5418 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
| 5419 | BUG_ON(!cache); /* Logic error */ |
| 5420 | |
| 5421 | pin_down_extent(root, cache, bytenr, num_bytes, reserved); |
| 5422 | |
| 5423 | btrfs_put_block_group(cache); |
| 5424 | return 0; |
| 5425 | } |
| 5426 | |
| 5427 | /* |
| 5428 | * this function must be called within transaction |
| 5429 | */ |
| 5430 | int btrfs_pin_extent_for_log_replay(struct btrfs_root *root, |
| 5431 | u64 bytenr, u64 num_bytes) |
| 5432 | { |
| 5433 | struct btrfs_block_group_cache *cache; |
| 5434 | int ret; |
| 5435 | |
| 5436 | cache = btrfs_lookup_block_group(root->fs_info, bytenr); |
| 5437 | if (!cache) |
| 5438 | return -EINVAL; |
| 5439 | |
| 5440 | /* |
| 5441 | * pull in the free space cache (if any) so that our pin |
| 5442 | * removes the free space from the cache. We have load_only set |
| 5443 | * to one because the slow code to read in the free extents does check |
| 5444 | * the pinned extents. |
| 5445 | */ |
| 5446 | cache_block_group(cache, 1); |
| 5447 | |
| 5448 | pin_down_extent(root, cache, bytenr, num_bytes, 0); |
| 5449 | |
| 5450 | /* remove us from the free space cache (if we're there at all) */ |
| 5451 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
| 5452 | btrfs_put_block_group(cache); |
| 5453 | return ret; |
| 5454 | } |
| 5455 | |
| 5456 | static int __exclude_logged_extent(struct btrfs_root *root, u64 start, u64 num_bytes) |
| 5457 | { |
| 5458 | int ret; |
| 5459 | struct btrfs_block_group_cache *block_group; |
| 5460 | struct btrfs_caching_control *caching_ctl; |
| 5461 | |
| 5462 | block_group = btrfs_lookup_block_group(root->fs_info, start); |
| 5463 | if (!block_group) |
| 5464 | return -EINVAL; |
| 5465 | |
| 5466 | cache_block_group(block_group, 0); |
| 5467 | caching_ctl = get_caching_control(block_group); |
| 5468 | |
| 5469 | if (!caching_ctl) { |
| 5470 | /* Logic error */ |
| 5471 | BUG_ON(!block_group_cache_done(block_group)); |
| 5472 | ret = btrfs_remove_free_space(block_group, start, num_bytes); |
| 5473 | } else { |
| 5474 | mutex_lock(&caching_ctl->mutex); |
| 5475 | |
| 5476 | if (start >= caching_ctl->progress) { |
| 5477 | ret = add_excluded_extent(root, start, num_bytes); |
| 5478 | } else if (start + num_bytes <= caching_ctl->progress) { |
| 5479 | ret = btrfs_remove_free_space(block_group, |
| 5480 | start, num_bytes); |
| 5481 | } else { |
| 5482 | num_bytes = caching_ctl->progress - start; |
| 5483 | ret = btrfs_remove_free_space(block_group, |
| 5484 | start, num_bytes); |
| 5485 | if (ret) |
| 5486 | goto out_lock; |
| 5487 | |
| 5488 | num_bytes = (start + num_bytes) - |
| 5489 | caching_ctl->progress; |
| 5490 | start = caching_ctl->progress; |
| 5491 | ret = add_excluded_extent(root, start, num_bytes); |
| 5492 | } |
| 5493 | out_lock: |
| 5494 | mutex_unlock(&caching_ctl->mutex); |
| 5495 | put_caching_control(caching_ctl); |
| 5496 | } |
| 5497 | btrfs_put_block_group(block_group); |
| 5498 | return ret; |
| 5499 | } |
| 5500 | |
| 5501 | int btrfs_exclude_logged_extents(struct btrfs_root *log, |
| 5502 | struct extent_buffer *eb) |
| 5503 | { |
| 5504 | struct btrfs_file_extent_item *item; |
| 5505 | struct btrfs_key key; |
| 5506 | int found_type; |
| 5507 | int i; |
| 5508 | |
| 5509 | if (!btrfs_fs_incompat(log->fs_info, MIXED_GROUPS)) |
| 5510 | return 0; |
| 5511 | |
| 5512 | for (i = 0; i < btrfs_header_nritems(eb); i++) { |
| 5513 | btrfs_item_key_to_cpu(eb, &key, i); |
| 5514 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
| 5515 | continue; |
| 5516 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); |
| 5517 | found_type = btrfs_file_extent_type(eb, item); |
| 5518 | if (found_type == BTRFS_FILE_EXTENT_INLINE) |
| 5519 | continue; |
| 5520 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) |
| 5521 | continue; |
| 5522 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); |
| 5523 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); |
| 5524 | __exclude_logged_extent(log, key.objectid, key.offset); |
| 5525 | } |
| 5526 | |
| 5527 | return 0; |
| 5528 | } |
| 5529 | |
| 5530 | /** |
| 5531 | * btrfs_update_reserved_bytes - update the block_group and space info counters |
| 5532 | * @cache: The cache we are manipulating |
| 5533 | * @num_bytes: The number of bytes in question |
| 5534 | * @reserve: One of the reservation enums |
| 5535 | * |
| 5536 | * This is called by the allocator when it reserves space, or by somebody who is |
| 5537 | * freeing space that was never actually used on disk. For example if you |
| 5538 | * reserve some space for a new leaf in transaction A and before transaction A |
| 5539 | * commits you free that leaf, you call this with reserve set to 0 in order to |
| 5540 | * clear the reservation. |
| 5541 | * |
| 5542 | * Metadata reservations should be called with RESERVE_ALLOC so we do the proper |
| 5543 | * ENOSPC accounting. For data we handle the reservation through clearing the |
| 5544 | * delalloc bits in the io_tree. We have to do this since we could end up |
| 5545 | * allocating less disk space for the amount of data we have reserved in the |
| 5546 | * case of compression. |
| 5547 | * |
| 5548 | * If this is a reservation and the block group has become read only we cannot |
| 5549 | * make the reservation and return -EAGAIN, otherwise this function always |
| 5550 | * succeeds. |
| 5551 | */ |
| 5552 | static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache, |
| 5553 | u64 num_bytes, int reserve) |
| 5554 | { |
| 5555 | struct btrfs_space_info *space_info = cache->space_info; |
| 5556 | int ret = 0; |
| 5557 | |
| 5558 | spin_lock(&space_info->lock); |
| 5559 | spin_lock(&cache->lock); |
| 5560 | if (reserve != RESERVE_FREE) { |
| 5561 | if (cache->ro) { |
| 5562 | ret = -EAGAIN; |
| 5563 | } else { |
| 5564 | cache->reserved += num_bytes; |
| 5565 | space_info->bytes_reserved += num_bytes; |
| 5566 | if (reserve == RESERVE_ALLOC) { |
| 5567 | trace_btrfs_space_reservation(cache->fs_info, |
| 5568 | "space_info", space_info->flags, |
| 5569 | num_bytes, 0); |
| 5570 | space_info->bytes_may_use -= num_bytes; |
| 5571 | } |
| 5572 | } |
| 5573 | } else { |
| 5574 | if (cache->ro) |
| 5575 | space_info->bytes_readonly += num_bytes; |
| 5576 | cache->reserved -= num_bytes; |
| 5577 | space_info->bytes_reserved -= num_bytes; |
| 5578 | } |
| 5579 | spin_unlock(&cache->lock); |
| 5580 | spin_unlock(&space_info->lock); |
| 5581 | return ret; |
| 5582 | } |
| 5583 | |
| 5584 | void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans, |
| 5585 | struct btrfs_root *root) |
| 5586 | { |
| 5587 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 5588 | struct btrfs_caching_control *next; |
| 5589 | struct btrfs_caching_control *caching_ctl; |
| 5590 | struct btrfs_block_group_cache *cache; |
| 5591 | struct btrfs_space_info *space_info; |
| 5592 | |
| 5593 | down_write(&fs_info->commit_root_sem); |
| 5594 | |
| 5595 | list_for_each_entry_safe(caching_ctl, next, |
| 5596 | &fs_info->caching_block_groups, list) { |
| 5597 | cache = caching_ctl->block_group; |
| 5598 | if (block_group_cache_done(cache)) { |
| 5599 | cache->last_byte_to_unpin = (u64)-1; |
| 5600 | list_del_init(&caching_ctl->list); |
| 5601 | put_caching_control(caching_ctl); |
| 5602 | } else { |
| 5603 | cache->last_byte_to_unpin = caching_ctl->progress; |
| 5604 | } |
| 5605 | } |
| 5606 | |
| 5607 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
| 5608 | fs_info->pinned_extents = &fs_info->freed_extents[1]; |
| 5609 | else |
| 5610 | fs_info->pinned_extents = &fs_info->freed_extents[0]; |
| 5611 | |
| 5612 | up_write(&fs_info->commit_root_sem); |
| 5613 | |
| 5614 | list_for_each_entry_rcu(space_info, &fs_info->space_info, list) |
| 5615 | percpu_counter_set(&space_info->total_bytes_pinned, 0); |
| 5616 | |
| 5617 | update_global_block_rsv(fs_info); |
| 5618 | } |
| 5619 | |
| 5620 | static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
| 5621 | { |
| 5622 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 5623 | struct btrfs_block_group_cache *cache = NULL; |
| 5624 | struct btrfs_space_info *space_info; |
| 5625 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
| 5626 | u64 len; |
| 5627 | bool readonly; |
| 5628 | |
| 5629 | while (start <= end) { |
| 5630 | readonly = false; |
| 5631 | if (!cache || |
| 5632 | start >= cache->key.objectid + cache->key.offset) { |
| 5633 | if (cache) |
| 5634 | btrfs_put_block_group(cache); |
| 5635 | cache = btrfs_lookup_block_group(fs_info, start); |
| 5636 | BUG_ON(!cache); /* Logic error */ |
| 5637 | } |
| 5638 | |
| 5639 | len = cache->key.objectid + cache->key.offset - start; |
| 5640 | len = min(len, end + 1 - start); |
| 5641 | |
| 5642 | if (start < cache->last_byte_to_unpin) { |
| 5643 | len = min(len, cache->last_byte_to_unpin - start); |
| 5644 | btrfs_add_free_space(cache, start, len); |
| 5645 | } |
| 5646 | |
| 5647 | start += len; |
| 5648 | space_info = cache->space_info; |
| 5649 | |
| 5650 | spin_lock(&space_info->lock); |
| 5651 | spin_lock(&cache->lock); |
| 5652 | cache->pinned -= len; |
| 5653 | space_info->bytes_pinned -= len; |
| 5654 | if (cache->ro) { |
| 5655 | space_info->bytes_readonly += len; |
| 5656 | readonly = true; |
| 5657 | } |
| 5658 | spin_unlock(&cache->lock); |
| 5659 | if (!readonly && global_rsv->space_info == space_info) { |
| 5660 | spin_lock(&global_rsv->lock); |
| 5661 | if (!global_rsv->full) { |
| 5662 | len = min(len, global_rsv->size - |
| 5663 | global_rsv->reserved); |
| 5664 | global_rsv->reserved += len; |
| 5665 | space_info->bytes_may_use += len; |
| 5666 | if (global_rsv->reserved >= global_rsv->size) |
| 5667 | global_rsv->full = 1; |
| 5668 | } |
| 5669 | spin_unlock(&global_rsv->lock); |
| 5670 | } |
| 5671 | spin_unlock(&space_info->lock); |
| 5672 | } |
| 5673 | |
| 5674 | if (cache) |
| 5675 | btrfs_put_block_group(cache); |
| 5676 | return 0; |
| 5677 | } |
| 5678 | |
| 5679 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, |
| 5680 | struct btrfs_root *root) |
| 5681 | { |
| 5682 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 5683 | struct extent_io_tree *unpin; |
| 5684 | u64 start; |
| 5685 | u64 end; |
| 5686 | int ret; |
| 5687 | |
| 5688 | if (trans->aborted) |
| 5689 | return 0; |
| 5690 | |
| 5691 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
| 5692 | unpin = &fs_info->freed_extents[1]; |
| 5693 | else |
| 5694 | unpin = &fs_info->freed_extents[0]; |
| 5695 | |
| 5696 | while (1) { |
| 5697 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
| 5698 | EXTENT_DIRTY, NULL); |
| 5699 | if (ret) |
| 5700 | break; |
| 5701 | |
| 5702 | if (btrfs_test_opt(root, DISCARD)) |
| 5703 | ret = btrfs_discard_extent(root, start, |
| 5704 | end + 1 - start, NULL); |
| 5705 | |
| 5706 | clear_extent_dirty(unpin, start, end, GFP_NOFS); |
| 5707 | unpin_extent_range(root, start, end); |
| 5708 | cond_resched(); |
| 5709 | } |
| 5710 | |
| 5711 | return 0; |
| 5712 | } |
| 5713 | |
| 5714 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, u64 num_bytes, |
| 5715 | u64 owner, u64 root_objectid) |
| 5716 | { |
| 5717 | struct btrfs_space_info *space_info; |
| 5718 | u64 flags; |
| 5719 | |
| 5720 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
| 5721 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) |
| 5722 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
| 5723 | else |
| 5724 | flags = BTRFS_BLOCK_GROUP_METADATA; |
| 5725 | } else { |
| 5726 | flags = BTRFS_BLOCK_GROUP_DATA; |
| 5727 | } |
| 5728 | |
| 5729 | space_info = __find_space_info(fs_info, flags); |
| 5730 | BUG_ON(!space_info); /* Logic bug */ |
| 5731 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); |
| 5732 | } |
| 5733 | |
| 5734 | |
| 5735 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
| 5736 | struct btrfs_root *root, |
| 5737 | u64 bytenr, u64 num_bytes, u64 parent, |
| 5738 | u64 root_objectid, u64 owner_objectid, |
| 5739 | u64 owner_offset, int refs_to_drop, |
| 5740 | struct btrfs_delayed_extent_op *extent_op, |
| 5741 | int no_quota) |
| 5742 | { |
| 5743 | struct btrfs_key key; |
| 5744 | struct btrfs_path *path; |
| 5745 | struct btrfs_fs_info *info = root->fs_info; |
| 5746 | struct btrfs_root *extent_root = info->extent_root; |
| 5747 | struct extent_buffer *leaf; |
| 5748 | struct btrfs_extent_item *ei; |
| 5749 | struct btrfs_extent_inline_ref *iref; |
| 5750 | int ret; |
| 5751 | int is_data; |
| 5752 | int extent_slot = 0; |
| 5753 | int found_extent = 0; |
| 5754 | int num_to_del = 1; |
| 5755 | u32 item_size; |
| 5756 | u64 refs; |
| 5757 | int last_ref = 0; |
| 5758 | enum btrfs_qgroup_operation_type type = BTRFS_QGROUP_OPER_SUB_EXCL; |
| 5759 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
| 5760 | SKINNY_METADATA); |
| 5761 | |
| 5762 | if (!info->quota_enabled || !is_fstree(root_objectid)) |
| 5763 | no_quota = 1; |
| 5764 | |
| 5765 | path = btrfs_alloc_path(); |
| 5766 | if (!path) |
| 5767 | return -ENOMEM; |
| 5768 | |
| 5769 | path->reada = 1; |
| 5770 | path->leave_spinning = 1; |
| 5771 | |
| 5772 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; |
| 5773 | BUG_ON(!is_data && refs_to_drop != 1); |
| 5774 | |
| 5775 | if (is_data) |
| 5776 | skinny_metadata = 0; |
| 5777 | |
| 5778 | ret = lookup_extent_backref(trans, extent_root, path, &iref, |
| 5779 | bytenr, num_bytes, parent, |
| 5780 | root_objectid, owner_objectid, |
| 5781 | owner_offset); |
| 5782 | if (ret == 0) { |
| 5783 | extent_slot = path->slots[0]; |
| 5784 | while (extent_slot >= 0) { |
| 5785 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
| 5786 | extent_slot); |
| 5787 | if (key.objectid != bytenr) |
| 5788 | break; |
| 5789 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
| 5790 | key.offset == num_bytes) { |
| 5791 | found_extent = 1; |
| 5792 | break; |
| 5793 | } |
| 5794 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
| 5795 | key.offset == owner_objectid) { |
| 5796 | found_extent = 1; |
| 5797 | break; |
| 5798 | } |
| 5799 | if (path->slots[0] - extent_slot > 5) |
| 5800 | break; |
| 5801 | extent_slot--; |
| 5802 | } |
| 5803 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 5804 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); |
| 5805 | if (found_extent && item_size < sizeof(*ei)) |
| 5806 | found_extent = 0; |
| 5807 | #endif |
| 5808 | if (!found_extent) { |
| 5809 | BUG_ON(iref); |
| 5810 | ret = remove_extent_backref(trans, extent_root, path, |
| 5811 | NULL, refs_to_drop, |
| 5812 | is_data, &last_ref); |
| 5813 | if (ret) { |
| 5814 | btrfs_abort_transaction(trans, extent_root, ret); |
| 5815 | goto out; |
| 5816 | } |
| 5817 | btrfs_release_path(path); |
| 5818 | path->leave_spinning = 1; |
| 5819 | |
| 5820 | key.objectid = bytenr; |
| 5821 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 5822 | key.offset = num_bytes; |
| 5823 | |
| 5824 | if (!is_data && skinny_metadata) { |
| 5825 | key.type = BTRFS_METADATA_ITEM_KEY; |
| 5826 | key.offset = owner_objectid; |
| 5827 | } |
| 5828 | |
| 5829 | ret = btrfs_search_slot(trans, extent_root, |
| 5830 | &key, path, -1, 1); |
| 5831 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
| 5832 | /* |
| 5833 | * Couldn't find our skinny metadata item, |
| 5834 | * see if we have ye olde extent item. |
| 5835 | */ |
| 5836 | path->slots[0]--; |
| 5837 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
| 5838 | path->slots[0]); |
| 5839 | if (key.objectid == bytenr && |
| 5840 | key.type == BTRFS_EXTENT_ITEM_KEY && |
| 5841 | key.offset == num_bytes) |
| 5842 | ret = 0; |
| 5843 | } |
| 5844 | |
| 5845 | if (ret > 0 && skinny_metadata) { |
| 5846 | skinny_metadata = false; |
| 5847 | key.objectid = bytenr; |
| 5848 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 5849 | key.offset = num_bytes; |
| 5850 | btrfs_release_path(path); |
| 5851 | ret = btrfs_search_slot(trans, extent_root, |
| 5852 | &key, path, -1, 1); |
| 5853 | } |
| 5854 | |
| 5855 | if (ret) { |
| 5856 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
| 5857 | ret, bytenr); |
| 5858 | if (ret > 0) |
| 5859 | btrfs_print_leaf(extent_root, |
| 5860 | path->nodes[0]); |
| 5861 | } |
| 5862 | if (ret < 0) { |
| 5863 | btrfs_abort_transaction(trans, extent_root, ret); |
| 5864 | goto out; |
| 5865 | } |
| 5866 | extent_slot = path->slots[0]; |
| 5867 | } |
| 5868 | } else if (WARN_ON(ret == -ENOENT)) { |
| 5869 | btrfs_print_leaf(extent_root, path->nodes[0]); |
| 5870 | btrfs_err(info, |
| 5871 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", |
| 5872 | bytenr, parent, root_objectid, owner_objectid, |
| 5873 | owner_offset); |
| 5874 | btrfs_abort_transaction(trans, extent_root, ret); |
| 5875 | goto out; |
| 5876 | } else { |
| 5877 | btrfs_abort_transaction(trans, extent_root, ret); |
| 5878 | goto out; |
| 5879 | } |
| 5880 | |
| 5881 | leaf = path->nodes[0]; |
| 5882 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
| 5883 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 5884 | if (item_size < sizeof(*ei)) { |
| 5885 | BUG_ON(found_extent || extent_slot != path->slots[0]); |
| 5886 | ret = convert_extent_item_v0(trans, extent_root, path, |
| 5887 | owner_objectid, 0); |
| 5888 | if (ret < 0) { |
| 5889 | btrfs_abort_transaction(trans, extent_root, ret); |
| 5890 | goto out; |
| 5891 | } |
| 5892 | |
| 5893 | btrfs_release_path(path); |
| 5894 | path->leave_spinning = 1; |
| 5895 | |
| 5896 | key.objectid = bytenr; |
| 5897 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 5898 | key.offset = num_bytes; |
| 5899 | |
| 5900 | ret = btrfs_search_slot(trans, extent_root, &key, path, |
| 5901 | -1, 1); |
| 5902 | if (ret) { |
| 5903 | btrfs_err(info, "umm, got %d back from search, was looking for %llu", |
| 5904 | ret, bytenr); |
| 5905 | btrfs_print_leaf(extent_root, path->nodes[0]); |
| 5906 | } |
| 5907 | if (ret < 0) { |
| 5908 | btrfs_abort_transaction(trans, extent_root, ret); |
| 5909 | goto out; |
| 5910 | } |
| 5911 | |
| 5912 | extent_slot = path->slots[0]; |
| 5913 | leaf = path->nodes[0]; |
| 5914 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
| 5915 | } |
| 5916 | #endif |
| 5917 | BUG_ON(item_size < sizeof(*ei)); |
| 5918 | ei = btrfs_item_ptr(leaf, extent_slot, |
| 5919 | struct btrfs_extent_item); |
| 5920 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
| 5921 | key.type == BTRFS_EXTENT_ITEM_KEY) { |
| 5922 | struct btrfs_tree_block_info *bi; |
| 5923 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); |
| 5924 | bi = (struct btrfs_tree_block_info *)(ei + 1); |
| 5925 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); |
| 5926 | } |
| 5927 | |
| 5928 | refs = btrfs_extent_refs(leaf, ei); |
| 5929 | if (refs < refs_to_drop) { |
| 5930 | btrfs_err(info, "trying to drop %d refs but we only have %Lu " |
| 5931 | "for bytenr %Lu\n", refs_to_drop, refs, bytenr); |
| 5932 | ret = -EINVAL; |
| 5933 | btrfs_abort_transaction(trans, extent_root, ret); |
| 5934 | goto out; |
| 5935 | } |
| 5936 | refs -= refs_to_drop; |
| 5937 | |
| 5938 | if (refs > 0) { |
| 5939 | type = BTRFS_QGROUP_OPER_SUB_SHARED; |
| 5940 | if (extent_op) |
| 5941 | __run_delayed_extent_op(extent_op, leaf, ei); |
| 5942 | /* |
| 5943 | * In the case of inline back ref, reference count will |
| 5944 | * be updated by remove_extent_backref |
| 5945 | */ |
| 5946 | if (iref) { |
| 5947 | BUG_ON(!found_extent); |
| 5948 | } else { |
| 5949 | btrfs_set_extent_refs(leaf, ei, refs); |
| 5950 | btrfs_mark_buffer_dirty(leaf); |
| 5951 | } |
| 5952 | if (found_extent) { |
| 5953 | ret = remove_extent_backref(trans, extent_root, path, |
| 5954 | iref, refs_to_drop, |
| 5955 | is_data, &last_ref); |
| 5956 | if (ret) { |
| 5957 | btrfs_abort_transaction(trans, extent_root, ret); |
| 5958 | goto out; |
| 5959 | } |
| 5960 | } |
| 5961 | add_pinned_bytes(root->fs_info, -num_bytes, owner_objectid, |
| 5962 | root_objectid); |
| 5963 | } else { |
| 5964 | if (found_extent) { |
| 5965 | BUG_ON(is_data && refs_to_drop != |
| 5966 | extent_data_ref_count(root, path, iref)); |
| 5967 | if (iref) { |
| 5968 | BUG_ON(path->slots[0] != extent_slot); |
| 5969 | } else { |
| 5970 | BUG_ON(path->slots[0] != extent_slot + 1); |
| 5971 | path->slots[0] = extent_slot; |
| 5972 | num_to_del = 2; |
| 5973 | } |
| 5974 | } |
| 5975 | |
| 5976 | last_ref = 1; |
| 5977 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
| 5978 | num_to_del); |
| 5979 | if (ret) { |
| 5980 | btrfs_abort_transaction(trans, extent_root, ret); |
| 5981 | goto out; |
| 5982 | } |
| 5983 | btrfs_release_path(path); |
| 5984 | |
| 5985 | if (is_data) { |
| 5986 | ret = btrfs_del_csums(trans, root, bytenr, num_bytes); |
| 5987 | if (ret) { |
| 5988 | btrfs_abort_transaction(trans, extent_root, ret); |
| 5989 | goto out; |
| 5990 | } |
| 5991 | } |
| 5992 | |
| 5993 | ret = update_block_group(root, bytenr, num_bytes, 0); |
| 5994 | if (ret) { |
| 5995 | btrfs_abort_transaction(trans, extent_root, ret); |
| 5996 | goto out; |
| 5997 | } |
| 5998 | } |
| 5999 | btrfs_release_path(path); |
| 6000 | |
| 6001 | /* Deal with the quota accounting */ |
| 6002 | if (!ret && last_ref && !no_quota) { |
| 6003 | int mod_seq = 0; |
| 6004 | |
| 6005 | if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID && |
| 6006 | type == BTRFS_QGROUP_OPER_SUB_SHARED) |
| 6007 | mod_seq = 1; |
| 6008 | |
| 6009 | ret = btrfs_qgroup_record_ref(trans, info, root_objectid, |
| 6010 | bytenr, num_bytes, type, |
| 6011 | mod_seq); |
| 6012 | } |
| 6013 | out: |
| 6014 | btrfs_free_path(path); |
| 6015 | return ret; |
| 6016 | } |
| 6017 | |
| 6018 | /* |
| 6019 | * when we free an block, it is possible (and likely) that we free the last |
| 6020 | * delayed ref for that extent as well. This searches the delayed ref tree for |
| 6021 | * a given extent, and if there are no other delayed refs to be processed, it |
| 6022 | * removes it from the tree. |
| 6023 | */ |
| 6024 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, |
| 6025 | struct btrfs_root *root, u64 bytenr) |
| 6026 | { |
| 6027 | struct btrfs_delayed_ref_head *head; |
| 6028 | struct btrfs_delayed_ref_root *delayed_refs; |
| 6029 | int ret = 0; |
| 6030 | |
| 6031 | delayed_refs = &trans->transaction->delayed_refs; |
| 6032 | spin_lock(&delayed_refs->lock); |
| 6033 | head = btrfs_find_delayed_ref_head(trans, bytenr); |
| 6034 | if (!head) |
| 6035 | goto out_delayed_unlock; |
| 6036 | |
| 6037 | spin_lock(&head->lock); |
| 6038 | if (rb_first(&head->ref_root)) |
| 6039 | goto out; |
| 6040 | |
| 6041 | if (head->extent_op) { |
| 6042 | if (!head->must_insert_reserved) |
| 6043 | goto out; |
| 6044 | btrfs_free_delayed_extent_op(head->extent_op); |
| 6045 | head->extent_op = NULL; |
| 6046 | } |
| 6047 | |
| 6048 | /* |
| 6049 | * waiting for the lock here would deadlock. If someone else has it |
| 6050 | * locked they are already in the process of dropping it anyway |
| 6051 | */ |
| 6052 | if (!mutex_trylock(&head->mutex)) |
| 6053 | goto out; |
| 6054 | |
| 6055 | /* |
| 6056 | * at this point we have a head with no other entries. Go |
| 6057 | * ahead and process it. |
| 6058 | */ |
| 6059 | head->node.in_tree = 0; |
| 6060 | rb_erase(&head->href_node, &delayed_refs->href_root); |
| 6061 | |
| 6062 | atomic_dec(&delayed_refs->num_entries); |
| 6063 | |
| 6064 | /* |
| 6065 | * we don't take a ref on the node because we're removing it from the |
| 6066 | * tree, so we just steal the ref the tree was holding. |
| 6067 | */ |
| 6068 | delayed_refs->num_heads--; |
| 6069 | if (head->processing == 0) |
| 6070 | delayed_refs->num_heads_ready--; |
| 6071 | head->processing = 0; |
| 6072 | spin_unlock(&head->lock); |
| 6073 | spin_unlock(&delayed_refs->lock); |
| 6074 | |
| 6075 | BUG_ON(head->extent_op); |
| 6076 | if (head->must_insert_reserved) |
| 6077 | ret = 1; |
| 6078 | |
| 6079 | mutex_unlock(&head->mutex); |
| 6080 | btrfs_put_delayed_ref(&head->node); |
| 6081 | return ret; |
| 6082 | out: |
| 6083 | spin_unlock(&head->lock); |
| 6084 | |
| 6085 | out_delayed_unlock: |
| 6086 | spin_unlock(&delayed_refs->lock); |
| 6087 | return 0; |
| 6088 | } |
| 6089 | |
| 6090 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
| 6091 | struct btrfs_root *root, |
| 6092 | struct extent_buffer *buf, |
| 6093 | u64 parent, int last_ref) |
| 6094 | { |
| 6095 | struct btrfs_block_group_cache *cache = NULL; |
| 6096 | int pin = 1; |
| 6097 | int ret; |
| 6098 | |
| 6099 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
| 6100 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
| 6101 | buf->start, buf->len, |
| 6102 | parent, root->root_key.objectid, |
| 6103 | btrfs_header_level(buf), |
| 6104 | BTRFS_DROP_DELAYED_REF, NULL, 0); |
| 6105 | BUG_ON(ret); /* -ENOMEM */ |
| 6106 | } |
| 6107 | |
| 6108 | if (!last_ref) |
| 6109 | return; |
| 6110 | |
| 6111 | cache = btrfs_lookup_block_group(root->fs_info, buf->start); |
| 6112 | |
| 6113 | if (btrfs_header_generation(buf) == trans->transid) { |
| 6114 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
| 6115 | ret = check_ref_cleanup(trans, root, buf->start); |
| 6116 | if (!ret) |
| 6117 | goto out; |
| 6118 | } |
| 6119 | |
| 6120 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
| 6121 | pin_down_extent(root, cache, buf->start, buf->len, 1); |
| 6122 | goto out; |
| 6123 | } |
| 6124 | |
| 6125 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); |
| 6126 | |
| 6127 | btrfs_add_free_space(cache, buf->start, buf->len); |
| 6128 | btrfs_update_reserved_bytes(cache, buf->len, RESERVE_FREE); |
| 6129 | trace_btrfs_reserved_extent_free(root, buf->start, buf->len); |
| 6130 | pin = 0; |
| 6131 | } |
| 6132 | out: |
| 6133 | if (pin) |
| 6134 | add_pinned_bytes(root->fs_info, buf->len, |
| 6135 | btrfs_header_level(buf), |
| 6136 | root->root_key.objectid); |
| 6137 | |
| 6138 | /* |
| 6139 | * Deleting the buffer, clear the corrupt flag since it doesn't matter |
| 6140 | * anymore. |
| 6141 | */ |
| 6142 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); |
| 6143 | btrfs_put_block_group(cache); |
| 6144 | } |
| 6145 | |
| 6146 | /* Can return -ENOMEM */ |
| 6147 | int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
| 6148 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, |
| 6149 | u64 owner, u64 offset, int no_quota) |
| 6150 | { |
| 6151 | int ret; |
| 6152 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 6153 | |
| 6154 | add_pinned_bytes(root->fs_info, num_bytes, owner, root_objectid); |
| 6155 | |
| 6156 | /* |
| 6157 | * tree log blocks never actually go into the extent allocation |
| 6158 | * tree, just update pinning info and exit early. |
| 6159 | */ |
| 6160 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
| 6161 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); |
| 6162 | /* unlocks the pinned mutex */ |
| 6163 | btrfs_pin_extent(root, bytenr, num_bytes, 1); |
| 6164 | ret = 0; |
| 6165 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
| 6166 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
| 6167 | num_bytes, |
| 6168 | parent, root_objectid, (int)owner, |
| 6169 | BTRFS_DROP_DELAYED_REF, NULL, no_quota); |
| 6170 | } else { |
| 6171 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
| 6172 | num_bytes, |
| 6173 | parent, root_objectid, owner, |
| 6174 | offset, BTRFS_DROP_DELAYED_REF, |
| 6175 | NULL, no_quota); |
| 6176 | } |
| 6177 | return ret; |
| 6178 | } |
| 6179 | |
| 6180 | static u64 stripe_align(struct btrfs_root *root, |
| 6181 | struct btrfs_block_group_cache *cache, |
| 6182 | u64 val, u64 num_bytes) |
| 6183 | { |
| 6184 | u64 ret = ALIGN(val, root->stripesize); |
| 6185 | return ret; |
| 6186 | } |
| 6187 | |
| 6188 | /* |
| 6189 | * when we wait for progress in the block group caching, its because |
| 6190 | * our allocation attempt failed at least once. So, we must sleep |
| 6191 | * and let some progress happen before we try again. |
| 6192 | * |
| 6193 | * This function will sleep at least once waiting for new free space to |
| 6194 | * show up, and then it will check the block group free space numbers |
| 6195 | * for our min num_bytes. Another option is to have it go ahead |
| 6196 | * and look in the rbtree for a free extent of a given size, but this |
| 6197 | * is a good start. |
| 6198 | * |
| 6199 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using |
| 6200 | * any of the information in this block group. |
| 6201 | */ |
| 6202 | static noinline void |
| 6203 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
| 6204 | u64 num_bytes) |
| 6205 | { |
| 6206 | struct btrfs_caching_control *caching_ctl; |
| 6207 | |
| 6208 | caching_ctl = get_caching_control(cache); |
| 6209 | if (!caching_ctl) |
| 6210 | return; |
| 6211 | |
| 6212 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
| 6213 | (cache->free_space_ctl->free_space >= num_bytes)); |
| 6214 | |
| 6215 | put_caching_control(caching_ctl); |
| 6216 | } |
| 6217 | |
| 6218 | static noinline int |
| 6219 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) |
| 6220 | { |
| 6221 | struct btrfs_caching_control *caching_ctl; |
| 6222 | int ret = 0; |
| 6223 | |
| 6224 | caching_ctl = get_caching_control(cache); |
| 6225 | if (!caching_ctl) |
| 6226 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
| 6227 | |
| 6228 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); |
| 6229 | if (cache->cached == BTRFS_CACHE_ERROR) |
| 6230 | ret = -EIO; |
| 6231 | put_caching_control(caching_ctl); |
| 6232 | return ret; |
| 6233 | } |
| 6234 | |
| 6235 | int __get_raid_index(u64 flags) |
| 6236 | { |
| 6237 | if (flags & BTRFS_BLOCK_GROUP_RAID10) |
| 6238 | return BTRFS_RAID_RAID10; |
| 6239 | else if (flags & BTRFS_BLOCK_GROUP_RAID1) |
| 6240 | return BTRFS_RAID_RAID1; |
| 6241 | else if (flags & BTRFS_BLOCK_GROUP_DUP) |
| 6242 | return BTRFS_RAID_DUP; |
| 6243 | else if (flags & BTRFS_BLOCK_GROUP_RAID0) |
| 6244 | return BTRFS_RAID_RAID0; |
| 6245 | else if (flags & BTRFS_BLOCK_GROUP_RAID5) |
| 6246 | return BTRFS_RAID_RAID5; |
| 6247 | else if (flags & BTRFS_BLOCK_GROUP_RAID6) |
| 6248 | return BTRFS_RAID_RAID6; |
| 6249 | |
| 6250 | return BTRFS_RAID_SINGLE; /* BTRFS_BLOCK_GROUP_SINGLE */ |
| 6251 | } |
| 6252 | |
| 6253 | int get_block_group_index(struct btrfs_block_group_cache *cache) |
| 6254 | { |
| 6255 | return __get_raid_index(cache->flags); |
| 6256 | } |
| 6257 | |
| 6258 | static const char *btrfs_raid_type_names[BTRFS_NR_RAID_TYPES] = { |
| 6259 | [BTRFS_RAID_RAID10] = "raid10", |
| 6260 | [BTRFS_RAID_RAID1] = "raid1", |
| 6261 | [BTRFS_RAID_DUP] = "dup", |
| 6262 | [BTRFS_RAID_RAID0] = "raid0", |
| 6263 | [BTRFS_RAID_SINGLE] = "single", |
| 6264 | [BTRFS_RAID_RAID5] = "raid5", |
| 6265 | [BTRFS_RAID_RAID6] = "raid6", |
| 6266 | }; |
| 6267 | |
| 6268 | static const char *get_raid_name(enum btrfs_raid_types type) |
| 6269 | { |
| 6270 | if (type >= BTRFS_NR_RAID_TYPES) |
| 6271 | return NULL; |
| 6272 | |
| 6273 | return btrfs_raid_type_names[type]; |
| 6274 | } |
| 6275 | |
| 6276 | enum btrfs_loop_type { |
| 6277 | LOOP_CACHING_NOWAIT = 0, |
| 6278 | LOOP_CACHING_WAIT = 1, |
| 6279 | LOOP_ALLOC_CHUNK = 2, |
| 6280 | LOOP_NO_EMPTY_SIZE = 3, |
| 6281 | }; |
| 6282 | |
| 6283 | /* |
| 6284 | * walks the btree of allocated extents and find a hole of a given size. |
| 6285 | * The key ins is changed to record the hole: |
| 6286 | * ins->objectid == start position |
| 6287 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
| 6288 | * ins->offset == the size of the hole. |
| 6289 | * Any available blocks before search_start are skipped. |
| 6290 | * |
| 6291 | * If there is no suitable free space, we will record the max size of |
| 6292 | * the free space extent currently. |
| 6293 | */ |
| 6294 | static noinline int find_free_extent(struct btrfs_root *orig_root, |
| 6295 | u64 num_bytes, u64 empty_size, |
| 6296 | u64 hint_byte, struct btrfs_key *ins, |
| 6297 | u64 flags) |
| 6298 | { |
| 6299 | int ret = 0; |
| 6300 | struct btrfs_root *root = orig_root->fs_info->extent_root; |
| 6301 | struct btrfs_free_cluster *last_ptr = NULL; |
| 6302 | struct btrfs_block_group_cache *block_group = NULL; |
| 6303 | u64 search_start = 0; |
| 6304 | u64 max_extent_size = 0; |
| 6305 | int empty_cluster = 2 * 1024 * 1024; |
| 6306 | struct btrfs_space_info *space_info; |
| 6307 | int loop = 0; |
| 6308 | int index = __get_raid_index(flags); |
| 6309 | int alloc_type = (flags & BTRFS_BLOCK_GROUP_DATA) ? |
| 6310 | RESERVE_ALLOC_NO_ACCOUNT : RESERVE_ALLOC; |
| 6311 | bool failed_cluster_refill = false; |
| 6312 | bool failed_alloc = false; |
| 6313 | bool use_cluster = true; |
| 6314 | bool have_caching_bg = false; |
| 6315 | |
| 6316 | WARN_ON(num_bytes < root->sectorsize); |
| 6317 | btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY); |
| 6318 | ins->objectid = 0; |
| 6319 | ins->offset = 0; |
| 6320 | |
| 6321 | trace_find_free_extent(orig_root, num_bytes, empty_size, flags); |
| 6322 | |
| 6323 | space_info = __find_space_info(root->fs_info, flags); |
| 6324 | if (!space_info) { |
| 6325 | btrfs_err(root->fs_info, "No space info for %llu", flags); |
| 6326 | return -ENOSPC; |
| 6327 | } |
| 6328 | |
| 6329 | /* |
| 6330 | * If the space info is for both data and metadata it means we have a |
| 6331 | * small filesystem and we can't use the clustering stuff. |
| 6332 | */ |
| 6333 | if (btrfs_mixed_space_info(space_info)) |
| 6334 | use_cluster = false; |
| 6335 | |
| 6336 | if (flags & BTRFS_BLOCK_GROUP_METADATA && use_cluster) { |
| 6337 | last_ptr = &root->fs_info->meta_alloc_cluster; |
| 6338 | if (!btrfs_test_opt(root, SSD)) |
| 6339 | empty_cluster = 64 * 1024; |
| 6340 | } |
| 6341 | |
| 6342 | if ((flags & BTRFS_BLOCK_GROUP_DATA) && use_cluster && |
| 6343 | btrfs_test_opt(root, SSD)) { |
| 6344 | last_ptr = &root->fs_info->data_alloc_cluster; |
| 6345 | } |
| 6346 | |
| 6347 | if (last_ptr) { |
| 6348 | spin_lock(&last_ptr->lock); |
| 6349 | if (last_ptr->block_group) |
| 6350 | hint_byte = last_ptr->window_start; |
| 6351 | spin_unlock(&last_ptr->lock); |
| 6352 | } |
| 6353 | |
| 6354 | search_start = max(search_start, first_logical_byte(root, 0)); |
| 6355 | search_start = max(search_start, hint_byte); |
| 6356 | |
| 6357 | if (!last_ptr) |
| 6358 | empty_cluster = 0; |
| 6359 | |
| 6360 | if (search_start == hint_byte) { |
| 6361 | block_group = btrfs_lookup_block_group(root->fs_info, |
| 6362 | search_start); |
| 6363 | /* |
| 6364 | * we don't want to use the block group if it doesn't match our |
| 6365 | * allocation bits, or if its not cached. |
| 6366 | * |
| 6367 | * However if we are re-searching with an ideal block group |
| 6368 | * picked out then we don't care that the block group is cached. |
| 6369 | */ |
| 6370 | if (block_group && block_group_bits(block_group, flags) && |
| 6371 | block_group->cached != BTRFS_CACHE_NO) { |
| 6372 | down_read(&space_info->groups_sem); |
| 6373 | if (list_empty(&block_group->list) || |
| 6374 | block_group->ro) { |
| 6375 | /* |
| 6376 | * someone is removing this block group, |
| 6377 | * we can't jump into the have_block_group |
| 6378 | * target because our list pointers are not |
| 6379 | * valid |
| 6380 | */ |
| 6381 | btrfs_put_block_group(block_group); |
| 6382 | up_read(&space_info->groups_sem); |
| 6383 | } else { |
| 6384 | index = get_block_group_index(block_group); |
| 6385 | goto have_block_group; |
| 6386 | } |
| 6387 | } else if (block_group) { |
| 6388 | btrfs_put_block_group(block_group); |
| 6389 | } |
| 6390 | } |
| 6391 | search: |
| 6392 | have_caching_bg = false; |
| 6393 | down_read(&space_info->groups_sem); |
| 6394 | list_for_each_entry(block_group, &space_info->block_groups[index], |
| 6395 | list) { |
| 6396 | u64 offset; |
| 6397 | int cached; |
| 6398 | |
| 6399 | btrfs_get_block_group(block_group); |
| 6400 | search_start = block_group->key.objectid; |
| 6401 | |
| 6402 | /* |
| 6403 | * this can happen if we end up cycling through all the |
| 6404 | * raid types, but we want to make sure we only allocate |
| 6405 | * for the proper type. |
| 6406 | */ |
| 6407 | if (!block_group_bits(block_group, flags)) { |
| 6408 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
| 6409 | BTRFS_BLOCK_GROUP_RAID1 | |
| 6410 | BTRFS_BLOCK_GROUP_RAID5 | |
| 6411 | BTRFS_BLOCK_GROUP_RAID6 | |
| 6412 | BTRFS_BLOCK_GROUP_RAID10; |
| 6413 | |
| 6414 | /* |
| 6415 | * if they asked for extra copies and this block group |
| 6416 | * doesn't provide them, bail. This does allow us to |
| 6417 | * fill raid0 from raid1. |
| 6418 | */ |
| 6419 | if ((flags & extra) && !(block_group->flags & extra)) |
| 6420 | goto loop; |
| 6421 | } |
| 6422 | |
| 6423 | have_block_group: |
| 6424 | cached = block_group_cache_done(block_group); |
| 6425 | if (unlikely(!cached)) { |
| 6426 | ret = cache_block_group(block_group, 0); |
| 6427 | BUG_ON(ret < 0); |
| 6428 | ret = 0; |
| 6429 | } |
| 6430 | |
| 6431 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
| 6432 | goto loop; |
| 6433 | if (unlikely(block_group->ro)) |
| 6434 | goto loop; |
| 6435 | |
| 6436 | /* |
| 6437 | * Ok we want to try and use the cluster allocator, so |
| 6438 | * lets look there |
| 6439 | */ |
| 6440 | if (last_ptr) { |
| 6441 | struct btrfs_block_group_cache *used_block_group; |
| 6442 | unsigned long aligned_cluster; |
| 6443 | /* |
| 6444 | * the refill lock keeps out other |
| 6445 | * people trying to start a new cluster |
| 6446 | */ |
| 6447 | spin_lock(&last_ptr->refill_lock); |
| 6448 | used_block_group = last_ptr->block_group; |
| 6449 | if (used_block_group != block_group && |
| 6450 | (!used_block_group || |
| 6451 | used_block_group->ro || |
| 6452 | !block_group_bits(used_block_group, flags))) |
| 6453 | goto refill_cluster; |
| 6454 | |
| 6455 | if (used_block_group != block_group) |
| 6456 | btrfs_get_block_group(used_block_group); |
| 6457 | |
| 6458 | offset = btrfs_alloc_from_cluster(used_block_group, |
| 6459 | last_ptr, |
| 6460 | num_bytes, |
| 6461 | used_block_group->key.objectid, |
| 6462 | &max_extent_size); |
| 6463 | if (offset) { |
| 6464 | /* we have a block, we're done */ |
| 6465 | spin_unlock(&last_ptr->refill_lock); |
| 6466 | trace_btrfs_reserve_extent_cluster(root, |
| 6467 | used_block_group, |
| 6468 | search_start, num_bytes); |
| 6469 | if (used_block_group != block_group) { |
| 6470 | btrfs_put_block_group(block_group); |
| 6471 | block_group = used_block_group; |
| 6472 | } |
| 6473 | goto checks; |
| 6474 | } |
| 6475 | |
| 6476 | WARN_ON(last_ptr->block_group != used_block_group); |
| 6477 | if (used_block_group != block_group) |
| 6478 | btrfs_put_block_group(used_block_group); |
| 6479 | refill_cluster: |
| 6480 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
| 6481 | * set up a new clusters, so lets just skip it |
| 6482 | * and let the allocator find whatever block |
| 6483 | * it can find. If we reach this point, we |
| 6484 | * will have tried the cluster allocator |
| 6485 | * plenty of times and not have found |
| 6486 | * anything, so we are likely way too |
| 6487 | * fragmented for the clustering stuff to find |
| 6488 | * anything. |
| 6489 | * |
| 6490 | * However, if the cluster is taken from the |
| 6491 | * current block group, release the cluster |
| 6492 | * first, so that we stand a better chance of |
| 6493 | * succeeding in the unclustered |
| 6494 | * allocation. */ |
| 6495 | if (loop >= LOOP_NO_EMPTY_SIZE && |
| 6496 | last_ptr->block_group != block_group) { |
| 6497 | spin_unlock(&last_ptr->refill_lock); |
| 6498 | goto unclustered_alloc; |
| 6499 | } |
| 6500 | |
| 6501 | /* |
| 6502 | * this cluster didn't work out, free it and |
| 6503 | * start over |
| 6504 | */ |
| 6505 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
| 6506 | |
| 6507 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
| 6508 | spin_unlock(&last_ptr->refill_lock); |
| 6509 | goto unclustered_alloc; |
| 6510 | } |
| 6511 | |
| 6512 | aligned_cluster = max_t(unsigned long, |
| 6513 | empty_cluster + empty_size, |
| 6514 | block_group->full_stripe_len); |
| 6515 | |
| 6516 | /* allocate a cluster in this block group */ |
| 6517 | ret = btrfs_find_space_cluster(root, block_group, |
| 6518 | last_ptr, search_start, |
| 6519 | num_bytes, |
| 6520 | aligned_cluster); |
| 6521 | if (ret == 0) { |
| 6522 | /* |
| 6523 | * now pull our allocation out of this |
| 6524 | * cluster |
| 6525 | */ |
| 6526 | offset = btrfs_alloc_from_cluster(block_group, |
| 6527 | last_ptr, |
| 6528 | num_bytes, |
| 6529 | search_start, |
| 6530 | &max_extent_size); |
| 6531 | if (offset) { |
| 6532 | /* we found one, proceed */ |
| 6533 | spin_unlock(&last_ptr->refill_lock); |
| 6534 | trace_btrfs_reserve_extent_cluster(root, |
| 6535 | block_group, search_start, |
| 6536 | num_bytes); |
| 6537 | goto checks; |
| 6538 | } |
| 6539 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
| 6540 | && !failed_cluster_refill) { |
| 6541 | spin_unlock(&last_ptr->refill_lock); |
| 6542 | |
| 6543 | failed_cluster_refill = true; |
| 6544 | wait_block_group_cache_progress(block_group, |
| 6545 | num_bytes + empty_cluster + empty_size); |
| 6546 | goto have_block_group; |
| 6547 | } |
| 6548 | |
| 6549 | /* |
| 6550 | * at this point we either didn't find a cluster |
| 6551 | * or we weren't able to allocate a block from our |
| 6552 | * cluster. Free the cluster we've been trying |
| 6553 | * to use, and go to the next block group |
| 6554 | */ |
| 6555 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
| 6556 | spin_unlock(&last_ptr->refill_lock); |
| 6557 | goto loop; |
| 6558 | } |
| 6559 | |
| 6560 | unclustered_alloc: |
| 6561 | spin_lock(&block_group->free_space_ctl->tree_lock); |
| 6562 | if (cached && |
| 6563 | block_group->free_space_ctl->free_space < |
| 6564 | num_bytes + empty_cluster + empty_size) { |
| 6565 | if (block_group->free_space_ctl->free_space > |
| 6566 | max_extent_size) |
| 6567 | max_extent_size = |
| 6568 | block_group->free_space_ctl->free_space; |
| 6569 | spin_unlock(&block_group->free_space_ctl->tree_lock); |
| 6570 | goto loop; |
| 6571 | } |
| 6572 | spin_unlock(&block_group->free_space_ctl->tree_lock); |
| 6573 | |
| 6574 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
| 6575 | num_bytes, empty_size, |
| 6576 | &max_extent_size); |
| 6577 | /* |
| 6578 | * If we didn't find a chunk, and we haven't failed on this |
| 6579 | * block group before, and this block group is in the middle of |
| 6580 | * caching and we are ok with waiting, then go ahead and wait |
| 6581 | * for progress to be made, and set failed_alloc to true. |
| 6582 | * |
| 6583 | * If failed_alloc is true then we've already waited on this |
| 6584 | * block group once and should move on to the next block group. |
| 6585 | */ |
| 6586 | if (!offset && !failed_alloc && !cached && |
| 6587 | loop > LOOP_CACHING_NOWAIT) { |
| 6588 | wait_block_group_cache_progress(block_group, |
| 6589 | num_bytes + empty_size); |
| 6590 | failed_alloc = true; |
| 6591 | goto have_block_group; |
| 6592 | } else if (!offset) { |
| 6593 | if (!cached) |
| 6594 | have_caching_bg = true; |
| 6595 | goto loop; |
| 6596 | } |
| 6597 | checks: |
| 6598 | search_start = stripe_align(root, block_group, |
| 6599 | offset, num_bytes); |
| 6600 | |
| 6601 | /* move on to the next group */ |
| 6602 | if (search_start + num_bytes > |
| 6603 | block_group->key.objectid + block_group->key.offset) { |
| 6604 | btrfs_add_free_space(block_group, offset, num_bytes); |
| 6605 | goto loop; |
| 6606 | } |
| 6607 | |
| 6608 | if (offset < search_start) |
| 6609 | btrfs_add_free_space(block_group, offset, |
| 6610 | search_start - offset); |
| 6611 | BUG_ON(offset > search_start); |
| 6612 | |
| 6613 | ret = btrfs_update_reserved_bytes(block_group, num_bytes, |
| 6614 | alloc_type); |
| 6615 | if (ret == -EAGAIN) { |
| 6616 | btrfs_add_free_space(block_group, offset, num_bytes); |
| 6617 | goto loop; |
| 6618 | } |
| 6619 | |
| 6620 | /* we are all good, lets return */ |
| 6621 | ins->objectid = search_start; |
| 6622 | ins->offset = num_bytes; |
| 6623 | |
| 6624 | trace_btrfs_reserve_extent(orig_root, block_group, |
| 6625 | search_start, num_bytes); |
| 6626 | btrfs_put_block_group(block_group); |
| 6627 | break; |
| 6628 | loop: |
| 6629 | failed_cluster_refill = false; |
| 6630 | failed_alloc = false; |
| 6631 | BUG_ON(index != get_block_group_index(block_group)); |
| 6632 | btrfs_put_block_group(block_group); |
| 6633 | } |
| 6634 | up_read(&space_info->groups_sem); |
| 6635 | |
| 6636 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
| 6637 | goto search; |
| 6638 | |
| 6639 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
| 6640 | goto search; |
| 6641 | |
| 6642 | /* |
| 6643 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
| 6644 | * caching kthreads as we move along |
| 6645 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
| 6646 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again |
| 6647 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try |
| 6648 | * again |
| 6649 | */ |
| 6650 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
| 6651 | index = 0; |
| 6652 | loop++; |
| 6653 | if (loop == LOOP_ALLOC_CHUNK) { |
| 6654 | struct btrfs_trans_handle *trans; |
| 6655 | int exist = 0; |
| 6656 | |
| 6657 | trans = current->journal_info; |
| 6658 | if (trans) |
| 6659 | exist = 1; |
| 6660 | else |
| 6661 | trans = btrfs_join_transaction(root); |
| 6662 | |
| 6663 | if (IS_ERR(trans)) { |
| 6664 | ret = PTR_ERR(trans); |
| 6665 | goto out; |
| 6666 | } |
| 6667 | |
| 6668 | ret = do_chunk_alloc(trans, root, flags, |
| 6669 | CHUNK_ALLOC_FORCE); |
| 6670 | /* |
| 6671 | * Do not bail out on ENOSPC since we |
| 6672 | * can do more things. |
| 6673 | */ |
| 6674 | if (ret < 0 && ret != -ENOSPC) |
| 6675 | btrfs_abort_transaction(trans, |
| 6676 | root, ret); |
| 6677 | else |
| 6678 | ret = 0; |
| 6679 | if (!exist) |
| 6680 | btrfs_end_transaction(trans, root); |
| 6681 | if (ret) |
| 6682 | goto out; |
| 6683 | } |
| 6684 | |
| 6685 | if (loop == LOOP_NO_EMPTY_SIZE) { |
| 6686 | empty_size = 0; |
| 6687 | empty_cluster = 0; |
| 6688 | } |
| 6689 | |
| 6690 | goto search; |
| 6691 | } else if (!ins->objectid) { |
| 6692 | ret = -ENOSPC; |
| 6693 | } else if (ins->objectid) { |
| 6694 | ret = 0; |
| 6695 | } |
| 6696 | out: |
| 6697 | if (ret == -ENOSPC) |
| 6698 | ins->offset = max_extent_size; |
| 6699 | return ret; |
| 6700 | } |
| 6701 | |
| 6702 | static void dump_space_info(struct btrfs_space_info *info, u64 bytes, |
| 6703 | int dump_block_groups) |
| 6704 | { |
| 6705 | struct btrfs_block_group_cache *cache; |
| 6706 | int index = 0; |
| 6707 | |
| 6708 | spin_lock(&info->lock); |
| 6709 | printk(KERN_INFO "BTRFS: space_info %llu has %llu free, is %sfull\n", |
| 6710 | info->flags, |
| 6711 | info->total_bytes - info->bytes_used - info->bytes_pinned - |
| 6712 | info->bytes_reserved - info->bytes_readonly, |
| 6713 | (info->full) ? "" : "not "); |
| 6714 | printk(KERN_INFO "BTRFS: space_info total=%llu, used=%llu, pinned=%llu, " |
| 6715 | "reserved=%llu, may_use=%llu, readonly=%llu\n", |
| 6716 | info->total_bytes, info->bytes_used, info->bytes_pinned, |
| 6717 | info->bytes_reserved, info->bytes_may_use, |
| 6718 | info->bytes_readonly); |
| 6719 | spin_unlock(&info->lock); |
| 6720 | |
| 6721 | if (!dump_block_groups) |
| 6722 | return; |
| 6723 | |
| 6724 | down_read(&info->groups_sem); |
| 6725 | again: |
| 6726 | list_for_each_entry(cache, &info->block_groups[index], list) { |
| 6727 | spin_lock(&cache->lock); |
| 6728 | printk(KERN_INFO "BTRFS: " |
| 6729 | "block group %llu has %llu bytes, " |
| 6730 | "%llu used %llu pinned %llu reserved %s\n", |
| 6731 | cache->key.objectid, cache->key.offset, |
| 6732 | btrfs_block_group_used(&cache->item), cache->pinned, |
| 6733 | cache->reserved, cache->ro ? "[readonly]" : ""); |
| 6734 | btrfs_dump_free_space(cache, bytes); |
| 6735 | spin_unlock(&cache->lock); |
| 6736 | } |
| 6737 | if (++index < BTRFS_NR_RAID_TYPES) |
| 6738 | goto again; |
| 6739 | up_read(&info->groups_sem); |
| 6740 | } |
| 6741 | |
| 6742 | int btrfs_reserve_extent(struct btrfs_root *root, |
| 6743 | u64 num_bytes, u64 min_alloc_size, |
| 6744 | u64 empty_size, u64 hint_byte, |
| 6745 | struct btrfs_key *ins, int is_data) |
| 6746 | { |
| 6747 | bool final_tried = false; |
| 6748 | u64 flags; |
| 6749 | int ret; |
| 6750 | |
| 6751 | flags = btrfs_get_alloc_profile(root, is_data); |
| 6752 | again: |
| 6753 | WARN_ON(num_bytes < root->sectorsize); |
| 6754 | ret = find_free_extent(root, num_bytes, empty_size, hint_byte, ins, |
| 6755 | flags); |
| 6756 | |
| 6757 | if (ret == -ENOSPC) { |
| 6758 | if (!final_tried && ins->offset) { |
| 6759 | num_bytes = min(num_bytes >> 1, ins->offset); |
| 6760 | num_bytes = round_down(num_bytes, root->sectorsize); |
| 6761 | num_bytes = max(num_bytes, min_alloc_size); |
| 6762 | if (num_bytes == min_alloc_size) |
| 6763 | final_tried = true; |
| 6764 | goto again; |
| 6765 | } else if (btrfs_test_opt(root, ENOSPC_DEBUG)) { |
| 6766 | struct btrfs_space_info *sinfo; |
| 6767 | |
| 6768 | sinfo = __find_space_info(root->fs_info, flags); |
| 6769 | btrfs_err(root->fs_info, "allocation failed flags %llu, wanted %llu", |
| 6770 | flags, num_bytes); |
| 6771 | if (sinfo) |
| 6772 | dump_space_info(sinfo, num_bytes, 1); |
| 6773 | } |
| 6774 | } |
| 6775 | |
| 6776 | return ret; |
| 6777 | } |
| 6778 | |
| 6779 | static int __btrfs_free_reserved_extent(struct btrfs_root *root, |
| 6780 | u64 start, u64 len, int pin) |
| 6781 | { |
| 6782 | struct btrfs_block_group_cache *cache; |
| 6783 | int ret = 0; |
| 6784 | |
| 6785 | cache = btrfs_lookup_block_group(root->fs_info, start); |
| 6786 | if (!cache) { |
| 6787 | btrfs_err(root->fs_info, "Unable to find block group for %llu", |
| 6788 | start); |
| 6789 | return -ENOSPC; |
| 6790 | } |
| 6791 | |
| 6792 | if (btrfs_test_opt(root, DISCARD)) |
| 6793 | ret = btrfs_discard_extent(root, start, len, NULL); |
| 6794 | |
| 6795 | if (pin) |
| 6796 | pin_down_extent(root, cache, start, len, 1); |
| 6797 | else { |
| 6798 | btrfs_add_free_space(cache, start, len); |
| 6799 | btrfs_update_reserved_bytes(cache, len, RESERVE_FREE); |
| 6800 | } |
| 6801 | btrfs_put_block_group(cache); |
| 6802 | |
| 6803 | trace_btrfs_reserved_extent_free(root, start, len); |
| 6804 | |
| 6805 | return ret; |
| 6806 | } |
| 6807 | |
| 6808 | int btrfs_free_reserved_extent(struct btrfs_root *root, |
| 6809 | u64 start, u64 len) |
| 6810 | { |
| 6811 | return __btrfs_free_reserved_extent(root, start, len, 0); |
| 6812 | } |
| 6813 | |
| 6814 | int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root, |
| 6815 | u64 start, u64 len) |
| 6816 | { |
| 6817 | return __btrfs_free_reserved_extent(root, start, len, 1); |
| 6818 | } |
| 6819 | |
| 6820 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
| 6821 | struct btrfs_root *root, |
| 6822 | u64 parent, u64 root_objectid, |
| 6823 | u64 flags, u64 owner, u64 offset, |
| 6824 | struct btrfs_key *ins, int ref_mod) |
| 6825 | { |
| 6826 | int ret; |
| 6827 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 6828 | struct btrfs_extent_item *extent_item; |
| 6829 | struct btrfs_extent_inline_ref *iref; |
| 6830 | struct btrfs_path *path; |
| 6831 | struct extent_buffer *leaf; |
| 6832 | int type; |
| 6833 | u32 size; |
| 6834 | |
| 6835 | if (parent > 0) |
| 6836 | type = BTRFS_SHARED_DATA_REF_KEY; |
| 6837 | else |
| 6838 | type = BTRFS_EXTENT_DATA_REF_KEY; |
| 6839 | |
| 6840 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
| 6841 | |
| 6842 | path = btrfs_alloc_path(); |
| 6843 | if (!path) |
| 6844 | return -ENOMEM; |
| 6845 | |
| 6846 | path->leave_spinning = 1; |
| 6847 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
| 6848 | ins, size); |
| 6849 | if (ret) { |
| 6850 | btrfs_free_path(path); |
| 6851 | return ret; |
| 6852 | } |
| 6853 | |
| 6854 | leaf = path->nodes[0]; |
| 6855 | extent_item = btrfs_item_ptr(leaf, path->slots[0], |
| 6856 | struct btrfs_extent_item); |
| 6857 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
| 6858 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); |
| 6859 | btrfs_set_extent_flags(leaf, extent_item, |
| 6860 | flags | BTRFS_EXTENT_FLAG_DATA); |
| 6861 | |
| 6862 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); |
| 6863 | btrfs_set_extent_inline_ref_type(leaf, iref, type); |
| 6864 | if (parent > 0) { |
| 6865 | struct btrfs_shared_data_ref *ref; |
| 6866 | ref = (struct btrfs_shared_data_ref *)(iref + 1); |
| 6867 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); |
| 6868 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); |
| 6869 | } else { |
| 6870 | struct btrfs_extent_data_ref *ref; |
| 6871 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); |
| 6872 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); |
| 6873 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); |
| 6874 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); |
| 6875 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); |
| 6876 | } |
| 6877 | |
| 6878 | btrfs_mark_buffer_dirty(path->nodes[0]); |
| 6879 | btrfs_free_path(path); |
| 6880 | |
| 6881 | /* Always set parent to 0 here since its exclusive anyway. */ |
| 6882 | ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid, |
| 6883 | ins->objectid, ins->offset, |
| 6884 | BTRFS_QGROUP_OPER_ADD_EXCL, 0); |
| 6885 | if (ret) |
| 6886 | return ret; |
| 6887 | |
| 6888 | ret = update_block_group(root, ins->objectid, ins->offset, 1); |
| 6889 | if (ret) { /* -ENOENT, logic error */ |
| 6890 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
| 6891 | ins->objectid, ins->offset); |
| 6892 | BUG(); |
| 6893 | } |
| 6894 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, ins->offset); |
| 6895 | return ret; |
| 6896 | } |
| 6897 | |
| 6898 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
| 6899 | struct btrfs_root *root, |
| 6900 | u64 parent, u64 root_objectid, |
| 6901 | u64 flags, struct btrfs_disk_key *key, |
| 6902 | int level, struct btrfs_key *ins, |
| 6903 | int no_quota) |
| 6904 | { |
| 6905 | int ret; |
| 6906 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 6907 | struct btrfs_extent_item *extent_item; |
| 6908 | struct btrfs_tree_block_info *block_info; |
| 6909 | struct btrfs_extent_inline_ref *iref; |
| 6910 | struct btrfs_path *path; |
| 6911 | struct extent_buffer *leaf; |
| 6912 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
| 6913 | u64 num_bytes = ins->offset; |
| 6914 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
| 6915 | SKINNY_METADATA); |
| 6916 | |
| 6917 | if (!skinny_metadata) |
| 6918 | size += sizeof(*block_info); |
| 6919 | |
| 6920 | path = btrfs_alloc_path(); |
| 6921 | if (!path) { |
| 6922 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, |
| 6923 | root->leafsize); |
| 6924 | return -ENOMEM; |
| 6925 | } |
| 6926 | |
| 6927 | path->leave_spinning = 1; |
| 6928 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
| 6929 | ins, size); |
| 6930 | if (ret) { |
| 6931 | btrfs_free_and_pin_reserved_extent(root, ins->objectid, |
| 6932 | root->leafsize); |
| 6933 | btrfs_free_path(path); |
| 6934 | return ret; |
| 6935 | } |
| 6936 | |
| 6937 | leaf = path->nodes[0]; |
| 6938 | extent_item = btrfs_item_ptr(leaf, path->slots[0], |
| 6939 | struct btrfs_extent_item); |
| 6940 | btrfs_set_extent_refs(leaf, extent_item, 1); |
| 6941 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); |
| 6942 | btrfs_set_extent_flags(leaf, extent_item, |
| 6943 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); |
| 6944 | |
| 6945 | if (skinny_metadata) { |
| 6946 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); |
| 6947 | num_bytes = root->leafsize; |
| 6948 | } else { |
| 6949 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); |
| 6950 | btrfs_set_tree_block_key(leaf, block_info, key); |
| 6951 | btrfs_set_tree_block_level(leaf, block_info, level); |
| 6952 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
| 6953 | } |
| 6954 | |
| 6955 | if (parent > 0) { |
| 6956 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
| 6957 | btrfs_set_extent_inline_ref_type(leaf, iref, |
| 6958 | BTRFS_SHARED_BLOCK_REF_KEY); |
| 6959 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); |
| 6960 | } else { |
| 6961 | btrfs_set_extent_inline_ref_type(leaf, iref, |
| 6962 | BTRFS_TREE_BLOCK_REF_KEY); |
| 6963 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); |
| 6964 | } |
| 6965 | |
| 6966 | btrfs_mark_buffer_dirty(leaf); |
| 6967 | btrfs_free_path(path); |
| 6968 | |
| 6969 | if (!no_quota) { |
| 6970 | ret = btrfs_qgroup_record_ref(trans, fs_info, root_objectid, |
| 6971 | ins->objectid, num_bytes, |
| 6972 | BTRFS_QGROUP_OPER_ADD_EXCL, 0); |
| 6973 | if (ret) |
| 6974 | return ret; |
| 6975 | } |
| 6976 | |
| 6977 | ret = update_block_group(root, ins->objectid, root->leafsize, 1); |
| 6978 | if (ret) { /* -ENOENT, logic error */ |
| 6979 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
| 6980 | ins->objectid, ins->offset); |
| 6981 | BUG(); |
| 6982 | } |
| 6983 | |
| 6984 | trace_btrfs_reserved_extent_alloc(root, ins->objectid, root->leafsize); |
| 6985 | return ret; |
| 6986 | } |
| 6987 | |
| 6988 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
| 6989 | struct btrfs_root *root, |
| 6990 | u64 root_objectid, u64 owner, |
| 6991 | u64 offset, struct btrfs_key *ins) |
| 6992 | { |
| 6993 | int ret; |
| 6994 | |
| 6995 | BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID); |
| 6996 | |
| 6997 | ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid, |
| 6998 | ins->offset, 0, |
| 6999 | root_objectid, owner, offset, |
| 7000 | BTRFS_ADD_DELAYED_EXTENT, NULL, 0); |
| 7001 | return ret; |
| 7002 | } |
| 7003 | |
| 7004 | /* |
| 7005 | * this is used by the tree logging recovery code. It records that |
| 7006 | * an extent has been allocated and makes sure to clear the free |
| 7007 | * space cache bits as well |
| 7008 | */ |
| 7009 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
| 7010 | struct btrfs_root *root, |
| 7011 | u64 root_objectid, u64 owner, u64 offset, |
| 7012 | struct btrfs_key *ins) |
| 7013 | { |
| 7014 | int ret; |
| 7015 | struct btrfs_block_group_cache *block_group; |
| 7016 | |
| 7017 | /* |
| 7018 | * Mixed block groups will exclude before processing the log so we only |
| 7019 | * need to do the exlude dance if this fs isn't mixed. |
| 7020 | */ |
| 7021 | if (!btrfs_fs_incompat(root->fs_info, MIXED_GROUPS)) { |
| 7022 | ret = __exclude_logged_extent(root, ins->objectid, ins->offset); |
| 7023 | if (ret) |
| 7024 | return ret; |
| 7025 | } |
| 7026 | |
| 7027 | block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid); |
| 7028 | if (!block_group) |
| 7029 | return -EINVAL; |
| 7030 | |
| 7031 | ret = btrfs_update_reserved_bytes(block_group, ins->offset, |
| 7032 | RESERVE_ALLOC_NO_ACCOUNT); |
| 7033 | BUG_ON(ret); /* logic error */ |
| 7034 | ret = alloc_reserved_file_extent(trans, root, 0, root_objectid, |
| 7035 | 0, owner, offset, ins, 1); |
| 7036 | btrfs_put_block_group(block_group); |
| 7037 | return ret; |
| 7038 | } |
| 7039 | |
| 7040 | static struct extent_buffer * |
| 7041 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
| 7042 | u64 bytenr, u32 blocksize, int level) |
| 7043 | { |
| 7044 | struct extent_buffer *buf; |
| 7045 | |
| 7046 | buf = btrfs_find_create_tree_block(root, bytenr, blocksize); |
| 7047 | if (!buf) |
| 7048 | return ERR_PTR(-ENOMEM); |
| 7049 | btrfs_set_header_generation(buf, trans->transid); |
| 7050 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
| 7051 | btrfs_tree_lock(buf); |
| 7052 | clean_tree_block(trans, root, buf); |
| 7053 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
| 7054 | |
| 7055 | btrfs_set_lock_blocking(buf); |
| 7056 | btrfs_set_buffer_uptodate(buf); |
| 7057 | |
| 7058 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
| 7059 | /* |
| 7060 | * we allow two log transactions at a time, use different |
| 7061 | * EXENT bit to differentiate dirty pages. |
| 7062 | */ |
| 7063 | if (root->log_transid % 2 == 0) |
| 7064 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
| 7065 | buf->start + buf->len - 1, GFP_NOFS); |
| 7066 | else |
| 7067 | set_extent_new(&root->dirty_log_pages, buf->start, |
| 7068 | buf->start + buf->len - 1, GFP_NOFS); |
| 7069 | } else { |
| 7070 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
| 7071 | buf->start + buf->len - 1, GFP_NOFS); |
| 7072 | } |
| 7073 | trans->blocks_used++; |
| 7074 | /* this returns a buffer locked for blocking */ |
| 7075 | return buf; |
| 7076 | } |
| 7077 | |
| 7078 | static struct btrfs_block_rsv * |
| 7079 | use_block_rsv(struct btrfs_trans_handle *trans, |
| 7080 | struct btrfs_root *root, u32 blocksize) |
| 7081 | { |
| 7082 | struct btrfs_block_rsv *block_rsv; |
| 7083 | struct btrfs_block_rsv *global_rsv = &root->fs_info->global_block_rsv; |
| 7084 | int ret; |
| 7085 | bool global_updated = false; |
| 7086 | |
| 7087 | block_rsv = get_block_rsv(trans, root); |
| 7088 | |
| 7089 | if (unlikely(block_rsv->size == 0)) |
| 7090 | goto try_reserve; |
| 7091 | again: |
| 7092 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
| 7093 | if (!ret) |
| 7094 | return block_rsv; |
| 7095 | |
| 7096 | if (block_rsv->failfast) |
| 7097 | return ERR_PTR(ret); |
| 7098 | |
| 7099 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
| 7100 | global_updated = true; |
| 7101 | update_global_block_rsv(root->fs_info); |
| 7102 | goto again; |
| 7103 | } |
| 7104 | |
| 7105 | if (btrfs_test_opt(root, ENOSPC_DEBUG)) { |
| 7106 | static DEFINE_RATELIMIT_STATE(_rs, |
| 7107 | DEFAULT_RATELIMIT_INTERVAL * 10, |
| 7108 | /*DEFAULT_RATELIMIT_BURST*/ 1); |
| 7109 | if (__ratelimit(&_rs)) |
| 7110 | WARN(1, KERN_DEBUG |
| 7111 | "BTRFS: block rsv returned %d\n", ret); |
| 7112 | } |
| 7113 | try_reserve: |
| 7114 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, |
| 7115 | BTRFS_RESERVE_NO_FLUSH); |
| 7116 | if (!ret) |
| 7117 | return block_rsv; |
| 7118 | /* |
| 7119 | * If we couldn't reserve metadata bytes try and use some from |
| 7120 | * the global reserve if its space type is the same as the global |
| 7121 | * reservation. |
| 7122 | */ |
| 7123 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
| 7124 | block_rsv->space_info == global_rsv->space_info) { |
| 7125 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
| 7126 | if (!ret) |
| 7127 | return global_rsv; |
| 7128 | } |
| 7129 | return ERR_PTR(ret); |
| 7130 | } |
| 7131 | |
| 7132 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
| 7133 | struct btrfs_block_rsv *block_rsv, u32 blocksize) |
| 7134 | { |
| 7135 | block_rsv_add_bytes(block_rsv, blocksize, 0); |
| 7136 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0); |
| 7137 | } |
| 7138 | |
| 7139 | /* |
| 7140 | * finds a free extent and does all the dirty work required for allocation |
| 7141 | * returns the key for the extent through ins, and a tree buffer for |
| 7142 | * the first block of the extent through buf. |
| 7143 | * |
| 7144 | * returns the tree buffer or NULL. |
| 7145 | */ |
| 7146 | struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans, |
| 7147 | struct btrfs_root *root, u32 blocksize, |
| 7148 | u64 parent, u64 root_objectid, |
| 7149 | struct btrfs_disk_key *key, int level, |
| 7150 | u64 hint, u64 empty_size) |
| 7151 | { |
| 7152 | struct btrfs_key ins; |
| 7153 | struct btrfs_block_rsv *block_rsv; |
| 7154 | struct extent_buffer *buf; |
| 7155 | u64 flags = 0; |
| 7156 | int ret; |
| 7157 | bool skinny_metadata = btrfs_fs_incompat(root->fs_info, |
| 7158 | SKINNY_METADATA); |
| 7159 | |
| 7160 | block_rsv = use_block_rsv(trans, root, blocksize); |
| 7161 | if (IS_ERR(block_rsv)) |
| 7162 | return ERR_CAST(block_rsv); |
| 7163 | |
| 7164 | ret = btrfs_reserve_extent(root, blocksize, blocksize, |
| 7165 | empty_size, hint, &ins, 0); |
| 7166 | if (ret) { |
| 7167 | unuse_block_rsv(root->fs_info, block_rsv, blocksize); |
| 7168 | return ERR_PTR(ret); |
| 7169 | } |
| 7170 | |
| 7171 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, |
| 7172 | blocksize, level); |
| 7173 | BUG_ON(IS_ERR(buf)); /* -ENOMEM */ |
| 7174 | |
| 7175 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { |
| 7176 | if (parent == 0) |
| 7177 | parent = ins.objectid; |
| 7178 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; |
| 7179 | } else |
| 7180 | BUG_ON(parent > 0); |
| 7181 | |
| 7182 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { |
| 7183 | struct btrfs_delayed_extent_op *extent_op; |
| 7184 | extent_op = btrfs_alloc_delayed_extent_op(); |
| 7185 | BUG_ON(!extent_op); /* -ENOMEM */ |
| 7186 | if (key) |
| 7187 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); |
| 7188 | else |
| 7189 | memset(&extent_op->key, 0, sizeof(extent_op->key)); |
| 7190 | extent_op->flags_to_set = flags; |
| 7191 | if (skinny_metadata) |
| 7192 | extent_op->update_key = 0; |
| 7193 | else |
| 7194 | extent_op->update_key = 1; |
| 7195 | extent_op->update_flags = 1; |
| 7196 | extent_op->is_data = 0; |
| 7197 | extent_op->level = level; |
| 7198 | |
| 7199 | ret = btrfs_add_delayed_tree_ref(root->fs_info, trans, |
| 7200 | ins.objectid, |
| 7201 | ins.offset, parent, root_objectid, |
| 7202 | level, BTRFS_ADD_DELAYED_EXTENT, |
| 7203 | extent_op, 0); |
| 7204 | BUG_ON(ret); /* -ENOMEM */ |
| 7205 | } |
| 7206 | return buf; |
| 7207 | } |
| 7208 | |
| 7209 | struct walk_control { |
| 7210 | u64 refs[BTRFS_MAX_LEVEL]; |
| 7211 | u64 flags[BTRFS_MAX_LEVEL]; |
| 7212 | struct btrfs_key update_progress; |
| 7213 | int stage; |
| 7214 | int level; |
| 7215 | int shared_level; |
| 7216 | int update_ref; |
| 7217 | int keep_locks; |
| 7218 | int reada_slot; |
| 7219 | int reada_count; |
| 7220 | int for_reloc; |
| 7221 | }; |
| 7222 | |
| 7223 | #define DROP_REFERENCE 1 |
| 7224 | #define UPDATE_BACKREF 2 |
| 7225 | |
| 7226 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
| 7227 | struct btrfs_root *root, |
| 7228 | struct walk_control *wc, |
| 7229 | struct btrfs_path *path) |
| 7230 | { |
| 7231 | u64 bytenr; |
| 7232 | u64 generation; |
| 7233 | u64 refs; |
| 7234 | u64 flags; |
| 7235 | u32 nritems; |
| 7236 | u32 blocksize; |
| 7237 | struct btrfs_key key; |
| 7238 | struct extent_buffer *eb; |
| 7239 | int ret; |
| 7240 | int slot; |
| 7241 | int nread = 0; |
| 7242 | |
| 7243 | if (path->slots[wc->level] < wc->reada_slot) { |
| 7244 | wc->reada_count = wc->reada_count * 2 / 3; |
| 7245 | wc->reada_count = max(wc->reada_count, 2); |
| 7246 | } else { |
| 7247 | wc->reada_count = wc->reada_count * 3 / 2; |
| 7248 | wc->reada_count = min_t(int, wc->reada_count, |
| 7249 | BTRFS_NODEPTRS_PER_BLOCK(root)); |
| 7250 | } |
| 7251 | |
| 7252 | eb = path->nodes[wc->level]; |
| 7253 | nritems = btrfs_header_nritems(eb); |
| 7254 | blocksize = btrfs_level_size(root, wc->level - 1); |
| 7255 | |
| 7256 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
| 7257 | if (nread >= wc->reada_count) |
| 7258 | break; |
| 7259 | |
| 7260 | cond_resched(); |
| 7261 | bytenr = btrfs_node_blockptr(eb, slot); |
| 7262 | generation = btrfs_node_ptr_generation(eb, slot); |
| 7263 | |
| 7264 | if (slot == path->slots[wc->level]) |
| 7265 | goto reada; |
| 7266 | |
| 7267 | if (wc->stage == UPDATE_BACKREF && |
| 7268 | generation <= root->root_key.offset) |
| 7269 | continue; |
| 7270 | |
| 7271 | /* We don't lock the tree block, it's OK to be racy here */ |
| 7272 | ret = btrfs_lookup_extent_info(trans, root, bytenr, |
| 7273 | wc->level - 1, 1, &refs, |
| 7274 | &flags); |
| 7275 | /* We don't care about errors in readahead. */ |
| 7276 | if (ret < 0) |
| 7277 | continue; |
| 7278 | BUG_ON(refs == 0); |
| 7279 | |
| 7280 | if (wc->stage == DROP_REFERENCE) { |
| 7281 | if (refs == 1) |
| 7282 | goto reada; |
| 7283 | |
| 7284 | if (wc->level == 1 && |
| 7285 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) |
| 7286 | continue; |
| 7287 | if (!wc->update_ref || |
| 7288 | generation <= root->root_key.offset) |
| 7289 | continue; |
| 7290 | btrfs_node_key_to_cpu(eb, &key, slot); |
| 7291 | ret = btrfs_comp_cpu_keys(&key, |
| 7292 | &wc->update_progress); |
| 7293 | if (ret < 0) |
| 7294 | continue; |
| 7295 | } else { |
| 7296 | if (wc->level == 1 && |
| 7297 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) |
| 7298 | continue; |
| 7299 | } |
| 7300 | reada: |
| 7301 | ret = readahead_tree_block(root, bytenr, blocksize, |
| 7302 | generation); |
| 7303 | if (ret) |
| 7304 | break; |
| 7305 | nread++; |
| 7306 | } |
| 7307 | wc->reada_slot = slot; |
| 7308 | } |
| 7309 | |
| 7310 | /* |
| 7311 | * helper to process tree block while walking down the tree. |
| 7312 | * |
| 7313 | * when wc->stage == UPDATE_BACKREF, this function updates |
| 7314 | * back refs for pointers in the block. |
| 7315 | * |
| 7316 | * NOTE: return value 1 means we should stop walking down. |
| 7317 | */ |
| 7318 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
| 7319 | struct btrfs_root *root, |
| 7320 | struct btrfs_path *path, |
| 7321 | struct walk_control *wc, int lookup_info) |
| 7322 | { |
| 7323 | int level = wc->level; |
| 7324 | struct extent_buffer *eb = path->nodes[level]; |
| 7325 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
| 7326 | int ret; |
| 7327 | |
| 7328 | if (wc->stage == UPDATE_BACKREF && |
| 7329 | btrfs_header_owner(eb) != root->root_key.objectid) |
| 7330 | return 1; |
| 7331 | |
| 7332 | /* |
| 7333 | * when reference count of tree block is 1, it won't increase |
| 7334 | * again. once full backref flag is set, we never clear it. |
| 7335 | */ |
| 7336 | if (lookup_info && |
| 7337 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || |
| 7338 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { |
| 7339 | BUG_ON(!path->locks[level]); |
| 7340 | ret = btrfs_lookup_extent_info(trans, root, |
| 7341 | eb->start, level, 1, |
| 7342 | &wc->refs[level], |
| 7343 | &wc->flags[level]); |
| 7344 | BUG_ON(ret == -ENOMEM); |
| 7345 | if (ret) |
| 7346 | return ret; |
| 7347 | BUG_ON(wc->refs[level] == 0); |
| 7348 | } |
| 7349 | |
| 7350 | if (wc->stage == DROP_REFERENCE) { |
| 7351 | if (wc->refs[level] > 1) |
| 7352 | return 1; |
| 7353 | |
| 7354 | if (path->locks[level] && !wc->keep_locks) { |
| 7355 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
| 7356 | path->locks[level] = 0; |
| 7357 | } |
| 7358 | return 0; |
| 7359 | } |
| 7360 | |
| 7361 | /* wc->stage == UPDATE_BACKREF */ |
| 7362 | if (!(wc->flags[level] & flag)) { |
| 7363 | BUG_ON(!path->locks[level]); |
| 7364 | ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc); |
| 7365 | BUG_ON(ret); /* -ENOMEM */ |
| 7366 | ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc); |
| 7367 | BUG_ON(ret); /* -ENOMEM */ |
| 7368 | ret = btrfs_set_disk_extent_flags(trans, root, eb->start, |
| 7369 | eb->len, flag, |
| 7370 | btrfs_header_level(eb), 0); |
| 7371 | BUG_ON(ret); /* -ENOMEM */ |
| 7372 | wc->flags[level] |= flag; |
| 7373 | } |
| 7374 | |
| 7375 | /* |
| 7376 | * the block is shared by multiple trees, so it's not good to |
| 7377 | * keep the tree lock |
| 7378 | */ |
| 7379 | if (path->locks[level] && level > 0) { |
| 7380 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
| 7381 | path->locks[level] = 0; |
| 7382 | } |
| 7383 | return 0; |
| 7384 | } |
| 7385 | |
| 7386 | /* |
| 7387 | * helper to process tree block pointer. |
| 7388 | * |
| 7389 | * when wc->stage == DROP_REFERENCE, this function checks |
| 7390 | * reference count of the block pointed to. if the block |
| 7391 | * is shared and we need update back refs for the subtree |
| 7392 | * rooted at the block, this function changes wc->stage to |
| 7393 | * UPDATE_BACKREF. if the block is shared and there is no |
| 7394 | * need to update back, this function drops the reference |
| 7395 | * to the block. |
| 7396 | * |
| 7397 | * NOTE: return value 1 means we should stop walking down. |
| 7398 | */ |
| 7399 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, |
| 7400 | struct btrfs_root *root, |
| 7401 | struct btrfs_path *path, |
| 7402 | struct walk_control *wc, int *lookup_info) |
| 7403 | { |
| 7404 | u64 bytenr; |
| 7405 | u64 generation; |
| 7406 | u64 parent; |
| 7407 | u32 blocksize; |
| 7408 | struct btrfs_key key; |
| 7409 | struct extent_buffer *next; |
| 7410 | int level = wc->level; |
| 7411 | int reada = 0; |
| 7412 | int ret = 0; |
| 7413 | |
| 7414 | generation = btrfs_node_ptr_generation(path->nodes[level], |
| 7415 | path->slots[level]); |
| 7416 | /* |
| 7417 | * if the lower level block was created before the snapshot |
| 7418 | * was created, we know there is no need to update back refs |
| 7419 | * for the subtree |
| 7420 | */ |
| 7421 | if (wc->stage == UPDATE_BACKREF && |
| 7422 | generation <= root->root_key.offset) { |
| 7423 | *lookup_info = 1; |
| 7424 | return 1; |
| 7425 | } |
| 7426 | |
| 7427 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); |
| 7428 | blocksize = btrfs_level_size(root, level - 1); |
| 7429 | |
| 7430 | next = btrfs_find_tree_block(root, bytenr, blocksize); |
| 7431 | if (!next) { |
| 7432 | next = btrfs_find_create_tree_block(root, bytenr, blocksize); |
| 7433 | if (!next) |
| 7434 | return -ENOMEM; |
| 7435 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
| 7436 | level - 1); |
| 7437 | reada = 1; |
| 7438 | } |
| 7439 | btrfs_tree_lock(next); |
| 7440 | btrfs_set_lock_blocking(next); |
| 7441 | |
| 7442 | ret = btrfs_lookup_extent_info(trans, root, bytenr, level - 1, 1, |
| 7443 | &wc->refs[level - 1], |
| 7444 | &wc->flags[level - 1]); |
| 7445 | if (ret < 0) { |
| 7446 | btrfs_tree_unlock(next); |
| 7447 | return ret; |
| 7448 | } |
| 7449 | |
| 7450 | if (unlikely(wc->refs[level - 1] == 0)) { |
| 7451 | btrfs_err(root->fs_info, "Missing references."); |
| 7452 | BUG(); |
| 7453 | } |
| 7454 | *lookup_info = 0; |
| 7455 | |
| 7456 | if (wc->stage == DROP_REFERENCE) { |
| 7457 | if (wc->refs[level - 1] > 1) { |
| 7458 | if (level == 1 && |
| 7459 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) |
| 7460 | goto skip; |
| 7461 | |
| 7462 | if (!wc->update_ref || |
| 7463 | generation <= root->root_key.offset) |
| 7464 | goto skip; |
| 7465 | |
| 7466 | btrfs_node_key_to_cpu(path->nodes[level], &key, |
| 7467 | path->slots[level]); |
| 7468 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); |
| 7469 | if (ret < 0) |
| 7470 | goto skip; |
| 7471 | |
| 7472 | wc->stage = UPDATE_BACKREF; |
| 7473 | wc->shared_level = level - 1; |
| 7474 | } |
| 7475 | } else { |
| 7476 | if (level == 1 && |
| 7477 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) |
| 7478 | goto skip; |
| 7479 | } |
| 7480 | |
| 7481 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
| 7482 | btrfs_tree_unlock(next); |
| 7483 | free_extent_buffer(next); |
| 7484 | next = NULL; |
| 7485 | *lookup_info = 1; |
| 7486 | } |
| 7487 | |
| 7488 | if (!next) { |
| 7489 | if (reada && level == 1) |
| 7490 | reada_walk_down(trans, root, wc, path); |
| 7491 | next = read_tree_block(root, bytenr, blocksize, generation); |
| 7492 | if (!next || !extent_buffer_uptodate(next)) { |
| 7493 | free_extent_buffer(next); |
| 7494 | return -EIO; |
| 7495 | } |
| 7496 | btrfs_tree_lock(next); |
| 7497 | btrfs_set_lock_blocking(next); |
| 7498 | } |
| 7499 | |
| 7500 | level--; |
| 7501 | BUG_ON(level != btrfs_header_level(next)); |
| 7502 | path->nodes[level] = next; |
| 7503 | path->slots[level] = 0; |
| 7504 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
| 7505 | wc->level = level; |
| 7506 | if (wc->level == 1) |
| 7507 | wc->reada_slot = 0; |
| 7508 | return 0; |
| 7509 | skip: |
| 7510 | wc->refs[level - 1] = 0; |
| 7511 | wc->flags[level - 1] = 0; |
| 7512 | if (wc->stage == DROP_REFERENCE) { |
| 7513 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { |
| 7514 | parent = path->nodes[level]->start; |
| 7515 | } else { |
| 7516 | BUG_ON(root->root_key.objectid != |
| 7517 | btrfs_header_owner(path->nodes[level])); |
| 7518 | parent = 0; |
| 7519 | } |
| 7520 | |
| 7521 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent, |
| 7522 | root->root_key.objectid, level - 1, 0, 0); |
| 7523 | BUG_ON(ret); /* -ENOMEM */ |
| 7524 | } |
| 7525 | btrfs_tree_unlock(next); |
| 7526 | free_extent_buffer(next); |
| 7527 | *lookup_info = 1; |
| 7528 | return 1; |
| 7529 | } |
| 7530 | |
| 7531 | /* |
| 7532 | * helper to process tree block while walking up the tree. |
| 7533 | * |
| 7534 | * when wc->stage == DROP_REFERENCE, this function drops |
| 7535 | * reference count on the block. |
| 7536 | * |
| 7537 | * when wc->stage == UPDATE_BACKREF, this function changes |
| 7538 | * wc->stage back to DROP_REFERENCE if we changed wc->stage |
| 7539 | * to UPDATE_BACKREF previously while processing the block. |
| 7540 | * |
| 7541 | * NOTE: return value 1 means we should stop walking up. |
| 7542 | */ |
| 7543 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, |
| 7544 | struct btrfs_root *root, |
| 7545 | struct btrfs_path *path, |
| 7546 | struct walk_control *wc) |
| 7547 | { |
| 7548 | int ret; |
| 7549 | int level = wc->level; |
| 7550 | struct extent_buffer *eb = path->nodes[level]; |
| 7551 | u64 parent = 0; |
| 7552 | |
| 7553 | if (wc->stage == UPDATE_BACKREF) { |
| 7554 | BUG_ON(wc->shared_level < level); |
| 7555 | if (level < wc->shared_level) |
| 7556 | goto out; |
| 7557 | |
| 7558 | ret = find_next_key(path, level + 1, &wc->update_progress); |
| 7559 | if (ret > 0) |
| 7560 | wc->update_ref = 0; |
| 7561 | |
| 7562 | wc->stage = DROP_REFERENCE; |
| 7563 | wc->shared_level = -1; |
| 7564 | path->slots[level] = 0; |
| 7565 | |
| 7566 | /* |
| 7567 | * check reference count again if the block isn't locked. |
| 7568 | * we should start walking down the tree again if reference |
| 7569 | * count is one. |
| 7570 | */ |
| 7571 | if (!path->locks[level]) { |
| 7572 | BUG_ON(level == 0); |
| 7573 | btrfs_tree_lock(eb); |
| 7574 | btrfs_set_lock_blocking(eb); |
| 7575 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
| 7576 | |
| 7577 | ret = btrfs_lookup_extent_info(trans, root, |
| 7578 | eb->start, level, 1, |
| 7579 | &wc->refs[level], |
| 7580 | &wc->flags[level]); |
| 7581 | if (ret < 0) { |
| 7582 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
| 7583 | path->locks[level] = 0; |
| 7584 | return ret; |
| 7585 | } |
| 7586 | BUG_ON(wc->refs[level] == 0); |
| 7587 | if (wc->refs[level] == 1) { |
| 7588 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
| 7589 | path->locks[level] = 0; |
| 7590 | return 1; |
| 7591 | } |
| 7592 | } |
| 7593 | } |
| 7594 | |
| 7595 | /* wc->stage == DROP_REFERENCE */ |
| 7596 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); |
| 7597 | |
| 7598 | if (wc->refs[level] == 1) { |
| 7599 | if (level == 0) { |
| 7600 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) |
| 7601 | ret = btrfs_dec_ref(trans, root, eb, 1, |
| 7602 | wc->for_reloc); |
| 7603 | else |
| 7604 | ret = btrfs_dec_ref(trans, root, eb, 0, |
| 7605 | wc->for_reloc); |
| 7606 | BUG_ON(ret); /* -ENOMEM */ |
| 7607 | } |
| 7608 | /* make block locked assertion in clean_tree_block happy */ |
| 7609 | if (!path->locks[level] && |
| 7610 | btrfs_header_generation(eb) == trans->transid) { |
| 7611 | btrfs_tree_lock(eb); |
| 7612 | btrfs_set_lock_blocking(eb); |
| 7613 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
| 7614 | } |
| 7615 | clean_tree_block(trans, root, eb); |
| 7616 | } |
| 7617 | |
| 7618 | if (eb == root->node) { |
| 7619 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) |
| 7620 | parent = eb->start; |
| 7621 | else |
| 7622 | BUG_ON(root->root_key.objectid != |
| 7623 | btrfs_header_owner(eb)); |
| 7624 | } else { |
| 7625 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) |
| 7626 | parent = path->nodes[level + 1]->start; |
| 7627 | else |
| 7628 | BUG_ON(root->root_key.objectid != |
| 7629 | btrfs_header_owner(path->nodes[level + 1])); |
| 7630 | } |
| 7631 | |
| 7632 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
| 7633 | out: |
| 7634 | wc->refs[level] = 0; |
| 7635 | wc->flags[level] = 0; |
| 7636 | return 0; |
| 7637 | } |
| 7638 | |
| 7639 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, |
| 7640 | struct btrfs_root *root, |
| 7641 | struct btrfs_path *path, |
| 7642 | struct walk_control *wc) |
| 7643 | { |
| 7644 | int level = wc->level; |
| 7645 | int lookup_info = 1; |
| 7646 | int ret; |
| 7647 | |
| 7648 | while (level >= 0) { |
| 7649 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
| 7650 | if (ret > 0) |
| 7651 | break; |
| 7652 | |
| 7653 | if (level == 0) |
| 7654 | break; |
| 7655 | |
| 7656 | if (path->slots[level] >= |
| 7657 | btrfs_header_nritems(path->nodes[level])) |
| 7658 | break; |
| 7659 | |
| 7660 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
| 7661 | if (ret > 0) { |
| 7662 | path->slots[level]++; |
| 7663 | continue; |
| 7664 | } else if (ret < 0) |
| 7665 | return ret; |
| 7666 | level = wc->level; |
| 7667 | } |
| 7668 | return 0; |
| 7669 | } |
| 7670 | |
| 7671 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
| 7672 | struct btrfs_root *root, |
| 7673 | struct btrfs_path *path, |
| 7674 | struct walk_control *wc, int max_level) |
| 7675 | { |
| 7676 | int level = wc->level; |
| 7677 | int ret; |
| 7678 | |
| 7679 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
| 7680 | while (level < max_level && path->nodes[level]) { |
| 7681 | wc->level = level; |
| 7682 | if (path->slots[level] + 1 < |
| 7683 | btrfs_header_nritems(path->nodes[level])) { |
| 7684 | path->slots[level]++; |
| 7685 | return 0; |
| 7686 | } else { |
| 7687 | ret = walk_up_proc(trans, root, path, wc); |
| 7688 | if (ret > 0) |
| 7689 | return 0; |
| 7690 | |
| 7691 | if (path->locks[level]) { |
| 7692 | btrfs_tree_unlock_rw(path->nodes[level], |
| 7693 | path->locks[level]); |
| 7694 | path->locks[level] = 0; |
| 7695 | } |
| 7696 | free_extent_buffer(path->nodes[level]); |
| 7697 | path->nodes[level] = NULL; |
| 7698 | level++; |
| 7699 | } |
| 7700 | } |
| 7701 | return 1; |
| 7702 | } |
| 7703 | |
| 7704 | /* |
| 7705 | * drop a subvolume tree. |
| 7706 | * |
| 7707 | * this function traverses the tree freeing any blocks that only |
| 7708 | * referenced by the tree. |
| 7709 | * |
| 7710 | * when a shared tree block is found. this function decreases its |
| 7711 | * reference count by one. if update_ref is true, this function |
| 7712 | * also make sure backrefs for the shared block and all lower level |
| 7713 | * blocks are properly updated. |
| 7714 | * |
| 7715 | * If called with for_reloc == 0, may exit early with -EAGAIN |
| 7716 | */ |
| 7717 | int btrfs_drop_snapshot(struct btrfs_root *root, |
| 7718 | struct btrfs_block_rsv *block_rsv, int update_ref, |
| 7719 | int for_reloc) |
| 7720 | { |
| 7721 | struct btrfs_path *path; |
| 7722 | struct btrfs_trans_handle *trans; |
| 7723 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
| 7724 | struct btrfs_root_item *root_item = &root->root_item; |
| 7725 | struct walk_control *wc; |
| 7726 | struct btrfs_key key; |
| 7727 | int err = 0; |
| 7728 | int ret; |
| 7729 | int level; |
| 7730 | bool root_dropped = false; |
| 7731 | |
| 7732 | path = btrfs_alloc_path(); |
| 7733 | if (!path) { |
| 7734 | err = -ENOMEM; |
| 7735 | goto out; |
| 7736 | } |
| 7737 | |
| 7738 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
| 7739 | if (!wc) { |
| 7740 | btrfs_free_path(path); |
| 7741 | err = -ENOMEM; |
| 7742 | goto out; |
| 7743 | } |
| 7744 | |
| 7745 | trans = btrfs_start_transaction(tree_root, 0); |
| 7746 | if (IS_ERR(trans)) { |
| 7747 | err = PTR_ERR(trans); |
| 7748 | goto out_free; |
| 7749 | } |
| 7750 | |
| 7751 | if (block_rsv) |
| 7752 | trans->block_rsv = block_rsv; |
| 7753 | |
| 7754 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
| 7755 | level = btrfs_header_level(root->node); |
| 7756 | path->nodes[level] = btrfs_lock_root_node(root); |
| 7757 | btrfs_set_lock_blocking(path->nodes[level]); |
| 7758 | path->slots[level] = 0; |
| 7759 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
| 7760 | memset(&wc->update_progress, 0, |
| 7761 | sizeof(wc->update_progress)); |
| 7762 | } else { |
| 7763 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
| 7764 | memcpy(&wc->update_progress, &key, |
| 7765 | sizeof(wc->update_progress)); |
| 7766 | |
| 7767 | level = root_item->drop_level; |
| 7768 | BUG_ON(level == 0); |
| 7769 | path->lowest_level = level; |
| 7770 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
| 7771 | path->lowest_level = 0; |
| 7772 | if (ret < 0) { |
| 7773 | err = ret; |
| 7774 | goto out_end_trans; |
| 7775 | } |
| 7776 | WARN_ON(ret > 0); |
| 7777 | |
| 7778 | /* |
| 7779 | * unlock our path, this is safe because only this |
| 7780 | * function is allowed to delete this snapshot |
| 7781 | */ |
| 7782 | btrfs_unlock_up_safe(path, 0); |
| 7783 | |
| 7784 | level = btrfs_header_level(root->node); |
| 7785 | while (1) { |
| 7786 | btrfs_tree_lock(path->nodes[level]); |
| 7787 | btrfs_set_lock_blocking(path->nodes[level]); |
| 7788 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
| 7789 | |
| 7790 | ret = btrfs_lookup_extent_info(trans, root, |
| 7791 | path->nodes[level]->start, |
| 7792 | level, 1, &wc->refs[level], |
| 7793 | &wc->flags[level]); |
| 7794 | if (ret < 0) { |
| 7795 | err = ret; |
| 7796 | goto out_end_trans; |
| 7797 | } |
| 7798 | BUG_ON(wc->refs[level] == 0); |
| 7799 | |
| 7800 | if (level == root_item->drop_level) |
| 7801 | break; |
| 7802 | |
| 7803 | btrfs_tree_unlock(path->nodes[level]); |
| 7804 | path->locks[level] = 0; |
| 7805 | WARN_ON(wc->refs[level] != 1); |
| 7806 | level--; |
| 7807 | } |
| 7808 | } |
| 7809 | |
| 7810 | wc->level = level; |
| 7811 | wc->shared_level = -1; |
| 7812 | wc->stage = DROP_REFERENCE; |
| 7813 | wc->update_ref = update_ref; |
| 7814 | wc->keep_locks = 0; |
| 7815 | wc->for_reloc = for_reloc; |
| 7816 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
| 7817 | |
| 7818 | while (1) { |
| 7819 | |
| 7820 | ret = walk_down_tree(trans, root, path, wc); |
| 7821 | if (ret < 0) { |
| 7822 | err = ret; |
| 7823 | break; |
| 7824 | } |
| 7825 | |
| 7826 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
| 7827 | if (ret < 0) { |
| 7828 | err = ret; |
| 7829 | break; |
| 7830 | } |
| 7831 | |
| 7832 | if (ret > 0) { |
| 7833 | BUG_ON(wc->stage != DROP_REFERENCE); |
| 7834 | break; |
| 7835 | } |
| 7836 | |
| 7837 | if (wc->stage == DROP_REFERENCE) { |
| 7838 | level = wc->level; |
| 7839 | btrfs_node_key(path->nodes[level], |
| 7840 | &root_item->drop_progress, |
| 7841 | path->slots[level]); |
| 7842 | root_item->drop_level = level; |
| 7843 | } |
| 7844 | |
| 7845 | BUG_ON(wc->level == 0); |
| 7846 | if (btrfs_should_end_transaction(trans, tree_root) || |
| 7847 | (!for_reloc && btrfs_need_cleaner_sleep(root))) { |
| 7848 | ret = btrfs_update_root(trans, tree_root, |
| 7849 | &root->root_key, |
| 7850 | root_item); |
| 7851 | if (ret) { |
| 7852 | btrfs_abort_transaction(trans, tree_root, ret); |
| 7853 | err = ret; |
| 7854 | goto out_end_trans; |
| 7855 | } |
| 7856 | |
| 7857 | btrfs_end_transaction_throttle(trans, tree_root); |
| 7858 | if (!for_reloc && btrfs_need_cleaner_sleep(root)) { |
| 7859 | pr_debug("BTRFS: drop snapshot early exit\n"); |
| 7860 | err = -EAGAIN; |
| 7861 | goto out_free; |
| 7862 | } |
| 7863 | |
| 7864 | trans = btrfs_start_transaction(tree_root, 0); |
| 7865 | if (IS_ERR(trans)) { |
| 7866 | err = PTR_ERR(trans); |
| 7867 | goto out_free; |
| 7868 | } |
| 7869 | if (block_rsv) |
| 7870 | trans->block_rsv = block_rsv; |
| 7871 | } |
| 7872 | } |
| 7873 | btrfs_release_path(path); |
| 7874 | if (err) |
| 7875 | goto out_end_trans; |
| 7876 | |
| 7877 | ret = btrfs_del_root(trans, tree_root, &root->root_key); |
| 7878 | if (ret) { |
| 7879 | btrfs_abort_transaction(trans, tree_root, ret); |
| 7880 | goto out_end_trans; |
| 7881 | } |
| 7882 | |
| 7883 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
| 7884 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
| 7885 | NULL, NULL); |
| 7886 | if (ret < 0) { |
| 7887 | btrfs_abort_transaction(trans, tree_root, ret); |
| 7888 | err = ret; |
| 7889 | goto out_end_trans; |
| 7890 | } else if (ret > 0) { |
| 7891 | /* if we fail to delete the orphan item this time |
| 7892 | * around, it'll get picked up the next time. |
| 7893 | * |
| 7894 | * The most common failure here is just -ENOENT. |
| 7895 | */ |
| 7896 | btrfs_del_orphan_item(trans, tree_root, |
| 7897 | root->root_key.objectid); |
| 7898 | } |
| 7899 | } |
| 7900 | |
| 7901 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
| 7902 | btrfs_drop_and_free_fs_root(tree_root->fs_info, root); |
| 7903 | } else { |
| 7904 | free_extent_buffer(root->node); |
| 7905 | free_extent_buffer(root->commit_root); |
| 7906 | btrfs_put_fs_root(root); |
| 7907 | } |
| 7908 | root_dropped = true; |
| 7909 | out_end_trans: |
| 7910 | btrfs_end_transaction_throttle(trans, tree_root); |
| 7911 | out_free: |
| 7912 | kfree(wc); |
| 7913 | btrfs_free_path(path); |
| 7914 | out: |
| 7915 | /* |
| 7916 | * So if we need to stop dropping the snapshot for whatever reason we |
| 7917 | * need to make sure to add it back to the dead root list so that we |
| 7918 | * keep trying to do the work later. This also cleans up roots if we |
| 7919 | * don't have it in the radix (like when we recover after a power fail |
| 7920 | * or unmount) so we don't leak memory. |
| 7921 | */ |
| 7922 | if (!for_reloc && root_dropped == false) |
| 7923 | btrfs_add_dead_root(root); |
| 7924 | if (err && err != -EAGAIN) |
| 7925 | btrfs_std_error(root->fs_info, err); |
| 7926 | return err; |
| 7927 | } |
| 7928 | |
| 7929 | /* |
| 7930 | * drop subtree rooted at tree block 'node'. |
| 7931 | * |
| 7932 | * NOTE: this function will unlock and release tree block 'node' |
| 7933 | * only used by relocation code |
| 7934 | */ |
| 7935 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
| 7936 | struct btrfs_root *root, |
| 7937 | struct extent_buffer *node, |
| 7938 | struct extent_buffer *parent) |
| 7939 | { |
| 7940 | struct btrfs_path *path; |
| 7941 | struct walk_control *wc; |
| 7942 | int level; |
| 7943 | int parent_level; |
| 7944 | int ret = 0; |
| 7945 | int wret; |
| 7946 | |
| 7947 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
| 7948 | |
| 7949 | path = btrfs_alloc_path(); |
| 7950 | if (!path) |
| 7951 | return -ENOMEM; |
| 7952 | |
| 7953 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
| 7954 | if (!wc) { |
| 7955 | btrfs_free_path(path); |
| 7956 | return -ENOMEM; |
| 7957 | } |
| 7958 | |
| 7959 | btrfs_assert_tree_locked(parent); |
| 7960 | parent_level = btrfs_header_level(parent); |
| 7961 | extent_buffer_get(parent); |
| 7962 | path->nodes[parent_level] = parent; |
| 7963 | path->slots[parent_level] = btrfs_header_nritems(parent); |
| 7964 | |
| 7965 | btrfs_assert_tree_locked(node); |
| 7966 | level = btrfs_header_level(node); |
| 7967 | path->nodes[level] = node; |
| 7968 | path->slots[level] = 0; |
| 7969 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
| 7970 | |
| 7971 | wc->refs[parent_level] = 1; |
| 7972 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
| 7973 | wc->level = level; |
| 7974 | wc->shared_level = -1; |
| 7975 | wc->stage = DROP_REFERENCE; |
| 7976 | wc->update_ref = 0; |
| 7977 | wc->keep_locks = 1; |
| 7978 | wc->for_reloc = 1; |
| 7979 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root); |
| 7980 | |
| 7981 | while (1) { |
| 7982 | wret = walk_down_tree(trans, root, path, wc); |
| 7983 | if (wret < 0) { |
| 7984 | ret = wret; |
| 7985 | break; |
| 7986 | } |
| 7987 | |
| 7988 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
| 7989 | if (wret < 0) |
| 7990 | ret = wret; |
| 7991 | if (wret != 0) |
| 7992 | break; |
| 7993 | } |
| 7994 | |
| 7995 | kfree(wc); |
| 7996 | btrfs_free_path(path); |
| 7997 | return ret; |
| 7998 | } |
| 7999 | |
| 8000 | static u64 update_block_group_flags(struct btrfs_root *root, u64 flags) |
| 8001 | { |
| 8002 | u64 num_devices; |
| 8003 | u64 stripped; |
| 8004 | |
| 8005 | /* |
| 8006 | * if restripe for this chunk_type is on pick target profile and |
| 8007 | * return, otherwise do the usual balance |
| 8008 | */ |
| 8009 | stripped = get_restripe_target(root->fs_info, flags); |
| 8010 | if (stripped) |
| 8011 | return extended_to_chunk(stripped); |
| 8012 | |
| 8013 | /* |
| 8014 | * we add in the count of missing devices because we want |
| 8015 | * to make sure that any RAID levels on a degraded FS |
| 8016 | * continue to be honored. |
| 8017 | */ |
| 8018 | num_devices = root->fs_info->fs_devices->rw_devices + |
| 8019 | root->fs_info->fs_devices->missing_devices; |
| 8020 | |
| 8021 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
| 8022 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
| 8023 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
| 8024 | |
| 8025 | if (num_devices == 1) { |
| 8026 | stripped |= BTRFS_BLOCK_GROUP_DUP; |
| 8027 | stripped = flags & ~stripped; |
| 8028 | |
| 8029 | /* turn raid0 into single device chunks */ |
| 8030 | if (flags & BTRFS_BLOCK_GROUP_RAID0) |
| 8031 | return stripped; |
| 8032 | |
| 8033 | /* turn mirroring into duplication */ |
| 8034 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | |
| 8035 | BTRFS_BLOCK_GROUP_RAID10)) |
| 8036 | return stripped | BTRFS_BLOCK_GROUP_DUP; |
| 8037 | } else { |
| 8038 | /* they already had raid on here, just return */ |
| 8039 | if (flags & stripped) |
| 8040 | return flags; |
| 8041 | |
| 8042 | stripped |= BTRFS_BLOCK_GROUP_DUP; |
| 8043 | stripped = flags & ~stripped; |
| 8044 | |
| 8045 | /* switch duplicated blocks with raid1 */ |
| 8046 | if (flags & BTRFS_BLOCK_GROUP_DUP) |
| 8047 | return stripped | BTRFS_BLOCK_GROUP_RAID1; |
| 8048 | |
| 8049 | /* this is drive concat, leave it alone */ |
| 8050 | } |
| 8051 | |
| 8052 | return flags; |
| 8053 | } |
| 8054 | |
| 8055 | static int set_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
| 8056 | { |
| 8057 | struct btrfs_space_info *sinfo = cache->space_info; |
| 8058 | u64 num_bytes; |
| 8059 | u64 min_allocable_bytes; |
| 8060 | int ret = -ENOSPC; |
| 8061 | |
| 8062 | |
| 8063 | /* |
| 8064 | * We need some metadata space and system metadata space for |
| 8065 | * allocating chunks in some corner cases until we force to set |
| 8066 | * it to be readonly. |
| 8067 | */ |
| 8068 | if ((sinfo->flags & |
| 8069 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && |
| 8070 | !force) |
| 8071 | min_allocable_bytes = 1 * 1024 * 1024; |
| 8072 | else |
| 8073 | min_allocable_bytes = 0; |
| 8074 | |
| 8075 | spin_lock(&sinfo->lock); |
| 8076 | spin_lock(&cache->lock); |
| 8077 | |
| 8078 | if (cache->ro) { |
| 8079 | ret = 0; |
| 8080 | goto out; |
| 8081 | } |
| 8082 | |
| 8083 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
| 8084 | cache->bytes_super - btrfs_block_group_used(&cache->item); |
| 8085 | |
| 8086 | if (sinfo->bytes_used + sinfo->bytes_reserved + sinfo->bytes_pinned + |
| 8087 | sinfo->bytes_may_use + sinfo->bytes_readonly + num_bytes + |
| 8088 | min_allocable_bytes <= sinfo->total_bytes) { |
| 8089 | sinfo->bytes_readonly += num_bytes; |
| 8090 | cache->ro = 1; |
| 8091 | ret = 0; |
| 8092 | } |
| 8093 | out: |
| 8094 | spin_unlock(&cache->lock); |
| 8095 | spin_unlock(&sinfo->lock); |
| 8096 | return ret; |
| 8097 | } |
| 8098 | |
| 8099 | int btrfs_set_block_group_ro(struct btrfs_root *root, |
| 8100 | struct btrfs_block_group_cache *cache) |
| 8101 | |
| 8102 | { |
| 8103 | struct btrfs_trans_handle *trans; |
| 8104 | u64 alloc_flags; |
| 8105 | int ret; |
| 8106 | |
| 8107 | BUG_ON(cache->ro); |
| 8108 | |
| 8109 | trans = btrfs_join_transaction(root); |
| 8110 | if (IS_ERR(trans)) |
| 8111 | return PTR_ERR(trans); |
| 8112 | |
| 8113 | alloc_flags = update_block_group_flags(root, cache->flags); |
| 8114 | if (alloc_flags != cache->flags) { |
| 8115 | ret = do_chunk_alloc(trans, root, alloc_flags, |
| 8116 | CHUNK_ALLOC_FORCE); |
| 8117 | if (ret < 0) |
| 8118 | goto out; |
| 8119 | } |
| 8120 | |
| 8121 | ret = set_block_group_ro(cache, 0); |
| 8122 | if (!ret) |
| 8123 | goto out; |
| 8124 | alloc_flags = get_alloc_profile(root, cache->space_info->flags); |
| 8125 | ret = do_chunk_alloc(trans, root, alloc_flags, |
| 8126 | CHUNK_ALLOC_FORCE); |
| 8127 | if (ret < 0) |
| 8128 | goto out; |
| 8129 | ret = set_block_group_ro(cache, 0); |
| 8130 | out: |
| 8131 | btrfs_end_transaction(trans, root); |
| 8132 | return ret; |
| 8133 | } |
| 8134 | |
| 8135 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
| 8136 | struct btrfs_root *root, u64 type) |
| 8137 | { |
| 8138 | u64 alloc_flags = get_alloc_profile(root, type); |
| 8139 | return do_chunk_alloc(trans, root, alloc_flags, |
| 8140 | CHUNK_ALLOC_FORCE); |
| 8141 | } |
| 8142 | |
| 8143 | /* |
| 8144 | * helper to account the unused space of all the readonly block group in the |
| 8145 | * list. takes mirrors into account. |
| 8146 | */ |
| 8147 | static u64 __btrfs_get_ro_block_group_free_space(struct list_head *groups_list) |
| 8148 | { |
| 8149 | struct btrfs_block_group_cache *block_group; |
| 8150 | u64 free_bytes = 0; |
| 8151 | int factor; |
| 8152 | |
| 8153 | list_for_each_entry(block_group, groups_list, list) { |
| 8154 | spin_lock(&block_group->lock); |
| 8155 | |
| 8156 | if (!block_group->ro) { |
| 8157 | spin_unlock(&block_group->lock); |
| 8158 | continue; |
| 8159 | } |
| 8160 | |
| 8161 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | |
| 8162 | BTRFS_BLOCK_GROUP_RAID10 | |
| 8163 | BTRFS_BLOCK_GROUP_DUP)) |
| 8164 | factor = 2; |
| 8165 | else |
| 8166 | factor = 1; |
| 8167 | |
| 8168 | free_bytes += (block_group->key.offset - |
| 8169 | btrfs_block_group_used(&block_group->item)) * |
| 8170 | factor; |
| 8171 | |
| 8172 | spin_unlock(&block_group->lock); |
| 8173 | } |
| 8174 | |
| 8175 | return free_bytes; |
| 8176 | } |
| 8177 | |
| 8178 | /* |
| 8179 | * helper to account the unused space of all the readonly block group in the |
| 8180 | * space_info. takes mirrors into account. |
| 8181 | */ |
| 8182 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
| 8183 | { |
| 8184 | int i; |
| 8185 | u64 free_bytes = 0; |
| 8186 | |
| 8187 | spin_lock(&sinfo->lock); |
| 8188 | |
| 8189 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
| 8190 | if (!list_empty(&sinfo->block_groups[i])) |
| 8191 | free_bytes += __btrfs_get_ro_block_group_free_space( |
| 8192 | &sinfo->block_groups[i]); |
| 8193 | |
| 8194 | spin_unlock(&sinfo->lock); |
| 8195 | |
| 8196 | return free_bytes; |
| 8197 | } |
| 8198 | |
| 8199 | void btrfs_set_block_group_rw(struct btrfs_root *root, |
| 8200 | struct btrfs_block_group_cache *cache) |
| 8201 | { |
| 8202 | struct btrfs_space_info *sinfo = cache->space_info; |
| 8203 | u64 num_bytes; |
| 8204 | |
| 8205 | BUG_ON(!cache->ro); |
| 8206 | |
| 8207 | spin_lock(&sinfo->lock); |
| 8208 | spin_lock(&cache->lock); |
| 8209 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
| 8210 | cache->bytes_super - btrfs_block_group_used(&cache->item); |
| 8211 | sinfo->bytes_readonly -= num_bytes; |
| 8212 | cache->ro = 0; |
| 8213 | spin_unlock(&cache->lock); |
| 8214 | spin_unlock(&sinfo->lock); |
| 8215 | } |
| 8216 | |
| 8217 | /* |
| 8218 | * checks to see if its even possible to relocate this block group. |
| 8219 | * |
| 8220 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its |
| 8221 | * ok to go ahead and try. |
| 8222 | */ |
| 8223 | int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr) |
| 8224 | { |
| 8225 | struct btrfs_block_group_cache *block_group; |
| 8226 | struct btrfs_space_info *space_info; |
| 8227 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; |
| 8228 | struct btrfs_device *device; |
| 8229 | struct btrfs_trans_handle *trans; |
| 8230 | u64 min_free; |
| 8231 | u64 dev_min = 1; |
| 8232 | u64 dev_nr = 0; |
| 8233 | u64 target; |
| 8234 | int index; |
| 8235 | int full = 0; |
| 8236 | int ret = 0; |
| 8237 | |
| 8238 | block_group = btrfs_lookup_block_group(root->fs_info, bytenr); |
| 8239 | |
| 8240 | /* odd, couldn't find the block group, leave it alone */ |
| 8241 | if (!block_group) |
| 8242 | return -1; |
| 8243 | |
| 8244 | min_free = btrfs_block_group_used(&block_group->item); |
| 8245 | |
| 8246 | /* no bytes used, we're good */ |
| 8247 | if (!min_free) |
| 8248 | goto out; |
| 8249 | |
| 8250 | space_info = block_group->space_info; |
| 8251 | spin_lock(&space_info->lock); |
| 8252 | |
| 8253 | full = space_info->full; |
| 8254 | |
| 8255 | /* |
| 8256 | * if this is the last block group we have in this space, we can't |
| 8257 | * relocate it unless we're able to allocate a new chunk below. |
| 8258 | * |
| 8259 | * Otherwise, we need to make sure we have room in the space to handle |
| 8260 | * all of the extents from this block group. If we can, we're good |
| 8261 | */ |
| 8262 | if ((space_info->total_bytes != block_group->key.offset) && |
| 8263 | (space_info->bytes_used + space_info->bytes_reserved + |
| 8264 | space_info->bytes_pinned + space_info->bytes_readonly + |
| 8265 | min_free < space_info->total_bytes)) { |
| 8266 | spin_unlock(&space_info->lock); |
| 8267 | goto out; |
| 8268 | } |
| 8269 | spin_unlock(&space_info->lock); |
| 8270 | |
| 8271 | /* |
| 8272 | * ok we don't have enough space, but maybe we have free space on our |
| 8273 | * devices to allocate new chunks for relocation, so loop through our |
| 8274 | * alloc devices and guess if we have enough space. if this block |
| 8275 | * group is going to be restriped, run checks against the target |
| 8276 | * profile instead of the current one. |
| 8277 | */ |
| 8278 | ret = -1; |
| 8279 | |
| 8280 | /* |
| 8281 | * index: |
| 8282 | * 0: raid10 |
| 8283 | * 1: raid1 |
| 8284 | * 2: dup |
| 8285 | * 3: raid0 |
| 8286 | * 4: single |
| 8287 | */ |
| 8288 | target = get_restripe_target(root->fs_info, block_group->flags); |
| 8289 | if (target) { |
| 8290 | index = __get_raid_index(extended_to_chunk(target)); |
| 8291 | } else { |
| 8292 | /* |
| 8293 | * this is just a balance, so if we were marked as full |
| 8294 | * we know there is no space for a new chunk |
| 8295 | */ |
| 8296 | if (full) |
| 8297 | goto out; |
| 8298 | |
| 8299 | index = get_block_group_index(block_group); |
| 8300 | } |
| 8301 | |
| 8302 | if (index == BTRFS_RAID_RAID10) { |
| 8303 | dev_min = 4; |
| 8304 | /* Divide by 2 */ |
| 8305 | min_free >>= 1; |
| 8306 | } else if (index == BTRFS_RAID_RAID1) { |
| 8307 | dev_min = 2; |
| 8308 | } else if (index == BTRFS_RAID_DUP) { |
| 8309 | /* Multiply by 2 */ |
| 8310 | min_free <<= 1; |
| 8311 | } else if (index == BTRFS_RAID_RAID0) { |
| 8312 | dev_min = fs_devices->rw_devices; |
| 8313 | do_div(min_free, dev_min); |
| 8314 | } |
| 8315 | |
| 8316 | /* We need to do this so that we can look at pending chunks */ |
| 8317 | trans = btrfs_join_transaction(root); |
| 8318 | if (IS_ERR(trans)) { |
| 8319 | ret = PTR_ERR(trans); |
| 8320 | goto out; |
| 8321 | } |
| 8322 | |
| 8323 | mutex_lock(&root->fs_info->chunk_mutex); |
| 8324 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
| 8325 | u64 dev_offset; |
| 8326 | |
| 8327 | /* |
| 8328 | * check to make sure we can actually find a chunk with enough |
| 8329 | * space to fit our block group in. |
| 8330 | */ |
| 8331 | if (device->total_bytes > device->bytes_used + min_free && |
| 8332 | !device->is_tgtdev_for_dev_replace) { |
| 8333 | ret = find_free_dev_extent(trans, device, min_free, |
| 8334 | &dev_offset, NULL); |
| 8335 | if (!ret) |
| 8336 | dev_nr++; |
| 8337 | |
| 8338 | if (dev_nr >= dev_min) |
| 8339 | break; |
| 8340 | |
| 8341 | ret = -1; |
| 8342 | } |
| 8343 | } |
| 8344 | mutex_unlock(&root->fs_info->chunk_mutex); |
| 8345 | btrfs_end_transaction(trans, root); |
| 8346 | out: |
| 8347 | btrfs_put_block_group(block_group); |
| 8348 | return ret; |
| 8349 | } |
| 8350 | |
| 8351 | static int find_first_block_group(struct btrfs_root *root, |
| 8352 | struct btrfs_path *path, struct btrfs_key *key) |
| 8353 | { |
| 8354 | int ret = 0; |
| 8355 | struct btrfs_key found_key; |
| 8356 | struct extent_buffer *leaf; |
| 8357 | int slot; |
| 8358 | |
| 8359 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
| 8360 | if (ret < 0) |
| 8361 | goto out; |
| 8362 | |
| 8363 | while (1) { |
| 8364 | slot = path->slots[0]; |
| 8365 | leaf = path->nodes[0]; |
| 8366 | if (slot >= btrfs_header_nritems(leaf)) { |
| 8367 | ret = btrfs_next_leaf(root, path); |
| 8368 | if (ret == 0) |
| 8369 | continue; |
| 8370 | if (ret < 0) |
| 8371 | goto out; |
| 8372 | break; |
| 8373 | } |
| 8374 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
| 8375 | |
| 8376 | if (found_key.objectid >= key->objectid && |
| 8377 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
| 8378 | ret = 0; |
| 8379 | goto out; |
| 8380 | } |
| 8381 | path->slots[0]++; |
| 8382 | } |
| 8383 | out: |
| 8384 | return ret; |
| 8385 | } |
| 8386 | |
| 8387 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
| 8388 | { |
| 8389 | struct btrfs_block_group_cache *block_group; |
| 8390 | u64 last = 0; |
| 8391 | |
| 8392 | while (1) { |
| 8393 | struct inode *inode; |
| 8394 | |
| 8395 | block_group = btrfs_lookup_first_block_group(info, last); |
| 8396 | while (block_group) { |
| 8397 | spin_lock(&block_group->lock); |
| 8398 | if (block_group->iref) |
| 8399 | break; |
| 8400 | spin_unlock(&block_group->lock); |
| 8401 | block_group = next_block_group(info->tree_root, |
| 8402 | block_group); |
| 8403 | } |
| 8404 | if (!block_group) { |
| 8405 | if (last == 0) |
| 8406 | break; |
| 8407 | last = 0; |
| 8408 | continue; |
| 8409 | } |
| 8410 | |
| 8411 | inode = block_group->inode; |
| 8412 | block_group->iref = 0; |
| 8413 | block_group->inode = NULL; |
| 8414 | spin_unlock(&block_group->lock); |
| 8415 | iput(inode); |
| 8416 | last = block_group->key.objectid + block_group->key.offset; |
| 8417 | btrfs_put_block_group(block_group); |
| 8418 | } |
| 8419 | } |
| 8420 | |
| 8421 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
| 8422 | { |
| 8423 | struct btrfs_block_group_cache *block_group; |
| 8424 | struct btrfs_space_info *space_info; |
| 8425 | struct btrfs_caching_control *caching_ctl; |
| 8426 | struct rb_node *n; |
| 8427 | |
| 8428 | down_write(&info->commit_root_sem); |
| 8429 | while (!list_empty(&info->caching_block_groups)) { |
| 8430 | caching_ctl = list_entry(info->caching_block_groups.next, |
| 8431 | struct btrfs_caching_control, list); |
| 8432 | list_del(&caching_ctl->list); |
| 8433 | put_caching_control(caching_ctl); |
| 8434 | } |
| 8435 | up_write(&info->commit_root_sem); |
| 8436 | |
| 8437 | spin_lock(&info->block_group_cache_lock); |
| 8438 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { |
| 8439 | block_group = rb_entry(n, struct btrfs_block_group_cache, |
| 8440 | cache_node); |
| 8441 | rb_erase(&block_group->cache_node, |
| 8442 | &info->block_group_cache_tree); |
| 8443 | spin_unlock(&info->block_group_cache_lock); |
| 8444 | |
| 8445 | down_write(&block_group->space_info->groups_sem); |
| 8446 | list_del(&block_group->list); |
| 8447 | up_write(&block_group->space_info->groups_sem); |
| 8448 | |
| 8449 | if (block_group->cached == BTRFS_CACHE_STARTED) |
| 8450 | wait_block_group_cache_done(block_group); |
| 8451 | |
| 8452 | /* |
| 8453 | * We haven't cached this block group, which means we could |
| 8454 | * possibly have excluded extents on this block group. |
| 8455 | */ |
| 8456 | if (block_group->cached == BTRFS_CACHE_NO || |
| 8457 | block_group->cached == BTRFS_CACHE_ERROR) |
| 8458 | free_excluded_extents(info->extent_root, block_group); |
| 8459 | |
| 8460 | btrfs_remove_free_space_cache(block_group); |
| 8461 | btrfs_put_block_group(block_group); |
| 8462 | |
| 8463 | spin_lock(&info->block_group_cache_lock); |
| 8464 | } |
| 8465 | spin_unlock(&info->block_group_cache_lock); |
| 8466 | |
| 8467 | /* now that all the block groups are freed, go through and |
| 8468 | * free all the space_info structs. This is only called during |
| 8469 | * the final stages of unmount, and so we know nobody is |
| 8470 | * using them. We call synchronize_rcu() once before we start, |
| 8471 | * just to be on the safe side. |
| 8472 | */ |
| 8473 | synchronize_rcu(); |
| 8474 | |
| 8475 | release_global_block_rsv(info); |
| 8476 | |
| 8477 | while (!list_empty(&info->space_info)) { |
| 8478 | int i; |
| 8479 | |
| 8480 | space_info = list_entry(info->space_info.next, |
| 8481 | struct btrfs_space_info, |
| 8482 | list); |
| 8483 | if (btrfs_test_opt(info->tree_root, ENOSPC_DEBUG)) { |
| 8484 | if (WARN_ON(space_info->bytes_pinned > 0 || |
| 8485 | space_info->bytes_reserved > 0 || |
| 8486 | space_info->bytes_may_use > 0)) { |
| 8487 | dump_space_info(space_info, 0, 0); |
| 8488 | } |
| 8489 | } |
| 8490 | list_del(&space_info->list); |
| 8491 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
| 8492 | struct kobject *kobj; |
| 8493 | kobj = &space_info->block_group_kobjs[i]; |
| 8494 | if (kobj->parent) { |
| 8495 | kobject_del(kobj); |
| 8496 | kobject_put(kobj); |
| 8497 | } |
| 8498 | } |
| 8499 | kobject_del(&space_info->kobj); |
| 8500 | kobject_put(&space_info->kobj); |
| 8501 | } |
| 8502 | return 0; |
| 8503 | } |
| 8504 | |
| 8505 | static void __link_block_group(struct btrfs_space_info *space_info, |
| 8506 | struct btrfs_block_group_cache *cache) |
| 8507 | { |
| 8508 | int index = get_block_group_index(cache); |
| 8509 | bool first = false; |
| 8510 | |
| 8511 | down_write(&space_info->groups_sem); |
| 8512 | if (list_empty(&space_info->block_groups[index])) |
| 8513 | first = true; |
| 8514 | list_add_tail(&cache->list, &space_info->block_groups[index]); |
| 8515 | up_write(&space_info->groups_sem); |
| 8516 | |
| 8517 | if (first) { |
| 8518 | struct kobject *kobj = &space_info->block_group_kobjs[index]; |
| 8519 | int ret; |
| 8520 | |
| 8521 | kobject_get(&space_info->kobj); /* put in release */ |
| 8522 | ret = kobject_add(kobj, &space_info->kobj, "%s", |
| 8523 | get_raid_name(index)); |
| 8524 | if (ret) { |
| 8525 | pr_warn("BTRFS: failed to add kobject for block cache. ignoring.\n"); |
| 8526 | kobject_put(&space_info->kobj); |
| 8527 | } |
| 8528 | } |
| 8529 | } |
| 8530 | |
| 8531 | static struct btrfs_block_group_cache * |
| 8532 | btrfs_create_block_group_cache(struct btrfs_root *root, u64 start, u64 size) |
| 8533 | { |
| 8534 | struct btrfs_block_group_cache *cache; |
| 8535 | |
| 8536 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
| 8537 | if (!cache) |
| 8538 | return NULL; |
| 8539 | |
| 8540 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
| 8541 | GFP_NOFS); |
| 8542 | if (!cache->free_space_ctl) { |
| 8543 | kfree(cache); |
| 8544 | return NULL; |
| 8545 | } |
| 8546 | |
| 8547 | cache->key.objectid = start; |
| 8548 | cache->key.offset = size; |
| 8549 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
| 8550 | |
| 8551 | cache->sectorsize = root->sectorsize; |
| 8552 | cache->fs_info = root->fs_info; |
| 8553 | cache->full_stripe_len = btrfs_full_stripe_len(root, |
| 8554 | &root->fs_info->mapping_tree, |
| 8555 | start); |
| 8556 | atomic_set(&cache->count, 1); |
| 8557 | spin_lock_init(&cache->lock); |
| 8558 | INIT_LIST_HEAD(&cache->list); |
| 8559 | INIT_LIST_HEAD(&cache->cluster_list); |
| 8560 | INIT_LIST_HEAD(&cache->new_bg_list); |
| 8561 | btrfs_init_free_space_ctl(cache); |
| 8562 | |
| 8563 | return cache; |
| 8564 | } |
| 8565 | |
| 8566 | int btrfs_read_block_groups(struct btrfs_root *root) |
| 8567 | { |
| 8568 | struct btrfs_path *path; |
| 8569 | int ret; |
| 8570 | struct btrfs_block_group_cache *cache; |
| 8571 | struct btrfs_fs_info *info = root->fs_info; |
| 8572 | struct btrfs_space_info *space_info; |
| 8573 | struct btrfs_key key; |
| 8574 | struct btrfs_key found_key; |
| 8575 | struct extent_buffer *leaf; |
| 8576 | int need_clear = 0; |
| 8577 | u64 cache_gen; |
| 8578 | |
| 8579 | root = info->extent_root; |
| 8580 | key.objectid = 0; |
| 8581 | key.offset = 0; |
| 8582 | btrfs_set_key_type(&key, BTRFS_BLOCK_GROUP_ITEM_KEY); |
| 8583 | path = btrfs_alloc_path(); |
| 8584 | if (!path) |
| 8585 | return -ENOMEM; |
| 8586 | path->reada = 1; |
| 8587 | |
| 8588 | cache_gen = btrfs_super_cache_generation(root->fs_info->super_copy); |
| 8589 | if (btrfs_test_opt(root, SPACE_CACHE) && |
| 8590 | btrfs_super_generation(root->fs_info->super_copy) != cache_gen) |
| 8591 | need_clear = 1; |
| 8592 | if (btrfs_test_opt(root, CLEAR_CACHE)) |
| 8593 | need_clear = 1; |
| 8594 | |
| 8595 | while (1) { |
| 8596 | ret = find_first_block_group(root, path, &key); |
| 8597 | if (ret > 0) |
| 8598 | break; |
| 8599 | if (ret != 0) |
| 8600 | goto error; |
| 8601 | |
| 8602 | leaf = path->nodes[0]; |
| 8603 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
| 8604 | |
| 8605 | cache = btrfs_create_block_group_cache(root, found_key.objectid, |
| 8606 | found_key.offset); |
| 8607 | if (!cache) { |
| 8608 | ret = -ENOMEM; |
| 8609 | goto error; |
| 8610 | } |
| 8611 | |
| 8612 | if (need_clear) { |
| 8613 | /* |
| 8614 | * When we mount with old space cache, we need to |
| 8615 | * set BTRFS_DC_CLEAR and set dirty flag. |
| 8616 | * |
| 8617 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we |
| 8618 | * truncate the old free space cache inode and |
| 8619 | * setup a new one. |
| 8620 | * b) Setting 'dirty flag' makes sure that we flush |
| 8621 | * the new space cache info onto disk. |
| 8622 | */ |
| 8623 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
| 8624 | if (btrfs_test_opt(root, SPACE_CACHE)) |
| 8625 | cache->dirty = 1; |
| 8626 | } |
| 8627 | |
| 8628 | read_extent_buffer(leaf, &cache->item, |
| 8629 | btrfs_item_ptr_offset(leaf, path->slots[0]), |
| 8630 | sizeof(cache->item)); |
| 8631 | cache->flags = btrfs_block_group_flags(&cache->item); |
| 8632 | |
| 8633 | key.objectid = found_key.objectid + found_key.offset; |
| 8634 | btrfs_release_path(path); |
| 8635 | |
| 8636 | /* |
| 8637 | * We need to exclude the super stripes now so that the space |
| 8638 | * info has super bytes accounted for, otherwise we'll think |
| 8639 | * we have more space than we actually do. |
| 8640 | */ |
| 8641 | ret = exclude_super_stripes(root, cache); |
| 8642 | if (ret) { |
| 8643 | /* |
| 8644 | * We may have excluded something, so call this just in |
| 8645 | * case. |
| 8646 | */ |
| 8647 | free_excluded_extents(root, cache); |
| 8648 | btrfs_put_block_group(cache); |
| 8649 | goto error; |
| 8650 | } |
| 8651 | |
| 8652 | /* |
| 8653 | * check for two cases, either we are full, and therefore |
| 8654 | * don't need to bother with the caching work since we won't |
| 8655 | * find any space, or we are empty, and we can just add all |
| 8656 | * the space in and be done with it. This saves us _alot_ of |
| 8657 | * time, particularly in the full case. |
| 8658 | */ |
| 8659 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { |
| 8660 | cache->last_byte_to_unpin = (u64)-1; |
| 8661 | cache->cached = BTRFS_CACHE_FINISHED; |
| 8662 | free_excluded_extents(root, cache); |
| 8663 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
| 8664 | cache->last_byte_to_unpin = (u64)-1; |
| 8665 | cache->cached = BTRFS_CACHE_FINISHED; |
| 8666 | add_new_free_space(cache, root->fs_info, |
| 8667 | found_key.objectid, |
| 8668 | found_key.objectid + |
| 8669 | found_key.offset); |
| 8670 | free_excluded_extents(root, cache); |
| 8671 | } |
| 8672 | |
| 8673 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
| 8674 | if (ret) { |
| 8675 | btrfs_remove_free_space_cache(cache); |
| 8676 | btrfs_put_block_group(cache); |
| 8677 | goto error; |
| 8678 | } |
| 8679 | |
| 8680 | ret = update_space_info(info, cache->flags, found_key.offset, |
| 8681 | btrfs_block_group_used(&cache->item), |
| 8682 | &space_info); |
| 8683 | if (ret) { |
| 8684 | btrfs_remove_free_space_cache(cache); |
| 8685 | spin_lock(&info->block_group_cache_lock); |
| 8686 | rb_erase(&cache->cache_node, |
| 8687 | &info->block_group_cache_tree); |
| 8688 | spin_unlock(&info->block_group_cache_lock); |
| 8689 | btrfs_put_block_group(cache); |
| 8690 | goto error; |
| 8691 | } |
| 8692 | |
| 8693 | cache->space_info = space_info; |
| 8694 | spin_lock(&cache->space_info->lock); |
| 8695 | cache->space_info->bytes_readonly += cache->bytes_super; |
| 8696 | spin_unlock(&cache->space_info->lock); |
| 8697 | |
| 8698 | __link_block_group(space_info, cache); |
| 8699 | |
| 8700 | set_avail_alloc_bits(root->fs_info, cache->flags); |
| 8701 | if (btrfs_chunk_readonly(root, cache->key.objectid)) |
| 8702 | set_block_group_ro(cache, 1); |
| 8703 | } |
| 8704 | |
| 8705 | list_for_each_entry_rcu(space_info, &root->fs_info->space_info, list) { |
| 8706 | if (!(get_alloc_profile(root, space_info->flags) & |
| 8707 | (BTRFS_BLOCK_GROUP_RAID10 | |
| 8708 | BTRFS_BLOCK_GROUP_RAID1 | |
| 8709 | BTRFS_BLOCK_GROUP_RAID5 | |
| 8710 | BTRFS_BLOCK_GROUP_RAID6 | |
| 8711 | BTRFS_BLOCK_GROUP_DUP))) |
| 8712 | continue; |
| 8713 | /* |
| 8714 | * avoid allocating from un-mirrored block group if there are |
| 8715 | * mirrored block groups. |
| 8716 | */ |
| 8717 | list_for_each_entry(cache, |
| 8718 | &space_info->block_groups[BTRFS_RAID_RAID0], |
| 8719 | list) |
| 8720 | set_block_group_ro(cache, 1); |
| 8721 | list_for_each_entry(cache, |
| 8722 | &space_info->block_groups[BTRFS_RAID_SINGLE], |
| 8723 | list) |
| 8724 | set_block_group_ro(cache, 1); |
| 8725 | } |
| 8726 | |
| 8727 | init_global_block_rsv(info); |
| 8728 | ret = 0; |
| 8729 | error: |
| 8730 | btrfs_free_path(path); |
| 8731 | return ret; |
| 8732 | } |
| 8733 | |
| 8734 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans, |
| 8735 | struct btrfs_root *root) |
| 8736 | { |
| 8737 | struct btrfs_block_group_cache *block_group, *tmp; |
| 8738 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
| 8739 | struct btrfs_block_group_item item; |
| 8740 | struct btrfs_key key; |
| 8741 | int ret = 0; |
| 8742 | |
| 8743 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, |
| 8744 | new_bg_list) { |
| 8745 | list_del_init(&block_group->new_bg_list); |
| 8746 | |
| 8747 | if (ret) |
| 8748 | continue; |
| 8749 | |
| 8750 | spin_lock(&block_group->lock); |
| 8751 | memcpy(&item, &block_group->item, sizeof(item)); |
| 8752 | memcpy(&key, &block_group->key, sizeof(key)); |
| 8753 | spin_unlock(&block_group->lock); |
| 8754 | |
| 8755 | ret = btrfs_insert_item(trans, extent_root, &key, &item, |
| 8756 | sizeof(item)); |
| 8757 | if (ret) |
| 8758 | btrfs_abort_transaction(trans, extent_root, ret); |
| 8759 | ret = btrfs_finish_chunk_alloc(trans, extent_root, |
| 8760 | key.objectid, key.offset); |
| 8761 | if (ret) |
| 8762 | btrfs_abort_transaction(trans, extent_root, ret); |
| 8763 | } |
| 8764 | } |
| 8765 | |
| 8766 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
| 8767 | struct btrfs_root *root, u64 bytes_used, |
| 8768 | u64 type, u64 chunk_objectid, u64 chunk_offset, |
| 8769 | u64 size) |
| 8770 | { |
| 8771 | int ret; |
| 8772 | struct btrfs_root *extent_root; |
| 8773 | struct btrfs_block_group_cache *cache; |
| 8774 | |
| 8775 | extent_root = root->fs_info->extent_root; |
| 8776 | |
| 8777 | btrfs_set_log_full_commit(root->fs_info, trans); |
| 8778 | |
| 8779 | cache = btrfs_create_block_group_cache(root, chunk_offset, size); |
| 8780 | if (!cache) |
| 8781 | return -ENOMEM; |
| 8782 | |
| 8783 | btrfs_set_block_group_used(&cache->item, bytes_used); |
| 8784 | btrfs_set_block_group_chunk_objectid(&cache->item, chunk_objectid); |
| 8785 | btrfs_set_block_group_flags(&cache->item, type); |
| 8786 | |
| 8787 | cache->flags = type; |
| 8788 | cache->last_byte_to_unpin = (u64)-1; |
| 8789 | cache->cached = BTRFS_CACHE_FINISHED; |
| 8790 | ret = exclude_super_stripes(root, cache); |
| 8791 | if (ret) { |
| 8792 | /* |
| 8793 | * We may have excluded something, so call this just in |
| 8794 | * case. |
| 8795 | */ |
| 8796 | free_excluded_extents(root, cache); |
| 8797 | btrfs_put_block_group(cache); |
| 8798 | return ret; |
| 8799 | } |
| 8800 | |
| 8801 | add_new_free_space(cache, root->fs_info, chunk_offset, |
| 8802 | chunk_offset + size); |
| 8803 | |
| 8804 | free_excluded_extents(root, cache); |
| 8805 | |
| 8806 | ret = btrfs_add_block_group_cache(root->fs_info, cache); |
| 8807 | if (ret) { |
| 8808 | btrfs_remove_free_space_cache(cache); |
| 8809 | btrfs_put_block_group(cache); |
| 8810 | return ret; |
| 8811 | } |
| 8812 | |
| 8813 | ret = update_space_info(root->fs_info, cache->flags, size, bytes_used, |
| 8814 | &cache->space_info); |
| 8815 | if (ret) { |
| 8816 | btrfs_remove_free_space_cache(cache); |
| 8817 | spin_lock(&root->fs_info->block_group_cache_lock); |
| 8818 | rb_erase(&cache->cache_node, |
| 8819 | &root->fs_info->block_group_cache_tree); |
| 8820 | spin_unlock(&root->fs_info->block_group_cache_lock); |
| 8821 | btrfs_put_block_group(cache); |
| 8822 | return ret; |
| 8823 | } |
| 8824 | update_global_block_rsv(root->fs_info); |
| 8825 | |
| 8826 | spin_lock(&cache->space_info->lock); |
| 8827 | cache->space_info->bytes_readonly += cache->bytes_super; |
| 8828 | spin_unlock(&cache->space_info->lock); |
| 8829 | |
| 8830 | __link_block_group(cache->space_info, cache); |
| 8831 | |
| 8832 | list_add_tail(&cache->new_bg_list, &trans->new_bgs); |
| 8833 | |
| 8834 | set_avail_alloc_bits(extent_root->fs_info, type); |
| 8835 | |
| 8836 | return 0; |
| 8837 | } |
| 8838 | |
| 8839 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
| 8840 | { |
| 8841 | u64 extra_flags = chunk_to_extended(flags) & |
| 8842 | BTRFS_EXTENDED_PROFILE_MASK; |
| 8843 | |
| 8844 | write_seqlock(&fs_info->profiles_lock); |
| 8845 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
| 8846 | fs_info->avail_data_alloc_bits &= ~extra_flags; |
| 8847 | if (flags & BTRFS_BLOCK_GROUP_METADATA) |
| 8848 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; |
| 8849 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
| 8850 | fs_info->avail_system_alloc_bits &= ~extra_flags; |
| 8851 | write_sequnlock(&fs_info->profiles_lock); |
| 8852 | } |
| 8853 | |
| 8854 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
| 8855 | struct btrfs_root *root, u64 group_start) |
| 8856 | { |
| 8857 | struct btrfs_path *path; |
| 8858 | struct btrfs_block_group_cache *block_group; |
| 8859 | struct btrfs_free_cluster *cluster; |
| 8860 | struct btrfs_root *tree_root = root->fs_info->tree_root; |
| 8861 | struct btrfs_key key; |
| 8862 | struct inode *inode; |
| 8863 | int ret; |
| 8864 | int index; |
| 8865 | int factor; |
| 8866 | |
| 8867 | root = root->fs_info->extent_root; |
| 8868 | |
| 8869 | block_group = btrfs_lookup_block_group(root->fs_info, group_start); |
| 8870 | BUG_ON(!block_group); |
| 8871 | BUG_ON(!block_group->ro); |
| 8872 | |
| 8873 | /* |
| 8874 | * Free the reserved super bytes from this block group before |
| 8875 | * remove it. |
| 8876 | */ |
| 8877 | free_excluded_extents(root, block_group); |
| 8878 | |
| 8879 | memcpy(&key, &block_group->key, sizeof(key)); |
| 8880 | index = get_block_group_index(block_group); |
| 8881 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
| 8882 | BTRFS_BLOCK_GROUP_RAID1 | |
| 8883 | BTRFS_BLOCK_GROUP_RAID10)) |
| 8884 | factor = 2; |
| 8885 | else |
| 8886 | factor = 1; |
| 8887 | |
| 8888 | /* make sure this block group isn't part of an allocation cluster */ |
| 8889 | cluster = &root->fs_info->data_alloc_cluster; |
| 8890 | spin_lock(&cluster->refill_lock); |
| 8891 | btrfs_return_cluster_to_free_space(block_group, cluster); |
| 8892 | spin_unlock(&cluster->refill_lock); |
| 8893 | |
| 8894 | /* |
| 8895 | * make sure this block group isn't part of a metadata |
| 8896 | * allocation cluster |
| 8897 | */ |
| 8898 | cluster = &root->fs_info->meta_alloc_cluster; |
| 8899 | spin_lock(&cluster->refill_lock); |
| 8900 | btrfs_return_cluster_to_free_space(block_group, cluster); |
| 8901 | spin_unlock(&cluster->refill_lock); |
| 8902 | |
| 8903 | path = btrfs_alloc_path(); |
| 8904 | if (!path) { |
| 8905 | ret = -ENOMEM; |
| 8906 | goto out; |
| 8907 | } |
| 8908 | |
| 8909 | inode = lookup_free_space_inode(tree_root, block_group, path); |
| 8910 | if (!IS_ERR(inode)) { |
| 8911 | ret = btrfs_orphan_add(trans, inode); |
| 8912 | if (ret) { |
| 8913 | btrfs_add_delayed_iput(inode); |
| 8914 | goto out; |
| 8915 | } |
| 8916 | clear_nlink(inode); |
| 8917 | /* One for the block groups ref */ |
| 8918 | spin_lock(&block_group->lock); |
| 8919 | if (block_group->iref) { |
| 8920 | block_group->iref = 0; |
| 8921 | block_group->inode = NULL; |
| 8922 | spin_unlock(&block_group->lock); |
| 8923 | iput(inode); |
| 8924 | } else { |
| 8925 | spin_unlock(&block_group->lock); |
| 8926 | } |
| 8927 | /* One for our lookup ref */ |
| 8928 | btrfs_add_delayed_iput(inode); |
| 8929 | } |
| 8930 | |
| 8931 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; |
| 8932 | key.offset = block_group->key.objectid; |
| 8933 | key.type = 0; |
| 8934 | |
| 8935 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); |
| 8936 | if (ret < 0) |
| 8937 | goto out; |
| 8938 | if (ret > 0) |
| 8939 | btrfs_release_path(path); |
| 8940 | if (ret == 0) { |
| 8941 | ret = btrfs_del_item(trans, tree_root, path); |
| 8942 | if (ret) |
| 8943 | goto out; |
| 8944 | btrfs_release_path(path); |
| 8945 | } |
| 8946 | |
| 8947 | spin_lock(&root->fs_info->block_group_cache_lock); |
| 8948 | rb_erase(&block_group->cache_node, |
| 8949 | &root->fs_info->block_group_cache_tree); |
| 8950 | |
| 8951 | if (root->fs_info->first_logical_byte == block_group->key.objectid) |
| 8952 | root->fs_info->first_logical_byte = (u64)-1; |
| 8953 | spin_unlock(&root->fs_info->block_group_cache_lock); |
| 8954 | |
| 8955 | down_write(&block_group->space_info->groups_sem); |
| 8956 | /* |
| 8957 | * we must use list_del_init so people can check to see if they |
| 8958 | * are still on the list after taking the semaphore |
| 8959 | */ |
| 8960 | list_del_init(&block_group->list); |
| 8961 | if (list_empty(&block_group->space_info->block_groups[index])) { |
| 8962 | kobject_del(&block_group->space_info->block_group_kobjs[index]); |
| 8963 | kobject_put(&block_group->space_info->block_group_kobjs[index]); |
| 8964 | clear_avail_alloc_bits(root->fs_info, block_group->flags); |
| 8965 | } |
| 8966 | up_write(&block_group->space_info->groups_sem); |
| 8967 | |
| 8968 | if (block_group->cached == BTRFS_CACHE_STARTED) |
| 8969 | wait_block_group_cache_done(block_group); |
| 8970 | |
| 8971 | btrfs_remove_free_space_cache(block_group); |
| 8972 | |
| 8973 | spin_lock(&block_group->space_info->lock); |
| 8974 | block_group->space_info->total_bytes -= block_group->key.offset; |
| 8975 | block_group->space_info->bytes_readonly -= block_group->key.offset; |
| 8976 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
| 8977 | spin_unlock(&block_group->space_info->lock); |
| 8978 | |
| 8979 | memcpy(&key, &block_group->key, sizeof(key)); |
| 8980 | |
| 8981 | btrfs_clear_space_info_full(root->fs_info); |
| 8982 | |
| 8983 | btrfs_put_block_group(block_group); |
| 8984 | btrfs_put_block_group(block_group); |
| 8985 | |
| 8986 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
| 8987 | if (ret > 0) |
| 8988 | ret = -EIO; |
| 8989 | if (ret < 0) |
| 8990 | goto out; |
| 8991 | |
| 8992 | ret = btrfs_del_item(trans, root, path); |
| 8993 | out: |
| 8994 | btrfs_free_path(path); |
| 8995 | return ret; |
| 8996 | } |
| 8997 | |
| 8998 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
| 8999 | { |
| 9000 | struct btrfs_space_info *space_info; |
| 9001 | struct btrfs_super_block *disk_super; |
| 9002 | u64 features; |
| 9003 | u64 flags; |
| 9004 | int mixed = 0; |
| 9005 | int ret; |
| 9006 | |
| 9007 | disk_super = fs_info->super_copy; |
| 9008 | if (!btrfs_super_root(disk_super)) |
| 9009 | return 1; |
| 9010 | |
| 9011 | features = btrfs_super_incompat_flags(disk_super); |
| 9012 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) |
| 9013 | mixed = 1; |
| 9014 | |
| 9015 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
| 9016 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); |
| 9017 | if (ret) |
| 9018 | goto out; |
| 9019 | |
| 9020 | if (mixed) { |
| 9021 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; |
| 9022 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); |
| 9023 | } else { |
| 9024 | flags = BTRFS_BLOCK_GROUP_METADATA; |
| 9025 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); |
| 9026 | if (ret) |
| 9027 | goto out; |
| 9028 | |
| 9029 | flags = BTRFS_BLOCK_GROUP_DATA; |
| 9030 | ret = update_space_info(fs_info, flags, 0, 0, &space_info); |
| 9031 | } |
| 9032 | out: |
| 9033 | return ret; |
| 9034 | } |
| 9035 | |
| 9036 | int btrfs_error_unpin_extent_range(struct btrfs_root *root, u64 start, u64 end) |
| 9037 | { |
| 9038 | return unpin_extent_range(root, start, end); |
| 9039 | } |
| 9040 | |
| 9041 | int btrfs_error_discard_extent(struct btrfs_root *root, u64 bytenr, |
| 9042 | u64 num_bytes, u64 *actual_bytes) |
| 9043 | { |
| 9044 | return btrfs_discard_extent(root, bytenr, num_bytes, actual_bytes); |
| 9045 | } |
| 9046 | |
| 9047 | int btrfs_trim_fs(struct btrfs_root *root, struct fstrim_range *range) |
| 9048 | { |
| 9049 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 9050 | struct btrfs_block_group_cache *cache = NULL; |
| 9051 | u64 group_trimmed; |
| 9052 | u64 start; |
| 9053 | u64 end; |
| 9054 | u64 trimmed = 0; |
| 9055 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
| 9056 | int ret = 0; |
| 9057 | |
| 9058 | /* |
| 9059 | * try to trim all FS space, our block group may start from non-zero. |
| 9060 | */ |
| 9061 | if (range->len == total_bytes) |
| 9062 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
| 9063 | else |
| 9064 | cache = btrfs_lookup_block_group(fs_info, range->start); |
| 9065 | |
| 9066 | while (cache) { |
| 9067 | if (cache->key.objectid >= (range->start + range->len)) { |
| 9068 | btrfs_put_block_group(cache); |
| 9069 | break; |
| 9070 | } |
| 9071 | |
| 9072 | start = max(range->start, cache->key.objectid); |
| 9073 | end = min(range->start + range->len, |
| 9074 | cache->key.objectid + cache->key.offset); |
| 9075 | |
| 9076 | if (end - start >= range->minlen) { |
| 9077 | if (!block_group_cache_done(cache)) { |
| 9078 | ret = cache_block_group(cache, 0); |
| 9079 | if (ret) { |
| 9080 | btrfs_put_block_group(cache); |
| 9081 | break; |
| 9082 | } |
| 9083 | ret = wait_block_group_cache_done(cache); |
| 9084 | if (ret) { |
| 9085 | btrfs_put_block_group(cache); |
| 9086 | break; |
| 9087 | } |
| 9088 | } |
| 9089 | ret = btrfs_trim_block_group(cache, |
| 9090 | &group_trimmed, |
| 9091 | start, |
| 9092 | end, |
| 9093 | range->minlen); |
| 9094 | |
| 9095 | trimmed += group_trimmed; |
| 9096 | if (ret) { |
| 9097 | btrfs_put_block_group(cache); |
| 9098 | break; |
| 9099 | } |
| 9100 | } |
| 9101 | |
| 9102 | cache = next_block_group(fs_info->tree_root, cache); |
| 9103 | } |
| 9104 | |
| 9105 | range->len = trimmed; |
| 9106 | return ret; |
| 9107 | } |
| 9108 | |
| 9109 | /* |
| 9110 | * btrfs_{start,end}_write() is similar to mnt_{want, drop}_write(), |
| 9111 | * they are used to prevent the some tasks writing data into the page cache |
| 9112 | * by nocow before the subvolume is snapshoted, but flush the data into |
| 9113 | * the disk after the snapshot creation. |
| 9114 | */ |
| 9115 | void btrfs_end_nocow_write(struct btrfs_root *root) |
| 9116 | { |
| 9117 | percpu_counter_dec(&root->subv_writers->counter); |
| 9118 | /* |
| 9119 | * Make sure counter is updated before we wake up |
| 9120 | * waiters. |
| 9121 | */ |
| 9122 | smp_mb(); |
| 9123 | if (waitqueue_active(&root->subv_writers->wait)) |
| 9124 | wake_up(&root->subv_writers->wait); |
| 9125 | } |
| 9126 | |
| 9127 | int btrfs_start_nocow_write(struct btrfs_root *root) |
| 9128 | { |
| 9129 | if (unlikely(atomic_read(&root->will_be_snapshoted))) |
| 9130 | return 0; |
| 9131 | |
| 9132 | percpu_counter_inc(&root->subv_writers->counter); |
| 9133 | /* |
| 9134 | * Make sure counter is updated before we check for snapshot creation. |
| 9135 | */ |
| 9136 | smp_mb(); |
| 9137 | if (unlikely(atomic_read(&root->will_be_snapshoted))) { |
| 9138 | btrfs_end_nocow_write(root); |
| 9139 | return 0; |
| 9140 | } |
| 9141 | return 1; |
| 9142 | } |