| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * Copyright (C) 2007 Oracle. All rights reserved. |
| 4 | */ |
| 5 | |
| 6 | #include <linux/sched.h> |
| 7 | #include <linux/sched/signal.h> |
| 8 | #include <linux/pagemap.h> |
| 9 | #include <linux/writeback.h> |
| 10 | #include <linux/blkdev.h> |
| 11 | #include <linux/sort.h> |
| 12 | #include <linux/rcupdate.h> |
| 13 | #include <linux/kthread.h> |
| 14 | #include <linux/slab.h> |
| 15 | #include <linux/ratelimit.h> |
| 16 | #include <linux/percpu_counter.h> |
| 17 | #include <linux/lockdep.h> |
| 18 | #include <linux/crc32c.h> |
| 19 | #include "tree-log.h" |
| 20 | #include "disk-io.h" |
| 21 | #include "print-tree.h" |
| 22 | #include "volumes.h" |
| 23 | #include "raid56.h" |
| 24 | #include "locking.h" |
| 25 | #include "free-space-cache.h" |
| 26 | #include "free-space-tree.h" |
| 27 | #include "math.h" |
| 28 | #include "sysfs.h" |
| 29 | #include "qgroup.h" |
| 30 | #include "ref-verify.h" |
| 31 | |
| 32 | #undef SCRAMBLE_DELAYED_REFS |
| 33 | |
| 34 | /* |
| 35 | * control flags for do_chunk_alloc's force field |
| 36 | * CHUNK_ALLOC_NO_FORCE means to only allocate a chunk |
| 37 | * if we really need one. |
| 38 | * |
| 39 | * CHUNK_ALLOC_LIMITED means to only try and allocate one |
| 40 | * if we have very few chunks already allocated. This is |
| 41 | * used as part of the clustering code to help make sure |
| 42 | * we have a good pool of storage to cluster in, without |
| 43 | * filling the FS with empty chunks |
| 44 | * |
| 45 | * CHUNK_ALLOC_FORCE means it must try to allocate one |
| 46 | * |
| 47 | */ |
| 48 | enum { |
| 49 | CHUNK_ALLOC_NO_FORCE = 0, |
| 50 | CHUNK_ALLOC_LIMITED = 1, |
| 51 | CHUNK_ALLOC_FORCE = 2, |
| 52 | }; |
| 53 | |
| 54 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
| 55 | struct btrfs_fs_info *fs_info, |
| 56 | struct btrfs_delayed_ref_node *node, u64 parent, |
| 57 | u64 root_objectid, u64 owner_objectid, |
| 58 | u64 owner_offset, int refs_to_drop, |
| 59 | struct btrfs_delayed_extent_op *extra_op); |
| 60 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
| 61 | struct extent_buffer *leaf, |
| 62 | struct btrfs_extent_item *ei); |
| 63 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
| 64 | struct btrfs_fs_info *fs_info, |
| 65 | u64 parent, u64 root_objectid, |
| 66 | u64 flags, u64 owner, u64 offset, |
| 67 | struct btrfs_key *ins, int ref_mod); |
| 68 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
| 69 | struct btrfs_delayed_ref_node *node, |
| 70 | struct btrfs_delayed_extent_op *extent_op); |
| 71 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
| 72 | struct btrfs_fs_info *fs_info, u64 flags, |
| 73 | int force); |
| 74 | static int find_next_key(struct btrfs_path *path, int level, |
| 75 | struct btrfs_key *key); |
| 76 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
| 77 | struct btrfs_space_info *info, u64 bytes, |
| 78 | int dump_block_groups); |
| 79 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
| 80 | u64 num_bytes); |
| 81 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, |
| 82 | struct btrfs_space_info *space_info, |
| 83 | u64 num_bytes); |
| 84 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, |
| 85 | struct btrfs_space_info *space_info, |
| 86 | u64 num_bytes); |
| 87 | |
| 88 | static noinline int |
| 89 | block_group_cache_done(struct btrfs_block_group_cache *cache) |
| 90 | { |
| 91 | smp_mb(); |
| 92 | return cache->cached == BTRFS_CACHE_FINISHED || |
| 93 | cache->cached == BTRFS_CACHE_ERROR; |
| 94 | } |
| 95 | |
| 96 | static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) |
| 97 | { |
| 98 | return (cache->flags & bits) == bits; |
| 99 | } |
| 100 | |
| 101 | void btrfs_get_block_group(struct btrfs_block_group_cache *cache) |
| 102 | { |
| 103 | atomic_inc(&cache->count); |
| 104 | } |
| 105 | |
| 106 | void btrfs_put_block_group(struct btrfs_block_group_cache *cache) |
| 107 | { |
| 108 | if (atomic_dec_and_test(&cache->count)) { |
| 109 | WARN_ON(cache->pinned > 0); |
| 110 | WARN_ON(cache->reserved > 0); |
| 111 | |
| 112 | /* |
| 113 | * If not empty, someone is still holding mutex of |
| 114 | * full_stripe_lock, which can only be released by caller. |
| 115 | * And it will definitely cause use-after-free when caller |
| 116 | * tries to release full stripe lock. |
| 117 | * |
| 118 | * No better way to resolve, but only to warn. |
| 119 | */ |
| 120 | WARN_ON(!RB_EMPTY_ROOT(&cache->full_stripe_locks_root.root)); |
| 121 | kfree(cache->free_space_ctl); |
| 122 | kfree(cache); |
| 123 | } |
| 124 | } |
| 125 | |
| 126 | /* |
| 127 | * this adds the block group to the fs_info rb tree for the block group |
| 128 | * cache |
| 129 | */ |
| 130 | static int btrfs_add_block_group_cache(struct btrfs_fs_info *info, |
| 131 | struct btrfs_block_group_cache *block_group) |
| 132 | { |
| 133 | struct rb_node **p; |
| 134 | struct rb_node *parent = NULL; |
| 135 | struct btrfs_block_group_cache *cache; |
| 136 | |
| 137 | spin_lock(&info->block_group_cache_lock); |
| 138 | p = &info->block_group_cache_tree.rb_node; |
| 139 | |
| 140 | while (*p) { |
| 141 | parent = *p; |
| 142 | cache = rb_entry(parent, struct btrfs_block_group_cache, |
| 143 | cache_node); |
| 144 | if (block_group->key.objectid < cache->key.objectid) { |
| 145 | p = &(*p)->rb_left; |
| 146 | } else if (block_group->key.objectid > cache->key.objectid) { |
| 147 | p = &(*p)->rb_right; |
| 148 | } else { |
| 149 | spin_unlock(&info->block_group_cache_lock); |
| 150 | return -EEXIST; |
| 151 | } |
| 152 | } |
| 153 | |
| 154 | rb_link_node(&block_group->cache_node, parent, p); |
| 155 | rb_insert_color(&block_group->cache_node, |
| 156 | &info->block_group_cache_tree); |
| 157 | |
| 158 | if (info->first_logical_byte > block_group->key.objectid) |
| 159 | info->first_logical_byte = block_group->key.objectid; |
| 160 | |
| 161 | spin_unlock(&info->block_group_cache_lock); |
| 162 | |
| 163 | return 0; |
| 164 | } |
| 165 | |
| 166 | /* |
| 167 | * This will return the block group at or after bytenr if contains is 0, else |
| 168 | * it will return the block group that contains the bytenr |
| 169 | */ |
| 170 | static struct btrfs_block_group_cache * |
| 171 | block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr, |
| 172 | int contains) |
| 173 | { |
| 174 | struct btrfs_block_group_cache *cache, *ret = NULL; |
| 175 | struct rb_node *n; |
| 176 | u64 end, start; |
| 177 | |
| 178 | spin_lock(&info->block_group_cache_lock); |
| 179 | n = info->block_group_cache_tree.rb_node; |
| 180 | |
| 181 | while (n) { |
| 182 | cache = rb_entry(n, struct btrfs_block_group_cache, |
| 183 | cache_node); |
| 184 | end = cache->key.objectid + cache->key.offset - 1; |
| 185 | start = cache->key.objectid; |
| 186 | |
| 187 | if (bytenr < start) { |
| 188 | if (!contains && (!ret || start < ret->key.objectid)) |
| 189 | ret = cache; |
| 190 | n = n->rb_left; |
| 191 | } else if (bytenr > start) { |
| 192 | if (contains && bytenr <= end) { |
| 193 | ret = cache; |
| 194 | break; |
| 195 | } |
| 196 | n = n->rb_right; |
| 197 | } else { |
| 198 | ret = cache; |
| 199 | break; |
| 200 | } |
| 201 | } |
| 202 | if (ret) { |
| 203 | btrfs_get_block_group(ret); |
| 204 | if (bytenr == 0 && info->first_logical_byte > ret->key.objectid) |
| 205 | info->first_logical_byte = ret->key.objectid; |
| 206 | } |
| 207 | spin_unlock(&info->block_group_cache_lock); |
| 208 | |
| 209 | return ret; |
| 210 | } |
| 211 | |
| 212 | static int add_excluded_extent(struct btrfs_fs_info *fs_info, |
| 213 | u64 start, u64 num_bytes) |
| 214 | { |
| 215 | u64 end = start + num_bytes - 1; |
| 216 | set_extent_bits(&fs_info->freed_extents[0], |
| 217 | start, end, EXTENT_UPTODATE); |
| 218 | set_extent_bits(&fs_info->freed_extents[1], |
| 219 | start, end, EXTENT_UPTODATE); |
| 220 | return 0; |
| 221 | } |
| 222 | |
| 223 | static void free_excluded_extents(struct btrfs_fs_info *fs_info, |
| 224 | struct btrfs_block_group_cache *cache) |
| 225 | { |
| 226 | u64 start, end; |
| 227 | |
| 228 | start = cache->key.objectid; |
| 229 | end = start + cache->key.offset - 1; |
| 230 | |
| 231 | clear_extent_bits(&fs_info->freed_extents[0], |
| 232 | start, end, EXTENT_UPTODATE); |
| 233 | clear_extent_bits(&fs_info->freed_extents[1], |
| 234 | start, end, EXTENT_UPTODATE); |
| 235 | } |
| 236 | |
| 237 | static int exclude_super_stripes(struct btrfs_fs_info *fs_info, |
| 238 | struct btrfs_block_group_cache *cache) |
| 239 | { |
| 240 | u64 bytenr; |
| 241 | u64 *logical; |
| 242 | int stripe_len; |
| 243 | int i, nr, ret; |
| 244 | |
| 245 | if (cache->key.objectid < BTRFS_SUPER_INFO_OFFSET) { |
| 246 | stripe_len = BTRFS_SUPER_INFO_OFFSET - cache->key.objectid; |
| 247 | cache->bytes_super += stripe_len; |
| 248 | ret = add_excluded_extent(fs_info, cache->key.objectid, |
| 249 | stripe_len); |
| 250 | if (ret) |
| 251 | return ret; |
| 252 | } |
| 253 | |
| 254 | for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { |
| 255 | bytenr = btrfs_sb_offset(i); |
| 256 | ret = btrfs_rmap_block(fs_info, cache->key.objectid, |
| 257 | bytenr, &logical, &nr, &stripe_len); |
| 258 | if (ret) |
| 259 | return ret; |
| 260 | |
| 261 | while (nr--) { |
| 262 | u64 start, len; |
| 263 | |
| 264 | if (logical[nr] > cache->key.objectid + |
| 265 | cache->key.offset) |
| 266 | continue; |
| 267 | |
| 268 | if (logical[nr] + stripe_len <= cache->key.objectid) |
| 269 | continue; |
| 270 | |
| 271 | start = logical[nr]; |
| 272 | if (start < cache->key.objectid) { |
| 273 | start = cache->key.objectid; |
| 274 | len = (logical[nr] + stripe_len) - start; |
| 275 | } else { |
| 276 | len = min_t(u64, stripe_len, |
| 277 | cache->key.objectid + |
| 278 | cache->key.offset - start); |
| 279 | } |
| 280 | |
| 281 | cache->bytes_super += len; |
| 282 | ret = add_excluded_extent(fs_info, start, len); |
| 283 | if (ret) { |
| 284 | kfree(logical); |
| 285 | return ret; |
| 286 | } |
| 287 | } |
| 288 | |
| 289 | kfree(logical); |
| 290 | } |
| 291 | return 0; |
| 292 | } |
| 293 | |
| 294 | static struct btrfs_caching_control * |
| 295 | get_caching_control(struct btrfs_block_group_cache *cache) |
| 296 | { |
| 297 | struct btrfs_caching_control *ctl; |
| 298 | |
| 299 | spin_lock(&cache->lock); |
| 300 | if (!cache->caching_ctl) { |
| 301 | spin_unlock(&cache->lock); |
| 302 | return NULL; |
| 303 | } |
| 304 | |
| 305 | ctl = cache->caching_ctl; |
| 306 | refcount_inc(&ctl->count); |
| 307 | spin_unlock(&cache->lock); |
| 308 | return ctl; |
| 309 | } |
| 310 | |
| 311 | static void put_caching_control(struct btrfs_caching_control *ctl) |
| 312 | { |
| 313 | if (refcount_dec_and_test(&ctl->count)) |
| 314 | kfree(ctl); |
| 315 | } |
| 316 | |
| 317 | #ifdef CONFIG_BTRFS_DEBUG |
| 318 | static void fragment_free_space(struct btrfs_block_group_cache *block_group) |
| 319 | { |
| 320 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
| 321 | u64 start = block_group->key.objectid; |
| 322 | u64 len = block_group->key.offset; |
| 323 | u64 chunk = block_group->flags & BTRFS_BLOCK_GROUP_METADATA ? |
| 324 | fs_info->nodesize : fs_info->sectorsize; |
| 325 | u64 step = chunk << 1; |
| 326 | |
| 327 | while (len > chunk) { |
| 328 | btrfs_remove_free_space(block_group, start, chunk); |
| 329 | start += step; |
| 330 | if (len < step) |
| 331 | len = 0; |
| 332 | else |
| 333 | len -= step; |
| 334 | } |
| 335 | } |
| 336 | #endif |
| 337 | |
| 338 | /* |
| 339 | * this is only called by cache_block_group, since we could have freed extents |
| 340 | * we need to check the pinned_extents for any extents that can't be used yet |
| 341 | * since their free space will be released as soon as the transaction commits. |
| 342 | */ |
| 343 | u64 add_new_free_space(struct btrfs_block_group_cache *block_group, |
| 344 | u64 start, u64 end) |
| 345 | { |
| 346 | struct btrfs_fs_info *info = block_group->fs_info; |
| 347 | u64 extent_start, extent_end, size, total_added = 0; |
| 348 | int ret; |
| 349 | |
| 350 | while (start < end) { |
| 351 | ret = find_first_extent_bit(info->pinned_extents, start, |
| 352 | &extent_start, &extent_end, |
| 353 | EXTENT_DIRTY | EXTENT_UPTODATE, |
| 354 | NULL); |
| 355 | if (ret) |
| 356 | break; |
| 357 | |
| 358 | if (extent_start <= start) { |
| 359 | start = extent_end + 1; |
| 360 | } else if (extent_start > start && extent_start < end) { |
| 361 | size = extent_start - start; |
| 362 | total_added += size; |
| 363 | ret = btrfs_add_free_space(block_group, start, |
| 364 | size); |
| 365 | BUG_ON(ret); /* -ENOMEM or logic error */ |
| 366 | start = extent_end + 1; |
| 367 | } else { |
| 368 | break; |
| 369 | } |
| 370 | } |
| 371 | |
| 372 | if (start < end) { |
| 373 | size = end - start; |
| 374 | total_added += size; |
| 375 | ret = btrfs_add_free_space(block_group, start, size); |
| 376 | BUG_ON(ret); /* -ENOMEM or logic error */ |
| 377 | } |
| 378 | |
| 379 | return total_added; |
| 380 | } |
| 381 | |
| 382 | static int load_extent_tree_free(struct btrfs_caching_control *caching_ctl) |
| 383 | { |
| 384 | struct btrfs_block_group_cache *block_group = caching_ctl->block_group; |
| 385 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
| 386 | struct btrfs_root *extent_root = fs_info->extent_root; |
| 387 | struct btrfs_path *path; |
| 388 | struct extent_buffer *leaf; |
| 389 | struct btrfs_key key; |
| 390 | u64 total_found = 0; |
| 391 | u64 last = 0; |
| 392 | u32 nritems; |
| 393 | int ret; |
| 394 | bool wakeup = true; |
| 395 | |
| 396 | path = btrfs_alloc_path(); |
| 397 | if (!path) |
| 398 | return -ENOMEM; |
| 399 | |
| 400 | last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET); |
| 401 | |
| 402 | #ifdef CONFIG_BTRFS_DEBUG |
| 403 | /* |
| 404 | * If we're fragmenting we don't want to make anybody think we can |
| 405 | * allocate from this block group until we've had a chance to fragment |
| 406 | * the free space. |
| 407 | */ |
| 408 | if (btrfs_should_fragment_free_space(block_group)) |
| 409 | wakeup = false; |
| 410 | #endif |
| 411 | /* |
| 412 | * We don't want to deadlock with somebody trying to allocate a new |
| 413 | * extent for the extent root while also trying to search the extent |
| 414 | * root to add free space. So we skip locking and search the commit |
| 415 | * root, since its read-only |
| 416 | */ |
| 417 | path->skip_locking = 1; |
| 418 | path->search_commit_root = 1; |
| 419 | path->reada = READA_FORWARD; |
| 420 | |
| 421 | key.objectid = last; |
| 422 | key.offset = 0; |
| 423 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 424 | |
| 425 | next: |
| 426 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
| 427 | if (ret < 0) |
| 428 | goto out; |
| 429 | |
| 430 | leaf = path->nodes[0]; |
| 431 | nritems = btrfs_header_nritems(leaf); |
| 432 | |
| 433 | while (1) { |
| 434 | if (btrfs_fs_closing(fs_info) > 1) { |
| 435 | last = (u64)-1; |
| 436 | break; |
| 437 | } |
| 438 | |
| 439 | if (path->slots[0] < nritems) { |
| 440 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
| 441 | } else { |
| 442 | ret = find_next_key(path, 0, &key); |
| 443 | if (ret) |
| 444 | break; |
| 445 | |
| 446 | if (need_resched() || |
| 447 | rwsem_is_contended(&fs_info->commit_root_sem)) { |
| 448 | if (wakeup) |
| 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 | mutex_lock(&caching_ctl->mutex); |
| 455 | down_read(&fs_info->commit_root_sem); |
| 456 | goto next; |
| 457 | } |
| 458 | |
| 459 | ret = btrfs_next_leaf(extent_root, path); |
| 460 | if (ret < 0) |
| 461 | goto out; |
| 462 | if (ret) |
| 463 | break; |
| 464 | leaf = path->nodes[0]; |
| 465 | nritems = btrfs_header_nritems(leaf); |
| 466 | continue; |
| 467 | } |
| 468 | |
| 469 | if (key.objectid < last) { |
| 470 | key.objectid = last; |
| 471 | key.offset = 0; |
| 472 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 473 | |
| 474 | if (wakeup) |
| 475 | caching_ctl->progress = last; |
| 476 | btrfs_release_path(path); |
| 477 | goto next; |
| 478 | } |
| 479 | |
| 480 | if (key.objectid < block_group->key.objectid) { |
| 481 | path->slots[0]++; |
| 482 | continue; |
| 483 | } |
| 484 | |
| 485 | if (key.objectid >= block_group->key.objectid + |
| 486 | block_group->key.offset) |
| 487 | break; |
| 488 | |
| 489 | if (key.type == BTRFS_EXTENT_ITEM_KEY || |
| 490 | key.type == BTRFS_METADATA_ITEM_KEY) { |
| 491 | total_found += add_new_free_space(block_group, last, |
| 492 | key.objectid); |
| 493 | if (key.type == BTRFS_METADATA_ITEM_KEY) |
| 494 | last = key.objectid + |
| 495 | fs_info->nodesize; |
| 496 | else |
| 497 | last = key.objectid + key.offset; |
| 498 | |
| 499 | if (total_found > CACHING_CTL_WAKE_UP) { |
| 500 | total_found = 0; |
| 501 | if (wakeup) |
| 502 | wake_up(&caching_ctl->wait); |
| 503 | } |
| 504 | } |
| 505 | path->slots[0]++; |
| 506 | } |
| 507 | ret = 0; |
| 508 | |
| 509 | total_found += add_new_free_space(block_group, last, |
| 510 | block_group->key.objectid + |
| 511 | block_group->key.offset); |
| 512 | caching_ctl->progress = (u64)-1; |
| 513 | |
| 514 | out: |
| 515 | btrfs_free_path(path); |
| 516 | return ret; |
| 517 | } |
| 518 | |
| 519 | static noinline void caching_thread(struct btrfs_work *work) |
| 520 | { |
| 521 | struct btrfs_block_group_cache *block_group; |
| 522 | struct btrfs_fs_info *fs_info; |
| 523 | struct btrfs_caching_control *caching_ctl; |
| 524 | int ret; |
| 525 | |
| 526 | caching_ctl = container_of(work, struct btrfs_caching_control, work); |
| 527 | block_group = caching_ctl->block_group; |
| 528 | fs_info = block_group->fs_info; |
| 529 | |
| 530 | mutex_lock(&caching_ctl->mutex); |
| 531 | down_read(&fs_info->commit_root_sem); |
| 532 | |
| 533 | if (btrfs_fs_compat_ro(fs_info, FREE_SPACE_TREE)) |
| 534 | ret = load_free_space_tree(caching_ctl); |
| 535 | else |
| 536 | ret = load_extent_tree_free(caching_ctl); |
| 537 | |
| 538 | spin_lock(&block_group->lock); |
| 539 | block_group->caching_ctl = NULL; |
| 540 | block_group->cached = ret ? BTRFS_CACHE_ERROR : BTRFS_CACHE_FINISHED; |
| 541 | spin_unlock(&block_group->lock); |
| 542 | |
| 543 | #ifdef CONFIG_BTRFS_DEBUG |
| 544 | if (btrfs_should_fragment_free_space(block_group)) { |
| 545 | u64 bytes_used; |
| 546 | |
| 547 | spin_lock(&block_group->space_info->lock); |
| 548 | spin_lock(&block_group->lock); |
| 549 | bytes_used = block_group->key.offset - |
| 550 | btrfs_block_group_used(&block_group->item); |
| 551 | block_group->space_info->bytes_used += bytes_used >> 1; |
| 552 | spin_unlock(&block_group->lock); |
| 553 | spin_unlock(&block_group->space_info->lock); |
| 554 | fragment_free_space(block_group); |
| 555 | } |
| 556 | #endif |
| 557 | |
| 558 | caching_ctl->progress = (u64)-1; |
| 559 | |
| 560 | up_read(&fs_info->commit_root_sem); |
| 561 | free_excluded_extents(fs_info, block_group); |
| 562 | mutex_unlock(&caching_ctl->mutex); |
| 563 | |
| 564 | wake_up(&caching_ctl->wait); |
| 565 | |
| 566 | put_caching_control(caching_ctl); |
| 567 | btrfs_put_block_group(block_group); |
| 568 | } |
| 569 | |
| 570 | static int cache_block_group(struct btrfs_block_group_cache *cache, |
| 571 | int load_cache_only) |
| 572 | { |
| 573 | DEFINE_WAIT(wait); |
| 574 | struct btrfs_fs_info *fs_info = cache->fs_info; |
| 575 | struct btrfs_caching_control *caching_ctl; |
| 576 | int ret = 0; |
| 577 | |
| 578 | caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS); |
| 579 | if (!caching_ctl) |
| 580 | return -ENOMEM; |
| 581 | |
| 582 | INIT_LIST_HEAD(&caching_ctl->list); |
| 583 | mutex_init(&caching_ctl->mutex); |
| 584 | init_waitqueue_head(&caching_ctl->wait); |
| 585 | caching_ctl->block_group = cache; |
| 586 | caching_ctl->progress = cache->key.objectid; |
| 587 | refcount_set(&caching_ctl->count, 1); |
| 588 | btrfs_init_work(&caching_ctl->work, btrfs_cache_helper, |
| 589 | caching_thread, NULL, NULL); |
| 590 | |
| 591 | spin_lock(&cache->lock); |
| 592 | /* |
| 593 | * This should be a rare occasion, but this could happen I think in the |
| 594 | * case where one thread starts to load the space cache info, and then |
| 595 | * some other thread starts a transaction commit which tries to do an |
| 596 | * allocation while the other thread is still loading the space cache |
| 597 | * info. The previous loop should have kept us from choosing this block |
| 598 | * group, but if we've moved to the state where we will wait on caching |
| 599 | * block groups we need to first check if we're doing a fast load here, |
| 600 | * so we can wait for it to finish, otherwise we could end up allocating |
| 601 | * from a block group who's cache gets evicted for one reason or |
| 602 | * another. |
| 603 | */ |
| 604 | while (cache->cached == BTRFS_CACHE_FAST) { |
| 605 | struct btrfs_caching_control *ctl; |
| 606 | |
| 607 | ctl = cache->caching_ctl; |
| 608 | refcount_inc(&ctl->count); |
| 609 | prepare_to_wait(&ctl->wait, &wait, TASK_UNINTERRUPTIBLE); |
| 610 | spin_unlock(&cache->lock); |
| 611 | |
| 612 | schedule(); |
| 613 | |
| 614 | finish_wait(&ctl->wait, &wait); |
| 615 | put_caching_control(ctl); |
| 616 | spin_lock(&cache->lock); |
| 617 | } |
| 618 | |
| 619 | if (cache->cached != BTRFS_CACHE_NO) { |
| 620 | spin_unlock(&cache->lock); |
| 621 | kfree(caching_ctl); |
| 622 | return 0; |
| 623 | } |
| 624 | WARN_ON(cache->caching_ctl); |
| 625 | cache->caching_ctl = caching_ctl; |
| 626 | cache->cached = BTRFS_CACHE_FAST; |
| 627 | spin_unlock(&cache->lock); |
| 628 | |
| 629 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) { |
| 630 | mutex_lock(&caching_ctl->mutex); |
| 631 | ret = load_free_space_cache(fs_info, cache); |
| 632 | |
| 633 | spin_lock(&cache->lock); |
| 634 | if (ret == 1) { |
| 635 | cache->caching_ctl = NULL; |
| 636 | cache->cached = BTRFS_CACHE_FINISHED; |
| 637 | cache->last_byte_to_unpin = (u64)-1; |
| 638 | caching_ctl->progress = (u64)-1; |
| 639 | } else { |
| 640 | if (load_cache_only) { |
| 641 | cache->caching_ctl = NULL; |
| 642 | cache->cached = BTRFS_CACHE_NO; |
| 643 | } else { |
| 644 | cache->cached = BTRFS_CACHE_STARTED; |
| 645 | cache->has_caching_ctl = 1; |
| 646 | } |
| 647 | } |
| 648 | spin_unlock(&cache->lock); |
| 649 | #ifdef CONFIG_BTRFS_DEBUG |
| 650 | if (ret == 1 && |
| 651 | btrfs_should_fragment_free_space(cache)) { |
| 652 | u64 bytes_used; |
| 653 | |
| 654 | spin_lock(&cache->space_info->lock); |
| 655 | spin_lock(&cache->lock); |
| 656 | bytes_used = cache->key.offset - |
| 657 | btrfs_block_group_used(&cache->item); |
| 658 | cache->space_info->bytes_used += bytes_used >> 1; |
| 659 | spin_unlock(&cache->lock); |
| 660 | spin_unlock(&cache->space_info->lock); |
| 661 | fragment_free_space(cache); |
| 662 | } |
| 663 | #endif |
| 664 | mutex_unlock(&caching_ctl->mutex); |
| 665 | |
| 666 | wake_up(&caching_ctl->wait); |
| 667 | if (ret == 1) { |
| 668 | put_caching_control(caching_ctl); |
| 669 | free_excluded_extents(fs_info, cache); |
| 670 | return 0; |
| 671 | } |
| 672 | } else { |
| 673 | /* |
| 674 | * We're either using the free space tree or no caching at all. |
| 675 | * Set cached to the appropriate value and wakeup any waiters. |
| 676 | */ |
| 677 | spin_lock(&cache->lock); |
| 678 | if (load_cache_only) { |
| 679 | cache->caching_ctl = NULL; |
| 680 | cache->cached = BTRFS_CACHE_NO; |
| 681 | } else { |
| 682 | cache->cached = BTRFS_CACHE_STARTED; |
| 683 | cache->has_caching_ctl = 1; |
| 684 | } |
| 685 | spin_unlock(&cache->lock); |
| 686 | wake_up(&caching_ctl->wait); |
| 687 | } |
| 688 | |
| 689 | if (load_cache_only) { |
| 690 | put_caching_control(caching_ctl); |
| 691 | return 0; |
| 692 | } |
| 693 | |
| 694 | down_write(&fs_info->commit_root_sem); |
| 695 | refcount_inc(&caching_ctl->count); |
| 696 | list_add_tail(&caching_ctl->list, &fs_info->caching_block_groups); |
| 697 | up_write(&fs_info->commit_root_sem); |
| 698 | |
| 699 | btrfs_get_block_group(cache); |
| 700 | |
| 701 | btrfs_queue_work(fs_info->caching_workers, &caching_ctl->work); |
| 702 | |
| 703 | return ret; |
| 704 | } |
| 705 | |
| 706 | /* |
| 707 | * return the block group that starts at or after bytenr |
| 708 | */ |
| 709 | static struct btrfs_block_group_cache * |
| 710 | btrfs_lookup_first_block_group(struct btrfs_fs_info *info, u64 bytenr) |
| 711 | { |
| 712 | return block_group_cache_tree_search(info, bytenr, 0); |
| 713 | } |
| 714 | |
| 715 | /* |
| 716 | * return the block group that contains the given bytenr |
| 717 | */ |
| 718 | struct btrfs_block_group_cache *btrfs_lookup_block_group( |
| 719 | struct btrfs_fs_info *info, |
| 720 | u64 bytenr) |
| 721 | { |
| 722 | return block_group_cache_tree_search(info, bytenr, 1); |
| 723 | } |
| 724 | |
| 725 | static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, |
| 726 | u64 flags) |
| 727 | { |
| 728 | struct list_head *head = &info->space_info; |
| 729 | struct btrfs_space_info *found; |
| 730 | |
| 731 | flags &= BTRFS_BLOCK_GROUP_TYPE_MASK; |
| 732 | |
| 733 | rcu_read_lock(); |
| 734 | list_for_each_entry_rcu(found, head, list) { |
| 735 | if (found->flags & flags) { |
| 736 | rcu_read_unlock(); |
| 737 | return found; |
| 738 | } |
| 739 | } |
| 740 | rcu_read_unlock(); |
| 741 | return NULL; |
| 742 | } |
| 743 | |
| 744 | static void add_pinned_bytes(struct btrfs_fs_info *fs_info, s64 num_bytes, |
| 745 | bool metadata, u64 root_objectid) |
| 746 | { |
| 747 | struct btrfs_space_info *space_info; |
| 748 | u64 flags; |
| 749 | |
| 750 | if (metadata) { |
| 751 | if (root_objectid == BTRFS_CHUNK_TREE_OBJECTID) |
| 752 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
| 753 | else |
| 754 | flags = BTRFS_BLOCK_GROUP_METADATA; |
| 755 | } else { |
| 756 | flags = BTRFS_BLOCK_GROUP_DATA; |
| 757 | } |
| 758 | |
| 759 | space_info = __find_space_info(fs_info, flags); |
| 760 | ASSERT(space_info); |
| 761 | percpu_counter_add(&space_info->total_bytes_pinned, num_bytes); |
| 762 | } |
| 763 | |
| 764 | /* |
| 765 | * after adding space to the filesystem, we need to clear the full flags |
| 766 | * on all the space infos. |
| 767 | */ |
| 768 | void btrfs_clear_space_info_full(struct btrfs_fs_info *info) |
| 769 | { |
| 770 | struct list_head *head = &info->space_info; |
| 771 | struct btrfs_space_info *found; |
| 772 | |
| 773 | rcu_read_lock(); |
| 774 | list_for_each_entry_rcu(found, head, list) |
| 775 | found->full = 0; |
| 776 | rcu_read_unlock(); |
| 777 | } |
| 778 | |
| 779 | /* simple helper to search for an existing data extent at a given offset */ |
| 780 | int btrfs_lookup_data_extent(struct btrfs_fs_info *fs_info, u64 start, u64 len) |
| 781 | { |
| 782 | int ret; |
| 783 | struct btrfs_key key; |
| 784 | struct btrfs_path *path; |
| 785 | |
| 786 | path = btrfs_alloc_path(); |
| 787 | if (!path) |
| 788 | return -ENOMEM; |
| 789 | |
| 790 | key.objectid = start; |
| 791 | key.offset = len; |
| 792 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 793 | ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0); |
| 794 | btrfs_free_path(path); |
| 795 | return ret; |
| 796 | } |
| 797 | |
| 798 | /* |
| 799 | * helper function to lookup reference count and flags of a tree block. |
| 800 | * |
| 801 | * the head node for delayed ref is used to store the sum of all the |
| 802 | * reference count modifications queued up in the rbtree. the head |
| 803 | * node may also store the extent flags to set. This way you can check |
| 804 | * to see what the reference count and extent flags would be if all of |
| 805 | * the delayed refs are not processed. |
| 806 | */ |
| 807 | int btrfs_lookup_extent_info(struct btrfs_trans_handle *trans, |
| 808 | struct btrfs_fs_info *fs_info, u64 bytenr, |
| 809 | u64 offset, int metadata, u64 *refs, u64 *flags) |
| 810 | { |
| 811 | struct btrfs_delayed_ref_head *head; |
| 812 | struct btrfs_delayed_ref_root *delayed_refs; |
| 813 | struct btrfs_path *path; |
| 814 | struct btrfs_extent_item *ei; |
| 815 | struct extent_buffer *leaf; |
| 816 | struct btrfs_key key; |
| 817 | u32 item_size; |
| 818 | u64 num_refs; |
| 819 | u64 extent_flags; |
| 820 | int ret; |
| 821 | |
| 822 | /* |
| 823 | * If we don't have skinny metadata, don't bother doing anything |
| 824 | * different |
| 825 | */ |
| 826 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) { |
| 827 | offset = fs_info->nodesize; |
| 828 | metadata = 0; |
| 829 | } |
| 830 | |
| 831 | path = btrfs_alloc_path(); |
| 832 | if (!path) |
| 833 | return -ENOMEM; |
| 834 | |
| 835 | if (!trans) { |
| 836 | path->skip_locking = 1; |
| 837 | path->search_commit_root = 1; |
| 838 | } |
| 839 | |
| 840 | search_again: |
| 841 | key.objectid = bytenr; |
| 842 | key.offset = offset; |
| 843 | if (metadata) |
| 844 | key.type = BTRFS_METADATA_ITEM_KEY; |
| 845 | else |
| 846 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 847 | |
| 848 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0); |
| 849 | if (ret < 0) |
| 850 | goto out_free; |
| 851 | |
| 852 | if (ret > 0 && metadata && key.type == BTRFS_METADATA_ITEM_KEY) { |
| 853 | if (path->slots[0]) { |
| 854 | path->slots[0]--; |
| 855 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
| 856 | path->slots[0]); |
| 857 | if (key.objectid == bytenr && |
| 858 | key.type == BTRFS_EXTENT_ITEM_KEY && |
| 859 | key.offset == fs_info->nodesize) |
| 860 | ret = 0; |
| 861 | } |
| 862 | } |
| 863 | |
| 864 | if (ret == 0) { |
| 865 | leaf = path->nodes[0]; |
| 866 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); |
| 867 | if (item_size >= sizeof(*ei)) { |
| 868 | ei = btrfs_item_ptr(leaf, path->slots[0], |
| 869 | struct btrfs_extent_item); |
| 870 | num_refs = btrfs_extent_refs(leaf, ei); |
| 871 | extent_flags = btrfs_extent_flags(leaf, ei); |
| 872 | } else { |
| 873 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 874 | struct btrfs_extent_item_v0 *ei0; |
| 875 | BUG_ON(item_size != sizeof(*ei0)); |
| 876 | ei0 = btrfs_item_ptr(leaf, path->slots[0], |
| 877 | struct btrfs_extent_item_v0); |
| 878 | num_refs = btrfs_extent_refs_v0(leaf, ei0); |
| 879 | /* FIXME: this isn't correct for data */ |
| 880 | extent_flags = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
| 881 | #else |
| 882 | BUG(); |
| 883 | #endif |
| 884 | } |
| 885 | BUG_ON(num_refs == 0); |
| 886 | } else { |
| 887 | num_refs = 0; |
| 888 | extent_flags = 0; |
| 889 | ret = 0; |
| 890 | } |
| 891 | |
| 892 | if (!trans) |
| 893 | goto out; |
| 894 | |
| 895 | delayed_refs = &trans->transaction->delayed_refs; |
| 896 | spin_lock(&delayed_refs->lock); |
| 897 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
| 898 | if (head) { |
| 899 | if (!mutex_trylock(&head->mutex)) { |
| 900 | refcount_inc(&head->refs); |
| 901 | spin_unlock(&delayed_refs->lock); |
| 902 | |
| 903 | btrfs_release_path(path); |
| 904 | |
| 905 | /* |
| 906 | * Mutex was contended, block until it's released and try |
| 907 | * again |
| 908 | */ |
| 909 | mutex_lock(&head->mutex); |
| 910 | mutex_unlock(&head->mutex); |
| 911 | btrfs_put_delayed_ref_head(head); |
| 912 | goto search_again; |
| 913 | } |
| 914 | spin_lock(&head->lock); |
| 915 | if (head->extent_op && head->extent_op->update_flags) |
| 916 | extent_flags |= head->extent_op->flags_to_set; |
| 917 | else |
| 918 | BUG_ON(num_refs == 0); |
| 919 | |
| 920 | num_refs += head->ref_mod; |
| 921 | spin_unlock(&head->lock); |
| 922 | mutex_unlock(&head->mutex); |
| 923 | } |
| 924 | spin_unlock(&delayed_refs->lock); |
| 925 | out: |
| 926 | WARN_ON(num_refs == 0); |
| 927 | if (refs) |
| 928 | *refs = num_refs; |
| 929 | if (flags) |
| 930 | *flags = extent_flags; |
| 931 | out_free: |
| 932 | btrfs_free_path(path); |
| 933 | return ret; |
| 934 | } |
| 935 | |
| 936 | /* |
| 937 | * Back reference rules. Back refs have three main goals: |
| 938 | * |
| 939 | * 1) differentiate between all holders of references to an extent so that |
| 940 | * when a reference is dropped we can make sure it was a valid reference |
| 941 | * before freeing the extent. |
| 942 | * |
| 943 | * 2) Provide enough information to quickly find the holders of an extent |
| 944 | * if we notice a given block is corrupted or bad. |
| 945 | * |
| 946 | * 3) Make it easy to migrate blocks for FS shrinking or storage pool |
| 947 | * maintenance. This is actually the same as #2, but with a slightly |
| 948 | * different use case. |
| 949 | * |
| 950 | * There are two kinds of back refs. The implicit back refs is optimized |
| 951 | * for pointers in non-shared tree blocks. For a given pointer in a block, |
| 952 | * back refs of this kind provide information about the block's owner tree |
| 953 | * and the pointer's key. These information allow us to find the block by |
| 954 | * b-tree searching. The full back refs is for pointers in tree blocks not |
| 955 | * referenced by their owner trees. The location of tree block is recorded |
| 956 | * in the back refs. Actually the full back refs is generic, and can be |
| 957 | * used in all cases the implicit back refs is used. The major shortcoming |
| 958 | * of the full back refs is its overhead. Every time a tree block gets |
| 959 | * COWed, we have to update back refs entry for all pointers in it. |
| 960 | * |
| 961 | * For a newly allocated tree block, we use implicit back refs for |
| 962 | * pointers in it. This means most tree related operations only involve |
| 963 | * implicit back refs. For a tree block created in old transaction, the |
| 964 | * only way to drop a reference to it is COW it. So we can detect the |
| 965 | * event that tree block loses its owner tree's reference and do the |
| 966 | * back refs conversion. |
| 967 | * |
| 968 | * When a tree block is COWed through a tree, there are four cases: |
| 969 | * |
| 970 | * The reference count of the block is one and the tree is the block's |
| 971 | * owner tree. Nothing to do in this case. |
| 972 | * |
| 973 | * The reference count of the block is one and the tree is not the |
| 974 | * block's owner tree. In this case, full back refs is used for pointers |
| 975 | * in the block. Remove these full back refs, add implicit back refs for |
| 976 | * every pointers in the new block. |
| 977 | * |
| 978 | * The reference count of the block is greater than one and the tree is |
| 979 | * the block's owner tree. In this case, implicit back refs is used for |
| 980 | * pointers in the block. Add full back refs for every pointers in the |
| 981 | * block, increase lower level extents' reference counts. The original |
| 982 | * implicit back refs are entailed to the new block. |
| 983 | * |
| 984 | * The reference count of the block is greater than one and the tree is |
| 985 | * not the block's owner tree. Add implicit back refs for every pointer in |
| 986 | * the new block, increase lower level extents' reference count. |
| 987 | * |
| 988 | * Back Reference Key composing: |
| 989 | * |
| 990 | * The key objectid corresponds to the first byte in the extent, |
| 991 | * The key type is used to differentiate between types of back refs. |
| 992 | * There are different meanings of the key offset for different types |
| 993 | * of back refs. |
| 994 | * |
| 995 | * File extents can be referenced by: |
| 996 | * |
| 997 | * - multiple snapshots, subvolumes, or different generations in one subvol |
| 998 | * - different files inside a single subvolume |
| 999 | * - different offsets inside a file (bookend extents in file.c) |
| 1000 | * |
| 1001 | * The extent ref structure for the implicit back refs has fields for: |
| 1002 | * |
| 1003 | * - Objectid of the subvolume root |
| 1004 | * - objectid of the file holding the reference |
| 1005 | * - original offset in the file |
| 1006 | * - how many bookend extents |
| 1007 | * |
| 1008 | * The key offset for the implicit back refs is hash of the first |
| 1009 | * three fields. |
| 1010 | * |
| 1011 | * The extent ref structure for the full back refs has field for: |
| 1012 | * |
| 1013 | * - number of pointers in the tree leaf |
| 1014 | * |
| 1015 | * The key offset for the implicit back refs is the first byte of |
| 1016 | * the tree leaf |
| 1017 | * |
| 1018 | * When a file extent is allocated, The implicit back refs is used. |
| 1019 | * the fields are filled in: |
| 1020 | * |
| 1021 | * (root_key.objectid, inode objectid, offset in file, 1) |
| 1022 | * |
| 1023 | * When a file extent is removed file truncation, we find the |
| 1024 | * corresponding implicit back refs and check the following fields: |
| 1025 | * |
| 1026 | * (btrfs_header_owner(leaf), inode objectid, offset in file) |
| 1027 | * |
| 1028 | * Btree extents can be referenced by: |
| 1029 | * |
| 1030 | * - Different subvolumes |
| 1031 | * |
| 1032 | * Both the implicit back refs and the full back refs for tree blocks |
| 1033 | * only consist of key. The key offset for the implicit back refs is |
| 1034 | * objectid of block's owner tree. The key offset for the full back refs |
| 1035 | * is the first byte of parent block. |
| 1036 | * |
| 1037 | * When implicit back refs is used, information about the lowest key and |
| 1038 | * level of the tree block are required. These information are stored in |
| 1039 | * tree block info structure. |
| 1040 | */ |
| 1041 | |
| 1042 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 1043 | static int convert_extent_item_v0(struct btrfs_trans_handle *trans, |
| 1044 | struct btrfs_fs_info *fs_info, |
| 1045 | struct btrfs_path *path, |
| 1046 | u64 owner, u32 extra_size) |
| 1047 | { |
| 1048 | struct btrfs_root *root = fs_info->extent_root; |
| 1049 | struct btrfs_extent_item *item; |
| 1050 | struct btrfs_extent_item_v0 *ei0; |
| 1051 | struct btrfs_extent_ref_v0 *ref0; |
| 1052 | struct btrfs_tree_block_info *bi; |
| 1053 | struct extent_buffer *leaf; |
| 1054 | struct btrfs_key key; |
| 1055 | struct btrfs_key found_key; |
| 1056 | u32 new_size = sizeof(*item); |
| 1057 | u64 refs; |
| 1058 | int ret; |
| 1059 | |
| 1060 | leaf = path->nodes[0]; |
| 1061 | BUG_ON(btrfs_item_size_nr(leaf, path->slots[0]) != sizeof(*ei0)); |
| 1062 | |
| 1063 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
| 1064 | ei0 = btrfs_item_ptr(leaf, path->slots[0], |
| 1065 | struct btrfs_extent_item_v0); |
| 1066 | refs = btrfs_extent_refs_v0(leaf, ei0); |
| 1067 | |
| 1068 | if (owner == (u64)-1) { |
| 1069 | while (1) { |
| 1070 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { |
| 1071 | ret = btrfs_next_leaf(root, path); |
| 1072 | if (ret < 0) |
| 1073 | return ret; |
| 1074 | BUG_ON(ret > 0); /* Corruption */ |
| 1075 | leaf = path->nodes[0]; |
| 1076 | } |
| 1077 | btrfs_item_key_to_cpu(leaf, &found_key, |
| 1078 | path->slots[0]); |
| 1079 | BUG_ON(key.objectid != found_key.objectid); |
| 1080 | if (found_key.type != BTRFS_EXTENT_REF_V0_KEY) { |
| 1081 | path->slots[0]++; |
| 1082 | continue; |
| 1083 | } |
| 1084 | ref0 = btrfs_item_ptr(leaf, path->slots[0], |
| 1085 | struct btrfs_extent_ref_v0); |
| 1086 | owner = btrfs_ref_objectid_v0(leaf, ref0); |
| 1087 | break; |
| 1088 | } |
| 1089 | } |
| 1090 | btrfs_release_path(path); |
| 1091 | |
| 1092 | if (owner < BTRFS_FIRST_FREE_OBJECTID) |
| 1093 | new_size += sizeof(*bi); |
| 1094 | |
| 1095 | new_size -= sizeof(*ei0); |
| 1096 | ret = btrfs_search_slot(trans, root, &key, path, |
| 1097 | new_size + extra_size, 1); |
| 1098 | if (ret < 0) |
| 1099 | return ret; |
| 1100 | BUG_ON(ret); /* Corruption */ |
| 1101 | |
| 1102 | btrfs_extend_item(fs_info, path, new_size); |
| 1103 | |
| 1104 | leaf = path->nodes[0]; |
| 1105 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
| 1106 | btrfs_set_extent_refs(leaf, item, refs); |
| 1107 | /* FIXME: get real generation */ |
| 1108 | btrfs_set_extent_generation(leaf, item, 0); |
| 1109 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
| 1110 | btrfs_set_extent_flags(leaf, item, |
| 1111 | BTRFS_EXTENT_FLAG_TREE_BLOCK | |
| 1112 | BTRFS_BLOCK_FLAG_FULL_BACKREF); |
| 1113 | bi = (struct btrfs_tree_block_info *)(item + 1); |
| 1114 | /* FIXME: get first key of the block */ |
| 1115 | memzero_extent_buffer(leaf, (unsigned long)bi, sizeof(*bi)); |
| 1116 | btrfs_set_tree_block_level(leaf, bi, (int)owner); |
| 1117 | } else { |
| 1118 | btrfs_set_extent_flags(leaf, item, BTRFS_EXTENT_FLAG_DATA); |
| 1119 | } |
| 1120 | btrfs_mark_buffer_dirty(leaf); |
| 1121 | return 0; |
| 1122 | } |
| 1123 | #endif |
| 1124 | |
| 1125 | /* |
| 1126 | * is_data == BTRFS_REF_TYPE_BLOCK, tree block type is required, |
| 1127 | * is_data == BTRFS_REF_TYPE_DATA, data type is requried, |
| 1128 | * is_data == BTRFS_REF_TYPE_ANY, either type is OK. |
| 1129 | */ |
| 1130 | int btrfs_get_extent_inline_ref_type(const struct extent_buffer *eb, |
| 1131 | struct btrfs_extent_inline_ref *iref, |
| 1132 | enum btrfs_inline_ref_type is_data) |
| 1133 | { |
| 1134 | int type = btrfs_extent_inline_ref_type(eb, iref); |
| 1135 | u64 offset = btrfs_extent_inline_ref_offset(eb, iref); |
| 1136 | |
| 1137 | if (type == BTRFS_TREE_BLOCK_REF_KEY || |
| 1138 | type == BTRFS_SHARED_BLOCK_REF_KEY || |
| 1139 | type == BTRFS_SHARED_DATA_REF_KEY || |
| 1140 | type == BTRFS_EXTENT_DATA_REF_KEY) { |
| 1141 | if (is_data == BTRFS_REF_TYPE_BLOCK) { |
| 1142 | if (type == BTRFS_TREE_BLOCK_REF_KEY) |
| 1143 | return type; |
| 1144 | if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
| 1145 | ASSERT(eb->fs_info); |
| 1146 | /* |
| 1147 | * Every shared one has parent tree |
| 1148 | * block, which must be aligned to |
| 1149 | * nodesize. |
| 1150 | */ |
| 1151 | if (offset && |
| 1152 | IS_ALIGNED(offset, eb->fs_info->nodesize)) |
| 1153 | return type; |
| 1154 | } |
| 1155 | } else if (is_data == BTRFS_REF_TYPE_DATA) { |
| 1156 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
| 1157 | return type; |
| 1158 | if (type == BTRFS_SHARED_DATA_REF_KEY) { |
| 1159 | ASSERT(eb->fs_info); |
| 1160 | /* |
| 1161 | * Every shared one has parent tree |
| 1162 | * block, which must be aligned to |
| 1163 | * nodesize. |
| 1164 | */ |
| 1165 | if (offset && |
| 1166 | IS_ALIGNED(offset, eb->fs_info->nodesize)) |
| 1167 | return type; |
| 1168 | } |
| 1169 | } else { |
| 1170 | ASSERT(is_data == BTRFS_REF_TYPE_ANY); |
| 1171 | return type; |
| 1172 | } |
| 1173 | } |
| 1174 | |
| 1175 | btrfs_print_leaf((struct extent_buffer *)eb); |
| 1176 | btrfs_err(eb->fs_info, "eb %llu invalid extent inline ref type %d", |
| 1177 | eb->start, type); |
| 1178 | WARN_ON(1); |
| 1179 | |
| 1180 | return BTRFS_REF_TYPE_INVALID; |
| 1181 | } |
| 1182 | |
| 1183 | static u64 hash_extent_data_ref(u64 root_objectid, u64 owner, u64 offset) |
| 1184 | { |
| 1185 | u32 high_crc = ~(u32)0; |
| 1186 | u32 low_crc = ~(u32)0; |
| 1187 | __le64 lenum; |
| 1188 | |
| 1189 | lenum = cpu_to_le64(root_objectid); |
| 1190 | high_crc = crc32c(high_crc, &lenum, sizeof(lenum)); |
| 1191 | lenum = cpu_to_le64(owner); |
| 1192 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
| 1193 | lenum = cpu_to_le64(offset); |
| 1194 | low_crc = crc32c(low_crc, &lenum, sizeof(lenum)); |
| 1195 | |
| 1196 | return ((u64)high_crc << 31) ^ (u64)low_crc; |
| 1197 | } |
| 1198 | |
| 1199 | static u64 hash_extent_data_ref_item(struct extent_buffer *leaf, |
| 1200 | struct btrfs_extent_data_ref *ref) |
| 1201 | { |
| 1202 | return hash_extent_data_ref(btrfs_extent_data_ref_root(leaf, ref), |
| 1203 | btrfs_extent_data_ref_objectid(leaf, ref), |
| 1204 | btrfs_extent_data_ref_offset(leaf, ref)); |
| 1205 | } |
| 1206 | |
| 1207 | static int match_extent_data_ref(struct extent_buffer *leaf, |
| 1208 | struct btrfs_extent_data_ref *ref, |
| 1209 | u64 root_objectid, u64 owner, u64 offset) |
| 1210 | { |
| 1211 | if (btrfs_extent_data_ref_root(leaf, ref) != root_objectid || |
| 1212 | btrfs_extent_data_ref_objectid(leaf, ref) != owner || |
| 1213 | btrfs_extent_data_ref_offset(leaf, ref) != offset) |
| 1214 | return 0; |
| 1215 | return 1; |
| 1216 | } |
| 1217 | |
| 1218 | static noinline int lookup_extent_data_ref(struct btrfs_trans_handle *trans, |
| 1219 | struct btrfs_path *path, |
| 1220 | u64 bytenr, u64 parent, |
| 1221 | u64 root_objectid, |
| 1222 | u64 owner, u64 offset) |
| 1223 | { |
| 1224 | struct btrfs_root *root = trans->fs_info->extent_root; |
| 1225 | struct btrfs_key key; |
| 1226 | struct btrfs_extent_data_ref *ref; |
| 1227 | struct extent_buffer *leaf; |
| 1228 | u32 nritems; |
| 1229 | int ret; |
| 1230 | int recow; |
| 1231 | int err = -ENOENT; |
| 1232 | |
| 1233 | key.objectid = bytenr; |
| 1234 | if (parent) { |
| 1235 | key.type = BTRFS_SHARED_DATA_REF_KEY; |
| 1236 | key.offset = parent; |
| 1237 | } else { |
| 1238 | key.type = BTRFS_EXTENT_DATA_REF_KEY; |
| 1239 | key.offset = hash_extent_data_ref(root_objectid, |
| 1240 | owner, offset); |
| 1241 | } |
| 1242 | again: |
| 1243 | recow = 0; |
| 1244 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
| 1245 | if (ret < 0) { |
| 1246 | err = ret; |
| 1247 | goto fail; |
| 1248 | } |
| 1249 | |
| 1250 | if (parent) { |
| 1251 | if (!ret) |
| 1252 | return 0; |
| 1253 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 1254 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
| 1255 | btrfs_release_path(path); |
| 1256 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
| 1257 | if (ret < 0) { |
| 1258 | err = ret; |
| 1259 | goto fail; |
| 1260 | } |
| 1261 | if (!ret) |
| 1262 | return 0; |
| 1263 | #endif |
| 1264 | goto fail; |
| 1265 | } |
| 1266 | |
| 1267 | leaf = path->nodes[0]; |
| 1268 | nritems = btrfs_header_nritems(leaf); |
| 1269 | while (1) { |
| 1270 | if (path->slots[0] >= nritems) { |
| 1271 | ret = btrfs_next_leaf(root, path); |
| 1272 | if (ret < 0) |
| 1273 | err = ret; |
| 1274 | if (ret) |
| 1275 | goto fail; |
| 1276 | |
| 1277 | leaf = path->nodes[0]; |
| 1278 | nritems = btrfs_header_nritems(leaf); |
| 1279 | recow = 1; |
| 1280 | } |
| 1281 | |
| 1282 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
| 1283 | if (key.objectid != bytenr || |
| 1284 | key.type != BTRFS_EXTENT_DATA_REF_KEY) |
| 1285 | goto fail; |
| 1286 | |
| 1287 | ref = btrfs_item_ptr(leaf, path->slots[0], |
| 1288 | struct btrfs_extent_data_ref); |
| 1289 | |
| 1290 | if (match_extent_data_ref(leaf, ref, root_objectid, |
| 1291 | owner, offset)) { |
| 1292 | if (recow) { |
| 1293 | btrfs_release_path(path); |
| 1294 | goto again; |
| 1295 | } |
| 1296 | err = 0; |
| 1297 | break; |
| 1298 | } |
| 1299 | path->slots[0]++; |
| 1300 | } |
| 1301 | fail: |
| 1302 | return err; |
| 1303 | } |
| 1304 | |
| 1305 | static noinline int insert_extent_data_ref(struct btrfs_trans_handle *trans, |
| 1306 | struct btrfs_path *path, |
| 1307 | u64 bytenr, u64 parent, |
| 1308 | u64 root_objectid, u64 owner, |
| 1309 | u64 offset, int refs_to_add) |
| 1310 | { |
| 1311 | struct btrfs_root *root = trans->fs_info->extent_root; |
| 1312 | struct btrfs_key key; |
| 1313 | struct extent_buffer *leaf; |
| 1314 | u32 size; |
| 1315 | u32 num_refs; |
| 1316 | int ret; |
| 1317 | |
| 1318 | key.objectid = bytenr; |
| 1319 | if (parent) { |
| 1320 | key.type = BTRFS_SHARED_DATA_REF_KEY; |
| 1321 | key.offset = parent; |
| 1322 | size = sizeof(struct btrfs_shared_data_ref); |
| 1323 | } else { |
| 1324 | key.type = BTRFS_EXTENT_DATA_REF_KEY; |
| 1325 | key.offset = hash_extent_data_ref(root_objectid, |
| 1326 | owner, offset); |
| 1327 | size = sizeof(struct btrfs_extent_data_ref); |
| 1328 | } |
| 1329 | |
| 1330 | ret = btrfs_insert_empty_item(trans, root, path, &key, size); |
| 1331 | if (ret && ret != -EEXIST) |
| 1332 | goto fail; |
| 1333 | |
| 1334 | leaf = path->nodes[0]; |
| 1335 | if (parent) { |
| 1336 | struct btrfs_shared_data_ref *ref; |
| 1337 | ref = btrfs_item_ptr(leaf, path->slots[0], |
| 1338 | struct btrfs_shared_data_ref); |
| 1339 | if (ret == 0) { |
| 1340 | btrfs_set_shared_data_ref_count(leaf, ref, refs_to_add); |
| 1341 | } else { |
| 1342 | num_refs = btrfs_shared_data_ref_count(leaf, ref); |
| 1343 | num_refs += refs_to_add; |
| 1344 | btrfs_set_shared_data_ref_count(leaf, ref, num_refs); |
| 1345 | } |
| 1346 | } else { |
| 1347 | struct btrfs_extent_data_ref *ref; |
| 1348 | while (ret == -EEXIST) { |
| 1349 | ref = btrfs_item_ptr(leaf, path->slots[0], |
| 1350 | struct btrfs_extent_data_ref); |
| 1351 | if (match_extent_data_ref(leaf, ref, root_objectid, |
| 1352 | owner, offset)) |
| 1353 | break; |
| 1354 | btrfs_release_path(path); |
| 1355 | key.offset++; |
| 1356 | ret = btrfs_insert_empty_item(trans, root, path, &key, |
| 1357 | size); |
| 1358 | if (ret && ret != -EEXIST) |
| 1359 | goto fail; |
| 1360 | |
| 1361 | leaf = path->nodes[0]; |
| 1362 | } |
| 1363 | ref = btrfs_item_ptr(leaf, path->slots[0], |
| 1364 | struct btrfs_extent_data_ref); |
| 1365 | if (ret == 0) { |
| 1366 | btrfs_set_extent_data_ref_root(leaf, ref, |
| 1367 | root_objectid); |
| 1368 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); |
| 1369 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); |
| 1370 | btrfs_set_extent_data_ref_count(leaf, ref, refs_to_add); |
| 1371 | } else { |
| 1372 | num_refs = btrfs_extent_data_ref_count(leaf, ref); |
| 1373 | num_refs += refs_to_add; |
| 1374 | btrfs_set_extent_data_ref_count(leaf, ref, num_refs); |
| 1375 | } |
| 1376 | } |
| 1377 | btrfs_mark_buffer_dirty(leaf); |
| 1378 | ret = 0; |
| 1379 | fail: |
| 1380 | btrfs_release_path(path); |
| 1381 | return ret; |
| 1382 | } |
| 1383 | |
| 1384 | static noinline int remove_extent_data_ref(struct btrfs_trans_handle *trans, |
| 1385 | struct btrfs_path *path, |
| 1386 | int refs_to_drop, int *last_ref) |
| 1387 | { |
| 1388 | struct btrfs_key key; |
| 1389 | struct btrfs_extent_data_ref *ref1 = NULL; |
| 1390 | struct btrfs_shared_data_ref *ref2 = NULL; |
| 1391 | struct extent_buffer *leaf; |
| 1392 | u32 num_refs = 0; |
| 1393 | int ret = 0; |
| 1394 | |
| 1395 | leaf = path->nodes[0]; |
| 1396 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
| 1397 | |
| 1398 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { |
| 1399 | ref1 = btrfs_item_ptr(leaf, path->slots[0], |
| 1400 | struct btrfs_extent_data_ref); |
| 1401 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); |
| 1402 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { |
| 1403 | ref2 = btrfs_item_ptr(leaf, path->slots[0], |
| 1404 | struct btrfs_shared_data_ref); |
| 1405 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); |
| 1406 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 1407 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { |
| 1408 | struct btrfs_extent_ref_v0 *ref0; |
| 1409 | ref0 = btrfs_item_ptr(leaf, path->slots[0], |
| 1410 | struct btrfs_extent_ref_v0); |
| 1411 | num_refs = btrfs_ref_count_v0(leaf, ref0); |
| 1412 | #endif |
| 1413 | } else { |
| 1414 | BUG(); |
| 1415 | } |
| 1416 | |
| 1417 | BUG_ON(num_refs < refs_to_drop); |
| 1418 | num_refs -= refs_to_drop; |
| 1419 | |
| 1420 | if (num_refs == 0) { |
| 1421 | ret = btrfs_del_item(trans, trans->fs_info->extent_root, path); |
| 1422 | *last_ref = 1; |
| 1423 | } else { |
| 1424 | if (key.type == BTRFS_EXTENT_DATA_REF_KEY) |
| 1425 | btrfs_set_extent_data_ref_count(leaf, ref1, num_refs); |
| 1426 | else if (key.type == BTRFS_SHARED_DATA_REF_KEY) |
| 1427 | btrfs_set_shared_data_ref_count(leaf, ref2, num_refs); |
| 1428 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 1429 | else { |
| 1430 | struct btrfs_extent_ref_v0 *ref0; |
| 1431 | ref0 = btrfs_item_ptr(leaf, path->slots[0], |
| 1432 | struct btrfs_extent_ref_v0); |
| 1433 | btrfs_set_ref_count_v0(leaf, ref0, num_refs); |
| 1434 | } |
| 1435 | #endif |
| 1436 | btrfs_mark_buffer_dirty(leaf); |
| 1437 | } |
| 1438 | return ret; |
| 1439 | } |
| 1440 | |
| 1441 | static noinline u32 extent_data_ref_count(struct btrfs_path *path, |
| 1442 | struct btrfs_extent_inline_ref *iref) |
| 1443 | { |
| 1444 | struct btrfs_key key; |
| 1445 | struct extent_buffer *leaf; |
| 1446 | struct btrfs_extent_data_ref *ref1; |
| 1447 | struct btrfs_shared_data_ref *ref2; |
| 1448 | u32 num_refs = 0; |
| 1449 | int type; |
| 1450 | |
| 1451 | leaf = path->nodes[0]; |
| 1452 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
| 1453 | if (iref) { |
| 1454 | /* |
| 1455 | * If type is invalid, we should have bailed out earlier than |
| 1456 | * this call. |
| 1457 | */ |
| 1458 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); |
| 1459 | ASSERT(type != BTRFS_REF_TYPE_INVALID); |
| 1460 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { |
| 1461 | ref1 = (struct btrfs_extent_data_ref *)(&iref->offset); |
| 1462 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); |
| 1463 | } else { |
| 1464 | ref2 = (struct btrfs_shared_data_ref *)(iref + 1); |
| 1465 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); |
| 1466 | } |
| 1467 | } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { |
| 1468 | ref1 = btrfs_item_ptr(leaf, path->slots[0], |
| 1469 | struct btrfs_extent_data_ref); |
| 1470 | num_refs = btrfs_extent_data_ref_count(leaf, ref1); |
| 1471 | } else if (key.type == BTRFS_SHARED_DATA_REF_KEY) { |
| 1472 | ref2 = btrfs_item_ptr(leaf, path->slots[0], |
| 1473 | struct btrfs_shared_data_ref); |
| 1474 | num_refs = btrfs_shared_data_ref_count(leaf, ref2); |
| 1475 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 1476 | } else if (key.type == BTRFS_EXTENT_REF_V0_KEY) { |
| 1477 | struct btrfs_extent_ref_v0 *ref0; |
| 1478 | ref0 = btrfs_item_ptr(leaf, path->slots[0], |
| 1479 | struct btrfs_extent_ref_v0); |
| 1480 | num_refs = btrfs_ref_count_v0(leaf, ref0); |
| 1481 | #endif |
| 1482 | } else { |
| 1483 | WARN_ON(1); |
| 1484 | } |
| 1485 | return num_refs; |
| 1486 | } |
| 1487 | |
| 1488 | static noinline int lookup_tree_block_ref(struct btrfs_trans_handle *trans, |
| 1489 | struct btrfs_path *path, |
| 1490 | u64 bytenr, u64 parent, |
| 1491 | u64 root_objectid) |
| 1492 | { |
| 1493 | struct btrfs_root *root = trans->fs_info->extent_root; |
| 1494 | struct btrfs_key key; |
| 1495 | int ret; |
| 1496 | |
| 1497 | key.objectid = bytenr; |
| 1498 | if (parent) { |
| 1499 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; |
| 1500 | key.offset = parent; |
| 1501 | } else { |
| 1502 | key.type = BTRFS_TREE_BLOCK_REF_KEY; |
| 1503 | key.offset = root_objectid; |
| 1504 | } |
| 1505 | |
| 1506 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
| 1507 | if (ret > 0) |
| 1508 | ret = -ENOENT; |
| 1509 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 1510 | if (ret == -ENOENT && parent) { |
| 1511 | btrfs_release_path(path); |
| 1512 | key.type = BTRFS_EXTENT_REF_V0_KEY; |
| 1513 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
| 1514 | if (ret > 0) |
| 1515 | ret = -ENOENT; |
| 1516 | } |
| 1517 | #endif |
| 1518 | return ret; |
| 1519 | } |
| 1520 | |
| 1521 | static noinline int insert_tree_block_ref(struct btrfs_trans_handle *trans, |
| 1522 | struct btrfs_path *path, |
| 1523 | u64 bytenr, u64 parent, |
| 1524 | u64 root_objectid) |
| 1525 | { |
| 1526 | struct btrfs_key key; |
| 1527 | int ret; |
| 1528 | |
| 1529 | key.objectid = bytenr; |
| 1530 | if (parent) { |
| 1531 | key.type = BTRFS_SHARED_BLOCK_REF_KEY; |
| 1532 | key.offset = parent; |
| 1533 | } else { |
| 1534 | key.type = BTRFS_TREE_BLOCK_REF_KEY; |
| 1535 | key.offset = root_objectid; |
| 1536 | } |
| 1537 | |
| 1538 | ret = btrfs_insert_empty_item(trans, trans->fs_info->extent_root, |
| 1539 | path, &key, 0); |
| 1540 | btrfs_release_path(path); |
| 1541 | return ret; |
| 1542 | } |
| 1543 | |
| 1544 | static inline int extent_ref_type(u64 parent, u64 owner) |
| 1545 | { |
| 1546 | int type; |
| 1547 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
| 1548 | if (parent > 0) |
| 1549 | type = BTRFS_SHARED_BLOCK_REF_KEY; |
| 1550 | else |
| 1551 | type = BTRFS_TREE_BLOCK_REF_KEY; |
| 1552 | } else { |
| 1553 | if (parent > 0) |
| 1554 | type = BTRFS_SHARED_DATA_REF_KEY; |
| 1555 | else |
| 1556 | type = BTRFS_EXTENT_DATA_REF_KEY; |
| 1557 | } |
| 1558 | return type; |
| 1559 | } |
| 1560 | |
| 1561 | static int find_next_key(struct btrfs_path *path, int level, |
| 1562 | struct btrfs_key *key) |
| 1563 | |
| 1564 | { |
| 1565 | for (; level < BTRFS_MAX_LEVEL; level++) { |
| 1566 | if (!path->nodes[level]) |
| 1567 | break; |
| 1568 | if (path->slots[level] + 1 >= |
| 1569 | btrfs_header_nritems(path->nodes[level])) |
| 1570 | continue; |
| 1571 | if (level == 0) |
| 1572 | btrfs_item_key_to_cpu(path->nodes[level], key, |
| 1573 | path->slots[level] + 1); |
| 1574 | else |
| 1575 | btrfs_node_key_to_cpu(path->nodes[level], key, |
| 1576 | path->slots[level] + 1); |
| 1577 | return 0; |
| 1578 | } |
| 1579 | return 1; |
| 1580 | } |
| 1581 | |
| 1582 | /* |
| 1583 | * look for inline back ref. if back ref is found, *ref_ret is set |
| 1584 | * to the address of inline back ref, and 0 is returned. |
| 1585 | * |
| 1586 | * if back ref isn't found, *ref_ret is set to the address where it |
| 1587 | * should be inserted, and -ENOENT is returned. |
| 1588 | * |
| 1589 | * if insert is true and there are too many inline back refs, the path |
| 1590 | * points to the extent item, and -EAGAIN is returned. |
| 1591 | * |
| 1592 | * NOTE: inline back refs are ordered in the same way that back ref |
| 1593 | * items in the tree are ordered. |
| 1594 | */ |
| 1595 | static noinline_for_stack |
| 1596 | int lookup_inline_extent_backref(struct btrfs_trans_handle *trans, |
| 1597 | struct btrfs_path *path, |
| 1598 | struct btrfs_extent_inline_ref **ref_ret, |
| 1599 | u64 bytenr, u64 num_bytes, |
| 1600 | u64 parent, u64 root_objectid, |
| 1601 | u64 owner, u64 offset, int insert) |
| 1602 | { |
| 1603 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 1604 | struct btrfs_root *root = fs_info->extent_root; |
| 1605 | struct btrfs_key key; |
| 1606 | struct extent_buffer *leaf; |
| 1607 | struct btrfs_extent_item *ei; |
| 1608 | struct btrfs_extent_inline_ref *iref; |
| 1609 | u64 flags; |
| 1610 | u64 item_size; |
| 1611 | unsigned long ptr; |
| 1612 | unsigned long end; |
| 1613 | int extra_size; |
| 1614 | int type; |
| 1615 | int want; |
| 1616 | int ret; |
| 1617 | int err = 0; |
| 1618 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
| 1619 | int needed; |
| 1620 | |
| 1621 | key.objectid = bytenr; |
| 1622 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 1623 | key.offset = num_bytes; |
| 1624 | |
| 1625 | want = extent_ref_type(parent, owner); |
| 1626 | if (insert) { |
| 1627 | extra_size = btrfs_extent_inline_ref_size(want); |
| 1628 | path->keep_locks = 1; |
| 1629 | } else |
| 1630 | extra_size = -1; |
| 1631 | |
| 1632 | /* |
| 1633 | * Owner is our level, so we can just add one to get the level for the |
| 1634 | * block we are interested in. |
| 1635 | */ |
| 1636 | if (skinny_metadata && owner < BTRFS_FIRST_FREE_OBJECTID) { |
| 1637 | key.type = BTRFS_METADATA_ITEM_KEY; |
| 1638 | key.offset = owner; |
| 1639 | } |
| 1640 | |
| 1641 | again: |
| 1642 | ret = btrfs_search_slot(trans, root, &key, path, extra_size, 1); |
| 1643 | if (ret < 0) { |
| 1644 | err = ret; |
| 1645 | goto out; |
| 1646 | } |
| 1647 | |
| 1648 | /* |
| 1649 | * We may be a newly converted file system which still has the old fat |
| 1650 | * extent entries for metadata, so try and see if we have one of those. |
| 1651 | */ |
| 1652 | if (ret > 0 && skinny_metadata) { |
| 1653 | skinny_metadata = false; |
| 1654 | if (path->slots[0]) { |
| 1655 | path->slots[0]--; |
| 1656 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
| 1657 | path->slots[0]); |
| 1658 | if (key.objectid == bytenr && |
| 1659 | key.type == BTRFS_EXTENT_ITEM_KEY && |
| 1660 | key.offset == num_bytes) |
| 1661 | ret = 0; |
| 1662 | } |
| 1663 | if (ret) { |
| 1664 | key.objectid = bytenr; |
| 1665 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 1666 | key.offset = num_bytes; |
| 1667 | btrfs_release_path(path); |
| 1668 | goto again; |
| 1669 | } |
| 1670 | } |
| 1671 | |
| 1672 | if (ret && !insert) { |
| 1673 | err = -ENOENT; |
| 1674 | goto out; |
| 1675 | } else if (WARN_ON(ret)) { |
| 1676 | err = -EIO; |
| 1677 | goto out; |
| 1678 | } |
| 1679 | |
| 1680 | leaf = path->nodes[0]; |
| 1681 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); |
| 1682 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 1683 | if (item_size < sizeof(*ei)) { |
| 1684 | if (!insert) { |
| 1685 | err = -ENOENT; |
| 1686 | goto out; |
| 1687 | } |
| 1688 | ret = convert_extent_item_v0(trans, fs_info, path, owner, |
| 1689 | extra_size); |
| 1690 | if (ret < 0) { |
| 1691 | err = ret; |
| 1692 | goto out; |
| 1693 | } |
| 1694 | leaf = path->nodes[0]; |
| 1695 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); |
| 1696 | } |
| 1697 | #endif |
| 1698 | BUG_ON(item_size < sizeof(*ei)); |
| 1699 | |
| 1700 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
| 1701 | flags = btrfs_extent_flags(leaf, ei); |
| 1702 | |
| 1703 | ptr = (unsigned long)(ei + 1); |
| 1704 | end = (unsigned long)ei + item_size; |
| 1705 | |
| 1706 | if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK && !skinny_metadata) { |
| 1707 | ptr += sizeof(struct btrfs_tree_block_info); |
| 1708 | BUG_ON(ptr > end); |
| 1709 | } |
| 1710 | |
| 1711 | if (owner >= BTRFS_FIRST_FREE_OBJECTID) |
| 1712 | needed = BTRFS_REF_TYPE_DATA; |
| 1713 | else |
| 1714 | needed = BTRFS_REF_TYPE_BLOCK; |
| 1715 | |
| 1716 | err = -ENOENT; |
| 1717 | while (1) { |
| 1718 | if (ptr >= end) { |
| 1719 | WARN_ON(ptr > end); |
| 1720 | break; |
| 1721 | } |
| 1722 | iref = (struct btrfs_extent_inline_ref *)ptr; |
| 1723 | type = btrfs_get_extent_inline_ref_type(leaf, iref, needed); |
| 1724 | if (type == BTRFS_REF_TYPE_INVALID) { |
| 1725 | err = -EINVAL; |
| 1726 | goto out; |
| 1727 | } |
| 1728 | |
| 1729 | if (want < type) |
| 1730 | break; |
| 1731 | if (want > type) { |
| 1732 | ptr += btrfs_extent_inline_ref_size(type); |
| 1733 | continue; |
| 1734 | } |
| 1735 | |
| 1736 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { |
| 1737 | struct btrfs_extent_data_ref *dref; |
| 1738 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); |
| 1739 | if (match_extent_data_ref(leaf, dref, root_objectid, |
| 1740 | owner, offset)) { |
| 1741 | err = 0; |
| 1742 | break; |
| 1743 | } |
| 1744 | if (hash_extent_data_ref_item(leaf, dref) < |
| 1745 | hash_extent_data_ref(root_objectid, owner, offset)) |
| 1746 | break; |
| 1747 | } else { |
| 1748 | u64 ref_offset; |
| 1749 | ref_offset = btrfs_extent_inline_ref_offset(leaf, iref); |
| 1750 | if (parent > 0) { |
| 1751 | if (parent == ref_offset) { |
| 1752 | err = 0; |
| 1753 | break; |
| 1754 | } |
| 1755 | if (ref_offset < parent) |
| 1756 | break; |
| 1757 | } else { |
| 1758 | if (root_objectid == ref_offset) { |
| 1759 | err = 0; |
| 1760 | break; |
| 1761 | } |
| 1762 | if (ref_offset < root_objectid) |
| 1763 | break; |
| 1764 | } |
| 1765 | } |
| 1766 | ptr += btrfs_extent_inline_ref_size(type); |
| 1767 | } |
| 1768 | if (err == -ENOENT && insert) { |
| 1769 | if (item_size + extra_size >= |
| 1770 | BTRFS_MAX_EXTENT_ITEM_SIZE(root)) { |
| 1771 | err = -EAGAIN; |
| 1772 | goto out; |
| 1773 | } |
| 1774 | /* |
| 1775 | * To add new inline back ref, we have to make sure |
| 1776 | * there is no corresponding back ref item. |
| 1777 | * For simplicity, we just do not add new inline back |
| 1778 | * ref if there is any kind of item for this block |
| 1779 | */ |
| 1780 | if (find_next_key(path, 0, &key) == 0 && |
| 1781 | key.objectid == bytenr && |
| 1782 | key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) { |
| 1783 | err = -EAGAIN; |
| 1784 | goto out; |
| 1785 | } |
| 1786 | } |
| 1787 | *ref_ret = (struct btrfs_extent_inline_ref *)ptr; |
| 1788 | out: |
| 1789 | if (insert) { |
| 1790 | path->keep_locks = 0; |
| 1791 | btrfs_unlock_up_safe(path, 1); |
| 1792 | } |
| 1793 | return err; |
| 1794 | } |
| 1795 | |
| 1796 | /* |
| 1797 | * helper to add new inline back ref |
| 1798 | */ |
| 1799 | static noinline_for_stack |
| 1800 | void setup_inline_extent_backref(struct btrfs_fs_info *fs_info, |
| 1801 | struct btrfs_path *path, |
| 1802 | struct btrfs_extent_inline_ref *iref, |
| 1803 | u64 parent, u64 root_objectid, |
| 1804 | u64 owner, u64 offset, int refs_to_add, |
| 1805 | struct btrfs_delayed_extent_op *extent_op) |
| 1806 | { |
| 1807 | struct extent_buffer *leaf; |
| 1808 | struct btrfs_extent_item *ei; |
| 1809 | unsigned long ptr; |
| 1810 | unsigned long end; |
| 1811 | unsigned long item_offset; |
| 1812 | u64 refs; |
| 1813 | int size; |
| 1814 | int type; |
| 1815 | |
| 1816 | leaf = path->nodes[0]; |
| 1817 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
| 1818 | item_offset = (unsigned long)iref - (unsigned long)ei; |
| 1819 | |
| 1820 | type = extent_ref_type(parent, owner); |
| 1821 | size = btrfs_extent_inline_ref_size(type); |
| 1822 | |
| 1823 | btrfs_extend_item(fs_info, path, size); |
| 1824 | |
| 1825 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
| 1826 | refs = btrfs_extent_refs(leaf, ei); |
| 1827 | refs += refs_to_add; |
| 1828 | btrfs_set_extent_refs(leaf, ei, refs); |
| 1829 | if (extent_op) |
| 1830 | __run_delayed_extent_op(extent_op, leaf, ei); |
| 1831 | |
| 1832 | ptr = (unsigned long)ei + item_offset; |
| 1833 | end = (unsigned long)ei + btrfs_item_size_nr(leaf, path->slots[0]); |
| 1834 | if (ptr < end - size) |
| 1835 | memmove_extent_buffer(leaf, ptr + size, ptr, |
| 1836 | end - size - ptr); |
| 1837 | |
| 1838 | iref = (struct btrfs_extent_inline_ref *)ptr; |
| 1839 | btrfs_set_extent_inline_ref_type(leaf, iref, type); |
| 1840 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { |
| 1841 | struct btrfs_extent_data_ref *dref; |
| 1842 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); |
| 1843 | btrfs_set_extent_data_ref_root(leaf, dref, root_objectid); |
| 1844 | btrfs_set_extent_data_ref_objectid(leaf, dref, owner); |
| 1845 | btrfs_set_extent_data_ref_offset(leaf, dref, offset); |
| 1846 | btrfs_set_extent_data_ref_count(leaf, dref, refs_to_add); |
| 1847 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { |
| 1848 | struct btrfs_shared_data_ref *sref; |
| 1849 | sref = (struct btrfs_shared_data_ref *)(iref + 1); |
| 1850 | btrfs_set_shared_data_ref_count(leaf, sref, refs_to_add); |
| 1851 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); |
| 1852 | } else if (type == BTRFS_SHARED_BLOCK_REF_KEY) { |
| 1853 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); |
| 1854 | } else { |
| 1855 | btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid); |
| 1856 | } |
| 1857 | btrfs_mark_buffer_dirty(leaf); |
| 1858 | } |
| 1859 | |
| 1860 | static int lookup_extent_backref(struct btrfs_trans_handle *trans, |
| 1861 | struct btrfs_fs_info *fs_info, |
| 1862 | struct btrfs_path *path, |
| 1863 | struct btrfs_extent_inline_ref **ref_ret, |
| 1864 | u64 bytenr, u64 num_bytes, u64 parent, |
| 1865 | u64 root_objectid, u64 owner, u64 offset) |
| 1866 | { |
| 1867 | int ret; |
| 1868 | |
| 1869 | ret = lookup_inline_extent_backref(trans, path, ref_ret, bytenr, |
| 1870 | num_bytes, parent, root_objectid, |
| 1871 | owner, offset, 0); |
| 1872 | if (ret != -ENOENT) |
| 1873 | return ret; |
| 1874 | |
| 1875 | btrfs_release_path(path); |
| 1876 | *ref_ret = NULL; |
| 1877 | |
| 1878 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
| 1879 | ret = lookup_tree_block_ref(trans, path, bytenr, parent, |
| 1880 | root_objectid); |
| 1881 | } else { |
| 1882 | ret = lookup_extent_data_ref(trans, path, bytenr, parent, |
| 1883 | root_objectid, owner, offset); |
| 1884 | } |
| 1885 | return ret; |
| 1886 | } |
| 1887 | |
| 1888 | /* |
| 1889 | * helper to update/remove inline back ref |
| 1890 | */ |
| 1891 | static noinline_for_stack |
| 1892 | void update_inline_extent_backref(struct btrfs_path *path, |
| 1893 | struct btrfs_extent_inline_ref *iref, |
| 1894 | int refs_to_mod, |
| 1895 | struct btrfs_delayed_extent_op *extent_op, |
| 1896 | int *last_ref) |
| 1897 | { |
| 1898 | struct extent_buffer *leaf = path->nodes[0]; |
| 1899 | struct btrfs_fs_info *fs_info = leaf->fs_info; |
| 1900 | struct btrfs_extent_item *ei; |
| 1901 | struct btrfs_extent_data_ref *dref = NULL; |
| 1902 | struct btrfs_shared_data_ref *sref = NULL; |
| 1903 | unsigned long ptr; |
| 1904 | unsigned long end; |
| 1905 | u32 item_size; |
| 1906 | int size; |
| 1907 | int type; |
| 1908 | u64 refs; |
| 1909 | |
| 1910 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
| 1911 | refs = btrfs_extent_refs(leaf, ei); |
| 1912 | WARN_ON(refs_to_mod < 0 && refs + refs_to_mod <= 0); |
| 1913 | refs += refs_to_mod; |
| 1914 | btrfs_set_extent_refs(leaf, ei, refs); |
| 1915 | if (extent_op) |
| 1916 | __run_delayed_extent_op(extent_op, leaf, ei); |
| 1917 | |
| 1918 | /* |
| 1919 | * If type is invalid, we should have bailed out after |
| 1920 | * lookup_inline_extent_backref(). |
| 1921 | */ |
| 1922 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_ANY); |
| 1923 | ASSERT(type != BTRFS_REF_TYPE_INVALID); |
| 1924 | |
| 1925 | if (type == BTRFS_EXTENT_DATA_REF_KEY) { |
| 1926 | dref = (struct btrfs_extent_data_ref *)(&iref->offset); |
| 1927 | refs = btrfs_extent_data_ref_count(leaf, dref); |
| 1928 | } else if (type == BTRFS_SHARED_DATA_REF_KEY) { |
| 1929 | sref = (struct btrfs_shared_data_ref *)(iref + 1); |
| 1930 | refs = btrfs_shared_data_ref_count(leaf, sref); |
| 1931 | } else { |
| 1932 | refs = 1; |
| 1933 | BUG_ON(refs_to_mod != -1); |
| 1934 | } |
| 1935 | |
| 1936 | BUG_ON(refs_to_mod < 0 && refs < -refs_to_mod); |
| 1937 | refs += refs_to_mod; |
| 1938 | |
| 1939 | if (refs > 0) { |
| 1940 | if (type == BTRFS_EXTENT_DATA_REF_KEY) |
| 1941 | btrfs_set_extent_data_ref_count(leaf, dref, refs); |
| 1942 | else |
| 1943 | btrfs_set_shared_data_ref_count(leaf, sref, refs); |
| 1944 | } else { |
| 1945 | *last_ref = 1; |
| 1946 | size = btrfs_extent_inline_ref_size(type); |
| 1947 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); |
| 1948 | ptr = (unsigned long)iref; |
| 1949 | end = (unsigned long)ei + item_size; |
| 1950 | if (ptr + size < end) |
| 1951 | memmove_extent_buffer(leaf, ptr, ptr + size, |
| 1952 | end - ptr - size); |
| 1953 | item_size -= size; |
| 1954 | btrfs_truncate_item(fs_info, path, item_size, 1); |
| 1955 | } |
| 1956 | btrfs_mark_buffer_dirty(leaf); |
| 1957 | } |
| 1958 | |
| 1959 | static noinline_for_stack |
| 1960 | int insert_inline_extent_backref(struct btrfs_trans_handle *trans, |
| 1961 | struct btrfs_fs_info *fs_info, |
| 1962 | struct btrfs_path *path, |
| 1963 | u64 bytenr, u64 num_bytes, u64 parent, |
| 1964 | u64 root_objectid, u64 owner, |
| 1965 | u64 offset, int refs_to_add, |
| 1966 | struct btrfs_delayed_extent_op *extent_op) |
| 1967 | { |
| 1968 | struct btrfs_extent_inline_ref *iref; |
| 1969 | int ret; |
| 1970 | |
| 1971 | ret = lookup_inline_extent_backref(trans, path, &iref, bytenr, |
| 1972 | num_bytes, parent, root_objectid, |
| 1973 | owner, offset, 1); |
| 1974 | if (ret == 0) { |
| 1975 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID); |
| 1976 | update_inline_extent_backref(path, iref, refs_to_add, |
| 1977 | extent_op, NULL); |
| 1978 | } else if (ret == -ENOENT) { |
| 1979 | setup_inline_extent_backref(fs_info, path, iref, parent, |
| 1980 | root_objectid, owner, offset, |
| 1981 | refs_to_add, extent_op); |
| 1982 | ret = 0; |
| 1983 | } |
| 1984 | return ret; |
| 1985 | } |
| 1986 | |
| 1987 | static int insert_extent_backref(struct btrfs_trans_handle *trans, |
| 1988 | struct btrfs_path *path, |
| 1989 | u64 bytenr, u64 parent, u64 root_objectid, |
| 1990 | u64 owner, u64 offset, int refs_to_add) |
| 1991 | { |
| 1992 | int ret; |
| 1993 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
| 1994 | BUG_ON(refs_to_add != 1); |
| 1995 | ret = insert_tree_block_ref(trans, path, bytenr, parent, |
| 1996 | root_objectid); |
| 1997 | } else { |
| 1998 | ret = insert_extent_data_ref(trans, path, bytenr, parent, |
| 1999 | root_objectid, owner, offset, |
| 2000 | refs_to_add); |
| 2001 | } |
| 2002 | return ret; |
| 2003 | } |
| 2004 | |
| 2005 | static int remove_extent_backref(struct btrfs_trans_handle *trans, |
| 2006 | struct btrfs_fs_info *fs_info, |
| 2007 | struct btrfs_path *path, |
| 2008 | struct btrfs_extent_inline_ref *iref, |
| 2009 | int refs_to_drop, int is_data, int *last_ref) |
| 2010 | { |
| 2011 | int ret = 0; |
| 2012 | |
| 2013 | BUG_ON(!is_data && refs_to_drop != 1); |
| 2014 | if (iref) { |
| 2015 | update_inline_extent_backref(path, iref, -refs_to_drop, NULL, |
| 2016 | last_ref); |
| 2017 | } else if (is_data) { |
| 2018 | ret = remove_extent_data_ref(trans, path, refs_to_drop, |
| 2019 | last_ref); |
| 2020 | } else { |
| 2021 | *last_ref = 1; |
| 2022 | ret = btrfs_del_item(trans, fs_info->extent_root, path); |
| 2023 | } |
| 2024 | return ret; |
| 2025 | } |
| 2026 | |
| 2027 | #define in_range(b, first, len) ((b) >= (first) && (b) < (first) + (len)) |
| 2028 | static int btrfs_issue_discard(struct block_device *bdev, u64 start, u64 len, |
| 2029 | u64 *discarded_bytes) |
| 2030 | { |
| 2031 | int j, ret = 0; |
| 2032 | u64 bytes_left, end; |
| 2033 | u64 aligned_start = ALIGN(start, 1 << 9); |
| 2034 | |
| 2035 | if (WARN_ON(start != aligned_start)) { |
| 2036 | len -= aligned_start - start; |
| 2037 | len = round_down(len, 1 << 9); |
| 2038 | start = aligned_start; |
| 2039 | } |
| 2040 | |
| 2041 | *discarded_bytes = 0; |
| 2042 | |
| 2043 | if (!len) |
| 2044 | return 0; |
| 2045 | |
| 2046 | end = start + len; |
| 2047 | bytes_left = len; |
| 2048 | |
| 2049 | /* Skip any superblocks on this device. */ |
| 2050 | for (j = 0; j < BTRFS_SUPER_MIRROR_MAX; j++) { |
| 2051 | u64 sb_start = btrfs_sb_offset(j); |
| 2052 | u64 sb_end = sb_start + BTRFS_SUPER_INFO_SIZE; |
| 2053 | u64 size = sb_start - start; |
| 2054 | |
| 2055 | if (!in_range(sb_start, start, bytes_left) && |
| 2056 | !in_range(sb_end, start, bytes_left) && |
| 2057 | !in_range(start, sb_start, BTRFS_SUPER_INFO_SIZE)) |
| 2058 | continue; |
| 2059 | |
| 2060 | /* |
| 2061 | * Superblock spans beginning of range. Adjust start and |
| 2062 | * try again. |
| 2063 | */ |
| 2064 | if (sb_start <= start) { |
| 2065 | start += sb_end - start; |
| 2066 | if (start > end) { |
| 2067 | bytes_left = 0; |
| 2068 | break; |
| 2069 | } |
| 2070 | bytes_left = end - start; |
| 2071 | continue; |
| 2072 | } |
| 2073 | |
| 2074 | if (size) { |
| 2075 | ret = blkdev_issue_discard(bdev, start >> 9, size >> 9, |
| 2076 | GFP_NOFS, 0); |
| 2077 | if (!ret) |
| 2078 | *discarded_bytes += size; |
| 2079 | else if (ret != -EOPNOTSUPP) |
| 2080 | return ret; |
| 2081 | } |
| 2082 | |
| 2083 | start = sb_end; |
| 2084 | if (start > end) { |
| 2085 | bytes_left = 0; |
| 2086 | break; |
| 2087 | } |
| 2088 | bytes_left = end - start; |
| 2089 | } |
| 2090 | |
| 2091 | if (bytes_left) { |
| 2092 | ret = blkdev_issue_discard(bdev, start >> 9, bytes_left >> 9, |
| 2093 | GFP_NOFS, 0); |
| 2094 | if (!ret) |
| 2095 | *discarded_bytes += bytes_left; |
| 2096 | } |
| 2097 | return ret; |
| 2098 | } |
| 2099 | |
| 2100 | int btrfs_discard_extent(struct btrfs_fs_info *fs_info, u64 bytenr, |
| 2101 | u64 num_bytes, u64 *actual_bytes) |
| 2102 | { |
| 2103 | int ret; |
| 2104 | u64 discarded_bytes = 0; |
| 2105 | struct btrfs_bio *bbio = NULL; |
| 2106 | |
| 2107 | |
| 2108 | /* |
| 2109 | * Avoid races with device replace and make sure our bbio has devices |
| 2110 | * associated to its stripes that don't go away while we are discarding. |
| 2111 | */ |
| 2112 | btrfs_bio_counter_inc_blocked(fs_info); |
| 2113 | /* Tell the block device(s) that the sectors can be discarded */ |
| 2114 | ret = btrfs_map_block(fs_info, BTRFS_MAP_DISCARD, bytenr, &num_bytes, |
| 2115 | &bbio, 0); |
| 2116 | /* Error condition is -ENOMEM */ |
| 2117 | if (!ret) { |
| 2118 | struct btrfs_bio_stripe *stripe = bbio->stripes; |
| 2119 | int i; |
| 2120 | |
| 2121 | |
| 2122 | for (i = 0; i < bbio->num_stripes; i++, stripe++) { |
| 2123 | u64 bytes; |
| 2124 | struct request_queue *req_q; |
| 2125 | |
| 2126 | if (!stripe->dev->bdev) { |
| 2127 | ASSERT(btrfs_test_opt(fs_info, DEGRADED)); |
| 2128 | continue; |
| 2129 | } |
| 2130 | req_q = bdev_get_queue(stripe->dev->bdev); |
| 2131 | if (!blk_queue_discard(req_q)) |
| 2132 | continue; |
| 2133 | |
| 2134 | ret = btrfs_issue_discard(stripe->dev->bdev, |
| 2135 | stripe->physical, |
| 2136 | stripe->length, |
| 2137 | &bytes); |
| 2138 | if (!ret) |
| 2139 | discarded_bytes += bytes; |
| 2140 | else if (ret != -EOPNOTSUPP) |
| 2141 | break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */ |
| 2142 | |
| 2143 | /* |
| 2144 | * Just in case we get back EOPNOTSUPP for some reason, |
| 2145 | * just ignore the return value so we don't screw up |
| 2146 | * people calling discard_extent. |
| 2147 | */ |
| 2148 | ret = 0; |
| 2149 | } |
| 2150 | btrfs_put_bbio(bbio); |
| 2151 | } |
| 2152 | btrfs_bio_counter_dec(fs_info); |
| 2153 | |
| 2154 | if (actual_bytes) |
| 2155 | *actual_bytes = discarded_bytes; |
| 2156 | |
| 2157 | |
| 2158 | if (ret == -EOPNOTSUPP) |
| 2159 | ret = 0; |
| 2160 | return ret; |
| 2161 | } |
| 2162 | |
| 2163 | /* Can return -ENOMEM */ |
| 2164 | int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
| 2165 | struct btrfs_root *root, |
| 2166 | u64 bytenr, u64 num_bytes, u64 parent, |
| 2167 | u64 root_objectid, u64 owner, u64 offset) |
| 2168 | { |
| 2169 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 2170 | int old_ref_mod, new_ref_mod; |
| 2171 | int ret; |
| 2172 | |
| 2173 | BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID && |
| 2174 | root_objectid == BTRFS_TREE_LOG_OBJECTID); |
| 2175 | |
| 2176 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, root_objectid, |
| 2177 | owner, offset, BTRFS_ADD_DELAYED_REF); |
| 2178 | |
| 2179 | if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
| 2180 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
| 2181 | num_bytes, parent, |
| 2182 | root_objectid, (int)owner, |
| 2183 | BTRFS_ADD_DELAYED_REF, NULL, |
| 2184 | &old_ref_mod, &new_ref_mod); |
| 2185 | } else { |
| 2186 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
| 2187 | num_bytes, parent, |
| 2188 | root_objectid, owner, offset, |
| 2189 | 0, BTRFS_ADD_DELAYED_REF, |
| 2190 | &old_ref_mod, &new_ref_mod); |
| 2191 | } |
| 2192 | |
| 2193 | if (ret == 0 && old_ref_mod < 0 && new_ref_mod >= 0) { |
| 2194 | bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID; |
| 2195 | |
| 2196 | add_pinned_bytes(fs_info, -num_bytes, metadata, root_objectid); |
| 2197 | } |
| 2198 | |
| 2199 | return ret; |
| 2200 | } |
| 2201 | |
| 2202 | /* |
| 2203 | * __btrfs_inc_extent_ref - insert backreference for a given extent |
| 2204 | * |
| 2205 | * @trans: Handle of transaction |
| 2206 | * |
| 2207 | * @node: The delayed ref node used to get the bytenr/length for |
| 2208 | * extent whose references are incremented. |
| 2209 | * |
| 2210 | * @parent: If this is a shared extent (BTRFS_SHARED_DATA_REF_KEY/ |
| 2211 | * BTRFS_SHARED_BLOCK_REF_KEY) then it holds the logical |
| 2212 | * bytenr of the parent block. Since new extents are always |
| 2213 | * created with indirect references, this will only be the case |
| 2214 | * when relocating a shared extent. In that case, root_objectid |
| 2215 | * will be BTRFS_TREE_RELOC_OBJECTID. Otheriwse, parent must |
| 2216 | * be 0 |
| 2217 | * |
| 2218 | * @root_objectid: The id of the root where this modification has originated, |
| 2219 | * this can be either one of the well-known metadata trees or |
| 2220 | * the subvolume id which references this extent. |
| 2221 | * |
| 2222 | * @owner: For data extents it is the inode number of the owning file. |
| 2223 | * For metadata extents this parameter holds the level in the |
| 2224 | * tree of the extent. |
| 2225 | * |
| 2226 | * @offset: For metadata extents the offset is ignored and is currently |
| 2227 | * always passed as 0. For data extents it is the fileoffset |
| 2228 | * this extent belongs to. |
| 2229 | * |
| 2230 | * @refs_to_add Number of references to add |
| 2231 | * |
| 2232 | * @extent_op Pointer to a structure, holding information necessary when |
| 2233 | * updating a tree block's flags |
| 2234 | * |
| 2235 | */ |
| 2236 | static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, |
| 2237 | struct btrfs_fs_info *fs_info, |
| 2238 | struct btrfs_delayed_ref_node *node, |
| 2239 | u64 parent, u64 root_objectid, |
| 2240 | u64 owner, u64 offset, int refs_to_add, |
| 2241 | struct btrfs_delayed_extent_op *extent_op) |
| 2242 | { |
| 2243 | struct btrfs_path *path; |
| 2244 | struct extent_buffer *leaf; |
| 2245 | struct btrfs_extent_item *item; |
| 2246 | struct btrfs_key key; |
| 2247 | u64 bytenr = node->bytenr; |
| 2248 | u64 num_bytes = node->num_bytes; |
| 2249 | u64 refs; |
| 2250 | int ret; |
| 2251 | |
| 2252 | path = btrfs_alloc_path(); |
| 2253 | if (!path) |
| 2254 | return -ENOMEM; |
| 2255 | |
| 2256 | path->reada = READA_FORWARD; |
| 2257 | path->leave_spinning = 1; |
| 2258 | /* this will setup the path even if it fails to insert the back ref */ |
| 2259 | ret = insert_inline_extent_backref(trans, fs_info, path, bytenr, |
| 2260 | num_bytes, parent, root_objectid, |
| 2261 | owner, offset, |
| 2262 | refs_to_add, extent_op); |
| 2263 | if ((ret < 0 && ret != -EAGAIN) || !ret) |
| 2264 | goto out; |
| 2265 | |
| 2266 | /* |
| 2267 | * Ok we had -EAGAIN which means we didn't have space to insert and |
| 2268 | * inline extent ref, so just update the reference count and add a |
| 2269 | * normal backref. |
| 2270 | */ |
| 2271 | leaf = path->nodes[0]; |
| 2272 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
| 2273 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
| 2274 | refs = btrfs_extent_refs(leaf, item); |
| 2275 | btrfs_set_extent_refs(leaf, item, refs + refs_to_add); |
| 2276 | if (extent_op) |
| 2277 | __run_delayed_extent_op(extent_op, leaf, item); |
| 2278 | |
| 2279 | btrfs_mark_buffer_dirty(leaf); |
| 2280 | btrfs_release_path(path); |
| 2281 | |
| 2282 | path->reada = READA_FORWARD; |
| 2283 | path->leave_spinning = 1; |
| 2284 | /* now insert the actual backref */ |
| 2285 | ret = insert_extent_backref(trans, path, bytenr, parent, root_objectid, |
| 2286 | owner, offset, refs_to_add); |
| 2287 | if (ret) |
| 2288 | btrfs_abort_transaction(trans, ret); |
| 2289 | out: |
| 2290 | btrfs_free_path(path); |
| 2291 | return ret; |
| 2292 | } |
| 2293 | |
| 2294 | static int run_delayed_data_ref(struct btrfs_trans_handle *trans, |
| 2295 | struct btrfs_fs_info *fs_info, |
| 2296 | struct btrfs_delayed_ref_node *node, |
| 2297 | struct btrfs_delayed_extent_op *extent_op, |
| 2298 | int insert_reserved) |
| 2299 | { |
| 2300 | int ret = 0; |
| 2301 | struct btrfs_delayed_data_ref *ref; |
| 2302 | struct btrfs_key ins; |
| 2303 | u64 parent = 0; |
| 2304 | u64 ref_root = 0; |
| 2305 | u64 flags = 0; |
| 2306 | |
| 2307 | ins.objectid = node->bytenr; |
| 2308 | ins.offset = node->num_bytes; |
| 2309 | ins.type = BTRFS_EXTENT_ITEM_KEY; |
| 2310 | |
| 2311 | ref = btrfs_delayed_node_to_data_ref(node); |
| 2312 | trace_run_delayed_data_ref(fs_info, node, ref, node->action); |
| 2313 | |
| 2314 | if (node->type == BTRFS_SHARED_DATA_REF_KEY) |
| 2315 | parent = ref->parent; |
| 2316 | ref_root = ref->root; |
| 2317 | |
| 2318 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
| 2319 | if (extent_op) |
| 2320 | flags |= extent_op->flags_to_set; |
| 2321 | ret = alloc_reserved_file_extent(trans, fs_info, |
| 2322 | parent, ref_root, flags, |
| 2323 | ref->objectid, ref->offset, |
| 2324 | &ins, node->ref_mod); |
| 2325 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
| 2326 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, parent, |
| 2327 | ref_root, ref->objectid, |
| 2328 | ref->offset, node->ref_mod, |
| 2329 | extent_op); |
| 2330 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
| 2331 | ret = __btrfs_free_extent(trans, fs_info, node, parent, |
| 2332 | ref_root, ref->objectid, |
| 2333 | ref->offset, node->ref_mod, |
| 2334 | extent_op); |
| 2335 | } else { |
| 2336 | BUG(); |
| 2337 | } |
| 2338 | return ret; |
| 2339 | } |
| 2340 | |
| 2341 | static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op, |
| 2342 | struct extent_buffer *leaf, |
| 2343 | struct btrfs_extent_item *ei) |
| 2344 | { |
| 2345 | u64 flags = btrfs_extent_flags(leaf, ei); |
| 2346 | if (extent_op->update_flags) { |
| 2347 | flags |= extent_op->flags_to_set; |
| 2348 | btrfs_set_extent_flags(leaf, ei, flags); |
| 2349 | } |
| 2350 | |
| 2351 | if (extent_op->update_key) { |
| 2352 | struct btrfs_tree_block_info *bi; |
| 2353 | BUG_ON(!(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)); |
| 2354 | bi = (struct btrfs_tree_block_info *)(ei + 1); |
| 2355 | btrfs_set_tree_block_key(leaf, bi, &extent_op->key); |
| 2356 | } |
| 2357 | } |
| 2358 | |
| 2359 | static int run_delayed_extent_op(struct btrfs_trans_handle *trans, |
| 2360 | struct btrfs_fs_info *fs_info, |
| 2361 | struct btrfs_delayed_ref_head *head, |
| 2362 | struct btrfs_delayed_extent_op *extent_op) |
| 2363 | { |
| 2364 | struct btrfs_key key; |
| 2365 | struct btrfs_path *path; |
| 2366 | struct btrfs_extent_item *ei; |
| 2367 | struct extent_buffer *leaf; |
| 2368 | u32 item_size; |
| 2369 | int ret; |
| 2370 | int err = 0; |
| 2371 | int metadata = !extent_op->is_data; |
| 2372 | |
| 2373 | if (trans->aborted) |
| 2374 | return 0; |
| 2375 | |
| 2376 | if (metadata && !btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
| 2377 | metadata = 0; |
| 2378 | |
| 2379 | path = btrfs_alloc_path(); |
| 2380 | if (!path) |
| 2381 | return -ENOMEM; |
| 2382 | |
| 2383 | key.objectid = head->bytenr; |
| 2384 | |
| 2385 | if (metadata) { |
| 2386 | key.type = BTRFS_METADATA_ITEM_KEY; |
| 2387 | key.offset = extent_op->level; |
| 2388 | } else { |
| 2389 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 2390 | key.offset = head->num_bytes; |
| 2391 | } |
| 2392 | |
| 2393 | again: |
| 2394 | path->reada = READA_FORWARD; |
| 2395 | path->leave_spinning = 1; |
| 2396 | ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 1); |
| 2397 | if (ret < 0) { |
| 2398 | err = ret; |
| 2399 | goto out; |
| 2400 | } |
| 2401 | if (ret > 0) { |
| 2402 | if (metadata) { |
| 2403 | if (path->slots[0] > 0) { |
| 2404 | path->slots[0]--; |
| 2405 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
| 2406 | path->slots[0]); |
| 2407 | if (key.objectid == head->bytenr && |
| 2408 | key.type == BTRFS_EXTENT_ITEM_KEY && |
| 2409 | key.offset == head->num_bytes) |
| 2410 | ret = 0; |
| 2411 | } |
| 2412 | if (ret > 0) { |
| 2413 | btrfs_release_path(path); |
| 2414 | metadata = 0; |
| 2415 | |
| 2416 | key.objectid = head->bytenr; |
| 2417 | key.offset = head->num_bytes; |
| 2418 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 2419 | goto again; |
| 2420 | } |
| 2421 | } else { |
| 2422 | err = -EIO; |
| 2423 | goto out; |
| 2424 | } |
| 2425 | } |
| 2426 | |
| 2427 | leaf = path->nodes[0]; |
| 2428 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); |
| 2429 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 2430 | if (item_size < sizeof(*ei)) { |
| 2431 | ret = convert_extent_item_v0(trans, fs_info, path, (u64)-1, 0); |
| 2432 | if (ret < 0) { |
| 2433 | err = ret; |
| 2434 | goto out; |
| 2435 | } |
| 2436 | leaf = path->nodes[0]; |
| 2437 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); |
| 2438 | } |
| 2439 | #endif |
| 2440 | BUG_ON(item_size < sizeof(*ei)); |
| 2441 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
| 2442 | __run_delayed_extent_op(extent_op, leaf, ei); |
| 2443 | |
| 2444 | btrfs_mark_buffer_dirty(leaf); |
| 2445 | out: |
| 2446 | btrfs_free_path(path); |
| 2447 | return err; |
| 2448 | } |
| 2449 | |
| 2450 | static int run_delayed_tree_ref(struct btrfs_trans_handle *trans, |
| 2451 | struct btrfs_fs_info *fs_info, |
| 2452 | struct btrfs_delayed_ref_node *node, |
| 2453 | struct btrfs_delayed_extent_op *extent_op, |
| 2454 | int insert_reserved) |
| 2455 | { |
| 2456 | int ret = 0; |
| 2457 | struct btrfs_delayed_tree_ref *ref; |
| 2458 | u64 parent = 0; |
| 2459 | u64 ref_root = 0; |
| 2460 | |
| 2461 | ref = btrfs_delayed_node_to_tree_ref(node); |
| 2462 | trace_run_delayed_tree_ref(fs_info, node, ref, node->action); |
| 2463 | |
| 2464 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
| 2465 | parent = ref->parent; |
| 2466 | ref_root = ref->root; |
| 2467 | |
| 2468 | if (node->ref_mod != 1) { |
| 2469 | btrfs_err(fs_info, |
| 2470 | "btree block(%llu) has %d references rather than 1: action %d ref_root %llu parent %llu", |
| 2471 | node->bytenr, node->ref_mod, node->action, ref_root, |
| 2472 | parent); |
| 2473 | return -EIO; |
| 2474 | } |
| 2475 | if (node->action == BTRFS_ADD_DELAYED_REF && insert_reserved) { |
| 2476 | BUG_ON(!extent_op || !extent_op->update_flags); |
| 2477 | ret = alloc_reserved_tree_block(trans, node, extent_op); |
| 2478 | } else if (node->action == BTRFS_ADD_DELAYED_REF) { |
| 2479 | ret = __btrfs_inc_extent_ref(trans, fs_info, node, |
| 2480 | parent, ref_root, |
| 2481 | ref->level, 0, 1, |
| 2482 | extent_op); |
| 2483 | } else if (node->action == BTRFS_DROP_DELAYED_REF) { |
| 2484 | ret = __btrfs_free_extent(trans, fs_info, node, |
| 2485 | parent, ref_root, |
| 2486 | ref->level, 0, 1, extent_op); |
| 2487 | } else { |
| 2488 | BUG(); |
| 2489 | } |
| 2490 | return ret; |
| 2491 | } |
| 2492 | |
| 2493 | /* helper function to actually process a single delayed ref entry */ |
| 2494 | static int run_one_delayed_ref(struct btrfs_trans_handle *trans, |
| 2495 | struct btrfs_fs_info *fs_info, |
| 2496 | struct btrfs_delayed_ref_node *node, |
| 2497 | struct btrfs_delayed_extent_op *extent_op, |
| 2498 | int insert_reserved) |
| 2499 | { |
| 2500 | int ret = 0; |
| 2501 | |
| 2502 | if (trans->aborted) { |
| 2503 | if (insert_reserved) |
| 2504 | btrfs_pin_extent(fs_info, node->bytenr, |
| 2505 | node->num_bytes, 1); |
| 2506 | return 0; |
| 2507 | } |
| 2508 | |
| 2509 | if (node->type == BTRFS_TREE_BLOCK_REF_KEY || |
| 2510 | node->type == BTRFS_SHARED_BLOCK_REF_KEY) |
| 2511 | ret = run_delayed_tree_ref(trans, fs_info, node, extent_op, |
| 2512 | insert_reserved); |
| 2513 | else if (node->type == BTRFS_EXTENT_DATA_REF_KEY || |
| 2514 | node->type == BTRFS_SHARED_DATA_REF_KEY) |
| 2515 | ret = run_delayed_data_ref(trans, fs_info, node, extent_op, |
| 2516 | insert_reserved); |
| 2517 | else |
| 2518 | BUG(); |
| 2519 | return ret; |
| 2520 | } |
| 2521 | |
| 2522 | static inline struct btrfs_delayed_ref_node * |
| 2523 | select_delayed_ref(struct btrfs_delayed_ref_head *head) |
| 2524 | { |
| 2525 | struct btrfs_delayed_ref_node *ref; |
| 2526 | |
| 2527 | if (RB_EMPTY_ROOT(&head->ref_tree)) |
| 2528 | return NULL; |
| 2529 | |
| 2530 | /* |
| 2531 | * Select a delayed ref of type BTRFS_ADD_DELAYED_REF first. |
| 2532 | * This is to prevent a ref count from going down to zero, which deletes |
| 2533 | * the extent item from the extent tree, when there still are references |
| 2534 | * to add, which would fail because they would not find the extent item. |
| 2535 | */ |
| 2536 | if (!list_empty(&head->ref_add_list)) |
| 2537 | return list_first_entry(&head->ref_add_list, |
| 2538 | struct btrfs_delayed_ref_node, add_list); |
| 2539 | |
| 2540 | ref = rb_entry(rb_first(&head->ref_tree), |
| 2541 | struct btrfs_delayed_ref_node, ref_node); |
| 2542 | ASSERT(list_empty(&ref->add_list)); |
| 2543 | return ref; |
| 2544 | } |
| 2545 | |
| 2546 | static void unselect_delayed_ref_head(struct btrfs_delayed_ref_root *delayed_refs, |
| 2547 | struct btrfs_delayed_ref_head *head) |
| 2548 | { |
| 2549 | spin_lock(&delayed_refs->lock); |
| 2550 | head->processing = 0; |
| 2551 | delayed_refs->num_heads_ready++; |
| 2552 | spin_unlock(&delayed_refs->lock); |
| 2553 | btrfs_delayed_ref_unlock(head); |
| 2554 | } |
| 2555 | |
| 2556 | static int cleanup_extent_op(struct btrfs_trans_handle *trans, |
| 2557 | struct btrfs_fs_info *fs_info, |
| 2558 | struct btrfs_delayed_ref_head *head) |
| 2559 | { |
| 2560 | struct btrfs_delayed_extent_op *extent_op = head->extent_op; |
| 2561 | int ret; |
| 2562 | |
| 2563 | if (!extent_op) |
| 2564 | return 0; |
| 2565 | head->extent_op = NULL; |
| 2566 | if (head->must_insert_reserved) { |
| 2567 | btrfs_free_delayed_extent_op(extent_op); |
| 2568 | return 0; |
| 2569 | } |
| 2570 | spin_unlock(&head->lock); |
| 2571 | ret = run_delayed_extent_op(trans, fs_info, head, extent_op); |
| 2572 | btrfs_free_delayed_extent_op(extent_op); |
| 2573 | return ret ? ret : 1; |
| 2574 | } |
| 2575 | |
| 2576 | static int cleanup_ref_head(struct btrfs_trans_handle *trans, |
| 2577 | struct btrfs_fs_info *fs_info, |
| 2578 | struct btrfs_delayed_ref_head *head) |
| 2579 | { |
| 2580 | struct btrfs_delayed_ref_root *delayed_refs; |
| 2581 | int ret; |
| 2582 | |
| 2583 | delayed_refs = &trans->transaction->delayed_refs; |
| 2584 | |
| 2585 | ret = cleanup_extent_op(trans, fs_info, head); |
| 2586 | if (ret < 0) { |
| 2587 | unselect_delayed_ref_head(delayed_refs, head); |
| 2588 | btrfs_debug(fs_info, "run_delayed_extent_op returned %d", ret); |
| 2589 | return ret; |
| 2590 | } else if (ret) { |
| 2591 | return ret; |
| 2592 | } |
| 2593 | |
| 2594 | /* |
| 2595 | * Need to drop our head ref lock and re-acquire the delayed ref lock |
| 2596 | * and then re-check to make sure nobody got added. |
| 2597 | */ |
| 2598 | spin_unlock(&head->lock); |
| 2599 | spin_lock(&delayed_refs->lock); |
| 2600 | spin_lock(&head->lock); |
| 2601 | if (!RB_EMPTY_ROOT(&head->ref_tree) || head->extent_op) { |
| 2602 | spin_unlock(&head->lock); |
| 2603 | spin_unlock(&delayed_refs->lock); |
| 2604 | return 1; |
| 2605 | } |
| 2606 | delayed_refs->num_heads--; |
| 2607 | rb_erase(&head->href_node, &delayed_refs->href_root); |
| 2608 | RB_CLEAR_NODE(&head->href_node); |
| 2609 | spin_unlock(&head->lock); |
| 2610 | spin_unlock(&delayed_refs->lock); |
| 2611 | atomic_dec(&delayed_refs->num_entries); |
| 2612 | |
| 2613 | trace_run_delayed_ref_head(fs_info, head, 0); |
| 2614 | |
| 2615 | if (head->total_ref_mod < 0) { |
| 2616 | struct btrfs_space_info *space_info; |
| 2617 | u64 flags; |
| 2618 | |
| 2619 | if (head->is_data) |
| 2620 | flags = BTRFS_BLOCK_GROUP_DATA; |
| 2621 | else if (head->is_system) |
| 2622 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
| 2623 | else |
| 2624 | flags = BTRFS_BLOCK_GROUP_METADATA; |
| 2625 | space_info = __find_space_info(fs_info, flags); |
| 2626 | ASSERT(space_info); |
| 2627 | percpu_counter_add(&space_info->total_bytes_pinned, |
| 2628 | -head->num_bytes); |
| 2629 | |
| 2630 | if (head->is_data) { |
| 2631 | spin_lock(&delayed_refs->lock); |
| 2632 | delayed_refs->pending_csums -= head->num_bytes; |
| 2633 | spin_unlock(&delayed_refs->lock); |
| 2634 | } |
| 2635 | } |
| 2636 | |
| 2637 | if (head->must_insert_reserved) { |
| 2638 | btrfs_pin_extent(fs_info, head->bytenr, |
| 2639 | head->num_bytes, 1); |
| 2640 | if (head->is_data) { |
| 2641 | ret = btrfs_del_csums(trans, fs_info, head->bytenr, |
| 2642 | head->num_bytes); |
| 2643 | } |
| 2644 | } |
| 2645 | |
| 2646 | /* Also free its reserved qgroup space */ |
| 2647 | btrfs_qgroup_free_delayed_ref(fs_info, head->qgroup_ref_root, |
| 2648 | head->qgroup_reserved); |
| 2649 | btrfs_delayed_ref_unlock(head); |
| 2650 | btrfs_put_delayed_ref_head(head); |
| 2651 | return 0; |
| 2652 | } |
| 2653 | |
| 2654 | /* |
| 2655 | * Returns 0 on success or if called with an already aborted transaction. |
| 2656 | * Returns -ENOMEM or -EIO on failure and will abort the transaction. |
| 2657 | */ |
| 2658 | static noinline int __btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
| 2659 | unsigned long nr) |
| 2660 | { |
| 2661 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 2662 | struct btrfs_delayed_ref_root *delayed_refs; |
| 2663 | struct btrfs_delayed_ref_node *ref; |
| 2664 | struct btrfs_delayed_ref_head *locked_ref = NULL; |
| 2665 | struct btrfs_delayed_extent_op *extent_op; |
| 2666 | ktime_t start = ktime_get(); |
| 2667 | int ret; |
| 2668 | unsigned long count = 0; |
| 2669 | unsigned long actual_count = 0; |
| 2670 | int must_insert_reserved = 0; |
| 2671 | |
| 2672 | delayed_refs = &trans->transaction->delayed_refs; |
| 2673 | while (1) { |
| 2674 | if (!locked_ref) { |
| 2675 | if (count >= nr) |
| 2676 | break; |
| 2677 | |
| 2678 | spin_lock(&delayed_refs->lock); |
| 2679 | locked_ref = btrfs_select_ref_head(trans); |
| 2680 | if (!locked_ref) { |
| 2681 | spin_unlock(&delayed_refs->lock); |
| 2682 | break; |
| 2683 | } |
| 2684 | |
| 2685 | /* grab the lock that says we are going to process |
| 2686 | * all the refs for this head */ |
| 2687 | ret = btrfs_delayed_ref_lock(trans, locked_ref); |
| 2688 | spin_unlock(&delayed_refs->lock); |
| 2689 | /* |
| 2690 | * we may have dropped the spin lock to get the head |
| 2691 | * mutex lock, and that might have given someone else |
| 2692 | * time to free the head. If that's true, it has been |
| 2693 | * removed from our list and we can move on. |
| 2694 | */ |
| 2695 | if (ret == -EAGAIN) { |
| 2696 | locked_ref = NULL; |
| 2697 | count++; |
| 2698 | continue; |
| 2699 | } |
| 2700 | } |
| 2701 | |
| 2702 | /* |
| 2703 | * We need to try and merge add/drops of the same ref since we |
| 2704 | * can run into issues with relocate dropping the implicit ref |
| 2705 | * and then it being added back again before the drop can |
| 2706 | * finish. If we merged anything we need to re-loop so we can |
| 2707 | * get a good ref. |
| 2708 | * Or we can get node references of the same type that weren't |
| 2709 | * merged when created due to bumps in the tree mod seq, and |
| 2710 | * we need to merge them to prevent adding an inline extent |
| 2711 | * backref before dropping it (triggering a BUG_ON at |
| 2712 | * insert_inline_extent_backref()). |
| 2713 | */ |
| 2714 | spin_lock(&locked_ref->lock); |
| 2715 | btrfs_merge_delayed_refs(trans, delayed_refs, locked_ref); |
| 2716 | |
| 2717 | ref = select_delayed_ref(locked_ref); |
| 2718 | |
| 2719 | if (ref && ref->seq && |
| 2720 | btrfs_check_delayed_seq(fs_info, ref->seq)) { |
| 2721 | spin_unlock(&locked_ref->lock); |
| 2722 | unselect_delayed_ref_head(delayed_refs, locked_ref); |
| 2723 | locked_ref = NULL; |
| 2724 | cond_resched(); |
| 2725 | count++; |
| 2726 | continue; |
| 2727 | } |
| 2728 | |
| 2729 | /* |
| 2730 | * We're done processing refs in this ref_head, clean everything |
| 2731 | * up and move on to the next ref_head. |
| 2732 | */ |
| 2733 | if (!ref) { |
| 2734 | ret = cleanup_ref_head(trans, fs_info, locked_ref); |
| 2735 | if (ret > 0 ) { |
| 2736 | /* We dropped our lock, we need to loop. */ |
| 2737 | ret = 0; |
| 2738 | continue; |
| 2739 | } else if (ret) { |
| 2740 | return ret; |
| 2741 | } |
| 2742 | locked_ref = NULL; |
| 2743 | count++; |
| 2744 | continue; |
| 2745 | } |
| 2746 | |
| 2747 | actual_count++; |
| 2748 | ref->in_tree = 0; |
| 2749 | rb_erase(&ref->ref_node, &locked_ref->ref_tree); |
| 2750 | RB_CLEAR_NODE(&ref->ref_node); |
| 2751 | if (!list_empty(&ref->add_list)) |
| 2752 | list_del(&ref->add_list); |
| 2753 | /* |
| 2754 | * When we play the delayed ref, also correct the ref_mod on |
| 2755 | * head |
| 2756 | */ |
| 2757 | switch (ref->action) { |
| 2758 | case BTRFS_ADD_DELAYED_REF: |
| 2759 | case BTRFS_ADD_DELAYED_EXTENT: |
| 2760 | locked_ref->ref_mod -= ref->ref_mod; |
| 2761 | break; |
| 2762 | case BTRFS_DROP_DELAYED_REF: |
| 2763 | locked_ref->ref_mod += ref->ref_mod; |
| 2764 | break; |
| 2765 | default: |
| 2766 | WARN_ON(1); |
| 2767 | } |
| 2768 | atomic_dec(&delayed_refs->num_entries); |
| 2769 | |
| 2770 | /* |
| 2771 | * Record the must-insert_reserved flag before we drop the spin |
| 2772 | * lock. |
| 2773 | */ |
| 2774 | must_insert_reserved = locked_ref->must_insert_reserved; |
| 2775 | locked_ref->must_insert_reserved = 0; |
| 2776 | |
| 2777 | extent_op = locked_ref->extent_op; |
| 2778 | locked_ref->extent_op = NULL; |
| 2779 | spin_unlock(&locked_ref->lock); |
| 2780 | |
| 2781 | ret = run_one_delayed_ref(trans, fs_info, ref, extent_op, |
| 2782 | must_insert_reserved); |
| 2783 | |
| 2784 | btrfs_free_delayed_extent_op(extent_op); |
| 2785 | if (ret) { |
| 2786 | unselect_delayed_ref_head(delayed_refs, locked_ref); |
| 2787 | btrfs_put_delayed_ref(ref); |
| 2788 | btrfs_debug(fs_info, "run_one_delayed_ref returned %d", |
| 2789 | ret); |
| 2790 | return ret; |
| 2791 | } |
| 2792 | |
| 2793 | btrfs_put_delayed_ref(ref); |
| 2794 | count++; |
| 2795 | cond_resched(); |
| 2796 | } |
| 2797 | |
| 2798 | /* |
| 2799 | * We don't want to include ref heads since we can have empty ref heads |
| 2800 | * and those will drastically skew our runtime down since we just do |
| 2801 | * accounting, no actual extent tree updates. |
| 2802 | */ |
| 2803 | if (actual_count > 0) { |
| 2804 | u64 runtime = ktime_to_ns(ktime_sub(ktime_get(), start)); |
| 2805 | u64 avg; |
| 2806 | |
| 2807 | /* |
| 2808 | * We weigh the current average higher than our current runtime |
| 2809 | * to avoid large swings in the average. |
| 2810 | */ |
| 2811 | spin_lock(&delayed_refs->lock); |
| 2812 | avg = fs_info->avg_delayed_ref_runtime * 3 + runtime; |
| 2813 | fs_info->avg_delayed_ref_runtime = avg >> 2; /* div by 4 */ |
| 2814 | spin_unlock(&delayed_refs->lock); |
| 2815 | } |
| 2816 | return 0; |
| 2817 | } |
| 2818 | |
| 2819 | #ifdef SCRAMBLE_DELAYED_REFS |
| 2820 | /* |
| 2821 | * Normally delayed refs get processed in ascending bytenr order. This |
| 2822 | * correlates in most cases to the order added. To expose dependencies on this |
| 2823 | * order, we start to process the tree in the middle instead of the beginning |
| 2824 | */ |
| 2825 | static u64 find_middle(struct rb_root *root) |
| 2826 | { |
| 2827 | struct rb_node *n = root->rb_node; |
| 2828 | struct btrfs_delayed_ref_node *entry; |
| 2829 | int alt = 1; |
| 2830 | u64 middle; |
| 2831 | u64 first = 0, last = 0; |
| 2832 | |
| 2833 | n = rb_first(root); |
| 2834 | if (n) { |
| 2835 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); |
| 2836 | first = entry->bytenr; |
| 2837 | } |
| 2838 | n = rb_last(root); |
| 2839 | if (n) { |
| 2840 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); |
| 2841 | last = entry->bytenr; |
| 2842 | } |
| 2843 | n = root->rb_node; |
| 2844 | |
| 2845 | while (n) { |
| 2846 | entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node); |
| 2847 | WARN_ON(!entry->in_tree); |
| 2848 | |
| 2849 | middle = entry->bytenr; |
| 2850 | |
| 2851 | if (alt) |
| 2852 | n = n->rb_left; |
| 2853 | else |
| 2854 | n = n->rb_right; |
| 2855 | |
| 2856 | alt = 1 - alt; |
| 2857 | } |
| 2858 | return middle; |
| 2859 | } |
| 2860 | #endif |
| 2861 | |
| 2862 | static inline u64 heads_to_leaves(struct btrfs_fs_info *fs_info, u64 heads) |
| 2863 | { |
| 2864 | u64 num_bytes; |
| 2865 | |
| 2866 | num_bytes = heads * (sizeof(struct btrfs_extent_item) + |
| 2867 | sizeof(struct btrfs_extent_inline_ref)); |
| 2868 | if (!btrfs_fs_incompat(fs_info, SKINNY_METADATA)) |
| 2869 | num_bytes += heads * sizeof(struct btrfs_tree_block_info); |
| 2870 | |
| 2871 | /* |
| 2872 | * We don't ever fill up leaves all the way so multiply by 2 just to be |
| 2873 | * closer to what we're really going to want to use. |
| 2874 | */ |
| 2875 | return div_u64(num_bytes, BTRFS_LEAF_DATA_SIZE(fs_info)); |
| 2876 | } |
| 2877 | |
| 2878 | /* |
| 2879 | * Takes the number of bytes to be csumm'ed and figures out how many leaves it |
| 2880 | * would require to store the csums for that many bytes. |
| 2881 | */ |
| 2882 | u64 btrfs_csum_bytes_to_leaves(struct btrfs_fs_info *fs_info, u64 csum_bytes) |
| 2883 | { |
| 2884 | u64 csum_size; |
| 2885 | u64 num_csums_per_leaf; |
| 2886 | u64 num_csums; |
| 2887 | |
| 2888 | csum_size = BTRFS_MAX_ITEM_SIZE(fs_info); |
| 2889 | num_csums_per_leaf = div64_u64(csum_size, |
| 2890 | (u64)btrfs_super_csum_size(fs_info->super_copy)); |
| 2891 | num_csums = div64_u64(csum_bytes, fs_info->sectorsize); |
| 2892 | num_csums += num_csums_per_leaf - 1; |
| 2893 | num_csums = div64_u64(num_csums, num_csums_per_leaf); |
| 2894 | return num_csums; |
| 2895 | } |
| 2896 | |
| 2897 | int btrfs_check_space_for_delayed_refs(struct btrfs_trans_handle *trans, |
| 2898 | struct btrfs_fs_info *fs_info) |
| 2899 | { |
| 2900 | struct btrfs_block_rsv *global_rsv; |
| 2901 | u64 num_heads = trans->transaction->delayed_refs.num_heads_ready; |
| 2902 | u64 csum_bytes = trans->transaction->delayed_refs.pending_csums; |
| 2903 | unsigned int num_dirty_bgs = trans->transaction->num_dirty_bgs; |
| 2904 | u64 num_bytes, num_dirty_bgs_bytes; |
| 2905 | int ret = 0; |
| 2906 | |
| 2907 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
| 2908 | num_heads = heads_to_leaves(fs_info, num_heads); |
| 2909 | if (num_heads > 1) |
| 2910 | num_bytes += (num_heads - 1) * fs_info->nodesize; |
| 2911 | num_bytes <<= 1; |
| 2912 | num_bytes += btrfs_csum_bytes_to_leaves(fs_info, csum_bytes) * |
| 2913 | fs_info->nodesize; |
| 2914 | num_dirty_bgs_bytes = btrfs_calc_trans_metadata_size(fs_info, |
| 2915 | num_dirty_bgs); |
| 2916 | global_rsv = &fs_info->global_block_rsv; |
| 2917 | |
| 2918 | /* |
| 2919 | * If we can't allocate any more chunks lets make sure we have _lots_ of |
| 2920 | * wiggle room since running delayed refs can create more delayed refs. |
| 2921 | */ |
| 2922 | if (global_rsv->space_info->full) { |
| 2923 | num_dirty_bgs_bytes <<= 1; |
| 2924 | num_bytes <<= 1; |
| 2925 | } |
| 2926 | |
| 2927 | spin_lock(&global_rsv->lock); |
| 2928 | if (global_rsv->reserved <= num_bytes + num_dirty_bgs_bytes) |
| 2929 | ret = 1; |
| 2930 | spin_unlock(&global_rsv->lock); |
| 2931 | return ret; |
| 2932 | } |
| 2933 | |
| 2934 | int btrfs_should_throttle_delayed_refs(struct btrfs_trans_handle *trans, |
| 2935 | struct btrfs_fs_info *fs_info) |
| 2936 | { |
| 2937 | u64 num_entries = |
| 2938 | atomic_read(&trans->transaction->delayed_refs.num_entries); |
| 2939 | u64 avg_runtime; |
| 2940 | u64 val; |
| 2941 | |
| 2942 | smp_mb(); |
| 2943 | avg_runtime = fs_info->avg_delayed_ref_runtime; |
| 2944 | val = num_entries * avg_runtime; |
| 2945 | if (val >= NSEC_PER_SEC) |
| 2946 | return 1; |
| 2947 | if (val >= NSEC_PER_SEC / 2) |
| 2948 | return 2; |
| 2949 | |
| 2950 | return btrfs_check_space_for_delayed_refs(trans, fs_info); |
| 2951 | } |
| 2952 | |
| 2953 | struct async_delayed_refs { |
| 2954 | struct btrfs_root *root; |
| 2955 | u64 transid; |
| 2956 | int count; |
| 2957 | int error; |
| 2958 | int sync; |
| 2959 | struct completion wait; |
| 2960 | struct btrfs_work work; |
| 2961 | }; |
| 2962 | |
| 2963 | static inline struct async_delayed_refs * |
| 2964 | to_async_delayed_refs(struct btrfs_work *work) |
| 2965 | { |
| 2966 | return container_of(work, struct async_delayed_refs, work); |
| 2967 | } |
| 2968 | |
| 2969 | static void delayed_ref_async_start(struct btrfs_work *work) |
| 2970 | { |
| 2971 | struct async_delayed_refs *async = to_async_delayed_refs(work); |
| 2972 | struct btrfs_trans_handle *trans; |
| 2973 | struct btrfs_fs_info *fs_info = async->root->fs_info; |
| 2974 | int ret; |
| 2975 | |
| 2976 | /* if the commit is already started, we don't need to wait here */ |
| 2977 | if (btrfs_transaction_blocked(fs_info)) |
| 2978 | goto done; |
| 2979 | |
| 2980 | trans = btrfs_join_transaction(async->root); |
| 2981 | if (IS_ERR(trans)) { |
| 2982 | async->error = PTR_ERR(trans); |
| 2983 | goto done; |
| 2984 | } |
| 2985 | |
| 2986 | /* |
| 2987 | * trans->sync means that when we call end_transaction, we won't |
| 2988 | * wait on delayed refs |
| 2989 | */ |
| 2990 | trans->sync = true; |
| 2991 | |
| 2992 | /* Don't bother flushing if we got into a different transaction */ |
| 2993 | if (trans->transid > async->transid) |
| 2994 | goto end; |
| 2995 | |
| 2996 | ret = btrfs_run_delayed_refs(trans, async->count); |
| 2997 | if (ret) |
| 2998 | async->error = ret; |
| 2999 | end: |
| 3000 | ret = btrfs_end_transaction(trans); |
| 3001 | if (ret && !async->error) |
| 3002 | async->error = ret; |
| 3003 | done: |
| 3004 | if (async->sync) |
| 3005 | complete(&async->wait); |
| 3006 | else |
| 3007 | kfree(async); |
| 3008 | } |
| 3009 | |
| 3010 | int btrfs_async_run_delayed_refs(struct btrfs_fs_info *fs_info, |
| 3011 | unsigned long count, u64 transid, int wait) |
| 3012 | { |
| 3013 | struct async_delayed_refs *async; |
| 3014 | int ret; |
| 3015 | |
| 3016 | async = kmalloc(sizeof(*async), GFP_NOFS); |
| 3017 | if (!async) |
| 3018 | return -ENOMEM; |
| 3019 | |
| 3020 | async->root = fs_info->tree_root; |
| 3021 | async->count = count; |
| 3022 | async->error = 0; |
| 3023 | async->transid = transid; |
| 3024 | if (wait) |
| 3025 | async->sync = 1; |
| 3026 | else |
| 3027 | async->sync = 0; |
| 3028 | init_completion(&async->wait); |
| 3029 | |
| 3030 | btrfs_init_work(&async->work, btrfs_extent_refs_helper, |
| 3031 | delayed_ref_async_start, NULL, NULL); |
| 3032 | |
| 3033 | btrfs_queue_work(fs_info->extent_workers, &async->work); |
| 3034 | |
| 3035 | if (wait) { |
| 3036 | wait_for_completion(&async->wait); |
| 3037 | ret = async->error; |
| 3038 | kfree(async); |
| 3039 | return ret; |
| 3040 | } |
| 3041 | return 0; |
| 3042 | } |
| 3043 | |
| 3044 | /* |
| 3045 | * this starts processing the delayed reference count updates and |
| 3046 | * extent insertions we have queued up so far. count can be |
| 3047 | * 0, which means to process everything in the tree at the start |
| 3048 | * of the run (but not newly added entries), or it can be some target |
| 3049 | * number you'd like to process. |
| 3050 | * |
| 3051 | * Returns 0 on success or if called with an aborted transaction |
| 3052 | * Returns <0 on error and aborts the transaction |
| 3053 | */ |
| 3054 | int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, |
| 3055 | unsigned long count) |
| 3056 | { |
| 3057 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 3058 | struct rb_node *node; |
| 3059 | struct btrfs_delayed_ref_root *delayed_refs; |
| 3060 | struct btrfs_delayed_ref_head *head; |
| 3061 | int ret; |
| 3062 | int run_all = count == (unsigned long)-1; |
| 3063 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
| 3064 | |
| 3065 | /* We'll clean this up in btrfs_cleanup_transaction */ |
| 3066 | if (trans->aborted) |
| 3067 | return 0; |
| 3068 | |
| 3069 | if (test_bit(BTRFS_FS_CREATING_FREE_SPACE_TREE, &fs_info->flags)) |
| 3070 | return 0; |
| 3071 | |
| 3072 | delayed_refs = &trans->transaction->delayed_refs; |
| 3073 | if (count == 0) |
| 3074 | count = atomic_read(&delayed_refs->num_entries) * 2; |
| 3075 | |
| 3076 | again: |
| 3077 | #ifdef SCRAMBLE_DELAYED_REFS |
| 3078 | delayed_refs->run_delayed_start = find_middle(&delayed_refs->root); |
| 3079 | #endif |
| 3080 | trans->can_flush_pending_bgs = false; |
| 3081 | ret = __btrfs_run_delayed_refs(trans, count); |
| 3082 | if (ret < 0) { |
| 3083 | btrfs_abort_transaction(trans, ret); |
| 3084 | return ret; |
| 3085 | } |
| 3086 | |
| 3087 | if (run_all) { |
| 3088 | if (!list_empty(&trans->new_bgs)) |
| 3089 | btrfs_create_pending_block_groups(trans); |
| 3090 | |
| 3091 | spin_lock(&delayed_refs->lock); |
| 3092 | node = rb_first(&delayed_refs->href_root); |
| 3093 | if (!node) { |
| 3094 | spin_unlock(&delayed_refs->lock); |
| 3095 | goto out; |
| 3096 | } |
| 3097 | head = rb_entry(node, struct btrfs_delayed_ref_head, |
| 3098 | href_node); |
| 3099 | refcount_inc(&head->refs); |
| 3100 | spin_unlock(&delayed_refs->lock); |
| 3101 | |
| 3102 | /* Mutex was contended, block until it's released and retry. */ |
| 3103 | mutex_lock(&head->mutex); |
| 3104 | mutex_unlock(&head->mutex); |
| 3105 | |
| 3106 | btrfs_put_delayed_ref_head(head); |
| 3107 | cond_resched(); |
| 3108 | goto again; |
| 3109 | } |
| 3110 | out: |
| 3111 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
| 3112 | return 0; |
| 3113 | } |
| 3114 | |
| 3115 | int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans, |
| 3116 | struct btrfs_fs_info *fs_info, |
| 3117 | u64 bytenr, u64 num_bytes, u64 flags, |
| 3118 | int level, int is_data) |
| 3119 | { |
| 3120 | struct btrfs_delayed_extent_op *extent_op; |
| 3121 | int ret; |
| 3122 | |
| 3123 | extent_op = btrfs_alloc_delayed_extent_op(); |
| 3124 | if (!extent_op) |
| 3125 | return -ENOMEM; |
| 3126 | |
| 3127 | extent_op->flags_to_set = flags; |
| 3128 | extent_op->update_flags = true; |
| 3129 | extent_op->update_key = false; |
| 3130 | extent_op->is_data = is_data ? true : false; |
| 3131 | extent_op->level = level; |
| 3132 | |
| 3133 | ret = btrfs_add_delayed_extent_op(fs_info, trans, bytenr, |
| 3134 | num_bytes, extent_op); |
| 3135 | if (ret) |
| 3136 | btrfs_free_delayed_extent_op(extent_op); |
| 3137 | return ret; |
| 3138 | } |
| 3139 | |
| 3140 | static noinline int check_delayed_ref(struct btrfs_root *root, |
| 3141 | struct btrfs_path *path, |
| 3142 | u64 objectid, u64 offset, u64 bytenr) |
| 3143 | { |
| 3144 | struct btrfs_delayed_ref_head *head; |
| 3145 | struct btrfs_delayed_ref_node *ref; |
| 3146 | struct btrfs_delayed_data_ref *data_ref; |
| 3147 | struct btrfs_delayed_ref_root *delayed_refs; |
| 3148 | struct btrfs_transaction *cur_trans; |
| 3149 | struct rb_node *node; |
| 3150 | int ret = 0; |
| 3151 | |
| 3152 | spin_lock(&root->fs_info->trans_lock); |
| 3153 | cur_trans = root->fs_info->running_transaction; |
| 3154 | if (cur_trans) |
| 3155 | refcount_inc(&cur_trans->use_count); |
| 3156 | spin_unlock(&root->fs_info->trans_lock); |
| 3157 | if (!cur_trans) |
| 3158 | return 0; |
| 3159 | |
| 3160 | delayed_refs = &cur_trans->delayed_refs; |
| 3161 | spin_lock(&delayed_refs->lock); |
| 3162 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
| 3163 | if (!head) { |
| 3164 | spin_unlock(&delayed_refs->lock); |
| 3165 | btrfs_put_transaction(cur_trans); |
| 3166 | return 0; |
| 3167 | } |
| 3168 | |
| 3169 | if (!mutex_trylock(&head->mutex)) { |
| 3170 | refcount_inc(&head->refs); |
| 3171 | spin_unlock(&delayed_refs->lock); |
| 3172 | |
| 3173 | btrfs_release_path(path); |
| 3174 | |
| 3175 | /* |
| 3176 | * Mutex was contended, block until it's released and let |
| 3177 | * caller try again |
| 3178 | */ |
| 3179 | mutex_lock(&head->mutex); |
| 3180 | mutex_unlock(&head->mutex); |
| 3181 | btrfs_put_delayed_ref_head(head); |
| 3182 | btrfs_put_transaction(cur_trans); |
| 3183 | return -EAGAIN; |
| 3184 | } |
| 3185 | spin_unlock(&delayed_refs->lock); |
| 3186 | |
| 3187 | spin_lock(&head->lock); |
| 3188 | /* |
| 3189 | * XXX: We should replace this with a proper search function in the |
| 3190 | * future. |
| 3191 | */ |
| 3192 | for (node = rb_first(&head->ref_tree); node; node = rb_next(node)) { |
| 3193 | ref = rb_entry(node, struct btrfs_delayed_ref_node, ref_node); |
| 3194 | /* If it's a shared ref we know a cross reference exists */ |
| 3195 | if (ref->type != BTRFS_EXTENT_DATA_REF_KEY) { |
| 3196 | ret = 1; |
| 3197 | break; |
| 3198 | } |
| 3199 | |
| 3200 | data_ref = btrfs_delayed_node_to_data_ref(ref); |
| 3201 | |
| 3202 | /* |
| 3203 | * If our ref doesn't match the one we're currently looking at |
| 3204 | * then we have a cross reference. |
| 3205 | */ |
| 3206 | if (data_ref->root != root->root_key.objectid || |
| 3207 | data_ref->objectid != objectid || |
| 3208 | data_ref->offset != offset) { |
| 3209 | ret = 1; |
| 3210 | break; |
| 3211 | } |
| 3212 | } |
| 3213 | spin_unlock(&head->lock); |
| 3214 | mutex_unlock(&head->mutex); |
| 3215 | btrfs_put_transaction(cur_trans); |
| 3216 | return ret; |
| 3217 | } |
| 3218 | |
| 3219 | static noinline int check_committed_ref(struct btrfs_root *root, |
| 3220 | struct btrfs_path *path, |
| 3221 | u64 objectid, u64 offset, u64 bytenr) |
| 3222 | { |
| 3223 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 3224 | struct btrfs_root *extent_root = fs_info->extent_root; |
| 3225 | struct extent_buffer *leaf; |
| 3226 | struct btrfs_extent_data_ref *ref; |
| 3227 | struct btrfs_extent_inline_ref *iref; |
| 3228 | struct btrfs_extent_item *ei; |
| 3229 | struct btrfs_key key; |
| 3230 | u32 item_size; |
| 3231 | int type; |
| 3232 | int ret; |
| 3233 | |
| 3234 | key.objectid = bytenr; |
| 3235 | key.offset = (u64)-1; |
| 3236 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 3237 | |
| 3238 | ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0); |
| 3239 | if (ret < 0) |
| 3240 | goto out; |
| 3241 | BUG_ON(ret == 0); /* Corruption */ |
| 3242 | |
| 3243 | ret = -ENOENT; |
| 3244 | if (path->slots[0] == 0) |
| 3245 | goto out; |
| 3246 | |
| 3247 | path->slots[0]--; |
| 3248 | leaf = path->nodes[0]; |
| 3249 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); |
| 3250 | |
| 3251 | if (key.objectid != bytenr || key.type != BTRFS_EXTENT_ITEM_KEY) |
| 3252 | goto out; |
| 3253 | |
| 3254 | ret = 1; |
| 3255 | item_size = btrfs_item_size_nr(leaf, path->slots[0]); |
| 3256 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 3257 | if (item_size < sizeof(*ei)) { |
| 3258 | WARN_ON(item_size != sizeof(struct btrfs_extent_item_v0)); |
| 3259 | goto out; |
| 3260 | } |
| 3261 | #endif |
| 3262 | ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item); |
| 3263 | |
| 3264 | if (item_size != sizeof(*ei) + |
| 3265 | btrfs_extent_inline_ref_size(BTRFS_EXTENT_DATA_REF_KEY)) |
| 3266 | goto out; |
| 3267 | |
| 3268 | if (btrfs_extent_generation(leaf, ei) <= |
| 3269 | btrfs_root_last_snapshot(&root->root_item)) |
| 3270 | goto out; |
| 3271 | |
| 3272 | iref = (struct btrfs_extent_inline_ref *)(ei + 1); |
| 3273 | |
| 3274 | type = btrfs_get_extent_inline_ref_type(leaf, iref, BTRFS_REF_TYPE_DATA); |
| 3275 | if (type != BTRFS_EXTENT_DATA_REF_KEY) |
| 3276 | goto out; |
| 3277 | |
| 3278 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); |
| 3279 | if (btrfs_extent_refs(leaf, ei) != |
| 3280 | btrfs_extent_data_ref_count(leaf, ref) || |
| 3281 | btrfs_extent_data_ref_root(leaf, ref) != |
| 3282 | root->root_key.objectid || |
| 3283 | btrfs_extent_data_ref_objectid(leaf, ref) != objectid || |
| 3284 | btrfs_extent_data_ref_offset(leaf, ref) != offset) |
| 3285 | goto out; |
| 3286 | |
| 3287 | ret = 0; |
| 3288 | out: |
| 3289 | return ret; |
| 3290 | } |
| 3291 | |
| 3292 | int btrfs_cross_ref_exist(struct btrfs_root *root, u64 objectid, u64 offset, |
| 3293 | u64 bytenr) |
| 3294 | { |
| 3295 | struct btrfs_path *path; |
| 3296 | int ret; |
| 3297 | int ret2; |
| 3298 | |
| 3299 | path = btrfs_alloc_path(); |
| 3300 | if (!path) |
| 3301 | return -ENOMEM; |
| 3302 | |
| 3303 | do { |
| 3304 | ret = check_committed_ref(root, path, objectid, |
| 3305 | offset, bytenr); |
| 3306 | if (ret && ret != -ENOENT) |
| 3307 | goto out; |
| 3308 | |
| 3309 | ret2 = check_delayed_ref(root, path, objectid, |
| 3310 | offset, bytenr); |
| 3311 | } while (ret2 == -EAGAIN); |
| 3312 | |
| 3313 | if (ret2 && ret2 != -ENOENT) { |
| 3314 | ret = ret2; |
| 3315 | goto out; |
| 3316 | } |
| 3317 | |
| 3318 | if (ret != -ENOENT || ret2 != -ENOENT) |
| 3319 | ret = 0; |
| 3320 | out: |
| 3321 | btrfs_free_path(path); |
| 3322 | if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID) |
| 3323 | WARN_ON(ret > 0); |
| 3324 | return ret; |
| 3325 | } |
| 3326 | |
| 3327 | static int __btrfs_mod_ref(struct btrfs_trans_handle *trans, |
| 3328 | struct btrfs_root *root, |
| 3329 | struct extent_buffer *buf, |
| 3330 | int full_backref, int inc) |
| 3331 | { |
| 3332 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 3333 | u64 bytenr; |
| 3334 | u64 num_bytes; |
| 3335 | u64 parent; |
| 3336 | u64 ref_root; |
| 3337 | u32 nritems; |
| 3338 | struct btrfs_key key; |
| 3339 | struct btrfs_file_extent_item *fi; |
| 3340 | int i; |
| 3341 | int level; |
| 3342 | int ret = 0; |
| 3343 | int (*process_func)(struct btrfs_trans_handle *, |
| 3344 | struct btrfs_root *, |
| 3345 | u64, u64, u64, u64, u64, u64); |
| 3346 | |
| 3347 | |
| 3348 | if (btrfs_is_testing(fs_info)) |
| 3349 | return 0; |
| 3350 | |
| 3351 | ref_root = btrfs_header_owner(buf); |
| 3352 | nritems = btrfs_header_nritems(buf); |
| 3353 | level = btrfs_header_level(buf); |
| 3354 | |
| 3355 | if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state) && level == 0) |
| 3356 | return 0; |
| 3357 | |
| 3358 | if (inc) |
| 3359 | process_func = btrfs_inc_extent_ref; |
| 3360 | else |
| 3361 | process_func = btrfs_free_extent; |
| 3362 | |
| 3363 | if (full_backref) |
| 3364 | parent = buf->start; |
| 3365 | else |
| 3366 | parent = 0; |
| 3367 | |
| 3368 | for (i = 0; i < nritems; i++) { |
| 3369 | if (level == 0) { |
| 3370 | btrfs_item_key_to_cpu(buf, &key, i); |
| 3371 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
| 3372 | continue; |
| 3373 | fi = btrfs_item_ptr(buf, i, |
| 3374 | struct btrfs_file_extent_item); |
| 3375 | if (btrfs_file_extent_type(buf, fi) == |
| 3376 | BTRFS_FILE_EXTENT_INLINE) |
| 3377 | continue; |
| 3378 | bytenr = btrfs_file_extent_disk_bytenr(buf, fi); |
| 3379 | if (bytenr == 0) |
| 3380 | continue; |
| 3381 | |
| 3382 | num_bytes = btrfs_file_extent_disk_num_bytes(buf, fi); |
| 3383 | key.offset -= btrfs_file_extent_offset(buf, fi); |
| 3384 | ret = process_func(trans, root, bytenr, num_bytes, |
| 3385 | parent, ref_root, key.objectid, |
| 3386 | key.offset); |
| 3387 | if (ret) |
| 3388 | goto fail; |
| 3389 | } else { |
| 3390 | bytenr = btrfs_node_blockptr(buf, i); |
| 3391 | num_bytes = fs_info->nodesize; |
| 3392 | ret = process_func(trans, root, bytenr, num_bytes, |
| 3393 | parent, ref_root, level - 1, 0); |
| 3394 | if (ret) |
| 3395 | goto fail; |
| 3396 | } |
| 3397 | } |
| 3398 | return 0; |
| 3399 | fail: |
| 3400 | return ret; |
| 3401 | } |
| 3402 | |
| 3403 | int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
| 3404 | struct extent_buffer *buf, int full_backref) |
| 3405 | { |
| 3406 | return __btrfs_mod_ref(trans, root, buf, full_backref, 1); |
| 3407 | } |
| 3408 | |
| 3409 | int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
| 3410 | struct extent_buffer *buf, int full_backref) |
| 3411 | { |
| 3412 | return __btrfs_mod_ref(trans, root, buf, full_backref, 0); |
| 3413 | } |
| 3414 | |
| 3415 | static int write_one_cache_group(struct btrfs_trans_handle *trans, |
| 3416 | struct btrfs_fs_info *fs_info, |
| 3417 | struct btrfs_path *path, |
| 3418 | struct btrfs_block_group_cache *cache) |
| 3419 | { |
| 3420 | int ret; |
| 3421 | struct btrfs_root *extent_root = fs_info->extent_root; |
| 3422 | unsigned long bi; |
| 3423 | struct extent_buffer *leaf; |
| 3424 | |
| 3425 | ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1); |
| 3426 | if (ret) { |
| 3427 | if (ret > 0) |
| 3428 | ret = -ENOENT; |
| 3429 | goto fail; |
| 3430 | } |
| 3431 | |
| 3432 | leaf = path->nodes[0]; |
| 3433 | bi = btrfs_item_ptr_offset(leaf, path->slots[0]); |
| 3434 | write_extent_buffer(leaf, &cache->item, bi, sizeof(cache->item)); |
| 3435 | btrfs_mark_buffer_dirty(leaf); |
| 3436 | fail: |
| 3437 | btrfs_release_path(path); |
| 3438 | return ret; |
| 3439 | |
| 3440 | } |
| 3441 | |
| 3442 | static struct btrfs_block_group_cache * |
| 3443 | next_block_group(struct btrfs_fs_info *fs_info, |
| 3444 | struct btrfs_block_group_cache *cache) |
| 3445 | { |
| 3446 | struct rb_node *node; |
| 3447 | |
| 3448 | spin_lock(&fs_info->block_group_cache_lock); |
| 3449 | |
| 3450 | /* If our block group was removed, we need a full search. */ |
| 3451 | if (RB_EMPTY_NODE(&cache->cache_node)) { |
| 3452 | const u64 next_bytenr = cache->key.objectid + cache->key.offset; |
| 3453 | |
| 3454 | spin_unlock(&fs_info->block_group_cache_lock); |
| 3455 | btrfs_put_block_group(cache); |
| 3456 | cache = btrfs_lookup_first_block_group(fs_info, next_bytenr); return cache; |
| 3457 | } |
| 3458 | node = rb_next(&cache->cache_node); |
| 3459 | btrfs_put_block_group(cache); |
| 3460 | if (node) { |
| 3461 | cache = rb_entry(node, struct btrfs_block_group_cache, |
| 3462 | cache_node); |
| 3463 | btrfs_get_block_group(cache); |
| 3464 | } else |
| 3465 | cache = NULL; |
| 3466 | spin_unlock(&fs_info->block_group_cache_lock); |
| 3467 | return cache; |
| 3468 | } |
| 3469 | |
| 3470 | static int cache_save_setup(struct btrfs_block_group_cache *block_group, |
| 3471 | struct btrfs_trans_handle *trans, |
| 3472 | struct btrfs_path *path) |
| 3473 | { |
| 3474 | struct btrfs_fs_info *fs_info = block_group->fs_info; |
| 3475 | struct btrfs_root *root = fs_info->tree_root; |
| 3476 | struct inode *inode = NULL; |
| 3477 | struct extent_changeset *data_reserved = NULL; |
| 3478 | u64 alloc_hint = 0; |
| 3479 | int dcs = BTRFS_DC_ERROR; |
| 3480 | u64 num_pages = 0; |
| 3481 | int retries = 0; |
| 3482 | int ret = 0; |
| 3483 | |
| 3484 | /* |
| 3485 | * If this block group is smaller than 100 megs don't bother caching the |
| 3486 | * block group. |
| 3487 | */ |
| 3488 | if (block_group->key.offset < (100 * SZ_1M)) { |
| 3489 | spin_lock(&block_group->lock); |
| 3490 | block_group->disk_cache_state = BTRFS_DC_WRITTEN; |
| 3491 | spin_unlock(&block_group->lock); |
| 3492 | return 0; |
| 3493 | } |
| 3494 | |
| 3495 | if (trans->aborted) |
| 3496 | return 0; |
| 3497 | again: |
| 3498 | inode = lookup_free_space_inode(fs_info, block_group, path); |
| 3499 | if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) { |
| 3500 | ret = PTR_ERR(inode); |
| 3501 | btrfs_release_path(path); |
| 3502 | goto out; |
| 3503 | } |
| 3504 | |
| 3505 | if (IS_ERR(inode)) { |
| 3506 | BUG_ON(retries); |
| 3507 | retries++; |
| 3508 | |
| 3509 | if (block_group->ro) |
| 3510 | goto out_free; |
| 3511 | |
| 3512 | ret = create_free_space_inode(fs_info, trans, block_group, |
| 3513 | path); |
| 3514 | if (ret) |
| 3515 | goto out_free; |
| 3516 | goto again; |
| 3517 | } |
| 3518 | |
| 3519 | /* |
| 3520 | * We want to set the generation to 0, that way if anything goes wrong |
| 3521 | * from here on out we know not to trust this cache when we load up next |
| 3522 | * time. |
| 3523 | */ |
| 3524 | BTRFS_I(inode)->generation = 0; |
| 3525 | ret = btrfs_update_inode(trans, root, inode); |
| 3526 | if (ret) { |
| 3527 | /* |
| 3528 | * So theoretically we could recover from this, simply set the |
| 3529 | * super cache generation to 0 so we know to invalidate the |
| 3530 | * cache, but then we'd have to keep track of the block groups |
| 3531 | * that fail this way so we know we _have_ to reset this cache |
| 3532 | * before the next commit or risk reading stale cache. So to |
| 3533 | * limit our exposure to horrible edge cases lets just abort the |
| 3534 | * transaction, this only happens in really bad situations |
| 3535 | * anyway. |
| 3536 | */ |
| 3537 | btrfs_abort_transaction(trans, ret); |
| 3538 | goto out_put; |
| 3539 | } |
| 3540 | WARN_ON(ret); |
| 3541 | |
| 3542 | /* We've already setup this transaction, go ahead and exit */ |
| 3543 | if (block_group->cache_generation == trans->transid && |
| 3544 | i_size_read(inode)) { |
| 3545 | dcs = BTRFS_DC_SETUP; |
| 3546 | goto out_put; |
| 3547 | } |
| 3548 | |
| 3549 | if (i_size_read(inode) > 0) { |
| 3550 | ret = btrfs_check_trunc_cache_free_space(fs_info, |
| 3551 | &fs_info->global_block_rsv); |
| 3552 | if (ret) |
| 3553 | goto out_put; |
| 3554 | |
| 3555 | ret = btrfs_truncate_free_space_cache(trans, NULL, inode); |
| 3556 | if (ret) |
| 3557 | goto out_put; |
| 3558 | } |
| 3559 | |
| 3560 | spin_lock(&block_group->lock); |
| 3561 | if (block_group->cached != BTRFS_CACHE_FINISHED || |
| 3562 | !btrfs_test_opt(fs_info, SPACE_CACHE)) { |
| 3563 | /* |
| 3564 | * don't bother trying to write stuff out _if_ |
| 3565 | * a) we're not cached, |
| 3566 | * b) we're with nospace_cache mount option, |
| 3567 | * c) we're with v2 space_cache (FREE_SPACE_TREE). |
| 3568 | */ |
| 3569 | dcs = BTRFS_DC_WRITTEN; |
| 3570 | spin_unlock(&block_group->lock); |
| 3571 | goto out_put; |
| 3572 | } |
| 3573 | spin_unlock(&block_group->lock); |
| 3574 | |
| 3575 | /* |
| 3576 | * We hit an ENOSPC when setting up the cache in this transaction, just |
| 3577 | * skip doing the setup, we've already cleared the cache so we're safe. |
| 3578 | */ |
| 3579 | if (test_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags)) { |
| 3580 | ret = -ENOSPC; |
| 3581 | goto out_put; |
| 3582 | } |
| 3583 | |
| 3584 | /* |
| 3585 | * Try to preallocate enough space based on how big the block group is. |
| 3586 | * Keep in mind this has to include any pinned space which could end up |
| 3587 | * taking up quite a bit since it's not folded into the other space |
| 3588 | * cache. |
| 3589 | */ |
| 3590 | num_pages = div_u64(block_group->key.offset, SZ_256M); |
| 3591 | if (!num_pages) |
| 3592 | num_pages = 1; |
| 3593 | |
| 3594 | num_pages *= 16; |
| 3595 | num_pages *= PAGE_SIZE; |
| 3596 | |
| 3597 | ret = btrfs_check_data_free_space(inode, &data_reserved, 0, num_pages); |
| 3598 | if (ret) |
| 3599 | goto out_put; |
| 3600 | |
| 3601 | ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages, |
| 3602 | num_pages, num_pages, |
| 3603 | &alloc_hint); |
| 3604 | /* |
| 3605 | * Our cache requires contiguous chunks so that we don't modify a bunch |
| 3606 | * of metadata or split extents when writing the cache out, which means |
| 3607 | * we can enospc if we are heavily fragmented in addition to just normal |
| 3608 | * out of space conditions. So if we hit this just skip setting up any |
| 3609 | * other block groups for this transaction, maybe we'll unpin enough |
| 3610 | * space the next time around. |
| 3611 | */ |
| 3612 | if (!ret) |
| 3613 | dcs = BTRFS_DC_SETUP; |
| 3614 | else if (ret == -ENOSPC) |
| 3615 | set_bit(BTRFS_TRANS_CACHE_ENOSPC, &trans->transaction->flags); |
| 3616 | |
| 3617 | out_put: |
| 3618 | iput(inode); |
| 3619 | out_free: |
| 3620 | btrfs_release_path(path); |
| 3621 | out: |
| 3622 | spin_lock(&block_group->lock); |
| 3623 | if (!ret && dcs == BTRFS_DC_SETUP) |
| 3624 | block_group->cache_generation = trans->transid; |
| 3625 | block_group->disk_cache_state = dcs; |
| 3626 | spin_unlock(&block_group->lock); |
| 3627 | |
| 3628 | extent_changeset_free(data_reserved); |
| 3629 | return ret; |
| 3630 | } |
| 3631 | |
| 3632 | int btrfs_setup_space_cache(struct btrfs_trans_handle *trans, |
| 3633 | struct btrfs_fs_info *fs_info) |
| 3634 | { |
| 3635 | struct btrfs_block_group_cache *cache, *tmp; |
| 3636 | struct btrfs_transaction *cur_trans = trans->transaction; |
| 3637 | struct btrfs_path *path; |
| 3638 | |
| 3639 | if (list_empty(&cur_trans->dirty_bgs) || |
| 3640 | !btrfs_test_opt(fs_info, SPACE_CACHE)) |
| 3641 | return 0; |
| 3642 | |
| 3643 | path = btrfs_alloc_path(); |
| 3644 | if (!path) |
| 3645 | return -ENOMEM; |
| 3646 | |
| 3647 | /* Could add new block groups, use _safe just in case */ |
| 3648 | list_for_each_entry_safe(cache, tmp, &cur_trans->dirty_bgs, |
| 3649 | dirty_list) { |
| 3650 | if (cache->disk_cache_state == BTRFS_DC_CLEAR) |
| 3651 | cache_save_setup(cache, trans, path); |
| 3652 | } |
| 3653 | |
| 3654 | btrfs_free_path(path); |
| 3655 | return 0; |
| 3656 | } |
| 3657 | |
| 3658 | /* |
| 3659 | * transaction commit does final block group cache writeback during a |
| 3660 | * critical section where nothing is allowed to change the FS. This is |
| 3661 | * required in order for the cache to actually match the block group, |
| 3662 | * but can introduce a lot of latency into the commit. |
| 3663 | * |
| 3664 | * So, btrfs_start_dirty_block_groups is here to kick off block group |
| 3665 | * cache IO. There's a chance we'll have to redo some of it if the |
| 3666 | * block group changes again during the commit, but it greatly reduces |
| 3667 | * the commit latency by getting rid of the easy block groups while |
| 3668 | * we're still allowing others to join the commit. |
| 3669 | */ |
| 3670 | int btrfs_start_dirty_block_groups(struct btrfs_trans_handle *trans) |
| 3671 | { |
| 3672 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 3673 | struct btrfs_block_group_cache *cache; |
| 3674 | struct btrfs_transaction *cur_trans = trans->transaction; |
| 3675 | int ret = 0; |
| 3676 | int should_put; |
| 3677 | struct btrfs_path *path = NULL; |
| 3678 | LIST_HEAD(dirty); |
| 3679 | struct list_head *io = &cur_trans->io_bgs; |
| 3680 | int num_started = 0; |
| 3681 | int loops = 0; |
| 3682 | |
| 3683 | spin_lock(&cur_trans->dirty_bgs_lock); |
| 3684 | if (list_empty(&cur_trans->dirty_bgs)) { |
| 3685 | spin_unlock(&cur_trans->dirty_bgs_lock); |
| 3686 | return 0; |
| 3687 | } |
| 3688 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
| 3689 | spin_unlock(&cur_trans->dirty_bgs_lock); |
| 3690 | |
| 3691 | again: |
| 3692 | /* |
| 3693 | * make sure all the block groups on our dirty list actually |
| 3694 | * exist |
| 3695 | */ |
| 3696 | btrfs_create_pending_block_groups(trans); |
| 3697 | |
| 3698 | if (!path) { |
| 3699 | path = btrfs_alloc_path(); |
| 3700 | if (!path) |
| 3701 | return -ENOMEM; |
| 3702 | } |
| 3703 | |
| 3704 | /* |
| 3705 | * cache_write_mutex is here only to save us from balance or automatic |
| 3706 | * removal of empty block groups deleting this block group while we are |
| 3707 | * writing out the cache |
| 3708 | */ |
| 3709 | mutex_lock(&trans->transaction->cache_write_mutex); |
| 3710 | while (!list_empty(&dirty)) { |
| 3711 | cache = list_first_entry(&dirty, |
| 3712 | struct btrfs_block_group_cache, |
| 3713 | dirty_list); |
| 3714 | /* |
| 3715 | * this can happen if something re-dirties a block |
| 3716 | * group that is already under IO. Just wait for it to |
| 3717 | * finish and then do it all again |
| 3718 | */ |
| 3719 | if (!list_empty(&cache->io_list)) { |
| 3720 | list_del_init(&cache->io_list); |
| 3721 | btrfs_wait_cache_io(trans, cache, path); |
| 3722 | btrfs_put_block_group(cache); |
| 3723 | } |
| 3724 | |
| 3725 | |
| 3726 | /* |
| 3727 | * btrfs_wait_cache_io uses the cache->dirty_list to decide |
| 3728 | * if it should update the cache_state. Don't delete |
| 3729 | * until after we wait. |
| 3730 | * |
| 3731 | * Since we're not running in the commit critical section |
| 3732 | * we need the dirty_bgs_lock to protect from update_block_group |
| 3733 | */ |
| 3734 | spin_lock(&cur_trans->dirty_bgs_lock); |
| 3735 | list_del_init(&cache->dirty_list); |
| 3736 | spin_unlock(&cur_trans->dirty_bgs_lock); |
| 3737 | |
| 3738 | should_put = 1; |
| 3739 | |
| 3740 | cache_save_setup(cache, trans, path); |
| 3741 | |
| 3742 | if (cache->disk_cache_state == BTRFS_DC_SETUP) { |
| 3743 | cache->io_ctl.inode = NULL; |
| 3744 | ret = btrfs_write_out_cache(fs_info, trans, |
| 3745 | cache, path); |
| 3746 | if (ret == 0 && cache->io_ctl.inode) { |
| 3747 | num_started++; |
| 3748 | should_put = 0; |
| 3749 | |
| 3750 | /* |
| 3751 | * The cache_write_mutex is protecting the |
| 3752 | * io_list, also refer to the definition of |
| 3753 | * btrfs_transaction::io_bgs for more details |
| 3754 | */ |
| 3755 | list_add_tail(&cache->io_list, io); |
| 3756 | } else { |
| 3757 | /* |
| 3758 | * if we failed to write the cache, the |
| 3759 | * generation will be bad and life goes on |
| 3760 | */ |
| 3761 | ret = 0; |
| 3762 | } |
| 3763 | } |
| 3764 | if (!ret) { |
| 3765 | ret = write_one_cache_group(trans, fs_info, |
| 3766 | path, cache); |
| 3767 | /* |
| 3768 | * Our block group might still be attached to the list |
| 3769 | * of new block groups in the transaction handle of some |
| 3770 | * other task (struct btrfs_trans_handle->new_bgs). This |
| 3771 | * means its block group item isn't yet in the extent |
| 3772 | * tree. If this happens ignore the error, as we will |
| 3773 | * try again later in the critical section of the |
| 3774 | * transaction commit. |
| 3775 | */ |
| 3776 | if (ret == -ENOENT) { |
| 3777 | ret = 0; |
| 3778 | spin_lock(&cur_trans->dirty_bgs_lock); |
| 3779 | if (list_empty(&cache->dirty_list)) { |
| 3780 | list_add_tail(&cache->dirty_list, |
| 3781 | &cur_trans->dirty_bgs); |
| 3782 | btrfs_get_block_group(cache); |
| 3783 | } |
| 3784 | spin_unlock(&cur_trans->dirty_bgs_lock); |
| 3785 | } else if (ret) { |
| 3786 | btrfs_abort_transaction(trans, ret); |
| 3787 | } |
| 3788 | } |
| 3789 | |
| 3790 | /* if its not on the io list, we need to put the block group */ |
| 3791 | if (should_put) |
| 3792 | btrfs_put_block_group(cache); |
| 3793 | |
| 3794 | if (ret) |
| 3795 | break; |
| 3796 | |
| 3797 | /* |
| 3798 | * Avoid blocking other tasks for too long. It might even save |
| 3799 | * us from writing caches for block groups that are going to be |
| 3800 | * removed. |
| 3801 | */ |
| 3802 | mutex_unlock(&trans->transaction->cache_write_mutex); |
| 3803 | mutex_lock(&trans->transaction->cache_write_mutex); |
| 3804 | } |
| 3805 | mutex_unlock(&trans->transaction->cache_write_mutex); |
| 3806 | |
| 3807 | /* |
| 3808 | * go through delayed refs for all the stuff we've just kicked off |
| 3809 | * and then loop back (just once) |
| 3810 | */ |
| 3811 | ret = btrfs_run_delayed_refs(trans, 0); |
| 3812 | if (!ret && loops == 0) { |
| 3813 | loops++; |
| 3814 | spin_lock(&cur_trans->dirty_bgs_lock); |
| 3815 | list_splice_init(&cur_trans->dirty_bgs, &dirty); |
| 3816 | /* |
| 3817 | * dirty_bgs_lock protects us from concurrent block group |
| 3818 | * deletes too (not just cache_write_mutex). |
| 3819 | */ |
| 3820 | if (!list_empty(&dirty)) { |
| 3821 | spin_unlock(&cur_trans->dirty_bgs_lock); |
| 3822 | goto again; |
| 3823 | } |
| 3824 | spin_unlock(&cur_trans->dirty_bgs_lock); |
| 3825 | } else if (ret < 0) { |
| 3826 | btrfs_cleanup_dirty_bgs(cur_trans, fs_info); |
| 3827 | } |
| 3828 | |
| 3829 | btrfs_free_path(path); |
| 3830 | return ret; |
| 3831 | } |
| 3832 | |
| 3833 | int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans, |
| 3834 | struct btrfs_fs_info *fs_info) |
| 3835 | { |
| 3836 | struct btrfs_block_group_cache *cache; |
| 3837 | struct btrfs_transaction *cur_trans = trans->transaction; |
| 3838 | int ret = 0; |
| 3839 | int should_put; |
| 3840 | struct btrfs_path *path; |
| 3841 | struct list_head *io = &cur_trans->io_bgs; |
| 3842 | int num_started = 0; |
| 3843 | |
| 3844 | path = btrfs_alloc_path(); |
| 3845 | if (!path) |
| 3846 | return -ENOMEM; |
| 3847 | |
| 3848 | /* |
| 3849 | * Even though we are in the critical section of the transaction commit, |
| 3850 | * we can still have concurrent tasks adding elements to this |
| 3851 | * transaction's list of dirty block groups. These tasks correspond to |
| 3852 | * endio free space workers started when writeback finishes for a |
| 3853 | * space cache, which run inode.c:btrfs_finish_ordered_io(), and can |
| 3854 | * allocate new block groups as a result of COWing nodes of the root |
| 3855 | * tree when updating the free space inode. The writeback for the space |
| 3856 | * caches is triggered by an earlier call to |
| 3857 | * btrfs_start_dirty_block_groups() and iterations of the following |
| 3858 | * loop. |
| 3859 | * Also we want to do the cache_save_setup first and then run the |
| 3860 | * delayed refs to make sure we have the best chance at doing this all |
| 3861 | * in one shot. |
| 3862 | */ |
| 3863 | spin_lock(&cur_trans->dirty_bgs_lock); |
| 3864 | while (!list_empty(&cur_trans->dirty_bgs)) { |
| 3865 | cache = list_first_entry(&cur_trans->dirty_bgs, |
| 3866 | struct btrfs_block_group_cache, |
| 3867 | dirty_list); |
| 3868 | |
| 3869 | /* |
| 3870 | * this can happen if cache_save_setup re-dirties a block |
| 3871 | * group that is already under IO. Just wait for it to |
| 3872 | * finish and then do it all again |
| 3873 | */ |
| 3874 | if (!list_empty(&cache->io_list)) { |
| 3875 | spin_unlock(&cur_trans->dirty_bgs_lock); |
| 3876 | list_del_init(&cache->io_list); |
| 3877 | btrfs_wait_cache_io(trans, cache, path); |
| 3878 | btrfs_put_block_group(cache); |
| 3879 | spin_lock(&cur_trans->dirty_bgs_lock); |
| 3880 | } |
| 3881 | |
| 3882 | /* |
| 3883 | * don't remove from the dirty list until after we've waited |
| 3884 | * on any pending IO |
| 3885 | */ |
| 3886 | list_del_init(&cache->dirty_list); |
| 3887 | spin_unlock(&cur_trans->dirty_bgs_lock); |
| 3888 | should_put = 1; |
| 3889 | |
| 3890 | cache_save_setup(cache, trans, path); |
| 3891 | |
| 3892 | if (!ret) |
| 3893 | ret = btrfs_run_delayed_refs(trans, |
| 3894 | (unsigned long) -1); |
| 3895 | |
| 3896 | if (!ret && cache->disk_cache_state == BTRFS_DC_SETUP) { |
| 3897 | cache->io_ctl.inode = NULL; |
| 3898 | ret = btrfs_write_out_cache(fs_info, trans, |
| 3899 | cache, path); |
| 3900 | if (ret == 0 && cache->io_ctl.inode) { |
| 3901 | num_started++; |
| 3902 | should_put = 0; |
| 3903 | list_add_tail(&cache->io_list, io); |
| 3904 | } else { |
| 3905 | /* |
| 3906 | * if we failed to write the cache, the |
| 3907 | * generation will be bad and life goes on |
| 3908 | */ |
| 3909 | ret = 0; |
| 3910 | } |
| 3911 | } |
| 3912 | if (!ret) { |
| 3913 | ret = write_one_cache_group(trans, fs_info, |
| 3914 | path, cache); |
| 3915 | /* |
| 3916 | * One of the free space endio workers might have |
| 3917 | * created a new block group while updating a free space |
| 3918 | * cache's inode (at inode.c:btrfs_finish_ordered_io()) |
| 3919 | * and hasn't released its transaction handle yet, in |
| 3920 | * which case the new block group is still attached to |
| 3921 | * its transaction handle and its creation has not |
| 3922 | * finished yet (no block group item in the extent tree |
| 3923 | * yet, etc). If this is the case, wait for all free |
| 3924 | * space endio workers to finish and retry. This is a |
| 3925 | * a very rare case so no need for a more efficient and |
| 3926 | * complex approach. |
| 3927 | */ |
| 3928 | if (ret == -ENOENT) { |
| 3929 | wait_event(cur_trans->writer_wait, |
| 3930 | atomic_read(&cur_trans->num_writers) == 1); |
| 3931 | ret = write_one_cache_group(trans, fs_info, |
| 3932 | path, cache); |
| 3933 | } |
| 3934 | if (ret) |
| 3935 | btrfs_abort_transaction(trans, ret); |
| 3936 | } |
| 3937 | |
| 3938 | /* if its not on the io list, we need to put the block group */ |
| 3939 | if (should_put) |
| 3940 | btrfs_put_block_group(cache); |
| 3941 | spin_lock(&cur_trans->dirty_bgs_lock); |
| 3942 | } |
| 3943 | spin_unlock(&cur_trans->dirty_bgs_lock); |
| 3944 | |
| 3945 | /* |
| 3946 | * Refer to the definition of io_bgs member for details why it's safe |
| 3947 | * to use it without any locking |
| 3948 | */ |
| 3949 | while (!list_empty(io)) { |
| 3950 | cache = list_first_entry(io, struct btrfs_block_group_cache, |
| 3951 | io_list); |
| 3952 | list_del_init(&cache->io_list); |
| 3953 | btrfs_wait_cache_io(trans, cache, path); |
| 3954 | btrfs_put_block_group(cache); |
| 3955 | } |
| 3956 | |
| 3957 | btrfs_free_path(path); |
| 3958 | return ret; |
| 3959 | } |
| 3960 | |
| 3961 | int btrfs_extent_readonly(struct btrfs_fs_info *fs_info, u64 bytenr) |
| 3962 | { |
| 3963 | struct btrfs_block_group_cache *block_group; |
| 3964 | int readonly = 0; |
| 3965 | |
| 3966 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
| 3967 | if (!block_group || block_group->ro) |
| 3968 | readonly = 1; |
| 3969 | if (block_group) |
| 3970 | btrfs_put_block_group(block_group); |
| 3971 | return readonly; |
| 3972 | } |
| 3973 | |
| 3974 | bool btrfs_inc_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
| 3975 | { |
| 3976 | struct btrfs_block_group_cache *bg; |
| 3977 | bool ret = true; |
| 3978 | |
| 3979 | bg = btrfs_lookup_block_group(fs_info, bytenr); |
| 3980 | if (!bg) |
| 3981 | return false; |
| 3982 | |
| 3983 | spin_lock(&bg->lock); |
| 3984 | if (bg->ro) |
| 3985 | ret = false; |
| 3986 | else |
| 3987 | atomic_inc(&bg->nocow_writers); |
| 3988 | spin_unlock(&bg->lock); |
| 3989 | |
| 3990 | /* no put on block group, done by btrfs_dec_nocow_writers */ |
| 3991 | if (!ret) |
| 3992 | btrfs_put_block_group(bg); |
| 3993 | |
| 3994 | return ret; |
| 3995 | |
| 3996 | } |
| 3997 | |
| 3998 | void btrfs_dec_nocow_writers(struct btrfs_fs_info *fs_info, u64 bytenr) |
| 3999 | { |
| 4000 | struct btrfs_block_group_cache *bg; |
| 4001 | |
| 4002 | bg = btrfs_lookup_block_group(fs_info, bytenr); |
| 4003 | ASSERT(bg); |
| 4004 | if (atomic_dec_and_test(&bg->nocow_writers)) |
| 4005 | wake_up_var(&bg->nocow_writers); |
| 4006 | /* |
| 4007 | * Once for our lookup and once for the lookup done by a previous call |
| 4008 | * to btrfs_inc_nocow_writers() |
| 4009 | */ |
| 4010 | btrfs_put_block_group(bg); |
| 4011 | btrfs_put_block_group(bg); |
| 4012 | } |
| 4013 | |
| 4014 | void btrfs_wait_nocow_writers(struct btrfs_block_group_cache *bg) |
| 4015 | { |
| 4016 | wait_var_event(&bg->nocow_writers, !atomic_read(&bg->nocow_writers)); |
| 4017 | } |
| 4018 | |
| 4019 | static const char *alloc_name(u64 flags) |
| 4020 | { |
| 4021 | switch (flags) { |
| 4022 | case BTRFS_BLOCK_GROUP_METADATA|BTRFS_BLOCK_GROUP_DATA: |
| 4023 | return "mixed"; |
| 4024 | case BTRFS_BLOCK_GROUP_METADATA: |
| 4025 | return "metadata"; |
| 4026 | case BTRFS_BLOCK_GROUP_DATA: |
| 4027 | return "data"; |
| 4028 | case BTRFS_BLOCK_GROUP_SYSTEM: |
| 4029 | return "system"; |
| 4030 | default: |
| 4031 | WARN_ON(1); |
| 4032 | return "invalid-combination"; |
| 4033 | }; |
| 4034 | } |
| 4035 | |
| 4036 | static int create_space_info(struct btrfs_fs_info *info, u64 flags) |
| 4037 | { |
| 4038 | |
| 4039 | struct btrfs_space_info *space_info; |
| 4040 | int i; |
| 4041 | int ret; |
| 4042 | |
| 4043 | space_info = kzalloc(sizeof(*space_info), GFP_NOFS); |
| 4044 | if (!space_info) |
| 4045 | return -ENOMEM; |
| 4046 | |
| 4047 | ret = percpu_counter_init(&space_info->total_bytes_pinned, 0, |
| 4048 | GFP_KERNEL); |
| 4049 | if (ret) { |
| 4050 | kfree(space_info); |
| 4051 | return ret; |
| 4052 | } |
| 4053 | |
| 4054 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) |
| 4055 | INIT_LIST_HEAD(&space_info->block_groups[i]); |
| 4056 | init_rwsem(&space_info->groups_sem); |
| 4057 | spin_lock_init(&space_info->lock); |
| 4058 | space_info->flags = flags & BTRFS_BLOCK_GROUP_TYPE_MASK; |
| 4059 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
| 4060 | init_waitqueue_head(&space_info->wait); |
| 4061 | INIT_LIST_HEAD(&space_info->ro_bgs); |
| 4062 | INIT_LIST_HEAD(&space_info->tickets); |
| 4063 | INIT_LIST_HEAD(&space_info->priority_tickets); |
| 4064 | |
| 4065 | ret = kobject_init_and_add(&space_info->kobj, &space_info_ktype, |
| 4066 | info->space_info_kobj, "%s", |
| 4067 | alloc_name(space_info->flags)); |
| 4068 | if (ret) { |
| 4069 | percpu_counter_destroy(&space_info->total_bytes_pinned); |
| 4070 | kfree(space_info); |
| 4071 | return ret; |
| 4072 | } |
| 4073 | |
| 4074 | list_add_rcu(&space_info->list, &info->space_info); |
| 4075 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
| 4076 | info->data_sinfo = space_info; |
| 4077 | |
| 4078 | return ret; |
| 4079 | } |
| 4080 | |
| 4081 | static void update_space_info(struct btrfs_fs_info *info, u64 flags, |
| 4082 | u64 total_bytes, u64 bytes_used, |
| 4083 | u64 bytes_readonly, |
| 4084 | struct btrfs_space_info **space_info) |
| 4085 | { |
| 4086 | struct btrfs_space_info *found; |
| 4087 | int factor; |
| 4088 | |
| 4089 | if (flags & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | |
| 4090 | BTRFS_BLOCK_GROUP_RAID10)) |
| 4091 | factor = 2; |
| 4092 | else |
| 4093 | factor = 1; |
| 4094 | |
| 4095 | found = __find_space_info(info, flags); |
| 4096 | ASSERT(found); |
| 4097 | spin_lock(&found->lock); |
| 4098 | found->total_bytes += total_bytes; |
| 4099 | found->disk_total += total_bytes * factor; |
| 4100 | found->bytes_used += bytes_used; |
| 4101 | found->disk_used += bytes_used * factor; |
| 4102 | found->bytes_readonly += bytes_readonly; |
| 4103 | if (total_bytes > 0) |
| 4104 | found->full = 0; |
| 4105 | space_info_add_new_bytes(info, found, total_bytes - |
| 4106 | bytes_used - bytes_readonly); |
| 4107 | spin_unlock(&found->lock); |
| 4108 | *space_info = found; |
| 4109 | } |
| 4110 | |
| 4111 | static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
| 4112 | { |
| 4113 | u64 extra_flags = chunk_to_extended(flags) & |
| 4114 | BTRFS_EXTENDED_PROFILE_MASK; |
| 4115 | |
| 4116 | write_seqlock(&fs_info->profiles_lock); |
| 4117 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
| 4118 | fs_info->avail_data_alloc_bits |= extra_flags; |
| 4119 | if (flags & BTRFS_BLOCK_GROUP_METADATA) |
| 4120 | fs_info->avail_metadata_alloc_bits |= extra_flags; |
| 4121 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
| 4122 | fs_info->avail_system_alloc_bits |= extra_flags; |
| 4123 | write_sequnlock(&fs_info->profiles_lock); |
| 4124 | } |
| 4125 | |
| 4126 | /* |
| 4127 | * returns target flags in extended format or 0 if restripe for this |
| 4128 | * chunk_type is not in progress |
| 4129 | * |
| 4130 | * should be called with balance_lock held |
| 4131 | */ |
| 4132 | static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags) |
| 4133 | { |
| 4134 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
| 4135 | u64 target = 0; |
| 4136 | |
| 4137 | if (!bctl) |
| 4138 | return 0; |
| 4139 | |
| 4140 | if (flags & BTRFS_BLOCK_GROUP_DATA && |
| 4141 | bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
| 4142 | target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target; |
| 4143 | } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM && |
| 4144 | bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
| 4145 | target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target; |
| 4146 | } else if (flags & BTRFS_BLOCK_GROUP_METADATA && |
| 4147 | bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
| 4148 | target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target; |
| 4149 | } |
| 4150 | |
| 4151 | return target; |
| 4152 | } |
| 4153 | |
| 4154 | /* |
| 4155 | * @flags: available profiles in extended format (see ctree.h) |
| 4156 | * |
| 4157 | * Returns reduced profile in chunk format. If profile changing is in |
| 4158 | * progress (either running or paused) picks the target profile (if it's |
| 4159 | * already available), otherwise falls back to plain reducing. |
| 4160 | */ |
| 4161 | static u64 btrfs_reduce_alloc_profile(struct btrfs_fs_info *fs_info, u64 flags) |
| 4162 | { |
| 4163 | u64 num_devices = fs_info->fs_devices->rw_devices; |
| 4164 | u64 target; |
| 4165 | u64 raid_type; |
| 4166 | u64 allowed = 0; |
| 4167 | |
| 4168 | /* |
| 4169 | * see if restripe for this chunk_type is in progress, if so |
| 4170 | * try to reduce to the target profile |
| 4171 | */ |
| 4172 | spin_lock(&fs_info->balance_lock); |
| 4173 | target = get_restripe_target(fs_info, flags); |
| 4174 | if (target) { |
| 4175 | /* pick target profile only if it's already available */ |
| 4176 | if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) { |
| 4177 | spin_unlock(&fs_info->balance_lock); |
| 4178 | return extended_to_chunk(target); |
| 4179 | } |
| 4180 | } |
| 4181 | spin_unlock(&fs_info->balance_lock); |
| 4182 | |
| 4183 | /* First, mask out the RAID levels which aren't possible */ |
| 4184 | for (raid_type = 0; raid_type < BTRFS_NR_RAID_TYPES; raid_type++) { |
| 4185 | if (num_devices >= btrfs_raid_array[raid_type].devs_min) |
| 4186 | allowed |= btrfs_raid_array[raid_type].bg_flag; |
| 4187 | } |
| 4188 | allowed &= flags; |
| 4189 | |
| 4190 | if (allowed & BTRFS_BLOCK_GROUP_RAID6) |
| 4191 | allowed = BTRFS_BLOCK_GROUP_RAID6; |
| 4192 | else if (allowed & BTRFS_BLOCK_GROUP_RAID5) |
| 4193 | allowed = BTRFS_BLOCK_GROUP_RAID5; |
| 4194 | else if (allowed & BTRFS_BLOCK_GROUP_RAID10) |
| 4195 | allowed = BTRFS_BLOCK_GROUP_RAID10; |
| 4196 | else if (allowed & BTRFS_BLOCK_GROUP_RAID1) |
| 4197 | allowed = BTRFS_BLOCK_GROUP_RAID1; |
| 4198 | else if (allowed & BTRFS_BLOCK_GROUP_RAID0) |
| 4199 | allowed = BTRFS_BLOCK_GROUP_RAID0; |
| 4200 | |
| 4201 | flags &= ~BTRFS_BLOCK_GROUP_PROFILE_MASK; |
| 4202 | |
| 4203 | return extended_to_chunk(flags | allowed); |
| 4204 | } |
| 4205 | |
| 4206 | static u64 get_alloc_profile(struct btrfs_fs_info *fs_info, u64 orig_flags) |
| 4207 | { |
| 4208 | unsigned seq; |
| 4209 | u64 flags; |
| 4210 | |
| 4211 | do { |
| 4212 | flags = orig_flags; |
| 4213 | seq = read_seqbegin(&fs_info->profiles_lock); |
| 4214 | |
| 4215 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
| 4216 | flags |= fs_info->avail_data_alloc_bits; |
| 4217 | else if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
| 4218 | flags |= fs_info->avail_system_alloc_bits; |
| 4219 | else if (flags & BTRFS_BLOCK_GROUP_METADATA) |
| 4220 | flags |= fs_info->avail_metadata_alloc_bits; |
| 4221 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
| 4222 | |
| 4223 | return btrfs_reduce_alloc_profile(fs_info, flags); |
| 4224 | } |
| 4225 | |
| 4226 | static u64 get_alloc_profile_by_root(struct btrfs_root *root, int data) |
| 4227 | { |
| 4228 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 4229 | u64 flags; |
| 4230 | u64 ret; |
| 4231 | |
| 4232 | if (data) |
| 4233 | flags = BTRFS_BLOCK_GROUP_DATA; |
| 4234 | else if (root == fs_info->chunk_root) |
| 4235 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
| 4236 | else |
| 4237 | flags = BTRFS_BLOCK_GROUP_METADATA; |
| 4238 | |
| 4239 | ret = get_alloc_profile(fs_info, flags); |
| 4240 | return ret; |
| 4241 | } |
| 4242 | |
| 4243 | u64 btrfs_data_alloc_profile(struct btrfs_fs_info *fs_info) |
| 4244 | { |
| 4245 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_DATA); |
| 4246 | } |
| 4247 | |
| 4248 | u64 btrfs_metadata_alloc_profile(struct btrfs_fs_info *fs_info) |
| 4249 | { |
| 4250 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
| 4251 | } |
| 4252 | |
| 4253 | u64 btrfs_system_alloc_profile(struct btrfs_fs_info *fs_info) |
| 4254 | { |
| 4255 | return get_alloc_profile(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
| 4256 | } |
| 4257 | |
| 4258 | static u64 btrfs_space_info_used(struct btrfs_space_info *s_info, |
| 4259 | bool may_use_included) |
| 4260 | { |
| 4261 | ASSERT(s_info); |
| 4262 | return s_info->bytes_used + s_info->bytes_reserved + |
| 4263 | s_info->bytes_pinned + s_info->bytes_readonly + |
| 4264 | (may_use_included ? s_info->bytes_may_use : 0); |
| 4265 | } |
| 4266 | |
| 4267 | int btrfs_alloc_data_chunk_ondemand(struct btrfs_inode *inode, u64 bytes) |
| 4268 | { |
| 4269 | struct btrfs_root *root = inode->root; |
| 4270 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 4271 | struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; |
| 4272 | u64 used; |
| 4273 | int ret = 0; |
| 4274 | int need_commit = 2; |
| 4275 | int have_pinned_space; |
| 4276 | |
| 4277 | /* make sure bytes are sectorsize aligned */ |
| 4278 | bytes = ALIGN(bytes, fs_info->sectorsize); |
| 4279 | |
| 4280 | if (btrfs_is_free_space_inode(inode)) { |
| 4281 | need_commit = 0; |
| 4282 | ASSERT(current->journal_info); |
| 4283 | } |
| 4284 | |
| 4285 | again: |
| 4286 | /* make sure we have enough space to handle the data first */ |
| 4287 | spin_lock(&data_sinfo->lock); |
| 4288 | used = btrfs_space_info_used(data_sinfo, true); |
| 4289 | |
| 4290 | if (used + bytes > data_sinfo->total_bytes) { |
| 4291 | struct btrfs_trans_handle *trans; |
| 4292 | |
| 4293 | /* |
| 4294 | * if we don't have enough free bytes in this space then we need |
| 4295 | * to alloc a new chunk. |
| 4296 | */ |
| 4297 | if (!data_sinfo->full) { |
| 4298 | u64 alloc_target; |
| 4299 | |
| 4300 | data_sinfo->force_alloc = CHUNK_ALLOC_FORCE; |
| 4301 | spin_unlock(&data_sinfo->lock); |
| 4302 | |
| 4303 | alloc_target = btrfs_data_alloc_profile(fs_info); |
| 4304 | /* |
| 4305 | * It is ugly that we don't call nolock join |
| 4306 | * transaction for the free space inode case here. |
| 4307 | * But it is safe because we only do the data space |
| 4308 | * reservation for the free space cache in the |
| 4309 | * transaction context, the common join transaction |
| 4310 | * just increase the counter of the current transaction |
| 4311 | * handler, doesn't try to acquire the trans_lock of |
| 4312 | * the fs. |
| 4313 | */ |
| 4314 | trans = btrfs_join_transaction(root); |
| 4315 | if (IS_ERR(trans)) |
| 4316 | return PTR_ERR(trans); |
| 4317 | |
| 4318 | ret = do_chunk_alloc(trans, fs_info, alloc_target, |
| 4319 | CHUNK_ALLOC_NO_FORCE); |
| 4320 | btrfs_end_transaction(trans); |
| 4321 | if (ret < 0) { |
| 4322 | if (ret != -ENOSPC) |
| 4323 | return ret; |
| 4324 | else { |
| 4325 | have_pinned_space = 1; |
| 4326 | goto commit_trans; |
| 4327 | } |
| 4328 | } |
| 4329 | |
| 4330 | goto again; |
| 4331 | } |
| 4332 | |
| 4333 | /* |
| 4334 | * If we don't have enough pinned space to deal with this |
| 4335 | * allocation, and no removed chunk in current transaction, |
| 4336 | * don't bother committing the transaction. |
| 4337 | */ |
| 4338 | have_pinned_space = percpu_counter_compare( |
| 4339 | &data_sinfo->total_bytes_pinned, |
| 4340 | used + bytes - data_sinfo->total_bytes); |
| 4341 | spin_unlock(&data_sinfo->lock); |
| 4342 | |
| 4343 | /* commit the current transaction and try again */ |
| 4344 | commit_trans: |
| 4345 | if (need_commit) { |
| 4346 | need_commit--; |
| 4347 | |
| 4348 | if (need_commit > 0) { |
| 4349 | btrfs_start_delalloc_roots(fs_info, -1); |
| 4350 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, |
| 4351 | (u64)-1); |
| 4352 | } |
| 4353 | |
| 4354 | trans = btrfs_join_transaction(root); |
| 4355 | if (IS_ERR(trans)) |
| 4356 | return PTR_ERR(trans); |
| 4357 | if (have_pinned_space >= 0 || |
| 4358 | test_bit(BTRFS_TRANS_HAVE_FREE_BGS, |
| 4359 | &trans->transaction->flags) || |
| 4360 | need_commit > 0) { |
| 4361 | ret = btrfs_commit_transaction(trans); |
| 4362 | if (ret) |
| 4363 | return ret; |
| 4364 | /* |
| 4365 | * The cleaner kthread might still be doing iput |
| 4366 | * operations. Wait for it to finish so that |
| 4367 | * more space is released. |
| 4368 | */ |
| 4369 | mutex_lock(&fs_info->cleaner_delayed_iput_mutex); |
| 4370 | mutex_unlock(&fs_info->cleaner_delayed_iput_mutex); |
| 4371 | goto again; |
| 4372 | } else { |
| 4373 | btrfs_end_transaction(trans); |
| 4374 | } |
| 4375 | } |
| 4376 | |
| 4377 | trace_btrfs_space_reservation(fs_info, |
| 4378 | "space_info:enospc", |
| 4379 | data_sinfo->flags, bytes, 1); |
| 4380 | return -ENOSPC; |
| 4381 | } |
| 4382 | data_sinfo->bytes_may_use += bytes; |
| 4383 | trace_btrfs_space_reservation(fs_info, "space_info", |
| 4384 | data_sinfo->flags, bytes, 1); |
| 4385 | spin_unlock(&data_sinfo->lock); |
| 4386 | |
| 4387 | return ret; |
| 4388 | } |
| 4389 | |
| 4390 | int btrfs_check_data_free_space(struct inode *inode, |
| 4391 | struct extent_changeset **reserved, u64 start, u64 len) |
| 4392 | { |
| 4393 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
| 4394 | int ret; |
| 4395 | |
| 4396 | /* align the range */ |
| 4397 | len = round_up(start + len, fs_info->sectorsize) - |
| 4398 | round_down(start, fs_info->sectorsize); |
| 4399 | start = round_down(start, fs_info->sectorsize); |
| 4400 | |
| 4401 | ret = btrfs_alloc_data_chunk_ondemand(BTRFS_I(inode), len); |
| 4402 | if (ret < 0) |
| 4403 | return ret; |
| 4404 | |
| 4405 | /* Use new btrfs_qgroup_reserve_data to reserve precious data space. */ |
| 4406 | ret = btrfs_qgroup_reserve_data(inode, reserved, start, len); |
| 4407 | if (ret < 0) |
| 4408 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
| 4409 | else |
| 4410 | ret = 0; |
| 4411 | return ret; |
| 4412 | } |
| 4413 | |
| 4414 | /* |
| 4415 | * Called if we need to clear a data reservation for this inode |
| 4416 | * Normally in a error case. |
| 4417 | * |
| 4418 | * This one will *NOT* use accurate qgroup reserved space API, just for case |
| 4419 | * which we can't sleep and is sure it won't affect qgroup reserved space. |
| 4420 | * Like clear_bit_hook(). |
| 4421 | */ |
| 4422 | void btrfs_free_reserved_data_space_noquota(struct inode *inode, u64 start, |
| 4423 | u64 len) |
| 4424 | { |
| 4425 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); |
| 4426 | struct btrfs_space_info *data_sinfo; |
| 4427 | |
| 4428 | /* Make sure the range is aligned to sectorsize */ |
| 4429 | len = round_up(start + len, fs_info->sectorsize) - |
| 4430 | round_down(start, fs_info->sectorsize); |
| 4431 | start = round_down(start, fs_info->sectorsize); |
| 4432 | |
| 4433 | data_sinfo = fs_info->data_sinfo; |
| 4434 | spin_lock(&data_sinfo->lock); |
| 4435 | if (WARN_ON(data_sinfo->bytes_may_use < len)) |
| 4436 | data_sinfo->bytes_may_use = 0; |
| 4437 | else |
| 4438 | data_sinfo->bytes_may_use -= len; |
| 4439 | trace_btrfs_space_reservation(fs_info, "space_info", |
| 4440 | data_sinfo->flags, len, 0); |
| 4441 | spin_unlock(&data_sinfo->lock); |
| 4442 | } |
| 4443 | |
| 4444 | /* |
| 4445 | * Called if we need to clear a data reservation for this inode |
| 4446 | * Normally in a error case. |
| 4447 | * |
| 4448 | * This one will handle the per-inode data rsv map for accurate reserved |
| 4449 | * space framework. |
| 4450 | */ |
| 4451 | void btrfs_free_reserved_data_space(struct inode *inode, |
| 4452 | struct extent_changeset *reserved, u64 start, u64 len) |
| 4453 | { |
| 4454 | struct btrfs_root *root = BTRFS_I(inode)->root; |
| 4455 | |
| 4456 | /* Make sure the range is aligned to sectorsize */ |
| 4457 | len = round_up(start + len, root->fs_info->sectorsize) - |
| 4458 | round_down(start, root->fs_info->sectorsize); |
| 4459 | start = round_down(start, root->fs_info->sectorsize); |
| 4460 | |
| 4461 | btrfs_free_reserved_data_space_noquota(inode, start, len); |
| 4462 | btrfs_qgroup_free_data(inode, reserved, start, len); |
| 4463 | } |
| 4464 | |
| 4465 | static void force_metadata_allocation(struct btrfs_fs_info *info) |
| 4466 | { |
| 4467 | struct list_head *head = &info->space_info; |
| 4468 | struct btrfs_space_info *found; |
| 4469 | |
| 4470 | rcu_read_lock(); |
| 4471 | list_for_each_entry_rcu(found, head, list) { |
| 4472 | if (found->flags & BTRFS_BLOCK_GROUP_METADATA) |
| 4473 | found->force_alloc = CHUNK_ALLOC_FORCE; |
| 4474 | } |
| 4475 | rcu_read_unlock(); |
| 4476 | } |
| 4477 | |
| 4478 | static inline u64 calc_global_rsv_need_space(struct btrfs_block_rsv *global) |
| 4479 | { |
| 4480 | return (global->size << 1); |
| 4481 | } |
| 4482 | |
| 4483 | static int should_alloc_chunk(struct btrfs_fs_info *fs_info, |
| 4484 | struct btrfs_space_info *sinfo, int force) |
| 4485 | { |
| 4486 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
| 4487 | u64 bytes_used = btrfs_space_info_used(sinfo, false); |
| 4488 | u64 thresh; |
| 4489 | |
| 4490 | if (force == CHUNK_ALLOC_FORCE) |
| 4491 | return 1; |
| 4492 | |
| 4493 | /* |
| 4494 | * We need to take into account the global rsv because for all intents |
| 4495 | * and purposes it's used space. Don't worry about locking the |
| 4496 | * global_rsv, it doesn't change except when the transaction commits. |
| 4497 | */ |
| 4498 | if (sinfo->flags & BTRFS_BLOCK_GROUP_METADATA) |
| 4499 | bytes_used += calc_global_rsv_need_space(global_rsv); |
| 4500 | |
| 4501 | /* |
| 4502 | * in limited mode, we want to have some free space up to |
| 4503 | * about 1% of the FS size. |
| 4504 | */ |
| 4505 | if (force == CHUNK_ALLOC_LIMITED) { |
| 4506 | thresh = btrfs_super_total_bytes(fs_info->super_copy); |
| 4507 | thresh = max_t(u64, SZ_64M, div_factor_fine(thresh, 1)); |
| 4508 | |
| 4509 | if (sinfo->total_bytes - bytes_used < thresh) |
| 4510 | return 1; |
| 4511 | } |
| 4512 | |
| 4513 | if (bytes_used + SZ_2M < div_factor(sinfo->total_bytes, 8)) |
| 4514 | return 0; |
| 4515 | return 1; |
| 4516 | } |
| 4517 | |
| 4518 | static u64 get_profile_num_devs(struct btrfs_fs_info *fs_info, u64 type) |
| 4519 | { |
| 4520 | u64 num_dev; |
| 4521 | |
| 4522 | if (type & (BTRFS_BLOCK_GROUP_RAID10 | |
| 4523 | BTRFS_BLOCK_GROUP_RAID0 | |
| 4524 | BTRFS_BLOCK_GROUP_RAID5 | |
| 4525 | BTRFS_BLOCK_GROUP_RAID6)) |
| 4526 | num_dev = fs_info->fs_devices->rw_devices; |
| 4527 | else if (type & BTRFS_BLOCK_GROUP_RAID1) |
| 4528 | num_dev = 2; |
| 4529 | else |
| 4530 | num_dev = 1; /* DUP or single */ |
| 4531 | |
| 4532 | return num_dev; |
| 4533 | } |
| 4534 | |
| 4535 | /* |
| 4536 | * If @is_allocation is true, reserve space in the system space info necessary |
| 4537 | * for allocating a chunk, otherwise if it's false, reserve space necessary for |
| 4538 | * removing a chunk. |
| 4539 | */ |
| 4540 | void check_system_chunk(struct btrfs_trans_handle *trans, |
| 4541 | struct btrfs_fs_info *fs_info, u64 type) |
| 4542 | { |
| 4543 | struct btrfs_space_info *info; |
| 4544 | u64 left; |
| 4545 | u64 thresh; |
| 4546 | int ret = 0; |
| 4547 | u64 num_devs; |
| 4548 | |
| 4549 | /* |
| 4550 | * Needed because we can end up allocating a system chunk and for an |
| 4551 | * atomic and race free space reservation in the chunk block reserve. |
| 4552 | */ |
| 4553 | lockdep_assert_held(&fs_info->chunk_mutex); |
| 4554 | |
| 4555 | info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
| 4556 | spin_lock(&info->lock); |
| 4557 | left = info->total_bytes - btrfs_space_info_used(info, true); |
| 4558 | spin_unlock(&info->lock); |
| 4559 | |
| 4560 | num_devs = get_profile_num_devs(fs_info, type); |
| 4561 | |
| 4562 | /* num_devs device items to update and 1 chunk item to add or remove */ |
| 4563 | thresh = btrfs_calc_trunc_metadata_size(fs_info, num_devs) + |
| 4564 | btrfs_calc_trans_metadata_size(fs_info, 1); |
| 4565 | |
| 4566 | if (left < thresh && btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
| 4567 | btrfs_info(fs_info, "left=%llu, need=%llu, flags=%llu", |
| 4568 | left, thresh, type); |
| 4569 | dump_space_info(fs_info, info, 0, 0); |
| 4570 | } |
| 4571 | |
| 4572 | if (left < thresh) { |
| 4573 | u64 flags = btrfs_system_alloc_profile(fs_info); |
| 4574 | |
| 4575 | /* |
| 4576 | * Ignore failure to create system chunk. We might end up not |
| 4577 | * needing it, as we might not need to COW all nodes/leafs from |
| 4578 | * the paths we visit in the chunk tree (they were already COWed |
| 4579 | * or created in the current transaction for example). |
| 4580 | */ |
| 4581 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
| 4582 | } |
| 4583 | |
| 4584 | if (!ret) { |
| 4585 | ret = btrfs_block_rsv_add(fs_info->chunk_root, |
| 4586 | &fs_info->chunk_block_rsv, |
| 4587 | thresh, BTRFS_RESERVE_NO_FLUSH); |
| 4588 | if (!ret) |
| 4589 | trans->chunk_bytes_reserved += thresh; |
| 4590 | } |
| 4591 | } |
| 4592 | |
| 4593 | /* |
| 4594 | * If force is CHUNK_ALLOC_FORCE: |
| 4595 | * - return 1 if it successfully allocates a chunk, |
| 4596 | * - return errors including -ENOSPC otherwise. |
| 4597 | * If force is NOT CHUNK_ALLOC_FORCE: |
| 4598 | * - return 0 if it doesn't need to allocate a new chunk, |
| 4599 | * - return 1 if it successfully allocates a chunk, |
| 4600 | * - return errors including -ENOSPC otherwise. |
| 4601 | */ |
| 4602 | static int do_chunk_alloc(struct btrfs_trans_handle *trans, |
| 4603 | struct btrfs_fs_info *fs_info, u64 flags, int force) |
| 4604 | { |
| 4605 | struct btrfs_space_info *space_info; |
| 4606 | int wait_for_alloc = 0; |
| 4607 | int ret = 0; |
| 4608 | |
| 4609 | /* Don't re-enter if we're already allocating a chunk */ |
| 4610 | if (trans->allocating_chunk) |
| 4611 | return -ENOSPC; |
| 4612 | |
| 4613 | space_info = __find_space_info(fs_info, flags); |
| 4614 | ASSERT(space_info); |
| 4615 | |
| 4616 | again: |
| 4617 | spin_lock(&space_info->lock); |
| 4618 | if (force < space_info->force_alloc) |
| 4619 | force = space_info->force_alloc; |
| 4620 | if (space_info->full) { |
| 4621 | if (should_alloc_chunk(fs_info, space_info, force)) |
| 4622 | ret = -ENOSPC; |
| 4623 | else |
| 4624 | ret = 0; |
| 4625 | spin_unlock(&space_info->lock); |
| 4626 | return ret; |
| 4627 | } |
| 4628 | |
| 4629 | if (!should_alloc_chunk(fs_info, space_info, force)) { |
| 4630 | spin_unlock(&space_info->lock); |
| 4631 | return 0; |
| 4632 | } else if (space_info->chunk_alloc) { |
| 4633 | wait_for_alloc = 1; |
| 4634 | } else { |
| 4635 | space_info->chunk_alloc = 1; |
| 4636 | } |
| 4637 | |
| 4638 | spin_unlock(&space_info->lock); |
| 4639 | |
| 4640 | mutex_lock(&fs_info->chunk_mutex); |
| 4641 | |
| 4642 | /* |
| 4643 | * The chunk_mutex is held throughout the entirety of a chunk |
| 4644 | * allocation, so once we've acquired the chunk_mutex we know that the |
| 4645 | * other guy is done and we need to recheck and see if we should |
| 4646 | * allocate. |
| 4647 | */ |
| 4648 | if (wait_for_alloc) { |
| 4649 | mutex_unlock(&fs_info->chunk_mutex); |
| 4650 | wait_for_alloc = 0; |
| 4651 | cond_resched(); |
| 4652 | goto again; |
| 4653 | } |
| 4654 | |
| 4655 | trans->allocating_chunk = true; |
| 4656 | |
| 4657 | /* |
| 4658 | * If we have mixed data/metadata chunks we want to make sure we keep |
| 4659 | * allocating mixed chunks instead of individual chunks. |
| 4660 | */ |
| 4661 | if (btrfs_mixed_space_info(space_info)) |
| 4662 | flags |= (BTRFS_BLOCK_GROUP_DATA | BTRFS_BLOCK_GROUP_METADATA); |
| 4663 | |
| 4664 | /* |
| 4665 | * if we're doing a data chunk, go ahead and make sure that |
| 4666 | * we keep a reasonable number of metadata chunks allocated in the |
| 4667 | * FS as well. |
| 4668 | */ |
| 4669 | if (flags & BTRFS_BLOCK_GROUP_DATA && fs_info->metadata_ratio) { |
| 4670 | fs_info->data_chunk_allocations++; |
| 4671 | if (!(fs_info->data_chunk_allocations % |
| 4672 | fs_info->metadata_ratio)) |
| 4673 | force_metadata_allocation(fs_info); |
| 4674 | } |
| 4675 | |
| 4676 | /* |
| 4677 | * Check if we have enough space in SYSTEM chunk because we may need |
| 4678 | * to update devices. |
| 4679 | */ |
| 4680 | check_system_chunk(trans, fs_info, flags); |
| 4681 | |
| 4682 | ret = btrfs_alloc_chunk(trans, fs_info, flags); |
| 4683 | trans->allocating_chunk = false; |
| 4684 | |
| 4685 | spin_lock(&space_info->lock); |
| 4686 | if (ret < 0) { |
| 4687 | if (ret == -ENOSPC) |
| 4688 | space_info->full = 1; |
| 4689 | else |
| 4690 | goto out; |
| 4691 | } else { |
| 4692 | ret = 1; |
| 4693 | } |
| 4694 | |
| 4695 | space_info->force_alloc = CHUNK_ALLOC_NO_FORCE; |
| 4696 | out: |
| 4697 | space_info->chunk_alloc = 0; |
| 4698 | spin_unlock(&space_info->lock); |
| 4699 | mutex_unlock(&fs_info->chunk_mutex); |
| 4700 | /* |
| 4701 | * When we allocate a new chunk we reserve space in the chunk block |
| 4702 | * reserve to make sure we can COW nodes/leafs in the chunk tree or |
| 4703 | * add new nodes/leafs to it if we end up needing to do it when |
| 4704 | * inserting the chunk item and updating device items as part of the |
| 4705 | * second phase of chunk allocation, performed by |
| 4706 | * btrfs_finish_chunk_alloc(). So make sure we don't accumulate a |
| 4707 | * large number of new block groups to create in our transaction |
| 4708 | * handle's new_bgs list to avoid exhausting the chunk block reserve |
| 4709 | * in extreme cases - like having a single transaction create many new |
| 4710 | * block groups when starting to write out the free space caches of all |
| 4711 | * the block groups that were made dirty during the lifetime of the |
| 4712 | * transaction. |
| 4713 | */ |
| 4714 | if (trans->can_flush_pending_bgs && |
| 4715 | trans->chunk_bytes_reserved >= (u64)SZ_2M) { |
| 4716 | btrfs_create_pending_block_groups(trans); |
| 4717 | btrfs_trans_release_chunk_metadata(trans); |
| 4718 | } |
| 4719 | return ret; |
| 4720 | } |
| 4721 | |
| 4722 | static int can_overcommit(struct btrfs_fs_info *fs_info, |
| 4723 | struct btrfs_space_info *space_info, u64 bytes, |
| 4724 | enum btrfs_reserve_flush_enum flush, |
| 4725 | bool system_chunk) |
| 4726 | { |
| 4727 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
| 4728 | u64 profile; |
| 4729 | u64 space_size; |
| 4730 | u64 avail; |
| 4731 | u64 used; |
| 4732 | |
| 4733 | /* Don't overcommit when in mixed mode. */ |
| 4734 | if (space_info->flags & BTRFS_BLOCK_GROUP_DATA) |
| 4735 | return 0; |
| 4736 | |
| 4737 | if (system_chunk) |
| 4738 | profile = btrfs_system_alloc_profile(fs_info); |
| 4739 | else |
| 4740 | profile = btrfs_metadata_alloc_profile(fs_info); |
| 4741 | |
| 4742 | used = btrfs_space_info_used(space_info, false); |
| 4743 | |
| 4744 | /* |
| 4745 | * We only want to allow over committing if we have lots of actual space |
| 4746 | * free, but if we don't have enough space to handle the global reserve |
| 4747 | * space then we could end up having a real enospc problem when trying |
| 4748 | * to allocate a chunk or some other such important allocation. |
| 4749 | */ |
| 4750 | spin_lock(&global_rsv->lock); |
| 4751 | space_size = calc_global_rsv_need_space(global_rsv); |
| 4752 | spin_unlock(&global_rsv->lock); |
| 4753 | if (used + space_size >= space_info->total_bytes) |
| 4754 | return 0; |
| 4755 | |
| 4756 | used += space_info->bytes_may_use; |
| 4757 | |
| 4758 | avail = atomic64_read(&fs_info->free_chunk_space); |
| 4759 | |
| 4760 | /* |
| 4761 | * If we have dup, raid1 or raid10 then only half of the free |
| 4762 | * space is actually useable. For raid56, the space info used |
| 4763 | * doesn't include the parity drive, so we don't have to |
| 4764 | * change the math |
| 4765 | */ |
| 4766 | if (profile & (BTRFS_BLOCK_GROUP_DUP | |
| 4767 | BTRFS_BLOCK_GROUP_RAID1 | |
| 4768 | BTRFS_BLOCK_GROUP_RAID10)) |
| 4769 | avail >>= 1; |
| 4770 | |
| 4771 | /* |
| 4772 | * If we aren't flushing all things, let us overcommit up to |
| 4773 | * 1/2th of the space. If we can flush, don't let us overcommit |
| 4774 | * too much, let it overcommit up to 1/8 of the space. |
| 4775 | */ |
| 4776 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
| 4777 | avail >>= 3; |
| 4778 | else |
| 4779 | avail >>= 1; |
| 4780 | |
| 4781 | if (used + bytes < space_info->total_bytes + avail) |
| 4782 | return 1; |
| 4783 | return 0; |
| 4784 | } |
| 4785 | |
| 4786 | static void btrfs_writeback_inodes_sb_nr(struct btrfs_fs_info *fs_info, |
| 4787 | unsigned long nr_pages, int nr_items) |
| 4788 | { |
| 4789 | struct super_block *sb = fs_info->sb; |
| 4790 | |
| 4791 | if (down_read_trylock(&sb->s_umount)) { |
| 4792 | writeback_inodes_sb_nr(sb, nr_pages, WB_REASON_FS_FREE_SPACE); |
| 4793 | up_read(&sb->s_umount); |
| 4794 | } else { |
| 4795 | /* |
| 4796 | * We needn't worry the filesystem going from r/w to r/o though |
| 4797 | * we don't acquire ->s_umount mutex, because the filesystem |
| 4798 | * should guarantee the delalloc inodes list be empty after |
| 4799 | * the filesystem is readonly(all dirty pages are written to |
| 4800 | * the disk). |
| 4801 | */ |
| 4802 | btrfs_start_delalloc_roots(fs_info, nr_items); |
| 4803 | if (!current->journal_info) |
| 4804 | btrfs_wait_ordered_roots(fs_info, nr_items, 0, (u64)-1); |
| 4805 | } |
| 4806 | } |
| 4807 | |
| 4808 | static inline u64 calc_reclaim_items_nr(struct btrfs_fs_info *fs_info, |
| 4809 | u64 to_reclaim) |
| 4810 | { |
| 4811 | u64 bytes; |
| 4812 | u64 nr; |
| 4813 | |
| 4814 | bytes = btrfs_calc_trans_metadata_size(fs_info, 1); |
| 4815 | nr = div64_u64(to_reclaim, bytes); |
| 4816 | if (!nr) |
| 4817 | nr = 1; |
| 4818 | return nr; |
| 4819 | } |
| 4820 | |
| 4821 | #define EXTENT_SIZE_PER_ITEM SZ_256K |
| 4822 | |
| 4823 | /* |
| 4824 | * shrink metadata reservation for delalloc |
| 4825 | */ |
| 4826 | static void shrink_delalloc(struct btrfs_fs_info *fs_info, u64 to_reclaim, |
| 4827 | u64 orig, bool wait_ordered) |
| 4828 | { |
| 4829 | struct btrfs_space_info *space_info; |
| 4830 | struct btrfs_trans_handle *trans; |
| 4831 | u64 delalloc_bytes; |
| 4832 | u64 max_reclaim; |
| 4833 | u64 items; |
| 4834 | long time_left; |
| 4835 | unsigned long nr_pages; |
| 4836 | int loops; |
| 4837 | |
| 4838 | /* Calc the number of the pages we need flush for space reservation */ |
| 4839 | items = calc_reclaim_items_nr(fs_info, to_reclaim); |
| 4840 | to_reclaim = items * EXTENT_SIZE_PER_ITEM; |
| 4841 | |
| 4842 | trans = (struct btrfs_trans_handle *)current->journal_info; |
| 4843 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
| 4844 | |
| 4845 | delalloc_bytes = percpu_counter_sum_positive( |
| 4846 | &fs_info->delalloc_bytes); |
| 4847 | if (delalloc_bytes == 0) { |
| 4848 | if (trans) |
| 4849 | return; |
| 4850 | if (wait_ordered) |
| 4851 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
| 4852 | return; |
| 4853 | } |
| 4854 | |
| 4855 | loops = 0; |
| 4856 | while (delalloc_bytes && loops < 3) { |
| 4857 | max_reclaim = min(delalloc_bytes, to_reclaim); |
| 4858 | nr_pages = max_reclaim >> PAGE_SHIFT; |
| 4859 | btrfs_writeback_inodes_sb_nr(fs_info, nr_pages, items); |
| 4860 | /* |
| 4861 | * We need to wait for the async pages to actually start before |
| 4862 | * we do anything. |
| 4863 | */ |
| 4864 | max_reclaim = atomic_read(&fs_info->async_delalloc_pages); |
| 4865 | if (!max_reclaim) |
| 4866 | goto skip_async; |
| 4867 | |
| 4868 | if (max_reclaim <= nr_pages) |
| 4869 | max_reclaim = 0; |
| 4870 | else |
| 4871 | max_reclaim -= nr_pages; |
| 4872 | |
| 4873 | wait_event(fs_info->async_submit_wait, |
| 4874 | atomic_read(&fs_info->async_delalloc_pages) <= |
| 4875 | (int)max_reclaim); |
| 4876 | skip_async: |
| 4877 | spin_lock(&space_info->lock); |
| 4878 | if (list_empty(&space_info->tickets) && |
| 4879 | list_empty(&space_info->priority_tickets)) { |
| 4880 | spin_unlock(&space_info->lock); |
| 4881 | break; |
| 4882 | } |
| 4883 | spin_unlock(&space_info->lock); |
| 4884 | |
| 4885 | loops++; |
| 4886 | if (wait_ordered && !trans) { |
| 4887 | btrfs_wait_ordered_roots(fs_info, items, 0, (u64)-1); |
| 4888 | } else { |
| 4889 | time_left = schedule_timeout_killable(1); |
| 4890 | if (time_left) |
| 4891 | break; |
| 4892 | } |
| 4893 | delalloc_bytes = percpu_counter_sum_positive( |
| 4894 | &fs_info->delalloc_bytes); |
| 4895 | } |
| 4896 | } |
| 4897 | |
| 4898 | struct reserve_ticket { |
| 4899 | u64 bytes; |
| 4900 | int error; |
| 4901 | struct list_head list; |
| 4902 | wait_queue_head_t wait; |
| 4903 | }; |
| 4904 | |
| 4905 | /** |
| 4906 | * maybe_commit_transaction - possibly commit the transaction if its ok to |
| 4907 | * @root - the root we're allocating for |
| 4908 | * @bytes - the number of bytes we want to reserve |
| 4909 | * @force - force the commit |
| 4910 | * |
| 4911 | * This will check to make sure that committing the transaction will actually |
| 4912 | * get us somewhere and then commit the transaction if it does. Otherwise it |
| 4913 | * will return -ENOSPC. |
| 4914 | */ |
| 4915 | static int may_commit_transaction(struct btrfs_fs_info *fs_info, |
| 4916 | struct btrfs_space_info *space_info) |
| 4917 | { |
| 4918 | struct reserve_ticket *ticket = NULL; |
| 4919 | struct btrfs_block_rsv *delayed_rsv = &fs_info->delayed_block_rsv; |
| 4920 | struct btrfs_trans_handle *trans; |
| 4921 | u64 bytes; |
| 4922 | |
| 4923 | trans = (struct btrfs_trans_handle *)current->journal_info; |
| 4924 | if (trans) |
| 4925 | return -EAGAIN; |
| 4926 | |
| 4927 | spin_lock(&space_info->lock); |
| 4928 | if (!list_empty(&space_info->priority_tickets)) |
| 4929 | ticket = list_first_entry(&space_info->priority_tickets, |
| 4930 | struct reserve_ticket, list); |
| 4931 | else if (!list_empty(&space_info->tickets)) |
| 4932 | ticket = list_first_entry(&space_info->tickets, |
| 4933 | struct reserve_ticket, list); |
| 4934 | bytes = (ticket) ? ticket->bytes : 0; |
| 4935 | spin_unlock(&space_info->lock); |
| 4936 | |
| 4937 | if (!bytes) |
| 4938 | return 0; |
| 4939 | |
| 4940 | /* See if there is enough pinned space to make this reservation */ |
| 4941 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
| 4942 | bytes) >= 0) |
| 4943 | goto commit; |
| 4944 | |
| 4945 | /* |
| 4946 | * See if there is some space in the delayed insertion reservation for |
| 4947 | * this reservation. |
| 4948 | */ |
| 4949 | if (space_info != delayed_rsv->space_info) |
| 4950 | return -ENOSPC; |
| 4951 | |
| 4952 | spin_lock(&delayed_rsv->lock); |
| 4953 | if (delayed_rsv->size > bytes) |
| 4954 | bytes = 0; |
| 4955 | else |
| 4956 | bytes -= delayed_rsv->size; |
| 4957 | spin_unlock(&delayed_rsv->lock); |
| 4958 | |
| 4959 | if (percpu_counter_compare(&space_info->total_bytes_pinned, |
| 4960 | bytes) < 0) { |
| 4961 | return -ENOSPC; |
| 4962 | } |
| 4963 | |
| 4964 | commit: |
| 4965 | trans = btrfs_join_transaction(fs_info->extent_root); |
| 4966 | if (IS_ERR(trans)) |
| 4967 | return -ENOSPC; |
| 4968 | |
| 4969 | return btrfs_commit_transaction(trans); |
| 4970 | } |
| 4971 | |
| 4972 | /* |
| 4973 | * Try to flush some data based on policy set by @state. This is only advisory |
| 4974 | * and may fail for various reasons. The caller is supposed to examine the |
| 4975 | * state of @space_info to detect the outcome. |
| 4976 | */ |
| 4977 | static void flush_space(struct btrfs_fs_info *fs_info, |
| 4978 | struct btrfs_space_info *space_info, u64 num_bytes, |
| 4979 | int state) |
| 4980 | { |
| 4981 | struct btrfs_root *root = fs_info->extent_root; |
| 4982 | struct btrfs_trans_handle *trans; |
| 4983 | int nr; |
| 4984 | int ret = 0; |
| 4985 | |
| 4986 | switch (state) { |
| 4987 | case FLUSH_DELAYED_ITEMS_NR: |
| 4988 | case FLUSH_DELAYED_ITEMS: |
| 4989 | if (state == FLUSH_DELAYED_ITEMS_NR) |
| 4990 | nr = calc_reclaim_items_nr(fs_info, num_bytes) * 2; |
| 4991 | else |
| 4992 | nr = -1; |
| 4993 | |
| 4994 | trans = btrfs_join_transaction(root); |
| 4995 | if (IS_ERR(trans)) { |
| 4996 | ret = PTR_ERR(trans); |
| 4997 | break; |
| 4998 | } |
| 4999 | ret = btrfs_run_delayed_items_nr(trans, nr); |
| 5000 | btrfs_end_transaction(trans); |
| 5001 | break; |
| 5002 | case FLUSH_DELALLOC: |
| 5003 | case FLUSH_DELALLOC_WAIT: |
| 5004 | shrink_delalloc(fs_info, num_bytes * 2, num_bytes, |
| 5005 | state == FLUSH_DELALLOC_WAIT); |
| 5006 | break; |
| 5007 | case ALLOC_CHUNK: |
| 5008 | trans = btrfs_join_transaction(root); |
| 5009 | if (IS_ERR(trans)) { |
| 5010 | ret = PTR_ERR(trans); |
| 5011 | break; |
| 5012 | } |
| 5013 | ret = do_chunk_alloc(trans, fs_info, |
| 5014 | btrfs_metadata_alloc_profile(fs_info), |
| 5015 | CHUNK_ALLOC_NO_FORCE); |
| 5016 | btrfs_end_transaction(trans); |
| 5017 | if (ret > 0 || ret == -ENOSPC) |
| 5018 | ret = 0; |
| 5019 | break; |
| 5020 | case COMMIT_TRANS: |
| 5021 | ret = may_commit_transaction(fs_info, space_info); |
| 5022 | break; |
| 5023 | default: |
| 5024 | ret = -ENOSPC; |
| 5025 | break; |
| 5026 | } |
| 5027 | |
| 5028 | trace_btrfs_flush_space(fs_info, space_info->flags, num_bytes, state, |
| 5029 | ret); |
| 5030 | return; |
| 5031 | } |
| 5032 | |
| 5033 | static inline u64 |
| 5034 | btrfs_calc_reclaim_metadata_size(struct btrfs_fs_info *fs_info, |
| 5035 | struct btrfs_space_info *space_info, |
| 5036 | bool system_chunk) |
| 5037 | { |
| 5038 | struct reserve_ticket *ticket; |
| 5039 | u64 used; |
| 5040 | u64 expected; |
| 5041 | u64 to_reclaim = 0; |
| 5042 | |
| 5043 | list_for_each_entry(ticket, &space_info->tickets, list) |
| 5044 | to_reclaim += ticket->bytes; |
| 5045 | list_for_each_entry(ticket, &space_info->priority_tickets, list) |
| 5046 | to_reclaim += ticket->bytes; |
| 5047 | if (to_reclaim) |
| 5048 | return to_reclaim; |
| 5049 | |
| 5050 | to_reclaim = min_t(u64, num_online_cpus() * SZ_1M, SZ_16M); |
| 5051 | if (can_overcommit(fs_info, space_info, to_reclaim, |
| 5052 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) |
| 5053 | return 0; |
| 5054 | |
| 5055 | used = btrfs_space_info_used(space_info, true); |
| 5056 | |
| 5057 | if (can_overcommit(fs_info, space_info, SZ_1M, |
| 5058 | BTRFS_RESERVE_FLUSH_ALL, system_chunk)) |
| 5059 | expected = div_factor_fine(space_info->total_bytes, 95); |
| 5060 | else |
| 5061 | expected = div_factor_fine(space_info->total_bytes, 90); |
| 5062 | |
| 5063 | if (used > expected) |
| 5064 | to_reclaim = used - expected; |
| 5065 | else |
| 5066 | to_reclaim = 0; |
| 5067 | to_reclaim = min(to_reclaim, space_info->bytes_may_use + |
| 5068 | space_info->bytes_reserved); |
| 5069 | return to_reclaim; |
| 5070 | } |
| 5071 | |
| 5072 | static inline int need_do_async_reclaim(struct btrfs_fs_info *fs_info, |
| 5073 | struct btrfs_space_info *space_info, |
| 5074 | u64 used, bool system_chunk) |
| 5075 | { |
| 5076 | u64 thresh = div_factor_fine(space_info->total_bytes, 98); |
| 5077 | |
| 5078 | /* If we're just plain full then async reclaim just slows us down. */ |
| 5079 | if ((space_info->bytes_used + space_info->bytes_reserved) >= thresh) |
| 5080 | return 0; |
| 5081 | |
| 5082 | if (!btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
| 5083 | system_chunk)) |
| 5084 | return 0; |
| 5085 | |
| 5086 | return (used >= thresh && !btrfs_fs_closing(fs_info) && |
| 5087 | !test_bit(BTRFS_FS_STATE_REMOUNTING, &fs_info->fs_state)); |
| 5088 | } |
| 5089 | |
| 5090 | static void wake_all_tickets(struct list_head *head) |
| 5091 | { |
| 5092 | struct reserve_ticket *ticket; |
| 5093 | |
| 5094 | while (!list_empty(head)) { |
| 5095 | ticket = list_first_entry(head, struct reserve_ticket, list); |
| 5096 | list_del_init(&ticket->list); |
| 5097 | ticket->error = -ENOSPC; |
| 5098 | wake_up(&ticket->wait); |
| 5099 | } |
| 5100 | } |
| 5101 | |
| 5102 | /* |
| 5103 | * This is for normal flushers, we can wait all goddamned day if we want to. We |
| 5104 | * will loop and continuously try to flush as long as we are making progress. |
| 5105 | * We count progress as clearing off tickets each time we have to loop. |
| 5106 | */ |
| 5107 | static void btrfs_async_reclaim_metadata_space(struct work_struct *work) |
| 5108 | { |
| 5109 | struct btrfs_fs_info *fs_info; |
| 5110 | struct btrfs_space_info *space_info; |
| 5111 | u64 to_reclaim; |
| 5112 | int flush_state; |
| 5113 | int commit_cycles = 0; |
| 5114 | u64 last_tickets_id; |
| 5115 | |
| 5116 | fs_info = container_of(work, struct btrfs_fs_info, async_reclaim_work); |
| 5117 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
| 5118 | |
| 5119 | spin_lock(&space_info->lock); |
| 5120 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
| 5121 | false); |
| 5122 | if (!to_reclaim) { |
| 5123 | space_info->flush = 0; |
| 5124 | spin_unlock(&space_info->lock); |
| 5125 | return; |
| 5126 | } |
| 5127 | last_tickets_id = space_info->tickets_id; |
| 5128 | spin_unlock(&space_info->lock); |
| 5129 | |
| 5130 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
| 5131 | do { |
| 5132 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
| 5133 | spin_lock(&space_info->lock); |
| 5134 | if (list_empty(&space_info->tickets)) { |
| 5135 | space_info->flush = 0; |
| 5136 | spin_unlock(&space_info->lock); |
| 5137 | return; |
| 5138 | } |
| 5139 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, |
| 5140 | space_info, |
| 5141 | false); |
| 5142 | if (last_tickets_id == space_info->tickets_id) { |
| 5143 | flush_state++; |
| 5144 | } else { |
| 5145 | last_tickets_id = space_info->tickets_id; |
| 5146 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
| 5147 | if (commit_cycles) |
| 5148 | commit_cycles--; |
| 5149 | } |
| 5150 | |
| 5151 | if (flush_state > COMMIT_TRANS) { |
| 5152 | commit_cycles++; |
| 5153 | if (commit_cycles > 2) { |
| 5154 | wake_all_tickets(&space_info->tickets); |
| 5155 | space_info->flush = 0; |
| 5156 | } else { |
| 5157 | flush_state = FLUSH_DELAYED_ITEMS_NR; |
| 5158 | } |
| 5159 | } |
| 5160 | spin_unlock(&space_info->lock); |
| 5161 | } while (flush_state <= COMMIT_TRANS); |
| 5162 | } |
| 5163 | |
| 5164 | void btrfs_init_async_reclaim_work(struct work_struct *work) |
| 5165 | { |
| 5166 | INIT_WORK(work, btrfs_async_reclaim_metadata_space); |
| 5167 | } |
| 5168 | |
| 5169 | static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, |
| 5170 | struct btrfs_space_info *space_info, |
| 5171 | struct reserve_ticket *ticket) |
| 5172 | { |
| 5173 | u64 to_reclaim; |
| 5174 | int flush_state = FLUSH_DELAYED_ITEMS_NR; |
| 5175 | |
| 5176 | spin_lock(&space_info->lock); |
| 5177 | to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info, |
| 5178 | false); |
| 5179 | if (!to_reclaim) { |
| 5180 | spin_unlock(&space_info->lock); |
| 5181 | return; |
| 5182 | } |
| 5183 | spin_unlock(&space_info->lock); |
| 5184 | |
| 5185 | do { |
| 5186 | flush_space(fs_info, space_info, to_reclaim, flush_state); |
| 5187 | flush_state++; |
| 5188 | spin_lock(&space_info->lock); |
| 5189 | if (ticket->bytes == 0) { |
| 5190 | spin_unlock(&space_info->lock); |
| 5191 | return; |
| 5192 | } |
| 5193 | spin_unlock(&space_info->lock); |
| 5194 | |
| 5195 | /* |
| 5196 | * Priority flushers can't wait on delalloc without |
| 5197 | * deadlocking. |
| 5198 | */ |
| 5199 | if (flush_state == FLUSH_DELALLOC || |
| 5200 | flush_state == FLUSH_DELALLOC_WAIT) |
| 5201 | flush_state = ALLOC_CHUNK; |
| 5202 | } while (flush_state < COMMIT_TRANS); |
| 5203 | } |
| 5204 | |
| 5205 | static int wait_reserve_ticket(struct btrfs_fs_info *fs_info, |
| 5206 | struct btrfs_space_info *space_info, |
| 5207 | struct reserve_ticket *ticket, u64 orig_bytes) |
| 5208 | |
| 5209 | { |
| 5210 | DEFINE_WAIT(wait); |
| 5211 | int ret = 0; |
| 5212 | |
| 5213 | spin_lock(&space_info->lock); |
| 5214 | while (ticket->bytes > 0 && ticket->error == 0) { |
| 5215 | ret = prepare_to_wait_event(&ticket->wait, &wait, TASK_KILLABLE); |
| 5216 | if (ret) { |
| 5217 | ret = -EINTR; |
| 5218 | break; |
| 5219 | } |
| 5220 | spin_unlock(&space_info->lock); |
| 5221 | |
| 5222 | schedule(); |
| 5223 | |
| 5224 | finish_wait(&ticket->wait, &wait); |
| 5225 | spin_lock(&space_info->lock); |
| 5226 | } |
| 5227 | if (!ret) |
| 5228 | ret = ticket->error; |
| 5229 | if (!list_empty(&ticket->list)) |
| 5230 | list_del_init(&ticket->list); |
| 5231 | if (ticket->bytes && ticket->bytes < orig_bytes) { |
| 5232 | u64 num_bytes = orig_bytes - ticket->bytes; |
| 5233 | space_info->bytes_may_use -= num_bytes; |
| 5234 | trace_btrfs_space_reservation(fs_info, "space_info", |
| 5235 | space_info->flags, num_bytes, 0); |
| 5236 | } |
| 5237 | spin_unlock(&space_info->lock); |
| 5238 | |
| 5239 | return ret; |
| 5240 | } |
| 5241 | |
| 5242 | /** |
| 5243 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space |
| 5244 | * @root - the root we're allocating for |
| 5245 | * @space_info - the space info we want to allocate from |
| 5246 | * @orig_bytes - the number of bytes we want |
| 5247 | * @flush - whether or not we can flush to make our reservation |
| 5248 | * |
| 5249 | * This will reserve orig_bytes number of bytes from the space info associated |
| 5250 | * with the block_rsv. If there is not enough space it will make an attempt to |
| 5251 | * flush out space to make room. It will do this by flushing delalloc if |
| 5252 | * possible or committing the transaction. If flush is 0 then no attempts to |
| 5253 | * regain reservations will be made and this will fail if there is not enough |
| 5254 | * space already. |
| 5255 | */ |
| 5256 | static int __reserve_metadata_bytes(struct btrfs_fs_info *fs_info, |
| 5257 | struct btrfs_space_info *space_info, |
| 5258 | u64 orig_bytes, |
| 5259 | enum btrfs_reserve_flush_enum flush, |
| 5260 | bool system_chunk) |
| 5261 | { |
| 5262 | struct reserve_ticket ticket; |
| 5263 | u64 used; |
| 5264 | int ret = 0; |
| 5265 | |
| 5266 | ASSERT(orig_bytes); |
| 5267 | ASSERT(!current->journal_info || flush != BTRFS_RESERVE_FLUSH_ALL); |
| 5268 | |
| 5269 | spin_lock(&space_info->lock); |
| 5270 | ret = -ENOSPC; |
| 5271 | used = btrfs_space_info_used(space_info, true); |
| 5272 | |
| 5273 | /* |
| 5274 | * If we have enough space then hooray, make our reservation and carry |
| 5275 | * on. If not see if we can overcommit, and if we can, hooray carry on. |
| 5276 | * If not things get more complicated. |
| 5277 | */ |
| 5278 | if (used + orig_bytes <= space_info->total_bytes) { |
| 5279 | space_info->bytes_may_use += orig_bytes; |
| 5280 | trace_btrfs_space_reservation(fs_info, "space_info", |
| 5281 | space_info->flags, orig_bytes, 1); |
| 5282 | ret = 0; |
| 5283 | } else if (can_overcommit(fs_info, space_info, orig_bytes, flush, |
| 5284 | system_chunk)) { |
| 5285 | space_info->bytes_may_use += orig_bytes; |
| 5286 | trace_btrfs_space_reservation(fs_info, "space_info", |
| 5287 | space_info->flags, orig_bytes, 1); |
| 5288 | ret = 0; |
| 5289 | } |
| 5290 | |
| 5291 | /* |
| 5292 | * If we couldn't make a reservation then setup our reservation ticket |
| 5293 | * and kick the async worker if it's not already running. |
| 5294 | * |
| 5295 | * If we are a priority flusher then we just need to add our ticket to |
| 5296 | * the list and we will do our own flushing further down. |
| 5297 | */ |
| 5298 | if (ret && flush != BTRFS_RESERVE_NO_FLUSH) { |
| 5299 | ticket.bytes = orig_bytes; |
| 5300 | ticket.error = 0; |
| 5301 | init_waitqueue_head(&ticket.wait); |
| 5302 | if (flush == BTRFS_RESERVE_FLUSH_ALL) { |
| 5303 | list_add_tail(&ticket.list, &space_info->tickets); |
| 5304 | if (!space_info->flush) { |
| 5305 | space_info->flush = 1; |
| 5306 | trace_btrfs_trigger_flush(fs_info, |
| 5307 | space_info->flags, |
| 5308 | orig_bytes, flush, |
| 5309 | "enospc"); |
| 5310 | queue_work(system_unbound_wq, |
| 5311 | &fs_info->async_reclaim_work); |
| 5312 | } |
| 5313 | } else { |
| 5314 | list_add_tail(&ticket.list, |
| 5315 | &space_info->priority_tickets); |
| 5316 | } |
| 5317 | } else if (!ret && space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
| 5318 | used += orig_bytes; |
| 5319 | /* |
| 5320 | * We will do the space reservation dance during log replay, |
| 5321 | * which means we won't have fs_info->fs_root set, so don't do |
| 5322 | * the async reclaim as we will panic. |
| 5323 | */ |
| 5324 | if (!test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags) && |
| 5325 | need_do_async_reclaim(fs_info, space_info, |
| 5326 | used, system_chunk) && |
| 5327 | !work_busy(&fs_info->async_reclaim_work)) { |
| 5328 | trace_btrfs_trigger_flush(fs_info, space_info->flags, |
| 5329 | orig_bytes, flush, "preempt"); |
| 5330 | queue_work(system_unbound_wq, |
| 5331 | &fs_info->async_reclaim_work); |
| 5332 | } |
| 5333 | } |
| 5334 | spin_unlock(&space_info->lock); |
| 5335 | if (!ret || flush == BTRFS_RESERVE_NO_FLUSH) |
| 5336 | return ret; |
| 5337 | |
| 5338 | if (flush == BTRFS_RESERVE_FLUSH_ALL) |
| 5339 | return wait_reserve_ticket(fs_info, space_info, &ticket, |
| 5340 | orig_bytes); |
| 5341 | |
| 5342 | ret = 0; |
| 5343 | priority_reclaim_metadata_space(fs_info, space_info, &ticket); |
| 5344 | spin_lock(&space_info->lock); |
| 5345 | if (ticket.bytes) { |
| 5346 | if (ticket.bytes < orig_bytes) { |
| 5347 | u64 num_bytes = orig_bytes - ticket.bytes; |
| 5348 | space_info->bytes_may_use -= num_bytes; |
| 5349 | trace_btrfs_space_reservation(fs_info, "space_info", |
| 5350 | space_info->flags, |
| 5351 | num_bytes, 0); |
| 5352 | |
| 5353 | } |
| 5354 | list_del_init(&ticket.list); |
| 5355 | ret = -ENOSPC; |
| 5356 | } |
| 5357 | spin_unlock(&space_info->lock); |
| 5358 | ASSERT(list_empty(&ticket.list)); |
| 5359 | return ret; |
| 5360 | } |
| 5361 | |
| 5362 | /** |
| 5363 | * reserve_metadata_bytes - try to reserve bytes from the block_rsv's space |
| 5364 | * @root - the root we're allocating for |
| 5365 | * @block_rsv - the block_rsv we're allocating for |
| 5366 | * @orig_bytes - the number of bytes we want |
| 5367 | * @flush - whether or not we can flush to make our reservation |
| 5368 | * |
| 5369 | * This will reserve orgi_bytes number of bytes from the space info associated |
| 5370 | * with the block_rsv. If there is not enough space it will make an attempt to |
| 5371 | * flush out space to make room. It will do this by flushing delalloc if |
| 5372 | * possible or committing the transaction. If flush is 0 then no attempts to |
| 5373 | * regain reservations will be made and this will fail if there is not enough |
| 5374 | * space already. |
| 5375 | */ |
| 5376 | static int reserve_metadata_bytes(struct btrfs_root *root, |
| 5377 | struct btrfs_block_rsv *block_rsv, |
| 5378 | u64 orig_bytes, |
| 5379 | enum btrfs_reserve_flush_enum flush) |
| 5380 | { |
| 5381 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 5382 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
| 5383 | int ret; |
| 5384 | bool system_chunk = (root == fs_info->chunk_root); |
| 5385 | |
| 5386 | ret = __reserve_metadata_bytes(fs_info, block_rsv->space_info, |
| 5387 | orig_bytes, flush, system_chunk); |
| 5388 | if (ret == -ENOSPC && |
| 5389 | unlikely(root->orphan_cleanup_state == ORPHAN_CLEANUP_STARTED)) { |
| 5390 | if (block_rsv != global_rsv && |
| 5391 | !block_rsv_use_bytes(global_rsv, orig_bytes)) |
| 5392 | ret = 0; |
| 5393 | } |
| 5394 | if (ret == -ENOSPC) { |
| 5395 | trace_btrfs_space_reservation(fs_info, "space_info:enospc", |
| 5396 | block_rsv->space_info->flags, |
| 5397 | orig_bytes, 1); |
| 5398 | |
| 5399 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) |
| 5400 | dump_space_info(fs_info, block_rsv->space_info, |
| 5401 | orig_bytes, 0); |
| 5402 | } |
| 5403 | return ret; |
| 5404 | } |
| 5405 | |
| 5406 | static struct btrfs_block_rsv *get_block_rsv( |
| 5407 | const struct btrfs_trans_handle *trans, |
| 5408 | const struct btrfs_root *root) |
| 5409 | { |
| 5410 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 5411 | struct btrfs_block_rsv *block_rsv = NULL; |
| 5412 | |
| 5413 | if (test_bit(BTRFS_ROOT_REF_COWS, &root->state) || |
| 5414 | (root == fs_info->csum_root && trans->adding_csums) || |
| 5415 | (root == fs_info->uuid_root)) |
| 5416 | block_rsv = trans->block_rsv; |
| 5417 | |
| 5418 | if (!block_rsv) |
| 5419 | block_rsv = root->block_rsv; |
| 5420 | |
| 5421 | if (!block_rsv) |
| 5422 | block_rsv = &fs_info->empty_block_rsv; |
| 5423 | |
| 5424 | return block_rsv; |
| 5425 | } |
| 5426 | |
| 5427 | static int block_rsv_use_bytes(struct btrfs_block_rsv *block_rsv, |
| 5428 | u64 num_bytes) |
| 5429 | { |
| 5430 | int ret = -ENOSPC; |
| 5431 | spin_lock(&block_rsv->lock); |
| 5432 | if (block_rsv->reserved >= num_bytes) { |
| 5433 | block_rsv->reserved -= num_bytes; |
| 5434 | if (block_rsv->reserved < block_rsv->size) |
| 5435 | block_rsv->full = 0; |
| 5436 | ret = 0; |
| 5437 | } |
| 5438 | spin_unlock(&block_rsv->lock); |
| 5439 | return ret; |
| 5440 | } |
| 5441 | |
| 5442 | static void block_rsv_add_bytes(struct btrfs_block_rsv *block_rsv, |
| 5443 | u64 num_bytes, int update_size) |
| 5444 | { |
| 5445 | spin_lock(&block_rsv->lock); |
| 5446 | block_rsv->reserved += num_bytes; |
| 5447 | if (update_size) |
| 5448 | block_rsv->size += num_bytes; |
| 5449 | else if (block_rsv->reserved >= block_rsv->size) |
| 5450 | block_rsv->full = 1; |
| 5451 | spin_unlock(&block_rsv->lock); |
| 5452 | } |
| 5453 | |
| 5454 | int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info, |
| 5455 | struct btrfs_block_rsv *dest, u64 num_bytes, |
| 5456 | int min_factor) |
| 5457 | { |
| 5458 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
| 5459 | u64 min_bytes; |
| 5460 | |
| 5461 | if (global_rsv->space_info != dest->space_info) |
| 5462 | return -ENOSPC; |
| 5463 | |
| 5464 | spin_lock(&global_rsv->lock); |
| 5465 | min_bytes = div_factor(global_rsv->size, min_factor); |
| 5466 | if (global_rsv->reserved < min_bytes + num_bytes) { |
| 5467 | spin_unlock(&global_rsv->lock); |
| 5468 | return -ENOSPC; |
| 5469 | } |
| 5470 | global_rsv->reserved -= num_bytes; |
| 5471 | if (global_rsv->reserved < global_rsv->size) |
| 5472 | global_rsv->full = 0; |
| 5473 | spin_unlock(&global_rsv->lock); |
| 5474 | |
| 5475 | block_rsv_add_bytes(dest, num_bytes, 1); |
| 5476 | return 0; |
| 5477 | } |
| 5478 | |
| 5479 | /* |
| 5480 | * This is for space we already have accounted in space_info->bytes_may_use, so |
| 5481 | * basically when we're returning space from block_rsv's. |
| 5482 | */ |
| 5483 | static void space_info_add_old_bytes(struct btrfs_fs_info *fs_info, |
| 5484 | struct btrfs_space_info *space_info, |
| 5485 | u64 num_bytes) |
| 5486 | { |
| 5487 | struct reserve_ticket *ticket; |
| 5488 | struct list_head *head; |
| 5489 | u64 used; |
| 5490 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_NO_FLUSH; |
| 5491 | bool check_overcommit = false; |
| 5492 | |
| 5493 | spin_lock(&space_info->lock); |
| 5494 | head = &space_info->priority_tickets; |
| 5495 | |
| 5496 | /* |
| 5497 | * If we are over our limit then we need to check and see if we can |
| 5498 | * overcommit, and if we can't then we just need to free up our space |
| 5499 | * and not satisfy any requests. |
| 5500 | */ |
| 5501 | used = btrfs_space_info_used(space_info, true); |
| 5502 | if (used - num_bytes >= space_info->total_bytes) |
| 5503 | check_overcommit = true; |
| 5504 | again: |
| 5505 | while (!list_empty(head) && num_bytes) { |
| 5506 | ticket = list_first_entry(head, struct reserve_ticket, |
| 5507 | list); |
| 5508 | /* |
| 5509 | * We use 0 bytes because this space is already reserved, so |
| 5510 | * adding the ticket space would be a double count. |
| 5511 | */ |
| 5512 | if (check_overcommit && |
| 5513 | !can_overcommit(fs_info, space_info, 0, flush, false)) |
| 5514 | break; |
| 5515 | if (num_bytes >= ticket->bytes) { |
| 5516 | list_del_init(&ticket->list); |
| 5517 | num_bytes -= ticket->bytes; |
| 5518 | ticket->bytes = 0; |
| 5519 | space_info->tickets_id++; |
| 5520 | wake_up(&ticket->wait); |
| 5521 | } else { |
| 5522 | ticket->bytes -= num_bytes; |
| 5523 | num_bytes = 0; |
| 5524 | } |
| 5525 | } |
| 5526 | |
| 5527 | if (num_bytes && head == &space_info->priority_tickets) { |
| 5528 | head = &space_info->tickets; |
| 5529 | flush = BTRFS_RESERVE_FLUSH_ALL; |
| 5530 | goto again; |
| 5531 | } |
| 5532 | space_info->bytes_may_use -= num_bytes; |
| 5533 | trace_btrfs_space_reservation(fs_info, "space_info", |
| 5534 | space_info->flags, num_bytes, 0); |
| 5535 | spin_unlock(&space_info->lock); |
| 5536 | } |
| 5537 | |
| 5538 | /* |
| 5539 | * This is for newly allocated space that isn't accounted in |
| 5540 | * space_info->bytes_may_use yet. So if we allocate a chunk or unpin an extent |
| 5541 | * we use this helper. |
| 5542 | */ |
| 5543 | static void space_info_add_new_bytes(struct btrfs_fs_info *fs_info, |
| 5544 | struct btrfs_space_info *space_info, |
| 5545 | u64 num_bytes) |
| 5546 | { |
| 5547 | struct reserve_ticket *ticket; |
| 5548 | struct list_head *head = &space_info->priority_tickets; |
| 5549 | |
| 5550 | again: |
| 5551 | while (!list_empty(head) && num_bytes) { |
| 5552 | ticket = list_first_entry(head, struct reserve_ticket, |
| 5553 | list); |
| 5554 | if (num_bytes >= ticket->bytes) { |
| 5555 | trace_btrfs_space_reservation(fs_info, "space_info", |
| 5556 | space_info->flags, |
| 5557 | ticket->bytes, 1); |
| 5558 | list_del_init(&ticket->list); |
| 5559 | num_bytes -= ticket->bytes; |
| 5560 | space_info->bytes_may_use += ticket->bytes; |
| 5561 | ticket->bytes = 0; |
| 5562 | space_info->tickets_id++; |
| 5563 | wake_up(&ticket->wait); |
| 5564 | } else { |
| 5565 | trace_btrfs_space_reservation(fs_info, "space_info", |
| 5566 | space_info->flags, |
| 5567 | num_bytes, 1); |
| 5568 | space_info->bytes_may_use += num_bytes; |
| 5569 | ticket->bytes -= num_bytes; |
| 5570 | num_bytes = 0; |
| 5571 | } |
| 5572 | } |
| 5573 | |
| 5574 | if (num_bytes && head == &space_info->priority_tickets) { |
| 5575 | head = &space_info->tickets; |
| 5576 | goto again; |
| 5577 | } |
| 5578 | } |
| 5579 | |
| 5580 | static u64 block_rsv_release_bytes(struct btrfs_fs_info *fs_info, |
| 5581 | struct btrfs_block_rsv *block_rsv, |
| 5582 | struct btrfs_block_rsv *dest, u64 num_bytes, |
| 5583 | u64 *qgroup_to_release_ret) |
| 5584 | { |
| 5585 | struct btrfs_space_info *space_info = block_rsv->space_info; |
| 5586 | u64 qgroup_to_release = 0; |
| 5587 | u64 ret; |
| 5588 | |
| 5589 | spin_lock(&block_rsv->lock); |
| 5590 | if (num_bytes == (u64)-1) { |
| 5591 | num_bytes = block_rsv->size; |
| 5592 | qgroup_to_release = block_rsv->qgroup_rsv_size; |
| 5593 | } |
| 5594 | block_rsv->size -= num_bytes; |
| 5595 | if (block_rsv->reserved >= block_rsv->size) { |
| 5596 | num_bytes = block_rsv->reserved - block_rsv->size; |
| 5597 | block_rsv->reserved = block_rsv->size; |
| 5598 | block_rsv->full = 1; |
| 5599 | } else { |
| 5600 | num_bytes = 0; |
| 5601 | } |
| 5602 | if (block_rsv->qgroup_rsv_reserved >= block_rsv->qgroup_rsv_size) { |
| 5603 | qgroup_to_release = block_rsv->qgroup_rsv_reserved - |
| 5604 | block_rsv->qgroup_rsv_size; |
| 5605 | block_rsv->qgroup_rsv_reserved = block_rsv->qgroup_rsv_size; |
| 5606 | } else { |
| 5607 | qgroup_to_release = 0; |
| 5608 | } |
| 5609 | spin_unlock(&block_rsv->lock); |
| 5610 | |
| 5611 | ret = num_bytes; |
| 5612 | if (num_bytes > 0) { |
| 5613 | if (dest) { |
| 5614 | spin_lock(&dest->lock); |
| 5615 | if (!dest->full) { |
| 5616 | u64 bytes_to_add; |
| 5617 | |
| 5618 | bytes_to_add = dest->size - dest->reserved; |
| 5619 | bytes_to_add = min(num_bytes, bytes_to_add); |
| 5620 | dest->reserved += bytes_to_add; |
| 5621 | if (dest->reserved >= dest->size) |
| 5622 | dest->full = 1; |
| 5623 | num_bytes -= bytes_to_add; |
| 5624 | } |
| 5625 | spin_unlock(&dest->lock); |
| 5626 | } |
| 5627 | if (num_bytes) |
| 5628 | space_info_add_old_bytes(fs_info, space_info, |
| 5629 | num_bytes); |
| 5630 | } |
| 5631 | if (qgroup_to_release_ret) |
| 5632 | *qgroup_to_release_ret = qgroup_to_release; |
| 5633 | return ret; |
| 5634 | } |
| 5635 | |
| 5636 | int btrfs_block_rsv_migrate(struct btrfs_block_rsv *src, |
| 5637 | struct btrfs_block_rsv *dst, u64 num_bytes, |
| 5638 | int update_size) |
| 5639 | { |
| 5640 | int ret; |
| 5641 | |
| 5642 | ret = block_rsv_use_bytes(src, num_bytes); |
| 5643 | if (ret) |
| 5644 | return ret; |
| 5645 | |
| 5646 | block_rsv_add_bytes(dst, num_bytes, update_size); |
| 5647 | return 0; |
| 5648 | } |
| 5649 | |
| 5650 | void btrfs_init_block_rsv(struct btrfs_block_rsv *rsv, unsigned short type) |
| 5651 | { |
| 5652 | memset(rsv, 0, sizeof(*rsv)); |
| 5653 | spin_lock_init(&rsv->lock); |
| 5654 | rsv->type = type; |
| 5655 | } |
| 5656 | |
| 5657 | void btrfs_init_metadata_block_rsv(struct btrfs_fs_info *fs_info, |
| 5658 | struct btrfs_block_rsv *rsv, |
| 5659 | unsigned short type) |
| 5660 | { |
| 5661 | btrfs_init_block_rsv(rsv, type); |
| 5662 | rsv->space_info = __find_space_info(fs_info, |
| 5663 | BTRFS_BLOCK_GROUP_METADATA); |
| 5664 | } |
| 5665 | |
| 5666 | struct btrfs_block_rsv *btrfs_alloc_block_rsv(struct btrfs_fs_info *fs_info, |
| 5667 | unsigned short type) |
| 5668 | { |
| 5669 | struct btrfs_block_rsv *block_rsv; |
| 5670 | |
| 5671 | block_rsv = kmalloc(sizeof(*block_rsv), GFP_NOFS); |
| 5672 | if (!block_rsv) |
| 5673 | return NULL; |
| 5674 | |
| 5675 | btrfs_init_metadata_block_rsv(fs_info, block_rsv, type); |
| 5676 | return block_rsv; |
| 5677 | } |
| 5678 | |
| 5679 | void btrfs_free_block_rsv(struct btrfs_fs_info *fs_info, |
| 5680 | struct btrfs_block_rsv *rsv) |
| 5681 | { |
| 5682 | if (!rsv) |
| 5683 | return; |
| 5684 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
| 5685 | kfree(rsv); |
| 5686 | } |
| 5687 | |
| 5688 | void __btrfs_free_block_rsv(struct btrfs_block_rsv *rsv) |
| 5689 | { |
| 5690 | kfree(rsv); |
| 5691 | } |
| 5692 | |
| 5693 | int btrfs_block_rsv_add(struct btrfs_root *root, |
| 5694 | struct btrfs_block_rsv *block_rsv, u64 num_bytes, |
| 5695 | enum btrfs_reserve_flush_enum flush) |
| 5696 | { |
| 5697 | int ret; |
| 5698 | |
| 5699 | if (num_bytes == 0) |
| 5700 | return 0; |
| 5701 | |
| 5702 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
| 5703 | if (!ret) { |
| 5704 | block_rsv_add_bytes(block_rsv, num_bytes, 1); |
| 5705 | return 0; |
| 5706 | } |
| 5707 | |
| 5708 | return ret; |
| 5709 | } |
| 5710 | |
| 5711 | int btrfs_block_rsv_check(struct btrfs_block_rsv *block_rsv, int min_factor) |
| 5712 | { |
| 5713 | u64 num_bytes = 0; |
| 5714 | int ret = -ENOSPC; |
| 5715 | |
| 5716 | if (!block_rsv) |
| 5717 | return 0; |
| 5718 | |
| 5719 | spin_lock(&block_rsv->lock); |
| 5720 | num_bytes = div_factor(block_rsv->size, min_factor); |
| 5721 | if (block_rsv->reserved >= num_bytes) |
| 5722 | ret = 0; |
| 5723 | spin_unlock(&block_rsv->lock); |
| 5724 | |
| 5725 | return ret; |
| 5726 | } |
| 5727 | |
| 5728 | int btrfs_block_rsv_refill(struct btrfs_root *root, |
| 5729 | struct btrfs_block_rsv *block_rsv, u64 min_reserved, |
| 5730 | enum btrfs_reserve_flush_enum flush) |
| 5731 | { |
| 5732 | u64 num_bytes = 0; |
| 5733 | int ret = -ENOSPC; |
| 5734 | |
| 5735 | if (!block_rsv) |
| 5736 | return 0; |
| 5737 | |
| 5738 | spin_lock(&block_rsv->lock); |
| 5739 | num_bytes = min_reserved; |
| 5740 | if (block_rsv->reserved >= num_bytes) |
| 5741 | ret = 0; |
| 5742 | else |
| 5743 | num_bytes -= block_rsv->reserved; |
| 5744 | spin_unlock(&block_rsv->lock); |
| 5745 | |
| 5746 | if (!ret) |
| 5747 | return 0; |
| 5748 | |
| 5749 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
| 5750 | if (!ret) { |
| 5751 | block_rsv_add_bytes(block_rsv, num_bytes, 0); |
| 5752 | return 0; |
| 5753 | } |
| 5754 | |
| 5755 | return ret; |
| 5756 | } |
| 5757 | |
| 5758 | /** |
| 5759 | * btrfs_inode_rsv_refill - refill the inode block rsv. |
| 5760 | * @inode - the inode we are refilling. |
| 5761 | * @flush - the flusing restriction. |
| 5762 | * |
| 5763 | * Essentially the same as btrfs_block_rsv_refill, except it uses the |
| 5764 | * block_rsv->size as the minimum size. We'll either refill the missing amount |
| 5765 | * or return if we already have enough space. This will also handle the resreve |
| 5766 | * tracepoint for the reserved amount. |
| 5767 | */ |
| 5768 | static int btrfs_inode_rsv_refill(struct btrfs_inode *inode, |
| 5769 | enum btrfs_reserve_flush_enum flush) |
| 5770 | { |
| 5771 | struct btrfs_root *root = inode->root; |
| 5772 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
| 5773 | u64 num_bytes = 0; |
| 5774 | u64 qgroup_num_bytes = 0; |
| 5775 | int ret = -ENOSPC; |
| 5776 | |
| 5777 | spin_lock(&block_rsv->lock); |
| 5778 | if (block_rsv->reserved < block_rsv->size) |
| 5779 | num_bytes = block_rsv->size - block_rsv->reserved; |
| 5780 | if (block_rsv->qgroup_rsv_reserved < block_rsv->qgroup_rsv_size) |
| 5781 | qgroup_num_bytes = block_rsv->qgroup_rsv_size - |
| 5782 | block_rsv->qgroup_rsv_reserved; |
| 5783 | spin_unlock(&block_rsv->lock); |
| 5784 | |
| 5785 | if (num_bytes == 0) |
| 5786 | return 0; |
| 5787 | |
| 5788 | ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_num_bytes, true); |
| 5789 | if (ret) |
| 5790 | return ret; |
| 5791 | ret = reserve_metadata_bytes(root, block_rsv, num_bytes, flush); |
| 5792 | if (!ret) { |
| 5793 | block_rsv_add_bytes(block_rsv, num_bytes, 0); |
| 5794 | trace_btrfs_space_reservation(root->fs_info, "delalloc", |
| 5795 | btrfs_ino(inode), num_bytes, 1); |
| 5796 | |
| 5797 | /* Don't forget to increase qgroup_rsv_reserved */ |
| 5798 | spin_lock(&block_rsv->lock); |
| 5799 | block_rsv->qgroup_rsv_reserved += qgroup_num_bytes; |
| 5800 | spin_unlock(&block_rsv->lock); |
| 5801 | } else |
| 5802 | btrfs_qgroup_free_meta_prealloc(root, qgroup_num_bytes); |
| 5803 | return ret; |
| 5804 | } |
| 5805 | |
| 5806 | /** |
| 5807 | * btrfs_inode_rsv_release - release any excessive reservation. |
| 5808 | * @inode - the inode we need to release from. |
| 5809 | * @qgroup_free - free or convert qgroup meta. |
| 5810 | * Unlike normal operation, qgroup meta reservation needs to know if we are |
| 5811 | * freeing qgroup reservation or just converting it into per-trans. Normally |
| 5812 | * @qgroup_free is true for error handling, and false for normal release. |
| 5813 | * |
| 5814 | * This is the same as btrfs_block_rsv_release, except that it handles the |
| 5815 | * tracepoint for the reservation. |
| 5816 | */ |
| 5817 | static void btrfs_inode_rsv_release(struct btrfs_inode *inode, bool qgroup_free) |
| 5818 | { |
| 5819 | struct btrfs_fs_info *fs_info = inode->root->fs_info; |
| 5820 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
| 5821 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
| 5822 | u64 released = 0; |
| 5823 | u64 qgroup_to_release = 0; |
| 5824 | |
| 5825 | /* |
| 5826 | * Since we statically set the block_rsv->size we just want to say we |
| 5827 | * are releasing 0 bytes, and then we'll just get the reservation over |
| 5828 | * the size free'd. |
| 5829 | */ |
| 5830 | released = block_rsv_release_bytes(fs_info, block_rsv, global_rsv, 0, |
| 5831 | &qgroup_to_release); |
| 5832 | if (released > 0) |
| 5833 | trace_btrfs_space_reservation(fs_info, "delalloc", |
| 5834 | btrfs_ino(inode), released, 0); |
| 5835 | if (qgroup_free) |
| 5836 | btrfs_qgroup_free_meta_prealloc(inode->root, qgroup_to_release); |
| 5837 | else |
| 5838 | btrfs_qgroup_convert_reserved_meta(inode->root, |
| 5839 | qgroup_to_release); |
| 5840 | } |
| 5841 | |
| 5842 | void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info, |
| 5843 | struct btrfs_block_rsv *block_rsv, |
| 5844 | u64 num_bytes) |
| 5845 | { |
| 5846 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
| 5847 | |
| 5848 | if (global_rsv == block_rsv || |
| 5849 | block_rsv->space_info != global_rsv->space_info) |
| 5850 | global_rsv = NULL; |
| 5851 | block_rsv_release_bytes(fs_info, block_rsv, global_rsv, num_bytes, NULL); |
| 5852 | } |
| 5853 | |
| 5854 | static void update_global_block_rsv(struct btrfs_fs_info *fs_info) |
| 5855 | { |
| 5856 | struct btrfs_block_rsv *block_rsv = &fs_info->global_block_rsv; |
| 5857 | struct btrfs_space_info *sinfo = block_rsv->space_info; |
| 5858 | u64 num_bytes; |
| 5859 | |
| 5860 | /* |
| 5861 | * The global block rsv is based on the size of the extent tree, the |
| 5862 | * checksum tree and the root tree. If the fs is empty we want to set |
| 5863 | * it to a minimal amount for safety. |
| 5864 | */ |
| 5865 | num_bytes = btrfs_root_used(&fs_info->extent_root->root_item) + |
| 5866 | btrfs_root_used(&fs_info->csum_root->root_item) + |
| 5867 | btrfs_root_used(&fs_info->tree_root->root_item); |
| 5868 | num_bytes = max_t(u64, num_bytes, SZ_16M); |
| 5869 | |
| 5870 | spin_lock(&sinfo->lock); |
| 5871 | spin_lock(&block_rsv->lock); |
| 5872 | |
| 5873 | block_rsv->size = min_t(u64, num_bytes, SZ_512M); |
| 5874 | |
| 5875 | if (block_rsv->reserved < block_rsv->size) { |
| 5876 | num_bytes = btrfs_space_info_used(sinfo, true); |
| 5877 | if (sinfo->total_bytes > num_bytes) { |
| 5878 | num_bytes = sinfo->total_bytes - num_bytes; |
| 5879 | num_bytes = min(num_bytes, |
| 5880 | block_rsv->size - block_rsv->reserved); |
| 5881 | block_rsv->reserved += num_bytes; |
| 5882 | sinfo->bytes_may_use += num_bytes; |
| 5883 | trace_btrfs_space_reservation(fs_info, "space_info", |
| 5884 | sinfo->flags, num_bytes, |
| 5885 | 1); |
| 5886 | } |
| 5887 | } else if (block_rsv->reserved > block_rsv->size) { |
| 5888 | num_bytes = block_rsv->reserved - block_rsv->size; |
| 5889 | sinfo->bytes_may_use -= num_bytes; |
| 5890 | trace_btrfs_space_reservation(fs_info, "space_info", |
| 5891 | sinfo->flags, num_bytes, 0); |
| 5892 | block_rsv->reserved = block_rsv->size; |
| 5893 | } |
| 5894 | |
| 5895 | if (block_rsv->reserved == block_rsv->size) |
| 5896 | block_rsv->full = 1; |
| 5897 | else |
| 5898 | block_rsv->full = 0; |
| 5899 | |
| 5900 | spin_unlock(&block_rsv->lock); |
| 5901 | spin_unlock(&sinfo->lock); |
| 5902 | } |
| 5903 | |
| 5904 | static void init_global_block_rsv(struct btrfs_fs_info *fs_info) |
| 5905 | { |
| 5906 | struct btrfs_space_info *space_info; |
| 5907 | |
| 5908 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_SYSTEM); |
| 5909 | fs_info->chunk_block_rsv.space_info = space_info; |
| 5910 | |
| 5911 | space_info = __find_space_info(fs_info, BTRFS_BLOCK_GROUP_METADATA); |
| 5912 | fs_info->global_block_rsv.space_info = space_info; |
| 5913 | fs_info->trans_block_rsv.space_info = space_info; |
| 5914 | fs_info->empty_block_rsv.space_info = space_info; |
| 5915 | fs_info->delayed_block_rsv.space_info = space_info; |
| 5916 | |
| 5917 | fs_info->extent_root->block_rsv = &fs_info->global_block_rsv; |
| 5918 | fs_info->csum_root->block_rsv = &fs_info->global_block_rsv; |
| 5919 | fs_info->dev_root->block_rsv = &fs_info->global_block_rsv; |
| 5920 | fs_info->tree_root->block_rsv = &fs_info->global_block_rsv; |
| 5921 | if (fs_info->quota_root) |
| 5922 | fs_info->quota_root->block_rsv = &fs_info->global_block_rsv; |
| 5923 | fs_info->chunk_root->block_rsv = &fs_info->chunk_block_rsv; |
| 5924 | |
| 5925 | update_global_block_rsv(fs_info); |
| 5926 | } |
| 5927 | |
| 5928 | static void release_global_block_rsv(struct btrfs_fs_info *fs_info) |
| 5929 | { |
| 5930 | block_rsv_release_bytes(fs_info, &fs_info->global_block_rsv, NULL, |
| 5931 | (u64)-1, NULL); |
| 5932 | WARN_ON(fs_info->trans_block_rsv.size > 0); |
| 5933 | WARN_ON(fs_info->trans_block_rsv.reserved > 0); |
| 5934 | WARN_ON(fs_info->chunk_block_rsv.size > 0); |
| 5935 | WARN_ON(fs_info->chunk_block_rsv.reserved > 0); |
| 5936 | WARN_ON(fs_info->delayed_block_rsv.size > 0); |
| 5937 | WARN_ON(fs_info->delayed_block_rsv.reserved > 0); |
| 5938 | } |
| 5939 | |
| 5940 | |
| 5941 | /* |
| 5942 | * To be called after all the new block groups attached to the transaction |
| 5943 | * handle have been created (btrfs_create_pending_block_groups()). |
| 5944 | */ |
| 5945 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) |
| 5946 | { |
| 5947 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 5948 | |
| 5949 | if (!trans->chunk_bytes_reserved) |
| 5950 | return; |
| 5951 | |
| 5952 | WARN_ON_ONCE(!list_empty(&trans->new_bgs)); |
| 5953 | |
| 5954 | block_rsv_release_bytes(fs_info, &fs_info->chunk_block_rsv, NULL, |
| 5955 | trans->chunk_bytes_reserved, NULL); |
| 5956 | trans->chunk_bytes_reserved = 0; |
| 5957 | } |
| 5958 | |
| 5959 | /* |
| 5960 | * btrfs_subvolume_reserve_metadata() - reserve space for subvolume operation |
| 5961 | * root: the root of the parent directory |
| 5962 | * rsv: block reservation |
| 5963 | * items: the number of items that we need do reservation |
| 5964 | * qgroup_reserved: used to return the reserved size in qgroup |
| 5965 | * |
| 5966 | * This function is used to reserve the space for snapshot/subvolume |
| 5967 | * creation and deletion. Those operations are different with the |
| 5968 | * common file/directory operations, they change two fs/file trees |
| 5969 | * and root tree, the number of items that the qgroup reserves is |
| 5970 | * different with the free space reservation. So we can not use |
| 5971 | * the space reservation mechanism in start_transaction(). |
| 5972 | */ |
| 5973 | int btrfs_subvolume_reserve_metadata(struct btrfs_root *root, |
| 5974 | struct btrfs_block_rsv *rsv, |
| 5975 | int items, |
| 5976 | bool use_global_rsv) |
| 5977 | { |
| 5978 | u64 num_bytes; |
| 5979 | int ret; |
| 5980 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 5981 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
| 5982 | |
| 5983 | if (test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) { |
| 5984 | /* One for parent inode, two for dir entries */ |
| 5985 | num_bytes = 3 * fs_info->nodesize; |
| 5986 | ret = btrfs_qgroup_reserve_meta_prealloc(root, num_bytes, true); |
| 5987 | if (ret) |
| 5988 | return ret; |
| 5989 | } else { |
| 5990 | num_bytes = 0; |
| 5991 | } |
| 5992 | |
| 5993 | num_bytes = btrfs_calc_trans_metadata_size(fs_info, items); |
| 5994 | rsv->space_info = __find_space_info(fs_info, |
| 5995 | BTRFS_BLOCK_GROUP_METADATA); |
| 5996 | ret = btrfs_block_rsv_add(root, rsv, num_bytes, |
| 5997 | BTRFS_RESERVE_FLUSH_ALL); |
| 5998 | |
| 5999 | if (ret == -ENOSPC && use_global_rsv) |
| 6000 | ret = btrfs_block_rsv_migrate(global_rsv, rsv, num_bytes, 1); |
| 6001 | |
| 6002 | if (ret && num_bytes) |
| 6003 | btrfs_qgroup_free_meta_prealloc(root, num_bytes); |
| 6004 | |
| 6005 | return ret; |
| 6006 | } |
| 6007 | |
| 6008 | void btrfs_subvolume_release_metadata(struct btrfs_fs_info *fs_info, |
| 6009 | struct btrfs_block_rsv *rsv) |
| 6010 | { |
| 6011 | btrfs_block_rsv_release(fs_info, rsv, (u64)-1); |
| 6012 | } |
| 6013 | |
| 6014 | static void btrfs_calculate_inode_block_rsv_size(struct btrfs_fs_info *fs_info, |
| 6015 | struct btrfs_inode *inode) |
| 6016 | { |
| 6017 | struct btrfs_block_rsv *block_rsv = &inode->block_rsv; |
| 6018 | u64 reserve_size = 0; |
| 6019 | u64 qgroup_rsv_size = 0; |
| 6020 | u64 csum_leaves; |
| 6021 | unsigned outstanding_extents; |
| 6022 | |
| 6023 | lockdep_assert_held(&inode->lock); |
| 6024 | outstanding_extents = inode->outstanding_extents; |
| 6025 | if (outstanding_extents) |
| 6026 | reserve_size = btrfs_calc_trans_metadata_size(fs_info, |
| 6027 | outstanding_extents + 1); |
| 6028 | csum_leaves = btrfs_csum_bytes_to_leaves(fs_info, |
| 6029 | inode->csum_bytes); |
| 6030 | reserve_size += btrfs_calc_trans_metadata_size(fs_info, |
| 6031 | csum_leaves); |
| 6032 | /* |
| 6033 | * For qgroup rsv, the calculation is very simple: |
| 6034 | * account one nodesize for each outstanding extent |
| 6035 | * |
| 6036 | * This is overestimating in most cases. |
| 6037 | */ |
| 6038 | qgroup_rsv_size = outstanding_extents * fs_info->nodesize; |
| 6039 | |
| 6040 | spin_lock(&block_rsv->lock); |
| 6041 | block_rsv->size = reserve_size; |
| 6042 | block_rsv->qgroup_rsv_size = qgroup_rsv_size; |
| 6043 | spin_unlock(&block_rsv->lock); |
| 6044 | } |
| 6045 | |
| 6046 | int btrfs_delalloc_reserve_metadata(struct btrfs_inode *inode, u64 num_bytes) |
| 6047 | { |
| 6048 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
| 6049 | unsigned nr_extents; |
| 6050 | enum btrfs_reserve_flush_enum flush = BTRFS_RESERVE_FLUSH_ALL; |
| 6051 | int ret = 0; |
| 6052 | bool delalloc_lock = true; |
| 6053 | |
| 6054 | /* If we are a free space inode we need to not flush since we will be in |
| 6055 | * the middle of a transaction commit. We also don't need the delalloc |
| 6056 | * mutex since we won't race with anybody. We need this mostly to make |
| 6057 | * lockdep shut its filthy mouth. |
| 6058 | * |
| 6059 | * If we have a transaction open (can happen if we call truncate_block |
| 6060 | * from truncate), then we need FLUSH_LIMIT so we don't deadlock. |
| 6061 | */ |
| 6062 | if (btrfs_is_free_space_inode(inode)) { |
| 6063 | flush = BTRFS_RESERVE_NO_FLUSH; |
| 6064 | delalloc_lock = false; |
| 6065 | } else { |
| 6066 | if (current->journal_info) |
| 6067 | flush = BTRFS_RESERVE_FLUSH_LIMIT; |
| 6068 | |
| 6069 | if (btrfs_transaction_in_commit(fs_info)) |
| 6070 | schedule_timeout(1); |
| 6071 | } |
| 6072 | |
| 6073 | if (delalloc_lock) |
| 6074 | mutex_lock(&inode->delalloc_mutex); |
| 6075 | |
| 6076 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
| 6077 | |
| 6078 | /* Add our new extents and calculate the new rsv size. */ |
| 6079 | spin_lock(&inode->lock); |
| 6080 | nr_extents = count_max_extents(num_bytes); |
| 6081 | btrfs_mod_outstanding_extents(inode, nr_extents); |
| 6082 | inode->csum_bytes += num_bytes; |
| 6083 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); |
| 6084 | spin_unlock(&inode->lock); |
| 6085 | |
| 6086 | ret = btrfs_inode_rsv_refill(inode, flush); |
| 6087 | if (unlikely(ret)) |
| 6088 | goto out_fail; |
| 6089 | |
| 6090 | if (delalloc_lock) |
| 6091 | mutex_unlock(&inode->delalloc_mutex); |
| 6092 | return 0; |
| 6093 | |
| 6094 | out_fail: |
| 6095 | spin_lock(&inode->lock); |
| 6096 | nr_extents = count_max_extents(num_bytes); |
| 6097 | btrfs_mod_outstanding_extents(inode, -nr_extents); |
| 6098 | inode->csum_bytes -= num_bytes; |
| 6099 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); |
| 6100 | spin_unlock(&inode->lock); |
| 6101 | |
| 6102 | btrfs_inode_rsv_release(inode, true); |
| 6103 | if (delalloc_lock) |
| 6104 | mutex_unlock(&inode->delalloc_mutex); |
| 6105 | return ret; |
| 6106 | } |
| 6107 | |
| 6108 | /** |
| 6109 | * btrfs_delalloc_release_metadata - release a metadata reservation for an inode |
| 6110 | * @inode: the inode to release the reservation for. |
| 6111 | * @num_bytes: the number of bytes we are releasing. |
| 6112 | * @qgroup_free: free qgroup reservation or convert it to per-trans reservation |
| 6113 | * |
| 6114 | * This will release the metadata reservation for an inode. This can be called |
| 6115 | * once we complete IO for a given set of bytes to release their metadata |
| 6116 | * reservations, or on error for the same reason. |
| 6117 | */ |
| 6118 | void btrfs_delalloc_release_metadata(struct btrfs_inode *inode, u64 num_bytes, |
| 6119 | bool qgroup_free) |
| 6120 | { |
| 6121 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
| 6122 | |
| 6123 | num_bytes = ALIGN(num_bytes, fs_info->sectorsize); |
| 6124 | spin_lock(&inode->lock); |
| 6125 | inode->csum_bytes -= num_bytes; |
| 6126 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); |
| 6127 | spin_unlock(&inode->lock); |
| 6128 | |
| 6129 | if (btrfs_is_testing(fs_info)) |
| 6130 | return; |
| 6131 | |
| 6132 | btrfs_inode_rsv_release(inode, qgroup_free); |
| 6133 | } |
| 6134 | |
| 6135 | /** |
| 6136 | * btrfs_delalloc_release_extents - release our outstanding_extents |
| 6137 | * @inode: the inode to balance the reservation for. |
| 6138 | * @num_bytes: the number of bytes we originally reserved with |
| 6139 | * @qgroup_free: do we need to free qgroup meta reservation or convert them. |
| 6140 | * |
| 6141 | * When we reserve space we increase outstanding_extents for the extents we may |
| 6142 | * add. Once we've set the range as delalloc or created our ordered extents we |
| 6143 | * have outstanding_extents to track the real usage, so we use this to free our |
| 6144 | * temporarily tracked outstanding_extents. This _must_ be used in conjunction |
| 6145 | * with btrfs_delalloc_reserve_metadata. |
| 6146 | */ |
| 6147 | void btrfs_delalloc_release_extents(struct btrfs_inode *inode, u64 num_bytes, |
| 6148 | bool qgroup_free) |
| 6149 | { |
| 6150 | struct btrfs_fs_info *fs_info = btrfs_sb(inode->vfs_inode.i_sb); |
| 6151 | unsigned num_extents; |
| 6152 | |
| 6153 | spin_lock(&inode->lock); |
| 6154 | num_extents = count_max_extents(num_bytes); |
| 6155 | btrfs_mod_outstanding_extents(inode, -num_extents); |
| 6156 | btrfs_calculate_inode_block_rsv_size(fs_info, inode); |
| 6157 | spin_unlock(&inode->lock); |
| 6158 | |
| 6159 | if (btrfs_is_testing(fs_info)) |
| 6160 | return; |
| 6161 | |
| 6162 | btrfs_inode_rsv_release(inode, qgroup_free); |
| 6163 | } |
| 6164 | |
| 6165 | /** |
| 6166 | * btrfs_delalloc_reserve_space - reserve data and metadata space for |
| 6167 | * delalloc |
| 6168 | * @inode: inode we're writing to |
| 6169 | * @start: start range we are writing to |
| 6170 | * @len: how long the range we are writing to |
| 6171 | * @reserved: mandatory parameter, record actually reserved qgroup ranges of |
| 6172 | * current reservation. |
| 6173 | * |
| 6174 | * This will do the following things |
| 6175 | * |
| 6176 | * o reserve space in data space info for num bytes |
| 6177 | * and reserve precious corresponding qgroup space |
| 6178 | * (Done in check_data_free_space) |
| 6179 | * |
| 6180 | * o reserve space for metadata space, based on the number of outstanding |
| 6181 | * extents and how much csums will be needed |
| 6182 | * also reserve metadata space in a per root over-reserve method. |
| 6183 | * o add to the inodes->delalloc_bytes |
| 6184 | * o add it to the fs_info's delalloc inodes list. |
| 6185 | * (Above 3 all done in delalloc_reserve_metadata) |
| 6186 | * |
| 6187 | * Return 0 for success |
| 6188 | * Return <0 for error(-ENOSPC or -EQUOT) |
| 6189 | */ |
| 6190 | int btrfs_delalloc_reserve_space(struct inode *inode, |
| 6191 | struct extent_changeset **reserved, u64 start, u64 len) |
| 6192 | { |
| 6193 | int ret; |
| 6194 | |
| 6195 | ret = btrfs_check_data_free_space(inode, reserved, start, len); |
| 6196 | if (ret < 0) |
| 6197 | return ret; |
| 6198 | ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), len); |
| 6199 | if (ret < 0) |
| 6200 | btrfs_free_reserved_data_space(inode, *reserved, start, len); |
| 6201 | return ret; |
| 6202 | } |
| 6203 | |
| 6204 | /** |
| 6205 | * btrfs_delalloc_release_space - release data and metadata space for delalloc |
| 6206 | * @inode: inode we're releasing space for |
| 6207 | * @start: start position of the space already reserved |
| 6208 | * @len: the len of the space already reserved |
| 6209 | * @release_bytes: the len of the space we consumed or didn't use |
| 6210 | * |
| 6211 | * This function will release the metadata space that was not used and will |
| 6212 | * decrement ->delalloc_bytes and remove it from the fs_info delalloc_inodes |
| 6213 | * list if there are no delalloc bytes left. |
| 6214 | * Also it will handle the qgroup reserved space. |
| 6215 | */ |
| 6216 | void btrfs_delalloc_release_space(struct inode *inode, |
| 6217 | struct extent_changeset *reserved, |
| 6218 | u64 start, u64 len, bool qgroup_free) |
| 6219 | { |
| 6220 | btrfs_delalloc_release_metadata(BTRFS_I(inode), len, qgroup_free); |
| 6221 | btrfs_free_reserved_data_space(inode, reserved, start, len); |
| 6222 | } |
| 6223 | |
| 6224 | static int update_block_group(struct btrfs_trans_handle *trans, |
| 6225 | struct btrfs_fs_info *info, u64 bytenr, |
| 6226 | u64 num_bytes, int alloc) |
| 6227 | { |
| 6228 | struct btrfs_block_group_cache *cache = NULL; |
| 6229 | u64 total = num_bytes; |
| 6230 | u64 old_val; |
| 6231 | u64 byte_in_group; |
| 6232 | int factor; |
| 6233 | |
| 6234 | /* block accounting for super block */ |
| 6235 | spin_lock(&info->delalloc_root_lock); |
| 6236 | old_val = btrfs_super_bytes_used(info->super_copy); |
| 6237 | if (alloc) |
| 6238 | old_val += num_bytes; |
| 6239 | else |
| 6240 | old_val -= num_bytes; |
| 6241 | btrfs_set_super_bytes_used(info->super_copy, old_val); |
| 6242 | spin_unlock(&info->delalloc_root_lock); |
| 6243 | |
| 6244 | while (total) { |
| 6245 | cache = btrfs_lookup_block_group(info, bytenr); |
| 6246 | if (!cache) |
| 6247 | return -ENOENT; |
| 6248 | if (cache->flags & (BTRFS_BLOCK_GROUP_DUP | |
| 6249 | BTRFS_BLOCK_GROUP_RAID1 | |
| 6250 | BTRFS_BLOCK_GROUP_RAID10)) |
| 6251 | factor = 2; |
| 6252 | else |
| 6253 | factor = 1; |
| 6254 | /* |
| 6255 | * If this block group has free space cache written out, we |
| 6256 | * need to make sure to load it if we are removing space. This |
| 6257 | * is because we need the unpinning stage to actually add the |
| 6258 | * space back to the block group, otherwise we will leak space. |
| 6259 | */ |
| 6260 | if (!alloc && cache->cached == BTRFS_CACHE_NO) |
| 6261 | cache_block_group(cache, 1); |
| 6262 | |
| 6263 | byte_in_group = bytenr - cache->key.objectid; |
| 6264 | WARN_ON(byte_in_group > cache->key.offset); |
| 6265 | |
| 6266 | spin_lock(&cache->space_info->lock); |
| 6267 | spin_lock(&cache->lock); |
| 6268 | |
| 6269 | if (btrfs_test_opt(info, SPACE_CACHE) && |
| 6270 | cache->disk_cache_state < BTRFS_DC_CLEAR) |
| 6271 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
| 6272 | |
| 6273 | old_val = btrfs_block_group_used(&cache->item); |
| 6274 | num_bytes = min(total, cache->key.offset - byte_in_group); |
| 6275 | if (alloc) { |
| 6276 | old_val += num_bytes; |
| 6277 | btrfs_set_block_group_used(&cache->item, old_val); |
| 6278 | cache->reserved -= num_bytes; |
| 6279 | cache->space_info->bytes_reserved -= num_bytes; |
| 6280 | cache->space_info->bytes_used += num_bytes; |
| 6281 | cache->space_info->disk_used += num_bytes * factor; |
| 6282 | spin_unlock(&cache->lock); |
| 6283 | spin_unlock(&cache->space_info->lock); |
| 6284 | } else { |
| 6285 | old_val -= num_bytes; |
| 6286 | btrfs_set_block_group_used(&cache->item, old_val); |
| 6287 | cache->pinned += num_bytes; |
| 6288 | cache->space_info->bytes_pinned += num_bytes; |
| 6289 | cache->space_info->bytes_used -= num_bytes; |
| 6290 | cache->space_info->disk_used -= num_bytes * factor; |
| 6291 | spin_unlock(&cache->lock); |
| 6292 | spin_unlock(&cache->space_info->lock); |
| 6293 | |
| 6294 | trace_btrfs_space_reservation(info, "pinned", |
| 6295 | cache->space_info->flags, |
| 6296 | num_bytes, 1); |
| 6297 | percpu_counter_add(&cache->space_info->total_bytes_pinned, |
| 6298 | num_bytes); |
| 6299 | set_extent_dirty(info->pinned_extents, |
| 6300 | bytenr, bytenr + num_bytes - 1, |
| 6301 | GFP_NOFS | __GFP_NOFAIL); |
| 6302 | } |
| 6303 | |
| 6304 | spin_lock(&trans->transaction->dirty_bgs_lock); |
| 6305 | if (list_empty(&cache->dirty_list)) { |
| 6306 | list_add_tail(&cache->dirty_list, |
| 6307 | &trans->transaction->dirty_bgs); |
| 6308 | trans->transaction->num_dirty_bgs++; |
| 6309 | btrfs_get_block_group(cache); |
| 6310 | } |
| 6311 | spin_unlock(&trans->transaction->dirty_bgs_lock); |
| 6312 | |
| 6313 | /* |
| 6314 | * No longer have used bytes in this block group, queue it for |
| 6315 | * deletion. We do this after adding the block group to the |
| 6316 | * dirty list to avoid races between cleaner kthread and space |
| 6317 | * cache writeout. |
| 6318 | */ |
| 6319 | if (!alloc && old_val == 0) { |
| 6320 | spin_lock(&info->unused_bgs_lock); |
| 6321 | if (list_empty(&cache->bg_list)) { |
| 6322 | btrfs_get_block_group(cache); |
| 6323 | trace_btrfs_add_unused_block_group(cache); |
| 6324 | list_add_tail(&cache->bg_list, |
| 6325 | &info->unused_bgs); |
| 6326 | } |
| 6327 | spin_unlock(&info->unused_bgs_lock); |
| 6328 | } |
| 6329 | |
| 6330 | btrfs_put_block_group(cache); |
| 6331 | total -= num_bytes; |
| 6332 | bytenr += num_bytes; |
| 6333 | } |
| 6334 | return 0; |
| 6335 | } |
| 6336 | |
| 6337 | static u64 first_logical_byte(struct btrfs_fs_info *fs_info, u64 search_start) |
| 6338 | { |
| 6339 | struct btrfs_block_group_cache *cache; |
| 6340 | u64 bytenr; |
| 6341 | |
| 6342 | spin_lock(&fs_info->block_group_cache_lock); |
| 6343 | bytenr = fs_info->first_logical_byte; |
| 6344 | spin_unlock(&fs_info->block_group_cache_lock); |
| 6345 | |
| 6346 | if (bytenr < (u64)-1) |
| 6347 | return bytenr; |
| 6348 | |
| 6349 | cache = btrfs_lookup_first_block_group(fs_info, search_start); |
| 6350 | if (!cache) |
| 6351 | return 0; |
| 6352 | |
| 6353 | bytenr = cache->key.objectid; |
| 6354 | btrfs_put_block_group(cache); |
| 6355 | |
| 6356 | return bytenr; |
| 6357 | } |
| 6358 | |
| 6359 | static int pin_down_extent(struct btrfs_fs_info *fs_info, |
| 6360 | struct btrfs_block_group_cache *cache, |
| 6361 | u64 bytenr, u64 num_bytes, int reserved) |
| 6362 | { |
| 6363 | spin_lock(&cache->space_info->lock); |
| 6364 | spin_lock(&cache->lock); |
| 6365 | cache->pinned += num_bytes; |
| 6366 | cache->space_info->bytes_pinned += num_bytes; |
| 6367 | if (reserved) { |
| 6368 | cache->reserved -= num_bytes; |
| 6369 | cache->space_info->bytes_reserved -= num_bytes; |
| 6370 | } |
| 6371 | spin_unlock(&cache->lock); |
| 6372 | spin_unlock(&cache->space_info->lock); |
| 6373 | |
| 6374 | trace_btrfs_space_reservation(fs_info, "pinned", |
| 6375 | cache->space_info->flags, num_bytes, 1); |
| 6376 | percpu_counter_add(&cache->space_info->total_bytes_pinned, num_bytes); |
| 6377 | set_extent_dirty(fs_info->pinned_extents, bytenr, |
| 6378 | bytenr + num_bytes - 1, GFP_NOFS | __GFP_NOFAIL); |
| 6379 | return 0; |
| 6380 | } |
| 6381 | |
| 6382 | /* |
| 6383 | * this function must be called within transaction |
| 6384 | */ |
| 6385 | int btrfs_pin_extent(struct btrfs_fs_info *fs_info, |
| 6386 | u64 bytenr, u64 num_bytes, int reserved) |
| 6387 | { |
| 6388 | struct btrfs_block_group_cache *cache; |
| 6389 | |
| 6390 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
| 6391 | BUG_ON(!cache); /* Logic error */ |
| 6392 | |
| 6393 | pin_down_extent(fs_info, cache, bytenr, num_bytes, reserved); |
| 6394 | |
| 6395 | btrfs_put_block_group(cache); |
| 6396 | return 0; |
| 6397 | } |
| 6398 | |
| 6399 | /* |
| 6400 | * this function must be called within transaction |
| 6401 | */ |
| 6402 | int btrfs_pin_extent_for_log_replay(struct btrfs_fs_info *fs_info, |
| 6403 | u64 bytenr, u64 num_bytes) |
| 6404 | { |
| 6405 | struct btrfs_block_group_cache *cache; |
| 6406 | int ret; |
| 6407 | |
| 6408 | cache = btrfs_lookup_block_group(fs_info, bytenr); |
| 6409 | if (!cache) |
| 6410 | return -EINVAL; |
| 6411 | |
| 6412 | /* |
| 6413 | * pull in the free space cache (if any) so that our pin |
| 6414 | * removes the free space from the cache. We have load_only set |
| 6415 | * to one because the slow code to read in the free extents does check |
| 6416 | * the pinned extents. |
| 6417 | */ |
| 6418 | cache_block_group(cache, 1); |
| 6419 | |
| 6420 | pin_down_extent(fs_info, cache, bytenr, num_bytes, 0); |
| 6421 | |
| 6422 | /* remove us from the free space cache (if we're there at all) */ |
| 6423 | ret = btrfs_remove_free_space(cache, bytenr, num_bytes); |
| 6424 | btrfs_put_block_group(cache); |
| 6425 | return ret; |
| 6426 | } |
| 6427 | |
| 6428 | static int __exclude_logged_extent(struct btrfs_fs_info *fs_info, |
| 6429 | u64 start, u64 num_bytes) |
| 6430 | { |
| 6431 | int ret; |
| 6432 | struct btrfs_block_group_cache *block_group; |
| 6433 | struct btrfs_caching_control *caching_ctl; |
| 6434 | |
| 6435 | block_group = btrfs_lookup_block_group(fs_info, start); |
| 6436 | if (!block_group) |
| 6437 | return -EINVAL; |
| 6438 | |
| 6439 | cache_block_group(block_group, 0); |
| 6440 | caching_ctl = get_caching_control(block_group); |
| 6441 | |
| 6442 | if (!caching_ctl) { |
| 6443 | /* Logic error */ |
| 6444 | BUG_ON(!block_group_cache_done(block_group)); |
| 6445 | ret = btrfs_remove_free_space(block_group, start, num_bytes); |
| 6446 | } else { |
| 6447 | mutex_lock(&caching_ctl->mutex); |
| 6448 | |
| 6449 | if (start >= caching_ctl->progress) { |
| 6450 | ret = add_excluded_extent(fs_info, start, num_bytes); |
| 6451 | } else if (start + num_bytes <= caching_ctl->progress) { |
| 6452 | ret = btrfs_remove_free_space(block_group, |
| 6453 | start, num_bytes); |
| 6454 | } else { |
| 6455 | num_bytes = caching_ctl->progress - start; |
| 6456 | ret = btrfs_remove_free_space(block_group, |
| 6457 | start, num_bytes); |
| 6458 | if (ret) |
| 6459 | goto out_lock; |
| 6460 | |
| 6461 | num_bytes = (start + num_bytes) - |
| 6462 | caching_ctl->progress; |
| 6463 | start = caching_ctl->progress; |
| 6464 | ret = add_excluded_extent(fs_info, start, num_bytes); |
| 6465 | } |
| 6466 | out_lock: |
| 6467 | mutex_unlock(&caching_ctl->mutex); |
| 6468 | put_caching_control(caching_ctl); |
| 6469 | } |
| 6470 | btrfs_put_block_group(block_group); |
| 6471 | return ret; |
| 6472 | } |
| 6473 | |
| 6474 | int btrfs_exclude_logged_extents(struct btrfs_fs_info *fs_info, |
| 6475 | struct extent_buffer *eb) |
| 6476 | { |
| 6477 | struct btrfs_file_extent_item *item; |
| 6478 | struct btrfs_key key; |
| 6479 | int found_type; |
| 6480 | int i; |
| 6481 | int ret = 0; |
| 6482 | |
| 6483 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) |
| 6484 | return 0; |
| 6485 | |
| 6486 | for (i = 0; i < btrfs_header_nritems(eb); i++) { |
| 6487 | btrfs_item_key_to_cpu(eb, &key, i); |
| 6488 | if (key.type != BTRFS_EXTENT_DATA_KEY) |
| 6489 | continue; |
| 6490 | item = btrfs_item_ptr(eb, i, struct btrfs_file_extent_item); |
| 6491 | found_type = btrfs_file_extent_type(eb, item); |
| 6492 | if (found_type == BTRFS_FILE_EXTENT_INLINE) |
| 6493 | continue; |
| 6494 | if (btrfs_file_extent_disk_bytenr(eb, item) == 0) |
| 6495 | continue; |
| 6496 | key.objectid = btrfs_file_extent_disk_bytenr(eb, item); |
| 6497 | key.offset = btrfs_file_extent_disk_num_bytes(eb, item); |
| 6498 | ret = __exclude_logged_extent(fs_info, key.objectid, key.offset); |
| 6499 | if (ret) |
| 6500 | break; |
| 6501 | } |
| 6502 | |
| 6503 | return ret; |
| 6504 | } |
| 6505 | |
| 6506 | static void |
| 6507 | btrfs_inc_block_group_reservations(struct btrfs_block_group_cache *bg) |
| 6508 | { |
| 6509 | atomic_inc(&bg->reservations); |
| 6510 | } |
| 6511 | |
| 6512 | void btrfs_dec_block_group_reservations(struct btrfs_fs_info *fs_info, |
| 6513 | const u64 start) |
| 6514 | { |
| 6515 | struct btrfs_block_group_cache *bg; |
| 6516 | |
| 6517 | bg = btrfs_lookup_block_group(fs_info, start); |
| 6518 | ASSERT(bg); |
| 6519 | if (atomic_dec_and_test(&bg->reservations)) |
| 6520 | wake_up_var(&bg->reservations); |
| 6521 | btrfs_put_block_group(bg); |
| 6522 | } |
| 6523 | |
| 6524 | void btrfs_wait_block_group_reservations(struct btrfs_block_group_cache *bg) |
| 6525 | { |
| 6526 | struct btrfs_space_info *space_info = bg->space_info; |
| 6527 | |
| 6528 | ASSERT(bg->ro); |
| 6529 | |
| 6530 | if (!(bg->flags & BTRFS_BLOCK_GROUP_DATA)) |
| 6531 | return; |
| 6532 | |
| 6533 | /* |
| 6534 | * Our block group is read only but before we set it to read only, |
| 6535 | * some task might have had allocated an extent from it already, but it |
| 6536 | * has not yet created a respective ordered extent (and added it to a |
| 6537 | * root's list of ordered extents). |
| 6538 | * Therefore wait for any task currently allocating extents, since the |
| 6539 | * block group's reservations counter is incremented while a read lock |
| 6540 | * on the groups' semaphore is held and decremented after releasing |
| 6541 | * the read access on that semaphore and creating the ordered extent. |
| 6542 | */ |
| 6543 | down_write(&space_info->groups_sem); |
| 6544 | up_write(&space_info->groups_sem); |
| 6545 | |
| 6546 | wait_var_event(&bg->reservations, !atomic_read(&bg->reservations)); |
| 6547 | } |
| 6548 | |
| 6549 | /** |
| 6550 | * btrfs_add_reserved_bytes - update the block_group and space info counters |
| 6551 | * @cache: The cache we are manipulating |
| 6552 | * @ram_bytes: The number of bytes of file content, and will be same to |
| 6553 | * @num_bytes except for the compress path. |
| 6554 | * @num_bytes: The number of bytes in question |
| 6555 | * @delalloc: The blocks are allocated for the delalloc write |
| 6556 | * |
| 6557 | * This is called by the allocator when it reserves space. If this is a |
| 6558 | * reservation and the block group has become read only we cannot make the |
| 6559 | * reservation and return -EAGAIN, otherwise this function always succeeds. |
| 6560 | */ |
| 6561 | static int btrfs_add_reserved_bytes(struct btrfs_block_group_cache *cache, |
| 6562 | u64 ram_bytes, u64 num_bytes, int delalloc) |
| 6563 | { |
| 6564 | struct btrfs_space_info *space_info = cache->space_info; |
| 6565 | int ret = 0; |
| 6566 | |
| 6567 | spin_lock(&space_info->lock); |
| 6568 | spin_lock(&cache->lock); |
| 6569 | if (cache->ro) { |
| 6570 | ret = -EAGAIN; |
| 6571 | } else { |
| 6572 | cache->reserved += num_bytes; |
| 6573 | space_info->bytes_reserved += num_bytes; |
| 6574 | |
| 6575 | trace_btrfs_space_reservation(cache->fs_info, |
| 6576 | "space_info", space_info->flags, |
| 6577 | ram_bytes, 0); |
| 6578 | space_info->bytes_may_use -= ram_bytes; |
| 6579 | if (delalloc) |
| 6580 | cache->delalloc_bytes += num_bytes; |
| 6581 | } |
| 6582 | spin_unlock(&cache->lock); |
| 6583 | spin_unlock(&space_info->lock); |
| 6584 | return ret; |
| 6585 | } |
| 6586 | |
| 6587 | /** |
| 6588 | * btrfs_free_reserved_bytes - update the block_group and space info counters |
| 6589 | * @cache: The cache we are manipulating |
| 6590 | * @num_bytes: The number of bytes in question |
| 6591 | * @delalloc: The blocks are allocated for the delalloc write |
| 6592 | * |
| 6593 | * This is called by somebody who is freeing space that was never actually used |
| 6594 | * on disk. For example if you reserve some space for a new leaf in transaction |
| 6595 | * A and before transaction A commits you free that leaf, you call this with |
| 6596 | * reserve set to 0 in order to clear the reservation. |
| 6597 | */ |
| 6598 | |
| 6599 | static int btrfs_free_reserved_bytes(struct btrfs_block_group_cache *cache, |
| 6600 | u64 num_bytes, int delalloc) |
| 6601 | { |
| 6602 | struct btrfs_space_info *space_info = cache->space_info; |
| 6603 | int ret = 0; |
| 6604 | |
| 6605 | spin_lock(&space_info->lock); |
| 6606 | spin_lock(&cache->lock); |
| 6607 | if (cache->ro) |
| 6608 | space_info->bytes_readonly += num_bytes; |
| 6609 | cache->reserved -= num_bytes; |
| 6610 | space_info->bytes_reserved -= num_bytes; |
| 6611 | |
| 6612 | if (delalloc) |
| 6613 | cache->delalloc_bytes -= num_bytes; |
| 6614 | spin_unlock(&cache->lock); |
| 6615 | spin_unlock(&space_info->lock); |
| 6616 | return ret; |
| 6617 | } |
| 6618 | void btrfs_prepare_extent_commit(struct btrfs_fs_info *fs_info) |
| 6619 | { |
| 6620 | struct btrfs_caching_control *next; |
| 6621 | struct btrfs_caching_control *caching_ctl; |
| 6622 | struct btrfs_block_group_cache *cache; |
| 6623 | |
| 6624 | down_write(&fs_info->commit_root_sem); |
| 6625 | |
| 6626 | list_for_each_entry_safe(caching_ctl, next, |
| 6627 | &fs_info->caching_block_groups, list) { |
| 6628 | cache = caching_ctl->block_group; |
| 6629 | if (block_group_cache_done(cache)) { |
| 6630 | cache->last_byte_to_unpin = (u64)-1; |
| 6631 | list_del_init(&caching_ctl->list); |
| 6632 | put_caching_control(caching_ctl); |
| 6633 | } else { |
| 6634 | cache->last_byte_to_unpin = caching_ctl->progress; |
| 6635 | } |
| 6636 | } |
| 6637 | |
| 6638 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
| 6639 | fs_info->pinned_extents = &fs_info->freed_extents[1]; |
| 6640 | else |
| 6641 | fs_info->pinned_extents = &fs_info->freed_extents[0]; |
| 6642 | |
| 6643 | up_write(&fs_info->commit_root_sem); |
| 6644 | |
| 6645 | update_global_block_rsv(fs_info); |
| 6646 | } |
| 6647 | |
| 6648 | /* |
| 6649 | * Returns the free cluster for the given space info and sets empty_cluster to |
| 6650 | * what it should be based on the mount options. |
| 6651 | */ |
| 6652 | static struct btrfs_free_cluster * |
| 6653 | fetch_cluster_info(struct btrfs_fs_info *fs_info, |
| 6654 | struct btrfs_space_info *space_info, u64 *empty_cluster) |
| 6655 | { |
| 6656 | struct btrfs_free_cluster *ret = NULL; |
| 6657 | |
| 6658 | *empty_cluster = 0; |
| 6659 | if (btrfs_mixed_space_info(space_info)) |
| 6660 | return ret; |
| 6661 | |
| 6662 | if (space_info->flags & BTRFS_BLOCK_GROUP_METADATA) { |
| 6663 | ret = &fs_info->meta_alloc_cluster; |
| 6664 | if (btrfs_test_opt(fs_info, SSD)) |
| 6665 | *empty_cluster = SZ_2M; |
| 6666 | else |
| 6667 | *empty_cluster = SZ_64K; |
| 6668 | } else if ((space_info->flags & BTRFS_BLOCK_GROUP_DATA) && |
| 6669 | btrfs_test_opt(fs_info, SSD_SPREAD)) { |
| 6670 | *empty_cluster = SZ_2M; |
| 6671 | ret = &fs_info->data_alloc_cluster; |
| 6672 | } |
| 6673 | |
| 6674 | return ret; |
| 6675 | } |
| 6676 | |
| 6677 | static int unpin_extent_range(struct btrfs_fs_info *fs_info, |
| 6678 | u64 start, u64 end, |
| 6679 | const bool return_free_space) |
| 6680 | { |
| 6681 | struct btrfs_block_group_cache *cache = NULL; |
| 6682 | struct btrfs_space_info *space_info; |
| 6683 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
| 6684 | struct btrfs_free_cluster *cluster = NULL; |
| 6685 | u64 len; |
| 6686 | u64 total_unpinned = 0; |
| 6687 | u64 empty_cluster = 0; |
| 6688 | bool readonly; |
| 6689 | |
| 6690 | while (start <= end) { |
| 6691 | readonly = false; |
| 6692 | if (!cache || |
| 6693 | start >= cache->key.objectid + cache->key.offset) { |
| 6694 | if (cache) |
| 6695 | btrfs_put_block_group(cache); |
| 6696 | total_unpinned = 0; |
| 6697 | cache = btrfs_lookup_block_group(fs_info, start); |
| 6698 | BUG_ON(!cache); /* Logic error */ |
| 6699 | |
| 6700 | cluster = fetch_cluster_info(fs_info, |
| 6701 | cache->space_info, |
| 6702 | &empty_cluster); |
| 6703 | empty_cluster <<= 1; |
| 6704 | } |
| 6705 | |
| 6706 | len = cache->key.objectid + cache->key.offset - start; |
| 6707 | len = min(len, end + 1 - start); |
| 6708 | |
| 6709 | if (start < cache->last_byte_to_unpin) { |
| 6710 | len = min(len, cache->last_byte_to_unpin - start); |
| 6711 | if (return_free_space) |
| 6712 | btrfs_add_free_space(cache, start, len); |
| 6713 | } |
| 6714 | |
| 6715 | start += len; |
| 6716 | total_unpinned += len; |
| 6717 | space_info = cache->space_info; |
| 6718 | |
| 6719 | /* |
| 6720 | * If this space cluster has been marked as fragmented and we've |
| 6721 | * unpinned enough in this block group to potentially allow a |
| 6722 | * cluster to be created inside of it go ahead and clear the |
| 6723 | * fragmented check. |
| 6724 | */ |
| 6725 | if (cluster && cluster->fragmented && |
| 6726 | total_unpinned > empty_cluster) { |
| 6727 | spin_lock(&cluster->lock); |
| 6728 | cluster->fragmented = 0; |
| 6729 | spin_unlock(&cluster->lock); |
| 6730 | } |
| 6731 | |
| 6732 | spin_lock(&space_info->lock); |
| 6733 | spin_lock(&cache->lock); |
| 6734 | cache->pinned -= len; |
| 6735 | space_info->bytes_pinned -= len; |
| 6736 | |
| 6737 | trace_btrfs_space_reservation(fs_info, "pinned", |
| 6738 | space_info->flags, len, 0); |
| 6739 | space_info->max_extent_size = 0; |
| 6740 | percpu_counter_add(&space_info->total_bytes_pinned, -len); |
| 6741 | if (cache->ro) { |
| 6742 | space_info->bytes_readonly += len; |
| 6743 | readonly = true; |
| 6744 | } |
| 6745 | spin_unlock(&cache->lock); |
| 6746 | if (!readonly && return_free_space && |
| 6747 | global_rsv->space_info == space_info) { |
| 6748 | u64 to_add = len; |
| 6749 | |
| 6750 | spin_lock(&global_rsv->lock); |
| 6751 | if (!global_rsv->full) { |
| 6752 | to_add = min(len, global_rsv->size - |
| 6753 | global_rsv->reserved); |
| 6754 | global_rsv->reserved += to_add; |
| 6755 | space_info->bytes_may_use += to_add; |
| 6756 | if (global_rsv->reserved >= global_rsv->size) |
| 6757 | global_rsv->full = 1; |
| 6758 | trace_btrfs_space_reservation(fs_info, |
| 6759 | "space_info", |
| 6760 | space_info->flags, |
| 6761 | to_add, 1); |
| 6762 | len -= to_add; |
| 6763 | } |
| 6764 | spin_unlock(&global_rsv->lock); |
| 6765 | /* Add to any tickets we may have */ |
| 6766 | if (len) |
| 6767 | space_info_add_new_bytes(fs_info, space_info, |
| 6768 | len); |
| 6769 | } |
| 6770 | spin_unlock(&space_info->lock); |
| 6771 | } |
| 6772 | |
| 6773 | if (cache) |
| 6774 | btrfs_put_block_group(cache); |
| 6775 | return 0; |
| 6776 | } |
| 6777 | |
| 6778 | int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans) |
| 6779 | { |
| 6780 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 6781 | struct btrfs_block_group_cache *block_group, *tmp; |
| 6782 | struct list_head *deleted_bgs; |
| 6783 | struct extent_io_tree *unpin; |
| 6784 | u64 start; |
| 6785 | u64 end; |
| 6786 | int ret; |
| 6787 | |
| 6788 | if (fs_info->pinned_extents == &fs_info->freed_extents[0]) |
| 6789 | unpin = &fs_info->freed_extents[1]; |
| 6790 | else |
| 6791 | unpin = &fs_info->freed_extents[0]; |
| 6792 | |
| 6793 | while (!trans->aborted) { |
| 6794 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
| 6795 | ret = find_first_extent_bit(unpin, 0, &start, &end, |
| 6796 | EXTENT_DIRTY, NULL); |
| 6797 | if (ret) { |
| 6798 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
| 6799 | break; |
| 6800 | } |
| 6801 | |
| 6802 | if (btrfs_test_opt(fs_info, DISCARD)) |
| 6803 | ret = btrfs_discard_extent(fs_info, start, |
| 6804 | end + 1 - start, NULL); |
| 6805 | |
| 6806 | clear_extent_dirty(unpin, start, end); |
| 6807 | unpin_extent_range(fs_info, start, end, true); |
| 6808 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
| 6809 | cond_resched(); |
| 6810 | } |
| 6811 | |
| 6812 | /* |
| 6813 | * Transaction is finished. We don't need the lock anymore. We |
| 6814 | * do need to clean up the block groups in case of a transaction |
| 6815 | * abort. |
| 6816 | */ |
| 6817 | deleted_bgs = &trans->transaction->deleted_bgs; |
| 6818 | list_for_each_entry_safe(block_group, tmp, deleted_bgs, bg_list) { |
| 6819 | u64 trimmed = 0; |
| 6820 | |
| 6821 | ret = -EROFS; |
| 6822 | if (!trans->aborted) |
| 6823 | ret = btrfs_discard_extent(fs_info, |
| 6824 | block_group->key.objectid, |
| 6825 | block_group->key.offset, |
| 6826 | &trimmed); |
| 6827 | |
| 6828 | list_del_init(&block_group->bg_list); |
| 6829 | btrfs_put_block_group_trimming(block_group); |
| 6830 | btrfs_put_block_group(block_group); |
| 6831 | |
| 6832 | if (ret) { |
| 6833 | const char *errstr = btrfs_decode_error(ret); |
| 6834 | btrfs_warn(fs_info, |
| 6835 | "discard failed while removing blockgroup: errno=%d %s", |
| 6836 | ret, errstr); |
| 6837 | } |
| 6838 | } |
| 6839 | |
| 6840 | return 0; |
| 6841 | } |
| 6842 | |
| 6843 | static int __btrfs_free_extent(struct btrfs_trans_handle *trans, |
| 6844 | struct btrfs_fs_info *info, |
| 6845 | struct btrfs_delayed_ref_node *node, u64 parent, |
| 6846 | u64 root_objectid, u64 owner_objectid, |
| 6847 | u64 owner_offset, int refs_to_drop, |
| 6848 | struct btrfs_delayed_extent_op *extent_op) |
| 6849 | { |
| 6850 | struct btrfs_key key; |
| 6851 | struct btrfs_path *path; |
| 6852 | struct btrfs_root *extent_root = info->extent_root; |
| 6853 | struct extent_buffer *leaf; |
| 6854 | struct btrfs_extent_item *ei; |
| 6855 | struct btrfs_extent_inline_ref *iref; |
| 6856 | int ret; |
| 6857 | int is_data; |
| 6858 | int extent_slot = 0; |
| 6859 | int found_extent = 0; |
| 6860 | int num_to_del = 1; |
| 6861 | u32 item_size; |
| 6862 | u64 refs; |
| 6863 | u64 bytenr = node->bytenr; |
| 6864 | u64 num_bytes = node->num_bytes; |
| 6865 | int last_ref = 0; |
| 6866 | bool skinny_metadata = btrfs_fs_incompat(info, SKINNY_METADATA); |
| 6867 | |
| 6868 | path = btrfs_alloc_path(); |
| 6869 | if (!path) |
| 6870 | return -ENOMEM; |
| 6871 | |
| 6872 | path->reada = READA_FORWARD; |
| 6873 | path->leave_spinning = 1; |
| 6874 | |
| 6875 | is_data = owner_objectid >= BTRFS_FIRST_FREE_OBJECTID; |
| 6876 | BUG_ON(!is_data && refs_to_drop != 1); |
| 6877 | |
| 6878 | if (is_data) |
| 6879 | skinny_metadata = false; |
| 6880 | |
| 6881 | ret = lookup_extent_backref(trans, info, path, &iref, |
| 6882 | bytenr, num_bytes, parent, |
| 6883 | root_objectid, owner_objectid, |
| 6884 | owner_offset); |
| 6885 | if (ret == 0) { |
| 6886 | extent_slot = path->slots[0]; |
| 6887 | while (extent_slot >= 0) { |
| 6888 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
| 6889 | extent_slot); |
| 6890 | if (key.objectid != bytenr) |
| 6891 | break; |
| 6892 | if (key.type == BTRFS_EXTENT_ITEM_KEY && |
| 6893 | key.offset == num_bytes) { |
| 6894 | found_extent = 1; |
| 6895 | break; |
| 6896 | } |
| 6897 | if (key.type == BTRFS_METADATA_ITEM_KEY && |
| 6898 | key.offset == owner_objectid) { |
| 6899 | found_extent = 1; |
| 6900 | break; |
| 6901 | } |
| 6902 | if (path->slots[0] - extent_slot > 5) |
| 6903 | break; |
| 6904 | extent_slot--; |
| 6905 | } |
| 6906 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 6907 | item_size = btrfs_item_size_nr(path->nodes[0], extent_slot); |
| 6908 | if (found_extent && item_size < sizeof(*ei)) |
| 6909 | found_extent = 0; |
| 6910 | #endif |
| 6911 | if (!found_extent) { |
| 6912 | BUG_ON(iref); |
| 6913 | ret = remove_extent_backref(trans, info, path, NULL, |
| 6914 | refs_to_drop, |
| 6915 | is_data, &last_ref); |
| 6916 | if (ret) { |
| 6917 | btrfs_abort_transaction(trans, ret); |
| 6918 | goto out; |
| 6919 | } |
| 6920 | btrfs_release_path(path); |
| 6921 | path->leave_spinning = 1; |
| 6922 | |
| 6923 | key.objectid = bytenr; |
| 6924 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 6925 | key.offset = num_bytes; |
| 6926 | |
| 6927 | if (!is_data && skinny_metadata) { |
| 6928 | key.type = BTRFS_METADATA_ITEM_KEY; |
| 6929 | key.offset = owner_objectid; |
| 6930 | } |
| 6931 | |
| 6932 | ret = btrfs_search_slot(trans, extent_root, |
| 6933 | &key, path, -1, 1); |
| 6934 | if (ret > 0 && skinny_metadata && path->slots[0]) { |
| 6935 | /* |
| 6936 | * Couldn't find our skinny metadata item, |
| 6937 | * see if we have ye olde extent item. |
| 6938 | */ |
| 6939 | path->slots[0]--; |
| 6940 | btrfs_item_key_to_cpu(path->nodes[0], &key, |
| 6941 | path->slots[0]); |
| 6942 | if (key.objectid == bytenr && |
| 6943 | key.type == BTRFS_EXTENT_ITEM_KEY && |
| 6944 | key.offset == num_bytes) |
| 6945 | ret = 0; |
| 6946 | } |
| 6947 | |
| 6948 | if (ret > 0 && skinny_metadata) { |
| 6949 | skinny_metadata = false; |
| 6950 | key.objectid = bytenr; |
| 6951 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 6952 | key.offset = num_bytes; |
| 6953 | btrfs_release_path(path); |
| 6954 | ret = btrfs_search_slot(trans, extent_root, |
| 6955 | &key, path, -1, 1); |
| 6956 | } |
| 6957 | |
| 6958 | if (ret) { |
| 6959 | btrfs_err(info, |
| 6960 | "umm, got %d back from search, was looking for %llu", |
| 6961 | ret, bytenr); |
| 6962 | if (ret > 0) |
| 6963 | btrfs_print_leaf(path->nodes[0]); |
| 6964 | } |
| 6965 | if (ret < 0) { |
| 6966 | btrfs_abort_transaction(trans, ret); |
| 6967 | goto out; |
| 6968 | } |
| 6969 | extent_slot = path->slots[0]; |
| 6970 | } |
| 6971 | } else if (WARN_ON(ret == -ENOENT)) { |
| 6972 | btrfs_print_leaf(path->nodes[0]); |
| 6973 | btrfs_err(info, |
| 6974 | "unable to find ref byte nr %llu parent %llu root %llu owner %llu offset %llu", |
| 6975 | bytenr, parent, root_objectid, owner_objectid, |
| 6976 | owner_offset); |
| 6977 | btrfs_abort_transaction(trans, ret); |
| 6978 | goto out; |
| 6979 | } else { |
| 6980 | btrfs_abort_transaction(trans, ret); |
| 6981 | goto out; |
| 6982 | } |
| 6983 | |
| 6984 | leaf = path->nodes[0]; |
| 6985 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
| 6986 | #ifdef BTRFS_COMPAT_EXTENT_TREE_V0 |
| 6987 | if (item_size < sizeof(*ei)) { |
| 6988 | BUG_ON(found_extent || extent_slot != path->slots[0]); |
| 6989 | ret = convert_extent_item_v0(trans, info, path, owner_objectid, |
| 6990 | 0); |
| 6991 | if (ret < 0) { |
| 6992 | btrfs_abort_transaction(trans, ret); |
| 6993 | goto out; |
| 6994 | } |
| 6995 | |
| 6996 | btrfs_release_path(path); |
| 6997 | path->leave_spinning = 1; |
| 6998 | |
| 6999 | key.objectid = bytenr; |
| 7000 | key.type = BTRFS_EXTENT_ITEM_KEY; |
| 7001 | key.offset = num_bytes; |
| 7002 | |
| 7003 | ret = btrfs_search_slot(trans, extent_root, &key, path, |
| 7004 | -1, 1); |
| 7005 | if (ret) { |
| 7006 | btrfs_err(info, |
| 7007 | "umm, got %d back from search, was looking for %llu", |
| 7008 | ret, bytenr); |
| 7009 | btrfs_print_leaf(path->nodes[0]); |
| 7010 | } |
| 7011 | if (ret < 0) { |
| 7012 | btrfs_abort_transaction(trans, ret); |
| 7013 | goto out; |
| 7014 | } |
| 7015 | |
| 7016 | extent_slot = path->slots[0]; |
| 7017 | leaf = path->nodes[0]; |
| 7018 | item_size = btrfs_item_size_nr(leaf, extent_slot); |
| 7019 | } |
| 7020 | #endif |
| 7021 | BUG_ON(item_size < sizeof(*ei)); |
| 7022 | ei = btrfs_item_ptr(leaf, extent_slot, |
| 7023 | struct btrfs_extent_item); |
| 7024 | if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID && |
| 7025 | key.type == BTRFS_EXTENT_ITEM_KEY) { |
| 7026 | struct btrfs_tree_block_info *bi; |
| 7027 | BUG_ON(item_size < sizeof(*ei) + sizeof(*bi)); |
| 7028 | bi = (struct btrfs_tree_block_info *)(ei + 1); |
| 7029 | WARN_ON(owner_objectid != btrfs_tree_block_level(leaf, bi)); |
| 7030 | } |
| 7031 | |
| 7032 | refs = btrfs_extent_refs(leaf, ei); |
| 7033 | if (refs < refs_to_drop) { |
| 7034 | btrfs_err(info, |
| 7035 | "trying to drop %d refs but we only have %Lu for bytenr %Lu", |
| 7036 | refs_to_drop, refs, bytenr); |
| 7037 | ret = -EINVAL; |
| 7038 | btrfs_abort_transaction(trans, ret); |
| 7039 | goto out; |
| 7040 | } |
| 7041 | refs -= refs_to_drop; |
| 7042 | |
| 7043 | if (refs > 0) { |
| 7044 | if (extent_op) |
| 7045 | __run_delayed_extent_op(extent_op, leaf, ei); |
| 7046 | /* |
| 7047 | * In the case of inline back ref, reference count will |
| 7048 | * be updated by remove_extent_backref |
| 7049 | */ |
| 7050 | if (iref) { |
| 7051 | BUG_ON(!found_extent); |
| 7052 | } else { |
| 7053 | btrfs_set_extent_refs(leaf, ei, refs); |
| 7054 | btrfs_mark_buffer_dirty(leaf); |
| 7055 | } |
| 7056 | if (found_extent) { |
| 7057 | ret = remove_extent_backref(trans, info, path, |
| 7058 | iref, refs_to_drop, |
| 7059 | is_data, &last_ref); |
| 7060 | if (ret) { |
| 7061 | btrfs_abort_transaction(trans, ret); |
| 7062 | goto out; |
| 7063 | } |
| 7064 | } |
| 7065 | } else { |
| 7066 | if (found_extent) { |
| 7067 | BUG_ON(is_data && refs_to_drop != |
| 7068 | extent_data_ref_count(path, iref)); |
| 7069 | if (iref) { |
| 7070 | BUG_ON(path->slots[0] != extent_slot); |
| 7071 | } else { |
| 7072 | BUG_ON(path->slots[0] != extent_slot + 1); |
| 7073 | path->slots[0] = extent_slot; |
| 7074 | num_to_del = 2; |
| 7075 | } |
| 7076 | } |
| 7077 | |
| 7078 | last_ref = 1; |
| 7079 | ret = btrfs_del_items(trans, extent_root, path, path->slots[0], |
| 7080 | num_to_del); |
| 7081 | if (ret) { |
| 7082 | btrfs_abort_transaction(trans, ret); |
| 7083 | goto out; |
| 7084 | } |
| 7085 | btrfs_release_path(path); |
| 7086 | |
| 7087 | if (is_data) { |
| 7088 | ret = btrfs_del_csums(trans, info, bytenr, num_bytes); |
| 7089 | if (ret) { |
| 7090 | btrfs_abort_transaction(trans, ret); |
| 7091 | goto out; |
| 7092 | } |
| 7093 | } |
| 7094 | |
| 7095 | ret = add_to_free_space_tree(trans, bytenr, num_bytes); |
| 7096 | if (ret) { |
| 7097 | btrfs_abort_transaction(trans, ret); |
| 7098 | goto out; |
| 7099 | } |
| 7100 | |
| 7101 | ret = update_block_group(trans, info, bytenr, num_bytes, 0); |
| 7102 | if (ret) { |
| 7103 | btrfs_abort_transaction(trans, ret); |
| 7104 | goto out; |
| 7105 | } |
| 7106 | } |
| 7107 | btrfs_release_path(path); |
| 7108 | |
| 7109 | out: |
| 7110 | btrfs_free_path(path); |
| 7111 | return ret; |
| 7112 | } |
| 7113 | |
| 7114 | /* |
| 7115 | * when we free an block, it is possible (and likely) that we free the last |
| 7116 | * delayed ref for that extent as well. This searches the delayed ref tree for |
| 7117 | * a given extent, and if there are no other delayed refs to be processed, it |
| 7118 | * removes it from the tree. |
| 7119 | */ |
| 7120 | static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, |
| 7121 | u64 bytenr) |
| 7122 | { |
| 7123 | struct btrfs_delayed_ref_head *head; |
| 7124 | struct btrfs_delayed_ref_root *delayed_refs; |
| 7125 | int ret = 0; |
| 7126 | |
| 7127 | delayed_refs = &trans->transaction->delayed_refs; |
| 7128 | spin_lock(&delayed_refs->lock); |
| 7129 | head = btrfs_find_delayed_ref_head(delayed_refs, bytenr); |
| 7130 | if (!head) |
| 7131 | goto out_delayed_unlock; |
| 7132 | |
| 7133 | spin_lock(&head->lock); |
| 7134 | if (!RB_EMPTY_ROOT(&head->ref_tree)) |
| 7135 | goto out; |
| 7136 | |
| 7137 | if (head->extent_op) { |
| 7138 | if (!head->must_insert_reserved) |
| 7139 | goto out; |
| 7140 | btrfs_free_delayed_extent_op(head->extent_op); |
| 7141 | head->extent_op = NULL; |
| 7142 | } |
| 7143 | |
| 7144 | /* |
| 7145 | * waiting for the lock here would deadlock. If someone else has it |
| 7146 | * locked they are already in the process of dropping it anyway |
| 7147 | */ |
| 7148 | if (!mutex_trylock(&head->mutex)) |
| 7149 | goto out; |
| 7150 | |
| 7151 | /* |
| 7152 | * at this point we have a head with no other entries. Go |
| 7153 | * ahead and process it. |
| 7154 | */ |
| 7155 | rb_erase(&head->href_node, &delayed_refs->href_root); |
| 7156 | RB_CLEAR_NODE(&head->href_node); |
| 7157 | atomic_dec(&delayed_refs->num_entries); |
| 7158 | |
| 7159 | /* |
| 7160 | * we don't take a ref on the node because we're removing it from the |
| 7161 | * tree, so we just steal the ref the tree was holding. |
| 7162 | */ |
| 7163 | delayed_refs->num_heads--; |
| 7164 | if (head->processing == 0) |
| 7165 | delayed_refs->num_heads_ready--; |
| 7166 | head->processing = 0; |
| 7167 | spin_unlock(&head->lock); |
| 7168 | spin_unlock(&delayed_refs->lock); |
| 7169 | |
| 7170 | BUG_ON(head->extent_op); |
| 7171 | if (head->must_insert_reserved) |
| 7172 | ret = 1; |
| 7173 | |
| 7174 | mutex_unlock(&head->mutex); |
| 7175 | btrfs_put_delayed_ref_head(head); |
| 7176 | return ret; |
| 7177 | out: |
| 7178 | spin_unlock(&head->lock); |
| 7179 | |
| 7180 | out_delayed_unlock: |
| 7181 | spin_unlock(&delayed_refs->lock); |
| 7182 | return 0; |
| 7183 | } |
| 7184 | |
| 7185 | void btrfs_free_tree_block(struct btrfs_trans_handle *trans, |
| 7186 | struct btrfs_root *root, |
| 7187 | struct extent_buffer *buf, |
| 7188 | u64 parent, int last_ref) |
| 7189 | { |
| 7190 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 7191 | int pin = 1; |
| 7192 | int ret; |
| 7193 | |
| 7194 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
| 7195 | int old_ref_mod, new_ref_mod; |
| 7196 | |
| 7197 | btrfs_ref_tree_mod(root, buf->start, buf->len, parent, |
| 7198 | root->root_key.objectid, |
| 7199 | btrfs_header_level(buf), 0, |
| 7200 | BTRFS_DROP_DELAYED_REF); |
| 7201 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, buf->start, |
| 7202 | buf->len, parent, |
| 7203 | root->root_key.objectid, |
| 7204 | btrfs_header_level(buf), |
| 7205 | BTRFS_DROP_DELAYED_REF, NULL, |
| 7206 | &old_ref_mod, &new_ref_mod); |
| 7207 | BUG_ON(ret); /* -ENOMEM */ |
| 7208 | pin = old_ref_mod >= 0 && new_ref_mod < 0; |
| 7209 | } |
| 7210 | |
| 7211 | if (last_ref && btrfs_header_generation(buf) == trans->transid) { |
| 7212 | struct btrfs_block_group_cache *cache; |
| 7213 | |
| 7214 | if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) { |
| 7215 | ret = check_ref_cleanup(trans, buf->start); |
| 7216 | if (!ret) |
| 7217 | goto out; |
| 7218 | } |
| 7219 | |
| 7220 | pin = 0; |
| 7221 | cache = btrfs_lookup_block_group(fs_info, buf->start); |
| 7222 | |
| 7223 | if (btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { |
| 7224 | pin_down_extent(fs_info, cache, buf->start, |
| 7225 | buf->len, 1); |
| 7226 | btrfs_put_block_group(cache); |
| 7227 | goto out; |
| 7228 | } |
| 7229 | |
| 7230 | WARN_ON(test_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)); |
| 7231 | |
| 7232 | btrfs_add_free_space(cache, buf->start, buf->len); |
| 7233 | btrfs_free_reserved_bytes(cache, buf->len, 0); |
| 7234 | btrfs_put_block_group(cache); |
| 7235 | trace_btrfs_reserved_extent_free(fs_info, buf->start, buf->len); |
| 7236 | } |
| 7237 | out: |
| 7238 | if (pin) |
| 7239 | add_pinned_bytes(fs_info, buf->len, true, |
| 7240 | root->root_key.objectid); |
| 7241 | |
| 7242 | if (last_ref) { |
| 7243 | /* |
| 7244 | * Deleting the buffer, clear the corrupt flag since it doesn't |
| 7245 | * matter anymore. |
| 7246 | */ |
| 7247 | clear_bit(EXTENT_BUFFER_CORRUPT, &buf->bflags); |
| 7248 | } |
| 7249 | } |
| 7250 | |
| 7251 | /* Can return -ENOMEM */ |
| 7252 | int btrfs_free_extent(struct btrfs_trans_handle *trans, |
| 7253 | struct btrfs_root *root, |
| 7254 | u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, |
| 7255 | u64 owner, u64 offset) |
| 7256 | { |
| 7257 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 7258 | int old_ref_mod, new_ref_mod; |
| 7259 | int ret; |
| 7260 | |
| 7261 | if (btrfs_is_testing(fs_info)) |
| 7262 | return 0; |
| 7263 | |
| 7264 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) |
| 7265 | btrfs_ref_tree_mod(root, bytenr, num_bytes, parent, |
| 7266 | root_objectid, owner, offset, |
| 7267 | BTRFS_DROP_DELAYED_REF); |
| 7268 | |
| 7269 | /* |
| 7270 | * tree log blocks never actually go into the extent allocation |
| 7271 | * tree, just update pinning info and exit early. |
| 7272 | */ |
| 7273 | if (root_objectid == BTRFS_TREE_LOG_OBJECTID) { |
| 7274 | WARN_ON(owner >= BTRFS_FIRST_FREE_OBJECTID); |
| 7275 | /* unlocks the pinned mutex */ |
| 7276 | btrfs_pin_extent(fs_info, bytenr, num_bytes, 1); |
| 7277 | old_ref_mod = new_ref_mod = 0; |
| 7278 | ret = 0; |
| 7279 | } else if (owner < BTRFS_FIRST_FREE_OBJECTID) { |
| 7280 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr, |
| 7281 | num_bytes, parent, |
| 7282 | root_objectid, (int)owner, |
| 7283 | BTRFS_DROP_DELAYED_REF, NULL, |
| 7284 | &old_ref_mod, &new_ref_mod); |
| 7285 | } else { |
| 7286 | ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr, |
| 7287 | num_bytes, parent, |
| 7288 | root_objectid, owner, offset, |
| 7289 | 0, BTRFS_DROP_DELAYED_REF, |
| 7290 | &old_ref_mod, &new_ref_mod); |
| 7291 | } |
| 7292 | |
| 7293 | if (ret == 0 && old_ref_mod >= 0 && new_ref_mod < 0) { |
| 7294 | bool metadata = owner < BTRFS_FIRST_FREE_OBJECTID; |
| 7295 | |
| 7296 | add_pinned_bytes(fs_info, num_bytes, metadata, root_objectid); |
| 7297 | } |
| 7298 | |
| 7299 | return ret; |
| 7300 | } |
| 7301 | |
| 7302 | /* |
| 7303 | * when we wait for progress in the block group caching, its because |
| 7304 | * our allocation attempt failed at least once. So, we must sleep |
| 7305 | * and let some progress happen before we try again. |
| 7306 | * |
| 7307 | * This function will sleep at least once waiting for new free space to |
| 7308 | * show up, and then it will check the block group free space numbers |
| 7309 | * for our min num_bytes. Another option is to have it go ahead |
| 7310 | * and look in the rbtree for a free extent of a given size, but this |
| 7311 | * is a good start. |
| 7312 | * |
| 7313 | * Callers of this must check if cache->cached == BTRFS_CACHE_ERROR before using |
| 7314 | * any of the information in this block group. |
| 7315 | */ |
| 7316 | static noinline void |
| 7317 | wait_block_group_cache_progress(struct btrfs_block_group_cache *cache, |
| 7318 | u64 num_bytes) |
| 7319 | { |
| 7320 | struct btrfs_caching_control *caching_ctl; |
| 7321 | |
| 7322 | caching_ctl = get_caching_control(cache); |
| 7323 | if (!caching_ctl) |
| 7324 | return; |
| 7325 | |
| 7326 | wait_event(caching_ctl->wait, block_group_cache_done(cache) || |
| 7327 | (cache->free_space_ctl->free_space >= num_bytes)); |
| 7328 | |
| 7329 | put_caching_control(caching_ctl); |
| 7330 | } |
| 7331 | |
| 7332 | static noinline int |
| 7333 | wait_block_group_cache_done(struct btrfs_block_group_cache *cache) |
| 7334 | { |
| 7335 | struct btrfs_caching_control *caching_ctl; |
| 7336 | int ret = 0; |
| 7337 | |
| 7338 | caching_ctl = get_caching_control(cache); |
| 7339 | if (!caching_ctl) |
| 7340 | return (cache->cached == BTRFS_CACHE_ERROR) ? -EIO : 0; |
| 7341 | |
| 7342 | wait_event(caching_ctl->wait, block_group_cache_done(cache)); |
| 7343 | if (cache->cached == BTRFS_CACHE_ERROR) |
| 7344 | ret = -EIO; |
| 7345 | put_caching_control(caching_ctl); |
| 7346 | return ret; |
| 7347 | } |
| 7348 | |
| 7349 | enum btrfs_loop_type { |
| 7350 | LOOP_CACHING_NOWAIT = 0, |
| 7351 | LOOP_CACHING_WAIT = 1, |
| 7352 | LOOP_ALLOC_CHUNK = 2, |
| 7353 | LOOP_NO_EMPTY_SIZE = 3, |
| 7354 | }; |
| 7355 | |
| 7356 | static inline void |
| 7357 | btrfs_lock_block_group(struct btrfs_block_group_cache *cache, |
| 7358 | int delalloc) |
| 7359 | { |
| 7360 | if (delalloc) |
| 7361 | down_read(&cache->data_rwsem); |
| 7362 | } |
| 7363 | |
| 7364 | static inline void |
| 7365 | btrfs_grab_block_group(struct btrfs_block_group_cache *cache, |
| 7366 | int delalloc) |
| 7367 | { |
| 7368 | btrfs_get_block_group(cache); |
| 7369 | if (delalloc) |
| 7370 | down_read(&cache->data_rwsem); |
| 7371 | } |
| 7372 | |
| 7373 | static struct btrfs_block_group_cache * |
| 7374 | btrfs_lock_cluster(struct btrfs_block_group_cache *block_group, |
| 7375 | struct btrfs_free_cluster *cluster, |
| 7376 | int delalloc) |
| 7377 | { |
| 7378 | struct btrfs_block_group_cache *used_bg = NULL; |
| 7379 | |
| 7380 | spin_lock(&cluster->refill_lock); |
| 7381 | while (1) { |
| 7382 | used_bg = cluster->block_group; |
| 7383 | if (!used_bg) |
| 7384 | return NULL; |
| 7385 | |
| 7386 | if (used_bg == block_group) |
| 7387 | return used_bg; |
| 7388 | |
| 7389 | btrfs_get_block_group(used_bg); |
| 7390 | |
| 7391 | if (!delalloc) |
| 7392 | return used_bg; |
| 7393 | |
| 7394 | if (down_read_trylock(&used_bg->data_rwsem)) |
| 7395 | return used_bg; |
| 7396 | |
| 7397 | spin_unlock(&cluster->refill_lock); |
| 7398 | |
| 7399 | /* We should only have one-level nested. */ |
| 7400 | down_read_nested(&used_bg->data_rwsem, SINGLE_DEPTH_NESTING); |
| 7401 | |
| 7402 | spin_lock(&cluster->refill_lock); |
| 7403 | if (used_bg == cluster->block_group) |
| 7404 | return used_bg; |
| 7405 | |
| 7406 | up_read(&used_bg->data_rwsem); |
| 7407 | btrfs_put_block_group(used_bg); |
| 7408 | } |
| 7409 | } |
| 7410 | |
| 7411 | static inline void |
| 7412 | btrfs_release_block_group(struct btrfs_block_group_cache *cache, |
| 7413 | int delalloc) |
| 7414 | { |
| 7415 | if (delalloc) |
| 7416 | up_read(&cache->data_rwsem); |
| 7417 | btrfs_put_block_group(cache); |
| 7418 | } |
| 7419 | |
| 7420 | /* |
| 7421 | * walks the btree of allocated extents and find a hole of a given size. |
| 7422 | * The key ins is changed to record the hole: |
| 7423 | * ins->objectid == start position |
| 7424 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
| 7425 | * ins->offset == the size of the hole. |
| 7426 | * Any available blocks before search_start are skipped. |
| 7427 | * |
| 7428 | * If there is no suitable free space, we will record the max size of |
| 7429 | * the free space extent currently. |
| 7430 | */ |
| 7431 | static noinline int find_free_extent(struct btrfs_fs_info *fs_info, |
| 7432 | u64 ram_bytes, u64 num_bytes, u64 empty_size, |
| 7433 | u64 hint_byte, struct btrfs_key *ins, |
| 7434 | u64 flags, int delalloc) |
| 7435 | { |
| 7436 | int ret = 0; |
| 7437 | struct btrfs_root *root = fs_info->extent_root; |
| 7438 | struct btrfs_free_cluster *last_ptr = NULL; |
| 7439 | struct btrfs_block_group_cache *block_group = NULL; |
| 7440 | u64 search_start = 0; |
| 7441 | u64 max_extent_size = 0; |
| 7442 | u64 empty_cluster = 0; |
| 7443 | struct btrfs_space_info *space_info; |
| 7444 | int loop = 0; |
| 7445 | int index = btrfs_bg_flags_to_raid_index(flags); |
| 7446 | bool failed_cluster_refill = false; |
| 7447 | bool failed_alloc = false; |
| 7448 | bool use_cluster = true; |
| 7449 | bool have_caching_bg = false; |
| 7450 | bool orig_have_caching_bg = false; |
| 7451 | bool full_search = false; |
| 7452 | |
| 7453 | WARN_ON(num_bytes < fs_info->sectorsize); |
| 7454 | ins->type = BTRFS_EXTENT_ITEM_KEY; |
| 7455 | ins->objectid = 0; |
| 7456 | ins->offset = 0; |
| 7457 | |
| 7458 | trace_find_free_extent(fs_info, num_bytes, empty_size, flags); |
| 7459 | |
| 7460 | space_info = __find_space_info(fs_info, flags); |
| 7461 | if (!space_info) { |
| 7462 | btrfs_err(fs_info, "No space info for %llu", flags); |
| 7463 | return -ENOSPC; |
| 7464 | } |
| 7465 | |
| 7466 | /* |
| 7467 | * If our free space is heavily fragmented we may not be able to make |
| 7468 | * big contiguous allocations, so instead of doing the expensive search |
| 7469 | * for free space, simply return ENOSPC with our max_extent_size so we |
| 7470 | * can go ahead and search for a more manageable chunk. |
| 7471 | * |
| 7472 | * If our max_extent_size is large enough for our allocation simply |
| 7473 | * disable clustering since we will likely not be able to find enough |
| 7474 | * space to create a cluster and induce latency trying. |
| 7475 | */ |
| 7476 | if (unlikely(space_info->max_extent_size)) { |
| 7477 | spin_lock(&space_info->lock); |
| 7478 | if (space_info->max_extent_size && |
| 7479 | num_bytes > space_info->max_extent_size) { |
| 7480 | ins->offset = space_info->max_extent_size; |
| 7481 | spin_unlock(&space_info->lock); |
| 7482 | return -ENOSPC; |
| 7483 | } else if (space_info->max_extent_size) { |
| 7484 | use_cluster = false; |
| 7485 | } |
| 7486 | spin_unlock(&space_info->lock); |
| 7487 | } |
| 7488 | |
| 7489 | last_ptr = fetch_cluster_info(fs_info, space_info, &empty_cluster); |
| 7490 | if (last_ptr) { |
| 7491 | spin_lock(&last_ptr->lock); |
| 7492 | if (last_ptr->block_group) |
| 7493 | hint_byte = last_ptr->window_start; |
| 7494 | if (last_ptr->fragmented) { |
| 7495 | /* |
| 7496 | * We still set window_start so we can keep track of the |
| 7497 | * last place we found an allocation to try and save |
| 7498 | * some time. |
| 7499 | */ |
| 7500 | hint_byte = last_ptr->window_start; |
| 7501 | use_cluster = false; |
| 7502 | } |
| 7503 | spin_unlock(&last_ptr->lock); |
| 7504 | } |
| 7505 | |
| 7506 | search_start = max(search_start, first_logical_byte(fs_info, 0)); |
| 7507 | search_start = max(search_start, hint_byte); |
| 7508 | if (search_start == hint_byte) { |
| 7509 | block_group = btrfs_lookup_block_group(fs_info, search_start); |
| 7510 | /* |
| 7511 | * we don't want to use the block group if it doesn't match our |
| 7512 | * allocation bits, or if its not cached. |
| 7513 | * |
| 7514 | * However if we are re-searching with an ideal block group |
| 7515 | * picked out then we don't care that the block group is cached. |
| 7516 | */ |
| 7517 | if (block_group && block_group_bits(block_group, flags) && |
| 7518 | block_group->cached != BTRFS_CACHE_NO) { |
| 7519 | down_read(&space_info->groups_sem); |
| 7520 | if (list_empty(&block_group->list) || |
| 7521 | block_group->ro) { |
| 7522 | /* |
| 7523 | * someone is removing this block group, |
| 7524 | * we can't jump into the have_block_group |
| 7525 | * target because our list pointers are not |
| 7526 | * valid |
| 7527 | */ |
| 7528 | btrfs_put_block_group(block_group); |
| 7529 | up_read(&space_info->groups_sem); |
| 7530 | } else { |
| 7531 | index = btrfs_bg_flags_to_raid_index( |
| 7532 | block_group->flags); |
| 7533 | btrfs_lock_block_group(block_group, delalloc); |
| 7534 | goto have_block_group; |
| 7535 | } |
| 7536 | } else if (block_group) { |
| 7537 | btrfs_put_block_group(block_group); |
| 7538 | } |
| 7539 | } |
| 7540 | search: |
| 7541 | have_caching_bg = false; |
| 7542 | if (index == 0 || index == btrfs_bg_flags_to_raid_index(flags)) |
| 7543 | full_search = true; |
| 7544 | down_read(&space_info->groups_sem); |
| 7545 | list_for_each_entry(block_group, &space_info->block_groups[index], |
| 7546 | list) { |
| 7547 | u64 offset; |
| 7548 | int cached; |
| 7549 | |
| 7550 | /* If the block group is read-only, we can skip it entirely. */ |
| 7551 | if (unlikely(block_group->ro)) |
| 7552 | continue; |
| 7553 | |
| 7554 | btrfs_grab_block_group(block_group, delalloc); |
| 7555 | search_start = block_group->key.objectid; |
| 7556 | |
| 7557 | /* |
| 7558 | * this can happen if we end up cycling through all the |
| 7559 | * raid types, but we want to make sure we only allocate |
| 7560 | * for the proper type. |
| 7561 | */ |
| 7562 | if (!block_group_bits(block_group, flags)) { |
| 7563 | u64 extra = BTRFS_BLOCK_GROUP_DUP | |
| 7564 | BTRFS_BLOCK_GROUP_RAID1 | |
| 7565 | BTRFS_BLOCK_GROUP_RAID5 | |
| 7566 | BTRFS_BLOCK_GROUP_RAID6 | |
| 7567 | BTRFS_BLOCK_GROUP_RAID10; |
| 7568 | |
| 7569 | /* |
| 7570 | * if they asked for extra copies and this block group |
| 7571 | * doesn't provide them, bail. This does allow us to |
| 7572 | * fill raid0 from raid1. |
| 7573 | */ |
| 7574 | if ((flags & extra) && !(block_group->flags & extra)) |
| 7575 | goto loop; |
| 7576 | } |
| 7577 | |
| 7578 | have_block_group: |
| 7579 | cached = block_group_cache_done(block_group); |
| 7580 | if (unlikely(!cached)) { |
| 7581 | have_caching_bg = true; |
| 7582 | ret = cache_block_group(block_group, 0); |
| 7583 | BUG_ON(ret < 0); |
| 7584 | ret = 0; |
| 7585 | } |
| 7586 | |
| 7587 | if (unlikely(block_group->cached == BTRFS_CACHE_ERROR)) |
| 7588 | goto loop; |
| 7589 | |
| 7590 | /* |
| 7591 | * Ok we want to try and use the cluster allocator, so |
| 7592 | * lets look there |
| 7593 | */ |
| 7594 | if (last_ptr && use_cluster) { |
| 7595 | struct btrfs_block_group_cache *used_block_group; |
| 7596 | unsigned long aligned_cluster; |
| 7597 | /* |
| 7598 | * the refill lock keeps out other |
| 7599 | * people trying to start a new cluster |
| 7600 | */ |
| 7601 | used_block_group = btrfs_lock_cluster(block_group, |
| 7602 | last_ptr, |
| 7603 | delalloc); |
| 7604 | if (!used_block_group) |
| 7605 | goto refill_cluster; |
| 7606 | |
| 7607 | if (used_block_group != block_group && |
| 7608 | (used_block_group->ro || |
| 7609 | !block_group_bits(used_block_group, flags))) |
| 7610 | goto release_cluster; |
| 7611 | |
| 7612 | offset = btrfs_alloc_from_cluster(used_block_group, |
| 7613 | last_ptr, |
| 7614 | num_bytes, |
| 7615 | used_block_group->key.objectid, |
| 7616 | &max_extent_size); |
| 7617 | if (offset) { |
| 7618 | /* we have a block, we're done */ |
| 7619 | spin_unlock(&last_ptr->refill_lock); |
| 7620 | trace_btrfs_reserve_extent_cluster( |
| 7621 | used_block_group, |
| 7622 | search_start, num_bytes); |
| 7623 | if (used_block_group != block_group) { |
| 7624 | btrfs_release_block_group(block_group, |
| 7625 | delalloc); |
| 7626 | block_group = used_block_group; |
| 7627 | } |
| 7628 | goto checks; |
| 7629 | } |
| 7630 | |
| 7631 | WARN_ON(last_ptr->block_group != used_block_group); |
| 7632 | release_cluster: |
| 7633 | /* If we are on LOOP_NO_EMPTY_SIZE, we can't |
| 7634 | * set up a new clusters, so lets just skip it |
| 7635 | * and let the allocator find whatever block |
| 7636 | * it can find. If we reach this point, we |
| 7637 | * will have tried the cluster allocator |
| 7638 | * plenty of times and not have found |
| 7639 | * anything, so we are likely way too |
| 7640 | * fragmented for the clustering stuff to find |
| 7641 | * anything. |
| 7642 | * |
| 7643 | * However, if the cluster is taken from the |
| 7644 | * current block group, release the cluster |
| 7645 | * first, so that we stand a better chance of |
| 7646 | * succeeding in the unclustered |
| 7647 | * allocation. */ |
| 7648 | if (loop >= LOOP_NO_EMPTY_SIZE && |
| 7649 | used_block_group != block_group) { |
| 7650 | spin_unlock(&last_ptr->refill_lock); |
| 7651 | btrfs_release_block_group(used_block_group, |
| 7652 | delalloc); |
| 7653 | goto unclustered_alloc; |
| 7654 | } |
| 7655 | |
| 7656 | /* |
| 7657 | * this cluster didn't work out, free it and |
| 7658 | * start over |
| 7659 | */ |
| 7660 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
| 7661 | |
| 7662 | if (used_block_group != block_group) |
| 7663 | btrfs_release_block_group(used_block_group, |
| 7664 | delalloc); |
| 7665 | refill_cluster: |
| 7666 | if (loop >= LOOP_NO_EMPTY_SIZE) { |
| 7667 | spin_unlock(&last_ptr->refill_lock); |
| 7668 | goto unclustered_alloc; |
| 7669 | } |
| 7670 | |
| 7671 | aligned_cluster = max_t(unsigned long, |
| 7672 | empty_cluster + empty_size, |
| 7673 | block_group->full_stripe_len); |
| 7674 | |
| 7675 | /* allocate a cluster in this block group */ |
| 7676 | ret = btrfs_find_space_cluster(fs_info, block_group, |
| 7677 | last_ptr, search_start, |
| 7678 | num_bytes, |
| 7679 | aligned_cluster); |
| 7680 | if (ret == 0) { |
| 7681 | /* |
| 7682 | * now pull our allocation out of this |
| 7683 | * cluster |
| 7684 | */ |
| 7685 | offset = btrfs_alloc_from_cluster(block_group, |
| 7686 | last_ptr, |
| 7687 | num_bytes, |
| 7688 | search_start, |
| 7689 | &max_extent_size); |
| 7690 | if (offset) { |
| 7691 | /* we found one, proceed */ |
| 7692 | spin_unlock(&last_ptr->refill_lock); |
| 7693 | trace_btrfs_reserve_extent_cluster( |
| 7694 | block_group, search_start, |
| 7695 | num_bytes); |
| 7696 | goto checks; |
| 7697 | } |
| 7698 | } else if (!cached && loop > LOOP_CACHING_NOWAIT |
| 7699 | && !failed_cluster_refill) { |
| 7700 | spin_unlock(&last_ptr->refill_lock); |
| 7701 | |
| 7702 | failed_cluster_refill = true; |
| 7703 | wait_block_group_cache_progress(block_group, |
| 7704 | num_bytes + empty_cluster + empty_size); |
| 7705 | goto have_block_group; |
| 7706 | } |
| 7707 | |
| 7708 | /* |
| 7709 | * at this point we either didn't find a cluster |
| 7710 | * or we weren't able to allocate a block from our |
| 7711 | * cluster. Free the cluster we've been trying |
| 7712 | * to use, and go to the next block group |
| 7713 | */ |
| 7714 | btrfs_return_cluster_to_free_space(NULL, last_ptr); |
| 7715 | spin_unlock(&last_ptr->refill_lock); |
| 7716 | goto loop; |
| 7717 | } |
| 7718 | |
| 7719 | unclustered_alloc: |
| 7720 | /* |
| 7721 | * We are doing an unclustered alloc, set the fragmented flag so |
| 7722 | * we don't bother trying to setup a cluster again until we get |
| 7723 | * more space. |
| 7724 | */ |
| 7725 | if (unlikely(last_ptr)) { |
| 7726 | spin_lock(&last_ptr->lock); |
| 7727 | last_ptr->fragmented = 1; |
| 7728 | spin_unlock(&last_ptr->lock); |
| 7729 | } |
| 7730 | if (cached) { |
| 7731 | struct btrfs_free_space_ctl *ctl = |
| 7732 | block_group->free_space_ctl; |
| 7733 | |
| 7734 | spin_lock(&ctl->tree_lock); |
| 7735 | if (ctl->free_space < |
| 7736 | num_bytes + empty_cluster + empty_size) { |
| 7737 | if (ctl->free_space > max_extent_size) |
| 7738 | max_extent_size = ctl->free_space; |
| 7739 | spin_unlock(&ctl->tree_lock); |
| 7740 | goto loop; |
| 7741 | } |
| 7742 | spin_unlock(&ctl->tree_lock); |
| 7743 | } |
| 7744 | |
| 7745 | offset = btrfs_find_space_for_alloc(block_group, search_start, |
| 7746 | num_bytes, empty_size, |
| 7747 | &max_extent_size); |
| 7748 | /* |
| 7749 | * If we didn't find a chunk, and we haven't failed on this |
| 7750 | * block group before, and this block group is in the middle of |
| 7751 | * caching and we are ok with waiting, then go ahead and wait |
| 7752 | * for progress to be made, and set failed_alloc to true. |
| 7753 | * |
| 7754 | * If failed_alloc is true then we've already waited on this |
| 7755 | * block group once and should move on to the next block group. |
| 7756 | */ |
| 7757 | if (!offset && !failed_alloc && !cached && |
| 7758 | loop > LOOP_CACHING_NOWAIT) { |
| 7759 | wait_block_group_cache_progress(block_group, |
| 7760 | num_bytes + empty_size); |
| 7761 | failed_alloc = true; |
| 7762 | goto have_block_group; |
| 7763 | } else if (!offset) { |
| 7764 | goto loop; |
| 7765 | } |
| 7766 | checks: |
| 7767 | search_start = ALIGN(offset, fs_info->stripesize); |
| 7768 | |
| 7769 | /* move on to the next group */ |
| 7770 | if (search_start + num_bytes > |
| 7771 | block_group->key.objectid + block_group->key.offset) { |
| 7772 | btrfs_add_free_space(block_group, offset, num_bytes); |
| 7773 | goto loop; |
| 7774 | } |
| 7775 | |
| 7776 | if (offset < search_start) |
| 7777 | btrfs_add_free_space(block_group, offset, |
| 7778 | search_start - offset); |
| 7779 | BUG_ON(offset > search_start); |
| 7780 | |
| 7781 | ret = btrfs_add_reserved_bytes(block_group, ram_bytes, |
| 7782 | num_bytes, delalloc); |
| 7783 | if (ret == -EAGAIN) { |
| 7784 | btrfs_add_free_space(block_group, offset, num_bytes); |
| 7785 | goto loop; |
| 7786 | } |
| 7787 | btrfs_inc_block_group_reservations(block_group); |
| 7788 | |
| 7789 | /* we are all good, lets return */ |
| 7790 | ins->objectid = search_start; |
| 7791 | ins->offset = num_bytes; |
| 7792 | |
| 7793 | trace_btrfs_reserve_extent(block_group, search_start, num_bytes); |
| 7794 | btrfs_release_block_group(block_group, delalloc); |
| 7795 | break; |
| 7796 | loop: |
| 7797 | failed_cluster_refill = false; |
| 7798 | failed_alloc = false; |
| 7799 | BUG_ON(btrfs_bg_flags_to_raid_index(block_group->flags) != |
| 7800 | index); |
| 7801 | btrfs_release_block_group(block_group, delalloc); |
| 7802 | cond_resched(); |
| 7803 | } |
| 7804 | up_read(&space_info->groups_sem); |
| 7805 | |
| 7806 | if ((loop == LOOP_CACHING_NOWAIT) && have_caching_bg |
| 7807 | && !orig_have_caching_bg) |
| 7808 | orig_have_caching_bg = true; |
| 7809 | |
| 7810 | if (!ins->objectid && loop >= LOOP_CACHING_WAIT && have_caching_bg) |
| 7811 | goto search; |
| 7812 | |
| 7813 | if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES) |
| 7814 | goto search; |
| 7815 | |
| 7816 | /* |
| 7817 | * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking |
| 7818 | * caching kthreads as we move along |
| 7819 | * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching |
| 7820 | * LOOP_ALLOC_CHUNK, force a chunk allocation and try again |
| 7821 | * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try |
| 7822 | * again |
| 7823 | */ |
| 7824 | if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) { |
| 7825 | index = 0; |
| 7826 | if (loop == LOOP_CACHING_NOWAIT) { |
| 7827 | /* |
| 7828 | * We want to skip the LOOP_CACHING_WAIT step if we |
| 7829 | * don't have any uncached bgs and we've already done a |
| 7830 | * full search through. |
| 7831 | */ |
| 7832 | if (orig_have_caching_bg || !full_search) |
| 7833 | loop = LOOP_CACHING_WAIT; |
| 7834 | else |
| 7835 | loop = LOOP_ALLOC_CHUNK; |
| 7836 | } else { |
| 7837 | loop++; |
| 7838 | } |
| 7839 | |
| 7840 | if (loop == LOOP_ALLOC_CHUNK) { |
| 7841 | struct btrfs_trans_handle *trans; |
| 7842 | int exist = 0; |
| 7843 | |
| 7844 | trans = current->journal_info; |
| 7845 | if (trans) |
| 7846 | exist = 1; |
| 7847 | else |
| 7848 | trans = btrfs_join_transaction(root); |
| 7849 | |
| 7850 | if (IS_ERR(trans)) { |
| 7851 | ret = PTR_ERR(trans); |
| 7852 | goto out; |
| 7853 | } |
| 7854 | |
| 7855 | ret = do_chunk_alloc(trans, fs_info, flags, |
| 7856 | CHUNK_ALLOC_FORCE); |
| 7857 | |
| 7858 | /* |
| 7859 | * If we can't allocate a new chunk we've already looped |
| 7860 | * through at least once, move on to the NO_EMPTY_SIZE |
| 7861 | * case. |
| 7862 | */ |
| 7863 | if (ret == -ENOSPC) |
| 7864 | loop = LOOP_NO_EMPTY_SIZE; |
| 7865 | |
| 7866 | /* |
| 7867 | * Do not bail out on ENOSPC since we |
| 7868 | * can do more things. |
| 7869 | */ |
| 7870 | if (ret < 0 && ret != -ENOSPC) |
| 7871 | btrfs_abort_transaction(trans, ret); |
| 7872 | else |
| 7873 | ret = 0; |
| 7874 | if (!exist) |
| 7875 | btrfs_end_transaction(trans); |
| 7876 | if (ret) |
| 7877 | goto out; |
| 7878 | } |
| 7879 | |
| 7880 | if (loop == LOOP_NO_EMPTY_SIZE) { |
| 7881 | /* |
| 7882 | * Don't loop again if we already have no empty_size and |
| 7883 | * no empty_cluster. |
| 7884 | */ |
| 7885 | if (empty_size == 0 && |
| 7886 | empty_cluster == 0) { |
| 7887 | ret = -ENOSPC; |
| 7888 | goto out; |
| 7889 | } |
| 7890 | empty_size = 0; |
| 7891 | empty_cluster = 0; |
| 7892 | } |
| 7893 | |
| 7894 | goto search; |
| 7895 | } else if (!ins->objectid) { |
| 7896 | ret = -ENOSPC; |
| 7897 | } else if (ins->objectid) { |
| 7898 | if (!use_cluster && last_ptr) { |
| 7899 | spin_lock(&last_ptr->lock); |
| 7900 | last_ptr->window_start = ins->objectid; |
| 7901 | spin_unlock(&last_ptr->lock); |
| 7902 | } |
| 7903 | ret = 0; |
| 7904 | } |
| 7905 | out: |
| 7906 | if (ret == -ENOSPC) { |
| 7907 | spin_lock(&space_info->lock); |
| 7908 | space_info->max_extent_size = max_extent_size; |
| 7909 | spin_unlock(&space_info->lock); |
| 7910 | ins->offset = max_extent_size; |
| 7911 | } |
| 7912 | return ret; |
| 7913 | } |
| 7914 | |
| 7915 | static void dump_space_info(struct btrfs_fs_info *fs_info, |
| 7916 | struct btrfs_space_info *info, u64 bytes, |
| 7917 | int dump_block_groups) |
| 7918 | { |
| 7919 | struct btrfs_block_group_cache *cache; |
| 7920 | int index = 0; |
| 7921 | |
| 7922 | spin_lock(&info->lock); |
| 7923 | btrfs_info(fs_info, "space_info %llu has %llu free, is %sfull", |
| 7924 | info->flags, |
| 7925 | info->total_bytes - btrfs_space_info_used(info, true), |
| 7926 | info->full ? "" : "not "); |
| 7927 | btrfs_info(fs_info, |
| 7928 | "space_info total=%llu, used=%llu, pinned=%llu, reserved=%llu, may_use=%llu, readonly=%llu", |
| 7929 | info->total_bytes, info->bytes_used, info->bytes_pinned, |
| 7930 | info->bytes_reserved, info->bytes_may_use, |
| 7931 | info->bytes_readonly); |
| 7932 | spin_unlock(&info->lock); |
| 7933 | |
| 7934 | if (!dump_block_groups) |
| 7935 | return; |
| 7936 | |
| 7937 | down_read(&info->groups_sem); |
| 7938 | again: |
| 7939 | list_for_each_entry(cache, &info->block_groups[index], list) { |
| 7940 | spin_lock(&cache->lock); |
| 7941 | btrfs_info(fs_info, |
| 7942 | "block group %llu has %llu bytes, %llu used %llu pinned %llu reserved %s", |
| 7943 | cache->key.objectid, cache->key.offset, |
| 7944 | btrfs_block_group_used(&cache->item), cache->pinned, |
| 7945 | cache->reserved, cache->ro ? "[readonly]" : ""); |
| 7946 | btrfs_dump_free_space(cache, bytes); |
| 7947 | spin_unlock(&cache->lock); |
| 7948 | } |
| 7949 | if (++index < BTRFS_NR_RAID_TYPES) |
| 7950 | goto again; |
| 7951 | up_read(&info->groups_sem); |
| 7952 | } |
| 7953 | |
| 7954 | /* |
| 7955 | * btrfs_reserve_extent - entry point to the extent allocator. Tries to find a |
| 7956 | * hole that is at least as big as @num_bytes. |
| 7957 | * |
| 7958 | * @root - The root that will contain this extent |
| 7959 | * |
| 7960 | * @ram_bytes - The amount of space in ram that @num_bytes take. This |
| 7961 | * is used for accounting purposes. This value differs |
| 7962 | * from @num_bytes only in the case of compressed extents. |
| 7963 | * |
| 7964 | * @num_bytes - Number of bytes to allocate on-disk. |
| 7965 | * |
| 7966 | * @min_alloc_size - Indicates the minimum amount of space that the |
| 7967 | * allocator should try to satisfy. In some cases |
| 7968 | * @num_bytes may be larger than what is required and if |
| 7969 | * the filesystem is fragmented then allocation fails. |
| 7970 | * However, the presence of @min_alloc_size gives a |
| 7971 | * chance to try and satisfy the smaller allocation. |
| 7972 | * |
| 7973 | * @empty_size - A hint that you plan on doing more COW. This is the |
| 7974 | * size in bytes the allocator should try to find free |
| 7975 | * next to the block it returns. This is just a hint and |
| 7976 | * may be ignored by the allocator. |
| 7977 | * |
| 7978 | * @hint_byte - Hint to the allocator to start searching above the byte |
| 7979 | * address passed. It might be ignored. |
| 7980 | * |
| 7981 | * @ins - This key is modified to record the found hole. It will |
| 7982 | * have the following values: |
| 7983 | * ins->objectid == start position |
| 7984 | * ins->flags = BTRFS_EXTENT_ITEM_KEY |
| 7985 | * ins->offset == the size of the hole. |
| 7986 | * |
| 7987 | * @is_data - Boolean flag indicating whether an extent is |
| 7988 | * allocated for data (true) or metadata (false) |
| 7989 | * |
| 7990 | * @delalloc - Boolean flag indicating whether this allocation is for |
| 7991 | * delalloc or not. If 'true' data_rwsem of block groups |
| 7992 | * is going to be acquired. |
| 7993 | * |
| 7994 | * |
| 7995 | * Returns 0 when an allocation succeeded or < 0 when an error occurred. In |
| 7996 | * case -ENOSPC is returned then @ins->offset will contain the size of the |
| 7997 | * largest available hole the allocator managed to find. |
| 7998 | */ |
| 7999 | int btrfs_reserve_extent(struct btrfs_root *root, u64 ram_bytes, |
| 8000 | u64 num_bytes, u64 min_alloc_size, |
| 8001 | u64 empty_size, u64 hint_byte, |
| 8002 | struct btrfs_key *ins, int is_data, int delalloc) |
| 8003 | { |
| 8004 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 8005 | bool final_tried = num_bytes == min_alloc_size; |
| 8006 | u64 flags; |
| 8007 | int ret; |
| 8008 | |
| 8009 | flags = get_alloc_profile_by_root(root, is_data); |
| 8010 | again: |
| 8011 | WARN_ON(num_bytes < fs_info->sectorsize); |
| 8012 | ret = find_free_extent(fs_info, ram_bytes, num_bytes, empty_size, |
| 8013 | hint_byte, ins, flags, delalloc); |
| 8014 | if (!ret && !is_data) { |
| 8015 | btrfs_dec_block_group_reservations(fs_info, ins->objectid); |
| 8016 | } else if (ret == -ENOSPC) { |
| 8017 | if (!final_tried && ins->offset) { |
| 8018 | num_bytes = min(num_bytes >> 1, ins->offset); |
| 8019 | num_bytes = round_down(num_bytes, |
| 8020 | fs_info->sectorsize); |
| 8021 | num_bytes = max(num_bytes, min_alloc_size); |
| 8022 | ram_bytes = num_bytes; |
| 8023 | if (num_bytes == min_alloc_size) |
| 8024 | final_tried = true; |
| 8025 | goto again; |
| 8026 | } else if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
| 8027 | struct btrfs_space_info *sinfo; |
| 8028 | |
| 8029 | sinfo = __find_space_info(fs_info, flags); |
| 8030 | btrfs_err(fs_info, |
| 8031 | "allocation failed flags %llu, wanted %llu", |
| 8032 | flags, num_bytes); |
| 8033 | if (sinfo) |
| 8034 | dump_space_info(fs_info, sinfo, num_bytes, 1); |
| 8035 | } |
| 8036 | } |
| 8037 | |
| 8038 | return ret; |
| 8039 | } |
| 8040 | |
| 8041 | static int __btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
| 8042 | u64 start, u64 len, |
| 8043 | int pin, int delalloc) |
| 8044 | { |
| 8045 | struct btrfs_block_group_cache *cache; |
| 8046 | int ret = 0; |
| 8047 | |
| 8048 | cache = btrfs_lookup_block_group(fs_info, start); |
| 8049 | if (!cache) { |
| 8050 | btrfs_err(fs_info, "Unable to find block group for %llu", |
| 8051 | start); |
| 8052 | return -ENOSPC; |
| 8053 | } |
| 8054 | |
| 8055 | if (pin) |
| 8056 | pin_down_extent(fs_info, cache, start, len, 1); |
| 8057 | else { |
| 8058 | if (btrfs_test_opt(fs_info, DISCARD)) |
| 8059 | ret = btrfs_discard_extent(fs_info, start, len, NULL); |
| 8060 | btrfs_add_free_space(cache, start, len); |
| 8061 | btrfs_free_reserved_bytes(cache, len, delalloc); |
| 8062 | trace_btrfs_reserved_extent_free(fs_info, start, len); |
| 8063 | } |
| 8064 | |
| 8065 | btrfs_put_block_group(cache); |
| 8066 | return ret; |
| 8067 | } |
| 8068 | |
| 8069 | int btrfs_free_reserved_extent(struct btrfs_fs_info *fs_info, |
| 8070 | u64 start, u64 len, int delalloc) |
| 8071 | { |
| 8072 | return __btrfs_free_reserved_extent(fs_info, start, len, 0, delalloc); |
| 8073 | } |
| 8074 | |
| 8075 | int btrfs_free_and_pin_reserved_extent(struct btrfs_fs_info *fs_info, |
| 8076 | u64 start, u64 len) |
| 8077 | { |
| 8078 | return __btrfs_free_reserved_extent(fs_info, start, len, 1, 0); |
| 8079 | } |
| 8080 | |
| 8081 | static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
| 8082 | struct btrfs_fs_info *fs_info, |
| 8083 | u64 parent, u64 root_objectid, |
| 8084 | u64 flags, u64 owner, u64 offset, |
| 8085 | struct btrfs_key *ins, int ref_mod) |
| 8086 | { |
| 8087 | int ret; |
| 8088 | struct btrfs_extent_item *extent_item; |
| 8089 | struct btrfs_extent_inline_ref *iref; |
| 8090 | struct btrfs_path *path; |
| 8091 | struct extent_buffer *leaf; |
| 8092 | int type; |
| 8093 | u32 size; |
| 8094 | |
| 8095 | if (parent > 0) |
| 8096 | type = BTRFS_SHARED_DATA_REF_KEY; |
| 8097 | else |
| 8098 | type = BTRFS_EXTENT_DATA_REF_KEY; |
| 8099 | |
| 8100 | size = sizeof(*extent_item) + btrfs_extent_inline_ref_size(type); |
| 8101 | |
| 8102 | path = btrfs_alloc_path(); |
| 8103 | if (!path) |
| 8104 | return -ENOMEM; |
| 8105 | |
| 8106 | path->leave_spinning = 1; |
| 8107 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
| 8108 | ins, size); |
| 8109 | if (ret) { |
| 8110 | btrfs_free_path(path); |
| 8111 | return ret; |
| 8112 | } |
| 8113 | |
| 8114 | leaf = path->nodes[0]; |
| 8115 | extent_item = btrfs_item_ptr(leaf, path->slots[0], |
| 8116 | struct btrfs_extent_item); |
| 8117 | btrfs_set_extent_refs(leaf, extent_item, ref_mod); |
| 8118 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); |
| 8119 | btrfs_set_extent_flags(leaf, extent_item, |
| 8120 | flags | BTRFS_EXTENT_FLAG_DATA); |
| 8121 | |
| 8122 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); |
| 8123 | btrfs_set_extent_inline_ref_type(leaf, iref, type); |
| 8124 | if (parent > 0) { |
| 8125 | struct btrfs_shared_data_ref *ref; |
| 8126 | ref = (struct btrfs_shared_data_ref *)(iref + 1); |
| 8127 | btrfs_set_extent_inline_ref_offset(leaf, iref, parent); |
| 8128 | btrfs_set_shared_data_ref_count(leaf, ref, ref_mod); |
| 8129 | } else { |
| 8130 | struct btrfs_extent_data_ref *ref; |
| 8131 | ref = (struct btrfs_extent_data_ref *)(&iref->offset); |
| 8132 | btrfs_set_extent_data_ref_root(leaf, ref, root_objectid); |
| 8133 | btrfs_set_extent_data_ref_objectid(leaf, ref, owner); |
| 8134 | btrfs_set_extent_data_ref_offset(leaf, ref, offset); |
| 8135 | btrfs_set_extent_data_ref_count(leaf, ref, ref_mod); |
| 8136 | } |
| 8137 | |
| 8138 | btrfs_mark_buffer_dirty(path->nodes[0]); |
| 8139 | btrfs_free_path(path); |
| 8140 | |
| 8141 | ret = remove_from_free_space_tree(trans, ins->objectid, ins->offset); |
| 8142 | if (ret) |
| 8143 | return ret; |
| 8144 | |
| 8145 | ret = update_block_group(trans, fs_info, ins->objectid, ins->offset, 1); |
| 8146 | if (ret) { /* -ENOENT, logic error */ |
| 8147 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
| 8148 | ins->objectid, ins->offset); |
| 8149 | BUG(); |
| 8150 | } |
| 8151 | trace_btrfs_reserved_extent_alloc(fs_info, ins->objectid, ins->offset); |
| 8152 | return ret; |
| 8153 | } |
| 8154 | |
| 8155 | static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans, |
| 8156 | struct btrfs_delayed_ref_node *node, |
| 8157 | struct btrfs_delayed_extent_op *extent_op) |
| 8158 | { |
| 8159 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 8160 | int ret; |
| 8161 | struct btrfs_extent_item *extent_item; |
| 8162 | struct btrfs_key extent_key; |
| 8163 | struct btrfs_tree_block_info *block_info; |
| 8164 | struct btrfs_extent_inline_ref *iref; |
| 8165 | struct btrfs_path *path; |
| 8166 | struct extent_buffer *leaf; |
| 8167 | struct btrfs_delayed_tree_ref *ref; |
| 8168 | u32 size = sizeof(*extent_item) + sizeof(*iref); |
| 8169 | u64 num_bytes; |
| 8170 | u64 flags = extent_op->flags_to_set; |
| 8171 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
| 8172 | |
| 8173 | ref = btrfs_delayed_node_to_tree_ref(node); |
| 8174 | |
| 8175 | extent_key.objectid = node->bytenr; |
| 8176 | if (skinny_metadata) { |
| 8177 | extent_key.offset = ref->level; |
| 8178 | extent_key.type = BTRFS_METADATA_ITEM_KEY; |
| 8179 | num_bytes = fs_info->nodesize; |
| 8180 | } else { |
| 8181 | extent_key.offset = node->num_bytes; |
| 8182 | extent_key.type = BTRFS_EXTENT_ITEM_KEY; |
| 8183 | size += sizeof(*block_info); |
| 8184 | num_bytes = node->num_bytes; |
| 8185 | } |
| 8186 | |
| 8187 | path = btrfs_alloc_path(); |
| 8188 | if (!path) { |
| 8189 | btrfs_free_and_pin_reserved_extent(fs_info, |
| 8190 | extent_key.objectid, |
| 8191 | fs_info->nodesize); |
| 8192 | return -ENOMEM; |
| 8193 | } |
| 8194 | |
| 8195 | path->leave_spinning = 1; |
| 8196 | ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path, |
| 8197 | &extent_key, size); |
| 8198 | if (ret) { |
| 8199 | btrfs_free_path(path); |
| 8200 | btrfs_free_and_pin_reserved_extent(fs_info, |
| 8201 | extent_key.objectid, |
| 8202 | fs_info->nodesize); |
| 8203 | return ret; |
| 8204 | } |
| 8205 | |
| 8206 | leaf = path->nodes[0]; |
| 8207 | extent_item = btrfs_item_ptr(leaf, path->slots[0], |
| 8208 | struct btrfs_extent_item); |
| 8209 | btrfs_set_extent_refs(leaf, extent_item, 1); |
| 8210 | btrfs_set_extent_generation(leaf, extent_item, trans->transid); |
| 8211 | btrfs_set_extent_flags(leaf, extent_item, |
| 8212 | flags | BTRFS_EXTENT_FLAG_TREE_BLOCK); |
| 8213 | |
| 8214 | if (skinny_metadata) { |
| 8215 | iref = (struct btrfs_extent_inline_ref *)(extent_item + 1); |
| 8216 | } else { |
| 8217 | block_info = (struct btrfs_tree_block_info *)(extent_item + 1); |
| 8218 | btrfs_set_tree_block_key(leaf, block_info, &extent_op->key); |
| 8219 | btrfs_set_tree_block_level(leaf, block_info, ref->level); |
| 8220 | iref = (struct btrfs_extent_inline_ref *)(block_info + 1); |
| 8221 | } |
| 8222 | |
| 8223 | if (node->type == BTRFS_SHARED_BLOCK_REF_KEY) { |
| 8224 | BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)); |
| 8225 | btrfs_set_extent_inline_ref_type(leaf, iref, |
| 8226 | BTRFS_SHARED_BLOCK_REF_KEY); |
| 8227 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->parent); |
| 8228 | } else { |
| 8229 | btrfs_set_extent_inline_ref_type(leaf, iref, |
| 8230 | BTRFS_TREE_BLOCK_REF_KEY); |
| 8231 | btrfs_set_extent_inline_ref_offset(leaf, iref, ref->root); |
| 8232 | } |
| 8233 | |
| 8234 | btrfs_mark_buffer_dirty(leaf); |
| 8235 | btrfs_free_path(path); |
| 8236 | |
| 8237 | ret = remove_from_free_space_tree(trans, extent_key.objectid, |
| 8238 | num_bytes); |
| 8239 | if (ret) |
| 8240 | return ret; |
| 8241 | |
| 8242 | ret = update_block_group(trans, fs_info, extent_key.objectid, |
| 8243 | fs_info->nodesize, 1); |
| 8244 | if (ret) { /* -ENOENT, logic error */ |
| 8245 | btrfs_err(fs_info, "update block group failed for %llu %llu", |
| 8246 | extent_key.objectid, extent_key.offset); |
| 8247 | BUG(); |
| 8248 | } |
| 8249 | |
| 8250 | trace_btrfs_reserved_extent_alloc(fs_info, extent_key.objectid, |
| 8251 | fs_info->nodesize); |
| 8252 | return ret; |
| 8253 | } |
| 8254 | |
| 8255 | int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans, |
| 8256 | struct btrfs_root *root, u64 owner, |
| 8257 | u64 offset, u64 ram_bytes, |
| 8258 | struct btrfs_key *ins) |
| 8259 | { |
| 8260 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 8261 | int ret; |
| 8262 | |
| 8263 | BUG_ON(root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
| 8264 | |
| 8265 | btrfs_ref_tree_mod(root, ins->objectid, ins->offset, 0, |
| 8266 | root->root_key.objectid, owner, offset, |
| 8267 | BTRFS_ADD_DELAYED_EXTENT); |
| 8268 | |
| 8269 | ret = btrfs_add_delayed_data_ref(fs_info, trans, ins->objectid, |
| 8270 | ins->offset, 0, |
| 8271 | root->root_key.objectid, owner, |
| 8272 | offset, ram_bytes, |
| 8273 | BTRFS_ADD_DELAYED_EXTENT, NULL, NULL); |
| 8274 | return ret; |
| 8275 | } |
| 8276 | |
| 8277 | /* |
| 8278 | * this is used by the tree logging recovery code. It records that |
| 8279 | * an extent has been allocated and makes sure to clear the free |
| 8280 | * space cache bits as well |
| 8281 | */ |
| 8282 | int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans, |
| 8283 | struct btrfs_fs_info *fs_info, |
| 8284 | u64 root_objectid, u64 owner, u64 offset, |
| 8285 | struct btrfs_key *ins) |
| 8286 | { |
| 8287 | int ret; |
| 8288 | struct btrfs_block_group_cache *block_group; |
| 8289 | struct btrfs_space_info *space_info; |
| 8290 | |
| 8291 | /* |
| 8292 | * Mixed block groups will exclude before processing the log so we only |
| 8293 | * need to do the exclude dance if this fs isn't mixed. |
| 8294 | */ |
| 8295 | if (!btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { |
| 8296 | ret = __exclude_logged_extent(fs_info, ins->objectid, |
| 8297 | ins->offset); |
| 8298 | if (ret) |
| 8299 | return ret; |
| 8300 | } |
| 8301 | |
| 8302 | block_group = btrfs_lookup_block_group(fs_info, ins->objectid); |
| 8303 | if (!block_group) |
| 8304 | return -EINVAL; |
| 8305 | |
| 8306 | space_info = block_group->space_info; |
| 8307 | spin_lock(&space_info->lock); |
| 8308 | spin_lock(&block_group->lock); |
| 8309 | space_info->bytes_reserved += ins->offset; |
| 8310 | block_group->reserved += ins->offset; |
| 8311 | spin_unlock(&block_group->lock); |
| 8312 | spin_unlock(&space_info->lock); |
| 8313 | |
| 8314 | ret = alloc_reserved_file_extent(trans, fs_info, 0, root_objectid, |
| 8315 | 0, owner, offset, ins, 1); |
| 8316 | btrfs_put_block_group(block_group); |
| 8317 | return ret; |
| 8318 | } |
| 8319 | |
| 8320 | static struct extent_buffer * |
| 8321 | btrfs_init_new_buffer(struct btrfs_trans_handle *trans, struct btrfs_root *root, |
| 8322 | u64 bytenr, int level) |
| 8323 | { |
| 8324 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 8325 | struct extent_buffer *buf; |
| 8326 | |
| 8327 | buf = btrfs_find_create_tree_block(fs_info, bytenr); |
| 8328 | if (IS_ERR(buf)) |
| 8329 | return buf; |
| 8330 | |
| 8331 | btrfs_set_header_generation(buf, trans->transid); |
| 8332 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level); |
| 8333 | btrfs_tree_lock(buf); |
| 8334 | clean_tree_block(fs_info, buf); |
| 8335 | clear_bit(EXTENT_BUFFER_STALE, &buf->bflags); |
| 8336 | |
| 8337 | btrfs_set_lock_blocking(buf); |
| 8338 | set_extent_buffer_uptodate(buf); |
| 8339 | |
| 8340 | if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { |
| 8341 | buf->log_index = root->log_transid % 2; |
| 8342 | /* |
| 8343 | * we allow two log transactions at a time, use different |
| 8344 | * EXENT bit to differentiate dirty pages. |
| 8345 | */ |
| 8346 | if (buf->log_index == 0) |
| 8347 | set_extent_dirty(&root->dirty_log_pages, buf->start, |
| 8348 | buf->start + buf->len - 1, GFP_NOFS); |
| 8349 | else |
| 8350 | set_extent_new(&root->dirty_log_pages, buf->start, |
| 8351 | buf->start + buf->len - 1); |
| 8352 | } else { |
| 8353 | buf->log_index = -1; |
| 8354 | set_extent_dirty(&trans->transaction->dirty_pages, buf->start, |
| 8355 | buf->start + buf->len - 1, GFP_NOFS); |
| 8356 | } |
| 8357 | trans->dirty = true; |
| 8358 | /* this returns a buffer locked for blocking */ |
| 8359 | return buf; |
| 8360 | } |
| 8361 | |
| 8362 | static struct btrfs_block_rsv * |
| 8363 | use_block_rsv(struct btrfs_trans_handle *trans, |
| 8364 | struct btrfs_root *root, u32 blocksize) |
| 8365 | { |
| 8366 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 8367 | struct btrfs_block_rsv *block_rsv; |
| 8368 | struct btrfs_block_rsv *global_rsv = &fs_info->global_block_rsv; |
| 8369 | int ret; |
| 8370 | bool global_updated = false; |
| 8371 | |
| 8372 | block_rsv = get_block_rsv(trans, root); |
| 8373 | |
| 8374 | if (unlikely(block_rsv->size == 0)) |
| 8375 | goto try_reserve; |
| 8376 | again: |
| 8377 | ret = block_rsv_use_bytes(block_rsv, blocksize); |
| 8378 | if (!ret) |
| 8379 | return block_rsv; |
| 8380 | |
| 8381 | if (block_rsv->failfast) |
| 8382 | return ERR_PTR(ret); |
| 8383 | |
| 8384 | if (block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL && !global_updated) { |
| 8385 | global_updated = true; |
| 8386 | update_global_block_rsv(fs_info); |
| 8387 | goto again; |
| 8388 | } |
| 8389 | |
| 8390 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
| 8391 | static DEFINE_RATELIMIT_STATE(_rs, |
| 8392 | DEFAULT_RATELIMIT_INTERVAL * 10, |
| 8393 | /*DEFAULT_RATELIMIT_BURST*/ 1); |
| 8394 | if (__ratelimit(&_rs)) |
| 8395 | WARN(1, KERN_DEBUG |
| 8396 | "BTRFS: block rsv returned %d\n", ret); |
| 8397 | } |
| 8398 | try_reserve: |
| 8399 | ret = reserve_metadata_bytes(root, block_rsv, blocksize, |
| 8400 | BTRFS_RESERVE_NO_FLUSH); |
| 8401 | if (!ret) |
| 8402 | return block_rsv; |
| 8403 | /* |
| 8404 | * If we couldn't reserve metadata bytes try and use some from |
| 8405 | * the global reserve if its space type is the same as the global |
| 8406 | * reservation. |
| 8407 | */ |
| 8408 | if (block_rsv->type != BTRFS_BLOCK_RSV_GLOBAL && |
| 8409 | block_rsv->space_info == global_rsv->space_info) { |
| 8410 | ret = block_rsv_use_bytes(global_rsv, blocksize); |
| 8411 | if (!ret) |
| 8412 | return global_rsv; |
| 8413 | } |
| 8414 | return ERR_PTR(ret); |
| 8415 | } |
| 8416 | |
| 8417 | static void unuse_block_rsv(struct btrfs_fs_info *fs_info, |
| 8418 | struct btrfs_block_rsv *block_rsv, u32 blocksize) |
| 8419 | { |
| 8420 | block_rsv_add_bytes(block_rsv, blocksize, 0); |
| 8421 | block_rsv_release_bytes(fs_info, block_rsv, NULL, 0, NULL); |
| 8422 | } |
| 8423 | |
| 8424 | /* |
| 8425 | * finds a free extent and does all the dirty work required for allocation |
| 8426 | * returns the tree buffer or an ERR_PTR on error. |
| 8427 | */ |
| 8428 | struct extent_buffer *btrfs_alloc_tree_block(struct btrfs_trans_handle *trans, |
| 8429 | struct btrfs_root *root, |
| 8430 | u64 parent, u64 root_objectid, |
| 8431 | const struct btrfs_disk_key *key, |
| 8432 | int level, u64 hint, |
| 8433 | u64 empty_size) |
| 8434 | { |
| 8435 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 8436 | struct btrfs_key ins; |
| 8437 | struct btrfs_block_rsv *block_rsv; |
| 8438 | struct extent_buffer *buf; |
| 8439 | struct btrfs_delayed_extent_op *extent_op; |
| 8440 | u64 flags = 0; |
| 8441 | int ret; |
| 8442 | u32 blocksize = fs_info->nodesize; |
| 8443 | bool skinny_metadata = btrfs_fs_incompat(fs_info, SKINNY_METADATA); |
| 8444 | |
| 8445 | #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS |
| 8446 | if (btrfs_is_testing(fs_info)) { |
| 8447 | buf = btrfs_init_new_buffer(trans, root, root->alloc_bytenr, |
| 8448 | level); |
| 8449 | if (!IS_ERR(buf)) |
| 8450 | root->alloc_bytenr += blocksize; |
| 8451 | return buf; |
| 8452 | } |
| 8453 | #endif |
| 8454 | |
| 8455 | block_rsv = use_block_rsv(trans, root, blocksize); |
| 8456 | if (IS_ERR(block_rsv)) |
| 8457 | return ERR_CAST(block_rsv); |
| 8458 | |
| 8459 | ret = btrfs_reserve_extent(root, blocksize, blocksize, blocksize, |
| 8460 | empty_size, hint, &ins, 0, 0); |
| 8461 | if (ret) |
| 8462 | goto out_unuse; |
| 8463 | |
| 8464 | buf = btrfs_init_new_buffer(trans, root, ins.objectid, level); |
| 8465 | if (IS_ERR(buf)) { |
| 8466 | ret = PTR_ERR(buf); |
| 8467 | goto out_free_reserved; |
| 8468 | } |
| 8469 | |
| 8470 | if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) { |
| 8471 | if (parent == 0) |
| 8472 | parent = ins.objectid; |
| 8473 | flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF; |
| 8474 | } else |
| 8475 | BUG_ON(parent > 0); |
| 8476 | |
| 8477 | if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { |
| 8478 | extent_op = btrfs_alloc_delayed_extent_op(); |
| 8479 | if (!extent_op) { |
| 8480 | ret = -ENOMEM; |
| 8481 | goto out_free_buf; |
| 8482 | } |
| 8483 | if (key) |
| 8484 | memcpy(&extent_op->key, key, sizeof(extent_op->key)); |
| 8485 | else |
| 8486 | memset(&extent_op->key, 0, sizeof(extent_op->key)); |
| 8487 | extent_op->flags_to_set = flags; |
| 8488 | extent_op->update_key = skinny_metadata ? false : true; |
| 8489 | extent_op->update_flags = true; |
| 8490 | extent_op->is_data = false; |
| 8491 | extent_op->level = level; |
| 8492 | |
| 8493 | btrfs_ref_tree_mod(root, ins.objectid, ins.offset, parent, |
| 8494 | root_objectid, level, 0, |
| 8495 | BTRFS_ADD_DELAYED_EXTENT); |
| 8496 | ret = btrfs_add_delayed_tree_ref(fs_info, trans, ins.objectid, |
| 8497 | ins.offset, parent, |
| 8498 | root_objectid, level, |
| 8499 | BTRFS_ADD_DELAYED_EXTENT, |
| 8500 | extent_op, NULL, NULL); |
| 8501 | if (ret) |
| 8502 | goto out_free_delayed; |
| 8503 | } |
| 8504 | return buf; |
| 8505 | |
| 8506 | out_free_delayed: |
| 8507 | btrfs_free_delayed_extent_op(extent_op); |
| 8508 | out_free_buf: |
| 8509 | free_extent_buffer(buf); |
| 8510 | out_free_reserved: |
| 8511 | btrfs_free_reserved_extent(fs_info, ins.objectid, ins.offset, 0); |
| 8512 | out_unuse: |
| 8513 | unuse_block_rsv(fs_info, block_rsv, blocksize); |
| 8514 | return ERR_PTR(ret); |
| 8515 | } |
| 8516 | |
| 8517 | struct walk_control { |
| 8518 | u64 refs[BTRFS_MAX_LEVEL]; |
| 8519 | u64 flags[BTRFS_MAX_LEVEL]; |
| 8520 | struct btrfs_key update_progress; |
| 8521 | int stage; |
| 8522 | int level; |
| 8523 | int shared_level; |
| 8524 | int update_ref; |
| 8525 | int keep_locks; |
| 8526 | int reada_slot; |
| 8527 | int reada_count; |
| 8528 | int for_reloc; |
| 8529 | }; |
| 8530 | |
| 8531 | #define DROP_REFERENCE 1 |
| 8532 | #define UPDATE_BACKREF 2 |
| 8533 | |
| 8534 | static noinline void reada_walk_down(struct btrfs_trans_handle *trans, |
| 8535 | struct btrfs_root *root, |
| 8536 | struct walk_control *wc, |
| 8537 | struct btrfs_path *path) |
| 8538 | { |
| 8539 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 8540 | u64 bytenr; |
| 8541 | u64 generation; |
| 8542 | u64 refs; |
| 8543 | u64 flags; |
| 8544 | u32 nritems; |
| 8545 | struct btrfs_key key; |
| 8546 | struct extent_buffer *eb; |
| 8547 | int ret; |
| 8548 | int slot; |
| 8549 | int nread = 0; |
| 8550 | |
| 8551 | if (path->slots[wc->level] < wc->reada_slot) { |
| 8552 | wc->reada_count = wc->reada_count * 2 / 3; |
| 8553 | wc->reada_count = max(wc->reada_count, 2); |
| 8554 | } else { |
| 8555 | wc->reada_count = wc->reada_count * 3 / 2; |
| 8556 | wc->reada_count = min_t(int, wc->reada_count, |
| 8557 | BTRFS_NODEPTRS_PER_BLOCK(fs_info)); |
| 8558 | } |
| 8559 | |
| 8560 | eb = path->nodes[wc->level]; |
| 8561 | nritems = btrfs_header_nritems(eb); |
| 8562 | |
| 8563 | for (slot = path->slots[wc->level]; slot < nritems; slot++) { |
| 8564 | if (nread >= wc->reada_count) |
| 8565 | break; |
| 8566 | |
| 8567 | cond_resched(); |
| 8568 | bytenr = btrfs_node_blockptr(eb, slot); |
| 8569 | generation = btrfs_node_ptr_generation(eb, slot); |
| 8570 | |
| 8571 | if (slot == path->slots[wc->level]) |
| 8572 | goto reada; |
| 8573 | |
| 8574 | if (wc->stage == UPDATE_BACKREF && |
| 8575 | generation <= root->root_key.offset) |
| 8576 | continue; |
| 8577 | |
| 8578 | /* We don't lock the tree block, it's OK to be racy here */ |
| 8579 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, |
| 8580 | wc->level - 1, 1, &refs, |
| 8581 | &flags); |
| 8582 | /* We don't care about errors in readahead. */ |
| 8583 | if (ret < 0) |
| 8584 | continue; |
| 8585 | BUG_ON(refs == 0); |
| 8586 | |
| 8587 | if (wc->stage == DROP_REFERENCE) { |
| 8588 | if (refs == 1) |
| 8589 | goto reada; |
| 8590 | |
| 8591 | if (wc->level == 1 && |
| 8592 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) |
| 8593 | continue; |
| 8594 | if (!wc->update_ref || |
| 8595 | generation <= root->root_key.offset) |
| 8596 | continue; |
| 8597 | btrfs_node_key_to_cpu(eb, &key, slot); |
| 8598 | ret = btrfs_comp_cpu_keys(&key, |
| 8599 | &wc->update_progress); |
| 8600 | if (ret < 0) |
| 8601 | continue; |
| 8602 | } else { |
| 8603 | if (wc->level == 1 && |
| 8604 | (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) |
| 8605 | continue; |
| 8606 | } |
| 8607 | reada: |
| 8608 | readahead_tree_block(fs_info, bytenr); |
| 8609 | nread++; |
| 8610 | } |
| 8611 | wc->reada_slot = slot; |
| 8612 | } |
| 8613 | |
| 8614 | /* |
| 8615 | * helper to process tree block while walking down the tree. |
| 8616 | * |
| 8617 | * when wc->stage == UPDATE_BACKREF, this function updates |
| 8618 | * back refs for pointers in the block. |
| 8619 | * |
| 8620 | * NOTE: return value 1 means we should stop walking down. |
| 8621 | */ |
| 8622 | static noinline int walk_down_proc(struct btrfs_trans_handle *trans, |
| 8623 | struct btrfs_root *root, |
| 8624 | struct btrfs_path *path, |
| 8625 | struct walk_control *wc, int lookup_info) |
| 8626 | { |
| 8627 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 8628 | int level = wc->level; |
| 8629 | struct extent_buffer *eb = path->nodes[level]; |
| 8630 | u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
| 8631 | int ret; |
| 8632 | |
| 8633 | if (wc->stage == UPDATE_BACKREF && |
| 8634 | btrfs_header_owner(eb) != root->root_key.objectid) |
| 8635 | return 1; |
| 8636 | |
| 8637 | /* |
| 8638 | * when reference count of tree block is 1, it won't increase |
| 8639 | * again. once full backref flag is set, we never clear it. |
| 8640 | */ |
| 8641 | if (lookup_info && |
| 8642 | ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) || |
| 8643 | (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag)))) { |
| 8644 | BUG_ON(!path->locks[level]); |
| 8645 | ret = btrfs_lookup_extent_info(trans, fs_info, |
| 8646 | eb->start, level, 1, |
| 8647 | &wc->refs[level], |
| 8648 | &wc->flags[level]); |
| 8649 | BUG_ON(ret == -ENOMEM); |
| 8650 | if (ret) |
| 8651 | return ret; |
| 8652 | BUG_ON(wc->refs[level] == 0); |
| 8653 | } |
| 8654 | |
| 8655 | if (wc->stage == DROP_REFERENCE) { |
| 8656 | if (wc->refs[level] > 1) |
| 8657 | return 1; |
| 8658 | |
| 8659 | if (path->locks[level] && !wc->keep_locks) { |
| 8660 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
| 8661 | path->locks[level] = 0; |
| 8662 | } |
| 8663 | return 0; |
| 8664 | } |
| 8665 | |
| 8666 | /* wc->stage == UPDATE_BACKREF */ |
| 8667 | if (!(wc->flags[level] & flag)) { |
| 8668 | BUG_ON(!path->locks[level]); |
| 8669 | ret = btrfs_inc_ref(trans, root, eb, 1); |
| 8670 | BUG_ON(ret); /* -ENOMEM */ |
| 8671 | ret = btrfs_dec_ref(trans, root, eb, 0); |
| 8672 | BUG_ON(ret); /* -ENOMEM */ |
| 8673 | ret = btrfs_set_disk_extent_flags(trans, fs_info, eb->start, |
| 8674 | eb->len, flag, |
| 8675 | btrfs_header_level(eb), 0); |
| 8676 | BUG_ON(ret); /* -ENOMEM */ |
| 8677 | wc->flags[level] |= flag; |
| 8678 | } |
| 8679 | |
| 8680 | /* |
| 8681 | * the block is shared by multiple trees, so it's not good to |
| 8682 | * keep the tree lock |
| 8683 | */ |
| 8684 | if (path->locks[level] && level > 0) { |
| 8685 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
| 8686 | path->locks[level] = 0; |
| 8687 | } |
| 8688 | return 0; |
| 8689 | } |
| 8690 | |
| 8691 | /* |
| 8692 | * helper to process tree block pointer. |
| 8693 | * |
| 8694 | * when wc->stage == DROP_REFERENCE, this function checks |
| 8695 | * reference count of the block pointed to. if the block |
| 8696 | * is shared and we need update back refs for the subtree |
| 8697 | * rooted at the block, this function changes wc->stage to |
| 8698 | * UPDATE_BACKREF. if the block is shared and there is no |
| 8699 | * need to update back, this function drops the reference |
| 8700 | * to the block. |
| 8701 | * |
| 8702 | * NOTE: return value 1 means we should stop walking down. |
| 8703 | */ |
| 8704 | static noinline int do_walk_down(struct btrfs_trans_handle *trans, |
| 8705 | struct btrfs_root *root, |
| 8706 | struct btrfs_path *path, |
| 8707 | struct walk_control *wc, int *lookup_info) |
| 8708 | { |
| 8709 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 8710 | u64 bytenr; |
| 8711 | u64 generation; |
| 8712 | u64 parent; |
| 8713 | u32 blocksize; |
| 8714 | struct btrfs_key key; |
| 8715 | struct btrfs_key first_key; |
| 8716 | struct extent_buffer *next; |
| 8717 | int level = wc->level; |
| 8718 | int reada = 0; |
| 8719 | int ret = 0; |
| 8720 | bool need_account = false; |
| 8721 | |
| 8722 | generation = btrfs_node_ptr_generation(path->nodes[level], |
| 8723 | path->slots[level]); |
| 8724 | /* |
| 8725 | * if the lower level block was created before the snapshot |
| 8726 | * was created, we know there is no need to update back refs |
| 8727 | * for the subtree |
| 8728 | */ |
| 8729 | if (wc->stage == UPDATE_BACKREF && |
| 8730 | generation <= root->root_key.offset) { |
| 8731 | *lookup_info = 1; |
| 8732 | return 1; |
| 8733 | } |
| 8734 | |
| 8735 | bytenr = btrfs_node_blockptr(path->nodes[level], path->slots[level]); |
| 8736 | btrfs_node_key_to_cpu(path->nodes[level], &first_key, |
| 8737 | path->slots[level]); |
| 8738 | blocksize = fs_info->nodesize; |
| 8739 | |
| 8740 | next = find_extent_buffer(fs_info, bytenr); |
| 8741 | if (!next) { |
| 8742 | next = btrfs_find_create_tree_block(fs_info, bytenr); |
| 8743 | if (IS_ERR(next)) |
| 8744 | return PTR_ERR(next); |
| 8745 | |
| 8746 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, next, |
| 8747 | level - 1); |
| 8748 | reada = 1; |
| 8749 | } |
| 8750 | btrfs_tree_lock(next); |
| 8751 | btrfs_set_lock_blocking(next); |
| 8752 | |
| 8753 | ret = btrfs_lookup_extent_info(trans, fs_info, bytenr, level - 1, 1, |
| 8754 | &wc->refs[level - 1], |
| 8755 | &wc->flags[level - 1]); |
| 8756 | if (ret < 0) |
| 8757 | goto out_unlock; |
| 8758 | |
| 8759 | if (unlikely(wc->refs[level - 1] == 0)) { |
| 8760 | btrfs_err(fs_info, "Missing references."); |
| 8761 | ret = -EIO; |
| 8762 | goto out_unlock; |
| 8763 | } |
| 8764 | *lookup_info = 0; |
| 8765 | |
| 8766 | if (wc->stage == DROP_REFERENCE) { |
| 8767 | if (wc->refs[level - 1] > 1) { |
| 8768 | need_account = true; |
| 8769 | if (level == 1 && |
| 8770 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) |
| 8771 | goto skip; |
| 8772 | |
| 8773 | if (!wc->update_ref || |
| 8774 | generation <= root->root_key.offset) |
| 8775 | goto skip; |
| 8776 | |
| 8777 | btrfs_node_key_to_cpu(path->nodes[level], &key, |
| 8778 | path->slots[level]); |
| 8779 | ret = btrfs_comp_cpu_keys(&key, &wc->update_progress); |
| 8780 | if (ret < 0) |
| 8781 | goto skip; |
| 8782 | |
| 8783 | wc->stage = UPDATE_BACKREF; |
| 8784 | wc->shared_level = level - 1; |
| 8785 | } |
| 8786 | } else { |
| 8787 | if (level == 1 && |
| 8788 | (wc->flags[0] & BTRFS_BLOCK_FLAG_FULL_BACKREF)) |
| 8789 | goto skip; |
| 8790 | } |
| 8791 | |
| 8792 | if (!btrfs_buffer_uptodate(next, generation, 0)) { |
| 8793 | btrfs_tree_unlock(next); |
| 8794 | free_extent_buffer(next); |
| 8795 | next = NULL; |
| 8796 | *lookup_info = 1; |
| 8797 | } |
| 8798 | |
| 8799 | if (!next) { |
| 8800 | if (reada && level == 1) |
| 8801 | reada_walk_down(trans, root, wc, path); |
| 8802 | next = read_tree_block(fs_info, bytenr, generation, level - 1, |
| 8803 | &first_key); |
| 8804 | if (IS_ERR(next)) { |
| 8805 | return PTR_ERR(next); |
| 8806 | } else if (!extent_buffer_uptodate(next)) { |
| 8807 | free_extent_buffer(next); |
| 8808 | return -EIO; |
| 8809 | } |
| 8810 | btrfs_tree_lock(next); |
| 8811 | btrfs_set_lock_blocking(next); |
| 8812 | } |
| 8813 | |
| 8814 | level--; |
| 8815 | ASSERT(level == btrfs_header_level(next)); |
| 8816 | if (level != btrfs_header_level(next)) { |
| 8817 | btrfs_err(root->fs_info, "mismatched level"); |
| 8818 | ret = -EIO; |
| 8819 | goto out_unlock; |
| 8820 | } |
| 8821 | path->nodes[level] = next; |
| 8822 | path->slots[level] = 0; |
| 8823 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
| 8824 | wc->level = level; |
| 8825 | if (wc->level == 1) |
| 8826 | wc->reada_slot = 0; |
| 8827 | return 0; |
| 8828 | skip: |
| 8829 | wc->refs[level - 1] = 0; |
| 8830 | wc->flags[level - 1] = 0; |
| 8831 | if (wc->stage == DROP_REFERENCE) { |
| 8832 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) { |
| 8833 | parent = path->nodes[level]->start; |
| 8834 | } else { |
| 8835 | ASSERT(root->root_key.objectid == |
| 8836 | btrfs_header_owner(path->nodes[level])); |
| 8837 | if (root->root_key.objectid != |
| 8838 | btrfs_header_owner(path->nodes[level])) { |
| 8839 | btrfs_err(root->fs_info, |
| 8840 | "mismatched block owner"); |
| 8841 | ret = -EIO; |
| 8842 | goto out_unlock; |
| 8843 | } |
| 8844 | parent = 0; |
| 8845 | } |
| 8846 | |
| 8847 | if (need_account) { |
| 8848 | ret = btrfs_qgroup_trace_subtree(trans, root, next, |
| 8849 | generation, level - 1); |
| 8850 | if (ret) { |
| 8851 | btrfs_err_rl(fs_info, |
| 8852 | "Error %d accounting shared subtree. Quota is out of sync, rescan required.", |
| 8853 | ret); |
| 8854 | } |
| 8855 | } |
| 8856 | ret = btrfs_free_extent(trans, root, bytenr, blocksize, |
| 8857 | parent, root->root_key.objectid, |
| 8858 | level - 1, 0); |
| 8859 | if (ret) |
| 8860 | goto out_unlock; |
| 8861 | } |
| 8862 | |
| 8863 | *lookup_info = 1; |
| 8864 | ret = 1; |
| 8865 | |
| 8866 | out_unlock: |
| 8867 | btrfs_tree_unlock(next); |
| 8868 | free_extent_buffer(next); |
| 8869 | |
| 8870 | return ret; |
| 8871 | } |
| 8872 | |
| 8873 | /* |
| 8874 | * helper to process tree block while walking up the tree. |
| 8875 | * |
| 8876 | * when wc->stage == DROP_REFERENCE, this function drops |
| 8877 | * reference count on the block. |
| 8878 | * |
| 8879 | * when wc->stage == UPDATE_BACKREF, this function changes |
| 8880 | * wc->stage back to DROP_REFERENCE if we changed wc->stage |
| 8881 | * to UPDATE_BACKREF previously while processing the block. |
| 8882 | * |
| 8883 | * NOTE: return value 1 means we should stop walking up. |
| 8884 | */ |
| 8885 | static noinline int walk_up_proc(struct btrfs_trans_handle *trans, |
| 8886 | struct btrfs_root *root, |
| 8887 | struct btrfs_path *path, |
| 8888 | struct walk_control *wc) |
| 8889 | { |
| 8890 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 8891 | int ret; |
| 8892 | int level = wc->level; |
| 8893 | struct extent_buffer *eb = path->nodes[level]; |
| 8894 | u64 parent = 0; |
| 8895 | |
| 8896 | if (wc->stage == UPDATE_BACKREF) { |
| 8897 | BUG_ON(wc->shared_level < level); |
| 8898 | if (level < wc->shared_level) |
| 8899 | goto out; |
| 8900 | |
| 8901 | ret = find_next_key(path, level + 1, &wc->update_progress); |
| 8902 | if (ret > 0) |
| 8903 | wc->update_ref = 0; |
| 8904 | |
| 8905 | wc->stage = DROP_REFERENCE; |
| 8906 | wc->shared_level = -1; |
| 8907 | path->slots[level] = 0; |
| 8908 | |
| 8909 | /* |
| 8910 | * check reference count again if the block isn't locked. |
| 8911 | * we should start walking down the tree again if reference |
| 8912 | * count is one. |
| 8913 | */ |
| 8914 | if (!path->locks[level]) { |
| 8915 | BUG_ON(level == 0); |
| 8916 | btrfs_tree_lock(eb); |
| 8917 | btrfs_set_lock_blocking(eb); |
| 8918 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
| 8919 | |
| 8920 | ret = btrfs_lookup_extent_info(trans, fs_info, |
| 8921 | eb->start, level, 1, |
| 8922 | &wc->refs[level], |
| 8923 | &wc->flags[level]); |
| 8924 | if (ret < 0) { |
| 8925 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
| 8926 | path->locks[level] = 0; |
| 8927 | return ret; |
| 8928 | } |
| 8929 | BUG_ON(wc->refs[level] == 0); |
| 8930 | if (wc->refs[level] == 1) { |
| 8931 | btrfs_tree_unlock_rw(eb, path->locks[level]); |
| 8932 | path->locks[level] = 0; |
| 8933 | return 1; |
| 8934 | } |
| 8935 | } |
| 8936 | } |
| 8937 | |
| 8938 | /* wc->stage == DROP_REFERENCE */ |
| 8939 | BUG_ON(wc->refs[level] > 1 && !path->locks[level]); |
| 8940 | |
| 8941 | if (wc->refs[level] == 1) { |
| 8942 | if (level == 0) { |
| 8943 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) |
| 8944 | ret = btrfs_dec_ref(trans, root, eb, 1); |
| 8945 | else |
| 8946 | ret = btrfs_dec_ref(trans, root, eb, 0); |
| 8947 | BUG_ON(ret); /* -ENOMEM */ |
| 8948 | ret = btrfs_qgroup_trace_leaf_items(trans, fs_info, eb); |
| 8949 | if (ret) { |
| 8950 | btrfs_err_rl(fs_info, |
| 8951 | "error %d accounting leaf items. Quota is out of sync, rescan required.", |
| 8952 | ret); |
| 8953 | } |
| 8954 | } |
| 8955 | /* make block locked assertion in clean_tree_block happy */ |
| 8956 | if (!path->locks[level] && |
| 8957 | btrfs_header_generation(eb) == trans->transid) { |
| 8958 | btrfs_tree_lock(eb); |
| 8959 | btrfs_set_lock_blocking(eb); |
| 8960 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
| 8961 | } |
| 8962 | clean_tree_block(fs_info, eb); |
| 8963 | } |
| 8964 | |
| 8965 | if (eb == root->node) { |
| 8966 | if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF) |
| 8967 | parent = eb->start; |
| 8968 | else |
| 8969 | BUG_ON(root->root_key.objectid != |
| 8970 | btrfs_header_owner(eb)); |
| 8971 | } else { |
| 8972 | if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF) |
| 8973 | parent = path->nodes[level + 1]->start; |
| 8974 | else |
| 8975 | BUG_ON(root->root_key.objectid != |
| 8976 | btrfs_header_owner(path->nodes[level + 1])); |
| 8977 | } |
| 8978 | |
| 8979 | btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1); |
| 8980 | out: |
| 8981 | wc->refs[level] = 0; |
| 8982 | wc->flags[level] = 0; |
| 8983 | return 0; |
| 8984 | } |
| 8985 | |
| 8986 | static noinline int walk_down_tree(struct btrfs_trans_handle *trans, |
| 8987 | struct btrfs_root *root, |
| 8988 | struct btrfs_path *path, |
| 8989 | struct walk_control *wc) |
| 8990 | { |
| 8991 | int level = wc->level; |
| 8992 | int lookup_info = 1; |
| 8993 | int ret; |
| 8994 | |
| 8995 | while (level >= 0) { |
| 8996 | ret = walk_down_proc(trans, root, path, wc, lookup_info); |
| 8997 | if (ret > 0) |
| 8998 | break; |
| 8999 | |
| 9000 | if (level == 0) |
| 9001 | break; |
| 9002 | |
| 9003 | if (path->slots[level] >= |
| 9004 | btrfs_header_nritems(path->nodes[level])) |
| 9005 | break; |
| 9006 | |
| 9007 | ret = do_walk_down(trans, root, path, wc, &lookup_info); |
| 9008 | if (ret > 0) { |
| 9009 | path->slots[level]++; |
| 9010 | continue; |
| 9011 | } else if (ret < 0) |
| 9012 | return ret; |
| 9013 | level = wc->level; |
| 9014 | } |
| 9015 | return 0; |
| 9016 | } |
| 9017 | |
| 9018 | static noinline int walk_up_tree(struct btrfs_trans_handle *trans, |
| 9019 | struct btrfs_root *root, |
| 9020 | struct btrfs_path *path, |
| 9021 | struct walk_control *wc, int max_level) |
| 9022 | { |
| 9023 | int level = wc->level; |
| 9024 | int ret; |
| 9025 | |
| 9026 | path->slots[level] = btrfs_header_nritems(path->nodes[level]); |
| 9027 | while (level < max_level && path->nodes[level]) { |
| 9028 | wc->level = level; |
| 9029 | if (path->slots[level] + 1 < |
| 9030 | btrfs_header_nritems(path->nodes[level])) { |
| 9031 | path->slots[level]++; |
| 9032 | return 0; |
| 9033 | } else { |
| 9034 | ret = walk_up_proc(trans, root, path, wc); |
| 9035 | if (ret > 0) |
| 9036 | return 0; |
| 9037 | |
| 9038 | if (path->locks[level]) { |
| 9039 | btrfs_tree_unlock_rw(path->nodes[level], |
| 9040 | path->locks[level]); |
| 9041 | path->locks[level] = 0; |
| 9042 | } |
| 9043 | free_extent_buffer(path->nodes[level]); |
| 9044 | path->nodes[level] = NULL; |
| 9045 | level++; |
| 9046 | } |
| 9047 | } |
| 9048 | return 1; |
| 9049 | } |
| 9050 | |
| 9051 | /* |
| 9052 | * drop a subvolume tree. |
| 9053 | * |
| 9054 | * this function traverses the tree freeing any blocks that only |
| 9055 | * referenced by the tree. |
| 9056 | * |
| 9057 | * when a shared tree block is found. this function decreases its |
| 9058 | * reference count by one. if update_ref is true, this function |
| 9059 | * also make sure backrefs for the shared block and all lower level |
| 9060 | * blocks are properly updated. |
| 9061 | * |
| 9062 | * If called with for_reloc == 0, may exit early with -EAGAIN |
| 9063 | */ |
| 9064 | int btrfs_drop_snapshot(struct btrfs_root *root, |
| 9065 | struct btrfs_block_rsv *block_rsv, int update_ref, |
| 9066 | int for_reloc) |
| 9067 | { |
| 9068 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 9069 | struct btrfs_path *path; |
| 9070 | struct btrfs_trans_handle *trans; |
| 9071 | struct btrfs_root *tree_root = fs_info->tree_root; |
| 9072 | struct btrfs_root_item *root_item = &root->root_item; |
| 9073 | struct walk_control *wc; |
| 9074 | struct btrfs_key key; |
| 9075 | int err = 0; |
| 9076 | int ret; |
| 9077 | int level; |
| 9078 | bool root_dropped = false; |
| 9079 | |
| 9080 | btrfs_debug(fs_info, "Drop subvolume %llu", root->objectid); |
| 9081 | |
| 9082 | path = btrfs_alloc_path(); |
| 9083 | if (!path) { |
| 9084 | err = -ENOMEM; |
| 9085 | goto out; |
| 9086 | } |
| 9087 | |
| 9088 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
| 9089 | if (!wc) { |
| 9090 | btrfs_free_path(path); |
| 9091 | err = -ENOMEM; |
| 9092 | goto out; |
| 9093 | } |
| 9094 | |
| 9095 | trans = btrfs_start_transaction(tree_root, 0); |
| 9096 | if (IS_ERR(trans)) { |
| 9097 | err = PTR_ERR(trans); |
| 9098 | goto out_free; |
| 9099 | } |
| 9100 | |
| 9101 | if (block_rsv) |
| 9102 | trans->block_rsv = block_rsv; |
| 9103 | |
| 9104 | if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { |
| 9105 | level = btrfs_header_level(root->node); |
| 9106 | path->nodes[level] = btrfs_lock_root_node(root); |
| 9107 | btrfs_set_lock_blocking(path->nodes[level]); |
| 9108 | path->slots[level] = 0; |
| 9109 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
| 9110 | memset(&wc->update_progress, 0, |
| 9111 | sizeof(wc->update_progress)); |
| 9112 | } else { |
| 9113 | btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); |
| 9114 | memcpy(&wc->update_progress, &key, |
| 9115 | sizeof(wc->update_progress)); |
| 9116 | |
| 9117 | level = root_item->drop_level; |
| 9118 | BUG_ON(level == 0); |
| 9119 | path->lowest_level = level; |
| 9120 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
| 9121 | path->lowest_level = 0; |
| 9122 | if (ret < 0) { |
| 9123 | err = ret; |
| 9124 | goto out_end_trans; |
| 9125 | } |
| 9126 | WARN_ON(ret > 0); |
| 9127 | |
| 9128 | /* |
| 9129 | * unlock our path, this is safe because only this |
| 9130 | * function is allowed to delete this snapshot |
| 9131 | */ |
| 9132 | btrfs_unlock_up_safe(path, 0); |
| 9133 | |
| 9134 | level = btrfs_header_level(root->node); |
| 9135 | while (1) { |
| 9136 | btrfs_tree_lock(path->nodes[level]); |
| 9137 | btrfs_set_lock_blocking(path->nodes[level]); |
| 9138 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
| 9139 | |
| 9140 | ret = btrfs_lookup_extent_info(trans, fs_info, |
| 9141 | path->nodes[level]->start, |
| 9142 | level, 1, &wc->refs[level], |
| 9143 | &wc->flags[level]); |
| 9144 | if (ret < 0) { |
| 9145 | err = ret; |
| 9146 | goto out_end_trans; |
| 9147 | } |
| 9148 | BUG_ON(wc->refs[level] == 0); |
| 9149 | |
| 9150 | if (level == root_item->drop_level) |
| 9151 | break; |
| 9152 | |
| 9153 | btrfs_tree_unlock(path->nodes[level]); |
| 9154 | path->locks[level] = 0; |
| 9155 | WARN_ON(wc->refs[level] != 1); |
| 9156 | level--; |
| 9157 | } |
| 9158 | } |
| 9159 | |
| 9160 | wc->level = level; |
| 9161 | wc->shared_level = -1; |
| 9162 | wc->stage = DROP_REFERENCE; |
| 9163 | wc->update_ref = update_ref; |
| 9164 | wc->keep_locks = 0; |
| 9165 | wc->for_reloc = for_reloc; |
| 9166 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
| 9167 | |
| 9168 | while (1) { |
| 9169 | |
| 9170 | ret = walk_down_tree(trans, root, path, wc); |
| 9171 | if (ret < 0) { |
| 9172 | err = ret; |
| 9173 | break; |
| 9174 | } |
| 9175 | |
| 9176 | ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL); |
| 9177 | if (ret < 0) { |
| 9178 | err = ret; |
| 9179 | break; |
| 9180 | } |
| 9181 | |
| 9182 | if (ret > 0) { |
| 9183 | BUG_ON(wc->stage != DROP_REFERENCE); |
| 9184 | break; |
| 9185 | } |
| 9186 | |
| 9187 | if (wc->stage == DROP_REFERENCE) { |
| 9188 | level = wc->level; |
| 9189 | btrfs_node_key(path->nodes[level], |
| 9190 | &root_item->drop_progress, |
| 9191 | path->slots[level]); |
| 9192 | root_item->drop_level = level; |
| 9193 | } |
| 9194 | |
| 9195 | BUG_ON(wc->level == 0); |
| 9196 | if (btrfs_should_end_transaction(trans) || |
| 9197 | (!for_reloc && btrfs_need_cleaner_sleep(fs_info))) { |
| 9198 | ret = btrfs_update_root(trans, tree_root, |
| 9199 | &root->root_key, |
| 9200 | root_item); |
| 9201 | if (ret) { |
| 9202 | btrfs_abort_transaction(trans, ret); |
| 9203 | err = ret; |
| 9204 | goto out_end_trans; |
| 9205 | } |
| 9206 | |
| 9207 | btrfs_end_transaction_throttle(trans); |
| 9208 | if (!for_reloc && btrfs_need_cleaner_sleep(fs_info)) { |
| 9209 | btrfs_debug(fs_info, |
| 9210 | "drop snapshot early exit"); |
| 9211 | err = -EAGAIN; |
| 9212 | goto out_free; |
| 9213 | } |
| 9214 | |
| 9215 | trans = btrfs_start_transaction(tree_root, 0); |
| 9216 | if (IS_ERR(trans)) { |
| 9217 | err = PTR_ERR(trans); |
| 9218 | goto out_free; |
| 9219 | } |
| 9220 | if (block_rsv) |
| 9221 | trans->block_rsv = block_rsv; |
| 9222 | } |
| 9223 | } |
| 9224 | btrfs_release_path(path); |
| 9225 | if (err) |
| 9226 | goto out_end_trans; |
| 9227 | |
| 9228 | ret = btrfs_del_root(trans, fs_info, &root->root_key); |
| 9229 | if (ret) { |
| 9230 | btrfs_abort_transaction(trans, ret); |
| 9231 | err = ret; |
| 9232 | goto out_end_trans; |
| 9233 | } |
| 9234 | |
| 9235 | if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { |
| 9236 | ret = btrfs_find_root(tree_root, &root->root_key, path, |
| 9237 | NULL, NULL); |
| 9238 | if (ret < 0) { |
| 9239 | btrfs_abort_transaction(trans, ret); |
| 9240 | err = ret; |
| 9241 | goto out_end_trans; |
| 9242 | } else if (ret > 0) { |
| 9243 | /* if we fail to delete the orphan item this time |
| 9244 | * around, it'll get picked up the next time. |
| 9245 | * |
| 9246 | * The most common failure here is just -ENOENT. |
| 9247 | */ |
| 9248 | btrfs_del_orphan_item(trans, tree_root, |
| 9249 | root->root_key.objectid); |
| 9250 | } |
| 9251 | } |
| 9252 | |
| 9253 | if (test_bit(BTRFS_ROOT_IN_RADIX, &root->state)) { |
| 9254 | btrfs_add_dropped_root(trans, root); |
| 9255 | } else { |
| 9256 | free_extent_buffer(root->node); |
| 9257 | free_extent_buffer(root->commit_root); |
| 9258 | btrfs_put_fs_root(root); |
| 9259 | } |
| 9260 | root_dropped = true; |
| 9261 | out_end_trans: |
| 9262 | btrfs_end_transaction_throttle(trans); |
| 9263 | out_free: |
| 9264 | kfree(wc); |
| 9265 | btrfs_free_path(path); |
| 9266 | out: |
| 9267 | /* |
| 9268 | * So if we need to stop dropping the snapshot for whatever reason we |
| 9269 | * need to make sure to add it back to the dead root list so that we |
| 9270 | * keep trying to do the work later. This also cleans up roots if we |
| 9271 | * don't have it in the radix (like when we recover after a power fail |
| 9272 | * or unmount) so we don't leak memory. |
| 9273 | */ |
| 9274 | if (!for_reloc && !root_dropped) |
| 9275 | btrfs_add_dead_root(root); |
| 9276 | if (err && err != -EAGAIN) |
| 9277 | btrfs_handle_fs_error(fs_info, err, NULL); |
| 9278 | return err; |
| 9279 | } |
| 9280 | |
| 9281 | /* |
| 9282 | * drop subtree rooted at tree block 'node'. |
| 9283 | * |
| 9284 | * NOTE: this function will unlock and release tree block 'node' |
| 9285 | * only used by relocation code |
| 9286 | */ |
| 9287 | int btrfs_drop_subtree(struct btrfs_trans_handle *trans, |
| 9288 | struct btrfs_root *root, |
| 9289 | struct extent_buffer *node, |
| 9290 | struct extent_buffer *parent) |
| 9291 | { |
| 9292 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 9293 | struct btrfs_path *path; |
| 9294 | struct walk_control *wc; |
| 9295 | int level; |
| 9296 | int parent_level; |
| 9297 | int ret = 0; |
| 9298 | int wret; |
| 9299 | |
| 9300 | BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); |
| 9301 | |
| 9302 | path = btrfs_alloc_path(); |
| 9303 | if (!path) |
| 9304 | return -ENOMEM; |
| 9305 | |
| 9306 | wc = kzalloc(sizeof(*wc), GFP_NOFS); |
| 9307 | if (!wc) { |
| 9308 | btrfs_free_path(path); |
| 9309 | return -ENOMEM; |
| 9310 | } |
| 9311 | |
| 9312 | btrfs_assert_tree_locked(parent); |
| 9313 | parent_level = btrfs_header_level(parent); |
| 9314 | extent_buffer_get(parent); |
| 9315 | path->nodes[parent_level] = parent; |
| 9316 | path->slots[parent_level] = btrfs_header_nritems(parent); |
| 9317 | |
| 9318 | btrfs_assert_tree_locked(node); |
| 9319 | level = btrfs_header_level(node); |
| 9320 | path->nodes[level] = node; |
| 9321 | path->slots[level] = 0; |
| 9322 | path->locks[level] = BTRFS_WRITE_LOCK_BLOCKING; |
| 9323 | |
| 9324 | wc->refs[parent_level] = 1; |
| 9325 | wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF; |
| 9326 | wc->level = level; |
| 9327 | wc->shared_level = -1; |
| 9328 | wc->stage = DROP_REFERENCE; |
| 9329 | wc->update_ref = 0; |
| 9330 | wc->keep_locks = 1; |
| 9331 | wc->for_reloc = 1; |
| 9332 | wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(fs_info); |
| 9333 | |
| 9334 | while (1) { |
| 9335 | wret = walk_down_tree(trans, root, path, wc); |
| 9336 | if (wret < 0) { |
| 9337 | ret = wret; |
| 9338 | break; |
| 9339 | } |
| 9340 | |
| 9341 | wret = walk_up_tree(trans, root, path, wc, parent_level); |
| 9342 | if (wret < 0) |
| 9343 | ret = wret; |
| 9344 | if (wret != 0) |
| 9345 | break; |
| 9346 | } |
| 9347 | |
| 9348 | kfree(wc); |
| 9349 | btrfs_free_path(path); |
| 9350 | return ret; |
| 9351 | } |
| 9352 | |
| 9353 | static u64 update_block_group_flags(struct btrfs_fs_info *fs_info, u64 flags) |
| 9354 | { |
| 9355 | u64 num_devices; |
| 9356 | u64 stripped; |
| 9357 | |
| 9358 | /* |
| 9359 | * if restripe for this chunk_type is on pick target profile and |
| 9360 | * return, otherwise do the usual balance |
| 9361 | */ |
| 9362 | stripped = get_restripe_target(fs_info, flags); |
| 9363 | if (stripped) |
| 9364 | return extended_to_chunk(stripped); |
| 9365 | |
| 9366 | num_devices = fs_info->fs_devices->rw_devices; |
| 9367 | |
| 9368 | stripped = BTRFS_BLOCK_GROUP_RAID0 | |
| 9369 | BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 | |
| 9370 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10; |
| 9371 | |
| 9372 | if (num_devices == 1) { |
| 9373 | stripped |= BTRFS_BLOCK_GROUP_DUP; |
| 9374 | stripped = flags & ~stripped; |
| 9375 | |
| 9376 | /* turn raid0 into single device chunks */ |
| 9377 | if (flags & BTRFS_BLOCK_GROUP_RAID0) |
| 9378 | return stripped; |
| 9379 | |
| 9380 | /* turn mirroring into duplication */ |
| 9381 | if (flags & (BTRFS_BLOCK_GROUP_RAID1 | |
| 9382 | BTRFS_BLOCK_GROUP_RAID10)) |
| 9383 | return stripped | BTRFS_BLOCK_GROUP_DUP; |
| 9384 | } else { |
| 9385 | /* they already had raid on here, just return */ |
| 9386 | if (flags & stripped) |
| 9387 | return flags; |
| 9388 | |
| 9389 | stripped |= BTRFS_BLOCK_GROUP_DUP; |
| 9390 | stripped = flags & ~stripped; |
| 9391 | |
| 9392 | /* switch duplicated blocks with raid1 */ |
| 9393 | if (flags & BTRFS_BLOCK_GROUP_DUP) |
| 9394 | return stripped | BTRFS_BLOCK_GROUP_RAID1; |
| 9395 | |
| 9396 | /* this is drive concat, leave it alone */ |
| 9397 | } |
| 9398 | |
| 9399 | return flags; |
| 9400 | } |
| 9401 | |
| 9402 | static int inc_block_group_ro(struct btrfs_block_group_cache *cache, int force) |
| 9403 | { |
| 9404 | struct btrfs_space_info *sinfo = cache->space_info; |
| 9405 | u64 num_bytes; |
| 9406 | u64 min_allocable_bytes; |
| 9407 | int ret = -ENOSPC; |
| 9408 | |
| 9409 | /* |
| 9410 | * We need some metadata space and system metadata space for |
| 9411 | * allocating chunks in some corner cases until we force to set |
| 9412 | * it to be readonly. |
| 9413 | */ |
| 9414 | if ((sinfo->flags & |
| 9415 | (BTRFS_BLOCK_GROUP_SYSTEM | BTRFS_BLOCK_GROUP_METADATA)) && |
| 9416 | !force) |
| 9417 | min_allocable_bytes = SZ_1M; |
| 9418 | else |
| 9419 | min_allocable_bytes = 0; |
| 9420 | |
| 9421 | spin_lock(&sinfo->lock); |
| 9422 | spin_lock(&cache->lock); |
| 9423 | |
| 9424 | if (cache->ro) { |
| 9425 | cache->ro++; |
| 9426 | ret = 0; |
| 9427 | goto out; |
| 9428 | } |
| 9429 | |
| 9430 | num_bytes = cache->key.offset - cache->reserved - cache->pinned - |
| 9431 | cache->bytes_super - btrfs_block_group_used(&cache->item); |
| 9432 | |
| 9433 | if (btrfs_space_info_used(sinfo, true) + num_bytes + |
| 9434 | min_allocable_bytes <= sinfo->total_bytes) { |
| 9435 | sinfo->bytes_readonly += num_bytes; |
| 9436 | cache->ro++; |
| 9437 | list_add_tail(&cache->ro_list, &sinfo->ro_bgs); |
| 9438 | ret = 0; |
| 9439 | } |
| 9440 | out: |
| 9441 | spin_unlock(&cache->lock); |
| 9442 | spin_unlock(&sinfo->lock); |
| 9443 | return ret; |
| 9444 | } |
| 9445 | |
| 9446 | int btrfs_inc_block_group_ro(struct btrfs_fs_info *fs_info, |
| 9447 | struct btrfs_block_group_cache *cache) |
| 9448 | |
| 9449 | { |
| 9450 | struct btrfs_trans_handle *trans; |
| 9451 | u64 alloc_flags; |
| 9452 | int ret; |
| 9453 | |
| 9454 | again: |
| 9455 | trans = btrfs_join_transaction(fs_info->extent_root); |
| 9456 | if (IS_ERR(trans)) |
| 9457 | return PTR_ERR(trans); |
| 9458 | |
| 9459 | /* |
| 9460 | * we're not allowed to set block groups readonly after the dirty |
| 9461 | * block groups cache has started writing. If it already started, |
| 9462 | * back off and let this transaction commit |
| 9463 | */ |
| 9464 | mutex_lock(&fs_info->ro_block_group_mutex); |
| 9465 | if (test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &trans->transaction->flags)) { |
| 9466 | u64 transid = trans->transid; |
| 9467 | |
| 9468 | mutex_unlock(&fs_info->ro_block_group_mutex); |
| 9469 | btrfs_end_transaction(trans); |
| 9470 | |
| 9471 | ret = btrfs_wait_for_commit(fs_info, transid); |
| 9472 | if (ret) |
| 9473 | return ret; |
| 9474 | goto again; |
| 9475 | } |
| 9476 | |
| 9477 | /* |
| 9478 | * if we are changing raid levels, try to allocate a corresponding |
| 9479 | * block group with the new raid level. |
| 9480 | */ |
| 9481 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
| 9482 | if (alloc_flags != cache->flags) { |
| 9483 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, |
| 9484 | CHUNK_ALLOC_FORCE); |
| 9485 | /* |
| 9486 | * ENOSPC is allowed here, we may have enough space |
| 9487 | * already allocated at the new raid level to |
| 9488 | * carry on |
| 9489 | */ |
| 9490 | if (ret == -ENOSPC) |
| 9491 | ret = 0; |
| 9492 | if (ret < 0) |
| 9493 | goto out; |
| 9494 | } |
| 9495 | |
| 9496 | ret = inc_block_group_ro(cache, 0); |
| 9497 | if (!ret) |
| 9498 | goto out; |
| 9499 | alloc_flags = get_alloc_profile(fs_info, cache->space_info->flags); |
| 9500 | ret = do_chunk_alloc(trans, fs_info, alloc_flags, |
| 9501 | CHUNK_ALLOC_FORCE); |
| 9502 | if (ret < 0) |
| 9503 | goto out; |
| 9504 | ret = inc_block_group_ro(cache, 0); |
| 9505 | out: |
| 9506 | if (cache->flags & BTRFS_BLOCK_GROUP_SYSTEM) { |
| 9507 | alloc_flags = update_block_group_flags(fs_info, cache->flags); |
| 9508 | mutex_lock(&fs_info->chunk_mutex); |
| 9509 | check_system_chunk(trans, fs_info, alloc_flags); |
| 9510 | mutex_unlock(&fs_info->chunk_mutex); |
| 9511 | } |
| 9512 | mutex_unlock(&fs_info->ro_block_group_mutex); |
| 9513 | |
| 9514 | btrfs_end_transaction(trans); |
| 9515 | return ret; |
| 9516 | } |
| 9517 | |
| 9518 | int btrfs_force_chunk_alloc(struct btrfs_trans_handle *trans, |
| 9519 | struct btrfs_fs_info *fs_info, u64 type) |
| 9520 | { |
| 9521 | u64 alloc_flags = get_alloc_profile(fs_info, type); |
| 9522 | |
| 9523 | return do_chunk_alloc(trans, fs_info, alloc_flags, CHUNK_ALLOC_FORCE); |
| 9524 | } |
| 9525 | |
| 9526 | /* |
| 9527 | * helper to account the unused space of all the readonly block group in the |
| 9528 | * space_info. takes mirrors into account. |
| 9529 | */ |
| 9530 | u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo) |
| 9531 | { |
| 9532 | struct btrfs_block_group_cache *block_group; |
| 9533 | u64 free_bytes = 0; |
| 9534 | int factor; |
| 9535 | |
| 9536 | /* It's df, we don't care if it's racy */ |
| 9537 | if (list_empty(&sinfo->ro_bgs)) |
| 9538 | return 0; |
| 9539 | |
| 9540 | spin_lock(&sinfo->lock); |
| 9541 | list_for_each_entry(block_group, &sinfo->ro_bgs, ro_list) { |
| 9542 | spin_lock(&block_group->lock); |
| 9543 | |
| 9544 | if (!block_group->ro) { |
| 9545 | spin_unlock(&block_group->lock); |
| 9546 | continue; |
| 9547 | } |
| 9548 | |
| 9549 | if (block_group->flags & (BTRFS_BLOCK_GROUP_RAID1 | |
| 9550 | BTRFS_BLOCK_GROUP_RAID10 | |
| 9551 | BTRFS_BLOCK_GROUP_DUP)) |
| 9552 | factor = 2; |
| 9553 | else |
| 9554 | factor = 1; |
| 9555 | |
| 9556 | free_bytes += (block_group->key.offset - |
| 9557 | btrfs_block_group_used(&block_group->item)) * |
| 9558 | factor; |
| 9559 | |
| 9560 | spin_unlock(&block_group->lock); |
| 9561 | } |
| 9562 | spin_unlock(&sinfo->lock); |
| 9563 | |
| 9564 | return free_bytes; |
| 9565 | } |
| 9566 | |
| 9567 | void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache) |
| 9568 | { |
| 9569 | struct btrfs_space_info *sinfo = cache->space_info; |
| 9570 | u64 num_bytes; |
| 9571 | |
| 9572 | BUG_ON(!cache->ro); |
| 9573 | |
| 9574 | spin_lock(&sinfo->lock); |
| 9575 | spin_lock(&cache->lock); |
| 9576 | if (!--cache->ro) { |
| 9577 | num_bytes = cache->key.offset - cache->reserved - |
| 9578 | cache->pinned - cache->bytes_super - |
| 9579 | btrfs_block_group_used(&cache->item); |
| 9580 | sinfo->bytes_readonly -= num_bytes; |
| 9581 | list_del_init(&cache->ro_list); |
| 9582 | } |
| 9583 | spin_unlock(&cache->lock); |
| 9584 | spin_unlock(&sinfo->lock); |
| 9585 | } |
| 9586 | |
| 9587 | /* |
| 9588 | * checks to see if its even possible to relocate this block group. |
| 9589 | * |
| 9590 | * @return - -1 if it's not a good idea to relocate this block group, 0 if its |
| 9591 | * ok to go ahead and try. |
| 9592 | */ |
| 9593 | int btrfs_can_relocate(struct btrfs_fs_info *fs_info, u64 bytenr) |
| 9594 | { |
| 9595 | struct btrfs_root *root = fs_info->extent_root; |
| 9596 | struct btrfs_block_group_cache *block_group; |
| 9597 | struct btrfs_space_info *space_info; |
| 9598 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
| 9599 | struct btrfs_device *device; |
| 9600 | struct btrfs_trans_handle *trans; |
| 9601 | u64 min_free; |
| 9602 | u64 dev_min = 1; |
| 9603 | u64 dev_nr = 0; |
| 9604 | u64 target; |
| 9605 | int debug; |
| 9606 | int index; |
| 9607 | int full = 0; |
| 9608 | int ret = 0; |
| 9609 | |
| 9610 | debug = btrfs_test_opt(fs_info, ENOSPC_DEBUG); |
| 9611 | |
| 9612 | block_group = btrfs_lookup_block_group(fs_info, bytenr); |
| 9613 | |
| 9614 | /* odd, couldn't find the block group, leave it alone */ |
| 9615 | if (!block_group) { |
| 9616 | if (debug) |
| 9617 | btrfs_warn(fs_info, |
| 9618 | "can't find block group for bytenr %llu", |
| 9619 | bytenr); |
| 9620 | return -1; |
| 9621 | } |
| 9622 | |
| 9623 | min_free = btrfs_block_group_used(&block_group->item); |
| 9624 | |
| 9625 | /* no bytes used, we're good */ |
| 9626 | if (!min_free) |
| 9627 | goto out; |
| 9628 | |
| 9629 | space_info = block_group->space_info; |
| 9630 | spin_lock(&space_info->lock); |
| 9631 | |
| 9632 | full = space_info->full; |
| 9633 | |
| 9634 | /* |
| 9635 | * if this is the last block group we have in this space, we can't |
| 9636 | * relocate it unless we're able to allocate a new chunk below. |
| 9637 | * |
| 9638 | * Otherwise, we need to make sure we have room in the space to handle |
| 9639 | * all of the extents from this block group. If we can, we're good |
| 9640 | */ |
| 9641 | if ((space_info->total_bytes != block_group->key.offset) && |
| 9642 | (btrfs_space_info_used(space_info, false) + min_free < |
| 9643 | space_info->total_bytes)) { |
| 9644 | spin_unlock(&space_info->lock); |
| 9645 | goto out; |
| 9646 | } |
| 9647 | spin_unlock(&space_info->lock); |
| 9648 | |
| 9649 | /* |
| 9650 | * ok we don't have enough space, but maybe we have free space on our |
| 9651 | * devices to allocate new chunks for relocation, so loop through our |
| 9652 | * alloc devices and guess if we have enough space. if this block |
| 9653 | * group is going to be restriped, run checks against the target |
| 9654 | * profile instead of the current one. |
| 9655 | */ |
| 9656 | ret = -1; |
| 9657 | |
| 9658 | /* |
| 9659 | * index: |
| 9660 | * 0: raid10 |
| 9661 | * 1: raid1 |
| 9662 | * 2: dup |
| 9663 | * 3: raid0 |
| 9664 | * 4: single |
| 9665 | */ |
| 9666 | target = get_restripe_target(fs_info, block_group->flags); |
| 9667 | if (target) { |
| 9668 | index = btrfs_bg_flags_to_raid_index(extended_to_chunk(target)); |
| 9669 | } else { |
| 9670 | /* |
| 9671 | * this is just a balance, so if we were marked as full |
| 9672 | * we know there is no space for a new chunk |
| 9673 | */ |
| 9674 | if (full) { |
| 9675 | if (debug) |
| 9676 | btrfs_warn(fs_info, |
| 9677 | "no space to alloc new chunk for block group %llu", |
| 9678 | block_group->key.objectid); |
| 9679 | goto out; |
| 9680 | } |
| 9681 | |
| 9682 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
| 9683 | } |
| 9684 | |
| 9685 | if (index == BTRFS_RAID_RAID10) { |
| 9686 | dev_min = 4; |
| 9687 | /* Divide by 2 */ |
| 9688 | min_free >>= 1; |
| 9689 | } else if (index == BTRFS_RAID_RAID1) { |
| 9690 | dev_min = 2; |
| 9691 | } else if (index == BTRFS_RAID_DUP) { |
| 9692 | /* Multiply by 2 */ |
| 9693 | min_free <<= 1; |
| 9694 | } else if (index == BTRFS_RAID_RAID0) { |
| 9695 | dev_min = fs_devices->rw_devices; |
| 9696 | min_free = div64_u64(min_free, dev_min); |
| 9697 | } |
| 9698 | |
| 9699 | /* We need to do this so that we can look at pending chunks */ |
| 9700 | trans = btrfs_join_transaction(root); |
| 9701 | if (IS_ERR(trans)) { |
| 9702 | ret = PTR_ERR(trans); |
| 9703 | goto out; |
| 9704 | } |
| 9705 | |
| 9706 | mutex_lock(&fs_info->chunk_mutex); |
| 9707 | list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { |
| 9708 | u64 dev_offset; |
| 9709 | |
| 9710 | /* |
| 9711 | * check to make sure we can actually find a chunk with enough |
| 9712 | * space to fit our block group in. |
| 9713 | */ |
| 9714 | if (device->total_bytes > device->bytes_used + min_free && |
| 9715 | !test_bit(BTRFS_DEV_STATE_REPLACE_TGT, &device->dev_state)) { |
| 9716 | ret = find_free_dev_extent(trans, device, min_free, |
| 9717 | &dev_offset, NULL); |
| 9718 | if (!ret) |
| 9719 | dev_nr++; |
| 9720 | |
| 9721 | if (dev_nr >= dev_min) |
| 9722 | break; |
| 9723 | |
| 9724 | ret = -1; |
| 9725 | } |
| 9726 | } |
| 9727 | if (debug && ret == -1) |
| 9728 | btrfs_warn(fs_info, |
| 9729 | "no space to allocate a new chunk for block group %llu", |
| 9730 | block_group->key.objectid); |
| 9731 | mutex_unlock(&fs_info->chunk_mutex); |
| 9732 | btrfs_end_transaction(trans); |
| 9733 | out: |
| 9734 | btrfs_put_block_group(block_group); |
| 9735 | return ret; |
| 9736 | } |
| 9737 | |
| 9738 | static int find_first_block_group(struct btrfs_fs_info *fs_info, |
| 9739 | struct btrfs_path *path, |
| 9740 | struct btrfs_key *key) |
| 9741 | { |
| 9742 | struct btrfs_root *root = fs_info->extent_root; |
| 9743 | int ret = 0; |
| 9744 | struct btrfs_key found_key; |
| 9745 | struct extent_buffer *leaf; |
| 9746 | int slot; |
| 9747 | |
| 9748 | ret = btrfs_search_slot(NULL, root, key, path, 0, 0); |
| 9749 | if (ret < 0) |
| 9750 | goto out; |
| 9751 | |
| 9752 | while (1) { |
| 9753 | slot = path->slots[0]; |
| 9754 | leaf = path->nodes[0]; |
| 9755 | if (slot >= btrfs_header_nritems(leaf)) { |
| 9756 | ret = btrfs_next_leaf(root, path); |
| 9757 | if (ret == 0) |
| 9758 | continue; |
| 9759 | if (ret < 0) |
| 9760 | goto out; |
| 9761 | break; |
| 9762 | } |
| 9763 | btrfs_item_key_to_cpu(leaf, &found_key, slot); |
| 9764 | |
| 9765 | if (found_key.objectid >= key->objectid && |
| 9766 | found_key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) { |
| 9767 | struct extent_map_tree *em_tree; |
| 9768 | struct extent_map *em; |
| 9769 | |
| 9770 | em_tree = &root->fs_info->mapping_tree.map_tree; |
| 9771 | read_lock(&em_tree->lock); |
| 9772 | em = lookup_extent_mapping(em_tree, found_key.objectid, |
| 9773 | found_key.offset); |
| 9774 | read_unlock(&em_tree->lock); |
| 9775 | if (!em) { |
| 9776 | btrfs_err(fs_info, |
| 9777 | "logical %llu len %llu found bg but no related chunk", |
| 9778 | found_key.objectid, found_key.offset); |
| 9779 | ret = -ENOENT; |
| 9780 | } else { |
| 9781 | ret = 0; |
| 9782 | } |
| 9783 | free_extent_map(em); |
| 9784 | goto out; |
| 9785 | } |
| 9786 | path->slots[0]++; |
| 9787 | } |
| 9788 | out: |
| 9789 | return ret; |
| 9790 | } |
| 9791 | |
| 9792 | void btrfs_put_block_group_cache(struct btrfs_fs_info *info) |
| 9793 | { |
| 9794 | struct btrfs_block_group_cache *block_group; |
| 9795 | u64 last = 0; |
| 9796 | |
| 9797 | while (1) { |
| 9798 | struct inode *inode; |
| 9799 | |
| 9800 | block_group = btrfs_lookup_first_block_group(info, last); |
| 9801 | while (block_group) { |
| 9802 | spin_lock(&block_group->lock); |
| 9803 | if (block_group->iref) |
| 9804 | break; |
| 9805 | spin_unlock(&block_group->lock); |
| 9806 | block_group = next_block_group(info, block_group); |
| 9807 | } |
| 9808 | if (!block_group) { |
| 9809 | if (last == 0) |
| 9810 | break; |
| 9811 | last = 0; |
| 9812 | continue; |
| 9813 | } |
| 9814 | |
| 9815 | inode = block_group->inode; |
| 9816 | block_group->iref = 0; |
| 9817 | block_group->inode = NULL; |
| 9818 | spin_unlock(&block_group->lock); |
| 9819 | ASSERT(block_group->io_ctl.inode == NULL); |
| 9820 | iput(inode); |
| 9821 | last = block_group->key.objectid + block_group->key.offset; |
| 9822 | btrfs_put_block_group(block_group); |
| 9823 | } |
| 9824 | } |
| 9825 | |
| 9826 | /* |
| 9827 | * Must be called only after stopping all workers, since we could have block |
| 9828 | * group caching kthreads running, and therefore they could race with us if we |
| 9829 | * freed the block groups before stopping them. |
| 9830 | */ |
| 9831 | int btrfs_free_block_groups(struct btrfs_fs_info *info) |
| 9832 | { |
| 9833 | struct btrfs_block_group_cache *block_group; |
| 9834 | struct btrfs_space_info *space_info; |
| 9835 | struct btrfs_caching_control *caching_ctl; |
| 9836 | struct rb_node *n; |
| 9837 | |
| 9838 | down_write(&info->commit_root_sem); |
| 9839 | while (!list_empty(&info->caching_block_groups)) { |
| 9840 | caching_ctl = list_entry(info->caching_block_groups.next, |
| 9841 | struct btrfs_caching_control, list); |
| 9842 | list_del(&caching_ctl->list); |
| 9843 | put_caching_control(caching_ctl); |
| 9844 | } |
| 9845 | up_write(&info->commit_root_sem); |
| 9846 | |
| 9847 | spin_lock(&info->unused_bgs_lock); |
| 9848 | while (!list_empty(&info->unused_bgs)) { |
| 9849 | block_group = list_first_entry(&info->unused_bgs, |
| 9850 | struct btrfs_block_group_cache, |
| 9851 | bg_list); |
| 9852 | list_del_init(&block_group->bg_list); |
| 9853 | btrfs_put_block_group(block_group); |
| 9854 | } |
| 9855 | spin_unlock(&info->unused_bgs_lock); |
| 9856 | |
| 9857 | spin_lock(&info->block_group_cache_lock); |
| 9858 | while ((n = rb_last(&info->block_group_cache_tree)) != NULL) { |
| 9859 | block_group = rb_entry(n, struct btrfs_block_group_cache, |
| 9860 | cache_node); |
| 9861 | rb_erase(&block_group->cache_node, |
| 9862 | &info->block_group_cache_tree); |
| 9863 | RB_CLEAR_NODE(&block_group->cache_node); |
| 9864 | spin_unlock(&info->block_group_cache_lock); |
| 9865 | |
| 9866 | down_write(&block_group->space_info->groups_sem); |
| 9867 | list_del(&block_group->list); |
| 9868 | up_write(&block_group->space_info->groups_sem); |
| 9869 | |
| 9870 | /* |
| 9871 | * We haven't cached this block group, which means we could |
| 9872 | * possibly have excluded extents on this block group. |
| 9873 | */ |
| 9874 | if (block_group->cached == BTRFS_CACHE_NO || |
| 9875 | block_group->cached == BTRFS_CACHE_ERROR) |
| 9876 | free_excluded_extents(info, block_group); |
| 9877 | |
| 9878 | btrfs_remove_free_space_cache(block_group); |
| 9879 | ASSERT(block_group->cached != BTRFS_CACHE_STARTED); |
| 9880 | ASSERT(list_empty(&block_group->dirty_list)); |
| 9881 | ASSERT(list_empty(&block_group->io_list)); |
| 9882 | ASSERT(list_empty(&block_group->bg_list)); |
| 9883 | ASSERT(atomic_read(&block_group->count) == 1); |
| 9884 | btrfs_put_block_group(block_group); |
| 9885 | |
| 9886 | spin_lock(&info->block_group_cache_lock); |
| 9887 | } |
| 9888 | spin_unlock(&info->block_group_cache_lock); |
| 9889 | |
| 9890 | /* now that all the block groups are freed, go through and |
| 9891 | * free all the space_info structs. This is only called during |
| 9892 | * the final stages of unmount, and so we know nobody is |
| 9893 | * using them. We call synchronize_rcu() once before we start, |
| 9894 | * just to be on the safe side. |
| 9895 | */ |
| 9896 | synchronize_rcu(); |
| 9897 | |
| 9898 | release_global_block_rsv(info); |
| 9899 | |
| 9900 | while (!list_empty(&info->space_info)) { |
| 9901 | int i; |
| 9902 | |
| 9903 | space_info = list_entry(info->space_info.next, |
| 9904 | struct btrfs_space_info, |
| 9905 | list); |
| 9906 | |
| 9907 | /* |
| 9908 | * Do not hide this behind enospc_debug, this is actually |
| 9909 | * important and indicates a real bug if this happens. |
| 9910 | */ |
| 9911 | if (WARN_ON(space_info->bytes_pinned > 0 || |
| 9912 | space_info->bytes_reserved > 0 || |
| 9913 | space_info->bytes_may_use > 0)) |
| 9914 | dump_space_info(info, space_info, 0, 0); |
| 9915 | list_del(&space_info->list); |
| 9916 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
| 9917 | struct kobject *kobj; |
| 9918 | kobj = space_info->block_group_kobjs[i]; |
| 9919 | space_info->block_group_kobjs[i] = NULL; |
| 9920 | if (kobj) { |
| 9921 | kobject_del(kobj); |
| 9922 | kobject_put(kobj); |
| 9923 | } |
| 9924 | } |
| 9925 | kobject_del(&space_info->kobj); |
| 9926 | kobject_put(&space_info->kobj); |
| 9927 | } |
| 9928 | return 0; |
| 9929 | } |
| 9930 | |
| 9931 | /* link_block_group will queue up kobjects to add when we're reclaim-safe */ |
| 9932 | void btrfs_add_raid_kobjects(struct btrfs_fs_info *fs_info) |
| 9933 | { |
| 9934 | struct btrfs_space_info *space_info; |
| 9935 | struct raid_kobject *rkobj; |
| 9936 | LIST_HEAD(list); |
| 9937 | int index; |
| 9938 | int ret = 0; |
| 9939 | |
| 9940 | spin_lock(&fs_info->pending_raid_kobjs_lock); |
| 9941 | list_splice_init(&fs_info->pending_raid_kobjs, &list); |
| 9942 | spin_unlock(&fs_info->pending_raid_kobjs_lock); |
| 9943 | |
| 9944 | list_for_each_entry(rkobj, &list, list) { |
| 9945 | space_info = __find_space_info(fs_info, rkobj->flags); |
| 9946 | index = btrfs_bg_flags_to_raid_index(rkobj->flags); |
| 9947 | |
| 9948 | ret = kobject_add(&rkobj->kobj, &space_info->kobj, |
| 9949 | "%s", get_raid_name(index)); |
| 9950 | if (ret) { |
| 9951 | kobject_put(&rkobj->kobj); |
| 9952 | break; |
| 9953 | } |
| 9954 | } |
| 9955 | if (ret) |
| 9956 | btrfs_warn(fs_info, |
| 9957 | "failed to add kobject for block cache, ignoring"); |
| 9958 | } |
| 9959 | |
| 9960 | static void link_block_group(struct btrfs_block_group_cache *cache) |
| 9961 | { |
| 9962 | struct btrfs_space_info *space_info = cache->space_info; |
| 9963 | struct btrfs_fs_info *fs_info = cache->fs_info; |
| 9964 | int index = btrfs_bg_flags_to_raid_index(cache->flags); |
| 9965 | bool first = false; |
| 9966 | |
| 9967 | down_write(&space_info->groups_sem); |
| 9968 | if (list_empty(&space_info->block_groups[index])) |
| 9969 | first = true; |
| 9970 | list_add_tail(&cache->list, &space_info->block_groups[index]); |
| 9971 | up_write(&space_info->groups_sem); |
| 9972 | |
| 9973 | if (first) { |
| 9974 | struct raid_kobject *rkobj = kzalloc(sizeof(*rkobj), GFP_NOFS); |
| 9975 | if (!rkobj) { |
| 9976 | btrfs_warn(cache->fs_info, |
| 9977 | "couldn't alloc memory for raid level kobject"); |
| 9978 | return; |
| 9979 | } |
| 9980 | rkobj->flags = cache->flags; |
| 9981 | kobject_init(&rkobj->kobj, &btrfs_raid_ktype); |
| 9982 | |
| 9983 | spin_lock(&fs_info->pending_raid_kobjs_lock); |
| 9984 | list_add_tail(&rkobj->list, &fs_info->pending_raid_kobjs); |
| 9985 | spin_unlock(&fs_info->pending_raid_kobjs_lock); |
| 9986 | space_info->block_group_kobjs[index] = &rkobj->kobj; |
| 9987 | } |
| 9988 | } |
| 9989 | |
| 9990 | static struct btrfs_block_group_cache * |
| 9991 | btrfs_create_block_group_cache(struct btrfs_fs_info *fs_info, |
| 9992 | u64 start, u64 size) |
| 9993 | { |
| 9994 | struct btrfs_block_group_cache *cache; |
| 9995 | |
| 9996 | cache = kzalloc(sizeof(*cache), GFP_NOFS); |
| 9997 | if (!cache) |
| 9998 | return NULL; |
| 9999 | |
| 10000 | cache->free_space_ctl = kzalloc(sizeof(*cache->free_space_ctl), |
| 10001 | GFP_NOFS); |
| 10002 | if (!cache->free_space_ctl) { |
| 10003 | kfree(cache); |
| 10004 | return NULL; |
| 10005 | } |
| 10006 | |
| 10007 | cache->key.objectid = start; |
| 10008 | cache->key.offset = size; |
| 10009 | cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
| 10010 | |
| 10011 | cache->fs_info = fs_info; |
| 10012 | cache->full_stripe_len = btrfs_full_stripe_len(fs_info, start); |
| 10013 | set_free_space_tree_thresholds(cache); |
| 10014 | |
| 10015 | atomic_set(&cache->count, 1); |
| 10016 | spin_lock_init(&cache->lock); |
| 10017 | init_rwsem(&cache->data_rwsem); |
| 10018 | INIT_LIST_HEAD(&cache->list); |
| 10019 | INIT_LIST_HEAD(&cache->cluster_list); |
| 10020 | INIT_LIST_HEAD(&cache->bg_list); |
| 10021 | INIT_LIST_HEAD(&cache->ro_list); |
| 10022 | INIT_LIST_HEAD(&cache->dirty_list); |
| 10023 | INIT_LIST_HEAD(&cache->io_list); |
| 10024 | btrfs_init_free_space_ctl(cache); |
| 10025 | atomic_set(&cache->trimming, 0); |
| 10026 | mutex_init(&cache->free_space_lock); |
| 10027 | btrfs_init_full_stripe_locks_tree(&cache->full_stripe_locks_root); |
| 10028 | |
| 10029 | return cache; |
| 10030 | } |
| 10031 | |
| 10032 | int btrfs_read_block_groups(struct btrfs_fs_info *info) |
| 10033 | { |
| 10034 | struct btrfs_path *path; |
| 10035 | int ret; |
| 10036 | struct btrfs_block_group_cache *cache; |
| 10037 | struct btrfs_space_info *space_info; |
| 10038 | struct btrfs_key key; |
| 10039 | struct btrfs_key found_key; |
| 10040 | struct extent_buffer *leaf; |
| 10041 | int need_clear = 0; |
| 10042 | u64 cache_gen; |
| 10043 | u64 feature; |
| 10044 | int mixed; |
| 10045 | |
| 10046 | feature = btrfs_super_incompat_flags(info->super_copy); |
| 10047 | mixed = !!(feature & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS); |
| 10048 | |
| 10049 | key.objectid = 0; |
| 10050 | key.offset = 0; |
| 10051 | key.type = BTRFS_BLOCK_GROUP_ITEM_KEY; |
| 10052 | path = btrfs_alloc_path(); |
| 10053 | if (!path) |
| 10054 | return -ENOMEM; |
| 10055 | path->reada = READA_FORWARD; |
| 10056 | |
| 10057 | cache_gen = btrfs_super_cache_generation(info->super_copy); |
| 10058 | if (btrfs_test_opt(info, SPACE_CACHE) && |
| 10059 | btrfs_super_generation(info->super_copy) != cache_gen) |
| 10060 | need_clear = 1; |
| 10061 | if (btrfs_test_opt(info, CLEAR_CACHE)) |
| 10062 | need_clear = 1; |
| 10063 | |
| 10064 | while (1) { |
| 10065 | ret = find_first_block_group(info, path, &key); |
| 10066 | if (ret > 0) |
| 10067 | break; |
| 10068 | if (ret != 0) |
| 10069 | goto error; |
| 10070 | |
| 10071 | leaf = path->nodes[0]; |
| 10072 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); |
| 10073 | |
| 10074 | cache = btrfs_create_block_group_cache(info, found_key.objectid, |
| 10075 | found_key.offset); |
| 10076 | if (!cache) { |
| 10077 | ret = -ENOMEM; |
| 10078 | goto error; |
| 10079 | } |
| 10080 | |
| 10081 | if (need_clear) { |
| 10082 | /* |
| 10083 | * When we mount with old space cache, we need to |
| 10084 | * set BTRFS_DC_CLEAR and set dirty flag. |
| 10085 | * |
| 10086 | * a) Setting 'BTRFS_DC_CLEAR' makes sure that we |
| 10087 | * truncate the old free space cache inode and |
| 10088 | * setup a new one. |
| 10089 | * b) Setting 'dirty flag' makes sure that we flush |
| 10090 | * the new space cache info onto disk. |
| 10091 | */ |
| 10092 | if (btrfs_test_opt(info, SPACE_CACHE)) |
| 10093 | cache->disk_cache_state = BTRFS_DC_CLEAR; |
| 10094 | } |
| 10095 | |
| 10096 | read_extent_buffer(leaf, &cache->item, |
| 10097 | btrfs_item_ptr_offset(leaf, path->slots[0]), |
| 10098 | sizeof(cache->item)); |
| 10099 | cache->flags = btrfs_block_group_flags(&cache->item); |
| 10100 | if (!mixed && |
| 10101 | ((cache->flags & BTRFS_BLOCK_GROUP_METADATA) && |
| 10102 | (cache->flags & BTRFS_BLOCK_GROUP_DATA))) { |
| 10103 | btrfs_err(info, |
| 10104 | "bg %llu is a mixed block group but filesystem hasn't enabled mixed block groups", |
| 10105 | cache->key.objectid); |
| 10106 | ret = -EINVAL; |
| 10107 | goto error; |
| 10108 | } |
| 10109 | |
| 10110 | key.objectid = found_key.objectid + found_key.offset; |
| 10111 | btrfs_release_path(path); |
| 10112 | |
| 10113 | /* |
| 10114 | * We need to exclude the super stripes now so that the space |
| 10115 | * info has super bytes accounted for, otherwise we'll think |
| 10116 | * we have more space than we actually do. |
| 10117 | */ |
| 10118 | ret = exclude_super_stripes(info, cache); |
| 10119 | if (ret) { |
| 10120 | /* |
| 10121 | * We may have excluded something, so call this just in |
| 10122 | * case. |
| 10123 | */ |
| 10124 | free_excluded_extents(info, cache); |
| 10125 | btrfs_put_block_group(cache); |
| 10126 | goto error; |
| 10127 | } |
| 10128 | |
| 10129 | /* |
| 10130 | * check for two cases, either we are full, and therefore |
| 10131 | * don't need to bother with the caching work since we won't |
| 10132 | * find any space, or we are empty, and we can just add all |
| 10133 | * the space in and be done with it. This saves us _alot_ of |
| 10134 | * time, particularly in the full case. |
| 10135 | */ |
| 10136 | if (found_key.offset == btrfs_block_group_used(&cache->item)) { |
| 10137 | cache->last_byte_to_unpin = (u64)-1; |
| 10138 | cache->cached = BTRFS_CACHE_FINISHED; |
| 10139 | free_excluded_extents(info, cache); |
| 10140 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
| 10141 | cache->last_byte_to_unpin = (u64)-1; |
| 10142 | cache->cached = BTRFS_CACHE_FINISHED; |
| 10143 | add_new_free_space(cache, found_key.objectid, |
| 10144 | found_key.objectid + |
| 10145 | found_key.offset); |
| 10146 | free_excluded_extents(info, cache); |
| 10147 | } |
| 10148 | |
| 10149 | ret = btrfs_add_block_group_cache(info, cache); |
| 10150 | if (ret) { |
| 10151 | btrfs_remove_free_space_cache(cache); |
| 10152 | btrfs_put_block_group(cache); |
| 10153 | goto error; |
| 10154 | } |
| 10155 | |
| 10156 | trace_btrfs_add_block_group(info, cache, 0); |
| 10157 | update_space_info(info, cache->flags, found_key.offset, |
| 10158 | btrfs_block_group_used(&cache->item), |
| 10159 | cache->bytes_super, &space_info); |
| 10160 | |
| 10161 | cache->space_info = space_info; |
| 10162 | |
| 10163 | link_block_group(cache); |
| 10164 | |
| 10165 | set_avail_alloc_bits(info, cache->flags); |
| 10166 | if (btrfs_chunk_readonly(info, cache->key.objectid)) { |
| 10167 | inc_block_group_ro(cache, 1); |
| 10168 | } else if (btrfs_block_group_used(&cache->item) == 0) { |
| 10169 | spin_lock(&info->unused_bgs_lock); |
| 10170 | /* Should always be true but just in case. */ |
| 10171 | if (list_empty(&cache->bg_list)) { |
| 10172 | btrfs_get_block_group(cache); |
| 10173 | trace_btrfs_add_unused_block_group(cache); |
| 10174 | list_add_tail(&cache->bg_list, |
| 10175 | &info->unused_bgs); |
| 10176 | } |
| 10177 | spin_unlock(&info->unused_bgs_lock); |
| 10178 | } |
| 10179 | } |
| 10180 | |
| 10181 | list_for_each_entry_rcu(space_info, &info->space_info, list) { |
| 10182 | if (!(get_alloc_profile(info, space_info->flags) & |
| 10183 | (BTRFS_BLOCK_GROUP_RAID10 | |
| 10184 | BTRFS_BLOCK_GROUP_RAID1 | |
| 10185 | BTRFS_BLOCK_GROUP_RAID5 | |
| 10186 | BTRFS_BLOCK_GROUP_RAID6 | |
| 10187 | BTRFS_BLOCK_GROUP_DUP))) |
| 10188 | continue; |
| 10189 | /* |
| 10190 | * avoid allocating from un-mirrored block group if there are |
| 10191 | * mirrored block groups. |
| 10192 | */ |
| 10193 | list_for_each_entry(cache, |
| 10194 | &space_info->block_groups[BTRFS_RAID_RAID0], |
| 10195 | list) |
| 10196 | inc_block_group_ro(cache, 1); |
| 10197 | list_for_each_entry(cache, |
| 10198 | &space_info->block_groups[BTRFS_RAID_SINGLE], |
| 10199 | list) |
| 10200 | inc_block_group_ro(cache, 1); |
| 10201 | } |
| 10202 | |
| 10203 | btrfs_add_raid_kobjects(info); |
| 10204 | init_global_block_rsv(info); |
| 10205 | ret = 0; |
| 10206 | error: |
| 10207 | btrfs_free_path(path); |
| 10208 | return ret; |
| 10209 | } |
| 10210 | |
| 10211 | void btrfs_create_pending_block_groups(struct btrfs_trans_handle *trans) |
| 10212 | { |
| 10213 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 10214 | struct btrfs_block_group_cache *block_group, *tmp; |
| 10215 | struct btrfs_root *extent_root = fs_info->extent_root; |
| 10216 | struct btrfs_block_group_item item; |
| 10217 | struct btrfs_key key; |
| 10218 | int ret = 0; |
| 10219 | bool can_flush_pending_bgs = trans->can_flush_pending_bgs; |
| 10220 | |
| 10221 | trans->can_flush_pending_bgs = false; |
| 10222 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { |
| 10223 | if (ret) |
| 10224 | goto next; |
| 10225 | |
| 10226 | spin_lock(&block_group->lock); |
| 10227 | memcpy(&item, &block_group->item, sizeof(item)); |
| 10228 | memcpy(&key, &block_group->key, sizeof(key)); |
| 10229 | spin_unlock(&block_group->lock); |
| 10230 | |
| 10231 | ret = btrfs_insert_item(trans, extent_root, &key, &item, |
| 10232 | sizeof(item)); |
| 10233 | if (ret) |
| 10234 | btrfs_abort_transaction(trans, ret); |
| 10235 | ret = btrfs_finish_chunk_alloc(trans, fs_info, key.objectid, |
| 10236 | key.offset); |
| 10237 | if (ret) |
| 10238 | btrfs_abort_transaction(trans, ret); |
| 10239 | add_block_group_free_space(trans, block_group); |
| 10240 | /* already aborted the transaction if it failed. */ |
| 10241 | next: |
| 10242 | list_del_init(&block_group->bg_list); |
| 10243 | } |
| 10244 | trans->can_flush_pending_bgs = can_flush_pending_bgs; |
| 10245 | } |
| 10246 | |
| 10247 | int btrfs_make_block_group(struct btrfs_trans_handle *trans, |
| 10248 | struct btrfs_fs_info *fs_info, u64 bytes_used, |
| 10249 | u64 type, u64 chunk_offset, u64 size) |
| 10250 | { |
| 10251 | struct btrfs_block_group_cache *cache; |
| 10252 | int ret; |
| 10253 | |
| 10254 | btrfs_set_log_full_commit(fs_info, trans); |
| 10255 | |
| 10256 | cache = btrfs_create_block_group_cache(fs_info, chunk_offset, size); |
| 10257 | if (!cache) |
| 10258 | return -ENOMEM; |
| 10259 | |
| 10260 | btrfs_set_block_group_used(&cache->item, bytes_used); |
| 10261 | btrfs_set_block_group_chunk_objectid(&cache->item, |
| 10262 | BTRFS_FIRST_CHUNK_TREE_OBJECTID); |
| 10263 | btrfs_set_block_group_flags(&cache->item, type); |
| 10264 | |
| 10265 | cache->flags = type; |
| 10266 | cache->last_byte_to_unpin = (u64)-1; |
| 10267 | cache->cached = BTRFS_CACHE_FINISHED; |
| 10268 | cache->needs_free_space = 1; |
| 10269 | ret = exclude_super_stripes(fs_info, cache); |
| 10270 | if (ret) { |
| 10271 | /* |
| 10272 | * We may have excluded something, so call this just in |
| 10273 | * case. |
| 10274 | */ |
| 10275 | free_excluded_extents(fs_info, cache); |
| 10276 | btrfs_put_block_group(cache); |
| 10277 | return ret; |
| 10278 | } |
| 10279 | |
| 10280 | add_new_free_space(cache, chunk_offset, chunk_offset + size); |
| 10281 | |
| 10282 | free_excluded_extents(fs_info, cache); |
| 10283 | |
| 10284 | #ifdef CONFIG_BTRFS_DEBUG |
| 10285 | if (btrfs_should_fragment_free_space(cache)) { |
| 10286 | u64 new_bytes_used = size - bytes_used; |
| 10287 | |
| 10288 | bytes_used += new_bytes_used >> 1; |
| 10289 | fragment_free_space(cache); |
| 10290 | } |
| 10291 | #endif |
| 10292 | /* |
| 10293 | * Ensure the corresponding space_info object is created and |
| 10294 | * assigned to our block group. We want our bg to be added to the rbtree |
| 10295 | * with its ->space_info set. |
| 10296 | */ |
| 10297 | cache->space_info = __find_space_info(fs_info, cache->flags); |
| 10298 | ASSERT(cache->space_info); |
| 10299 | |
| 10300 | ret = btrfs_add_block_group_cache(fs_info, cache); |
| 10301 | if (ret) { |
| 10302 | btrfs_remove_free_space_cache(cache); |
| 10303 | btrfs_put_block_group(cache); |
| 10304 | return ret; |
| 10305 | } |
| 10306 | |
| 10307 | /* |
| 10308 | * Now that our block group has its ->space_info set and is inserted in |
| 10309 | * the rbtree, update the space info's counters. |
| 10310 | */ |
| 10311 | trace_btrfs_add_block_group(fs_info, cache, 1); |
| 10312 | update_space_info(fs_info, cache->flags, size, bytes_used, |
| 10313 | cache->bytes_super, &cache->space_info); |
| 10314 | update_global_block_rsv(fs_info); |
| 10315 | |
| 10316 | link_block_group(cache); |
| 10317 | |
| 10318 | list_add_tail(&cache->bg_list, &trans->new_bgs); |
| 10319 | |
| 10320 | set_avail_alloc_bits(fs_info, type); |
| 10321 | return 0; |
| 10322 | } |
| 10323 | |
| 10324 | static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags) |
| 10325 | { |
| 10326 | u64 extra_flags = chunk_to_extended(flags) & |
| 10327 | BTRFS_EXTENDED_PROFILE_MASK; |
| 10328 | |
| 10329 | write_seqlock(&fs_info->profiles_lock); |
| 10330 | if (flags & BTRFS_BLOCK_GROUP_DATA) |
| 10331 | fs_info->avail_data_alloc_bits &= ~extra_flags; |
| 10332 | if (flags & BTRFS_BLOCK_GROUP_METADATA) |
| 10333 | fs_info->avail_metadata_alloc_bits &= ~extra_flags; |
| 10334 | if (flags & BTRFS_BLOCK_GROUP_SYSTEM) |
| 10335 | fs_info->avail_system_alloc_bits &= ~extra_flags; |
| 10336 | write_sequnlock(&fs_info->profiles_lock); |
| 10337 | } |
| 10338 | |
| 10339 | int btrfs_remove_block_group(struct btrfs_trans_handle *trans, |
| 10340 | struct btrfs_fs_info *fs_info, u64 group_start, |
| 10341 | struct extent_map *em) |
| 10342 | { |
| 10343 | struct btrfs_root *root = fs_info->extent_root; |
| 10344 | struct btrfs_path *path; |
| 10345 | struct btrfs_block_group_cache *block_group; |
| 10346 | struct btrfs_free_cluster *cluster; |
| 10347 | struct btrfs_root *tree_root = fs_info->tree_root; |
| 10348 | struct btrfs_key key; |
| 10349 | struct inode *inode; |
| 10350 | struct kobject *kobj = NULL; |
| 10351 | int ret; |
| 10352 | int index; |
| 10353 | int factor; |
| 10354 | struct btrfs_caching_control *caching_ctl = NULL; |
| 10355 | bool remove_em; |
| 10356 | |
| 10357 | block_group = btrfs_lookup_block_group(fs_info, group_start); |
| 10358 | BUG_ON(!block_group); |
| 10359 | BUG_ON(!block_group->ro); |
| 10360 | |
| 10361 | trace_btrfs_remove_block_group(block_group); |
| 10362 | /* |
| 10363 | * Free the reserved super bytes from this block group before |
| 10364 | * remove it. |
| 10365 | */ |
| 10366 | free_excluded_extents(fs_info, block_group); |
| 10367 | btrfs_free_ref_tree_range(fs_info, block_group->key.objectid, |
| 10368 | block_group->key.offset); |
| 10369 | |
| 10370 | memcpy(&key, &block_group->key, sizeof(key)); |
| 10371 | index = btrfs_bg_flags_to_raid_index(block_group->flags); |
| 10372 | if (block_group->flags & (BTRFS_BLOCK_GROUP_DUP | |
| 10373 | BTRFS_BLOCK_GROUP_RAID1 | |
| 10374 | BTRFS_BLOCK_GROUP_RAID10)) |
| 10375 | factor = 2; |
| 10376 | else |
| 10377 | factor = 1; |
| 10378 | |
| 10379 | /* make sure this block group isn't part of an allocation cluster */ |
| 10380 | cluster = &fs_info->data_alloc_cluster; |
| 10381 | spin_lock(&cluster->refill_lock); |
| 10382 | btrfs_return_cluster_to_free_space(block_group, cluster); |
| 10383 | spin_unlock(&cluster->refill_lock); |
| 10384 | |
| 10385 | /* |
| 10386 | * make sure this block group isn't part of a metadata |
| 10387 | * allocation cluster |
| 10388 | */ |
| 10389 | cluster = &fs_info->meta_alloc_cluster; |
| 10390 | spin_lock(&cluster->refill_lock); |
| 10391 | btrfs_return_cluster_to_free_space(block_group, cluster); |
| 10392 | spin_unlock(&cluster->refill_lock); |
| 10393 | |
| 10394 | path = btrfs_alloc_path(); |
| 10395 | if (!path) { |
| 10396 | ret = -ENOMEM; |
| 10397 | goto out; |
| 10398 | } |
| 10399 | |
| 10400 | /* |
| 10401 | * get the inode first so any iput calls done for the io_list |
| 10402 | * aren't the final iput (no unlinks allowed now) |
| 10403 | */ |
| 10404 | inode = lookup_free_space_inode(fs_info, block_group, path); |
| 10405 | |
| 10406 | mutex_lock(&trans->transaction->cache_write_mutex); |
| 10407 | /* |
| 10408 | * make sure our free spache cache IO is done before remove the |
| 10409 | * free space inode |
| 10410 | */ |
| 10411 | spin_lock(&trans->transaction->dirty_bgs_lock); |
| 10412 | if (!list_empty(&block_group->io_list)) { |
| 10413 | list_del_init(&block_group->io_list); |
| 10414 | |
| 10415 | WARN_ON(!IS_ERR(inode) && inode != block_group->io_ctl.inode); |
| 10416 | |
| 10417 | spin_unlock(&trans->transaction->dirty_bgs_lock); |
| 10418 | btrfs_wait_cache_io(trans, block_group, path); |
| 10419 | btrfs_put_block_group(block_group); |
| 10420 | spin_lock(&trans->transaction->dirty_bgs_lock); |
| 10421 | } |
| 10422 | |
| 10423 | if (!list_empty(&block_group->dirty_list)) { |
| 10424 | list_del_init(&block_group->dirty_list); |
| 10425 | btrfs_put_block_group(block_group); |
| 10426 | } |
| 10427 | spin_unlock(&trans->transaction->dirty_bgs_lock); |
| 10428 | mutex_unlock(&trans->transaction->cache_write_mutex); |
| 10429 | |
| 10430 | if (!IS_ERR(inode)) { |
| 10431 | ret = btrfs_orphan_add(trans, BTRFS_I(inode)); |
| 10432 | if (ret) { |
| 10433 | btrfs_add_delayed_iput(inode); |
| 10434 | goto out; |
| 10435 | } |
| 10436 | clear_nlink(inode); |
| 10437 | /* One for the block groups ref */ |
| 10438 | spin_lock(&block_group->lock); |
| 10439 | if (block_group->iref) { |
| 10440 | block_group->iref = 0; |
| 10441 | block_group->inode = NULL; |
| 10442 | spin_unlock(&block_group->lock); |
| 10443 | iput(inode); |
| 10444 | } else { |
| 10445 | spin_unlock(&block_group->lock); |
| 10446 | } |
| 10447 | /* One for our lookup ref */ |
| 10448 | btrfs_add_delayed_iput(inode); |
| 10449 | } |
| 10450 | |
| 10451 | key.objectid = BTRFS_FREE_SPACE_OBJECTID; |
| 10452 | key.offset = block_group->key.objectid; |
| 10453 | key.type = 0; |
| 10454 | |
| 10455 | ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1); |
| 10456 | if (ret < 0) |
| 10457 | goto out; |
| 10458 | if (ret > 0) |
| 10459 | btrfs_release_path(path); |
| 10460 | if (ret == 0) { |
| 10461 | ret = btrfs_del_item(trans, tree_root, path); |
| 10462 | if (ret) |
| 10463 | goto out; |
| 10464 | btrfs_release_path(path); |
| 10465 | } |
| 10466 | |
| 10467 | spin_lock(&fs_info->block_group_cache_lock); |
| 10468 | rb_erase(&block_group->cache_node, |
| 10469 | &fs_info->block_group_cache_tree); |
| 10470 | RB_CLEAR_NODE(&block_group->cache_node); |
| 10471 | |
| 10472 | if (fs_info->first_logical_byte == block_group->key.objectid) |
| 10473 | fs_info->first_logical_byte = (u64)-1; |
| 10474 | spin_unlock(&fs_info->block_group_cache_lock); |
| 10475 | |
| 10476 | down_write(&block_group->space_info->groups_sem); |
| 10477 | /* |
| 10478 | * we must use list_del_init so people can check to see if they |
| 10479 | * are still on the list after taking the semaphore |
| 10480 | */ |
| 10481 | list_del_init(&block_group->list); |
| 10482 | if (list_empty(&block_group->space_info->block_groups[index])) { |
| 10483 | kobj = block_group->space_info->block_group_kobjs[index]; |
| 10484 | block_group->space_info->block_group_kobjs[index] = NULL; |
| 10485 | clear_avail_alloc_bits(fs_info, block_group->flags); |
| 10486 | } |
| 10487 | up_write(&block_group->space_info->groups_sem); |
| 10488 | if (kobj) { |
| 10489 | kobject_del(kobj); |
| 10490 | kobject_put(kobj); |
| 10491 | } |
| 10492 | |
| 10493 | if (block_group->has_caching_ctl) |
| 10494 | caching_ctl = get_caching_control(block_group); |
| 10495 | if (block_group->cached == BTRFS_CACHE_STARTED) |
| 10496 | wait_block_group_cache_done(block_group); |
| 10497 | if (block_group->has_caching_ctl) { |
| 10498 | down_write(&fs_info->commit_root_sem); |
| 10499 | if (!caching_ctl) { |
| 10500 | struct btrfs_caching_control *ctl; |
| 10501 | |
| 10502 | list_for_each_entry(ctl, |
| 10503 | &fs_info->caching_block_groups, list) |
| 10504 | if (ctl->block_group == block_group) { |
| 10505 | caching_ctl = ctl; |
| 10506 | refcount_inc(&caching_ctl->count); |
| 10507 | break; |
| 10508 | } |
| 10509 | } |
| 10510 | if (caching_ctl) |
| 10511 | list_del_init(&caching_ctl->list); |
| 10512 | up_write(&fs_info->commit_root_sem); |
| 10513 | if (caching_ctl) { |
| 10514 | /* Once for the caching bgs list and once for us. */ |
| 10515 | put_caching_control(caching_ctl); |
| 10516 | put_caching_control(caching_ctl); |
| 10517 | } |
| 10518 | } |
| 10519 | |
| 10520 | spin_lock(&trans->transaction->dirty_bgs_lock); |
| 10521 | if (!list_empty(&block_group->dirty_list)) { |
| 10522 | WARN_ON(1); |
| 10523 | } |
| 10524 | if (!list_empty(&block_group->io_list)) { |
| 10525 | WARN_ON(1); |
| 10526 | } |
| 10527 | spin_unlock(&trans->transaction->dirty_bgs_lock); |
| 10528 | btrfs_remove_free_space_cache(block_group); |
| 10529 | |
| 10530 | spin_lock(&block_group->space_info->lock); |
| 10531 | list_del_init(&block_group->ro_list); |
| 10532 | |
| 10533 | if (btrfs_test_opt(fs_info, ENOSPC_DEBUG)) { |
| 10534 | WARN_ON(block_group->space_info->total_bytes |
| 10535 | < block_group->key.offset); |
| 10536 | WARN_ON(block_group->space_info->bytes_readonly |
| 10537 | < block_group->key.offset); |
| 10538 | WARN_ON(block_group->space_info->disk_total |
| 10539 | < block_group->key.offset * factor); |
| 10540 | } |
| 10541 | block_group->space_info->total_bytes -= block_group->key.offset; |
| 10542 | block_group->space_info->bytes_readonly -= block_group->key.offset; |
| 10543 | block_group->space_info->disk_total -= block_group->key.offset * factor; |
| 10544 | |
| 10545 | spin_unlock(&block_group->space_info->lock); |
| 10546 | |
| 10547 | memcpy(&key, &block_group->key, sizeof(key)); |
| 10548 | |
| 10549 | mutex_lock(&fs_info->chunk_mutex); |
| 10550 | if (!list_empty(&em->list)) { |
| 10551 | /* We're in the transaction->pending_chunks list. */ |
| 10552 | free_extent_map(em); |
| 10553 | } |
| 10554 | spin_lock(&block_group->lock); |
| 10555 | block_group->removed = 1; |
| 10556 | /* |
| 10557 | * At this point trimming can't start on this block group, because we |
| 10558 | * removed the block group from the tree fs_info->block_group_cache_tree |
| 10559 | * so no one can't find it anymore and even if someone already got this |
| 10560 | * block group before we removed it from the rbtree, they have already |
| 10561 | * incremented block_group->trimming - if they didn't, they won't find |
| 10562 | * any free space entries because we already removed them all when we |
| 10563 | * called btrfs_remove_free_space_cache(). |
| 10564 | * |
| 10565 | * And we must not remove the extent map from the fs_info->mapping_tree |
| 10566 | * to prevent the same logical address range and physical device space |
| 10567 | * ranges from being reused for a new block group. This is because our |
| 10568 | * fs trim operation (btrfs_trim_fs() / btrfs_ioctl_fitrim()) is |
| 10569 | * completely transactionless, so while it is trimming a range the |
| 10570 | * currently running transaction might finish and a new one start, |
| 10571 | * allowing for new block groups to be created that can reuse the same |
| 10572 | * physical device locations unless we take this special care. |
| 10573 | * |
| 10574 | * There may also be an implicit trim operation if the file system |
| 10575 | * is mounted with -odiscard. The same protections must remain |
| 10576 | * in place until the extents have been discarded completely when |
| 10577 | * the transaction commit has completed. |
| 10578 | */ |
| 10579 | remove_em = (atomic_read(&block_group->trimming) == 0); |
| 10580 | /* |
| 10581 | * Make sure a trimmer task always sees the em in the pinned_chunks list |
| 10582 | * if it sees block_group->removed == 1 (needs to lock block_group->lock |
| 10583 | * before checking block_group->removed). |
| 10584 | */ |
| 10585 | if (!remove_em) { |
| 10586 | /* |
| 10587 | * Our em might be in trans->transaction->pending_chunks which |
| 10588 | * is protected by fs_info->chunk_mutex ([lock|unlock]_chunks), |
| 10589 | * and so is the fs_info->pinned_chunks list. |
| 10590 | * |
| 10591 | * So at this point we must be holding the chunk_mutex to avoid |
| 10592 | * any races with chunk allocation (more specifically at |
| 10593 | * volumes.c:contains_pending_extent()), to ensure it always |
| 10594 | * sees the em, either in the pending_chunks list or in the |
| 10595 | * pinned_chunks list. |
| 10596 | */ |
| 10597 | list_move_tail(&em->list, &fs_info->pinned_chunks); |
| 10598 | } |
| 10599 | spin_unlock(&block_group->lock); |
| 10600 | |
| 10601 | if (remove_em) { |
| 10602 | struct extent_map_tree *em_tree; |
| 10603 | |
| 10604 | em_tree = &fs_info->mapping_tree.map_tree; |
| 10605 | write_lock(&em_tree->lock); |
| 10606 | /* |
| 10607 | * The em might be in the pending_chunks list, so make sure the |
| 10608 | * chunk mutex is locked, since remove_extent_mapping() will |
| 10609 | * delete us from that list. |
| 10610 | */ |
| 10611 | remove_extent_mapping(em_tree, em); |
| 10612 | write_unlock(&em_tree->lock); |
| 10613 | /* once for the tree */ |
| 10614 | free_extent_map(em); |
| 10615 | } |
| 10616 | |
| 10617 | mutex_unlock(&fs_info->chunk_mutex); |
| 10618 | |
| 10619 | ret = remove_block_group_free_space(trans, block_group); |
| 10620 | if (ret) |
| 10621 | goto out; |
| 10622 | |
| 10623 | btrfs_put_block_group(block_group); |
| 10624 | btrfs_put_block_group(block_group); |
| 10625 | |
| 10626 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
| 10627 | if (ret > 0) |
| 10628 | ret = -EIO; |
| 10629 | if (ret < 0) |
| 10630 | goto out; |
| 10631 | |
| 10632 | ret = btrfs_del_item(trans, root, path); |
| 10633 | out: |
| 10634 | btrfs_free_path(path); |
| 10635 | return ret; |
| 10636 | } |
| 10637 | |
| 10638 | struct btrfs_trans_handle * |
| 10639 | btrfs_start_trans_remove_block_group(struct btrfs_fs_info *fs_info, |
| 10640 | const u64 chunk_offset) |
| 10641 | { |
| 10642 | struct extent_map_tree *em_tree = &fs_info->mapping_tree.map_tree; |
| 10643 | struct extent_map *em; |
| 10644 | struct map_lookup *map; |
| 10645 | unsigned int num_items; |
| 10646 | |
| 10647 | read_lock(&em_tree->lock); |
| 10648 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
| 10649 | read_unlock(&em_tree->lock); |
| 10650 | ASSERT(em && em->start == chunk_offset); |
| 10651 | |
| 10652 | /* |
| 10653 | * We need to reserve 3 + N units from the metadata space info in order |
| 10654 | * to remove a block group (done at btrfs_remove_chunk() and at |
| 10655 | * btrfs_remove_block_group()), which are used for: |
| 10656 | * |
| 10657 | * 1 unit for adding the free space inode's orphan (located in the tree |
| 10658 | * of tree roots). |
| 10659 | * 1 unit for deleting the block group item (located in the extent |
| 10660 | * tree). |
| 10661 | * 1 unit for deleting the free space item (located in tree of tree |
| 10662 | * roots). |
| 10663 | * N units for deleting N device extent items corresponding to each |
| 10664 | * stripe (located in the device tree). |
| 10665 | * |
| 10666 | * In order to remove a block group we also need to reserve units in the |
| 10667 | * system space info in order to update the chunk tree (update one or |
| 10668 | * more device items and remove one chunk item), but this is done at |
| 10669 | * btrfs_remove_chunk() through a call to check_system_chunk(). |
| 10670 | */ |
| 10671 | map = em->map_lookup; |
| 10672 | num_items = 3 + map->num_stripes; |
| 10673 | free_extent_map(em); |
| 10674 | |
| 10675 | return btrfs_start_transaction_fallback_global_rsv(fs_info->extent_root, |
| 10676 | num_items, 1); |
| 10677 | } |
| 10678 | |
| 10679 | /* |
| 10680 | * Process the unused_bgs list and remove any that don't have any allocated |
| 10681 | * space inside of them. |
| 10682 | */ |
| 10683 | void btrfs_delete_unused_bgs(struct btrfs_fs_info *fs_info) |
| 10684 | { |
| 10685 | struct btrfs_block_group_cache *block_group; |
| 10686 | struct btrfs_space_info *space_info; |
| 10687 | struct btrfs_trans_handle *trans; |
| 10688 | int ret = 0; |
| 10689 | |
| 10690 | if (!test_bit(BTRFS_FS_OPEN, &fs_info->flags)) |
| 10691 | return; |
| 10692 | |
| 10693 | spin_lock(&fs_info->unused_bgs_lock); |
| 10694 | while (!list_empty(&fs_info->unused_bgs)) { |
| 10695 | u64 start, end; |
| 10696 | int trimming; |
| 10697 | |
| 10698 | block_group = list_first_entry(&fs_info->unused_bgs, |
| 10699 | struct btrfs_block_group_cache, |
| 10700 | bg_list); |
| 10701 | list_del_init(&block_group->bg_list); |
| 10702 | |
| 10703 | space_info = block_group->space_info; |
| 10704 | |
| 10705 | if (ret || btrfs_mixed_space_info(space_info)) { |
| 10706 | btrfs_put_block_group(block_group); |
| 10707 | continue; |
| 10708 | } |
| 10709 | spin_unlock(&fs_info->unused_bgs_lock); |
| 10710 | |
| 10711 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
| 10712 | |
| 10713 | /* Don't want to race with allocators so take the groups_sem */ |
| 10714 | down_write(&space_info->groups_sem); |
| 10715 | spin_lock(&block_group->lock); |
| 10716 | if (block_group->reserved || |
| 10717 | btrfs_block_group_used(&block_group->item) || |
| 10718 | block_group->ro || |
| 10719 | list_is_singular(&block_group->list)) { |
| 10720 | /* |
| 10721 | * We want to bail if we made new allocations or have |
| 10722 | * outstanding allocations in this block group. We do |
| 10723 | * the ro check in case balance is currently acting on |
| 10724 | * this block group. |
| 10725 | */ |
| 10726 | trace_btrfs_skip_unused_block_group(block_group); |
| 10727 | spin_unlock(&block_group->lock); |
| 10728 | up_write(&space_info->groups_sem); |
| 10729 | goto next; |
| 10730 | } |
| 10731 | spin_unlock(&block_group->lock); |
| 10732 | |
| 10733 | /* We don't want to force the issue, only flip if it's ok. */ |
| 10734 | ret = inc_block_group_ro(block_group, 0); |
| 10735 | up_write(&space_info->groups_sem); |
| 10736 | if (ret < 0) { |
| 10737 | ret = 0; |
| 10738 | goto next; |
| 10739 | } |
| 10740 | |
| 10741 | /* |
| 10742 | * Want to do this before we do anything else so we can recover |
| 10743 | * properly if we fail to join the transaction. |
| 10744 | */ |
| 10745 | trans = btrfs_start_trans_remove_block_group(fs_info, |
| 10746 | block_group->key.objectid); |
| 10747 | if (IS_ERR(trans)) { |
| 10748 | btrfs_dec_block_group_ro(block_group); |
| 10749 | ret = PTR_ERR(trans); |
| 10750 | goto next; |
| 10751 | } |
| 10752 | |
| 10753 | /* |
| 10754 | * We could have pending pinned extents for this block group, |
| 10755 | * just delete them, we don't care about them anymore. |
| 10756 | */ |
| 10757 | start = block_group->key.objectid; |
| 10758 | end = start + block_group->key.offset - 1; |
| 10759 | /* |
| 10760 | * Hold the unused_bg_unpin_mutex lock to avoid racing with |
| 10761 | * btrfs_finish_extent_commit(). If we are at transaction N, |
| 10762 | * another task might be running finish_extent_commit() for the |
| 10763 | * previous transaction N - 1, and have seen a range belonging |
| 10764 | * to the block group in freed_extents[] before we were able to |
| 10765 | * clear the whole block group range from freed_extents[]. This |
| 10766 | * means that task can lookup for the block group after we |
| 10767 | * unpinned it from freed_extents[] and removed it, leading to |
| 10768 | * a BUG_ON() at btrfs_unpin_extent_range(). |
| 10769 | */ |
| 10770 | mutex_lock(&fs_info->unused_bg_unpin_mutex); |
| 10771 | ret = clear_extent_bits(&fs_info->freed_extents[0], start, end, |
| 10772 | EXTENT_DIRTY); |
| 10773 | if (ret) { |
| 10774 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
| 10775 | btrfs_dec_block_group_ro(block_group); |
| 10776 | goto end_trans; |
| 10777 | } |
| 10778 | ret = clear_extent_bits(&fs_info->freed_extents[1], start, end, |
| 10779 | EXTENT_DIRTY); |
| 10780 | if (ret) { |
| 10781 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
| 10782 | btrfs_dec_block_group_ro(block_group); |
| 10783 | goto end_trans; |
| 10784 | } |
| 10785 | mutex_unlock(&fs_info->unused_bg_unpin_mutex); |
| 10786 | |
| 10787 | /* Reset pinned so btrfs_put_block_group doesn't complain */ |
| 10788 | spin_lock(&space_info->lock); |
| 10789 | spin_lock(&block_group->lock); |
| 10790 | |
| 10791 | space_info->bytes_pinned -= block_group->pinned; |
| 10792 | space_info->bytes_readonly += block_group->pinned; |
| 10793 | percpu_counter_add(&space_info->total_bytes_pinned, |
| 10794 | -block_group->pinned); |
| 10795 | block_group->pinned = 0; |
| 10796 | |
| 10797 | spin_unlock(&block_group->lock); |
| 10798 | spin_unlock(&space_info->lock); |
| 10799 | |
| 10800 | /* DISCARD can flip during remount */ |
| 10801 | trimming = btrfs_test_opt(fs_info, DISCARD); |
| 10802 | |
| 10803 | /* Implicit trim during transaction commit. */ |
| 10804 | if (trimming) |
| 10805 | btrfs_get_block_group_trimming(block_group); |
| 10806 | |
| 10807 | /* |
| 10808 | * Btrfs_remove_chunk will abort the transaction if things go |
| 10809 | * horribly wrong. |
| 10810 | */ |
| 10811 | ret = btrfs_remove_chunk(trans, fs_info, |
| 10812 | block_group->key.objectid); |
| 10813 | |
| 10814 | if (ret) { |
| 10815 | if (trimming) |
| 10816 | btrfs_put_block_group_trimming(block_group); |
| 10817 | goto end_trans; |
| 10818 | } |
| 10819 | |
| 10820 | /* |
| 10821 | * If we're not mounted with -odiscard, we can just forget |
| 10822 | * about this block group. Otherwise we'll need to wait |
| 10823 | * until transaction commit to do the actual discard. |
| 10824 | */ |
| 10825 | if (trimming) { |
| 10826 | spin_lock(&fs_info->unused_bgs_lock); |
| 10827 | /* |
| 10828 | * A concurrent scrub might have added us to the list |
| 10829 | * fs_info->unused_bgs, so use a list_move operation |
| 10830 | * to add the block group to the deleted_bgs list. |
| 10831 | */ |
| 10832 | list_move(&block_group->bg_list, |
| 10833 | &trans->transaction->deleted_bgs); |
| 10834 | spin_unlock(&fs_info->unused_bgs_lock); |
| 10835 | btrfs_get_block_group(block_group); |
| 10836 | } |
| 10837 | end_trans: |
| 10838 | btrfs_end_transaction(trans); |
| 10839 | next: |
| 10840 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
| 10841 | btrfs_put_block_group(block_group); |
| 10842 | spin_lock(&fs_info->unused_bgs_lock); |
| 10843 | } |
| 10844 | spin_unlock(&fs_info->unused_bgs_lock); |
| 10845 | } |
| 10846 | |
| 10847 | int btrfs_init_space_info(struct btrfs_fs_info *fs_info) |
| 10848 | { |
| 10849 | struct btrfs_super_block *disk_super; |
| 10850 | u64 features; |
| 10851 | u64 flags; |
| 10852 | int mixed = 0; |
| 10853 | int ret; |
| 10854 | |
| 10855 | disk_super = fs_info->super_copy; |
| 10856 | if (!btrfs_super_root(disk_super)) |
| 10857 | return -EINVAL; |
| 10858 | |
| 10859 | features = btrfs_super_incompat_flags(disk_super); |
| 10860 | if (features & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) |
| 10861 | mixed = 1; |
| 10862 | |
| 10863 | flags = BTRFS_BLOCK_GROUP_SYSTEM; |
| 10864 | ret = create_space_info(fs_info, flags); |
| 10865 | if (ret) |
| 10866 | goto out; |
| 10867 | |
| 10868 | if (mixed) { |
| 10869 | flags = BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_DATA; |
| 10870 | ret = create_space_info(fs_info, flags); |
| 10871 | } else { |
| 10872 | flags = BTRFS_BLOCK_GROUP_METADATA; |
| 10873 | ret = create_space_info(fs_info, flags); |
| 10874 | if (ret) |
| 10875 | goto out; |
| 10876 | |
| 10877 | flags = BTRFS_BLOCK_GROUP_DATA; |
| 10878 | ret = create_space_info(fs_info, flags); |
| 10879 | } |
| 10880 | out: |
| 10881 | return ret; |
| 10882 | } |
| 10883 | |
| 10884 | int btrfs_error_unpin_extent_range(struct btrfs_fs_info *fs_info, |
| 10885 | u64 start, u64 end) |
| 10886 | { |
| 10887 | return unpin_extent_range(fs_info, start, end, false); |
| 10888 | } |
| 10889 | |
| 10890 | /* |
| 10891 | * It used to be that old block groups would be left around forever. |
| 10892 | * Iterating over them would be enough to trim unused space. Since we |
| 10893 | * now automatically remove them, we also need to iterate over unallocated |
| 10894 | * space. |
| 10895 | * |
| 10896 | * We don't want a transaction for this since the discard may take a |
| 10897 | * substantial amount of time. We don't require that a transaction be |
| 10898 | * running, but we do need to take a running transaction into account |
| 10899 | * to ensure that we're not discarding chunks that were released in |
| 10900 | * the current transaction. |
| 10901 | * |
| 10902 | * Holding the chunks lock will prevent other threads from allocating |
| 10903 | * or releasing chunks, but it won't prevent a running transaction |
| 10904 | * from committing and releasing the memory that the pending chunks |
| 10905 | * list head uses. For that, we need to take a reference to the |
| 10906 | * transaction. |
| 10907 | */ |
| 10908 | static int btrfs_trim_free_extents(struct btrfs_device *device, |
| 10909 | u64 minlen, u64 *trimmed) |
| 10910 | { |
| 10911 | u64 start = 0, len = 0; |
| 10912 | int ret; |
| 10913 | |
| 10914 | *trimmed = 0; |
| 10915 | |
| 10916 | /* Not writeable = nothing to do. */ |
| 10917 | if (!test_bit(BTRFS_DEV_STATE_WRITEABLE, &device->dev_state)) |
| 10918 | return 0; |
| 10919 | |
| 10920 | /* No free space = nothing to do. */ |
| 10921 | if (device->total_bytes <= device->bytes_used) |
| 10922 | return 0; |
| 10923 | |
| 10924 | ret = 0; |
| 10925 | |
| 10926 | while (1) { |
| 10927 | struct btrfs_fs_info *fs_info = device->fs_info; |
| 10928 | struct btrfs_transaction *trans; |
| 10929 | u64 bytes; |
| 10930 | |
| 10931 | ret = mutex_lock_interruptible(&fs_info->chunk_mutex); |
| 10932 | if (ret) |
| 10933 | return ret; |
| 10934 | |
| 10935 | down_read(&fs_info->commit_root_sem); |
| 10936 | |
| 10937 | spin_lock(&fs_info->trans_lock); |
| 10938 | trans = fs_info->running_transaction; |
| 10939 | if (trans) |
| 10940 | refcount_inc(&trans->use_count); |
| 10941 | spin_unlock(&fs_info->trans_lock); |
| 10942 | |
| 10943 | ret = find_free_dev_extent_start(trans, device, minlen, start, |
| 10944 | &start, &len); |
| 10945 | if (trans) |
| 10946 | btrfs_put_transaction(trans); |
| 10947 | |
| 10948 | if (ret) { |
| 10949 | up_read(&fs_info->commit_root_sem); |
| 10950 | mutex_unlock(&fs_info->chunk_mutex); |
| 10951 | if (ret == -ENOSPC) |
| 10952 | ret = 0; |
| 10953 | break; |
| 10954 | } |
| 10955 | |
| 10956 | ret = btrfs_issue_discard(device->bdev, start, len, &bytes); |
| 10957 | up_read(&fs_info->commit_root_sem); |
| 10958 | mutex_unlock(&fs_info->chunk_mutex); |
| 10959 | |
| 10960 | if (ret) |
| 10961 | break; |
| 10962 | |
| 10963 | start += len; |
| 10964 | *trimmed += bytes; |
| 10965 | |
| 10966 | if (fatal_signal_pending(current)) { |
| 10967 | ret = -ERESTARTSYS; |
| 10968 | break; |
| 10969 | } |
| 10970 | |
| 10971 | cond_resched(); |
| 10972 | } |
| 10973 | |
| 10974 | return ret; |
| 10975 | } |
| 10976 | |
| 10977 | int btrfs_trim_fs(struct btrfs_fs_info *fs_info, struct fstrim_range *range) |
| 10978 | { |
| 10979 | struct btrfs_block_group_cache *cache = NULL; |
| 10980 | struct btrfs_device *device; |
| 10981 | struct list_head *devices; |
| 10982 | u64 group_trimmed; |
| 10983 | u64 start; |
| 10984 | u64 end; |
| 10985 | u64 trimmed = 0; |
| 10986 | u64 total_bytes = btrfs_super_total_bytes(fs_info->super_copy); |
| 10987 | int ret = 0; |
| 10988 | |
| 10989 | /* |
| 10990 | * try to trim all FS space, our block group may start from non-zero. |
| 10991 | */ |
| 10992 | if (range->len == total_bytes) |
| 10993 | cache = btrfs_lookup_first_block_group(fs_info, range->start); |
| 10994 | else |
| 10995 | cache = btrfs_lookup_block_group(fs_info, range->start); |
| 10996 | |
| 10997 | while (cache) { |
| 10998 | if (cache->key.objectid >= (range->start + range->len)) { |
| 10999 | btrfs_put_block_group(cache); |
| 11000 | break; |
| 11001 | } |
| 11002 | |
| 11003 | start = max(range->start, cache->key.objectid); |
| 11004 | end = min(range->start + range->len, |
| 11005 | cache->key.objectid + cache->key.offset); |
| 11006 | |
| 11007 | if (end - start >= range->minlen) { |
| 11008 | if (!block_group_cache_done(cache)) { |
| 11009 | ret = cache_block_group(cache, 0); |
| 11010 | if (ret) { |
| 11011 | btrfs_put_block_group(cache); |
| 11012 | break; |
| 11013 | } |
| 11014 | ret = wait_block_group_cache_done(cache); |
| 11015 | if (ret) { |
| 11016 | btrfs_put_block_group(cache); |
| 11017 | break; |
| 11018 | } |
| 11019 | } |
| 11020 | ret = btrfs_trim_block_group(cache, |
| 11021 | &group_trimmed, |
| 11022 | start, |
| 11023 | end, |
| 11024 | range->minlen); |
| 11025 | |
| 11026 | trimmed += group_trimmed; |
| 11027 | if (ret) { |
| 11028 | btrfs_put_block_group(cache); |
| 11029 | break; |
| 11030 | } |
| 11031 | } |
| 11032 | |
| 11033 | cache = next_block_group(fs_info, cache); |
| 11034 | } |
| 11035 | |
| 11036 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
| 11037 | devices = &fs_info->fs_devices->alloc_list; |
| 11038 | list_for_each_entry(device, devices, dev_alloc_list) { |
| 11039 | ret = btrfs_trim_free_extents(device, range->minlen, |
| 11040 | &group_trimmed); |
| 11041 | if (ret) |
| 11042 | break; |
| 11043 | |
| 11044 | trimmed += group_trimmed; |
| 11045 | } |
| 11046 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
| 11047 | |
| 11048 | range->len = trimmed; |
| 11049 | return ret; |
| 11050 | } |
| 11051 | |
| 11052 | /* |
| 11053 | * btrfs_{start,end}_write_no_snapshotting() are similar to |
| 11054 | * mnt_{want,drop}_write(), they are used to prevent some tasks from writing |
| 11055 | * data into the page cache through nocow before the subvolume is snapshoted, |
| 11056 | * but flush the data into disk after the snapshot creation, or to prevent |
| 11057 | * operations while snapshotting is ongoing and that cause the snapshot to be |
| 11058 | * inconsistent (writes followed by expanding truncates for example). |
| 11059 | */ |
| 11060 | void btrfs_end_write_no_snapshotting(struct btrfs_root *root) |
| 11061 | { |
| 11062 | percpu_counter_dec(&root->subv_writers->counter); |
| 11063 | cond_wake_up(&root->subv_writers->wait); |
| 11064 | } |
| 11065 | |
| 11066 | int btrfs_start_write_no_snapshotting(struct btrfs_root *root) |
| 11067 | { |
| 11068 | if (atomic_read(&root->will_be_snapshotted)) |
| 11069 | return 0; |
| 11070 | |
| 11071 | percpu_counter_inc(&root->subv_writers->counter); |
| 11072 | /* |
| 11073 | * Make sure counter is updated before we check for snapshot creation. |
| 11074 | */ |
| 11075 | smp_mb(); |
| 11076 | if (atomic_read(&root->will_be_snapshotted)) { |
| 11077 | btrfs_end_write_no_snapshotting(root); |
| 11078 | return 0; |
| 11079 | } |
| 11080 | return 1; |
| 11081 | } |
| 11082 | |
| 11083 | void btrfs_wait_for_snapshot_creation(struct btrfs_root *root) |
| 11084 | { |
| 11085 | while (true) { |
| 11086 | int ret; |
| 11087 | |
| 11088 | ret = btrfs_start_write_no_snapshotting(root); |
| 11089 | if (ret) |
| 11090 | break; |
| 11091 | wait_var_event(&root->will_be_snapshotted, |
| 11092 | !atomic_read(&root->will_be_snapshotted)); |
| 11093 | } |
| 11094 | } |