| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * Copyright (C) 2007 Oracle. All rights reserved. |
| 4 | */ |
| 5 | |
| 6 | #include <linux/fs.h> |
| 7 | #include <linux/slab.h> |
| 8 | #include <linux/sched.h> |
| 9 | #include <linux/sched/mm.h> |
| 10 | #include <linux/writeback.h> |
| 11 | #include <linux/pagemap.h> |
| 12 | #include <linux/blkdev.h> |
| 13 | #include <linux/uuid.h> |
| 14 | #include <linux/timekeeping.h> |
| 15 | #include "misc.h" |
| 16 | #include "ctree.h" |
| 17 | #include "disk-io.h" |
| 18 | #include "transaction.h" |
| 19 | #include "locking.h" |
| 20 | #include "tree-log.h" |
| 21 | #include "volumes.h" |
| 22 | #include "dev-replace.h" |
| 23 | #include "qgroup.h" |
| 24 | #include "block-group.h" |
| 25 | #include "space-info.h" |
| 26 | #include "zoned.h" |
| 27 | #include "fs.h" |
| 28 | #include "accessors.h" |
| 29 | #include "extent-tree.h" |
| 30 | #include "root-tree.h" |
| 31 | #include "defrag.h" |
| 32 | #include "dir-item.h" |
| 33 | #include "uuid-tree.h" |
| 34 | #include "ioctl.h" |
| 35 | #include "relocation.h" |
| 36 | #include "scrub.h" |
| 37 | |
| 38 | static struct kmem_cache *btrfs_trans_handle_cachep; |
| 39 | |
| 40 | #define BTRFS_ROOT_TRANS_TAG 0 |
| 41 | |
| 42 | /* |
| 43 | * Transaction states and transitions |
| 44 | * |
| 45 | * No running transaction (fs tree blocks are not modified) |
| 46 | * | |
| 47 | * | To next stage: |
| 48 | * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). |
| 49 | * V |
| 50 | * Transaction N [[TRANS_STATE_RUNNING]] |
| 51 | * | |
| 52 | * | New trans handles can be attached to transaction N by calling all |
| 53 | * | start_transaction() variants. |
| 54 | * | |
| 55 | * | To next stage: |
| 56 | * | Call btrfs_commit_transaction() on any trans handle attached to |
| 57 | * | transaction N |
| 58 | * V |
| 59 | * Transaction N [[TRANS_STATE_COMMIT_START]] |
| 60 | * | |
| 61 | * | Will wait for previous running transaction to completely finish if there |
| 62 | * | is one |
| 63 | * | |
| 64 | * | Then one of the following happes: |
| 65 | * | - Wait for all other trans handle holders to release. |
| 66 | * | The btrfs_commit_transaction() caller will do the commit work. |
| 67 | * | - Wait for current transaction to be committed by others. |
| 68 | * | Other btrfs_commit_transaction() caller will do the commit work. |
| 69 | * | |
| 70 | * | At this stage, only btrfs_join_transaction*() variants can attach |
| 71 | * | to this running transaction. |
| 72 | * | All other variants will wait for current one to finish and attach to |
| 73 | * | transaction N+1. |
| 74 | * | |
| 75 | * | To next stage: |
| 76 | * | Caller is chosen to commit transaction N, and all other trans handle |
| 77 | * | haven been released. |
| 78 | * V |
| 79 | * Transaction N [[TRANS_STATE_COMMIT_DOING]] |
| 80 | * | |
| 81 | * | The heavy lifting transaction work is started. |
| 82 | * | From running delayed refs (modifying extent tree) to creating pending |
| 83 | * | snapshots, running qgroups. |
| 84 | * | In short, modify supporting trees to reflect modifications of subvolume |
| 85 | * | trees. |
| 86 | * | |
| 87 | * | At this stage, all start_transaction() calls will wait for this |
| 88 | * | transaction to finish and attach to transaction N+1. |
| 89 | * | |
| 90 | * | To next stage: |
| 91 | * | Until all supporting trees are updated. |
| 92 | * V |
| 93 | * Transaction N [[TRANS_STATE_UNBLOCKED]] |
| 94 | * | Transaction N+1 |
| 95 | * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] |
| 96 | * | need to write them back to disk and update | |
| 97 | * | super blocks. | |
| 98 | * | | |
| 99 | * | At this stage, new transaction is allowed to | |
| 100 | * | start. | |
| 101 | * | All new start_transaction() calls will be | |
| 102 | * | attached to transid N+1. | |
| 103 | * | | |
| 104 | * | To next stage: | |
| 105 | * | Until all tree blocks are super blocks are | |
| 106 | * | written to block devices | |
| 107 | * V | |
| 108 | * Transaction N [[TRANS_STATE_COMPLETED]] V |
| 109 | * All tree blocks and super blocks are written. Transaction N+1 |
| 110 | * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] |
| 111 | * data structures will be cleaned up. | Life goes on |
| 112 | */ |
| 113 | static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { |
| 114 | [TRANS_STATE_RUNNING] = 0U, |
| 115 | [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), |
| 116 | [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | |
| 117 | __TRANS_ATTACH | |
| 118 | __TRANS_JOIN | |
| 119 | __TRANS_JOIN_NOSTART), |
| 120 | [TRANS_STATE_UNBLOCKED] = (__TRANS_START | |
| 121 | __TRANS_ATTACH | |
| 122 | __TRANS_JOIN | |
| 123 | __TRANS_JOIN_NOLOCK | |
| 124 | __TRANS_JOIN_NOSTART), |
| 125 | [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | |
| 126 | __TRANS_ATTACH | |
| 127 | __TRANS_JOIN | |
| 128 | __TRANS_JOIN_NOLOCK | |
| 129 | __TRANS_JOIN_NOSTART), |
| 130 | [TRANS_STATE_COMPLETED] = (__TRANS_START | |
| 131 | __TRANS_ATTACH | |
| 132 | __TRANS_JOIN | |
| 133 | __TRANS_JOIN_NOLOCK | |
| 134 | __TRANS_JOIN_NOSTART), |
| 135 | }; |
| 136 | |
| 137 | void btrfs_put_transaction(struct btrfs_transaction *transaction) |
| 138 | { |
| 139 | WARN_ON(refcount_read(&transaction->use_count) == 0); |
| 140 | if (refcount_dec_and_test(&transaction->use_count)) { |
| 141 | BUG_ON(!list_empty(&transaction->list)); |
| 142 | WARN_ON(!RB_EMPTY_ROOT( |
| 143 | &transaction->delayed_refs.href_root.rb_root)); |
| 144 | WARN_ON(!RB_EMPTY_ROOT( |
| 145 | &transaction->delayed_refs.dirty_extent_root)); |
| 146 | if (transaction->delayed_refs.pending_csums) |
| 147 | btrfs_err(transaction->fs_info, |
| 148 | "pending csums is %llu", |
| 149 | transaction->delayed_refs.pending_csums); |
| 150 | /* |
| 151 | * If any block groups are found in ->deleted_bgs then it's |
| 152 | * because the transaction was aborted and a commit did not |
| 153 | * happen (things failed before writing the new superblock |
| 154 | * and calling btrfs_finish_extent_commit()), so we can not |
| 155 | * discard the physical locations of the block groups. |
| 156 | */ |
| 157 | while (!list_empty(&transaction->deleted_bgs)) { |
| 158 | struct btrfs_block_group *cache; |
| 159 | |
| 160 | cache = list_first_entry(&transaction->deleted_bgs, |
| 161 | struct btrfs_block_group, |
| 162 | bg_list); |
| 163 | list_del_init(&cache->bg_list); |
| 164 | btrfs_unfreeze_block_group(cache); |
| 165 | btrfs_put_block_group(cache); |
| 166 | } |
| 167 | WARN_ON(!list_empty(&transaction->dev_update_list)); |
| 168 | kfree(transaction); |
| 169 | } |
| 170 | } |
| 171 | |
| 172 | static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) |
| 173 | { |
| 174 | struct btrfs_transaction *cur_trans = trans->transaction; |
| 175 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 176 | struct btrfs_root *root, *tmp; |
| 177 | |
| 178 | /* |
| 179 | * At this point no one can be using this transaction to modify any tree |
| 180 | * and no one can start another transaction to modify any tree either. |
| 181 | */ |
| 182 | ASSERT(cur_trans->state == TRANS_STATE_COMMIT_DOING); |
| 183 | |
| 184 | down_write(&fs_info->commit_root_sem); |
| 185 | |
| 186 | if (test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) |
| 187 | fs_info->last_reloc_trans = trans->transid; |
| 188 | |
| 189 | list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, |
| 190 | dirty_list) { |
| 191 | list_del_init(&root->dirty_list); |
| 192 | free_extent_buffer(root->commit_root); |
| 193 | root->commit_root = btrfs_root_node(root); |
| 194 | extent_io_tree_release(&root->dirty_log_pages); |
| 195 | btrfs_qgroup_clean_swapped_blocks(root); |
| 196 | } |
| 197 | |
| 198 | /* We can free old roots now. */ |
| 199 | spin_lock(&cur_trans->dropped_roots_lock); |
| 200 | while (!list_empty(&cur_trans->dropped_roots)) { |
| 201 | root = list_first_entry(&cur_trans->dropped_roots, |
| 202 | struct btrfs_root, root_list); |
| 203 | list_del_init(&root->root_list); |
| 204 | spin_unlock(&cur_trans->dropped_roots_lock); |
| 205 | btrfs_free_log(trans, root); |
| 206 | btrfs_drop_and_free_fs_root(fs_info, root); |
| 207 | spin_lock(&cur_trans->dropped_roots_lock); |
| 208 | } |
| 209 | spin_unlock(&cur_trans->dropped_roots_lock); |
| 210 | |
| 211 | up_write(&fs_info->commit_root_sem); |
| 212 | } |
| 213 | |
| 214 | static inline void extwriter_counter_inc(struct btrfs_transaction *trans, |
| 215 | unsigned int type) |
| 216 | { |
| 217 | if (type & TRANS_EXTWRITERS) |
| 218 | atomic_inc(&trans->num_extwriters); |
| 219 | } |
| 220 | |
| 221 | static inline void extwriter_counter_dec(struct btrfs_transaction *trans, |
| 222 | unsigned int type) |
| 223 | { |
| 224 | if (type & TRANS_EXTWRITERS) |
| 225 | atomic_dec(&trans->num_extwriters); |
| 226 | } |
| 227 | |
| 228 | static inline void extwriter_counter_init(struct btrfs_transaction *trans, |
| 229 | unsigned int type) |
| 230 | { |
| 231 | atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); |
| 232 | } |
| 233 | |
| 234 | static inline int extwriter_counter_read(struct btrfs_transaction *trans) |
| 235 | { |
| 236 | return atomic_read(&trans->num_extwriters); |
| 237 | } |
| 238 | |
| 239 | /* |
| 240 | * To be called after doing the chunk btree updates right after allocating a new |
| 241 | * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a |
| 242 | * chunk after all chunk btree updates and after finishing the second phase of |
| 243 | * chunk allocation (btrfs_create_pending_block_groups()) in case some block |
| 244 | * group had its chunk item insertion delayed to the second phase. |
| 245 | */ |
| 246 | void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) |
| 247 | { |
| 248 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 249 | |
| 250 | if (!trans->chunk_bytes_reserved) |
| 251 | return; |
| 252 | |
| 253 | btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, |
| 254 | trans->chunk_bytes_reserved, NULL); |
| 255 | trans->chunk_bytes_reserved = 0; |
| 256 | } |
| 257 | |
| 258 | /* |
| 259 | * either allocate a new transaction or hop into the existing one |
| 260 | */ |
| 261 | static noinline int join_transaction(struct btrfs_fs_info *fs_info, |
| 262 | unsigned int type) |
| 263 | { |
| 264 | struct btrfs_transaction *cur_trans; |
| 265 | |
| 266 | spin_lock(&fs_info->trans_lock); |
| 267 | loop: |
| 268 | /* The file system has been taken offline. No new transactions. */ |
| 269 | if (BTRFS_FS_ERROR(fs_info)) { |
| 270 | spin_unlock(&fs_info->trans_lock); |
| 271 | return -EROFS; |
| 272 | } |
| 273 | |
| 274 | cur_trans = fs_info->running_transaction; |
| 275 | if (cur_trans) { |
| 276 | if (TRANS_ABORTED(cur_trans)) { |
| 277 | spin_unlock(&fs_info->trans_lock); |
| 278 | return cur_trans->aborted; |
| 279 | } |
| 280 | if (btrfs_blocked_trans_types[cur_trans->state] & type) { |
| 281 | spin_unlock(&fs_info->trans_lock); |
| 282 | return -EBUSY; |
| 283 | } |
| 284 | refcount_inc(&cur_trans->use_count); |
| 285 | atomic_inc(&cur_trans->num_writers); |
| 286 | extwriter_counter_inc(cur_trans, type); |
| 287 | spin_unlock(&fs_info->trans_lock); |
| 288 | btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); |
| 289 | btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); |
| 290 | return 0; |
| 291 | } |
| 292 | spin_unlock(&fs_info->trans_lock); |
| 293 | |
| 294 | /* |
| 295 | * If we are ATTACH, we just want to catch the current transaction, |
| 296 | * and commit it. If there is no transaction, just return ENOENT. |
| 297 | */ |
| 298 | if (type == TRANS_ATTACH) |
| 299 | return -ENOENT; |
| 300 | |
| 301 | /* |
| 302 | * JOIN_NOLOCK only happens during the transaction commit, so |
| 303 | * it is impossible that ->running_transaction is NULL |
| 304 | */ |
| 305 | BUG_ON(type == TRANS_JOIN_NOLOCK); |
| 306 | |
| 307 | cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); |
| 308 | if (!cur_trans) |
| 309 | return -ENOMEM; |
| 310 | |
| 311 | btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); |
| 312 | btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); |
| 313 | |
| 314 | spin_lock(&fs_info->trans_lock); |
| 315 | if (fs_info->running_transaction) { |
| 316 | /* |
| 317 | * someone started a transaction after we unlocked. Make sure |
| 318 | * to redo the checks above |
| 319 | */ |
| 320 | btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); |
| 321 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); |
| 322 | kfree(cur_trans); |
| 323 | goto loop; |
| 324 | } else if (BTRFS_FS_ERROR(fs_info)) { |
| 325 | spin_unlock(&fs_info->trans_lock); |
| 326 | btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); |
| 327 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); |
| 328 | kfree(cur_trans); |
| 329 | return -EROFS; |
| 330 | } |
| 331 | |
| 332 | cur_trans->fs_info = fs_info; |
| 333 | atomic_set(&cur_trans->pending_ordered, 0); |
| 334 | init_waitqueue_head(&cur_trans->pending_wait); |
| 335 | atomic_set(&cur_trans->num_writers, 1); |
| 336 | extwriter_counter_init(cur_trans, type); |
| 337 | init_waitqueue_head(&cur_trans->writer_wait); |
| 338 | init_waitqueue_head(&cur_trans->commit_wait); |
| 339 | cur_trans->state = TRANS_STATE_RUNNING; |
| 340 | /* |
| 341 | * One for this trans handle, one so it will live on until we |
| 342 | * commit the transaction. |
| 343 | */ |
| 344 | refcount_set(&cur_trans->use_count, 2); |
| 345 | cur_trans->flags = 0; |
| 346 | cur_trans->start_time = ktime_get_seconds(); |
| 347 | |
| 348 | memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); |
| 349 | |
| 350 | cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; |
| 351 | cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; |
| 352 | atomic_set(&cur_trans->delayed_refs.num_entries, 0); |
| 353 | |
| 354 | /* |
| 355 | * although the tree mod log is per file system and not per transaction, |
| 356 | * the log must never go across transaction boundaries. |
| 357 | */ |
| 358 | smp_mb(); |
| 359 | if (!list_empty(&fs_info->tree_mod_seq_list)) |
| 360 | WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); |
| 361 | if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) |
| 362 | WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); |
| 363 | atomic64_set(&fs_info->tree_mod_seq, 0); |
| 364 | |
| 365 | spin_lock_init(&cur_trans->delayed_refs.lock); |
| 366 | |
| 367 | INIT_LIST_HEAD(&cur_trans->pending_snapshots); |
| 368 | INIT_LIST_HEAD(&cur_trans->dev_update_list); |
| 369 | INIT_LIST_HEAD(&cur_trans->switch_commits); |
| 370 | INIT_LIST_HEAD(&cur_trans->dirty_bgs); |
| 371 | INIT_LIST_HEAD(&cur_trans->io_bgs); |
| 372 | INIT_LIST_HEAD(&cur_trans->dropped_roots); |
| 373 | mutex_init(&cur_trans->cache_write_mutex); |
| 374 | spin_lock_init(&cur_trans->dirty_bgs_lock); |
| 375 | INIT_LIST_HEAD(&cur_trans->deleted_bgs); |
| 376 | spin_lock_init(&cur_trans->dropped_roots_lock); |
| 377 | INIT_LIST_HEAD(&cur_trans->releasing_ebs); |
| 378 | spin_lock_init(&cur_trans->releasing_ebs_lock); |
| 379 | list_add_tail(&cur_trans->list, &fs_info->trans_list); |
| 380 | extent_io_tree_init(fs_info, &cur_trans->dirty_pages, |
| 381 | IO_TREE_TRANS_DIRTY_PAGES); |
| 382 | extent_io_tree_init(fs_info, &cur_trans->pinned_extents, |
| 383 | IO_TREE_FS_PINNED_EXTENTS); |
| 384 | fs_info->generation++; |
| 385 | cur_trans->transid = fs_info->generation; |
| 386 | fs_info->running_transaction = cur_trans; |
| 387 | cur_trans->aborted = 0; |
| 388 | spin_unlock(&fs_info->trans_lock); |
| 389 | |
| 390 | return 0; |
| 391 | } |
| 392 | |
| 393 | /* |
| 394 | * This does all the record keeping required to make sure that a shareable root |
| 395 | * is properly recorded in a given transaction. This is required to make sure |
| 396 | * the old root from before we joined the transaction is deleted when the |
| 397 | * transaction commits. |
| 398 | */ |
| 399 | static int record_root_in_trans(struct btrfs_trans_handle *trans, |
| 400 | struct btrfs_root *root, |
| 401 | int force) |
| 402 | { |
| 403 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 404 | int ret = 0; |
| 405 | |
| 406 | if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && |
| 407 | root->last_trans < trans->transid) || force) { |
| 408 | WARN_ON(!force && root->commit_root != root->node); |
| 409 | |
| 410 | /* |
| 411 | * see below for IN_TRANS_SETUP usage rules |
| 412 | * we have the reloc mutex held now, so there |
| 413 | * is only one writer in this function |
| 414 | */ |
| 415 | set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
| 416 | |
| 417 | /* make sure readers find IN_TRANS_SETUP before |
| 418 | * they find our root->last_trans update |
| 419 | */ |
| 420 | smp_wmb(); |
| 421 | |
| 422 | spin_lock(&fs_info->fs_roots_radix_lock); |
| 423 | if (root->last_trans == trans->transid && !force) { |
| 424 | spin_unlock(&fs_info->fs_roots_radix_lock); |
| 425 | return 0; |
| 426 | } |
| 427 | radix_tree_tag_set(&fs_info->fs_roots_radix, |
| 428 | (unsigned long)root->root_key.objectid, |
| 429 | BTRFS_ROOT_TRANS_TAG); |
| 430 | spin_unlock(&fs_info->fs_roots_radix_lock); |
| 431 | root->last_trans = trans->transid; |
| 432 | |
| 433 | /* this is pretty tricky. We don't want to |
| 434 | * take the relocation lock in btrfs_record_root_in_trans |
| 435 | * unless we're really doing the first setup for this root in |
| 436 | * this transaction. |
| 437 | * |
| 438 | * Normally we'd use root->last_trans as a flag to decide |
| 439 | * if we want to take the expensive mutex. |
| 440 | * |
| 441 | * But, we have to set root->last_trans before we |
| 442 | * init the relocation root, otherwise, we trip over warnings |
| 443 | * in ctree.c. The solution used here is to flag ourselves |
| 444 | * with root IN_TRANS_SETUP. When this is 1, we're still |
| 445 | * fixing up the reloc trees and everyone must wait. |
| 446 | * |
| 447 | * When this is zero, they can trust root->last_trans and fly |
| 448 | * through btrfs_record_root_in_trans without having to take the |
| 449 | * lock. smp_wmb() makes sure that all the writes above are |
| 450 | * done before we pop in the zero below |
| 451 | */ |
| 452 | ret = btrfs_init_reloc_root(trans, root); |
| 453 | smp_mb__before_atomic(); |
| 454 | clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); |
| 455 | } |
| 456 | return ret; |
| 457 | } |
| 458 | |
| 459 | |
| 460 | void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, |
| 461 | struct btrfs_root *root) |
| 462 | { |
| 463 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 464 | struct btrfs_transaction *cur_trans = trans->transaction; |
| 465 | |
| 466 | /* Add ourselves to the transaction dropped list */ |
| 467 | spin_lock(&cur_trans->dropped_roots_lock); |
| 468 | list_add_tail(&root->root_list, &cur_trans->dropped_roots); |
| 469 | spin_unlock(&cur_trans->dropped_roots_lock); |
| 470 | |
| 471 | /* Make sure we don't try to update the root at commit time */ |
| 472 | spin_lock(&fs_info->fs_roots_radix_lock); |
| 473 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
| 474 | (unsigned long)root->root_key.objectid, |
| 475 | BTRFS_ROOT_TRANS_TAG); |
| 476 | spin_unlock(&fs_info->fs_roots_radix_lock); |
| 477 | } |
| 478 | |
| 479 | int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, |
| 480 | struct btrfs_root *root) |
| 481 | { |
| 482 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 483 | int ret; |
| 484 | |
| 485 | if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) |
| 486 | return 0; |
| 487 | |
| 488 | /* |
| 489 | * see record_root_in_trans for comments about IN_TRANS_SETUP usage |
| 490 | * and barriers |
| 491 | */ |
| 492 | smp_rmb(); |
| 493 | if (root->last_trans == trans->transid && |
| 494 | !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) |
| 495 | return 0; |
| 496 | |
| 497 | mutex_lock(&fs_info->reloc_mutex); |
| 498 | ret = record_root_in_trans(trans, root, 0); |
| 499 | mutex_unlock(&fs_info->reloc_mutex); |
| 500 | |
| 501 | return ret; |
| 502 | } |
| 503 | |
| 504 | static inline int is_transaction_blocked(struct btrfs_transaction *trans) |
| 505 | { |
| 506 | return (trans->state >= TRANS_STATE_COMMIT_START && |
| 507 | trans->state < TRANS_STATE_UNBLOCKED && |
| 508 | !TRANS_ABORTED(trans)); |
| 509 | } |
| 510 | |
| 511 | /* wait for commit against the current transaction to become unblocked |
| 512 | * when this is done, it is safe to start a new transaction, but the current |
| 513 | * transaction might not be fully on disk. |
| 514 | */ |
| 515 | static void wait_current_trans(struct btrfs_fs_info *fs_info) |
| 516 | { |
| 517 | struct btrfs_transaction *cur_trans; |
| 518 | |
| 519 | spin_lock(&fs_info->trans_lock); |
| 520 | cur_trans = fs_info->running_transaction; |
| 521 | if (cur_trans && is_transaction_blocked(cur_trans)) { |
| 522 | refcount_inc(&cur_trans->use_count); |
| 523 | spin_unlock(&fs_info->trans_lock); |
| 524 | |
| 525 | btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); |
| 526 | wait_event(fs_info->transaction_wait, |
| 527 | cur_trans->state >= TRANS_STATE_UNBLOCKED || |
| 528 | TRANS_ABORTED(cur_trans)); |
| 529 | btrfs_put_transaction(cur_trans); |
| 530 | } else { |
| 531 | spin_unlock(&fs_info->trans_lock); |
| 532 | } |
| 533 | } |
| 534 | |
| 535 | static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) |
| 536 | { |
| 537 | if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) |
| 538 | return 0; |
| 539 | |
| 540 | if (type == TRANS_START) |
| 541 | return 1; |
| 542 | |
| 543 | return 0; |
| 544 | } |
| 545 | |
| 546 | static inline bool need_reserve_reloc_root(struct btrfs_root *root) |
| 547 | { |
| 548 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 549 | |
| 550 | if (!fs_info->reloc_ctl || |
| 551 | !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || |
| 552 | root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || |
| 553 | root->reloc_root) |
| 554 | return false; |
| 555 | |
| 556 | return true; |
| 557 | } |
| 558 | |
| 559 | static struct btrfs_trans_handle * |
| 560 | start_transaction(struct btrfs_root *root, unsigned int num_items, |
| 561 | unsigned int type, enum btrfs_reserve_flush_enum flush, |
| 562 | bool enforce_qgroups) |
| 563 | { |
| 564 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 565 | struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; |
| 566 | struct btrfs_trans_handle *h; |
| 567 | struct btrfs_transaction *cur_trans; |
| 568 | u64 num_bytes = 0; |
| 569 | u64 qgroup_reserved = 0; |
| 570 | bool reloc_reserved = false; |
| 571 | bool do_chunk_alloc = false; |
| 572 | int ret; |
| 573 | |
| 574 | if (BTRFS_FS_ERROR(fs_info)) |
| 575 | return ERR_PTR(-EROFS); |
| 576 | |
| 577 | if (current->journal_info) { |
| 578 | WARN_ON(type & TRANS_EXTWRITERS); |
| 579 | h = current->journal_info; |
| 580 | refcount_inc(&h->use_count); |
| 581 | WARN_ON(refcount_read(&h->use_count) > 2); |
| 582 | h->orig_rsv = h->block_rsv; |
| 583 | h->block_rsv = NULL; |
| 584 | goto got_it; |
| 585 | } |
| 586 | |
| 587 | /* |
| 588 | * Do the reservation before we join the transaction so we can do all |
| 589 | * the appropriate flushing if need be. |
| 590 | */ |
| 591 | if (num_items && root != fs_info->chunk_root) { |
| 592 | struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; |
| 593 | u64 delayed_refs_bytes = 0; |
| 594 | |
| 595 | qgroup_reserved = num_items * fs_info->nodesize; |
| 596 | ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, |
| 597 | enforce_qgroups); |
| 598 | if (ret) |
| 599 | return ERR_PTR(ret); |
| 600 | |
| 601 | /* |
| 602 | * We want to reserve all the bytes we may need all at once, so |
| 603 | * we only do 1 enospc flushing cycle per transaction start. We |
| 604 | * accomplish this by simply assuming we'll do num_items worth |
| 605 | * of delayed refs updates in this trans handle, and refill that |
| 606 | * amount for whatever is missing in the reserve. |
| 607 | */ |
| 608 | num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); |
| 609 | if (flush == BTRFS_RESERVE_FLUSH_ALL && |
| 610 | !btrfs_block_rsv_full(delayed_refs_rsv)) { |
| 611 | delayed_refs_bytes = btrfs_calc_delayed_ref_bytes(fs_info, |
| 612 | num_items); |
| 613 | num_bytes += delayed_refs_bytes; |
| 614 | } |
| 615 | |
| 616 | /* |
| 617 | * Do the reservation for the relocation root creation |
| 618 | */ |
| 619 | if (need_reserve_reloc_root(root)) { |
| 620 | num_bytes += fs_info->nodesize; |
| 621 | reloc_reserved = true; |
| 622 | } |
| 623 | |
| 624 | ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes, flush); |
| 625 | if (ret) |
| 626 | goto reserve_fail; |
| 627 | if (delayed_refs_bytes) { |
| 628 | btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, |
| 629 | delayed_refs_bytes); |
| 630 | num_bytes -= delayed_refs_bytes; |
| 631 | } |
| 632 | |
| 633 | if (rsv->space_info->force_alloc) |
| 634 | do_chunk_alloc = true; |
| 635 | } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && |
| 636 | !btrfs_block_rsv_full(delayed_refs_rsv)) { |
| 637 | /* |
| 638 | * Some people call with btrfs_start_transaction(root, 0) |
| 639 | * because they can be throttled, but have some other mechanism |
| 640 | * for reserving space. We still want these guys to refill the |
| 641 | * delayed block_rsv so just add 1 items worth of reservation |
| 642 | * here. |
| 643 | */ |
| 644 | ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); |
| 645 | if (ret) |
| 646 | goto reserve_fail; |
| 647 | } |
| 648 | again: |
| 649 | h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); |
| 650 | if (!h) { |
| 651 | ret = -ENOMEM; |
| 652 | goto alloc_fail; |
| 653 | } |
| 654 | |
| 655 | /* |
| 656 | * If we are JOIN_NOLOCK we're already committing a transaction and |
| 657 | * waiting on this guy, so we don't need to do the sb_start_intwrite |
| 658 | * because we're already holding a ref. We need this because we could |
| 659 | * have raced in and did an fsync() on a file which can kick a commit |
| 660 | * and then we deadlock with somebody doing a freeze. |
| 661 | * |
| 662 | * If we are ATTACH, it means we just want to catch the current |
| 663 | * transaction and commit it, so we needn't do sb_start_intwrite(). |
| 664 | */ |
| 665 | if (type & __TRANS_FREEZABLE) |
| 666 | sb_start_intwrite(fs_info->sb); |
| 667 | |
| 668 | if (may_wait_transaction(fs_info, type)) |
| 669 | wait_current_trans(fs_info); |
| 670 | |
| 671 | do { |
| 672 | ret = join_transaction(fs_info, type); |
| 673 | if (ret == -EBUSY) { |
| 674 | wait_current_trans(fs_info); |
| 675 | if (unlikely(type == TRANS_ATTACH || |
| 676 | type == TRANS_JOIN_NOSTART)) |
| 677 | ret = -ENOENT; |
| 678 | } |
| 679 | } while (ret == -EBUSY); |
| 680 | |
| 681 | if (ret < 0) |
| 682 | goto join_fail; |
| 683 | |
| 684 | cur_trans = fs_info->running_transaction; |
| 685 | |
| 686 | h->transid = cur_trans->transid; |
| 687 | h->transaction = cur_trans; |
| 688 | refcount_set(&h->use_count, 1); |
| 689 | h->fs_info = root->fs_info; |
| 690 | |
| 691 | h->type = type; |
| 692 | INIT_LIST_HEAD(&h->new_bgs); |
| 693 | |
| 694 | smp_mb(); |
| 695 | if (cur_trans->state >= TRANS_STATE_COMMIT_START && |
| 696 | may_wait_transaction(fs_info, type)) { |
| 697 | current->journal_info = h; |
| 698 | btrfs_commit_transaction(h); |
| 699 | goto again; |
| 700 | } |
| 701 | |
| 702 | if (num_bytes) { |
| 703 | trace_btrfs_space_reservation(fs_info, "transaction", |
| 704 | h->transid, num_bytes, 1); |
| 705 | h->block_rsv = &fs_info->trans_block_rsv; |
| 706 | h->bytes_reserved = num_bytes; |
| 707 | h->reloc_reserved = reloc_reserved; |
| 708 | } |
| 709 | |
| 710 | got_it: |
| 711 | if (!current->journal_info) |
| 712 | current->journal_info = h; |
| 713 | |
| 714 | /* |
| 715 | * If the space_info is marked ALLOC_FORCE then we'll get upgraded to |
| 716 | * ALLOC_FORCE the first run through, and then we won't allocate for |
| 717 | * anybody else who races in later. We don't care about the return |
| 718 | * value here. |
| 719 | */ |
| 720 | if (do_chunk_alloc && num_bytes) { |
| 721 | u64 flags = h->block_rsv->space_info->flags; |
| 722 | |
| 723 | btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), |
| 724 | CHUNK_ALLOC_NO_FORCE); |
| 725 | } |
| 726 | |
| 727 | /* |
| 728 | * btrfs_record_root_in_trans() needs to alloc new extents, and may |
| 729 | * call btrfs_join_transaction() while we're also starting a |
| 730 | * transaction. |
| 731 | * |
| 732 | * Thus it need to be called after current->journal_info initialized, |
| 733 | * or we can deadlock. |
| 734 | */ |
| 735 | ret = btrfs_record_root_in_trans(h, root); |
| 736 | if (ret) { |
| 737 | /* |
| 738 | * The transaction handle is fully initialized and linked with |
| 739 | * other structures so it needs to be ended in case of errors, |
| 740 | * not just freed. |
| 741 | */ |
| 742 | btrfs_end_transaction(h); |
| 743 | return ERR_PTR(ret); |
| 744 | } |
| 745 | |
| 746 | return h; |
| 747 | |
| 748 | join_fail: |
| 749 | if (type & __TRANS_FREEZABLE) |
| 750 | sb_end_intwrite(fs_info->sb); |
| 751 | kmem_cache_free(btrfs_trans_handle_cachep, h); |
| 752 | alloc_fail: |
| 753 | if (num_bytes) |
| 754 | btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, |
| 755 | num_bytes, NULL); |
| 756 | reserve_fail: |
| 757 | btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); |
| 758 | return ERR_PTR(ret); |
| 759 | } |
| 760 | |
| 761 | struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, |
| 762 | unsigned int num_items) |
| 763 | { |
| 764 | return start_transaction(root, num_items, TRANS_START, |
| 765 | BTRFS_RESERVE_FLUSH_ALL, true); |
| 766 | } |
| 767 | |
| 768 | struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( |
| 769 | struct btrfs_root *root, |
| 770 | unsigned int num_items) |
| 771 | { |
| 772 | return start_transaction(root, num_items, TRANS_START, |
| 773 | BTRFS_RESERVE_FLUSH_ALL_STEAL, false); |
| 774 | } |
| 775 | |
| 776 | struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) |
| 777 | { |
| 778 | return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, |
| 779 | true); |
| 780 | } |
| 781 | |
| 782 | struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) |
| 783 | { |
| 784 | return start_transaction(root, 0, TRANS_JOIN_NOLOCK, |
| 785 | BTRFS_RESERVE_NO_FLUSH, true); |
| 786 | } |
| 787 | |
| 788 | /* |
| 789 | * Similar to regular join but it never starts a transaction when none is |
| 790 | * running or after waiting for the current one to finish. |
| 791 | */ |
| 792 | struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) |
| 793 | { |
| 794 | return start_transaction(root, 0, TRANS_JOIN_NOSTART, |
| 795 | BTRFS_RESERVE_NO_FLUSH, true); |
| 796 | } |
| 797 | |
| 798 | /* |
| 799 | * btrfs_attach_transaction() - catch the running transaction |
| 800 | * |
| 801 | * It is used when we want to commit the current the transaction, but |
| 802 | * don't want to start a new one. |
| 803 | * |
| 804 | * Note: If this function return -ENOENT, it just means there is no |
| 805 | * running transaction. But it is possible that the inactive transaction |
| 806 | * is still in the memory, not fully on disk. If you hope there is no |
| 807 | * inactive transaction in the fs when -ENOENT is returned, you should |
| 808 | * invoke |
| 809 | * btrfs_attach_transaction_barrier() |
| 810 | */ |
| 811 | struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) |
| 812 | { |
| 813 | return start_transaction(root, 0, TRANS_ATTACH, |
| 814 | BTRFS_RESERVE_NO_FLUSH, true); |
| 815 | } |
| 816 | |
| 817 | /* |
| 818 | * btrfs_attach_transaction_barrier() - catch the running transaction |
| 819 | * |
| 820 | * It is similar to the above function, the difference is this one |
| 821 | * will wait for all the inactive transactions until they fully |
| 822 | * complete. |
| 823 | */ |
| 824 | struct btrfs_trans_handle * |
| 825 | btrfs_attach_transaction_barrier(struct btrfs_root *root) |
| 826 | { |
| 827 | struct btrfs_trans_handle *trans; |
| 828 | |
| 829 | trans = start_transaction(root, 0, TRANS_ATTACH, |
| 830 | BTRFS_RESERVE_NO_FLUSH, true); |
| 831 | if (trans == ERR_PTR(-ENOENT)) |
| 832 | btrfs_wait_for_commit(root->fs_info, 0); |
| 833 | |
| 834 | return trans; |
| 835 | } |
| 836 | |
| 837 | /* Wait for a transaction commit to reach at least the given state. */ |
| 838 | static noinline void wait_for_commit(struct btrfs_transaction *commit, |
| 839 | const enum btrfs_trans_state min_state) |
| 840 | { |
| 841 | struct btrfs_fs_info *fs_info = commit->fs_info; |
| 842 | u64 transid = commit->transid; |
| 843 | bool put = false; |
| 844 | |
| 845 | /* |
| 846 | * At the moment this function is called with min_state either being |
| 847 | * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED. |
| 848 | */ |
| 849 | if (min_state == TRANS_STATE_COMPLETED) |
| 850 | btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); |
| 851 | else |
| 852 | btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); |
| 853 | |
| 854 | while (1) { |
| 855 | wait_event(commit->commit_wait, commit->state >= min_state); |
| 856 | if (put) |
| 857 | btrfs_put_transaction(commit); |
| 858 | |
| 859 | if (min_state < TRANS_STATE_COMPLETED) |
| 860 | break; |
| 861 | |
| 862 | /* |
| 863 | * A transaction isn't really completed until all of the |
| 864 | * previous transactions are completed, but with fsync we can |
| 865 | * end up with SUPER_COMMITTED transactions before a COMPLETED |
| 866 | * transaction. Wait for those. |
| 867 | */ |
| 868 | |
| 869 | spin_lock(&fs_info->trans_lock); |
| 870 | commit = list_first_entry_or_null(&fs_info->trans_list, |
| 871 | struct btrfs_transaction, |
| 872 | list); |
| 873 | if (!commit || commit->transid > transid) { |
| 874 | spin_unlock(&fs_info->trans_lock); |
| 875 | break; |
| 876 | } |
| 877 | refcount_inc(&commit->use_count); |
| 878 | put = true; |
| 879 | spin_unlock(&fs_info->trans_lock); |
| 880 | } |
| 881 | } |
| 882 | |
| 883 | int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) |
| 884 | { |
| 885 | struct btrfs_transaction *cur_trans = NULL, *t; |
| 886 | int ret = 0; |
| 887 | |
| 888 | if (transid) { |
| 889 | if (transid <= fs_info->last_trans_committed) |
| 890 | goto out; |
| 891 | |
| 892 | /* find specified transaction */ |
| 893 | spin_lock(&fs_info->trans_lock); |
| 894 | list_for_each_entry(t, &fs_info->trans_list, list) { |
| 895 | if (t->transid == transid) { |
| 896 | cur_trans = t; |
| 897 | refcount_inc(&cur_trans->use_count); |
| 898 | ret = 0; |
| 899 | break; |
| 900 | } |
| 901 | if (t->transid > transid) { |
| 902 | ret = 0; |
| 903 | break; |
| 904 | } |
| 905 | } |
| 906 | spin_unlock(&fs_info->trans_lock); |
| 907 | |
| 908 | /* |
| 909 | * The specified transaction doesn't exist, or we |
| 910 | * raced with btrfs_commit_transaction |
| 911 | */ |
| 912 | if (!cur_trans) { |
| 913 | if (transid > fs_info->last_trans_committed) |
| 914 | ret = -EINVAL; |
| 915 | goto out; |
| 916 | } |
| 917 | } else { |
| 918 | /* find newest transaction that is committing | committed */ |
| 919 | spin_lock(&fs_info->trans_lock); |
| 920 | list_for_each_entry_reverse(t, &fs_info->trans_list, |
| 921 | list) { |
| 922 | if (t->state >= TRANS_STATE_COMMIT_START) { |
| 923 | if (t->state == TRANS_STATE_COMPLETED) |
| 924 | break; |
| 925 | cur_trans = t; |
| 926 | refcount_inc(&cur_trans->use_count); |
| 927 | break; |
| 928 | } |
| 929 | } |
| 930 | spin_unlock(&fs_info->trans_lock); |
| 931 | if (!cur_trans) |
| 932 | goto out; /* nothing committing|committed */ |
| 933 | } |
| 934 | |
| 935 | wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); |
| 936 | btrfs_put_transaction(cur_trans); |
| 937 | out: |
| 938 | return ret; |
| 939 | } |
| 940 | |
| 941 | void btrfs_throttle(struct btrfs_fs_info *fs_info) |
| 942 | { |
| 943 | wait_current_trans(fs_info); |
| 944 | } |
| 945 | |
| 946 | bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) |
| 947 | { |
| 948 | struct btrfs_transaction *cur_trans = trans->transaction; |
| 949 | |
| 950 | if (cur_trans->state >= TRANS_STATE_COMMIT_START || |
| 951 | test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) |
| 952 | return true; |
| 953 | |
| 954 | if (btrfs_check_space_for_delayed_refs(trans->fs_info)) |
| 955 | return true; |
| 956 | |
| 957 | return !!btrfs_block_rsv_check(&trans->fs_info->global_block_rsv, 50); |
| 958 | } |
| 959 | |
| 960 | static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) |
| 961 | |
| 962 | { |
| 963 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 964 | |
| 965 | if (!trans->block_rsv) { |
| 966 | ASSERT(!trans->bytes_reserved); |
| 967 | return; |
| 968 | } |
| 969 | |
| 970 | if (!trans->bytes_reserved) |
| 971 | return; |
| 972 | |
| 973 | ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); |
| 974 | trace_btrfs_space_reservation(fs_info, "transaction", |
| 975 | trans->transid, trans->bytes_reserved, 0); |
| 976 | btrfs_block_rsv_release(fs_info, trans->block_rsv, |
| 977 | trans->bytes_reserved, NULL); |
| 978 | trans->bytes_reserved = 0; |
| 979 | } |
| 980 | |
| 981 | static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, |
| 982 | int throttle) |
| 983 | { |
| 984 | struct btrfs_fs_info *info = trans->fs_info; |
| 985 | struct btrfs_transaction *cur_trans = trans->transaction; |
| 986 | int err = 0; |
| 987 | |
| 988 | if (refcount_read(&trans->use_count) > 1) { |
| 989 | refcount_dec(&trans->use_count); |
| 990 | trans->block_rsv = trans->orig_rsv; |
| 991 | return 0; |
| 992 | } |
| 993 | |
| 994 | btrfs_trans_release_metadata(trans); |
| 995 | trans->block_rsv = NULL; |
| 996 | |
| 997 | btrfs_create_pending_block_groups(trans); |
| 998 | |
| 999 | btrfs_trans_release_chunk_metadata(trans); |
| 1000 | |
| 1001 | if (trans->type & __TRANS_FREEZABLE) |
| 1002 | sb_end_intwrite(info->sb); |
| 1003 | |
| 1004 | WARN_ON(cur_trans != info->running_transaction); |
| 1005 | WARN_ON(atomic_read(&cur_trans->num_writers) < 1); |
| 1006 | atomic_dec(&cur_trans->num_writers); |
| 1007 | extwriter_counter_dec(cur_trans, trans->type); |
| 1008 | |
| 1009 | cond_wake_up(&cur_trans->writer_wait); |
| 1010 | |
| 1011 | btrfs_lockdep_release(info, btrfs_trans_num_extwriters); |
| 1012 | btrfs_lockdep_release(info, btrfs_trans_num_writers); |
| 1013 | |
| 1014 | btrfs_put_transaction(cur_trans); |
| 1015 | |
| 1016 | if (current->journal_info == trans) |
| 1017 | current->journal_info = NULL; |
| 1018 | |
| 1019 | if (throttle) |
| 1020 | btrfs_run_delayed_iputs(info); |
| 1021 | |
| 1022 | if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) { |
| 1023 | wake_up_process(info->transaction_kthread); |
| 1024 | if (TRANS_ABORTED(trans)) |
| 1025 | err = trans->aborted; |
| 1026 | else |
| 1027 | err = -EROFS; |
| 1028 | } |
| 1029 | |
| 1030 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
| 1031 | return err; |
| 1032 | } |
| 1033 | |
| 1034 | int btrfs_end_transaction(struct btrfs_trans_handle *trans) |
| 1035 | { |
| 1036 | return __btrfs_end_transaction(trans, 0); |
| 1037 | } |
| 1038 | |
| 1039 | int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) |
| 1040 | { |
| 1041 | return __btrfs_end_transaction(trans, 1); |
| 1042 | } |
| 1043 | |
| 1044 | /* |
| 1045 | * when btree blocks are allocated, they have some corresponding bits set for |
| 1046 | * them in one of two extent_io trees. This is used to make sure all of |
| 1047 | * those extents are sent to disk but does not wait on them |
| 1048 | */ |
| 1049 | int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, |
| 1050 | struct extent_io_tree *dirty_pages, int mark) |
| 1051 | { |
| 1052 | int err = 0; |
| 1053 | int werr = 0; |
| 1054 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
| 1055 | struct extent_state *cached_state = NULL; |
| 1056 | u64 start = 0; |
| 1057 | u64 end; |
| 1058 | |
| 1059 | atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
| 1060 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
| 1061 | mark, &cached_state)) { |
| 1062 | bool wait_writeback = false; |
| 1063 | |
| 1064 | err = convert_extent_bit(dirty_pages, start, end, |
| 1065 | EXTENT_NEED_WAIT, |
| 1066 | mark, &cached_state); |
| 1067 | /* |
| 1068 | * convert_extent_bit can return -ENOMEM, which is most of the |
| 1069 | * time a temporary error. So when it happens, ignore the error |
| 1070 | * and wait for writeback of this range to finish - because we |
| 1071 | * failed to set the bit EXTENT_NEED_WAIT for the range, a call |
| 1072 | * to __btrfs_wait_marked_extents() would not know that |
| 1073 | * writeback for this range started and therefore wouldn't |
| 1074 | * wait for it to finish - we don't want to commit a |
| 1075 | * superblock that points to btree nodes/leafs for which |
| 1076 | * writeback hasn't finished yet (and without errors). |
| 1077 | * We cleanup any entries left in the io tree when committing |
| 1078 | * the transaction (through extent_io_tree_release()). |
| 1079 | */ |
| 1080 | if (err == -ENOMEM) { |
| 1081 | err = 0; |
| 1082 | wait_writeback = true; |
| 1083 | } |
| 1084 | if (!err) |
| 1085 | err = filemap_fdatawrite_range(mapping, start, end); |
| 1086 | if (err) |
| 1087 | werr = err; |
| 1088 | else if (wait_writeback) |
| 1089 | werr = filemap_fdatawait_range(mapping, start, end); |
| 1090 | free_extent_state(cached_state); |
| 1091 | cached_state = NULL; |
| 1092 | cond_resched(); |
| 1093 | start = end + 1; |
| 1094 | } |
| 1095 | atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); |
| 1096 | return werr; |
| 1097 | } |
| 1098 | |
| 1099 | /* |
| 1100 | * when btree blocks are allocated, they have some corresponding bits set for |
| 1101 | * them in one of two extent_io trees. This is used to make sure all of |
| 1102 | * those extents are on disk for transaction or log commit. We wait |
| 1103 | * on all the pages and clear them from the dirty pages state tree |
| 1104 | */ |
| 1105 | static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, |
| 1106 | struct extent_io_tree *dirty_pages) |
| 1107 | { |
| 1108 | int err = 0; |
| 1109 | int werr = 0; |
| 1110 | struct address_space *mapping = fs_info->btree_inode->i_mapping; |
| 1111 | struct extent_state *cached_state = NULL; |
| 1112 | u64 start = 0; |
| 1113 | u64 end; |
| 1114 | |
| 1115 | while (!find_first_extent_bit(dirty_pages, start, &start, &end, |
| 1116 | EXTENT_NEED_WAIT, &cached_state)) { |
| 1117 | /* |
| 1118 | * Ignore -ENOMEM errors returned by clear_extent_bit(). |
| 1119 | * When committing the transaction, we'll remove any entries |
| 1120 | * left in the io tree. For a log commit, we don't remove them |
| 1121 | * after committing the log because the tree can be accessed |
| 1122 | * concurrently - we do it only at transaction commit time when |
| 1123 | * it's safe to do it (through extent_io_tree_release()). |
| 1124 | */ |
| 1125 | err = clear_extent_bit(dirty_pages, start, end, |
| 1126 | EXTENT_NEED_WAIT, &cached_state); |
| 1127 | if (err == -ENOMEM) |
| 1128 | err = 0; |
| 1129 | if (!err) |
| 1130 | err = filemap_fdatawait_range(mapping, start, end); |
| 1131 | if (err) |
| 1132 | werr = err; |
| 1133 | free_extent_state(cached_state); |
| 1134 | cached_state = NULL; |
| 1135 | cond_resched(); |
| 1136 | start = end + 1; |
| 1137 | } |
| 1138 | if (err) |
| 1139 | werr = err; |
| 1140 | return werr; |
| 1141 | } |
| 1142 | |
| 1143 | static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, |
| 1144 | struct extent_io_tree *dirty_pages) |
| 1145 | { |
| 1146 | bool errors = false; |
| 1147 | int err; |
| 1148 | |
| 1149 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); |
| 1150 | if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) |
| 1151 | errors = true; |
| 1152 | |
| 1153 | if (errors && !err) |
| 1154 | err = -EIO; |
| 1155 | return err; |
| 1156 | } |
| 1157 | |
| 1158 | int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) |
| 1159 | { |
| 1160 | struct btrfs_fs_info *fs_info = log_root->fs_info; |
| 1161 | struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; |
| 1162 | bool errors = false; |
| 1163 | int err; |
| 1164 | |
| 1165 | ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); |
| 1166 | |
| 1167 | err = __btrfs_wait_marked_extents(fs_info, dirty_pages); |
| 1168 | if ((mark & EXTENT_DIRTY) && |
| 1169 | test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) |
| 1170 | errors = true; |
| 1171 | |
| 1172 | if ((mark & EXTENT_NEW) && |
| 1173 | test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) |
| 1174 | errors = true; |
| 1175 | |
| 1176 | if (errors && !err) |
| 1177 | err = -EIO; |
| 1178 | return err; |
| 1179 | } |
| 1180 | |
| 1181 | /* |
| 1182 | * When btree blocks are allocated the corresponding extents are marked dirty. |
| 1183 | * This function ensures such extents are persisted on disk for transaction or |
| 1184 | * log commit. |
| 1185 | * |
| 1186 | * @trans: transaction whose dirty pages we'd like to write |
| 1187 | */ |
| 1188 | static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) |
| 1189 | { |
| 1190 | int ret; |
| 1191 | int ret2; |
| 1192 | struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; |
| 1193 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 1194 | struct blk_plug plug; |
| 1195 | |
| 1196 | blk_start_plug(&plug); |
| 1197 | ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); |
| 1198 | blk_finish_plug(&plug); |
| 1199 | ret2 = btrfs_wait_extents(fs_info, dirty_pages); |
| 1200 | |
| 1201 | extent_io_tree_release(&trans->transaction->dirty_pages); |
| 1202 | |
| 1203 | if (ret) |
| 1204 | return ret; |
| 1205 | else if (ret2) |
| 1206 | return ret2; |
| 1207 | else |
| 1208 | return 0; |
| 1209 | } |
| 1210 | |
| 1211 | /* |
| 1212 | * this is used to update the root pointer in the tree of tree roots. |
| 1213 | * |
| 1214 | * But, in the case of the extent allocation tree, updating the root |
| 1215 | * pointer may allocate blocks which may change the root of the extent |
| 1216 | * allocation tree. |
| 1217 | * |
| 1218 | * So, this loops and repeats and makes sure the cowonly root didn't |
| 1219 | * change while the root pointer was being updated in the metadata. |
| 1220 | */ |
| 1221 | static int update_cowonly_root(struct btrfs_trans_handle *trans, |
| 1222 | struct btrfs_root *root) |
| 1223 | { |
| 1224 | int ret; |
| 1225 | u64 old_root_bytenr; |
| 1226 | u64 old_root_used; |
| 1227 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 1228 | struct btrfs_root *tree_root = fs_info->tree_root; |
| 1229 | |
| 1230 | old_root_used = btrfs_root_used(&root->root_item); |
| 1231 | |
| 1232 | while (1) { |
| 1233 | old_root_bytenr = btrfs_root_bytenr(&root->root_item); |
| 1234 | if (old_root_bytenr == root->node->start && |
| 1235 | old_root_used == btrfs_root_used(&root->root_item)) |
| 1236 | break; |
| 1237 | |
| 1238 | btrfs_set_root_node(&root->root_item, root->node); |
| 1239 | ret = btrfs_update_root(trans, tree_root, |
| 1240 | &root->root_key, |
| 1241 | &root->root_item); |
| 1242 | if (ret) |
| 1243 | return ret; |
| 1244 | |
| 1245 | old_root_used = btrfs_root_used(&root->root_item); |
| 1246 | } |
| 1247 | |
| 1248 | return 0; |
| 1249 | } |
| 1250 | |
| 1251 | /* |
| 1252 | * update all the cowonly tree roots on disk |
| 1253 | * |
| 1254 | * The error handling in this function may not be obvious. Any of the |
| 1255 | * failures will cause the file system to go offline. We still need |
| 1256 | * to clean up the delayed refs. |
| 1257 | */ |
| 1258 | static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) |
| 1259 | { |
| 1260 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 1261 | struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; |
| 1262 | struct list_head *io_bgs = &trans->transaction->io_bgs; |
| 1263 | struct list_head *next; |
| 1264 | struct extent_buffer *eb; |
| 1265 | int ret; |
| 1266 | |
| 1267 | /* |
| 1268 | * At this point no one can be using this transaction to modify any tree |
| 1269 | * and no one can start another transaction to modify any tree either. |
| 1270 | */ |
| 1271 | ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); |
| 1272 | |
| 1273 | eb = btrfs_lock_root_node(fs_info->tree_root); |
| 1274 | ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, |
| 1275 | 0, &eb, BTRFS_NESTING_COW); |
| 1276 | btrfs_tree_unlock(eb); |
| 1277 | free_extent_buffer(eb); |
| 1278 | |
| 1279 | if (ret) |
| 1280 | return ret; |
| 1281 | |
| 1282 | ret = btrfs_run_dev_stats(trans); |
| 1283 | if (ret) |
| 1284 | return ret; |
| 1285 | ret = btrfs_run_dev_replace(trans); |
| 1286 | if (ret) |
| 1287 | return ret; |
| 1288 | ret = btrfs_run_qgroups(trans); |
| 1289 | if (ret) |
| 1290 | return ret; |
| 1291 | |
| 1292 | ret = btrfs_setup_space_cache(trans); |
| 1293 | if (ret) |
| 1294 | return ret; |
| 1295 | |
| 1296 | again: |
| 1297 | while (!list_empty(&fs_info->dirty_cowonly_roots)) { |
| 1298 | struct btrfs_root *root; |
| 1299 | next = fs_info->dirty_cowonly_roots.next; |
| 1300 | list_del_init(next); |
| 1301 | root = list_entry(next, struct btrfs_root, dirty_list); |
| 1302 | clear_bit(BTRFS_ROOT_DIRTY, &root->state); |
| 1303 | |
| 1304 | list_add_tail(&root->dirty_list, |
| 1305 | &trans->transaction->switch_commits); |
| 1306 | ret = update_cowonly_root(trans, root); |
| 1307 | if (ret) |
| 1308 | return ret; |
| 1309 | } |
| 1310 | |
| 1311 | /* Now flush any delayed refs generated by updating all of the roots */ |
| 1312 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
| 1313 | if (ret) |
| 1314 | return ret; |
| 1315 | |
| 1316 | while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { |
| 1317 | ret = btrfs_write_dirty_block_groups(trans); |
| 1318 | if (ret) |
| 1319 | return ret; |
| 1320 | |
| 1321 | /* |
| 1322 | * We're writing the dirty block groups, which could generate |
| 1323 | * delayed refs, which could generate more dirty block groups, |
| 1324 | * so we want to keep this flushing in this loop to make sure |
| 1325 | * everything gets run. |
| 1326 | */ |
| 1327 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
| 1328 | if (ret) |
| 1329 | return ret; |
| 1330 | } |
| 1331 | |
| 1332 | if (!list_empty(&fs_info->dirty_cowonly_roots)) |
| 1333 | goto again; |
| 1334 | |
| 1335 | /* Update dev-replace pointer once everything is committed */ |
| 1336 | fs_info->dev_replace.committed_cursor_left = |
| 1337 | fs_info->dev_replace.cursor_left_last_write_of_item; |
| 1338 | |
| 1339 | return 0; |
| 1340 | } |
| 1341 | |
| 1342 | /* |
| 1343 | * If we had a pending drop we need to see if there are any others left in our |
| 1344 | * dead roots list, and if not clear our bit and wake any waiters. |
| 1345 | */ |
| 1346 | void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info) |
| 1347 | { |
| 1348 | /* |
| 1349 | * We put the drop in progress roots at the front of the list, so if the |
| 1350 | * first entry doesn't have UNFINISHED_DROP set we can wake everybody |
| 1351 | * up. |
| 1352 | */ |
| 1353 | spin_lock(&fs_info->trans_lock); |
| 1354 | if (!list_empty(&fs_info->dead_roots)) { |
| 1355 | struct btrfs_root *root = list_first_entry(&fs_info->dead_roots, |
| 1356 | struct btrfs_root, |
| 1357 | root_list); |
| 1358 | if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) { |
| 1359 | spin_unlock(&fs_info->trans_lock); |
| 1360 | return; |
| 1361 | } |
| 1362 | } |
| 1363 | spin_unlock(&fs_info->trans_lock); |
| 1364 | |
| 1365 | btrfs_wake_unfinished_drop(fs_info); |
| 1366 | } |
| 1367 | |
| 1368 | /* |
| 1369 | * dead roots are old snapshots that need to be deleted. This allocates |
| 1370 | * a dirty root struct and adds it into the list of dead roots that need to |
| 1371 | * be deleted |
| 1372 | */ |
| 1373 | void btrfs_add_dead_root(struct btrfs_root *root) |
| 1374 | { |
| 1375 | struct btrfs_fs_info *fs_info = root->fs_info; |
| 1376 | |
| 1377 | spin_lock(&fs_info->trans_lock); |
| 1378 | if (list_empty(&root->root_list)) { |
| 1379 | btrfs_grab_root(root); |
| 1380 | |
| 1381 | /* We want to process the partially complete drops first. */ |
| 1382 | if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) |
| 1383 | list_add(&root->root_list, &fs_info->dead_roots); |
| 1384 | else |
| 1385 | list_add_tail(&root->root_list, &fs_info->dead_roots); |
| 1386 | } |
| 1387 | spin_unlock(&fs_info->trans_lock); |
| 1388 | } |
| 1389 | |
| 1390 | /* |
| 1391 | * Update each subvolume root and its relocation root, if it exists, in the tree |
| 1392 | * of tree roots. Also free log roots if they exist. |
| 1393 | */ |
| 1394 | static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) |
| 1395 | { |
| 1396 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 1397 | struct btrfs_root *gang[8]; |
| 1398 | int i; |
| 1399 | int ret; |
| 1400 | |
| 1401 | /* |
| 1402 | * At this point no one can be using this transaction to modify any tree |
| 1403 | * and no one can start another transaction to modify any tree either. |
| 1404 | */ |
| 1405 | ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); |
| 1406 | |
| 1407 | spin_lock(&fs_info->fs_roots_radix_lock); |
| 1408 | while (1) { |
| 1409 | ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, |
| 1410 | (void **)gang, 0, |
| 1411 | ARRAY_SIZE(gang), |
| 1412 | BTRFS_ROOT_TRANS_TAG); |
| 1413 | if (ret == 0) |
| 1414 | break; |
| 1415 | for (i = 0; i < ret; i++) { |
| 1416 | struct btrfs_root *root = gang[i]; |
| 1417 | int ret2; |
| 1418 | |
| 1419 | /* |
| 1420 | * At this point we can neither have tasks logging inodes |
| 1421 | * from a root nor trying to commit a log tree. |
| 1422 | */ |
| 1423 | ASSERT(atomic_read(&root->log_writers) == 0); |
| 1424 | ASSERT(atomic_read(&root->log_commit[0]) == 0); |
| 1425 | ASSERT(atomic_read(&root->log_commit[1]) == 0); |
| 1426 | |
| 1427 | radix_tree_tag_clear(&fs_info->fs_roots_radix, |
| 1428 | (unsigned long)root->root_key.objectid, |
| 1429 | BTRFS_ROOT_TRANS_TAG); |
| 1430 | spin_unlock(&fs_info->fs_roots_radix_lock); |
| 1431 | |
| 1432 | btrfs_free_log(trans, root); |
| 1433 | ret2 = btrfs_update_reloc_root(trans, root); |
| 1434 | if (ret2) |
| 1435 | return ret2; |
| 1436 | |
| 1437 | /* see comments in should_cow_block() */ |
| 1438 | clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
| 1439 | smp_mb__after_atomic(); |
| 1440 | |
| 1441 | if (root->commit_root != root->node) { |
| 1442 | list_add_tail(&root->dirty_list, |
| 1443 | &trans->transaction->switch_commits); |
| 1444 | btrfs_set_root_node(&root->root_item, |
| 1445 | root->node); |
| 1446 | } |
| 1447 | |
| 1448 | ret2 = btrfs_update_root(trans, fs_info->tree_root, |
| 1449 | &root->root_key, |
| 1450 | &root->root_item); |
| 1451 | if (ret2) |
| 1452 | return ret2; |
| 1453 | spin_lock(&fs_info->fs_roots_radix_lock); |
| 1454 | btrfs_qgroup_free_meta_all_pertrans(root); |
| 1455 | } |
| 1456 | } |
| 1457 | spin_unlock(&fs_info->fs_roots_radix_lock); |
| 1458 | return 0; |
| 1459 | } |
| 1460 | |
| 1461 | /* |
| 1462 | * defrag a given btree. |
| 1463 | * Every leaf in the btree is read and defragged. |
| 1464 | */ |
| 1465 | int btrfs_defrag_root(struct btrfs_root *root) |
| 1466 | { |
| 1467 | struct btrfs_fs_info *info = root->fs_info; |
| 1468 | struct btrfs_trans_handle *trans; |
| 1469 | int ret; |
| 1470 | |
| 1471 | if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) |
| 1472 | return 0; |
| 1473 | |
| 1474 | while (1) { |
| 1475 | trans = btrfs_start_transaction(root, 0); |
| 1476 | if (IS_ERR(trans)) { |
| 1477 | ret = PTR_ERR(trans); |
| 1478 | break; |
| 1479 | } |
| 1480 | |
| 1481 | ret = btrfs_defrag_leaves(trans, root); |
| 1482 | |
| 1483 | btrfs_end_transaction(trans); |
| 1484 | btrfs_btree_balance_dirty(info); |
| 1485 | cond_resched(); |
| 1486 | |
| 1487 | if (btrfs_fs_closing(info) || ret != -EAGAIN) |
| 1488 | break; |
| 1489 | |
| 1490 | if (btrfs_defrag_cancelled(info)) { |
| 1491 | btrfs_debug(info, "defrag_root cancelled"); |
| 1492 | ret = -EAGAIN; |
| 1493 | break; |
| 1494 | } |
| 1495 | } |
| 1496 | clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); |
| 1497 | return ret; |
| 1498 | } |
| 1499 | |
| 1500 | /* |
| 1501 | * Do all special snapshot related qgroup dirty hack. |
| 1502 | * |
| 1503 | * Will do all needed qgroup inherit and dirty hack like switch commit |
| 1504 | * roots inside one transaction and write all btree into disk, to make |
| 1505 | * qgroup works. |
| 1506 | */ |
| 1507 | static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, |
| 1508 | struct btrfs_root *src, |
| 1509 | struct btrfs_root *parent, |
| 1510 | struct btrfs_qgroup_inherit *inherit, |
| 1511 | u64 dst_objectid) |
| 1512 | { |
| 1513 | struct btrfs_fs_info *fs_info = src->fs_info; |
| 1514 | int ret; |
| 1515 | |
| 1516 | /* |
| 1517 | * Save some performance in the case that qgroups are not |
| 1518 | * enabled. If this check races with the ioctl, rescan will |
| 1519 | * kick in anyway. |
| 1520 | */ |
| 1521 | if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) |
| 1522 | return 0; |
| 1523 | |
| 1524 | /* |
| 1525 | * Ensure dirty @src will be committed. Or, after coming |
| 1526 | * commit_fs_roots() and switch_commit_roots(), any dirty but not |
| 1527 | * recorded root will never be updated again, causing an outdated root |
| 1528 | * item. |
| 1529 | */ |
| 1530 | ret = record_root_in_trans(trans, src, 1); |
| 1531 | if (ret) |
| 1532 | return ret; |
| 1533 | |
| 1534 | /* |
| 1535 | * btrfs_qgroup_inherit relies on a consistent view of the usage for the |
| 1536 | * src root, so we must run the delayed refs here. |
| 1537 | * |
| 1538 | * However this isn't particularly fool proof, because there's no |
| 1539 | * synchronization keeping us from changing the tree after this point |
| 1540 | * before we do the qgroup_inherit, or even from making changes while |
| 1541 | * we're doing the qgroup_inherit. But that's a problem for the future, |
| 1542 | * for now flush the delayed refs to narrow the race window where the |
| 1543 | * qgroup counters could end up wrong. |
| 1544 | */ |
| 1545 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
| 1546 | if (ret) { |
| 1547 | btrfs_abort_transaction(trans, ret); |
| 1548 | return ret; |
| 1549 | } |
| 1550 | |
| 1551 | ret = commit_fs_roots(trans); |
| 1552 | if (ret) |
| 1553 | goto out; |
| 1554 | ret = btrfs_qgroup_account_extents(trans); |
| 1555 | if (ret < 0) |
| 1556 | goto out; |
| 1557 | |
| 1558 | /* Now qgroup are all updated, we can inherit it to new qgroups */ |
| 1559 | ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, |
| 1560 | inherit); |
| 1561 | if (ret < 0) |
| 1562 | goto out; |
| 1563 | |
| 1564 | /* |
| 1565 | * Now we do a simplified commit transaction, which will: |
| 1566 | * 1) commit all subvolume and extent tree |
| 1567 | * To ensure all subvolume and extent tree have a valid |
| 1568 | * commit_root to accounting later insert_dir_item() |
| 1569 | * 2) write all btree blocks onto disk |
| 1570 | * This is to make sure later btree modification will be cowed |
| 1571 | * Or commit_root can be populated and cause wrong qgroup numbers |
| 1572 | * In this simplified commit, we don't really care about other trees |
| 1573 | * like chunk and root tree, as they won't affect qgroup. |
| 1574 | * And we don't write super to avoid half committed status. |
| 1575 | */ |
| 1576 | ret = commit_cowonly_roots(trans); |
| 1577 | if (ret) |
| 1578 | goto out; |
| 1579 | switch_commit_roots(trans); |
| 1580 | ret = btrfs_write_and_wait_transaction(trans); |
| 1581 | if (ret) |
| 1582 | btrfs_handle_fs_error(fs_info, ret, |
| 1583 | "Error while writing out transaction for qgroup"); |
| 1584 | |
| 1585 | out: |
| 1586 | /* |
| 1587 | * Force parent root to be updated, as we recorded it before so its |
| 1588 | * last_trans == cur_transid. |
| 1589 | * Or it won't be committed again onto disk after later |
| 1590 | * insert_dir_item() |
| 1591 | */ |
| 1592 | if (!ret) |
| 1593 | ret = record_root_in_trans(trans, parent, 1); |
| 1594 | return ret; |
| 1595 | } |
| 1596 | |
| 1597 | /* |
| 1598 | * new snapshots need to be created at a very specific time in the |
| 1599 | * transaction commit. This does the actual creation. |
| 1600 | * |
| 1601 | * Note: |
| 1602 | * If the error which may affect the commitment of the current transaction |
| 1603 | * happens, we should return the error number. If the error which just affect |
| 1604 | * the creation of the pending snapshots, just return 0. |
| 1605 | */ |
| 1606 | static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, |
| 1607 | struct btrfs_pending_snapshot *pending) |
| 1608 | { |
| 1609 | |
| 1610 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 1611 | struct btrfs_key key; |
| 1612 | struct btrfs_root_item *new_root_item; |
| 1613 | struct btrfs_root *tree_root = fs_info->tree_root; |
| 1614 | struct btrfs_root *root = pending->root; |
| 1615 | struct btrfs_root *parent_root; |
| 1616 | struct btrfs_block_rsv *rsv; |
| 1617 | struct inode *parent_inode = pending->dir; |
| 1618 | struct btrfs_path *path; |
| 1619 | struct btrfs_dir_item *dir_item; |
| 1620 | struct extent_buffer *tmp; |
| 1621 | struct extent_buffer *old; |
| 1622 | struct timespec64 cur_time; |
| 1623 | int ret = 0; |
| 1624 | u64 to_reserve = 0; |
| 1625 | u64 index = 0; |
| 1626 | u64 objectid; |
| 1627 | u64 root_flags; |
| 1628 | unsigned int nofs_flags; |
| 1629 | struct fscrypt_name fname; |
| 1630 | |
| 1631 | ASSERT(pending->path); |
| 1632 | path = pending->path; |
| 1633 | |
| 1634 | ASSERT(pending->root_item); |
| 1635 | new_root_item = pending->root_item; |
| 1636 | |
| 1637 | /* |
| 1638 | * We're inside a transaction and must make sure that any potential |
| 1639 | * allocations with GFP_KERNEL in fscrypt won't recurse back to |
| 1640 | * filesystem. |
| 1641 | */ |
| 1642 | nofs_flags = memalloc_nofs_save(); |
| 1643 | pending->error = fscrypt_setup_filename(parent_inode, |
| 1644 | &pending->dentry->d_name, 0, |
| 1645 | &fname); |
| 1646 | memalloc_nofs_restore(nofs_flags); |
| 1647 | if (pending->error) |
| 1648 | goto free_pending; |
| 1649 | |
| 1650 | pending->error = btrfs_get_free_objectid(tree_root, &objectid); |
| 1651 | if (pending->error) |
| 1652 | goto free_fname; |
| 1653 | |
| 1654 | /* |
| 1655 | * Make qgroup to skip current new snapshot's qgroupid, as it is |
| 1656 | * accounted by later btrfs_qgroup_inherit(). |
| 1657 | */ |
| 1658 | btrfs_set_skip_qgroup(trans, objectid); |
| 1659 | |
| 1660 | btrfs_reloc_pre_snapshot(pending, &to_reserve); |
| 1661 | |
| 1662 | if (to_reserve > 0) { |
| 1663 | pending->error = btrfs_block_rsv_add(fs_info, |
| 1664 | &pending->block_rsv, |
| 1665 | to_reserve, |
| 1666 | BTRFS_RESERVE_NO_FLUSH); |
| 1667 | if (pending->error) |
| 1668 | goto clear_skip_qgroup; |
| 1669 | } |
| 1670 | |
| 1671 | key.objectid = objectid; |
| 1672 | key.offset = (u64)-1; |
| 1673 | key.type = BTRFS_ROOT_ITEM_KEY; |
| 1674 | |
| 1675 | rsv = trans->block_rsv; |
| 1676 | trans->block_rsv = &pending->block_rsv; |
| 1677 | trans->bytes_reserved = trans->block_rsv->reserved; |
| 1678 | trace_btrfs_space_reservation(fs_info, "transaction", |
| 1679 | trans->transid, |
| 1680 | trans->bytes_reserved, 1); |
| 1681 | parent_root = BTRFS_I(parent_inode)->root; |
| 1682 | ret = record_root_in_trans(trans, parent_root, 0); |
| 1683 | if (ret) |
| 1684 | goto fail; |
| 1685 | cur_time = current_time(parent_inode); |
| 1686 | |
| 1687 | /* |
| 1688 | * insert the directory item |
| 1689 | */ |
| 1690 | ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); |
| 1691 | BUG_ON(ret); /* -ENOMEM */ |
| 1692 | |
| 1693 | /* check if there is a file/dir which has the same name. */ |
| 1694 | dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, |
| 1695 | btrfs_ino(BTRFS_I(parent_inode)), |
| 1696 | &fname.disk_name, 0); |
| 1697 | if (dir_item != NULL && !IS_ERR(dir_item)) { |
| 1698 | pending->error = -EEXIST; |
| 1699 | goto dir_item_existed; |
| 1700 | } else if (IS_ERR(dir_item)) { |
| 1701 | ret = PTR_ERR(dir_item); |
| 1702 | btrfs_abort_transaction(trans, ret); |
| 1703 | goto fail; |
| 1704 | } |
| 1705 | btrfs_release_path(path); |
| 1706 | |
| 1707 | /* |
| 1708 | * pull in the delayed directory update |
| 1709 | * and the delayed inode item |
| 1710 | * otherwise we corrupt the FS during |
| 1711 | * snapshot |
| 1712 | */ |
| 1713 | ret = btrfs_run_delayed_items(trans); |
| 1714 | if (ret) { /* Transaction aborted */ |
| 1715 | btrfs_abort_transaction(trans, ret); |
| 1716 | goto fail; |
| 1717 | } |
| 1718 | |
| 1719 | ret = record_root_in_trans(trans, root, 0); |
| 1720 | if (ret) { |
| 1721 | btrfs_abort_transaction(trans, ret); |
| 1722 | goto fail; |
| 1723 | } |
| 1724 | btrfs_set_root_last_snapshot(&root->root_item, trans->transid); |
| 1725 | memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); |
| 1726 | btrfs_check_and_init_root_item(new_root_item); |
| 1727 | |
| 1728 | root_flags = btrfs_root_flags(new_root_item); |
| 1729 | if (pending->readonly) |
| 1730 | root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; |
| 1731 | else |
| 1732 | root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; |
| 1733 | btrfs_set_root_flags(new_root_item, root_flags); |
| 1734 | |
| 1735 | btrfs_set_root_generation_v2(new_root_item, |
| 1736 | trans->transid); |
| 1737 | generate_random_guid(new_root_item->uuid); |
| 1738 | memcpy(new_root_item->parent_uuid, root->root_item.uuid, |
| 1739 | BTRFS_UUID_SIZE); |
| 1740 | if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { |
| 1741 | memset(new_root_item->received_uuid, 0, |
| 1742 | sizeof(new_root_item->received_uuid)); |
| 1743 | memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); |
| 1744 | memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); |
| 1745 | btrfs_set_root_stransid(new_root_item, 0); |
| 1746 | btrfs_set_root_rtransid(new_root_item, 0); |
| 1747 | } |
| 1748 | btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); |
| 1749 | btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); |
| 1750 | btrfs_set_root_otransid(new_root_item, trans->transid); |
| 1751 | |
| 1752 | old = btrfs_lock_root_node(root); |
| 1753 | ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, |
| 1754 | BTRFS_NESTING_COW); |
| 1755 | if (ret) { |
| 1756 | btrfs_tree_unlock(old); |
| 1757 | free_extent_buffer(old); |
| 1758 | btrfs_abort_transaction(trans, ret); |
| 1759 | goto fail; |
| 1760 | } |
| 1761 | |
| 1762 | ret = btrfs_copy_root(trans, root, old, &tmp, objectid); |
| 1763 | /* clean up in any case */ |
| 1764 | btrfs_tree_unlock(old); |
| 1765 | free_extent_buffer(old); |
| 1766 | if (ret) { |
| 1767 | btrfs_abort_transaction(trans, ret); |
| 1768 | goto fail; |
| 1769 | } |
| 1770 | /* see comments in should_cow_block() */ |
| 1771 | set_bit(BTRFS_ROOT_FORCE_COW, &root->state); |
| 1772 | smp_wmb(); |
| 1773 | |
| 1774 | btrfs_set_root_node(new_root_item, tmp); |
| 1775 | /* record when the snapshot was created in key.offset */ |
| 1776 | key.offset = trans->transid; |
| 1777 | ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); |
| 1778 | btrfs_tree_unlock(tmp); |
| 1779 | free_extent_buffer(tmp); |
| 1780 | if (ret) { |
| 1781 | btrfs_abort_transaction(trans, ret); |
| 1782 | goto fail; |
| 1783 | } |
| 1784 | |
| 1785 | /* |
| 1786 | * insert root back/forward references |
| 1787 | */ |
| 1788 | ret = btrfs_add_root_ref(trans, objectid, |
| 1789 | parent_root->root_key.objectid, |
| 1790 | btrfs_ino(BTRFS_I(parent_inode)), index, |
| 1791 | &fname.disk_name); |
| 1792 | if (ret) { |
| 1793 | btrfs_abort_transaction(trans, ret); |
| 1794 | goto fail; |
| 1795 | } |
| 1796 | |
| 1797 | key.offset = (u64)-1; |
| 1798 | pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); |
| 1799 | if (IS_ERR(pending->snap)) { |
| 1800 | ret = PTR_ERR(pending->snap); |
| 1801 | pending->snap = NULL; |
| 1802 | btrfs_abort_transaction(trans, ret); |
| 1803 | goto fail; |
| 1804 | } |
| 1805 | |
| 1806 | ret = btrfs_reloc_post_snapshot(trans, pending); |
| 1807 | if (ret) { |
| 1808 | btrfs_abort_transaction(trans, ret); |
| 1809 | goto fail; |
| 1810 | } |
| 1811 | |
| 1812 | /* |
| 1813 | * Do special qgroup accounting for snapshot, as we do some qgroup |
| 1814 | * snapshot hack to do fast snapshot. |
| 1815 | * To co-operate with that hack, we do hack again. |
| 1816 | * Or snapshot will be greatly slowed down by a subtree qgroup rescan |
| 1817 | */ |
| 1818 | ret = qgroup_account_snapshot(trans, root, parent_root, |
| 1819 | pending->inherit, objectid); |
| 1820 | if (ret < 0) |
| 1821 | goto fail; |
| 1822 | |
| 1823 | ret = btrfs_insert_dir_item(trans, &fname.disk_name, |
| 1824 | BTRFS_I(parent_inode), &key, BTRFS_FT_DIR, |
| 1825 | index); |
| 1826 | /* We have check then name at the beginning, so it is impossible. */ |
| 1827 | BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); |
| 1828 | if (ret) { |
| 1829 | btrfs_abort_transaction(trans, ret); |
| 1830 | goto fail; |
| 1831 | } |
| 1832 | |
| 1833 | btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + |
| 1834 | fname.disk_name.len * 2); |
| 1835 | parent_inode->i_mtime = current_time(parent_inode); |
| 1836 | parent_inode->i_ctime = parent_inode->i_mtime; |
| 1837 | ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); |
| 1838 | if (ret) { |
| 1839 | btrfs_abort_transaction(trans, ret); |
| 1840 | goto fail; |
| 1841 | } |
| 1842 | ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, |
| 1843 | BTRFS_UUID_KEY_SUBVOL, |
| 1844 | objectid); |
| 1845 | if (ret) { |
| 1846 | btrfs_abort_transaction(trans, ret); |
| 1847 | goto fail; |
| 1848 | } |
| 1849 | if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { |
| 1850 | ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, |
| 1851 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, |
| 1852 | objectid); |
| 1853 | if (ret && ret != -EEXIST) { |
| 1854 | btrfs_abort_transaction(trans, ret); |
| 1855 | goto fail; |
| 1856 | } |
| 1857 | } |
| 1858 | |
| 1859 | fail: |
| 1860 | pending->error = ret; |
| 1861 | dir_item_existed: |
| 1862 | trans->block_rsv = rsv; |
| 1863 | trans->bytes_reserved = 0; |
| 1864 | clear_skip_qgroup: |
| 1865 | btrfs_clear_skip_qgroup(trans); |
| 1866 | free_fname: |
| 1867 | fscrypt_free_filename(&fname); |
| 1868 | free_pending: |
| 1869 | kfree(new_root_item); |
| 1870 | pending->root_item = NULL; |
| 1871 | btrfs_free_path(path); |
| 1872 | pending->path = NULL; |
| 1873 | |
| 1874 | return ret; |
| 1875 | } |
| 1876 | |
| 1877 | /* |
| 1878 | * create all the snapshots we've scheduled for creation |
| 1879 | */ |
| 1880 | static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) |
| 1881 | { |
| 1882 | struct btrfs_pending_snapshot *pending, *next; |
| 1883 | struct list_head *head = &trans->transaction->pending_snapshots; |
| 1884 | int ret = 0; |
| 1885 | |
| 1886 | list_for_each_entry_safe(pending, next, head, list) { |
| 1887 | list_del(&pending->list); |
| 1888 | ret = create_pending_snapshot(trans, pending); |
| 1889 | if (ret) |
| 1890 | break; |
| 1891 | } |
| 1892 | return ret; |
| 1893 | } |
| 1894 | |
| 1895 | static void update_super_roots(struct btrfs_fs_info *fs_info) |
| 1896 | { |
| 1897 | struct btrfs_root_item *root_item; |
| 1898 | struct btrfs_super_block *super; |
| 1899 | |
| 1900 | super = fs_info->super_copy; |
| 1901 | |
| 1902 | root_item = &fs_info->chunk_root->root_item; |
| 1903 | super->chunk_root = root_item->bytenr; |
| 1904 | super->chunk_root_generation = root_item->generation; |
| 1905 | super->chunk_root_level = root_item->level; |
| 1906 | |
| 1907 | root_item = &fs_info->tree_root->root_item; |
| 1908 | super->root = root_item->bytenr; |
| 1909 | super->generation = root_item->generation; |
| 1910 | super->root_level = root_item->level; |
| 1911 | if (btrfs_test_opt(fs_info, SPACE_CACHE)) |
| 1912 | super->cache_generation = root_item->generation; |
| 1913 | else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) |
| 1914 | super->cache_generation = 0; |
| 1915 | if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) |
| 1916 | super->uuid_tree_generation = root_item->generation; |
| 1917 | } |
| 1918 | |
| 1919 | int btrfs_transaction_in_commit(struct btrfs_fs_info *info) |
| 1920 | { |
| 1921 | struct btrfs_transaction *trans; |
| 1922 | int ret = 0; |
| 1923 | |
| 1924 | spin_lock(&info->trans_lock); |
| 1925 | trans = info->running_transaction; |
| 1926 | if (trans) |
| 1927 | ret = (trans->state >= TRANS_STATE_COMMIT_START); |
| 1928 | spin_unlock(&info->trans_lock); |
| 1929 | return ret; |
| 1930 | } |
| 1931 | |
| 1932 | int btrfs_transaction_blocked(struct btrfs_fs_info *info) |
| 1933 | { |
| 1934 | struct btrfs_transaction *trans; |
| 1935 | int ret = 0; |
| 1936 | |
| 1937 | spin_lock(&info->trans_lock); |
| 1938 | trans = info->running_transaction; |
| 1939 | if (trans) |
| 1940 | ret = is_transaction_blocked(trans); |
| 1941 | spin_unlock(&info->trans_lock); |
| 1942 | return ret; |
| 1943 | } |
| 1944 | |
| 1945 | void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans) |
| 1946 | { |
| 1947 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 1948 | struct btrfs_transaction *cur_trans; |
| 1949 | |
| 1950 | /* Kick the transaction kthread. */ |
| 1951 | set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); |
| 1952 | wake_up_process(fs_info->transaction_kthread); |
| 1953 | |
| 1954 | /* take transaction reference */ |
| 1955 | cur_trans = trans->transaction; |
| 1956 | refcount_inc(&cur_trans->use_count); |
| 1957 | |
| 1958 | btrfs_end_transaction(trans); |
| 1959 | |
| 1960 | /* |
| 1961 | * Wait for the current transaction commit to start and block |
| 1962 | * subsequent transaction joins |
| 1963 | */ |
| 1964 | btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); |
| 1965 | wait_event(fs_info->transaction_blocked_wait, |
| 1966 | cur_trans->state >= TRANS_STATE_COMMIT_START || |
| 1967 | TRANS_ABORTED(cur_trans)); |
| 1968 | btrfs_put_transaction(cur_trans); |
| 1969 | } |
| 1970 | |
| 1971 | static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) |
| 1972 | { |
| 1973 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 1974 | struct btrfs_transaction *cur_trans = trans->transaction; |
| 1975 | |
| 1976 | WARN_ON(refcount_read(&trans->use_count) > 1); |
| 1977 | |
| 1978 | btrfs_abort_transaction(trans, err); |
| 1979 | |
| 1980 | spin_lock(&fs_info->trans_lock); |
| 1981 | |
| 1982 | /* |
| 1983 | * If the transaction is removed from the list, it means this |
| 1984 | * transaction has been committed successfully, so it is impossible |
| 1985 | * to call the cleanup function. |
| 1986 | */ |
| 1987 | BUG_ON(list_empty(&cur_trans->list)); |
| 1988 | |
| 1989 | if (cur_trans == fs_info->running_transaction) { |
| 1990 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
| 1991 | spin_unlock(&fs_info->trans_lock); |
| 1992 | |
| 1993 | /* |
| 1994 | * The thread has already released the lockdep map as reader |
| 1995 | * already in btrfs_commit_transaction(). |
| 1996 | */ |
| 1997 | btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); |
| 1998 | wait_event(cur_trans->writer_wait, |
| 1999 | atomic_read(&cur_trans->num_writers) == 1); |
| 2000 | |
| 2001 | spin_lock(&fs_info->trans_lock); |
| 2002 | } |
| 2003 | |
| 2004 | /* |
| 2005 | * Now that we know no one else is still using the transaction we can |
| 2006 | * remove the transaction from the list of transactions. This avoids |
| 2007 | * the transaction kthread from cleaning up the transaction while some |
| 2008 | * other task is still using it, which could result in a use-after-free |
| 2009 | * on things like log trees, as it forces the transaction kthread to |
| 2010 | * wait for this transaction to be cleaned up by us. |
| 2011 | */ |
| 2012 | list_del_init(&cur_trans->list); |
| 2013 | |
| 2014 | spin_unlock(&fs_info->trans_lock); |
| 2015 | |
| 2016 | btrfs_cleanup_one_transaction(trans->transaction, fs_info); |
| 2017 | |
| 2018 | spin_lock(&fs_info->trans_lock); |
| 2019 | if (cur_trans == fs_info->running_transaction) |
| 2020 | fs_info->running_transaction = NULL; |
| 2021 | spin_unlock(&fs_info->trans_lock); |
| 2022 | |
| 2023 | if (trans->type & __TRANS_FREEZABLE) |
| 2024 | sb_end_intwrite(fs_info->sb); |
| 2025 | btrfs_put_transaction(cur_trans); |
| 2026 | btrfs_put_transaction(cur_trans); |
| 2027 | |
| 2028 | trace_btrfs_transaction_commit(fs_info); |
| 2029 | |
| 2030 | if (current->journal_info == trans) |
| 2031 | current->journal_info = NULL; |
| 2032 | |
| 2033 | /* |
| 2034 | * If relocation is running, we can't cancel scrub because that will |
| 2035 | * result in a deadlock. Before relocating a block group, relocation |
| 2036 | * pauses scrub, then starts and commits a transaction before unpausing |
| 2037 | * scrub. If the transaction commit is being done by the relocation |
| 2038 | * task or triggered by another task and the relocation task is waiting |
| 2039 | * for the commit, and we end up here due to an error in the commit |
| 2040 | * path, then calling btrfs_scrub_cancel() will deadlock, as we are |
| 2041 | * asking for scrub to stop while having it asked to be paused higher |
| 2042 | * above in relocation code. |
| 2043 | */ |
| 2044 | if (!test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) |
| 2045 | btrfs_scrub_cancel(fs_info); |
| 2046 | |
| 2047 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
| 2048 | } |
| 2049 | |
| 2050 | /* |
| 2051 | * Release reserved delayed ref space of all pending block groups of the |
| 2052 | * transaction and remove them from the list |
| 2053 | */ |
| 2054 | static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) |
| 2055 | { |
| 2056 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 2057 | struct btrfs_block_group *block_group, *tmp; |
| 2058 | |
| 2059 | list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { |
| 2060 | btrfs_delayed_refs_rsv_release(fs_info, 1); |
| 2061 | list_del_init(&block_group->bg_list); |
| 2062 | } |
| 2063 | } |
| 2064 | |
| 2065 | static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) |
| 2066 | { |
| 2067 | /* |
| 2068 | * We use try_to_writeback_inodes_sb() here because if we used |
| 2069 | * btrfs_start_delalloc_roots we would deadlock with fs freeze. |
| 2070 | * Currently are holding the fs freeze lock, if we do an async flush |
| 2071 | * we'll do btrfs_join_transaction() and deadlock because we need to |
| 2072 | * wait for the fs freeze lock. Using the direct flushing we benefit |
| 2073 | * from already being in a transaction and our join_transaction doesn't |
| 2074 | * have to re-take the fs freeze lock. |
| 2075 | * |
| 2076 | * Note that try_to_writeback_inodes_sb() will only trigger writeback |
| 2077 | * if it can read lock sb->s_umount. It will always be able to lock it, |
| 2078 | * except when the filesystem is being unmounted or being frozen, but in |
| 2079 | * those cases sync_filesystem() is called, which results in calling |
| 2080 | * writeback_inodes_sb() while holding a write lock on sb->s_umount. |
| 2081 | * Note that we don't call writeback_inodes_sb() directly, because it |
| 2082 | * will emit a warning if sb->s_umount is not locked. |
| 2083 | */ |
| 2084 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
| 2085 | try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); |
| 2086 | return 0; |
| 2087 | } |
| 2088 | |
| 2089 | static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) |
| 2090 | { |
| 2091 | if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) |
| 2092 | btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); |
| 2093 | } |
| 2094 | |
| 2095 | /* |
| 2096 | * Add a pending snapshot associated with the given transaction handle to the |
| 2097 | * respective handle. This must be called after the transaction commit started |
| 2098 | * and while holding fs_info->trans_lock. |
| 2099 | * This serves to guarantee a caller of btrfs_commit_transaction() that it can |
| 2100 | * safely free the pending snapshot pointer in case btrfs_commit_transaction() |
| 2101 | * returns an error. |
| 2102 | */ |
| 2103 | static void add_pending_snapshot(struct btrfs_trans_handle *trans) |
| 2104 | { |
| 2105 | struct btrfs_transaction *cur_trans = trans->transaction; |
| 2106 | |
| 2107 | if (!trans->pending_snapshot) |
| 2108 | return; |
| 2109 | |
| 2110 | lockdep_assert_held(&trans->fs_info->trans_lock); |
| 2111 | ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_START); |
| 2112 | |
| 2113 | list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots); |
| 2114 | } |
| 2115 | |
| 2116 | static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval) |
| 2117 | { |
| 2118 | fs_info->commit_stats.commit_count++; |
| 2119 | fs_info->commit_stats.last_commit_dur = interval; |
| 2120 | fs_info->commit_stats.max_commit_dur = |
| 2121 | max_t(u64, fs_info->commit_stats.max_commit_dur, interval); |
| 2122 | fs_info->commit_stats.total_commit_dur += interval; |
| 2123 | } |
| 2124 | |
| 2125 | int btrfs_commit_transaction(struct btrfs_trans_handle *trans) |
| 2126 | { |
| 2127 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 2128 | struct btrfs_transaction *cur_trans = trans->transaction; |
| 2129 | struct btrfs_transaction *prev_trans = NULL; |
| 2130 | int ret; |
| 2131 | ktime_t start_time; |
| 2132 | ktime_t interval; |
| 2133 | |
| 2134 | ASSERT(refcount_read(&trans->use_count) == 1); |
| 2135 | btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); |
| 2136 | |
| 2137 | clear_bit(BTRFS_FS_NEED_TRANS_COMMIT, &fs_info->flags); |
| 2138 | |
| 2139 | /* Stop the commit early if ->aborted is set */ |
| 2140 | if (TRANS_ABORTED(cur_trans)) { |
| 2141 | ret = cur_trans->aborted; |
| 2142 | goto lockdep_trans_commit_start_release; |
| 2143 | } |
| 2144 | |
| 2145 | btrfs_trans_release_metadata(trans); |
| 2146 | trans->block_rsv = NULL; |
| 2147 | |
| 2148 | /* |
| 2149 | * We only want one transaction commit doing the flushing so we do not |
| 2150 | * waste a bunch of time on lock contention on the extent root node. |
| 2151 | */ |
| 2152 | if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, |
| 2153 | &cur_trans->delayed_refs.flags)) { |
| 2154 | /* |
| 2155 | * Make a pass through all the delayed refs we have so far. |
| 2156 | * Any running threads may add more while we are here. |
| 2157 | */ |
| 2158 | ret = btrfs_run_delayed_refs(trans, 0); |
| 2159 | if (ret) |
| 2160 | goto lockdep_trans_commit_start_release; |
| 2161 | } |
| 2162 | |
| 2163 | btrfs_create_pending_block_groups(trans); |
| 2164 | |
| 2165 | if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { |
| 2166 | int run_it = 0; |
| 2167 | |
| 2168 | /* this mutex is also taken before trying to set |
| 2169 | * block groups readonly. We need to make sure |
| 2170 | * that nobody has set a block group readonly |
| 2171 | * after a extents from that block group have been |
| 2172 | * allocated for cache files. btrfs_set_block_group_ro |
| 2173 | * will wait for the transaction to commit if it |
| 2174 | * finds BTRFS_TRANS_DIRTY_BG_RUN set. |
| 2175 | * |
| 2176 | * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure |
| 2177 | * only one process starts all the block group IO. It wouldn't |
| 2178 | * hurt to have more than one go through, but there's no |
| 2179 | * real advantage to it either. |
| 2180 | */ |
| 2181 | mutex_lock(&fs_info->ro_block_group_mutex); |
| 2182 | if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, |
| 2183 | &cur_trans->flags)) |
| 2184 | run_it = 1; |
| 2185 | mutex_unlock(&fs_info->ro_block_group_mutex); |
| 2186 | |
| 2187 | if (run_it) { |
| 2188 | ret = btrfs_start_dirty_block_groups(trans); |
| 2189 | if (ret) |
| 2190 | goto lockdep_trans_commit_start_release; |
| 2191 | } |
| 2192 | } |
| 2193 | |
| 2194 | spin_lock(&fs_info->trans_lock); |
| 2195 | if (cur_trans->state >= TRANS_STATE_COMMIT_START) { |
| 2196 | enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; |
| 2197 | |
| 2198 | add_pending_snapshot(trans); |
| 2199 | |
| 2200 | spin_unlock(&fs_info->trans_lock); |
| 2201 | refcount_inc(&cur_trans->use_count); |
| 2202 | |
| 2203 | if (trans->in_fsync) |
| 2204 | want_state = TRANS_STATE_SUPER_COMMITTED; |
| 2205 | |
| 2206 | btrfs_trans_state_lockdep_release(fs_info, |
| 2207 | BTRFS_LOCKDEP_TRANS_COMMIT_START); |
| 2208 | ret = btrfs_end_transaction(trans); |
| 2209 | wait_for_commit(cur_trans, want_state); |
| 2210 | |
| 2211 | if (TRANS_ABORTED(cur_trans)) |
| 2212 | ret = cur_trans->aborted; |
| 2213 | |
| 2214 | btrfs_put_transaction(cur_trans); |
| 2215 | |
| 2216 | return ret; |
| 2217 | } |
| 2218 | |
| 2219 | cur_trans->state = TRANS_STATE_COMMIT_START; |
| 2220 | wake_up(&fs_info->transaction_blocked_wait); |
| 2221 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); |
| 2222 | |
| 2223 | if (cur_trans->list.prev != &fs_info->trans_list) { |
| 2224 | enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; |
| 2225 | |
| 2226 | if (trans->in_fsync) |
| 2227 | want_state = TRANS_STATE_SUPER_COMMITTED; |
| 2228 | |
| 2229 | prev_trans = list_entry(cur_trans->list.prev, |
| 2230 | struct btrfs_transaction, list); |
| 2231 | if (prev_trans->state < want_state) { |
| 2232 | refcount_inc(&prev_trans->use_count); |
| 2233 | spin_unlock(&fs_info->trans_lock); |
| 2234 | |
| 2235 | wait_for_commit(prev_trans, want_state); |
| 2236 | |
| 2237 | ret = READ_ONCE(prev_trans->aborted); |
| 2238 | |
| 2239 | btrfs_put_transaction(prev_trans); |
| 2240 | if (ret) |
| 2241 | goto lockdep_release; |
| 2242 | } else { |
| 2243 | spin_unlock(&fs_info->trans_lock); |
| 2244 | } |
| 2245 | } else { |
| 2246 | spin_unlock(&fs_info->trans_lock); |
| 2247 | /* |
| 2248 | * The previous transaction was aborted and was already removed |
| 2249 | * from the list of transactions at fs_info->trans_list. So we |
| 2250 | * abort to prevent writing a new superblock that reflects a |
| 2251 | * corrupt state (pointing to trees with unwritten nodes/leafs). |
| 2252 | */ |
| 2253 | if (BTRFS_FS_ERROR(fs_info)) { |
| 2254 | ret = -EROFS; |
| 2255 | goto lockdep_release; |
| 2256 | } |
| 2257 | } |
| 2258 | |
| 2259 | /* |
| 2260 | * Get the time spent on the work done by the commit thread and not |
| 2261 | * the time spent waiting on a previous commit |
| 2262 | */ |
| 2263 | start_time = ktime_get_ns(); |
| 2264 | |
| 2265 | extwriter_counter_dec(cur_trans, trans->type); |
| 2266 | |
| 2267 | ret = btrfs_start_delalloc_flush(fs_info); |
| 2268 | if (ret) |
| 2269 | goto lockdep_release; |
| 2270 | |
| 2271 | ret = btrfs_run_delayed_items(trans); |
| 2272 | if (ret) |
| 2273 | goto lockdep_release; |
| 2274 | |
| 2275 | /* |
| 2276 | * The thread has started/joined the transaction thus it holds the |
| 2277 | * lockdep map as a reader. It has to release it before acquiring the |
| 2278 | * lockdep map as a writer. |
| 2279 | */ |
| 2280 | btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); |
| 2281 | btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters); |
| 2282 | wait_event(cur_trans->writer_wait, |
| 2283 | extwriter_counter_read(cur_trans) == 0); |
| 2284 | |
| 2285 | /* some pending stuffs might be added after the previous flush. */ |
| 2286 | ret = btrfs_run_delayed_items(trans); |
| 2287 | if (ret) { |
| 2288 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); |
| 2289 | goto cleanup_transaction; |
| 2290 | } |
| 2291 | |
| 2292 | btrfs_wait_delalloc_flush(fs_info); |
| 2293 | |
| 2294 | /* |
| 2295 | * Wait for all ordered extents started by a fast fsync that joined this |
| 2296 | * transaction. Otherwise if this transaction commits before the ordered |
| 2297 | * extents complete we lose logged data after a power failure. |
| 2298 | */ |
| 2299 | btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered); |
| 2300 | wait_event(cur_trans->pending_wait, |
| 2301 | atomic_read(&cur_trans->pending_ordered) == 0); |
| 2302 | |
| 2303 | btrfs_scrub_pause(fs_info); |
| 2304 | /* |
| 2305 | * Ok now we need to make sure to block out any other joins while we |
| 2306 | * commit the transaction. We could have started a join before setting |
| 2307 | * COMMIT_DOING so make sure to wait for num_writers to == 1 again. |
| 2308 | */ |
| 2309 | spin_lock(&fs_info->trans_lock); |
| 2310 | add_pending_snapshot(trans); |
| 2311 | cur_trans->state = TRANS_STATE_COMMIT_DOING; |
| 2312 | spin_unlock(&fs_info->trans_lock); |
| 2313 | |
| 2314 | /* |
| 2315 | * The thread has started/joined the transaction thus it holds the |
| 2316 | * lockdep map as a reader. It has to release it before acquiring the |
| 2317 | * lockdep map as a writer. |
| 2318 | */ |
| 2319 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); |
| 2320 | btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); |
| 2321 | wait_event(cur_trans->writer_wait, |
| 2322 | atomic_read(&cur_trans->num_writers) == 1); |
| 2323 | |
| 2324 | /* |
| 2325 | * Make lockdep happy by acquiring the state locks after |
| 2326 | * btrfs_trans_num_writers is released. If we acquired the state locks |
| 2327 | * before releasing the btrfs_trans_num_writers lock then lockdep would |
| 2328 | * complain because we did not follow the reverse order unlocking rule. |
| 2329 | */ |
| 2330 | btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); |
| 2331 | btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); |
| 2332 | btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); |
| 2333 | |
| 2334 | /* |
| 2335 | * We've started the commit, clear the flag in case we were triggered to |
| 2336 | * do an async commit but somebody else started before the transaction |
| 2337 | * kthread could do the work. |
| 2338 | */ |
| 2339 | clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); |
| 2340 | |
| 2341 | if (TRANS_ABORTED(cur_trans)) { |
| 2342 | ret = cur_trans->aborted; |
| 2343 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); |
| 2344 | goto scrub_continue; |
| 2345 | } |
| 2346 | /* |
| 2347 | * the reloc mutex makes sure that we stop |
| 2348 | * the balancing code from coming in and moving |
| 2349 | * extents around in the middle of the commit |
| 2350 | */ |
| 2351 | mutex_lock(&fs_info->reloc_mutex); |
| 2352 | |
| 2353 | /* |
| 2354 | * We needn't worry about the delayed items because we will |
| 2355 | * deal with them in create_pending_snapshot(), which is the |
| 2356 | * core function of the snapshot creation. |
| 2357 | */ |
| 2358 | ret = create_pending_snapshots(trans); |
| 2359 | if (ret) |
| 2360 | goto unlock_reloc; |
| 2361 | |
| 2362 | /* |
| 2363 | * We insert the dir indexes of the snapshots and update the inode |
| 2364 | * of the snapshots' parents after the snapshot creation, so there |
| 2365 | * are some delayed items which are not dealt with. Now deal with |
| 2366 | * them. |
| 2367 | * |
| 2368 | * We needn't worry that this operation will corrupt the snapshots, |
| 2369 | * because all the tree which are snapshoted will be forced to COW |
| 2370 | * the nodes and leaves. |
| 2371 | */ |
| 2372 | ret = btrfs_run_delayed_items(trans); |
| 2373 | if (ret) |
| 2374 | goto unlock_reloc; |
| 2375 | |
| 2376 | ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); |
| 2377 | if (ret) |
| 2378 | goto unlock_reloc; |
| 2379 | |
| 2380 | /* |
| 2381 | * make sure none of the code above managed to slip in a |
| 2382 | * delayed item |
| 2383 | */ |
| 2384 | btrfs_assert_delayed_root_empty(fs_info); |
| 2385 | |
| 2386 | WARN_ON(cur_trans != trans->transaction); |
| 2387 | |
| 2388 | ret = commit_fs_roots(trans); |
| 2389 | if (ret) |
| 2390 | goto unlock_reloc; |
| 2391 | |
| 2392 | /* commit_fs_roots gets rid of all the tree log roots, it is now |
| 2393 | * safe to free the root of tree log roots |
| 2394 | */ |
| 2395 | btrfs_free_log_root_tree(trans, fs_info); |
| 2396 | |
| 2397 | /* |
| 2398 | * Since fs roots are all committed, we can get a quite accurate |
| 2399 | * new_roots. So let's do quota accounting. |
| 2400 | */ |
| 2401 | ret = btrfs_qgroup_account_extents(trans); |
| 2402 | if (ret < 0) |
| 2403 | goto unlock_reloc; |
| 2404 | |
| 2405 | ret = commit_cowonly_roots(trans); |
| 2406 | if (ret) |
| 2407 | goto unlock_reloc; |
| 2408 | |
| 2409 | /* |
| 2410 | * The tasks which save the space cache and inode cache may also |
| 2411 | * update ->aborted, check it. |
| 2412 | */ |
| 2413 | if (TRANS_ABORTED(cur_trans)) { |
| 2414 | ret = cur_trans->aborted; |
| 2415 | goto unlock_reloc; |
| 2416 | } |
| 2417 | |
| 2418 | cur_trans = fs_info->running_transaction; |
| 2419 | |
| 2420 | btrfs_set_root_node(&fs_info->tree_root->root_item, |
| 2421 | fs_info->tree_root->node); |
| 2422 | list_add_tail(&fs_info->tree_root->dirty_list, |
| 2423 | &cur_trans->switch_commits); |
| 2424 | |
| 2425 | btrfs_set_root_node(&fs_info->chunk_root->root_item, |
| 2426 | fs_info->chunk_root->node); |
| 2427 | list_add_tail(&fs_info->chunk_root->dirty_list, |
| 2428 | &cur_trans->switch_commits); |
| 2429 | |
| 2430 | if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { |
| 2431 | btrfs_set_root_node(&fs_info->block_group_root->root_item, |
| 2432 | fs_info->block_group_root->node); |
| 2433 | list_add_tail(&fs_info->block_group_root->dirty_list, |
| 2434 | &cur_trans->switch_commits); |
| 2435 | } |
| 2436 | |
| 2437 | switch_commit_roots(trans); |
| 2438 | |
| 2439 | ASSERT(list_empty(&cur_trans->dirty_bgs)); |
| 2440 | ASSERT(list_empty(&cur_trans->io_bgs)); |
| 2441 | update_super_roots(fs_info); |
| 2442 | |
| 2443 | btrfs_set_super_log_root(fs_info->super_copy, 0); |
| 2444 | btrfs_set_super_log_root_level(fs_info->super_copy, 0); |
| 2445 | memcpy(fs_info->super_for_commit, fs_info->super_copy, |
| 2446 | sizeof(*fs_info->super_copy)); |
| 2447 | |
| 2448 | btrfs_commit_device_sizes(cur_trans); |
| 2449 | |
| 2450 | clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); |
| 2451 | clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); |
| 2452 | |
| 2453 | btrfs_trans_release_chunk_metadata(trans); |
| 2454 | |
| 2455 | /* |
| 2456 | * Before changing the transaction state to TRANS_STATE_UNBLOCKED and |
| 2457 | * setting fs_info->running_transaction to NULL, lock tree_log_mutex to |
| 2458 | * make sure that before we commit our superblock, no other task can |
| 2459 | * start a new transaction and commit a log tree before we commit our |
| 2460 | * superblock. Anyone trying to commit a log tree locks this mutex before |
| 2461 | * writing its superblock. |
| 2462 | */ |
| 2463 | mutex_lock(&fs_info->tree_log_mutex); |
| 2464 | |
| 2465 | spin_lock(&fs_info->trans_lock); |
| 2466 | cur_trans->state = TRANS_STATE_UNBLOCKED; |
| 2467 | fs_info->running_transaction = NULL; |
| 2468 | spin_unlock(&fs_info->trans_lock); |
| 2469 | mutex_unlock(&fs_info->reloc_mutex); |
| 2470 | |
| 2471 | wake_up(&fs_info->transaction_wait); |
| 2472 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); |
| 2473 | |
| 2474 | /* If we have features changed, wake up the cleaner to update sysfs. */ |
| 2475 | if (test_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags) && |
| 2476 | fs_info->cleaner_kthread) |
| 2477 | wake_up_process(fs_info->cleaner_kthread); |
| 2478 | |
| 2479 | ret = btrfs_write_and_wait_transaction(trans); |
| 2480 | if (ret) { |
| 2481 | btrfs_handle_fs_error(fs_info, ret, |
| 2482 | "Error while writing out transaction"); |
| 2483 | mutex_unlock(&fs_info->tree_log_mutex); |
| 2484 | goto scrub_continue; |
| 2485 | } |
| 2486 | |
| 2487 | /* |
| 2488 | * At this point, we should have written all the tree blocks allocated |
| 2489 | * in this transaction. So it's now safe to free the redirtyied extent |
| 2490 | * buffers. |
| 2491 | */ |
| 2492 | btrfs_free_redirty_list(cur_trans); |
| 2493 | |
| 2494 | ret = write_all_supers(fs_info, 0); |
| 2495 | /* |
| 2496 | * the super is written, we can safely allow the tree-loggers |
| 2497 | * to go about their business |
| 2498 | */ |
| 2499 | mutex_unlock(&fs_info->tree_log_mutex); |
| 2500 | if (ret) |
| 2501 | goto scrub_continue; |
| 2502 | |
| 2503 | /* |
| 2504 | * We needn't acquire the lock here because there is no other task |
| 2505 | * which can change it. |
| 2506 | */ |
| 2507 | cur_trans->state = TRANS_STATE_SUPER_COMMITTED; |
| 2508 | wake_up(&cur_trans->commit_wait); |
| 2509 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); |
| 2510 | |
| 2511 | btrfs_finish_extent_commit(trans); |
| 2512 | |
| 2513 | if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) |
| 2514 | btrfs_clear_space_info_full(fs_info); |
| 2515 | |
| 2516 | fs_info->last_trans_committed = cur_trans->transid; |
| 2517 | /* |
| 2518 | * We needn't acquire the lock here because there is no other task |
| 2519 | * which can change it. |
| 2520 | */ |
| 2521 | cur_trans->state = TRANS_STATE_COMPLETED; |
| 2522 | wake_up(&cur_trans->commit_wait); |
| 2523 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); |
| 2524 | |
| 2525 | spin_lock(&fs_info->trans_lock); |
| 2526 | list_del_init(&cur_trans->list); |
| 2527 | spin_unlock(&fs_info->trans_lock); |
| 2528 | |
| 2529 | btrfs_put_transaction(cur_trans); |
| 2530 | btrfs_put_transaction(cur_trans); |
| 2531 | |
| 2532 | if (trans->type & __TRANS_FREEZABLE) |
| 2533 | sb_end_intwrite(fs_info->sb); |
| 2534 | |
| 2535 | trace_btrfs_transaction_commit(fs_info); |
| 2536 | |
| 2537 | interval = ktime_get_ns() - start_time; |
| 2538 | |
| 2539 | btrfs_scrub_continue(fs_info); |
| 2540 | |
| 2541 | if (current->journal_info == trans) |
| 2542 | current->journal_info = NULL; |
| 2543 | |
| 2544 | kmem_cache_free(btrfs_trans_handle_cachep, trans); |
| 2545 | |
| 2546 | update_commit_stats(fs_info, interval); |
| 2547 | |
| 2548 | return ret; |
| 2549 | |
| 2550 | unlock_reloc: |
| 2551 | mutex_unlock(&fs_info->reloc_mutex); |
| 2552 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); |
| 2553 | scrub_continue: |
| 2554 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); |
| 2555 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); |
| 2556 | btrfs_scrub_continue(fs_info); |
| 2557 | cleanup_transaction: |
| 2558 | btrfs_trans_release_metadata(trans); |
| 2559 | btrfs_cleanup_pending_block_groups(trans); |
| 2560 | btrfs_trans_release_chunk_metadata(trans); |
| 2561 | trans->block_rsv = NULL; |
| 2562 | btrfs_warn(fs_info, "Skipping commit of aborted transaction."); |
| 2563 | if (current->journal_info == trans) |
| 2564 | current->journal_info = NULL; |
| 2565 | cleanup_transaction(trans, ret); |
| 2566 | |
| 2567 | return ret; |
| 2568 | |
| 2569 | lockdep_release: |
| 2570 | btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); |
| 2571 | btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); |
| 2572 | goto cleanup_transaction; |
| 2573 | |
| 2574 | lockdep_trans_commit_start_release: |
| 2575 | btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); |
| 2576 | btrfs_end_transaction(trans); |
| 2577 | return ret; |
| 2578 | } |
| 2579 | |
| 2580 | /* |
| 2581 | * return < 0 if error |
| 2582 | * 0 if there are no more dead_roots at the time of call |
| 2583 | * 1 there are more to be processed, call me again |
| 2584 | * |
| 2585 | * The return value indicates there are certainly more snapshots to delete, but |
| 2586 | * if there comes a new one during processing, it may return 0. We don't mind, |
| 2587 | * because btrfs_commit_super will poke cleaner thread and it will process it a |
| 2588 | * few seconds later. |
| 2589 | */ |
| 2590 | int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info) |
| 2591 | { |
| 2592 | struct btrfs_root *root; |
| 2593 | int ret; |
| 2594 | |
| 2595 | spin_lock(&fs_info->trans_lock); |
| 2596 | if (list_empty(&fs_info->dead_roots)) { |
| 2597 | spin_unlock(&fs_info->trans_lock); |
| 2598 | return 0; |
| 2599 | } |
| 2600 | root = list_first_entry(&fs_info->dead_roots, |
| 2601 | struct btrfs_root, root_list); |
| 2602 | list_del_init(&root->root_list); |
| 2603 | spin_unlock(&fs_info->trans_lock); |
| 2604 | |
| 2605 | btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); |
| 2606 | |
| 2607 | btrfs_kill_all_delayed_nodes(root); |
| 2608 | |
| 2609 | if (btrfs_header_backref_rev(root->node) < |
| 2610 | BTRFS_MIXED_BACKREF_REV) |
| 2611 | ret = btrfs_drop_snapshot(root, 0, 0); |
| 2612 | else |
| 2613 | ret = btrfs_drop_snapshot(root, 1, 0); |
| 2614 | |
| 2615 | btrfs_put_root(root); |
| 2616 | return (ret < 0) ? 0 : 1; |
| 2617 | } |
| 2618 | |
| 2619 | /* |
| 2620 | * We only mark the transaction aborted and then set the file system read-only. |
| 2621 | * This will prevent new transactions from starting or trying to join this |
| 2622 | * one. |
| 2623 | * |
| 2624 | * This means that error recovery at the call site is limited to freeing |
| 2625 | * any local memory allocations and passing the error code up without |
| 2626 | * further cleanup. The transaction should complete as it normally would |
| 2627 | * in the call path but will return -EIO. |
| 2628 | * |
| 2629 | * We'll complete the cleanup in btrfs_end_transaction and |
| 2630 | * btrfs_commit_transaction. |
| 2631 | */ |
| 2632 | void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans, |
| 2633 | const char *function, |
| 2634 | unsigned int line, int errno, bool first_hit) |
| 2635 | { |
| 2636 | struct btrfs_fs_info *fs_info = trans->fs_info; |
| 2637 | |
| 2638 | WRITE_ONCE(trans->aborted, errno); |
| 2639 | WRITE_ONCE(trans->transaction->aborted, errno); |
| 2640 | if (first_hit && errno == -ENOSPC) |
| 2641 | btrfs_dump_space_info_for_trans_abort(fs_info); |
| 2642 | /* Wake up anybody who may be waiting on this transaction */ |
| 2643 | wake_up(&fs_info->transaction_wait); |
| 2644 | wake_up(&fs_info->transaction_blocked_wait); |
| 2645 | __btrfs_handle_fs_error(fs_info, function, line, errno, NULL); |
| 2646 | } |
| 2647 | |
| 2648 | int __init btrfs_transaction_init(void) |
| 2649 | { |
| 2650 | btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", |
| 2651 | sizeof(struct btrfs_trans_handle), 0, |
| 2652 | SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); |
| 2653 | if (!btrfs_trans_handle_cachep) |
| 2654 | return -ENOMEM; |
| 2655 | return 0; |
| 2656 | } |
| 2657 | |
| 2658 | void __cold btrfs_transaction_exit(void) |
| 2659 | { |
| 2660 | kmem_cache_destroy(btrfs_trans_handle_cachep); |
| 2661 | } |