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 or TRANS_JOIN_NOSTART, we just want to catch the 296 * current transaction, and commit it. If there is no transaction, just 297 * return ENOENT. 298 */ 299 if (type == TRANS_ATTACH || type == TRANS_JOIN_NOSTART) 300 return -ENOENT; 301 302 /* 303 * JOIN_NOLOCK only happens during the transaction commit, so 304 * it is impossible that ->running_transaction is NULL 305 */ 306 BUG_ON(type == TRANS_JOIN_NOLOCK); 307 308 cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); 309 if (!cur_trans) 310 return -ENOMEM; 311 312 btrfs_lockdep_acquire(fs_info, btrfs_trans_num_writers); 313 btrfs_lockdep_acquire(fs_info, btrfs_trans_num_extwriters); 314 315 spin_lock(&fs_info->trans_lock); 316 if (fs_info->running_transaction) { 317 /* 318 * someone started a transaction after we unlocked. Make sure 319 * to redo the checks above 320 */ 321 btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 322 btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 323 kfree(cur_trans); 324 goto loop; 325 } else if (BTRFS_FS_ERROR(fs_info)) { 326 spin_unlock(&fs_info->trans_lock); 327 btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 328 btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 329 kfree(cur_trans); 330 return -EROFS; 331 } 332 333 cur_trans->fs_info = fs_info; 334 atomic_set(&cur_trans->pending_ordered, 0); 335 init_waitqueue_head(&cur_trans->pending_wait); 336 atomic_set(&cur_trans->num_writers, 1); 337 extwriter_counter_init(cur_trans, type); 338 init_waitqueue_head(&cur_trans->writer_wait); 339 init_waitqueue_head(&cur_trans->commit_wait); 340 cur_trans->state = TRANS_STATE_RUNNING; 341 /* 342 * One for this trans handle, one so it will live on until we 343 * commit the transaction. 344 */ 345 refcount_set(&cur_trans->use_count, 2); 346 cur_trans->flags = 0; 347 cur_trans->start_time = ktime_get_seconds(); 348 349 memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); 350 351 cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; 352 cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; 353 atomic_set(&cur_trans->delayed_refs.num_entries, 0); 354 355 /* 356 * although the tree mod log is per file system and not per transaction, 357 * the log must never go across transaction boundaries. 358 */ 359 smp_mb(); 360 if (!list_empty(&fs_info->tree_mod_seq_list)) 361 WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); 362 if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) 363 WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); 364 atomic64_set(&fs_info->tree_mod_seq, 0); 365 366 spin_lock_init(&cur_trans->delayed_refs.lock); 367 368 INIT_LIST_HEAD(&cur_trans->pending_snapshots); 369 INIT_LIST_HEAD(&cur_trans->dev_update_list); 370 INIT_LIST_HEAD(&cur_trans->switch_commits); 371 INIT_LIST_HEAD(&cur_trans->dirty_bgs); 372 INIT_LIST_HEAD(&cur_trans->io_bgs); 373 INIT_LIST_HEAD(&cur_trans->dropped_roots); 374 mutex_init(&cur_trans->cache_write_mutex); 375 spin_lock_init(&cur_trans->dirty_bgs_lock); 376 INIT_LIST_HEAD(&cur_trans->deleted_bgs); 377 spin_lock_init(&cur_trans->dropped_roots_lock); 378 list_add_tail(&cur_trans->list, &fs_info->trans_list); 379 extent_io_tree_init(fs_info, &cur_trans->dirty_pages, 380 IO_TREE_TRANS_DIRTY_PAGES); 381 extent_io_tree_init(fs_info, &cur_trans->pinned_extents, 382 IO_TREE_FS_PINNED_EXTENTS); 383 fs_info->generation++; 384 cur_trans->transid = fs_info->generation; 385 fs_info->running_transaction = cur_trans; 386 cur_trans->aborted = 0; 387 spin_unlock(&fs_info->trans_lock); 388 389 return 0; 390 } 391 392 /* 393 * This does all the record keeping required to make sure that a shareable root 394 * is properly recorded in a given transaction. This is required to make sure 395 * the old root from before we joined the transaction is deleted when the 396 * transaction commits. 397 */ 398 static int record_root_in_trans(struct btrfs_trans_handle *trans, 399 struct btrfs_root *root, 400 int force) 401 { 402 struct btrfs_fs_info *fs_info = root->fs_info; 403 int ret = 0; 404 405 if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 406 root->last_trans < trans->transid) || force) { 407 WARN_ON(!force && root->commit_root != root->node); 408 409 /* 410 * see below for IN_TRANS_SETUP usage rules 411 * we have the reloc mutex held now, so there 412 * is only one writer in this function 413 */ 414 set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 415 416 /* make sure readers find IN_TRANS_SETUP before 417 * they find our root->last_trans update 418 */ 419 smp_wmb(); 420 421 spin_lock(&fs_info->fs_roots_radix_lock); 422 if (root->last_trans == trans->transid && !force) { 423 spin_unlock(&fs_info->fs_roots_radix_lock); 424 return 0; 425 } 426 radix_tree_tag_set(&fs_info->fs_roots_radix, 427 (unsigned long)root->root_key.objectid, 428 BTRFS_ROOT_TRANS_TAG); 429 spin_unlock(&fs_info->fs_roots_radix_lock); 430 root->last_trans = trans->transid; 431 432 /* this is pretty tricky. We don't want to 433 * take the relocation lock in btrfs_record_root_in_trans 434 * unless we're really doing the first setup for this root in 435 * this transaction. 436 * 437 * Normally we'd use root->last_trans as a flag to decide 438 * if we want to take the expensive mutex. 439 * 440 * But, we have to set root->last_trans before we 441 * init the relocation root, otherwise, we trip over warnings 442 * in ctree.c. The solution used here is to flag ourselves 443 * with root IN_TRANS_SETUP. When this is 1, we're still 444 * fixing up the reloc trees and everyone must wait. 445 * 446 * When this is zero, they can trust root->last_trans and fly 447 * through btrfs_record_root_in_trans without having to take the 448 * lock. smp_wmb() makes sure that all the writes above are 449 * done before we pop in the zero below 450 */ 451 ret = btrfs_init_reloc_root(trans, root); 452 smp_mb__before_atomic(); 453 clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 454 } 455 return ret; 456 } 457 458 459 void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, 460 struct btrfs_root *root) 461 { 462 struct btrfs_fs_info *fs_info = root->fs_info; 463 struct btrfs_transaction *cur_trans = trans->transaction; 464 465 /* Add ourselves to the transaction dropped list */ 466 spin_lock(&cur_trans->dropped_roots_lock); 467 list_add_tail(&root->root_list, &cur_trans->dropped_roots); 468 spin_unlock(&cur_trans->dropped_roots_lock); 469 470 /* Make sure we don't try to update the root at commit time */ 471 spin_lock(&fs_info->fs_roots_radix_lock); 472 radix_tree_tag_clear(&fs_info->fs_roots_radix, 473 (unsigned long)root->root_key.objectid, 474 BTRFS_ROOT_TRANS_TAG); 475 spin_unlock(&fs_info->fs_roots_radix_lock); 476 } 477 478 int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 479 struct btrfs_root *root) 480 { 481 struct btrfs_fs_info *fs_info = root->fs_info; 482 int ret; 483 484 if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 485 return 0; 486 487 /* 488 * see record_root_in_trans for comments about IN_TRANS_SETUP usage 489 * and barriers 490 */ 491 smp_rmb(); 492 if (root->last_trans == trans->transid && 493 !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) 494 return 0; 495 496 mutex_lock(&fs_info->reloc_mutex); 497 ret = record_root_in_trans(trans, root, 0); 498 mutex_unlock(&fs_info->reloc_mutex); 499 500 return ret; 501 } 502 503 static inline int is_transaction_blocked(struct btrfs_transaction *trans) 504 { 505 return (trans->state >= TRANS_STATE_COMMIT_START && 506 trans->state < TRANS_STATE_UNBLOCKED && 507 !TRANS_ABORTED(trans)); 508 } 509 510 /* wait for commit against the current transaction to become unblocked 511 * when this is done, it is safe to start a new transaction, but the current 512 * transaction might not be fully on disk. 513 */ 514 static void wait_current_trans(struct btrfs_fs_info *fs_info) 515 { 516 struct btrfs_transaction *cur_trans; 517 518 spin_lock(&fs_info->trans_lock); 519 cur_trans = fs_info->running_transaction; 520 if (cur_trans && is_transaction_blocked(cur_trans)) { 521 refcount_inc(&cur_trans->use_count); 522 spin_unlock(&fs_info->trans_lock); 523 524 btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 525 wait_event(fs_info->transaction_wait, 526 cur_trans->state >= TRANS_STATE_UNBLOCKED || 527 TRANS_ABORTED(cur_trans)); 528 btrfs_put_transaction(cur_trans); 529 } else { 530 spin_unlock(&fs_info->trans_lock); 531 } 532 } 533 534 static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) 535 { 536 if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) 537 return 0; 538 539 if (type == TRANS_START) 540 return 1; 541 542 return 0; 543 } 544 545 static inline bool need_reserve_reloc_root(struct btrfs_root *root) 546 { 547 struct btrfs_fs_info *fs_info = root->fs_info; 548 549 if (!fs_info->reloc_ctl || 550 !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || 551 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 552 root->reloc_root) 553 return false; 554 555 return true; 556 } 557 558 static struct btrfs_trans_handle * 559 start_transaction(struct btrfs_root *root, unsigned int num_items, 560 unsigned int type, enum btrfs_reserve_flush_enum flush, 561 bool enforce_qgroups) 562 { 563 struct btrfs_fs_info *fs_info = root->fs_info; 564 struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 565 struct btrfs_trans_handle *h; 566 struct btrfs_transaction *cur_trans; 567 u64 num_bytes = 0; 568 u64 qgroup_reserved = 0; 569 bool reloc_reserved = false; 570 bool do_chunk_alloc = false; 571 int ret; 572 573 if (BTRFS_FS_ERROR(fs_info)) 574 return ERR_PTR(-EROFS); 575 576 if (current->journal_info) { 577 WARN_ON(type & TRANS_EXTWRITERS); 578 h = current->journal_info; 579 refcount_inc(&h->use_count); 580 WARN_ON(refcount_read(&h->use_count) > 2); 581 h->orig_rsv = h->block_rsv; 582 h->block_rsv = NULL; 583 goto got_it; 584 } 585 586 /* 587 * Do the reservation before we join the transaction so we can do all 588 * the appropriate flushing if need be. 589 */ 590 if (num_items && root != fs_info->chunk_root) { 591 struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; 592 u64 delayed_refs_bytes = 0; 593 594 qgroup_reserved = num_items * fs_info->nodesize; 595 /* 596 * Use prealloc for now, as there might be a currently running 597 * transaction that could free this reserved space prematurely 598 * by committing. 599 */ 600 ret = btrfs_qgroup_reserve_meta_prealloc(root, qgroup_reserved, 601 enforce_qgroups, false); 602 if (ret) 603 return ERR_PTR(ret); 604 605 /* 606 * We want to reserve all the bytes we may need all at once, so 607 * we only do 1 enospc flushing cycle per transaction start. We 608 * accomplish this by simply assuming we'll do num_items worth 609 * of delayed refs updates in this trans handle, and refill that 610 * amount for whatever is missing in the reserve. 611 */ 612 num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); 613 if (flush == BTRFS_RESERVE_FLUSH_ALL && 614 !btrfs_block_rsv_full(delayed_refs_rsv)) { 615 delayed_refs_bytes = btrfs_calc_delayed_ref_bytes(fs_info, 616 num_items); 617 num_bytes += delayed_refs_bytes; 618 } 619 620 /* 621 * Do the reservation for the relocation root creation 622 */ 623 if (need_reserve_reloc_root(root)) { 624 num_bytes += fs_info->nodesize; 625 reloc_reserved = true; 626 } 627 628 ret = btrfs_block_rsv_add(fs_info, rsv, num_bytes, flush); 629 if (ret) 630 goto reserve_fail; 631 if (delayed_refs_bytes) { 632 btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, 633 delayed_refs_bytes); 634 num_bytes -= delayed_refs_bytes; 635 } 636 637 if (rsv->space_info->force_alloc) 638 do_chunk_alloc = true; 639 } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && 640 !btrfs_block_rsv_full(delayed_refs_rsv)) { 641 /* 642 * Some people call with btrfs_start_transaction(root, 0) 643 * because they can be throttled, but have some other mechanism 644 * for reserving space. We still want these guys to refill the 645 * delayed block_rsv so just add 1 items worth of reservation 646 * here. 647 */ 648 ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); 649 if (ret) 650 goto reserve_fail; 651 } 652 again: 653 h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); 654 if (!h) { 655 ret = -ENOMEM; 656 goto alloc_fail; 657 } 658 659 /* 660 * If we are JOIN_NOLOCK we're already committing a transaction and 661 * waiting on this guy, so we don't need to do the sb_start_intwrite 662 * because we're already holding a ref. We need this because we could 663 * have raced in and did an fsync() on a file which can kick a commit 664 * and then we deadlock with somebody doing a freeze. 665 * 666 * If we are ATTACH, it means we just want to catch the current 667 * transaction and commit it, so we needn't do sb_start_intwrite(). 668 */ 669 if (type & __TRANS_FREEZABLE) 670 sb_start_intwrite(fs_info->sb); 671 672 if (may_wait_transaction(fs_info, type)) 673 wait_current_trans(fs_info); 674 675 do { 676 ret = join_transaction(fs_info, type); 677 if (ret == -EBUSY) { 678 wait_current_trans(fs_info); 679 if (unlikely(type == TRANS_ATTACH || 680 type == TRANS_JOIN_NOSTART)) 681 ret = -ENOENT; 682 } 683 } while (ret == -EBUSY); 684 685 if (ret < 0) 686 goto join_fail; 687 688 cur_trans = fs_info->running_transaction; 689 690 h->transid = cur_trans->transid; 691 h->transaction = cur_trans; 692 refcount_set(&h->use_count, 1); 693 h->fs_info = root->fs_info; 694 695 h->type = type; 696 INIT_LIST_HEAD(&h->new_bgs); 697 698 smp_mb(); 699 if (cur_trans->state >= TRANS_STATE_COMMIT_START && 700 may_wait_transaction(fs_info, type)) { 701 current->journal_info = h; 702 btrfs_commit_transaction(h); 703 goto again; 704 } 705 706 if (num_bytes) { 707 trace_btrfs_space_reservation(fs_info, "transaction", 708 h->transid, num_bytes, 1); 709 h->block_rsv = &fs_info->trans_block_rsv; 710 h->bytes_reserved = num_bytes; 711 h->reloc_reserved = reloc_reserved; 712 } 713 714 /* 715 * Now that we have found a transaction to be a part of, convert the 716 * qgroup reservation from prealloc to pertrans. A different transaction 717 * can't race in and free our pertrans out from under us. 718 */ 719 if (qgroup_reserved) 720 btrfs_qgroup_convert_reserved_meta(root, qgroup_reserved); 721 722 got_it: 723 if (!current->journal_info) 724 current->journal_info = h; 725 726 /* 727 * If the space_info is marked ALLOC_FORCE then we'll get upgraded to 728 * ALLOC_FORCE the first run through, and then we won't allocate for 729 * anybody else who races in later. We don't care about the return 730 * value here. 731 */ 732 if (do_chunk_alloc && num_bytes) { 733 u64 flags = h->block_rsv->space_info->flags; 734 735 btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), 736 CHUNK_ALLOC_NO_FORCE); 737 } 738 739 /* 740 * btrfs_record_root_in_trans() needs to alloc new extents, and may 741 * call btrfs_join_transaction() while we're also starting a 742 * transaction. 743 * 744 * Thus it need to be called after current->journal_info initialized, 745 * or we can deadlock. 746 */ 747 ret = btrfs_record_root_in_trans(h, root); 748 if (ret) { 749 /* 750 * The transaction handle is fully initialized and linked with 751 * other structures so it needs to be ended in case of errors, 752 * not just freed. 753 */ 754 btrfs_end_transaction(h); 755 return ERR_PTR(ret); 756 } 757 758 return h; 759 760 join_fail: 761 if (type & __TRANS_FREEZABLE) 762 sb_end_intwrite(fs_info->sb); 763 kmem_cache_free(btrfs_trans_handle_cachep, h); 764 alloc_fail: 765 if (num_bytes) 766 btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, 767 num_bytes, NULL); 768 reserve_fail: 769 btrfs_qgroup_free_meta_prealloc(root, qgroup_reserved); 770 return ERR_PTR(ret); 771 } 772 773 struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 774 unsigned int num_items) 775 { 776 return start_transaction(root, num_items, TRANS_START, 777 BTRFS_RESERVE_FLUSH_ALL, true); 778 } 779 780 struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( 781 struct btrfs_root *root, 782 unsigned int num_items) 783 { 784 return start_transaction(root, num_items, TRANS_START, 785 BTRFS_RESERVE_FLUSH_ALL_STEAL, false); 786 } 787 788 struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) 789 { 790 return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, 791 true); 792 } 793 794 struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) 795 { 796 return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 797 BTRFS_RESERVE_NO_FLUSH, true); 798 } 799 800 /* 801 * Similar to regular join but it never starts a transaction when none is 802 * running or after waiting for the current one to finish. 803 */ 804 struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) 805 { 806 return start_transaction(root, 0, TRANS_JOIN_NOSTART, 807 BTRFS_RESERVE_NO_FLUSH, true); 808 } 809 810 /* 811 * btrfs_attach_transaction() - catch the running transaction 812 * 813 * It is used when we want to commit the current the transaction, but 814 * don't want to start a new one. 815 * 816 * Note: If this function return -ENOENT, it just means there is no 817 * running transaction. But it is possible that the inactive transaction 818 * is still in the memory, not fully on disk. If you hope there is no 819 * inactive transaction in the fs when -ENOENT is returned, you should 820 * invoke 821 * btrfs_attach_transaction_barrier() 822 */ 823 struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) 824 { 825 return start_transaction(root, 0, TRANS_ATTACH, 826 BTRFS_RESERVE_NO_FLUSH, true); 827 } 828 829 /* 830 * btrfs_attach_transaction_barrier() - catch the running transaction 831 * 832 * It is similar to the above function, the difference is this one 833 * will wait for all the inactive transactions until they fully 834 * complete. 835 */ 836 struct btrfs_trans_handle * 837 btrfs_attach_transaction_barrier(struct btrfs_root *root) 838 { 839 struct btrfs_trans_handle *trans; 840 841 trans = start_transaction(root, 0, TRANS_ATTACH, 842 BTRFS_RESERVE_NO_FLUSH, true); 843 if (trans == ERR_PTR(-ENOENT)) { 844 int ret; 845 846 ret = btrfs_wait_for_commit(root->fs_info, 0); 847 if (ret) 848 return ERR_PTR(ret); 849 } 850 851 return trans; 852 } 853 854 /* Wait for a transaction commit to reach at least the given state. */ 855 static noinline void wait_for_commit(struct btrfs_transaction *commit, 856 const enum btrfs_trans_state min_state) 857 { 858 struct btrfs_fs_info *fs_info = commit->fs_info; 859 u64 transid = commit->transid; 860 bool put = false; 861 862 /* 863 * At the moment this function is called with min_state either being 864 * TRANS_STATE_COMPLETED or TRANS_STATE_SUPER_COMMITTED. 865 */ 866 if (min_state == TRANS_STATE_COMPLETED) 867 btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 868 else 869 btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 870 871 while (1) { 872 wait_event(commit->commit_wait, commit->state >= min_state); 873 if (put) 874 btrfs_put_transaction(commit); 875 876 if (min_state < TRANS_STATE_COMPLETED) 877 break; 878 879 /* 880 * A transaction isn't really completed until all of the 881 * previous transactions are completed, but with fsync we can 882 * end up with SUPER_COMMITTED transactions before a COMPLETED 883 * transaction. Wait for those. 884 */ 885 886 spin_lock(&fs_info->trans_lock); 887 commit = list_first_entry_or_null(&fs_info->trans_list, 888 struct btrfs_transaction, 889 list); 890 if (!commit || commit->transid > transid) { 891 spin_unlock(&fs_info->trans_lock); 892 break; 893 } 894 refcount_inc(&commit->use_count); 895 put = true; 896 spin_unlock(&fs_info->trans_lock); 897 } 898 } 899 900 int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) 901 { 902 struct btrfs_transaction *cur_trans = NULL, *t; 903 int ret = 0; 904 905 if (transid) { 906 if (transid <= fs_info->last_trans_committed) 907 goto out; 908 909 /* find specified transaction */ 910 spin_lock(&fs_info->trans_lock); 911 list_for_each_entry(t, &fs_info->trans_list, list) { 912 if (t->transid == transid) { 913 cur_trans = t; 914 refcount_inc(&cur_trans->use_count); 915 ret = 0; 916 break; 917 } 918 if (t->transid > transid) { 919 ret = 0; 920 break; 921 } 922 } 923 spin_unlock(&fs_info->trans_lock); 924 925 /* 926 * The specified transaction doesn't exist, or we 927 * raced with btrfs_commit_transaction 928 */ 929 if (!cur_trans) { 930 if (transid > fs_info->last_trans_committed) 931 ret = -EINVAL; 932 goto out; 933 } 934 } else { 935 /* find newest transaction that is committing | committed */ 936 spin_lock(&fs_info->trans_lock); 937 list_for_each_entry_reverse(t, &fs_info->trans_list, 938 list) { 939 if (t->state >= TRANS_STATE_COMMIT_START) { 940 if (t->state == TRANS_STATE_COMPLETED) 941 break; 942 cur_trans = t; 943 refcount_inc(&cur_trans->use_count); 944 break; 945 } 946 } 947 spin_unlock(&fs_info->trans_lock); 948 if (!cur_trans) 949 goto out; /* nothing committing|committed */ 950 } 951 952 wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); 953 ret = cur_trans->aborted; 954 btrfs_put_transaction(cur_trans); 955 out: 956 return ret; 957 } 958 959 void btrfs_throttle(struct btrfs_fs_info *fs_info) 960 { 961 wait_current_trans(fs_info); 962 } 963 964 bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) 965 { 966 struct btrfs_transaction *cur_trans = trans->transaction; 967 968 if (cur_trans->state >= TRANS_STATE_COMMIT_START || 969 test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) 970 return true; 971 972 if (btrfs_check_space_for_delayed_refs(trans->fs_info)) 973 return true; 974 975 return !!btrfs_block_rsv_check(&trans->fs_info->global_block_rsv, 50); 976 } 977 978 static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) 979 980 { 981 struct btrfs_fs_info *fs_info = trans->fs_info; 982 983 if (!trans->block_rsv) { 984 ASSERT(!trans->bytes_reserved); 985 return; 986 } 987 988 if (!trans->bytes_reserved) 989 return; 990 991 ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); 992 trace_btrfs_space_reservation(fs_info, "transaction", 993 trans->transid, trans->bytes_reserved, 0); 994 btrfs_block_rsv_release(fs_info, trans->block_rsv, 995 trans->bytes_reserved, NULL); 996 trans->bytes_reserved = 0; 997 } 998 999 static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 1000 int throttle) 1001 { 1002 struct btrfs_fs_info *info = trans->fs_info; 1003 struct btrfs_transaction *cur_trans = trans->transaction; 1004 int err = 0; 1005 1006 if (refcount_read(&trans->use_count) > 1) { 1007 refcount_dec(&trans->use_count); 1008 trans->block_rsv = trans->orig_rsv; 1009 return 0; 1010 } 1011 1012 btrfs_trans_release_metadata(trans); 1013 trans->block_rsv = NULL; 1014 1015 btrfs_create_pending_block_groups(trans); 1016 1017 btrfs_trans_release_chunk_metadata(trans); 1018 1019 if (trans->type & __TRANS_FREEZABLE) 1020 sb_end_intwrite(info->sb); 1021 1022 WARN_ON(cur_trans != info->running_transaction); 1023 WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 1024 atomic_dec(&cur_trans->num_writers); 1025 extwriter_counter_dec(cur_trans, trans->type); 1026 1027 cond_wake_up(&cur_trans->writer_wait); 1028 1029 btrfs_lockdep_release(info, btrfs_trans_num_extwriters); 1030 btrfs_lockdep_release(info, btrfs_trans_num_writers); 1031 1032 btrfs_put_transaction(cur_trans); 1033 1034 if (current->journal_info == trans) 1035 current->journal_info = NULL; 1036 1037 if (throttle) 1038 btrfs_run_delayed_iputs(info); 1039 1040 if (TRANS_ABORTED(trans) || BTRFS_FS_ERROR(info)) { 1041 wake_up_process(info->transaction_kthread); 1042 if (TRANS_ABORTED(trans)) 1043 err = trans->aborted; 1044 else 1045 err = -EROFS; 1046 } 1047 1048 kmem_cache_free(btrfs_trans_handle_cachep, trans); 1049 return err; 1050 } 1051 1052 int btrfs_end_transaction(struct btrfs_trans_handle *trans) 1053 { 1054 return __btrfs_end_transaction(trans, 0); 1055 } 1056 1057 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) 1058 { 1059 return __btrfs_end_transaction(trans, 1); 1060 } 1061 1062 /* 1063 * when btree blocks are allocated, they have some corresponding bits set for 1064 * them in one of two extent_io trees. This is used to make sure all of 1065 * those extents are sent to disk but does not wait on them 1066 */ 1067 int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, 1068 struct extent_io_tree *dirty_pages, int mark) 1069 { 1070 int err = 0; 1071 int werr = 0; 1072 struct address_space *mapping = fs_info->btree_inode->i_mapping; 1073 struct extent_state *cached_state = NULL; 1074 u64 start = 0; 1075 u64 end; 1076 1077 while (find_first_extent_bit(dirty_pages, start, &start, &end, 1078 mark, &cached_state)) { 1079 bool wait_writeback = false; 1080 1081 err = convert_extent_bit(dirty_pages, start, end, 1082 EXTENT_NEED_WAIT, 1083 mark, &cached_state); 1084 /* 1085 * convert_extent_bit can return -ENOMEM, which is most of the 1086 * time a temporary error. So when it happens, ignore the error 1087 * and wait for writeback of this range to finish - because we 1088 * failed to set the bit EXTENT_NEED_WAIT for the range, a call 1089 * to __btrfs_wait_marked_extents() would not know that 1090 * writeback for this range started and therefore wouldn't 1091 * wait for it to finish - we don't want to commit a 1092 * superblock that points to btree nodes/leafs for which 1093 * writeback hasn't finished yet (and without errors). 1094 * We cleanup any entries left in the io tree when committing 1095 * the transaction (through extent_io_tree_release()). 1096 */ 1097 if (err == -ENOMEM) { 1098 err = 0; 1099 wait_writeback = true; 1100 } 1101 if (!err) 1102 err = filemap_fdatawrite_range(mapping, start, end); 1103 if (err) 1104 werr = err; 1105 else if (wait_writeback) 1106 werr = filemap_fdatawait_range(mapping, start, end); 1107 free_extent_state(cached_state); 1108 cached_state = NULL; 1109 cond_resched(); 1110 start = end + 1; 1111 } 1112 return werr; 1113 } 1114 1115 /* 1116 * when btree blocks are allocated, they have some corresponding bits set for 1117 * them in one of two extent_io trees. This is used to make sure all of 1118 * those extents are on disk for transaction or log commit. We wait 1119 * on all the pages and clear them from the dirty pages state tree 1120 */ 1121 static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, 1122 struct extent_io_tree *dirty_pages) 1123 { 1124 int err = 0; 1125 int werr = 0; 1126 struct address_space *mapping = fs_info->btree_inode->i_mapping; 1127 struct extent_state *cached_state = NULL; 1128 u64 start = 0; 1129 u64 end; 1130 1131 while (find_first_extent_bit(dirty_pages, start, &start, &end, 1132 EXTENT_NEED_WAIT, &cached_state)) { 1133 /* 1134 * Ignore -ENOMEM errors returned by clear_extent_bit(). 1135 * When committing the transaction, we'll remove any entries 1136 * left in the io tree. For a log commit, we don't remove them 1137 * after committing the log because the tree can be accessed 1138 * concurrently - we do it only at transaction commit time when 1139 * it's safe to do it (through extent_io_tree_release()). 1140 */ 1141 err = clear_extent_bit(dirty_pages, start, end, 1142 EXTENT_NEED_WAIT, &cached_state); 1143 if (err == -ENOMEM) 1144 err = 0; 1145 if (!err) 1146 err = filemap_fdatawait_range(mapping, start, end); 1147 if (err) 1148 werr = err; 1149 free_extent_state(cached_state); 1150 cached_state = NULL; 1151 cond_resched(); 1152 start = end + 1; 1153 } 1154 if (err) 1155 werr = err; 1156 return werr; 1157 } 1158 1159 static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, 1160 struct extent_io_tree *dirty_pages) 1161 { 1162 bool errors = false; 1163 int err; 1164 1165 err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1166 if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) 1167 errors = true; 1168 1169 if (errors && !err) 1170 err = -EIO; 1171 return err; 1172 } 1173 1174 int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) 1175 { 1176 struct btrfs_fs_info *fs_info = log_root->fs_info; 1177 struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; 1178 bool errors = false; 1179 int err; 1180 1181 ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 1182 1183 err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1184 if ((mark & EXTENT_DIRTY) && 1185 test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) 1186 errors = true; 1187 1188 if ((mark & EXTENT_NEW) && 1189 test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) 1190 errors = true; 1191 1192 if (errors && !err) 1193 err = -EIO; 1194 return err; 1195 } 1196 1197 /* 1198 * When btree blocks are allocated the corresponding extents are marked dirty. 1199 * This function ensures such extents are persisted on disk for transaction or 1200 * log commit. 1201 * 1202 * @trans: transaction whose dirty pages we'd like to write 1203 */ 1204 static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) 1205 { 1206 int ret; 1207 int ret2; 1208 struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; 1209 struct btrfs_fs_info *fs_info = trans->fs_info; 1210 struct blk_plug plug; 1211 1212 blk_start_plug(&plug); 1213 ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); 1214 blk_finish_plug(&plug); 1215 ret2 = btrfs_wait_extents(fs_info, dirty_pages); 1216 1217 extent_io_tree_release(&trans->transaction->dirty_pages); 1218 1219 if (ret) 1220 return ret; 1221 else if (ret2) 1222 return ret2; 1223 else 1224 return 0; 1225 } 1226 1227 /* 1228 * this is used to update the root pointer in the tree of tree roots. 1229 * 1230 * But, in the case of the extent allocation tree, updating the root 1231 * pointer may allocate blocks which may change the root of the extent 1232 * allocation tree. 1233 * 1234 * So, this loops and repeats and makes sure the cowonly root didn't 1235 * change while the root pointer was being updated in the metadata. 1236 */ 1237 static int update_cowonly_root(struct btrfs_trans_handle *trans, 1238 struct btrfs_root *root) 1239 { 1240 int ret; 1241 u64 old_root_bytenr; 1242 u64 old_root_used; 1243 struct btrfs_fs_info *fs_info = root->fs_info; 1244 struct btrfs_root *tree_root = fs_info->tree_root; 1245 1246 old_root_used = btrfs_root_used(&root->root_item); 1247 1248 while (1) { 1249 old_root_bytenr = btrfs_root_bytenr(&root->root_item); 1250 if (old_root_bytenr == root->node->start && 1251 old_root_used == btrfs_root_used(&root->root_item)) 1252 break; 1253 1254 btrfs_set_root_node(&root->root_item, root->node); 1255 ret = btrfs_update_root(trans, tree_root, 1256 &root->root_key, 1257 &root->root_item); 1258 if (ret) 1259 return ret; 1260 1261 old_root_used = btrfs_root_used(&root->root_item); 1262 } 1263 1264 return 0; 1265 } 1266 1267 /* 1268 * update all the cowonly tree roots on disk 1269 * 1270 * The error handling in this function may not be obvious. Any of the 1271 * failures will cause the file system to go offline. We still need 1272 * to clean up the delayed refs. 1273 */ 1274 static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) 1275 { 1276 struct btrfs_fs_info *fs_info = trans->fs_info; 1277 struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; 1278 struct list_head *io_bgs = &trans->transaction->io_bgs; 1279 struct list_head *next; 1280 struct extent_buffer *eb; 1281 int ret; 1282 1283 /* 1284 * At this point no one can be using this transaction to modify any tree 1285 * and no one can start another transaction to modify any tree either. 1286 */ 1287 ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1288 1289 eb = btrfs_lock_root_node(fs_info->tree_root); 1290 ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 1291 0, &eb, BTRFS_NESTING_COW); 1292 btrfs_tree_unlock(eb); 1293 free_extent_buffer(eb); 1294 1295 if (ret) 1296 return ret; 1297 1298 ret = btrfs_run_dev_stats(trans); 1299 if (ret) 1300 return ret; 1301 ret = btrfs_run_dev_replace(trans); 1302 if (ret) 1303 return ret; 1304 ret = btrfs_run_qgroups(trans); 1305 if (ret) 1306 return ret; 1307 1308 ret = btrfs_setup_space_cache(trans); 1309 if (ret) 1310 return ret; 1311 1312 again: 1313 while (!list_empty(&fs_info->dirty_cowonly_roots)) { 1314 struct btrfs_root *root; 1315 next = fs_info->dirty_cowonly_roots.next; 1316 list_del_init(next); 1317 root = list_entry(next, struct btrfs_root, dirty_list); 1318 clear_bit(BTRFS_ROOT_DIRTY, &root->state); 1319 1320 list_add_tail(&root->dirty_list, 1321 &trans->transaction->switch_commits); 1322 ret = update_cowonly_root(trans, root); 1323 if (ret) 1324 return ret; 1325 } 1326 1327 /* Now flush any delayed refs generated by updating all of the roots */ 1328 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1329 if (ret) 1330 return ret; 1331 1332 while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { 1333 ret = btrfs_write_dirty_block_groups(trans); 1334 if (ret) 1335 return ret; 1336 1337 /* 1338 * We're writing the dirty block groups, which could generate 1339 * delayed refs, which could generate more dirty block groups, 1340 * so we want to keep this flushing in this loop to make sure 1341 * everything gets run. 1342 */ 1343 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1344 if (ret) 1345 return ret; 1346 } 1347 1348 if (!list_empty(&fs_info->dirty_cowonly_roots)) 1349 goto again; 1350 1351 /* Update dev-replace pointer once everything is committed */ 1352 fs_info->dev_replace.committed_cursor_left = 1353 fs_info->dev_replace.cursor_left_last_write_of_item; 1354 1355 return 0; 1356 } 1357 1358 /* 1359 * If we had a pending drop we need to see if there are any others left in our 1360 * dead roots list, and if not clear our bit and wake any waiters. 1361 */ 1362 void btrfs_maybe_wake_unfinished_drop(struct btrfs_fs_info *fs_info) 1363 { 1364 /* 1365 * We put the drop in progress roots at the front of the list, so if the 1366 * first entry doesn't have UNFINISHED_DROP set we can wake everybody 1367 * up. 1368 */ 1369 spin_lock(&fs_info->trans_lock); 1370 if (!list_empty(&fs_info->dead_roots)) { 1371 struct btrfs_root *root = list_first_entry(&fs_info->dead_roots, 1372 struct btrfs_root, 1373 root_list); 1374 if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) { 1375 spin_unlock(&fs_info->trans_lock); 1376 return; 1377 } 1378 } 1379 spin_unlock(&fs_info->trans_lock); 1380 1381 btrfs_wake_unfinished_drop(fs_info); 1382 } 1383 1384 /* 1385 * dead roots are old snapshots that need to be deleted. This allocates 1386 * a dirty root struct and adds it into the list of dead roots that need to 1387 * be deleted 1388 */ 1389 void btrfs_add_dead_root(struct btrfs_root *root) 1390 { 1391 struct btrfs_fs_info *fs_info = root->fs_info; 1392 1393 spin_lock(&fs_info->trans_lock); 1394 if (list_empty(&root->root_list)) { 1395 btrfs_grab_root(root); 1396 1397 /* We want to process the partially complete drops first. */ 1398 if (test_bit(BTRFS_ROOT_UNFINISHED_DROP, &root->state)) 1399 list_add(&root->root_list, &fs_info->dead_roots); 1400 else 1401 list_add_tail(&root->root_list, &fs_info->dead_roots); 1402 } 1403 spin_unlock(&fs_info->trans_lock); 1404 } 1405 1406 /* 1407 * Update each subvolume root and its relocation root, if it exists, in the tree 1408 * of tree roots. Also free log roots if they exist. 1409 */ 1410 static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) 1411 { 1412 struct btrfs_fs_info *fs_info = trans->fs_info; 1413 struct btrfs_root *gang[8]; 1414 int i; 1415 int ret; 1416 1417 /* 1418 * At this point no one can be using this transaction to modify any tree 1419 * and no one can start another transaction to modify any tree either. 1420 */ 1421 ASSERT(trans->transaction->state == TRANS_STATE_COMMIT_DOING); 1422 1423 spin_lock(&fs_info->fs_roots_radix_lock); 1424 while (1) { 1425 ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 1426 (void **)gang, 0, 1427 ARRAY_SIZE(gang), 1428 BTRFS_ROOT_TRANS_TAG); 1429 if (ret == 0) 1430 break; 1431 for (i = 0; i < ret; i++) { 1432 struct btrfs_root *root = gang[i]; 1433 int ret2; 1434 1435 /* 1436 * At this point we can neither have tasks logging inodes 1437 * from a root nor trying to commit a log tree. 1438 */ 1439 ASSERT(atomic_read(&root->log_writers) == 0); 1440 ASSERT(atomic_read(&root->log_commit[0]) == 0); 1441 ASSERT(atomic_read(&root->log_commit[1]) == 0); 1442 1443 radix_tree_tag_clear(&fs_info->fs_roots_radix, 1444 (unsigned long)root->root_key.objectid, 1445 BTRFS_ROOT_TRANS_TAG); 1446 spin_unlock(&fs_info->fs_roots_radix_lock); 1447 1448 btrfs_free_log(trans, root); 1449 ret2 = btrfs_update_reloc_root(trans, root); 1450 if (ret2) 1451 return ret2; 1452 1453 /* see comments in should_cow_block() */ 1454 clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1455 smp_mb__after_atomic(); 1456 1457 if (root->commit_root != root->node) { 1458 list_add_tail(&root->dirty_list, 1459 &trans->transaction->switch_commits); 1460 btrfs_set_root_node(&root->root_item, 1461 root->node); 1462 } 1463 1464 ret2 = btrfs_update_root(trans, fs_info->tree_root, 1465 &root->root_key, 1466 &root->root_item); 1467 if (ret2) 1468 return ret2; 1469 spin_lock(&fs_info->fs_roots_radix_lock); 1470 btrfs_qgroup_free_meta_all_pertrans(root); 1471 } 1472 } 1473 spin_unlock(&fs_info->fs_roots_radix_lock); 1474 return 0; 1475 } 1476 1477 /* 1478 * defrag a given btree. 1479 * Every leaf in the btree is read and defragged. 1480 */ 1481 int btrfs_defrag_root(struct btrfs_root *root) 1482 { 1483 struct btrfs_fs_info *info = root->fs_info; 1484 struct btrfs_trans_handle *trans; 1485 int ret; 1486 1487 if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) 1488 return 0; 1489 1490 while (1) { 1491 trans = btrfs_start_transaction(root, 0); 1492 if (IS_ERR(trans)) { 1493 ret = PTR_ERR(trans); 1494 break; 1495 } 1496 1497 ret = btrfs_defrag_leaves(trans, root); 1498 1499 btrfs_end_transaction(trans); 1500 btrfs_btree_balance_dirty(info); 1501 cond_resched(); 1502 1503 if (btrfs_fs_closing(info) || ret != -EAGAIN) 1504 break; 1505 1506 if (btrfs_defrag_cancelled(info)) { 1507 btrfs_debug(info, "defrag_root cancelled"); 1508 ret = -EAGAIN; 1509 break; 1510 } 1511 } 1512 clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); 1513 return ret; 1514 } 1515 1516 /* 1517 * Do all special snapshot related qgroup dirty hack. 1518 * 1519 * Will do all needed qgroup inherit and dirty hack like switch commit 1520 * roots inside one transaction and write all btree into disk, to make 1521 * qgroup works. 1522 */ 1523 static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, 1524 struct btrfs_root *src, 1525 struct btrfs_root *parent, 1526 struct btrfs_qgroup_inherit *inherit, 1527 u64 dst_objectid) 1528 { 1529 struct btrfs_fs_info *fs_info = src->fs_info; 1530 int ret; 1531 1532 /* 1533 * Save some performance in the case that qgroups are not 1534 * enabled. If this check races with the ioctl, rescan will 1535 * kick in anyway. 1536 */ 1537 if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) 1538 return 0; 1539 1540 /* 1541 * Ensure dirty @src will be committed. Or, after coming 1542 * commit_fs_roots() and switch_commit_roots(), any dirty but not 1543 * recorded root will never be updated again, causing an outdated root 1544 * item. 1545 */ 1546 ret = record_root_in_trans(trans, src, 1); 1547 if (ret) 1548 return ret; 1549 1550 /* 1551 * btrfs_qgroup_inherit relies on a consistent view of the usage for the 1552 * src root, so we must run the delayed refs here. 1553 * 1554 * However this isn't particularly fool proof, because there's no 1555 * synchronization keeping us from changing the tree after this point 1556 * before we do the qgroup_inherit, or even from making changes while 1557 * we're doing the qgroup_inherit. But that's a problem for the future, 1558 * for now flush the delayed refs to narrow the race window where the 1559 * qgroup counters could end up wrong. 1560 */ 1561 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1562 if (ret) { 1563 btrfs_abort_transaction(trans, ret); 1564 return ret; 1565 } 1566 1567 ret = commit_fs_roots(trans); 1568 if (ret) 1569 goto out; 1570 ret = btrfs_qgroup_account_extents(trans); 1571 if (ret < 0) 1572 goto out; 1573 1574 /* Now qgroup are all updated, we can inherit it to new qgroups */ 1575 ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, 1576 inherit); 1577 if (ret < 0) 1578 goto out; 1579 1580 /* 1581 * Now we do a simplified commit transaction, which will: 1582 * 1) commit all subvolume and extent tree 1583 * To ensure all subvolume and extent tree have a valid 1584 * commit_root to accounting later insert_dir_item() 1585 * 2) write all btree blocks onto disk 1586 * This is to make sure later btree modification will be cowed 1587 * Or commit_root can be populated and cause wrong qgroup numbers 1588 * In this simplified commit, we don't really care about other trees 1589 * like chunk and root tree, as they won't affect qgroup. 1590 * And we don't write super to avoid half committed status. 1591 */ 1592 ret = commit_cowonly_roots(trans); 1593 if (ret) 1594 goto out; 1595 switch_commit_roots(trans); 1596 ret = btrfs_write_and_wait_transaction(trans); 1597 if (ret) 1598 btrfs_handle_fs_error(fs_info, ret, 1599 "Error while writing out transaction for qgroup"); 1600 1601 out: 1602 /* 1603 * Force parent root to be updated, as we recorded it before so its 1604 * last_trans == cur_transid. 1605 * Or it won't be committed again onto disk after later 1606 * insert_dir_item() 1607 */ 1608 if (!ret) 1609 ret = record_root_in_trans(trans, parent, 1); 1610 return ret; 1611 } 1612 1613 /* 1614 * new snapshots need to be created at a very specific time in the 1615 * transaction commit. This does the actual creation. 1616 * 1617 * Note: 1618 * If the error which may affect the commitment of the current transaction 1619 * happens, we should return the error number. If the error which just affect 1620 * the creation of the pending snapshots, just return 0. 1621 */ 1622 static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 1623 struct btrfs_pending_snapshot *pending) 1624 { 1625 1626 struct btrfs_fs_info *fs_info = trans->fs_info; 1627 struct btrfs_key key; 1628 struct btrfs_root_item *new_root_item; 1629 struct btrfs_root *tree_root = fs_info->tree_root; 1630 struct btrfs_root *root = pending->root; 1631 struct btrfs_root *parent_root; 1632 struct btrfs_block_rsv *rsv; 1633 struct inode *parent_inode = pending->dir; 1634 struct btrfs_path *path; 1635 struct btrfs_dir_item *dir_item; 1636 struct extent_buffer *tmp; 1637 struct extent_buffer *old; 1638 struct timespec64 cur_time; 1639 int ret = 0; 1640 u64 to_reserve = 0; 1641 u64 index = 0; 1642 u64 objectid; 1643 u64 root_flags; 1644 unsigned int nofs_flags; 1645 struct fscrypt_name fname; 1646 1647 ASSERT(pending->path); 1648 path = pending->path; 1649 1650 ASSERT(pending->root_item); 1651 new_root_item = pending->root_item; 1652 1653 /* 1654 * We're inside a transaction and must make sure that any potential 1655 * allocations with GFP_KERNEL in fscrypt won't recurse back to 1656 * filesystem. 1657 */ 1658 nofs_flags = memalloc_nofs_save(); 1659 pending->error = fscrypt_setup_filename(parent_inode, 1660 &pending->dentry->d_name, 0, 1661 &fname); 1662 memalloc_nofs_restore(nofs_flags); 1663 if (pending->error) 1664 goto free_pending; 1665 1666 pending->error = btrfs_get_free_objectid(tree_root, &objectid); 1667 if (pending->error) 1668 goto free_fname; 1669 1670 /* 1671 * Make qgroup to skip current new snapshot's qgroupid, as it is 1672 * accounted by later btrfs_qgroup_inherit(). 1673 */ 1674 btrfs_set_skip_qgroup(trans, objectid); 1675 1676 btrfs_reloc_pre_snapshot(pending, &to_reserve); 1677 1678 if (to_reserve > 0) { 1679 pending->error = btrfs_block_rsv_add(fs_info, 1680 &pending->block_rsv, 1681 to_reserve, 1682 BTRFS_RESERVE_NO_FLUSH); 1683 if (pending->error) 1684 goto clear_skip_qgroup; 1685 } 1686 1687 key.objectid = objectid; 1688 key.offset = (u64)-1; 1689 key.type = BTRFS_ROOT_ITEM_KEY; 1690 1691 rsv = trans->block_rsv; 1692 trans->block_rsv = &pending->block_rsv; 1693 trans->bytes_reserved = trans->block_rsv->reserved; 1694 trace_btrfs_space_reservation(fs_info, "transaction", 1695 trans->transid, 1696 trans->bytes_reserved, 1); 1697 parent_root = BTRFS_I(parent_inode)->root; 1698 ret = record_root_in_trans(trans, parent_root, 0); 1699 if (ret) 1700 goto fail; 1701 cur_time = current_time(parent_inode); 1702 1703 /* 1704 * insert the directory item 1705 */ 1706 ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); 1707 if (ret) { 1708 btrfs_abort_transaction(trans, ret); 1709 goto fail; 1710 } 1711 1712 /* check if there is a file/dir which has the same name. */ 1713 dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, 1714 btrfs_ino(BTRFS_I(parent_inode)), 1715 &fname.disk_name, 0); 1716 if (dir_item != NULL && !IS_ERR(dir_item)) { 1717 pending->error = -EEXIST; 1718 goto dir_item_existed; 1719 } else if (IS_ERR(dir_item)) { 1720 ret = PTR_ERR(dir_item); 1721 btrfs_abort_transaction(trans, ret); 1722 goto fail; 1723 } 1724 btrfs_release_path(path); 1725 1726 /* 1727 * pull in the delayed directory update 1728 * and the delayed inode item 1729 * otherwise we corrupt the FS during 1730 * snapshot 1731 */ 1732 ret = btrfs_run_delayed_items(trans); 1733 if (ret) { /* Transaction aborted */ 1734 btrfs_abort_transaction(trans, ret); 1735 goto fail; 1736 } 1737 1738 ret = record_root_in_trans(trans, root, 0); 1739 if (ret) { 1740 btrfs_abort_transaction(trans, ret); 1741 goto fail; 1742 } 1743 btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 1744 memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 1745 btrfs_check_and_init_root_item(new_root_item); 1746 1747 root_flags = btrfs_root_flags(new_root_item); 1748 if (pending->readonly) 1749 root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 1750 else 1751 root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 1752 btrfs_set_root_flags(new_root_item, root_flags); 1753 1754 btrfs_set_root_generation_v2(new_root_item, 1755 trans->transid); 1756 generate_random_guid(new_root_item->uuid); 1757 memcpy(new_root_item->parent_uuid, root->root_item.uuid, 1758 BTRFS_UUID_SIZE); 1759 if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { 1760 memset(new_root_item->received_uuid, 0, 1761 sizeof(new_root_item->received_uuid)); 1762 memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); 1763 memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); 1764 btrfs_set_root_stransid(new_root_item, 0); 1765 btrfs_set_root_rtransid(new_root_item, 0); 1766 } 1767 btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); 1768 btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); 1769 btrfs_set_root_otransid(new_root_item, trans->transid); 1770 1771 old = btrfs_lock_root_node(root); 1772 ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, 1773 BTRFS_NESTING_COW); 1774 if (ret) { 1775 btrfs_tree_unlock(old); 1776 free_extent_buffer(old); 1777 btrfs_abort_transaction(trans, ret); 1778 goto fail; 1779 } 1780 1781 ret = btrfs_copy_root(trans, root, old, &tmp, objectid); 1782 /* clean up in any case */ 1783 btrfs_tree_unlock(old); 1784 free_extent_buffer(old); 1785 if (ret) { 1786 btrfs_abort_transaction(trans, ret); 1787 goto fail; 1788 } 1789 /* see comments in should_cow_block() */ 1790 set_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1791 smp_wmb(); 1792 1793 btrfs_set_root_node(new_root_item, tmp); 1794 /* record when the snapshot was created in key.offset */ 1795 key.offset = trans->transid; 1796 ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1797 btrfs_tree_unlock(tmp); 1798 free_extent_buffer(tmp); 1799 if (ret) { 1800 btrfs_abort_transaction(trans, ret); 1801 goto fail; 1802 } 1803 1804 /* 1805 * insert root back/forward references 1806 */ 1807 ret = btrfs_add_root_ref(trans, objectid, 1808 parent_root->root_key.objectid, 1809 btrfs_ino(BTRFS_I(parent_inode)), index, 1810 &fname.disk_name); 1811 if (ret) { 1812 btrfs_abort_transaction(trans, ret); 1813 goto fail; 1814 } 1815 1816 key.offset = (u64)-1; 1817 pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); 1818 if (IS_ERR(pending->snap)) { 1819 ret = PTR_ERR(pending->snap); 1820 pending->snap = NULL; 1821 btrfs_abort_transaction(trans, ret); 1822 goto fail; 1823 } 1824 1825 ret = btrfs_reloc_post_snapshot(trans, pending); 1826 if (ret) { 1827 btrfs_abort_transaction(trans, ret); 1828 goto fail; 1829 } 1830 1831 /* 1832 * Do special qgroup accounting for snapshot, as we do some qgroup 1833 * snapshot hack to do fast snapshot. 1834 * To co-operate with that hack, we do hack again. 1835 * Or snapshot will be greatly slowed down by a subtree qgroup rescan 1836 */ 1837 ret = qgroup_account_snapshot(trans, root, parent_root, 1838 pending->inherit, objectid); 1839 if (ret < 0) 1840 goto fail; 1841 1842 ret = btrfs_insert_dir_item(trans, &fname.disk_name, 1843 BTRFS_I(parent_inode), &key, BTRFS_FT_DIR, 1844 index); 1845 /* We have check then name at the beginning, so it is impossible. */ 1846 BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); 1847 if (ret) { 1848 btrfs_abort_transaction(trans, ret); 1849 goto fail; 1850 } 1851 1852 btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + 1853 fname.disk_name.len * 2); 1854 parent_inode->i_mtime = current_time(parent_inode); 1855 parent_inode->i_ctime = parent_inode->i_mtime; 1856 ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); 1857 if (ret) { 1858 btrfs_abort_transaction(trans, ret); 1859 goto fail; 1860 } 1861 ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, 1862 BTRFS_UUID_KEY_SUBVOL, 1863 objectid); 1864 if (ret) { 1865 btrfs_abort_transaction(trans, ret); 1866 goto fail; 1867 } 1868 if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { 1869 ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, 1870 BTRFS_UUID_KEY_RECEIVED_SUBVOL, 1871 objectid); 1872 if (ret && ret != -EEXIST) { 1873 btrfs_abort_transaction(trans, ret); 1874 goto fail; 1875 } 1876 } 1877 1878 fail: 1879 pending->error = ret; 1880 dir_item_existed: 1881 trans->block_rsv = rsv; 1882 trans->bytes_reserved = 0; 1883 clear_skip_qgroup: 1884 btrfs_clear_skip_qgroup(trans); 1885 free_fname: 1886 fscrypt_free_filename(&fname); 1887 free_pending: 1888 kfree(new_root_item); 1889 pending->root_item = NULL; 1890 btrfs_free_path(path); 1891 pending->path = NULL; 1892 1893 return ret; 1894 } 1895 1896 /* 1897 * create all the snapshots we've scheduled for creation 1898 */ 1899 static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) 1900 { 1901 struct btrfs_pending_snapshot *pending, *next; 1902 struct list_head *head = &trans->transaction->pending_snapshots; 1903 int ret = 0; 1904 1905 list_for_each_entry_safe(pending, next, head, list) { 1906 list_del(&pending->list); 1907 ret = create_pending_snapshot(trans, pending); 1908 if (ret) 1909 break; 1910 } 1911 return ret; 1912 } 1913 1914 static void update_super_roots(struct btrfs_fs_info *fs_info) 1915 { 1916 struct btrfs_root_item *root_item; 1917 struct btrfs_super_block *super; 1918 1919 super = fs_info->super_copy; 1920 1921 root_item = &fs_info->chunk_root->root_item; 1922 super->chunk_root = root_item->bytenr; 1923 super->chunk_root_generation = root_item->generation; 1924 super->chunk_root_level = root_item->level; 1925 1926 root_item = &fs_info->tree_root->root_item; 1927 super->root = root_item->bytenr; 1928 super->generation = root_item->generation; 1929 super->root_level = root_item->level; 1930 if (btrfs_test_opt(fs_info, SPACE_CACHE)) 1931 super->cache_generation = root_item->generation; 1932 else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) 1933 super->cache_generation = 0; 1934 if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) 1935 super->uuid_tree_generation = root_item->generation; 1936 } 1937 1938 int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1939 { 1940 struct btrfs_transaction *trans; 1941 int ret = 0; 1942 1943 spin_lock(&info->trans_lock); 1944 trans = info->running_transaction; 1945 if (trans) 1946 ret = (trans->state >= TRANS_STATE_COMMIT_START); 1947 spin_unlock(&info->trans_lock); 1948 return ret; 1949 } 1950 1951 int btrfs_transaction_blocked(struct btrfs_fs_info *info) 1952 { 1953 struct btrfs_transaction *trans; 1954 int ret = 0; 1955 1956 spin_lock(&info->trans_lock); 1957 trans = info->running_transaction; 1958 if (trans) 1959 ret = is_transaction_blocked(trans); 1960 spin_unlock(&info->trans_lock); 1961 return ret; 1962 } 1963 1964 void btrfs_commit_transaction_async(struct btrfs_trans_handle *trans) 1965 { 1966 struct btrfs_fs_info *fs_info = trans->fs_info; 1967 struct btrfs_transaction *cur_trans; 1968 1969 /* Kick the transaction kthread. */ 1970 set_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 1971 wake_up_process(fs_info->transaction_kthread); 1972 1973 /* take transaction reference */ 1974 cur_trans = trans->transaction; 1975 refcount_inc(&cur_trans->use_count); 1976 1977 btrfs_end_transaction(trans); 1978 1979 /* 1980 * Wait for the current transaction commit to start and block 1981 * subsequent transaction joins 1982 */ 1983 btrfs_might_wait_for_state(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 1984 wait_event(fs_info->transaction_blocked_wait, 1985 cur_trans->state >= TRANS_STATE_COMMIT_START || 1986 TRANS_ABORTED(cur_trans)); 1987 btrfs_put_transaction(cur_trans); 1988 } 1989 1990 static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) 1991 { 1992 struct btrfs_fs_info *fs_info = trans->fs_info; 1993 struct btrfs_transaction *cur_trans = trans->transaction; 1994 1995 WARN_ON(refcount_read(&trans->use_count) > 1); 1996 1997 btrfs_abort_transaction(trans, err); 1998 1999 spin_lock(&fs_info->trans_lock); 2000 2001 /* 2002 * If the transaction is removed from the list, it means this 2003 * transaction has been committed successfully, so it is impossible 2004 * to call the cleanup function. 2005 */ 2006 BUG_ON(list_empty(&cur_trans->list)); 2007 2008 if (cur_trans == fs_info->running_transaction) { 2009 cur_trans->state = TRANS_STATE_COMMIT_DOING; 2010 spin_unlock(&fs_info->trans_lock); 2011 2012 /* 2013 * The thread has already released the lockdep map as reader 2014 * already in btrfs_commit_transaction(). 2015 */ 2016 btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 2017 wait_event(cur_trans->writer_wait, 2018 atomic_read(&cur_trans->num_writers) == 1); 2019 2020 spin_lock(&fs_info->trans_lock); 2021 } 2022 2023 /* 2024 * Now that we know no one else is still using the transaction we can 2025 * remove the transaction from the list of transactions. This avoids 2026 * the transaction kthread from cleaning up the transaction while some 2027 * other task is still using it, which could result in a use-after-free 2028 * on things like log trees, as it forces the transaction kthread to 2029 * wait for this transaction to be cleaned up by us. 2030 */ 2031 list_del_init(&cur_trans->list); 2032 2033 spin_unlock(&fs_info->trans_lock); 2034 2035 btrfs_cleanup_one_transaction(trans->transaction, fs_info); 2036 2037 spin_lock(&fs_info->trans_lock); 2038 if (cur_trans == fs_info->running_transaction) 2039 fs_info->running_transaction = NULL; 2040 spin_unlock(&fs_info->trans_lock); 2041 2042 if (trans->type & __TRANS_FREEZABLE) 2043 sb_end_intwrite(fs_info->sb); 2044 btrfs_put_transaction(cur_trans); 2045 btrfs_put_transaction(cur_trans); 2046 2047 trace_btrfs_transaction_commit(fs_info); 2048 2049 if (current->journal_info == trans) 2050 current->journal_info = NULL; 2051 2052 /* 2053 * If relocation is running, we can't cancel scrub because that will 2054 * result in a deadlock. Before relocating a block group, relocation 2055 * pauses scrub, then starts and commits a transaction before unpausing 2056 * scrub. If the transaction commit is being done by the relocation 2057 * task or triggered by another task and the relocation task is waiting 2058 * for the commit, and we end up here due to an error in the commit 2059 * path, then calling btrfs_scrub_cancel() will deadlock, as we are 2060 * asking for scrub to stop while having it asked to be paused higher 2061 * above in relocation code. 2062 */ 2063 if (!test_bit(BTRFS_FS_RELOC_RUNNING, &fs_info->flags)) 2064 btrfs_scrub_cancel(fs_info); 2065 2066 kmem_cache_free(btrfs_trans_handle_cachep, trans); 2067 } 2068 2069 /* 2070 * Release reserved delayed ref space of all pending block groups of the 2071 * transaction and remove them from the list 2072 */ 2073 static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) 2074 { 2075 struct btrfs_fs_info *fs_info = trans->fs_info; 2076 struct btrfs_block_group *block_group, *tmp; 2077 2078 list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { 2079 btrfs_delayed_refs_rsv_release(fs_info, 1); 2080 list_del_init(&block_group->bg_list); 2081 } 2082 } 2083 2084 static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) 2085 { 2086 /* 2087 * We use try_to_writeback_inodes_sb() here because if we used 2088 * btrfs_start_delalloc_roots we would deadlock with fs freeze. 2089 * Currently are holding the fs freeze lock, if we do an async flush 2090 * we'll do btrfs_join_transaction() and deadlock because we need to 2091 * wait for the fs freeze lock. Using the direct flushing we benefit 2092 * from already being in a transaction and our join_transaction doesn't 2093 * have to re-take the fs freeze lock. 2094 * 2095 * Note that try_to_writeback_inodes_sb() will only trigger writeback 2096 * if it can read lock sb->s_umount. It will always be able to lock it, 2097 * except when the filesystem is being unmounted or being frozen, but in 2098 * those cases sync_filesystem() is called, which results in calling 2099 * writeback_inodes_sb() while holding a write lock on sb->s_umount. 2100 * Note that we don't call writeback_inodes_sb() directly, because it 2101 * will emit a warning if sb->s_umount is not locked. 2102 */ 2103 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 2104 try_to_writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); 2105 return 0; 2106 } 2107 2108 static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) 2109 { 2110 if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 2111 btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 2112 } 2113 2114 /* 2115 * Add a pending snapshot associated with the given transaction handle to the 2116 * respective handle. This must be called after the transaction commit started 2117 * and while holding fs_info->trans_lock. 2118 * This serves to guarantee a caller of btrfs_commit_transaction() that it can 2119 * safely free the pending snapshot pointer in case btrfs_commit_transaction() 2120 * returns an error. 2121 */ 2122 static void add_pending_snapshot(struct btrfs_trans_handle *trans) 2123 { 2124 struct btrfs_transaction *cur_trans = trans->transaction; 2125 2126 if (!trans->pending_snapshot) 2127 return; 2128 2129 lockdep_assert_held(&trans->fs_info->trans_lock); 2130 ASSERT(cur_trans->state >= TRANS_STATE_COMMIT_START); 2131 2132 list_add(&trans->pending_snapshot->list, &cur_trans->pending_snapshots); 2133 } 2134 2135 static void update_commit_stats(struct btrfs_fs_info *fs_info, ktime_t interval) 2136 { 2137 fs_info->commit_stats.commit_count++; 2138 fs_info->commit_stats.last_commit_dur = interval; 2139 fs_info->commit_stats.max_commit_dur = 2140 max_t(u64, fs_info->commit_stats.max_commit_dur, interval); 2141 fs_info->commit_stats.total_commit_dur += interval; 2142 } 2143 2144 int btrfs_commit_transaction(struct btrfs_trans_handle *trans) 2145 { 2146 struct btrfs_fs_info *fs_info = trans->fs_info; 2147 struct btrfs_transaction *cur_trans = trans->transaction; 2148 struct btrfs_transaction *prev_trans = NULL; 2149 int ret; 2150 ktime_t start_time; 2151 ktime_t interval; 2152 2153 ASSERT(refcount_read(&trans->use_count) == 1); 2154 btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 2155 2156 clear_bit(BTRFS_FS_NEED_TRANS_COMMIT, &fs_info->flags); 2157 2158 /* Stop the commit early if ->aborted is set */ 2159 if (TRANS_ABORTED(cur_trans)) { 2160 ret = cur_trans->aborted; 2161 goto lockdep_trans_commit_start_release; 2162 } 2163 2164 btrfs_trans_release_metadata(trans); 2165 trans->block_rsv = NULL; 2166 2167 /* 2168 * We only want one transaction commit doing the flushing so we do not 2169 * waste a bunch of time on lock contention on the extent root node. 2170 */ 2171 if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, 2172 &cur_trans->delayed_refs.flags)) { 2173 /* 2174 * Make a pass through all the delayed refs we have so far. 2175 * Any running threads may add more while we are here. 2176 */ 2177 ret = btrfs_run_delayed_refs(trans, 0); 2178 if (ret) 2179 goto lockdep_trans_commit_start_release; 2180 } 2181 2182 btrfs_create_pending_block_groups(trans); 2183 2184 if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { 2185 int run_it = 0; 2186 2187 /* this mutex is also taken before trying to set 2188 * block groups readonly. We need to make sure 2189 * that nobody has set a block group readonly 2190 * after a extents from that block group have been 2191 * allocated for cache files. btrfs_set_block_group_ro 2192 * will wait for the transaction to commit if it 2193 * finds BTRFS_TRANS_DIRTY_BG_RUN set. 2194 * 2195 * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure 2196 * only one process starts all the block group IO. It wouldn't 2197 * hurt to have more than one go through, but there's no 2198 * real advantage to it either. 2199 */ 2200 mutex_lock(&fs_info->ro_block_group_mutex); 2201 if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, 2202 &cur_trans->flags)) 2203 run_it = 1; 2204 mutex_unlock(&fs_info->ro_block_group_mutex); 2205 2206 if (run_it) { 2207 ret = btrfs_start_dirty_block_groups(trans); 2208 if (ret) 2209 goto lockdep_trans_commit_start_release; 2210 } 2211 } 2212 2213 spin_lock(&fs_info->trans_lock); 2214 if (cur_trans->state >= TRANS_STATE_COMMIT_START) { 2215 enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2216 2217 add_pending_snapshot(trans); 2218 2219 spin_unlock(&fs_info->trans_lock); 2220 refcount_inc(&cur_trans->use_count); 2221 2222 if (trans->in_fsync) 2223 want_state = TRANS_STATE_SUPER_COMMITTED; 2224 2225 btrfs_trans_state_lockdep_release(fs_info, 2226 BTRFS_LOCKDEP_TRANS_COMMIT_START); 2227 ret = btrfs_end_transaction(trans); 2228 wait_for_commit(cur_trans, want_state); 2229 2230 if (TRANS_ABORTED(cur_trans)) 2231 ret = cur_trans->aborted; 2232 2233 btrfs_put_transaction(cur_trans); 2234 2235 return ret; 2236 } 2237 2238 cur_trans->state = TRANS_STATE_COMMIT_START; 2239 wake_up(&fs_info->transaction_blocked_wait); 2240 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 2241 2242 if (cur_trans->list.prev != &fs_info->trans_list) { 2243 enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2244 2245 if (trans->in_fsync) 2246 want_state = TRANS_STATE_SUPER_COMMITTED; 2247 2248 prev_trans = list_entry(cur_trans->list.prev, 2249 struct btrfs_transaction, list); 2250 if (prev_trans->state < want_state) { 2251 refcount_inc(&prev_trans->use_count); 2252 spin_unlock(&fs_info->trans_lock); 2253 2254 wait_for_commit(prev_trans, want_state); 2255 2256 ret = READ_ONCE(prev_trans->aborted); 2257 2258 btrfs_put_transaction(prev_trans); 2259 if (ret) 2260 goto lockdep_release; 2261 } else { 2262 spin_unlock(&fs_info->trans_lock); 2263 } 2264 } else { 2265 spin_unlock(&fs_info->trans_lock); 2266 /* 2267 * The previous transaction was aborted and was already removed 2268 * from the list of transactions at fs_info->trans_list. So we 2269 * abort to prevent writing a new superblock that reflects a 2270 * corrupt state (pointing to trees with unwritten nodes/leafs). 2271 */ 2272 if (BTRFS_FS_ERROR(fs_info)) { 2273 ret = -EROFS; 2274 goto lockdep_release; 2275 } 2276 } 2277 2278 /* 2279 * Get the time spent on the work done by the commit thread and not 2280 * the time spent waiting on a previous commit 2281 */ 2282 start_time = ktime_get_ns(); 2283 2284 extwriter_counter_dec(cur_trans, trans->type); 2285 2286 ret = btrfs_start_delalloc_flush(fs_info); 2287 if (ret) 2288 goto lockdep_release; 2289 2290 ret = btrfs_run_delayed_items(trans); 2291 if (ret) 2292 goto lockdep_release; 2293 2294 /* 2295 * The thread has started/joined the transaction thus it holds the 2296 * lockdep map as a reader. It has to release it before acquiring the 2297 * lockdep map as a writer. 2298 */ 2299 btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 2300 btrfs_might_wait_for_event(fs_info, btrfs_trans_num_extwriters); 2301 wait_event(cur_trans->writer_wait, 2302 extwriter_counter_read(cur_trans) == 0); 2303 2304 /* some pending stuffs might be added after the previous flush. */ 2305 ret = btrfs_run_delayed_items(trans); 2306 if (ret) { 2307 btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2308 goto cleanup_transaction; 2309 } 2310 2311 btrfs_wait_delalloc_flush(fs_info); 2312 2313 /* 2314 * Wait for all ordered extents started by a fast fsync that joined this 2315 * transaction. Otherwise if this transaction commits before the ordered 2316 * extents complete we lose logged data after a power failure. 2317 */ 2318 btrfs_might_wait_for_event(fs_info, btrfs_trans_pending_ordered); 2319 wait_event(cur_trans->pending_wait, 2320 atomic_read(&cur_trans->pending_ordered) == 0); 2321 2322 btrfs_scrub_pause(fs_info); 2323 /* 2324 * Ok now we need to make sure to block out any other joins while we 2325 * commit the transaction. We could have started a join before setting 2326 * COMMIT_DOING so make sure to wait for num_writers to == 1 again. 2327 */ 2328 spin_lock(&fs_info->trans_lock); 2329 add_pending_snapshot(trans); 2330 cur_trans->state = TRANS_STATE_COMMIT_DOING; 2331 spin_unlock(&fs_info->trans_lock); 2332 2333 /* 2334 * The thread has started/joined the transaction thus it holds the 2335 * lockdep map as a reader. It has to release it before acquiring the 2336 * lockdep map as a writer. 2337 */ 2338 btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2339 btrfs_might_wait_for_event(fs_info, btrfs_trans_num_writers); 2340 wait_event(cur_trans->writer_wait, 2341 atomic_read(&cur_trans->num_writers) == 1); 2342 2343 /* 2344 * Make lockdep happy by acquiring the state locks after 2345 * btrfs_trans_num_writers is released. If we acquired the state locks 2346 * before releasing the btrfs_trans_num_writers lock then lockdep would 2347 * complain because we did not follow the reverse order unlocking rule. 2348 */ 2349 btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 2350 btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 2351 btrfs_trans_state_lockdep_acquire(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 2352 2353 /* 2354 * We've started the commit, clear the flag in case we were triggered to 2355 * do an async commit but somebody else started before the transaction 2356 * kthread could do the work. 2357 */ 2358 clear_bit(BTRFS_FS_COMMIT_TRANS, &fs_info->flags); 2359 2360 if (TRANS_ABORTED(cur_trans)) { 2361 ret = cur_trans->aborted; 2362 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 2363 goto scrub_continue; 2364 } 2365 /* 2366 * the reloc mutex makes sure that we stop 2367 * the balancing code from coming in and moving 2368 * extents around in the middle of the commit 2369 */ 2370 mutex_lock(&fs_info->reloc_mutex); 2371 2372 /* 2373 * We needn't worry about the delayed items because we will 2374 * deal with them in create_pending_snapshot(), which is the 2375 * core function of the snapshot creation. 2376 */ 2377 ret = create_pending_snapshots(trans); 2378 if (ret) 2379 goto unlock_reloc; 2380 2381 /* 2382 * We insert the dir indexes of the snapshots and update the inode 2383 * of the snapshots' parents after the snapshot creation, so there 2384 * are some delayed items which are not dealt with. Now deal with 2385 * them. 2386 * 2387 * We needn't worry that this operation will corrupt the snapshots, 2388 * because all the tree which are snapshoted will be forced to COW 2389 * the nodes and leaves. 2390 */ 2391 ret = btrfs_run_delayed_items(trans); 2392 if (ret) 2393 goto unlock_reloc; 2394 2395 ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 2396 if (ret) 2397 goto unlock_reloc; 2398 2399 /* 2400 * make sure none of the code above managed to slip in a 2401 * delayed item 2402 */ 2403 btrfs_assert_delayed_root_empty(fs_info); 2404 2405 WARN_ON(cur_trans != trans->transaction); 2406 2407 ret = commit_fs_roots(trans); 2408 if (ret) 2409 goto unlock_reloc; 2410 2411 /* commit_fs_roots gets rid of all the tree log roots, it is now 2412 * safe to free the root of tree log roots 2413 */ 2414 btrfs_free_log_root_tree(trans, fs_info); 2415 2416 /* 2417 * Since fs roots are all committed, we can get a quite accurate 2418 * new_roots. So let's do quota accounting. 2419 */ 2420 ret = btrfs_qgroup_account_extents(trans); 2421 if (ret < 0) 2422 goto unlock_reloc; 2423 2424 ret = commit_cowonly_roots(trans); 2425 if (ret) 2426 goto unlock_reloc; 2427 2428 /* 2429 * The tasks which save the space cache and inode cache may also 2430 * update ->aborted, check it. 2431 */ 2432 if (TRANS_ABORTED(cur_trans)) { 2433 ret = cur_trans->aborted; 2434 goto unlock_reloc; 2435 } 2436 2437 cur_trans = fs_info->running_transaction; 2438 2439 btrfs_set_root_node(&fs_info->tree_root->root_item, 2440 fs_info->tree_root->node); 2441 list_add_tail(&fs_info->tree_root->dirty_list, 2442 &cur_trans->switch_commits); 2443 2444 btrfs_set_root_node(&fs_info->chunk_root->root_item, 2445 fs_info->chunk_root->node); 2446 list_add_tail(&fs_info->chunk_root->dirty_list, 2447 &cur_trans->switch_commits); 2448 2449 if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) { 2450 btrfs_set_root_node(&fs_info->block_group_root->root_item, 2451 fs_info->block_group_root->node); 2452 list_add_tail(&fs_info->block_group_root->dirty_list, 2453 &cur_trans->switch_commits); 2454 } 2455 2456 switch_commit_roots(trans); 2457 2458 ASSERT(list_empty(&cur_trans->dirty_bgs)); 2459 ASSERT(list_empty(&cur_trans->io_bgs)); 2460 update_super_roots(fs_info); 2461 2462 btrfs_set_super_log_root(fs_info->super_copy, 0); 2463 btrfs_set_super_log_root_level(fs_info->super_copy, 0); 2464 memcpy(fs_info->super_for_commit, fs_info->super_copy, 2465 sizeof(*fs_info->super_copy)); 2466 2467 btrfs_commit_device_sizes(cur_trans); 2468 2469 clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); 2470 clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); 2471 2472 btrfs_trans_release_chunk_metadata(trans); 2473 2474 /* 2475 * Before changing the transaction state to TRANS_STATE_UNBLOCKED and 2476 * setting fs_info->running_transaction to NULL, lock tree_log_mutex to 2477 * make sure that before we commit our superblock, no other task can 2478 * start a new transaction and commit a log tree before we commit our 2479 * superblock. Anyone trying to commit a log tree locks this mutex before 2480 * writing its superblock. 2481 */ 2482 mutex_lock(&fs_info->tree_log_mutex); 2483 2484 spin_lock(&fs_info->trans_lock); 2485 cur_trans->state = TRANS_STATE_UNBLOCKED; 2486 fs_info->running_transaction = NULL; 2487 spin_unlock(&fs_info->trans_lock); 2488 mutex_unlock(&fs_info->reloc_mutex); 2489 2490 wake_up(&fs_info->transaction_wait); 2491 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 2492 2493 /* If we have features changed, wake up the cleaner to update sysfs. */ 2494 if (test_bit(BTRFS_FS_FEATURE_CHANGED, &fs_info->flags) && 2495 fs_info->cleaner_kthread) 2496 wake_up_process(fs_info->cleaner_kthread); 2497 2498 ret = btrfs_write_and_wait_transaction(trans); 2499 if (ret) { 2500 btrfs_handle_fs_error(fs_info, ret, 2501 "Error while writing out transaction"); 2502 mutex_unlock(&fs_info->tree_log_mutex); 2503 goto scrub_continue; 2504 } 2505 2506 ret = write_all_supers(fs_info, 0); 2507 /* 2508 * the super is written, we can safely allow the tree-loggers 2509 * to go about their business 2510 */ 2511 mutex_unlock(&fs_info->tree_log_mutex); 2512 if (ret) 2513 goto scrub_continue; 2514 2515 /* 2516 * We needn't acquire the lock here because there is no other task 2517 * which can change it. 2518 */ 2519 cur_trans->state = TRANS_STATE_SUPER_COMMITTED; 2520 wake_up(&cur_trans->commit_wait); 2521 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 2522 2523 btrfs_finish_extent_commit(trans); 2524 2525 if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) 2526 btrfs_clear_space_info_full(fs_info); 2527 2528 fs_info->last_trans_committed = cur_trans->transid; 2529 /* 2530 * We needn't acquire the lock here because there is no other task 2531 * which can change it. 2532 */ 2533 cur_trans->state = TRANS_STATE_COMPLETED; 2534 wake_up(&cur_trans->commit_wait); 2535 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 2536 2537 spin_lock(&fs_info->trans_lock); 2538 list_del_init(&cur_trans->list); 2539 spin_unlock(&fs_info->trans_lock); 2540 2541 btrfs_put_transaction(cur_trans); 2542 btrfs_put_transaction(cur_trans); 2543 2544 if (trans->type & __TRANS_FREEZABLE) 2545 sb_end_intwrite(fs_info->sb); 2546 2547 trace_btrfs_transaction_commit(fs_info); 2548 2549 interval = ktime_get_ns() - start_time; 2550 2551 btrfs_scrub_continue(fs_info); 2552 2553 if (current->journal_info == trans) 2554 current->journal_info = NULL; 2555 2556 kmem_cache_free(btrfs_trans_handle_cachep, trans); 2557 2558 update_commit_stats(fs_info, interval); 2559 2560 return ret; 2561 2562 unlock_reloc: 2563 mutex_unlock(&fs_info->reloc_mutex); 2564 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_UNBLOCKED); 2565 scrub_continue: 2566 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_SUPER_COMMITTED); 2567 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMPLETED); 2568 btrfs_scrub_continue(fs_info); 2569 cleanup_transaction: 2570 btrfs_trans_release_metadata(trans); 2571 btrfs_cleanup_pending_block_groups(trans); 2572 btrfs_trans_release_chunk_metadata(trans); 2573 trans->block_rsv = NULL; 2574 btrfs_warn(fs_info, "Skipping commit of aborted transaction."); 2575 if (current->journal_info == trans) 2576 current->journal_info = NULL; 2577 cleanup_transaction(trans, ret); 2578 2579 return ret; 2580 2581 lockdep_release: 2582 btrfs_lockdep_release(fs_info, btrfs_trans_num_extwriters); 2583 btrfs_lockdep_release(fs_info, btrfs_trans_num_writers); 2584 goto cleanup_transaction; 2585 2586 lockdep_trans_commit_start_release: 2587 btrfs_trans_state_lockdep_release(fs_info, BTRFS_LOCKDEP_TRANS_COMMIT_START); 2588 btrfs_end_transaction(trans); 2589 return ret; 2590 } 2591 2592 /* 2593 * return < 0 if error 2594 * 0 if there are no more dead_roots at the time of call 2595 * 1 there are more to be processed, call me again 2596 * 2597 * The return value indicates there are certainly more snapshots to delete, but 2598 * if there comes a new one during processing, it may return 0. We don't mind, 2599 * because btrfs_commit_super will poke cleaner thread and it will process it a 2600 * few seconds later. 2601 */ 2602 int btrfs_clean_one_deleted_snapshot(struct btrfs_fs_info *fs_info) 2603 { 2604 struct btrfs_root *root; 2605 int ret; 2606 2607 spin_lock(&fs_info->trans_lock); 2608 if (list_empty(&fs_info->dead_roots)) { 2609 spin_unlock(&fs_info->trans_lock); 2610 return 0; 2611 } 2612 root = list_first_entry(&fs_info->dead_roots, 2613 struct btrfs_root, root_list); 2614 list_del_init(&root->root_list); 2615 spin_unlock(&fs_info->trans_lock); 2616 2617 btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); 2618 2619 btrfs_kill_all_delayed_nodes(root); 2620 2621 if (btrfs_header_backref_rev(root->node) < 2622 BTRFS_MIXED_BACKREF_REV) 2623 ret = btrfs_drop_snapshot(root, 0, 0); 2624 else 2625 ret = btrfs_drop_snapshot(root, 1, 0); 2626 2627 btrfs_put_root(root); 2628 return (ret < 0) ? 0 : 1; 2629 } 2630 2631 /* 2632 * We only mark the transaction aborted and then set the file system read-only. 2633 * This will prevent new transactions from starting or trying to join this 2634 * one. 2635 * 2636 * This means that error recovery at the call site is limited to freeing 2637 * any local memory allocations and passing the error code up without 2638 * further cleanup. The transaction should complete as it normally would 2639 * in the call path but will return -EIO. 2640 * 2641 * We'll complete the cleanup in btrfs_end_transaction and 2642 * btrfs_commit_transaction. 2643 */ 2644 void __cold __btrfs_abort_transaction(struct btrfs_trans_handle *trans, 2645 const char *function, 2646 unsigned int line, int errno, bool first_hit) 2647 { 2648 struct btrfs_fs_info *fs_info = trans->fs_info; 2649 2650 WRITE_ONCE(trans->aborted, errno); 2651 WRITE_ONCE(trans->transaction->aborted, errno); 2652 if (first_hit && errno == -ENOSPC) 2653 btrfs_dump_space_info_for_trans_abort(fs_info); 2654 /* Wake up anybody who may be waiting on this transaction */ 2655 wake_up(&fs_info->transaction_wait); 2656 wake_up(&fs_info->transaction_blocked_wait); 2657 __btrfs_handle_fs_error(fs_info, function, line, errno, NULL); 2658 } 2659 2660 int __init btrfs_transaction_init(void) 2661 { 2662 btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle", 2663 sizeof(struct btrfs_trans_handle), 0, 2664 SLAB_TEMPORARY | SLAB_MEM_SPREAD, NULL); 2665 if (!btrfs_trans_handle_cachep) 2666 return -ENOMEM; 2667 return 0; 2668 } 2669 2670 void __cold btrfs_transaction_exit(void) 2671 { 2672 kmem_cache_destroy(btrfs_trans_handle_cachep); 2673 } 2674