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