1c1d7c514SDavid Sterba // SPDX-License-Identifier: GPL-2.0 26cbd5570SChris Mason /* 36cbd5570SChris Mason * Copyright (C) 2007 Oracle. All rights reserved. 46cbd5570SChris Mason */ 56cbd5570SChris Mason 679154b1bSChris Mason #include <linux/fs.h> 75a0e3ad6STejun Heo #include <linux/slab.h> 834088780SChris Mason #include <linux/sched.h> 9d3c2fdcfSChris Mason #include <linux/writeback.h> 105f39d397SChris Mason #include <linux/pagemap.h> 115f2cc086SChris Mason #include <linux/blkdev.h> 128ea05e3aSAlexander Block #include <linux/uuid.h> 13602cbe91SDavid Sterba #include "misc.h" 1479154b1bSChris Mason #include "ctree.h" 1579154b1bSChris Mason #include "disk-io.h" 1679154b1bSChris Mason #include "transaction.h" 17925baeddSChris Mason #include "locking.h" 18e02119d5SChris Mason #include "tree-log.h" 19733f4fbbSStefan Behrens #include "volumes.h" 208dabb742SStefan Behrens #include "dev-replace.h" 21fcebe456SJosef Bacik #include "qgroup.h" 22aac0023cSJosef Bacik #include "block-group.h" 239c343784SJosef Bacik #include "space-info.h" 24d3575156SNaohiro Aota #include "zoned.h" 2579154b1bSChris Mason 260f7d52f4SChris Mason #define BTRFS_ROOT_TRANS_TAG 0 270f7d52f4SChris Mason 2861c047b5SQu Wenruo /* 2961c047b5SQu Wenruo * Transaction states and transitions 3061c047b5SQu Wenruo * 3161c047b5SQu Wenruo * No running transaction (fs tree blocks are not modified) 3261c047b5SQu Wenruo * | 3361c047b5SQu Wenruo * | To next stage: 3461c047b5SQu Wenruo * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). 3561c047b5SQu Wenruo * V 3661c047b5SQu Wenruo * Transaction N [[TRANS_STATE_RUNNING]] 3761c047b5SQu Wenruo * | 3861c047b5SQu Wenruo * | New trans handles can be attached to transaction N by calling all 3961c047b5SQu Wenruo * | start_transaction() variants. 4061c047b5SQu Wenruo * | 4161c047b5SQu Wenruo * | To next stage: 4261c047b5SQu Wenruo * | Call btrfs_commit_transaction() on any trans handle attached to 4361c047b5SQu Wenruo * | transaction N 4461c047b5SQu Wenruo * V 4561c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_START]] 4661c047b5SQu Wenruo * | 4761c047b5SQu Wenruo * | Will wait for previous running transaction to completely finish if there 4861c047b5SQu Wenruo * | is one 4961c047b5SQu Wenruo * | 5061c047b5SQu Wenruo * | Then one of the following happes: 5161c047b5SQu Wenruo * | - Wait for all other trans handle holders to release. 5261c047b5SQu Wenruo * | The btrfs_commit_transaction() caller will do the commit work. 5361c047b5SQu Wenruo * | - Wait for current transaction to be committed by others. 5461c047b5SQu Wenruo * | Other btrfs_commit_transaction() caller will do the commit work. 5561c047b5SQu Wenruo * | 5661c047b5SQu Wenruo * | At this stage, only btrfs_join_transaction*() variants can attach 5761c047b5SQu Wenruo * | to this running transaction. 5861c047b5SQu Wenruo * | All other variants will wait for current one to finish and attach to 5961c047b5SQu Wenruo * | transaction N+1. 6061c047b5SQu Wenruo * | 6161c047b5SQu Wenruo * | To next stage: 6261c047b5SQu Wenruo * | Caller is chosen to commit transaction N, and all other trans handle 6361c047b5SQu Wenruo * | haven been released. 6461c047b5SQu Wenruo * V 6561c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_DOING]] 6661c047b5SQu Wenruo * | 6761c047b5SQu Wenruo * | The heavy lifting transaction work is started. 6861c047b5SQu Wenruo * | From running delayed refs (modifying extent tree) to creating pending 6961c047b5SQu Wenruo * | snapshots, running qgroups. 7061c047b5SQu Wenruo * | In short, modify supporting trees to reflect modifications of subvolume 7161c047b5SQu Wenruo * | trees. 7261c047b5SQu Wenruo * | 7361c047b5SQu Wenruo * | At this stage, all start_transaction() calls will wait for this 7461c047b5SQu Wenruo * | transaction to finish and attach to transaction N+1. 7561c047b5SQu Wenruo * | 7661c047b5SQu Wenruo * | To next stage: 7761c047b5SQu Wenruo * | Until all supporting trees are updated. 7861c047b5SQu Wenruo * V 7961c047b5SQu Wenruo * Transaction N [[TRANS_STATE_UNBLOCKED]] 8061c047b5SQu Wenruo * | Transaction N+1 8161c047b5SQu Wenruo * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] 8261c047b5SQu Wenruo * | need to write them back to disk and update | 8361c047b5SQu Wenruo * | super blocks. | 8461c047b5SQu Wenruo * | | 8561c047b5SQu Wenruo * | At this stage, new transaction is allowed to | 8661c047b5SQu Wenruo * | start. | 8761c047b5SQu Wenruo * | All new start_transaction() calls will be | 8861c047b5SQu Wenruo * | attached to transid N+1. | 8961c047b5SQu Wenruo * | | 9061c047b5SQu Wenruo * | To next stage: | 9161c047b5SQu Wenruo * | Until all tree blocks are super blocks are | 9261c047b5SQu Wenruo * | written to block devices | 9361c047b5SQu Wenruo * V | 9461c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMPLETED]] V 9561c047b5SQu Wenruo * All tree blocks and super blocks are written. Transaction N+1 9661c047b5SQu Wenruo * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] 9761c047b5SQu Wenruo * data structures will be cleaned up. | Life goes on 9861c047b5SQu Wenruo */ 99e8c9f186SDavid Sterba static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { 1004a9d8bdeSMiao Xie [TRANS_STATE_RUNNING] = 0U, 101bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), 102bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | 1034a9d8bdeSMiao Xie __TRANS_ATTACH | 104a6d155d2SFilipe Manana __TRANS_JOIN | 105a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 106bcf3a3e7SNikolay Borisov [TRANS_STATE_UNBLOCKED] = (__TRANS_START | 1074a9d8bdeSMiao Xie __TRANS_ATTACH | 1084a9d8bdeSMiao Xie __TRANS_JOIN | 109a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 110a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 111d0c2f4faSFilipe Manana [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | 112d0c2f4faSFilipe Manana __TRANS_ATTACH | 113d0c2f4faSFilipe Manana __TRANS_JOIN | 114d0c2f4faSFilipe Manana __TRANS_JOIN_NOLOCK | 115d0c2f4faSFilipe Manana __TRANS_JOIN_NOSTART), 116bcf3a3e7SNikolay Borisov [TRANS_STATE_COMPLETED] = (__TRANS_START | 1174a9d8bdeSMiao Xie __TRANS_ATTACH | 1184a9d8bdeSMiao Xie __TRANS_JOIN | 119a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 120a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 1214a9d8bdeSMiao Xie }; 1224a9d8bdeSMiao Xie 123724e2315SJosef Bacik void btrfs_put_transaction(struct btrfs_transaction *transaction) 12479154b1bSChris Mason { 1259b64f57dSElena Reshetova WARN_ON(refcount_read(&transaction->use_count) == 0); 1269b64f57dSElena Reshetova if (refcount_dec_and_test(&transaction->use_count)) { 127a4abeea4SJosef Bacik BUG_ON(!list_empty(&transaction->list)); 1285c9d028bSLiu Bo WARN_ON(!RB_EMPTY_ROOT( 1295c9d028bSLiu Bo &transaction->delayed_refs.href_root.rb_root)); 13081f7eb00SJeff Mahoney WARN_ON(!RB_EMPTY_ROOT( 13181f7eb00SJeff Mahoney &transaction->delayed_refs.dirty_extent_root)); 1321262133bSJosef Bacik if (transaction->delayed_refs.pending_csums) 133ab8d0fc4SJeff Mahoney btrfs_err(transaction->fs_info, 134ab8d0fc4SJeff Mahoney "pending csums is %llu", 1351262133bSJosef Bacik transaction->delayed_refs.pending_csums); 1367785a663SFilipe Manana /* 1377785a663SFilipe Manana * If any block groups are found in ->deleted_bgs then it's 1387785a663SFilipe Manana * because the transaction was aborted and a commit did not 1397785a663SFilipe Manana * happen (things failed before writing the new superblock 1407785a663SFilipe Manana * and calling btrfs_finish_extent_commit()), so we can not 1417785a663SFilipe Manana * discard the physical locations of the block groups. 1427785a663SFilipe Manana */ 1437785a663SFilipe Manana while (!list_empty(&transaction->deleted_bgs)) { 14432da5386SDavid Sterba struct btrfs_block_group *cache; 1457785a663SFilipe Manana 1467785a663SFilipe Manana cache = list_first_entry(&transaction->deleted_bgs, 14732da5386SDavid Sterba struct btrfs_block_group, 1487785a663SFilipe Manana bg_list); 1497785a663SFilipe Manana list_del_init(&cache->bg_list); 1506b7304afSFilipe Manana btrfs_unfreeze_block_group(cache); 1517785a663SFilipe Manana btrfs_put_block_group(cache); 1527785a663SFilipe Manana } 153bbbf7243SNikolay Borisov WARN_ON(!list_empty(&transaction->dev_update_list)); 1544b5faeacSDavid Sterba kfree(transaction); 15579154b1bSChris Mason } 15678fae27eSChris Mason } 15779154b1bSChris Mason 158889bfa39SJosef Bacik static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) 159817d52f8SJosef Bacik { 160889bfa39SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 16116916a88SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1629e351cc8SJosef Bacik struct btrfs_root *root, *tmp; 16327d56e62SJosef Bacik struct btrfs_caching_control *caching_ctl, *next; 1649e351cc8SJosef Bacik 1659e351cc8SJosef Bacik down_write(&fs_info->commit_root_sem); 166889bfa39SJosef Bacik list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, 1679e351cc8SJosef Bacik dirty_list) { 1689e351cc8SJosef Bacik list_del_init(&root->dirty_list); 169817d52f8SJosef Bacik free_extent_buffer(root->commit_root); 170817d52f8SJosef Bacik root->commit_root = btrfs_root_node(root); 17141e7acd3SNikolay Borisov extent_io_tree_release(&root->dirty_log_pages); 172370a11b8SQu Wenruo btrfs_qgroup_clean_swapped_blocks(root); 1739e351cc8SJosef Bacik } 1742b9dbef2SJosef Bacik 1752b9dbef2SJosef Bacik /* We can free old roots now. */ 176889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 177889bfa39SJosef Bacik while (!list_empty(&cur_trans->dropped_roots)) { 178889bfa39SJosef Bacik root = list_first_entry(&cur_trans->dropped_roots, 1792b9dbef2SJosef Bacik struct btrfs_root, root_list); 1802b9dbef2SJosef Bacik list_del_init(&root->root_list); 181889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 182889bfa39SJosef Bacik btrfs_free_log(trans, root); 1832b9dbef2SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, root); 184889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 1852b9dbef2SJosef Bacik } 186889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 18727d56e62SJosef Bacik 18827d56e62SJosef Bacik /* 18927d56e62SJosef Bacik * We have to update the last_byte_to_unpin under the commit_root_sem, 19027d56e62SJosef Bacik * at the same time we swap out the commit roots. 19127d56e62SJosef Bacik * 19227d56e62SJosef Bacik * This is because we must have a real view of the last spot the caching 19327d56e62SJosef Bacik * kthreads were while caching. Consider the following views of the 19427d56e62SJosef Bacik * extent tree for a block group 19527d56e62SJosef Bacik * 19627d56e62SJosef Bacik * commit root 19727d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 19827d56e62SJosef Bacik * |\\\\| |\\\\|\\\\| |\\\\|\\\\| 19927d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 20027d56e62SJosef Bacik * 0 1 2 3 4 5 6 7 20127d56e62SJosef Bacik * 20227d56e62SJosef Bacik * new commit root 20327d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 20427d56e62SJosef Bacik * | | | |\\\\| | |\\\\| 20527d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 20627d56e62SJosef Bacik * 0 1 2 3 4 5 6 7 20727d56e62SJosef Bacik * 20827d56e62SJosef Bacik * If the cache_ctl->progress was at 3, then we are only allowed to 20927d56e62SJosef Bacik * unpin [0,1) and [2,3], because the caching thread has already 21027d56e62SJosef Bacik * processed those extents. We are not allowed to unpin [5,6), because 21127d56e62SJosef Bacik * the caching thread will re-start it's search from 3, and thus find 21227d56e62SJosef Bacik * the hole from [4,6) to add to the free space cache. 21327d56e62SJosef Bacik */ 214bbb86a37SJosef Bacik spin_lock(&fs_info->block_group_cache_lock); 21527d56e62SJosef Bacik list_for_each_entry_safe(caching_ctl, next, 21627d56e62SJosef Bacik &fs_info->caching_block_groups, list) { 21727d56e62SJosef Bacik struct btrfs_block_group *cache = caching_ctl->block_group; 21827d56e62SJosef Bacik 21927d56e62SJosef Bacik if (btrfs_block_group_done(cache)) { 22027d56e62SJosef Bacik cache->last_byte_to_unpin = (u64)-1; 22127d56e62SJosef Bacik list_del_init(&caching_ctl->list); 22227d56e62SJosef Bacik btrfs_put_caching_control(caching_ctl); 22327d56e62SJosef Bacik } else { 22427d56e62SJosef Bacik cache->last_byte_to_unpin = caching_ctl->progress; 22527d56e62SJosef Bacik } 22627d56e62SJosef Bacik } 227bbb86a37SJosef Bacik spin_unlock(&fs_info->block_group_cache_lock); 2289e351cc8SJosef Bacik up_write(&fs_info->commit_root_sem); 229817d52f8SJosef Bacik } 230817d52f8SJosef Bacik 2310860adfdSMiao Xie static inline void extwriter_counter_inc(struct btrfs_transaction *trans, 2320860adfdSMiao Xie unsigned int type) 2330860adfdSMiao Xie { 2340860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2350860adfdSMiao Xie atomic_inc(&trans->num_extwriters); 2360860adfdSMiao Xie } 2370860adfdSMiao Xie 2380860adfdSMiao Xie static inline void extwriter_counter_dec(struct btrfs_transaction *trans, 2390860adfdSMiao Xie unsigned int type) 2400860adfdSMiao Xie { 2410860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2420860adfdSMiao Xie atomic_dec(&trans->num_extwriters); 2430860adfdSMiao Xie } 2440860adfdSMiao Xie 2450860adfdSMiao Xie static inline void extwriter_counter_init(struct btrfs_transaction *trans, 2460860adfdSMiao Xie unsigned int type) 2470860adfdSMiao Xie { 2480860adfdSMiao Xie atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); 2490860adfdSMiao Xie } 2500860adfdSMiao Xie 2510860adfdSMiao Xie static inline int extwriter_counter_read(struct btrfs_transaction *trans) 2520860adfdSMiao Xie { 2530860adfdSMiao Xie return atomic_read(&trans->num_extwriters); 254178260b2SMiao Xie } 255178260b2SMiao Xie 256d352ac68SChris Mason /* 257*79bd3712SFilipe Manana * To be called after doing the chunk btree updates right after allocating a new 258*79bd3712SFilipe Manana * chunk (after btrfs_chunk_alloc_add_chunk_item() is called), when removing a 259*79bd3712SFilipe Manana * chunk after all chunk btree updates and after finishing the second phase of 260*79bd3712SFilipe Manana * chunk allocation (btrfs_create_pending_block_groups()) in case some block 261*79bd3712SFilipe Manana * group had its chunk item insertion delayed to the second phase. 262fb6dea26SJosef Bacik */ 263fb6dea26SJosef Bacik void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) 264fb6dea26SJosef Bacik { 265fb6dea26SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 266fb6dea26SJosef Bacik 267fb6dea26SJosef Bacik if (!trans->chunk_bytes_reserved) 268fb6dea26SJosef Bacik return; 269fb6dea26SJosef Bacik 270fb6dea26SJosef Bacik btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, 27163f018beSNikolay Borisov trans->chunk_bytes_reserved, NULL); 272fb6dea26SJosef Bacik trans->chunk_bytes_reserved = 0; 273fb6dea26SJosef Bacik } 274fb6dea26SJosef Bacik 275fb6dea26SJosef Bacik /* 276d352ac68SChris Mason * either allocate a new transaction or hop into the existing one 277d352ac68SChris Mason */ 2782ff7e61eSJeff Mahoney static noinline int join_transaction(struct btrfs_fs_info *fs_info, 2792ff7e61eSJeff Mahoney unsigned int type) 28079154b1bSChris Mason { 28179154b1bSChris Mason struct btrfs_transaction *cur_trans; 282a4abeea4SJosef Bacik 28319ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 284d43317dcSChris Mason loop: 28549b25e05SJeff Mahoney /* The file system has been taken offline. No new transactions. */ 28687533c47SMiao Xie if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 28719ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 28849b25e05SJeff Mahoney return -EROFS; 28949b25e05SJeff Mahoney } 29049b25e05SJeff Mahoney 29119ae4e81SJan Schmidt cur_trans = fs_info->running_transaction; 292a4abeea4SJosef Bacik if (cur_trans) { 293bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 29419ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 29549b25e05SJeff Mahoney return cur_trans->aborted; 296871383beSDavid Sterba } 2974a9d8bdeSMiao Xie if (btrfs_blocked_trans_types[cur_trans->state] & type) { 298178260b2SMiao Xie spin_unlock(&fs_info->trans_lock); 299178260b2SMiao Xie return -EBUSY; 300178260b2SMiao Xie } 3019b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 302a4abeea4SJosef Bacik atomic_inc(&cur_trans->num_writers); 3030860adfdSMiao Xie extwriter_counter_inc(cur_trans, type); 30419ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 305a4abeea4SJosef Bacik return 0; 306a4abeea4SJosef Bacik } 30719ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 308a4abeea4SJosef Bacik 309354aa0fbSMiao Xie /* 310354aa0fbSMiao Xie * If we are ATTACH, we just want to catch the current transaction, 311354aa0fbSMiao Xie * and commit it. If there is no transaction, just return ENOENT. 312354aa0fbSMiao Xie */ 313354aa0fbSMiao Xie if (type == TRANS_ATTACH) 314354aa0fbSMiao Xie return -ENOENT; 315354aa0fbSMiao Xie 3164a9d8bdeSMiao Xie /* 3174a9d8bdeSMiao Xie * JOIN_NOLOCK only happens during the transaction commit, so 3184a9d8bdeSMiao Xie * it is impossible that ->running_transaction is NULL 3194a9d8bdeSMiao Xie */ 3204a9d8bdeSMiao Xie BUG_ON(type == TRANS_JOIN_NOLOCK); 3214a9d8bdeSMiao Xie 3224b5faeacSDavid Sterba cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); 323db5b493aSTsutomu Itoh if (!cur_trans) 324db5b493aSTsutomu Itoh return -ENOMEM; 325d43317dcSChris Mason 32619ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 32719ae4e81SJan Schmidt if (fs_info->running_transaction) { 328d43317dcSChris Mason /* 329d43317dcSChris Mason * someone started a transaction after we unlocked. Make sure 3304a9d8bdeSMiao Xie * to redo the checks above 331d43317dcSChris Mason */ 3324b5faeacSDavid Sterba kfree(cur_trans); 333d43317dcSChris Mason goto loop; 33487533c47SMiao Xie } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 335e4b50e14SDan Carpenter spin_unlock(&fs_info->trans_lock); 3364b5faeacSDavid Sterba kfree(cur_trans); 3377b8b92afSJosef Bacik return -EROFS; 338a4abeea4SJosef Bacik } 339d43317dcSChris Mason 340ab8d0fc4SJeff Mahoney cur_trans->fs_info = fs_info; 34148778179SFilipe Manana atomic_set(&cur_trans->pending_ordered, 0); 34248778179SFilipe Manana init_waitqueue_head(&cur_trans->pending_wait); 34313c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 3440860adfdSMiao Xie extwriter_counter_init(cur_trans, type); 34579154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 34679154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 3474a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_RUNNING; 348a4abeea4SJosef Bacik /* 349a4abeea4SJosef Bacik * One for this trans handle, one so it will live on until we 350a4abeea4SJosef Bacik * commit the transaction. 351a4abeea4SJosef Bacik */ 3529b64f57dSElena Reshetova refcount_set(&cur_trans->use_count, 2); 3533204d33cSJosef Bacik cur_trans->flags = 0; 354afd48513SArnd Bergmann cur_trans->start_time = ktime_get_seconds(); 35556bec294SChris Mason 356a099d0fdSAlexandru Moise memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); 357a099d0fdSAlexandru Moise 3585c9d028bSLiu Bo cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; 3593368d001SQu Wenruo cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; 360d7df2c79SJosef Bacik atomic_set(&cur_trans->delayed_refs.num_entries, 0); 36120b297d6SJan Schmidt 36220b297d6SJan Schmidt /* 36320b297d6SJan Schmidt * although the tree mod log is per file system and not per transaction, 36420b297d6SJan Schmidt * the log must never go across transaction boundaries. 36520b297d6SJan Schmidt */ 36620b297d6SJan Schmidt smp_mb(); 36731b1a2bdSJulia Lawall if (!list_empty(&fs_info->tree_mod_seq_list)) 3685d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); 36931b1a2bdSJulia Lawall if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) 3705d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); 371fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 37220b297d6SJan Schmidt 37356bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 37456bec294SChris Mason 3753063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 376bbbf7243SNikolay Borisov INIT_LIST_HEAD(&cur_trans->dev_update_list); 3779e351cc8SJosef Bacik INIT_LIST_HEAD(&cur_trans->switch_commits); 378ce93ec54SJosef Bacik INIT_LIST_HEAD(&cur_trans->dirty_bgs); 3791bbc621eSChris Mason INIT_LIST_HEAD(&cur_trans->io_bgs); 3802b9dbef2SJosef Bacik INIT_LIST_HEAD(&cur_trans->dropped_roots); 3811bbc621eSChris Mason mutex_init(&cur_trans->cache_write_mutex); 382ce93ec54SJosef Bacik spin_lock_init(&cur_trans->dirty_bgs_lock); 383e33e17eeSJeff Mahoney INIT_LIST_HEAD(&cur_trans->deleted_bgs); 3842b9dbef2SJosef Bacik spin_lock_init(&cur_trans->dropped_roots_lock); 385d3575156SNaohiro Aota INIT_LIST_HEAD(&cur_trans->releasing_ebs); 386d3575156SNaohiro Aota spin_lock_init(&cur_trans->releasing_ebs_lock); 38719ae4e81SJan Schmidt list_add_tail(&cur_trans->list, &fs_info->trans_list); 388c258d6e3SQu Wenruo extent_io_tree_init(fs_info, &cur_trans->dirty_pages, 38943eb5f29SQu Wenruo IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode); 390fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &cur_trans->pinned_extents, 391fe119a6eSNikolay Borisov IO_TREE_FS_PINNED_EXTENTS, NULL); 39219ae4e81SJan Schmidt fs_info->generation++; 39319ae4e81SJan Schmidt cur_trans->transid = fs_info->generation; 39419ae4e81SJan Schmidt fs_info->running_transaction = cur_trans; 39549b25e05SJeff Mahoney cur_trans->aborted = 0; 39619ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 39715ee9bc7SJosef Bacik 39879154b1bSChris Mason return 0; 39979154b1bSChris Mason } 40079154b1bSChris Mason 401d352ac68SChris Mason /* 40292a7cc42SQu Wenruo * This does all the record keeping required to make sure that a shareable root 40392a7cc42SQu Wenruo * is properly recorded in a given transaction. This is required to make sure 40492a7cc42SQu Wenruo * the old root from before we joined the transaction is deleted when the 40592a7cc42SQu Wenruo * transaction commits. 406d352ac68SChris Mason */ 4077585717fSChris Mason static int record_root_in_trans(struct btrfs_trans_handle *trans, 4086426c7adSQu Wenruo struct btrfs_root *root, 4096426c7adSQu Wenruo int force) 4106702ed49SChris Mason { 4110b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 41203a7e111SJosef Bacik int ret = 0; 4130b246afaSJeff Mahoney 41492a7cc42SQu Wenruo if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 4156426c7adSQu Wenruo root->last_trans < trans->transid) || force) { 4160b246afaSJeff Mahoney WARN_ON(root == fs_info->extent_root); 4174d31778aSQu Wenruo WARN_ON(!force && root->commit_root != root->node); 4185d4f98a2SYan Zheng 4197585717fSChris Mason /* 42027cdeb70SMiao Xie * see below for IN_TRANS_SETUP usage rules 4217585717fSChris Mason * we have the reloc mutex held now, so there 4227585717fSChris Mason * is only one writer in this function 4237585717fSChris Mason */ 42427cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4257585717fSChris Mason 42627cdeb70SMiao Xie /* make sure readers find IN_TRANS_SETUP before 4277585717fSChris Mason * they find our root->last_trans update 4287585717fSChris Mason */ 4297585717fSChris Mason smp_wmb(); 4307585717fSChris Mason 4310b246afaSJeff Mahoney spin_lock(&fs_info->fs_roots_radix_lock); 4326426c7adSQu Wenruo if (root->last_trans == trans->transid && !force) { 4330b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 434a4abeea4SJosef Bacik return 0; 435a4abeea4SJosef Bacik } 4360b246afaSJeff Mahoney radix_tree_tag_set(&fs_info->fs_roots_radix, 4376702ed49SChris Mason (unsigned long)root->root_key.objectid, 4386702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 4390b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 4407585717fSChris Mason root->last_trans = trans->transid; 4417585717fSChris Mason 4427585717fSChris Mason /* this is pretty tricky. We don't want to 4437585717fSChris Mason * take the relocation lock in btrfs_record_root_in_trans 4447585717fSChris Mason * unless we're really doing the first setup for this root in 4457585717fSChris Mason * this transaction. 4467585717fSChris Mason * 4477585717fSChris Mason * Normally we'd use root->last_trans as a flag to decide 4487585717fSChris Mason * if we want to take the expensive mutex. 4497585717fSChris Mason * 4507585717fSChris Mason * But, we have to set root->last_trans before we 4517585717fSChris Mason * init the relocation root, otherwise, we trip over warnings 4527585717fSChris Mason * in ctree.c. The solution used here is to flag ourselves 45327cdeb70SMiao Xie * with root IN_TRANS_SETUP. When this is 1, we're still 4547585717fSChris Mason * fixing up the reloc trees and everyone must wait. 4557585717fSChris Mason * 4567585717fSChris Mason * When this is zero, they can trust root->last_trans and fly 4577585717fSChris Mason * through btrfs_record_root_in_trans without having to take the 4587585717fSChris Mason * lock. smp_wmb() makes sure that all the writes above are 4597585717fSChris Mason * done before we pop in the zero below 4607585717fSChris Mason */ 46103a7e111SJosef Bacik ret = btrfs_init_reloc_root(trans, root); 462c7548af6SChris Mason smp_mb__before_atomic(); 46327cdeb70SMiao Xie clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4646702ed49SChris Mason } 46503a7e111SJosef Bacik return ret; 4666702ed49SChris Mason } 4675d4f98a2SYan Zheng 4687585717fSChris Mason 4692b9dbef2SJosef Bacik void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, 4702b9dbef2SJosef Bacik struct btrfs_root *root) 4712b9dbef2SJosef Bacik { 4720b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4732b9dbef2SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 4742b9dbef2SJosef Bacik 4752b9dbef2SJosef Bacik /* Add ourselves to the transaction dropped list */ 4762b9dbef2SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 4772b9dbef2SJosef Bacik list_add_tail(&root->root_list, &cur_trans->dropped_roots); 4782b9dbef2SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 4792b9dbef2SJosef Bacik 4802b9dbef2SJosef Bacik /* Make sure we don't try to update the root at commit time */ 4810b246afaSJeff Mahoney spin_lock(&fs_info->fs_roots_radix_lock); 4820b246afaSJeff Mahoney radix_tree_tag_clear(&fs_info->fs_roots_radix, 4832b9dbef2SJosef Bacik (unsigned long)root->root_key.objectid, 4842b9dbef2SJosef Bacik BTRFS_ROOT_TRANS_TAG); 4850b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 4862b9dbef2SJosef Bacik } 4872b9dbef2SJosef Bacik 4887585717fSChris Mason int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 4897585717fSChris Mason struct btrfs_root *root) 4907585717fSChris Mason { 4910b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4921409e6ccSJosef Bacik int ret; 4930b246afaSJeff Mahoney 49492a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 4957585717fSChris Mason return 0; 4967585717fSChris Mason 4977585717fSChris Mason /* 49827cdeb70SMiao Xie * see record_root_in_trans for comments about IN_TRANS_SETUP usage 4997585717fSChris Mason * and barriers 5007585717fSChris Mason */ 5017585717fSChris Mason smp_rmb(); 5027585717fSChris Mason if (root->last_trans == trans->transid && 50327cdeb70SMiao Xie !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) 5047585717fSChris Mason return 0; 5057585717fSChris Mason 5060b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 5071409e6ccSJosef Bacik ret = record_root_in_trans(trans, root, 0); 5080b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 5097585717fSChris Mason 5101409e6ccSJosef Bacik return ret; 5117585717fSChris Mason } 5127585717fSChris Mason 5134a9d8bdeSMiao Xie static inline int is_transaction_blocked(struct btrfs_transaction *trans) 5144a9d8bdeSMiao Xie { 5153296bf56SQu Wenruo return (trans->state >= TRANS_STATE_COMMIT_START && 516501407aaSJosef Bacik trans->state < TRANS_STATE_UNBLOCKED && 517bf31f87fSDavid Sterba !TRANS_ABORTED(trans)); 5184a9d8bdeSMiao Xie } 5194a9d8bdeSMiao Xie 520d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 521d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 522d352ac68SChris Mason * transaction might not be fully on disk. 523d352ac68SChris Mason */ 5242ff7e61eSJeff Mahoney static void wait_current_trans(struct btrfs_fs_info *fs_info) 52579154b1bSChris Mason { 526f9295749SChris Mason struct btrfs_transaction *cur_trans; 52779154b1bSChris Mason 5280b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5290b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 5304a9d8bdeSMiao Xie if (cur_trans && is_transaction_blocked(cur_trans)) { 5319b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 5320b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 53372d63ed6SLi Zefan 5340b246afaSJeff Mahoney wait_event(fs_info->transaction_wait, 535501407aaSJosef Bacik cur_trans->state >= TRANS_STATE_UNBLOCKED || 536bf31f87fSDavid Sterba TRANS_ABORTED(cur_trans)); 537724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 538a4abeea4SJosef Bacik } else { 5390b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 540f9295749SChris Mason } 54137d1aeeeSChris Mason } 54237d1aeeeSChris Mason 5432ff7e61eSJeff Mahoney static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) 54437d1aeeeSChris Mason { 5450b246afaSJeff Mahoney if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) 546a4abeea4SJosef Bacik return 0; 547a4abeea4SJosef Bacik 54892e2f7e3SNikolay Borisov if (type == TRANS_START) 549a4abeea4SJosef Bacik return 1; 550a4abeea4SJosef Bacik 551a22285a6SYan, Zheng return 0; 552a22285a6SYan, Zheng } 553a22285a6SYan, Zheng 55420dd2cbfSMiao Xie static inline bool need_reserve_reloc_root(struct btrfs_root *root) 55520dd2cbfSMiao Xie { 5560b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5570b246afaSJeff Mahoney 5580b246afaSJeff Mahoney if (!fs_info->reloc_ctl || 55992a7cc42SQu Wenruo !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || 56020dd2cbfSMiao Xie root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 56120dd2cbfSMiao Xie root->reloc_root) 56220dd2cbfSMiao Xie return false; 56320dd2cbfSMiao Xie 56420dd2cbfSMiao Xie return true; 56520dd2cbfSMiao Xie } 56620dd2cbfSMiao Xie 56708e007d2SMiao Xie static struct btrfs_trans_handle * 5685aed1dd8SAlexandru Moise start_transaction(struct btrfs_root *root, unsigned int num_items, 569003d7c59SJeff Mahoney unsigned int type, enum btrfs_reserve_flush_enum flush, 570003d7c59SJeff Mahoney bool enforce_qgroups) 571a22285a6SYan, Zheng { 5720b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 573ba2c4d4eSJosef Bacik struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 574a22285a6SYan, Zheng struct btrfs_trans_handle *h; 575a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 576b5009945SJosef Bacik u64 num_bytes = 0; 577c5567237SArne Jansen u64 qgroup_reserved = 0; 57820dd2cbfSMiao Xie bool reloc_reserved = false; 5799c343784SJosef Bacik bool do_chunk_alloc = false; 58020dd2cbfSMiao Xie int ret; 581acce952bSliubo 5820b246afaSJeff Mahoney if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) 583acce952bSliubo return ERR_PTR(-EROFS); 5842a1eb461SJosef Bacik 58546c4e71eSFilipe Manana if (current->journal_info) { 5860860adfdSMiao Xie WARN_ON(type & TRANS_EXTWRITERS); 5872a1eb461SJosef Bacik h = current->journal_info; 588b50fff81SDavid Sterba refcount_inc(&h->use_count); 589b50fff81SDavid Sterba WARN_ON(refcount_read(&h->use_count) > 2); 5902a1eb461SJosef Bacik h->orig_rsv = h->block_rsv; 5912a1eb461SJosef Bacik h->block_rsv = NULL; 5922a1eb461SJosef Bacik goto got_it; 5932a1eb461SJosef Bacik } 594b5009945SJosef Bacik 595b5009945SJosef Bacik /* 596b5009945SJosef Bacik * Do the reservation before we join the transaction so we can do all 597b5009945SJosef Bacik * the appropriate flushing if need be. 598b5009945SJosef Bacik */ 599003d7c59SJeff Mahoney if (num_items && root != fs_info->chunk_root) { 600ba2c4d4eSJosef Bacik struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; 601ba2c4d4eSJosef Bacik u64 delayed_refs_bytes = 0; 602ba2c4d4eSJosef Bacik 6030b246afaSJeff Mahoney qgroup_reserved = num_items * fs_info->nodesize; 604733e03a0SQu Wenruo ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, 605003d7c59SJeff Mahoney enforce_qgroups); 606c5567237SArne Jansen if (ret) 607c5567237SArne Jansen return ERR_PTR(ret); 608c5567237SArne Jansen 609ba2c4d4eSJosef Bacik /* 610ba2c4d4eSJosef Bacik * We want to reserve all the bytes we may need all at once, so 611ba2c4d4eSJosef Bacik * we only do 1 enospc flushing cycle per transaction start. We 612ba2c4d4eSJosef Bacik * accomplish this by simply assuming we'll do 2 x num_items 613ba2c4d4eSJosef Bacik * worth of delayed refs updates in this trans handle, and 614ba2c4d4eSJosef Bacik * refill that amount for whatever is missing in the reserve. 615ba2c4d4eSJosef Bacik */ 6162bd36e7bSJosef Bacik num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); 6177f9fe614SJosef Bacik if (flush == BTRFS_RESERVE_FLUSH_ALL && 6187f9fe614SJosef Bacik delayed_refs_rsv->full == 0) { 619ba2c4d4eSJosef Bacik delayed_refs_bytes = num_bytes; 620ba2c4d4eSJosef Bacik num_bytes <<= 1; 621ba2c4d4eSJosef Bacik } 622ba2c4d4eSJosef Bacik 62320dd2cbfSMiao Xie /* 62420dd2cbfSMiao Xie * Do the reservation for the relocation root creation 62520dd2cbfSMiao Xie */ 626ee39b432SDavid Sterba if (need_reserve_reloc_root(root)) { 6270b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 62820dd2cbfSMiao Xie reloc_reserved = true; 62920dd2cbfSMiao Xie } 63020dd2cbfSMiao Xie 631ba2c4d4eSJosef Bacik ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush); 632ba2c4d4eSJosef Bacik if (ret) 633ba2c4d4eSJosef Bacik goto reserve_fail; 634ba2c4d4eSJosef Bacik if (delayed_refs_bytes) { 635ba2c4d4eSJosef Bacik btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, 636ba2c4d4eSJosef Bacik delayed_refs_bytes); 637ba2c4d4eSJosef Bacik num_bytes -= delayed_refs_bytes; 638ba2c4d4eSJosef Bacik } 6399c343784SJosef Bacik 6409c343784SJosef Bacik if (rsv->space_info->force_alloc) 6419c343784SJosef Bacik do_chunk_alloc = true; 642ba2c4d4eSJosef Bacik } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && 643ba2c4d4eSJosef Bacik !delayed_refs_rsv->full) { 644ba2c4d4eSJosef Bacik /* 645ba2c4d4eSJosef Bacik * Some people call with btrfs_start_transaction(root, 0) 646ba2c4d4eSJosef Bacik * because they can be throttled, but have some other mechanism 647ba2c4d4eSJosef Bacik * for reserving space. We still want these guys to refill the 648ba2c4d4eSJosef Bacik * delayed block_rsv so just add 1 items worth of reservation 649ba2c4d4eSJosef Bacik * here. 650ba2c4d4eSJosef Bacik */ 651ba2c4d4eSJosef Bacik ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); 652b5009945SJosef Bacik if (ret) 653843fcf35SMiao Xie goto reserve_fail; 654b5009945SJosef Bacik } 655a22285a6SYan, Zheng again: 656f2f767e7SAlexandru Moise h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); 657843fcf35SMiao Xie if (!h) { 658843fcf35SMiao Xie ret = -ENOMEM; 659843fcf35SMiao Xie goto alloc_fail; 660843fcf35SMiao Xie } 661a22285a6SYan, Zheng 66298114659SJosef Bacik /* 66398114659SJosef Bacik * If we are JOIN_NOLOCK we're already committing a transaction and 66498114659SJosef Bacik * waiting on this guy, so we don't need to do the sb_start_intwrite 66598114659SJosef Bacik * because we're already holding a ref. We need this because we could 66698114659SJosef Bacik * have raced in and did an fsync() on a file which can kick a commit 66798114659SJosef Bacik * and then we deadlock with somebody doing a freeze. 668354aa0fbSMiao Xie * 669354aa0fbSMiao Xie * If we are ATTACH, it means we just want to catch the current 670354aa0fbSMiao Xie * transaction and commit it, so we needn't do sb_start_intwrite(). 67198114659SJosef Bacik */ 6720860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 6730b246afaSJeff Mahoney sb_start_intwrite(fs_info->sb); 674b2b5ef5cSJan Kara 6752ff7e61eSJeff Mahoney if (may_wait_transaction(fs_info, type)) 6762ff7e61eSJeff Mahoney wait_current_trans(fs_info); 677a22285a6SYan, Zheng 678a4abeea4SJosef Bacik do { 6792ff7e61eSJeff Mahoney ret = join_transaction(fs_info, type); 680178260b2SMiao Xie if (ret == -EBUSY) { 6812ff7e61eSJeff Mahoney wait_current_trans(fs_info); 682a6d155d2SFilipe Manana if (unlikely(type == TRANS_ATTACH || 683a6d155d2SFilipe Manana type == TRANS_JOIN_NOSTART)) 684178260b2SMiao Xie ret = -ENOENT; 685178260b2SMiao Xie } 686a4abeea4SJosef Bacik } while (ret == -EBUSY); 687a4abeea4SJosef Bacik 688a43f7f82SLiu Bo if (ret < 0) 689843fcf35SMiao Xie goto join_fail; 6900f7d52f4SChris Mason 6910b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 692a22285a6SYan, Zheng 693a22285a6SYan, Zheng h->transid = cur_trans->transid; 694a22285a6SYan, Zheng h->transaction = cur_trans; 695d13603efSArne Jansen h->root = root; 696b50fff81SDavid Sterba refcount_set(&h->use_count, 1); 69764b63580SJeff Mahoney h->fs_info = root->fs_info; 6987174109cSQu Wenruo 699a698d075SMiao Xie h->type = type; 700ea658badSJosef Bacik INIT_LIST_HEAD(&h->new_bgs); 701b7ec40d7SChris Mason 702a22285a6SYan, Zheng smp_mb(); 7033296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START && 7042ff7e61eSJeff Mahoney may_wait_transaction(fs_info, type)) { 705abdd2e80SFilipe Manana current->journal_info = h; 7063a45bb20SJeff Mahoney btrfs_commit_transaction(h); 707a22285a6SYan, Zheng goto again; 708a22285a6SYan, Zheng } 7099ed74f2dSJosef Bacik 710b5009945SJosef Bacik if (num_bytes) { 7110b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 7122bcc0328SLiu Bo h->transid, num_bytes, 1); 7130b246afaSJeff Mahoney h->block_rsv = &fs_info->trans_block_rsv; 714b5009945SJosef Bacik h->bytes_reserved = num_bytes; 71520dd2cbfSMiao Xie h->reloc_reserved = reloc_reserved; 716a22285a6SYan, Zheng } 717a22285a6SYan, Zheng 7182a1eb461SJosef Bacik got_it: 719bcf3a3e7SNikolay Borisov if (!current->journal_info) 720a22285a6SYan, Zheng current->journal_info = h; 721fcc99734SQu Wenruo 722fcc99734SQu Wenruo /* 7239c343784SJosef Bacik * If the space_info is marked ALLOC_FORCE then we'll get upgraded to 7249c343784SJosef Bacik * ALLOC_FORCE the first run through, and then we won't allocate for 7259c343784SJosef Bacik * anybody else who races in later. We don't care about the return 7269c343784SJosef Bacik * value here. 7279c343784SJosef Bacik */ 7289c343784SJosef Bacik if (do_chunk_alloc && num_bytes) { 7299c343784SJosef Bacik u64 flags = h->block_rsv->space_info->flags; 7309c343784SJosef Bacik 7319c343784SJosef Bacik btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), 7329c343784SJosef Bacik CHUNK_ALLOC_NO_FORCE); 7339c343784SJosef Bacik } 7349c343784SJosef Bacik 7359c343784SJosef Bacik /* 736fcc99734SQu Wenruo * btrfs_record_root_in_trans() needs to alloc new extents, and may 737fcc99734SQu Wenruo * call btrfs_join_transaction() while we're also starting a 738fcc99734SQu Wenruo * transaction. 739fcc99734SQu Wenruo * 740fcc99734SQu Wenruo * Thus it need to be called after current->journal_info initialized, 741fcc99734SQu Wenruo * or we can deadlock. 742fcc99734SQu Wenruo */ 74368075ea8SJosef Bacik ret = btrfs_record_root_in_trans(h, root); 74468075ea8SJosef Bacik if (ret) { 74568075ea8SJosef Bacik /* 74668075ea8SJosef Bacik * The transaction handle is fully initialized and linked with 74768075ea8SJosef Bacik * other structures so it needs to be ended in case of errors, 74868075ea8SJosef Bacik * not just freed. 74968075ea8SJosef Bacik */ 75068075ea8SJosef Bacik btrfs_end_transaction(h); 75168075ea8SJosef Bacik return ERR_PTR(ret); 75268075ea8SJosef Bacik } 753fcc99734SQu Wenruo 75479154b1bSChris Mason return h; 755843fcf35SMiao Xie 756843fcf35SMiao Xie join_fail: 7570860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 7580b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 759843fcf35SMiao Xie kmem_cache_free(btrfs_trans_handle_cachep, h); 760843fcf35SMiao Xie alloc_fail: 761843fcf35SMiao Xie if (num_bytes) 7622ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, 76363f018beSNikolay Borisov num_bytes, NULL); 764843fcf35SMiao Xie reserve_fail: 765733e03a0SQu Wenruo btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); 766843fcf35SMiao Xie return ERR_PTR(ret); 76779154b1bSChris Mason } 76879154b1bSChris Mason 769f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 7705aed1dd8SAlexandru Moise unsigned int num_items) 771f9295749SChris Mason { 77208e007d2SMiao Xie return start_transaction(root, num_items, TRANS_START, 773003d7c59SJeff Mahoney BTRFS_RESERVE_FLUSH_ALL, true); 774f9295749SChris Mason } 775003d7c59SJeff Mahoney 7768eab77ffSFilipe Manana struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( 7778eab77ffSFilipe Manana struct btrfs_root *root, 7787f9fe614SJosef Bacik unsigned int num_items) 7798eab77ffSFilipe Manana { 7807f9fe614SJosef Bacik return start_transaction(root, num_items, TRANS_START, 7817f9fe614SJosef Bacik BTRFS_RESERVE_FLUSH_ALL_STEAL, false); 7828eab77ffSFilipe Manana } 7838407aa46SMiao Xie 7847a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) 785f9295749SChris Mason { 786003d7c59SJeff Mahoney return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, 787003d7c59SJeff Mahoney true); 788f9295749SChris Mason } 789f9295749SChris Mason 7908d510121SNikolay Borisov struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) 7910af3d00bSJosef Bacik { 792575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 793003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 7940af3d00bSJosef Bacik } 7950af3d00bSJosef Bacik 796d4edf39bSMiao Xie /* 797a6d155d2SFilipe Manana * Similar to regular join but it never starts a transaction when none is 798a6d155d2SFilipe Manana * running or after waiting for the current one to finish. 799a6d155d2SFilipe Manana */ 800a6d155d2SFilipe Manana struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) 801a6d155d2SFilipe Manana { 802a6d155d2SFilipe Manana return start_transaction(root, 0, TRANS_JOIN_NOSTART, 803a6d155d2SFilipe Manana BTRFS_RESERVE_NO_FLUSH, true); 804a6d155d2SFilipe Manana } 805a6d155d2SFilipe Manana 806a6d155d2SFilipe Manana /* 807d4edf39bSMiao Xie * btrfs_attach_transaction() - catch the running transaction 808d4edf39bSMiao Xie * 809d4edf39bSMiao Xie * It is used when we want to commit the current the transaction, but 810d4edf39bSMiao Xie * don't want to start a new one. 811d4edf39bSMiao Xie * 812d4edf39bSMiao Xie * Note: If this function return -ENOENT, it just means there is no 813d4edf39bSMiao Xie * running transaction. But it is possible that the inactive transaction 814d4edf39bSMiao Xie * is still in the memory, not fully on disk. If you hope there is no 815d4edf39bSMiao Xie * inactive transaction in the fs when -ENOENT is returned, you should 816d4edf39bSMiao Xie * invoke 817d4edf39bSMiao Xie * btrfs_attach_transaction_barrier() 818d4edf39bSMiao Xie */ 819354aa0fbSMiao Xie struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) 82060376ce4SJosef Bacik { 821575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_ATTACH, 822003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 82360376ce4SJosef Bacik } 82460376ce4SJosef Bacik 825d4edf39bSMiao Xie /* 82690b6d283SWang Sheng-Hui * btrfs_attach_transaction_barrier() - catch the running transaction 827d4edf39bSMiao Xie * 82852042d8eSAndrea Gelmini * It is similar to the above function, the difference is this one 829d4edf39bSMiao Xie * will wait for all the inactive transactions until they fully 830d4edf39bSMiao Xie * complete. 831d4edf39bSMiao Xie */ 832d4edf39bSMiao Xie struct btrfs_trans_handle * 833d4edf39bSMiao Xie btrfs_attach_transaction_barrier(struct btrfs_root *root) 834d4edf39bSMiao Xie { 835d4edf39bSMiao Xie struct btrfs_trans_handle *trans; 836d4edf39bSMiao Xie 837575a75d6SAlexandru Moise trans = start_transaction(root, 0, TRANS_ATTACH, 838003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 8398d9e220cSAl Viro if (trans == ERR_PTR(-ENOENT)) 8402ff7e61eSJeff Mahoney btrfs_wait_for_commit(root->fs_info, 0); 841d4edf39bSMiao Xie 842d4edf39bSMiao Xie return trans; 843d4edf39bSMiao Xie } 844d4edf39bSMiao Xie 845d0c2f4faSFilipe Manana /* Wait for a transaction commit to reach at least the given state. */ 846d0c2f4faSFilipe Manana static noinline void wait_for_commit(struct btrfs_transaction *commit, 847d0c2f4faSFilipe Manana const enum btrfs_trans_state min_state) 84889ce8a63SChris Mason { 849d0c2f4faSFilipe Manana wait_event(commit->commit_wait, commit->state >= min_state); 85089ce8a63SChris Mason } 85189ce8a63SChris Mason 8522ff7e61eSJeff Mahoney int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) 85346204592SSage Weil { 85446204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 8558cd2807fSMiao Xie int ret = 0; 85646204592SSage Weil 85746204592SSage Weil if (transid) { 8580b246afaSJeff Mahoney if (transid <= fs_info->last_trans_committed) 859a4abeea4SJosef Bacik goto out; 86046204592SSage Weil 86146204592SSage Weil /* find specified transaction */ 8620b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8630b246afaSJeff Mahoney list_for_each_entry(t, &fs_info->trans_list, list) { 86446204592SSage Weil if (t->transid == transid) { 86546204592SSage Weil cur_trans = t; 8669b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 8678cd2807fSMiao Xie ret = 0; 86846204592SSage Weil break; 86946204592SSage Weil } 8708cd2807fSMiao Xie if (t->transid > transid) { 8718cd2807fSMiao Xie ret = 0; 87246204592SSage Weil break; 87346204592SSage Weil } 8748cd2807fSMiao Xie } 8750b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 87642383020SSage Weil 87742383020SSage Weil /* 87842383020SSage Weil * The specified transaction doesn't exist, or we 87942383020SSage Weil * raced with btrfs_commit_transaction 88042383020SSage Weil */ 88142383020SSage Weil if (!cur_trans) { 8820b246afaSJeff Mahoney if (transid > fs_info->last_trans_committed) 88342383020SSage Weil ret = -EINVAL; 8848cd2807fSMiao Xie goto out; 88542383020SSage Weil } 88646204592SSage Weil } else { 88746204592SSage Weil /* find newest transaction that is committing | committed */ 8880b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8890b246afaSJeff Mahoney list_for_each_entry_reverse(t, &fs_info->trans_list, 89046204592SSage Weil list) { 8914a9d8bdeSMiao Xie if (t->state >= TRANS_STATE_COMMIT_START) { 8924a9d8bdeSMiao Xie if (t->state == TRANS_STATE_COMPLETED) 8933473f3c0SJosef Bacik break; 89446204592SSage Weil cur_trans = t; 8959b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 89646204592SSage Weil break; 89746204592SSage Weil } 89846204592SSage Weil } 8990b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 90046204592SSage Weil if (!cur_trans) 901a4abeea4SJosef Bacik goto out; /* nothing committing|committed */ 90246204592SSage Weil } 90346204592SSage Weil 904d0c2f4faSFilipe Manana wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); 905724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 906a4abeea4SJosef Bacik out: 90746204592SSage Weil return ret; 90846204592SSage Weil } 90946204592SSage Weil 9102ff7e61eSJeff Mahoney void btrfs_throttle(struct btrfs_fs_info *fs_info) 91137d1aeeeSChris Mason { 9122ff7e61eSJeff Mahoney wait_current_trans(fs_info); 91337d1aeeeSChris Mason } 91437d1aeeeSChris Mason 9158a8f4deaSNikolay Borisov static bool should_end_transaction(struct btrfs_trans_handle *trans) 9168929ecfaSYan, Zheng { 9172ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 9180b246afaSJeff Mahoney 91964403612SJosef Bacik if (btrfs_check_space_for_delayed_refs(fs_info)) 9208a8f4deaSNikolay Borisov return true; 92136ba022aSJosef Bacik 9222ff7e61eSJeff Mahoney return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); 9238929ecfaSYan, Zheng } 9248929ecfaSYan, Zheng 925a2633b6aSNikolay Borisov bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) 9268929ecfaSYan, Zheng { 9278929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9288929ecfaSYan, Zheng 9293296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START || 930e19eb11fSJosef Bacik test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) 931a2633b6aSNikolay Borisov return true; 9328929ecfaSYan, Zheng 9332ff7e61eSJeff Mahoney return should_end_transaction(trans); 9348929ecfaSYan, Zheng } 9358929ecfaSYan, Zheng 936dc60c525SNikolay Borisov static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) 937dc60c525SNikolay Borisov 9380e34693fSNikolay Borisov { 939dc60c525SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 940dc60c525SNikolay Borisov 9410e34693fSNikolay Borisov if (!trans->block_rsv) { 9420e34693fSNikolay Borisov ASSERT(!trans->bytes_reserved); 9430e34693fSNikolay Borisov return; 9440e34693fSNikolay Borisov } 9450e34693fSNikolay Borisov 9460e34693fSNikolay Borisov if (!trans->bytes_reserved) 9470e34693fSNikolay Borisov return; 9480e34693fSNikolay Borisov 9490e34693fSNikolay Borisov ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); 9500e34693fSNikolay Borisov trace_btrfs_space_reservation(fs_info, "transaction", 9510e34693fSNikolay Borisov trans->transid, trans->bytes_reserved, 0); 9520e34693fSNikolay Borisov btrfs_block_rsv_release(fs_info, trans->block_rsv, 95363f018beSNikolay Borisov trans->bytes_reserved, NULL); 9540e34693fSNikolay Borisov trans->bytes_reserved = 0; 9550e34693fSNikolay Borisov } 9560e34693fSNikolay Borisov 95789ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 9583a45bb20SJeff Mahoney int throttle) 95979154b1bSChris Mason { 9603a45bb20SJeff Mahoney struct btrfs_fs_info *info = trans->fs_info; 9618929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9624edc2ca3SDave Jones int err = 0; 963d6e4a428SChris Mason 964b50fff81SDavid Sterba if (refcount_read(&trans->use_count) > 1) { 965b50fff81SDavid Sterba refcount_dec(&trans->use_count); 9662a1eb461SJosef Bacik trans->block_rsv = trans->orig_rsv; 9672a1eb461SJosef Bacik return 0; 9682a1eb461SJosef Bacik } 9692a1eb461SJosef Bacik 970dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 9714c13d758SJosef Bacik trans->block_rsv = NULL; 972c5567237SArne Jansen 9736c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 974ea658badSJosef Bacik 9754fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 9764fbcdf66SFilipe Manana 9770860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 9780b246afaSJeff Mahoney sb_end_intwrite(info->sb); 9796df7881aSJosef Bacik 9808929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 98113c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 98213c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 9830860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 98489ce8a63SChris Mason 985093258e6SDavid Sterba cond_wake_up(&cur_trans->writer_wait); 986724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 9879ed74f2dSJosef Bacik 9889ed74f2dSJosef Bacik if (current->journal_info == trans) 9899ed74f2dSJosef Bacik current->journal_info = NULL; 990ab78c84dSChris Mason 99124bbcf04SYan, Zheng if (throttle) 9922ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(info); 99324bbcf04SYan, Zheng 994bf31f87fSDavid Sterba if (TRANS_ABORTED(trans) || 9950b246afaSJeff Mahoney test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) { 9964e121c06SJosef Bacik wake_up_process(info->transaction_kthread); 997fbabd4a3SJosef Bacik if (TRANS_ABORTED(trans)) 998fbabd4a3SJosef Bacik err = trans->aborted; 999fbabd4a3SJosef Bacik else 1000fbabd4a3SJosef Bacik err = -EROFS; 10014e121c06SJosef Bacik } 100249b25e05SJeff Mahoney 10034edc2ca3SDave Jones kmem_cache_free(btrfs_trans_handle_cachep, trans); 10044edc2ca3SDave Jones return err; 100579154b1bSChris Mason } 100679154b1bSChris Mason 10073a45bb20SJeff Mahoney int btrfs_end_transaction(struct btrfs_trans_handle *trans) 100889ce8a63SChris Mason { 10093a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 0); 101089ce8a63SChris Mason } 101189ce8a63SChris Mason 10123a45bb20SJeff Mahoney int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) 101389ce8a63SChris Mason { 10143a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 1); 101516cdcec7SMiao Xie } 101616cdcec7SMiao Xie 1017d352ac68SChris Mason /* 1018d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1019d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1020690587d1SChris Mason * those extents are sent to disk but does not wait on them 1021d352ac68SChris Mason */ 10222ff7e61eSJeff Mahoney int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, 10238cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 102479154b1bSChris Mason { 1025777e6bd7SChris Mason int err = 0; 10267c4452b9SChris Mason int werr = 0; 10270b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1028e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1029777e6bd7SChris Mason u64 start = 0; 10305f39d397SChris Mason u64 end; 10317c4452b9SChris Mason 10326300463bSLiu Bo atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); 10331728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1034e6138876SJosef Bacik mark, &cached_state)) { 1035663dfbb0SFilipe Manana bool wait_writeback = false; 1036663dfbb0SFilipe Manana 1037663dfbb0SFilipe Manana err = convert_extent_bit(dirty_pages, start, end, 1038663dfbb0SFilipe Manana EXTENT_NEED_WAIT, 1039210aa277SDavid Sterba mark, &cached_state); 1040663dfbb0SFilipe Manana /* 1041663dfbb0SFilipe Manana * convert_extent_bit can return -ENOMEM, which is most of the 1042663dfbb0SFilipe Manana * time a temporary error. So when it happens, ignore the error 1043663dfbb0SFilipe Manana * and wait for writeback of this range to finish - because we 1044663dfbb0SFilipe Manana * failed to set the bit EXTENT_NEED_WAIT for the range, a call 1045bf89d38fSJeff Mahoney * to __btrfs_wait_marked_extents() would not know that 1046bf89d38fSJeff Mahoney * writeback for this range started and therefore wouldn't 1047bf89d38fSJeff Mahoney * wait for it to finish - we don't want to commit a 1048bf89d38fSJeff Mahoney * superblock that points to btree nodes/leafs for which 1049bf89d38fSJeff Mahoney * writeback hasn't finished yet (and without errors). 1050663dfbb0SFilipe Manana * We cleanup any entries left in the io tree when committing 105141e7acd3SNikolay Borisov * the transaction (through extent_io_tree_release()). 1052663dfbb0SFilipe Manana */ 1053663dfbb0SFilipe Manana if (err == -ENOMEM) { 1054663dfbb0SFilipe Manana err = 0; 1055663dfbb0SFilipe Manana wait_writeback = true; 1056663dfbb0SFilipe Manana } 1057663dfbb0SFilipe Manana if (!err) 10581728366eSJosef Bacik err = filemap_fdatawrite_range(mapping, start, end); 10597c4452b9SChris Mason if (err) 10607c4452b9SChris Mason werr = err; 1061663dfbb0SFilipe Manana else if (wait_writeback) 1062663dfbb0SFilipe Manana werr = filemap_fdatawait_range(mapping, start, end); 1063e38e2ed7SFilipe Manana free_extent_state(cached_state); 1064663dfbb0SFilipe Manana cached_state = NULL; 10651728366eSJosef Bacik cond_resched(); 10661728366eSJosef Bacik start = end + 1; 10677c4452b9SChris Mason } 10686300463bSLiu Bo atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); 1069690587d1SChris Mason return werr; 1070690587d1SChris Mason } 1071690587d1SChris Mason 1072690587d1SChris Mason /* 1073690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1074690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1075690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 1076690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 1077690587d1SChris Mason */ 1078bf89d38fSJeff Mahoney static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, 1079bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1080690587d1SChris Mason { 1081690587d1SChris Mason int err = 0; 1082690587d1SChris Mason int werr = 0; 10830b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1084e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1085690587d1SChris Mason u64 start = 0; 1086690587d1SChris Mason u64 end; 1087690587d1SChris Mason 10881728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1089e6138876SJosef Bacik EXTENT_NEED_WAIT, &cached_state)) { 1090663dfbb0SFilipe Manana /* 1091663dfbb0SFilipe Manana * Ignore -ENOMEM errors returned by clear_extent_bit(). 1092663dfbb0SFilipe Manana * When committing the transaction, we'll remove any entries 1093663dfbb0SFilipe Manana * left in the io tree. For a log commit, we don't remove them 1094663dfbb0SFilipe Manana * after committing the log because the tree can be accessed 1095663dfbb0SFilipe Manana * concurrently - we do it only at transaction commit time when 109641e7acd3SNikolay Borisov * it's safe to do it (through extent_io_tree_release()). 1097663dfbb0SFilipe Manana */ 1098663dfbb0SFilipe Manana err = clear_extent_bit(dirty_pages, start, end, 1099ae0f1625SDavid Sterba EXTENT_NEED_WAIT, 0, 0, &cached_state); 1100663dfbb0SFilipe Manana if (err == -ENOMEM) 1101663dfbb0SFilipe Manana err = 0; 1102663dfbb0SFilipe Manana if (!err) 11031728366eSJosef Bacik err = filemap_fdatawait_range(mapping, start, end); 1104777e6bd7SChris Mason if (err) 1105777e6bd7SChris Mason werr = err; 1106e38e2ed7SFilipe Manana free_extent_state(cached_state); 1107e38e2ed7SFilipe Manana cached_state = NULL; 1108777e6bd7SChris Mason cond_resched(); 11091728366eSJosef Bacik start = end + 1; 1110777e6bd7SChris Mason } 11117c4452b9SChris Mason if (err) 11127c4452b9SChris Mason werr = err; 1113bf89d38fSJeff Mahoney return werr; 1114bf89d38fSJeff Mahoney } 1115656f30dbSFilipe Manana 1116b9fae2ebSFilipe Manana static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, 1117bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1118bf89d38fSJeff Mahoney { 1119bf89d38fSJeff Mahoney bool errors = false; 1120bf89d38fSJeff Mahoney int err; 1121bf89d38fSJeff Mahoney 1122bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1123bf89d38fSJeff Mahoney if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) 1124bf89d38fSJeff Mahoney errors = true; 1125bf89d38fSJeff Mahoney 1126bf89d38fSJeff Mahoney if (errors && !err) 1127bf89d38fSJeff Mahoney err = -EIO; 1128bf89d38fSJeff Mahoney return err; 1129bf89d38fSJeff Mahoney } 1130bf89d38fSJeff Mahoney 1131bf89d38fSJeff Mahoney int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) 1132bf89d38fSJeff Mahoney { 1133bf89d38fSJeff Mahoney struct btrfs_fs_info *fs_info = log_root->fs_info; 1134bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; 1135bf89d38fSJeff Mahoney bool errors = false; 1136bf89d38fSJeff Mahoney int err; 1137bf89d38fSJeff Mahoney 1138bf89d38fSJeff Mahoney ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 1139bf89d38fSJeff Mahoney 1140bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1141656f30dbSFilipe Manana if ((mark & EXTENT_DIRTY) && 11420b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) 1143656f30dbSFilipe Manana errors = true; 1144656f30dbSFilipe Manana 1145656f30dbSFilipe Manana if ((mark & EXTENT_NEW) && 11460b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) 1147656f30dbSFilipe Manana errors = true; 1148656f30dbSFilipe Manana 1149bf89d38fSJeff Mahoney if (errors && !err) 1150bf89d38fSJeff Mahoney err = -EIO; 1151bf89d38fSJeff Mahoney return err; 115279154b1bSChris Mason } 115379154b1bSChris Mason 1154690587d1SChris Mason /* 1155c9b577c0SNikolay Borisov * When btree blocks are allocated the corresponding extents are marked dirty. 1156c9b577c0SNikolay Borisov * This function ensures such extents are persisted on disk for transaction or 1157c9b577c0SNikolay Borisov * log commit. 1158c9b577c0SNikolay Borisov * 1159c9b577c0SNikolay Borisov * @trans: transaction whose dirty pages we'd like to write 1160690587d1SChris Mason */ 116170458a58SNikolay Borisov static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) 1162d0c803c4SChris Mason { 1163663dfbb0SFilipe Manana int ret; 1164c9b577c0SNikolay Borisov int ret2; 1165c9b577c0SNikolay Borisov struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; 116670458a58SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1167c9b577c0SNikolay Borisov struct blk_plug plug; 1168663dfbb0SFilipe Manana 1169c9b577c0SNikolay Borisov blk_start_plug(&plug); 1170c9b577c0SNikolay Borisov ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); 1171c9b577c0SNikolay Borisov blk_finish_plug(&plug); 1172c9b577c0SNikolay Borisov ret2 = btrfs_wait_extents(fs_info, dirty_pages); 1173c9b577c0SNikolay Borisov 117441e7acd3SNikolay Borisov extent_io_tree_release(&trans->transaction->dirty_pages); 1175663dfbb0SFilipe Manana 1176c9b577c0SNikolay Borisov if (ret) 1177663dfbb0SFilipe Manana return ret; 1178c9b577c0SNikolay Borisov else if (ret2) 1179c9b577c0SNikolay Borisov return ret2; 1180c9b577c0SNikolay Borisov else 1181c9b577c0SNikolay Borisov return 0; 1182d0c803c4SChris Mason } 1183d0c803c4SChris Mason 1184d352ac68SChris Mason /* 1185d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 1186d352ac68SChris Mason * 1187d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 1188d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 1189d352ac68SChris Mason * allocation tree. 1190d352ac68SChris Mason * 1191d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 1192d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 1193d352ac68SChris Mason */ 11940b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 119579154b1bSChris Mason struct btrfs_root *root) 119679154b1bSChris Mason { 119779154b1bSChris Mason int ret; 11980b86a832SChris Mason u64 old_root_bytenr; 119986b9f2ecSYan, Zheng u64 old_root_used; 12000b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12010b246afaSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 120279154b1bSChris Mason 120386b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 120456bec294SChris Mason 120579154b1bSChris Mason while (1) { 12060b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 120786b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 1208ea526d18SJosef Bacik old_root_used == btrfs_root_used(&root->root_item)) 120979154b1bSChris Mason break; 121087ef2bb4SChris Mason 12115d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 121279154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 12130b86a832SChris Mason &root->root_key, 12140b86a832SChris Mason &root->root_item); 121549b25e05SJeff Mahoney if (ret) 121649b25e05SJeff Mahoney return ret; 121756bec294SChris Mason 121886b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 1219e7070be1SJosef Bacik } 1220276e680dSYan Zheng 12210b86a832SChris Mason return 0; 12220b86a832SChris Mason } 12230b86a832SChris Mason 1224d352ac68SChris Mason /* 1225d352ac68SChris Mason * update all the cowonly tree roots on disk 122649b25e05SJeff Mahoney * 122749b25e05SJeff Mahoney * The error handling in this function may not be obvious. Any of the 122849b25e05SJeff Mahoney * failures will cause the file system to go offline. We still need 122949b25e05SJeff Mahoney * to clean up the delayed refs. 1230d352ac68SChris Mason */ 12319386d8bcSNikolay Borisov static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) 12320b86a832SChris Mason { 12339386d8bcSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1234ea526d18SJosef Bacik struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; 12351bbc621eSChris Mason struct list_head *io_bgs = &trans->transaction->io_bgs; 12360b86a832SChris Mason struct list_head *next; 123784234f3aSYan Zheng struct extent_buffer *eb; 123856bec294SChris Mason int ret; 123984234f3aSYan Zheng 124084234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 124149b25e05SJeff Mahoney ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 12429631e4ccSJosef Bacik 0, &eb, BTRFS_NESTING_COW); 124384234f3aSYan Zheng btrfs_tree_unlock(eb); 124484234f3aSYan Zheng free_extent_buffer(eb); 12450b86a832SChris Mason 124649b25e05SJeff Mahoney if (ret) 124749b25e05SJeff Mahoney return ret; 124849b25e05SJeff Mahoney 1249196c9d8dSDavid Sterba ret = btrfs_run_dev_stats(trans); 1250c16ce190SJosef Bacik if (ret) 1251c16ce190SJosef Bacik return ret; 12522b584c68SDavid Sterba ret = btrfs_run_dev_replace(trans); 1253c16ce190SJosef Bacik if (ret) 1254c16ce190SJosef Bacik return ret; 1255280f8bd2SLu Fengqi ret = btrfs_run_qgroups(trans); 1256c16ce190SJosef Bacik if (ret) 1257c16ce190SJosef Bacik return ret; 1258546adb0dSJan Schmidt 1259bbebb3e0SDavid Sterba ret = btrfs_setup_space_cache(trans); 1260dcdf7f6dSJosef Bacik if (ret) 1261dcdf7f6dSJosef Bacik return ret; 1262dcdf7f6dSJosef Bacik 1263ea526d18SJosef Bacik again: 12640b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 12652ff7e61eSJeff Mahoney struct btrfs_root *root; 12660b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 12670b86a832SChris Mason list_del_init(next); 12680b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 1269e7070be1SJosef Bacik clear_bit(BTRFS_ROOT_DIRTY, &root->state); 127087ef2bb4SChris Mason 12719e351cc8SJosef Bacik if (root != fs_info->extent_root) 12729e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 12739e351cc8SJosef Bacik &trans->transaction->switch_commits); 127449b25e05SJeff Mahoney ret = update_cowonly_root(trans, root); 127549b25e05SJeff Mahoney if (ret) 127649b25e05SJeff Mahoney return ret; 1277488bc2a2SJosef Bacik } 1278488bc2a2SJosef Bacik 1279488bc2a2SJosef Bacik /* Now flush any delayed refs generated by updating all of the roots */ 1280c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1281ea526d18SJosef Bacik if (ret) 1282ea526d18SJosef Bacik return ret; 1283276e680dSYan Zheng 12841bbc621eSChris Mason while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { 12855742d15fSDavid Sterba ret = btrfs_write_dirty_block_groups(trans); 1286ea526d18SJosef Bacik if (ret) 1287ea526d18SJosef Bacik return ret; 1288488bc2a2SJosef Bacik 1289488bc2a2SJosef Bacik /* 1290488bc2a2SJosef Bacik * We're writing the dirty block groups, which could generate 1291488bc2a2SJosef Bacik * delayed refs, which could generate more dirty block groups, 1292488bc2a2SJosef Bacik * so we want to keep this flushing in this loop to make sure 1293488bc2a2SJosef Bacik * everything gets run. 1294488bc2a2SJosef Bacik */ 1295c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1296ea526d18SJosef Bacik if (ret) 1297ea526d18SJosef Bacik return ret; 1298ea526d18SJosef Bacik } 1299ea526d18SJosef Bacik 1300ea526d18SJosef Bacik if (!list_empty(&fs_info->dirty_cowonly_roots)) 1301ea526d18SJosef Bacik goto again; 1302ea526d18SJosef Bacik 13039e351cc8SJosef Bacik list_add_tail(&fs_info->extent_root->dirty_list, 13049e351cc8SJosef Bacik &trans->transaction->switch_commits); 13059f6cbcbbSDavid Sterba 13069f6cbcbbSDavid Sterba /* Update dev-replace pointer once everything is committed */ 13079f6cbcbbSDavid Sterba fs_info->dev_replace.committed_cursor_left = 13089f6cbcbbSDavid Sterba fs_info->dev_replace.cursor_left_last_write_of_item; 13098dabb742SStefan Behrens 131079154b1bSChris Mason return 0; 131179154b1bSChris Mason } 131279154b1bSChris Mason 1313d352ac68SChris Mason /* 1314d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 1315d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 1316d352ac68SChris Mason * be deleted 1317d352ac68SChris Mason */ 1318cfad392bSJosef Bacik void btrfs_add_dead_root(struct btrfs_root *root) 13195eda7b5eSChris Mason { 13200b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13210b246afaSJeff Mahoney 13220b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1323dc9492c1SJosef Bacik if (list_empty(&root->root_list)) { 1324dc9492c1SJosef Bacik btrfs_grab_root(root); 13250b246afaSJeff Mahoney list_add_tail(&root->root_list, &fs_info->dead_roots); 1326dc9492c1SJosef Bacik } 13270b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 13285eda7b5eSChris Mason } 13295eda7b5eSChris Mason 1330d352ac68SChris Mason /* 13315d4f98a2SYan Zheng * update all the cowonly tree roots on disk 1332d352ac68SChris Mason */ 13337e4443d9SNikolay Borisov static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) 13340f7d52f4SChris Mason { 13357e4443d9SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 13360f7d52f4SChris Mason struct btrfs_root *gang[8]; 13370f7d52f4SChris Mason int i; 13380f7d52f4SChris Mason int ret; 133954aa1f4dSChris Mason 1340a4abeea4SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 13410f7d52f4SChris Mason while (1) { 13425d4f98a2SYan Zheng ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 13435d4f98a2SYan Zheng (void **)gang, 0, 13440f7d52f4SChris Mason ARRAY_SIZE(gang), 13450f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 13460f7d52f4SChris Mason if (ret == 0) 13470f7d52f4SChris Mason break; 13480f7d52f4SChris Mason for (i = 0; i < ret; i++) { 13495b4aacefSJeff Mahoney struct btrfs_root *root = gang[i]; 13504f4317c1SJosef Bacik int ret2; 13514f4317c1SJosef Bacik 13525d4f98a2SYan Zheng radix_tree_tag_clear(&fs_info->fs_roots_radix, 13532619ba1fSChris Mason (unsigned long)root->root_key.objectid, 13540f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 1355a4abeea4SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 135631153d81SYan Zheng 1357e02119d5SChris Mason btrfs_free_log(trans, root); 13582dd8298eSJosef Bacik ret2 = btrfs_update_reloc_root(trans, root); 13592dd8298eSJosef Bacik if (ret2) 13602dd8298eSJosef Bacik return ret2; 1361e02119d5SChris Mason 1362f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 136327cdeb70SMiao Xie clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1364c7548af6SChris Mason smp_mb__after_atomic(); 1365f1ebcc74SLiu Bo 1366978d910dSYan Zheng if (root->commit_root != root->node) { 13679e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 13689e351cc8SJosef Bacik &trans->transaction->switch_commits); 1369978d910dSYan Zheng btrfs_set_root_node(&root->root_item, 1370978d910dSYan Zheng root->node); 1371978d910dSYan Zheng } 137231153d81SYan Zheng 13734f4317c1SJosef Bacik ret2 = btrfs_update_root(trans, fs_info->tree_root, 13740f7d52f4SChris Mason &root->root_key, 13750f7d52f4SChris Mason &root->root_item); 13764f4317c1SJosef Bacik if (ret2) 13774f4317c1SJosef Bacik return ret2; 1378a4abeea4SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 1379733e03a0SQu Wenruo btrfs_qgroup_free_meta_all_pertrans(root); 13800f7d52f4SChris Mason } 13819f3a7427SChris Mason } 1382a4abeea4SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 13834f4317c1SJosef Bacik return 0; 13840f7d52f4SChris Mason } 13850f7d52f4SChris Mason 1386d352ac68SChris Mason /* 1387de78b51aSEric Sandeen * defrag a given btree. 1388de78b51aSEric Sandeen * Every leaf in the btree is read and defragged. 1389d352ac68SChris Mason */ 1390de78b51aSEric Sandeen int btrfs_defrag_root(struct btrfs_root *root) 1391e9d0b13bSChris Mason { 1392e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 1393e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 13948929ecfaSYan, Zheng int ret; 1395e9d0b13bSChris Mason 139627cdeb70SMiao Xie if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) 1397e9d0b13bSChris Mason return 0; 13988929ecfaSYan, Zheng 13996b80053dSChris Mason while (1) { 14008929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 14016819703fSDavid Sterba if (IS_ERR(trans)) { 14026819703fSDavid Sterba ret = PTR_ERR(trans); 14036819703fSDavid Sterba break; 14046819703fSDavid Sterba } 14058929ecfaSYan, Zheng 1406de78b51aSEric Sandeen ret = btrfs_defrag_leaves(trans, root); 14078929ecfaSYan, Zheng 14083a45bb20SJeff Mahoney btrfs_end_transaction(trans); 14092ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(info); 1410e9d0b13bSChris Mason cond_resched(); 1411e9d0b13bSChris Mason 1412ab8d0fc4SJeff Mahoney if (btrfs_fs_closing(info) || ret != -EAGAIN) 1413e9d0b13bSChris Mason break; 1414210549ebSDavid Sterba 1415ab8d0fc4SJeff Mahoney if (btrfs_defrag_cancelled(info)) { 1416ab8d0fc4SJeff Mahoney btrfs_debug(info, "defrag_root cancelled"); 1417210549ebSDavid Sterba ret = -EAGAIN; 1418210549ebSDavid Sterba break; 1419210549ebSDavid Sterba } 1420e9d0b13bSChris Mason } 142127cdeb70SMiao Xie clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); 14228929ecfaSYan, Zheng return ret; 1423e9d0b13bSChris Mason } 1424e9d0b13bSChris Mason 1425d352ac68SChris Mason /* 14266426c7adSQu Wenruo * Do all special snapshot related qgroup dirty hack. 14276426c7adSQu Wenruo * 14286426c7adSQu Wenruo * Will do all needed qgroup inherit and dirty hack like switch commit 14296426c7adSQu Wenruo * roots inside one transaction and write all btree into disk, to make 14306426c7adSQu Wenruo * qgroup works. 14316426c7adSQu Wenruo */ 14326426c7adSQu Wenruo static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, 14336426c7adSQu Wenruo struct btrfs_root *src, 14346426c7adSQu Wenruo struct btrfs_root *parent, 14356426c7adSQu Wenruo struct btrfs_qgroup_inherit *inherit, 14366426c7adSQu Wenruo u64 dst_objectid) 14376426c7adSQu Wenruo { 14386426c7adSQu Wenruo struct btrfs_fs_info *fs_info = src->fs_info; 14396426c7adSQu Wenruo int ret; 14406426c7adSQu Wenruo 14416426c7adSQu Wenruo /* 14426426c7adSQu Wenruo * Save some performance in the case that qgroups are not 14436426c7adSQu Wenruo * enabled. If this check races with the ioctl, rescan will 14446426c7adSQu Wenruo * kick in anyway. 14456426c7adSQu Wenruo */ 14469ea6e2b5SDavid Sterba if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) 14476426c7adSQu Wenruo return 0; 14486426c7adSQu Wenruo 14496426c7adSQu Wenruo /* 145052042d8eSAndrea Gelmini * Ensure dirty @src will be committed. Or, after coming 14514d31778aSQu Wenruo * commit_fs_roots() and switch_commit_roots(), any dirty but not 14524d31778aSQu Wenruo * recorded root will never be updated again, causing an outdated root 14534d31778aSQu Wenruo * item. 14544d31778aSQu Wenruo */ 14551c442d22SJosef Bacik ret = record_root_in_trans(trans, src, 1); 14561c442d22SJosef Bacik if (ret) 14571c442d22SJosef Bacik return ret; 14584d31778aSQu Wenruo 14594d31778aSQu Wenruo /* 14602a4d84c1SJosef Bacik * btrfs_qgroup_inherit relies on a consistent view of the usage for the 14612a4d84c1SJosef Bacik * src root, so we must run the delayed refs here. 14622a4d84c1SJosef Bacik * 14632a4d84c1SJosef Bacik * However this isn't particularly fool proof, because there's no 14642a4d84c1SJosef Bacik * synchronization keeping us from changing the tree after this point 14652a4d84c1SJosef Bacik * before we do the qgroup_inherit, or even from making changes while 14662a4d84c1SJosef Bacik * we're doing the qgroup_inherit. But that's a problem for the future, 14672a4d84c1SJosef Bacik * for now flush the delayed refs to narrow the race window where the 14682a4d84c1SJosef Bacik * qgroup counters could end up wrong. 14692a4d84c1SJosef Bacik */ 14702a4d84c1SJosef Bacik ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 14712a4d84c1SJosef Bacik if (ret) { 14722a4d84c1SJosef Bacik btrfs_abort_transaction(trans, ret); 147344365827SNaohiro Aota return ret; 14742a4d84c1SJosef Bacik } 14752a4d84c1SJosef Bacik 14762a4d84c1SJosef Bacik /* 14776426c7adSQu Wenruo * We are going to commit transaction, see btrfs_commit_transaction() 14786426c7adSQu Wenruo * comment for reason locking tree_log_mutex 14796426c7adSQu Wenruo */ 14806426c7adSQu Wenruo mutex_lock(&fs_info->tree_log_mutex); 14816426c7adSQu Wenruo 14827e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 14836426c7adSQu Wenruo if (ret) 14846426c7adSQu Wenruo goto out; 1485460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 14866426c7adSQu Wenruo if (ret < 0) 14876426c7adSQu Wenruo goto out; 14886426c7adSQu Wenruo 14896426c7adSQu Wenruo /* Now qgroup are all updated, we can inherit it to new qgroups */ 1490a9377422SLu Fengqi ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, 14916426c7adSQu Wenruo inherit); 14926426c7adSQu Wenruo if (ret < 0) 14936426c7adSQu Wenruo goto out; 14946426c7adSQu Wenruo 14956426c7adSQu Wenruo /* 14966426c7adSQu Wenruo * Now we do a simplified commit transaction, which will: 14976426c7adSQu Wenruo * 1) commit all subvolume and extent tree 14986426c7adSQu Wenruo * To ensure all subvolume and extent tree have a valid 14996426c7adSQu Wenruo * commit_root to accounting later insert_dir_item() 15006426c7adSQu Wenruo * 2) write all btree blocks onto disk 15016426c7adSQu Wenruo * This is to make sure later btree modification will be cowed 15026426c7adSQu Wenruo * Or commit_root can be populated and cause wrong qgroup numbers 15036426c7adSQu Wenruo * In this simplified commit, we don't really care about other trees 15046426c7adSQu Wenruo * like chunk and root tree, as they won't affect qgroup. 15056426c7adSQu Wenruo * And we don't write super to avoid half committed status. 15066426c7adSQu Wenruo */ 15079386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 15086426c7adSQu Wenruo if (ret) 15096426c7adSQu Wenruo goto out; 1510889bfa39SJosef Bacik switch_commit_roots(trans); 151170458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 15126426c7adSQu Wenruo if (ret) 1513f7af3934SDavid Sterba btrfs_handle_fs_error(fs_info, ret, 15146426c7adSQu Wenruo "Error while writing out transaction for qgroup"); 15156426c7adSQu Wenruo 15166426c7adSQu Wenruo out: 15176426c7adSQu Wenruo mutex_unlock(&fs_info->tree_log_mutex); 15186426c7adSQu Wenruo 15196426c7adSQu Wenruo /* 15206426c7adSQu Wenruo * Force parent root to be updated, as we recorded it before so its 15216426c7adSQu Wenruo * last_trans == cur_transid. 15226426c7adSQu Wenruo * Or it won't be committed again onto disk after later 15236426c7adSQu Wenruo * insert_dir_item() 15246426c7adSQu Wenruo */ 15256426c7adSQu Wenruo if (!ret) 15261c442d22SJosef Bacik ret = record_root_in_trans(trans, parent, 1); 15276426c7adSQu Wenruo return ret; 15286426c7adSQu Wenruo } 15296426c7adSQu Wenruo 15306426c7adSQu Wenruo /* 1531d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 1532aec8030aSMiao Xie * transaction commit. This does the actual creation. 1533aec8030aSMiao Xie * 1534aec8030aSMiao Xie * Note: 1535aec8030aSMiao Xie * If the error which may affect the commitment of the current transaction 1536aec8030aSMiao Xie * happens, we should return the error number. If the error which just affect 1537aec8030aSMiao Xie * the creation of the pending snapshots, just return 0. 1538d352ac68SChris Mason */ 153980b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 15403063d29fSChris Mason struct btrfs_pending_snapshot *pending) 15413063d29fSChris Mason { 154208d50ca3SNikolay Borisov 154308d50ca3SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 15443063d29fSChris Mason struct btrfs_key key; 154580b6794dSChris Mason struct btrfs_root_item *new_root_item; 15463063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 15473063d29fSChris Mason struct btrfs_root *root = pending->root; 15486bdb72deSSage Weil struct btrfs_root *parent_root; 154998c9942aSLiu Bo struct btrfs_block_rsv *rsv; 15506bdb72deSSage Weil struct inode *parent_inode; 155142874b3dSMiao Xie struct btrfs_path *path; 155242874b3dSMiao Xie struct btrfs_dir_item *dir_item; 1553a22285a6SYan, Zheng struct dentry *dentry; 15543063d29fSChris Mason struct extent_buffer *tmp; 1555925baeddSChris Mason struct extent_buffer *old; 155695582b00SDeepa Dinamani struct timespec64 cur_time; 1557aec8030aSMiao Xie int ret = 0; 1558d68fc57bSYan, Zheng u64 to_reserve = 0; 15596bdb72deSSage Weil u64 index = 0; 1560a22285a6SYan, Zheng u64 objectid; 1561b83cc969SLi Zefan u64 root_flags; 15623063d29fSChris Mason 15638546b570SDavid Sterba ASSERT(pending->path); 15648546b570SDavid Sterba path = pending->path; 156542874b3dSMiao Xie 1566b0c0ea63SDavid Sterba ASSERT(pending->root_item); 1567b0c0ea63SDavid Sterba new_root_item = pending->root_item; 1568a22285a6SYan, Zheng 1569543068a2SNikolay Borisov pending->error = btrfs_get_free_objectid(tree_root, &objectid); 1570aec8030aSMiao Xie if (pending->error) 15716fa9700eSMiao Xie goto no_free_objectid; 15723063d29fSChris Mason 1573d6726335SQu Wenruo /* 1574d6726335SQu Wenruo * Make qgroup to skip current new snapshot's qgroupid, as it is 1575d6726335SQu Wenruo * accounted by later btrfs_qgroup_inherit(). 1576d6726335SQu Wenruo */ 1577d6726335SQu Wenruo btrfs_set_skip_qgroup(trans, objectid); 1578d6726335SQu Wenruo 1579147d256eSZhaolei btrfs_reloc_pre_snapshot(pending, &to_reserve); 1580d68fc57bSYan, Zheng 1581d68fc57bSYan, Zheng if (to_reserve > 0) { 1582aec8030aSMiao Xie pending->error = btrfs_block_rsv_add(root, 1583aec8030aSMiao Xie &pending->block_rsv, 158408e007d2SMiao Xie to_reserve, 158508e007d2SMiao Xie BTRFS_RESERVE_NO_FLUSH); 1586aec8030aSMiao Xie if (pending->error) 1587d6726335SQu Wenruo goto clear_skip_qgroup; 1588d68fc57bSYan, Zheng } 1589d68fc57bSYan, Zheng 15903063d29fSChris Mason key.objectid = objectid; 1591a22285a6SYan, Zheng key.offset = (u64)-1; 1592a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 15933063d29fSChris Mason 15946fa9700eSMiao Xie rsv = trans->block_rsv; 1595a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 15962382c5ccSLiu Bo trans->bytes_reserved = trans->block_rsv->reserved; 15970b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 159888d3a5aaSJosef Bacik trans->transid, 159988d3a5aaSJosef Bacik trans->bytes_reserved, 1); 1600a22285a6SYan, Zheng dentry = pending->dentry; 1601e9662f70SMiao Xie parent_inode = pending->dir; 1602a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 1603f0118cb6SJosef Bacik ret = record_root_in_trans(trans, parent_root, 0); 1604f0118cb6SJosef Bacik if (ret) 1605f0118cb6SJosef Bacik goto fail; 1606c2050a45SDeepa Dinamani cur_time = current_time(parent_inode); 160704b285f3SDeepa Dinamani 16086bdb72deSSage Weil /* 16096bdb72deSSage Weil * insert the directory item 16106bdb72deSSage Weil */ 1611877574e2SNikolay Borisov ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); 161249b25e05SJeff Mahoney BUG_ON(ret); /* -ENOMEM */ 161342874b3dSMiao Xie 161442874b3dSMiao Xie /* check if there is a file/dir which has the same name. */ 161542874b3dSMiao Xie dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, 16164a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), 161742874b3dSMiao Xie dentry->d_name.name, 161842874b3dSMiao Xie dentry->d_name.len, 0); 161942874b3dSMiao Xie if (dir_item != NULL && !IS_ERR(dir_item)) { 1620fe66a05aSChris Mason pending->error = -EEXIST; 1621aec8030aSMiao Xie goto dir_item_existed; 162242874b3dSMiao Xie } else if (IS_ERR(dir_item)) { 162342874b3dSMiao Xie ret = PTR_ERR(dir_item); 162466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16258732d44fSMiao Xie goto fail; 162679787eaaSJeff Mahoney } 162742874b3dSMiao Xie btrfs_release_path(path); 16286bdb72deSSage Weil 1629e999376fSChris Mason /* 1630e999376fSChris Mason * pull in the delayed directory update 1631e999376fSChris Mason * and the delayed inode item 1632e999376fSChris Mason * otherwise we corrupt the FS during 1633e999376fSChris Mason * snapshot 1634e999376fSChris Mason */ 1635e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 16368732d44fSMiao Xie if (ret) { /* Transaction aborted */ 163766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16388732d44fSMiao Xie goto fail; 16398732d44fSMiao Xie } 1640e999376fSChris Mason 1641f0118cb6SJosef Bacik ret = record_root_in_trans(trans, root, 0); 1642f0118cb6SJosef Bacik if (ret) { 1643f0118cb6SJosef Bacik btrfs_abort_transaction(trans, ret); 1644f0118cb6SJosef Bacik goto fail; 1645f0118cb6SJosef Bacik } 16466bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 16476bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 164808fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 16496bdb72deSSage Weil 1650b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 1651b83cc969SLi Zefan if (pending->readonly) 1652b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 1653b83cc969SLi Zefan else 1654b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 1655b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 1656b83cc969SLi Zefan 16578ea05e3aSAlexander Block btrfs_set_root_generation_v2(new_root_item, 16588ea05e3aSAlexander Block trans->transid); 1659807fc790SAndy Shevchenko generate_random_guid(new_root_item->uuid); 16608ea05e3aSAlexander Block memcpy(new_root_item->parent_uuid, root->root_item.uuid, 16618ea05e3aSAlexander Block BTRFS_UUID_SIZE); 166270023da2SStefan Behrens if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { 166370023da2SStefan Behrens memset(new_root_item->received_uuid, 0, 166470023da2SStefan Behrens sizeof(new_root_item->received_uuid)); 16658ea05e3aSAlexander Block memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); 16668ea05e3aSAlexander Block memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); 16678ea05e3aSAlexander Block btrfs_set_root_stransid(new_root_item, 0); 16688ea05e3aSAlexander Block btrfs_set_root_rtransid(new_root_item, 0); 166970023da2SStefan Behrens } 16703cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); 16713cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); 167270023da2SStefan Behrens btrfs_set_root_otransid(new_root_item, trans->transid); 16738ea05e3aSAlexander Block 1674925baeddSChris Mason old = btrfs_lock_root_node(root); 16759631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, 16769631e4ccSJosef Bacik BTRFS_NESTING_COW); 167779787eaaSJeff Mahoney if (ret) { 167879787eaaSJeff Mahoney btrfs_tree_unlock(old); 167979787eaaSJeff Mahoney free_extent_buffer(old); 168066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16818732d44fSMiao Xie goto fail; 168279787eaaSJeff Mahoney } 168349b25e05SJeff Mahoney 168449b25e05SJeff Mahoney ret = btrfs_copy_root(trans, root, old, &tmp, objectid); 168579787eaaSJeff Mahoney /* clean up in any case */ 1686925baeddSChris Mason btrfs_tree_unlock(old); 1687925baeddSChris Mason free_extent_buffer(old); 16888732d44fSMiao Xie if (ret) { 168966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16908732d44fSMiao Xie goto fail; 16918732d44fSMiao Xie } 1692f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 169327cdeb70SMiao Xie set_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1694f1ebcc74SLiu Bo smp_wmb(); 1695f1ebcc74SLiu Bo 16965d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 1697a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 1698a22285a6SYan, Zheng key.offset = trans->transid; 1699a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1700925baeddSChris Mason btrfs_tree_unlock(tmp); 17013063d29fSChris Mason free_extent_buffer(tmp); 17028732d44fSMiao Xie if (ret) { 170366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17048732d44fSMiao Xie goto fail; 17058732d44fSMiao Xie } 17060660b5afSChris Mason 1707a22285a6SYan, Zheng /* 1708a22285a6SYan, Zheng * insert root back/forward references 1709a22285a6SYan, Zheng */ 17106025c19fSLu Fengqi ret = btrfs_add_root_ref(trans, objectid, 1711a22285a6SYan, Zheng parent_root->root_key.objectid, 17124a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), index, 1713a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len); 17148732d44fSMiao Xie if (ret) { 171566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17168732d44fSMiao Xie goto fail; 17178732d44fSMiao Xie } 1718a22285a6SYan, Zheng 1719a22285a6SYan, Zheng key.offset = (u64)-1; 17202dfb1e43SQu Wenruo pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); 172179787eaaSJeff Mahoney if (IS_ERR(pending->snap)) { 172279787eaaSJeff Mahoney ret = PTR_ERR(pending->snap); 17232d892ccdSFilipe Manana pending->snap = NULL; 172466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17258732d44fSMiao Xie goto fail; 172679787eaaSJeff Mahoney } 1727d68fc57bSYan, Zheng 172849b25e05SJeff Mahoney ret = btrfs_reloc_post_snapshot(trans, pending); 17298732d44fSMiao Xie if (ret) { 173066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17318732d44fSMiao Xie goto fail; 17328732d44fSMiao Xie } 1733361048f5SMiao Xie 17346426c7adSQu Wenruo /* 17356426c7adSQu Wenruo * Do special qgroup accounting for snapshot, as we do some qgroup 17366426c7adSQu Wenruo * snapshot hack to do fast snapshot. 17376426c7adSQu Wenruo * To co-operate with that hack, we do hack again. 17386426c7adSQu Wenruo * Or snapshot will be greatly slowed down by a subtree qgroup rescan 17396426c7adSQu Wenruo */ 17406426c7adSQu Wenruo ret = qgroup_account_snapshot(trans, root, parent_root, 17416426c7adSQu Wenruo pending->inherit, objectid); 17426426c7adSQu Wenruo if (ret < 0) 17436426c7adSQu Wenruo goto fail; 17446426c7adSQu Wenruo 1745684572dfSLu Fengqi ret = btrfs_insert_dir_item(trans, dentry->d_name.name, 1746684572dfSLu Fengqi dentry->d_name.len, BTRFS_I(parent_inode), 1747684572dfSLu Fengqi &key, BTRFS_FT_DIR, index); 174842874b3dSMiao Xie /* We have check then name at the beginning, so it is impossible. */ 17499c52057cSChris Mason BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); 17508732d44fSMiao Xie if (ret) { 175166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17528732d44fSMiao Xie goto fail; 17538732d44fSMiao Xie } 175442874b3dSMiao Xie 17556ef06d27SNikolay Borisov btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + 175642874b3dSMiao Xie dentry->d_name.len * 2); 175704b285f3SDeepa Dinamani parent_inode->i_mtime = parent_inode->i_ctime = 1758c2050a45SDeepa Dinamani current_time(parent_inode); 1759729f7961SNikolay Borisov ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); 1760dd5f9615SStefan Behrens if (ret) { 176166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1762dd5f9615SStefan Behrens goto fail; 1763dd5f9615SStefan Behrens } 1764807fc790SAndy Shevchenko ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, 1765807fc790SAndy Shevchenko BTRFS_UUID_KEY_SUBVOL, 1766cdb345a8SLu Fengqi objectid); 1767dd5f9615SStefan Behrens if (ret) { 176866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1769dd5f9615SStefan Behrens goto fail; 1770dd5f9615SStefan Behrens } 1771dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { 1772cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, 1773dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 1774dd5f9615SStefan Behrens objectid); 1775dd5f9615SStefan Behrens if (ret && ret != -EEXIST) { 177666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1777dd5f9615SStefan Behrens goto fail; 1778dd5f9615SStefan Behrens } 1779dd5f9615SStefan Behrens } 1780d6726335SQu Wenruo 17813063d29fSChris Mason fail: 1782aec8030aSMiao Xie pending->error = ret; 1783aec8030aSMiao Xie dir_item_existed: 178498c9942aSLiu Bo trans->block_rsv = rsv; 17852382c5ccSLiu Bo trans->bytes_reserved = 0; 1786d6726335SQu Wenruo clear_skip_qgroup: 1787d6726335SQu Wenruo btrfs_clear_skip_qgroup(trans); 17886fa9700eSMiao Xie no_free_objectid: 17896fa9700eSMiao Xie kfree(new_root_item); 1790b0c0ea63SDavid Sterba pending->root_item = NULL; 179142874b3dSMiao Xie btrfs_free_path(path); 17928546b570SDavid Sterba pending->path = NULL; 17938546b570SDavid Sterba 179449b25e05SJeff Mahoney return ret; 17953063d29fSChris Mason } 17963063d29fSChris Mason 1797d352ac68SChris Mason /* 1798d352ac68SChris Mason * create all the snapshots we've scheduled for creation 1799d352ac68SChris Mason */ 180008d50ca3SNikolay Borisov static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) 18013063d29fSChris Mason { 1802aec8030aSMiao Xie struct btrfs_pending_snapshot *pending, *next; 18033063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 1804aec8030aSMiao Xie int ret = 0; 18053de4586cSChris Mason 1806aec8030aSMiao Xie list_for_each_entry_safe(pending, next, head, list) { 1807aec8030aSMiao Xie list_del(&pending->list); 180808d50ca3SNikolay Borisov ret = create_pending_snapshot(trans, pending); 1809aec8030aSMiao Xie if (ret) 1810aec8030aSMiao Xie break; 1811aec8030aSMiao Xie } 1812aec8030aSMiao Xie return ret; 18133de4586cSChris Mason } 18143de4586cSChris Mason 18152ff7e61eSJeff Mahoney static void update_super_roots(struct btrfs_fs_info *fs_info) 18165d4f98a2SYan Zheng { 18175d4f98a2SYan Zheng struct btrfs_root_item *root_item; 18185d4f98a2SYan Zheng struct btrfs_super_block *super; 18195d4f98a2SYan Zheng 18200b246afaSJeff Mahoney super = fs_info->super_copy; 18215d4f98a2SYan Zheng 18220b246afaSJeff Mahoney root_item = &fs_info->chunk_root->root_item; 1823093e037cSDavid Sterba super->chunk_root = root_item->bytenr; 1824093e037cSDavid Sterba super->chunk_root_generation = root_item->generation; 1825093e037cSDavid Sterba super->chunk_root_level = root_item->level; 18265d4f98a2SYan Zheng 18270b246afaSJeff Mahoney root_item = &fs_info->tree_root->root_item; 1828093e037cSDavid Sterba super->root = root_item->bytenr; 1829093e037cSDavid Sterba super->generation = root_item->generation; 1830093e037cSDavid Sterba super->root_level = root_item->level; 18310b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, SPACE_CACHE)) 1832093e037cSDavid Sterba super->cache_generation = root_item->generation; 183394846229SBoris Burkov else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) 183494846229SBoris Burkov super->cache_generation = 0; 18350b246afaSJeff Mahoney if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) 1836093e037cSDavid Sterba super->uuid_tree_generation = root_item->generation; 18375d4f98a2SYan Zheng } 18385d4f98a2SYan Zheng 1839f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1840f36f3042SChris Mason { 18414a9d8bdeSMiao Xie struct btrfs_transaction *trans; 1842f36f3042SChris Mason int ret = 0; 18434a9d8bdeSMiao Xie 1844a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 18454a9d8bdeSMiao Xie trans = info->running_transaction; 18464a9d8bdeSMiao Xie if (trans) 18474a9d8bdeSMiao Xie ret = (trans->state >= TRANS_STATE_COMMIT_START); 1848a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 1849f36f3042SChris Mason return ret; 1850f36f3042SChris Mason } 1851f36f3042SChris Mason 18528929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 18538929ecfaSYan, Zheng { 18544a9d8bdeSMiao Xie struct btrfs_transaction *trans; 18558929ecfaSYan, Zheng int ret = 0; 18564a9d8bdeSMiao Xie 1857a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 18584a9d8bdeSMiao Xie trans = info->running_transaction; 18594a9d8bdeSMiao Xie if (trans) 18604a9d8bdeSMiao Xie ret = is_transaction_blocked(trans); 1861a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 18628929ecfaSYan, Zheng return ret; 18638929ecfaSYan, Zheng } 18648929ecfaSYan, Zheng 1865bb9c12c9SSage Weil /* 1866bb9c12c9SSage Weil * commit transactions asynchronously. once btrfs_commit_transaction_async 1867bb9c12c9SSage Weil * returns, any subsequent transaction will not be allowed to join. 1868bb9c12c9SSage Weil */ 1869bb9c12c9SSage Weil struct btrfs_async_commit { 1870bb9c12c9SSage Weil struct btrfs_trans_handle *newtrans; 18717892b5afSMiao Xie struct work_struct work; 1872bb9c12c9SSage Weil }; 1873bb9c12c9SSage Weil 1874bb9c12c9SSage Weil static void do_async_commit(struct work_struct *work) 1875bb9c12c9SSage Weil { 1876bb9c12c9SSage Weil struct btrfs_async_commit *ac = 18777892b5afSMiao Xie container_of(work, struct btrfs_async_commit, work); 1878bb9c12c9SSage Weil 18796fc4e354SSage Weil /* 18806fc4e354SSage Weil * We've got freeze protection passed with the transaction. 18816fc4e354SSage Weil * Tell lockdep about it. 18826fc4e354SSage Weil */ 1883b1a06a4bSLiu Bo if (ac->newtrans->type & __TRANS_FREEZABLE) 18843a45bb20SJeff Mahoney __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS); 18856fc4e354SSage Weil 1886e209db7aSSage Weil current->journal_info = ac->newtrans; 1887e209db7aSSage Weil 18883a45bb20SJeff Mahoney btrfs_commit_transaction(ac->newtrans); 1889bb9c12c9SSage Weil kfree(ac); 1890bb9c12c9SSage Weil } 1891bb9c12c9SSage Weil 189232cc4f87SDavid Sterba int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans) 1893bb9c12c9SSage Weil { 18943a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 1895bb9c12c9SSage Weil struct btrfs_async_commit *ac; 1896bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1897bb9c12c9SSage Weil 1898bb9c12c9SSage Weil ac = kmalloc(sizeof(*ac), GFP_NOFS); 1899db5b493aSTsutomu Itoh if (!ac) 1900db5b493aSTsutomu Itoh return -ENOMEM; 1901bb9c12c9SSage Weil 19027892b5afSMiao Xie INIT_WORK(&ac->work, do_async_commit); 19033a45bb20SJeff Mahoney ac->newtrans = btrfs_join_transaction(trans->root); 19043612b495STsutomu Itoh if (IS_ERR(ac->newtrans)) { 19053612b495STsutomu Itoh int err = PTR_ERR(ac->newtrans); 19063612b495STsutomu Itoh kfree(ac); 19073612b495STsutomu Itoh return err; 19083612b495STsutomu Itoh } 1909bb9c12c9SSage Weil 1910bb9c12c9SSage Weil /* take transaction reference */ 1911bb9c12c9SSage Weil cur_trans = trans->transaction; 19129b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 1913bb9c12c9SSage Weil 19143a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19156fc4e354SSage Weil 19166fc4e354SSage Weil /* 19176fc4e354SSage Weil * Tell lockdep we've released the freeze rwsem, since the 19186fc4e354SSage Weil * async commit thread will be the one to unlock it. 19196fc4e354SSage Weil */ 1920b1a06a4bSLiu Bo if (ac->newtrans->type & __TRANS_FREEZABLE) 19210b246afaSJeff Mahoney __sb_writers_release(fs_info->sb, SB_FREEZE_FS); 19226fc4e354SSage Weil 19237892b5afSMiao Xie schedule_work(&ac->work); 1924ae5d29d4SDavid Sterba /* 1925ae5d29d4SDavid Sterba * Wait for the current transaction commit to start and block 1926ae5d29d4SDavid Sterba * subsequent transaction joins 1927ae5d29d4SDavid Sterba */ 1928ae5d29d4SDavid Sterba wait_event(fs_info->transaction_blocked_wait, 1929ae5d29d4SDavid Sterba cur_trans->state >= TRANS_STATE_COMMIT_START || 1930ae5d29d4SDavid Sterba TRANS_ABORTED(cur_trans)); 193138e88054SSage Weil if (current->journal_info == trans) 193238e88054SSage Weil current->journal_info = NULL; 193338e88054SSage Weil 1934724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1935bb9c12c9SSage Weil return 0; 1936bb9c12c9SSage Weil } 1937bb9c12c9SSage Weil 193849b25e05SJeff Mahoney 193997cb39bbSNikolay Borisov static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) 194049b25e05SJeff Mahoney { 194197cb39bbSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 194249b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 194349b25e05SJeff Mahoney 1944b50fff81SDavid Sterba WARN_ON(refcount_read(&trans->use_count) > 1); 194549b25e05SJeff Mahoney 194666642832SJeff Mahoney btrfs_abort_transaction(trans, err); 19477b8b92afSJosef Bacik 19480b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 194966b6135bSLiu Bo 195025d8c284SMiao Xie /* 195125d8c284SMiao Xie * If the transaction is removed from the list, it means this 195225d8c284SMiao Xie * transaction has been committed successfully, so it is impossible 195325d8c284SMiao Xie * to call the cleanup function. 195425d8c284SMiao Xie */ 195525d8c284SMiao Xie BUG_ON(list_empty(&cur_trans->list)); 195666b6135bSLiu Bo 19570b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) { 19584a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 19590b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1960f094ac32SLiu Bo wait_event(cur_trans->writer_wait, 1961f094ac32SLiu Bo atomic_read(&cur_trans->num_writers) == 1); 1962f094ac32SLiu Bo 19630b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1964d7096fc3SJosef Bacik } 1965061dde82SFilipe Manana 1966061dde82SFilipe Manana /* 1967061dde82SFilipe Manana * Now that we know no one else is still using the transaction we can 1968061dde82SFilipe Manana * remove the transaction from the list of transactions. This avoids 1969061dde82SFilipe Manana * the transaction kthread from cleaning up the transaction while some 1970061dde82SFilipe Manana * other task is still using it, which could result in a use-after-free 1971061dde82SFilipe Manana * on things like log trees, as it forces the transaction kthread to 1972061dde82SFilipe Manana * wait for this transaction to be cleaned up by us. 1973061dde82SFilipe Manana */ 1974061dde82SFilipe Manana list_del_init(&cur_trans->list); 1975061dde82SFilipe Manana 19760b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 197749b25e05SJeff Mahoney 19782ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(trans->transaction, fs_info); 197949b25e05SJeff Mahoney 19800b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 19810b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) 19820b246afaSJeff Mahoney fs_info->running_transaction = NULL; 19830b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 19844a9d8bdeSMiao Xie 1985e0228285SJosef Bacik if (trans->type & __TRANS_FREEZABLE) 19860b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 1987724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1988724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 198949b25e05SJeff Mahoney 199097cb39bbSNikolay Borisov trace_btrfs_transaction_commit(trans->root); 199149b25e05SJeff Mahoney 199249b25e05SJeff Mahoney if (current->journal_info == trans) 199349b25e05SJeff Mahoney current->journal_info = NULL; 19940b246afaSJeff Mahoney btrfs_scrub_cancel(fs_info); 199549b25e05SJeff Mahoney 199649b25e05SJeff Mahoney kmem_cache_free(btrfs_trans_handle_cachep, trans); 199749b25e05SJeff Mahoney } 199849b25e05SJeff Mahoney 1999c7cc64a9SDavid Sterba /* 2000c7cc64a9SDavid Sterba * Release reserved delayed ref space of all pending block groups of the 2001c7cc64a9SDavid Sterba * transaction and remove them from the list 2002c7cc64a9SDavid Sterba */ 2003c7cc64a9SDavid Sterba static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) 2004c7cc64a9SDavid Sterba { 2005c7cc64a9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 200632da5386SDavid Sterba struct btrfs_block_group *block_group, *tmp; 2007c7cc64a9SDavid Sterba 2008c7cc64a9SDavid Sterba list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { 2009c7cc64a9SDavid Sterba btrfs_delayed_refs_rsv_release(fs_info, 1); 2010c7cc64a9SDavid Sterba list_del_init(&block_group->bg_list); 2011c7cc64a9SDavid Sterba } 2012c7cc64a9SDavid Sterba } 2013c7cc64a9SDavid Sterba 201488090ad3SFilipe Manana static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) 201582436617SMiao Xie { 2016ce8ea7ccSJosef Bacik /* 2017ce8ea7ccSJosef Bacik * We use writeback_inodes_sb here because if we used 2018ce8ea7ccSJosef Bacik * btrfs_start_delalloc_roots we would deadlock with fs freeze. 2019ce8ea7ccSJosef Bacik * Currently are holding the fs freeze lock, if we do an async flush 2020ce8ea7ccSJosef Bacik * we'll do btrfs_join_transaction() and deadlock because we need to 2021ce8ea7ccSJosef Bacik * wait for the fs freeze lock. Using the direct flushing we benefit 2022ce8ea7ccSJosef Bacik * from already being in a transaction and our join_transaction doesn't 2023ce8ea7ccSJosef Bacik * have to re-take the fs freeze lock. 2024ce8ea7ccSJosef Bacik */ 202588090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 2026ce8ea7ccSJosef Bacik writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); 202782436617SMiao Xie return 0; 202882436617SMiao Xie } 202982436617SMiao Xie 203088090ad3SFilipe Manana static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) 203182436617SMiao Xie { 203288090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 20336374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 203482436617SMiao Xie } 203582436617SMiao Xie 20363a45bb20SJeff Mahoney int btrfs_commit_transaction(struct btrfs_trans_handle *trans) 203779154b1bSChris Mason { 20383a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 203949b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 20408fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 204125287e0aSMiao Xie int ret; 204279154b1bSChris Mason 204335b814f3SNikolay Borisov ASSERT(refcount_read(&trans->use_count) == 1); 204435b814f3SNikolay Borisov 20458d25a086SMiao Xie /* Stop the commit early if ->aborted is set */ 2046bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 204725287e0aSMiao Xie ret = cur_trans->aborted; 20483a45bb20SJeff Mahoney btrfs_end_transaction(trans); 2049e4a2bcacSJosef Bacik return ret; 205025287e0aSMiao Xie } 205149b25e05SJeff Mahoney 2052f45c752bSJosef Bacik btrfs_trans_release_metadata(trans); 2053f45c752bSJosef Bacik trans->block_rsv = NULL; 2054f45c752bSJosef Bacik 2055e19eb11fSJosef Bacik /* 2056e19eb11fSJosef Bacik * We only want one transaction commit doing the flushing so we do not 2057e19eb11fSJosef Bacik * waste a bunch of time on lock contention on the extent root node. 2058e19eb11fSJosef Bacik */ 2059e19eb11fSJosef Bacik if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, 2060e19eb11fSJosef Bacik &cur_trans->delayed_refs.flags)) { 2061e19eb11fSJosef Bacik /* 2062e19eb11fSJosef Bacik * Make a pass through all the delayed refs we have so far. 2063e19eb11fSJosef Bacik * Any running threads may add more while we are here. 206456bec294SChris Mason */ 2065c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, 0); 2066e4a2bcacSJosef Bacik if (ret) { 20673a45bb20SJeff Mahoney btrfs_end_transaction(trans); 2068e4a2bcacSJosef Bacik return ret; 2069e4a2bcacSJosef Bacik } 2070e19eb11fSJosef Bacik } 207156bec294SChris Mason 20726c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 2073ea658badSJosef Bacik 20743204d33cSJosef Bacik if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { 20751bbc621eSChris Mason int run_it = 0; 20761bbc621eSChris Mason 20771bbc621eSChris Mason /* this mutex is also taken before trying to set 20781bbc621eSChris Mason * block groups readonly. We need to make sure 20791bbc621eSChris Mason * that nobody has set a block group readonly 20801bbc621eSChris Mason * after a extents from that block group have been 20811bbc621eSChris Mason * allocated for cache files. btrfs_set_block_group_ro 20821bbc621eSChris Mason * will wait for the transaction to commit if it 20833204d33cSJosef Bacik * finds BTRFS_TRANS_DIRTY_BG_RUN set. 20841bbc621eSChris Mason * 20853204d33cSJosef Bacik * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure 20863204d33cSJosef Bacik * only one process starts all the block group IO. It wouldn't 20871bbc621eSChris Mason * hurt to have more than one go through, but there's no 20881bbc621eSChris Mason * real advantage to it either. 20891bbc621eSChris Mason */ 20900b246afaSJeff Mahoney mutex_lock(&fs_info->ro_block_group_mutex); 20913204d33cSJosef Bacik if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, 20923204d33cSJosef Bacik &cur_trans->flags)) 20931bbc621eSChris Mason run_it = 1; 20940b246afaSJeff Mahoney mutex_unlock(&fs_info->ro_block_group_mutex); 20951bbc621eSChris Mason 2096f9cacae3SNikolay Borisov if (run_it) { 209721217054SNikolay Borisov ret = btrfs_start_dirty_block_groups(trans); 20981bbc621eSChris Mason if (ret) { 20993a45bb20SJeff Mahoney btrfs_end_transaction(trans); 21001bbc621eSChris Mason return ret; 21011bbc621eSChris Mason } 2102f9cacae3SNikolay Borisov } 2103f9cacae3SNikolay Borisov } 21041bbc621eSChris Mason 21050b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 21064a9d8bdeSMiao Xie if (cur_trans->state >= TRANS_STATE_COMMIT_START) { 2107d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2108d0c2f4faSFilipe Manana 21090b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 21109b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 2111ccd467d6SChris Mason 2112d0c2f4faSFilipe Manana if (trans->in_fsync) 2113d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2114d0c2f4faSFilipe Manana ret = btrfs_end_transaction(trans); 2115d0c2f4faSFilipe Manana wait_for_commit(cur_trans, want_state); 211615ee9bc7SJosef Bacik 2117bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) 2118b4924a0fSLiu Bo ret = cur_trans->aborted; 2119b4924a0fSLiu Bo 2120724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 212115ee9bc7SJosef Bacik 212249b25e05SJeff Mahoney return ret; 212379154b1bSChris Mason } 21244313b399SChris Mason 21254a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 21260b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 2127bb9c12c9SSage Weil 21280b246afaSJeff Mahoney if (cur_trans->list.prev != &fs_info->trans_list) { 2129d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2130d0c2f4faSFilipe Manana 2131d0c2f4faSFilipe Manana if (trans->in_fsync) 2132d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2133d0c2f4faSFilipe Manana 2134ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 2135ccd467d6SChris Mason struct btrfs_transaction, list); 2136d0c2f4faSFilipe Manana if (prev_trans->state < want_state) { 21379b64f57dSElena Reshetova refcount_inc(&prev_trans->use_count); 21380b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2139ccd467d6SChris Mason 2140d0c2f4faSFilipe Manana wait_for_commit(prev_trans, want_state); 2141d0c2f4faSFilipe Manana 2142bf31f87fSDavid Sterba ret = READ_ONCE(prev_trans->aborted); 2143ccd467d6SChris Mason 2144724e2315SJosef Bacik btrfs_put_transaction(prev_trans); 21451f9b8c8fSFilipe Manana if (ret) 21461f9b8c8fSFilipe Manana goto cleanup_transaction; 2147a4abeea4SJosef Bacik } else { 21480b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2149ccd467d6SChris Mason } 2150a4abeea4SJosef Bacik } else { 21510b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2152cb2d3dadSFilipe Manana /* 2153cb2d3dadSFilipe Manana * The previous transaction was aborted and was already removed 2154cb2d3dadSFilipe Manana * from the list of transactions at fs_info->trans_list. So we 2155cb2d3dadSFilipe Manana * abort to prevent writing a new superblock that reflects a 2156cb2d3dadSFilipe Manana * corrupt state (pointing to trees with unwritten nodes/leafs). 2157cb2d3dadSFilipe Manana */ 2158cb2d3dadSFilipe Manana if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) { 2159cb2d3dadSFilipe Manana ret = -EROFS; 2160cb2d3dadSFilipe Manana goto cleanup_transaction; 2161cb2d3dadSFilipe Manana } 2162ccd467d6SChris Mason } 216315ee9bc7SJosef Bacik 21640860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 21650860adfdSMiao Xie 216688090ad3SFilipe Manana ret = btrfs_start_delalloc_flush(fs_info); 216782436617SMiao Xie if (ret) 216882436617SMiao Xie goto cleanup_transaction; 216982436617SMiao Xie 2170e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 217149b25e05SJeff Mahoney if (ret) 217249b25e05SJeff Mahoney goto cleanup_transaction; 217316cdcec7SMiao Xie 2174581227d0SMiao Xie wait_event(cur_trans->writer_wait, 2175581227d0SMiao Xie extwriter_counter_read(cur_trans) == 0); 2176ed3b3d31SChris Mason 2177581227d0SMiao Xie /* some pending stuffs might be added after the previous flush. */ 2178e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 2179ca469637SMiao Xie if (ret) 2180ca469637SMiao Xie goto cleanup_transaction; 2181ca469637SMiao Xie 218288090ad3SFilipe Manana btrfs_wait_delalloc_flush(fs_info); 2183cb7ab021SWang Shilong 218448778179SFilipe Manana /* 218548778179SFilipe Manana * Wait for all ordered extents started by a fast fsync that joined this 218648778179SFilipe Manana * transaction. Otherwise if this transaction commits before the ordered 218748778179SFilipe Manana * extents complete we lose logged data after a power failure. 218848778179SFilipe Manana */ 218948778179SFilipe Manana wait_event(cur_trans->pending_wait, 219048778179SFilipe Manana atomic_read(&cur_trans->pending_ordered) == 0); 219148778179SFilipe Manana 21922ff7e61eSJeff Mahoney btrfs_scrub_pause(fs_info); 2193ed0ca140SJosef Bacik /* 2194ed0ca140SJosef Bacik * Ok now we need to make sure to block out any other joins while we 2195ed0ca140SJosef Bacik * commit the transaction. We could have started a join before setting 21964a9d8bdeSMiao Xie * COMMIT_DOING so make sure to wait for num_writers to == 1 again. 2197ed0ca140SJosef Bacik */ 21980b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 21994a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 22000b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2201ed0ca140SJosef Bacik wait_event(cur_trans->writer_wait, 2202ed0ca140SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 220315ee9bc7SJosef Bacik 2204bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 22052cba30f1SMiao Xie ret = cur_trans->aborted; 22066cf7f77eSWang Shilong goto scrub_continue; 22072cba30f1SMiao Xie } 22087585717fSChris Mason /* 22097585717fSChris Mason * the reloc mutex makes sure that we stop 22107585717fSChris Mason * the balancing code from coming in and moving 22117585717fSChris Mason * extents around in the middle of the commit 22127585717fSChris Mason */ 22130b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 22147585717fSChris Mason 221542874b3dSMiao Xie /* 221642874b3dSMiao Xie * We needn't worry about the delayed items because we will 221742874b3dSMiao Xie * deal with them in create_pending_snapshot(), which is the 221842874b3dSMiao Xie * core function of the snapshot creation. 221942874b3dSMiao Xie */ 222008d50ca3SNikolay Borisov ret = create_pending_snapshots(trans); 222156e9f6eaSDavid Sterba if (ret) 222256e9f6eaSDavid Sterba goto unlock_reloc; 22233063d29fSChris Mason 222442874b3dSMiao Xie /* 222542874b3dSMiao Xie * We insert the dir indexes of the snapshots and update the inode 222642874b3dSMiao Xie * of the snapshots' parents after the snapshot creation, so there 222742874b3dSMiao Xie * are some delayed items which are not dealt with. Now deal with 222842874b3dSMiao Xie * them. 222942874b3dSMiao Xie * 223042874b3dSMiao Xie * We needn't worry that this operation will corrupt the snapshots, 223142874b3dSMiao Xie * because all the tree which are snapshoted will be forced to COW 223242874b3dSMiao Xie * the nodes and leaves. 223342874b3dSMiao Xie */ 2234e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 223556e9f6eaSDavid Sterba if (ret) 223656e9f6eaSDavid Sterba goto unlock_reloc; 223716cdcec7SMiao Xie 2238c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 223956e9f6eaSDavid Sterba if (ret) 224056e9f6eaSDavid Sterba goto unlock_reloc; 224156bec294SChris Mason 2242e999376fSChris Mason /* 2243e999376fSChris Mason * make sure none of the code above managed to slip in a 2244e999376fSChris Mason * delayed item 2245e999376fSChris Mason */ 2246ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 2247e999376fSChris Mason 22482c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 2249dc17ff8fSChris Mason 2250e02119d5SChris Mason /* btrfs_commit_tree_roots is responsible for getting the 2251e02119d5SChris Mason * various roots consistent with each other. Every pointer 2252e02119d5SChris Mason * in the tree of tree roots has to point to the most up to date 2253e02119d5SChris Mason * root for every subvolume and other tree. So, we have to keep 2254e02119d5SChris Mason * the tree logging code from jumping in and changing any 2255e02119d5SChris Mason * of the trees. 2256e02119d5SChris Mason * 2257e02119d5SChris Mason * At this point in the commit, there can't be any tree-log 2258e02119d5SChris Mason * writers, but a little lower down we drop the trans mutex 2259e02119d5SChris Mason * and let new people in. By holding the tree_log_mutex 2260e02119d5SChris Mason * from now until after the super is written, we avoid races 2261e02119d5SChris Mason * with the tree-log code. 2262e02119d5SChris Mason */ 22630b246afaSJeff Mahoney mutex_lock(&fs_info->tree_log_mutex); 22641a40e23bSZheng Yan 22657e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 226656e9f6eaSDavid Sterba if (ret) 226756e9f6eaSDavid Sterba goto unlock_tree_log; 226854aa1f4dSChris Mason 22693818aea2SQu Wenruo /* 22707e1876acSDavid Sterba * Since the transaction is done, we can apply the pending changes 22717e1876acSDavid Sterba * before the next transaction. 22723818aea2SQu Wenruo */ 22730b246afaSJeff Mahoney btrfs_apply_pending_changes(fs_info); 22743818aea2SQu Wenruo 22755d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 2276e02119d5SChris Mason * safe to free the root of tree log roots 2277e02119d5SChris Mason */ 22780b246afaSJeff Mahoney btrfs_free_log_root_tree(trans, fs_info); 2279e02119d5SChris Mason 22800ed4792aSQu Wenruo /* 22810ed4792aSQu Wenruo * Since fs roots are all committed, we can get a quite accurate 22820ed4792aSQu Wenruo * new_roots. So let's do quota accounting. 22830ed4792aSQu Wenruo */ 2284460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 228556e9f6eaSDavid Sterba if (ret < 0) 228656e9f6eaSDavid Sterba goto unlock_tree_log; 22870ed4792aSQu Wenruo 22889386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 228956e9f6eaSDavid Sterba if (ret) 229056e9f6eaSDavid Sterba goto unlock_tree_log; 229154aa1f4dSChris Mason 22922cba30f1SMiao Xie /* 22932cba30f1SMiao Xie * The tasks which save the space cache and inode cache may also 22942cba30f1SMiao Xie * update ->aborted, check it. 22952cba30f1SMiao Xie */ 2296bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 22972cba30f1SMiao Xie ret = cur_trans->aborted; 229856e9f6eaSDavid Sterba goto unlock_tree_log; 22992cba30f1SMiao Xie } 23002cba30f1SMiao Xie 23010b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 23025f39d397SChris Mason 23030b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->tree_root->root_item, 23040b246afaSJeff Mahoney fs_info->tree_root->node); 23050b246afaSJeff Mahoney list_add_tail(&fs_info->tree_root->dirty_list, 23069e351cc8SJosef Bacik &cur_trans->switch_commits); 23075d4f98a2SYan Zheng 23080b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->chunk_root->root_item, 23090b246afaSJeff Mahoney fs_info->chunk_root->node); 23100b246afaSJeff Mahoney list_add_tail(&fs_info->chunk_root->dirty_list, 23119e351cc8SJosef Bacik &cur_trans->switch_commits); 23129e351cc8SJosef Bacik 2313889bfa39SJosef Bacik switch_commit_roots(trans); 23145d4f98a2SYan Zheng 2315ce93ec54SJosef Bacik ASSERT(list_empty(&cur_trans->dirty_bgs)); 23161bbc621eSChris Mason ASSERT(list_empty(&cur_trans->io_bgs)); 23172ff7e61eSJeff Mahoney update_super_roots(fs_info); 2318e02119d5SChris Mason 23190b246afaSJeff Mahoney btrfs_set_super_log_root(fs_info->super_copy, 0); 23200b246afaSJeff Mahoney btrfs_set_super_log_root_level(fs_info->super_copy, 0); 23210b246afaSJeff Mahoney memcpy(fs_info->super_for_commit, fs_info->super_copy, 23220b246afaSJeff Mahoney sizeof(*fs_info->super_copy)); 2323ccd467d6SChris Mason 2324bbbf7243SNikolay Borisov btrfs_commit_device_sizes(cur_trans); 2325935e5cc9SMiao Xie 23260b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); 23270b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); 2328656f30dbSFilipe Manana 23294fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 23304fbcdf66SFilipe Manana 23310b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 23324a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 23330b246afaSJeff Mahoney fs_info->running_transaction = NULL; 23340b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 23350b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2336b7ec40d7SChris Mason 23370b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 2338e6dcd2dcSChris Mason 233970458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 234049b25e05SJeff Mahoney if (ret) { 23410b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 234208748810SDavid Sterba "Error while writing out transaction"); 234356e9f6eaSDavid Sterba /* 234456e9f6eaSDavid Sterba * reloc_mutex has been unlocked, tree_log_mutex is still held 234556e9f6eaSDavid Sterba * but we can't jump to unlock_tree_log causing double unlock 234656e9f6eaSDavid Sterba */ 23470b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 23486cf7f77eSWang Shilong goto scrub_continue; 234949b25e05SJeff Mahoney } 235049b25e05SJeff Mahoney 2351d3575156SNaohiro Aota /* 2352d3575156SNaohiro Aota * At this point, we should have written all the tree blocks allocated 2353d3575156SNaohiro Aota * in this transaction. So it's now safe to free the redirtyied extent 2354d3575156SNaohiro Aota * buffers. 2355d3575156SNaohiro Aota */ 2356d3575156SNaohiro Aota btrfs_free_redirty_list(cur_trans); 2357d3575156SNaohiro Aota 2358eece6a9cSDavid Sterba ret = write_all_supers(fs_info, 0); 2359e02119d5SChris Mason /* 2360e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 2361e02119d5SChris Mason * to go about their business 2362e02119d5SChris Mason */ 23630b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 2364c1f32b7cSAnand Jain if (ret) 2365c1f32b7cSAnand Jain goto scrub_continue; 2366e02119d5SChris Mason 2367d0c2f4faSFilipe Manana /* 2368d0c2f4faSFilipe Manana * We needn't acquire the lock here because there is no other task 2369d0c2f4faSFilipe Manana * which can change it. 2370d0c2f4faSFilipe Manana */ 2371d0c2f4faSFilipe Manana cur_trans->state = TRANS_STATE_SUPER_COMMITTED; 2372d0c2f4faSFilipe Manana wake_up(&cur_trans->commit_wait); 2373d0c2f4faSFilipe Manana 23745ead2dd0SNikolay Borisov btrfs_finish_extent_commit(trans); 23754313b399SChris Mason 23763204d33cSJosef Bacik if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) 23770b246afaSJeff Mahoney btrfs_clear_space_info_full(fs_info); 237813212b54SZhao Lei 23790b246afaSJeff Mahoney fs_info->last_trans_committed = cur_trans->transid; 23804a9d8bdeSMiao Xie /* 23814a9d8bdeSMiao Xie * We needn't acquire the lock here because there is no other task 23824a9d8bdeSMiao Xie * which can change it. 23834a9d8bdeSMiao Xie */ 23844a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMPLETED; 23852c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 23863de4586cSChris Mason 23870b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 238813c5a93eSJosef Bacik list_del_init(&cur_trans->list); 23890b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2390a4abeea4SJosef Bacik 2391724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2392724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 239358176a96SJosef Bacik 23940860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 23950b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2396b2b5ef5cSJan Kara 23973a45bb20SJeff Mahoney trace_btrfs_transaction_commit(trans->root); 23981abe9b8aSliubo 23992ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 2400a2de733cSArne Jansen 24019ed74f2dSJosef Bacik if (current->journal_info == trans) 24029ed74f2dSJosef Bacik current->journal_info = NULL; 24039ed74f2dSJosef Bacik 24042c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 240524bbcf04SYan, Zheng 240679154b1bSChris Mason return ret; 240749b25e05SJeff Mahoney 240856e9f6eaSDavid Sterba unlock_tree_log: 240956e9f6eaSDavid Sterba mutex_unlock(&fs_info->tree_log_mutex); 241056e9f6eaSDavid Sterba unlock_reloc: 241156e9f6eaSDavid Sterba mutex_unlock(&fs_info->reloc_mutex); 24126cf7f77eSWang Shilong scrub_continue: 24132ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 241449b25e05SJeff Mahoney cleanup_transaction: 2415dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 2416c7cc64a9SDavid Sterba btrfs_cleanup_pending_block_groups(trans); 24174fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 24180e721106SJosef Bacik trans->block_rsv = NULL; 24190b246afaSJeff Mahoney btrfs_warn(fs_info, "Skipping commit of aborted transaction."); 242049b25e05SJeff Mahoney if (current->journal_info == trans) 242149b25e05SJeff Mahoney current->journal_info = NULL; 242297cb39bbSNikolay Borisov cleanup_transaction(trans, ret); 242349b25e05SJeff Mahoney 242449b25e05SJeff Mahoney return ret; 242579154b1bSChris Mason } 242679154b1bSChris Mason 2427d352ac68SChris Mason /* 24289d1a2a3aSDavid Sterba * return < 0 if error 24299d1a2a3aSDavid Sterba * 0 if there are no more dead_roots at the time of call 24309d1a2a3aSDavid Sterba * 1 there are more to be processed, call me again 24319d1a2a3aSDavid Sterba * 24329d1a2a3aSDavid Sterba * The return value indicates there are certainly more snapshots to delete, but 24339d1a2a3aSDavid Sterba * if there comes a new one during processing, it may return 0. We don't mind, 24349d1a2a3aSDavid Sterba * because btrfs_commit_super will poke cleaner thread and it will process it a 24359d1a2a3aSDavid Sterba * few seconds later. 2436d352ac68SChris Mason */ 24379d1a2a3aSDavid Sterba int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) 2438e9d0b13bSChris Mason { 24399d1a2a3aSDavid Sterba int ret; 24405d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = root->fs_info; 2441e9d0b13bSChris Mason 2442a4abeea4SJosef Bacik spin_lock(&fs_info->trans_lock); 24439d1a2a3aSDavid Sterba if (list_empty(&fs_info->dead_roots)) { 24449d1a2a3aSDavid Sterba spin_unlock(&fs_info->trans_lock); 24459d1a2a3aSDavid Sterba return 0; 24469d1a2a3aSDavid Sterba } 24479d1a2a3aSDavid Sterba root = list_first_entry(&fs_info->dead_roots, 24489d1a2a3aSDavid Sterba struct btrfs_root, root_list); 2449cfad392bSJosef Bacik list_del_init(&root->root_list); 2450a4abeea4SJosef Bacik spin_unlock(&fs_info->trans_lock); 24515d4f98a2SYan Zheng 24524fd786e6SMisono Tomohiro btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); 245376dda93cSYan, Zheng 245416cdcec7SMiao Xie btrfs_kill_all_delayed_nodes(root); 245516cdcec7SMiao Xie 245676dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 245776dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 24580078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 0, 0); 245976dda93cSYan, Zheng else 24600078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 1, 0); 246132471dc2SDavid Sterba 2462dc9492c1SJosef Bacik btrfs_put_root(root); 24636596a928SJosef Bacik return (ret < 0) ? 0 : 1; 2464e9d0b13bSChris Mason } 2465572d9ab7SDavid Sterba 2466572d9ab7SDavid Sterba void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) 2467572d9ab7SDavid Sterba { 2468572d9ab7SDavid Sterba unsigned long prev; 2469572d9ab7SDavid Sterba unsigned long bit; 2470572d9ab7SDavid Sterba 24716c9fe14fSQu Wenruo prev = xchg(&fs_info->pending_changes, 0); 2472572d9ab7SDavid Sterba if (!prev) 2473572d9ab7SDavid Sterba return; 2474572d9ab7SDavid Sterba 2475d51033d0SDavid Sterba bit = 1 << BTRFS_PENDING_COMMIT; 2476d51033d0SDavid Sterba if (prev & bit) 2477d51033d0SDavid Sterba btrfs_debug(fs_info, "pending commit done"); 2478d51033d0SDavid Sterba prev &= ~bit; 2479d51033d0SDavid Sterba 2480572d9ab7SDavid Sterba if (prev) 2481572d9ab7SDavid Sterba btrfs_warn(fs_info, 2482572d9ab7SDavid Sterba "unknown pending changes left 0x%lx, ignoring", prev); 2483572d9ab7SDavid Sterba } 2484