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" 24*d3575156SNaohiro 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 /* 257fb6dea26SJosef Bacik * To be called after all the new block groups attached to the transaction 258fb6dea26SJosef Bacik * handle have been created (btrfs_create_pending_block_groups()). 259fb6dea26SJosef Bacik */ 260fb6dea26SJosef Bacik void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) 261fb6dea26SJosef Bacik { 262fb6dea26SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 263fb6dea26SJosef Bacik 264fb6dea26SJosef Bacik if (!trans->chunk_bytes_reserved) 265fb6dea26SJosef Bacik return; 266fb6dea26SJosef Bacik 267fb6dea26SJosef Bacik WARN_ON_ONCE(!list_empty(&trans->new_bgs)); 268fb6dea26SJosef Bacik 269fb6dea26SJosef Bacik btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, 27063f018beSNikolay Borisov trans->chunk_bytes_reserved, NULL); 271fb6dea26SJosef Bacik trans->chunk_bytes_reserved = 0; 272fb6dea26SJosef Bacik } 273fb6dea26SJosef Bacik 274fb6dea26SJosef Bacik /* 275d352ac68SChris Mason * either allocate a new transaction or hop into the existing one 276d352ac68SChris Mason */ 2772ff7e61eSJeff Mahoney static noinline int join_transaction(struct btrfs_fs_info *fs_info, 2782ff7e61eSJeff Mahoney unsigned int type) 27979154b1bSChris Mason { 28079154b1bSChris Mason struct btrfs_transaction *cur_trans; 281a4abeea4SJosef Bacik 28219ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 283d43317dcSChris Mason loop: 28449b25e05SJeff Mahoney /* The file system has been taken offline. No new transactions. */ 28587533c47SMiao Xie if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 28619ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 28749b25e05SJeff Mahoney return -EROFS; 28849b25e05SJeff Mahoney } 28949b25e05SJeff Mahoney 29019ae4e81SJan Schmidt cur_trans = fs_info->running_transaction; 291a4abeea4SJosef Bacik if (cur_trans) { 292bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 29319ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 29449b25e05SJeff Mahoney return cur_trans->aborted; 295871383beSDavid Sterba } 2964a9d8bdeSMiao Xie if (btrfs_blocked_trans_types[cur_trans->state] & type) { 297178260b2SMiao Xie spin_unlock(&fs_info->trans_lock); 298178260b2SMiao Xie return -EBUSY; 299178260b2SMiao Xie } 3009b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 301a4abeea4SJosef Bacik atomic_inc(&cur_trans->num_writers); 3020860adfdSMiao Xie extwriter_counter_inc(cur_trans, type); 30319ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 304a4abeea4SJosef Bacik return 0; 305a4abeea4SJosef Bacik } 30619ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 307a4abeea4SJosef Bacik 308354aa0fbSMiao Xie /* 309354aa0fbSMiao Xie * If we are ATTACH, we just want to catch the current transaction, 310354aa0fbSMiao Xie * and commit it. If there is no transaction, just return ENOENT. 311354aa0fbSMiao Xie */ 312354aa0fbSMiao Xie if (type == TRANS_ATTACH) 313354aa0fbSMiao Xie return -ENOENT; 314354aa0fbSMiao Xie 3154a9d8bdeSMiao Xie /* 3164a9d8bdeSMiao Xie * JOIN_NOLOCK only happens during the transaction commit, so 3174a9d8bdeSMiao Xie * it is impossible that ->running_transaction is NULL 3184a9d8bdeSMiao Xie */ 3194a9d8bdeSMiao Xie BUG_ON(type == TRANS_JOIN_NOLOCK); 3204a9d8bdeSMiao Xie 3214b5faeacSDavid Sterba cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); 322db5b493aSTsutomu Itoh if (!cur_trans) 323db5b493aSTsutomu Itoh return -ENOMEM; 324d43317dcSChris Mason 32519ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 32619ae4e81SJan Schmidt if (fs_info->running_transaction) { 327d43317dcSChris Mason /* 328d43317dcSChris Mason * someone started a transaction after we unlocked. Make sure 3294a9d8bdeSMiao Xie * to redo the checks above 330d43317dcSChris Mason */ 3314b5faeacSDavid Sterba kfree(cur_trans); 332d43317dcSChris Mason goto loop; 33387533c47SMiao Xie } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 334e4b50e14SDan Carpenter spin_unlock(&fs_info->trans_lock); 3354b5faeacSDavid Sterba kfree(cur_trans); 3367b8b92afSJosef Bacik return -EROFS; 337a4abeea4SJosef Bacik } 338d43317dcSChris Mason 339ab8d0fc4SJeff Mahoney cur_trans->fs_info = fs_info; 34048778179SFilipe Manana atomic_set(&cur_trans->pending_ordered, 0); 34148778179SFilipe Manana init_waitqueue_head(&cur_trans->pending_wait); 34213c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 3430860adfdSMiao Xie extwriter_counter_init(cur_trans, type); 34479154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 34579154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 3464a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_RUNNING; 347a4abeea4SJosef Bacik /* 348a4abeea4SJosef Bacik * One for this trans handle, one so it will live on until we 349a4abeea4SJosef Bacik * commit the transaction. 350a4abeea4SJosef Bacik */ 3519b64f57dSElena Reshetova refcount_set(&cur_trans->use_count, 2); 3523204d33cSJosef Bacik cur_trans->flags = 0; 353afd48513SArnd Bergmann cur_trans->start_time = ktime_get_seconds(); 35456bec294SChris Mason 355a099d0fdSAlexandru Moise memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); 356a099d0fdSAlexandru Moise 3575c9d028bSLiu Bo cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; 3583368d001SQu Wenruo cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; 359d7df2c79SJosef Bacik atomic_set(&cur_trans->delayed_refs.num_entries, 0); 36020b297d6SJan Schmidt 36120b297d6SJan Schmidt /* 36220b297d6SJan Schmidt * although the tree mod log is per file system and not per transaction, 36320b297d6SJan Schmidt * the log must never go across transaction boundaries. 36420b297d6SJan Schmidt */ 36520b297d6SJan Schmidt smp_mb(); 36631b1a2bdSJulia Lawall if (!list_empty(&fs_info->tree_mod_seq_list)) 3675d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); 36831b1a2bdSJulia Lawall if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) 3695d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); 370fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 37120b297d6SJan Schmidt 37256bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 37356bec294SChris Mason 3743063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 375bbbf7243SNikolay Borisov INIT_LIST_HEAD(&cur_trans->dev_update_list); 3769e351cc8SJosef Bacik INIT_LIST_HEAD(&cur_trans->switch_commits); 377ce93ec54SJosef Bacik INIT_LIST_HEAD(&cur_trans->dirty_bgs); 3781bbc621eSChris Mason INIT_LIST_HEAD(&cur_trans->io_bgs); 3792b9dbef2SJosef Bacik INIT_LIST_HEAD(&cur_trans->dropped_roots); 3801bbc621eSChris Mason mutex_init(&cur_trans->cache_write_mutex); 381ce93ec54SJosef Bacik spin_lock_init(&cur_trans->dirty_bgs_lock); 382e33e17eeSJeff Mahoney INIT_LIST_HEAD(&cur_trans->deleted_bgs); 3832b9dbef2SJosef Bacik spin_lock_init(&cur_trans->dropped_roots_lock); 384*d3575156SNaohiro Aota INIT_LIST_HEAD(&cur_trans->releasing_ebs); 385*d3575156SNaohiro Aota spin_lock_init(&cur_trans->releasing_ebs_lock); 38619ae4e81SJan Schmidt list_add_tail(&cur_trans->list, &fs_info->trans_list); 387c258d6e3SQu Wenruo extent_io_tree_init(fs_info, &cur_trans->dirty_pages, 38843eb5f29SQu Wenruo IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode); 389fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &cur_trans->pinned_extents, 390fe119a6eSNikolay Borisov IO_TREE_FS_PINNED_EXTENTS, NULL); 39119ae4e81SJan Schmidt fs_info->generation++; 39219ae4e81SJan Schmidt cur_trans->transid = fs_info->generation; 39319ae4e81SJan Schmidt fs_info->running_transaction = cur_trans; 39449b25e05SJeff Mahoney cur_trans->aborted = 0; 39519ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 39615ee9bc7SJosef Bacik 39779154b1bSChris Mason return 0; 39879154b1bSChris Mason } 39979154b1bSChris Mason 400d352ac68SChris Mason /* 40192a7cc42SQu Wenruo * This does all the record keeping required to make sure that a shareable root 40292a7cc42SQu Wenruo * is properly recorded in a given transaction. This is required to make sure 40392a7cc42SQu Wenruo * the old root from before we joined the transaction is deleted when the 40492a7cc42SQu Wenruo * transaction commits. 405d352ac68SChris Mason */ 4067585717fSChris Mason static int record_root_in_trans(struct btrfs_trans_handle *trans, 4076426c7adSQu Wenruo struct btrfs_root *root, 4086426c7adSQu Wenruo int force) 4096702ed49SChris Mason { 4100b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4110b246afaSJeff Mahoney 41292a7cc42SQu Wenruo if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 4136426c7adSQu Wenruo root->last_trans < trans->transid) || force) { 4140b246afaSJeff Mahoney WARN_ON(root == fs_info->extent_root); 4154d31778aSQu Wenruo WARN_ON(!force && root->commit_root != root->node); 4165d4f98a2SYan Zheng 4177585717fSChris Mason /* 41827cdeb70SMiao Xie * see below for IN_TRANS_SETUP usage rules 4197585717fSChris Mason * we have the reloc mutex held now, so there 4207585717fSChris Mason * is only one writer in this function 4217585717fSChris Mason */ 42227cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4237585717fSChris Mason 42427cdeb70SMiao Xie /* make sure readers find IN_TRANS_SETUP before 4257585717fSChris Mason * they find our root->last_trans update 4267585717fSChris Mason */ 4277585717fSChris Mason smp_wmb(); 4287585717fSChris Mason 4290b246afaSJeff Mahoney spin_lock(&fs_info->fs_roots_radix_lock); 4306426c7adSQu Wenruo if (root->last_trans == trans->transid && !force) { 4310b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 432a4abeea4SJosef Bacik return 0; 433a4abeea4SJosef Bacik } 4340b246afaSJeff Mahoney radix_tree_tag_set(&fs_info->fs_roots_radix, 4356702ed49SChris Mason (unsigned long)root->root_key.objectid, 4366702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 4370b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 4387585717fSChris Mason root->last_trans = trans->transid; 4397585717fSChris Mason 4407585717fSChris Mason /* this is pretty tricky. We don't want to 4417585717fSChris Mason * take the relocation lock in btrfs_record_root_in_trans 4427585717fSChris Mason * unless we're really doing the first setup for this root in 4437585717fSChris Mason * this transaction. 4447585717fSChris Mason * 4457585717fSChris Mason * Normally we'd use root->last_trans as a flag to decide 4467585717fSChris Mason * if we want to take the expensive mutex. 4477585717fSChris Mason * 4487585717fSChris Mason * But, we have to set root->last_trans before we 4497585717fSChris Mason * init the relocation root, otherwise, we trip over warnings 4507585717fSChris Mason * in ctree.c. The solution used here is to flag ourselves 45127cdeb70SMiao Xie * with root IN_TRANS_SETUP. When this is 1, we're still 4527585717fSChris Mason * fixing up the reloc trees and everyone must wait. 4537585717fSChris Mason * 4547585717fSChris Mason * When this is zero, they can trust root->last_trans and fly 4557585717fSChris Mason * through btrfs_record_root_in_trans without having to take the 4567585717fSChris Mason * lock. smp_wmb() makes sure that all the writes above are 4577585717fSChris Mason * done before we pop in the zero below 4587585717fSChris Mason */ 4595d4f98a2SYan Zheng btrfs_init_reloc_root(trans, root); 460c7548af6SChris Mason smp_mb__before_atomic(); 46127cdeb70SMiao Xie clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4626702ed49SChris Mason } 4635d4f98a2SYan Zheng return 0; 4646702ed49SChris Mason } 4655d4f98a2SYan Zheng 4667585717fSChris Mason 4672b9dbef2SJosef Bacik void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, 4682b9dbef2SJosef Bacik struct btrfs_root *root) 4692b9dbef2SJosef Bacik { 4700b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4712b9dbef2SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 4722b9dbef2SJosef Bacik 4732b9dbef2SJosef Bacik /* Add ourselves to the transaction dropped list */ 4742b9dbef2SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 4752b9dbef2SJosef Bacik list_add_tail(&root->root_list, &cur_trans->dropped_roots); 4762b9dbef2SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 4772b9dbef2SJosef Bacik 4782b9dbef2SJosef Bacik /* Make sure we don't try to update the root at commit time */ 4790b246afaSJeff Mahoney spin_lock(&fs_info->fs_roots_radix_lock); 4800b246afaSJeff Mahoney radix_tree_tag_clear(&fs_info->fs_roots_radix, 4812b9dbef2SJosef Bacik (unsigned long)root->root_key.objectid, 4822b9dbef2SJosef Bacik BTRFS_ROOT_TRANS_TAG); 4830b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 4842b9dbef2SJosef Bacik } 4852b9dbef2SJosef Bacik 4867585717fSChris Mason int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 4877585717fSChris Mason struct btrfs_root *root) 4887585717fSChris Mason { 4890b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4900b246afaSJeff Mahoney 49192a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 4927585717fSChris Mason return 0; 4937585717fSChris Mason 4947585717fSChris Mason /* 49527cdeb70SMiao Xie * see record_root_in_trans for comments about IN_TRANS_SETUP usage 4967585717fSChris Mason * and barriers 4977585717fSChris Mason */ 4987585717fSChris Mason smp_rmb(); 4997585717fSChris Mason if (root->last_trans == trans->transid && 50027cdeb70SMiao Xie !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) 5017585717fSChris Mason return 0; 5027585717fSChris Mason 5030b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 5046426c7adSQu Wenruo record_root_in_trans(trans, root, 0); 5050b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 5067585717fSChris Mason 5077585717fSChris Mason return 0; 5087585717fSChris Mason } 5097585717fSChris Mason 5104a9d8bdeSMiao Xie static inline int is_transaction_blocked(struct btrfs_transaction *trans) 5114a9d8bdeSMiao Xie { 5123296bf56SQu Wenruo return (trans->state >= TRANS_STATE_COMMIT_START && 513501407aaSJosef Bacik trans->state < TRANS_STATE_UNBLOCKED && 514bf31f87fSDavid Sterba !TRANS_ABORTED(trans)); 5154a9d8bdeSMiao Xie } 5164a9d8bdeSMiao Xie 517d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 518d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 519d352ac68SChris Mason * transaction might not be fully on disk. 520d352ac68SChris Mason */ 5212ff7e61eSJeff Mahoney static void wait_current_trans(struct btrfs_fs_info *fs_info) 52279154b1bSChris Mason { 523f9295749SChris Mason struct btrfs_transaction *cur_trans; 52479154b1bSChris Mason 5250b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5260b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 5274a9d8bdeSMiao Xie if (cur_trans && is_transaction_blocked(cur_trans)) { 5289b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 5290b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 53072d63ed6SLi Zefan 5310b246afaSJeff Mahoney wait_event(fs_info->transaction_wait, 532501407aaSJosef Bacik cur_trans->state >= TRANS_STATE_UNBLOCKED || 533bf31f87fSDavid Sterba TRANS_ABORTED(cur_trans)); 534724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 535a4abeea4SJosef Bacik } else { 5360b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 537f9295749SChris Mason } 53837d1aeeeSChris Mason } 53937d1aeeeSChris Mason 5402ff7e61eSJeff Mahoney static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) 54137d1aeeeSChris Mason { 5420b246afaSJeff Mahoney if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) 543a4abeea4SJosef Bacik return 0; 544a4abeea4SJosef Bacik 54592e2f7e3SNikolay Borisov if (type == TRANS_START) 546a4abeea4SJosef Bacik return 1; 547a4abeea4SJosef Bacik 548a22285a6SYan, Zheng return 0; 549a22285a6SYan, Zheng } 550a22285a6SYan, Zheng 55120dd2cbfSMiao Xie static inline bool need_reserve_reloc_root(struct btrfs_root *root) 55220dd2cbfSMiao Xie { 5530b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5540b246afaSJeff Mahoney 5550b246afaSJeff Mahoney if (!fs_info->reloc_ctl || 55692a7cc42SQu Wenruo !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || 55720dd2cbfSMiao Xie root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 55820dd2cbfSMiao Xie root->reloc_root) 55920dd2cbfSMiao Xie return false; 56020dd2cbfSMiao Xie 56120dd2cbfSMiao Xie return true; 56220dd2cbfSMiao Xie } 56320dd2cbfSMiao Xie 56408e007d2SMiao Xie static struct btrfs_trans_handle * 5655aed1dd8SAlexandru Moise start_transaction(struct btrfs_root *root, unsigned int num_items, 566003d7c59SJeff Mahoney unsigned int type, enum btrfs_reserve_flush_enum flush, 567003d7c59SJeff Mahoney bool enforce_qgroups) 568a22285a6SYan, Zheng { 5690b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 570ba2c4d4eSJosef Bacik struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 571a22285a6SYan, Zheng struct btrfs_trans_handle *h; 572a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 573b5009945SJosef Bacik u64 num_bytes = 0; 574c5567237SArne Jansen u64 qgroup_reserved = 0; 57520dd2cbfSMiao Xie bool reloc_reserved = false; 5769c343784SJosef Bacik bool do_chunk_alloc = false; 57720dd2cbfSMiao Xie int ret; 578acce952bSliubo 57946c4e71eSFilipe Manana /* Send isn't supposed to start transactions. */ 5802755a0deSDavid Sterba ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); 58146c4e71eSFilipe Manana 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; 700d9a0540aSFilipe Manana h->can_flush_pending_bgs = true; 701ea658badSJosef Bacik INIT_LIST_HEAD(&h->new_bgs); 702b7ec40d7SChris Mason 703a22285a6SYan, Zheng smp_mb(); 7043296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START && 7052ff7e61eSJeff Mahoney may_wait_transaction(fs_info, type)) { 706abdd2e80SFilipe Manana current->journal_info = h; 7073a45bb20SJeff Mahoney btrfs_commit_transaction(h); 708a22285a6SYan, Zheng goto again; 709a22285a6SYan, Zheng } 7109ed74f2dSJosef Bacik 711b5009945SJosef Bacik if (num_bytes) { 7120b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 7132bcc0328SLiu Bo h->transid, num_bytes, 1); 7140b246afaSJeff Mahoney h->block_rsv = &fs_info->trans_block_rsv; 715b5009945SJosef Bacik h->bytes_reserved = num_bytes; 71620dd2cbfSMiao Xie h->reloc_reserved = reloc_reserved; 717a22285a6SYan, Zheng } 718a22285a6SYan, Zheng 7192a1eb461SJosef Bacik got_it: 720bcf3a3e7SNikolay Borisov if (!current->journal_info) 721a22285a6SYan, Zheng current->journal_info = h; 722fcc99734SQu Wenruo 723fcc99734SQu Wenruo /* 7249c343784SJosef Bacik * If the space_info is marked ALLOC_FORCE then we'll get upgraded to 7259c343784SJosef Bacik * ALLOC_FORCE the first run through, and then we won't allocate for 7269c343784SJosef Bacik * anybody else who races in later. We don't care about the return 7279c343784SJosef Bacik * value here. 7289c343784SJosef Bacik */ 7299c343784SJosef Bacik if (do_chunk_alloc && num_bytes) { 7309c343784SJosef Bacik u64 flags = h->block_rsv->space_info->flags; 7319c343784SJosef Bacik 7329c343784SJosef Bacik btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), 7339c343784SJosef Bacik CHUNK_ALLOC_NO_FORCE); 7349c343784SJosef Bacik } 7359c343784SJosef Bacik 7369c343784SJosef Bacik /* 737fcc99734SQu Wenruo * btrfs_record_root_in_trans() needs to alloc new extents, and may 738fcc99734SQu Wenruo * call btrfs_join_transaction() while we're also starting a 739fcc99734SQu Wenruo * transaction. 740fcc99734SQu Wenruo * 741fcc99734SQu Wenruo * Thus it need to be called after current->journal_info initialized, 742fcc99734SQu Wenruo * or we can deadlock. 743fcc99734SQu Wenruo */ 744fcc99734SQu Wenruo btrfs_record_root_in_trans(h, root); 745fcc99734SQu Wenruo 74679154b1bSChris Mason return h; 747843fcf35SMiao Xie 748843fcf35SMiao Xie join_fail: 7490860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 7500b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 751843fcf35SMiao Xie kmem_cache_free(btrfs_trans_handle_cachep, h); 752843fcf35SMiao Xie alloc_fail: 753843fcf35SMiao Xie if (num_bytes) 7542ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, 75563f018beSNikolay Borisov num_bytes, NULL); 756843fcf35SMiao Xie reserve_fail: 757733e03a0SQu Wenruo btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); 758843fcf35SMiao Xie return ERR_PTR(ret); 75979154b1bSChris Mason } 76079154b1bSChris Mason 761f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 7625aed1dd8SAlexandru Moise unsigned int num_items) 763f9295749SChris Mason { 76408e007d2SMiao Xie return start_transaction(root, num_items, TRANS_START, 765003d7c59SJeff Mahoney BTRFS_RESERVE_FLUSH_ALL, true); 766f9295749SChris Mason } 767003d7c59SJeff Mahoney 7688eab77ffSFilipe Manana struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( 7698eab77ffSFilipe Manana struct btrfs_root *root, 7707f9fe614SJosef Bacik unsigned int num_items) 7718eab77ffSFilipe Manana { 7727f9fe614SJosef Bacik return start_transaction(root, num_items, TRANS_START, 7737f9fe614SJosef Bacik BTRFS_RESERVE_FLUSH_ALL_STEAL, false); 7748eab77ffSFilipe Manana } 7758407aa46SMiao Xie 7767a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) 777f9295749SChris Mason { 778003d7c59SJeff Mahoney return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, 779003d7c59SJeff Mahoney true); 780f9295749SChris Mason } 781f9295749SChris Mason 7828d510121SNikolay Borisov struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) 7830af3d00bSJosef Bacik { 784575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 785003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 7860af3d00bSJosef Bacik } 7870af3d00bSJosef Bacik 788d4edf39bSMiao Xie /* 789a6d155d2SFilipe Manana * Similar to regular join but it never starts a transaction when none is 790a6d155d2SFilipe Manana * running or after waiting for the current one to finish. 791a6d155d2SFilipe Manana */ 792a6d155d2SFilipe Manana struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) 793a6d155d2SFilipe Manana { 794a6d155d2SFilipe Manana return start_transaction(root, 0, TRANS_JOIN_NOSTART, 795a6d155d2SFilipe Manana BTRFS_RESERVE_NO_FLUSH, true); 796a6d155d2SFilipe Manana } 797a6d155d2SFilipe Manana 798a6d155d2SFilipe Manana /* 799d4edf39bSMiao Xie * btrfs_attach_transaction() - catch the running transaction 800d4edf39bSMiao Xie * 801d4edf39bSMiao Xie * It is used when we want to commit the current the transaction, but 802d4edf39bSMiao Xie * don't want to start a new one. 803d4edf39bSMiao Xie * 804d4edf39bSMiao Xie * Note: If this function return -ENOENT, it just means there is no 805d4edf39bSMiao Xie * running transaction. But it is possible that the inactive transaction 806d4edf39bSMiao Xie * is still in the memory, not fully on disk. If you hope there is no 807d4edf39bSMiao Xie * inactive transaction in the fs when -ENOENT is returned, you should 808d4edf39bSMiao Xie * invoke 809d4edf39bSMiao Xie * btrfs_attach_transaction_barrier() 810d4edf39bSMiao Xie */ 811354aa0fbSMiao Xie struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) 81260376ce4SJosef Bacik { 813575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_ATTACH, 814003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 81560376ce4SJosef Bacik } 81660376ce4SJosef Bacik 817d4edf39bSMiao Xie /* 81890b6d283SWang Sheng-Hui * btrfs_attach_transaction_barrier() - catch the running transaction 819d4edf39bSMiao Xie * 82052042d8eSAndrea Gelmini * It is similar to the above function, the difference is this one 821d4edf39bSMiao Xie * will wait for all the inactive transactions until they fully 822d4edf39bSMiao Xie * complete. 823d4edf39bSMiao Xie */ 824d4edf39bSMiao Xie struct btrfs_trans_handle * 825d4edf39bSMiao Xie btrfs_attach_transaction_barrier(struct btrfs_root *root) 826d4edf39bSMiao Xie { 827d4edf39bSMiao Xie struct btrfs_trans_handle *trans; 828d4edf39bSMiao Xie 829575a75d6SAlexandru Moise trans = start_transaction(root, 0, TRANS_ATTACH, 830003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 8318d9e220cSAl Viro if (trans == ERR_PTR(-ENOENT)) 8322ff7e61eSJeff Mahoney btrfs_wait_for_commit(root->fs_info, 0); 833d4edf39bSMiao Xie 834d4edf39bSMiao Xie return trans; 835d4edf39bSMiao Xie } 836d4edf39bSMiao Xie 837d0c2f4faSFilipe Manana /* Wait for a transaction commit to reach at least the given state. */ 838d0c2f4faSFilipe Manana static noinline void wait_for_commit(struct btrfs_transaction *commit, 839d0c2f4faSFilipe Manana const enum btrfs_trans_state min_state) 84089ce8a63SChris Mason { 841d0c2f4faSFilipe Manana wait_event(commit->commit_wait, commit->state >= min_state); 84289ce8a63SChris Mason } 84389ce8a63SChris Mason 8442ff7e61eSJeff Mahoney int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) 84546204592SSage Weil { 84646204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 8478cd2807fSMiao Xie int ret = 0; 84846204592SSage Weil 84946204592SSage Weil if (transid) { 8500b246afaSJeff Mahoney if (transid <= fs_info->last_trans_committed) 851a4abeea4SJosef Bacik goto out; 85246204592SSage Weil 85346204592SSage Weil /* find specified transaction */ 8540b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8550b246afaSJeff Mahoney list_for_each_entry(t, &fs_info->trans_list, list) { 85646204592SSage Weil if (t->transid == transid) { 85746204592SSage Weil cur_trans = t; 8589b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 8598cd2807fSMiao Xie ret = 0; 86046204592SSage Weil break; 86146204592SSage Weil } 8628cd2807fSMiao Xie if (t->transid > transid) { 8638cd2807fSMiao Xie ret = 0; 86446204592SSage Weil break; 86546204592SSage Weil } 8668cd2807fSMiao Xie } 8670b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 86842383020SSage Weil 86942383020SSage Weil /* 87042383020SSage Weil * The specified transaction doesn't exist, or we 87142383020SSage Weil * raced with btrfs_commit_transaction 87242383020SSage Weil */ 87342383020SSage Weil if (!cur_trans) { 8740b246afaSJeff Mahoney if (transid > fs_info->last_trans_committed) 87542383020SSage Weil ret = -EINVAL; 8768cd2807fSMiao Xie goto out; 87742383020SSage Weil } 87846204592SSage Weil } else { 87946204592SSage Weil /* find newest transaction that is committing | committed */ 8800b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8810b246afaSJeff Mahoney list_for_each_entry_reverse(t, &fs_info->trans_list, 88246204592SSage Weil list) { 8834a9d8bdeSMiao Xie if (t->state >= TRANS_STATE_COMMIT_START) { 8844a9d8bdeSMiao Xie if (t->state == TRANS_STATE_COMPLETED) 8853473f3c0SJosef Bacik break; 88646204592SSage Weil cur_trans = t; 8879b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 88846204592SSage Weil break; 88946204592SSage Weil } 89046204592SSage Weil } 8910b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 89246204592SSage Weil if (!cur_trans) 893a4abeea4SJosef Bacik goto out; /* nothing committing|committed */ 89446204592SSage Weil } 89546204592SSage Weil 896d0c2f4faSFilipe Manana wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); 897724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 898a4abeea4SJosef Bacik out: 89946204592SSage Weil return ret; 90046204592SSage Weil } 90146204592SSage Weil 9022ff7e61eSJeff Mahoney void btrfs_throttle(struct btrfs_fs_info *fs_info) 90337d1aeeeSChris Mason { 9042ff7e61eSJeff Mahoney wait_current_trans(fs_info); 90537d1aeeeSChris Mason } 90637d1aeeeSChris Mason 9078a8f4deaSNikolay Borisov static bool should_end_transaction(struct btrfs_trans_handle *trans) 9088929ecfaSYan, Zheng { 9092ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 9100b246afaSJeff Mahoney 91164403612SJosef Bacik if (btrfs_check_space_for_delayed_refs(fs_info)) 9128a8f4deaSNikolay Borisov return true; 91336ba022aSJosef Bacik 9142ff7e61eSJeff Mahoney return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); 9158929ecfaSYan, Zheng } 9168929ecfaSYan, Zheng 917a2633b6aSNikolay Borisov bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) 9188929ecfaSYan, Zheng { 9198929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9208929ecfaSYan, Zheng 9213296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START || 922e19eb11fSJosef Bacik test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) 923a2633b6aSNikolay Borisov return true; 9248929ecfaSYan, Zheng 9252ff7e61eSJeff Mahoney return should_end_transaction(trans); 9268929ecfaSYan, Zheng } 9278929ecfaSYan, Zheng 928dc60c525SNikolay Borisov static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) 929dc60c525SNikolay Borisov 9300e34693fSNikolay Borisov { 931dc60c525SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 932dc60c525SNikolay Borisov 9330e34693fSNikolay Borisov if (!trans->block_rsv) { 9340e34693fSNikolay Borisov ASSERT(!trans->bytes_reserved); 9350e34693fSNikolay Borisov return; 9360e34693fSNikolay Borisov } 9370e34693fSNikolay Borisov 9380e34693fSNikolay Borisov if (!trans->bytes_reserved) 9390e34693fSNikolay Borisov return; 9400e34693fSNikolay Borisov 9410e34693fSNikolay Borisov ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); 9420e34693fSNikolay Borisov trace_btrfs_space_reservation(fs_info, "transaction", 9430e34693fSNikolay Borisov trans->transid, trans->bytes_reserved, 0); 9440e34693fSNikolay Borisov btrfs_block_rsv_release(fs_info, trans->block_rsv, 94563f018beSNikolay Borisov trans->bytes_reserved, NULL); 9460e34693fSNikolay Borisov trans->bytes_reserved = 0; 9470e34693fSNikolay Borisov } 9480e34693fSNikolay Borisov 94989ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 9503a45bb20SJeff Mahoney int throttle) 95179154b1bSChris Mason { 9523a45bb20SJeff Mahoney struct btrfs_fs_info *info = trans->fs_info; 9538929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9544edc2ca3SDave Jones int err = 0; 955d6e4a428SChris Mason 956b50fff81SDavid Sterba if (refcount_read(&trans->use_count) > 1) { 957b50fff81SDavid Sterba refcount_dec(&trans->use_count); 9582a1eb461SJosef Bacik trans->block_rsv = trans->orig_rsv; 9592a1eb461SJosef Bacik return 0; 9602a1eb461SJosef Bacik } 9612a1eb461SJosef Bacik 962dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 9634c13d758SJosef Bacik trans->block_rsv = NULL; 964c5567237SArne Jansen 9656c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 966ea658badSJosef Bacik 9674fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 9684fbcdf66SFilipe Manana 9690860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 9700b246afaSJeff Mahoney sb_end_intwrite(info->sb); 9716df7881aSJosef Bacik 9728929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 97313c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 97413c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 9750860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 97689ce8a63SChris Mason 977093258e6SDavid Sterba cond_wake_up(&cur_trans->writer_wait); 978724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 9799ed74f2dSJosef Bacik 9809ed74f2dSJosef Bacik if (current->journal_info == trans) 9819ed74f2dSJosef Bacik current->journal_info = NULL; 982ab78c84dSChris Mason 98324bbcf04SYan, Zheng if (throttle) 9842ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(info); 98524bbcf04SYan, Zheng 986bf31f87fSDavid Sterba if (TRANS_ABORTED(trans) || 9870b246afaSJeff Mahoney test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) { 9884e121c06SJosef Bacik wake_up_process(info->transaction_kthread); 989fbabd4a3SJosef Bacik if (TRANS_ABORTED(trans)) 990fbabd4a3SJosef Bacik err = trans->aborted; 991fbabd4a3SJosef Bacik else 992fbabd4a3SJosef Bacik err = -EROFS; 9934e121c06SJosef Bacik } 99449b25e05SJeff Mahoney 9954edc2ca3SDave Jones kmem_cache_free(btrfs_trans_handle_cachep, trans); 9964edc2ca3SDave Jones return err; 99779154b1bSChris Mason } 99879154b1bSChris Mason 9993a45bb20SJeff Mahoney int btrfs_end_transaction(struct btrfs_trans_handle *trans) 100089ce8a63SChris Mason { 10013a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 0); 100289ce8a63SChris Mason } 100389ce8a63SChris Mason 10043a45bb20SJeff Mahoney int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) 100589ce8a63SChris Mason { 10063a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 1); 100716cdcec7SMiao Xie } 100816cdcec7SMiao Xie 1009d352ac68SChris Mason /* 1010d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1011d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1012690587d1SChris Mason * those extents are sent to disk but does not wait on them 1013d352ac68SChris Mason */ 10142ff7e61eSJeff Mahoney int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, 10158cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 101679154b1bSChris Mason { 1017777e6bd7SChris Mason int err = 0; 10187c4452b9SChris Mason int werr = 0; 10190b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1020e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1021777e6bd7SChris Mason u64 start = 0; 10225f39d397SChris Mason u64 end; 10237c4452b9SChris Mason 10246300463bSLiu Bo atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); 10251728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1026e6138876SJosef Bacik mark, &cached_state)) { 1027663dfbb0SFilipe Manana bool wait_writeback = false; 1028663dfbb0SFilipe Manana 1029663dfbb0SFilipe Manana err = convert_extent_bit(dirty_pages, start, end, 1030663dfbb0SFilipe Manana EXTENT_NEED_WAIT, 1031210aa277SDavid Sterba mark, &cached_state); 1032663dfbb0SFilipe Manana /* 1033663dfbb0SFilipe Manana * convert_extent_bit can return -ENOMEM, which is most of the 1034663dfbb0SFilipe Manana * time a temporary error. So when it happens, ignore the error 1035663dfbb0SFilipe Manana * and wait for writeback of this range to finish - because we 1036663dfbb0SFilipe Manana * failed to set the bit EXTENT_NEED_WAIT for the range, a call 1037bf89d38fSJeff Mahoney * to __btrfs_wait_marked_extents() would not know that 1038bf89d38fSJeff Mahoney * writeback for this range started and therefore wouldn't 1039bf89d38fSJeff Mahoney * wait for it to finish - we don't want to commit a 1040bf89d38fSJeff Mahoney * superblock that points to btree nodes/leafs for which 1041bf89d38fSJeff Mahoney * writeback hasn't finished yet (and without errors). 1042663dfbb0SFilipe Manana * We cleanup any entries left in the io tree when committing 104341e7acd3SNikolay Borisov * the transaction (through extent_io_tree_release()). 1044663dfbb0SFilipe Manana */ 1045663dfbb0SFilipe Manana if (err == -ENOMEM) { 1046663dfbb0SFilipe Manana err = 0; 1047663dfbb0SFilipe Manana wait_writeback = true; 1048663dfbb0SFilipe Manana } 1049663dfbb0SFilipe Manana if (!err) 10501728366eSJosef Bacik err = filemap_fdatawrite_range(mapping, start, end); 10517c4452b9SChris Mason if (err) 10527c4452b9SChris Mason werr = err; 1053663dfbb0SFilipe Manana else if (wait_writeback) 1054663dfbb0SFilipe Manana werr = filemap_fdatawait_range(mapping, start, end); 1055e38e2ed7SFilipe Manana free_extent_state(cached_state); 1056663dfbb0SFilipe Manana cached_state = NULL; 10571728366eSJosef Bacik cond_resched(); 10581728366eSJosef Bacik start = end + 1; 10597c4452b9SChris Mason } 10606300463bSLiu Bo atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); 1061690587d1SChris Mason return werr; 1062690587d1SChris Mason } 1063690587d1SChris Mason 1064690587d1SChris Mason /* 1065690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1066690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1067690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 1068690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 1069690587d1SChris Mason */ 1070bf89d38fSJeff Mahoney static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, 1071bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1072690587d1SChris Mason { 1073690587d1SChris Mason int err = 0; 1074690587d1SChris Mason int werr = 0; 10750b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1076e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1077690587d1SChris Mason u64 start = 0; 1078690587d1SChris Mason u64 end; 1079690587d1SChris Mason 10801728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1081e6138876SJosef Bacik EXTENT_NEED_WAIT, &cached_state)) { 1082663dfbb0SFilipe Manana /* 1083663dfbb0SFilipe Manana * Ignore -ENOMEM errors returned by clear_extent_bit(). 1084663dfbb0SFilipe Manana * When committing the transaction, we'll remove any entries 1085663dfbb0SFilipe Manana * left in the io tree. For a log commit, we don't remove them 1086663dfbb0SFilipe Manana * after committing the log because the tree can be accessed 1087663dfbb0SFilipe Manana * concurrently - we do it only at transaction commit time when 108841e7acd3SNikolay Borisov * it's safe to do it (through extent_io_tree_release()). 1089663dfbb0SFilipe Manana */ 1090663dfbb0SFilipe Manana err = clear_extent_bit(dirty_pages, start, end, 1091ae0f1625SDavid Sterba EXTENT_NEED_WAIT, 0, 0, &cached_state); 1092663dfbb0SFilipe Manana if (err == -ENOMEM) 1093663dfbb0SFilipe Manana err = 0; 1094663dfbb0SFilipe Manana if (!err) 10951728366eSJosef Bacik err = filemap_fdatawait_range(mapping, start, end); 1096777e6bd7SChris Mason if (err) 1097777e6bd7SChris Mason werr = err; 1098e38e2ed7SFilipe Manana free_extent_state(cached_state); 1099e38e2ed7SFilipe Manana cached_state = NULL; 1100777e6bd7SChris Mason cond_resched(); 11011728366eSJosef Bacik start = end + 1; 1102777e6bd7SChris Mason } 11037c4452b9SChris Mason if (err) 11047c4452b9SChris Mason werr = err; 1105bf89d38fSJeff Mahoney return werr; 1106bf89d38fSJeff Mahoney } 1107656f30dbSFilipe Manana 1108b9fae2ebSFilipe Manana static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, 1109bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1110bf89d38fSJeff Mahoney { 1111bf89d38fSJeff Mahoney bool errors = false; 1112bf89d38fSJeff Mahoney int err; 1113bf89d38fSJeff Mahoney 1114bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1115bf89d38fSJeff Mahoney if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) 1116bf89d38fSJeff Mahoney errors = true; 1117bf89d38fSJeff Mahoney 1118bf89d38fSJeff Mahoney if (errors && !err) 1119bf89d38fSJeff Mahoney err = -EIO; 1120bf89d38fSJeff Mahoney return err; 1121bf89d38fSJeff Mahoney } 1122bf89d38fSJeff Mahoney 1123bf89d38fSJeff Mahoney int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) 1124bf89d38fSJeff Mahoney { 1125bf89d38fSJeff Mahoney struct btrfs_fs_info *fs_info = log_root->fs_info; 1126bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; 1127bf89d38fSJeff Mahoney bool errors = false; 1128bf89d38fSJeff Mahoney int err; 1129bf89d38fSJeff Mahoney 1130bf89d38fSJeff Mahoney ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 1131bf89d38fSJeff Mahoney 1132bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1133656f30dbSFilipe Manana if ((mark & EXTENT_DIRTY) && 11340b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) 1135656f30dbSFilipe Manana errors = true; 1136656f30dbSFilipe Manana 1137656f30dbSFilipe Manana if ((mark & EXTENT_NEW) && 11380b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) 1139656f30dbSFilipe Manana errors = true; 1140656f30dbSFilipe Manana 1141bf89d38fSJeff Mahoney if (errors && !err) 1142bf89d38fSJeff Mahoney err = -EIO; 1143bf89d38fSJeff Mahoney return err; 114479154b1bSChris Mason } 114579154b1bSChris Mason 1146690587d1SChris Mason /* 1147c9b577c0SNikolay Borisov * When btree blocks are allocated the corresponding extents are marked dirty. 1148c9b577c0SNikolay Borisov * This function ensures such extents are persisted on disk for transaction or 1149c9b577c0SNikolay Borisov * log commit. 1150c9b577c0SNikolay Borisov * 1151c9b577c0SNikolay Borisov * @trans: transaction whose dirty pages we'd like to write 1152690587d1SChris Mason */ 115370458a58SNikolay Borisov static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) 1154d0c803c4SChris Mason { 1155663dfbb0SFilipe Manana int ret; 1156c9b577c0SNikolay Borisov int ret2; 1157c9b577c0SNikolay Borisov struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; 115870458a58SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1159c9b577c0SNikolay Borisov struct blk_plug plug; 1160663dfbb0SFilipe Manana 1161c9b577c0SNikolay Borisov blk_start_plug(&plug); 1162c9b577c0SNikolay Borisov ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); 1163c9b577c0SNikolay Borisov blk_finish_plug(&plug); 1164c9b577c0SNikolay Borisov ret2 = btrfs_wait_extents(fs_info, dirty_pages); 1165c9b577c0SNikolay Borisov 116641e7acd3SNikolay Borisov extent_io_tree_release(&trans->transaction->dirty_pages); 1167663dfbb0SFilipe Manana 1168c9b577c0SNikolay Borisov if (ret) 1169663dfbb0SFilipe Manana return ret; 1170c9b577c0SNikolay Borisov else if (ret2) 1171c9b577c0SNikolay Borisov return ret2; 1172c9b577c0SNikolay Borisov else 1173c9b577c0SNikolay Borisov return 0; 1174d0c803c4SChris Mason } 1175d0c803c4SChris Mason 1176d352ac68SChris Mason /* 1177d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 1178d352ac68SChris Mason * 1179d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 1180d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 1181d352ac68SChris Mason * allocation tree. 1182d352ac68SChris Mason * 1183d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 1184d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 1185d352ac68SChris Mason */ 11860b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 118779154b1bSChris Mason struct btrfs_root *root) 118879154b1bSChris Mason { 118979154b1bSChris Mason int ret; 11900b86a832SChris Mason u64 old_root_bytenr; 119186b9f2ecSYan, Zheng u64 old_root_used; 11920b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 11930b246afaSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 119479154b1bSChris Mason 119586b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 119656bec294SChris Mason 119779154b1bSChris Mason while (1) { 11980b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 119986b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 1200ea526d18SJosef Bacik old_root_used == btrfs_root_used(&root->root_item)) 120179154b1bSChris Mason break; 120287ef2bb4SChris Mason 12035d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 120479154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 12050b86a832SChris Mason &root->root_key, 12060b86a832SChris Mason &root->root_item); 120749b25e05SJeff Mahoney if (ret) 120849b25e05SJeff Mahoney return ret; 120956bec294SChris Mason 121086b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 1211e7070be1SJosef Bacik } 1212276e680dSYan Zheng 12130b86a832SChris Mason return 0; 12140b86a832SChris Mason } 12150b86a832SChris Mason 1216d352ac68SChris Mason /* 1217d352ac68SChris Mason * update all the cowonly tree roots on disk 121849b25e05SJeff Mahoney * 121949b25e05SJeff Mahoney * The error handling in this function may not be obvious. Any of the 122049b25e05SJeff Mahoney * failures will cause the file system to go offline. We still need 122149b25e05SJeff Mahoney * to clean up the delayed refs. 1222d352ac68SChris Mason */ 12239386d8bcSNikolay Borisov static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) 12240b86a832SChris Mason { 12259386d8bcSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1226ea526d18SJosef Bacik struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; 12271bbc621eSChris Mason struct list_head *io_bgs = &trans->transaction->io_bgs; 12280b86a832SChris Mason struct list_head *next; 122984234f3aSYan Zheng struct extent_buffer *eb; 123056bec294SChris Mason int ret; 123184234f3aSYan Zheng 123284234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 123349b25e05SJeff Mahoney ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 12349631e4ccSJosef Bacik 0, &eb, BTRFS_NESTING_COW); 123584234f3aSYan Zheng btrfs_tree_unlock(eb); 123684234f3aSYan Zheng free_extent_buffer(eb); 12370b86a832SChris Mason 123849b25e05SJeff Mahoney if (ret) 123949b25e05SJeff Mahoney return ret; 124049b25e05SJeff Mahoney 1241196c9d8dSDavid Sterba ret = btrfs_run_dev_stats(trans); 1242c16ce190SJosef Bacik if (ret) 1243c16ce190SJosef Bacik return ret; 12442b584c68SDavid Sterba ret = btrfs_run_dev_replace(trans); 1245c16ce190SJosef Bacik if (ret) 1246c16ce190SJosef Bacik return ret; 1247280f8bd2SLu Fengqi ret = btrfs_run_qgroups(trans); 1248c16ce190SJosef Bacik if (ret) 1249c16ce190SJosef Bacik return ret; 1250546adb0dSJan Schmidt 1251bbebb3e0SDavid Sterba ret = btrfs_setup_space_cache(trans); 1252dcdf7f6dSJosef Bacik if (ret) 1253dcdf7f6dSJosef Bacik return ret; 1254dcdf7f6dSJosef Bacik 1255ea526d18SJosef Bacik again: 12560b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 12572ff7e61eSJeff Mahoney struct btrfs_root *root; 12580b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 12590b86a832SChris Mason list_del_init(next); 12600b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 1261e7070be1SJosef Bacik clear_bit(BTRFS_ROOT_DIRTY, &root->state); 126287ef2bb4SChris Mason 12639e351cc8SJosef Bacik if (root != fs_info->extent_root) 12649e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 12659e351cc8SJosef Bacik &trans->transaction->switch_commits); 126649b25e05SJeff Mahoney ret = update_cowonly_root(trans, root); 126749b25e05SJeff Mahoney if (ret) 126849b25e05SJeff Mahoney return ret; 1269488bc2a2SJosef Bacik } 1270488bc2a2SJosef Bacik 1271488bc2a2SJosef Bacik /* Now flush any delayed refs generated by updating all of the roots */ 1272c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1273ea526d18SJosef Bacik if (ret) 1274ea526d18SJosef Bacik return ret; 1275276e680dSYan Zheng 12761bbc621eSChris Mason while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { 12775742d15fSDavid Sterba ret = btrfs_write_dirty_block_groups(trans); 1278ea526d18SJosef Bacik if (ret) 1279ea526d18SJosef Bacik return ret; 1280488bc2a2SJosef Bacik 1281488bc2a2SJosef Bacik /* 1282488bc2a2SJosef Bacik * We're writing the dirty block groups, which could generate 1283488bc2a2SJosef Bacik * delayed refs, which could generate more dirty block groups, 1284488bc2a2SJosef Bacik * so we want to keep this flushing in this loop to make sure 1285488bc2a2SJosef Bacik * everything gets run. 1286488bc2a2SJosef Bacik */ 1287c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1288ea526d18SJosef Bacik if (ret) 1289ea526d18SJosef Bacik return ret; 1290ea526d18SJosef Bacik } 1291ea526d18SJosef Bacik 1292ea526d18SJosef Bacik if (!list_empty(&fs_info->dirty_cowonly_roots)) 1293ea526d18SJosef Bacik goto again; 1294ea526d18SJosef Bacik 12959e351cc8SJosef Bacik list_add_tail(&fs_info->extent_root->dirty_list, 12969e351cc8SJosef Bacik &trans->transaction->switch_commits); 12979f6cbcbbSDavid Sterba 12989f6cbcbbSDavid Sterba /* Update dev-replace pointer once everything is committed */ 12999f6cbcbbSDavid Sterba fs_info->dev_replace.committed_cursor_left = 13009f6cbcbbSDavid Sterba fs_info->dev_replace.cursor_left_last_write_of_item; 13018dabb742SStefan Behrens 130279154b1bSChris Mason return 0; 130379154b1bSChris Mason } 130479154b1bSChris Mason 1305d352ac68SChris Mason /* 1306d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 1307d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 1308d352ac68SChris Mason * be deleted 1309d352ac68SChris Mason */ 1310cfad392bSJosef Bacik void btrfs_add_dead_root(struct btrfs_root *root) 13115eda7b5eSChris Mason { 13120b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13130b246afaSJeff Mahoney 13140b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1315dc9492c1SJosef Bacik if (list_empty(&root->root_list)) { 1316dc9492c1SJosef Bacik btrfs_grab_root(root); 13170b246afaSJeff Mahoney list_add_tail(&root->root_list, &fs_info->dead_roots); 1318dc9492c1SJosef Bacik } 13190b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 13205eda7b5eSChris Mason } 13215eda7b5eSChris Mason 1322d352ac68SChris Mason /* 13235d4f98a2SYan Zheng * update all the cowonly tree roots on disk 1324d352ac68SChris Mason */ 13257e4443d9SNikolay Borisov static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) 13260f7d52f4SChris Mason { 13277e4443d9SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 13280f7d52f4SChris Mason struct btrfs_root *gang[8]; 13290f7d52f4SChris Mason int i; 13300f7d52f4SChris Mason int ret; 133154aa1f4dSChris Mason 1332a4abeea4SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 13330f7d52f4SChris Mason while (1) { 13345d4f98a2SYan Zheng ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 13355d4f98a2SYan Zheng (void **)gang, 0, 13360f7d52f4SChris Mason ARRAY_SIZE(gang), 13370f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 13380f7d52f4SChris Mason if (ret == 0) 13390f7d52f4SChris Mason break; 13400f7d52f4SChris Mason for (i = 0; i < ret; i++) { 13415b4aacefSJeff Mahoney struct btrfs_root *root = gang[i]; 13424f4317c1SJosef Bacik int ret2; 13434f4317c1SJosef Bacik 13445d4f98a2SYan Zheng radix_tree_tag_clear(&fs_info->fs_roots_radix, 13452619ba1fSChris Mason (unsigned long)root->root_key.objectid, 13460f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 1347a4abeea4SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 134831153d81SYan Zheng 1349e02119d5SChris Mason btrfs_free_log(trans, root); 13505d4f98a2SYan Zheng btrfs_update_reloc_root(trans, root); 1351e02119d5SChris Mason 1352f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 135327cdeb70SMiao Xie clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1354c7548af6SChris Mason smp_mb__after_atomic(); 1355f1ebcc74SLiu Bo 1356978d910dSYan Zheng if (root->commit_root != root->node) { 13579e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 13589e351cc8SJosef Bacik &trans->transaction->switch_commits); 1359978d910dSYan Zheng btrfs_set_root_node(&root->root_item, 1360978d910dSYan Zheng root->node); 1361978d910dSYan Zheng } 136231153d81SYan Zheng 13634f4317c1SJosef Bacik ret2 = btrfs_update_root(trans, fs_info->tree_root, 13640f7d52f4SChris Mason &root->root_key, 13650f7d52f4SChris Mason &root->root_item); 13664f4317c1SJosef Bacik if (ret2) 13674f4317c1SJosef Bacik return ret2; 1368a4abeea4SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 1369733e03a0SQu Wenruo btrfs_qgroup_free_meta_all_pertrans(root); 13700f7d52f4SChris Mason } 13719f3a7427SChris Mason } 1372a4abeea4SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 13734f4317c1SJosef Bacik return 0; 13740f7d52f4SChris Mason } 13750f7d52f4SChris Mason 1376d352ac68SChris Mason /* 1377de78b51aSEric Sandeen * defrag a given btree. 1378de78b51aSEric Sandeen * Every leaf in the btree is read and defragged. 1379d352ac68SChris Mason */ 1380de78b51aSEric Sandeen int btrfs_defrag_root(struct btrfs_root *root) 1381e9d0b13bSChris Mason { 1382e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 1383e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 13848929ecfaSYan, Zheng int ret; 1385e9d0b13bSChris Mason 138627cdeb70SMiao Xie if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) 1387e9d0b13bSChris Mason return 0; 13888929ecfaSYan, Zheng 13896b80053dSChris Mason while (1) { 13908929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 13918929ecfaSYan, Zheng if (IS_ERR(trans)) 13928929ecfaSYan, Zheng return PTR_ERR(trans); 13938929ecfaSYan, Zheng 1394de78b51aSEric Sandeen ret = btrfs_defrag_leaves(trans, root); 13958929ecfaSYan, Zheng 13963a45bb20SJeff Mahoney btrfs_end_transaction(trans); 13972ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(info); 1398e9d0b13bSChris Mason cond_resched(); 1399e9d0b13bSChris Mason 1400ab8d0fc4SJeff Mahoney if (btrfs_fs_closing(info) || ret != -EAGAIN) 1401e9d0b13bSChris Mason break; 1402210549ebSDavid Sterba 1403ab8d0fc4SJeff Mahoney if (btrfs_defrag_cancelled(info)) { 1404ab8d0fc4SJeff Mahoney btrfs_debug(info, "defrag_root cancelled"); 1405210549ebSDavid Sterba ret = -EAGAIN; 1406210549ebSDavid Sterba break; 1407210549ebSDavid Sterba } 1408e9d0b13bSChris Mason } 140927cdeb70SMiao Xie clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); 14108929ecfaSYan, Zheng return ret; 1411e9d0b13bSChris Mason } 1412e9d0b13bSChris Mason 1413d352ac68SChris Mason /* 14146426c7adSQu Wenruo * Do all special snapshot related qgroup dirty hack. 14156426c7adSQu Wenruo * 14166426c7adSQu Wenruo * Will do all needed qgroup inherit and dirty hack like switch commit 14176426c7adSQu Wenruo * roots inside one transaction and write all btree into disk, to make 14186426c7adSQu Wenruo * qgroup works. 14196426c7adSQu Wenruo */ 14206426c7adSQu Wenruo static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, 14216426c7adSQu Wenruo struct btrfs_root *src, 14226426c7adSQu Wenruo struct btrfs_root *parent, 14236426c7adSQu Wenruo struct btrfs_qgroup_inherit *inherit, 14246426c7adSQu Wenruo u64 dst_objectid) 14256426c7adSQu Wenruo { 14266426c7adSQu Wenruo struct btrfs_fs_info *fs_info = src->fs_info; 14276426c7adSQu Wenruo int ret; 14286426c7adSQu Wenruo 14296426c7adSQu Wenruo /* 14306426c7adSQu Wenruo * Save some performance in the case that qgroups are not 14316426c7adSQu Wenruo * enabled. If this check races with the ioctl, rescan will 14326426c7adSQu Wenruo * kick in anyway. 14336426c7adSQu Wenruo */ 14349ea6e2b5SDavid Sterba if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) 14356426c7adSQu Wenruo return 0; 14366426c7adSQu Wenruo 14376426c7adSQu Wenruo /* 143852042d8eSAndrea Gelmini * Ensure dirty @src will be committed. Or, after coming 14394d31778aSQu Wenruo * commit_fs_roots() and switch_commit_roots(), any dirty but not 14404d31778aSQu Wenruo * recorded root will never be updated again, causing an outdated root 14414d31778aSQu Wenruo * item. 14424d31778aSQu Wenruo */ 14434d31778aSQu Wenruo record_root_in_trans(trans, src, 1); 14444d31778aSQu Wenruo 14454d31778aSQu Wenruo /* 14462a4d84c1SJosef Bacik * btrfs_qgroup_inherit relies on a consistent view of the usage for the 14472a4d84c1SJosef Bacik * src root, so we must run the delayed refs here. 14482a4d84c1SJosef Bacik * 14492a4d84c1SJosef Bacik * However this isn't particularly fool proof, because there's no 14502a4d84c1SJosef Bacik * synchronization keeping us from changing the tree after this point 14512a4d84c1SJosef Bacik * before we do the qgroup_inherit, or even from making changes while 14522a4d84c1SJosef Bacik * we're doing the qgroup_inherit. But that's a problem for the future, 14532a4d84c1SJosef Bacik * for now flush the delayed refs to narrow the race window where the 14542a4d84c1SJosef Bacik * qgroup counters could end up wrong. 14552a4d84c1SJosef Bacik */ 14562a4d84c1SJosef Bacik ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 14572a4d84c1SJosef Bacik if (ret) { 14582a4d84c1SJosef Bacik btrfs_abort_transaction(trans, ret); 14592a4d84c1SJosef Bacik goto out; 14602a4d84c1SJosef Bacik } 14612a4d84c1SJosef Bacik 14622a4d84c1SJosef Bacik /* 14636426c7adSQu Wenruo * We are going to commit transaction, see btrfs_commit_transaction() 14646426c7adSQu Wenruo * comment for reason locking tree_log_mutex 14656426c7adSQu Wenruo */ 14666426c7adSQu Wenruo mutex_lock(&fs_info->tree_log_mutex); 14676426c7adSQu Wenruo 14687e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 14696426c7adSQu Wenruo if (ret) 14706426c7adSQu Wenruo goto out; 1471460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 14726426c7adSQu Wenruo if (ret < 0) 14736426c7adSQu Wenruo goto out; 14746426c7adSQu Wenruo 14756426c7adSQu Wenruo /* Now qgroup are all updated, we can inherit it to new qgroups */ 1476a9377422SLu Fengqi ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, 14776426c7adSQu Wenruo inherit); 14786426c7adSQu Wenruo if (ret < 0) 14796426c7adSQu Wenruo goto out; 14806426c7adSQu Wenruo 14816426c7adSQu Wenruo /* 14826426c7adSQu Wenruo * Now we do a simplified commit transaction, which will: 14836426c7adSQu Wenruo * 1) commit all subvolume and extent tree 14846426c7adSQu Wenruo * To ensure all subvolume and extent tree have a valid 14856426c7adSQu Wenruo * commit_root to accounting later insert_dir_item() 14866426c7adSQu Wenruo * 2) write all btree blocks onto disk 14876426c7adSQu Wenruo * This is to make sure later btree modification will be cowed 14886426c7adSQu Wenruo * Or commit_root can be populated and cause wrong qgroup numbers 14896426c7adSQu Wenruo * In this simplified commit, we don't really care about other trees 14906426c7adSQu Wenruo * like chunk and root tree, as they won't affect qgroup. 14916426c7adSQu Wenruo * And we don't write super to avoid half committed status. 14926426c7adSQu Wenruo */ 14939386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 14946426c7adSQu Wenruo if (ret) 14956426c7adSQu Wenruo goto out; 1496889bfa39SJosef Bacik switch_commit_roots(trans); 149770458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 14986426c7adSQu Wenruo if (ret) 1499f7af3934SDavid Sterba btrfs_handle_fs_error(fs_info, ret, 15006426c7adSQu Wenruo "Error while writing out transaction for qgroup"); 15016426c7adSQu Wenruo 15026426c7adSQu Wenruo out: 15036426c7adSQu Wenruo mutex_unlock(&fs_info->tree_log_mutex); 15046426c7adSQu Wenruo 15056426c7adSQu Wenruo /* 15066426c7adSQu Wenruo * Force parent root to be updated, as we recorded it before so its 15076426c7adSQu Wenruo * last_trans == cur_transid. 15086426c7adSQu Wenruo * Or it won't be committed again onto disk after later 15096426c7adSQu Wenruo * insert_dir_item() 15106426c7adSQu Wenruo */ 15116426c7adSQu Wenruo if (!ret) 15126426c7adSQu Wenruo record_root_in_trans(trans, parent, 1); 15136426c7adSQu Wenruo return ret; 15146426c7adSQu Wenruo } 15156426c7adSQu Wenruo 15166426c7adSQu Wenruo /* 1517d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 1518aec8030aSMiao Xie * transaction commit. This does the actual creation. 1519aec8030aSMiao Xie * 1520aec8030aSMiao Xie * Note: 1521aec8030aSMiao Xie * If the error which may affect the commitment of the current transaction 1522aec8030aSMiao Xie * happens, we should return the error number. If the error which just affect 1523aec8030aSMiao Xie * the creation of the pending snapshots, just return 0. 1524d352ac68SChris Mason */ 152580b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 15263063d29fSChris Mason struct btrfs_pending_snapshot *pending) 15273063d29fSChris Mason { 152808d50ca3SNikolay Borisov 152908d50ca3SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 15303063d29fSChris Mason struct btrfs_key key; 153180b6794dSChris Mason struct btrfs_root_item *new_root_item; 15323063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 15333063d29fSChris Mason struct btrfs_root *root = pending->root; 15346bdb72deSSage Weil struct btrfs_root *parent_root; 153598c9942aSLiu Bo struct btrfs_block_rsv *rsv; 15366bdb72deSSage Weil struct inode *parent_inode; 153742874b3dSMiao Xie struct btrfs_path *path; 153842874b3dSMiao Xie struct btrfs_dir_item *dir_item; 1539a22285a6SYan, Zheng struct dentry *dentry; 15403063d29fSChris Mason struct extent_buffer *tmp; 1541925baeddSChris Mason struct extent_buffer *old; 154295582b00SDeepa Dinamani struct timespec64 cur_time; 1543aec8030aSMiao Xie int ret = 0; 1544d68fc57bSYan, Zheng u64 to_reserve = 0; 15456bdb72deSSage Weil u64 index = 0; 1546a22285a6SYan, Zheng u64 objectid; 1547b83cc969SLi Zefan u64 root_flags; 15483063d29fSChris Mason 15498546b570SDavid Sterba ASSERT(pending->path); 15508546b570SDavid Sterba path = pending->path; 155142874b3dSMiao Xie 1552b0c0ea63SDavid Sterba ASSERT(pending->root_item); 1553b0c0ea63SDavid Sterba new_root_item = pending->root_item; 1554a22285a6SYan, Zheng 1555543068a2SNikolay Borisov pending->error = btrfs_get_free_objectid(tree_root, &objectid); 1556aec8030aSMiao Xie if (pending->error) 15576fa9700eSMiao Xie goto no_free_objectid; 15583063d29fSChris Mason 1559d6726335SQu Wenruo /* 1560d6726335SQu Wenruo * Make qgroup to skip current new snapshot's qgroupid, as it is 1561d6726335SQu Wenruo * accounted by later btrfs_qgroup_inherit(). 1562d6726335SQu Wenruo */ 1563d6726335SQu Wenruo btrfs_set_skip_qgroup(trans, objectid); 1564d6726335SQu Wenruo 1565147d256eSZhaolei btrfs_reloc_pre_snapshot(pending, &to_reserve); 1566d68fc57bSYan, Zheng 1567d68fc57bSYan, Zheng if (to_reserve > 0) { 1568aec8030aSMiao Xie pending->error = btrfs_block_rsv_add(root, 1569aec8030aSMiao Xie &pending->block_rsv, 157008e007d2SMiao Xie to_reserve, 157108e007d2SMiao Xie BTRFS_RESERVE_NO_FLUSH); 1572aec8030aSMiao Xie if (pending->error) 1573d6726335SQu Wenruo goto clear_skip_qgroup; 1574d68fc57bSYan, Zheng } 1575d68fc57bSYan, Zheng 15763063d29fSChris Mason key.objectid = objectid; 1577a22285a6SYan, Zheng key.offset = (u64)-1; 1578a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 15793063d29fSChris Mason 15806fa9700eSMiao Xie rsv = trans->block_rsv; 1581a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 15822382c5ccSLiu Bo trans->bytes_reserved = trans->block_rsv->reserved; 15830b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 158488d3a5aaSJosef Bacik trans->transid, 158588d3a5aaSJosef Bacik trans->bytes_reserved, 1); 1586a22285a6SYan, Zheng dentry = pending->dentry; 1587e9662f70SMiao Xie parent_inode = pending->dir; 1588a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 15896426c7adSQu Wenruo record_root_in_trans(trans, parent_root, 0); 1590a22285a6SYan, Zheng 1591c2050a45SDeepa Dinamani cur_time = current_time(parent_inode); 159204b285f3SDeepa Dinamani 15936bdb72deSSage Weil /* 15946bdb72deSSage Weil * insert the directory item 15956bdb72deSSage Weil */ 1596877574e2SNikolay Borisov ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); 159749b25e05SJeff Mahoney BUG_ON(ret); /* -ENOMEM */ 159842874b3dSMiao Xie 159942874b3dSMiao Xie /* check if there is a file/dir which has the same name. */ 160042874b3dSMiao Xie dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, 16014a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), 160242874b3dSMiao Xie dentry->d_name.name, 160342874b3dSMiao Xie dentry->d_name.len, 0); 160442874b3dSMiao Xie if (dir_item != NULL && !IS_ERR(dir_item)) { 1605fe66a05aSChris Mason pending->error = -EEXIST; 1606aec8030aSMiao Xie goto dir_item_existed; 160742874b3dSMiao Xie } else if (IS_ERR(dir_item)) { 160842874b3dSMiao Xie ret = PTR_ERR(dir_item); 160966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16108732d44fSMiao Xie goto fail; 161179787eaaSJeff Mahoney } 161242874b3dSMiao Xie btrfs_release_path(path); 16136bdb72deSSage Weil 1614e999376fSChris Mason /* 1615e999376fSChris Mason * pull in the delayed directory update 1616e999376fSChris Mason * and the delayed inode item 1617e999376fSChris Mason * otherwise we corrupt the FS during 1618e999376fSChris Mason * snapshot 1619e999376fSChris Mason */ 1620e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 16218732d44fSMiao Xie if (ret) { /* Transaction aborted */ 162266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16238732d44fSMiao Xie goto fail; 16248732d44fSMiao Xie } 1625e999376fSChris Mason 16266426c7adSQu Wenruo record_root_in_trans(trans, root, 0); 16276bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 16286bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 162908fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 16306bdb72deSSage Weil 1631b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 1632b83cc969SLi Zefan if (pending->readonly) 1633b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 1634b83cc969SLi Zefan else 1635b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 1636b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 1637b83cc969SLi Zefan 16388ea05e3aSAlexander Block btrfs_set_root_generation_v2(new_root_item, 16398ea05e3aSAlexander Block trans->transid); 1640807fc790SAndy Shevchenko generate_random_guid(new_root_item->uuid); 16418ea05e3aSAlexander Block memcpy(new_root_item->parent_uuid, root->root_item.uuid, 16428ea05e3aSAlexander Block BTRFS_UUID_SIZE); 164370023da2SStefan Behrens if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { 164470023da2SStefan Behrens memset(new_root_item->received_uuid, 0, 164570023da2SStefan Behrens sizeof(new_root_item->received_uuid)); 16468ea05e3aSAlexander Block memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); 16478ea05e3aSAlexander Block memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); 16488ea05e3aSAlexander Block btrfs_set_root_stransid(new_root_item, 0); 16498ea05e3aSAlexander Block btrfs_set_root_rtransid(new_root_item, 0); 165070023da2SStefan Behrens } 16513cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); 16523cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); 165370023da2SStefan Behrens btrfs_set_root_otransid(new_root_item, trans->transid); 16548ea05e3aSAlexander Block 1655925baeddSChris Mason old = btrfs_lock_root_node(root); 16569631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, 16579631e4ccSJosef Bacik BTRFS_NESTING_COW); 165879787eaaSJeff Mahoney if (ret) { 165979787eaaSJeff Mahoney btrfs_tree_unlock(old); 166079787eaaSJeff Mahoney free_extent_buffer(old); 166166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16628732d44fSMiao Xie goto fail; 166379787eaaSJeff Mahoney } 166449b25e05SJeff Mahoney 166549b25e05SJeff Mahoney ret = btrfs_copy_root(trans, root, old, &tmp, objectid); 166679787eaaSJeff Mahoney /* clean up in any case */ 1667925baeddSChris Mason btrfs_tree_unlock(old); 1668925baeddSChris Mason free_extent_buffer(old); 16698732d44fSMiao Xie if (ret) { 167066642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16718732d44fSMiao Xie goto fail; 16728732d44fSMiao Xie } 1673f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 167427cdeb70SMiao Xie set_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1675f1ebcc74SLiu Bo smp_wmb(); 1676f1ebcc74SLiu Bo 16775d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 1678a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 1679a22285a6SYan, Zheng key.offset = trans->transid; 1680a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1681925baeddSChris Mason btrfs_tree_unlock(tmp); 16823063d29fSChris Mason free_extent_buffer(tmp); 16838732d44fSMiao Xie if (ret) { 168466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16858732d44fSMiao Xie goto fail; 16868732d44fSMiao Xie } 16870660b5afSChris Mason 1688a22285a6SYan, Zheng /* 1689a22285a6SYan, Zheng * insert root back/forward references 1690a22285a6SYan, Zheng */ 16916025c19fSLu Fengqi ret = btrfs_add_root_ref(trans, objectid, 1692a22285a6SYan, Zheng parent_root->root_key.objectid, 16934a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), index, 1694a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len); 16958732d44fSMiao Xie if (ret) { 169666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16978732d44fSMiao Xie goto fail; 16988732d44fSMiao Xie } 1699a22285a6SYan, Zheng 1700a22285a6SYan, Zheng key.offset = (u64)-1; 17012dfb1e43SQu Wenruo pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); 170279787eaaSJeff Mahoney if (IS_ERR(pending->snap)) { 170379787eaaSJeff Mahoney ret = PTR_ERR(pending->snap); 17042d892ccdSFilipe Manana pending->snap = NULL; 170566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17068732d44fSMiao Xie goto fail; 170779787eaaSJeff Mahoney } 1708d68fc57bSYan, Zheng 170949b25e05SJeff Mahoney ret = btrfs_reloc_post_snapshot(trans, pending); 17108732d44fSMiao Xie if (ret) { 171166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17128732d44fSMiao Xie goto fail; 17138732d44fSMiao Xie } 1714361048f5SMiao Xie 17156426c7adSQu Wenruo /* 17166426c7adSQu Wenruo * Do special qgroup accounting for snapshot, as we do some qgroup 17176426c7adSQu Wenruo * snapshot hack to do fast snapshot. 17186426c7adSQu Wenruo * To co-operate with that hack, we do hack again. 17196426c7adSQu Wenruo * Or snapshot will be greatly slowed down by a subtree qgroup rescan 17206426c7adSQu Wenruo */ 17216426c7adSQu Wenruo ret = qgroup_account_snapshot(trans, root, parent_root, 17226426c7adSQu Wenruo pending->inherit, objectid); 17236426c7adSQu Wenruo if (ret < 0) 17246426c7adSQu Wenruo goto fail; 17256426c7adSQu Wenruo 1726684572dfSLu Fengqi ret = btrfs_insert_dir_item(trans, dentry->d_name.name, 1727684572dfSLu Fengqi dentry->d_name.len, BTRFS_I(parent_inode), 1728684572dfSLu Fengqi &key, BTRFS_FT_DIR, index); 172942874b3dSMiao Xie /* We have check then name at the beginning, so it is impossible. */ 17309c52057cSChris Mason BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); 17318732d44fSMiao Xie if (ret) { 173266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17338732d44fSMiao Xie goto fail; 17348732d44fSMiao Xie } 173542874b3dSMiao Xie 17366ef06d27SNikolay Borisov btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + 173742874b3dSMiao Xie dentry->d_name.len * 2); 173804b285f3SDeepa Dinamani parent_inode->i_mtime = parent_inode->i_ctime = 1739c2050a45SDeepa Dinamani current_time(parent_inode); 1740729f7961SNikolay Borisov ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); 1741dd5f9615SStefan Behrens if (ret) { 174266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1743dd5f9615SStefan Behrens goto fail; 1744dd5f9615SStefan Behrens } 1745807fc790SAndy Shevchenko ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, 1746807fc790SAndy Shevchenko BTRFS_UUID_KEY_SUBVOL, 1747cdb345a8SLu Fengqi objectid); 1748dd5f9615SStefan Behrens if (ret) { 174966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1750dd5f9615SStefan Behrens goto fail; 1751dd5f9615SStefan Behrens } 1752dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { 1753cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, 1754dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 1755dd5f9615SStefan Behrens objectid); 1756dd5f9615SStefan Behrens if (ret && ret != -EEXIST) { 175766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1758dd5f9615SStefan Behrens goto fail; 1759dd5f9615SStefan Behrens } 1760dd5f9615SStefan Behrens } 1761d6726335SQu Wenruo 17623063d29fSChris Mason fail: 1763aec8030aSMiao Xie pending->error = ret; 1764aec8030aSMiao Xie dir_item_existed: 176598c9942aSLiu Bo trans->block_rsv = rsv; 17662382c5ccSLiu Bo trans->bytes_reserved = 0; 1767d6726335SQu Wenruo clear_skip_qgroup: 1768d6726335SQu Wenruo btrfs_clear_skip_qgroup(trans); 17696fa9700eSMiao Xie no_free_objectid: 17706fa9700eSMiao Xie kfree(new_root_item); 1771b0c0ea63SDavid Sterba pending->root_item = NULL; 177242874b3dSMiao Xie btrfs_free_path(path); 17738546b570SDavid Sterba pending->path = NULL; 17748546b570SDavid Sterba 177549b25e05SJeff Mahoney return ret; 17763063d29fSChris Mason } 17773063d29fSChris Mason 1778d352ac68SChris Mason /* 1779d352ac68SChris Mason * create all the snapshots we've scheduled for creation 1780d352ac68SChris Mason */ 178108d50ca3SNikolay Borisov static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) 17823063d29fSChris Mason { 1783aec8030aSMiao Xie struct btrfs_pending_snapshot *pending, *next; 17843063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 1785aec8030aSMiao Xie int ret = 0; 17863de4586cSChris Mason 1787aec8030aSMiao Xie list_for_each_entry_safe(pending, next, head, list) { 1788aec8030aSMiao Xie list_del(&pending->list); 178908d50ca3SNikolay Borisov ret = create_pending_snapshot(trans, pending); 1790aec8030aSMiao Xie if (ret) 1791aec8030aSMiao Xie break; 1792aec8030aSMiao Xie } 1793aec8030aSMiao Xie return ret; 17943de4586cSChris Mason } 17953de4586cSChris Mason 17962ff7e61eSJeff Mahoney static void update_super_roots(struct btrfs_fs_info *fs_info) 17975d4f98a2SYan Zheng { 17985d4f98a2SYan Zheng struct btrfs_root_item *root_item; 17995d4f98a2SYan Zheng struct btrfs_super_block *super; 18005d4f98a2SYan Zheng 18010b246afaSJeff Mahoney super = fs_info->super_copy; 18025d4f98a2SYan Zheng 18030b246afaSJeff Mahoney root_item = &fs_info->chunk_root->root_item; 1804093e037cSDavid Sterba super->chunk_root = root_item->bytenr; 1805093e037cSDavid Sterba super->chunk_root_generation = root_item->generation; 1806093e037cSDavid Sterba super->chunk_root_level = root_item->level; 18075d4f98a2SYan Zheng 18080b246afaSJeff Mahoney root_item = &fs_info->tree_root->root_item; 1809093e037cSDavid Sterba super->root = root_item->bytenr; 1810093e037cSDavid Sterba super->generation = root_item->generation; 1811093e037cSDavid Sterba super->root_level = root_item->level; 18120b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, SPACE_CACHE)) 1813093e037cSDavid Sterba super->cache_generation = root_item->generation; 181494846229SBoris Burkov else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) 181594846229SBoris Burkov super->cache_generation = 0; 18160b246afaSJeff Mahoney if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) 1817093e037cSDavid Sterba super->uuid_tree_generation = root_item->generation; 18185d4f98a2SYan Zheng } 18195d4f98a2SYan Zheng 1820f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1821f36f3042SChris Mason { 18224a9d8bdeSMiao Xie struct btrfs_transaction *trans; 1823f36f3042SChris Mason int ret = 0; 18244a9d8bdeSMiao Xie 1825a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 18264a9d8bdeSMiao Xie trans = info->running_transaction; 18274a9d8bdeSMiao Xie if (trans) 18284a9d8bdeSMiao Xie ret = (trans->state >= TRANS_STATE_COMMIT_START); 1829a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 1830f36f3042SChris Mason return ret; 1831f36f3042SChris Mason } 1832f36f3042SChris Mason 18338929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 18348929ecfaSYan, Zheng { 18354a9d8bdeSMiao Xie struct btrfs_transaction *trans; 18368929ecfaSYan, Zheng int ret = 0; 18374a9d8bdeSMiao Xie 1838a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 18394a9d8bdeSMiao Xie trans = info->running_transaction; 18404a9d8bdeSMiao Xie if (trans) 18414a9d8bdeSMiao Xie ret = is_transaction_blocked(trans); 1842a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 18438929ecfaSYan, Zheng return ret; 18448929ecfaSYan, Zheng } 18458929ecfaSYan, Zheng 1846bb9c12c9SSage Weil /* 1847bb9c12c9SSage Weil * wait for the current transaction commit to start and block subsequent 1848bb9c12c9SSage Weil * transaction joins 1849bb9c12c9SSage Weil */ 18502ff7e61eSJeff Mahoney static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info, 1851bb9c12c9SSage Weil struct btrfs_transaction *trans) 1852bb9c12c9SSage Weil { 18532ff7e61eSJeff Mahoney wait_event(fs_info->transaction_blocked_wait, 1854bf31f87fSDavid Sterba trans->state >= TRANS_STATE_COMMIT_START || 1855bf31f87fSDavid Sterba TRANS_ABORTED(trans)); 1856bb9c12c9SSage Weil } 1857bb9c12c9SSage Weil 1858bb9c12c9SSage Weil /* 1859bb9c12c9SSage Weil * wait for the current transaction to start and then become unblocked. 1860bb9c12c9SSage Weil * caller holds ref. 1861bb9c12c9SSage Weil */ 18622ff7e61eSJeff Mahoney static void wait_current_trans_commit_start_and_unblock( 18632ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 1864bb9c12c9SSage Weil struct btrfs_transaction *trans) 1865bb9c12c9SSage Weil { 18662ff7e61eSJeff Mahoney wait_event(fs_info->transaction_wait, 1867bf31f87fSDavid Sterba trans->state >= TRANS_STATE_UNBLOCKED || 1868bf31f87fSDavid Sterba TRANS_ABORTED(trans)); 1869bb9c12c9SSage Weil } 1870bb9c12c9SSage Weil 1871bb9c12c9SSage Weil /* 1872bb9c12c9SSage Weil * commit transactions asynchronously. once btrfs_commit_transaction_async 1873bb9c12c9SSage Weil * returns, any subsequent transaction will not be allowed to join. 1874bb9c12c9SSage Weil */ 1875bb9c12c9SSage Weil struct btrfs_async_commit { 1876bb9c12c9SSage Weil struct btrfs_trans_handle *newtrans; 18777892b5afSMiao Xie struct work_struct work; 1878bb9c12c9SSage Weil }; 1879bb9c12c9SSage Weil 1880bb9c12c9SSage Weil static void do_async_commit(struct work_struct *work) 1881bb9c12c9SSage Weil { 1882bb9c12c9SSage Weil struct btrfs_async_commit *ac = 18837892b5afSMiao Xie container_of(work, struct btrfs_async_commit, work); 1884bb9c12c9SSage Weil 18856fc4e354SSage Weil /* 18866fc4e354SSage Weil * We've got freeze protection passed with the transaction. 18876fc4e354SSage Weil * Tell lockdep about it. 18886fc4e354SSage Weil */ 1889b1a06a4bSLiu Bo if (ac->newtrans->type & __TRANS_FREEZABLE) 18903a45bb20SJeff Mahoney __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS); 18916fc4e354SSage Weil 1892e209db7aSSage Weil current->journal_info = ac->newtrans; 1893e209db7aSSage Weil 18943a45bb20SJeff Mahoney btrfs_commit_transaction(ac->newtrans); 1895bb9c12c9SSage Weil kfree(ac); 1896bb9c12c9SSage Weil } 1897bb9c12c9SSage Weil 1898bb9c12c9SSage Weil int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, 1899bb9c12c9SSage Weil int wait_for_unblock) 1900bb9c12c9SSage Weil { 19013a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 1902bb9c12c9SSage Weil struct btrfs_async_commit *ac; 1903bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1904bb9c12c9SSage Weil 1905bb9c12c9SSage Weil ac = kmalloc(sizeof(*ac), GFP_NOFS); 1906db5b493aSTsutomu Itoh if (!ac) 1907db5b493aSTsutomu Itoh return -ENOMEM; 1908bb9c12c9SSage Weil 19097892b5afSMiao Xie INIT_WORK(&ac->work, do_async_commit); 19103a45bb20SJeff Mahoney ac->newtrans = btrfs_join_transaction(trans->root); 19113612b495STsutomu Itoh if (IS_ERR(ac->newtrans)) { 19123612b495STsutomu Itoh int err = PTR_ERR(ac->newtrans); 19133612b495STsutomu Itoh kfree(ac); 19143612b495STsutomu Itoh return err; 19153612b495STsutomu Itoh } 1916bb9c12c9SSage Weil 1917bb9c12c9SSage Weil /* take transaction reference */ 1918bb9c12c9SSage Weil cur_trans = trans->transaction; 19199b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 1920bb9c12c9SSage Weil 19213a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19226fc4e354SSage Weil 19236fc4e354SSage Weil /* 19246fc4e354SSage Weil * Tell lockdep we've released the freeze rwsem, since the 19256fc4e354SSage Weil * async commit thread will be the one to unlock it. 19266fc4e354SSage Weil */ 1927b1a06a4bSLiu Bo if (ac->newtrans->type & __TRANS_FREEZABLE) 19280b246afaSJeff Mahoney __sb_writers_release(fs_info->sb, SB_FREEZE_FS); 19296fc4e354SSage Weil 19307892b5afSMiao Xie schedule_work(&ac->work); 1931bb9c12c9SSage Weil 1932bb9c12c9SSage Weil /* wait for transaction to start and unblock */ 1933bb9c12c9SSage Weil if (wait_for_unblock) 19342ff7e61eSJeff Mahoney wait_current_trans_commit_start_and_unblock(fs_info, cur_trans); 1935bb9c12c9SSage Weil else 19362ff7e61eSJeff Mahoney wait_current_trans_commit_start(fs_info, cur_trans); 1937bb9c12c9SSage Weil 193838e88054SSage Weil if (current->journal_info == trans) 193938e88054SSage Weil current->journal_info = NULL; 194038e88054SSage Weil 1941724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1942bb9c12c9SSage Weil return 0; 1943bb9c12c9SSage Weil } 1944bb9c12c9SSage Weil 194549b25e05SJeff Mahoney 194697cb39bbSNikolay Borisov static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) 194749b25e05SJeff Mahoney { 194897cb39bbSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 194949b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 195049b25e05SJeff Mahoney 1951b50fff81SDavid Sterba WARN_ON(refcount_read(&trans->use_count) > 1); 195249b25e05SJeff Mahoney 195366642832SJeff Mahoney btrfs_abort_transaction(trans, err); 19547b8b92afSJosef Bacik 19550b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 195666b6135bSLiu Bo 195725d8c284SMiao Xie /* 195825d8c284SMiao Xie * If the transaction is removed from the list, it means this 195925d8c284SMiao Xie * transaction has been committed successfully, so it is impossible 196025d8c284SMiao Xie * to call the cleanup function. 196125d8c284SMiao Xie */ 196225d8c284SMiao Xie BUG_ON(list_empty(&cur_trans->list)); 196366b6135bSLiu Bo 196449b25e05SJeff Mahoney list_del_init(&cur_trans->list); 19650b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) { 19664a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 19670b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1968f094ac32SLiu Bo wait_event(cur_trans->writer_wait, 1969f094ac32SLiu Bo atomic_read(&cur_trans->num_writers) == 1); 1970f094ac32SLiu Bo 19710b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1972d7096fc3SJosef Bacik } 19730b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 197449b25e05SJeff Mahoney 19752ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(trans->transaction, fs_info); 197649b25e05SJeff Mahoney 19770b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 19780b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) 19790b246afaSJeff Mahoney fs_info->running_transaction = NULL; 19800b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 19814a9d8bdeSMiao Xie 1982e0228285SJosef Bacik if (trans->type & __TRANS_FREEZABLE) 19830b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 1984724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1985724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 198649b25e05SJeff Mahoney 198797cb39bbSNikolay Borisov trace_btrfs_transaction_commit(trans->root); 198849b25e05SJeff Mahoney 198949b25e05SJeff Mahoney if (current->journal_info == trans) 199049b25e05SJeff Mahoney current->journal_info = NULL; 19910b246afaSJeff Mahoney btrfs_scrub_cancel(fs_info); 199249b25e05SJeff Mahoney 199349b25e05SJeff Mahoney kmem_cache_free(btrfs_trans_handle_cachep, trans); 199449b25e05SJeff Mahoney } 199549b25e05SJeff Mahoney 1996c7cc64a9SDavid Sterba /* 1997c7cc64a9SDavid Sterba * Release reserved delayed ref space of all pending block groups of the 1998c7cc64a9SDavid Sterba * transaction and remove them from the list 1999c7cc64a9SDavid Sterba */ 2000c7cc64a9SDavid Sterba static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) 2001c7cc64a9SDavid Sterba { 2002c7cc64a9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 200332da5386SDavid Sterba struct btrfs_block_group *block_group, *tmp; 2004c7cc64a9SDavid Sterba 2005c7cc64a9SDavid Sterba list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { 2006c7cc64a9SDavid Sterba btrfs_delayed_refs_rsv_release(fs_info, 1); 2007c7cc64a9SDavid Sterba list_del_init(&block_group->bg_list); 2008c7cc64a9SDavid Sterba } 2009c7cc64a9SDavid Sterba } 2010c7cc64a9SDavid Sterba 201188090ad3SFilipe Manana static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) 201282436617SMiao Xie { 2013ce8ea7ccSJosef Bacik /* 2014ce8ea7ccSJosef Bacik * We use writeback_inodes_sb here because if we used 2015ce8ea7ccSJosef Bacik * btrfs_start_delalloc_roots we would deadlock with fs freeze. 2016ce8ea7ccSJosef Bacik * Currently are holding the fs freeze lock, if we do an async flush 2017ce8ea7ccSJosef Bacik * we'll do btrfs_join_transaction() and deadlock because we need to 2018ce8ea7ccSJosef Bacik * wait for the fs freeze lock. Using the direct flushing we benefit 2019ce8ea7ccSJosef Bacik * from already being in a transaction and our join_transaction doesn't 2020ce8ea7ccSJosef Bacik * have to re-take the fs freeze lock. 2021ce8ea7ccSJosef Bacik */ 202288090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 2023ce8ea7ccSJosef Bacik writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); 202482436617SMiao Xie return 0; 202582436617SMiao Xie } 202682436617SMiao Xie 202788090ad3SFilipe Manana static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) 202882436617SMiao Xie { 202988090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 20306374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 203182436617SMiao Xie } 203282436617SMiao Xie 20333a45bb20SJeff Mahoney int btrfs_commit_transaction(struct btrfs_trans_handle *trans) 203479154b1bSChris Mason { 20353a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 203649b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 20378fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 203825287e0aSMiao Xie int ret; 203979154b1bSChris Mason 204035b814f3SNikolay Borisov ASSERT(refcount_read(&trans->use_count) == 1); 204135b814f3SNikolay Borisov 2042d62b23c9SJosef Bacik /* 2043d62b23c9SJosef Bacik * Some places just start a transaction to commit it. We need to make 2044d62b23c9SJosef Bacik * sure that if this commit fails that the abort code actually marks the 2045d62b23c9SJosef Bacik * transaction as failed, so set trans->dirty to make the abort code do 2046d62b23c9SJosef Bacik * the right thing. 2047d62b23c9SJosef Bacik */ 2048d62b23c9SJosef Bacik trans->dirty = true; 2049d62b23c9SJosef Bacik 20508d25a086SMiao Xie /* Stop the commit early if ->aborted is set */ 2051bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 205225287e0aSMiao Xie ret = cur_trans->aborted; 20533a45bb20SJeff Mahoney btrfs_end_transaction(trans); 2054e4a2bcacSJosef Bacik return ret; 205525287e0aSMiao Xie } 205649b25e05SJeff Mahoney 2057f45c752bSJosef Bacik btrfs_trans_release_metadata(trans); 2058f45c752bSJosef Bacik trans->block_rsv = NULL; 2059f45c752bSJosef Bacik 2060e19eb11fSJosef Bacik /* 2061e19eb11fSJosef Bacik * We only want one transaction commit doing the flushing so we do not 2062e19eb11fSJosef Bacik * waste a bunch of time on lock contention on the extent root node. 2063e19eb11fSJosef Bacik */ 2064e19eb11fSJosef Bacik if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, 2065e19eb11fSJosef Bacik &cur_trans->delayed_refs.flags)) { 2066e19eb11fSJosef Bacik /* 2067e19eb11fSJosef Bacik * Make a pass through all the delayed refs we have so far. 2068e19eb11fSJosef Bacik * Any running threads may add more while we are here. 206956bec294SChris Mason */ 2070c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, 0); 2071e4a2bcacSJosef Bacik if (ret) { 20723a45bb20SJeff Mahoney btrfs_end_transaction(trans); 2073e4a2bcacSJosef Bacik return ret; 2074e4a2bcacSJosef Bacik } 2075e19eb11fSJosef Bacik } 207656bec294SChris Mason 20776c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 2078ea658badSJosef Bacik 20793204d33cSJosef Bacik if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { 20801bbc621eSChris Mason int run_it = 0; 20811bbc621eSChris Mason 20821bbc621eSChris Mason /* this mutex is also taken before trying to set 20831bbc621eSChris Mason * block groups readonly. We need to make sure 20841bbc621eSChris Mason * that nobody has set a block group readonly 20851bbc621eSChris Mason * after a extents from that block group have been 20861bbc621eSChris Mason * allocated for cache files. btrfs_set_block_group_ro 20871bbc621eSChris Mason * will wait for the transaction to commit if it 20883204d33cSJosef Bacik * finds BTRFS_TRANS_DIRTY_BG_RUN set. 20891bbc621eSChris Mason * 20903204d33cSJosef Bacik * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure 20913204d33cSJosef Bacik * only one process starts all the block group IO. It wouldn't 20921bbc621eSChris Mason * hurt to have more than one go through, but there's no 20931bbc621eSChris Mason * real advantage to it either. 20941bbc621eSChris Mason */ 20950b246afaSJeff Mahoney mutex_lock(&fs_info->ro_block_group_mutex); 20963204d33cSJosef Bacik if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, 20973204d33cSJosef Bacik &cur_trans->flags)) 20981bbc621eSChris Mason run_it = 1; 20990b246afaSJeff Mahoney mutex_unlock(&fs_info->ro_block_group_mutex); 21001bbc621eSChris Mason 2101f9cacae3SNikolay Borisov if (run_it) { 210221217054SNikolay Borisov ret = btrfs_start_dirty_block_groups(trans); 21031bbc621eSChris Mason if (ret) { 21043a45bb20SJeff Mahoney btrfs_end_transaction(trans); 21051bbc621eSChris Mason return ret; 21061bbc621eSChris Mason } 2107f9cacae3SNikolay Borisov } 2108f9cacae3SNikolay Borisov } 21091bbc621eSChris Mason 21100b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 21114a9d8bdeSMiao Xie if (cur_trans->state >= TRANS_STATE_COMMIT_START) { 2112d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2113d0c2f4faSFilipe Manana 21140b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 21159b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 2116ccd467d6SChris Mason 2117d0c2f4faSFilipe Manana if (trans->in_fsync) 2118d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2119d0c2f4faSFilipe Manana ret = btrfs_end_transaction(trans); 2120d0c2f4faSFilipe Manana wait_for_commit(cur_trans, want_state); 212115ee9bc7SJosef Bacik 2122bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) 2123b4924a0fSLiu Bo ret = cur_trans->aborted; 2124b4924a0fSLiu Bo 2125724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 212615ee9bc7SJosef Bacik 212749b25e05SJeff Mahoney return ret; 212879154b1bSChris Mason } 21294313b399SChris Mason 21304a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 21310b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 2132bb9c12c9SSage Weil 21330b246afaSJeff Mahoney if (cur_trans->list.prev != &fs_info->trans_list) { 2134d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2135d0c2f4faSFilipe Manana 2136d0c2f4faSFilipe Manana if (trans->in_fsync) 2137d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2138d0c2f4faSFilipe Manana 2139ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 2140ccd467d6SChris Mason struct btrfs_transaction, list); 2141d0c2f4faSFilipe Manana if (prev_trans->state < want_state) { 21429b64f57dSElena Reshetova refcount_inc(&prev_trans->use_count); 21430b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2144ccd467d6SChris Mason 2145d0c2f4faSFilipe Manana wait_for_commit(prev_trans, want_state); 2146d0c2f4faSFilipe Manana 2147bf31f87fSDavid Sterba ret = READ_ONCE(prev_trans->aborted); 2148ccd467d6SChris Mason 2149724e2315SJosef Bacik btrfs_put_transaction(prev_trans); 21501f9b8c8fSFilipe Manana if (ret) 21511f9b8c8fSFilipe Manana goto cleanup_transaction; 2152a4abeea4SJosef Bacik } else { 21530b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2154ccd467d6SChris Mason } 2155a4abeea4SJosef Bacik } else { 21560b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2157cb2d3dadSFilipe Manana /* 2158cb2d3dadSFilipe Manana * The previous transaction was aborted and was already removed 2159cb2d3dadSFilipe Manana * from the list of transactions at fs_info->trans_list. So we 2160cb2d3dadSFilipe Manana * abort to prevent writing a new superblock that reflects a 2161cb2d3dadSFilipe Manana * corrupt state (pointing to trees with unwritten nodes/leafs). 2162cb2d3dadSFilipe Manana */ 2163cb2d3dadSFilipe Manana if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) { 2164cb2d3dadSFilipe Manana ret = -EROFS; 2165cb2d3dadSFilipe Manana goto cleanup_transaction; 2166cb2d3dadSFilipe Manana } 2167ccd467d6SChris Mason } 216815ee9bc7SJosef Bacik 21690860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 21700860adfdSMiao Xie 217188090ad3SFilipe Manana ret = btrfs_start_delalloc_flush(fs_info); 217282436617SMiao Xie if (ret) 217382436617SMiao Xie goto cleanup_transaction; 217482436617SMiao Xie 2175e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 217649b25e05SJeff Mahoney if (ret) 217749b25e05SJeff Mahoney goto cleanup_transaction; 217816cdcec7SMiao Xie 2179581227d0SMiao Xie wait_event(cur_trans->writer_wait, 2180581227d0SMiao Xie extwriter_counter_read(cur_trans) == 0); 2181ed3b3d31SChris Mason 2182581227d0SMiao Xie /* some pending stuffs might be added after the previous flush. */ 2183e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 2184ca469637SMiao Xie if (ret) 2185ca469637SMiao Xie goto cleanup_transaction; 2186ca469637SMiao Xie 218788090ad3SFilipe Manana btrfs_wait_delalloc_flush(fs_info); 2188cb7ab021SWang Shilong 218948778179SFilipe Manana /* 219048778179SFilipe Manana * Wait for all ordered extents started by a fast fsync that joined this 219148778179SFilipe Manana * transaction. Otherwise if this transaction commits before the ordered 219248778179SFilipe Manana * extents complete we lose logged data after a power failure. 219348778179SFilipe Manana */ 219448778179SFilipe Manana wait_event(cur_trans->pending_wait, 219548778179SFilipe Manana atomic_read(&cur_trans->pending_ordered) == 0); 219648778179SFilipe Manana 21972ff7e61eSJeff Mahoney btrfs_scrub_pause(fs_info); 2198ed0ca140SJosef Bacik /* 2199ed0ca140SJosef Bacik * Ok now we need to make sure to block out any other joins while we 2200ed0ca140SJosef Bacik * commit the transaction. We could have started a join before setting 22014a9d8bdeSMiao Xie * COMMIT_DOING so make sure to wait for num_writers to == 1 again. 2202ed0ca140SJosef Bacik */ 22030b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 22044a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 22050b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2206ed0ca140SJosef Bacik wait_event(cur_trans->writer_wait, 2207ed0ca140SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 220815ee9bc7SJosef Bacik 2209bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 22102cba30f1SMiao Xie ret = cur_trans->aborted; 22116cf7f77eSWang Shilong goto scrub_continue; 22122cba30f1SMiao Xie } 22137585717fSChris Mason /* 22147585717fSChris Mason * the reloc mutex makes sure that we stop 22157585717fSChris Mason * the balancing code from coming in and moving 22167585717fSChris Mason * extents around in the middle of the commit 22177585717fSChris Mason */ 22180b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 22197585717fSChris Mason 222042874b3dSMiao Xie /* 222142874b3dSMiao Xie * We needn't worry about the delayed items because we will 222242874b3dSMiao Xie * deal with them in create_pending_snapshot(), which is the 222342874b3dSMiao Xie * core function of the snapshot creation. 222442874b3dSMiao Xie */ 222508d50ca3SNikolay Borisov ret = create_pending_snapshots(trans); 222656e9f6eaSDavid Sterba if (ret) 222756e9f6eaSDavid Sterba goto unlock_reloc; 22283063d29fSChris Mason 222942874b3dSMiao Xie /* 223042874b3dSMiao Xie * We insert the dir indexes of the snapshots and update the inode 223142874b3dSMiao Xie * of the snapshots' parents after the snapshot creation, so there 223242874b3dSMiao Xie * are some delayed items which are not dealt with. Now deal with 223342874b3dSMiao Xie * them. 223442874b3dSMiao Xie * 223542874b3dSMiao Xie * We needn't worry that this operation will corrupt the snapshots, 223642874b3dSMiao Xie * because all the tree which are snapshoted will be forced to COW 223742874b3dSMiao Xie * the nodes and leaves. 223842874b3dSMiao Xie */ 2239e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 224056e9f6eaSDavid Sterba if (ret) 224156e9f6eaSDavid Sterba goto unlock_reloc; 224216cdcec7SMiao Xie 2243c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 224456e9f6eaSDavid Sterba if (ret) 224556e9f6eaSDavid Sterba goto unlock_reloc; 224656bec294SChris Mason 2247e999376fSChris Mason /* 2248e999376fSChris Mason * make sure none of the code above managed to slip in a 2249e999376fSChris Mason * delayed item 2250e999376fSChris Mason */ 2251ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 2252e999376fSChris Mason 22532c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 2254dc17ff8fSChris Mason 2255e02119d5SChris Mason /* btrfs_commit_tree_roots is responsible for getting the 2256e02119d5SChris Mason * various roots consistent with each other. Every pointer 2257e02119d5SChris Mason * in the tree of tree roots has to point to the most up to date 2258e02119d5SChris Mason * root for every subvolume and other tree. So, we have to keep 2259e02119d5SChris Mason * the tree logging code from jumping in and changing any 2260e02119d5SChris Mason * of the trees. 2261e02119d5SChris Mason * 2262e02119d5SChris Mason * At this point in the commit, there can't be any tree-log 2263e02119d5SChris Mason * writers, but a little lower down we drop the trans mutex 2264e02119d5SChris Mason * and let new people in. By holding the tree_log_mutex 2265e02119d5SChris Mason * from now until after the super is written, we avoid races 2266e02119d5SChris Mason * with the tree-log code. 2267e02119d5SChris Mason */ 22680b246afaSJeff Mahoney mutex_lock(&fs_info->tree_log_mutex); 22691a40e23bSZheng Yan 22707e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 227156e9f6eaSDavid Sterba if (ret) 227256e9f6eaSDavid Sterba goto unlock_tree_log; 227354aa1f4dSChris Mason 22743818aea2SQu Wenruo /* 22757e1876acSDavid Sterba * Since the transaction is done, we can apply the pending changes 22767e1876acSDavid Sterba * before the next transaction. 22773818aea2SQu Wenruo */ 22780b246afaSJeff Mahoney btrfs_apply_pending_changes(fs_info); 22793818aea2SQu Wenruo 22805d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 2281e02119d5SChris Mason * safe to free the root of tree log roots 2282e02119d5SChris Mason */ 22830b246afaSJeff Mahoney btrfs_free_log_root_tree(trans, fs_info); 2284e02119d5SChris Mason 22850ed4792aSQu Wenruo /* 22860ed4792aSQu Wenruo * Since fs roots are all committed, we can get a quite accurate 22870ed4792aSQu Wenruo * new_roots. So let's do quota accounting. 22880ed4792aSQu Wenruo */ 2289460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 229056e9f6eaSDavid Sterba if (ret < 0) 229156e9f6eaSDavid Sterba goto unlock_tree_log; 22920ed4792aSQu Wenruo 22939386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 229456e9f6eaSDavid Sterba if (ret) 229556e9f6eaSDavid Sterba goto unlock_tree_log; 229654aa1f4dSChris Mason 22972cba30f1SMiao Xie /* 22982cba30f1SMiao Xie * The tasks which save the space cache and inode cache may also 22992cba30f1SMiao Xie * update ->aborted, check it. 23002cba30f1SMiao Xie */ 2301bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 23022cba30f1SMiao Xie ret = cur_trans->aborted; 230356e9f6eaSDavid Sterba goto unlock_tree_log; 23042cba30f1SMiao Xie } 23052cba30f1SMiao Xie 23060b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 23075f39d397SChris Mason 23080b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->tree_root->root_item, 23090b246afaSJeff Mahoney fs_info->tree_root->node); 23100b246afaSJeff Mahoney list_add_tail(&fs_info->tree_root->dirty_list, 23119e351cc8SJosef Bacik &cur_trans->switch_commits); 23125d4f98a2SYan Zheng 23130b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->chunk_root->root_item, 23140b246afaSJeff Mahoney fs_info->chunk_root->node); 23150b246afaSJeff Mahoney list_add_tail(&fs_info->chunk_root->dirty_list, 23169e351cc8SJosef Bacik &cur_trans->switch_commits); 23179e351cc8SJosef Bacik 2318889bfa39SJosef Bacik switch_commit_roots(trans); 23195d4f98a2SYan Zheng 2320ce93ec54SJosef Bacik ASSERT(list_empty(&cur_trans->dirty_bgs)); 23211bbc621eSChris Mason ASSERT(list_empty(&cur_trans->io_bgs)); 23222ff7e61eSJeff Mahoney update_super_roots(fs_info); 2323e02119d5SChris Mason 23240b246afaSJeff Mahoney btrfs_set_super_log_root(fs_info->super_copy, 0); 23250b246afaSJeff Mahoney btrfs_set_super_log_root_level(fs_info->super_copy, 0); 23260b246afaSJeff Mahoney memcpy(fs_info->super_for_commit, fs_info->super_copy, 23270b246afaSJeff Mahoney sizeof(*fs_info->super_copy)); 2328ccd467d6SChris Mason 2329bbbf7243SNikolay Borisov btrfs_commit_device_sizes(cur_trans); 2330935e5cc9SMiao Xie 23310b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); 23320b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); 2333656f30dbSFilipe Manana 23344fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 23354fbcdf66SFilipe Manana 23360b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 23374a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 23380b246afaSJeff Mahoney fs_info->running_transaction = NULL; 23390b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 23400b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2341b7ec40d7SChris Mason 23420b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 2343e6dcd2dcSChris Mason 234470458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 234549b25e05SJeff Mahoney if (ret) { 23460b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 234708748810SDavid Sterba "Error while writing out transaction"); 234856e9f6eaSDavid Sterba /* 234956e9f6eaSDavid Sterba * reloc_mutex has been unlocked, tree_log_mutex is still held 235056e9f6eaSDavid Sterba * but we can't jump to unlock_tree_log causing double unlock 235156e9f6eaSDavid Sterba */ 23520b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 23536cf7f77eSWang Shilong goto scrub_continue; 235449b25e05SJeff Mahoney } 235549b25e05SJeff Mahoney 2356*d3575156SNaohiro Aota /* 2357*d3575156SNaohiro Aota * At this point, we should have written all the tree blocks allocated 2358*d3575156SNaohiro Aota * in this transaction. So it's now safe to free the redirtyied extent 2359*d3575156SNaohiro Aota * buffers. 2360*d3575156SNaohiro Aota */ 2361*d3575156SNaohiro Aota btrfs_free_redirty_list(cur_trans); 2362*d3575156SNaohiro Aota 2363eece6a9cSDavid Sterba ret = write_all_supers(fs_info, 0); 2364e02119d5SChris Mason /* 2365e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 2366e02119d5SChris Mason * to go about their business 2367e02119d5SChris Mason */ 23680b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 2369c1f32b7cSAnand Jain if (ret) 2370c1f32b7cSAnand Jain goto scrub_continue; 2371e02119d5SChris Mason 2372d0c2f4faSFilipe Manana /* 2373d0c2f4faSFilipe Manana * We needn't acquire the lock here because there is no other task 2374d0c2f4faSFilipe Manana * which can change it. 2375d0c2f4faSFilipe Manana */ 2376d0c2f4faSFilipe Manana cur_trans->state = TRANS_STATE_SUPER_COMMITTED; 2377d0c2f4faSFilipe Manana wake_up(&cur_trans->commit_wait); 2378d0c2f4faSFilipe Manana 23795ead2dd0SNikolay Borisov btrfs_finish_extent_commit(trans); 23804313b399SChris Mason 23813204d33cSJosef Bacik if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) 23820b246afaSJeff Mahoney btrfs_clear_space_info_full(fs_info); 238313212b54SZhao Lei 23840b246afaSJeff Mahoney fs_info->last_trans_committed = cur_trans->transid; 23854a9d8bdeSMiao Xie /* 23864a9d8bdeSMiao Xie * We needn't acquire the lock here because there is no other task 23874a9d8bdeSMiao Xie * which can change it. 23884a9d8bdeSMiao Xie */ 23894a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMPLETED; 23902c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 23913de4586cSChris Mason 23920b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 239313c5a93eSJosef Bacik list_del_init(&cur_trans->list); 23940b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2395a4abeea4SJosef Bacik 2396724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2397724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 239858176a96SJosef Bacik 23990860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 24000b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2401b2b5ef5cSJan Kara 24023a45bb20SJeff Mahoney trace_btrfs_transaction_commit(trans->root); 24031abe9b8aSliubo 24042ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 2405a2de733cSArne Jansen 24069ed74f2dSJosef Bacik if (current->journal_info == trans) 24079ed74f2dSJosef Bacik current->journal_info = NULL; 24089ed74f2dSJosef Bacik 24092c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 241024bbcf04SYan, Zheng 241179154b1bSChris Mason return ret; 241249b25e05SJeff Mahoney 241356e9f6eaSDavid Sterba unlock_tree_log: 241456e9f6eaSDavid Sterba mutex_unlock(&fs_info->tree_log_mutex); 241556e9f6eaSDavid Sterba unlock_reloc: 241656e9f6eaSDavid Sterba mutex_unlock(&fs_info->reloc_mutex); 24176cf7f77eSWang Shilong scrub_continue: 24182ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 241949b25e05SJeff Mahoney cleanup_transaction: 2420dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 2421c7cc64a9SDavid Sterba btrfs_cleanup_pending_block_groups(trans); 24224fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 24230e721106SJosef Bacik trans->block_rsv = NULL; 24240b246afaSJeff Mahoney btrfs_warn(fs_info, "Skipping commit of aborted transaction."); 242549b25e05SJeff Mahoney if (current->journal_info == trans) 242649b25e05SJeff Mahoney current->journal_info = NULL; 242797cb39bbSNikolay Borisov cleanup_transaction(trans, ret); 242849b25e05SJeff Mahoney 242949b25e05SJeff Mahoney return ret; 243079154b1bSChris Mason } 243179154b1bSChris Mason 2432d352ac68SChris Mason /* 24339d1a2a3aSDavid Sterba * return < 0 if error 24349d1a2a3aSDavid Sterba * 0 if there are no more dead_roots at the time of call 24359d1a2a3aSDavid Sterba * 1 there are more to be processed, call me again 24369d1a2a3aSDavid Sterba * 24379d1a2a3aSDavid Sterba * The return value indicates there are certainly more snapshots to delete, but 24389d1a2a3aSDavid Sterba * if there comes a new one during processing, it may return 0. We don't mind, 24399d1a2a3aSDavid Sterba * because btrfs_commit_super will poke cleaner thread and it will process it a 24409d1a2a3aSDavid Sterba * few seconds later. 2441d352ac68SChris Mason */ 24429d1a2a3aSDavid Sterba int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) 2443e9d0b13bSChris Mason { 24449d1a2a3aSDavid Sterba int ret; 24455d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = root->fs_info; 2446e9d0b13bSChris Mason 2447a4abeea4SJosef Bacik spin_lock(&fs_info->trans_lock); 24489d1a2a3aSDavid Sterba if (list_empty(&fs_info->dead_roots)) { 24499d1a2a3aSDavid Sterba spin_unlock(&fs_info->trans_lock); 24509d1a2a3aSDavid Sterba return 0; 24519d1a2a3aSDavid Sterba } 24529d1a2a3aSDavid Sterba root = list_first_entry(&fs_info->dead_roots, 24539d1a2a3aSDavid Sterba struct btrfs_root, root_list); 2454cfad392bSJosef Bacik list_del_init(&root->root_list); 2455a4abeea4SJosef Bacik spin_unlock(&fs_info->trans_lock); 24565d4f98a2SYan Zheng 24574fd786e6SMisono Tomohiro btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); 245876dda93cSYan, Zheng 245916cdcec7SMiao Xie btrfs_kill_all_delayed_nodes(root); 246016cdcec7SMiao Xie 246176dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 246276dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 24630078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 0, 0); 246476dda93cSYan, Zheng else 24650078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 1, 0); 246632471dc2SDavid Sterba 2467dc9492c1SJosef Bacik btrfs_put_root(root); 24686596a928SJosef Bacik return (ret < 0) ? 0 : 1; 2469e9d0b13bSChris Mason } 2470572d9ab7SDavid Sterba 2471572d9ab7SDavid Sterba void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) 2472572d9ab7SDavid Sterba { 2473572d9ab7SDavid Sterba unsigned long prev; 2474572d9ab7SDavid Sterba unsigned long bit; 2475572d9ab7SDavid Sterba 24766c9fe14fSQu Wenruo prev = xchg(&fs_info->pending_changes, 0); 2477572d9ab7SDavid Sterba if (!prev) 2478572d9ab7SDavid Sterba return; 2479572d9ab7SDavid Sterba 2480d51033d0SDavid Sterba bit = 1 << BTRFS_PENDING_COMMIT; 2481d51033d0SDavid Sterba if (prev & bit) 2482d51033d0SDavid Sterba btrfs_debug(fs_info, "pending commit done"); 2483d51033d0SDavid Sterba prev &= ~bit; 2484d51033d0SDavid Sterba 2485572d9ab7SDavid Sterba if (prev) 2486572d9ab7SDavid Sterba btrfs_warn(fs_info, 2487572d9ab7SDavid Sterba "unknown pending changes left 0x%lx, ignoring", prev); 2488572d9ab7SDavid Sterba } 2489