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" 2479154b1bSChris Mason 250f7d52f4SChris Mason #define BTRFS_ROOT_TRANS_TAG 0 260f7d52f4SChris Mason 2761c047b5SQu Wenruo /* 2861c047b5SQu Wenruo * Transaction states and transitions 2961c047b5SQu Wenruo * 3061c047b5SQu Wenruo * No running transaction (fs tree blocks are not modified) 3161c047b5SQu Wenruo * | 3261c047b5SQu Wenruo * | To next stage: 3361c047b5SQu Wenruo * | Call start_transaction() variants. Except btrfs_join_transaction_nostart(). 3461c047b5SQu Wenruo * V 3561c047b5SQu Wenruo * Transaction N [[TRANS_STATE_RUNNING]] 3661c047b5SQu Wenruo * | 3761c047b5SQu Wenruo * | New trans handles can be attached to transaction N by calling all 3861c047b5SQu Wenruo * | start_transaction() variants. 3961c047b5SQu Wenruo * | 4061c047b5SQu Wenruo * | To next stage: 4161c047b5SQu Wenruo * | Call btrfs_commit_transaction() on any trans handle attached to 4261c047b5SQu Wenruo * | transaction N 4361c047b5SQu Wenruo * V 4461c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_START]] 4561c047b5SQu Wenruo * | 4661c047b5SQu Wenruo * | Will wait for previous running transaction to completely finish if there 4761c047b5SQu Wenruo * | is one 4861c047b5SQu Wenruo * | 4961c047b5SQu Wenruo * | Then one of the following happes: 5061c047b5SQu Wenruo * | - Wait for all other trans handle holders to release. 5161c047b5SQu Wenruo * | The btrfs_commit_transaction() caller will do the commit work. 5261c047b5SQu Wenruo * | - Wait for current transaction to be committed by others. 5361c047b5SQu Wenruo * | Other btrfs_commit_transaction() caller will do the commit work. 5461c047b5SQu Wenruo * | 5561c047b5SQu Wenruo * | At this stage, only btrfs_join_transaction*() variants can attach 5661c047b5SQu Wenruo * | to this running transaction. 5761c047b5SQu Wenruo * | All other variants will wait for current one to finish and attach to 5861c047b5SQu Wenruo * | transaction N+1. 5961c047b5SQu Wenruo * | 6061c047b5SQu Wenruo * | To next stage: 6161c047b5SQu Wenruo * | Caller is chosen to commit transaction N, and all other trans handle 6261c047b5SQu Wenruo * | haven been released. 6361c047b5SQu Wenruo * V 6461c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMMIT_DOING]] 6561c047b5SQu Wenruo * | 6661c047b5SQu Wenruo * | The heavy lifting transaction work is started. 6761c047b5SQu Wenruo * | From running delayed refs (modifying extent tree) to creating pending 6861c047b5SQu Wenruo * | snapshots, running qgroups. 6961c047b5SQu Wenruo * | In short, modify supporting trees to reflect modifications of subvolume 7061c047b5SQu Wenruo * | trees. 7161c047b5SQu Wenruo * | 7261c047b5SQu Wenruo * | At this stage, all start_transaction() calls will wait for this 7361c047b5SQu Wenruo * | transaction to finish and attach to transaction N+1. 7461c047b5SQu Wenruo * | 7561c047b5SQu Wenruo * | To next stage: 7661c047b5SQu Wenruo * | Until all supporting trees are updated. 7761c047b5SQu Wenruo * V 7861c047b5SQu Wenruo * Transaction N [[TRANS_STATE_UNBLOCKED]] 7961c047b5SQu Wenruo * | Transaction N+1 8061c047b5SQu Wenruo * | All needed trees are modified, thus we only [[TRANS_STATE_RUNNING]] 8161c047b5SQu Wenruo * | need to write them back to disk and update | 8261c047b5SQu Wenruo * | super blocks. | 8361c047b5SQu Wenruo * | | 8461c047b5SQu Wenruo * | At this stage, new transaction is allowed to | 8561c047b5SQu Wenruo * | start. | 8661c047b5SQu Wenruo * | All new start_transaction() calls will be | 8761c047b5SQu Wenruo * | attached to transid N+1. | 8861c047b5SQu Wenruo * | | 8961c047b5SQu Wenruo * | To next stage: | 9061c047b5SQu Wenruo * | Until all tree blocks are super blocks are | 9161c047b5SQu Wenruo * | written to block devices | 9261c047b5SQu Wenruo * V | 9361c047b5SQu Wenruo * Transaction N [[TRANS_STATE_COMPLETED]] V 9461c047b5SQu Wenruo * All tree blocks and super blocks are written. Transaction N+1 9561c047b5SQu Wenruo * This transaction is finished and all its [[TRANS_STATE_COMMIT_START]] 9661c047b5SQu Wenruo * data structures will be cleaned up. | Life goes on 9761c047b5SQu Wenruo */ 98e8c9f186SDavid Sterba static const unsigned int btrfs_blocked_trans_types[TRANS_STATE_MAX] = { 994a9d8bdeSMiao Xie [TRANS_STATE_RUNNING] = 0U, 100bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_START] = (__TRANS_START | __TRANS_ATTACH), 101bcf3a3e7SNikolay Borisov [TRANS_STATE_COMMIT_DOING] = (__TRANS_START | 1024a9d8bdeSMiao Xie __TRANS_ATTACH | 103a6d155d2SFilipe Manana __TRANS_JOIN | 104a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 105bcf3a3e7SNikolay Borisov [TRANS_STATE_UNBLOCKED] = (__TRANS_START | 1064a9d8bdeSMiao Xie __TRANS_ATTACH | 1074a9d8bdeSMiao Xie __TRANS_JOIN | 108a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 109a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 110*d0c2f4faSFilipe Manana [TRANS_STATE_SUPER_COMMITTED] = (__TRANS_START | 111*d0c2f4faSFilipe Manana __TRANS_ATTACH | 112*d0c2f4faSFilipe Manana __TRANS_JOIN | 113*d0c2f4faSFilipe Manana __TRANS_JOIN_NOLOCK | 114*d0c2f4faSFilipe Manana __TRANS_JOIN_NOSTART), 115bcf3a3e7SNikolay Borisov [TRANS_STATE_COMPLETED] = (__TRANS_START | 1164a9d8bdeSMiao Xie __TRANS_ATTACH | 1174a9d8bdeSMiao Xie __TRANS_JOIN | 118a6d155d2SFilipe Manana __TRANS_JOIN_NOLOCK | 119a6d155d2SFilipe Manana __TRANS_JOIN_NOSTART), 1204a9d8bdeSMiao Xie }; 1214a9d8bdeSMiao Xie 122724e2315SJosef Bacik void btrfs_put_transaction(struct btrfs_transaction *transaction) 12379154b1bSChris Mason { 1249b64f57dSElena Reshetova WARN_ON(refcount_read(&transaction->use_count) == 0); 1259b64f57dSElena Reshetova if (refcount_dec_and_test(&transaction->use_count)) { 126a4abeea4SJosef Bacik BUG_ON(!list_empty(&transaction->list)); 1275c9d028bSLiu Bo WARN_ON(!RB_EMPTY_ROOT( 1285c9d028bSLiu Bo &transaction->delayed_refs.href_root.rb_root)); 12981f7eb00SJeff Mahoney WARN_ON(!RB_EMPTY_ROOT( 13081f7eb00SJeff Mahoney &transaction->delayed_refs.dirty_extent_root)); 1311262133bSJosef Bacik if (transaction->delayed_refs.pending_csums) 132ab8d0fc4SJeff Mahoney btrfs_err(transaction->fs_info, 133ab8d0fc4SJeff Mahoney "pending csums is %llu", 1341262133bSJosef Bacik transaction->delayed_refs.pending_csums); 1357785a663SFilipe Manana /* 1367785a663SFilipe Manana * If any block groups are found in ->deleted_bgs then it's 1377785a663SFilipe Manana * because the transaction was aborted and a commit did not 1387785a663SFilipe Manana * happen (things failed before writing the new superblock 1397785a663SFilipe Manana * and calling btrfs_finish_extent_commit()), so we can not 1407785a663SFilipe Manana * discard the physical locations of the block groups. 1417785a663SFilipe Manana */ 1427785a663SFilipe Manana while (!list_empty(&transaction->deleted_bgs)) { 14332da5386SDavid Sterba struct btrfs_block_group *cache; 1447785a663SFilipe Manana 1457785a663SFilipe Manana cache = list_first_entry(&transaction->deleted_bgs, 14632da5386SDavid Sterba struct btrfs_block_group, 1477785a663SFilipe Manana bg_list); 1487785a663SFilipe Manana list_del_init(&cache->bg_list); 1496b7304afSFilipe Manana btrfs_unfreeze_block_group(cache); 1507785a663SFilipe Manana btrfs_put_block_group(cache); 1517785a663SFilipe Manana } 152bbbf7243SNikolay Borisov WARN_ON(!list_empty(&transaction->dev_update_list)); 1534b5faeacSDavid Sterba kfree(transaction); 15479154b1bSChris Mason } 15578fae27eSChris Mason } 15679154b1bSChris Mason 157889bfa39SJosef Bacik static noinline void switch_commit_roots(struct btrfs_trans_handle *trans) 158817d52f8SJosef Bacik { 159889bfa39SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 16016916a88SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1619e351cc8SJosef Bacik struct btrfs_root *root, *tmp; 16227d56e62SJosef Bacik struct btrfs_caching_control *caching_ctl, *next; 1639e351cc8SJosef Bacik 1649e351cc8SJosef Bacik down_write(&fs_info->commit_root_sem); 165889bfa39SJosef Bacik list_for_each_entry_safe(root, tmp, &cur_trans->switch_commits, 1669e351cc8SJosef Bacik dirty_list) { 1679e351cc8SJosef Bacik list_del_init(&root->dirty_list); 168817d52f8SJosef Bacik free_extent_buffer(root->commit_root); 169817d52f8SJosef Bacik root->commit_root = btrfs_root_node(root); 17041e7acd3SNikolay Borisov extent_io_tree_release(&root->dirty_log_pages); 171370a11b8SQu Wenruo btrfs_qgroup_clean_swapped_blocks(root); 1729e351cc8SJosef Bacik } 1732b9dbef2SJosef Bacik 1742b9dbef2SJosef Bacik /* We can free old roots now. */ 175889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 176889bfa39SJosef Bacik while (!list_empty(&cur_trans->dropped_roots)) { 177889bfa39SJosef Bacik root = list_first_entry(&cur_trans->dropped_roots, 1782b9dbef2SJosef Bacik struct btrfs_root, root_list); 1792b9dbef2SJosef Bacik list_del_init(&root->root_list); 180889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 181889bfa39SJosef Bacik btrfs_free_log(trans, root); 1822b9dbef2SJosef Bacik btrfs_drop_and_free_fs_root(fs_info, root); 183889bfa39SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 1842b9dbef2SJosef Bacik } 185889bfa39SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 18627d56e62SJosef Bacik 18727d56e62SJosef Bacik /* 18827d56e62SJosef Bacik * We have to update the last_byte_to_unpin under the commit_root_sem, 18927d56e62SJosef Bacik * at the same time we swap out the commit roots. 19027d56e62SJosef Bacik * 19127d56e62SJosef Bacik * This is because we must have a real view of the last spot the caching 19227d56e62SJosef Bacik * kthreads were while caching. Consider the following views of the 19327d56e62SJosef Bacik * extent tree for a block group 19427d56e62SJosef Bacik * 19527d56e62SJosef Bacik * commit root 19627d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 19727d56e62SJosef Bacik * |\\\\| |\\\\|\\\\| |\\\\|\\\\| 19827d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 19927d56e62SJosef Bacik * 0 1 2 3 4 5 6 7 20027d56e62SJosef Bacik * 20127d56e62SJosef Bacik * new commit root 20227d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 20327d56e62SJosef Bacik * | | | |\\\\| | |\\\\| 20427d56e62SJosef Bacik * +----+----+----+----+----+----+----+ 20527d56e62SJosef Bacik * 0 1 2 3 4 5 6 7 20627d56e62SJosef Bacik * 20727d56e62SJosef Bacik * If the cache_ctl->progress was at 3, then we are only allowed to 20827d56e62SJosef Bacik * unpin [0,1) and [2,3], because the caching thread has already 20927d56e62SJosef Bacik * processed those extents. We are not allowed to unpin [5,6), because 21027d56e62SJosef Bacik * the caching thread will re-start it's search from 3, and thus find 21127d56e62SJosef Bacik * the hole from [4,6) to add to the free space cache. 21227d56e62SJosef Bacik */ 213bbb86a37SJosef Bacik spin_lock(&fs_info->block_group_cache_lock); 21427d56e62SJosef Bacik list_for_each_entry_safe(caching_ctl, next, 21527d56e62SJosef Bacik &fs_info->caching_block_groups, list) { 21627d56e62SJosef Bacik struct btrfs_block_group *cache = caching_ctl->block_group; 21727d56e62SJosef Bacik 21827d56e62SJosef Bacik if (btrfs_block_group_done(cache)) { 21927d56e62SJosef Bacik cache->last_byte_to_unpin = (u64)-1; 22027d56e62SJosef Bacik list_del_init(&caching_ctl->list); 22127d56e62SJosef Bacik btrfs_put_caching_control(caching_ctl); 22227d56e62SJosef Bacik } else { 22327d56e62SJosef Bacik cache->last_byte_to_unpin = caching_ctl->progress; 22427d56e62SJosef Bacik } 22527d56e62SJosef Bacik } 226bbb86a37SJosef Bacik spin_unlock(&fs_info->block_group_cache_lock); 2279e351cc8SJosef Bacik up_write(&fs_info->commit_root_sem); 228817d52f8SJosef Bacik } 229817d52f8SJosef Bacik 2300860adfdSMiao Xie static inline void extwriter_counter_inc(struct btrfs_transaction *trans, 2310860adfdSMiao Xie unsigned int type) 2320860adfdSMiao Xie { 2330860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2340860adfdSMiao Xie atomic_inc(&trans->num_extwriters); 2350860adfdSMiao Xie } 2360860adfdSMiao Xie 2370860adfdSMiao Xie static inline void extwriter_counter_dec(struct btrfs_transaction *trans, 2380860adfdSMiao Xie unsigned int type) 2390860adfdSMiao Xie { 2400860adfdSMiao Xie if (type & TRANS_EXTWRITERS) 2410860adfdSMiao Xie atomic_dec(&trans->num_extwriters); 2420860adfdSMiao Xie } 2430860adfdSMiao Xie 2440860adfdSMiao Xie static inline void extwriter_counter_init(struct btrfs_transaction *trans, 2450860adfdSMiao Xie unsigned int type) 2460860adfdSMiao Xie { 2470860adfdSMiao Xie atomic_set(&trans->num_extwriters, ((type & TRANS_EXTWRITERS) ? 1 : 0)); 2480860adfdSMiao Xie } 2490860adfdSMiao Xie 2500860adfdSMiao Xie static inline int extwriter_counter_read(struct btrfs_transaction *trans) 2510860adfdSMiao Xie { 2520860adfdSMiao Xie return atomic_read(&trans->num_extwriters); 253178260b2SMiao Xie } 254178260b2SMiao Xie 255d352ac68SChris Mason /* 256fb6dea26SJosef Bacik * To be called after all the new block groups attached to the transaction 257fb6dea26SJosef Bacik * handle have been created (btrfs_create_pending_block_groups()). 258fb6dea26SJosef Bacik */ 259fb6dea26SJosef Bacik void btrfs_trans_release_chunk_metadata(struct btrfs_trans_handle *trans) 260fb6dea26SJosef Bacik { 261fb6dea26SJosef Bacik struct btrfs_fs_info *fs_info = trans->fs_info; 262fb6dea26SJosef Bacik 263fb6dea26SJosef Bacik if (!trans->chunk_bytes_reserved) 264fb6dea26SJosef Bacik return; 265fb6dea26SJosef Bacik 266fb6dea26SJosef Bacik WARN_ON_ONCE(!list_empty(&trans->new_bgs)); 267fb6dea26SJosef Bacik 268fb6dea26SJosef Bacik btrfs_block_rsv_release(fs_info, &fs_info->chunk_block_rsv, 26963f018beSNikolay Borisov trans->chunk_bytes_reserved, NULL); 270fb6dea26SJosef Bacik trans->chunk_bytes_reserved = 0; 271fb6dea26SJosef Bacik } 272fb6dea26SJosef Bacik 273fb6dea26SJosef Bacik /* 274d352ac68SChris Mason * either allocate a new transaction or hop into the existing one 275d352ac68SChris Mason */ 2762ff7e61eSJeff Mahoney static noinline int join_transaction(struct btrfs_fs_info *fs_info, 2772ff7e61eSJeff Mahoney unsigned int type) 27879154b1bSChris Mason { 27979154b1bSChris Mason struct btrfs_transaction *cur_trans; 280a4abeea4SJosef Bacik 28119ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 282d43317dcSChris Mason loop: 28349b25e05SJeff Mahoney /* The file system has been taken offline. No new transactions. */ 28487533c47SMiao Xie if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 28519ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 28649b25e05SJeff Mahoney return -EROFS; 28749b25e05SJeff Mahoney } 28849b25e05SJeff Mahoney 28919ae4e81SJan Schmidt cur_trans = fs_info->running_transaction; 290a4abeea4SJosef Bacik if (cur_trans) { 291bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 29219ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 29349b25e05SJeff Mahoney return cur_trans->aborted; 294871383beSDavid Sterba } 2954a9d8bdeSMiao Xie if (btrfs_blocked_trans_types[cur_trans->state] & type) { 296178260b2SMiao Xie spin_unlock(&fs_info->trans_lock); 297178260b2SMiao Xie return -EBUSY; 298178260b2SMiao Xie } 2999b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 300a4abeea4SJosef Bacik atomic_inc(&cur_trans->num_writers); 3010860adfdSMiao Xie extwriter_counter_inc(cur_trans, type); 30219ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 303a4abeea4SJosef Bacik return 0; 304a4abeea4SJosef Bacik } 30519ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 306a4abeea4SJosef Bacik 307354aa0fbSMiao Xie /* 308354aa0fbSMiao Xie * If we are ATTACH, we just want to catch the current transaction, 309354aa0fbSMiao Xie * and commit it. If there is no transaction, just return ENOENT. 310354aa0fbSMiao Xie */ 311354aa0fbSMiao Xie if (type == TRANS_ATTACH) 312354aa0fbSMiao Xie return -ENOENT; 313354aa0fbSMiao Xie 3144a9d8bdeSMiao Xie /* 3154a9d8bdeSMiao Xie * JOIN_NOLOCK only happens during the transaction commit, so 3164a9d8bdeSMiao Xie * it is impossible that ->running_transaction is NULL 3174a9d8bdeSMiao Xie */ 3184a9d8bdeSMiao Xie BUG_ON(type == TRANS_JOIN_NOLOCK); 3194a9d8bdeSMiao Xie 3204b5faeacSDavid Sterba cur_trans = kmalloc(sizeof(*cur_trans), GFP_NOFS); 321db5b493aSTsutomu Itoh if (!cur_trans) 322db5b493aSTsutomu Itoh return -ENOMEM; 323d43317dcSChris Mason 32419ae4e81SJan Schmidt spin_lock(&fs_info->trans_lock); 32519ae4e81SJan Schmidt if (fs_info->running_transaction) { 326d43317dcSChris Mason /* 327d43317dcSChris Mason * someone started a transaction after we unlocked. Make sure 3284a9d8bdeSMiao Xie * to redo the checks above 329d43317dcSChris Mason */ 3304b5faeacSDavid Sterba kfree(cur_trans); 331d43317dcSChris Mason goto loop; 33287533c47SMiao Xie } else if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) { 333e4b50e14SDan Carpenter spin_unlock(&fs_info->trans_lock); 3344b5faeacSDavid Sterba kfree(cur_trans); 3357b8b92afSJosef Bacik return -EROFS; 336a4abeea4SJosef Bacik } 337d43317dcSChris Mason 338ab8d0fc4SJeff Mahoney cur_trans->fs_info = fs_info; 33948778179SFilipe Manana atomic_set(&cur_trans->pending_ordered, 0); 34048778179SFilipe Manana init_waitqueue_head(&cur_trans->pending_wait); 34113c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 3420860adfdSMiao Xie extwriter_counter_init(cur_trans, type); 34379154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 34479154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 3454a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_RUNNING; 346a4abeea4SJosef Bacik /* 347a4abeea4SJosef Bacik * One for this trans handle, one so it will live on until we 348a4abeea4SJosef Bacik * commit the transaction. 349a4abeea4SJosef Bacik */ 3509b64f57dSElena Reshetova refcount_set(&cur_trans->use_count, 2); 3513204d33cSJosef Bacik cur_trans->flags = 0; 352afd48513SArnd Bergmann cur_trans->start_time = ktime_get_seconds(); 35356bec294SChris Mason 354a099d0fdSAlexandru Moise memset(&cur_trans->delayed_refs, 0, sizeof(cur_trans->delayed_refs)); 355a099d0fdSAlexandru Moise 3565c9d028bSLiu Bo cur_trans->delayed_refs.href_root = RB_ROOT_CACHED; 3573368d001SQu Wenruo cur_trans->delayed_refs.dirty_extent_root = RB_ROOT; 358d7df2c79SJosef Bacik atomic_set(&cur_trans->delayed_refs.num_entries, 0); 35920b297d6SJan Schmidt 36020b297d6SJan Schmidt /* 36120b297d6SJan Schmidt * although the tree mod log is per file system and not per transaction, 36220b297d6SJan Schmidt * the log must never go across transaction boundaries. 36320b297d6SJan Schmidt */ 36420b297d6SJan Schmidt smp_mb(); 36531b1a2bdSJulia Lawall if (!list_empty(&fs_info->tree_mod_seq_list)) 3665d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_seq_list not empty when creating a fresh transaction\n"); 36731b1a2bdSJulia Lawall if (!RB_EMPTY_ROOT(&fs_info->tree_mod_log)) 3685d163e0eSJeff Mahoney WARN(1, KERN_ERR "BTRFS: tree_mod_log rb tree not empty when creating a fresh transaction\n"); 369fc36ed7eSJan Schmidt atomic64_set(&fs_info->tree_mod_seq, 0); 37020b297d6SJan Schmidt 37156bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 37256bec294SChris Mason 3733063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 374bbbf7243SNikolay Borisov INIT_LIST_HEAD(&cur_trans->dev_update_list); 3759e351cc8SJosef Bacik INIT_LIST_HEAD(&cur_trans->switch_commits); 376ce93ec54SJosef Bacik INIT_LIST_HEAD(&cur_trans->dirty_bgs); 3771bbc621eSChris Mason INIT_LIST_HEAD(&cur_trans->io_bgs); 3782b9dbef2SJosef Bacik INIT_LIST_HEAD(&cur_trans->dropped_roots); 3791bbc621eSChris Mason mutex_init(&cur_trans->cache_write_mutex); 380ce93ec54SJosef Bacik spin_lock_init(&cur_trans->dirty_bgs_lock); 381e33e17eeSJeff Mahoney INIT_LIST_HEAD(&cur_trans->deleted_bgs); 3822b9dbef2SJosef Bacik spin_lock_init(&cur_trans->dropped_roots_lock); 38319ae4e81SJan Schmidt list_add_tail(&cur_trans->list, &fs_info->trans_list); 384c258d6e3SQu Wenruo extent_io_tree_init(fs_info, &cur_trans->dirty_pages, 38543eb5f29SQu Wenruo IO_TREE_TRANS_DIRTY_PAGES, fs_info->btree_inode); 386fe119a6eSNikolay Borisov extent_io_tree_init(fs_info, &cur_trans->pinned_extents, 387fe119a6eSNikolay Borisov IO_TREE_FS_PINNED_EXTENTS, NULL); 38819ae4e81SJan Schmidt fs_info->generation++; 38919ae4e81SJan Schmidt cur_trans->transid = fs_info->generation; 39019ae4e81SJan Schmidt fs_info->running_transaction = cur_trans; 39149b25e05SJeff Mahoney cur_trans->aborted = 0; 39219ae4e81SJan Schmidt spin_unlock(&fs_info->trans_lock); 39315ee9bc7SJosef Bacik 39479154b1bSChris Mason return 0; 39579154b1bSChris Mason } 39679154b1bSChris Mason 397d352ac68SChris Mason /* 39892a7cc42SQu Wenruo * This does all the record keeping required to make sure that a shareable root 39992a7cc42SQu Wenruo * is properly recorded in a given transaction. This is required to make sure 40092a7cc42SQu Wenruo * the old root from before we joined the transaction is deleted when the 40192a7cc42SQu Wenruo * transaction commits. 402d352ac68SChris Mason */ 4037585717fSChris Mason static int record_root_in_trans(struct btrfs_trans_handle *trans, 4046426c7adSQu Wenruo struct btrfs_root *root, 4056426c7adSQu Wenruo int force) 4066702ed49SChris Mason { 4070b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4080b246afaSJeff Mahoney 40992a7cc42SQu Wenruo if ((test_bit(BTRFS_ROOT_SHAREABLE, &root->state) && 4106426c7adSQu Wenruo root->last_trans < trans->transid) || force) { 4110b246afaSJeff Mahoney WARN_ON(root == fs_info->extent_root); 4124d31778aSQu Wenruo WARN_ON(!force && root->commit_root != root->node); 4135d4f98a2SYan Zheng 4147585717fSChris Mason /* 41527cdeb70SMiao Xie * see below for IN_TRANS_SETUP usage rules 4167585717fSChris Mason * we have the reloc mutex held now, so there 4177585717fSChris Mason * is only one writer in this function 4187585717fSChris Mason */ 41927cdeb70SMiao Xie set_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4207585717fSChris Mason 42127cdeb70SMiao Xie /* make sure readers find IN_TRANS_SETUP before 4227585717fSChris Mason * they find our root->last_trans update 4237585717fSChris Mason */ 4247585717fSChris Mason smp_wmb(); 4257585717fSChris Mason 4260b246afaSJeff Mahoney spin_lock(&fs_info->fs_roots_radix_lock); 4276426c7adSQu Wenruo if (root->last_trans == trans->transid && !force) { 4280b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 429a4abeea4SJosef Bacik return 0; 430a4abeea4SJosef Bacik } 4310b246afaSJeff Mahoney radix_tree_tag_set(&fs_info->fs_roots_radix, 4326702ed49SChris Mason (unsigned long)root->root_key.objectid, 4336702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 4340b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 4357585717fSChris Mason root->last_trans = trans->transid; 4367585717fSChris Mason 4377585717fSChris Mason /* this is pretty tricky. We don't want to 4387585717fSChris Mason * take the relocation lock in btrfs_record_root_in_trans 4397585717fSChris Mason * unless we're really doing the first setup for this root in 4407585717fSChris Mason * this transaction. 4417585717fSChris Mason * 4427585717fSChris Mason * Normally we'd use root->last_trans as a flag to decide 4437585717fSChris Mason * if we want to take the expensive mutex. 4447585717fSChris Mason * 4457585717fSChris Mason * But, we have to set root->last_trans before we 4467585717fSChris Mason * init the relocation root, otherwise, we trip over warnings 4477585717fSChris Mason * in ctree.c. The solution used here is to flag ourselves 44827cdeb70SMiao Xie * with root IN_TRANS_SETUP. When this is 1, we're still 4497585717fSChris Mason * fixing up the reloc trees and everyone must wait. 4507585717fSChris Mason * 4517585717fSChris Mason * When this is zero, they can trust root->last_trans and fly 4527585717fSChris Mason * through btrfs_record_root_in_trans without having to take the 4537585717fSChris Mason * lock. smp_wmb() makes sure that all the writes above are 4547585717fSChris Mason * done before we pop in the zero below 4557585717fSChris Mason */ 4565d4f98a2SYan Zheng btrfs_init_reloc_root(trans, root); 457c7548af6SChris Mason smp_mb__before_atomic(); 45827cdeb70SMiao Xie clear_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state); 4596702ed49SChris Mason } 4605d4f98a2SYan Zheng return 0; 4616702ed49SChris Mason } 4625d4f98a2SYan Zheng 4637585717fSChris Mason 4642b9dbef2SJosef Bacik void btrfs_add_dropped_root(struct btrfs_trans_handle *trans, 4652b9dbef2SJosef Bacik struct btrfs_root *root) 4662b9dbef2SJosef Bacik { 4670b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4682b9dbef2SJosef Bacik struct btrfs_transaction *cur_trans = trans->transaction; 4692b9dbef2SJosef Bacik 4702b9dbef2SJosef Bacik /* Add ourselves to the transaction dropped list */ 4712b9dbef2SJosef Bacik spin_lock(&cur_trans->dropped_roots_lock); 4722b9dbef2SJosef Bacik list_add_tail(&root->root_list, &cur_trans->dropped_roots); 4732b9dbef2SJosef Bacik spin_unlock(&cur_trans->dropped_roots_lock); 4742b9dbef2SJosef Bacik 4752b9dbef2SJosef Bacik /* Make sure we don't try to update the root at commit time */ 4760b246afaSJeff Mahoney spin_lock(&fs_info->fs_roots_radix_lock); 4770b246afaSJeff Mahoney radix_tree_tag_clear(&fs_info->fs_roots_radix, 4782b9dbef2SJosef Bacik (unsigned long)root->root_key.objectid, 4792b9dbef2SJosef Bacik BTRFS_ROOT_TRANS_TAG); 4800b246afaSJeff Mahoney spin_unlock(&fs_info->fs_roots_radix_lock); 4812b9dbef2SJosef Bacik } 4822b9dbef2SJosef Bacik 4837585717fSChris Mason int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 4847585717fSChris Mason struct btrfs_root *root) 4857585717fSChris Mason { 4860b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 4870b246afaSJeff Mahoney 48892a7cc42SQu Wenruo if (!test_bit(BTRFS_ROOT_SHAREABLE, &root->state)) 4897585717fSChris Mason return 0; 4907585717fSChris Mason 4917585717fSChris Mason /* 49227cdeb70SMiao Xie * see record_root_in_trans for comments about IN_TRANS_SETUP usage 4937585717fSChris Mason * and barriers 4947585717fSChris Mason */ 4957585717fSChris Mason smp_rmb(); 4967585717fSChris Mason if (root->last_trans == trans->transid && 49727cdeb70SMiao Xie !test_bit(BTRFS_ROOT_IN_TRANS_SETUP, &root->state)) 4987585717fSChris Mason return 0; 4997585717fSChris Mason 5000b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 5016426c7adSQu Wenruo record_root_in_trans(trans, root, 0); 5020b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 5037585717fSChris Mason 5047585717fSChris Mason return 0; 5057585717fSChris Mason } 5067585717fSChris Mason 5074a9d8bdeSMiao Xie static inline int is_transaction_blocked(struct btrfs_transaction *trans) 5084a9d8bdeSMiao Xie { 5093296bf56SQu Wenruo return (trans->state >= TRANS_STATE_COMMIT_START && 510501407aaSJosef Bacik trans->state < TRANS_STATE_UNBLOCKED && 511bf31f87fSDavid Sterba !TRANS_ABORTED(trans)); 5124a9d8bdeSMiao Xie } 5134a9d8bdeSMiao Xie 514d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 515d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 516d352ac68SChris Mason * transaction might not be fully on disk. 517d352ac68SChris Mason */ 5182ff7e61eSJeff Mahoney static void wait_current_trans(struct btrfs_fs_info *fs_info) 51979154b1bSChris Mason { 520f9295749SChris Mason struct btrfs_transaction *cur_trans; 52179154b1bSChris Mason 5220b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 5230b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 5244a9d8bdeSMiao Xie if (cur_trans && is_transaction_blocked(cur_trans)) { 5259b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 5260b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 52772d63ed6SLi Zefan 5280b246afaSJeff Mahoney wait_event(fs_info->transaction_wait, 529501407aaSJosef Bacik cur_trans->state >= TRANS_STATE_UNBLOCKED || 530bf31f87fSDavid Sterba TRANS_ABORTED(cur_trans)); 531724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 532a4abeea4SJosef Bacik } else { 5330b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 534f9295749SChris Mason } 53537d1aeeeSChris Mason } 53637d1aeeeSChris Mason 5372ff7e61eSJeff Mahoney static int may_wait_transaction(struct btrfs_fs_info *fs_info, int type) 53837d1aeeeSChris Mason { 5390b246afaSJeff Mahoney if (test_bit(BTRFS_FS_LOG_RECOVERING, &fs_info->flags)) 540a4abeea4SJosef Bacik return 0; 541a4abeea4SJosef Bacik 54292e2f7e3SNikolay Borisov if (type == TRANS_START) 543a4abeea4SJosef Bacik return 1; 544a4abeea4SJosef Bacik 545a22285a6SYan, Zheng return 0; 546a22285a6SYan, Zheng } 547a22285a6SYan, Zheng 54820dd2cbfSMiao Xie static inline bool need_reserve_reloc_root(struct btrfs_root *root) 54920dd2cbfSMiao Xie { 5500b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 5510b246afaSJeff Mahoney 5520b246afaSJeff Mahoney if (!fs_info->reloc_ctl || 55392a7cc42SQu Wenruo !test_bit(BTRFS_ROOT_SHAREABLE, &root->state) || 55420dd2cbfSMiao Xie root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID || 55520dd2cbfSMiao Xie root->reloc_root) 55620dd2cbfSMiao Xie return false; 55720dd2cbfSMiao Xie 55820dd2cbfSMiao Xie return true; 55920dd2cbfSMiao Xie } 56020dd2cbfSMiao Xie 56108e007d2SMiao Xie static struct btrfs_trans_handle * 5625aed1dd8SAlexandru Moise start_transaction(struct btrfs_root *root, unsigned int num_items, 563003d7c59SJeff Mahoney unsigned int type, enum btrfs_reserve_flush_enum flush, 564003d7c59SJeff Mahoney bool enforce_qgroups) 565a22285a6SYan, Zheng { 5660b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 567ba2c4d4eSJosef Bacik struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv; 568a22285a6SYan, Zheng struct btrfs_trans_handle *h; 569a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 570b5009945SJosef Bacik u64 num_bytes = 0; 571c5567237SArne Jansen u64 qgroup_reserved = 0; 57220dd2cbfSMiao Xie bool reloc_reserved = false; 5739c343784SJosef Bacik bool do_chunk_alloc = false; 57420dd2cbfSMiao Xie int ret; 575acce952bSliubo 57646c4e71eSFilipe Manana /* Send isn't supposed to start transactions. */ 5772755a0deSDavid Sterba ASSERT(current->journal_info != BTRFS_SEND_TRANS_STUB); 57846c4e71eSFilipe Manana 5790b246afaSJeff Mahoney if (test_bit(BTRFS_FS_STATE_ERROR, &fs_info->fs_state)) 580acce952bSliubo return ERR_PTR(-EROFS); 5812a1eb461SJosef Bacik 58246c4e71eSFilipe Manana if (current->journal_info) { 5830860adfdSMiao Xie WARN_ON(type & TRANS_EXTWRITERS); 5842a1eb461SJosef Bacik h = current->journal_info; 585b50fff81SDavid Sterba refcount_inc(&h->use_count); 586b50fff81SDavid Sterba WARN_ON(refcount_read(&h->use_count) > 2); 5872a1eb461SJosef Bacik h->orig_rsv = h->block_rsv; 5882a1eb461SJosef Bacik h->block_rsv = NULL; 5892a1eb461SJosef Bacik goto got_it; 5902a1eb461SJosef Bacik } 591b5009945SJosef Bacik 592b5009945SJosef Bacik /* 593b5009945SJosef Bacik * Do the reservation before we join the transaction so we can do all 594b5009945SJosef Bacik * the appropriate flushing if need be. 595b5009945SJosef Bacik */ 596003d7c59SJeff Mahoney if (num_items && root != fs_info->chunk_root) { 597ba2c4d4eSJosef Bacik struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv; 598ba2c4d4eSJosef Bacik u64 delayed_refs_bytes = 0; 599ba2c4d4eSJosef Bacik 6000b246afaSJeff Mahoney qgroup_reserved = num_items * fs_info->nodesize; 601733e03a0SQu Wenruo ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved, 602003d7c59SJeff Mahoney enforce_qgroups); 603c5567237SArne Jansen if (ret) 604c5567237SArne Jansen return ERR_PTR(ret); 605c5567237SArne Jansen 606ba2c4d4eSJosef Bacik /* 607ba2c4d4eSJosef Bacik * We want to reserve all the bytes we may need all at once, so 608ba2c4d4eSJosef Bacik * we only do 1 enospc flushing cycle per transaction start. We 609ba2c4d4eSJosef Bacik * accomplish this by simply assuming we'll do 2 x num_items 610ba2c4d4eSJosef Bacik * worth of delayed refs updates in this trans handle, and 611ba2c4d4eSJosef Bacik * refill that amount for whatever is missing in the reserve. 612ba2c4d4eSJosef Bacik */ 6132bd36e7bSJosef Bacik num_bytes = btrfs_calc_insert_metadata_size(fs_info, num_items); 6147f9fe614SJosef Bacik if (flush == BTRFS_RESERVE_FLUSH_ALL && 6157f9fe614SJosef Bacik delayed_refs_rsv->full == 0) { 616ba2c4d4eSJosef Bacik delayed_refs_bytes = num_bytes; 617ba2c4d4eSJosef Bacik num_bytes <<= 1; 618ba2c4d4eSJosef Bacik } 619ba2c4d4eSJosef Bacik 62020dd2cbfSMiao Xie /* 62120dd2cbfSMiao Xie * Do the reservation for the relocation root creation 62220dd2cbfSMiao Xie */ 623ee39b432SDavid Sterba if (need_reserve_reloc_root(root)) { 6240b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 62520dd2cbfSMiao Xie reloc_reserved = true; 62620dd2cbfSMiao Xie } 62720dd2cbfSMiao Xie 628ba2c4d4eSJosef Bacik ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush); 629ba2c4d4eSJosef Bacik if (ret) 630ba2c4d4eSJosef Bacik goto reserve_fail; 631ba2c4d4eSJosef Bacik if (delayed_refs_bytes) { 632ba2c4d4eSJosef Bacik btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv, 633ba2c4d4eSJosef Bacik delayed_refs_bytes); 634ba2c4d4eSJosef Bacik num_bytes -= delayed_refs_bytes; 635ba2c4d4eSJosef Bacik } 6369c343784SJosef Bacik 6379c343784SJosef Bacik if (rsv->space_info->force_alloc) 6389c343784SJosef Bacik do_chunk_alloc = true; 639ba2c4d4eSJosef Bacik } else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL && 640ba2c4d4eSJosef Bacik !delayed_refs_rsv->full) { 641ba2c4d4eSJosef Bacik /* 642ba2c4d4eSJosef Bacik * Some people call with btrfs_start_transaction(root, 0) 643ba2c4d4eSJosef Bacik * because they can be throttled, but have some other mechanism 644ba2c4d4eSJosef Bacik * for reserving space. We still want these guys to refill the 645ba2c4d4eSJosef Bacik * delayed block_rsv so just add 1 items worth of reservation 646ba2c4d4eSJosef Bacik * here. 647ba2c4d4eSJosef Bacik */ 648ba2c4d4eSJosef Bacik ret = btrfs_delayed_refs_rsv_refill(fs_info, flush); 649b5009945SJosef Bacik if (ret) 650843fcf35SMiao Xie goto reserve_fail; 651b5009945SJosef Bacik } 652a22285a6SYan, Zheng again: 653f2f767e7SAlexandru Moise h = kmem_cache_zalloc(btrfs_trans_handle_cachep, GFP_NOFS); 654843fcf35SMiao Xie if (!h) { 655843fcf35SMiao Xie ret = -ENOMEM; 656843fcf35SMiao Xie goto alloc_fail; 657843fcf35SMiao Xie } 658a22285a6SYan, Zheng 65998114659SJosef Bacik /* 66098114659SJosef Bacik * If we are JOIN_NOLOCK we're already committing a transaction and 66198114659SJosef Bacik * waiting on this guy, so we don't need to do the sb_start_intwrite 66298114659SJosef Bacik * because we're already holding a ref. We need this because we could 66398114659SJosef Bacik * have raced in and did an fsync() on a file which can kick a commit 66498114659SJosef Bacik * and then we deadlock with somebody doing a freeze. 665354aa0fbSMiao Xie * 666354aa0fbSMiao Xie * If we are ATTACH, it means we just want to catch the current 667354aa0fbSMiao Xie * transaction and commit it, so we needn't do sb_start_intwrite(). 66898114659SJosef Bacik */ 6690860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 6700b246afaSJeff Mahoney sb_start_intwrite(fs_info->sb); 671b2b5ef5cSJan Kara 6722ff7e61eSJeff Mahoney if (may_wait_transaction(fs_info, type)) 6732ff7e61eSJeff Mahoney wait_current_trans(fs_info); 674a22285a6SYan, Zheng 675a4abeea4SJosef Bacik do { 6762ff7e61eSJeff Mahoney ret = join_transaction(fs_info, type); 677178260b2SMiao Xie if (ret == -EBUSY) { 6782ff7e61eSJeff Mahoney wait_current_trans(fs_info); 679a6d155d2SFilipe Manana if (unlikely(type == TRANS_ATTACH || 680a6d155d2SFilipe Manana type == TRANS_JOIN_NOSTART)) 681178260b2SMiao Xie ret = -ENOENT; 682178260b2SMiao Xie } 683a4abeea4SJosef Bacik } while (ret == -EBUSY); 684a4abeea4SJosef Bacik 685a43f7f82SLiu Bo if (ret < 0) 686843fcf35SMiao Xie goto join_fail; 6870f7d52f4SChris Mason 6880b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 689a22285a6SYan, Zheng 690a22285a6SYan, Zheng h->transid = cur_trans->transid; 691a22285a6SYan, Zheng h->transaction = cur_trans; 692d13603efSArne Jansen h->root = root; 693b50fff81SDavid Sterba refcount_set(&h->use_count, 1); 69464b63580SJeff Mahoney h->fs_info = root->fs_info; 6957174109cSQu Wenruo 696a698d075SMiao Xie h->type = type; 697d9a0540aSFilipe Manana h->can_flush_pending_bgs = true; 698ea658badSJosef Bacik INIT_LIST_HEAD(&h->new_bgs); 699b7ec40d7SChris Mason 700a22285a6SYan, Zheng smp_mb(); 7013296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START && 7022ff7e61eSJeff Mahoney may_wait_transaction(fs_info, type)) { 703abdd2e80SFilipe Manana current->journal_info = h; 7043a45bb20SJeff Mahoney btrfs_commit_transaction(h); 705a22285a6SYan, Zheng goto again; 706a22285a6SYan, Zheng } 7079ed74f2dSJosef Bacik 708b5009945SJosef Bacik if (num_bytes) { 7090b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 7102bcc0328SLiu Bo h->transid, num_bytes, 1); 7110b246afaSJeff Mahoney h->block_rsv = &fs_info->trans_block_rsv; 712b5009945SJosef Bacik h->bytes_reserved = num_bytes; 71320dd2cbfSMiao Xie h->reloc_reserved = reloc_reserved; 714a22285a6SYan, Zheng } 715a22285a6SYan, Zheng 7162a1eb461SJosef Bacik got_it: 717bcf3a3e7SNikolay Borisov if (!current->journal_info) 718a22285a6SYan, Zheng current->journal_info = h; 719fcc99734SQu Wenruo 720fcc99734SQu Wenruo /* 7219c343784SJosef Bacik * If the space_info is marked ALLOC_FORCE then we'll get upgraded to 7229c343784SJosef Bacik * ALLOC_FORCE the first run through, and then we won't allocate for 7239c343784SJosef Bacik * anybody else who races in later. We don't care about the return 7249c343784SJosef Bacik * value here. 7259c343784SJosef Bacik */ 7269c343784SJosef Bacik if (do_chunk_alloc && num_bytes) { 7279c343784SJosef Bacik u64 flags = h->block_rsv->space_info->flags; 7289c343784SJosef Bacik 7299c343784SJosef Bacik btrfs_chunk_alloc(h, btrfs_get_alloc_profile(fs_info, flags), 7309c343784SJosef Bacik CHUNK_ALLOC_NO_FORCE); 7319c343784SJosef Bacik } 7329c343784SJosef Bacik 7339c343784SJosef Bacik /* 734fcc99734SQu Wenruo * btrfs_record_root_in_trans() needs to alloc new extents, and may 735fcc99734SQu Wenruo * call btrfs_join_transaction() while we're also starting a 736fcc99734SQu Wenruo * transaction. 737fcc99734SQu Wenruo * 738fcc99734SQu Wenruo * Thus it need to be called after current->journal_info initialized, 739fcc99734SQu Wenruo * or we can deadlock. 740fcc99734SQu Wenruo */ 741fcc99734SQu Wenruo btrfs_record_root_in_trans(h, root); 742fcc99734SQu Wenruo 74379154b1bSChris Mason return h; 744843fcf35SMiao Xie 745843fcf35SMiao Xie join_fail: 7460860adfdSMiao Xie if (type & __TRANS_FREEZABLE) 7470b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 748843fcf35SMiao Xie kmem_cache_free(btrfs_trans_handle_cachep, h); 749843fcf35SMiao Xie alloc_fail: 750843fcf35SMiao Xie if (num_bytes) 7512ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, &fs_info->trans_block_rsv, 75263f018beSNikolay Borisov num_bytes, NULL); 753843fcf35SMiao Xie reserve_fail: 754733e03a0SQu Wenruo btrfs_qgroup_free_meta_pertrans(root, qgroup_reserved); 755843fcf35SMiao Xie return ERR_PTR(ret); 75679154b1bSChris Mason } 75779154b1bSChris Mason 758f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 7595aed1dd8SAlexandru Moise unsigned int num_items) 760f9295749SChris Mason { 76108e007d2SMiao Xie return start_transaction(root, num_items, TRANS_START, 762003d7c59SJeff Mahoney BTRFS_RESERVE_FLUSH_ALL, true); 763f9295749SChris Mason } 764003d7c59SJeff Mahoney 7658eab77ffSFilipe Manana struct btrfs_trans_handle *btrfs_start_transaction_fallback_global_rsv( 7668eab77ffSFilipe Manana struct btrfs_root *root, 7677f9fe614SJosef Bacik unsigned int num_items) 7688eab77ffSFilipe Manana { 7697f9fe614SJosef Bacik return start_transaction(root, num_items, TRANS_START, 7707f9fe614SJosef Bacik BTRFS_RESERVE_FLUSH_ALL_STEAL, false); 7718eab77ffSFilipe Manana } 7728407aa46SMiao Xie 7737a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) 774f9295749SChris Mason { 775003d7c59SJeff Mahoney return start_transaction(root, 0, TRANS_JOIN, BTRFS_RESERVE_NO_FLUSH, 776003d7c59SJeff Mahoney true); 777f9295749SChris Mason } 778f9295749SChris Mason 7798d510121SNikolay Borisov struct btrfs_trans_handle *btrfs_join_transaction_spacecache(struct btrfs_root *root) 7800af3d00bSJosef Bacik { 781575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_JOIN_NOLOCK, 782003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 7830af3d00bSJosef Bacik } 7840af3d00bSJosef Bacik 785d4edf39bSMiao Xie /* 786a6d155d2SFilipe Manana * Similar to regular join but it never starts a transaction when none is 787a6d155d2SFilipe Manana * running or after waiting for the current one to finish. 788a6d155d2SFilipe Manana */ 789a6d155d2SFilipe Manana struct btrfs_trans_handle *btrfs_join_transaction_nostart(struct btrfs_root *root) 790a6d155d2SFilipe Manana { 791a6d155d2SFilipe Manana return start_transaction(root, 0, TRANS_JOIN_NOSTART, 792a6d155d2SFilipe Manana BTRFS_RESERVE_NO_FLUSH, true); 793a6d155d2SFilipe Manana } 794a6d155d2SFilipe Manana 795a6d155d2SFilipe Manana /* 796d4edf39bSMiao Xie * btrfs_attach_transaction() - catch the running transaction 797d4edf39bSMiao Xie * 798d4edf39bSMiao Xie * It is used when we want to commit the current the transaction, but 799d4edf39bSMiao Xie * don't want to start a new one. 800d4edf39bSMiao Xie * 801d4edf39bSMiao Xie * Note: If this function return -ENOENT, it just means there is no 802d4edf39bSMiao Xie * running transaction. But it is possible that the inactive transaction 803d4edf39bSMiao Xie * is still in the memory, not fully on disk. If you hope there is no 804d4edf39bSMiao Xie * inactive transaction in the fs when -ENOENT is returned, you should 805d4edf39bSMiao Xie * invoke 806d4edf39bSMiao Xie * btrfs_attach_transaction_barrier() 807d4edf39bSMiao Xie */ 808354aa0fbSMiao Xie struct btrfs_trans_handle *btrfs_attach_transaction(struct btrfs_root *root) 80960376ce4SJosef Bacik { 810575a75d6SAlexandru Moise return start_transaction(root, 0, TRANS_ATTACH, 811003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 81260376ce4SJosef Bacik } 81360376ce4SJosef Bacik 814d4edf39bSMiao Xie /* 81590b6d283SWang Sheng-Hui * btrfs_attach_transaction_barrier() - catch the running transaction 816d4edf39bSMiao Xie * 81752042d8eSAndrea Gelmini * It is similar to the above function, the difference is this one 818d4edf39bSMiao Xie * will wait for all the inactive transactions until they fully 819d4edf39bSMiao Xie * complete. 820d4edf39bSMiao Xie */ 821d4edf39bSMiao Xie struct btrfs_trans_handle * 822d4edf39bSMiao Xie btrfs_attach_transaction_barrier(struct btrfs_root *root) 823d4edf39bSMiao Xie { 824d4edf39bSMiao Xie struct btrfs_trans_handle *trans; 825d4edf39bSMiao Xie 826575a75d6SAlexandru Moise trans = start_transaction(root, 0, TRANS_ATTACH, 827003d7c59SJeff Mahoney BTRFS_RESERVE_NO_FLUSH, true); 8288d9e220cSAl Viro if (trans == ERR_PTR(-ENOENT)) 8292ff7e61eSJeff Mahoney btrfs_wait_for_commit(root->fs_info, 0); 830d4edf39bSMiao Xie 831d4edf39bSMiao Xie return trans; 832d4edf39bSMiao Xie } 833d4edf39bSMiao Xie 834*d0c2f4faSFilipe Manana /* Wait for a transaction commit to reach at least the given state. */ 835*d0c2f4faSFilipe Manana static noinline void wait_for_commit(struct btrfs_transaction *commit, 836*d0c2f4faSFilipe Manana const enum btrfs_trans_state min_state) 83789ce8a63SChris Mason { 838*d0c2f4faSFilipe Manana wait_event(commit->commit_wait, commit->state >= min_state); 83989ce8a63SChris Mason } 84089ce8a63SChris Mason 8412ff7e61eSJeff Mahoney int btrfs_wait_for_commit(struct btrfs_fs_info *fs_info, u64 transid) 84246204592SSage Weil { 84346204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 8448cd2807fSMiao Xie int ret = 0; 84546204592SSage Weil 84646204592SSage Weil if (transid) { 8470b246afaSJeff Mahoney if (transid <= fs_info->last_trans_committed) 848a4abeea4SJosef Bacik goto out; 84946204592SSage Weil 85046204592SSage Weil /* find specified transaction */ 8510b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8520b246afaSJeff Mahoney list_for_each_entry(t, &fs_info->trans_list, list) { 85346204592SSage Weil if (t->transid == transid) { 85446204592SSage Weil cur_trans = t; 8559b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 8568cd2807fSMiao Xie ret = 0; 85746204592SSage Weil break; 85846204592SSage Weil } 8598cd2807fSMiao Xie if (t->transid > transid) { 8608cd2807fSMiao Xie ret = 0; 86146204592SSage Weil break; 86246204592SSage Weil } 8638cd2807fSMiao Xie } 8640b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 86542383020SSage Weil 86642383020SSage Weil /* 86742383020SSage Weil * The specified transaction doesn't exist, or we 86842383020SSage Weil * raced with btrfs_commit_transaction 86942383020SSage Weil */ 87042383020SSage Weil if (!cur_trans) { 8710b246afaSJeff Mahoney if (transid > fs_info->last_trans_committed) 87242383020SSage Weil ret = -EINVAL; 8738cd2807fSMiao Xie goto out; 87442383020SSage Weil } 87546204592SSage Weil } else { 87646204592SSage Weil /* find newest transaction that is committing | committed */ 8770b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 8780b246afaSJeff Mahoney list_for_each_entry_reverse(t, &fs_info->trans_list, 87946204592SSage Weil list) { 8804a9d8bdeSMiao Xie if (t->state >= TRANS_STATE_COMMIT_START) { 8814a9d8bdeSMiao Xie if (t->state == TRANS_STATE_COMPLETED) 8823473f3c0SJosef Bacik break; 88346204592SSage Weil cur_trans = t; 8849b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 88546204592SSage Weil break; 88646204592SSage Weil } 88746204592SSage Weil } 8880b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 88946204592SSage Weil if (!cur_trans) 890a4abeea4SJosef Bacik goto out; /* nothing committing|committed */ 89146204592SSage Weil } 89246204592SSage Weil 893*d0c2f4faSFilipe Manana wait_for_commit(cur_trans, TRANS_STATE_COMPLETED); 894724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 895a4abeea4SJosef Bacik out: 89646204592SSage Weil return ret; 89746204592SSage Weil } 89846204592SSage Weil 8992ff7e61eSJeff Mahoney void btrfs_throttle(struct btrfs_fs_info *fs_info) 90037d1aeeeSChris Mason { 9012ff7e61eSJeff Mahoney wait_current_trans(fs_info); 90237d1aeeeSChris Mason } 90337d1aeeeSChris Mason 9048a8f4deaSNikolay Borisov static bool should_end_transaction(struct btrfs_trans_handle *trans) 9058929ecfaSYan, Zheng { 9062ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 9070b246afaSJeff Mahoney 90864403612SJosef Bacik if (btrfs_check_space_for_delayed_refs(fs_info)) 9098a8f4deaSNikolay Borisov return true; 91036ba022aSJosef Bacik 9112ff7e61eSJeff Mahoney return !!btrfs_block_rsv_check(&fs_info->global_block_rsv, 5); 9128929ecfaSYan, Zheng } 9138929ecfaSYan, Zheng 914a2633b6aSNikolay Borisov bool btrfs_should_end_transaction(struct btrfs_trans_handle *trans) 9158929ecfaSYan, Zheng { 9168929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9178929ecfaSYan, Zheng 9183296bf56SQu Wenruo if (cur_trans->state >= TRANS_STATE_COMMIT_START || 919e19eb11fSJosef Bacik test_bit(BTRFS_DELAYED_REFS_FLUSHING, &cur_trans->delayed_refs.flags)) 920a2633b6aSNikolay Borisov return true; 9218929ecfaSYan, Zheng 9222ff7e61eSJeff Mahoney return should_end_transaction(trans); 9238929ecfaSYan, Zheng } 9248929ecfaSYan, Zheng 925dc60c525SNikolay Borisov static void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans) 926dc60c525SNikolay Borisov 9270e34693fSNikolay Borisov { 928dc60c525SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 929dc60c525SNikolay Borisov 9300e34693fSNikolay Borisov if (!trans->block_rsv) { 9310e34693fSNikolay Borisov ASSERT(!trans->bytes_reserved); 9320e34693fSNikolay Borisov return; 9330e34693fSNikolay Borisov } 9340e34693fSNikolay Borisov 9350e34693fSNikolay Borisov if (!trans->bytes_reserved) 9360e34693fSNikolay Borisov return; 9370e34693fSNikolay Borisov 9380e34693fSNikolay Borisov ASSERT(trans->block_rsv == &fs_info->trans_block_rsv); 9390e34693fSNikolay Borisov trace_btrfs_space_reservation(fs_info, "transaction", 9400e34693fSNikolay Borisov trans->transid, trans->bytes_reserved, 0); 9410e34693fSNikolay Borisov btrfs_block_rsv_release(fs_info, trans->block_rsv, 94263f018beSNikolay Borisov trans->bytes_reserved, NULL); 9430e34693fSNikolay Borisov trans->bytes_reserved = 0; 9440e34693fSNikolay Borisov } 9450e34693fSNikolay Borisov 94689ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 9473a45bb20SJeff Mahoney int throttle) 94879154b1bSChris Mason { 9493a45bb20SJeff Mahoney struct btrfs_fs_info *info = trans->fs_info; 9508929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 9514edc2ca3SDave Jones int err = 0; 952d6e4a428SChris Mason 953b50fff81SDavid Sterba if (refcount_read(&trans->use_count) > 1) { 954b50fff81SDavid Sterba refcount_dec(&trans->use_count); 9552a1eb461SJosef Bacik trans->block_rsv = trans->orig_rsv; 9562a1eb461SJosef Bacik return 0; 9572a1eb461SJosef Bacik } 9582a1eb461SJosef Bacik 959dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 9604c13d758SJosef Bacik trans->block_rsv = NULL; 961c5567237SArne Jansen 9626c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 963ea658badSJosef Bacik 9644fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 9654fbcdf66SFilipe Manana 9660860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 9670b246afaSJeff Mahoney sb_end_intwrite(info->sb); 9686df7881aSJosef Bacik 9698929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 97013c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 97113c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 9720860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 97389ce8a63SChris Mason 974093258e6SDavid Sterba cond_wake_up(&cur_trans->writer_wait); 975724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 9769ed74f2dSJosef Bacik 9779ed74f2dSJosef Bacik if (current->journal_info == trans) 9789ed74f2dSJosef Bacik current->journal_info = NULL; 979ab78c84dSChris Mason 98024bbcf04SYan, Zheng if (throttle) 9812ff7e61eSJeff Mahoney btrfs_run_delayed_iputs(info); 98224bbcf04SYan, Zheng 983bf31f87fSDavid Sterba if (TRANS_ABORTED(trans) || 9840b246afaSJeff Mahoney test_bit(BTRFS_FS_STATE_ERROR, &info->fs_state)) { 9854e121c06SJosef Bacik wake_up_process(info->transaction_kthread); 986fbabd4a3SJosef Bacik if (TRANS_ABORTED(trans)) 987fbabd4a3SJosef Bacik err = trans->aborted; 988fbabd4a3SJosef Bacik else 989fbabd4a3SJosef Bacik err = -EROFS; 9904e121c06SJosef Bacik } 99149b25e05SJeff Mahoney 9924edc2ca3SDave Jones kmem_cache_free(btrfs_trans_handle_cachep, trans); 9934edc2ca3SDave Jones return err; 99479154b1bSChris Mason } 99579154b1bSChris Mason 9963a45bb20SJeff Mahoney int btrfs_end_transaction(struct btrfs_trans_handle *trans) 99789ce8a63SChris Mason { 9983a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 0); 99989ce8a63SChris Mason } 100089ce8a63SChris Mason 10013a45bb20SJeff Mahoney int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans) 100289ce8a63SChris Mason { 10033a45bb20SJeff Mahoney return __btrfs_end_transaction(trans, 1); 100416cdcec7SMiao Xie } 100516cdcec7SMiao Xie 1006d352ac68SChris Mason /* 1007d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1008d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1009690587d1SChris Mason * those extents are sent to disk but does not wait on them 1010d352ac68SChris Mason */ 10112ff7e61eSJeff Mahoney int btrfs_write_marked_extents(struct btrfs_fs_info *fs_info, 10128cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 101379154b1bSChris Mason { 1014777e6bd7SChris Mason int err = 0; 10157c4452b9SChris Mason int werr = 0; 10160b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1017e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1018777e6bd7SChris Mason u64 start = 0; 10195f39d397SChris Mason u64 end; 10207c4452b9SChris Mason 10216300463bSLiu Bo atomic_inc(&BTRFS_I(fs_info->btree_inode)->sync_writers); 10221728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1023e6138876SJosef Bacik mark, &cached_state)) { 1024663dfbb0SFilipe Manana bool wait_writeback = false; 1025663dfbb0SFilipe Manana 1026663dfbb0SFilipe Manana err = convert_extent_bit(dirty_pages, start, end, 1027663dfbb0SFilipe Manana EXTENT_NEED_WAIT, 1028210aa277SDavid Sterba mark, &cached_state); 1029663dfbb0SFilipe Manana /* 1030663dfbb0SFilipe Manana * convert_extent_bit can return -ENOMEM, which is most of the 1031663dfbb0SFilipe Manana * time a temporary error. So when it happens, ignore the error 1032663dfbb0SFilipe Manana * and wait for writeback of this range to finish - because we 1033663dfbb0SFilipe Manana * failed to set the bit EXTENT_NEED_WAIT for the range, a call 1034bf89d38fSJeff Mahoney * to __btrfs_wait_marked_extents() would not know that 1035bf89d38fSJeff Mahoney * writeback for this range started and therefore wouldn't 1036bf89d38fSJeff Mahoney * wait for it to finish - we don't want to commit a 1037bf89d38fSJeff Mahoney * superblock that points to btree nodes/leafs for which 1038bf89d38fSJeff Mahoney * writeback hasn't finished yet (and without errors). 1039663dfbb0SFilipe Manana * We cleanup any entries left in the io tree when committing 104041e7acd3SNikolay Borisov * the transaction (through extent_io_tree_release()). 1041663dfbb0SFilipe Manana */ 1042663dfbb0SFilipe Manana if (err == -ENOMEM) { 1043663dfbb0SFilipe Manana err = 0; 1044663dfbb0SFilipe Manana wait_writeback = true; 1045663dfbb0SFilipe Manana } 1046663dfbb0SFilipe Manana if (!err) 10471728366eSJosef Bacik err = filemap_fdatawrite_range(mapping, start, end); 10487c4452b9SChris Mason if (err) 10497c4452b9SChris Mason werr = err; 1050663dfbb0SFilipe Manana else if (wait_writeback) 1051663dfbb0SFilipe Manana werr = filemap_fdatawait_range(mapping, start, end); 1052e38e2ed7SFilipe Manana free_extent_state(cached_state); 1053663dfbb0SFilipe Manana cached_state = NULL; 10541728366eSJosef Bacik cond_resched(); 10551728366eSJosef Bacik start = end + 1; 10567c4452b9SChris Mason } 10576300463bSLiu Bo atomic_dec(&BTRFS_I(fs_info->btree_inode)->sync_writers); 1058690587d1SChris Mason return werr; 1059690587d1SChris Mason } 1060690587d1SChris Mason 1061690587d1SChris Mason /* 1062690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 1063690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 1064690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 1065690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 1066690587d1SChris Mason */ 1067bf89d38fSJeff Mahoney static int __btrfs_wait_marked_extents(struct btrfs_fs_info *fs_info, 1068bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1069690587d1SChris Mason { 1070690587d1SChris Mason int err = 0; 1071690587d1SChris Mason int werr = 0; 10720b246afaSJeff Mahoney struct address_space *mapping = fs_info->btree_inode->i_mapping; 1073e6138876SJosef Bacik struct extent_state *cached_state = NULL; 1074690587d1SChris Mason u64 start = 0; 1075690587d1SChris Mason u64 end; 1076690587d1SChris Mason 10771728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 1078e6138876SJosef Bacik EXTENT_NEED_WAIT, &cached_state)) { 1079663dfbb0SFilipe Manana /* 1080663dfbb0SFilipe Manana * Ignore -ENOMEM errors returned by clear_extent_bit(). 1081663dfbb0SFilipe Manana * When committing the transaction, we'll remove any entries 1082663dfbb0SFilipe Manana * left in the io tree. For a log commit, we don't remove them 1083663dfbb0SFilipe Manana * after committing the log because the tree can be accessed 1084663dfbb0SFilipe Manana * concurrently - we do it only at transaction commit time when 108541e7acd3SNikolay Borisov * it's safe to do it (through extent_io_tree_release()). 1086663dfbb0SFilipe Manana */ 1087663dfbb0SFilipe Manana err = clear_extent_bit(dirty_pages, start, end, 1088ae0f1625SDavid Sterba EXTENT_NEED_WAIT, 0, 0, &cached_state); 1089663dfbb0SFilipe Manana if (err == -ENOMEM) 1090663dfbb0SFilipe Manana err = 0; 1091663dfbb0SFilipe Manana if (!err) 10921728366eSJosef Bacik err = filemap_fdatawait_range(mapping, start, end); 1093777e6bd7SChris Mason if (err) 1094777e6bd7SChris Mason werr = err; 1095e38e2ed7SFilipe Manana free_extent_state(cached_state); 1096e38e2ed7SFilipe Manana cached_state = NULL; 1097777e6bd7SChris Mason cond_resched(); 10981728366eSJosef Bacik start = end + 1; 1099777e6bd7SChris Mason } 11007c4452b9SChris Mason if (err) 11017c4452b9SChris Mason werr = err; 1102bf89d38fSJeff Mahoney return werr; 1103bf89d38fSJeff Mahoney } 1104656f30dbSFilipe Manana 1105b9fae2ebSFilipe Manana static int btrfs_wait_extents(struct btrfs_fs_info *fs_info, 1106bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages) 1107bf89d38fSJeff Mahoney { 1108bf89d38fSJeff Mahoney bool errors = false; 1109bf89d38fSJeff Mahoney int err; 1110bf89d38fSJeff Mahoney 1111bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1112bf89d38fSJeff Mahoney if (test_and_clear_bit(BTRFS_FS_BTREE_ERR, &fs_info->flags)) 1113bf89d38fSJeff Mahoney errors = true; 1114bf89d38fSJeff Mahoney 1115bf89d38fSJeff Mahoney if (errors && !err) 1116bf89d38fSJeff Mahoney err = -EIO; 1117bf89d38fSJeff Mahoney return err; 1118bf89d38fSJeff Mahoney } 1119bf89d38fSJeff Mahoney 1120bf89d38fSJeff Mahoney int btrfs_wait_tree_log_extents(struct btrfs_root *log_root, int mark) 1121bf89d38fSJeff Mahoney { 1122bf89d38fSJeff Mahoney struct btrfs_fs_info *fs_info = log_root->fs_info; 1123bf89d38fSJeff Mahoney struct extent_io_tree *dirty_pages = &log_root->dirty_log_pages; 1124bf89d38fSJeff Mahoney bool errors = false; 1125bf89d38fSJeff Mahoney int err; 1126bf89d38fSJeff Mahoney 1127bf89d38fSJeff Mahoney ASSERT(log_root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID); 1128bf89d38fSJeff Mahoney 1129bf89d38fSJeff Mahoney err = __btrfs_wait_marked_extents(fs_info, dirty_pages); 1130656f30dbSFilipe Manana if ((mark & EXTENT_DIRTY) && 11310b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags)) 1132656f30dbSFilipe Manana errors = true; 1133656f30dbSFilipe Manana 1134656f30dbSFilipe Manana if ((mark & EXTENT_NEW) && 11350b246afaSJeff Mahoney test_and_clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags)) 1136656f30dbSFilipe Manana errors = true; 1137656f30dbSFilipe Manana 1138bf89d38fSJeff Mahoney if (errors && !err) 1139bf89d38fSJeff Mahoney err = -EIO; 1140bf89d38fSJeff Mahoney return err; 114179154b1bSChris Mason } 114279154b1bSChris Mason 1143690587d1SChris Mason /* 1144c9b577c0SNikolay Borisov * When btree blocks are allocated the corresponding extents are marked dirty. 1145c9b577c0SNikolay Borisov * This function ensures such extents are persisted on disk for transaction or 1146c9b577c0SNikolay Borisov * log commit. 1147c9b577c0SNikolay Borisov * 1148c9b577c0SNikolay Borisov * @trans: transaction whose dirty pages we'd like to write 1149690587d1SChris Mason */ 115070458a58SNikolay Borisov static int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans) 1151d0c803c4SChris Mason { 1152663dfbb0SFilipe Manana int ret; 1153c9b577c0SNikolay Borisov int ret2; 1154c9b577c0SNikolay Borisov struct extent_io_tree *dirty_pages = &trans->transaction->dirty_pages; 115570458a58SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1156c9b577c0SNikolay Borisov struct blk_plug plug; 1157663dfbb0SFilipe Manana 1158c9b577c0SNikolay Borisov blk_start_plug(&plug); 1159c9b577c0SNikolay Borisov ret = btrfs_write_marked_extents(fs_info, dirty_pages, EXTENT_DIRTY); 1160c9b577c0SNikolay Borisov blk_finish_plug(&plug); 1161c9b577c0SNikolay Borisov ret2 = btrfs_wait_extents(fs_info, dirty_pages); 1162c9b577c0SNikolay Borisov 116341e7acd3SNikolay Borisov extent_io_tree_release(&trans->transaction->dirty_pages); 1164663dfbb0SFilipe Manana 1165c9b577c0SNikolay Borisov if (ret) 1166663dfbb0SFilipe Manana return ret; 1167c9b577c0SNikolay Borisov else if (ret2) 1168c9b577c0SNikolay Borisov return ret2; 1169c9b577c0SNikolay Borisov else 1170c9b577c0SNikolay Borisov return 0; 1171d0c803c4SChris Mason } 1172d0c803c4SChris Mason 1173d352ac68SChris Mason /* 1174d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 1175d352ac68SChris Mason * 1176d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 1177d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 1178d352ac68SChris Mason * allocation tree. 1179d352ac68SChris Mason * 1180d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 1181d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 1182d352ac68SChris Mason */ 11830b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 118479154b1bSChris Mason struct btrfs_root *root) 118579154b1bSChris Mason { 118679154b1bSChris Mason int ret; 11870b86a832SChris Mason u64 old_root_bytenr; 118886b9f2ecSYan, Zheng u64 old_root_used; 11890b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 11900b246afaSJeff Mahoney struct btrfs_root *tree_root = fs_info->tree_root; 119179154b1bSChris Mason 119286b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 119356bec294SChris Mason 119479154b1bSChris Mason while (1) { 11950b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 119686b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 1197ea526d18SJosef Bacik old_root_used == btrfs_root_used(&root->root_item)) 119879154b1bSChris Mason break; 119987ef2bb4SChris Mason 12005d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 120179154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 12020b86a832SChris Mason &root->root_key, 12030b86a832SChris Mason &root->root_item); 120449b25e05SJeff Mahoney if (ret) 120549b25e05SJeff Mahoney return ret; 120656bec294SChris Mason 120786b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 1208e7070be1SJosef Bacik } 1209276e680dSYan Zheng 12100b86a832SChris Mason return 0; 12110b86a832SChris Mason } 12120b86a832SChris Mason 1213d352ac68SChris Mason /* 1214d352ac68SChris Mason * update all the cowonly tree roots on disk 121549b25e05SJeff Mahoney * 121649b25e05SJeff Mahoney * The error handling in this function may not be obvious. Any of the 121749b25e05SJeff Mahoney * failures will cause the file system to go offline. We still need 121849b25e05SJeff Mahoney * to clean up the delayed refs. 1219d352ac68SChris Mason */ 12209386d8bcSNikolay Borisov static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans) 12210b86a832SChris Mason { 12229386d8bcSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 1223ea526d18SJosef Bacik struct list_head *dirty_bgs = &trans->transaction->dirty_bgs; 12241bbc621eSChris Mason struct list_head *io_bgs = &trans->transaction->io_bgs; 12250b86a832SChris Mason struct list_head *next; 122684234f3aSYan Zheng struct extent_buffer *eb; 122756bec294SChris Mason int ret; 122884234f3aSYan Zheng 122984234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 123049b25e05SJeff Mahoney ret = btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 12319631e4ccSJosef Bacik 0, &eb, BTRFS_NESTING_COW); 123284234f3aSYan Zheng btrfs_tree_unlock(eb); 123384234f3aSYan Zheng free_extent_buffer(eb); 12340b86a832SChris Mason 123549b25e05SJeff Mahoney if (ret) 123649b25e05SJeff Mahoney return ret; 123749b25e05SJeff Mahoney 1238196c9d8dSDavid Sterba ret = btrfs_run_dev_stats(trans); 1239c16ce190SJosef Bacik if (ret) 1240c16ce190SJosef Bacik return ret; 12412b584c68SDavid Sterba ret = btrfs_run_dev_replace(trans); 1242c16ce190SJosef Bacik if (ret) 1243c16ce190SJosef Bacik return ret; 1244280f8bd2SLu Fengqi ret = btrfs_run_qgroups(trans); 1245c16ce190SJosef Bacik if (ret) 1246c16ce190SJosef Bacik return ret; 1247546adb0dSJan Schmidt 1248bbebb3e0SDavid Sterba ret = btrfs_setup_space_cache(trans); 1249dcdf7f6dSJosef Bacik if (ret) 1250dcdf7f6dSJosef Bacik return ret; 1251dcdf7f6dSJosef Bacik 1252ea526d18SJosef Bacik again: 12530b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 12542ff7e61eSJeff Mahoney struct btrfs_root *root; 12550b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 12560b86a832SChris Mason list_del_init(next); 12570b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 1258e7070be1SJosef Bacik clear_bit(BTRFS_ROOT_DIRTY, &root->state); 125987ef2bb4SChris Mason 12609e351cc8SJosef Bacik if (root != fs_info->extent_root) 12619e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 12629e351cc8SJosef Bacik &trans->transaction->switch_commits); 126349b25e05SJeff Mahoney ret = update_cowonly_root(trans, root); 126449b25e05SJeff Mahoney if (ret) 126549b25e05SJeff Mahoney return ret; 1266488bc2a2SJosef Bacik } 1267488bc2a2SJosef Bacik 1268488bc2a2SJosef Bacik /* Now flush any delayed refs generated by updating all of the roots */ 1269c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1270ea526d18SJosef Bacik if (ret) 1271ea526d18SJosef Bacik return ret; 1272276e680dSYan Zheng 12731bbc621eSChris Mason while (!list_empty(dirty_bgs) || !list_empty(io_bgs)) { 12745742d15fSDavid Sterba ret = btrfs_write_dirty_block_groups(trans); 1275ea526d18SJosef Bacik if (ret) 1276ea526d18SJosef Bacik return ret; 1277488bc2a2SJosef Bacik 1278488bc2a2SJosef Bacik /* 1279488bc2a2SJosef Bacik * We're writing the dirty block groups, which could generate 1280488bc2a2SJosef Bacik * delayed refs, which could generate more dirty block groups, 1281488bc2a2SJosef Bacik * so we want to keep this flushing in this loop to make sure 1282488bc2a2SJosef Bacik * everything gets run. 1283488bc2a2SJosef Bacik */ 1284c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 1285ea526d18SJosef Bacik if (ret) 1286ea526d18SJosef Bacik return ret; 1287ea526d18SJosef Bacik } 1288ea526d18SJosef Bacik 1289ea526d18SJosef Bacik if (!list_empty(&fs_info->dirty_cowonly_roots)) 1290ea526d18SJosef Bacik goto again; 1291ea526d18SJosef Bacik 12929e351cc8SJosef Bacik list_add_tail(&fs_info->extent_root->dirty_list, 12939e351cc8SJosef Bacik &trans->transaction->switch_commits); 12949f6cbcbbSDavid Sterba 12959f6cbcbbSDavid Sterba /* Update dev-replace pointer once everything is committed */ 12969f6cbcbbSDavid Sterba fs_info->dev_replace.committed_cursor_left = 12979f6cbcbbSDavid Sterba fs_info->dev_replace.cursor_left_last_write_of_item; 12988dabb742SStefan Behrens 129979154b1bSChris Mason return 0; 130079154b1bSChris Mason } 130179154b1bSChris Mason 1302d352ac68SChris Mason /* 1303d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 1304d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 1305d352ac68SChris Mason * be deleted 1306d352ac68SChris Mason */ 1307cfad392bSJosef Bacik void btrfs_add_dead_root(struct btrfs_root *root) 13085eda7b5eSChris Mason { 13090b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13100b246afaSJeff Mahoney 13110b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1312dc9492c1SJosef Bacik if (list_empty(&root->root_list)) { 1313dc9492c1SJosef Bacik btrfs_grab_root(root); 13140b246afaSJeff Mahoney list_add_tail(&root->root_list, &fs_info->dead_roots); 1315dc9492c1SJosef Bacik } 13160b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 13175eda7b5eSChris Mason } 13185eda7b5eSChris Mason 1319d352ac68SChris Mason /* 13205d4f98a2SYan Zheng * update all the cowonly tree roots on disk 1321d352ac68SChris Mason */ 13227e4443d9SNikolay Borisov static noinline int commit_fs_roots(struct btrfs_trans_handle *trans) 13230f7d52f4SChris Mason { 13247e4443d9SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 13250f7d52f4SChris Mason struct btrfs_root *gang[8]; 13260f7d52f4SChris Mason int i; 13270f7d52f4SChris Mason int ret; 132854aa1f4dSChris Mason 1329a4abeea4SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 13300f7d52f4SChris Mason while (1) { 13315d4f98a2SYan Zheng ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 13325d4f98a2SYan Zheng (void **)gang, 0, 13330f7d52f4SChris Mason ARRAY_SIZE(gang), 13340f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 13350f7d52f4SChris Mason if (ret == 0) 13360f7d52f4SChris Mason break; 13370f7d52f4SChris Mason for (i = 0; i < ret; i++) { 13385b4aacefSJeff Mahoney struct btrfs_root *root = gang[i]; 13394f4317c1SJosef Bacik int ret2; 13404f4317c1SJosef Bacik 13415d4f98a2SYan Zheng radix_tree_tag_clear(&fs_info->fs_roots_radix, 13422619ba1fSChris Mason (unsigned long)root->root_key.objectid, 13430f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 1344a4abeea4SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 134531153d81SYan Zheng 1346e02119d5SChris Mason btrfs_free_log(trans, root); 13475d4f98a2SYan Zheng btrfs_update_reloc_root(trans, root); 1348e02119d5SChris Mason 1349f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 135027cdeb70SMiao Xie clear_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1351c7548af6SChris Mason smp_mb__after_atomic(); 1352f1ebcc74SLiu Bo 1353978d910dSYan Zheng if (root->commit_root != root->node) { 13549e351cc8SJosef Bacik list_add_tail(&root->dirty_list, 13559e351cc8SJosef Bacik &trans->transaction->switch_commits); 1356978d910dSYan Zheng btrfs_set_root_node(&root->root_item, 1357978d910dSYan Zheng root->node); 1358978d910dSYan Zheng } 135931153d81SYan Zheng 13604f4317c1SJosef Bacik ret2 = btrfs_update_root(trans, fs_info->tree_root, 13610f7d52f4SChris Mason &root->root_key, 13620f7d52f4SChris Mason &root->root_item); 13634f4317c1SJosef Bacik if (ret2) 13644f4317c1SJosef Bacik return ret2; 1365a4abeea4SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 1366733e03a0SQu Wenruo btrfs_qgroup_free_meta_all_pertrans(root); 13670f7d52f4SChris Mason } 13689f3a7427SChris Mason } 1369a4abeea4SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 13704f4317c1SJosef Bacik return 0; 13710f7d52f4SChris Mason } 13720f7d52f4SChris Mason 1373d352ac68SChris Mason /* 1374de78b51aSEric Sandeen * defrag a given btree. 1375de78b51aSEric Sandeen * Every leaf in the btree is read and defragged. 1376d352ac68SChris Mason */ 1377de78b51aSEric Sandeen int btrfs_defrag_root(struct btrfs_root *root) 1378e9d0b13bSChris Mason { 1379e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 1380e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 13818929ecfaSYan, Zheng int ret; 1382e9d0b13bSChris Mason 138327cdeb70SMiao Xie if (test_and_set_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state)) 1384e9d0b13bSChris Mason return 0; 13858929ecfaSYan, Zheng 13866b80053dSChris Mason while (1) { 13878929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 13888929ecfaSYan, Zheng if (IS_ERR(trans)) 13898929ecfaSYan, Zheng return PTR_ERR(trans); 13908929ecfaSYan, Zheng 1391de78b51aSEric Sandeen ret = btrfs_defrag_leaves(trans, root); 13928929ecfaSYan, Zheng 13933a45bb20SJeff Mahoney btrfs_end_transaction(trans); 13942ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(info); 1395e9d0b13bSChris Mason cond_resched(); 1396e9d0b13bSChris Mason 1397ab8d0fc4SJeff Mahoney if (btrfs_fs_closing(info) || ret != -EAGAIN) 1398e9d0b13bSChris Mason break; 1399210549ebSDavid Sterba 1400ab8d0fc4SJeff Mahoney if (btrfs_defrag_cancelled(info)) { 1401ab8d0fc4SJeff Mahoney btrfs_debug(info, "defrag_root cancelled"); 1402210549ebSDavid Sterba ret = -EAGAIN; 1403210549ebSDavid Sterba break; 1404210549ebSDavid Sterba } 1405e9d0b13bSChris Mason } 140627cdeb70SMiao Xie clear_bit(BTRFS_ROOT_DEFRAG_RUNNING, &root->state); 14078929ecfaSYan, Zheng return ret; 1408e9d0b13bSChris Mason } 1409e9d0b13bSChris Mason 1410d352ac68SChris Mason /* 14116426c7adSQu Wenruo * Do all special snapshot related qgroup dirty hack. 14126426c7adSQu Wenruo * 14136426c7adSQu Wenruo * Will do all needed qgroup inherit and dirty hack like switch commit 14146426c7adSQu Wenruo * roots inside one transaction and write all btree into disk, to make 14156426c7adSQu Wenruo * qgroup works. 14166426c7adSQu Wenruo */ 14176426c7adSQu Wenruo static int qgroup_account_snapshot(struct btrfs_trans_handle *trans, 14186426c7adSQu Wenruo struct btrfs_root *src, 14196426c7adSQu Wenruo struct btrfs_root *parent, 14206426c7adSQu Wenruo struct btrfs_qgroup_inherit *inherit, 14216426c7adSQu Wenruo u64 dst_objectid) 14226426c7adSQu Wenruo { 14236426c7adSQu Wenruo struct btrfs_fs_info *fs_info = src->fs_info; 14246426c7adSQu Wenruo int ret; 14256426c7adSQu Wenruo 14266426c7adSQu Wenruo /* 14276426c7adSQu Wenruo * Save some performance in the case that qgroups are not 14286426c7adSQu Wenruo * enabled. If this check races with the ioctl, rescan will 14296426c7adSQu Wenruo * kick in anyway. 14306426c7adSQu Wenruo */ 14319ea6e2b5SDavid Sterba if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags)) 14326426c7adSQu Wenruo return 0; 14336426c7adSQu Wenruo 14346426c7adSQu Wenruo /* 143552042d8eSAndrea Gelmini * Ensure dirty @src will be committed. Or, after coming 14364d31778aSQu Wenruo * commit_fs_roots() and switch_commit_roots(), any dirty but not 14374d31778aSQu Wenruo * recorded root will never be updated again, causing an outdated root 14384d31778aSQu Wenruo * item. 14394d31778aSQu Wenruo */ 14404d31778aSQu Wenruo record_root_in_trans(trans, src, 1); 14414d31778aSQu Wenruo 14424d31778aSQu Wenruo /* 14432a4d84c1SJosef Bacik * btrfs_qgroup_inherit relies on a consistent view of the usage for the 14442a4d84c1SJosef Bacik * src root, so we must run the delayed refs here. 14452a4d84c1SJosef Bacik * 14462a4d84c1SJosef Bacik * However this isn't particularly fool proof, because there's no 14472a4d84c1SJosef Bacik * synchronization keeping us from changing the tree after this point 14482a4d84c1SJosef Bacik * before we do the qgroup_inherit, or even from making changes while 14492a4d84c1SJosef Bacik * we're doing the qgroup_inherit. But that's a problem for the future, 14502a4d84c1SJosef Bacik * for now flush the delayed refs to narrow the race window where the 14512a4d84c1SJosef Bacik * qgroup counters could end up wrong. 14522a4d84c1SJosef Bacik */ 14532a4d84c1SJosef Bacik ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 14542a4d84c1SJosef Bacik if (ret) { 14552a4d84c1SJosef Bacik btrfs_abort_transaction(trans, ret); 14562a4d84c1SJosef Bacik goto out; 14572a4d84c1SJosef Bacik } 14582a4d84c1SJosef Bacik 14592a4d84c1SJosef Bacik /* 14606426c7adSQu Wenruo * We are going to commit transaction, see btrfs_commit_transaction() 14616426c7adSQu Wenruo * comment for reason locking tree_log_mutex 14626426c7adSQu Wenruo */ 14636426c7adSQu Wenruo mutex_lock(&fs_info->tree_log_mutex); 14646426c7adSQu Wenruo 14657e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 14666426c7adSQu Wenruo if (ret) 14676426c7adSQu Wenruo goto out; 1468460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 14696426c7adSQu Wenruo if (ret < 0) 14706426c7adSQu Wenruo goto out; 14716426c7adSQu Wenruo 14726426c7adSQu Wenruo /* Now qgroup are all updated, we can inherit it to new qgroups */ 1473a9377422SLu Fengqi ret = btrfs_qgroup_inherit(trans, src->root_key.objectid, dst_objectid, 14746426c7adSQu Wenruo inherit); 14756426c7adSQu Wenruo if (ret < 0) 14766426c7adSQu Wenruo goto out; 14776426c7adSQu Wenruo 14786426c7adSQu Wenruo /* 14796426c7adSQu Wenruo * Now we do a simplified commit transaction, which will: 14806426c7adSQu Wenruo * 1) commit all subvolume and extent tree 14816426c7adSQu Wenruo * To ensure all subvolume and extent tree have a valid 14826426c7adSQu Wenruo * commit_root to accounting later insert_dir_item() 14836426c7adSQu Wenruo * 2) write all btree blocks onto disk 14846426c7adSQu Wenruo * This is to make sure later btree modification will be cowed 14856426c7adSQu Wenruo * Or commit_root can be populated and cause wrong qgroup numbers 14866426c7adSQu Wenruo * In this simplified commit, we don't really care about other trees 14876426c7adSQu Wenruo * like chunk and root tree, as they won't affect qgroup. 14886426c7adSQu Wenruo * And we don't write super to avoid half committed status. 14896426c7adSQu Wenruo */ 14909386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 14916426c7adSQu Wenruo if (ret) 14926426c7adSQu Wenruo goto out; 1493889bfa39SJosef Bacik switch_commit_roots(trans); 149470458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 14956426c7adSQu Wenruo if (ret) 1496f7af3934SDavid Sterba btrfs_handle_fs_error(fs_info, ret, 14976426c7adSQu Wenruo "Error while writing out transaction for qgroup"); 14986426c7adSQu Wenruo 14996426c7adSQu Wenruo out: 15006426c7adSQu Wenruo mutex_unlock(&fs_info->tree_log_mutex); 15016426c7adSQu Wenruo 15026426c7adSQu Wenruo /* 15036426c7adSQu Wenruo * Force parent root to be updated, as we recorded it before so its 15046426c7adSQu Wenruo * last_trans == cur_transid. 15056426c7adSQu Wenruo * Or it won't be committed again onto disk after later 15066426c7adSQu Wenruo * insert_dir_item() 15076426c7adSQu Wenruo */ 15086426c7adSQu Wenruo if (!ret) 15096426c7adSQu Wenruo record_root_in_trans(trans, parent, 1); 15106426c7adSQu Wenruo return ret; 15116426c7adSQu Wenruo } 15126426c7adSQu Wenruo 15136426c7adSQu Wenruo /* 1514d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 1515aec8030aSMiao Xie * transaction commit. This does the actual creation. 1516aec8030aSMiao Xie * 1517aec8030aSMiao Xie * Note: 1518aec8030aSMiao Xie * If the error which may affect the commitment of the current transaction 1519aec8030aSMiao Xie * happens, we should return the error number. If the error which just affect 1520aec8030aSMiao Xie * the creation of the pending snapshots, just return 0. 1521d352ac68SChris Mason */ 152280b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 15233063d29fSChris Mason struct btrfs_pending_snapshot *pending) 15243063d29fSChris Mason { 152508d50ca3SNikolay Borisov 152608d50ca3SNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 15273063d29fSChris Mason struct btrfs_key key; 152880b6794dSChris Mason struct btrfs_root_item *new_root_item; 15293063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 15303063d29fSChris Mason struct btrfs_root *root = pending->root; 15316bdb72deSSage Weil struct btrfs_root *parent_root; 153298c9942aSLiu Bo struct btrfs_block_rsv *rsv; 15336bdb72deSSage Weil struct inode *parent_inode; 153442874b3dSMiao Xie struct btrfs_path *path; 153542874b3dSMiao Xie struct btrfs_dir_item *dir_item; 1536a22285a6SYan, Zheng struct dentry *dentry; 15373063d29fSChris Mason struct extent_buffer *tmp; 1538925baeddSChris Mason struct extent_buffer *old; 153995582b00SDeepa Dinamani struct timespec64 cur_time; 1540aec8030aSMiao Xie int ret = 0; 1541d68fc57bSYan, Zheng u64 to_reserve = 0; 15426bdb72deSSage Weil u64 index = 0; 1543a22285a6SYan, Zheng u64 objectid; 1544b83cc969SLi Zefan u64 root_flags; 15453063d29fSChris Mason 15468546b570SDavid Sterba ASSERT(pending->path); 15478546b570SDavid Sterba path = pending->path; 154842874b3dSMiao Xie 1549b0c0ea63SDavid Sterba ASSERT(pending->root_item); 1550b0c0ea63SDavid Sterba new_root_item = pending->root_item; 1551a22285a6SYan, Zheng 1552543068a2SNikolay Borisov pending->error = btrfs_get_free_objectid(tree_root, &objectid); 1553aec8030aSMiao Xie if (pending->error) 15546fa9700eSMiao Xie goto no_free_objectid; 15553063d29fSChris Mason 1556d6726335SQu Wenruo /* 1557d6726335SQu Wenruo * Make qgroup to skip current new snapshot's qgroupid, as it is 1558d6726335SQu Wenruo * accounted by later btrfs_qgroup_inherit(). 1559d6726335SQu Wenruo */ 1560d6726335SQu Wenruo btrfs_set_skip_qgroup(trans, objectid); 1561d6726335SQu Wenruo 1562147d256eSZhaolei btrfs_reloc_pre_snapshot(pending, &to_reserve); 1563d68fc57bSYan, Zheng 1564d68fc57bSYan, Zheng if (to_reserve > 0) { 1565aec8030aSMiao Xie pending->error = btrfs_block_rsv_add(root, 1566aec8030aSMiao Xie &pending->block_rsv, 156708e007d2SMiao Xie to_reserve, 156808e007d2SMiao Xie BTRFS_RESERVE_NO_FLUSH); 1569aec8030aSMiao Xie if (pending->error) 1570d6726335SQu Wenruo goto clear_skip_qgroup; 1571d68fc57bSYan, Zheng } 1572d68fc57bSYan, Zheng 15733063d29fSChris Mason key.objectid = objectid; 1574a22285a6SYan, Zheng key.offset = (u64)-1; 1575a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 15763063d29fSChris Mason 15776fa9700eSMiao Xie rsv = trans->block_rsv; 1578a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 15792382c5ccSLiu Bo trans->bytes_reserved = trans->block_rsv->reserved; 15800b246afaSJeff Mahoney trace_btrfs_space_reservation(fs_info, "transaction", 158188d3a5aaSJosef Bacik trans->transid, 158288d3a5aaSJosef Bacik trans->bytes_reserved, 1); 1583a22285a6SYan, Zheng dentry = pending->dentry; 1584e9662f70SMiao Xie parent_inode = pending->dir; 1585a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 15866426c7adSQu Wenruo record_root_in_trans(trans, parent_root, 0); 1587a22285a6SYan, Zheng 1588c2050a45SDeepa Dinamani cur_time = current_time(parent_inode); 158904b285f3SDeepa Dinamani 15906bdb72deSSage Weil /* 15916bdb72deSSage Weil * insert the directory item 15926bdb72deSSage Weil */ 1593877574e2SNikolay Borisov ret = btrfs_set_inode_index(BTRFS_I(parent_inode), &index); 159449b25e05SJeff Mahoney BUG_ON(ret); /* -ENOMEM */ 159542874b3dSMiao Xie 159642874b3dSMiao Xie /* check if there is a file/dir which has the same name. */ 159742874b3dSMiao Xie dir_item = btrfs_lookup_dir_item(NULL, parent_root, path, 15984a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), 159942874b3dSMiao Xie dentry->d_name.name, 160042874b3dSMiao Xie dentry->d_name.len, 0); 160142874b3dSMiao Xie if (dir_item != NULL && !IS_ERR(dir_item)) { 1602fe66a05aSChris Mason pending->error = -EEXIST; 1603aec8030aSMiao Xie goto dir_item_existed; 160442874b3dSMiao Xie } else if (IS_ERR(dir_item)) { 160542874b3dSMiao Xie ret = PTR_ERR(dir_item); 160666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16078732d44fSMiao Xie goto fail; 160879787eaaSJeff Mahoney } 160942874b3dSMiao Xie btrfs_release_path(path); 16106bdb72deSSage Weil 1611e999376fSChris Mason /* 1612e999376fSChris Mason * pull in the delayed directory update 1613e999376fSChris Mason * and the delayed inode item 1614e999376fSChris Mason * otherwise we corrupt the FS during 1615e999376fSChris Mason * snapshot 1616e999376fSChris Mason */ 1617e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 16188732d44fSMiao Xie if (ret) { /* Transaction aborted */ 161966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16208732d44fSMiao Xie goto fail; 16218732d44fSMiao Xie } 1622e999376fSChris Mason 16236426c7adSQu Wenruo record_root_in_trans(trans, root, 0); 16246bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 16256bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 162608fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 16276bdb72deSSage Weil 1628b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 1629b83cc969SLi Zefan if (pending->readonly) 1630b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 1631b83cc969SLi Zefan else 1632b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 1633b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 1634b83cc969SLi Zefan 16358ea05e3aSAlexander Block btrfs_set_root_generation_v2(new_root_item, 16368ea05e3aSAlexander Block trans->transid); 1637807fc790SAndy Shevchenko generate_random_guid(new_root_item->uuid); 16388ea05e3aSAlexander Block memcpy(new_root_item->parent_uuid, root->root_item.uuid, 16398ea05e3aSAlexander Block BTRFS_UUID_SIZE); 164070023da2SStefan Behrens if (!(root_flags & BTRFS_ROOT_SUBVOL_RDONLY)) { 164170023da2SStefan Behrens memset(new_root_item->received_uuid, 0, 164270023da2SStefan Behrens sizeof(new_root_item->received_uuid)); 16438ea05e3aSAlexander Block memset(&new_root_item->stime, 0, sizeof(new_root_item->stime)); 16448ea05e3aSAlexander Block memset(&new_root_item->rtime, 0, sizeof(new_root_item->rtime)); 16458ea05e3aSAlexander Block btrfs_set_root_stransid(new_root_item, 0); 16468ea05e3aSAlexander Block btrfs_set_root_rtransid(new_root_item, 0); 164770023da2SStefan Behrens } 16483cae210fSQu Wenruo btrfs_set_stack_timespec_sec(&new_root_item->otime, cur_time.tv_sec); 16493cae210fSQu Wenruo btrfs_set_stack_timespec_nsec(&new_root_item->otime, cur_time.tv_nsec); 165070023da2SStefan Behrens btrfs_set_root_otransid(new_root_item, trans->transid); 16518ea05e3aSAlexander Block 1652925baeddSChris Mason old = btrfs_lock_root_node(root); 16539631e4ccSJosef Bacik ret = btrfs_cow_block(trans, root, old, NULL, 0, &old, 16549631e4ccSJosef Bacik BTRFS_NESTING_COW); 165579787eaaSJeff Mahoney if (ret) { 165679787eaaSJeff Mahoney btrfs_tree_unlock(old); 165779787eaaSJeff Mahoney free_extent_buffer(old); 165866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16598732d44fSMiao Xie goto fail; 166079787eaaSJeff Mahoney } 166149b25e05SJeff Mahoney 166249b25e05SJeff Mahoney ret = btrfs_copy_root(trans, root, old, &tmp, objectid); 166379787eaaSJeff Mahoney /* clean up in any case */ 1664925baeddSChris Mason btrfs_tree_unlock(old); 1665925baeddSChris Mason free_extent_buffer(old); 16668732d44fSMiao Xie if (ret) { 166766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16688732d44fSMiao Xie goto fail; 16698732d44fSMiao Xie } 1670f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 167127cdeb70SMiao Xie set_bit(BTRFS_ROOT_FORCE_COW, &root->state); 1672f1ebcc74SLiu Bo smp_wmb(); 1673f1ebcc74SLiu Bo 16745d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 1675a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 1676a22285a6SYan, Zheng key.offset = trans->transid; 1677a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 1678925baeddSChris Mason btrfs_tree_unlock(tmp); 16793063d29fSChris Mason free_extent_buffer(tmp); 16808732d44fSMiao Xie if (ret) { 168166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16828732d44fSMiao Xie goto fail; 16838732d44fSMiao Xie } 16840660b5afSChris Mason 1685a22285a6SYan, Zheng /* 1686a22285a6SYan, Zheng * insert root back/forward references 1687a22285a6SYan, Zheng */ 16886025c19fSLu Fengqi ret = btrfs_add_root_ref(trans, objectid, 1689a22285a6SYan, Zheng parent_root->root_key.objectid, 16904a0cc7caSNikolay Borisov btrfs_ino(BTRFS_I(parent_inode)), index, 1691a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len); 16928732d44fSMiao Xie if (ret) { 169366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 16948732d44fSMiao Xie goto fail; 16958732d44fSMiao Xie } 1696a22285a6SYan, Zheng 1697a22285a6SYan, Zheng key.offset = (u64)-1; 16982dfb1e43SQu Wenruo pending->snap = btrfs_get_new_fs_root(fs_info, objectid, pending->anon_dev); 169979787eaaSJeff Mahoney if (IS_ERR(pending->snap)) { 170079787eaaSJeff Mahoney ret = PTR_ERR(pending->snap); 17012d892ccdSFilipe Manana pending->snap = NULL; 170266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17038732d44fSMiao Xie goto fail; 170479787eaaSJeff Mahoney } 1705d68fc57bSYan, Zheng 170649b25e05SJeff Mahoney ret = btrfs_reloc_post_snapshot(trans, pending); 17078732d44fSMiao Xie if (ret) { 170866642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17098732d44fSMiao Xie goto fail; 17108732d44fSMiao Xie } 1711361048f5SMiao Xie 17126426c7adSQu Wenruo /* 17136426c7adSQu Wenruo * Do special qgroup accounting for snapshot, as we do some qgroup 17146426c7adSQu Wenruo * snapshot hack to do fast snapshot. 17156426c7adSQu Wenruo * To co-operate with that hack, we do hack again. 17166426c7adSQu Wenruo * Or snapshot will be greatly slowed down by a subtree qgroup rescan 17176426c7adSQu Wenruo */ 17186426c7adSQu Wenruo ret = qgroup_account_snapshot(trans, root, parent_root, 17196426c7adSQu Wenruo pending->inherit, objectid); 17206426c7adSQu Wenruo if (ret < 0) 17216426c7adSQu Wenruo goto fail; 17226426c7adSQu Wenruo 1723684572dfSLu Fengqi ret = btrfs_insert_dir_item(trans, dentry->d_name.name, 1724684572dfSLu Fengqi dentry->d_name.len, BTRFS_I(parent_inode), 1725684572dfSLu Fengqi &key, BTRFS_FT_DIR, index); 172642874b3dSMiao Xie /* We have check then name at the beginning, so it is impossible. */ 17279c52057cSChris Mason BUG_ON(ret == -EEXIST || ret == -EOVERFLOW); 17288732d44fSMiao Xie if (ret) { 172966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 17308732d44fSMiao Xie goto fail; 17318732d44fSMiao Xie } 173242874b3dSMiao Xie 17336ef06d27SNikolay Borisov btrfs_i_size_write(BTRFS_I(parent_inode), parent_inode->i_size + 173442874b3dSMiao Xie dentry->d_name.len * 2); 173504b285f3SDeepa Dinamani parent_inode->i_mtime = parent_inode->i_ctime = 1736c2050a45SDeepa Dinamani current_time(parent_inode); 1737729f7961SNikolay Borisov ret = btrfs_update_inode_fallback(trans, parent_root, BTRFS_I(parent_inode)); 1738dd5f9615SStefan Behrens if (ret) { 173966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1740dd5f9615SStefan Behrens goto fail; 1741dd5f9615SStefan Behrens } 1742807fc790SAndy Shevchenko ret = btrfs_uuid_tree_add(trans, new_root_item->uuid, 1743807fc790SAndy Shevchenko BTRFS_UUID_KEY_SUBVOL, 1744cdb345a8SLu Fengqi objectid); 1745dd5f9615SStefan Behrens if (ret) { 174666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1747dd5f9615SStefan Behrens goto fail; 1748dd5f9615SStefan Behrens } 1749dd5f9615SStefan Behrens if (!btrfs_is_empty_uuid(new_root_item->received_uuid)) { 1750cdb345a8SLu Fengqi ret = btrfs_uuid_tree_add(trans, new_root_item->received_uuid, 1751dd5f9615SStefan Behrens BTRFS_UUID_KEY_RECEIVED_SUBVOL, 1752dd5f9615SStefan Behrens objectid); 1753dd5f9615SStefan Behrens if (ret && ret != -EEXIST) { 175466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 1755dd5f9615SStefan Behrens goto fail; 1756dd5f9615SStefan Behrens } 1757dd5f9615SStefan Behrens } 1758d6726335SQu Wenruo 17593063d29fSChris Mason fail: 1760aec8030aSMiao Xie pending->error = ret; 1761aec8030aSMiao Xie dir_item_existed: 176298c9942aSLiu Bo trans->block_rsv = rsv; 17632382c5ccSLiu Bo trans->bytes_reserved = 0; 1764d6726335SQu Wenruo clear_skip_qgroup: 1765d6726335SQu Wenruo btrfs_clear_skip_qgroup(trans); 17666fa9700eSMiao Xie no_free_objectid: 17676fa9700eSMiao Xie kfree(new_root_item); 1768b0c0ea63SDavid Sterba pending->root_item = NULL; 176942874b3dSMiao Xie btrfs_free_path(path); 17708546b570SDavid Sterba pending->path = NULL; 17718546b570SDavid Sterba 177249b25e05SJeff Mahoney return ret; 17733063d29fSChris Mason } 17743063d29fSChris Mason 1775d352ac68SChris Mason /* 1776d352ac68SChris Mason * create all the snapshots we've scheduled for creation 1777d352ac68SChris Mason */ 177808d50ca3SNikolay Borisov static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans) 17793063d29fSChris Mason { 1780aec8030aSMiao Xie struct btrfs_pending_snapshot *pending, *next; 17813063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 1782aec8030aSMiao Xie int ret = 0; 17833de4586cSChris Mason 1784aec8030aSMiao Xie list_for_each_entry_safe(pending, next, head, list) { 1785aec8030aSMiao Xie list_del(&pending->list); 178608d50ca3SNikolay Borisov ret = create_pending_snapshot(trans, pending); 1787aec8030aSMiao Xie if (ret) 1788aec8030aSMiao Xie break; 1789aec8030aSMiao Xie } 1790aec8030aSMiao Xie return ret; 17913de4586cSChris Mason } 17923de4586cSChris Mason 17932ff7e61eSJeff Mahoney static void update_super_roots(struct btrfs_fs_info *fs_info) 17945d4f98a2SYan Zheng { 17955d4f98a2SYan Zheng struct btrfs_root_item *root_item; 17965d4f98a2SYan Zheng struct btrfs_super_block *super; 17975d4f98a2SYan Zheng 17980b246afaSJeff Mahoney super = fs_info->super_copy; 17995d4f98a2SYan Zheng 18000b246afaSJeff Mahoney root_item = &fs_info->chunk_root->root_item; 1801093e037cSDavid Sterba super->chunk_root = root_item->bytenr; 1802093e037cSDavid Sterba super->chunk_root_generation = root_item->generation; 1803093e037cSDavid Sterba super->chunk_root_level = root_item->level; 18045d4f98a2SYan Zheng 18050b246afaSJeff Mahoney root_item = &fs_info->tree_root->root_item; 1806093e037cSDavid Sterba super->root = root_item->bytenr; 1807093e037cSDavid Sterba super->generation = root_item->generation; 1808093e037cSDavid Sterba super->root_level = root_item->level; 18090b246afaSJeff Mahoney if (btrfs_test_opt(fs_info, SPACE_CACHE)) 1810093e037cSDavid Sterba super->cache_generation = root_item->generation; 181194846229SBoris Burkov else if (test_bit(BTRFS_FS_CLEANUP_SPACE_CACHE_V1, &fs_info->flags)) 181294846229SBoris Burkov super->cache_generation = 0; 18130b246afaSJeff Mahoney if (test_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags)) 1814093e037cSDavid Sterba super->uuid_tree_generation = root_item->generation; 18155d4f98a2SYan Zheng } 18165d4f98a2SYan Zheng 1817f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1818f36f3042SChris Mason { 18194a9d8bdeSMiao Xie struct btrfs_transaction *trans; 1820f36f3042SChris Mason int ret = 0; 18214a9d8bdeSMiao Xie 1822a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 18234a9d8bdeSMiao Xie trans = info->running_transaction; 18244a9d8bdeSMiao Xie if (trans) 18254a9d8bdeSMiao Xie ret = (trans->state >= TRANS_STATE_COMMIT_START); 1826a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 1827f36f3042SChris Mason return ret; 1828f36f3042SChris Mason } 1829f36f3042SChris Mason 18308929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 18318929ecfaSYan, Zheng { 18324a9d8bdeSMiao Xie struct btrfs_transaction *trans; 18338929ecfaSYan, Zheng int ret = 0; 18344a9d8bdeSMiao Xie 1835a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 18364a9d8bdeSMiao Xie trans = info->running_transaction; 18374a9d8bdeSMiao Xie if (trans) 18384a9d8bdeSMiao Xie ret = is_transaction_blocked(trans); 1839a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 18408929ecfaSYan, Zheng return ret; 18418929ecfaSYan, Zheng } 18428929ecfaSYan, Zheng 1843bb9c12c9SSage Weil /* 1844bb9c12c9SSage Weil * wait for the current transaction commit to start and block subsequent 1845bb9c12c9SSage Weil * transaction joins 1846bb9c12c9SSage Weil */ 18472ff7e61eSJeff Mahoney static void wait_current_trans_commit_start(struct btrfs_fs_info *fs_info, 1848bb9c12c9SSage Weil struct btrfs_transaction *trans) 1849bb9c12c9SSage Weil { 18502ff7e61eSJeff Mahoney wait_event(fs_info->transaction_blocked_wait, 1851bf31f87fSDavid Sterba trans->state >= TRANS_STATE_COMMIT_START || 1852bf31f87fSDavid Sterba TRANS_ABORTED(trans)); 1853bb9c12c9SSage Weil } 1854bb9c12c9SSage Weil 1855bb9c12c9SSage Weil /* 1856bb9c12c9SSage Weil * wait for the current transaction to start and then become unblocked. 1857bb9c12c9SSage Weil * caller holds ref. 1858bb9c12c9SSage Weil */ 18592ff7e61eSJeff Mahoney static void wait_current_trans_commit_start_and_unblock( 18602ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info, 1861bb9c12c9SSage Weil struct btrfs_transaction *trans) 1862bb9c12c9SSage Weil { 18632ff7e61eSJeff Mahoney wait_event(fs_info->transaction_wait, 1864bf31f87fSDavid Sterba trans->state >= TRANS_STATE_UNBLOCKED || 1865bf31f87fSDavid Sterba TRANS_ABORTED(trans)); 1866bb9c12c9SSage Weil } 1867bb9c12c9SSage Weil 1868bb9c12c9SSage Weil /* 1869bb9c12c9SSage Weil * commit transactions asynchronously. once btrfs_commit_transaction_async 1870bb9c12c9SSage Weil * returns, any subsequent transaction will not be allowed to join. 1871bb9c12c9SSage Weil */ 1872bb9c12c9SSage Weil struct btrfs_async_commit { 1873bb9c12c9SSage Weil struct btrfs_trans_handle *newtrans; 18747892b5afSMiao Xie struct work_struct work; 1875bb9c12c9SSage Weil }; 1876bb9c12c9SSage Weil 1877bb9c12c9SSage Weil static void do_async_commit(struct work_struct *work) 1878bb9c12c9SSage Weil { 1879bb9c12c9SSage Weil struct btrfs_async_commit *ac = 18807892b5afSMiao Xie container_of(work, struct btrfs_async_commit, work); 1881bb9c12c9SSage Weil 18826fc4e354SSage Weil /* 18836fc4e354SSage Weil * We've got freeze protection passed with the transaction. 18846fc4e354SSage Weil * Tell lockdep about it. 18856fc4e354SSage Weil */ 1886b1a06a4bSLiu Bo if (ac->newtrans->type & __TRANS_FREEZABLE) 18873a45bb20SJeff Mahoney __sb_writers_acquired(ac->newtrans->fs_info->sb, SB_FREEZE_FS); 18886fc4e354SSage Weil 1889e209db7aSSage Weil current->journal_info = ac->newtrans; 1890e209db7aSSage Weil 18913a45bb20SJeff Mahoney btrfs_commit_transaction(ac->newtrans); 1892bb9c12c9SSage Weil kfree(ac); 1893bb9c12c9SSage Weil } 1894bb9c12c9SSage Weil 1895bb9c12c9SSage Weil int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, 1896bb9c12c9SSage Weil int wait_for_unblock) 1897bb9c12c9SSage Weil { 18983a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 1899bb9c12c9SSage Weil struct btrfs_async_commit *ac; 1900bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1901bb9c12c9SSage Weil 1902bb9c12c9SSage Weil ac = kmalloc(sizeof(*ac), GFP_NOFS); 1903db5b493aSTsutomu Itoh if (!ac) 1904db5b493aSTsutomu Itoh return -ENOMEM; 1905bb9c12c9SSage Weil 19067892b5afSMiao Xie INIT_WORK(&ac->work, do_async_commit); 19073a45bb20SJeff Mahoney ac->newtrans = btrfs_join_transaction(trans->root); 19083612b495STsutomu Itoh if (IS_ERR(ac->newtrans)) { 19093612b495STsutomu Itoh int err = PTR_ERR(ac->newtrans); 19103612b495STsutomu Itoh kfree(ac); 19113612b495STsutomu Itoh return err; 19123612b495STsutomu Itoh } 1913bb9c12c9SSage Weil 1914bb9c12c9SSage Weil /* take transaction reference */ 1915bb9c12c9SSage Weil cur_trans = trans->transaction; 19169b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 1917bb9c12c9SSage Weil 19183a45bb20SJeff Mahoney btrfs_end_transaction(trans); 19196fc4e354SSage Weil 19206fc4e354SSage Weil /* 19216fc4e354SSage Weil * Tell lockdep we've released the freeze rwsem, since the 19226fc4e354SSage Weil * async commit thread will be the one to unlock it. 19236fc4e354SSage Weil */ 1924b1a06a4bSLiu Bo if (ac->newtrans->type & __TRANS_FREEZABLE) 19250b246afaSJeff Mahoney __sb_writers_release(fs_info->sb, SB_FREEZE_FS); 19266fc4e354SSage Weil 19277892b5afSMiao Xie schedule_work(&ac->work); 1928bb9c12c9SSage Weil 1929bb9c12c9SSage Weil /* wait for transaction to start and unblock */ 1930bb9c12c9SSage Weil if (wait_for_unblock) 19312ff7e61eSJeff Mahoney wait_current_trans_commit_start_and_unblock(fs_info, cur_trans); 1932bb9c12c9SSage Weil else 19332ff7e61eSJeff Mahoney wait_current_trans_commit_start(fs_info, cur_trans); 1934bb9c12c9SSage Weil 193538e88054SSage Weil if (current->journal_info == trans) 193638e88054SSage Weil current->journal_info = NULL; 193738e88054SSage Weil 1938724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1939bb9c12c9SSage Weil return 0; 1940bb9c12c9SSage Weil } 1941bb9c12c9SSage Weil 194249b25e05SJeff Mahoney 194397cb39bbSNikolay Borisov static void cleanup_transaction(struct btrfs_trans_handle *trans, int err) 194449b25e05SJeff Mahoney { 194597cb39bbSNikolay Borisov struct btrfs_fs_info *fs_info = trans->fs_info; 194649b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 194749b25e05SJeff Mahoney 1948b50fff81SDavid Sterba WARN_ON(refcount_read(&trans->use_count) > 1); 194949b25e05SJeff Mahoney 195066642832SJeff Mahoney btrfs_abort_transaction(trans, err); 19517b8b92afSJosef Bacik 19520b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 195366b6135bSLiu Bo 195425d8c284SMiao Xie /* 195525d8c284SMiao Xie * If the transaction is removed from the list, it means this 195625d8c284SMiao Xie * transaction has been committed successfully, so it is impossible 195725d8c284SMiao Xie * to call the cleanup function. 195825d8c284SMiao Xie */ 195925d8c284SMiao Xie BUG_ON(list_empty(&cur_trans->list)); 196066b6135bSLiu Bo 196149b25e05SJeff Mahoney list_del_init(&cur_trans->list); 19620b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) { 19634a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 19640b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1965f094ac32SLiu Bo wait_event(cur_trans->writer_wait, 1966f094ac32SLiu Bo atomic_read(&cur_trans->num_writers) == 1); 1967f094ac32SLiu Bo 19680b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1969d7096fc3SJosef Bacik } 19700b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 197149b25e05SJeff Mahoney 19722ff7e61eSJeff Mahoney btrfs_cleanup_one_transaction(trans->transaction, fs_info); 197349b25e05SJeff Mahoney 19740b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 19750b246afaSJeff Mahoney if (cur_trans == fs_info->running_transaction) 19760b246afaSJeff Mahoney fs_info->running_transaction = NULL; 19770b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 19784a9d8bdeSMiao Xie 1979e0228285SJosef Bacik if (trans->type & __TRANS_FREEZABLE) 19800b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 1981724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 1982724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 198349b25e05SJeff Mahoney 198497cb39bbSNikolay Borisov trace_btrfs_transaction_commit(trans->root); 198549b25e05SJeff Mahoney 198649b25e05SJeff Mahoney if (current->journal_info == trans) 198749b25e05SJeff Mahoney current->journal_info = NULL; 19880b246afaSJeff Mahoney btrfs_scrub_cancel(fs_info); 198949b25e05SJeff Mahoney 199049b25e05SJeff Mahoney kmem_cache_free(btrfs_trans_handle_cachep, trans); 199149b25e05SJeff Mahoney } 199249b25e05SJeff Mahoney 1993c7cc64a9SDavid Sterba /* 1994c7cc64a9SDavid Sterba * Release reserved delayed ref space of all pending block groups of the 1995c7cc64a9SDavid Sterba * transaction and remove them from the list 1996c7cc64a9SDavid Sterba */ 1997c7cc64a9SDavid Sterba static void btrfs_cleanup_pending_block_groups(struct btrfs_trans_handle *trans) 1998c7cc64a9SDavid Sterba { 1999c7cc64a9SDavid Sterba struct btrfs_fs_info *fs_info = trans->fs_info; 200032da5386SDavid Sterba struct btrfs_block_group *block_group, *tmp; 2001c7cc64a9SDavid Sterba 2002c7cc64a9SDavid Sterba list_for_each_entry_safe(block_group, tmp, &trans->new_bgs, bg_list) { 2003c7cc64a9SDavid Sterba btrfs_delayed_refs_rsv_release(fs_info, 1); 2004c7cc64a9SDavid Sterba list_del_init(&block_group->bg_list); 2005c7cc64a9SDavid Sterba } 2006c7cc64a9SDavid Sterba } 2007c7cc64a9SDavid Sterba 200888090ad3SFilipe Manana static inline int btrfs_start_delalloc_flush(struct btrfs_fs_info *fs_info) 200982436617SMiao Xie { 2010ce8ea7ccSJosef Bacik /* 2011ce8ea7ccSJosef Bacik * We use writeback_inodes_sb here because if we used 2012ce8ea7ccSJosef Bacik * btrfs_start_delalloc_roots we would deadlock with fs freeze. 2013ce8ea7ccSJosef Bacik * Currently are holding the fs freeze lock, if we do an async flush 2014ce8ea7ccSJosef Bacik * we'll do btrfs_join_transaction() and deadlock because we need to 2015ce8ea7ccSJosef Bacik * wait for the fs freeze lock. Using the direct flushing we benefit 2016ce8ea7ccSJosef Bacik * from already being in a transaction and our join_transaction doesn't 2017ce8ea7ccSJosef Bacik * have to re-take the fs freeze lock. 2018ce8ea7ccSJosef Bacik */ 201988090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 2020ce8ea7ccSJosef Bacik writeback_inodes_sb(fs_info->sb, WB_REASON_SYNC); 202182436617SMiao Xie return 0; 202282436617SMiao Xie } 202382436617SMiao Xie 202488090ad3SFilipe Manana static inline void btrfs_wait_delalloc_flush(struct btrfs_fs_info *fs_info) 202582436617SMiao Xie { 202688090ad3SFilipe Manana if (btrfs_test_opt(fs_info, FLUSHONCOMMIT)) 20276374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 0, (u64)-1); 202882436617SMiao Xie } 202982436617SMiao Xie 20303a45bb20SJeff Mahoney int btrfs_commit_transaction(struct btrfs_trans_handle *trans) 203179154b1bSChris Mason { 20323a45bb20SJeff Mahoney struct btrfs_fs_info *fs_info = trans->fs_info; 203349b25e05SJeff Mahoney struct btrfs_transaction *cur_trans = trans->transaction; 20348fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 203525287e0aSMiao Xie int ret; 203679154b1bSChris Mason 203735b814f3SNikolay Borisov ASSERT(refcount_read(&trans->use_count) == 1); 203835b814f3SNikolay Borisov 2039d62b23c9SJosef Bacik /* 2040d62b23c9SJosef Bacik * Some places just start a transaction to commit it. We need to make 2041d62b23c9SJosef Bacik * sure that if this commit fails that the abort code actually marks the 2042d62b23c9SJosef Bacik * transaction as failed, so set trans->dirty to make the abort code do 2043d62b23c9SJosef Bacik * the right thing. 2044d62b23c9SJosef Bacik */ 2045d62b23c9SJosef Bacik trans->dirty = true; 2046d62b23c9SJosef Bacik 20478d25a086SMiao Xie /* Stop the commit early if ->aborted is set */ 2048bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 204925287e0aSMiao Xie ret = cur_trans->aborted; 20503a45bb20SJeff Mahoney btrfs_end_transaction(trans); 2051e4a2bcacSJosef Bacik return ret; 205225287e0aSMiao Xie } 205349b25e05SJeff Mahoney 2054f45c752bSJosef Bacik btrfs_trans_release_metadata(trans); 2055f45c752bSJosef Bacik trans->block_rsv = NULL; 2056f45c752bSJosef Bacik 2057e19eb11fSJosef Bacik /* 2058e19eb11fSJosef Bacik * We only want one transaction commit doing the flushing so we do not 2059e19eb11fSJosef Bacik * waste a bunch of time on lock contention on the extent root node. 2060e19eb11fSJosef Bacik */ 2061e19eb11fSJosef Bacik if (!test_and_set_bit(BTRFS_DELAYED_REFS_FLUSHING, 2062e19eb11fSJosef Bacik &cur_trans->delayed_refs.flags)) { 2063e19eb11fSJosef Bacik /* 2064e19eb11fSJosef Bacik * Make a pass through all the delayed refs we have so far. 2065e19eb11fSJosef Bacik * Any running threads may add more while we are here. 206656bec294SChris Mason */ 2067c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, 0); 2068e4a2bcacSJosef Bacik if (ret) { 20693a45bb20SJeff Mahoney btrfs_end_transaction(trans); 2070e4a2bcacSJosef Bacik return ret; 2071e4a2bcacSJosef Bacik } 2072e19eb11fSJosef Bacik } 207356bec294SChris Mason 20746c686b35SNikolay Borisov btrfs_create_pending_block_groups(trans); 2075ea658badSJosef Bacik 20763204d33cSJosef Bacik if (!test_bit(BTRFS_TRANS_DIRTY_BG_RUN, &cur_trans->flags)) { 20771bbc621eSChris Mason int run_it = 0; 20781bbc621eSChris Mason 20791bbc621eSChris Mason /* this mutex is also taken before trying to set 20801bbc621eSChris Mason * block groups readonly. We need to make sure 20811bbc621eSChris Mason * that nobody has set a block group readonly 20821bbc621eSChris Mason * after a extents from that block group have been 20831bbc621eSChris Mason * allocated for cache files. btrfs_set_block_group_ro 20841bbc621eSChris Mason * will wait for the transaction to commit if it 20853204d33cSJosef Bacik * finds BTRFS_TRANS_DIRTY_BG_RUN set. 20861bbc621eSChris Mason * 20873204d33cSJosef Bacik * The BTRFS_TRANS_DIRTY_BG_RUN flag is also used to make sure 20883204d33cSJosef Bacik * only one process starts all the block group IO. It wouldn't 20891bbc621eSChris Mason * hurt to have more than one go through, but there's no 20901bbc621eSChris Mason * real advantage to it either. 20911bbc621eSChris Mason */ 20920b246afaSJeff Mahoney mutex_lock(&fs_info->ro_block_group_mutex); 20933204d33cSJosef Bacik if (!test_and_set_bit(BTRFS_TRANS_DIRTY_BG_RUN, 20943204d33cSJosef Bacik &cur_trans->flags)) 20951bbc621eSChris Mason run_it = 1; 20960b246afaSJeff Mahoney mutex_unlock(&fs_info->ro_block_group_mutex); 20971bbc621eSChris Mason 2098f9cacae3SNikolay Borisov if (run_it) { 209921217054SNikolay Borisov ret = btrfs_start_dirty_block_groups(trans); 21001bbc621eSChris Mason if (ret) { 21013a45bb20SJeff Mahoney btrfs_end_transaction(trans); 21021bbc621eSChris Mason return ret; 21031bbc621eSChris Mason } 2104f9cacae3SNikolay Borisov } 2105f9cacae3SNikolay Borisov } 21061bbc621eSChris Mason 21070b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 21084a9d8bdeSMiao Xie if (cur_trans->state >= TRANS_STATE_COMMIT_START) { 2109*d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2110*d0c2f4faSFilipe Manana 21110b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 21129b64f57dSElena Reshetova refcount_inc(&cur_trans->use_count); 2113ccd467d6SChris Mason 2114*d0c2f4faSFilipe Manana if (trans->in_fsync) 2115*d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2116*d0c2f4faSFilipe Manana ret = btrfs_end_transaction(trans); 2117*d0c2f4faSFilipe Manana wait_for_commit(cur_trans, want_state); 211815ee9bc7SJosef Bacik 2119bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) 2120b4924a0fSLiu Bo ret = cur_trans->aborted; 2121b4924a0fSLiu Bo 2122724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 212315ee9bc7SJosef Bacik 212449b25e05SJeff Mahoney return ret; 212579154b1bSChris Mason } 21264313b399SChris Mason 21274a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_START; 21280b246afaSJeff Mahoney wake_up(&fs_info->transaction_blocked_wait); 2129bb9c12c9SSage Weil 21300b246afaSJeff Mahoney if (cur_trans->list.prev != &fs_info->trans_list) { 2131*d0c2f4faSFilipe Manana enum btrfs_trans_state want_state = TRANS_STATE_COMPLETED; 2132*d0c2f4faSFilipe Manana 2133*d0c2f4faSFilipe Manana if (trans->in_fsync) 2134*d0c2f4faSFilipe Manana want_state = TRANS_STATE_SUPER_COMMITTED; 2135*d0c2f4faSFilipe Manana 2136ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 2137ccd467d6SChris Mason struct btrfs_transaction, list); 2138*d0c2f4faSFilipe Manana if (prev_trans->state < want_state) { 21399b64f57dSElena Reshetova refcount_inc(&prev_trans->use_count); 21400b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2141ccd467d6SChris Mason 2142*d0c2f4faSFilipe Manana wait_for_commit(prev_trans, want_state); 2143*d0c2f4faSFilipe Manana 2144bf31f87fSDavid Sterba ret = READ_ONCE(prev_trans->aborted); 2145ccd467d6SChris Mason 2146724e2315SJosef Bacik btrfs_put_transaction(prev_trans); 21471f9b8c8fSFilipe Manana if (ret) 21481f9b8c8fSFilipe Manana goto cleanup_transaction; 2149a4abeea4SJosef Bacik } else { 21500b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2151ccd467d6SChris Mason } 2152a4abeea4SJosef Bacik } else { 21530b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2154cb2d3dadSFilipe Manana /* 2155cb2d3dadSFilipe Manana * The previous transaction was aborted and was already removed 2156cb2d3dadSFilipe Manana * from the list of transactions at fs_info->trans_list. So we 2157cb2d3dadSFilipe Manana * abort to prevent writing a new superblock that reflects a 2158cb2d3dadSFilipe Manana * corrupt state (pointing to trees with unwritten nodes/leafs). 2159cb2d3dadSFilipe Manana */ 2160cb2d3dadSFilipe Manana if (test_bit(BTRFS_FS_STATE_TRANS_ABORTED, &fs_info->fs_state)) { 2161cb2d3dadSFilipe Manana ret = -EROFS; 2162cb2d3dadSFilipe Manana goto cleanup_transaction; 2163cb2d3dadSFilipe Manana } 2164ccd467d6SChris Mason } 216515ee9bc7SJosef Bacik 21660860adfdSMiao Xie extwriter_counter_dec(cur_trans, trans->type); 21670860adfdSMiao Xie 216888090ad3SFilipe Manana ret = btrfs_start_delalloc_flush(fs_info); 216982436617SMiao Xie if (ret) 217082436617SMiao Xie goto cleanup_transaction; 217182436617SMiao Xie 2172e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 217349b25e05SJeff Mahoney if (ret) 217449b25e05SJeff Mahoney goto cleanup_transaction; 217516cdcec7SMiao Xie 2176581227d0SMiao Xie wait_event(cur_trans->writer_wait, 2177581227d0SMiao Xie extwriter_counter_read(cur_trans) == 0); 2178ed3b3d31SChris Mason 2179581227d0SMiao Xie /* some pending stuffs might be added after the previous flush. */ 2180e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 2181ca469637SMiao Xie if (ret) 2182ca469637SMiao Xie goto cleanup_transaction; 2183ca469637SMiao Xie 218488090ad3SFilipe Manana btrfs_wait_delalloc_flush(fs_info); 2185cb7ab021SWang Shilong 218648778179SFilipe Manana /* 218748778179SFilipe Manana * Wait for all ordered extents started by a fast fsync that joined this 218848778179SFilipe Manana * transaction. Otherwise if this transaction commits before the ordered 218948778179SFilipe Manana * extents complete we lose logged data after a power failure. 219048778179SFilipe Manana */ 219148778179SFilipe Manana wait_event(cur_trans->pending_wait, 219248778179SFilipe Manana atomic_read(&cur_trans->pending_ordered) == 0); 219348778179SFilipe Manana 21942ff7e61eSJeff Mahoney btrfs_scrub_pause(fs_info); 2195ed0ca140SJosef Bacik /* 2196ed0ca140SJosef Bacik * Ok now we need to make sure to block out any other joins while we 2197ed0ca140SJosef Bacik * commit the transaction. We could have started a join before setting 21984a9d8bdeSMiao Xie * COMMIT_DOING so make sure to wait for num_writers to == 1 again. 2199ed0ca140SJosef Bacik */ 22000b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 22014a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMMIT_DOING; 22020b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2203ed0ca140SJosef Bacik wait_event(cur_trans->writer_wait, 2204ed0ca140SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 220515ee9bc7SJosef Bacik 2206bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 22072cba30f1SMiao Xie ret = cur_trans->aborted; 22086cf7f77eSWang Shilong goto scrub_continue; 22092cba30f1SMiao Xie } 22107585717fSChris Mason /* 22117585717fSChris Mason * the reloc mutex makes sure that we stop 22127585717fSChris Mason * the balancing code from coming in and moving 22137585717fSChris Mason * extents around in the middle of the commit 22147585717fSChris Mason */ 22150b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 22167585717fSChris Mason 221742874b3dSMiao Xie /* 221842874b3dSMiao Xie * We needn't worry about the delayed items because we will 221942874b3dSMiao Xie * deal with them in create_pending_snapshot(), which is the 222042874b3dSMiao Xie * core function of the snapshot creation. 222142874b3dSMiao Xie */ 222208d50ca3SNikolay Borisov ret = create_pending_snapshots(trans); 222356e9f6eaSDavid Sterba if (ret) 222456e9f6eaSDavid Sterba goto unlock_reloc; 22253063d29fSChris Mason 222642874b3dSMiao Xie /* 222742874b3dSMiao Xie * We insert the dir indexes of the snapshots and update the inode 222842874b3dSMiao Xie * of the snapshots' parents after the snapshot creation, so there 222942874b3dSMiao Xie * are some delayed items which are not dealt with. Now deal with 223042874b3dSMiao Xie * them. 223142874b3dSMiao Xie * 223242874b3dSMiao Xie * We needn't worry that this operation will corrupt the snapshots, 223342874b3dSMiao Xie * because all the tree which are snapshoted will be forced to COW 223442874b3dSMiao Xie * the nodes and leaves. 223542874b3dSMiao Xie */ 2236e5c304e6SNikolay Borisov ret = btrfs_run_delayed_items(trans); 223756e9f6eaSDavid Sterba if (ret) 223856e9f6eaSDavid Sterba goto unlock_reloc; 223916cdcec7SMiao Xie 2240c79a70b1SNikolay Borisov ret = btrfs_run_delayed_refs(trans, (unsigned long)-1); 224156e9f6eaSDavid Sterba if (ret) 224256e9f6eaSDavid Sterba goto unlock_reloc; 224356bec294SChris Mason 2244e999376fSChris Mason /* 2245e999376fSChris Mason * make sure none of the code above managed to slip in a 2246e999376fSChris Mason * delayed item 2247e999376fSChris Mason */ 2248ccdf9b30SJeff Mahoney btrfs_assert_delayed_root_empty(fs_info); 2249e999376fSChris Mason 22502c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 2251dc17ff8fSChris Mason 2252e02119d5SChris Mason /* btrfs_commit_tree_roots is responsible for getting the 2253e02119d5SChris Mason * various roots consistent with each other. Every pointer 2254e02119d5SChris Mason * in the tree of tree roots has to point to the most up to date 2255e02119d5SChris Mason * root for every subvolume and other tree. So, we have to keep 2256e02119d5SChris Mason * the tree logging code from jumping in and changing any 2257e02119d5SChris Mason * of the trees. 2258e02119d5SChris Mason * 2259e02119d5SChris Mason * At this point in the commit, there can't be any tree-log 2260e02119d5SChris Mason * writers, but a little lower down we drop the trans mutex 2261e02119d5SChris Mason * and let new people in. By holding the tree_log_mutex 2262e02119d5SChris Mason * from now until after the super is written, we avoid races 2263e02119d5SChris Mason * with the tree-log code. 2264e02119d5SChris Mason */ 22650b246afaSJeff Mahoney mutex_lock(&fs_info->tree_log_mutex); 22661a40e23bSZheng Yan 22677e4443d9SNikolay Borisov ret = commit_fs_roots(trans); 226856e9f6eaSDavid Sterba if (ret) 226956e9f6eaSDavid Sterba goto unlock_tree_log; 227054aa1f4dSChris Mason 22713818aea2SQu Wenruo /* 22727e1876acSDavid Sterba * Since the transaction is done, we can apply the pending changes 22737e1876acSDavid Sterba * before the next transaction. 22743818aea2SQu Wenruo */ 22750b246afaSJeff Mahoney btrfs_apply_pending_changes(fs_info); 22763818aea2SQu Wenruo 22775d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 2278e02119d5SChris Mason * safe to free the root of tree log roots 2279e02119d5SChris Mason */ 22800b246afaSJeff Mahoney btrfs_free_log_root_tree(trans, fs_info); 2281e02119d5SChris Mason 22820ed4792aSQu Wenruo /* 22830ed4792aSQu Wenruo * Since fs roots are all committed, we can get a quite accurate 22840ed4792aSQu Wenruo * new_roots. So let's do quota accounting. 22850ed4792aSQu Wenruo */ 2286460fb20aSNikolay Borisov ret = btrfs_qgroup_account_extents(trans); 228756e9f6eaSDavid Sterba if (ret < 0) 228856e9f6eaSDavid Sterba goto unlock_tree_log; 22890ed4792aSQu Wenruo 22909386d8bcSNikolay Borisov ret = commit_cowonly_roots(trans); 229156e9f6eaSDavid Sterba if (ret) 229256e9f6eaSDavid Sterba goto unlock_tree_log; 229354aa1f4dSChris Mason 22942cba30f1SMiao Xie /* 22952cba30f1SMiao Xie * The tasks which save the space cache and inode cache may also 22962cba30f1SMiao Xie * update ->aborted, check it. 22972cba30f1SMiao Xie */ 2298bf31f87fSDavid Sterba if (TRANS_ABORTED(cur_trans)) { 22992cba30f1SMiao Xie ret = cur_trans->aborted; 230056e9f6eaSDavid Sterba goto unlock_tree_log; 23012cba30f1SMiao Xie } 23022cba30f1SMiao Xie 23030b246afaSJeff Mahoney cur_trans = fs_info->running_transaction; 23045f39d397SChris Mason 23050b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->tree_root->root_item, 23060b246afaSJeff Mahoney fs_info->tree_root->node); 23070b246afaSJeff Mahoney list_add_tail(&fs_info->tree_root->dirty_list, 23089e351cc8SJosef Bacik &cur_trans->switch_commits); 23095d4f98a2SYan Zheng 23100b246afaSJeff Mahoney btrfs_set_root_node(&fs_info->chunk_root->root_item, 23110b246afaSJeff Mahoney fs_info->chunk_root->node); 23120b246afaSJeff Mahoney list_add_tail(&fs_info->chunk_root->dirty_list, 23139e351cc8SJosef Bacik &cur_trans->switch_commits); 23149e351cc8SJosef Bacik 2315889bfa39SJosef Bacik switch_commit_roots(trans); 23165d4f98a2SYan Zheng 2317ce93ec54SJosef Bacik ASSERT(list_empty(&cur_trans->dirty_bgs)); 23181bbc621eSChris Mason ASSERT(list_empty(&cur_trans->io_bgs)); 23192ff7e61eSJeff Mahoney update_super_roots(fs_info); 2320e02119d5SChris Mason 23210b246afaSJeff Mahoney btrfs_set_super_log_root(fs_info->super_copy, 0); 23220b246afaSJeff Mahoney btrfs_set_super_log_root_level(fs_info->super_copy, 0); 23230b246afaSJeff Mahoney memcpy(fs_info->super_for_commit, fs_info->super_copy, 23240b246afaSJeff Mahoney sizeof(*fs_info->super_copy)); 2325ccd467d6SChris Mason 2326bbbf7243SNikolay Borisov btrfs_commit_device_sizes(cur_trans); 2327935e5cc9SMiao Xie 23280b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG1_ERR, &fs_info->flags); 23290b246afaSJeff Mahoney clear_bit(BTRFS_FS_LOG2_ERR, &fs_info->flags); 2330656f30dbSFilipe Manana 23314fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 23324fbcdf66SFilipe Manana 23330b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 23344a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_UNBLOCKED; 23350b246afaSJeff Mahoney fs_info->running_transaction = NULL; 23360b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 23370b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2338b7ec40d7SChris Mason 23390b246afaSJeff Mahoney wake_up(&fs_info->transaction_wait); 2340e6dcd2dcSChris Mason 234170458a58SNikolay Borisov ret = btrfs_write_and_wait_transaction(trans); 234249b25e05SJeff Mahoney if (ret) { 23430b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, 234408748810SDavid Sterba "Error while writing out transaction"); 234556e9f6eaSDavid Sterba /* 234656e9f6eaSDavid Sterba * reloc_mutex has been unlocked, tree_log_mutex is still held 234756e9f6eaSDavid Sterba * but we can't jump to unlock_tree_log causing double unlock 234856e9f6eaSDavid Sterba */ 23490b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 23506cf7f77eSWang Shilong goto scrub_continue; 235149b25e05SJeff Mahoney } 235249b25e05SJeff Mahoney 2353eece6a9cSDavid Sterba ret = write_all_supers(fs_info, 0); 2354e02119d5SChris Mason /* 2355e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 2356e02119d5SChris Mason * to go about their business 2357e02119d5SChris Mason */ 23580b246afaSJeff Mahoney mutex_unlock(&fs_info->tree_log_mutex); 2359c1f32b7cSAnand Jain if (ret) 2360c1f32b7cSAnand Jain goto scrub_continue; 2361e02119d5SChris Mason 2362*d0c2f4faSFilipe Manana /* 2363*d0c2f4faSFilipe Manana * We needn't acquire the lock here because there is no other task 2364*d0c2f4faSFilipe Manana * which can change it. 2365*d0c2f4faSFilipe Manana */ 2366*d0c2f4faSFilipe Manana cur_trans->state = TRANS_STATE_SUPER_COMMITTED; 2367*d0c2f4faSFilipe Manana wake_up(&cur_trans->commit_wait); 2368*d0c2f4faSFilipe Manana 23695ead2dd0SNikolay Borisov btrfs_finish_extent_commit(trans); 23704313b399SChris Mason 23713204d33cSJosef Bacik if (test_bit(BTRFS_TRANS_HAVE_FREE_BGS, &cur_trans->flags)) 23720b246afaSJeff Mahoney btrfs_clear_space_info_full(fs_info); 237313212b54SZhao Lei 23740b246afaSJeff Mahoney fs_info->last_trans_committed = cur_trans->transid; 23754a9d8bdeSMiao Xie /* 23764a9d8bdeSMiao Xie * We needn't acquire the lock here because there is no other task 23774a9d8bdeSMiao Xie * which can change it. 23784a9d8bdeSMiao Xie */ 23794a9d8bdeSMiao Xie cur_trans->state = TRANS_STATE_COMPLETED; 23802c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 23813de4586cSChris Mason 23820b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 238313c5a93eSJosef Bacik list_del_init(&cur_trans->list); 23840b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 2385a4abeea4SJosef Bacik 2386724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 2387724e2315SJosef Bacik btrfs_put_transaction(cur_trans); 238858176a96SJosef Bacik 23890860adfdSMiao Xie if (trans->type & __TRANS_FREEZABLE) 23900b246afaSJeff Mahoney sb_end_intwrite(fs_info->sb); 2391b2b5ef5cSJan Kara 23923a45bb20SJeff Mahoney trace_btrfs_transaction_commit(trans->root); 23931abe9b8aSliubo 23942ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 2395a2de733cSArne Jansen 23969ed74f2dSJosef Bacik if (current->journal_info == trans) 23979ed74f2dSJosef Bacik current->journal_info = NULL; 23989ed74f2dSJosef Bacik 23992c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 240024bbcf04SYan, Zheng 240179154b1bSChris Mason return ret; 240249b25e05SJeff Mahoney 240356e9f6eaSDavid Sterba unlock_tree_log: 240456e9f6eaSDavid Sterba mutex_unlock(&fs_info->tree_log_mutex); 240556e9f6eaSDavid Sterba unlock_reloc: 240656e9f6eaSDavid Sterba mutex_unlock(&fs_info->reloc_mutex); 24076cf7f77eSWang Shilong scrub_continue: 24082ff7e61eSJeff Mahoney btrfs_scrub_continue(fs_info); 240949b25e05SJeff Mahoney cleanup_transaction: 2410dc60c525SNikolay Borisov btrfs_trans_release_metadata(trans); 2411c7cc64a9SDavid Sterba btrfs_cleanup_pending_block_groups(trans); 24124fbcdf66SFilipe Manana btrfs_trans_release_chunk_metadata(trans); 24130e721106SJosef Bacik trans->block_rsv = NULL; 24140b246afaSJeff Mahoney btrfs_warn(fs_info, "Skipping commit of aborted transaction."); 241549b25e05SJeff Mahoney if (current->journal_info == trans) 241649b25e05SJeff Mahoney current->journal_info = NULL; 241797cb39bbSNikolay Borisov cleanup_transaction(trans, ret); 241849b25e05SJeff Mahoney 241949b25e05SJeff Mahoney return ret; 242079154b1bSChris Mason } 242179154b1bSChris Mason 2422d352ac68SChris Mason /* 24239d1a2a3aSDavid Sterba * return < 0 if error 24249d1a2a3aSDavid Sterba * 0 if there are no more dead_roots at the time of call 24259d1a2a3aSDavid Sterba * 1 there are more to be processed, call me again 24269d1a2a3aSDavid Sterba * 24279d1a2a3aSDavid Sterba * The return value indicates there are certainly more snapshots to delete, but 24289d1a2a3aSDavid Sterba * if there comes a new one during processing, it may return 0. We don't mind, 24299d1a2a3aSDavid Sterba * because btrfs_commit_super will poke cleaner thread and it will process it a 24309d1a2a3aSDavid Sterba * few seconds later. 2431d352ac68SChris Mason */ 24329d1a2a3aSDavid Sterba int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root) 2433e9d0b13bSChris Mason { 24349d1a2a3aSDavid Sterba int ret; 24355d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = root->fs_info; 2436e9d0b13bSChris Mason 2437a4abeea4SJosef Bacik spin_lock(&fs_info->trans_lock); 24389d1a2a3aSDavid Sterba if (list_empty(&fs_info->dead_roots)) { 24399d1a2a3aSDavid Sterba spin_unlock(&fs_info->trans_lock); 24409d1a2a3aSDavid Sterba return 0; 24419d1a2a3aSDavid Sterba } 24429d1a2a3aSDavid Sterba root = list_first_entry(&fs_info->dead_roots, 24439d1a2a3aSDavid Sterba struct btrfs_root, root_list); 2444cfad392bSJosef Bacik list_del_init(&root->root_list); 2445a4abeea4SJosef Bacik spin_unlock(&fs_info->trans_lock); 24465d4f98a2SYan Zheng 24474fd786e6SMisono Tomohiro btrfs_debug(fs_info, "cleaner removing %llu", root->root_key.objectid); 244876dda93cSYan, Zheng 244916cdcec7SMiao Xie btrfs_kill_all_delayed_nodes(root); 245016cdcec7SMiao Xie 245176dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 245276dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 24530078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 0, 0); 245476dda93cSYan, Zheng else 24550078a9f9SNikolay Borisov ret = btrfs_drop_snapshot(root, 1, 0); 245632471dc2SDavid Sterba 2457dc9492c1SJosef Bacik btrfs_put_root(root); 24586596a928SJosef Bacik return (ret < 0) ? 0 : 1; 2459e9d0b13bSChris Mason } 2460572d9ab7SDavid Sterba 2461572d9ab7SDavid Sterba void btrfs_apply_pending_changes(struct btrfs_fs_info *fs_info) 2462572d9ab7SDavid Sterba { 2463572d9ab7SDavid Sterba unsigned long prev; 2464572d9ab7SDavid Sterba unsigned long bit; 2465572d9ab7SDavid Sterba 24666c9fe14fSQu Wenruo prev = xchg(&fs_info->pending_changes, 0); 2467572d9ab7SDavid Sterba if (!prev) 2468572d9ab7SDavid Sterba return; 2469572d9ab7SDavid Sterba 2470d51033d0SDavid Sterba bit = 1 << BTRFS_PENDING_COMMIT; 2471d51033d0SDavid Sterba if (prev & bit) 2472d51033d0SDavid Sterba btrfs_debug(fs_info, "pending commit done"); 2473d51033d0SDavid Sterba prev &= ~bit; 2474d51033d0SDavid Sterba 2475572d9ab7SDavid Sterba if (prev) 2476572d9ab7SDavid Sterba btrfs_warn(fs_info, 2477572d9ab7SDavid Sterba "unknown pending changes left 0x%lx, ignoring", prev); 2478572d9ab7SDavid Sterba } 2479