16cbd5570SChris Mason /* 26cbd5570SChris Mason * Copyright (C) 2007 Oracle. All rights reserved. 36cbd5570SChris Mason * 46cbd5570SChris Mason * This program is free software; you can redistribute it and/or 56cbd5570SChris Mason * modify it under the terms of the GNU General Public 66cbd5570SChris Mason * License v2 as published by the Free Software Foundation. 76cbd5570SChris Mason * 86cbd5570SChris Mason * This program is distributed in the hope that it will be useful, 96cbd5570SChris Mason * but WITHOUT ANY WARRANTY; without even the implied warranty of 106cbd5570SChris Mason * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 116cbd5570SChris Mason * General Public License for more details. 126cbd5570SChris Mason * 136cbd5570SChris Mason * You should have received a copy of the GNU General Public 146cbd5570SChris Mason * License along with this program; if not, write to the 156cbd5570SChris Mason * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 166cbd5570SChris Mason * Boston, MA 021110-1307, USA. 176cbd5570SChris Mason */ 186cbd5570SChris Mason 1979154b1bSChris Mason #include <linux/fs.h> 205a0e3ad6STejun Heo #include <linux/slab.h> 2134088780SChris Mason #include <linux/sched.h> 22d3c2fdcfSChris Mason #include <linux/writeback.h> 235f39d397SChris Mason #include <linux/pagemap.h> 245f2cc086SChris Mason #include <linux/blkdev.h> 2579154b1bSChris Mason #include "ctree.h" 2679154b1bSChris Mason #include "disk-io.h" 2779154b1bSChris Mason #include "transaction.h" 28925baeddSChris Mason #include "locking.h" 29e02119d5SChris Mason #include "tree-log.h" 30581bb050SLi Zefan #include "inode-map.h" 3179154b1bSChris Mason 320f7d52f4SChris Mason #define BTRFS_ROOT_TRANS_TAG 0 330f7d52f4SChris Mason 3480b6794dSChris Mason static noinline void put_transaction(struct btrfs_transaction *transaction) 3579154b1bSChris Mason { 3613c5a93eSJosef Bacik WARN_ON(atomic_read(&transaction->use_count) == 0); 3713c5a93eSJosef Bacik if (atomic_dec_and_test(&transaction->use_count)) { 38a4abeea4SJosef Bacik BUG_ON(!list_empty(&transaction->list)); 392c90e5d6SChris Mason memset(transaction, 0, sizeof(*transaction)); 402c90e5d6SChris Mason kmem_cache_free(btrfs_transaction_cachep, transaction); 4179154b1bSChris Mason } 4278fae27eSChris Mason } 4379154b1bSChris Mason 44817d52f8SJosef Bacik static noinline void switch_commit_root(struct btrfs_root *root) 45817d52f8SJosef Bacik { 46817d52f8SJosef Bacik free_extent_buffer(root->commit_root); 47817d52f8SJosef Bacik root->commit_root = btrfs_root_node(root); 48817d52f8SJosef Bacik } 49817d52f8SJosef Bacik 50d352ac68SChris Mason /* 51d352ac68SChris Mason * either allocate a new transaction or hop into the existing one 52d352ac68SChris Mason */ 53a4abeea4SJosef Bacik static noinline int join_transaction(struct btrfs_root *root, int nofail) 5479154b1bSChris Mason { 5579154b1bSChris Mason struct btrfs_transaction *cur_trans; 56a4abeea4SJosef Bacik 57a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 58d43317dcSChris Mason loop: 59a4abeea4SJosef Bacik if (root->fs_info->trans_no_join) { 60a4abeea4SJosef Bacik if (!nofail) { 61a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 62a4abeea4SJosef Bacik return -EBUSY; 63a4abeea4SJosef Bacik } 64a4abeea4SJosef Bacik } 65a4abeea4SJosef Bacik 6679154b1bSChris Mason cur_trans = root->fs_info->running_transaction; 67a4abeea4SJosef Bacik if (cur_trans) { 68a4abeea4SJosef Bacik atomic_inc(&cur_trans->use_count); 69a4abeea4SJosef Bacik atomic_inc(&cur_trans->num_writers); 70a4abeea4SJosef Bacik cur_trans->num_joined++; 71a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 72a4abeea4SJosef Bacik return 0; 73a4abeea4SJosef Bacik } 74a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 75a4abeea4SJosef Bacik 76a4abeea4SJosef Bacik cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); 77db5b493aSTsutomu Itoh if (!cur_trans) 78db5b493aSTsutomu Itoh return -ENOMEM; 79d43317dcSChris Mason 80a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 81a4abeea4SJosef Bacik if (root->fs_info->running_transaction) { 82d43317dcSChris Mason /* 83d43317dcSChris Mason * someone started a transaction after we unlocked. Make sure 84d43317dcSChris Mason * to redo the trans_no_join checks above 85d43317dcSChris Mason */ 86a4abeea4SJosef Bacik kmem_cache_free(btrfs_transaction_cachep, cur_trans); 87a4abeea4SJosef Bacik cur_trans = root->fs_info->running_transaction; 88d43317dcSChris Mason goto loop; 89a4abeea4SJosef Bacik } 90d43317dcSChris Mason 9113c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 9215ee9bc7SJosef Bacik cur_trans->num_joined = 0; 9379154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 9479154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 9579154b1bSChris Mason cur_trans->in_commit = 0; 96f9295749SChris Mason cur_trans->blocked = 0; 97a4abeea4SJosef Bacik /* 98a4abeea4SJosef Bacik * One for this trans handle, one so it will live on until we 99a4abeea4SJosef Bacik * commit the transaction. 100a4abeea4SJosef Bacik */ 101a4abeea4SJosef Bacik atomic_set(&cur_trans->use_count, 2); 10279154b1bSChris Mason cur_trans->commit_done = 0; 10308607c1bSChris Mason cur_trans->start_time = get_seconds(); 10456bec294SChris Mason 1056bef4d31SEric Paris cur_trans->delayed_refs.root = RB_ROOT; 10656bec294SChris Mason cur_trans->delayed_refs.num_entries = 0; 107c3e69d58SChris Mason cur_trans->delayed_refs.num_heads_ready = 0; 108c3e69d58SChris Mason cur_trans->delayed_refs.num_heads = 0; 10956bec294SChris Mason cur_trans->delayed_refs.flushing = 0; 110c3e69d58SChris Mason cur_trans->delayed_refs.run_delayed_start = 0; 111a4abeea4SJosef Bacik spin_lock_init(&cur_trans->commit_lock); 11256bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 11356bec294SChris Mason 1143063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 1158fd17795SChris Mason list_add_tail(&cur_trans->list, &root->fs_info->trans_list); 116d1310b2eSChris Mason extent_io_tree_init(&cur_trans->dirty_pages, 117f993c883SDavid Sterba root->fs_info->btree_inode->i_mapping); 118a4abeea4SJosef Bacik root->fs_info->generation++; 119a4abeea4SJosef Bacik cur_trans->transid = root->fs_info->generation; 12048ec2cf8SChris Mason root->fs_info->running_transaction = cur_trans; 121a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 12215ee9bc7SJosef Bacik 12379154b1bSChris Mason return 0; 12479154b1bSChris Mason } 12579154b1bSChris Mason 126d352ac68SChris Mason /* 127d397712bSChris Mason * this does all the record keeping required to make sure that a reference 128d397712bSChris Mason * counted root is properly recorded in a given transaction. This is required 129d397712bSChris Mason * to make sure the old root from before we joined the transaction is deleted 130d397712bSChris Mason * when the transaction commits 131d352ac68SChris Mason */ 1327585717fSChris Mason static int record_root_in_trans(struct btrfs_trans_handle *trans, 1335d4f98a2SYan Zheng struct btrfs_root *root) 1346702ed49SChris Mason { 1355d4f98a2SYan Zheng if (root->ref_cows && root->last_trans < trans->transid) { 1366702ed49SChris Mason WARN_ON(root == root->fs_info->extent_root); 1375d4f98a2SYan Zheng WARN_ON(root->commit_root != root->node); 1385d4f98a2SYan Zheng 1397585717fSChris Mason /* 1407585717fSChris Mason * see below for in_trans_setup usage rules 1417585717fSChris Mason * we have the reloc mutex held now, so there 1427585717fSChris Mason * is only one writer in this function 1437585717fSChris Mason */ 1447585717fSChris Mason root->in_trans_setup = 1; 1457585717fSChris Mason 1467585717fSChris Mason /* make sure readers find in_trans_setup before 1477585717fSChris Mason * they find our root->last_trans update 1487585717fSChris Mason */ 1497585717fSChris Mason smp_wmb(); 1507585717fSChris Mason 151a4abeea4SJosef Bacik spin_lock(&root->fs_info->fs_roots_radix_lock); 152a4abeea4SJosef Bacik if (root->last_trans == trans->transid) { 153a4abeea4SJosef Bacik spin_unlock(&root->fs_info->fs_roots_radix_lock); 154a4abeea4SJosef Bacik return 0; 155a4abeea4SJosef Bacik } 1566702ed49SChris Mason radix_tree_tag_set(&root->fs_info->fs_roots_radix, 1576702ed49SChris Mason (unsigned long)root->root_key.objectid, 1586702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 159a4abeea4SJosef Bacik spin_unlock(&root->fs_info->fs_roots_radix_lock); 1607585717fSChris Mason root->last_trans = trans->transid; 1617585717fSChris Mason 1627585717fSChris Mason /* this is pretty tricky. We don't want to 1637585717fSChris Mason * take the relocation lock in btrfs_record_root_in_trans 1647585717fSChris Mason * unless we're really doing the first setup for this root in 1657585717fSChris Mason * this transaction. 1667585717fSChris Mason * 1677585717fSChris Mason * Normally we'd use root->last_trans as a flag to decide 1687585717fSChris Mason * if we want to take the expensive mutex. 1697585717fSChris Mason * 1707585717fSChris Mason * But, we have to set root->last_trans before we 1717585717fSChris Mason * init the relocation root, otherwise, we trip over warnings 1727585717fSChris Mason * in ctree.c. The solution used here is to flag ourselves 1737585717fSChris Mason * with root->in_trans_setup. When this is 1, we're still 1747585717fSChris Mason * fixing up the reloc trees and everyone must wait. 1757585717fSChris Mason * 1767585717fSChris Mason * When this is zero, they can trust root->last_trans and fly 1777585717fSChris Mason * through btrfs_record_root_in_trans without having to take the 1787585717fSChris Mason * lock. smp_wmb() makes sure that all the writes above are 1797585717fSChris Mason * done before we pop in the zero below 1807585717fSChris Mason */ 1815d4f98a2SYan Zheng btrfs_init_reloc_root(trans, root); 1827585717fSChris Mason smp_wmb(); 1837585717fSChris Mason root->in_trans_setup = 0; 1846702ed49SChris Mason } 1855d4f98a2SYan Zheng return 0; 1866702ed49SChris Mason } 1875d4f98a2SYan Zheng 1887585717fSChris Mason 1897585717fSChris Mason int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 1907585717fSChris Mason struct btrfs_root *root) 1917585717fSChris Mason { 1927585717fSChris Mason if (!root->ref_cows) 1937585717fSChris Mason return 0; 1947585717fSChris Mason 1957585717fSChris Mason /* 1967585717fSChris Mason * see record_root_in_trans for comments about in_trans_setup usage 1977585717fSChris Mason * and barriers 1987585717fSChris Mason */ 1997585717fSChris Mason smp_rmb(); 2007585717fSChris Mason if (root->last_trans == trans->transid && 2017585717fSChris Mason !root->in_trans_setup) 2027585717fSChris Mason return 0; 2037585717fSChris Mason 2047585717fSChris Mason mutex_lock(&root->fs_info->reloc_mutex); 2057585717fSChris Mason record_root_in_trans(trans, root); 2067585717fSChris Mason mutex_unlock(&root->fs_info->reloc_mutex); 2077585717fSChris Mason 2087585717fSChris Mason return 0; 2097585717fSChris Mason } 2107585717fSChris Mason 211d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 212d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 213d352ac68SChris Mason * transaction might not be fully on disk. 214d352ac68SChris Mason */ 21537d1aeeeSChris Mason static void wait_current_trans(struct btrfs_root *root) 21679154b1bSChris Mason { 217f9295749SChris Mason struct btrfs_transaction *cur_trans; 21879154b1bSChris Mason 219a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 220f9295749SChris Mason cur_trans = root->fs_info->running_transaction; 22137d1aeeeSChris Mason if (cur_trans && cur_trans->blocked) { 22213c5a93eSJosef Bacik atomic_inc(&cur_trans->use_count); 223a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 22472d63ed6SLi Zefan 22572d63ed6SLi Zefan wait_event(root->fs_info->transaction_wait, 22672d63ed6SLi Zefan !cur_trans->blocked); 227f9295749SChris Mason put_transaction(cur_trans); 228a4abeea4SJosef Bacik } else { 229a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 230f9295749SChris Mason } 23137d1aeeeSChris Mason } 23237d1aeeeSChris Mason 233249ac1e5SJosef Bacik enum btrfs_trans_type { 234249ac1e5SJosef Bacik TRANS_START, 235249ac1e5SJosef Bacik TRANS_JOIN, 236249ac1e5SJosef Bacik TRANS_USERSPACE, 2370af3d00bSJosef Bacik TRANS_JOIN_NOLOCK, 238249ac1e5SJosef Bacik }; 239249ac1e5SJosef Bacik 240a22285a6SYan, Zheng static int may_wait_transaction(struct btrfs_root *root, int type) 24137d1aeeeSChris Mason { 242a4abeea4SJosef Bacik if (root->fs_info->log_root_recovering) 243a4abeea4SJosef Bacik return 0; 244a4abeea4SJosef Bacik 245a4abeea4SJosef Bacik if (type == TRANS_USERSPACE) 246a22285a6SYan, Zheng return 1; 247a4abeea4SJosef Bacik 248a4abeea4SJosef Bacik if (type == TRANS_START && 249a4abeea4SJosef Bacik !atomic_read(&root->fs_info->open_ioctl_trans)) 250a4abeea4SJosef Bacik return 1; 251a4abeea4SJosef Bacik 252a22285a6SYan, Zheng return 0; 253a22285a6SYan, Zheng } 254a22285a6SYan, Zheng 255a22285a6SYan, Zheng static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, 256a22285a6SYan, Zheng u64 num_items, int type) 257a22285a6SYan, Zheng { 258a22285a6SYan, Zheng struct btrfs_trans_handle *h; 259a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 260b5009945SJosef Bacik u64 num_bytes = 0; 261a22285a6SYan, Zheng int ret; 262acce952bSliubo 263acce952bSliubo if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) 264acce952bSliubo return ERR_PTR(-EROFS); 2652a1eb461SJosef Bacik 2662a1eb461SJosef Bacik if (current->journal_info) { 2672a1eb461SJosef Bacik WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK); 2682a1eb461SJosef Bacik h = current->journal_info; 2692a1eb461SJosef Bacik h->use_count++; 2702a1eb461SJosef Bacik h->orig_rsv = h->block_rsv; 2712a1eb461SJosef Bacik h->block_rsv = NULL; 2722a1eb461SJosef Bacik goto got_it; 2732a1eb461SJosef Bacik } 274b5009945SJosef Bacik 275b5009945SJosef Bacik /* 276b5009945SJosef Bacik * Do the reservation before we join the transaction so we can do all 277b5009945SJosef Bacik * the appropriate flushing if need be. 278b5009945SJosef Bacik */ 279b5009945SJosef Bacik if (num_items > 0 && root != root->fs_info->chunk_root) { 280b5009945SJosef Bacik num_bytes = btrfs_calc_trans_metadata_size(root, num_items); 2814a92b1b8SJosef Bacik ret = btrfs_block_rsv_add(root, 282b5009945SJosef Bacik &root->fs_info->trans_block_rsv, 283b5009945SJosef Bacik num_bytes); 284b5009945SJosef Bacik if (ret) 285b5009945SJosef Bacik return ERR_PTR(ret); 286b5009945SJosef Bacik } 287a22285a6SYan, Zheng again: 288a22285a6SYan, Zheng h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); 289a22285a6SYan, Zheng if (!h) 290a22285a6SYan, Zheng return ERR_PTR(-ENOMEM); 291a22285a6SYan, Zheng 292a22285a6SYan, Zheng if (may_wait_transaction(root, type)) 29337d1aeeeSChris Mason wait_current_trans(root); 294a22285a6SYan, Zheng 295a4abeea4SJosef Bacik do { 296a4abeea4SJosef Bacik ret = join_transaction(root, type == TRANS_JOIN_NOLOCK); 297a4abeea4SJosef Bacik if (ret == -EBUSY) 298a4abeea4SJosef Bacik wait_current_trans(root); 299a4abeea4SJosef Bacik } while (ret == -EBUSY); 300a4abeea4SJosef Bacik 301db5b493aSTsutomu Itoh if (ret < 0) { 3026e8df2aeSYoshinori Sano kmem_cache_free(btrfs_trans_handle_cachep, h); 303db5b493aSTsutomu Itoh return ERR_PTR(ret); 304db5b493aSTsutomu Itoh } 3050f7d52f4SChris Mason 306a22285a6SYan, Zheng cur_trans = root->fs_info->running_transaction; 307a22285a6SYan, Zheng 308a22285a6SYan, Zheng h->transid = cur_trans->transid; 309a22285a6SYan, Zheng h->transaction = cur_trans; 31079154b1bSChris Mason h->blocks_used = 0; 311a22285a6SYan, Zheng h->bytes_reserved = 0; 31256bec294SChris Mason h->delayed_ref_updates = 0; 3132a1eb461SJosef Bacik h->use_count = 1; 314f0486c68SYan, Zheng h->block_rsv = NULL; 3152a1eb461SJosef Bacik h->orig_rsv = NULL; 316b7ec40d7SChris Mason 317a22285a6SYan, Zheng smp_mb(); 318a22285a6SYan, Zheng if (cur_trans->blocked && may_wait_transaction(root, type)) { 319a22285a6SYan, Zheng btrfs_commit_transaction(h, root); 320a22285a6SYan, Zheng goto again; 321a22285a6SYan, Zheng } 3229ed74f2dSJosef Bacik 323b5009945SJosef Bacik if (num_bytes) { 324b5009945SJosef Bacik h->block_rsv = &root->fs_info->trans_block_rsv; 325b5009945SJosef Bacik h->bytes_reserved = num_bytes; 326a22285a6SYan, Zheng } 327a22285a6SYan, Zheng 3282a1eb461SJosef Bacik got_it: 329a4abeea4SJosef Bacik btrfs_record_root_in_trans(h, root); 330a22285a6SYan, Zheng 331a22285a6SYan, Zheng if (!current->journal_info && type != TRANS_USERSPACE) 332a22285a6SYan, Zheng current->journal_info = h; 33379154b1bSChris Mason return h; 33479154b1bSChris Mason } 33579154b1bSChris Mason 336f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 337a22285a6SYan, Zheng int num_items) 338f9295749SChris Mason { 339a22285a6SYan, Zheng return start_transaction(root, num_items, TRANS_START); 340f9295749SChris Mason } 3417a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) 342f9295749SChris Mason { 343a22285a6SYan, Zheng return start_transaction(root, 0, TRANS_JOIN); 344f9295749SChris Mason } 345f9295749SChris Mason 3467a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) 3470af3d00bSJosef Bacik { 3480af3d00bSJosef Bacik return start_transaction(root, 0, TRANS_JOIN_NOLOCK); 3490af3d00bSJosef Bacik } 3500af3d00bSJosef Bacik 3517a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) 3529ca9ee09SSage Weil { 3537a7eaa40SJosef Bacik return start_transaction(root, 0, TRANS_USERSPACE); 3549ca9ee09SSage Weil } 3559ca9ee09SSage Weil 356d352ac68SChris Mason /* wait for a transaction commit to be fully complete */ 357b9c8300cSLi Zefan static noinline void wait_for_commit(struct btrfs_root *root, 35889ce8a63SChris Mason struct btrfs_transaction *commit) 35989ce8a63SChris Mason { 36072d63ed6SLi Zefan wait_event(commit->commit_wait, commit->commit_done); 36189ce8a63SChris Mason } 36289ce8a63SChris Mason 36346204592SSage Weil int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) 36446204592SSage Weil { 36546204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 36646204592SSage Weil int ret; 36746204592SSage Weil 36846204592SSage Weil ret = 0; 36946204592SSage Weil if (transid) { 37046204592SSage Weil if (transid <= root->fs_info->last_trans_committed) 371a4abeea4SJosef Bacik goto out; 37246204592SSage Weil 37346204592SSage Weil /* find specified transaction */ 374a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 37546204592SSage Weil list_for_each_entry(t, &root->fs_info->trans_list, list) { 37646204592SSage Weil if (t->transid == transid) { 37746204592SSage Weil cur_trans = t; 378a4abeea4SJosef Bacik atomic_inc(&cur_trans->use_count); 37946204592SSage Weil break; 38046204592SSage Weil } 38146204592SSage Weil if (t->transid > transid) 38246204592SSage Weil break; 38346204592SSage Weil } 384a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 38546204592SSage Weil ret = -EINVAL; 38646204592SSage Weil if (!cur_trans) 387a4abeea4SJosef Bacik goto out; /* bad transid */ 38846204592SSage Weil } else { 38946204592SSage Weil /* find newest transaction that is committing | committed */ 390a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 39146204592SSage Weil list_for_each_entry_reverse(t, &root->fs_info->trans_list, 39246204592SSage Weil list) { 39346204592SSage Weil if (t->in_commit) { 39446204592SSage Weil if (t->commit_done) 3953473f3c0SJosef Bacik break; 39646204592SSage Weil cur_trans = t; 397a4abeea4SJosef Bacik atomic_inc(&cur_trans->use_count); 39846204592SSage Weil break; 39946204592SSage Weil } 40046204592SSage Weil } 401a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 40246204592SSage Weil if (!cur_trans) 403a4abeea4SJosef Bacik goto out; /* nothing committing|committed */ 40446204592SSage Weil } 40546204592SSage Weil 40646204592SSage Weil wait_for_commit(root, cur_trans); 40746204592SSage Weil 40846204592SSage Weil put_transaction(cur_trans); 40946204592SSage Weil ret = 0; 410a4abeea4SJosef Bacik out: 41146204592SSage Weil return ret; 41246204592SSage Weil } 41346204592SSage Weil 41437d1aeeeSChris Mason void btrfs_throttle(struct btrfs_root *root) 41537d1aeeeSChris Mason { 416a4abeea4SJosef Bacik if (!atomic_read(&root->fs_info->open_ioctl_trans)) 41737d1aeeeSChris Mason wait_current_trans(root); 41837d1aeeeSChris Mason } 41937d1aeeeSChris Mason 4208929ecfaSYan, Zheng static int should_end_transaction(struct btrfs_trans_handle *trans, 4218929ecfaSYan, Zheng struct btrfs_root *root) 4228929ecfaSYan, Zheng { 4238929ecfaSYan, Zheng int ret; 42436ba022aSJosef Bacik 42536ba022aSJosef Bacik ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); 4268929ecfaSYan, Zheng return ret ? 1 : 0; 4278929ecfaSYan, Zheng } 4288929ecfaSYan, Zheng 4298929ecfaSYan, Zheng int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, 4308929ecfaSYan, Zheng struct btrfs_root *root) 4318929ecfaSYan, Zheng { 4328929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 4339c8d86dbSJosef Bacik struct btrfs_block_rsv *rsv = trans->block_rsv; 4348929ecfaSYan, Zheng int updates; 4358929ecfaSYan, Zheng 436a4abeea4SJosef Bacik smp_mb(); 4378929ecfaSYan, Zheng if (cur_trans->blocked || cur_trans->delayed_refs.flushing) 4388929ecfaSYan, Zheng return 1; 4398929ecfaSYan, Zheng 4409c8d86dbSJosef Bacik /* 4419c8d86dbSJosef Bacik * We need to do this in case we're deleting csums so the global block 4429c8d86dbSJosef Bacik * rsv get's used instead of the csum block rsv. 4439c8d86dbSJosef Bacik */ 4449c8d86dbSJosef Bacik trans->block_rsv = NULL; 4459c8d86dbSJosef Bacik 4468929ecfaSYan, Zheng updates = trans->delayed_ref_updates; 4478929ecfaSYan, Zheng trans->delayed_ref_updates = 0; 4488929ecfaSYan, Zheng if (updates) 4498929ecfaSYan, Zheng btrfs_run_delayed_refs(trans, root, updates); 4508929ecfaSYan, Zheng 4519c8d86dbSJosef Bacik trans->block_rsv = rsv; 4529c8d86dbSJosef Bacik 4538929ecfaSYan, Zheng return should_end_transaction(trans, root); 4548929ecfaSYan, Zheng } 4558929ecfaSYan, Zheng 45689ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 4570af3d00bSJosef Bacik struct btrfs_root *root, int throttle, int lock) 45879154b1bSChris Mason { 4598929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 460ab78c84dSChris Mason struct btrfs_fs_info *info = root->fs_info; 461c3e69d58SChris Mason int count = 0; 462d6e4a428SChris Mason 4632a1eb461SJosef Bacik if (--trans->use_count) { 4642a1eb461SJosef Bacik trans->block_rsv = trans->orig_rsv; 4652a1eb461SJosef Bacik return 0; 4662a1eb461SJosef Bacik } 4672a1eb461SJosef Bacik 468b24e03dbSJosef Bacik btrfs_trans_release_metadata(trans, root); 4694c13d758SJosef Bacik trans->block_rsv = NULL; 470203bf287SChris Mason while (count < 2) { 471c3e69d58SChris Mason unsigned long cur = trans->delayed_ref_updates; 472c3e69d58SChris Mason trans->delayed_ref_updates = 0; 473c3e69d58SChris Mason if (cur && 474c3e69d58SChris Mason trans->transaction->delayed_refs.num_heads_ready > 64) { 475c3e69d58SChris Mason trans->delayed_ref_updates = 0; 476c3e69d58SChris Mason btrfs_run_delayed_refs(trans, root, cur); 477c3e69d58SChris Mason } else { 478c3e69d58SChris Mason break; 479c3e69d58SChris Mason } 480c3e69d58SChris Mason count++; 48156bec294SChris Mason } 48256bec294SChris Mason 483a4abeea4SJosef Bacik if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && 484a4abeea4SJosef Bacik should_end_transaction(trans, root)) { 4858929ecfaSYan, Zheng trans->transaction->blocked = 1; 486a4abeea4SJosef Bacik smp_wmb(); 487a4abeea4SJosef Bacik } 4888929ecfaSYan, Zheng 4890af3d00bSJosef Bacik if (lock && cur_trans->blocked && !cur_trans->in_commit) { 49081317fdeSJosef Bacik if (throttle) { 49181317fdeSJosef Bacik /* 49281317fdeSJosef Bacik * We may race with somebody else here so end up having 49381317fdeSJosef Bacik * to call end_transaction on ourselves again, so inc 49481317fdeSJosef Bacik * our use_count. 49581317fdeSJosef Bacik */ 49681317fdeSJosef Bacik trans->use_count++; 4978929ecfaSYan, Zheng return btrfs_commit_transaction(trans, root); 49881317fdeSJosef Bacik } else { 4998929ecfaSYan, Zheng wake_up_process(info->transaction_kthread); 5008929ecfaSYan, Zheng } 50181317fdeSJosef Bacik } 5028929ecfaSYan, Zheng 5038929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 50413c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 50513c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 50689ce8a63SChris Mason 50799d16cbcSSage Weil smp_mb(); 50879154b1bSChris Mason if (waitqueue_active(&cur_trans->writer_wait)) 50979154b1bSChris Mason wake_up(&cur_trans->writer_wait); 51079154b1bSChris Mason put_transaction(cur_trans); 5119ed74f2dSJosef Bacik 5129ed74f2dSJosef Bacik if (current->journal_info == trans) 5139ed74f2dSJosef Bacik current->journal_info = NULL; 514d6025579SChris Mason memset(trans, 0, sizeof(*trans)); 5152c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 516ab78c84dSChris Mason 51724bbcf04SYan, Zheng if (throttle) 51824bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 51924bbcf04SYan, Zheng 52079154b1bSChris Mason return 0; 52179154b1bSChris Mason } 52279154b1bSChris Mason 52389ce8a63SChris Mason int btrfs_end_transaction(struct btrfs_trans_handle *trans, 52489ce8a63SChris Mason struct btrfs_root *root) 52589ce8a63SChris Mason { 52616cdcec7SMiao Xie int ret; 52716cdcec7SMiao Xie 52816cdcec7SMiao Xie ret = __btrfs_end_transaction(trans, root, 0, 1); 52916cdcec7SMiao Xie if (ret) 53016cdcec7SMiao Xie return ret; 53116cdcec7SMiao Xie return 0; 53289ce8a63SChris Mason } 53389ce8a63SChris Mason 53489ce8a63SChris Mason int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, 53589ce8a63SChris Mason struct btrfs_root *root) 53689ce8a63SChris Mason { 53716cdcec7SMiao Xie int ret; 53816cdcec7SMiao Xie 53916cdcec7SMiao Xie ret = __btrfs_end_transaction(trans, root, 1, 1); 54016cdcec7SMiao Xie if (ret) 54116cdcec7SMiao Xie return ret; 54216cdcec7SMiao Xie return 0; 5430af3d00bSJosef Bacik } 5440af3d00bSJosef Bacik 5450af3d00bSJosef Bacik int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, 5460af3d00bSJosef Bacik struct btrfs_root *root) 5470af3d00bSJosef Bacik { 54816cdcec7SMiao Xie int ret; 54916cdcec7SMiao Xie 55016cdcec7SMiao Xie ret = __btrfs_end_transaction(trans, root, 0, 0); 55116cdcec7SMiao Xie if (ret) 55216cdcec7SMiao Xie return ret; 55316cdcec7SMiao Xie return 0; 55416cdcec7SMiao Xie } 55516cdcec7SMiao Xie 55616cdcec7SMiao Xie int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, 55716cdcec7SMiao Xie struct btrfs_root *root) 55816cdcec7SMiao Xie { 55916cdcec7SMiao Xie return __btrfs_end_transaction(trans, root, 1, 1); 56089ce8a63SChris Mason } 56189ce8a63SChris Mason 562d352ac68SChris Mason /* 563d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 564d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 565690587d1SChris Mason * those extents are sent to disk but does not wait on them 566d352ac68SChris Mason */ 567690587d1SChris Mason int btrfs_write_marked_extents(struct btrfs_root *root, 5688cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 56979154b1bSChris Mason { 570777e6bd7SChris Mason int err = 0; 5717c4452b9SChris Mason int werr = 0; 5721728366eSJosef Bacik struct address_space *mapping = root->fs_info->btree_inode->i_mapping; 573777e6bd7SChris Mason u64 start = 0; 5745f39d397SChris Mason u64 end; 5757c4452b9SChris Mason 5761728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 5771728366eSJosef Bacik mark)) { 5781728366eSJosef Bacik convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark, 5791728366eSJosef Bacik GFP_NOFS); 5801728366eSJosef Bacik err = filemap_fdatawrite_range(mapping, start, end); 5817c4452b9SChris Mason if (err) 5827c4452b9SChris Mason werr = err; 5831728366eSJosef Bacik cond_resched(); 5841728366eSJosef Bacik start = end + 1; 5857c4452b9SChris Mason } 586690587d1SChris Mason if (err) 587690587d1SChris Mason werr = err; 588690587d1SChris Mason return werr; 589690587d1SChris Mason } 590690587d1SChris Mason 591690587d1SChris Mason /* 592690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 593690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 594690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 595690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 596690587d1SChris Mason */ 597690587d1SChris Mason int btrfs_wait_marked_extents(struct btrfs_root *root, 5988cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 599690587d1SChris Mason { 600690587d1SChris Mason int err = 0; 601690587d1SChris Mason int werr = 0; 6021728366eSJosef Bacik struct address_space *mapping = root->fs_info->btree_inode->i_mapping; 603690587d1SChris Mason u64 start = 0; 604690587d1SChris Mason u64 end; 605690587d1SChris Mason 6061728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 6071728366eSJosef Bacik EXTENT_NEED_WAIT)) { 6081728366eSJosef Bacik clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS); 6091728366eSJosef Bacik err = filemap_fdatawait_range(mapping, start, end); 610777e6bd7SChris Mason if (err) 611777e6bd7SChris Mason werr = err; 612777e6bd7SChris Mason cond_resched(); 6131728366eSJosef Bacik start = end + 1; 614777e6bd7SChris Mason } 6157c4452b9SChris Mason if (err) 6167c4452b9SChris Mason werr = err; 6177c4452b9SChris Mason return werr; 61879154b1bSChris Mason } 61979154b1bSChris Mason 620690587d1SChris Mason /* 621690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 622690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 623690587d1SChris Mason * those extents are on disk for transaction or log commit 624690587d1SChris Mason */ 625690587d1SChris Mason int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, 6268cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 627690587d1SChris Mason { 628690587d1SChris Mason int ret; 629690587d1SChris Mason int ret2; 630690587d1SChris Mason 6318cef4e16SYan, Zheng ret = btrfs_write_marked_extents(root, dirty_pages, mark); 6328cef4e16SYan, Zheng ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); 633bf0da8c1SChris Mason 634bf0da8c1SChris Mason if (ret) 635bf0da8c1SChris Mason return ret; 636bf0da8c1SChris Mason if (ret2) 637bf0da8c1SChris Mason return ret2; 638bf0da8c1SChris Mason return 0; 639690587d1SChris Mason } 640690587d1SChris Mason 641d0c803c4SChris Mason int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, 642d0c803c4SChris Mason struct btrfs_root *root) 643d0c803c4SChris Mason { 644d0c803c4SChris Mason if (!trans || !trans->transaction) { 645d0c803c4SChris Mason struct inode *btree_inode; 646d0c803c4SChris Mason btree_inode = root->fs_info->btree_inode; 647d0c803c4SChris Mason return filemap_write_and_wait(btree_inode->i_mapping); 648d0c803c4SChris Mason } 649d0c803c4SChris Mason return btrfs_write_and_wait_marked_extents(root, 6508cef4e16SYan, Zheng &trans->transaction->dirty_pages, 6518cef4e16SYan, Zheng EXTENT_DIRTY); 652d0c803c4SChris Mason } 653d0c803c4SChris Mason 654d352ac68SChris Mason /* 655d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 656d352ac68SChris Mason * 657d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 658d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 659d352ac68SChris Mason * allocation tree. 660d352ac68SChris Mason * 661d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 662d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 663d352ac68SChris Mason */ 6640b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 66579154b1bSChris Mason struct btrfs_root *root) 66679154b1bSChris Mason { 66779154b1bSChris Mason int ret; 6680b86a832SChris Mason u64 old_root_bytenr; 66986b9f2ecSYan, Zheng u64 old_root_used; 6700b86a832SChris Mason struct btrfs_root *tree_root = root->fs_info->tree_root; 67179154b1bSChris Mason 67286b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 6730b86a832SChris Mason btrfs_write_dirty_block_groups(trans, root); 67456bec294SChris Mason 67579154b1bSChris Mason while (1) { 6760b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 67786b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 67886b9f2ecSYan, Zheng old_root_used == btrfs_root_used(&root->root_item)) 67979154b1bSChris Mason break; 68087ef2bb4SChris Mason 6815d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 68279154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 6830b86a832SChris Mason &root->root_key, 6840b86a832SChris Mason &root->root_item); 68579154b1bSChris Mason BUG_ON(ret); 68656bec294SChris Mason 68786b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 6884a8c9a62SYan Zheng ret = btrfs_write_dirty_block_groups(trans, root); 68956bec294SChris Mason BUG_ON(ret); 6900b86a832SChris Mason } 691276e680dSYan Zheng 692276e680dSYan Zheng if (root != root->fs_info->extent_root) 693817d52f8SJosef Bacik switch_commit_root(root); 694276e680dSYan Zheng 6950b86a832SChris Mason return 0; 6960b86a832SChris Mason } 6970b86a832SChris Mason 698d352ac68SChris Mason /* 699d352ac68SChris Mason * update all the cowonly tree roots on disk 700d352ac68SChris Mason */ 7015d4f98a2SYan Zheng static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, 7020b86a832SChris Mason struct btrfs_root *root) 7030b86a832SChris Mason { 7040b86a832SChris Mason struct btrfs_fs_info *fs_info = root->fs_info; 7050b86a832SChris Mason struct list_head *next; 70684234f3aSYan Zheng struct extent_buffer *eb; 70756bec294SChris Mason int ret; 70884234f3aSYan Zheng 70956bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 71056bec294SChris Mason BUG_ON(ret); 71187ef2bb4SChris Mason 71284234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 7139fa8cfe7SChris Mason btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); 71484234f3aSYan Zheng btrfs_tree_unlock(eb); 71584234f3aSYan Zheng free_extent_buffer(eb); 7160b86a832SChris Mason 71756bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 71856bec294SChris Mason BUG_ON(ret); 71987ef2bb4SChris Mason 7200b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 7210b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 7220b86a832SChris Mason list_del_init(next); 7230b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 72487ef2bb4SChris Mason 7250b86a832SChris Mason update_cowonly_root(trans, root); 72679154b1bSChris Mason } 727276e680dSYan Zheng 728276e680dSYan Zheng down_write(&fs_info->extent_commit_sem); 729276e680dSYan Zheng switch_commit_root(fs_info->extent_root); 730276e680dSYan Zheng up_write(&fs_info->extent_commit_sem); 731276e680dSYan Zheng 73279154b1bSChris Mason return 0; 73379154b1bSChris Mason } 73479154b1bSChris Mason 735d352ac68SChris Mason /* 736d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 737d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 738d352ac68SChris Mason * be deleted 739d352ac68SChris Mason */ 7405d4f98a2SYan Zheng int btrfs_add_dead_root(struct btrfs_root *root) 7415eda7b5eSChris Mason { 742a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 7435d4f98a2SYan Zheng list_add(&root->root_list, &root->fs_info->dead_roots); 744a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 7455eda7b5eSChris Mason return 0; 7465eda7b5eSChris Mason } 7475eda7b5eSChris Mason 748d352ac68SChris Mason /* 7495d4f98a2SYan Zheng * update all the cowonly tree roots on disk 750d352ac68SChris Mason */ 7515d4f98a2SYan Zheng static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, 7525d4f98a2SYan Zheng struct btrfs_root *root) 7530f7d52f4SChris Mason { 7540f7d52f4SChris Mason struct btrfs_root *gang[8]; 7555d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = root->fs_info; 7560f7d52f4SChris Mason int i; 7570f7d52f4SChris Mason int ret; 75854aa1f4dSChris Mason int err = 0; 75954aa1f4dSChris Mason 760a4abeea4SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 7610f7d52f4SChris Mason while (1) { 7625d4f98a2SYan Zheng ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 7635d4f98a2SYan Zheng (void **)gang, 0, 7640f7d52f4SChris Mason ARRAY_SIZE(gang), 7650f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 7660f7d52f4SChris Mason if (ret == 0) 7670f7d52f4SChris Mason break; 7680f7d52f4SChris Mason for (i = 0; i < ret; i++) { 7690f7d52f4SChris Mason root = gang[i]; 7705d4f98a2SYan Zheng radix_tree_tag_clear(&fs_info->fs_roots_radix, 7712619ba1fSChris Mason (unsigned long)root->root_key.objectid, 7720f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 773a4abeea4SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 77431153d81SYan Zheng 775e02119d5SChris Mason btrfs_free_log(trans, root); 7765d4f98a2SYan Zheng btrfs_update_reloc_root(trans, root); 777d68fc57bSYan, Zheng btrfs_orphan_commit_root(trans, root); 778e02119d5SChris Mason 77982d5902dSLi Zefan btrfs_save_ino_cache(root, trans); 78082d5902dSLi Zefan 781f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 782f1ebcc74SLiu Bo root->force_cow = 0; 783f1ebcc74SLiu Bo smp_wmb(); 784f1ebcc74SLiu Bo 785978d910dSYan Zheng if (root->commit_root != root->node) { 786581bb050SLi Zefan mutex_lock(&root->fs_commit_mutex); 787817d52f8SJosef Bacik switch_commit_root(root); 788581bb050SLi Zefan btrfs_unpin_free_ino(root); 789581bb050SLi Zefan mutex_unlock(&root->fs_commit_mutex); 790581bb050SLi Zefan 791978d910dSYan Zheng btrfs_set_root_node(&root->root_item, 792978d910dSYan Zheng root->node); 793978d910dSYan Zheng } 79431153d81SYan Zheng 7955d4f98a2SYan Zheng err = btrfs_update_root(trans, fs_info->tree_root, 7960f7d52f4SChris Mason &root->root_key, 7970f7d52f4SChris Mason &root->root_item); 798a4abeea4SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 79954aa1f4dSChris Mason if (err) 80054aa1f4dSChris Mason break; 8010f7d52f4SChris Mason } 8029f3a7427SChris Mason } 803a4abeea4SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 80454aa1f4dSChris Mason return err; 8050f7d52f4SChris Mason } 8060f7d52f4SChris Mason 807d352ac68SChris Mason /* 808d352ac68SChris Mason * defrag a given btree. If cacheonly == 1, this won't read from the disk, 809d352ac68SChris Mason * otherwise every leaf in the btree is read and defragged. 810d352ac68SChris Mason */ 811e9d0b13bSChris Mason int btrfs_defrag_root(struct btrfs_root *root, int cacheonly) 812e9d0b13bSChris Mason { 813e9d0b13bSChris Mason struct btrfs_fs_info *info = root->fs_info; 814e9d0b13bSChris Mason struct btrfs_trans_handle *trans; 8158929ecfaSYan, Zheng int ret; 816d3c2fdcfSChris Mason unsigned long nr; 817e9d0b13bSChris Mason 8188929ecfaSYan, Zheng if (xchg(&root->defrag_running, 1)) 819e9d0b13bSChris Mason return 0; 8208929ecfaSYan, Zheng 8216b80053dSChris Mason while (1) { 8228929ecfaSYan, Zheng trans = btrfs_start_transaction(root, 0); 8238929ecfaSYan, Zheng if (IS_ERR(trans)) 8248929ecfaSYan, Zheng return PTR_ERR(trans); 8258929ecfaSYan, Zheng 826e9d0b13bSChris Mason ret = btrfs_defrag_leaves(trans, root, cacheonly); 8278929ecfaSYan, Zheng 828d3c2fdcfSChris Mason nr = trans->blocks_used; 829e9d0b13bSChris Mason btrfs_end_transaction(trans, root); 830d3c2fdcfSChris Mason btrfs_btree_balance_dirty(info->tree_root, nr); 831e9d0b13bSChris Mason cond_resched(); 832e9d0b13bSChris Mason 8337841cb28SDavid Sterba if (btrfs_fs_closing(root->fs_info) || ret != -EAGAIN) 834e9d0b13bSChris Mason break; 835e9d0b13bSChris Mason } 836e9d0b13bSChris Mason root->defrag_running = 0; 8378929ecfaSYan, Zheng return ret; 838e9d0b13bSChris Mason } 839e9d0b13bSChris Mason 840d352ac68SChris Mason /* 841d352ac68SChris Mason * new snapshots need to be created at a very specific time in the 842d352ac68SChris Mason * transaction commit. This does the actual creation 843d352ac68SChris Mason */ 84480b6794dSChris Mason static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans, 8453063d29fSChris Mason struct btrfs_fs_info *fs_info, 8463063d29fSChris Mason struct btrfs_pending_snapshot *pending) 8473063d29fSChris Mason { 8483063d29fSChris Mason struct btrfs_key key; 84980b6794dSChris Mason struct btrfs_root_item *new_root_item; 8503063d29fSChris Mason struct btrfs_root *tree_root = fs_info->tree_root; 8513063d29fSChris Mason struct btrfs_root *root = pending->root; 8526bdb72deSSage Weil struct btrfs_root *parent_root; 85398c9942aSLiu Bo struct btrfs_block_rsv *rsv; 8546bdb72deSSage Weil struct inode *parent_inode; 8556a912213SJosef Bacik struct dentry *parent; 856a22285a6SYan, Zheng struct dentry *dentry; 8573063d29fSChris Mason struct extent_buffer *tmp; 858925baeddSChris Mason struct extent_buffer *old; 8593063d29fSChris Mason int ret; 860d68fc57bSYan, Zheng u64 to_reserve = 0; 8616bdb72deSSage Weil u64 index = 0; 862a22285a6SYan, Zheng u64 objectid; 863b83cc969SLi Zefan u64 root_flags; 8643063d29fSChris Mason 86598c9942aSLiu Bo rsv = trans->block_rsv; 86698c9942aSLiu Bo 86780b6794dSChris Mason new_root_item = kmalloc(sizeof(*new_root_item), GFP_NOFS); 86880b6794dSChris Mason if (!new_root_item) { 869a22285a6SYan, Zheng pending->error = -ENOMEM; 87080b6794dSChris Mason goto fail; 87180b6794dSChris Mason } 872a22285a6SYan, Zheng 873581bb050SLi Zefan ret = btrfs_find_free_objectid(tree_root, &objectid); 874a22285a6SYan, Zheng if (ret) { 875a22285a6SYan, Zheng pending->error = ret; 8763063d29fSChris Mason goto fail; 877a22285a6SYan, Zheng } 8783063d29fSChris Mason 8793fd0a558SYan, Zheng btrfs_reloc_pre_snapshot(trans, pending, &to_reserve); 880d68fc57bSYan, Zheng 881d68fc57bSYan, Zheng if (to_reserve > 0) { 88262f30c54SMiao Xie ret = btrfs_block_rsv_add_noflush(root, &pending->block_rsv, 8838bb8ab2eSJosef Bacik to_reserve); 884d68fc57bSYan, Zheng if (ret) { 885d68fc57bSYan, Zheng pending->error = ret; 886d68fc57bSYan, Zheng goto fail; 887d68fc57bSYan, Zheng } 888d68fc57bSYan, Zheng } 889d68fc57bSYan, Zheng 8903063d29fSChris Mason key.objectid = objectid; 891a22285a6SYan, Zheng key.offset = (u64)-1; 892a22285a6SYan, Zheng key.type = BTRFS_ROOT_ITEM_KEY; 8933063d29fSChris Mason 894a22285a6SYan, Zheng trans->block_rsv = &pending->block_rsv; 8956bdb72deSSage Weil 896a22285a6SYan, Zheng dentry = pending->dentry; 8976a912213SJosef Bacik parent = dget_parent(dentry); 8986a912213SJosef Bacik parent_inode = parent->d_inode; 899a22285a6SYan, Zheng parent_root = BTRFS_I(parent_inode)->root; 9007585717fSChris Mason record_root_in_trans(trans, parent_root); 901a22285a6SYan, Zheng 9026bdb72deSSage Weil /* 9036bdb72deSSage Weil * insert the directory item 9046bdb72deSSage Weil */ 9056bdb72deSSage Weil ret = btrfs_set_inode_index(parent_inode, &index); 9066bdb72deSSage Weil BUG_ON(ret); 9076bdb72deSSage Weil ret = btrfs_insert_dir_item(trans, parent_root, 908a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len, 90916cdcec7SMiao Xie parent_inode, &key, 910a22285a6SYan, Zheng BTRFS_FT_DIR, index); 9116bdb72deSSage Weil BUG_ON(ret); 9126bdb72deSSage Weil 913a22285a6SYan, Zheng btrfs_i_size_write(parent_inode, parent_inode->i_size + 914a22285a6SYan, Zheng dentry->d_name.len * 2); 9156bdb72deSSage Weil ret = btrfs_update_inode(trans, parent_root, parent_inode); 9166bdb72deSSage Weil BUG_ON(ret); 9176bdb72deSSage Weil 918e999376fSChris Mason /* 919e999376fSChris Mason * pull in the delayed directory update 920e999376fSChris Mason * and the delayed inode item 921e999376fSChris Mason * otherwise we corrupt the FS during 922e999376fSChris Mason * snapshot 923e999376fSChris Mason */ 924e999376fSChris Mason ret = btrfs_run_delayed_items(trans, root); 925e999376fSChris Mason BUG_ON(ret); 926e999376fSChris Mason 9277585717fSChris Mason record_root_in_trans(trans, root); 9286bdb72deSSage Weil btrfs_set_root_last_snapshot(&root->root_item, trans->transid); 9296bdb72deSSage Weil memcpy(new_root_item, &root->root_item, sizeof(*new_root_item)); 93008fe4db1SLi Zefan btrfs_check_and_init_root_item(new_root_item); 9316bdb72deSSage Weil 932b83cc969SLi Zefan root_flags = btrfs_root_flags(new_root_item); 933b83cc969SLi Zefan if (pending->readonly) 934b83cc969SLi Zefan root_flags |= BTRFS_ROOT_SUBVOL_RDONLY; 935b83cc969SLi Zefan else 936b83cc969SLi Zefan root_flags &= ~BTRFS_ROOT_SUBVOL_RDONLY; 937b83cc969SLi Zefan btrfs_set_root_flags(new_root_item, root_flags); 938b83cc969SLi Zefan 939925baeddSChris Mason old = btrfs_lock_root_node(root); 9409fa8cfe7SChris Mason btrfs_cow_block(trans, root, old, NULL, 0, &old); 9415d4f98a2SYan Zheng btrfs_set_lock_blocking(old); 9423063d29fSChris Mason 943925baeddSChris Mason btrfs_copy_root(trans, root, old, &tmp, objectid); 944925baeddSChris Mason btrfs_tree_unlock(old); 945925baeddSChris Mason free_extent_buffer(old); 9463063d29fSChris Mason 947f1ebcc74SLiu Bo /* see comments in should_cow_block() */ 948f1ebcc74SLiu Bo root->force_cow = 1; 949f1ebcc74SLiu Bo smp_wmb(); 950f1ebcc74SLiu Bo 9515d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 952a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 953a22285a6SYan, Zheng key.offset = trans->transid; 954a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 955925baeddSChris Mason btrfs_tree_unlock(tmp); 9563063d29fSChris Mason free_extent_buffer(tmp); 9570660b5afSChris Mason BUG_ON(ret); 9580660b5afSChris Mason 959a22285a6SYan, Zheng /* 960a22285a6SYan, Zheng * insert root back/forward references 961a22285a6SYan, Zheng */ 962a22285a6SYan, Zheng ret = btrfs_add_root_ref(trans, tree_root, objectid, 963a22285a6SYan, Zheng parent_root->root_key.objectid, 96433345d01SLi Zefan btrfs_ino(parent_inode), index, 965a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len); 966a22285a6SYan, Zheng BUG_ON(ret); 9676a912213SJosef Bacik dput(parent); 968a22285a6SYan, Zheng 969a22285a6SYan, Zheng key.offset = (u64)-1; 970a22285a6SYan, Zheng pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); 971a22285a6SYan, Zheng BUG_ON(IS_ERR(pending->snap)); 972d68fc57bSYan, Zheng 9733fd0a558SYan, Zheng btrfs_reloc_post_snapshot(trans, pending); 9743063d29fSChris Mason fail: 9756bdb72deSSage Weil kfree(new_root_item); 97698c9942aSLiu Bo trans->block_rsv = rsv; 977a22285a6SYan, Zheng btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); 978a22285a6SYan, Zheng return 0; 9793063d29fSChris Mason } 9803063d29fSChris Mason 981d352ac68SChris Mason /* 982d352ac68SChris Mason * create all the snapshots we've scheduled for creation 983d352ac68SChris Mason */ 98480b6794dSChris Mason static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, 9853063d29fSChris Mason struct btrfs_fs_info *fs_info) 9863063d29fSChris Mason { 9873063d29fSChris Mason struct btrfs_pending_snapshot *pending; 9883063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 9893de4586cSChris Mason int ret; 9903de4586cSChris Mason 991c6e30871SQinghuang Feng list_for_each_entry(pending, head, list) { 9923de4586cSChris Mason ret = create_pending_snapshot(trans, fs_info, pending); 9933de4586cSChris Mason BUG_ON(ret); 9943de4586cSChris Mason } 9953de4586cSChris Mason return 0; 9963de4586cSChris Mason } 9973de4586cSChris Mason 9985d4f98a2SYan Zheng static void update_super_roots(struct btrfs_root *root) 9995d4f98a2SYan Zheng { 10005d4f98a2SYan Zheng struct btrfs_root_item *root_item; 10015d4f98a2SYan Zheng struct btrfs_super_block *super; 10025d4f98a2SYan Zheng 10036c41761fSDavid Sterba super = root->fs_info->super_copy; 10045d4f98a2SYan Zheng 10055d4f98a2SYan Zheng root_item = &root->fs_info->chunk_root->root_item; 10065d4f98a2SYan Zheng super->chunk_root = root_item->bytenr; 10075d4f98a2SYan Zheng super->chunk_root_generation = root_item->generation; 10085d4f98a2SYan Zheng super->chunk_root_level = root_item->level; 10095d4f98a2SYan Zheng 10105d4f98a2SYan Zheng root_item = &root->fs_info->tree_root->root_item; 10115d4f98a2SYan Zheng super->root = root_item->bytenr; 10125d4f98a2SYan Zheng super->generation = root_item->generation; 10135d4f98a2SYan Zheng super->root_level = root_item->level; 101473bc1876SJosef Bacik if (btrfs_test_opt(root, SPACE_CACHE)) 10150af3d00bSJosef Bacik super->cache_generation = root_item->generation; 10165d4f98a2SYan Zheng } 10175d4f98a2SYan Zheng 1018f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1019f36f3042SChris Mason { 1020f36f3042SChris Mason int ret = 0; 1021a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 1022f36f3042SChris Mason if (info->running_transaction) 1023f36f3042SChris Mason ret = info->running_transaction->in_commit; 1024a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 1025f36f3042SChris Mason return ret; 1026f36f3042SChris Mason } 1027f36f3042SChris Mason 10288929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 10298929ecfaSYan, Zheng { 10308929ecfaSYan, Zheng int ret = 0; 1031a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 10328929ecfaSYan, Zheng if (info->running_transaction) 10338929ecfaSYan, Zheng ret = info->running_transaction->blocked; 1034a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 10358929ecfaSYan, Zheng return ret; 10368929ecfaSYan, Zheng } 10378929ecfaSYan, Zheng 1038bb9c12c9SSage Weil /* 1039bb9c12c9SSage Weil * wait for the current transaction commit to start and block subsequent 1040bb9c12c9SSage Weil * transaction joins 1041bb9c12c9SSage Weil */ 1042bb9c12c9SSage Weil static void wait_current_trans_commit_start(struct btrfs_root *root, 1043bb9c12c9SSage Weil struct btrfs_transaction *trans) 1044bb9c12c9SSage Weil { 104572d63ed6SLi Zefan wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit); 1046bb9c12c9SSage Weil } 1047bb9c12c9SSage Weil 1048bb9c12c9SSage Weil /* 1049bb9c12c9SSage Weil * wait for the current transaction to start and then become unblocked. 1050bb9c12c9SSage Weil * caller holds ref. 1051bb9c12c9SSage Weil */ 1052bb9c12c9SSage Weil static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, 1053bb9c12c9SSage Weil struct btrfs_transaction *trans) 1054bb9c12c9SSage Weil { 105572d63ed6SLi Zefan wait_event(root->fs_info->transaction_wait, 105672d63ed6SLi Zefan trans->commit_done || (trans->in_commit && !trans->blocked)); 1057bb9c12c9SSage Weil } 1058bb9c12c9SSage Weil 1059bb9c12c9SSage Weil /* 1060bb9c12c9SSage Weil * commit transactions asynchronously. once btrfs_commit_transaction_async 1061bb9c12c9SSage Weil * returns, any subsequent transaction will not be allowed to join. 1062bb9c12c9SSage Weil */ 1063bb9c12c9SSage Weil struct btrfs_async_commit { 1064bb9c12c9SSage Weil struct btrfs_trans_handle *newtrans; 1065bb9c12c9SSage Weil struct btrfs_root *root; 1066bb9c12c9SSage Weil struct delayed_work work; 1067bb9c12c9SSage Weil }; 1068bb9c12c9SSage Weil 1069bb9c12c9SSage Weil static void do_async_commit(struct work_struct *work) 1070bb9c12c9SSage Weil { 1071bb9c12c9SSage Weil struct btrfs_async_commit *ac = 1072bb9c12c9SSage Weil container_of(work, struct btrfs_async_commit, work.work); 1073bb9c12c9SSage Weil 1074bb9c12c9SSage Weil btrfs_commit_transaction(ac->newtrans, ac->root); 1075bb9c12c9SSage Weil kfree(ac); 1076bb9c12c9SSage Weil } 1077bb9c12c9SSage Weil 1078bb9c12c9SSage Weil int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, 1079bb9c12c9SSage Weil struct btrfs_root *root, 1080bb9c12c9SSage Weil int wait_for_unblock) 1081bb9c12c9SSage Weil { 1082bb9c12c9SSage Weil struct btrfs_async_commit *ac; 1083bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1084bb9c12c9SSage Weil 1085bb9c12c9SSage Weil ac = kmalloc(sizeof(*ac), GFP_NOFS); 1086db5b493aSTsutomu Itoh if (!ac) 1087db5b493aSTsutomu Itoh return -ENOMEM; 1088bb9c12c9SSage Weil 1089bb9c12c9SSage Weil INIT_DELAYED_WORK(&ac->work, do_async_commit); 1090bb9c12c9SSage Weil ac->root = root; 10917a7eaa40SJosef Bacik ac->newtrans = btrfs_join_transaction(root); 10923612b495STsutomu Itoh if (IS_ERR(ac->newtrans)) { 10933612b495STsutomu Itoh int err = PTR_ERR(ac->newtrans); 10943612b495STsutomu Itoh kfree(ac); 10953612b495STsutomu Itoh return err; 10963612b495STsutomu Itoh } 1097bb9c12c9SSage Weil 1098bb9c12c9SSage Weil /* take transaction reference */ 1099bb9c12c9SSage Weil cur_trans = trans->transaction; 110013c5a93eSJosef Bacik atomic_inc(&cur_trans->use_count); 1101bb9c12c9SSage Weil 1102bb9c12c9SSage Weil btrfs_end_transaction(trans, root); 1103bb9c12c9SSage Weil schedule_delayed_work(&ac->work, 0); 1104bb9c12c9SSage Weil 1105bb9c12c9SSage Weil /* wait for transaction to start and unblock */ 1106bb9c12c9SSage Weil if (wait_for_unblock) 1107bb9c12c9SSage Weil wait_current_trans_commit_start_and_unblock(root, cur_trans); 1108bb9c12c9SSage Weil else 1109bb9c12c9SSage Weil wait_current_trans_commit_start(root, cur_trans); 1110bb9c12c9SSage Weil 111138e88054SSage Weil if (current->journal_info == trans) 111238e88054SSage Weil current->journal_info = NULL; 111338e88054SSage Weil 111438e88054SSage Weil put_transaction(cur_trans); 1115bb9c12c9SSage Weil return 0; 1116bb9c12c9SSage Weil } 1117bb9c12c9SSage Weil 1118bb9c12c9SSage Weil /* 1119bb9c12c9SSage Weil * btrfs_transaction state sequence: 1120bb9c12c9SSage Weil * in_commit = 0, blocked = 0 (initial) 1121bb9c12c9SSage Weil * in_commit = 1, blocked = 1 1122bb9c12c9SSage Weil * blocked = 0 1123bb9c12c9SSage Weil * commit_done = 1 1124bb9c12c9SSage Weil */ 112579154b1bSChris Mason int btrfs_commit_transaction(struct btrfs_trans_handle *trans, 112679154b1bSChris Mason struct btrfs_root *root) 112779154b1bSChris Mason { 112815ee9bc7SJosef Bacik unsigned long joined = 0; 112979154b1bSChris Mason struct btrfs_transaction *cur_trans; 11308fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 113179154b1bSChris Mason DEFINE_WAIT(wait); 113215ee9bc7SJosef Bacik int ret; 113389573b9cSChris Mason int should_grow = 0; 113489573b9cSChris Mason unsigned long now = get_seconds(); 1135dccae999SSage Weil int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); 113679154b1bSChris Mason 11375a3f23d5SChris Mason btrfs_run_ordered_operations(root, 0); 11385a3f23d5SChris Mason 1139b24e03dbSJosef Bacik btrfs_trans_release_metadata(trans, root); 11409c8d86dbSJosef Bacik trans->block_rsv = NULL; 11419c8d86dbSJosef Bacik 114256bec294SChris Mason /* make a pass through all the delayed refs we have so far 114356bec294SChris Mason * any runnings procs may add more while we are here 114456bec294SChris Mason */ 114556bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, 0); 114656bec294SChris Mason BUG_ON(ret); 114756bec294SChris Mason 1148b7ec40d7SChris Mason cur_trans = trans->transaction; 114956bec294SChris Mason /* 115056bec294SChris Mason * set the flushing flag so procs in this transaction have to 115156bec294SChris Mason * start sending their work down. 115256bec294SChris Mason */ 1153b7ec40d7SChris Mason cur_trans->delayed_refs.flushing = 1; 115456bec294SChris Mason 1155c3e69d58SChris Mason ret = btrfs_run_delayed_refs(trans, root, 0); 115656bec294SChris Mason BUG_ON(ret); 115756bec294SChris Mason 1158a4abeea4SJosef Bacik spin_lock(&cur_trans->commit_lock); 1159b7ec40d7SChris Mason if (cur_trans->in_commit) { 1160a4abeea4SJosef Bacik spin_unlock(&cur_trans->commit_lock); 116113c5a93eSJosef Bacik atomic_inc(&cur_trans->use_count); 116279154b1bSChris Mason btrfs_end_transaction(trans, root); 1163ccd467d6SChris Mason 1164b9c8300cSLi Zefan wait_for_commit(root, cur_trans); 116515ee9bc7SJosef Bacik 116679154b1bSChris Mason put_transaction(cur_trans); 116715ee9bc7SJosef Bacik 116879154b1bSChris Mason return 0; 116979154b1bSChris Mason } 11704313b399SChris Mason 11712c90e5d6SChris Mason trans->transaction->in_commit = 1; 1172f9295749SChris Mason trans->transaction->blocked = 1; 1173a4abeea4SJosef Bacik spin_unlock(&cur_trans->commit_lock); 1174bb9c12c9SSage Weil wake_up(&root->fs_info->transaction_blocked_wait); 1175bb9c12c9SSage Weil 1176a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 1177ccd467d6SChris Mason if (cur_trans->list.prev != &root->fs_info->trans_list) { 1178ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 1179ccd467d6SChris Mason struct btrfs_transaction, list); 1180ccd467d6SChris Mason if (!prev_trans->commit_done) { 118113c5a93eSJosef Bacik atomic_inc(&prev_trans->use_count); 1182a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1183ccd467d6SChris Mason 1184ccd467d6SChris Mason wait_for_commit(root, prev_trans); 1185ccd467d6SChris Mason 118615ee9bc7SJosef Bacik put_transaction(prev_trans); 1187a4abeea4SJosef Bacik } else { 1188a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1189ccd467d6SChris Mason } 1190a4abeea4SJosef Bacik } else { 1191a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1192ccd467d6SChris Mason } 119315ee9bc7SJosef Bacik 119489573b9cSChris Mason if (now < cur_trans->start_time || now - cur_trans->start_time < 1) 119589573b9cSChris Mason should_grow = 1; 119689573b9cSChris Mason 119715ee9bc7SJosef Bacik do { 11987ea394f1SYan Zheng int snap_pending = 0; 1199a4abeea4SJosef Bacik 120015ee9bc7SJosef Bacik joined = cur_trans->num_joined; 12017ea394f1SYan Zheng if (!list_empty(&trans->transaction->pending_snapshots)) 12027ea394f1SYan Zheng snap_pending = 1; 12037ea394f1SYan Zheng 12042c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 120515ee9bc7SJosef Bacik 12060bdb1db2SSage Weil if (flush_on_commit || snap_pending) { 120724bbcf04SYan, Zheng btrfs_start_delalloc_inodes(root, 1); 120824bbcf04SYan, Zheng ret = btrfs_wait_ordered_extents(root, 0, 1); 1209ebecd3d9SSage Weil BUG_ON(ret); 12107ea394f1SYan Zheng } 12117ea394f1SYan Zheng 121216cdcec7SMiao Xie ret = btrfs_run_delayed_items(trans, root); 121316cdcec7SMiao Xie BUG_ON(ret); 121416cdcec7SMiao Xie 12155a3f23d5SChris Mason /* 12165a3f23d5SChris Mason * rename don't use btrfs_join_transaction, so, once we 12175a3f23d5SChris Mason * set the transaction to blocked above, we aren't going 12185a3f23d5SChris Mason * to get any new ordered operations. We can safely run 12195a3f23d5SChris Mason * it here and no for sure that nothing new will be added 12205a3f23d5SChris Mason * to the list 12215a3f23d5SChris Mason */ 12225a3f23d5SChris Mason btrfs_run_ordered_operations(root, 1); 12235a3f23d5SChris Mason 1224ed3b3d31SChris Mason prepare_to_wait(&cur_trans->writer_wait, &wait, 1225ed3b3d31SChris Mason TASK_UNINTERRUPTIBLE); 1226ed3b3d31SChris Mason 122713c5a93eSJosef Bacik if (atomic_read(&cur_trans->num_writers) > 1) 122899d16cbcSSage Weil schedule_timeout(MAX_SCHEDULE_TIMEOUT); 122999d16cbcSSage Weil else if (should_grow) 123099d16cbcSSage Weil schedule_timeout(1); 123115ee9bc7SJosef Bacik 123215ee9bc7SJosef Bacik finish_wait(&cur_trans->writer_wait, &wait); 1233ed0ca140SJosef Bacik } while (atomic_read(&cur_trans->num_writers) > 1 || 1234ed0ca140SJosef Bacik (should_grow && cur_trans->num_joined != joined)); 1235ed0ca140SJosef Bacik 1236ed0ca140SJosef Bacik /* 1237ed0ca140SJosef Bacik * Ok now we need to make sure to block out any other joins while we 1238ed0ca140SJosef Bacik * commit the transaction. We could have started a join before setting 1239ed0ca140SJosef Bacik * no_join so make sure to wait for num_writers to == 1 again. 1240ed0ca140SJosef Bacik */ 1241a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 1242a4abeea4SJosef Bacik root->fs_info->trans_no_join = 1; 1243a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1244ed0ca140SJosef Bacik wait_event(cur_trans->writer_wait, 1245ed0ca140SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 124615ee9bc7SJosef Bacik 12477585717fSChris Mason /* 12487585717fSChris Mason * the reloc mutex makes sure that we stop 12497585717fSChris Mason * the balancing code from coming in and moving 12507585717fSChris Mason * extents around in the middle of the commit 12517585717fSChris Mason */ 12527585717fSChris Mason mutex_lock(&root->fs_info->reloc_mutex); 12537585717fSChris Mason 1254e999376fSChris Mason ret = btrfs_run_delayed_items(trans, root); 12553063d29fSChris Mason BUG_ON(ret); 12563063d29fSChris Mason 1257e999376fSChris Mason ret = create_pending_snapshots(trans, root->fs_info); 125816cdcec7SMiao Xie BUG_ON(ret); 125916cdcec7SMiao Xie 126056bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 126156bec294SChris Mason BUG_ON(ret); 126256bec294SChris Mason 1263e999376fSChris Mason /* 1264e999376fSChris Mason * make sure none of the code above managed to slip in a 1265e999376fSChris Mason * delayed item 1266e999376fSChris Mason */ 1267e999376fSChris Mason btrfs_assert_delayed_root_empty(root); 1268e999376fSChris Mason 12692c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 1270dc17ff8fSChris Mason 1271a2de733cSArne Jansen btrfs_scrub_pause(root); 1272e02119d5SChris Mason /* btrfs_commit_tree_roots is responsible for getting the 1273e02119d5SChris Mason * various roots consistent with each other. Every pointer 1274e02119d5SChris Mason * in the tree of tree roots has to point to the most up to date 1275e02119d5SChris Mason * root for every subvolume and other tree. So, we have to keep 1276e02119d5SChris Mason * the tree logging code from jumping in and changing any 1277e02119d5SChris Mason * of the trees. 1278e02119d5SChris Mason * 1279e02119d5SChris Mason * At this point in the commit, there can't be any tree-log 1280e02119d5SChris Mason * writers, but a little lower down we drop the trans mutex 1281e02119d5SChris Mason * and let new people in. By holding the tree_log_mutex 1282e02119d5SChris Mason * from now until after the super is written, we avoid races 1283e02119d5SChris Mason * with the tree-log code. 1284e02119d5SChris Mason */ 1285e02119d5SChris Mason mutex_lock(&root->fs_info->tree_log_mutex); 12861a40e23bSZheng Yan 12875d4f98a2SYan Zheng ret = commit_fs_roots(trans, root); 128854aa1f4dSChris Mason BUG_ON(ret); 128954aa1f4dSChris Mason 12905d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 1291e02119d5SChris Mason * safe to free the root of tree log roots 1292e02119d5SChris Mason */ 1293e02119d5SChris Mason btrfs_free_log_root_tree(trans, root->fs_info); 1294e02119d5SChris Mason 12955d4f98a2SYan Zheng ret = commit_cowonly_roots(trans, root); 129679154b1bSChris Mason BUG_ON(ret); 129754aa1f4dSChris Mason 129811833d66SYan Zheng btrfs_prepare_extent_commit(trans, root); 129911833d66SYan Zheng 130078fae27eSChris Mason cur_trans = root->fs_info->running_transaction; 13015f39d397SChris Mason 13025d4f98a2SYan Zheng btrfs_set_root_node(&root->fs_info->tree_root->root_item, 13035d4f98a2SYan Zheng root->fs_info->tree_root->node); 1304817d52f8SJosef Bacik switch_commit_root(root->fs_info->tree_root); 13055d4f98a2SYan Zheng 13065d4f98a2SYan Zheng btrfs_set_root_node(&root->fs_info->chunk_root->root_item, 13075d4f98a2SYan Zheng root->fs_info->chunk_root->node); 1308817d52f8SJosef Bacik switch_commit_root(root->fs_info->chunk_root); 13095d4f98a2SYan Zheng 13105d4f98a2SYan Zheng update_super_roots(root); 1311e02119d5SChris Mason 1312e02119d5SChris Mason if (!root->fs_info->log_root_recovering) { 13136c41761fSDavid Sterba btrfs_set_super_log_root(root->fs_info->super_copy, 0); 13146c41761fSDavid Sterba btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); 1315e02119d5SChris Mason } 1316e02119d5SChris Mason 13176c41761fSDavid Sterba memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, 13186c41761fSDavid Sterba sizeof(*root->fs_info->super_copy)); 1319ccd467d6SChris Mason 1320f9295749SChris Mason trans->transaction->blocked = 0; 1321a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 1322a4abeea4SJosef Bacik root->fs_info->running_transaction = NULL; 1323a4abeea4SJosef Bacik root->fs_info->trans_no_join = 0; 1324a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 13257585717fSChris Mason mutex_unlock(&root->fs_info->reloc_mutex); 1326b7ec40d7SChris Mason 1327f9295749SChris Mason wake_up(&root->fs_info->transaction_wait); 1328e6dcd2dcSChris Mason 132979154b1bSChris Mason ret = btrfs_write_and_wait_transaction(trans, root); 133079154b1bSChris Mason BUG_ON(ret); 1331a512bbf8SYan Zheng write_ctree_super(trans, root, 0); 13324313b399SChris Mason 1333e02119d5SChris Mason /* 1334e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 1335e02119d5SChris Mason * to go about their business 1336e02119d5SChris Mason */ 1337e02119d5SChris Mason mutex_unlock(&root->fs_info->tree_log_mutex); 1338e02119d5SChris Mason 133911833d66SYan Zheng btrfs_finish_extent_commit(trans, root); 13404313b399SChris Mason 13412c90e5d6SChris Mason cur_trans->commit_done = 1; 1342b7ec40d7SChris Mason 134315ee9bc7SJosef Bacik root->fs_info->last_trans_committed = cur_trans->transid; 1344817d52f8SJosef Bacik 13452c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 13463de4586cSChris Mason 1347a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 134813c5a93eSJosef Bacik list_del_init(&cur_trans->list); 1349a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1350a4abeea4SJosef Bacik 135179154b1bSChris Mason put_transaction(cur_trans); 135278fae27eSChris Mason put_transaction(cur_trans); 135358176a96SJosef Bacik 13541abe9b8aSliubo trace_btrfs_transaction_commit(root); 13551abe9b8aSliubo 1356a2de733cSArne Jansen btrfs_scrub_continue(root); 1357a2de733cSArne Jansen 13589ed74f2dSJosef Bacik if (current->journal_info == trans) 13599ed74f2dSJosef Bacik current->journal_info = NULL; 13609ed74f2dSJosef Bacik 13612c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 136224bbcf04SYan, Zheng 136324bbcf04SYan, Zheng if (current != root->fs_info->transaction_kthread) 136424bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 136524bbcf04SYan, Zheng 136679154b1bSChris Mason return ret; 136779154b1bSChris Mason } 136879154b1bSChris Mason 1369d352ac68SChris Mason /* 1370d352ac68SChris Mason * interface function to delete all the snapshots we have scheduled for deletion 1371d352ac68SChris Mason */ 1372e9d0b13bSChris Mason int btrfs_clean_old_snapshots(struct btrfs_root *root) 1373e9d0b13bSChris Mason { 13745d4f98a2SYan Zheng LIST_HEAD(list); 13755d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = root->fs_info; 1376e9d0b13bSChris Mason 1377a4abeea4SJosef Bacik spin_lock(&fs_info->trans_lock); 13785d4f98a2SYan Zheng list_splice_init(&fs_info->dead_roots, &list); 1379a4abeea4SJosef Bacik spin_unlock(&fs_info->trans_lock); 13805d4f98a2SYan Zheng 13815d4f98a2SYan Zheng while (!list_empty(&list)) { 13825d4f98a2SYan Zheng root = list_entry(list.next, struct btrfs_root, root_list); 138376dda93cSYan, Zheng list_del(&root->root_list); 138476dda93cSYan, Zheng 138516cdcec7SMiao Xie btrfs_kill_all_delayed_nodes(root); 138616cdcec7SMiao Xie 138776dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 138876dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 13893fd0a558SYan, Zheng btrfs_drop_snapshot(root, NULL, 0); 139076dda93cSYan, Zheng else 13913fd0a558SYan, Zheng btrfs_drop_snapshot(root, NULL, 1); 1392e9d0b13bSChris Mason } 1393e9d0b13bSChris Mason return 0; 1394e9d0b13bSChris Mason } 1395