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); 58a4abeea4SJosef Bacik if (root->fs_info->trans_no_join) { 59a4abeea4SJosef Bacik if (!nofail) { 60a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 61a4abeea4SJosef Bacik return -EBUSY; 62a4abeea4SJosef Bacik } 63a4abeea4SJosef Bacik } 64a4abeea4SJosef Bacik 6579154b1bSChris Mason cur_trans = root->fs_info->running_transaction; 66a4abeea4SJosef Bacik if (cur_trans) { 67a4abeea4SJosef Bacik atomic_inc(&cur_trans->use_count); 68a4abeea4SJosef Bacik atomic_inc(&cur_trans->num_writers); 69a4abeea4SJosef Bacik cur_trans->num_joined++; 70a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 71a4abeea4SJosef Bacik return 0; 72a4abeea4SJosef Bacik } 73a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 74a4abeea4SJosef Bacik 75a4abeea4SJosef Bacik cur_trans = kmem_cache_alloc(btrfs_transaction_cachep, GFP_NOFS); 76db5b493aSTsutomu Itoh if (!cur_trans) 77db5b493aSTsutomu Itoh return -ENOMEM; 78a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 79a4abeea4SJosef Bacik if (root->fs_info->running_transaction) { 80a4abeea4SJosef Bacik kmem_cache_free(btrfs_transaction_cachep, cur_trans); 81a4abeea4SJosef Bacik cur_trans = root->fs_info->running_transaction; 82a4abeea4SJosef Bacik atomic_inc(&cur_trans->use_count); 83a4abeea4SJosef Bacik atomic_inc(&cur_trans->num_writers); 84a4abeea4SJosef Bacik cur_trans->num_joined++; 85a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 86a4abeea4SJosef Bacik return 0; 87a4abeea4SJosef Bacik } 8813c5a93eSJosef Bacik atomic_set(&cur_trans->num_writers, 1); 8915ee9bc7SJosef Bacik cur_trans->num_joined = 0; 9079154b1bSChris Mason init_waitqueue_head(&cur_trans->writer_wait); 9179154b1bSChris Mason init_waitqueue_head(&cur_trans->commit_wait); 9279154b1bSChris Mason cur_trans->in_commit = 0; 93f9295749SChris Mason cur_trans->blocked = 0; 94a4abeea4SJosef Bacik /* 95a4abeea4SJosef Bacik * One for this trans handle, one so it will live on until we 96a4abeea4SJosef Bacik * commit the transaction. 97a4abeea4SJosef Bacik */ 98a4abeea4SJosef Bacik atomic_set(&cur_trans->use_count, 2); 9979154b1bSChris Mason cur_trans->commit_done = 0; 10008607c1bSChris Mason cur_trans->start_time = get_seconds(); 10156bec294SChris Mason 1026bef4d31SEric Paris cur_trans->delayed_refs.root = RB_ROOT; 10356bec294SChris Mason cur_trans->delayed_refs.num_entries = 0; 104c3e69d58SChris Mason cur_trans->delayed_refs.num_heads_ready = 0; 105c3e69d58SChris Mason cur_trans->delayed_refs.num_heads = 0; 10656bec294SChris Mason cur_trans->delayed_refs.flushing = 0; 107c3e69d58SChris Mason cur_trans->delayed_refs.run_delayed_start = 0; 108a4abeea4SJosef Bacik spin_lock_init(&cur_trans->commit_lock); 10956bec294SChris Mason spin_lock_init(&cur_trans->delayed_refs.lock); 11056bec294SChris Mason 1113063d29fSChris Mason INIT_LIST_HEAD(&cur_trans->pending_snapshots); 1128fd17795SChris Mason list_add_tail(&cur_trans->list, &root->fs_info->trans_list); 113d1310b2eSChris Mason extent_io_tree_init(&cur_trans->dirty_pages, 114f993c883SDavid Sterba root->fs_info->btree_inode->i_mapping); 115a4abeea4SJosef Bacik root->fs_info->generation++; 116a4abeea4SJosef Bacik cur_trans->transid = root->fs_info->generation; 11748ec2cf8SChris Mason root->fs_info->running_transaction = cur_trans; 118a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 11915ee9bc7SJosef Bacik 12079154b1bSChris Mason return 0; 12179154b1bSChris Mason } 12279154b1bSChris Mason 123d352ac68SChris Mason /* 124d397712bSChris Mason * this does all the record keeping required to make sure that a reference 125d397712bSChris Mason * counted root is properly recorded in a given transaction. This is required 126d397712bSChris Mason * to make sure the old root from before we joined the transaction is deleted 127d397712bSChris Mason * when the transaction commits 128d352ac68SChris Mason */ 1297585717fSChris Mason static int record_root_in_trans(struct btrfs_trans_handle *trans, 1305d4f98a2SYan Zheng struct btrfs_root *root) 1316702ed49SChris Mason { 1325d4f98a2SYan Zheng if (root->ref_cows && root->last_trans < trans->transid) { 1336702ed49SChris Mason WARN_ON(root == root->fs_info->extent_root); 1345d4f98a2SYan Zheng WARN_ON(root->commit_root != root->node); 1355d4f98a2SYan Zheng 1367585717fSChris Mason /* 1377585717fSChris Mason * see below for in_trans_setup usage rules 1387585717fSChris Mason * we have the reloc mutex held now, so there 1397585717fSChris Mason * is only one writer in this function 1407585717fSChris Mason */ 1417585717fSChris Mason root->in_trans_setup = 1; 1427585717fSChris Mason 1437585717fSChris Mason /* make sure readers find in_trans_setup before 1447585717fSChris Mason * they find our root->last_trans update 1457585717fSChris Mason */ 1467585717fSChris Mason smp_wmb(); 1477585717fSChris Mason 148a4abeea4SJosef Bacik spin_lock(&root->fs_info->fs_roots_radix_lock); 149a4abeea4SJosef Bacik if (root->last_trans == trans->transid) { 150a4abeea4SJosef Bacik spin_unlock(&root->fs_info->fs_roots_radix_lock); 151a4abeea4SJosef Bacik return 0; 152a4abeea4SJosef Bacik } 1536702ed49SChris Mason radix_tree_tag_set(&root->fs_info->fs_roots_radix, 1546702ed49SChris Mason (unsigned long)root->root_key.objectid, 1556702ed49SChris Mason BTRFS_ROOT_TRANS_TAG); 156a4abeea4SJosef Bacik spin_unlock(&root->fs_info->fs_roots_radix_lock); 1577585717fSChris Mason root->last_trans = trans->transid; 1587585717fSChris Mason 1597585717fSChris Mason /* this is pretty tricky. We don't want to 1607585717fSChris Mason * take the relocation lock in btrfs_record_root_in_trans 1617585717fSChris Mason * unless we're really doing the first setup for this root in 1627585717fSChris Mason * this transaction. 1637585717fSChris Mason * 1647585717fSChris Mason * Normally we'd use root->last_trans as a flag to decide 1657585717fSChris Mason * if we want to take the expensive mutex. 1667585717fSChris Mason * 1677585717fSChris Mason * But, we have to set root->last_trans before we 1687585717fSChris Mason * init the relocation root, otherwise, we trip over warnings 1697585717fSChris Mason * in ctree.c. The solution used here is to flag ourselves 1707585717fSChris Mason * with root->in_trans_setup. When this is 1, we're still 1717585717fSChris Mason * fixing up the reloc trees and everyone must wait. 1727585717fSChris Mason * 1737585717fSChris Mason * When this is zero, they can trust root->last_trans and fly 1747585717fSChris Mason * through btrfs_record_root_in_trans without having to take the 1757585717fSChris Mason * lock. smp_wmb() makes sure that all the writes above are 1767585717fSChris Mason * done before we pop in the zero below 1777585717fSChris Mason */ 1785d4f98a2SYan Zheng btrfs_init_reloc_root(trans, root); 1797585717fSChris Mason smp_wmb(); 1807585717fSChris Mason root->in_trans_setup = 0; 1816702ed49SChris Mason } 1825d4f98a2SYan Zheng return 0; 1836702ed49SChris Mason } 1845d4f98a2SYan Zheng 1857585717fSChris Mason 1867585717fSChris Mason int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans, 1877585717fSChris Mason struct btrfs_root *root) 1887585717fSChris Mason { 1897585717fSChris Mason if (!root->ref_cows) 1907585717fSChris Mason return 0; 1917585717fSChris Mason 1927585717fSChris Mason /* 1937585717fSChris Mason * see record_root_in_trans for comments about in_trans_setup usage 1947585717fSChris Mason * and barriers 1957585717fSChris Mason */ 1967585717fSChris Mason smp_rmb(); 1977585717fSChris Mason if (root->last_trans == trans->transid && 1987585717fSChris Mason !root->in_trans_setup) 1997585717fSChris Mason return 0; 2007585717fSChris Mason 2017585717fSChris Mason mutex_lock(&root->fs_info->reloc_mutex); 2027585717fSChris Mason record_root_in_trans(trans, root); 2037585717fSChris Mason mutex_unlock(&root->fs_info->reloc_mutex); 2047585717fSChris Mason 2057585717fSChris Mason return 0; 2067585717fSChris Mason } 2077585717fSChris Mason 208d352ac68SChris Mason /* wait for commit against the current transaction to become unblocked 209d352ac68SChris Mason * when this is done, it is safe to start a new transaction, but the current 210d352ac68SChris Mason * transaction might not be fully on disk. 211d352ac68SChris Mason */ 21237d1aeeeSChris Mason static void wait_current_trans(struct btrfs_root *root) 21379154b1bSChris Mason { 214f9295749SChris Mason struct btrfs_transaction *cur_trans; 21579154b1bSChris Mason 216a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 217f9295749SChris Mason cur_trans = root->fs_info->running_transaction; 21837d1aeeeSChris Mason if (cur_trans && cur_trans->blocked) { 21913c5a93eSJosef Bacik atomic_inc(&cur_trans->use_count); 220a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 22172d63ed6SLi Zefan 22272d63ed6SLi Zefan wait_event(root->fs_info->transaction_wait, 22372d63ed6SLi Zefan !cur_trans->blocked); 224f9295749SChris Mason put_transaction(cur_trans); 225a4abeea4SJosef Bacik } else { 226a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 227f9295749SChris Mason } 22837d1aeeeSChris Mason } 22937d1aeeeSChris Mason 230249ac1e5SJosef Bacik enum btrfs_trans_type { 231249ac1e5SJosef Bacik TRANS_START, 232249ac1e5SJosef Bacik TRANS_JOIN, 233249ac1e5SJosef Bacik TRANS_USERSPACE, 2340af3d00bSJosef Bacik TRANS_JOIN_NOLOCK, 235249ac1e5SJosef Bacik }; 236249ac1e5SJosef Bacik 237a22285a6SYan, Zheng static int may_wait_transaction(struct btrfs_root *root, int type) 23837d1aeeeSChris Mason { 239a4abeea4SJosef Bacik if (root->fs_info->log_root_recovering) 240a4abeea4SJosef Bacik return 0; 241a4abeea4SJosef Bacik 242a4abeea4SJosef Bacik if (type == TRANS_USERSPACE) 243a22285a6SYan, Zheng return 1; 244a4abeea4SJosef Bacik 245a4abeea4SJosef Bacik if (type == TRANS_START && 246a4abeea4SJosef Bacik !atomic_read(&root->fs_info->open_ioctl_trans)) 247a4abeea4SJosef Bacik return 1; 248a4abeea4SJosef Bacik 249a22285a6SYan, Zheng return 0; 250a22285a6SYan, Zheng } 251a22285a6SYan, Zheng 252a22285a6SYan, Zheng static struct btrfs_trans_handle *start_transaction(struct btrfs_root *root, 253a22285a6SYan, Zheng u64 num_items, int type) 254a22285a6SYan, Zheng { 255a22285a6SYan, Zheng struct btrfs_trans_handle *h; 256a22285a6SYan, Zheng struct btrfs_transaction *cur_trans; 257b5009945SJosef Bacik u64 num_bytes = 0; 258a22285a6SYan, Zheng int ret; 259acce952bSliubo 260acce952bSliubo if (root->fs_info->fs_state & BTRFS_SUPER_FLAG_ERROR) 261acce952bSliubo return ERR_PTR(-EROFS); 2622a1eb461SJosef Bacik 2632a1eb461SJosef Bacik if (current->journal_info) { 2642a1eb461SJosef Bacik WARN_ON(type != TRANS_JOIN && type != TRANS_JOIN_NOLOCK); 2652a1eb461SJosef Bacik h = current->journal_info; 2662a1eb461SJosef Bacik h->use_count++; 2672a1eb461SJosef Bacik h->orig_rsv = h->block_rsv; 2682a1eb461SJosef Bacik h->block_rsv = NULL; 2692a1eb461SJosef Bacik goto got_it; 2702a1eb461SJosef Bacik } 271b5009945SJosef Bacik 272b5009945SJosef Bacik /* 273b5009945SJosef Bacik * Do the reservation before we join the transaction so we can do all 274b5009945SJosef Bacik * the appropriate flushing if need be. 275b5009945SJosef Bacik */ 276b5009945SJosef Bacik if (num_items > 0 && root != root->fs_info->chunk_root) { 277b5009945SJosef Bacik num_bytes = btrfs_calc_trans_metadata_size(root, num_items); 2784a92b1b8SJosef Bacik ret = btrfs_block_rsv_add(root, 279b5009945SJosef Bacik &root->fs_info->trans_block_rsv, 280b5009945SJosef Bacik num_bytes); 281b5009945SJosef Bacik if (ret) 282b5009945SJosef Bacik return ERR_PTR(ret); 283b5009945SJosef Bacik } 284a22285a6SYan, Zheng again: 285a22285a6SYan, Zheng h = kmem_cache_alloc(btrfs_trans_handle_cachep, GFP_NOFS); 286a22285a6SYan, Zheng if (!h) 287a22285a6SYan, Zheng return ERR_PTR(-ENOMEM); 288a22285a6SYan, Zheng 289a22285a6SYan, Zheng if (may_wait_transaction(root, type)) 29037d1aeeeSChris Mason wait_current_trans(root); 291a22285a6SYan, Zheng 292a4abeea4SJosef Bacik do { 293a4abeea4SJosef Bacik ret = join_transaction(root, type == TRANS_JOIN_NOLOCK); 294a4abeea4SJosef Bacik if (ret == -EBUSY) 295a4abeea4SJosef Bacik wait_current_trans(root); 296a4abeea4SJosef Bacik } while (ret == -EBUSY); 297a4abeea4SJosef Bacik 298db5b493aSTsutomu Itoh if (ret < 0) { 2996e8df2aeSYoshinori Sano kmem_cache_free(btrfs_trans_handle_cachep, h); 300db5b493aSTsutomu Itoh return ERR_PTR(ret); 301db5b493aSTsutomu Itoh } 3020f7d52f4SChris Mason 303a22285a6SYan, Zheng cur_trans = root->fs_info->running_transaction; 304a22285a6SYan, Zheng 305a22285a6SYan, Zheng h->transid = cur_trans->transid; 306a22285a6SYan, Zheng h->transaction = cur_trans; 30779154b1bSChris Mason h->blocks_used = 0; 308a22285a6SYan, Zheng h->bytes_reserved = 0; 30956bec294SChris Mason h->delayed_ref_updates = 0; 3102a1eb461SJosef Bacik h->use_count = 1; 311f0486c68SYan, Zheng h->block_rsv = NULL; 3122a1eb461SJosef Bacik h->orig_rsv = NULL; 313b7ec40d7SChris Mason 314a22285a6SYan, Zheng smp_mb(); 315a22285a6SYan, Zheng if (cur_trans->blocked && may_wait_transaction(root, type)) { 316a22285a6SYan, Zheng btrfs_commit_transaction(h, root); 317a22285a6SYan, Zheng goto again; 318a22285a6SYan, Zheng } 3199ed74f2dSJosef Bacik 320b5009945SJosef Bacik if (num_bytes) { 321b5009945SJosef Bacik h->block_rsv = &root->fs_info->trans_block_rsv; 322b5009945SJosef Bacik h->bytes_reserved = num_bytes; 323a22285a6SYan, Zheng } 324a22285a6SYan, Zheng 3252a1eb461SJosef Bacik got_it: 326a4abeea4SJosef Bacik btrfs_record_root_in_trans(h, root); 327a22285a6SYan, Zheng 328a22285a6SYan, Zheng if (!current->journal_info && type != TRANS_USERSPACE) 329a22285a6SYan, Zheng current->journal_info = h; 33079154b1bSChris Mason return h; 33179154b1bSChris Mason } 33279154b1bSChris Mason 333f9295749SChris Mason struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root, 334a22285a6SYan, Zheng int num_items) 335f9295749SChris Mason { 336a22285a6SYan, Zheng return start_transaction(root, num_items, TRANS_START); 337f9295749SChris Mason } 3387a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root) 339f9295749SChris Mason { 340a22285a6SYan, Zheng return start_transaction(root, 0, TRANS_JOIN); 341f9295749SChris Mason } 342f9295749SChris Mason 3437a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root *root) 3440af3d00bSJosef Bacik { 3450af3d00bSJosef Bacik return start_transaction(root, 0, TRANS_JOIN_NOLOCK); 3460af3d00bSJosef Bacik } 3470af3d00bSJosef Bacik 3487a7eaa40SJosef Bacik struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *root) 3499ca9ee09SSage Weil { 3507a7eaa40SJosef Bacik return start_transaction(root, 0, TRANS_USERSPACE); 3519ca9ee09SSage Weil } 3529ca9ee09SSage Weil 353d352ac68SChris Mason /* wait for a transaction commit to be fully complete */ 354b9c8300cSLi Zefan static noinline void wait_for_commit(struct btrfs_root *root, 35589ce8a63SChris Mason struct btrfs_transaction *commit) 35689ce8a63SChris Mason { 35772d63ed6SLi Zefan wait_event(commit->commit_wait, commit->commit_done); 35889ce8a63SChris Mason } 35989ce8a63SChris Mason 36046204592SSage Weil int btrfs_wait_for_commit(struct btrfs_root *root, u64 transid) 36146204592SSage Weil { 36246204592SSage Weil struct btrfs_transaction *cur_trans = NULL, *t; 36346204592SSage Weil int ret; 36446204592SSage Weil 36546204592SSage Weil ret = 0; 36646204592SSage Weil if (transid) { 36746204592SSage Weil if (transid <= root->fs_info->last_trans_committed) 368a4abeea4SJosef Bacik goto out; 36946204592SSage Weil 37046204592SSage Weil /* find specified transaction */ 371a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 37246204592SSage Weil list_for_each_entry(t, &root->fs_info->trans_list, list) { 37346204592SSage Weil if (t->transid == transid) { 37446204592SSage Weil cur_trans = t; 375a4abeea4SJosef Bacik atomic_inc(&cur_trans->use_count); 37646204592SSage Weil break; 37746204592SSage Weil } 37846204592SSage Weil if (t->transid > transid) 37946204592SSage Weil break; 38046204592SSage Weil } 381a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 38246204592SSage Weil ret = -EINVAL; 38346204592SSage Weil if (!cur_trans) 384a4abeea4SJosef Bacik goto out; /* bad transid */ 38546204592SSage Weil } else { 38646204592SSage Weil /* find newest transaction that is committing | committed */ 387a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 38846204592SSage Weil list_for_each_entry_reverse(t, &root->fs_info->trans_list, 38946204592SSage Weil list) { 39046204592SSage Weil if (t->in_commit) { 39146204592SSage Weil if (t->commit_done) 3923473f3c0SJosef Bacik break; 39346204592SSage Weil cur_trans = t; 394a4abeea4SJosef Bacik atomic_inc(&cur_trans->use_count); 39546204592SSage Weil break; 39646204592SSage Weil } 39746204592SSage Weil } 398a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 39946204592SSage Weil if (!cur_trans) 400a4abeea4SJosef Bacik goto out; /* nothing committing|committed */ 40146204592SSage Weil } 40246204592SSage Weil 40346204592SSage Weil wait_for_commit(root, cur_trans); 40446204592SSage Weil 40546204592SSage Weil put_transaction(cur_trans); 40646204592SSage Weil ret = 0; 407a4abeea4SJosef Bacik out: 40846204592SSage Weil return ret; 40946204592SSage Weil } 41046204592SSage Weil 41137d1aeeeSChris Mason void btrfs_throttle(struct btrfs_root *root) 41237d1aeeeSChris Mason { 413a4abeea4SJosef Bacik if (!atomic_read(&root->fs_info->open_ioctl_trans)) 41437d1aeeeSChris Mason wait_current_trans(root); 41537d1aeeeSChris Mason } 41637d1aeeeSChris Mason 4178929ecfaSYan, Zheng static int should_end_transaction(struct btrfs_trans_handle *trans, 4188929ecfaSYan, Zheng struct btrfs_root *root) 4198929ecfaSYan, Zheng { 4208929ecfaSYan, Zheng int ret; 42136ba022aSJosef Bacik 42236ba022aSJosef Bacik ret = btrfs_block_rsv_check(root, &root->fs_info->global_block_rsv, 5); 4238929ecfaSYan, Zheng return ret ? 1 : 0; 4248929ecfaSYan, Zheng } 4258929ecfaSYan, Zheng 4268929ecfaSYan, Zheng int btrfs_should_end_transaction(struct btrfs_trans_handle *trans, 4278929ecfaSYan, Zheng struct btrfs_root *root) 4288929ecfaSYan, Zheng { 4298929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 4309c8d86dbSJosef Bacik struct btrfs_block_rsv *rsv = trans->block_rsv; 4318929ecfaSYan, Zheng int updates; 4328929ecfaSYan, Zheng 433a4abeea4SJosef Bacik smp_mb(); 4348929ecfaSYan, Zheng if (cur_trans->blocked || cur_trans->delayed_refs.flushing) 4358929ecfaSYan, Zheng return 1; 4368929ecfaSYan, Zheng 4379c8d86dbSJosef Bacik /* 4389c8d86dbSJosef Bacik * We need to do this in case we're deleting csums so the global block 4399c8d86dbSJosef Bacik * rsv get's used instead of the csum block rsv. 4409c8d86dbSJosef Bacik */ 4419c8d86dbSJosef Bacik trans->block_rsv = NULL; 4429c8d86dbSJosef Bacik 4438929ecfaSYan, Zheng updates = trans->delayed_ref_updates; 4448929ecfaSYan, Zheng trans->delayed_ref_updates = 0; 4458929ecfaSYan, Zheng if (updates) 4468929ecfaSYan, Zheng btrfs_run_delayed_refs(trans, root, updates); 4478929ecfaSYan, Zheng 4489c8d86dbSJosef Bacik trans->block_rsv = rsv; 4499c8d86dbSJosef Bacik 4508929ecfaSYan, Zheng return should_end_transaction(trans, root); 4518929ecfaSYan, Zheng } 4528929ecfaSYan, Zheng 45389ce8a63SChris Mason static int __btrfs_end_transaction(struct btrfs_trans_handle *trans, 4540af3d00bSJosef Bacik struct btrfs_root *root, int throttle, int lock) 45579154b1bSChris Mason { 4568929ecfaSYan, Zheng struct btrfs_transaction *cur_trans = trans->transaction; 457ab78c84dSChris Mason struct btrfs_fs_info *info = root->fs_info; 458c3e69d58SChris Mason int count = 0; 459d6e4a428SChris Mason 4602a1eb461SJosef Bacik if (--trans->use_count) { 4612a1eb461SJosef Bacik trans->block_rsv = trans->orig_rsv; 4622a1eb461SJosef Bacik return 0; 4632a1eb461SJosef Bacik } 4642a1eb461SJosef Bacik 465b24e03dbSJosef Bacik btrfs_trans_release_metadata(trans, root); 4664c13d758SJosef Bacik trans->block_rsv = NULL; 467c3e69d58SChris Mason while (count < 4) { 468c3e69d58SChris Mason unsigned long cur = trans->delayed_ref_updates; 469c3e69d58SChris Mason trans->delayed_ref_updates = 0; 470c3e69d58SChris Mason if (cur && 471c3e69d58SChris Mason trans->transaction->delayed_refs.num_heads_ready > 64) { 472c3e69d58SChris Mason trans->delayed_ref_updates = 0; 473b7ec40d7SChris Mason 474b7ec40d7SChris Mason /* 475b7ec40d7SChris Mason * do a full flush if the transaction is trying 476b7ec40d7SChris Mason * to close 477b7ec40d7SChris Mason */ 478b7ec40d7SChris Mason if (trans->transaction->delayed_refs.flushing) 479b7ec40d7SChris Mason cur = 0; 480c3e69d58SChris Mason btrfs_run_delayed_refs(trans, root, cur); 481c3e69d58SChris Mason } else { 482c3e69d58SChris Mason break; 483c3e69d58SChris Mason } 484c3e69d58SChris Mason count++; 48556bec294SChris Mason } 48656bec294SChris Mason 487a4abeea4SJosef Bacik if (lock && !atomic_read(&root->fs_info->open_ioctl_trans) && 488a4abeea4SJosef Bacik should_end_transaction(trans, root)) { 4898929ecfaSYan, Zheng trans->transaction->blocked = 1; 490a4abeea4SJosef Bacik smp_wmb(); 491a4abeea4SJosef Bacik } 4928929ecfaSYan, Zheng 4930af3d00bSJosef Bacik if (lock && cur_trans->blocked && !cur_trans->in_commit) { 49481317fdeSJosef Bacik if (throttle) { 49581317fdeSJosef Bacik /* 49681317fdeSJosef Bacik * We may race with somebody else here so end up having 49781317fdeSJosef Bacik * to call end_transaction on ourselves again, so inc 49881317fdeSJosef Bacik * our use_count. 49981317fdeSJosef Bacik */ 50081317fdeSJosef Bacik trans->use_count++; 5018929ecfaSYan, Zheng return btrfs_commit_transaction(trans, root); 50281317fdeSJosef Bacik } else { 5038929ecfaSYan, Zheng wake_up_process(info->transaction_kthread); 5048929ecfaSYan, Zheng } 50581317fdeSJosef Bacik } 5068929ecfaSYan, Zheng 5078929ecfaSYan, Zheng WARN_ON(cur_trans != info->running_transaction); 50813c5a93eSJosef Bacik WARN_ON(atomic_read(&cur_trans->num_writers) < 1); 50913c5a93eSJosef Bacik atomic_dec(&cur_trans->num_writers); 51089ce8a63SChris Mason 51199d16cbcSSage Weil smp_mb(); 51279154b1bSChris Mason if (waitqueue_active(&cur_trans->writer_wait)) 51379154b1bSChris Mason wake_up(&cur_trans->writer_wait); 51479154b1bSChris Mason put_transaction(cur_trans); 5159ed74f2dSJosef Bacik 5169ed74f2dSJosef Bacik if (current->journal_info == trans) 5179ed74f2dSJosef Bacik current->journal_info = NULL; 518d6025579SChris Mason memset(trans, 0, sizeof(*trans)); 5192c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 520ab78c84dSChris Mason 52124bbcf04SYan, Zheng if (throttle) 52224bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 52324bbcf04SYan, Zheng 52479154b1bSChris Mason return 0; 52579154b1bSChris Mason } 52679154b1bSChris Mason 52789ce8a63SChris Mason int btrfs_end_transaction(struct btrfs_trans_handle *trans, 52889ce8a63SChris Mason struct btrfs_root *root) 52989ce8a63SChris Mason { 53016cdcec7SMiao Xie int ret; 53116cdcec7SMiao Xie 53216cdcec7SMiao Xie ret = __btrfs_end_transaction(trans, root, 0, 1); 53316cdcec7SMiao Xie if (ret) 53416cdcec7SMiao Xie return ret; 53516cdcec7SMiao Xie return 0; 53689ce8a63SChris Mason } 53789ce8a63SChris Mason 53889ce8a63SChris Mason int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans, 53989ce8a63SChris Mason struct btrfs_root *root) 54089ce8a63SChris Mason { 54116cdcec7SMiao Xie int ret; 54216cdcec7SMiao Xie 54316cdcec7SMiao Xie ret = __btrfs_end_transaction(trans, root, 1, 1); 54416cdcec7SMiao Xie if (ret) 54516cdcec7SMiao Xie return ret; 54616cdcec7SMiao Xie return 0; 5470af3d00bSJosef Bacik } 5480af3d00bSJosef Bacik 5490af3d00bSJosef Bacik int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans, 5500af3d00bSJosef Bacik struct btrfs_root *root) 5510af3d00bSJosef Bacik { 55216cdcec7SMiao Xie int ret; 55316cdcec7SMiao Xie 55416cdcec7SMiao Xie ret = __btrfs_end_transaction(trans, root, 0, 0); 55516cdcec7SMiao Xie if (ret) 55616cdcec7SMiao Xie return ret; 55716cdcec7SMiao Xie return 0; 55816cdcec7SMiao Xie } 55916cdcec7SMiao Xie 56016cdcec7SMiao Xie int btrfs_end_transaction_dmeta(struct btrfs_trans_handle *trans, 56116cdcec7SMiao Xie struct btrfs_root *root) 56216cdcec7SMiao Xie { 56316cdcec7SMiao Xie return __btrfs_end_transaction(trans, root, 1, 1); 56489ce8a63SChris Mason } 56589ce8a63SChris Mason 566d352ac68SChris Mason /* 567d352ac68SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 568d352ac68SChris Mason * them in one of two extent_io trees. This is used to make sure all of 569690587d1SChris Mason * those extents are sent to disk but does not wait on them 570d352ac68SChris Mason */ 571690587d1SChris Mason int btrfs_write_marked_extents(struct btrfs_root *root, 5728cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 57379154b1bSChris Mason { 574777e6bd7SChris Mason int err = 0; 5757c4452b9SChris Mason int werr = 0; 5761728366eSJosef Bacik struct address_space *mapping = root->fs_info->btree_inode->i_mapping; 577777e6bd7SChris Mason u64 start = 0; 5785f39d397SChris Mason u64 end; 5797c4452b9SChris Mason 5801728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 5811728366eSJosef Bacik mark)) { 5821728366eSJosef Bacik convert_extent_bit(dirty_pages, start, end, EXTENT_NEED_WAIT, mark, 5831728366eSJosef Bacik GFP_NOFS); 5841728366eSJosef Bacik err = filemap_fdatawrite_range(mapping, start, end); 5857c4452b9SChris Mason if (err) 5867c4452b9SChris Mason werr = err; 5871728366eSJosef Bacik cond_resched(); 5881728366eSJosef Bacik start = end + 1; 5897c4452b9SChris Mason } 590690587d1SChris Mason if (err) 591690587d1SChris Mason werr = err; 592690587d1SChris Mason return werr; 593690587d1SChris Mason } 594690587d1SChris Mason 595690587d1SChris Mason /* 596690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 597690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 598690587d1SChris Mason * those extents are on disk for transaction or log commit. We wait 599690587d1SChris Mason * on all the pages and clear them from the dirty pages state tree 600690587d1SChris Mason */ 601690587d1SChris Mason int btrfs_wait_marked_extents(struct btrfs_root *root, 6028cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 603690587d1SChris Mason { 604690587d1SChris Mason int err = 0; 605690587d1SChris Mason int werr = 0; 6061728366eSJosef Bacik struct address_space *mapping = root->fs_info->btree_inode->i_mapping; 607690587d1SChris Mason u64 start = 0; 608690587d1SChris Mason u64 end; 609690587d1SChris Mason 6101728366eSJosef Bacik while (!find_first_extent_bit(dirty_pages, start, &start, &end, 6111728366eSJosef Bacik EXTENT_NEED_WAIT)) { 6121728366eSJosef Bacik clear_extent_bits(dirty_pages, start, end, EXTENT_NEED_WAIT, GFP_NOFS); 6131728366eSJosef Bacik err = filemap_fdatawait_range(mapping, start, end); 614777e6bd7SChris Mason if (err) 615777e6bd7SChris Mason werr = err; 616777e6bd7SChris Mason cond_resched(); 6171728366eSJosef Bacik start = end + 1; 618777e6bd7SChris Mason } 6197c4452b9SChris Mason if (err) 6207c4452b9SChris Mason werr = err; 6217c4452b9SChris Mason return werr; 62279154b1bSChris Mason } 62379154b1bSChris Mason 624690587d1SChris Mason /* 625690587d1SChris Mason * when btree blocks are allocated, they have some corresponding bits set for 626690587d1SChris Mason * them in one of two extent_io trees. This is used to make sure all of 627690587d1SChris Mason * those extents are on disk for transaction or log commit 628690587d1SChris Mason */ 629690587d1SChris Mason int btrfs_write_and_wait_marked_extents(struct btrfs_root *root, 6308cef4e16SYan, Zheng struct extent_io_tree *dirty_pages, int mark) 631690587d1SChris Mason { 632690587d1SChris Mason int ret; 633690587d1SChris Mason int ret2; 634690587d1SChris Mason 6358cef4e16SYan, Zheng ret = btrfs_write_marked_extents(root, dirty_pages, mark); 6368cef4e16SYan, Zheng ret2 = btrfs_wait_marked_extents(root, dirty_pages, mark); 637bf0da8c1SChris Mason 638bf0da8c1SChris Mason if (ret) 639bf0da8c1SChris Mason return ret; 640bf0da8c1SChris Mason if (ret2) 641bf0da8c1SChris Mason return ret2; 642bf0da8c1SChris Mason return 0; 643690587d1SChris Mason } 644690587d1SChris Mason 645d0c803c4SChris Mason int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans, 646d0c803c4SChris Mason struct btrfs_root *root) 647d0c803c4SChris Mason { 648d0c803c4SChris Mason if (!trans || !trans->transaction) { 649d0c803c4SChris Mason struct inode *btree_inode; 650d0c803c4SChris Mason btree_inode = root->fs_info->btree_inode; 651d0c803c4SChris Mason return filemap_write_and_wait(btree_inode->i_mapping); 652d0c803c4SChris Mason } 653d0c803c4SChris Mason return btrfs_write_and_wait_marked_extents(root, 6548cef4e16SYan, Zheng &trans->transaction->dirty_pages, 6558cef4e16SYan, Zheng EXTENT_DIRTY); 656d0c803c4SChris Mason } 657d0c803c4SChris Mason 658d352ac68SChris Mason /* 659d352ac68SChris Mason * this is used to update the root pointer in the tree of tree roots. 660d352ac68SChris Mason * 661d352ac68SChris Mason * But, in the case of the extent allocation tree, updating the root 662d352ac68SChris Mason * pointer may allocate blocks which may change the root of the extent 663d352ac68SChris Mason * allocation tree. 664d352ac68SChris Mason * 665d352ac68SChris Mason * So, this loops and repeats and makes sure the cowonly root didn't 666d352ac68SChris Mason * change while the root pointer was being updated in the metadata. 667d352ac68SChris Mason */ 6680b86a832SChris Mason static int update_cowonly_root(struct btrfs_trans_handle *trans, 66979154b1bSChris Mason struct btrfs_root *root) 67079154b1bSChris Mason { 67179154b1bSChris Mason int ret; 6720b86a832SChris Mason u64 old_root_bytenr; 67386b9f2ecSYan, Zheng u64 old_root_used; 6740b86a832SChris Mason struct btrfs_root *tree_root = root->fs_info->tree_root; 67579154b1bSChris Mason 67686b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 6770b86a832SChris Mason btrfs_write_dirty_block_groups(trans, root); 67856bec294SChris Mason 67979154b1bSChris Mason while (1) { 6800b86a832SChris Mason old_root_bytenr = btrfs_root_bytenr(&root->root_item); 68186b9f2ecSYan, Zheng if (old_root_bytenr == root->node->start && 68286b9f2ecSYan, Zheng old_root_used == btrfs_root_used(&root->root_item)) 68379154b1bSChris Mason break; 68487ef2bb4SChris Mason 6855d4f98a2SYan Zheng btrfs_set_root_node(&root->root_item, root->node); 68679154b1bSChris Mason ret = btrfs_update_root(trans, tree_root, 6870b86a832SChris Mason &root->root_key, 6880b86a832SChris Mason &root->root_item); 68979154b1bSChris Mason BUG_ON(ret); 69056bec294SChris Mason 69186b9f2ecSYan, Zheng old_root_used = btrfs_root_used(&root->root_item); 6924a8c9a62SYan Zheng ret = btrfs_write_dirty_block_groups(trans, root); 69356bec294SChris Mason BUG_ON(ret); 6940b86a832SChris Mason } 695276e680dSYan Zheng 696276e680dSYan Zheng if (root != root->fs_info->extent_root) 697817d52f8SJosef Bacik switch_commit_root(root); 698276e680dSYan Zheng 6990b86a832SChris Mason return 0; 7000b86a832SChris Mason } 7010b86a832SChris Mason 702d352ac68SChris Mason /* 703d352ac68SChris Mason * update all the cowonly tree roots on disk 704d352ac68SChris Mason */ 7055d4f98a2SYan Zheng static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans, 7060b86a832SChris Mason struct btrfs_root *root) 7070b86a832SChris Mason { 7080b86a832SChris Mason struct btrfs_fs_info *fs_info = root->fs_info; 7090b86a832SChris Mason struct list_head *next; 71084234f3aSYan Zheng struct extent_buffer *eb; 71156bec294SChris Mason int ret; 71284234f3aSYan Zheng 71356bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 71456bec294SChris Mason BUG_ON(ret); 71587ef2bb4SChris Mason 71684234f3aSYan Zheng eb = btrfs_lock_root_node(fs_info->tree_root); 7179fa8cfe7SChris Mason btrfs_cow_block(trans, fs_info->tree_root, eb, NULL, 0, &eb); 71884234f3aSYan Zheng btrfs_tree_unlock(eb); 71984234f3aSYan Zheng free_extent_buffer(eb); 7200b86a832SChris Mason 72156bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 72256bec294SChris Mason BUG_ON(ret); 72387ef2bb4SChris Mason 7240b86a832SChris Mason while (!list_empty(&fs_info->dirty_cowonly_roots)) { 7250b86a832SChris Mason next = fs_info->dirty_cowonly_roots.next; 7260b86a832SChris Mason list_del_init(next); 7270b86a832SChris Mason root = list_entry(next, struct btrfs_root, dirty_list); 72887ef2bb4SChris Mason 7290b86a832SChris Mason update_cowonly_root(trans, root); 73079154b1bSChris Mason } 731276e680dSYan Zheng 732276e680dSYan Zheng down_write(&fs_info->extent_commit_sem); 733276e680dSYan Zheng switch_commit_root(fs_info->extent_root); 734276e680dSYan Zheng up_write(&fs_info->extent_commit_sem); 735276e680dSYan Zheng 73679154b1bSChris Mason return 0; 73779154b1bSChris Mason } 73879154b1bSChris Mason 739d352ac68SChris Mason /* 740d352ac68SChris Mason * dead roots are old snapshots that need to be deleted. This allocates 741d352ac68SChris Mason * a dirty root struct and adds it into the list of dead roots that need to 742d352ac68SChris Mason * be deleted 743d352ac68SChris Mason */ 7445d4f98a2SYan Zheng int btrfs_add_dead_root(struct btrfs_root *root) 7455eda7b5eSChris Mason { 746a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 7475d4f98a2SYan Zheng list_add(&root->root_list, &root->fs_info->dead_roots); 748a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 7495eda7b5eSChris Mason return 0; 7505eda7b5eSChris Mason } 7515eda7b5eSChris Mason 752d352ac68SChris Mason /* 7535d4f98a2SYan Zheng * update all the cowonly tree roots on disk 754d352ac68SChris Mason */ 7555d4f98a2SYan Zheng static noinline int commit_fs_roots(struct btrfs_trans_handle *trans, 7565d4f98a2SYan Zheng struct btrfs_root *root) 7570f7d52f4SChris Mason { 7580f7d52f4SChris Mason struct btrfs_root *gang[8]; 7595d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = root->fs_info; 7600f7d52f4SChris Mason int i; 7610f7d52f4SChris Mason int ret; 76254aa1f4dSChris Mason int err = 0; 76354aa1f4dSChris Mason 764a4abeea4SJosef Bacik spin_lock(&fs_info->fs_roots_radix_lock); 7650f7d52f4SChris Mason while (1) { 7665d4f98a2SYan Zheng ret = radix_tree_gang_lookup_tag(&fs_info->fs_roots_radix, 7675d4f98a2SYan Zheng (void **)gang, 0, 7680f7d52f4SChris Mason ARRAY_SIZE(gang), 7690f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 7700f7d52f4SChris Mason if (ret == 0) 7710f7d52f4SChris Mason break; 7720f7d52f4SChris Mason for (i = 0; i < ret; i++) { 7730f7d52f4SChris Mason root = gang[i]; 7745d4f98a2SYan Zheng radix_tree_tag_clear(&fs_info->fs_roots_radix, 7752619ba1fSChris Mason (unsigned long)root->root_key.objectid, 7760f7d52f4SChris Mason BTRFS_ROOT_TRANS_TAG); 777a4abeea4SJosef Bacik spin_unlock(&fs_info->fs_roots_radix_lock); 77831153d81SYan Zheng 779e02119d5SChris Mason btrfs_free_log(trans, root); 7805d4f98a2SYan Zheng btrfs_update_reloc_root(trans, root); 781d68fc57bSYan, Zheng btrfs_orphan_commit_root(trans, root); 782e02119d5SChris Mason 78382d5902dSLi Zefan btrfs_save_ino_cache(root, trans); 78482d5902dSLi Zefan 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) { 8824a92b1b8SJosef Bacik ret = btrfs_block_rsv_add(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 9475d4f98a2SYan Zheng btrfs_set_root_node(new_root_item, tmp); 948a22285a6SYan, Zheng /* record when the snapshot was created in key.offset */ 949a22285a6SYan, Zheng key.offset = trans->transid; 950a22285a6SYan, Zheng ret = btrfs_insert_root(trans, tree_root, &key, new_root_item); 951925baeddSChris Mason btrfs_tree_unlock(tmp); 9523063d29fSChris Mason free_extent_buffer(tmp); 9530660b5afSChris Mason BUG_ON(ret); 9540660b5afSChris Mason 955a22285a6SYan, Zheng /* 956a22285a6SYan, Zheng * insert root back/forward references 957a22285a6SYan, Zheng */ 958a22285a6SYan, Zheng ret = btrfs_add_root_ref(trans, tree_root, objectid, 959a22285a6SYan, Zheng parent_root->root_key.objectid, 96033345d01SLi Zefan btrfs_ino(parent_inode), index, 961a22285a6SYan, Zheng dentry->d_name.name, dentry->d_name.len); 962a22285a6SYan, Zheng BUG_ON(ret); 9636a912213SJosef Bacik dput(parent); 964a22285a6SYan, Zheng 965a22285a6SYan, Zheng key.offset = (u64)-1; 966a22285a6SYan, Zheng pending->snap = btrfs_read_fs_root_no_name(root->fs_info, &key); 967a22285a6SYan, Zheng BUG_ON(IS_ERR(pending->snap)); 968d68fc57bSYan, Zheng 9693fd0a558SYan, Zheng btrfs_reloc_post_snapshot(trans, pending); 9703063d29fSChris Mason fail: 9716bdb72deSSage Weil kfree(new_root_item); 97298c9942aSLiu Bo trans->block_rsv = rsv; 973a22285a6SYan, Zheng btrfs_block_rsv_release(root, &pending->block_rsv, (u64)-1); 974a22285a6SYan, Zheng return 0; 9753063d29fSChris Mason } 9763063d29fSChris Mason 977d352ac68SChris Mason /* 978d352ac68SChris Mason * create all the snapshots we've scheduled for creation 979d352ac68SChris Mason */ 98080b6794dSChris Mason static noinline int create_pending_snapshots(struct btrfs_trans_handle *trans, 9813063d29fSChris Mason struct btrfs_fs_info *fs_info) 9823063d29fSChris Mason { 9833063d29fSChris Mason struct btrfs_pending_snapshot *pending; 9843063d29fSChris Mason struct list_head *head = &trans->transaction->pending_snapshots; 9853de4586cSChris Mason int ret; 9863de4586cSChris Mason 987c6e30871SQinghuang Feng list_for_each_entry(pending, head, list) { 9883de4586cSChris Mason ret = create_pending_snapshot(trans, fs_info, pending); 9893de4586cSChris Mason BUG_ON(ret); 9903de4586cSChris Mason } 9913de4586cSChris Mason return 0; 9923de4586cSChris Mason } 9933de4586cSChris Mason 9945d4f98a2SYan Zheng static void update_super_roots(struct btrfs_root *root) 9955d4f98a2SYan Zheng { 9965d4f98a2SYan Zheng struct btrfs_root_item *root_item; 9975d4f98a2SYan Zheng struct btrfs_super_block *super; 9985d4f98a2SYan Zheng 9996c41761fSDavid Sterba super = root->fs_info->super_copy; 10005d4f98a2SYan Zheng 10015d4f98a2SYan Zheng root_item = &root->fs_info->chunk_root->root_item; 10025d4f98a2SYan Zheng super->chunk_root = root_item->bytenr; 10035d4f98a2SYan Zheng super->chunk_root_generation = root_item->generation; 10045d4f98a2SYan Zheng super->chunk_root_level = root_item->level; 10055d4f98a2SYan Zheng 10065d4f98a2SYan Zheng root_item = &root->fs_info->tree_root->root_item; 10075d4f98a2SYan Zheng super->root = root_item->bytenr; 10085d4f98a2SYan Zheng super->generation = root_item->generation; 10095d4f98a2SYan Zheng super->root_level = root_item->level; 101073bc1876SJosef Bacik if (btrfs_test_opt(root, SPACE_CACHE)) 10110af3d00bSJosef Bacik super->cache_generation = root_item->generation; 10125d4f98a2SYan Zheng } 10135d4f98a2SYan Zheng 1014f36f3042SChris Mason int btrfs_transaction_in_commit(struct btrfs_fs_info *info) 1015f36f3042SChris Mason { 1016f36f3042SChris Mason int ret = 0; 1017a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 1018f36f3042SChris Mason if (info->running_transaction) 1019f36f3042SChris Mason ret = info->running_transaction->in_commit; 1020a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 1021f36f3042SChris Mason return ret; 1022f36f3042SChris Mason } 1023f36f3042SChris Mason 10248929ecfaSYan, Zheng int btrfs_transaction_blocked(struct btrfs_fs_info *info) 10258929ecfaSYan, Zheng { 10268929ecfaSYan, Zheng int ret = 0; 1027a4abeea4SJosef Bacik spin_lock(&info->trans_lock); 10288929ecfaSYan, Zheng if (info->running_transaction) 10298929ecfaSYan, Zheng ret = info->running_transaction->blocked; 1030a4abeea4SJosef Bacik spin_unlock(&info->trans_lock); 10318929ecfaSYan, Zheng return ret; 10328929ecfaSYan, Zheng } 10338929ecfaSYan, Zheng 1034bb9c12c9SSage Weil /* 1035bb9c12c9SSage Weil * wait for the current transaction commit to start and block subsequent 1036bb9c12c9SSage Weil * transaction joins 1037bb9c12c9SSage Weil */ 1038bb9c12c9SSage Weil static void wait_current_trans_commit_start(struct btrfs_root *root, 1039bb9c12c9SSage Weil struct btrfs_transaction *trans) 1040bb9c12c9SSage Weil { 104172d63ed6SLi Zefan wait_event(root->fs_info->transaction_blocked_wait, trans->in_commit); 1042bb9c12c9SSage Weil } 1043bb9c12c9SSage Weil 1044bb9c12c9SSage Weil /* 1045bb9c12c9SSage Weil * wait for the current transaction to start and then become unblocked. 1046bb9c12c9SSage Weil * caller holds ref. 1047bb9c12c9SSage Weil */ 1048bb9c12c9SSage Weil static void wait_current_trans_commit_start_and_unblock(struct btrfs_root *root, 1049bb9c12c9SSage Weil struct btrfs_transaction *trans) 1050bb9c12c9SSage Weil { 105172d63ed6SLi Zefan wait_event(root->fs_info->transaction_wait, 105272d63ed6SLi Zefan trans->commit_done || (trans->in_commit && !trans->blocked)); 1053bb9c12c9SSage Weil } 1054bb9c12c9SSage Weil 1055bb9c12c9SSage Weil /* 1056bb9c12c9SSage Weil * commit transactions asynchronously. once btrfs_commit_transaction_async 1057bb9c12c9SSage Weil * returns, any subsequent transaction will not be allowed to join. 1058bb9c12c9SSage Weil */ 1059bb9c12c9SSage Weil struct btrfs_async_commit { 1060bb9c12c9SSage Weil struct btrfs_trans_handle *newtrans; 1061bb9c12c9SSage Weil struct btrfs_root *root; 1062bb9c12c9SSage Weil struct delayed_work work; 1063bb9c12c9SSage Weil }; 1064bb9c12c9SSage Weil 1065bb9c12c9SSage Weil static void do_async_commit(struct work_struct *work) 1066bb9c12c9SSage Weil { 1067bb9c12c9SSage Weil struct btrfs_async_commit *ac = 1068bb9c12c9SSage Weil container_of(work, struct btrfs_async_commit, work.work); 1069bb9c12c9SSage Weil 1070bb9c12c9SSage Weil btrfs_commit_transaction(ac->newtrans, ac->root); 1071bb9c12c9SSage Weil kfree(ac); 1072bb9c12c9SSage Weil } 1073bb9c12c9SSage Weil 1074bb9c12c9SSage Weil int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans, 1075bb9c12c9SSage Weil struct btrfs_root *root, 1076bb9c12c9SSage Weil int wait_for_unblock) 1077bb9c12c9SSage Weil { 1078bb9c12c9SSage Weil struct btrfs_async_commit *ac; 1079bb9c12c9SSage Weil struct btrfs_transaction *cur_trans; 1080bb9c12c9SSage Weil 1081bb9c12c9SSage Weil ac = kmalloc(sizeof(*ac), GFP_NOFS); 1082db5b493aSTsutomu Itoh if (!ac) 1083db5b493aSTsutomu Itoh return -ENOMEM; 1084bb9c12c9SSage Weil 1085bb9c12c9SSage Weil INIT_DELAYED_WORK(&ac->work, do_async_commit); 1086bb9c12c9SSage Weil ac->root = root; 10877a7eaa40SJosef Bacik ac->newtrans = btrfs_join_transaction(root); 10883612b495STsutomu Itoh if (IS_ERR(ac->newtrans)) { 10893612b495STsutomu Itoh int err = PTR_ERR(ac->newtrans); 10903612b495STsutomu Itoh kfree(ac); 10913612b495STsutomu Itoh return err; 10923612b495STsutomu Itoh } 1093bb9c12c9SSage Weil 1094bb9c12c9SSage Weil /* take transaction reference */ 1095bb9c12c9SSage Weil cur_trans = trans->transaction; 109613c5a93eSJosef Bacik atomic_inc(&cur_trans->use_count); 1097bb9c12c9SSage Weil 1098bb9c12c9SSage Weil btrfs_end_transaction(trans, root); 1099bb9c12c9SSage Weil schedule_delayed_work(&ac->work, 0); 1100bb9c12c9SSage Weil 1101bb9c12c9SSage Weil /* wait for transaction to start and unblock */ 1102bb9c12c9SSage Weil if (wait_for_unblock) 1103bb9c12c9SSage Weil wait_current_trans_commit_start_and_unblock(root, cur_trans); 1104bb9c12c9SSage Weil else 1105bb9c12c9SSage Weil wait_current_trans_commit_start(root, cur_trans); 1106bb9c12c9SSage Weil 110738e88054SSage Weil if (current->journal_info == trans) 110838e88054SSage Weil current->journal_info = NULL; 110938e88054SSage Weil 111038e88054SSage Weil put_transaction(cur_trans); 1111bb9c12c9SSage Weil return 0; 1112bb9c12c9SSage Weil } 1113bb9c12c9SSage Weil 1114bb9c12c9SSage Weil /* 1115bb9c12c9SSage Weil * btrfs_transaction state sequence: 1116bb9c12c9SSage Weil * in_commit = 0, blocked = 0 (initial) 1117bb9c12c9SSage Weil * in_commit = 1, blocked = 1 1118bb9c12c9SSage Weil * blocked = 0 1119bb9c12c9SSage Weil * commit_done = 1 1120bb9c12c9SSage Weil */ 112179154b1bSChris Mason int btrfs_commit_transaction(struct btrfs_trans_handle *trans, 112279154b1bSChris Mason struct btrfs_root *root) 112379154b1bSChris Mason { 112415ee9bc7SJosef Bacik unsigned long joined = 0; 112579154b1bSChris Mason struct btrfs_transaction *cur_trans; 11268fd17795SChris Mason struct btrfs_transaction *prev_trans = NULL; 112779154b1bSChris Mason DEFINE_WAIT(wait); 112815ee9bc7SJosef Bacik int ret; 112989573b9cSChris Mason int should_grow = 0; 113089573b9cSChris Mason unsigned long now = get_seconds(); 1131dccae999SSage Weil int flush_on_commit = btrfs_test_opt(root, FLUSHONCOMMIT); 113279154b1bSChris Mason 11335a3f23d5SChris Mason btrfs_run_ordered_operations(root, 0); 11345a3f23d5SChris Mason 1135b24e03dbSJosef Bacik btrfs_trans_release_metadata(trans, root); 11369c8d86dbSJosef Bacik trans->block_rsv = NULL; 11379c8d86dbSJosef Bacik 113856bec294SChris Mason /* make a pass through all the delayed refs we have so far 113956bec294SChris Mason * any runnings procs may add more while we are here 114056bec294SChris Mason */ 114156bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, 0); 114256bec294SChris Mason BUG_ON(ret); 114356bec294SChris Mason 1144b7ec40d7SChris Mason cur_trans = trans->transaction; 114556bec294SChris Mason /* 114656bec294SChris Mason * set the flushing flag so procs in this transaction have to 114756bec294SChris Mason * start sending their work down. 114856bec294SChris Mason */ 1149b7ec40d7SChris Mason cur_trans->delayed_refs.flushing = 1; 115056bec294SChris Mason 1151c3e69d58SChris Mason ret = btrfs_run_delayed_refs(trans, root, 0); 115256bec294SChris Mason BUG_ON(ret); 115356bec294SChris Mason 1154a4abeea4SJosef Bacik spin_lock(&cur_trans->commit_lock); 1155b7ec40d7SChris Mason if (cur_trans->in_commit) { 1156a4abeea4SJosef Bacik spin_unlock(&cur_trans->commit_lock); 115713c5a93eSJosef Bacik atomic_inc(&cur_trans->use_count); 115879154b1bSChris Mason btrfs_end_transaction(trans, root); 1159ccd467d6SChris Mason 1160b9c8300cSLi Zefan wait_for_commit(root, cur_trans); 116115ee9bc7SJosef Bacik 116279154b1bSChris Mason put_transaction(cur_trans); 116315ee9bc7SJosef Bacik 116479154b1bSChris Mason return 0; 116579154b1bSChris Mason } 11664313b399SChris Mason 11672c90e5d6SChris Mason trans->transaction->in_commit = 1; 1168f9295749SChris Mason trans->transaction->blocked = 1; 1169a4abeea4SJosef Bacik spin_unlock(&cur_trans->commit_lock); 1170bb9c12c9SSage Weil wake_up(&root->fs_info->transaction_blocked_wait); 1171bb9c12c9SSage Weil 1172a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 1173ccd467d6SChris Mason if (cur_trans->list.prev != &root->fs_info->trans_list) { 1174ccd467d6SChris Mason prev_trans = list_entry(cur_trans->list.prev, 1175ccd467d6SChris Mason struct btrfs_transaction, list); 1176ccd467d6SChris Mason if (!prev_trans->commit_done) { 117713c5a93eSJosef Bacik atomic_inc(&prev_trans->use_count); 1178a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1179ccd467d6SChris Mason 1180ccd467d6SChris Mason wait_for_commit(root, prev_trans); 1181ccd467d6SChris Mason 118215ee9bc7SJosef Bacik put_transaction(prev_trans); 1183a4abeea4SJosef Bacik } else { 1184a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1185ccd467d6SChris Mason } 1186a4abeea4SJosef Bacik } else { 1187a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1188ccd467d6SChris Mason } 118915ee9bc7SJosef Bacik 119089573b9cSChris Mason if (now < cur_trans->start_time || now - cur_trans->start_time < 1) 119189573b9cSChris Mason should_grow = 1; 119289573b9cSChris Mason 119315ee9bc7SJosef Bacik do { 11947ea394f1SYan Zheng int snap_pending = 0; 1195a4abeea4SJosef Bacik 119615ee9bc7SJosef Bacik joined = cur_trans->num_joined; 11977ea394f1SYan Zheng if (!list_empty(&trans->transaction->pending_snapshots)) 11987ea394f1SYan Zheng snap_pending = 1; 11997ea394f1SYan Zheng 12002c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 120115ee9bc7SJosef Bacik 12020bdb1db2SSage Weil if (flush_on_commit || snap_pending) { 120324bbcf04SYan, Zheng btrfs_start_delalloc_inodes(root, 1); 120424bbcf04SYan, Zheng ret = btrfs_wait_ordered_extents(root, 0, 1); 1205ebecd3d9SSage Weil BUG_ON(ret); 12067ea394f1SYan Zheng } 12077ea394f1SYan Zheng 120816cdcec7SMiao Xie ret = btrfs_run_delayed_items(trans, root); 120916cdcec7SMiao Xie BUG_ON(ret); 121016cdcec7SMiao Xie 12115a3f23d5SChris Mason /* 12125a3f23d5SChris Mason * rename don't use btrfs_join_transaction, so, once we 12135a3f23d5SChris Mason * set the transaction to blocked above, we aren't going 12145a3f23d5SChris Mason * to get any new ordered operations. We can safely run 12155a3f23d5SChris Mason * it here and no for sure that nothing new will be added 12165a3f23d5SChris Mason * to the list 12175a3f23d5SChris Mason */ 12185a3f23d5SChris Mason btrfs_run_ordered_operations(root, 1); 12195a3f23d5SChris Mason 1220ed3b3d31SChris Mason prepare_to_wait(&cur_trans->writer_wait, &wait, 1221ed3b3d31SChris Mason TASK_UNINTERRUPTIBLE); 1222ed3b3d31SChris Mason 122313c5a93eSJosef Bacik if (atomic_read(&cur_trans->num_writers) > 1) 122499d16cbcSSage Weil schedule_timeout(MAX_SCHEDULE_TIMEOUT); 122599d16cbcSSage Weil else if (should_grow) 122699d16cbcSSage Weil schedule_timeout(1); 122715ee9bc7SJosef Bacik 122815ee9bc7SJosef Bacik finish_wait(&cur_trans->writer_wait, &wait); 1229ed0ca140SJosef Bacik } while (atomic_read(&cur_trans->num_writers) > 1 || 1230ed0ca140SJosef Bacik (should_grow && cur_trans->num_joined != joined)); 1231ed0ca140SJosef Bacik 1232ed0ca140SJosef Bacik /* 1233ed0ca140SJosef Bacik * Ok now we need to make sure to block out any other joins while we 1234ed0ca140SJosef Bacik * commit the transaction. We could have started a join before setting 1235ed0ca140SJosef Bacik * no_join so make sure to wait for num_writers to == 1 again. 1236ed0ca140SJosef Bacik */ 1237a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 1238a4abeea4SJosef Bacik root->fs_info->trans_no_join = 1; 1239a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1240ed0ca140SJosef Bacik wait_event(cur_trans->writer_wait, 1241ed0ca140SJosef Bacik atomic_read(&cur_trans->num_writers) == 1); 124215ee9bc7SJosef Bacik 12437585717fSChris Mason /* 12447585717fSChris Mason * the reloc mutex makes sure that we stop 12457585717fSChris Mason * the balancing code from coming in and moving 12467585717fSChris Mason * extents around in the middle of the commit 12477585717fSChris Mason */ 12487585717fSChris Mason mutex_lock(&root->fs_info->reloc_mutex); 12497585717fSChris Mason 1250e999376fSChris Mason ret = btrfs_run_delayed_items(trans, root); 12513063d29fSChris Mason BUG_ON(ret); 12523063d29fSChris Mason 1253e999376fSChris Mason ret = create_pending_snapshots(trans, root->fs_info); 125416cdcec7SMiao Xie BUG_ON(ret); 125516cdcec7SMiao Xie 125656bec294SChris Mason ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1); 125756bec294SChris Mason BUG_ON(ret); 125856bec294SChris Mason 1259e999376fSChris Mason /* 1260e999376fSChris Mason * make sure none of the code above managed to slip in a 1261e999376fSChris Mason * delayed item 1262e999376fSChris Mason */ 1263e999376fSChris Mason btrfs_assert_delayed_root_empty(root); 1264e999376fSChris Mason 12652c90e5d6SChris Mason WARN_ON(cur_trans != trans->transaction); 1266dc17ff8fSChris Mason 1267a2de733cSArne Jansen btrfs_scrub_pause(root); 1268e02119d5SChris Mason /* btrfs_commit_tree_roots is responsible for getting the 1269e02119d5SChris Mason * various roots consistent with each other. Every pointer 1270e02119d5SChris Mason * in the tree of tree roots has to point to the most up to date 1271e02119d5SChris Mason * root for every subvolume and other tree. So, we have to keep 1272e02119d5SChris Mason * the tree logging code from jumping in and changing any 1273e02119d5SChris Mason * of the trees. 1274e02119d5SChris Mason * 1275e02119d5SChris Mason * At this point in the commit, there can't be any tree-log 1276e02119d5SChris Mason * writers, but a little lower down we drop the trans mutex 1277e02119d5SChris Mason * and let new people in. By holding the tree_log_mutex 1278e02119d5SChris Mason * from now until after the super is written, we avoid races 1279e02119d5SChris Mason * with the tree-log code. 1280e02119d5SChris Mason */ 1281e02119d5SChris Mason mutex_lock(&root->fs_info->tree_log_mutex); 12821a40e23bSZheng Yan 12835d4f98a2SYan Zheng ret = commit_fs_roots(trans, root); 128454aa1f4dSChris Mason BUG_ON(ret); 128554aa1f4dSChris Mason 12865d4f98a2SYan Zheng /* commit_fs_roots gets rid of all the tree log roots, it is now 1287e02119d5SChris Mason * safe to free the root of tree log roots 1288e02119d5SChris Mason */ 1289e02119d5SChris Mason btrfs_free_log_root_tree(trans, root->fs_info); 1290e02119d5SChris Mason 12915d4f98a2SYan Zheng ret = commit_cowonly_roots(trans, root); 129279154b1bSChris Mason BUG_ON(ret); 129354aa1f4dSChris Mason 129411833d66SYan Zheng btrfs_prepare_extent_commit(trans, root); 129511833d66SYan Zheng 129678fae27eSChris Mason cur_trans = root->fs_info->running_transaction; 12975f39d397SChris Mason 12985d4f98a2SYan Zheng btrfs_set_root_node(&root->fs_info->tree_root->root_item, 12995d4f98a2SYan Zheng root->fs_info->tree_root->node); 1300817d52f8SJosef Bacik switch_commit_root(root->fs_info->tree_root); 13015d4f98a2SYan Zheng 13025d4f98a2SYan Zheng btrfs_set_root_node(&root->fs_info->chunk_root->root_item, 13035d4f98a2SYan Zheng root->fs_info->chunk_root->node); 1304817d52f8SJosef Bacik switch_commit_root(root->fs_info->chunk_root); 13055d4f98a2SYan Zheng 13065d4f98a2SYan Zheng update_super_roots(root); 1307e02119d5SChris Mason 1308e02119d5SChris Mason if (!root->fs_info->log_root_recovering) { 13096c41761fSDavid Sterba btrfs_set_super_log_root(root->fs_info->super_copy, 0); 13106c41761fSDavid Sterba btrfs_set_super_log_root_level(root->fs_info->super_copy, 0); 1311e02119d5SChris Mason } 1312e02119d5SChris Mason 13136c41761fSDavid Sterba memcpy(root->fs_info->super_for_commit, root->fs_info->super_copy, 13146c41761fSDavid Sterba sizeof(*root->fs_info->super_copy)); 1315ccd467d6SChris Mason 1316f9295749SChris Mason trans->transaction->blocked = 0; 1317a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 1318a4abeea4SJosef Bacik root->fs_info->running_transaction = NULL; 1319a4abeea4SJosef Bacik root->fs_info->trans_no_join = 0; 1320a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 13217585717fSChris Mason mutex_unlock(&root->fs_info->reloc_mutex); 1322b7ec40d7SChris Mason 1323f9295749SChris Mason wake_up(&root->fs_info->transaction_wait); 1324e6dcd2dcSChris Mason 132579154b1bSChris Mason ret = btrfs_write_and_wait_transaction(trans, root); 132679154b1bSChris Mason BUG_ON(ret); 1327a512bbf8SYan Zheng write_ctree_super(trans, root, 0); 13284313b399SChris Mason 1329e02119d5SChris Mason /* 1330e02119d5SChris Mason * the super is written, we can safely allow the tree-loggers 1331e02119d5SChris Mason * to go about their business 1332e02119d5SChris Mason */ 1333e02119d5SChris Mason mutex_unlock(&root->fs_info->tree_log_mutex); 1334e02119d5SChris Mason 133511833d66SYan Zheng btrfs_finish_extent_commit(trans, root); 13364313b399SChris Mason 13372c90e5d6SChris Mason cur_trans->commit_done = 1; 1338b7ec40d7SChris Mason 133915ee9bc7SJosef Bacik root->fs_info->last_trans_committed = cur_trans->transid; 1340817d52f8SJosef Bacik 13412c90e5d6SChris Mason wake_up(&cur_trans->commit_wait); 13423de4586cSChris Mason 1343a4abeea4SJosef Bacik spin_lock(&root->fs_info->trans_lock); 134413c5a93eSJosef Bacik list_del_init(&cur_trans->list); 1345a4abeea4SJosef Bacik spin_unlock(&root->fs_info->trans_lock); 1346a4abeea4SJosef Bacik 134779154b1bSChris Mason put_transaction(cur_trans); 134878fae27eSChris Mason put_transaction(cur_trans); 134958176a96SJosef Bacik 13501abe9b8aSliubo trace_btrfs_transaction_commit(root); 13511abe9b8aSliubo 1352a2de733cSArne Jansen btrfs_scrub_continue(root); 1353a2de733cSArne Jansen 13549ed74f2dSJosef Bacik if (current->journal_info == trans) 13559ed74f2dSJosef Bacik current->journal_info = NULL; 13569ed74f2dSJosef Bacik 13572c90e5d6SChris Mason kmem_cache_free(btrfs_trans_handle_cachep, trans); 135824bbcf04SYan, Zheng 135924bbcf04SYan, Zheng if (current != root->fs_info->transaction_kthread) 136024bbcf04SYan, Zheng btrfs_run_delayed_iputs(root); 136124bbcf04SYan, Zheng 136279154b1bSChris Mason return ret; 136379154b1bSChris Mason } 136479154b1bSChris Mason 1365d352ac68SChris Mason /* 1366d352ac68SChris Mason * interface function to delete all the snapshots we have scheduled for deletion 1367d352ac68SChris Mason */ 1368e9d0b13bSChris Mason int btrfs_clean_old_snapshots(struct btrfs_root *root) 1369e9d0b13bSChris Mason { 13705d4f98a2SYan Zheng LIST_HEAD(list); 13715d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = root->fs_info; 1372e9d0b13bSChris Mason 1373a4abeea4SJosef Bacik spin_lock(&fs_info->trans_lock); 13745d4f98a2SYan Zheng list_splice_init(&fs_info->dead_roots, &list); 1375a4abeea4SJosef Bacik spin_unlock(&fs_info->trans_lock); 13765d4f98a2SYan Zheng 13775d4f98a2SYan Zheng while (!list_empty(&list)) { 13785d4f98a2SYan Zheng root = list_entry(list.next, struct btrfs_root, root_list); 137976dda93cSYan, Zheng list_del(&root->root_list); 138076dda93cSYan, Zheng 138116cdcec7SMiao Xie btrfs_kill_all_delayed_nodes(root); 138216cdcec7SMiao Xie 138376dda93cSYan, Zheng if (btrfs_header_backref_rev(root->node) < 138476dda93cSYan, Zheng BTRFS_MIXED_BACKREF_REV) 13853fd0a558SYan, Zheng btrfs_drop_snapshot(root, NULL, 0); 138676dda93cSYan, Zheng else 13873fd0a558SYan, Zheng btrfs_drop_snapshot(root, NULL, 1); 1388e9d0b13bSChris Mason } 1389e9d0b13bSChris Mason return 0; 1390e9d0b13bSChris Mason } 1391