1ac27a0ecSDave Kleikamp /* 2617ba13bSMingming Cao * linux/fs/ext4/fsync.c 3ac27a0ecSDave Kleikamp * 4ac27a0ecSDave Kleikamp * Copyright (C) 1993 Stephen Tweedie (sct@redhat.com) 5ac27a0ecSDave Kleikamp * from 6ac27a0ecSDave Kleikamp * Copyright (C) 1992 Remy Card (card@masi.ibp.fr) 7ac27a0ecSDave Kleikamp * Laboratoire MASI - Institut Blaise Pascal 8ac27a0ecSDave Kleikamp * Universite Pierre et Marie Curie (Paris VI) 9ac27a0ecSDave Kleikamp * from 10ac27a0ecSDave Kleikamp * linux/fs/minix/truncate.c Copyright (C) 1991, 1992 Linus Torvalds 11ac27a0ecSDave Kleikamp * 12617ba13bSMingming Cao * ext4fs fsync primitive 13ac27a0ecSDave Kleikamp * 14ac27a0ecSDave Kleikamp * Big-endian to little-endian byte-swapping/bitmaps by 15ac27a0ecSDave Kleikamp * David S. Miller (davem@caip.rutgers.edu), 1995 16ac27a0ecSDave Kleikamp * 17ac27a0ecSDave Kleikamp * Removed unnecessary code duplication for little endian machines 18ac27a0ecSDave Kleikamp * and excessive __inline__s. 19ac27a0ecSDave Kleikamp * Andi Kleen, 1997 20ac27a0ecSDave Kleikamp * 21ac27a0ecSDave Kleikamp * Major simplications and cleanup - we only need to do the metadata, because 22ac27a0ecSDave Kleikamp * we can depend on generic_block_fdatasync() to sync the data blocks. 23ac27a0ecSDave Kleikamp */ 24ac27a0ecSDave Kleikamp 25ac27a0ecSDave Kleikamp #include <linux/time.h> 26ac27a0ecSDave Kleikamp #include <linux/fs.h> 27ac27a0ecSDave Kleikamp #include <linux/sched.h> 28ac27a0ecSDave Kleikamp #include <linux/writeback.h> 29dab291afSMingming Cao #include <linux/jbd2.h> 30d755fb38SEric Sandeen #include <linux/blkdev.h> 319bffad1eSTheodore Ts'o 323dcf5451SChristoph Hellwig #include "ext4.h" 333dcf5451SChristoph Hellwig #include "ext4_jbd2.h" 34ac27a0ecSDave Kleikamp 359bffad1eSTheodore Ts'o #include <trace/events/ext4.h> 369bffad1eSTheodore Ts'o 37ac27a0ecSDave Kleikamp /* 3814ece102SFrank Mayhar * If we're not journaling and this is a just-created file, we have to 3914ece102SFrank Mayhar * sync our parent directory (if it was freshly created) since 4014ece102SFrank Mayhar * otherwise it will only be written by writeback, leaving a huge 4114ece102SFrank Mayhar * window during which a crash may lose the file. This may apply for 4214ece102SFrank Mayhar * the parent directory's parent as well, and so on recursively, if 4314ece102SFrank Mayhar * they are also freshly created. 4414ece102SFrank Mayhar */ 450893ed45SCurt Wohlgemuth static int ext4_sync_parent(struct inode *inode) 4614ece102SFrank Mayhar { 4714ece102SFrank Mayhar struct dentry *dentry = NULL; 48d59729f4STheodore Ts'o struct inode *next; 490893ed45SCurt Wohlgemuth int ret = 0; 5014ece102SFrank Mayhar 51d59729f4STheodore Ts'o if (!ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) 52d59729f4STheodore Ts'o return 0; 53d59729f4STheodore Ts'o inode = igrab(inode); 54d59729f4STheodore Ts'o while (ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) { 5514ece102SFrank Mayhar ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY); 569f713878SAl Viro dentry = d_find_any_alias(inode); 57d59729f4STheodore Ts'o if (!dentry) 5814ece102SFrank Mayhar break; 59d59729f4STheodore Ts'o next = igrab(dentry->d_parent->d_inode); 60d59729f4STheodore Ts'o dput(dentry); 61d59729f4STheodore Ts'o if (!next) 62d59729f4STheodore Ts'o break; 63d59729f4STheodore Ts'o iput(inode); 64d59729f4STheodore Ts'o inode = next; 650893ed45SCurt Wohlgemuth ret = sync_mapping_buffers(inode->i_mapping); 660893ed45SCurt Wohlgemuth if (ret) 670893ed45SCurt Wohlgemuth break; 6864744e03SGuo Chao ret = sync_inode_metadata(inode, 1); 690893ed45SCurt Wohlgemuth if (ret) 700893ed45SCurt Wohlgemuth break; 7114ece102SFrank Mayhar } 72d59729f4STheodore Ts'o iput(inode); 730893ed45SCurt Wohlgemuth return ret; 7414ece102SFrank Mayhar } 7514ece102SFrank Mayhar 7614ece102SFrank Mayhar /* 77617ba13bSMingming Cao * akpm: A new design for ext4_sync_file(). 78ac27a0ecSDave Kleikamp * 79ac27a0ecSDave Kleikamp * This is only called from sys_fsync(), sys_fdatasync() and sys_msync(). 80ac27a0ecSDave Kleikamp * There cannot be a transaction open by this task. 81ac27a0ecSDave Kleikamp * Another task could have dirtied this inode. Its data can be in any 82ac27a0ecSDave Kleikamp * state in the journalling system. 83ac27a0ecSDave Kleikamp * 84ac27a0ecSDave Kleikamp * What we do is just kick off a commit and wait on it. This will snapshot the 85ac27a0ecSDave Kleikamp * inode to disk. 86ac27a0ecSDave Kleikamp */ 87ac27a0ecSDave Kleikamp 8802c24a82SJosef Bacik int ext4_sync_file(struct file *file, loff_t start, loff_t end, int datasync) 89ac27a0ecSDave Kleikamp { 907ea80859SChristoph Hellwig struct inode *inode = file->f_mapping->host; 91b436b9beSJan Kara struct ext4_inode_info *ei = EXT4_I(inode); 92d755fb38SEric Sandeen journal_t *journal = EXT4_SB(inode->i_sb)->s_journal; 9337b10dd0SJan Kara int ret = 0, err; 94b436b9beSJan Kara tid_t commit_tid; 9593628ffbSJan Kara bool needs_barrier = false; 96ac27a0ecSDave Kleikamp 97ac39849dSAneesh Kumar K.V J_ASSERT(ext4_journal_current_handle() == NULL); 98ac27a0ecSDave Kleikamp 990562e0baSJiaying Zhang trace_ext4_sync_file_enter(file, datasync); 100ede86cc4STheodore Ts'o 10137b10dd0SJan Kara if (inode->i_sb->s_flags & MS_RDONLY) 10292e6222dSJan Kara goto out; 10337b10dd0SJan Kara 10437b10dd0SJan Kara if (!journal) { 10537b10dd0SJan Kara ret = generic_file_fsync(file, start, end, datasync); 10637b10dd0SJan Kara if (!ret && !hlist_empty(&inode->i_dentry)) 10737b10dd0SJan Kara ret = ext4_sync_parent(inode); 10892e6222dSJan Kara goto out; 10937b10dd0SJan Kara } 11037b10dd0SJan Kara 11102c24a82SJosef Bacik ret = filemap_write_and_wait_range(inode->i_mapping, start, end); 11202c24a82SJosef Bacik if (ret) 11302c24a82SJosef Bacik return ret; 114ac27a0ecSDave Kleikamp /* 115b436b9beSJan Kara * data=writeback,ordered: 116ac27a0ecSDave Kleikamp * The caller's filemap_fdatawrite()/wait will sync the data. 117b436b9beSJan Kara * Metadata is in the journal, we wait for proper transaction to 118b436b9beSJan Kara * commit here. 119ac27a0ecSDave Kleikamp * 120ac27a0ecSDave Kleikamp * data=journal: 121ac27a0ecSDave Kleikamp * filemap_fdatawrite won't do anything (the buffers are clean). 122617ba13bSMingming Cao * ext4_force_commit will write the file data into the journal and 123ac27a0ecSDave Kleikamp * will wait on that. 124ac27a0ecSDave Kleikamp * filemap_fdatawait() will encounter a ton of newly-dirtied pages 125ac27a0ecSDave Kleikamp * (they were dirtied by commit). But that's OK - the blocks are 126ac27a0ecSDave Kleikamp * safe in-journal, which is all fsync() needs to ensure. 127ac27a0ecSDave Kleikamp */ 1280562e0baSJiaying Zhang if (ext4_should_journal_data(inode)) { 1290562e0baSJiaying Zhang ret = ext4_force_commit(inode->i_sb); 1300562e0baSJiaying Zhang goto out; 1310562e0baSJiaying Zhang } 132ac27a0ecSDave Kleikamp 133b436b9beSJan Kara commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid; 13493628ffbSJan Kara if (journal->j_flags & JBD2_BARRIER && 13593628ffbSJan Kara !jbd2_trans_will_send_data_barrier(journal, commit_tid)) 13693628ffbSJan Kara needs_barrier = true; 137d76a3a77STheodore Ts'o ret = jbd2_complete_transaction(journal, commit_tid); 138a4a39040STheodore Ts'o if (needs_barrier) { 139a4a39040STheodore Ts'o err = blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL); 140a4a39040STheodore Ts'o if (!ret) 141a4a39040STheodore Ts'o ret = err; 142a4a39040STheodore Ts'o } 1430562e0baSJiaying Zhang out: 1440562e0baSJiaying Zhang trace_ext4_sync_file_exit(inode, ret); 145ac27a0ecSDave Kleikamp return ret; 146ac27a0ecSDave Kleikamp } 147