| 1 | |
| 2 | |
| 3 | * linux/fs/ext4/fsync.c
|
| 4 | *
|
| 5 | * Copyright (C) 1993 Stephen Tweedie ([email protected])
|
| 6 | * from
|
| 7 | * Copyright (C) 1992 Remy Card ([email protected])
|
| 8 | * Laboratoire MASI - Institut Blaise Pascal
|
| 9 | * Universite Pierre et Marie Curie (Paris VI)
|
| 10 | * from
|
| 11 | * linux/fs/minix/truncate.c Copyright (C) 1991, 1992 Linus Torvalds
|
| 12 | *
|
| 13 | * ext4fs fsync primitive
|
| 14 | *
|
| 15 | * Big-endian to little-endian byte-swapping/bitmaps by
|
| 16 | * David S. Miller ([email protected]), 1995
|
| 17 | *
|
| 18 | * Removed unnecessary code duplication for little endian machines
|
| 19 | * and excessive __inline__s.
|
| 20 | * Andi Kleen, 1997
|
| 21 | *
|
| 22 | * Major simplications and cleanup - we only need to do the metadata, because
|
| 23 | * we can depend on generic_block_fdatasync() to sync the data blocks.
|
| 24 |
|
| 25 |
|
| 26 | #include <linux/time.h>
|
| 27 | #include <linux/fs.h>
|
| 28 | #include <linux/sched.h>
|
| 29 | #include <linux/writeback.h>
|
| 30 | #include <linux/blkdev.h>
|
| 31 | #include <linux/buffer_head.h>
|
| 32 |
|
| 33 | #include "ext4.h"
|
| 34 | #include "ext4_jbd2.h"
|
| 35 |
|
| 36 | #include <trace/events/ext4.h>
|
| 37 |
|
| 38 | |
| 39 | * If we're not journaling and this is a just-created file, we have to
|
| 40 | * sync our parent directory (if it was freshly created) since
|
| 41 | * otherwise it will only be written by writeback, leaving a huge
|
| 42 | * window during which a crash may lose the file. This may apply for
|
| 43 | * the parent directory's parent as well, and so on recursively, if
|
| 44 | * they are also freshly created.
|
| 45 |
|
| 46 | static int ext4_sync_parent(struct inode *inode)
|
| 47 | {
|
| 48 | struct dentry *dentry, *next;
|
| 49 | int ret = 0;
|
| 50 |
|
| 51 | if (!ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY))
|
| 52 | return 0;
|
| 53 | dentry = d_find_any_alias(inode);
|
| 54 | if (!dentry)
|
| 55 | return 0;
|
| 56 | while (ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
|
| 57 | ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
|
| 58 |
|
| 59 | next = dget_parent(dentry);
|
| 60 | dput(dentry);
|
| 61 | dentry = next;
|
| 62 | inode = dentry->d_inode;
|
| 63 |
|
| 64 | |
| 65 | * The directory inode may have gone through rmdir by now. But
|
| 66 | * the inode itself and its blocks are still allocated (we hold
|
| 67 | * a reference to the inode via its dentry), so it didn't go
|
| 68 | * through ext4_evict_inode()) and so we are safe to flush
|
| 69 | * metadata blocks and the inode.
|
| 70 |
|
| 71 | ret = sync_inode_metadata(inode, 1);
|
| 72 | if (ret)
|
| 73 | break;
|
| 74 | }
|
| 75 | dput(dentry);
|
| 76 | return ret;
|
| 77 | }
|
| 78 |
|
| 79 | static int ext4_fsync_nojournal(struct file *file, loff_t start, loff_t end,
|
| 80 | int datasync, bool *needs_barrier)
|
| 81 | {
|
| 82 | struct inode *inode = file->f_inode;
|
| 83 | int ret;
|
| 84 |
|
| 85 | ret = sync_inode_metadata(inode, 1);
|
| 86 | if (ret)
|
| 87 | return ret;
|
| 88 | ret = ext4_sync_parent(inode);
|
| 89 |
|
| 90 | if (test_opt(inode->i_sb, BARRIER))
|
| 91 | *needs_barrier = true;
|
| 92 |
|
| 93 | return ret;
|
| 94 | }
|
| 95 |
|
| 96 | static int ext4_fsync_journal(struct inode *inode, bool datasync,
|
| 97 | bool *needs_barrier)
|
| 98 | {
|
| 99 | struct ext4_inode_info *ei = EXT4_I(inode);
|
| 100 | journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
|
| 101 | tid_t commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid;
|
| 102 |
|
| 103 | |
| 104 | * Fastcommit does not really support fsync on directories or other
|
| 105 | * special files. Force a full commit.
|
| 106 |
|
| 107 | if (!S_ISREG(inode->i_mode))
|
| 108 | return ext4_force_commit(inode->i_sb);
|
| 109 |
|
| 110 | if (journal->j_flags & JBD2_BARRIER &&
|
| 111 | !jbd2_trans_will_send_data_barrier(journal, commit_tid))
|
| 112 | *needs_barrier = true;
|
| 113 |
|
| 114 | return ext4_fc_commit(journal, commit_tid);
|
| 115 | }
|
| 116 |
|
| 117 | |
| 118 | * akpm: A new design for ext4_sync_file().
|
| 119 | *
|
| 120 | * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
|
| 121 | * There cannot be a transaction open by this task.
|
| 122 | * Another task could have dirtied this inode. Its data can be in any
|
| 123 | * state in the journalling system.
|
| 124 | *
|
| 125 | * What we do is just kick off a commit and wait on it. This will snapshot the
|
| 126 | * inode to disk.
|
| 127 |
|
| 128 | int ext4_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
|
| 129 | {
|
| 130 | int ret = 0, err;
|
| 131 | bool needs_barrier = false;
|
| 132 | struct inode *inode = file->f_mapping->host;
|
| 133 |
|
| 134 | ret = ext4_emergency_state(inode->i_sb);
|
| 135 | if (unlikely(ret))
|
| 136 | return ret;
|
| 137 |
|
| 138 | ASSERT(ext4_journal_current_handle() == NULL);
|
| 139 |
|
| 140 | trace_ext4_sync_file_enter(file, datasync);
|
| 141 |
|
| 142 | if (sb_rdonly(inode->i_sb))
|
| 143 | goto out;
|
| 144 |
|
| 145 | ret = file_write_and_wait_range(file, start, end);
|
| 146 | if (ret)
|
| 147 | goto out;
|
| 148 |
|
| 149 | if (!EXT4_SB(inode->i_sb)->s_journal) {
|
| 150 | ret = ext4_fsync_nojournal(file, start, end, datasync,
|
| 151 | &needs_barrier);
|
| 152 | if (needs_barrier)
|
| 153 | goto issue_flush;
|
| 154 | goto out;
|
| 155 | }
|
| 156 |
|
| 157 | |
| 158 | * The caller's filemap_fdatawrite()/wait will sync the data.
|
| 159 | * Metadata is in the journal, we wait for proper transaction to
|
| 160 | * commit here.
|
| 161 |
|
| 162 | ret = ext4_fsync_journal(inode, datasync, &needs_barrier);
|
| 163 |
|
| 164 | issue_flush:
|
| 165 | if (needs_barrier) {
|
| 166 | err = blkdev_issue_flush(inode->i_sb->s_bdev);
|
| 167 | if (!ret)
|
| 168 | ret = err;
|
| 169 | }
|
| 170 | out:
|
| 171 | err = file_check_and_advance_wb_err(file);
|
| 172 | if (ret == 0)
|
| 173 | ret = err;
|
| 174 | trace_ext4_sync_file_exit(inode, ret);
|
| 175 | return ret;
|
| 176 | }
|