We cannot rely on buffer dirty bits during fsync because pdflush can come
before fsync is called and clear dirty bits without forcing a transaction
commit. What we do is that we track which transaction has last changed
the inode and which transaction last changed allocation and force it to
disk on fsync.
CC: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
Signed-off-by: Jan Kara <jack@xxxxxxx>
---
fs/ext3/fsync.c | 36 ++++++++++++++++--------------------
fs/ext3/inode.c | 35 ++++++++++++++++++++++++++++++++++-
include/linux/ext3_fs_i.h | 8 ++++++++
3 files changed, 58 insertions(+), 21 deletions(-)
diff --git a/fs/ext3/fsync.c b/fs/ext3/fsync.c
index 451d166..8209f26 100644
--- a/fs/ext3/fsync.c
+++ b/fs/ext3/fsync.c
@@ -46,19 +46,21 @@
int ext3_sync_file(struct file * file, struct dentry *dentry, int datasync)
{
struct inode *inode = dentry->d_inode;
+ struct ext3_inode_info *ei = EXT3_I(inode);
+ journal_t *journal = EXT3_SB(inode->i_sb)->s_journal;
int ret = 0;
+ tid_t commit_tid;
+
+ if (inode->i_sb->s_flags & MS_RDONLY)
+ return 0;
J_ASSERT(ext3_journal_current_handle() == NULL);
/*
- * data=writeback:
+ * data=writeback,ordered:
* The caller's filemap_fdatawrite()/wait will sync the data.
- * sync_inode() will sync the metadata
- *
- * data=ordered:
- * The caller's filemap_fdatawrite() will write the data and
- * sync_inode() will write the inode if it is dirty. Then the caller's
- * filemap_fdatawait() will wait on the pages.
+ * Metadata is in the journal, we wait for a proper transaction
+ * to commit here.
*
* data=journal:
* filemap_fdatawrite won't do anything (the buffers are clean).
@@ -73,22 +75,16 @@ int ext3_sync_file(struct file * file, struct dentry *dentry, int datasync)
goto out;
}
- if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
- goto flush;
+ if (datasync)
+ commit_tid = atomic_read(&ei->i_datasync_tid);
+ else
+ commit_tid = atomic_read(&ei->i_sync_tid);
- /*
- * The VFS has written the file data. If the inode is unaltered
- * then we need not start a commit.
- */
- if (inode->i_state & (I_DIRTY_SYNC|I_DIRTY_DATASYNC)) {
- struct writeback_control wbc = {
- .sync_mode = WB_SYNC_ALL,
- .nr_to_write = 0, /* sys_fsync did this */
- };
- ret = sync_inode(inode, &wbc);
+ if (log_start_commit(journal, commit_tid)) {
+ log_wait_commit(journal, commit_tid);
goto out;
}
-flush:
+
/*
* In case we didn't commit a transaction, we have to flush
* disk caches manually so that data really is on persistent
diff --git a/fs/ext3/inode.c b/fs/ext3/inode.c
index acf1b14..0a956a8 100644
--- a/fs/ext3/inode.c
+++ b/fs/ext3/inode.c
@@ -699,8 +699,9 @@ static int ext3_splice_branch(handle_t *handle, struct inode *inode,
int err = 0;
struct ext3_block_alloc_info *block_i;
ext3_fsblk_t current_block;
+ struct ext3_inode_info *ei = EXT3_I(inode);
- block_i = EXT3_I(inode)->i_block_alloc_info;
+ block_i = ei->i_block_alloc_info;
/*
* If we're splicing into a [td]indirect block (as opposed to the
* inode) then we need to get write access to the [td]indirect block
@@ -741,6 +742,8 @@ static int ext3_splice_branch(handle_t *handle, struct inode *inode,
inode->i_ctime = CURRENT_TIME_SEC;
ext3_mark_inode_dirty(handle, inode);
+ /* ext3_mark_inode_dirty already updated i_sync_tid */
+ atomic_set(&ei->i_datasync_tid, handle->h_transaction->t_tid);
/* had we spliced it onto indirect block? */
if (where->bh) {
@@ -2750,6 +2753,8 @@ struct inode *ext3_iget(struct super_block *sb, unsigned long ino)
struct ext3_inode_info *ei;
struct buffer_head *bh;
struct inode *inode;
+ journal_t *journal = EXT3_SB(sb)->s_journal;
+ transaction_t *transaction;
long ret;
int block;
@@ -2827,6 +2832,33 @@ struct inode *ext3_iget(struct super_block *sb, unsigned long ino)
ei->i_data[block] = raw_inode->i_block[block];
INIT_LIST_HEAD(&ei->i_orphan);
+ /*
+ * Set transaction id's of transactions that have to be committed
+ * to finish f[data]sync. We set them to currently running transaction
+ * as we cannot be sure that the inode or some of its metadata isn't
+ * part of the transaction - the inode could have been reclaimed and
+ * now it is reread from disk.
+ */
+ if (journal) {
+ tid_t tid;
+
+ spin_lock(&journal->j_state_lock);
+ if (journal->j_running_transaction)
+ transaction = journal->j_running_transaction;
+ else
+ transaction = journal->j_committing_transaction;
+ if (transaction)
+ tid = transaction->t_tid;
+ else
+ tid = journal->j_commit_sequence;
+ spin_unlock(&journal->j_state_lock);
+ atomic_set(&ei->i_sync_tid, tid);
+ atomic_set(&ei->i_datasync_tid, tid);
+ } else {
+ atomic_set(&ei->i_sync_tid, 0);
+ atomic_set(&ei->i_datasync_tid, 0);
+ }
+
if (inode->i_ino >= EXT3_FIRST_INO(inode->i_sb) + 1 &&
EXT3_INODE_SIZE(inode->i_sb) > EXT3_GOOD_OLD_INODE_SIZE) {
/*
@@ -3011,6 +3043,7 @@ again:
err = rc;
ei->i_state &= ~EXT3_STATE_NEW;
+ atomic_set(&ei->i_sync_tid, handle->h_transaction->t_tid);
out_brelse:
brelse (bh);
ext3_std_error(inode->i_sb, err);
diff --git a/include/linux/ext3_fs_i.h b/include/linux/ext3_fs_i.h
index ca1bfe9..93e7428 100644
--- a/include/linux/ext3_fs_i.h
+++ b/include/linux/ext3_fs_i.h
@@ -137,6 +137,14 @@ struct ext3_inode_info {
* by other means, so we have truncate_mutex.
*/
struct mutex truncate_mutex;
+
+ /*
+ * Transactions that contain inode's metadata needed to complete
+ * fsync and fdatasync, respectively.
+ */
+ atomic_t i_sync_tid;
+ atomic_t i_datasync_tid;
+
struct inode vfs_inode;
};
--
1.6.0.2
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
