Commit 2b5adb1d authored by Dave Kleikamp's avatar Dave Kleikamp Committed by Linus Torvalds

[PATCH] Prevent deadlock in JFS when flushing data during commit

This prevents a deadlock in JFS.
parent 2e48a176
...@@ -1155,19 +1155,30 @@ int txCommit(tid_t tid, /* transaction identifier */ ...@@ -1155,19 +1155,30 @@ int txCommit(tid_t tid, /* transaction identifier */
jfs_ip = JFS_IP(ip); jfs_ip = JFS_IP(ip);
/* /*
* BUGBUG - Should we call filemap_fdatawrite here instead * BUGBUG - This code has temporarily been removed. The
* of fsync_inode_data? * intent is to ensure that any file data is written before
* If we do, we have a deadlock condition since we may end * the metadata is committed to the journal. This prevents
* up recursively calling jfs_get_block with the IWRITELOCK * uninitialized data from appearing in a file after the
* held. We may be able to do away with IWRITELOCK while * journal has been replayed. (The uninitialized data
* committing transactions and use i_sem instead. * could be sensitive data removed by another user.)
*/ *
if ((!S_ISDIR(ip->i_mode)) * The problem now is that we are holding the IWRITELOCK
&& (tblk->flag & COMMIT_DELETE) == 0) { * on the inode, and calling filemap_fdatawrite on an
filemap_fdatawait(ip->i_mapping); * unmapped page will cause a deadlock in jfs_get_block.
filemap_fdatawrite(ip->i_mapping); *
filemap_fdatawait(ip->i_mapping); * The long term solution is to pare down the use of
} * IWRITELOCK. We are currently holding it too long.
* We could also be smarter about which data pages need
* to be written before the transaction is committed and
* when we don't need to worry about it at all.
*
* if ((!S_ISDIR(ip->i_mode))
* && (tblk->flag & COMMIT_DELETE) == 0) {
* filemap_fdatawait(ip->i_mapping);
* filemap_fdatawrite(ip->i_mapping);
* filemap_fdatawait(ip->i_mapping);
* }
*/
/* /*
* Mark inode as not dirty. It will still be on the dirty * Mark inode as not dirty. It will still be on the dirty
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment