logfs: fix "Kernel BUG at readwrite.c:1193"
authorPrasad Joshi <prasadjoshi124@gmail.com>
Wed, 22 Dec 2010 01:24:20 +0000 (17:24 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 23 Dec 2010 03:43:33 +0000 (19:43 -0800)
This happens when __logfs_create() tries to write a new inode to the disk
which is full.

__logfs_create() associates the transaction pointer with inode.  During
the logfs_write_inode() function call chain this transaction pointer is
moved from inode to page->private using function move_inode_to_page
(do_write_inode() -> inode_to_page() -> move_inode_to_page)

When the write inode fails, the transaction is aborted and iput is called
on the failed inode.  During delete_inode the same transaction pointer
associated with the page is getting used.  Thus causing kernel BUG.

The patch checks for error in write_inode() and restores the page->private
to NULL.

Addresses https://bugzilla.kernel.org/show_bug.cgi?id=20162

Signed-off-by: Prasad Joshi <prasadjoshi124@gmail.com>
Cc: Joern Engel <joern@logfs.org>
Cc: Florian Mickler <florian@mickler.org>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Maciej Rutecki <maciej.rutecki@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
fs/logfs/readwrite.c

index 6127baf..ee99a9f 100644 (file)
@@ -1994,6 +1994,9 @@ static int do_write_inode(struct inode *inode)
 
        /* FIXME: transaction is part of logfs_block now.  Is that enough? */
        err = logfs_write_buf(master_inode, page, 0);
+       if (err)
+               move_page_to_inode(inode, page);
+
        logfs_put_write_page(page);
        return err;
 }