Re: BUG: KASAN: use-after-free in xfs_iflush_cluster+0x9d7/0xaf0

[Dave Chinner <david@fromorbit.com>]

On Mon, Dec 14, 2015 at 09:15:26PM +0100, Andrea Gelmini wrote:
> On Tue, Dec 15, 2015 at 06:54:22AM +1100, Dave Chinner wrote:
> > What line of code does this address correspond to in your kernel?
> > 
> > 	xfs_iflush_cluster+0x9d7
> 
> gelma@glen:~/dev/kernel/v4.4.x$ git grep -Iin xfs_iflush_cluster
> fs/xfs/xfs_inode.c:3179:xfs_iflush_cluster(
> fs/xfs/xfs_inode.c:3414:	error = xfs_iflush_cluster(ip, bp);

If that was what I needed, I wouldn't have needed to ask. :/

I need the translation of the memory address to line number, not the
line number of function call. This requires translation from your
built kernel object file. e.g. on a kernel I just built:

$ gdb vmlinux
....
(gdb) l *(xfs_iflush_cluster+0x9d7)
0xffffffff814df647 is in xfs_bulkstat_one_int (fs/xfs/xfs_itable.c:110).
105             buf->bs_dmevmask = dic->di_dmevmask;
106             buf->bs_dmstate = dic->di_dmstate;
107             buf->bs_aextents = dic->di_anextents;
108             buf->bs_forkoff = XFS_IFORK_BOFF(ip);
109
110             switch (dic->di_format) {
111             case XFS_DINODE_FMT_DEV:
112                     buf->bs_rdev = ip->i_df.if_u2.if_rdev;
113                     buf->bs_blksize = BLKDEV_IOSIZE;
114                     buf->bs_blocks = 0;

That's clearly not code in xfs_iflush_cluster() or any function that
xfs_iflush_cluster() calls. Indeed, xfs_iflush_cluster() is only
0x411 bytes long on that kernel, so there's more than 2x the amount
of code in that function in your instrumented kernel than mine.

Hence I need the address-to-line number translation from your kernel
to tell me what line of code is being tripped over.

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com