xfs
[Top] [All Lists]

Anyone grok pf_batch_read() in xfs_repair?

To: xfs-oss <xfs@xxxxxxxxxxx>
Subject: Anyone grok pf_batch_read() in xfs_repair?
From: Eric Sandeen <sandeen@xxxxxxxxxxx>
Date: Tue, 27 Aug 2013 14:57:03 -0500
Delivered-to: xfs@xxxxxxxxxxx
User-agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.8; rv:17.0) Gecko/20130801 Thunderbird/17.0.8
There's a pretty clear buffer overflow in pf_batch_read() but I'm having
a hard time wrapping my head around what it's supposed to be doing:

<snip>
                num = 0;
                if (which == PF_SECONDARY) {
                        bplist[0] = btree_find(args->io_queue, 0, &fsbno);
                        max_fsbno = MIN(fsbno + pf_max_fsbs,
                                                        args->last_bno_read);
                } else {
                        bplist[0] = btree_find(args->io_queue,
                                                args->last_bno_read, &fsbno);
                        max_fsbno = fsbno + pf_max_fsbs;
                }
                while (bplist[num] && num < MAX_BUFS && fsbno < max_fsbno) {
                        if (which != PF_META_ONLY ||
                            !B_IS_INODE(XFS_BUF_PRIORITY(bplist[num])))
                                num++;
                        bplist[num] = btree_lookup_next(args->io_queue, &fsbno);
                }
<snip>

That while loop is busted; we can increase num from (MAX_BUFS-1) to MAX_BUFS,
and then index the bplist array with MAX_BUFS, which is 1 past its end -
this corrupts memory.  Derp!

But I'm just not grokking what this is supposed to be doing.

If we come in w/ (which == PF_META_ONLY) we'll immediately overwrite
bplist[0] set in the if/else case above.  It all seems weird.
Anybody understand this offhand or should I just keep digging?

-Eric

<Prev in Thread] Current Thread [Next in Thread>