block: make bio_check_eod work for zero sized devices

commit 3eb9694 upstream.

Since the dawn of time bio_check_eod has a check for a non-zero size of
the device.  This doesn't really make any sense as we never want to send
I/O to a device that's been set to zero size, or never moved out of that.

I am a bit surprised we haven't caught this for a long time, but the
removal of the extra validation inside of zram caused syzbot to trip
over this issue recently.  I've added a Fixes tag for that commit, but
the issue really goes back way before git history.

Fixes: 9fe95ba ("zram: remove valid_io_request")
Reported-by: syzbot+b8d61a58b7c7ebd2c8e0@syzkaller.appspotmail.com
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20230524060538.1593686-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>

Backport note for 6.1.y:

The same underlying bug affects 6.1.y even though 9fe95ba is not in
that tree.  The syzbot reproducer attaches a socket to /dev/nbd0 without
setting its capacity, then mounts it as BEFS.  BEFS calls sb_bread() while
bdev_nr_sectors() is zero.  The maxsector guard makes bio_check_eod() accept
that read, which reaches NBD and waits for a response instead of failing
with -EIO.  syzbot reproduced the resulting __bread_gfp() hang on 6.1.y and
identified this upstream commit by fix bisection.

Reported-by: syzbot+2aca91e1d3ae43aef10c@syzkaller.appspotmail.com
Closes: https://syzkaller.appspot.com/bug?extid=2aca91e1d3ae43aef10c
Signed-off-by: Miguel García <miguelgarciaroman8@gmail.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>

Author: Christoph Hellwig

block/blk-core.c

@@ -515,7 +515,7 @@ static inline int bio_check_eod(struct bio *bio)
 	sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
 	unsigned int nr_sectors = bio_sectors(bio);
 
-	if (nr_sectors && maxsector &&
+	if (nr_sectors &&
 	    (nr_sectors > maxsector ||
 	     bio->bi_iter.bi_sector > maxsector - nr_sectors)) {
 		pr_info_ratelimited("%s: attempt to access beyond end of device\n"