From owner-cvs-all  Sun Aug 22  9:25: 1 1999
Delivered-To: cvs-all@freebsd.org
Received: from godzilla.zeta.org.au (godzilla.zeta.org.au [203.26.10.9])
	by hub.freebsd.org (Postfix) with ESMTP
	id CB4BB15518; Sun, 22 Aug 1999 09:24:00 -0700 (PDT)
	(envelope-from bde@godzilla.zeta.org.au)
Received: (from bde@localhost)
	by godzilla.zeta.org.au (8.8.7/8.8.7) id CAA30375;
	Mon, 23 Aug 1999 02:21:56 +1000
Date: Mon, 23 Aug 1999 02:21:56 +1000
From: Bruce Evans <bde@zeta.org.au>
Message-Id: <199908221621.CAA30375@godzilla.zeta.org.au>
To: bde@zeta.org.au, phk@critter.freebsd.dk
Subject: Re: cvs commit: src/sys/kern kern_physio.c
Cc: cvs-all@FreeBSD.org, cvs-committers@FreeBSD.org,
	green@FreeBSD.org
Sender: owner-cvs-all@FreeBSD.ORG
Precedence: bulk

>>btodb() probably shouldn't cast to daddr_t.  This mainly prevents its
>>use in code like the above.  It is not a feature that it may prevent
>>warnings about truncation when off_t's are assigned to daddr_t's.
>
>btodb() probably shouldn't be used for off_t...

It seems to be used mostly to convert fs block sizes to "disk blocks".
The places that use it for off_t's seem to mostly have the same bug
that you fixed.  There's one in vfs_aio.c, one in vinumio.c, and 3 in
each of spec_vnops.c and devfs_vnops.c.  Only he ones for getpages in
the last to files have a bounds check, and it is verbose and broken
(I forgot the sign bit in daddr_t and have been sitting on the fix
for a year or 2) :-(.  It is used with a long arg in a few places.
Longs are the same as off_t's on alphas, so overflows can't be ruled
out by compile-time checking.

There don't seem to be enough checks for uio_offset < 0.  Negative offsets
may be set by lseek(), and I think they normally get passed to quite low
level routines.  ffs_read() handles them dubiously by interpreting them
as large unsigned offsets and returning EFBIG.  POSIX.1 seems to require
disallowing bogus seeks so that read() and write() don't have to worry
about bogus offsets (there is no documented error return for attempts
to read from bogus offsets in POSIX.  FreeBSD's read.2 documents EINVAL
for attempts to read from negative offsets, but negative offsets may
be non-bogus for some files, e.g., /dev/mem on a 64-bit system).

Bruce


To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe cvs-all" in the body of the message