Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 1 Oct 2012 12:21:13 +0300
From:      Konstantin Belousov <kostikbel@gmail.com>
To:        John Baldwin <jhb@freebsd.org>
Cc:        arch@freebsd.org
Subject:   Re: stdio and short file descriptors revisited
Message-ID:  <20121001092113.GH35915@deviant.kiev.zoral.com.ua>
In-Reply-To: <201209281847.39663.jhb@freebsd.org>
References:  <201209281847.39663.jhb@freebsd.org>

next in thread | previous in thread | raw e-mail | index | archive | help

--Zlo2GrDIozf6aQFY
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Fri, Sep 28, 2012 at 06:47:39PM -0400, John Baldwin wrote:
> Four years or so ago I cleaned up some of the stdio internals as fallout =
=66rom=20
> running into problems with stdio using a short instead of an int to hold =
file=20
> descriptors.  Back then I got sidetracked with attempting to make FILE op=
aque=20
> and ended up never getting around to bumping _file from a short to an int=
=2E  I=20
> recently ran back into the SHRT_MAX limit at work again and came up with =
a=20
> patch to fix this.
>=20
> To preserve the ABI, it is necessary to leave the existing short _file in=
=20
> place and add a new int _file to the end of the FILE structure.  Also, fo=
r old=20
> applications, the old _file (_ofile in the patch) must still be valid.  T=
he=20
> approach I have taken is to bump the symbol version for routines that cre=
ate=20
> FILE objects with a non-fake _file (fopen, fdopen, and freopen).  The old=
=20
> FBSD_1.0 variants still fail if an fd is greater than SHRT_MAX (and thus=
=20
> cannot be safely stored in _ofile).  The new FBSD_1.3 variants assign to =
both=20
> _file and _ofile if the fd is less than SHRT_MAX.  I also changed fileno()
> to no longer be an inline macro in <stdio.h> but to always be a function =
call=20
> going forward.
>=20
> If folks think this is ok, I'll hack up a modified version that hides _fi=
le
> from outside consumers (rename it to _nfile or some such) and send it for=
 a
> ports-exp run before committing to make sure there aren't any 3rd party a=
pps
> accessing _file directly.
>=20
> http://www.FreeBSD.org/~jhb/patches/stdio_file.patch

The corner case left unhandled is the situation where we have a dso which
is linked against FBSD_1.0 version of libc, but which gets FILE * as an
API argument for some of its exported routines. If the implementation
uses fileno(3), it would fail. I have no idea how to fix this, most
likely, the issue is not fixable at all. Workaround seems to be to force
the __isthreaded to 1. Might be, as an ugly hack, some flag could be
added to the stdbuf(1), if anybody cares enough.

Otherwise, the patch looks good.


--Zlo2GrDIozf6aQFY
Content-Type: application/pgp-signature

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (FreeBSD)

iEYEARECAAYFAlBpYIkACgkQC3+MBN1Mb4gCHwCg1XHzpsvs6LeerJaEtZXC87/x
0dEAoPOL19RkL5gOVOc0lvvzKTsMFjY+
=6tu+
-----END PGP SIGNATURE-----

--Zlo2GrDIozf6aQFY--



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20121001092113.GH35915>