Date: Sat, 10 Mar 2012 15:00:36 GMT From: Ed Schouten <ed@80386.nl> To: freebsd-bugs@FreeBSD.org Subject: Re: bin/143017: watch(8): fatal: cannot attach to tty Message-ID: <201203101500.q2AF0aKC045266@freefall.freebsd.org>
next in thread | raw e-mail | index | archive | help
The following reply was made to PR bin/143017; it has been noted by GNATS.
From: Ed Schouten <ed@80386.nl>
To: Mark <markjdb@gmail.com>
Cc: bug-followup@FreeBSD.org, dummy@smtp.ru
Subject: Re: bin/143017: watch(8): fatal: cannot attach to tty
Date: Sat, 10 Mar 2012 15:59:56 +0100
--z0eOaCaDLjvTGF2l
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable
Hello Mark,
* Mark <markjdb@gmail.com>, 20120310 09:50:
> How about something like the attached change to snp(4)? It has snp
> detach from the tty when the input to SNPSTTY is -1, which is what the
> man page says is supposed to happen.
This isn't the right approach. It would break when a snoop descriptor is
used by two or more threads. Imagine what would happen if one thread
would call read(), while the other would call SNPSTTY. The only way you
could implement this safely, is by locking down all the system calls
using for example an sx lock.
I think the easiest way to fix this would be by simply patching watch(8)
to not reuse snoop descriptors to switch between TTYs. The following
(ugly) patch should make watch(8) work properly:
Index: usr.sbin/watch/watch.c
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
--- usr.sbin/watch/watch.c (Revision 232781)
+++ usr.sbin/watch/watch.c (Arbeitskopie)
@@ -206,10 +206,9 @@
static void
detach_snp(void)
{
- int fd;
=20
- fd =3D -1;
- ioctl(snp_io, SNPSTTY, &fd);
+ close(snp_io);
+ snp_io =3D open_snp();
}
=20
static void
@@ -217,6 +216,7 @@
{
int snp_tty;
=20
+ detach_snp();
snp_tty =3D open(dev_name, O_RDONLY | O_NONBLOCK);
if (snp_tty < 0)
fatal(EX_DATAERR, "can't open device");
If someone would write a nice set of patches to clean up the watch(8)
source code (and fix this bug), I'd be more than happy to commit them to
SVN. In fact, I think we should simplify the watch(8) utility. In my
opinion the `interactive mode' is pretty useless.
--=20
Ed Schouten <ed@80386.nl>
WWW: http://80386.nl/
--z0eOaCaDLjvTGF2l
Content-Type: application/pgp-signature
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (FreeBSD)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=K2kY
-----END PGP SIGNATURE-----
--z0eOaCaDLjvTGF2l--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201203101500.q2AF0aKC045266>