This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] Add the --with-tty-gid configure option [BZ #24941]
- From: "Gabriel F. T. Gomes" <gabriel at inconstante dot net dot br>
- To: Florian Weimer <fweimer at redhat dot com>
- Cc: <libc-alpha at sourceware dot org>
- Date: Thu, 31 Oct 2019 17:41:37 -0300
- Subject: Re: [PATCH] Add the --with-tty-gid configure option [BZ #24941]
- References: <87h862kdyx.fsf@oldenburg2.str.redhat.com>
Hi, Florian,
Thanks for, in the bug report [1], helping me understand the rationale for
this patch (the problem was well described here, anyway, I had just failed
to notice the details).
[1] https://sourceware.org/bugzilla/show_bug.cgi?id=24941#c3
On Tue, 27 Aug 2019, Florian Weimer wrote:
>2019-08-27 Florian Weimer <fweimer@redhat.com>
>
> [BZ #24941]
> Add the --with-tty-gid configure option.
> * config.make.in (define-tty-gid-cflags): New variable.
> * configure.ac (--with-tty-gid): New option.
> * manual/install.texi (Configuring and compiling): Document
> --with-tty-gid.
> * sysdeps/unix/Makefile [$(subdir) == login] (CFLAGS-grantpt.c):
> Add $(define-tty-gid-cflags).
> * sysdeps/unix/grantpt.c (grantpt): Use TTY_GID constant for gid
> if defined.
> * configure, INSTALL: Regeneration.
I thinks this needs a new commit message, preferably one that summarizes
why this patch is useful... Something in the lines of:
When, in multi-threaded applications, a forked process tries to call
grantpt, it could hang in an internal NSS lock (__nss_database_lookup2),
if the fork happened while NSS was used in another thread. This patch
adds the --with-tty-gid configure option, which avoids the hang, by
avoiding calls to NSS functions.
Other than that, the patch looks good to me.
Reviewed-by: Gabriel F. T. Gomes <gabrielftg@linux.ibm.com>
Thanks
>+@item --with-tty-gid=@var{GID}
>+Use @var{GID} as the value for the group ID of the @samp{tty} group.
>+This enables a more efficient implementation of the @code{grantpt}
>+function, and the function will also work reliably in a forked
>+subprocess of a multi-threaded process.
OK.
>+#ifdef TTY_GID
>+ /* TTY_GID is set by the --with-tty-gid configure option. */
>+ enum { gid = TTY_GID };
>+#else
> static int tty_gid = -1;
> if (__glibc_unlikely (tty_gid == -1))
> {
>@@ -154,6 +158,7 @@ grantpt (int fd)
> tty_gid = p->gr_gid;
> }
> gid_t gid = tty_gid == -1 ? __getgid () : tty_gid;
>+#endif
OK.