cvs commit: src/lib/libthr/thread thr_join.c

[Don Lewis]

On 28 May, Mike Makonnen wrote:
> On Tue, 27 May 2003 18:55:52 -0400
> Alexander Kabaev <ak03 at gte.com> wrote:
> 
>> 
>> konsole has similar problems when run with libkse. The ktrace + source
>> code inspection convinced me that it simply abort()'a itself, because
>> ttyname() function returns NULL in makePty() function. 
> 
> Yes, seems like it. More comments below.
> 
>> 
>> I do not think either threading library has anything to do with the
>> failure.
> 
> It doesn't look like it's a threading issue, but since libc_r doesn't bring out
> this behaviour in Konsole, they must be doing (or not doing) something (directly
> or indirectly) that triggers it.
> 
> I have finally taken the plunge and installed KDE, and I can reproduce the bug.
> However, it is unclear to me exactly what the problem is. To complicate things
> even further it doesn't even look look Konsole uses threads, and instead plays
> games with fork(2). I think we need someone who is familiar with the code to
> help out because I am having a hard time following it. Just to add a couple of
> data points:
> 
> It crashes because tcgetattr(3) called from libc/ttyname_unthreaded() returns
> ENOTTY to TEPty::makePty().
> 
> When it does crash it is also usually preceded by:
> 	konsole: cannot chown /dev/ttypc.
> 	Reason: Operation not permitted
> which is as a result of this code snippet from TEPty.cpp:openPty()
> 
>     if (openpty(&master_fd, &slave_fd, NULL, NULL, NULL) == 0) {
>       m_MasterFd = master_fd;
>       m_SlaveFd = slave_fd;
> #ifdef HAVE_PTSNAME
>       strncpy(ptynam, ptsname(master_fd), 50);
> 	...
> 	[snip]
> 	...
>       if (fchown(slave_fd, (uid_t) -1, gid) < 0)
>       {
>          m_bNeedGrantPty = true;
>          fprintf(stderr,"konsole: cannot chown %s.\n",ttynam);
>          perror("Reason");
>       }
> 
> 
> The fact that it works sometimes and abort()s at others suggests that there may
> be a race somewhere, but where...

According to the openpty() and fchown() man pages, the fchown() call
will fail with that error if 

	konsole isn't running with an effective UID of super-user at that
        point.
        
AND

	The slave device either isn't owned by the user or gid isn't
        a member of the user's group.