__kernel_vsyscall () hangs in SIGCHLD handler

__kernel_vsyscall () hangs in SIGCHLD handler

[John Z. Bohach]

Is there some reason that syslog() sleeps in __kernel_vsyscall() when 
invoked from a signal handler?  Is it that I am not allowed to call any 
system calls from inside a signal handler?

I use syslog() from a daemon client/server sys. app. that (tries) to log 
whenever a child exits.  I've registered a sigChld handler via 
sigaction(using SA_SIGINFO semantics), and everything works fine and is 
fully functional, including calls to syslog() in the normal program 
flow.  However, if I try to use syslog() from inside the signal handler 
itself, it never returns.

Tracing the program with gdb, I get:

 Attaching to program: lfsad, process 8845
`system-supplied DSO at 0xffffe000' has disappeared; keeping its 
symbols.
[Thread debugging using libthread_db enabled]
[New Thread -1211890000 (LWP 8845)]
Loaded symbols for /usr/lib/libxml2.so.2
Loaded symbols for /lib/libz.so.1
Loaded symbols for /lib/libm.so.6
Loaded symbols for /lib/librt.so.1
Loaded symbols for /lib/libc.so.6
Loaded symbols for /lib/libdl.so.2
Loaded symbols for /lib/ld-linux.so.2
Loaded symbols for /lib/libpthread.so.0
Reading symbols from /lib/libnss_files.so.2...done.
Loaded symbols for /lib/libnss_files.so.2
0xffffe410 in __kernel_vsyscall ()
(gdb) bt
#0  0xffffe410 in __kernel_vsyscall ()
#1  0xb7d2457e in __lll_mutex_lock_wait () from /lib/libc.so.6
#2  0xb7d14d6d in _L_mutex_lock_621 () from /lib/libc.so.6
#3  0x08511b18 in ?? ()
#4  0xbf86b64d in ?? ()
#5  0xbf86b628 in ?? ()
#6  0xbf86b578 in ?? ()
#7  0xb7d76320 in ?? () from /lib/libc.so.6
#8  0xfbad8001 in ?? ()
#9  0x00000014 in ?? ()
#10 0x00000000 in ?? ()

If I check its status in /proc/8845/status, I see that its sleeping.  If 
I remove the call to syslog() in the signal handler, everything is 
fine, no sleeping, and the rest of the syslog() work fine.

Am I doing something wrong?

Thanks,
John

Re: __kernel_vsyscall () hangs in SIGCHLD handler

[Ulrich Drepper]

On 9/26/07, John Z. Bohach <jzb2@aexorsyst.com> wrote:
> Is there some reason that syslog() sleeps in __kernel_vsyscall() when
> invoked from a signal handler?

Only very few functions are allowed to be called from signal handlers.
 This is clearly spelled out in the POSIX spec.  Section XSH 2.4.3
lists the allowed functions.  syslog of course is not on it.

Re: __kernel_vsyscall () hangs in SIGCHLD handler

[John Z. Bohach]

On Wednesday 26 September 2007 10:03:33 Ulrich Drepper wrote:
> On 9/26/07, John Z. Bohach <jzb2@aexorsyst.com> wrote:
> > Is there some reason that syslog() sleeps in __kernel_vsyscall()
> > when invoked from a signal handler?
>
> Only very few functions are allowed to be called from signal
> handlers. This is clearly spelled out in the POSIX spec.  Section XSH
> 2.4.3 lists the allowed functions.  syslog of course is not on it. -
> To unsubscribe from this list: send the line "unsubscribe
> linux-kernel" in the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

Thank you for this information...also thanks to Giacomo who answered me 
offline...

--john

Re: __kernel_vsyscall () hangs in SIGCHLD handler

[Mikael Pettersson]

Ulrich Drepper writes:
 > On 9/26/07, John Z. Bohach <jzb2@aexorsyst.com> wrote:
 > > Is there some reason that syslog() sleeps in __kernel_vsyscall() when
 > > invoked from a signal handler?
 > 
 > Only very few functions are allowed to be called from signal handlers.
 >  This is clearly spelled out in the POSIX spec.  Section XSH 2.4.3
 > lists the allowed functions.  syslog of course is not on it.

The Linux kernel itself imposes no restrictions on what you can
do in user-space signal handlers.

However, user-space is a different story. The interrupted thread
may have held a lock, in which case calling code from the signal
handler that tries to take the same lock may result in a deadlock.
Or the thread may be been in the process of updating some private
data, in which case calling code from the signal handler that tries
to access that data may result in data corruption. And then there's
libc which may wrap your signal handler with code doing unspecified
libc magic. And so on, the possibilities for failure are endless :-)