Libc thread cancelation safety?

Libc thread cancelation safety?

[Dean Anderson]

I have noticed that printf in glibc (glibc 2.3.5, i386) is not pthread
cancellation safe, though printf on Solaris is cancellation safe.  
Specifically, canceling a thread in printf leads to deadlock.

What should one expect regarding cancelation safety in standard
libraries?  Is this a bug in glibc?

I haven't done a lot of testing of other functions in the standard
library, so I don't know if this affects more than just printf and
friends.

Working around it is easy, but I'm not sure how many functions one needs
to do this for...'Everything' might be a bit tedious.

static int safe_printf(char *fmt, ...)
{
    va_list ap;
    int cancelstate;
    int ret, status;

    va_start(ap, fmt);
    status = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &cancelstate);
    ret = vprintf(fmt, ap);
    status = pthread_setcancelstate(cancelstate, &cancelstate);
    return (ret);
}

#define printf safe_printf

I have a test program if anyone is interested.

		--Dean



-- 
Av8 Internet   Prepared to pay a premium for better service?
www.av8.net         faster, more reliable, better service
617 344 9000   

Re: Libc thread cancelation safety?

[Florian Weimer]

* Dean Anderson:

> I have noticed that printf in glibc (glibc 2.3.5, i386) is not pthread
> cancellation safe, though printf on Solaris is cancellation safe.  
> Specifically, canceling a thread in printf leads to deadlock.
>
> What should one expect regarding cancelation safety in standard
> libraries?  Is this a bug in glibc?

Asynchronous cancellation is inherently unsafe.  I can't find a full
list of functions which may be called from a thread which is subject
to asynchronous cancellation right now, but it is undoubtly very short
and does not include anything which touches the heap or uses the stdio
I/O facilities.  Most bare system calls (read, write, close, pipe)
should be fine, though.

Re: Libc thread cancelation safety?

[Ian Lance Taylor]

Dean Anderson <dean@av8.com> writes:

> I have noticed that printf in glibc (glibc 2.3.5, i386) is not pthread
> cancellation safe, though printf on Solaris is cancellation safe.  
> Specifically, canceling a thread in printf leads to deadlock.
>
> What should one expect regarding cancelation safety in standard
> libraries?  Is this a bug in glibc?

This is probably better directed to a glibc mailing list.  See
http://www.gnu.org/software/libc/ .

That said, I doubt that any standard requires that printf be
async-cancel safe.

Ian