signal handling issue.

signal handling issue.

[Alex Davis]

There appears to be a change between linux 2.4 and 2.6
in how signals are handled. As a test, I wrote the program
below:

#include <stdio.h>
#include <signal.h>
#include <setjmp.h>

static jmp_buf env;

static void handler(int s) {
        printf("caught signal %d\n", s);
        longjmp(env, 1);
}

int main() {
        int * p = 0;

        printf("write\n");
        signal(SIGSEGV, handler);
        if ( ! setjmp(env) )
        {
                *p = 0;
        }

        printf("read\n");
        signal(SIGSEGV, handler);
        if ( ! setjmp(env) )
        {
                int a = *p;
        }
        return 0;
}

When run on 2.4.26, the program prints:

write
caught signal 11
read
caught signal 11


Which (I think) is expected, but when run on 2.6.5,
the program prints:

write
caught signal 11
read
Segmentation fault

It's as if the second call to signal is being ignored.
Is this a bug or a feature?

-Alex



=====
I code, therefore I am


	
		
__________________________________
Do you Yahoo!?
SBC Yahoo! - Internet access at a great low price.
http://promo.yahoo.com/sbc/

Re: signal handling issue.

[Ulrich Drepper]

Fix your code:

> static void handler(int s) {
>         printf("caught signal %d\n", s);
>         longjmp(env, 1);
> }

Use siglonjmp() and sigsetjmp().  You are not allowed to use longjmp()
to jump from a signal handler.

[How many times have I explained this here now?]

-- 
➧ Ulrich Drepper ➧ Red Hat, Inc. ➧ 444 Castro St ➧ Mountain View, CA ❖

Re: signal handling issue.

[Richard B. Johnson]

On Tue, 18 May 2004, Alex Davis wrote:

> There appears to be a change between linux 2.4 and 2.6
> in how signals are handled. As a test, I wrote the program
> below:
>
> #include <stdio.h>
> #include <signal.h>
> #include <setjmp.h>
>
> static jmp_buf env;
>
> static void handler(int s) {
>         printf("caught signal %d\n", s);
>         longjmp(env, 1);
> }

[SNIPPED...]

A a couple years ago they changed the rules. You can't longjmp
from a signal handler anymore. There's some other function
call with 'sig' in it, like siglongjmp or something like
that. Ahh... here it is:

/usr/include/setjmp.h:extern void siglongjmp __P ((sigjmp_buf __env, int __val))


Cheers,
Dick Johnson
Penguin : Linux version 2.4.26 on an i686 machine (5557.45 BogoMips).
            Note 96.31% of all statistics are fiction.

Re: signal handling issue.

[Alex Davis]

Some additional info: the 2.6.5 system has gcc 3.3.3,
the 2.4.26 system has gcc 3.3.1. Both systems have
glibc 2.3.2.

-Alex

=====
I code, therefore I am


	
		
__________________________________
Do you Yahoo!?
SBC Yahoo! - Internet access at a great low price.
http://promo.yahoo.com/sbc/

Re: signal handling issue.

[Alex Davis]

>Use siglonjmp() and sigsetjmp().  You are not allowed to use longjmp()
>to jump from a signal handler.

Is this documented anywhere??

=====
I code, therefore I am


	
		
__________________________________
Do you Yahoo!?
SBC Yahoo! - Internet access at a great low price.
http://promo.yahoo.com/sbc/