signal handling

问题:

I'm just playing with signal in Mac OS X.

Why does the following code not produce the default behavior of SIGSEGV after my signal handler has finished? Under Linux, the code works fine.

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

void err_crash();

void my_signal_handler (int signal)
{
    fprintf(stderr, "signal == %i\n", signal);
    fflush(stderr);
    err_crash();
}

void err_crash()
{
    static int count= 0;
    if (count)
        signal(SIGSEGV, SIG_DFL);       /* break recursion loop if we recursed */
    count++;

    // one of the two should really crash ;)
    ((void)(*((int volatile *)NULL)));
    *((int *)NULL) = 42;

    abort();                            /* in case we're not crashed yet... */
}

int main ()
{
    signal(SIGSEGV, my_signal_handler);
    err_crash();
    return 0;
}

EDIT: The output I get is the following:

bonecrusher:devel sw$ g++ signal_problems.cpp -o foo
bonecrusher:devel sw$ ./foo 
signal == 11
^C
bonecrusher:devel sw$

The problem is that I want that the program terminates after the output of signal == 11, but it rans forever and I have to interrupt it.

回答1:

This actually caused me brain freeze for a few minutes, and the reason why one should never use signal() in this day and age only grew stronger in me.

First of all, from the man pages for signal()

The behavior of signal() varies across UNIX versions, and has also varied historically across different versions of Linux. Avoid its use: use sigaction(2) instead.

and further down :

• If the disposition is set to a function, then first either the disposition is reset to SIG_DFL, or the signal is blocked (see Portability below), and then handler is called with argument signum. If invocation of the handler caused the signal to be blocked, then the signal is unblocked upon return from the handler.

In the original Unix systems, when a handler was installed, the disposition was reset to SIG_DFL, did not block incoming signals of the same type, and then it ran the handler function. System V provided this, and the linux kernel does the same.

This means that, once the code is run on a linux system, once second exception is called, it will exit directly.

Now to the fun part. BSD tried to improve this behaviour. From the man pages again:

On BSD, when a signal handler is invoked, the signal disposition is not reset, and further instances of the signal are blocked from being delivered while the handler is executing.

And since mac osx is partly based on BSD, once the code is run on a mac osx, once second exception is called, it will be pending and wait for the handler of the first exception to exit. But since you will never exit, you have a deadlock.

Thats why one should use sigaction() instead and never signal().

Now to some tips:

Handlers should be short, and return quickly. If you are performing calculations and calling other functions you are probably doing something wrong. Signals are not a substitute for an event driven framework.

Calling functions that are not async-safe is bad. Consider what would happen if an exception happened during a call to fprintf, and inside the handler fprintf was called again. Both the signal handlers and the programs data could be corrupted since they operate on the stream itself.

Some more reading : "Do" and "Don't" inside A Signal Handler

回答2:

According to POSIX:

If any of the SIGFPE, SIGILL, SIGSEGV, or SIGBUS signals are generated while they are blocked the result is undefined, unless the signal was generated by the kill() function, the sigqueue() function, or the raise() function.

Because SIGSEGV is blocked while in the SIGSEGV signal handler, the result is undefined in this case and any behavior is valid. If you don't want it to be blocked you could install the signal handler using sigaction() with the SA_NODEFER flag, or use sigprocmask() to unblock the signal within the signal handler.

回答3:

Run it in a debugger and single-step the instructions that you expect to crash.

There is no guarantee that writing to invalid addresses must create a segmentation fault. Perhaps Mac OS X maps those addresses for you, and you're overwriting something benign.