[PATCH 2/3] Test how PTRACE_SETOPTIONS support works

[Roland McGrath]

> Since PTRACE_O_TRACECLONE/*FORK were introduced to kernel
> at the same time, so this test seems enough for these 3 options.

I agree.

> +#ifdef LINUX
> +/* Test whether kernel support PTRACE_O_TRACECLONE et al options.
> + * First fork a new child, call ptrace with PTRACE_SETOPTIONS on it,
> + * and then see which options are supported on this kernel.
> + */
> +int test_ptrace_setoptions()

There is no reason this can't be static, so make it static.
Also, we're using prototypes now, so make its argument list (void).

> +	unsigned int test_options = PTRACE_O_TRACEFORK;

IMHO using an unchanging variable for this is less clear, not more.
Just use the constant in the actual call below.

> +		kill(getpid(), SIGSTOP);

Might as well use raise, though it's the same.

> +		if ((pid = fork()) < 0) {
> +			exit(1);
> +		}
> +		exit(0);

Not that it really matters here where the parent hasn't used stdio output
calls beforehand, but a fork child should use _exit rather than exit.

> +	}
> +	else {
> +		int status, tracee_pid, wait_errno;
> +		int no_child = 0;
> +		while(1) {

Put a space after the while keyword.

> +			if (tracee_pid != pid)
> +				ptrace(PTRACE_CONT, tracee_pid, 0, 0);

You check every other error, so might as well check the ptrace result too.

> +				ptrace(PTRACE_SYSCALL, pid, 0, 0);

Likewise here.

> +#ifdef LINUX
> +	if (test_ptrace_setoptions() < 0) {
> +		fprintf(stderr, "Test for options supported by PTRACE_SETOPTIONS\
> +			failed, give up using this feature\n");
> +		ptrace_setoptions = 0;

There is no reason to do the test at all unless followfork is set (-f/-F).


Thanks,
Roland