reopen STDIN and STDOUT as dev/null

[Maxime2]

It is better to reopen STDIN and STDOUT as well as they may be used in a third-party application using ccan.

[Maxime2]

But STDIN is read only while STDOUT and STDERR are write only.

[rustyrussell]

Good point! A couple of typos to fix is all it needs then!

[Maxime2]

Nope. If the new fd is already open, dup2() would close it before reusing. If the old fd is equal to the new fd dup2() does nothing. So there is no point to close fd if it is below or equal to STDERR_FILENO.

[rustyrussell]

Maxim Zakharov [email protected] writes:

Nope. If the new fd is already open, dup2() would close it before reusing. If the old fd is equal to the new fd dup2() does nothing. So there is no point to close fd if it is below or equal to STDERR_FILENO.

Well, if you're assuming the continuity of stdin, stdout, and stderr, then how about this:

    if (open("/dev/null", O_RDONLY) != STDIN_FILENO)
            return false;
    /* Save some kernel memory by dup() for stdout & stderr */
    if (dup(open("/dev/null", O_WRONLY)) != STDERR_FILENO)
            return false;

Cheers, Rusty.

[Maxime2]

It looks very good, but perhaps we need to add a comment that this code is rely on unix's property to always use lowest unused file descriptor. As I remember, Windows do not state such guaranty.

[rustyrussell]

Maxim Zakharov [email protected] writes:

It looks very good, but perhaps we need to add a comment that this code is rely on unix's property to always use lowest unused file descriptor. As I remember, Windows do not state such guaranty.

Hmm, does this code even work on Windows at all? If so, might be nice to do it the long way, but it'll need a comment explaining the reason for the > test instead of == :)

Thanks, Rusty.