Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sat, 20 Dec 2014 20:38:58 -0500
From: Rich Felker <dalias@...c.org>
To: musl@...ts.openwall.com
Subject: Re: Add login_tty

On Sat, Dec 20, 2014 at 07:58:21PM -0500, Rich Felker wrote:
> On Mon, Nov 03, 2014 at 07:29:54PM +0100, Felix Janda wrote:
> > Thanks for the review. Below a new version.
> 
> Sorry I didn't get around to reviewing this right away.
> 
> > #include <pty.h>
> > #include <utmp.h>
> > #include <unistd.h>
> > 
> > int forkpty(int *m, char *name, const struct termios *tio, const struct winsize *ws)
> > {
> > 	int s, ec, p[2];
> > 	pid_t pid;
> > 
> > 	if (openpty(m, &s, name, tio, ws) < 0) return -1;
> > 	if (pipe2(p, O_CLOEXEC)) {
> > 		close(s);
> > 		goto fail;
> > 	}
> > 
> > 	pid = fork();
> > 	if (!pid) {
> > 		close(*m);
> > 		close(p[0]);
> > 		ec = login_tty(s);
> 
> login_tty could end up closing the pipe if stdin/out/err were
> initially closed in the parent, since p[1] might be 0/1/2 in that
> case. I think we need to check for this and move p[1] to a new fd in
> that case (and fail if that fails) before calling login_tty.

Actually this is a non-issue, since login_tty has committed itself to
returning success by the time it dup2's over top of file descriptors
0/1/2.

However I noticed another small issue:

> > 		while (write(p[1], &ec, sizeof ec) < 0);

This is writing -1, not the errno value.

> > 		if (ec) _exit(127);
> > 		close(p[1]);
> > 		return 0;
> > 	}
> > 	close(s);
> > 	close(p[1]);
> > 	if (pid > 0) read(p[0], &ec, sizeof ec);
> 
> This read probably needs to retry-loop, in case the parent has
> interrupting signal handlers.

I'm working on an improvement and I think it's better to just block
signals for the whole function. Then the retry loop wouldn't be
needed. The reason is that we don't want to allow a signal handler to
run in a child process that "never existed" from the application's
perspective.

> 
> > 	close(p[0]);
> > 	if (pid > 0) {
> > 		if (!ec) return pid;
> > 		waitpid(pid, &(int){0}, 0);
> 
> I think waitpid could in principle fail too, but it probably shouldn't
> since the process is already dead at the time waitpid is called.

Then the retry is unneeded here too.

I've got a draft based on these comments that I'll post soon for
review.

Rich

Powered by blists - more mailing lists

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.