linux-kernel - Re: [PATCH 0/1] devpts: use dynamic

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CA+55aFzqedKjjq+FDF3AEgyQfqQRHyR-6PTR_BTvyyQLqnPwPw@mail.gmail.com>
Date:   Wed, 16 Aug 2017 11:26:37 -0700
From:   Linus Torvalds <torvalds@...ux-foundation.org>
To:     Christian Brauner <christian.brauner@...ntu.com>
Cc:     Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        "Serge E. Hallyn" <serge@...lyn.com>,
        Al Viro <viro@...iv.linux.org.uk>
Subject: Re: [PATCH 0/1] devpts: use dynamic_dname() to generate proc name

On Wed, Aug 16, 2017 at 10:12 AM, Christian Brauner
<christian.brauner@...ntu.com> wrote:
>
> Recently the kernel has implemented the TIOCGPTPEER ioctl() call which allows
> users to retrieve an fd for the slave side of a pty solely based on the
> corresponding fd for the master side. The ioctl()-based fd retrieval however
> causes the "/proc/<pid>/fd/<pty-slave-fd>" symlink to point to the wrong dentry.
> Currently, it will always point to "/". The following simple program can be used
> to illustrate the problem when run on a system that implements the TIOCGPTPEER
> ioctl() call.

I think your patch is wrong - we need to actually use the *right*
path, rather than hardcode "/dev/pts/%d" in there.

Hardcoding "/dev/pts/%d" is something that user space can already do.
The kernel can and should do better.

That "dynamic_dname()" helper is for things that don't really have a
real pathname at all, so things like pipes and other file descriptors
that were opened without an associated entry in a filesystem (sockets,
other "special" inodes with no filesystem backing).

For things like pts slaves, we actually *do* have a real pathname -
and we should expose it. If the devpts filesystem is mounted somewhere
else than /dev/pts, it should give that correct pathname.

I also think your test program is a bit buggy, and silly:

>         ret = snprintf(path, 4096, "/proc/self/fd/%d", slave);
>         if (ret < 0 || ret >= 4096)
>                 goto on_error;
>
>         ret = readlink(path, buf, sizeof(buf));
>         if (ret < 0)
>                 goto on_error;
>         printf("I point to \"%s\"\n", buf);

This just smells wrong to me. And not just because you don't
NUL-terminate the readlink() return value (or just use "%.*s" with
ret, buf) .

We should just give people a better way to get the pathname than that
readlink on /proc/self/fd/<fd> thing. It's kind of ridiculous that we
don't. We already have a "getcwd()" system call that only does it for
cwd.

So I think we could/should just add a system call for 'fdname()' or similar.

                Linus