linux-kernel - Re: [PATCH 0/9] overlay: fix inconsistency of ro file after copy-up

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: <20170219091430.GL29622@ZenIV.linux.org.uk>
Date:   Sun, 19 Feb 2017 09:14:41 +0000
From:   Al Viro <viro@...IV.linux.org.uk>
To:     Miklos Szeredi <mszeredi@...hat.com>
Cc:     linux-unionfs@...r.kernel.org, linux-kernel@...r.kernel.org,
        linux-fsdevel@...r.kernel.org
Subject: Re: [PATCH 0/9] overlay: fix inconsistency of ro file after copy-up

On Fri, Feb 17, 2017 at 05:09:29PM +0100, Miklos Szeredi wrote:
> A file is opened for read-only, opened read-write (resulting in a copy up)
> and modified.  The data read back from the the read-only fd will be stale
> in this case (the read-only file descriptor still refers to the lower,
> unmodified file).
> 
> This patchset fixes issues related to this corner case.  This is a
> requirement from various parties for accepting overlayfs as a "POSIX"
> filesystem.
> 
> When an operation (read, mmap, fsync) is done on an overlay fd opened
> read-only that is referring to a lower file, check if it has been copied up
> in the mean time.  If so, open the upper file and use that for the operation.
> 
> To make the performance impact minimal for non-overlay case, use a flag in
> file->f_mode to indicate that this is an overlay file.

This is one hell of a DoS vector - it's really easy to eat tons of struct
file that way.  Preparatory parts of that series make sense on their own,
but your "let's allocate a struct file, call ->open() and schedule that
struct file for closing upon the exit to userland on each kernel_read()"
is not.