linux-kernel - Re: [AppArmor 19/45] Add struct vfsmount parameters to vfs

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-next>] [day] [month] [year] [list]

Message-Id: <E1IntXa-0001JL-Qw@be1.7eggert.dyndns.org>
Date:	Fri, 02 Nov 2007 11:14:14 +0100
From:	Bodo Eggert <7eggert@....de>
To:	Al Viro <viro@....linux.org.uk>, John Johansen <jjohansen@...e.de>,
	akpm@...ux-foundation.org, linux-kernel@...r.kernel.org,
	linux-security-module@...r.kernel.org, Tony Jones <tonyj@...e.de>,
	Andreas Gruenbacher <agruen@...e.de>
Subject: Re: [AppArmor 19/45] Add struct vfsmount parameters to vfs_rename()

Al Viro <viro@....linux.org.uk> wrote:
> On Fri, Oct 26, 2007 at 11:23:53AM -0700, John Johansen wrote:

>> In the current code, both vfsmounts are always identical, and so one of
>> the two should go, agreed.
>> 
>> The thought behind passing both vfsmounts was that they could differ but
>> point to the same super_block, in which case renames would still be
>> possible at least from a filesystem point of view. The essential
>> restriction here is that both files must be on the same device; the vfs
>> restriction of not allowing cross-mount renames is arbitrary.
> 
> It's called "access control".  Pathname-based one, BTW.  And yes, it's
> 100% deliberate.

I doubt anybody uses bind mounts & co instead of symlinks in order to
prevent rename() while still allowing to move files by either copying
or by using the source file in the bound directory. At least I expected
bind mounted directories to behave like symlinked ones, minus the problems
of symlinks. 

Since this feature only protects you from rename(src/foo,dst/foo) if
1) There is no way to access src and dst in the same mount space
2) src and dst are writebale by the attacker
3) Unlinking src/foo is OK
4) Renaming src/foo is OK as long as it's within the same mount as foo
5) Symlinking src/foo to dst/foo is OK
6) Creating dst/foo having a different owner is OK
7) Having dst/foo with the original content and owner from src/foo is _not_ OK
8) Moon crashes on earth
, I'd rather like to have a fast mv.

>> Cross-mount renames are not allowed currently, and granted, they may not
>> be very useful, either.
> 
> <raised brows>
> Excuse me, but IIRC LSM was supposed to _add_ restrictions, not to remove
> existing security checks.

Security checks as in "we built a steel door into that Chinese paper wall"?

As far as I understand, the restriction would not be removed by the LSM
explicitely allowing it, but by the fixed vfs then being able to handle
cross-mountpoint-renames. Maybe yo'll want to keep the ability for the users
who use bind mounts in order to not allow rename() ... both of them.-)

/me prepares for the impact of a large round object on earth.

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/