linux-kernel - Re: [PATCH v8 1/4] Landlock: Add abstract unix socket connect restriction

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 for Android: free password hash cracker in your pocket

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

Message-ID: <20240807.Be5aiChaf8ie@digikod.net>
Date: Wed, 7 Aug 2024 16:44:36 +0200
From: Mickaël Salaün <mic@...ikod.net>
To: Jann Horn <jannh@...gle.com>
Cc: Tahera Fahimi <fahimitahera@...il.com>, outreachy@...ts.linux.dev, 
	gnoack@...gle.com, paul@...l-moore.com, jmorris@...ei.org, serge@...lyn.com, 
	linux-security-module@...r.kernel.org, linux-kernel@...r.kernel.org, bjorn3_gh@...tonmail.com, 
	netdev@...r.kernel.org
Subject: Re: [PATCH v8 1/4] Landlock: Add abstract unix socket connect
 restriction

On Wed, Aug 07, 2024 at 03:45:18PM +0200, Jann Horn wrote:
> On Wed, Aug 7, 2024 at 9:21 AM Mickaël Salaün <mic@...ikod.net> wrote:
> > On Tue, Aug 06, 2024 at 10:46:43PM +0200, Jann Horn wrote:
> > > I think adding something like this change on top of your code would
> > > make it more concise (though this is entirely untested):
> > >
> > > --- /tmp/a      2024-08-06 22:37:33.800158308 +0200
> > > +++ /tmp/b      2024-08-06 22:44:49.539314039 +0200
> > > @@ -15,25 +15,12 @@
> > >           * client_layer must be a signed integer with greater capacity than
> > >           * client->num_layers to ensure the following loop stops.
> > >           */
> > >          BUILD_BUG_ON(sizeof(client_layer) > sizeof(client->num_layers));
> > >
> > > -        if (!server) {
> > > -                /*
> > > -                 * Walks client's parent domains and checks that none of these
> > > -                 * domains are scoped.
> > > -                 */
> > > -                for (; client_layer >= 0; client_layer--) {
> > > -                        if (landlock_get_scope_mask(client, client_layer) &
> > > -                            scope)
> > > -                                return true;
> > > -                }
> > > -                return false;
> > > -        }
> >
> > This loop is redundant with the following one, but it makes sure there
> > is no issue nor inconsistencies with the server or server_walker
> > pointers.  That's the only approach I found to make sure we don't go
> > through a path that could use an incorrect pointer, and makes the code
> > easy to review.
> 
> My view is that this is a duplication of logic for one particular
> special case - after all, you can also end up walking up to the same
> state (client_layer==-1, server_layer==-1, client_walker==NULL,
> server_walker==NULL) with the loop at the bottom.

Indeed

> 
> But I guess my preference for more concise code is kinda subjective -
> if you prefer the more verbose version, I'm fine with that too.
> 
> > > -
> > > -        server_layer = server->num_layers - 1;
> > > -        server_walker = server->hierarchy;
> > > +        server_layer = server ? (server->num_layers - 1) : -1;
> > > +        server_walker = server ? server->hierarchy : NULL;
> >
> > We would need to change the last loop to avoid a null pointer deref.
> 
> Why? The first loop would either exit or walk the client_walker up
> until client_layer is -1 and client_walker is NULL; the second loop
> wouldn't do anything because the walkers are at the same layer; the
> third loop's body wouldn't be executed because client_layer is -1.

Correct, I missed that client_layer would always be greater than
server_layer (-1).

Tahera, could you please take Jann's proposal?


> 
> The case where the server is not in any Landlock domain is just one
> subcase of the more general case "client and server do not have a
> common ancestor domain".
> 
> > >
> > >          /*
> > >           * Walks client's parent domains down to the same hierarchy level as
> > >           * the server's domain, and checks that none of these client's parent
> > >           * domains are scoped.
> > >
>