Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Tue, 4 Apr 2023 12:45:06 +0200
From:   Oleg Nesterov <oleg@redhat.com>
To:     Catalin Marinas <catalin.marinas@arm.com>
Cc:     Gregory Price <gourry.memverge@gmail.com>,
        linux-kernel@vger.kernel.org, linux-doc@vger.kernel.org,
        linux-arch@vger.kernel.org, avagin@gmail.com, peterz@infradead.org,
        luto@kernel.org, krisman@collabora.com, tglx@linutronix.de,
        corbet@lwn.net, shuah@kernel.org, arnd@arndb.de,
        Gregory Price <gregory.price@memverge.com>
Subject: Re: [PATCH v15 2/4] syscall user dispatch: untag selector addresses
 before access_ok
Message-ID: <20230404104506.GA24740@redhat.com>
References: <20230330212121.1688-1-gregory.price@memverge.com>
 <20230330212121.1688-3-gregory.price@memverge.com>
 <ZCYP+4gRZDqC0lRo@arm.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <ZCYP+4gRZDqC0lRo@arm.com>
User-Agent: Mutt/1.5.24 (2015-08-30)
Precedence: bulk

Catalin,

doesn't this mean that access_ok() on arm64 could use
untagged_addr(addr) unconditionally without any security risk?

On 03/30, Catalin Marinas wrote:
>
> On Thu, Mar 30, 2023 at 05:21:22PM -0400, Gregory Price wrote:
> > diff --git a/kernel/entry/syscall_user_dispatch.c b/kernel/entry/syscall_user_dispatch.c
> > index 22396b234854..16086226b41c 100644
> > --- a/kernel/entry/syscall_user_dispatch.c
> > +++ b/kernel/entry/syscall_user_dispatch.c
> > @@ -87,7 +87,18 @@ static int task_set_syscall_user_dispatch(struct task_struct *task, unsigned lon
> >  		if (offset && offset + len <= offset)
> >  			return -EINVAL;
> >
> > -		if (selector && !access_ok(selector, sizeof(*selector)))
> > +		/*
> > +		 * access_ok will clear memory tags for tagged addresses on tasks where
> > +		 * memory tagging is enabled.  To enable a tracer to set a tracee's
> > +		 * selector not in the same tagging state, the selector address must be
> > +		 * untagged for access_ok, otherwise an untagged tracer will always fail
> > +		 * to set a tagged tracee's selector.
> > +		 *
> > +		 * The result of this is that a tagged tracer may be capable of setting
> > +		 * an invalid address, and the tracee will SIGSEGV on the next syscall.
> > +		 * This is equivalent to a task setting a bad selector (selector=0x1).
> > +		 */
>
> I'd drop the last paragraph above. Even without tagged pointers, a tracer
> can set an invalid address (as you already mentioned) but the phrasing
> some implies (to me) that if we did it differently, the tracer would not be
> able to set an invalid pointer.
> 
> Either way,
> 
> Reviewed-by: Catalin Marinas <catalin.marinas@arm.com>
> 
> -- 
> Catalin
>