Received-SPF: pass (google.com: domain of linux-crypto+bounces-1903-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45d1:ec00::1 as permitted sender) client-ip=2604:1380:45d1:ec00::1;
Date: Wed, 7 Feb 2024 20:36:10 -0800
From: Eric Biggers <ebiggers@kernel.org>
To: Mikulas Patocka <mpatocka@redhat.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>,
	"David S. Miller" <davem@davemloft.net>,
	linux-crypto@vger.kernel.org, dm-devel@lists.linux.dev
Subject: Re: A question about modifying the buffer under authenticated
 encryption
Message-ID: <20240208043610.GB85799@sol.localdomain>
References: <f22dae2c-9cac-a63-fff-3b0b7305be6@redhat.com>
 <20240207004723.GA35324@sol.localdomain>
 <1a4713fc-62c7-4a8f-e28a-14fc5d04977@redhat.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1a4713fc-62c7-4a8f-e28a-14fc5d04977@redhat.com>

On Wed, Feb 07, 2024 at 01:51:51PM +0100, Mikulas Patocka wrote:
> 
> 
> On Tue, 6 Feb 2024, Eric Biggers wrote:
> 
> > On Tue, Feb 06, 2024 at 10:46:59PM +0100, Mikulas Patocka wrote:
> > > Hi
> > > 
> > > I'm trying to fix some problems in dm-crypt that it may report 
> > > authentication failures when the user reads data with O_DIRECT and 
> > > modifies the read buffer while it is being read.
> > > 
> > > I'd like to ask you:
> > > 
> > > 1. If the authenticated encryption encrypts a message, reading from 
> > >    buffer1 and writing to buffer2 - and buffer1 changes while reading from 
> > >    it - is it possible that it generates invalid authentication tag?
> > > 
> > > 2. If the authenticated encryption decrypts a message, reading from 
> > >    buffer1 and writing to buffer2 - and buffer2 changes while writing to 
> > >    it - is is possible that it reports authentication tag mismatch?
> > > 
> > 
> > Yes, both scenarios are possible.  But it depends on the AEAD algorithm and how
> > it happens to be implemented, and on whether the data overlaps or not.
> > 
> > This is very much a "don't do that" sort of thing.
> > 
> > - Eric
> 
> I see. So I will copy the data to a kernel buffer before encryption or 
> decryption.
> 
> I assume that authenticated encryption or decryption using the same buffer 
> as a source and as a destination should be ok. Right?
> 

The crypto_aead API allows the source and destination to overlap, yes.

- Eric