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;
Message-ID: <01595c2fa5958253f08c07e316435abe9f32e305.camel@redhat.com>
Subject: Re: Is ->sendmsg() allowed to change the msghdr struct it is given?
From:   Paolo Abeni <pabeni@redhat.com>
To:     David Howells <dhowells@redhat.com>
Cc:     Jakub Kicinski <kuba@kernel.org>,
        Ilya Dryomov <idryomov@gmail.com>,
        "David S. Miller" <davem@davemloft.net>,
        Eric Dumazet <edumazet@google.com>, ceph-devel@vger.kernel.org,
        netdev@vger.kernel.org, linux-kernel@vger.kernel.org
Date:   Tue, 27 Jun 2023 15:51:43 +0200
In-Reply-To: <3132610.1687871361@warthog.procyon.org.uk>
References: <b0a0cb0fac4ebdc23f01d183a9de10731dc90093.camel@redhat.com>
         <3112097.1687814081@warthog.procyon.org.uk>
         <20230626142257.6e14a801@kernel.org>
         <3132610.1687871361@warthog.procyon.org.uk>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.46.4 (3.46.4-1.fc37) 
MIME-Version: 1.0
Precedence: bulk

On Tue, 2023-06-27 at 14:09 +0100, David Howells wrote:
> Paolo Abeni <pabeni@redhat.com> wrote:
>=20
> > udp_sendmsg() can set the MSG_TRUNC bit in msg->msg_flags, so I guess
> > that kind of actions are sort of allowed. Still, AFAICS, the kernel
> > based msghdr is not copied back to the user-space, so such change
> > should be almost a no-op in practice.
> >=20
> > @David: which would be the end goal for such action?
>=20
> Various places in the kernel use sock_sendmsg() - and I've added a bunch =
more
> with the MSG_SPLICE_PAGES patches.  For some of the things I've added, th=
ere's
> a loop which used to call ->sendpage() and now calls sock_sendmsg().  In =
most
> of those places, msghdr will get reset each time round the loop - but not=
 in
> all cases.
>=20
> Of particular immediate interest is net/ceph/messenger_v2.c.  If you go t=
o:
>=20
> 	https://lore.kernel.org/r/3111635.1687813501@warthog.procyon.org.uk/
>=20
> and look at the resultant code:
>=20
> 	static int do_sendmsg(struct socket *sock, struct iov_iter *it)
> 	{
> 		struct msghdr msg =3D { .msg_flags =3D CEPH_MSG_FLAGS };
> 		int ret;
>=20
> 		msg.msg_iter =3D *it;
> 		while (iov_iter_count(it)) {
> 			ret =3D sock_sendmsg(sock, &msg);
> 			if (ret <=3D 0) {
> 				if (ret =3D=3D -EAGAIN)
> 					ret =3D 0;
> 				return ret;
> 			}
>=20
> 			iov_iter_advance(it, ret);
> 		}
>=20
> 		WARN_ON(msg_data_left(&msg));
> 		return 1;
> 	}
>=20
> for example.  It could/would malfunction if sendmsg() is allowed to modif=
y
> msghdr - or if it doesn't update msg_iter.  Likewise:
>=20
> 	static int do_try_sendpage(struct socket *sock, struct iov_iter *it)
> 	{
> 		struct msghdr msg =3D { .msg_flags =3D CEPH_MSG_FLAGS };
> 		struct bio_vec bv;
> 		int ret;
>=20
> 		if (WARN_ON(!iov_iter_is_bvec(it)))
> 			return -EINVAL;
>=20
> 		while (iov_iter_count(it)) {
> 			/* iov_iter_iovec() for ITER_BVEC */
> 			bvec_set_page(&bv, it->bvec->bv_page,
> 				      min(iov_iter_count(it),
> 					  it->bvec->bv_len - it->iov_offset),
> 				      it->bvec->bv_offset + it->iov_offset);
>=20
> 			/*
> 			 * MSG_SPLICE_PAGES cannot properly handle pages with
> 			 * page_count =3D=3D 0, we need to fall back to sendmsg if
> 			 * that's the case.
> 			 *
> 			 * Same goes for slab pages: skb_can_coalesce() allows
> 			 * coalescing neighboring slab objects into a single frag
> 			 * which triggers one of hardened usercopy checks.
> 			 */
> 			if (sendpage_ok(bv.bv_page))
> 				msg.msg_flags |=3D MSG_SPLICE_PAGES;
> 			else
> 				msg.msg_flags &=3D ~MSG_SPLICE_PAGES;
>=20
> 			iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bv, 1, bv.bv_len);
> 			ret =3D sock_sendmsg(sock, &msg);
> 			if (ret <=3D 0) {
> 				if (ret =3D=3D -EAGAIN)
> 					ret =3D 0;
> 				return ret;
> 			}
>=20
> 			iov_iter_advance(it, ret);
> 		}
>=20
> 		return 1;
> 	}
>=20
> could be similarly affected if ->sendmsg() mucks about with msg_flags.

With some help from the compiler - locally changing the proto_ops and
proto sendmsg definition and handling the fallout - I found the
following:

- mptcp_sendmsg() is clearing unsupported msg_flags=20
  (I should have recalled this one even without much testing ;)

- udpv4_sendmsg() is setting msg_name/msg_namelen
- tls_device_sendmsg() is clearing MSG_SPLICE_PAGE when zerocopy is not
supported
- unix_seqpacket_sendmsg() is clearing msg_namelen

I could have missed something, but the above looks safe for the use-
case you mentioned.

Cheers,

Paolo