From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:43812) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1dPTST-0006QP-5L for qemu-devel@nongnu.org; Mon, 26 Jun 2017 08:49:40 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1dPTSP-0006ve-EJ for qemu-devel@nongnu.org; Mon, 26 Jun 2017 08:49:37 -0400 Received: from mx1.redhat.com ([209.132.183.28]:50742) by eggs.gnu.org with esmtps (TLS1.0:DHE_RSA_AES_256_CBC_SHA1:32) (Exim 4.71) (envelope-from ) id 1dPTSP-0006vL-5B for qemu-devel@nongnu.org; Mon, 26 Jun 2017 08:49:33 -0400 Date: Mon, 26 Jun 2017 13:49:22 +0100 From: "Daniel P. Berrange" Message-ID: <20170626124922.GO495@redhat.com> Reply-To: "Daniel P. Berrange" References: <51d7f54d100e9dedecf6dc65691ca65adfc8394f.1498213152.git-series.knut.omang@oracle.com> <20170626102254.GG495@redhat.com> <1498480368.3341.43.camel@oracle.com> MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Disposition: inline In-Reply-To: <1498480368.3341.43.camel@oracle.com> Content-Transfer-Encoding: quoted-printable Subject: Re: [Qemu-devel] [PATCH v4 4/4] sockets: Handle race condition between binds to the same port List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Knut Omang Cc: Gerd Hoffmann , Paolo Bonzini , qemu-devel@nongnu.org On Mon, Jun 26, 2017 at 02:32:48PM +0200, Knut Omang wrote: > On Mon, 2017-06-26 at 11:22 +0100, Daniel P. Berrange wrote: > > On Fri, Jun 23, 2017 at 12:31:08PM +0200, Knut Omang wrote: > > > If an offset of ports is specified to the inet_listen_saddr functio= n(), > > > and two or more processes tries to bind from these ports at the sam= e time, > > > occasionally more than one process may be able to bind to the same > > > port. The condition is detected by listen() but too late to avoid a= failure. > > >=C2=A0 > > > This function is called by socket_listen() and used > > > by all socket listening code in QEMU, so all cases where any form o= f dynamic > > > port selection is used should be subject to this issue. > > >=C2=A0 > > > Add code to close and re-establish the socket when this > > > condition is observed, hiding the race condition from the user. > > >=C2=A0 > > > This has been developed and tested by means of the > > > test-listen unit test in the previous commit. > > > Enable the test for make check now that it passes. > > >=C2=A0 > > > Signed-off-by: Knut Omang > > > Reviewed-by: Bhavesh Davda > > > Reviewed-by: Yuval Shaia > > > Reviewed-by: Girish Moodalbail > > > --- > > >=C2=A0=C2=A0tests/Makefile.include |=C2=A0=C2=A02 +- > > >=C2=A0=C2=A0util/qemu-sockets.c=C2=A0=C2=A0=C2=A0=C2=A0| 68 ++++++++= ++++++++++++++++++++++++----------- > > >=C2=A0=C2=A02 files changed, 53 insertions(+), 17 deletions(-) > > >=C2=A0 > > > diff --git a/tests/Makefile.include b/tests/Makefile.include > > > index 22bb97e..c38f94e 100644 > > > --- a/tests/Makefile.include > > > +++ b/tests/Makefile.include > > > @@ -127,7 +127,7 @@ check-unit-y +=3D tests/test-bufferiszero$(EXES= UF) > > >=C2=A0=C2=A0gcov-files-check-bufferiszero-y =3D util/bufferiszero.c > > >=C2=A0=C2=A0check-unit-y +=3D tests/test-uuid$(EXESUF) > > >=C2=A0=C2=A0check-unit-y +=3D tests/ptimer-test$(EXESUF) > > > -#check-unit-y +=3D tests/test-listen$(EXESUF) > > > +check-unit-y +=3D tests/test-listen$(EXESUF) > > >=C2=A0=C2=A0gcov-files-ptimer-test-y =3D hw/core/ptimer.c > > >=C2=A0=C2=A0check-unit-y +=3D tests/test-qapi-util$(EXESUF) > > >=C2=A0=C2=A0gcov-files-test-qapi-util-y =3D qapi/qapi-util.c > > > diff --git a/util/qemu-sockets.c b/util/qemu-sockets.c > > > index 48b9319..7b118b4 100644 > > > --- a/util/qemu-sockets.c > > > +++ b/util/qemu-sockets.c > > > @@ -201,6 +201,42 @@ static int try_bind(int socket, InetSocketAddr= ess *saddr, struct > > addrinfo *e) > > >=C2=A0=C2=A0#endif > > >=C2=A0=C2=A0} > > >=C2=A0=C2=A0 > > > +static int try_bind_listen(int *socket, InetSocketAddress *saddr, > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2= =A0=C2=A0=C2=A0=C2=A0struct addrinfo *e, int port, Error **errp) > > > +{ > > > +=C2=A0=C2=A0=C2=A0=C2=A0int s =3D *socket; > > > +=C2=A0=C2=A0=C2=A0=C2=A0int ret; > > > + > > > +=C2=A0=C2=A0=C2=A0=C2=A0inet_setport(e, port); > > > +=C2=A0=C2=A0=C2=A0=C2=A0ret =3D try_bind(s, saddr, e); > > > +=C2=A0=C2=A0=C2=A0=C2=A0if (ret) { > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0if (errno !=3D EAD= DRINUSE) { > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0error_setg_errno(errp, errno, "Failed to bind socket"); > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0} > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0return errno; > > > +=C2=A0=C2=A0=C2=A0=C2=A0} > > > +=C2=A0=C2=A0=C2=A0=C2=A0if (listen(s, 1) =3D=3D 0) { > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0return 0; > > > +=C2=A0=C2=A0=C2=A0=C2=A0} > > > +=C2=A0=C2=A0=C2=A0=C2=A0if (errno =3D=3D EADDRINUSE) { > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0/* We got to bind = the socket to a port but someone else managed > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0* to bind to= the same port and beat us to listen on it! > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0* Recreate t= he socket and return EADDRINUSE to preserve the > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0* expected s= tate by the caller: > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0*/ > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0closesocket(s); > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0s =3D create_fast_= reuse_socket(e, errp); > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0if (s < 0) { > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0return errno; > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0} > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0*socket =3D s; > >=20 > > I don't really like this at all - if we need to close + recreate the > > socket, IMHO that should remain the job of the caller, since it owns > > the socket FD ultimately. >=20 > Normally I would agree, but this is a very unlikely situation.=C2=A0I c= onsidered moving the > complexity out to the caller, even to recreate for every call, but foun= d those solutions > to be inferior as they do not in any way confine the problem, and cause= the handling of > the common cases to be much less readable. It's going to be some trade-= offs here. >=20 > As long as the caller is aware of (by the reference call) that the sock= et in use may > change, this is in my view a clean (as clean as possible) abstraction t= hat simplifies the > logic at the next level. My intention is to make the common, good case = as readable as > possible and hide some of the complexity of these=20 > unlikely error scenarios inside the new functions - divide and conquer.= . >=20 > >=20 > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0errno =3D EADDRINU= SE; > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0return errno; > > > +=C2=A0=C2=A0=C2=A0=C2=A0} > > > +=C2=A0=C2=A0=C2=A0=C2=A0error_setg_errno(errp, errno, "Failed to l= isten on socket"); > > > +=C2=A0=C2=A0=C2=A0=C2=A0return errno; > > > +} > > > + > > >=C2=A0=C2=A0static int inet_listen_saddr(InetSocketAddress *saddr, > > >=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2= =A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0int port_offset, > > >=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2= =A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0bool update_addr, > > > @@ -210,7 +246,9 @@ static int inet_listen_saddr(InetSocketAddress = *saddr, > > >=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0char port[33]; > > >=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0char uaddr[INET6_ADDRSTRLEN+1]; > > >=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0char uport[33]; > > > -=C2=A0=C2=A0=C2=A0=C2=A0int slisten, rc, port_min, port_max, p; > > > +=C2=A0=C2=A0=C2=A0=C2=A0int rc, port_min, port_max, p; > > > +=C2=A0=C2=A0=C2=A0=C2=A0int slisten =3D 0; > > > +=C2=A0=C2=A0=C2=A0=C2=A0int saved_errno =3D 0; > > >=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0Error *err =3D NULL; > > >=C2=A0=C2=A0 > > >=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0memset(&ai,0, sizeof(ai)); > > > @@ -276,28 +314,26 @@ static int inet_listen_saddr(InetSocketAddres= s *saddr, > >=20 > > Just above this line is the original 'create_fast_reuse_socket' call. > >=20 > > I'd suggest that we push that call down into the body of the loop > > below: > >=20 > > >=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0port_min= =3D inet_getport(e); > > >=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0port_max= =3D saddr->has_to ? saddr->to + port_offset : port_min; > > >=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0for (p =3D= port_min; p <=3D port_max; p++) { > > > -=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0inet_setport(e, p); > > > -=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0if (try_bind(slisten, saddr, e) >=3D 0) { > > > -=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0goto listen; > > > -=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0} > > > -=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0if (p =3D=3D port_max) { > > > -=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0if (!e->ai_next) { > > > -=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0error_setg_errno(er= rp, errno, "Failed to bind socket"); > > > -=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0} > > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0= =C2=A0int eno =3D try_bind_listen(&slisten, saddr, e, p, &err); > >=20 > > Which would mean try_bind_listen no longer needs the magic to close + > > recreate the socket. > >=20 > > The only cost of doing this is that you end up closing + recreating t= he > > socket after bind hits EADDRINUSE, as well as after listen() hits it. >=20 > The problem with this approach in my opinion is that one has to underst= and the > fix for the problem I am trying to solve here in order to read the main= code,=C2=A0 > even though this is a very special case. Everyone reading the code woul= d ask themselves > the question 'why do they recreate the socket here?' and then be forced= to ready the > details of try_bind_listen anyway, or we would need additional comments= . That's easily solved by adding a comment /* We recreate the socket FD on each iteration because if bind succeeds & listen fails, we can't bind again on the same socket FD */ > The idea behind the abstractions I have used here is to hide the detail= s inside functions, > but leave them with an as clean as possible (although not ideal) interf= ace that=20 > makes the overall logic more readable. I think the result is actually harder to understand, because of the peculiar way the function closes & reopens the socket FD belonging to the caller, and the error handling is really very unclear and buggy as a result too. > > I think that's acceptable tradeoff for simpler code, since this is no= t > > a performance critical operation. >=20 > Also should we perhaps worry about any side effects of creating and clo= sing a lot of > sockets unnecessary? What side effects ? I don't think there are any - since this is server side, not client side, we're not leaving any state around in timed waits or similar. Regards, Daniel --=20 |: https://berrange.com -o- https://www.flickr.com/photos/dberran= ge :| |: https://libvirt.org -o- https://fstop138.berrange.c= om :| |: https://entangle-photo.org -o- https://www.instagram.com/dberran= ge :|