RE: [PATCH v2 1/1] vhost: tolerate file descriptor in REM_MEM_REG msg

Internal Use - Confidential
> -----Original Message-----
> From: David Marchand <[email protected]>
> Sent: Thursday, July 30, 2026 11:22 PM
> To: Bathija, Pravin <[email protected]>; stephen=
AEA-networkplumber.org
> Cc: [email protected]; [email protected];
> [email protected]; [email protected]; thomas@=
monjalon.net
> Subject: Re: [PATCH v2 1/1] vhost: tolerate file descriptor i=
n REM_MEM_REG
> msg
>
>
> [EXTERNAL EMAIL]
>
> On Fri, 31 Jul 2026 at 05:14, <[email protected]> w=
rote:
> >
> > From: Pravin M Bathija <[email protected]>
> >
> > The vhost-user specification (vhost-user.rst) states that no fi=
le
> > descriptors SHOULD be passed with VHOST_USER_REM_ME=
M_REG.
> However, it
> > also says: "For compatibility with existing incorrect imple=
mentations,
> > the back-end MAY accept messages with one file descriptor.  If =
a file
> > descriptor is passed, the back-end MUST close it without using =
it
> > otherwise."
> >
> > Some front-ends, notably libblkio, reuse the same message-build=
ing
> > helper for both ADD_MEM_REG and REM_MEM_REG and=
 unconditionally
> attach
> > the mapping fd.  The previous implementation rejected any REM
F8-MEM_REG
> > carrying a file descriptor with the error:
> >
> >         expect 0 FDs for request VHOST_USER_REM_MEM=
_REG, received 1
> >
> > This broke teardown and memory region hot-swap with these front=
-ends.
> >
> > To reproduce, run any libblkio (v1.5.0) application using the
> > virtio-blk-vhost-user driver against a DPDK vhost back-end.  Th=
e
> > connection is dropped during cleanup or whenever a memory regio=
n is
> > unmapped and remapped.
>
> Is libblkio fixed now?
>
> I am not a fan of such compatibility fix, having to accept one buggy =
client...
>

Yes, the libblkio fix is ready and will be submitted upstream within a day =
or so.
It stops sending the fd with REM_MEM_REG.

That said, this DPDK fix stands on its own regardless of libblkio. The vhos=
t-user
specification explicitly anticipates this situation and requires back-ends =
to handle it:

"For compatibility with existing incorrect implementations, the back-en=
d MAY accept messages
with one file descriptor. If a file descriptor is passed, the back-end MUST=
 close it without
using it otherwise."

While the immediate motivation was libblkio, the spec's compatibility claus=
e was written for
exactly this situation — any front-end could make the same mistake. QEM=
U's libvhost-user
reference implementation already tolerates it (see vu_rem_mem_r=
eg).

The fix is 2 lines with no downside: accept the message, close the fd. Reje=
cting it drops the
connection entirely, which is a disproportionate response to a harmless ext=
ra fd.

> >
> > QEMU's libvhost-user reference back-end (vu_rem_mem_=
-reg) already
> > tolerates zero or one fd in this message.  Align DPDK's behavio=
r with
> > both the specification's compatibility clause and the reference
> > implementation by accepting the message and closing any unexpec=
ted fd.
> >
> > Tested with:
> >  - QEMU VM bring-up with runtime add/remove memory regions via
> >    QEMU monitor
> >  - QEMU post-copy live migration between source and destination
> >  - SPDK vhost-blk with libblkio (fio libblkio engine, write + =
md5
> >    verify)
> >  - libblkio alloc-mem-region and map-mem-region tests exercisin=
g
> >    ADD_MEM_REG / REM_MEM_REG / ADD_MEM_=
REG cycles against SPDK
> >    vhost-blk
> >
> > Fixes: 1d730eea6a42 ("vhost: add memory region handlers"=
-)
> > Cc: [email protected]
> >
> > Signed-off-by: Pravin M Bathija <[email protected]=
m>
> > ---
> >  doc/guides/rel_notes/release_26_07.rst | 8 +=
-++++++
> >  lib/vhost/vhost_user.c                 | 9 +++++=
-++-
> >  2 files changed, 16 insertions(+), 1 deletion(-)
> >
> > diff --git a/doc/guides/rel_notes/release_26_07.rst
> > b/doc/guides/rel_notes/release_26_07.rst
> > index 3d18ba2dd0..86042bd5e3 100644
> > --- a/doc/guides/rel_notes/release_26_07.rst
> > +++ b/doc/guides/rel_notes/release_26_07.rst
>
> We are working on 26.*11* ...
>
>
> > @@ -85,6 +85,14 @@ New Features
> >    to support adding and removing memory regions without resett=
ing
> >    the whole guest memory map.
> >
> > +* **Fixed vhost REM_MEM_REG compatibility w=
ith front-ends sending
> > +file descriptors.**
> > +
> > +  The vhost-user ``REM_MEM_REG`` message =
handler now tolerates  an
> > + unexpected file descriptor sent by the front-end,  closing i=
t
> > + without use as permitted by the vhost-user specification.
> > +  This fixes interoperability with front-ends such as libblki=
o  that
> > + unconditionally attach a file descriptor to the message.
> > +
>
> ... and regardless of the version this was put in, this is not a feat=
ure.
>
> We usually don't put Fixes: in RN.
> The AI bot nags too much.
>
>
> --
> David Marchand