Re: [PATCH v1] linux-user: Add option to run `execve`d programs through QEMU

[Ilya Leoshkevich <iii@linux.ibm.com>]

On Wed, 2024-10-02 at 10:10 -0500, Noah Goldstein wrote:
> On Wed, Oct 2, 2024 at 9:53 AM Ilya Leoshkevich <iii@linux.ibm.com>
> wrote:
> > 
> > On Wed, 2024-10-02 at 09:44 -0500, Noah Goldstein wrote:
> > > On Wed, Oct 2, 2024 at 9:38 AM Ilya Leoshkevich
> > > <iii@linux.ibm.com>
> > > wrote:
> > > > 
> > > > On Wed, 2024-10-02 at 16:08 +0200, Laurent Vivier wrote:
> > > > > Le 02/10/2024 à 10:08, Ilya Leoshkevich a écrit :
> > > > > > On Fri, 2024-08-30 at 15:36 -0700, Noah Goldstein wrote:
> > > > > > > The new option '-qemu-children' makes it so that on
> > > > > > > `execve`
> > > > > > > the
> > > > > > > child
> > > > > > > process will be launch by the same `qemu` executable that
> > > > > > > is
> > > > > > > currently
> > > > > > > running along with its current commandline arguments.
> > > > > > > 
> > > > > > > The motivation for the change is to make it so that
> > > > > > > plugins
> > > > > > > running
> > > > > > > through `qemu` can continue to run on children.  Why not
> > > > > > > just
> > > > > > > `binfmt`?: Plugins can be desirable regardless of
> > > > > > > system/architecture
> > > > > > > emulation, and can sometimes be useful for elf files that
> > > > > > > can
> > > > > > > run
> > > > > > > natively. Enabling `binfmt` for all natively runnable elf
> > > > > > > files
> > > > > > > may
> > > > > > > not be desirable.
> > > > > > 
> > > > > > Another reason to have this is that one may not have root
> > > > > > permissions
> > > > > > to configure binfmt-misc.
> > > > > 
> > > > > A little note on that: binfmt_misc is now part of the user
> > > > > namespace
> > > > > (since linux v6.7), so you can
> > > > > configure binfmt_misc as a non root user in a given namepace.
> > > > > 
> > > > > There is helper to use it with unshare from util-linux, you
> > > > > can
> > > > > do
> > > > > things like that:
> > > > > 
> > > > >    With 'F' flag, load the interpreter from the initial
> > > > > namespace:
> > > > > 
> > > > >      $ /bin/qemu-m68k-static --version
> > > > >      qemu-m68k version 8.2.2 (qemu-8.2.2-1.fc40)
> > > > >      Copyright (c) 2003-2023 Fabrice Bellard and the QEMU
> > > > > Project
> > > > > developers
> > > > >      $ unshare --map-root-user --fork --pid
> > > > > --load-interp=":qemu-
> > > > > m68k:M::\\x7fELF\\x01\\x02\\x01\\x00\\x00\\x00\\x00\\x00\\x00
> > > > > \\x0
> > > > > 0\\x
> > > > > 00\\x00\\x00\\x02\\x00\\x04:\\xff\\xff\\xff\\xff\\xff\\xff\\x
> > > > > fe\\
> > > > > x00\
> > > > > \xff\\xff\\xff\\xff\\xff\\xff\\xff\\xff\\xff\\xfe\\xff\\xff:/
> > > > > bin/
> > > > > qemu
> > > > > -m68k-static:OCF"
> > > > > --root=chroot/m68k/sid
> > > > >      # QEMU_VERSION= ls
> > > > >      qemu-m68k version 8.2.2 (qemu-8.2.2-1.fc40)
> > > > >      Copyright (c) 2003-2023 Fabrice Bellard and the QEMU
> > > > > Project
> > > > > developers
> > > > >      # /qemu-m68k  --version
> > > > >      qemu-m68k version 8.0.50 (v8.0.0-340-gb1cff5e2da95)
> > > > >      Copyright (c) 2003-2022 Fabrice Bellard and the QEMU
> > > > > Project
> > > > > developers
> > > > > 
> > > > >    Without 'F' flag, from inside the namespace:
> > > > > 
> > > > >      $ unshare --map-root-user --fork --pid
> > > > > --load-interp=":qemu-
> > > > > m68k:M::\\x7fELF\\x01\\x02\\x01\\x00\\x00\\x00\\x00\\x00\\x00
> > > > > \\x0
> > > > > 0\\x
> > > > > 00\\x00\\x00\\x02\\x00\\x04:\\xff\\xff\\xff\\xff\\xff\\xff\\x
> > > > > fe\\
> > > > > x00\
> > > > > \xff\\xff\\xff\\xff\\xff\\xff\\xff\\xff\\xff\\xfe\\xff\\xff:/
> > > > > qemu
> > > > > -
> > > > > m68k:OC"
> > > > > --root=chroot/m68k/sid
> > > > >      # QEMU_VERSION= ls
> > > > >      qemu-m68k version 8.0.50 (v8.0.0-340-gb1cff5e2da95)
> > > > >      Copyright (c) 2003-2022 Fabrice Bellard and the QEMU
> > > > > Project
> > > > > developers
> > > > >      # /qemu-m68k  --version
> > > > >      qemu-m68k version 8.0.50 (v8.0.0-340-gb1cff5e2da95)
> > > > >      Copyright (c) 2003-2022 Fabrice Bellard and the QEMU
> > > > > Project
> > > > > developers
> > > > > 
> > > > > Thanks,
> > > > > Laurent
> > > > > 
> > > > 
> > > > Thanks for posting this, I wasn't aware of this feature and it
> > > > looks
> > > > really useful.
> > > > 
> > > > IIUC it also resolves the main problem this patch is dealing
> > > > with:
> > > 
> > > I might misunderstand, but I don't think it does in the sense
> > > that it still might not be desirable to use the same qemu flags
> > > for the entire class of executables.
> > > 
> > > I.e the original motivating case was wanting to attach
> > > some plugins to a process and its children and AFAICT
> > > binfmt still doesn't give that level of control.
> > 
> > I think if you start a process in a user namespace, which has a
> > binfmt_misc handler for a certain class of binaries, then this
> > handler
> > will affect only this process and its children, and not the rest of
> > the
> > system.
> 
> It won't also affect other binaries in the user namespace?

It would, but you should be able to create a user namespace just
for your program. It should also be possible to nest user namespaces.

> > > >   Enabling `binfmt` for all natively runnable elf files may
> > > >   not be desirable.