Re: [PATCH 5.2 36/37] vhost: block speculation of translated descriptors

[Thomas Backlund]

Den 14-09-2019 kl. 11:08, skrev Greg Kroah-Hartman:
> On Sat, Sep 14, 2019 at 09:15:48AM +0200, Stefan Lippers-Hollmann wrote:
> > Hi
> >
> > On 2019-09-14, Greg Kroah-Hartman wrote:
> > > On Sat, Sep 14, 2019 at 02:54:11AM +0200, Stefan Lippers-Hollmann wrote:
> > > > On 2019-09-13, Greg Kroah-Hartman wrote:
> > > > > From: Michael S. Tsirkin <mst@xxxxxxxxxx>
> > > > >
> > > > > commit a89db445fbd7f1f8457b03759aa7343fa530ef6b upstream.
> > > > >
> > > > > iovec addresses coming from vhost are assumed to be
> > > > > pre-validated, but in fact can be speculated to a value
> > > > > out of range.
> > > > >
> > > > > Userspace address are later validated with array_index_nospec so we can
> > > > > be sure kernel info does not leak through these addresses, but vhost
> > > > > must also not leak userspace info outside the allowed memory table to
> > > > > guests.
> > > > >
> > > > > Following the defence in depth principle, make sure
> > > > > the address is not validated out of node range.
> > [...]
> > > Do you have the same problem with Linus's tree right now?
> >
> > Actually, yes I do (I had not tested i386 for 5.3~ within the last ~2
> > weeks, only amd64). Very similar kernel config, same compiler versions
> > but built in a slightly different environment (built directly on the bare
> > iron, in a amd64 multilib userspace, rather than a pure-i386 chroot on an
> > amd64 kernel).
> >
> > $ git describe
> > v5.3-rc8-36-ga7f89616b737
> >
> > $ LANG= make ARCH=x86 -j1 bzImage modules
> >    CALL    scripts/checksyscalls.sh
> >    CALL    scripts/atomic/check-atomics.sh
> >    CHK     include/generated/compile.h
> >    CHK     kernel/kheaders_data.tar.xz
> >    CC [M]  drivers/vhost/vhost.o
> > In file included from ./include/linux/export.h:45,
> >                   from ./include/linux/linkage.h:7,
> >                   from ./include/linux/kernel.h:8,
> >                   from ./include/linux/list.h:9,
> >                   from ./include/linux/wait.h:7,
> >                   from ./include/linux/eventfd.h:13,
> >                   from drivers/vhost/vhost.c:13:
> > drivers/vhost/vhost.c: In function 'translate_desc':
> > ./include/linux/compiler.h:350:38: error: call to '__compiletime_assert_2076' declared with attribute error: BUILD_BUG_ON failed: sizeof(_s) > sizeof(long)
> >    350 |  _compiletime_assert(condition, msg, __compiletime_assert_, __LINE__)
> >        |                                      ^
> > ./include/linux/compiler.h:331:4: note: in definition of macro '__compiletime_assert'
> >    331 |    prefix ## suffix();    \
> >        |    ^~~~~~
> > ./include/linux/compiler.h:350:2: note: in expansion of macro '_compiletime_assert'
> >    350 |  _compiletime_assert(condition, msg, __compiletime_assert_, __LINE__)
> >        |  ^~~~~~~~~~~~~~~~~~~
> > ./include/linux/build_bug.h:39:37: note: in expansion of macro 'compiletime_assert'
> >     39 | #define BUILD_BUG_ON_MSG(cond, msg) compiletime_assert(!(cond), msg)
> >        |                                     ^~~~~~~~~~~~~~~~~~
> > ./include/linux/build_bug.h:50:2: note: in expansion of macro 'BUILD_BUG_ON_MSG'
> >     50 |  BUILD_BUG_ON_MSG(condition, "BUILD_BUG_ON failed: " #condition)
> >        |  ^~~~~~~~~~~~~~~~
> > ./include/linux/nospec.h:56:2: note: in expansion of macro 'BUILD_BUG_ON'
> >     56 |  BUILD_BUG_ON(sizeof(_s) > sizeof(long));   \
> >        |  ^~~~~~~~~~~~
> > drivers/vhost/vhost.c:2076:5: note: in expansion of macro 'array_index_nospec'
> >   2076 |     array_index_nospec((unsigned long)(addr - node->start),
> >        |     ^~~~~~~~~~~~~~~~~~
> > make[2]: *** [scripts/Makefile.build:281: drivers/vhost/vhost.o] Error 1
> > make[1]: *** [scripts/Makefile.build:497: drivers/vhost] Error 2
> > make: *** [Makefile:1083: drivers] Error 2
> >
> > $ git revert a89db445fbd7f1f8457b03759aa7343fa530ef6b
> >
> > $ LANG= make ARCH=x86 -j16 bzImage modules
> >    CALL    scripts/atomic/check-atomics.sh
> >    CALL    scripts/checksyscalls.sh
> >    CHK     include/generated/compile.h
> >    CHK     kernel/kheaders_data.tar.xz
> >    Building modules, stage 2.
> > Kernel: arch/x86/boot/bzImage is ready  (#1)
> >    MODPOST 3464 modules
> >
> > $ echo $?
> > 0
> >
> > $ find . -name vhost\\.ko
> > ./drivers/vhost/vhost.ko
> >
> > I've attached the affected kernel config for v5.3~/ i386.
>
> Ok, good, we will be "bug compatible" at the very least now :)
>
> When this gets fixed in Linus's tree we can backport the fix here as
> well.  The number of users of that compiler version/configuration is
> probably pretty low at the moment to want to hold off on this fix.

This is now reverted upstream:

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=0d4a3f2abbef73b9e5bb5f12213c275565473588

--
Thomas