Hi, Keir,
Do you mean that in a multi-thread process, one thread issues an I/O
operation, and in the time slot that just after the processor has fetched the
instruction, validated the access, but before Xen re-fetches the instruction
for emulation, another thread steals that I/O instruction and replace it with a
new one? Maybe we can regard it as a kind of attack...
This could be happen in theory, but I think other instruction emulation may
also have this problem. In your last sentence, do you mean that we still need
to do an entire I/O permission check (including CPL, IOPL, and TSS I/O bitmap)
in x86_emulate() for safety consideration? Thanks! :-)
Best regards,
-- Dongxiao
-----Original Message-----
From: Keir Fraser [mailto:keir.fraser@xxxxxxxxxxxxx]
Sent: 2008年3月18日 16:46
To: Xu, Dongxiao; Cui, Dexuan; xen-devel@xxxxxxxxxxxxxxxxxxx
Subject: Re: [Xen-devel] [PATCH]Fix the bug of guest os installationfailure and
win2k boot failure
We're on the same page now, except that I realised there is a TOCTTOU race
introduced by relying on the processor's permission check while re-fetching
the instruction from scratch in the hypervisor. This allows, in theory, a
multi-threaded process to rewrite the I/O-port accessing instruction after
the processor has fetched the instruction, and validated the access, but
before Xen re-fetches the instruction for emulation. Possibly we do not care
too much about this, since the process must already have some
I/O-port-access permissions, but equally I don't expect we fall into the
TSS-bitmap check all that often, it's not that hard to implement, and then
we are definitely safe.
-- Keir
On 18/3/08 01:49, "Xu, Dongxiao" <dongxiao.xu@xxxxxxxxx> wrote:
> Hi, Keir,
> Now I understand what you mean. read_io, write_io, inject_hw_exception
> callbacks are not defined within the multi.c. So I/O instructions will not be
> emulated by it. Thanks for your comments. And the new patch is attached.
>
> Best regards,
> -- Dongxiao
>
> -----Original Message-----
> From: Keir Fraser [mailto:keir.fraser@xxxxxxxxxxxxx]
> Sent: 2008年3月17日 19:21
> To: Cui, Dexuan; Xu, Dongxiao; xen-devel@xxxxxxxxxxxxxxxxxxx
> Subject: Re: [Xen-devel] [PATCH]Fix the bug of guest os installationfailure
> and win2k boot failure
>
> On 17/3/08 11:16, "Cui, Dexuan" <dexuan.cui@xxxxxxxxx> wrote:
>
>>> I think you misunderstand. The shadow emulator *never* emulates I/O
>>> port accesses or exception deliveries. Those callback functions are
>>> simply not implemented and are left as NULL.
>> Those callback functions -- what are they? -- do you mean the following?
>> static struct x86_emulate_ops hvm_emulate_ops = {
>> ....
>> .read_io = hvmemul_read_io,
>> .write_io = hvmemul_write_io,
>> ...
>> };
>
> Yes indeed. Also, crucially, .inject_hw_exception. Without that
> x86_emulate() is unable to inject any exception into the guest, and will
> instead return X86EMUL_UNHANDLEABLE.
>
> -- Keir
>
>
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel
|
Home