Re: [PATCH RFC] seccomp: Implement syscall isolation based on memory areas

From: Matthew Wilcox
Date: Sun May 31 2020 - 13:32:03 EST

On Sun, May 31, 2020 at 08:10:18PM +0300, Paul Gofman wrote:
> On 5/31/20 19:49, Matthew Wilcox wrote:
> > On Sun, May 31, 2020 at 03:39:33PM +0300, Paul Gofman wrote:
> >>> Paul (cc'ed) is the wine expert, but my understanding is that memory
> >>> allocation and initial program load of the emulated binary will go
> >>> through wine. It does the allocation and mark the vma accordingly
> >>> before returning the allocated range to the windows application.
> >> Yes, exactly. Pretty much any memory allocation which Wine does needs
> >> syscalls (if those are ever encountered later during executing code from
> >> those areas) to be trapped by Wine and passed to Wine's implementation
> >> of the corresponding Windows API function. Linux native libraries
> >> loading and memory allocations performed by them go outside of Wine control.
> > I don't like Gabriel's approach very much. Could we do something like
> > issue a syscall before executing a Windows region and then issue another
> > syscall when exiting? If so, we could switch the syscall entry point (ie
> > change MSR_LSTAR). I'm thinking something like a personality() syscall.
> > But maybe that would be too high an overhead.
> IIRC Gabriel had such idea that we discussed. We can potentially track
> the boundary between the Windows and native code exectution. But issuing
> syscall every time we cross that boundary may have a prohibitive
> performance impact, that happens way too often. What we could do is to
> put the flag somewhere, but that flag has to be per thread. E. g., we
> could use Linux gs: based thread local storage, or fs: based address
> (that's what Windows using for thread local data and thus Wine maintains
> also). If Seccomp filters could access such a memory location (fetch a
> byte from there and put into the structure accessible by BPF_LD) we
> could use SECCOMP_MODE_FILTER, I think.

If it's the cost of the syscall that's the problem, there are ways
around that. We'd still want a personality() call to indicate that
the syscall handler should look (somewhere) to determine the current
personality, but that could be issued at the start of execution rather
than when we switch between Windows & Linux code.