Xen devel - Jan 2008 - Mapping another domain''s user memory

Hello,

Is it possible to map another domain''s user-level memory into a  
different domain''s kernel?  (For clarity, let the source domain be the
one with the user-space pages that a destination domain wants to write  
to.)

I tried to do that by first getting a grant for the user-space memory  
in the source domain as follows:

	page_start = user_buf;
	buffer_mfn = virt_to_mfn(page_start);
	gnttab_grant_foreign_access_ref(grant_ref, info->dev->otherend_id,  
buffer_mfn, 0);

But the virt_to_mfn call asserted, in include/asm-x86_64/mach-xen/asm/ 
maddr.h:pfn_to_mfn.
The assertion that failed was: BUG_ON(end_pfn && pfn >= end_pfn);

So a few questions:
1) is it even possible to map a domain''s user-space page into another  
domain?
2) for the above code snippet to work, must I install some mappings in  
the source domain first?
3) does the blktap driver do something similar? I tried looking at the  
blktap driver, because it looked like it might help.  But I''m new to  
the Xen memory management code so it was a bit difficult to follow.
4) I looked at fs/filemap.c:file_read_actor and how it calls kmap  
before calling __copy_to_user.  Would calling kmap help?   
__copy_to_user is a bit of a mystery to me (granted, no pun intended,  
I didn''t dig deep enough into the assembly code to understand what  
it''s doing.)
5) any other suggestions?

Thanks,
Mike

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On Thu, 2008-01-31 at 09:57 -0500, Michael Abd-El-Malek
wrote:
> Hello,
> 
> Is it possible to map another domain''s user-level memory into a  
> different domain''s kernel?  (For clarity, let the source domain be
the
> one with the user-space pages that a destination domain wants to write  
> to.)
> 
> I tried to do that by first getting a grant for the user-space memory  
> in the source domain as follows:
> 
> 	page_start = user_buf;
> 	buffer_mfn = virt_to_mfn(page_start);
> 	gnttab_grant_foreign_access_ref(grant_ref, info->dev->otherend_id,  
> buffer_mfn, 0);
this functions wants a guest page frame number, not the mfn, e.g. via
page_to_pfn().
> But the virt_to_mfn call asserted, in include/asm-x86_64/mach-xen/asm/ 
> maddr.h:pfn_to_mfn.
> The assertion that failed was: BUG_ON(end_pfn && pfn >=
end_pfn);
> 
> So a few questions:
> 1) is it even possible to map a domain''s user-space page into
another
> domain?
yes, any, by dom0, or else via grants.
> 2) for the above code snippet to work, must I install some mappings in  
> the source domain first?
no, afaik it does not have to be mapped, just owned.
> 3) does the blktap driver do something similar? I tried looking at the  
> blktap driver, because it looked like it might help.  But I''m new
to
> the Xen memory management code so it was a bit difficult to follow.
not sure about blktap, but {blk|net}front should give you examples of
how the above code is supposed to be used. grants are typically created
by the frontend, as the backend drivers won''t dedicate memory of the
driver domain to that purpose.

regards,
daniel

-- 
Daniel Stodden
LRR     -      Lehrstuhl für Rechnertechnik und Rechnerorganisation
Institut für Informatik der TU München             D-85748 Garching
http://www.lrr.in.tum.de/~stodden         mailto:stodden@cs.tum.edu
PGP Fingerprint: F5A4 1575 4C56 E26A 0B33  3D80 457E 82AE B0D8 735B



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Hi Mike,

On Jan 31, 2008 2:57 PM, Michael Abd-El-Malek <mabdelmalek@cmu.edu>
wrote:
> I tried to do that by first getting a grant for the user-space memory
> in the source domain as follows:
>
>         page_start = user_buf;
>         buffer_mfn = virt_to_mfn(page_start);
>         gnttab_grant_foreign_access_ref(grant_ref,
info->dev->otherend_id,
> buffer_mfn, 0);
>
> But the virt_to_mfn call asserted, in include/asm-x86_64/mach-xen/asm/
> maddr.h:pfn_to_mfn.
> The assertion that failed was: BUG_ON(end_pfn && pfn >=
end_pfn);
You have another problem here, because virt_to_mfn (and virt_to_page
etc.) operate on kernel virtual addresses, where converting to a
physical address effectively just involves subtracting an offset. User
virtual addresses are handled differently, and you''ll have to walk the
page table to find the PTE and from that obtain an MFN.

To respond to Daniel''s point, from my reading of the code, the grant
table entry takes a "GMFN", which is equivalent to an MFN on PV guests
without shadow translation. Therefore it shouldn''t be necessary to
translate this to a pseudo-physical frame number.
> So a few questions:
> 1) is it even possible to map a domain''s user-space page into
another
> domain?
I don''t see any problem in principle, though you''ll have to
make sure
that the page is locked in memory.
> 3) does the blktap driver do something similar? I tried looking at the
> blktap driver, because it looked like it might help.  But I''m new
to
> the Xen memory management code so it was a bit difficult to follow.
blktap maps memory that was granted by another domain''s kernel (in the
blkfront driver) into user-space. Therefore, on the destination side,
this is probably more complicated than you need. However, if you did
want to map the grant into user-space, you could use gntdev, which
employs the same mechanism as blktap.

Regards,

Derek Murray.

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On Jan 31, 2008, at 11:16 AM, Derek Murray wrote:
> On Jan 31, 2008 2:57 PM, Michael Abd-El-Malek <mabdelmalek@cmu.edu>  
> wrote:
>> I tried to do that by first getting a grant for the user-space memory
>> in the source domain as follows:
>>
>>        page_start = user_buf;
>>        buffer_mfn = virt_to_mfn(page_start);
>>        gnttab_grant_foreign_access_ref(grant_ref, info->dev- 
>> >otherend_id,
>> buffer_mfn, 0);
>>
>> But the virt_to_mfn call asserted, in include/asm-x86_64/mach-xen/ 
>> asm/
>> maddr.h:pfn_to_mfn.
>> The assertion that failed was: BUG_ON(end_pfn && pfn >=
end_pfn);
>
> You have another problem here, because virt_to_mfn (and virt_to_page
> etc.) operate on kernel virtual addresses, where converting to a
> physical address effectively just involves subtracting an offset. User
> virtual addresses are handled differently, and you''ll have to walk
the
> page table to find the PTE and from that obtain an MFN.
Thanks for your help Derek.  Fortunately I didn''t have to muck around  
with the page tables.  get_user_pages already does this.  So I called  
that function to get the relevant pages for the user pointer, kmap''ed  
those pages, got the virtual address using virt_to_mfn, and Xen''s  
grants worked.

Cheers,
Mike

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel