[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xen-devel] [PATCH L1TF v8 9/9] common/grant_table: block speculative out-of-bound accesses


  • To: Jan Beulich <JBeulich@xxxxxxxx>
  • From: Norbert Manthey <nmanthey@xxxxxxxxx>
  • Date: Mon, 4 Mar 2019 09:15:07 +0100
  • Autocrypt: addr=nmanthey@xxxxxxxxx; prefer-encrypt=mutual; keydata= mQINBFoJQc0BEADM8Z7hB7AnW6ErbSMsYkKh4HLAPfoM+wt7Fd7axHurcOgFJEBOY2gz0isR /EDiGxYyTgxt5PZHJIfra0OqXRbWuLltbjhJACbu35eaAo8UM4/awgtYx3O1UCbIlvHGsYDg kXjF8bBrVjPu0+g55XizX6ot/YPAgmWTdH8qXoLYVZVWJilKlTqpYEVvarSn/BVgCbIsQIps K93sOTN9eJKDSqHvbkgKl9XG3WsZ703431egIpIZpfN0zZtzumdZONb7LiodcFHJ717vvd89 3Hv2bYv8QLSfYsZcSnyU0NVzbPhb1WtaduwXwNmnX1qHJuExzr8EnRT1pyhVSqouxt+xkKbV QD9r+cWLChumg3g9bDLzyrOTlEfAUNxIqbzSt03CRR43dWgfgGiLDcrqC2b1QR886WDpz4ok xX3fdLaqN492s/3c59qCGNG30ebAj8AbV+v551rsfEba+IWTvvoQnbstc6vKJCc2uG8rom5o eHG/bP1Ug2ht6m/0uWRyFq9C27fpU9+FDhb0ZsT4UwOCbthe35/wBZUg72zDpT/h5lm64G6C 0TRqYRgYcltlP705BJafsymmAXOZ1nTCuXnYAB9G9LzZcKKq5q0rP0kp7KRDbniirCUfp7jK VpPCOUEc3tS1RdCCSeWNuVgzLnJdR8W2h9StuEbb7hW4aFhwRQARAQABtCROb3JiZXJ0IE1h bnRoZXkgPG5tYW50aGV5QGFtYXpvbi5kZT6JAj0EEwEIACcFAloJQc0CGyMFCQlmAYAFCwkI BwIGFQgJCgsCBBYCAwECHgECF4AACgkQZ+8yS8zN62ajmQ/6AlChoY5UlnUaH/jgcabyAfUC XayHgCcpL1SoMKvc2rCA8PF0fza3Ep2Sw0idLqC/LyAYbI6gMYavSZsLcsvY6KYAZKeaEriG 7R6cSdrbmRcKpPjwvv4iY6G0DBTeaqfNjGe1ECY8u522LprDQVquysJIf3YaEyxoK/cLSb0c kjzpqI1P9Vh+8BQb5H9gWpakbhFIwbRGHdAF1roT7tezmEshFS0IURJ2ZFEI+ZgWgtl1MBwN sBt65im7x5gDo25h8A5xC9gLXTc4j3tk+3huaZjUJ9mCbtI12djVtspjNvDyUPQ5Mxw2Jwar C3/ZC+Nkb+VlymmErpnEUZNltcq8gsdYND4TlNbZ2JhD0ibiYFQPkyuCVUiVtimXfh6po9Yt OkE0DIgEngxMYfTTx01Zf6iwrbi49eHd/eQQw3zG5nn+yZsEG8UcP1SCrUma8p93KiKOedoL n43kTg4RscdZMjj4v6JkISBcGTR4uotMYP4M0zwjklnFXPmrZ6/E5huzUpH9B7ZIe/SUu8Ur xww/4dN6rfqbNzMxmya8VGlEQZgUMWcck+cPrRLB09ZOk4zq9i/yaHDEZA1HNOfQ9UCevXV5 7seXSX7PCY6WDAdsT3+FuaoQ7UoWN3rdpb+064QKZ0FsHeGzUd7MZtlgU4EKrh25mTSNZYRs nTz2zT/J33e5Ag0EWglBzQEQAKioD1gSELj3Y47NE11oPkzWWdxKZdVr8B8VMu6nVAAGFRSf Dms4ZmwGY27skMmOH2srnZyTfm9FaTKr8RI+71Fh9nfB9PMmwzA7OIY9nD73/HqPywzTTleG MlALmnuY6xFRSDmqmvxDHgWyzB4TgPWt8+hW3+TJKCx2RgLAdSuULZla4lia+NlS8WNRUDGK sFJCCB3BW5I/cocfpBEUqLbbmnPuD9UKpEnFcYWD9YaDNcBTjSc7iDsvtpdrBXg5VETOz/TQ /CmVs9h/5zug8O4bXxHEEJpCAxs4cGKxowBqx/XJfkwdWeo/LdaeR+LRbXvq4A32HSkyj9sV vygwt2OFEk493JGik8qtAA/oPvuqVPJGacxmZ7zKR12c0mnKCHiexFJzFbC7MSiUhhe8nNiM p6Sl6EZmsTUXhV2bd2M12Bqcss3TTJ1AcW04T4HYHVCSxwl0dVfcf3TIaH0BSPiwFxz0FjMk 10umoRvUhYYoYpPFCz8dujXBlfB8q2tnHltEfoi/EIptt1BMNzTYkHKArj8Fwjf6K+nQ3a8p 1cWfkYpA5bRqbhbplzpa0u1Ex0hZk6pka0qcVgqmH31O2OcSsqeKfUfHkzj3Q6dmuwm1je/f HWH9N1gDPEp1RB5bIxPnOG1Z4SNl9oVQJhc4qoJiqbvkciivYcH7u2CBkboFABEBAAGJAiUE GAEIAA8FAloJQc0CGwwFCQlmAYAACgkQZ+8yS8zN62YU9Q//WTnN28aBX1EhDidVho80Ql2b tV1cDRh/vWTcM4qoM8vzW4+F/Ive6wDVAJ7zwAv8F8WPzy+acxtHLkyYk14M6VZ1eSy0kV0+ RZQdQ+nPtlb1MoDKw2N5zhvs8A+WD8xjDIA9i21hQ/BNILUBINuYKyR19448/41szmYIEhuJ R2fHoLzNdXNKWQnN3/NPTuvpjcrkXKJm2k32qfiys9KBcZX8/GpuMCc9hMuymzOr+YlXo4z4 1xarEJoPOQOXnrmxN4Y30/qmf70KHLZ0GQccIm/o/XSOvNGluaYv0ZVJXHoCnYvTbi0eYvz5 OfOcndqLOfboq9kVHC6Yye1DLNGjIVoShJGSsphxOx2ryGjHwhzqDrLiRkV82gh6dUHKxBWd DXfirT8a4Gz/tY9PMxan67aSxQ5ONpXe7g7FrfrAMe91XRTf50G3rHb8+AqZfxZJFrBn+06i p1cthq7rJSlYCqna2FedTUT+tK1hU9O0aK4ZYYcRzuTRxjd4gKAWDzJ1F/MQ12ftrfCAvs7U sVbXv2TndGIleMnheYv1pIrXEm0+sdz5v91l2/TmvkyyWT8s2ksuZis9luh+OubeLxHq090C hfavI9WxhitfYVsfo2kr3EotGG1MnW+cOkCIX68w+3ZS4nixZyJ/TBa7RcTDNr+gjbiGMtd9 pEddsOqYwOs=
  • Cc: Juergen Gross <jgross@xxxxxxxx>, Tim Deegan <tim@xxxxxxx>, Stefano Stabellini <sstabellini@xxxxxxxxxx>, Wei Liu <wei.liu2@xxxxxxxxxx>, Konrad Rzeszutek Wilk <konrad.wilk@xxxxxxxxxx>, George Dunlap <George.Dunlap@xxxxxxxxxxxxx>, Andrew Cooper <andrew.cooper3@xxxxxxxxxx>, Ian Jackson <Ian.Jackson@xxxxxxxxxxxxx>, Dario Faggioli <dfaggioli@xxxxxxxx>, Martin Pohlack <mpohlack@xxxxxxxxx>, wipawel@xxxxxxxxx, Julien Grall <julien.grall@xxxxxxx>, David Woodhouse <dwmw@xxxxxxxxxxxx>, "Martin Mazein\(amazein\)" <amazein@xxxxxxxxx>, xen-devel <xen-devel@xxxxxxxxxxxxxxxxxxxx>, Julian Stecklina <jsteckli@xxxxxxxxx>, Bjoern Doebel <doebel@xxxxxxxxx>
  • Delivery-date: Mon, 04 Mar 2019 08:15:31 +0000
  • List-id: Xen developer discussion <xen-devel.lists.xenproject.org>
  • Openpgp: preference=signencrypt

On 2/28/19 11:00, Jan Beulich wrote:
>>>> On 27.02.19 at 14:01, <nmanthey@xxxxxxxxx> wrote:
>> On 2/25/19 17:46, Jan Beulich wrote:
>>> I would really like to ask that I (or someone else) don't need to
>>> go through and list remaining version checks again - after all I
>>> had done so for v6 already, and I didn't go through all of them
>>> again for v7 assuming that you would have worked through the
>>> entire set.
>> So, here is the annotation for all of them. Anyone that I did not
>> include in the list has been fixed in previous versions, or will be
>> fixed in the next version:
>>
>> git grep -np version common/grant_table.c
>>
>> common/grant_table.c:831:static int _set_status(unsigned gt_version,
>> common/grant_table.c:840:    if ( gt_version == 1 )
>>
>> -> I do not see how out-of-bound accesses happen in the called functions
>> there.
> Both functions get shah passed into them, which may point to the
> other version's layout. Earlier fences don't help speculation on this
> conditional.
Whenever this function is called, the shah field has the same structure
for both versions. The v2 grant entry has a plain header hdr, which is
used here, and which is forwarded to the set_status method. Another
property that we never talked about here is that we only care about
cross cache-line accesses. As soon as any byte is touched on a cache
line, the whole line is brought into the cache. As the code around the
set_status method already pulls in the corresponding cache line, that
code looks okay to me from a L1TF perspective.
>
>> common/grant_table.c=1444=unmap_common_complete(struct
>> gnttab_unmap_common *op)
>> common/grant_table.c:1469:    if ( rgt->gt_version == 1 )
>>
>> -> I do not see how to be exploitable, as the shared_entry_header call
>> above just used an lfence.
> Again, earlier fences don't suppress subsequent speculation on
> an independent conditional.

Yes, agreed. However, in the current version, the array is allocated
already for the one branch, and the flags are in the cache for both
versions as well. The above evaluate_nospec that protects the evaluation
of the variable "done" above prevents out-of-bound values for ref. Now,
if you want me to be prepared for the future where people might move the
allocation of the status array from the generic code to the v2 specific
code, I would have to add another lfence here today already.

>
>> common/grant_table.c=1761=gnttab_grow_table(struct domain *d, unsigned
>> int req_nr_frames)
>> common/grant_table.c:1800:    /* Status pages - version 2 */
>> common/grant_table.c:1801:    if ( gt->gt_version > 1 )
>>
>> -> I do not see how out-of-bound access could happen. This calls
>> gnttab_populate_status_frames that allocates pages and should not touch
>> more memory than before
> We've been talking about the speculation window being perhaps
> hundreds of insns. It may be purely theoretical, but speculation all
> the way through alloc_xenheap_page() would lead to a speculative
> store to gt->status[]. _Right now_ that array gets allocated in
> grant_table_init(), but I can't see why that couldn't be moved to
> gnttab_set_version(), so the access above could be a latent issue.

I agree that this might be a latent issue, and in case people modify the
code, the effect wrt L1TF mitigation might have to be taken into
account. While we have a specific problematic code snippet here, that
statement is true for all the other mitigations as well. Since I do not
see a problem with the code today, and I do not like to pay the penalty
of the lfence to protect against something that might not happen in the
future, I would not add the lfence here.

Best,
Norbert

>
> I'll skip going into details of further ones, assuming that some of
> them follow patterns above.
>
> Jan
>
>




Amazon Development Center Germany GmbH
Krausenstr. 38
10117 Berlin
Geschaeftsfuehrer: Christian Schlaeger, Ralf Herbrich
Ust-ID: DE 289 237 879
Eingetragen am Amtsgericht Charlottenburg HRB 149173 B

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.