[PATCH] powerpc/fault: fix wrong KUAP fault for IO_URING
Jens Axboe
axboe at kernel.dk
Fri Feb 5 04:03:34 AEDT 2021
On 2/4/21 10:01 AM, Aneesh Kumar K.V wrote:
> On 2/4/21 10:23 PM, Jens Axboe wrote:
>> On 2/1/21 11:30 PM, Aneesh Kumar K.V wrote:
>>> On 2/2/21 11:50 AM, Christophe Leroy wrote:
>>>>
>>>>
>>>> Le 02/02/2021 à 07:16, Aneesh Kumar K.V a écrit :
>>>>> On 2/2/21 11:32 AM, Christophe Leroy wrote:
>>>>>>
>>>>>>
>>>>>> Le 02/02/2021 à 06:55, Aneesh Kumar K.V a écrit :
>>>>>>> Aneesh Kumar K.V <aneesh.kumar at linux.ibm.com> writes:
>>>>>>>
>>>>>>>> Nicholas Piggin <npiggin at gmail.com> writes:
>>>>>>>>
>>>>>>>>> Excerpts from Michael Ellerman's message of January 30, 2021 9:22 pm:
>>>>>>>>>> Christophe Leroy <christophe.leroy at csgroup.eu> writes:
>>>>>>>>>>> +Aneesh
>>>>>>>>>>>
>>>>>>>>>>> Le 29/01/2021 à 07:52, Zorro Lang a écrit :
>>>>>>>>>> ..
>>>>>>>>>>>> [ 96.200296] ------------[ cut here ]------------
>>>>>>>>>>>> [ 96.200304] Bug: Read fault blocked by KUAP!
>>>>>>>>>>>> [ 96.200309] WARNING: CPU: 3 PID: 1876 at
>>>>>>>>>>>> arch/powerpc/mm/fault.c:229 bad_kernel_fault+0x180/0x310
>>>>>>>>>>>
>>>>>>>>>>>> [ 96.200734] NIP [c000000000849424]
>>>>>>>>>>>> fault_in_pages_readable+0x104/0x350
>>>>>>>>>>>> [ 96.200741] LR [c00000000084952c]
>>>>>>>>>>>> fault_in_pages_readable+0x20c/0x350
>>>>>>>>>>>> [ 96.200747] --- interrupt: 300
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Problem happens in a section where userspace access is supposed
>>>>>>>>>>> to be granted, so the patch you
>>>>>>>>>>> proposed is definitely not the right fix.
>>>>>>>>>>>
>>>>>>>>>>> c000000000849408: 2c 01 00 4c isync
>>>>>>>>>>> c00000000084940c: a6 03 3d 7d mtspr 29,r9 <== granting
>>>>>>>>>>> userspace access permission
>>>>>>>>>>> c000000000849410: 2c 01 00 4c isync
>>>>>>>>>>> c000000000849414: 00 00 36 e9 ld r9,0(r22)
>>>>>>>>>>> c000000000849418: 20 00 29 81 lwz r9,32(r9)
>>>>>>>>>>> c00000000084941c: 00 02 29 71 andi. r9,r9,512
>>>>>>>>>>> c000000000849420: 78 d3 5e 7f mr r30,r26
>>>>>>>>>>> ==> c000000000849424: 00 00 bf 8b lbz r29,0(r31) <==
>>>>>>>>>>> accessing userspace
>>>>>>>>>>> c000000000849428: 10 00 82 41 beq c000000000849438
>>>>>>>>>>> <fault_in_pages_readable+0x118>
>>>>>>>>>>> c00000000084942c: 2c 01 00 4c isync
>>>>>>>>>>> c000000000849430: a6 03 bd 7e mtspr 29,r21 <==
>>>>>>>>>>> clearing userspace access permission
>>>>>>>>>>> c000000000849434: 2c 01 00 4c isync
>>>>>>>>>>>
>>>>>>>>>>> My first guess is that the problem is linked to the following
>>>>>>>>>>> function, see the comment
>>>>>>>>>>>
>>>>>>>>>>> /*
>>>>>>>>>>> * For kernel thread that doesn't have thread.regs return
>>>>>>>>>>> * default AMR/IAMR values.
>>>>>>>>>>> */
>>>>>>>>>>> static inline u64 current_thread_amr(void)
>>>>>>>>>>> {
>>>>>>>>>>> if (current->thread.regs)
>>>>>>>>>>> return current->thread.regs->amr;
>>>>>>>>>>> return AMR_KUAP_BLOCKED;
>>>>>>>>>>> }
>>>>>>>>>>>
>>>>>>>>>>> Above function was introduced by commit 48a8ab4eeb82
>>>>>>>>>>> ("powerpc/book3s64/pkeys: Don't update SPRN_AMR
>>>>>>>>>>> when in kernel mode")
>>>>>>>>>>
>>>>>>>>>> Yeah that's a bit of a curly one.
>>>>>>>>>>
>>>>>>>>>> At some point io_uring did kthread_use_mm(), which is supposed to
>>>>>>>>>> mean
>>>>>>>>>> the kthread can operate on behalf of the original process that
>>>>>>>>>> submitted
>>>>>>>>>> the IO.
>>>>>>>>>>
>>>>>>>>>> But because KUAP is implemented using memory protection keys, it
>>>>>>>>>> depends
>>>>>>>>>> on the value of the AMR register, which is not part of the mm,
>>>>>>>>>> it's in
>>>>>>>>>> thread.regs->amr.
>>>>>>>>>>
>>>>>>>>>> And what's worse by the time we're in kthread_use_mm() we no
>>>>>>>>>> longer have
>>>>>>>>>> access to the thread.regs->amr of the original process that
>>>>>>>>>> submitted
>>>>>>>>>> the IO.
>>>>>>>>>>
>>>>>>>>>> We also can't simply move the AMR into the mm, precisely because
>>>>>>>>>> it's
>>>>>>>>>> per thread, not per mm.
>>>>>>>>>>
>>>>>>>>>> So TBH I don't know how we're going to fix this.
>>>>>>>>>>
>>>>>>>>>> I guess we could return AMR=unblocked for kernel threads, but that's
>>>>>>>>>> arguably a bug because it allows a process to circumvent memory
>>>>>>>>>> keys by
>>>>>>>>>> asking the kernel to do the access.
>>>>>>>>>
>>>>>>>>> We shouldn't need to inherit AMR should we? We only need it to be
>>>>>>>>> locked
>>>>>>>>> for kernel threads until it's explicitly unlocked -- nothing mm
>>>>>>>>> specific
>>>>>>>>> there. I think current_thread_amr could return 0 for kernel
>>>>>>>>> threads? Or
>>>>>>>>> I would even avoid using that function for allow_user_access and open
>>>>>>>>> code the kthread case and remove it from current_thread_amr().
>>>>>>>>>
>>>>>>>>> Thanks,
>>>>>>>>> Nick
>>>>>>>>
>>>>>>>
>>>>>>> updated one
>>>>>>>
>>>>>>> From 8fdb0680f983940d61f91da8252b13c8d3e8ebee Mon Sep 17 00:00:00 2001
>>>>>>> From: "Aneesh Kumar K.V" <aneesh.kumar at linux.ibm.com>
>>>>>>> Date: Tue, 2 Feb 2021 09:23:38 +0530
>>>>>>> Subject: [PATCH v2] powerpc/kuap: Allow kernel thread to access
>>>>>>> userspace
>>>>>>> after kthread_use_mm
>>>>>>>
>>>>>>> This fix the bad fault reported by KUAP when io_wqe_worker access
>>>>>>> userspace.
>>>>>>>
>>>>>>> Bug: Read fault blocked by KUAP!
>>>>>>> WARNING: CPU: 1 PID: 101841 at arch/powerpc/mm/fault.c:229
>>>>>>> __do_page_fault+0x6b4/0xcd0
>>>>>>> NIP [c00000000009e7e4] __do_page_fault+0x6b4/0xcd0
>>>>>>> LR [c00000000009e7e0] __do_page_fault+0x6b0/0xcd0
>>>>>>> ..........
>>>>>>> Call Trace:
>>>>>>> [c000000016367330] [c00000000009e7e0] __do_page_fault+0x6b0/0xcd0
>>>>>>> (unreliable)
>>>>>>> [c0000000163673e0] [c00000000009ee3c] do_page_fault+0x3c/0x120
>>>>>>> [c000000016367430] [c00000000000c848] handle_page_fault+0x10/0x2c
>>>>>>> --- interrupt: 300 at iov_iter_fault_in_readable+0x148/0x6f0
>>>>>>> ..........
>>>>>>> NIP [c0000000008e8228] iov_iter_fault_in_readable+0x148/0x6f0
>>>>>>> LR [c0000000008e834c] iov_iter_fault_in_readable+0x26c/0x6f0
>>>>>>> interrupt: 300
>>>>>>> [c0000000163677e0] [c0000000007154a0] iomap_write_actor+0xc0/0x280
>>>>>>> [c000000016367880] [c00000000070fc94] iomap_apply+0x1c4/0x780
>>>>>>> [c000000016367990] [c000000000710330]
>>>>>>> iomap_file_buffered_write+0xa0/0x120
>>>>>>> [c0000000163679e0] [c00800000040791c]
>>>>>>> xfs_file_buffered_aio_write+0x314/0x5e0 [xfs]
>>>>>>> [c000000016367a90] [c0000000006d74bc] io_write+0x10c/0x460
>>>>>>> [c000000016367bb0] [c0000000006d80e4] io_issue_sqe+0x8d4/0x1200
>>>>>>> [c000000016367c70] [c0000000006d8ad0] io_wq_submit_work+0xc0/0x250
>>>>>>> [c000000016367cb0] [c0000000006e2578]
>>>>>>> io_worker_handle_work+0x498/0x800
>>>>>>> [c000000016367d40] [c0000000006e2cdc] io_wqe_worker+0x3fc/0x4f0
>>>>>>> [c000000016367da0] [c0000000001cb0a4] kthread+0x1c4/0x1d0
>>>>>>> [c000000016367e10] [c00000000000dbf0]
>>>>>>> ret_from_kernel_thread+0x5c/0x6c
>>>>>>>
>>>>>>> The kernel consider thread AMR value for kernel thread to be
>>>>>>> AMR_KUAP_BLOCKED. Hence access to userspace is denied. This
>>>>>>> of course not correct and we should allow userspace access after
>>>>>>> kthread_use_mm(). To be precise, kthread_use_mm() should inherit the
>>>>>>> AMR value of the operating address space. But, the AMR value is
>>>>>>> thread-specific and we inherit the address space and not thread
>>>>>>> access restrictions. Because of this ignore AMR value when accessing
>>>>>>> userspace via kernel thread.
>>>>>>>
>>>>>>> Signed-off-by: Aneesh Kumar K.V <aneesh.kumar at linux.ibm.com>
>>>>>>> ---
>>>>>>> Changes from v1:
>>>>>>> * Address review feedback from Nick
>>>>>>>
>>>>>>> arch/powerpc/include/asm/book3s/64/kup.h | 8 +++++++-
>>>>>>> 1 file changed, 7 insertions(+), 1 deletion(-)
>>>>>>>
>>>>>>> diff --git a/arch/powerpc/include/asm/book3s/64/kup.h
>>>>>>> b/arch/powerpc/include/asm/book3s/64/kup.h
>>>>>>> index f50f72e535aa..95f4df99249e 100644
>>>>>>> --- a/arch/powerpc/include/asm/book3s/64/kup.h
>>>>>>> +++ b/arch/powerpc/include/asm/book3s/64/kup.h
>>>>>>> @@ -384,7 +384,13 @@ static __always_inline void
>>>>>>> allow_user_access(void __user *to, const void __user
>>>>>>> // This is written so we can resolve to a single case at build
>>>>>>> time
>>>>>>> BUILD_BUG_ON(!__builtin_constant_p(dir));
>>>>>>> - if (mmu_has_feature(MMU_FTR_PKEY))
>>>>>>> + /*
>>>>>>> + * if it is a kthread that did kthread_use_mm() don't
>>>>>>> + * use current_thread_amr().
>>>>>>
>>>>>> According to include/linux/sched.h, PF_KTHREAD means /* I am a kernel
>>>>>> thread */
>>>>>> It doesn't seem to be related to kthread_use_mm()
>>>>>
>>>>> That should be a sufficient check here. if we did reach here without
>>>>> calling kthread_user_mm, we will crash on access because we don't have
>>>>> a mm attached to the current process. a kernel thread with
>>>>> kthread_use_mm has
>>>>
>>>> Ok but then the comment doesn't match the check.
>>>
>>>
>>> I was trying to be explict in the comment that we expect the thread to
>>> have done kthread_use_mm().
>>>
>>>>
>>>> And also the comment in current_thread_amr() is then misleading.
>>>>
>>>> Why not do the current->flags & PF_KTHREAD check in current_thread_amr()
>>>> and return 0 in that case instead of BLOCKED ?
>>>
>>> In my view currrent_thread_amr() is more generic and we want to be
>>> explicit there that a kernel thread AMR is KUAP_BLOCKED. Only when we
>>> call allow user access, we relax the AMR value.
>>
>> Just following up on this, as I'd hate to have 5.11 released with this
>> bug in it for powerpc. It'll obviously also affect other cases of a
>> kernel thread faulting after having done kthread_use_mm(), though I'm
>> not sure how widespread that is. In any case, it'll leave io_uring
>> mostly broken on powerpc if this isn't patched for release.
>>
>
> I am waiting for test feedback on the change I posted earlier. I am also
> running a regression run myself. Once that is complete i will post the
> patch as a separate email.
Perfect, sounds good!
--
Jens Axboe
More information about the Linuxppc-dev
mailing list