Re: [PATCH 0/2] Enable page table write protection

[Andrew Fish <afish@apple.com>]

> On Nov 29, 2017, at 5:16 PM, Wang, Jian J <jian.j.wang@intel.com> wrote:
> 
> When you split page tables, you need to allocate new pages for new page table.
> Since you have new page tables added, you need to mark them to be read-only
> as well, right? When you do this, the page table for the memory newly allocated
> might still needs to be split. This is the worst case but there's still chance of it,
> right? We cannot guarantee, theoretically, the page table for the newly allocated
> page tables is in the same as just split ones. So, in worst case, once we want to
> mark new page table to be read-only, we need to split the page table managing
> those memory, and if we need to do split, we need to allocate more new pages.
> This might fall into a recursive situation until all the smallest page table are created.
> In practice it's almost impossible to let it happen but the chances are we will fall into 
> such recursive situation more than one level.
> 

Why not just have the page table update code turn off the write protect when it is modifying tables, and then turn it back on? Or am I missing something?

Thanks,

Andrew Fish

>> -----Original Message-----
>> From: Yao, Jiewen
>> Sent: Thursday, November 30, 2017 8:52 AM
>> To: Wang, Jian J <jian.j.wang@intel.com>
>> Cc: edk2-devel@lists.01.org
>> Subject: RE: [edk2] [PATCH 0/2] Enable page table write protection
>> 
>> -- whenever you're trying to mark one page used as page table to be read-only,
>> you need to split its page table in advance
>> [Jiewen] Sorry, I do not quite understand the problem statement.
>> To set a page to be ReadOnly, you just flip the R/W bit in the page entry.
>> The step I expect is:
>> 1) You split page table.
>> 2) Fill the data structure in the new entry.
>> 3) Flip R/W bit (<== This is the new step).
>> 4) FlushPageTable.
>> 
>> Thank you
>> Yao Jiewen
>> 
>>> -----Original Message-----
>>> From: Wang, Jian J
>>> Sent: Thursday, November 30, 2017 8:44 AM
>>> To: Yao, Jiewen <jiewen.yao@intel.com>
>>> Cc: edk2-devel@lists.01.org
>>> Subject: RE: [edk2] [PATCH 0/2] Enable page table write protection
>>> 
>>> I did think of protecting new page table in cpu driver. But I think there's risks
>>> to do it, which is that there might be a chance that, whenever you're trying to
>>> mark one page used as page table to be read-only, you need to split its page
>>> table in advance, and again and again, until all page tables are for 4KB pages.
>>> 
>>> Although this is a rare case, or the page table "split" will just happen for a few
>>> times, it will slow down the boot process a little bit anyway. So I though  it's
>>> safer not to protect new page tables created after DxeIpl.
>>> 
>>> Maybe it's just over worrying about it. Maybe we just need to add a PCD to
>>> turn on/off it just in case. Do you have any ideas in mind?
>>> 
>>>> -----Original Message-----
>>>> From: Yao, Jiewen
>>>> Sent: Wednesday, November 29, 2017 9:35 PM
>>>> To: Wang, Jian J <jian.j.wang@intel.com>
>>>> Cc: edk2-devel@lists.01.org
>>>> Subject: Re: [edk2] [PATCH 0/2] Enable page table write protection
>>>> 
>>>> I do think we need update CPU driver to protect new allocated split page.
>>>> 
>>>> If you verified in shell env, I think it will exposed, please add a test to trigger
>>>> page split in CPU driver.
>>>> 
>>>> I recommend to write some unit test to parse page table in shell.
>>>> 
>>>> thank you!
>>>> Yao, Jiewen
>>>> 
>>>> 
>>>>> 在 2017年11月29日，下午8:15，Wang, Jian J <jian.j.wang@intel.com>
>>> 写
>>>> 道：
>>>>> 
>>>>> It's in the DxeIplPeim.
>>>>> 
>>>>> By the way, there's an issue in this patch. I forgot to protect page table for
>>> 32-
>>>> bit mode.
>>>>> So this patch works only for 64-bit mode. I'll add it in v2 patch.
>>>>> 
>>>>>> -----Original Message-----
>>>>>> From: Yao, Jiewen
>>>>>> Sent: Wednesday, November 29, 2017 6:56 PM
>>>>>> To: Wang, Jian J <jian.j.wang@intel.com>
>>>>>> Cc: edk2-devel@lists.01.org
>>>>>> Subject: Re: [edk2] [PATCH 0/2] Enable page table write protection
>>>>>> 
>>>>>> Is this code in CPU driver?
>>>>>> 
>>>>>> thank you!
>>>>>> Yao, Jiewen
>>>>>> 
>>>>>> 
>>>>>>> 在 2017年11月29日，下午6:24，Wang, Jian J
>> <jian.j.wang@intel.com>
>>>> 写
>>>>>> 道：
>>>>>>> 
>>>>>>> Yes, I validated them manually with JTAG debug tool.
>>>>>>> 
>>>>>>> if ((L3PageTable[Index3] & IA32_PG_PS) != 0) {
>>>>>>> // 1G page. Split to 2M.
>>>>>>> L2PageTable = AllocatePages (1);
>>>>>>> ASSERT (L2PageTable != NULL);
>>>>>>>  PhysicalAddress = L3PageTable[Index3] &
>>> PAGING_1G_ADDRESS_MASK_64;
>>>>>>>  for (Index = 0; Index < EFI_PAGE_SIZE/sizeof (UINT64); ++Index) {
>>>>>>>   L2PageTable[Index] = PhysicalAddress  | AddressEncMask |
>>>>>>>                        IA32_PG_PS | IA32_PG_P | IA32_PG_RW;
>>>>>>>   PhysicalAddress += SIZE_2MB;
>>>>>>> }
>>>>>>> L3PageTable[Index3] = (UINT64) (UINTN) L2PageTable |
>>> AddressEncMask |
>>>>>>>                                        IA32_PG_P |
>>> IA32_PG_RW;
>>>>>>> SetPageReadOnly (PageTableBase,
>>>>>> (EFI_PHYSICAL_ADDRESS)(UINTN)L2PageTable);
>>>>>>> }
>>>>>>> 
>>>>>>> The newly allocated page table is set in the SetPageReadOnly() itself
>>>>>> recursively, like
>>>>>>> above code in which L2PageTable is allocated and then set it to be
>>> read-only
>>>>>> after
>>>>>>> initializing the table content.
>>>>>>> 
>>>>>>>> -----Original Message-----
>>>>>>>> From: Yao, Jiewen
>>>>>>>> Sent: Wednesday, November 29, 2017 5:16 PM
>>>>>>>> To: Wang, Jian J <jian.j.wang@intel.com>
>>>>>>>> Cc: edk2-devel@lists.01.org
>>>>>>>> Subject: Re: [edk2] [PATCH 0/2] Enable page table write protection
>>>>>>>> 
>>>>>>>> Thanks.
>>>>>>>> 
>>>>>>>> May I know if this is validated in uefi shell, that all page table is
>> readonly?
>>>>>>>> 
>>>>>>>> I did not find the code to set new allocated split page to be readonly.
>> Can
>>>> you
>>>>>>>> give me a hand on that?
>>>>>>>> 
>>>>>>>> thank you!
>>>>>>>> Yao, Jiewen
>>>>>>>> 
>>>>>>>> 
>>>>>>>>> 在 2017年11月29日，下午4:47，Jian J Wang
>>> <jian.j.wang@intel.com>
>>>>>> 写
>>>>>>>> 道：
>>>>>>>>> 
>>>>>>>>> Write Protect feature (CR0.WP) is always enabled in driver
>>>>>>>> UefiCpuPkg/CpuDxe.
>>>>>>>>> But the memory pages used for page table are not set as read-only in
>>> the
>>>>>>>> driver
>>>>>>>>> DxeIplPeim, after the paging is setup. This might jeopardize the page
>>>> table
>>>>>>>>> integrity if there's buffer overflow occured in other part of system.
>>>>>>>>> 
>>>>>>>>> This patch series will change this situation by clearing R/W bit in page
>>>>>> attribute
>>>>>>>>> of the pages used as page table.
>>>>>>>>> 
>>>>>>>>> Validation works include booting Windows (10/server 2016) and Linux
>>>>>>>> (Fedora/Ubuntu)
>>>>>>>>> on OVMF and Intel real platform.
>>>>>>>>> 
>>>>>>>>> Jian J Wang (2):
>>>>>>>>> UefiCpuPkg/CpuDxe: Check CR0.WP before changing page table
>>>>>>>>> MdeModulePkg/DxeIpl: Mark page table as read-only
>>>>>>>>> 
>>>>>>>>> MdeModulePkg/Core/DxeIplPeim/X64/VirtualMemory.c | 166
>>>>>>>> +++++++++++++++++++++++
>>>>>>>>> MdeModulePkg/Core/DxeIplPeim/X64/VirtualMemory.h |  14 ++
>>>>>>>>> UefiCpuPkg/CpuDxe/CpuPageTable.c                 |  65
>>> ++++++++-
>>>>>>>>> 3 files changed, 241 insertions(+), 4 deletions(-)
>>>>>>>>> 
>>>>>>>>> --
>>>>>>>>> 2.14.1.windows.1
>>>>>>>>> 
>>>>>>>>> _______________________________________________
>>>>>>>>> edk2-devel mailing list
>>>>>>>>> edk2-devel@lists.01.org
>>>>>>>>> https://lists.01.org/mailman/listinfo/edk2-devel
> _______________________________________________
> edk2-devel mailing list
> edk2-devel@lists.01.org
> https://lists.01.org/mailman/listinfo/edk2-devel