Excerpts from Kefeng Wang's message of December 25, 2021 12:05 pm:
>
> On 2021/12/24 21:18, Christophe Leroy wrote:
>>
>> Le 24/12/2021 à 08:06, Kefeng Wang a écrit :
>>> On 2021/12/24 14:01, Christophe Leroy wrote:
>>>> Le 23/12/2021 à 11:21, Kefeng Wang a écrit :
>>>>> This reverts commit 517e1fbeb65f5eade8d14f46ac365db6c75aea9b.
>>>>>
>>>>> usercopy: Kernel memory exposure attempt detected from SLUB
>>>>> object not in SLUB page?! (offset 0, size 1048)!
>>>>> kernel BUG at mm/usercopy.c:99
>>>>> ...
>>>>> usercopy_abort+0x64/0xa0 (unreliable)
>>>>> __check_heap_object+0x168/0x190
>>>>> __check_object_size+0x1a0/0x200
>>>>> dev_ethtool+0x2494/0x2b20
>>>>> dev_ioctl+0x5d0/0x770
>>>>> sock_do_ioctl+0xf0/0x1d0
>>>>> sock_ioctl+0x3ec/0x5a0
>>>>> __se_sys_ioctl+0xf0/0x160
>>>>> system_call_exception+0xfc/0x1f0
>>>>> system_call_common+0xf8/0x200
>>>>>
>>>>> When run ethtool eth0, the BUG occurred, the code shows below,
>>>>>
>>>>> data = vzalloc(array_size(gstrings.len, ETH_GSTRING_LEN));
>>>>> copy_to_user(useraddr, data, gstrings.len * ETH_GSTRING_LEN))
>>>>>
>>>>> The data is alloced by vmalloc(), virt_addr_valid(ptr) will return true
>>>>> on PowerPC64, which leads to the panic, add back the
>>>>> is_vmalloc_or_module()
>>>>> check to fix it.
>>>> Is it expected that virt_addr_valid() returns true on PPC64 for
>>>> vmalloc'ed memory ? If that's the case it also means that
>>>> CONFIG_DEBUG_VIRTUAL won't work as expected either.
>>> Our product reports this bug to me, after let them do some test,
>>>
>>> I found virt_addr_valid return true for vmalloc'ed memory on their board.
>>>
>>> I think DEBUG_VIRTUAL could not be work well too, but I can't test it.
>>>
>>>> If it is unexpected, I think you should fix PPC64 instead of adding this
>>>> hack back. Maybe the ARM64 fix can be used as a starting point, see
>>>> commit 68dd8ef32162 ("arm64: memory: Fix virt_addr_valid() using
>>>> __is_lm_address()")
>>> Yes， I check the history, fix virt_addr_valid() on PowerPC is what I
>>> firstly want to do,
>>>
>>> but I am not familiar with PPC, and also HARDENED_USERCOPY on other's
>>> ARCHs could
>>>
>>> has this issue too, so I add the workaround back.
>>>
>>>
>>> 1) PPC maintainer/expert, any suggestion ?
>>>
>>> 2) Maybe we could add some check to WARN this scenario.
>>>
>>> --- a/mm/usercopy.c
>>> +++ b/mm/usercopy.c
>>> @@ -229,6 +229,8 @@ static inline void check_heap_object(const void
>>> *ptr, unsigned long n,
>>> if (!virt_addr_valid(ptr))
>>> return;
>>>
>>> + WARN_ON_ONCE(is_vmalloc_or_module_addr(ptr));
>
>>>
>>>> In the meantime, can you provide more information on your config,
>>>> especially which memory model is used ?
>>> Some useful configs,
>>>
>>> CONFIG_PPC64=y
>>> CONFIG_PPC_BOOK3E_64=y
>>> CONFIG_E5500_CPU=y
>>> CONFIG_TARGET_CPU_BOOL=y
>>> CONFIG_PPC_BOOK3E=y
>>> CONFIG_E500=y
>>> CONFIG_PPC_E500MC=y
>>> CONFIG_PPC_FPU=y
>>> CONFIG_FSL_EMB_PERFMON=y
>>> CONFIG_FSL_EMB_PERF_EVENT=y
>>> CONFIG_FSL_EMB_PERF_EVENT_E500=y
>>> CONFIG_BOOKE=y
>>> CONFIG_PPC_FSL_BOOK3E=y
>>> CONFIG_PTE_64BIT=y
>>> CONFIG_PHYS_64BIT=y
>>> CONFIG_PPC_MMU_NOHASH=y
>>> CONFIG_PPC_BOOK3E_MMU=y
>>> CONFIG_SELECT_MEMORY_MODEL=y
>>> CONFIG_FLATMEM_MANUAL=y
>>> CONFIG_FLATMEM=y
>>> CONFIG_FLAT_NODE_MEM_MAP=y
>>> CONFIG_SPARSEMEM_VMEMMAP_ENABLE=y
>>>
>> OK so it is PPC64 book3e and with flatmem.
>>
>> The problem is virt_to_pfn() which uses __pa()
>>
>> __pa(x) on PPC64 is (x) & 0x0fffffffffffffffUL
>>
>> And on book3e/64 we have
>>
>> VMALLOC_START = KERN_VIRT_START = ASM_CONST(0x8000000000000000)
>>
>>
>> It means that __pa() will return a valid PFN for VMALLOCed addresses.
>>
>>
>> So an additional check is required in virt_addr_valid(), maybe check
>> that (kaddr & PAGE_OFFSET) == PAGE_OFFSET
>>
>> Can you try that ?
>>
>> #define virt_addr_valid(kaddr) ((kaddr & PAGE_OFFSET) == PAGE_OFFSET &&
>> pfn_valid(virt_to_pfn(kaddr)))
>
> I got this commit,
>
> commit 4dd7554a6456d124c85e0a4ad156625b71390b5c
>
> Author: Nicholas Piggin <npiggin@xxxxxxxxx>
> Date: Wed Jul 24 18:46:37 2019 +1000
>
> powerpc/64: Add VIRTUAL_BUG_ON checks for __va and __pa addresses
>
> Ensure __va is given a physical address below PAGE_OFFSET, and __pa is
> given a virtual address above PAGE_OFFSET.
>
> It has check the PAGE_OFFSET in __pa, will test it and resend the
> patch(with above warning changes).
What did you get with this commit? Is this what causes the crash?
riscv for example with flatmem also relies on pfn_valid to do the right
thing, so as far as I can see the check should exclude vmalloc addresses
and it's just a matter of virt_addr_valid not to give virt_to_pfn an
address < PAGE_OFFSET.
If we take riscv's implementation
diff --git a/arch/powerpc/include/asm/page.h b/arch/powerpc/include/asm/page.h
index 254687258f42..7713188516a6 100644
--- a/arch/powerpc/include/asm/page.h
+++ b/arch/powerpc/include/asm/page.h
@@ -132,7 +132,10 @@ static inline bool pfn_valid(unsigned long pfn)
#define virt_to_page(kaddr) pfn_to_page(virt_to_pfn(kaddr))
#define pfn_to_kaddr(pfn) __va((pfn) << PAGE_SHIFT)
-#define virt_addr_valid(kaddr) pfn_valid(virt_to_pfn(kaddr))
+#define virt_addr_valid(vaddr) ({ \
+ unsigned long _addr = (unsigned long)vaddr; \
+ (unsigned long)(_addr) >= PAGE_OFFSET && pfn_valid(virt_to_pfn(_addr)); \
+})
/*
* On Book-E parts we need __va to parse the device tree and we can't
