Re: [PATCH] Revert "mm/usercopy: Drop extra is_vmalloc_or_module() check"

[Christophe Leroy]

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.

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()")

In the meantime, can you provide more information on your config, 
especially which memory model is used ?

Christophe