On 07/14/2017 10:05 PM, Linus Torvalds wrote:
> On Fri, Jul 14, 2017 at 7:25 AM, Dave Jones <davej@xxxxxxxxxxxxxxxxx> wrote:
>> On Thu, Jul 13, 2017 at 05:16:24PM -0400, Anna Schumaker wrote:
>> >
>> > git://git.linux-nfs.org/projects/anna/linux-nfs.git tags/nfs-for-4.13-1
>>
>> Since this landed, I'm seeing this during boot..
>>
>> ==================================================================
>> BUG: KASAN: global-out-of-bounds in strscpy+0x4a/0x230
>> Read of size 8 at addr ffffffffb4eeaf20 by task nfsd/688
>
> Is KASAN aware that strscpy() does the word-at-a-time optimistic reads
> of the sources?
>
Nope.
> The problem may be that the source is initialized from the global
> string "nfsd", and KASAN may be unhappy abotu the fact that we read 8
> bytes from a 5-byte string (four plus NUL) as we do the word-at-a-time
> strscpy..
>
Right.
> That said, we do check the size first (because we also *write* 8 bytes
> at a time), so maybe KASAN shouldn't even need to care.
>
Perhaps we could fallback to unoptimzed copy for KASAN case by setting max = 0
in strscpy().
> Hmm. it really looks to me like this is actually a compiler bug (I'm
> using current gcc in F26, which is gcc-7.1.1 - I'm assuming DaveJ is
> the same).
>
> This is the source code in __ip_map_lookup:
>
> struct ip_map ip;
> .....
> strcpy(ip.m_class, class);
>
> and "m_class" is 8 bytes in size:
>
> struct ip_map {
> ...
> char m_class[8]; /* e.g. "nfsd" */
> ...
>
> yet when I look at the generated code for __ip_map_lookup, I see
>
> movl $32, %edx #,
> movq %r13, %rsi # class,
> leaq 48(%rax), %rdi #, tmp126
> call strscpy #
>
> what's the bug here? Look at that third argume8nt - %rdx. It is
> initialized to 32.
>
> WTF?
>
It's not a compiler bug, it's a bug in our strcpy().
Whoever wrote this strcpy() into strscpy() code apparently didn't read carefully
enough gcc manual about __builtin_object_size().
Summary from https://gcc.gnu.org/onlinedocs/gcc/Object-Size-Checking.html :
__builtin_object_size(ptr, type) returns a constant number of bytes from 'ptr' to the end of the object 'ptr'
pointer points to. "type" is an integer constant from 0 to 3. If the least significant bit is clear, objects
are whole variables, if it is set, a closest surrounding subobject is considered the object a pointer points to.
The second bit determines if maximum or minimum of remaining bytes is computed.
We have type = 0 in strcpy(), so the least significant bit is clear. So the 'ptr' is considered as a pointer to the whole
variable i.e. pointer to struct ip_map ip;
And the number of bytes from 'ip.m_class' to the end of the ip object is exactly 32.
I suppose that changing the type to 1 should fix this bug.
> The code to turn "strcpy()" into "strscpy()" should pick the *smaller*
> of the two object sizes as the size argument. How the hell is that
> size argument 32?
>
> Am I missing something? DaveJ, do you see the same?
>
> Linus
>
