Re: fs: NULL deref in atime_needs_update

["Mickaël Salaün" <mic@digikod.net>]

[-- Attachment #1: Type: text/plain, Size: 1286 bytes --]


On 20/02/2016 04:54, Al Viro wrote:
> On Sat, Feb 20, 2016 at 03:21:27AM +0000, Al Viro wrote:
>> On Fri, Feb 19, 2016 at 08:32:10PM +0100, Dmitry Vyukov wrote:
>>>> BUG: unable to handle kernel NULL pointer dereference at 0000000000000050
>>
>> NULL inode->i_sb, by the look of the offset, but I really don't understand
>> where the hell is that code doing (or how is that instruction going to
>> generate dereferencing of 0x50, for that matter).
> 
> BTW, Mickaël's trace *does* make sense and it's definitely NULL inode->i_sb
> (inode itself - in %rsi, inode->i_sb - in %rdx, offset of s_flags is 0x50,
> the line in question is
>         if ((inode->i_sb->s_flags & MS_NODIRATIME) && S_ISDIR(inode->i_mode))
> 
> What I don't understand is what could possibly have NULL ->i_sb in *any*
> instance of struct inode.
> 

I can also trigger bugs with a bad inode pointer dereference in atime_needs_update: if (inode->i_flags & S_NOATIME)

I think the bug may be somewhere in the nd->depth handling (when its value is 0) in fs/namei.c:get_link(): struct saved *last = nd->stack + nd->depth - 1

Moreover, is it intentional that touch_atime() is called by get_link() even if the access (previously checked with security_file_open(), e.g. by do_last) is denied?


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 455 bytes --]