Re: fs: out of bounds on stack in iov_iter_advance

[Jens Axboe <axboe@kernel.dk>]

On 11/10/2015 07:41 PM, Jens Axboe wrote:
> On 11/10/2015 07:40 PM, Jens Axboe wrote:
>> On 11/10/2015 07:31 PM, Linus Torvalds wrote:
>>> On Tue, Nov 10, 2015 at 6:25 PM, Jens Axboe <axboe@kernel.dk> wrote:
>>>> On Tue, Nov 10 2015, Linus Torvalds wrote:
>>>>> Al, ping?
>>>>>
>>>>> On Thu, Nov 5, 2015 at 7:38 PM, Linus Torvalds
>>>>> <torvalds@linux-foundation.org> wrote:
>>>>>> On Thu, Nov 5, 2015 at 6:19 PM, Al Viro <viro@zeniv.linux.org.uk>
>>>>>> wrote:
>>>>>>>
>>>>>>> How are we going to handle that one?  I can put it into mainline
>>>>>>> pull
>>>>>>> request via vfs.git, with Cc: stable, but if e.g. Jens prefers to
>>>>>>> take it
>>>>>>> via the block tree, I'll be glad to leave it for him to deal with.
>>>>>>
>>>>>> Put it in the vfs tree (I'm hoping for a pull request soon..)
>>>>>>
>>>>>> I pulled the block trees from Jens yesterday, so there is presumably
>>>>>> nothing pending there right now.
>>>>>
>>>>> Apparently my "hoping for a pull request soon" was ridiculously
>>>>> optimistic.
>>>>>
>>>>> Al, looking at the most recent linux-next, most of the vfs commits
>>>>> there seem to be committed in the last day or two. I'm getting the
>>>>> feeling that that is all 4.5 material by now.
>>>>>
>>>>> Should I just take the iov patch as-is, since apparently no vfs pull
>>>>> request is happening this merge cycle? And no, I'm not taking
>>>>> "developed during the second week of the merge window, and sent in the
>>>>> last few days of it". I'm done with that.
>>>>
>>>> I've got 8 other patches pending for a post core merge, just waiting
>>>> for
>>>> the last core pull request to go in. I haven't seen this iov iter fix,
>>>> though.
>>>
>>> It was in this thread, looked like this (without the whitespace damage):
>>>
>>>      dax_io(): don't let non-error value escape via retval instead of
>>> EFAULT
>>>
>>>      Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
>>>      ---
>>>      diff --git a/fs/dax.c b/fs/dax.c
>>>      index a86d3cc..7b653e9 100644
>>>      --- a/fs/dax.c
>>>      +++ b/fs/dax.c
>>>      @@ -169,8 +169,10 @@ static ssize_t dax_io(struct inode *inode,
>>> struct iov_iter *iter,
>>>                      else
>>>                              len = iov_iter_zero(max - pos, iter);
>>>
>>>      -               if (!len)
>>>      +               if (!len) {
>>>      +                       retval = -EFAULT;
>>>                              break;
>>>      +               }
>>>
>>>                      pos += len;
>>>                      addr += len;
>>>
>>>
>>> although I don't think I saw a confirmation that that was what Sasha
>>> actually hit (but Sasha had narrowed it down to DAX, so it looks
>>> possible/likely)
>>
>> I found it right after sending that email. Patch looks pretty straight
>> forward, at least from the case of max - pos != 0 and len == 0 on
>> return. Might be cleaner to add a
>>
>> if (retval < 0)
>>      break;
>>
>> check, that should be the case where max == pos anyway. But we'd
>> potentially return -Exx into -EFAULT for that case with the patch.
>>
>> Hmm?
>
> So we already do that, in the 'if' above. I think the patch looks fine.

Queued up. Unless Al objects, it'll be part of the 'for-linus' pull 
later this week.

-- 
Jens Axboe