Re: [PATCH v4 1/2] hugetlb: use f_mode & FMODE_HUGETLBFS to identify hugetlbfs files

[Mike Kravetz <mike.kravetz@oracle.com>]

On 6/12/20 11:53 PM, Amir Goldstein wrote:
>>> Incidentally, can a hugetlbfs be a lower layer, while the upper one
>>> is a normal filesystem?  What should happen on copyup?
>>
>> Yes, that seems to work as expected.  When accessed for write the hugetlb
>> file is copied to the normal filesystem.
>>
>> The BUG found by syzbot actually has a single hugetlbfs as both lower and
>> upper.  With the BUG 'fixed', I am not exactly sure what the expected
>> behavior is in this case.  I may be wrong, but I would expect any operations
>> that can be performed on a stand alone hugetlbfs to also be performed on
>> the overlay.  However, mmap() still fails.  I will look into it.
>>
>> I also looked at normal filesystem lower and hugetlbfs upper.  Yes, overlayfs
>> allows this.  This is somewhat 'interesting' as write() is not supported in
>> hugetlbfs.  Writing to files in the overlay actually ended up writing to
>> files in the lower filesystem.  That seems wrong, but overlayfs is new to me.
>>
> 
> I am not sure how that happened, but I think that ovl_open_realfile()
> needs to fixup f_mode flags FMODE_CAN_WRITE | FMODE_CAN_READ
> after open_with_fake_path().
> 
>> Earlier in the discussion of these issues, Colin Walters asked "Is there any
>> actual valid use case for mounting an overlayfs on top of hugetlbfs?"  I can
>> not think of one.  Perhaps we should consider limiting the ways in which
>> hugetlbfs can be used in overlayfs?  Preventing it from being an upper
>> filesystem might be a good start?  Or, do people think making hugetlbfs and
>> overlayfs play nice together is useful?
> 
> If people think that making hugetlbfs and overlayfs play nice together maybe
> they should work on this problem. It doesn't look like either
> hugetlbfs developers
> nor overlayfs developers care much about the combination.

Thanks Amir,

As a hugetlbfs developer, I do not know of a use case for interoperability
with overlayfs.  So yes, I am not too interested in making them work well
together.  However, if there was an actual use case I would be more than
happy to consider doing the work.  Just hate to put effort into fixing up
two 'special' filesystems for functionality that may not be used.

I can't speak for overlayfs developers.

> Your concern, I assume, is fixing the syzbot issue.

That is the primary concern.  We should not BUG!  After fixing that up, Al
asked how these things worked together.  I honestly did not look at
interoperability before that.  I am not sure if anyone has done that in the
past.

> I agree with Colin's remark about adding limitations, but it would be a shame
> if overlay had to special case hugetlbfs. It would have been better if we could
> find a property of hugetlbfs that makes it inapplicable for overlayfs
> upper/lower
> or stacking fs in general.
> 
> The simplest thing for you to do in order to shush syzbot is what procfs does:
>         /*
>          * procfs isn't actually a stacking filesystem; however, there is
>          * too much magic going on inside it to permit stacking things on
>          * top of it
>          */
>         s->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
> 
> Currently, the only in-tree stacking fs are overlayfs and ecryptfs, but there
> are some out of tree implementations as well (shiftfs).
> So you may only take that option if you do not care about the combination
> of hugetlbfs with any of the above.
> 
> overlayfs support of mmap is not as good as one might hope.
> overlayfs.rst says:
> "If a file residing on a lower layer is opened for read-only and then
>  memory mapped with MAP_SHARED, then subsequent changes to
>  the file are not reflected in the memory mapping."
> 
> So if I were you, I wouldn't go trying to fix overlayfs-huguetlb interop...

Thanks again,

I'll look at something as simple as s_stack_depth.
-- 
Mike Kravetz