Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: Re: kernel BUG at mm/hugetlb.c:LINE!
To:     Miklos Szeredi <miklos@szeredi.hu>
Cc:     syzbot <syzbot+d6ec23007e951dadf3de@syzkaller.appspotmail.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org,
        linux-mm@kvack.org, Miklos Szeredi <mszeredi@redhat.com>,
        syzkaller-bugs <syzkaller-bugs@googlegroups.com>,
        Al Viro <viro@zeniv.linux.org.uk>
References: <000000000000b4684e05a2968ca6@google.com>
 <aa7812b8-60ae-8578-40db-e71ad766b4d3@oracle.com>
 <CAJfpegtVca6H1JPW00OF-7sCwpomMCo=A2qr5K=9uGKEGjEp3w@mail.gmail.com>
From:   Mike Kravetz <mike.kravetz@oracle.com>
Message-ID: <d32b8579-04a3-2a9b-cd54-1d581c63332e@oracle.com>
Date:   Tue, 12 May 2020 11:11:18 -0700
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:68.0) Gecko/20100101
 Thunderbird/68.7.0
MIME-Version: 1.0
In-Reply-To: <CAJfpegtVca6H1JPW00OF-7sCwpomMCo=A2qr5K=9uGKEGjEp3w@mail.gmail.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 5/12/20 8:04 AM, Miklos Szeredi wrote:
> On Tue, Apr 7, 2020 at 12:06 AM Mike Kravetz <mike.kravetz@oracle.com> wrote:
>> On 4/5/20 8:06 PM, syzbot wrote:
>>
>> The routine is_file_hugepages() is just comparing the file ops to huegtlbfs:
>>
>>         if (file->f_op == &hugetlbfs_file_operations)
>>                 return true;
>>
>> Since the file is in an overlayfs, file->f_op == ovl_file_operations.
>> Therefore, length will not be rounded up to huge page size and we create a
>> mapping with incorrect size which leads to the BUG.
>>
>> Because of the code in mmap, the hugetlbfs mmap() routine assumes length is
>> rounded to a huge page size.  I can easily add a check to hugetlbfs mmap
>> to validate length and return -EINVAL.  However, I think we really want to
>> do the 'round up' earlier in mmap.  This is because the man page says:
>>
>>    Huge page (Huge TLB) mappings
>>        For mappings that employ huge pages, the requirements for the arguments
>>        of  mmap()  and munmap() differ somewhat from the requirements for map‐
>>        pings that use the native system page size.
>>
>>        For mmap(), offset must be a multiple of the underlying huge page size.
>>        The system automatically aligns length to be a multiple of the underly‐
>>        ing huge page size.
>>
>> Since the location for the mapping is chosen BEFORE getting to the hugetlbfs
>> mmap routine, we can not wait until then to round up the length.  Is there a
>> defined way to go from a struct file * to the underlying filesystem so we
>> can continue to do the 'round up' in early mmap code?
> 
> That's easy enough:
> 
> static inline struct file *real_file(struct file *file)
> {
>     return file->f_op != ovl_file_operations ? file : file->private_data;
> }
> 
> But adding more filesystem specific code to generic code does not
> sound like the cleanest way to solve this...

We can incorporate the above 'real_file' functionality in the filesystem
specific routine is_file_hugepages(), and I think that would address this
specific issue.  I'll code that up.

>> One other thing I noticed with overlayfs is that it does not contain a
>> specific get_unmapped_area file_operations routine.  I would expect it to at
>> least check for and use the get_unmapped_area of the underlying filesystem?
>> Can someone comment if this is by design?
> 
> Not sure.  What exactly is f_op->get_unmapped_area supposed to do?
> 

IIUC, filesystems can define their own routines to get addresses for mmap
operations.  Quite a few filesystems define get_unmapped_area.

The generic mmap code does the following,

	get_area = current->mm->get_unmapped_area;
	if (file) {
		if (file->f_op->get_unmapped_area)
			get_area = file->f_op->get_unmapped_area;
	} else if (flags & MAP_SHARED) {
		/*
		 * mmap_region() will call shmem_zero_setup() to create a file,
		 * so use shmem's get_unmapped_area in case it can be huge.
		 * do_mmap_pgoff() will clear pgoff, so match alignment.
		 */
		pgoff = 0;
		get_area = shmem_get_unmapped_area;
	}

	addr = get_area(file, addr, len, pgoff, flags);

If the filesystem provides a get_unmapped_area, it will use it.  I beleive
overlayfs prevents this from happening for the underlying filesystem.

Perhaps we do need to add something like a call 'real_file' to this generic
code?  I can't think of any other way to get to the underlying filesystem
get_unmapped_area here.
-- 
Mike Kravetz