Hello,

On Fri 02-02-18 15:04:11, Liu Bo wrote:

On Thu, Jan 25, 2018 at 12:57:27PM +0100, Jan Kara wrote:

quoted

Hello,

this is about a problem I have identified last month and for which I still
don't have good solution. Some discussion of the problem happened here [1]
where also technical details are posted but culprit of the problem is
relatively simple: Lots of places in kernel (fs code, writeback logic,
stable-pages framework for DIF/DIX) assume that file pages in page cache
can be modified either via write(2), truncate(2), fallocate(2) or similar
code paths explicitely manipulating with file space or via a writeable
mapping into page tables. In particular we assume that if we block all the
above paths by taking proper locks, block page faults, and unmap (/ map
read-only) the page, it cannot be modified. But this assumption is violated
by get_user_pages() users (such as direct IO or RDMA drivers - and we've
got reports from such users of weird things happening).

The problem with GUP users is that they acquire page reference (at that
point page is writeably mapped into page tables) and some time in future
(which can be quite far in case of RDMA) page contents gets modified and
page marked dirty.

I got a question here, when you say 'page contents gets modified', do
you mean that GUP users modify the page content?

Yes.

I have another story about GUP users who use direct-IO, qemu sometimes
doesn't work well with btrfs when checksum enabled and reports
checksum failures when guest OS doesn't use stable pages, where it is
not GUP users but the original file/mapping that may be changing the
page content in flight.

OK, but that is kind of expected, isn't it? The whole purpose of 'stable
pages' is exactly to modifying pages while IO is in flight. So if a device
image is backed by a storage (filesystem in this case) which checksums
data, qemu should present it to the guest as a block device supporting
DIF/DIX and thus requiring stable pages...

								Honza
-- 
Jan Kara [off-list ref]
SUSE Labs, CR