On Mon, Apr 19, 2021 at 04:20:08PM +0100, Matthew Wilcox wrote: > On Tue, Apr 13, 2021 at 01:28:44PM +0200, Jan Kara wrote: > > Also when writing the documentation I came across one question: Do we mandate > > i_mapping_sem for truncate + hole punch for all filesystems or just for > > filesystems that support hole punching (or other complex fallocate operations)? > > I wrote the documentation so that we require every filesystem to use > > i_mapping_sem. This makes locking rules simpler, we can also add asserts when > > all filesystems are converted. The downside is that simple filesystems now pay > > the overhead of the locking unnecessary for them. The overhead is small > > (uncontended rwsem acquisition for truncate) so I don't think we care and the > > simplicity is worth it but I wanted to spell this out. > > What do we actually get in return for supporting these complex fallocate > operations? Someone added them for a reason, but does that reason > actually benefit me? Other than running xfstests, how many times has > holepunch been called on your laptop in the last week? Quite a lot, actually. > I don't want to > incur even one extra instruction per I/O operation to support something > that happens twice a week; that's a bad tradeoff. Hole punching gets into all sorts of interesting places. For example, did you know that issuing fstrim (discards) or "write zeroes" on a file-backed loopback device will issue hole punches to the underlying file? nvmet does the same. Userspace iscsi server implementations (e.g. TGT) do the same thing and have for a long time. NFSv4 servers issue hole punching based on client side requests, too. Then there's Kubernetes management tools. Samba. Qemu. Libvirt. Mysql. Network-Manager. Gluster. Chromium. RocksDB. Swift. Systemd. The list of core system infrastructure we have that uses hole punching is quite large... So, really, hole punching is something that happens a lot and in many unexpected places. You can argue that your laptop doesn't use it, but that really doesn't matter in the bigger scheme of things. Hole punching is something applications expect to work and not corrupt data.... > Can we implement holepunch as a NOP? Or return -ENOTTY? Those both > seem like better solutions than adding an extra rwsem to every inode. We've already added this extra i_rwsem to ext4 and XFS - it's a sunk cost for almost every production machine out there in the wild. It needs to be made generic so we can optimise the implementation and not have to implement a unicorn in every filesystem to work around the fact the page cache and page faults have no internal serialisation mechanism against filesystem operations that directly manipulate and invalidate large ranges of the backing storage the page cache sits over. > Failing that, is there a bigger hammer we can use on the holepunch side > (eg preventing all concurrent accesses while the holepunch is happening) > to reduce the overhead on the read side? That's what we currently do and what Jan is trying to refine.... Cheers, Dave. -- Dave Chinner david@xxxxxxxxxxxxx
