On Tue, Aug 06, 2024 at 10:05:50PM +0800, Yafang Shao wrote: > On Tue, Aug 6, 2024 at 9:24 PM Jan Kara <jack@xxxxxxx> wrote: > > On Tue 06-08-24 19:54:58, Yafang Shao wrote: > > > Its guarantee is clear: > > > > > > : I/O is intended to be atomic to ordinary files and pipes and FIFOs. > > > : Atomic means that all the bytes from a single operation that started > > > : out together end up together, without interleaving from other I/O > > > : operations. > > > > Oh, I understand why XFS does locking this way and I'm well aware this is > > a requirement in POSIX. However, as you have experienced, it has a > > significant performance cost for certain workloads (at least with simple > > locking protocol we have now) and history shows users rather want the extra > > performance at the cost of being a bit more careful in userspace. So I > > don't see any filesystem switching to XFS behavior until we have a > > performant range locking primitive. > > > > > What this flag does is avoid waiting for this type of lock if it > > > exists. Maybe we should consider a more descriptive name like > > > RWF_NOATOMICWAIT, RWF_NOFSLOCK, or RWF_NOPOSIXWAIT? Naming is always > > > challenging. > > > > Aha, OK. So you want the flag to mean "I don't care about POSIX read-write > > exclusion". I'm still not convinced the flag is a great idea but > > RWF_NOWRITEEXCLUSION could perhaps better describe the intent of the flag. > > That's better. Should we proceed with implementing this new flag? It > provides users with an option to avoid this type of issue. No. If we are going to add a flag like that, the fix to XFS isn't to use IOCB_NOWAIT on reads, it's to use shared locking for buffered writes just like we do for direct IO. IOWs, this flag would be needed on -writes-, not reads, and at that point we may as well just change XFS to do shared buffered writes for -everyone- so it is consistent with all other Linux filesystems. Indeed, last time Amir brought this up, I suggested that shared buffered write locking in XFS was the simplest way forward. Given that we use large folios now, small IOs get mapped to a single folio and so will still have the same write vs overlapping write exclusion behaviour most all the time. However, since then we've moved to using shared IO locking for cloning files. A clone does not modify data, so read IO is allowed during the clone. If we move writes to use shared locking, this breaks file cloning. We would have to move cloning back to to using exclusive locking, and that's going to cause performance and IO latency regressions for applications using clones with concurrent IO (e.g. VM image snapshots in cloud infrastruction). Hence the only viable solution to all these different competing "we need exclusive access to a range of the file whilst allowing other concurrent IO" issues is to move to range locking for IO exclusion.... -Dave. -- Dave Chinner david@xxxxxxxxxxxxx
