RE: [PATCH 5/5] ceph: Fix write storm on fscrypted files

[Viacheslav Dubeyko <Slava.Dubeyko@ibm.com>]

On Mon, 2026-01-05 at 22:53 -0800, Sam Edwards wrote:
> On Mon, Jan 5, 2026 at 2:34 PM Viacheslav Dubeyko <Slava.Dubeyko@ibm.com> wrote:
> > 
> > On Tue, 2025-12-30 at 18:43 -0800, Sam Edwards wrote:
> > > CephFS stores file data across multiple RADOS objects. An object is the
> > > atomic unit of storage, so the writeback code must clean only folios
> > > that belong to the same object with each OSD request.
> > > 
> > > CephFS also supports RAID0-style striping of file contents: if enabled,
> > > each object stores multiple unbroken "stripe units" covering different
> > > portions of the file; if disabled, a "stripe unit" is simply the whole
> > > object. The stripe unit is (usually) reported as the inode's block size.
> > > 
> > > Though the writeback logic could, in principle, lock all dirty folios
> > > belonging to the same object, its current design is to lock only a
> > > single stripe unit at a time. Ever since this code was first written,
> > > it has determined this size by checking the inode's block size.
> > > However, the relatively-new fscrypt support needed to reduce the block
> > > size for encrypted inodes to the crypto block size (see 'fixes' commit),
> > > which causes an unnecessarily high number of write operations (~1024x as
> > > many, with 4MiB objects) and grossly degraded performance.
> 
> Hi Slava,
> 
> > Do you have any benchmarking results that prove your point?
> 
> I haven't done any "real" benchmarking for this change. On my setup
> (closer to a home server than a typical Ceph deployment), sequential
> write throughput increased from ~1.7 to ~66 MB/s with this patch
> applied. I don't consider this single datapoint representative, so
> rather than presenting it as a general benchmark in the commit
> message, I chose the qualitative wording "grossly degraded
> performance." Actual impact will vary depending on workload, disk
> type, OSD count, etc.
> 
> Those curious about the bug's performance impact in their environment
> can find out without enabling fscrypt, using: mount -o wsize=4096
> 
> However, the core rationale for my claim is based on principles, not
> on measurements: batching writes into fewer operations necessarily
> spreads per-operation overhead across more bytes. So this change
> removes an artificial per-op bottleneck on sequential write
> performance. The exact impact varies, but the patch does improve
> (fscrypt-enabled) write throughput in nearly every case.
> 

If you claim in commit message that "this patch fixes performance degradation",
then you MUST share the evidence (benchmarking results). Otherwise, you cannot
make such statements in commit message. Yes, it could be a good fix but please
don't make a promise of performance improvement. Because, end-users have very
different environments and workloads. And what could work on your side may not
work for other use-cases and environments. Potentially, you could describe your
environment, workload and to share your estimation/expectation of potential
performance improvement.

Thanks,
Slava.

> Warm regards,
> Sam
> 
> 
> > 
> > Thanks,
> > Slava.
> > 
> > > 
> > > Fix this (and clarify intent) by using i_layout.stripe_unit directly in
> > > ceph_define_write_size() so that encrypted inodes are written back with
> > > the same number of operations as if they were unencrypted.
> > > 
> > > Fixes: 94af0470924c ("ceph: add some fscrypt guardrails")
> > > Cc: stable@vger.kernel.org
> > > Signed-off-by: Sam Edwards <CFSworks@gmail.com>
> > > ---
> > >  fs/ceph/addr.c | 3 ++-
> > >  1 file changed, 2 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
> > > index b3569d44d510..cb1da8e27c2b 100644
> > > --- a/fs/ceph/addr.c
> > > +++ b/fs/ceph/addr.c
> > > @@ -1000,7 +1000,8 @@ unsigned int ceph_define_write_size(struct address_space *mapping)
> > >  {
> > >       struct inode *inode = mapping->host;
> > >       struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
> > > -     unsigned int wsize = i_blocksize(inode);
> > > +     struct ceph_inode_info *ci = ceph_inode(inode);
> > > +     unsigned int wsize = ci->i_layout.stripe_unit;
> > > 
> > >       if (fsc->mount_options->wsize < wsize)
> > >               wsize = fsc->mount_options->wsize;