Re: [linux-next:master] [VFS/nfsd/cachefiles/ovl] 7ab96df840: WARNING:at_fs/dcache.c:#umount_check

[Oliver Sang <oliver.sang@intel.com>]

hi, NeilBrown,

On Thu, Nov 27, 2025 at 07:51:18AM +1100, NeilBrown wrote:
> On Wed, 26 Nov 2025, Amir Goldstein wrote:
> > On Wed, Nov 26, 2025 at 11:42 AM Christian Brauner <brauner@kernel.org> wrote:
> > >
> > > On Tue, Nov 25, 2025 at 09:48:18PM +0800, kernel test robot wrote:
> > > >
> > > > Hello,
> > > >
> > > > kernel test robot noticed "WARNING:at_fs/dcache.c:#umount_check" on:
> > > >
> > > > commit: 7ab96df840e60eb933abfe65fc5fe44e72f16dc0 ("VFS/nfsd/cachefiles/ovl: add start_creating() and end_creating()")
> > > > https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git master
> > > >
> > > > [test failed on linux-next/master d724c6f85e80a23ed46b7ebc6e38b527c09d64f5]
> > >
> > > Neil, can you please take a look at this soon?
> > > I plan on sending the batch of PRs for this cycle on Friday.
> > >
> > > >
> > > > in testcase: filebench
> > > > version: filebench-x86_64-22620e6-1_20251009
> > > > with following parameters:
> > > >
> > > >       disk: 1SSD
> > > >       fs: ext4
> > > >       fs2: nfsv4
> > > >       test: ratelimcopyfiles.f
> > > >       cpufreq_governor: performance
> > > >
> > 
> > Test is copying to nfsv4 so that's the immediate suspect.
> > WARN_ON is in unmount of ext4, but I suspect that nfs
> > was loop mounted for the test.
> > 
> > FWIW, nfsd_proc_create() looks very suspicious.
> > 
> > nfsd_create_locked() does end_creating() internally (internal API change)
> > but nfsd_create_locked() still does end_creating() regardless.
> 
> Thanks for looking at this Amir.  That omission in nfsproc.c is
> certainly part of the problem but not all of it.
> By skipping the end_creating() there, we avoid a duplicate unlock, but
> also lose a dput() which we need.  Both callers of nfsd_create_locked()
> have the same problem.
> I think this should fix it.  The resulting code is a bit ugly but I can
> fix that with the nfsd team once this gets upstream.
> 
> (FYI nfsd_proc_create() is only used for NFSv2 and as it was an nfsv4 test,
>  that could wouldn't have been run)
> 
> Thanks,
> NeilBrown

we tested below patch, confirmed it could fix the issues in our original
reports. thanks!

Tested-by: kernel test robot <oliver.sang@intel.com>

> 
> diff --git a/fs/nfsd/nfsproc.c b/fs/nfsd/nfsproc.c
> index 28f03a6a3cc3..481e789a7697 100644
> --- a/fs/nfsd/nfsproc.c
> +++ b/fs/nfsd/nfsproc.c
> @@ -407,6 +407,9 @@ nfsd_proc_create(struct svc_rqst *rqstp)
>  		/* File doesn't exist. Create it and set attrs */
>  		resp->status = nfsd_create_locked(rqstp, dirfhp, &attrs, type,
>  						  rdev, newfhp);
> +		/* nfsd_create_locked() unlocked the parent */
> +		dput(dchild);
> +		goto out_write;
>  	} else if (type == S_IFREG) {
>  		dprintk("nfsd:   existing %s, valid=%x, size=%ld\n",
>  			argp->name, attr->ia_valid, (long) attr->ia_size);
> diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c
> index 145f1c8d124d..4688f3fd59e2 100644
> --- a/fs/nfsd/vfs.c
> +++ b/fs/nfsd/vfs.c
> @@ -1633,16 +1633,14 @@ nfsd_create(struct svc_rqst *rqstp, struct svc_fh *fhp,
>  		return nfserrno(host_err);
>  
>  	err = fh_compose(resfhp, fhp->fh_export, dchild, fhp);
> -	/*
> -	 * We unconditionally drop our ref to dchild as fh_compose will have
> -	 * already grabbed its own ref for it.
> -	 */
>  	if (err)
>  		goto out_unlock;
>  	err = fh_fill_pre_attrs(fhp);
>  	if (err != nfs_ok)
>  		goto out_unlock;
>  	err = nfsd_create_locked(rqstp, fhp, attrs, type, rdev, resfhp);
> +	/* nfsd_create_locked() unlocked the parent */
> +	dput(dchild);
>  	return err;
>  
>  out_unlock: