[PATCH v1] NFSD: return NFS4ERR_EXIST for a guarded OPEN of a non-regular object

[PATCH v1] NFSD: return NFS4ERR_EXIST for a guarded OPEN of a non-regular object

[Chuck Lever]

Commit db8b567ffa98 ("nfsd: (almost) always open file in
nfsd4_create_file()") moved the regular-file type check ahead of the
createmode logic. A GUARDED4, EXCLUSIVE4, or EXCLUSIVE4_1 OPEN that
finds a directory or symbolic link under the requested name now fails
with NFS4ERR_ISDIR or NFS4ERR_SYMLINK instead of NFS4ERR_EXIST.

RFC 8881 Section 18.16.3 requires NFS4ERR_EXIST when a guarded create
finds any object under that name, or when an exclusive create's
verifier does not match. The type-specific errors in Section 18.16.4
apply to opening an existing file, which only UNCHECKED4 does.

Fixes: db8b567ffa98 ("nfsd: (almost) always open file in nfsd4_create_file()")
Signed-off-by: Chuck Lever <cel@kernel.org>
---
 fs/nfsd/nfs4proc.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

Neil, I'll apply this to nfsd-testing. It should be in time to get
into nfsd-next before v7.4.


diff --git a/fs/nfsd/nfs4proc.c b/fs/nfsd/nfs4proc.c
index bb74eef43938..3a82af381a8d 100644
--- a/fs/nfsd/nfs4proc.c
+++ b/fs/nfsd/nfs4proc.c
@@ -430,7 +430,17 @@ nfsd4_create_file(struct svc_rqst *rqstp, struct svc_fh *fhp,
 				       oflags,
 				       open->op_iattr.ia_mode);
 	if (IS_ERR(open->op_filp)) {
-		status = nfserrno(PTR_ERR(open->op_filp));
+		int hosterr = PTR_ERR(open->op_filp);
+
+		if (open->op_createmode != NFS4_CREATE_UNCHECKED) {
+			switch (hosterr) {
+			case -EISDIR:
+			case -ELOOP:
+			case -EFTYPE:
+				hosterr = -EEXIST;
+			}
+		}
+		status = nfserrno(hosterr);
 		open->op_filp = NULL;
 		if (status == nfserr_noent && create_status)
 			status = create_status;
-- 
2.54.0

Re: [PATCH v1] NFSD: return NFS4ERR_EXIST for a guarded OPEN of a non-regular object

[NeilBrown]

On Mon, 07 Sep 2026, Chuck Lever wrote:
> Commit db8b567ffa98 ("nfsd: (almost) always open file in
> nfsd4_create_file()") moved the regular-file type check ahead of the
> createmode logic. A GUARDED4, EXCLUSIVE4, or EXCLUSIVE4_1 OPEN that
> finds a directory or symbolic link under the requested name now fails
> with NFS4ERR_ISDIR or NFS4ERR_SYMLINK instead of NFS4ERR_EXIST.
> 
> RFC 8881 Section 18.16.3 requires NFS4ERR_EXIST when a guarded create
> finds any object under that name, or when an exclusive create's
> verifier does not match. The type-specific errors in Section 18.16.4
> apply to opening an existing file, which only UNCHECKED4 does.
> 
> Fixes: db8b567ffa98 ("nfsd: (almost) always open file in nfsd4_create_file()")
> Signed-off-by: Chuck Lever <cel@kernel.org>
> ---
>  fs/nfsd/nfs4proc.c | 12 +++++++++++-
>  1 file changed, 11 insertions(+), 1 deletion(-)
> 
> Neil, I'll apply this to nfsd-testing. It should be in time to get
> into nfsd-next before v7.4.
> 
> 
> diff --git a/fs/nfsd/nfs4proc.c b/fs/nfsd/nfs4proc.c
> index bb74eef43938..3a82af381a8d 100644
> --- a/fs/nfsd/nfs4proc.c
> +++ b/fs/nfsd/nfs4proc.c
> @@ -430,7 +430,17 @@ nfsd4_create_file(struct svc_rqst *rqstp, struct svc_fh *fhp,
>  				       oflags,
>  				       open->op_iattr.ia_mode);
>  	if (IS_ERR(open->op_filp)) {
> -		status = nfserrno(PTR_ERR(open->op_filp));
> +		int hosterr = PTR_ERR(open->op_filp);
> +
> +		if (open->op_createmode != NFS4_CREATE_UNCHECKED) {
> +			switch (hosterr) {
> +			case -EISDIR:
> +			case -ELOOP:
> +			case -EFTYPE:
> +				hosterr = -EEXIST;
> +			}
> +		}
> +		status = nfserrno(hosterr);

Thanks but I don't think this is the best place to fix it.
I think it would be best to put the fix in do_lookup_open() so that
it's behaviour more closely matches vfs_lookup_open().

I'll send an alternate patch.

Thanks,
NeilBrown


>  		open->op_filp = NULL;
>  		if (status == nfserr_noent && create_status)
>  			status = create_status;
> -- 
> 2.54.0
> 
>