[PATCH V5 0/1] NFSD: offer write delegation for OPEN with OPEN4_SHARE_ACCESS only

[PATCH V5 0/1] NFSD: offer write delegation for OPEN with OPEN4_SHARE_ACCESS only

[Dai Ngo]

From RFC8881 does not explicitly state that server must grant write
delegation to OPEN with OPEN4_SHARE_ACCESS_WRITE only. However there
are text in the RFC that implies it is up to the server implementation
to offer write delegation for OPEN with OPEN4_SHARE_ACCESS_WRITE only.

Section 9.1.2:

  "In the case of READ, the server may perform the corresponding
   check on the access mode, or it may choose to allow READ for
   OPEN4_SHARE_ACCESS_WRITE, to accommodate clients whose WRITE
   implementation may unavoidably do reads (e.g., due to buffer
   cache constraints)."

Also in section 10.4.1

  "Similarly, when closing a file opened for OPEN4_SHARE_ACCESS_WRITE/
   OPEN4_SHARE_ACCESS_BOTH and if an OPEN_DELEGATE_WRITE delegation
   is in effect"

When a write delegation is granted for OPEN with OPEN4_SHARE_ACCESS_WRITE,
a new pair of nfsd_file and struct file are allocated with read access
and added to nfs4_file's fi_fds[]. This is to allow the client to use
the delegation stateid to do reads.

No additional actions is needed when the delegation is returned. The
nfsd_file for read remains attached to the nfs4_file and is freed when
the open stateid is closed.

The use of separate nfsd_file for read was suggested by Chuck.

I also tried the suggestion by Jeff which is to "acquire a R/W open from
the get-go instead when you intend to grant a delegation". To implement
this approach, in nfs4_get_vfs_file OPEN4_SHARE_ACCESS_READ was added to
op_share_access before computing the 'oflag' and 'access' flag. This
forces the rest of the code to process the OPEN as if it was sent with
access mode OPEN4_SHARE_ACCESS_WRITE|OPEN4_SHARE_ACCESS_READ.

This mostly works except a case where the file was created with 0222
permission and the user credential of OPEN is not the same as the owner
of the file. In this case the OPEN fails with NFS4ERR_ACCESS. This is
because nfsd_permission was called with (MAY_READ | MAY_WRITE) mask
and the file permission is 0222.

I don't see any simple fix for this nfsd_permission issue. Basically
the access mode has to be rdwr when creating the nfsd_file but the access
mode passed to nfsd_permission to check should be rdonly (the original
OPEN's op_share_access).

v4: reacted to Jeff's comment
v5: merged both patches into one to avoid potential bisect problem.
    fixed some nits from Jeff's comments.

 fs/nfsd/nfs4state.c | 76 ++++++++++++++++++++++++++++++------------------
 1 file changed, 48 insertions(+), 28 deletions(-)

[PATCH V5 1/1] NFSD: Offer write delegation for OPEN with OPEN4_SHARE_ACCESS_WRITE

[Dai Ngo]

RFC8881, section 9.1.2 says:

  "In the case of READ, the server may perform the corresponding
   check on the access mode, or it may choose to allow READ for
   OPEN4_SHARE_ACCESS_WRITE, to accommodate clients whose WRITE
   implementation may unavoidably do reads (e.g., due to buffer
   constraints)."

and in section 10.4.1:
   "Similarly, when closing a file opened for OPEN4_SHARE_ACCESS_WRITE/
   OPEN4_SHARE_ACCESS_BOTH and if an OPEN_DELEGATE_WRITE delegation
   is in effect"

This patch allow READ using write delegation stateid granted on OPENs
with OPEN4_SHARE_ACCESS_WRITE only, to accommodate clients whose WRITE
implementation may unavoidably do (e.g., due to buffer cache
constraints).

For write delegation granted for OPEN with OPEN4_SHARE_ACCESS_WRITE
a new nfsd_file and a struct file are allocated to use for reads.
The nfsd_file is freed when the file is closed by release_all_access.

Suggested-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Dai Ngo <dai.ngo@oracle.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
---
 fs/nfsd/nfs4state.c | 76 ++++++++++++++++++++++++++++-----------------
 1 file changed, 48 insertions(+), 28 deletions(-)

diff --git a/fs/nfsd/nfs4state.c b/fs/nfsd/nfs4state.c
index 0f97f2c62b3a..4c5338c50f0c 100644
--- a/fs/nfsd/nfs4state.c
+++ b/fs/nfsd/nfs4state.c
@@ -633,18 +633,6 @@ find_readable_file(struct nfs4_file *f)
 	return ret;
 }
 
-static struct nfsd_file *
-find_rw_file(struct nfs4_file *f)
-{
-	struct nfsd_file *ret;
-
-	spin_lock(&f->fi_lock);
-	ret = nfsd_file_get(f->fi_fds[O_RDWR]);
-	spin_unlock(&f->fi_lock);
-
-	return ret;
-}
-
 struct nfsd_file *
 find_any_file(struct nfs4_file *f)
 {
@@ -5987,14 +5975,19 @@ nfs4_set_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
 	 *  "An OPEN_DELEGATE_WRITE delegation allows the client to handle,
 	 *   on its own, all opens."
 	 *
-	 * Furthermore the client can use a write delegation for most READ
-	 * operations as well, so we require a O_RDWR file here.
+	 * Furthermore, section 9.1.2 says:
+	 *
+	 *  "In the case of READ, the server may perform the corresponding
+	 *  check on the access mode, or it may choose to allow READ for
+	 *  OPEN4_SHARE_ACCESS_WRITE, to accommodate clients whose WRITE
+	 *  implementation may unavoidably do reads (e.g., due to buffer
+	 *  cache constraints)."
 	 *
-	 * Offer a write delegation in the case of a BOTH open, and ensure
-	 * we get the O_RDWR descriptor.
+	 *  We choose to offer a write delegation for OPEN with the
+	 *  OPEN4_SHARE_ACCESS_WRITE access mode to accommodate such clients.
 	 */
-	if ((open->op_share_access & NFS4_SHARE_ACCESS_BOTH) == NFS4_SHARE_ACCESS_BOTH) {
-		nf = find_rw_file(fp);
+	if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE) {
+		nf = find_writeable_file(fp);
 		dl_type = deleg_ts ? OPEN_DELEGATE_WRITE_ATTRS_DELEG : OPEN_DELEGATE_WRITE;
 	}
 
@@ -6116,7 +6109,7 @@ static bool
 nfs4_delegation_stat(struct nfs4_delegation *dp, struct svc_fh *currentfh,
 		     struct kstat *stat)
 {
-	struct nfsd_file *nf = find_rw_file(dp->dl_stid.sc_file);
+	struct nfsd_file *nf = find_writeable_file(dp->dl_stid.sc_file);
 	struct path path;
 	int rc;
 
@@ -6134,6 +6127,34 @@ nfs4_delegation_stat(struct nfs4_delegation *dp, struct svc_fh *currentfh,
 	return rc == 0;
 }
 
+/*
+ * Add NFS4_SHARE_ACCESS_READ to the write delegation granted on OPEN
+ * with NFS4_SHARE_ACCESS_WRITE by allocating separate nfsd_file and
+ * struct file to be used for read with delegation stateid.
+ *
+ */
+static bool
+nfsd4_add_rdaccess_to_wrdeleg(struct svc_rqst *rqstp, struct nfsd4_open *open,
+			      struct svc_fh *fh, struct nfs4_ol_stateid *stp)
+{
+	struct nfs4_file *fp;
+	struct nfsd_file *nf = NULL;
+
+	if ((open->op_share_access & NFS4_SHARE_ACCESS_BOTH) ==
+			NFS4_SHARE_ACCESS_WRITE) {
+		if (nfsd_file_acquire_opened(rqstp, fh, NFSD_MAY_READ, NULL, &nf))
+			return (false);
+		fp = stp->st_stid.sc_file;
+		spin_lock(&fp->fi_lock);
+		__nfs4_file_get_access(fp, NFS4_SHARE_ACCESS_READ);
+		set_access(NFS4_SHARE_ACCESS_READ, stp);
+		fp = stp->st_stid.sc_file;
+		fp->fi_fds[O_RDONLY] = nf;
+		spin_unlock(&fp->fi_lock);
+	}
+	return true;
+}
+
 /*
  * The Linux NFS server does not offer write delegations to NFSv4.0
  * clients in order to avoid conflicts between write delegations and
@@ -6159,8 +6180,9 @@ nfs4_delegation_stat(struct nfs4_delegation *dp, struct svc_fh *currentfh,
  * open or lock state.
  */
 static void
-nfs4_open_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
-		     struct svc_fh *currentfh)
+nfs4_open_delegation(struct svc_rqst *rqstp, struct nfsd4_open *open,
+		     struct nfs4_ol_stateid *stp, struct svc_fh *currentfh,
+		     struct svc_fh *fh)
 {
 	bool deleg_ts = open->op_deleg_want & OPEN4_SHARE_ACCESS_WANT_DELEG_TIMESTAMPS;
 	struct nfs4_openowner *oo = openowner(stp->st_stateowner);
@@ -6205,7 +6227,8 @@ nfs4_open_delegation(struct nfsd4_open *open, struct nfs4_ol_stateid *stp,
 	memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
 
 	if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE) {
-		if (!nfs4_delegation_stat(dp, currentfh, &stat)) {
+		if (!nfsd4_add_rdaccess_to_wrdeleg(rqstp, open, fh, stp) ||
+				!nfs4_delegation_stat(dp, currentfh, &stat)) {
 			nfs4_put_stid(&dp->dl_stid);
 			destroy_delegation(dp);
 			goto out_no_deleg;
@@ -6361,7 +6384,8 @@ nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nf
 	* Attempt to hand out a delegation. No error return, because the
 	* OPEN succeeds even if we fail.
 	*/
-	nfs4_open_delegation(open, stp, &resp->cstate.current_fh);
+	nfs4_open_delegation(rqstp, open, stp,
+		&resp->cstate.current_fh, current_fh);
 
 	/*
 	 * If there is an existing open stateid, it must be updated and
@@ -7063,7 +7087,7 @@ nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
 		return_revoked = true;
 	if (typemask & SC_TYPE_DELEG)
 		/* Always allow REVOKED for DELEG so we can
-		 * retturn the appropriate error.
+		 * return the appropriate error.
 		 */
 		statusmask |= SC_STATUS_REVOKED;
 
@@ -7106,10 +7130,6 @@ nfs4_find_file(struct nfs4_stid *s, int flags)
 
 	switch (s->sc_type) {
 	case SC_TYPE_DELEG:
-		spin_lock(&s->sc_file->fi_lock);
-		ret = nfsd_file_get(s->sc_file->fi_deleg_file);
-		spin_unlock(&s->sc_file->fi_lock);
-		break;
 	case SC_TYPE_OPEN:
 	case SC_TYPE_LOCK:
 		if (flags & RD_STATE)
-- 
2.43.5

Re: [PATCH V5 0/1] NFSD: offer write delegation for OPEN with OPEN4_SHARE_ACCESS only

[cel]

From: Chuck Lever <chuck.lever@oracle.com>

On Wed, 05 Mar 2025 14:05:07 -0800, Dai Ngo wrote:
> >From RFC8881 does not explicitly state that server must grant write
> delegation to OPEN with OPEN4_SHARE_ACCESS_WRITE only. However there
> are text in the RFC that implies it is up to the server implementation
> to offer write delegation for OPEN with OPEN4_SHARE_ACCESS_WRITE only.
> 
> Section 9.1.2:
> 
> [...]

Applied v5 to nfsd-testing, thanks!

v4 has been reverted.

[1/1] NFSD: Offer write delegation for OPEN with OPEN4_SHARE_ACCESS_WRITE
      commit: 70b39d56ded1c6ae51757b182047153cbb6522d2

--
Chuck Lever