[PATCH v2 0/4] NFS: Add directory delegation support for CB_NOTIFY

[PATCH v2 0/4] NFS: Add directory delegation support for CB_NOTIFY

[Anna Schumaker]

From: Anna Schumaker <anna.schumaker@hammerspace.com>

These patches add support for the CB_NOTIFY callback operation from NFS
servers. Editing directory order or placement in our caches isn't easy
or practical the way things are structured now. Instead, I use CB_NOTIFY
as an indication that the client needs to refresh its view of the
directory.

Changes in v2:
 * Fix signed / unsigned type mixups
 * Don't try to share lookup code between NOTIFY4_REMOVE and NOTIFY4_ADD
 * Add handling for the NOTIFY4_ADD 'na_have_old_entry' case

Thoughts?
Anna


Anna Schumaker (4):
  NFS: Add an outline for the CB_NOTIFY operation
  NFS: Add support for CB_NOTIFY4_REMOVE_ENTRY
  NFS: Add support for CB_NOTIFY4_ADD_ENTRY
  NFS: Add support for CB_NOTIFY4_RENAME_ENTRY

 fs/nfs/callback.h      |  51 ++++++++++++
 fs/nfs/callback_proc.c | 147 ++++++++++++++++++++++++++++++++++
 fs/nfs/callback_xdr.c  | 175 ++++++++++++++++++++++++++++++++++++++++-
 fs/nfs/nfs4_fs.h       |   2 +
 fs/nfs/nfs4xdr.c       |  33 +++++++-
 include/linux/nfs4.h   |   6 ++
 6 files changed, 409 insertions(+), 5 deletions(-)

-- 
2.55.0

[PATCH v2 1/4] NFS: Add an outline for the CB_NOTIFY operation

[Anna Schumaker]

From: Anna Schumaker <anna.schumaker@oracle.com>

Support for individual notification types will be added over the next
few patches.

Signed-off-by: Anna Schumaker <anna.schumaker@oracle.com>
Signed-off-by: Anna Schumaker <anna.schumaker@hammerspace.com>
---
v2: Change res to a ssize_t
---
 fs/nfs/callback.h      | 14 +++++++++++
 fs/nfs/callback_proc.c |  9 +++++++
 fs/nfs/callback_xdr.c  | 54 +++++++++++++++++++++++++++++++++++++++++-
 fs/nfs/nfs4xdr.c       |  2 +-
 include/linux/nfs4.h   |  3 +++
 5 files changed, 80 insertions(+), 2 deletions(-)

diff --git a/fs/nfs/callback.h b/fs/nfs/callback.h
index 2a721c422d48..f7cc5b6931bf 100644
--- a/fs/nfs/callback.h
+++ b/fs/nfs/callback.h
@@ -143,6 +143,20 @@ struct cb_layoutrecallargs {
 extern __be32 nfs4_callback_layoutrecall(void *argp, void *resp,
 		struct cb_process_state *cps);
 
+struct cb_notify_changes {
+	u32 notify_mask;
+};
+
+struct cb_notifyargs {
+	nfs4_stateid			cna_stateid;
+	struct nfs_fh			cna_fh;
+	uint32_t			cna_n_changes;
+	struct cb_notify_changes	*cna_changes;
+};
+
+extern __be32 nfs4_callback_notify(void *argp, void *resp,
+				   struct cb_process_state *cps);
+
 struct cb_devicenotifyitem {
 	uint32_t		cbd_notify_type;
 	uint32_t		cbd_layout_type;
diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c
index 3fb10c8e4271..01600d75f2d9 100644
--- a/fs/nfs/callback_proc.c
+++ b/fs/nfs/callback_proc.c
@@ -369,6 +369,15 @@ static void pnfs_recall_all_layouts(struct nfs_client *clp,
 	do_callback_layoutrecall(clp, &args, cps);
 }
 
+__be32 nfs4_callback_notify(void *argp, void *resp,
+			    struct cb_process_state *cps)
+{
+	struct cb_notifyargs *args = argp;
+
+	kfree(args->cna_changes);
+	return 0;
+}
+
 __be32 nfs4_callback_devicenotify(void *argp, void *resp,
 				  struct cb_process_state *cps)
 {
diff --git a/fs/nfs/callback_xdr.c b/fs/nfs/callback_xdr.c
index 4382baddc9ee..8d822a3ee248 100644
--- a/fs/nfs/callback_xdr.c
+++ b/fs/nfs/callback_xdr.c
@@ -31,6 +31,7 @@
 #define CB_OP_RECALL_RES_MAXSZ		(CB_OP_HDR_RES_MAXSZ)
 
 #define CB_OP_LAYOUTRECALL_RES_MAXSZ	(CB_OP_HDR_RES_MAXSZ)
+#define CB_OP_NOTIFY_RES_MAXSZ		(CB_OP_HDR_RES_MAXSZ)
 #define CB_OP_DEVICENOTIFY_RES_MAXSZ	(CB_OP_HDR_RES_MAXSZ)
 #define CB_OP_SEQUENCE_RES_MAXSZ	(CB_OP_HDR_RES_MAXSZ + \
 					 NFS4_MAX_SESSIONID_LEN + \
@@ -251,6 +252,52 @@ static __be32 decode_layoutrecall_args(struct svc_rqst *rqstp,
 	return 0;
 }
 
+static
+__be32 decode_notify_args(struct svc_rqst *rqstp,
+			  struct xdr_stream *xdr,
+			  void *argp)
+{
+	struct cb_notifyargs *args = argp;
+	__be32 status;
+	ssize_t res;
+	__be32 *p;
+	int i;
+
+	status = decode_stateid(xdr, &args->cna_stateid);
+	if (unlikely(status != 0))
+		return status;
+
+	status = decode_fh(xdr, &args->cna_fh);
+	if (unlikely(status != 0))
+		return status;
+
+	p = xdr_inline_decode(xdr, 4);
+	if (unlikely(!p))
+		goto out;
+
+	args->cna_n_changes = ntohl(*p++);
+	args->cna_changes = kmalloc_array(args->cna_n_changes,
+					  sizeof(*args->cna_changes),
+					  GFP_KERNEL);
+	if (!args->cna_changes)
+		goto out;
+
+	for (i = 0; i < args->cna_n_changes; i++) {
+		struct cb_notify_changes *change = &args->cna_changes[i];
+
+		res = xdr_stream_decode_uint32_array(xdr,
+						     &change->notify_mask, 1);
+		if (unlikely(res < 0))
+			goto err;
+	}
+
+	return 0;
+err:
+	kfree(args->cna_changes);
+out:
+	return htonl(NFS4ERR_BADXDR);
+}
+
 static
 __be32 decode_devicenotify_args(struct svc_rqst *rqstp,
 				struct xdr_stream *xdr,
@@ -797,12 +844,12 @@ preprocess_nfs41_op(int nop, unsigned int op_nr, struct callback_op **op)
 	case OP_CB_RECALL_ANY:
 	case OP_CB_RECALL_SLOT:
 	case OP_CB_LAYOUTRECALL:
+	case OP_CB_NOTIFY:
 	case OP_CB_NOTIFY_DEVICEID:
 	case OP_CB_NOTIFY_LOCK:
 		*op = &callback_ops[op_nr];
 		break;
 
-	case OP_CB_NOTIFY:
 	case OP_CB_PUSH_DELEG:
 	case OP_CB_RECALLABLE_OBJ_AVAIL:
 	case OP_CB_WANTS_CANCELLED:
@@ -1034,6 +1081,11 @@ static struct callback_op callback_ops[] = {
 		.decode_args = decode_layoutrecall_args,
 		.res_maxsize = CB_OP_LAYOUTRECALL_RES_MAXSZ,
 	},
+	[OP_CB_NOTIFY] = {
+		.process_op = nfs4_callback_notify,
+		.decode_args = decode_notify_args,
+		.res_maxsize = CB_OP_NOTIFY_RES_MAXSZ,
+	},
 	[OP_CB_NOTIFY_DEVICEID] = {
 		.process_op = nfs4_callback_devicenotify,
 		.decode_args = decode_devicenotify_args,
diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
index fc049ce4ba8a..5c51691fff03 100644
--- a/fs/nfs/nfs4xdr.c
+++ b/fs/nfs/nfs4xdr.c
@@ -2012,7 +2012,7 @@ static void
 encode_get_dir_delegation(struct xdr_stream *xdr, struct compound_hdr *hdr)
 {
 	struct timespec64 ts = { 0, 0 };
-	u32 notifications[1] = { 0 };
+	u32 notifications[1] = { CB_NOTIFY4_GFLAG_EXTEND };
 	u32 attributes[1] = { 0 };
 	__be32 *p;
 
diff --git a/include/linux/nfs4.h b/include/linux/nfs4.h
index 1a3981c26b23..2e33f2ffdc31 100644
--- a/include/linux/nfs4.h
+++ b/include/linux/nfs4.h
@@ -491,6 +491,9 @@ enum {
 #define THRESHOLD_RD_IO                 (1UL << 2)
 #define THRESHOLD_WR_IO                 (1UL << 3)
 
+/* Directory Delegation / CB_NOTIFY bits */
+#define CB_NOTIFY4_GFLAG_EXTEND		(1UL << 6)
+
 #define NFSPROC4_NULL 0
 #define NFSPROC4_COMPOUND 1
 #define NFS4_VERSION 4
-- 
2.55.0

[PATCH v2 2/4] NFS: Add support for CB_NOTIFY4_REMOVE_ENTRY

[Anna Schumaker]

From: Anna Schumaker <anna.schumaker@oracle.com>

When the server tells us that a directory entry has been removed then we
need to take that as an indication that our knowledge of the directory
has changed and needs to be refreshed.

Signed-off-by: Anna Schumaker <anna.schumaker@oracle.com>
Signed-off-by: Anna Schumaker <anna.schumaker@hammerspace.com>
---
v2:
  * Initialize 'res' to 0 in nfs4_callback_notify()
  * Add missing error handling
  * Remove nfs4_cb_notify_lookup()
---
 fs/nfs/callback.h      | 16 ++++++++++-
 fs/nfs/callback_proc.c | 63 +++++++++++++++++++++++++++++++++++++++++-
 fs/nfs/callback_xdr.c  | 62 +++++++++++++++++++++++++++++++++++++++++
 fs/nfs/nfs4xdr.c       |  3 +-
 include/linux/nfs4.h   |  1 +
 5 files changed, 142 insertions(+), 3 deletions(-)

diff --git a/fs/nfs/callback.h b/fs/nfs/callback.h
index f7cc5b6931bf..3740c999bb82 100644
--- a/fs/nfs/callback.h
+++ b/fs/nfs/callback.h
@@ -143,8 +143,22 @@ struct cb_layoutrecallargs {
 extern __be32 nfs4_callback_layoutrecall(void *argp, void *resp,
 		struct cb_process_state *cps);
 
+struct cb_notify_entry {
+	u32			ne_namelen;
+	const char		*ne_name;
+	struct nfs_fattr	ne_attrs;
+};
+
+struct cb_notify_remove {
+	struct cb_notify_entry	nrm_old_entry;
+	u64			nrm_old_entry_cookie;
+};
+
 struct cb_notify_changes {
-	u32 notify_mask;
+	u32	notify_mask;
+	union {
+		struct cb_notify_remove notify_remove;
+	};
 };
 
 struct cb_notifyargs {
diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c
index 01600d75f2d9..d8ad3cd792df 100644
--- a/fs/nfs/callback_proc.c
+++ b/fs/nfs/callback_proc.c
@@ -9,6 +9,7 @@
 
 #include <linux/errno.h>
 #include <linux/math.h>
+#include <linux/namei.h>
 #include <linux/nfs4.h>
 #include <linux/nfs_fs.h>
 #include <linux/slab.h>
@@ -369,13 +370,73 @@ static void pnfs_recall_all_layouts(struct nfs_client *clp,
 	do_callback_layoutrecall(clp, &args, cps);
 }
 
+static __be32 nfs4_cb_notify_remove(struct cb_process_state *cps,
+				    struct dentry *parent,
+				    struct cb_notify_remove *cb_remove)
+{
+	struct qstr filename = QSTR_INIT(cb_remove->nrm_old_entry.ne_name,
+					 cb_remove->nrm_old_entry.ne_namelen);
+	struct dentry *child;
+
+	filename.hash = full_name_hash(parent, filename.name, filename.len);
+	child = d_lookup(parent, &filename);
+	if (IS_ERR_OR_NULL(child))
+		return htonl(NFS4ERR_BADHANDLE);
+
+	d_drop(child);
+	dput(child);
+
+	nfs_set_cache_invalid(parent->d_inode, NFS_INO_INVALID_DATA);
+	return 0;
+}
+
 __be32 nfs4_callback_notify(void *argp, void *resp,
 			    struct cb_process_state *cps)
 {
 	struct cb_notifyargs *args = argp;
+	struct dentry *parent;
+	struct inode *inode;
+	unsigned int i;
+	__be32 res = 0;
 
+	if (!cps->clp) {
+		res = htonl(NFS4ERR_OP_NOT_IN_SESSION);
+		goto out;
+	}
+
+	inode = nfs_delegation_find_inode(cps->clp, &args->cna_fh);
+	if (IS_ERR(inode)) {
+		res = htonl(NFS4ERR_BADHANDLE);
+		goto out;
+	}
+	parent = d_find_alias(inode);
+	if (!parent)
+		goto out_iput;
+
+	for (i = 0; i < args->cna_n_changes; i++) {
+		struct cb_notify_changes *change = &args->cna_changes[i];
+
+		switch (change->notify_mask) {
+		case CB_NOTIFY4_REMOVE_ENTRY:
+			res = nfs4_cb_notify_remove(cps, parent,
+						    &change->notify_remove);
+			break;
+		default:
+			res = htonl(NFS4ERR_NOTSUPP);
+			goto out_dput;
+		}
+
+		if (unlikely(res != 0))
+			break;
+	}
+
+out_dput:
+	dput(parent);
+out_iput:
+	nfs_iput_and_deactive(inode);
+out:
 	kfree(args->cna_changes);
-	return 0;
+	return res;
 }
 
 __be32 nfs4_callback_devicenotify(void *argp, void *resp,
diff --git a/fs/nfs/callback_xdr.c b/fs/nfs/callback_xdr.c
index 8d822a3ee248..612d215ac66e 100644
--- a/fs/nfs/callback_xdr.c
+++ b/fs/nfs/callback_xdr.c
@@ -252,6 +252,50 @@ static __be32 decode_layoutrecall_args(struct svc_rqst *rqstp,
 	return 0;
 }
 
+static __be32 decode_notify_entry(struct xdr_stream *xdr,
+				  struct cb_notify_entry *args)
+{
+	uint32_t bitmap[3] = {0};
+	__be32 status;
+	u32 attrlen;
+	__be32 *p;
+
+	status = decode_string(xdr, &args->ne_namelen, &args->ne_name,
+				NFS4_OPAQUE_LIMIT);
+	if (unlikely(status != 0))
+		return status;
+
+	status = decode_bitmap(xdr, bitmap);
+	if (unlikely(status != 0))
+		return status;
+
+	p = xdr_inline_decode(xdr, 4);
+	if (unlikely(!p))
+		return htonl(NFS4ERR_BADXDR);
+
+	attrlen = be32_to_cpup(p);
+	if (attrlen != 0)
+		return htonl(NFS4ERR_BADXDR);
+	return 0;
+}
+
+static __be32 decode_notify_remove(struct xdr_stream *xdr,
+				   struct cb_notify_remove *args)
+{
+	__be32 status;
+	__be32 *p;
+
+	status = decode_notify_entry(xdr, &args->nrm_old_entry);
+	if (unlikely(status != 0))
+		return status;
+
+	p = xdr_inline_decode(xdr, 8);
+	if (unlikely(!p))
+		return htonl(NFS4ERR_BADXDR);
+	xdr_decode_hyper(p, &args->nrm_old_entry_cookie);
+	return 0;
+}
+
 static
 __be32 decode_notify_args(struct svc_rqst *rqstp,
 			  struct xdr_stream *xdr,
@@ -289,6 +333,24 @@ __be32 decode_notify_args(struct svc_rqst *rqstp,
 						     &change->notify_mask, 1);
 		if (unlikely(res < 0))
 			goto err;
+
+		/* Decode opaque size */
+		p = xdr_inline_decode(xdr, 4);
+		if (unlikely(!p))
+			goto err;
+		res = ntohl(*p);
+
+		switch (change->notify_mask) {
+		case CB_NOTIFY4_REMOVE_ENTRY:
+			status = decode_notify_remove(xdr,
+						      &change->notify_remove);
+			break;
+		default:
+			goto err;
+		}
+
+		if (unlikely(status != 0))
+			goto err;
 	}
 
 	return 0;
diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
index 5c51691fff03..e275fbfd395e 100644
--- a/fs/nfs/nfs4xdr.c
+++ b/fs/nfs/nfs4xdr.c
@@ -2012,7 +2012,8 @@ static void
 encode_get_dir_delegation(struct xdr_stream *xdr, struct compound_hdr *hdr)
 {
 	struct timespec64 ts = { 0, 0 };
-	u32 notifications[1] = { CB_NOTIFY4_GFLAG_EXTEND };
+	u32 notifications[1] = { CB_NOTIFY4_REMOVE_ENTRY |
+				 CB_NOTIFY4_GFLAG_EXTEND };
 	u32 attributes[1] = { 0 };
 	__be32 *p;
 
diff --git a/include/linux/nfs4.h b/include/linux/nfs4.h
index 2e33f2ffdc31..8de3ea283767 100644
--- a/include/linux/nfs4.h
+++ b/include/linux/nfs4.h
@@ -492,6 +492,7 @@ enum {
 #define THRESHOLD_WR_IO                 (1UL << 3)
 
 /* Directory Delegation / CB_NOTIFY bits */
+#define CB_NOTIFY4_REMOVE_ENTRY		(1UL << 2)
 #define CB_NOTIFY4_GFLAG_EXTEND		(1UL << 6)
 
 #define NFSPROC4_NULL 0
-- 
2.55.0

[PATCH v2 3/4] NFS: Add support for CB_NOTIFY4_ADD_ENTRY

[Anna Schumaker]

From: Anna Schumaker <anna.schumaker@oracle.com>

When the server tells us that a directory entry has been created, we
need to instantiate a new dentry and add it to the dcache on the client.

Signed-off-by: Anna Schumaker <anna.schumaker@oracle.com>
Signed-off-by: Anna Schumaker <anna.schumaker@hammerspace.com>

---
v2:
 * Handle the na_have_old_entry case
 * Initialize fattr->valid to 0 before decoding
 * Properly check the return code from decode_fattr_cb()
 * Rewrite nfs4_cb_notify_add() to closer resemble nfs_prime_dcache()
---
 fs/nfs/callback.h      | 17 +++++++++++
 fs/nfs/callback_proc.c | 61 +++++++++++++++++++++++++++++++++++++
 fs/nfs/callback_xdr.c  | 69 ++++++++++++++++++++++++++++++++++--------
 fs/nfs/nfs4_fs.h       |  2 ++
 fs/nfs/nfs4xdr.c       | 29 ++++++++++++++++--
 include/linux/nfs4.h   |  1 +
 6 files changed, 164 insertions(+), 15 deletions(-)

diff --git a/fs/nfs/callback.h b/fs/nfs/callback.h
index 3740c999bb82..d3cc3e5df776 100644
--- a/fs/nfs/callback.h
+++ b/fs/nfs/callback.h
@@ -147,6 +147,12 @@ struct cb_notify_entry {
 	u32			ne_namelen;
 	const char		*ne_name;
 	struct nfs_fattr	ne_attrs;
+	struct nfs_fh		ne_fh;
+};
+
+struct cb_notify_prev_entry {
+	struct cb_notify_entry	pe_prev_entry;
+	u64			pe_prev_entry_cookie;
 };
 
 struct cb_notify_remove {
@@ -154,10 +160,21 @@ struct cb_notify_remove {
 	u64			nrm_old_entry_cookie;
 };
 
+struct cb_notify_add {
+	bool			na_have_old_entry;
+	struct cb_notify_remove	na_old_entry;
+	struct cb_notify_entry	na_new_entry;
+	bool			na_have_new_entry_cookie;
+	u64			na_new_entry_cookie;
+	bool			na_have_prev_entry;
+	bool			na_last_entry;
+};
+
 struct cb_notify_changes {
 	u32	notify_mask;
 	union {
 		struct cb_notify_remove notify_remove;
+		struct cb_notify_add	notify_add;
 	};
 };
 
diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c
index d8ad3cd792df..9ac786f258c2 100644
--- a/fs/nfs/callback_proc.c
+++ b/fs/nfs/callback_proc.c
@@ -390,6 +390,63 @@ static __be32 nfs4_cb_notify_remove(struct cb_process_state *cps,
 	return 0;
 }
 
+static __be32 nfs4_cb_notify_add(struct cb_process_state *cps,
+				 struct dentry *parent,
+				 struct cb_notify_add *cb_add)
+{
+	struct nfs_entry entry = {
+		.cookie = cb_add->na_new_entry_cookie,
+		.name = cb_add->na_new_entry.ne_name,
+		.len = cb_add->na_new_entry.ne_namelen,
+		.eof = cb_add->na_last_entry,
+		.fh = &cb_add->na_new_entry.ne_fh,
+		.fattr = &cb_add->na_new_entry.ne_attrs,
+		.server = NFS_SERVER(d_inode(parent)),
+	};
+	struct qstr filename = QSTR_INIT(entry.name, entry.len);
+	struct dentry *dentry, *alias;
+	struct inode *inode;
+
+	if (cb_add->na_have_old_entry)
+		nfs4_cb_notify_remove(cps, parent, &cb_add->na_old_entry);
+
+	filename.hash = full_name_hash(parent, filename.name, filename.len);
+	dentry = d_lookup(parent, &filename);
+	if (!dentry) {
+		dentry = d_alloc_parallel(parent, &filename);
+		if (IS_ERR(dentry))
+			goto out;
+	}
+
+	if (!d_in_lookup(dentry)) {
+		nfs_set_verifier(dentry, parent->d_time);
+		if (!nfs_refresh_inode(d_inode(dentry), entry.fattr))
+			nfs_setsecurity(d_inode(dentry), entry.fattr);
+		goto out;
+	}
+
+	if (!entry.fh->size) {
+		d_lookup_done(dentry);
+		goto out;
+	}
+
+	nfs_set_verifier(dentry, parent->d_time);
+	inode = nfs_fhget(parent->d_sb, entry.fh, entry.fattr);
+	alias = d_splice_alias(inode, dentry);
+	d_lookup_done(dentry);
+
+	if (IS_ERR_OR_NULL(alias))
+		goto out;
+	nfs_set_verifier(alias, parent->d_time);
+	dput(dentry);
+	dentry = alias;
+
+out:
+	nfs_set_cache_invalid(parent->d_inode, NFS_INO_INVALID_DATA);
+	dput(dentry);
+	return 0;
+}
+
 __be32 nfs4_callback_notify(void *argp, void *resp,
 			    struct cb_process_state *cps)
 {
@@ -421,6 +478,10 @@ __be32 nfs4_callback_notify(void *argp, void *resp,
 			res = nfs4_cb_notify_remove(cps, parent,
 						    &change->notify_remove);
 			break;
+		case CB_NOTIFY4_ADD_ENTRY:
+			res = nfs4_cb_notify_add(cps, parent,
+						 &change->notify_add);
+			break;
 		default:
 			res = htonl(NFS4ERR_NOTSUPP);
 			goto out_dput;
diff --git a/fs/nfs/callback_xdr.c b/fs/nfs/callback_xdr.c
index 612d215ac66e..39a879e82312 100644
--- a/fs/nfs/callback_xdr.c
+++ b/fs/nfs/callback_xdr.c
@@ -257,26 +257,20 @@ static __be32 decode_notify_entry(struct xdr_stream *xdr,
 {
 	uint32_t bitmap[3] = {0};
 	__be32 status;
-	u32 attrlen;
-	__be32 *p;
 
 	status = decode_string(xdr, &args->ne_namelen, &args->ne_name,
 				NFS4_OPAQUE_LIMIT);
 	if (unlikely(status != 0))
 		return status;
 
-	status = decode_bitmap(xdr, bitmap);
+	status = xdr_stream_decode_uint32_array(xdr, bitmap, 3);
+	if (unlikely(status == 0 || status > 3))
+		return htonl(NFS4ERR_BADXDR);
+
+	status = decode_fattr_cb(xdr, bitmap, &args->ne_attrs, &args->ne_fh);
 	if (unlikely(status != 0))
-		return status;
-
-	p = xdr_inline_decode(xdr, 4);
-	if (unlikely(!p))
 		return htonl(NFS4ERR_BADXDR);
-
-	attrlen = be32_to_cpup(p);
-	if (attrlen != 0)
-		return htonl(NFS4ERR_BADXDR);
-	return 0;
+	return status;
 }
 
 static __be32 decode_notify_remove(struct xdr_stream *xdr,
@@ -296,6 +290,54 @@ static __be32 decode_notify_remove(struct xdr_stream *xdr,
 	return 0;
 }
 
+static __be32 decode_notify_add(struct xdr_stream *xdr,
+				struct cb_notify_add *args)
+{
+	__be32 status;
+	__be32 *p;
+
+	p = xdr_inline_decode(xdr, 4);
+	if (unlikely(!p))
+		return htonl(NFS4ERR_BADXDR);
+	args->na_have_old_entry = ntohl(*p);
+
+	if (args->na_have_old_entry) {
+		status = decode_notify_remove(xdr, &args->na_old_entry);
+		if (unlikely(status != 0))
+			return status;
+	}
+
+	status = decode_notify_entry(xdr, &args->na_new_entry);
+	if (unlikely(status != 0))
+		return status;
+
+	p = xdr_inline_decode(xdr, 4);
+	if (unlikely(!p))
+		return htonl(NFS4ERR_BADXDR);
+	args->na_have_new_entry_cookie = ntohl(*p);
+
+	if (args->na_have_new_entry_cookie) {
+		p = xdr_inline_decode(xdr, 8);
+		if (unlikely(!p))
+			return htonl(NFS4ERR_BADXDR);
+
+		xdr_decode_hyper(p, &args->na_new_entry_cookie);
+	}
+
+	p = xdr_inline_decode(xdr, 4);
+	if (unlikely(!p))
+		return htonl(NFS4ERR_BADXDR);
+	args->na_have_prev_entry = ntohl(*p);
+
+	WARN_ONCE(args->na_have_prev_entry, "NFS: Have prev entry unimplemented\n");
+
+	p = xdr_inline_decode(xdr, 4);
+	if (unlikely(!p))
+		return htonl(NFS4ERR_BADXDR);
+	args->na_last_entry = ntohl(*p);
+	return 0;
+}
+
 static
 __be32 decode_notify_args(struct svc_rqst *rqstp,
 			  struct xdr_stream *xdr,
@@ -345,6 +387,9 @@ __be32 decode_notify_args(struct svc_rqst *rqstp,
 			status = decode_notify_remove(xdr,
 						      &change->notify_remove);
 			break;
+		case CB_NOTIFY4_ADD_ENTRY:
+			status = decode_notify_add(xdr, &change->notify_add);
+			break;
 		default:
 			goto err;
 		}
diff --git a/fs/nfs/nfs4_fs.h b/fs/nfs/nfs4_fs.h
index b48e5b87cb2a..b029c9cefc38 100644
--- a/fs/nfs/nfs4_fs.h
+++ b/fs/nfs/nfs4_fs.h
@@ -586,6 +586,8 @@ extern const u32 nfs42_maxlistxattrs_overhead;
 #endif
 
 struct nfs4_mount_data;
+int decode_fattr_cb(struct xdr_stream *xdr, uint32_t *bitmap,
+		    struct nfs_fattr *fattr, struct nfs_fh *fhandle);
 
 /* callback_xdr.c */
 extern const struct svc_version nfs4_callback_version1;
diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
index e275fbfd395e..9a7c4d23aea7 100644
--- a/fs/nfs/nfs4xdr.c
+++ b/fs/nfs/nfs4xdr.c
@@ -2013,8 +2013,11 @@ encode_get_dir_delegation(struct xdr_stream *xdr, struct compound_hdr *hdr)
 {
 	struct timespec64 ts = { 0, 0 };
 	u32 notifications[1] = { CB_NOTIFY4_REMOVE_ENTRY |
+				 CB_NOTIFY4_ADD_ENTRY |
 				 CB_NOTIFY4_GFLAG_EXTEND };
-	u32 attributes[1] = { 0 };
+	u32 child_attrs[1] = { FATTR4_WORD0_FSID | FATTR4_WORD0_FILEID |
+			       FATTR4_WORD0_TYPE | FATTR4_WORD0_FILEHANDLE };
+	u32 dir_attrs[1] = { 0 };
 	__be32 *p;
 
 	encode_op_hdr(xdr, OP_GET_DIR_DELEGATION, decode_get_dir_deleg_maxsz, hdr);
@@ -2030,10 +2033,10 @@ encode_get_dir_delegation(struct xdr_stream *xdr, struct compound_hdr *hdr)
 	xdr_encode_nfstime4(p, &ts);
 
 	/* Requested child attributes */
-	xdr_encode_bitmap4(xdr, attributes, ARRAY_SIZE(attributes));
+	xdr_encode_bitmap4(xdr, child_attrs, ARRAY_SIZE(child_attrs));
 
 	/* Requested dir attributes */
-	xdr_encode_bitmap4(xdr, attributes, ARRAY_SIZE(attributes));
+	xdr_encode_bitmap4(xdr, dir_attrs, ARRAY_SIZE(dir_attrs));
 }
 
 static void
@@ -4919,6 +4922,26 @@ static int decode_getfattr(struct xdr_stream *xdr, struct nfs_fattr *fattr,
 	return decode_getfattr_generic(xdr, fattr, NULL, NULL, server);
 }
 
+int decode_fattr_cb(struct xdr_stream *xdr, uint32_t *bitmap,
+			     struct nfs_fattr *fattr, struct nfs_fh *fhandle)
+{
+	unsigned int savep;
+	uint32_t attrlen;
+	int status;
+
+	fattr->valid = 0;
+
+	status = decode_attr_length(xdr, &attrlen, &savep);
+	if (status < 0)
+		return status;
+
+	status = decode_getfattr_attrs(xdr, bitmap, fattr, fhandle, NULL, NULL);
+	if (status < 0)
+		return status;
+
+	return verify_attr_len(xdr, savep, attrlen);
+}
+
 /*
  * Decode potentially multiple layout types.
  */
diff --git a/include/linux/nfs4.h b/include/linux/nfs4.h
index 8de3ea283767..4b507a93aa38 100644
--- a/include/linux/nfs4.h
+++ b/include/linux/nfs4.h
@@ -493,6 +493,7 @@ enum {
 
 /* Directory Delegation / CB_NOTIFY bits */
 #define CB_NOTIFY4_REMOVE_ENTRY		(1UL << 2)
+#define CB_NOTIFY4_ADD_ENTRY		(1UL << 3)
 #define CB_NOTIFY4_GFLAG_EXTEND		(1UL << 6)
 
 #define NFSPROC4_NULL 0
-- 
2.55.0

[PATCH v2 4/4] NFS: Add support for CB_NOTIFY4_RENAME_ENTRY

[Anna Schumaker]

From: Anna Schumaker <anna.schumaker@oracle.com>

Renaming an entry essentially combines a remove with an add, allowing us
to reuse most of the code from the previous patches to do the work for
us.

Signed-off-by: Anna Schumaker <anna.schumaker@oracle.com>
Signed-off-by: Anna Schumaker <anna.schumaker@hammerspace.com>
---
 fs/nfs/callback.h      |  6 ++++++
 fs/nfs/callback_proc.c | 16 ++++++++++++++++
 fs/nfs/callback_xdr.c  | 14 ++++++++++++++
 fs/nfs/nfs4xdr.c       |  1 +
 include/linux/nfs4.h   |  1 +
 5 files changed, 38 insertions(+)

diff --git a/fs/nfs/callback.h b/fs/nfs/callback.h
index d3cc3e5df776..019ee9eed9fe 100644
--- a/fs/nfs/callback.h
+++ b/fs/nfs/callback.h
@@ -170,11 +170,17 @@ struct cb_notify_add {
 	bool			na_last_entry;
 };
 
+struct cb_notify_rename {
+	struct cb_notify_remove	nrn_old_entry;
+	struct cb_notify_add	nrn_new_entry;
+};
+
 struct cb_notify_changes {
 	u32	notify_mask;
 	union {
 		struct cb_notify_remove notify_remove;
 		struct cb_notify_add	notify_add;
+		struct cb_notify_rename	notify_rename;
 	};
 };
 
diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c
index 9ac786f258c2..2e533c8e297a 100644
--- a/fs/nfs/callback_proc.c
+++ b/fs/nfs/callback_proc.c
@@ -447,6 +447,18 @@ static __be32 nfs4_cb_notify_add(struct cb_process_state *cps,
 	return 0;
 }
 
+static __be32 nfs4_cb_notify_rename(struct cb_process_state *cps,
+				    struct dentry *parent,
+				    struct cb_notify_rename *cb_rename)
+{
+	__be32 status;
+
+	status = nfs4_cb_notify_remove(cps, parent, &cb_rename->nrn_old_entry);
+	if (status != 0)
+		return status;
+	return nfs4_cb_notify_add(cps, parent, &cb_rename->nrn_new_entry);
+}
+
 __be32 nfs4_callback_notify(void *argp, void *resp,
 			    struct cb_process_state *cps)
 {
@@ -482,6 +494,10 @@ __be32 nfs4_callback_notify(void *argp, void *resp,
 			res = nfs4_cb_notify_add(cps, parent,
 						 &change->notify_add);
 			break;
+		case CB_NOTIFY4_RENAME_ENTRY:
+			res = nfs4_cb_notify_rename(cps, parent,
+						    &change->notify_rename);
+			break;
 		default:
 			res = htonl(NFS4ERR_NOTSUPP);
 			goto out_dput;
diff --git a/fs/nfs/callback_xdr.c b/fs/nfs/callback_xdr.c
index 39a879e82312..e787657c5db0 100644
--- a/fs/nfs/callback_xdr.c
+++ b/fs/nfs/callback_xdr.c
@@ -338,6 +338,17 @@ static __be32 decode_notify_add(struct xdr_stream *xdr,
 	return 0;
 }
 
+static __be32 decode_notify_rename(struct xdr_stream *xdr,
+				   struct cb_notify_rename *args)
+{
+	__be32 status;
+
+	status = decode_notify_remove(xdr, &args->nrn_old_entry);
+	if (unlikely(status != 0))
+		return status;
+	return decode_notify_add(xdr, &args->nrn_new_entry);
+}
+
 static
 __be32 decode_notify_args(struct svc_rqst *rqstp,
 			  struct xdr_stream *xdr,
@@ -390,6 +401,9 @@ __be32 decode_notify_args(struct svc_rqst *rqstp,
 		case CB_NOTIFY4_ADD_ENTRY:
 			status = decode_notify_add(xdr, &change->notify_add);
 			break;
+		case CB_NOTIFY4_RENAME_ENTRY:
+			status = decode_notify_rename(xdr, &change->notify_rename);
+			break;
 		default:
 			goto err;
 		}
diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
index 9a7c4d23aea7..6d102ebf5e6d 100644
--- a/fs/nfs/nfs4xdr.c
+++ b/fs/nfs/nfs4xdr.c
@@ -2014,6 +2014,7 @@ encode_get_dir_delegation(struct xdr_stream *xdr, struct compound_hdr *hdr)
 	struct timespec64 ts = { 0, 0 };
 	u32 notifications[1] = { CB_NOTIFY4_REMOVE_ENTRY |
 				 CB_NOTIFY4_ADD_ENTRY |
+				 CB_NOTIFY4_RENAME_ENTRY |
 				 CB_NOTIFY4_GFLAG_EXTEND };
 	u32 child_attrs[1] = { FATTR4_WORD0_FSID | FATTR4_WORD0_FILEID |
 			       FATTR4_WORD0_TYPE | FATTR4_WORD0_FILEHANDLE };
diff --git a/include/linux/nfs4.h b/include/linux/nfs4.h
index 4b507a93aa38..27f083546421 100644
--- a/include/linux/nfs4.h
+++ b/include/linux/nfs4.h
@@ -494,6 +494,7 @@ enum {
 /* Directory Delegation / CB_NOTIFY bits */
 #define CB_NOTIFY4_REMOVE_ENTRY		(1UL << 2)
 #define CB_NOTIFY4_ADD_ENTRY		(1UL << 3)
+#define CB_NOTIFY4_RENAME_ENTRY		(1UL << 4)
 #define CB_NOTIFY4_GFLAG_EXTEND		(1UL << 6)
 
 #define NFSPROC4_NULL 0
-- 
2.55.0

Re: [PATCH v2 2/4] NFS: Add support for CB_NOTIFY4_REMOVE_ENTRY

[Jeff Layton]

On Tue, 2026-08-25 at 11:02 -0400, Anna Schumaker wrote:
> From: Anna Schumaker <anna.schumaker@oracle.com>
> 
> When the server tells us that a directory entry has been removed then we
> need to take that as an indication that our knowledge of the directory
> has changed and needs to be refreshed.
> 
> Signed-off-by: Anna Schumaker <anna.schumaker@oracle.com>
> Signed-off-by: Anna Schumaker <anna.schumaker@hammerspace.com>
> ---
> v2:
>   * Initialize 'res' to 0 in nfs4_callback_notify()
>   * Add missing error handling
>   * Remove nfs4_cb_notify_lookup()
> ---
>  fs/nfs/callback.h      | 16 ++++++++++-
>  fs/nfs/callback_proc.c | 63 +++++++++++++++++++++++++++++++++++++++++-
>  fs/nfs/callback_xdr.c  | 62 +++++++++++++++++++++++++++++++++++++++++
>  fs/nfs/nfs4xdr.c       |  3 +-
>  include/linux/nfs4.h   |  1 +
>  5 files changed, 142 insertions(+), 3 deletions(-)
> 
> diff --git a/fs/nfs/callback.h b/fs/nfs/callback.h
> index f7cc5b6931bf..3740c999bb82 100644
> --- a/fs/nfs/callback.h
> +++ b/fs/nfs/callback.h
> @@ -143,8 +143,22 @@ struct cb_layoutrecallargs {
>  extern __be32 nfs4_callback_layoutrecall(void *argp, void *resp,
>  		struct cb_process_state *cps);
>  
> +struct cb_notify_entry {
> +	u32			ne_namelen;
> +	const char		*ne_name;
> +	struct nfs_fattr	ne_attrs;
> +};
> +
> +struct cb_notify_remove {
> +	struct cb_notify_entry	nrm_old_entry;
> +	u64			nrm_old_entry_cookie;
> +};
> +
>  struct cb_notify_changes {
> -	u32 notify_mask;
> +	u32	notify_mask;
> +	union {
> +		struct cb_notify_remove notify_remove;
> +	};
>  };
>  
>  struct cb_notifyargs {
> diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c
> index 01600d75f2d9..d8ad3cd792df 100644
> --- a/fs/nfs/callback_proc.c
> +++ b/fs/nfs/callback_proc.c
> @@ -9,6 +9,7 @@
>  
>  #include <linux/errno.h>
>  #include <linux/math.h>
> +#include <linux/namei.h>
>  #include <linux/nfs4.h>
>  #include <linux/nfs_fs.h>
>  #include <linux/slab.h>
> @@ -369,13 +370,73 @@ static void pnfs_recall_all_layouts(struct nfs_client *clp,
>  	do_callback_layoutrecall(clp, &args, cps);
>  }
>  
> +static __be32 nfs4_cb_notify_remove(struct cb_process_state *cps,
> +				    struct dentry *parent,
> +				    struct cb_notify_remove *cb_remove)
> +{
> +	struct qstr filename = QSTR_INIT(cb_remove->nrm_old_entry.ne_name,
> +					 cb_remove->nrm_old_entry.ne_namelen);
> +	struct dentry *child;
> +
> +	filename.hash = full_name_hash(parent, filename.name, filename.len);
> +	child = d_lookup(parent, &filename);
> +	if (IS_ERR_OR_NULL(child))
> +		return htonl(NFS4ERR_BADHANDLE);
> +
> +	d_drop(child);

Should this instead make this a negative dentry?

I think this will just drop the existing dentry and it'll trigger a
lookup next time we end up trying to hit it.

> +	dput(child);
> +
> +	nfs_set_cache_invalid(parent->d_inode, NFS_INO_INVALID_DATA);
> +	return 0;
> +}
> +
>  __be32 nfs4_callback_notify(void *argp, void *resp,
>  			    struct cb_process_state *cps)
>  {
>  	struct cb_notifyargs *args = argp;
> +	struct dentry *parent;
> +	struct inode *inode;
> +	unsigned int i;
> +	__be32 res = 0;
>  
> +	if (!cps->clp) {
> +		res = htonl(NFS4ERR_OP_NOT_IN_SESSION);
> +		goto out;
> +	}
> +
> +	inode = nfs_delegation_find_inode(cps->clp, &args->cna_fh);
> +	if (IS_ERR(inode)) {
> +		res = htonl(NFS4ERR_BADHANDLE);
> +		goto out;
> +	}
> +	parent = d_find_alias(inode);
> +	if (!parent)
> +		goto out_iput;
> +
> +	for (i = 0; i < args->cna_n_changes; i++) {
> +		struct cb_notify_changes *change = &args->cna_changes[i];
> +
> +		switch (change->notify_mask) {
> +		case CB_NOTIFY4_REMOVE_ENTRY:
> +			res = nfs4_cb_notify_remove(cps, parent,
> +						    &change->notify_remove);
> +			break;
> +		default:
> +			res = htonl(NFS4ERR_NOTSUPP);
> +			goto out_dput;
> +		}
> +
> +		if (unlikely(res != 0))
> +			break;
> +	}
> +
> +out_dput:
> +	dput(parent);
> +out_iput:
> +	nfs_iput_and_deactive(inode);
> +out:
>  	kfree(args->cna_changes);
> -	return 0;
> +	return res;
>  }
>  
>  __be32 nfs4_callback_devicenotify(void *argp, void *resp,
> diff --git a/fs/nfs/callback_xdr.c b/fs/nfs/callback_xdr.c
> index 8d822a3ee248..612d215ac66e 100644
> --- a/fs/nfs/callback_xdr.c
> +++ b/fs/nfs/callback_xdr.c
> @@ -252,6 +252,50 @@ static __be32 decode_layoutrecall_args(struct svc_rqst *rqstp,
>  	return 0;
>  }
>  
> +static __be32 decode_notify_entry(struct xdr_stream *xdr,
> +				  struct cb_notify_entry *args)
> +{
> +	uint32_t bitmap[3] = {0};
> +	__be32 status;
> +	u32 attrlen;
> +	__be32 *p;
> +
> +	status = decode_string(xdr, &args->ne_namelen, &args->ne_name,
> +				NFS4_OPAQUE_LIMIT);
> +	if (unlikely(status != 0))
> +		return status;
> +
> +	status = decode_bitmap(xdr, bitmap);
> +	if (unlikely(status != 0))
> +		return status;
> +
> +	p = xdr_inline_decode(xdr, 4);
> +	if (unlikely(!p))
> +		return htonl(NFS4ERR_BADXDR);
> +
> +	attrlen = be32_to_cpup(p);
> +	if (attrlen != 0)
> +		return htonl(NFS4ERR_BADXDR);
> +	return 0;
> +}
> +
> +static __be32 decode_notify_remove(struct xdr_stream *xdr,
> +				   struct cb_notify_remove *args)
> +{
> +	__be32 status;
> +	__be32 *p;
> +
> +	status = decode_notify_entry(xdr, &args->nrm_old_entry);
> +	if (unlikely(status != 0))
> +		return status;
> +
> +	p = xdr_inline_decode(xdr, 8);
> +	if (unlikely(!p))
> +		return htonl(NFS4ERR_BADXDR);
> +	xdr_decode_hyper(p, &args->nrm_old_entry_cookie);
> +	return 0;
> +}
> +
>  static
>  __be32 decode_notify_args(struct svc_rqst *rqstp,
>  			  struct xdr_stream *xdr,
> @@ -289,6 +333,24 @@ __be32 decode_notify_args(struct svc_rqst *rqstp,
>  						     &change->notify_mask, 1);
>  		if (unlikely(res < 0))
>  			goto err;
> +
> +		/* Decode opaque size */
> +		p = xdr_inline_decode(xdr, 4);
> +		if (unlikely(!p))
> +			goto err;
> +		res = ntohl(*p);
> +
> +		switch (change->notify_mask) {
> +		case CB_NOTIFY4_REMOVE_ENTRY:
> +			status = decode_notify_remove(xdr,
> +						      &change->notify_remove);
> +			break;
> +		default:
> +			goto err;
> +		}
> +
> +		if (unlikely(status != 0))
> +			goto err;
>  	}
>  
>  	return 0;
> diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
> index 5c51691fff03..e275fbfd395e 100644
> --- a/fs/nfs/nfs4xdr.c
> +++ b/fs/nfs/nfs4xdr.c
> @@ -2012,7 +2012,8 @@ static void
>  encode_get_dir_delegation(struct xdr_stream *xdr, struct compound_hdr *hdr)
>  {
>  	struct timespec64 ts = { 0, 0 };
> -	u32 notifications[1] = { CB_NOTIFY4_GFLAG_EXTEND };
> +	u32 notifications[1] = { CB_NOTIFY4_REMOVE_ENTRY |
> +				 CB_NOTIFY4_GFLAG_EXTEND };
>  	u32 attributes[1] = { 0 };
>  	__be32 *p;
>  
> diff --git a/include/linux/nfs4.h b/include/linux/nfs4.h
> index 2e33f2ffdc31..8de3ea283767 100644
> --- a/include/linux/nfs4.h
> +++ b/include/linux/nfs4.h
> @@ -492,6 +492,7 @@ enum {
>  #define THRESHOLD_WR_IO                 (1UL << 3)
>  
>  /* Directory Delegation / CB_NOTIFY bits */
> +#define CB_NOTIFY4_REMOVE_ENTRY		(1UL << 2)
>  #define CB_NOTIFY4_GFLAG_EXTEND		(1UL << 6)
>  
>  #define NFSPROC4_NULL 0

-- 
Jeff Layton <jlayton@kernel.org>