Skip to contentSkip to conversation

NFSD: Rebase dropped patches for block/scsi layout

PATCH In review Sergey Bashirov 3 participants 5 messages Jul 21, 2025

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

cover NFSD: Rebase dropped patches for block/scsi layout

This series resubmits two patches that were dropped due to conflicts after the deviceid4 handling rework. The first patch is rebased. Only the decoding of bex.vol_id, which has the deviceid4 type, is updated. The second patch doesn't change, it just depends on the first one.

---
Sergey Bashirov (2):
NFSD: Implement large extent array support in pNFS
NFSD: Fix last write offset handling in layoutcommit

 fs/nfsd/blocklayout.c    | 25 ++++++------
 fs/nfsd/blocklayoutxdr.c | 83 +++++++++++++++++++++++++++-------------
 fs/nfsd/blocklayoutxdr.h |  4 +-
 fs/nfsd/nfs4proc.c       | 32 ++++++++--------
 fs/nfsd/nfs4xdr.c        | 11 +++---
 fs/nfsd/pnfs.h           |  1 +
 fs/nfsd/xdr4.h           |  3 +-
 7 files changed, 95 insertions(+), 64 deletions(-)
Signature · 1 line
2.43.0
#1 lore raw

PATCH 1/2 NFSD: Implement large extent array support in pNFS

When pNFS client in the block or scsi layout mode sends layoutcommit to MDS, a variable length array of modified extents is supplied within the request. This patch allows the server to accept such extent arrays if they do not fit within single memory page.

The issue can be reproduced when writing to a 1GB file using FIO with O_DIRECT, 4K block and large I/O depth without preallocation of the file. In this case, the server returns NFSERR_BADXDR to the client.

---
 fs/nfsd/blocklayout.c    | 20 ++++++----
 fs/nfsd/blocklayoutxdr.c | 83 +++++++++++++++++++++++++++-------------
 fs/nfsd/blocklayoutxdr.h |  4 +-
 fs/nfsd/nfs4proc.c       |  2 +-
 fs/nfsd/nfs4xdr.c        | 11 +++---
 fs/nfsd/pnfs.h           |  1 +
 fs/nfsd/xdr4.h           |  3 +-
 7 files changed, 78 insertions(+), 46 deletions(-)
fs/nfsd/blocklayout.c +12 -8
index 19078a043e85c..4c936132eb440 100644
@@ -173,16 +173,18 @@ nfsd4_block_proc_getdeviceinfo(struct super_block *sb,
 }
 
 static __be32
-nfsd4_block_proc_layoutcommit(struct inode *inode,
+nfsd4_block_proc_layoutcommit(struct inode *inode, struct svc_rqst *rqstp,
 		struct nfsd4_layoutcommit *lcp)
 {
 	struct iomap *iomaps;
 	int nr_iomaps;
 	__be32 nfserr;
 
-	nfserr = nfsd4_block_decode_layoutupdate(lcp->lc_up_layout,
-			lcp->lc_up_len, &iomaps, &nr_iomaps,
-			i_blocksize(inode));
+	rqstp->rq_arg = lcp->lc_up_layout;
+	svcxdr_init_decode(rqstp);
+
+	nfserr = nfsd4_block_decode_layoutupdate(&rqstp->rq_arg_stream,
+			&iomaps, &nr_iomaps, i_blocksize(inode));
 	if (nfserr != nfs_ok)
 		return nfserr;
 
@@ -313,16 +315,18 @@ nfsd4_scsi_proc_getdeviceinfo(struct super_block *sb,
 	return nfserrno(nfsd4_block_get_device_info_scsi(sb, clp, gdp));
 }
 static __be32
-nfsd4_scsi_proc_layoutcommit(struct inode *inode,
+nfsd4_scsi_proc_layoutcommit(struct inode *inode, struct svc_rqst *rqstp,
 		struct nfsd4_layoutcommit *lcp)
 {
 	struct iomap *iomaps;
 	int nr_iomaps;
 	__be32 nfserr;
 
-	nfserr = nfsd4_scsi_decode_layoutupdate(lcp->lc_up_layout,
-			lcp->lc_up_len, &iomaps, &nr_iomaps,
-			i_blocksize(inode));
+	rqstp->rq_arg = lcp->lc_up_layout;
+	svcxdr_init_decode(rqstp);
+
+	nfserr = nfsd4_scsi_decode_layoutupdate(&rqstp->rq_arg_stream,
+			&iomaps, &nr_iomaps, i_blocksize(inode));
 	if (nfserr != nfs_ok)
 		return nfserr;
 
fs/nfsd/blocklayoutxdr.c +56 -27

163 lines not on this page · raw message

fs/nfsd/blocklayoutxdr.h +2 -2
index 15b3569f3d9ad..7d25ef689671f 100644
@@ -54,9 +54,9 @@ __be32 nfsd4_block_encode_getdeviceinfo(struct xdr_stream *xdr,
 		const struct nfsd4_getdeviceinfo *gdp);
 __be32 nfsd4_block_encode_layoutget(struct xdr_stream *xdr,
 		const struct nfsd4_layoutget *lgp);
-__be32 nfsd4_block_decode_layoutupdate(__be32 *p, u32 len,
+__be32 nfsd4_block_decode_layoutupdate(struct xdr_stream *xdr,
 		struct iomap **iomapp, int *nr_iomapsp, u32 block_size);
-__be32 nfsd4_scsi_decode_layoutupdate(__be32 *p, u32 len,
+__be32 nfsd4_scsi_decode_layoutupdate(struct xdr_stream *xdr,
 		struct iomap **iomapp, int *nr_iomapsp, u32 block_size);
 
 #endif /* _NFSD_BLOCKLAYOUTXDR_H */
fs/nfsd/nfs4proc.c +1 -1
index 04b8856d06157..656b2e7d88407 100644
@@ -2520,7 +2520,7 @@ nfsd4_layoutcommit(struct svc_rqst *rqstp,
 		lcp->lc_size_chg = false;
 	}
 
-	nfserr = ops->proc_layoutcommit(inode, lcp);
+	nfserr = ops->proc_layoutcommit(inode, rqstp, lcp);
 	nfs4_put_stid(&ls->ls_stid);
 out:
 	return nfserr;
fs/nfsd/nfs4xdr.c +5 -6
index cbbb61fcdd499..8b68f74a8cf08 100644
@@ -592,6 +592,8 @@ static __be32
 nfsd4_decode_layoutupdate4(struct nfsd4_compoundargs *argp,
 			   struct nfsd4_layoutcommit *lcp)
 {
+	u32 len;
+
 	if (xdr_stream_decode_u32(argp->xdr, &lcp->lc_layout_type) < 0)
 		return nfserr_bad_xdr;
 	if (lcp->lc_layout_type < LAYOUT_NFSV4_1_FILES)
@@ -599,13 +601,10 @@ nfsd4_decode_layoutupdate4(struct nfsd4_compoundargs *argp,
 	if (lcp->lc_layout_type >= LAYOUT_TYPE_MAX)
 		return nfserr_bad_xdr;
 
-	if (xdr_stream_decode_u32(argp->xdr, &lcp->lc_up_len) < 0)
+	if (xdr_stream_decode_u32(argp->xdr, &len) < 0)
+		return nfserr_bad_xdr;
+	if (!xdr_stream_subsegment(argp->xdr, &lcp->lc_up_layout, len))
 		return nfserr_bad_xdr;
-	if (lcp->lc_up_len > 0) {
-		lcp->lc_up_layout = xdr_inline_decode(argp->xdr, lcp->lc_up_len);
-		if (!lcp->lc_up_layout)
-			return nfserr_bad_xdr;
-	}
 
 	return nfs_ok;
 }
fs/nfsd/pnfs.h +1 -0
index 925817f669176..dfd411d1f363f 100644
@@ -35,6 +35,7 @@ struct nfsd4_layout_ops {
 			const struct nfsd4_layoutget *lgp);
 
 	__be32 (*proc_layoutcommit)(struct inode *inode,
+			struct svc_rqst *rqstp,
 			struct nfsd4_layoutcommit *lcp);
 
 	void (*fence_client)(struct nfs4_layout_stateid *ls,
fs/nfsd/xdr4.h +1 -2
index e65b552bf5f5a..d4b48602b2b0c 100644
@@ -664,8 +664,7 @@ struct nfsd4_layoutcommit {
 	u64			lc_last_wr;	/* request */
 	struct timespec64	lc_mtime;	/* request */
 	u32			lc_layout_type;	/* request */
-	u32			lc_up_len;	/* layout length */
-	void			*lc_up_layout;	/* decoded by callback */
+	struct xdr_buf		lc_up_layout;	/* decoded by callback */
 	bool			lc_size_chg;	/* response */
 	u64			lc_newsize;	/* response */
 };
Signature · 1 line
2.43.0
#2 lore raw

PATCH 2/2 NFSD: Fix last write offset handling in layoutcommit

In reply to Sergey Bashirov

The data type of loca_last_write_offset is newoffset4 and is switched on a boolean value, no_newoffset, that indicates if a previous write occurred or not. If no_newoffset is FALSE, an offset is not given. This means that client does not try to update the file size. Thus, server should not try to calculate new file size and check if it fits into the segment range. See RFC 8881, section 12.5.4.2.

Sometimes the current incorrect logic may cause clients to hang when trying to sync an inode. If layoutcommit fails, the client marks the inode as dirty again.

---
 fs/nfsd/blocklayout.c |  5 ++---
 fs/nfsd/nfs4proc.c    | 30 +++++++++++++++---------------
 2 files changed, 17 insertions(+), 18 deletions(-)
fs/nfsd/blocklayout.c +2 -3
index 4c936132eb440..0822d8a119c6f 100644
@@ -118,7 +118,6 @@ nfsd4_block_commit_blocks(struct inode *inode, struct nfsd4_layoutcommit *lcp,
 		struct iomap *iomaps, int nr_iomaps)
 {
 	struct timespec64 mtime = inode_get_mtime(inode);
-	loff_t new_size = lcp->lc_last_wr + 1;
 	struct iattr iattr = { .ia_valid = 0 };
 	int error;
 
@@ -128,9 +127,9 @@ nfsd4_block_commit_blocks(struct inode *inode, struct nfsd4_layoutcommit *lcp,
 	iattr.ia_valid |= ATTR_ATIME | ATTR_CTIME | ATTR_MTIME;
 	iattr.ia_atime = iattr.ia_ctime = iattr.ia_mtime = lcp->lc_mtime;
 
-	if (new_size > i_size_read(inode)) {
+	if (lcp->lc_size_chg) {
 		iattr.ia_valid |= ATTR_SIZE;
-		iattr.ia_size = new_size;
+		iattr.ia_size = lcp->lc_newsize;
 	}
 
 	error = inode->i_sb->s_export_op->commit_blocks(inode, iomaps,
fs/nfsd/nfs4proc.c +15 -15
index 656b2e7d88407..7043fc475458d 100644
@@ -2475,7 +2475,6 @@ nfsd4_layoutcommit(struct svc_rqst *rqstp,
 	const struct nfsd4_layout_seg *seg = &lcp->lc_seg;
 	struct svc_fh *current_fh = &cstate->current_fh;
 	const struct nfsd4_layout_ops *ops;
-	loff_t new_size = lcp->lc_last_wr + 1;
 	struct inode *inode;
 	struct nfs4_layout_stateid *ls;
 	__be32 nfserr;
@@ -2491,13 +2490,21 @@ nfsd4_layoutcommit(struct svc_rqst *rqstp,
 		goto out;
 	inode = d_inode(current_fh->fh_dentry);
 
-	nfserr = nfserr_inval;
-	if (new_size <= seg->offset)
-		goto out;
-	if (new_size > seg->offset + seg->length)
-		goto out;
-	if (!lcp->lc_newoffset && new_size > i_size_read(inode))
-		goto out;
+	lcp->lc_size_chg = false;
+	if (lcp->lc_newoffset) {
+		loff_t new_size = lcp->lc_last_wr + 1;
+
+		nfserr = nfserr_inval;
+		if (new_size <= seg->offset)
+			goto out;
+		if (new_size > seg->offset + seg->length)
+			goto out;
+
+		if (new_size > i_size_read(inode)) {
+			lcp->lc_size_chg = true;
+			lcp->lc_newsize = new_size;
+		}
+	}
 
 	nfserr = nfsd4_preprocess_layout_stateid(rqstp, cstate, &lcp->lc_sid,
 						false, lcp->lc_layout_type,
@@ -2513,13 +2520,6 @@ nfsd4_layoutcommit(struct svc_rqst *rqstp,
 	/* LAYOUTCOMMIT does not require any serialization */
 	mutex_unlock(&ls->ls_mutex);
 
-	if (new_size > i_size_read(inode)) {
-		lcp->lc_size_chg = true;
-		lcp->lc_newsize = new_size;
-	} else {
-		lcp->lc_size_chg = false;
-	}
-
 	nfserr = ops->proc_layoutcommit(inode, rqstp, lcp);
 	nfs4_put_stid(&ls->ls_stid);
 out:
Signature · 1 line
2.43.0
maintainer Reviewed-by on v1 2/2
#3 lore raw

Re: [PATCH 2/2] NFSD: Fix last write offset handling in layoutcommit

Quoted 3 lines · Sergey Bashirov

The data type of loca_last_write_offset is newoffset4 and is switched on a boolean value, no_newoffset, that indicates if a previous write occurred or not. If no_newoffset is FALSE, an offset is not given.

Gah, what a confusing label in the spec. I went down the rabbit hole of trying to understand why "no_newoffset" being TRUE means that you do get an offset. The "no_" in this case is just a prefix for "union newoffset", and not a negation of the flag.

Quoted 88 lines

This means that client does not try to update the file size. Thus, server should not try to calculate new file size and check if it fits into the segment range. See RFC 8881, section 12.5.4.2.

Sometimes the current incorrect logic may cause clients to hang when trying to sync an inode. If layoutcommit fails, the client marks the inode as dirty again.

---
 fs/nfsd/blocklayout.c |  5 ++---
 fs/nfsd/nfs4proc.c    | 30 +++++++++++++++---------------
 2 files changed, 17 insertions(+), 18 deletions(-)
fs/nfsd/blocklayout.c +2 -3
index 4c936132eb440..0822d8a119c6f 100644
@@ -118,7 +118,6 @@ nfsd4_block_commit_blocks(struct inode *inode, struct nfsd4_layoutcommit *lcp,
 		struct iomap *iomaps, int nr_iomaps)
 {
 	struct timespec64 mtime = inode_get_mtime(inode);
-	loff_t new_size = lcp->lc_last_wr + 1;
 	struct iattr iattr = { .ia_valid = 0 };
 	int error;
 
@@ -128,9 +127,9 @@ nfsd4_block_commit_blocks(struct inode *inode, struct nfsd4_layoutcommit *lcp,
 	iattr.ia_valid |= ATTR_ATIME | ATTR_CTIME | ATTR_MTIME;
 	iattr.ia_atime = iattr.ia_ctime = iattr.ia_mtime = lcp->lc_mtime;
 
-	if (new_size > i_size_read(inode)) {
+	if (lcp->lc_size_chg) {
 		iattr.ia_valid |= ATTR_SIZE;
-		iattr.ia_size = new_size;
+		iattr.ia_size = lcp->lc_newsize;
 	}
 
 	error = inode->i_sb->s_export_op->commit_blocks(inode, iomaps,
fs/nfsd/nfs4proc.c +15 -8
index 656b2e7d88407..7043fc475458d 100644
@@ -2475,7 +2475,6 @@ nfsd4_layoutcommit(struct svc_rqst *rqstp,
 	const struct nfsd4_layout_seg *seg = &lcp->lc_seg;
 	struct svc_fh *current_fh = &cstate->current_fh;
 	const struct nfsd4_layout_ops *ops;
-	loff_t new_size = lcp->lc_last_wr + 1;
 	struct inode *inode;
 	struct nfs4_layout_stateid *ls;
 	__be32 nfserr;
@@ -2491,13 +2490,21 @@ nfsd4_layoutcommit(struct svc_rqst *rqstp,
 		goto out;
 	inode = d_inode(current_fh->fh_dentry);
 
-	nfserr = nfserr_inval;
-	if (new_size <= seg->offset)
-		goto out;
-	if (new_size > seg->offset + seg->length)
-		goto out;
-	if (!lcp->lc_newoffset && new_size > i_size_read(inode))
-		goto out;
+	lcp->lc_size_chg = false;
+	if (lcp->lc_newoffset) {
+		loff_t new_size = lcp->lc_last_wr + 1;
+
+		nfserr = nfserr_inval;
+		if (new_size <= seg->offset)
+			goto out;
+		if (new_size > seg->offset + seg->length)
+			goto out;
+
+		if (new_size > i_size_read(inode)) {
+			lcp->lc_size_chg = true;
+			lcp->lc_newsize = new_size;
+		}
+	}
 
 	nfserr = nfsd4_preprocess_layout_stateid(rqstp, cstate, &lcp->lc_sid,
 						false, lcp->lc_layout_type,
@@ -2513,13 +2520,6 @@ nfsd4_layoutcommit(struct svc_rqst *rqstp,
 	/* LAYOUTCOMMIT does not require any serialization */
 	mutex_unlock(&ls->ls_mutex);
 

fs/nfsd/nfs4proc.c nfsd4_layoutcommit(struct svc_rqst *rqstp, · line 2525

Quoted 10 lines
fs/nfsd/nfs4proc.c +0 -7
-	if (new_size > i_size_read(inode)) {
-		lcp->lc_size_chg = true;
-		lcp->lc_newsize = new_size;
-	} else {
-		lcp->lc_size_chg = false;
-	}
-
 	nfserr = ops->proc_layoutcommit(inode, rqstp, lcp);
 	nfs4_put_stid(&ls->ls_stid);
 out:

Code looks correct to me though.

maintainer
#4 lore raw
In reply to Sergey Bashirov

From: Chuck Lever <[email protected]>

Quoted 6 lines · Sergey Bashirov

This series resubmits two patches that were dropped due to conflicts after the deviceid4 handling rework. The first patch is rebased. Only the decoding of bex.vol_id, which has the deviceid4 type, is updated. The second patch doesn't change, it just depends on the first one.

Applied to nfsd-testing, thanks!

[1/2] NFSD: Implement large extent array support in pNFS

      commit: 0c8cc94a6dc2b200617094c0016cd46f01c12490
[2/2] NFSD: Fix last write offset handling in layoutcommit
      commit: f91f48cedc8ef26d55a548e4e3879070d06906f6

--
Chuck Lever