Skip to contentSkip to conversation

nfs/blocklayout: allow pNFS SCSI layout without pNFS block layout

PATCH In review Scott Mayhew 2 participants 2 messages Aug 27 – Aug 31

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 nfs/blocklayout: allow pNFS SCSI layout without pNFS block layout

The pNFS block-volume layout (LAYOUT_BLOCK_VOLUME, RFC 5663) and the pNFS SCSI layout (LAYOUT_SCSI, RFC 8154) are built from a single blocklayout driver. Until now CONFIG_PNFS_BLOCK was an unprompted symbol that pulled in both layout types whenever NFS_V4 and BLK_DEV_DM were enabled, so a kernel could not offer SCSI layout without also enabling block layout and device-mapper.

Split the two into separate, independently selectable options. Both types share almost the entire driver (read/write path, extent tree, and the device tree including the composite SLICE/CONCAT/STRIPE volume types), so keep a single blocklayoutdriver module and compile only the type-specific pieces in or out:

- block.c SIMPLE volume parsing, which resolves the device via the
userspace pipefs daemon (rpc_pipefs.c) - block only.
- scsi.c SCSI leaf volume parsing and persistent-reservation
register/unregister - SCSI only.

The remaining type-specific call sites in the shared core are guarded with IS_ENABLED() so the reference to a not-built symbol is eliminated, and the now cross-file helpers (bl_map_simple, bl_parse_simple, bl_parse_scsi, bl_register_scsi, bl_unregister_scsi) are made global with prototypes in blocklayout.h.

CONFIG_PNFS_BLOCK and CONFIG_PNFS_SCSI_LAYOUT are now prompted booleans that only select which volume types are compiled in. Module linkage is carried by a single internal tristate, CONFIG_PNFS_BLOCK_LAYOUT_DRIVER, which follows NFS_V4; making the two options booleans avoids a y/m mismatch where a built-in composite would silently drop a module-only object. SCSI layout defaults on and no longer depends on BLK_DEV_DM; block layout now defaults off.

blocklayout.h no longer includes device-mapper.h (it was unused); it includes blkdev.h for the block_device/sector_t/BLK_OPEN_* definitions used in the header.

Assisted-by Claude:claude-opus-4-8
---
 fs/nfs/Kconfig                   |  20 ++-
 fs/nfs/Makefile                  |   2 +-
 fs/nfs/blocklayout/Makefile      |  12 +-
 fs/nfs/blocklayout/block.c       |  40 ++++++
 fs/nfs/blocklayout/blocklayout.c |  43 ++++---
 fs/nfs/blocklayout/blocklayout.h |  14 ++-
 fs/nfs/blocklayout/dev.c         | 209 +++----------------------------
 fs/nfs/blocklayout/scsi.c        | 175 ++++++++++++++++++++++++++
 8 files changed, 300 insertions(+), 215 deletions(-)
 create mode 100644 fs/nfs/blocklayout/block.c
 create mode 100644 fs/nfs/blocklayout/scsi.c
fs/nfs/Kconfig +19 -1
index 6bb30543eff0..9166ac623ba8 100644
@@ -122,9 +122,27 @@ config PNFS_FILE_LAYOUT
 	default NFS_V4
 
 config PNFS_BLOCK
-	tristate
+	bool "pNFS block layout support"
 	depends on NFS_V4 && BLK_DEV_DM
+	default n
+	help
+	  Enable support for the pNFS block-volume layout type (RFC 5663).
+
+config PNFS_SCSI_LAYOUT
+	bool "pNFS SCSI layout support"
+	depends on NFS_V4
 	default NFS_V4
+	help
+	  Enable support for the pNFS SCSI layout type (RFC 8154). This does
+	  not require device-mapper and can be enabled without PNFS_BLOCK.
+
+# Internal: the shared block/scsi layout driver module, built when either layout
+# type is enabled. The two options above are booleans that only select which
+# volume types are compiled in; this symbol is the single modularity knob and
+# follows NFS_V4, so the composite module never lands in both obj-y and obj-m.
+config PNFS_BLOCK_LAYOUT_DRIVER
+	tristate
+	default NFS_V4 if PNFS_BLOCK || PNFS_SCSI_LAYOUT
 
 config PNFS_FLEXFILE_LAYOUT
 	tristate
fs/nfs/Makefile +1 -1
index c895521f27f3..49c06b759b60 100644
@@ -34,5 +34,5 @@ nfsv4-$(CONFIG_NFS_V4_0)	+= nfs40client.o nfs40proc.o
 nfsv4-$(CONFIG_NFS_V4_2)	+= nfs42proc.o nfs42xattr.o
 
 obj-$(CONFIG_PNFS_FILE_LAYOUT) += filelayout/
-obj-$(CONFIG_PNFS_BLOCK) += blocklayout/
+obj-$(CONFIG_PNFS_BLOCK_LAYOUT_DRIVER) += blocklayout/
 obj-$(CONFIG_PNFS_FLEXFILE_LAYOUT) += flexfilelayout/
fs/nfs/blocklayout/Makefile +10 -2
index 7668a1bfb5fa..832012d1c1fc 100644
@@ -2,6 +2,14 @@
 #
 # Makefile for the pNFS block layout driver kernel module
 #
-obj-$(CONFIG_PNFS_BLOCK) += blocklayoutdriver.o
+obj-$(CONFIG_PNFS_BLOCK_LAYOUT_DRIVER) += blocklayoutdriver.o
 
-blocklayoutdriver-y += blocklayout.o dev.o extent_tree.o rpc_pipefs.o
+# Shared core: device tree incl. composite SLICE/CONCAT/STRIPE volumes (used by
+# both layout types), extent tree, common layout ops, init/exit.
+blocklayoutdriver-y += blocklayout.o dev.o extent_tree.o
+
+# Block-layout-specific: SIMPLE volumes resolved via the userspace pipefs daemon.
+blocklayoutdriver-$(CONFIG_PNFS_BLOCK) += block.o rpc_pipefs.o
+
+# SCSI-layout-specific: SCSI leaf volume parse and PR register/unregister.
+blocklayoutdriver-$(CONFIG_PNFS_SCSI_LAYOUT) += scsi.o
fs/nfs/blocklayout/block.c new file +40 -0
new file mode 100644
index 000000000000..530373a71c98
@@ -0,0 +1,40 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (c) 2014-2016 Christoph Hellwig.
+ */
+#include <linux/blkdev.h>
+#include <linux/nfs4.h>
+#include <linux/nfs_fs.h>
+#include <linux/nfs_xdr.h>
+
+#include "blocklayout.h"
+
+#define NFSDBG_FACILITY		NFSDBG_PNFS_LD
+
+int
+bl_parse_simple(struct nfs_server *server, struct pnfs_block_dev *d,
+		struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask)
+{
+	struct pnfs_block_volume *v = &volumes[idx];
+	struct file *bdev_file;
+	dev_t dev;
+
+	dev = bl_resolve_deviceid(server, v, gfp_mask);
+	if (!dev)
+		return -EIO;
+
+	bdev_file = bdev_file_open_by_dev(dev, BLK_OPEN_READ | BLK_OPEN_WRITE,
+				       NULL, NULL);
+	if (IS_ERR(bdev_file)) {
+		printk(KERN_WARNING "pNFS: failed to open device %d:%d (%ld)\n",
+			MAJOR(dev), MINOR(dev), PTR_ERR(bdev_file));
+		return PTR_ERR(bdev_file);
+	}
+	d->bdev_file = bdev_file;
+	d->len = bdev_nr_bytes(file_bdev(bdev_file));
+	d->map = bl_map_simple;
+
+	printk(KERN_INFO "pNFS: using block device %s\n",
+		file_bdev(bdev_file)->bd_disk->disk_name);
+	return 0;
+}
fs/nfs/blocklayout/blocklayout.c +28 -15

68 lines not on this page · raw message

fs/nfs/blocklayout/blocklayout.h +13 -1
index 6da40ca19570..07740bbfeb67 100644
@@ -32,7 +32,7 @@
 #ifndef FS_NFS_NFS4BLOCKLAYOUT_H
 #define FS_NFS_NFS4BLOCKLAYOUT_H
 
-#include <linux/device-mapper.h>
+#include <linux/blkdev.h>
 #include <linux/nfs_fs.h>
 #include <linux/sunrpc/rpc_pipe_fs.h>
 
@@ -182,6 +182,18 @@ bool bl_register_dev(struct pnfs_block_dev *d);
 struct nfs4_deviceid_node *bl_alloc_deviceid_node(struct nfs_server *server,
 		struct pnfs_device *pdev, gfp_t gfp_mask);
 void bl_free_deviceid_node(struct nfs4_deviceid_node *d);
+bool bl_map_simple(struct pnfs_block_dev *dev, u64 offset,
+		struct pnfs_block_dev_map *map);
+
+/* block.c */
+int bl_parse_simple(struct nfs_server *server, struct pnfs_block_dev *d,
+		struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask);
+
+/* scsi.c */
+int bl_parse_scsi(struct nfs_server *server, struct pnfs_block_dev *d,
+		struct pnfs_block_volume *volumes, int idx, gfp_t gfp_mask);
+bool bl_register_scsi(struct pnfs_block_dev *dev);
+void bl_unregister_scsi(struct pnfs_block_dev *dev);
 
 /* extent_tree.c */
 int ext_tree_insert(struct pnfs_block_layout *bl,
fs/nfs/blocklayout/dev.c +14 -195

263 lines not on this page · raw message

fs/nfs/blocklayout/scsi.c new file +175 -0

175 lines not on this page · raw message

Signature · 1 line
2.55.0
#1 lore raw

Supporting SCSI layout with block layout makes totaly sense, but please do it without all the gracious renaming and annoying AI slop commit log. Take a look at show nfsd handled this with CONFIG_NFSD_BLOCKLAYOUT and CONFIG_NFSD_SCSILAYOUT as that is the model that applies here as well.