Skip to contentSkip to conversation

merge struct block_device and struct hd_struct v2

PATCH In review Christoph Hellwig 7 participants 90 messages Nov 24, 2020 – Nov 26, 2020

Subsystems (27)
Memory management VFS SCSI USB +23 more subsystems
lore
Share & tools
lore

Christoph Hellwig

author maintainer
#0 lore raw

Hi Jens,

this series cleans up our main per-device node data structure by merging the block_device and hd_struct data structures that have the same scope, but different life times. The main effect (besides removing lots of code) is that instead of having two device sizes that need complex synchronization there is just one now.

Note that this now includes the previous "misc cleanups" series as I had to fix up a thing in there with the changed patch ordering.

The first patch already is in 5.10-rc, but not in for-5.11/block

A git tree is available here:

git://git.infradead.org/users/hch/block.git bdev-lookup

Gitweb:

http://git.infradead.org/users/hch/block.git/shortlog/refs/heads/bdev-lookup

Changes since v1:
- spelling fixes
- fix error unwinding in __alloc_disk_node
- use bdev_is_partition in a few more places
- don't send the RESIZE=1 uevent for hidden gendisks
- rename __bdget_disk to disk_find_part
- drop a bcache patch
- some patch reordering
- add more refactoring
- use rcu protection to prevent racing with a disk going away in blkdev_get
- split up some of the big patches into many small ones
- clean up the freeze_bdev interface

Diffstat:

 block/bio.c                                  |    6 
 block/blk-cgroup.c                           |   50 -
 block/blk-core.c                             |   68 +-
 block/blk-flush.c                            |    2 
 block/blk-iocost.c                           |   36 -
 block/blk-lib.c                              |    2 
 block/blk-merge.c                            |    2 
 block/blk-mq.c                               |    9 
 block/blk-mq.h                               |    7 
 block/blk.h                                  |   84 ---
 block/genhd.c                                |  467 ++++-------------
 block/ioctl.c                                |   14 
 block/partitions/core.c                      |  252 +++------
 drivers/block/drbd/drbd_receiver.c           |    2 
 drivers/block/drbd/drbd_worker.c             |    3 
 drivers/block/loop.c                         |   24 
 drivers/block/mtip32xx/mtip32xx.c            |   15 
 drivers/block/mtip32xx/mtip32xx.h            |    2 
 drivers/block/nbd.c                          |    6 
 drivers/block/xen-blkback/common.h           |    4 
 drivers/block/xen-blkfront.c                 |   20 
 drivers/block/zram/zram_drv.c                |   87 ---
 drivers/block/zram/zram_drv.h                |    1 
 drivers/md/bcache/request.c                  |    4 
 drivers/md/bcache/super.c                    |   29 -
 drivers/md/dm-core.h                         |    7 
 drivers/md/dm-table.c                        |    9 
 drivers/md/dm.c                              |   45 -
 drivers/md/md.c                              |    8 
 drivers/mtd/mtdsuper.c                       |   17 
 drivers/nvme/target/admin-cmd.c              |   20 
 drivers/s390/block/dasd.c                    |    8 
 drivers/s390/block/dasd_ioctl.c              |    9 
 drivers/scsi/scsicam.c                       |    2 
 drivers/target/target_core_file.c            |    6 
 drivers/target/target_core_pscsi.c           |    7 
 drivers/usb/gadget/function/storage_common.c |    8 
 fs/block_dev.c                               |  730 +++++++++------------------
 fs/btrfs/sysfs.c                             |   15 
 fs/btrfs/volumes.c                           |   13 
 fs/buffer.c                                  |    2 
 fs/ext4/ioctl.c                              |    2 
 fs/ext4/super.c                              |   18 
 fs/ext4/sysfs.c                              |   10 
 fs/f2fs/checkpoint.c                         |    5 
 fs/f2fs/f2fs.h                               |    2 
 fs/f2fs/file.c                               |   14 
 fs/f2fs/super.c                              |    8 
 fs/f2fs/sysfs.c                              |    9 
 fs/inode.c                                   |    3 
 fs/internal.h                                |    7 
 fs/io_uring.c                                |   10 
 fs/pipe.c                                    |    5 
 fs/pstore/blk.c                              |    2 
 fs/quota/quota.c                             |   40 +
 fs/statfs.c                                  |    2 
 fs/super.c                                   |   86 ---
 fs/xfs/xfs_fsops.c                           |    7 
 include/linux/blk-cgroup.h                   |    4 
 include/linux/blk_types.h                    |   24 
 include/linux/blkdev.h                       |   27 
 include/linux/fs.h                           |    5 
 include/linux/genhd.h                        |  110 ----
 include/linux/part_stat.h                    |   45 -
 init/do_mounts.c                             |  271 ++++------
 kernel/trace/blktrace.c                      |   54 -
 mm/filemap.c                                 |   13 
 67 files changed, 957 insertions(+), 1928 deletions(-)
author maintainer
#1 lore raw

PATCH 1/45 blk-cgroup: fix a hd_struct leak in blkcg_fill_root_iostats

disk_get_part needs to be paired with a disk_put_part.

Fixes ef45fe470e1 ("blk-cgroup: show global disk stats in root cgroup io.stat")
---
 block/blk-cgroup.c | 1 +
 1 file changed, 1 insertion(+)
block/blk-cgroup.c +1 -0
index c68bdf58c9a6e1..54fbe1e80cc41a 100644
@@ -849,6 +849,7 @@ static void blkcg_fill_root_iostats(void)
 			blkg_iostat_set(&blkg->iostat.cur, &tmp);
 			u64_stats_update_end(&blkg->iostat.sync);
 		}
+		disk_put_part(part);
 	}
 }
 
Signature · 1 line
2.29.2
Reviewed-by on v1 1/45
#2 lore raw

Re: [PATCH 01/45] blk-cgroup: fix a hd_struct leak in blkcg_fill_root_iostats

Quoted 8 lines · Christoph Hellwig

disk_get_part needs to be paired with a disk_put_part.

Fixes ef45fe470e1 ("blk-cgroup: show global disk stats in root cgroup io.stat")
Quoted 4 lines
---
  block/blk-cgroup.c | 1 +
  1 file changed, 1 insertion(+)

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
author maintainer
#3 lore raw

PATCH 2/45 filemap: consistently use ->f_mapping over ->i_mapping

In reply to Christoph Hellwig

Use file->f_mapping in all remaining places that have a struct file available to properly handle the case where inode->i_mapping != file_inode(file)->i_mapping.

---
 mm/filemap.c | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)
mm/filemap.c +6 -7
index d5e7c2029d16b4..4f583489aa3c2a 100644
@@ -2886,14 +2886,14 @@ EXPORT_SYMBOL(filemap_map_pages);
 
 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
 {
+	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
 	struct page *page = vmf->page;
-	struct inode *inode = file_inode(vmf->vma->vm_file);
 	vm_fault_t ret = VM_FAULT_LOCKED;
 
-	sb_start_pagefault(inode->i_sb);
+	sb_start_pagefault(mapping->host->i_sb);
 	file_update_time(vmf->vma->vm_file);
 	lock_page(page);
-	if (page->mapping != inode->i_mapping) {
+	if (page->mapping != mapping) {
 		unlock_page(page);
 		ret = VM_FAULT_NOPAGE;
 		goto out;
@@ -2906,7 +2906,7 @@ vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
 	set_page_dirty(page);
 	wait_for_stable_page(page);
 out:
-	sb_end_pagefault(inode->i_sb);
+	sb_end_pagefault(mapping->host->i_sb);
 	return ret;
 }
 
@@ -3149,10 +3149,9 @@ void dio_warn_stale_pagecache(struct file *filp)
 {
 	static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
 	char pathname[128];
-	struct inode *inode = file_inode(filp);
 	char *path;
 
-	errseq_set(&inode->i_mapping->wb_err, -EIO);
+	errseq_set(&filp->f_mapping->wb_err, -EIO);
 	if (__ratelimit(&_rs)) {
 		path = file_path(filp, pathname, sizeof(pathname));
 		if (IS_ERR(path))
@@ -3179,7 +3178,7 @@ generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
 
 	if (iocb->ki_flags & IOCB_NOWAIT) {
 		/* If there are pages to writeback, return */
-		if (filemap_range_has_page(inode->i_mapping, pos,
+		if (filemap_range_has_page(file->f_mapping, pos,
 					   pos + write_len - 1))
 			return -EAGAIN;
 	} else {
Signature · 1 line
2.29.2
Reviewed-by on v1 2/45
#5 lore raw

Re: [PATCH 02/45] filemap: consistently use ->f_mapping over ->i_mapping

In reply to Christoph Hellwig
Quoted 9 lines · Christoph Hellwig

Use file->f_mapping in all remaining places that have a struct file available to properly handle the case where inode->i_mapping != file_inode(file)->i_mapping.

---
  mm/filemap.c | 13 ++++++-------
  1 file changed, 6 insertions(+), 7 deletions(-)

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
maintainer Reviewed-by on v1 2/45
#6 lore raw

Re: [PATCH 02/45] filemap: consistently use ->f_mapping over ->i_mapping

In reply to Christoph Hellwig
Quoted 5 lines · Christoph Hellwig

Use file->f_mapping in all remaining places that have a struct file available to properly handle the case where inode->i_mapping != file_inode(file)->i_mapping.

Looks good to me. You can add:

Honza

Quoted 59 lines
---
 mm/filemap.c | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)
mm/filemap.c +6 -7
index d5e7c2029d16b4..4f583489aa3c2a 100644
@@ -2886,14 +2886,14 @@ EXPORT_SYMBOL(filemap_map_pages);
 
 vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
 {
+	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
 	struct page *page = vmf->page;
-	struct inode *inode = file_inode(vmf->vma->vm_file);
 	vm_fault_t ret = VM_FAULT_LOCKED;
 
-	sb_start_pagefault(inode->i_sb);
+	sb_start_pagefault(mapping->host->i_sb);
 	file_update_time(vmf->vma->vm_file);
 	lock_page(page);
-	if (page->mapping != inode->i_mapping) {
+	if (page->mapping != mapping) {
 		unlock_page(page);
 		ret = VM_FAULT_NOPAGE;
 		goto out;
@@ -2906,7 +2906,7 @@ vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
 	set_page_dirty(page);
 	wait_for_stable_page(page);
 out:
-	sb_end_pagefault(inode->i_sb);
+	sb_end_pagefault(mapping->host->i_sb);
 	return ret;
 }
 
@@ -3149,10 +3149,9 @@ void dio_warn_stale_pagecache(struct file *filp)
 {
 	static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
 	char pathname[128];
-	struct inode *inode = file_inode(filp);
 	char *path;
 
-	errseq_set(&inode->i_mapping->wb_err, -EIO);
+	errseq_set(&filp->f_mapping->wb_err, -EIO);
 	if (__ratelimit(&_rs)) {
 		path = file_path(filp, pathname, sizeof(pathname));
 		if (IS_ERR(path))
@@ -3179,7 +3178,7 @@ generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
 
 	if (iocb->ki_flags & IOCB_NOWAIT) {
 		/* If there are pages to writeback, return */
-		if (filemap_range_has_page(inode->i_mapping, pos,
+		if (filemap_range_has_page(file->f_mapping, pos,
 					   pos + write_len - 1))
 			return -EAGAIN;
 	} else {
Signature · 1 line
2.29.2
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author maintainer
#7 lore raw

PATCH 3/45 fs: remove get_super_thawed and get_super_exclusive_thawed

In reply to Christoph Hellwig

Just open code the wait in the only caller of both functions.

---
 fs/internal.h      |  2 ++
 fs/quota/quota.c   | 31 +++++++++++++++++++++-------
 fs/super.c         | 51 ++--------------------------------------------
 include/linux/fs.h |  4 +---
 4 files changed, 29 insertions(+), 59 deletions(-)
fs/internal.h +2 -0
index a7cd0f64faa4ab..47be21dfeebef5 100644
@@ -114,7 +114,9 @@ extern struct file *alloc_empty_file_noaccount(int, const struct cred *);
  */
 extern int reconfigure_super(struct fs_context *);
 extern bool trylock_super(struct super_block *sb);
+struct super_block *__get_super(struct block_device *bdev, bool excl);
 extern struct super_block *user_get_super(dev_t);
+void put_super(struct super_block *sb);
 extern bool mount_capable(struct fs_context *);
 
 /*
fs/quota/quota.c +24 -7
index 9af95c7a0bbe3c..f3d32b0d9008f2 100644
@@ -20,6 +20,7 @@
 #include <linux/writeback.h>
 #include <linux/nospec.h>
 #include "compat.h"
+#include "../internal.h"
 
 static int check_quotactl_permission(struct super_block *sb, int type, int cmd,
 				     qid_t id)
@@ -868,6 +869,7 @@ static struct super_block *quotactl_block(const char __user *special, int cmd)
 	struct block_device *bdev;
 	struct super_block *sb;
 	struct filename *tmp = getname(special);
+	bool excl = false, thawed = false;
 
 	if (IS_ERR(tmp))
 		return ERR_CAST(tmp);
@@ -875,17 +877,32 @@ static struct super_block *quotactl_block(const char __user *special, int cmd)
 	putname(tmp);
 	if (IS_ERR(bdev))
 		return ERR_CAST(bdev);
-	if (quotactl_cmd_onoff(cmd))
-		sb = get_super_exclusive_thawed(bdev);
-	else if (quotactl_cmd_write(cmd))
-		sb = get_super_thawed(bdev);
-	else
-		sb = get_super(bdev);
+
+	if (quotactl_cmd_onoff(cmd)) {
+		excl = true;
+		thawed = true;
+	} else if (quotactl_cmd_write(cmd)) {
+		thawed = true;
+	}
+
+retry:
+	sb = __get_super(bdev, excl);
+	if (thawed && sb && sb->s_writers.frozen != SB_UNFROZEN) {
+		if (excl)
+			up_write(&sb->s_umount);
+		else
+			up_read(&sb->s_umount);
+		wait_event(sb->s_writers.wait_unfrozen,
+			   sb->s_writers.frozen == SB_UNFROZEN);
+		put_super(sb);
+		goto retry;
+	}
+
 	bdput(bdev);
 	if (!sb)
 		return ERR_PTR(-ENODEV);
-
 	return sb;
+
 #else
 	return ERR_PTR(-ENODEV);
 #endif
fs/super.c +2 -49

69 lines not on this page · raw message

include/linux/fs.h +1 -3
index 8667d0cdc71e76..a61df0dd4f1989 100644
@@ -1409,7 +1409,7 @@ enum {
 
 struct sb_writers {
 	int				frozen;		/* Is sb frozen? */
-	wait_queue_head_t		wait_unfrozen;	/* for get_super_thawed() */
+	wait_queue_head_t		wait_unfrozen;	/* wait for thaw */
 	struct percpu_rw_semaphore	rw_sem[SB_FREEZE_LEVELS];
 };
 
@@ -3132,8 +3132,6 @@ extern struct file_system_type *get_filesystem(struct file_system_type *fs);
 extern void put_filesystem(struct file_system_type *fs);
 extern struct file_system_type *get_fs_type(const char *name);
 extern struct super_block *get_super(struct block_device *);
-extern struct super_block *get_super_thawed(struct block_device *);
-extern struct super_block *get_super_exclusive_thawed(struct block_device *bdev);
 extern struct super_block *get_active_super(struct block_device *bdev);
 extern void drop_super(struct super_block *sb);
 extern void drop_super_exclusive(struct super_block *sb);
Signature · 1 line
2.29.2
Reviewed-by on v1 3/45
#8 lore raw

Re: [PATCH 03/45] fs: remove get_super_thawed and get_super_exclusive_thawed

Quoted 10 lines · Christoph Hellwig

Just open code the wait in the only caller of both functions.

---
  fs/internal.h      |  2 ++
  fs/quota/quota.c   | 31 +++++++++++++++++++++-------
  fs/super.c         | 51 ++--------------------------------------------
  include/linux/fs.h |  4 +---
  4 files changed, 29 insertions(+), 59 deletions(-)

Cheers,

Hannes

Signature · 4 lines
Dr. Hannes Reinecke                Kernel Storage Architect
[email protected]                              +49 911 74053 688
SUSE Software Solutions GmbH, Maxfeldstr. 5, 90409 Nürnberg
HRB 36809 (AG Nürnberg), Geschäftsführer: Felix Imendörffer
maintainer Reviewed-by on v1 3/45
#9 lore raw

Re: [PATCH 03/45] fs: remove get_super_thawed and get_super_exclusive_thawed

In reply to Christoph Hellwig
Quoted 3 lines · Christoph Hellwig

Just open code the wait in the only caller of both functions.

Fine by me. You can add:

Honza

Quoted 182 lines
---
 fs/internal.h      |  2 ++
 fs/quota/quota.c   | 31 +++++++++++++++++++++-------
 fs/super.c         | 51 ++--------------------------------------------
 include/linux/fs.h |  4 +---
 4 files changed, 29 insertions(+), 59 deletions(-)
fs/internal.h +2 -0
index a7cd0f64faa4ab..47be21dfeebef5 100644
@@ -114,7 +114,9 @@ extern struct file *alloc_empty_file_noaccount(int, const struct cred *);
  */
 extern int reconfigure_super(struct fs_context *);
 extern bool trylock_super(struct super_block *sb);
+struct super_block *__get_super(struct block_device *bdev, bool excl);
 extern struct super_block *user_get_super(dev_t);
+void put_super(struct super_block *sb);
 extern bool mount_capable(struct fs_context *);
 
 /*
fs/quota/quota.c +24 -7
index 9af95c7a0bbe3c..f3d32b0d9008f2 100644
@@ -20,6 +20,7 @@
 #include <linux/writeback.h>
 #include <linux/nospec.h>
 #include "compat.h"
+#include "../internal.h"
 
 static int check_quotactl_permission(struct super_block *sb, int type, int cmd,
 				     qid_t id)
@@ -868,6 +869,7 @@ static struct super_block *quotactl_block(const char __user *special, int cmd)
 	struct block_device *bdev;
 	struct super_block *sb;
 	struct filename *tmp = getname(special);
+	bool excl = false, thawed = false;
 
 	if (IS_ERR(tmp))
 		return ERR_CAST(tmp);
@@ -875,17 +877,32 @@ static struct super_block *quotactl_block(const char __user *special, int cmd)
 	putname(tmp);
 	if (IS_ERR(bdev))
 		return ERR_CAST(bdev);
-	if (quotactl_cmd_onoff(cmd))
-		sb = get_super_exclusive_thawed(bdev);
-	else if (quotactl_cmd_write(cmd))
-		sb = get_super_thawed(bdev);
-	else
-		sb = get_super(bdev);
+
+	if (quotactl_cmd_onoff(cmd)) {
+		excl = true;
+		thawed = true;
+	} else if (quotactl_cmd_write(cmd)) {
+		thawed = true;
+	}
+
+retry:
+	sb = __get_super(bdev, excl);
+	if (thawed && sb && sb->s_writers.frozen != SB_UNFROZEN) {
+		if (excl)
+			up_write(&sb->s_umount);
+		else
+			up_read(&sb->s_umount);
+		wait_event(sb->s_writers.wait_unfrozen,
+			   sb->s_writers.frozen == SB_UNFROZEN);
+		put_super(sb);
+		goto retry;
+	}
+
 	bdput(bdev);
 	if (!sb)
 		return ERR_PTR(-ENODEV);
-
 	return sb;
+
 #else
 	return ERR_PTR(-ENODEV);
 #endif
fs/super.c +2 -49
index 98bb0629ee108e..343e5c1e538d2a 100644
@@ -307,7 +307,7 @@ static void __put_super(struct super_block *s)
  *	Drops a temporary reference, frees superblock if there's no
  *	references left.
  */
-static void put_super(struct super_block *sb)
+void put_super(struct super_block *sb)
 {
 	spin_lock(&sb_lock);
 	__put_super(sb);
@@ -740,7 +740,7 @@ void iterate_supers_type(struct file_system_type *type,
 
 EXPORT_SYMBOL(iterate_supers_type);
 
-static struct super_block *__get_super(struct block_device *bdev, bool excl)
+struct super_block *__get_super(struct block_device *bdev, bool excl)
 {
 	struct super_block *sb;
 
@@ -789,53 +789,6 @@ struct super_block *get_super(struct block_device *bdev)
 }
 EXPORT_SYMBOL(get_super);
 
-static struct super_block *__get_super_thawed(struct block_device *bdev,
-					      bool excl)
-{
-	while (1) {
-		struct super_block *s = __get_super(bdev, excl);
-		if (!s || s->s_writers.frozen == SB_UNFROZEN)
-			return s;
-		if (!excl)
-			up_read(&s->s_umount);
-		else
-			up_write(&s->s_umount);
-		wait_event(s->s_writers.wait_unfrozen,
-			   s->s_writers.frozen == SB_UNFROZEN);
-		put_super(s);
-	}
-}
-
-/**
- *	get_super_thawed - get thawed superblock of a device
- *	@bdev: device to get the superblock for
- *
- *	Scans the superblock list and finds the superblock of the file system
- *	mounted on the device. The superblock is returned once it is thawed
- *	(or immediately if it was not frozen). %NULL is returned if no match
- *	is found.
- */
-struct super_block *get_super_thawed(struct block_device *bdev)
-{
-	return __get_super_thawed(bdev, false);
-}
-EXPORT_SYMBOL(get_super_thawed);
-
-/**
- *	get_super_exclusive_thawed - get thawed superblock of a device
- *	@bdev: device to get the superblock for
- *
- *	Scans the superblock list and finds the superblock of the file system
- *	mounted on the device. The superblock is returned once it is thawed
- *	(or immediately if it was not frozen) and s_umount semaphore is held
- *	in exclusive mode. %NULL is returned if no match is found.
- */
-struct super_block *get_super_exclusive_thawed(struct block_device *bdev)
-{
-	return __get_super_thawed(bdev, true);
-}
-EXPORT_SYMBOL(get_super_exclusive_thawed);
-
 /**
  * get_active_super - get an active reference to the superblock of a device
  * @bdev: device to get the superblock for
include/linux/fs.h +1 -3
index 8667d0cdc71e76..a61df0dd4f1989 100644
@@ -1409,7 +1409,7 @@ enum {
 
 struct sb_writers {
 	int				frozen;		/* Is sb frozen? */
-	wait_queue_head_t		wait_unfrozen;	/* for get_super_thawed() */
+	wait_queue_head_t		wait_unfrozen;	/* wait for thaw */
 	struct percpu_rw_semaphore	rw_sem[SB_FREEZE_LEVELS];
 };
 
@@ -3132,8 +3132,6 @@ extern struct file_system_type *get_filesystem(struct file_system_type *fs);
 extern void put_filesystem(struct file_system_type *fs);
 extern struct file_system_type *get_fs_type(const char *name);
 extern struct super_block *get_super(struct block_device *);
-extern struct super_block *get_super_thawed(struct block_device *);
-extern struct super_block *get_super_exclusive_thawed(struct block_device *bdev);
 extern struct super_block *get_active_super(struct block_device *bdev);
 extern void drop_super(struct super_block *sb);
 extern void drop_super_exclusive(struct super_block *sb);
Signature · 1 line
2.29.2
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author maintainer
#10 lore raw

PATCH 4/45 fs: simplify freeze_bdev/thaw_bdev

In reply to Christoph Hellwig

Store the frozen superblock in struct block_device to avoid the awkward interface that can return a sb only used a cookie, an ERR_PTR or NULL.

---
 drivers/md/dm-core.h      |  5 -----
 drivers/md/dm.c           | 20 ++++++--------------
 fs/block_dev.c            | 39 ++++++++++++++++-----------------------
 fs/buffer.c               |  2 +-
 fs/ext4/ioctl.c           |  2 +-
 fs/f2fs/file.c            | 14 +++++---------
 fs/xfs/xfs_fsops.c        |  7 ++-----
 include/linux/blk_types.h |  1 +
 include/linux/blkdev.h    |  4 ++--
 9 files changed, 34 insertions(+), 60 deletions(-)
drivers/md/dm-core.h +0 -5
index d522093cb39dda..aace147effcacb 100644
@@ -96,11 +96,6 @@ struct mapped_device {
 	 */
 	struct workqueue_struct *wq;
 
-	/*
-	 * freeze/thaw support require holding onto a super block
-	 */
-	struct super_block *frozen_sb;
-
 	/* forced geometry settings */
 	struct hd_geometry geometry;
 
drivers/md/dm.c +6 -14
index 54739f1b579bc8..50541d336c719b 100644
@@ -2392,27 +2392,19 @@ static int lock_fs(struct mapped_device *md)
 {
 	int r;
 
-	WARN_ON(md->frozen_sb);
+	WARN_ON(test_bit(DMF_FROZEN, &md->flags));
 
-	md->frozen_sb = freeze_bdev(md->bdev);
-	if (IS_ERR(md->frozen_sb)) {
-		r = PTR_ERR(md->frozen_sb);
-		md->frozen_sb = NULL;
-		return r;
-	}
-
-	set_bit(DMF_FROZEN, &md->flags);
-
-	return 0;
+	r = freeze_bdev(md->bdev);
+	if (!r)
+		set_bit(DMF_FROZEN, &md->flags);
+	return r;
 }
 
 static void unlock_fs(struct mapped_device *md)
 {
 	if (!test_bit(DMF_FROZEN, &md->flags))
 		return;
-
-	thaw_bdev(md->bdev, md->frozen_sb);
-	md->frozen_sb = NULL;
+	thaw_bdev(md->bdev);
 	clear_bit(DMF_FROZEN, &md->flags);
 }
 
fs/block_dev.c +16 -23
index d8664f5c1ff669..60492620d51866 100644
@@ -548,55 +548,47 @@ EXPORT_SYMBOL(fsync_bdev);
  * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
  * actually.
  */
-struct super_block *freeze_bdev(struct block_device *bdev)
+int freeze_bdev(struct block_device *bdev)
 {
 	struct super_block *sb;
 	int error = 0;
 
 	mutex_lock(&bdev->bd_fsfreeze_mutex);
-	if (++bdev->bd_fsfreeze_count > 1) {
-		/*
-		 * We don't even need to grab a reference - the first call
-		 * to freeze_bdev grab an active reference and only the last
-		 * thaw_bdev drops it.
-		 */
-		sb = get_super(bdev);
-		if (sb)
-			drop_super(sb);
-		mutex_unlock(&bdev->bd_fsfreeze_mutex);
-		return sb;
-	}
+	if (++bdev->bd_fsfreeze_count > 1)
+		goto done;
 
 	sb = get_active_super(bdev);
 	if (!sb)
-		goto out;
+		goto sync;
 	if (sb->s_op->freeze_super)
 		error = sb->s_op->freeze_super(sb);
 	else
 		error = freeze_super(sb);
+	deactivate_super(sb);
+
 	if (error) {
-		deactivate_super(sb);
 		bdev->bd_fsfreeze_count--;
-		mutex_unlock(&bdev->bd_fsfreeze_mutex);
-		return ERR_PTR(error);
+		goto done;
 	}
-	deactivate_super(sb);
- out:
+	bdev->bd_fsfreeze_sb = sb;
+
+sync:
 	sync_blockdev(bdev);
+done:
 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
-	return sb;	/* thaw_bdev releases s->s_umount */
+	return error;	/* thaw_bdev releases s->s_umount */
 }
 EXPORT_SYMBOL(freeze_bdev);
 
 /**
  * thaw_bdev  -- unlock filesystem
  * @bdev:	blockdevice to unlock
- * @sb:		associated superblock
  *
  * Unlocks the filesystem and marks it writeable again after freeze_bdev().
  */
-int thaw_bdev(struct block_device *bdev, struct super_block *sb)
+int thaw_bdev(struct block_device *bdev)
 {
+	struct super_block *sb;
 	int error = -EINVAL;
 
 	mutex_lock(&bdev->bd_fsfreeze_mutex);
@@ -607,6 +599,7 @@ int thaw_bdev(struct block_device *bdev, struct super_block *sb)
 	if (--bdev->bd_fsfreeze_count > 0)
 		goto out;
 
+	sb = bdev->bd_fsfreeze_sb;
 	if (!sb)
 		goto out;
 
@@ -618,7 +611,7 @@ int thaw_bdev(struct block_device *bdev, struct super_block *sb)
 		bdev->bd_fsfreeze_count++;
 out:
 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
-	return error;
+	return 0;
 }
 EXPORT_SYMBOL(thaw_bdev);
 
fs/buffer.c +1 -1
index 23f645657488ba..a7595ada9400ff 100644
@@ -523,7 +523,7 @@ static int osync_buffers_list(spinlock_t *lock, struct list_head *list)
 
 void emergency_thaw_bdev(struct super_block *sb)
 {
-	while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb))
+	while (sb->s_bdev && !thaw_bdev(sb->s_bdev))
 		printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev);
 }
 
fs/ext4/ioctl.c +1 -1
index f0381876a7e5b0..524e134324475e 100644
@@ -624,7 +624,7 @@ static int ext4_shutdown(struct super_block *sb, unsigned long arg)
 	case EXT4_GOING_FLAGS_DEFAULT:
 		freeze_bdev(sb->s_bdev);
 		set_bit(EXT4_FLAGS_SHUTDOWN, &sbi->s_ext4_flags);
-		thaw_bdev(sb->s_bdev, sb);
+		thaw_bdev(sb->s_bdev);
 		break;
 	case EXT4_GOING_FLAGS_LOGFLUSH:
 		set_bit(EXT4_FLAGS_SHUTDOWN, &sbi->s_ext4_flags);
fs/f2fs/file.c +5 -9
index ee861c6d9ff026..a9fc482a0e60a5 100644
@@ -2230,16 +2230,12 @@ static int f2fs_ioc_shutdown(struct file *filp, unsigned long arg)
 
 	switch (in) {
 	case F2FS_GOING_DOWN_FULLSYNC:
-		sb = freeze_bdev(sb->s_bdev);
-		if (IS_ERR(sb)) {
-			ret = PTR_ERR(sb);
+		ret = freeze_bdev(sb->s_bdev);
+		if (ret)
 			goto out;
-		}
-		if (sb) {
-			f2fs_stop_checkpoint(sbi, false);
-			set_sbi_flag(sbi, SBI_IS_SHUTDOWN);
-			thaw_bdev(sb->s_bdev, sb);
-		}
+		f2fs_stop_checkpoint(sbi, false);
+		set_sbi_flag(sbi, SBI_IS_SHUTDOWN);
+		thaw_bdev(sb->s_bdev);
 		break;
 	case F2FS_GOING_DOWN_METASYNC:
 		/* do checkpoint only */
fs/xfs/xfs_fsops.c +2 -5
index ef1d5bb88b93ab..b7c5783a031c69 100644
@@ -433,13 +433,10 @@ xfs_fs_goingdown(
 {
 	switch (inflags) {
 	case XFS_FSOP_GOING_FLAGS_DEFAULT: {
-		struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
-
-		if (sb && !IS_ERR(sb)) {
+		if (!freeze_bdev(mp->m_super->s_bdev)) {
 			xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
-			thaw_bdev(sb->s_bdev, sb);
+			thaw_bdev(mp->m_super->s_bdev);
 		}
-
 		break;
 	}
 	case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
include/linux/blk_types.h +1 -0
index d9b69bbde5cc54..ebfb4e7c1fd125 100644
@@ -46,6 +46,7 @@ struct block_device {
 	int			bd_fsfreeze_count;
 	/* Mutex for freeze */
 	struct mutex		bd_fsfreeze_mutex;
+	struct super_block	*bd_fsfreeze_sb;
 } __randomize_layout;
 
 /*
include/linux/blkdev.h +2 -2
index 05b346a68c2eee..12810a19edebc4 100644
@@ -2020,7 +2020,7 @@ static inline int sync_blockdev(struct block_device *bdev)
 #endif
 int fsync_bdev(struct block_device *bdev);
 
-struct super_block *freeze_bdev(struct block_device *bdev);
-int thaw_bdev(struct block_device *bdev, struct super_block *sb);
+int freeze_bdev(struct block_device *bdev);
+int thaw_bdev(struct block_device *bdev);
 
 #endif /* _LINUX_BLKDEV_H */
Signature · 1 line
2.29.2
Acked-by on v1 4/45
#11 lore raw

Re: [PATCH 04/45] fs: simplify freeze_bdev/thaw_bdev

Quoted 11 lines · Christoph Hellwig

Store the frozen superblock in struct block_device to avoid the awkward interface that can return a sb only used a cookie, an ERR_PTR or NULL.

---
  drivers/md/dm-core.h      |  5 -----
  drivers/md/dm.c           | 20 ++++++--------------
  fs/block_dev.c            | 39 ++++++++++++++++-----------------------
  fs/buffer.c               |  2 +-
  fs/ext4/ioctl.c           |  2 +-
  fs/f2fs/file.c            | 14 +++++---------

For f2fs part,

Thanks,

Quoted 4 lines
  fs/xfs/xfs_fsops.c        |  7 ++-----
  include/linux/blk_types.h |  1 +
  include/linux/blkdev.h    |  4 ++--
  9 files changed, 34 insertions(+), 60 deletions(-)
maintainer
#12 lore raw

Re: [PATCH 04/45] fs: simplify freeze_bdev/thaw_bdev

In reply to Christoph Hellwig
Quoted 4 lines · Christoph Hellwig

Store the frozen superblock in struct block_device to avoid the awkward interface that can return a sb only used a cookie, an ERR_PTR or NULL.

Some comments below...

Quoted 47 lines
fs/block_dev.c +7 -17
index d8664f5c1ff669..60492620d51866 100644
@@ -548,55 +548,47 @@ EXPORT_SYMBOL(fsync_bdev);
  * count down in thaw_bdev(). When it becomes 0, thaw_bdev() will unfreeze
  * actually.
  */
-struct super_block *freeze_bdev(struct block_device *bdev)
+int freeze_bdev(struct block_device *bdev)
 {
 	struct super_block *sb;
 	int error = 0;
 
 	mutex_lock(&bdev->bd_fsfreeze_mutex);
-	if (++bdev->bd_fsfreeze_count > 1) {
-		/*
-		 * We don't even need to grab a reference - the first call
-		 * to freeze_bdev grab an active reference and only the last
-		 * thaw_bdev drops it.
-		 */
-		sb = get_super(bdev);
-		if (sb)
-			drop_super(sb);
-		mutex_unlock(&bdev->bd_fsfreeze_mutex);
-		return sb;
-	}
+	if (++bdev->bd_fsfreeze_count > 1)
+		goto done;
 
 	sb = get_active_super(bdev);
 	if (!sb)
-		goto out;
+		goto sync;
 	if (sb->s_op->freeze_super)
 		error = sb->s_op->freeze_super(sb);
 	else
 		error = freeze_super(sb);
+	deactivate_super(sb);
+
 	if (error) {
-		deactivate_super(sb);
 		bdev->bd_fsfreeze_count--;
-		mutex_unlock(&bdev->bd_fsfreeze_mutex);
-		return ERR_PTR(error);
+		goto done;
 	}

fs/block_dev.c EXPORT_SYMBOL(fsync_bdev); · line 579

Quoted 10 lines
fs/block_dev.c +5 -3
-	deactivate_super(sb);
- out:
+	bdev->bd_fsfreeze_sb = sb;
+
+sync:
 	sync_blockdev(bdev);
+done:
 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
-	return sb;	/* thaw_bdev releases s->s_umount */
+	return error;	/* thaw_bdev releases s->s_umount */

The comment about thaw_bdev() seems to be stale? At least I don't see what it's speaking about...

Quoted 26 lines
fs/block_dev.c +3 -2
 }
 EXPORT_SYMBOL(freeze_bdev);
 
 /**
  * thaw_bdev  -- unlock filesystem
  * @bdev:	blockdevice to unlock
- * @sb:		associated superblock
  *
  * Unlocks the filesystem and marks it writeable again after freeze_bdev().
  */
-int thaw_bdev(struct block_device *bdev, struct super_block *sb)
+int thaw_bdev(struct block_device *bdev)
 {
+	struct super_block *sb;
 	int error = -EINVAL;
 
 	mutex_lock(&bdev->bd_fsfreeze_mutex);
@@ -607,6 +599,7 @@ int thaw_bdev(struct block_device *bdev, struct super_block *sb)
 	if (--bdev->bd_fsfreeze_count > 0)
 		goto out;
 
+	sb = bdev->bd_fsfreeze_sb;
 	if (!sb)
 		goto out;
 

fs/block_dev.c int thaw_bdev(struct block_device *bdev, struct super_block *sb) · line 614

Quoted 5 lines
fs/block_dev.c +1 -1
@@ -618,7 +611,7 @@ int thaw_bdev(struct block_device *bdev, struct super_block *sb)
 		bdev->bd_fsfreeze_count++;
 out:
 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
-	return error;
+	return 0;

But we now won't return -EINVAL if this gets called e.g. with bd_fsfreeze_count == 0, right?

Honza

Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author maintainer
#13 lore raw

Re: [PATCH 04/45] fs: simplify freeze_bdev/thaw_bdev

Quoted 6 lines · Jan Kara

fs/block_dev.c EXPORT_SYMBOL(fsync_bdev); · line 579

Quoted 3 lines
fs/block_dev.c +1 -1
 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
-	return sb;	/* thaw_bdev releases s->s_umount */
+	return error;	/* thaw_bdev releases s->s_umount */

The comment about thaw_bdev() seems to be stale? At least I don't see what it's speaking about...

Yes, this comment seems long stale. I think in the very early days we held s_umount on frozen file system, which caused all kinds of problems.

Quoted 6 lines

fs/block_dev.c int thaw_bdev(struct block_device *bdev, struct super_block *sb) · line 614

Quoted 3 lines
fs/block_dev.c +1 -1
 	mutex_unlock(&bdev->bd_fsfreeze_mutex);
-	return error;
+	return 0;

But we now won't return -EINVAL if this gets called e.g. with bd_fsfreeze_count == 0, right?

Yes. I had tried to drop the return value as all the freeze_bdev calls ignored it. But I had missed the unpaired emergency thaw and put it back and messed this up..

author maintainer
#14 lore raw

PATCH 5/45 mtip32xx: remove the call to fsync_bdev on removal

In reply to Christoph Hellwig

del_gendisk already calls fsync_bdev for every partition, no need to do this twice.

---
 drivers/block/mtip32xx/mtip32xx.c | 15 ---------------
 drivers/block/mtip32xx/mtip32xx.h |  2 --
 2 files changed, 17 deletions(-)
drivers/block/mtip32xx/mtip32xx.c +0 -15
index 153e2cdecb4d40..53ac59d19ae530 100644
@@ -3687,7 +3687,6 @@ static int mtip_block_initialize(struct driver_data *dd)
 	/* Enable the block device and add it to /dev */
 	device_add_disk(&dd->pdev->dev, dd->disk, NULL);
 
-	dd->bdev = bdget_disk(dd->disk, 0);
 	/*
 	 * Now that the disk is active, initialize any sysfs attributes
 	 * managed by the protocol layer.
@@ -3721,9 +3720,6 @@ static int mtip_block_initialize(struct driver_data *dd)
 	return rv;
 
 kthread_run_error:
-	bdput(dd->bdev);
-	dd->bdev = NULL;
-
 	/* Delete our gendisk. This also removes the device from /dev */
 	del_gendisk(dd->disk);
 
@@ -3804,14 +3800,6 @@ static int mtip_block_remove(struct driver_data *dd)
 	blk_mq_tagset_busy_iter(&dd->tags, mtip_no_dev_cleanup, dd);
 	blk_mq_unquiesce_queue(dd->queue);
 
-	/*
-	 * Delete our gendisk structure. This also removes the device
-	 * from /dev
-	 */
-	if (dd->bdev) {
-		bdput(dd->bdev);
-		dd->bdev = NULL;
-	}
 	if (dd->disk) {
 		if (test_bit(MTIP_DDF_INIT_DONE_BIT, &dd->dd_flag))
 			del_gendisk(dd->disk);
@@ -4206,9 +4194,6 @@ static void mtip_pci_remove(struct pci_dev *pdev)
 	} while (atomic_read(&dd->irq_workers_active) != 0 &&
 		time_before(jiffies, to));
 
-	if (!dd->sr)
-		fsync_bdev(dd->bdev);
-
 	if (atomic_read(&dd->irq_workers_active) != 0) {
 		dev_warn(&dd->pdev->dev,
 			"Completion workers still active!\n");
drivers/block/mtip32xx/mtip32xx.h +0 -2
index e22a7f0523bf30..88f4206310e4c8 100644
@@ -463,8 +463,6 @@ struct driver_data {
 
 	int isr_binding;
 
-	struct block_device *bdev;
-
 	struct list_head online_list; /* linkage for online list */
 
 	struct list_head remove_list; /* linkage for removing list */
Signature · 1 line
2.29.2
maintainer Reviewed-by on v1 5/45
#15 lore raw

Re: [PATCH 05/45] mtip32xx: remove the call to fsync_bdev on removal

Quoted 4 lines · Christoph Hellwig

del_gendisk already calls fsync_bdev for every partition, no need to do this twice.

Makes sense to me. You can add:

Honza

Quoted 68 lines
---
 drivers/block/mtip32xx/mtip32xx.c | 15 ---------------
 drivers/block/mtip32xx/mtip32xx.h |  2 --
 2 files changed, 17 deletions(-)
drivers/block/mtip32xx/mtip32xx.c +0 -15
index 153e2cdecb4d40..53ac59d19ae530 100644
@@ -3687,7 +3687,6 @@ static int mtip_block_initialize(struct driver_data *dd)
 	/* Enable the block device and add it to /dev */
 	device_add_disk(&dd->pdev->dev, dd->disk, NULL);
 
-	dd->bdev = bdget_disk(dd->disk, 0);
 	/*
 	 * Now that the disk is active, initialize any sysfs attributes
 	 * managed by the protocol layer.
@@ -3721,9 +3720,6 @@ static int mtip_block_initialize(struct driver_data *dd)
 	return rv;
 
 kthread_run_error:
-	bdput(dd->bdev);
-	dd->bdev = NULL;
-
 	/* Delete our gendisk. This also removes the device from /dev */
 	del_gendisk(dd->disk);
 
@@ -3804,14 +3800,6 @@ static int mtip_block_remove(struct driver_data *dd)
 	blk_mq_tagset_busy_iter(&dd->tags, mtip_no_dev_cleanup, dd);
 	blk_mq_unquiesce_queue(dd->queue);
 
-	/*
-	 * Delete our gendisk structure. This also removes the device
-	 * from /dev
-	 */
-	if (dd->bdev) {
-		bdput(dd->bdev);
-		dd->bdev = NULL;
-	}
 	if (dd->disk) {
 		if (test_bit(MTIP_DDF_INIT_DONE_BIT, &dd->dd_flag))
 			del_gendisk(dd->disk);
@@ -4206,9 +4194,6 @@ static void mtip_pci_remove(struct pci_dev *pdev)
 	} while (atomic_read(&dd->irq_workers_active) != 0 &&
 		time_before(jiffies, to));
 
-	if (!dd->sr)
-		fsync_bdev(dd->bdev);
-
 	if (atomic_read(&dd->irq_workers_active) != 0) {
 		dev_warn(&dd->pdev->dev,
 			"Completion workers still active!\n");
drivers/block/mtip32xx/mtip32xx.h +0 -2
index e22a7f0523bf30..88f4206310e4c8 100644
@@ -463,8 +463,6 @@ struct driver_data {
 
 	int isr_binding;
 
-	struct block_device *bdev;
-
 	struct list_head online_list; /* linkage for online list */
 
 	struct list_head remove_list; /* linkage for removing list */
Signature · 1 line
2.29.2
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author maintainer
#16 lore raw

PATCH 6/45 zram: remove the claim mechanism

In reply to Christoph Hellwig

The zram claim mechanism was added to ensure no new opens come in during teardown. But the proper way to archive that is to call del_gendisk first, which takes care of all that. Once del_gendisk is called in the right place, the reset side can also be simplified as no I/O can be outstanding on a block device that is not open.

---
 drivers/block/zram/zram_drv.c | 72 ++++++++---------------------------
 1 file changed, 15 insertions(+), 57 deletions(-)
drivers/block/zram/zram_drv.c +15 -57
index 6d15d51cee2b7e..2e6d75ec1afddb 100644
@@ -1756,64 +1756,33 @@ static ssize_t disksize_store(struct device *dev,
 static ssize_t reset_store(struct device *dev,
 		struct device_attribute *attr, const char *buf, size_t len)
 {
-	int ret;
-	unsigned short do_reset;
-	struct zram *zram;
+	struct zram *zram = dev_to_zram(dev);
 	struct block_device *bdev;
+	unsigned short do_reset;
+	int ret = 0;
 
 	ret = kstrtou16(buf, 10, &do_reset);
 	if (ret)
 		return ret;
-
 	if (!do_reset)
 		return -EINVAL;
 
-	zram = dev_to_zram(dev);
 	bdev = bdget_disk(zram->disk, 0);
 	if (!bdev)
 		return -ENOMEM;
 
 	mutex_lock(&bdev->bd_mutex);
-	/* Do not reset an active device or claimed device */
-	if (bdev->bd_openers || zram->claim) {
-		mutex_unlock(&bdev->bd_mutex);
-		bdput(bdev);
-		return -EBUSY;
-	}
-
-	/* From now on, anyone can't open /dev/zram[0-9] */
-	zram->claim = true;
+	if (bdev->bd_openers)
+		ret = -EBUSY;
+	else
+		zram_reset_device(zram);
 	mutex_unlock(&bdev->bd_mutex);
-
-	/* Make sure all the pending I/O are finished */
-	fsync_bdev(bdev);
-	zram_reset_device(zram);
 	bdput(bdev);
 
-	mutex_lock(&bdev->bd_mutex);
-	zram->claim = false;
-	mutex_unlock(&bdev->bd_mutex);
-
-	return len;
-}
-
-static int zram_open(struct block_device *bdev, fmode_t mode)
-{
-	int ret = 0;
-	struct zram *zram;
-
-	WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
-
-	zram = bdev->bd_disk->private_data;
-	/* zram was claimed to reset so open request fails */
-	if (zram->claim)
-		ret = -EBUSY;
-
-	return ret;
+	return ret ? ret : len;
 }
 
 static const struct block_device_operations zram_devops = {
-	.open = zram_open,
 	.submit_bio = zram_submit_bio,
 	.swap_slot_free_notify = zram_slot_free_notify,
 	.rw_page = zram_rw_page,
@@ -1821,7 +1790,6 @@ static const struct block_device_operations zram_devops = {
 };
 
 static const struct block_device_operations zram_wb_devops = {
-	.open = zram_open,
 	.submit_bio = zram_submit_bio,
 	.swap_slot_free_notify = zram_slot_free_notify,
 	.owner = THIS_MODULE
@@ -1974,32 +1942,22 @@ static int zram_add(void)
 
 static int zram_remove(struct zram *zram)
 {
-	struct block_device *bdev;
-
-	bdev = bdget_disk(zram->disk, 0);
-	if (!bdev)
-		return -ENOMEM;
+	struct block_device *bdev = bdget_disk(zram->disk, 0);
 
-	mutex_lock(&bdev->bd_mutex);
-	if (bdev->bd_openers || zram->claim) {
-		mutex_unlock(&bdev->bd_mutex);
+	if (bdev) {
+		if (bdev->bd_openers) {
+			bdput(bdev);
+			return -EBUSY;
+		}
 		bdput(bdev);
-		return -EBUSY;
 	}
 
-	zram->claim = true;
-	mutex_unlock(&bdev->bd_mutex);
-
+	del_gendisk(zram->disk);
 	zram_debugfs_unregister(zram);
-
-	/* Make sure all the pending I/O are finished */
-	fsync_bdev(bdev);
 	zram_reset_device(zram);
-	bdput(bdev);
 
 	pr_info("Removed device: %s\n", zram->disk->disk_name);
 
-	del_gendisk(zram->disk);
 	blk_cleanup_queue(zram->disk->queue);
 	put_disk(zram->disk);
 	kfree(zram);
Signature · 1 line
2.29.2
maintainer
#17 lore raw

Re: [PATCH 06/45] zram: remove the claim mechanism

Quoted 2 lines · Christoph Hellwig

The zram claim mechanism was added to ensure no new opens come in during teardown. But the proper way to archive that is to call

					  ^^^ achieve
Quoted 5 lines

del_gendisk first, which takes care of all that. Once del_gendisk is called in the right place, the reset side can also be simplified as no I/O can be outstanding on a block device that is not open.

Otherwise I didn't find anything obviously wrong with the patch but I don't feel confident enough with zram to really give you my reviewed-by on this one.

Honza

Quoted 132 lines
---
 drivers/block/zram/zram_drv.c | 72 ++++++++---------------------------
 1 file changed, 15 insertions(+), 57 deletions(-)
drivers/block/zram/zram_drv.c +15 -57
index 6d15d51cee2b7e..2e6d75ec1afddb 100644
@@ -1756,64 +1756,33 @@ static ssize_t disksize_store(struct device *dev,
 static ssize_t reset_store(struct device *dev,
 		struct device_attribute *attr, const char *buf, size_t len)
 {
-	int ret;
-	unsigned short do_reset;
-	struct zram *zram;
+	struct zram *zram = dev_to_zram(dev);
 	struct block_device *bdev;
+	unsigned short do_reset;
+	int ret = 0;
 
 	ret = kstrtou16(buf, 10, &do_reset);
 	if (ret)
 		return ret;
-
 	if (!do_reset)
 		return -EINVAL;
 
-	zram = dev_to_zram(dev);
 	bdev = bdget_disk(zram->disk, 0);
 	if (!bdev)
 		return -ENOMEM;
 
 	mutex_lock(&bdev->bd_mutex);
-	/* Do not reset an active device or claimed device */
-	if (bdev->bd_openers || zram->claim) {
-		mutex_unlock(&bdev->bd_mutex);
-		bdput(bdev);
-		return -EBUSY;
-	}
-
-	/* From now on, anyone can't open /dev/zram[0-9] */
-	zram->claim = true;
+	if (bdev->bd_openers)
+		ret = -EBUSY;
+	else
+		zram_reset_device(zram);
 	mutex_unlock(&bdev->bd_mutex);
-
-	/* Make sure all the pending I/O are finished */
-	fsync_bdev(bdev);
-	zram_reset_device(zram);
 	bdput(bdev);
 
-	mutex_lock(&bdev->bd_mutex);
-	zram->claim = false;
-	mutex_unlock(&bdev->bd_mutex);
-
-	return len;
-}
-
-static int zram_open(struct block_device *bdev, fmode_t mode)
-{
-	int ret = 0;
-	struct zram *zram;
-
-	WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
-
-	zram = bdev->bd_disk->private_data;
-	/* zram was claimed to reset so open request fails */
-	if (zram->claim)
-		ret = -EBUSY;
-
-	return ret;
+	return ret ? ret : len;
 }
 
 static const struct block_device_operations zram_devops = {
-	.open = zram_open,
 	.submit_bio = zram_submit_bio,
 	.swap_slot_free_notify = zram_slot_free_notify,
 	.rw_page = zram_rw_page,
@@ -1821,7 +1790,6 @@ static const struct block_device_operations zram_devops = {
 };
 
 static const struct block_device_operations zram_wb_devops = {
-	.open = zram_open,
 	.submit_bio = zram_submit_bio,
 	.swap_slot_free_notify = zram_slot_free_notify,
 	.owner = THIS_MODULE
@@ -1974,32 +1942,22 @@ static int zram_add(void)
 
 static int zram_remove(struct zram *zram)
 {
-	struct block_device *bdev;
-
-	bdev = bdget_disk(zram->disk, 0);
-	if (!bdev)
-		return -ENOMEM;
+	struct block_device *bdev = bdget_disk(zram->disk, 0);
 
-	mutex_lock(&bdev->bd_mutex);
-	if (bdev->bd_openers || zram->claim) {
-		mutex_unlock(&bdev->bd_mutex);
+	if (bdev) {
+		if (bdev->bd_openers) {
+			bdput(bdev);
+			return -EBUSY;
+		}
 		bdput(bdev);
-		return -EBUSY;
 	}
 
-	zram->claim = true;
-	mutex_unlock(&bdev->bd_mutex);
-
+	del_gendisk(zram->disk);
 	zram_debugfs_unregister(zram);
-
-	/* Make sure all the pending I/O are finished */
-	fsync_bdev(bdev);
 	zram_reset_device(zram);
-	bdput(bdev);
 
 	pr_info("Removed device: %s\n", zram->disk->disk_name);
 
-	del_gendisk(zram->disk);
 	blk_cleanup_queue(zram->disk->queue);
 	put_disk(zram->disk);
 	kfree(zram);
Signature · 1 line
2.29.2
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author maintainer
#18 lore raw

PATCH 7/45 zram: do not call set_blocksize

In reply to Christoph Hellwig

set_blocksize is used by file systems to use their preferred buffer cache block size. Block drivers should not set it.

---
 drivers/block/zram/zram_drv.c | 11 +----------
 drivers/block/zram/zram_drv.h |  1 -
 2 files changed, 1 insertion(+), 11 deletions(-)
drivers/block/zram/zram_drv.c +1 -10
index 2e6d75ec1afddb..88baa6158eaee1 100644
@@ -403,13 +403,10 @@ static void reset_bdev(struct zram *zram)
 		return;
 
 	bdev = zram->bdev;
-	if (zram->old_block_size)
-		set_blocksize(bdev, zram->old_block_size);
 	blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 	/* hope filp_close flush all of IO */
 	filp_close(zram->backing_dev, NULL);
 	zram->backing_dev = NULL;
-	zram->old_block_size = 0;
 	zram->bdev = NULL;
 	zram->disk->fops = &zram_devops;
 	kvfree(zram->bitmap);
@@ -454,7 +451,7 @@ static ssize_t backing_dev_store(struct device *dev,
 	struct file *backing_dev = NULL;
 	struct inode *inode;
 	struct address_space *mapping;
-	unsigned int bitmap_sz, old_block_size = 0;
+	unsigned int bitmap_sz;
 	unsigned long nr_pages, *bitmap = NULL;
 	struct block_device *bdev = NULL;
 	int err;
@@ -509,14 +506,8 @@ static ssize_t backing_dev_store(struct device *dev,
 		goto out;
 	}
 
-	old_block_size = block_size(bdev);
-	err = set_blocksize(bdev, PAGE_SIZE);
-	if (err)
-		goto out;
-
 	reset_bdev(zram);
 
-	zram->old_block_size = old_block_size;
 	zram->bdev = bdev;
 	zram->backing_dev = backing_dev;
 	zram->bitmap = bitmap;
drivers/block/zram/zram_drv.h +0 -1
index f2fd46daa76045..712354a4207c77 100644
@@ -118,7 +118,6 @@ struct zram {
 	bool wb_limit_enable;
 	u64 bd_wb_limit;
 	struct block_device *bdev;
-	unsigned int old_block_size;
 	unsigned long *bitmap;
 	unsigned long nr_pages;
 #endif
Signature · 1 line
2.29.2
author maintainer
#19 lore raw

PATCH 8/45 loop: do not call set_blocksize

In reply to Christoph Hellwig

set_blocksize is used by file systems to use their preferred buffer cache block size. Block drivers should not set it.

---
 drivers/block/loop.c | 3 ---
 1 file changed, 3 deletions(-)
drivers/block/loop.c +0 -3
index 9a27d4f1c08aac..b42c728620c9e4 100644
@@ -1164,9 +1164,6 @@ static int loop_configure(struct loop_device *lo, fmode_t mode,
 	size = get_loop_size(lo, file);
 	loop_set_size(lo, size);
 
-	set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
-		      block_size(inode->i_bdev) : PAGE_SIZE);
-
 	lo->lo_state = Lo_bound;
 	if (part_shift)
 		lo->lo_flags |= LO_FLAGS_PARTSCAN;
Signature · 1 line
2.29.2
maintainer Reviewed-by on v1 8/45
#20 lore raw

Re: [PATCH 08/45] loop: do not call set_blocksize

Quoted 4 lines · Christoph Hellwig

set_blocksize is used by file systems to use their preferred buffer cache block size. Block drivers should not set it.

Looks good to me. You can add:

Honza

Quoted 21 lines
---
 drivers/block/loop.c | 3 ---
 1 file changed, 3 deletions(-)
drivers/block/loop.c +0 -3
index 9a27d4f1c08aac..b42c728620c9e4 100644
@@ -1164,9 +1164,6 @@ static int loop_configure(struct loop_device *lo, fmode_t mode,
 	size = get_loop_size(lo, file);
 	loop_set_size(lo, size);
 
-	set_blocksize(bdev, S_ISBLK(inode->i_mode) ?
-		      block_size(inode->i_bdev) : PAGE_SIZE);
-
 	lo->lo_state = Lo_bound;
 	if (part_shift)
 		lo->lo_flags |= LO_FLAGS_PARTSCAN;
Signature · 1 line
2.29.2
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author maintainer
#21 lore raw

PATCH 9/45 dm: simplify flush_bio initialization in __send_empty_flush

In reply to Christoph Hellwig

We don't really need the struct block_device to initialize a bio. So switch from using bio_set_dev to manually setting up bi_disk (bi_partno will always be zero and has been cleared by bio_init already).

---
 drivers/md/dm.c | 12 +++---------
 1 file changed, 3 insertions(+), 9 deletions(-)
drivers/md/dm.c +3 -9
index 50541d336c719b..ab0a8335f098d9 100644
@@ -1422,18 +1422,12 @@ static int __send_empty_flush(struct clone_info *ci)
 	 */
 	bio_init(&flush_bio, NULL, 0);
 	flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
+	flush_bio.bi_disk = ci->io->md->disk;
+	bio_associate_blkg(&flush_bio);
+
 	ci->bio = &flush_bio;
 	ci->sector_count = 0;
 
-	/*
-	 * Empty flush uses a statically initialized bio, as the base for
-	 * cloning.  However, blkg association requires that a bdev is
-	 * associated with a gendisk, which doesn't happen until the bdev is
-	 * opened.  So, blkg association is done at issue time of the flush
-	 * rather than when the device is created in alloc_dev().
-	 */
-	bio_set_dev(ci->bio, ci->io->md->bdev);
-
 	BUG_ON(bio_has_data(ci->bio));
 	while ((ti = dm_table_get_target(ci->map, target_nr++)))
 		__send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
Signature · 1 line
2.29.2
author maintainer
#22 lore raw

PATCH 10/45 dm: remove the block_device reference in struct mapped_device

In reply to Christoph Hellwig

Get rid of the long-lasting struct block_device reference in struct mapped_device. The only remaining user is the freeze code, where we can trivially look up the block device at freeze time and release the reference at thaw time.

---
 drivers/md/dm-core.h |  2 --
 drivers/md/dm.c      | 25 ++++++++++++++-----------
 2 files changed, 14 insertions(+), 13 deletions(-)
drivers/md/dm-core.h +0 -2
index aace147effcacb..086d293c2b036c 100644
@@ -102,8 +102,6 @@ struct mapped_device {
 	/* kobject and completion */
 	struct dm_kobject_holder kobj_holder;
 
-	struct block_device *bdev;
-
 	struct dm_stats stats;
 
 	/* for blk-mq request-based DM support */
drivers/md/dm.c +14 -11
index ab0a8335f098d9..48051db006f30c 100644
@@ -1744,11 +1744,6 @@ static void cleanup_mapped_device(struct mapped_device *md)
 
 	cleanup_srcu_struct(&md->io_barrier);
 
-	if (md->bdev) {
-		bdput(md->bdev);
-		md->bdev = NULL;
-	}
-
 	mutex_destroy(&md->suspend_lock);
 	mutex_destroy(&md->type_lock);
 	mutex_destroy(&md->table_devices_lock);
@@ -1840,10 +1835,6 @@ static struct mapped_device *alloc_dev(int minor)
 	if (!md->wq)
 		goto bad;
 
-	md->bdev = bdget_disk(md->disk, 0);
-	if (!md->bdev)
-		goto bad;
-
 	dm_stats_init(&md->stats);
 
 	/* Populate the mapping, nobody knows we exist yet */
@@ -2384,11 +2375,16 @@ struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
  */
 static int lock_fs(struct mapped_device *md)
 {
+	struct block_device *bdev;
 	int r;
 
 	WARN_ON(test_bit(DMF_FROZEN, &md->flags));
 
-	r = freeze_bdev(md->bdev);
+	bdev = bdget_disk(md->disk, 0);
+	if (!bdev)
+		return -ENOMEM;
+	r = freeze_bdev(bdev);
+	bdput(bdev);
 	if (!r)
 		set_bit(DMF_FROZEN, &md->flags);
 	return r;
@@ -2396,9 +2392,16 @@ static int lock_fs(struct mapped_device *md)
 
 static void unlock_fs(struct mapped_device *md)
 {
+	struct block_device *bdev;
+
 	if (!test_bit(DMF_FROZEN, &md->flags))
 		return;
-	thaw_bdev(md->bdev);
+
+	bdev = bdget_disk(md->disk, 0);
+	if (!bdev)
+		return;
+	thaw_bdev(bdev);
+	bdput(bdev);
 	clear_bit(DMF_FROZEN, &md->flags);
 }
 
Signature · 1 line
2.29.2
author maintainer
#23 lore raw

PATCH 11/45 block: remove a duplicate __disk_get_part prototype

In reply to Christoph Hellwig
---
 include/linux/genhd.h | 1 -
 1 file changed, 1 deletion(-)
include/linux/genhd.h +0 -1
index 46553d6d602563..22f5b9fd96f8bf 100644
@@ -250,7 +250,6 @@ static inline dev_t part_devt(struct hd_struct *part)
 	return part_to_dev(part)->devt;
 }
 
-extern struct hd_struct *__disk_get_part(struct gendisk *disk, int partno);
 extern struct hd_struct *disk_get_part(struct gendisk *disk, int partno);
 
 static inline void disk_put_part(struct hd_struct *part)
Signature · 1 line
2.29.2
author maintainer
#25 lore raw

PATCH 12/45 block: remove a superflous check in blkpg_do_ioctl

In reply to Christoph Hellwig

sector_t is now always a u64, so this check is not needed.

---
 block/ioctl.c | 9 ---------
 1 file changed, 9 deletions(-)
block/ioctl.c +0 -9
index 6b785181344fe1..0c09bb7a6ff35f 100644
@@ -35,15 +35,6 @@ static int blkpg_do_ioctl(struct block_device *bdev,
 	start = p.start >> SECTOR_SHIFT;
 	length = p.length >> SECTOR_SHIFT;
 
-	/* check for fit in a hd_struct */
-	if (sizeof(sector_t) < sizeof(long long)) {
-		long pstart = start, plength = length;
-
-		if (pstart != start || plength != length || pstart < 0 ||
-		    plength < 0 || p.pno > 65535)
-			return -EINVAL;
-	}
-
 	switch (op) {
 	case BLKPG_ADD_PARTITION:
 		/* check if partition is aligned to blocksize */
Signature · 1 line
2.29.2
maintainer Reviewed-by on v1 12/45
#27 lore raw

Re: [PATCH 12/45] block: remove a superflous check in blkpg_do_ioctl

In reply to Christoph Hellwig
Quoted 3 lines · Christoph Hellwig

sector_t is now always a u64, so this check is not needed.

Looks good. You can add:

Honza

Quoted 27 lines
---
 block/ioctl.c | 9 ---------
 1 file changed, 9 deletions(-)
block/ioctl.c +0 -9
index 6b785181344fe1..0c09bb7a6ff35f 100644
@@ -35,15 +35,6 @@ static int blkpg_do_ioctl(struct block_device *bdev,
 	start = p.start >> SECTOR_SHIFT;
 	length = p.length >> SECTOR_SHIFT;
 
-	/* check for fit in a hd_struct */
-	if (sizeof(sector_t) < sizeof(long long)) {
-		long pstart = start, plength = length;
-
-		if (pstart != start || plength != length || pstart < 0 ||
-		    plength < 0 || p.pno > 65535)
-			return -EINVAL;
-	}
-
 	switch (op) {
 	case BLKPG_ADD_PARTITION:
 		/* check if partition is aligned to blocksize */
Signature · 1 line
2.29.2
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
author maintainer
#28 lore raw

PATCH 13/45 block: add a bdev_kobj helper

In reply to Christoph Hellwig

Add a little helper to find the kobject for a struct block_device.

---
 drivers/md/bcache/super.c |  7 ++-----
 drivers/md/md.c           |  4 +---
 fs/block_dev.c            |  6 +++---
 fs/btrfs/sysfs.c          | 15 +++------------
 include/linux/blk_types.h |  3 +++
 5 files changed, 12 insertions(+), 23 deletions(-)
drivers/md/bcache/super.c +2 -5
index 46a00134a36ae1..a6a5e21e4fd136 100644
@@ -1447,8 +1447,7 @@ static int register_bdev(struct cache_sb *sb, struct cache_sb_disk *sb_disk,
 		goto err;
 
 	err = "error creating kobject";
-	if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj,
-			"bcache"))
+	if (kobject_add(&dc->disk.kobj, bdev_kobj(bdev), "bcache"))
 		goto err;
 	if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
 		goto err;
@@ -2342,9 +2341,7 @@ static int register_cache(struct cache_sb *sb, struct cache_sb_disk *sb_disk,
 		goto err;
 	}
 
-	if (kobject_add(&ca->kobj,
-			&part_to_dev(bdev->bd_part)->kobj,
-			"bcache")) {
+	if (kobject_add(&ca->kobj, bdev_kobj(bdev), "bcache")) {
 		err = "error calling kobject_add";
 		ret = -ENOMEM;
 		goto out;
drivers/md/md.c +1 -3
index b2edf5e0f965b5..7ce6047c856ea2 100644
@@ -2414,7 +2414,6 @@ EXPORT_SYMBOL(md_integrity_add_rdev);
 static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
 {
 	char b[BDEVNAME_SIZE];
-	struct kobject *ko;
 	int err;
 
 	/* prevent duplicates */
@@ -2477,9 +2476,8 @@ static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
 	if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
 		goto fail;
 
-	ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
 	/* failure here is OK */
-	err = sysfs_create_link(&rdev->kobj, ko, "block");
+	err = sysfs_create_link(&rdev->kobj, bdev_kobj(rdev->bdev), "block");
 	rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
 	rdev->sysfs_unack_badblocks =
 		sysfs_get_dirent_safe(rdev->kobj.sd, "unacknowledged_bad_blocks");
fs/block_dev.c +3 -3
index 60492620d51866..f6a2a06ad262fa 100644
@@ -1242,7 +1242,7 @@ int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
 	holder->disk = disk;
 	holder->refcnt = 1;
 
-	ret = add_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
+	ret = add_symlink(disk->slave_dir, bdev_kobj(bdev));
 	if (ret)
 		goto out_free;
 
@@ -1259,7 +1259,7 @@ int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
 	goto out_unlock;
 
 out_del:
-	del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
+	del_symlink(disk->slave_dir, bdev_kobj(bdev));
 out_free:
 	kfree(holder);
 out_unlock:
@@ -1287,7 +1287,7 @@ void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
 	holder = bd_find_holder_disk(bdev, disk);
 
 	if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
-		del_symlink(disk->slave_dir, &part_to_dev(bdev->bd_part)->kobj);
+		del_symlink(disk->slave_dir, bdev_kobj(bdev));
 		del_symlink(bdev->bd_part->holder_dir,
 			    &disk_to_dev(disk)->kobj);
 		kobject_put(bdev->bd_part->holder_dir);
fs/btrfs/sysfs.c +3 -12
index 279d9262b676d4..24b6c6dc69000a 100644
@@ -1232,8 +1232,6 @@ int btrfs_sysfs_add_space_info_type(struct btrfs_fs_info *fs_info,
 
 void btrfs_sysfs_remove_device(struct btrfs_device *device)
 {
-	struct hd_struct *disk;
-	struct kobject *disk_kobj;
 	struct kobject *devices_kobj;
 
 	/*
@@ -1243,11 +1241,8 @@ void btrfs_sysfs_remove_device(struct btrfs_device *device)
 	devices_kobj = device->fs_info->fs_devices->devices_kobj;
 	ASSERT(devices_kobj);
 
-	if (device->bdev) {
-		disk = device->bdev->bd_part;
-		disk_kobj = &part_to_dev(disk)->kobj;
-		sysfs_remove_link(devices_kobj, disk_kobj->name);
-	}
+	if (device->bdev)
+		sysfs_remove_link(devices_kobj, bdev_kobj(device->bdev)->name);
 
 	if (device->devid_kobj.state_initialized) {
 		kobject_del(&device->devid_kobj);
@@ -1353,11 +1348,7 @@ int btrfs_sysfs_add_device(struct btrfs_device *device)
 	nofs_flag = memalloc_nofs_save();
 
 	if (device->bdev) {
-		struct hd_struct *disk;
-		struct kobject *disk_kobj;
-
-		disk = device->bdev->bd_part;
-		disk_kobj = &part_to_dev(disk)->kobj;
+		struct kobject *disk_kobj = bdev_kobj(device->bdev);
 
 		ret = sysfs_create_link(devices_kobj, disk_kobj, disk_kobj->name);
 		if (ret) {
include/linux/blk_types.h +3 -0
index ebfb4e7c1fd125..9698f459cc65c9 100644
@@ -49,6 +49,9 @@ struct block_device {
 	struct super_block	*bd_fsfreeze_sb;
 } __randomize_layout;
 
+#define bdev_kobj(_bdev) \
+	(&part_to_dev((_bdev)->bd_part)->kobj)
+
 /*
  * Block error status values.  See block/blk-core:blk_errors for the details.
  * Alpha cannot write a byte atomically, so we need to use 32-bit value.
Signature · 1 line
2.29.2
#29 lore raw

Re: [PATCH 13/45] block: add a bdev_kobj helper

Quoted 6 lines · Christoph Hellwig

Add a little helper to find the kobject for a struct block_device.

For the bcache part, Acked-by: Coly Li <[email protected]>

Thanks.

Coly Li

Quoted 33 lines
---
 drivers/md/bcache/super.c |  7 ++-----
 drivers/md/md.c           |  4 +---
 fs/block_dev.c            |  6 +++---
 fs/btrfs/sysfs.c          | 15 +++------------
 include/linux/blk_types.h |  3 +++
 5 files changed, 12 insertions(+), 23 deletions(-)
drivers/md/bcache/super.c +2 -5
index 46a00134a36ae1..a6a5e21e4fd136 100644
@@ -1447,8 +1447,7 @@ static int register_bdev(struct cache_sb *sb, struct cache_sb_disk *sb_disk,
 		goto err;
 
 	err = "error creating kobject";
-	if (kobject_add(&dc->disk.kobj, &part_to_dev(bdev->bd_part)->kobj,
-			"bcache"))
+	if (kobject_add(&dc->disk.kobj, bdev_kobj(bdev), "bcache"))
 		goto err;
 	if (bch_cache_accounting_add_kobjs(&dc->accounting, &dc->disk.kobj))
 		goto err;
@@ -2342,9 +2341,7 @@ static int register_cache(struct cache_sb *sb, struct cache_sb_disk *sb_disk,
 		goto err;
 	}
 
-	if (kobject_add(&ca->kobj,
-			&part_to_dev(bdev->bd_part)->kobj,
-			"bcache")) {
+	if (kobject_add(&ca->kobj, bdev_kobj(bdev), "bcache")) {
 		err = "error calling kobject_add";
 		ret = -ENOMEM;
 		goto out;

[snipped]

author maintainer
#31 lore raw

PATCH 14/45 block: use disk_part_iter_exit in disk_part_iter_next

In reply to Christoph Hellwig

Call disk_part_iter_exit in disk_part_iter_next instead of duplicating the functionality.

---
 block/genhd.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
block/genhd.c +1 -2
index 4e039524f92b8f..0bd9c41dd4cb69 100644
@@ -227,8 +227,7 @@ struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
 	int inc, end;
 
 	/* put the last partition */
-	disk_put_part(piter->part);
-	piter->part = NULL;
+	disk_part_iter_exit(piter);
 
 	/* get part_tbl */
 	rcu_read_lock();
Signature · 1 line
2.29.2
maintainer Acked-by on v1 14/45
#32 lore raw

Re: [PATCH 14/45] block: use disk_part_iter_exit in disk_part_iter_next

Quoted 2 lines · Christoph Hellwig

Call disk_part_iter_exit in disk_part_iter_next instead of duplicating the functionality.

Signature · 1 line
tejun
author maintainer
#33 lore raw

PATCH 15/45 block: use put_device in put_disk

In reply to Christoph Hellwig

Use put_device to put the device instead of poking into the internals and using kobject_put.

---
 block/genhd.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
block/genhd.c +1 -1
index 0bd9c41dd4cb69..f46e89226fdf91 100644
@@ -1803,7 +1803,7 @@ EXPORT_SYMBOL(__alloc_disk_node);
 void put_disk(struct gendisk *disk)
 {
 	if (disk)
-		kobject_put(&disk_to_dev(disk)->kobj);
+		put_device(disk_to_dev(disk));
 }
 EXPORT_SYMBOL(put_disk);
 
Signature · 1 line
2.29.2
maintainer Acked-by on v1 15/45
#34 lore raw

Re: [PATCH 15/45] block: use put_device in put_disk

Quoted 7 lines · Christoph Hellwig

Use put_device to put the device instead of poking into the internals and using kobject_put.

Signature · 1 line
tejun
author maintainer
#35 lore raw

PATCH 16/45 block: change the hash used for looking up block devices

In reply to Christoph Hellwig

Adding the minor to the major creates tons of pointless conflicts. Just use the dev_t itself, which is 32-bits and thus is guaranteed to fit into ino_t.

---
 fs/block_dev.c | 26 ++------------------------
 1 file changed, 2 insertions(+), 24 deletions(-)
fs/block_dev.c +2 -24
index f6a2a06ad262fa..437f67e12b2838 100644
@@ -863,35 +863,12 @@ void __init bdev_cache_init(void)
 	blockdev_superblock = bd_mnt->mnt_sb;   /* For writeback */
 }
 
-/*
- * Most likely _very_ bad one - but then it's hardly critical for small
- * /dev and can be fixed when somebody will need really large one.
- * Keep in mind that it will be fed through icache hash function too.
- */
-static inline unsigned long hash(dev_t dev)
-{
-	return MAJOR(dev)+MINOR(dev);
-}
-
-static int bdev_test(struct inode *inode, void *data)
-{
-	return BDEV_I(inode)->bdev.bd_dev == *(dev_t *)data;
-}
-
-static int bdev_set(struct inode *inode, void *data)
-{
-	BDEV_I(inode)->bdev.bd_dev = *(dev_t *)data;
-	return 0;
-}
-
 static struct block_device *bdget(dev_t dev)
 {
 	struct block_device *bdev;
 	struct inode *inode;
 
-	inode = iget5_locked(blockdev_superblock, hash(dev),
-			bdev_test, bdev_set, &dev);
-
+	inode = iget_locked(blockdev_superblock, dev);
 	if (!inode)
 		return NULL;
 
@@ -903,6 +880,7 @@ static struct block_device *bdget(dev_t dev)
 		bdev->bd_super = NULL;
 		bdev->bd_inode = inode;
 		bdev->bd_part_count = 0;
+		bdev->bd_dev = dev;
 		inode->i_mode = S_IFBLK;
 		inode->i_rdev = dev;
 		inode->i_bdev = bdev;
Signature · 1 line
2.29.2
maintainer Acked-by on v1 16/45
#36 lore raw

Re: [PATCH 16/45] block: change the hash used for looking up block devices

Quoted 7 lines · Christoph Hellwig

Adding the minor to the major creates tons of pointless conflicts. Just use the dev_t itself, which is 32-bits and thus is guaranteed to fit into ino_t.

Signature · 1 line
tejun
author maintainer
#37 lore raw

PATCH 17/45 init: refactor name_to_dev_t

In reply to Christoph Hellwig

Split each case into a self-contained helper, and move the block dependent code entirely under the pre-existing #ifdef CONFIG_BLOCK. This allows to remove the blk_lookup_devt stub in genhd.h.

---
 include/linux/genhd.h |   7 +-
 init/do_mounts.c      | 183 +++++++++++++++++++++---------------------
 2 files changed, 91 insertions(+), 99 deletions(-)
include/linux/genhd.h +1 -6
index 22f5b9fd96f8bf..ca5e356084c353 100644
@@ -388,18 +388,13 @@ static inline void bd_unlink_disk_holder(struct block_device *bdev,
 }
 #endif /* CONFIG_SYSFS */
 
+dev_t blk_lookup_devt(const char *name, int partno);
 #ifdef CONFIG_BLOCK
 void printk_all_partitions(void);
-dev_t blk_lookup_devt(const char *name, int partno);
 #else /* CONFIG_BLOCK */
 static inline void printk_all_partitions(void)
 {
 }
-static inline dev_t blk_lookup_devt(const char *name, int partno)
-{
-	dev_t devt = MKDEV(0, 0);
-	return devt;
-}
 #endif /* CONFIG_BLOCK */
 
 #endif /* _LINUX_GENHD_H */
init/do_mounts.c +90 -93

205 lines not on this page · raw message

Signature · 1 line
2.29.2
maintainer Acked-by on v1 17/45
#38 lore raw

Re: [PATCH 17/45] init: refactor name_to_dev_t

Quoted 8 lines · Christoph Hellwig

Split each case into a self-contained helper, and move the block dependent code entirely under the pre-existing #ifdef CONFIG_BLOCK. This allows to remove the blk_lookup_devt stub in genhd.h.

Signature · 1 line
tejun
author maintainer
#39 lore raw

PATCH 18/45 init: refactor devt_from_partuuid

In reply to Christoph Hellwig

The code in devt_from_partuuid is very convoluted. Refactor a bit by sanitizing the goto and variable name usage.

---
 init/do_mounts.c | 68 ++++++++++++++++++++++--------------------------
 1 file changed, 31 insertions(+), 37 deletions(-)
init/do_mounts.c +31 -37
index aef2f24461c7f1..afa26a4028d25e 100644
@@ -105,13 +105,10 @@ static int match_dev_by_uuid(struct device *dev, const void *data)
  */
 static dev_t devt_from_partuuid(const char *uuid_str)
 {
-	dev_t res = 0;
 	struct uuidcmp cmp;
 	struct device *dev = NULL;
-	struct gendisk *disk;
-	struct hd_struct *part;
+	dev_t devt = 0;
 	int offset = 0;
-	bool clear_root_wait = false;
 	char *slash;
 
 	cmp.uuid = uuid_str;
@@ -120,52 +117,49 @@ static dev_t devt_from_partuuid(const char *uuid_str)
 	/* Check for optional partition number offset attributes. */
 	if (slash) {
 		char c = 0;
+
 		/* Explicitly fail on poor PARTUUID syntax. */
-		if (sscanf(slash + 1,
-			   "PARTNROFF=%d%c", &offset, &c) != 1) {
-			clear_root_wait = true;
-			goto done;
-		}
+		if (sscanf(slash + 1, "PARTNROFF=%d%c", &offset, &c) != 1)
+			goto clear_root_wait;
 		cmp.len = slash - uuid_str;
 	} else {
 		cmp.len = strlen(uuid_str);
 	}
 
-	if (!cmp.len) {
-		clear_root_wait = true;
-		goto done;
-	}
+	if (!cmp.len)
+		goto clear_root_wait;
 
-	dev = class_find_device(&block_class, NULL, &cmp,
-				&match_dev_by_uuid);
+	dev = class_find_device(&block_class, NULL, &cmp, &match_dev_by_uuid);
 	if (!dev)
-		goto done;
-
-	res = dev->devt;
+		return 0;
 
-	/* Attempt to find the partition by offset. */
-	if (!offset)
-		goto no_offset;
+	if (offset) {
+		/*
+		 * Attempt to find the requested partition by adding an offset
+		 * to the partition number found by UUID.
+		 */
+		struct hd_struct *part;
 
-	res = 0;
-	disk = part_to_disk(dev_to_part(dev));
-	part = disk_get_part(disk, dev_to_part(dev)->partno + offset);
-	if (part) {
-		res = part_devt(part);
-		put_device(part_to_dev(part));
+		part = disk_get_part(dev_to_disk(dev),
+				     dev_to_part(dev)->partno + offset);
+		if (part) {
+			devt = part_devt(part);
+			put_device(part_to_dev(part));
+		}
+	} else {
+		devt = dev->devt;
 	}
 
-no_offset:
 	put_device(dev);
-done:
-	if (clear_root_wait) {
-		pr_err("VFS: PARTUUID= is invalid.\n"
-		       "Expected PARTUUID=<valid-uuid-id>[/PARTNROFF=%%d]\n");
-		if (root_wait)
-			pr_err("Disabling rootwait; root= is invalid.\n");
-		root_wait = 0;
-	}
-	return res;
+	return devt;
+
+clear_root_wait:
+	pr_err("VFS: PARTUUID= is invalid.\n"
+	       "Expected PARTUUID=<valid-uuid-id>[/PARTNROFF=%%d]\n");
+	if (root_wait)
+		pr_err("Disabling rootwait; root= is invalid.\n");
+	root_wait = 0;
+	return 0;
 }
 
 /**
Signature · 1 line
2.29.2
author maintainer
#41 lore raw

PATCH 19/45 init: cleanup match_dev_by_uuid and match_dev_by_label

In reply to Christoph Hellwig

Avoid a totally pointless goto label, and use the same style of comparism for both helpers.

---
 init/do_mounts.c | 18 ++++++------------
 1 file changed, 6 insertions(+), 12 deletions(-)
init/do_mounts.c +6 -12
index afa26a4028d25e..5879edf083b318 100644
@@ -79,15 +79,10 @@ static int match_dev_by_uuid(struct device *dev, const void *data)
 	const struct uuidcmp *cmp = data;
 	struct hd_struct *part = dev_to_part(dev);
 
-	if (!part->info)
-		goto no_match;
-
-	if (strncasecmp(cmp->uuid, part->info->uuid, cmp->len))
-		goto no_match;
-
+	if (!part->info ||
+	    strncasecmp(cmp->uuid, part->info->uuid, cmp->len))
+		return 0;
 	return 1;
-no_match:
-	return 0;
 }
 
 /**
@@ -174,10 +169,9 @@ static int match_dev_by_label(struct device *dev, const void *data)
 	const char *label = data;
 	struct hd_struct *part = dev_to_part(dev);
 
-	if (part->info && !strcmp(label, part->info->volname))
-		return 1;
-
-	return 0;
+	if (!part->info || strcmp(label, part->info->volname))
+		return 0;
+	return 1;
 }
 
 static dev_t devt_from_partlabel(const char *label)
Signature · 1 line
2.29.2
maintainer Acked-by on v1 19/45
#42 lore raw

Re: [PATCH 19/45] init: cleanup match_dev_by_uuid and match_dev_by_label

Quoted 7 lines · Christoph Hellwig

Avoid a totally pointless goto label, and use the same style of comparism for both helpers.

Signature · 1 line
tejun
author maintainer
#43 lore raw

PATCH 20/45 block: refactor __blkdev_put

In reply to Christoph Hellwig

Reorder the code to have one big section for the last close, and to use bdev_is_partition.

---
 fs/block_dev.c | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)
fs/block_dev.c +7 -7
index 437f67e12b2838..88847839ef0102 100644
@@ -1738,22 +1738,22 @@ static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
 		WARN_ON_ONCE(bdev->bd_holders);
 		sync_blockdev(bdev);
 		kill_bdev(bdev);
-
 		bdev_write_inode(bdev);
-	}
-	if (bdev->bd_contains == bdev) {
-		if (disk->fops->release)
+
+		if (!bdev_is_partition(bdev) && disk->fops->release)
 			disk->fops->release(disk, mode);
-	}
-	if (!bdev->bd_openers) {
+
 		disk_put_part(bdev->bd_part);
 		bdev->bd_part = NULL;
 		bdev->bd_disk = NULL;
-		if (bdev != bdev->bd_contains)
+		if (bdev_is_partition(bdev))
 			victim = bdev->bd_contains;
 		bdev->bd_contains = NULL;
 
 		put_disk_and_module(disk);
+	} else {
+		if (!bdev_is_partition(bdev) && disk->fops->release)
+			disk->fops->release(disk, mode);
 	}
 	mutex_unlock(&bdev->bd_mutex);
 	bdput(bdev);
Signature · 1 line
2.29.2
author maintainer
#45 lore raw

PATCH 21/45 block: refactor blkdev_get

In reply to Christoph Hellwig

Move more code that is only run on the outer open but not the open of the underlying whole device when opening a partition into blkdev_get, which leads to a much easier to follow structure.

This allows to simplify the disk and module refcounting so that one reference is held for each open, similar to what we do with normal file operations.

---
 fs/block_dev.c | 185 +++++++++++++++++++++++--------------------------
 1 file changed, 86 insertions(+), 99 deletions(-)
fs/block_dev.c +86 -99
index 88847839ef0102..2ffa11a95f10db 100644
@@ -1403,46 +1403,12 @@ EXPORT_SYMBOL_GPL(bdev_disk_changed);
  *  mutex_lock(part->bd_mutex)
  *    mutex_lock_nested(whole->bd_mutex, 1)
  */
-
-static int __blkdev_get(struct block_device *bdev, fmode_t mode, void *holder,
-		int for_part)
+static int __blkdev_get(struct block_device *bdev, struct gendisk *disk,
+		int partno, fmode_t mode)
 {
-	struct block_device *whole = NULL, *claiming = NULL;
-	struct gendisk *disk;
 	int ret;
-	int partno;
-	bool first_open = false, unblock_events = true, need_restart;
-
- restart:
-	need_restart = false;
-	ret = -ENXIO;
-	disk = bdev_get_gendisk(bdev, &partno);
-	if (!disk)
-		goto out;
-
-	if (partno) {
-		whole = bdget_disk(disk, 0);
-		if (!whole) {
-			ret = -ENOMEM;
-			goto out_put_disk;
-		}
-	}
 
-	if (!for_part && (mode & FMODE_EXCL)) {
-		WARN_ON_ONCE(!holder);
-		if (whole)
-			claiming = whole;
-		else
-			claiming = bdev;
-		ret = bd_prepare_to_claim(bdev, claiming, holder);
-		if (ret)
-			goto out_put_whole;
-	}
-
-	disk_block_events(disk);
-	mutex_lock_nested(&bdev->bd_mutex, for_part);
 	if (!bdev->bd_openers) {
-		first_open = true;
 		bdev->bd_disk = disk;
 		bdev->bd_contains = bdev;
 		bdev->bd_partno = partno;
@@ -1454,15 +1420,8 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode, void *holder,
 				goto out_clear;
 
 			ret = 0;
-			if (disk->fops->open) {
+			if (disk->fops->open)
 				ret = disk->fops->open(bdev, mode);
-				/*
-				 * If we lost a race with 'disk' being deleted,
-				 * try again.  See md.c
-				 */
-				if (ret == -ERESTARTSYS)
-					need_restart = true;
-			}
 
 			if (!ret) {
 				bd_set_nr_sectors(bdev, get_capacity(disk));
@@ -1482,14 +1441,23 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode, void *holder,
 			if (ret)
 				goto out_clear;
 		} else {
-			BUG_ON(for_part);
-			ret = __blkdev_get(whole, mode, NULL, 1);
-			if (ret)
+			struct block_device *whole = bdget_disk(disk, 0);
+
+			mutex_lock_nested(&whole->bd_mutex, 1);
+			ret = __blkdev_get(whole, disk, 0, mode);
+			if (ret) {
+				mutex_unlock(&whole->bd_mutex);
+				bdput(whole);
 				goto out_clear;
-			bdev->bd_contains = bdgrab(whole);
+			}
+			whole->bd_part_count++;
+			mutex_unlock(&whole->bd_mutex);
+
+			bdev->bd_contains = whole;
 			bdev->bd_part = disk_get_part(disk, partno);
 			if (!(disk->flags & GENHD_FL_UP) ||
 			    !bdev->bd_part || !bdev->bd_part->nr_sects) {
+				__blkdev_put(whole, mode, 1);
 				ret = -ENXIO;
 				goto out_clear;
 			}
@@ -1509,58 +1477,17 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode, void *holder,
 			    (!ret || ret == -ENOMEDIUM))
 				bdev_disk_changed(bdev, ret == -ENOMEDIUM);
 			if (ret)
-				goto out_unlock_bdev;
+				return ret;
 		}
 	}
 	bdev->bd_openers++;
-	if (for_part)
-		bdev->bd_part_count++;
-	if (claiming)
-		bd_finish_claiming(bdev, claiming, holder);
-
-	/*
-	 * Block event polling for write claims if requested.  Any write holder
-	 * makes the write_holder state stick until all are released.  This is
-	 * good enough and tracking individual writeable reference is too
-	 * fragile given the way @mode is used in blkdev_get/put().
-	 */
-	if (claiming && (mode & FMODE_WRITE) && !bdev->bd_write_holder &&
-	    (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
-		bdev->bd_write_holder = true;
-		unblock_events = false;
-	}
-	mutex_unlock(&bdev->bd_mutex);
-
-	if (unblock_events)
-		disk_unblock_events(disk);
-
-	/* only one opener holds refs to the module and disk */
-	if (!first_open)
-		put_disk_and_module(disk);
-	if (whole)
-		bdput(whole);
 	return 0;
 
  out_clear:
 	disk_put_part(bdev->bd_part);
 	bdev->bd_disk = NULL;
 	bdev->bd_part = NULL;
-	if (bdev != bdev->bd_contains)
-		__blkdev_put(bdev->bd_contains, mode, 1);
 	bdev->bd_contains = NULL;
- out_unlock_bdev:
-	if (claiming)
-		bd_abort_claiming(bdev, claiming, holder);
-	mutex_unlock(&bdev->bd_mutex);
-	disk_unblock_events(disk);
- out_put_whole:
- 	if (whole)
-		bdput(whole);
- out_put_disk:
-	put_disk_and_module(disk);
-	if (need_restart)
-		goto restart;
- out:
 	return ret;
 }
 
@@ -1585,7 +1512,12 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode, void *holder,
  */
 static int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
 {
-	int ret, perm = 0;
+	struct block_device *claiming;
+	bool unblock_events = true;
+	struct gendisk *disk;
+	int perm = 0;
+	int partno;
+	int ret;
 
 	if (mode & FMODE_READ)
 		perm |= MAY_READ;
@@ -1595,13 +1527,67 @@ static int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
 	if (ret)
 		goto bdput;
 
-	ret =__blkdev_get(bdev, mode, holder, 0);
-	if (ret)
+	/*
+	 * If we lost a race with 'disk' being deleted, try again.  See md.c.
+	 */
+retry:
+	ret = -ENXIO;
+	disk = bdev_get_gendisk(bdev, &partno);
+	if (!disk)
 		goto bdput;
-	return 0;
 
+	if (mode & FMODE_EXCL) {
+		WARN_ON_ONCE(!holder);
+	
+		ret = -ENOMEM;
+		claiming = bdget_disk(disk, 0);
+		if (!claiming)
+			goto put_disk;
+		ret = bd_prepare_to_claim(bdev, claiming, holder);
+		if (ret)
+			goto put_claiming;
+	}
+
+	disk_block_events(disk);
+
+	mutex_lock(&bdev->bd_mutex);
+	ret =__blkdev_get(bdev, disk, partno, mode);
+	if (!(mode & FMODE_EXCL)) {
+		; /* nothing to do here */
+	} else if (ret) {
+		bd_abort_claiming(bdev, claiming, holder);
+	} else {
+		bd_finish_claiming(bdev, claiming, holder);
+
+		/*
+		 * Block event polling for write claims if requested.  Any write
+		 * holder makes the write_holder state stick until all are
+		 * released.  This is good enough and tracking individual
+		 * writeable reference is too fragile given the way @mode is
+		 * used in blkdev_get/put().
+		 */
+		if ((mode & FMODE_WRITE) && !bdev->bd_write_holder &&
+		    (disk->flags & GENHD_FL_BLOCK_EVENTS_ON_EXCL_WRITE)) {
+			bdev->bd_write_holder = true;
+			unblock_events = false;
+		}
+	}
+	mutex_unlock(&bdev->bd_mutex);
+
+	if (unblock_events)
+		disk_unblock_events(disk);
+
+put_claiming:
+	if (mode & FMODE_EXCL)
+		bdput(claiming);
+put_disk:
+	if (ret)
+		put_disk_and_module(disk);
+	if (ret == -ERESTARTSYS)
+		goto retry;
 bdput:
-	bdput(bdev);
+	if (ret)
+		bdput(bdev);
 	return ret;
 }
 
@@ -1749,8 +1735,6 @@ static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
 		if (bdev_is_partition(bdev))
 			victim = bdev->bd_contains;
 		bdev->bd_contains = NULL;
-
-		put_disk_and_module(disk);
 	} else {
 		if (!bdev_is_partition(bdev) && disk->fops->release)
 			disk->fops->release(disk, mode);
@@ -1763,6 +1747,8 @@ static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
 
 void blkdev_put(struct block_device *bdev, fmode_t mode)
 {
+	struct gendisk *disk = bdev->bd_disk;
+
 	mutex_lock(&bdev->bd_mutex);
 
 	if (mode & FMODE_EXCL) {
@@ -1791,7 +1777,7 @@ void blkdev_put(struct block_device *bdev, fmode_t mode)
 		 * unblock evpoll if it was a write holder.
 		 */
 		if (bdev_free && bdev->bd_write_holder) {
-			disk_unblock_events(bdev->bd_disk);
+			disk_unblock_events(disk);
 			bdev->bd_write_holder = false;
 		}
 	}
@@ -1801,11 +1787,12 @@ void blkdev_put(struct block_device *bdev, fmode_t mode)
 	 * event.  This is to ensure detection of media removal commanded
 	 * from userland - e.g. eject(1).
 	 */
-	disk_flush_events(bdev->bd_disk, DISK_EVENT_MEDIA_CHANGE);
+	disk_flush_events(disk, DISK_EVENT_MEDIA_CHANGE);
 
 	mutex_unlock(&bdev->bd_mutex);
 
 	__blkdev_put(bdev, mode, 0);
+	put_disk_and_module(disk);
 }
 EXPORT_SYMBOL(blkdev_put);
 
Signature · 1 line
2.29.2
maintainer Acked-by on v1 21/45
#46 lore raw

Re: [PATCH 21/45] block: refactor blkdev_get

Quoted 9 lines · Christoph Hellwig

Move more code that is only run on the outer open but not the open of the underlying whole device when opening a partition into blkdev_get, which leads to a much easier to follow structure.

This allows to simplify the disk and module refcounting so that one reference is held for each open, similar to what we do with normal file operations.

Thanks.

Signature · 1 line
tejun
author maintainer
#47 lore raw

PATCH 22/45 block: opencode devcgroup_inode_permission

In reply to Christoph Hellwig

Just call devcgroup_check_permission to avoid various superflous checks and a double conversion of the access flags.

---
 fs/block_dev.c | 10 ++++------
 1 file changed, 4 insertions(+), 6 deletions(-)
fs/block_dev.c +4 -6
index 2ffa11a95f10db..1e35faf6dad42c 100644
@@ -1515,15 +1515,13 @@ static int blkdev_get(struct block_device *bdev, fmode_t mode, void *holder)
 	struct block_device *claiming;
 	bool unblock_events = true;
 	struct gendisk *disk;
-	int perm = 0;
 	int partno;
 	int ret;
 
-	if (mode & FMODE_READ)
-		perm |= MAY_READ;
-	if (mode & FMODE_WRITE)
-		perm |= MAY_WRITE;
-	ret = devcgroup_inode_permission(bdev->bd_inode, perm);
+	ret = devcgroup_check_permission(DEVCG_DEV_BLOCK,
+			imajor(bdev->bd_inode), iminor(bdev->bd_inode),
+			((mode & FMODE_READ) ? DEVCG_ACC_READ : 0) |
+			((mode & FMODE_WRITE) ? DEVCG_ACC_WRITE : 0));
 	if (ret)
 		goto bdput;
 
Signature · 1 line
2.29.2
author maintainer
#49 lore raw

PATCH 23/45 block: remove i_bdev

In reply to Christoph Hellwig

Switch the block device lookup interfaces to directly work with a dev_t so that struct block_device references are only acquired by the blkdev_get variants (and the blk-cgroup special case). This means that we not don't need an extra reference in the inode and can generally simplify handling of struct block_device to keep the lookups contained in the core block layer code.

---
 block/ioctl.c                                |   3 +-
 drivers/block/loop.c                         |   8 +-
 drivers/md/bcache/super.c                    |  20 +-
 drivers/md/dm-table.c                        |   9 +-
 drivers/mtd/mtdsuper.c                       |  17 +-
 drivers/target/target_core_file.c            |   6 +-
 drivers/usb/gadget/function/storage_common.c |   8 +-
 fs/block_dev.c                               | 195 +++++--------------
 fs/btrfs/volumes.c                           |  13 +-
 fs/inode.c                                   |   3 -
 fs/internal.h                                |   7 +-
 fs/io_uring.c                                |  10 +-
 fs/pipe.c                                    |   5 +-
 fs/quota/quota.c                             |  19 +-
 fs/statfs.c                                  |   2 +-
 fs/super.c                                   |  37 ++--
 include/linux/blkdev.h                       |   2 +-
 include/linux/fs.h                           |   1 -
 18 files changed, 114 insertions(+), 251 deletions(-)
block/ioctl.c +1 -2
index 0c09bb7a6ff35f..a6d8171221c7dc 100644
@@ -590,8 +590,7 @@ long compat_blkdev_ioctl(struct file *file, unsigned cmd, unsigned long arg)
 {
 	int ret;
 	void __user *argp = compat_ptr(arg);
-	struct inode *inode = file->f_mapping->host;
-	struct block_device *bdev = inode->i_bdev;
+	struct block_device *bdev = I_BDEV(file->f_mapping->host);
 	struct gendisk *disk = bdev->bd_disk;
 	fmode_t mode = file->f_mode;
 	loff_t size;
drivers/block/loop.c +3 -5
index b42c728620c9e4..26c7aafba7c5f8 100644
@@ -675,10 +675,10 @@ static int loop_validate_file(struct file *file, struct block_device *bdev)
 	while (is_loop_device(f)) {
 		struct loop_device *l;
 
-		if (f->f_mapping->host->i_bdev == bdev)
+		if (f->f_mapping->host->i_rdev == bdev->bd_dev)
 			return -EBADF;
 
-		l = f->f_mapping->host->i_bdev->bd_disk->private_data;
+		l = I_BDEV(f->f_mapping->host)->bd_disk->private_data;
 		if (l->lo_state != Lo_bound) {
 			return -EINVAL;
 		}
@@ -885,9 +885,7 @@ static void loop_config_discard(struct loop_device *lo)
 	 * file-backed loop devices: discarded regions read back as zero.
 	 */
 	if (S_ISBLK(inode->i_mode) && !lo->lo_encrypt_key_size) {
-		struct request_queue *backingq;
-
-		backingq = bdev_get_queue(inode->i_bdev);
+		struct request_queue *backingq = bdev_get_queue(I_BDEV(inode));
 
 		max_discard_sectors = backingq->limits.max_write_zeroes_sectors;
 		granularity = backingq->limits.discard_granularity ?:
drivers/md/bcache/super.c +11 -9
index a6a5e21e4fd136..c55d3c58a7ef55 100644
@@ -2380,38 +2380,38 @@ kobj_attribute_write(register,		register_bcache);
 kobj_attribute_write(register_quiet,	register_bcache);
 kobj_attribute_write(pendings_cleanup,	bch_pending_bdevs_cleanup);
 
-static bool bch_is_open_backing(struct block_device *bdev)
+static bool bch_is_open_backing(dev_t dev)
 {
 	struct cache_set *c, *tc;
 	struct cached_dev *dc, *t;
 
 	list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
 		list_for_each_entry_safe(dc, t, &c->cached_devs, list)
-			if (dc->bdev == bdev)
+			if (dc->bdev->bd_dev == dev)
 				return true;
 	list_for_each_entry_safe(dc, t, &uncached_devices, list)
-		if (dc->bdev == bdev)
+		if (dc->bdev->bd_dev == dev)
 			return true;
 	return false;
 }
 
-static bool bch_is_open_cache(struct block_device *bdev)
+static bool bch_is_open_cache(dev_t dev)
 {
 	struct cache_set *c, *tc;
 
 	list_for_each_entry_safe(c, tc, &bch_cache_sets, list) {
 		struct cache *ca = c->cache;
 
-		if (ca->bdev == bdev)
+		if (ca->bdev->bd_dev == dev)
 			return true;
 	}
 
 	return false;
 }
 
-static bool bch_is_open(struct block_device *bdev)
+static bool bch_is_open(dev_t dev)
 {
-	return bch_is_open_cache(bdev) || bch_is_open_backing(bdev);
+	return bch_is_open_cache(dev) || bch_is_open_backing(dev);
 }
 
 struct async_reg_args {
@@ -2535,9 +2535,11 @@ static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
 				  sb);
 	if (IS_ERR(bdev)) {
 		if (bdev == ERR_PTR(-EBUSY)) {
-			bdev = lookup_bdev(strim(path));
+			dev_t dev;
+
 			mutex_lock(&bch_register_lock);
-			if (!IS_ERR(bdev) && bch_is_open(bdev))
+			if (lookup_bdev(strim(path), &dev) == 0 &&
+			    bch_is_open(dev))
 				err = "device already registered";
 			else
 				err = "device busy";
drivers/md/dm-table.c +1 -8
index ce543b761be7b2..dea67772171053 100644
@@ -348,16 +348,9 @@ static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
 dev_t dm_get_dev_t(const char *path)
 {
 	dev_t dev;
-	struct block_device *bdev;
 
-	bdev = lookup_bdev(path);
-	if (IS_ERR(bdev))
+	if (lookup_bdev(path, &dev))
 		dev = name_to_dev_t(path);
-	else {
-		dev = bdev->bd_dev;
-		bdput(bdev);
-	}
-
 	return dev;
 }
 EXPORT_SYMBOL_GPL(dm_get_dev_t);
drivers/mtd/mtdsuper.c +6 -11
index c3e2098372f2e5..38b6aa849c6383 100644
@@ -120,8 +120,8 @@ int get_tree_mtd(struct fs_context *fc,
 				struct fs_context *fc))
 {
 #ifdef CONFIG_BLOCK
-	struct block_device *bdev;
-	int ret, major;
+	dev_t dev;
+	int ret;
 #endif
 	int mtdnr;
 
@@ -169,20 +169,15 @@ int get_tree_mtd(struct fs_context *fc,
 	/* try the old way - the hack where we allowed users to mount
 	 * /dev/mtdblock$(n) but didn't actually _use_ the blockdev
 	 */
-	bdev = lookup_bdev(fc->source);
-	if (IS_ERR(bdev)) {
-		ret = PTR_ERR(bdev);
+	ret = lookup_bdev(fc->source, &dev);
+	if (ret) {
 		errorf(fc, "MTD: Couldn't look up '%s': %d", fc->source, ret);
 		return ret;
 	}
 	pr_debug("MTDSB: lookup_bdev() returned 0\n");
 
-	major = MAJOR(bdev->bd_dev);
-	mtdnr = MINOR(bdev->bd_dev);
-	bdput(bdev);
-
-	if (major == MTD_BLOCK_MAJOR)
-		return mtd_get_sb_by_nr(fc, mtdnr, fill_super);
+	if (MAJOR(dev) == MTD_BLOCK_MAJOR)
+		return mtd_get_sb_by_nr(fc, MINOR(dev), fill_super);
 
 #endif /* CONFIG_BLOCK */
 
drivers/target/target_core_file.c +3 -3
index 7143d03f0e027e..b0cb5b95e892d3 100644
@@ -133,10 +133,10 @@ static int fd_configure_device(struct se_device *dev)
 	 */
 	inode = file->f_mapping->host;
 	if (S_ISBLK(inode->i_mode)) {
-		struct request_queue *q = bdev_get_queue(inode->i_bdev);
+		struct request_queue *q = bdev_get_queue(I_BDEV(inode));
 		unsigned long long dev_size;
 
-		fd_dev->fd_block_size = bdev_logical_block_size(inode->i_bdev);
+		fd_dev->fd_block_size = bdev_logical_block_size(I_BDEV(inode));
 		/*
 		 * Determine the number of bytes from i_size_read() minus
 		 * one (1) logical sector from underlying struct block_device
@@ -559,7 +559,7 @@ fd_execute_unmap(struct se_cmd *cmd, sector_t lba, sector_t nolb)
 
 	if (S_ISBLK(inode->i_mode)) {
 		/* The backend is block device, use discard */
-		struct block_device *bdev = inode->i_bdev;
+		struct block_device *bdev = I_BDEV(inode);
 		struct se_device *dev = cmd->se_dev;
 
 		ret = blkdev_issue_discard(bdev,
drivers/usb/gadget/function/storage_common.c +4 -4
index f7e6c42558eb76..b859a158a4140e 100644
@@ -204,7 +204,7 @@ int fsg_lun_open(struct fsg_lun *curlun, const char *filename)
 	if (!(filp->f_mode & FMODE_WRITE))
 		ro = 1;
 
-	inode = file_inode(filp);
+	inode = filp->f_mapping->host;
 	if ((!S_ISREG(inode->i_mode) && !S_ISBLK(inode->i_mode))) {
 		LINFO(curlun, "invalid file type: %s\n", filename);
 		goto out;
@@ -221,7 +221,7 @@ int fsg_lun_open(struct fsg_lun *curlun, const char *filename)
 	if (!(filp->f_mode & FMODE_CAN_WRITE))
 		ro = 1;
 
-	size = i_size_read(inode->i_mapping->host);
+	size = i_size_read(inode);
 	if (size < 0) {
 		LINFO(curlun, "unable to find file size: %s\n", filename);
 		rc = (int) size;
@@ -231,8 +231,8 @@ int fsg_lun_open(struct fsg_lun *curlun, const char *filename)
 	if (curlun->cdrom) {
 		blksize = 2048;
 		blkbits = 11;
-	} else if (inode->i_bdev) {
-		blksize = bdev_logical_block_size(inode->i_bdev);
+	} else if (S_ISBLK(inode->i_mode)) {
+		blksize = bdev_logical_block_size(I_BDEV(inode));
 		blkbits = blksize_bits(blksize);
 	} else {
 		blksize = 512;
fs/block_dev.c +49 -146

299 lines not on this page · raw message

fs/btrfs/volumes.c +6 -7
index a6406b3b8c2b4f..fbc4b58228f784 100644
@@ -929,16 +929,16 @@ static noinline struct btrfs_device *device_list_add(const char *path,
 		 * make sure it's the same device if the device is mounted
 		 */
 		if (device->bdev) {
-			struct block_device *path_bdev;
+			int error;
+			dev_t path_dev;
 
-			path_bdev = lookup_bdev(path);
-			if (IS_ERR(path_bdev)) {
+			error = lookup_bdev(path, &path_dev);
+			if (error) {
 				mutex_unlock(&fs_devices->device_list_mutex);
-				return ERR_CAST(path_bdev);
+				return ERR_PTR(error);
 			}
 
-			if (device->bdev != path_bdev) {
-				bdput(path_bdev);
+			if (device->bdev->bd_dev != path_dev) {
 				mutex_unlock(&fs_devices->device_list_mutex);
 				btrfs_warn_in_rcu(device->fs_info,
 	"duplicate device %s devid %llu generation %llu scanned by %s (%d)",
@@ -947,7 +947,6 @@ static noinline struct btrfs_device *device_list_add(const char *path,
 						  task_pid_nr(current));
 				return ERR_PTR(-EEXIST);
 			}
-			bdput(path_bdev);
 			btrfs_info_in_rcu(device->fs_info,
 	"devid %llu device path %s changed to %s scanned by %s (%d)",
 					  devid, rcu_str_deref(device->name),
fs/inode.c +0 -3
index 9d78c37b00b817..cb008acf0efdb8 100644
@@ -155,7 +155,6 @@ int inode_init_always(struct super_block *sb, struct inode *inode)
 	inode->i_bytes = 0;
 	inode->i_generation = 0;
 	inode->i_pipe = NULL;
-	inode->i_bdev = NULL;
 	inode->i_cdev = NULL;
 	inode->i_link = NULL;
 	inode->i_dir_seq = 0;
@@ -580,8 +579,6 @@ static void evict(struct inode *inode)
 		truncate_inode_pages_final(&inode->i_data);
 		clear_inode(inode);
 	}
-	if (S_ISBLK(inode->i_mode) && inode->i_bdev)
-		bd_forget(inode);
 	if (S_ISCHR(inode->i_mode) && inode->i_cdev)
 		cd_forget(inode);
 
fs/internal.h +1 -6
index 47be21dfeebef5..53f890446e7508 100644
@@ -25,7 +25,6 @@ extern void __init bdev_cache_init(void);
 extern int __sync_blockdev(struct block_device *bdev, int wait);
 void iterate_bdevs(void (*)(struct block_device *, void *), void *);
 void emergency_thaw_bdev(struct super_block *sb);
-void bd_forget(struct inode *inode);
 #else
 static inline void bdev_cache_init(void)
 {
@@ -43,9 +42,6 @@ static inline int emergency_thaw_bdev(struct super_block *sb)
 {
 	return 0;
 }
-static inline void bd_forget(struct inode *inode)
-{
-}
 #endif /* CONFIG_BLOCK */
 
 /*
@@ -114,8 +110,7 @@ extern struct file *alloc_empty_file_noaccount(int, const struct cred *);
  */
 extern int reconfigure_super(struct fs_context *);
 extern bool trylock_super(struct super_block *sb);
-struct super_block *__get_super(struct block_device *bdev, bool excl);
-extern struct super_block *user_get_super(dev_t);
+struct super_block *user_get_super(dev_t, bool excl);
 void put_super(struct super_block *sb);
 extern bool mount_capable(struct fs_context *);
 
fs/io_uring.c +4 -6
index 4ead291b2976f3..8f13c0417f940c 100644
@@ -2716,11 +2716,7 @@ static struct file *__io_file_get(struct io_submit_state *state, int fd)
 
 static bool io_bdev_nowait(struct block_device *bdev)
 {
-#ifdef CONFIG_BLOCK
 	return !bdev || blk_queue_nowait(bdev_get_queue(bdev));
-#else
-	return true;
-#endif
 }
 
 /*
@@ -2733,14 +2729,16 @@ static bool io_file_supports_async(struct file *file, int rw)
 	umode_t mode = file_inode(file)->i_mode;
 
 	if (S_ISBLK(mode)) {
-		if (io_bdev_nowait(file->f_inode->i_bdev))
+		if (IS_ENABLED(CONFIG_BLOCK) &&
+		    io_bdev_nowait(I_BDEV(file->f_mapping->host)))
 			return true;
 		return false;
 	}
 	if (S_ISCHR(mode) || S_ISSOCK(mode))
 		return true;
 	if (S_ISREG(mode)) {
-		if (io_bdev_nowait(file->f_inode->i_sb->s_bdev) &&
+		if (IS_ENABLED(CONFIG_BLOCK) &&
+		    io_bdev_nowait(file->f_inode->i_sb->s_bdev) &&
 		    file->f_op != &io_uring_fops)
 			return true;
 		return false;
fs/pipe.c +2 -3
index 0ac197658a2d6e..c5989cfd564d45 100644
@@ -1342,9 +1342,8 @@ static long pipe_set_size(struct pipe_inode_info *pipe, unsigned long arg)
 }
 
 /*
- * After the inode slimming patch, i_pipe/i_bdev/i_cdev share the same
- * location, so checking ->i_pipe is not enough to verify that this is a
- * pipe.
+ * Note that i_pipe and i_cdev share the same location, so checking ->i_pipe is
+ * not enough to verify that this is a pipe.
  */
 struct pipe_inode_info *get_pipe_info(struct file *file, bool for_splice)
 {
fs/quota/quota.c +9 -10
index f3d32b0d9008f2..6d16b2be5ac4a3 100644
@@ -866,17 +866,18 @@ static bool quotactl_cmd_onoff(int cmd)
 static struct super_block *quotactl_block(const char __user *special, int cmd)
 {
 #ifdef CONFIG_BLOCK
-	struct block_device *bdev;
 	struct super_block *sb;
 	struct filename *tmp = getname(special);
 	bool excl = false, thawed = false;
+	int error;
+	dev_t dev;
 
 	if (IS_ERR(tmp))
 		return ERR_CAST(tmp);
-	bdev = lookup_bdev(tmp->name);
+	error = lookup_bdev(tmp->name, &dev);
 	putname(tmp);
-	if (IS_ERR(bdev))
-		return ERR_CAST(bdev);
+	if (error)
+		return ERR_PTR(error);
 
 	if (quotactl_cmd_onoff(cmd)) {
 		excl = true;
@@ -886,8 +887,10 @@ static struct super_block *quotactl_block(const char __user *special, int cmd)
 	}
 
 retry:
-	sb = __get_super(bdev, excl);
-	if (thawed && sb && sb->s_writers.frozen != SB_UNFROZEN) {
+	sb = user_get_super(dev, excl);
+	if (!sb)
+		return ERR_PTR(-ENODEV);
+	if (thawed && sb->s_writers.frozen != SB_UNFROZEN) {
 		if (excl)
 			up_write(&sb->s_umount);
 		else
@@ -897,10 +900,6 @@ static struct super_block *quotactl_block(const char __user *special, int cmd)
 		put_super(sb);
 		goto retry;
 	}
-
-	bdput(bdev);
-	if (!sb)
-		return ERR_PTR(-ENODEV);
 	return sb;
 
 #else
fs/statfs.c +1 -1
index 59f33752c1311f..68cb077887504f 100644
@@ -235,7 +235,7 @@ SYSCALL_DEFINE3(fstatfs64, unsigned int, fd, size_t, sz, struct statfs64 __user
 
 static int vfs_ustat(dev_t dev, struct kstatfs *sbuf)
 {
-	struct super_block *s = user_get_super(dev);
+	struct super_block *s = user_get_super(dev, false);
 	int err;
 	if (!s)
 		return -EINVAL;
fs/super.c +12 -25
index 343e5c1e538d2a..7a1611e5d0f45d 100644
@@ -740,7 +740,7 @@ void iterate_supers_type(struct file_system_type *type,
 
 EXPORT_SYMBOL(iterate_supers_type);
 
-struct super_block *__get_super(struct block_device *bdev, bool excl)
+struct super_block *get_super(struct block_device *bdev)
 {
 	struct super_block *sb;
 
@@ -755,17 +755,11 @@ struct super_block *__get_super(struct block_device *bdev, bool excl)
 		if (sb->s_bdev == bdev) {
 			sb->s_count++;
 			spin_unlock(&sb_lock);
-			if (!excl)
-				down_read(&sb->s_umount);
-			else
-				down_write(&sb->s_umount);
+			down_read(&sb->s_umount);
 			/* still alive? */
 			if (sb->s_root && (sb->s_flags & SB_BORN))
 				return sb;
-			if (!excl)
-				up_read(&sb->s_umount);
-			else
-				up_write(&sb->s_umount);
+			up_read(&sb->s_umount);
 			/* nope, got unmounted */
 			spin_lock(&sb_lock);
 			__put_super(sb);
@@ -776,19 +770,6 @@ struct super_block *__get_super(struct block_device *bdev, bool excl)
 	return NULL;
 }
 
-/**
- *	get_super - get the superblock of a device
- *	@bdev: device to get the superblock for
- *
- *	Scans the superblock list and finds the superblock of the file system
- *	mounted on the device given. %NULL is returned if no match is found.
- */
-struct super_block *get_super(struct block_device *bdev)
-{
-	return __get_super(bdev, false);
-}
-EXPORT_SYMBOL(get_super);
-
 /**
  * get_active_super - get an active reference to the superblock of a device
  * @bdev: device to get the superblock for
@@ -820,7 +801,7 @@ struct super_block *get_active_super(struct block_device *bdev)
 	return NULL;
 }
 
-struct super_block *user_get_super(dev_t dev)
+struct super_block *user_get_super(dev_t dev, bool excl)
 {
 	struct super_block *sb;
 
@@ -832,11 +813,17 @@ struct super_block *user_get_super(dev_t dev)
 		if (sb->s_dev ==  dev) {
 			sb->s_count++;
 			spin_unlock(&sb_lock);
-			down_read(&sb->s_umount);
+			if (excl)
+				down_write(&sb->s_umount);
+			else
+				down_read(&sb->s_umount);
 			/* still alive? */
 			if (sb->s_root && (sb->s_flags & SB_BORN))
 				return sb;
-			up_read(&sb->s_umount);
+			if (excl)
+				up_write(&sb->s_umount);
+			else
+				up_read(&sb->s_umount);
 			/* nope, got unmounted */
 			spin_lock(&sb_lock);
 			__put_super(sb);
include/linux/blkdev.h +1 -1
index 12810a19edebc4..bdd7339bcda462 100644
@@ -1973,7 +1973,7 @@ int bdev_read_only(struct block_device *bdev);
 int set_blocksize(struct block_device *bdev, int size);
 
 const char *bdevname(struct block_device *bdev, char *buffer);
-struct block_device *lookup_bdev(const char *);
+int lookup_bdev(const char *pathname, dev_t *dev);
 
 void blkdev_show(struct seq_file *seqf, off_t offset);
 
include/linux/fs.h +0 -1
index a61df0dd4f1989..b0b358309657ba 100644
@@ -696,7 +696,6 @@ struct inode {
 	struct list_head	i_devices;
 	union {
 		struct pipe_inode_info	*i_pipe;
-		struct block_device	*i_bdev;
 		struct cdev		*i_cdev;
 		char			*i_link;
 		unsigned		i_dir_seq;
Signature · 1 line
2.29.2
#50 lore raw

Re: [PATCH 23/45] block: remove i_bdev

Quoted 8 lines · Christoph Hellwig

Switch the block device lookup interfaces to directly work with a dev_t so that struct block_device references are only acquired by the blkdev_get variants (and the blk-cgroup special case). This means that we not don't need an extra reference in the inode and can generally simplify handling of struct block_device to keep the lookups contained in the core block layer code.

For the bcache part, Acked-by: Coly Li <[email protected]>

Thanks.

Coly Li

Quoted 21 lines
---
 block/ioctl.c                                |   3 +-
 drivers/block/loop.c                         |   8 +-
 drivers/md/bcache/super.c                    |  20 +-
 drivers/md/dm-table.c                        |   9 +-
 drivers/mtd/mtdsuper.c                       |  17 +-
 drivers/target/target_core_file.c            |   6 +-
 drivers/usb/gadget/function/storage_common.c |   8 +-
 fs/block_dev.c                               | 195 +++++--------------
 fs/btrfs/volumes.c                           |  13 +-
 fs/inode.c                                   |   3 -
 fs/internal.h                                |   7 +-
 fs/io_uring.c                                |  10 +-
 fs/pipe.c                                    |   5 +-
 fs/quota/quota.c                             |  19 +-
 fs/statfs.c                                  |   2 +-
 fs/super.c                                   |  37 ++--
 include/linux/blkdev.h                       |   2 +-
 include/linux/fs.h                           |   1 -
 18 files changed, 114 insertions(+), 251 deletions(-)

[snipped]

Quoted 64 lines
drivers/md/bcache/super.c +11 -9
index a6a5e21e4fd136..c55d3c58a7ef55 100644
@@ -2380,38 +2380,38 @@ kobj_attribute_write(register,		register_bcache);
 kobj_attribute_write(register_quiet,	register_bcache);
 kobj_attribute_write(pendings_cleanup,	bch_pending_bdevs_cleanup);
 
-static bool bch_is_open_backing(struct block_device *bdev)
+static bool bch_is_open_backing(dev_t dev)
 {
 	struct cache_set *c, *tc;
 	struct cached_dev *dc, *t;
 
 	list_for_each_entry_safe(c, tc, &bch_cache_sets, list)
 		list_for_each_entry_safe(dc, t, &c->cached_devs, list)
-			if (dc->bdev == bdev)
+			if (dc->bdev->bd_dev == dev)
 				return true;
 	list_for_each_entry_safe(dc, t, &uncached_devices, list)
-		if (dc->bdev == bdev)
+		if (dc->bdev->bd_dev == dev)
 			return true;
 	return false;
 }
 
-static bool bch_is_open_cache(struct block_device *bdev)
+static bool bch_is_open_cache(dev_t dev)
 {
 	struct cache_set *c, *tc;
 
 	list_for_each_entry_safe(c, tc, &bch_cache_sets, list) {
 		struct cache *ca = c->cache;
 
-		if (ca->bdev == bdev)
+		if (ca->bdev->bd_dev == dev)
 			return true;
 	}
 
 	return false;
 }
 
-static bool bch_is_open(struct block_device *bdev)
+static bool bch_is_open(dev_t dev)
 {
-	return bch_is_open_cache(bdev) || bch_is_open_backing(bdev);
+	return bch_is_open_cache(dev) || bch_is_open_backing(dev);
 }
 
 struct async_reg_args {
@@ -2535,9 +2535,11 @@ static ssize_t register_bcache(struct kobject *k, struct kobj_attribute *attr,
 				  sb);
 	if (IS_ERR(bdev)) {
 		if (bdev == ERR_PTR(-EBUSY)) {
-			bdev = lookup_bdev(strim(path));
+			dev_t dev;
+
 			mutex_lock(&bch_register_lock);
-			if (!IS_ERR(bdev) && bch_is_open(bdev))
+			if (lookup_bdev(strim(path), &dev) == 0 &&
+			    bch_is_open(dev))
 				err = "device already registered";
 			else
 				err = "device busy";

[snipped]

maintainer Acked-by on v1 23/45
#51 lore raw

Re: [PATCH 23/45] block: remove i_bdev

In reply to Christoph Hellwig
Quoted 4 lines · Christoph Hellwig

Switch the block device lookup interfaces to directly work with a dev_t so that struct block_device references are only acquired by the blkdev_get variants (and the blk-cgroup special case). This means that we not don't need an extra reference in the inode and can generally

     ^
     now
Quoted 4 lines

simplify handling of struct block_device to keep the lookups contained in the core block layer code.

...

Quoted 7 lines
fs/block_dev.c +0 -3
@@ -1689,14 +1599,12 @@ static int blkdev_open(struct inode * inode, struct file * filp)
 	if ((filp->f_flags & O_ACCMODE) == 3)
 		filp->f_mode |= FMODE_WRITE_IOCTL;
 
-	bdev = bd_acquire(inode);
-	if (bdev == NULL)
-		return -ENOMEM;

fs/block_dev.c static int blkdev_open(struct inode * inode, struct file * filp) · line 1608

Quoted 10 lines
fs/block_dev.c +4 -3
-
+	bdev = blkdev_get_by_dev(inode->i_rdev, filp->f_mode, filp);
+	if (IS_ERR(bdev))
+		return PTR_ERR(bdev);
 	filp->f_mapping = bdev->bd_inode->i_mapping;
 	filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
-
-	return blkdev_get(bdev, filp->f_mode, filp);
+	return 0;
 }

I was wondering whether losing the stale bdev flushing in bd_acquire() would cause user-visible behavior changes but can't see how it would given that userland has no way of holding onto a specific instance of block inode. Maybe it's something worth mentioning in the commit message?

Other than that, for the block part:

Thanks.

Signature · 1 line
tejun
author maintainer
#52 lore raw

Re: [PATCH 23/45] block: remove i_bdev

Quoted 26 lines · Tejun Heo
Quoted 4 lines · Christoph Hellwig

Switch the block device lookup interfaces to directly work with a dev_t so that struct block_device references are only acquired by the blkdev_get variants (and the blk-cgroup special case). This means that we not don't need an extra reference in the inode and can generally

     ^
     now
Quoted 4 lines

simplify handling of struct block_device to keep the lookups contained in the core block layer code.

...

Quoted 14 lines
fs/block_dev.c +3 -5
@@ -1689,14 +1599,12 @@ static int blkdev_open(struct inode * inode, struct file * filp)
 	if ((filp->f_flags & O_ACCMODE) == 3)
 		filp->f_mode |= FMODE_WRITE_IOCTL;
 
-	bdev = bd_acquire(inode);
-	if (bdev == NULL)
-		return -ENOMEM;
-
+	bdev = blkdev_get_by_dev(inode->i_rdev, filp->f_mode, filp);
+	if (IS_ERR(bdev))
+		return PTR_ERR(bdev);
 	filp->f_mapping = bdev->bd_inode->i_mapping;
 	filp->f_wb_err = filemap_sample_wb_err(filp->f_mapping);
-
Quoted 8 lines

fs/block_dev.c static int blkdev_open(struct inode * inode, struct file * filp) · line 1608

Quoted 3 lines
fs/block_dev.c +1 -1
-	return blkdev_get(bdev, filp->f_mode, filp);
+	return 0;
 }

I was wondering whether losing the stale bdev flushing in bd_acquire() would cause user-visible behavior changes but can't see how it would given that userland has no way of holding onto a specific instance of block inode. Maybe it's something worth mentioning in the commit message?

With stale bdev flushing do you mean the call to bd_forget if i_bdev exists but is unhashed? It doesn't actually flush anything but just detaches the old bdev from the inode so that the new one can be attached. That problem goes away by definition if we don't attach the bdev to the inode.

maintainer
#53 lore raw

Re: [PATCH 23/45] block: remove i_bdev

Hello,

Quoted 7 lines · Christoph Hellwig
Quoted 4 lines

I was wondering whether losing the stale bdev flushing in bd_acquire() would cause user-visible behavior changes but can't see how it would given that userland has no way of holding onto a specific instance of block inode. Maybe it's something worth mentioning in the commit message?

With stale bdev flushing do you mean the call to bd_forget if i_bdev exists but is unhashed? It doesn't actually flush anything but

Yeah.

Quoted 3 lines

just detaches the old bdev from the inode so that the new one can be attached. That problem goes away by definition if we don't attach the bdev to the inode.

Yeah, I think so. Was just wondering whether the problem actually goes away.

Thanks.

Signature · 1 line
tejun
author maintainer
#54 lore raw

PATCH 24/45 blk-cgroup: stop abusing get_gendisk

In reply to Christoph Hellwig

Properly open the device instead of relying on deep internals by using get_gendisk. Note that this uses FMODE_NDELAY without either FMODE_READ or FMODE_WRITE, which is a special open mode to allow for opening without media access, thus avoiding unexpexted interactions especially on removable media.

---
 block/blk-cgroup.c         | 42 +++++++++++++++++++-------------------
 block/blk-iocost.c         | 36 ++++++++++++++++----------------
 include/linux/blk-cgroup.h |  4 ++--
 3 files changed, 41 insertions(+), 41 deletions(-)
block/blk-cgroup.c +21 -21
index 54fbe1e80cc41a..23437b96ea41e6 100644
@@ -556,22 +556,22 @@ static struct blkcg_gq *blkg_lookup_check(struct blkcg *blkcg,
 }
 
 /**
- * blkg_conf_prep - parse and prepare for per-blkg config update
+ * blkcg_conf_open_bdev - parse and open bdev for per-blkg config update
  * @inputp: input string pointer
  *
  * Parse the device node prefix part, MAJ:MIN, of per-blkg config update
- * from @input and get and return the matching gendisk.  *@inputp is
+ * from @input and get and return the matching bdev.  *@inputp is
  * updated to point past the device node prefix.  Returns an ERR_PTR()
  * value on error.
  *
  * Use this function iff blkg_conf_prep() can't be used for some reason.
  */
-struct gendisk *blkcg_conf_get_disk(char **inputp)
+struct block_device *blkcg_conf_open_bdev(char **inputp)
 {
 	char *input = *inputp;
 	unsigned int major, minor;
-	struct gendisk *disk;
-	int key_len, part;
+	struct block_device *bdev;
+	int key_len;
 
 	if (sscanf(input, "%u:%u%n", &major, &minor, &key_len) != 2)
 		return ERR_PTR(-EINVAL);
@@ -581,16 +581,16 @@ struct gendisk *blkcg_conf_get_disk(char **inputp)
 		return ERR_PTR(-EINVAL);
 	input = skip_spaces(input);
 
-	disk = get_gendisk(MKDEV(major, minor), &part);
-	if (!disk)
+	bdev = blkdev_get_by_dev(MKDEV(major, minor), FMODE_NDELAY, NULL);
+	if (!bdev)
 		return ERR_PTR(-ENODEV);
-	if (part) {
-		put_disk_and_module(disk);
+	if (bdev_is_partition(bdev)) {
+		blkdev_put(bdev, FMODE_NDELAY);
 		return ERR_PTR(-ENODEV);
 	}
 
 	*inputp = input;
-	return disk;
+	return bdev;
 }
 
 /**
@@ -607,18 +607,18 @@ struct gendisk *blkcg_conf_get_disk(char **inputp)
  */
 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
 		   char *input, struct blkg_conf_ctx *ctx)
-	__acquires(rcu) __acquires(&disk->queue->queue_lock)
+	__acquires(rcu) __acquires(&bdev->bd_disk->queue->queue_lock)
 {
-	struct gendisk *disk;
+	struct block_device *bdev;
 	struct request_queue *q;
 	struct blkcg_gq *blkg;
 	int ret;
 
-	disk = blkcg_conf_get_disk(&input);
-	if (IS_ERR(disk))
-		return PTR_ERR(disk);
+	bdev = blkcg_conf_open_bdev(&input);
+	if (IS_ERR(bdev))
+		return PTR_ERR(bdev);
 
-	q = disk->queue;
+	q = bdev->bd_disk->queue;
 
 	rcu_read_lock();
 	spin_lock_irq(&q->queue_lock);
@@ -689,7 +689,7 @@ int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
 			goto success;
 	}
 success:
-	ctx->disk = disk;
+	ctx->bdev = bdev;
 	ctx->blkg = blkg;
 	ctx->body = input;
 	return 0;
@@ -700,7 +700,7 @@ int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
 	spin_unlock_irq(&q->queue_lock);
 	rcu_read_unlock();
 fail:
-	put_disk_and_module(disk);
+	blkdev_put(bdev, FMODE_NDELAY);
 	/*
 	 * If queue was bypassing, we should retry.  Do so after a
 	 * short msleep().  It isn't strictly necessary but queue
@@ -723,11 +723,11 @@ EXPORT_SYMBOL_GPL(blkg_conf_prep);
  * with blkg_conf_prep().
  */
 void blkg_conf_finish(struct blkg_conf_ctx *ctx)
-	__releases(&ctx->disk->queue->queue_lock) __releases(rcu)
+	__releases(&ctx->bdev->bd_disk->queue->queue_lock) __releases(rcu)
 {
-	spin_unlock_irq(&ctx->disk->queue->queue_lock);
+	spin_unlock_irq(&ctx->bdev->bd_disk->queue->queue_lock);
 	rcu_read_unlock();
-	put_disk_and_module(ctx->disk);
+	blkdev_put(ctx->bdev, FMODE_NDELAY);
 }
 EXPORT_SYMBOL_GPL(blkg_conf_finish);
 
block/blk-iocost.c +18 -18
index bbe86d1199dc5b..9f219718e9813c 100644
@@ -3120,23 +3120,23 @@ static const match_table_t qos_tokens = {
 static ssize_t ioc_qos_write(struct kernfs_open_file *of, char *input,
 			     size_t nbytes, loff_t off)
 {
-	struct gendisk *disk;
+	struct block_device *bdev;
 	struct ioc *ioc;
 	u32 qos[NR_QOS_PARAMS];
 	bool enable, user;
 	char *p;
 	int ret;
 
-	disk = blkcg_conf_get_disk(&input);
-	if (IS_ERR(disk))
-		return PTR_ERR(disk);
+	bdev = blkcg_conf_open_bdev(&input);
+	if (IS_ERR(bdev))
+		return PTR_ERR(bdev);
 
-	ioc = q_to_ioc(disk->queue);
+	ioc = q_to_ioc(bdev->bd_disk->queue);
 	if (!ioc) {
-		ret = blk_iocost_init(disk->queue);
+		ret = blk_iocost_init(bdev->bd_disk->queue);
 		if (ret)
 			goto err;
-		ioc = q_to_ioc(disk->queue);
+		ioc = q_to_ioc(bdev->bd_disk->queue);
 	}
 
 	spin_lock_irq(&ioc->lock);
@@ -3231,12 +3231,12 @@ static ssize_t ioc_qos_write(struct kernfs_open_file *of, char *input,
 	ioc_refresh_params(ioc, true);
 	spin_unlock_irq(&ioc->lock);
 
-	put_disk_and_module(disk);
+	blkdev_put(bdev, FMODE_NDELAY);
 	return nbytes;
 einval:
 	ret = -EINVAL;
 err:
-	put_disk_and_module(disk);
+	blkdev_put(bdev, FMODE_NDELAY);
 	return ret;
 }
 
@@ -3287,23 +3287,23 @@ static const match_table_t i_lcoef_tokens = {
 static ssize_t ioc_cost_model_write(struct kernfs_open_file *of, char *input,
 				    size_t nbytes, loff_t off)
 {
-	struct gendisk *disk;
+	struct block_device *bdev;
 	struct ioc *ioc;
 	u64 u[NR_I_LCOEFS];
 	bool user;
 	char *p;
 	int ret;
 
-	disk = blkcg_conf_get_disk(&input);
-	if (IS_ERR(disk))
-		return PTR_ERR(disk);
+	bdev = blkcg_conf_open_bdev(&input);
+	if (IS_ERR(bdev))
+		return PTR_ERR(bdev);
 
-	ioc = q_to_ioc(disk->queue);
+	ioc = q_to_ioc(bdev->bd_disk->queue);
 	if (!ioc) {
-		ret = blk_iocost_init(disk->queue);
+		ret = blk_iocost_init(bdev->bd_disk->queue);
 		if (ret)
 			goto err;
-		ioc = q_to_ioc(disk->queue);
+		ioc = q_to_ioc(bdev->bd_disk->queue);
 	}
 
 	spin_lock_irq(&ioc->lock);
@@ -3356,13 +3356,13 @@ static ssize_t ioc_cost_model_write(struct kernfs_open_file *of, char *input,
 	ioc_refresh_params(ioc, true);
 	spin_unlock_irq(&ioc->lock);
 
-	put_disk_and_module(disk);
+	blkdev_put(bdev, FMODE_NDELAY);
 	return nbytes;
 
 einval:
 	ret = -EINVAL;
 err:
-	put_disk_and_module(disk);
+	blkdev_put(bdev, FMODE_NDELAY);
 	return ret;
 }
 
include/linux/blk-cgroup.h +2 -2
index c8fc9792ac776d..b9f3c246c3c908 100644
@@ -197,12 +197,12 @@ void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
 
 struct blkg_conf_ctx {
-	struct gendisk			*disk;
+	struct block_device		*bdev;
 	struct blkcg_gq			*blkg;
 	char				*body;
 };
 
-struct gendisk *blkcg_conf_get_disk(char **inputp);
+struct block_device *blkcg_conf_open_bdev(char **inputp);
 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
 		   char *input, struct blkg_conf_ctx *ctx);
 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
Signature · 1 line
2.29.2
maintainer
#55 lore raw

Re: [PATCH 24/45] blk-cgroup: stop abusing get_gendisk

Quoted 5 lines · Christoph Hellwig

Properly open the device instead of relying on deep internals by using get_gendisk. Note that this uses FMODE_NDELAY without either FMODE_READ or FMODE_WRITE, which is a special open mode to allow for opening without media access, thus avoiding unexpexted interactions especially on removable media.

I'm not sure FMODE_NDELAY does that. e.g. sd_open() does media change check and full revalidation including disk spin up regadless of NDELAY and it's odd and can lead to nasty surprises to require cgroup configuration updates to wait for SCSI EH.

Thanks.

Signature · 1 line
tejun
author maintainer
#56 lore raw

PATCH 25/45 block: reference struct block_device from struct hd_struct

In reply to Christoph Hellwig

To simplify block device lookup and a few other upcoming areas, make sure that we always have a struct block_device available for each disk and each partition. The only downside of this is that each device and partition uses a little more memory. The upside will be that a lot of code can be simplified.

With that all we need to look up the block device is to lookup the inode and do a few sanity checks on the gendisk, instead of the separate lookup for the gendisk. These checks are in a new RCU critical section and the disk is now freed using kfree_rcu().

As part of the change switch bdget() to only find existing block devices, given that we know that the block_device structure must be allocated at probe / partition scan time.

---
 block/blk.h             |   2 +-
 block/genhd.c           | 212 ++++------------------------------------
 block/partitions/core.c |  29 +++---
 fs/block_dev.c          | 171 ++++++++++++++++++--------------
 include/linux/blkdev.h  |   2 +
 include/linux/genhd.h   |   6 +-
 6 files changed, 140 insertions(+), 282 deletions(-)
block/blk.h +1 -1
index dfab98465db9a5..c4839abcfa27eb 100644
@@ -352,7 +352,6 @@ struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector);
 
 int blk_alloc_devt(struct hd_struct *part, dev_t *devt);
 void blk_free_devt(dev_t devt);
-void blk_invalidate_devt(dev_t devt);
 char *disk_name(struct gendisk *hd, int partno, char *buf);
 #define ADDPART_FLAG_NONE	0
 #define ADDPART_FLAG_RAID	1
@@ -384,6 +383,7 @@ static inline void hd_free_part(struct hd_struct *part)
 {
 	free_percpu(part->dkstats);
 	kfree(part->info);
+	bdput(part->bdev);
 	percpu_ref_exit(&part->ref);
 }
 
block/genhd.c +21 -191

314 lines not on this page · raw message

block/partitions/core.c +17 -12
index a02e224115943d..508b46da53eee5 100644
@@ -340,12 +340,11 @@ void delete_partition(struct hd_struct *part)
 	device_del(part_to_dev(part));
 
 	/*
-	 * Remove gendisk pointer from idr so that it cannot be looked up
-	 * while RCU period before freeing gendisk is running to prevent
-	 * use-after-free issues. Note that the device number stays
-	 * "in-use" until we really free the gendisk.
+	 * Remove the block device from the inode hash, so that it cannot be
+	 * looked up while waiting for the RCU grace period.
 	 */
-	blk_invalidate_devt(part_devt(part));
+	remove_inode_hash(part->bdev->bd_inode);
+
 	percpu_ref_kill(&part->ref);
 }
 
@@ -368,6 +367,7 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 	dev_t devt = MKDEV(0, 0);
 	struct device *ddev = disk_to_dev(disk);
 	struct device *pdev;
+	struct block_device *bdev;
 	struct disk_part_tbl *ptbl;
 	const char *dname;
 	int err;
@@ -402,11 +402,15 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 	if (!p)
 		return ERR_PTR(-EBUSY);
 
+	err = -ENOMEM;
 	p->dkstats = alloc_percpu(struct disk_stats);
-	if (!p->dkstats) {
-		err = -ENOMEM;
+	if (!p->dkstats)
 		goto out_free;
-	}
+
+	bdev = bdev_alloc(disk, partno);
+	if (!bdev)
+		goto out_free_stats;
+	p->bdev = bdev;
 
 	hd_sects_seq_init(p);
 	pdev = part_to_dev(p);
@@ -420,10 +424,8 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 		struct partition_meta_info *pinfo;
 
 		pinfo = kzalloc_node(sizeof(*pinfo), GFP_KERNEL, disk->node_id);
-		if (!pinfo) {
-			err = -ENOMEM;
-			goto out_free_stats;
-		}
+		if (!pinfo)
+			goto out_bdput;
 		memcpy(pinfo, info, sizeof(*info));
 		p->info = pinfo;
 	}
@@ -470,6 +472,7 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 	}
 
 	/* everything is up and running, commence */
+	bdev_add(bdev, devt);
 	rcu_assign_pointer(ptbl->part[partno], p);
 
 	/* suppress uevent if the disk suppresses it */
@@ -479,6 +482,8 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 
 out_free_info:
 	kfree(p->info);
+out_bdput:
+	bdput(bdev);
 out_free_stats:
 	free_percpu(p->dkstats);
 out_free:
fs/block_dev.c +96 -75

274 lines not on this page · raw message

include/linux/blkdev.h +2 -0
index bdd7339bcda462..48e5a8393cd793 100644
@@ -1994,6 +1994,8 @@ void bd_abort_claiming(struct block_device *bdev, struct block_device *whole,
 		void *holder);
 void blkdev_put(struct block_device *bdev, fmode_t mode);
 
+struct block_device *bdev_alloc(struct gendisk *disk, u8 partno);
+void bdev_add(struct block_device *bdev, dev_t dev);
 struct block_device *I_BDEV(struct inode *inode);
 struct block_device *bdget_part(struct hd_struct *part);
 struct block_device *bdgrab(struct block_device *bdev);
include/linux/genhd.h +3 -3
index ca5e356084c353..d068e46f9086ae 100644
@@ -65,6 +65,7 @@ struct hd_struct {
 	struct disk_stats __percpu *dkstats;
 	struct percpu_ref ref;
 
+	struct block_device *bdev;
 	struct device __dev;
 	struct kobject *holder_dir;
 	int policy, partno;
@@ -193,7 +194,6 @@ struct gendisk {
 	int flags;
 	unsigned long state;
 #define GD_NEED_PART_SCAN		0
-	struct rw_semaphore lookup_sem;
 	struct kobject *slave_dir;
 
 	struct timer_rand_state *random;
@@ -207,6 +207,7 @@ struct gendisk {
 #endif
 	int node_id;
 	struct badblocks *bb;
+	struct rcu_head rcu;
 	struct lockdep_map lockdep_map;
 };
 
@@ -300,7 +301,6 @@ static inline void add_disk_no_queue_reg(struct gendisk *disk)
 }
 
 extern void del_gendisk(struct gendisk *gp);
-extern struct gendisk *get_gendisk(dev_t dev, int *partno);
 extern struct block_device *bdget_disk(struct gendisk *disk, int partno);
 
 extern void set_disk_ro(struct gendisk *disk, int flag);
@@ -338,7 +338,6 @@ int blk_drop_partitions(struct block_device *bdev);
 
 extern struct gendisk *__alloc_disk_node(int minors, int node_id);
 extern void put_disk(struct gendisk *disk);
-extern void put_disk_and_module(struct gendisk *disk);
 
 #define alloc_disk_node(minors, node_id)				\
 ({									\
@@ -389,6 +388,7 @@ static inline void bd_unlink_disk_holder(struct block_device *bdev,
 #endif /* CONFIG_SYSFS */
 
 dev_t blk_lookup_devt(const char *name, int partno);
+void blk_request_module(dev_t devt);
 #ifdef CONFIG_BLOCK
 void printk_all_partitions(void);
 #else /* CONFIG_BLOCK */
Signature · 1 line
2.29.2
maintainer
#57 lore raw

Re: [PATCH 25/45] block: reference struct block_device from struct hd_struct

Hello,

Please see lkml.kernel.org/r/[email protected] for a few nits on the previous version.

Quoted 10 lines · Christoph Hellwig

To simplify block device lookup and a few other upcoming areas, make sure that we always have a struct block_device available for each disk and each partition. The only downside of this is that each device and partition uses a little more memory. The upside will be that a lot of code can be simplified.

With that all we need to look up the block device is to lookup the inode and do a few sanity checks on the gendisk, instead of the separate lookup for the gendisk. These checks are in a new RCU critical section and the disk is now freed using kfree_rcu().

I might be confused but am wondering whether RCU is needed. It's currently used to ensure that gendisk is accessible in the blkdev_get path but wouldn't it be possible to simply pin gendisk from block_devices? The gendisk and hd_structs hold the base refs of the block_devices and in turn the block_devices pin the gendisk. When the gendisk gets deleted, it puts the base refs of the block devices but stays around while the block devices are being accessed.

Also, would it make sense to separate out lookup_sem removal? I *think* it's there to ensure that the same bdev doesn't get associated with old and new gendisks at the same time but can't wrap my head around how it works exactly. I can see that this may not be needed once the lifetimes of gendisk and block_devices are tied together but that may warrant a bit more explanation.

Thanks.

Signature · 1 line
tejun
author maintainer
#58 lore raw

Re: [PATCH 25/45] block: reference struct block_device from struct hd_struct

Quoted 4 lines · Tejun Heo

Hello,

Please see lkml.kernel.org/r/[email protected] for a few nits on the previous version.

Thanks, I've addressed the mapping_set_gfp mask nit and updated the commit log. As Jan pointed also pointed out I think we do need the remove_inode_hash.

Quoted 7 lines

I might be confused but am wondering whether RCU is needed. It's currently used to ensure that gendisk is accessible in the blkdev_get path but wouldn't it be possible to simply pin gendisk from block_devices? The gendisk and hd_structs hold the base refs of the block_devices and in turn the block_devices pin the gendisk. When the gendisk gets deleted, it puts the base refs of the block devices but stays around while the block devices are being accessed.

Yes, that sounds sensible. I'll take a look.

Quoted 6 lines

Also, would it make sense to separate out lookup_sem removal? I *think* it's there to ensure that the same bdev doesn't get associated with old and new gendisks at the same time but can't wrap my head around how it works exactly. I can see that this may not be needed once the lifetimes of gendisk and block_devices are tied together but that may warrant a bit more explanation.

Jan added lookup_sem in commit 56c0908c855afbb to prevent a three way race between del_gendisk and blkdev_open due to the weird bdev vs gendisk lifetime rules. None of that can happen with the new lookup scheme.

maintainer
#59 lore raw

Re: [PATCH 25/45] block: reference struct block_device from struct hd_struct

Hello,

Quoted 9 lines · Christoph Hellwig
Quoted 4 lines · Tejun Heo

Hello,

Please see lkml.kernel.org/r/[email protected] for a few nits on the previous version.

Thanks, I've addressed the mapping_set_gfp mask nit and updated the commit log. As Jan pointed also pointed out I think we do need the remove_inode_hash.

Agreed. It'd be nice to replace the stale comment.

Quoted 11 lines
Quoted 6 lines

Also, would it make sense to separate out lookup_sem removal? I *think* it's there to ensure that the same bdev doesn't get associated with old and new gendisks at the same time but can't wrap my head around how it works exactly. I can see that this may not be needed once the lifetimes of gendisk and block_devices are tied together but that may warrant a bit more explanation.

Jan added lookup_sem in commit 56c0908c855afbb to prevent a three way race between del_gendisk and blkdev_open due to the weird bdev vs gendisk lifetime rules. None of that can happen with the new lookup scheme.

Understood. I think it'd be worthwhile to note that in the commit log.

Thanks.

Signature · 1 line
tejun
author maintainer
#60 lore raw

Re: [PATCH 25/45] block: reference struct block_device from struct hd_struct

Quoted 1 line · Tejun Heo

Agreed. It'd be nice to replace the stale comment.

I've updated the comment.

Quoted 6 lines
Quoted 4 lines

Jan added lookup_sem in commit 56c0908c855afbb to prevent a three way race between del_gendisk and blkdev_open due to the weird bdev vs gendisk lifetime rules. None of that can happen with the new lookup scheme.

Understood. I think it'd be worthwhile to note that in the commit log.

So after the idea of just holding a disk reference from each bdev did not work out very well I've reworked this a bit, and lookup_sem stays, but becomes global to protect the disk lookup (similar to the previous global bdev_map_lock).

author maintainer
#61 lore raw

PATCH 26/45 block: remove ->bd_contains

In reply to Christoph Hellwig

Now that each hd_struct has a reference to the corresponding block_device, there is no need for the bd_contains pointer. Add a bdev_whole() helper to look up the whole device block_device struture instead.

---
 drivers/block/loop.c      |  2 +-
 drivers/scsi/scsicam.c    |  2 +-
 fs/block_dev.c            | 22 ++++++++--------------
 include/linux/blk_types.h |  4 +++-
 4 files changed, 13 insertions(+), 17 deletions(-)
drivers/block/loop.c +1 -1
index 26c7aafba7c5f8..c0df88b3300c41 100644
@@ -1088,7 +1088,7 @@ static int loop_configure(struct loop_device *lo, fmode_t mode,
 	 * here to avoid changing device under exclusive owner.
 	 */
 	if (!(mode & FMODE_EXCL)) {
-		claimed_bdev = bdev->bd_contains;
+		claimed_bdev = bdev_whole(bdev);
 		error = bd_prepare_to_claim(bdev, claimed_bdev, loop_configure);
 		if (error)
 			goto out_putf;
drivers/scsi/scsicam.c +1 -1
index 682cf08ab04153..f1553a453616fd 100644
@@ -32,7 +32,7 @@
  */
 unsigned char *scsi_bios_ptable(struct block_device *dev)
 {
-	struct address_space *mapping = dev->bd_contains->bd_inode->i_mapping;
+	struct address_space *mapping = bdev_whole(dev)->bd_inode->i_mapping;
 	unsigned char *res = NULL;
 	struct page *page;
 
fs/block_dev.c +8 -14
index b9ee8fe5acd570..e8d7de5fae00a9 100644
@@ -119,7 +119,7 @@ int truncate_bdev_range(struct block_device *bdev, fmode_t mode,
 	 * under live filesystem.
 	 */
 	if (!(mode & FMODE_EXCL)) {
-		claimed_bdev = bdev->bd_contains;
+		claimed_bdev = bdev_whole(bdev);
 		err = bd_prepare_to_claim(bdev, claimed_bdev,
 					  truncate_bdev_range);
 		if (err)
@@ -879,7 +879,6 @@ struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
 	spin_lock_init(&bdev->bd_size_lock);
 	bdev->bd_disk = disk;
 	bdev->bd_partno = partno;
-	bdev->bd_contains = NULL;
 	bdev->bd_super = NULL;
 	bdev->bd_inode = inode;
 	bdev->bd_part_count = 0;
@@ -1342,9 +1341,7 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode)
 	int ret;
 
 	if (!bdev->bd_openers) {
-		bdev->bd_contains = bdev;
-
-		if (!bdev->bd_partno) {
+		if (!bdev_is_partition(bdev)) {
 			ret = -ENXIO;
 			bdev->bd_part = disk_get_part(disk, 0);
 			if (!bdev->bd_part)
@@ -1384,7 +1381,6 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode)
 			whole->bd_part_count++;
 			mutex_unlock(&whole->bd_mutex);
 
-			bdev->bd_contains = whole;
 			bdev->bd_part = disk_get_part(disk, bdev->bd_partno);
 			if (!bdev->bd_part || !bdev->bd_part->nr_sects) {
 				__blkdev_put(whole, mode, 1);
@@ -1398,7 +1394,7 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode)
 		if (bdev->bd_bdi == &noop_backing_dev_info)
 			bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
 	} else {
-		if (bdev->bd_contains == bdev) {
+		if (!bdev_is_partition(bdev)) {
 			ret = 0;
 			if (bdev->bd_disk->fops->open)
 				ret = bdev->bd_disk->fops->open(bdev, mode);
@@ -1416,7 +1412,6 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode)
  out_clear:
 	disk_put_part(bdev->bd_part);
 	bdev->bd_part = NULL;
-	bdev->bd_contains = NULL;
 	return ret;
 }
 
@@ -1659,8 +1654,7 @@ static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
 		disk_put_part(bdev->bd_part);
 		bdev->bd_part = NULL;
 		if (bdev_is_partition(bdev))
-			victim = bdev->bd_contains;
-		bdev->bd_contains = NULL;
+			victim = bdev_whole(bdev);
 	} else {
 		if (!bdev_is_partition(bdev) && disk->fops->release)
 			disk->fops->release(disk, mode);
@@ -1678,6 +1672,7 @@ void blkdev_put(struct block_device *bdev, fmode_t mode)
 	mutex_lock(&bdev->bd_mutex);
 
 	if (mode & FMODE_EXCL) {
+		struct block_device *whole = bdev_whole(bdev);
 		bool bdev_free;
 
 		/*
@@ -1688,13 +1683,12 @@ void blkdev_put(struct block_device *bdev, fmode_t mode)
 		spin_lock(&bdev_lock);
 
 		WARN_ON_ONCE(--bdev->bd_holders < 0);
-		WARN_ON_ONCE(--bdev->bd_contains->bd_holders < 0);
+		WARN_ON_ONCE(--whole->bd_holders < 0);
 
-		/* bd_contains might point to self, check in a separate step */
 		if ((bdev_free = !bdev->bd_holders))
 			bdev->bd_holder = NULL;
-		if (!bdev->bd_contains->bd_holders)
-			bdev->bd_contains->bd_holder = NULL;
+		if (!whole->bd_holders)
+			whole->bd_holder = NULL;
 
 		spin_unlock(&bdev_lock);
 
include/linux/blk_types.h +3 -1
index 9698f459cc65c9..2e0a9bd9688d28 100644
@@ -32,7 +32,6 @@ struct block_device {
 #ifdef CONFIG_SYSFS
 	struct list_head	bd_holder_disks;
 #endif
-	struct block_device *	bd_contains;
 	u8			bd_partno;
 	struct hd_struct *	bd_part;
 	/* number of times partitions within this device have been opened. */
@@ -49,6 +48,9 @@ struct block_device {
 	struct super_block	*bd_fsfreeze_sb;
 } __randomize_layout;
 
+#define bdev_whole(_bdev) \
+	((_bdev)->bd_disk->part0.bdev)
+
 #define bdev_kobj(_bdev) \
 	(&part_to_dev((_bdev)->bd_part)->kobj)
 
Signature · 1 line
2.29.2
maintainer Acked-by on v1 26/45
#62 lore raw

Re: [PATCH 26/45] block: remove ->bd_contains

Quoted 8 lines · Christoph Hellwig

Now that each hd_struct has a reference to the corresponding block_device, there is no need for the bd_contains pointer. Add a bdev_whole() helper to look up the whole device block_device struture instead.

Signature · 1 line
tejun
author maintainer
#63 lore raw

PATCH 27/45 block: simplify the block device claiming interface

In reply to Christoph Hellwig

Stop passing the whole device as a separate argument given that it can be trivially deducted and cleanup the !holder debug check.

---
 drivers/block/loop.c   | 12 +++++-----
 fs/block_dev.c         | 51 +++++++++++++++---------------------------
 include/linux/blkdev.h |  6 ++---
 3 files changed, 25 insertions(+), 44 deletions(-)
drivers/block/loop.c +5 -7
index c0df88b3300c41..d643c67be6acea 100644
@@ -1069,7 +1069,6 @@ static int loop_configure(struct loop_device *lo, fmode_t mode,
 	struct file	*file;
 	struct inode	*inode;
 	struct address_space *mapping;
-	struct block_device *claimed_bdev = NULL;
 	int		error;
 	loff_t		size;
 	bool		partscan;
@@ -1088,8 +1087,7 @@ static int loop_configure(struct loop_device *lo, fmode_t mode,
 	 * here to avoid changing device under exclusive owner.
 	 */
 	if (!(mode & FMODE_EXCL)) {
-		claimed_bdev = bdev_whole(bdev);
-		error = bd_prepare_to_claim(bdev, claimed_bdev, loop_configure);
+		error = bd_prepare_to_claim(bdev, loop_configure);
 		if (error)
 			goto out_putf;
 	}
@@ -1176,15 +1174,15 @@ static int loop_configure(struct loop_device *lo, fmode_t mode,
 	mutex_unlock(&loop_ctl_mutex);
 	if (partscan)
 		loop_reread_partitions(lo, bdev);
-	if (claimed_bdev)
-		bd_abort_claiming(bdev, claimed_bdev, loop_configure);
+	if (!(mode & FMODE_EXCL))
+		bd_abort_claiming(bdev, loop_configure);
 	return 0;
 
 out_unlock:
 	mutex_unlock(&loop_ctl_mutex);
 out_bdev:
-	if (claimed_bdev)
-		bd_abort_claiming(bdev, claimed_bdev, loop_configure);
+	if (!(mode & FMODE_EXCL))
+		bd_abort_claiming(bdev, loop_configure);
 out_putf:
 	fput(file);
 out:
fs/block_dev.c +18 -33

131 lines not on this page · raw message

include/linux/blkdev.h +2 -4
index 48e5a8393cd793..db1b11d6d07568 100644
@@ -1988,10 +1988,8 @@ void blkdev_show(struct seq_file *seqf, off_t offset);
 struct block_device *blkdev_get_by_path(const char *path, fmode_t mode,
 		void *holder);
 struct block_device *blkdev_get_by_dev(dev_t dev, fmode_t mode, void *holder);
-int bd_prepare_to_claim(struct block_device *bdev, struct block_device *whole,
-		void *holder);
-void bd_abort_claiming(struct block_device *bdev, struct block_device *whole,
-		void *holder);
+int bd_prepare_to_claim(struct block_device *bdev, void *holder);
+void bd_abort_claiming(struct block_device *bdev, void *holder);
 void blkdev_put(struct block_device *bdev, fmode_t mode);
 
 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno);
Signature · 1 line
2.29.2
maintainer Acked-by on v1 27/45
#64 lore raw

Re: [PATCH 27/45] block: simplify the block device claiming interface

Quoted 6 lines · Christoph Hellwig

Stop passing the whole device as a separate argument given that it can be trivially deducted and cleanup the !holder debug check.

Signature · 1 line
tejun
author maintainer
#65 lore raw

PATCH 28/45 block: simplify part_to_disk

In reply to Christoph Hellwig

Now that struct hd_struct has a block_device pointer use that to find the disk.

---
 include/linux/genhd.h | 10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)
include/linux/genhd.h +3 -7
index d068e46f9086ae..dcf86a3d4dedc4 100644
@@ -219,13 +219,9 @@ struct gendisk {
 
 static inline struct gendisk *part_to_disk(struct hd_struct *part)
 {
-	if (likely(part)) {
-		if (part->partno)
-			return dev_to_disk(part_to_dev(part)->parent);
-		else
-			return dev_to_disk(part_to_dev(part));
-	}
-	return NULL;
+	if (unlikely(!part))
+		return NULL;
+	return part->bdev->bd_disk;
 }
 
 static inline int disk_max_parts(struct gendisk *disk)
Signature · 1 line
2.29.2
author maintainer
#67 lore raw

PATCH 29/45 block: initialize struct block_device in bdev_alloc

In reply to Christoph Hellwig

Don't play tricks with slab constructors as bdev structures tends to not get reused very much, and this makes the code a lot less error prone.

---
 fs/block_dev.c | 22 +++++++++-------------
 1 file changed, 9 insertions(+), 13 deletions(-)
fs/block_dev.c +9 -13
index 43a0fda982c879..1e5c6d0eb92677 100644
@@ -784,20 +784,11 @@ static void bdev_free_inode(struct inode *inode)
 	kmem_cache_free(bdev_cachep, BDEV_I(inode));
 }
 
-static void init_once(void *foo)
+static void init_once(void *data)
 {
-	struct bdev_inode *ei = (struct bdev_inode *) foo;
-	struct block_device *bdev = &ei->bdev;
+	struct bdev_inode *ei = data;
 
-	memset(bdev, 0, sizeof(*bdev));
-	mutex_init(&bdev->bd_mutex);
-#ifdef CONFIG_SYSFS
-	INIT_LIST_HEAD(&bdev->bd_holder_disks);
-#endif
-	bdev->bd_bdi = &noop_backing_dev_info;
 	inode_init_once(&ei->vfs_inode);
-	/* Initialize mutex for freeze. */
-	mutex_init(&bdev->bd_fsfreeze_mutex);
 }
 
 static void bdev_evict_inode(struct inode *inode)
@@ -872,12 +863,17 @@ struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
 	inode->i_data.a_ops = &def_blk_aops;
 
 	bdev = I_BDEV(inode);
+	memset(bdev, 0, sizeof(*bdev));
+	mutex_init(&bdev->bd_mutex);
+	mutex_init(&bdev->bd_fsfreeze_mutex);
 	spin_lock_init(&bdev->bd_size_lock);
 	bdev->bd_disk = disk;
 	bdev->bd_partno = partno;
-	bdev->bd_super = NULL;
 	bdev->bd_inode = inode;
-	bdev->bd_part_count = 0;
+	bdev->bd_bdi = &noop_backing_dev_info;
+#ifdef CONFIG_SYSFS
+	INIT_LIST_HEAD(&bdev->bd_holder_disks);
+#endif
 	return bdev;
 }
 
Signature · 1 line
2.29.2
maintainer Acked-by on v1 29/45
#68 lore raw

Re: [PATCH 29/45] block: initialize struct block_device in bdev_alloc

Quoted 4 lines · Christoph Hellwig

Don't play tricks with slab constructors as bdev structures tends to not get reused very much, and this makes the code a lot less error prone.

Signature · 1 line
tejun
author maintainer
#69 lore raw

PATCH 30/45 block: remove the nr_sects field in struct hd_struct

In reply to Christoph Hellwig

Now that the hd_struct always has a block device attached to it, there is no need for having two size field that just get out of sync.

Additional the field in hd_struct did not use proper serializiation, possibly allowing for torn writes. By only using the block_device field this problem also gets fixed.

---
 block/bio.c                        |  4 +-
 block/blk-core.c                   |  2 +-
 block/blk.h                        | 53 ----------------------
 block/genhd.c                      | 55 +++++++++++-----------
 block/partitions/core.c            | 17 ++++---
 drivers/block/loop.c               |  1 -
 drivers/block/nbd.c                |  2 +-
 drivers/block/xen-blkback/common.h |  4 +-
 drivers/md/bcache/super.c          |  2 +-
 drivers/s390/block/dasd_ioctl.c    |  4 +-
 drivers/target/target_core_pscsi.c |  7 +--
 fs/block_dev.c                     | 73 +-----------------------------
 fs/f2fs/super.c                    |  2 +-
 fs/pstore/blk.c                    |  2 +-
 include/linux/genhd.h              | 29 +++---------
 kernel/trace/blktrace.c            |  2 +-
 16 files changed, 60 insertions(+), 199 deletions(-)
block/bio.c +2 -2
index fa01bef35bb1fe..669bb47a31988e 100644
@@ -613,8 +613,8 @@ void guard_bio_eod(struct bio *bio)
 	rcu_read_lock();
 	part = __disk_get_part(bio->bi_disk, bio->bi_partno);
 	if (part)
-		maxsector = part_nr_sects_read(part);
-	else
+		maxsector = bdev_nr_sectors(part->bdev);
+	else	
 		maxsector = get_capacity(bio->bi_disk);
 	rcu_read_unlock();
 
block/blk-core.c +1 -1
index 2db8bda43b6e6d..988f45094a387b 100644
@@ -755,7 +755,7 @@ static inline int blk_partition_remap(struct bio *bio)
 		goto out;
 
 	if (bio_sectors(bio)) {
-		if (bio_check_eod(bio, part_nr_sects_read(p)))
+		if (bio_check_eod(bio, bdev_nr_sectors(p->bdev)))
 			goto out;
 		bio->bi_iter.bi_sector += p->start_sect;
 		trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
block/blk.h +0 -53

59 lines not on this page · raw message

block/genhd.c +29 -26

106 lines not on this page · raw message

block/partitions/core.c +11 -6
index 508b46da53eee5..92ffa55bdfddfd 100644
@@ -85,6 +85,13 @@ static int (*check_part[])(struct parsed_partitions *) = {
 	NULL
 };
 
+static void bdev_set_nr_sectors(struct block_device *bdev, sector_t sectors)
+{
+	spin_lock(&bdev->bd_size_lock);
+	i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
+	spin_unlock(&bdev->bd_size_lock);
+}
+
 static struct parsed_partitions *allocate_partitions(struct gendisk *hd)
 {
 	struct parsed_partitions *state;
@@ -295,7 +302,7 @@ static void hd_struct_free_work(struct work_struct *work)
 	put_device(disk_to_dev(disk));
 
 	part->start_sect = 0;
-	part->nr_sects = 0;
+	bdev_set_nr_sectors(part->bdev, 0);
 	part_stat_set_all(part, 0);
 	put_device(part_to_dev(part));
 }
@@ -412,11 +419,10 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 		goto out_free_stats;
 	p->bdev = bdev;
 
-	hd_sects_seq_init(p);
 	pdev = part_to_dev(p);
 
 	p->start_sect = start;
-	p->nr_sects = len;
+	bdev_set_nr_sectors(bdev, len);
 	p->partno = partno;
 	p->policy = get_disk_ro(disk);
 
@@ -509,7 +515,7 @@ static bool partition_overlaps(struct gendisk *disk, sector_t start,
 	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
 	while ((part = disk_part_iter_next(&piter))) {
 		if (part->partno == skip_partno ||
-		    start >= part->start_sect + part->nr_sects ||
+		    start >= part->start_sect + bdev_nr_sectors(part->bdev) ||
 		    start + length <= part->start_sect)
 			continue;
 		overlap = true;
@@ -600,8 +606,7 @@ int bdev_resize_partition(struct block_device *bdev, int partno,
 	if (partition_overlaps(bdev->bd_disk, start, length, partno))
 		goto out_unlock;
 
-	part_nr_sects_write(part, length);
-	bd_set_nr_sectors(bdevp, length);
+	bdev_set_nr_sectors(bdevp, length);
 
 	ret = 0;
 out_unlock:
drivers/block/loop.c +0 -1
index d643c67be6acea..d2ce1ddc192d78 100644
@@ -1241,7 +1241,6 @@ static int __loop_clr_fd(struct loop_device *lo, bool release)
 	set_capacity(lo->lo_disk, 0);
 	loop_sysfs_exit(lo);
 	if (bdev) {
-		bd_set_nr_sectors(bdev, 0);
 		/* let user-space know about this change */
 		kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, KOBJ_CHANGE);
 	}
drivers/block/nbd.c +1 -1
index 45b0423ef2c53d..014683968ce174 100644
@@ -1132,7 +1132,7 @@ static void nbd_bdev_reset(struct block_device *bdev)
 {
 	if (bdev->bd_openers > 1)
 		return;
-	bd_set_nr_sectors(bdev, 0);
+	set_capacity(bdev->bd_disk, 0);
 }
 
 static void nbd_parse_flags(struct nbd_device *nbd)
drivers/block/xen-blkback/common.h +1 -3
index c6ea5d38c509a6..0762db247b41b3 100644
@@ -358,9 +358,7 @@ struct pending_req {
 };
 
 
-#define vbd_sz(_v)	((_v)->bdev->bd_part ? \
-			 (_v)->bdev->bd_part->nr_sects : \
-			  get_capacity((_v)->bdev->bd_disk))
+#define vbd_sz(_v)	bdev_nr_sectors((_v)->bdev)
 
 #define xen_blkif_get(_b) (atomic_inc(&(_b)->refcnt))
 #define xen_blkif_put(_b)				\
drivers/md/bcache/super.c +1 -1
index c55d3c58a7ef55..04fa40868fbe10 100644
@@ -1408,7 +1408,7 @@ static int cached_dev_init(struct cached_dev *dc, unsigned int block_size)
 			q->limits.raid_partial_stripes_expensive;
 
 	ret = bcache_device_init(&dc->disk, block_size,
-			 dc->bdev->bd_part->nr_sects - dc->sb.data_offset,
+			 bdev_nr_sectors(dc->bdev) - dc->sb.data_offset,
 			 dc->bdev, &bcache_cached_ops);
 	if (ret)
 		return ret;
drivers/s390/block/dasd_ioctl.c +1 -3
index 3359559517bfcf..304eba1acf163c 100644
@@ -54,8 +54,6 @@ dasd_ioctl_enable(struct block_device *bdev)
 		return -ENODEV;
 
 	dasd_enable_device(base);
-	/* Formatting the dasd device can change the capacity. */
-	bd_set_nr_sectors(bdev, get_capacity(base->block->gdp));
 	dasd_put_device(base);
 	return 0;
 }
@@ -88,7 +86,7 @@ dasd_ioctl_disable(struct block_device *bdev)
 	 * Set i_size to zero, since read, write, etc. check against this
 	 * value.
 	 */
-	bd_set_nr_sectors(bdev, 0);
+	set_capacity(bdev->bd_disk, 0);
 	dasd_put_device(base);
 	return 0;
 }
drivers/target/target_core_pscsi.c +1 -6
index 4e37fa9b409d52..a70c33c49f0960 100644
@@ -1027,12 +1027,7 @@ static u32 pscsi_get_device_type(struct se_device *dev)
 
 static sector_t pscsi_get_blocks(struct se_device *dev)
 {
-	struct pscsi_dev_virt *pdv = PSCSI_DEV(dev);
-
-	if (pdv->pdv_bd && pdv->pdv_bd->bd_part)
-		return pdv->pdv_bd->bd_part->nr_sects;
-
-	return 0;
+	return bdev_nr_sectors(PSCSI_DEV(dev)->pdv_bd);
 }
 
 static void pscsi_req_done(struct request *req, blk_status_t status)
fs/block_dev.c +2 -71

102 lines not on this page · raw message

fs/f2fs/super.c +1 -1
index 00eff2f5180790..d4e7fab352bacb 100644
@@ -3151,7 +3151,7 @@ static int f2fs_report_zone_cb(struct blk_zone *zone, unsigned int idx,
 static int init_blkz_info(struct f2fs_sb_info *sbi, int devi)
 {
 	struct block_device *bdev = FDEV(devi).bdev;
-	sector_t nr_sectors = bdev->bd_part->nr_sects;
+	sector_t nr_sectors = bdev_nr_sectors(bdev);
 	struct f2fs_report_zones_args rep_zone_arg;
 	int ret;
 
fs/pstore/blk.c +1 -1
index fcd5563dde063c..777a26f7bbe2aa 100644
@@ -245,7 +245,7 @@ static struct block_device *psblk_get_bdev(void *holder,
 			return bdev;
 	}
 
-	nr_sects = part_nr_sects_read(bdev->bd_part);
+	nr_sects = bdev_nr_sectors(bdev);
 	if (!nr_sects) {
 		pr_err("not enough space for '%s'\n", blkdev);
 		blkdev_put(bdev, mode);
include/linux/genhd.h +7 -22
index dcf86a3d4dedc4..0dbd254bca51aa 100644
@@ -52,15 +52,6 @@ struct partition_meta_info {
 
 struct hd_struct {
 	sector_t start_sect;
-	/*
-	 * nr_sects is protected by sequence counter. One might extend a
-	 * partition while IO is happening to it and update of nr_sects
-	 * can be non-atomic on 32bit machines with 64bit sector_t.
-	 */
-	sector_t nr_sects;
-#if BITS_PER_LONG==32 && defined(CONFIG_SMP)
-	seqcount_t nr_sects_seq;
-#endif
 	unsigned long stamp;
 	struct disk_stats __percpu *dkstats;
 	struct percpu_ref ref;
@@ -255,13 +246,6 @@ static inline void disk_put_part(struct hd_struct *part)
 		put_device(part_to_dev(part));
 }
 
-static inline void hd_sects_seq_init(struct hd_struct *p)
-{
-#if BITS_PER_LONG==32 && defined(CONFIG_SMP)
-	seqcount_init(&p->nr_sects_seq);
-#endif
-}
-
 /*
  * Smarter partition iterator without context limits.
  */
@@ -319,13 +303,15 @@ static inline sector_t get_start_sect(struct block_device *bdev)
 {
 	return bdev->bd_part->start_sect;
 }
-static inline sector_t get_capacity(struct gendisk *disk)
+
+static inline sector_t bdev_nr_sectors(struct block_device *bdev)
 {
-	return disk->part0.nr_sects;
+	return i_size_read(bdev->bd_inode) >> 9;
 }
-static inline void set_capacity(struct gendisk *disk, sector_t size)
+
+static inline sector_t get_capacity(struct gendisk *disk)
 {
-	disk->part0.nr_sects = size;
+	return bdev_nr_sectors(disk->part0.bdev);
 }
 
 int bdev_disk_changed(struct block_device *bdev, bool invalidate);
@@ -359,10 +345,9 @@ int __register_blkdev(unsigned int major, const char *name,
 	__register_blkdev(major, name, NULL)
 void unregister_blkdev(unsigned int major, const char *name);
 
-void revalidate_disk_size(struct gendisk *disk, bool verbose);
 bool bdev_check_media_change(struct block_device *bdev);
 int __invalidate_device(struct block_device *bdev, bool kill_dirty);
-void bd_set_nr_sectors(struct block_device *bdev, sector_t sectors);
+void set_capacity(struct gendisk *disk, sector_t size);
 
 /* for drivers/char/raw.c: */
 int blkdev_ioctl(struct block_device *, fmode_t, unsigned, unsigned long);
kernel/trace/blktrace.c +1 -1
index f1022945e3460b..7076d588a50d69 100644
@@ -465,7 +465,7 @@ static void blk_trace_setup_lba(struct blk_trace *bt,
 
 	if (part) {
 		bt->start_lba = part->start_sect;
-		bt->end_lba = part->start_sect + part->nr_sects;
+		bt->end_lba = part->start_sect + bdev_nr_sectors(bdev);
 	} else {
 		bt->start_lba = 0;
 		bt->end_lba = -1ULL;
Signature · 1 line
2.29.2
#70 lore raw

Re: [PATCH 30/45] block: remove the nr_sects field in struct hd_struct

Quoted 9 lines · Christoph Hellwig

Now that the hd_struct always has a block device attached to it, there is no need for having two size field that just get out of sync.

Additional the field in hd_struct did not use proper serializiation, possibly allowing for torn writes. By only using the block_device field this problem also gets fixed.

For the bcache part, Acked-by: Coly Li <[email protected]>

Thanks.

Coly Li

Quoted 19 lines
---
 block/bio.c                        |  4 +-
 block/blk-core.c                   |  2 +-
 block/blk.h                        | 53 ----------------------
 block/genhd.c                      | 55 +++++++++++-----------
 block/partitions/core.c            | 17 ++++---
 drivers/block/loop.c               |  1 -
 drivers/block/nbd.c                |  2 +-
 drivers/block/xen-blkback/common.h |  4 +-
 drivers/md/bcache/super.c          |  2 +-
 drivers/s390/block/dasd_ioctl.c    |  4 +-
 drivers/target/target_core_pscsi.c |  7 +--
 fs/block_dev.c                     | 73 +-----------------------------
 fs/f2fs/super.c                    |  2 +-
 fs/pstore/blk.c                    |  2 +-
 include/linux/genhd.h              | 29 +++---------
 kernel/trace/blktrace.c            |  2 +-
 16 files changed, 60 insertions(+), 199 deletions(-)

[snipped]

Quoted 13 lines
drivers/md/bcache/super.c +1 -1
index c55d3c58a7ef55..04fa40868fbe10 100644
@@ -1408,7 +1408,7 @@ static int cached_dev_init(struct cached_dev *dc, unsigned int block_size)
 			q->limits.raid_partial_stripes_expensive;
 
 	ret = bcache_device_init(&dc->disk, block_size,
-			 dc->bdev->bd_part->nr_sects - dc->sb.data_offset,
+			 bdev_nr_sectors(dc->bdev) - dc->sb.data_offset,
 			 dc->bdev, &bcache_cached_ops);
 	if (ret)
 		return ret;

[snipped]

Acked-by on v1 30/45
#71 lore raw

Re: [PATCH 30/45] block: remove the nr_sects field in struct hd_struct

In reply to Christoph Hellwig
Quoted 9 lines · Christoph Hellwig

Now that the hd_struct always has a block device attached to it, there is no need for having two size field that just get out of sync.

Additional the field in hd_struct did not use proper serializiation, possibly allowing for torn writes. By only using the block_device field this problem also gets fixed.

Quoted 14 lines
---
  block/bio.c                        |  4 +-
  block/blk-core.c                   |  2 +-
  block/blk.h                        | 53 ----------------------
  block/genhd.c                      | 55 +++++++++++-----------
  block/partitions/core.c            | 17 ++++---
  drivers/block/loop.c               |  1 -
  drivers/block/nbd.c                |  2 +-
  drivers/block/xen-blkback/common.h |  4 +-
  drivers/md/bcache/super.c          |  2 +-
  drivers/s390/block/dasd_ioctl.c    |  4 +-
  drivers/target/target_core_pscsi.c |  7 +--
  fs/block_dev.c                     | 73 +-----------------------------
  fs/f2fs/super.c                    |  2 +-

For f2fs part,

Thanks,

Quoted 4 lines
  fs/pstore/blk.c                    |  2 +-
  include/linux/genhd.h              | 29 +++---------
  kernel/trace/blktrace.c            |  2 +-
  16 files changed, 60 insertions(+), 199 deletions(-)
author maintainer
#72 lore raw

PATCH 31/45 block: move disk stat accounting to struct block_device

In reply to Christoph Hellwig

Move the dkstats and stamp field to struct block_device in preparation of killing struct hd_struct.

---
 block/blk-cgroup.c        |  2 +-
 block/blk-core.c          |  4 ++--
 block/blk.h               |  1 -
 block/genhd.c             | 14 ++++----------
 block/partitions/core.c   |  9 +--------
 fs/block_dev.c            | 10 ++++++++++
 include/linux/blk_types.h |  2 ++
 include/linux/genhd.h     |  2 --
 include/linux/part_stat.h | 38 +++++++++++++++++++-------------------
 9 files changed, 39 insertions(+), 43 deletions(-)
block/blk-cgroup.c +1 -1
index 23437b96ea41e6..a598f86e014137 100644
@@ -830,7 +830,7 @@ static void blkcg_fill_root_iostats(void)
 		for_each_possible_cpu(cpu) {
 			struct disk_stats *cpu_dkstats;
 
-			cpu_dkstats = per_cpu_ptr(part->dkstats, cpu);
+			cpu_dkstats = per_cpu_ptr(part->bdev->bd_stats, cpu);
 			tmp.ios[BLKG_IOSTAT_READ] +=
 				cpu_dkstats->ios[STAT_READ];
 			tmp.ios[BLKG_IOSTAT_WRITE] +=
block/blk-core.c +2 -2
index 988f45094a387b..d2c9cb24e087f3 100644
@@ -1264,9 +1264,9 @@ static void update_io_ticks(struct hd_struct *part, unsigned long now, bool end)
 {
 	unsigned long stamp;
 again:
-	stamp = READ_ONCE(part->stamp);
+	stamp = READ_ONCE(part->bdev->bd_stamp);
 	if (unlikely(stamp != now)) {
-		if (likely(cmpxchg(&part->stamp, stamp, now) == stamp))
+		if (likely(cmpxchg(&part->bdev->bd_stamp, stamp, now) == stamp))
 			__part_stat_add(part, io_ticks, end ? now - stamp : 1);
 	}
 	if (part->partno) {
block/blk.h +0 -1
index 09cee7024fb43e..3f801f6e86f8a1 100644
@@ -381,7 +381,6 @@ static inline void hd_struct_put(struct hd_struct *part)
 
 static inline void hd_free_part(struct hd_struct *part)
 {
-	free_percpu(part->dkstats);
 	kfree(part->info);
 	bdput(part->bdev);
 	percpu_ref_exit(&part->ref);
block/genhd.c +4 -10
index 8ace0628ac20b7..0c0458367da7e4 100644
@@ -102,7 +102,7 @@ static void part_stat_read_all(struct hd_struct *part, struct disk_stats *stat)
 
 	memset(stat, 0, sizeof(struct disk_stats));
 	for_each_possible_cpu(cpu) {
-		struct disk_stats *ptr = per_cpu_ptr(part->dkstats, cpu);
+		struct disk_stats *ptr = per_cpu_ptr(part->bdev->bd_stats, cpu);
 		int group;
 
 		for (group = 0; group < NR_STAT_GROUPS; group++) {
@@ -875,7 +875,7 @@ void del_gendisk(struct gendisk *disk)
 	kobject_put(disk->slave_dir);
 
 	part_stat_set_all(&disk->part0, 0);
-	disk->part0.stamp = 0;
+	disk->part0.bdev->bd_stamp = 0;
 	if (!sysfs_deprecated)
 		sysfs_remove_link(block_depr, dev_name(disk_to_dev(disk)));
 	pm_runtime_set_memalloc_noio(disk_to_dev(disk), false);
@@ -1612,19 +1612,15 @@ struct gendisk *__alloc_disk_node(int minors, int node_id)
 	if (!disk->part0.bdev)
 		goto out_free_disk;
 
-	disk->part0.dkstats = alloc_percpu(struct disk_stats);
-	if (!disk->part0.dkstats)
-		goto out_bdput;
-
 	disk->node_id = node_id;
 	if (disk_expand_part_tbl(disk, 0))
-		goto out_free_bdstats;
+		goto out_bdput;
 
 	ptbl = rcu_dereference_protected(disk->part_tbl, 1);
 	rcu_assign_pointer(ptbl->part[0], &disk->part0);
 
 	if (hd_ref_init(&disk->part0))
-		goto out_free_bdstats;
+		goto out_bdput;
 
 	disk->minors = minors;
 	rand_initialize_disk(disk);
@@ -1633,8 +1629,6 @@ struct gendisk *__alloc_disk_node(int minors, int node_id)
 	device_initialize(disk_to_dev(disk));
 	return disk;
 
-out_free_bdstats:
-	free_percpu(disk->part0.dkstats);
 out_bdput:
 	bdput(disk->part0.bdev);
 out_free_disk:
block/partitions/core.c +1 -8
index 92ffa55bdfddfd..c3a4870bfb123d 100644
@@ -409,14 +409,9 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 	if (!p)
 		return ERR_PTR(-EBUSY);
 
-	err = -ENOMEM;
-	p->dkstats = alloc_percpu(struct disk_stats);
-	if (!p->dkstats)
-		goto out_free;
-
 	bdev = bdev_alloc(disk, partno);
 	if (!bdev)
-		goto out_free_stats;
+		goto out_free;
 	p->bdev = bdev;
 
 	pdev = part_to_dev(p);
@@ -490,8 +485,6 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 	kfree(p->info);
 out_bdput:
 	bdput(bdev);
-out_free_stats:
-	free_percpu(p->dkstats);
 out_free:
 	kfree(p);
 	return ERR_PTR(err);
fs/block_dev.c +10 -0
index 02536d9fa29945..0427e6fa59556f 100644
@@ -32,6 +32,7 @@
 #include <linux/cleancache.h>
 #include <linux/task_io_accounting_ops.h>
 #include <linux/falloc.h>
+#include <linux/part_stat.h>
 #include <linux/uaccess.h>
 #include <linux/suspend.h>
 #include "internal.h"
@@ -781,6 +782,10 @@ static struct inode *bdev_alloc_inode(struct super_block *sb)
 
 static void bdev_free_inode(struct inode *inode)
 {
+	struct block_device *bdev = I_BDEV(inode);
+
+	free_percpu(bdev->bd_stats);
+
 	kmem_cache_free(bdev_cachep, BDEV_I(inode));
 }
 
@@ -874,6 +879,11 @@ struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
 #ifdef CONFIG_SYSFS
 	INIT_LIST_HEAD(&bdev->bd_holder_disks);
 #endif
+	bdev->bd_stats = alloc_percpu(struct disk_stats);
+	if (!bdev->bd_stats) {
+		iput(inode);
+		return NULL;
+	}
 	return bdev;
 }
 
include/linux/blk_types.h +2 -0
index 2e0a9bd9688d28..520011b95276fb 100644
@@ -20,6 +20,8 @@ typedef void (bio_end_io_t) (struct bio *);
 struct bio_crypt_ctx;
 
 struct block_device {
+	struct disk_stats __percpu *bd_stats;
+	unsigned long		bd_stamp;
 	dev_t			bd_dev;
 	int			bd_openers;
 	struct inode *		bd_inode;	/* will die */
include/linux/genhd.h +0 -2
index 0dbd254bca51aa..dcc40c8217d095 100644
@@ -52,8 +52,6 @@ struct partition_meta_info {
 
 struct hd_struct {
 	sector_t start_sect;
-	unsigned long stamp;
-	struct disk_stats __percpu *dkstats;
 	struct percpu_ref ref;
 
 	struct block_device *bdev;
include/linux/part_stat.h +19 -19
index 24125778ef3ec7..87ad60106e1db0 100644
@@ -25,17 +25,17 @@ struct disk_stats {
 #define part_stat_unlock()	preempt_enable()
 
 #define part_stat_get_cpu(part, field, cpu)				\
-	(per_cpu_ptr((part)->dkstats, (cpu))->field)
+	(per_cpu_ptr((part)->bdev->bd_stats, (cpu))->field)
 
 #define part_stat_get(part, field)					\
 	part_stat_get_cpu(part, field, smp_processor_id())
 
 #define part_stat_read(part, field)					\
 ({									\
-	typeof((part)->dkstats->field) res = 0;				\
+	typeof((part)->bdev->bd_stats->field) res = 0;			\
 	unsigned int _cpu;						\
 	for_each_possible_cpu(_cpu)					\
-		res += per_cpu_ptr((part)->dkstats, _cpu)->field;	\
+		res += per_cpu_ptr((part)->bdev->bd_stats, _cpu)->field; \
 	res;								\
 })
 
@@ -44,7 +44,7 @@ static inline void part_stat_set_all(struct hd_struct *part, int value)
 	int i;
 
 	for_each_possible_cpu(i)
-		memset(per_cpu_ptr(part->dkstats, i), value,
+		memset(per_cpu_ptr(part->bdev->bd_stats, i), value,
 				sizeof(struct disk_stats));
 }
 
@@ -54,7 +54,7 @@ static inline void part_stat_set_all(struct hd_struct *part, int value)
 	 part_stat_read(part, field[STAT_DISCARD]))
 
 #define __part_stat_add(part, field, addnd)				\
-	__this_cpu_add((part)->dkstats->field, addnd)
+	__this_cpu_add((part)->bdev->bd_stats->field, addnd)
 
 #define part_stat_add(part, field, addnd)	do {			\
 	__part_stat_add((part), field, addnd);				\
@@ -63,20 +63,20 @@ static inline void part_stat_set_all(struct hd_struct *part, int value)
 				field, addnd);				\
 } while (0)
 
-#define part_stat_dec(gendiskp, field)					\
-	part_stat_add(gendiskp, field, -1)
-#define part_stat_inc(gendiskp, field)					\
-	part_stat_add(gendiskp, field, 1)
-#define part_stat_sub(gendiskp, field, subnd)				\
-	part_stat_add(gendiskp, field, -subnd)
+#define part_stat_dec(part, field)					\
+	part_stat_add(part, field, -1)
+#define part_stat_inc(part, field)					\
+	part_stat_add(part, field, 1)
+#define part_stat_sub(part, field, subnd)				\
+	part_stat_add(part, field, -subnd)
 
-#define part_stat_local_dec(gendiskp, field)				\
-	local_dec(&(part_stat_get(gendiskp, field)))
-#define part_stat_local_inc(gendiskp, field)				\
-	local_inc(&(part_stat_get(gendiskp, field)))
-#define part_stat_local_read(gendiskp, field)				\
-	local_read(&(part_stat_get(gendiskp, field)))
-#define part_stat_local_read_cpu(gendiskp, field, cpu)			\
-	local_read(&(part_stat_get_cpu(gendiskp, field, cpu)))
+#define part_stat_local_dec(part, field)				\
+	local_dec(&(part_stat_get(part, field)))
+#define part_stat_local_inc(part, field)				\
+	local_inc(&(part_stat_get(part, field)))
+#define part_stat_local_read(part, field)				\
+	local_read(&(part_stat_get(part, field)))
+#define part_stat_local_read_cpu(part, field, cpu)			\
+	local_read(&(part_stat_get_cpu(part, field, cpu)))
 
 #endif /* _LINUX_PART_STAT_H */
Signature · 1 line
2.29.2
author maintainer
#73 lore raw

PATCH 32/45 block: move the start_sect field to struct block_device

In reply to Christoph Hellwig

Move the start_sect field to struct block_device in preparation of killing struct hd_struct.

---
 block/blk-core.c          |  5 +++--
 block/blk-lib.c           |  2 +-
 block/genhd.c             |  4 ++--
 block/partitions/core.c   | 17 +++++++++--------
 include/linux/blk_types.h |  1 +
 include/linux/blkdev.h    |  4 ++--
 include/linux/genhd.h     |  3 +--
 kernel/trace/blktrace.c   | 11 +++--------
 8 files changed, 22 insertions(+), 25 deletions(-)
block/blk-core.c +3 -2
index d2c9cb24e087f3..9a3793d5ce38d4 100644
@@ -757,9 +757,10 @@ static inline int blk_partition_remap(struct bio *bio)
 	if (bio_sectors(bio)) {
 		if (bio_check_eod(bio, bdev_nr_sectors(p->bdev)))
 			goto out;
-		bio->bi_iter.bi_sector += p->start_sect;
+		bio->bi_iter.bi_sector += p->bdev->bd_start_sect;
 		trace_block_bio_remap(bio->bi_disk->queue, bio, part_devt(p),
-				      bio->bi_iter.bi_sector - p->start_sect);
+				      bio->bi_iter.bi_sector -
+				      p->bdev->bd_start_sect);
 	}
 	bio->bi_partno = 0;
 	ret = 0;
block/blk-lib.c +1 -1
index e90614fd8d6a42..752f9c7220622a 100644
@@ -65,7 +65,7 @@ int __blkdev_issue_discard(struct block_device *bdev, sector_t sector,
 
 	/* In case the discard request is in a partition */
 	if (bdev_is_partition(bdev))
-		part_offset = bdev->bd_part->start_sect;
+		part_offset = bdev->bd_start_sect;
 
 	while (nr_sects) {
 		sector_t granularity_aligned_lba, req_sects;
block/genhd.c +2 -2
index 0c0458367da7e4..8212a2dd10ec4e 100644
@@ -295,8 +295,8 @@ EXPORT_SYMBOL_GPL(disk_part_iter_exit);
 
 static inline int sector_in_part(struct hd_struct *part, sector_t sector)
 {
-	return part->start_sect <= sector &&
-		sector < part->start_sect + bdev_nr_sectors(part->bdev);
+	return part->bdev->bd_start_sect <= sector &&
+		sector < part->bdev->bd_start_sect + bdev_nr_sectors(part->bdev);
 }
 
 /**
block/partitions/core.c +9 -8
index c3a4870bfb123d..aa4b836374b037 100644
@@ -192,7 +192,7 @@ static ssize_t part_start_show(struct device *dev,
 {
 	struct hd_struct *p = dev_to_part(dev);
 
-	return sprintf(buf, "%llu\n",(unsigned long long)p->start_sect);
+	return sprintf(buf, "%llu\n",(unsigned long long)p->bdev->bd_start_sect);
 }
 
 static ssize_t part_ro_show(struct device *dev,
@@ -209,7 +209,7 @@ static ssize_t part_alignment_offset_show(struct device *dev,
 
 	return sprintf(buf, "%u\n",
 		queue_limit_alignment_offset(&part_to_disk(p)->queue->limits,
-				p->start_sect));
+				p->bdev->bd_start_sect));
 }
 
 static ssize_t part_discard_alignment_show(struct device *dev,
@@ -219,7 +219,7 @@ static ssize_t part_discard_alignment_show(struct device *dev,
 
 	return sprintf(buf, "%u\n",
 		queue_limit_discard_alignment(&part_to_disk(p)->queue->limits,
-				p->start_sect));
+				p->bdev->bd_start_sect));
 }
 
 static DEVICE_ATTR(partition, 0444, part_partition_show, NULL);
@@ -301,7 +301,7 @@ static void hd_struct_free_work(struct work_struct *work)
 	 */
 	put_device(disk_to_dev(disk));
 
-	part->start_sect = 0;
+	part->bdev->bd_start_sect = 0;
 	bdev_set_nr_sectors(part->bdev, 0);
 	part_stat_set_all(part, 0);
 	put_device(part_to_dev(part));
@@ -416,7 +416,7 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 
 	pdev = part_to_dev(p);
 
-	p->start_sect = start;
+	bdev->bd_start_sect = start;
 	bdev_set_nr_sectors(bdev, len);
 	p->partno = partno;
 	p->policy = get_disk_ro(disk);
@@ -508,8 +508,9 @@ static bool partition_overlaps(struct gendisk *disk, sector_t start,
 	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
 	while ((part = disk_part_iter_next(&piter))) {
 		if (part->partno == skip_partno ||
-		    start >= part->start_sect + bdev_nr_sectors(part->bdev) ||
-		    start + length <= part->start_sect)
+		    start >= part->bdev->bd_start_sect +
+			bdev_nr_sectors(part->bdev) ||
+		    start + length <= part->bdev->bd_start_sect)
 			continue;
 		overlap = true;
 		break;
@@ -592,7 +593,7 @@ int bdev_resize_partition(struct block_device *bdev, int partno,
 	mutex_lock_nested(&bdev->bd_mutex, 1);
 
 	ret = -EINVAL;
-	if (start != part->start_sect)
+	if (start != part->bdev->bd_start_sect)
 		goto out_unlock;
 
 	ret = -EBUSY;
include/linux/blk_types.h +1 -0
index 520011b95276fb..a690008f60cd92 100644
@@ -20,6 +20,7 @@ typedef void (bio_end_io_t) (struct bio *);
 struct bio_crypt_ctx;
 
 struct block_device {
+	sector_t		bd_start_sect;
 	struct disk_stats __percpu *bd_stats;
 	unsigned long		bd_stamp;
 	dev_t			bd_dev;
include/linux/blkdev.h +2 -2
index db1b11d6d07568..8fc0b266610f7f 100644
@@ -1488,7 +1488,7 @@ static inline int bdev_alignment_offset(struct block_device *bdev)
 		return -1;
 	if (bdev_is_partition(bdev))
 		return queue_limit_alignment_offset(&q->limits,
-				bdev->bd_part->start_sect);
+				bdev->bd_start_sect);
 	return q->limits.alignment_offset;
 }
 
@@ -1529,7 +1529,7 @@ static inline int bdev_discard_alignment(struct block_device *bdev)
 
 	if (bdev_is_partition(bdev))
 		return queue_limit_discard_alignment(&q->limits,
-				bdev->bd_part->start_sect);
+				bdev->bd_start_sect);
 	return q->limits.discard_alignment;
 }
 
include/linux/genhd.h +1 -2
index dcc40c8217d095..a9d64da474233f 100644
@@ -51,7 +51,6 @@ struct partition_meta_info {
 };
 
 struct hd_struct {
-	sector_t start_sect;
 	struct percpu_ref ref;
 
 	struct block_device *bdev;
@@ -299,7 +298,7 @@ extern void rand_initialize_disk(struct gendisk *disk);
 
 static inline sector_t get_start_sect(struct block_device *bdev)
 {
-	return bdev->bd_part->start_sect;
+	return bdev->bd_start_sect;
 }
 
 static inline sector_t bdev_nr_sectors(struct block_device *bdev)
kernel/trace/blktrace.c +3 -8
index 7076d588a50d69..8a723a91ec5a06 100644
@@ -458,14 +458,9 @@ static struct rchan_callbacks blk_relay_callbacks = {
 static void blk_trace_setup_lba(struct blk_trace *bt,
 				struct block_device *bdev)
 {
-	struct hd_struct *part = NULL;
-
-	if (bdev)
-		part = bdev->bd_part;
-
-	if (part) {
-		bt->start_lba = part->start_sect;
-		bt->end_lba = part->start_sect + bdev_nr_sectors(bdev);
+	if (bdev) {
+		bt->start_lba = bdev->bd_start_sect;
+		bt->end_lba = bdev->bd_start_sect + bdev_nr_sectors(bdev);
 	} else {
 		bt->start_lba = 0;
 		bt->end_lba = -1ULL;
Signature · 1 line
2.29.2
author maintainer
#74 lore raw

PATCH 33/45 block: move the partition_meta_info to struct block_device

In reply to Christoph Hellwig

Move the partition_meta_info to struct block_device in preparation for killing struct hd_struct.

---
 block/blk.h               |  1 -
 block/genhd.c             |  3 ++-
 block/partitions/core.c   | 18 +++++++-----------
 fs/block_dev.c            |  1 +
 include/linux/blk_types.h |  2 ++
 include/linux/genhd.h     |  1 -
 init/do_mounts.c          |  7 ++++---
 7 files changed, 16 insertions(+), 17 deletions(-)
block/blk.h +0 -1
index 3f801f6e86f8a1..0bd4b58bcbaf77 100644
@@ -381,7 +381,6 @@ static inline void hd_struct_put(struct hd_struct *part)
 
 static inline void hd_free_part(struct hd_struct *part)
 {
-	kfree(part->info);
 	bdput(part->bdev);
 	percpu_ref_exit(&part->ref);
 }
block/genhd.c +2 -1
index 8212a2dd10ec4e..20c7bf6d091e94 100644
@@ -994,7 +994,8 @@ void __init printk_all_partitions(void)
 			       bdevt_str(part_devt(part), devt_buf),
 			       bdev_nr_sectors(part->bdev) >> 1,
 			       disk_name(disk, part->partno, name_buf),
-			       part->info ? part->info->uuid : "");
+			       part->bdev->bd_meta_info ?
+					part->bdev->bd_meta_info->uuid : "");
 			if (is_part0) {
 				if (dev->parent && dev->parent->driver)
 					printk(" driver: %s\n",
block/partitions/core.c +7 -11
index aa4b836374b037..e24673b4cba61f 100644
@@ -275,8 +275,9 @@ static int part_uevent(struct device *dev, struct kobj_uevent_env *env)
 	struct hd_struct *part = dev_to_part(dev);
 
 	add_uevent_var(env, "PARTN=%u", part->partno);
-	if (part->info && part->info->volname[0])
-		add_uevent_var(env, "PARTNAME=%s", part->info->volname);
+	if (part->bdev->bd_meta_info && part->bdev->bd_meta_info->volname[0])
+		add_uevent_var(env, "PARTNAME=%s",
+			       part->bdev->bd_meta_info->volname);
 	return 0;
 }
 
@@ -422,13 +423,10 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 	p->policy = get_disk_ro(disk);
 
 	if (info) {
-		struct partition_meta_info *pinfo;
-
-		pinfo = kzalloc_node(sizeof(*pinfo), GFP_KERNEL, disk->node_id);
-		if (!pinfo)
+		err = -ENOMEM;
+		bdev->bd_meta_info = kmemdup(info, sizeof(*info), GFP_KERNEL);
+		if (!bdev->bd_meta_info)
 			goto out_bdput;
-		memcpy(pinfo, info, sizeof(*info));
-		p->info = pinfo;
 	}
 
 	dname = dev_name(ddev);
@@ -444,7 +442,7 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 
 	err = blk_alloc_devt(p, &devt);
 	if (err)
-		goto out_free_info;
+		goto out_bdput;
 	pdev->devt = devt;
 
 	/* delay uevent until 'holders' subdir is created */
@@ -481,8 +479,6 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 		kobject_uevent(&pdev->kobj, KOBJ_ADD);
 	return p;
 
-out_free_info:
-	kfree(p->info);
 out_bdput:
 	bdput(bdev);
 out_free:
fs/block_dev.c +1 -0
index 0427e6fa59556f..2393395201aa6c 100644
@@ -785,6 +785,7 @@ static void bdev_free_inode(struct inode *inode)
 	struct block_device *bdev = I_BDEV(inode);
 
 	free_percpu(bdev->bd_stats);
+	kfree(bdev->bd_meta_info);
 
 	kmem_cache_free(bdev_cachep, BDEV_I(inode));
 }
include/linux/blk_types.h +2 -0
index a690008f60cd92..2f8ede04e5a94c 100644
@@ -49,6 +49,8 @@ struct block_device {
 	/* Mutex for freeze */
 	struct mutex		bd_fsfreeze_mutex;
 	struct super_block	*bd_fsfreeze_sb;
+
+	struct partition_meta_info *bd_meta_info;
 } __randomize_layout;
 
 #define bdev_whole(_bdev) \
include/linux/genhd.h +0 -1
index a9d64da474233f..1e52f38b719db3 100644
@@ -57,7 +57,6 @@ struct hd_struct {
 	struct device __dev;
 	struct kobject *holder_dir;
 	int policy, partno;
-	struct partition_meta_info *info;
 #ifdef CONFIG_FAIL_MAKE_REQUEST
 	int make_it_fail;
 #endif
init/do_mounts.c +4 -3
index 5879edf083b318..368ccb71850126 100644
@@ -79,8 +79,8 @@ static int match_dev_by_uuid(struct device *dev, const void *data)
 	const struct uuidcmp *cmp = data;
 	struct hd_struct *part = dev_to_part(dev);
 
-	if (!part->info ||
-	    strncasecmp(cmp->uuid, part->info->uuid, cmp->len))
+	if (!part->bdev->bd_meta_info ||
+	    strncasecmp(cmp->uuid, part->bdev->bd_meta_info->uuid, cmp->len))
 		return 0;
 	return 1;
 }
@@ -169,7 +169,8 @@ static int match_dev_by_label(struct device *dev, const void *data)
 	const char *label = data;
 	struct hd_struct *part = dev_to_part(dev);
 
-	if (!part->info || strcmp(label, part->info->volname))
+	if (!part->bdev->bd_meta_info ||
+	    strcmp(label, part->bdev->bd_meta_info->volname))
 		return 0;
 	return 1;
 }
Signature · 1 line
2.29.2
author maintainer
#75 lore raw

PATCH 34/45 block: move holder_dir to struct block_device

In reply to Christoph Hellwig

Move the holder_dir field to struct block_device in preparation for kill struct hd_struct.

---
 block/genhd.c             |  5 +++--
 block/partitions/core.c   |  8 ++++----
 fs/block_dev.c            | 11 +++++------
 include/linux/blk_types.h |  1 +
 include/linux/genhd.h     |  1 -
 5 files changed, 13 insertions(+), 13 deletions(-)
block/genhd.c +3 -2
index 20c7bf6d091e94..a991f0122e53d8 100644
@@ -671,7 +671,8 @@ static void register_disk(struct device *parent, struct gendisk *disk,
 	 */
 	pm_runtime_set_memalloc_noio(ddev, true);
 
-	disk->part0.holder_dir = kobject_create_and_add("holders", &ddev->kobj);
+	disk->part0.bdev->bd_holder_dir =
+			kobject_create_and_add("holders", &ddev->kobj);
 	disk->slave_dir = kobject_create_and_add("slaves", &ddev->kobj);
 
 	if (disk->flags & GENHD_FL_HIDDEN) {
@@ -871,7 +872,7 @@ void del_gendisk(struct gendisk *disk)
 
 	blk_unregister_queue(disk);
 
-	kobject_put(disk->part0.holder_dir);
+	kobject_put(disk->part0.bdev->bd_holder_dir);
 	kobject_put(disk->slave_dir);
 
 	part_stat_set_all(&disk->part0, 0);
block/partitions/core.c +4 -4
index e24673b4cba61f..ff60a14ed4dcdd 100644
@@ -344,7 +344,7 @@ void delete_partition(struct hd_struct *part)
 	 */
 	get_device(disk_to_dev(disk));
 	rcu_assign_pointer(ptbl->part[part->partno], NULL);
-	kobject_put(part->holder_dir);
+	kobject_put(part->bdev->bd_holder_dir);
 	device_del(part_to_dev(part));
 
 	/*
@@ -452,8 +452,8 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 		goto out_put;
 
 	err = -ENOMEM;
-	p->holder_dir = kobject_create_and_add("holders", &pdev->kobj);
-	if (!p->holder_dir)
+	bdev->bd_holder_dir = kobject_create_and_add("holders", &pdev->kobj);
+	if (!bdev->bd_holder_dir)
 		goto out_del;
 
 	dev_set_uevent_suppress(pdev, 0);
@@ -487,7 +487,7 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 out_remove_file:
 	device_remove_file(pdev, &dev_attr_whole_disk);
 out_del:
-	kobject_put(p->holder_dir);
+	kobject_put(bdev->bd_holder_dir);
 	device_del(pdev);
 out_put:
 	put_device(pdev);
fs/block_dev.c +5 -6
index 2393395201aa6c..0b8d6009486643 100644
@@ -1137,7 +1137,7 @@ int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
 	WARN_ON_ONCE(!bdev->bd_holder);
 
 	/* FIXME: remove the following once add_disk() handles errors */
-	if (WARN_ON(!disk->slave_dir || !bdev->bd_part->holder_dir))
+	if (WARN_ON(!disk->slave_dir || !bdev->bd_holder_dir))
 		goto out_unlock;
 
 	holder = bd_find_holder_disk(bdev, disk);
@@ -1160,14 +1160,14 @@ int bd_link_disk_holder(struct block_device *bdev, struct gendisk *disk)
 	if (ret)
 		goto out_free;
 
-	ret = add_symlink(bdev->bd_part->holder_dir, &disk_to_dev(disk)->kobj);
+	ret = add_symlink(bdev->bd_holder_dir, &disk_to_dev(disk)->kobj);
 	if (ret)
 		goto out_del;
 	/*
 	 * bdev could be deleted beneath us which would implicitly destroy
 	 * the holder directory.  Hold on to it.
 	 */
-	kobject_get(bdev->bd_part->holder_dir);
+	kobject_get(bdev->bd_holder_dir);
 
 	list_add(&holder->list, &bdev->bd_holder_disks);
 	goto out_unlock;
@@ -1202,9 +1202,8 @@ void bd_unlink_disk_holder(struct block_device *bdev, struct gendisk *disk)
 
 	if (!WARN_ON_ONCE(holder == NULL) && !--holder->refcnt) {
 		del_symlink(disk->slave_dir, bdev_kobj(bdev));
-		del_symlink(bdev->bd_part->holder_dir,
-			    &disk_to_dev(disk)->kobj);
-		kobject_put(bdev->bd_part->holder_dir);
+		del_symlink(bdev->bd_holder_dir, &disk_to_dev(disk)->kobj);
+		kobject_put(bdev->bd_holder_dir);
 		list_del_init(&holder->list);
 		kfree(holder);
 	}
include/linux/blk_types.h +1 -0
index 2f8ede04e5a94c..c0591e52d7d7ce 100644
@@ -35,6 +35,7 @@ struct block_device {
 #ifdef CONFIG_SYSFS
 	struct list_head	bd_holder_disks;
 #endif
+	struct kobject		*bd_holder_dir;
 	u8			bd_partno;
 	struct hd_struct *	bd_part;
 	/* number of times partitions within this device have been opened. */
include/linux/genhd.h +0 -1
index 1e52f38b719db3..c2a8cf12c5cab5 100644
@@ -55,7 +55,6 @@ struct hd_struct {
 
 	struct block_device *bdev;
 	struct device __dev;
-	struct kobject *holder_dir;
 	int policy, partno;
 #ifdef CONFIG_FAIL_MAKE_REQUEST
 	int make_it_fail;
Signature · 1 line
2.29.2
author maintainer
#76 lore raw

PATCH 35/45 block: move make_it_fail to struct block_device

In reply to Christoph Hellwig

Move the make_it_fail flag to struct block_device an turn it into a bool in preparation of killing struct hd_struct.

---
 block/blk-core.c          | 3 ++-
 block/genhd.c             | 4 ++--
 include/linux/blk_types.h | 3 +++
 include/linux/genhd.h     | 3 ---
 4 files changed, 7 insertions(+), 6 deletions(-)
block/blk-core.c +2 -1
index 9a3793d5ce38d4..9121390be97a76 100644
@@ -668,7 +668,8 @@ __setup("fail_make_request=", setup_fail_make_request);
 
 static bool should_fail_request(struct hd_struct *part, unsigned int bytes)
 {
-	return part->make_it_fail && should_fail(&fail_make_request, bytes);
+	return part->bdev->bd_make_it_fail &&
+		should_fail(&fail_make_request, bytes);
 }
 
 static int __init fail_make_request_debugfs(void)
block/genhd.c +2 -2
index a991f0122e53d8..8c734a4e8ff31c 100644
@@ -1276,7 +1276,7 @@ ssize_t part_fail_show(struct device *dev,
 {
 	struct hd_struct *p = dev_to_part(dev);
 
-	return sprintf(buf, "%d\n", p->make_it_fail);
+	return sprintf(buf, "%d\n", p->bdev->bd_make_it_fail);
 }
 
 ssize_t part_fail_store(struct device *dev,
@@ -1287,7 +1287,7 @@ ssize_t part_fail_store(struct device *dev,
 	int i;
 
 	if (count > 0 && sscanf(buf, "%d", &i) > 0)
-		p->make_it_fail = (i == 0) ? 0 : 1;
+		p->pdev->bd_make_it_fail = (i == 0) ? 0 : 1;
 
 	return count;
 }
include/linux/blk_types.h +3 -0
index c0591e52d7d7ce..b237f1e4081405 100644
@@ -52,6 +52,9 @@ struct block_device {
 	struct super_block	*bd_fsfreeze_sb;
 
 	struct partition_meta_info *bd_meta_info;
+#ifdef CONFIG_FAIL_MAKE_REQUEST
+	bool			bd_make_it_fail;
+#endif
 } __randomize_layout;
 
 #define bdev_whole(_bdev) \
include/linux/genhd.h +0 -3
index c2a8cf12c5cab5..5d46ea7be7e4f0 100644
@@ -56,9 +56,6 @@ struct hd_struct {
 	struct block_device *bdev;
 	struct device __dev;
 	int policy, partno;
-#ifdef CONFIG_FAIL_MAKE_REQUEST
-	int make_it_fail;
-#endif
 	struct rcu_work rcu_work;
 };
 
Signature · 1 line
2.29.2
author maintainer
#77 lore raw

PATCH 36/45 block: move the policy field to struct block_device

In reply to Christoph Hellwig

Move the policy field to struct block_device and rename it to the more descriptive bd_read_only. Also turn the field into a bool as it is used as such.

---
 block/blk-core.c          | 2 +-
 block/genhd.c             | 8 ++++----
 block/ioctl.c             | 2 +-
 block/partitions/core.c   | 4 ++--
 include/linux/blk_types.h | 1 +
 include/linux/genhd.h     | 4 ++--
 6 files changed, 11 insertions(+), 10 deletions(-)
block/blk-core.c +1 -1
index 9121390be97a76..d64ffcb6f9ae5d 100644
@@ -696,7 +696,7 @@ static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
 {
 	const int op = bio_op(bio);
 
-	if (part->policy && op_is_write(op)) {
+	if (part->bdev->bd_read_only && op_is_write(op)) {
 		char b[BDEVNAME_SIZE];
 
 		if (op_is_flush(bio->bi_opf) && !bio_sectors(bio))
block/genhd.c +4 -4
index 8c734a4e8ff31c..8aed77cc8ad169 100644
@@ -1671,14 +1671,14 @@ void set_disk_ro(struct gendisk *disk, int flag)
 	struct disk_part_iter piter;
 	struct hd_struct *part;
 
-	if (disk->part0.policy != flag) {
+	if (disk->part0.bdev->bd_read_only != flag) {
 		set_disk_ro_uevent(disk, flag);
-		disk->part0.policy = flag;
+		disk->part0.bdev->bd_read_only = flag;
 	}
 
 	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
 	while ((part = disk_part_iter_next(&piter)))
-		part->policy = flag;
+		part->bdev->bd_read_only = flag;
 	disk_part_iter_exit(&piter);
 }
 
@@ -1688,7 +1688,7 @@ int bdev_read_only(struct block_device *bdev)
 {
 	if (!bdev)
 		return 0;
-	return bdev->bd_part->policy;
+	return bdev->bd_read_only;
 }
 
 EXPORT_SYMBOL(bdev_read_only);
block/ioctl.c +1 -1
index a6d8171221c7dc..d61d652078f41c 100644
@@ -345,7 +345,7 @@ static int blkdev_roset(struct block_device *bdev, fmode_t mode,
 		if (ret)
 			return ret;
 	}
-	bdev->bd_part->policy = n;
+	bdev->bd_read_only = n;
 	return 0;
 }
 
block/partitions/core.c +2 -2
index ff60a14ed4dcdd..fd00428e437a63 100644
@@ -199,7 +199,7 @@ static ssize_t part_ro_show(struct device *dev,
 			    struct device_attribute *attr, char *buf)
 {
 	struct hd_struct *p = dev_to_part(dev);
-	return sprintf(buf, "%d\n", p->policy ? 1 : 0);
+	return sprintf(buf, "%d\n", p->bdev->bd_read_only);
 }
 
 static ssize_t part_alignment_offset_show(struct device *dev,
@@ -420,7 +420,7 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 	bdev->bd_start_sect = start;
 	bdev_set_nr_sectors(bdev, len);
 	p->partno = partno;
-	p->policy = get_disk_ro(disk);
+	bdev->bd_read_only = get_disk_ro(disk);
 
 	if (info) {
 		err = -ENOMEM;
include/linux/blk_types.h +1 -0
index b237f1e4081405..758cf71c9aa2a6 100644
@@ -23,6 +23,7 @@ struct block_device {
 	sector_t		bd_start_sect;
 	struct disk_stats __percpu *bd_stats;
 	unsigned long		bd_stamp;
+	bool			bd_read_only;	/* read-only policy */
 	dev_t			bd_dev;
 	int			bd_openers;
 	struct inode *		bd_inode;	/* will die */
include/linux/genhd.h +2 -2
index 5d46ea7be7e4f0..08d00b526b0a3b 100644
@@ -55,7 +55,7 @@ struct hd_struct {
 
 	struct block_device *bdev;
 	struct device __dev;
-	int policy, partno;
+	int partno;
 	struct rcu_work rcu_work;
 };
 
@@ -279,7 +279,7 @@ extern void set_disk_ro(struct gendisk *disk, int flag);
 
 static inline int get_disk_ro(struct gendisk *disk)
 {
-	return disk->part0.policy;
+	return disk->part0.bdev->bd_read_only;
 }
 
 extern void disk_block_events(struct gendisk *disk);
Signature · 1 line
2.29.2
author maintainer
#78 lore raw

PATCH 37/45 block: allocate struct hd_struct as part of struct bdev_inode

In reply to Christoph Hellwig

Allocate hd_struct together with struct block_device to pre-load the lifetime rule changes in preparation of merging the two structures.

Note that part0 was previously embedded into struct gendisk, but is a separate allocation now, and already points to the block_device instead of the hd_struct. The lifetime of struct gendisk is still controlled by the struct device embedded in the part0 hd_struct.

---
 block/blk-core.c                   | 16 ++++---
 block/blk-flush.c                  |  2 +-
 block/blk-merge.c                  |  2 -
 block/blk.h                        | 21 ----------
 block/genhd.c                      | 50 +++++++++-------------
 block/partitions/core.c            | 67 +++---------------------------
 drivers/block/drbd/drbd_receiver.c |  2 +-
 drivers/block/drbd/drbd_worker.c   |  3 +-
 drivers/block/zram/zram_drv.c      |  2 +-
 drivers/md/dm.c                    |  4 +-
 drivers/md/md.c                    |  2 +-
 fs/block_dev.c                     | 37 +++++------------
 include/linux/blk_types.h          |  2 +-
 include/linux/genhd.h              | 14 +++----
 include/linux/part_stat.h          |  4 +-
 15 files changed, 60 insertions(+), 168 deletions(-)
block/blk-core.c +7 -9
index d64ffcb6f9ae5d..9ea70275fc1cfe 100644
@@ -714,7 +714,8 @@ static inline bool bio_check_ro(struct bio *bio, struct hd_struct *part)
 
 static noinline int should_fail_bio(struct bio *bio)
 {
-	if (should_fail_request(&bio->bi_disk->part0, bio->bi_iter.bi_size))
+	if (should_fail_request(bio->bi_disk->part0->bd_part,
+			bio->bi_iter.bi_size))
 		return -EIO;
 	return 0;
 }
@@ -831,7 +832,7 @@ static noinline_for_stack bool submit_bio_checks(struct bio *bio)
 		if (unlikely(blk_partition_remap(bio)))
 			goto end_io;
 	} else {
-		if (unlikely(bio_check_ro(bio, &bio->bi_disk->part0)))
+		if (unlikely(bio_check_ro(bio, bio->bi_disk->part0->bd_part)))
 			goto end_io;
 		if (unlikely(bio_check_eod(bio, get_capacity(bio->bi_disk))))
 			goto end_io;
@@ -1203,7 +1204,7 @@ blk_status_t blk_insert_cloned_request(struct request_queue *q, struct request *
 		return ret;
 
 	if (rq->rq_disk &&
-	    should_fail_request(&rq->rq_disk->part0, blk_rq_bytes(rq)))
+	    should_fail_request(rq->rq_disk->part0->bd_part, blk_rq_bytes(rq)))
 		return BLK_STS_IOERR;
 
 	if (blk_crypto_insert_cloned_request(rq))
@@ -1272,7 +1273,7 @@ static void update_io_ticks(struct hd_struct *part, unsigned long now, bool end)
 			__part_stat_add(part, io_ticks, end ? now - stamp : 1);
 	}
 	if (part->partno) {
-		part = &part_to_disk(part)->part0;
+		part = part_to_disk(part)->part0->bd_part;
 		goto again;
 	}
 }
@@ -1309,8 +1310,6 @@ void blk_account_io_done(struct request *req, u64 now)
 		part_stat_inc(part, ios[sgrp]);
 		part_stat_add(part, nsecs[sgrp], now - req->start_time_ns);
 		part_stat_unlock();
-
-		hd_struct_put(part);
 	}
 }
 
@@ -1354,7 +1353,7 @@ EXPORT_SYMBOL_GPL(part_start_io_acct);
 unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
 				 unsigned int op)
 {
-	return __part_start_io_acct(&disk->part0, sectors, op);
+	return __part_start_io_acct(disk->part0->bd_part, sectors, op);
 }
 EXPORT_SYMBOL(disk_start_io_acct);
 
@@ -1376,14 +1375,13 @@ void part_end_io_acct(struct hd_struct *part, struct bio *bio,
 		      unsigned long start_time)
 {
 	__part_end_io_acct(part, bio_op(bio), start_time);
-	hd_struct_put(part);
 }
 EXPORT_SYMBOL_GPL(part_end_io_acct);
 
 void disk_end_io_acct(struct gendisk *disk, unsigned int op,
 		      unsigned long start_time)
 {
-	__part_end_io_acct(&disk->part0, op, start_time);
+	__part_end_io_acct(disk->part0->bd_part, op, start_time);
 }
 EXPORT_SYMBOL(disk_end_io_acct);
 
block/blk-flush.c +1 -1
index e32958f0b68750..fcd0a60574dff8 100644
@@ -139,7 +139,7 @@ static void blk_flush_queue_rq(struct request *rq, bool add_front)
 
 static void blk_account_io_flush(struct request *rq)
 {
-	struct hd_struct *part = &rq->rq_disk->part0;
+	struct hd_struct *part = rq->rq_disk->part0->bd_part;
 
 	part_stat_lock();
 	part_stat_inc(part, ios[STAT_FLUSH]);
block/blk-merge.c +0 -2
index bcf5e458060337..cb351ab9b77dbd 100644
@@ -683,8 +683,6 @@ static void blk_account_io_merge_request(struct request *req)
 		part_stat_lock();
 		part_stat_inc(req->part, merges[op_stat_group(req_op(req))]);
 		part_stat_unlock();
-
-		hd_struct_put(req->part);
 	}
 }
 
block/blk.h +0 -21
index 0bd4b58bcbaf77..32ac41f7557fcc 100644
@@ -363,27 +363,6 @@ int bdev_del_partition(struct block_device *bdev, int partno);
 int bdev_resize_partition(struct block_device *bdev, int partno,
 		sector_t start, sector_t length);
 int disk_expand_part_tbl(struct gendisk *disk, int target);
-int hd_ref_init(struct hd_struct *part);
-
-/* no need to get/put refcount of part0 */
-static inline int hd_struct_try_get(struct hd_struct *part)
-{
-	if (part->partno)
-		return percpu_ref_tryget_live(&part->ref);
-	return 1;
-}
-
-static inline void hd_struct_put(struct hd_struct *part)
-{
-	if (part->partno)
-		percpu_ref_put(&part->ref);
-}
-
-static inline void hd_free_part(struct hd_struct *part)
-{
-	bdput(part->bdev);
-	percpu_ref_exit(&part->ref);
-}
 
 int bio_add_hw_page(struct request_queue *q, struct bio *bio,
 		struct page *page, unsigned int len, unsigned int offset,
block/genhd.c +19 -31

142 lines not on this page · raw message

block/partitions/core.c +6 -61

115 lines not on this page · raw message

drivers/block/drbd/drbd_receiver.c +1 -1
index dc333dbe523281..9e5c2fdfda3629 100644
@@ -2802,7 +2802,7 @@ bool drbd_rs_c_min_rate_throttle(struct drbd_device *device)
 	if (c_min_rate == 0)
 		return false;
 
-	curr_events = (int)part_stat_read_accum(&disk->part0, sectors) -
+	curr_events = (int)part_stat_read_accum(disk->part0->bd_part, sectors) -
 			atomic_read(&device->rs_sect_ev);
 
 	if (atomic_read(&device->ap_actlog_cnt)
drivers/block/drbd/drbd_worker.c +2 -1
index ba56f3f05312f0..343f56b86bb766 100644
@@ -1678,7 +1678,8 @@ void drbd_rs_controller_reset(struct drbd_device *device)
 	atomic_set(&device->rs_sect_in, 0);
 	atomic_set(&device->rs_sect_ev, 0);
 	device->rs_in_flight = 0;
-	device->rs_last_events = (int)part_stat_read_accum(&disk->part0, sectors);
+	device->rs_last_events =
+		(int)part_stat_read_accum(disk->part0->bd_part, sectors);
 
 	/* Updating the RCU protected object in place is necessary since
 	   this function gets called from atomic context.
drivers/block/zram/zram_drv.c +1 -1
index 88baa6158eaee1..153858734cd47d 100644
@@ -1687,7 +1687,7 @@ static void zram_reset_device(struct zram *zram)
 	zram->disksize = 0;
 
 	set_capacity_and_notify(zram->disk, 0);
-	part_stat_set_all(&zram->disk->part0, 0);
+	part_stat_set_all(zram->disk->part0->bd_part, 0);
 
 	up_write(&zram->init_lock);
 	/* I/O operation under all of CPU are done so let's free */
drivers/md/dm.c +2 -2
index 48051db006f30c..1b2db4d530ea71 100644
@@ -1607,7 +1607,7 @@ static blk_qc_t __split_and_process_bio(struct mapped_device *md,
 				 * (by eliminating DM's splitting and just using bio_split)
 				 */
 				part_stat_lock();
-				__dm_part_stat_sub(&dm_disk(md)->part0,
+				__dm_part_stat_sub(dm_disk(md)->part0->bd_part,
 						   sectors[op_stat_group(bio_op(bio))], ci.sector_count);
 				part_stat_unlock();
 
@@ -2242,7 +2242,7 @@ EXPORT_SYMBOL_GPL(dm_put);
 static bool md_in_flight_bios(struct mapped_device *md)
 {
 	int cpu;
-	struct hd_struct *part = &dm_disk(md)->part0;
+	struct hd_struct *part = dm_disk(md)->part0->bd_part;
 	long sum = 0;
 
 	for_each_possible_cpu(cpu) {
drivers/md/md.c +1 -1
index 7ce6047c856ea2..3696c2d77a4dd7 100644
@@ -8441,7 +8441,7 @@ static int is_mddev_idle(struct mddev *mddev, int init)
 	rcu_read_lock();
 	rdev_for_each_rcu(rdev, mddev) {
 		struct gendisk *disk = rdev->bdev->bd_disk;
-		curr_events = (int)part_stat_read_accum(&disk->part0, sectors) -
+		curr_events = (int)part_stat_read_accum(disk->part0->bd_part, sectors) -
 			      atomic_read(&disk->sync_io);
 		/* sync IO will cause sync_io to increase before the disk_stats
 		 * as sync_io is counted when a request starts, and
fs/block_dev.c +11 -26
index 0b8d6009486643..1538b20ca4bd43 100644
@@ -39,6 +39,7 @@
 
 struct bdev_inode {
 	struct block_device bdev;
+	struct hd_struct hd;
 	struct inode vfs_inode;
 };
 
@@ -885,6 +886,9 @@ struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
 		iput(inode);
 		return NULL;
 	}
+	bdev->bd_part = &BDEV_I(inode)->hd;
+	memset(bdev->bd_part, 0, sizeof(*bdev->bd_part));
+	bdev->bd_part->bdev = bdev;
 	return bdev;
 }
 
@@ -1275,15 +1279,10 @@ EXPORT_SYMBOL_GPL(bdev_disk_changed);
 static int __blkdev_get(struct block_device *bdev, fmode_t mode)
 {
 	struct gendisk *disk = bdev->bd_disk;
-	int ret;
+	int ret = 0;
 
 	if (!bdev->bd_openers) {
 		if (!bdev_is_partition(bdev)) {
-			ret = -ENXIO;
-			bdev->bd_part = disk_get_part(disk, 0);
-			if (!bdev->bd_part)
-				goto out_clear;
-
 			ret = 0;
 			if (disk->fops->open)
 				ret = disk->fops->open(bdev, mode);
@@ -1302,7 +1301,7 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode)
 				bdev_disk_changed(bdev, ret == -ENOMEDIUM);
 
 			if (ret)
-				goto out_clear;
+				return ret;
 		} else {
 			struct block_device *whole = bdget_disk(disk, 0);
 
@@ -1311,16 +1310,14 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode)
 			if (ret) {
 				mutex_unlock(&whole->bd_mutex);
 				bdput(whole);
-				goto out_clear;
+				return ret;
 			}
 			whole->bd_part_count++;
 			mutex_unlock(&whole->bd_mutex);
 
-			bdev->bd_part = disk_get_part(disk, bdev->bd_partno);
 			if (!bdev_nr_sectors(bdev)) {
 				__blkdev_put(whole, mode, 1);
-				ret = -ENXIO;
-				goto out_clear;
+				return -ENXIO;
 			}
 			set_init_blocksize(bdev);
 		}
@@ -1329,7 +1326,6 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode)
 			bdev->bd_bdi = bdi_get(disk->queue->backing_dev_info);
 	} else {
 		if (!bdev_is_partition(bdev)) {
-			ret = 0;
 			if (bdev->bd_disk->fops->open)
 				ret = bdev->bd_disk->fops->open(bdev, mode);
 			/* the same as first opener case, read comment there */
@@ -1342,11 +1338,6 @@ static int __blkdev_get(struct block_device *bdev, fmode_t mode)
 	}
 	bdev->bd_openers++;
 	return 0;
-
- out_clear:
-	disk_put_part(bdev->bd_part);
-	bdev->bd_part = NULL;
-	return ret;
 }
 
 static struct block_device *get_bdev_disk_and_module(dev_t dev)
@@ -1569,18 +1560,12 @@ static void __blkdev_put(struct block_device *bdev, fmode_t mode, int for_part)
 		sync_blockdev(bdev);
 		kill_bdev(bdev);
 		bdev_write_inode(bdev);
-
-		if (!bdev_is_partition(bdev) && disk->fops->release)
-			disk->fops->release(disk, mode);
-
-		disk_put_part(bdev->bd_part);
-		bdev->bd_part = NULL;
 		if (bdev_is_partition(bdev))
 			victim = bdev_whole(bdev);
-	} else {
-		if (!bdev_is_partition(bdev) && disk->fops->release)
-			disk->fops->release(disk, mode);
 	}
+
+	if (!bdev_is_partition(bdev) && disk->fops->release)
+		disk->fops->release(disk, mode);
 	mutex_unlock(&bdev->bd_mutex);
 	bdput(bdev);
 	if (victim)
include/linux/blk_types.h +1 -1
index 758cf71c9aa2a6..6edea5c1625909 100644
@@ -59,7 +59,7 @@ struct block_device {
 } __randomize_layout;
 
 #define bdev_whole(_bdev) \
-	((_bdev)->bd_disk->part0.bdev)
+	((_bdev)->bd_disk->part0)
 
 #define bdev_kobj(_bdev) \
 	(&part_to_dev((_bdev)->bd_part)->kobj)
include/linux/genhd.h +6 -8
index 08d00b526b0a3b..6e16c264439bdb 100644
@@ -19,11 +19,12 @@
 #include <linux/blk_types.h>
 #include <asm/local.h>
 
-#define dev_to_disk(device)	container_of((device), struct gendisk, part0.__dev)
 #define dev_to_part(device)	container_of((device), struct hd_struct, __dev)
-#define disk_to_dev(disk)	(&(disk)->part0.__dev)
 #define part_to_dev(part)	(&((part)->__dev))
 
+#define dev_to_disk(device)	(dev_to_part(device)->bdev->bd_disk)
+#define disk_to_dev(disk)	(part_to_dev((disk)->part0->bd_part))
+
 extern const struct device_type disk_type;
 extern struct device_type part_type;
 extern struct class block_class;
@@ -51,12 +52,9 @@ struct partition_meta_info {
 };
 
 struct hd_struct {
-	struct percpu_ref ref;
-
 	struct block_device *bdev;
 	struct device __dev;
 	int partno;
-	struct rcu_work rcu_work;
 };
 
 /**
@@ -168,7 +166,7 @@ struct gendisk {
 	 * helpers.
 	 */
 	struct disk_part_tbl __rcu *part_tbl;
-	struct hd_struct part0;
+	struct block_device *part0;
 
 	const struct block_device_operations *fops;
 	struct request_queue *queue;
@@ -279,7 +277,7 @@ extern void set_disk_ro(struct gendisk *disk, int flag);
 
 static inline int get_disk_ro(struct gendisk *disk)
 {
-	return disk->part0.bdev->bd_read_only;
+	return disk->part0->bd_read_only;
 }
 
 extern void disk_block_events(struct gendisk *disk);
@@ -303,7 +301,7 @@ static inline sector_t bdev_nr_sectors(struct block_device *bdev)
 
 static inline sector_t get_capacity(struct gendisk *disk)
 {
-	return bdev_nr_sectors(disk->part0.bdev);
+	return bdev_nr_sectors(disk->part0);
 }
 
 int bdev_disk_changed(struct block_device *bdev, bool invalidate);
include/linux/part_stat.h +2 -2
index 87ad60106e1db0..680de036691ef9 100644
@@ -59,8 +59,8 @@ static inline void part_stat_set_all(struct hd_struct *part, int value)
 #define part_stat_add(part, field, addnd)	do {			\
 	__part_stat_add((part), field, addnd);				\
 	if ((part)->partno)						\
-		__part_stat_add(&part_to_disk((part))->part0,		\
-				field, addnd);				\
+		__part_stat_add(part_to_disk((part))->part0->bd_part,	\
+			field, addnd); \
 } while (0)
 
 #define part_stat_dec(part, field)					\
Signature · 1 line
2.29.2
author maintainer
#79 lore raw

PATCH 38/45 block: switch partition lookup to use struct block_device

In reply to Christoph Hellwig

Use struct block_device to lookup partitions on a disk. This removes all usage of struct hd_struct from the I/O path, and this allows removing the percpu refcount in struct hd_struct.

---
 block/bio.c                        |  4 +-
 block/blk-core.c                   | 66 ++++++++++++++----------------
 block/blk-flush.c                  |  2 +-
 block/blk-mq.c                     |  9 ++--
 block/blk-mq.h                     |  7 ++--
 block/blk.h                        |  4 +-
 block/genhd.c                      | 56 +++++++++++++------------
 block/partitions/core.c            |  7 +---
 drivers/block/drbd/drbd_receiver.c |  2 +-
 drivers/block/drbd/drbd_worker.c   |  2 +-
 drivers/block/zram/zram_drv.c      |  2 +-
 drivers/md/bcache/request.c        |  4 +-
 drivers/md/dm.c                    |  4 +-
 drivers/md/md.c                    |  4 +-
 drivers/nvme/target/admin-cmd.c    | 20 ++++-----
 fs/ext4/super.c                    | 18 +++-----
 fs/ext4/sysfs.c                    | 10 +----
 fs/f2fs/f2fs.h                     |  2 +-
 fs/f2fs/super.c                    |  6 +--
 include/linux/blkdev.h             |  8 ++--
 include/linux/genhd.h              |  4 +-
 include/linux/part_stat.h          | 17 ++++----
 22 files changed, 120 insertions(+), 138 deletions(-)
block/bio.c +2 -2
index 669bb47a31988e..ebb18136b86f2f 100644
@@ -608,12 +608,12 @@ void bio_truncate(struct bio *bio, unsigned new_size)
 void guard_bio_eod(struct bio *bio)
 {
 	sector_t maxsector;
-	struct hd_struct *part;
+	struct block_device *part;
 
 	rcu_read_lock();
 	part = __disk_get_part(bio->bi_disk, bio->bi_partno);
 	if (part)
-		maxsector = bdev_nr_sectors(part->bdev);
+		maxsector = bdev_nr_sectors(part);
 	else	
 		maxsector = get_capacity(bio->bi_disk);
 	rcu_read_unlock();
block/blk-core.c +30 -36

187 lines not on this page · raw message

block/blk-flush.c +1 -1
index fcd0a60574dff8..9507dcdd58814c 100644
@@ -139,7 +139,7 @@ static void blk_flush_queue_rq(struct request *rq, bool add_front)
 
 static void blk_account_io_flush(struct request *rq)
 {
-	struct hd_struct *part = rq->rq_disk->part0->bd_part;
+	struct block_device *part = rq->rq_disk->part0;
 
 	part_stat_lock();
 	part_stat_inc(part, ios[STAT_FLUSH]);
block/blk-mq.c +5 -4
index 55bcee5dc0320c..a2593748fa5342 100644
@@ -95,7 +95,7 @@ static void blk_mq_hctx_clear_pending(struct blk_mq_hw_ctx *hctx,
 }
 
 struct mq_inflight {
-	struct hd_struct *part;
+	struct block_device *part;
 	unsigned int inflight[2];
 };
 
@@ -111,7 +111,8 @@ static bool blk_mq_check_inflight(struct blk_mq_hw_ctx *hctx,
 	return true;
 }
 
-unsigned int blk_mq_in_flight(struct request_queue *q, struct hd_struct *part)
+unsigned int blk_mq_in_flight(struct request_queue *q,
+		struct block_device *part)
 {
 	struct mq_inflight mi = { .part = part };
 
@@ -120,8 +121,8 @@ unsigned int blk_mq_in_flight(struct request_queue *q, struct hd_struct *part)
 	return mi.inflight[0] + mi.inflight[1];
 }
 
-void blk_mq_in_flight_rw(struct request_queue *q, struct hd_struct *part,
-			 unsigned int inflight[2])
+void blk_mq_in_flight_rw(struct request_queue *q, struct block_device *part,
+		unsigned int inflight[2])
 {
 	struct mq_inflight mi = { .part = part };
 
block/blk-mq.h +4 -3
index a52703c98b7736..c696515766c780 100644
@@ -182,9 +182,10 @@ static inline bool blk_mq_hw_queue_mapped(struct blk_mq_hw_ctx *hctx)
 	return hctx->nr_ctx && hctx->tags;
 }
 
-unsigned int blk_mq_in_flight(struct request_queue *q, struct hd_struct *part);
-void blk_mq_in_flight_rw(struct request_queue *q, struct hd_struct *part,
-			 unsigned int inflight[2]);
+unsigned int blk_mq_in_flight(struct request_queue *q,
+		struct block_device *part);
+void blk_mq_in_flight_rw(struct request_queue *q, struct block_device *part,
+		unsigned int inflight[2]);
 
 static inline void blk_mq_put_dispatch_budget(struct request_queue *q)
 {
block/blk.h +2 -2
index 32ac41f7557fcc..d5bf8f3a078186 100644
@@ -215,7 +215,7 @@ static inline void elevator_exit(struct request_queue *q,
 	__elevator_exit(q, e);
 }
 
-struct hd_struct *__disk_get_part(struct gendisk *disk, int partno);
+struct block_device *__disk_get_part(struct gendisk *disk, int partno);
 
 ssize_t part_size_show(struct device *dev, struct device_attribute *attr,
 		char *buf);
@@ -348,7 +348,7 @@ void blk_queue_free_zone_bitmaps(struct request_queue *q);
 static inline void blk_queue_free_zone_bitmaps(struct request_queue *q) {}
 #endif
 
-struct hd_struct *disk_map_sector_rcu(struct gendisk *disk, sector_t sector);
+struct block_device *disk_map_sector_rcu(struct gendisk *disk, sector_t sector);
 
 int blk_alloc_devt(struct hd_struct *part, dev_t *devt);
 void blk_free_devt(dev_t devt);
block/genhd.c +30 -26

153 lines not on this page · raw message

block/partitions/core.c +2 -5
index 4b0352cb29e132..8beab9e7727e27 100644
@@ -298,12 +298,9 @@ void delete_partition(struct hd_struct *part)
 	struct disk_part_tbl *ptbl =
 		rcu_dereference_protected(disk->part_tbl, 1);
 
-	/*
-	 * ->part_tbl is referenced in this part's release handler, so
-	 *  we have to hold the disk device
-	 */
 	rcu_assign_pointer(ptbl->part[part->partno], NULL);
 	rcu_assign_pointer(ptbl->last_lookup, NULL);
+
 	kobject_put(part->bdev->bd_holder_dir);
 	device_del(part_to_dev(part));
 
@@ -421,7 +418,7 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 
 	/* everything is up and running, commence */
 	bdev_add(bdev, devt);
-	rcu_assign_pointer(ptbl->part[partno], p);
+	rcu_assign_pointer(ptbl->part[partno], bdev);
 
 	/* suppress uevent if the disk suppresses it */
 	if (!dev_get_uevent_suppress(ddev))
drivers/block/drbd/drbd_receiver.c +1 -1
index 9e5c2fdfda3629..09c86ef3f0fd93 100644
@@ -2802,7 +2802,7 @@ bool drbd_rs_c_min_rate_throttle(struct drbd_device *device)
 	if (c_min_rate == 0)
 		return false;
 
-	curr_events = (int)part_stat_read_accum(disk->part0->bd_part, sectors) -
+	curr_events = (int)part_stat_read_accum(disk->part0, sectors) -
 			atomic_read(&device->rs_sect_ev);
 
 	if (atomic_read(&device->ap_actlog_cnt)
drivers/block/drbd/drbd_worker.c +1 -1
index 343f56b86bb766..02044ab7f767d5 100644
@@ -1679,7 +1679,7 @@ void drbd_rs_controller_reset(struct drbd_device *device)
 	atomic_set(&device->rs_sect_ev, 0);
 	device->rs_in_flight = 0;
 	device->rs_last_events =
-		(int)part_stat_read_accum(disk->part0->bd_part, sectors);
+		(int)part_stat_read_accum(disk->part0, sectors);
 
 	/* Updating the RCU protected object in place is necessary since
 	   this function gets called from atomic context.
drivers/block/zram/zram_drv.c +1 -1
index 153858734cd47d..01757f9578dcb8 100644
@@ -1687,7 +1687,7 @@ static void zram_reset_device(struct zram *zram)
 	zram->disksize = 0;
 
 	set_capacity_and_notify(zram->disk, 0);
-	part_stat_set_all(zram->disk->part0->bd_part, 0);
+	part_stat_set_all(zram->disk->part0, 0);
 
 	up_write(&zram->init_lock);
 	/* I/O operation under all of CPU are done so let's free */
drivers/md/bcache/request.c +2 -2
index afac8d07c1bd00..85b1f2a9b72d68 100644
@@ -475,7 +475,7 @@ struct search {
 	unsigned int		read_dirty_data:1;
 	unsigned int		cache_missed:1;
 
-	struct hd_struct	*part;
+	struct block_device	*part;
 	unsigned long		start_time;
 
 	struct btree_op		op;
@@ -1073,7 +1073,7 @@ struct detached_dev_io_private {
 	unsigned long		start_time;
 	bio_end_io_t		*bi_end_io;
 	void			*bi_private;
-	struct hd_struct	*part;
+	struct block_device	*part;
 };
 
 static void detached_dev_end_io(struct bio *bio)
drivers/md/dm.c +2 -2
index 1b2db4d530ea71..176adcff56b380 100644
@@ -1607,7 +1607,7 @@ static blk_qc_t __split_and_process_bio(struct mapped_device *md,
 				 * (by eliminating DM's splitting and just using bio_split)
 				 */
 				part_stat_lock();
-				__dm_part_stat_sub(dm_disk(md)->part0->bd_part,
+				__dm_part_stat_sub(dm_disk(md)->part0,
 						   sectors[op_stat_group(bio_op(bio))], ci.sector_count);
 				part_stat_unlock();
 
@@ -2242,7 +2242,7 @@ EXPORT_SYMBOL_GPL(dm_put);
 static bool md_in_flight_bios(struct mapped_device *md)
 {
 	int cpu;
-	struct hd_struct *part = dm_disk(md)->part0->bd_part;
+	struct block_device *part = dm_disk(md)->part0;
 	long sum = 0;
 
 	for_each_possible_cpu(cpu) {
drivers/md/md.c +2 -2
index 3696c2d77a4dd7..0065736f05b428 100644
@@ -464,7 +464,7 @@ struct md_io {
 	bio_end_io_t *orig_bi_end_io;
 	void *orig_bi_private;
 	unsigned long start_time;
-	struct hd_struct *part;
+	struct block_device *part;
 };
 
 static void md_end_io(struct bio *bio)
@@ -8441,7 +8441,7 @@ static int is_mddev_idle(struct mddev *mddev, int init)
 	rcu_read_lock();
 	rdev_for_each_rcu(rdev, mddev) {
 		struct gendisk *disk = rdev->bdev->bd_disk;
-		curr_events = (int)part_stat_read_accum(disk->part0->bd_part, sectors) -
+		curr_events = (int)part_stat_read_accum(disk->part0, sectors) -
 			      atomic_read(&disk->sync_io);
 		/* sync IO will cause sync_io to increase before the disk_stats
 		 * as sync_io is counted when a request starts, and
drivers/nvme/target/admin-cmd.c +10 -10
index dca34489a1dc9e..8d90235e4fcc5a 100644
@@ -89,12 +89,12 @@ static u16 nvmet_get_smart_log_nsid(struct nvmet_req *req,
 	if (!ns->bdev)
 		goto out;
 
-	host_reads = part_stat_read(ns->bdev->bd_part, ios[READ]);
-	data_units_read = DIV_ROUND_UP(part_stat_read(ns->bdev->bd_part,
-		sectors[READ]), 1000);
-	host_writes = part_stat_read(ns->bdev->bd_part, ios[WRITE]);
-	data_units_written = DIV_ROUND_UP(part_stat_read(ns->bdev->bd_part,
-		sectors[WRITE]), 1000);
+	host_reads = part_stat_read(ns->bdev, ios[READ]);
+	data_units_read =
+		DIV_ROUND_UP(part_stat_read(ns->bdev, sectors[READ]), 1000);
+	host_writes = part_stat_read(ns->bdev, ios[WRITE]);
+	data_units_written =
+		DIV_ROUND_UP(part_stat_read(ns->bdev, sectors[WRITE]), 1000);
 
 	put_unaligned_le64(host_reads, &slog->host_reads[0]);
 	put_unaligned_le64(data_units_read, &slog->data_units_read[0]);
@@ -120,12 +120,12 @@ static u16 nvmet_get_smart_log_all(struct nvmet_req *req,
 		/* we don't have the right data for file backed ns */
 		if (!ns->bdev)
 			continue;
-		host_reads += part_stat_read(ns->bdev->bd_part, ios[READ]);
+		host_reads += part_stat_read(ns->bdev, ios[READ]);
 		data_units_read += DIV_ROUND_UP(
-			part_stat_read(ns->bdev->bd_part, sectors[READ]), 1000);
-		host_writes += part_stat_read(ns->bdev->bd_part, ios[WRITE]);
+			part_stat_read(ns->bdev, sectors[READ]), 1000);
+		host_writes += part_stat_read(ns->bdev, ios[WRITE]);
 		data_units_written += DIV_ROUND_UP(
-			part_stat_read(ns->bdev->bd_part, sectors[WRITE]), 1000);
+			part_stat_read(ns->bdev, sectors[WRITE]), 1000);
 	}
 
 	put_unaligned_le64(host_reads, &slog->host_reads[0]);
fs/ext4/super.c +6 -12
index 6633b20224d509..c303a0ff0b1701 100644
@@ -4048,9 +4048,8 @@ static int ext4_fill_super(struct super_block *sb, void *data, int silent)
 	sbi->s_sb = sb;
 	sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
 	sbi->s_sb_block = sb_block;
-	if (sb->s_bdev->bd_part)
-		sbi->s_sectors_written_start =
-			part_stat_read(sb->s_bdev->bd_part, sectors[STAT_WRITE]);
+	sbi->s_sectors_written_start =
+		part_stat_read(sb->s_bdev, sectors[STAT_WRITE]);
 
 	/* Cleanup superblock name */
 	strreplace(sb->s_id, '/', '!');
@@ -5509,15 +5508,10 @@ static int ext4_commit_super(struct super_block *sb, int sync)
 	 */
 	if (!(sb->s_flags & SB_RDONLY))
 		ext4_update_tstamp(es, s_wtime);
-	if (sb->s_bdev->bd_part)
-		es->s_kbytes_written =
-			cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
-			    ((part_stat_read(sb->s_bdev->bd_part,
-					     sectors[STAT_WRITE]) -
-			      EXT4_SB(sb)->s_sectors_written_start) >> 1));
-	else
-		es->s_kbytes_written =
-			cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
+	es->s_kbytes_written =
+		cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
+		    ((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) -
+		      EXT4_SB(sb)->s_sectors_written_start) >> 1));
 	if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
 		ext4_free_blocks_count_set(es,
 			EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
fs/ext4/sysfs.c +2 -8
index 4e27fe6ed3ae6a..075aa3a19ff5f1 100644
@@ -62,11 +62,8 @@ static ssize_t session_write_kbytes_show(struct ext4_sb_info *sbi, char *buf)
 {
 	struct super_block *sb = sbi->s_buddy_cache->i_sb;
 
-	if (!sb->s_bdev->bd_part)
-		return snprintf(buf, PAGE_SIZE, "0\n");
 	return snprintf(buf, PAGE_SIZE, "%lu\n",
-			(part_stat_read(sb->s_bdev->bd_part,
-					sectors[STAT_WRITE]) -
+			(part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) -
 			 sbi->s_sectors_written_start) >> 1);
 }
 
@@ -74,12 +71,9 @@ static ssize_t lifetime_write_kbytes_show(struct ext4_sb_info *sbi, char *buf)
 {
 	struct super_block *sb = sbi->s_buddy_cache->i_sb;
 
-	if (!sb->s_bdev->bd_part)
-		return snprintf(buf, PAGE_SIZE, "0\n");
 	return snprintf(buf, PAGE_SIZE, "%llu\n",
 			(unsigned long long)(sbi->s_kbytes_written +
-			((part_stat_read(sb->s_bdev->bd_part,
-					 sectors[STAT_WRITE]) -
+			((part_stat_read(sb->s_bdev, sectors[STAT_WRITE]) -
 			  EXT4_SB(sb)->s_sectors_written_start) >> 1)));
 }
 
fs/f2fs/f2fs.h +1 -1
index cb700d79729680..49681a8d2b14a5 100644
@@ -1675,7 +1675,7 @@ static inline bool f2fs_is_multi_device(struct f2fs_sb_info *sbi)
  * and the return value is in kbytes. s is of struct f2fs_sb_info.
  */
 #define BD_PART_WRITTEN(s)						 \
-(((u64)part_stat_read((s)->sb->s_bdev->bd_part, sectors[STAT_WRITE]) -   \
+	(((u64)part_stat_read((s)->sb->s_bdev, sectors[STAT_WRITE]) -   \
 		(s)->sectors_written_start) >> 1)
 
 static inline void f2fs_update_time(struct f2fs_sb_info *sbi, int type)
fs/f2fs/super.c +2 -4
index d4e7fab352bacb..af9f449da64bac 100644
@@ -3700,10 +3700,8 @@ static int f2fs_fill_super(struct super_block *sb, void *data, int silent)
 	}
 
 	/* For write statistics */
-	if (sb->s_bdev->bd_part)
-		sbi->sectors_written_start =
-			(u64)part_stat_read(sb->s_bdev->bd_part,
-					    sectors[STAT_WRITE]);
+	sbi->sectors_written_start =
+		(u64)part_stat_read(sb->s_bdev, sectors[STAT_WRITE]);
 
 	/* Read accumulated write IO statistics if exists */
 	seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
include/linux/blkdev.h +4 -4
index 8fc0b266610f7f..0e83989b9678c3 100644
@@ -191,7 +191,7 @@ struct request {
 	};
 
 	struct gendisk *rq_disk;
-	struct hd_struct *part;
+	struct block_device *part;
 #ifdef CONFIG_BLK_RQ_ALLOC_TIME
 	/* Time that the first bio started allocating this request. */
 	u64 alloc_time_ns;
@@ -1943,9 +1943,9 @@ unsigned long disk_start_io_acct(struct gendisk *disk, unsigned int sectors,
 void disk_end_io_acct(struct gendisk *disk, unsigned int op,
 		unsigned long start_time);
 
-unsigned long part_start_io_acct(struct gendisk *disk, struct hd_struct **part,
-				 struct bio *bio);
-void part_end_io_acct(struct hd_struct *part, struct bio *bio,
+unsigned long part_start_io_acct(struct gendisk *disk,
+		struct block_device **part, struct bio *bio);
+void part_end_io_acct(struct block_device *part, struct bio *bio,
 		      unsigned long start_time);
 
 /**
include/linux/genhd.h +2 -2
index 6e16c264439bdb..77443a1031e373 100644
@@ -131,8 +131,8 @@ enum {
 struct disk_part_tbl {
 	struct rcu_head rcu_head;
 	int len;
-	struct hd_struct __rcu *last_lookup;
-	struct hd_struct __rcu *part[];
+	struct block_device __rcu *last_lookup;
+	struct block_device __rcu *part[];
 };
 
 struct disk_events;
include/linux/part_stat.h +8 -9
index 680de036691ef9..d2558121d48c00 100644
@@ -25,26 +25,26 @@ struct disk_stats {
 #define part_stat_unlock()	preempt_enable()
 
 #define part_stat_get_cpu(part, field, cpu)				\
-	(per_cpu_ptr((part)->bdev->bd_stats, (cpu))->field)
+	(per_cpu_ptr((part)->bd_stats, (cpu))->field)
 
 #define part_stat_get(part, field)					\
 	part_stat_get_cpu(part, field, smp_processor_id())
 
 #define part_stat_read(part, field)					\
 ({									\
-	typeof((part)->bdev->bd_stats->field) res = 0;			\
+	typeof((part)->bd_stats->field) res = 0;			\
 	unsigned int _cpu;						\
 	for_each_possible_cpu(_cpu)					\
-		res += per_cpu_ptr((part)->bdev->bd_stats, _cpu)->field; \
+		res += per_cpu_ptr((part)->bd_stats, _cpu)->field; \
 	res;								\
 })
 
-static inline void part_stat_set_all(struct hd_struct *part, int value)
+static inline void part_stat_set_all(struct block_device *part, int value)
 {
 	int i;
 
 	for_each_possible_cpu(i)
-		memset(per_cpu_ptr(part->bdev->bd_stats, i), value,
+		memset(per_cpu_ptr(part->bd_stats, i), value,
 				sizeof(struct disk_stats));
 }
 
@@ -54,13 +54,12 @@ static inline void part_stat_set_all(struct hd_struct *part, int value)
 	 part_stat_read(part, field[STAT_DISCARD]))
 
 #define __part_stat_add(part, field, addnd)				\
-	__this_cpu_add((part)->bdev->bd_stats->field, addnd)
+	__this_cpu_add((part)->bd_stats->field, addnd)
 
 #define part_stat_add(part, field, addnd)	do {			\
 	__part_stat_add((part), field, addnd);				\
-	if ((part)->partno)						\
-		__part_stat_add(part_to_disk((part))->part0->bd_part,	\
-			field, addnd); \
+	if ((part)->bd_partno)						\
+		__part_stat_add(bdev_whole(part), field, addnd);	\
 } while (0)
 
 #define part_stat_dec(part, field)					\
Signature · 1 line
2.29.2
#80 lore raw

Re: [PATCH 38/45] block: switch partition lookup to use struct block_device

Quoted 5 lines · Christoph Hellwig

Use struct block_device to lookup partitions on a disk. This removes all usage of struct hd_struct from the I/O path, and this allows removing the percpu refcount in struct hd_struct.

For the bcache part, Acked-by: Coly Li <[email protected]>

Quoted 24 lines
---
 block/bio.c                        |  4 +-
 block/blk-core.c                   | 66 ++++++++++++++----------------
 block/blk-flush.c                  |  2 +-
 block/blk-mq.c                     |  9 ++--
 block/blk-mq.h                     |  7 ++--
 block/blk.h                        |  4 +-
 block/genhd.c                      | 56 +++++++++++++------------
 block/partitions/core.c            |  7 +---
 drivers/block/drbd/drbd_receiver.c |  2 +-
 drivers/block/drbd/drbd_worker.c   |  2 +-
 drivers/block/zram/zram_drv.c      |  2 +-
 drivers/md/bcache/request.c        |  4 +-
 drivers/md/dm.c                    |  4 +-
 drivers/md/md.c                    |  4 +-
 drivers/nvme/target/admin-cmd.c    | 20 ++++-----
 fs/ext4/super.c                    | 18 +++-----
 fs/ext4/sysfs.c                    | 10 +----
 fs/f2fs/f2fs.h                     |  2 +-
 fs/f2fs/super.c                    |  6 +--
 include/linux/blkdev.h             |  8 ++--
 include/linux/genhd.h              |  4 +-
 include/linux/part_stat.h          | 17 ++++----
 22 files changed, 120 insertions(+), 138 deletions(-)

[snipped]

Quoted 20 lines
drivers/md/bcache/request.c +2 -2
index afac8d07c1bd00..85b1f2a9b72d68 100644
@@ -475,7 +475,7 @@ struct search {
 	unsigned int		read_dirty_data:1;
 	unsigned int		cache_missed:1;
 
-	struct hd_struct	*part;
+	struct block_device	*part;
 	unsigned long		start_time;
 
 	struct btree_op		op;
@@ -1073,7 +1073,7 @@ struct detached_dev_io_private {
 	unsigned long		start_time;
 	bio_end_io_t		*bi_end_io;
 	void			*bi_private;
-	struct hd_struct	*part;
+	struct block_device	*part;
 };

[snipped]

Acked-by on v1 38/45
#81 lore raw

Re: [PATCH 38/45] block: switch partition lookup to use struct block_device

In reply to Christoph Hellwig
Quoted 17 lines · Christoph Hellwig

Use struct block_device to lookup partitions on a disk. This removes all usage of struct hd_struct from the I/O path, and this allows removing the percpu refcount in struct hd_struct.

---
  block/bio.c                        |  4 +-
  block/blk-core.c                   | 66 ++++++++++++++----------------
  block/blk-flush.c                  |  2 +-
  block/blk-mq.c                     |  9 ++--
  block/blk-mq.h                     |  7 ++--
  block/blk.h                        |  4 +-
  block/genhd.c                      | 56 +++++++++++++------------
  block/partitions/core.c            |  7 +---
  drivers/block/drbd/drbd_receiver.c |  2 +-
  drivers/block/drbd/drbd_worker.c   |  2 +-
  drivers/block/zram/zram_drv.c      |  2 +-
Quoted 8 lines
  drivers/md/bcache/request.c        |  4 +-
  drivers/md/dm.c                    |  4 +-
  drivers/md/md.c                    |  4 +-
  drivers/nvme/target/admin-cmd.c    | 20 ++++-----
  fs/ext4/super.c                    | 18 +++-----
  fs/ext4/sysfs.c                    | 10 +----
  fs/f2fs/f2fs.h                     |  2 +-
  fs/f2fs/super.c                    |  6 +--

For f2fs part,

Thanks,

Quoted 4 lines
  include/linux/blkdev.h             |  8 ++--
  include/linux/genhd.h              |  4 +-
  include/linux/part_stat.h          | 17 ++++----
  22 files changed, 120 insertions(+), 138 deletions(-)
author maintainer
#82 lore raw

PATCH 39/45 block: remove the partno field from struct hd_struct

In reply to Christoph Hellwig

Just use the bd_partno field in struct block_device everywhere.

---
 block/genhd.c           | 12 ++++++------
 block/partitions/core.c |  9 ++++-----
 include/linux/genhd.h   |  1 -
 init/do_mounts.c        |  2 +-
 4 files changed, 11 insertions(+), 13 deletions(-)
block/genhd.c +6 -6
index 197120c0c60f23..60004bc8ba5b56 100644
@@ -576,8 +576,8 @@ int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
 	int idx;
 
 	/* in consecutive minor range? */
-	if (part->partno < disk->minors) {
-		*devt = MKDEV(disk->major, disk->first_minor + part->partno);
+	if (part->bdev->bd_partno < disk->minors) {
+		*devt = MKDEV(disk->major, disk->first_minor + part->bdev->bd_partno);
 		return 0;
 	}
 
@@ -847,7 +847,7 @@ void del_gendisk(struct gendisk *disk)
 	disk_part_iter_init(&piter, disk,
 			     DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
 	while ((part = disk_part_iter_next(&piter))) {
-		invalidate_partition(disk, part->partno);
+		invalidate_partition(disk, part->bdev->bd_partno);
 		delete_partition(part);
 	}
 	disk_part_iter_exit(&piter);
@@ -989,7 +989,7 @@ void __init printk_all_partitions(void)
 			printk("%s%s %10llu %s %s", is_part0 ? "" : "  ",
 			       bdevt_str(part_devt(part), devt_buf),
 			       bdev_nr_sectors(part->bdev) >> 1,
-			       disk_name(disk, part->partno, name_buf),
+			       disk_name(disk, part->bdev->bd_partno, name_buf),
 			       part->bdev->bd_meta_info ?
 					part->bdev->bd_meta_info->uuid : "");
 			if (is_part0) {
@@ -1083,7 +1083,7 @@ static int show_partition(struct seq_file *seqf, void *v)
 		seq_printf(seqf, "%4d  %7d %10llu %s\n",
 			   MAJOR(part_devt(part)), MINOR(part_devt(part)),
 			   bdev_nr_sectors(part->bdev) >> 1,
-			   disk_name(sgp, part->partno, buf));
+			   disk_name(sgp, part->bdev->bd_partno, buf));
 	disk_part_iter_exit(&piter);
 
 	return 0;
@@ -1506,7 +1506,7 @@ static int diskstats_show(struct seq_file *seqf, void *v)
 			   "%lu %u"
 			   "\n",
 			   MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
-			   disk_name(gp, hd->partno, buf),
+			   disk_name(gp, hd->bdev->bd_partno, buf),
 			   stat.ios[STAT_READ],
 			   stat.merges[STAT_READ],
 			   stat.sectors[STAT_READ],
block/partitions/core.c +4 -5
index 8beab9e7727e27..ee4f4e3237aa2d 100644
@@ -184,7 +184,7 @@ static ssize_t part_partition_show(struct device *dev,
 {
 	struct hd_struct *p = dev_to_part(dev);
 
-	return sprintf(buf, "%d\n", p->partno);
+	return sprintf(buf, "%d\n", p->bdev->bd_partno);
 }
 
 static ssize_t part_start_show(struct device *dev,
@@ -274,7 +274,7 @@ static int part_uevent(struct device *dev, struct kobj_uevent_env *env)
 {
 	struct hd_struct *part = dev_to_part(dev);
 
-	add_uevent_var(env, "PARTN=%u", part->partno);
+	add_uevent_var(env, "PARTN=%u", part->bdev->bd_partno);
 	if (part->bdev->bd_meta_info && part->bdev->bd_meta_info->volname[0])
 		add_uevent_var(env, "PARTNAME=%s",
 			       part->bdev->bd_meta_info->volname);
@@ -298,7 +298,7 @@ void delete_partition(struct hd_struct *part)
 	struct disk_part_tbl *ptbl =
 		rcu_dereference_protected(disk->part_tbl, 1);
 
-	rcu_assign_pointer(ptbl->part[part->partno], NULL);
+	rcu_assign_pointer(ptbl->part[part->bdev->bd_partno], NULL);
 	rcu_assign_pointer(ptbl->last_lookup, NULL);
 
 	kobject_put(part->bdev->bd_holder_dir);
@@ -372,7 +372,6 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 
 	bdev->bd_start_sect = start;
 	bdev_set_nr_sectors(bdev, len);
-	p->partno = partno;
 	bdev->bd_read_only = get_disk_ro(disk);
 
 	if (info) {
@@ -445,7 +444,7 @@ static bool partition_overlaps(struct gendisk *disk, sector_t start,
 
 	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
 	while ((part = disk_part_iter_next(&piter))) {
-		if (part->partno == skip_partno ||
+		if (part->bdev->bd_partno == skip_partno ||
 		    start >= part->bdev->bd_start_sect +
 			bdev_nr_sectors(part->bdev) ||
 		    start + length <= part->bdev->bd_start_sect)
include/linux/genhd.h +0 -1
index 77443a1031e373..afe27a0b8f5dd8 100644
@@ -54,7 +54,6 @@ struct partition_meta_info {
 struct hd_struct {
 	struct block_device *bdev;
 	struct device __dev;
-	int partno;
 };
 
 /**
init/do_mounts.c +1 -1
index 368ccb71850126..86bef93e72ebd6 100644
@@ -136,7 +136,7 @@ static dev_t devt_from_partuuid(const char *uuid_str)
 		struct hd_struct *part;
 
 		part = disk_get_part(dev_to_disk(dev),
-				     dev_to_part(dev)->partno + offset);
+				     dev_to_part(dev)->bdev->bd_partno + offset);
 		if (part) {
 			devt = part_devt(part);
 			put_device(part_to_dev(part));
Signature · 1 line
2.29.2
author maintainer
#83 lore raw

PATCH 40/45 block: pass a block_device to blk_alloc_devt

In reply to Christoph Hellwig

Pass the block_device actually needed instead of the hd_struct.

---
 block/blk.h             |  2 +-
 block/genhd.c           | 14 +++++++-------
 block/partitions/core.c |  2 +-
 3 files changed, 9 insertions(+), 9 deletions(-)
block/blk.h +1 -1
index d5bf8f3a078186..9657c6da7c770c 100644
@@ -350,7 +350,7 @@ static inline void blk_queue_free_zone_bitmaps(struct request_queue *q) {}
 
 struct block_device *disk_map_sector_rcu(struct gendisk *disk, sector_t sector);
 
-int blk_alloc_devt(struct hd_struct *part, dev_t *devt);
+int blk_alloc_devt(struct block_device *part, dev_t *devt);
 void blk_free_devt(dev_t devt);
 char *disk_name(struct gendisk *hd, int partno, char *buf);
 #define ADDPART_FLAG_NONE	0
block/genhd.c +7 -7
index 60004bc8ba5b56..498c816e90df64 100644
@@ -557,8 +557,8 @@ static int blk_mangle_minor(int minor)
 }
 
 /**
- * blk_alloc_devt - allocate a dev_t for a partition
- * @part: partition to allocate dev_t for
+ * blk_alloc_devt - allocate a dev_t for a block device
+ * @bdev: block device to allocate dev_t for
  * @devt: out parameter for resulting dev_t
  *
  * Allocate a dev_t for block device.
@@ -570,14 +570,14 @@ static int blk_mangle_minor(int minor)
  * CONTEXT:
  * Might sleep.
  */
-int blk_alloc_devt(struct hd_struct *part, dev_t *devt)
+int blk_alloc_devt(struct block_device *bdev, dev_t *devt)
 {
-	struct gendisk *disk = part_to_disk(part);
+	struct gendisk *disk = bdev->bd_disk;
 	int idx;
 
 	/* in consecutive minor range? */
-	if (part->bdev->bd_partno < disk->minors) {
-		*devt = MKDEV(disk->major, disk->first_minor + part->bdev->bd_partno);
+	if (bdev->bd_partno < disk->minors) {
+		*devt = MKDEV(disk->major, disk->first_minor + bdev->bd_partno);
 		return 0;
 	}
 
@@ -733,7 +733,7 @@ static void __device_add_disk(struct device *parent, struct gendisk *disk,
 
 	disk->flags |= GENHD_FL_UP;
 
-	retval = blk_alloc_devt(disk->part0->bd_part, &devt);
+	retval = blk_alloc_devt(disk->part0, &devt);
 	if (retval) {
 		WARN_ON(1);
 		return;
block/partitions/core.c +1 -1
index ee4f4e3237aa2d..45fed1108d4425 100644
@@ -392,7 +392,7 @@ static struct hd_struct *add_partition(struct gendisk *disk, int partno,
 	pdev->type = &part_type;
 	pdev->parent = ddev;
 
-	err = blk_alloc_devt(p, &devt);
+	err = blk_alloc_devt(bdev, &devt);
 	if (err)
 		goto out_bdput;
 	pdev->devt = devt;
Signature · 1 line
2.29.2
author maintainer
#84 lore raw

PATCH 41/45 block: pass a block_device to invalidate_partition

In reply to Christoph Hellwig

Pass the block_device actually needed instead of looking it up using bdget_disk.

---
 block/genhd.c | 13 +++----------
 1 file changed, 3 insertions(+), 10 deletions(-)
block/genhd.c +3 -10
index 498c816e90df64..2985740eab084b 100644
@@ -790,14 +790,8 @@ void device_add_disk_no_queue_reg(struct device *parent, struct gendisk *disk)
 }
 EXPORT_SYMBOL(device_add_disk_no_queue_reg);
 
-static void invalidate_partition(struct gendisk *disk, int partno)
+static void invalidate_partition(struct block_device *bdev)
 {
-	struct block_device *bdev;
-
-	bdev = bdget_disk(disk, partno);
-	if (!bdev)
-		return;
-
 	fsync_bdev(bdev);
 	__invalidate_device(bdev, true);
 
@@ -806,7 +800,6 @@ static void invalidate_partition(struct gendisk *disk, int partno)
 	 * as last inode reference is dropped.
 	 */
 	remove_inode_hash(bdev->bd_inode);
-	bdput(bdev);
 }
 
 /**
@@ -847,12 +840,12 @@ void del_gendisk(struct gendisk *disk)
 	disk_part_iter_init(&piter, disk,
 			     DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
 	while ((part = disk_part_iter_next(&piter))) {
-		invalidate_partition(disk, part->bdev->bd_partno);
+		invalidate_partition(part->bdev);
 		delete_partition(part);
 	}
 	disk_part_iter_exit(&piter);
 
-	invalidate_partition(disk, 0);
+	invalidate_partition(disk->part0);
 	set_capacity(disk, 0);
 
 	if (!(disk->flags & GENHD_FL_HIDDEN)) {
Signature · 1 line
2.29.2
author maintainer
#85 lore raw

PATCH 42/45 block: switch disk_part_iter_* to use a struct block_device

In reply to Christoph Hellwig

Switch the partition iter infrastructure to iterate over block_device references instead of hd_struct ones mostly used to get at the block_device.

---
 block/genhd.c             | 57 +++++++++++++++++++--------------------
 block/partitions/core.c   | 13 +++++----
 drivers/s390/block/dasd.c |  8 +++---
 include/linux/genhd.h     |  4 +--
 4 files changed, 40 insertions(+), 42 deletions(-)
block/genhd.c +28 -29
index 2985740eab084b..f0bf10192066ac 100644
@@ -231,7 +231,7 @@ EXPORT_SYMBOL_GPL(disk_part_iter_init);
  * CONTEXT:
  * Don't care.
  */
-struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
+struct block_device *disk_part_iter_next(struct disk_part_iter *piter)
 {
 	struct disk_part_tbl *ptbl;
 	int inc, end;
@@ -269,8 +269,7 @@ struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter)
 		      piter->idx == 0))
 			continue;
 
-		get_device(part_to_dev(part->bd_part));
-		piter->part = part->bd_part;
+		piter->part = bdgrab(part);
 		piter->idx += inc;
 		break;
 	}
@@ -292,7 +291,8 @@ EXPORT_SYMBOL_GPL(disk_part_iter_next);
  */
 void disk_part_iter_exit(struct disk_part_iter *piter)
 {
-	disk_put_part(piter->part);
+	if (piter->part)
+		bdput(piter->part);
 	piter->part = NULL;
 }
 EXPORT_SYMBOL_GPL(disk_part_iter_exit);
@@ -333,7 +333,6 @@ struct block_device *disk_map_sector_rcu(struct gendisk *disk, sector_t sector)
 
 	for (i = 1; i < ptbl->len; i++) {
 		part = rcu_dereference(ptbl->part[i]);
-
 		if (part && sector_in_part(part, sector)) {
 			rcu_assign_pointer(ptbl->last_lookup, part);
 			goto out_unlock;
@@ -634,7 +633,7 @@ static void register_disk(struct device *parent, struct gendisk *disk,
 {
 	struct device *ddev = disk_to_dev(disk);
 	struct disk_part_iter piter;
-	struct hd_struct *part;
+	struct block_device *part;
 	int err;
 
 	ddev->parent = parent;
@@ -684,7 +683,7 @@ static void register_disk(struct device *parent, struct gendisk *disk,
 	/* announce possible partitions */
 	disk_part_iter_init(&piter, disk, 0);
 	while ((part = disk_part_iter_next(&piter)))
-		kobject_uevent(&part_to_dev(part)->kobj, KOBJ_ADD);
+		kobject_uevent(bdev_kobj(part), KOBJ_ADD);
 	disk_part_iter_exit(&piter);
 
 	if (disk->queue->backing_dev_info->dev) {
@@ -824,7 +823,7 @@ static void invalidate_partition(struct block_device *bdev)
 void del_gendisk(struct gendisk *disk)
 {
 	struct disk_part_iter piter;
-	struct hd_struct *part;
+	struct block_device *part;
 
 	might_sleep();
 
@@ -840,8 +839,8 @@ void del_gendisk(struct gendisk *disk)
 	disk_part_iter_init(&piter, disk,
 			     DISK_PITER_INCL_EMPTY | DISK_PITER_REVERSE);
 	while ((part = disk_part_iter_next(&piter))) {
-		invalidate_partition(part->bdev);
-		delete_partition(part);
+		invalidate_partition(part);
+		delete_partition(part->bd_part);
 	}
 	disk_part_iter_exit(&piter);
 
@@ -958,7 +957,7 @@ void __init printk_all_partitions(void)
 	while ((dev = class_dev_iter_next(&iter))) {
 		struct gendisk *disk = dev_to_disk(dev);
 		struct disk_part_iter piter;
-		struct hd_struct *part;
+		struct block_device *part;
 		char name_buf[BDEVNAME_SIZE];
 		char devt_buf[BDEVT_SIZE];
 
@@ -977,14 +976,14 @@ void __init printk_all_partitions(void)
 		 */
 		disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 		while ((part = disk_part_iter_next(&piter))) {
-			bool is_part0 = part == disk->part0->bd_part;
+			bool is_part0 = part == disk->part0;
 
 			printk("%s%s %10llu %s %s", is_part0 ? "" : "  ",
-			       bdevt_str(part_devt(part), devt_buf),
-			       bdev_nr_sectors(part->bdev) >> 1,
-			       disk_name(disk, part->bdev->bd_partno, name_buf),
-			       part->bdev->bd_meta_info ?
-					part->bdev->bd_meta_info->uuid : "");
+			       bdevt_str(part->bd_dev, devt_buf),
+			       bdev_nr_sectors(part) >> 1,
+			       disk_name(disk, part->bd_partno, name_buf),
+			       part->bd_meta_info ?
+					part->bd_meta_info->uuid : "");
 			if (is_part0) {
 				if (dev->parent && dev->parent->driver)
 					printk(" driver: %s\n",
@@ -1060,7 +1059,7 @@ static int show_partition(struct seq_file *seqf, void *v)
 {
 	struct gendisk *sgp = v;
 	struct disk_part_iter piter;
-	struct hd_struct *part;
+	struct block_device *part;
 	char buf[BDEVNAME_SIZE];
 
 	/* Don't show non-partitionable removeable devices or empty devices */
@@ -1074,9 +1073,9 @@ static int show_partition(struct seq_file *seqf, void *v)
 	disk_part_iter_init(&piter, sgp, DISK_PITER_INCL_PART0);
 	while ((part = disk_part_iter_next(&piter)))
 		seq_printf(seqf, "%4d  %7d %10llu %s\n",
-			   MAJOR(part_devt(part)), MINOR(part_devt(part)),
-			   bdev_nr_sectors(part->bdev) >> 1,
-			   disk_name(sgp, part->bdev->bd_partno, buf));
+			   MAJOR(part->bd_dev), MINOR(part->bd_dev),
+			   bdev_nr_sectors(part) >> 1,
+			   disk_name(sgp, part->bd_partno, buf));
 	disk_part_iter_exit(&piter);
 
 	return 0;
@@ -1470,7 +1469,7 @@ static int diskstats_show(struct seq_file *seqf, void *v)
 {
 	struct gendisk *gp = v;
 	struct disk_part_iter piter;
-	struct hd_struct *hd;
+	struct block_device *hd;
 	char buf[BDEVNAME_SIZE];
 	unsigned int inflight;
 	struct disk_stats stat;
@@ -1485,11 +1484,11 @@ static int diskstats_show(struct seq_file *seqf, void *v)
 
 	disk_part_iter_init(&piter, gp, DISK_PITER_INCL_EMPTY_PART0);
 	while ((hd = disk_part_iter_next(&piter))) {
-		part_stat_read_all(hd, &stat);
+		part_stat_read_all(hd->bd_part, &stat);
 		if (queue_is_mq(gp->queue))
-			inflight = blk_mq_in_flight(gp->queue, hd->bdev);
+			inflight = blk_mq_in_flight(gp->queue, hd);
 		else
-			inflight = part_in_flight(hd->bdev);
+			inflight = part_in_flight(hd);
 
 		seq_printf(seqf, "%4d %7d %s "
 			   "%lu %lu %lu %u "
@@ -1498,8 +1497,8 @@ static int diskstats_show(struct seq_file *seqf, void *v)
 			   "%lu %lu %lu %u "
 			   "%lu %u"
 			   "\n",
-			   MAJOR(part_devt(hd)), MINOR(part_devt(hd)),
-			   disk_name(gp, hd->bdev->bd_partno, buf),
+			   MAJOR(hd->bd_dev), MINOR(hd->bd_dev),
+			   disk_name(gp, hd->bd_partno, buf),
 			   stat.ios[STAT_READ],
 			   stat.merges[STAT_READ],
 			   stat.sectors[STAT_READ],
@@ -1654,7 +1653,7 @@ static void set_disk_ro_uevent(struct gendisk *gd, int ro)
 void set_disk_ro(struct gendisk *disk, int flag)
 {
 	struct disk_part_iter piter;
-	struct hd_struct *part;
+	struct block_device *part;
 
 	if (disk->part0->bd_read_only != flag) {
 		set_disk_ro_uevent(disk, flag);
@@ -1663,7 +1662,7 @@ void set_disk_ro(struct gendisk *disk, int flag)
 
 	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
 	while ((part = disk_part_iter_next(&piter)))
-		part->bdev->bd_read_only = flag;
+		part->bd_read_only = flag;
 	disk_part_iter_exit(&piter);
 }
 
block/partitions/core.c +6 -7
index 45fed1108d4425..c189ebd4569812 100644
@@ -439,15 +439,14 @@ static bool partition_overlaps(struct gendisk *disk, sector_t start,
 		sector_t length, int skip_partno)
 {
 	struct disk_part_iter piter;
-	struct hd_struct *part;
+	struct block_device *part;
 	bool overlap = false;
 
 	disk_part_iter_init(&piter, disk, DISK_PITER_INCL_EMPTY);
 	while ((part = disk_part_iter_next(&piter))) {
-		if (part->bdev->bd_partno == skip_partno ||
-		    start >= part->bdev->bd_start_sect +
-			bdev_nr_sectors(part->bdev) ||
-		    start + length <= part->bdev->bd_start_sect)
+		if (part->bd_partno == skip_partno ||
+		    start >= part->bd_start_sect + bdev_nr_sectors(part) ||
+		    start + length <= part->bd_start_sect)
 			continue;
 		overlap = true;
 		break;
@@ -568,7 +567,7 @@ static bool disk_unlock_native_capacity(struct gendisk *disk)
 int blk_drop_partitions(struct block_device *bdev)
 {
 	struct disk_part_iter piter;
-	struct hd_struct *part;
+	struct block_device *part;
 
 	if (bdev->bd_part_count)
 		return -EBUSY;
@@ -578,7 +577,7 @@ int blk_drop_partitions(struct block_device *bdev)
 
 	disk_part_iter_init(&piter, bdev->bd_disk, DISK_PITER_INCL_EMPTY);
 	while ((part = disk_part_iter_next(&piter)))
-		delete_partition(part);
+		delete_partition(part->bd_part);
 	disk_part_iter_exit(&piter);
 
 	return 0;
drivers/s390/block/dasd.c +4 -4
index db24e04ee9781e..1825fa8d05a780 100644
@@ -432,7 +432,7 @@ dasd_state_ready_to_online(struct dasd_device * device)
 {
 	struct gendisk *disk;
 	struct disk_part_iter piter;
-	struct hd_struct *part;
+	struct block_device *part;
 
 	device->state = DASD_STATE_ONLINE;
 	if (device->block) {
@@ -445,7 +445,7 @@ dasd_state_ready_to_online(struct dasd_device * device)
 		disk = device->block->bdev->bd_disk;
 		disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 		while ((part = disk_part_iter_next(&piter)))
-			kobject_uevent(&part_to_dev(part)->kobj, KOBJ_CHANGE);
+			kobject_uevent(bdev_kobj(part), KOBJ_CHANGE);
 		disk_part_iter_exit(&piter);
 	}
 	return 0;
@@ -459,7 +459,7 @@ static int dasd_state_online_to_ready(struct dasd_device *device)
 	int rc;
 	struct gendisk *disk;
 	struct disk_part_iter piter;
-	struct hd_struct *part;
+	struct block_device *part;
 
 	if (device->discipline->online_to_ready) {
 		rc = device->discipline->online_to_ready(device);
@@ -472,7 +472,7 @@ static int dasd_state_online_to_ready(struct dasd_device *device)
 		disk = device->block->bdev->bd_disk;
 		disk_part_iter_init(&piter, disk, DISK_PITER_INCL_PART0);
 		while ((part = disk_part_iter_next(&piter)))
-			kobject_uevent(&part_to_dev(part)->kobj, KOBJ_CHANGE);
+			kobject_uevent(bdev_kobj(part), KOBJ_CHANGE);
 		disk_part_iter_exit(&piter);
 	}
 	return 0;
include/linux/genhd.h +2 -2
index afe27a0b8f5dd8..de2ee5ffeefe45 100644
@@ -245,14 +245,14 @@ static inline void disk_put_part(struct hd_struct *part)
 
 struct disk_part_iter {
 	struct gendisk		*disk;
-	struct hd_struct	*part;
+	struct block_device	*part;
 	int			idx;
 	unsigned int		flags;
 };
 
 extern void disk_part_iter_init(struct disk_part_iter *piter,
 				 struct gendisk *disk, unsigned int flags);
-extern struct hd_struct *disk_part_iter_next(struct disk_part_iter *piter);
+struct block_device *disk_part_iter_next(struct disk_part_iter *piter);
 extern void disk_part_iter_exit(struct disk_part_iter *piter);
 extern bool disk_has_partitions(struct gendisk *disk);
 
Signature · 1 line
2.29.2
author maintainer
#86 lore raw

PATCH 43/45 f2fs: remove a few bd_part checks

In reply to Christoph Hellwig

bd_part is never NULL for a block device in use by a file system, so remove the checks.

---
 fs/f2fs/checkpoint.c | 5 +----
 fs/f2fs/sysfs.c      | 9 ---------
 2 files changed, 1 insertion(+), 13 deletions(-)
fs/f2fs/checkpoint.c +1 -4
index 023462e80e58d5..54a1905af052cc 100644
@@ -1395,7 +1395,6 @@ static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 	__u32 crc32 = 0;
 	int i;
 	int cp_payload_blks = __cp_payload(sbi);
-	struct super_block *sb = sbi->sb;
 	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
 	u64 kbytes_written;
 	int err;
@@ -1489,9 +1488,7 @@ static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
 	start_blk += data_sum_blocks;
 
 	/* Record write statistics in the hot node summary */
-	kbytes_written = sbi->kbytes_written;
-	if (sb->s_bdev->bd_part)
-		kbytes_written += BD_PART_WRITTEN(sbi);
+	kbytes_written = sbi->kbytes_written + BD_PART_WRITTEN(sbi);
 
 	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
 
fs/f2fs/sysfs.c +0 -9
index ec77ccfea923dc..24e876e849c512 100644
@@ -90,11 +90,6 @@ static ssize_t free_segments_show(struct f2fs_attr *a,
 static ssize_t lifetime_write_kbytes_show(struct f2fs_attr *a,
 		struct f2fs_sb_info *sbi, char *buf)
 {
-	struct super_block *sb = sbi->sb;
-
-	if (!sb->s_bdev->bd_part)
-		return sprintf(buf, "0\n");
-
 	return sprintf(buf, "%llu\n",
 			(unsigned long long)(sbi->kbytes_written +
 			BD_PART_WRITTEN(sbi)));
@@ -103,12 +98,8 @@ static ssize_t lifetime_write_kbytes_show(struct f2fs_attr *a,
 static ssize_t features_show(struct f2fs_attr *a,
 		struct f2fs_sb_info *sbi, char *buf)
 {
-	struct super_block *sb = sbi->sb;
 	int len = 0;
 
-	if (!sb->s_bdev->bd_part)
-		return sprintf(buf, "0\n");
-
 	if (f2fs_sb_has_encrypt(sbi))
 		len += scnprintf(buf, PAGE_SIZE - len, "%s",
 						"encryption");
Signature · 1 line
2.29.2
author maintainer
#88 lore raw

PATCH 44/45 block: merge struct block_device and struct hd_struct

In reply to Christoph Hellwig

Instead of having two structures that represent each block device with different life time rules, merge them into a single one. This also greatly simplifies the reference counting rules, as we can use the inode reference count as the main reference count for the new struct block_device, with the device model reference front ending it for device model interaction. The percpu refcount in struct hd_struct is entirely gone given that struct block_device must be opened and thus valid for the duration of the I/O.

---
 block/blk-cgroup.c        |   9 ++-
 block/blk.h               |   2 +-
 block/genhd.c             |  86 +++++++++-------------------
 block/partitions/core.c   | 116 +++++++++++++++-----------------------
 fs/block_dev.c            |   9 ---
 include/linux/blk_types.h |   8 ++-
 include/linux/blkdev.h    |   1 -
 include/linux/genhd.h     |  40 +++----------
 init/do_mounts.c          |  21 ++++---
 kernel/trace/blktrace.c   |  43 +++-----------
 10 files changed, 109 insertions(+), 226 deletions(-)
block/blk-cgroup.c +4 -5
index a598f86e014137..99e0e0b3e347de 100644
@@ -820,9 +820,9 @@ static void blkcg_fill_root_iostats(void)
 
 	class_dev_iter_init(&iter, &block_class, NULL, &disk_type);
 	while ((dev = class_dev_iter_next(&iter))) {
-		struct gendisk *disk = dev_to_disk(dev);
-		struct hd_struct *part = disk_get_part(disk, 0);
-		struct blkcg_gq *blkg = blk_queue_root_blkg(disk->queue);
+		struct block_device *bdev = dev_to_bdev(dev);
+		struct blkcg_gq *blkg =
+			blk_queue_root_blkg(bdev->bd_disk->queue);
 		struct blkg_iostat tmp;
 		int cpu;
 
@@ -830,7 +830,7 @@ static void blkcg_fill_root_iostats(void)
 		for_each_possible_cpu(cpu) {
 			struct disk_stats *cpu_dkstats;
 
-			cpu_dkstats = per_cpu_ptr(part->bdev->bd_stats, cpu);
+			cpu_dkstats = per_cpu_ptr(bdev->bd_stats, cpu);
 			tmp.ios[BLKG_IOSTAT_READ] +=
 				cpu_dkstats->ios[STAT_READ];
 			tmp.ios[BLKG_IOSTAT_WRITE] +=
@@ -849,7 +849,6 @@ static void blkcg_fill_root_iostats(void)
 			blkg_iostat_set(&blkg->iostat.cur, &tmp);
 			u64_stats_update_end(&blkg->iostat.sync);
 		}
-		disk_put_part(part);
 	}
 }
 
block/blk.h +1 -1
index 9657c6da7c770c..98f0b1ae264120 100644
@@ -356,7 +356,7 @@ char *disk_name(struct gendisk *hd, int partno, char *buf);
 #define ADDPART_FLAG_NONE	0
 #define ADDPART_FLAG_RAID	1
 #define ADDPART_FLAG_WHOLEDISK	2
-void delete_partition(struct hd_struct *part);
+void delete_partition(struct block_device *part);
 int bdev_add_partition(struct block_device *bdev, int partno,
 		sector_t start, sector_t length);
 int bdev_del_partition(struct block_device *bdev, int partno);
block/genhd.c +26 -60

176 lines not on this page · raw message

block/partitions/core.c +45 -71

271 lines not on this page · raw message

fs/block_dev.c +0 -9
index 1538b20ca4bd43..08171478473ead 100644
@@ -39,7 +39,6 @@
 
 struct bdev_inode {
 	struct block_device bdev;
-	struct hd_struct hd;
 	struct inode vfs_inode;
 };
 
@@ -886,9 +885,6 @@ struct block_device *bdev_alloc(struct gendisk *disk, u8 partno)
 		iput(inode);
 		return NULL;
 	}
-	bdev->bd_part = &BDEV_I(inode)->hd;
-	memset(bdev->bd_part, 0, sizeof(*bdev->bd_part));
-	bdev->bd_part->bdev = bdev;
 	return bdev;
 }
 
@@ -921,11 +917,6 @@ struct block_device *bdgrab(struct block_device *bdev)
 }
 EXPORT_SYMBOL(bdgrab);
 
-struct block_device *bdget_part(struct hd_struct *part)
-{
-	return bdget(part_devt(part));
-}
-
 long nr_blockdev_pages(void)
 {
 	struct inode *inode;
include/linux/blk_types.h +6 -2
index 6edea5c1625909..866f74261b3ba8 100644
@@ -8,6 +8,7 @@
 
 #include <linux/types.h>
 #include <linux/bvec.h>
+#include <linux/device.h>
 #include <linux/ktime.h>
 
 struct bio_set;
@@ -30,6 +31,7 @@ struct block_device {
 	struct super_block *	bd_super;
 	struct mutex		bd_mutex;	/* open/close mutex */
 	void *			bd_claiming;
+	struct device		bd_device;
 	void *			bd_holder;
 	int			bd_holders;
 	bool			bd_write_holder;
@@ -38,7 +40,6 @@ struct block_device {
 #endif
 	struct kobject		*bd_holder_dir;
 	u8			bd_partno;
-	struct hd_struct *	bd_part;
 	/* number of times partitions within this device have been opened. */
 	unsigned		bd_part_count;
 
@@ -61,8 +62,11 @@ struct block_device {
 #define bdev_whole(_bdev) \
 	((_bdev)->bd_disk->part0)
 
+#define dev_to_bdev(device) \
+	container_of((device), struct block_device, bd_device)
+
 #define bdev_kobj(_bdev) \
-	(&part_to_dev((_bdev)->bd_part)->kobj)
+	(&((_bdev)->bd_device.kobj))
 
 /*
  * Block error status values.  See block/blk-core:blk_errors for the details.
include/linux/blkdev.h +0 -1
index 0e83989b9678c3..f390446919b745 100644
@@ -1995,7 +1995,6 @@ void blkdev_put(struct block_device *bdev, fmode_t mode);
 struct block_device *bdev_alloc(struct gendisk *disk, u8 partno);
 void bdev_add(struct block_device *bdev, dev_t dev);
 struct block_device *I_BDEV(struct inode *inode);
-struct block_device *bdget_part(struct hd_struct *part);
 struct block_device *bdgrab(struct block_device *bdev);
 void bdput(struct block_device *);
 
include/linux/genhd.h +9 -31
index de2ee5ffeefe45..769dc22f574797 100644
@@ -19,12 +19,6 @@
 #include <linux/blk_types.h>
 #include <asm/local.h>
 
-#define dev_to_part(device)	container_of((device), struct hd_struct, __dev)
-#define part_to_dev(part)	(&((part)->__dev))
-
-#define dev_to_disk(device)	(dev_to_part(device)->bdev->bd_disk)
-#define disk_to_dev(disk)	(part_to_dev((disk)->part0->bd_part))
-
 extern const struct device_type disk_type;
 extern struct device_type part_type;
 extern struct class block_class;
@@ -51,11 +45,6 @@ struct partition_meta_info {
 	u8 volname[PARTITION_META_INFO_VOLNAMELTH];
 };
 
-struct hd_struct {
-	struct block_device *bdev;
-	struct device __dev;
-};
-
 /**
  * DOC: genhd capability flags
  *
@@ -191,19 +180,21 @@ struct gendisk {
 	struct lockdep_map lockdep_map;
 };
 
+/*
+ * The gendisk is refcounted by the part0 block_device, and the bd_device
+ * therein is also used for device model presentation in sysfs.
+ */
+#define dev_to_disk(device) \
+	(dev_to_bdev(device)->bd_disk)
+#define disk_to_dev(disk) \
+	(&((disk)->part0->bd_device))
+
 #if IS_REACHABLE(CONFIG_CDROM)
 #define disk_to_cdi(disk)	((disk)->cdi)
 #else
 #define disk_to_cdi(disk)	NULL
 #endif
 
-static inline struct gendisk *part_to_disk(struct hd_struct *part)
-{
-	if (unlikely(!part))
-		return NULL;
-	return part->bdev->bd_disk;
-}
-
 static inline int disk_max_parts(struct gendisk *disk)
 {
 	if (disk->flags & GENHD_FL_EXT_DEVT)
@@ -222,19 +213,6 @@ static inline dev_t disk_devt(struct gendisk *disk)
 	return MKDEV(disk->major, disk->first_minor);
 }
 
-static inline dev_t part_devt(struct hd_struct *part)
-{
-	return part_to_dev(part)->devt;
-}
-
-extern struct hd_struct *disk_get_part(struct gendisk *disk, int partno);
-
-static inline void disk_put_part(struct hd_struct *part)
-{
-	if (likely(part))
-		put_device(part_to_dev(part));
-}
-
 /*
  * Smarter partition iterator without context limits.
  */
init/do_mounts.c +10 -11
index 86bef93e72ebd6..a78e44ee6adb8d 100644
@@ -76,11 +76,11 @@ struct uuidcmp {
  */
 static int match_dev_by_uuid(struct device *dev, const void *data)
 {
+	struct block_device *bdev = dev_to_bdev(dev);
 	const struct uuidcmp *cmp = data;
-	struct hd_struct *part = dev_to_part(dev);
 
-	if (!part->bdev->bd_meta_info ||
-	    strncasecmp(cmp->uuid, part->bdev->bd_meta_info->uuid, cmp->len))
+	if (!bdev->bd_meta_info ||
+	    strncasecmp(cmp->uuid, bdev->bd_meta_info->uuid, cmp->len))
 		return 0;
 	return 1;
 }
@@ -133,13 +133,13 @@ static dev_t devt_from_partuuid(const char *uuid_str)
 		 * Attempt to find the requested partition by adding an offset
 		 * to the partition number found by UUID.
 		 */
-		struct hd_struct *part;
+		struct block_device *part;
 
-		part = disk_get_part(dev_to_disk(dev),
-				     dev_to_part(dev)->bdev->bd_partno + offset);
+		part = bdget_disk(dev_to_disk(dev),
+				  dev_to_bdev(dev)->bd_partno + offset);
 		if (part) {
-			devt = part_devt(part);
-			put_device(part_to_dev(part));
+			devt = part->bd_dev;
+			bdput(part);
 		}
 	} else {
 		devt = dev->devt;
@@ -166,11 +166,10 @@ static dev_t devt_from_partuuid(const char *uuid_str)
  */
 static int match_dev_by_label(struct device *dev, const void *data)
 {
+	struct block_device *bdev = dev_to_bdev(dev);
 	const char *label = data;
-	struct hd_struct *part = dev_to_part(dev);
 
-	if (!part->bdev->bd_meta_info ||
-	    strcmp(label, part->bdev->bd_meta_info->volname))
+	if (!bdev->bd_meta_info || strcmp(label, bdev->bd_meta_info->volname))
 		return 0;
 	return 1;
 }
kernel/trace/blktrace.c +8 -35

80 lines not on this page · raw message

Signature · 1 line
2.29.2
author maintainer
#89 lore raw

PATCH 45/45 block: stop using bdget_disk for partition 0

In reply to Christoph Hellwig

We can just dereference the point in struct gendisk instead. Also remove the now unused export.

---
 block/genhd.c                   |  1 -
 drivers/block/nbd.c             |  4 +---
 drivers/block/xen-blkfront.c    | 20 +++++---------------
 drivers/block/zram/zram_drv.c   | 18 +++---------------
 drivers/md/dm.c                 | 16 ++--------------
 drivers/s390/block/dasd_ioctl.c |  5 ++---
 6 files changed, 13 insertions(+), 51 deletions(-)
block/genhd.c +0 -1
index e1f67b9951f0ec..9a3ad1895888c6 100644
@@ -912,7 +912,6 @@ struct block_device *bdget_disk(struct gendisk *disk, int partno)
 
 	return bdev;
 }
-EXPORT_SYMBOL(bdget_disk);
 
 /*
  * print a full list of all partitions - intended for places where the root
drivers/block/nbd.c +1 -3
index 014683968ce174..92f84ed0ba9eb6 100644
@@ -1488,12 +1488,10 @@ static int nbd_open(struct block_device *bdev, fmode_t mode)
 static void nbd_release(struct gendisk *disk, fmode_t mode)
 {
 	struct nbd_device *nbd = disk->private_data;
-	struct block_device *bdev = bdget_disk(disk, 0);
 
 	if (test_bit(NBD_RT_DISCONNECT_ON_CLOSE, &nbd->config->runtime_flags) &&
-			bdev->bd_openers == 0)
+			disk->part0->bd_openers == 0)
 		nbd_disconnect_and_put(nbd);
-	bdput(bdev);
 
 	nbd_config_put(nbd);
 	nbd_put(nbd);
drivers/block/xen-blkfront.c +5 -15
index 79521e33d30ed5..188e0b47534bcf 100644
@@ -2153,7 +2153,7 @@ static void blkfront_closing(struct blkfront_info *info)
 	}
 
 	if (info->gd)
-		bdev = bdget_disk(info->gd, 0);
+		bdev = bdgrab(info->gd->part0);
 
 	mutex_unlock(&info->mutex);
 
@@ -2518,7 +2518,7 @@ static int blkfront_remove(struct xenbus_device *xbdev)
 
 	disk = info->gd;
 	if (disk)
-		bdev = bdget_disk(disk, 0);
+		bdev = bdgrab(disk->part0);
 
 	info->xbdev = NULL;
 	mutex_unlock(&info->mutex);
@@ -2595,19 +2595,11 @@ static int blkif_open(struct block_device *bdev, fmode_t mode)
 static void blkif_release(struct gendisk *disk, fmode_t mode)
 {
 	struct blkfront_info *info = disk->private_data;
-	struct block_device *bdev;
 	struct xenbus_device *xbdev;
 
 	mutex_lock(&blkfront_mutex);
-
-	bdev = bdget_disk(disk, 0);
-
-	if (!bdev) {
-		WARN(1, "Block device %s yanked out from us!\n", disk->disk_name);
+	if (disk->part0->bd_openers)
 		goto out_mutex;
-	}
-	if (bdev->bd_openers)
-		goto out;
 
 	/*
 	 * Check if we have been instructed to close. We will have
@@ -2619,7 +2611,7 @@ static void blkif_release(struct gendisk *disk, fmode_t mode)
 
 	if (xbdev && xbdev->state == XenbusStateClosing) {
 		/* pending switch to state closed */
-		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
+		dev_info(disk_to_dev(disk), "releasing disk\n");
 		xlvbd_release_gendisk(info);
 		xenbus_frontend_closed(info->xbdev);
  	}
@@ -2628,14 +2620,12 @@ static void blkif_release(struct gendisk *disk, fmode_t mode)
 
 	if (!xbdev) {
 		/* sudden device removal */
-		dev_info(disk_to_dev(bdev->bd_disk), "releasing disk\n");
+		dev_info(disk_to_dev(disk), "releasing disk\n");
 		xlvbd_release_gendisk(info);
 		disk->private_data = NULL;
 		free_info(info);
 	}
 
-out:
-	bdput(bdev);
 out_mutex:
 	mutex_unlock(&blkfront_mutex);
 }
drivers/block/zram/zram_drv.c +3 -15
index 01757f9578dcb8..56024905bd242c 100644
@@ -1748,7 +1748,7 @@ static ssize_t reset_store(struct device *dev,
 		struct device_attribute *attr, const char *buf, size_t len)
 {
 	struct zram *zram = dev_to_zram(dev);
-	struct block_device *bdev;
+	struct block_device *bdev = zram->disk->part0;
 	unsigned short do_reset;
 	int ret = 0;
 
@@ -1758,17 +1758,12 @@ static ssize_t reset_store(struct device *dev,
 	if (!do_reset)
 		return -EINVAL;
 
-	bdev = bdget_disk(zram->disk, 0);
-	if (!bdev)
-		return -ENOMEM;
-
 	mutex_lock(&bdev->bd_mutex);
 	if (bdev->bd_openers)
 		ret = -EBUSY;
 	else
 		zram_reset_device(zram);
 	mutex_unlock(&bdev->bd_mutex);
-	bdput(bdev);
 
 	return ret ? ret : len;
 }
@@ -1933,15 +1928,8 @@ static int zram_add(void)
 
 static int zram_remove(struct zram *zram)
 {
-	struct block_device *bdev = bdget_disk(zram->disk, 0);
-
-	if (bdev) {
-		if (bdev->bd_openers) {
-			bdput(bdev);
-			return -EBUSY;
-		}
-		bdput(bdev);
-	}
+	if (zram->disk->part0->bd_openers)
+		return -EBUSY;
 
 	del_gendisk(zram->disk);
 	zram_debugfs_unregister(zram);
drivers/md/dm.c +2 -14
index 176adcff56b380..ed7e836efbcdbc 100644
@@ -2375,16 +2375,11 @@ struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
  */
 static int lock_fs(struct mapped_device *md)
 {
-	struct block_device *bdev;
 	int r;
 
 	WARN_ON(test_bit(DMF_FROZEN, &md->flags));
 
-	bdev = bdget_disk(md->disk, 0);
-	if (!bdev)
-		return -ENOMEM;
-	r = freeze_bdev(bdev);
-	bdput(bdev);
+	r = freeze_bdev(md->disk->part0);
 	if (!r)
 		set_bit(DMF_FROZEN, &md->flags);
 	return r;
@@ -2392,16 +2387,9 @@ static int lock_fs(struct mapped_device *md)
 
 static void unlock_fs(struct mapped_device *md)
 {
-	struct block_device *bdev;
-
 	if (!test_bit(DMF_FROZEN, &md->flags))
 		return;
-
-	bdev = bdget_disk(md->disk, 0);
-	if (!bdev)
-		return;
-	thaw_bdev(bdev);
-	bdput(bdev);
+	thaw_bdev(md->disk->part0);
 	clear_bit(DMF_FROZEN, &md->flags);
 }
 
drivers/s390/block/dasd_ioctl.c +2 -3
index 304eba1acf163c..9f642440894655 100644
@@ -220,9 +220,8 @@ dasd_format(struct dasd_block *block, struct format_data_t *fdata)
 	 * enabling the device later.
 	 */
 	if (fdata->start_unit == 0) {
-		struct block_device *bdev = bdget_disk(block->gdp, 0);
-		bdev->bd_inode->i_blkbits = blksize_bits(fdata->blksize);
-		bdput(bdev);
+		block->gdp->part0->bd_inode->i_blkbits =
+			blksize_bits(fdata->blksize);
 	}
 
 	rc = base->discipline->format_device(base, fdata, 1);
Signature · 1 line
2.29.2