Skip to contentSkip to conversation

Use inode_lock/unlock class of provided APIs in filesystems

PATCH RESEND In review Ritesh Harjani 3 participants 7 messages Jan 1, 2020 – May 1, 2020

Subsystems (7)
lore
Share & tools
lore
#0 lore raw

cover Use inode_lock/unlock class of provided APIs in filesystems

Al, any comments?
Resending this after adding Reviewed-by/Acked-by tags.

From previous version:-
Matthew Wilcox in [1] suggested that it will be a good idea to define some missing API instead of directly using i_rwsem in filesystems drivers for lock/unlock/downgrade purposes.

This patch does that work. No functionality change in this patch.

After this there are only lockdep class of APIs at certain places in filesystems which are directly using i_rwsem and second is XFS, but it seems to be anyway defining it's own xfs_ilock/iunlock set of APIs and 'iolock' naming convention for this lock.

[1]: https://www.spinics.net/lists/linux-ext4/msg68689.html

Ritesh Harjani (1):
fs: Use inode_lock/unlock class of provided APIs in filesystems

 fs/btrfs/delayed-inode.c |  2 +-
 fs/btrfs/ioctl.c         |  4 ++--
 fs/ceph/io.c             | 24 ++++++++++++------------
 fs/nfs/io.c              | 24 ++++++++++++------------
 fs/orangefs/file.c       |  4 ++--
 fs/overlayfs/readdir.c   |  2 +-
 fs/readdir.c             |  4 ++--
 include/linux/fs.h       | 21 +++++++++++++++++++++
 8 files changed, 53 insertions(+), 32 deletions(-)
Signature · 1 line
2.21.0
#1 lore raw

PATCH RESEND 1/1 fs: Use inode_lock/unlock class of provided APIs in filesystems

This defines 4 more APIs which some of the filesystem needs and reduces the direct use of i_rwsem in filesystem drivers. Instead those are replaced with inode_lock/unlock_** APIs.

---
 fs/btrfs/delayed-inode.c |  2 +-
 fs/btrfs/ioctl.c         |  4 ++--
 fs/ceph/io.c             | 24 ++++++++++++------------
 fs/nfs/io.c              | 24 ++++++++++++------------
 fs/orangefs/file.c       |  4 ++--
 fs/overlayfs/readdir.c   |  2 +-
 fs/readdir.c             |  4 ++--
 include/linux/fs.h       | 21 +++++++++++++++++++++
 8 files changed, 53 insertions(+), 32 deletions(-)
fs/btrfs/delayed-inode.c +1 -1
index d3e15e1d4a91..c3e92f2fd915 100644
@@ -1644,7 +1644,7 @@ void btrfs_readdir_put_delayed_items(struct inode *inode,
 	 * The VFS is going to do up_read(), so we need to downgrade back to a
 	 * read lock.
 	 */
-	downgrade_write(&inode->i_rwsem);
+	inode_lock_downgrade(inode);
 }
 
 int btrfs_should_delete_dir_index(struct list_head *del_list,
fs/btrfs/ioctl.c +2 -2
index 18e328ce4b54..caa80372992d 100644
@@ -961,7 +961,7 @@ static noinline int btrfs_mksubvol(const struct path *parent,
 	struct dentry *dentry;
 	int error;
 
-	error = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
+	error = inode_lock_killable_nested(dir, I_MUTEX_PARENT);
 	if (error == -EINTR)
 		return error;
 
@@ -2869,7 +2869,7 @@ static noinline int btrfs_ioctl_snap_destroy(struct file *file,
 		goto out;
 
 
-	err = down_write_killable_nested(&dir->i_rwsem, I_MUTEX_PARENT);
+	err = inode_lock_killable_nested(dir, I_MUTEX_PARENT);
 	if (err == -EINTR)
 		goto out_drop_write;
 	dentry = lookup_one_len(vol_args->name, parent, namelen);
fs/ceph/io.c +12 -12
index 97602ea92ff4..e94186259c2e 100644
@@ -53,14 +53,14 @@ ceph_start_io_read(struct inode *inode)
 	struct ceph_inode_info *ci = ceph_inode(inode);
 
 	/* Be an optimist! */
-	down_read(&inode->i_rwsem);
+	inode_lock_shared(inode);
 	if (!(READ_ONCE(ci->i_ceph_flags) & CEPH_I_ODIRECT))
 		return;
-	up_read(&inode->i_rwsem);
+	inode_unlock_shared(inode);
 	/* Slow path.... */
-	down_write(&inode->i_rwsem);
+	inode_lock(inode);
 	ceph_block_o_direct(ci, inode);
-	downgrade_write(&inode->i_rwsem);
+	inode_lock_downgrade(inode);
 }
 
 /**
@@ -73,7 +73,7 @@ ceph_start_io_read(struct inode *inode)
 void
 ceph_end_io_read(struct inode *inode)
 {
-	up_read(&inode->i_rwsem);
+	inode_unlock_shared(inode);
 }
 
 /**
@@ -86,7 +86,7 @@ ceph_end_io_read(struct inode *inode)
 void
 ceph_start_io_write(struct inode *inode)
 {
-	down_write(&inode->i_rwsem);
+	inode_lock(inode);
 	ceph_block_o_direct(ceph_inode(inode), inode);
 }
 
@@ -100,7 +100,7 @@ ceph_start_io_write(struct inode *inode)
 void
 ceph_end_io_write(struct inode *inode)
 {
-	up_write(&inode->i_rwsem);
+	inode_unlock(inode);
 }
 
 /* Call with exclusively locked inode->i_rwsem */
@@ -139,14 +139,14 @@ ceph_start_io_direct(struct inode *inode)
 	struct ceph_inode_info *ci = ceph_inode(inode);
 
 	/* Be an optimist! */
-	down_read(&inode->i_rwsem);
+	inode_lock_shared(inode);
 	if (READ_ONCE(ci->i_ceph_flags) & CEPH_I_ODIRECT)
 		return;
-	up_read(&inode->i_rwsem);
+	inode_unlock_shared(inode);
 	/* Slow path.... */
-	down_write(&inode->i_rwsem);
+	inode_lock(inode);
 	ceph_block_buffered(ci, inode);
-	downgrade_write(&inode->i_rwsem);
+	inode_lock_downgrade(inode);
 }
 
 /**
@@ -159,5 +159,5 @@ ceph_start_io_direct(struct inode *inode)
 void
 ceph_end_io_direct(struct inode *inode)
 {
-	up_read(&inode->i_rwsem);
+	inode_unlock_shared(inode);
 }
fs/nfs/io.c +12 -12
index 5088fda9b453..bf5ed7bea59d 100644
@@ -44,14 +44,14 @@ nfs_start_io_read(struct inode *inode)
 {
 	struct nfs_inode *nfsi = NFS_I(inode);
 	/* Be an optimist! */
-	down_read(&inode->i_rwsem);
+	inode_lock_shared(inode);
 	if (test_bit(NFS_INO_ODIRECT, &nfsi->flags) == 0)
 		return;
-	up_read(&inode->i_rwsem);
+	inode_unlock_shared(inode);
 	/* Slow path.... */
-	down_write(&inode->i_rwsem);
+	inode_lock(inode);
 	nfs_block_o_direct(nfsi, inode);
-	downgrade_write(&inode->i_rwsem);
+	inode_lock_downgrade(inode);
 }
 
 /**
@@ -64,7 +64,7 @@ nfs_start_io_read(struct inode *inode)
 void
 nfs_end_io_read(struct inode *inode)
 {
-	up_read(&inode->i_rwsem);
+	inode_unlock_shared(inode);
 }
 
 /**
@@ -77,7 +77,7 @@ nfs_end_io_read(struct inode *inode)
 void
 nfs_start_io_write(struct inode *inode)
 {
-	down_write(&inode->i_rwsem);
+	inode_lock(inode);
 	nfs_block_o_direct(NFS_I(inode), inode);
 }
 
@@ -91,7 +91,7 @@ nfs_start_io_write(struct inode *inode)
 void
 nfs_end_io_write(struct inode *inode)
 {
-	up_write(&inode->i_rwsem);
+	inode_unlock(inode);
 }
 
 /* Call with exclusively locked inode->i_rwsem */
@@ -124,14 +124,14 @@ nfs_start_io_direct(struct inode *inode)
 {
 	struct nfs_inode *nfsi = NFS_I(inode);
 	/* Be an optimist! */
-	down_read(&inode->i_rwsem);
+	inode_lock_shared(inode);
 	if (test_bit(NFS_INO_ODIRECT, &nfsi->flags) != 0)
 		return;
-	up_read(&inode->i_rwsem);
+	inode_unlock_shared(inode);
 	/* Slow path.... */
-	down_write(&inode->i_rwsem);
+	inode_lock(inode);
 	nfs_block_buffered(nfsi, inode);
-	downgrade_write(&inode->i_rwsem);
+	inode_lock_downgrade(inode);
 }
 
 /**
@@ -144,5 +144,5 @@ nfs_start_io_direct(struct inode *inode)
 void
 nfs_end_io_direct(struct inode *inode)
 {
-	up_read(&inode->i_rwsem);
+	inode_unlock_shared(inode);
 }
fs/orangefs/file.c +2 -2
index c740159d9ad1..369ace12249d 100644
@@ -363,14 +363,14 @@ static ssize_t orangefs_file_read_iter(struct kiocb *iocb,
 		ro->blksiz = iter->count;
 	}
 
-	down_read(&file_inode(iocb->ki_filp)->i_rwsem);
+	inode_lock_shared(file_inode(iocb->ki_filp));
 	ret = orangefs_revalidate_mapping(file_inode(iocb->ki_filp));
 	if (ret)
 		goto out;
 
 	ret = generic_file_read_iter(iocb, iter);
 out:
-	up_read(&file_inode(iocb->ki_filp)->i_rwsem);
+	inode_unlock_shared(file_inode(iocb->ki_filp));
 	return ret;
 }
 
fs/overlayfs/readdir.c +1 -1
index 47a91c9733a5..c203e73160b0 100644
@@ -273,7 +273,7 @@ static int ovl_check_whiteouts(struct dentry *dir, struct ovl_readdir_data *rdd)
 
 	old_cred = ovl_override_creds(rdd->dentry->d_sb);
 
-	err = down_write_killable(&dir->d_inode->i_rwsem);
+	err = inode_lock_killable(dir->d_inode);
 	if (!err) {
 		while (rdd->first_maybe_whiteout) {
 			p = rdd->first_maybe_whiteout;
fs/readdir.c +2 -2
index d26d5ea4de7b..10a34efa0af0 100644
@@ -52,9 +52,9 @@ int iterate_dir(struct file *file, struct dir_context *ctx)
 		goto out;
 
 	if (shared)
-		res = down_read_killable(&inode->i_rwsem);
+		res = inode_lock_shared_killable(inode);
 	else
-		res = down_write_killable(&inode->i_rwsem);
+		res = inode_lock_killable(inode);
 	if (res)
 		goto out;
 
include/linux/fs.h +21 -0
index 98e0349adb52..2b407464fac1 100644
@@ -831,6 +831,27 @@ static inline void inode_lock_shared_nested(struct inode *inode, unsigned subcla
 	down_read_nested(&inode->i_rwsem, subclass);
 }
 
+static inline void inode_lock_downgrade(struct inode *inode)
+{
+	downgrade_write(&inode->i_rwsem);
+}
+
+static inline int inode_lock_killable(struct inode *inode)
+{
+	return down_write_killable(&inode->i_rwsem);
+}
+
+static inline int inode_lock_shared_killable(struct inode *inode)
+{
+	return down_read_killable(&inode->i_rwsem);
+}
+
+static inline int inode_lock_killable_nested(struct inode *inode,
+					     unsigned subclass)
+{
+	return down_write_killable_nested(&inode->i_rwsem, subclass);
+}
+
 void lock_two_nondirectories(struct inode *, struct inode*);
 void unlock_two_nondirectories(struct inode *, struct inode*);
 
Signature · 1 line
2.21.0
maintainer
#2 lore raw
In reply to Ritesh Harjani

Ritesh -

I just loaded your patch on top of 5.5-rc4 and it looks fine to me and xfstests :-) ... I pointed ftrace at the orangefs function you modified while xfstests was running, and it got called about a jillion times...

-Mike

Quoted 35 lines · Ritesh Harjani

Al, any comments?
Resending this after adding Reviewed-by/Acked-by tags.

From previous version:-
Matthew Wilcox in [1] suggested that it will be a good idea to define some missing API instead of directly using i_rwsem in filesystems drivers for lock/unlock/downgrade purposes.

This patch does that work. No functionality change in this patch.

After this there are only lockdep class of APIs at certain places in filesystems which are directly using i_rwsem and second is XFS, but it seems to be anyway defining it's own xfs_ilock/iunlock set of APIs and 'iolock' naming convention for this lock.

[1]: https://www.spinics.net/lists/linux-ext4/msg68689.html

Ritesh Harjani (1):
fs: Use inode_lock/unlock class of provided APIs in filesystems

 fs/btrfs/delayed-inode.c |  2 +-
 fs/btrfs/ioctl.c         |  4 ++--
 fs/ceph/io.c             | 24 ++++++++++++------------
 fs/nfs/io.c              | 24 ++++++++++++------------
 fs/orangefs/file.c       |  4 ++--
 fs/overlayfs/readdir.c   |  2 +-
 fs/readdir.c             |  4 ++--
 include/linux/fs.h       | 21 +++++++++++++++++++++
 8 files changed, 53 insertions(+), 32 deletions(-)

--
2.21.0

#3 lore raw
Quoted 6 lines · Mike Marshall

Ritesh -

I just loaded your patch on top of 5.5-rc4 and it looks fine to me and xfstests :-) ... I pointed ftrace at the orangefs function you modified while xfstests was running, and it got called about a jillion times...

Thanks Mike for testing this. Shall I add your Tested-by?

-ritesh

Quoted 39 lines

-Mike

Quoted 35 lines · Ritesh Harjani

Al, any comments?
Resending this after adding Reviewed-by/Acked-by tags.

From previous version:-
Matthew Wilcox in [1] suggested that it will be a good idea to define some missing API instead of directly using i_rwsem in filesystems drivers for lock/unlock/downgrade purposes.

This patch does that work. No functionality change in this patch.

After this there are only lockdep class of APIs at certain places in filesystems which are directly using i_rwsem and second is XFS, but it seems to be anyway defining it's own xfs_ilock/iunlock set of APIs and 'iolock' naming convention for this lock.

[1]: https://www.spinics.net/lists/linux-ext4/msg68689.html

Ritesh Harjani (1):
fs: Use inode_lock/unlock class of provided APIs in filesystems

  fs/btrfs/delayed-inode.c |  2 +-
  fs/btrfs/ioctl.c         |  4 ++--
  fs/ceph/io.c             | 24 ++++++++++++------------
  fs/nfs/io.c              | 24 ++++++++++++------------
  fs/orangefs/file.c       |  4 ++--
  fs/overlayfs/readdir.c   |  2 +-
  fs/readdir.c             |  4 ++--
  include/linux/fs.h       | 21 +++++++++++++++++++++
  8 files changed, 53 insertions(+), 32 deletions(-)

--
2.21.0

maintainer
#4 lore raw

Yes, please do.

-Mike

Quoted 56 lines · Ritesh Harjani
Quoted 6 lines · Mike Marshall

Ritesh -

I just loaded your patch on top of 5.5-rc4 and it looks fine to me and xfstests :-) ... I pointed ftrace at the orangefs function you modified while xfstests was running, and it got called about a jillion times...

Thanks Mike for testing this. Shall I add your Tested-by?

-ritesh

Quoted 39 lines

-Mike

Quoted 35 lines · Ritesh Harjani

Al, any comments?
Resending this after adding Reviewed-by/Acked-by tags.

From previous version:-
Matthew Wilcox in [1] suggested that it will be a good idea to define some missing API instead of directly using i_rwsem in filesystems drivers for lock/unlock/downgrade purposes.

This patch does that work. No functionality change in this patch.

After this there are only lockdep class of APIs at certain places in filesystems which are directly using i_rwsem and second is XFS, but it seems to be anyway defining it's own xfs_ilock/iunlock set of APIs and 'iolock' naming convention for this lock.

[1]: https://www.spinics.net/lists/linux-ext4/msg68689.html

Ritesh Harjani (1):
fs: Use inode_lock/unlock class of provided APIs in filesystems

  fs/btrfs/delayed-inode.c |  2 +-
  fs/btrfs/ioctl.c         |  4 ++--
  fs/ceph/io.c             | 24 ++++++++++++------------
  fs/nfs/io.c              | 24 ++++++++++++------------
  fs/orangefs/file.c       |  4 ++--
  fs/overlayfs/readdir.c   |  2 +-
  fs/readdir.c             |  4 ++--
  include/linux/fs.h       | 21 +++++++++++++++++++++
  8 files changed, 53 insertions(+), 32 deletions(-)

--
2.21.0

maintainer
#5 lore raw
In reply to Ritesh Harjani
Quoted 2 lines · Ritesh Harjani

Al, any comments?
Resending this after adding Reviewed-by/Acked-by tags.

.... argh. My apologies - that got fallen through the cracks. Could you rebase and resend it?

#6 lore raw
Quoted 7 lines · Al Viro
Quoted 2 lines · Ritesh Harjani

Al, any comments?
Resending this after adding Reviewed-by/Acked-by tags.

.... argh. My apologies - that got fallen through the cracks. Could you rebase and resend it?

Np.
Sure, will rebase and resend.

-ritesh