Skip to contentSkip to conversation

ceph: set FMODE_CAN_ODIRECT instead of a dummy direct_IO method

PATCH In review Christoph Hellwig 2 participants 4 messages Jun 12, 2023 – Jun 13, 2023 Ceph client

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ceph: set FMODE_CAN_ODIRECT instead of a dummy direct_IO method

Since commit a2ad63daa88b ("VFS: add FMODE_CAN_ODIRECT file flag") file systems can just set the FMODE_CAN_ODIRECT flag at open time instead of wiring up a dummy direct_IO method to indicate support for direct I/O. Do that for ceph so that noop_direct_IO can eventually be removed.

---
 fs/ceph/addr.c | 1 -
 fs/ceph/file.c | 2 ++
 2 files changed, 2 insertions(+), 1 deletion(-)
fs/ceph/addr.c +0 -1
index 6bb251a4d613eb..19c0f42540b600 100644
@@ -1401,7 +1401,6 @@ const struct address_space_operations ceph_aops = {
 	.dirty_folio = ceph_dirty_folio,
 	.invalidate_folio = ceph_invalidate_folio,
 	.release_folio = ceph_release_folio,
-	.direct_IO = noop_direct_IO,
 };
 
 static void ceph_block_sigs(sigset_t *oldset)
fs/ceph/file.c +2 -0
index f4d8bf7dec88a8..314c5d5971bf4a 100644
@@ -368,6 +368,8 @@ int ceph_open(struct inode *inode, struct file *file)
 	flags = file->f_flags & ~(O_CREAT|O_EXCL);
 	if (S_ISDIR(inode->i_mode))
 		flags = O_DIRECTORY;  /* mds likes to know */
+	if (S_ISREG(inode->i_mode))
+		file->f_mode |= FMODE_CAN_ODIRECT;
 
 	dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
 	     ceph_vinop(inode), file, flags, file->f_flags);
Signature · 1 line
2.39.2
#1 lore raw
Quoted 28 lines · Christoph Hellwig

Since commit a2ad63daa88b ("VFS: add FMODE_CAN_ODIRECT file flag") file systems can just set the FMODE_CAN_ODIRECT flag at open time instead of wiring up a dummy direct_IO method to indicate support for direct I/O. Do that for ceph so that noop_direct_IO can eventually be removed.

---
  fs/ceph/addr.c | 1 -
  fs/ceph/file.c | 2 ++
  2 files changed, 2 insertions(+), 1 deletion(-)
fs/ceph/addr.c +0 -1
index 6bb251a4d613eb..19c0f42540b600 100644
@@ -1401,7 +1401,6 @@ const struct address_space_operations ceph_aops = {
  	.dirty_folio = ceph_dirty_folio,
  	.invalidate_folio = ceph_invalidate_folio,
  	.release_folio = ceph_release_folio,
-	.direct_IO = noop_direct_IO,
  };
 
  static void ceph_block_sigs(sigset_t *oldset)
fs/ceph/file.c +0 -0
index f4d8bf7dec88a8..314c5d5971bf4a 100644

fs/ceph/file.c int ceph_open(struct inode *inode, struct file *file) · line 373

Quoted 6 lines
fs/ceph/file.c +2 -0
@@ -368,6 +368,8 @@ int ceph_open(struct inode *inode, struct file *file)
  	flags = file->f_flags & ~(O_CREAT|O_EXCL);
  	if (S_ISDIR(inode->i_mode))
  		flags = O_DIRECTORY;  /* mds likes to know */
+	if (S_ISREG(inode->i_mode))
+		file->f_mode |= FMODE_CAN_ODIRECT;
 

Shouldn't we do the same in 'ceph_atomic_open()' too ?

Thanks

- Xiubo

Quoted 2 lines
fs/ceph/file.c +0 -0
  	dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
  	     ceph_vinop(inode), file, flags, file->f_flags);
#2 lore raw
In reply to Christoph Hellwig
Quoted 28 lines · Christoph Hellwig

Since commit a2ad63daa88b ("VFS: add FMODE_CAN_ODIRECT file flag") file systems can just set the FMODE_CAN_ODIRECT flag at open time instead of wiring up a dummy direct_IO method to indicate support for direct I/O. Do that for ceph so that noop_direct_IO can eventually be removed.

---
  fs/ceph/addr.c | 1 -
  fs/ceph/file.c | 2 ++
  2 files changed, 2 insertions(+), 1 deletion(-)
fs/ceph/addr.c +0 -1
index 6bb251a4d613eb..19c0f42540b600 100644
@@ -1401,7 +1401,6 @@ const struct address_space_operations ceph_aops = {
  	.dirty_folio = ceph_dirty_folio,
  	.invalidate_folio = ceph_invalidate_folio,
  	.release_folio = ceph_release_folio,
-	.direct_IO = noop_direct_IO,
  };
 
  static void ceph_block_sigs(sigset_t *oldset)
fs/ceph/file.c +0 -0
index f4d8bf7dec88a8..314c5d5971bf4a 100644

fs/ceph/file.c int ceph_open(struct inode *inode, struct file *file) · line 371

Quoted 4 lines
fs/ceph/file.c +1 -0
@@ -368,6 +368,8 @@ int ceph_open(struct inode *inode, struct file *file)
  	flags = file->f_flags & ~(O_CREAT|O_EXCL);
  	if (S_ISDIR(inode->i_mode))
  		flags = O_DIRECTORY;  /* mds likes to know */
+	if (S_ISREG(inode->i_mode))

BTW, the commit a2ad63daa88b ("VFS: add FMODE_CAN_ODIRECT file flag") doesn't check the S_ISREG, and I couldn't see this commit and NFS confine it to regular files, is that okay ?

Thanks

- Xiubo

Quoted 4 lines
fs/ceph/file.c +1 -0
+		file->f_mode |= FMODE_CAN_ODIRECT;
 
  	dout("open inode %p ino %llx.%llx file %p flags %d (%d)\n", inode,
  	     ceph_vinop(inode), file, flags, file->f_flags);
#3 lore raw
Quoted 5 lines · Xiubo Li

fs/ceph/file.c int ceph_open(struct inode *inode, struct file *file) · line 372

Quoted 3 lines
fs/ceph/file.c +2 -0
+	if (S_ISREG(inode->i_mode))
+		file->f_mode |= FMODE_CAN_ODIRECT;

Shouldn't we do the same in 'ceph_atomic_open()' too ?

Yes, probably. Or I really need to press ahead and move the flag to file operations..

Quoted 11 lines

fs/ceph/file.c int ceph_open(struct inode *inode, struct file *file) · line 371

Quoted 7 lines
fs/ceph/file.c +1 -0
@@ -368,6 +368,8 @@ int ceph_open(struct inode *inode, struct file *file)
  	flags = file->f_flags & ~(O_CREAT|O_EXCL);
  	if (S_ISDIR(inode->i_mode))
  		flags = O_DIRECTORY;  /* mds likes to know */
+	if (S_ISREG(inode->i_mode))

BTW, the commit a2ad63daa88b ("VFS: add FMODE_CAN_ODIRECT file flag") doesn't check the S_ISREG, and I couldn't see this commit and NFS confine it to regular files, is that okay ?

It doesn't have to. ->direct_IO was previously set only for regular files (and block devices in the block code). So it makes sense to do the same for the flag.