Skip to contentSkip to conversation

vfs: don't unnecessarily clone write access for writable fds

PATCH Superseded Eric Biggers 2 participants 5 messages Jun 11, 2020 – Sep 22, 2020 VFS

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 vfs: don't unnecessarily clone write access for writable fds

From: Eric Biggers <[email protected]>

There's no need for mnt_want_write_file() to increment mnt_writers when the file is already open for writing, provided that mnt_drop_write_file() is changed to conditionally decrement it.

We seem to have ended up in the current situation because mnt_want_write_file() used to be paired with mnt_drop_write(), due to mnt_drop_write_file() not having been added yet. So originally mnt_want_write_file() had to always increment mnt_writers.

But later mnt_drop_write_file() was added, and all callers of mnt_want_write_file() were paired with it. This makes the compatibility between mnt_want_write_file() and mnt_drop_write() no longer necessary.

Therefore, make __mnt_want_write_file() and __mnt_drop_write_file() skip incrementing mnt_writers on files already open for writing. This removes the only caller of mnt_clone_write(), so remove that too.

---

v2: keep the check for emergency r/o remounts.

 fs/namespace.c        | 53 ++++++++++++++++---------------------------
 include/linux/mount.h |  1 -
 2 files changed, 20 insertions(+), 34 deletions(-)
fs/namespace.c +20 -33
index f30ed401cc6d7a..b2052b2f3d87f6 100644
@@ -359,51 +359,37 @@ int mnt_want_write(struct vfsmount *m)
 }
 EXPORT_SYMBOL_GPL(mnt_want_write);
 
-/**
- * mnt_clone_write - get write access to a mount
- * @mnt: the mount on which to take a write
- *
- * This is effectively like mnt_want_write, except
- * it must only be used to take an extra write reference
- * on a mountpoint that we already know has a write reference
- * on it. This allows some optimisation.
- *
- * After finished, mnt_drop_write must be called as usual to
- * drop the reference.
- */
-int mnt_clone_write(struct vfsmount *mnt)
-{
-	/* superblock may be r/o */
-	if (__mnt_is_readonly(mnt))
-		return -EROFS;
-	preempt_disable();
-	mnt_inc_writers(real_mount(mnt));
-	preempt_enable();
-	return 0;
-}
-EXPORT_SYMBOL_GPL(mnt_clone_write);
-
 /**
  * __mnt_want_write_file - get write access to a file's mount
  * @file: the file who's mount on which to take a write
  *
- * This is like __mnt_want_write, but it takes a file and can
- * do some optimisations if the file is open for write already
+ * This is like __mnt_want_write, but if the file is already open for writing it
+ * skips incrementing mnt_writers (since the open file already has a reference)
+ * and instead only does the check for emergency r/o remounts.  This must be
+ * paired with __mnt_drop_write_file.
  */
 int __mnt_want_write_file(struct file *file)
 {
-	if (!(file->f_mode & FMODE_WRITER))
-		return __mnt_want_write(file->f_path.mnt);
-	else
-		return mnt_clone_write(file->f_path.mnt);
+	if (file->f_mode & FMODE_WRITER) {
+		/*
+		 * Superblock may have become readonly while there are still
+		 * writable fd's, e.g. due to a fs error with errors=remount-ro
+		 */
+		if (__mnt_is_readonly(file->f_path.mnt))
+			return -EROFS;
+		return 0;
+	}
+	return __mnt_want_write(file->f_path.mnt);
 }
 
 /**
  * mnt_want_write_file - get write access to a file's mount
  * @file: the file who's mount on which to take a write
  *
- * This is like mnt_want_write, but it takes a file and can
- * do some optimisations if the file is open for write already
+ * This is like mnt_want_write, but if the file is already open for writing it
+ * skips incrementing mnt_writers (since the open file already has a reference)
+ * and instead only does the freeze protection and the check for emergency r/o
+ * remounts.  This must be paired with mnt_drop_write_file.
  */
 int mnt_want_write_file(struct file *file)
 {
@@ -449,7 +435,8 @@ EXPORT_SYMBOL_GPL(mnt_drop_write);
 
 void __mnt_drop_write_file(struct file *file)
 {
-	__mnt_drop_write(file->f_path.mnt);
+	if (!(file->f_mode & FMODE_WRITER))
+		__mnt_drop_write(file->f_path.mnt);
 }
 
 void mnt_drop_write_file(struct file *file)
include/linux/mount.h +0 -1
index de657bd211fa64..29d216f927c28c 100644
@@ -78,7 +78,6 @@ struct path;
 
 extern int mnt_want_write(struct vfsmount *mnt);
 extern int mnt_want_write_file(struct file *file);
-extern int mnt_clone_write(struct vfsmount *mnt);
 extern void mnt_drop_write(struct vfsmount *mnt);
 extern void mnt_drop_write_file(struct file *file);
 extern void mntput(struct vfsmount *mnt);
Signature · 1 line
2.26.2
#1 lore raw
Quoted 15 lines · Eric Biggers

From: Eric Biggers <[email protected]>

There's no need for mnt_want_write_file() to increment mnt_writers when the file is already open for writing, provided that mnt_drop_write_file() is changed to conditionally decrement it.

We seem to have ended up in the current situation because mnt_want_write_file() used to be paired with mnt_drop_write(), due to mnt_drop_write_file() not having been added yet. So originally mnt_want_write_file() had to always increment mnt_writers.

But later mnt_drop_write_file() was added, and all callers of mnt_want_write_file() were paired with it. This makes the compatibility between mnt_want_write_file() and mnt_drop_write() no longer necessary.

Quoted 5 lines

Therefore, make __mnt_want_write_file() and __mnt_drop_write_file() skip incrementing mnt_writers on files already open for writing. This removes the only caller of mnt_clone_write(), so remove that too.

Al, any thoughts on this patch?

- Eric

#2 lore raw
Quoted 16 lines · Eric Biggers
Quoted 15 lines · Eric Biggers

From: Eric Biggers <[email protected]>

There's no need for mnt_want_write_file() to increment mnt_writers when the file is already open for writing, provided that mnt_drop_write_file() is changed to conditionally decrement it.

We seem to have ended up in the current situation because mnt_want_write_file() used to be paired with mnt_drop_write(), due to mnt_drop_write_file() not having been added yet. So originally mnt_want_write_file() had to always increment mnt_writers.

But later mnt_drop_write_file() was added, and all callers of mnt_want_write_file() were paired with it. This makes the compatibility between mnt_want_write_file() and mnt_drop_write() no longer necessary.

Quoted 8 lines
Quoted 5 lines

Therefore, make __mnt_want_write_file() and __mnt_drop_write_file() skip incrementing mnt_writers on files already open for writing. This removes the only caller of mnt_clone_write(), so remove that too.

Al, any thoughts on this patch?

Ping?

maintainer
#3 lore raw
Quoted 8 lines · Eric Biggers
Quoted 7 lines · Eric Biggers
Quoted 6 lines · Eric Biggers

From: Eric Biggers <[email protected]>

There's no need for mnt_want_write_file() to increment mnt_writers when the file is already open for writing, provided that mnt_drop_write_file() is changed to conditionally decrement it.

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

We seem to have ended up in the current situation because mnt_want_write_file() used to be paired with mnt_drop_write(), due to mnt_drop_write_file() not having been added yet. So originally mnt_want_write_file() had to always increment mnt_writers.

But later mnt_drop_write_file() was added, and all callers of mnt_want_write_file() were paired with it. This makes the compatibility between mnt_want_write_file() and mnt_drop_write() no longer necessary.

Umm... That really needs to be put into D/f/porting; this kind of rule changes (from "it used to work both ways" to "things quietly break if you use the old variant") should come with explicit statement in there.

I'm certainly fine with unexporting mnt_clone_write() and making the damn thing static, but as for the rest I would put an explicit "don't pair mnt_drop_write() with mnt_want_write_file()" and wait for a cycle.

#4 lore raw
Quoted 18 lines · Al Viro
Quoted 8 lines · Eric Biggers
Quoted 7 lines · Eric Biggers
Quoted 6 lines · Eric Biggers

From: Eric Biggers <[email protected]>

There's no need for mnt_want_write_file() to increment mnt_writers when the file is already open for writing, provided that mnt_drop_write_file() is changed to conditionally decrement it.

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

We seem to have ended up in the current situation because mnt_want_write_file() used to be paired with mnt_drop_write(), due to mnt_drop_write_file() not having been added yet. So originally mnt_want_write_file() had to always increment mnt_writers.

But later mnt_drop_write_file() was added, and all callers of mnt_want_write_file() were paired with it. This makes the compatibility between mnt_want_write_file() and mnt_drop_write() no longer necessary.

Quoted 7 lines

Umm... That really needs to be put into D/f/porting; this kind of rule changes (from "it used to work both ways" to "things quietly break if you use the old variant") should come with explicit statement in there.

I'm certainly fine with unexporting mnt_clone_write() and making the damn thing static, but as for the rest I would put an explicit "don't pair mnt_drop_write() with mnt_want_write_file()" and wait for a cycle.

Is there any point in waiting a cycle between adding the note to Documentation/filesystems/porting.rst and making the behavior change? It seems that all the other notes just get added at the same time the change is made.

- Eric