fsnotify: optimize FMODE_NONOTIFY_PERM for the common cases

The most unlikely watched permission event is FAN_ACCESS_PERM, because
at the time that it was introduced there were no evictable ignore mark,
so subscribing to FAN_ACCESS_PERM would have incured a very high
overhead.

Yet, when we set the fmode to FMODE_NOTIFY_HSM(), we never skip trying
to send FAN_ACCESS_PERM, which is almost always a waste of cycles.

We got to this logic because of bundling FAN_OPEN*_PERM and
FAN_ACCESS_PERM in the same category and because FAN_OPEN_PERM is a
commonly used event.

By open coding fsnotify_open_perm() in fsnotify_open_perm_and_set_mode(),
we no longer need to regard FAN_OPEN*_PERM when calculating fmode.

This leaves the case of having pre-content events and not having any
other permission event in the object masks a more likely case than the
other way around.

Rework the fmode macros and code so that their meaning now refers only
to hooks on an already open file:

- FMODE_NOTIFY_NONE()		skip all events
- FMODE_NOTIFY_ACCESS_PERM()	send all permission events including
  				FAN_ACCESS_PERM
- FMODE_NOTIFY_HSM()		send pre-content permission events

Signed-off-by: Amir Goldstein <[email protected]>
Signed-off-by: Jan Kara <[email protected]>
Link: https://patch.msgid.link/[email protected]

Author: amir73il

fs/notify/fsnotify.c

@@ -199,8 +199,8 @@ static bool fsnotify_event_needs_parent(struct inode *inode, __u32 mnt_mask,
 }
 
 /* Are there any inode/mount/sb objects that watch for these events? */
-static inline bool fsnotify_object_watched(struct inode *inode, __u32 mnt_mask,
-					   __u32 mask)
+static inline __u32 fsnotify_object_watched(struct inode *inode, __u32 mnt_mask,
+					    __u32 mask)
 {
 	__u32 marks_mask = READ_ONCE(inode->i_fsnotify_mask) | mnt_mask |
 			   READ_ONCE(inode->i_sb->s_fsnotify_mask);
@@ -665,7 +665,7 @@ int fsnotify_open_perm_and_set_mode(struct file *file)
 {
 	struct dentry *dentry = file->f_path.dentry, *parent;
 	struct super_block *sb = dentry->d_sb;
-	__u32 mnt_mask, p_mask;
+	__u32 mnt_mask, p_mask = 0;
 
 	/* Is it a file opened by fanotify? */
 	if (FMODE_FSNOTIFY_NONE(file->f_mode))
@@ -683,45 +683,60 @@ int fsnotify_open_perm_and_set_mode(struct file *file)
 	}
 
 	/*
-	 * If there are permission event watchers but no pre-content event
-	 * watchers, set FMODE_NONOTIFY | FMODE_NONOTIFY_PERM to indicate that.
+	 * OK, there are some permission event watchers. Check if anybody is
+	 * watching for permission events on *this* file.
 	 */
-	if ((!d_is_dir(dentry) && !d_is_reg(dentry)) ||
-	    likely(!fsnotify_sb_has_priority_watchers(sb,
-						FSNOTIFY_PRIO_PRE_CONTENT))) {
-		file_set_fsnotify_mode(file, FMODE_NONOTIFY |
-				       FMODE_NONOTIFY_PERM);
-		goto open_perm;
+	mnt_mask = READ_ONCE(real_mount(file->f_path.mnt)->mnt_fsnotify_mask);
+	p_mask = fsnotify_object_watched(d_inode(dentry), mnt_mask,
+					 ALL_FSNOTIFY_PERM_EVENTS);
+	if (dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED) {
+		parent = dget_parent(dentry);
+		p_mask |= fsnotify_inode_watches_children(d_inode(parent));
+		dput(parent);
 	}
 
 	/*
-	 * OK, there are some pre-content watchers. Check if anybody is
-	 * watching for pre-content events on *this* file.
+	 * Legacy FAN_ACCESS_PERM events have very high performance overhead,
+	 * so unlikely to be used in the wild. If they are used there will be
+	 * no optimizations at all.
 	 */
-	mnt_mask = READ_ONCE(real_mount(file->f_path.mnt)->mnt_fsnotify_mask);
-	if (unlikely(fsnotify_object_watched(d_inode(dentry), mnt_mask,
-				     FSNOTIFY_PRE_CONTENT_EVENTS))) {
-		/* Enable pre-content events */
+	if (unlikely(p_mask & FS_ACCESS_PERM)) {
+		/* Enable all permission and pre-content events */
 		file_set_fsnotify_mode(file, 0);
 		goto open_perm;
 	}
 
-	/* Is parent watching for pre-content events on this file? */
-	if (dentry->d_flags & DCACHE_FSNOTIFY_PARENT_WATCHED) {
-		parent = dget_parent(dentry);
-		p_mask = fsnotify_inode_watches_children(d_inode(parent));
-		dput(parent);
-		if (p_mask & FSNOTIFY_PRE_CONTENT_EVENTS) {
-			/* Enable pre-content events */
-			file_set_fsnotify_mode(file, 0);
-			goto open_perm;
-		}
+	/*
+	 * Pre-content events are only supported on regular files.
+	 * If there are pre-content event watchers and no permission access
+	 * watchers, set FMODE_NONOTIFY | FMODE_NONOTIFY_PERM to indicate that.
+	 * That is the common case with HSM service.
+	 */
+	if (d_is_reg(dentry) && (p_mask & FSNOTIFY_PRE_CONTENT_EVENTS)) {
+		file_set_fsnotify_mode(file, FMODE_NONOTIFY |
+					     FMODE_NONOTIFY_PERM);
+		goto open_perm;
 	}
-	/* Nobody watching for pre-content events from this file */
-	file_set_fsnotify_mode(file, FMODE_NONOTIFY | FMODE_NONOTIFY_PERM);
+
+	/* Nobody watching permission and pre-content events on this file */
+	file_set_fsnotify_mode(file, FMODE_NONOTIFY_PERM);
 
 open_perm:
-	return fsnotify_open_perm(file);
+	/*
+	 * Send open perm events depending on object masks and regardless of
+	 * FMODE_NONOTIFY_PERM.
+	 */
+	if (file->f_flags & __FMODE_EXEC && p_mask & FS_OPEN_EXEC_PERM) {
+		int ret = fsnotify_path(&file->f_path, FS_OPEN_EXEC_PERM);
+
+		if (ret)
+			return ret;
+	}
+
+	if (p_mask & FS_OPEN_PERM)
+		return fsnotify_path(&file->f_path, FS_OPEN_PERM);
+
+	return 0;
 }
 #endif
 

include/linux/fs.h

@@ -200,26 +200,26 @@ typedef int (dio_iodone_t)(struct kiocb *iocb, loff_t offset,
 
 /*
  * The two FMODE_NONOTIFY* define which fsnotify events should not be generated
- * for a file. These are the possible values of (f->f_mode &
- * FMODE_FSNOTIFY_MASK) and their meaning:
+ * for an open file. These are the possible values of
+ * (f->f_mode & FMODE_FSNOTIFY_MASK) and their meaning:
  *
  * FMODE_NONOTIFY - suppress all (incl. non-permission) events.
  * FMODE_NONOTIFY_PERM - suppress permission (incl. pre-content) events.
- * FMODE_NONOTIFY | FMODE_NONOTIFY_PERM - suppress only pre-content events.
+ * FMODE_NONOTIFY | FMODE_NONOTIFY_PERM - suppress only FAN_ACCESS_PERM.
  */
 #define FMODE_FSNOTIFY_MASK \
 	(FMODE_NONOTIFY | FMODE_NONOTIFY_PERM)
 
 #define FMODE_FSNOTIFY_NONE(mode) \
 	((mode & FMODE_FSNOTIFY_MASK) == FMODE_NONOTIFY)
 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
-#define FMODE_FSNOTIFY_PERM(mode) \
+#define FMODE_FSNOTIFY_HSM(mode) \
 	((mode & FMODE_FSNOTIFY_MASK) == 0 || \
 	 (mode & FMODE_FSNOTIFY_MASK) == (FMODE_NONOTIFY | FMODE_NONOTIFY_PERM))
-#define FMODE_FSNOTIFY_HSM(mode) \
+#define FMODE_FSNOTIFY_ACCESS_PERM(mode) \
 	((mode & FMODE_FSNOTIFY_MASK) == 0)
 #else
-#define FMODE_FSNOTIFY_PERM(mode)	0
+#define FMODE_FSNOTIFY_ACCESS_PERM(mode) 0
 #define FMODE_FSNOTIFY_HSM(mode)	0
 #endif
 

include/linux/fsnotify.h

@@ -147,9 +147,6 @@ static inline int fsnotify_file_area_perm(struct file *file, int perm_mask,
 	if (!(perm_mask & (MAY_READ | MAY_WRITE | MAY_ACCESS)))
 		return 0;
 
-	if (likely(!FMODE_FSNOTIFY_PERM(file->f_mode)))
-		return 0;
-
 	/*
 	 * read()/write() and other types of access generate pre-content events.
 	 */
@@ -160,7 +157,8 @@ static inline int fsnotify_file_area_perm(struct file *file, int perm_mask,
 			return ret;
 	}
 
-	if (!(perm_mask & MAY_READ))
+	if (!(perm_mask & MAY_READ) ||
+	    likely(!FMODE_FSNOTIFY_ACCESS_PERM(file->f_mode)))
 		return 0;
 
 	/*
@@ -208,22 +206,6 @@ static inline int fsnotify_file_perm(struct file *file, int perm_mask)
 	return fsnotify_file_area_perm(file, perm_mask, NULL, 0);
 }
 
-/*
- * fsnotify_open_perm - permission hook before file open
- */
-static inline int fsnotify_open_perm(struct file *file)
-{
-	int ret;
-
-	if (file->f_flags & __FMODE_EXEC) {
-		ret = fsnotify_path(&file->f_path, FS_OPEN_EXEC_PERM);
-		if (ret)
-			return ret;
-	}
-
-	return fsnotify_path(&file->f_path, FS_OPEN_PERM);
-}
-
 #else
 static inline int fsnotify_open_perm_and_set_mode(struct file *file)
 {
@@ -251,11 +233,6 @@ static inline int fsnotify_file_perm(struct file *file, int perm_mask)
 {
 	return 0;
 }
-
-static inline int fsnotify_open_perm(struct file *file)
-{
-	return 0;
-}
 #endif
 
 /*