fsnotify: create helpers for group mark_mutex lock

Create helpers to take and release the group mark_mutex lock.

Define a flag FSNOTIFY_GROUP_NOFS in fsnotify_group that determines
if the mark_mutex lock is fs reclaim safe or not. If not safe, the
lock helpers take the lock and disable direct fs reclaim.

In that case we annotate the mutex with a different lockdep class to
express to lockdep that an allocation of mark of an fs reclaim safe group
may take the group lock of another "NOFS" group to evict inodes.

For now, converted only the callers in common code and no backend
defines the NOFS flag. It is intended to be set by fanotify for
evictable marks support.

Link: https://lore.kernel.org/r/[e-mail]
Suggested-by: Jan Kara <[e-mail]>
Link: https://lore.kernel.org/r/[e-mail]/
Signed-off-by: Amir Goldstein <[e-mail]>
Signed-off-by: Jan Kara <[e-mail]>

fs/notify/fdinfo.c

···
 	struct fsnotify_group *group = f->private_data;
 	struct fsnotify_mark *mark;
 
-	mutex_lock(&group->mark_mutex);
+	fsnotify_group_lock(group);
 	list_for_each_entry(mark, &group->marks_list, g_list) {
 		show(m, mark);
 		if (seq_has_overflowed(m))
 			break;
 	}
-	mutex_unlock(&group->mark_mutex);
+	fsnotify_group_unlock(group);
 }
 
 #if defined(CONFIG_EXPORTFS)

fs/notify/group.c

···
 				const struct fsnotify_ops *ops,
 				int flags, gfp_t gfp)
 {
+	static struct lock_class_key nofs_marks_lock;
 	struct fsnotify_group *group;
 
 	group = kzalloc(sizeof(struct fsnotify_group), gfp);
···
 
 	group->ops = ops;
 	group->flags = flags;
+	/*
+	 * For most backends, eviction of inode with a mark is not expected,
+	 * because marks hold a refcount on the inode against eviction.
+	 *
+	 * Use a different lockdep class for groups that support evictable
+	 * inode marks, because with evictable marks, mark_mutex is NOT
+	 * fs-reclaim safe - the mutex is taken when evicting inodes.
+	 */
+	if (flags & FSNOTIFY_GROUP_NOFS)
+		lockdep_set_class(&group->mark_mutex, &nofs_marks_lock);
 
 	return group;
 }

fs/notify/mark.c

···
  */
 void fsnotify_detach_mark(struct fsnotify_mark *mark)
 {
-	struct fsnotify_group *group = mark->group;
-
-	WARN_ON_ONCE(!mutex_is_locked(&group->mark_mutex));
+	fsnotify_group_assert_locked(mark->group);
 	WARN_ON_ONCE(!srcu_read_lock_held(&fsnotify_mark_srcu) &&
 		     refcount_read(&mark->refcnt) < 1 +
 			!!(mark->flags & FSNOTIFY_MARK_FLAG_ATTACHED));
···
 void fsnotify_destroy_mark(struct fsnotify_mark *mark,
 			   struct fsnotify_group *group)
 {
-	mutex_lock(&group->mark_mutex);
+	fsnotify_group_lock(group);
 	fsnotify_detach_mark(mark);
-	mutex_unlock(&group->mark_mutex);
+	fsnotify_group_unlock(group);
 	fsnotify_free_mark(mark);
 }
 EXPORT_SYMBOL_GPL(fsnotify_destroy_mark);
···
 	struct fsnotify_group *group = mark->group;
 	int ret = 0;
 
-	BUG_ON(!mutex_is_locked(&group->mark_mutex));
+	fsnotify_group_assert_locked(group);
 
 	/*
 	 * LOCKING ORDER!!!!
···
 	int ret;
 	struct fsnotify_group *group = mark->group;
 
-	mutex_lock(&group->mark_mutex);
+	fsnotify_group_lock(group);
 	ret = fsnotify_add_mark_locked(mark, connp, obj_type, add_flags, fsid);
-	mutex_unlock(&group->mark_mutex);
+	fsnotify_group_unlock(group);
 	return ret;
 }
 EXPORT_SYMBOL_GPL(fsnotify_add_mark);
···
 	 * move marks to free to to_free list in one go and then free marks in
 	 * to_free list one by one.
 	 */
-	mutex_lock(&group->mark_mutex);
+	fsnotify_group_lock(group);
 	list_for_each_entry_safe(mark, lmark, &group->marks_list, g_list) {
 		if (mark->connector->type == obj_type)
 			list_move(&mark->g_list, &to_free);
 	}
-	mutex_unlock(&group->mark_mutex);
+	fsnotify_group_unlock(group);
 
 clear:
 	while (1) {
-		mutex_lock(&group->mark_mutex);
+		fsnotify_group_lock(group);
 		if (list_empty(head)) {
-			mutex_unlock(&group->mark_mutex);
+			fsnotify_group_unlock(group);
 			break;
 		}
 		mark = list_first_entry(head, struct fsnotify_mark, g_list);
 		fsnotify_get_mark(mark);
 		fsnotify_detach_mark(mark);
-		mutex_unlock(&group->mark_mutex);
+		fsnotify_group_unlock(group);
 		fsnotify_free_mark(mark);
 		fsnotify_put_mark(mark);
 	}

include/linux/fsnotify_backend.h

···
 #include <linux/user_namespace.h>
 #include <linux/refcount.h>
 #include <linux/mempool.h>
+#include <linux/sched/mm.h>
 
 /*
  * IN_* from inotfy.h lines up EXACTLY with FS_*, this is so we can easily
···
 
 #define FSNOTIFY_GROUP_USER	0x01 /* user allocated group */
 #define FSNOTIFY_GROUP_DUPS	0x02 /* allow multiple marks per object */
+#define FSNOTIFY_GROUP_NOFS	0x04 /* group lock is not direct reclaim safe */
 	int flags;
+	unsigned int owner_flags;	/* stored flags of mark_mutex owner */
 
 	/* stores all fastpath marks assoc with this group so they can be cleaned on unregister */
 	struct mutex mark_mutex;	/* protect marks_list */
···
 #endif /* CONFIG_FANOTIFY */
 	};
 };
+
+/*
+ * These helpers are used to prevent deadlock when reclaiming inodes with
+ * evictable marks of the same group that is allocating a new mark.
+ */
+static inline void fsnotify_group_lock(struct fsnotify_group *group)
+{
+	mutex_lock(&group->mark_mutex);
+	if (group->flags & FSNOTIFY_GROUP_NOFS)
+		group->owner_flags = memalloc_nofs_save();
+}
+
+static inline void fsnotify_group_unlock(struct fsnotify_group *group)
+{
+	if (group->flags & FSNOTIFY_GROUP_NOFS)
+		memalloc_nofs_restore(group->owner_flags);
+	mutex_unlock(&group->mark_mutex);
+}
+
+static inline void fsnotify_group_assert_locked(struct fsnotify_group *group)
+{
+	WARN_ON_ONCE(!mutex_is_locked(&group->mark_mutex));
+	if (group->flags & FSNOTIFY_GROUP_NOFS)
+		WARN_ON_ONCE(!(current->flags & PF_MEMALLOC_NOFS));
+}
 
 /* When calling fsnotify tell it if the data is a path or inode */
 enum fsnotify_data_type {