Add zfs_sb_prune_aliases() function

For kernels which do not implement a per-suberblock shrinker,
those older than Linux 3.1, the shrink_dcache_parent() function
was used to attempt to reclaim dentries.  This was found not be
entirely reliable and could lead to performance issues on older
kernels running meta-data heavy workloads.

To address this issue a zfs_sb_prune_aliases() function has been
added to implement this functionality.  It relies on traversing
the list of znodes for a filesystem and adding them to a private
list with a reference held.  The private list can then be safely
walked outside the z_znodes_lock to prune dentires and drop the
last reference so the inode can be freed.

This provides the same synchronous behavior as the per-filesystem
shrinker and has the advantage of depending on only long standing
interfaces.

Signed-off-by: Brian Behlendorf <[email protected]>
Signed-off-by: Tim Chase <[email protected]>
Closes #3501

3 files changed, 84 insertions, 11 deletions

diff --git a/config/kernel-d-prune-aliases.m4 b/config/kernel-d-prune-aliases.m4
new file mode 100644
index 000000000..d9c521b1d
--- /dev/null
+++ b/config/kernel-d-prune-aliases.m4
@@ -0,0 +1,19 @@
+dnl #
+dnl # 2.6.12 API change
+dnl # d_prune_aliases() helper function available.
+dnl #
+AC_DEFUN([ZFS_AC_KERNEL_D_PRUNE_ALIASES],
+	[AC_MSG_CHECKING([whether d_prune_aliases() is available])
+	ZFS_LINUX_TRY_COMPILE_SYMBOL([
+		#include <linux/dcache.h>
+	], [
+		struct inode *ip = NULL;
+		d_prune_aliases(ip);
+	], [d_prune_aliases], [fs/dcache.c], [
+		AC_MSG_RESULT(yes)
+		AC_DEFINE(HAVE_D_PRUNE_ALIASES, 1,
+		          [d_prune_aliases() is available])
+	], [
+		AC_MSG_RESULT(no)
+	])
+])
diff --git a/config/kernel.m4 b/config/kernel.m4
index fe42e171e..51f8a2bf7 100644
--- a/config/kernel.m4
+++ b/config/kernel.m4
@@ -78,6 +78,7 @@ AC_DEFUN([ZFS_AC_CONFIG_KERNEL], [
 	ZFS_AC_KERNEL_INSERT_INODE_LOCKED
 	ZFS_AC_KERNEL_D_MAKE_ROOT
 	ZFS_AC_KERNEL_D_OBTAIN_ALIAS
+	ZFS_AC_KERNEL_D_PRUNE_ALIASES
 	ZFS_AC_KERNEL_D_SET_D_OP
 	ZFS_AC_KERNEL_D_REVALIDATE_NAMEIDATA
 	ZFS_AC_KERNEL_CONST_DENTRY_OPERATIONS
diff --git a/module/zfs/zfs_vfsops.c b/module/zfs/zfs_vfsops.c
index 88f655a8c..ae1bc324b 100644
--- a/module/zfs/zfs_vfsops.c
+++ b/module/zfs/zfs_vfsops.c
@@ -1072,6 +1072,67 @@ zfs_root(zfs_sb_t *zsb, struct inode **ipp)
 }
 EXPORT_SYMBOL(zfs_root);
 
+#if !defined(HAVE_SPLIT_SHRINKER_CALLBACK) && !defined(HAVE_SHRINK) && \
+	defined(HAVE_D_PRUNE_ALIASES)
+/*
+ * Linux kernels older than 3.1 do not support a per-filesystem shrinker.
+ * To accommodate this we must improvise and manually walk the list of znodes
+ * attempting to prune dentries in order to be able to drop the inodes.
+ *
+ * To avoid scanning the same znodes multiple times they are always rotated
+ * to the end of the z_all_znodes list.  New znodes are inserted at the
+ * end of the list so we're always scanning the oldest znodes first.
+ */
+static int
+zfs_sb_prune_aliases(zfs_sb_t *zsb, unsigned long nr_to_scan)
+{
+	znode_t **zp_array, *zp;
+	int max_array = MIN(nr_to_scan, PAGE_SIZE * 8 / sizeof (znode_t *));
+	int objects = 0;
+	int i = 0, j = 0;
+
+	zp_array = kmem_zalloc(max_array * sizeof (znode_t *), KM_SLEEP);
+
+	mutex_enter(&zsb->z_znodes_lock);
+	while ((zp = list_head(&zsb->z_all_znodes)) != NULL) {
+
+		if ((i++ > nr_to_scan) || (j >= max_array))
+			break;
+
+		ASSERT(list_link_active(&zp->z_link_node));
+		list_remove(&zsb->z_all_znodes, zp);
+		list_insert_tail(&zsb->z_all_znodes, zp);
+
+		/* Skip active znodes and .zfs entries */
+		if (MUTEX_HELD(&zp->z_lock) || zp->z_is_ctldir)
+			continue;
+
+		if (igrab(ZTOI(zp)) == NULL)
+			continue;
+
+		zp_array[j] = zp;
+		j++;
+	}
+	mutex_exit(&zsb->z_znodes_lock);
+
+	for (i = 0; i < j; i++) {
+		zp = zp_array[i];
+
+		ASSERT3P(zp, !=, NULL);
+		d_prune_aliases(ZTOI(zp));
+
+		if (atomic_read(&ZTOI(zp)->i_count) == 1)
+			objects++;
+
+		iput(ZTOI(zp));
+	}
+
+	kmem_free(zp_array, max_array * sizeof (znode_t *));
+
+	return (objects);
+}
+#endif /* HAVE_D_PRUNE_ALIASES */
+
 /*
  * The ARC has requested that the filesystem drop entries from the dentry
  * and inode caches.  This can occur when the ARC needs to free meta data
@@ -1106,18 +1167,10 @@ zfs_sb_prune(struct super_block *sb, unsigned long nr_to_scan, int *objects)
 	*objects = (*shrinker->scan_objects)(shrinker, &sc);
 #elif defined(HAVE_SHRINK)
 	*objects = (*shrinker->shrink)(shrinker, &sc);
+#elif defined(HAVE_D_PRUNE_ALIASES)
+	*objects = zfs_sb_prune_aliases(zsb, nr_to_scan);
 #else
-	/*
-	 * Linux kernels older than 3.1 do not support a per-filesystem
-	 * shrinker.  Therefore, we must fall back to the only available
-	 * interface which is to discard all unused dentries and inodes.
-	 * This behavior clearly isn't ideal but it's required so the ARC
-	 * may free memory.  The performance impact is mitigated by the
-	 * fact that the frequently accessed dentry and inode buffers will
-	 * still be in the ARC making them relatively cheap to recreate.
-	 */
-	*objects = 0;
-	shrink_dcache_parent(sb->s_root);
+#error "No available dentry and inode cache pruning mechanism."
 #endif
 	ZFS_EXIT(zsb);