initial merge with 3.2.72

3 files changed, 1789 insertions, 0 deletions

diff --git a/fs/btrfs/backref.c b/fs/btrfs/backref.c
new file mode 100644
index 0000000..22c64ff
--- /dev/null
+++ b/fs/btrfs/backref.c
@@ -0,0 +1,776 @@
+/*
+ * Copyright (C) 2011 STRATO.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include "ctree.h"
+#include "disk-io.h"
+#include "backref.h"
+
+struct __data_ref {
+	struct list_head list;
+	u64 inum;
+	u64 root;
+	u64 extent_data_item_offset;
+};
+
+struct __shared_ref {
+	struct list_head list;
+	u64 disk_byte;
+};
+
+static int __inode_info(u64 inum, u64 ioff, u8 key_type,
+			struct btrfs_root *fs_root, struct btrfs_path *path,
+			struct btrfs_key *found_key)
+{
+	int ret;
+	struct btrfs_key key;
+	struct extent_buffer *eb;
+
+	key.type = key_type;
+	key.objectid = inum;
+	key.offset = ioff;
+
+	ret = btrfs_search_slot(NULL, fs_root, &key, path, 0, 0);
+	if (ret < 0)
+		return ret;
+
+	eb = path->nodes[0];
+	if (ret && path->slots[0] >= btrfs_header_nritems(eb)) {
+		ret = btrfs_next_leaf(fs_root, path);
+		if (ret)
+			return ret;
+		eb = path->nodes[0];
+	}
+
+	btrfs_item_key_to_cpu(eb, found_key, path->slots[0]);
+	if (found_key->type != key.type || found_key->objectid != key.objectid)
+		return 1;
+
+	return 0;
+}
+
+/*
+ * this makes the path point to (inum INODE_ITEM ioff)
+ */
+int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
+			struct btrfs_path *path)
+{
+	struct btrfs_key key;
+	return __inode_info(inum, ioff, BTRFS_INODE_ITEM_KEY, fs_root, path,
+				&key);
+}
+
+static int inode_ref_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
+				struct btrfs_path *path,
+				struct btrfs_key *found_key)
+{
+	return __inode_info(inum, ioff, BTRFS_INODE_REF_KEY, fs_root, path,
+				found_key);
+}
+
+/*
+ * this iterates to turn a btrfs_inode_ref into a full filesystem path. elements
+ * of the path are separated by '/' and the path is guaranteed to be
+ * 0-terminated. the path is only given within the current file system.
+ * Therefore, it never starts with a '/'. the caller is responsible to provide
+ * "size" bytes in "dest". the dest buffer will be filled backwards. finally,
+ * the start point of the resulting string is returned. this pointer is within
+ * dest, normally.
+ * in case the path buffer would overflow, the pointer is decremented further
+ * as if output was written to the buffer, though no more output is actually
+ * generated. that way, the caller can determine how much space would be
+ * required for the path to fit into the buffer. in that case, the returned
+ * value will be smaller than dest. callers must check this!
+ */
+static char *iref_to_path(struct btrfs_root *fs_root, struct btrfs_path *path,
+				struct btrfs_inode_ref *iref,
+				struct extent_buffer *eb_in, u64 parent,
+				char *dest, u32 size)
+{
+	u32 len;
+	int slot;
+	u64 next_inum;
+	int ret;
+	s64 bytes_left = size - 1;
+	struct extent_buffer *eb = eb_in;
+	struct btrfs_key found_key;
+
+	if (bytes_left >= 0)
+		dest[bytes_left] = '\0';
+
+	while (1) {
+		len = btrfs_inode_ref_name_len(eb, iref);
+		bytes_left -= len;
+		if (bytes_left >= 0)
+			read_extent_buffer(eb, dest + bytes_left,
+						(unsigned long)(iref + 1), len);
+		if (eb != eb_in)
+			free_extent_buffer(eb);
+		ret = inode_ref_info(parent, 0, fs_root, path, &found_key);
+		if (ret)
+			break;
+		next_inum = found_key.offset;
+
+		/* regular exit ahead */
+		if (parent == next_inum)
+			break;
+
+		slot = path->slots[0];
+		eb = path->nodes[0];
+		/* make sure we can use eb after releasing the path */
+		if (eb != eb_in)
+			atomic_inc(&eb->refs);
+		btrfs_release_path(path);
+
+		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
+		parent = next_inum;
+		--bytes_left;
+		if (bytes_left >= 0)
+			dest[bytes_left] = '/';
+	}
+
+	btrfs_release_path(path);
+
+	if (ret)
+		return ERR_PTR(ret);
+
+	return dest + bytes_left;
+}
+
+/*
+ * this makes the path point to (logical EXTENT_ITEM *)
+ * returns BTRFS_EXTENT_FLAG_DATA for data, BTRFS_EXTENT_FLAG_TREE_BLOCK for
+ * tree blocks and <0 on error.
+ */
+int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
+			struct btrfs_path *path, struct btrfs_key *found_key)
+{
+	int ret;
+	u64 flags;
+	u32 item_size;
+	struct extent_buffer *eb;
+	struct btrfs_extent_item *ei;
+	struct btrfs_key key;
+
+	key.type = BTRFS_EXTENT_ITEM_KEY;
+	key.objectid = logical;
+	key.offset = (u64)-1;
+
+	ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
+	if (ret < 0)
+		return ret;
+	ret = btrfs_previous_item(fs_info->extent_root, path,
+					0, BTRFS_EXTENT_ITEM_KEY);
+	if (ret < 0)
+		return ret;
+
+	btrfs_item_key_to_cpu(path->nodes[0], found_key, path->slots[0]);
+	if (found_key->type != BTRFS_EXTENT_ITEM_KEY ||
+	    found_key->objectid > logical ||
+	    found_key->objectid + found_key->offset <= logical)
+		return -ENOENT;
+
+	eb = path->nodes[0];
+	item_size = btrfs_item_size_nr(eb, path->slots[0]);
+	BUG_ON(item_size < sizeof(*ei));
+
+	ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
+	flags = btrfs_extent_flags(eb, ei);
+
+	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
+		return BTRFS_EXTENT_FLAG_TREE_BLOCK;
+	if (flags & BTRFS_EXTENT_FLAG_DATA)
+		return BTRFS_EXTENT_FLAG_DATA;
+
+	return -EIO;
+}
+
+/*
+ * helper function to iterate extent inline refs. ptr must point to a 0 value
+ * for the first call and may be modified. it is used to track state.
+ * if more refs exist, 0 is returned and the next call to
+ * __get_extent_inline_ref must pass the modified ptr parameter to get the
+ * next ref. after the last ref was processed, 1 is returned.
+ * returns <0 on error
+ */
+static int __get_extent_inline_ref(unsigned long *ptr, struct extent_buffer *eb,
+				struct btrfs_extent_item *ei, u32 item_size,
+				struct btrfs_extent_inline_ref **out_eiref,
+				int *out_type)
+{
+	unsigned long end;
+	u64 flags;
+	struct btrfs_tree_block_info *info;
+
+	if (!*ptr) {
+		/* first call */
+		flags = btrfs_extent_flags(eb, ei);
+		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
+			info = (struct btrfs_tree_block_info *)(ei + 1);
+			*out_eiref =
+				(struct btrfs_extent_inline_ref *)(info + 1);
+		} else {
+			*out_eiref = (struct btrfs_extent_inline_ref *)(ei + 1);
+		}
+		*ptr = (unsigned long)*out_eiref;
+		if ((void *)*ptr >= (void *)ei + item_size)
+			return -ENOENT;
+	}
+
+	end = (unsigned long)ei + item_size;
+	*out_eiref = (struct btrfs_extent_inline_ref *)*ptr;
+	*out_type = btrfs_extent_inline_ref_type(eb, *out_eiref);
+
+	*ptr += btrfs_extent_inline_ref_size(*out_type);
+	WARN_ON(*ptr > end);
+	if (*ptr == end)
+		return 1; /* last */
+
+	return 0;
+}
+
+/*
+ * reads the tree block backref for an extent. tree level and root are returned
+ * through out_level and out_root. ptr must point to a 0 value for the first
+ * call and may be modified (see __get_extent_inline_ref comment).
+ * returns 0 if data was provided, 1 if there was no more data to provide or
+ * <0 on error.
+ */
+int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
+				struct btrfs_extent_item *ei, u32 item_size,
+				u64 *out_root, u8 *out_level)
+{
+	int ret;
+	int type;
+	struct btrfs_tree_block_info *info;
+	struct btrfs_extent_inline_ref *eiref;
+
+	if (*ptr == (unsigned long)-1)
+		return 1;
+
+	while (1) {
+		ret = __get_extent_inline_ref(ptr, eb, ei, item_size,
+						&eiref, &type);
+		if (ret < 0)
+			return ret;
+
+		if (type == BTRFS_TREE_BLOCK_REF_KEY ||
+		    type == BTRFS_SHARED_BLOCK_REF_KEY)
+			break;
+
+		if (ret == 1)
+			return 1;
+	}
+
+	/* we can treat both ref types equally here */
+	info = (struct btrfs_tree_block_info *)(ei + 1);
+	*out_root = btrfs_extent_inline_ref_offset(eb, eiref);
+	*out_level = btrfs_tree_block_level(eb, info);
+
+	if (ret == 1)
+		*ptr = (unsigned long)-1;
+
+	return 0;
+}
+
+static int __data_list_add(struct list_head *head, u64 inum,
+				u64 extent_data_item_offset, u64 root)
+{
+	struct __data_ref *ref;
+
+	ref = kmalloc(sizeof(*ref), GFP_NOFS);
+	if (!ref)
+		return -ENOMEM;
+
+	ref->inum = inum;
+	ref->extent_data_item_offset = extent_data_item_offset;
+	ref->root = root;
+	list_add_tail(&ref->list, head);
+
+	return 0;
+}
+
+static int __data_list_add_eb(struct list_head *head, struct extent_buffer *eb,
+				struct btrfs_extent_data_ref *dref)
+{
+	return __data_list_add(head, btrfs_extent_data_ref_objectid(eb, dref),
+				btrfs_extent_data_ref_offset(eb, dref),
+				btrfs_extent_data_ref_root(eb, dref));
+}
+
+static int __shared_list_add(struct list_head *head, u64 disk_byte)
+{
+	struct __shared_ref *ref;
+
+	ref = kmalloc(sizeof(*ref), GFP_NOFS);
+	if (!ref)
+		return -ENOMEM;
+
+	ref->disk_byte = disk_byte;
+	list_add_tail(&ref->list, head);
+
+	return 0;
+}
+
+static int __iter_shared_inline_ref_inodes(struct btrfs_fs_info *fs_info,
+					   u64 logical, u64 inum,
+					   u64 extent_data_item_offset,
+					   u64 extent_offset,
+					   struct btrfs_path *path,
+					   struct list_head *data_refs,
+					   iterate_extent_inodes_t *iterate,
+					   void *ctx)
+{
+	u64 ref_root;
+	u32 item_size;
+	struct btrfs_key key;
+	struct extent_buffer *eb;
+	struct btrfs_extent_item *ei;
+	struct btrfs_extent_inline_ref *eiref;
+	struct __data_ref *ref;
+	int ret;
+	int type;
+	int last;
+	unsigned long ptr = 0;
+
+	WARN_ON(!list_empty(data_refs));
+	ret = extent_from_logical(fs_info, logical, path, &key);
+	if (ret & BTRFS_EXTENT_FLAG_DATA)
+		ret = -EIO;
+	if (ret < 0)
+		goto out;
+
+	eb = path->nodes[0];
+	ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
+	item_size = btrfs_item_size_nr(eb, path->slots[0]);
+
+	ret = 0;
+	ref_root = 0;
+	/*
+	 * as done in iterate_extent_inodes, we first build a list of refs to
+	 * iterate, then free the path and then iterate them to avoid deadlocks.
+	 */
+	do {
+		last = __get_extent_inline_ref(&ptr, eb, ei, item_size,
+						&eiref, &type);
+		if (last < 0) {
+			ret = last;
+			goto out;
+		}
+		if (type == BTRFS_TREE_BLOCK_REF_KEY ||
+		    type == BTRFS_SHARED_BLOCK_REF_KEY) {
+			ref_root = btrfs_extent_inline_ref_offset(eb, eiref);
+			ret = __data_list_add(data_refs, inum,
+						extent_data_item_offset,
+						ref_root);
+		}
+	} while (!ret && !last);
+
+	btrfs_release_path(path);
+
+	if (ref_root == 0) {
+		printk(KERN_ERR "btrfs: failed to find tree block ref "
+			"for shared data backref %llu\n", logical);
+		WARN_ON(1);
+		ret = -EIO;
+	}
+
+out:
+	while (!list_empty(data_refs)) {
+		ref = list_first_entry(data_refs, struct __data_ref, list);
+		list_del(&ref->list);
+		if (!ret)
+			ret = iterate(ref->inum, extent_offset +
+					ref->extent_data_item_offset,
+					ref->root, ctx);
+		kfree(ref);
+	}
+
+	return ret;
+}
+
+static int __iter_shared_inline_ref(struct btrfs_fs_info *fs_info,
+				    u64 logical, u64 orig_extent_item_objectid,
+				    u64 extent_offset, struct btrfs_path *path,
+				    struct list_head *data_refs,
+				    iterate_extent_inodes_t *iterate,
+				    void *ctx)
+{
+	u64 disk_byte;
+	struct btrfs_key key;
+	struct btrfs_file_extent_item *fi;
+	struct extent_buffer *eb;
+	int slot;
+	int nritems;
+	int ret;
+	int found = 0;
+
+	eb = read_tree_block(fs_info->tree_root, logical,
+				fs_info->tree_root->leafsize, 0);
+	if (!eb)
+		return -EIO;
+
+	/*
+	 * from the shared data ref, we only have the leaf but we need
+	 * the key. thus, we must look into all items and see that we
+	 * find one (some) with a reference to our extent item.
+	 */
+	nritems = btrfs_header_nritems(eb);
+	for (slot = 0; slot < nritems; ++slot) {
+		btrfs_item_key_to_cpu(eb, &key, slot);
+		if (key.type != BTRFS_EXTENT_DATA_KEY)
+			continue;
+		fi = btrfs_item_ptr(eb, slot, struct btrfs_file_extent_item);
+		if (!fi) {
+			free_extent_buffer(eb);
+			return -EIO;
+		}
+		disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
+		if (disk_byte != orig_extent_item_objectid) {
+			if (found)
+				break;
+			else
+				continue;
+		}
+		++found;
+		ret = __iter_shared_inline_ref_inodes(fs_info, logical,
+							key.objectid,
+							key.offset,
+							extent_offset, path,
+							data_refs,
+							iterate, ctx);
+		if (ret)
+			break;
+	}
+
+	if (!found) {
+		printk(KERN_ERR "btrfs: failed to follow shared data backref "
+			"to parent %llu\n", logical);
+		WARN_ON(1);
+		ret = -EIO;
+	}
+
+	free_extent_buffer(eb);
+	return ret;
+}
+
+/*
+ * calls iterate() for every inode that references the extent identified by
+ * the given parameters. will use the path given as a parameter and return it
+ * released.
+ * when the iterator function returns a non-zero value, iteration stops.
+ */
+int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
+				struct btrfs_path *path,
+				u64 extent_item_objectid,
+				u64 extent_offset,
+				iterate_extent_inodes_t *iterate, void *ctx)
+{
+	unsigned long ptr = 0;
+	int last;
+	int ret;
+	int type;
+	u64 logical;
+	u32 item_size;
+	struct btrfs_extent_inline_ref *eiref;
+	struct btrfs_extent_data_ref *dref;
+	struct extent_buffer *eb;
+	struct btrfs_extent_item *ei;
+	struct btrfs_key key;
+	struct list_head data_refs = LIST_HEAD_INIT(data_refs);
+	struct list_head shared_refs = LIST_HEAD_INIT(shared_refs);
+	struct __data_ref *ref_d;
+	struct __shared_ref *ref_s;
+
+	eb = path->nodes[0];
+	ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
+	item_size = btrfs_item_size_nr(eb, path->slots[0]);
+
+	/* first we iterate the inline refs, ... */
+	do {
+		last = __get_extent_inline_ref(&ptr, eb, ei, item_size,
+						&eiref, &type);
+		if (last == -ENOENT) {
+			ret = 0;
+			break;
+		}
+		if (last < 0) {
+			ret = last;
+			break;
+		}
+
+		if (type == BTRFS_EXTENT_DATA_REF_KEY) {
+			dref = (struct btrfs_extent_data_ref *)(&eiref->offset);
+			ret = __data_list_add_eb(&data_refs, eb, dref);
+		} else if (type == BTRFS_SHARED_DATA_REF_KEY) {
+			logical = btrfs_extent_inline_ref_offset(eb, eiref);
+			ret = __shared_list_add(&shared_refs, logical);
+		}
+	} while (!ret && !last);
+
+	/* ... then we proceed to in-tree references and ... */
+	while (!ret) {
+		++path->slots[0];
+		if (path->slots[0] > btrfs_header_nritems(eb)) {
+			ret = btrfs_next_leaf(fs_info->extent_root, path);
+			if (ret) {
+				if (ret == 1)
+					ret = 0; /* we're done */
+				break;
+			}
+			eb = path->nodes[0];
+		}
+		btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
+		if (key.objectid != extent_item_objectid)
+			break;
+		if (key.type == BTRFS_EXTENT_DATA_REF_KEY) {
+			dref = btrfs_item_ptr(eb, path->slots[0],
+						struct btrfs_extent_data_ref);
+			ret = __data_list_add_eb(&data_refs, eb, dref);
+		} else if (key.type == BTRFS_SHARED_DATA_REF_KEY) {
+			ret = __shared_list_add(&shared_refs, key.offset);
+		}
+	}
+
+	btrfs_release_path(path);
+
+	/*
+	 * ... only at the very end we can process the refs we found. this is
+	 * because the iterator function we call is allowed to make tree lookups
+	 * and we have to avoid deadlocks. additionally, we need more tree
+	 * lookups ourselves for shared data refs.
+	 */
+	while (!list_empty(&data_refs)) {
+		ref_d = list_first_entry(&data_refs, struct __data_ref, list);
+		list_del(&ref_d->list);
+		if (!ret)
+			ret = iterate(ref_d->inum, extent_offset +
+					ref_d->extent_data_item_offset,
+					ref_d->root, ctx);
+		kfree(ref_d);
+	}
+
+	while (!list_empty(&shared_refs)) {
+		ref_s = list_first_entry(&shared_refs, struct __shared_ref,
+					list);
+		list_del(&ref_s->list);
+		if (!ret)
+			ret = __iter_shared_inline_ref(fs_info,
+							ref_s->disk_byte,
+							extent_item_objectid,
+							extent_offset, path,
+							&data_refs,
+							iterate, ctx);
+		kfree(ref_s);
+	}
+
+	return ret;
+}
+
+int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
+				struct btrfs_path *path,
+				iterate_extent_inodes_t *iterate, void *ctx)
+{
+	int ret;
+	u64 offset;
+	struct btrfs_key found_key;
+
+	ret = extent_from_logical(fs_info, logical, path,
+					&found_key);
+	if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK)
+		ret = -EINVAL;
+	if (ret < 0)
+		return ret;
+
+	offset = logical - found_key.objectid;
+	ret = iterate_extent_inodes(fs_info, path, found_key.objectid,
+					offset, iterate, ctx);
+
+	return ret;
+}
+
+static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
+				struct btrfs_path *path,
+				iterate_irefs_t *iterate, void *ctx)
+{
+	int ret;
+	int slot;
+	u32 cur;
+	u32 len;
+	u32 name_len;
+	u64 parent = 0;
+	int found = 0;
+	struct extent_buffer *eb;
+	struct btrfs_item *item;
+	struct btrfs_inode_ref *iref;
+	struct btrfs_key found_key;
+
+	while (1) {
+		ret = inode_ref_info(inum, parent ? parent+1 : 0, fs_root, path,
+					&found_key);
+		if (ret < 0)
+			break;
+		if (ret) {
+			ret = found ? 0 : -ENOENT;
+			break;
+		}
+		++found;
+
+		parent = found_key.offset;
+		slot = path->slots[0];
+		eb = path->nodes[0];
+		/* make sure we can use eb after releasing the path */
+		atomic_inc(&eb->refs);
+		btrfs_release_path(path);
+
+		item = btrfs_item_nr(eb, slot);
+		iref = btrfs_item_ptr(eb, slot, struct btrfs_inode_ref);
+
+		for (cur = 0; cur < btrfs_item_size(eb, item); cur += len) {
+			name_len = btrfs_inode_ref_name_len(eb, iref);
+			/* path must be released before calling iterate()! */
+			ret = iterate(parent, iref, eb, ctx);
+			if (ret) {
+				free_extent_buffer(eb);
+				break;
+			}
+			len = sizeof(*iref) + name_len;
+			iref = (struct btrfs_inode_ref *)((char *)iref + len);
+		}
+		free_extent_buffer(eb);
+	}
+
+	btrfs_release_path(path);
+
+	return ret;
+}
+
+/*
+ * returns 0 if the path could be dumped (probably truncated)
+ * returns <0 in case of an error
+ */
+static int inode_to_path(u64 inum, struct btrfs_inode_ref *iref,
+				struct extent_buffer *eb, void *ctx)
+{
+	struct inode_fs_paths *ipath = ctx;
+	char *fspath;
+	char *fspath_min;
+	int i = ipath->fspath->elem_cnt;
+	const int s_ptr = sizeof(char *);
+	u32 bytes_left;
+
+	bytes_left = ipath->fspath->bytes_left > s_ptr ?
+					ipath->fspath->bytes_left - s_ptr : 0;
+
+	fspath_min = (char *)ipath->fspath->val + (i + 1) * s_ptr;
+	fspath = iref_to_path(ipath->fs_root, ipath->btrfs_path, iref, eb,
+				inum, fspath_min, bytes_left);
+	if (IS_ERR(fspath))
+		return PTR_ERR(fspath);
+
+	if (fspath > fspath_min) {
+		ipath->fspath->val[i] = (u64)(unsigned long)fspath;
+		++ipath->fspath->elem_cnt;
+		ipath->fspath->bytes_left = fspath - fspath_min;
+	} else {
+		++ipath->fspath->elem_missed;
+		ipath->fspath->bytes_missing += fspath_min - fspath;
+		ipath->fspath->bytes_left = 0;
+	}
+
+	return 0;
+}
+
+/*
+ * this dumps all file system paths to the inode into the ipath struct, provided
+ * is has been created large enough. each path is zero-terminated and accessed
+ * from ipath->fspath->val[i].
+ * when it returns, there are ipath->fspath->elem_cnt number of paths available
+ * in ipath->fspath->val[]. when the allocated space wasn't sufficient, the
+ * number of missed paths in recored in ipath->fspath->elem_missed, otherwise,
+ * it's zero. ipath->fspath->bytes_missing holds the number of bytes that would
+ * have been needed to return all paths.
+ */
+int paths_from_inode(u64 inum, struct inode_fs_paths *ipath)
+{
+	return iterate_irefs(inum, ipath->fs_root, ipath->btrfs_path,
+				inode_to_path, ipath);
+}
+
+/*
+ * allocates space to return multiple file system paths for an inode.
+ * total_bytes to allocate are passed, note that space usable for actual path
+ * information will be total_bytes - sizeof(struct inode_fs_paths).
+ * the returned pointer must be freed with free_ipath() in the end.
+ */
+struct btrfs_data_container *init_data_container(u32 total_bytes)
+{
+	struct btrfs_data_container *data;
+	size_t alloc_bytes;
+
+	alloc_bytes = max_t(size_t, total_bytes, sizeof(*data));
+	data = kmalloc(alloc_bytes, GFP_NOFS);
+	if (!data)
+		return ERR_PTR(-ENOMEM);
+
+	if (total_bytes >= sizeof(*data)) {
+		data->bytes_left = total_bytes - sizeof(*data);
+		data->bytes_missing = 0;
+	} else {
+		data->bytes_missing = sizeof(*data) - total_bytes;
+		data->bytes_left = 0;
+	}
+
+	data->elem_cnt = 0;
+	data->elem_missed = 0;
+
+	return data;
+}
+
+/*
+ * allocates space to return multiple file system paths for an inode.
+ * total_bytes to allocate are passed, note that space usable for actual path
+ * information will be total_bytes - sizeof(struct inode_fs_paths).
+ * the returned pointer must be freed with free_ipath() in the end.
+ */
+struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
+					struct btrfs_path *path)
+{
+	struct inode_fs_paths *ifp;
+	struct btrfs_data_container *fspath;
+
+	fspath = init_data_container(total_bytes);
+	if (IS_ERR(fspath))
+		return (void *)fspath;
+
+	ifp = kmalloc(sizeof(*ifp), GFP_NOFS);
+	if (!ifp) {
+		kfree(fspath);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	ifp->btrfs_path = path;
+	ifp->fspath = fspath;
+	ifp->fs_root = fs_root;
+
+	return ifp;
+}
+
+void free_ipath(struct inode_fs_paths *ipath)
+{
+	kfree(ipath);
+}
diff --git a/fs/btrfs/backref.h b/fs/btrfs/backref.h
new file mode 100644
index 0000000..9261883
--- /dev/null
+++ b/fs/btrfs/backref.h
@@ -0,0 +1,62 @@
+/*
+ * Copyright (C) 2011 STRATO.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#ifndef __BTRFS_BACKREF__
+#define __BTRFS_BACKREF__
+
+#include "ioctl.h"
+
+struct inode_fs_paths {
+	struct btrfs_path		*btrfs_path;
+	struct btrfs_root		*fs_root;
+	struct btrfs_data_container	*fspath;
+};
+
+typedef int (iterate_extent_inodes_t)(u64 inum, u64 offset, u64 root,
+		void *ctx);
+typedef int (iterate_irefs_t)(u64 parent, struct btrfs_inode_ref *iref,
+				struct extent_buffer *eb, void *ctx);
+
+int inode_item_info(u64 inum, u64 ioff, struct btrfs_root *fs_root,
+			struct btrfs_path *path);
+
+int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
+			struct btrfs_path *path, struct btrfs_key *found_key);
+
+int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
+				struct btrfs_extent_item *ei, u32 item_size,
+				u64 *out_root, u8 *out_level);
+
+int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
+				struct btrfs_path *path,
+				u64 extent_item_objectid,
+				u64 extent_offset,
+				iterate_extent_inodes_t *iterate, void *ctx);
+
+int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
+				struct btrfs_path *path,
+				iterate_extent_inodes_t *iterate, void *ctx);
+
+int paths_from_inode(u64 inum, struct inode_fs_paths *ipath);
+
+struct btrfs_data_container *init_data_container(u32 total_bytes);
+struct inode_fs_paths *init_ipath(s32 total_bytes, struct btrfs_root *fs_root,
+					struct btrfs_path *path);
+void free_ipath(struct inode_fs_paths *ipath);
+
+#endif
diff --git a/fs/btrfs/reada.c b/fs/btrfs/reada.c
new file mode 100644
index 0000000..2373b39
--- /dev/null
+++ b/fs/btrfs/reada.c
@@ -0,0 +1,951 @@
+/*
+ * Copyright (C) 2011 STRATO.  All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public
+ * License v2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public
+ * License along with this program; if not, write to the
+ * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+ * Boston, MA 021110-1307, USA.
+ */
+
+#include <linux/sched.h>
+#include <linux/pagemap.h>
+#include <linux/writeback.h>
+#include <linux/blkdev.h>
+#include <linux/rbtree.h>
+#include <linux/slab.h>
+#include <linux/workqueue.h>
+#include "ctree.h"
+#include "volumes.h"
+#include "disk-io.h"
+#include "transaction.h"
+
+#undef DEBUG
+
+/*
+ * This is the implementation for the generic read ahead framework.
+ *
+ * To trigger a readahead, btrfs_reada_add must be called. It will start
+ * a read ahead for the given range [start, end) on tree root. The returned
+ * handle can either be used to wait on the readahead to finish
+ * (btrfs_reada_wait), or to send it to the background (btrfs_reada_detach).
+ *
+ * The read ahead works as follows:
+ * On btrfs_reada_add, the root of the tree is inserted into a radix_tree.
+ * reada_start_machine will then search for extents to prefetch and trigger
+ * some reads. When a read finishes for a node, all contained node/leaf
+ * pointers that lie in the given range will also be enqueued. The reads will
+ * be triggered in sequential order, thus giving a big win over a naive
+ * enumeration. It will also make use of multi-device layouts. Each disk
+ * will have its on read pointer and all disks will by utilized in parallel.
+ * Also will no two disks read both sides of a mirror simultaneously, as this
+ * would waste seeking capacity. Instead both disks will read different parts
+ * of the filesystem.
+ * Any number of readaheads can be started in parallel. The read order will be
+ * determined globally, i.e. 2 parallel readaheads will normally finish faster
+ * than the 2 started one after another.
+ */
+
+#define MAX_MIRRORS 2
+#define MAX_IN_FLIGHT 6
+
+struct reada_extctl {
+	struct list_head	list;
+	struct reada_control	*rc;
+	u64			generation;
+};
+
+struct reada_extent {
+	u64			logical;
+	struct btrfs_key	top;
+	u32			blocksize;
+	int			err;
+	struct list_head	extctl;
+	struct kref		refcnt;
+	spinlock_t		lock;
+	struct reada_zone	*zones[MAX_MIRRORS];
+	int			nzones;
+	struct btrfs_device	*scheduled_for;
+};
+
+struct reada_zone {
+	u64			start;
+	u64			end;
+	u64			elems;
+	struct list_head	list;
+	spinlock_t		lock;
+	int			locked;
+	struct btrfs_device	*device;
+	struct btrfs_device	*devs[MAX_MIRRORS]; /* full list, incl self */
+	int			ndevs;
+	struct kref		refcnt;
+};
+
+struct reada_machine_work {
+	struct btrfs_work	work;
+	struct btrfs_fs_info	*fs_info;
+};
+
+static void reada_extent_put(struct btrfs_fs_info *, struct reada_extent *);
+static void reada_control_release(struct kref *kref);
+static void reada_zone_release(struct kref *kref);
+static void reada_start_machine(struct btrfs_fs_info *fs_info);
+static void __reada_start_machine(struct btrfs_fs_info *fs_info);
+
+static int reada_add_block(struct reada_control *rc, u64 logical,
+			   struct btrfs_key *top, int level, u64 generation);
+
+/* recurses */
+/* in case of err, eb might be NULL */
+static int __readahead_hook(struct btrfs_root *root, struct extent_buffer *eb,
+			    u64 start, int err)
+{
+	int level = 0;
+	int nritems;
+	int i;
+	u64 bytenr;
+	u64 generation;
+	struct reada_extent *re;
+	struct btrfs_fs_info *fs_info = root->fs_info;
+	struct list_head list;
+	unsigned long index = start >> PAGE_CACHE_SHIFT;
+	struct btrfs_device *for_dev;
+
+	if (eb)
+		level = btrfs_header_level(eb);
+
+	/* find extent */
+	spin_lock(&fs_info->reada_lock);
+	re = radix_tree_lookup(&fs_info->reada_tree, index);
+	if (re)
+		kref_get(&re->refcnt);
+	spin_unlock(&fs_info->reada_lock);
+
+	if (!re)
+		return -1;
+
+	spin_lock(&re->lock);
+	/*
+	 * just take the full list from the extent. afterwards we
+	 * don't need the lock anymore
+	 */
+	list_replace_init(&re->extctl, &list);
+	for_dev = re->scheduled_for;
+	re->scheduled_for = NULL;
+	spin_unlock(&re->lock);
+
+	if (err == 0) {
+		nritems = level ? btrfs_header_nritems(eb) : 0;
+		generation = btrfs_header_generation(eb);
+		/*
+		 * FIXME: currently we just set nritems to 0 if this is a leaf,
+		 * effectively ignoring the content. In a next step we could
+		 * trigger more readahead depending from the content, e.g.
+		 * fetch the checksums for the extents in the leaf.
+		 */
+	} else {
+		/*
+		 * this is the error case, the extent buffer has not been
+		 * read correctly. We won't access anything from it and
+		 * just cleanup our data structures. Effectively this will
+		 * cut the branch below this node from read ahead.
+		 */
+		nritems = 0;
+		generation = 0;
+	}
+
+	for (i = 0; i < nritems; i++) {
+		struct reada_extctl *rec;
+		u64 n_gen;
+		struct btrfs_key key;
+		struct btrfs_key next_key;
+
+		btrfs_node_key_to_cpu(eb, &key, i);
+		if (i + 1 < nritems)
+			btrfs_node_key_to_cpu(eb, &next_key, i + 1);
+		else
+			next_key = re->top;
+		bytenr = btrfs_node_blockptr(eb, i);
+		n_gen = btrfs_node_ptr_generation(eb, i);
+
+		list_for_each_entry(rec, &list, list) {
+			struct reada_control *rc = rec->rc;
+
+			/*
+			 * if the generation doesn't match, just ignore this
+			 * extctl. This will probably cut off a branch from
+			 * prefetch. Alternatively one could start a new (sub-)
+			 * prefetch for this branch, starting again from root.
+			 * FIXME: move the generation check out of this loop
+			 */
+#ifdef DEBUG
+			if (rec->generation != generation) {
+				printk(KERN_DEBUG "generation mismatch for "
+						"(%llu,%d,%llu) %llu != %llu\n",
+				       key.objectid, key.type, key.offset,
+				       rec->generation, generation);
+			}
+#endif
+			if (rec->generation == generation &&
+			    btrfs_comp_cpu_keys(&key, &rc->key_end) < 0 &&
+			    btrfs_comp_cpu_keys(&next_key, &rc->key_start) > 0)
+				reada_add_block(rc, bytenr, &next_key,
+						level - 1, n_gen);
+		}
+	}
+	/*
+	 * free extctl records
+	 */
+	while (!list_empty(&list)) {
+		struct reada_control *rc;
+		struct reada_extctl *rec;
+
+		rec = list_first_entry(&list, struct reada_extctl, list);
+		list_del(&rec->list);
+		rc = rec->rc;
+		kfree(rec);
+
+		kref_get(&rc->refcnt);
+		if (atomic_dec_and_test(&rc->elems)) {
+			kref_put(&rc->refcnt, reada_control_release);
+			wake_up(&rc->wait);
+		}
+		kref_put(&rc->refcnt, reada_control_release);
+
+		reada_extent_put(fs_info, re);	/* one ref for each entry */
+	}
+	reada_extent_put(fs_info, re);	/* our ref */
+	if (for_dev)
+		atomic_dec(&for_dev->reada_in_flight);
+
+	return 0;
+}
+
+/*
+ * start is passed separately in case eb in NULL, which may be the case with
+ * failed I/O
+ */
+int btree_readahead_hook(struct btrfs_root *root, struct extent_buffer *eb,
+			 u64 start, int err)
+{
+	int ret;
+
+	ret = __readahead_hook(root, eb, start, err);
+
+	reada_start_machine(root->fs_info);
+
+	return ret;
+}
+
+static struct reada_zone *reada_find_zone(struct btrfs_fs_info *fs_info,
+					  struct btrfs_device *dev, u64 logical,
+					  struct btrfs_bio *bbio)
+{
+	int ret;
+	int looped = 0;
+	struct reada_zone *zone;
+	struct btrfs_block_group_cache *cache = NULL;
+	u64 start;
+	u64 end;
+	int i;
+
+again:
+	zone = NULL;
+	spin_lock(&fs_info->reada_lock);
+	ret = radix_tree_gang_lookup(&dev->reada_zones, (void **)&zone,
+				     logical >> PAGE_CACHE_SHIFT, 1);
+	if (ret == 1)
+		kref_get(&zone->refcnt);
+	spin_unlock(&fs_info->reada_lock);
+
+	if (ret == 1) {
+		if (logical >= zone->start && logical < zone->end)
+			return zone;
+		spin_lock(&fs_info->reada_lock);
+		kref_put(&zone->refcnt, reada_zone_release);
+		spin_unlock(&fs_info->reada_lock);
+	}
+
+	if (looped)
+		return NULL;
+
+	cache = btrfs_lookup_block_group(fs_info, logical);
+	if (!cache)
+		return NULL;
+
+	start = cache->key.objectid;
+	end = start + cache->key.offset - 1;
+	btrfs_put_block_group(cache);
+
+	zone = kzalloc(sizeof(*zone), GFP_NOFS);
+	if (!zone)
+		return NULL;
+
+	zone->start = start;
+	zone->end = end;
+	INIT_LIST_HEAD(&zone->list);
+	spin_lock_init(&zone->lock);
+	zone->locked = 0;
+	kref_init(&zone->refcnt);
+	zone->elems = 0;
+	zone->device = dev; /* our device always sits at index 0 */
+	for (i = 0; i < bbio->num_stripes; ++i) {
+		/* bounds have already been checked */
+		zone->devs[i] = bbio->stripes[i].dev;
+	}
+	zone->ndevs = bbio->num_stripes;
+
+	spin_lock(&fs_info->reada_lock);
+	ret = radix_tree_insert(&dev->reada_zones,
+				(unsigned long)zone->end >> PAGE_CACHE_SHIFT,
+				zone);
+	spin_unlock(&fs_info->reada_lock);
+
+	if (ret) {
+		kfree(zone);
+		looped = 1;
+		goto again;
+	}
+
+	return zone;
+}
+
+static struct reada_extent *reada_find_extent(struct btrfs_root *root,
+					      u64 logical,
+					      struct btrfs_key *top, int level)
+{
+	int ret;
+	int looped = 0;
+	struct reada_extent *re = NULL;
+	struct btrfs_fs_info *fs_info = root->fs_info;
+	struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
+	struct btrfs_bio *bbio = NULL;
+	struct btrfs_device *dev;
+	u32 blocksize;
+	u64 length;
+	int nzones = 0;
+	int i;
+	unsigned long index = logical >> PAGE_CACHE_SHIFT;
+
+again:
+	spin_lock(&fs_info->reada_lock);
+	re = radix_tree_lookup(&fs_info->reada_tree, index);
+	if (re)
+		kref_get(&re->refcnt);
+	spin_unlock(&fs_info->reada_lock);
+
+	if (re || looped)
+		return re;
+
+	re = kzalloc(sizeof(*re), GFP_NOFS);
+	if (!re)
+		return NULL;
+
+	blocksize = btrfs_level_size(root, level);
+	re->logical = logical;
+	re->blocksize = blocksize;
+	re->top = *top;
+	INIT_LIST_HEAD(&re->extctl);
+	spin_lock_init(&re->lock);
+	kref_init(&re->refcnt);
+
+	/*
+	 * map block
+	 */
+	length = blocksize;
+	ret = btrfs_map_block(map_tree, REQ_WRITE, logical, &length, &bbio, 0);
+	if (ret || !bbio || length < blocksize)
+		goto error;
+
+	if (bbio->num_stripes > MAX_MIRRORS) {
+		printk(KERN_ERR "btrfs readahead: more than %d copies not "
+				"supported", MAX_MIRRORS);
+		goto error;
+	}
+
+	for (nzones = 0; nzones < bbio->num_stripes; ++nzones) {
+		struct reada_zone *zone;
+
+		dev = bbio->stripes[nzones].dev;
+		zone = reada_find_zone(fs_info, dev, logical, bbio);
+		if (!zone)
+			break;
+
+		re->zones[nzones] = zone;
+		spin_lock(&zone->lock);
+		if (!zone->elems)
+			kref_get(&zone->refcnt);
+		++zone->elems;
+		spin_unlock(&zone->lock);
+		spin_lock(&fs_info->reada_lock);
+		kref_put(&zone->refcnt, reada_zone_release);
+		spin_unlock(&fs_info->reada_lock);
+	}
+	re->nzones = nzones;
+	if (nzones == 0) {
+		/* not a single zone found, error and out */
+		goto error;
+	}
+
+	/* insert extent in reada_tree + all per-device trees, all or nothing */
+	spin_lock(&fs_info->reada_lock);
+	ret = radix_tree_insert(&fs_info->reada_tree, index, re);
+	if (ret) {
+		spin_unlock(&fs_info->reada_lock);
+		if (ret != -ENOMEM) {
+			/* someone inserted the extent in the meantime */
+			looped = 1;
+		}
+		goto error;
+	}
+	for (i = 0; i < nzones; ++i) {
+		dev = bbio->stripes[i].dev;
+		ret = radix_tree_insert(&dev->reada_extents, index, re);
+		if (ret) {
+			while (--i >= 0) {
+				dev = bbio->stripes[i].dev;
+				BUG_ON(dev == NULL);
+				radix_tree_delete(&dev->reada_extents, index);
+			}
+			BUG_ON(fs_info == NULL);
+			radix_tree_delete(&fs_info->reada_tree, index);
+			spin_unlock(&fs_info->reada_lock);
+			goto error;
+		}
+	}
+	spin_unlock(&fs_info->reada_lock);
+
+	kfree(bbio);
+	return re;
+
+error:
+	while (nzones) {
+		struct reada_zone *zone;
+
+		--nzones;
+		zone = re->zones[nzones];
+		kref_get(&zone->refcnt);
+		spin_lock(&zone->lock);
+		--zone->elems;
+		if (zone->elems == 0) {
+			/*
+			 * no fs_info->reada_lock needed, as this can't be
+			 * the last ref
+			 */
+			kref_put(&zone->refcnt, reada_zone_release);
+		}
+		spin_unlock(&zone->lock);
+
+		spin_lock(&fs_info->reada_lock);
+		kref_put(&zone->refcnt, reada_zone_release);
+		spin_unlock(&fs_info->reada_lock);
+	}
+	kfree(bbio);
+	kfree(re);
+	if (looped)
+		goto again;
+	return NULL;
+}
+
+static void reada_kref_dummy(struct kref *kr)
+{
+}
+
+static void reada_extent_put(struct btrfs_fs_info *fs_info,
+			     struct reada_extent *re)
+{
+	int i;
+	unsigned long index = re->logical >> PAGE_CACHE_SHIFT;
+
+	spin_lock(&fs_info->reada_lock);
+	if (!kref_put(&re->refcnt, reada_kref_dummy)) {
+		spin_unlock(&fs_info->reada_lock);
+		return;
+	}
+
+	radix_tree_delete(&fs_info->reada_tree, index);
+	for (i = 0; i < re->nzones; ++i) {
+		struct reada_zone *zone = re->zones[i];
+
+		radix_tree_delete(&zone->device->reada_extents, index);
+	}
+
+	spin_unlock(&fs_info->reada_lock);
+
+	for (i = 0; i < re->nzones; ++i) {
+		struct reada_zone *zone = re->zones[i];
+
+		kref_get(&zone->refcnt);
+		spin_lock(&zone->lock);
+		--zone->elems;
+		if (zone->elems == 0) {
+			/* no fs_info->reada_lock needed, as this can't be
+			 * the last ref */
+			kref_put(&zone->refcnt, reada_zone_release);
+		}
+		spin_unlock(&zone->lock);
+
+		spin_lock(&fs_info->reada_lock);
+		kref_put(&zone->refcnt, reada_zone_release);
+		spin_unlock(&fs_info->reada_lock);
+	}
+	if (re->scheduled_for)
+		atomic_dec(&re->scheduled_for->reada_in_flight);
+
+	kfree(re);
+}
+
+static void reada_zone_release(struct kref *kref)
+{
+	struct reada_zone *zone = container_of(kref, struct reada_zone, refcnt);
+
+	radix_tree_delete(&zone->device->reada_zones,
+			  zone->end >> PAGE_CACHE_SHIFT);
+
+	kfree(zone);
+}
+
+static void reada_control_release(struct kref *kref)
+{
+	struct reada_control *rc = container_of(kref, struct reada_control,
+						refcnt);
+
+	kfree(rc);
+}
+
+static int reada_add_block(struct reada_control *rc, u64 logical,
+			   struct btrfs_key *top, int level, u64 generation)
+{
+	struct btrfs_root *root = rc->root;
+	struct reada_extent *re;
+	struct reada_extctl *rec;
+
+	re = reada_find_extent(root, logical, top, level); /* takes one ref */
+	if (!re)
+		return -1;
+
+	rec = kzalloc(sizeof(*rec), GFP_NOFS);
+	if (!rec) {
+		reada_extent_put(root->fs_info, re);
+		return -1;
+	}
+
+	rec->rc = rc;
+	rec->generation = generation;
+	atomic_inc(&rc->elems);
+
+	spin_lock(&re->lock);
+	list_add_tail(&rec->list, &re->extctl);
+	spin_unlock(&re->lock);
+
+	/* leave the ref on the extent */
+
+	return 0;
+}
+
+/*
+ * called with fs_info->reada_lock held
+ */
+static void reada_peer_zones_set_lock(struct reada_zone *zone, int lock)
+{
+	int i;
+	unsigned long index = zone->end >> PAGE_CACHE_SHIFT;
+
+	for (i = 0; i < zone->ndevs; ++i) {
+		struct reada_zone *peer;
+		peer = radix_tree_lookup(&zone->devs[i]->reada_zones, index);
+		if (peer && peer->device != zone->device)
+			peer->locked = lock;
+	}
+}
+
+/*
+ * called with fs_info->reada_lock held
+ */
+static int reada_pick_zone(struct btrfs_device *dev)
+{
+	struct reada_zone *top_zone = NULL;
+	struct reada_zone *top_locked_zone = NULL;
+	u64 top_elems = 0;
+	u64 top_locked_elems = 0;
+	unsigned long index = 0;
+	int ret;
+
+	if (dev->reada_curr_zone) {
+		reada_peer_zones_set_lock(dev->reada_curr_zone, 0);
+		kref_put(&dev->reada_curr_zone->refcnt, reada_zone_release);
+		dev->reada_curr_zone = NULL;
+	}
+	/* pick the zone with the most elements */
+	while (1) {
+		struct reada_zone *zone;
+
+		ret = radix_tree_gang_lookup(&dev->reada_zones,
+					     (void **)&zone, index, 1);
+		if (ret == 0)
+			break;
+		index = (zone->end >> PAGE_CACHE_SHIFT) + 1;
+		if (zone->locked) {
+			if (zone->elems > top_locked_elems) {
+				top_locked_elems = zone->elems;
+				top_locked_zone = zone;
+			}
+		} else {
+			if (zone->elems > top_elems) {
+				top_elems = zone->elems;
+				top_zone = zone;
+			}
+		}
+	}
+	if (top_zone)
+		dev->reada_curr_zone = top_zone;
+	else if (top_locked_zone)
+		dev->reada_curr_zone = top_locked_zone;
+	else
+		return 0;
+
+	dev->reada_next = dev->reada_curr_zone->start;
+	kref_get(&dev->reada_curr_zone->refcnt);
+	reada_peer_zones_set_lock(dev->reada_curr_zone, 1);
+
+	return 1;
+}
+
+static int reada_start_machine_dev(struct btrfs_fs_info *fs_info,
+				   struct btrfs_device *dev)
+{
+	struct reada_extent *re = NULL;
+	int mirror_num = 0;
+	struct extent_buffer *eb = NULL;
+	u64 logical;
+	u32 blocksize;
+	int ret;
+	int i;
+	int need_kick = 0;
+
+	spin_lock(&fs_info->reada_lock);
+	if (dev->reada_curr_zone == NULL) {
+		ret = reada_pick_zone(dev);
+		if (!ret) {
+			spin_unlock(&fs_info->reada_lock);
+			return 0;
+		}
+	}
+	/*
+	 * FIXME currently we issue the reads one extent at a time. If we have
+	 * a contiguous block of extents, we could also coagulate them or use
+	 * plugging to speed things up
+	 */
+	ret = radix_tree_gang_lookup(&dev->reada_extents, (void **)&re,
+				     dev->reada_next >> PAGE_CACHE_SHIFT, 1);
+	if (ret == 0 || re->logical >= dev->reada_curr_zone->end) {
+		ret = reada_pick_zone(dev);
+		if (!ret) {
+			spin_unlock(&fs_info->reada_lock);
+			return 0;
+		}
+		re = NULL;
+		ret = radix_tree_gang_lookup(&dev->reada_extents, (void **)&re,
+					dev->reada_next >> PAGE_CACHE_SHIFT, 1);
+	}
+	if (ret == 0) {
+		spin_unlock(&fs_info->reada_lock);
+		return 0;
+	}
+	dev->reada_next = re->logical + re->blocksize;
+	kref_get(&re->refcnt);
+
+	spin_unlock(&fs_info->reada_lock);
+
+	/*
+	 * find mirror num
+	 */
+	for (i = 0; i < re->nzones; ++i) {
+		if (re->zones[i]->device == dev) {
+			mirror_num = i + 1;
+			break;
+		}
+	}
+	logical = re->logical;
+	blocksize = re->blocksize;
+
+	spin_lock(&re->lock);
+	if (re->scheduled_for == NULL) {
+		re->scheduled_for = dev;
+		need_kick = 1;
+	}
+	spin_unlock(&re->lock);
+
+	reada_extent_put(fs_info, re);
+
+	if (!need_kick)
+		return 0;
+
+	atomic_inc(&dev->reada_in_flight);
+	ret = reada_tree_block_flagged(fs_info->extent_root, logical, blocksize,
+			 mirror_num, &eb);
+	if (ret)
+		__readahead_hook(fs_info->extent_root, NULL, logical, ret);
+	else if (eb)
+		__readahead_hook(fs_info->extent_root, eb, eb->start, ret);
+
+	if (eb)
+		free_extent_buffer(eb);
+
+	return 1;
+
+}
+
+static void reada_start_machine_worker(struct btrfs_work *work)
+{
+	struct reada_machine_work *rmw;
+	struct btrfs_fs_info *fs_info;
+
+	rmw = container_of(work, struct reada_machine_work, work);
+	fs_info = rmw->fs_info;
+
+	kfree(rmw);
+
+	__reada_start_machine(fs_info);
+}
+
+static void __reada_start_machine(struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_device *device;
+	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
+	u64 enqueued;
+	u64 total = 0;
+	int i;
+
+	do {
+		enqueued = 0;
+		list_for_each_entry(device, &fs_devices->devices, dev_list) {
+			if (atomic_read(&device->reada_in_flight) <
+			    MAX_IN_FLIGHT)
+				enqueued += reada_start_machine_dev(fs_info,
+								    device);
+		}
+		total += enqueued;
+	} while (enqueued && total < 10000);
+
+	if (enqueued == 0)
+		return;
+
+	/*
+	 * If everything is already in the cache, this is effectively single
+	 * threaded. To a) not hold the caller for too long and b) to utilize
+	 * more cores, we broke the loop above after 10000 iterations and now
+	 * enqueue to workers to finish it. This will distribute the load to
+	 * the cores.
+	 */
+	for (i = 0; i < 2; ++i)
+		reada_start_machine(fs_info);
+}
+
+static void reada_start_machine(struct btrfs_fs_info *fs_info)
+{
+	struct reada_machine_work *rmw;
+
+	rmw = kzalloc(sizeof(*rmw), GFP_NOFS);
+	if (!rmw) {
+		/* FIXME we cannot handle this properly right now */
+		BUG();
+	}
+	rmw->work.func = reada_start_machine_worker;
+	rmw->fs_info = fs_info;
+
+	btrfs_queue_worker(&fs_info->readahead_workers, &rmw->work);
+}
+
+#ifdef DEBUG
+static void dump_devs(struct btrfs_fs_info *fs_info, int all)
+{
+	struct btrfs_device *device;
+	struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
+	unsigned long index;
+	int ret;
+	int i;
+	int j;
+	int cnt;
+
+	spin_lock(&fs_info->reada_lock);
+	list_for_each_entry(device, &fs_devices->devices, dev_list) {
+		printk(KERN_DEBUG "dev %lld has %d in flight\n", device->devid,
+			atomic_read(&device->reada_in_flight));
+		index = 0;
+		while (1) {
+			struct reada_zone *zone;
+			ret = radix_tree_gang_lookup(&device->reada_zones,
+						     (void **)&zone, index, 1);
+			if (ret == 0)
+				break;
+			printk(KERN_DEBUG "  zone %llu-%llu elems %llu locked "
+				"%d devs", zone->start, zone->end, zone->elems,
+				zone->locked);
+			for (j = 0; j < zone->ndevs; ++j) {
+				printk(KERN_CONT " %lld",
+					zone->devs[j]->devid);
+			}
+			if (device->reada_curr_zone == zone)
+				printk(KERN_CONT " curr off %llu",
+					device->reada_next - zone->start);
+			printk(KERN_CONT "\n");
+			index = (zone->end >> PAGE_CACHE_SHIFT) + 1;
+		}
+		cnt = 0;
+		index = 0;
+		while (all) {
+			struct reada_extent *re = NULL;
+
+			ret = radix_tree_gang_lookup(&device->reada_extents,
+						     (void **)&re, index, 1);
+			if (ret == 0)
+				break;
+			printk(KERN_DEBUG
+				"  re: logical %llu size %u empty %d for %lld",
+				re->logical, re->blocksize,
+				list_empty(&re->extctl), re->scheduled_for ?
+				re->scheduled_for->devid : -1);
+
+			for (i = 0; i < re->nzones; ++i) {
+				printk(KERN_CONT " zone %llu-%llu devs",
+					re->zones[i]->start,
+					re->zones[i]->end);
+				for (j = 0; j < re->zones[i]->ndevs; ++j) {
+					printk(KERN_CONT " %lld",
+						re->zones[i]->devs[j]->devid);
+				}
+			}
+			printk(KERN_CONT "\n");
+			index = (re->logical >> PAGE_CACHE_SHIFT) + 1;
+			if (++cnt > 15)
+				break;
+		}
+	}
+
+	index = 0;
+	cnt = 0;
+	while (all) {
+		struct reada_extent *re = NULL;
+
+		ret = radix_tree_gang_lookup(&fs_info->reada_tree, (void **)&re,
+					     index, 1);
+		if (ret == 0)
+			break;
+		if (!re->scheduled_for) {
+			index = (re->logical >> PAGE_CACHE_SHIFT) + 1;
+			continue;
+		}
+		printk(KERN_DEBUG
+			"re: logical %llu size %u list empty %d for %lld",
+			re->logical, re->blocksize, list_empty(&re->extctl),
+			re->scheduled_for ? re->scheduled_for->devid : -1);
+		for (i = 0; i < re->nzones; ++i) {
+			printk(KERN_CONT " zone %llu-%llu devs",
+				re->zones[i]->start,
+				re->zones[i]->end);
+			for (i = 0; i < re->nzones; ++i) {
+				printk(KERN_CONT " zone %llu-%llu devs",
+					re->zones[i]->start,
+					re->zones[i]->end);
+				for (j = 0; j < re->zones[i]->ndevs; ++j) {
+					printk(KERN_CONT " %lld",
+						re->zones[i]->devs[j]->devid);
+				}
+			}
+		}
+		printk(KERN_CONT "\n");
+		index = (re->logical >> PAGE_CACHE_SHIFT) + 1;
+	}
+	spin_unlock(&fs_info->reada_lock);
+}
+#endif
+
+/*
+ * interface
+ */
+struct reada_control *btrfs_reada_add(struct btrfs_root *root,
+			struct btrfs_key *key_start, struct btrfs_key *key_end)
+{
+	struct reada_control *rc;
+	u64 start;
+	u64 generation;
+	int level;
+	struct extent_buffer *node;
+	static struct btrfs_key max_key = {
+		.objectid = (u64)-1,
+		.type = (u8)-1,
+		.offset = (u64)-1
+	};
+
+	rc = kzalloc(sizeof(*rc), GFP_NOFS);
+	if (!rc)
+		return ERR_PTR(-ENOMEM);
+
+	rc->root = root;
+	rc->key_start = *key_start;
+	rc->key_end = *key_end;
+	atomic_set(&rc->elems, 0);
+	init_waitqueue_head(&rc->wait);
+	kref_init(&rc->refcnt);
+	kref_get(&rc->refcnt); /* one ref for having elements */
+
+	node = btrfs_root_node(root);
+	start = node->start;
+	level = btrfs_header_level(node);
+	generation = btrfs_header_generation(node);
+	free_extent_buffer(node);
+
+	reada_add_block(rc, start, &max_key, level, generation);
+
+	reada_start_machine(root->fs_info);
+
+	return rc;
+}
+
+#ifdef DEBUG
+int btrfs_reada_wait(void *handle)
+{
+	struct reada_control *rc = handle;
+
+	while (atomic_read(&rc->elems)) {
+		wait_event_timeout(rc->wait, atomic_read(&rc->elems) == 0,
+				   5 * HZ);
+		dump_devs(rc->root->fs_info, rc->elems < 10 ? 1 : 0);
+	}
+
+	dump_devs(rc->root->fs_info, rc->elems < 10 ? 1 : 0);
+
+	kref_put(&rc->refcnt, reada_control_release);
+
+	return 0;
+}
+#else
+int btrfs_reada_wait(void *handle)
+{
+	struct reada_control *rc = handle;
+
+	while (atomic_read(&rc->elems)) {
+		wait_event(rc->wait, atomic_read(&rc->elems) == 0);
+	}
+
+	kref_put(&rc->refcnt, reada_control_release);
+
+	return 0;
+}
+#endif
+
+void btrfs_reada_detach(void *handle)
+{
+	struct reada_control *rc = handle;
+
+	kref_put(&rc->refcnt, reada_control_release);
+}