[PATCH v1 00/10] Btrfs: backref walking rewrite

[PATCH v1 00/10] Btrfs: backref walking rewrite

[Jan Schmidt]

This patch series is a major rewrite of the backref walking code. The patch
series Arne sent some weeks ago for quota groups had a very interesting
function, find_all_roots. I took this from him together with the bits needed
for find_all_roots to work and replaced a major part of the code in backref.c
with it.

It can be pulled from
	git://git.jan-o-sch.net/btrfs-unstable for-chris
There's also a gitweb for that repo on
	http://git.jan-o-sch.net/?p=btrfs-unstable

My old backref code had several problems:
- it relied on a consistent state of the trees in memory
- it ignored delayed refs
- it only featured rudimentary locking
- it could miss some references depending on the tree layout

The biggest advantage is, that we're now able to do reliable backref resolving,
even on busy file systems. So we've got benefits for:
- the existing "btrfs inspect-internal" commands
- aforementioned qgroups (patches on the list)
- "btrfs send" (currently in development)
- snapshot-aware defrag
- ... possibly more to come

Splitting the needed bits out of Arne's code was a quite intrusive operation. In
case this goes into 3.3, any of us will soon make a rebased version of the
qgroup patch set. Things corrected/changed in Arne's code along the way:
- don't assume INODE_ITEMs and the corresponding EXTENT_DATA items are in the
  same leaf (use the correct EXTENT_DATA_KEY for tree searches)
- don't assume all EXTENT_DATA items with the same backref for the same inode
  are in the same leaf (__resolve_indirect_refs can now add more refs)
- added missing key and level to prelim lists for shared block refs
- delayed ref sequence locking ability without wasting sequence numbers
- waitqueue instead of busy waiting for more delayed refs

As this touches a critical part of the file system, I also did some speed
benchmarks. It turns out that dbench shows no performance decrease on my
hardware. I can do more tests if desired.

By the way: this patch series fixes xfstest 278 (to be published soon) :-)

-Jan

Arne Jansen (6):
  Btrfs: generic data structure to build unique lists
  Btrfs: mark delayed refs as for cow
  Btrfs: always save ref_root in delayed refs
  Btrfs: add nested locking mode for paths
  Btrfs: add sequence numbers to delayed refs
  Btrfs: put back delayed refs that are too new

Jan Schmidt (4):
  Btrfs: added helper btrfs_next_item()
  Btrfs: add waitqueue instead of doing busy waiting for more delayed
    refs
  Btrfs: added btrfs_find_all_roots()
  Btrfs: new backref walking code

 fs/btrfs/Makefile      |    2 +-
 fs/btrfs/backref.c     | 1132 +++++++++++++++++++++++++++++++++++++-----------
 fs/btrfs/backref.h     |    5 +
 fs/btrfs/ctree.c       |   64 ++--
 fs/btrfs/ctree.h       |   25 +-
 fs/btrfs/delayed-ref.c |  153 +++++--
 fs/btrfs/delayed-ref.h |  104 ++++-
 fs/btrfs/disk-io.c     |    3 +-
 fs/btrfs/extent-tree.c |  187 ++++++--
 fs/btrfs/extent_io.c   |    1 +
 fs/btrfs/extent_io.h   |    2 +
 fs/btrfs/file.c        |   10 +-
 fs/btrfs/inode.c       |    2 +-
 fs/btrfs/ioctl.c       |   13 +-
 fs/btrfs/locking.c     |   51 ++-
 fs/btrfs/locking.h     |    2 +-
 fs/btrfs/relocation.c  |   18 +-
 fs/btrfs/scrub.c       |    7 +-
 fs/btrfs/transaction.c |    9 +-
 fs/btrfs/tree-log.c    |    2 +-
 fs/btrfs/ulist.c       |  220 ++++++++++
 fs/btrfs/ulist.h       |   68 +++
 22 files changed, 1651 insertions(+), 429 deletions(-)
 create mode 100644 fs/btrfs/ulist.c
 create mode 100644 fs/btrfs/ulist.h

-- 
1.7.3.4

[PATCH v1 01/10] Btrfs: generic data structure to build unique lists

[Jan Schmidt]

From: Arne Jansen <sensille@gmx.net>

ulist is a generic data structures to hold a collection of unique u64
values. The only operations it supports is adding to the list and
enumerating it.

It is possible to store an auxiliary value along with the key. The
implementation is preliminary and can probably be sped up significantly.

It is used by btrfs_find_all_roots() quota to translate recursions into
iterative loops.

Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
---
 fs/btrfs/Makefile |    2 +-
 fs/btrfs/ulist.c  |  220 +++++++++++++++++++++++++++++++++++++++++++++++++++++
 fs/btrfs/ulist.h  |   68 ++++++++++++++++
 3 files changed, 289 insertions(+), 1 deletions(-)

diff --git a/fs/btrfs/Makefile b/fs/btrfs/Makefile
index c0ddfd2..7079840 100644
--- a/fs/btrfs/Makefile
+++ b/fs/btrfs/Makefile
@@ -8,6 +8,6 @@ btrfs-y += super.o ctree.o extent-tree.o print-tree.o root-tree.o dir-item.o \
 	   extent_io.o volumes.o async-thread.o ioctl.o locking.o orphan.o \
 	   export.o tree-log.o free-space-cache.o zlib.o lzo.o \
 	   compression.o delayed-ref.o relocation.o delayed-inode.o scrub.o \
-	   reada.o backref.o
+	   reada.o backref.o ulist.o
 
 btrfs-$(CONFIG_BTRFS_FS_POSIX_ACL) += acl.o
diff --git a/fs/btrfs/ulist.c b/fs/btrfs/ulist.c
new file mode 100644
index 0000000..12f5147
--- /dev/null
+++ b/fs/btrfs/ulist.c
@@ -0,0 +1,220 @@
+/*
+ * Copyright (C) 2011 STRATO AG
+ * written by Arne Jansen <sensille@gmx.net>
+ * Distributed under the GNU GPL license version 2.
+ */
+
+#include <linux/slab.h>
+#include <linux/module.h>
+#include "ulist.h"
+
+/*
+ * ulist is a generic data structure to hold a collection of unique u64
+ * values. The only operations it supports is adding to the list and
+ * enumerating it.
+ * It is possible to store an auxiliary value along with the key.
+ *
+ * The implementation is preliminary and can probably be sped up
+ * significantly. A first step would be to store the values in an rbtree
+ * as soon as ULIST_SIZE is exceeded.
+ *
+ * A sample usage for ulists is the enumeration of directed graphs without
+ * visiting a node twice. The pseudo-code could look like this:
+ *
+ * ulist = ulist_alloc();
+ * ulist_add(ulist, root);
+ * elem = NULL;
+ *
+ * while ((elem = ulist_next(ulist, elem)) {
+ * 	for (all child nodes n in elem)
+ *		ulist_add(ulist, n);
+ *	do something useful with the node;
+ * }
+ * ulist_free(ulist);
+ *
+ * This assumes the graph nodes are adressable by u64. This stems from the
+ * usage for tree enumeration in btrfs, where the logical addresses are
+ * 64 bit.
+ *
+ * It is also useful for tree enumeration which could be done elegantly
+ * recursively, but is not possible due to kernel stack limitations. The
+ * loop would be similar to the above.
+ */
+
+/**
+ * ulist_init - freshly initialize a ulist
+ * @ulist:	the ulist to initialize
+ *
+ * Note: don't use this function to init an already used ulist, use
+ * ulist_reinit instead.
+ */
+void ulist_init(struct ulist *ulist)
+{
+	ulist->nnodes = 0;
+	ulist->nodes = ulist->int_nodes;
+	ulist->nodes_alloced = ULIST_SIZE;
+}
+EXPORT_SYMBOL(ulist_init);
+
+/**
+ * ulist_fini - free up additionally allocated memory for the ulist
+ * @ulist:	the ulist from which to free the additional memory
+ *
+ * This is useful in cases where the base 'struct ulist' has been statically
+ * allocated.
+ */
+void ulist_fini(struct ulist *ulist)
+{
+	/*
+	 * The first ULIST_SIZE elements are stored inline in struct ulist.
+	 * Only if more elements are alocated they need to be freed.
+	 */
+	if (ulist->nodes_alloced > ULIST_SIZE)
+		kfree(ulist->nodes);
+	ulist->nodes_alloced = 0;	/* in case ulist_fini is called twice */
+}
+EXPORT_SYMBOL(ulist_fini);
+
+/**
+ * ulist_reinit - prepare a ulist for reuse
+ * @ulist:	ulist to be reused
+ *
+ * Free up all additional memory allocated for the list elements and reinit
+ * the ulist.
+ */
+void ulist_reinit(struct ulist *ulist)
+{
+	ulist_fini(ulist);
+	ulist_init(ulist);
+}
+EXPORT_SYMBOL(ulist_reinit);
+
+/**
+ * ulist_alloc - dynamically allocate a ulist
+ * @gfp_mask:	allocation flags to for base allocation
+ *
+ * The allocated ulist will be returned in an initialized state.
+ */
+struct ulist *ulist_alloc(unsigned long gfp_mask)
+{
+	struct ulist *ulist = kmalloc(sizeof(*ulist), gfp_mask);
+
+	if (!ulist)
+		return NULL;
+
+	ulist_init(ulist);
+
+	return ulist;
+}
+EXPORT_SYMBOL(ulist_alloc);
+
+/**
+ * ulist_free - free dynamically allocated ulist
+ * @ulist:	ulist to free
+ *
+ * It is not necessary to call ulist_fini before.
+ */
+void ulist_free(struct ulist *ulist)
+{
+	if (!ulist)
+		return;
+	ulist_fini(ulist);
+	kfree(ulist);
+}
+EXPORT_SYMBOL(ulist_free);
+
+/**
+ * ulist_add - add an element to the ulist
+ * @ulist:	ulist to add the element to
+ * @val:	value to add to ulist
+ * @aux:	auxiliary value to store along with val
+ * @gfp_mask:	flags to use for allocation
+ *
+ * Note: locking must be provided by the caller. In case of rwlocks write
+ *       locking is needed
+ *
+ * Add an element to a ulist. The @val will only be added if it doesn't
+ * already exist. If it is added, the auxiliary value @aux is stored along with
+ * it. In case @val already exists in the ulist, @aux is ignored, even if
+ * it differs from the already stored value.
+ *
+ * ulist_add returns 0 if @val already exists in ulist and 1 if @val has been
+ * inserted.
+ * In case of allocation failure -ENOMEM is returned and the ulist stays
+ * unaltered.
+ */
+int ulist_add(struct ulist *ulist, u64 val, unsigned long aux,
+	      unsigned long gfp_mask)
+{
+	int i;
+
+	for (i = 0; i < ulist->nnodes; ++i) {
+		if (ulist->nodes[i].val == val)
+			return 0;
+	}
+
+	if (ulist->nnodes >= ulist->nodes_alloced) {
+		u64 new_alloced = ulist->nodes_alloced + 128;
+		struct ulist_node *new_nodes;
+		void *old = NULL;
+
+		/*
+		 * if nodes_alloced == ULIST_SIZE no memory has been allocated
+		 * yet, so pass NULL to krealloc
+		 */
+		if (ulist->nodes_alloced > ULIST_SIZE)
+			old = ulist->nodes;
+
+		new_nodes = krealloc(old, sizeof(*new_nodes) * new_alloced,
+				     gfp_mask);
+		if (!new_nodes)
+			return -ENOMEM;
+
+		if (!old)
+			memcpy(new_nodes, ulist->int_nodes,
+			       sizeof(ulist->int_nodes));
+
+		ulist->nodes = new_nodes;
+		ulist->nodes_alloced = new_alloced;
+	}
+	ulist->nodes[ulist->nnodes].val = val;
+	ulist->nodes[ulist->nnodes].aux = aux;
+	++ulist->nnodes;
+
+	return 1;
+}
+EXPORT_SYMBOL(ulist_add);
+
+/**
+ * ulist_next - iterate ulist
+ * @ulist:	ulist to iterate
+ * @prev:	previously returned element or %NULL to start iteration
+ *
+ * Note: locking must be provided by the caller. In case of rwlocks only read
+ *       locking is needed
+ *
+ * This function is used to iterate an ulist. The iteration is started with
+ * @prev = %NULL. It returns the next element from the ulist or %NULL when the
+ * end is reached. No guarantee is made with respect to the order in which
+ * the elements are returned. They might neither be returned in order of
+ * addition nor in ascending order.
+ * It is allowed to call ulist_add during an enumeration. Newly added items
+ * are guaranteed to show up in the running enumeration.
+ */
+struct ulist_node *ulist_next(struct ulist *ulist, struct ulist_node *prev)
+{
+	int next;
+
+	if (ulist->nnodes == 0)
+		return NULL;
+
+	if (!prev)
+		return &ulist->nodes[0];
+
+	next = (prev - ulist->nodes) + 1;
+	if (next < 0 || next >= ulist->nnodes)
+		return NULL;
+
+	return &ulist->nodes[next];
+}
+EXPORT_SYMBOL(ulist_next);
diff --git a/fs/btrfs/ulist.h b/fs/btrfs/ulist.h
new file mode 100644
index 0000000..2e25dec
--- /dev/null
+++ b/fs/btrfs/ulist.h
@@ -0,0 +1,68 @@
+/*
+ * Copyright (C) 2011 STRATO AG
+ * written by Arne Jansen <sensille@gmx.net>
+ * Distributed under the GNU GPL license version 2.
+ *
+ */
+
+#ifndef __ULIST__
+#define __ULIST__
+
+/*
+ * ulist is a generic data structure to hold a collection of unique u64
+ * values. The only operations it supports is adding to the list and
+ * enumerating it.
+ * It is possible to store an auxiliary value along with the key.
+ *
+ * The implementation is preliminary and can probably be sped up
+ * significantly. A first step would be to store the values in an rbtree
+ * as soon as ULIST_SIZE is exceeded.
+ */
+
+/*
+ * number of elements statically allocated inside struct ulist
+ */
+#define ULIST_SIZE 16
+
+/*
+ * element of the list
+ */
+struct ulist_node {
+	u64 val;		/* value to store */
+	unsigned long aux;	/* auxiliary value saved along with the val */
+};
+
+struct ulist {
+	/*
+	 * number of elements stored in list
+	 */
+	unsigned long nnodes;
+
+	/*
+	 * number of nodes we already have room for
+	 */
+	unsigned long nodes_alloced;
+
+	/*
+	 * pointer to the array storing the elements. The first ULIST_SIZE
+	 * elements are stored inline. In this case the it points to int_nodes.
+	 * After exceeding ULIST_SIZE, dynamic memory is allocated.
+	 */
+	struct ulist_node *nodes;
+
+	/*
+	 * inline storage space for the first ULIST_SIZE entries
+	 */
+	struct ulist_node int_nodes[ULIST_SIZE];
+};
+
+void ulist_init(struct ulist *ulist);
+void ulist_fini(struct ulist *ulist);
+void ulist_reinit(struct ulist *ulist);
+struct ulist *ulist_alloc(unsigned long gfp_mask);
+void ulist_free(struct ulist *ulist);
+int ulist_add(struct ulist *ulist, u64 val, unsigned long aux,
+	      unsigned long gfp_mask);
+struct ulist_node *ulist_next(struct ulist *ulist, struct ulist_node *prev);
+
+#endif
-- 
1.7.3.4

[PATCH v1 02/10] Btrfs: added helper btrfs_next_item()

[Jan Schmidt]

btrfs_next_item() makes the btrfs path point to the next item, crossing leaf
boundaries if needed.

Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
---
 fs/btrfs/ctree.h |    7 +++++++
 1 files changed, 7 insertions(+), 0 deletions(-)

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 50634abe..3e4a07b 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -2482,6 +2482,13 @@ static inline int btrfs_insert_empty_item(struct btrfs_trans_handle *trans,
 }
 
 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path);
+static inline int btrfs_next_item(struct btrfs_root *root, struct btrfs_path *p)
+{
+	++p->slots[0];
+	if (p->slots[0] >= btrfs_header_nritems(p->nodes[0]))
+		return btrfs_next_leaf(root, p);
+	return 0;
+}
 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path);
 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf);
 void btrfs_drop_snapshot(struct btrfs_root *root,
-- 
1.7.3.4

[PATCH v1 03/10] Btrfs: mark delayed refs as for cow

[Jan Schmidt]

From: Arne Jansen <sensille@gmx.net>

Add a for_cow parameter to add_delayed_*_ref and pass the appropriate value
from every call site. The for_cow parameter will later on be used to
determine if a ref will change anything with respect to qgroups.

Delayed refs coming from relocation are always counted as for_cow, as they
don't change subvol quota.

Also pass in the fs_info for later use.

btrfs_find_all_roots() will use this as an optimization, as changes that are
for_cow will not change anything with respect to which root points to a
certain leaf. Thus, we don't need to add the current sequence number to
those delayed refs.

Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
---
 fs/btrfs/ctree.c       |   42 ++++++++++----------
 fs/btrfs/ctree.h       |   17 ++++----
 fs/btrfs/delayed-ref.c |   50 +++++++++++++++---------
 fs/btrfs/delayed-ref.h |   15 +++++--
 fs/btrfs/disk-io.c     |    3 +-
 fs/btrfs/extent-tree.c |  101 ++++++++++++++++++++++++++++-------------------
 fs/btrfs/file.c        |   10 ++--
 fs/btrfs/inode.c       |    2 +-
 fs/btrfs/ioctl.c       |    5 +-
 fs/btrfs/relocation.c  |   18 +++++----
 fs/btrfs/transaction.c |    4 +-
 fs/btrfs/tree-log.c    |    2 +-
 12 files changed, 155 insertions(+), 114 deletions(-)

diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index dede441..0639a55 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -240,7 +240,7 @@ int btrfs_copy_root(struct btrfs_trans_handle *trans,
 
 	cow = btrfs_alloc_free_block(trans, root, buf->len, 0,
 				     new_root_objectid, &disk_key, level,
-				     buf->start, 0);
+				     buf->start, 0, 1);
 	if (IS_ERR(cow))
 		return PTR_ERR(cow);
 
@@ -261,9 +261,9 @@ int btrfs_copy_root(struct btrfs_trans_handle *trans,
 
 	WARN_ON(btrfs_header_generation(buf) > trans->transid);
 	if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
-		ret = btrfs_inc_ref(trans, root, cow, 1);
+		ret = btrfs_inc_ref(trans, root, cow, 1, 1);
 	else
-		ret = btrfs_inc_ref(trans, root, cow, 0);
+		ret = btrfs_inc_ref(trans, root, cow, 0, 1);
 
 	if (ret)
 		return ret;
@@ -350,14 +350,14 @@ static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
 		if ((owner == root->root_key.objectid ||
 		     root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
 		    !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
-			ret = btrfs_inc_ref(trans, root, buf, 1);
+			ret = btrfs_inc_ref(trans, root, buf, 1, 1);
 			BUG_ON(ret);
 
 			if (root->root_key.objectid ==
 			    BTRFS_TREE_RELOC_OBJECTID) {
-				ret = btrfs_dec_ref(trans, root, buf, 0);
+				ret = btrfs_dec_ref(trans, root, buf, 0, 1);
 				BUG_ON(ret);
-				ret = btrfs_inc_ref(trans, root, cow, 1);
+				ret = btrfs_inc_ref(trans, root, cow, 1, 1);
 				BUG_ON(ret);
 			}
 			new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
@@ -365,9 +365,9 @@ static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
 
 			if (root->root_key.objectid ==
 			    BTRFS_TREE_RELOC_OBJECTID)
-				ret = btrfs_inc_ref(trans, root, cow, 1);
+				ret = btrfs_inc_ref(trans, root, cow, 1, 1);
 			else
-				ret = btrfs_inc_ref(trans, root, cow, 0);
+				ret = btrfs_inc_ref(trans, root, cow, 0, 1);
 			BUG_ON(ret);
 		}
 		if (new_flags != 0) {
@@ -381,11 +381,11 @@ static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
 		if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
 			if (root->root_key.objectid ==
 			    BTRFS_TREE_RELOC_OBJECTID)
-				ret = btrfs_inc_ref(trans, root, cow, 1);
+				ret = btrfs_inc_ref(trans, root, cow, 1, 1);
 			else
-				ret = btrfs_inc_ref(trans, root, cow, 0);
+				ret = btrfs_inc_ref(trans, root, cow, 0, 1);
 			BUG_ON(ret);
-			ret = btrfs_dec_ref(trans, root, buf, 1);
+			ret = btrfs_dec_ref(trans, root, buf, 1, 1);
 			BUG_ON(ret);
 		}
 		clean_tree_block(trans, root, buf);
@@ -446,7 +446,7 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 
 	cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
 				     root->root_key.objectid, &disk_key,
-				     level, search_start, empty_size);
+				     level, search_start, empty_size, 1);
 	if (IS_ERR(cow))
 		return PTR_ERR(cow);
 
@@ -484,7 +484,7 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 		rcu_assign_pointer(root->node, cow);
 
 		btrfs_free_tree_block(trans, root, buf, parent_start,
-				      last_ref);
+				      last_ref, 1);
 		free_extent_buffer(buf);
 		add_root_to_dirty_list(root);
 	} else {
@@ -500,7 +500,7 @@ static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
 					      trans->transid);
 		btrfs_mark_buffer_dirty(parent);
 		btrfs_free_tree_block(trans, root, buf, parent_start,
-				      last_ref);
+				      last_ref, 1);
 	}
 	if (unlock_orig)
 		btrfs_tree_unlock(buf);
@@ -957,7 +957,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
 		free_extent_buffer(mid);
 
 		root_sub_used(root, mid->len);
-		btrfs_free_tree_block(trans, root, mid, 0, 1);
+		btrfs_free_tree_block(trans, root, mid, 0, 1, 0);
 		/* once for the root ptr */
 		free_extent_buffer(mid);
 		return 0;
@@ -1015,7 +1015,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
 			if (wret)
 				ret = wret;
 			root_sub_used(root, right->len);
-			btrfs_free_tree_block(trans, root, right, 0, 1);
+			btrfs_free_tree_block(trans, root, right, 0, 1, 0);
 			free_extent_buffer(right);
 			right = NULL;
 		} else {
@@ -1055,7 +1055,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
 		if (wret)
 			ret = wret;
 		root_sub_used(root, mid->len);
-		btrfs_free_tree_block(trans, root, mid, 0, 1);
+		btrfs_free_tree_block(trans, root, mid, 0, 1, 0);
 		free_extent_buffer(mid);
 		mid = NULL;
 	} else {
@@ -2089,7 +2089,7 @@ static noinline int insert_new_root(struct btrfs_trans_handle *trans,
 
 	c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
 				   root->root_key.objectid, &lower_key,
-				   level, root->node->start, 0);
+				   level, root->node->start, 0, 0);
 	if (IS_ERR(c))
 		return PTR_ERR(c);
 
@@ -2216,7 +2216,7 @@ static noinline int split_node(struct btrfs_trans_handle *trans,
 
 	split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
 					root->root_key.objectid,
-					&disk_key, level, c->start, 0);
+					&disk_key, level, c->start, 0, 0);
 	if (IS_ERR(split))
 		return PTR_ERR(split);
 
@@ -2970,7 +2970,7 @@ again:
 
 	right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
 					root->root_key.objectid,
-					&disk_key, 0, l->start, 0);
+					&disk_key, 0, l->start, 0, 0);
 	if (IS_ERR(right))
 		return PTR_ERR(right);
 
@@ -3781,7 +3781,7 @@ static noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
 
 	root_sub_used(root, leaf->len);
 
-	btrfs_free_tree_block(trans, root, leaf, 0, 1);
+	btrfs_free_tree_block(trans, root, leaf, 0, 1, 0);
 	return 0;
 }
 /*
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 3e4a07b..543f60b 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -2277,11 +2277,11 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
 					struct btrfs_root *root, u32 blocksize,
 					u64 parent, u64 root_objectid,
 					struct btrfs_disk_key *key, int level,
-					u64 hint, u64 empty_size);
+					u64 hint, u64 empty_size, int for_cow);
 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
 			   struct btrfs_root *root,
 			   struct extent_buffer *buf,
-			   u64 parent, int last_ref);
+			   u64 parent, int last_ref, int for_cow);
 struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
 					    struct btrfs_root *root,
 					    u64 bytenr, u32 blocksize,
@@ -2301,17 +2301,17 @@ int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
 				  u64 search_end, struct btrfs_key *ins,
 				  u64 data);
 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
-		  struct extent_buffer *buf, int full_backref);
+		  struct extent_buffer *buf, int full_backref, int for_cow);
 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
-		  struct extent_buffer *buf, int full_backref);
+		  struct extent_buffer *buf, int full_backref, int for_cow);
 int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
 				struct btrfs_root *root,
 				u64 bytenr, u64 num_bytes, u64 flags,
 				int is_data);
 int btrfs_free_extent(struct btrfs_trans_handle *trans,
 		      struct btrfs_root *root,
-		      u64 bytenr, u64 num_bytes, u64 parent,
-		      u64 root_objectid, u64 owner, u64 offset);
+		      u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
+		      u64 owner, u64 offset, int for_cow);
 
 int btrfs_free_reserved_extent(struct btrfs_root *root, u64 start, u64 len);
 int btrfs_free_and_pin_reserved_extent(struct btrfs_root *root,
@@ -2323,7 +2323,7 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 			 struct btrfs_root *root,
 			 u64 bytenr, u64 num_bytes, u64 parent,
-			 u64 root_objectid, u64 owner, u64 offset);
+			 u64 root_objectid, u64 owner, u64 offset, int for_cow);
 
 int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
 				    struct btrfs_root *root);
@@ -2492,7 +2492,8 @@ static inline int btrfs_next_item(struct btrfs_root *root, struct btrfs_path *p)
 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path);
 int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf);
 void btrfs_drop_snapshot(struct btrfs_root *root,
-			 struct btrfs_block_rsv *block_rsv, int update_ref);
+			 struct btrfs_block_rsv *block_rsv, int update_ref,
+			 int for_reloc);
 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
 			struct btrfs_root *root,
 			struct extent_buffer *node,
diff --git a/fs/btrfs/delayed-ref.c b/fs/btrfs/delayed-ref.c
index 125cf76..3a0f0ab 100644
--- a/fs/btrfs/delayed-ref.c
+++ b/fs/btrfs/delayed-ref.c
@@ -390,7 +390,8 @@ update_existing_head_ref(struct btrfs_delayed_ref_node *existing,
  * this does all the dirty work in terms of maintaining the correct
  * overall modification count.
  */
-static noinline int add_delayed_ref_head(struct btrfs_trans_handle *trans,
+static noinline int add_delayed_ref_head(struct btrfs_fs_info *fs_info,
+					struct btrfs_trans_handle *trans,
 					struct btrfs_delayed_ref_node *ref,
 					u64 bytenr, u64 num_bytes,
 					int action, int is_data)
@@ -468,10 +469,12 @@ static noinline int add_delayed_ref_head(struct btrfs_trans_handle *trans,
 /*
  * helper to insert a delayed tree ref into the rbtree.
  */
-static noinline int add_delayed_tree_ref(struct btrfs_trans_handle *trans,
+static noinline int add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
+					 struct btrfs_trans_handle *trans,
 					 struct btrfs_delayed_ref_node *ref,
 					 u64 bytenr, u64 num_bytes, u64 parent,
-					 u64 ref_root, int level, int action)
+					 u64 ref_root, int level, int action,
+					 int for_cow)
 {
 	struct btrfs_delayed_ref_node *existing;
 	struct btrfs_delayed_tree_ref *full_ref;
@@ -522,11 +525,12 @@ static noinline int add_delayed_tree_ref(struct btrfs_trans_handle *trans,
 /*
  * helper to insert a delayed data ref into the rbtree.
  */
-static noinline int add_delayed_data_ref(struct btrfs_trans_handle *trans,
+static noinline int add_delayed_data_ref(struct btrfs_fs_info *fs_info,
+					 struct btrfs_trans_handle *trans,
 					 struct btrfs_delayed_ref_node *ref,
 					 u64 bytenr, u64 num_bytes, u64 parent,
 					 u64 ref_root, u64 owner, u64 offset,
-					 int action)
+					 int action, int for_cow)
 {
 	struct btrfs_delayed_ref_node *existing;
 	struct btrfs_delayed_data_ref *full_ref;
@@ -554,6 +558,7 @@ static noinline int add_delayed_data_ref(struct btrfs_trans_handle *trans,
 		full_ref->root = ref_root;
 		ref->type = BTRFS_EXTENT_DATA_REF_KEY;
 	}
+
 	full_ref->objectid = owner;
 	full_ref->offset = offset;
 
@@ -580,10 +585,12 @@ static noinline int add_delayed_data_ref(struct btrfs_trans_handle *trans,
  * to make sure the delayed ref is eventually processed before this
  * transaction commits.
  */
-int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
+int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
+			       struct btrfs_trans_handle *trans,
 			       u64 bytenr, u64 num_bytes, u64 parent,
 			       u64 ref_root,  int level, int action,
-			       struct btrfs_delayed_extent_op *extent_op)
+			       struct btrfs_delayed_extent_op *extent_op,
+			       int for_cow)
 {
 	struct btrfs_delayed_tree_ref *ref;
 	struct btrfs_delayed_ref_head *head_ref;
@@ -610,12 +617,13 @@ int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
 	 * insert both the head node and the new ref without dropping
 	 * the spin lock
 	 */
-	ret = add_delayed_ref_head(trans, &head_ref->node, bytenr, num_bytes,
-				   action, 0);
+	ret = add_delayed_ref_head(fs_info, trans, &head_ref->node, bytenr,
+				   num_bytes, action, 0);
 	BUG_ON(ret);
 
-	ret = add_delayed_tree_ref(trans, &ref->node, bytenr, num_bytes,
-				   parent, ref_root, level, action);
+	ret = add_delayed_tree_ref(fs_info, trans, &ref->node, bytenr,
+				   num_bytes, parent, ref_root, level, action,
+				   for_cow);
 	BUG_ON(ret);
 	spin_unlock(&delayed_refs->lock);
 	return 0;
@@ -624,11 +632,13 @@ int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
 /*
  * add a delayed data ref. it's similar to btrfs_add_delayed_tree_ref.
  */
-int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
+int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info,
+			       struct btrfs_trans_handle *trans,
 			       u64 bytenr, u64 num_bytes,
 			       u64 parent, u64 ref_root,
 			       u64 owner, u64 offset, int action,
-			       struct btrfs_delayed_extent_op *extent_op)
+			       struct btrfs_delayed_extent_op *extent_op,
+			       int for_cow)
 {
 	struct btrfs_delayed_data_ref *ref;
 	struct btrfs_delayed_ref_head *head_ref;
@@ -655,18 +665,20 @@ int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
 	 * insert both the head node and the new ref without dropping
 	 * the spin lock
 	 */
-	ret = add_delayed_ref_head(trans, &head_ref->node, bytenr, num_bytes,
-				   action, 1);
+	ret = add_delayed_ref_head(fs_info, trans, &head_ref->node, bytenr,
+				   num_bytes, action, 1);
 	BUG_ON(ret);
 
-	ret = add_delayed_data_ref(trans, &ref->node, bytenr, num_bytes,
-				   parent, ref_root, owner, offset, action);
+	ret = add_delayed_data_ref(fs_info, trans, &ref->node, bytenr,
+				   num_bytes, parent, ref_root, owner, offset,
+				   action, for_cow);
 	BUG_ON(ret);
 	spin_unlock(&delayed_refs->lock);
 	return 0;
 }
 
-int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
+int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info,
+				struct btrfs_trans_handle *trans,
 				u64 bytenr, u64 num_bytes,
 				struct btrfs_delayed_extent_op *extent_op)
 {
@@ -683,7 +695,7 @@ int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
 	delayed_refs = &trans->transaction->delayed_refs;
 	spin_lock(&delayed_refs->lock);
 
-	ret = add_delayed_ref_head(trans, &head_ref->node, bytenr,
+	ret = add_delayed_ref_head(fs_info, trans, &head_ref->node, bytenr,
 				   num_bytes, BTRFS_UPDATE_DELAYED_HEAD,
 				   extent_op->is_data);
 	BUG_ON(ret);
diff --git a/fs/btrfs/delayed-ref.h b/fs/btrfs/delayed-ref.h
index e287e3b..8316bff 100644
--- a/fs/btrfs/delayed-ref.h
+++ b/fs/btrfs/delayed-ref.h
@@ -151,16 +151,21 @@ static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
 	}
 }
 
-int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
+int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
+			       struct btrfs_trans_handle *trans,
 			       u64 bytenr, u64 num_bytes, u64 parent,
 			       u64 ref_root, int level, int action,
-			       struct btrfs_delayed_extent_op *extent_op);
-int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
+			       struct btrfs_delayed_extent_op *extent_op,
+			       int for_cow);
+int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info,
+			       struct btrfs_trans_handle *trans,
 			       u64 bytenr, u64 num_bytes,
 			       u64 parent, u64 ref_root,
 			       u64 owner, u64 offset, int action,
-			       struct btrfs_delayed_extent_op *extent_op);
-int btrfs_add_delayed_extent_op(struct btrfs_trans_handle *trans,
+			       struct btrfs_delayed_extent_op *extent_op,
+			       int for_cow);
+int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info,
+				struct btrfs_trans_handle *trans,
 				u64 bytenr, u64 num_bytes,
 				struct btrfs_delayed_extent_op *extent_op);
 
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 94abc25..6f8cd17 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -1243,7 +1243,8 @@ static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
 	root->ref_cows = 0;
 
 	leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
-				      BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
+				      BTRFS_TREE_LOG_OBJECTID, NULL,
+				      0, 0, 0, 0);
 	if (IS_ERR(leaf)) {
 		kfree(root);
 		return ERR_CAST(leaf);
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 813c6bb..dc8b9a8 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -1872,20 +1872,24 @@ static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
 			 struct btrfs_root *root,
 			 u64 bytenr, u64 num_bytes, u64 parent,
-			 u64 root_objectid, u64 owner, u64 offset)
+			 u64 root_objectid, u64 owner, u64 offset, int for_cow)
 {
 	int ret;
+	struct btrfs_fs_info *fs_info = root->fs_info;
+
 	BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID &&
 	       root_objectid == BTRFS_TREE_LOG_OBJECTID);
 
 	if (owner < BTRFS_FIRST_FREE_OBJECTID) {
-		ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
+		ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
+					num_bytes,
 					parent, root_objectid, (int)owner,
-					BTRFS_ADD_DELAYED_REF, NULL);
+					BTRFS_ADD_DELAYED_REF, NULL, for_cow);
 	} else {
-		ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
+		ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
+					num_bytes,
 					parent, root_objectid, owner, offset,
-					BTRFS_ADD_DELAYED_REF, NULL);
+					BTRFS_ADD_DELAYED_REF, NULL, for_cow);
 	}
 	return ret;
 }
@@ -2405,7 +2409,8 @@ int btrfs_set_disk_extent_flags(struct btrfs_trans_handle *trans,
 	extent_op->update_key = 0;
 	extent_op->is_data = is_data ? 1 : 0;
 
-	ret = btrfs_add_delayed_extent_op(trans, bytenr, num_bytes, extent_op);
+	ret = btrfs_add_delayed_extent_op(root->fs_info, trans, bytenr,
+					  num_bytes, extent_op);
 	if (ret)
 		kfree(extent_op);
 	return ret;
@@ -2590,7 +2595,7 @@ out:
 static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
 			   struct btrfs_root *root,
 			   struct extent_buffer *buf,
-			   int full_backref, int inc)
+			   int full_backref, int inc, int for_cow)
 {
 	u64 bytenr;
 	u64 num_bytes;
@@ -2603,7 +2608,7 @@ static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
 	int level;
 	int ret = 0;
 	int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *,
-			    u64, u64, u64, u64, u64, u64);
+			    u64, u64, u64, u64, u64, u64, int);
 
 	ref_root = btrfs_header_owner(buf);
 	nritems = btrfs_header_nritems(buf);
@@ -2640,14 +2645,15 @@ static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
 			key.offset -= btrfs_file_extent_offset(buf, fi);
 			ret = process_func(trans, root, bytenr, num_bytes,
 					   parent, ref_root, key.objectid,
-					   key.offset);
+					   key.offset, for_cow);
 			if (ret)
 				goto fail;
 		} else {
 			bytenr = btrfs_node_blockptr(buf, i);
 			num_bytes = btrfs_level_size(root, level - 1);
 			ret = process_func(trans, root, bytenr, num_bytes,
-					   parent, ref_root, level - 1, 0);
+					   parent, ref_root, level - 1, 0,
+					   for_cow);
 			if (ret)
 				goto fail;
 		}
@@ -2659,15 +2665,15 @@ fail:
 }
 
 int btrfs_inc_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
-		  struct extent_buffer *buf, int full_backref)
+		  struct extent_buffer *buf, int full_backref, int for_cow)
 {
-	return __btrfs_mod_ref(trans, root, buf, full_backref, 1);
+	return __btrfs_mod_ref(trans, root, buf, full_backref, 1, for_cow);
 }
 
 int btrfs_dec_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root,
-		  struct extent_buffer *buf, int full_backref)
+		  struct extent_buffer *buf, int full_backref, int for_cow)
 {
-	return __btrfs_mod_ref(trans, root, buf, full_backref, 0);
+	return __btrfs_mod_ref(trans, root, buf, full_backref, 0, for_cow);
 }
 
 static int write_one_cache_group(struct btrfs_trans_handle *trans,
@@ -4937,16 +4943,17 @@ out:
 void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
 			   struct btrfs_root *root,
 			   struct extent_buffer *buf,
-			   u64 parent, int last_ref)
+			   u64 parent, int last_ref, int for_cow)
 {
 	struct btrfs_block_group_cache *cache = NULL;
 	int ret;
 
 	if (root->root_key.objectid != BTRFS_TREE_LOG_OBJECTID) {
-		ret = btrfs_add_delayed_tree_ref(trans, buf->start, buf->len,
-						parent, root->root_key.objectid,
-						btrfs_header_level(buf),
-						BTRFS_DROP_DELAYED_REF, NULL);
+		ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
+					buf->start, buf->len,
+					parent, root->root_key.objectid,
+					btrfs_header_level(buf),
+					BTRFS_DROP_DELAYED_REF, NULL, for_cow);
 		BUG_ON(ret);
 	}
 
@@ -4981,12 +4988,12 @@ out:
 	btrfs_put_block_group(cache);
 }
 
-int btrfs_free_extent(struct btrfs_trans_handle *trans,
-		      struct btrfs_root *root,
-		      u64 bytenr, u64 num_bytes, u64 parent,
-		      u64 root_objectid, u64 owner, u64 offset)
+int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
+		      u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
+		      u64 owner, u64 offset, int for_cow)
 {
 	int ret;
+	struct btrfs_fs_info *fs_info = root->fs_info;
 
 	/*
 	 * tree log blocks never actually go into the extent allocation
@@ -4998,14 +5005,17 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans,
 		btrfs_pin_extent(root, bytenr, num_bytes, 1);
 		ret = 0;
 	} else if (owner < BTRFS_FIRST_FREE_OBJECTID) {
-		ret = btrfs_add_delayed_tree_ref(trans, bytenr, num_bytes,
+		ret = btrfs_add_delayed_tree_ref(fs_info, trans, bytenr,
+					num_bytes,
 					parent, root_objectid, (int)owner,
-					BTRFS_DROP_DELAYED_REF, NULL);
+					BTRFS_DROP_DELAYED_REF, NULL, for_cow);
 		BUG_ON(ret);
 	} else {
-		ret = btrfs_add_delayed_data_ref(trans, bytenr, num_bytes,
-					parent, root_objectid, owner,
-					offset, BTRFS_DROP_DELAYED_REF, NULL);
+		ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
+						num_bytes,
+						parent, root_objectid, owner,
+						offset, BTRFS_DROP_DELAYED_REF,
+						NULL, for_cow);
 		BUG_ON(ret);
 	}
 	return ret;
@@ -5826,9 +5836,10 @@ int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 
 	BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
 
-	ret = btrfs_add_delayed_data_ref(trans, ins->objectid, ins->offset,
-					 0, root_objectid, owner, offset,
-					 BTRFS_ADD_DELAYED_EXTENT, NULL);
+	ret = btrfs_add_delayed_data_ref(root->fs_info, trans, ins->objectid,
+					 ins->offset, 0,
+					 root_objectid, owner, offset,
+					 BTRFS_ADD_DELAYED_EXTENT, NULL, 0);
 	return ret;
 }
 
@@ -5998,7 +6009,7 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
 					struct btrfs_root *root, u32 blocksize,
 					u64 parent, u64 root_objectid,
 					struct btrfs_disk_key *key, int level,
-					u64 hint, u64 empty_size)
+					u64 hint, u64 empty_size, int for_cow)
 {
 	struct btrfs_key ins;
 	struct btrfs_block_rsv *block_rsv;
@@ -6042,10 +6053,11 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
 		extent_op->update_flags = 1;
 		extent_op->is_data = 0;
 
-		ret = btrfs_add_delayed_tree_ref(trans, ins.objectid,
+		ret = btrfs_add_delayed_tree_ref(root->fs_info, trans,
+					ins.objectid,
 					ins.offset, parent, root_objectid,
 					level, BTRFS_ADD_DELAYED_EXTENT,
-					extent_op);
+					extent_op, for_cow);
 		BUG_ON(ret);
 	}
 	return buf;
@@ -6062,6 +6074,7 @@ struct walk_control {
 	int keep_locks;
 	int reada_slot;
 	int reada_count;
+	int for_reloc;
 };
 
 #define DROP_REFERENCE	1
@@ -6200,9 +6213,9 @@ static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
 	/* wc->stage == UPDATE_BACKREF */
 	if (!(wc->flags[level] & flag)) {
 		BUG_ON(!path->locks[level]);
-		ret = btrfs_inc_ref(trans, root, eb, 1);
+		ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
 		BUG_ON(ret);
-		ret = btrfs_dec_ref(trans, root, eb, 0);
+		ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
 		BUG_ON(ret);
 		ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
 						  eb->len, flag, 0);
@@ -6346,7 +6359,7 @@ skip:
 		}
 
 		ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
-					root->root_key.objectid, level - 1, 0);
+				root->root_key.objectid, level - 1, 0, 0);
 		BUG_ON(ret);
 	}
 	btrfs_tree_unlock(next);
@@ -6420,9 +6433,11 @@ static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
 	if (wc->refs[level] == 1) {
 		if (level == 0) {
 			if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
-				ret = btrfs_dec_ref(trans, root, eb, 1);
+				ret = btrfs_dec_ref(trans, root, eb, 1,
+						    wc->for_reloc);
 			else
-				ret = btrfs_dec_ref(trans, root, eb, 0);
+				ret = btrfs_dec_ref(trans, root, eb, 0,
+						    wc->for_reloc);
 			BUG_ON(ret);
 		}
 		/* make block locked assertion in clean_tree_block happy */
@@ -6449,7 +6464,7 @@ static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
 			       btrfs_header_owner(path->nodes[level + 1]));
 	}
 
-	btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1);
+	btrfs_free_tree_block(trans, root, eb, parent, wc->refs[level] == 1, 0);
 out:
 	wc->refs[level] = 0;
 	wc->flags[level] = 0;
@@ -6533,7 +6548,8 @@ static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
  * blocks are properly updated.
  */
 void btrfs_drop_snapshot(struct btrfs_root *root,
-			 struct btrfs_block_rsv *block_rsv, int update_ref)
+			 struct btrfs_block_rsv *block_rsv, int update_ref,
+			 int for_reloc)
 {
 	struct btrfs_path *path;
 	struct btrfs_trans_handle *trans;
@@ -6621,6 +6637,7 @@ void btrfs_drop_snapshot(struct btrfs_root *root,
 	wc->stage = DROP_REFERENCE;
 	wc->update_ref = update_ref;
 	wc->keep_locks = 0;
+	wc->for_reloc = for_reloc;
 	wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
 
 	while (1) {
@@ -6705,6 +6722,7 @@ out:
  * drop subtree rooted at tree block 'node'.
  *
  * NOTE: this function will unlock and release tree block 'node'
+ * only used by relocation code
  */
 int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
 			struct btrfs_root *root,
@@ -6749,6 +6767,7 @@ int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
 	wc->stage = DROP_REFERENCE;
 	wc->update_ref = 0;
 	wc->keep_locks = 1;
+	wc->for_reloc = 1;
 	wc->reada_count = BTRFS_NODEPTRS_PER_BLOCK(root);
 
 	while (1) {
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index f2e9282..d2b60ed 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -678,7 +678,7 @@ next_slot:
 						disk_bytenr, num_bytes, 0,
 						root->root_key.objectid,
 						new_key.objectid,
-						start - extent_offset);
+						start - extent_offset, 0);
 				BUG_ON(ret);
 				*hint_byte = disk_bytenr;
 			}
@@ -753,7 +753,7 @@ next_slot:
 						disk_bytenr, num_bytes, 0,
 						root->root_key.objectid,
 						key.objectid, key.offset -
-						extent_offset);
+						extent_offset, 0);
 				BUG_ON(ret);
 				inode_sub_bytes(inode,
 						extent_end - key.offset);
@@ -962,7 +962,7 @@ again:
 
 		ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
 					   root->root_key.objectid,
-					   ino, orig_offset);
+					   ino, orig_offset, 0);
 		BUG_ON(ret);
 
 		if (split == start) {
@@ -989,7 +989,7 @@ again:
 		del_nr++;
 		ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 					0, root->root_key.objectid,
-					ino, orig_offset);
+					ino, orig_offset, 0);
 		BUG_ON(ret);
 	}
 	other_start = 0;
@@ -1006,7 +1006,7 @@ again:
 		del_nr++;
 		ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 					0, root->root_key.objectid,
-					ino, orig_offset);
+					ino, orig_offset, 0);
 		BUG_ON(ret);
 	}
 	if (del_nr == 0) {
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index c5ccec2..ea81938 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -3139,7 +3139,7 @@ delete:
 			ret = btrfs_free_extent(trans, root, extent_start,
 						extent_num_bytes, 0,
 						btrfs_header_owner(leaf),
-						ino, extent_offset);
+						ino, extent_offset, 0);
 			BUG_ON(ret);
 		}
 
diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 72d4616..c48f2e9 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -358,7 +358,7 @@ static noinline int create_subvol(struct btrfs_root *root,
 		return PTR_ERR(trans);
 
 	leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
-				      0, objectid, NULL, 0, 0, 0);
+				      0, objectid, NULL, 0, 0, 0, 0);
 	if (IS_ERR(leaf)) {
 		ret = PTR_ERR(leaf);
 		goto fail;
@@ -2425,7 +2425,8 @@ static noinline long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
 							disko, diskl, 0,
 							root->root_key.objectid,
 							btrfs_ino(inode),
-							new_key.offset - datao);
+							new_key.offset - datao,
+							0);
 					BUG_ON(ret);
 				}
 			} else if (type == BTRFS_FILE_EXTENT_INLINE) {
diff --git a/fs/btrfs/relocation.c b/fs/btrfs/relocation.c
index dff29d5..8c1aae2 100644
--- a/fs/btrfs/relocation.c
+++ b/fs/btrfs/relocation.c
@@ -1604,12 +1604,12 @@ int replace_file_extents(struct btrfs_trans_handle *trans,
 		ret = btrfs_inc_extent_ref(trans, root, new_bytenr,
 					   num_bytes, parent,
 					   btrfs_header_owner(leaf),
-					   key.objectid, key.offset);
+					   key.objectid, key.offset, 1);
 		BUG_ON(ret);
 
 		ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 					parent, btrfs_header_owner(leaf),
-					key.objectid, key.offset);
+					key.objectid, key.offset, 1);
 		BUG_ON(ret);
 	}
 	if (dirty)
@@ -1778,21 +1778,23 @@ again:
 
 		ret = btrfs_inc_extent_ref(trans, src, old_bytenr, blocksize,
 					path->nodes[level]->start,
-					src->root_key.objectid, level - 1, 0);
+					src->root_key.objectid, level - 1, 0,
+					1);
 		BUG_ON(ret);
 		ret = btrfs_inc_extent_ref(trans, dest, new_bytenr, blocksize,
 					0, dest->root_key.objectid, level - 1,
-					0);
+					0, 1);
 		BUG_ON(ret);
 
 		ret = btrfs_free_extent(trans, src, new_bytenr, blocksize,
 					path->nodes[level]->start,
-					src->root_key.objectid, level - 1, 0);
+					src->root_key.objectid, level - 1, 0,
+					1);
 		BUG_ON(ret);
 
 		ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize,
 					0, dest->root_key.objectid, level - 1,
-					0);
+					0, 1);
 		BUG_ON(ret);
 
 		btrfs_unlock_up_safe(path, 0);
@@ -2244,7 +2246,7 @@ again:
 		} else {
 			list_del_init(&reloc_root->root_list);
 		}
-		btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0);
+		btrfs_drop_snapshot(reloc_root, rc->block_rsv, 0, 1);
 	}
 
 	if (found) {
@@ -2558,7 +2560,7 @@ static int do_relocation(struct btrfs_trans_handle *trans,
 						node->eb->start, blocksize,
 						upper->eb->start,
 						btrfs_header_owner(upper->eb),
-						node->level, 0);
+						node->level, 0, 1);
 			BUG_ON(ret);
 
 			ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 81376d9..a2bfedc 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -1393,9 +1393,9 @@ int btrfs_clean_old_snapshots(struct btrfs_root *root)
 
 		if (btrfs_header_backref_rev(root->node) <
 		    BTRFS_MIXED_BACKREF_REV)
-			btrfs_drop_snapshot(root, NULL, 0);
+			btrfs_drop_snapshot(root, NULL, 0, 0);
 		else
-			btrfs_drop_snapshot(root, NULL, 1);
+			btrfs_drop_snapshot(root, NULL, 1, 0);
 	}
 	return 0;
 }
diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
index f4d81c0..fce7b9e 100644
--- a/fs/btrfs/tree-log.c
+++ b/fs/btrfs/tree-log.c
@@ -589,7 +589,7 @@ static noinline int replay_one_extent(struct btrfs_trans_handle *trans,
 				ret = btrfs_inc_extent_ref(trans, root,
 						ins.objectid, ins.offset,
 						0, root->root_key.objectid,
-						key->objectid, offset);
+						key->objectid, offset, 0);
 				BUG_ON(ret);
 			} else {
 				/*
-- 
1.7.3.4

[PATCH v1 04/10] Btrfs: always save ref_root in delayed refs

[Jan Schmidt]

From: Arne Jansen <sensille@gmx.net>

For consistent backref walking and (later) qgroup calculation the
information to which root a delayed ref belongs is useful even for shared
refs.

Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
---
 fs/btrfs/delayed-ref.c |   18 ++++++++----------
 fs/btrfs/delayed-ref.h |   12 ++++--------
 2 files changed, 12 insertions(+), 18 deletions(-)

diff --git a/fs/btrfs/delayed-ref.c b/fs/btrfs/delayed-ref.c
index 3a0f0ab..babd37b 100644
--- a/fs/btrfs/delayed-ref.c
+++ b/fs/btrfs/delayed-ref.c
@@ -495,13 +495,12 @@ static noinline int add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
 	ref->in_tree = 1;
 
 	full_ref = btrfs_delayed_node_to_tree_ref(ref);
-	if (parent) {
-		full_ref->parent = parent;
+	full_ref->parent = parent;
+	full_ref->root = ref_root;
+	if (parent)
 		ref->type = BTRFS_SHARED_BLOCK_REF_KEY;
-	} else {
-		full_ref->root = ref_root;
+	else
 		ref->type = BTRFS_TREE_BLOCK_REF_KEY;
-	}
 	full_ref->level = level;
 
 	trace_btrfs_delayed_tree_ref(ref, full_ref, action);
@@ -551,13 +550,12 @@ static noinline int add_delayed_data_ref(struct btrfs_fs_info *fs_info,
 	ref->in_tree = 1;
 
 	full_ref = btrfs_delayed_node_to_data_ref(ref);
-	if (parent) {
-		full_ref->parent = parent;
+	full_ref->parent = parent;
+	full_ref->root = ref_root;
+	if (parent)
 		ref->type = BTRFS_SHARED_DATA_REF_KEY;
-	} else {
-		full_ref->root = ref_root;
+	else
 		ref->type = BTRFS_EXTENT_DATA_REF_KEY;
-	}
 
 	full_ref->objectid = owner;
 	full_ref->offset = offset;
diff --git a/fs/btrfs/delayed-ref.h b/fs/btrfs/delayed-ref.h
index 8316bff..a5fb2bc 100644
--- a/fs/btrfs/delayed-ref.h
+++ b/fs/btrfs/delayed-ref.h
@@ -98,19 +98,15 @@ struct btrfs_delayed_ref_head {
 
 struct btrfs_delayed_tree_ref {
 	struct btrfs_delayed_ref_node node;
-	union {
-		u64 root;
-		u64 parent;
-	};
+	u64 root;
+	u64 parent;
 	int level;
 };
 
 struct btrfs_delayed_data_ref {
 	struct btrfs_delayed_ref_node node;
-	union {
-		u64 root;
-		u64 parent;
-	};
+	u64 root;
+	u64 parent;
 	u64 objectid;
 	u64 offset;
 };
-- 
1.7.3.4

[PATCH v1 05/10] Btrfs: add nested locking mode for paths

[Jan Schmidt]

From: Arne Jansen <sensille@gmx.net>

This patch adds the possibilty to read-lock an extent even if it is already
write-locked from the same thread. btrfs_find_all_roots() needs this
capability.

Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
---
 fs/btrfs/ctree.c     |   22 ++++++++++++--------
 fs/btrfs/ctree.h     |    1 +
 fs/btrfs/extent_io.c |    1 +
 fs/btrfs/extent_io.h |    2 +
 fs/btrfs/locking.c   |   51 +++++++++++++++++++++++++++++++++++++++++++++++--
 fs/btrfs/locking.h   |    2 +-
 6 files changed, 66 insertions(+), 13 deletions(-)

diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index 0639a55..d0cd67e 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -186,13 +186,14 @@ struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
  * tree until you end up with a lock on the root.  A locked buffer
  * is returned, with a reference held.
  */
-struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
+struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root,
+						int nested)
 {
 	struct extent_buffer *eb;
 
 	while (1) {
 		eb = btrfs_root_node(root);
-		btrfs_tree_read_lock(eb);
+		btrfs_tree_read_lock(eb, nested);
 		if (eb == root->node)
 			break;
 		btrfs_tree_read_unlock(eb);
@@ -1637,6 +1638,7 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
 	/* everything at write_lock_level or lower must be write locked */
 	int write_lock_level = 0;
 	u8 lowest_level = 0;
+	int nested = p->nested;
 
 	lowest_level = p->lowest_level;
 	WARN_ON(lowest_level && ins_len > 0);
@@ -1678,8 +1680,9 @@ again:
 		b = root->commit_root;
 		extent_buffer_get(b);
 		level = btrfs_header_level(b);
+		BUG_ON(p->skip_locking && nested);
 		if (!p->skip_locking)
-			btrfs_tree_read_lock(b);
+			btrfs_tree_read_lock(b, 0);
 	} else {
 		if (p->skip_locking) {
 			b = btrfs_root_node(root);
@@ -1688,7 +1691,7 @@ again:
 			/* we don't know the level of the root node
 			 * until we actually have it read locked
 			 */
-			b = btrfs_read_lock_root_node(root);
+			b = btrfs_read_lock_root_node(root, nested);
 			level = btrfs_header_level(b);
 			if (level <= write_lock_level) {
 				/* whoops, must trade for write lock */
@@ -1827,7 +1830,8 @@ cow_done:
 					err = btrfs_try_tree_read_lock(b);
 					if (!err) {
 						btrfs_set_path_blocking(p);
-						btrfs_tree_read_lock(b);
+						btrfs_tree_read_lock(b,
+								     nested);
 						btrfs_clear_path_blocking(p, b,
 								  BTRFS_READ_LOCK);
 					}
@@ -3972,7 +3976,7 @@ int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
 
 	WARN_ON(!path->keep_locks);
 again:
-	cur = btrfs_read_lock_root_node(root);
+	cur = btrfs_read_lock_root_node(root, 0);
 	level = btrfs_header_level(cur);
 	WARN_ON(path->nodes[level]);
 	path->nodes[level] = cur;
@@ -4066,7 +4070,7 @@ find_next_key:
 		cur = read_node_slot(root, cur, slot);
 		BUG_ON(!cur);
 
-		btrfs_tree_read_lock(cur);
+		btrfs_tree_read_lock(cur, 0);
 
 		path->locks[level - 1] = BTRFS_READ_LOCK;
 		path->nodes[level - 1] = cur;
@@ -4260,7 +4264,7 @@ again:
 			ret = btrfs_try_tree_read_lock(next);
 			if (!ret) {
 				btrfs_set_path_blocking(path);
-				btrfs_tree_read_lock(next);
+				btrfs_tree_read_lock(next, 0);
 				btrfs_clear_path_blocking(path, next,
 							  BTRFS_READ_LOCK);
 			}
@@ -4297,7 +4301,7 @@ again:
 			ret = btrfs_try_tree_read_lock(next);
 			if (!ret) {
 				btrfs_set_path_blocking(path);
-				btrfs_tree_read_lock(next);
+				btrfs_tree_read_lock(next, 0);
 				btrfs_clear_path_blocking(path, next,
 							  BTRFS_READ_LOCK);
 			}
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 543f60b..34a6941 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -530,6 +530,7 @@ struct btrfs_path {
 	unsigned int skip_locking:1;
 	unsigned int leave_spinning:1;
 	unsigned int search_commit_root:1;
+	unsigned int nested:1;
 };
 
 /*
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c
index be1bf62..dd8d140 100644
--- a/fs/btrfs/extent_io.c
+++ b/fs/btrfs/extent_io.c
@@ -3571,6 +3571,7 @@ static struct extent_buffer *__alloc_extent_buffer(struct extent_io_tree *tree,
 	atomic_set(&eb->blocking_writers, 0);
 	atomic_set(&eb->spinning_readers, 0);
 	atomic_set(&eb->spinning_writers, 0);
+	eb->lock_nested = 0;
 	init_waitqueue_head(&eb->write_lock_wq);
 	init_waitqueue_head(&eb->read_lock_wq);
 
diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h
index 7604c30..bc6a042cb 100644
--- a/fs/btrfs/extent_io.h
+++ b/fs/btrfs/extent_io.h
@@ -129,6 +129,7 @@ struct extent_buffer {
 	struct list_head leak_list;
 	struct rcu_head rcu_head;
 	atomic_t refs;
+	pid_t lock_owner;
 
 	/* count of read lock holders on the extent buffer */
 	atomic_t write_locks;
@@ -137,6 +138,7 @@ struct extent_buffer {
 	atomic_t blocking_readers;
 	atomic_t spinning_readers;
 	atomic_t spinning_writers;
+	int lock_nested;
 
 	/* protects write locks */
 	rwlock_t lock;
diff --git a/fs/btrfs/locking.c b/fs/btrfs/locking.c
index d77b67c..3cd604f 100644
--- a/fs/btrfs/locking.c
+++ b/fs/btrfs/locking.c
@@ -33,6 +33,14 @@ void btrfs_assert_tree_read_locked(struct extent_buffer *eb);
  */
 void btrfs_set_lock_blocking_rw(struct extent_buffer *eb, int rw)
 {
+	if (eb->lock_nested) {
+		read_lock(&eb->lock);
+		if (eb->lock_nested && current->pid == eb->lock_owner) {
+			read_unlock(&eb->lock);
+			return;
+		}
+		read_unlock(&eb->lock);
+	}
 	if (rw == BTRFS_WRITE_LOCK) {
 		if (atomic_read(&eb->blocking_writers) == 0) {
 			WARN_ON(atomic_read(&eb->spinning_writers) != 1);
@@ -57,6 +65,14 @@ void btrfs_set_lock_blocking_rw(struct extent_buffer *eb, int rw)
  */
 void btrfs_clear_lock_blocking_rw(struct extent_buffer *eb, int rw)
 {
+	if (eb->lock_nested) {
+		read_lock(&eb->lock);
+		if (&eb->lock_nested && current->pid == eb->lock_owner) {
+			read_unlock(&eb->lock);
+			return;
+		}
+		read_unlock(&eb->lock);
+	}
 	if (rw == BTRFS_WRITE_LOCK_BLOCKING) {
 		BUG_ON(atomic_read(&eb->blocking_writers) != 1);
 		write_lock(&eb->lock);
@@ -78,15 +94,24 @@ void btrfs_clear_lock_blocking_rw(struct extent_buffer *eb, int rw)
  * take a spinning read lock.  This will wait for any blocking
  * writers
  */
-void btrfs_tree_read_lock(struct extent_buffer *eb)
+void btrfs_tree_read_lock(struct extent_buffer *eb, int nested)
 {
 again:
+	if (nested) {
+		read_lock(&eb->lock);
+		if (atomic_read(&eb->blocking_writers) &&
+		    current->pid == eb->lock_owner) {
+			BUG_ON(eb->lock_nested);
+			eb->lock_nested = 1;
+			read_unlock(&eb->lock);
+			return;
+		}
+		read_unlock(&eb->lock);
+	}
 	wait_event(eb->write_lock_wq, atomic_read(&eb->blocking_writers) == 0);
 	read_lock(&eb->lock);
 	if (atomic_read(&eb->blocking_writers)) {
 		read_unlock(&eb->lock);
-		wait_event(eb->write_lock_wq,
-			   atomic_read(&eb->blocking_writers) == 0);
 		goto again;
 	}
 	atomic_inc(&eb->read_locks);
@@ -129,6 +154,7 @@ int btrfs_try_tree_write_lock(struct extent_buffer *eb)
 	}
 	atomic_inc(&eb->write_locks);
 	atomic_inc(&eb->spinning_writers);
+	eb->lock_owner = current->pid;
 	return 1;
 }
 
@@ -137,6 +163,15 @@ int btrfs_try_tree_write_lock(struct extent_buffer *eb)
  */
 void btrfs_tree_read_unlock(struct extent_buffer *eb)
 {
+	if (eb->lock_nested) {
+		read_lock(&eb->lock);
+		if (eb->lock_nested && current->pid == eb->lock_owner) {
+			eb->lock_nested = 0;
+			read_unlock(&eb->lock);
+			return;
+		}
+		read_unlock(&eb->lock);
+	}
 	btrfs_assert_tree_read_locked(eb);
 	WARN_ON(atomic_read(&eb->spinning_readers) == 0);
 	atomic_dec(&eb->spinning_readers);
@@ -149,6 +184,15 @@ void btrfs_tree_read_unlock(struct extent_buffer *eb)
  */
 void btrfs_tree_read_unlock_blocking(struct extent_buffer *eb)
 {
+	if (eb->lock_nested) {
+		read_lock(&eb->lock);
+		if (eb->lock_nested && current->pid == eb->lock_owner) {
+			eb->lock_nested = 0;
+			read_unlock(&eb->lock);
+			return;
+		}
+		read_unlock(&eb->lock);
+	}
 	btrfs_assert_tree_read_locked(eb);
 	WARN_ON(atomic_read(&eb->blocking_readers) == 0);
 	if (atomic_dec_and_test(&eb->blocking_readers))
@@ -181,6 +225,7 @@ again:
 	WARN_ON(atomic_read(&eb->spinning_writers));
 	atomic_inc(&eb->spinning_writers);
 	atomic_inc(&eb->write_locks);
+	eb->lock_owner = current->pid;
 	return 0;
 }
 
diff --git a/fs/btrfs/locking.h b/fs/btrfs/locking.h
index 17247dd..c6d69e5 100644
--- a/fs/btrfs/locking.h
+++ b/fs/btrfs/locking.h
@@ -28,7 +28,7 @@ int btrfs_tree_lock(struct extent_buffer *eb);
 int btrfs_tree_unlock(struct extent_buffer *eb);
 int btrfs_try_spin_lock(struct extent_buffer *eb);
 
-void btrfs_tree_read_lock(struct extent_buffer *eb);
+void btrfs_tree_read_lock(struct extent_buffer *eb, int nested);
 void btrfs_tree_read_unlock(struct extent_buffer *eb);
 void btrfs_tree_read_unlock_blocking(struct extent_buffer *eb);
 void btrfs_set_lock_blocking_rw(struct extent_buffer *eb, int rw);
-- 
1.7.3.4

Re: [PATCH v1 05/10] Btrfs: add nested locking mode for paths

[Chris Mason]

On Thu, Dec 22, 2011 at 05:03:19PM +0100, Jan Schmidt wrote:
> From: Arne Jansen <sensille@gmx.net>
> 
> This patch adds the possibilty to read-lock an extent even if it is already
> write-locked from the same thread. btrfs_find_all_roots() needs this
> capability.

I'd rather not add a nested flag to the locking code, lets just make the
nesting explicitly allowed.

You shouldn't need locks around lock->owner.  Either your process owns
the lock (and it won't change away from your pid), or you don't own it
and it won't be your pid.  Just make sure the owner field gets cleared
when you do your final unlock.

So, if you are the owner of a write lock, you can add more write locks
or a read lock as required.

Could you please describe the case where btrfs_find_all_roots needs
this?

-chris

Re: [PATCH v1 05/10] Btrfs: add nested locking mode for paths

[Jan Schmidt]

Am 22.12.2011 20:16, schrieb Chris Mason:
> On Thu, Dec 22, 2011 at 05:03:19PM +0100, Jan Schmidt wrote:
>> From: Arne Jansen<sensille@gmx.net>
>>
>> This patch adds the possibilty to read-lock an extent even if it is already
>> write-locked from the same thread. btrfs_find_all_roots() needs this
>> capability.
>
> I'd rather not add a nested flag to the locking code, lets just make the
> nesting explicitly allowed.
>
> You shouldn't need locks around lock->owner.  Either your process owns
> the lock (and it won't change away from your pid), or you don't own it
> and it won't be your pid.  Just make sure the owner field gets cleared
> when you do your final unlock.

Sounds good. I'll make a new patch the next days and send it. I'll also 
force-push it into the for-chris branch mentioned in the cover letter. 
So if anybody pulled that already, be warned.

> So, if you are the owner of a write lock, you can add more write locks
> or a read lock as required.
>
> Could you please describe the case where btrfs_find_all_roots needs
> this?

That's part of the patch set I ripped out of Arne's qgroup patches (I 
hope he speaks up in case my explanation is garbage). The idea is that 
find_all_roots should be callable even if you are currently holding a 
lock on the data extent or the leaf in question. In order to resolve 
non-shared refs, __resolve_indirect_ref needs so call btrfs_search_slot 
on the fs-tree and that would deadlock if we called it while holding any 
of the locks.

The current application for find_all_roots does not strictly require 
this bit, the caller won't have any of these locks (although I 
experimented with a version that locked the extent before calling 
find_all_roots while writing this patch set). But qgroups and btrfs send 
will both depend on that. I decided to include that bit here to avoid 
we're having multiple patch sets adding the same code later on.

-Jan

[PATCH v1 06/10] Btrfs: add sequence numbers to delayed refs

[Jan Schmidt]

From: Arne Jansen <sensille@gmx.net>

Sequence numbers are needed to reconstruct the backrefs of a given extent to
a certain point in time. The total set of backrefs consist of the set of
backrefs recorded on disk plus the enqueued delayed refs for it that existed
at that moment.

This patch also adds a list that records all delayed refs which are
currently in the process of being added.

When walking all refs of an extent in btrfs_find_all_roots(), we freeze the
current state of delayed refs, honor anythinh up to this point and prevent
processing newer delayed refs to assert consistency.

Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
---
 fs/btrfs/delayed-ref.c |   34 +++++++++++++++++++++++
 fs/btrfs/delayed-ref.h |   70 ++++++++++++++++++++++++++++++++++++++++++++++++
 fs/btrfs/transaction.c |    4 +++
 3 files changed, 108 insertions(+), 0 deletions(-)

diff --git a/fs/btrfs/delayed-ref.c b/fs/btrfs/delayed-ref.c
index babd37b..a405db0 100644
--- a/fs/btrfs/delayed-ref.c
+++ b/fs/btrfs/delayed-ref.c
@@ -101,6 +101,11 @@ static int comp_entry(struct btrfs_delayed_ref_node *ref2,
 		return -1;
 	if (ref1->type > ref2->type)
 		return 1;
+	/* merging of sequenced refs is not allowed */
+	if (ref1->seq < ref2->seq)
+		return -1;
+	if (ref1->seq > ref2->seq)
+		return 1;
 	if (ref1->type == BTRFS_TREE_BLOCK_REF_KEY ||
 	    ref1->type == BTRFS_SHARED_BLOCK_REF_KEY) {
 		return comp_tree_refs(btrfs_delayed_node_to_tree_ref(ref2),
@@ -209,6 +214,24 @@ int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
 	return 0;
 }
 
+int btrfs_check_delayed_seq(struct btrfs_delayed_ref_root *delayed_refs,
+			    u64 seq)
+{
+	struct seq_list *elem;
+
+	assert_spin_locked(&delayed_refs->lock);
+	if (list_empty(&delayed_refs->seq_head))
+		return 0;
+
+	elem = list_first_entry(&delayed_refs->seq_head, struct seq_list, list);
+	if (seq >= elem->seq) {
+		pr_debug("holding back delayed_ref %llu, lowest is %llu (%p)\n",
+			 seq, elem->seq, delayed_refs);
+		return 1;
+	}
+	return 0;
+}
+
 int btrfs_find_ref_cluster(struct btrfs_trans_handle *trans,
 			   struct list_head *cluster, u64 start)
 {
@@ -438,6 +461,7 @@ static noinline int add_delayed_ref_head(struct btrfs_fs_info *fs_info,
 	ref->action  = 0;
 	ref->is_head = 1;
 	ref->in_tree = 1;
+	ref->seq = 0;
 
 	head_ref = btrfs_delayed_node_to_head(ref);
 	head_ref->must_insert_reserved = must_insert_reserved;
@@ -479,6 +503,7 @@ static noinline int add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
 	struct btrfs_delayed_ref_node *existing;
 	struct btrfs_delayed_tree_ref *full_ref;
 	struct btrfs_delayed_ref_root *delayed_refs;
+	u64 seq = 0;
 
 	if (action == BTRFS_ADD_DELAYED_EXTENT)
 		action = BTRFS_ADD_DELAYED_REF;
@@ -494,6 +519,10 @@ static noinline int add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
 	ref->is_head = 0;
 	ref->in_tree = 1;
 
+	if (need_ref_seq(for_cow, ref_root))
+		seq = inc_delayed_seq(delayed_refs);
+	ref->seq = seq;
+
 	full_ref = btrfs_delayed_node_to_tree_ref(ref);
 	full_ref->parent = parent;
 	full_ref->root = ref_root;
@@ -534,6 +563,7 @@ static noinline int add_delayed_data_ref(struct btrfs_fs_info *fs_info,
 	struct btrfs_delayed_ref_node *existing;
 	struct btrfs_delayed_data_ref *full_ref;
 	struct btrfs_delayed_ref_root *delayed_refs;
+	u64 seq = 0;
 
 	if (action == BTRFS_ADD_DELAYED_EXTENT)
 		action = BTRFS_ADD_DELAYED_REF;
@@ -549,6 +579,10 @@ static noinline int add_delayed_data_ref(struct btrfs_fs_info *fs_info,
 	ref->is_head = 0;
 	ref->in_tree = 1;
 
+	if (need_ref_seq(for_cow, ref_root))
+		seq = inc_delayed_seq(delayed_refs);
+	ref->seq = seq;
+
 	full_ref = btrfs_delayed_node_to_data_ref(ref);
 	full_ref->parent = parent;
 	full_ref->root = ref_root;
diff --git a/fs/btrfs/delayed-ref.h b/fs/btrfs/delayed-ref.h
index a5fb2bc..174416f 100644
--- a/fs/btrfs/delayed-ref.h
+++ b/fs/btrfs/delayed-ref.h
@@ -33,6 +33,9 @@ struct btrfs_delayed_ref_node {
 	/* the size of the extent */
 	u64 num_bytes;
 
+	/* seq number to keep track of insertion order */
+	u64 seq;
+
 	/* ref count on this data structure */
 	atomic_t refs;
 
@@ -136,6 +139,20 @@ struct btrfs_delayed_ref_root {
 	int flushing;
 
 	u64 run_delayed_start;
+
+	/*
+	 * seq number of delayed refs. We need to know if a backref was being
+	 * added before the currently processed ref or afterwards.
+	 */
+	u64 seq;
+
+	/*
+	 * seq_list holds a list of all seq numbers that are currently being
+	 * added to the list. While walking backrefs (btrfs_find_all_roots,
+	 * qgroups), which might take some time, no newer ref must be processed,
+	 * as it might influence the outcome of the walk.
+	 */
+	struct list_head seq_head;
 };
 
 static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
@@ -171,6 +188,59 @@ int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans,
 			   struct btrfs_delayed_ref_head *head);
 int btrfs_find_ref_cluster(struct btrfs_trans_handle *trans,
 			   struct list_head *cluster, u64 search_start);
+
+struct seq_list {
+	struct list_head list;
+	u64 seq;
+};
+
+static inline u64 inc_delayed_seq(struct btrfs_delayed_ref_root *delayed_refs)
+{
+	assert_spin_locked(&delayed_refs->lock);
+	++delayed_refs->seq;
+	return delayed_refs->seq;
+}
+
+static inline void
+btrfs_get_delayed_seq(struct btrfs_delayed_ref_root *delayed_refs,
+		      struct seq_list *elem)
+{
+	assert_spin_locked(&delayed_refs->lock);
+	elem->seq = delayed_refs->seq;
+	list_add_tail(&elem->list, &delayed_refs->seq_head);
+}
+
+static inline void
+btrfs_put_delayed_seq(struct btrfs_delayed_ref_root *delayed_refs,
+		      struct seq_list *elem)
+{
+	spin_lock(&delayed_refs->lock);
+	list_del(&elem->list);
+	spin_unlock(&delayed_refs->lock);
+}
+
+int btrfs_check_delayed_seq(struct btrfs_delayed_ref_root *delayed_refs,
+			    u64 seq);
+
+/*
+ * delayed refs with a ref_seq > 0 must be held back during backref walking.
+ * this only applies to items in one of the fs-trees. for_cow items never need
+ * to be held back, so they won't get a ref_seq number.
+ */
+static inline int need_ref_seq(int for_cow, u64 rootid)
+{
+	if (for_cow)
+		return 0;
+
+	if (rootid == BTRFS_FS_TREE_OBJECTID)
+		return 1;
+
+	if ((s64)rootid >= (s64)BTRFS_FIRST_FREE_OBJECTID)
+		return 1;
+
+	return 0;
+}
+
 /*
  * a node might live in a head or a regular ref, this lets you
  * test for the proper type to use.
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index a2bfedc..31a7393 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -36,6 +36,8 @@ static noinline void put_transaction(struct btrfs_transaction *transaction)
 	WARN_ON(atomic_read(&transaction->use_count) == 0);
 	if (atomic_dec_and_test(&transaction->use_count)) {
 		BUG_ON(!list_empty(&transaction->list));
+		WARN_ON(transaction->delayed_refs.root.rb_node);
+		WARN_ON(!list_empty(&transaction->delayed_refs.seq_head));
 		memset(transaction, 0, sizeof(*transaction));
 		kmem_cache_free(btrfs_transaction_cachep, transaction);
 	}
@@ -108,8 +110,10 @@ loop:
 	cur_trans->delayed_refs.num_heads = 0;
 	cur_trans->delayed_refs.flushing = 0;
 	cur_trans->delayed_refs.run_delayed_start = 0;
+	cur_trans->delayed_refs.seq = 1;
 	spin_lock_init(&cur_trans->commit_lock);
 	spin_lock_init(&cur_trans->delayed_refs.lock);
+	INIT_LIST_HEAD(&cur_trans->delayed_refs.seq_head);
 
 	INIT_LIST_HEAD(&cur_trans->pending_snapshots);
 	list_add_tail(&cur_trans->list, &root->fs_info->trans_list);
-- 
1.7.3.4

[PATCH v1 07/10] Btrfs: put back delayed refs that are too new

[Jan Schmidt]

From: Arne Jansen <sensille@gmx.net>

When processing a delayed ref, first check if there are still old refs in
the process of being added. If so, put this ref back to the tree. To avoid
looping on this ref, choose a newer one in the next loop.
btrfs_find_ref_cluster has to take care of that.

Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
---
 fs/btrfs/delayed-ref.c |   43 +++++++++++++++++++++++++------------------
 fs/btrfs/extent-tree.c |   27 ++++++++++++++++++++++-----
 2 files changed, 47 insertions(+), 23 deletions(-)

diff --git a/fs/btrfs/delayed-ref.c b/fs/btrfs/delayed-ref.c
index a405db0..ee18198 100644
--- a/fs/btrfs/delayed-ref.c
+++ b/fs/btrfs/delayed-ref.c
@@ -155,16 +155,22 @@ static struct btrfs_delayed_ref_node *tree_insert(struct rb_root *root,
 
 /*
  * find an head entry based on bytenr. This returns the delayed ref
- * head if it was able to find one, or NULL if nothing was in that spot
+ * head if it was able to find one, or NULL if nothing was in that spot.
+ * If return_bigger is given, the next bigger entry is returned if no exact
+ * match is found.
  */
 static struct btrfs_delayed_ref_node *find_ref_head(struct rb_root *root,
 				  u64 bytenr,
-				  struct btrfs_delayed_ref_node **last)
+				  struct btrfs_delayed_ref_node **last,
+				  int return_bigger)
 {
-	struct rb_node *n = root->rb_node;
+	struct rb_node *n;
 	struct btrfs_delayed_ref_node *entry;
-	int cmp;
+	int cmp = 0;
 
+again:
+	n = root->rb_node;
+	entry = NULL;
 	while (n) {
 		entry = rb_entry(n, struct btrfs_delayed_ref_node, rb_node);
 		WARN_ON(!entry->in_tree);
@@ -187,6 +193,19 @@ static struct btrfs_delayed_ref_node *find_ref_head(struct rb_root *root,
 		else
 			return entry;
 	}
+	if (entry && return_bigger) {
+		if (cmp > 0) {
+			n = rb_next(&entry->rb_node);
+			if (!n)
+				n = rb_first(root);
+			entry = rb_entry(n, struct btrfs_delayed_ref_node,
+					 rb_node);
+			bytenr = entry->bytenr;
+			return_bigger = 0;
+			goto again;
+		}
+		return entry;
+	}
 	return NULL;
 }
 
@@ -246,20 +265,8 @@ int btrfs_find_ref_cluster(struct btrfs_trans_handle *trans,
 		node = rb_first(&delayed_refs->root);
 	} else {
 		ref = NULL;
-		find_ref_head(&delayed_refs->root, start, &ref);
+		find_ref_head(&delayed_refs->root, start + 1, &ref, 1);
 		if (ref) {
-			struct btrfs_delayed_ref_node *tmp;
-
-			node = rb_prev(&ref->rb_node);
-			while (node) {
-				tmp = rb_entry(node,
-					       struct btrfs_delayed_ref_node,
-					       rb_node);
-				if (tmp->bytenr < start)
-					break;
-				ref = tmp;
-				node = rb_prev(&ref->rb_node);
-			}
 			node = &ref->rb_node;
 		} else
 			node = rb_first(&delayed_refs->root);
@@ -748,7 +755,7 @@ btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr)
 	struct btrfs_delayed_ref_root *delayed_refs;
 
 	delayed_refs = &trans->transaction->delayed_refs;
-	ref = find_ref_head(&delayed_refs->root, bytenr, NULL);
+	ref = find_ref_head(&delayed_refs->root, bytenr, NULL, 0);
 	if (ref)
 		return btrfs_delayed_node_to_head(ref);
 	return NULL;
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index dc8b9a8..bbcca12 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2237,6 +2237,28 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
 		}
 
 		/*
+		 * locked_ref is the head node, so we have to go one
+		 * node back for any delayed ref updates
+		 */
+		ref = select_delayed_ref(locked_ref);
+
+		if (ref && ref->seq &&
+		    btrfs_check_delayed_seq(delayed_refs, ref->seq)) {
+			/*
+			 * there are still refs with lower seq numbers in the
+			 * process of being added. Don't run this ref yet.
+			 */
+			list_del_init(&locked_ref->cluster);
+			mutex_unlock(&locked_ref->mutex);
+			locked_ref = NULL;
+			delayed_refs->num_heads_ready++;
+			spin_unlock(&delayed_refs->lock);
+			cond_resched();
+			spin_lock(&delayed_refs->lock);
+			continue;
+		}
+
+		/*
 		 * record the must insert reserved flag before we
 		 * drop the spin lock.
 		 */
@@ -2246,11 +2268,6 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
 		extent_op = locked_ref->extent_op;
 		locked_ref->extent_op = NULL;
 
-		/*
-		 * locked_ref is the head node, so we have to go one
-		 * node back for any delayed ref updates
-		 */
-		ref = select_delayed_ref(locked_ref);
 		if (!ref) {
 			/* All delayed refs have been processed, Go ahead
 			 * and send the head node to run_one_delayed_ref,
-- 
1.7.3.4

[PATCH v1 08/10] Btrfs: add waitqueue instead of doing busy waiting for more delayed refs

[Jan Schmidt]

Now that we may be holding back delayed refs for a limited period, we
might end up having no runnable delayed refs. Without this commit, we'd
do busy waiting in that thread until another (runnable) ref arives.
Instead, we're detecting this situation and use a waitqueue, such that
we only try to run more refs after
	a) another runnable ref was added  or
	b) delayed refs are no longer held back

Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
---
 fs/btrfs/delayed-ref.c |    8 ++++++
 fs/btrfs/delayed-ref.h |    7 +++++
 fs/btrfs/extent-tree.c |   59 +++++++++++++++++++++++++++++++++++++++++++++++-
 fs/btrfs/transaction.c |    1 +
 4 files changed, 74 insertions(+), 1 deletions(-)

diff --git a/fs/btrfs/delayed-ref.c b/fs/btrfs/delayed-ref.c
index ee18198..66e4f29 100644
--- a/fs/btrfs/delayed-ref.c
+++ b/fs/btrfs/delayed-ref.c
@@ -664,6 +664,9 @@ int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info,
 				   num_bytes, parent, ref_root, level, action,
 				   for_cow);
 	BUG_ON(ret);
+	if (!need_ref_seq(for_cow, ref_root) &&
+	    waitqueue_active(&delayed_refs->seq_wait))
+		wake_up(&delayed_refs->seq_wait);
 	spin_unlock(&delayed_refs->lock);
 	return 0;
 }
@@ -712,6 +715,9 @@ int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info,
 				   num_bytes, parent, ref_root, owner, offset,
 				   action, for_cow);
 	BUG_ON(ret);
+	if (!need_ref_seq(for_cow, ref_root) &&
+	    waitqueue_active(&delayed_refs->seq_wait))
+		wake_up(&delayed_refs->seq_wait);
 	spin_unlock(&delayed_refs->lock);
 	return 0;
 }
@@ -739,6 +745,8 @@ int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info,
 				   extent_op->is_data);
 	BUG_ON(ret);
 
+	if (waitqueue_active(&delayed_refs->seq_wait))
+		wake_up(&delayed_refs->seq_wait);
 	spin_unlock(&delayed_refs->lock);
 	return 0;
 }
diff --git a/fs/btrfs/delayed-ref.h b/fs/btrfs/delayed-ref.h
index 174416f..d8f244d 100644
--- a/fs/btrfs/delayed-ref.h
+++ b/fs/btrfs/delayed-ref.h
@@ -153,6 +153,12 @@ struct btrfs_delayed_ref_root {
 	 * as it might influence the outcome of the walk.
 	 */
 	struct list_head seq_head;
+
+	/*
+	 * when the only refs we have in the list must not be processed, we want
+	 * to wait for more refs to show up or for the end of backref walking.
+	 */
+	wait_queue_head_t seq_wait;
 };
 
 static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref)
@@ -216,6 +222,7 @@ btrfs_put_delayed_seq(struct btrfs_delayed_ref_root *delayed_refs,
 {
 	spin_lock(&delayed_refs->lock);
 	list_del(&elem->list);
+	wake_up(&delayed_refs->seq_wait);
 	spin_unlock(&delayed_refs->lock);
 }
 
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index bbcca12..0a435e2 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2300,7 +2300,12 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
 		ref->in_tree = 0;
 		rb_erase(&ref->rb_node, &delayed_refs->root);
 		delayed_refs->num_entries--;
-
+		/*
+		 * we modified num_entries, but as we're currently running
+		 * delayed refs, skip
+		 *     wake_up(&delayed_refs->seq_wait);
+		 * here.
+		 */
 		spin_unlock(&delayed_refs->lock);
 
 		ret = run_one_delayed_ref(trans, root, ref, extent_op,
@@ -2317,6 +2322,23 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
 	return count;
 }
 
+
+static void wait_for_more_refs(struct btrfs_delayed_ref_root *delayed_refs,
+			unsigned long num_refs)
+{
+	struct list_head *first_seq = delayed_refs->seq_head.next;
+
+	spin_unlock(&delayed_refs->lock);
+	pr_debug("waiting for more refs (num %ld, first %p)\n",
+		 num_refs, first_seq);
+	wait_event(delayed_refs->seq_wait,
+		   num_refs != delayed_refs->num_entries ||
+		   delayed_refs->seq_head.next != first_seq);
+	pr_debug("done waiting for more refs (num %ld, first %p)\n",
+		 delayed_refs->num_entries, delayed_refs->seq_head.next);
+	spin_lock(&delayed_refs->lock);
+}
+
 /*
  * this starts processing the delayed reference count updates and
  * extent insertions we have queued up so far.  count can be
@@ -2332,8 +2354,11 @@ int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
 	struct btrfs_delayed_ref_node *ref;
 	struct list_head cluster;
 	int ret;
+	u64 delayed_start;
 	int run_all = count == (unsigned long)-1;
 	int run_most = 0;
+	unsigned long num_refs = 0;
+	int consider_waiting;
 
 	if (root == root->fs_info->extent_root)
 		root = root->fs_info->tree_root;
@@ -2341,6 +2366,7 @@ int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
 	delayed_refs = &trans->transaction->delayed_refs;
 	INIT_LIST_HEAD(&cluster);
 again:
+	consider_waiting = 0;
 	spin_lock(&delayed_refs->lock);
 	if (count == 0) {
 		count = delayed_refs->num_entries * 2;
@@ -2357,11 +2383,35 @@ again:
 		 * of refs to process starting at the first one we are able to
 		 * lock
 		 */
+		delayed_start = delayed_refs->run_delayed_start;
 		ret = btrfs_find_ref_cluster(trans, &cluster,
 					     delayed_refs->run_delayed_start);
 		if (ret)
 			break;
 
+		if (delayed_start >= delayed_refs->run_delayed_start) {
+			if (consider_waiting == 0) {
+				/*
+				 * btrfs_find_ref_cluster looped. let's do one
+				 * more cycle. if we don't run any delayed ref
+				 * during that cycle (because we can't because
+				 * all of them are blocked) and if the number of
+				 * refs doesn't change, we avoid busy waiting.
+				 */
+				consider_waiting = 1;
+				num_refs = delayed_refs->num_entries;
+			} else {
+				wait_for_more_refs(delayed_refs, num_refs);
+				/*
+				 * after waiting, things have changed. we
+				 * dropped the lock and someone else might have
+				 * run some refs, built new clusters and so on.
+				 * therefore, we restart staleness detection.
+				 */
+				consider_waiting = 0;
+			}
+		}
+
 		ret = run_clustered_refs(trans, root, &cluster);
 		BUG_ON(ret < 0);
 
@@ -2369,6 +2419,11 @@ again:
 
 		if (count == 0)
 			break;
+
+		if (ret || delayed_refs->run_delayed_start == 0) {
+			/* refs were run, let's reset staleness detection */
+			consider_waiting = 0;
+		}
 	}
 
 	if (run_all) {
@@ -4933,6 +4988,8 @@ static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans,
 	rb_erase(&head->node.rb_node, &delayed_refs->root);
 
 	delayed_refs->num_entries--;
+	if (waitqueue_active(&delayed_refs->seq_wait))
+		wake_up(&delayed_refs->seq_wait);
 
 	/*
 	 * we don't take a ref on the node because we're removing it from the
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 31a7393..04c5c7c 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -111,6 +111,7 @@ loop:
 	cur_trans->delayed_refs.flushing = 0;
 	cur_trans->delayed_refs.run_delayed_start = 0;
 	cur_trans->delayed_refs.seq = 1;
+	init_waitqueue_head(&cur_trans->delayed_refs.seq_wait);
 	spin_lock_init(&cur_trans->commit_lock);
 	spin_lock_init(&cur_trans->delayed_refs.lock);
 	INIT_LIST_HEAD(&cur_trans->delayed_refs.seq_head);
-- 
1.7.3.4

[PATCH v1 09/10] Btrfs: added btrfs_find_all_roots()

[Jan Schmidt]

This function gets a byte number (a data extent), collects all the leafs
pointing to it and walks up the trees to find all fs roots pointing to those
leafs. It also returns the list of all leafs pointing to that extent.

It does proper locking for the involved trees, can be used on busy file
systems and honors delayed refs.

Signed-off-by: Arne Jansen <sensille@gmx.net>
Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
---
 fs/btrfs/backref.c |  784 ++++++++++++++++++++++++++++++++++++++++++++++++++++
 fs/btrfs/backref.h |    5 +
 2 files changed, 789 insertions(+), 0 deletions(-)

diff --git a/fs/btrfs/backref.c b/fs/btrfs/backref.c
index 22c64ff..e01790e 100644
--- a/fs/btrfs/backref.c
+++ b/fs/btrfs/backref.c
@@ -19,6 +19,9 @@
 #include "ctree.h"
 #include "disk-io.h"
 #include "backref.h"
+#include "ulist.h"
+#include "transaction.h"
+#include "delayed-ref.h"
 
 struct __data_ref {
 	struct list_head list;
@@ -32,6 +35,787 @@ struct __shared_ref {
 	u64 disk_byte;
 };
 
+/*
+ * this structure records all encountered refs on the way up to the root
+ */
+struct __prelim_ref {
+	struct list_head list;
+	u64 root_id;
+	struct btrfs_key key;
+	int level;
+	int count;
+	u64 parent;
+	u64 wanted_disk_byte;
+};
+
+static int __add_prelim_ref(struct list_head *head, u64 root_id,
+			    struct btrfs_key *key, int level, u64 parent,
+			    u64 wanted_disk_byte, int count)
+{
+	struct __prelim_ref *ref;
+
+	/* in case we're adding delayed refs, we're holding the refs spinlock */
+	ref = kmalloc(sizeof(*ref), GFP_ATOMIC);
+	if (!ref)
+		return -ENOMEM;
+
+	ref->root_id = root_id;
+	if (key)
+		ref->key = *key;
+	else
+		memset(&ref->key, 0, sizeof(ref->key));
+
+	ref->level = level;
+	ref->count = count;
+	ref->parent = parent;
+	ref->wanted_disk_byte = wanted_disk_byte;
+	list_add_tail(&ref->list, head);
+
+	return 0;
+}
+
+static int add_all_parents(struct btrfs_root *root, struct btrfs_path *path,
+				struct ulist *parents,
+				struct extent_buffer *eb, int level,
+				u64 wanted_objectid, u64 wanted_disk_byte)
+{
+	int ret;
+	int slot;
+	struct btrfs_file_extent_item *fi;
+	struct btrfs_key key;
+	u64 disk_byte;
+
+add_parent:
+	ret = ulist_add(parents, eb->start, 0, GFP_NOFS);
+	if (ret < 0)
+		return ret;
+
+	if (level != 0)
+		return 0;
+
+	/*
+	 * if the current leaf is full with EXTENT_DATA items, we must
+	 * check the next one if that holds a reference as well.
+	 * ref->count cannot be used to skip this check.
+	 * repeat this until we don't find any additional EXTENT_DATA items.
+	 */
+	while (1) {
+		ret = btrfs_next_leaf(root, path);
+		if (ret < 0)
+			return ret;
+		if (ret)
+			return 0;
+
+		eb = path->nodes[0];
+		for (slot = 0; slot < btrfs_header_nritems(eb); ++slot) {
+			btrfs_item_key_to_cpu(eb, &key, slot);
+			if (key.objectid != wanted_objectid ||
+			    key.type != BTRFS_EXTENT_DATA_KEY)
+				return 0;
+			fi = btrfs_item_ptr(eb, slot,
+						struct btrfs_file_extent_item);
+			disk_byte = btrfs_file_extent_disk_bytenr(eb, fi);
+			if (disk_byte == wanted_disk_byte)
+				goto add_parent;
+		}
+	}
+
+	return 0;
+}
+
+/*
+ * resolve an indirect backref in the form (root_id, key, level)
+ * to a logical address
+ */
+static int __resolve_indirect_ref(struct btrfs_fs_info *fs_info,
+					struct __prelim_ref *ref,
+					struct ulist *parents)
+{
+	struct btrfs_path *path;
+	struct btrfs_root *root;
+	struct btrfs_key root_key;
+	struct btrfs_key key = {0};
+	struct extent_buffer *eb;
+	int ret = 0;
+	int root_level;
+	int level = ref->level;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	root_key.objectid = ref->root_id;
+	root_key.type = BTRFS_ROOT_ITEM_KEY;
+	root_key.offset = (u64)-1;
+	root = btrfs_read_fs_root_no_name(fs_info, &root_key);
+	if (IS_ERR(root)) {
+		ret = PTR_ERR(root);
+		goto out;
+	}
+
+	rcu_read_lock();
+	root_level = btrfs_header_level(root->node);
+	rcu_read_unlock();
+
+	if (root_level + 1 == level)
+		goto out;
+
+	path->lowest_level = level;
+	path->nested = 1;
+	ret = btrfs_search_slot(NULL, root, &ref->key, path, 0, 0);
+	pr_debug("search slot in root %llu (level %d, ref count %d) returned "
+		 "%d for key (%llu %u %llu)\n",
+		 (unsigned long long)ref->root_id, level, ref->count, ret,
+		 (unsigned long long)ref->key.objectid, ref->key.type,
+		 (unsigned long long)ref->key.offset);
+	if (ret < 0)
+		goto out;
+
+	eb = path->nodes[level];
+	if (!eb) {
+		WARN_ON(1);
+		ret = 1;
+		goto out;
+	}
+
+	if (level == 0) {
+		if (ret == 1 && path->slots[0] >= btrfs_header_nritems(eb)) {
+			ret = btrfs_next_leaf(root, path);
+			if (ret)
+				goto out;
+			eb = path->nodes[0];
+		}
+
+		btrfs_item_key_to_cpu(eb, &key, path->slots[0]);
+	}
+
+	/* the last two parameters will only be used for level == 0 */
+	ret = add_all_parents(root, path, parents, eb, level, key.objectid,
+				ref->wanted_disk_byte);
+out:
+	btrfs_free_path(path);
+	return ret;
+}
+
+/*
+ * resolve all indirect backrefs from the list
+ */
+static int __resolve_indirect_refs(struct btrfs_fs_info *fs_info,
+				   struct list_head *head)
+{
+	int err;
+	int ret = 0;
+	struct __prelim_ref *ref;
+	struct __prelim_ref *ref_safe;
+	struct __prelim_ref *new_ref;
+	struct ulist *parents;
+	struct ulist_node *node;
+
+	parents = ulist_alloc(GFP_NOFS);
+	if (!parents)
+		return -ENOMEM;
+
+	/*
+	 * _safe allows us to insert directly after the current item without
+	 * iterating over the newly inserted items.
+	 * we're also allowed to re-assign ref during iteration.
+	 */
+	list_for_each_entry_safe(ref, ref_safe, head, list) {
+		if (ref->parent)	/* already direct */
+			continue;
+		if (ref->count == 0)
+			continue;
+		err = __resolve_indirect_ref(fs_info, ref, parents);
+		if (err) {
+			if (ret == 0)
+				ret = err;
+			continue;
+		}
+
+		/* we put the first parent into the ref at hand */
+		node = ulist_next(parents, NULL);
+		ref->parent = node ? node->val : 0;
+
+		/* additional parents require new refs being added here */
+		while ((node = ulist_next(parents, node))) {
+			new_ref = kmalloc(sizeof(*new_ref), GFP_NOFS);
+			if (!new_ref) {
+				ret = -ENOMEM;
+				break;
+			}
+			memcpy(new_ref, ref, sizeof(*ref));
+			new_ref->parent = node->val;
+			list_add(&new_ref->list, &ref->list);
+		}
+		ulist_reinit(parents);
+	}
+
+	ulist_free(parents);
+	return ret;
+}
+
+/*
+ * merge two lists of backrefs and adjust counts accordingly
+ *
+ * mode = 1: merge identical keys, if key is set
+ * mode = 2: merge identical parents
+ */
+static int __merge_refs(struct list_head *head, int mode)
+{
+	struct list_head *pos1;
+
+	list_for_each(pos1, head) {
+		struct list_head *n2;
+		struct list_head *pos2;
+		struct __prelim_ref *ref1;
+
+		ref1 = list_entry(pos1, struct __prelim_ref, list);
+
+		if (mode == 1 && ref1->key.type == 0)
+			continue;
+		for (pos2 = pos1->next, n2 = pos2->next; pos2 != head;
+		     pos2 = n2, n2 = pos2->next) {
+			struct __prelim_ref *ref2;
+
+			ref2 = list_entry(pos2, struct __prelim_ref, list);
+
+			if (mode == 1) {
+				if (memcmp(&ref1->key, &ref2->key,
+					   sizeof(ref1->key)) ||
+				    ref1->level != ref2->level ||
+				    ref1->root_id != ref2->root_id)
+					continue;
+				ref1->count += ref2->count;
+			} else {
+				if (ref1->parent != ref2->parent)
+					continue;
+				ref1->count += ref2->count;
+			}
+			list_del(&ref2->list);
+			kfree(ref2);
+		}
+
+	}
+	return 0;
+}
+
+/*
+ * add all currently queued delayed refs from this head whose seq nr is
+ * smaller or equal that seq to the list
+ */
+static int __add_delayed_refs(struct btrfs_delayed_ref_head *head, u64 seq,
+			      struct btrfs_key *info_key,
+			      struct list_head *prefs)
+{
+	struct btrfs_delayed_extent_op *extent_op = head->extent_op;
+	struct rb_node *n = &head->node.rb_node;
+	int sgn;
+	int ret;
+
+	if (extent_op && extent_op->update_key)
+		btrfs_disk_key_to_cpu(info_key, &extent_op->key);
+
+	while ((n = rb_prev(n))) {
+		struct btrfs_delayed_ref_node *node;
+		node = rb_entry(n, struct btrfs_delayed_ref_node,
+				rb_node);
+		if (node->bytenr != head->node.bytenr)
+			break;
+		WARN_ON(node->is_head);
+
+		if (node->seq > seq)
+			continue;
+
+		switch (node->action) {
+		case BTRFS_ADD_DELAYED_EXTENT:
+		case BTRFS_UPDATE_DELAYED_HEAD:
+			WARN_ON(1);
+			continue;
+		case BTRFS_ADD_DELAYED_REF:
+			sgn = 1;
+			break;
+		case BTRFS_DROP_DELAYED_REF:
+			sgn = -1;
+			break;
+		default:
+			BUG_ON(1);
+		}
+		switch (node->type) {
+		case BTRFS_TREE_BLOCK_REF_KEY: {
+			struct btrfs_delayed_tree_ref *ref;
+
+			ref = btrfs_delayed_node_to_tree_ref(node);
+			ret = __add_prelim_ref(prefs, ref->root, info_key,
+					       ref->level + 1, 0, node->bytenr,
+					       node->ref_mod * sgn);
+			break;
+		}
+		case BTRFS_SHARED_BLOCK_REF_KEY: {
+			struct btrfs_delayed_tree_ref *ref;
+
+			ref = btrfs_delayed_node_to_tree_ref(node);
+			ret = __add_prelim_ref(prefs, ref->root, info_key,
+					       ref->level + 1, ref->parent,
+					       node->bytenr,
+					       node->ref_mod * sgn);
+			break;
+		}
+		case BTRFS_EXTENT_DATA_REF_KEY: {
+			struct btrfs_delayed_data_ref *ref;
+			struct btrfs_key key;
+
+			ref = btrfs_delayed_node_to_data_ref(node);
+
+			key.objectid = ref->objectid;
+			key.type = BTRFS_EXTENT_DATA_KEY;
+			key.offset = ref->offset;
+			ret = __add_prelim_ref(prefs, ref->root, &key, 0, 0,
+					       node->bytenr,
+					       node->ref_mod * sgn);
+			break;
+		}
+		case BTRFS_SHARED_DATA_REF_KEY: {
+			struct btrfs_delayed_data_ref *ref;
+			struct btrfs_key key;
+
+			ref = btrfs_delayed_node_to_data_ref(node);
+
+			key.objectid = ref->objectid;
+			key.type = BTRFS_EXTENT_DATA_KEY;
+			key.offset = ref->offset;
+			ret = __add_prelim_ref(prefs, ref->root, &key, 0,
+					       ref->parent, node->bytenr,
+					       node->ref_mod * sgn);
+			break;
+		}
+		default:
+			WARN_ON(1);
+		}
+		BUG_ON(ret);
+	}
+
+	return 0;
+}
+
+/*
+ * add all inline backrefs for bytenr to the list
+ */
+static int __add_inline_refs(struct btrfs_fs_info *fs_info,
+			     struct btrfs_path *path, u64 bytenr,
+			     struct btrfs_key *info_key, int *info_level,
+			     struct list_head *prefs)
+{
+	int ret;
+	int slot;
+	struct extent_buffer *leaf;
+	struct btrfs_key key;
+	unsigned long ptr;
+	unsigned long end;
+	struct btrfs_extent_item *ei;
+	u64 flags;
+	u64 item_size;
+
+	/*
+	 * enumerate all inline refs
+	 */
+	leaf = path->nodes[0];
+	slot = path->slots[0] - 1;
+
+	item_size = btrfs_item_size_nr(leaf, slot);
+	BUG_ON(item_size < sizeof(*ei));
+
+	ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item);
+	flags = btrfs_extent_flags(leaf, ei);
+
+	ptr = (unsigned long)(ei + 1);
+	end = (unsigned long)ei + item_size;
+
+	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
+		struct btrfs_tree_block_info *info;
+		struct btrfs_disk_key disk_key;
+
+		info = (struct btrfs_tree_block_info *)ptr;
+		*info_level = btrfs_tree_block_level(leaf, info);
+		btrfs_tree_block_key(leaf, info, &disk_key);
+		btrfs_disk_key_to_cpu(info_key, &disk_key);
+		ptr += sizeof(struct btrfs_tree_block_info);
+		BUG_ON(ptr > end);
+	} else {
+		BUG_ON(!(flags & BTRFS_EXTENT_FLAG_DATA));
+	}
+
+	while (ptr < end) {
+		struct btrfs_extent_inline_ref *iref;
+		u64 offset;
+		int type;
+
+		iref = (struct btrfs_extent_inline_ref *)ptr;
+		type = btrfs_extent_inline_ref_type(leaf, iref);
+		offset = btrfs_extent_inline_ref_offset(leaf, iref);
+
+		switch (type) {
+		case BTRFS_SHARED_BLOCK_REF_KEY:
+			ret = __add_prelim_ref(prefs, 0, info_key,
+						*info_level + 1, offset,
+						bytenr, 1);
+			break;
+		case BTRFS_SHARED_DATA_REF_KEY: {
+			struct btrfs_shared_data_ref *sdref;
+			int count;
+
+			sdref = (struct btrfs_shared_data_ref *)(iref + 1);
+			count = btrfs_shared_data_ref_count(leaf, sdref);
+			ret = __add_prelim_ref(prefs, 0, NULL, 0, offset,
+					       bytenr, count);
+			break;
+		}
+		case BTRFS_TREE_BLOCK_REF_KEY:
+			ret = __add_prelim_ref(prefs, offset, info_key,
+					       *info_level + 1, 0, bytenr, 1);
+			break;
+		case BTRFS_EXTENT_DATA_REF_KEY: {
+			struct btrfs_extent_data_ref *dref;
+			int count;
+			u64 root;
+
+			dref = (struct btrfs_extent_data_ref *)(&iref->offset);
+			count = btrfs_extent_data_ref_count(leaf, dref);
+			key.objectid = btrfs_extent_data_ref_objectid(leaf,
+								      dref);
+			key.type = BTRFS_EXTENT_DATA_KEY;
+			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
+			root = btrfs_extent_data_ref_root(leaf, dref);
+			ret = __add_prelim_ref(prefs, root, &key, 0, 0, bytenr,
+						count);
+			break;
+		}
+		default:
+			WARN_ON(1);
+		}
+		BUG_ON(ret);
+		ptr += btrfs_extent_inline_ref_size(type);
+	}
+
+	return 0;
+}
+
+/*
+ * add all non-inline backrefs for bytenr to the list
+ */
+static int __add_keyed_refs(struct btrfs_fs_info *fs_info,
+			    struct btrfs_path *path, u64 bytenr,
+			    struct btrfs_key *info_key, int info_level,
+			    struct list_head *prefs)
+{
+	struct btrfs_root *extent_root = fs_info->extent_root;
+	int ret;
+	int slot;
+	struct extent_buffer *leaf;
+	struct btrfs_key key;
+
+	while (1) {
+		ret = btrfs_next_item(extent_root, path);
+		if (ret < 0)
+			break;
+		if (ret) {
+			ret = 0;
+			break;
+		}
+
+		slot = path->slots[0];
+		leaf = path->nodes[0];
+		btrfs_item_key_to_cpu(leaf, &key, slot);
+
+		if (key.objectid != bytenr)
+			break;
+		if (key.type < BTRFS_TREE_BLOCK_REF_KEY)
+			continue;
+		if (key.type > BTRFS_SHARED_DATA_REF_KEY)
+			break;
+
+		switch (key.type) {
+		case BTRFS_SHARED_BLOCK_REF_KEY:
+			ret = __add_prelim_ref(prefs, 0, info_key,
+						info_level + 1, key.offset,
+						bytenr, 1);
+			break;
+		case BTRFS_SHARED_DATA_REF_KEY: {
+			struct btrfs_shared_data_ref *sdref;
+			int count;
+
+			sdref = btrfs_item_ptr(leaf, slot,
+					      struct btrfs_shared_data_ref);
+			count = btrfs_shared_data_ref_count(leaf, sdref);
+			ret = __add_prelim_ref(prefs, 0, NULL, 0, key.offset,
+						bytenr, count);
+			break;
+		}
+		case BTRFS_TREE_BLOCK_REF_KEY:
+			ret = __add_prelim_ref(prefs, key.offset, info_key,
+						info_level + 1, 0, bytenr, 1);
+			break;
+		case BTRFS_EXTENT_DATA_REF_KEY: {
+			struct btrfs_extent_data_ref *dref;
+			int count;
+			u64 root;
+
+			dref = btrfs_item_ptr(leaf, slot,
+					      struct btrfs_extent_data_ref);
+			count = btrfs_extent_data_ref_count(leaf, dref);
+			key.objectid = btrfs_extent_data_ref_objectid(leaf,
+								      dref);
+			key.type = BTRFS_EXTENT_DATA_KEY;
+			key.offset = btrfs_extent_data_ref_offset(leaf, dref);
+			root = btrfs_extent_data_ref_root(leaf, dref);
+			ret = __add_prelim_ref(prefs, root, &key, 0, 0,
+						bytenr, count);
+			break;
+		}
+		default:
+			WARN_ON(1);
+		}
+		BUG_ON(ret);
+	}
+
+	return ret;
+}
+
+/*
+ * this adds all existing backrefs (inline backrefs, backrefs and delayed
+ * refs) for the given bytenr to the refs list, merges duplicates and resolves
+ * indirect refs to their parent bytenr.
+ * When roots are found, they're added to the roots list
+ *
+ * FIXME some caching might speed things up
+ */
+static int find_parent_nodes(struct btrfs_trans_handle *trans,
+			     struct btrfs_fs_info *fs_info, u64 bytenr,
+			     u64 seq, struct ulist *refs, struct ulist *roots)
+{
+	struct btrfs_key key;
+	struct btrfs_path *path;
+	struct btrfs_key info_key = { 0 };
+	struct btrfs_delayed_ref_root *delayed_refs = NULL;
+	struct btrfs_delayed_ref_head *head = NULL;
+	int info_level = 0;
+	int ret;
+	struct list_head prefs_delayed;
+	struct list_head prefs;
+	struct __prelim_ref *ref;
+
+	INIT_LIST_HEAD(&prefs);
+	INIT_LIST_HEAD(&prefs_delayed);
+
+	key.objectid = bytenr;
+	key.type = BTRFS_EXTENT_ITEM_KEY;
+	key.offset = (u64)-1;
+
+	path = btrfs_alloc_path();
+	if (!path)
+		return -ENOMEM;
+
+	/*
+	 * grab both a lock on the path and a lock on the delayed ref head.
+	 * We need both to get a consistent picture of how the refs look
+	 * at a specified point in time
+	 */
+again:
+	ret = btrfs_search_slot(trans, fs_info->extent_root, &key, path, 0, 0);
+	if (ret < 0)
+		goto out;
+	BUG_ON(ret == 0);
+
+	/*
+	 * look if there are updates for this ref queued and lock the head
+	 */
+	delayed_refs = &trans->transaction->delayed_refs;
+	spin_lock(&delayed_refs->lock);
+	head = btrfs_find_delayed_ref_head(trans, bytenr);
+	if (head) {
+		if (!mutex_trylock(&head->mutex)) {
+			atomic_inc(&head->node.refs);
+			spin_unlock(&delayed_refs->lock);
+
+			btrfs_release_path(path);
+
+			/*
+			 * Mutex was contended, block until it's
+			 * released and try again
+			 */
+			mutex_lock(&head->mutex);
+			mutex_unlock(&head->mutex);
+			btrfs_put_delayed_ref(&head->node);
+			goto again;
+		}
+		ret = __add_delayed_refs(head, seq, &info_key, &prefs_delayed);
+		if (ret)
+			goto out;
+	}
+	spin_unlock(&delayed_refs->lock);
+
+	if (path->slots[0]) {
+		struct extent_buffer *leaf;
+		int slot;
+
+		leaf = path->nodes[0];
+		slot = path->slots[0] - 1;
+		btrfs_item_key_to_cpu(leaf, &key, slot);
+		if (key.objectid == bytenr &&
+		    key.type == BTRFS_EXTENT_ITEM_KEY) {
+			ret = __add_inline_refs(fs_info, path, bytenr,
+						&info_key, &info_level, &prefs);
+			if (ret)
+				goto out;
+			ret = __add_keyed_refs(fs_info, path, bytenr, &info_key,
+					       info_level, &prefs);
+			if (ret)
+				goto out;
+		}
+	}
+	btrfs_release_path(path);
+
+	/*
+	 * when adding the delayed refs above, the info_key might not have
+	 * been known yet. Go over the list and replace the missing keys
+	 */
+	list_for_each_entry(ref, &prefs_delayed, list) {
+		if ((ref->key.offset | ref->key.type | ref->key.objectid) == 0)
+			memcpy(&ref->key, &info_key, sizeof(ref->key));
+	}
+	list_splice_init(&prefs_delayed, &prefs);
+
+	ret = __merge_refs(&prefs, 1);
+	if (ret)
+		goto out;
+
+	ret = __resolve_indirect_refs(fs_info, &prefs);
+	if (ret)
+		goto out;
+
+	ret = __merge_refs(&prefs, 2);
+	if (ret)
+		goto out;
+
+	while (!list_empty(&prefs)) {
+		ref = list_first_entry(&prefs, struct __prelim_ref, list);
+		list_del(&ref->list);
+		if (ref->count < 0)
+			WARN_ON(1);
+		if (ref->count && ref->root_id && ref->parent == 0) {
+			/* no parent == root of tree */
+			ret = ulist_add(roots, ref->root_id, 0, GFP_NOFS);
+			BUG_ON(ret < 0);
+		}
+		if (ref->count && ref->parent) {
+			ret = ulist_add(refs, ref->parent, 0, GFP_NOFS);
+			BUG_ON(ret < 0);
+		}
+		kfree(ref);
+	}
+
+out:
+	if (head)
+		mutex_unlock(&head->mutex);
+	btrfs_free_path(path);
+	while (!list_empty(&prefs)) {
+		ref = list_first_entry(&prefs, struct __prelim_ref, list);
+		list_del(&ref->list);
+		kfree(ref);
+	}
+	while (!list_empty(&prefs_delayed)) {
+		ref = list_first_entry(&prefs_delayed, struct __prelim_ref,
+				       list);
+		list_del(&ref->list);
+		kfree(ref);
+	}
+
+	return ret;
+}
+
+/*
+ * Finds all leafs with a reference to the specified combination of bytenr and
+ * offset. key_list_head will point to a list of corresponding keys (caller must
+ * free each list element). The leafs will be stored in the leafs ulist, which
+ * must be freed with ulist_free.
+ *
+ * returns 0 on success, <0 on error
+ */
+static int btrfs_find_all_leafs(struct btrfs_trans_handle *trans,
+				struct btrfs_fs_info *fs_info, u64 bytenr,
+				u64 num_bytes, u64 seq, struct ulist **leafs)
+{
+	struct ulist *tmp;
+	int ret;
+
+	tmp = ulist_alloc(GFP_NOFS);
+	if (!tmp)
+		return -ENOMEM;
+	*leafs = ulist_alloc(GFP_NOFS);
+	if (!*leafs) {
+		ulist_free(tmp);
+		return -ENOMEM;
+	}
+
+	ret = find_parent_nodes(trans, fs_info, bytenr, seq, *leafs, tmp);
+	ulist_free(tmp);
+
+	if (ret < 0 && ret != -ENOENT) {
+		ulist_free(*leafs);
+		return ret;
+	}
+
+	return 0;
+}
+
+/*
+ * walk all backrefs for a given extent to find all roots that reference this
+ * extent. Walking a backref means finding all extents that reference this
+ * extent and in turn walk the backrefs of those, too. Naturally this is a
+ * recursive process, but here it is implemented in an iterative fashion: We
+ * find all referencing extents for the extent in question and put them on a
+ * list. In turn, we find all referencing extents for those, further appending
+ * to the list. The way we iterate the list allows adding more elements after
+ * the current while iterating. The process stops when we reach the end of the
+ * list. Found roots are added to the roots list.
+ *
+ * returns 0 on success, < 0 on error.
+ */
+int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
+				struct btrfs_fs_info *fs_info, u64 bytenr,
+				u64 num_bytes, u64 seq, struct ulist **roots)
+{
+	struct ulist *tmp;
+	struct ulist_node *node = NULL;
+	int ret;
+
+	tmp = ulist_alloc(GFP_NOFS);
+	if (!tmp)
+		return -ENOMEM;
+	*roots = ulist_alloc(GFP_NOFS);
+	if (!*roots) {
+		ulist_free(tmp);
+		return -ENOMEM;
+	}
+
+	while (1) {
+		ret = find_parent_nodes(trans, fs_info, bytenr, seq,
+					tmp, *roots);
+		if (ret < 0 && ret != -ENOENT) {
+			ulist_free(tmp);
+			ulist_free(*roots);
+			return ret;
+		}
+		node = ulist_next(tmp, node);
+		if (!node)
+			break;
+		bytenr = node->val;
+	}
+
+	ulist_free(tmp);
+	return 0;
+}
+
+
 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)
diff --git a/fs/btrfs/backref.h b/fs/btrfs/backref.h
index 9261883..d00dfa9 100644
--- a/fs/btrfs/backref.h
+++ b/fs/btrfs/backref.h
@@ -20,6 +20,7 @@
 #define __BTRFS_BACKREF__
 
 #include "ioctl.h"
+#include "ulist.h"
 
 struct inode_fs_paths {
 	struct btrfs_path		*btrfs_path;
@@ -54,6 +55,10 @@ int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
 
 int paths_from_inode(u64 inum, struct inode_fs_paths *ipath);
 
+int btrfs_find_all_roots(struct btrfs_trans_handle *trans,
+				struct btrfs_fs_info *fs_info, u64 bytenr,
+				u64 num_bytes, u64 seq, struct ulist **roots);
+
 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);
-- 
1.7.3.4

[PATCH v1 10/10] Btrfs: new backref walking code

[Jan Schmidt]

The old backref iteration code could only safely be used on commit roots.
Besides this limitation, it had bugs in finding the roots for these
references. This commit replaces large parts of it by btrfs_find_all_roots()
which a) really finds all roots and the correct roots, b) works correctly
under heavy file system load, c) considers delayed refs.

Signed-off-by: Jan Schmidt <list.btrfs@jan-o-sch.net>
---
 fs/btrfs/backref.c |  354 +++++++++++++++-------------------------------------
 fs/btrfs/ioctl.c   |    8 +-
 fs/btrfs/scrub.c   |    7 +-
 3 files changed, 107 insertions(+), 262 deletions(-)

diff --git a/fs/btrfs/backref.c b/fs/btrfs/backref.c
index e01790e..2fdb4b1 100644
--- a/fs/btrfs/backref.c
+++ b/fs/btrfs/backref.c
@@ -23,18 +23,6 @@
 #include "transaction.h"
 #include "delayed-ref.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;
-};
-
 /*
  * this structure records all encountered refs on the way up to the root
  */
@@ -965,8 +953,11 @@ int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
 	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)
+	    found_key->objectid + found_key->offset <= logical) {
+		pr_debug("logical %llu is not within any extent\n",
+			 (unsigned long long)logical);
 		return -ENOENT;
+	}
 
 	eb = path->nodes[0];
 	item_size = btrfs_item_size_nr(eb, path->slots[0]);
@@ -975,6 +966,13 @@ int extent_from_logical(struct btrfs_fs_info *fs_info, u64 logical,
 	ei = btrfs_item_ptr(eb, path->slots[0], struct btrfs_extent_item);
 	flags = btrfs_extent_flags(eb, ei);
 
+	pr_debug("logical %llu is at position %llu within the extent (%llu "
+		 "EXTENT_ITEM %llu) flags %#llx size %u\n",
+		 (unsigned long long)logical,
+		 (unsigned long long)(logical - found_key->objectid),
+		 (unsigned long long)found_key->objectid,
+		 (unsigned long long)found_key->offset,
+		 (unsigned long long)flags, item_size);
 	if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)
 		return BTRFS_EXTENT_FLAG_TREE_BLOCK;
 	if (flags & BTRFS_EXTENT_FLAG_DATA)
@@ -1071,128 +1069,11 @@ int tree_backref_for_extent(unsigned long *ptr, struct extent_buffer *eb,
 	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)
+static int iterate_leaf_refs(struct btrfs_fs_info *fs_info,
+				struct btrfs_path *path, u64 logical,
+				u64 orig_extent_item_objectid,
+				u64 extent_item_pos, u64 root,
+				iterate_extent_inodes_t *iterate, void *ctx)
 {
 	u64 disk_byte;
 	struct btrfs_key key;
@@ -1200,8 +1081,10 @@ static int __iter_shared_inline_ref(struct btrfs_fs_info *fs_info,
 	struct extent_buffer *eb;
 	int slot;
 	int nritems;
-	int ret;
-	int found = 0;
+	int ret = 0;
+	int extent_type;
+	u64 data_offset;
+	u64 data_len;
 
 	eb = read_tree_block(fs_info->tree_root, logical,
 				fs_info->tree_root->leafsize, 0);
@@ -1219,149 +1102,99 @@ static int __iter_shared_inline_ref(struct btrfs_fs_info *fs_info,
 		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;
-		}
+		extent_type = btrfs_file_extent_type(eb, fi);
+		if (extent_type == BTRFS_FILE_EXTENT_INLINE)
+			continue;
+		/* don't skip BTRFS_FILE_EXTENT_PREALLOC, we can handle that */
 		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 (disk_byte != orig_extent_item_objectid)
+			continue;
 
-	if (!found) {
-		printk(KERN_ERR "btrfs: failed to follow shared data backref "
-			"to parent %llu\n", logical);
-		WARN_ON(1);
-		ret = -EIO;
+		data_offset = btrfs_file_extent_offset(eb, fi);
+		data_len = btrfs_file_extent_num_bytes(eb, fi);
+
+		if (extent_item_pos < data_offset ||
+		    extent_item_pos >= data_offset + data_len)
+			continue;
+
+		pr_debug("ref for %llu resolved, key (%llu EXTEND_DATA %llu), "
+				"root %llu\n", orig_extent_item_objectid,
+				key.objectid, key.offset, root);
+		ret = iterate(key.objectid,
+				key.offset + (extent_item_pos - data_offset),
+				root, ctx);
+		if (ret) {
+			pr_debug("stopping iteration because ret=%d\n", ret);
+			break;
+		}
 	}
 
 	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.
+ * the given parameters.
  * when the iterator function returns a non-zero value, iteration stops.
+ * path is guaranteed to be in released state when iterate() is called.
  */
 int iterate_extent_inodes(struct btrfs_fs_info *fs_info,
 				struct btrfs_path *path,
-				u64 extent_item_objectid,
-				u64 extent_offset,
+				u64 extent_item_objectid, u64 extent_item_pos,
 				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;
+	struct btrfs_trans_handle *trans;
+	struct ulist *refs;
+	struct ulist *roots;
+	struct ulist_node *ref_node = NULL;
+	struct ulist_node *root_node = NULL;
+	struct seq_list seq_elem;
+	struct btrfs_delayed_ref_root *delayed_refs;
 
-	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;
-		}
+	trans = btrfs_join_transaction(fs_info->extent_root);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
 
-		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);
+	pr_debug("resolving all inodes for extent %llu\n",
+			extent_item_objectid);
 
-	/* ... 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);
-		}
-	}
+	delayed_refs = &trans->transaction->delayed_refs;
+	spin_lock(&delayed_refs->lock);
+	btrfs_get_delayed_seq(delayed_refs, &seq_elem);
+	spin_unlock(&delayed_refs->lock);
 
-	btrfs_release_path(path);
+	ret = btrfs_find_all_leafs(trans, fs_info, extent_item_objectid,
+				   extent_item_pos, seq_elem.seq,
+				   &refs);
 
-	/*
-	 * ... 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);
-	}
+	if (ret)
+		goto out;
 
-	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);
+	while (!ret && (ref_node = ulist_next(refs, ref_node))) {
+		ret = btrfs_find_all_roots(trans, fs_info, ref_node->val, -1,
+						seq_elem.seq, &roots);
+		if (ret)
+			break;
+		while (!ret && (root_node = ulist_next(roots, root_node))) {
+			pr_debug("root %llu references leaf %llu\n",
+					root_node->val, ref_node->val);
+			ret = iterate_leaf_refs(fs_info, path, ref_node->val,
+						extent_item_objectid,
+						extent_item_pos, root_node->val,
+						iterate, ctx);
+		}
 	}
 
+	ulist_free(refs);
+	ulist_free(roots);
+out:
+	btrfs_put_delayed_seq(delayed_refs, &seq_elem);
+	btrfs_end_transaction(trans, fs_info->extent_root);
 	return ret;
 }
 
@@ -1370,19 +1203,20 @@ int iterate_inodes_from_logical(u64 logical, struct btrfs_fs_info *fs_info,
 				iterate_extent_inodes_t *iterate, void *ctx)
 {
 	int ret;
-	u64 offset;
+	u64 extent_item_pos;
 	struct btrfs_key found_key;
 
 	ret = extent_from_logical(fs_info, logical, path,
 					&found_key);
+	btrfs_release_path(path);
 	if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK)
 		ret = -EINVAL;
 	if (ret < 0)
 		return ret;
 
-	offset = logical - found_key.objectid;
+	extent_item_pos = logical - found_key.objectid;
 	ret = iterate_extent_inodes(fs_info, path, found_key.objectid,
-					offset, iterate, ctx);
+					extent_item_pos, iterate, ctx);
 
 	return ret;
 }
@@ -1427,6 +1261,10 @@ static int iterate_irefs(u64 inum, struct btrfs_root *fs_root,
 		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()! */
+			pr_debug("following ref at offset %u for inode %llu in "
+				 "tree %llu\n", cur,
+				 (unsigned long long)found_key.objectid,
+				 (unsigned long long)fs_root->objectid);
 			ret = iterate(parent, iref, eb, ctx);
 			if (ret) {
 				free_extent_buffer(eb);
@@ -1467,10 +1305,14 @@ static int inode_to_path(u64 inum, struct btrfs_inode_ref *iref,
 		return PTR_ERR(fspath);
 
 	if (fspath > fspath_min) {
+		pr_debug("path resolved: %s\n", fspath);
 		ipath->fspath->val[i] = (u64)(unsigned long)fspath;
 		++ipath->fspath->elem_cnt;
 		ipath->fspath->bytes_left = fspath - fspath_min;
 	} else {
+		pr_debug("missed path, not enough space. missing bytes: %ld, "
+			 "constructed so far: %s\n", fspath_min - fspath,
+			 fspath_min);
 		++ipath->fspath->elem_missed;
 		ipath->fspath->bytes_missing += fspath_min - fspath;
 		ipath->fspath->bytes_left = 0;
diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index c48f2e9..9b05268 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -2976,7 +2976,7 @@ static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
 {
 	int ret = 0;
 	int size;
-	u64 extent_offset;
+	u64 extent_item_pos;
 	struct btrfs_ioctl_logical_ino_args *loi;
 	struct btrfs_data_container *inodes = NULL;
 	struct btrfs_path *path = NULL;
@@ -3007,15 +3007,17 @@ static long btrfs_ioctl_logical_to_ino(struct btrfs_root *root,
 	}
 
 	ret = extent_from_logical(root->fs_info, loi->logical, path, &key);
+	btrfs_release_path(path);
 
 	if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK)
 		ret = -ENOENT;
 	if (ret < 0)
 		goto out;
 
-	extent_offset = loi->logical - key.objectid;
+	extent_item_pos = loi->logical - key.objectid;
 	ret = iterate_extent_inodes(root->fs_info, path, key.objectid,
-					extent_offset, build_ino_list, inodes);
+					extent_item_pos, build_ino_list,
+					inodes);
 
 	if (ret < 0)
 		goto out;
diff --git a/fs/btrfs/scrub.c b/fs/btrfs/scrub.c
index c27bcb6..b5edff2 100644
--- a/fs/btrfs/scrub.c
+++ b/fs/btrfs/scrub.c
@@ -309,7 +309,7 @@ static void scrub_print_warning(const char *errstr, struct scrub_bio *sbio,
 	u8 ref_level;
 	unsigned long ptr = 0;
 	const int bufsize = 4096;
-	u64 extent_offset;
+	u64 extent_item_pos;
 
 	path = btrfs_alloc_path();
 
@@ -329,12 +329,13 @@ static void scrub_print_warning(const char *errstr, struct scrub_bio *sbio,
 	if (ret < 0)
 		goto out;
 
-	extent_offset = swarn.logical - found_key.objectid;
+	extent_item_pos = swarn.logical - found_key.objectid;
 	swarn.extent_item_size = found_key.offset;
 
 	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]);
+	btrfs_release_path(path);
 
 	if (ret & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 		do {
@@ -351,7 +352,7 @@ static void scrub_print_warning(const char *errstr, struct scrub_bio *sbio,
 	} else {
 		swarn.path = path;
 		iterate_extent_inodes(fs_info, path, found_key.objectid,
-					extent_offset,
+					extent_item_pos,
 					scrub_print_warning_inode, &swarn);
 	}
 
-- 
1.7.3.4