[PATCH 0/6] Freespace tree repair support

[PATCH 0/6] Freespace tree repair support

[Nikolay Borisov]

Hello, 

This patchset adds support to btrfs check to repair corrupted freespace tree. 
Once this has been merged then we can start thinking of making the freespace
tree the default freespace cache and deprecate the existing v1 cache in new 
filesystems. 

This series depend on the already sent delayed refs. In order to facilitate 
testing the 2 series have been integrated and posted at :

   https://github.com/lorddoskias/btrfs-progs.git dev/delayed-refs-fst

The main thing with this code is that the repair will always regenerate the 
FST with extents since it I haven't ported the code which deals with 
detecting that bitmaps are more space efficient. This is not a big deal since 
the kernel will do it.

The first patch is here for completeness' sake and was only used in a previous 
incarnation of this code but since I've re-worked the way the code is supposed 
to work it can be dropped. I don't have strong opinions either way. It just 
teaches btrfs_read_fs_root to return the freespace root

Patch 2 adds some low-level bit manipulation primitives which are necessary to 
support bitmap based FST. My initial idea was to completely ommit the bitmap 
support but in testing it turned out to be less pain to just include it from 
the get-go. Otherwise the filesystem created in the test case in patch 6 
couldn't be worked on. 

Patch 3 is mostly copy/paste from the kernel code bringin in necessary 
functions to repair the freespace tree. I've only omitted locking code since 
in userspace we are always single-threaded, also there is no support currently 
when regenerating FST to convert it to bitmap.

Patch 4 updates the compat_ro bitmask to allow reading an FST filesystem in 
repair mode.

Patch 5 adds the core code which utilizes everything added up until this point. 
It's not that big and should be fairly easy to review. 

Patch 6 is a test-case which creates a filesystem with FST enabled, populates it 
with some files and then corrupts a bitmap and an extent records and tries to 
repair them. It's passing for me. 


Nikolay Borisov (6):
  btrfs-progs: Add support for freespace tree in btrfs_read_fs_root
  btrfs-progs: Add extent buffer bitmap manipulation infrastructure
  btrfs-progs: Pull free space tree related code from kernel
  btrfs-progs: Add freespace tree as compat_ro supported feature
  btrfs-progs: check: Add support for freespace tree fixing
  btrfs-progs: tests: Test for FST corruption detection/repair

 check/main.c                                      |  61 +-
 ctree.c                                           |  77 ++
 ctree.h                                           |   8 +-
 disk-io.c                                         |   3 +
 extent-tree.c                                     |   9 +
 extent_io.c                                       |  39 +
 extent_io.h                                       |  15 +
 free-space-tree.c                                 | 893 +++++++++++++++++++++-
 free-space-tree.h                                 |  10 +-
 tests/fsck-tests/035-freespacetree-repair/test.sh |  79 ++
 10 files changed, 1163 insertions(+), 31 deletions(-)
 create mode 100755 tests/fsck-tests/035-freespacetree-repair/test.sh

-- 
2.7.4

[PATCH 1/6] btrfs-progs: Add support for freespace tree in btrfs_read_fs_root

[Nikolay Borisov]

For completeness sake add code to btrfs_read_fs_root so that it can
handle the freespace tree.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
---
 disk-io.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/disk-io.c b/disk-io.c
index 8da6e3ce5fc8..9ad826b83b3e 100644
--- a/disk-io.c
+++ b/disk-io.c
@@ -664,6 +664,9 @@ struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
 	if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
 		return fs_info->quota_enabled ? fs_info->quota_root :
 				ERR_PTR(-ENOENT);
+	if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
+        return fs_info->free_space_root ? fs_info->free_space_root :
+                                                  ERR_PTR(-ENOENT);
 
 	BUG_ON(location->objectid == BTRFS_TREE_RELOC_OBJECTID ||
 	       location->offset != (u64)-1);
-- 
2.7.4

[PATCH 2/6] btrfs-progs: Add extent buffer bitmap manipulation infrastructure

[Nikolay Borisov]

Those functions are in preparation for adding the freespace tree
repair code since it needs to be able to deal with bitmap based fsts.
This patch adds extent_buffer_bitmap_set and extent_buffer_bitmap_clear
functions. Since in userspace we don't have to deal with page mappings
their implementation is vastly simplified by simply setting each bit in
the passed range.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
---
 extent_io.c | 39 +++++++++++++++++++++++++++++++++++++++
 extent_io.h | 15 +++++++++++++++
 2 files changed, 54 insertions(+)

diff --git a/extent_io.c b/extent_io.c
index 198492699438..568a12f7084b 100644
--- a/extent_io.c
+++ b/extent_io.c
@@ -204,6 +204,45 @@ static int clear_state_bit(struct extent_io_tree *tree,
 	return ret;
 }
 
+/**
+ * extent_buffer_bitmap_set - set an area of a bitmap
+ * @eb: the extent buffer
+ * @start: offset of the bitmap item in the extent buffer
+ * @pos: bit number of the first bit
+ * @len: number of bits to set
+ */
+void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
+                              unsigned long pos, unsigned long len)
+{
+        u8 *kaddr = (u8 *)eb->data + start;
+
+        while (len) {
+		le_set_bit(pos, kaddr);
+		pos++;
+		len--;
+        }
+}
+
+
+/**
+ * extent_buffer_bitmap_clear - clear an area of a bitmap
+ * @eb: the extent buffer
+ * @start: offset of the bitmap item in the extent buffer
+ * @pos: bit number of the first bit
+ * @len: number of bits to clear
+ */
+void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
+                                unsigned long pos, unsigned long len)
+{
+        u8 *kaddr = (u8 *)eb->data + start;
+
+        while (len) {
+		le_clear_bit(pos, kaddr);
+		pos++;
+		len--;
+        }
+}
+
 /*
  * clear some bits on a range in the tree.
  */
diff --git a/extent_io.h b/extent_io.h
index d407d93d617e..f9097911f5ef 100644
--- a/extent_io.h
+++ b/extent_io.h
@@ -68,6 +68,17 @@ static inline int le_test_bit(int nr, const u8 *addr)
 	return 1U & (addr[BIT_BYTE(nr)] >> (nr & (BITS_PER_BYTE-1)));
 }
 
+
+static inline void le_set_bit(int nr, u8 *addr)
+{
+	addr[BIT_BYTE(nr)] |= (1U << (nr & (BITS_PER_BYTE-1)));
+}
+
+static inline void le_clear_bit(int nr, u8 *addr)
+{
+	addr[BIT_BYTE(nr)] &= ~(1U << (nr & (BITS_PER_BYTE-1)));
+}
+
 struct btrfs_fs_info;
 
 struct extent_io_tree {
@@ -175,4 +186,8 @@ int read_data_from_disk(struct btrfs_fs_info *info, void *buf, u64 offset,
 			u64 bytes, int mirror);
 int write_data_to_disk(struct btrfs_fs_info *info, void *buf, u64 offset,
 		       u64 bytes, int mirror);
+void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
+                                unsigned long pos, unsigned long len);
+void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
+                              unsigned long pos, unsigned long len);
 #endif
-- 
2.7.4

[PATCH 3/6] btrfs-progs: Pull free space tree related code from kernel

[Nikolay Borisov]

To help implement free space tree checker in user space some kernel
function are necessary, namely iterating/deleting/adding freespace
items, some internal search functions. Functions to populate a block
group based on the extent tree. The code is largely copy/paste from
the kernel with locking eliminated (i.e free_space_lock). It supports
reading/writing of both bitmap and extent based FST trees.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
---
 ctree.c           |  77 +++++
 ctree.h           |   4 +
 free-space-tree.c | 893 +++++++++++++++++++++++++++++++++++++++++++++++++++++-
 free-space-tree.h |  10 +-
 4 files changed, 977 insertions(+), 7 deletions(-)

diff --git a/ctree.c b/ctree.c
index d8a6883aa85f..aa1568620205 100644
--- a/ctree.c
+++ b/ctree.c
@@ -1226,6 +1226,83 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
 }
 
 /*
+ * helper to use instead of search slot if no exact match is needed but
+ * instead the next or previous item should be returned.
+ * When find_higher is true, the next higher item is returned, the next lower
+ * otherwise.
+ * When return_any and find_higher are both true, and no higher item is found,
+ * return the next lower instead.
+ * When return_any is true and find_higher is false, and no lower item is found,
+ * return the next higher instead.
+ * It returns 0 if any item is found, 1 if none is found (tree empty), and
+ * < 0 on error
+ */
+int btrfs_search_slot_for_read(struct btrfs_root *root,
+                               const struct btrfs_key *key,
+                               struct btrfs_path *p, int find_higher,
+                               int return_any)
+{
+        int ret;
+        struct extent_buffer *leaf;
+
+again:
+        ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
+        if (ret <= 0)
+                return ret;
+        /*
+         * a return value of 1 means the path is at the position where the
+         * item should be inserted. Normally this is the next bigger item,
+         * but in case the previous item is the last in a leaf, path points
+         * to the first free slot in the previous leaf, i.e. at an invalid
+         * item.
+         */
+        leaf = p->nodes[0];
+
+        if (find_higher) {
+                if (p->slots[0] >= btrfs_header_nritems(leaf)) {
+                        ret = btrfs_next_leaf(root, p);
+                        if (ret <= 0)
+                                return ret;
+                        if (!return_any)
+                                return 1;
+                        /*
+                         * no higher item found, return the next
+                         * lower instead
+                         */
+                        return_any = 0;
+                        find_higher = 0;
+                        btrfs_release_path(p);
+                        goto again;
+                }
+        } else {
+                if (p->slots[0] == 0) {
+                        ret = btrfs_prev_leaf(root, p);
+                        if (ret < 0)
+                                return ret;
+                        if (!ret) {
+                                leaf = p->nodes[0];
+                                if (p->slots[0] == btrfs_header_nritems(leaf))
+                                        p->slots[0]--;
+                                return 0;
+                        }
+                        if (!return_any)
+                                return 1;
+                        /*
+                         * no lower item found, return the next
+                         * higher instead
+                         */
+                        return_any = 0;
+                        find_higher = 1;
+                        btrfs_release_path(p);
+                        goto again;
+                } else {
+                        --p->slots[0];
+                }
+        }
+        return 0;
+}
+
+/*
  * adjust the pointers going up the tree, starting at level
  * making sure the right key of each node is points to 'key'.
  * This is used after shifting pointers to the left, so it stops
diff --git a/ctree.h b/ctree.h
index 3e9ca2ca8432..ade883fecbd6 100644
--- a/ctree.h
+++ b/ctree.h
@@ -2619,6 +2619,10 @@ int btrfs_split_item(struct btrfs_trans_handle *trans,
 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
 		      *root, struct btrfs_key *key, struct btrfs_path *p, int
 		      ins_len, int cow);
+int btrfs_search_slot_for_read(struct btrfs_root *root,
+                               const struct btrfs_key *key,
+                               struct btrfs_path *p, int find_higher,
+                               int return_any);
 int btrfs_find_item(struct btrfs_root *fs_root, struct btrfs_path *found_path,
 		u64 iobjectid, u64 ioff, u8 key_type,
 		struct btrfs_key *found_key);
diff --git a/free-space-tree.c b/free-space-tree.c
index 139a031e8483..6acb71378110 100644
--- a/free-space-tree.c
+++ b/free-space-tree.c
@@ -47,8 +47,7 @@ search_free_space_info(struct btrfs_trans_handle *trans,
 }
 
 static int free_space_test_bit(struct btrfs_block_group_cache *block_group,
-			       struct btrfs_path *path, u64 offset,
-			       u64 sectorsize)
+			       struct btrfs_path *path, u64 offset)
 {
 	struct extent_buffer *leaf;
 	struct btrfs_key key;
@@ -64,10 +63,756 @@ static int free_space_test_bit(struct btrfs_block_group_cache *block_group,
 	ASSERT(offset >= found_start && offset < found_end);
 
 	ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
-	i = (offset - found_start) / sectorsize;
+	i = (offset - found_start) / leaf->fs_info->sectorsize;
 	return !!extent_buffer_test_bit(leaf, ptr, i);
 }
 
+/*
+ * btrfs_search_slot() but we're looking for the greatest key less than the
+ * passed key.
+ */
+static int btrfs_search_prev_slot(struct btrfs_trans_handle *trans,
+                                  struct btrfs_root *root,
+                                  struct btrfs_key *key, struct btrfs_path *p,
+                                  int ins_len, int cow)
+{
+	int ret;
+
+	ret = btrfs_search_slot(trans, root, key, p, ins_len, cow);
+	if (ret < 0)
+		return ret;
+
+	if (ret == 0) {
+		ASSERT(0);
+		return -EIO;
+	}
+
+	if (p->slots[0] == 0) {
+		ASSERT(0);
+		return -EIO;
+	}
+	p->slots[0]--;
+
+	return 0;
+}
+
+static int add_new_free_space_info(struct btrfs_trans_handle *trans,
+                                   struct btrfs_block_group_cache *block_group,
+                                   struct btrfs_path *path)
+{
+	struct btrfs_root *root = trans->fs_info->free_space_root;
+	struct btrfs_free_space_info *info;
+	struct btrfs_key key;
+	struct extent_buffer *leaf;
+	int ret;
+
+	key.objectid = block_group->key.objectid;
+	key.type = BTRFS_FREE_SPACE_INFO_KEY;
+	key.offset = block_group->key.offset;
+
+	ret = btrfs_insert_empty_item(trans, root, path, &key, sizeof(*info));
+	if (ret)
+		goto out;
+
+	leaf = path->nodes[0];
+	info = btrfs_item_ptr(leaf, path->slots[0],
+	                      struct btrfs_free_space_info);
+	btrfs_set_free_space_extent_count(leaf, info, 0);
+	btrfs_set_free_space_flags(leaf, info, 0);
+	btrfs_mark_buffer_dirty(leaf);
+
+	ret = 0;
+out:
+	btrfs_release_path(path);
+	return ret;
+}
+
+static int update_free_space_extent_count(struct btrfs_trans_handle *trans,
+                                          struct btrfs_block_group_cache *block_group,
+                                          struct btrfs_path *path,
+                                          int new_extents)
+{
+	struct btrfs_free_space_info *info;
+	u32 extent_count;
+	int ret = 0;
+
+	if (new_extents == 0)
+		return 0;
+
+	info = search_free_space_info(trans, trans->fs_info, block_group, path,
+				1);
+	if (IS_ERR(info)) {
+		ret = PTR_ERR(info);
+		goto out;
+	}
+	extent_count = btrfs_free_space_extent_count(path->nodes[0], info);
+
+	extent_count += new_extents;
+	btrfs_set_free_space_extent_count(path->nodes[0], info, extent_count);
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+	btrfs_release_path(path);
+out:
+	return ret;
+}
+
+
+static void free_space_set_bits(struct btrfs_block_group_cache *block_group,
+                                struct btrfs_path *path, u64 *start, u64 *size,
+                                int bit)
+{
+        struct extent_buffer *leaf = path->nodes[0];
+        struct btrfs_fs_info *fs_info = leaf->fs_info;
+        struct btrfs_key key;
+        u64 end = *start + *size;
+        u64 found_start, found_end;
+        unsigned long ptr, first, last;
+
+        btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
+        ASSERT(key.type == BTRFS_FREE_SPACE_BITMAP_KEY);
+
+        found_start = key.objectid;
+        found_end = key.objectid + key.offset;
+        ASSERT(*start >= found_start && *start < found_end);
+        ASSERT(end > found_start);
+
+        if (end > found_end)
+                end = found_end;
+
+        ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
+        first = (*start - found_start) / fs_info->sectorsize;
+        last = (end - found_start) / fs_info->sectorsize;
+        if (bit)
+                extent_buffer_bitmap_set(leaf, ptr, first, last - first);
+        else
+                extent_buffer_bitmap_clear(leaf, ptr, first, last - first);
+        btrfs_mark_buffer_dirty(leaf);
+
+        *size -= end - *start;
+        *start = end;
+}
+
+/*
+ * We can't use btrfs_next_item() in modify_free_space_bitmap() because
+ * btrfs_next_leaf() doesn't get the path for writing. We can forgo the fancy
+ * tree walking in btrfs_next_leaf() anyways because we know exactly what we're
+ * looking for.
+ */
+static int free_space_next_bitmap(struct btrfs_trans_handle *trans,
+                                  struct btrfs_root *root, struct btrfs_path *p)
+{
+	struct btrfs_key key;
+
+	if (p->slots[0] + 1 < btrfs_header_nritems(p->nodes[0])) {
+		p->slots[0]++;
+		return 0;
+	}
+
+	btrfs_item_key_to_cpu(p->nodes[0], &key, p->slots[0]);
+	btrfs_release_path(p);
+
+	key.objectid += key.offset;
+	key.type = (u8)-1;
+	key.offset = (u64)-1;
+
+	return btrfs_search_prev_slot(trans, root, &key, p, 0, 1);
+}
+
+/*
+ * If remove is 1, then we are removing free space, thus clearing bits in the
+ * bitmap. If remove is 0, then we are adding free space, thus setting bits in
+ * the bitmap.
+ */
+static int modify_free_space_bitmap(struct btrfs_trans_handle *trans,
+                                    struct btrfs_block_group_cache *block_group,
+                                    struct btrfs_path *path,
+                                    u64 start, u64 size, int remove)
+{
+        struct btrfs_root *root = trans->fs_info->free_space_root;
+        struct btrfs_key key;
+        u64 end = start + size;
+        u64 cur_start, cur_size;
+        int prev_bit, next_bit;
+        int new_extents;
+        int ret;
+
+        /*
+         * Read the bit for the block immediately before the extent of space if
+         * that block is within the block group.
+         */
+        if (start > block_group->key.objectid) {
+                u64 prev_block = start - trans->fs_info->sectorsize;
+
+                key.objectid = prev_block;
+                key.type = (u8)-1;
+                key.offset = (u64)-1;
+
+                ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
+                if (ret)
+                        goto out;
+
+                prev_bit = free_space_test_bit(block_group, path, prev_block);
+
+                /* The previous block may have been in the previous bitmap. */
+                btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+                if (start >= key.objectid + key.offset) {
+                        ret = free_space_next_bitmap(trans, root, path);
+                        if (ret)
+                                goto out;
+                }
+        } else {
+                key.objectid = start;
+                key.type = (u8)-1;
+                key.offset = (u64)-1;
+
+                ret = btrfs_search_prev_slot(trans, root, &key, path, 0, 1);
+                if (ret)
+                        goto out;
+
+                prev_bit = -1;
+        }
+
+        /*
+         * Iterate over all of the bitmaps overlapped by the extent of space,
+         * clearing/setting bits as required.
+         */
+        cur_start = start;
+        cur_size = size;
+        while (1) {
+                free_space_set_bits(block_group, path, &cur_start, &cur_size,
+                                    !remove);
+                if (cur_size == 0)
+                        break;
+                ret = free_space_next_bitmap(trans, root, path);
+                if (ret)
+                        goto out;
+        }
+
+	/*
+         * Read the bit for the block immediately after the extent of space if
+         * that block is within the block group.
+         */
+        if (end < block_group->key.objectid + block_group->key.offset) {
+                /* The next block may be in the next bitmap. */
+                btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+                if (end >= key.objectid + key.offset) {
+                        ret = free_space_next_bitmap(trans, root, path);
+                        if (ret)
+                                goto out;
+                }
+
+                next_bit = free_space_test_bit(block_group, path, end);
+        } else {
+                next_bit = -1;
+        }
+
+        if (remove) {
+                new_extents = -1;
+                if (prev_bit == 1) {
+                        /* Leftover on the left. */
+                        new_extents++;
+                }
+                if (next_bit == 1) {
+                        /* Leftover on the right. */
+                        new_extents++;
+                }
+        } else {
+                new_extents = 1;
+                if (prev_bit == 1) {
+                        /* Merging with neighbor on the left. */
+                        new_extents--;
+                }
+                if (next_bit == 1) {
+                        /* Merging with neighbor on the right. */
+                        new_extents--;
+                }
+        }
+
+        btrfs_release_path(path);
+        ret = update_free_space_extent_count(trans, block_group, path,
+                                             new_extents);
+
+out:
+        return ret;
+}
+
+static int remove_free_space_extent(struct btrfs_trans_handle *trans,
+				    struct btrfs_block_group_cache *block_group,
+				    struct btrfs_path *path,
+				    u64 start, u64 size)
+{
+	struct btrfs_root *root = trans->fs_info->free_space_root;
+	struct btrfs_key key;
+	u64 found_start, found_end;
+	u64 end = start + size;
+	int new_extents = -1;
+	int ret;
+
+	key.objectid = start;
+	key.type = (u8)-1;
+	key.offset = (u64)-1;
+
+	ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
+	if (ret)
+		goto out;
+
+	btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+	ASSERT(key.type == BTRFS_FREE_SPACE_EXTENT_KEY);
+
+	found_start = key.objectid;
+	found_end = key.objectid + key.offset;
+	ASSERT(start >= found_start && end <= found_end);
+
+	/*
+	 * Okay, now that we've found the free space extent which contains the
+	 * free space that we are removing, there are four cases:
+	 *
+	 * 1. We're using the whole extent: delete the key we found and
+	 * decrement the free space extent count.
+	 * 2. We are using part of the extent starting at the beginning: delete
+	 * the key we found and insert a new key representing the leftover at
+	 * the end. There is no net change in the number of extents.
+	 * 3. We are using part of the extent ending at the end: delete the key
+	 * we found and insert a new key representing the leftover at the
+	 * beginning. There is no net change in the number of extents.
+	 * 4. We are using part of the extent in the middle: delete the key we
+	 * found and insert two new keys representing the leftovers on each
+	 * side. Where we used to have one extent, we now have two, so increment
+	 * the extent count. We may need to convert the block group to bitmaps
+	 * as a result.
+	 */
+
+	/* Delete the existing key (cases 1-4). */
+	ret = btrfs_del_item(trans, root, path);
+	if (ret)
+		goto out;
+
+	/* Add a key for leftovers at the beginning (cases 3 and 4). */
+	if (start > found_start) {
+		key.objectid = found_start;
+		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
+		key.offset = start - found_start;
+
+		btrfs_release_path(path);
+		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
+		if (ret)
+			goto out;
+		new_extents++;
+	}
+
+	/* Add a key for leftovers at the end (cases 2 and 4). */
+	if (end < found_end) {
+		key.objectid = end;
+		key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
+		key.offset = found_end - end;
+
+		btrfs_release_path(path);
+		ret = btrfs_insert_empty_item(trans, root, path, &key, 0);
+		if (ret)
+			goto out;
+		new_extents++;
+	}
+
+	btrfs_release_path(path);
+	ret = update_free_space_extent_count(trans, block_group, path,
+					     new_extents);
+
+out:
+	return ret;
+}
+
+int __remove_from_free_space_tree(struct btrfs_trans_handle *trans,
+                                  struct btrfs_block_group_cache *block_group,
+                                  struct btrfs_path *path, u64 start, u64 size)
+{
+	struct btrfs_free_space_info *info;
+	u32 flags;
+
+	info = search_free_space_info(NULL, trans->fs_info, block_group, path,
+	                              0);
+	if (IS_ERR(info))
+		return PTR_ERR(info);
+	flags = btrfs_free_space_flags(path->nodes[0], info);
+	btrfs_release_path(path);
+
+	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
+		return modify_free_space_bitmap(trans, block_group, path,
+	                                        start, size, 1);
+	} else {
+		return remove_free_space_extent(trans, block_group, path,
+	                                        start, size);
+	}
+}
+
+int remove_from_free_space_tree(struct btrfs_trans_handle *trans,
+				u64 start, u64 size)
+{
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_path *path;
+	int ret;
+
+	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
+		return 0;
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	block_group = btrfs_lookup_block_group(trans->fs_info, start);
+	if (!block_group) {
+		ASSERT(0);
+		ret = -ENOENT;
+		goto out;
+	}
+
+	ret = __remove_from_free_space_tree(trans, block_group, path, start,
+					    size);
+out:
+	btrfs_free_path(path);
+	if (ret)
+		btrfs_abort_transaction(trans, ret);
+	return ret;
+}
+
+static int add_free_space_extent(struct btrfs_trans_handle *trans,
+                                 struct btrfs_block_group_cache *block_group,
+                                 struct btrfs_path *path,
+                                 u64 start, u64 size)
+{
+        struct btrfs_root *root = trans->fs_info->free_space_root;
+        struct btrfs_key key, new_key;
+        u64 found_start, found_end;
+        u64 end = start + size;
+        int new_extents = 1;
+        int ret;
+
+        /*
+         * We are adding a new extent of free space, but we need to merge
+         * extents. There are four cases here:
+         *
+         * 1. The new extent does not have any immediate neighbors to merge
+         * with: add the new key and increment the free space extent count. We
+         * may need to convert the block group to bitmaps as a result.
+         * 2. The new extent has an immediate neighbor before it: remove the
+         * previous key and insert a new key combining both of them. There is no
+         * net change in the number of extents.
+         * 3. The new extent has an immediate neighbor after it: remove the next
+         * key and insert a new key combining both of them. There is no net
+         * change in the number of extents.
+         * 4. The new extent has immediate neighbors on both sides: remove both
+         * of the keys and insert a new key combining all of them. Where we used
+         * to have two extents, we now have one, so decrement the extent count.
+         */
+
+        new_key.objectid = start;
+        new_key.type = BTRFS_FREE_SPACE_EXTENT_KEY;
+        new_key.offset = size;
+
+        /* Search for a neighbor on the left. */
+        if (start == block_group->key.objectid)
+                goto right;
+        key.objectid = start - 1;
+        key.type = (u8)-1;
+        key.offset = (u64)-1;
+
+        ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
+        if (ret)
+                goto out;
+
+        btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+        if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
+                ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
+                btrfs_release_path(path);
+                goto right;
+        }
+
+        found_start = key.objectid;
+        found_end = key.objectid + key.offset;
+        ASSERT(found_start >= block_group->key.objectid &&
+               found_end > block_group->key.objectid);
+        ASSERT(found_start < start && found_end <= start);
+
+        /*
+         * Delete the neighbor on the left and absorb it into the new key (cases
+         * 2 and 4).
+         */
+        if (found_end == start) {
+                ret = btrfs_del_item(trans, root, path);
+                if (ret)
+                        goto out;
+                new_key.objectid = found_start;
+                new_key.offset += key.offset;
+                new_extents--;
+        }
+        btrfs_release_path(path);
+right:
+        /* Search for a neighbor on the right. */
+        if (end == block_group->key.objectid + block_group->key.offset)
+                goto insert;
+        key.objectid = end;
+        key.type = (u8)-1;
+        key.offset = (u64)-1;
+
+        ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
+        if (ret)
+                goto out;
+
+        btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+        if (key.type != BTRFS_FREE_SPACE_EXTENT_KEY) {
+                ASSERT(key.type == BTRFS_FREE_SPACE_INFO_KEY);
+                btrfs_release_path(path);
+                goto insert;
+        }
+
+        found_start = key.objectid;
+        found_end = key.objectid + key.offset;
+        ASSERT(found_start >= block_group->key.objectid &&
+               found_end > block_group->key.objectid);
+        ASSERT((found_start < start && found_end <= start) ||
+               (found_start >= end && found_end > end));
+
+        /*
+         * Delete the neighbor on the right and absorb it into the new key
+         * (cases 3 and 4).
+         */
+        if (found_start == end) {
+                ret = btrfs_del_item(trans, root, path);
+                if (ret)
+                        goto out;
+                new_key.offset += key.offset;
+                new_extents--;
+        }
+        btrfs_release_path(path);
+
+insert:
+        /* Insert the new key (cases 1-4). */
+        ret = btrfs_insert_empty_item(trans, root, path, &new_key, 0);
+        if (ret)
+                goto out;
+
+        btrfs_release_path(path);
+        ret = update_free_space_extent_count(trans, block_group, path,
+                                             new_extents);
+
+out:
+        return ret;
+}
+
+int __add_to_free_space_tree(struct btrfs_trans_handle *trans,
+                             struct btrfs_block_group_cache *block_group,
+                             struct btrfs_path *path, u64 start, u64 size)
+{
+	struct btrfs_fs_info *fs_info = trans->fs_info;
+	struct btrfs_free_space_info *info;
+	u32 flags;
+
+	info = search_free_space_info(NULL, fs_info, block_group, path, 0);
+	if (IS_ERR(info))
+	        return PTR_ERR(info);
+	flags = btrfs_free_space_flags(path->nodes[0], info);
+	btrfs_release_path(path);
+
+	if (flags & BTRFS_FREE_SPACE_USING_BITMAPS) {
+	        return modify_free_space_bitmap(trans, block_group, path,
+	                                        start, size, 0);
+	} else {
+	        return add_free_space_extent(trans, block_group, path, start,
+	                                     size);
+	}
+}
+
+
+int add_to_free_space_tree(struct btrfs_trans_handle *trans,
+			   u64 start, u64 size)
+{
+	struct btrfs_block_group_cache *block_group;
+	struct btrfs_path *path;
+	int ret;
+
+	if (!btrfs_fs_compat_ro(trans->fs_info, FREE_SPACE_TREE))
+		return 0;
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	block_group = btrfs_lookup_block_group(trans->fs_info, start);
+	if (!block_group) {
+		ASSERT(0);
+		ret = -ENOENT;
+		goto out;
+	}
+
+	ret = __add_to_free_space_tree(trans, block_group, path, start, size);
+out:
+	btrfs_free_path(path);
+	if (ret)
+		btrfs_abort_transaction(trans, ret);
+	return ret;
+}
+
+int populate_free_space_tree(struct btrfs_trans_handle *trans,
+			     struct btrfs_block_group_cache *block_group)
+{
+        struct btrfs_root *extent_root = trans->fs_info->extent_root;
+        struct btrfs_path *path, *path2;
+        struct btrfs_key key;
+        u64 start, end;
+        int ret;
+
+        path = btrfs_alloc_path();
+        if (!path)
+                return -ENOMEM;
+        path->reada = READA_FORWARD;
+
+        path2 = btrfs_alloc_path();
+        if (!path2) {
+                btrfs_free_path(path);
+                return -ENOMEM;
+        }
+
+        ret = add_new_free_space_info(trans, block_group, path2);
+        if (ret)
+                goto out;
+
+        /*
+         * Iterate through all of the extent and metadata items in this block
+         * group, adding the free space between them and the free space at the
+         * end. Note that EXTENT_ITEM and METADATA_ITEM are less than
+         * BLOCK_GROUP_ITEM, so an extent may precede the block group that it's
+         * contained in.
+         */
+        key.objectid = block_group->key.objectid;
+        key.type = BTRFS_EXTENT_ITEM_KEY;
+        key.offset = 0;
+
+        ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
+        if (ret < 0)
+                goto out;
+        ASSERT(ret == 0);
+
+        start = block_group->key.objectid;
+        end = block_group->key.objectid + block_group->key.offset;
+        while (1) {
+                btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
+
+                if (key.type == BTRFS_EXTENT_ITEM_KEY ||
+                    key.type == BTRFS_METADATA_ITEM_KEY) {
+                        if (key.objectid >= end)
+                                break;
+
+                        if (start < key.objectid) {
+                                ret = __add_to_free_space_tree(trans,
+                                                               block_group,
+                                                               path2, start,
+                                                               key.objectid -
+                                                               start);
+                                if (ret)
+                                        goto out;
+                        }
+                        start = key.objectid;
+                        if (key.type == BTRFS_METADATA_ITEM_KEY)
+                                start += trans->fs_info->nodesize;
+                        else
+                                start += key.offset;
+                } else if (key.type == BTRFS_BLOCK_GROUP_ITEM_KEY) {
+                        if (key.objectid != block_group->key.objectid)
+                                break;
+                }
+
+                ret = btrfs_next_item(extent_root, path);
+                if (ret < 0)
+                        goto out;
+                if (ret)
+                        break;
+        }
+        if (start < end) {
+                ret = __add_to_free_space_tree(trans, block_group, path2,
+                                               start, end - start);
+                if (ret)
+                        goto out;
+        }
+
+        ret = 0;
+out:
+        btrfs_free_path(path2);
+        btrfs_free_path(path);
+        return ret;
+}
+
+int remove_block_group_free_space(struct btrfs_trans_handle *trans,
+				  struct btrfs_block_group_cache *block_group)
+{
+	struct btrfs_root *root = trans->fs_info->free_space_root;
+	struct btrfs_path *path;
+	struct btrfs_key key, found_key;
+	struct extent_buffer *leaf;
+	u64 start, end;
+	int done = 0, nr;
+	int ret;
+
+	path = btrfs_alloc_path();
+	if (!path) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	start = block_group->key.objectid;
+	end = block_group->key.objectid + block_group->key.offset;
+
+	key.objectid = end - 1;
+	key.type = (u8)-1;
+	key.offset = (u64)-1;
+
+	while (!done) {
+		ret = btrfs_search_prev_slot(trans, root, &key, path, -1, 1);
+		if (ret)
+			goto out;
+
+		leaf = path->nodes[0];
+		nr = 0;
+		path->slots[0]++;
+		while (path->slots[0] > 0) {
+			btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0] - 1);
+
+			if (found_key.type == BTRFS_FREE_SPACE_INFO_KEY) {
+				ASSERT(found_key.objectid == block_group->key.objectid);
+				ASSERT(found_key.offset == block_group->key.offset);
+				done = 1;
+				nr++;
+				path->slots[0]--;
+				break;
+			} else if (found_key.type == BTRFS_FREE_SPACE_EXTENT_KEY ||
+				   found_key.type == BTRFS_FREE_SPACE_BITMAP_KEY) {
+				ASSERT(found_key.objectid >= start);
+				ASSERT(found_key.objectid < end);
+				ASSERT(found_key.objectid + found_key.offset <= end);
+				nr++;
+				path->slots[0]--;
+			} else {
+				ASSERT(0);
+			}
+		}
+
+		ret = btrfs_del_items(trans, root, path, path->slots[0], nr);
+		if (ret)
+			goto out;
+		btrfs_release_path(path);
+	}
+
+	ret = 0;
+out:
+	btrfs_free_path(path);
+	if (ret)
+		btrfs_abort_transaction(trans, ret);
+	return ret;
+}
 static int clear_free_space_tree(struct btrfs_trans_handle *trans,
 				 struct btrfs_root *root)
 {
@@ -204,8 +949,8 @@ static int load_free_space_bitmaps(struct btrfs_fs_info *fs_info,
 
 		offset = key.objectid;
 		while (offset < key.objectid + key.offset) {
-			bit = free_space_test_bit(block_group, path, offset,
-						  fs_info->sectorsize);
+			bit = free_space_test_bit(block_group, path, offset);
+
 			if (prev_bit == 0 && bit == 1) {
 				extent_start = offset;
 			} else if (prev_bit == 1 && bit == 0) {
@@ -320,6 +1065,142 @@ static int load_free_space_extents(struct btrfs_fs_info *fs_info,
 	return ret;
 }
 
+struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
+                                     struct btrfs_fs_info *fs_info,
+                                     u64 objectid)
+{
+	struct extent_buffer *leaf;
+	struct btrfs_root *tree_root = fs_info->tree_root;
+	struct btrfs_root *root;
+	struct btrfs_key key;
+	int ret = 0;
+
+	root = kzalloc(sizeof(*root), GFP_KERNEL);
+	if (!root)
+		return ERR_PTR(-ENOMEM);
+
+	btrfs_setup_root(root, fs_info, objectid);
+	root->root_key.objectid = objectid;
+	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
+	root->root_key.offset = 0;
+
+	leaf = btrfs_alloc_free_block(trans, root, fs_info->nodesize, objectid, NULL, 0, 0, 0);
+	if (IS_ERR(leaf)) {
+		ret = PTR_ERR(leaf);
+		leaf = NULL;
+		goto fail;
+	}
+
+	memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
+	btrfs_set_header_bytenr(leaf, leaf->start);
+	btrfs_set_header_generation(leaf, trans->transid);
+	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
+	btrfs_set_header_owner(leaf, objectid);
+	root->node = leaf;
+	write_extent_buffer(leaf, fs_info->fsid, btrfs_header_fsid(), BTRFS_FSID_SIZE);
+	write_extent_buffer(leaf, fs_info->chunk_tree_uuid,
+			    btrfs_header_chunk_tree_uuid(leaf),
+			    BTRFS_UUID_SIZE);
+	btrfs_mark_buffer_dirty(leaf);
+
+	extent_buffer_get(root->node);
+	root->commit_root = root->node;
+	root->track_dirty = 1;
+
+	root->root_item.flags = 0;
+	root->root_item.byte_limit = 0;
+	btrfs_set_root_bytenr(&root->root_item, leaf->start);
+	btrfs_set_root_generation(&root->root_item, trans->transid);
+	btrfs_set_root_level(&root->root_item, 0);
+	btrfs_set_root_refs(&root->root_item, 1);
+	btrfs_set_root_used(&root->root_item, leaf->len);
+	btrfs_set_root_last_snapshot(&root->root_item, 0);
+	btrfs_set_root_dirid(&root->root_item, 0);
+	memset(root->root_item.uuid, 0, BTRFS_UUID_SIZE);
+	root->root_item.drop_level = 0;
+
+	key.objectid = objectid;
+	key.type = BTRFS_ROOT_ITEM_KEY;
+	key.offset = 0;
+	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
+	if (ret)
+		goto fail;
+
+	return root;
+
+fail:
+	if (leaf)
+		free_extent_buffer(leaf);
+
+	kfree(root);
+	return ERR_PTR(ret);
+}
+
+#define btrfs_set_fs_compat_ro(__fs_info, opt) \
+        __btrfs_set_fs_compat_ro((__fs_info), BTRFS_FEATURE_COMPAT_RO_##opt)
+
+static inline void __btrfs_set_fs_compat_ro(struct btrfs_fs_info *fs_info,
+                                            u64 flag)
+{
+	struct btrfs_super_block *disk_super;
+	u64 features;
+
+	disk_super = fs_info->super_copy;
+	features = btrfs_super_compat_ro_flags(disk_super);
+	if (!(features & flag)) {
+		features = btrfs_super_compat_ro_flags(disk_super);
+		if (!(features & flag)) {
+			features |= flag;
+			btrfs_set_super_compat_ro_flags(disk_super, features);
+		}
+	}
+}
+
+int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
+{
+	struct btrfs_trans_handle *trans;
+	struct btrfs_root *tree_root = fs_info->tree_root;
+	struct btrfs_root *free_space_root;
+	struct btrfs_block_group_cache *block_group;
+	u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
+	int ret;
+
+	trans = btrfs_start_transaction(tree_root, 0);
+	if (IS_ERR(trans))
+		return PTR_ERR(trans);
+
+	free_space_root = btrfs_create_tree(trans, fs_info,
+					    BTRFS_FREE_SPACE_TREE_OBJECTID);
+	if (IS_ERR(free_space_root)) {
+		ret = PTR_ERR(free_space_root);
+		goto abort;
+	}
+	fs_info->free_space_root = free_space_root;
+
+	do {
+		block_group = btrfs_lookup_first_block_group(fs_info, start);
+		if (!block_group)
+			break;
+		start = block_group->key.objectid + block_group->key.offset;
+		ret = populate_free_space_tree(trans, block_group);
+		if (ret)
+			goto abort;
+	} while (block_group);
+
+	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE);
+	btrfs_set_fs_compat_ro(fs_info, FREE_SPACE_TREE_VALID);
+
+	ret = btrfs_commit_transaction(trans, tree_root);
+	if (ret)
+		return ret;
+
+	return 0;
+
+abort:
+	btrfs_abort_transaction(trans, ret);
+	return ret;
+}
+
 int load_free_space_tree(struct btrfs_fs_info *fs_info,
 			 struct btrfs_block_group_cache *block_group)
 {
@@ -332,7 +1213,7 @@ int load_free_space_tree(struct btrfs_fs_info *fs_info,
 	path = btrfs_alloc_path();
 	if (!path)
 		return -ENOMEM;
-	path->reada = 1;
+	path->reada = READA_BACK;
 
 	info = search_free_space_info(NULL, fs_info, block_group, path, 0);
 	if (IS_ERR(info)) {
diff --git a/free-space-tree.h b/free-space-tree.h
index 4845f13e6808..0509e9ae21ae 100644
--- a/free-space-tree.h
+++ b/free-space-tree.h
@@ -22,5 +22,13 @@
 int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info);
 int load_free_space_tree(struct btrfs_fs_info *fs_info,
 			 struct btrfs_block_group_cache *block_group);
-
+int populate_free_space_tree(struct btrfs_trans_handle *trans,
+			     struct btrfs_block_group_cache *block_group);
+int remove_block_group_free_space(struct btrfs_trans_handle *trans,
+				  struct btrfs_block_group_cache *block_group);
+int add_to_free_space_tree(struct btrfs_trans_handle *trans, u64 start,
+			   u64 size);
+int remove_from_free_space_tree(struct btrfs_trans_handle *trans, u64 start,
+				u64 size);
+int btrfs_create_free_space_tree(struct btrfs_fs_info *info);
 #endif
-- 
2.7.4

[PATCH 4/6] btrfs-progs: Add freespace tree as compat_ro supported feature

[Nikolay Borisov]

The RO_FREE_SPACE_TREE(_VALID) flags are required in order to be able
to open an FST filesystem in repair mode. Add them to
BTRFS_FEATURE_COMPAT_RO_SUPP.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
---
 ctree.h | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/ctree.h b/ctree.h
index ade883fecbd6..ef05e8122982 100644
--- a/ctree.h
+++ b/ctree.h
@@ -497,7 +497,9 @@ struct btrfs_super_block {
  * added here until read-write support for the free space tree is implemented in
  * btrfs-progs.
  */
-#define BTRFS_FEATURE_COMPAT_RO_SUPP		0ULL
+#define BTRFS_FEATURE_COMPAT_RO_SUPP			\
+	(BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE |	\
+	 BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE_VALID)
 
 #define BTRFS_FEATURE_INCOMPAT_SUPP			\
 	(BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF |		\
-- 
2.7.4

[PATCH 5/6] btrfs-progs: check: Add support for freespace tree fixing

[Nikolay Borisov]

Now that all the prerequisite code for proper support of free space
tree repair is in, it's time to wire it in. This is achieved by first
hooking the freespace tree to the __free_extent/alloc_reserved_tree_block
functions. And then introducing a wrapper function to contains the
existing check_space_cache and the newly introduced repair code.
Finally, it's important to note that FST repair code first clears the
existing FST in case of any problem found and rebuilds it from scratch.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
---
 check/main.c  | 61 +++++++++++++++++++++++++++++++++++++----------------------
 extent-tree.c |  9 +++++++++
 2 files changed, 47 insertions(+), 23 deletions(-)

diff --git a/check/main.c b/check/main.c
index 3a5efaf615a9..44d734ff4254 100644
--- a/check/main.c
+++ b/check/main.c
@@ -5321,19 +5321,6 @@ static int check_space_cache(struct btrfs_root *root)
 	int ret;
 	int error = 0;
 
-	if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
-	    btrfs_super_generation(root->fs_info->super_copy) !=
-	    btrfs_super_cache_generation(root->fs_info->super_copy)) {
-		printf("cache and super generation don't match, space cache "
-		       "will be invalidated\n");
-		return 0;
-	}
-
-	if (ctx.progress_enabled) {
-		ctx.tp = TASK_FREE_SPACE;
-		task_start(ctx.info);
-	}
-
 	while (1) {
 		cache = btrfs_lookup_first_block_group(root->fs_info, start);
 		if (!cache)
@@ -5383,11 +5370,11 @@ static int check_space_cache(struct btrfs_root *root)
 		}
 	}
 
-	task_stop(ctx.info);
 
 	return error ? -EINVAL : 0;
 }
 
+
 /*
  * Check data checksum for [@bytenr, @bytenr + @num_bytes).
  *
@@ -9338,7 +9325,6 @@ static int do_clear_free_space_cache(struct btrfs_fs_info *fs_info,
 			ret = 1;
 			goto close_out;
 		}
-		printf("Clearing free space cache\n");
 		ret = clear_free_space_cache(fs_info);
 		if (ret) {
 			error("failed to clear free space cache");
@@ -9365,6 +9351,41 @@ static int do_clear_free_space_cache(struct btrfs_fs_info *fs_info,
 	return ret;
 }
 
+static int validate_free_space_cache(struct btrfs_root *root)
+{
+
+	int ret;
+
+	if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
+	    btrfs_super_generation(root->fs_info->super_copy) !=
+	    btrfs_super_cache_generation(root->fs_info->super_copy)) {
+		printf("cache and super generation don't match, space cache "
+		       "will be invalidated\n");
+		return 0;
+	}
+
+	if (ctx.progress_enabled) {
+		ctx.tp = TASK_FREE_SPACE;
+		task_start(ctx.info);
+	}
+
+	ret = check_space_cache(root);
+	if (ret && btrfs_fs_compat_ro(global_info, FREE_SPACE_TREE)
+	    && repair) {
+		ret = do_clear_free_space_cache(global_info, 2);
+		if (ret)
+			goto out;
+
+		ret = btrfs_create_free_space_tree(global_info);
+		if (ret)
+			error("couldn't repair freespace tree");
+	}
+
+out:
+	task_stop(ctx.info);
+	return ret ? -EINVAL : 0;
+}
+
 const char * const cmd_check_usage[] = {
 	"btrfs check [options] <device>",
 	"Check structural integrity of a filesystem (unmounted).",
@@ -9768,15 +9789,9 @@ int cmd_check(int argc, char **argv)
 		else
 			fprintf(stderr, "checking free space cache\n");
 	}
-	ret = check_space_cache(root);
+
+	ret = validate_free_space_cache(root);
 	err |= !!ret;
-	if (ret) {
-		if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
-			error("errors found in free space tree");
-		else
-			error("errors found in free space cache");
-		goto out;
-	}
 
 	/*
 	 * We used to have to have these hole extents in between our real
diff --git a/extent-tree.c b/extent-tree.c
index b9d51b388c9a..40117f81352e 100644
--- a/extent-tree.c
+++ b/extent-tree.c
@@ -29,6 +29,7 @@
 #include "crc32c.h"
 #include "volumes.h"
 #include "free-space-cache.h"
+#include "free-space-tree.h"
 #include "utils.h"
 
 #define PENDING_EXTENT_INSERT 0
@@ -2292,6 +2293,11 @@ static int __free_extent(struct btrfs_trans_handle *trans,
 			BUG_ON(ret);
 		}
 
+		ret = add_to_free_space_tree(trans, bytenr, num_bytes);
+		if (ret) {
+			goto fail;
+		}
+
 		update_block_group(trans->fs_info, bytenr, num_bytes, 0,
 				   mark_free);
 	}
@@ -2630,6 +2636,9 @@ static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
 	btrfs_mark_buffer_dirty(leaf);
 	btrfs_free_path(path);
 
+	ret = remove_from_free_space_tree(trans, ins->objectid, fs_info->nodesize);
+	if (ret)
+		return ret;
 	ret = update_block_group(fs_info, ins->objectid, fs_info->nodesize,
 				 1, 0);
 	return ret;
-- 
2.7.4

[PATCH 6/6] btrfs-progs: tests: Test for FST corruption detection/repair

[Nikolay Borisov]

Simple test case which preps a filesystem, then corrupts the FST and
finally repairs it. Tests both extent based and bitmap based FSTs.

Signed-off-by: Nikolay Borisov <nborisov@suse.com>
---
 tests/fsck-tests/035-freespacetree-repair/test.sh | 79 +++++++++++++++++++++++
 1 file changed, 79 insertions(+)
 create mode 100755 tests/fsck-tests/035-freespacetree-repair/test.sh

diff --git a/tests/fsck-tests/035-freespacetree-repair/test.sh b/tests/fsck-tests/035-freespacetree-repair/test.sh
new file mode 100755
index 000000000000..e19d997b3d6a
--- /dev/null
+++ b/tests/fsck-tests/035-freespacetree-repair/test.sh
@@ -0,0 +1,79 @@
+#!/bin/bash
+# Corrupt a filesystem that is using freespace tree and then ensure that 
+# btrfs check is able to repair it. This tests correct detection/repair of 
+# both a FREE_SPACE_EXTENT based FST and a FREE_SPACE_BITMAP based FST. 
+
+source "$TEST_TOP/common"
+
+repair_and_verify()
+{
+	# since repairing entails allocating a block, which in turn implies 
+	# FST modification another btrfs check is required to ensure that 
+	# FST modification logic is correct. 
+	run_check $SUDO_HELPER "$TOP/btrfs" check --repair "$TEST_DEV"
+	run_check $SUDO_HELPER "$TOP/btrfs" check "$TEST_DEV"
+}
+
+# wrapper for btrfs-corrupt-item
+# $1: Type of item we want to corrupt - extent or bitmap
+corrupt_fst_item()
+{
+	local type
+	local objectid
+	local offset
+	type="$1"
+
+	if [[ $type == "bitmap" ]]; then 
+		type=200
+		objectid=$("$TOP/btrfs" inspect-internal dump-tree -t 10 "$TEST_DEV" | \
+			grep -o "[[:digit:]]* FREE_SPACE_BITMAP [[:digit:]]*" | \
+			cut -d' ' -f1 | tail -2 | head -1)
+		offset=$("$TOP/btrfs" inspect-internal dump-tree -t 10 "$TEST_DEV" | \
+			grep -o "[[:digit:]]* FREE_SPACE_BITMAP [[:digit:]]*" | \
+			cut -d' ' -f3 |tail -2 | head -1)
+		echo "Corrupting $objectid,FREE_SPACE_BITMAP,$offset"
+	elif [[ $type == "extent" ]]; then
+		type=199
+		objectid=$("$TOP/btrfs" inspect-internal dump-tree -t 10 "$TEST_DEV" | \
+			grep -o "[[:digit:]]* FREE_SPACE_EXTENT [[:digit:]]*" | \
+			cut -d' ' -f1 | tail -2 | head -1)
+		offset=$("$TOP/btrfs" inspect-internal dump-tree -t 10 "$TEST_DEV" | \
+			grep -o "[[:digit:]]* FREE_SPACE_EXTENT [[:digit:]]*" | \
+			cut -d' ' -f3 | tail -2 | head -1)
+		echo "Corrupting $objectid,FREE_SPACE_EXTENT,$offset"
+	else
+		_fail "Unknown item type for corruption"
+	fi
+	
+
+	run_check "$TOP/btrfs-corrupt-block" -r 10 -K "$objectid,$type,$offset" \
+		-f offset "$TEST_DEV"
+}
+
+check_prereq btrfs
+check_prereq mkfs.btrfs
+check_global_prereq grep
+check_global_prereq tail 
+check_global_prereq head
+check_global_prereq cut
+
+setup_root_helper
+prepare_test_dev 256M
+
+run_check "$TOP/mkfs.btrfs" -n 4k -f "$TEST_DEV"
+run_check_mount_test_dev -oclear_cache,space_cache=v2
+
+#create files which will populate the FST
+for i in {1..3000}; do
+	fallocate -l 4k "$TEST_MNT/file.$i" 
+done
+
+run_check_umount_test_dev
+
+#now corrupt one of the bitmap items 
+corrupt_fst_item "bitmap"
+check_image "$TEST_DEV"
+
+#now corrupt an extent 
+corrupt_fst_item "extent"
+check_image "$TEST_DEV"
-- 
2.7.4

Re: [PATCH 1/6] btrfs-progs: Add support for freespace tree in btrfs_read_fs_root

[Omar Sandoval]

On Fri, Jun 15, 2018 at 02:05:57PM +0300, Nikolay Borisov wrote:
> For completeness sake add code to btrfs_read_fs_root so that it can
> handle the freespace tree.

Reviewed-by: Omar Sandoval <osandov@fb.com>

> Signed-off-by: Nikolay Borisov <nborisov@suse.com>
> ---
>  disk-io.c | 3 +++
>  1 file changed, 3 insertions(+)
> 
> diff --git a/disk-io.c b/disk-io.c
> index 8da6e3ce5fc8..9ad826b83b3e 100644
> --- a/disk-io.c
> +++ b/disk-io.c
> @@ -664,6 +664,9 @@ struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
>  	if (location->objectid == BTRFS_QUOTA_TREE_OBJECTID)
>  		return fs_info->quota_enabled ? fs_info->quota_root :
>  				ERR_PTR(-ENOENT);
> +	if (location->objectid == BTRFS_FREE_SPACE_TREE_OBJECTID)
> +        return fs_info->free_space_root ? fs_info->free_space_root :
> +                                                  ERR_PTR(-ENOENT);
>  
>  	BUG_ON(location->objectid == BTRFS_TREE_RELOC_OBJECTID ||
>  	       location->offset != (u64)-1);
> -- 
> 2.7.4
> 

Re: [PATCH 2/6] btrfs-progs: Add extent buffer bitmap manipulation infrastructure

[Omar Sandoval]

On Fri, Jun 15, 2018 at 02:05:58PM +0300, Nikolay Borisov wrote:
> Those functions are in preparation for adding the freespace tree
> repair code since it needs to be able to deal with bitmap based fsts.
> This patch adds extent_buffer_bitmap_set and extent_buffer_bitmap_clear
> functions. Since in userspace we don't have to deal with page mappings
> their implementation is vastly simplified by simply setting each bit in
> the passed range.
> 
> Signed-off-by: Nikolay Borisov <nborisov@suse.com>
> ---
>  extent_io.c | 39 +++++++++++++++++++++++++++++++++++++++
>  extent_io.h | 15 +++++++++++++++
>  2 files changed, 54 insertions(+)
> 
> diff --git a/extent_io.c b/extent_io.c
> index 198492699438..568a12f7084b 100644
> --- a/extent_io.c
> +++ b/extent_io.c
> @@ -204,6 +204,45 @@ static int clear_state_bit(struct extent_io_tree *tree,
>  	return ret;
>  }
>  
> +/**
> + * extent_buffer_bitmap_set - set an area of a bitmap
> + * @eb: the extent buffer
> + * @start: offset of the bitmap item in the extent buffer
> + * @pos: bit number of the first bit
> + * @len: number of bits to set
> + */
> +void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
> +                              unsigned long pos, unsigned long len)
> +{
> +        u8 *kaddr = (u8 *)eb->data + start;
> +
> +        while (len) {
> +		le_set_bit(pos, kaddr);
> +		pos++;
> +		len--;
> +        }
> +}
> +
> +
> +/**
> + * extent_buffer_bitmap_clear - clear an area of a bitmap
> + * @eb: the extent buffer
> + * @start: offset of the bitmap item in the extent buffer
> + * @pos: bit number of the first bit
> + * @len: number of bits to clear
> + */
> +void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
> +                                unsigned long pos, unsigned long len)
> +{
> +        u8 *kaddr = (u8 *)eb->data + start;
> +
> +        while (len) {
> +		le_clear_bit(pos, kaddr);
> +		pos++;
> +		len--;
> +        }
> +}
> +

Hm, bit-by-bit is pretty slow. We don't have to worry about the kmap
nonsense, but we can still do this byte-by-byte:

/**
 * extent_buffer_bitmap_set - set an area of a bitmap
 * @eb: the extent buffer
 * @start: offset of the bitmap item in the extent buffer
 * @pos: bit number of the first bit
 * @len: number of bits to set
 */
void extent_buffer_bitmap_set(struct extent_buffer *eb, unsigned long start,
			      unsigned long pos, unsigned long len)
{
        u8 *p = (u8 *)eb->data + start + BIT_BYTE(pos);
	const unsigned int size = pos + len;
	int bits_to_set = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
	u8 mask_to_set = BITMAP_FIRST_BYTE_MASK(pos);

	while (len >= bits_to_set) {
		*p |= mask_to_set;
		len -= bits_to_set;
		bits_to_set = BITS_PER_BYTE;
		mask_to_set = ~0;
		p++;
	}
	if (len) {
		mask_to_set &= BITMAP_LAST_BYTE_MASK(size);
		*p |= mask_to_set;
	}
}

/**
 * extent_buffer_bitmap_clear - clear an area of a bitmap
 * @eb: the extent buffer
 * @start: offset of the bitmap item in the extent buffer
 * @pos: bit number of the first bit
 * @len: number of bits to clear
 */
void extent_buffer_bitmap_clear(struct extent_buffer *eb, unsigned long start,
				unsigned long pos, unsigned long len)
{
        u8 *p = (u8 *)eb->data + start + BIT_BYTE(pos);
	const unsigned int size = pos + len;
	int bits_to_clear = BITS_PER_BYTE - (pos % BITS_PER_BYTE);
	u8 mask_to_clear = BITMAP_FIRST_BYTE_MASK(pos);

	while (len >= bits_to_clear) {
		*p &= ~mask_to_clear;
		len -= bits_to_clear;
		bits_to_clear = BITS_PER_BYTE;
		mask_to_clear = ~0;
		p++;
	}
	if (len) {
		mask_to_clear &= BITMAP_LAST_BYTE_MASK(size);
		*p &= ~mask_to_clear;
	}
}

I'm 95% sure that's right ;) Compare to __bitmap_set() and
__bitmap_clear() in the kernel.

Re: [PATCH 3/6] btrfs-progs: Pull free space tree related code from kernel

[Omar Sandoval]

On Fri, Jun 15, 2018 at 02:05:59PM +0300, Nikolay Borisov wrote:
> To help implement free space tree checker in user space some kernel
> function are necessary, namely iterating/deleting/adding freespace
> items, some internal search functions. Functions to populate a block
> group based on the extent tree. The code is largely copy/paste from
> the kernel with locking eliminated (i.e free_space_lock). It supports
> reading/writing of both bitmap and extent based FST trees.
> 
> Signed-off-by: Nikolay Borisov <nborisov@suse.com>

Why doesn't this include the bitmap <-> extent conversions? If we end up
rebuilding the free space tree, we're never going to use the bitmap
format, which sucks if the free space is fragmented.

Re: [PATCH 3/6] btrfs-progs: Pull free space tree related code from kernel

[Nikolay Borisov]

On 21.09.2018 23:19, Omar Sandoval wrote:
> On Fri, Jun 15, 2018 at 02:05:59PM +0300, Nikolay Borisov wrote:
>> To help implement free space tree checker in user space some kernel
>> function are necessary, namely iterating/deleting/adding freespace
>> items, some internal search functions. Functions to populate a block
>> group based on the extent tree. The code is largely copy/paste from
>> the kernel with locking eliminated (i.e free_space_lock). It supports
>> reading/writing of both bitmap and extent based FST trees.
>>
>> Signed-off-by: Nikolay Borisov <nborisov@suse.com>
> 
> Why doesn't this include the bitmap <-> extent conversions? If we end up
> rebuilding the free space tree, we're never going to use the bitmap
> format, which sucks if the free space is fragmented.
> 

The idea was to have *something* working initially and then add as
deemed appropriate.

Re: [PATCH 4/6] btrfs-progs: Add freespace tree as compat_ro supported feature

[Omar Sandoval]

On Fri, Jun 15, 2018 at 02:06:00PM +0300, Nikolay Borisov wrote:
> The RO_FREE_SPACE_TREE(_VALID) flags are required in order to be able
> to open an FST filesystem in repair mode. Add them to
> BTRFS_FEATURE_COMPAT_RO_SUPP.
> 
> Signed-off-by: Nikolay Borisov <nborisov@suse.com>
> ---
>  ctree.h | 4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> diff --git a/ctree.h b/ctree.h
> index ade883fecbd6..ef05e8122982 100644
> --- a/ctree.h
> +++ b/ctree.h
> @@ -497,7 +497,9 @@ struct btrfs_super_block {
>   * added here until read-write support for the free space tree is implemented in
>   * btrfs-progs.
>   */
> -#define BTRFS_FEATURE_COMPAT_RO_SUPP		0ULL
> +#define BTRFS_FEATURE_COMPAT_RO_SUPP			\
> +	(BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE |	\
> +	 BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE_VALID)


Have you tested whether btrfs-progs commands that modify the filesystem
(e.g., btrfstune or btrfs fi label) work with this series? Because that
is a requirement for claiming that we support this bit (at which point
we can delete the comment above). Also, this needs to happen _after_ we
hook up the free space tree with the extent tree.

See here for some historical context:
https://www.mail-archive.com/linux-btrfs@vger.kernel.org/msg57738.html

Re: [PATCH 5/6] btrfs-progs: check: Add support for freespace tree fixing

[Omar Sandoval]

On Fri, Jun 15, 2018 at 02:06:01PM +0300, Nikolay Borisov wrote:
> Now that all the prerequisite code for proper support of free space
> tree repair is in, it's time to wire it in. This is achieved by first
> hooking the freespace tree to the __free_extent/alloc_reserved_tree_block
> functions. And then introducing a wrapper function to contains the
> existing check_space_cache and the newly introduced repair code.
> Finally, it's important to note that FST repair code first clears the
> existing FST in case of any problem found and rebuilds it from scratch.
> 
> Signed-off-by: Nikolay Borisov <nborisov@suse.com>
> ---
>  check/main.c  | 61 +++++++++++++++++++++++++++++++++++++----------------------
>  extent-tree.c |  9 +++++++++
>  2 files changed, 47 insertions(+), 23 deletions(-)
> 
> diff --git a/check/main.c b/check/main.c
> index 3a5efaf615a9..44d734ff4254 100644
> --- a/check/main.c
> +++ b/check/main.c
> @@ -5321,19 +5321,6 @@ static int check_space_cache(struct btrfs_root *root)
>  	int ret;
>  	int error = 0;
>  
> -	if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
> -	    btrfs_super_generation(root->fs_info->super_copy) !=
> -	    btrfs_super_cache_generation(root->fs_info->super_copy)) {
> -		printf("cache and super generation don't match, space cache "
> -		       "will be invalidated\n");
> -		return 0;
> -	}
> -
> -	if (ctx.progress_enabled) {
> -		ctx.tp = TASK_FREE_SPACE;
> -		task_start(ctx.info);
> -	}
> -
>  	while (1) {
>  		cache = btrfs_lookup_first_block_group(root->fs_info, start);
>  		if (!cache)
> @@ -5383,11 +5370,11 @@ static int check_space_cache(struct btrfs_root *root)
>  		}
>  	}
>  
> -	task_stop(ctx.info);
>  
>  	return error ? -EINVAL : 0;
>  }
>  
> +

Stray newline.

>  /*
>   * Check data checksum for [@bytenr, @bytenr + @num_bytes).
>   *
> @@ -9338,7 +9325,6 @@ static int do_clear_free_space_cache(struct btrfs_fs_info *fs_info,
>  			ret = 1;
>  			goto close_out;
>  		}
> -		printf("Clearing free space cache\n");
>  		ret = clear_free_space_cache(fs_info);
>  		if (ret) {
>  			error("failed to clear free space cache");
> @@ -9365,6 +9351,41 @@ static int do_clear_free_space_cache(struct btrfs_fs_info *fs_info,
>  	return ret;
>  }
>  
> +static int validate_free_space_cache(struct btrfs_root *root)
> +{
> +
> +	int ret;
> +
> +	if (btrfs_super_cache_generation(root->fs_info->super_copy) != -1ULL &&
> +	    btrfs_super_generation(root->fs_info->super_copy) !=
> +	    btrfs_super_cache_generation(root->fs_info->super_copy)) {
> +		printf("cache and super generation don't match, space cache "
> +		       "will be invalidated\n");
> +		return 0;
> +	}
> +
> +	if (ctx.progress_enabled) {
> +		ctx.tp = TASK_FREE_SPACE;
> +		task_start(ctx.info);
> +	}
> +
> +	ret = check_space_cache(root);
> +	if (ret && btrfs_fs_compat_ro(global_info, FREE_SPACE_TREE)
> +	    && repair) {
> +		ret = do_clear_free_space_cache(global_info, 2);
> +		if (ret)
> +			goto out;
> +
> +		ret = btrfs_create_free_space_tree(global_info);
> +		if (ret)
> +			error("couldn't repair freespace tree");
> +	}
> +
> +out:
> +	task_stop(ctx.info);
> +	return ret ? -EINVAL : 0;
> +}
> +
>  const char * const cmd_check_usage[] = {
>  	"btrfs check [options] <device>",
>  	"Check structural integrity of a filesystem (unmounted).",
> @@ -9768,15 +9789,9 @@ int cmd_check(int argc, char **argv)
>  		else
>  			fprintf(stderr, "checking free space cache\n");
>  	}
> -	ret = check_space_cache(root);
> +
> +	ret = validate_free_space_cache(root);
>  	err |= !!ret;
> -	if (ret) {
> -		if (btrfs_fs_compat_ro(info, FREE_SPACE_TREE))
> -			error("errors found in free space tree");
> -		else
> -			error("errors found in free space cache");
> -		goto out;
> -	}
>  
>  	/*
>  	 * We used to have to have these hole extents in between our real

This approach seems reasonable.

> diff --git a/extent-tree.c b/extent-tree.c
> index b9d51b388c9a..40117f81352e 100644
> --- a/extent-tree.c
> +++ b/extent-tree.c
> @@ -29,6 +29,7 @@
>  #include "crc32c.h"
>  #include "volumes.h"
>  #include "free-space-cache.h"
> +#include "free-space-tree.h"
>  #include "utils.h"
>  
>  #define PENDING_EXTENT_INSERT 0
> @@ -2292,6 +2293,11 @@ static int __free_extent(struct btrfs_trans_handle *trans,
>  			BUG_ON(ret);
>  		}
>  
> +		ret = add_to_free_space_tree(trans, bytenr, num_bytes);
> +		if (ret) {
> +			goto fail;
> +		}
> +
>  		update_block_group(trans->fs_info, bytenr, num_bytes, 0,
>  				   mark_free);
>  	}
> @@ -2630,6 +2636,9 @@ static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
>  	btrfs_mark_buffer_dirty(leaf);
>  	btrfs_free_path(path);
>  
> +	ret = remove_from_free_space_tree(trans, ins->objectid, fs_info->nodesize);
> +	if (ret)
> +		return ret;
>  	ret = update_block_group(fs_info, ins->objectid, fs_info->nodesize,
>  				 1, 0);
>  	return ret;

Related to my comment on patch 4, the extent-tree.c changes should be a
separate patch that comes before we add the compat bit, assuming it
works with commands that open_ctree(..., OPEN_CTREE_WRITES) and modify
the fs.