V1->V2 - This includes my previous fix for the Incompat flags that I posted earlier - Fix kunmap() to use the page and not the vaddr, I messed this up when going from kmap_atomic to kmap This patch series introduces the ability for btrfs to store the free space cache ondisk to make the caching of a block group much quicker. Previously we had to search the entire extent-tree to look for gaps everytime we wanted to allocate in a block group. This approach instead dumps all of the free space cache to disk for every dirtied block group each time we commit the transaction. This is a disk format change, but in order to use the feature you will have to mount with -o space_cache, and then from then on you won''t be able to use old kernels with your filesystem. You can pull these patches from my git tree git://git.kernel.org/pub/scm/linux/kernel/git/josef/btrfs-work.git and they should pull right onto Chris''s btrfs-unstable tree. Please test this as it''s a big change and I can only test so much. That being said this has been run through xfstests thoroughly and I''ve been running it on a VM with btrfs as root. Thanks, Josef -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Josef Bacik
2010-Sep-22 18:39 UTC
[PATCH 1/3] Btrfs: create special free space cache inode FORMAT CHANGE
In order to save free space cache, we need an inode to hold the data, and we
need a special item to point at the right inode for the right block group. So
first, create a special item that will point to the right inode, and the number
of extent entries we will have and the number of bitmaps we will have. We
truncate and pre-allocate space everytime to make sure it''s uptodate.
This is a disk format change, HOWEVER, you need to mount with -o space_cache in
order to get it, otherwise this feature won''t be enabled.
Signed-off-by: Josef Bacik <josef@redhat.com>
---
fs/btrfs/ctree.h | 71 +++++++++++++-
fs/btrfs/disk-io.c | 3 +-
fs/btrfs/extent-tree.c | 220 +++++++++++++++++++++++++++++++++++++++++-
fs/btrfs/free-space-cache.c | 155 ++++++++++++++++++++++++++++++
fs/btrfs/free-space-cache.h | 11 ++
fs/btrfs/inode.c | 92 +++++++++++++++---
fs/btrfs/relocation.c | 91 +++++++++++++++++-
fs/btrfs/super.c | 11 ++-
fs/btrfs/transaction.c | 41 ++++++--
fs/btrfs/transaction.h | 4 +
10 files changed, 654 insertions(+), 45 deletions(-)
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index e9bf864..621bf90 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -99,6 +99,9 @@ struct btrfs_ordered_sum;
*/
#define BTRFS_EXTENT_CSUM_OBJECTID -10ULL
+/* For storing free space cache */
+#define BTRFS_FREE_SPACE_OBJECTID -11ULL
+
/* dummy objectid represents multiple objectids */
#define BTRFS_MULTIPLE_OBJECTIDS -255ULL
@@ -265,6 +268,22 @@ struct btrfs_chunk {
/* additional stripes go here */
} __attribute__ ((__packed__));
+#define BTRFS_FREE_SPACE_EXTENT 1
+#define BTRFS_FREE_SPACE_BITMAP 2
+
+struct btrfs_free_space_entry {
+ __le64 offset;
+ __le64 bytes;
+ u8 type;
+} __attribute__ ((__packed__));
+
+struct btrfs_free_space_header {
+ struct btrfs_disk_key location;
+ __le64 generation;
+ __le64 num_entries;
+ __le64 num_bitmaps;
+} __attribute__ ((__packed__));
+
static inline unsigned long btrfs_chunk_item_size(int num_stripes)
{
BUG_ON(num_stripes == 0);
@@ -375,13 +394,15 @@ struct btrfs_super_block {
* ones specified below then we will fail to mount
*/
#define BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF (1ULL << 0)
-#define BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL (2ULL << 0)
+#define BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL (1ULL << 1)
+#define BTRFS_FEATURE_INCOMPAT_SPACE_CACHE (1ULL << 2)
#define BTRFS_FEATURE_COMPAT_SUPP 0ULL
#define BTRFS_FEATURE_COMPAT_RO_SUPP 0ULL
-#define BTRFS_FEATURE_INCOMPAT_SUPP \
- (BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF | \
- BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL)
+#define BTRFS_FEATURE_INCOMPAT_SUPP \
+ (BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF | \
+ BTRFS_FEATURE_INCOMPAT_DEFAULT_SUBVOL | \
+ BTRFS_FEATURE_INCOMPAT_SPACE_CACHE)
/*
* A leaf is full of items. offset and size tell us where to find
@@ -750,6 +771,14 @@ enum btrfs_caching_type {
BTRFS_CACHE_FINISHED = 2,
};
+enum btrfs_disk_cache_state {
+ BTRFS_DC_WRITTEN = 0,
+ BTRFS_DC_ERROR = 1,
+ BTRFS_DC_CLEAR = 2,
+ BTRFS_DC_SETUP = 3,
+ BTRFS_DC_NEED_WRITE = 4,
+};
+
struct btrfs_caching_control {
struct list_head list;
struct mutex mutex;
@@ -763,6 +792,7 @@ struct btrfs_block_group_cache {
struct btrfs_key key;
struct btrfs_block_group_item item;
struct btrfs_fs_info *fs_info;
+ struct inode *inode;
spinlock_t lock;
u64 pinned;
u64 reserved;
@@ -773,8 +803,11 @@ struct btrfs_block_group_cache {
int extents_thresh;
int free_extents;
int total_bitmaps;
- int ro;
- int dirty;
+ int ro:1;
+ int dirty:1;
+ int iref:1;
+
+ int disk_cache_state;
/* cache tracking stuff */
int cached;
@@ -1665,6 +1698,27 @@ static inline void btrfs_set_dir_item_key(struct
extent_buffer *eb,
write_eb_member(eb, item, struct btrfs_dir_item, location, key);
}
+BTRFS_SETGET_FUNCS(free_space_entries, struct btrfs_free_space_header,
+ num_entries, 64);
+BTRFS_SETGET_FUNCS(free_space_bitmaps, struct btrfs_free_space_header,
+ num_bitmaps, 64);
+BTRFS_SETGET_FUNCS(free_space_generation, struct btrfs_free_space_header,
+ generation, 64);
+
+static inline void btrfs_free_space_key(struct extent_buffer *eb,
+ struct btrfs_free_space_header *h,
+ struct btrfs_disk_key *key)
+{
+ read_eb_member(eb, h, struct btrfs_free_space_header, location, key);
+}
+
+static inline void btrfs_set_free_space_key(struct extent_buffer *eb,
+ struct btrfs_free_space_header *h,
+ struct btrfs_disk_key *key)
+{
+ write_eb_member(eb, h, struct btrfs_free_space_header, location, key);
+}
+
/* struct btrfs_disk_key */
BTRFS_SETGET_STACK_FUNCS(disk_key_objectid, struct btrfs_disk_key,
objectid, 64);
@@ -2115,6 +2169,7 @@ int btrfs_set_block_group_ro(struct btrfs_root *root,
struct btrfs_block_group_cache *cache);
int btrfs_set_block_group_rw(struct btrfs_root *root,
struct btrfs_block_group_cache *cache);
+void btrfs_put_block_group_cache(struct btrfs_fs_info *info);
/* ctree.c */
int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
int level, int *slot);
@@ -2426,6 +2481,10 @@ void btrfs_run_delayed_iputs(struct btrfs_root *root);
int btrfs_prealloc_file_range(struct inode *inode, int mode,
u64 start, u64 num_bytes, u64 min_size,
loff_t actual_len, u64 *alloc_hint);
+int btrfs_prealloc_file_range_trans(struct inode *inode,
+ struct btrfs_trans_handle *trans, int mode,
+ u64 start, u64 num_bytes, u64 min_size,
+ loff_t actual_len, u64 *alloc_hint);
extern const struct dentry_operations btrfs_dentry_operations;
/* ioctl.c */
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 34f7c37..163ee07 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -1685,7 +1685,6 @@ struct btrfs_root *open_ctree(struct super_block *sb,
__setup_root(4096, 4096, 4096, 4096, tree_root,
fs_info, BTRFS_ROOT_TREE_OBJECTID);
-
bh = btrfs_read_dev_super(fs_devices->latest_bdev);
if (!bh)
goto fail_iput;
@@ -1993,6 +1992,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
if (!(sb->s_flags & MS_RDONLY)) {
down_read(&fs_info->cleanup_work_sem);
btrfs_orphan_cleanup(fs_info->fs_root);
+ btrfs_orphan_cleanup(fs_info->tree_root);
up_read(&fs_info->cleanup_work_sem);
}
@@ -2421,6 +2421,7 @@ int close_ctree(struct btrfs_root *root)
fs_info->closing = 1;
smp_mb();
+ btrfs_put_block_group_cache(fs_info);
if (!(fs_info->sb->s_flags & MS_RDONLY)) {
ret = btrfs_commit_super(root);
if (ret)
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index a46b64d..b8151ad 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2688,6 +2688,109 @@ next_block_group(struct btrfs_root *root,
return cache;
}
+static int cache_save_setup(struct btrfs_block_group_cache *block_group,
+ struct btrfs_trans_handle *trans,
+ struct btrfs_path *path)
+{
+ struct btrfs_root *root = block_group->fs_info->tree_root;
+ struct inode *inode = NULL;
+ u64 alloc_hint = 0;
+ int num_pages = 0;
+ int retries = 0;
+ int ret = 0;
+
+ /*
+ * If this block group is smaller than 100 megs don''t bother caching
the
+ * block group.
+ */
+ if (block_group->key.offset < (100 * 1024 * 1024)) {
+ spin_lock(&block_group->lock);
+ block_group->disk_cache_state = BTRFS_DC_WRITTEN;
+ spin_unlock(&block_group->lock);
+ return 0;
+ }
+
+again:
+ inode = lookup_free_space_inode(root, block_group, path);
+ if (IS_ERR(inode) && PTR_ERR(inode) != -ENOENT) {
+ ret = PTR_ERR(inode);
+ btrfs_release_path(root, path);
+ goto out;
+ }
+
+ if (IS_ERR(inode)) {
+ BUG_ON(retries);
+ retries++;
+
+ if (block_group->ro)
+ goto out_free;
+
+ ret = create_free_space_inode(root, trans, block_group, path);
+ if (ret)
+ goto out_free;
+ goto again;
+ }
+
+ /*
+ * We want to set the generation to 0, that way if anything goes wrong
+ * from here on out we know not to trust this cache when we load up next
+ * time.
+ */
+ BTRFS_I(inode)->generation = 0;
+ ret = btrfs_update_inode(trans, root, inode);
+ WARN_ON(ret);
+
+ if (i_size_read(inode) > 0) {
+ ret = btrfs_truncate_free_space_cache(root, trans, path,
+ inode);
+ if (ret)
+ goto out_put;
+ }
+
+ spin_lock(&block_group->lock);
+ if (block_group->cached != BTRFS_CACHE_FINISHED) {
+ spin_unlock(&block_group->lock);
+ goto out_put;
+ }
+ spin_unlock(&block_group->lock);
+
+ num_pages = (int)div64_u64(block_group->key.offset, 1024 * 1024 * 1024);
+ if (!num_pages)
+ num_pages = 1;
+
+ /*
+ * Just to make absolutely sure we have enough space, we''re going to
+ * preallocate 12 pages worth of space for each block group. In
+ * practice we ought to use at most 8, but we need extra space so we can
+ * add our header and have a terminator between the extents and the
+ * bitmaps.
+ */
+ num_pages *= 16;
+ num_pages *= PAGE_CACHE_SIZE;
+
+ ret = btrfs_check_data_free_space(inode, num_pages);
+ if (ret)
+ goto out_put;
+
+ ret = btrfs_prealloc_file_range_trans(inode, trans, 0, 0, num_pages,
+ num_pages, num_pages,
+ &alloc_hint);
+ btrfs_free_reserved_data_space(inode, num_pages);
+out_put:
+ iput(inode);
+out_free:
+ btrfs_release_path(root, path);
+out:
+ spin_lock(&block_group->lock);
+ if (ret)
+ block_group->disk_cache_state = BTRFS_DC_ERROR;
+ else
+ block_group->disk_cache_state = BTRFS_DC_SETUP;
+ spin_unlock(&block_group->lock);
+
+ return ret;
+}
+
int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
struct btrfs_root *root)
{
@@ -2700,6 +2803,25 @@ int btrfs_write_dirty_block_groups(struct
btrfs_trans_handle *trans,
if (!path)
return -ENOMEM;
+again:
+ while (1) {
+ cache = btrfs_lookup_first_block_group(root->fs_info, last);
+ while (cache) {
+ if (cache->disk_cache_state == BTRFS_DC_CLEAR)
+ break;
+ cache = next_block_group(root, cache);
+ }
+ if (!cache) {
+ if (last == 0)
+ break;
+ last = 0;
+ continue;
+ }
+ err = cache_save_setup(cache, trans, path);
+ last = cache->key.objectid + cache->key.offset;
+ btrfs_put_block_group(cache);
+ }
+
while (1) {
if (last == 0) {
err = btrfs_run_delayed_refs(trans, root,
@@ -2709,6 +2831,11 @@ int btrfs_write_dirty_block_groups(struct
btrfs_trans_handle *trans,
cache = btrfs_lookup_first_block_group(root->fs_info, last);
while (cache) {
+ if (cache->disk_cache_state == BTRFS_DC_CLEAR) {
+ btrfs_put_block_group(cache);
+ goto again;
+ }
+
if (cache->dirty)
break;
cache = next_block_group(root, cache);
@@ -2883,11 +3010,16 @@ int btrfs_check_data_free_space(struct inode *inode, u64
bytes)
struct btrfs_space_info *data_sinfo;
struct btrfs_root *root = BTRFS_I(inode)->root;
u64 used;
- int ret = 0, committed = 0;
+ int ret = 0, committed = 0, alloc_chunk = 1;
/* make sure bytes are sectorsize aligned */
bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize -
1);
+ if (root == root->fs_info->tree_root) {
+ alloc_chunk = 0;
+ committed = 1;
+ }
+
data_sinfo = BTRFS_I(inode)->space_info;
if (!data_sinfo)
goto alloc;
@@ -2906,7 +3038,7 @@ again:
* if we don''t have enough free bytes in this space then we need
* to alloc a new chunk.
*/
- if (!data_sinfo->full) {
+ if (!data_sinfo->full && alloc_chunk) {
u64 alloc_target;
data_sinfo->force_alloc = 1;
@@ -3777,12 +3909,13 @@ static int update_block_group(struct btrfs_trans_handle
*trans,
struct btrfs_root *root,
u64 bytenr, u64 num_bytes, int alloc)
{
- struct btrfs_block_group_cache *cache;
+ struct btrfs_block_group_cache *cache = NULL;
struct btrfs_fs_info *info = root->fs_info;
- int factor;
+ u64 features = btrfs_super_incompat_flags(&info->super_copy);
u64 total = num_bytes;
u64 old_val;
u64 byte_in_group;
+ int factor;
/* block accounting for super block */
spin_lock(&info->delalloc_lock);
@@ -3804,11 +3937,17 @@ static int update_block_group(struct btrfs_trans_handle
*trans,
factor = 2;
else
factor = 1;
+
byte_in_group = bytenr - cache->key.objectid;
WARN_ON(byte_in_group > cache->key.offset);
spin_lock(&cache->space_info->lock);
spin_lock(&cache->lock);
+
+ if (cache->disk_cache_state < BTRFS_DC_CLEAR &&
+ features & BTRFS_FEATURE_INCOMPAT_SPACE_CACHE)
+ cache->disk_cache_state = BTRFS_DC_CLEAR;
+
cache->dirty = 1;
old_val = btrfs_block_group_used(&cache->item);
num_bytes = min(total, cache->key.offset - byte_in_group);
@@ -7814,6 +7953,39 @@ out:
return ret;
}
+void btrfs_put_block_group_cache(struct btrfs_fs_info *info)
+{
+ struct btrfs_block_group_cache *block_group;
+ u64 last = 0;
+
+ while (1) {
+ struct inode *inode;
+
+ block_group = btrfs_lookup_first_block_group(info, last);
+ while (block_group) {
+ spin_lock(&block_group->lock);
+ if (block_group->iref)
+ break;
+ spin_unlock(&block_group->lock);
+ block_group = next_block_group(info->tree_root,
+ block_group);
+ }
+ if (!block_group) {
+ if (last == 0)
+ break;
+ last = 0;
+ continue;
+ }
+
+ inode = block_group->inode;
+ block_group->iref = 0;
+ spin_unlock(&block_group->lock);
+ iput(inode);
+ last = block_group->key.objectid + block_group->key.offset;
+ btrfs_put_block_group(block_group);
+ }
+}
+
int btrfs_free_block_groups(struct btrfs_fs_info *info)
{
struct btrfs_block_group_cache *block_group;
@@ -8032,6 +8204,7 @@ int btrfs_make_block_group(struct btrfs_trans_handle
*trans,
cache->key.offset = size;
cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
cache->sectorsize = root->sectorsize;
+ cache->fs_info = root->fs_info;
/*
* we only want to have 32k of ram per block group for keeping track
@@ -8088,7 +8261,9 @@ int btrfs_remove_block_group(struct btrfs_trans_handle
*trans,
struct btrfs_path *path;
struct btrfs_block_group_cache *block_group;
struct btrfs_free_cluster *cluster;
+ struct btrfs_root *tree_root = root->fs_info->tree_root;
struct btrfs_key key;
+ struct inode *inode;
int ret;
root = root->fs_info->extent_root;
@@ -8097,8 +8272,6 @@ int btrfs_remove_block_group(struct btrfs_trans_handle
*trans,
BUG_ON(!block_group);
BUG_ON(!block_group->ro);
- memcpy(&key, &block_group->key, sizeof(key));
-
/* make sure this block group isn''t part of an allocation cluster */
cluster = &root->fs_info->data_alloc_cluster;
spin_lock(&cluster->refill_lock);
@@ -8117,6 +8290,39 @@ int btrfs_remove_block_group(struct btrfs_trans_handle
*trans,
path = btrfs_alloc_path();
BUG_ON(!path);
+ inode = lookup_free_space_inode(root, block_group, path);
+ if (!IS_ERR(inode)) {
+ btrfs_orphan_add(trans, inode);
+ clear_nlink(inode);
+ /* One for the block groups ref */
+ spin_lock(&block_group->lock);
+ if (block_group->iref) {
+ block_group->inode = NULL;
+ spin_unlock(&block_group->lock);
+ iput(inode);
+ } else {
+ spin_unlock(&block_group->lock);
+ }
+ /* One for our lookup ref */
+ iput(inode);
+ }
+
+ key.objectid = BTRFS_FREE_SPACE_OBJECTID;
+ key.offset = block_group->key.objectid;
+ key.type = 0;
+
+ ret = btrfs_search_slot(trans, tree_root, &key, path, -1, 1);
+ if (ret < 0)
+ goto out;
+ if (ret > 0)
+ btrfs_release_path(tree_root, path);
+ if (ret == 0) {
+ ret = btrfs_del_item(trans, tree_root, path);
+ if (ret)
+ goto out;
+ btrfs_release_path(tree_root, path);
+ }
+
spin_lock(&root->fs_info->block_group_cache_lock);
rb_erase(&block_group->cache_node,
&root->fs_info->block_group_cache_tree);
@@ -8140,6 +8346,8 @@ int btrfs_remove_block_group(struct btrfs_trans_handle
*trans,
block_group->space_info->bytes_readonly -= block_group->key.offset;
spin_unlock(&block_group->space_info->lock);
+ memcpy(&key, &block_group->key, sizeof(key));
+
btrfs_clear_space_info_full(root->fs_info);
btrfs_put_block_group(block_group);
diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
index f488fac..05efcc7 100644
--- a/fs/btrfs/free-space-cache.c
+++ b/fs/btrfs/free-space-cache.c
@@ -23,10 +23,165 @@
#include "ctree.h"
#include "free-space-cache.h"
#include "transaction.h"
+#include "disk-io.h"
#define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8)
#define MAX_CACHE_BYTES_PER_GIG (32 * 1024)
+struct inode *lookup_free_space_inode(struct btrfs_root *root,
+ struct btrfs_block_group_cache
+ *block_group, struct btrfs_path *path)
+{
+ struct btrfs_key key;
+ struct btrfs_key location;
+ struct btrfs_disk_key disk_key;
+ struct btrfs_free_space_header *header;
+ struct extent_buffer *leaf;
+ struct inode *inode = NULL;
+ int ret;
+
+ spin_lock(&block_group->lock);
+ if (block_group->inode)
+ inode = igrab(block_group->inode);
+ spin_unlock(&block_group->lock);
+ if (inode)
+ return inode;
+
+ key.objectid = BTRFS_FREE_SPACE_OBJECTID;
+ key.offset = block_group->key.objectid;
+ key.type = 0;
+
+ ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+ if (ret < 0)
+ return ERR_PTR(ret);
+ if (ret > 0) {
+ btrfs_release_path(root, path);
+ return ERR_PTR(-ENOENT);
+ }
+
+ leaf = path->nodes[0];
+ header = btrfs_item_ptr(leaf, path->slots[0],
+ struct btrfs_free_space_header);
+ btrfs_free_space_key(leaf, header, &disk_key);
+ btrfs_disk_key_to_cpu(&location, &disk_key);
+ btrfs_release_path(root, path);
+
+ inode = btrfs_iget(root->fs_info->sb, &location, root, NULL);
+ if (!inode)
+ return ERR_PTR(-ENOENT);
+ if (IS_ERR(inode))
+ return inode;
+ if (is_bad_inode(inode)) {
+ iput(inode);
+ return ERR_PTR(-ENOENT);
+ }
+
+ spin_lock(&block_group->lock);
+ if (!root->fs_info->closing) {
+ block_group->inode = igrab(inode);
+ block_group->iref = 1;
+ }
+ spin_unlock(&block_group->lock);
+
+ return inode;
+}
+
+int create_free_space_inode(struct btrfs_root *root,
+ struct btrfs_trans_handle *trans,
+ struct btrfs_block_group_cache *block_group,
+ struct btrfs_path *path)
+{
+ struct btrfs_key key;
+ struct btrfs_disk_key disk_key;
+ struct btrfs_free_space_header *header;
+ struct btrfs_inode_item *inode_item;
+ struct extent_buffer *leaf;
+ u64 objectid;
+ int ret;
+
+ ret = btrfs_find_free_objectid(trans, root, 0, &objectid);
+ if (ret < 0)
+ return ret;
+
+ ret = btrfs_insert_empty_inode(trans, root, path, objectid);
+ if (ret)
+ return ret;
+
+ leaf = path->nodes[0];
+ inode_item = btrfs_item_ptr(leaf, path->slots[0],
+ struct btrfs_inode_item);
+ btrfs_item_key(leaf, &disk_key, path->slots[0]);
+ memset_extent_buffer(leaf, 0, (unsigned long)inode_item,
+ sizeof(*inode_item));
+ btrfs_set_inode_generation(leaf, inode_item, trans->transid);
+ btrfs_set_inode_size(leaf, inode_item, 0);
+ btrfs_set_inode_nbytes(leaf, inode_item, 0);
+ btrfs_set_inode_uid(leaf, inode_item, 0);
+ btrfs_set_inode_gid(leaf, inode_item, 0);
+ btrfs_set_inode_mode(leaf, inode_item, S_IFREG | 0600);
+ btrfs_set_inode_flags(leaf, inode_item, BTRFS_INODE_NOCOMPRESS |
+ BTRFS_INODE_PREALLOC | BTRFS_INODE_NODATASUM);
+ btrfs_set_inode_nlink(leaf, inode_item, 1);
+ btrfs_set_inode_transid(leaf, inode_item, trans->transid);
+ btrfs_set_inode_block_group(leaf, inode_item,
+ block_group->key.objectid);
+ btrfs_mark_buffer_dirty(leaf);
+ btrfs_release_path(root, path);
+
+ key.objectid = BTRFS_FREE_SPACE_OBJECTID;
+ key.offset = block_group->key.objectid;
+ key.type = 0;
+
+ ret = btrfs_insert_empty_item(trans, root, path, &key,
+ sizeof(struct btrfs_free_space_header));
+ if (ret < 0) {
+ btrfs_release_path(root, path);
+ return ret;
+ }
+ leaf = path->nodes[0];
+ header = btrfs_item_ptr(leaf, path->slots[0],
+ struct btrfs_free_space_header);
+ memset_extent_buffer(leaf, 0, (unsigned long)header, sizeof(*header));
+ btrfs_set_free_space_key(leaf, header, &disk_key);
+ btrfs_mark_buffer_dirty(leaf);
+ btrfs_release_path(root, path);
+
+ return 0;
+}
+
+int btrfs_truncate_free_space_cache(struct btrfs_root *root,
+ struct btrfs_trans_handle *trans,
+ struct btrfs_path *path,
+ struct inode *inode)
+{
+ loff_t oldsize;
+ int ret = 0;
+
+ trans->block_rsv = root->orphan_block_rsv;
+ ret = btrfs_block_rsv_check(trans, root,
+ root->orphan_block_rsv,
+ 0, 5);
+ if (ret)
+ return ret;
+
+ oldsize = i_size_read(inode);
+ btrfs_i_size_write(inode, 0);
+ truncate_pagecache(inode, oldsize, 0);
+
+ /*
+ * We don''t need an orphan item because truncating the free space
cache
+ * will never be split across transactions.
+ */
+ ret = btrfs_truncate_inode_items(trans, root, inode,
+ 0, BTRFS_EXTENT_DATA_KEY);
+ if (ret) {
+ WARN_ON(1);
+ return ret;
+ }
+
+ return btrfs_update_inode(trans, root, inode);
+}
+
static inline unsigned long offset_to_bit(u64 bitmap_start, u64 sectorsize,
u64 offset)
{
diff --git a/fs/btrfs/free-space-cache.h b/fs/btrfs/free-space-cache.h
index 890a8e7..45be29e 100644
--- a/fs/btrfs/free-space-cache.h
+++ b/fs/btrfs/free-space-cache.h
@@ -27,6 +27,17 @@ struct btrfs_free_space {
struct list_head list;
};
+struct inode *lookup_free_space_inode(struct btrfs_root *root,
+ struct btrfs_block_group_cache
+ *block_group, struct btrfs_path *path);
+int create_free_space_inode(struct btrfs_root *root,
+ struct btrfs_trans_handle *trans,
+ struct btrfs_block_group_cache *block_group,
+ struct btrfs_path *path);
+int btrfs_truncate_free_space_cache(struct btrfs_root *root,
+ struct btrfs_trans_handle *trans,
+ struct btrfs_path *path,
+ struct inode *inode);
int btrfs_add_free_space(struct btrfs_block_group_cache *block_group,
u64 bytenr, u64 size);
int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index f08427c..16b708a 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -1700,6 +1700,7 @@ static int btrfs_finish_ordered_io(struct inode *inode,
u64 start, u64 end)
ordered_extent->len);
BUG_ON(ret);
} else {
+ BUG_ON(root == root->fs_info->tree_root);
ret = insert_reserved_file_extent(trans, inode,
ordered_extent->file_offset,
ordered_extent->start,
@@ -3197,7 +3198,7 @@ int btrfs_truncate_inode_items(struct btrfs_trans_handle
*trans,
BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
- if (root->ref_cows)
+ if (root->ref_cows || root == root->fs_info->tree_root)
btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
path = btrfs_alloc_path();
@@ -3345,7 +3346,8 @@ delete:
} else {
break;
}
- if (found_extent && root->ref_cows) {
+ if (found_extent && (root->ref_cows ||
+ root == root->fs_info->tree_root)) {
btrfs_set_path_blocking(path);
ret = btrfs_free_extent(trans, root, extent_start,
extent_num_bytes, 0,
@@ -3883,7 +3885,14 @@ static void inode_tree_del(struct inode *inode)
}
spin_unlock(&root->inode_lock);
- if (empty && btrfs_root_refs(&root->root_item) == 0) {
+ /*
+ * Free space cache has inodes in the tree root, but the tree root has a
+ * root_refs of 0, so this could end up dropping the tree root as a
+ * snapshot, so we need the extra !root->fs_info->tree_root check to
+ * make sure we don''t drop it.
+ */
+ if (empty && btrfs_root_refs(&root->root_item) == 0 &&
+ root != root->fs_info->tree_root) {
synchronize_srcu(&root->fs_info->subvol_srcu);
spin_lock(&root->inode_lock);
empty = RB_EMPTY_ROOT(&root->inode_tree);
@@ -4277,14 +4286,24 @@ int btrfs_write_inode(struct inode *inode, struct
writeback_control *wbc)
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_trans_handle *trans;
int ret = 0;
+ bool nolock = false;
if (BTRFS_I(inode)->dummy_inode)
return 0;
+ smp_mb();
+ nolock = (root->fs_info->closing && root ==
root->fs_info->tree_root);
+
if (wbc->sync_mode == WB_SYNC_ALL) {
- trans = btrfs_join_transaction(root, 1);
+ if (nolock)
+ trans = btrfs_join_transaction_nolock(root, 1);
+ else
+ trans = btrfs_join_transaction(root, 1);
btrfs_set_trans_block_group(trans, inode);
- ret = btrfs_commit_transaction(trans, root);
+ if (nolock)
+ ret = btrfs_end_transaction_nolock(trans, root);
+ else
+ ret = btrfs_commit_transaction(trans, root);
}
return ret;
}
@@ -6312,6 +6331,21 @@ void btrfs_destroy_inode(struct inode *inode)
spin_unlock(&root->fs_info->ordered_extent_lock);
}
+ if (root == root->fs_info->tree_root) {
+ struct btrfs_block_group_cache *block_group;
+
+ block_group = btrfs_lookup_block_group(root->fs_info,
+ BTRFS_I(inode)->block_group);
+ if (block_group && block_group->inode == inode) {
+ spin_lock(&block_group->lock);
+ block_group->inode = NULL;
+ spin_unlock(&block_group->lock);
+ btrfs_put_block_group(block_group);
+ } else if (block_group) {
+ btrfs_put_block_group(block_group);
+ }
+ }
+
spin_lock(&root->orphan_lock);
if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
@@ -6343,7 +6377,8 @@ free:
void btrfs_drop_inode(struct inode *inode)
{
struct btrfs_root *root = BTRFS_I(inode)->root;
- if (inode->i_nlink > 0 &&
btrfs_root_refs(&root->root_item) == 0)
+ if (inode->i_nlink > 0 &&
(btrfs_root_refs(&root->root_item) == 0 &&
+ root != root->fs_info->tree_root))
generic_delete_inode(inode);
else
generic_drop_inode(inode);
@@ -6760,27 +6795,33 @@ out_unlock:
return err;
}
-int btrfs_prealloc_file_range(struct inode *inode, int mode,
- u64 start, u64 num_bytes, u64 min_size,
- loff_t actual_len, u64 *alloc_hint)
+static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
+ u64 start, u64 num_bytes, u64 min_size,
+ loff_t actual_len, u64 *alloc_hint,
+ struct btrfs_trans_handle *trans)
{
- struct btrfs_trans_handle *trans;
struct btrfs_root *root = BTRFS_I(inode)->root;
struct btrfs_key ins;
u64 cur_offset = start;
int ret = 0;
+ bool own_trans = true;
+ if (trans)
+ own_trans = false;
while (num_bytes > 0) {
- trans = btrfs_start_transaction(root, 3);
- if (IS_ERR(trans)) {
- ret = PTR_ERR(trans);
- break;
+ if (own_trans) {
+ trans = btrfs_start_transaction(root, 3);
+ if (IS_ERR(trans)) {
+ ret = PTR_ERR(trans);
+ break;
+ }
}
ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
0, *alloc_hint, (u64)-1, &ins, 1);
if (ret) {
- btrfs_end_transaction(trans, root);
+ if (own_trans)
+ btrfs_end_transaction(trans, root);
break;
}
@@ -6813,11 +6854,30 @@ int btrfs_prealloc_file_range(struct inode *inode, int
mode,
ret = btrfs_update_inode(trans, root, inode);
BUG_ON(ret);
- btrfs_end_transaction(trans, root);
+ if (own_trans)
+ btrfs_end_transaction(trans, root);
}
return ret;
}
+int btrfs_prealloc_file_range(struct inode *inode, int mode,
+ u64 start, u64 num_bytes, u64 min_size,
+ loff_t actual_len, u64 *alloc_hint)
+{
+ return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
+ min_size, actual_len, alloc_hint,
+ NULL);
+}
+
+int btrfs_prealloc_file_range_trans(struct inode *inode,
+ struct btrfs_trans_handle *trans, int mode,
+ u64 start, u64 num_bytes, u64 min_size,
+ loff_t actual_len, u64 *alloc_hint)
+{
+ return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
+ min_size, actual_len, alloc_hint, trans);
+}
+
static long btrfs_fallocate(struct inode *inode, int mode,
loff_t offset, loff_t len)
{
diff --git a/fs/btrfs/relocation.c b/fs/btrfs/relocation.c
index b37d723..af339ee 100644
--- a/fs/btrfs/relocation.c
+++ b/fs/btrfs/relocation.c
@@ -29,6 +29,7 @@
#include "locking.h"
#include "btrfs_inode.h"
#include "async-thread.h"
+#include "free-space-cache.h"
/*
* backref_node, mapping_node and tree_block start with this
@@ -3191,6 +3192,54 @@ static int block_use_full_backref(struct reloc_control
*rc,
return ret;
}
+static int delete_block_group_cache(struct btrfs_fs_info *fs_info,
+ struct inode *inode, u64 ino)
+{
+ struct btrfs_key key;
+ struct btrfs_path *path;
+ struct btrfs_root *root = fs_info->tree_root;
+ struct btrfs_trans_handle *trans;
+ unsigned long nr;
+ int ret = 0;
+
+ if (inode)
+ goto truncate;
+
+ key.objectid = ino;
+ key.type = BTRFS_INODE_ITEM_KEY;
+ key.offset = 0;
+
+ inode = btrfs_iget(fs_info->sb, &key, root, NULL);
+ if (!inode || IS_ERR(inode) || is_bad_inode(inode)) {
+ if (inode && !IS_ERR(inode))
+ iput(inode);
+ return -ENOENT;
+ }
+
+truncate:
+ path = btrfs_alloc_path();
+ if (!path) {
+ ret = -ENOMEM;
+ goto out;
+ }
+
+ trans = btrfs_join_transaction(root, 0);
+ if (IS_ERR(trans)) {
+ btrfs_free_path(path);
+ goto out;
+ }
+
+ ret = btrfs_truncate_free_space_cache(root, trans, path, inode);
+
+ btrfs_free_path(path);
+ nr = trans->blocks_used;
+ btrfs_end_transaction(trans, root);
+ btrfs_btree_balance_dirty(root, nr);
+out:
+ iput(inode);
+ return ret;
+}
+
/*
* helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY
* this function scans fs tree to find blocks reference the data extent
@@ -3217,15 +3266,27 @@ static int find_data_references(struct reloc_control
*rc,
int counted;
int ret;
- path = btrfs_alloc_path();
- if (!path)
- return -ENOMEM;
-
ref_root = btrfs_extent_data_ref_root(leaf, ref);
ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref);
ref_offset = btrfs_extent_data_ref_offset(leaf, ref);
ref_count = btrfs_extent_data_ref_count(leaf, ref);
+ /*
+ * This is an extent belonging to the free space cache, lets just delete
+ * it and redo the search.
+ */
+ if (ref_root == BTRFS_ROOT_TREE_OBJECTID) {
+ ret = delete_block_group_cache(rc->extent_root->fs_info,
+ NULL, ref_objectid);
+ if (ret != -ENOENT)
+ return ret;
+ ret = 0;
+ }
+
+ path = btrfs_alloc_path();
+ if (!path)
+ return -ENOMEM;
+
root = read_fs_root(rc->extent_root->fs_info, ref_root);
if (IS_ERR(root)) {
err = PTR_ERR(root);
@@ -3860,6 +3921,8 @@ int btrfs_relocate_block_group(struct btrfs_root
*extent_root, u64 group_start)
{
struct btrfs_fs_info *fs_info = extent_root->fs_info;
struct reloc_control *rc;
+ struct inode *inode;
+ struct btrfs_path *path;
int ret;
int rw = 0;
int err = 0;
@@ -3882,6 +3945,26 @@ int btrfs_relocate_block_group(struct btrfs_root
*extent_root, u64 group_start)
rw = 1;
}
+ path = btrfs_alloc_path();
+ if (!path) {
+ err = -ENOMEM;
+ goto out;
+ }
+
+ inode = lookup_free_space_inode(fs_info->tree_root, rc->block_group,
+ path);
+ btrfs_free_path(path);
+
+ if (!IS_ERR(inode))
+ ret = delete_block_group_cache(fs_info, inode, 0);
+ else
+ ret = PTR_ERR(inode);
+
+ if (ret && ret != -ENOENT) {
+ err = ret;
+ goto out;
+ }
+
rc->data_inode = create_reloc_inode(fs_info, rc->block_group);
if (IS_ERR(rc->data_inode)) {
err = PTR_ERR(rc->data_inode);
diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
index 859ddaa..6a625e9 100644
--- a/fs/btrfs/super.c
+++ b/fs/btrfs/super.c
@@ -68,7 +68,7 @@ enum {
Opt_nodatacow, Opt_max_inline, Opt_alloc_start, Opt_nobarrier, Opt_ssd,
Opt_nossd, Opt_ssd_spread, Opt_thread_pool, Opt_noacl, Opt_compress,
Opt_compress_force, Opt_notreelog, Opt_ratio, Opt_flushoncommit,
- Opt_discard, Opt_err,
+ Opt_discard, Opt_space_cache, Opt_err,
};
static match_table_t tokens = {
@@ -92,6 +92,7 @@ static match_table_t tokens = {
{Opt_flushoncommit, "flushoncommit"},
{Opt_ratio, "metadata_ratio=%d"},
{Opt_discard, "discard"},
+ {Opt_space_cache, "space_cache"},
{Opt_err, NULL},
};
@@ -104,6 +105,8 @@ int btrfs_parse_options(struct btrfs_root *root, char
*options)
struct btrfs_fs_info *info = root->fs_info;
substring_t args[MAX_OPT_ARGS];
char *p, *num, *orig;
+ struct btrfs_super_block *disk_super;
+ u64 features;
int intarg;
int ret = 0;
@@ -235,6 +238,12 @@ int btrfs_parse_options(struct btrfs_root *root, char
*options)
case Opt_discard:
btrfs_set_opt(info->mount_opt, DISCARD);
break;
+ case Opt_space_cache:
+ disk_super = &info->super_copy;
+ features = btrfs_super_incompat_flags(disk_super);
+ features |= BTRFS_FEATURE_INCOMPAT_SPACE_CACHE;
+ btrfs_set_super_incompat_flags(disk_super, features);
+ break;
case Opt_err:
printk(KERN_INFO "btrfs: unrecognized mount option "
"''%s''\n", p);
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 66e4c66..db0da35 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -163,6 +163,7 @@ enum btrfs_trans_type {
TRANS_START,
TRANS_JOIN,
TRANS_USERSPACE,
+ TRANS_JOIN_NOLOCK,
};
static int may_wait_transaction(struct btrfs_root *root, int type)
@@ -186,7 +187,8 @@ again:
if (!h)
return ERR_PTR(-ENOMEM);
- mutex_lock(&root->fs_info->trans_mutex);
+ if (type != TRANS_JOIN_NOLOCK)
+ mutex_lock(&root->fs_info->trans_mutex);
if (may_wait_transaction(root, type))
wait_current_trans(root);
@@ -195,7 +197,8 @@ again:
cur_trans = root->fs_info->running_transaction;
cur_trans->use_count++;
- mutex_unlock(&root->fs_info->trans_mutex);
+ if (type != TRANS_JOIN_NOLOCK)
+ mutex_unlock(&root->fs_info->trans_mutex);
h->transid = cur_trans->transid;
h->transaction = cur_trans;
@@ -224,9 +227,11 @@ again:
}
}
- mutex_lock(&root->fs_info->trans_mutex);
+ if (type != TRANS_JOIN_NOLOCK)
+ mutex_lock(&root->fs_info->trans_mutex);
record_root_in_trans(h, root);
- mutex_unlock(&root->fs_info->trans_mutex);
+ if (type != TRANS_JOIN_NOLOCK)
+ mutex_unlock(&root->fs_info->trans_mutex);
if (!current->journal_info && type != TRANS_USERSPACE)
current->journal_info = h;
@@ -244,6 +249,12 @@ struct btrfs_trans_handle *btrfs_join_transaction(struct
btrfs_root *root,
return start_transaction(root, 0, TRANS_JOIN);
}
+struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root
*root,
+ int num_blocks)
+{
+ return start_transaction(root, 0, TRANS_JOIN_NOLOCK);
+}
+
struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
int num_blocks)
{
@@ -348,7 +359,7 @@ int btrfs_should_end_transaction(struct btrfs_trans_handle
*trans,
}
static int __btrfs_end_transaction(struct btrfs_trans_handle *trans,
- struct btrfs_root *root, int throttle)
+ struct btrfs_root *root, int throttle, int lock)
{
struct btrfs_transaction *cur_trans = trans->transaction;
struct btrfs_fs_info *info = root->fs_info;
@@ -376,18 +387,19 @@ static int __btrfs_end_transaction(struct
btrfs_trans_handle *trans,
btrfs_trans_release_metadata(trans, root);
- if (!root->fs_info->open_ioctl_trans &&
+ if (lock && !root->fs_info->open_ioctl_trans &&
should_end_transaction(trans, root))
trans->transaction->blocked = 1;
- if (cur_trans->blocked && !cur_trans->in_commit) {
+ if (lock && cur_trans->blocked && !cur_trans->in_commit)
{
if (throttle)
return btrfs_commit_transaction(trans, root);
else
wake_up_process(info->transaction_kthread);
}
- mutex_lock(&info->trans_mutex);
+ if (lock)
+ mutex_lock(&info->trans_mutex);
WARN_ON(cur_trans != info->running_transaction);
WARN_ON(cur_trans->num_writers < 1);
cur_trans->num_writers--;
@@ -395,7 +407,8 @@ static int __btrfs_end_transaction(struct btrfs_trans_handle
*trans,
if (waitqueue_active(&cur_trans->writer_wait))
wake_up(&cur_trans->writer_wait);
put_transaction(cur_trans);
- mutex_unlock(&info->trans_mutex);
+ if (lock)
+ mutex_unlock(&info->trans_mutex);
if (current->journal_info == trans)
current->journal_info = NULL;
@@ -411,13 +424,19 @@ static int __btrfs_end_transaction(struct
btrfs_trans_handle *trans,
int btrfs_end_transaction(struct btrfs_trans_handle *trans,
struct btrfs_root *root)
{
- return __btrfs_end_transaction(trans, root, 0);
+ return __btrfs_end_transaction(trans, root, 0, 1);
}
int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans,
struct btrfs_root *root)
{
- return __btrfs_end_transaction(trans, root, 1);
+ return __btrfs_end_transaction(trans, root, 1, 1);
+}
+
+int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root)
+{
+ return __btrfs_end_transaction(trans, root, 0, 0);
}
/*
diff --git a/fs/btrfs/transaction.h b/fs/btrfs/transaction.h
index e104986..15f83e1 100644
--- a/fs/btrfs/transaction.h
+++ b/fs/btrfs/transaction.h
@@ -87,10 +87,14 @@ static inline void btrfs_set_inode_last_trans(struct
btrfs_trans_handle *trans,
int btrfs_end_transaction(struct btrfs_trans_handle *trans,
struct btrfs_root *root);
+int btrfs_end_transaction_nolock(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root);
struct btrfs_trans_handle *btrfs_start_transaction(struct btrfs_root *root,
int num_items);
struct btrfs_trans_handle *btrfs_join_transaction(struct btrfs_root *root,
int num_blocks);
+struct btrfs_trans_handle *btrfs_join_transaction_nolock(struct btrfs_root
*root,
+ int num_blocks);
struct btrfs_trans_handle *btrfs_start_ioctl_transaction(struct btrfs_root *r,
int num_blocks);
int btrfs_write_and_wait_transaction(struct btrfs_trans_handle *trans,
--
1.6.6.1
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs"
in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
This is a simple bit, just dump the free space cache out to our preallocated
inode when we''re writing out dirty block groups. There are a bunch of
changes
in inode.c in order to account for special cases. Mostly when we''re
doing the
writeout we''re holding trans_mutex, so we need to use the nolock
transacation
functions. Also we can''t do asynchronous completions since the async
thread
could be blocked on already completed IO waiting for the transaction lock. This
has been tested with xfstests and btrfs filesystem balance, as well as my ENOSPC
tests. Thanks,
Signed-off-by: Josef Bacik <josef@redhat.com>
---
fs/btrfs/ctree.h | 1 +
fs/btrfs/disk-io.c | 17 +++-
fs/btrfs/extent-tree.c | 48 ++++++++
fs/btrfs/free-space-cache.c | 275 +++++++++++++++++++++++++++++++++++++++++++
fs/btrfs/free-space-cache.h | 5 +
fs/btrfs/inode.c | 60 ++++++++--
6 files changed, 393 insertions(+), 13 deletions(-)
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 621bf90..1ecd8f6 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -982,6 +982,7 @@ struct btrfs_fs_info {
struct btrfs_workers endio_meta_workers;
struct btrfs_workers endio_meta_write_workers;
struct btrfs_workers endio_write_workers;
+ struct btrfs_workers endio_freespace_worker;
struct btrfs_workers submit_workers;
/*
* fixup workers take dirty pages that didn''t properly go through
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 163ee07..20d4056 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -481,9 +481,12 @@ static void end_workqueue_bio(struct bio *bio, int err)
end_io_wq->work.flags = 0;
if (bio->bi_rw & (1 << BIO_RW)) {
- if (end_io_wq->metadata)
+ if (end_io_wq->metadata == 1)
btrfs_queue_worker(&fs_info->endio_meta_write_workers,
&end_io_wq->work);
+ else if (end_io_wq->metadata == 2)
+ btrfs_queue_worker(&fs_info->endio_freespace_worker,
+ &end_io_wq->work);
else
btrfs_queue_worker(&fs_info->endio_write_workers,
&end_io_wq->work);
@@ -497,6 +500,13 @@ static void end_workqueue_bio(struct bio *bio, int err)
}
}
+/*
+ * For the metadata arg you want
+ *
+ * 0 - if data
+ * 1 - if normal metadta
+ * 2 - if writing to the free space cache area
+ */
int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
int metadata)
{
@@ -1774,6 +1784,8 @@ struct btrfs_root *open_ctree(struct super_block *sb,
btrfs_init_workers(&fs_info->endio_write_workers,
"endio-write",
fs_info->thread_pool_size,
&fs_info->generic_worker);
+ btrfs_init_workers(&fs_info->endio_freespace_worker,
"freespace-write",
+ 1, &fs_info->generic_worker);
/*
* endios are largely parallel and should have a very
@@ -1794,6 +1806,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
btrfs_start_workers(&fs_info->endio_meta_workers, 1);
btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
btrfs_start_workers(&fs_info->endio_write_workers, 1);
+ btrfs_start_workers(&fs_info->endio_freespace_worker, 1);
fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
@@ -2035,6 +2048,7 @@ fail_sb_buffer:
btrfs_stop_workers(&fs_info->endio_meta_workers);
btrfs_stop_workers(&fs_info->endio_meta_write_workers);
btrfs_stop_workers(&fs_info->endio_write_workers);
+ btrfs_stop_workers(&fs_info->endio_freespace_worker);
btrfs_stop_workers(&fs_info->submit_workers);
fail_iput:
invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
@@ -2468,6 +2482,7 @@ int close_ctree(struct btrfs_root *root)
btrfs_stop_workers(&fs_info->endio_meta_workers);
btrfs_stop_workers(&fs_info->endio_meta_write_workers);
btrfs_stop_workers(&fs_info->endio_write_workers);
+ btrfs_stop_workers(&fs_info->endio_freespace_worker);
btrfs_stop_workers(&fs_info->submit_workers);
btrfs_close_devices(fs_info->fs_devices);
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index b8151ad..6994423 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -2847,6 +2847,8 @@ again:
continue;
}
+ if (cache->disk_cache_state == BTRFS_DC_SETUP)
+ cache->disk_cache_state = BTRFS_DC_NEED_WRITE;
cache->dirty = 0;
last = cache->key.objectid + cache->key.offset;
@@ -2855,6 +2857,52 @@ again:
btrfs_put_block_group(cache);
}
+ while (1) {
+ /*
+ * I don''t think this is needed since we''re just marking
our
+ * preallocated extent as written, but just in case it can''t
+ * hurt.
+ */
+ if (last == 0) {
+ err = btrfs_run_delayed_refs(trans, root,
+ (unsigned long)-1);
+ BUG_ON(err);
+ }
+
+ cache = btrfs_lookup_first_block_group(root->fs_info, last);
+ while (cache) {
+ /*
+ * Really this shouldn''t happen, but it could if we
+ * couldn''t write the entire preallocated extent and
+ * splitting the extent resulted in a new block.
+ */
+ if (cache->dirty) {
+ btrfs_put_block_group(cache);
+ goto again;
+ }
+ if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
+ break;
+ cache = next_block_group(root, cache);
+ }
+ if (!cache) {
+ if (last == 0)
+ break;
+ last = 0;
+ continue;
+ }
+
+ btrfs_write_out_cache(root, trans, cache, path);
+
+ /*
+ * If we didn''t have an error then the cache state is still
+ * NEED_WRITE, so we can set it to WRITTEN.
+ */
+ if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
+ cache->disk_cache_state = BTRFS_DC_WRITTEN;
+ last = cache->key.objectid + cache->key.offset;
+ btrfs_put_block_group(cache);
+ }
+
btrfs_free_path(path);
return 0;
}
diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
index 05efcc7..8fe0aee 100644
--- a/fs/btrfs/free-space-cache.c
+++ b/fs/btrfs/free-space-cache.c
@@ -28,6 +28,11 @@
#define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8)
#define MAX_CACHE_BYTES_PER_GIG (32 * 1024)
+static void recalculate_thresholds(struct btrfs_block_group_cache
+ *block_group);
+static int link_free_space(struct btrfs_block_group_cache *block_group,
+ struct btrfs_free_space *info);
+
struct inode *lookup_free_space_inode(struct btrfs_root *root,
struct btrfs_block_group_cache
*block_group, struct btrfs_path *path)
@@ -182,6 +187,276 @@ int btrfs_truncate_free_space_cache(struct btrfs_root
*root,
return btrfs_update_inode(trans, root, inode);
}
+int btrfs_write_out_cache(struct btrfs_root *root,
+ struct btrfs_trans_handle *trans,
+ struct btrfs_block_group_cache *block_group,
+ struct btrfs_path *path)
+{
+ struct btrfs_free_space_header *header;
+ struct extent_buffer *leaf;
+ struct inode *inode;
+ struct rb_node *node;
+ struct list_head *pos, *n;
+ struct list_head bitmap_list;
+ struct btrfs_key key;
+ u64 bytes = 0;
+ u32 *crc, *checksums;
+ pgoff_t index = 0, last_index = 0;
+ unsigned long first_page_offset;
+ int num_checksums;
+ int entries = 0;
+ int bitmaps = 0;
+ int ret = 0;
+
+ root = root->fs_info->tree_root;
+
+ INIT_LIST_HEAD(&bitmap_list);
+
+ spin_lock(&block_group->lock);
+ if (block_group->disk_cache_state < BTRFS_DC_SETUP) {
+ spin_unlock(&block_group->lock);
+ return 0;
+ }
+ spin_unlock(&block_group->lock);
+
+ inode = lookup_free_space_inode(root, block_group, path);
+ if (IS_ERR(inode))
+ return 0;
+
+ if (!i_size_read(inode)) {
+ iput(inode);
+ return 0;
+ }
+
+ last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
+ filemap_write_and_wait(inode->i_mapping);
+ btrfs_wait_ordered_range(inode, inode->i_size &
+ ~(root->sectorsize - 1), (u64)-1);
+
+ /* We need a checksum per page. */
+ num_checksums = i_size_read(inode) / PAGE_CACHE_SIZE;
+ crc = checksums = kzalloc(sizeof(u32) * num_checksums, GFP_NOFS);
+ if (!crc) {
+ iput(inode);
+ return 0;
+ }
+
+ /* Since the first page has all of our checksums and our generation we
+ * need to calculate the offset into the page that we can start writing
+ * our entries.
+ */
+ first_page_offset = (sizeof(u32) * num_checksums) + sizeof(u64);
+
+ node = rb_first(&block_group->free_space_offset);
+ if (!node)
+ goto out_free;
+
+ do {
+ struct btrfs_free_space_entry *entry;
+ void *addr;
+ struct page *page;
+ unsigned long offset = 0;
+ unsigned long start_offset = 0;
+
+ if (index == 0) {
+ start_offset = first_page_offset;
+ offset = start_offset;
+ }
+
+ page = grab_cache_page(inode->i_mapping, index);
+ if (!page)
+ goto out_free;
+
+ addr = kmap(page);
+ entry = addr + start_offset;
+
+ memset(addr, 0, PAGE_CACHE_SIZE);
+ while (1) {
+ struct btrfs_free_space *e;
+
+ e = rb_entry(node, struct btrfs_free_space, offset_index);
+ entries++;
+
+ entry->offset = cpu_to_le64(e->offset);
+ entry->bytes = cpu_to_le64(e->bytes);
+ if (e->bitmap) {
+ entry->type = BTRFS_FREE_SPACE_BITMAP;
+ list_add_tail(&e->list, &bitmap_list);
+ bitmaps++;
+ } else {
+ entry->type = BTRFS_FREE_SPACE_EXTENT;
+ }
+ node = rb_next(node);
+ if (!node)
+ break;
+ offset += sizeof(struct btrfs_free_space_entry);
+ if (offset + sizeof(struct btrfs_free_space_entry) >+
PAGE_CACHE_SIZE)
+ break;
+ entry++;
+ }
+ *crc = ~(u32)0;
+ *crc = btrfs_csum_data(root, addr + start_offset, *crc,
+ PAGE_CACHE_SIZE - start_offset);
+ kunmap(page);
+
+ btrfs_csum_final(*crc, (char *)crc);
+ crc++;
+
+ bytes += PAGE_CACHE_SIZE;
+
+ ClearPageChecked(page);
+ set_page_extent_mapped(page);
+ SetPageUptodate(page);
+ set_page_dirty(page);
+ unlock_page(page);
+ page_cache_release(page);
+ index++;
+ } while (node);
+
+ list_for_each_safe(pos, n, &bitmap_list) {
+ struct page *page;
+ void *addr;
+ struct btrfs_free_space *entry + list_entry(pos, struct btrfs_free_space,
list);
+
+ page = grab_cache_page(inode->i_mapping, index);
+ if (!page)
+ goto out_free;
+
+ addr = kmap(page);
+ memcpy(addr, entry->bitmap, PAGE_CACHE_SIZE);
+ *crc = ~(u32)0;
+ *crc = btrfs_csum_data(root, addr, *crc, PAGE_CACHE_SIZE);
+ kunmap(page);
+ btrfs_csum_final(*crc, (char *)crc);
+ crc++;
+ bytes += PAGE_CACHE_SIZE;
+
+ ClearPageChecked(page);
+ set_page_extent_mapped(page);
+ SetPageUptodate(page);
+ set_page_dirty(page);
+ unlock_page(page);
+ page_cache_release(page);
+ list_del_init(&entry->list);
+ index++;
+ }
+
+ while (index <= last_index) {
+ struct page *page;
+ void *addr;
+
+ page = grab_cache_page(inode->i_mapping, index);
+ if (!page) {
+ /* Since this is just an optimization so we don''t split
+ * the preallocated extent, don''t freak-out, just
+ * break.
+ */
+ break;
+ }
+ addr = kmap(page);
+ memset(addr, 0, PAGE_CACHE_SIZE);
+ kunmap(page);
+ ClearPageChecked(page);
+ set_page_extent_mapped(page);
+ SetPageUptodate(page);
+ set_page_dirty(page);
+ unlock_page(page);
+ page_cache_release(page);
+ bytes += PAGE_CACHE_SIZE;
+ index++;
+ }
+
+ btrfs_set_extent_delalloc(inode, 0, bytes - 1, NULL);
+
+ {
+ struct page *page;
+ void *addr;
+ u64 *gen;
+
+ page = grab_cache_page(inode->i_mapping, 0);
+ if (!page)
+ goto out_free;
+
+ if (!PageUptodate(page)) {
+ btrfs_readpage(NULL, page);
+ lock_page(page);
+ if (!PageUptodate(page)) {
+ unlock_page(page);
+ page_cache_release(page);
+ goto out_free;
+ }
+ }
+
+ addr = kmap(page);
+ memcpy(addr, checksums, sizeof(u32) * num_checksums);
+ gen = addr + (sizeof(u32) * num_checksums);
+ *gen = trans->transid;
+ kunmap(page);
+ ClearPageChecked(page);
+ set_page_extent_mapped(page);
+ SetPageUptodate(page);
+ set_page_dirty(page);
+ unlock_page(page);
+ page_cache_release(page);
+ }
+ BTRFS_I(inode)->generation = trans->transid;
+ filemap_write_and_wait(inode->i_mapping);
+
+ key.objectid = BTRFS_FREE_SPACE_OBJECTID;
+ key.offset = block_group->key.objectid;
+ key.type = 0;
+
+ ret = btrfs_search_slot(trans, root, &key, path, 1, 1);
+ if (ret < 0) {
+ ret = 0;
+ clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1,
+ EXTENT_DIRTY | EXTENT_DELALLOC |
+ EXTENT_DO_ACCOUNTING, 0, 0, NULL, GFP_NOFS);
+ goto out_free;
+ }
+ leaf = path->nodes[0];
+ if (ret > 0) {
+ struct btrfs_key found_key;
+ BUG_ON(!path->slots[0]);
+ path->slots[0]--;
+ btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
+ if (found_key.objectid != BTRFS_FREE_SPACE_OBJECTID ||
+ found_key.offset != block_group->key.objectid) {
+ ret = 0;
+ clear_extent_bit(&BTRFS_I(inode)->io_tree, 0, bytes - 1,
+ EXTENT_DIRTY | EXTENT_DELALLOC |
+ EXTENT_DO_ACCOUNTING, 0, 0, NULL,
+ GFP_NOFS);
+ btrfs_release_path(root, path);
+ goto out_free;
+ }
+ }
+ header = btrfs_item_ptr(leaf, path->slots[0],
+ struct btrfs_free_space_header);
+ btrfs_set_free_space_entries(leaf, header, entries);
+ btrfs_set_free_space_bitmaps(leaf, header, bitmaps);
+ btrfs_set_free_space_generation(leaf, header, trans->transid);
+ btrfs_mark_buffer_dirty(leaf);
+ btrfs_release_path(root, path);
+
+ ret = 1;
+
+out_free:
+ if (ret == 0) {
+ invalidate_inode_pages2_range(inode->i_mapping, 0, index);
+ spin_lock(&block_group->lock);
+ block_group->disk_cache_state = BTRFS_DC_ERROR;
+ spin_unlock(&block_group->lock);
+ BTRFS_I(inode)->generation = 0;
+ }
+ kfree(checksums);
+ btrfs_update_inode(trans, root, inode);
+ iput(inode);
+ return ret;
+}
+
static inline unsigned long offset_to_bit(u64 bitmap_start, u64 sectorsize,
u64 offset)
{
diff --git a/fs/btrfs/free-space-cache.h b/fs/btrfs/free-space-cache.h
index 45be29e..189f740 100644
--- a/fs/btrfs/free-space-cache.h
+++ b/fs/btrfs/free-space-cache.h
@@ -34,10 +34,15 @@ int create_free_space_inode(struct btrfs_root *root,
struct btrfs_trans_handle *trans,
struct btrfs_block_group_cache *block_group,
struct btrfs_path *path);
+
int btrfs_truncate_free_space_cache(struct btrfs_root *root,
struct btrfs_trans_handle *trans,
struct btrfs_path *path,
struct inode *inode);
+int btrfs_write_out_cache(struct btrfs_root *root,
+ struct btrfs_trans_handle *trans,
+ struct btrfs_block_group_cache *block_group,
+ struct btrfs_path *path);
int btrfs_add_free_space(struct btrfs_block_group_cache *block_group,
u64 bytenr, u64 size);
int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 16b708a..1d17518 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -764,6 +764,7 @@ static noinline int cow_file_range(struct inode *inode,
struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
int ret = 0;
+ BUG_ON(root == root->fs_info->tree_root);
trans = btrfs_join_transaction(root, 1);
BUG_ON(!trans);
btrfs_set_trans_block_group(trans, inode);
@@ -1035,10 +1036,16 @@ static noinline int run_delalloc_nocow(struct inode
*inode,
int type;
int nocow;
int check_prev = 1;
+ bool nolock = false;
path = btrfs_alloc_path();
BUG_ON(!path);
- trans = btrfs_join_transaction(root, 1);
+ if (root == root->fs_info->tree_root) {
+ nolock = true;
+ trans = btrfs_join_transaction_nolock(root, 1);
+ } else {
+ trans = btrfs_join_transaction(root, 1);
+ }
BUG_ON(!trans);
cow_start = (u64)-1;
@@ -1211,8 +1218,13 @@ out_check:
BUG_ON(ret);
}
- ret = btrfs_end_transaction(trans, root);
- BUG_ON(ret);
+ if (nolock) {
+ ret = btrfs_end_transaction_nolock(trans, root);
+ BUG_ON(ret);
+ } else {
+ ret = btrfs_end_transaction(trans, root);
+ BUG_ON(ret);
+ }
btrfs_free_path(path);
return 0;
}
@@ -1289,6 +1301,8 @@ static int btrfs_set_bit_hook(struct inode *inode,
if (!(state->state & EXTENT_DELALLOC) && (*bits &
EXTENT_DELALLOC)) {
struct btrfs_root *root = BTRFS_I(inode)->root;
u64 len = state->end + 1 - state->start;
+ int do_list = (root->root_key.objectid !+
BTRFS_ROOT_TREE_OBJECTID);
if (*bits & EXTENT_FIRST_DELALLOC)
*bits &= ~EXTENT_FIRST_DELALLOC;
@@ -1298,7 +1312,7 @@ static int btrfs_set_bit_hook(struct inode *inode,
spin_lock(&root->fs_info->delalloc_lock);
BTRFS_I(inode)->delalloc_bytes += len;
root->fs_info->delalloc_bytes += len;
- if (list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
+ if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
&root->fs_info->delalloc_inodes);
}
@@ -1321,6 +1335,8 @@ static int btrfs_clear_bit_hook(struct inode *inode,
if ((state->state & EXTENT_DELALLOC) && (*bits &
EXTENT_DELALLOC)) {
struct btrfs_root *root = BTRFS_I(inode)->root;
u64 len = state->end + 1 - state->start;
+ int do_list = (root->root_key.objectid !+
BTRFS_ROOT_TREE_OBJECTID);
if (*bits & EXTENT_FIRST_DELALLOC)
*bits &= ~EXTENT_FIRST_DELALLOC;
@@ -1330,14 +1346,15 @@ static int btrfs_clear_bit_hook(struct inode *inode,
if (*bits & EXTENT_DO_ACCOUNTING)
btrfs_delalloc_release_metadata(inode, len);
- if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID)
+ if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
+ && do_list)
btrfs_free_reserved_data_space(inode, len);
spin_lock(&root->fs_info->delalloc_lock);
root->fs_info->delalloc_bytes -= len;
BTRFS_I(inode)->delalloc_bytes -= len;
- if (BTRFS_I(inode)->delalloc_bytes == 0 &&
+ if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
!list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
list_del_init(&BTRFS_I(inode)->delalloc_inodes);
}
@@ -1426,7 +1443,10 @@ static int btrfs_submit_bio_hook(struct inode *inode, int
rw, struct bio *bio,
skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
- ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
+ if (root == root->fs_info->tree_root)
+ ret = btrfs_bio_wq_end_io(root->fs_info, bio, 2);
+ else
+ ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
BUG_ON(ret);
if (!(rw & (1 << BIO_RW))) {
@@ -1662,6 +1682,7 @@ static int btrfs_finish_ordered_io(struct inode *inode,
u64 start, u64 end)
struct extent_state *cached_state = NULL;
int compressed = 0;
int ret;
+ bool nolock = false;
ret = btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
end - start + 1);
@@ -1669,11 +1690,17 @@ static int btrfs_finish_ordered_io(struct inode *inode,
u64 start, u64 end)
return 0;
BUG_ON(!ordered_extent);
+ nolock = (root == root->fs_info->tree_root);
+
if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
BUG_ON(!list_empty(&ordered_extent->list));
ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
if (!ret) {
- trans = btrfs_join_transaction(root, 1);
+ if (nolock)
+ trans = btrfs_join_transaction_nolock(root, 1);
+ else
+ trans = btrfs_join_transaction(root, 1);
+ BUG_ON(!trans);
btrfs_set_trans_block_group(trans, inode);
trans->block_rsv = &root->fs_info->delalloc_block_rsv;
ret = btrfs_update_inode(trans, root, inode);
@@ -1686,7 +1713,10 @@ static int btrfs_finish_ordered_io(struct inode *inode,
u64 start, u64 end)
ordered_extent->file_offset + ordered_extent->len - 1,
0, &cached_state, GFP_NOFS);
- trans = btrfs_join_transaction(root, 1);
+ if (nolock)
+ trans = btrfs_join_transaction_nolock(root, 1);
+ else
+ trans = btrfs_join_transaction(root, 1);
btrfs_set_trans_block_group(trans, inode);
trans->block_rsv = &root->fs_info->delalloc_block_rsv;
@@ -1725,9 +1755,15 @@ static int btrfs_finish_ordered_io(struct inode *inode,
u64 start, u64 end)
ret = btrfs_update_inode(trans, root, inode);
BUG_ON(ret);
out:
- btrfs_delalloc_release_metadata(inode, ordered_extent->len);
- if (trans)
- btrfs_end_transaction(trans, root);
+ if (nolock) {
+ if (trans)
+ btrfs_end_transaction_nolock(trans, root);
+ } else {
+ btrfs_delalloc_release_metadata(inode, ordered_extent->len);
+ if (trans)
+ btrfs_end_transaction(trans, root);
+ }
+
/* once for us */
btrfs_put_ordered_extent(ordered_extent);
/* once for the tree */
--
1.6.6.1
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs"
in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
This patch actually loads the free space cache if it exists. The only thing
that really changes here is that we need to cache the block group if
we''re going
to remove an extent from it. Previously we did not do this since the caching
kthread would pick it up. With the on disk cache we don''t have this
luxury so
we need to make sure we read the on disk cache in first, and then remove the
extent, that way when the extent is unpinned the free space is added to the
block group. This has been tested with all sorts of things.
Signed-off-by: Josef Bacik <josef@redhat.com>
---
fs/btrfs/extent-tree.c | 50 +++++++-
fs/btrfs/free-space-cache.c | 296 +++++++++++++++++++++++++++++++++++++++++++
fs/btrfs/free-space-cache.h | 2 +
3 files changed, 345 insertions(+), 3 deletions(-)
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 6994423..8db9330 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -421,7 +421,9 @@ err:
return 0;
}
-static int cache_block_group(struct btrfs_block_group_cache *cache)
+static int cache_block_group(struct btrfs_block_group_cache *cache,
+ struct btrfs_trans_handle *trans,
+ int load_cache_only)
{
struct btrfs_fs_info *fs_info = cache->fs_info;
struct btrfs_caching_control *caching_ctl;
@@ -432,6 +434,36 @@ static int cache_block_group(struct btrfs_block_group_cache
*cache)
if (cache->cached != BTRFS_CACHE_NO)
return 0;
+ /*
+ * We can''t do the read from on-disk cache during a commit since we
need
+ * to have the normal tree locking.
+ */
+ if (!trans->transaction->in_commit) {
+ spin_lock(&cache->lock);
+ if (cache->cached != BTRFS_CACHE_NO) {
+ spin_unlock(&cache->lock);
+ return 0;
+ }
+ cache->cached = BTRFS_CACHE_STARTED;
+ spin_unlock(&cache->lock);
+
+ ret = load_free_space_cache(fs_info, cache);
+
+ spin_lock(&cache->lock);
+ if (ret == 1) {
+ cache->cached = BTRFS_CACHE_FINISHED;
+ cache->last_byte_to_unpin = (u64)-1;
+ } else {
+ cache->cached = BTRFS_CACHE_NO;
+ }
+ spin_unlock(&cache->lock);
+ if (ret == 1)
+ return 0;
+ }
+
+ if (load_cache_only)
+ return 0;
+
caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_KERNEL);
BUG_ON(!caching_ctl);
@@ -3985,6 +4017,14 @@ static int update_block_group(struct btrfs_trans_handle
*trans,
factor = 2;
else
factor = 1;
+ /*
+ * If this block group has free space cache written out, we
+ * need to make sure to load it if we are removing space. This
+ * is because we need the unpinning stage to actually add the
+ * space back to the block group, otherwise we will leak space.
+ */
+ if (!alloc && cache->cached == BTRFS_CACHE_NO)
+ cache_block_group(cache, trans, 1);
byte_in_group = bytenr - cache->key.objectid;
WARN_ON(byte_in_group > cache->key.offset);
@@ -4829,6 +4869,10 @@ have_block_group:
if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
u64 free_percent;
+ ret = cache_block_group(block_group, trans, 1);
+ if (block_group->cached == BTRFS_CACHE_FINISHED)
+ goto have_block_group;
+
free_percent = btrfs_block_group_used(&block_group->item);
free_percent *= 100;
free_percent = div64_u64(free_percent,
@@ -4849,7 +4893,7 @@ have_block_group:
if (loop > LOOP_CACHING_NOWAIT ||
(loop > LOOP_FIND_IDEAL &&
atomic_read(&space_info->caching_threads) < 2)) {
- ret = cache_block_group(block_group);
+ ret = cache_block_group(block_group, trans, 0);
BUG_ON(ret);
}
found_uncached_bg = true;
@@ -5406,7 +5450,7 @@ int btrfs_alloc_logged_file_extent(struct
btrfs_trans_handle *trans,
u64 num_bytes = ins->offset;
block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
- cache_block_group(block_group);
+ cache_block_group(block_group, trans, 0);
caching_ctl = get_caching_control(block_group);
if (!caching_ctl) {
diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
index 8fe0aee..c2e8c71 100644
--- a/fs/btrfs/free-space-cache.c
+++ b/fs/btrfs/free-space-cache.c
@@ -187,6 +187,302 @@ int btrfs_truncate_free_space_cache(struct btrfs_root
*root,
return btrfs_update_inode(trans, root, inode);
}
+static int readahead_cache(struct inode *inode)
+{
+ struct file_ra_state *ra;
+ unsigned long last_index;
+
+ ra = kzalloc(sizeof(*ra), GFP_NOFS);
+ if (!ra)
+ return -ENOMEM;
+
+ file_ra_state_init(ra, inode->i_mapping);
+ last_index = (i_size_read(inode) - 1) >> PAGE_CACHE_SHIFT;
+
+ page_cache_sync_readahead(inode->i_mapping, ra, NULL, 0, last_index);
+
+ kfree(ra);
+
+ return 0;
+}
+
+int load_free_space_cache(struct btrfs_fs_info *fs_info,
+ struct btrfs_block_group_cache *block_group)
+{
+ struct btrfs_root *root = fs_info->tree_root;
+ struct inode *inode;
+ struct btrfs_free_space_header *header;
+ struct extent_buffer *leaf;
+ struct page *page;
+ struct btrfs_path *path;
+ u32 *checksums = NULL, *crc;
+ char *disk_crcs = NULL;
+ struct btrfs_key key;
+ struct list_head bitmaps;
+ u64 num_entries;
+ u64 num_bitmaps;
+ u64 generation;
+ u32 cur_crc = ~(u32)0;
+ pgoff_t index = 0;
+ unsigned long first_page_offset;
+ int num_checksums;
+ int ret = 0;
+
+ /*
+ * If we''re unmounting then just return, since this does a search on
the
+ * normal root and not the commit root and we could deadlock.
+ */
+ smp_mb();
+ if (fs_info->closing)
+ return 0;
+
+ /*
+ * If this block group has been marked to be cleared for one reason or
+ * another then we can''t trust the on disk cache, so just return.
+ */
+ spin_lock(&block_group->lock);
+ if (block_group->disk_cache_state != BTRFS_DC_WRITTEN) {
+ printk(KERN_ERR "not reading block group %llu, dcs is %d\n",
block_group->key.objectid,
+ block_group->disk_cache_state);
+ spin_unlock(&block_group->lock);
+ return 0;
+ }
+ spin_unlock(&block_group->lock);
+
+ INIT_LIST_HEAD(&bitmaps);
+
+ path = btrfs_alloc_path();
+ if (!path)
+ return 0;
+
+ inode = lookup_free_space_inode(root, block_group, path);
+ if (IS_ERR(inode)) {
+ btrfs_free_path(path);
+ return 0;
+ }
+
+ /* Nothing in the space cache, goodbye */
+ if (!i_size_read(inode)) {
+ btrfs_free_path(path);
+ goto out;
+ }
+
+ key.objectid = BTRFS_FREE_SPACE_OBJECTID;
+ key.offset = block_group->key.objectid;
+ key.type = 0;
+
+ ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+ if (ret) {
+ btrfs_free_path(path);
+ goto out;
+ }
+
+ leaf = path->nodes[0];
+ header = btrfs_item_ptr(leaf, path->slots[0],
+ struct btrfs_free_space_header);
+ num_entries = btrfs_free_space_entries(leaf, header);
+ num_bitmaps = btrfs_free_space_bitmaps(leaf, header);
+ generation = btrfs_free_space_generation(leaf, header);
+ btrfs_free_path(path);
+
+ if (BTRFS_I(inode)->generation != generation) {
+ printk(KERN_ERR "btrfs: free space inode generation (%llu) did"
+ " not match free space cache generation (%llu) for "
+ "block group %llu\n",
+ (unsigned long long)BTRFS_I(inode)->generation,
+ (unsigned long long)generation,
+ (unsigned long long)block_group->key.objectid);
+ goto out;
+ }
+
+ if (!num_entries)
+ goto out;
+
+ /* Setup everything for doing checksumming */
+ num_checksums = i_size_read(inode) / PAGE_CACHE_SIZE;
+ checksums = crc = kzalloc(sizeof(u32) * num_checksums, GFP_NOFS);
+ if (!checksums)
+ goto out;
+ first_page_offset = (sizeof(u32) * num_checksums) + sizeof(u64);
+ disk_crcs = kzalloc(first_page_offset, GFP_NOFS);
+ if (!disk_crcs)
+ goto out;
+
+ ret = readahead_cache(inode);
+ if (ret) {
+ ret = 0;
+ goto out;
+ }
+
+ while (1) {
+ struct btrfs_free_space_entry *entry;
+ struct btrfs_free_space *e;
+ void *addr;
+ unsigned long offset = 0;
+ unsigned long start_offset = 0;
+ int need_loop = 0;
+
+ if (!num_entries && !num_bitmaps)
+ break;
+
+ if (index == 0) {
+ start_offset = first_page_offset;
+ offset = start_offset;
+ }
+
+ page = grab_cache_page(inode->i_mapping, index);
+ if (!page) {
+ ret = 0;
+ goto free_cache;
+ }
+
+ if (!PageUptodate(page)) {
+ btrfs_readpage(NULL, page);
+ lock_page(page);
+ if (!PageUptodate(page)) {
+ unlock_page(page);
+ page_cache_release(page);
+ printk(KERN_ERR "btrfs: error reading free "
+ "space cache: %llu\n",
+ (unsigned long long)
+ block_group->key.objectid);
+ goto free_cache;
+ }
+ }
+ addr = kmap(page);
+
+ if (index == 0) {
+ u64 *gen;
+
+ memcpy(disk_crcs, addr, first_page_offset);
+ gen = addr + (sizeof(u32) * num_checksums);
+ if (*gen != BTRFS_I(inode)->generation) {
+ printk(KERN_ERR "btrfs: space cache generation"
+ " (%llu) does not match inode (%llu) "
+ "for block group %llu\n",
+ (unsigned long long)*gen,
+ (unsigned long long)
+ BTRFS_I(inode)->generation,
+ (unsigned long long)
+ block_group->key.objectid);
+ kunmap(page);
+ unlock_page(page);
+ page_cache_release(page);
+ goto free_cache;
+ }
+ crc = (u32 *)disk_crcs;
+ }
+ entry = addr + start_offset;
+
+ /* First lets check our crc before we do anything fun */
+ cur_crc = ~(u32)0;
+ cur_crc = btrfs_csum_data(root, addr + start_offset, cur_crc,
+ PAGE_CACHE_SIZE - start_offset);
+ btrfs_csum_final(cur_crc, (char *)&cur_crc);
+ if (cur_crc != *crc) {
+ printk(KERN_ERR "btrfs: crc mismatch for page %lu in "
+ "block group %llu\n", index,
+ (unsigned long long)block_group->key.objectid);
+ kunmap(page);
+ unlock_page(page);
+ page_cache_release(page);
+ goto free_cache;
+ }
+ crc++;
+
+ while (1) {
+ if (!num_entries)
+ break;
+
+ need_loop = 1;
+ e = kzalloc(sizeof(struct btrfs_free_space), GFP_NOFS);
+ if (!e) {
+ kunmap(page);
+ unlock_page(page);
+ page_cache_release(page);
+ goto free_cache;
+ }
+
+ e->offset = le64_to_cpu(entry->offset);
+ e->bytes = le64_to_cpu(entry->bytes);
+ if (!e->bytes) {
+ kunmap(page);
+ kfree(e);
+ unlock_page(page);
+ page_cache_release(page);
+ goto free_cache;
+ }
+
+ if (entry->type == BTRFS_FREE_SPACE_EXTENT) {
+ spin_lock(&block_group->tree_lock);
+ ret = link_free_space(block_group, e);
+ spin_unlock(&block_group->tree_lock);
+ BUG_ON(ret);
+ } else {
+ e->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
+ if (!e->bitmap) {
+ kunmap(page);
+ kfree(e);
+ unlock_page(page);
+ page_cache_release(page);
+ goto free_cache;
+ }
+ spin_lock(&block_group->tree_lock);
+ ret = link_free_space(block_group, e);
+ block_group->total_bitmaps++;
+ recalculate_thresholds(block_group);
+ spin_unlock(&block_group->tree_lock);
+ list_add_tail(&e->list, &bitmaps);
+ }
+
+ num_entries--;
+ offset += sizeof(struct btrfs_free_space_entry);
+ if (offset + sizeof(struct btrfs_free_space_entry) >+
PAGE_CACHE_SIZE)
+ break;
+ entry++;
+ }
+
+ /*
+ * We read an entry out of this page, we need to move on to the
+ * next page.
+ */
+ if (need_loop) {
+ kunmap(page);
+ goto next;
+ }
+
+ /*
+ * We add the bitmaps at the end of the entries in order that
+ * the bitmap entries are added to the cache.
+ */
+ e = list_entry(bitmaps.next, struct btrfs_free_space, list);
+ list_del_init(&e->list);
+ memcpy(e->bitmap, addr, PAGE_CACHE_SIZE);
+ kunmap(page);
+ num_bitmaps--;
+next:
+ unlock_page(page);
+ page_cache_release(page);
+ index++;
+ }
+
+ ret = 1;
+out:
+ kfree(checksums);
+ kfree(disk_crcs);
+ iput(inode);
+ return ret;
+
+free_cache:
+ /* This cache is bogus, make sure it gets cleared */
+ spin_lock(&block_group->lock);
+ block_group->disk_cache_state = BTRFS_DC_CLEAR;
+ spin_unlock(&block_group->lock);
+ btrfs_remove_free_space_cache(block_group);
+ goto out;
+}
+
int btrfs_write_out_cache(struct btrfs_root *root,
struct btrfs_trans_handle *trans,
struct btrfs_block_group_cache *block_group,
diff --git a/fs/btrfs/free-space-cache.h b/fs/btrfs/free-space-cache.h
index 189f740..e49ca5c 100644
--- a/fs/btrfs/free-space-cache.h
+++ b/fs/btrfs/free-space-cache.h
@@ -39,6 +39,8 @@ int btrfs_truncate_free_space_cache(struct btrfs_root *root,
struct btrfs_trans_handle *trans,
struct btrfs_path *path,
struct inode *inode);
+int load_free_space_cache(struct btrfs_fs_info *fs_info,
+ struct btrfs_block_group_cache *block_group);
int btrfs_write_out_cache(struct btrfs_root *root,
struct btrfs_trans_handle *trans,
struct btrfs_block_group_cache *block_group,
--
1.6.6.1
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs"
in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html