summaryrefslogtreecommitdiffstats
path: root/fs/btrfs
diff options
context:
space:
mode:
Diffstat (limited to 'fs/btrfs')
-rw-r--r--fs/btrfs/acl.c44
-rw-r--r--fs/btrfs/async-thread.c6
-rw-r--r--fs/btrfs/btrfs_inode.h4
-rw-r--r--fs/btrfs/compression.c1
-rw-r--r--fs/btrfs/ctree.c121
-rw-r--r--fs/btrfs/ctree.h32
-rw-r--r--fs/btrfs/disk-io.c15
-rw-r--r--fs/btrfs/extent-tree.c1096
-rw-r--r--fs/btrfs/file.c6
-rw-r--r--fs/btrfs/free-space-cache.c1058
-rw-r--r--fs/btrfs/free-space-cache.h8
-rw-r--r--fs/btrfs/inode.c49
-rw-r--r--fs/btrfs/ioctl.c7
-rw-r--r--fs/btrfs/print-tree.c6
-rw-r--r--fs/btrfs/relocation.c17
-rw-r--r--fs/btrfs/super.c1
-rw-r--r--fs/btrfs/transaction.c60
-rw-r--r--fs/btrfs/transaction.h1
-rw-r--r--fs/btrfs/tree-log.c2
-rw-r--r--fs/btrfs/volumes.c46
-rw-r--r--fs/btrfs/zlib.c6
21 files changed, 1872 insertions, 714 deletions
diff --git a/fs/btrfs/acl.c b/fs/btrfs/acl.c
index 603972576f0..f128427b995 100644
--- a/fs/btrfs/acl.c
+++ b/fs/btrfs/acl.c
@@ -29,51 +29,28 @@
#ifdef CONFIG_FS_POSIX_ACL
-static void btrfs_update_cached_acl(struct inode *inode,
- struct posix_acl **p_acl,
- struct posix_acl *acl)
-{
- spin_lock(&inode->i_lock);
- if (*p_acl && *p_acl != BTRFS_ACL_NOT_CACHED)
- posix_acl_release(*p_acl);
- *p_acl = posix_acl_dup(acl);
- spin_unlock(&inode->i_lock);
-}
-
static struct posix_acl *btrfs_get_acl(struct inode *inode, int type)
{
int size;
const char *name;
char *value = NULL;
- struct posix_acl *acl = NULL, **p_acl;
+ struct posix_acl *acl;
+
+ acl = get_cached_acl(inode, type);
+ if (acl != ACL_NOT_CACHED)
+ return acl;
switch (type) {
case ACL_TYPE_ACCESS:
name = POSIX_ACL_XATTR_ACCESS;
- p_acl = &BTRFS_I(inode)->i_acl;
break;
case ACL_TYPE_DEFAULT:
name = POSIX_ACL_XATTR_DEFAULT;
- p_acl = &BTRFS_I(inode)->i_default_acl;
break;
default:
- return ERR_PTR(-EINVAL);
+ BUG();
}
- /* Handle the cached NULL acl case without locking */
- acl = ACCESS_ONCE(*p_acl);
- if (!acl)
- return acl;
-
- spin_lock(&inode->i_lock);
- acl = *p_acl;
- if (acl != BTRFS_ACL_NOT_CACHED)
- acl = posix_acl_dup(acl);
- spin_unlock(&inode->i_lock);
-
- if (acl != BTRFS_ACL_NOT_CACHED)
- return acl;
-
size = __btrfs_getxattr(inode, name, "", 0);
if (size > 0) {
value = kzalloc(size, GFP_NOFS);
@@ -82,13 +59,13 @@ static struct posix_acl *btrfs_get_acl(struct inode *inode, int type)
size = __btrfs_getxattr(inode, name, value, size);
if (size > 0) {
acl = posix_acl_from_xattr(value, size);
- btrfs_update_cached_acl(inode, p_acl, acl);
+ set_cached_acl(inode, type, acl);
}
kfree(value);
} else if (size == -ENOENT || size == -ENODATA || size == 0) {
/* FIXME, who returns -ENOENT? I think nobody */
acl = NULL;
- btrfs_update_cached_acl(inode, p_acl, acl);
+ set_cached_acl(inode, type, acl);
} else {
acl = ERR_PTR(-EIO);
}
@@ -121,7 +98,6 @@ static int btrfs_set_acl(struct inode *inode, struct posix_acl *acl, int type)
{
int ret, size = 0;
const char *name;
- struct posix_acl **p_acl;
char *value = NULL;
mode_t mode;
@@ -141,13 +117,11 @@ static int btrfs_set_acl(struct inode *inode, struct posix_acl *acl, int type)
ret = 0;
inode->i_mode = mode;
name = POSIX_ACL_XATTR_ACCESS;
- p_acl = &BTRFS_I(inode)->i_acl;
break;
case ACL_TYPE_DEFAULT:
if (!S_ISDIR(inode->i_mode))
return acl ? -EINVAL : 0;
name = POSIX_ACL_XATTR_DEFAULT;
- p_acl = &BTRFS_I(inode)->i_default_acl;
break;
default:
return -EINVAL;
@@ -172,7 +146,7 @@ out:
kfree(value);
if (!ret)
- btrfs_update_cached_acl(inode, p_acl, acl);
+ set_cached_acl(inode, type, acl);
return ret;
}
diff --git a/fs/btrfs/async-thread.c b/fs/btrfs/async-thread.c
index 7f88628a1a7..019e8af449a 100644
--- a/fs/btrfs/async-thread.c
+++ b/fs/btrfs/async-thread.c
@@ -299,8 +299,8 @@ int btrfs_start_workers(struct btrfs_workers *workers, int num_workers)
"btrfs-%s-%d", workers->name,
workers->num_workers + i);
if (IS_ERR(worker->task)) {
- kfree(worker);
ret = PTR_ERR(worker->task);
+ kfree(worker);
goto fail;
}
@@ -424,11 +424,11 @@ int btrfs_requeue_work(struct btrfs_work *work)
* list
*/
if (worker->idle) {
- spin_lock_irqsave(&worker->workers->lock, flags);
+ spin_lock(&worker->workers->lock);
worker->idle = 0;
list_move_tail(&worker->worker_list,
&worker->workers->worker_list);
- spin_unlock_irqrestore(&worker->workers->lock, flags);
+ spin_unlock(&worker->workers->lock);
}
if (!worker->working) {
wake = 1;
diff --git a/fs/btrfs/btrfs_inode.h b/fs/btrfs/btrfs_inode.h
index acb4f351758..ea1ea0af8c0 100644
--- a/fs/btrfs/btrfs_inode.h
+++ b/fs/btrfs/btrfs_inode.h
@@ -53,10 +53,6 @@ struct btrfs_inode {
/* used to order data wrt metadata */
struct btrfs_ordered_inode_tree ordered_tree;
- /* standard acl pointers */
- struct posix_acl *i_acl;
- struct posix_acl *i_default_acl;
-
/* for keeping track of orphaned inodes */
struct list_head i_orphan;
diff --git a/fs/btrfs/compression.c b/fs/btrfs/compression.c
index de1e2fd3208..9d8ba4d54a3 100644
--- a/fs/btrfs/compression.c
+++ b/fs/btrfs/compression.c
@@ -26,7 +26,6 @@
#include <linux/time.h>
#include <linux/init.h>
#include <linux/string.h>
-#include <linux/smp_lock.h>
#include <linux/backing-dev.h>
#include <linux/mpage.h>
#include <linux/swap.h>
diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index 60a45f3a4e9..3fdcc0512d3 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -557,19 +557,7 @@ static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
btrfs_disk_key_to_cpu(&k1, disk);
- if (k1.objectid > k2->objectid)
- return 1;
- if (k1.objectid < k2->objectid)
- return -1;
- if (k1.type > k2->type)
- return 1;
- if (k1.type < k2->type)
- return -1;
- if (k1.offset > k2->offset)
- return 1;
- if (k1.offset < k2->offset)
- return -1;
- return 0;
+ return btrfs_comp_cpu_keys(&k1, k2);
}
/*
@@ -1052,9 +1040,6 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
return 0;
- if (btrfs_header_nritems(mid) > 2)
- return 0;
-
if (btrfs_header_nritems(mid) < 2)
err_on_enospc = 1;
@@ -1701,6 +1686,7 @@ int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
struct extent_buffer *b;
int slot;
int ret;
+ int err;
int level;
int lowest_unlock = 1;
u8 lowest_level = 0;
@@ -1737,8 +1723,6 @@ again:
p->locks[level] = 1;
if (cow) {
- int wret;
-
/*
* if we don't really need to cow this block
* then we don't want to set the path blocking,
@@ -1749,12 +1733,12 @@ again:
btrfs_set_path_blocking(p);
- wret = btrfs_cow_block(trans, root, b,
- p->nodes[level + 1],
- p->slots[level + 1], &b);
- if (wret) {
+ err = btrfs_cow_block(trans, root, b,
+ p->nodes[level + 1],
+ p->slots[level + 1], &b);
+ if (err) {
free_extent_buffer(b);
- ret = wret;
+ ret = err;
goto done;
}
}
@@ -1793,41 +1777,45 @@ cow_done:
ret = bin_search(b, key, level, &slot);
if (level != 0) {
- if (ret && slot > 0)
+ int dec = 0;
+ if (ret && slot > 0) {
+ dec = 1;
slot -= 1;
+ }
p->slots[level] = slot;
- ret = setup_nodes_for_search(trans, root, p, b, level,
+ err = setup_nodes_for_search(trans, root, p, b, level,
ins_len);
- if (ret == -EAGAIN)
+ if (err == -EAGAIN)
goto again;
- else if (ret)
+ if (err) {
+ ret = err;
goto done;
+ }
b = p->nodes[level];
slot = p->slots[level];
unlock_up(p, level, lowest_unlock);
- /* this is only true while dropping a snapshot */
if (level == lowest_level) {
- ret = 0;
+ if (dec)
+ p->slots[level]++;
goto done;
}
- ret = read_block_for_search(trans, root, p,
+ err = read_block_for_search(trans, root, p,
&b, level, slot, key);
- if (ret == -EAGAIN)
+ if (err == -EAGAIN)
goto again;
-
- if (ret == -EIO)
+ if (err) {
+ ret = err;
goto done;
+ }
if (!p->skip_locking) {
- int lret;
-
btrfs_clear_path_blocking(p, NULL);
- lret = btrfs_try_spin_lock(b);
+ err = btrfs_try_spin_lock(b);
- if (!lret) {
+ if (!err) {
btrfs_set_path_blocking(p);
btrfs_tree_lock(b);
btrfs_clear_path_blocking(p, b);
@@ -1837,16 +1825,14 @@ cow_done:
p->slots[level] = slot;
if (ins_len > 0 &&
btrfs_leaf_free_space(root, b) < ins_len) {
- int sret;
-
btrfs_set_path_blocking(p);
- sret = split_leaf(trans, root, key,
- p, ins_len, ret == 0);
+ err = split_leaf(trans, root, key,
+ p, ins_len, ret == 0);
btrfs_clear_path_blocking(p, NULL);
- BUG_ON(sret > 0);
- if (sret) {
- ret = sret;
+ BUG_ON(err > 0);
+ if (err) {
+ ret = err;
goto done;
}
}
@@ -3807,7 +3793,7 @@ int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
}
/* delete the leaf if it is mostly empty */
- if (used < BTRFS_LEAF_DATA_SIZE(root) / 2) {
+ if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
/* push_leaf_left fixes the path.
* make sure the path still points to our leaf
* for possible call to del_ptr below
@@ -4042,10 +4028,9 @@ out:
* calling this function.
*/
int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
- struct btrfs_key *key, int lowest_level,
+ struct btrfs_key *key, int level,
int cache_only, u64 min_trans)
{
- int level = lowest_level;
int slot;
struct extent_buffer *c;
@@ -4058,11 +4043,40 @@ int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
c = path->nodes[level];
next:
if (slot >= btrfs_header_nritems(c)) {
- level++;
- if (level == BTRFS_MAX_LEVEL)
+ int ret;
+ int orig_lowest;
+ struct btrfs_key cur_key;
+ if (level + 1 >= BTRFS_MAX_LEVEL ||
+ !path->nodes[level + 1])
return 1;
- continue;
+
+ if (path->locks[level + 1]) {
+ level++;
+ continue;
+ }
+
+ slot = btrfs_header_nritems(c) - 1;
+ if (level == 0)
+ btrfs_item_key_to_cpu(c, &cur_key, slot);
+ else
+ btrfs_node_key_to_cpu(c, &cur_key, slot);
+
+ orig_lowest = path->lowest_level;
+ btrfs_release_path(root, path);
+ path->lowest_level = level;
+ ret = btrfs_search_slot(NULL, root, &cur_key, path,
+ 0, 0);
+ path->lowest_level = orig_lowest;
+ if (ret < 0)
+ return ret;
+
+ c = path->nodes[level];
+ slot = path->slots[level];
+ if (ret == 0)
+ slot++;
+ goto next;
}
+
if (level == 0)
btrfs_item_key_to_cpu(c, key, slot);
else {
@@ -4146,7 +4160,8 @@ again:
* advance the path if there are now more items available.
*/
if (nritems > 0 && path->slots[0] < nritems - 1) {
- path->slots[0]++;
+ if (ret == 0)
+ path->slots[0]++;
ret = 0;
goto done;
}
@@ -4278,10 +4293,10 @@ int btrfs_previous_item(struct btrfs_root *root,
path->slots[0]--;
btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
- if (found_key.type == type)
- return 0;
if (found_key.objectid < min_objectid)
break;
+ if (found_key.type == type)
+ return 0;
if (found_key.objectid == min_objectid &&
found_key.type < type)
break;
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 03441a99ea3..837435ce84c 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -41,8 +41,6 @@ struct btrfs_ordered_sum;
#define BTRFS_MAGIC "_BHRfS_M"
-#define BTRFS_ACL_NOT_CACHED ((void *)-1)
-
#define BTRFS_MAX_LEVEL 8
#define BTRFS_COMPAT_EXTENT_TREE_V0
@@ -483,7 +481,7 @@ struct btrfs_shared_data_ref {
struct btrfs_extent_inline_ref {
u8 type;
- u64 offset;
+ __le64 offset;
} __attribute__ ((__packed__));
/* old style backrefs item */
@@ -691,6 +689,7 @@ struct btrfs_space_info {
struct list_head block_groups;
spinlock_t lock;
struct rw_semaphore groups_sem;
+ atomic_t caching_threads;
};
/*
@@ -709,6 +708,9 @@ struct btrfs_free_cluster {
/* first extent starting offset */
u64 window_start;
+ /* if this cluster simply points at a bitmap in the block group */
+ bool points_to_bitmap;
+
struct btrfs_block_group_cache *block_group;
/*
* when a cluster is allocated from a block group, we put the
@@ -718,24 +720,37 @@ struct btrfs_free_cluster {
struct list_head block_group_list;
};
+enum btrfs_caching_type {
+ BTRFS_CACHE_NO = 0,
+ BTRFS_CACHE_STARTED = 1,
+ BTRFS_CACHE_FINISHED = 2,
+};
+
struct btrfs_block_group_cache {
struct btrfs_key key;
struct btrfs_block_group_item item;
+ struct btrfs_fs_info *fs_info;
spinlock_t lock;
- struct mutex cache_mutex;
u64 pinned;
u64 reserved;
u64 flags;
- int cached;
+ u64 sectorsize;
+ int extents_thresh;
+ int free_extents;
+ int total_bitmaps;
int ro;
int dirty;
+ /* cache tracking stuff */
+ wait_queue_head_t caching_q;
+ int cached;
+
struct btrfs_space_info *space_info;
/* free space cache stuff */
spinlock_t tree_lock;
- struct rb_root free_space_bytes;
struct rb_root free_space_offset;
+ u64 free_space;
/* block group cache stuff */
struct rb_node cache_node;
@@ -810,6 +825,7 @@ struct btrfs_fs_info {
struct mutex drop_mutex;
struct mutex volume_mutex;
struct mutex tree_reloc_mutex;
+ struct rw_semaphore extent_commit_sem;
/*
* this protects the ordered operations list only while we are
@@ -1990,6 +2006,7 @@ void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
u64 bytes);
void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
u64 bytes);
+void btrfs_free_pinned_extents(struct btrfs_fs_info *info);
/* ctree.c */
int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
int level, int *slot);
@@ -2076,8 +2093,7 @@ static inline int btrfs_insert_empty_item(struct btrfs_trans_handle *trans,
int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path);
int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path);
int btrfs_leaf_free_space(struct btrfs_root *root, struct extent_buffer *leaf);
-int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
- *root);
+int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref);
int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
struct btrfs_root *root,
struct extent_buffer *node,
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index d28d29c95f7..e83be2e4602 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -1639,6 +1639,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
mutex_init(&fs_info->cleaner_mutex);
mutex_init(&fs_info->volume_mutex);
mutex_init(&fs_info->tree_reloc_mutex);
+ init_rwsem(&fs_info->extent_commit_sem);
btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
@@ -1799,6 +1800,11 @@ struct btrfs_root *open_ctree(struct super_block *sb,
btrfs_super_chunk_root(disk_super),
blocksize, generation);
BUG_ON(!chunk_root->node);
+ if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
+ printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
+ sb->s_id);
+ goto fail_chunk_root;
+ }
btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
chunk_root->commit_root = btrfs_root_node(chunk_root);
@@ -1826,6 +1832,11 @@ struct btrfs_root *open_ctree(struct super_block *sb,
blocksize, generation);
if (!tree_root->node)
goto fail_chunk_root;
+ if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
+ printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
+ sb->s_id);
+ goto fail_tree_root;
+ }
btrfs_set_root_node(&tree_root->root_item, tree_root->node);
tree_root->commit_root = btrfs_root_node(tree_root);
@@ -2322,6 +2333,9 @@ int close_ctree(struct btrfs_root *root)
printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
}
+ fs_info->closing = 2;
+ smp_mb();
+
if (fs_info->delalloc_bytes) {
printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
(unsigned long long)fs_info->delalloc_bytes);
@@ -2343,6 +2357,7 @@ int close_ctree(struct btrfs_root *root)
free_extent_buffer(root->fs_info->csum_root->commit_root);
btrfs_free_block_groups(root->fs_info);
+ btrfs_free_pinned_extents(root->fs_info);
del_fs_roots(fs_info);
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index edc7d208c5c..72a2b9c28e9 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -21,6 +21,7 @@
#include <linux/blkdev.h>
#include <linux/sort.h>
#include <linux/rcupdate.h>
+#include <linux/kthread.h>
#include "compat.h"
#include "hash.h"
#include "ctree.h"
@@ -61,6 +62,13 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans,
struct btrfs_root *extent_root, u64 alloc_bytes,
u64 flags, int force);
+static noinline int
+block_group_cache_done(struct btrfs_block_group_cache *cache)
+{
+ smp_mb();
+ return cache->cached == BTRFS_CACHE_FINISHED;
+}
+
static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits)
{
return (cache->flags & bits) == bits;
@@ -146,20 +154,70 @@ block_group_cache_tree_search(struct btrfs_fs_info *info, u64 bytenr,
}
/*
+ * We always set EXTENT_LOCKED for the super mirror extents so we don't
+ * overwrite them, so those bits need to be unset. Also, if we are unmounting
+ * with pinned extents still sitting there because we had a block group caching,
+ * we need to clear those now, since we are done.
+ */
+void btrfs_free_pinned_extents(struct btrfs_fs_info *info)
+{
+ u64 start, end, last = 0;
+ int ret;
+
+ while (1) {
+ ret = find_first_extent_bit(&info->pinned_extents, last,
+ &start, &end,
+ EXTENT_LOCKED|EXTENT_DIRTY);
+ if (ret)
+ break;
+
+ clear_extent_bits(&info->pinned_extents, start, end,
+ EXTENT_LOCKED|EXTENT_DIRTY, GFP_NOFS);
+ last = end+1;
+ }
+}
+
+static int remove_sb_from_cache(struct btrfs_root *root,
+ struct btrfs_block_group_cache *cache)
+{
+ struct btrfs_fs_info *fs_info = root->fs_info;
+ u64 bytenr;
+ u64 *logical;
+ int stripe_len;
+ int i, nr, ret;
+
+ for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
+ bytenr = btrfs_sb_offset(i);
+ ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
+ cache->key.objectid, bytenr,
+ 0, &logical, &nr, &stripe_len);
+ BUG_ON(ret);
+ while (nr--) {
+ try_lock_extent(&fs_info->pinned_extents,
+ logical[nr],
+ logical[nr] + stripe_len - 1, GFP_NOFS);
+ }
+ kfree(logical);
+ }
+
+ return 0;
+}
+
+/*
* this is only called by cache_block_group, since we could have freed extents
* we need to check the pinned_extents for any extents that can't be used yet
* since their free space will be released as soon as the transaction commits.
*/
-static int add_new_free_space(struct btrfs_block_group_cache *block_group,
+static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
struct btrfs_fs_info *info, u64 start, u64 end)
{
- u64 extent_start, extent_end, size;
+ u64 extent_start, extent_end, size, total_added = 0;
int ret;
while (start < end) {
ret = find_first_extent_bit(&info->pinned_extents, start,
&extent_start, &extent_end,
- EXTENT_DIRTY);
+ EXTENT_DIRTY|EXTENT_LOCKED);
if (ret)
break;
@@ -167,6 +225,7 @@ static int add_new_free_space(struct btrfs_block_group_cache *block_group,
start = extent_end + 1;
} else if (extent_start > start && extent_start < end) {
size = extent_start - start;
+ total_added += size;
ret = btrfs_add_free_space(block_group, start,
size);
BUG_ON(ret);
@@ -178,84 +237,93 @@ static int add_new_free_space(struct btrfs_block_group_cache *block_group,
if (start < end) {
size = end - start;
+ total_added += size;
ret = btrfs_add_free_space(block_group, start, size);
BUG_ON(ret);
}
- return 0;
+ return total_added;
}
-static int remove_sb_from_cache(struct btrfs_root *root,
- struct btrfs_block_group_cache *cache)
-{
- u64 bytenr;
- u64 *logical;
- int stripe_len;
- int i, nr, ret;
-
- for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
- bytenr = btrfs_sb_offset(i);
- ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
- cache->key.objectid, bytenr, 0,
- &logical, &nr, &stripe_len);
- BUG_ON(ret);
- while (nr--) {
- btrfs_remove_free_space(cache, logical[nr],
- stripe_len);
- }
- kfree(logical);
- }
- return 0;
-}
-
-static int cache_block_group(struct btrfs_root *root,
- struct btrfs_block_group_cache *block_group)
+static int caching_kthread(void *data)
{
+ struct btrfs_block_group_cache *block_group = data;
+ struct btrfs_fs_info *fs_info = block_group->fs_info;
+ u64 last = 0;
struct btrfs_path *path;
int ret = 0;
struct btrfs_key key;
struct extent_buffer *leaf;
int slot;
- u64 last;
+ u64 total_found = 0;
- if (!block_group)
- return 0;
-
- root = root->fs_info->extent_root;
-
- if (block_group->cached)
- return 0;
+ BUG_ON(!fs_info);
path = btrfs_alloc_path();
if (!path)
return -ENOMEM;
- path->reada = 2;
+ atomic_inc(&block_group->space_info->caching_threads);
+ last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
/*
- * we get into deadlocks with paths held by callers of this function.
- * since the alloc_mutex is protecting things right now, just
- * skip the locking here
+ * We don't want to deadlock with somebody trying to allocate a new
+ * extent for the extent root while also trying to search the extent
+ * root to add free space. So we skip locking and search the commit
+ * root, since its read-only
*/
path->skip_locking = 1;
- last = max_t(u64, block_group->key.objectid, BTRFS_SUPER_INFO_OFFSET);
+ path->search_commit_root = 1;
+ path->reada = 2;
+
key.objectid = last;
key.offset = 0;
btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY);
- ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+again:
+ /* need to make sure the commit_root doesn't disappear */
+ down_read(&fs_info->extent_commit_sem);
+
+ ret = btrfs_search_slot(NULL, fs_info->extent_root, &key, path, 0, 0);
if (ret < 0)
goto err;
while (1) {
+ smp_mb();
+ if (block_group->fs_info->closing > 1) {
+ last = (u64)-1;
+ break;
+ }
+
leaf = path->nodes[0];
slot = path->slots[0];
if (slot >= btrfs_header_nritems(leaf)) {
- ret = btrfs_next_leaf(root, path);
+ ret = btrfs_next_leaf(fs_info->extent_root, path);
if (ret < 0)
goto err;
- if (ret == 0)
- continue;
- else
+ else if (ret)
break;
+
+ if (need_resched() ||
+ btrfs_transaction_in_commit(fs_info)) {
+ leaf = path->nodes[0];
+
+ /* this shouldn't happen, but if the
+ * leaf is empty just move on.
+ */
+ if (btrfs_header_nritems(leaf) == 0)
+ break;
+ /*
+ * we need to copy the key out so that
+ * we are sure the next search advances
+ * us forward in the btree.
+ */
+ btrfs_item_key_to_cpu(leaf, &key, 0);
+ btrfs_release_path(fs_info->extent_root, path);
+ up_read(&fs_info->extent_commit_sem);
+ schedule_timeout(1);
+ goto again;
+ }
+
+ continue;
}
btrfs_item_key_to_cpu(leaf, &key, slot);
if (key.objectid < block_group->key.objectid)
@@ -266,24 +334,59 @@ static int cache_block_group(struct btrfs_root *root,
break;
if (btrfs_key_type(&key) == BTRFS_EXTENT_ITEM_KEY) {
- add_new_free_space(block_group, root->fs_info, last,
- key.objectid);
-
+ total_found += add_new_free_space(block_group,
+ fs_info, last,
+ key.objectid);
last = key.objectid + key.offset;
}
+
+ if (total_found > (1024 * 1024 * 2)) {
+ total_found = 0;
+ wake_up(&block_group->caching_q);
+ }
next:
path->slots[0]++;
}
+ ret = 0;
- add_new_free_space(block_group, root->fs_info, last,
- block_group->key.objectid +
- block_group->key.offset);
+ total_found += add_new_free_space(block_group, fs_info, last,
+ block_group->key.objectid +
+ block_group->key.offset);
+
+ spin_lock(&block_group->lock);
+ block_group->cached = BTRFS_CACHE_FINISHED;
+ spin_unlock(&block_group->lock);
- block_group->cached = 1;
- remove_sb_from_cache(root, block_group);
- ret = 0;
err:
btrfs_free_path(path);
+ up_read(&fs_info->extent_commit_sem);
+ atomic_dec(&block_group->space_info->caching_threads);
+ wake_up(&block_group->caching_q);
+
+ return 0;
+}
+
+static int cache_block_group(struct btrfs_block_group_cache *cache)
+{
+ struct task_struct *tsk;
+ int ret = 0;
+
+ spin_lock(&cache->lock);
+ if (cache->cached != BTRFS_CACHE_NO) {
+ spin_unlock(&cache->lock);
+ return ret;
+ }
+ cache->cached = BTRFS_CACHE_STARTED;
+ spin_unlock(&cache->lock);
+
+ tsk = kthread_run(caching_kthread, cache, "btrfs-cache-%llu\n",
+ cache->key.objectid);
+ if (IS_ERR(tsk)) {
+ ret = PTR_ERR(tsk);
+ printk(KERN_ERR "error running thread %d\n", ret);
+ BUG();
+ }
+
return ret;
}
@@ -990,15 +1093,13 @@ static inline int extent_ref_type(u64 parent, u64 owner)
return type;
}
-static int find_next_key(struct btrfs_path *path, struct btrfs_key *key)
+static int find_next_key(struct btrfs_path *path, int level,
+ struct btrfs_key *key)
{
- int level;
- BUG_ON(!path->keep_locks);
- for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
+ for (; level < BTRFS_MAX_LEVEL; level++) {
if (!path->nodes[level])
break;
- btrfs_assert_tree_locked(path->nodes[level]);
if (path->slots[level] + 1 >=
btrfs_header_nritems(path->nodes[level]))
continue;
@@ -1158,7 +1259,8 @@ int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
* For simplicity, we just do not add new inline back
* ref if there is any kind of item for this block
*/
- if (find_next_key(path, &key) == 0 && key.objectid == bytenr &&
+ if (find_next_key(path, 0, &key) == 0 &&
+ key.objectid == bytenr &&
key.type < BTRFS_BLOCK_GROUP_ITEM_KEY) {
err = -EAGAIN;
goto out;
@@ -2388,13 +2490,29 @@ fail:
}
+static struct btrfs_block_group_cache *
+next_block_group(struct btrfs_root *root,
+ struct btrfs_block_group_cache *cache)
+{
+ struct rb_node *node;
+ spin_lock(&root->fs_info->block_group_cache_lock);
+ node = rb_next(&cache->cache_node);
+ btrfs_put_block_group(cache);
+ if (node) {
+ cache = rb_entry(node, struct btrfs_block_group_cache,
+ cache_node);
+ atomic_inc(&cache->count);
+ } else
+ cache = NULL;
+ spin_unlock(&root->fs_info->block_group_cache_lock);
+ return cache;
+}
+
int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
struct btrfs_root *root)
{
- struct btrfs_block_group_cache *cache, *entry;
- struct rb_node *n;
+ struct btrfs_block_group_cache *cache;
int err = 0;
- int werr = 0;
struct btrfs_path *path;
u64 last = 0;
@@ -2403,39 +2521,35 @@ int btrfs_write_dirty_block_groups(struct btrfs_trans_handle *trans,
return -ENOMEM;
while (1) {
- cache = NULL;
- spin_lock(&root->fs_info->block_group_cache_lock);
- for (n = rb_first(&root->fs_info->block_group_cache_tree);
- n; n = rb_next(n)) {
- entry = rb_entry(n, struct btrfs_block_group_cache,
- cache_node);
- if (entry->dirty) {
- cache = entry;
- break;
- }
+ if (last == 0) {
+ err = btrfs_run_delayed_refs(trans, root,
+ (unsigned long)-1);
+ BUG_ON(err);
}
- spin_unlock(&root->fs_info->block_group_cache_lock);
- if (!cache)
- break;
+ cache = btrfs_lookup_first_block_group(root->fs_info, last);
+ while (cache) {
+ if (cache->dirty)
+ break;
+ cache = next_block_group(root, cache);
+ }
+ if (!cache) {
+ if (last == 0)
+ break;
+ last = 0;
+ continue;
+ }
cache->dirty = 0;
- last += cache->key.offset;
+ last = cache->key.objectid + cache->key.offset;
- err = write_one_cache_group(trans, root,
- path, cache);
- /*
- * if we fail to write the cache group, we want
- * to keep it marked dirty in hopes that a later
- * write will work
- */
- if (err) {
- werr = err;
- continue;
- }
+ err = write_one_cache_group(trans, root, path, cache);
+ BUG_ON(err);
+ btrfs_put_block_group(cache);
}
+
btrfs_free_path(path);
- return werr;
+ return 0;
}
int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
@@ -2485,6 +2599,7 @@ static int update_space_info(struct btrfs_fs_info *info, u64 flags,
found->force_alloc = 0;
*space_info = found;
list_add_rcu(&found->list, &info->space_info);
+ atomic_set(&found->caching_threads, 0);
return 0;
}
@@ -2697,7 +2812,7 @@ again:
printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes"
", %llu bytes_used, %llu bytes_reserved, "
- "%llu bytes_pinned, %llu bytes_readonly, %llu may use"
+ "%llu bytes_pinned, %llu bytes_readonly, %llu may use "
"%llu total\n", (unsigned long long)bytes,
(unsigned long long)data_sinfo->bytes_delalloc,
(unsigned long long)data_sinfo->bytes_used,
@@ -2948,13 +3063,9 @@ int btrfs_update_pinned_extents(struct btrfs_root *root,
struct btrfs_block_group_cache *cache;
struct btrfs_fs_info *fs_info = root->fs_info;
- if (pin) {
+ if (pin)
set_extent_dirty(&fs_info->pinned_extents,
bytenr, bytenr + num - 1, GFP_NOFS);
- } else {
- clear_extent_dirty(&fs_info->pinned_extents,
- bytenr, bytenr + num - 1, GFP_NOFS);
- }
while (num > 0) {
cache = btrfs_lookup_block_group(fs_info, bytenr);
@@ -2970,14 +3081,34 @@ int btrfs_update_pinned_extents(struct btrfs_root *root,
spin_unlock(&cache->space_info->lock);
fs_info->total_pinned += len;
} else {
+ int unpin = 0;
+
+ /*
+ * in order to not race with the block group caching, we
+ * only want to unpin the extent if we are cached. If
+ * we aren't cached, we want to start async caching this
+ * block group so we can free the extent the next time
+ * around.
+ */
spin_lock(&cache->space_info->lock);
spin_lock(&cache->lock);
- cache->pinned -= len;
- cache->space_info->bytes_pinned -= len;
+ unpin = (cache->cached == BTRFS_CACHE_FINISHED);
+ if (likely(unpin)) {
+ cache->pinned -= len;
+ cache->space_info->bytes_pinned -= len;
+ fs_info->total_pinned -= len;
+ }
spin_unlock(&cache->lock);
spin_unlock(&cache->space_info->lock);
- fs_info->total_pinned -= len;
- if (cache->cached)
+
+ if (likely(unpin))
+ clear_extent_dirty(&fs_info->pinned_extents,
+ bytenr, bytenr + len -1,
+ GFP_NOFS);
+ else
+ cache_block_group(cache);
+
+ if (unpin)
btrfs_add_free_space(cache, bytenr, len);
}
btrfs_put_block_group(cache);
@@ -3031,6 +3162,7 @@ int btrfs_copy_pinned(struct btrfs_root *root, struct extent_io_tree *copy)
&start, &end, EXTENT_DIRTY);
if (ret)
break;
+
set_extent_dirty(copy, start, end, GFP_NOFS);
last = end + 1;
}
@@ -3059,6 +3191,7 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
cond_resched();
}
+
return ret;
}
@@ -3437,6 +3570,45 @@ static u64 stripe_align(struct btrfs_root *root, u64 val)
}
/*
+ * when we wait for progress in the block group caching, its because
+ * our allocation attempt failed at least once. So, we must sleep
+ * and let some progress happen before we try again.
+ *
+ * This function will sleep at least once waiting for new free space to
+ * show up, and then it will check the block group free space numbers
+ * for our min num_bytes. Another option is to have it go ahead
+ * and look in the rbtree for a free extent of a given size, but this
+ * is a good start.
+ */
+static noinline int
+wait_block_group_cache_progress(struct btrfs_block_group_cache *cache,
+ u64 num_bytes)
+{
+ DEFINE_WAIT(wait);
+
+ prepare_to_wait(&cache->caching_q, &wait, TASK_UNINTERRUPTIBLE);
+
+ if (block_group_cache_done(cache)) {
+ finish_wait(&cache->caching_q, &wait);
+ return 0;
+ }
+ schedule();
+ finish_wait(&cache->caching_q, &wait);
+
+ wait_event(cache->caching_q, block_group_cache_done(cache) ||
+ (cache->free_space >= num_bytes));
+ return 0;
+}
+
+enum btrfs_loop_type {
+ LOOP_CACHED_ONLY = 0,
+ LOOP_CACHING_NOWAIT = 1,
+ LOOP_CACHING_WAIT = 2,
+ LOOP_ALLOC_CHUNK = 3,
+ LOOP_NO_EMPTY_SIZE = 4,
+};
+
+/*
* walks the btree of allocated extents and find a hole of a given size.
* The key ins is changed to record the hole:
* ins->objectid == block start
@@ -3461,6 +3633,7 @@ static noinline int find_free_extent(struct btrfs_trans_handle *trans,
struct btrfs_space_info *space_info;
int last_ptr_loop = 0;
int loop = 0;
+ bool found_uncached_bg = false;
WARN_ON(num_bytes < root->sectorsize);
btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
@@ -3492,15 +3665,18 @@ static noinline int find_free_extent(struct btrfs_trans_handle *trans,
search_start = max(search_start, first_logical_byte(root, 0));
search_start = max(search_start, hint_byte);
- if (!last_ptr) {
+ if (!last_ptr)
empty_cluster = 0;
- loop = 1;
- }
if (search_start == hint_byte) {
block_group = btrfs_lookup_block_group(root->fs_info,
search_start);
- if (block_group && block_group_bits(block_group, data)) {
+ /*
+ * we don't want to use the block group if it doesn't match our
+ * allocation bits, or if its not cached.
+ */
+ if (block_group && block_group_bits(block_group, data) &&
+ block_group_cache_done(block_group)) {
down_read(&space_info->groups_sem);
if (list_empty(&block_group->list) ||
block_group->ro) {
@@ -3523,21 +3699,35 @@ search:
down_read(&space_info->groups_sem);
list_for_each_entry(block_group, &space_info->block_groups, list) {
u64 offset;
+ int cached;
atomic_inc(&block_group->count);
search_start = block_group->key.objectid;
have_block_group:
- if (unlikely(!block_group->cached)) {
- mutex_lock(&block_group->cache_mutex);
- ret = cache_block_group(root, block_group);
- mutex_unlock(&block_group->cache_mutex);
- if (ret) {
- btrfs_put_block_group(block_group);
- break;
+ if (unlikely(block_group->cached == BTRFS_CACHE_NO)) {
+ /*
+ * we want to start caching kthreads, but not too many
+ * right off the bat so we don't overwhelm the system,
+ * so only start them if there are less than 2 and we're
+ * in the initial allocation phase.
+ */
+ if (loop > LOOP_CACHING_NOWAIT ||
+ atomic_read(&space_info->caching_threads) < 2) {
+ ret = cache_block_group(block_group);
+ BUG_ON(ret);
}
}
+ cached = block_group_cache_done(block_group);
+ if (unlikely(!cached)) {
+ found_uncached_bg = true;
+
+ /* if we only want cached bgs, loop */
+ if (loop == LOOP_CACHED_ONLY)
+ goto loop;
+ }
+
if (unlikely(block_group->ro))
goto loop;
@@ -3616,14 +3806,21 @@ refill_cluster:
spin_unlock(&last_ptr->refill_lock);
goto checks;
}
+ } else if (!cached && loop > LOOP_CACHING_NOWAIT) {
+ spin_unlock(&last_ptr->refill_lock);
+
+ wait_block_group_cache_progress(block_group,
+ num_bytes + empty_cluster + empty_size);
+ goto have_block_group;
}
+
/*
* at this point we either didn't find a cluster
* or we weren't able to allocate a block from our
* cluster. Free the cluster we've been trying
* to use, and go to the next block group
*/
- if (loop < 2) {
+ if (loop < LOOP_NO_EMPTY_SIZE) {
btrfs_return_cluster_to_free_space(NULL,
last_ptr);
spin_unlock(&last_ptr->refill_lock);
@@ -3634,11 +3831,17 @@ refill_cluster:
offset = btrfs_find_space_for_alloc(block_group, search_start,
num_bytes, empty_size);
- if (!offset)
+ if (!offset && (cached || (!cached &&
+ loop == LOOP_CACHING_NOWAIT))) {
goto loop;
+ } else if (!offset && (!cached &&
+ loop > LOOP_CACHING_NOWAIT)) {
+ wait_block_group_cache_progress(block_group,
+ num_bytes + empty_size);
+ goto have_block_group;
+ }
checks:
search_start = stripe_align(root, offset);
-
/* move on to the next group */
if (search_start + num_bytes >= search_end) {
btrfs_add_free_space(block_group, offset, num_bytes);
@@ -3684,13 +3887,26 @@ loop:
}
up_read(&space_info->groups_sem);
- /* loop == 0, try to find a clustered alloc in every block group
- * loop == 1, try again after forcing a chunk allocation
- * loop == 2, set empty_size and empty_cluster to 0 and try again
+ /* LOOP_CACHED_ONLY, only search fully cached block groups
+ * LOOP_CACHING_NOWAIT, search partially cached block groups, but
+ * dont wait foR them to finish caching
+ * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
+ * LOOP_ALLOC_CHUNK, force a chunk allocation and try again
+ * LOOP_NO_EMPTY_SIZE, set empty_size and empty_cluster to 0 and try
+ * again
*/
- if (!ins->objectid && loop < 3 &&
- (empty_size || empty_cluster || allowed_chunk_alloc)) {
- if (loop >= 2) {
+ if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE &&
+ (found_uncached_bg || empty_size || empty_cluster ||
+ allowed_chunk_alloc)) {
+ if (found_uncached_bg) {
+ found_uncached_bg = false;
+ if (loop < LOOP_CACHING_WAIT) {
+ loop++;
+ goto search;
+ }
+ }
+
+ if (loop == LOOP_ALLOC_CHUNK) {
empty_size = 0;
empty_cluster = 0;
}
@@ -3703,7 +3919,7 @@ loop:
space_info->force_alloc = 1;
}
- if (loop < 3) {
+ if (loop < LOOP_NO_EMPTY_SIZE) {
loop++;
goto search;
}
@@ -3799,7 +4015,7 @@ again:
num_bytes, data, 1);
goto again;
}
- if (ret) {
+ if (ret == -ENOSPC) {
struct btrfs_space_info *sinfo;
sinfo = __find_space_info(root->fs_info, data);
@@ -3807,7 +4023,6 @@ again:
"wanted %llu\n", (unsigned long long)data,
(unsigned long long)num_bytes);
dump_space_info(sinfo, num_bytes);
- BUG();
}
return ret;
@@ -3845,7 +4060,9 @@ int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size,
empty_size, hint_byte, search_end, ins,
data);
- update_reserved_extents(root, ins->objectid, ins->offset, 1);
+ if (!ret)
+ update_reserved_extents(root, ins->objectid, ins->offset, 1);
+
return ret;
}
@@ -4007,9 +4224,9 @@ int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
struct btrfs_block_group_cache *block_group;
block_group = btrfs_lookup_block_group(root->fs_info, ins->objectid);
- mutex_lock(&block_group->cache_mutex);
- cache_block_group(root, block_group);
- mutex_unlock(&block_group->cache_mutex);
+ cache_block_group(block_group);
+ wait_event(block_group->caching_q,
+ block_group_cache_done(block_group));
ret = btrfs_remove_free_space(block_group, ins->objectid,
ins->offset);
@@ -4040,7 +4257,8 @@ static int alloc_tree_block(struct btrfs_trans_handle *trans,
ret = __btrfs_reserve_extent(trans, root, num_bytes, num_bytes,
empty_size, hint_byte, search_end,
ins, 0);
- BUG_ON(ret);
+ if (ret)
+ return ret;
if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
if (parent == 0)
@@ -4128,6 +4346,7 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
return buf;
}
+#if 0
int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
struct btrfs_root *root, struct extent_buffer *leaf)
{
@@ -4171,8 +4390,6 @@ int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans,
return 0;
}
-#if 0
-
static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans,
struct btrfs_root *root,
struct btrfs_leaf_ref *ref)
@@ -4553,262 +4770,471 @@ out:
}
#endif
+struct walk_control {
+ u64 refs[BTRFS_MAX_LEVEL];
+ u64 flags[BTRFS_MAX_LEVEL];
+ struct btrfs_key update_progress;
+ int stage;
+ int level;
+ int shared_level;
+ int update_ref;
+ int keep_locks;
+};
+
+#define DROP_REFERENCE 1
+#define UPDATE_BACKREF 2
+
/*
- * helper function for drop_subtree, this function is similar to
- * walk_down_tree. The main difference is that it checks reference
- * counts while tree blocks are locked.
+ * hepler to process tree block while walking down the tree.
+ *
+ * when wc->stage == DROP_REFERENCE, this function checks
+ * reference count of the block. if the block is shared and
+ * we need update back refs for the subtree rooted at the
+ * block, this function changes wc->stage to UPDATE_BACKREF
+ *
+ * when wc->stage == UPDATE_BACKREF, this function updates
+ * back refs for pointers in the block.
+ *
+ * NOTE: return value 1 means we should stop walking down.
*/
-static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
+static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
struct btrfs_root *root,
- struct btrfs_path *path, int *level)
+ struct btrfs_path *path,
+ struct walk_control *wc)
{
- struct extent_buffer *next;
- struct extent_buffer *cur;
- struct extent_buffer *parent;
- u64 bytenr;
- u64 ptr_gen;
- u64 refs;
- u64 flags;
- u32 blocksize;
+ int level = wc->level;
+ struct extent_buffer *eb = path->nodes[level];
+ struct btrfs_key key;
+ u64 flag = BTRFS_BLOCK_FLAG_FULL_BACKREF;
int ret;
- cur = path->nodes[*level];
- ret = btrfs_lookup_extent_info(trans, root, cur->start, cur->len,
- &refs, &flags);
- BUG_ON(ret);
- if (refs > 1)
- goto out;
+ if (wc->stage == UPDATE_BACKREF &&
+ btrfs_header_owner(eb) != root->root_key.objectid)
+ return 1;
- BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
+ /*
+ * when reference count of tree block is 1, it won't increase
+ * again. once full backref flag is set, we never clear it.
+ */
+ if ((wc->stage == DROP_REFERENCE && wc->refs[level] != 1) ||
+ (wc->stage == UPDATE_BACKREF && !(wc->flags[level] & flag))) {
+ BUG_ON(!path->locks[level]);
+ ret = btrfs_lookup_extent_info(trans, root,
+ eb->start, eb->len,
+ &wc->refs[level],
+ &wc->flags[level]);
+ BUG_ON(ret);
+ BUG_ON(wc->refs[level] == 0);
+ }
- while (*level >= 0) {
- cur = path->nodes[*level];
- if (*level == 0) {
- ret = btrfs_drop_leaf_ref(trans, root, cur);
- BUG_ON(ret);
- clean_tree_block(trans, root, cur);
- break;
- }
- if (path->slots[*level] >= btrfs_header_nritems(cur)) {
- clean_tree_block(trans, root, cur);
- break;
+ if (wc->stage == DROP_REFERENCE &&
+ wc->update_ref && wc->refs[level] > 1) {
+ BUG_ON(eb == root->node);
+ BUG_ON(path->slots[level] > 0);
+ if (level == 0)
+ btrfs_item_key_to_cpu(eb, &key, path->slots[level]);
+ else
+ btrfs_node_key_to_cpu(eb, &key, path->slots[level]);
+ if (btrfs_header_owner(eb) == root->root_key.objectid &&
+ btrfs_comp_cpu_keys(&key, &wc->update_progress) >= 0) {
+ wc->stage = UPDATE_BACKREF;
+ wc->shared_level = level;
}
+ }
- bytenr = btrfs_node_blockptr(cur, path->slots[*level]);
- blocksize = btrfs_level_size(root, *level - 1);
- ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]);
+ if (wc->stage == DROP_REFERENCE) {
+ if (wc->refs[level] > 1)
+ return 1;
- next = read_tree_block(root, bytenr, blocksize, ptr_gen);
- btrfs_tree_lock(next);
- btrfs_set_lock_blocking(next);
+ if (path->locks[level] && !wc->keep_locks) {
+ btrfs_tree_unlock(eb);
+ path->locks[level] = 0;
+ }
+ return 0;
+ }
- ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
- &refs, &flags);
+ /* wc->stage == UPDATE_BACKREF */
+ if (!(wc->flags[level] & flag)) {
+ BUG_ON(!path->locks[level]);
+ ret = btrfs_inc_ref(trans, root, eb, 1);
BUG_ON(ret);
- if (refs > 1) {
- parent = path->nodes[*level];
- ret = btrfs_free_extent(trans, root, bytenr,
- blocksize, parent->start,
- btrfs_header_owner(parent),
- *level - 1, 0);
+ ret = btrfs_dec_ref(trans, root, eb, 0);
+ BUG_ON(ret);
+ ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
+ eb->len, flag, 0);
+ BUG_ON(ret);
+ wc->flags[level] |= flag;
+ }
+
+ /*
+ * the block is shared by multiple trees, so it's not good to
+ * keep the tree lock
+ */
+ if (path->locks[level] && level > 0) {
+ btrfs_tree_unlock(eb);
+ path->locks[level] = 0;
+ }
+ return 0;
+}
+
+/*
+ * hepler to process tree block while walking up the tree.
+ *
+ * when wc->stage == DROP_REFERENCE, this function drops
+ * reference count on the block.
+ *
+ * when wc->stage == UPDATE_BACKREF, this function changes
+ * wc->stage back to DROP_REFERENCE if we changed wc->stage
+ * to UPDATE_BACKREF previously while processing the block.
+ *
+ * NOTE: return value 1 means we should stop walking up.
+ */
+static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root,
+ struct btrfs_path *path,
+ struct walk_control *wc)
+{
+ int ret = 0;
+ int level = wc->level;
+ struct extent_buffer *eb = path->nodes[level];
+ u64 parent = 0;
+
+ if (wc->stage == UPDATE_BACKREF) {
+ BUG_ON(wc->shared_level < level);
+ if (level < wc->shared_level)
+ goto out;
+
+ BUG_ON(wc->refs[level] <= 1);
+ ret = find_next_key(path, level + 1, &wc->update_progress);
+ if (ret > 0)
+ wc->update_ref = 0;
+
+ wc->stage = DROP_REFERENCE;
+ wc->shared_level = -1;
+ path->slots[level] = 0;
+
+ /*
+ * check reference count again if the block isn't locked.
+ * we should start walking down the tree again if reference
+ * count is one.
+ */
+ if (!path->locks[level]) {
+ BUG_ON(level == 0);
+ btrfs_tree_lock(eb);
+ btrfs_set_lock_blocking(eb);
+ path->locks[level] = 1;
+
+ ret = btrfs_lookup_extent_info(trans, root,
+ eb->start, eb->len,
+ &wc->refs[level],
+ &wc->flags[level]);
BUG_ON(ret);
- path->slots[*level]++;
- btrfs_tree_unlock(next);
- free_extent_buffer(next);
- continue;
+ BUG_ON(wc->refs[level] == 0);
+ if (wc->refs[level] == 1) {
+ btrfs_tree_unlock(eb);
+ path->locks[level] = 0;
+ return 1;
+ }
+ } else {
+ BUG_ON(level != 0);
}
+ }
- BUG_ON(!(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
+ /* wc->stage == DROP_REFERENCE */
+ BUG_ON(wc->refs[level] > 1 && !path->locks[level]);
- *level = btrfs_header_level(next);
- path->nodes[*level] = next;
- path->slots[*level] = 0;
- path->locks[*level] = 1;
- cond_resched();
+ if (wc->refs[level] == 1) {
+ if (level == 0) {
+ if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
+ ret = btrfs_dec_ref(trans, root, eb, 1);
+ else
+ ret = btrfs_dec_ref(trans, root, eb, 0);
+ BUG_ON(ret);
+ }
+ /* make block locked assertion in clean_tree_block happy */
+ if (!path->locks[level] &&
+ btrfs_header_generation(eb) == trans->transid) {
+ btrfs_tree_lock(eb);
+ btrfs_set_lock_blocking(eb);
+ path->locks[level] = 1;
+ }
+ clean_tree_block(trans, root, eb);
}
-out:
- if (path->nodes[*level] == root->node)
- parent = path->nodes[*level];
- else
- parent = path->nodes[*level + 1];
- bytenr = path->nodes[*level]->start;
- blocksize = path->nodes[*level]->len;
- ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent->start,
- btrfs_header_owner(parent), *level, 0);
+ if (eb == root->node) {
+ if (wc->flags[level] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
+ parent = eb->start;
+ else
+ BUG_ON(root->root_key.objectid !=
+ btrfs_header_owner(eb));
+ } else {
+ if (wc->flags[level + 1] & BTRFS_BLOCK_FLAG_FULL_BACKREF)
+ parent = path->nodes[level + 1]->start;
+ else
+ BUG_ON(root->root_key.objectid !=
+ btrfs_header_owner(path->nodes[level + 1]));
+ }
+
+ ret = btrfs_free_extent(trans, root, eb->start, eb->len, parent,
+ root->root_key.objectid, level, 0);
BUG_ON(ret);
+out:
+ wc->refs[level] = 0;
+ wc->flags[level] = 0;
+ return ret;
+}
+
+static noinline int walk_down_tree(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root,
+ struct btrfs_path *path,
+ struct walk_control *wc)
+{
+ struct extent_buffer *next;
+ struct extent_buffer *cur;
+ u64 bytenr;
+ u64 ptr_gen;
+ u32 blocksize;
+ int level = wc->level;
+ int ret;
+
+ while (level >= 0) {
+ cur = path->nodes[level];
+ BUG_ON(path->slots[level] >= btrfs_header_nritems(cur));
+
+ ret = walk_down_proc(trans, root, path, wc);
+ if (ret > 0)
+ break;
+
+ if (level == 0)
+ break;
+
+ bytenr = btrfs_node_blockptr(cur, path->slots[level]);
+ blocksize = btrfs_level_size(root, level - 1);
+ ptr_gen = btrfs_node_ptr_generation(cur, path->slots[level]);
+
+ next = read_tree_block(root, bytenr, blocksize, ptr_gen);
+ btrfs_tree_lock(next);
+ btrfs_set_lock_blocking(next);
- if (path->locks[*level]) {
- btrfs_tree_unlock(path->nodes[*level]);
- path->locks[*level] = 0;
+ level--;
+ BUG_ON(level != btrfs_header_level(next));
+ path->nodes[level] = next;
+ path->slots[level] = 0;
+ path->locks[level] = 1;
+ wc->level = level;
}
- free_extent_buffer(path->nodes[*level]);
- path->nodes[*level] = NULL;
- *level += 1;
- cond_resched();
return 0;
}
-/*
- * helper for dropping snapshots. This walks back up the tree in the path
- * to find the first node higher up where we haven't yet gone through
- * all the slots
- */
static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
struct btrfs_root *root,
struct btrfs_path *path,
- int *level, int max_level)
+ struct walk_control *wc, int max_level)
{
- struct btrfs_root_item *root_item = &root->root_item;
- int i;
- int slot;
+ int level = wc->level;
int ret;
- for (i = *level; i < max_level && path->nodes[i]; i++) {
- slot = path->slots[i];
- if (slot + 1 < btrfs_header_nritems(path->nodes[i])) {
- /*
- * there is more work to do in this level.
- * Update the drop_progress marker to reflect
- * the work we've done so far, and then bump
- * the slot number
- */
- path->slots[i]++;
- WARN_ON(*level == 0);
- if (max_level == BTRFS_MAX_LEVEL) {
- btrfs_node_key(path->nodes[i],
- &root_item->drop_progress,
- path->slots[i]);
- root_item->drop_level = i;
- }
- *level = i;
+ path->slots[level] = btrfs_header_nritems(path->nodes[level]);
+ while (level < max_level && path->nodes[level]) {
+ wc->level = level;
+ if (path->slots[level] + 1 <
+ btrfs_header_nritems(path->nodes[level])) {
+ path->slots[level]++;
return 0;
} else {
- struct extent_buffer *parent;
-
- /*
- * this whole node is done, free our reference
- * on it and go up one level
- */
- if (path->nodes[*level] == root->node)
- parent = path->nodes[*level];
- else
- parent = path->nodes[*level + 1];
+ ret = walk_up_proc(trans, root, path, wc);
+ if (ret > 0)
+ return 0;
- clean_tree_block(trans, root, path->nodes[i]);
- ret = btrfs_free_extent(trans, root,
- path->nodes[i]->start,
- path->nodes[i]->len,
- parent->start,
- btrfs_header_owner(parent),
- *level, 0);
- BUG_ON(ret);
- if (path->locks[*level]) {
- btrfs_tree_unlock(path->nodes[i]);
- path->locks[i] = 0;
+ if (path->locks[level]) {
+ btrfs_tree_unlock(path->nodes[level]);
+ path->locks[level] = 0;
}
- free_extent_buffer(path->nodes[i]);
- path->nodes[i] = NULL;
- *level = i + 1;
+ free_extent_buffer(path->nodes[level]);
+ path->nodes[level] = NULL;
+ level++;
}
}
return 1;
}
/*
- * drop the reference count on the tree rooted at 'snap'. This traverses
- * the tree freeing any blocks that have a ref count of zero after being
- * decremented.
+ * drop a subvolume tree.
+ *
+ * this function traverses the tree freeing any blocks that only
+ * referenced by the tree.
+ *
+ * when a shared tree block is found. this function decreases its
+ * reference count by one. if update_ref is true, this function
+ * also make sure backrefs for the shared block and all lower level
+ * blocks are properly updated.
*/
-int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root
- *root)
+int btrfs_drop_snapshot(struct btrfs_root *root, int update_ref)
{
- int ret = 0;
- int wret;
- int level;
struct btrfs_path *path;
- int update_count;
+ struct btrfs_trans_handle *trans;
+ struct btrfs_root *tree_root = root->fs_info->tree_root;
struct btrfs_root_item *root_item = &root->root_item;
+ struct walk_control *wc;
+ struct btrfs_key key;
+ int err = 0;
+ int ret;
+ int level;
path = btrfs_alloc_path();
BUG_ON(!path);
- level = btrfs_header_level(root->node);
+ wc = kzalloc(sizeof(*wc), GFP_NOFS);
+ BUG_ON(!wc);
+
+ trans = btrfs_start_transaction(tree_root, 1);
+
if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) {
+ level = btrfs_header_level(root->node);
path->nodes[level] = btrfs_lock_root_node(root);
btrfs_set_lock_blocking(path->nodes[level]);
path->slots[level] = 0;
path->locks[level] = 1;
+ memset(&wc->update_progress, 0,
+ sizeof(wc->update_progress));
} else {
- struct btrfs_key key;
- struct btrfs_disk_key found_key;
- struct extent_buffer *node;
-
btrfs_disk_key_to_cpu(&key, &root_item->drop_progress);
+ memcpy(&wc->update_progress, &key,
+ sizeof(wc->update_progress));
+
level = root_item->drop_level;
+ BUG_ON(level == 0);
path->lowest_level = level;
- wret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
- if (wret < 0) {
- ret = wret;
+ ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
+ path->lowest_level = 0;
+ if (ret < 0) {
+ err = ret;
goto out;
}
- node = path->nodes[level];
- btrfs_node_key(node, &found_key, path->slots[level]);
- WARN_ON(memcmp(&found_key, &root_item->drop_progress,
- sizeof(found_key)));
+ btrfs_node_key_to_cpu(path->nodes[level], &key,
+ path->slots[level]);
+ WARN_ON(memcmp(&key, &wc->update_progress, sizeof(key)));
+
/*
* unlock our path, this is safe because only this
* function is allowed to delete this snapshot
*/
btrfs_unlock_up_safe(path, 0);
+
+ level = btrfs_header_level(root->node);
+ while (1) {
+ btrfs_tree_lock(path->nodes[level]);
+ btrfs_set_lock_blocking(path->nodes[level]);
+
+ ret = btrfs_lookup_extent_info(trans, root,
+ path->nodes[level]->start,
+ path->nodes[level]->len,
+ &wc->refs[level],
+ &wc->flags[level]);
+ BUG_ON(ret);
+ BUG_ON(wc->refs[level] == 0);
+
+ if (level == root_item->drop_level)
+ break;
+
+ btrfs_tree_unlock(path->nodes[level]);
+ WARN_ON(wc->refs[level] != 1);
+ level--;
+ }
}
+
+ wc->level = level;
+ wc->shared_level = -1;
+ wc->stage = DROP_REFERENCE;
+ wc->update_ref = update_ref;
+ wc->keep_locks = 0;
+
while (1) {
- unsigned long update;
- wret = walk_down_tree(trans, root, path, &level);
- if (wret > 0)
+ ret = walk_down_tree(trans, root, path, wc);
+ if (ret < 0) {
+ err = ret;
break;
- if (wret < 0)
- ret = wret;
+ }
- wret = walk_up_tree(trans, root, path, &level,
- BTRFS_MAX_LEVEL);
- if (wret > 0)
+ ret = walk_up_tree(trans, root, path, wc, BTRFS_MAX_LEVEL);
+ if (ret < 0) {
+ err = ret;
break;
- if (wret < 0)
- ret = wret;
- if (trans->transaction->in_commit ||
- trans->transaction->delayed_refs.flushing) {
- ret = -EAGAIN;
+ }
+
+ if (ret > 0) {
+ BUG_ON(wc->stage != DROP_REFERENCE);
break;
}
- for (update_count = 0; update_count < 16; update_count++) {
+
+ if (wc->stage == DROP_REFERENCE) {
+ level = wc->level;
+ btrfs_node_key(path->nodes[level],
+ &root_item->drop_progress,
+ path->slots[level]);
+ root_item->drop_level = level;
+ }
+
+ BUG_ON(wc->level == 0);
+ if (trans->transaction->in_commit ||
+ trans->transaction->delayed_refs.flushing) {
+ ret = btrfs_update_root(trans, tree_root,
+ &root->root_key,
+ root_item);
+ BUG_ON(ret);
+
+ btrfs_end_transaction(trans, tree_root);
+ trans = btrfs_start_transaction(tree_root, 1);
+ } else {
+ unsigned long update;
update = trans->delayed_ref_updates;
trans->delayed_ref_updates = 0;
if (update)
- btrfs_run_delayed_refs(trans, root, update);
- else
- break;
+ btrfs_run_delayed_refs(trans, tree_root,
+ update);
}
}
+ btrfs_release_path(root, path);
+ BUG_ON(err);
+
+ ret = btrfs_del_root(trans, tree_root, &root->root_key);
+ BUG_ON(ret);
+
+ free_extent_buffer(root->node);
+ free_extent_buffer(root->commit_root);
+ kfree(root);
out:
+ btrfs_end_transaction(trans, tree_root);
+ kfree(wc);
btrfs_free_path(path);
- return ret;
+ return err;
}
+/*
+ * drop subtree rooted at tree block 'node'.
+ *
+ * NOTE: this function will unlock and release tree block 'node'
+ */
int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
struct btrfs_root *root,
struct extent_buffer *node,
struct extent_buffer *parent)
{
struct btrfs_path *path;
+ struct walk_control *wc;
int level;
int parent_level;
int ret = 0;
int wret;
+ BUG_ON(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID);
+
path = btrfs_alloc_path();
BUG_ON(!path);
+ wc = kzalloc(sizeof(*wc), GFP_NOFS);
+ BUG_ON(!wc);
+
btrfs_assert_tree_locked(parent);
parent_level = btrfs_header_level(parent);
extent_buffer_get(parent);
@@ -4817,24 +5243,33 @@ int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
btrfs_assert_tree_locked(node);
level = btrfs_header_level(node);
- extent_buffer_get(node);
path->nodes[level] = node;
path->slots[level] = 0;
+ path->locks[level] = 1;
+
+ wc->refs[parent_level] = 1;
+ wc->flags[parent_level] = BTRFS_BLOCK_FLAG_FULL_BACKREF;
+ wc->level = level;
+ wc->shared_level = -1;
+ wc->stage = DROP_REFERENCE;
+ wc->update_ref = 0;
+ wc->keep_locks = 1;
while (1) {
- wret = walk_down_tree(trans, root, path, &level);
- if (wret < 0)
+ wret = walk_down_tree(trans, root, path, wc);
+ if (wret < 0) {
ret = wret;
- if (wret != 0)
break;
+ }
- wret = walk_up_tree(trans, root, path, &level, parent_level);
+ wret = walk_up_tree(trans, root, path, wc, parent_level);
if (wret < 0)
ret = wret;
if (wret != 0)
break;
}
+ kfree(wc);
btrfs_free_path(path);
return ret;
}
@@ -6739,11 +7174,16 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info)
&info->block_group_cache_tree);
spin_unlock(&info->block_group_cache_lock);
- btrfs_remove_free_space_cache(block_group);
down_write(&block_group->space_info->groups_sem);
list_del(&block_group->list);
up_write(&block_group->space_info->groups_sem);
+ if (block_group->cached == BTRFS_CACHE_STARTED)
+ wait_event(block_group->caching_q,
+ block_group_cache_done(block_group));
+
+ btrfs_remove_free_space_cache(block_group);
+
WARN_ON(atomic_read(&block_group->count) != 1);
kfree(block_group);
@@ -6809,9 +7249,19 @@ int btrfs_read_block_groups(struct btrfs_root *root)
atomic_set(&cache->count, 1);
spin_lock_init(&cache->lock);
spin_lock_init(&cache->tree_lock);
- mutex_init(&cache->cache_mutex);
+ cache->fs_info = info;
+ init_waitqueue_head(&cache->caching_q);
INIT_LIST_HEAD(&cache->list);
INIT_LIST_HEAD(&cache->cluster_list);
+
+ /*
+ * we only want to have 32k of ram per block group for keeping
+ * track of free space, and if we pass 1/2 of that we want to
+ * start converting things over to using bitmaps
+ */
+ cache->extents_thresh = ((1024 * 32) / 2) /
+ sizeof(struct btrfs_free_space);
+
read_extent_buffer(leaf, &cache->item,
btrfs_item_ptr_offset(leaf, path->slots[0]),
sizeof(cache->item));
@@ -6820,6 +7270,26 @@ int btrfs_read_block_groups(struct btrfs_root *root)
key.objectid = found_key.objectid + found_key.offset;
btrfs_release_path(root, path);
cache->flags = btrfs_block_group_flags(&cache->item);
+ cache->sectorsize = root->sectorsize;
+
+ remove_sb_from_cache(root, cache);
+
+ /*
+ * check for two cases, either we are full, and therefore
+ * don't need to bother with the caching work since we won't
+ * find any space, or we are empty, and we can just add all
+ * the space in and be done with it. This saves us _alot_ of
+ * time, particularly in the full case.
+ */
+ if (found_key.offset == btrfs_block_group_used(&cache->item)) {
+ cache->cached = BTRFS_CACHE_FINISHED;
+ } else if (btrfs_block_group_used(&cache->item) == 0) {
+ cache->cached = BTRFS_CACHE_FINISHED;
+ add_new_free_space(cache, root->fs_info,
+ found_key.objectid,
+ found_key.objectid +
+ found_key.offset);
+ }
ret = update_space_info(info, cache->flags, found_key.offset,
btrfs_block_group_used(&cache->item),
@@ -6863,10 +7333,19 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
cache->key.objectid = chunk_offset;
cache->key.offset = size;
cache->key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
+ cache->sectorsize = root->sectorsize;
+
+ /*
+ * we only want to have 32k of ram per block group for keeping track
+ * of free space, and if we pass 1/2 of that we want to start
+ * converting things over to using bitmaps
+ */
+ cache->extents_thresh = ((1024 * 32) / 2) /
+ sizeof(struct btrfs_free_space);
atomic_set(&cache->count, 1);
spin_lock_init(&cache->lock);
spin_lock_init(&cache->tree_lock);
- mutex_init(&cache->cache_mutex);
+ init_waitqueue_head(&cache->caching_q);
INIT_LIST_HEAD(&cache->list);
INIT_LIST_HEAD(&cache->cluster_list);
@@ -6875,6 +7354,12 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
cache->flags = type;
btrfs_set_block_group_flags(&cache->item, type);
+ cache->cached = BTRFS_CACHE_FINISHED;
+ remove_sb_from_cache(root, cache);
+
+ add_new_free_space(cache, root->fs_info, chunk_offset,
+ chunk_offset + size);
+
ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
&cache->space_info);
BUG_ON(ret);
@@ -6933,7 +7418,7 @@ int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
rb_erase(&block_group->cache_node,
&root->fs_info->block_group_cache_tree);
spin_unlock(&root->fs_info->block_group_cache_lock);
- btrfs_remove_free_space_cache(block_group);
+
down_write(&block_group->space_info->groups_sem);
/*
* we must use list_del_init so people can check to see if they
@@ -6942,11 +7427,18 @@ int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
list_del_init(&block_group->list);
up_write(&block_group->space_info->groups_sem);
+ if (block_group->cached == BTRFS_CACHE_STARTED)
+ wait_event(block_group->caching_q,
+ block_group_cache_done(block_group));
+
+ btrfs_remove_free_space_cache(block_group);
+
spin_lock(&block_group->space_info->lock);
block_group->space_info->total_bytes -= block_group->key.offset;
block_group->space_info->bytes_readonly -= block_group->key.offset;
spin_unlock(&block_group->space_info->lock);
- block_group->space_info->full = 0;
+
+ btrfs_clear_space_info_full(root->fs_info);
btrfs_put_block_group(block_group);
btrfs_put_block_group(block_group);
diff --git a/fs/btrfs/file.c b/fs/btrfs/file.c
index 126477eaecf..4b833972273 100644
--- a/fs/btrfs/file.c
+++ b/fs/btrfs/file.c
@@ -22,7 +22,6 @@
#include <linux/time.h>
#include <linux/init.h>
#include <linux/string.h>
-#include <linux/smp_lock.h>
#include <linux/backing-dev.h>
#include <linux/mpage.h>
#include <linux/swap.h>
@@ -151,7 +150,10 @@ static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
}
if (end_pos > isize) {
i_size_write(inode, end_pos);
- btrfs_update_inode(trans, root, inode);
+ /* we've only changed i_size in ram, and we haven't updated
+ * the disk i_size. There is no need to log the inode
+ * at this time.
+ */
}
err = btrfs_end_transaction(trans, root);
out_unlock:
diff --git a/fs/btrfs/free-space-cache.c b/fs/btrfs/free-space-cache.c
index 4538e48581a..5edcee3a617 100644
--- a/fs/btrfs/free-space-cache.c
+++ b/fs/btrfs/free-space-cache.c
@@ -16,45 +16,46 @@
* Boston, MA 021110-1307, USA.
*/
+#include <linux/pagemap.h>
#include <linux/sched.h>
+#include <linux/math64.h>
#include "ctree.h"
#include "free-space-cache.h"
#include "transaction.h"
-struct btrfs_free_space {
- struct rb_node bytes_index;
- struct rb_node offset_index;
- u64 offset;
- u64 bytes;
-};
+#define BITS_PER_BITMAP (PAGE_CACHE_SIZE * 8)
+#define MAX_CACHE_BYTES_PER_GIG (32 * 1024)
-static int tree_insert_offset(struct rb_root *root, u64 offset,
- struct rb_node *node)
+static inline unsigned long offset_to_bit(u64 bitmap_start, u64 sectorsize,
+ u64 offset)
{
- struct rb_node **p = &root->rb_node;
- struct rb_node *parent = NULL;
- struct btrfs_free_space *info;
+ BUG_ON(offset < bitmap_start);
+ offset -= bitmap_start;
+ return (unsigned long)(div64_u64(offset, sectorsize));
+}
- while (*p) {
- parent = *p;
- info = rb_entry(parent, struct btrfs_free_space, offset_index);
+static inline unsigned long bytes_to_bits(u64 bytes, u64 sectorsize)
+{
+ return (unsigned long)(div64_u64(bytes, sectorsize));
+}
- if (offset < info->offset)
- p = &(*p)->rb_left;
- else if (offset > info->offset)
- p = &(*p)->rb_right;
- else
- return -EEXIST;
- }
+static inline u64 offset_to_bitmap(struct btrfs_block_group_cache *block_group,
+ u64 offset)
+{
+ u64 bitmap_start;
+ u64 bytes_per_bitmap;
- rb_link_node(node, parent, p);
- rb_insert_color(node, root);
+ bytes_per_bitmap = BITS_PER_BITMAP * block_group->sectorsize;
+ bitmap_start = offset - block_group->key.objectid;
+ bitmap_start = div64_u64(bitmap_start, bytes_per_bitmap);
+ bitmap_start *= bytes_per_bitmap;
+ bitmap_start += block_group->key.objectid;
- return 0;
+ return bitmap_start;
}
-static int tree_insert_bytes(struct rb_root *root, u64 bytes,
- struct rb_node *node)
+static int tree_insert_offset(struct rb_root *root, u64 offset,
+ struct rb_node *node, int bitmap)
{
struct rb_node **p = &root->rb_node;
struct rb_node *parent = NULL;
@@ -62,12 +63,34 @@ static int tree_insert_bytes(struct rb_root *root, u64 bytes,
while (*p) {
parent = *p;
- info = rb_entry(parent, struct btrfs_free_space, bytes_index);
+ info = rb_entry(parent, struct btrfs_free_space, offset_index);
- if (bytes < info->bytes)
+ if (offset < info->offset) {
p = &(*p)->rb_left;
- else
+ } else if (offset > info->offset) {
p = &(*p)->rb_right;
+ } else {
+ /*
+ * we could have a bitmap entry and an extent entry
+ * share the same offset. If this is the case, we want
+ * the extent entry to always be found first if we do a
+ * linear search through the tree, since we want to have
+ * the quickest allocation time, and allocating from an
+ * extent is faster than allocating from a bitmap. So
+ * if we're inserting a bitmap and we find an entry at
+ * this offset, we want to go right, or after this entry
+ * logically. If we are inserting an extent and we've
+ * found a bitmap, we want to go left, or before
+ * logically.
+ */
+ if (bitmap) {
+ WARN_ON(info->bitmap);
+ p = &(*p)->rb_right;
+ } else {
+ WARN_ON(!info->bitmap);
+ p = &(*p)->rb_left;
+ }
+ }
}
rb_link_node(node, parent, p);
@@ -79,110 +102,143 @@ static int tree_insert_bytes(struct rb_root *root, u64 bytes,
/*
* searches the tree for the given offset.
*
- * fuzzy == 1: this is used for allocations where we are given a hint of where
- * to look for free space. Because the hint may not be completely on an offset
- * mark, or the hint may no longer point to free space we need to fudge our
- * results a bit. So we look for free space starting at or after offset with at
- * least bytes size. We prefer to find as close to the given offset as we can.
- * Also if the offset is within a free space range, then we will return the free
- * space that contains the given offset, which means we can return a free space
- * chunk with an offset before the provided offset.
- *
- * fuzzy == 0: this is just a normal tree search. Give us the free space that
- * starts at the given offset which is at least bytes size, and if its not there
- * return NULL.
+ * fuzzy - If this is set, then we are trying to make an allocation, and we just
+ * want a section that has at least bytes size and comes at or after the given
+ * offset.
*/
-static struct btrfs_free_space *tree_search_offset(struct rb_root *root,
- u64 offset, u64 bytes,
- int fuzzy)
+static struct btrfs_free_space *
+tree_search_offset(struct btrfs_block_group_cache *block_group,
+ u64 offset, int bitmap_only, int fuzzy)
{
- struct rb_node *n = root->rb_node;
- struct btrfs_free_space *entry, *ret = NULL;
+ struct rb_node *n = block_group->free_space_offset.rb_node;
+ struct btrfs_free_space *entry, *prev = NULL;
+
+ /* find entry that is closest to the 'offset' */
+ while (1) {
+ if (!n) {
+ entry = NULL;
+ break;
+ }
- while (n) {
entry = rb_entry(n, struct btrfs_free_space, offset_index);
+ prev = entry;
- if (offset < entry->offset) {
- if (fuzzy &&
- (!ret || entry->offset < ret->offset) &&
- (bytes <= entry->bytes))
- ret = entry;
+ if (offset < entry->offset)
n = n->rb_left;
- } else if (offset > entry->offset) {
- if (fuzzy &&
- (entry->offset + entry->bytes - 1) >= offset &&
- bytes <= entry->bytes) {
- ret = entry;
- break;
- }
+ else if (offset > entry->offset)
n = n->rb_right;
- } else {
- if (bytes > entry->bytes) {
- n = n->rb_right;
- continue;
- }
- ret = entry;
+ else
break;
- }
}
- return ret;
-}
+ if (bitmap_only) {
+ if (!entry)
+ return NULL;
+ if (entry->bitmap)
+ return entry;
-/*
- * return a chunk at least bytes size, as close to offset that we can get.
- */
-static struct btrfs_free_space *tree_search_bytes(struct rb_root *root,
- u64 offset, u64 bytes)
-{
- struct rb_node *n = root->rb_node;
- struct btrfs_free_space *entry, *ret = NULL;
-
- while (n) {
- entry = rb_entry(n, struct btrfs_free_space, bytes_index);
+ /*
+ * bitmap entry and extent entry may share same offset,
+ * in that case, bitmap entry comes after extent entry.
+ */
+ n = rb_next(n);
+ if (!n)
+ return NULL;
+ entry = rb_entry(n, struct btrfs_free_space, offset_index);
+ if (entry->offset != offset)
+ return NULL;
- if (bytes < entry->bytes) {
+ WARN_ON(!entry->bitmap);
+ return entry;
+ } else if (entry) {
+ if (entry->bitmap) {
/*
- * We prefer to get a hole size as close to the size we
- * are asking for so we don't take small slivers out of
- * huge holes, but we also want to get as close to the
- * offset as possible so we don't have a whole lot of
- * fragmentation.
+ * if previous extent entry covers the offset,
+ * we should return it instead of the bitmap entry
*/
- if (offset <= entry->offset) {
- if (!ret)
- ret = entry;
- else if (entry->bytes < ret->bytes)
- ret = entry;
- else if (entry->offset < ret->offset)
- ret = entry;
+ n = &entry->offset_index;
+ while (1) {
+ n = rb_prev(n);
+ if (!n)
+ break;
+ prev = rb_entry(n, struct btrfs_free_space,
+ offset_index);
+ if (!prev->bitmap) {
+ if (prev->offset + prev->bytes > offset)
+ entry = prev;
+ break;
+ }
}
- n = n->rb_left;
- } else if (bytes > entry->bytes) {
- n = n->rb_right;
+ }
+ return entry;
+ }
+
+ if (!prev)
+ return NULL;
+
+ /* find last entry before the 'offset' */
+ entry = prev;
+ if (entry->offset > offset) {
+ n = rb_prev(&entry->offset_index);
+ if (n) {
+ entry = rb_entry(n, struct btrfs_free_space,
+ offset_index);
+ BUG_ON(entry->offset > offset);
} else {
- /*
- * Ok we may have multiple chunks of the wanted size,
- * so we don't want to take the first one we find, we
- * want to take the one closest to our given offset, so
- * keep searching just in case theres a better match.
- */
- n = n->rb_right;
- if (offset > entry->offset)
- continue;
- else if (!ret || entry->offset < ret->offset)
- ret = entry;
+ if (fuzzy)
+ return entry;
+ else
+ return NULL;
}
}
- return ret;
+ if (entry->bitmap) {
+ n = &entry->offset_index;
+ while (1) {
+ n = rb_prev(n);
+ if (!n)
+ break;
+ prev = rb_entry(n, struct btrfs_free_space,
+ offset_index);
+ if (!prev->bitmap) {
+ if (prev->offset + prev->bytes > offset)
+ return prev;
+ break;
+ }
+ }
+ if (entry->offset + BITS_PER_BITMAP *
+ block_group->sectorsize > offset)
+ return entry;
+ } else if (entry->offset + entry->bytes > offset)
+ return entry;
+
+ if (!fuzzy)
+ return NULL;
+
+ while (1) {
+ if (entry->bitmap) {
+ if (entry->offset + BITS_PER_BITMAP *
+ block_group->sectorsize > offset)
+ break;
+ } else {
+ if (entry->offset + entry->bytes > offset)
+ break;
+ }
+
+ n = rb_next(&entry->offset_index);
+ if (!n)
+ return NULL;
+ entry = rb_entry(n, struct btrfs_free_space, offset_index);
+ }
+ return entry;
}
static void unlink_free_space(struct btrfs_block_group_cache *block_group,
struct btrfs_free_space *info)
{
rb_erase(&info->offset_index, &block_group->free_space_offset);
- rb_erase(&info->bytes_index, &block_group->free_space_bytes);
+ block_group->free_extents--;
+ block_group->free_space -= info->bytes;
}
static int link_free_space(struct btrfs_block_group_cache *block_group,
@@ -190,17 +246,353 @@ static int link_free_space(struct btrfs_block_group_cache *block_group,
{
int ret = 0;
-
- BUG_ON(!info->bytes);
+ BUG_ON(!info->bitmap && !info->bytes);
ret = tree_insert_offset(&block_group->free_space_offset, info->offset,
- &info->offset_index);
+ &info->offset_index, (info->bitmap != NULL));
if (ret)
return ret;
- ret = tree_insert_bytes(&block_group->free_space_bytes, info->bytes,
- &info->bytes_index);
- if (ret)
- return ret;
+ block_group->free_space += info->bytes;
+ block_group->free_extents++;
+ return ret;
+}
+
+static void recalculate_thresholds(struct btrfs_block_group_cache *block_group)
+{
+ u64 max_bytes, possible_bytes;
+
+ /*
+ * The goal is to keep the total amount of memory used per 1gb of space
+ * at or below 32k, so we need to adjust how much memory we allow to be
+ * used by extent based free space tracking
+ */
+ max_bytes = MAX_CACHE_BYTES_PER_GIG *
+ (div64_u64(block_group->key.offset, 1024 * 1024 * 1024));
+
+ possible_bytes = (block_group->total_bitmaps * PAGE_CACHE_SIZE) +
+ (sizeof(struct btrfs_free_space) *
+ block_group->extents_thresh);
+
+ if (possible_bytes > max_bytes) {
+ int extent_bytes = max_bytes -
+ (block_group->total_bitmaps * PAGE_CACHE_SIZE);
+
+ if (extent_bytes <= 0) {
+ block_group->extents_thresh = 0;
+ return;
+ }
+
+ block_group->extents_thresh = extent_bytes /
+ (sizeof(struct btrfs_free_space));
+ }
+}
+
+static void bitmap_clear_bits(struct btrfs_block_group_cache *block_group,
+ struct btrfs_free_space *info, u64 offset,
+ u64 bytes)
+{
+ unsigned long start, end;
+ unsigned long i;
+
+ start = offset_to_bit(info->offset, block_group->sectorsize, offset);
+ end = start + bytes_to_bits(bytes, block_group->sectorsize);
+ BUG_ON(end > BITS_PER_BITMAP);
+
+ for (i = start; i < end; i++)
+ clear_bit(i, info->bitmap);
+
+ info->bytes -= bytes;
+ block_group->free_space -= bytes;
+}
+
+static void bitmap_set_bits(struct btrfs_block_group_cache *block_group,
+ struct btrfs_free_space *info, u64 offset,
+ u64 bytes)
+{
+ unsigned long start, end;
+ unsigned long i;
+
+ start = offset_to_bit(info->offset, block_group->sectorsize, offset);
+ end = start + bytes_to_bits(bytes, block_group->sectorsize);
+ BUG_ON(end > BITS_PER_BITMAP);
+
+ for (i = start; i < end; i++)
+ set_bit(i, info->bitmap);
+
+ info->bytes += bytes;
+ block_group->free_space += bytes;
+}
+
+static int search_bitmap(struct btrfs_block_group_cache *block_group,
+ struct btrfs_free_space *bitmap_info, u64 *offset,
+ u64 *bytes)
+{
+ unsigned long found_bits = 0;
+ unsigned long bits, i;
+ unsigned long next_zero;
+
+ i = offset_to_bit(bitmap_info->offset, block_group->sectorsize,
+ max_t(u64, *offset, bitmap_info->offset));
+ bits = bytes_to_bits(*bytes, block_group->sectorsize);
+
+ for (i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i);
+ i < BITS_PER_BITMAP;
+ i = find_next_bit(bitmap_info->bitmap, BITS_PER_BITMAP, i + 1)) {
+ next_zero = find_next_zero_bit(bitmap_info->bitmap,
+ BITS_PER_BITMAP, i);
+ if ((next_zero - i) >= bits) {
+ found_bits = next_zero - i;
+ break;
+ }
+ i = next_zero;
+ }
+
+ if (found_bits) {
+ *offset = (u64)(i * block_group->sectorsize) +
+ bitmap_info->offset;
+ *bytes = (u64)(found_bits) * block_group->sectorsize;
+ return 0;
+ }
+
+ return -1;
+}
+
+static struct btrfs_free_space *find_free_space(struct btrfs_block_group_cache
+ *block_group, u64 *offset,
+ u64 *bytes, int debug)
+{
+ struct btrfs_free_space *entry;
+ struct rb_node *node;
+ int ret;
+
+ if (!block_group->free_space_offset.rb_node)
+ return NULL;
+
+ entry = tree_search_offset(block_group,
+ offset_to_bitmap(block_group, *offset),
+ 0, 1);
+ if (!entry)
+ return NULL;
+
+ for (node = &entry->offset_index; node; node = rb_next(node)) {
+ entry = rb_entry(node, struct btrfs_free_space, offset_index);
+ if (entry->bytes < *bytes)
+ continue;
+
+ if (entry->bitmap) {
+ ret = search_bitmap(block_group, entry, offset, bytes);
+ if (!ret)
+ return entry;
+ continue;
+ }
+
+ *offset = entry->offset;
+ *bytes = entry->bytes;
+ return entry;
+ }
+
+ return NULL;
+}
+
+static void add_new_bitmap(struct btrfs_block_group_cache *block_group,
+ struct btrfs_free_space *info, u64 offset)
+{
+ u64 bytes_per_bg = BITS_PER_BITMAP * block_group->sectorsize;
+ int max_bitmaps = (int)div64_u64(block_group->key.offset +
+ bytes_per_bg - 1, bytes_per_bg);
+ BUG_ON(block_group->total_bitmaps >= max_bitmaps);
+
+ info->offset = offset_to_bitmap(block_group, offset);
+ link_free_space(block_group, info);
+ block_group->total_bitmaps++;
+
+ recalculate_thresholds(block_group);
+}
+
+static noinline int remove_from_bitmap(struct btrfs_block_group_cache *block_group,
+ struct btrfs_free_space *bitmap_info,
+ u64 *offset, u64 *bytes)
+{
+ u64 end;
+ u64 search_start, search_bytes;
+ int ret;
+
+again:
+ end = bitmap_info->offset +
+ (u64)(BITS_PER_BITMAP * block_group->sectorsize) - 1;
+
+ /*
+ * XXX - this can go away after a few releases.
+ *
+ * since the only user of btrfs_remove_free_space is the tree logging
+ * stuff, and the only way to test that is under crash conditions, we
+ * want to have this debug stuff here just in case somethings not
+ * working. Search the bitmap for the space we are trying to use to
+ * make sure its actually there. If its not there then we need to stop
+ * because something has gone wrong.
+ */
+ search_start = *offset;
+ search_bytes = *bytes;
+ ret = search_bitmap(block_group, bitmap_info, &search_start,
+ &search_bytes);
+ BUG_ON(ret < 0 || search_start != *offset);
+
+ if (*offset > bitmap_info->offset && *offset + *bytes > end) {
+ bitmap_clear_bits(block_group, bitmap_info, *offset,
+ end - *offset + 1);
+ *bytes -= end - *offset + 1;
+ *offset = end + 1;
+ } else if (*offset >= bitmap_info->offset && *offset + *bytes <= end) {
+ bitmap_clear_bits(block_group, bitmap_info, *offset, *bytes);
+ *bytes = 0;
+ }
+
+ if (*bytes) {
+ struct rb_node *next = rb_next(&bitmap_info->offset_index);
+ if (!bitmap_info->bytes) {
+ unlink_free_space(block_group, bitmap_info);
+ kfree(bitmap_info->bitmap);
+ kfree(bitmap_info);
+ block_group->total_bitmaps--;
+ recalculate_thresholds(block_group);
+ }
+
+ /*
+ * no entry after this bitmap, but we still have bytes to
+ * remove, so something has gone wrong.
+ */
+ if (!next)
+ return -EINVAL;
+
+ bitmap_info = rb_entry(next, struct btrfs_free_space,
+ offset_index);
+
+ /*
+ * if the next entry isn't a bitmap we need to return to let the
+ * extent stuff do its work.
+ */
+ if (!bitmap_info->bitmap)
+ return -EAGAIN;
+
+ /*
+ * Ok the next item is a bitmap, but it may not actually hold
+ * the information for the rest of this free space stuff, so
+ * look for it, and if we don't find it return so we can try
+ * everything over again.
+ */
+ search_start = *offset;
+ search_bytes = *bytes;
+ ret = search_bitmap(block_group, bitmap_info, &search_start,
+ &search_bytes);
+ if (ret < 0 || search_start != *offset)
+ return -EAGAIN;
+
+ goto again;
+ } else if (!bitmap_info->bytes) {
+ unlink_free_space(block_group, bitmap_info);
+ kfree(bitmap_info->bitmap);
+ kfree(bitmap_info);
+ block_group->total_bitmaps--;
+ recalculate_thresholds(block_group);
+ }
+
+ return 0;
+}
+
+static int insert_into_bitmap(struct btrfs_block_group_cache *block_group,
+ struct btrfs_free_space *info)
+{
+ struct btrfs_free_space *bitmap_info;
+ int added = 0;
+ u64 bytes, offset, end;
+ int ret;
+
+ /*
+ * If we are below the extents threshold then we can add this as an
+ * extent, and don't have to deal with the bitmap
+ */
+ if (block_group->free_extents < block_group->extents_thresh &&
+ info->bytes > block_group->sectorsize * 4)
+ return 0;
+
+ /*
+ * some block groups are so tiny they can't be enveloped by a bitmap, so
+ * don't even bother to create a bitmap for this
+ */
+ if (BITS_PER_BITMAP * block_group->sectorsize >
+ block_group->key.offset)
+ return 0;
+
+ bytes = info->bytes;
+ offset = info->offset;
+
+again:
+ bitmap_info = tree_search_offset(block_group,
+ offset_to_bitmap(block_group, offset),
+ 1, 0);
+ if (!bitmap_info) {
+ BUG_ON(added);
+ goto new_bitmap;
+ }
+
+ end = bitmap_info->offset +
+ (u64)(BITS_PER_BITMAP * block_group->sectorsize);
+
+ if (offset >= bitmap_info->offset && offset + bytes > end) {
+ bitmap_set_bits(block_group, bitmap_info, offset,
+ end - offset);
+ bytes -= end - offset;
+ offset = end;
+ added = 0;
+ } else if (offset >= bitmap_info->offset && offset + bytes <= end) {
+ bitmap_set_bits(block_group, bitmap_info, offset, bytes);
+ bytes = 0;
+ } else {
+ BUG();
+ }
+
+ if (!bytes) {
+ ret = 1;
+ goto out;
+ } else
+ goto again;
+
+new_bitmap:
+ if (info && info->bitmap) {
+ add_new_bitmap(block_group, info, offset);
+ added = 1;
+ info = NULL;
+ goto again;
+ } else {
+ spin_unlock(&block_group->tree_lock);
+
+ /* no pre-allocated info, allocate a new one */
+ if (!info) {
+ info = kzalloc(sizeof(struct btrfs_free_space),
+ GFP_NOFS);
+ if (!info) {
+ spin_lock(&block_group->tree_lock);
+ ret = -ENOMEM;
+ goto out;
+ }
+ }
+
+ /* allocate the bitmap */
+ info->bitmap = kzalloc(PAGE_CACHE_SIZE, GFP_NOFS);
+ spin_lock(&block_group->tree_lock);
+ if (!info->bitmap) {
+ ret = -ENOMEM;
+ goto out;
+ }
+ goto again;
+ }
+
+out:
+ if (info) {
+ if (info->bitmap)
+ kfree(info->bitmap);
+ kfree(info);
+ }
return ret;
}
@@ -208,8 +600,8 @@ static int link_free_space(struct btrfs_block_group_cache *block_group,
int btrfs_add_free_space(struct btrfs_block_group_cache *block_group,
u64 offset, u64 bytes)
{
- struct btrfs_free_space *right_info;
- struct btrfs_free_space *left_info;
+ struct btrfs_free_space *right_info = NULL;
+ struct btrfs_free_space *left_info = NULL;
struct btrfs_free_space *info = NULL;
int ret = 0;
@@ -227,18 +619,38 @@ int btrfs_add_free_space(struct btrfs_block_group_cache *block_group,
* are adding, if there is remove that struct and add a new one to
* cover the entire range
*/
- right_info = tree_search_offset(&block_group->free_space_offset,
- offset+bytes, 0, 0);
- left_info = tree_search_offset(&block_group->free_space_offset,
- offset-1, 0, 1);
+ right_info = tree_search_offset(block_group, offset + bytes, 0, 0);
+ if (right_info && rb_prev(&right_info->offset_index))
+ left_info = rb_entry(rb_prev(&right_info->offset_index),
+ struct btrfs_free_space, offset_index);
+ else
+ left_info = tree_search_offset(block_group, offset - 1, 0, 0);
+
+ /*
+ * If there was no extent directly to the left or right of this new
+ * extent then we know we're going to have to allocate a new extent, so
+ * before we do that see if we need to drop this into a bitmap
+ */
+ if ((!left_info || left_info->bitmap) &&
+ (!right_info || right_info->bitmap)) {
+ ret = insert_into_bitmap(block_group, info);
+
+ if (ret < 0) {
+ goto out;
+ } else if (ret) {
+ ret = 0;
+ goto out;
+ }
+ }
- if (right_info) {
+ if (right_info && !right_info->bitmap) {
unlink_free_space(block_group, right_info);
info->bytes += right_info->bytes;
kfree(right_info);
}
- if (left_info && left_info->offset + left_info->bytes == offset) {
+ if (left_info && !left_info->bitmap &&
+ left_info->offset + left_info->bytes == offset) {
unlink_free_space(block_group, left_info);
info->offset = left_info->offset;
info->bytes += left_info->bytes;
@@ -248,11 +660,11 @@ int btrfs_add_free_space(struct btrfs_block_group_cache *block_group,
ret = link_free_space(block_group, info);
if (ret)
kfree(info);
-
+out:
spin_unlock(&block_group->tree_lock);
if (ret) {
- printk(KERN_ERR "btrfs: unable to add free space :%d\n", ret);
+ printk(KERN_CRIT "btrfs: unable to add free space :%d\n", ret);
BUG_ON(ret == -EEXIST);
}
@@ -263,40 +675,74 @@ int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
u64 offset, u64 bytes)
{
struct btrfs_free_space *info;
+ struct btrfs_free_space *next_info = NULL;
int ret = 0;
spin_lock(&block_group->tree_lock);
- info = tree_search_offset(&block_group->free_space_offset, offset, 0,
- 1);
- if (info && info->offset == offset) {
- if (info->bytes < bytes) {
- printk(KERN_ERR "Found free space at %llu, size %llu,"
- "trying to use %llu\n",
- (unsigned long long)info->offset,
- (unsigned long long)info->bytes,
- (unsigned long long)bytes);
+again:
+ info = tree_search_offset(block_group, offset, 0, 0);
+ if (!info) {
+ /*
+ * oops didn't find an extent that matched the space we wanted
+ * to remove, look for a bitmap instead
+ */
+ info = tree_search_offset(block_group,
+ offset_to_bitmap(block_group, offset),
+ 1, 0);
+ if (!info) {
+ WARN_ON(1);
+ goto out_lock;
+ }
+ }
+
+ if (info->bytes < bytes && rb_next(&info->offset_index)) {
+ u64 end;
+ next_info = rb_entry(rb_next(&info->offset_index),
+ struct btrfs_free_space,
+ offset_index);
+
+ if (next_info->bitmap)
+ end = next_info->offset + BITS_PER_BITMAP *
+ block_group->sectorsize - 1;
+ else
+ end = next_info->offset + next_info->bytes;
+
+ if (next_info->bytes < bytes ||
+ next_info->offset > offset || offset > end) {
+ printk(KERN_CRIT "Found free space at %llu, size %llu,"
+ " trying to use %llu\n",
+ (unsigned long long)info->offset,
+ (unsigned long long)info->bytes,
+ (unsigned long long)bytes);
WARN_ON(1);
ret = -EINVAL;
- spin_unlock(&block_group->tree_lock);
- goto out;
+ goto out_lock;
}
- unlink_free_space(block_group, info);
- if (info->bytes == bytes) {
- kfree(info);
- spin_unlock(&block_group->tree_lock);
- goto out;
+ info = next_info;
+ }
+
+ if (info->bytes == bytes) {
+ unlink_free_space(block_group, info);
+ if (info->bitmap) {
+ kfree(info->bitmap);
+ block_group->total_bitmaps--;
}
+ kfree(info);
+ goto out_lock;
+ }
+ if (!info->bitmap && info->offset == offset) {
+ unlink_free_space(block_group, info);
info->offset += bytes;
info->bytes -= bytes;
+ link_free_space(block_group, info);
+ goto out_lock;
+ }
- ret = link_free_space(block_group, info);
- spin_unlock(&block_group->tree_lock);
- BUG_ON(ret);
- } else if (info && info->offset < offset &&
- info->offset + info->bytes >= offset + bytes) {
+ if (!info->bitmap && info->offset <= offset &&
+ info->offset + info->bytes >= offset + bytes) {
u64 old_start = info->offset;
/*
* we're freeing space in the middle of the info,
@@ -312,7 +758,9 @@ int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
info->offset = offset + bytes;
info->bytes = old_end - info->offset;
ret = link_free_space(block_group, info);
- BUG_ON(ret);
+ WARN_ON(ret);
+ if (ret)
+ goto out_lock;
} else {
/* the hole we're creating ends at the end
* of the info struct, just free the info
@@ -320,32 +768,22 @@ int btrfs_remove_free_space(struct btrfs_block_group_cache *block_group,
kfree(info);
}
spin_unlock(&block_group->tree_lock);
- /* step two, insert a new info struct to cover anything
- * before the hole
+
+ /* step two, insert a new info struct to cover
+ * anything before the hole
*/
ret = btrfs_add_free_space(block_group, old_start,
offset - old_start);
- BUG_ON(ret);
- } else {
- spin_unlock(&block_group->tree_lock);
- if (!info) {
- printk(KERN_ERR "couldn't find space %llu to free\n",
- (unsigned long long)offset);
- printk(KERN_ERR "cached is %d, offset %llu bytes %llu\n",
- block_group->cached,
- (unsigned long long)block_group->key.objectid,
- (unsigned long long)block_group->key.offset);
- btrfs_dump_free_space(block_group, bytes);
- } else if (info) {
- printk(KERN_ERR "hmm, found offset=%llu bytes=%llu, "
- "but wanted offset=%llu bytes=%llu\n",
- (unsigned long long)info->offset,
- (unsigned long long)info->bytes,
- (unsigned long long)offset,
- (unsigned long long)bytes);
- }
- WARN_ON(1);
+ WARN_ON(ret);
+ goto out;
}
+
+ ret = remove_from_bitmap(block_group, info, &offset, &bytes);
+ if (ret == -EAGAIN)
+ goto again;
+ BUG_ON(ret);
+out_lock:
+ spin_unlock(&block_group->tree_lock);
out:
return ret;
}
@@ -361,10 +799,13 @@ void btrfs_dump_free_space(struct btrfs_block_group_cache *block_group,
info = rb_entry(n, struct btrfs_free_space, offset_index);
if (info->bytes >= bytes)
count++;
- printk(KERN_ERR "entry offset %llu, bytes %llu\n",
+ printk(KERN_CRIT "entry offset %llu, bytes %llu, bitmap %s\n",
(unsigned long long)info->offset,
- (unsigned long long)info->bytes);
+ (unsigned long long)info->bytes,
+ (info->bitmap) ? "yes" : "no");
}
+ printk(KERN_INFO "block group has cluster?: %s\n",
+ list_empty(&block_group->cluster_list) ? "no" : "yes");
printk(KERN_INFO "%d blocks of free space at or bigger than bytes is"
"\n", count);
}
@@ -397,26 +838,35 @@ __btrfs_return_cluster_to_free_space(
{
struct btrfs_free_space *entry;
struct rb_node *node;
+ bool bitmap;
spin_lock(&cluster->lock);
if (cluster->block_group != block_group)
goto out;
+ bitmap = cluster->points_to_bitmap;
+ cluster->block_group = NULL;
cluster->window_start = 0;
+ list_del_init(&cluster->block_group_list);
+ cluster->points_to_bitmap = false;
+
+ if (bitmap)
+ goto out;
+
node = rb_first(&cluster->root);
- while(node) {
+ while (node) {
entry = rb_entry(node, struct btrfs_free_space, offset_index);
node = rb_next(&entry->offset_index);
rb_erase(&entry->offset_index, &cluster->root);
- link_free_space(block_group, entry);
+ BUG_ON(entry->bitmap);
+ tree_insert_offset(&block_group->free_space_offset,
+ entry->offset, &entry->offset_index, 0);
}
- list_del_init(&cluster->block_group_list);
-
- btrfs_put_block_group(cluster->block_group);
- cluster->block_group = NULL;
cluster->root.rb_node = NULL;
+
out:
spin_unlock(&cluster->lock);
+ btrfs_put_block_group(block_group);
return 0;
}
@@ -425,20 +875,28 @@ void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
struct btrfs_free_space *info;
struct rb_node *node;
struct btrfs_free_cluster *cluster;
- struct btrfs_free_cluster *safe;
+ struct list_head *head;
spin_lock(&block_group->tree_lock);
-
- list_for_each_entry_safe(cluster, safe, &block_group->cluster_list,
- block_group_list) {
+ while ((head = block_group->cluster_list.next) !=
+ &block_group->cluster_list) {
+ cluster = list_entry(head, struct btrfs_free_cluster,
+ block_group_list);
WARN_ON(cluster->block_group != block_group);
__btrfs_return_cluster_to_free_space(block_group, cluster);
+ if (need_resched()) {
+ spin_unlock(&block_group->tree_lock);
+ cond_resched();
+ spin_lock(&block_group->tree_lock);
+ }
}
- while ((node = rb_last(&block_group->free_space_bytes)) != NULL) {
- info = rb_entry(node, struct btrfs_free_space, bytes_index);
+ while ((node = rb_last(&block_group->free_space_offset)) != NULL) {
+ info = rb_entry(node, struct btrfs_free_space, offset_index);
unlink_free_space(block_group, info);
+ if (info->bitmap)
+ kfree(info->bitmap);
kfree(info);
if (need_resched()) {
spin_unlock(&block_group->tree_lock);
@@ -446,6 +904,7 @@ void btrfs_remove_free_space_cache(struct btrfs_block_group_cache *block_group)
spin_lock(&block_group->tree_lock);
}
}
+
spin_unlock(&block_group->tree_lock);
}
@@ -453,25 +912,35 @@ u64 btrfs_find_space_for_alloc(struct btrfs_block_group_cache *block_group,
u64 offset, u64 bytes, u64 empty_size)
{
struct btrfs_free_space *entry = NULL;
+ u64 bytes_search = bytes + empty_size;
u64 ret = 0;
spin_lock(&block_group->tree_lock);
- entry = tree_search_offset(&block_group->free_space_offset, offset,
- bytes + empty_size, 1);
+ entry = find_free_space(block_group, &offset, &bytes_search, 0);
if (!entry)
- entry = tree_search_bytes(&block_group->free_space_bytes,
- offset, bytes + empty_size);
- if (entry) {
+ goto out;
+
+ ret = offset;
+ if (entry->bitmap) {
+ bitmap_clear_bits(block_group, entry, offset, bytes);
+ if (!entry->bytes) {
+ unlink_free_space(block_group, entry);
+ kfree(entry->bitmap);
+ kfree(entry);
+ block_group->total_bitmaps--;
+ recalculate_thresholds(block_group);
+ }
+ } else {
unlink_free_space(block_group, entry);
- ret = entry->offset;
entry->offset += bytes;
entry->bytes -= bytes;
-
if (!entry->bytes)
kfree(entry);
else
link_free_space(block_group, entry);
}
+
+out:
spin_unlock(&block_group->tree_lock);
return ret;
@@ -517,6 +986,54 @@ int btrfs_return_cluster_to_free_space(
return ret;
}
+static u64 btrfs_alloc_from_bitmap(struct btrfs_block_group_cache *block_group,
+ struct btrfs_free_cluster *cluster,
+ u64 bytes, u64 min_start)
+{
+ struct btrfs_free_space *entry;
+ int err;
+ u64 search_start = cluster->window_start;
+ u64 search_bytes = bytes;
+ u64 ret = 0;
+
+ spin_lock(&block_group->tree_lock);
+ spin_lock(&cluster->lock);
+
+ if (!cluster->points_to_bitmap)
+ goto out;
+
+ if (cluster->block_group != block_group)
+ goto out;
+
+ /*
+ * search_start is the beginning of the bitmap, but at some point it may
+ * be a good idea to point to the actual start of the free area in the
+ * bitmap, so do the offset_to_bitmap trick anyway, and set bitmap_only
+ * to 1 to make sure we get the bitmap entry
+ */
+ entry = tree_search_offset(block_group,
+ offset_to_bitmap(block_group, search_start),
+ 1, 0);
+ if (!entry || !entry->bitmap)
+ goto out;
+
+ search_start = min_start;
+ search_bytes = bytes;
+
+ err = search_bitmap(block_group, entry, &search_start,
+ &search_bytes);
+ if (err)
+ goto out;
+
+ ret = search_start;
+ bitmap_clear_bits(block_group, entry, ret, bytes);
+out:
+ spin_unlock(&cluster->lock);
+ spin_unlock(&block_group->tree_lock);
+
+ return ret;
+}
+
/*
* given a cluster, try to allocate 'bytes' from it, returns 0
* if it couldn't find anything suitably large, or a logical disk offset
@@ -530,6 +1047,10 @@ u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
struct rb_node *node;
u64 ret = 0;
+ if (cluster->points_to_bitmap)
+ return btrfs_alloc_from_bitmap(block_group, cluster, bytes,
+ min_start);
+
spin_lock(&cluster->lock);
if (bytes > cluster->max_size)
goto out;
@@ -567,9 +1088,73 @@ u64 btrfs_alloc_from_cluster(struct btrfs_block_group_cache *block_group,
}
out:
spin_unlock(&cluster->lock);
+
return ret;
}
+static int btrfs_bitmap_cluster(struct btrfs_block_group_cache *block_group,
+ struct btrfs_free_space *entry,
+ struct btrfs_free_cluster *cluster,
+ u64 offset, u64 bytes, u64 min_bytes)
+{
+ unsigned long next_zero;
+ unsigned long i;
+ unsigned long search_bits;
+ unsigned long total_bits;
+ unsigned long found_bits;
+ unsigned long start = 0;
+ unsigned long total_found = 0;
+ bool found = false;
+
+ i = offset_to_bit(entry->offset, block_group->sectorsize,
+ max_t(u64, offset, entry->offset));
+ search_bits = bytes_to_bits(min_bytes, block_group->sectorsize);
+ total_bits = bytes_to_bits(bytes, block_group->sectorsize);
+
+again:
+ found_bits = 0;
+ for (i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i);
+ i < BITS_PER_BITMAP;
+ i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, i + 1)) {
+ next_zero = find_next_zero_bit(entry->bitmap,
+ BITS_PER_BITMAP, i);
+ if (next_zero - i >= search_bits) {
+ found_bits = next_zero - i;
+ break;
+ }
+ i = next_zero;
+ }
+
+ if (!found_bits)
+ return -1;
+
+ if (!found) {
+ start = i;
+ found = true;
+ }
+
+ total_found += found_bits;
+
+ if (cluster->max_size < found_bits * block_group->sectorsize)
+ cluster->max_size = found_bits * block_group->sectorsize;
+
+ if (total_found < total_bits) {
+ i = find_next_bit(entry->bitmap, BITS_PER_BITMAP, next_zero);
+ if (i - start > total_bits * 2) {
+ total_found = 0;
+ cluster->max_size = 0;
+ found = false;
+ }
+ goto again;
+ }
+
+ cluster->window_start = start * block_group->sectorsize +
+ entry->offset;
+ cluster->points_to_bitmap = true;
+
+ return 0;
+}
+
/*
* here we try to find a cluster of blocks in a block group. The goal
* is to find at least bytes free and up to empty_size + bytes free.
@@ -587,12 +1172,12 @@ int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
struct btrfs_free_space *entry = NULL;
struct rb_node *node;
struct btrfs_free_space *next;
- struct btrfs_free_space *last;
+ struct btrfs_free_space *last = NULL;
u64 min_bytes;
u64 window_start;
u64 window_free;
u64 max_extent = 0;
- int total_retries = 0;
+ bool found_bitmap = false;
int ret;
/* for metadata, allow allocates with more holes */
@@ -620,31 +1205,80 @@ int btrfs_find_space_cluster(struct btrfs_trans_handle *trans,
goto out;
}
again:
- min_bytes = min(min_bytes, bytes + empty_size);
- entry = tree_search_bytes(&block_group->free_space_bytes,
- offset, min_bytes);
+ entry = tree_search_offset(block_group, offset, found_bitmap, 1);
if (!entry) {
ret = -ENOSPC;
goto out;
}
+
+ /*
+ * If found_bitmap is true, we exhausted our search for extent entries,
+ * and we just want to search all of the bitmaps that we can find, and
+ * ignore any extent entries we find.
+ */
+ while (entry->bitmap || found_bitmap ||
+ (!entry->bitmap && entry->bytes < min_bytes)) {
+ struct rb_node *node = rb_next(&entry->offset_index);
+
+ if (entry->bitmap && entry->bytes > bytes + empty_size) {
+ ret = btrfs_bitmap_cluster(block_group, entry, cluster,
+ offset, bytes + empty_size,
+ min_bytes);
+ if (!ret)
+ goto got_it;
+ }
+
+ if (!node) {
+ ret = -ENOSPC;
+ goto out;
+ }
+ entry = rb_entry(node, struct btrfs_free_space, offset_index);
+ }
+
+ /*
+ * We already searched all the extent entries from the passed in offset
+ * to the end and didn't find enough space for the cluster, and we also
+ * didn't find any bitmaps that met our criteria, just go ahead and exit
+ */
+ if (found_bitmap) {
+ ret = -ENOSPC;
+ goto out;
+ }
+
+ cluster->points_to_bitmap = false;
window_start = entry->offset;
window_free = entry->bytes;
last = entry;
max_extent = entry->bytes;
- while(1) {
+ while (1) {
/* out window is just right, lets fill it */
if (window_free >= bytes + empty_size)
break;
node = rb_next(&last->offset_index);
if (!node) {
+ if (found_bitmap)
+ goto again;
ret = -ENOSPC;
goto out;
}
next = rb_entry(node, struct btrfs_free_space, offset_index);
/*
+ * we found a bitmap, so if this search doesn't result in a
+ * cluster, we know to go and search again for the bitmaps and
+ * start looking for space there
+ */
+ if (next->bitmap) {
+ if (!found_bitmap)
+ offset = next->offset;
+ found_bitmap = true;
+ last = next;
+ continue;
+ }
+
+ /*
* we haven't filled the empty size and the window is
* very large. reset and try again
*/
@@ -655,19 +1289,6 @@ again:
window_free = entry->bytes;
last = entry;
max_extent = 0;
- total_retries++;
- if (total_retries % 64 == 0) {
- if (min_bytes >= (bytes + empty_size)) {
- ret = -ENOSPC;
- goto out;
- }
- /*
- * grow our allocation a bit, we're not having
- * much luck
- */
- min_bytes *= 2;
- goto again;
- }
} else {
last = next;
window_free += next->bytes;
@@ -685,11 +1306,19 @@ again:
* The cluster includes an rbtree, but only uses the offset index
* of each free space cache entry.
*/
- while(1) {
+ while (1) {
node = rb_next(&entry->offset_index);
- unlink_free_space(block_group, entry);
+ if (entry->bitmap && node) {
+ entry = rb_entry(node, struct btrfs_free_space,
+ offset_index);
+ continue;
+ } else if (entry->bitmap && !node) {
+ break;
+ }
+
+ rb_erase(&entry->offset_index, &block_group->free_space_offset);
ret = tree_insert_offset(&cluster->root, entry->offset,
- &entry->offset_index);
+ &entry->offset_index, 0);
BUG_ON(ret);
if (!node || entry == last)
@@ -697,8 +1326,10 @@ again:
entry = rb_entry(node, struct btrfs_free_space, offset_index);
}
- ret = 0;
+
cluster->max_size = max_extent;
+got_it:
+ ret = 0;
atomic_inc(&block_group->count);
list_add_tail(&cluster->block_group_list, &block_group->cluster_list);
cluster->block_group = block_group;
@@ -718,6 +1349,7 @@ void btrfs_init_free_cluster(struct btrfs_free_cluster *cluster)
spin_lock_init(&cluster->refill_lock);
cluster->root.rb_node = NULL;
cluster->max_size = 0;
+ cluster->points_to_bitmap = false;
INIT_LIST_HEAD(&cluster->block_group_list);
cluster->block_group = NULL;
}
diff --git a/fs/btrfs/free-space-cache.h b/fs/btrfs/free-space-cache.h
index 266fb876405..890a8e79011 100644
--- a/fs/btrfs/free-space-cache.h
+++ b/fs/btrfs/free-space-cache.h
@@ -19,6 +19,14 @@
#ifndef __BTRFS_FREE_SPACE_CACHE
#define __BTRFS_FREE_SPACE_CACHE
+struct btrfs_free_space {
+ struct rb_node offset_index;
+ u64 offset;
+ u64 bytes;
+ unsigned long *bitmap;
+ struct list_head list;
+};
+
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 8612b3a0981..272b9b2bea8 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -26,7 +26,6 @@
#include <linux/time.h>
#include <linux/init.h>
#include <linux/string.h>
-#include <linux/smp_lock.h>
#include <linux/backing-dev.h>
#include <linux/mpage.h>
#include <linux/swap.h>
@@ -2122,10 +2121,8 @@ static void btrfs_read_locked_inode(struct inode *inode)
* any xattrs or acls
*/
maybe_acls = acls_after_inode_item(leaf, path->slots[0], inode->i_ino);
- if (!maybe_acls) {
- BTRFS_I(inode)->i_acl = NULL;
- BTRFS_I(inode)->i_default_acl = NULL;
- }
+ if (!maybe_acls)
+ cache_no_acl(inode);
BTRFS_I(inode)->block_group = btrfs_find_block_group(root, 0,
alloc_group_block, 0);
@@ -2606,8 +2603,8 @@ noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
if (root->ref_cows)
btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
path = btrfs_alloc_path();
- path->reada = -1;
BUG_ON(!path);
+ path->reada = -1;
/* FIXME, add redo link to tree so we don't leak on crash */
key.objectid = inode->i_ino;
@@ -3141,9 +3138,6 @@ static noinline void init_btrfs_i(struct inode *inode)
{
struct btrfs_inode *bi = BTRFS_I(inode);
- bi->i_acl = BTRFS_ACL_NOT_CACHED;
- bi->i_default_acl = BTRFS_ACL_NOT_CACHED;
-
bi->generation = 0;
bi->sequence = 0;
bi->last_trans = 0;
@@ -3585,12 +3579,6 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
owner = 1;
BTRFS_I(inode)->block_group =
btrfs_find_block_group(root, 0, alloc_hint, owner);
- if ((mode & S_IFREG)) {
- if (btrfs_test_opt(root, NODATASUM))
- BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
- if (btrfs_test_opt(root, NODATACOW))
- BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
- }
key[0].objectid = objectid;
btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
@@ -3645,6 +3633,13 @@ static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
btrfs_inherit_iflags(inode, dir);
+ if ((mode & S_IFREG)) {
+ if (btrfs_test_opt(root, NODATASUM))
+ BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
+ if (btrfs_test_opt(root, NODATACOW))
+ BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
+ }
+
insert_inode_hash(inode);
inode_tree_add(inode);
return inode;
@@ -4640,8 +4635,6 @@ struct inode *btrfs_alloc_inode(struct super_block *sb)
ei->last_trans = 0;
ei->logged_trans = 0;
btrfs_ordered_inode_tree_init(&ei->ordered_tree);
- ei->i_acl = BTRFS_ACL_NOT_CACHED;
- ei->i_default_acl = BTRFS_ACL_NOT_CACHED;
INIT_LIST_HEAD(&ei->i_orphan);
INIT_LIST_HEAD(&ei->ordered_operations);
return &ei->vfs_inode;
@@ -4655,13 +4648,6 @@ void btrfs_destroy_inode(struct inode *inode)
WARN_ON(!list_empty(&inode->i_dentry));
WARN_ON(inode->i_data.nrpages);
- if (BTRFS_I(inode)->i_acl &&
- BTRFS_I(inode)->i_acl != BTRFS_ACL_NOT_CACHED)
- posix_acl_release(BTRFS_I(inode)->i_acl);
- if (BTRFS_I(inode)->i_default_acl &&
- BTRFS_I(inode)->i_default_acl != BTRFS_ACL_NOT_CACHED)
- posix_acl_release(BTRFS_I(inode)->i_default_acl);
-
/*
* Make sure we're properly removed from the ordered operation
* lists.
@@ -4799,8 +4785,7 @@ static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
* and the replacement file is large. Start IO on it now so
* we don't add too much work to the end of the transaction
*/
- if (new_inode && old_inode && S_ISREG(old_inode->i_mode) &&
- new_inode->i_size &&
+ if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
filemap_flush(old_inode->i_mapping);
@@ -5096,6 +5081,7 @@ static long btrfs_fallocate(struct inode *inode, int mode,
u64 mask = BTRFS_I(inode)->root->sectorsize - 1;
struct extent_map *em;
struct btrfs_trans_handle *trans;
+ struct btrfs_root *root;
int ret;
alloc_start = offset & ~mask;
@@ -5114,6 +5100,13 @@ static long btrfs_fallocate(struct inode *inode, int mode,
goto out;
}
+ root = BTRFS_I(inode)->root;
+
+ ret = btrfs_check_data_free_space(root, inode,
+ alloc_end - alloc_start);
+ if (ret)
+ goto out;
+
locked_end = alloc_end - 1;
while (1) {
struct btrfs_ordered_extent *ordered;
@@ -5121,7 +5114,7 @@ static long btrfs_fallocate(struct inode *inode, int mode,
trans = btrfs_start_transaction(BTRFS_I(inode)->root, 1);
if (!trans) {
ret = -EIO;
- goto out;
+ goto out_free;
}
/* the extent lock is ordered inside the running
@@ -5182,6 +5175,8 @@ static long btrfs_fallocate(struct inode *inode, int mode,
GFP_NOFS);
btrfs_end_transaction(trans, BTRFS_I(inode)->root);
+out_free:
+ btrfs_free_reserved_data_space(root, inode, alloc_end - alloc_start);
out:
mutex_unlock(&inode->i_mutex);
return ret;
diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index eff18f5b536..bd88f25889f 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -27,7 +27,6 @@
#include <linux/time.h>
#include <linux/init.h>
#include <linux/string.h>
-#include <linux/smp_lock.h>
#include <linux/backing-dev.h>
#include <linux/mount.h>
#include <linux/mpage.h>
@@ -1028,7 +1027,8 @@ static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
struct btrfs_file_extent_item);
comp = btrfs_file_extent_compression(leaf, extent);
type = btrfs_file_extent_type(leaf, extent);
- if (type == BTRFS_FILE_EXTENT_REG) {
+ if (type == BTRFS_FILE_EXTENT_REG ||
+ type == BTRFS_FILE_EXTENT_PREALLOC) {
disko = btrfs_file_extent_disk_bytenr(leaf,
extent);
diskl = btrfs_file_extent_disk_num_bytes(leaf,
@@ -1051,7 +1051,8 @@ static long btrfs_ioctl_clone(struct file *file, unsigned long srcfd,
new_key.objectid = inode->i_ino;
new_key.offset = key.offset + destoff - off;
- if (type == BTRFS_FILE_EXTENT_REG) {
+ if (type == BTRFS_FILE_EXTENT_REG ||
+ type == BTRFS_FILE_EXTENT_PREALLOC) {
ret = btrfs_insert_empty_item(trans, root, path,
&new_key, size);
if (ret)
diff --git a/fs/btrfs/print-tree.c b/fs/btrfs/print-tree.c
index 6d6523da0a3..0d126be22b6 100644
--- a/fs/btrfs/print-tree.c
+++ b/fs/btrfs/print-tree.c
@@ -309,7 +309,7 @@ void btrfs_print_tree(struct btrfs_root *root, struct extent_buffer *c)
}
printk(KERN_INFO "node %llu level %d total ptrs %d free spc %u\n",
(unsigned long long)btrfs_header_bytenr(c),
- btrfs_header_level(c), nr,
+ level, nr,
(u32)BTRFS_NODEPTRS_PER_BLOCK(root) - nr);
for (i = 0; i < nr; i++) {
btrfs_node_key_to_cpu(c, &key, i);
@@ -326,10 +326,10 @@ void btrfs_print_tree(struct btrfs_root *root, struct extent_buffer *c)
btrfs_level_size(root, level - 1),
btrfs_node_ptr_generation(c, i));
if (btrfs_is_leaf(next) &&
- btrfs_header_level(c) != 1)
+ level != 1)
BUG();
if (btrfs_header_level(next) !=
- btrfs_header_level(c) - 1)
+ level - 1)
BUG();
btrfs_print_tree(root, next);
free_extent_buffer(next);
diff --git a/fs/btrfs/relocation.c b/fs/btrfs/relocation.c
index b23dc209ae1..c04f7f21260 100644
--- a/fs/btrfs/relocation.c
+++ b/fs/btrfs/relocation.c
@@ -670,6 +670,8 @@ again:
err = ret;
goto out;
}
+ if (ret > 0 && path2->slots[level] > 0)
+ path2->slots[level]--;
eb = path2->nodes[level];
WARN_ON(btrfs_node_blockptr(eb, path2->slots[level]) !=
@@ -1609,6 +1611,7 @@ static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
BUG_ON(level == 0);
path->lowest_level = level;
ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0);
+ path->lowest_level = 0;
if (ret < 0) {
btrfs_free_path(path);
return ret;
@@ -1788,7 +1791,7 @@ static void merge_func(struct btrfs_work *work)
btrfs_end_transaction(trans, root);
}
- btrfs_drop_dead_root(reloc_root);
+ btrfs_drop_snapshot(reloc_root, 0);
if (atomic_dec_and_test(async->num_pending))
complete(async->done);
@@ -2075,9 +2078,6 @@ static int do_relocation(struct btrfs_trans_handle *trans,
ret = btrfs_drop_subtree(trans, root, eb, upper->eb);
BUG_ON(ret);
-
- btrfs_tree_unlock(eb);
- free_extent_buffer(eb);
}
if (!lowest) {
btrfs_tree_unlock(upper->eb);
@@ -2553,8 +2553,13 @@ int relocate_inode_pages(struct inode *inode, u64 start, u64 len)
last_index = (start + len - 1) >> PAGE_CACHE_SHIFT;
/* make sure the dirty trick played by the caller work */
- ret = invalidate_inode_pages2_range(inode->i_mapping,
- first_index, last_index);
+ while (1) {
+ ret = invalidate_inode_pages2_range(inode->i_mapping,
+ first_index, last_index);
+ if (ret != -EBUSY)
+ break;
+ schedule_timeout(HZ/10);
+ }
if (ret)
goto out_unlock;
diff --git a/fs/btrfs/super.c b/fs/btrfs/super.c
index 9f179d4832d..6d6d06cb6df 100644
--- a/fs/btrfs/super.c
+++ b/fs/btrfs/super.c
@@ -26,7 +26,6 @@
#include <linux/init.h>
#include <linux/seq_file.h>
#include <linux/string.h>
-#include <linux/smp_lock.h>
#include <linux/backing-dev.h>
#include <linux/mount.h>
#include <linux/mpage.h>
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 4e83457ea25..cdbb5022da5 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -40,6 +40,12 @@ static noinline void put_transaction(struct btrfs_transaction *transaction)
}
}
+static noinline void switch_commit_root(struct btrfs_root *root)
+{
+ free_extent_buffer(root->commit_root);
+ root->commit_root = btrfs_root_node(root);
+}
+
/*
* either allocate a new transaction or hop into the existing one
*/
@@ -444,9 +450,6 @@ static int update_cowonly_root(struct btrfs_trans_handle *trans,
btrfs_write_dirty_block_groups(trans, root);
- ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
- BUG_ON(ret);
-
while (1) {
old_root_bytenr = btrfs_root_bytenr(&root->root_item);
if (old_root_bytenr == root->node->start)
@@ -457,13 +460,14 @@ static int update_cowonly_root(struct btrfs_trans_handle *trans,
&root->root_key,
&root->root_item);
BUG_ON(ret);
- btrfs_write_dirty_block_groups(trans, root);
- ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
+ ret = btrfs_write_dirty_block_groups(trans, root);
BUG_ON(ret);
}
- free_extent_buffer(root->commit_root);
- root->commit_root = btrfs_root_node(root);
+
+ if (root != root->fs_info->extent_root)
+ switch_commit_root(root);
+
return 0;
}
@@ -495,10 +499,12 @@ static noinline int commit_cowonly_roots(struct btrfs_trans_handle *trans,
root = list_entry(next, struct btrfs_root, dirty_list);
update_cowonly_root(trans, root);
-
- ret = btrfs_run_delayed_refs(trans, root, (unsigned long)-1);
- BUG_ON(ret);
}
+
+ down_write(&fs_info->extent_commit_sem);
+ switch_commit_root(fs_info->extent_root);
+ up_write(&fs_info->extent_commit_sem);
+
return 0;
}
@@ -544,8 +550,7 @@ static noinline int commit_fs_roots(struct btrfs_trans_handle *trans,
btrfs_update_reloc_root(trans, root);
if (root->commit_root != root->node) {
- free_extent_buffer(root->commit_root);
- root->commit_root = btrfs_root_node(root);
+ switch_commit_root(root);
btrfs_set_root_node(&root->root_item,
root->node);
}
@@ -593,6 +598,7 @@ int btrfs_defrag_root(struct btrfs_root *root, int cacheonly)
return 0;
}
+#if 0
/*
* when dropping snapshots, we generate a ton of delayed refs, and it makes
* sense not to join the transaction while it is trying to flush the current
@@ -681,6 +687,7 @@ int btrfs_drop_dead_root(struct btrfs_root *root)
btrfs_btree_balance_dirty(tree_root, nr);
return ret;
}
+#endif
/*
* new snapshots need to be created at a very specific time in the
@@ -850,6 +857,16 @@ static void update_super_roots(struct btrfs_root *root)
super->root_level = root_item->level;
}
+int btrfs_transaction_in_commit(struct btrfs_fs_info *info)
+{
+ int ret = 0;
+ spin_lock(&info->new_trans_lock);
+ if (info->running_transaction)
+ ret = info->running_transaction->in_commit;
+ spin_unlock(&info->new_trans_lock);
+ return ret;
+}
+
int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
struct btrfs_root *root)
{
@@ -941,9 +958,11 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
mutex_unlock(&root->fs_info->trans_mutex);
- if (flush_on_commit || snap_pending) {
- if (flush_on_commit)
- btrfs_start_delalloc_inodes(root);
+ if (flush_on_commit) {
+ btrfs_start_delalloc_inodes(root);
+ ret = btrfs_wait_ordered_extents(root, 0);
+ BUG_ON(ret);
+ } else if (snap_pending) {
ret = btrfs_wait_ordered_extents(root, 1);
BUG_ON(ret);
}
@@ -1007,15 +1026,11 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
btrfs_set_root_node(&root->fs_info->tree_root->root_item,
root->fs_info->tree_root->node);
- free_extent_buffer(root->fs_info->tree_root->commit_root);
- root->fs_info->tree_root->commit_root =
- btrfs_root_node(root->fs_info->tree_root);
+ switch_commit_root(root->fs_info->tree_root);
btrfs_set_root_node(&root->fs_info->chunk_root->root_item,
root->fs_info->chunk_root->node);
- free_extent_buffer(root->fs_info->chunk_root->commit_root);
- root->fs_info->chunk_root->commit_root =
- btrfs_root_node(root->fs_info->chunk_root);
+ switch_commit_root(root->fs_info->chunk_root);
update_super_roots(root);
@@ -1055,6 +1070,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
cur_trans->commit_done = 1;
root->fs_info->last_trans_committed = cur_trans->transid;
+
wake_up(&cur_trans->commit_wait);
put_transaction(cur_trans);
@@ -1081,7 +1097,7 @@ int btrfs_clean_old_snapshots(struct btrfs_root *root)
while (!list_empty(&list)) {
root = list_entry(list.next, struct btrfs_root, root_list);
list_del_init(&root->root_list);
- btrfs_drop_dead_root(root);
+ btrfs_drop_snapshot(root, 0);
}
return 0;
}
diff --git a/fs/btrfs/transaction.h b/fs/btrfs/transaction.h
index 961c3ee5a2e..663c6740491 100644
--- a/fs/btrfs/transaction.h
+++ b/fs/btrfs/transaction.h
@@ -107,4 +107,5 @@ int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
struct btrfs_root *root);
int btrfs_write_and_wait_marked_extents(struct btrfs_root *root,
struct extent_io_tree *dirty_pages);
+int btrfs_transaction_in_commit(struct btrfs_fs_info *info);
#endif
diff --git a/fs/btrfs/tree-log.c b/fs/btrfs/tree-log.c
index c13922206d1..d91b0de7c50 100644
--- a/fs/btrfs/tree-log.c
+++ b/fs/btrfs/tree-log.c
@@ -797,7 +797,7 @@ static noinline int add_inode_ref(struct btrfs_trans_handle *trans,
return -ENOENT;
inode = read_one_inode(root, key->objectid);
- BUG_ON(!dir);
+ BUG_ON(!inode);
ref_ptr = btrfs_item_ptr_offset(eb, slot);
ref_end = ref_ptr + btrfs_item_size_nr(eb, slot);
diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index 3ab80e9cd76..5dbefd11b4a 100644
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -721,7 +721,8 @@ error:
*/
static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
struct btrfs_device *device,
- u64 num_bytes, u64 *start)
+ u64 num_bytes, u64 *start,
+ u64 *max_avail)
{
struct btrfs_key key;
struct btrfs_root *root = device->dev_root;
@@ -758,9 +759,13 @@ static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans,
ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
if (ret < 0)
goto error;
- ret = btrfs_previous_item(root, path, 0, key.type);
- if (ret < 0)
- goto error;
+ if (ret > 0) {
+ ret = btrfs_previous_item(root, path, key.objectid, key.type);
+ if (ret < 0)
+ goto error;
+ if (ret > 0)
+ start_found = 1;
+ }
l = path->nodes[0];
btrfs_item_key_to_cpu(l, &key, path->slots[0]);
while (1) {
@@ -803,6 +808,10 @@ no_more_items:
if (last_byte < search_start)
last_byte = search_start;
hole_size = key.offset - last_byte;
+
+ if (hole_size > *max_avail)
+ *max_avail = hole_size;
+
if (key.offset > last_byte &&
hole_size >= num_bytes) {
*start = last_byte;
@@ -1621,6 +1630,7 @@ static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
device->fs_devices->total_rw_bytes += diff;
device->total_bytes = new_size;
+ device->disk_total_bytes = new_size;
btrfs_clear_space_info_full(device->dev_root->fs_info);
return btrfs_update_device(trans, device);
@@ -2007,7 +2017,7 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
goto done;
if (ret) {
ret = 0;
- goto done;
+ break;
}
l = path->nodes[0];
@@ -2015,7 +2025,7 @@ int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
btrfs_item_key_to_cpu(l, &key, path->slots[0]);
if (key.objectid != device->devid)
- goto done;
+ break;
dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
length = btrfs_dev_extent_length(l, dev_extent);
@@ -2171,6 +2181,7 @@ static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
max_chunk_size);
again:
+ max_avail = 0;
if (!map || map->num_stripes != num_stripes) {
kfree(map);
map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
@@ -2219,7 +2230,8 @@ again:
if (device->in_fs_metadata && avail >= min_free) {
ret = find_free_dev_extent(trans, device,
- min_free, &dev_offset);
+ min_free, &dev_offset,
+ &max_avail);
if (ret == 0) {
list_move_tail(&device->dev_alloc_list,
&private_devs);
@@ -2795,26 +2807,6 @@ int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
}
}
- for (i = 0; i > nr; i++) {
- struct btrfs_multi_bio *multi;
- struct btrfs_bio_stripe *stripe;
- int ret;
-
- length = 1;
- ret = btrfs_map_block(map_tree, WRITE, buf[i],
- &length, &multi, 0);
- BUG_ON(ret);
-
- stripe = multi->stripes;
- for (j = 0; j < multi->num_stripes; j++) {
- if (stripe->physical >= physical &&
- physical < stripe->physical + length)
- break;
- }
- BUG_ON(j >= multi->num_stripes);
- kfree(multi);
- }
-
*logical = buf;
*naddrs = nr;
*stripe_len = map->stripe_len;
diff --git a/fs/btrfs/zlib.c b/fs/btrfs/zlib.c
index ecfbce836d3..3e2b90eaa23 100644
--- a/fs/btrfs/zlib.c
+++ b/fs/btrfs/zlib.c
@@ -208,7 +208,7 @@ int btrfs_zlib_compress_pages(struct address_space *mapping,
*total_in = 0;
workspace = find_zlib_workspace();
- if (!workspace)
+ if (IS_ERR(workspace))
return -1;
if (Z_OK != zlib_deflateInit(&workspace->def_strm, 3)) {
@@ -366,7 +366,7 @@ int btrfs_zlib_decompress_biovec(struct page **pages_in,
char *kaddr;
workspace = find_zlib_workspace();
- if (!workspace)
+ if (IS_ERR(workspace))
return -ENOMEM;
data_in = kmap(pages_in[page_in_index]);
@@ -547,7 +547,7 @@ int btrfs_zlib_decompress(unsigned char *data_in,
return -ENOMEM;
workspace = find_zlib_workspace();
- if (!workspace)
+ if (IS_ERR(workspace))
return -ENOMEM;
workspace->inf_strm.next_in = data_in;
generated by cgit v1.2.3-70-g09d2 (git 2.46.0) at 2024-12-11 23:39:16 +0000