Btrfs: early reference counting
Signed-off-by: Chris Mason <chris.mason@oracle.com>
diff --git a/fs/btrfs/ctree.c b/fs/btrfs/ctree.c
index 3c5f4c2..1118986 100644
--- a/fs/btrfs/ctree.c
+++ b/fs/btrfs/ctree.c
@@ -34,6 +34,37 @@
memset(p, 0, sizeof(*p));
}
+int btrfs_cow_block(struct ctree_root *root,
+ struct tree_buffer *buf,
+ struct tree_buffer *parent,
+ int parent_slot,
+ struct tree_buffer **cow_ret)
+{
+ struct tree_buffer *cow;
+
+ if (!list_empty(&buf->dirty)) {
+ *cow_ret = buf;
+ return 0;
+ }
+ cow = alloc_free_block(root);
+ memcpy(&cow->node, &buf->node, sizeof(buf->node));
+ cow->node.header.blocknr = cow->blocknr;
+ *cow_ret = cow;
+ if (buf == root->node) {
+ root->node = cow;
+ cow->count++;
+ tree_block_release(root, buf);
+ } else {
+ parent->node.blockptrs[parent_slot] = cow->blocknr;
+ BUG_ON(list_empty(&parent->dirty));
+ }
+ if (0 && root != root->extent_root && !is_leaf(cow->node.header.flags)) {
+ btrfs_inc_ref(root, cow);
+ }
+ tree_block_release(root, buf);
+ return 0;
+}
+
/*
* The leaf data grows from end-to-front in the node.
* this returns the address of the start of the last item,
@@ -263,6 +294,8 @@
/* first, try to make some room in the middle buffer */
if (left_buf) {
+ btrfs_cow_block(root, left_buf, parent_buf,
+ pslot - 1, &left_buf);
left = &left_buf->node;
orig_slot += left->header.nritems;
wret = push_node_left(root, left_buf, mid_buf);
@@ -274,6 +307,8 @@
* then try to empty the right most buffer into the middle
*/
if (right_buf) {
+ btrfs_cow_block(root, right_buf, parent_buf,
+ pslot + 1, &right_buf);
right = &right_buf->node;
wret = push_node_left(root, mid_buf, right_buf);
if (wret < 0)
@@ -293,9 +328,7 @@
} else {
memcpy(parent->keys + pslot + 1, right->keys,
sizeof(struct key));
- wret = dirty_tree_block(root, parent_buf);
- if (wret)
- ret = wret;
+ BUG_ON(list_empty(&parent_buf->dirty));
}
}
if (mid->header.nritems == 1) {
@@ -330,9 +363,7 @@
} else {
/* update the parent key to reflect our changes */
memcpy(parent->keys + pslot, mid->keys, sizeof(struct key));
- wret = dirty_tree_block(root, parent_buf);
- if (wret)
- ret = wret;
+ BUG_ON(list_empty(&parent_buf->dirty));
}
/* update the path */
@@ -375,9 +406,10 @@
* possible)
*/
int search_slot(struct ctree_root *root, struct key *key,
- struct ctree_path *p, int ins_len)
+ struct ctree_path *p, int ins_len, int cow)
{
struct tree_buffer *b;
+ struct tree_buffer *cow_buf;
struct node *c;
int slot;
int ret;
@@ -387,8 +419,15 @@
b = root->node;
b->count++;
while (b) {
+ level = node_level(b->node.header.flags);
+ if (cow) {
+ int wret;
+ wret = btrfs_cow_block(root, b, p->nodes[level + 1],
+ p->slots[level + 1], &cow_buf);
+ b = cow_buf;
+ }
+ BUG_ON(!cow && ins_len);
c = &b->node;
- level = node_level(c->header.flags);
p->nodes[level] = b;
ret = check_block(p, level);
if (ret)
@@ -453,7 +492,6 @@
{
int i;
int ret = 0;
- int wret;
for (i = level; i < MAX_LEVEL; i++) {
struct node *t;
int tslot = path->slots[i];
@@ -461,9 +499,7 @@
break;
t = &path->nodes[i]->node;
memcpy(t->keys + tslot, key, sizeof(*key));
- wret = dirty_tree_block(root, path->nodes[i]);
- if (wret)
- ret = wret;
+ BUG_ON(list_empty(&path->nodes[i]->dirty));
if (tslot != 0)
break;
}
@@ -486,7 +522,6 @@
int src_nritems;
int dst_nritems;
int ret = 0;
- int wret;
src_nritems = src->header.nritems;
dst_nritems = dst->header.nritems;
@@ -511,13 +546,8 @@
src->header.nritems -= push_items;
dst->header.nritems += push_items;
- wret = dirty_tree_block(root, src_buf);
- if (wret < 0)
- ret = wret;
-
- wret = dirty_tree_block(root, dst_buf);
- if (wret < 0)
- ret = wret;
+ BUG_ON(list_empty(&src_buf->dirty));
+ BUG_ON(list_empty(&dst_buf->dirty));
return ret;
}
@@ -541,7 +571,6 @@
int src_nritems;
int dst_nritems;
int ret = 0;
- int wret;
src_nritems = src->header.nritems;
dst_nritems = dst->header.nritems;
@@ -569,13 +598,8 @@
src->header.nritems -= push_items;
dst->header.nritems += push_items;
- wret = dirty_tree_block(root, src_buf);
- if (wret < 0)
- ret = wret;
-
- wret = dirty_tree_block(root, dst_buf);
- if (wret < 0)
- ret = wret;
+ BUG_ON(list_empty(&src_buf->dirty));
+ BUG_ON(list_empty(&dst_buf->dirty));
return ret;
}
@@ -615,7 +639,6 @@
tree_block_release(root, root->node);
root->node = t;
t->count++;
- dirty_tree_block(root, t);
path->nodes[level] = t;
path->slots[level] = 0;
return 0;
@@ -655,7 +678,7 @@
lower->header.nritems++;
if (lower->keys[1].objectid == 0)
BUG();
- dirty_tree_block(root, path->nodes[level]);
+ BUG_ON(list_empty(&path->nodes[level]->dirty));
return 0;
}
@@ -701,12 +724,7 @@
c->header.nritems = mid;
ret = 0;
- wret = dirty_tree_block(root, t);
- if (wret)
- ret = wret;
- wret = dirty_tree_block(root, split_buffer);
- if (wret)
- ret = wret;
+ BUG_ON(list_empty(&t->dirty));
wret = insert_ptr(root, path, split->keys, split_buffer->blocknr,
path->slots[level + 1] + 1, level + 1);
if (wret)
@@ -778,6 +796,15 @@
tree_block_release(root, right_buf);
return 1;
}
+ /* cow and double check */
+ btrfs_cow_block(root, right_buf, upper, slot + 1, &right_buf);
+ right = &right_buf->leaf;
+ free_space = leaf_free_space(right);
+ if (free_space < data_size + sizeof(struct item)) {
+ tree_block_release(root, right_buf);
+ return 1;
+ }
+
for (i = left->header.nritems - 1; i >= 0; i--) {
item = left->items + i;
if (path->slots[0] == i)
@@ -818,11 +845,12 @@
}
left->header.nritems -= push_items;
- dirty_tree_block(root, left_buf);
- dirty_tree_block(root, right_buf);
+ BUG_ON(list_empty(&left_buf->dirty));
+ BUG_ON(list_empty(&right_buf->dirty));
memcpy(upper->node.keys + slot + 1,
&right->items[0].key, sizeof(struct key));
- dirty_tree_block(root, upper);
+ BUG_ON(list_empty(&upper->dirty));
+
/* then fixup the leaf pointer in the path */
if (path->slots[0] >= left->header.nritems) {
path->slots[0] -= left->header.nritems;
@@ -869,6 +897,16 @@
tree_block_release(root, t);
return 1;
}
+
+ /* cow and double check */
+ btrfs_cow_block(root, t, path->nodes[1], slot - 1, &t);
+ left = &t->leaf;
+ free_space = leaf_free_space(left);
+ if (free_space < data_size + sizeof(struct item)) {
+ tree_block_release(root, t);
+ return 1;
+ }
+
for (i = 0; i < right->header.nritems; i++) {
item = right->items + i;
if (path->slots[0] == i)
@@ -912,12 +950,8 @@
push_space = right->items[i].offset;
}
- wret = dirty_tree_block(root, t);
- if (wret)
- ret = wret;
- wret = dirty_tree_block(root, right_buf);
- if (wret)
- ret = wret;
+ BUG_ON(list_empty(&t->dirty));
+ BUG_ON(list_empty(&right_buf->dirty));
wret = fixup_low_keys(root, path, &right->items[0].key, 1);
if (wret)
@@ -968,6 +1002,7 @@
if (wret < 0)
return wret;
}
+
l_buf = path->nodes[0];
l = &l_buf->leaf;
@@ -1022,13 +1057,8 @@
right_buffer->blocknr, path->slots[1] + 1, 1);
if (wret)
ret = wret;
- wret = dirty_tree_block(root, right_buffer);
- if (wret)
- ret = wret;
- wret = dirty_tree_block(root, l_buf);
- if (wret)
- ret = wret;
-
+ BUG_ON(list_empty(&right_buffer->dirty));
+ BUG_ON(list_empty(&l_buf->dirty));
BUG_ON(path->slots[0] != slot);
if (mid <= slot) {
tree_block_release(root, path->nodes[0]);
@@ -1049,7 +1079,6 @@
void *data, int data_size)
{
int ret = 0;
- int wret;
int slot;
int slot_orig;
struct leaf *leaf;
@@ -1062,7 +1091,7 @@
if (!root->node)
BUG();
init_path(&path);
- ret = search_slot(root, key, &path, data_size);
+ ret = search_slot(root, key, &path, data_size, 1);
if (ret == 0) {
release_path(root, &path);
return -EEXIST;
@@ -1114,10 +1143,7 @@
if (slot == 0)
ret = fixup_low_keys(root, &path, key, 1);
- wret = dirty_tree_block(root, leaf_buf);
- if (wret)
- ret = wret;
-
+ BUG_ON(list_empty(&leaf_buf->dirty));
if (leaf_free_space(leaf) < 0)
BUG();
check_leaf(&path, 0);
@@ -1162,9 +1188,7 @@
if (wret)
ret = wret;
}
- wret = dirty_tree_block(root, parent);
- if (wret)
- ret = wret;
+ BUG_ON(list_empty(&parent->dirty));
return ret;
}
@@ -1205,7 +1229,7 @@
if (leaf->header.nritems == 0) {
if (leaf_buf == root->node) {
leaf->header.flags = node_level(0);
- dirty_tree_block(root, leaf_buf);
+ BUG_ON(list_empty(&leaf_buf->dirty));
} else {
clean_tree_block(root, leaf_buf);
wret = del_ptr(root, path, 1, path->slots[1]);
@@ -1223,9 +1247,7 @@
if (wret)
ret = wret;
}
- wret = dirty_tree_block(root, leaf_buf);
- if (wret)
- ret = wret;
+ BUG_ON(list_empty(&leaf_buf->dirty));
/* delete the leaf if it is mostly empty */
if (used < LEAF_DATA_SIZE / 3) {
@@ -1304,3 +1326,4 @@
return 0;
}
+
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 6b4dabd..9fe8ba6 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -142,8 +142,9 @@
};
struct tree_buffer *alloc_free_block(struct ctree_root *root);
+int btrfs_inc_ref(struct ctree_root *root, struct tree_buffer *buf);
int free_extent(struct ctree_root *root, u64 blocknr, u64 num_blocks);
-int search_slot(struct ctree_root *root, struct key *key, struct ctree_path *p, int ins_len);
+int search_slot(struct ctree_root *root, struct key *key, struct ctree_path *p, int ins_len, int cow);
void release_path(struct ctree_root *root, struct ctree_path *p);
void init_path(struct ctree_path *p);
int del_item(struct ctree_root *root, struct ctree_path *path);
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index b1a8149..0e1c31e 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -260,6 +260,8 @@
if (buf->count < 0)
BUG();
if (buf->count == 0) {
+ BUG_ON(!list_empty(&buf->cache));
+ BUG_ON(!list_empty(&buf->dirty));
if (!radix_tree_lookup(&root->cache_radix, buf->blocknr))
BUG();
radix_tree_delete(&root->cache_radix, buf->blocknr);
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index fb21444..25d9cd1 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -15,6 +15,39 @@
*/
#define CTREE_EXTENT_PENDING 0
+static int inc_block_ref(struct ctree_root *root, u64 blocknr)
+{
+ struct ctree_path path;
+ int ret;
+ struct key key;
+ struct leaf *l;
+ struct extent_item *item;
+ init_path(&path);
+ key.objectid = blocknr;
+ key.flags = 0;
+ key.offset = 1;
+ ret = search_slot(root->extent_root, &key, &path, 0, 1);
+ BUG_ON(ret != 0);
+ l = &path.nodes[0]->leaf;
+ item = (struct extent_item *)(l->data +
+ l->items[path.slots[0]].offset);
+ item->refs++;
+ BUG_ON(list_empty(&path.nodes[0]->dirty));
+ release_path(root->extent_root, &path);
+ return 0;
+}
+
+int btrfs_inc_ref(struct ctree_root *root, struct tree_buffer *buf)
+{
+ u64 blocknr;
+ int i;
+ for (i = 0; i < buf->node.header.nritems; i++) {
+ blocknr = buf->node.blockptrs[i];
+ inc_block_ref(root, blocknr);
+ }
+ return 0;
+}
+
/*
* find all the blocks marked as pending in the radix tree and remove
* them from the extent map
@@ -39,7 +72,7 @@
key.flags = 0;
key.offset = 1;
init_path(&path);
- ret = search_slot(extent_root, &key, &path, -1);
+ ret = search_slot(extent_root, &key, &path, -1, 1);
if (ret) {
print_tree(extent_root, extent_root->node);
printf("unable to find %Lu\n", key.objectid);
@@ -83,7 +116,7 @@
return 0;
}
init_path(&path);
- ret = search_slot(extent_root, &key, &path, -1);
+ ret = search_slot(extent_root, &key, &path, -1, 1);
if (ret) {
print_tree(extent_root, extent_root->node);
printf("failed to find %Lu\n", key.objectid);
@@ -124,7 +157,7 @@
ins->offset = 0;
ins->flags = 0;
start_found = 0;
- ret = search_slot(root, ins, &path, 0);
+ ret = search_slot(root, ins, &path, 0, 0);
if (ret < 0)
goto error;
@@ -221,6 +254,8 @@
ret = insert_item(extent_root, &key, &item,
sizeof(item));
if (ret) {
+ printf("%Lu already in tree\n", key.objectid);
+ print_tree(extent_root, extent_root->node);
BUG();
// FIXME undo it and return sane
return ret;
@@ -228,6 +263,7 @@
radix_tree_tag_clear(&extent_root->cache_radix,
gang[i]->blocknr,
CTREE_EXTENT_PENDING);
+ printf("%Lu is not pending\n", gang[i]->blocknr);
tree_block_release(extent_root, gang[i]);
}
}
@@ -266,15 +302,18 @@
if (pending_ret)
return pending_ret;
*buf = find_tree_block(root, ins->objectid);
+ dirty_tree_block(root, *buf);
return 0;
}
/* we're allocating an extent for the extent tree, don't recurse */
BUG_ON(ins->offset != 1);
*buf = find_tree_block(root, ins->objectid);
BUG_ON(!*buf);
+ printf("%Lu is pending\n", ins->objectid);
radix_tree_tag_set(&root->cache_radix, ins->objectid,
CTREE_EXTENT_PENDING);
(*buf)->count++;
+ dirty_tree_block(root, *buf);
return 0;
}
diff --git a/fs/btrfs/quick-test.c b/fs/btrfs/quick-test.c
index dbd00c3..8255f79 100644
--- a/fs/btrfs/quick-test.c
+++ b/fs/btrfs/quick-test.c
@@ -19,7 +19,7 @@
int i;
int num;
int ret;
- int run_size = 100000;
+ int run_size = 1024;
int max_key = 100000000;
int tree_size = 0;
struct ctree_path path;
@@ -57,7 +57,7 @@
init_path(&path);
if (i % 10000 == 0)
fprintf(stderr, "search %d:%d\n", num, i);
- ret = search_slot(root, &ins, &path, 0);
+ ret = search_slot(root, &ins, &path, 0, 0);
if (ret) {
print_tree(root, root->node);
printf("unable to find %d\n", num);
@@ -79,7 +79,7 @@
num = next_key(i, max_key);
ins.objectid = num;
init_path(&path);
- ret = search_slot(root, &ins, &path, -1);
+ ret = search_slot(root, &ins, &path, -1, 1);
if (!ret) {
if (i % 10000 == 0)
fprintf(stderr, "del %d:%d\n", num, i);
@@ -117,7 +117,7 @@
init_path(&path);
if (i % 10000 == 0)
fprintf(stderr, "search %d:%d\n", num, i);
- ret = search_slot(root, &ins, &path, 0);
+ ret = search_slot(root, &ins, &path, 0, 0);
if (ret) {
print_tree(root, root->node);
printf("unable to find %d\n", num);
@@ -131,7 +131,7 @@
int slot;
ins.objectid = (u64)-1;
init_path(&path);
- ret = search_slot(root, &ins, &path, -1);
+ ret = search_slot(root, &ins, &path, -1, 1);
if (ret == 0)
BUG();
diff --git a/fs/btrfs/random-test.c b/fs/btrfs/random-test.c
index 53245c50..dcc852a 100644
--- a/fs/btrfs/random-test.c
+++ b/fs/btrfs/random-test.c
@@ -93,7 +93,7 @@
ret = setup_key(radix, &key, 1);
if (ret < 0)
return 0;
- ret = search_slot(root, &key, &path, -1);
+ ret = search_slot(root, &key, &path, -1, 1);
if (ret)
goto error;
ret = del_item(root, &path);
@@ -118,7 +118,7 @@
ret = setup_key(radix, &key, 1);
if (ret < 0)
return 0;
- ret = search_slot(root, &key, &path, 0);
+ ret = search_slot(root, &key, &path, 0, 1);
release_path(root, &path);
if (ret)
goto error;
@@ -137,7 +137,7 @@
ret = setup_key(radix, &key, 0);
if (ret < 0)
return ret;
- ret = search_slot(root, &key, &path, 0);
+ ret = search_slot(root, &key, &path, 0, 0);
release_path(root, &path);
if (ret <= 0)
goto error;
@@ -163,7 +163,7 @@
key.objectid = (unsigned long)-1;
while(nr-- >= 0) {
init_path(&path);
- ret = search_slot(root, &key, &path, -1);
+ ret = search_slot(root, &key, &path, -1, 1);
if (ret < 0) {
release_path(root, &path);
return ret;
@@ -216,7 +216,7 @@
return ret;
}
}
- if (i % 10000 == 0) {
+ if (i && i % 10000 == 0) {
printf("bigfill %d\n", i);
}
if (!keep_running)
@@ -263,7 +263,7 @@
key.objectid = (unsigned long)-1;
while(1) {
init_path(&path);
- ret = search_slot(root, &key, &path, 0);
+ ret = search_slot(root, &key, &path, 0, 0);
if (ret < 0) {
release_path(root, &path);
return ret;