BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
return 0;
+ if (trans->transaction->delayed_refs.flushing &&
+ btrfs_header_nritems(mid) > 2)
+ return 0;
+
if (btrfs_header_nritems(mid) < 2)
err_on_enospc = 1;
* readahead one full node of leaves, finding things that are close
* to the block in 'slot', and triggering ra on them.
*/
-static noinline void reada_for_search(struct btrfs_root *root,
- struct btrfs_path *path,
- int level, int slot, u64 objectid)
+static void reada_for_search(struct btrfs_root *root,
+ struct btrfs_path *path,
+ int level, int slot, u64 objectid)
{
struct extent_buffer *node;
struct btrfs_disk_key disk_key;
int ret = 0;
int blocksize;
- parent = path->nodes[level - 1];
+ parent = path->nodes[level + 1];
if (!parent)
return 0;
nritems = btrfs_header_nritems(parent);
- slot = path->slots[level];
+ slot = path->slots[level + 1];
blocksize = btrfs_level_size(root, level);
if (slot > 0) {
block1 = 0;
free_extent_buffer(eb);
}
- if (slot < nritems) {
+ if (slot + 1 < nritems) {
block2 = btrfs_node_blockptr(parent, slot + 1);
gen = btrfs_node_ptr_generation(parent, slot + 1);
eb = btrfs_find_tree_block(root, block2, blocksize);
}
if (block1 || block2) {
ret = -EAGAIN;
+
+ /* release the whole path */
btrfs_release_path(root, path);
+
+ /* read the blocks */
if (block1)
readahead_tree_block(root, block1, blocksize, 0);
if (block2)
eb = read_tree_block(root, block1, blocksize, 0);
free_extent_buffer(eb);
}
- if (block1) {
+ if (block2) {
eb = read_tree_block(root, block2, blocksize, 0);
free_extent_buffer(eb);
}
}
/*
+ * helper function for btrfs_search_slot. The goal is to find a block
+ * in cache without setting the path to blocking. If we find the block
+ * we return zero and the path is unchanged.
+ *
+ * If we can't find the block, we set the path blocking and do some
+ * reada. -EAGAIN is returned and the search must be repeated.
+ */
+static int
+read_block_for_search(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root, struct btrfs_path *p,
+ struct extent_buffer **eb_ret, int level, int slot,
+ struct btrfs_key *key)
+{
+ u64 blocknr;
+ u64 gen;
+ u32 blocksize;
+ struct extent_buffer *b = *eb_ret;
+ struct extent_buffer *tmp;
+
+ blocknr = btrfs_node_blockptr(b, slot);
+ gen = btrfs_node_ptr_generation(b, slot);
+ blocksize = btrfs_level_size(root, level - 1);
+
+ tmp = btrfs_find_tree_block(root, blocknr, blocksize);
+ if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
+ *eb_ret = tmp;
+ return 0;
+ }
+
+ /*
+ * reduce lock contention at high levels
+ * of the btree by dropping locks before
+ * we read.
+ */
+ btrfs_unlock_up_safe(p, level + 1);
+ btrfs_set_path_blocking(p);
+
+ if (tmp)
+ free_extent_buffer(tmp);
+ if (p->reada)
+ reada_for_search(root, p, level, slot, key->objectid);
+
+ btrfs_release_path(NULL, p);
+ tmp = read_tree_block(root, blocknr, blocksize, gen);
+ if (tmp)
+ free_extent_buffer(tmp);
+ return -EAGAIN;
+}
+
+/*
+ * helper function for btrfs_search_slot. This does all of the checks
+ * for node-level blocks and does any balancing required based on
+ * the ins_len.
+ *
+ * If no extra work was required, zero is returned. If we had to
+ * drop the path, -EAGAIN is returned and btrfs_search_slot must
+ * start over
+ */
+static int
+setup_nodes_for_search(struct btrfs_trans_handle *trans,
+ struct btrfs_root *root, struct btrfs_path *p,
+ struct extent_buffer *b, int level, int ins_len)
+{
+ int ret;
+ if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
+ BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
+ int sret;
+
+ sret = reada_for_balance(root, p, level);
+ if (sret)
+ goto again;
+
+ btrfs_set_path_blocking(p);
+ sret = split_node(trans, root, p, level);
+ btrfs_clear_path_blocking(p, NULL);
+
+ BUG_ON(sret > 0);
+ if (sret) {
+ ret = sret;
+ goto done;
+ }
+ b = p->nodes[level];
+ } else if (ins_len < 0 && btrfs_header_nritems(b) <
+ BTRFS_NODEPTRS_PER_BLOCK(root) / 4) {
+ int sret;
+
+ sret = reada_for_balance(root, p, level);
+ if (sret)
+ goto again;
+
+ btrfs_set_path_blocking(p);
+ sret = balance_level(trans, root, p, level);
+ btrfs_clear_path_blocking(p, NULL);
+
+ if (sret) {
+ ret = sret;
+ goto done;
+ }
+ b = p->nodes[level];
+ if (!b) {
+ btrfs_release_path(NULL, p);
+ goto again;
+ }
+ BUG_ON(btrfs_header_nritems(b) == 1);
+ }
+ return 0;
+
+again:
+ ret = -EAGAIN;
+done:
+ return ret;
+}
+
+/*
* look for key in the tree. path is filled in with nodes along the way
* if key is found, we return zero and you can find the item in the leaf
* level of the path (level 0)
ins_len, int cow)
{
struct extent_buffer *b;
- struct extent_buffer *tmp;
int slot;
int ret;
int level;
- int should_reada = p->reada;
int lowest_unlock = 1;
- int blocksize;
u8 lowest_level = 0;
- u64 blocknr;
- u64 gen;
lowest_level = p->lowest_level;
WARN_ON(lowest_level && ins_len > 0);
if (cow) {
int wret;
- /* is a cow on this block not required */
+ /*
+ * if we don't really need to cow this block
+ * then we don't want to set the path blocking,
+ * so we test it here
+ */
if (btrfs_header_generation(b) == trans->transid &&
btrfs_header_owner(b) == root->root_key.objectid &&
!btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) {
if (ret && slot > 0)
slot -= 1;
p->slots[level] = slot;
- if ((p->search_for_split || ins_len > 0) &&
- btrfs_header_nritems(b) >=
- BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
- int sret;
-
- sret = reada_for_balance(root, p, level);
- if (sret)
- goto again;
-
- btrfs_set_path_blocking(p);
- sret = split_node(trans, root, p, level);
- btrfs_clear_path_blocking(p, NULL);
-
- BUG_ON(sret > 0);
- if (sret) {
- ret = sret;
- goto done;
- }
- b = p->nodes[level];
- slot = p->slots[level];
- } else if (ins_len < 0 &&
- btrfs_header_nritems(b) <
- BTRFS_NODEPTRS_PER_BLOCK(root) / 4) {
- int sret;
-
- sret = reada_for_balance(root, p, level);
- if (sret)
- goto again;
-
- btrfs_set_path_blocking(p);
- sret = balance_level(trans, root, p, level);
- btrfs_clear_path_blocking(p, NULL);
+ ret = setup_nodes_for_search(trans, root, p, b, level,
+ ins_len);
+ if (ret == -EAGAIN)
+ goto again;
+ else if (ret)
+ goto done;
+ b = p->nodes[level];
+ slot = p->slots[level];
- if (sret) {
- ret = sret;
- goto done;
- }
- b = p->nodes[level];
- if (!b) {
- btrfs_release_path(NULL, p);
- goto again;
- }
- slot = p->slots[level];
- BUG_ON(btrfs_header_nritems(b) == 1);
- }
unlock_up(p, level, lowest_unlock);
/* this is only true while dropping a snapshot */
goto done;
}
- blocknr = btrfs_node_blockptr(b, slot);
- gen = btrfs_node_ptr_generation(b, slot);
- blocksize = btrfs_level_size(root, level - 1);
+ ret = read_block_for_search(trans, root, p,
+ &b, level, slot, key);
+ if (ret == -EAGAIN)
+ goto again;
- tmp = btrfs_find_tree_block(root, blocknr, blocksize);
- if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
- b = tmp;
- } else {
- /*
- * reduce lock contention at high levels
- * of the btree by dropping locks before
- * we read.
- */
- if (level > 0) {
- btrfs_release_path(NULL, p);
- if (tmp)
- free_extent_buffer(tmp);
- if (should_reada)
- reada_for_search(root, p,
- level, slot,
- key->objectid);
-
- tmp = read_tree_block(root, blocknr,
- blocksize, gen);
- if (tmp)
- free_extent_buffer(tmp);
- goto again;
- } else {
- btrfs_set_path_blocking(p);
- if (tmp)
- free_extent_buffer(tmp);
- if (should_reada)
- reada_for_search(root, p,
- level, slot,
- key->objectid);
- b = read_node_slot(root, b, slot);
- }
- }
if (!p->skip_locking) {
int lret;
* we don't really know what they plan on doing with the path
* from here on, so for now just mark it as blocking
*/
- btrfs_set_path_blocking(p);
+ if (!p->leave_spinning)
+ btrfs_set_path_blocking(p);
return ret;
}
BUG_ON(!path->nodes[level]);
lower = path->nodes[level];
nritems = btrfs_header_nritems(lower);
- if (slot > nritems)
- BUG();
+ BUG_ON(slot > nritems);
if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
BUG();
if (slot != nritems) {
ret = insert_new_root(trans, root, path, level + 1);
if (ret)
return ret;
- } else {
+ } else if (!trans->transaction->delayed_refs.flushing) {
ret = push_nodes_for_insert(trans, root, path, level);
c = path->nodes[level];
if (!ret && btrfs_header_nritems(c) <
int num_doubles = 0;
/* first try to make some room by pushing left and right */
- if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
+ if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY &&
+ !trans->transaction->delayed_refs.flushing) {
wret = push_leaf_right(trans, root, path, data_size, 0);
if (wret < 0)
return wret;
return -EAGAIN;
}
+ btrfs_set_path_blocking(path);
ret = split_leaf(trans, root, &orig_key, path,
sizeof(struct btrfs_item), 1);
path->keep_locks = 0;
BUG_ON(ret);
+ btrfs_unlock_up_safe(path, 1);
+ leaf = path->nodes[0];
+ BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
+
+split:
/*
* make sure any changes to the path from split_leaf leave it
* in a blocking state
*/
btrfs_set_path_blocking(path);
- leaf = path->nodes[0];
- BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
-
-split:
item = btrfs_item_nr(leaf, path->slots[0]);
orig_offset = btrfs_item_offset(leaf, item);
item_size = btrfs_item_size(leaf, item);
-
buf = kmalloc(item_size, GFP_NOFS);
read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
path->slots[0]), item_size);
}
btrfs_set_header_nritems(leaf, nritems + nr);
- btrfs_mark_buffer_dirty(leaf);
ret = 0;
if (slot == 0) {
btrfs_cpu_key_to_disk(&disk_key, cpu_key);
ret = fixup_low_keys(trans, root, path, &disk_key, 1);
}
+ btrfs_unlock_up_safe(path, 1);
+ btrfs_mark_buffer_dirty(leaf);
if (btrfs_leaf_free_space(root, leaf) < 0) {
btrfs_print_leaf(root, leaf);
total_data, total_size, nr);
out:
- btrfs_unlock_up_safe(path, 1);
return ret;
}
}
/* delete the leaf if it is mostly empty */
- if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
+ if (used < BTRFS_LEAF_DATA_SIZE(root) / 4 &&
+ !trans->transaction->delayed_refs.flushing) {
/* push_leaf_left fixes the path.
* make sure the path still points to our leaf
* for possible call to del_ptr below
slot = path->slots[1];
extent_buffer_get(leaf);
+ btrfs_set_path_blocking(path);
wret = push_leaf_left(trans, root, path, 1, 1);
if (wret < 0 && wret != -ENOSPC)
ret = wret;
int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
{
int slot;
- int level = 1;
+ int level;
struct extent_buffer *c;
- struct extent_buffer *next = NULL;
+ struct extent_buffer *next;
struct btrfs_key key;
u32 nritems;
int ret;
+ int old_spinning = path->leave_spinning;
+ int force_blocking = 0;
nritems = btrfs_header_nritems(path->nodes[0]);
if (nritems == 0)
return 1;
- btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
+ /*
+ * we take the blocks in an order that upsets lockdep. Using
+ * blocking mode is the only way around it.
+ */
+#ifdef CONFIG_DEBUG_LOCK_ALLOC
+ force_blocking = 1;
+#endif
+ btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
+again:
+ level = 1;
+ next = NULL;
btrfs_release_path(root, path);
+
path->keep_locks = 1;
+
+ if (!force_blocking)
+ path->leave_spinning = 1;
+
ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
path->keep_locks = 0;
if (ret < 0)
return ret;
- btrfs_set_path_blocking(path);
nritems = btrfs_header_nritems(path->nodes[0]);
/*
* by releasing the path above we dropped all our locks. A balance
*/
if (nritems > 0 && path->slots[0] < nritems - 1) {
path->slots[0]++;
+ ret = 0;
goto done;
}
while (level < BTRFS_MAX_LEVEL) {
- if (!path->nodes[level])
- return 1;
+ if (!path->nodes[level]) {
+ ret = 1;
+ goto done;
+ }
slot = path->slots[level] + 1;
c = path->nodes[level];
if (slot >= btrfs_header_nritems(c)) {
level++;
- if (level == BTRFS_MAX_LEVEL)
- return 1;
+ if (level == BTRFS_MAX_LEVEL) {
+ ret = 1;
+ goto done;
+ }
continue;
}
free_extent_buffer(next);
}
- /* the path was set to blocking above */
- if (level == 1 && (path->locks[1] || path->skip_locking) &&
- path->reada)
- reada_for_search(root, path, level, slot, 0);
+ next = c;
+ ret = read_block_for_search(NULL, root, path, &next, level,
+ slot, &key);
+ if (ret == -EAGAIN)
+ goto again;
- next = read_node_slot(root, c, slot);
if (!path->skip_locking) {
- btrfs_assert_tree_locked(c);
- btrfs_tree_lock(next);
- btrfs_set_lock_blocking(next);
+ ret = btrfs_try_spin_lock(next);
+ if (!ret) {
+ btrfs_set_path_blocking(path);
+ btrfs_tree_lock(next);
+ if (!force_blocking)
+ btrfs_clear_path_blocking(path, next);
+ }
+ if (force_blocking)
+ btrfs_set_lock_blocking(next);
}
break;
}
c = path->nodes[level];
if (path->locks[level])
btrfs_tree_unlock(c);
+
free_extent_buffer(c);
path->nodes[level] = next;
path->slots[level] = 0;
if (!path->skip_locking)
path->locks[level] = 1;
+
if (!level)
break;
- btrfs_set_path_blocking(path);
- if (level == 1 && path->locks[1] && path->reada)
- reada_for_search(root, path, level, slot, 0);
- next = read_node_slot(root, next, 0);
+ ret = read_block_for_search(NULL, root, path, &next, level,
+ 0, &key);
+ if (ret == -EAGAIN)
+ goto again;
+
if (!path->skip_locking) {
btrfs_assert_tree_locked(path->nodes[level]);
- btrfs_tree_lock(next);
- btrfs_set_lock_blocking(next);
+ ret = btrfs_try_spin_lock(next);
+ if (!ret) {
+ btrfs_set_path_blocking(path);
+ btrfs_tree_lock(next);
+ if (!force_blocking)
+ btrfs_clear_path_blocking(path, next);
+ }
+ if (force_blocking)
+ btrfs_set_lock_blocking(next);
}
}
+ ret = 0;
done:
unlock_up(path, 0, 1);
- return 0;
+ path->leave_spinning = old_spinning;
+ if (!old_spinning)
+ btrfs_set_path_blocking(path);
+
+ return ret;
}
/*
git://ftp.safe.ca / safe / jmp / linux-2.6 / blobdiff