Btrfs: balance_level checks !child after access
[pandora-kernel.git] / fs / btrfs / ctree.c
index f6f7a6a..6674692 100644 (file)
@@ -54,6 +54,31 @@ struct btrfs_path *btrfs_alloc_path(void)
        return path;
 }
 
+/*
+ * set all locked nodes in the path to blocking locks.  This should
+ * be done before scheduling
+ */
+noinline void btrfs_set_path_blocking(struct btrfs_path *p)
+{
+       int i;
+       for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
+               if (p->nodes[i] && p->locks[i])
+                       btrfs_set_lock_blocking(p->nodes[i]);
+       }
+}
+
+/*
+ * reset all the locked nodes in the patch to spinning locks.
+ */
+noinline void btrfs_clear_path_blocking(struct btrfs_path *p)
+{
+       int i;
+       for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
+               if (p->nodes[i] && p->locks[i])
+                       btrfs_clear_lock_blocking(p->nodes[i]);
+       }
+}
+
 /* this also releases the path */
 void btrfs_free_path(struct btrfs_path *p)
 {
@@ -67,7 +92,7 @@ void btrfs_free_path(struct btrfs_path *p)
  *
  * It is safe to call this on paths that no locks or extent buffers held.
  */
-void noinline btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
+noinline void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
 {
        int i;
 
@@ -112,7 +137,7 @@ struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
 {
        struct extent_buffer *eb;
 
-       while(1) {
+       while (1) {
                eb = btrfs_root_node(root);
                btrfs_tree_lock(eb);
 
@@ -202,22 +227,22 @@ int btrfs_copy_root(struct btrfs_trans_handle *trans,
 }
 
 /*
- * does the dirty work in cow of a single block.  The parent block
- * (if supplied) is updated to point to the new cow copy.  The new
- * buffer is marked dirty and returned locked.  If you modify the block
- * it needs to be marked dirty again.
+ * does the dirty work in cow of a single block.  The parent block (if
+ * supplied) is updated to point to the new cow copy.  The new buffer is marked
+ * dirty and returned locked.  If you modify the block it needs to be marked
+ * dirty again.
  *
  * search_start -- an allocation hint for the new block
  *
- * empty_size -- a hint that you plan on doing more cow.  This is the size in bytes
- * the allocator should try to find free next to the block it returns.  This is
- * just a hint and may be ignored by the allocator.
+ * empty_size -- a hint that you plan on doing more cow.  This is the size in
+ * bytes the allocator should try to find free next to the block it returns.
+ * This is just a hint and may be ignored by the allocator.
  *
  * prealloc_dest -- if you have already reserved a destination for the cow,
- * this uses that block instead of allocating a new one.  btrfs_alloc_reserved_extent
- * is used to finish the allocation.
+ * this uses that block instead of allocating a new one.
+ * btrfs_alloc_reserved_extent is used to finish the allocation.
  */
-static int noinline __btrfs_cow_block(struct btrfs_trans_handle *trans,
+static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
                             struct btrfs_root *root,
                             struct extent_buffer *buf,
                             struct extent_buffer *parent, int parent_slot,
@@ -272,6 +297,8 @@ static int noinline __btrfs_cow_block(struct btrfs_trans_handle *trans,
        if (IS_ERR(cow))
                return PTR_ERR(cow);
 
+       /* cow is set to blocking by btrfs_init_new_buffer */
+
        copy_extent_buffer(cow, buf, 0, 0, cow->len);
        btrfs_set_header_bytenr(cow, cow->start);
        btrfs_set_header_generation(cow, trans->transid);
@@ -366,7 +393,7 @@ static int noinline __btrfs_cow_block(struct btrfs_trans_handle *trans,
  * This version of it has extra checks so that a block isn't cow'd more than
  * once per transaction, as long as it hasn't been written yet
  */
-int noinline btrfs_cow_block(struct btrfs_trans_handle *trans,
+noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
                    struct btrfs_root *root, struct extent_buffer *buf,
                    struct extent_buffer *parent, int parent_slot,
                    struct extent_buffer **cow_ret, u64 prealloc_dest)
@@ -375,27 +402,33 @@ int noinline btrfs_cow_block(struct btrfs_trans_handle *trans,
        int ret;
 
        if (trans->transaction != root->fs_info->running_transaction) {
-               printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
+               printk(KERN_CRIT "trans %llu running %llu\n",
+                      (unsigned long long)trans->transid,
+                      (unsigned long long)
                       root->fs_info->running_transaction->transid);
                WARN_ON(1);
        }
        if (trans->transid != root->fs_info->generation) {
-               printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
-                      root->fs_info->generation);
+               printk(KERN_CRIT "trans %llu running %llu\n",
+                      (unsigned long long)trans->transid,
+                      (unsigned long long)root->fs_info->generation);
                WARN_ON(1);
        }
 
-       spin_lock(&root->fs_info->hash_lock);
        if (btrfs_header_generation(buf) == trans->transid &&
            btrfs_header_owner(buf) == root->root_key.objectid &&
            !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
                *cow_ret = buf;
-               spin_unlock(&root->fs_info->hash_lock);
                WARN_ON(prealloc_dest);
                return 0;
        }
-       spin_unlock(&root->fs_info->hash_lock);
+
        search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
+
+       if (parent)
+               btrfs_set_lock_blocking(parent);
+       btrfs_set_lock_blocking(buf);
+
        ret = __btrfs_cow_block(trans, root, buf, parent,
                                 parent_slot, cow_ret, search_start, 0,
                                 prealloc_dest);
@@ -489,16 +522,10 @@ int btrfs_realloc_node(struct btrfs_trans_handle *trans,
        if (cache_only && parent_level != 1)
                return 0;
 
-       if (trans->transaction != root->fs_info->running_transaction) {
-               printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
-                      root->fs_info->running_transaction->transid);
+       if (trans->transaction != root->fs_info->running_transaction)
                WARN_ON(1);
-       }
-       if (trans->transid != root->fs_info->generation) {
-               printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
-                      root->fs_info->generation);
+       if (trans->transid != root->fs_info->generation)
                WARN_ON(1);
-       }
 
        parent_nritems = btrfs_header_nritems(parent);
        blocksize = btrfs_level_size(root, parent_level - 1);
@@ -507,6 +534,8 @@ int btrfs_realloc_node(struct btrfs_trans_handle *trans,
        if (parent_nritems == 1)
                return 0;
 
+       btrfs_set_lock_blocking(parent);
+
        for (i = start_slot; i < end_slot; i++) {
                int close = 1;
 
@@ -567,6 +596,7 @@ int btrfs_realloc_node(struct btrfs_trans_handle *trans,
                        search_start = last_block;
 
                btrfs_tree_lock(cur);
+               btrfs_set_lock_blocking(cur);
                err = __btrfs_cow_block(trans, root, cur, parent, i,
                                        &cur, search_start,
                                        min(16 * blocksize,
@@ -681,51 +711,18 @@ static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
                BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
                       btrfs_header_bytenr(leaf));
        }
-#if 0
-       for (i = 0; nritems > 1 && i < nritems - 2; i++) {
-               btrfs_item_key_to_cpu(leaf, &cpukey, i + 1);
-               btrfs_item_key(leaf, &leaf_key, i);
-               if (comp_keys(&leaf_key, &cpukey) >= 0) {
-                       btrfs_print_leaf(root, leaf);
-                       printk("slot %d offset bad key\n", i);
-                       BUG_ON(1);
-               }
-               if (btrfs_item_offset_nr(leaf, i) !=
-                       btrfs_item_end_nr(leaf, i + 1)) {
-                       btrfs_print_leaf(root, leaf);
-                       printk("slot %d offset bad\n", i);
-                       BUG_ON(1);
-               }
-               if (i == 0) {
-                       if (btrfs_item_offset_nr(leaf, i) +
-                              btrfs_item_size_nr(leaf, i) !=
-                              BTRFS_LEAF_DATA_SIZE(root)) {
-                               btrfs_print_leaf(root, leaf);
-                               printk("slot %d first offset bad\n", i);
-                               BUG_ON(1);
-                       }
-               }
-       }
-       if (nritems > 0) {
-               if (btrfs_item_size_nr(leaf, nritems - 1) > 4096) {
-                               btrfs_print_leaf(root, leaf);
-                               printk("slot %d bad size \n", nritems - 1);
-                               BUG_ON(1);
-               }
-       }
-#endif
        if (slot != 0 && slot < nritems - 1) {
                btrfs_item_key(leaf, &leaf_key, slot);
                btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
                if (comp_keys(&leaf_key, &cpukey) <= 0) {
                        btrfs_print_leaf(root, leaf);
-                       printk("slot %d offset bad key\n", slot);
+                       printk(KERN_CRIT "slot %d offset bad key\n", slot);
                        BUG_ON(1);
                }
                if (btrfs_item_offset_nr(leaf, slot - 1) !=
                       btrfs_item_end_nr(leaf, slot)) {
                        btrfs_print_leaf(root, leaf);
-                       printk("slot %d offset bad\n", slot);
+                       printk(KERN_CRIT "slot %d offset bad\n", slot);
                        BUG_ON(1);
                }
        }
@@ -736,7 +733,7 @@ static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
                if (btrfs_item_offset_nr(leaf, slot) !=
                        btrfs_item_end_nr(leaf, slot + 1)) {
                        btrfs_print_leaf(root, leaf);
-                       printk("slot %d offset bad\n", slot);
+                       printk(KERN_CRIT "slot %d offset bad\n", slot);
                        BUG_ON(1);
                }
        }
@@ -745,30 +742,10 @@ static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
        return 0;
 }
 
-static int noinline check_block(struct btrfs_root *root,
+static noinline int check_block(struct btrfs_root *root,
                                struct btrfs_path *path, int level)
 {
-       u64 found_start;
        return 0;
-       if (btrfs_header_level(path->nodes[level]) != level)
-           printk("warning: bad level %Lu wanted %d found %d\n",
-                  path->nodes[level]->start, level,
-                  btrfs_header_level(path->nodes[level]));
-       found_start = btrfs_header_bytenr(path->nodes[level]);
-       if (found_start != path->nodes[level]->start) {
-           printk("warning: bad bytentr %Lu found %Lu\n",
-                  path->nodes[level]->start, found_start);
-       }
-#if 0
-       struct extent_buffer *buf = path->nodes[level];
-
-       if (memcmp_extent_buffer(buf, root->fs_info->fsid,
-                                (unsigned long)btrfs_header_fsid(buf),
-                                BTRFS_FSID_SIZE)) {
-               printk("warning bad block %Lu\n", buf->start);
-               return 1;
-       }
-#endif
        if (level == 0)
                return check_leaf(root, path, level);
        return check_node(root, path, level);
@@ -802,7 +779,7 @@ static noinline int generic_bin_search(struct extent_buffer *eb,
        unsigned long map_len = 0;
        int err;
 
-       while(low < high) {
+       while (low < high) {
                mid = (low + high) / 2;
                offset = p + mid * item_size;
 
@@ -918,6 +895,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
                return 0;
 
        mid = path->nodes[level];
+
        WARN_ON(!path->locks[level]);
        WARN_ON(btrfs_header_generation(mid) != trans->transid);
 
@@ -939,8 +917,9 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
 
                /* promote the child to a root */
                child = read_node_slot(root, mid, 0);
-               btrfs_tree_lock(child);
                BUG_ON(!child);
+               btrfs_tree_lock(child);
+               btrfs_set_lock_blocking(child);
                ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 0);
                BUG_ON(ret);
 
@@ -956,6 +935,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
 
                add_root_to_dirty_list(root);
                btrfs_tree_unlock(child);
+
                path->locks[level] = 0;
                path->nodes[level] = NULL;
                clean_tree_block(trans, root, mid);
@@ -980,6 +960,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
        left = read_node_slot(root, parent, pslot - 1);
        if (left) {
                btrfs_tree_lock(left);
+               btrfs_set_lock_blocking(left);
                wret = btrfs_cow_block(trans, root, left,
                                       parent, pslot - 1, &left, 0);
                if (wret) {
@@ -990,6 +971,7 @@ static noinline int balance_level(struct btrfs_trans_handle *trans,
        right = read_node_slot(root, parent, pslot + 1);
        if (right) {
                btrfs_tree_lock(right);
+               btrfs_set_lock_blocking(right);
                wret = btrfs_cow_block(trans, root, right,
                                       parent, pslot + 1, &right, 0);
                if (wret) {
@@ -1130,7 +1112,7 @@ enospc:
  * when they are completely full.  This is also done top down, so we
  * have to be pessimistic.
  */
-static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
+static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
                                          struct btrfs_root *root,
                                          struct btrfs_path *path, int level)
 {
@@ -1165,6 +1147,8 @@ static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
                u32 left_nr;
 
                btrfs_tree_lock(left);
+               btrfs_set_lock_blocking(left);
+
                left_nr = btrfs_header_nritems(left);
                if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
                        wret = 1;
@@ -1211,7 +1195,10 @@ static int noinline push_nodes_for_insert(struct btrfs_trans_handle *trans,
         */
        if (right) {
                u32 right_nr;
+
                btrfs_tree_lock(right);
+               btrfs_set_lock_blocking(right);
+
                right_nr = btrfs_header_nritems(right);
                if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
                        wret = 1;
@@ -1266,8 +1253,7 @@ static noinline void reada_for_search(struct btrfs_root *root,
        struct btrfs_disk_key disk_key;
        u32 nritems;
        u64 search;
-       u64 lowest_read;
-       u64 highest_read;
+       u64 target;
        u64 nread = 0;
        int direction = path->reada;
        struct extent_buffer *eb;
@@ -1291,12 +1277,11 @@ static noinline void reada_for_search(struct btrfs_root *root,
                return;
        }
 
-       highest_read = search;
-       lowest_read = search;
+       target = search;
 
        nritems = btrfs_header_nritems(node);
        nr = slot;
-       while(1) {
+       while (1) {
                if (direction < 0) {
                        if (nr == 0)
                                break;
@@ -1312,38 +1297,92 @@ static noinline void reada_for_search(struct btrfs_root *root,
                                break;
                }
                search = btrfs_node_blockptr(node, nr);
-               if ((search >= lowest_read && search <= highest_read) ||
-                   (search < lowest_read && lowest_read - search <= 16384) ||
-                   (search > highest_read && search - highest_read <= 16384)) {
+               if ((search <= target && target - search <= 65536) ||
+                   (search > target && search - target <= 65536)) {
                        readahead_tree_block(root, search, blocksize,
                                     btrfs_node_ptr_generation(node, nr));
                        nread += blocksize;
                }
                nscan++;
-               if (path->reada < 2 && (nread > (64 * 1024) || nscan > 32))
-                       break;
-               if(nread > (256 * 1024) || nscan > 128)
+               if ((nread > 65536 || nscan > 32))
                        break;
+       }
+}
 
-               if (search < lowest_read)
-                       lowest_read = search;
-               if (search > highest_read)
-                       highest_read = search;
+/*
+ * returns -EAGAIN if it had to drop the path, or zero if everything was in
+ * cache
+ */
+static noinline int reada_for_balance(struct btrfs_root *root,
+                                     struct btrfs_path *path, int level)
+{
+       int slot;
+       int nritems;
+       struct extent_buffer *parent;
+       struct extent_buffer *eb;
+       u64 gen;
+       u64 block1 = 0;
+       u64 block2 = 0;
+       int ret = 0;
+       int blocksize;
+
+       parent = path->nodes[level - 1];
+       if (!parent)
+               return 0;
+
+       nritems = btrfs_header_nritems(parent);
+       slot = path->slots[level];
+       blocksize = btrfs_level_size(root, level);
+
+       if (slot > 0) {
+               block1 = btrfs_node_blockptr(parent, slot - 1);
+               gen = btrfs_node_ptr_generation(parent, slot - 1);
+               eb = btrfs_find_tree_block(root, block1, blocksize);
+               if (eb && btrfs_buffer_uptodate(eb, gen))
+                       block1 = 0;
+               free_extent_buffer(eb);
+       }
+       if (slot < 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 (eb && btrfs_buffer_uptodate(eb, gen))
+                       block2 = 0;
+               free_extent_buffer(eb);
+       }
+       if (block1 || block2) {
+               ret = -EAGAIN;
+               btrfs_release_path(root, path);
+               if (block1)
+                       readahead_tree_block(root, block1, blocksize, 0);
+               if (block2)
+                       readahead_tree_block(root, block2, blocksize, 0);
+
+               if (block1) {
+                       eb = read_tree_block(root, block1, blocksize, 0);
+                       free_extent_buffer(eb);
+               }
+               if (block1) {
+                       eb = read_tree_block(root, block2, blocksize, 0);
+                       free_extent_buffer(eb);
+               }
        }
+       return ret;
 }
 
+
 /*
- * when we walk down the tree, it is usually safe to unlock the higher layers in
- * the tree.  The exceptions are when our path goes through slot 0, because operations
- * on the tree might require changing key pointers higher up in the tree.
+ * when we walk down the tree, it is usually safe to unlock the higher layers
+ * in the tree.  The exceptions are when our path goes through slot 0, because
+ * operations on the tree might require changing key pointers higher up in the
+ * tree.
  *
- * callers might also have set path->keep_locks, which tells this code to
- * keep the lock if the path points to the last slot in the block.  This is
- * part of walking through the tree, and selecting the next slot in the higher
- * block.
+ * callers might also have set path->keep_locks, which tells this code to keep
+ * the lock if the path points to the last slot in the block.  This is part of
+ * walking through the tree, and selecting the next slot in the higher block.
  *
- * lowest_unlock sets the lowest level in the tree we're allowed to unlock.
- * so if lowest_unlock is 1, level 0 won't be unlocked
+ * lowest_unlock sets the lowest level in the tree we're allowed to unlock.  so
+ * if lowest_unlock is 1, level 0 won't be unlocked
  */
 static noinline void unlock_up(struct btrfs_path *path, int level,
                               int lowest_unlock)
@@ -1382,6 +1421,32 @@ static noinline void unlock_up(struct btrfs_path *path, int level,
        }
 }
 
+/*
+ * This releases any locks held in the path starting at level and
+ * going all the way up to the root.
+ *
+ * btrfs_search_slot will keep the lock held on higher nodes in a few
+ * corner cases, such as COW of the block at slot zero in the node.  This
+ * ignores those rules, and it should only be called when there are no
+ * more updates to be done higher up in the tree.
+ */
+noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
+{
+       int i;
+
+       if (path->keep_locks || path->lowest_level)
+               return;
+
+       for (i = level; i < BTRFS_MAX_LEVEL; i++) {
+               if (!path->nodes[i])
+                       continue;
+               if (!path->locks[i])
+                       continue;
+               btrfs_tree_unlock(path->nodes[i]);
+               path->locks[i] = 0;
+       }
+}
+
 /*
  * 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
@@ -1442,32 +1507,30 @@ again:
                        int wret;
 
                        /* is a cow on this block not required */
-                       spin_lock(&root->fs_info->hash_lock);
                        if (btrfs_header_generation(b) == trans->transid &&
                            btrfs_header_owner(b) == root->root_key.objectid &&
                            !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) {
-                               spin_unlock(&root->fs_info->hash_lock);
                                goto cow_done;
                        }
-                       spin_unlock(&root->fs_info->hash_lock);
 
                        /* ok, we have to cow, is our old prealloc the right
                         * size?
                         */
                        if (prealloc_block.objectid &&
                            prealloc_block.offset != b->len) {
+                               btrfs_release_path(root, p);
                                btrfs_free_reserved_extent(root,
                                           prealloc_block.objectid,
                                           prealloc_block.offset);
                                prealloc_block.objectid = 0;
+                               goto again;
                        }
 
                        /*
                         * for higher level blocks, try not to allocate blocks
                         * with the block and the parent locks held.
                         */
-                       if (level > 1 && !prealloc_block.objectid &&
-                           btrfs_path_lock_waiting(p, level)) {
+                       if (level > 0 && !prealloc_block.objectid) {
                                u32 size = b->len;
                                u64 hint = b->start;
 
@@ -1480,6 +1543,8 @@ again:
                                goto again;
                        }
 
+                       btrfs_set_path_blocking(p);
+
                        wret = btrfs_cow_block(trans, root, b,
                                               p->nodes[level + 1],
                                               p->slots[level + 1],
@@ -1501,6 +1566,22 @@ cow_done:
                if (!p->skip_locking)
                        p->locks[level] = 1;
 
+               btrfs_clear_path_blocking(p);
+
+               /*
+                * we have a lock on b and as long as we aren't changing
+                * the tree, there is no way to for the items in b to change.
+                * It is safe to drop the lock on our parent before we
+                * go through the expensive btree search on b.
+                *
+                * If cow is true, then we might be changing slot zero,
+                * which may require changing the parent.  So, we can't
+                * drop the lock until after we know which slot we're
+                * operating on.
+                */
+               if (!cow)
+                       btrfs_unlock_up_safe(p, level + 1);
+
                ret = check_block(root, p, level);
                if (ret) {
                        ret = -1;
@@ -1508,6 +1589,7 @@ cow_done:
                }
 
                ret = bin_search(b, key, level, &slot);
+
                if (level != 0) {
                        if (ret && slot > 0)
                                slot -= 1;
@@ -1515,7 +1597,16 @@ cow_done:
                        if ((p->search_for_split || ins_len > 0) &&
                            btrfs_header_nritems(b) >=
                            BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
-                               int sret = split_node(trans, root, p, level);
+                               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);
+
                                BUG_ON(sret > 0);
                                if (sret) {
                                        ret = sret;
@@ -1523,9 +1614,19 @@ cow_done:
                                }
                                b = p->nodes[level];
                                slot = p->slots[level];
-                       } else if (ins_len < 0) {
-                               int sret = balance_level(trans, root, p,
-                                                        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);
+
                                if (sret) {
                                        ret = sret;
                                        goto done;
@@ -1559,7 +1660,7 @@ cow_done:
                                 * of the btree by dropping locks before
                                 * we read.
                                 */
-                               if (level > 1) {
+                               if (level > 0) {
                                        btrfs_release_path(NULL, p);
                                        if (tmp)
                                                free_extent_buffer(tmp);
@@ -1574,6 +1675,7 @@ cow_done:
                                                free_extent_buffer(tmp);
                                        goto again;
                                } else {
+                                       btrfs_set_path_blocking(p);
                                        if (tmp)
                                                free_extent_buffer(tmp);
                                        if (should_reada)
@@ -1583,14 +1685,29 @@ cow_done:
                                        b = read_node_slot(root, b, slot);
                                }
                        }
-                       if (!p->skip_locking)
-                               btrfs_tree_lock(b);
+                       if (!p->skip_locking) {
+                               int lret;
+
+                               btrfs_clear_path_blocking(p);
+                               lret = btrfs_try_spin_lock(b);
+
+                               if (!lret) {
+                                       btrfs_set_path_blocking(p);
+                                       btrfs_tree_lock(b);
+                                       btrfs_clear_path_blocking(p);
+                               }
+                       }
                } else {
                        p->slots[level] = slot;
-                       if (ins_len > 0 && btrfs_leaf_free_space(root, b) <
-                           sizeof(struct btrfs_item) + ins_len) {
-                               int sret = split_leaf(trans, root, key,
+                       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);
+                               btrfs_clear_path_blocking(p);
+
                                BUG_ON(sret > 0);
                                if (sret) {
                                        ret = sret;
@@ -1604,12 +1721,16 @@ cow_done:
        }
        ret = 1;
 done:
+       /*
+        * 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 (prealloc_block.objectid) {
                btrfs_free_reserved_extent(root,
                           prealloc_block.objectid,
                           prealloc_block.offset);
        }
-
        return ret;
 }
 
@@ -1633,6 +1754,8 @@ int btrfs_merge_path(struct btrfs_trans_handle *trans,
        ret = btrfs_cow_block(trans, root, eb, NULL, 0, &eb, 0);
        BUG_ON(ret);
 
+       btrfs_set_lock_blocking(eb);
+
        parent = eb;
        while (1) {
                level = btrfs_header_level(parent);
@@ -1657,6 +1780,7 @@ int btrfs_merge_path(struct btrfs_trans_handle *trans,
                        eb = read_tree_block(root, bytenr, blocksize,
                                             generation);
                        btrfs_tree_lock(eb);
+                       btrfs_set_lock_blocking(eb);
                }
 
                /*
@@ -1681,6 +1805,7 @@ int btrfs_merge_path(struct btrfs_trans_handle *trans,
                                eb = read_tree_block(root, bytenr, blocksize,
                                                generation);
                                btrfs_tree_lock(eb);
+                               btrfs_set_lock_blocking(eb);
                        }
 
                        ret = btrfs_cow_block(trans, root, eb, parent, slot,
@@ -1832,9 +1957,8 @@ static int push_node_left(struct btrfs_trans_handle *trans,
        if (!empty && src_nritems <= 8)
                return 1;
 
-       if (push_items <= 0) {
+       if (push_items <= 0)
                return 1;
-       }
 
        if (empty) {
                push_items = min(src_nritems, push_items);
@@ -1854,7 +1978,7 @@ static int push_node_left(struct btrfs_trans_handle *trans,
        copy_extent_buffer(dst, src,
                           btrfs_node_key_ptr_offset(dst_nritems),
                           btrfs_node_key_ptr_offset(0),
-                          push_items * sizeof(struct btrfs_key_ptr));
+                          push_items * sizeof(struct btrfs_key_ptr));
 
        if (push_items < src_nritems) {
                memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
@@ -1899,19 +2023,16 @@ static int balance_node_right(struct btrfs_trans_handle *trans,
        src_nritems = btrfs_header_nritems(src);
        dst_nritems = btrfs_header_nritems(dst);
        push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
-       if (push_items <= 0) {
+       if (push_items <= 0)
                return 1;
-       }
 
-       if (src_nritems < 4) {
+       if (src_nritems < 4)
                return 1;
-       }
 
        max_push = src_nritems / 2 + 1;
        /* don't try to empty the node */
-       if (max_push >= src_nritems) {
+       if (max_push >= src_nritems)
                return 1;
-       }
 
        if (max_push < push_items)
                push_items = max_push;
@@ -1924,7 +2045,7 @@ static int balance_node_right(struct btrfs_trans_handle *trans,
        copy_extent_buffer(dst, src,
                           btrfs_node_key_ptr_offset(0),
                           btrfs_node_key_ptr_offset(src_nritems - push_items),
-                          push_items * sizeof(struct btrfs_key_ptr));
+                          push_items * sizeof(struct btrfs_key_ptr));
 
        btrfs_set_header_nritems(src, src_nritems - push_items);
        btrfs_set_header_nritems(dst, dst_nritems + push_items);
@@ -1945,7 +2066,7 @@ static int balance_node_right(struct btrfs_trans_handle *trans,
  *
  * returns zero on success or < 0 on failure.
  */
-static int noinline insert_new_root(struct btrfs_trans_handle *trans,
+static noinline int insert_new_root(struct btrfs_trans_handle *trans,
                           struct btrfs_root *root,
                           struct btrfs_path *path, int level)
 {
@@ -2176,14 +2297,15 @@ static int leaf_space_used(struct extent_buffer *l, int start, int nr)
  * the start of the leaf data.  IOW, how much room
  * the leaf has left for both items and data
  */
-int noinline btrfs_leaf_free_space(struct btrfs_root *root,
+noinline int btrfs_leaf_free_space(struct btrfs_root *root,
                                   struct extent_buffer *leaf)
 {
        int nritems = btrfs_header_nritems(leaf);
        int ret;
        ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
        if (ret < 0) {
-               printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
+               printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
+                      "used %d nritems %d\n",
                       ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
                       leaf_space_used(leaf, 0, nritems), nritems);
        }
@@ -2219,9 +2341,9 @@ static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
        int ret;
 
        slot = path->slots[1];
-       if (!path->nodes[1]) {
+       if (!path->nodes[1])
                return 1;
-       }
+
        upper = path->nodes[1];
        if (slot >= btrfs_header_nritems(upper) - 1)
                return 1;
@@ -2230,8 +2352,10 @@ static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
 
        right = read_node_slot(root, upper, slot + 1);
        btrfs_tree_lock(right);
+       btrfs_set_lock_blocking(right);
+
        free_space = btrfs_leaf_free_space(root, right);
-       if (free_space < data_size + sizeof(struct btrfs_item))
+       if (free_space < data_size)
                goto out_unlock;
 
        /* cow and double check */
@@ -2241,7 +2365,7 @@ static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
                goto out_unlock;
 
        free_space = btrfs_leaf_free_space(root, right);
-       if (free_space < data_size + sizeof(struct btrfs_item))
+       if (free_space < data_size)
                goto out_unlock;
 
        left_nritems = btrfs_header_nritems(left);
@@ -2254,7 +2378,7 @@ static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
                nr = 1;
 
        if (path->slots[0] >= left_nritems)
-               push_space += data_size + sizeof(*item);
+               push_space += data_size;
 
        i = left_nritems - 1;
        while (i >= nr) {
@@ -2271,7 +2395,7 @@ static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
                }
 
                if (path->slots[0] == i)
-                       push_space += data_size + sizeof(*item);
+                       push_space += data_size;
 
                if (!left->map_token) {
                        map_extent_buffer(left, (unsigned long)item,
@@ -2418,16 +2542,17 @@ static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
                return 1;
 
        right_nritems = btrfs_header_nritems(right);
-       if (right_nritems == 0) {
+       if (right_nritems == 0)
                return 1;
-       }
 
        WARN_ON(!btrfs_tree_locked(path->nodes[1]));
 
        left = read_node_slot(root, path->nodes[1], slot - 1);
        btrfs_tree_lock(left);
+       btrfs_set_lock_blocking(left);
+
        free_space = btrfs_leaf_free_space(root, left);
-       if (free_space < data_size + sizeof(struct btrfs_item)) {
+       if (free_space < data_size) {
                ret = 1;
                goto out;
        }
@@ -2442,7 +2567,7 @@ static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
        }
 
        free_space = btrfs_leaf_free_space(root, left);
-       if (free_space < data_size + sizeof(struct btrfs_item)) {
+       if (free_space < data_size) {
                ret = 1;
                goto out;
        }
@@ -2473,7 +2598,7 @@ static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
                }
 
                if (path->slots[0] == i)
-                       push_space += data_size + sizeof(*item);
+                       push_space += data_size;
 
                this_item_size = btrfs_item_size(right, item);
                if (this_item_size + sizeof(*item) + push_space > free_space)
@@ -2502,7 +2627,7 @@ static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
                           push_items * sizeof(struct btrfs_item));
 
        push_space = BTRFS_LEAF_DATA_SIZE(root) -
-                    btrfs_item_offset_nr(right, push_items -1);
+                    btrfs_item_offset_nr(right, push_items - 1);
 
        copy_extent_buffer(left, right, btrfs_leaf_data(left) +
                     leaf_data_end(root, left) - push_space,
@@ -2510,7 +2635,7 @@ static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
                     btrfs_item_offset_nr(right, push_items - 1),
                     push_space);
        old_left_nritems = btrfs_header_nritems(left);
-       BUG_ON(old_left_nritems < 0);
+       BUG_ON(old_left_nritems <= 0);
 
        old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
        for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
@@ -2537,7 +2662,8 @@ static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
 
        /* fixup right node */
        if (push_items > right_nritems) {
-               printk("push items %d nr %u\n", push_items, right_nritems);
+               printk(KERN_CRIT "push items %d nr %u\n", push_items,
+                      right_nritems);
                WARN_ON(1);
        }
 
@@ -2628,7 +2754,6 @@ static noinline int split_leaf(struct btrfs_trans_handle *trans,
        int mid;
        int slot;
        struct extent_buffer *right;
-       int space_needed = data_size + sizeof(struct btrfs_item);
        int data_copy_size;
        int rt_data_off;
        int i;
@@ -2638,15 +2763,11 @@ static noinline int split_leaf(struct btrfs_trans_handle *trans,
        int num_doubles = 0;
        struct btrfs_disk_key disk_key;
 
-       if (extend && data_size)
-               space_needed = data_size;
-
        /* first try to make some room by pushing left and right */
        if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
                wret = push_leaf_right(trans, root, path, data_size, 0);
-               if (wret < 0) {
+               if (wret < 0)
                        return wret;
-               }
                if (wret) {
                        wret = push_leaf_left(trans, root, path, data_size, 0);
                        if (wret < 0)
@@ -2655,7 +2776,7 @@ static noinline int split_leaf(struct btrfs_trans_handle *trans,
                l = path->nodes[0];
 
                /* did the pushes work? */
-               if (btrfs_leaf_free_space(root, l) >= space_needed)
+               if (btrfs_leaf_free_space(root, l) >= data_size)
                        return 0;
        }
 
@@ -2669,7 +2790,7 @@ again:
        l = path->nodes[0];
        slot = path->slots[0];
        nritems = btrfs_header_nritems(l);
-       mid = (nritems + 1)/ 2;
+       mid = (nritems + 1) / 2;
 
        right = btrfs_alloc_free_block(trans, root, root->leafsize,
                                        path->nodes[1]->start,
@@ -2694,7 +2815,7 @@ again:
                            BTRFS_UUID_SIZE);
        if (mid <= slot) {
                if (nritems == 1 ||
-                   leaf_space_used(l, mid, nritems - mid) + space_needed >
+                   leaf_space_used(l, mid, nritems - mid) + data_size >
                        BTRFS_LEAF_DATA_SIZE(root)) {
                        if (slot >= nritems) {
                                btrfs_cpu_key_to_disk(&disk_key, ins_key);
@@ -2716,12 +2837,12 @@ again:
                        mid = slot;
                        if (mid != nritems &&
                            leaf_space_used(l, mid, nritems - mid) +
-                           space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
+                           data_size > BTRFS_LEAF_DATA_SIZE(root)) {
                                double_split = 1;
                        }
                }
        } else {
-               if (leaf_space_used(l, 0, mid + 1) + space_needed >
+               if (leaf_space_used(l, 0, mid) + data_size >
                        BTRFS_LEAF_DATA_SIZE(root)) {
                        if (!extend && data_size && slot == 0) {
                                btrfs_cpu_key_to_disk(&disk_key, ins_key);
@@ -2738,7 +2859,7 @@ again:
                                path->slots[0] = 0;
                                if (path->slots[1] == 0) {
                                        wret = fixup_low_keys(trans, root,
-                                                  path, &disk_key, 1);
+                                                     path, &disk_key, 1);
                                        if (wret)
                                                ret = wret;
                                }
@@ -2750,7 +2871,7 @@ again:
                                mid = slot;
                                if (mid != nritems &&
                                    leaf_space_used(l, mid, nritems - mid) +
-                                   space_needed > BTRFS_LEAF_DATA_SIZE(root)) {
+                                   data_size > BTRFS_LEAF_DATA_SIZE(root)) {
                                        double_split = 1;
                                }
                        }
@@ -2883,10 +3004,17 @@ int btrfs_split_item(struct btrfs_trans_handle *trans,
                return -EAGAIN;
        }
 
-       ret = split_leaf(trans, root, &orig_key, path, 0, 0);
+       ret = split_leaf(trans, root, &orig_key, path,
+                        sizeof(struct btrfs_item), 1);
        path->keep_locks = 0;
        BUG_ON(ret);
 
+       /*
+        * 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));
 
@@ -3036,8 +3164,8 @@ int btrfs_truncate_item(struct btrfs_trans_handle *trans,
                            BTRFS_FILE_EXTENT_INLINE) {
                                ptr = btrfs_item_ptr_offset(leaf, slot);
                                memmove_extent_buffer(leaf, ptr,
-                                       (unsigned long)fi,
-                                       offsetof(struct btrfs_file_extent_item,
+                                     (unsigned long)fi,
+                                     offsetof(struct btrfs_file_extent_item,
                                                 disk_bytenr));
                        }
                }
@@ -3099,7 +3227,8 @@ int btrfs_extend_item(struct btrfs_trans_handle *trans,
        BUG_ON(slot < 0);
        if (slot >= nritems) {
                btrfs_print_leaf(root, leaf);
-               printk("slot %d too large, nritems %d\n", slot, nritems);
+               printk(KERN_CRIT "slot %d too large, nritems %d\n",
+                      slot, nritems);
                BUG_ON(1);
        }
 
@@ -3169,14 +3298,17 @@ int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
        struct btrfs_disk_key disk_key;
        struct btrfs_key found_key;
 
-       found_key.objectid = 0;
-       nr = min_t(int, nr, BTRFS_NODEPTRS_PER_BLOCK(root));
-
-       for (i = 0; i < nr; i++)
+       for (i = 0; i < nr; i++) {
+               if (total_size + data_size[i] + sizeof(struct btrfs_item) >
+                   BTRFS_LEAF_DATA_SIZE(root)) {
+                       break;
+                       nr = i;
+               }
                total_data += data_size[i];
+               total_size += data_size[i] + sizeof(struct btrfs_item);
+       }
+       BUG_ON(nr == 0);
 
-       total_data = min_t(u32, total_data, BTRFS_LEAF_DATA_SIZE(root));
-       total_size = total_data + (nr * sizeof(struct btrfs_item));
        ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
        if (ret == 0)
                return -EEXIST;
@@ -3218,7 +3350,7 @@ int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
 
                if (old_data < data_end) {
                        btrfs_print_leaf(root, leaf);
-                       printk("slot %d old_data %d data_end %d\n",
+                       printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
                               slot, old_data, data_end);
                        BUG_ON(1);
                }
@@ -3317,9 +3449,8 @@ int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
        unsigned int data_end;
        struct btrfs_disk_key disk_key;
 
-       for (i = 0; i < nr; i++) {
+       for (i = 0; i < nr; i++)
                total_data += data_size[i];
-       }
 
        total_size = total_data + (nr * sizeof(struct btrfs_item));
        ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
@@ -3336,7 +3467,7 @@ int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
 
        if (btrfs_leaf_free_space(root, leaf) < total_size) {
                btrfs_print_leaf(root, leaf);
-               printk("not enough freespace need %u have %d\n",
+               printk(KERN_CRIT "not enough freespace need %u have %d\n",
                       total_size, btrfs_leaf_free_space(root, leaf));
                BUG();
        }
@@ -3349,7 +3480,7 @@ int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
 
                if (old_data < data_end) {
                        btrfs_print_leaf(root, leaf);
-                       printk("slot %d old_data %d data_end %d\n",
+                       printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
                               slot, old_data, data_end);
                        BUG_ON(1);
                }
@@ -3413,6 +3544,7 @@ int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
                BUG();
        }
 out:
+       btrfs_unlock_up_safe(path, 1);
        return ret;
 }
 
@@ -3457,7 +3589,7 @@ static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
        int wret;
 
        nritems = btrfs_header_nritems(parent);
-       if (slot != nritems -1) {
+       if (slot != nritems - 1) {
                memmove_extent_buffer(parent,
                              btrfs_node_key_ptr_offset(slot),
                              btrfs_node_key_ptr_offset(slot + 1),
@@ -3500,15 +3632,22 @@ noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
 {
        int ret;
        u64 root_gen = btrfs_header_generation(path->nodes[1]);
+       u64 parent_start = path->nodes[1]->start;
+       u64 parent_owner = btrfs_header_owner(path->nodes[1]);
 
        ret = del_ptr(trans, root, path, 1, path->slots[1]);
        if (ret)
                return ret;
 
+       /*
+        * btrfs_free_extent is expensive, we want to make sure we
+        * aren't holding any locks when we call it
+        */
+       btrfs_unlock_up_safe(path, 0);
+
        ret = btrfs_free_extent(trans, root, bytenr,
                                btrfs_level_size(root, 0),
-                               path->nodes[1]->start,
-                               btrfs_header_owner(path->nodes[1]),
+                               parent_start, parent_owner,
                                root_gen, 0, 1);
        return ret;
 }
@@ -3614,7 +3753,8 @@ int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 
                        if (btrfs_header_nritems(leaf) == 0) {
                                path->slots[1] = slot;
-                               ret = btrfs_del_leaf(trans, root, path, leaf->start);
+                               ret = btrfs_del_leaf(trans, root, path,
+                                                    leaf->start);
                                BUG_ON(ret);
                                free_extent_buffer(leaf);
                        } else {
@@ -3717,7 +3857,7 @@ again:
                ret = 1;
                goto out;
        }
-       while(1) {
+       while (1) {
                nritems = btrfs_header_nritems(cur);
                level = btrfs_header_level(cur);
                sret = bin_search(cur, min_key, level, &slot);
@@ -3738,7 +3878,7 @@ again:
                 * min_trans parameters.  If it isn't in cache or is too
                 * old, skip to the next one.
                 */
-               while(slot < nritems) {
+               while (slot < nritems) {
                        u64 blockptr;
                        u64 gen;
                        struct extent_buffer *tmp;
@@ -3779,12 +3919,14 @@ find_next_key:
                 */
                if (slot >= nritems) {
                        path->slots[level] = slot;
+                       btrfs_set_path_blocking(path);
                        sret = btrfs_find_next_key(root, path, min_key, level,
                                                  cache_only, min_trans);
                        if (sret == 0) {
                                btrfs_release_path(root, path);
                                goto again;
                        } else {
+                               btrfs_clear_path_blocking(path);
                                goto out;
                        }
                }
@@ -3796,16 +3938,20 @@ find_next_key:
                        unlock_up(path, level, 1);
                        goto out;
                }
+               btrfs_set_path_blocking(path);
                cur = read_node_slot(root, cur, slot);
 
                btrfs_tree_lock(cur);
+
                path->locks[level - 1] = 1;
                path->nodes[level - 1] = cur;
                unlock_up(path, level, 1);
+               btrfs_clear_path_blocking(path);
        }
 out:
        if (ret == 0)
                memcpy(min_key, &found_key, sizeof(found_key));
+       btrfs_set_path_blocking(path);
        return ret;
 }
 
@@ -3830,7 +3976,7 @@ int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
        struct extent_buffer *c;
 
        WARN_ON(!path->keep_locks);
-       while(level < BTRFS_MAX_LEVEL) {
+       while (level < BTRFS_MAX_LEVEL) {
                if (!path->nodes[level])
                        return 1;
 
@@ -3839,9 +3985,8 @@ int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
 next:
                if (slot >= btrfs_header_nritems(c)) {
                        level++;
-                       if (level == BTRFS_MAX_LEVEL) {
+                       if (level == BTRFS_MAX_LEVEL)
                                return 1;
-                       }
                        continue;
                }
                if (level == 0)
@@ -3889,9 +4034,8 @@ int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
        int ret;
 
        nritems = btrfs_header_nritems(path->nodes[0]);
-       if (nritems == 0) {
+       if (nritems == 0)
                return 1;
-       }
 
        btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
 
@@ -3903,6 +4047,7 @@ int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
        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
@@ -3915,7 +4060,7 @@ int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
                goto done;
        }
 
-       while(level < BTRFS_MAX_LEVEL) {
+       while (level < BTRFS_MAX_LEVEL) {
                if (!path->nodes[level])
                        return 1;
 
@@ -3923,9 +4068,8 @@ int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
                c = path->nodes[level];
                if (slot >= btrfs_header_nritems(c)) {
                        level++;
-                       if (level == BTRFS_MAX_LEVEL) {
+                       if (level == BTRFS_MAX_LEVEL)
                                return 1;
-                       }
                        continue;
                }
 
@@ -3934,6 +4078,7 @@ int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
                        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);
@@ -3942,11 +4087,12 @@ int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
                if (!path->skip_locking) {
                        WARN_ON(!btrfs_tree_locked(c));
                        btrfs_tree_lock(next);
+                       btrfs_set_lock_blocking(next);
                }
                break;
        }
        path->slots[level] = slot;
-       while(1) {
+       while (1) {
                level--;
                c = path->nodes[level];
                if (path->locks[level])
@@ -3958,12 +4104,15 @@ int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
                        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);
                if (!path->skip_locking) {
                        WARN_ON(!btrfs_tree_locked(path->nodes[level]));
                        btrfs_tree_lock(next);
+                       btrfs_set_lock_blocking(next);
                }
        }
 done:
@@ -3986,8 +4135,9 @@ int btrfs_previous_item(struct btrfs_root *root,
        u32 nritems;
        int ret;
 
-       while(1) {
+       while (1) {
                if (path->slots[0] == 0) {
+                       btrfs_set_path_blocking(path);
                        ret = btrfs_prev_leaf(root, path);
                        if (ret != 0)
                                return ret;