Btrfs: use commit root when loading free space cache
[pandora-kernel.git] / fs / btrfs / extent-tree.c
index 37e0a80..2b35f8d 100644 (file)
@@ -245,7 +245,7 @@ static int exclude_super_stripes(struct btrfs_root *root,
                cache->bytes_super += stripe_len;
                ret = add_excluded_extent(root, cache->key.objectid,
                                          stripe_len);
-               BUG_ON(ret);
+               BUG_ON(ret); /* -ENOMEM */
        }
 
        for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
@@ -253,13 +253,13 @@ static int exclude_super_stripes(struct btrfs_root *root,
                ret = btrfs_rmap_block(&root->fs_info->mapping_tree,
                                       cache->key.objectid, bytenr,
                                       0, &logical, &nr, &stripe_len);
-               BUG_ON(ret);
+               BUG_ON(ret); /* -ENOMEM */
 
                while (nr--) {
                        cache->bytes_super += stripe_len;
                        ret = add_excluded_extent(root, logical[nr],
                                                  stripe_len);
-                       BUG_ON(ret);
+                       BUG_ON(ret); /* -ENOMEM */
                }
 
                kfree(logical);
@@ -321,7 +321,7 @@ static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
                        total_added += size;
                        ret = btrfs_add_free_space(block_group, start,
                                                   size);
-                       BUG_ON(ret);
+                       BUG_ON(ret); /* -ENOMEM or logic error */
                        start = extent_end + 1;
                } else {
                        break;
@@ -332,7 +332,7 @@ static u64 add_new_free_space(struct btrfs_block_group_cache *block_group,
                size = end - start;
                total_added += size;
                ret = btrfs_add_free_space(block_group, start, size);
-               BUG_ON(ret);
+               BUG_ON(ret); /* -ENOMEM or logic error */
        }
 
        return total_added;
@@ -474,7 +474,8 @@ static int cache_block_group(struct btrfs_block_group_cache *cache,
        int ret = 0;
 
        caching_ctl = kzalloc(sizeof(*caching_ctl), GFP_NOFS);
-       BUG_ON(!caching_ctl);
+       if (!caching_ctl)
+               return -ENOMEM;
 
        INIT_LIST_HEAD(&caching_ctl->list);
        mutex_init(&caching_ctl->mutex);
@@ -528,9 +529,7 @@ static int cache_block_group(struct btrfs_block_group_cache *cache,
         * allocate blocks for the tree root we can't do the fast caching since
         * we likely hold important locks.
         */
-       if (trans && (!trans->transaction->in_commit) &&
-           (root && root != root->fs_info->tree_root) &&
-           btrfs_test_opt(root, SPACE_CACHE)) {
+       if (fs_info->mount_opt & BTRFS_MOUNT_SPACE_CACHE) {
                ret = load_free_space_cache(fs_info, cache);
 
                spin_lock(&cache->lock);
@@ -982,7 +981,7 @@ static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
                                ret = btrfs_next_leaf(root, path);
                                if (ret < 0)
                                        return ret;
-                               BUG_ON(ret > 0);
+                               BUG_ON(ret > 0); /* Corruption */
                                leaf = path->nodes[0];
                        }
                        btrfs_item_key_to_cpu(leaf, &found_key,
@@ -1008,9 +1007,9 @@ static int convert_extent_item_v0(struct btrfs_trans_handle *trans,
                                new_size + extra_size, 1);
        if (ret < 0)
                return ret;
-       BUG_ON(ret);
+       BUG_ON(ret); /* Corruption */
 
-       ret = btrfs_extend_item(trans, root, path, new_size);
+       btrfs_extend_item(trans, root, path, new_size);
 
        leaf = path->nodes[0];
        item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
@@ -1478,7 +1477,11 @@ int lookup_inline_extent_backref(struct btrfs_trans_handle *trans,
                err = ret;
                goto out;
        }
-       BUG_ON(ret);
+       if (ret && !insert) {
+               err = -ENOENT;
+               goto out;
+       }
+       BUG_ON(ret); /* Corruption */
 
        leaf = path->nodes[0];
        item_size = btrfs_item_size_nr(leaf, path->slots[0]);
@@ -1592,13 +1595,13 @@ out:
  * helper to add new inline back ref
  */
 static noinline_for_stack
-int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
-                               struct btrfs_root *root,
-                               struct btrfs_path *path,
-                               struct btrfs_extent_inline_ref *iref,
-                               u64 parent, u64 root_objectid,
-                               u64 owner, u64 offset, int refs_to_add,
-                               struct btrfs_delayed_extent_op *extent_op)
+void setup_inline_extent_backref(struct btrfs_trans_handle *trans,
+                                struct btrfs_root *root,
+                                struct btrfs_path *path,
+                                struct btrfs_extent_inline_ref *iref,
+                                u64 parent, u64 root_objectid,
+                                u64 owner, u64 offset, int refs_to_add,
+                                struct btrfs_delayed_extent_op *extent_op)
 {
        struct extent_buffer *leaf;
        struct btrfs_extent_item *ei;
@@ -1608,7 +1611,6 @@ int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
        u64 refs;
        int size;
        int type;
-       int ret;
 
        leaf = path->nodes[0];
        ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
@@ -1617,7 +1619,7 @@ int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
        type = extent_ref_type(parent, owner);
        size = btrfs_extent_inline_ref_size(type);
 
-       ret = btrfs_extend_item(trans, root, path, size);
+       btrfs_extend_item(trans, root, path, size);
 
        ei = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_item);
        refs = btrfs_extent_refs(leaf, ei);
@@ -1652,7 +1654,6 @@ int setup_inline_extent_backref(struct btrfs_trans_handle *trans,
                btrfs_set_extent_inline_ref_offset(leaf, iref, root_objectid);
        }
        btrfs_mark_buffer_dirty(leaf);
-       return 0;
 }
 
 static int lookup_extent_backref(struct btrfs_trans_handle *trans,
@@ -1687,12 +1688,12 @@ static int lookup_extent_backref(struct btrfs_trans_handle *trans,
  * helper to update/remove inline back ref
  */
 static noinline_for_stack
-int update_inline_extent_backref(struct btrfs_trans_handle *trans,
-                                struct btrfs_root *root,
-                                struct btrfs_path *path,
-                                struct btrfs_extent_inline_ref *iref,
-                                int refs_to_mod,
-                                struct btrfs_delayed_extent_op *extent_op)
+void update_inline_extent_backref(struct btrfs_trans_handle *trans,
+                                 struct btrfs_root *root,
+                                 struct btrfs_path *path,
+                                 struct btrfs_extent_inline_ref *iref,
+                                 int refs_to_mod,
+                                 struct btrfs_delayed_extent_op *extent_op)
 {
        struct extent_buffer *leaf;
        struct btrfs_extent_item *ei;
@@ -1703,7 +1704,6 @@ int update_inline_extent_backref(struct btrfs_trans_handle *trans,
        u32 item_size;
        int size;
        int type;
-       int ret;
        u64 refs;
 
        leaf = path->nodes[0];
@@ -1745,10 +1745,9 @@ int update_inline_extent_backref(struct btrfs_trans_handle *trans,
                        memmove_extent_buffer(leaf, ptr, ptr + size,
                                              end - ptr - size);
                item_size -= size;
-               ret = btrfs_truncate_item(trans, root, path, item_size, 1);
+               btrfs_truncate_item(trans, root, path, item_size, 1);
        }
        btrfs_mark_buffer_dirty(leaf);
-       return 0;
 }
 
 static noinline_for_stack
@@ -1768,13 +1767,13 @@ int insert_inline_extent_backref(struct btrfs_trans_handle *trans,
                                           root_objectid, owner, offset, 1);
        if (ret == 0) {
                BUG_ON(owner < BTRFS_FIRST_FREE_OBJECTID);
-               ret = update_inline_extent_backref(trans, root, path, iref,
-                                                  refs_to_add, extent_op);
+               update_inline_extent_backref(trans, root, path, iref,
+                                            refs_to_add, extent_op);
        } else if (ret == -ENOENT) {
-               ret = setup_inline_extent_backref(trans, root, path, iref,
-                                                 parent, root_objectid,
-                                                 owner, offset, refs_to_add,
-                                                 extent_op);
+               setup_inline_extent_backref(trans, root, path, iref, parent,
+                                           root_objectid, owner, offset,
+                                           refs_to_add, extent_op);
+               ret = 0;
        }
        return ret;
 }
@@ -1804,12 +1803,12 @@ static int remove_extent_backref(struct btrfs_trans_handle *trans,
                                 struct btrfs_extent_inline_ref *iref,
                                 int refs_to_drop, int is_data)
 {
-       int ret;
+       int ret = 0;
 
        BUG_ON(!is_data && refs_to_drop != 1);
        if (iref) {
-               ret = update_inline_extent_backref(trans, root, path, iref,
-                                                  -refs_to_drop, NULL);
+               update_inline_extent_backref(trans, root, path, iref,
+                                            -refs_to_drop, NULL);
        } else if (is_data) {
                ret = remove_extent_data_ref(trans, root, path, refs_to_drop);
        } else {
@@ -1835,6 +1834,7 @@ static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
        /* Tell the block device(s) that the sectors can be discarded */
        ret = btrfs_map_block(&root->fs_info->mapping_tree, REQ_DISCARD,
                              bytenr, &num_bytes, &bbio, 0);
+       /* Error condition is -ENOMEM */
        if (!ret) {
                struct btrfs_bio_stripe *stripe = bbio->stripes;
                int i;
@@ -1850,7 +1850,7 @@ static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
                        if (!ret)
                                discarded_bytes += stripe->length;
                        else if (ret != -EOPNOTSUPP)
-                               break;
+                               break; /* Logic errors or -ENOMEM, or -EIO but I don't know how that could happen JDM */
 
                        /*
                         * Just in case we get back EOPNOTSUPP for some reason,
@@ -1869,6 +1869,7 @@ static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr,
        return ret;
 }
 
+/* Can return -ENOMEM */
 int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
                         struct btrfs_root *root,
                         u64 bytenr, u64 num_bytes, u64 parent,
@@ -1944,7 +1945,8 @@ static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans,
        ret = insert_extent_backref(trans, root->fs_info->extent_root,
                                    path, bytenr, parent, root_objectid,
                                    owner, offset, refs_to_add);
-       BUG_ON(ret);
+       if (ret)
+               btrfs_abort_transaction(trans, root, ret);
 out:
        btrfs_free_path(path);
        return err;
@@ -2031,6 +2033,9 @@ static int run_delayed_extent_op(struct btrfs_trans_handle *trans,
        int ret;
        int err = 0;
 
+       if (trans->aborted)
+               return 0;
+
        path = btrfs_alloc_path();
        if (!path)
                return -ENOMEM;
@@ -2128,7 +2133,11 @@ static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
                               struct btrfs_delayed_extent_op *extent_op,
                               int insert_reserved)
 {
-       int ret;
+       int ret = 0;
+
+       if (trans->aborted)
+               return 0;
+
        if (btrfs_delayed_ref_is_head(node)) {
                struct btrfs_delayed_ref_head *head;
                /*
@@ -2146,11 +2155,10 @@ static int run_one_delayed_ref(struct btrfs_trans_handle *trans,
                                ret = btrfs_del_csums(trans, root,
                                                      node->bytenr,
                                                      node->num_bytes);
-                               BUG_ON(ret);
                        }
                }
                mutex_unlock(&head->mutex);
-               return 0;
+               return ret;
        }
 
        if (node->type == BTRFS_TREE_BLOCK_REF_KEY ||
@@ -2197,6 +2205,10 @@ again:
        return NULL;
 }
 
+/*
+ * Returns 0 on success or if called with an already aborted transaction.
+ * Returns -ENOMEM or -EIO on failure and will abort the transaction.
+ */
 static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
                                       struct btrfs_root *root,
                                       struct list_head *cluster)
@@ -2285,9 +2297,13 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
 
                                ret = run_delayed_extent_op(trans, root,
                                                            ref, extent_op);
-                               BUG_ON(ret);
                                kfree(extent_op);
 
+                               if (ret) {
+                                       printk(KERN_DEBUG "btrfs: run_delayed_extent_op returned %d\n", ret);
+                                       return ret;
+                               }
+
                                goto next;
                        }
 
@@ -2308,11 +2324,16 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans,
 
                ret = run_one_delayed_ref(trans, root, ref, extent_op,
                                          must_insert_reserved);
-               BUG_ON(ret);
 
                btrfs_put_delayed_ref(ref);
                kfree(extent_op);
                count++;
+
+               if (ret) {
+                       printk(KERN_DEBUG "btrfs: run_one_delayed_ref returned %d\n", ret);
+                       return ret;
+               }
+
 next:
                do_chunk_alloc(trans, root->fs_info->extent_root,
                               2 * 1024 * 1024,
@@ -2347,6 +2368,9 @@ static void wait_for_more_refs(struct btrfs_delayed_ref_root *delayed_refs,
  * 0, which means to process everything in the tree at the start
  * of the run (but not newly added entries), or it can be some target
  * number you'd like to process.
+ *
+ * Returns 0 on success or if called with an aborted transaction
+ * Returns <0 on error and aborts the transaction
  */
 int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
                           struct btrfs_root *root, unsigned long count)
@@ -2362,6 +2386,10 @@ int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans,
        unsigned long num_refs = 0;
        int consider_waiting;
 
+       /* We'll clean this up in btrfs_cleanup_transaction */
+       if (trans->aborted)
+               return 0;
+
        if (root == root->fs_info->extent_root)
                root = root->fs_info->tree_root;
 
@@ -2419,7 +2447,11 @@ again:
                }
 
                ret = run_clustered_refs(trans, root, &cluster);
-               BUG_ON(ret < 0);
+               if (ret < 0) {
+                       spin_unlock(&delayed_refs->lock);
+                       btrfs_abort_transaction(trans, root, ret);
+                       return ret;
+               }
 
                count -= min_t(unsigned long, ret, count);
 
@@ -2584,7 +2616,7 @@ static noinline int check_committed_ref(struct btrfs_trans_handle *trans,
        ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
        if (ret < 0)
                goto out;
-       BUG_ON(ret == 0);
+       BUG_ON(ret == 0); /* Corruption */
 
        ret = -ENOENT;
        if (path->slots[0] == 0)
@@ -2738,7 +2770,6 @@ static int __btrfs_mod_ref(struct btrfs_trans_handle *trans,
        }
        return 0;
 fail:
-       BUG();
        return ret;
 }
 
@@ -2767,7 +2798,7 @@ static int write_one_cache_group(struct btrfs_trans_handle *trans,
        ret = btrfs_search_slot(trans, extent_root, &cache->key, path, 0, 1);
        if (ret < 0)
                goto fail;
-       BUG_ON(ret);
+       BUG_ON(ret); /* Corruption */
 
        leaf = path->nodes[0];
        bi = btrfs_item_ptr_offset(leaf, path->slots[0]);
@@ -2775,8 +2806,10 @@ static int write_one_cache_group(struct btrfs_trans_handle *trans,
        btrfs_mark_buffer_dirty(leaf);
        btrfs_release_path(path);
 fail:
-       if (ret)
+       if (ret) {
+               btrfs_abort_transaction(trans, root, ret);
                return ret;
+       }
        return 0;
 
 }
@@ -2949,7 +2982,8 @@ again:
                if (last == 0) {
                        err = btrfs_run_delayed_refs(trans, root,
                                                     (unsigned long)-1);
-                       BUG_ON(err);
+                       if (err) /* File system offline */
+                               goto out;
                }
 
                cache = btrfs_lookup_first_block_group(root->fs_info, last);
@@ -2976,7 +3010,9 @@ again:
                last = cache->key.objectid + cache->key.offset;
 
                err = write_one_cache_group(trans, root, path, cache);
-               BUG_ON(err);
+               if (err) /* File system offline */
+                       goto out;
+
                btrfs_put_block_group(cache);
        }
 
@@ -2989,7 +3025,8 @@ again:
                if (last == 0) {
                        err = btrfs_run_delayed_refs(trans, root,
                                                     (unsigned long)-1);
-                       BUG_ON(err);
+                       if (err) /* File system offline */
+                               goto out;
                }
 
                cache = btrfs_lookup_first_block_group(root->fs_info, last);
@@ -3014,20 +3051,21 @@ again:
                        continue;
                }
 
-               btrfs_write_out_cache(root, trans, cache, path);
+               err = btrfs_write_out_cache(root, trans, cache, path);
 
                /*
                 * If we didn't have an error then the cache state is still
                 * NEED_WRITE, so we can set it to WRITTEN.
                 */
-               if (cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
+               if (!err && cache->disk_cache_state == BTRFS_DC_NEED_WRITE)
                        cache->disk_cache_state = BTRFS_DC_WRITTEN;
                last = cache->key.objectid + cache->key.offset;
                btrfs_put_block_group(cache);
        }
+out:
 
        btrfs_free_path(path);
-       return 0;
+       return err;
 }
 
 int btrfs_extent_readonly(struct btrfs_root *root, u64 bytenr)
@@ -3098,11 +3136,8 @@ static int update_space_info(struct btrfs_fs_info *info, u64 flags,
 
 static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
 {
-       u64 extra_flags = flags & BTRFS_BLOCK_GROUP_PROFILE_MASK;
-
-       /* chunk -> extended profile */
-       if (extra_flags == 0)
-               extra_flags = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
+       u64 extra_flags = chunk_to_extended(flags) &
+                               BTRFS_EXTENDED_PROFILE_MASK;
 
        if (flags & BTRFS_BLOCK_GROUP_DATA)
                fs_info->avail_data_alloc_bits |= extra_flags;
@@ -3112,6 +3147,34 @@ static void set_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
                fs_info->avail_system_alloc_bits |= extra_flags;
 }
 
+/*
+ * returns target flags in extended format or 0 if restripe for this
+ * chunk_type is not in progress
+ *
+ * should be called with either volume_mutex or balance_lock held
+ */
+static u64 get_restripe_target(struct btrfs_fs_info *fs_info, u64 flags)
+{
+       struct btrfs_balance_control *bctl = fs_info->balance_ctl;
+       u64 target = 0;
+
+       if (!bctl)
+               return 0;
+
+       if (flags & BTRFS_BLOCK_GROUP_DATA &&
+           bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
+               target = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
+       } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
+                  bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
+               target = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
+       } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
+                  bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
+               target = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
+       }
+
+       return target;
+}
+
 /*
  * @flags: available profiles in extended format (see ctree.h)
  *
@@ -3128,31 +3191,19 @@ u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
         */
        u64 num_devices = root->fs_info->fs_devices->rw_devices +
                root->fs_info->fs_devices->missing_devices;
+       u64 target;
 
-       /* pick restriper's target profile if it's available */
+       /*
+        * see if restripe for this chunk_type is in progress, if so
+        * try to reduce to the target profile
+        */
        spin_lock(&root->fs_info->balance_lock);
-       if (root->fs_info->balance_ctl) {
-               struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
-               u64 tgt = 0;
-
-               if ((flags & BTRFS_BLOCK_GROUP_DATA) &&
-                   (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
-                   (flags & bctl->data.target)) {
-                       tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
-               } else if ((flags & BTRFS_BLOCK_GROUP_SYSTEM) &&
-                          (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
-                          (flags & bctl->sys.target)) {
-                       tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
-               } else if ((flags & BTRFS_BLOCK_GROUP_METADATA) &&
-                          (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
-                          (flags & bctl->meta.target)) {
-                       tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
-               }
-
-               if (tgt) {
+       target = get_restripe_target(root->fs_info, flags);
+       if (target) {
+               /* pick target profile only if it's already available */
+               if ((flags & target) & BTRFS_EXTENDED_PROFILE_MASK) {
                        spin_unlock(&root->fs_info->balance_lock);
-                       flags = tgt;
-                       goto out;
+                       return extended_to_chunk(target);
                }
        }
        spin_unlock(&root->fs_info->balance_lock);
@@ -3180,10 +3231,7 @@ u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
                flags &= ~BTRFS_BLOCK_GROUP_RAID0;
        }
 
-out:
-       /* extended -> chunk profile */
-       flags &= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE;
-       return flags;
+       return extended_to_chunk(flags);
 }
 
 static u64 get_alloc_profile(struct btrfs_root *root, u64 flags)
@@ -3312,8 +3360,7 @@ commit_trans:
        }
        data_sinfo->bytes_may_use += bytes;
        trace_btrfs_space_reservation(root->fs_info, "space_info",
-                                     (u64)(unsigned long)data_sinfo,
-                                     bytes, 1);
+                                     data_sinfo->flags, bytes, 1);
        spin_unlock(&data_sinfo->lock);
 
        return 0;
@@ -3334,8 +3381,7 @@ void btrfs_free_reserved_data_space(struct inode *inode, u64 bytes)
        spin_lock(&data_sinfo->lock);
        data_sinfo->bytes_may_use -= bytes;
        trace_btrfs_space_reservation(root->fs_info, "space_info",
-                                     (u64)(unsigned long)data_sinfo,
-                                     bytes, 0);
+                                     data_sinfo->flags, bytes, 0);
        spin_unlock(&data_sinfo->lock);
 }
 
@@ -3396,6 +3442,50 @@ static int should_alloc_chunk(struct btrfs_root *root,
        return 1;
 }
 
+static u64 get_system_chunk_thresh(struct btrfs_root *root, u64 type)
+{
+       u64 num_dev;
+
+       if (type & BTRFS_BLOCK_GROUP_RAID10 ||
+           type & BTRFS_BLOCK_GROUP_RAID0)
+               num_dev = root->fs_info->fs_devices->rw_devices;
+       else if (type & BTRFS_BLOCK_GROUP_RAID1)
+               num_dev = 2;
+       else
+               num_dev = 1;    /* DUP or single */
+
+       /* metadata for updaing devices and chunk tree */
+       return btrfs_calc_trans_metadata_size(root, num_dev + 1);
+}
+
+static void check_system_chunk(struct btrfs_trans_handle *trans,
+                              struct btrfs_root *root, u64 type)
+{
+       struct btrfs_space_info *info;
+       u64 left;
+       u64 thresh;
+
+       info = __find_space_info(root->fs_info, BTRFS_BLOCK_GROUP_SYSTEM);
+       spin_lock(&info->lock);
+       left = info->total_bytes - info->bytes_used - info->bytes_pinned -
+               info->bytes_reserved - info->bytes_readonly;
+       spin_unlock(&info->lock);
+
+       thresh = get_system_chunk_thresh(root, type);
+       if (left < thresh && btrfs_test_opt(root, ENOSPC_DEBUG)) {
+               printk(KERN_INFO "left=%llu, need=%llu, flags=%llu\n",
+                      left, thresh, type);
+               dump_space_info(info, 0, 0);
+       }
+
+       if (left < thresh) {
+               u64 flags;
+
+               flags = btrfs_get_alloc_profile(root->fs_info->chunk_root, 0);
+               btrfs_alloc_chunk(trans, root, flags);
+       }
+}
+
 static int do_chunk_alloc(struct btrfs_trans_handle *trans,
                          struct btrfs_root *extent_root, u64 alloc_bytes,
                          u64 flags, int force)
@@ -3405,15 +3495,13 @@ static int do_chunk_alloc(struct btrfs_trans_handle *trans,
        int wait_for_alloc = 0;
        int ret = 0;
 
-       BUG_ON(!profile_is_valid(flags, 0));
-
        space_info = __find_space_info(extent_root->fs_info, flags);
        if (!space_info) {
                ret = update_space_info(extent_root->fs_info, flags,
                                        0, 0, &space_info);
-               BUG_ON(ret);
+               BUG_ON(ret); /* -ENOMEM */
        }
-       BUG_ON(!space_info);
+       BUG_ON(!space_info); /* Logic error */
 
 again:
        spin_lock(&space_info->lock);
@@ -3468,6 +3556,12 @@ again:
                        force_metadata_allocation(fs_info);
        }
 
+       /*
+        * Check if we have enough space in SYSTEM chunk because we may need
+        * to update devices.
+        */
+       check_system_chunk(trans, extent_root, flags);
+
        ret = btrfs_alloc_chunk(trans, extent_root, flags);
        if (ret < 0 && ret != -ENOSPC)
                goto out;
@@ -3678,8 +3772,10 @@ again:
                ret = wait_event_interruptible(space_info->wait,
                                               !space_info->flush);
                /* Must have been interrupted, return */
-               if (ret)
+               if (ret) {
+                       printk(KERN_DEBUG "btrfs: %s returning -EINTR\n", __func__);
                        return -EINTR;
+               }
 
                spin_lock(&space_info->lock);
        }
@@ -3700,9 +3796,7 @@ again:
                if (used + orig_bytes <= space_info->total_bytes) {
                        space_info->bytes_may_use += orig_bytes;
                        trace_btrfs_space_reservation(root->fs_info,
-                                             "space_info",
-                                             (u64)(unsigned long)space_info,
-                                             orig_bytes, 1);
+                               "space_info", space_info->flags, orig_bytes, 1);
                        ret = 0;
                } else {
                        /*
@@ -3771,9 +3865,7 @@ again:
                if (used + num_bytes < space_info->total_bytes + avail) {
                        space_info->bytes_may_use += orig_bytes;
                        trace_btrfs_space_reservation(root->fs_info,
-                                             "space_info",
-                                             (u64)(unsigned long)space_info,
-                                             orig_bytes, 1);
+                               "space_info", space_info->flags, orig_bytes, 1);
                        ret = 0;
                } else {
                        wait_ordered = true;
@@ -3836,8 +3928,9 @@ out:
        return ret;
 }
 
-static struct btrfs_block_rsv *get_block_rsv(struct btrfs_trans_handle *trans,
-                                            struct btrfs_root *root)
+static struct btrfs_block_rsv *get_block_rsv(
+                                       const struct btrfs_trans_handle *trans,
+                                       const struct btrfs_root *root)
 {
        struct btrfs_block_rsv *block_rsv = NULL;
 
@@ -3918,8 +4011,7 @@ static void block_rsv_release_bytes(struct btrfs_fs_info *fs_info,
                        spin_lock(&space_info->lock);
                        space_info->bytes_may_use -= num_bytes;
                        trace_btrfs_space_reservation(fs_info, "space_info",
-                                             (u64)(unsigned long)space_info,
-                                             num_bytes, 0);
+                                       space_info->flags, num_bytes, 0);
                        space_info->reservation_progress++;
                        spin_unlock(&space_info->lock);
                }
@@ -4110,7 +4202,7 @@ static u64 calc_global_metadata_size(struct btrfs_fs_info *fs_info)
        num_bytes += div64_u64(data_used + meta_used, 50);
 
        if (num_bytes * 3 > meta_used)
-               num_bytes = div64_u64(meta_used, 3) * 2;
+               num_bytes = div64_u64(meta_used, 3);
 
        return ALIGN(num_bytes, fs_info->extent_root->leafsize << 10);
 }
@@ -4137,14 +4229,14 @@ static void update_global_block_rsv(struct btrfs_fs_info *fs_info)
                block_rsv->reserved += num_bytes;
                sinfo->bytes_may_use += num_bytes;
                trace_btrfs_space_reservation(fs_info, "space_info",
-                                     (u64)(unsigned long)sinfo, num_bytes, 1);
+                                     sinfo->flags, num_bytes, 1);
        }
 
        if (block_rsv->reserved >= block_rsv->size) {
                num_bytes = block_rsv->reserved - block_rsv->size;
                sinfo->bytes_may_use -= num_bytes;
                trace_btrfs_space_reservation(fs_info, "space_info",
-                                     (u64)(unsigned long)sinfo, num_bytes, 0);
+                                     sinfo->flags, num_bytes, 0);
                sinfo->reservation_progress++;
                block_rsv->reserved = block_rsv->size;
                block_rsv->full = 1;
@@ -4198,12 +4290,12 @@ void btrfs_trans_release_metadata(struct btrfs_trans_handle *trans,
                return;
 
        trace_btrfs_space_reservation(root->fs_info, "transaction",
-                                     (u64)(unsigned long)trans,
-                                     trans->bytes_reserved, 0);
+                                     trans->transid, trans->bytes_reserved, 0);
        btrfs_block_rsv_release(root, trans->block_rsv, trans->bytes_reserved);
        trans->bytes_reserved = 0;
 }
 
+/* Can only return 0 or -ENOSPC */
 int btrfs_orphan_reserve_metadata(struct btrfs_trans_handle *trans,
                                  struct inode *inode)
 {
@@ -4540,7 +4632,7 @@ static int update_block_group(struct btrfs_trans_handle *trans,
        while (total) {
                cache = btrfs_lookup_block_group(info, bytenr);
                if (!cache)
-                       return -1;
+                       return -ENOENT;
                if (cache->flags & (BTRFS_BLOCK_GROUP_DUP |
                                    BTRFS_BLOCK_GROUP_RAID1 |
                                    BTRFS_BLOCK_GROUP_RAID10))
@@ -4643,7 +4735,7 @@ int btrfs_pin_extent(struct btrfs_root *root,
        struct btrfs_block_group_cache *cache;
 
        cache = btrfs_lookup_block_group(root->fs_info, bytenr);
-       BUG_ON(!cache);
+       BUG_ON(!cache); /* Logic error */
 
        pin_down_extent(root, cache, bytenr, num_bytes, reserved);
 
@@ -4661,7 +4753,7 @@ int btrfs_pin_extent_for_log_replay(struct btrfs_trans_handle *trans,
        struct btrfs_block_group_cache *cache;
 
        cache = btrfs_lookup_block_group(root->fs_info, bytenr);
-       BUG_ON(!cache);
+       BUG_ON(!cache); /* Logic error */
 
        /*
         * pull in the free space cache (if any) so that our pin
@@ -4706,6 +4798,7 @@ static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
 {
        struct btrfs_space_info *space_info = cache->space_info;
        int ret = 0;
+
        spin_lock(&space_info->lock);
        spin_lock(&cache->lock);
        if (reserve != RESERVE_FREE) {
@@ -4716,9 +4809,8 @@ static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
                        space_info->bytes_reserved += num_bytes;
                        if (reserve == RESERVE_ALLOC) {
                                trace_btrfs_space_reservation(cache->fs_info,
-                                             "space_info",
-                                             (u64)(unsigned long)space_info,
-                                             num_bytes, 0);
+                                               "space_info", space_info->flags,
+                                               num_bytes, 0);
                                space_info->bytes_may_use -= num_bytes;
                        }
                }
@@ -4734,7 +4826,7 @@ static int btrfs_update_reserved_bytes(struct btrfs_block_group_cache *cache,
        return ret;
 }
 
-int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
+void btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
                                struct btrfs_root *root)
 {
        struct btrfs_fs_info *fs_info = root->fs_info;
@@ -4764,7 +4856,6 @@ int btrfs_prepare_extent_commit(struct btrfs_trans_handle *trans,
        up_write(&fs_info->extent_commit_sem);
 
        update_global_block_rsv(fs_info);
-       return 0;
 }
 
 static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
@@ -4779,7 +4870,7 @@ static int unpin_extent_range(struct btrfs_root *root, u64 start, u64 end)
                        if (cache)
                                btrfs_put_block_group(cache);
                        cache = btrfs_lookup_block_group(fs_info, start);
-                       BUG_ON(!cache);
+                       BUG_ON(!cache); /* Logic error */
                }
 
                len = cache->key.objectid + cache->key.offset - start;
@@ -4816,6 +4907,9 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans,
        u64 end;
        int ret;
 
+       if (trans->aborted)
+               return 0;
+
        if (fs_info->pinned_extents == &fs_info->freed_extents[0])
                unpin = &fs_info->freed_extents[1];
        else
@@ -4901,7 +4995,8 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
                        ret = remove_extent_backref(trans, extent_root, path,
                                                    NULL, refs_to_drop,
                                                    is_data);
-                       BUG_ON(ret);
+                       if (ret)
+                               goto abort;
                        btrfs_release_path(path);
                        path->leave_spinning = 1;
 
@@ -4919,10 +5014,11 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
                                        btrfs_print_leaf(extent_root,
                                                         path->nodes[0]);
                        }
-                       BUG_ON(ret);
+                       if (ret < 0)
+                               goto abort;
                        extent_slot = path->slots[0];
                }
-       } else {
+       } else if (ret == -ENOENT) {
                btrfs_print_leaf(extent_root, path->nodes[0]);
                WARN_ON(1);
                printk(KERN_ERR "btrfs unable to find ref byte nr %llu "
@@ -4932,6 +5028,8 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
                       (unsigned long long)root_objectid,
                       (unsigned long long)owner_objectid,
                       (unsigned long long)owner_offset);
+       } else {
+               goto abort;
        }
 
        leaf = path->nodes[0];
@@ -4941,7 +5039,8 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
                BUG_ON(found_extent || extent_slot != path->slots[0]);
                ret = convert_extent_item_v0(trans, extent_root, path,
                                             owner_objectid, 0);
-               BUG_ON(ret < 0);
+               if (ret < 0)
+                       goto abort;
 
                btrfs_release_path(path);
                path->leave_spinning = 1;
@@ -4958,7 +5057,8 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
                               (unsigned long long)bytenr);
                        btrfs_print_leaf(extent_root, path->nodes[0]);
                }
-               BUG_ON(ret);
+               if (ret < 0)
+                       goto abort;
                extent_slot = path->slots[0];
                leaf = path->nodes[0];
                item_size = btrfs_item_size_nr(leaf, extent_slot);
@@ -4995,7 +5095,8 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
                        ret = remove_extent_backref(trans, extent_root, path,
                                                    iref, refs_to_drop,
                                                    is_data);
-                       BUG_ON(ret);
+                       if (ret)
+                               goto abort;
                }
        } else {
                if (found_extent) {
@@ -5012,23 +5113,27 @@ static int __btrfs_free_extent(struct btrfs_trans_handle *trans,
 
                ret = btrfs_del_items(trans, extent_root, path, path->slots[0],
                                      num_to_del);
-               BUG_ON(ret);
+               if (ret)
+                       goto abort;
                btrfs_release_path(path);
 
                if (is_data) {
                        ret = btrfs_del_csums(trans, root, bytenr, num_bytes);
-                       BUG_ON(ret);
-               } else {
-                       invalidate_mapping_pages(info->btree_inode->i_mapping,
-                            bytenr >> PAGE_CACHE_SHIFT,
-                            (bytenr + num_bytes - 1) >> PAGE_CACHE_SHIFT);
+                       if (ret)
+                               goto abort;
                }
 
                ret = update_block_group(trans, root, bytenr, num_bytes, 0);
-               BUG_ON(ret);
+               if (ret)
+                       goto abort;
        }
+out:
        btrfs_free_path(path);
        return ret;
+
+abort:
+       btrfs_abort_transaction(trans, extent_root, ret);
+       goto out;
 }
 
 /*
@@ -5124,7 +5229,7 @@ void btrfs_free_tree_block(struct btrfs_trans_handle *trans,
                                        parent, root->root_key.objectid,
                                        btrfs_header_level(buf),
                                        BTRFS_DROP_DELAYED_REF, NULL, for_cow);
-               BUG_ON(ret);
+               BUG_ON(ret); /* -ENOMEM */
        }
 
        if (!last_ref)
@@ -5158,6 +5263,7 @@ out:
        btrfs_put_block_group(cache);
 }
 
+/* Can return -ENOMEM */
 int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
                      u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid,
                      u64 owner, u64 offset, int for_cow)
@@ -5179,14 +5285,12 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root,
                                        num_bytes,
                                        parent, root_objectid, (int)owner,
                                        BTRFS_DROP_DELAYED_REF, NULL, for_cow);
-               BUG_ON(ret);
        } else {
                ret = btrfs_add_delayed_data_ref(fs_info, trans, bytenr,
                                                num_bytes,
                                                parent, root_objectid, owner,
                                                offset, BTRFS_DROP_DELAYED_REF,
                                                NULL, for_cow);
-               BUG_ON(ret);
        }
        return ret;
 }
@@ -5243,28 +5347,34 @@ wait_block_group_cache_done(struct btrfs_block_group_cache *cache)
        return 0;
 }
 
-static int get_block_group_index(struct btrfs_block_group_cache *cache)
+static int __get_block_group_index(u64 flags)
 {
        int index;
-       if (cache->flags & BTRFS_BLOCK_GROUP_RAID10)
+
+       if (flags & BTRFS_BLOCK_GROUP_RAID10)
                index = 0;
-       else if (cache->flags & BTRFS_BLOCK_GROUP_RAID1)
+       else if (flags & BTRFS_BLOCK_GROUP_RAID1)
                index = 1;
-       else if (cache->flags & BTRFS_BLOCK_GROUP_DUP)
+       else if (flags & BTRFS_BLOCK_GROUP_DUP)
                index = 2;
-       else if (cache->flags & BTRFS_BLOCK_GROUP_RAID0)
+       else if (flags & BTRFS_BLOCK_GROUP_RAID0)
                index = 3;
        else
                index = 4;
+
        return index;
 }
 
+static int get_block_group_index(struct btrfs_block_group_cache *cache)
+{
+       return __get_block_group_index(cache->flags);
+}
+
 enum btrfs_loop_type {
-       LOOP_FIND_IDEAL = 0,
-       LOOP_CACHING_NOWAIT = 1,
-       LOOP_CACHING_WAIT = 2,
-       LOOP_ALLOC_CHUNK = 3,
-       LOOP_NO_EMPTY_SIZE = 4,
+       LOOP_CACHING_NOWAIT = 0,
+       LOOP_CACHING_WAIT = 1,
+       LOOP_ALLOC_CHUNK = 2,
+       LOOP_NO_EMPTY_SIZE = 3,
 };
 
 /*
@@ -5278,7 +5388,6 @@ enum btrfs_loop_type {
 static noinline int find_free_extent(struct btrfs_trans_handle *trans,
                                     struct btrfs_root *orig_root,
                                     u64 num_bytes, u64 empty_size,
-                                    u64 search_start, u64 search_end,
                                     u64 hint_byte, struct btrfs_key *ins,
                                     u64 data)
 {
@@ -5287,6 +5396,7 @@ static noinline int find_free_extent(struct btrfs_trans_handle *trans,
        struct btrfs_free_cluster *last_ptr = NULL;
        struct btrfs_block_group_cache *block_group = NULL;
        struct btrfs_block_group_cache *used_block_group;
+       u64 search_start = 0;
        int empty_cluster = 2 * 1024 * 1024;
        int allowed_chunk_alloc = 0;
        int done_chunk_alloc = 0;
@@ -5300,8 +5410,6 @@ static noinline int find_free_extent(struct btrfs_trans_handle *trans,
        bool failed_alloc = false;
        bool use_cluster = true;
        bool have_caching_bg = false;
-       u64 ideal_cache_percent = 0;
-       u64 ideal_cache_offset = 0;
 
        WARN_ON(num_bytes < root->sectorsize);
        btrfs_set_key_type(ins, BTRFS_EXTENT_ITEM_KEY);
@@ -5351,7 +5459,6 @@ static noinline int find_free_extent(struct btrfs_trans_handle *trans,
                empty_cluster = 0;
 
        if (search_start == hint_byte) {
-ideal_cache:
                block_group = btrfs_lookup_block_group(root->fs_info,
                                                       search_start);
                used_block_group = block_group;
@@ -5363,8 +5470,7 @@ ideal_cache:
                 * picked out then we don't care that the block group is cached.
                 */
                if (block_group && block_group_bits(block_group, data) &&
-                   (block_group->cached != BTRFS_CACHE_NO ||
-                    search_start == ideal_cache_offset)) {
+                   block_group->cached != BTRFS_CACHE_NO) {
                        down_read(&space_info->groups_sem);
                        if (list_empty(&block_group->list) ||
                            block_group->ro) {
@@ -5418,44 +5524,13 @@ search:
 have_block_group:
                cached = block_group_cache_done(block_group);
                if (unlikely(!cached)) {
-                       u64 free_percent;
-
                        found_uncached_bg = true;
                        ret = cache_block_group(block_group, trans,
-                                               orig_root, 1);
-                       if (block_group->cached == BTRFS_CACHE_FINISHED)
-                               goto alloc;
-
-                       free_percent = btrfs_block_group_used(&block_group->item);
-                       free_percent *= 100;
-                       free_percent = div64_u64(free_percent,
-                                                block_group->key.offset);
-                       free_percent = 100 - free_percent;
-                       if (free_percent > ideal_cache_percent &&
-                           likely(!block_group->ro)) {
-                               ideal_cache_offset = block_group->key.objectid;
-                               ideal_cache_percent = free_percent;
-                       }
-
-                       /*
-                        * The caching workers are limited to 2 threads, so we
-                        * can queue as much work as we care to.
-                        */
-                       if (loop > LOOP_FIND_IDEAL) {
-                               ret = cache_block_group(block_group, trans,
-                                                       orig_root, 0);
-                               BUG_ON(ret);
-                       }
-
-                       /*
-                        * If loop is set for cached only, try the next block
-                        * group.
-                        */
-                       if (loop == LOOP_FIND_IDEAL)
-                               goto loop;
+                                               orig_root, 0);
+                       BUG_ON(ret < 0);
+                       ret = 0;
                }
 
-alloc:
                if (unlikely(block_group->ro))
                        goto loop;
 
@@ -5606,11 +5681,6 @@ unclustered_alloc:
                }
 checks:
                search_start = stripe_align(root, offset);
-               /* move on to the next group */
-               if (search_start + num_bytes >= search_end) {
-                       btrfs_add_free_space(used_block_group, offset, num_bytes);
-                       goto loop;
-               }
 
                /* move on to the next group */
                if (search_start + num_bytes >
@@ -5661,9 +5731,7 @@ loop:
        if (!ins->objectid && ++index < BTRFS_NR_RAID_TYPES)
                goto search;
 
-       /* LOOP_FIND_IDEAL, only search caching/cached bg's, and don't wait for
-        *                      for them to make caching progress.  Also
-        *                      determine the best possible bg to cache
+       /*
         * LOOP_CACHING_NOWAIT, search partially cached block groups, kicking
         *                      caching kthreads as we move along
         * LOOP_CACHING_WAIT, search everything, and wait if our bg is caching
@@ -5673,50 +5741,17 @@ loop:
         */
        if (!ins->objectid && loop < LOOP_NO_EMPTY_SIZE) {
                index = 0;
-               if (loop == LOOP_FIND_IDEAL && found_uncached_bg) {
-                       found_uncached_bg = false;
-                       loop++;
-                       if (!ideal_cache_percent)
-                               goto search;
-
-                       /*
-                        * 1 of the following 2 things have happened so far
-                        *
-                        * 1) We found an ideal block group for caching that
-                        * is mostly full and will cache quickly, so we might
-                        * as well wait for it.
-                        *
-                        * 2) We searched for cached only and we didn't find
-                        * anything, and we didn't start any caching kthreads
-                        * either, so chances are we will loop through and
-                        * start a couple caching kthreads, and then come back
-                        * around and just wait for them.  This will be slower
-                        * because we will have 2 caching kthreads reading at
-                        * the same time when we could have just started one
-                        * and waited for it to get far enough to give us an
-                        * allocation, so go ahead and go to the wait caching
-                        * loop.
-                        */
-                       loop = LOOP_CACHING_WAIT;
-                       search_start = ideal_cache_offset;
-                       ideal_cache_percent = 0;
-                       goto ideal_cache;
-               } else if (loop == LOOP_FIND_IDEAL) {
-                       /*
-                        * Didn't find a uncached bg, wait on anything we find
-                        * next.
-                        */
-                       loop = LOOP_CACHING_WAIT;
-                       goto search;
-               }
-
                loop++;
-
                if (loop == LOOP_ALLOC_CHUNK) {
                       if (allowed_chunk_alloc) {
                                ret = do_chunk_alloc(trans, root, num_bytes +
                                                     2 * 1024 * 1024, data,
                                                     CHUNK_ALLOC_LIMITED);
+                               if (ret < 0) {
+                                       btrfs_abort_transaction(trans,
+                                                               root, ret);
+                                       goto out;
+                               }
                                allowed_chunk_alloc = 0;
                                if (ret == 1)
                                        done_chunk_alloc = 1;
@@ -5745,6 +5780,7 @@ loop:
        } else if (ins->objectid) {
                ret = 0;
        }
+out:
 
        return ret;
 }
@@ -5798,12 +5834,10 @@ int btrfs_reserve_extent(struct btrfs_trans_handle *trans,
                         struct btrfs_root *root,
                         u64 num_bytes, u64 min_alloc_size,
                         u64 empty_size, u64 hint_byte,
-                        u64 search_end, struct btrfs_key *ins,
-                        u64 data)
+                        struct btrfs_key *ins, u64 data)
 {
        bool final_tried = false;
        int ret;
-       u64 search_start = 0;
 
        data = btrfs_get_alloc_profile(root, data);
 again:
@@ -5811,23 +5845,31 @@ again:
         * the only place that sets empty_size is btrfs_realloc_node, which
         * is not called recursively on allocations
         */
-       if (empty_size || root->ref_cows)
+       if (empty_size || root->ref_cows) {
                ret = do_chunk_alloc(trans, root->fs_info->extent_root,
                                     num_bytes + 2 * 1024 * 1024, data,
                                     CHUNK_ALLOC_NO_FORCE);
+               if (ret < 0 && ret != -ENOSPC) {
+                       btrfs_abort_transaction(trans, root, ret);
+                       return ret;
+               }
+       }
 
        WARN_ON(num_bytes < root->sectorsize);
        ret = find_free_extent(trans, root, num_bytes, empty_size,
-                              search_start, search_end, hint_byte,
-                              ins, data);
+                              hint_byte, ins, data);
 
        if (ret == -ENOSPC) {
                if (!final_tried) {
                        num_bytes = num_bytes >> 1;
                        num_bytes = num_bytes & ~(root->sectorsize - 1);
                        num_bytes = max(num_bytes, min_alloc_size);
-                       do_chunk_alloc(trans, root->fs_info->extent_root,
+                       ret = do_chunk_alloc(trans, root->fs_info->extent_root,
                                       num_bytes, data, CHUNK_ALLOC_FORCE);
+                       if (ret < 0 && ret != -ENOSPC) {
+                               btrfs_abort_transaction(trans, root, ret);
+                               return ret;
+                       }
                        if (num_bytes == min_alloc_size)
                                final_tried = true;
                        goto again;
@@ -5838,7 +5880,8 @@ again:
                        printk(KERN_ERR "btrfs allocation failed flags %llu, "
                               "wanted %llu\n", (unsigned long long)data,
                               (unsigned long long)num_bytes);
-                       dump_space_info(sinfo, num_bytes, 1);
+                       if (sinfo)
+                               dump_space_info(sinfo, num_bytes, 1);
                }
        }
 
@@ -5917,7 +5960,10 @@ static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
        path->leave_spinning = 1;
        ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
                                      ins, size);
-       BUG_ON(ret);
+       if (ret) {
+               btrfs_free_path(path);
+               return ret;
+       }
 
        leaf = path->nodes[0];
        extent_item = btrfs_item_ptr(leaf, path->slots[0],
@@ -5947,7 +5993,7 @@ static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
        btrfs_free_path(path);
 
        ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
-       if (ret) {
+       if (ret) { /* -ENOENT, logic error */
                printk(KERN_ERR "btrfs update block group failed for %llu "
                       "%llu\n", (unsigned long long)ins->objectid,
                       (unsigned long long)ins->offset);
@@ -5978,7 +6024,10 @@ static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
        path->leave_spinning = 1;
        ret = btrfs_insert_empty_item(trans, fs_info->extent_root, path,
                                      ins, size);
-       BUG_ON(ret);
+       if (ret) {
+               btrfs_free_path(path);
+               return ret;
+       }
 
        leaf = path->nodes[0];
        extent_item = btrfs_item_ptr(leaf, path->slots[0],
@@ -6008,7 +6057,7 @@ static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
        btrfs_free_path(path);
 
        ret = update_block_group(trans, root, ins->objectid, ins->offset, 1);
-       if (ret) {
+       if (ret) { /* -ENOENT, logic error */
                printk(KERN_ERR "btrfs update block group failed for %llu "
                       "%llu\n", (unsigned long long)ins->objectid,
                       (unsigned long long)ins->offset);
@@ -6056,28 +6105,28 @@ int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
        if (!caching_ctl) {
                BUG_ON(!block_group_cache_done(block_group));
                ret = btrfs_remove_free_space(block_group, start, num_bytes);
-               BUG_ON(ret);
+               BUG_ON(ret); /* -ENOMEM */
        } else {
                mutex_lock(&caching_ctl->mutex);
 
                if (start >= caching_ctl->progress) {
                        ret = add_excluded_extent(root, start, num_bytes);
-                       BUG_ON(ret);
+                       BUG_ON(ret); /* -ENOMEM */
                } else if (start + num_bytes <= caching_ctl->progress) {
                        ret = btrfs_remove_free_space(block_group,
                                                      start, num_bytes);
-                       BUG_ON(ret);
+                       BUG_ON(ret); /* -ENOMEM */
                } else {
                        num_bytes = caching_ctl->progress - start;
                        ret = btrfs_remove_free_space(block_group,
                                                      start, num_bytes);
-                       BUG_ON(ret);
+                       BUG_ON(ret); /* -ENOMEM */
 
                        start = caching_ctl->progress;
                        num_bytes = ins->objectid + ins->offset -
                                    caching_ctl->progress;
                        ret = add_excluded_extent(root, start, num_bytes);
-                       BUG_ON(ret);
+                       BUG_ON(ret); /* -ENOMEM */
                }
 
                mutex_unlock(&caching_ctl->mutex);
@@ -6086,7 +6135,7 @@ int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
 
        ret = btrfs_update_reserved_bytes(block_group, ins->offset,
                                          RESERVE_ALLOC_NO_ACCOUNT);
-       BUG_ON(ret);
+       BUG_ON(ret); /* logic error */
        btrfs_put_block_group(block_group);
        ret = alloc_reserved_file_extent(trans, root, 0, root_objectid,
                                         0, owner, offset, ins, 1);
@@ -6107,6 +6156,7 @@ struct extent_buffer *btrfs_init_new_buffer(struct btrfs_trans_handle *trans,
        btrfs_set_buffer_lockdep_class(root->root_key.objectid, buf, level);
        btrfs_tree_lock(buf);
        clean_tree_block(trans, root, buf);
+       clear_bit(EXTENT_BUFFER_STALE, &buf->bflags);
 
        btrfs_set_lock_blocking(buf);
        btrfs_set_buffer_uptodate(buf);
@@ -6214,7 +6264,7 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
                return ERR_CAST(block_rsv);
 
        ret = btrfs_reserve_extent(trans, root, blocksize, blocksize,
-                                  empty_size, hint, (u64)-1, &ins, 0);
+                                  empty_size, hint, &ins, 0);
        if (ret) {
                unuse_block_rsv(root->fs_info, block_rsv, blocksize);
                return ERR_PTR(ret);
@@ -6222,7 +6272,7 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
 
        buf = btrfs_init_new_buffer(trans, root, ins.objectid,
                                    blocksize, level);
-       BUG_ON(IS_ERR(buf));
+       BUG_ON(IS_ERR(buf)); /* -ENOMEM */
 
        if (root_objectid == BTRFS_TREE_RELOC_OBJECTID) {
                if (parent == 0)
@@ -6234,7 +6284,7 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
        if (root_objectid != BTRFS_TREE_LOG_OBJECTID) {
                struct btrfs_delayed_extent_op *extent_op;
                extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS);
-               BUG_ON(!extent_op);
+               BUG_ON(!extent_op); /* -ENOMEM */
                if (key)
                        memcpy(&extent_op->key, key, sizeof(extent_op->key));
                else
@@ -6249,7 +6299,7 @@ struct extent_buffer *btrfs_alloc_free_block(struct btrfs_trans_handle *trans,
                                        ins.offset, parent, root_objectid,
                                        level, BTRFS_ADD_DELAYED_EXTENT,
                                        extent_op, for_cow);
-               BUG_ON(ret);
+               BUG_ON(ret); /* -ENOMEM */
        }
        return buf;
 }
@@ -6319,7 +6369,9 @@ static noinline void reada_walk_down(struct btrfs_trans_handle *trans,
                /* We don't lock the tree block, it's OK to be racy here */
                ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
                                               &refs, &flags);
-               BUG_ON(ret);
+               /* We don't care about errors in readahead. */
+               if (ret < 0)
+                       continue;
                BUG_ON(refs == 0);
 
                if (wc->stage == DROP_REFERENCE) {
@@ -6386,7 +6438,9 @@ static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
                                               eb->start, eb->len,
                                               &wc->refs[level],
                                               &wc->flags[level]);
-               BUG_ON(ret);
+               BUG_ON(ret == -ENOMEM);
+               if (ret)
+                       return ret;
                BUG_ON(wc->refs[level] == 0);
        }
 
@@ -6405,12 +6459,12 @@ static noinline int walk_down_proc(struct btrfs_trans_handle *trans,
        if (!(wc->flags[level] & flag)) {
                BUG_ON(!path->locks[level]);
                ret = btrfs_inc_ref(trans, root, eb, 1, wc->for_reloc);
-               BUG_ON(ret);
+               BUG_ON(ret); /* -ENOMEM */
                ret = btrfs_dec_ref(trans, root, eb, 0, wc->for_reloc);
-               BUG_ON(ret);
+               BUG_ON(ret); /* -ENOMEM */
                ret = btrfs_set_disk_extent_flags(trans, root, eb->start,
                                                  eb->len, flag, 0);
-               BUG_ON(ret);
+               BUG_ON(ret); /* -ENOMEM */
                wc->flags[level] |= flag;
        }
 
@@ -6482,7 +6536,11 @@ static noinline int do_walk_down(struct btrfs_trans_handle *trans,
        ret = btrfs_lookup_extent_info(trans, root, bytenr, blocksize,
                                       &wc->refs[level - 1],
                                       &wc->flags[level - 1]);
-       BUG_ON(ret);
+       if (ret < 0) {
+               btrfs_tree_unlock(next);
+               return ret;
+       }
+
        BUG_ON(wc->refs[level - 1] == 0);
        *lookup_info = 0;
 
@@ -6551,7 +6609,7 @@ skip:
 
                ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent,
                                root->root_key.objectid, level - 1, 0, 0);
-               BUG_ON(ret);
+               BUG_ON(ret); /* -ENOMEM */
        }
        btrfs_tree_unlock(next);
        free_extent_buffer(next);
@@ -6609,7 +6667,10 @@ static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
                                                       eb->start, eb->len,
                                                       &wc->refs[level],
                                                       &wc->flags[level]);
-                       BUG_ON(ret);
+                       if (ret < 0) {
+                               btrfs_tree_unlock_rw(eb, path->locks[level]);
+                               return ret;
+                       }
                        BUG_ON(wc->refs[level] == 0);
                        if (wc->refs[level] == 1) {
                                btrfs_tree_unlock_rw(eb, path->locks[level]);
@@ -6629,7 +6690,7 @@ static noinline int walk_up_proc(struct btrfs_trans_handle *trans,
                        else
                                ret = btrfs_dec_ref(trans, root, eb, 0,
                                                    wc->for_reloc);
-                       BUG_ON(ret);
+                       BUG_ON(ret); /* -ENOMEM */
                }
                /* make block locked assertion in clean_tree_block happy */
                if (!path->locks[level] &&
@@ -6738,7 +6799,7 @@ static noinline int walk_up_tree(struct btrfs_trans_handle *trans,
  * also make sure backrefs for the shared block and all lower level
  * blocks are properly updated.
  */
-void btrfs_drop_snapshot(struct btrfs_root *root,
+int btrfs_drop_snapshot(struct btrfs_root *root,
                         struct btrfs_block_rsv *block_rsv, int update_ref,
                         int for_reloc)
 {
@@ -6766,7 +6827,10 @@ void btrfs_drop_snapshot(struct btrfs_root *root,
        }
 
        trans = btrfs_start_transaction(tree_root, 0);
-       BUG_ON(IS_ERR(trans));
+       if (IS_ERR(trans)) {
+               err = PTR_ERR(trans);
+               goto out_free;
+       }
 
        if (block_rsv)
                trans->block_rsv = block_rsv;
@@ -6791,7 +6855,7 @@ void btrfs_drop_snapshot(struct btrfs_root *root,
                path->lowest_level = 0;
                if (ret < 0) {
                        err = ret;
-                       goto out_free;
+                       goto out_end_trans;
                }
                WARN_ON(ret > 0);
 
@@ -6811,7 +6875,10 @@ void btrfs_drop_snapshot(struct btrfs_root *root,
                                                path->nodes[level]->len,
                                                &wc->refs[level],
                                                &wc->flags[level]);
-                       BUG_ON(ret);
+                       if (ret < 0) {
+                               err = ret;
+                               goto out_end_trans;
+                       }
                        BUG_ON(wc->refs[level] == 0);
 
                        if (level == root_item->drop_level)
@@ -6862,26 +6929,40 @@ void btrfs_drop_snapshot(struct btrfs_root *root,
                        ret = btrfs_update_root(trans, tree_root,
                                                &root->root_key,
                                                root_item);
-                       BUG_ON(ret);
+                       if (ret) {
+                               btrfs_abort_transaction(trans, tree_root, ret);
+                               err = ret;
+                               goto out_end_trans;
+                       }
 
                        btrfs_end_transaction_throttle(trans, tree_root);
                        trans = btrfs_start_transaction(tree_root, 0);
-                       BUG_ON(IS_ERR(trans));
+                       if (IS_ERR(trans)) {
+                               err = PTR_ERR(trans);
+                               goto out_free;
+                       }
                        if (block_rsv)
                                trans->block_rsv = block_rsv;
                }
        }
        btrfs_release_path(path);
-       BUG_ON(err);
+       if (err)
+               goto out_end_trans;
 
        ret = btrfs_del_root(trans, tree_root, &root->root_key);
-       BUG_ON(ret);
+       if (ret) {
+               btrfs_abort_transaction(trans, tree_root, ret);
+               goto out_end_trans;
+       }
 
        if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) {
                ret = btrfs_find_last_root(tree_root, root->root_key.objectid,
                                           NULL, NULL);
-               BUG_ON(ret < 0);
-               if (ret > 0) {
+               if (ret < 0) {
+                       btrfs_abort_transaction(trans, tree_root, ret);
+                       err = ret;
+                       goto out_end_trans;
+               } else if (ret > 0) {
                        /* if we fail to delete the orphan item this time
                         * around, it'll get picked up the next time.
                         *
@@ -6899,14 +6980,15 @@ void btrfs_drop_snapshot(struct btrfs_root *root,
                free_extent_buffer(root->commit_root);
                kfree(root);
        }
-out_free:
+out_end_trans:
        btrfs_end_transaction_throttle(trans, tree_root);
+out_free:
        kfree(wc);
        btrfs_free_path(path);
 out:
        if (err)
                btrfs_std_error(root->fs_info, err);
-       return;
+       return err;
 }
 
 /*
@@ -6983,31 +7065,15 @@ int btrfs_drop_subtree(struct btrfs_trans_handle *trans,
 static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
 {
        u64 num_devices;
-       u64 stripped = BTRFS_BLOCK_GROUP_RAID0 |
-               BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
+       u64 stripped;
 
-       if (root->fs_info->balance_ctl) {
-               struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
-               u64 tgt = 0;
-
-               /* pick restriper's target profile and return */
-               if (flags & BTRFS_BLOCK_GROUP_DATA &&
-                   bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) {
-                       tgt = BTRFS_BLOCK_GROUP_DATA | bctl->data.target;
-               } else if (flags & BTRFS_BLOCK_GROUP_SYSTEM &&
-                          bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
-                       tgt = BTRFS_BLOCK_GROUP_SYSTEM | bctl->sys.target;
-               } else if (flags & BTRFS_BLOCK_GROUP_METADATA &&
-                          bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) {
-                       tgt = BTRFS_BLOCK_GROUP_METADATA | bctl->meta.target;
-               }
-
-               if (tgt) {
-                       /* extended -> chunk profile */
-                       tgt &= ~BTRFS_AVAIL_ALLOC_BIT_SINGLE;
-                       return tgt;
-               }
-       }
+       /*
+        * if restripe for this chunk_type is on pick target profile and
+        * return, otherwise do the usual balance
+        */
+       stripped = get_restripe_target(root->fs_info, flags);
+       if (stripped)
+               return extended_to_chunk(stripped);
 
        /*
         * we add in the count of missing devices because we want
@@ -7017,6 +7083,9 @@ static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
        num_devices = root->fs_info->fs_devices->rw_devices +
                root->fs_info->fs_devices->missing_devices;
 
+       stripped = BTRFS_BLOCK_GROUP_RAID0 |
+               BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10;
+
        if (num_devices == 1) {
                stripped |= BTRFS_BLOCK_GROUP_DUP;
                stripped = flags & ~stripped;
@@ -7029,7 +7098,6 @@ static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
                if (flags & (BTRFS_BLOCK_GROUP_RAID1 |
                             BTRFS_BLOCK_GROUP_RAID10))
                        return stripped | BTRFS_BLOCK_GROUP_DUP;
-               return flags;
        } else {
                /* they already had raid on here, just return */
                if (flags & stripped)
@@ -7042,9 +7110,9 @@ static u64 update_block_group_flags(struct btrfs_root *root, u64 flags)
                if (flags & BTRFS_BLOCK_GROUP_DUP)
                        return stripped | BTRFS_BLOCK_GROUP_RAID1;
 
-               /* turn single device chunks into raid0 */
-               return stripped | BTRFS_BLOCK_GROUP_RAID0;
+               /* this is drive concat, leave it alone */
        }
+
        return flags;
 }
 
@@ -7103,12 +7171,16 @@ int btrfs_set_block_group_ro(struct btrfs_root *root,
        BUG_ON(cache->ro);
 
        trans = btrfs_join_transaction(root);
-       BUG_ON(IS_ERR(trans));
+       if (IS_ERR(trans))
+               return PTR_ERR(trans);
 
        alloc_flags = update_block_group_flags(root, cache->flags);
-       if (alloc_flags != cache->flags)
-               do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
-                              CHUNK_ALLOC_FORCE);
+       if (alloc_flags != cache->flags) {
+               ret = do_chunk_alloc(trans, root, 2 * 1024 * 1024, alloc_flags,
+                                    CHUNK_ALLOC_FORCE);
+               if (ret < 0)
+                       goto out;
+       }
 
        ret = set_block_group_ro(cache, 0);
        if (!ret)
@@ -7188,7 +7260,7 @@ u64 btrfs_account_ro_block_groups_free_space(struct btrfs_space_info *sinfo)
        return free_bytes;
 }
 
-int btrfs_set_block_group_rw(struct btrfs_root *root,
+void btrfs_set_block_group_rw(struct btrfs_root *root,
                              struct btrfs_block_group_cache *cache)
 {
        struct btrfs_space_info *sinfo = cache->space_info;
@@ -7204,7 +7276,6 @@ int btrfs_set_block_group_rw(struct btrfs_root *root,
        cache->ro = 0;
        spin_unlock(&cache->lock);
        spin_unlock(&sinfo->lock);
-       return 0;
 }
 
 /*
@@ -7222,6 +7293,7 @@ int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
        u64 min_free;
        u64 dev_min = 1;
        u64 dev_nr = 0;
+       u64 target;
        int index;
        int full = 0;
        int ret = 0;
@@ -7262,13 +7334,11 @@ int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
        /*
         * ok we don't have enough space, but maybe we have free space on our
         * devices to allocate new chunks for relocation, so loop through our
-        * alloc devices and guess if we have enough space.  However, if we
-        * were marked as full, then we know there aren't enough chunks, and we
-        * can just return.
+        * alloc devices and guess if we have enough space.  if this block
+        * group is going to be restriped, run checks against the target
+        * profile instead of the current one.
         */
        ret = -1;
-       if (full)
-               goto out;
 
        /*
         * index:
@@ -7278,7 +7348,20 @@ int btrfs_can_relocate(struct btrfs_root *root, u64 bytenr)
         *      3: raid0
         *      4: single
         */
-       index = get_block_group_index(block_group);
+       target = get_restripe_target(root->fs_info, block_group->flags);
+       if (target) {
+               index = __get_block_group_index(extended_to_chunk(target));
+       } else {
+               /*
+                * this is just a balance, so if we were marked as full
+                * we know there is no space for a new chunk
+                */
+               if (full)
+                       goto out;
+
+               index = get_block_group_index(block_group);
+       }
+
        if (index == 0) {
                dev_min = 4;
                /* Divide by 2 */
@@ -7572,7 +7655,7 @@ int btrfs_read_block_groups(struct btrfs_root *root)
                ret = update_space_info(info, cache->flags, found_key.offset,
                                        btrfs_block_group_used(&cache->item),
                                        &space_info);
-               BUG_ON(ret);
+               BUG_ON(ret); /* -ENOMEM */
                cache->space_info = space_info;
                spin_lock(&cache->space_info->lock);
                cache->space_info->bytes_readonly += cache->bytes_super;
@@ -7581,7 +7664,7 @@ int btrfs_read_block_groups(struct btrfs_root *root)
                __link_block_group(space_info, cache);
 
                ret = btrfs_add_block_group_cache(root->fs_info, cache);
-               BUG_ON(ret);
+               BUG_ON(ret); /* Logic error */
 
                set_avail_alloc_bits(root->fs_info, cache->flags);
                if (btrfs_chunk_readonly(root, cache->key.objectid))
@@ -7663,7 +7746,7 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
 
        ret = update_space_info(root->fs_info, cache->flags, size, bytes_used,
                                &cache->space_info);
-       BUG_ON(ret);
+       BUG_ON(ret); /* -ENOMEM */
        update_global_block_rsv(root->fs_info);
 
        spin_lock(&cache->space_info->lock);
@@ -7673,11 +7756,14 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
        __link_block_group(cache->space_info, cache);
 
        ret = btrfs_add_block_group_cache(root->fs_info, cache);
-       BUG_ON(ret);
+       BUG_ON(ret); /* Logic error */
 
        ret = btrfs_insert_item(trans, extent_root, &cache->key, &cache->item,
                                sizeof(cache->item));
-       BUG_ON(ret);
+       if (ret) {
+               btrfs_abort_transaction(trans, extent_root, ret);
+               return ret;
+       }
 
        set_avail_alloc_bits(extent_root->fs_info, type);
 
@@ -7686,11 +7772,8 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans,
 
 static void clear_avail_alloc_bits(struct btrfs_fs_info *fs_info, u64 flags)
 {
-       u64 extra_flags = flags & BTRFS_BLOCK_GROUP_PROFILE_MASK;
-
-       /* chunk -> extended profile */
-       if (extra_flags == 0)
-               extra_flags = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
+       u64 extra_flags = chunk_to_extended(flags) &
+                               BTRFS_EXTENDED_PROFILE_MASK;
 
        if (flags & BTRFS_BLOCK_GROUP_DATA)
                fs_info->avail_data_alloc_bits &= ~extra_flags;
@@ -7758,7 +7841,10 @@ int btrfs_remove_block_group(struct btrfs_trans_handle *trans,
        inode = lookup_free_space_inode(tree_root, block_group, path);
        if (!IS_ERR(inode)) {
                ret = btrfs_orphan_add(trans, inode);
-               BUG_ON(ret);
+               if (ret) {
+                       btrfs_add_delayed_iput(inode);
+                       goto out;
+               }
                clear_nlink(inode);
                /* One for the block groups ref */
                spin_lock(&block_group->lock);