X-Git-Url: http://pilppa.org/gitweb/gitweb.cgi?a=blobdiff_plain;f=fs%2Fbtrfs%2Fextent-tree.c;h=8933d15a240fa0bdc079598d99e7ee739e12728d;hb=b9473439d3e84d9fc1a0a83faca69cc1b7566341;hp=0a5d796c9f7ee3086bd3c8bccdfc63b59d674e0f;hpb=b5482d475c6eff1ebc0b1cee73421ef26f1d046c;p=linux-2.6-omap-h63xx.git diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index 0a5d796c9f7..8933d15a240 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -20,6 +20,7 @@ #include #include #include +#include #include "compat.h" #include "hash.h" #include "crc32c.h" @@ -48,17 +49,27 @@ struct pending_extent_op { int del; }; -static int finish_current_insert(struct btrfs_trans_handle *trans, - struct btrfs_root *extent_root, int all); -static int del_pending_extents(struct btrfs_trans_handle *trans, - struct btrfs_root *extent_root, int all); -static int pin_down_bytes(struct btrfs_trans_handle *trans, - struct btrfs_root *root, - u64 bytenr, u64 num_bytes, int is_data); +static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, + struct btrfs_root *root, u64 parent, + u64 root_objectid, u64 ref_generation, + u64 owner, struct btrfs_key *ins, + int ref_mod); +static int update_reserved_extents(struct btrfs_root *root, + u64 bytenr, u64 num, int reserve); static int update_block_group(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 bytenr, u64 num_bytes, int alloc, int mark_free); +static noinline int __btrfs_free_extent(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 bytenr, u64 num_bytes, u64 parent, + u64 root_objectid, u64 ref_generation, + u64 owner_objectid, int pin, + int ref_to_drop); + +static int do_chunk_alloc(struct btrfs_trans_handle *trans, + struct btrfs_root *extent_root, u64 alloc_bytes, + u64 flags, int force); static int block_group_bits(struct btrfs_block_group_cache *cache, u64 bits) { @@ -326,13 +337,33 @@ static struct btrfs_space_info *__find_space_info(struct btrfs_fs_info *info, { struct list_head *head = &info->space_info; struct btrfs_space_info *found; - list_for_each_entry(found, head, list) { - if (found->flags == flags) + + rcu_read_lock(); + list_for_each_entry_rcu(found, head, list) { + if (found->flags == flags) { + rcu_read_unlock(); return found; + } } + rcu_read_unlock(); return NULL; } +/* + * after adding space to the filesystem, we need to clear the full flags + * on all the space infos. + */ +void btrfs_clear_space_info_full(struct btrfs_fs_info *info) +{ + struct list_head *head = &info->space_info; + struct btrfs_space_info *found; + + rcu_read_lock(); + list_for_each_entry_rcu(found, head, list) + found->full = 0; + rcu_read_unlock(); +} + static u64 div_factor(u64 num, int factor) { if (factor == 10) @@ -529,262 +560,13 @@ out: return ret; } -/* - * updates all the backrefs that are pending on update_list for the - * extent_root - */ -static noinline int update_backrefs(struct btrfs_trans_handle *trans, - struct btrfs_root *extent_root, - struct btrfs_path *path, - struct list_head *update_list) -{ - struct btrfs_key key; - struct btrfs_extent_ref *ref; - struct btrfs_fs_info *info = extent_root->fs_info; - struct pending_extent_op *op; - struct extent_buffer *leaf; - int ret = 0; - struct list_head *cur = update_list->next; - u64 ref_objectid; - u64 ref_root = extent_root->root_key.objectid; - - op = list_entry(cur, struct pending_extent_op, list); - -search: - key.objectid = op->bytenr; - key.type = BTRFS_EXTENT_REF_KEY; - key.offset = op->orig_parent; - - ret = btrfs_search_slot(trans, extent_root, &key, path, 0, 1); - BUG_ON(ret); - - leaf = path->nodes[0]; - -loop: - ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); - - ref_objectid = btrfs_ref_objectid(leaf, ref); - - if (btrfs_ref_root(leaf, ref) != ref_root || - btrfs_ref_generation(leaf, ref) != op->orig_generation || - (ref_objectid != op->level && - ref_objectid != BTRFS_MULTIPLE_OBJECTIDS)) { - printk(KERN_ERR "btrfs couldn't find %llu, parent %llu, " - "root %llu, owner %u\n", - (unsigned long long)op->bytenr, - (unsigned long long)op->orig_parent, - (unsigned long long)ref_root, op->level); - btrfs_print_leaf(extent_root, leaf); - BUG(); - } - - key.objectid = op->bytenr; - key.offset = op->parent; - key.type = BTRFS_EXTENT_REF_KEY; - ret = btrfs_set_item_key_safe(trans, extent_root, path, &key); - BUG_ON(ret); - ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); - btrfs_set_ref_generation(leaf, ref, op->generation); - - cur = cur->next; - - list_del_init(&op->list); - unlock_extent(&info->extent_ins, op->bytenr, - op->bytenr + op->num_bytes - 1, GFP_NOFS); - kfree(op); - - if (cur == update_list) { - btrfs_mark_buffer_dirty(path->nodes[0]); - btrfs_release_path(extent_root, path); - goto out; - } - - op = list_entry(cur, struct pending_extent_op, list); - - path->slots[0]++; - while (path->slots[0] < btrfs_header_nritems(leaf)) { - btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); - if (key.objectid == op->bytenr && - key.type == BTRFS_EXTENT_REF_KEY) - goto loop; - path->slots[0]++; - } - - btrfs_mark_buffer_dirty(path->nodes[0]); - btrfs_release_path(extent_root, path); - goto search; - -out: - return 0; -} - -static noinline int insert_extents(struct btrfs_trans_handle *trans, - struct btrfs_root *extent_root, - struct btrfs_path *path, - struct list_head *insert_list, int nr) -{ - struct btrfs_key *keys; - u32 *data_size; - struct pending_extent_op *op; - struct extent_buffer *leaf; - struct list_head *cur = insert_list->next; - struct btrfs_fs_info *info = extent_root->fs_info; - u64 ref_root = extent_root->root_key.objectid; - int i = 0, last = 0, ret; - int total = nr * 2; - - if (!nr) - return 0; - - keys = kzalloc(total * sizeof(struct btrfs_key), GFP_NOFS); - if (!keys) - return -ENOMEM; - - data_size = kzalloc(total * sizeof(u32), GFP_NOFS); - if (!data_size) { - kfree(keys); - return -ENOMEM; - } - - list_for_each_entry(op, insert_list, list) { - keys[i].objectid = op->bytenr; - keys[i].offset = op->num_bytes; - keys[i].type = BTRFS_EXTENT_ITEM_KEY; - data_size[i] = sizeof(struct btrfs_extent_item); - i++; - - keys[i].objectid = op->bytenr; - keys[i].offset = op->parent; - keys[i].type = BTRFS_EXTENT_REF_KEY; - data_size[i] = sizeof(struct btrfs_extent_ref); - i++; - } - - op = list_entry(cur, struct pending_extent_op, list); - i = 0; - while (i < total) { - int c; - ret = btrfs_insert_some_items(trans, extent_root, path, - keys+i, data_size+i, total-i); - BUG_ON(ret < 0); - - if (last && ret > 1) - BUG(); - - leaf = path->nodes[0]; - for (c = 0; c < ret; c++) { - int ref_first = keys[i].type == BTRFS_EXTENT_REF_KEY; - - /* - * if the first item we inserted was a backref, then - * the EXTENT_ITEM will be the odd c's, else it will - * be the even c's - */ - if ((ref_first && (c % 2)) || - (!ref_first && !(c % 2))) { - struct btrfs_extent_item *itm; - - itm = btrfs_item_ptr(leaf, path->slots[0] + c, - struct btrfs_extent_item); - btrfs_set_extent_refs(path->nodes[0], itm, 1); - op->del++; - } else { - struct btrfs_extent_ref *ref; - - ref = btrfs_item_ptr(leaf, path->slots[0] + c, - struct btrfs_extent_ref); - btrfs_set_ref_root(leaf, ref, ref_root); - btrfs_set_ref_generation(leaf, ref, - op->generation); - btrfs_set_ref_objectid(leaf, ref, op->level); - btrfs_set_ref_num_refs(leaf, ref, 1); - op->del++; - } - - /* - * using del to see when its ok to free up the - * pending_extent_op. In the case where we insert the - * last item on the list in order to help do batching - * we need to not free the extent op until we actually - * insert the extent_item - */ - if (op->del == 2) { - unlock_extent(&info->extent_ins, op->bytenr, - op->bytenr + op->num_bytes - 1, - GFP_NOFS); - cur = cur->next; - list_del_init(&op->list); - kfree(op); - if (cur != insert_list) - op = list_entry(cur, - struct pending_extent_op, - list); - } - } - btrfs_mark_buffer_dirty(leaf); - btrfs_release_path(extent_root, path); - - /* - * Ok backref's and items usually go right next to eachother, - * but if we could only insert 1 item that means that we - * inserted on the end of a leaf, and we have no idea what may - * be on the next leaf so we just play it safe. In order to - * try and help this case we insert the last thing on our - * insert list so hopefully it will end up being the last - * thing on the leaf and everything else will be before it, - * which will let us insert a whole bunch of items at the same - * time. - */ - if (ret == 1 && !last && (i + ret < total)) { - /* - * last: where we will pick up the next time around - * i: our current key to insert, will be total - 1 - * cur: the current op we are screwing with - * op: duh - */ - last = i + ret; - i = total - 1; - cur = insert_list->prev; - op = list_entry(cur, struct pending_extent_op, list); - } else if (last) { - /* - * ok we successfully inserted the last item on the - * list, lets reset everything - * - * i: our current key to insert, so where we left off - * last time - * last: done with this - * cur: the op we are messing with - * op: duh - * total: since we inserted the last key, we need to - * decrement total so we dont overflow - */ - i = last; - last = 0; - total--; - if (i < total) { - cur = insert_list->next; - op = list_entry(cur, struct pending_extent_op, - list); - } - } else { - i += ret; - } - - cond_resched(); - } - ret = 0; - kfree(keys); - kfree(data_size); - return ret; -} - static noinline int insert_extent_backref(struct btrfs_trans_handle *trans, struct btrfs_root *root, struct btrfs_path *path, u64 bytenr, u64 parent, u64 ref_root, u64 ref_generation, - u64 owner_objectid) + u64 owner_objectid, + int refs_to_add) { struct btrfs_key key; struct extent_buffer *leaf; @@ -804,9 +586,10 @@ static noinline int insert_extent_backref(struct btrfs_trans_handle *trans, btrfs_set_ref_root(leaf, ref, ref_root); btrfs_set_ref_generation(leaf, ref, ref_generation); btrfs_set_ref_objectid(leaf, ref, owner_objectid); - btrfs_set_ref_num_refs(leaf, ref, 1); + btrfs_set_ref_num_refs(leaf, ref, refs_to_add); } else if (ret == -EEXIST) { u64 existing_owner; + BUG_ON(owner_objectid < BTRFS_FIRST_FREE_OBJECTID); leaf = path->nodes[0]; ref = btrfs_item_ptr(leaf, path->slots[0], @@ -820,7 +603,7 @@ static noinline int insert_extent_backref(struct btrfs_trans_handle *trans, num_refs = btrfs_ref_num_refs(leaf, ref); BUG_ON(num_refs == 0); - btrfs_set_ref_num_refs(leaf, ref, num_refs + 1); + btrfs_set_ref_num_refs(leaf, ref, num_refs + refs_to_add); existing_owner = btrfs_ref_objectid(leaf, ref); if (existing_owner != owner_objectid && @@ -832,6 +615,7 @@ static noinline int insert_extent_backref(struct btrfs_trans_handle *trans, } else { goto out; } + btrfs_unlock_up_safe(path, 1); btrfs_mark_buffer_dirty(path->nodes[0]); out: btrfs_release_path(root, path); @@ -840,7 +624,8 @@ out: static noinline int remove_extent_backref(struct btrfs_trans_handle *trans, struct btrfs_root *root, - struct btrfs_path *path) + struct btrfs_path *path, + int refs_to_drop) { struct extent_buffer *leaf; struct btrfs_extent_ref *ref; @@ -850,8 +635,8 @@ static noinline int remove_extent_backref(struct btrfs_trans_handle *trans, leaf = path->nodes[0]; ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); num_refs = btrfs_ref_num_refs(leaf, ref); - BUG_ON(num_refs == 0); - num_refs -= 1; + BUG_ON(num_refs < refs_to_drop); + num_refs -= refs_to_drop; if (num_refs == 0) { ret = btrfs_del_item(trans, root, path); } else { @@ -902,332 +687,28 @@ static int btrfs_discard_extent(struct btrfs_root *root, u64 bytenr, #endif } -static noinline int free_extents(struct btrfs_trans_handle *trans, - struct btrfs_root *extent_root, - struct list_head *del_list) -{ - struct btrfs_fs_info *info = extent_root->fs_info; - struct btrfs_path *path; - struct btrfs_key key, found_key; - struct extent_buffer *leaf; - struct list_head *cur; - struct pending_extent_op *op; - struct btrfs_extent_item *ei; - int ret, num_to_del, extent_slot = 0, found_extent = 0; - u32 refs; - u64 bytes_freed = 0; - - path = btrfs_alloc_path(); - if (!path) - return -ENOMEM; - path->reada = 1; - -search: - /* search for the backref for the current ref we want to delete */ - cur = del_list->next; - op = list_entry(cur, struct pending_extent_op, list); - ret = lookup_extent_backref(trans, extent_root, path, op->bytenr, - op->orig_parent, - extent_root->root_key.objectid, - op->orig_generation, op->level, 1); - if (ret) { - printk(KERN_ERR "btrfs unable to find backref byte nr %llu " - "root %llu gen %llu owner %u\n", - (unsigned long long)op->bytenr, - (unsigned long long)extent_root->root_key.objectid, - (unsigned long long)op->orig_generation, op->level); - btrfs_print_leaf(extent_root, path->nodes[0]); - WARN_ON(1); - goto out; - } - - extent_slot = path->slots[0]; - num_to_del = 1; - found_extent = 0; - - /* - * if we aren't the first item on the leaf we can move back one and see - * if our ref is right next to our extent item - */ - if (likely(extent_slot)) { - extent_slot--; - btrfs_item_key_to_cpu(path->nodes[0], &found_key, - extent_slot); - if (found_key.objectid == op->bytenr && - found_key.type == BTRFS_EXTENT_ITEM_KEY && - found_key.offset == op->num_bytes) { - num_to_del++; - found_extent = 1; - } - } - - /* - * if we didn't find the extent we need to delete the backref and then - * search for the extent item key so we can update its ref count - */ - if (!found_extent) { - key.objectid = op->bytenr; - key.type = BTRFS_EXTENT_ITEM_KEY; - key.offset = op->num_bytes; - - ret = remove_extent_backref(trans, extent_root, path); - BUG_ON(ret); - btrfs_release_path(extent_root, path); - ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); - BUG_ON(ret); - extent_slot = path->slots[0]; - } - - /* this is where we update the ref count for the extent */ - leaf = path->nodes[0]; - ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item); - refs = btrfs_extent_refs(leaf, ei); - BUG_ON(refs == 0); - refs--; - btrfs_set_extent_refs(leaf, ei, refs); - - btrfs_mark_buffer_dirty(leaf); - - /* - * This extent needs deleting. The reason cur_slot is extent_slot + - * num_to_del is because extent_slot points to the slot where the extent - * is, and if the backref was not right next to the extent we will be - * deleting at least 1 item, and will want to start searching at the - * slot directly next to extent_slot. However if we did find the - * backref next to the extent item them we will be deleting at least 2 - * items and will want to start searching directly after the ref slot - */ - if (!refs) { - struct list_head *pos, *n, *end; - int cur_slot = extent_slot+num_to_del; - u64 super_used; - u64 root_used; - - path->slots[0] = extent_slot; - bytes_freed = op->num_bytes; - - mutex_lock(&info->pinned_mutex); - ret = pin_down_bytes(trans, extent_root, op->bytenr, - op->num_bytes, op->level >= - BTRFS_FIRST_FREE_OBJECTID); - mutex_unlock(&info->pinned_mutex); - BUG_ON(ret < 0); - op->del = ret; - - /* - * we need to see if we can delete multiple things at once, so - * start looping through the list of extents we are wanting to - * delete and see if their extent/backref's are right next to - * eachother and the extents only have 1 ref - */ - for (pos = cur->next; pos != del_list; pos = pos->next) { - struct pending_extent_op *tmp; - - tmp = list_entry(pos, struct pending_extent_op, list); - - /* we only want to delete extent+ref at this stage */ - if (cur_slot >= btrfs_header_nritems(leaf) - 1) - break; - - btrfs_item_key_to_cpu(leaf, &found_key, cur_slot); - if (found_key.objectid != tmp->bytenr || - found_key.type != BTRFS_EXTENT_ITEM_KEY || - found_key.offset != tmp->num_bytes) - break; - - /* check to make sure this extent only has one ref */ - ei = btrfs_item_ptr(leaf, cur_slot, - struct btrfs_extent_item); - if (btrfs_extent_refs(leaf, ei) != 1) - break; - - btrfs_item_key_to_cpu(leaf, &found_key, cur_slot+1); - if (found_key.objectid != tmp->bytenr || - found_key.type != BTRFS_EXTENT_REF_KEY || - found_key.offset != tmp->orig_parent) - break; - - /* - * the ref is right next to the extent, we can set the - * ref count to 0 since we will delete them both now - */ - btrfs_set_extent_refs(leaf, ei, 0); - - /* pin down the bytes for this extent */ - mutex_lock(&info->pinned_mutex); - ret = pin_down_bytes(trans, extent_root, tmp->bytenr, - tmp->num_bytes, tmp->level >= - BTRFS_FIRST_FREE_OBJECTID); - mutex_unlock(&info->pinned_mutex); - BUG_ON(ret < 0); - - /* - * use the del field to tell if we need to go ahead and - * free up the extent when we delete the item or not. - */ - tmp->del = ret; - bytes_freed += tmp->num_bytes; - - num_to_del += 2; - cur_slot += 2; - } - end = pos; - - /* update the free space counters */ - spin_lock(&info->delalloc_lock); - super_used = btrfs_super_bytes_used(&info->super_copy); - btrfs_set_super_bytes_used(&info->super_copy, - super_used - bytes_freed); - - root_used = btrfs_root_used(&extent_root->root_item); - btrfs_set_root_used(&extent_root->root_item, - root_used - bytes_freed); - spin_unlock(&info->delalloc_lock); - - /* delete the items */ - ret = btrfs_del_items(trans, extent_root, path, - path->slots[0], num_to_del); - BUG_ON(ret); - - /* - * loop through the extents we deleted and do the cleanup work - * on them - */ - for (pos = cur, n = pos->next; pos != end; - pos = n, n = pos->next) { - struct pending_extent_op *tmp; - tmp = list_entry(pos, struct pending_extent_op, list); - - /* - * remember tmp->del tells us wether or not we pinned - * down the extent - */ - ret = update_block_group(trans, extent_root, - tmp->bytenr, tmp->num_bytes, 0, - tmp->del); - BUG_ON(ret); - - list_del_init(&tmp->list); - unlock_extent(&info->extent_ins, tmp->bytenr, - tmp->bytenr + tmp->num_bytes - 1, - GFP_NOFS); - kfree(tmp); - } - } else if (refs && found_extent) { - /* - * the ref and extent were right next to eachother, but the - * extent still has a ref, so just free the backref and keep - * going - */ - ret = remove_extent_backref(trans, extent_root, path); - BUG_ON(ret); - - list_del_init(&op->list); - unlock_extent(&info->extent_ins, op->bytenr, - op->bytenr + op->num_bytes - 1, GFP_NOFS); - kfree(op); - } else { - /* - * the extent has multiple refs and the backref we were looking - * for was not right next to it, so just unlock and go next, - * we're good to go - */ - list_del_init(&op->list); - unlock_extent(&info->extent_ins, op->bytenr, - op->bytenr + op->num_bytes - 1, GFP_NOFS); - kfree(op); - } - - btrfs_release_path(extent_root, path); - if (!list_empty(del_list)) - goto search; - -out: - btrfs_free_path(path); - return ret; -} - static int __btrfs_update_extent_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 bytenr, + u64 num_bytes, u64 orig_parent, u64 parent, u64 orig_root, u64 ref_root, u64 orig_generation, u64 ref_generation, u64 owner_objectid) { int ret; - struct btrfs_root *extent_root = root->fs_info->extent_root; - struct btrfs_path *path; - - if (root == root->fs_info->extent_root) { - struct pending_extent_op *extent_op; - u64 num_bytes; - - BUG_ON(owner_objectid >= BTRFS_MAX_LEVEL); - num_bytes = btrfs_level_size(root, (int)owner_objectid); - mutex_lock(&root->fs_info->extent_ins_mutex); - if (test_range_bit(&root->fs_info->extent_ins, bytenr, - bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) { - u64 priv; - ret = get_state_private(&root->fs_info->extent_ins, - bytenr, &priv); - BUG_ON(ret); - extent_op = (struct pending_extent_op *) - (unsigned long)priv; - BUG_ON(extent_op->parent != orig_parent); - BUG_ON(extent_op->generation != orig_generation); + int pin = owner_objectid < BTRFS_FIRST_FREE_OBJECTID; - extent_op->parent = parent; - extent_op->generation = ref_generation; - } else { - extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); - BUG_ON(!extent_op); - - extent_op->type = PENDING_BACKREF_UPDATE; - extent_op->bytenr = bytenr; - extent_op->num_bytes = num_bytes; - extent_op->parent = parent; - extent_op->orig_parent = orig_parent; - extent_op->generation = ref_generation; - extent_op->orig_generation = orig_generation; - extent_op->level = (int)owner_objectid; - INIT_LIST_HEAD(&extent_op->list); - extent_op->del = 0; - - set_extent_bits(&root->fs_info->extent_ins, - bytenr, bytenr + num_bytes - 1, - EXTENT_WRITEBACK, GFP_NOFS); - set_state_private(&root->fs_info->extent_ins, - bytenr, (unsigned long)extent_op); - } - mutex_unlock(&root->fs_info->extent_ins_mutex); - return 0; - } - - path = btrfs_alloc_path(); - if (!path) - return -ENOMEM; - ret = lookup_extent_backref(trans, extent_root, path, - bytenr, orig_parent, orig_root, - orig_generation, owner_objectid, 1); - if (ret) - goto out; - ret = remove_extent_backref(trans, extent_root, path); - if (ret) - goto out; - ret = insert_extent_backref(trans, extent_root, path, bytenr, - parent, ref_root, ref_generation, - owner_objectid); + ret = btrfs_update_delayed_ref(trans, bytenr, num_bytes, + orig_parent, parent, orig_root, + ref_root, orig_generation, + ref_generation, owner_objectid, pin); BUG_ON(ret); - finish_current_insert(trans, extent_root, 0); - del_pending_extents(trans, extent_root, 0); -out: - btrfs_free_path(path); return ret; } int btrfs_update_extent_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 bytenr, - u64 orig_parent, u64 parent, + u64 num_bytes, u64 orig_parent, u64 parent, u64 ref_root, u64 ref_generation, u64 owner_objectid) { @@ -1235,19 +716,35 @@ int btrfs_update_extent_ref(struct btrfs_trans_handle *trans, if (ref_root == BTRFS_TREE_LOG_OBJECTID && owner_objectid < BTRFS_FIRST_FREE_OBJECTID) return 0; - ret = __btrfs_update_extent_ref(trans, root, bytenr, orig_parent, - parent, ref_root, ref_root, - ref_generation, ref_generation, - owner_objectid); + + ret = __btrfs_update_extent_ref(trans, root, bytenr, num_bytes, + orig_parent, parent, ref_root, + ref_root, ref_generation, + ref_generation, owner_objectid); return ret; } - static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 bytenr, + u64 num_bytes, u64 orig_parent, u64 parent, u64 orig_root, u64 ref_root, u64 orig_generation, u64 ref_generation, u64 owner_objectid) +{ + int ret; + + ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent, ref_root, + ref_generation, owner_objectid, + BTRFS_ADD_DELAYED_REF, 0); + BUG_ON(ret); + return ret; +} + +static noinline_for_stack int add_extent_ref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, u64 bytenr, + u64 num_bytes, u64 parent, u64 ref_root, + u64 ref_generation, u64 owner_objectid, + int refs_to_add) { struct btrfs_path *path; int ret; @@ -1261,17 +758,24 @@ static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, return -ENOMEM; path->reada = 1; + path->leave_spinning = 1; key.objectid = bytenr; key.type = BTRFS_EXTENT_ITEM_KEY; - key.offset = (u64)-1; + key.offset = num_bytes; - ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, - 0, 1); - if (ret < 0) + /* first find the extent item and update its reference count */ + ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, + path, 0, 1); + if (ret < 0) { + btrfs_set_path_blocking(path); return ret; - BUG_ON(ret == 0 || path->slots[0] == 0); + } - path->slots[0]--; + if (ret > 0) { + WARN_ON(1); + btrfs_free_path(path); + return -EIO; + } l = path->nodes[0]; btrfs_item_key_to_cpu(l, &key, path->slots[0]); @@ -1285,21 +789,24 @@ static int __btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, BUG_ON(key.type != BTRFS_EXTENT_ITEM_KEY); item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); + refs = btrfs_extent_refs(l, item); - btrfs_set_extent_refs(l, item, refs + 1); + btrfs_set_extent_refs(l, item, refs + refs_to_add); + btrfs_unlock_up_safe(path, 1); + btrfs_mark_buffer_dirty(path->nodes[0]); btrfs_release_path(root->fs_info->extent_root, path); path->reada = 1; + path->leave_spinning = 1; + + /* now insert the actual backref */ ret = insert_extent_backref(trans, root->fs_info->extent_root, path, bytenr, parent, ref_root, ref_generation, - owner_objectid); + owner_objectid, refs_to_add); BUG_ON(ret); - finish_current_insert(trans, root->fs_info->extent_root, 0); - del_pending_extents(trans, root->fs_info->extent_root, 0); - btrfs_free_path(path); return 0; } @@ -1314,68 +821,278 @@ int btrfs_inc_extent_ref(struct btrfs_trans_handle *trans, if (ref_root == BTRFS_TREE_LOG_OBJECTID && owner_objectid < BTRFS_FIRST_FREE_OBJECTID) return 0; - ret = __btrfs_inc_extent_ref(trans, root, bytenr, 0, parent, + + ret = __btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0, parent, 0, ref_root, 0, ref_generation, owner_objectid); return ret; } -int btrfs_extent_post_op(struct btrfs_trans_handle *trans, - struct btrfs_root *root) +static int drop_delayed_ref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_delayed_ref_node *node) +{ + int ret = 0; + struct btrfs_delayed_ref *ref = btrfs_delayed_node_to_ref(node); + + BUG_ON(node->ref_mod == 0); + ret = __btrfs_free_extent(trans, root, node->bytenr, node->num_bytes, + node->parent, ref->root, ref->generation, + ref->owner_objectid, ref->pin, node->ref_mod); + + return ret; +} + +/* helper function to actually process a single delayed ref entry */ +static noinline int run_one_delayed_ref(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct btrfs_delayed_ref_node *node, + int insert_reserved) { - u64 start; - u64 end; int ret; + struct btrfs_delayed_ref *ref; + + if (node->parent == (u64)-1) { + struct btrfs_delayed_ref_head *head; + /* + * we've hit the end of the chain and we were supposed + * to insert this extent into the tree. But, it got + * deleted before we ever needed to insert it, so all + * we have to do is clean up the accounting + */ + if (insert_reserved) { + update_reserved_extents(root, node->bytenr, + node->num_bytes, 0); + } + head = btrfs_delayed_node_to_head(node); + mutex_unlock(&head->mutex); + return 0; + } - while(1) { - finish_current_insert(trans, root->fs_info->extent_root, 1); - del_pending_extents(trans, root->fs_info->extent_root, 1); + ref = btrfs_delayed_node_to_ref(node); + if (ref->action == BTRFS_ADD_DELAYED_REF) { + if (insert_reserved) { + struct btrfs_key ins; - /* is there more work to do? */ - ret = find_first_extent_bit(&root->fs_info->pending_del, - 0, &start, &end, EXTENT_WRITEBACK); - if (!ret) - continue; - ret = find_first_extent_bit(&root->fs_info->extent_ins, - 0, &start, &end, EXTENT_WRITEBACK); - if (!ret) - continue; - break; + ins.objectid = node->bytenr; + ins.offset = node->num_bytes; + ins.type = BTRFS_EXTENT_ITEM_KEY; + + /* record the full extent allocation */ + ret = __btrfs_alloc_reserved_extent(trans, root, + node->parent, ref->root, + ref->generation, ref->owner_objectid, + &ins, node->ref_mod); + update_reserved_extents(root, node->bytenr, + node->num_bytes, 0); + } else { + /* just add one backref */ + ret = add_extent_ref(trans, root, node->bytenr, + node->num_bytes, + node->parent, ref->root, ref->generation, + ref->owner_objectid, node->ref_mod); + } + BUG_ON(ret); + } else if (ref->action == BTRFS_DROP_DELAYED_REF) { + WARN_ON(insert_reserved); + ret = drop_delayed_ref(trans, root, node); } return 0; } -int btrfs_lookup_extent_ref(struct btrfs_trans_handle *trans, - struct btrfs_root *root, u64 bytenr, - u64 num_bytes, u32 *refs) +static noinline struct btrfs_delayed_ref_node * +select_delayed_ref(struct btrfs_delayed_ref_head *head) { - struct btrfs_path *path; + struct rb_node *node; + struct btrfs_delayed_ref_node *ref; + int action = BTRFS_ADD_DELAYED_REF; +again: + /* + * select delayed ref of type BTRFS_ADD_DELAYED_REF first. + * this prevents ref count from going down to zero when + * there still are pending delayed ref. + */ + node = rb_prev(&head->node.rb_node); + while (1) { + if (!node) + break; + ref = rb_entry(node, struct btrfs_delayed_ref_node, + rb_node); + if (ref->bytenr != head->node.bytenr) + break; + if (btrfs_delayed_node_to_ref(ref)->action == action) + return ref; + node = rb_prev(node); + } + if (action == BTRFS_ADD_DELAYED_REF) { + action = BTRFS_DROP_DELAYED_REF; + goto again; + } + return NULL; +} + +static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + struct list_head *cluster) +{ + struct btrfs_delayed_ref_root *delayed_refs; + struct btrfs_delayed_ref_node *ref; + struct btrfs_delayed_ref_head *locked_ref = NULL; int ret; - struct btrfs_key key; - struct extent_buffer *l; - struct btrfs_extent_item *item; + int count = 0; + int must_insert_reserved = 0; - WARN_ON(num_bytes < root->sectorsize); - path = btrfs_alloc_path(); - path->reada = 1; - key.objectid = bytenr; - key.offset = num_bytes; - btrfs_set_key_type(&key, BTRFS_EXTENT_ITEM_KEY); - ret = btrfs_search_slot(trans, root->fs_info->extent_root, &key, path, - 0, 0); - if (ret < 0) - goto out; - if (ret != 0) { - btrfs_print_leaf(root, path->nodes[0]); - printk(KERN_INFO "btrfs failed to find block number %llu\n", - (unsigned long long)bytenr); - BUG(); + delayed_refs = &trans->transaction->delayed_refs; + while (1) { + if (!locked_ref) { + /* pick a new head ref from the cluster list */ + if (list_empty(cluster)) + break; + + locked_ref = list_entry(cluster->next, + struct btrfs_delayed_ref_head, cluster); + + /* grab the lock that says we are going to process + * all the refs for this head */ + ret = btrfs_delayed_ref_lock(trans, locked_ref); + + /* + * we may have dropped the spin lock to get the head + * mutex lock, and that might have given someone else + * time to free the head. If that's true, it has been + * removed from our list and we can move on. + */ + if (ret == -EAGAIN) { + locked_ref = NULL; + count++; + continue; + } + } + + /* + * record the must insert reserved flag before we + * drop the spin lock. + */ + must_insert_reserved = locked_ref->must_insert_reserved; + locked_ref->must_insert_reserved = 0; + + /* + * locked_ref is the head node, so we have to go one + * node back for any delayed ref updates + */ + ref = select_delayed_ref(locked_ref); + if (!ref) { + /* All delayed refs have been processed, Go ahead + * and send the head node to run_one_delayed_ref, + * so that any accounting fixes can happen + */ + ref = &locked_ref->node; + list_del_init(&locked_ref->cluster); + locked_ref = NULL; + } + + ref->in_tree = 0; + rb_erase(&ref->rb_node, &delayed_refs->root); + delayed_refs->num_entries--; + spin_unlock(&delayed_refs->lock); + + ret = run_one_delayed_ref(trans, root, ref, + must_insert_reserved); + BUG_ON(ret); + btrfs_put_delayed_ref(ref); + + count++; + cond_resched(); + spin_lock(&delayed_refs->lock); + } + return count; +} + +/* + * this starts processing the delayed reference count updates and + * extent insertions we have queued up so far. count can be + * 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. + */ +int btrfs_run_delayed_refs(struct btrfs_trans_handle *trans, + struct btrfs_root *root, unsigned long count) +{ + struct rb_node *node; + struct btrfs_delayed_ref_root *delayed_refs; + struct btrfs_delayed_ref_node *ref; + struct list_head cluster; + int ret; + int run_all = count == (unsigned long)-1; + int run_most = 0; + + if (root == root->fs_info->extent_root) + root = root->fs_info->tree_root; + + delayed_refs = &trans->transaction->delayed_refs; + INIT_LIST_HEAD(&cluster); +again: + spin_lock(&delayed_refs->lock); + if (count == 0) { + count = delayed_refs->num_entries * 2; + run_most = 1; + } + while (1) { + if (!(run_all || run_most) && + delayed_refs->num_heads_ready < 64) + break; + + /* + * go find something we can process in the rbtree. We start at + * the beginning of the tree, and then build a cluster + * of refs to process starting at the first one we are able to + * lock + */ + ret = btrfs_find_ref_cluster(trans, &cluster, + delayed_refs->run_delayed_start); + if (ret) + break; + + ret = run_clustered_refs(trans, root, &cluster); + BUG_ON(ret < 0); + + count -= min_t(unsigned long, ret, count); + + if (count == 0) + break; + } + + if (run_all) { + node = rb_first(&delayed_refs->root); + if (!node) + goto out; + count = (unsigned long)-1; + + while (node) { + ref = rb_entry(node, struct btrfs_delayed_ref_node, + rb_node); + if (btrfs_delayed_ref_is_head(ref)) { + struct btrfs_delayed_ref_head *head; + + head = btrfs_delayed_node_to_head(ref); + atomic_inc(&ref->refs); + + spin_unlock(&delayed_refs->lock); + mutex_lock(&head->mutex); + mutex_unlock(&head->mutex); + + btrfs_put_delayed_ref(ref); + cond_resched(); + goto again; + } + node = rb_next(node); + } + spin_unlock(&delayed_refs->lock); + schedule_timeout(1); + goto again; } - l = path->nodes[0]; - item = btrfs_item_ptr(l, path->slots[0], struct btrfs_extent_item); - *refs = btrfs_extent_refs(l, item); out: - btrfs_free_path(path); + spin_unlock(&delayed_refs->lock); return 0; } @@ -1599,7 +1316,7 @@ noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans, int refi = 0; int slot; int (*process_func)(struct btrfs_trans_handle *, struct btrfs_root *, - u64, u64, u64, u64, u64, u64, u64, u64); + u64, u64, u64, u64, u64, u64, u64, u64, u64); ref_root = btrfs_header_owner(buf); ref_generation = btrfs_header_generation(buf); @@ -1671,12 +1388,19 @@ noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans, if (level == 0) { btrfs_item_key_to_cpu(buf, &key, slot); + fi = btrfs_item_ptr(buf, slot, + struct btrfs_file_extent_item); + + bytenr = btrfs_file_extent_disk_bytenr(buf, fi); + if (bytenr == 0) + continue; ret = process_func(trans, root, bytenr, - orig_buf->start, buf->start, - orig_root, ref_root, - orig_generation, ref_generation, - key.objectid); + btrfs_file_extent_disk_num_bytes(buf, fi), + orig_buf->start, buf->start, + orig_root, ref_root, + orig_generation, ref_generation, + key.objectid); if (ret) { faili = slot; @@ -1684,7 +1408,7 @@ noinline int btrfs_inc_ref(struct btrfs_trans_handle *trans, goto fail; } } else { - ret = process_func(trans, root, bytenr, + ret = process_func(trans, root, bytenr, buf->len, orig_buf->start, buf->start, orig_root, ref_root, orig_generation, ref_generation, @@ -1761,17 +1485,17 @@ int btrfs_update_ref(struct btrfs_trans_handle *trans, if (bytenr == 0) continue; ret = __btrfs_update_extent_ref(trans, root, bytenr, - orig_buf->start, buf->start, - orig_root, ref_root, - orig_generation, ref_generation, - key.objectid); + btrfs_file_extent_disk_num_bytes(buf, fi), + orig_buf->start, buf->start, + orig_root, ref_root, orig_generation, + ref_generation, key.objectid); if (ret) goto fail; } else { bytenr = btrfs_node_blockptr(buf, slot); ret = __btrfs_update_extent_ref(trans, root, bytenr, - orig_buf->start, buf->start, - orig_root, ref_root, + buf->len, orig_buf->start, + buf->start, orig_root, ref_root, orig_generation, ref_generation, level - 1); if (ret) @@ -1790,7 +1514,6 @@ static int write_one_cache_group(struct btrfs_trans_handle *trans, struct btrfs_block_group_cache *cache) { int ret; - int pending_ret; struct btrfs_root *extent_root = root->fs_info->extent_root; unsigned long bi; struct extent_buffer *leaf; @@ -1806,12 +1529,8 @@ static int write_one_cache_group(struct btrfs_trans_handle *trans, btrfs_mark_buffer_dirty(leaf); btrfs_release_path(extent_root, path); fail: - finish_current_insert(trans, extent_root, 0); - pending_ret = del_pending_extents(trans, extent_root, 0); if (ret) return ret; - if (pending_ret) - return pending_ret; return 0; } @@ -1899,7 +1618,6 @@ static int update_space_info(struct btrfs_fs_info *info, u64 flags, if (!found) return -ENOMEM; - list_add(&found->list, &info->space_info); INIT_LIST_HEAD(&found->block_groups); init_rwsem(&found->groups_sem); spin_lock_init(&found->lock); @@ -1909,9 +1627,11 @@ static int update_space_info(struct btrfs_fs_info *info, u64 flags, found->bytes_pinned = 0; found->bytes_reserved = 0; found->bytes_readonly = 0; + found->bytes_delalloc = 0; found->full = 0; found->force_alloc = 0; *space_info = found; + list_add_rcu(&found->list, &info->space_info); return 0; } @@ -1944,32 +1664,259 @@ static void set_block_group_readonly(struct btrfs_block_group_cache *cache) spin_unlock(&cache->space_info->lock); } -u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) +u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags) +{ + u64 num_devices = root->fs_info->fs_devices->rw_devices; + + if (num_devices == 1) + flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); + if (num_devices < 4) + flags &= ~BTRFS_BLOCK_GROUP_RAID10; + + if ((flags & BTRFS_BLOCK_GROUP_DUP) && + (flags & (BTRFS_BLOCK_GROUP_RAID1 | + BTRFS_BLOCK_GROUP_RAID10))) { + flags &= ~BTRFS_BLOCK_GROUP_DUP; + } + + if ((flags & BTRFS_BLOCK_GROUP_RAID1) && + (flags & BTRFS_BLOCK_GROUP_RAID10)) { + flags &= ~BTRFS_BLOCK_GROUP_RAID1; + } + + if ((flags & BTRFS_BLOCK_GROUP_RAID0) && + ((flags & BTRFS_BLOCK_GROUP_RAID1) | + (flags & BTRFS_BLOCK_GROUP_RAID10) | + (flags & BTRFS_BLOCK_GROUP_DUP))) + flags &= ~BTRFS_BLOCK_GROUP_RAID0; + return flags; +} + +static u64 btrfs_get_alloc_profile(struct btrfs_root *root, u64 data) +{ + struct btrfs_fs_info *info = root->fs_info; + u64 alloc_profile; + + if (data) { + alloc_profile = info->avail_data_alloc_bits & + info->data_alloc_profile; + data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; + } else if (root == root->fs_info->chunk_root) { + alloc_profile = info->avail_system_alloc_bits & + info->system_alloc_profile; + data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; + } else { + alloc_profile = info->avail_metadata_alloc_bits & + info->metadata_alloc_profile; + data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; + } + + return btrfs_reduce_alloc_profile(root, data); +} + +void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *inode) +{ + u64 alloc_target; + + alloc_target = btrfs_get_alloc_profile(root, 1); + BTRFS_I(inode)->space_info = __find_space_info(root->fs_info, + alloc_target); +} + +/* + * for now this just makes sure we have at least 5% of our metadata space free + * for use. + */ +int btrfs_check_metadata_free_space(struct btrfs_root *root) +{ + struct btrfs_fs_info *info = root->fs_info; + struct btrfs_space_info *meta_sinfo; + u64 alloc_target, thresh; + int committed = 0, ret; + + /* get the space info for where the metadata will live */ + alloc_target = btrfs_get_alloc_profile(root, 0); + meta_sinfo = __find_space_info(info, alloc_target); + +again: + spin_lock(&meta_sinfo->lock); + if (!meta_sinfo->full) + thresh = meta_sinfo->total_bytes * 80; + else + thresh = meta_sinfo->total_bytes * 95; + + do_div(thresh, 100); + + if (meta_sinfo->bytes_used + meta_sinfo->bytes_reserved + + meta_sinfo->bytes_pinned + meta_sinfo->bytes_readonly > thresh) { + struct btrfs_trans_handle *trans; + if (!meta_sinfo->full) { + meta_sinfo->force_alloc = 1; + spin_unlock(&meta_sinfo->lock); + + trans = btrfs_start_transaction(root, 1); + if (!trans) + return -ENOMEM; + + ret = do_chunk_alloc(trans, root->fs_info->extent_root, + 2 * 1024 * 1024, alloc_target, 0); + btrfs_end_transaction(trans, root); + goto again; + } + spin_unlock(&meta_sinfo->lock); + + if (!committed) { + committed = 1; + trans = btrfs_join_transaction(root, 1); + if (!trans) + return -ENOMEM; + ret = btrfs_commit_transaction(trans, root); + if (ret) + return ret; + goto again; + } + return -ENOSPC; + } + spin_unlock(&meta_sinfo->lock); + + return 0; +} + +/* + * This will check the space that the inode allocates from to make sure we have + * enough space for bytes. + */ +int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode, + u64 bytes) +{ + struct btrfs_space_info *data_sinfo; + int ret = 0, committed = 0; + + /* make sure bytes are sectorsize aligned */ + bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); + + data_sinfo = BTRFS_I(inode)->space_info; +again: + /* make sure we have enough space to handle the data first */ + spin_lock(&data_sinfo->lock); + if (data_sinfo->total_bytes - data_sinfo->bytes_used - + data_sinfo->bytes_delalloc - data_sinfo->bytes_reserved - + data_sinfo->bytes_pinned - data_sinfo->bytes_readonly - + data_sinfo->bytes_may_use < bytes) { + struct btrfs_trans_handle *trans; + + /* + * if we don't have enough free bytes in this space then we need + * to alloc a new chunk. + */ + if (!data_sinfo->full) { + u64 alloc_target; + + data_sinfo->force_alloc = 1; + spin_unlock(&data_sinfo->lock); + + alloc_target = btrfs_get_alloc_profile(root, 1); + trans = btrfs_start_transaction(root, 1); + if (!trans) + return -ENOMEM; + + ret = do_chunk_alloc(trans, root->fs_info->extent_root, + bytes + 2 * 1024 * 1024, + alloc_target, 0); + btrfs_end_transaction(trans, root); + if (ret) + return ret; + goto again; + } + spin_unlock(&data_sinfo->lock); + + /* commit the current transaction and try again */ + if (!committed) { + committed = 1; + trans = btrfs_join_transaction(root, 1); + if (!trans) + return -ENOMEM; + ret = btrfs_commit_transaction(trans, root); + if (ret) + return ret; + goto again; + } + + printk(KERN_ERR "no space left, need %llu, %llu delalloc bytes" + ", %llu bytes_used, %llu bytes_reserved, " + "%llu bytes_pinned, %llu bytes_readonly, %llu may use" + "%llu total\n", bytes, data_sinfo->bytes_delalloc, + data_sinfo->bytes_used, data_sinfo->bytes_reserved, + data_sinfo->bytes_pinned, data_sinfo->bytes_readonly, + data_sinfo->bytes_may_use, data_sinfo->total_bytes); + return -ENOSPC; + } + data_sinfo->bytes_may_use += bytes; + BTRFS_I(inode)->reserved_bytes += bytes; + spin_unlock(&data_sinfo->lock); + + return btrfs_check_metadata_free_space(root); +} + +/* + * if there was an error for whatever reason after calling + * btrfs_check_data_free_space, call this so we can cleanup the counters. + */ +void btrfs_free_reserved_data_space(struct btrfs_root *root, + struct inode *inode, u64 bytes) { - u64 num_devices = root->fs_info->fs_devices->rw_devices; + struct btrfs_space_info *data_sinfo; - if (num_devices == 1) - flags &= ~(BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID0); - if (num_devices < 4) - flags &= ~BTRFS_BLOCK_GROUP_RAID10; + /* make sure bytes are sectorsize aligned */ + bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1); - if ((flags & BTRFS_BLOCK_GROUP_DUP) && - (flags & (BTRFS_BLOCK_GROUP_RAID1 | - BTRFS_BLOCK_GROUP_RAID10))) { - flags &= ~BTRFS_BLOCK_GROUP_DUP; - } + data_sinfo = BTRFS_I(inode)->space_info; + spin_lock(&data_sinfo->lock); + data_sinfo->bytes_may_use -= bytes; + BTRFS_I(inode)->reserved_bytes -= bytes; + spin_unlock(&data_sinfo->lock); +} - if ((flags & BTRFS_BLOCK_GROUP_RAID1) && - (flags & BTRFS_BLOCK_GROUP_RAID10)) { - flags &= ~BTRFS_BLOCK_GROUP_RAID1; +/* called when we are adding a delalloc extent to the inode's io_tree */ +void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode, + u64 bytes) +{ + struct btrfs_space_info *data_sinfo; + + /* get the space info for where this inode will be storing its data */ + data_sinfo = BTRFS_I(inode)->space_info; + + /* make sure we have enough space to handle the data first */ + spin_lock(&data_sinfo->lock); + data_sinfo->bytes_delalloc += bytes; + + /* + * we are adding a delalloc extent without calling + * btrfs_check_data_free_space first. This happens on a weird + * writepage condition, but shouldn't hurt our accounting + */ + if (unlikely(bytes > BTRFS_I(inode)->reserved_bytes)) { + data_sinfo->bytes_may_use -= BTRFS_I(inode)->reserved_bytes; + BTRFS_I(inode)->reserved_bytes = 0; + } else { + data_sinfo->bytes_may_use -= bytes; + BTRFS_I(inode)->reserved_bytes -= bytes; } - if ((flags & BTRFS_BLOCK_GROUP_RAID0) && - ((flags & BTRFS_BLOCK_GROUP_RAID1) | - (flags & BTRFS_BLOCK_GROUP_RAID10) | - (flags & BTRFS_BLOCK_GROUP_DUP))) - flags &= ~BTRFS_BLOCK_GROUP_RAID0; - return flags; + spin_unlock(&data_sinfo->lock); +} + +/* called when we are clearing an delalloc extent from the inode's io_tree */ +void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode, + u64 bytes) +{ + struct btrfs_space_info *info; + + info = BTRFS_I(inode)->space_info; + + spin_lock(&info->lock); + info->bytes_delalloc -= bytes; + spin_unlock(&info->lock); } static int do_chunk_alloc(struct btrfs_trans_handle *trans, @@ -2108,6 +2055,8 @@ int btrfs_update_pinned_extents(struct btrfs_root *root, clear_extent_dirty(&fs_info->pinned_extents, bytenr, bytenr + num - 1, GFP_NOFS); } + mutex_unlock(&root->fs_info->pinned_mutex); + while (num > 0) { cache = btrfs_lookup_block_group(fs_info, bytenr); BUG_ON(!cache); @@ -2199,8 +2148,8 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, u64 end; int ret; - mutex_lock(&root->fs_info->pinned_mutex); while (1) { + mutex_lock(&root->fs_info->pinned_mutex); ret = find_first_extent_bit(unpin, 0, &start, &end, EXTENT_DIRTY); if (ret) @@ -2208,209 +2157,21 @@ int btrfs_finish_extent_commit(struct btrfs_trans_handle *trans, ret = btrfs_discard_extent(root, start, end + 1 - start); + /* unlocks the pinned mutex */ btrfs_update_pinned_extents(root, start, end + 1 - start, 0); clear_extent_dirty(unpin, start, end, GFP_NOFS); - if (need_resched()) { - mutex_unlock(&root->fs_info->pinned_mutex); - cond_resched(); - mutex_lock(&root->fs_info->pinned_mutex); - } + cond_resched(); } mutex_unlock(&root->fs_info->pinned_mutex); return ret; } -static int finish_current_insert(struct btrfs_trans_handle *trans, - struct btrfs_root *extent_root, int all) -{ - u64 start; - u64 end; - u64 priv; - u64 search = 0; - struct btrfs_fs_info *info = extent_root->fs_info; - struct btrfs_path *path; - struct pending_extent_op *extent_op, *tmp; - struct list_head insert_list, update_list; - int ret; - int num_inserts = 0, max_inserts, restart = 0; - - path = btrfs_alloc_path(); - INIT_LIST_HEAD(&insert_list); - INIT_LIST_HEAD(&update_list); - - max_inserts = extent_root->leafsize / - (2 * sizeof(struct btrfs_key) + 2 * sizeof(struct btrfs_item) + - sizeof(struct btrfs_extent_ref) + - sizeof(struct btrfs_extent_item)); -again: - mutex_lock(&info->extent_ins_mutex); - while (1) { - ret = find_first_extent_bit(&info->extent_ins, search, &start, - &end, EXTENT_WRITEBACK); - if (ret) { - if (restart && !num_inserts && - list_empty(&update_list)) { - restart = 0; - search = 0; - continue; - } - break; - } - - ret = try_lock_extent(&info->extent_ins, start, end, GFP_NOFS); - if (!ret) { - if (all) - restart = 1; - search = end + 1; - if (need_resched()) { - mutex_unlock(&info->extent_ins_mutex); - cond_resched(); - mutex_lock(&info->extent_ins_mutex); - } - continue; - } - - ret = get_state_private(&info->extent_ins, start, &priv); - BUG_ON(ret); - extent_op = (struct pending_extent_op *)(unsigned long) priv; - - if (extent_op->type == PENDING_EXTENT_INSERT) { - num_inserts++; - list_add_tail(&extent_op->list, &insert_list); - search = end + 1; - if (num_inserts == max_inserts) { - restart = 1; - break; - } - } else if (extent_op->type == PENDING_BACKREF_UPDATE) { - list_add_tail(&extent_op->list, &update_list); - search = end + 1; - } else { - BUG(); - } - } - - /* - * process the update list, clear the writeback bit for it, and if - * somebody marked this thing for deletion then just unlock it and be - * done, the free_extents will handle it - */ - list_for_each_entry_safe(extent_op, tmp, &update_list, list) { - clear_extent_bits(&info->extent_ins, extent_op->bytenr, - extent_op->bytenr + extent_op->num_bytes - 1, - EXTENT_WRITEBACK, GFP_NOFS); - if (extent_op->del) { - list_del_init(&extent_op->list); - unlock_extent(&info->extent_ins, extent_op->bytenr, - extent_op->bytenr + extent_op->num_bytes - - 1, GFP_NOFS); - kfree(extent_op); - } - } - mutex_unlock(&info->extent_ins_mutex); - - /* - * still have things left on the update list, go ahead an update - * everything - */ - if (!list_empty(&update_list)) { - ret = update_backrefs(trans, extent_root, path, &update_list); - BUG_ON(ret); - - /* we may have COW'ed new blocks, so lets start over */ - if (all) - restart = 1; - } - - /* - * if no inserts need to be done, but we skipped some extents and we - * need to make sure everything is cleaned then reset everything and - * go back to the beginning - */ - if (!num_inserts && restart) { - search = 0; - restart = 0; - INIT_LIST_HEAD(&update_list); - INIT_LIST_HEAD(&insert_list); - goto again; - } else if (!num_inserts) { - goto out; - } - - /* - * process the insert extents list. Again if we are deleting this - * extent, then just unlock it, pin down the bytes if need be, and be - * done with it. Saves us from having to actually insert the extent - * into the tree and then subsequently come along and delete it - */ - mutex_lock(&info->extent_ins_mutex); - list_for_each_entry_safe(extent_op, tmp, &insert_list, list) { - clear_extent_bits(&info->extent_ins, extent_op->bytenr, - extent_op->bytenr + extent_op->num_bytes - 1, - EXTENT_WRITEBACK, GFP_NOFS); - if (extent_op->del) { - u64 used; - list_del_init(&extent_op->list); - unlock_extent(&info->extent_ins, extent_op->bytenr, - extent_op->bytenr + extent_op->num_bytes - - 1, GFP_NOFS); - - mutex_lock(&extent_root->fs_info->pinned_mutex); - ret = pin_down_bytes(trans, extent_root, - extent_op->bytenr, - extent_op->num_bytes, 0); - mutex_unlock(&extent_root->fs_info->pinned_mutex); - - spin_lock(&info->delalloc_lock); - used = btrfs_super_bytes_used(&info->super_copy); - btrfs_set_super_bytes_used(&info->super_copy, - used - extent_op->num_bytes); - used = btrfs_root_used(&extent_root->root_item); - btrfs_set_root_used(&extent_root->root_item, - used - extent_op->num_bytes); - spin_unlock(&info->delalloc_lock); - - ret = update_block_group(trans, extent_root, - extent_op->bytenr, - extent_op->num_bytes, - 0, ret > 0); - BUG_ON(ret); - kfree(extent_op); - num_inserts--; - } - } - mutex_unlock(&info->extent_ins_mutex); - - ret = insert_extents(trans, extent_root, path, &insert_list, - num_inserts); - BUG_ON(ret); - - /* - * if restart is set for whatever reason we need to go back and start - * searching through the pending list again. - * - * We just inserted some extents, which could have resulted in new - * blocks being allocated, which would result in new blocks needing - * updates, so if all is set we _must_ restart to get the updated - * blocks. - */ - if (restart || all) { - INIT_LIST_HEAD(&insert_list); - INIT_LIST_HEAD(&update_list); - search = 0; - restart = 0; - num_inserts = 0; - goto again; - } -out: - btrfs_free_path(path); - return 0; -} - static int pin_down_bytes(struct btrfs_trans_handle *trans, struct btrfs_root *root, - u64 bytenr, u64 num_bytes, int is_data) + struct btrfs_path *path, + u64 bytenr, u64 num_bytes, int is_data, + struct extent_buffer **must_clean) { int err = 0; struct extent_buffer *buf; @@ -2433,17 +2194,19 @@ static int pin_down_bytes(struct btrfs_trans_handle *trans, u64 header_transid = btrfs_header_generation(buf); if (header_owner != BTRFS_TREE_LOG_OBJECTID && header_owner != BTRFS_TREE_RELOC_OBJECTID && + header_owner != BTRFS_DATA_RELOC_TREE_OBJECTID && header_transid == trans->transid && !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) { - clean_tree_block(NULL, root, buf); - btrfs_tree_unlock(buf); - free_extent_buffer(buf); + *must_clean = buf; return 1; } btrfs_tree_unlock(buf); } free_extent_buffer(buf); pinit: + btrfs_set_path_blocking(path); + mutex_lock(&root->fs_info->pinned_mutex); + /* unlocks the pinned mutex */ btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); BUG_ON(err < 0); @@ -2457,7 +2220,8 @@ static int __free_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 bytenr, u64 num_bytes, u64 parent, u64 root_objectid, u64 ref_generation, - u64 owner_objectid, int pin, int mark_free) + u64 owner_objectid, int pin, int mark_free, + int refs_to_drop) { struct btrfs_path *path; struct btrfs_key key; @@ -2479,6 +2243,7 @@ static int __free_extent(struct btrfs_trans_handle *trans, return -ENOMEM; path->reada = 1; + path->leave_spinning = 1; ret = lookup_extent_backref(trans, extent_root, path, bytenr, parent, root_objectid, ref_generation, owner_objectid, 1); @@ -2500,9 +2265,11 @@ static int __free_extent(struct btrfs_trans_handle *trans, break; } if (!found_extent) { - ret = remove_extent_backref(trans, extent_root, path); + ret = remove_extent_backref(trans, extent_root, path, + refs_to_drop); BUG_ON(ret); btrfs_release_path(extent_root, path); + path->leave_spinning = 1; ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); if (ret) { @@ -2518,8 +2285,9 @@ static int __free_extent(struct btrfs_trans_handle *trans, btrfs_print_leaf(extent_root, path->nodes[0]); WARN_ON(1); printk(KERN_ERR "btrfs unable to find ref byte nr %llu " - "root %llu gen %llu owner %llu\n", + "parent %llu root %llu gen %llu owner %llu\n", (unsigned long long)bytenr, + (unsigned long long)parent, (unsigned long long)root_objectid, (unsigned long long)ref_generation, (unsigned long long)owner_objectid); @@ -2529,17 +2297,23 @@ static int __free_extent(struct btrfs_trans_handle *trans, ei = btrfs_item_ptr(leaf, extent_slot, struct btrfs_extent_item); refs = btrfs_extent_refs(leaf, ei); - BUG_ON(refs == 0); - refs -= 1; - btrfs_set_extent_refs(leaf, ei, refs); + /* + * we're not allowed to delete the extent item if there + * are other delayed ref updates pending + */ + + BUG_ON(refs < refs_to_drop); + refs -= refs_to_drop; + btrfs_set_extent_refs(leaf, ei, refs); btrfs_mark_buffer_dirty(leaf); - if (refs == 0 && found_extent && path->slots[0] == extent_slot + 1) { + if (refs == 0 && found_extent && + path->slots[0] == extent_slot + 1) { struct btrfs_extent_ref *ref; ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_extent_ref); - BUG_ON(btrfs_ref_num_refs(leaf, ref) != 1); + BUG_ON(btrfs_ref_num_refs(leaf, ref) != refs_to_drop); /* if the back ref and the extent are next to each other * they get deleted below in one shot */ @@ -2547,11 +2321,13 @@ static int __free_extent(struct btrfs_trans_handle *trans, num_to_del = 2; } else if (found_extent) { /* otherwise delete the extent back ref */ - ret = remove_extent_backref(trans, extent_root, path); + ret = remove_extent_backref(trans, extent_root, path, + refs_to_drop); BUG_ON(ret); /* if refs are 0, we need to setup the path for deletion */ if (refs == 0) { btrfs_release_path(extent_root, path); + path->leave_spinning = 1; ret = btrfs_search_slot(trans, extent_root, &key, path, -1, 1); BUG_ON(ret); @@ -2561,16 +2337,18 @@ static int __free_extent(struct btrfs_trans_handle *trans, if (refs == 0) { u64 super_used; u64 root_used; + struct extent_buffer *must_clean = NULL; if (pin) { - mutex_lock(&root->fs_info->pinned_mutex); - ret = pin_down_bytes(trans, root, bytenr, num_bytes, - owner_objectid >= BTRFS_FIRST_FREE_OBJECTID); - mutex_unlock(&root->fs_info->pinned_mutex); + ret = pin_down_bytes(trans, root, path, + bytenr, num_bytes, + owner_objectid >= BTRFS_FIRST_FREE_OBJECTID, + &must_clean); if (ret > 0) mark_free = 1; BUG_ON(ret < 0); } + /* block accounting for super block */ spin_lock(&info->delalloc_lock); super_used = btrfs_super_bytes_used(&info->super_copy); @@ -2582,11 +2360,27 @@ static int __free_extent(struct btrfs_trans_handle *trans, btrfs_set_root_used(&root->root_item, root_used - num_bytes); spin_unlock(&info->delalloc_lock); + + /* + * it is going to be very rare for someone to be waiting + * on the block we're freeing. del_items might need to + * schedule, so rather than get fancy, just force it + * to blocking here + */ + if (must_clean) + btrfs_set_lock_blocking(must_clean); + ret = btrfs_del_items(trans, extent_root, path, path->slots[0], num_to_del); BUG_ON(ret); btrfs_release_path(extent_root, path); + if (must_clean) { + clean_tree_block(NULL, root, must_clean); + btrfs_tree_unlock(must_clean); + free_extent_buffer(must_clean); + } + if (owner_objectid >= BTRFS_FIRST_FREE_OBJECTID) { ret = btrfs_del_csums(trans, root, bytenr, num_bytes); BUG_ON(ret); @@ -2597,218 +2391,103 @@ static int __free_extent(struct btrfs_trans_handle *trans, BUG_ON(ret); } btrfs_free_path(path); - finish_current_insert(trans, extent_root, 0); return ret; } /* - * find all the blocks marked as pending in the radix tree and remove - * them from the extent map + * remove an extent from the root, returns 0 on success */ -static int del_pending_extents(struct btrfs_trans_handle *trans, - struct btrfs_root *extent_root, int all) +static int __btrfs_free_extent(struct btrfs_trans_handle *trans, + struct btrfs_root *root, + u64 bytenr, u64 num_bytes, u64 parent, + u64 root_objectid, u64 ref_generation, + u64 owner_objectid, int pin, + int refs_to_drop) { - int ret; - int err = 0; - u64 start; - u64 end; - u64 priv; - u64 search = 0; - int nr = 0, skipped = 0; - struct extent_io_tree *pending_del; - struct extent_io_tree *extent_ins; - struct pending_extent_op *extent_op; - struct btrfs_fs_info *info = extent_root->fs_info; - struct list_head delete_list; - - INIT_LIST_HEAD(&delete_list); - extent_ins = &extent_root->fs_info->extent_ins; - pending_del = &extent_root->fs_info->pending_del; - -again: - mutex_lock(&info->extent_ins_mutex); - while (1) { - ret = find_first_extent_bit(pending_del, search, &start, &end, - EXTENT_WRITEBACK); - if (ret) { - if (all && skipped && !nr) { - search = 0; - skipped = 0; - continue; - } - mutex_unlock(&info->extent_ins_mutex); - break; - } - - ret = try_lock_extent(extent_ins, start, end, GFP_NOFS); - if (!ret) { - search = end+1; - skipped = 1; - - if (need_resched()) { - mutex_unlock(&info->extent_ins_mutex); - cond_resched(); - mutex_lock(&info->extent_ins_mutex); - } - - continue; - } - BUG_ON(ret < 0); - - ret = get_state_private(pending_del, start, &priv); - BUG_ON(ret); - extent_op = (struct pending_extent_op *)(unsigned long)priv; - - clear_extent_bits(pending_del, start, end, EXTENT_WRITEBACK, - GFP_NOFS); - if (!test_range_bit(extent_ins, start, end, - EXTENT_WRITEBACK, 0)) { - list_add_tail(&extent_op->list, &delete_list); - nr++; - } else { - kfree(extent_op); - - ret = get_state_private(&info->extent_ins, start, - &priv); - BUG_ON(ret); - extent_op = (struct pending_extent_op *) - (unsigned long)priv; - - clear_extent_bits(&info->extent_ins, start, end, - EXTENT_WRITEBACK, GFP_NOFS); - - if (extent_op->type == PENDING_BACKREF_UPDATE) { - list_add_tail(&extent_op->list, &delete_list); - search = end + 1; - nr++; - continue; - } - - mutex_lock(&extent_root->fs_info->pinned_mutex); - ret = pin_down_bytes(trans, extent_root, start, - end + 1 - start, 0); - mutex_unlock(&extent_root->fs_info->pinned_mutex); - - ret = update_block_group(trans, extent_root, start, - end + 1 - start, 0, ret > 0); - - unlock_extent(extent_ins, start, end, GFP_NOFS); - BUG_ON(ret); - kfree(extent_op); - } - if (ret) - err = ret; - - search = end + 1; - - if (need_resched()) { - mutex_unlock(&info->extent_ins_mutex); - cond_resched(); - mutex_lock(&info->extent_ins_mutex); - } - } + WARN_ON(num_bytes < root->sectorsize); - if (nr) { - ret = free_extents(trans, extent_root, &delete_list); - BUG_ON(ret); - } + /* + * if metadata always pin + * if data pin when any transaction has committed this + */ + if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID || + ref_generation != trans->transid) + pin = 1; - if (all && skipped) { - INIT_LIST_HEAD(&delete_list); - search = 0; - nr = 0; - goto again; - } + if (ref_generation != trans->transid) + pin = 1; - if (!err) - finish_current_insert(trans, extent_root, 0); - return err; + return __free_extent(trans, root, bytenr, num_bytes, parent, + root_objectid, ref_generation, + owner_objectid, pin, pin == 0, refs_to_drop); } /* - * remove an extent from the root, returns 0 on success + * when we free an extent, it is possible (and likely) that we free the last + * delayed ref for that extent as well. This searches the delayed ref tree for + * a given extent, and if there are no other delayed refs to be processed, it + * removes it from the tree. */ -static int __btrfs_free_extent(struct btrfs_trans_handle *trans, - struct btrfs_root *root, - u64 bytenr, u64 num_bytes, u64 parent, - u64 root_objectid, u64 ref_generation, - u64 owner_objectid, int pin) +static noinline int check_ref_cleanup(struct btrfs_trans_handle *trans, + struct btrfs_root *root, u64 bytenr) { - struct btrfs_root *extent_root = root->fs_info->extent_root; - int pending_ret; + struct btrfs_delayed_ref_head *head; + struct btrfs_delayed_ref_root *delayed_refs; + struct btrfs_delayed_ref_node *ref; + struct rb_node *node; int ret; - WARN_ON(num_bytes < root->sectorsize); - if (root == extent_root) { - struct pending_extent_op *extent_op = NULL; - - mutex_lock(&root->fs_info->extent_ins_mutex); - if (test_range_bit(&root->fs_info->extent_ins, bytenr, - bytenr + num_bytes - 1, EXTENT_WRITEBACK, 0)) { - u64 priv; - ret = get_state_private(&root->fs_info->extent_ins, - bytenr, &priv); - BUG_ON(ret); - extent_op = (struct pending_extent_op *) - (unsigned long)priv; + delayed_refs = &trans->transaction->delayed_refs; + spin_lock(&delayed_refs->lock); + head = btrfs_find_delayed_ref_head(trans, bytenr); + if (!head) + goto out; - extent_op->del = 1; - if (extent_op->type == PENDING_EXTENT_INSERT) { - mutex_unlock(&root->fs_info->extent_ins_mutex); - return 0; - } - } + node = rb_prev(&head->node.rb_node); + if (!node) + goto out; - if (extent_op) { - ref_generation = extent_op->orig_generation; - parent = extent_op->orig_parent; - } + ref = rb_entry(node, struct btrfs_delayed_ref_node, rb_node); - extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); - BUG_ON(!extent_op); - - extent_op->type = PENDING_EXTENT_DELETE; - extent_op->bytenr = bytenr; - extent_op->num_bytes = num_bytes; - extent_op->parent = parent; - extent_op->orig_parent = parent; - extent_op->generation = ref_generation; - extent_op->orig_generation = ref_generation; - extent_op->level = (int)owner_objectid; - INIT_LIST_HEAD(&extent_op->list); - extent_op->del = 0; - - set_extent_bits(&root->fs_info->pending_del, - bytenr, bytenr + num_bytes - 1, - EXTENT_WRITEBACK, GFP_NOFS); - set_state_private(&root->fs_info->pending_del, - bytenr, (unsigned long)extent_op); - mutex_unlock(&root->fs_info->extent_ins_mutex); - return 0; - } - /* if metadata always pin */ - if (owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { - if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID) { - mutex_lock(&root->fs_info->pinned_mutex); - btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); - mutex_unlock(&root->fs_info->pinned_mutex); - update_reserved_extents(root, bytenr, num_bytes, 0); - return 0; - } - pin = 1; - } + /* there are still entries for this ref, we can't drop it */ + if (ref->bytenr == bytenr) + goto out; - /* if data pin when any transaction has committed this */ - if (ref_generation != trans->transid) - pin = 1; + /* + * waiting for the lock here would deadlock. If someone else has it + * locked they are already in the process of dropping it anyway + */ + if (!mutex_trylock(&head->mutex)) + goto out; - ret = __free_extent(trans, root, bytenr, num_bytes, parent, - root_objectid, ref_generation, - owner_objectid, pin, pin == 0); + /* + * at this point we have a head with no other entries. Go + * ahead and process it. + */ + head->node.in_tree = 0; + rb_erase(&head->node.rb_node, &delayed_refs->root); + + delayed_refs->num_entries--; + + /* + * we don't take a ref on the node because we're removing it from the + * tree, so we just steal the ref the tree was holding. + */ + delayed_refs->num_heads--; + if (list_empty(&head->cluster)) + delayed_refs->num_heads_ready--; + + list_del_init(&head->cluster); + spin_unlock(&delayed_refs->lock); - finish_current_insert(trans, root->fs_info->extent_root, 0); - pending_ret = del_pending_extents(trans, root->fs_info->extent_root, 0); - return ret ? ret : pending_ret; + ret = run_one_delayed_ref(trans, root->fs_info->tree_root, + &head->node, head->must_insert_reserved); + BUG_ON(ret); + btrfs_put_delayed_ref(&head->node); + return 0; +out: + spin_unlock(&delayed_refs->lock); + return 0; } int btrfs_free_extent(struct btrfs_trans_handle *trans, @@ -2819,9 +2498,30 @@ int btrfs_free_extent(struct btrfs_trans_handle *trans, { int ret; - ret = __btrfs_free_extent(trans, root, bytenr, num_bytes, parent, - root_objectid, ref_generation, - owner_objectid, pin); + /* + * tree log blocks never actually go into the extent allocation + * tree, just update pinning info and exit early. + * + * data extents referenced by the tree log do need to have + * their reference counts bumped. + */ + if (root->root_key.objectid == BTRFS_TREE_LOG_OBJECTID && + owner_objectid < BTRFS_FIRST_FREE_OBJECTID) { + mutex_lock(&root->fs_info->pinned_mutex); + + /* unlocks the pinned mutex */ + btrfs_update_pinned_extents(root, bytenr, num_bytes, 1); + update_reserved_extents(root, bytenr, num_bytes, 0); + ret = 0; + } else { + ret = btrfs_add_delayed_ref(trans, bytenr, num_bytes, parent, + root_objectid, ref_generation, + owner_objectid, + BTRFS_DROP_DELAYED_REF, 1); + BUG_ON(ret); + ret = check_ref_cleanup(trans, root, bytenr); + BUG_ON(ret); + } return ret; } @@ -3105,6 +2805,10 @@ static void dump_space_info(struct btrfs_space_info *info, u64 bytes) (unsigned long long)(info->total_bytes - info->bytes_used - info->bytes_pinned - info->bytes_reserved), (info->full) ? "" : "not "); + printk(KERN_INFO "space_info total=%llu, pinned=%llu, delalloc=%llu," + " may_use=%llu, used=%llu\n", info->total_bytes, + info->bytes_pinned, info->bytes_delalloc, info->bytes_may_use, + info->bytes_used); down_read(&info->groups_sem); list_for_each_entry(cache, &info->block_groups, list) { @@ -3131,24 +2835,10 @@ static int __btrfs_reserve_extent(struct btrfs_trans_handle *trans, { int ret; u64 search_start = 0; - u64 alloc_profile; struct btrfs_fs_info *info = root->fs_info; - if (data) { - alloc_profile = info->avail_data_alloc_bits & - info->data_alloc_profile; - data = BTRFS_BLOCK_GROUP_DATA | alloc_profile; - } else if (root == root->fs_info->chunk_root) { - alloc_profile = info->avail_system_alloc_bits & - info->system_alloc_profile; - data = BTRFS_BLOCK_GROUP_SYSTEM | alloc_profile; - } else { - alloc_profile = info->avail_metadata_alloc_bits & - info->metadata_alloc_profile; - data = BTRFS_BLOCK_GROUP_METADATA | alloc_profile; - } + data = btrfs_get_alloc_profile(root, data); again: - data = btrfs_reduce_alloc_profile(root, data); /* * the only place that sets empty_size is btrfs_realloc_node, which * is not called recursively on allocations @@ -3232,10 +2922,10 @@ int btrfs_reserve_extent(struct btrfs_trans_handle *trans, static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, struct btrfs_root *root, u64 parent, u64 root_objectid, u64 ref_generation, - u64 owner, struct btrfs_key *ins) + u64 owner, struct btrfs_key *ins, + int ref_mod) { int ret; - int pending_ret; u64 super_used; u64 root_used; u64 num_bytes = ins->offset; @@ -3260,33 +2950,6 @@ static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, btrfs_set_root_used(&root->root_item, root_used + num_bytes); spin_unlock(&info->delalloc_lock); - if (root == extent_root) { - struct pending_extent_op *extent_op; - - extent_op = kmalloc(sizeof(*extent_op), GFP_NOFS); - BUG_ON(!extent_op); - - extent_op->type = PENDING_EXTENT_INSERT; - extent_op->bytenr = ins->objectid; - extent_op->num_bytes = ins->offset; - extent_op->parent = parent; - extent_op->orig_parent = 0; - extent_op->generation = ref_generation; - extent_op->orig_generation = 0; - extent_op->level = (int)owner; - INIT_LIST_HEAD(&extent_op->list); - extent_op->del = 0; - - mutex_lock(&root->fs_info->extent_ins_mutex); - set_extent_bits(&root->fs_info->extent_ins, ins->objectid, - ins->objectid + ins->offset - 1, - EXTENT_WRITEBACK, GFP_NOFS); - set_state_private(&root->fs_info->extent_ins, - ins->objectid, (unsigned long)extent_op); - mutex_unlock(&root->fs_info->extent_ins_mutex); - goto update_block; - } - memcpy(&keys[0], ins, sizeof(*ins)); keys[1].objectid = ins->objectid; keys[1].type = BTRFS_EXTENT_REF_KEY; @@ -3297,37 +2960,31 @@ static int __btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, path = btrfs_alloc_path(); BUG_ON(!path); + path->leave_spinning = 1; ret = btrfs_insert_empty_items(trans, extent_root, path, keys, sizes, 2); BUG_ON(ret); extent_item = btrfs_item_ptr(path->nodes[0], path->slots[0], struct btrfs_extent_item); - btrfs_set_extent_refs(path->nodes[0], extent_item, 1); + btrfs_set_extent_refs(path->nodes[0], extent_item, ref_mod); ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1, struct btrfs_extent_ref); btrfs_set_ref_root(path->nodes[0], ref, root_objectid); btrfs_set_ref_generation(path->nodes[0], ref, ref_generation); btrfs_set_ref_objectid(path->nodes[0], ref, owner); - btrfs_set_ref_num_refs(path->nodes[0], ref, 1); + btrfs_set_ref_num_refs(path->nodes[0], ref, ref_mod); btrfs_mark_buffer_dirty(path->nodes[0]); trans->alloc_exclude_start = 0; trans->alloc_exclude_nr = 0; btrfs_free_path(path); - finish_current_insert(trans, extent_root, 0); - pending_ret = del_pending_extents(trans, extent_root, 0); if (ret) goto out; - if (pending_ret) { - ret = pending_ret; - goto out; - } -update_block: ret = update_block_group(trans, root, ins->objectid, ins->offset, 1, 0); if (ret) { @@ -3349,9 +3006,12 @@ int btrfs_alloc_reserved_extent(struct btrfs_trans_handle *trans, if (root_objectid == BTRFS_TREE_LOG_OBJECTID) return 0; - ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid, - ref_generation, owner, ins); - update_reserved_extents(root, ins->objectid, ins->offset, 0); + + ret = btrfs_add_delayed_ref(trans, ins->objectid, + ins->offset, parent, root_objectid, + ref_generation, owner, + BTRFS_ADD_DELAYED_EXTENT, 0); + BUG_ON(ret); return ret; } @@ -3378,7 +3038,7 @@ int btrfs_alloc_logged_extent(struct btrfs_trans_handle *trans, BUG_ON(ret); put_block_group(block_group); ret = __btrfs_alloc_reserved_extent(trans, root, parent, root_objectid, - ref_generation, owner, ins); + ref_generation, owner, ins, 1); return ret; } @@ -3397,20 +3057,18 @@ int btrfs_alloc_extent(struct btrfs_trans_handle *trans, u64 search_end, struct btrfs_key *ins, u64 data) { int ret; - ret = __btrfs_reserve_extent(trans, root, num_bytes, min_alloc_size, empty_size, hint_byte, search_end, ins, data); BUG_ON(ret); if (root_objectid != BTRFS_TREE_LOG_OBJECTID) { - ret = __btrfs_alloc_reserved_extent(trans, root, parent, - root_objectid, ref_generation, - owner_objectid, ins); + ret = btrfs_add_delayed_ref(trans, ins->objectid, + ins->offset, parent, root_objectid, + ref_generation, owner_objectid, + BTRFS_ADD_DELAYED_EXTENT, 0); BUG_ON(ret); - - } else { - update_reserved_extents(root, ins->objectid, ins->offset, 1); } + update_reserved_extents(root, ins->objectid, ins->offset, 1); return ret; } @@ -3546,7 +3204,7 @@ int btrfs_drop_leaf_ref(struct btrfs_trans_handle *trans, fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item); - ret = __btrfs_free_extent(trans, root, disk_bytenr, + ret = btrfs_free_extent(trans, root, disk_bytenr, btrfs_file_extent_disk_num_bytes(leaf, fi), leaf->start, leaf_owner, leaf_generation, key.objectid, 0); @@ -3586,7 +3244,7 @@ static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans, */ for (i = 0; i < ref->nritems; i++) { info = ref->extents + sorted[i].slot; - ret = __btrfs_free_extent(trans, root, info->bytenr, + ret = btrfs_free_extent(trans, root, info->bytenr, info->num_bytes, ref->bytenr, ref->owner, ref->generation, info->objectid, 0); @@ -3603,12 +3261,13 @@ static noinline int cache_drop_leaf_ref(struct btrfs_trans_handle *trans, return 0; } -static int drop_snap_lookup_refcount(struct btrfs_root *root, u64 start, +static int drop_snap_lookup_refcount(struct btrfs_trans_handle *trans, + struct btrfs_root *root, u64 start, u64 len, u32 *refs) { int ret; - ret = btrfs_lookup_extent_ref(NULL, root, start, len, refs); + ret = btrfs_lookup_extent_ref(trans, root, start, len, refs); BUG_ON(ret); #if 0 /* some debugging code in case we see problems here */ @@ -3716,7 +3375,8 @@ static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans, * we just decrement it below and don't update any * of the refs the leaf points to. */ - ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs); + ret = drop_snap_lookup_refcount(trans, root, bytenr, + blocksize, &refs); BUG_ON(ret); if (refs != 1) continue; @@ -3767,7 +3427,7 @@ static noinline int drop_level_one_refs(struct btrfs_trans_handle *trans, */ for (i = 0; i < refi; i++) { bytenr = sorted[i].bytenr; - ret = __btrfs_free_extent(trans, root, bytenr, + ret = btrfs_free_extent(trans, root, bytenr, blocksize, eb->start, root_owner, root_gen, 0, 1); BUG_ON(ret); @@ -3810,7 +3470,7 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans, WARN_ON(*level < 0); WARN_ON(*level >= BTRFS_MAX_LEVEL); - ret = drop_snap_lookup_refcount(root, path->nodes[*level]->start, + ret = drop_snap_lookup_refcount(trans, root, path->nodes[*level]->start, path->nodes[*level]->len, &refs); BUG_ON(ret); if (refs > 1) @@ -3861,7 +3521,8 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans, ptr_gen = btrfs_node_ptr_generation(cur, path->slots[*level]); blocksize = btrfs_level_size(root, *level - 1); - ret = drop_snap_lookup_refcount(root, bytenr, blocksize, &refs); + ret = drop_snap_lookup_refcount(trans, root, bytenr, + blocksize, &refs); BUG_ON(ret); /* @@ -3876,7 +3537,7 @@ static noinline int walk_down_tree(struct btrfs_trans_handle *trans, root_gen = btrfs_header_generation(parent); path->slots[*level]++; - ret = __btrfs_free_extent(trans, root, bytenr, + ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent->start, root_owner, root_gen, *level - 1, 1); @@ -3922,7 +3583,7 @@ out: * cleanup and free the reference on the last node * we processed */ - ret = __btrfs_free_extent(trans, root, bytenr, blocksize, + ret = btrfs_free_extent(trans, root, bytenr, blocksize, parent->start, root_owner, root_gen, *level, 1); free_extent_buffer(path->nodes[*level]); @@ -4111,6 +3772,7 @@ int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root struct btrfs_path *path; int i; int orig_level; + int update_count; struct btrfs_root_item *root_item = &root->root_item; WARN_ON(!mutex_is_locked(&root->fs_info->drop_mutex)); @@ -4152,6 +3814,7 @@ int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root } } while (1) { + unsigned long update; wret = walk_down_tree(trans, root, path, &level); if (wret > 0) break; @@ -4164,12 +3827,21 @@ int btrfs_drop_snapshot(struct btrfs_trans_handle *trans, struct btrfs_root break; if (wret < 0) ret = wret; - if (trans->transaction->in_commit) { + if (trans->transaction->in_commit || + trans->transaction->delayed_refs.flushing) { ret = -EAGAIN; break; } atomic_inc(&root->fs_info->throttle_gen); wake_up(&root->fs_info->transaction_throttle); + for (update_count = 0; update_count < 16; update_count++) { + update = trans->delayed_ref_updates; + trans->delayed_ref_updates = 0; + if (update) + btrfs_run_delayed_refs(trans, root, update); + else + break; + } } for (i = 0; i <= orig_level; i++) { if (path->nodes[i]) { @@ -4196,13 +3868,13 @@ int btrfs_drop_subtree(struct btrfs_trans_handle *trans, path = btrfs_alloc_path(); BUG_ON(!path); - BUG_ON(!btrfs_tree_locked(parent)); + btrfs_assert_tree_locked(parent); parent_level = btrfs_header_level(parent); extent_buffer_get(parent); path->nodes[parent_level] = parent; path->slots[parent_level] = btrfs_header_nritems(parent); - BUG_ON(!btrfs_tree_locked(node)); + btrfs_assert_tree_locked(node); level = btrfs_header_level(node); extent_buffer_get(node); path->nodes[level] = node; @@ -5214,6 +4886,7 @@ static noinline int replace_extents_in_leaf(struct btrfs_trans_handle *trans, root->root_key.objectid, trans->transid, key.objectid); BUG_ON(ret); + ret = btrfs_free_extent(trans, root, bytenr, num_bytes, leaf->start, btrfs_header_owner(leaf), @@ -5525,9 +5198,6 @@ static noinline int relocate_tree_block(struct btrfs_trans_handle *trans, ref_path, NULL, NULL); BUG_ON(ret); - if (root == root->fs_info->extent_root) - btrfs_extent_post_op(trans, root); - return 0; } @@ -5795,6 +5465,7 @@ static int __insert_orphan_inode(struct btrfs_trans_handle *trans, if (!path) return -ENOMEM; + path->leave_spinning = 1; ret = btrfs_insert_empty_inode(trans, root, path, objectid); if (ret) goto out; @@ -5965,6 +5636,9 @@ again: btrfs_remove_leaf_refs(info->tree_root, (u64)-1, 1); mutex_unlock(&root->fs_info->cleaner_mutex); + trans = btrfs_start_transaction(info->tree_root, 1); + btrfs_commit_transaction(trans, info->tree_root); + while (1) { ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); if (ret < 0) @@ -6098,6 +5772,7 @@ out: int btrfs_free_block_groups(struct btrfs_fs_info *info) { struct btrfs_block_group_cache *block_group; + struct btrfs_space_info *space_info; struct rb_node *n; spin_lock(&info->block_group_cache_lock); @@ -6119,6 +5794,23 @@ int btrfs_free_block_groups(struct btrfs_fs_info *info) spin_lock(&info->block_group_cache_lock); } spin_unlock(&info->block_group_cache_lock); + + /* now that all the block groups are freed, go through and + * free all the space_info structs. This is only called during + * the final stages of unmount, and so we know nobody is + * using them. We call synchronize_rcu() once before we start, + * just to be on the safe side. + */ + synchronize_rcu(); + + while(!list_empty(&info->space_info)) { + space_info = list_entry(info->space_info.next, + struct btrfs_space_info, + list); + + list_del(&space_info->list); + kfree(space_info); + } return 0; } @@ -6239,9 +5931,6 @@ int btrfs_make_block_group(struct btrfs_trans_handle *trans, sizeof(cache->item)); BUG_ON(ret); - finish_current_insert(trans, extent_root, 0); - ret = del_pending_extents(trans, extent_root, 0); - BUG_ON(ret); set_avail_alloc_bits(extent_root->fs_info, type); return 0;