Btrfs: Make truncate(2) more ENOSPC friendly
authorYan, Zheng <zheng.yan@oracle.com>
Thu, 12 Nov 2009 09:35:36 +0000 (09:35 +0000)
committerChris Mason <chris.mason@oracle.com>
Thu, 17 Dec 2009 17:33:34 +0000 (12:33 -0500)
truncating and deleting regular files are unbound operations,
so it's not good to do them in a single transaction. This
patch makes btrfs_truncate and btrfs_delete_inode start a
new transaction after all items in a tree leaf are deleted.

Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
fs/btrfs/inode.c
fs/btrfs/relocation.c

index 8d8baaa..dcec42e 100644 (file)
@@ -2848,37 +2848,40 @@ out:
  * min_type is the minimum key type to truncate down to.  If set to 0, this
  * will kill all the items on this inode, including the INODE_ITEM_KEY.
  */
-noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
-                                       struct btrfs_root *root,
-                                       struct inode *inode,
-                                       u64 new_size, u32 min_type)
+int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
+                              struct btrfs_root *root,
+                              struct inode *inode,
+                              u64 new_size, u32 min_type)
 {
-       int ret;
        struct btrfs_path *path;
-       struct btrfs_key key;
-       struct btrfs_key found_key;
-       u32 found_type = (u8)-1;
        struct extent_buffer *leaf;
        struct btrfs_file_extent_item *fi;
+       struct btrfs_key key;
+       struct btrfs_key found_key;
        u64 extent_start = 0;
        u64 extent_num_bytes = 0;
        u64 extent_offset = 0;
        u64 item_end = 0;
+       u64 mask = root->sectorsize - 1;
+       u32 found_type = (u8)-1;
        int found_extent;
        int del_item;
        int pending_del_nr = 0;
        int pending_del_slot = 0;
        int extent_type = -1;
        int encoding;
-       u64 mask = root->sectorsize - 1;
+       int ret;
+       int err = 0;
+
+       BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
 
        if (root->ref_cows)
                btrfs_drop_extent_cache(inode, new_size & (~mask), (u64)-1, 0);
+
        path = btrfs_alloc_path();
        BUG_ON(!path);
        path->reada = -1;
 
-       /* FIXME, add redo link to tree so we don't leak on crash */
        key.objectid = inode->i_ino;
        key.offset = (u64)-1;
        key.type = (u8)-1;
@@ -2886,17 +2889,17 @@ noinline int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
 search_again:
        path->leave_spinning = 1;
        ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
-       if (ret < 0)
-               goto error;
+       if (ret < 0) {
+               err = ret;
+               goto out;
+       }
 
        if (ret > 0) {
                /* there are no items in the tree for us to truncate, we're
                 * done
                 */
-               if (path->slots[0] == 0) {
-                       ret = 0;
-                       goto error;
-               }
+               if (path->slots[0] == 0)
+                       goto out;
                path->slots[0]--;
        }
 
@@ -2931,28 +2934,17 @@ search_again:
                        }
                        item_end--;
                }
-               if (item_end < new_size) {
-                       if (found_type == BTRFS_DIR_ITEM_KEY)
-                               found_type = BTRFS_INODE_ITEM_KEY;
-                       else if (found_type == BTRFS_EXTENT_ITEM_KEY)
-                               found_type = BTRFS_EXTENT_DATA_KEY;
-                       else if (found_type == BTRFS_EXTENT_DATA_KEY)
-                               found_type = BTRFS_XATTR_ITEM_KEY;
-                       else if (found_type == BTRFS_XATTR_ITEM_KEY)
-                               found_type = BTRFS_INODE_REF_KEY;
-                       else if (found_type)
-                               found_type--;
-                       else
+               if (found_type > min_type) {
+                       del_item = 1;
+               } else {
+                       if (item_end < new_size)
                                break;
-                       btrfs_set_key_type(&key, found_type);
-                       goto next;
+                       if (found_key.offset >= new_size)
+                               del_item = 1;
+                       else
+                               del_item = 0;
                }
-               if (found_key.offset >= new_size)
-                       del_item = 1;
-               else
-                       del_item = 0;
                found_extent = 0;
-
                /* FIXME, shrink the extent if the ref count is only 1 */
                if (found_type != BTRFS_EXTENT_DATA_KEY)
                        goto delete;
@@ -3039,42 +3031,36 @@ delete:
                                                inode->i_ino, extent_offset);
                        BUG_ON(ret);
                }
-next:
-               if (path->slots[0] == 0) {
-                       if (pending_del_nr)
-                               goto del_pending;
-                       btrfs_release_path(root, path);
-                       if (found_type == BTRFS_INODE_ITEM_KEY)
-                               break;
-                       goto search_again;
-               }
 
-               path->slots[0]--;
-               if (pending_del_nr &&
-                   path->slots[0] + 1 != pending_del_slot) {
-                       struct btrfs_key debug;
-del_pending:
-                       btrfs_item_key_to_cpu(path->nodes[0], &debug,
-                                             pending_del_slot);
-                       ret = btrfs_del_items(trans, root, path,
-                                             pending_del_slot,
-                                             pending_del_nr);
-                       BUG_ON(ret);
-                       pending_del_nr = 0;
+               if (found_type == BTRFS_INODE_ITEM_KEY)
+                       break;
+
+               if (path->slots[0] == 0 ||
+                   path->slots[0] != pending_del_slot) {
+                       if (root->ref_cows) {
+                               err = -EAGAIN;
+                               goto out;
+                       }
+                       if (pending_del_nr) {
+                               ret = btrfs_del_items(trans, root, path,
+                                               pending_del_slot,
+                                               pending_del_nr);
+                               BUG_ON(ret);
+                               pending_del_nr = 0;
+                       }
                        btrfs_release_path(root, path);
-                       if (found_type == BTRFS_INODE_ITEM_KEY)
-                               break;
                        goto search_again;
+               } else {
+                       path->slots[0]--;
                }
        }
-       ret = 0;
-error:
+out:
        if (pending_del_nr) {
                ret = btrfs_del_items(trans, root, path, pending_del_slot,
                                      pending_del_nr);
        }
        btrfs_free_path(path);
-       return ret;
+       return err;
 }
 
 /*
@@ -3194,10 +3180,6 @@ int btrfs_cont_expand(struct inode *inode, loff_t size)
        if (size <= hole_start)
                return 0;
 
-       err = btrfs_truncate_page(inode->i_mapping, inode->i_size);
-       if (err)
-               return err;
-
        while (1) {
                struct btrfs_ordered_extent *ordered;
                btrfs_wait_ordered_range(inode, hole_start,
@@ -3210,9 +3192,6 @@ int btrfs_cont_expand(struct inode *inode, loff_t size)
                btrfs_put_ordered_extent(ordered);
        }
 
-       trans = btrfs_start_transaction(root, 1);
-       btrfs_set_trans_block_group(trans, inode);
-
        cur_offset = hole_start;
        while (1) {
                em = btrfs_get_extent(inode, NULL, 0, cur_offset,
@@ -3220,38 +3199,120 @@ int btrfs_cont_expand(struct inode *inode, loff_t size)
                BUG_ON(IS_ERR(em) || !em);
                last_byte = min(extent_map_end(em), block_end);
                last_byte = (last_byte + mask) & ~mask;
-               if (test_bit(EXTENT_FLAG_VACANCY, &em->flags)) {
+               if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
                        u64 hint_byte = 0;
                        hole_size = last_byte - cur_offset;
-                       err = btrfs_drop_extents(trans, inode, cur_offset,
-                                                cur_offset + hole_size,
-                                                &hint_byte, 1);
-                       if (err)
-                               break;
 
-                       err = btrfs_reserve_metadata_space(root, 1);
+                       err = btrfs_reserve_metadata_space(root, 2);
                        if (err)
                                break;
 
+                       trans = btrfs_start_transaction(root, 1);
+                       btrfs_set_trans_block_group(trans, inode);
+
+                       err = btrfs_drop_extents(trans, inode, cur_offset,
+                                                cur_offset + hole_size,
+                                                &hint_byte, 1);
+                       BUG_ON(err);
+
                        err = btrfs_insert_file_extent(trans, root,
                                        inode->i_ino, cur_offset, 0,
                                        0, hole_size, 0, hole_size,
                                        0, 0, 0);
+                       BUG_ON(err);
+
                        btrfs_drop_extent_cache(inode, hole_start,
                                        last_byte - 1, 0);
-                       btrfs_unreserve_metadata_space(root, 1);
+
+                       btrfs_end_transaction(trans, root);
+                       btrfs_unreserve_metadata_space(root, 2);
                }
                free_extent_map(em);
                cur_offset = last_byte;
-               if (err || cur_offset >= block_end)
+               if (cur_offset >= block_end)
                        break;
        }
 
-       btrfs_end_transaction(trans, root);
        unlock_extent(io_tree, hole_start, block_end - 1, GFP_NOFS);
        return err;
 }
 
+static int btrfs_setattr_size(struct inode *inode, struct iattr *attr)
+{
+       struct btrfs_root *root = BTRFS_I(inode)->root;
+       struct btrfs_trans_handle *trans;
+       unsigned long nr;
+       int ret;
+
+       if (attr->ia_size == inode->i_size)
+               return 0;
+
+       if (attr->ia_size > inode->i_size) {
+               unsigned long limit;
+               limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
+               if (attr->ia_size > inode->i_sb->s_maxbytes)
+                       return -EFBIG;
+               if (limit != RLIM_INFINITY && attr->ia_size > limit) {
+                       send_sig(SIGXFSZ, current, 0);
+                       return -EFBIG;
+               }
+       }
+
+       ret = btrfs_reserve_metadata_space(root, 1);
+       if (ret)
+               return ret;
+
+       trans = btrfs_start_transaction(root, 1);
+       btrfs_set_trans_block_group(trans, inode);
+
+       ret = btrfs_orphan_add(trans, inode);
+       BUG_ON(ret);
+
+       nr = trans->blocks_used;
+       btrfs_end_transaction(trans, root);
+       btrfs_unreserve_metadata_space(root, 1);
+       btrfs_btree_balance_dirty(root, nr);
+
+       if (attr->ia_size > inode->i_size) {
+               ret = btrfs_cont_expand(inode, attr->ia_size);
+               if (ret) {
+                       btrfs_truncate(inode);
+                       return ret;
+               }
+
+               i_size_write(inode, attr->ia_size);
+               btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
+
+               trans = btrfs_start_transaction(root, 1);
+               btrfs_set_trans_block_group(trans, inode);
+
+               ret = btrfs_update_inode(trans, root, inode);
+               BUG_ON(ret);
+               if (inode->i_nlink > 0) {
+                       ret = btrfs_orphan_del(trans, inode);
+                       BUG_ON(ret);
+               }
+               nr = trans->blocks_used;
+               btrfs_end_transaction(trans, root);
+               btrfs_btree_balance_dirty(root, nr);
+               return 0;
+       }
+
+       /*
+        * We're truncating a file that used to have good data down to
+        * zero. Make sure it gets into the ordered flush list so that
+        * any new writes get down to disk quickly.
+        */
+       if (attr->ia_size == 0)
+               BTRFS_I(inode)->ordered_data_close = 1;
+
+       /* we don't support swapfiles, so vmtruncate shouldn't fail */
+       ret = vmtruncate(inode, attr->ia_size);
+       BUG_ON(ret);
+
+       return 0;
+}
+
 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
 {
        struct inode *inode = dentry->d_inode;
@@ -3262,23 +3323,14 @@ static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
                return err;
 
        if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
-               if (attr->ia_size > inode->i_size) {
-                       err = btrfs_cont_expand(inode, attr->ia_size);
-                       if (err)
-                               return err;
-               } else if (inode->i_size > 0 &&
-                          attr->ia_size == 0) {
-
-                       /* we're truncating a file that used to have good
-                        * data down to zero.  Make sure it gets into
-                        * the ordered flush list so that any new writes
-                        * get down to disk quickly.
-                        */
-                       BTRFS_I(inode)->ordered_data_close = 1;
-               }
+               err = btrfs_setattr_size(inode, attr);
+               if (err)
+                       return err;
        }
+       attr->ia_valid &= ~ATTR_SIZE;
 
-       err = inode_setattr(inode, attr);
+       if (attr->ia_valid)
+               err = inode_setattr(inode, attr);
 
        if (!err && ((attr->ia_valid & ATTR_MODE)))
                err = btrfs_acl_chmod(inode);
@@ -3310,30 +3362,32 @@ void btrfs_delete_inode(struct inode *inode)
        }
 
        btrfs_i_size_write(inode, 0);
-       trans = btrfs_join_transaction(root, 1);
 
-       btrfs_set_trans_block_group(trans, inode);
-       ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size, 0);
-       if (ret) {
-               btrfs_orphan_del(NULL, inode);
-               goto no_delete_lock;
-       }
+       while (1) {
+               trans = btrfs_start_transaction(root, 1);
+               btrfs_set_trans_block_group(trans, inode);
+               ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
 
-       btrfs_orphan_del(trans, inode);
+               if (ret != -EAGAIN)
+                       break;
 
-       nr = trans->blocks_used;
-       clear_inode(inode);
+               nr = trans->blocks_used;
+               btrfs_end_transaction(trans, root);
+               trans = NULL;
+               btrfs_btree_balance_dirty(root, nr);
+       }
 
-       btrfs_end_transaction(trans, root);
-       btrfs_btree_balance_dirty(root, nr);
-       return;
+       if (ret == 0) {
+               ret = btrfs_orphan_del(trans, inode);
+               BUG_ON(ret);
+       }
 
-no_delete_lock:
        nr = trans->blocks_used;
        btrfs_end_transaction(trans, root);
        btrfs_btree_balance_dirty(root, nr);
 no_delete:
        clear_inode(inode);
+       return;
 }
 
 /*
@@ -5097,17 +5151,20 @@ static void btrfs_truncate(struct inode *inode)
        unsigned long nr;
        u64 mask = root->sectorsize - 1;
 
-       if (!S_ISREG(inode->i_mode))
-               return;
-       if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
+       if (!S_ISREG(inode->i_mode)) {
+               WARN_ON(1);
                return;
+       }
 
        ret = btrfs_truncate_page(inode->i_mapping, inode->i_size);
        if (ret)
                return;
+
        btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
+       btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
 
        trans = btrfs_start_transaction(root, 1);
+       btrfs_set_trans_block_group(trans, inode);
 
        /*
         * setattr is responsible for setting the ordered_data_close flag,
@@ -5129,21 +5186,32 @@ static void btrfs_truncate(struct inode *inode)
        if (inode->i_size == 0 && BTRFS_I(inode)->ordered_data_close)
                btrfs_add_ordered_operation(trans, root, inode);
 
-       btrfs_set_trans_block_group(trans, inode);
-       btrfs_i_size_write(inode, inode->i_size);
+       while (1) {
+               ret = btrfs_truncate_inode_items(trans, root, inode,
+                                                inode->i_size,
+                                                BTRFS_EXTENT_DATA_KEY);
+               if (ret != -EAGAIN)
+                       break;
 
-       ret = btrfs_orphan_add(trans, inode);
-       if (ret)
-               goto out;
-       /* FIXME, add redo link to tree so we don't leak on crash */
-       ret = btrfs_truncate_inode_items(trans, root, inode, inode->i_size,
-                                     BTRFS_EXTENT_DATA_KEY);
-       btrfs_update_inode(trans, root, inode);
+               ret = btrfs_update_inode(trans, root, inode);
+               BUG_ON(ret);
+
+               nr = trans->blocks_used;
+               btrfs_end_transaction(trans, root);
+               btrfs_btree_balance_dirty(root, nr);
+
+               trans = btrfs_start_transaction(root, 1);
+               btrfs_set_trans_block_group(trans, inode);
+       }
 
-       ret = btrfs_orphan_del(trans, inode);
+       if (ret == 0 && inode->i_nlink > 0) {
+               ret = btrfs_orphan_del(trans, inode);
+               BUG_ON(ret);
+       }
+
+       ret = btrfs_update_inode(trans, root, inode);
        BUG_ON(ret);
 
-out:
        nr = trans->blocks_used;
        ret = btrfs_end_transaction_throttle(trans, root);
        BUG_ON(ret);
@@ -5240,9 +5308,9 @@ void btrfs_destroy_inode(struct inode *inode)
 
        spin_lock(&root->list_lock);
        if (!list_empty(&BTRFS_I(inode)->i_orphan)) {
-               printk(KERN_ERR "BTRFS: inode %lu: inode still on the orphan"
-                      " list\n", inode->i_ino);
-               dump_stack();
+               printk(KERN_INFO "BTRFS: inode %lu still on the orphan list\n",
+                      inode->i_ino);
+               list_del_init(&BTRFS_I(inode)->i_orphan);
        }
        spin_unlock(&root->list_lock);
 
index 975fdd3..f2aa53d 100644 (file)
@@ -1561,6 +1561,20 @@ static int invalidate_extent_cache(struct btrfs_root *root,
        return 0;
 }
 
+static void put_inodes(struct list_head *list)
+{
+       struct inodevec *ivec;
+       while (!list_empty(list)) {
+               ivec = list_entry(list->next, struct inodevec, list);
+               list_del(&ivec->list);
+               while (ivec->nr > 0) {
+                       ivec->nr--;
+                       iput(ivec->inode[ivec->nr]);
+               }
+               kfree(ivec);
+       }
+}
+
 static int find_next_key(struct btrfs_path *path, int level,
                         struct btrfs_key *key)
 
@@ -1723,6 +1737,11 @@ static noinline_for_stack int merge_reloc_root(struct reloc_control *rc,
 
                btrfs_btree_balance_dirty(root, nr);
 
+               /*
+                * put inodes outside transaction, otherwise we may deadlock.
+                */
+               put_inodes(&inode_list);
+
                if (replaced && rc->stage == UPDATE_DATA_PTRS)
                        invalidate_extent_cache(root, &key, &next_key);
        }
@@ -1752,19 +1771,7 @@ out:
 
        btrfs_btree_balance_dirty(root, nr);
 
-       /*
-        * put inodes while we aren't holding the tree locks
-        */
-       while (!list_empty(&inode_list)) {
-               struct inodevec *ivec;
-               ivec = list_entry(inode_list.next, struct inodevec, list);
-               list_del(&ivec->list);
-               while (ivec->nr > 0) {
-                       ivec->nr--;
-                       iput(ivec->inode[ivec->nr]);
-               }
-               kfree(ivec);
-       }
+       put_inodes(&inode_list);
 
        if (replaced && rc->stage == UPDATE_DATA_PTRS)
                invalidate_extent_cache(root, &key, &next_key);