Btrfs: add better -ENOSPC handling

This is a step in the direction of better -ENOSPC handling.  Instead of
checking the global bytes counter we check the space_info bytes counters to
make sure we have enough space.

If we don't we go ahead and try to allocate a new chunk, and then if that fails
we return -ENOSPC.  This patch adds two counters to btrfs_space_info,
bytes_delalloc and bytes_may_use.

bytes_delalloc account for extents we've actually setup for delalloc and will
be allocated at some point down the line. 

bytes_may_use is to keep track of how many bytes we may use for delalloc at
some point.  When we actually set the extent_bit for the delalloc bytes we
subtract the reserved bytes from the bytes_may_use counter.  This keeps us from
not actually being able to allocate space for any delalloc bytes.

Signed-off-by: Josef Bacik <jbacik@redhat.com>

fs/btrfs/btrfs_inode.h

@@ -66,6 +66,9 @@ struct btrfs_inode {
 	 */
 	struct list_head delalloc_inodes;
 
+	/* the space_info for where this inode's data allocations are done */
+	struct btrfs_space_info *space_info;
+
 	/* full 64 bit generation number, struct vfs_inode doesn't have a big
 	 * enough field for this.
 	 */
@@ -94,6 +97,11 @@ struct btrfs_inode {
 	 */
 	u64 delalloc_bytes;
 
+	/* total number of bytes that may be used for this inode for
+	 * delalloc
+	 */
+	u64 reserved_bytes;
+
 	/*
 	 * the size of the file stored in the metadata on disk.  data=ordered
 	 * means the in-memory i_size might be larger than the size on disk

fs/btrfs/ctree.h

@@ -596,13 +596,27 @@ struct btrfs_block_group_item {
 
 struct btrfs_space_info {
 	u64 flags;
-	u64 total_bytes;
-	u64 bytes_used;
-	u64 bytes_pinned;
-	u64 bytes_reserved;
-	u64 bytes_readonly;
-	int full;
-	int force_alloc;
+
+	u64 total_bytes;	/* total bytes in the space */
+	u64 bytes_used;		/* total bytes used on disk */
+	u64 bytes_pinned;	/* total bytes pinned, will be freed when the
+				   transaction finishes */
+	u64 bytes_reserved;	/* total bytes the allocator has reserved for
+				   current allocations */
+	u64 bytes_readonly;	/* total bytes that are read only */
+
+	/* delalloc accounting */
+	u64 bytes_delalloc;	/* number of bytes reserved for allocation,
+				   this space is not necessarily reserved yet
+				   by the allocator */
+	u64 bytes_may_use;	/* number of bytes that may be used for
+				   delalloc */
+
+	int full;		/* indicates that we cannot allocate any more
+				   chunks for this space */
+	int force_alloc;	/* set if we need to force a chunk alloc for
+				   this space */
+
 	struct list_head list;
 
 	/* for block groups in our same type */
@@ -1782,6 +1796,16 @@ int btrfs_add_dead_reloc_root(struct btrfs_root *root);
 int btrfs_cleanup_reloc_trees(struct btrfs_root *root);
 int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len);
 u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags);
+void btrfs_set_inode_space_info(struct btrfs_root *root, struct inode *ionde);
+int btrfs_check_metadata_free_space(struct btrfs_root *root);
+int btrfs_check_data_free_space(struct btrfs_root *root, struct inode *inode,
+				u64 bytes);
+void btrfs_free_reserved_data_space(struct btrfs_root *root,
+				    struct inode *inode, u64 bytes);
+void btrfs_delalloc_reserve_space(struct btrfs_root *root, struct inode *inode,
+				 u64 bytes);
+void btrfs_delalloc_free_space(struct btrfs_root *root, struct inode *inode,
+			      u64 bytes);
 /* ctree.c */
 int btrfs_previous_item(struct btrfs_root *root,
 			struct btrfs_path *path, u64 min_objectid,
@@ -2027,8 +2051,6 @@ int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
 unsigned long btrfs_force_ra(struct address_space *mapping,
 			      struct file_ra_state *ra, struct file *file,
 			      pgoff_t offset, pgoff_t last_index);
-int btrfs_check_free_space(struct btrfs_root *root, u64 num_required,
-			   int for_del);
 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct page *page);
 int btrfs_readpage(struct file *file, struct page *page);
 void btrfs_delete_inode(struct inode *inode);

fs/btrfs/extent-tree.c

@@ -60,6 +60,10 @@ static int update_block_group(struct btrfs_trans_handle *trans,
 			      u64 bytenr, u64 num_bytes, int alloc,
 			      int mark_free);
 
+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)
 {
 	return (cache->flags & bits) == bits;
@@ -1909,6 +1913,7 @@ 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;
@@ -1972,6 +1977,196 @@ u64 btrfs_reduce_alloc_profile(struct btrfs_root *root, u64 flags)
 	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;
+
+	/* 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);
+
+	/*
+	 * if the metadata area isn't maxed out then there is no sense in
+	 * checking how much is used, since we can always allocate a new chunk
+	 */
+	if (!meta_sinfo->full)
+		return 0;
+
+	spin_lock(&meta_sinfo->lock);
+	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) {
+		spin_unlock(&meta_sinfo->lock);
+		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;
+
+	/* 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) {
+		/*
+		 * 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;
+			struct btrfs_trans_handle *trans;
+
+			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);
+		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)
+{
+	struct btrfs_space_info *data_sinfo;
+
+	/* make sure bytes are sectorsize aligned */
+	bytes = (bytes + root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
+
+	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);
+}
+
+/* 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;
+	}
+
+	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,
 			  struct btrfs_root *extent_root, u64 alloc_bytes,
 			  u64 flags, int force)
@@ -3105,6 +3300,10 @@ 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 +3330,10 @@ 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

fs/btrfs/file.c

@@ -1091,28 +1091,36 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 		WARN_ON(num_pages > nrptrs);
 		memset(pages, 0, sizeof(struct page *) * nrptrs);
 
-		ret = btrfs_check_free_space(root, write_bytes, 0);
+		ret = btrfs_check_data_free_space(root, inode, write_bytes);
 		if (ret)
 			goto out;
 
 		ret = prepare_pages(root, file, pages, num_pages,
 				    pos, first_index, last_index,
 				    write_bytes);
-		if (ret)
+		if (ret) {
+			btrfs_free_reserved_data_space(root, inode,
+						       write_bytes);
 			goto out;
+		}
 
 		ret = btrfs_copy_from_user(pos, num_pages,
 					   write_bytes, pages, buf);
 		if (ret) {
+			btrfs_free_reserved_data_space(root, inode,
+						       write_bytes);
 			btrfs_drop_pages(pages, num_pages);
 			goto out;
 		}
 
 		ret = dirty_and_release_pages(NULL, root, file, pages,
 					      num_pages, pos, write_bytes);
 		btrfs_drop_pages(pages, num_pages);
-		if (ret)
+		if (ret) {
+			btrfs_free_reserved_data_space(root, inode,
+						       write_bytes);
 			goto out;
+		}
 
 		if (will_write) {
 			btrfs_fdatawrite_range(inode->i_mapping, pos,
@@ -1136,6 +1144,8 @@ static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 	}
 out:
 	mutex_unlock(&inode->i_mutex);
+	if (ret)
+		err = ret;
 
 out_nolock:
 	kfree(pages);