xfs: introduce the CoW fork

Introduce a new in-core fork for storing copy-on-write delalloc
reservations and allocated extents that are in the process of being
written out.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>

1 files changed, 114 insertions, 0 deletions

diff --git a/fs/xfs/xfs_reflink.c b/fs/xfs/xfs_reflink.c
new file mode 100644
index 000000000000..7adbb83df040
--- /dev/null
+++ b/fs/xfs/xfs_reflink.c
@@ -0,0 +1,114 @@
+/*
+ * Copyright (C) 2016 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <darrick.wong@oracle.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#include "xfs.h"
+#include "xfs_fs.h"
+#include "xfs_shared.h"
+#include "xfs_format.h"
+#include "xfs_log_format.h"
+#include "xfs_trans_resv.h"
+#include "xfs_mount.h"
+#include "xfs_defer.h"
+#include "xfs_da_format.h"
+#include "xfs_da_btree.h"
+#include "xfs_inode.h"
+#include "xfs_trans.h"
+#include "xfs_inode_item.h"
+#include "xfs_bmap.h"
+#include "xfs_bmap_util.h"
+#include "xfs_error.h"
+#include "xfs_dir2.h"
+#include "xfs_dir2_priv.h"
+#include "xfs_ioctl.h"
+#include "xfs_trace.h"
+#include "xfs_log.h"
+#include "xfs_icache.h"
+#include "xfs_pnfs.h"
+#include "xfs_refcount_btree.h"
+#include "xfs_refcount.h"
+#include "xfs_bmap_btree.h"
+#include "xfs_trans_space.h"
+#include "xfs_bit.h"
+#include "xfs_alloc.h"
+#include "xfs_quota_defs.h"
+#include "xfs_quota.h"
+#include "xfs_btree.h"
+#include "xfs_bmap_btree.h"
+#include "xfs_reflink.h"
+
+/*
+ * Copy on Write of Shared Blocks
+ *
+ * XFS must preserve "the usual" file semantics even when two files share
+ * the same physical blocks.  This means that a write to one file must not
+ * alter the blocks in a different file; the way that we'll do that is
+ * through the use of a copy-on-write mechanism.  At a high level, that
+ * means that when we want to write to a shared block, we allocate a new
+ * block, write the data to the new block, and if that succeeds we map the
+ * new block into the file.
+ *
+ * XFS provides a "delayed allocation" mechanism that defers the allocation
+ * of disk blocks to dirty-but-not-yet-mapped file blocks as long as
+ * possible.  This reduces fragmentation by enabling the filesystem to ask
+ * for bigger chunks less often, which is exactly what we want for CoW.
+ *
+ * The delalloc mechanism begins when the kernel wants to make a block
+ * writable (write_begin or page_mkwrite).  If the offset is not mapped, we
+ * create a delalloc mapping, which is a regular in-core extent, but without
+ * a real startblock.  (For delalloc mappings, the startblock encodes both
+ * a flag that this is a delalloc mapping, and a worst-case estimate of how
+ * many blocks might be required to put the mapping into the BMBT.)  delalloc
+ * mappings are a reservation against the free space in the filesystem;
+ * adjacent mappings can also be combined into fewer larger mappings.
+ *
+ * When dirty pages are being written out (typically in writepage), the
+ * delalloc reservations are converted into real mappings by allocating
+ * blocks and replacing the delalloc mapping with real ones.  A delalloc
+ * mapping can be replaced by several real ones if the free space is
+ * fragmented.
+ *
+ * We want to adapt the delalloc mechanism for copy-on-write, since the
+ * write paths are similar.  The first two steps (creating the reservation
+ * and allocating the blocks) are exactly the same as delalloc except that
+ * the mappings must be stored in a separate CoW fork because we do not want
+ * to disturb the mapping in the data fork until we're sure that the write
+ * succeeded.  IO completion in this case is the process of removing the old
+ * mapping from the data fork and moving the new mapping from the CoW fork to
+ * the data fork.  This will be discussed shortly.
+ *
+ * For now, unaligned directio writes will be bounced back to the page cache.
+ * Block-aligned directio writes will use the same mechanism as buffered
+ * writes.
+ *
+ * CoW remapping must be done after the data block write completes,
+ * because we don't want to destroy the old data fork map until we're sure
+ * the new block has been written.  Since the new mappings are kept in a
+ * separate fork, we can simply iterate these mappings to find the ones
+ * that cover the file blocks that we just CoW'd.  For each extent, simply
+ * unmap the corresponding range in the data fork, map the new range into
+ * the data fork, and remove the extent from the CoW fork.
+ *
+ * Since the remapping operation can be applied to an arbitrary file
+ * range, we record the need for the remap step as a flag in the ioend
+ * instead of declaring a new IO type.  This is required for direct io
+ * because we only have ioend for the whole dio, and we have to be able to
+ * remember the presence of unwritten blocks and CoW blocks with a single
+ * ioend structure.  Better yet, the more ground we can cover with one
+ * ioend, the better.
+ */