[PATCH 06/10] xfs: always allocate blocks as unwritten for file data

From: Christoph Hellwig <hch@lst.de>
To: linux-xfs@vger.kernel.org
Subject: [PATCH 06/10] xfs: always allocate blocks as unwritten for file data
Date: Mon, 17 Sep 2018 22:53:50 +0200	[thread overview]
Message-ID: <20180917205354.15401-7-hch@lst.de> (raw)
In-Reply-To: <20180917205354.15401-1-hch@lst.de>

XFS historically had a small race that could lead to exposing
uninitialized data in case of a crash.  If we are filling holes using
buffered I/O we convert the delayed allocation to a real allocation
before writing out the data.  If we crash after the blocks were
allocated, but before the data was written this could lead to reading
uninitialized blocks (or leaked data from a previous allocation that was
reused).  Now that we have the CIL logging extent format changes is
cheap, so we can switch to always allocating blocks as unwritten.
Note that this is not be strictly necessary for writes that append
beyond i_size, but given that we have to log a transaction in that
case anyway we might as well give all block allocations a uniform
treatment.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/xfs/xfs_aops.c  | 3 +--
 fs/xfs/xfs_aops.h  | 2 --
 fs/xfs/xfs_iomap.c | 4 ++--
 3 files changed, 3 insertions(+), 6 deletions(-)

diff --git a/fs/xfs/xfs_aops.c b/fs/xfs/xfs_aops.c
index 338b9d9984e0..775cdcfe70c2 100644
--- a/fs/xfs/xfs_aops.c
+++ b/fs/xfs/xfs_aops.c
@@ -437,8 +437,7 @@ xfs_map_blocks(
 			imap.br_blockcount = cow_fsb - imap.br_startoff;
 
 		if (isnullstartblock(imap.br_startblock)) {
-			/* got a delalloc extent */
-			wpc->io_type = XFS_IO_DELALLOC;
+			wpc->io_type = XFS_IO_UNWRITTEN;
 			goto allocate_blocks;
 		}
 
diff --git a/fs/xfs/xfs_aops.h b/fs/xfs/xfs_aops.h
index 494b4338446e..f0710c54cf68 100644
--- a/fs/xfs/xfs_aops.h
+++ b/fs/xfs/xfs_aops.h
@@ -13,7 +13,6 @@ extern struct bio_set xfs_ioend_bioset;
  */
 enum {
 	XFS_IO_HOLE,		/* covers region without any block allocation */
-	XFS_IO_DELALLOC,	/* covers delalloc region */
 	XFS_IO_UNWRITTEN,	/* covers allocated but uninitialized data */
 	XFS_IO_OVERWRITE,	/* covers already allocated extent */
 	XFS_IO_COW,		/* covers copy-on-write extent */
@@ -21,7 +20,6 @@ enum {
 
 #define XFS_IO_TYPES \
 	{ XFS_IO_HOLE,			"hole" },	\
-	{ XFS_IO_DELALLOC,		"delalloc" },	\
 	{ XFS_IO_UNWRITTEN,		"unwritten" },	\
 	{ XFS_IO_OVERWRITE,		"overwrite" },	\
 	{ XFS_IO_COW,			"CoW" }
diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
index 854b91080002..e12ff5e9a8ec 100644
--- a/fs/xfs/xfs_iomap.c
+++ b/fs/xfs/xfs_iomap.c
@@ -687,11 +687,11 @@ xfs_iomap_write_allocate(
 	xfs_trans_t	*tp;
 	int		nimaps;
 	int		error = 0;
-	int		flags = XFS_BMAPI_DELALLOC;
+	int		flags = XFS_BMAPI_DELALLOC | XFS_BMAPI_PREALLOC;
 	int		nres;
 
 	if (whichfork == XFS_COW_FORK)
-		flags |= XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC;
+		flags |= XFS_BMAPI_COWFORK;
 
 	/*
 	 * Make sure that the dquots are there.
-- 
2.18.0
