Linux-Fsdevel Archive on lore.kernel.org
 help / color / Atom feed
From: Christoph Hellwig <hch@lst.de>
To: xfs@oss.sgi.com
Cc: linux-nvdimm@ml01.01.org, linux-fsdevel@vger.kernel.org
Subject: [PATCH 7/8] xfs: split direct I/O and DAX path
Date: Wed, 22 Jun 2016 17:27:15 +0200
Message-ID: <1466609236-23801-8-git-send-email-hch@lst.de> (raw)
In-Reply-To: <1466609236-23801-1-git-send-email-hch@lst.de>

So far the DAX code overloaded the direct I/O code path.  There is very little
in common between the two, and untangling them allows to clean up both variants.

As a ѕide effect we also get separate trace points for both I/O types.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 fs/xfs/xfs_file.c  | 139 ++++++++++++++++++++++++++++++++++++++++++-----------
 fs/xfs/xfs_trace.h |   2 +
 2 files changed, 112 insertions(+), 29 deletions(-)

diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
index 24b267d..0e74325 100644
--- a/fs/xfs/xfs_file.c
+++ b/fs/xfs/xfs_file.c
@@ -262,13 +262,11 @@ xfs_file_dio_aio_read(
 	else
 		target = ip->i_mount->m_ddev_targp;
 
-	if (!IS_DAX(inode)) {
-		/* DIO must be aligned to device logical sector size */
-		if ((iocb->ki_pos | count) & target->bt_logical_sectormask) {
-			if (iocb->ki_pos == isize)
-				return 0;
-			return -EINVAL;
-		}
+	/* DIO must be aligned to device logical sector size */
+	if ((iocb->ki_pos | count) & target->bt_logical_sectormask) {
+		if (iocb->ki_pos == isize)
+			return 0;
+		return -EINVAL;
 	}
 
 	/*
@@ -317,13 +315,37 @@ xfs_file_dio_aio_read(
 	}
 
 	data = *to;
-	if (IS_DAX(inode)) {
-		ret = dax_do_io(iocb, inode, &data, xfs_get_blocks_direct,
-				NULL, 0);
-	} else {
-		ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data,
-				xfs_get_blocks_direct, NULL, NULL, 0);
+	ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data,
+			xfs_get_blocks_direct, NULL, NULL, 0);
+	if (ret > 0) {
+		iocb->ki_pos += ret;
+		iov_iter_advance(to, ret);
 	}
+	xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
+
+	file_accessed(iocb->ki_filp);
+	return ret;
+}
+
+STATIC ssize_t
+xfs_file_dax_read(
+	struct kiocb		*iocb,
+	struct iov_iter		*to)
+{
+	struct address_space	*mapping = iocb->ki_filp->f_mapping;
+	struct inode		*inode = mapping->host;
+	struct xfs_inode	*ip = XFS_I(inode);
+	struct iov_iter		data = *to;
+	size_t			count = iov_iter_count(to);
+	ssize_t			ret = 0;
+
+	trace_xfs_file_dax_read(ip, count, iocb->ki_pos);
+
+	if (!count)
+		return 0; /* skip atime */
+
+	xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
+	ret = dax_do_io(iocb, inode, &data, xfs_get_blocks_direct, NULL, 0);
 	if (ret > 0) {
 		iocb->ki_pos += ret;
 		iov_iter_advance(to, ret);
@@ -356,7 +378,8 @@ xfs_file_read_iter(
 	struct kiocb		*iocb,
 	struct iov_iter		*to)
 {
-	struct xfs_mount	*mp = XFS_I(file_inode(iocb->ki_filp))->i_mount;
+	struct inode		*inode = file_inode(iocb->ki_filp);
+	struct xfs_mount	*mp = XFS_I(inode)->i_mount;
 	ssize_t			ret = 0;
 
 	XFS_STATS_INC(mp, xs_read_calls);
@@ -364,7 +387,9 @@ xfs_file_read_iter(
 	if (XFS_FORCED_SHUTDOWN(mp))
 		return -EIO;
 
-	if (iocb->ki_flags & IOCB_DIRECT)
+	if (IS_DAX(inode))
+		ret = xfs_file_dax_read(iocb, to);
+	else if (iocb->ki_flags & IOCB_DIRECT)
 		ret = xfs_file_dio_aio_read(iocb, to);
 	else
 		ret = xfs_file_buffered_aio_read(iocb, to);
@@ -586,8 +611,7 @@ xfs_file_dio_aio_write(
 					mp->m_rtdev_targp : mp->m_ddev_targp;
 
 	/* DIO must be aligned to device logical sector size */
-	if (!IS_DAX(inode) &&
-	    ((iocb->ki_pos | count) & target->bt_logical_sectormask))
+	if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
 		return -EINVAL;
 
 	/* "unaligned" here means not aligned to a filesystem block */
@@ -656,14 +680,9 @@ xfs_file_dio_aio_write(
 	trace_xfs_file_direct_write(ip, count, iocb->ki_pos);
 
 	data = *from;
-	if (IS_DAX(inode)) {
-		ret = dax_do_io(iocb, inode, &data, xfs_get_blocks_direct,
-				xfs_end_io_direct_write, 0);
-	} else {
-		ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data,
-				xfs_get_blocks_direct, xfs_end_io_direct_write,
-				NULL, DIO_ASYNC_EXTEND);
-	}
+	ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data,
+			xfs_get_blocks_direct, xfs_end_io_direct_write,
+			NULL, DIO_ASYNC_EXTEND);
 
 	/* see generic_file_direct_write() for why this is necessary */
 	if (mapping->nrpages) {
@@ -680,10 +699,70 @@ out:
 	xfs_rw_iunlock(ip, iolock);
 
 	/*
-	 * No fallback to buffered IO on errors for XFS. DAX can result in
-	 * partial writes, but direct IO will either complete fully or fail.
+	 * No fallback to buffered IO on errors for XFS, direct IO will either
+	 * complete fully or fail.
+	 */
+	ASSERT(ret < 0 || ret == count);
+	return ret;
+}
+
+STATIC ssize_t
+xfs_file_dax_write(
+	struct kiocb		*iocb,
+	struct iov_iter		*from)
+{
+	struct address_space	*mapping = iocb->ki_filp->f_mapping;
+	struct inode		*inode = mapping->host;
+	struct xfs_inode	*ip = XFS_I(inode);
+	struct xfs_mount	*mp = ip->i_mount;
+	ssize_t			ret = 0;
+	int			unaligned_io = 0;
+	int			iolock;
+	struct iov_iter		data;
+
+	/* "unaligned" here means not aligned to a filesystem block */
+	if ((iocb->ki_pos & mp->m_blockmask) ||
+	    ((iocb->ki_pos + iov_iter_count(from)) & mp->m_blockmask)) {
+		unaligned_io = 1;
+		iolock = XFS_IOLOCK_EXCL;
+	} else if (mapping->nrpages) {
+		iolock = XFS_IOLOCK_EXCL;
+	} else {
+		iolock = XFS_IOLOCK_SHARED;
+	}
+	xfs_rw_ilock(ip, iolock);
+
+	ret = xfs_file_aio_write_checks(iocb, from, &iolock);
+	if (ret)
+		goto out;
+
+	/*
+	 * Yes, even DAX files can have page cache attached to them:  A zeroed
+	 * page is inserted into the pagecache when we have to serve a write
+	 * fault on a hole.  It should never be dirtied and can simply be
+	 * dropped from the pagecache once we get real data for the page.
 	 */
-	ASSERT(ret < 0 || ret == count || IS_DAX(VFS_I(ip)));
+	if (mapping->nrpages) {
+		ret = invalidate_inode_pages2(mapping);
+		WARN_ON_ONCE(ret);
+	}
+
+	if (iolock == XFS_IOLOCK_EXCL && !unaligned_io) {
+		xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
+		iolock = XFS_IOLOCK_SHARED;
+	}
+
+	trace_xfs_file_dax_write(ip, iov_iter_count(from), iocb->ki_pos);
+
+	data = *from;
+	ret = dax_do_io(iocb, inode, &data, xfs_get_blocks_direct,
+			xfs_end_io_direct_write, 0);
+	if (ret > 0) {
+		iocb->ki_pos += ret;
+		iov_iter_advance(from, ret);
+	}
+out:
+	xfs_rw_iunlock(ip, iolock);
 	return ret;
 }
 
@@ -765,7 +844,9 @@ xfs_file_write_iter(
 	if (XFS_FORCED_SHUTDOWN(ip->i_mount))
 		return -EIO;
 
-	if ((iocb->ki_flags & IOCB_DIRECT) || IS_DAX(inode))
+	if (IS_DAX(inode))
+		ret = xfs_file_dax_write(iocb, from);
+	else if (iocb->ki_flags & IOCB_DIRECT)
 		ret = xfs_file_dio_aio_write(iocb, from);
 	else
 		ret = xfs_file_buffered_aio_write(iocb, from);
diff --git a/fs/xfs/xfs_trace.h b/fs/xfs/xfs_trace.h
index 2504f94..1451690 100644
--- a/fs/xfs/xfs_trace.h
+++ b/fs/xfs/xfs_trace.h
@@ -1165,8 +1165,10 @@ DEFINE_EVENT(xfs_file_class, name,	\
 	TP_ARGS(ip, count, offset))
 DEFINE_RW_EVENT(xfs_file_buffered_read);
 DEFINE_RW_EVENT(xfs_file_direct_read);
+DEFINE_RW_EVENT(xfs_file_dax_read);
 DEFINE_RW_EVENT(xfs_file_buffered_write);
 DEFINE_RW_EVENT(xfs_file_direct_write);
+DEFINE_RW_EVENT(xfs_file_dax_write);
 DEFINE_RW_EVENT(xfs_file_splice_read);
 
 DECLARE_EVENT_CLASS(xfs_page_class,
-- 
2.1.4


  parent reply index

Thread overview: 27+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-06-22 15:27 xfs: untangle the direct I/O and DAX path, fix DAX locking Christoph Hellwig
2016-06-22 15:27 ` [PATCH 1/8] xfs: don't pass ioflags around in the ioctl path Christoph Hellwig
2016-06-22 15:27 ` [PATCH 2/8] xfs: kill ioflags Christoph Hellwig
2016-06-22 15:27 ` [PATCH 3/8] xfs: remove s_maxbytes enforcement in xfs_file_read_iter Christoph Hellwig
2016-06-22 15:27 ` [PATCH 4/8] xfs: split xfs_file_read_iter into buffered and direct I/O helpers Christoph Hellwig
2016-06-22 15:27 ` [PATCH 5/8] xfs: stop using generic_file_read_iter for direct I/O Christoph Hellwig
2016-06-22 15:27 ` [PATCH 6/8] xfs: direct calls in the direct I/O path Christoph Hellwig
2016-06-22 15:27 ` Christoph Hellwig [this message]
2016-09-29  2:53   ` [PATCH 7/8] xfs: split direct I/O and DAX path Darrick J. Wong
2016-09-29  8:38     ` aio completions vs file_accessed race, was: " Christoph Hellwig
2016-09-29 20:18       ` Christoph Hellwig
2016-09-29 20:18         ` Christoph Hellwig
2016-09-29 20:33           ` Darrick J. Wong
2016-06-22 15:27 ` [PATCH 8/8] xfs: fix locking for DAX writes Christoph Hellwig
2016-06-23 14:22   ` Boaz Harrosh
2016-06-23 23:24 ` xfs: untangle the direct I/O and DAX path, fix DAX locking Dave Chinner
2016-06-24  1:14   ` Dan Williams
2016-06-24  7:13     ` Dave Chinner
2016-06-24  7:31       ` Christoph Hellwig
2016-06-24  7:26   ` Christoph Hellwig
2016-06-24 23:00     ` Dave Chinner
2016-06-28 13:10       ` Christoph Hellwig
2016-06-28 13:27         ` Boaz Harrosh
2016-06-28 13:39           ` Christoph Hellwig
2016-06-28 13:56             ` Boaz Harrosh
2016-06-28 15:39               ` Christoph Hellwig
2016-06-29 12:23                 ` Boaz Harrosh

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1466609236-23801-8-git-send-email-hch@lst.de \
    --to=hch@lst.de \
    --cc=linux-fsdevel@vger.kernel.org \
    --cc=linux-nvdimm@ml01.01.org \
    --cc=xfs@oss.sgi.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

Linux-Fsdevel Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/linux-fsdevel/0 linux-fsdevel/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 linux-fsdevel linux-fsdevel/ https://lore.kernel.org/linux-fsdevel \
		linux-fsdevel@vger.kernel.org
	public-inbox-index linux-fsdevel

Example config snippet for mirrors

Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.kernel.vger.linux-fsdevel


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git