[PATCH v9 0/5] gfs2 iomap write support

[PATCH v9 0/5] gfs2 iomap write support

[Andreas Gruenbacher]

Here's another update of my gfs2 iomap write patches on top of:

  git://git.infradead.org/users/hch/xfs.git gfs2-iomap


Changes since v8:

 * Rebase on top of Christoph's branch.

 * New patch to add private pointer to struct iomap. 


Changes since v7:

 * Drop patch that allows to unconditionally mark the inode dirty in
   __generic_write_end and mark the inode dirty in
   iomap_write_inline_data instead.

 * Call the page_write_end hook even for IOMAP_INLINE mappings.

 * Drop patch "iomap: Put struct iomap_ops into struct iomap".


Significant changes since v6:

 * Switch to Christoph's patch for splitting out __generic_write_end and 
   put the inode-dirtying change in a separate commit.

 * Switch froma mandatory write_end operation to an optional
   page_write_end hook.

 * Add a patch to pass struct iomap_ops in struct iomap at the end for
   demonstration purposes.


Significant changes since v5:

 * Initial gfs2 specific cleanups split off and posted separately.

 * New patch for generic iomap inline data handling.

 * Iomap write_begin operation removed.


These patches can be found here:

  https://git.kernel.org/pub/scm/linux/kernel/git/gfs2/linux-gfs2.git/log/?h=iomap-write


Thanks,
Andreas

Andreas Gruenbacher (5):
  iomap: add private pointer to struct iomap
  gfs2: iomap buffered write support
  gfs2: gfs2_extent_length cleanup
  gfs2: iomap direct I/O support
  gfs2: Remove gfs2_write_{begin,end}

 fs/gfs2/aops.c        | 331 ++-------------------------------------
 fs/gfs2/aops.h        |  22 +++
 fs/gfs2/bmap.c        | 352 ++++++++++++++++++++++++++++++++++++------
 fs/gfs2/file.c        | 168 ++++++++++++++++++--
 include/linux/iomap.h |   1 +
 5 files changed, 500 insertions(+), 374 deletions(-)
 create mode 100644 fs/gfs2/aops.h

-- 
2.17.1

[Cluster-devel] [PATCH v9 0/5] gfs2 iomap write support

[Andreas Gruenbacher]

Here's another update of my gfs2 iomap write patches on top of:

  git://git.infradead.org/users/hch/xfs.git gfs2-iomap


Changes since v8:

 * Rebase on top of Christoph's branch.

 * New patch to add private pointer to struct iomap. 


Changes since v7:

 * Drop patch that allows to unconditionally mark the inode dirty in
   __generic_write_end and mark the inode dirty in
   iomap_write_inline_data instead.

 * Call the page_write_end hook even for IOMAP_INLINE mappings.

 * Drop patch "iomap: Put struct iomap_ops into struct iomap".


Significant changes since v6:

 * Switch to Christoph's patch for splitting out __generic_write_end and 
   put the inode-dirtying change in a separate commit.

 * Switch froma mandatory write_end operation to an optional
   page_write_end hook.

 * Add a patch to pass struct iomap_ops in struct iomap at the end for
   demonstration purposes.


Significant changes since v5:

 * Initial gfs2 specific cleanups split off and posted separately.

 * New patch for generic iomap inline data handling.

 * Iomap write_begin operation removed.


These patches can be found here:

  https://git.kernel.org/pub/scm/linux/kernel/git/gfs2/linux-gfs2.git/log/?h=iomap-write


Thanks,
Andreas

Andreas Gruenbacher (5):
  iomap: add private pointer to struct iomap
  gfs2: iomap buffered write support
  gfs2: gfs2_extent_length cleanup
  gfs2: iomap direct I/O support
  gfs2: Remove gfs2_write_{begin,end}

 fs/gfs2/aops.c        | 331 ++-------------------------------------
 fs/gfs2/aops.h        |  22 +++
 fs/gfs2/bmap.c        | 352 ++++++++++++++++++++++++++++++++++++------
 fs/gfs2/file.c        | 168 ++++++++++++++++++--
 include/linux/iomap.h |   1 +
 5 files changed, 500 insertions(+), 374 deletions(-)
 create mode 100644 fs/gfs2/aops.h

-- 
2.17.1

[PATCH v9 1/5] iomap: add private pointer to struct iomap

[Andreas Gruenbacher]

Add a private pointer to struct iomap to allow filesystems to pass data
from iomap_begin to iomap_end.  Will be used by gfs2 for passing on the
on-disk inode buffer head.

Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
---
 include/linux/iomap.h | 1 +
 1 file changed, 1 insertion(+)

diff --git a/include/linux/iomap.h b/include/linux/iomap.h
index 43cc9208599f..9962dcb836b9 100644
--- a/include/linux/iomap.h
+++ b/include/linux/iomap.h
@@ -59,6 +59,7 @@ struct iomap {
 		struct dax_device   *dax_dev;
 		void		    *inline_data;
 	};
+	void			*private; /* filesystem private */
 
 	/*
 	 * Called when finished processing a page in the mapping returned in
-- 
2.17.1

[Cluster-devel] [PATCH v9 1/5] iomap: add private pointer to struct iomap

[Andreas Gruenbacher]

Add a private pointer to struct iomap to allow filesystems to pass data
from iomap_begin to iomap_end.  Will be used by gfs2 for passing on the
on-disk inode buffer head.

Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
---
 include/linux/iomap.h | 1 +
 1 file changed, 1 insertion(+)

diff --git a/include/linux/iomap.h b/include/linux/iomap.h
index 43cc9208599f..9962dcb836b9 100644
--- a/include/linux/iomap.h
+++ b/include/linux/iomap.h
@@ -59,6 +59,7 @@ struct iomap {
 		struct dax_device   *dax_dev;
 		void		    *inline_data;
 	};
+	void			*private; /* filesystem private */
 
 	/*
 	 * Called when finished processing a page in the mapping returned in
-- 
2.17.1

Re: [PATCH v9 1/5] iomap: add private pointer to struct iomap

[Andreas Gruenbacher]

Hi Christoph,

On 15 June 2018 at 14:19, Andreas Gruenbacher <agruenba@redhat.com> wrote:
> Add a private pointer to struct iomap to allow filesystems to pass data
> from iomap_begin to iomap_end.  Will be used by gfs2 for passing on the
> on-disk inode buffer head.
>
> Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
> ---
>  include/linux/iomap.h | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/include/linux/iomap.h b/include/linux/iomap.h
> index 43cc9208599f..9962dcb836b9 100644
> --- a/include/linux/iomap.h
> +++ b/include/linux/iomap.h
> @@ -59,6 +59,7 @@ struct iomap {
>                 struct dax_device   *dax_dev;
>                 void                *inline_data;
>         };
> +       void                    *private; /* filesystem private */
>
>         /*
>          * Called when finished processing a page in the mapping returned in
> --
> 2.17.1

any objections? If not, could you please add this to the shared branch?

Thanks,
Andreas

[Cluster-devel] [PATCH v9 1/5] iomap: add private pointer to struct iomap

[Andreas Gruenbacher]

Hi Christoph,

On 15 June 2018 at 14:19, Andreas Gruenbacher <agruenba@redhat.com> wrote:
> Add a private pointer to struct iomap to allow filesystems to pass data
> from iomap_begin to iomap_end.  Will be used by gfs2 for passing on the
> on-disk inode buffer head.
>
> Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
> ---
>  include/linux/iomap.h | 1 +
>  1 file changed, 1 insertion(+)
>
> diff --git a/include/linux/iomap.h b/include/linux/iomap.h
> index 43cc9208599f..9962dcb836b9 100644
> --- a/include/linux/iomap.h
> +++ b/include/linux/iomap.h
> @@ -59,6 +59,7 @@ struct iomap {
>                 struct dax_device   *dax_dev;
>                 void                *inline_data;
>         };
> +       void                    *private; /* filesystem private */
>
>         /*
>          * Called when finished processing a page in the mapping returned in
> --
> 2.17.1

any objections? If not, could you please add this to the shared branch?

Thanks,
Andreas

Re: [PATCH v9 1/5] iomap: add private pointer to struct iomap

[Christoph Hellwig]

On Fri, Jun 15, 2018 at 02:19:18PM +0200, Andreas Gruenbacher wrote:
> Add a private pointer to struct iomap to allow filesystems to pass data
> from iomap_begin to iomap_end.  Will be used by gfs2 for passing on the
> on-disk inode buffer head.
> 
> Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>

Looks good,

Reviewed-by: Christoph Hellwig <hch@lst.de>

[Cluster-devel] [PATCH v9 1/5] iomap: add private pointer to struct iomap

[Christoph Hellwig]

On Fri, Jun 15, 2018 at 02:19:18PM +0200, Andreas Gruenbacher wrote:
> Add a private pointer to struct iomap to allow filesystems to pass data
> from iomap_begin to iomap_end.  Will be used by gfs2 for passing on the
> on-disk inode buffer head.
> 
> Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>

Looks good,

Reviewed-by: Christoph Hellwig <hch@lst.de>

[PATCH v9 2/5] gfs2: iomap buffered write support

[Andreas Gruenbacher]

With the traditional page-based writes, blocks are allocated separately
for each page written to.  With iomap writes, we can allocate a lot more
blocks at once, with a fraction of the allocation overhead for each
page.

Split calculating the number of blocks that can be allocated at a given
position (gfs2_alloc_size) off from gfs2_iomap_alloc: that size
determines the number of blocks to allocate and reserve in the journal.

In gfs2_iomap_alloc, set the type of newly allocated extents to
IOMAP_MAPPED so that iomap_to_bh will set the bh states correctly:
otherwise, the bhs would not be marked as mapped, confusing
__mpage_writepage.  This means that we need to check for the IOMAP_F_NEW
flag in fallocate_chunk now.

Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
---
 fs/gfs2/aops.c |  20 ++--
 fs/gfs2/aops.h |  22 ++++
 fs/gfs2/bmap.c | 317 ++++++++++++++++++++++++++++++++++++++++++++-----
 fs/gfs2/file.c |  46 +++++--
 4 files changed, 359 insertions(+), 46 deletions(-)
 create mode 100644 fs/gfs2/aops.h

diff --git a/fs/gfs2/aops.c b/fs/gfs2/aops.c
index 35f5ee23566d..ecfbca9c88ff 100644
--- a/fs/gfs2/aops.c
+++ b/fs/gfs2/aops.c
@@ -22,6 +22,7 @@
 #include <linux/backing-dev.h>
 #include <linux/uio.h>
 #include <trace/events/writeback.h>
+#include <linux/sched/signal.h>
 
 #include "gfs2.h"
 #include "incore.h"
@@ -36,10 +37,11 @@
 #include "super.h"
 #include "util.h"
 #include "glops.h"
+#include "aops.h"
 
 
-static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
-				   unsigned int from, unsigned int len)
+void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
+			    unsigned int from, unsigned int len)
 {
 	struct buffer_head *head = page_buffers(page);
 	unsigned int bsize = head->b_size;
@@ -462,7 +464,7 @@ static int gfs2_jdata_writepages(struct address_space *mapping,
  * Returns: errno
  */
 
-static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
+int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
 {
 	struct buffer_head *dibh;
 	u64 dsize = i_size_read(&ip->i_inode);
@@ -776,7 +778,7 @@ static int gfs2_write_begin(struct file *file, struct address_space *mapping,
  * adjust_fs_space - Adjusts the free space available due to gfs2_grow
  * @inode: the rindex inode
  */
-static void adjust_fs_space(struct inode *inode)
+void adjust_fs_space(struct inode *inode)
 {
 	struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
@@ -822,11 +824,11 @@ static void adjust_fs_space(struct inode *inode)
  * This copies the data from the page into the inode block after
  * the inode data structure itself.
  *
- * Returns: errno
+ * Returns: copied bytes or errno
  */
-static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
-				  loff_t pos, unsigned copied,
-				  struct page *page)
+int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
+			   loff_t pos, unsigned copied,
+			   struct page *page)
 {
 	struct gfs2_inode *ip = GFS2_I(inode);
 	u64 to = pos + copied;
@@ -865,7 +867,7 @@ static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
  * The main write_end function for GFS2. We just put our locking around the VFS
  * provided functions.
  *
- * Returns: errno
+ * Returns: copied bytes or errno
  */
 
 static int gfs2_write_end(struct file *file, struct address_space *mapping,
diff --git a/fs/gfs2/aops.h b/fs/gfs2/aops.h
new file mode 100644
index 000000000000..976bb32dd405
--- /dev/null
+++ b/fs/gfs2/aops.h
@@ -0,0 +1,22 @@
+/*
+ * Copyright (C) 2017 Red Hat, Inc.  All rights reserved.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License version 2.
+ */
+
+#ifndef __AOPS_DOT_H__
+#define __AOPS_DOT_H__
+
+#include "incore.h"
+
+extern int stuffed_readpage(struct gfs2_inode *ip, struct page *page);
+extern int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
+				  loff_t pos, unsigned copied,
+				  struct page *page);
+extern void adjust_fs_space(struct inode *inode);
+extern void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
+				   unsigned int from, unsigned int len);
+
+#endif /* __AOPS_DOT_H__ */
diff --git a/fs/gfs2/bmap.c b/fs/gfs2/bmap.c
index ed6699705c13..704a9150dcbc 100644
--- a/fs/gfs2/bmap.c
+++ b/fs/gfs2/bmap.c
@@ -28,6 +28,7 @@
 #include "trans.h"
 #include "dir.h"
 #include "util.h"
+#include "aops.h"
 #include "trace_gfs2.h"
 
 /* This doesn't need to be that large as max 64 bit pointers in a 4k
@@ -41,6 +42,8 @@ struct metapath {
 	int mp_aheight; /* actual height (lookup height) */
 };
 
+static int punch_hole(struct gfs2_inode *ip, u64 offset, u64 length);
+
 /**
  * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page
  * @ip: the inode
@@ -389,7 +392,7 @@ static int fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, int h)
 	return mp->mp_aheight - x - 1;
 }
 
-static inline void release_metapath(struct metapath *mp)
+static void release_metapath(struct metapath *mp)
 {
 	int i;
 
@@ -397,6 +400,7 @@ static inline void release_metapath(struct metapath *mp)
 		if (mp->mp_bh[i] == NULL)
 			break;
 		brelse(mp->mp_bh[i]);
+		mp->mp_bh[i] = NULL;
 	}
 }
 
@@ -609,11 +613,13 @@ enum alloc_state {
  *  ii) Indirect blocks to fill in lower part of the metadata tree
  * iii) Data blocks
  *
- * The function is in two parts. The first part works out the total
- * number of blocks which we need. The second part does the actual
- * allocation asking for an extent at a time (if enough contiguous free
- * blocks are available, there will only be one request per bmap call)
- * and uses the state machine to initialise the blocks in order.
+ * This function is called after gfs2_iomap_get, which works out the
+ * total number of blocks which we need via gfs2_alloc_size.
+ *
+ * We then do the actual allocation asking for an extent at a time (if
+ * enough contiguous free blocks are available, there will only be one
+ * allocation request per call) and uses the state machine to initialise
+ * the blocks in order.
  *
  * Right now, this function will allocate at most one indirect block
  * worth of data -- with a default block size of 4K, that's slightly
@@ -633,39 +639,26 @@ static int gfs2_iomap_alloc(struct inode *inode, struct iomap *iomap,
 	struct buffer_head *dibh = mp->mp_bh[0];
 	u64 bn;
 	unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0;
-	unsigned dblks = 0;
-	unsigned ptrs_per_blk;
+	size_t dblks = iomap->length >> inode->i_blkbits;
 	const unsigned end_of_metadata = mp->mp_fheight - 1;
 	int ret;
 	enum alloc_state state;
 	__be64 *ptr;
 	__be64 zero_bn = 0;
-	size_t maxlen = iomap->length >> inode->i_blkbits;
 
 	BUG_ON(mp->mp_aheight < 1);
 	BUG_ON(dibh == NULL);
+	BUG_ON(dblks < 1);
 
 	gfs2_trans_add_meta(ip->i_gl, dibh);
 
 	down_write(&ip->i_rw_mutex);
 
 	if (mp->mp_fheight == mp->mp_aheight) {
-		struct buffer_head *bh;
-		int eob;
-
-		/* Bottom indirect block exists, find unalloced extent size */
-		ptr = metapointer(end_of_metadata, mp);
-		bh = mp->mp_bh[end_of_metadata];
-		dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr,
-					   maxlen, &eob);
-		BUG_ON(dblks < 1);
+		/* Bottom indirect block exists */
 		state = ALLOC_DATA;
 	} else {
 		/* Need to allocate indirect blocks */
-		ptrs_per_blk = mp->mp_fheight > 1 ? sdp->sd_inptrs :
-			sdp->sd_diptrs;
-		dblks = min(maxlen, (size_t)(ptrs_per_blk -
-					     mp->mp_list[end_of_metadata]));
 		if (mp->mp_fheight == ip->i_height) {
 			/* Writing into existing tree, extend tree down */
 			iblks = mp->mp_fheight - mp->mp_aheight;
@@ -750,6 +743,7 @@ static int gfs2_iomap_alloc(struct inode *inode, struct iomap *iomap,
 		}
 	} while (iomap->addr == IOMAP_NULL_ADDR);
 
+	iomap->type = IOMAP_MAPPED;
 	iomap->length = (u64)dblks << inode->i_blkbits;
 	ip->i_height = mp->mp_fheight;
 	gfs2_add_inode_blocks(&ip->i_inode, alloced);
@@ -759,19 +753,64 @@ static int gfs2_iomap_alloc(struct inode *inode, struct iomap *iomap,
 	return ret;
 }
 
-static void gfs2_stuffed_iomap(struct inode *inode, struct iomap *iomap)
+static void gfs2_stuffed_iomap(struct inode *inode, struct iomap *iomap,
+			       u64 length)
 {
 	struct gfs2_inode *ip = GFS2_I(inode);
 
 	iomap->addr = (ip->i_no_addr << inode->i_blkbits) +
 		      sizeof(struct gfs2_dinode);
 	iomap->offset = 0;
-	iomap->length = i_size_read(inode);
+	iomap->length = length;
 	iomap->type = IOMAP_INLINE;
 }
 
 #define IOMAP_F_GFS2_BOUNDARY IOMAP_F_PRIVATE
 
+/**
+ * gfs2_alloc_size - Compute the maximum allocation size
+ * @inode: The inode
+ * @mp: The metapath
+ * @size: Requested size in blocks
+ *
+ * Compute the maximum size of the next allocation at @mp.
+ *
+ * Returns: size in blocks
+ */
+static u64 gfs2_alloc_size(struct inode *inode, struct metapath *mp, u64 size)
+{
+	struct gfs2_inode *ip = GFS2_I(inode);
+	struct gfs2_sbd *sdp = GFS2_SB(inode);
+	const __be64 *first, *ptr, *end;
+
+	/*
+	 * For writes to stuffed files, this function is called twice via
+	 * gfs2_iomap_get, before and after unstuffing. The size we return the
+	 * first time needs to be large enough to get the reservation and
+	 * allocation sizes right.  The size we return the second time must
+	 * be exact or else gfs2_iomap_alloc won't do the right thing.
+	 */
+
+	if (gfs2_is_stuffed(ip) || mp->mp_fheight != mp->mp_aheight) {
+		unsigned maxsize = mp->mp_fheight > 1 ?
+			sdp->sd_inptrs : sdp->sd_diptrs;
+		maxsize -= mp->mp_list[mp->mp_fheight - 1];
+		if (size > maxsize)
+			size = maxsize;
+		return size;
+	}
+
+	first = metapointer(ip->i_height - 1, mp);
+	end = metaend(ip->i_height - 1, mp);
+	if (end - first > size)
+		end = first + size;
+	for (ptr = first; ptr < end; ptr++) {
+		if (*ptr)
+			break;
+	}
+	return ptr - first;
+}
+
 /**
  * gfs2_iomap_get - Map blocks from an inode to disk blocks
  * @inode: The inode
@@ -805,10 +844,15 @@ static int gfs2_iomap_get(struct inode *inode, loff_t pos, loff_t length,
 		if (flags & IOMAP_REPORT) {
 			if (pos >= i_size_read(inode))
 				return -ENOENT;
-			gfs2_stuffed_iomap(inode, iomap);
+			gfs2_stuffed_iomap(inode, iomap,
+					   i_size_read(inode));
+			return 0;
+		}
+		if (pos + length <= gfs2_max_stuffed_size(ip)) {
+			gfs2_stuffed_iomap(inode, iomap,
+					   gfs2_max_stuffed_size(ip));
 			return 0;
 		}
-		BUG_ON(!(flags & IOMAP_WRITE));
 	}
 	lblock = pos >> inode->i_blkbits;
 	iomap->offset = lblock << inode->i_blkbits;
@@ -868,10 +912,173 @@ static int gfs2_iomap_get(struct inode *inode, loff_t pos, loff_t length,
 			ret = gfs2_hole_size(inode, lblock, len, mp, iomap);
 		else
 			iomap->length = size - pos;
+	} else if (flags & IOMAP_WRITE) {
+		u64 size;
+		size = gfs2_alloc_size(inode, mp, len) << inode->i_blkbits;
+		if (size < iomap->length)
+			iomap->length = size;
+	} else {
+		if (height == ip->i_height)
+			ret = gfs2_hole_size(inode, lblock, len, mp, iomap);
 	}
 	goto out;
 }
 
+static int gfs2_write_lock(struct inode *inode)
+{
+	struct gfs2_inode *ip = GFS2_I(inode);
+	struct gfs2_sbd *sdp = GFS2_SB(inode);
+	int error;
+
+	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
+	error = gfs2_glock_nq(&ip->i_gh);
+	if (error)
+		goto out_uninit;
+	if (&ip->i_inode == sdp->sd_rindex) {
+		struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
+
+		error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
+					   GL_NOCACHE, &m_ip->i_gh);
+		if (error)
+			goto out_unlock;
+	}
+	return 0;
+
+out_unlock:
+	gfs2_glock_dq(&ip->i_gh);
+out_uninit:
+	gfs2_holder_uninit(&ip->i_gh);
+	return error;
+}
+
+static void gfs2_write_unlock(struct inode *inode)
+{
+	struct gfs2_inode *ip = GFS2_I(inode);
+	struct gfs2_sbd *sdp = GFS2_SB(inode);
+
+	if (&ip->i_inode == sdp->sd_rindex) {
+		struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
+
+		gfs2_glock_dq_uninit(&m_ip->i_gh);
+	}
+	gfs2_glock_dq_uninit(&ip->i_gh);
+}
+
+static void gfs2_iomap_journaled_page_done(struct inode *inode, loff_t pos,
+				unsigned copied, struct page *page,
+				struct iomap *iomap)
+{
+	struct gfs2_inode *ip = GFS2_I(inode);
+
+	gfs2_page_add_databufs(ip, page, offset_in_page(pos), copied);
+}
+
+static int gfs2_iomap_begin_write(struct inode *inode, loff_t pos, loff_t length,
+				  unsigned flags, struct iomap *iomap)
+{
+	struct metapath mp = { .mp_aheight = 1, };
+	struct gfs2_inode *ip = GFS2_I(inode);
+	struct gfs2_sbd *sdp = GFS2_SB(inode);
+	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
+	bool unstuff, alloc_required;
+	struct buffer_head *dibh = NULL;
+	int ret;
+
+	ret = gfs2_write_lock(inode);
+	if (ret)
+		return ret;
+
+	unstuff = gfs2_is_stuffed(ip) &&
+		  pos + length > gfs2_max_stuffed_size(ip);
+
+	ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
+	if (ret)
+		goto out_release;
+
+	dibh = mp.mp_bh[0];
+	if (dibh) {
+		get_bh(dibh);
+	} else {
+		ret = gfs2_meta_inode_buffer(ip, &dibh);
+		if (ret)
+			goto out_release;
+	}
+	iomap->private = dibh;
+
+	if (iomap->type == IOMAP_INLINE)
+		iomap->inline_data = dibh->b_data + sizeof(struct gfs2_dinode);
+
+	alloc_required = unstuff || iomap->type == IOMAP_HOLE;
+
+	if (alloc_required || gfs2_is_jdata(ip))
+		gfs2_write_calc_reserv(ip, iomap->length, &data_blocks, &ind_blocks);
+
+	if (alloc_required) {
+		struct gfs2_alloc_parms ap = { .target = data_blocks + ind_blocks };
+		ret = gfs2_quota_lock_check(ip, &ap);
+		if (ret)
+			goto out_release;
+
+		ret = gfs2_inplace_reserve(ip, &ap);
+		if (ret)
+			goto out_qunlock;
+	}
+
+	rblocks = RES_DINODE + ind_blocks;
+	if (gfs2_is_jdata(ip))
+		rblocks += data_blocks;
+	if (ind_blocks || data_blocks)
+		rblocks += RES_STATFS + RES_QUOTA;
+	if (inode == sdp->sd_rindex)
+		rblocks += 2 * RES_STATFS;
+	if (alloc_required)
+		rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
+
+	ret = gfs2_trans_begin(sdp, rblocks, iomap->length >> inode->i_blkbits);
+	if (ret)
+		goto out_trans_fail;
+
+	if (unstuff) {
+		ret = gfs2_unstuff_dinode(ip, NULL);
+		if (ret)
+			goto out_trans_end;
+		release_metapath(&mp);
+		ret = gfs2_iomap_get(inode, iomap->offset, iomap->length,
+				     flags, iomap, &mp);
+		if (ret)
+			goto out_trans_end;
+	}
+
+	if (iomap->type == IOMAP_HOLE) {
+		ret = gfs2_iomap_alloc(inode, iomap, flags, &mp);
+		if (ret) {
+			gfs2_trans_end(sdp);
+			gfs2_inplace_release(ip);
+			punch_hole(ip, iomap->offset, iomap->length);
+			goto out_qunlock;
+		}
+	}
+	release_metapath(&mp);
+	if (gfs2_is_jdata(ip))
+		iomap->page_done = gfs2_iomap_journaled_page_done;
+	return 0;
+
+out_trans_end:
+	gfs2_trans_end(sdp);
+out_trans_fail:
+	if (alloc_required)
+		gfs2_inplace_release(ip);
+out_qunlock:
+	if (alloc_required)
+		gfs2_quota_unlock(ip);
+out_release:
+	release_metapath(&mp);
+	if (dibh)
+		brelse(dibh);
+	gfs2_write_unlock(inode);
+	return ret;
+}
+
 static int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length,
 			    unsigned flags, struct iomap *iomap)
 {
@@ -881,10 +1088,7 @@ static int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length,
 
 	trace_gfs2_iomap_start(ip, pos, length, flags);
 	if (flags & IOMAP_WRITE) {
-		ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
-		if (!ret && iomap->type == IOMAP_HOLE)
-			ret = gfs2_iomap_alloc(inode, iomap, flags, &mp);
-		release_metapath(&mp);
+		ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap);
 	} else {
 		ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
 		release_metapath(&mp);
@@ -893,8 +1097,61 @@ static int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length,
 	return ret;
 }
 
+static int gfs2_iomap_end(struct inode *inode, loff_t pos, loff_t length,
+			  ssize_t written, unsigned flags, struct iomap *iomap)
+{
+	struct gfs2_inode *ip = GFS2_I(inode);
+	struct gfs2_sbd *sdp = GFS2_SB(inode);
+	struct gfs2_trans *tr = current->journal_info;
+	struct buffer_head *dibh;
+
+	if (!(flags & IOMAP_WRITE))
+		return 0;
+
+	dibh = iomap->private;
+	if (iomap->type == IOMAP_INLINE) {
+		gfs2_trans_add_meta(ip->i_gl, dibh);
+		goto out;
+	}
+
+	gfs2_ordered_add_inode(ip);
+
+	if (tr->tr_num_buf_new)
+		__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
+	else {
+		gfs2_trans_add_meta(ip->i_gl, dibh);
+	}
+
+out:
+	brelse(dibh);
+	if (inode == sdp->sd_rindex) {
+		adjust_fs_space(inode);
+		sdp->sd_rindex_uptodate = 0;
+	}
+
+	gfs2_trans_end(sdp);
+	gfs2_inplace_release(ip);
+
+	if (length != written && (iomap->flags & IOMAP_F_NEW)) {
+		/* Deallocate blocks that were just allocated. */
+		loff_t blockmask = i_blocksize(inode) - 1;
+		loff_t end = (pos + length) & ~blockmask;
+		pos = (pos + written + blockmask) & ~blockmask;
+		if (pos < end) {
+			truncate_pagecache_range(inode, pos, end - 1);
+			punch_hole(ip, pos, end - pos);
+		}
+	}
+
+	if (ip->i_qadata && ip->i_qadata->qa_qd_num)
+		gfs2_quota_unlock(ip);
+	gfs2_write_unlock(inode);
+	return 0;
+}
+
 const struct iomap_ops gfs2_iomap_ops = {
 	.iomap_begin = gfs2_iomap_begin,
+	.iomap_end = gfs2_iomap_end,
 };
 
 /**
diff --git a/fs/gfs2/file.c b/fs/gfs2/file.c
index 7137db7b0119..16dd395479a5 100644
--- a/fs/gfs2/file.c
+++ b/fs/gfs2/file.c
@@ -26,10 +26,12 @@
 #include <linux/dlm.h>
 #include <linux/dlm_plock.h>
 #include <linux/delay.h>
+#include <linux/backing-dev.h>
 
 #include "gfs2.h"
 #include "incore.h"
 #include "bmap.h"
+#include "aops.h"
 #include "dir.h"
 #include "glock.h"
 #include "glops.h"
@@ -691,9 +693,7 @@ static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
 /**
  * gfs2_file_write_iter - Perform a write to a file
  * @iocb: The io context
- * @iov: The data to write
- * @nr_segs: Number of @iov segments
- * @pos: The file position
+ * @from: The data to write
  *
  * We have to do a lock/unlock here to refresh the inode size for
  * O_APPEND writes, otherwise we can land up writing at the wrong
@@ -705,8 +705,9 @@ static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 {
 	struct file *file = iocb->ki_filp;
-	struct gfs2_inode *ip = GFS2_I(file_inode(file));
-	int ret;
+	struct inode *inode = file_inode(file);
+	struct gfs2_inode *ip = GFS2_I(inode);
+	ssize_t ret;
 
 	ret = gfs2_rsqa_alloc(ip);
 	if (ret)
@@ -723,7 +724,38 @@ static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 		gfs2_glock_dq_uninit(&gh);
 	}
 
-	return generic_file_write_iter(iocb, from);
+	if (iocb->ki_flags & IOCB_DIRECT)
+		return generic_file_write_iter(iocb, from);
+
+	inode_lock(inode);
+	ret = generic_write_checks(iocb, from);
+	if (ret <= 0)
+		goto out;
+
+	/* We can write back this queue in page reclaim */
+	current->backing_dev_info = inode_to_bdi(inode);
+
+	ret = file_remove_privs(file);
+	if (ret)
+		goto out2;
+
+	ret = file_update_time(file);
+	if (ret)
+		goto out2;
+
+	ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
+
+out2:
+	current->backing_dev_info = NULL;
+out:
+	inode_unlock(inode);
+	if (likely(ret > 0)) {
+		iocb->ki_pos += ret;
+
+		/* Handle various SYNC-type writes */
+		ret = generic_write_sync(iocb, ret);
+	}
+	return ret;
 }
 
 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
@@ -754,7 +786,7 @@ static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
 		if (error)
 			goto out;
 		offset = iomap.offset + iomap.length;
-		if (iomap.type != IOMAP_HOLE)
+		if (!(iomap.flags & IOMAP_F_NEW))
 			continue;
 		error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
 					 iomap.length >> inode->i_blkbits,
-- 
2.17.1

[Cluster-devel] [PATCH v9 2/5] gfs2: iomap buffered write support

[Andreas Gruenbacher]

With the traditional page-based writes, blocks are allocated separately
for each page written to.  With iomap writes, we can allocate a lot more
blocks at once, with a fraction of the allocation overhead for each
page.

Split calculating the number of blocks that can be allocated at a given
position (gfs2_alloc_size) off from gfs2_iomap_alloc: that size
determines the number of blocks to allocate and reserve in the journal.

In gfs2_iomap_alloc, set the type of newly allocated extents to
IOMAP_MAPPED so that iomap_to_bh will set the bh states correctly:
otherwise, the bhs would not be marked as mapped, confusing
__mpage_writepage.  This means that we need to check for the IOMAP_F_NEW
flag in fallocate_chunk now.

Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
---
 fs/gfs2/aops.c |  20 ++--
 fs/gfs2/aops.h |  22 ++++
 fs/gfs2/bmap.c | 317 ++++++++++++++++++++++++++++++++++++++++++++-----
 fs/gfs2/file.c |  46 +++++--
 4 files changed, 359 insertions(+), 46 deletions(-)
 create mode 100644 fs/gfs2/aops.h

diff --git a/fs/gfs2/aops.c b/fs/gfs2/aops.c
index 35f5ee23566d..ecfbca9c88ff 100644
--- a/fs/gfs2/aops.c
+++ b/fs/gfs2/aops.c
@@ -22,6 +22,7 @@
 #include <linux/backing-dev.h>
 #include <linux/uio.h>
 #include <trace/events/writeback.h>
+#include <linux/sched/signal.h>
 
 #include "gfs2.h"
 #include "incore.h"
@@ -36,10 +37,11 @@
 #include "super.h"
 #include "util.h"
 #include "glops.h"
+#include "aops.h"
 
 
-static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
-				   unsigned int from, unsigned int len)
+void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
+			    unsigned int from, unsigned int len)
 {
 	struct buffer_head *head = page_buffers(page);
 	unsigned int bsize = head->b_size;
@@ -462,7 +464,7 @@ static int gfs2_jdata_writepages(struct address_space *mapping,
  * Returns: errno
  */
 
-static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
+int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
 {
 	struct buffer_head *dibh;
 	u64 dsize = i_size_read(&ip->i_inode);
@@ -776,7 +778,7 @@ static int gfs2_write_begin(struct file *file, struct address_space *mapping,
  * adjust_fs_space - Adjusts the free space available due to gfs2_grow
  * @inode: the rindex inode
  */
-static void adjust_fs_space(struct inode *inode)
+void adjust_fs_space(struct inode *inode)
 {
 	struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
 	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
@@ -822,11 +824,11 @@ static void adjust_fs_space(struct inode *inode)
  * This copies the data from the page into the inode block after
  * the inode data structure itself.
  *
- * Returns: errno
+ * Returns: copied bytes or errno
  */
-static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
-				  loff_t pos, unsigned copied,
-				  struct page *page)
+int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
+			   loff_t pos, unsigned copied,
+			   struct page *page)
 {
 	struct gfs2_inode *ip = GFS2_I(inode);
 	u64 to = pos + copied;
@@ -865,7 +867,7 @@ static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
  * The main write_end function for GFS2. We just put our locking around the VFS
  * provided functions.
  *
- * Returns: errno
+ * Returns: copied bytes or errno
  */
 
 static int gfs2_write_end(struct file *file, struct address_space *mapping,
diff --git a/fs/gfs2/aops.h b/fs/gfs2/aops.h
new file mode 100644
index 000000000000..976bb32dd405
--- /dev/null
+++ b/fs/gfs2/aops.h
@@ -0,0 +1,22 @@
+/*
+ * Copyright (C) 2017 Red Hat, Inc.  All rights reserved.
+ *
+ * This copyrighted material is made available to anyone wishing to use,
+ * modify, copy, or redistribute it subject to the terms and conditions
+ * of the GNU General Public License version 2.
+ */
+
+#ifndef __AOPS_DOT_H__
+#define __AOPS_DOT_H__
+
+#include "incore.h"
+
+extern int stuffed_readpage(struct gfs2_inode *ip, struct page *page);
+extern int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
+				  loff_t pos, unsigned copied,
+				  struct page *page);
+extern void adjust_fs_space(struct inode *inode);
+extern void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
+				   unsigned int from, unsigned int len);
+
+#endif /* __AOPS_DOT_H__ */
diff --git a/fs/gfs2/bmap.c b/fs/gfs2/bmap.c
index ed6699705c13..704a9150dcbc 100644
--- a/fs/gfs2/bmap.c
+++ b/fs/gfs2/bmap.c
@@ -28,6 +28,7 @@
 #include "trans.h"
 #include "dir.h"
 #include "util.h"
+#include "aops.h"
 #include "trace_gfs2.h"
 
 /* This doesn't need to be that large as max 64 bit pointers in a 4k
@@ -41,6 +42,8 @@ struct metapath {
 	int mp_aheight; /* actual height (lookup height) */
 };
 
+static int punch_hole(struct gfs2_inode *ip, u64 offset, u64 length);
+
 /**
  * gfs2_unstuffer_page - unstuff a stuffed inode into a block cached by a page
  * @ip: the inode
@@ -389,7 +392,7 @@ static int fillup_metapath(struct gfs2_inode *ip, struct metapath *mp, int h)
 	return mp->mp_aheight - x - 1;
 }
 
-static inline void release_metapath(struct metapath *mp)
+static void release_metapath(struct metapath *mp)
 {
 	int i;
 
@@ -397,6 +400,7 @@ static inline void release_metapath(struct metapath *mp)
 		if (mp->mp_bh[i] == NULL)
 			break;
 		brelse(mp->mp_bh[i]);
+		mp->mp_bh[i] = NULL;
 	}
 }
 
@@ -609,11 +613,13 @@ enum alloc_state {
  *  ii) Indirect blocks to fill in lower part of the metadata tree
  * iii) Data blocks
  *
- * The function is in two parts. The first part works out the total
- * number of blocks which we need. The second part does the actual
- * allocation asking for an extent at a time (if enough contiguous free
- * blocks are available, there will only be one request per bmap call)
- * and uses the state machine to initialise the blocks in order.
+ * This function is called after gfs2_iomap_get, which works out the
+ * total number of blocks which we need via gfs2_alloc_size.
+ *
+ * We then do the actual allocation asking for an extent at a time (if
+ * enough contiguous free blocks are available, there will only be one
+ * allocation request per call) and uses the state machine to initialise
+ * the blocks in order.
  *
  * Right now, this function will allocate at most one indirect block
  * worth of data -- with a default block size of 4K, that's slightly
@@ -633,39 +639,26 @@ static int gfs2_iomap_alloc(struct inode *inode, struct iomap *iomap,
 	struct buffer_head *dibh = mp->mp_bh[0];
 	u64 bn;
 	unsigned n, i, blks, alloced = 0, iblks = 0, branch_start = 0;
-	unsigned dblks = 0;
-	unsigned ptrs_per_blk;
+	size_t dblks = iomap->length >> inode->i_blkbits;
 	const unsigned end_of_metadata = mp->mp_fheight - 1;
 	int ret;
 	enum alloc_state state;
 	__be64 *ptr;
 	__be64 zero_bn = 0;
-	size_t maxlen = iomap->length >> inode->i_blkbits;
 
 	BUG_ON(mp->mp_aheight < 1);
 	BUG_ON(dibh == NULL);
+	BUG_ON(dblks < 1);
 
 	gfs2_trans_add_meta(ip->i_gl, dibh);
 
 	down_write(&ip->i_rw_mutex);
 
 	if (mp->mp_fheight == mp->mp_aheight) {
-		struct buffer_head *bh;
-		int eob;
-
-		/* Bottom indirect block exists, find unalloced extent size */
-		ptr = metapointer(end_of_metadata, mp);
-		bh = mp->mp_bh[end_of_metadata];
-		dblks = gfs2_extent_length(bh->b_data, bh->b_size, ptr,
-					   maxlen, &eob);
-		BUG_ON(dblks < 1);
+		/* Bottom indirect block exists */
 		state = ALLOC_DATA;
 	} else {
 		/* Need to allocate indirect blocks */
-		ptrs_per_blk = mp->mp_fheight > 1 ? sdp->sd_inptrs :
-			sdp->sd_diptrs;
-		dblks = min(maxlen, (size_t)(ptrs_per_blk -
-					     mp->mp_list[end_of_metadata]));
 		if (mp->mp_fheight == ip->i_height) {
 			/* Writing into existing tree, extend tree down */
 			iblks = mp->mp_fheight - mp->mp_aheight;
@@ -750,6 +743,7 @@ static int gfs2_iomap_alloc(struct inode *inode, struct iomap *iomap,
 		}
 	} while (iomap->addr == IOMAP_NULL_ADDR);
 
+	iomap->type = IOMAP_MAPPED;
 	iomap->length = (u64)dblks << inode->i_blkbits;
 	ip->i_height = mp->mp_fheight;
 	gfs2_add_inode_blocks(&ip->i_inode, alloced);
@@ -759,19 +753,64 @@ static int gfs2_iomap_alloc(struct inode *inode, struct iomap *iomap,
 	return ret;
 }
 
-static void gfs2_stuffed_iomap(struct inode *inode, struct iomap *iomap)
+static void gfs2_stuffed_iomap(struct inode *inode, struct iomap *iomap,
+			       u64 length)
 {
 	struct gfs2_inode *ip = GFS2_I(inode);
 
 	iomap->addr = (ip->i_no_addr << inode->i_blkbits) +
 		      sizeof(struct gfs2_dinode);
 	iomap->offset = 0;
-	iomap->length = i_size_read(inode);
+	iomap->length = length;
 	iomap->type = IOMAP_INLINE;
 }
 
 #define IOMAP_F_GFS2_BOUNDARY IOMAP_F_PRIVATE
 
+/**
+ * gfs2_alloc_size - Compute the maximum allocation size
+ * @inode: The inode
+ * @mp: The metapath
+ * @size: Requested size in blocks
+ *
+ * Compute the maximum size of the next allocation at @mp.
+ *
+ * Returns: size in blocks
+ */
+static u64 gfs2_alloc_size(struct inode *inode, struct metapath *mp, u64 size)
+{
+	struct gfs2_inode *ip = GFS2_I(inode);
+	struct gfs2_sbd *sdp = GFS2_SB(inode);
+	const __be64 *first, *ptr, *end;
+
+	/*
+	 * For writes to stuffed files, this function is called twice via
+	 * gfs2_iomap_get, before and after unstuffing. The size we return the
+	 * first time needs to be large enough to get the reservation and
+	 * allocation sizes right.  The size we return the second time must
+	 * be exact or else gfs2_iomap_alloc won't do the right thing.
+	 */
+
+	if (gfs2_is_stuffed(ip) || mp->mp_fheight != mp->mp_aheight) {
+		unsigned maxsize = mp->mp_fheight > 1 ?
+			sdp->sd_inptrs : sdp->sd_diptrs;
+		maxsize -= mp->mp_list[mp->mp_fheight - 1];
+		if (size > maxsize)
+			size = maxsize;
+		return size;
+	}
+
+	first = metapointer(ip->i_height - 1, mp);
+	end = metaend(ip->i_height - 1, mp);
+	if (end - first > size)
+		end = first + size;
+	for (ptr = first; ptr < end; ptr++) {
+		if (*ptr)
+			break;
+	}
+	return ptr - first;
+}
+
 /**
  * gfs2_iomap_get - Map blocks from an inode to disk blocks
  * @inode: The inode
@@ -805,10 +844,15 @@ static int gfs2_iomap_get(struct inode *inode, loff_t pos, loff_t length,
 		if (flags & IOMAP_REPORT) {
 			if (pos >= i_size_read(inode))
 				return -ENOENT;
-			gfs2_stuffed_iomap(inode, iomap);
+			gfs2_stuffed_iomap(inode, iomap,
+					   i_size_read(inode));
+			return 0;
+		}
+		if (pos + length <= gfs2_max_stuffed_size(ip)) {
+			gfs2_stuffed_iomap(inode, iomap,
+					   gfs2_max_stuffed_size(ip));
 			return 0;
 		}
-		BUG_ON(!(flags & IOMAP_WRITE));
 	}
 	lblock = pos >> inode->i_blkbits;
 	iomap->offset = lblock << inode->i_blkbits;
@@ -868,10 +912,173 @@ static int gfs2_iomap_get(struct inode *inode, loff_t pos, loff_t length,
 			ret = gfs2_hole_size(inode, lblock, len, mp, iomap);
 		else
 			iomap->length = size - pos;
+	} else if (flags & IOMAP_WRITE) {
+		u64 size;
+		size = gfs2_alloc_size(inode, mp, len) << inode->i_blkbits;
+		if (size < iomap->length)
+			iomap->length = size;
+	} else {
+		if (height == ip->i_height)
+			ret = gfs2_hole_size(inode, lblock, len, mp, iomap);
 	}
 	goto out;
 }
 
+static int gfs2_write_lock(struct inode *inode)
+{
+	struct gfs2_inode *ip = GFS2_I(inode);
+	struct gfs2_sbd *sdp = GFS2_SB(inode);
+	int error;
+
+	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
+	error = gfs2_glock_nq(&ip->i_gh);
+	if (error)
+		goto out_uninit;
+	if (&ip->i_inode == sdp->sd_rindex) {
+		struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
+
+		error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
+					   GL_NOCACHE, &m_ip->i_gh);
+		if (error)
+			goto out_unlock;
+	}
+	return 0;
+
+out_unlock:
+	gfs2_glock_dq(&ip->i_gh);
+out_uninit:
+	gfs2_holder_uninit(&ip->i_gh);
+	return error;
+}
+
+static void gfs2_write_unlock(struct inode *inode)
+{
+	struct gfs2_inode *ip = GFS2_I(inode);
+	struct gfs2_sbd *sdp = GFS2_SB(inode);
+
+	if (&ip->i_inode == sdp->sd_rindex) {
+		struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
+
+		gfs2_glock_dq_uninit(&m_ip->i_gh);
+	}
+	gfs2_glock_dq_uninit(&ip->i_gh);
+}
+
+static void gfs2_iomap_journaled_page_done(struct inode *inode, loff_t pos,
+				unsigned copied, struct page *page,
+				struct iomap *iomap)
+{
+	struct gfs2_inode *ip = GFS2_I(inode);
+
+	gfs2_page_add_databufs(ip, page, offset_in_page(pos), copied);
+}
+
+static int gfs2_iomap_begin_write(struct inode *inode, loff_t pos, loff_t length,
+				  unsigned flags, struct iomap *iomap)
+{
+	struct metapath mp = { .mp_aheight = 1, };
+	struct gfs2_inode *ip = GFS2_I(inode);
+	struct gfs2_sbd *sdp = GFS2_SB(inode);
+	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
+	bool unstuff, alloc_required;
+	struct buffer_head *dibh = NULL;
+	int ret;
+
+	ret = gfs2_write_lock(inode);
+	if (ret)
+		return ret;
+
+	unstuff = gfs2_is_stuffed(ip) &&
+		  pos + length > gfs2_max_stuffed_size(ip);
+
+	ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
+	if (ret)
+		goto out_release;
+
+	dibh = mp.mp_bh[0];
+	if (dibh) {
+		get_bh(dibh);
+	} else {
+		ret = gfs2_meta_inode_buffer(ip, &dibh);
+		if (ret)
+			goto out_release;
+	}
+	iomap->private = dibh;
+
+	if (iomap->type == IOMAP_INLINE)
+		iomap->inline_data = dibh->b_data + sizeof(struct gfs2_dinode);
+
+	alloc_required = unstuff || iomap->type == IOMAP_HOLE;
+
+	if (alloc_required || gfs2_is_jdata(ip))
+		gfs2_write_calc_reserv(ip, iomap->length, &data_blocks, &ind_blocks);
+
+	if (alloc_required) {
+		struct gfs2_alloc_parms ap = { .target = data_blocks + ind_blocks };
+		ret = gfs2_quota_lock_check(ip, &ap);
+		if (ret)
+			goto out_release;
+
+		ret = gfs2_inplace_reserve(ip, &ap);
+		if (ret)
+			goto out_qunlock;
+	}
+
+	rblocks = RES_DINODE + ind_blocks;
+	if (gfs2_is_jdata(ip))
+		rblocks += data_blocks;
+	if (ind_blocks || data_blocks)
+		rblocks += RES_STATFS + RES_QUOTA;
+	if (inode == sdp->sd_rindex)
+		rblocks += 2 * RES_STATFS;
+	if (alloc_required)
+		rblocks += gfs2_rg_blocks(ip, data_blocks + ind_blocks);
+
+	ret = gfs2_trans_begin(sdp, rblocks, iomap->length >> inode->i_blkbits);
+	if (ret)
+		goto out_trans_fail;
+
+	if (unstuff) {
+		ret = gfs2_unstuff_dinode(ip, NULL);
+		if (ret)
+			goto out_trans_end;
+		release_metapath(&mp);
+		ret = gfs2_iomap_get(inode, iomap->offset, iomap->length,
+				     flags, iomap, &mp);
+		if (ret)
+			goto out_trans_end;
+	}
+
+	if (iomap->type == IOMAP_HOLE) {
+		ret = gfs2_iomap_alloc(inode, iomap, flags, &mp);
+		if (ret) {
+			gfs2_trans_end(sdp);
+			gfs2_inplace_release(ip);
+			punch_hole(ip, iomap->offset, iomap->length);
+			goto out_qunlock;
+		}
+	}
+	release_metapath(&mp);
+	if (gfs2_is_jdata(ip))
+		iomap->page_done = gfs2_iomap_journaled_page_done;
+	return 0;
+
+out_trans_end:
+	gfs2_trans_end(sdp);
+out_trans_fail:
+	if (alloc_required)
+		gfs2_inplace_release(ip);
+out_qunlock:
+	if (alloc_required)
+		gfs2_quota_unlock(ip);
+out_release:
+	release_metapath(&mp);
+	if (dibh)
+		brelse(dibh);
+	gfs2_write_unlock(inode);
+	return ret;
+}
+
 static int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length,
 			    unsigned flags, struct iomap *iomap)
 {
@@ -881,10 +1088,7 @@ static int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length,
 
 	trace_gfs2_iomap_start(ip, pos, length, flags);
 	if (flags & IOMAP_WRITE) {
-		ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
-		if (!ret && iomap->type == IOMAP_HOLE)
-			ret = gfs2_iomap_alloc(inode, iomap, flags, &mp);
-		release_metapath(&mp);
+		ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap);
 	} else {
 		ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
 		release_metapath(&mp);
@@ -893,8 +1097,61 @@ static int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length,
 	return ret;
 }
 
+static int gfs2_iomap_end(struct inode *inode, loff_t pos, loff_t length,
+			  ssize_t written, unsigned flags, struct iomap *iomap)
+{
+	struct gfs2_inode *ip = GFS2_I(inode);
+	struct gfs2_sbd *sdp = GFS2_SB(inode);
+	struct gfs2_trans *tr = current->journal_info;
+	struct buffer_head *dibh;
+
+	if (!(flags & IOMAP_WRITE))
+		return 0;
+
+	dibh = iomap->private;
+	if (iomap->type == IOMAP_INLINE) {
+		gfs2_trans_add_meta(ip->i_gl, dibh);
+		goto out;
+	}
+
+	gfs2_ordered_add_inode(ip);
+
+	if (tr->tr_num_buf_new)
+		__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
+	else {
+		gfs2_trans_add_meta(ip->i_gl, dibh);
+	}
+
+out:
+	brelse(dibh);
+	if (inode == sdp->sd_rindex) {
+		adjust_fs_space(inode);
+		sdp->sd_rindex_uptodate = 0;
+	}
+
+	gfs2_trans_end(sdp);
+	gfs2_inplace_release(ip);
+
+	if (length != written && (iomap->flags & IOMAP_F_NEW)) {
+		/* Deallocate blocks that were just allocated. */
+		loff_t blockmask = i_blocksize(inode) - 1;
+		loff_t end = (pos + length) & ~blockmask;
+		pos = (pos + written + blockmask) & ~blockmask;
+		if (pos < end) {
+			truncate_pagecache_range(inode, pos, end - 1);
+			punch_hole(ip, pos, end - pos);
+		}
+	}
+
+	if (ip->i_qadata && ip->i_qadata->qa_qd_num)
+		gfs2_quota_unlock(ip);
+	gfs2_write_unlock(inode);
+	return 0;
+}
+
 const struct iomap_ops gfs2_iomap_ops = {
 	.iomap_begin = gfs2_iomap_begin,
+	.iomap_end = gfs2_iomap_end,
 };
 
 /**
diff --git a/fs/gfs2/file.c b/fs/gfs2/file.c
index 7137db7b0119..16dd395479a5 100644
--- a/fs/gfs2/file.c
+++ b/fs/gfs2/file.c
@@ -26,10 +26,12 @@
 #include <linux/dlm.h>
 #include <linux/dlm_plock.h>
 #include <linux/delay.h>
+#include <linux/backing-dev.h>
 
 #include "gfs2.h"
 #include "incore.h"
 #include "bmap.h"
+#include "aops.h"
 #include "dir.h"
 #include "glock.h"
 #include "glops.h"
@@ -691,9 +693,7 @@ static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
 /**
  * gfs2_file_write_iter - Perform a write to a file
  * @iocb: The io context
- * @iov: The data to write
- * @nr_segs: Number of @iov segments
- * @pos: The file position
+ * @from: The data to write
  *
  * We have to do a lock/unlock here to refresh the inode size for
  * O_APPEND writes, otherwise we can land up writing at the wrong
@@ -705,8 +705,9 @@ static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
 static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 {
 	struct file *file = iocb->ki_filp;
-	struct gfs2_inode *ip = GFS2_I(file_inode(file));
-	int ret;
+	struct inode *inode = file_inode(file);
+	struct gfs2_inode *ip = GFS2_I(inode);
+	ssize_t ret;
 
 	ret = gfs2_rsqa_alloc(ip);
 	if (ret)
@@ -723,7 +724,38 @@ static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 		gfs2_glock_dq_uninit(&gh);
 	}
 
-	return generic_file_write_iter(iocb, from);
+	if (iocb->ki_flags & IOCB_DIRECT)
+		return generic_file_write_iter(iocb, from);
+
+	inode_lock(inode);
+	ret = generic_write_checks(iocb, from);
+	if (ret <= 0)
+		goto out;
+
+	/* We can write back this queue in page reclaim */
+	current->backing_dev_info = inode_to_bdi(inode);
+
+	ret = file_remove_privs(file);
+	if (ret)
+		goto out2;
+
+	ret = file_update_time(file);
+	if (ret)
+		goto out2;
+
+	ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
+
+out2:
+	current->backing_dev_info = NULL;
+out:
+	inode_unlock(inode);
+	if (likely(ret > 0)) {
+		iocb->ki_pos += ret;
+
+		/* Handle various SYNC-type writes */
+		ret = generic_write_sync(iocb, ret);
+	}
+	return ret;
 }
 
 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
@@ -754,7 +786,7 @@ static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
 		if (error)
 			goto out;
 		offset = iomap.offset + iomap.length;
-		if (iomap.type != IOMAP_HOLE)
+		if (!(iomap.flags & IOMAP_F_NEW))
 			continue;
 		error = sb_issue_zeroout(sb, iomap.addr >> inode->i_blkbits,
 					 iomap.length >> inode->i_blkbits,
-- 
2.17.1

Re: [PATCH v9 2/5] gfs2: iomap buffered write support

[Christoph Hellwig]

> +/*
> + * Copyright (C) 2017 Red Hat, Inc.  All rights reserved.
> + *
> + * This copyrighted material is made available to anyone wishing to use,
> + * modify, copy, or redistribute it subject to the terms and conditions
> + * of the GNU General Public License version 2.
> + */

Replace with an SPDX tag, please.

> +static int gfs2_iomap_begin_write(struct inode *inode, loff_t pos, loff_t length,

overly long line.

> +		gfs2_write_calc_reserv(ip, iomap->length, &data_blocks, &ind_blocks);

plus a few more down the function.

> +	if (iocb->ki_flags & IOCB_DIRECT)
> +		return generic_file_write_iter(iocb, from);

This looks odd, as generic_file_write_iter duplicate the code
below to a large extent, but it's probably ok.

> @@ -754,7 +786,7 @@ static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
>  		if (error)
>  			goto out;
>  		offset = iomap.offset + iomap.length;
> -		if (iomap.type != IOMAP_HOLE)
> +		if (!(iomap.flags & IOMAP_F_NEW))
>  			continue;

shouldn't this be a separate patch?

[Cluster-devel] [PATCH v9 2/5] gfs2: iomap buffered write support

[Christoph Hellwig]

> +/*
> + * Copyright (C) 2017 Red Hat, Inc.  All rights reserved.
> + *
> + * This copyrighted material is made available to anyone wishing to use,
> + * modify, copy, or redistribute it subject to the terms and conditions
> + * of the GNU General Public License version 2.
> + */

Replace with an SPDX tag, please.

> +static int gfs2_iomap_begin_write(struct inode *inode, loff_t pos, loff_t length,

overly long line.

> +		gfs2_write_calc_reserv(ip, iomap->length, &data_blocks, &ind_blocks);

plus a few more down the function.

> +	if (iocb->ki_flags & IOCB_DIRECT)
> +		return generic_file_write_iter(iocb, from);

This looks odd, as generic_file_write_iter duplicate the code
below to a large extent, but it's probably ok.

> @@ -754,7 +786,7 @@ static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
>  		if (error)
>  			goto out;
>  		offset = iomap.offset + iomap.length;
> -		if (iomap.type != IOMAP_HOLE)
> +		if (!(iomap.flags & IOMAP_F_NEW))
>  			continue;

shouldn't this be a separate patch?

[PATCH v9 3/5] gfs2: gfs2_extent_length cleanup

[Andreas Gruenbacher]

Now that gfs2_extent_length is no longer used for determining the size
of a hole and always with an upper size limit, the function can be
simplified.

Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
---
 fs/gfs2/bmap.c | 24 ++++++++----------------
 1 file changed, 8 insertions(+), 16 deletions(-)

diff --git a/fs/gfs2/bmap.c b/fs/gfs2/bmap.c
index 704a9150dcbc..b70ef98a9637 100644
--- a/fs/gfs2/bmap.c
+++ b/fs/gfs2/bmap.c
@@ -406,22 +406,17 @@ static void release_metapath(struct metapath *mp)
 
 /**
  * gfs2_extent_length - Returns length of an extent of blocks
- * @start: Start of the buffer
- * @len: Length of the buffer in bytes
- * @ptr: Current position in the buffer
- * @limit: Max extent length to return (0 = unlimited)
+ * @bh: The metadata block
+ * @ptr: Current position in @bh
+ * @limit: Max extent length to return
  * @eob: Set to 1 if we hit "end of block"
  *
- * If the first block is zero (unallocated) it will return the number of
- * unallocated blocks in the extent, otherwise it will return the number
- * of contiguous blocks in the extent.
- *
  * Returns: The length of the extent (minimum of one block)
  */
 
-static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, size_t limit, int *eob)
+static inline unsigned int gfs2_extent_length(struct buffer_head *bh, __be64 *ptr, size_t limit, int *eob)
 {
-	const __be64 *end = (start + len);
+	const __be64 *end = (__be64 *)(bh->b_data + bh->b_size);
 	const __be64 *first = ptr;
 	u64 d = be64_to_cpu(*ptr);
 
@@ -430,14 +425,11 @@ static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __b
 		ptr++;
 		if (ptr >= end)
 			break;
-		if (limit && --limit == 0)
-			break;
-		if (d)
-			d++;
+		d++;
 	} while(be64_to_cpu(*ptr) == d);
 	if (ptr >= end)
 		*eob = 1;
-	return (ptr - first);
+	return ptr - first;
 }
 
 typedef const __be64 *(*gfs2_metadata_walker)(
@@ -884,7 +876,7 @@ static int gfs2_iomap_get(struct inode *inode, loff_t pos, loff_t length,
 		goto do_alloc;
 
 	bh = mp->mp_bh[ip->i_height - 1];
-	len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, len, &eob);
+	len = gfs2_extent_length(bh, ptr, len, &eob);
 
 	iomap->addr = be64_to_cpu(*ptr) << inode->i_blkbits;
 	iomap->length = len << inode->i_blkbits;
-- 
2.17.1

[Cluster-devel] [PATCH v9 3/5] gfs2: gfs2_extent_length cleanup

[Andreas Gruenbacher]

Now that gfs2_extent_length is no longer used for determining the size
of a hole and always with an upper size limit, the function can be
simplified.

Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
---
 fs/gfs2/bmap.c | 24 ++++++++----------------
 1 file changed, 8 insertions(+), 16 deletions(-)

diff --git a/fs/gfs2/bmap.c b/fs/gfs2/bmap.c
index 704a9150dcbc..b70ef98a9637 100644
--- a/fs/gfs2/bmap.c
+++ b/fs/gfs2/bmap.c
@@ -406,22 +406,17 @@ static void release_metapath(struct metapath *mp)
 
 /**
  * gfs2_extent_length - Returns length of an extent of blocks
- * @start: Start of the buffer
- * @len: Length of the buffer in bytes
- * @ptr: Current position in the buffer
- * @limit: Max extent length to return (0 = unlimited)
+ * @bh: The metadata block
+ * @ptr: Current position in @bh
+ * @limit: Max extent length to return
  * @eob: Set to 1 if we hit "end of block"
  *
- * If the first block is zero (unallocated) it will return the number of
- * unallocated blocks in the extent, otherwise it will return the number
- * of contiguous blocks in the extent.
- *
  * Returns: The length of the extent (minimum of one block)
  */
 
-static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __be64 *ptr, size_t limit, int *eob)
+static inline unsigned int gfs2_extent_length(struct buffer_head *bh, __be64 *ptr, size_t limit, int *eob)
 {
-	const __be64 *end = (start + len);
+	const __be64 *end = (__be64 *)(bh->b_data + bh->b_size);
 	const __be64 *first = ptr;
 	u64 d = be64_to_cpu(*ptr);
 
@@ -430,14 +425,11 @@ static inline unsigned int gfs2_extent_length(void *start, unsigned int len, __b
 		ptr++;
 		if (ptr >= end)
 			break;
-		if (limit && --limit == 0)
-			break;
-		if (d)
-			d++;
+		d++;
 	} while(be64_to_cpu(*ptr) == d);
 	if (ptr >= end)
 		*eob = 1;
-	return (ptr - first);
+	return ptr - first;
 }
 
 typedef const __be64 *(*gfs2_metadata_walker)(
@@ -884,7 +876,7 @@ static int gfs2_iomap_get(struct inode *inode, loff_t pos, loff_t length,
 		goto do_alloc;
 
 	bh = mp->mp_bh[ip->i_height - 1];
-	len = gfs2_extent_length(bh->b_data, bh->b_size, ptr, len, &eob);
+	len = gfs2_extent_length(bh, ptr, len, &eob);
 
 	iomap->addr = be64_to_cpu(*ptr) << inode->i_blkbits;
 	iomap->length = len << inode->i_blkbits;
-- 
2.17.1

[PATCH v9 4/5] gfs2: iomap direct I/O support

[Andreas Gruenbacher]

The page unmapping previously done in gfs2_direct_IO is now done
generically in iomap_dio_rw.

Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
---
 fs/gfs2/aops.c | 100 +----------------------------------
 fs/gfs2/bmap.c |  15 +++++-
 fs/gfs2/file.c | 138 +++++++++++++++++++++++++++++++++++++++++++++----
 3 files changed, 143 insertions(+), 110 deletions(-)

diff --git a/fs/gfs2/aops.c b/fs/gfs2/aops.c
index ecfbca9c88ff..1054cc4a96db 100644
--- a/fs/gfs2/aops.c
+++ b/fs/gfs2/aops.c
@@ -84,12 +84,6 @@ static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
 	return 0;
 }
 
-static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
-				 struct buffer_head *bh_result, int create)
-{
-	return gfs2_block_map(inode, lblock, bh_result, 0);
-}
-
 /**
  * gfs2_writepage_common - Common bits of writepage
  * @page: The page to be written
@@ -1024,96 +1018,6 @@ static void gfs2_invalidatepage(struct page *page, unsigned int offset,
 		try_to_release_page(page, 0);
 }
 
-/**
- * gfs2_ok_for_dio - check that dio is valid on this file
- * @ip: The inode
- * @offset: The offset at which we are reading or writing
- *
- * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
- *          1 (to accept the i/o request)
- */
-static int gfs2_ok_for_dio(struct gfs2_inode *ip, loff_t offset)
-{
-	/*
-	 * Should we return an error here? I can't see that O_DIRECT for
-	 * a stuffed file makes any sense. For now we'll silently fall
-	 * back to buffered I/O
-	 */
-	if (gfs2_is_stuffed(ip))
-		return 0;
-
-	if (offset >= i_size_read(&ip->i_inode))
-		return 0;
-	return 1;
-}
-
-
-
-static ssize_t gfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
-{
-	struct file *file = iocb->ki_filp;
-	struct inode *inode = file->f_mapping->host;
-	struct address_space *mapping = inode->i_mapping;
-	struct gfs2_inode *ip = GFS2_I(inode);
-	loff_t offset = iocb->ki_pos;
-	struct gfs2_holder gh;
-	int rv;
-
-	/*
-	 * Deferred lock, even if its a write, since we do no allocation
-	 * on this path. All we need change is atime, and this lock mode
-	 * ensures that other nodes have flushed their buffered read caches
-	 * (i.e. their page cache entries for this inode). We do not,
-	 * unfortunately have the option of only flushing a range like
-	 * the VFS does.
-	 */
-	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
-	rv = gfs2_glock_nq(&gh);
-	if (rv)
-		goto out_uninit;
-	rv = gfs2_ok_for_dio(ip, offset);
-	if (rv != 1)
-		goto out; /* dio not valid, fall back to buffered i/o */
-
-	/*
-	 * Now since we are holding a deferred (CW) lock at this point, you
-	 * might be wondering why this is ever needed. There is a case however
-	 * where we've granted a deferred local lock against a cached exclusive
-	 * glock. That is ok provided all granted local locks are deferred, but
-	 * it also means that it is possible to encounter pages which are
-	 * cached and possibly also mapped. So here we check for that and sort
-	 * them out ahead of the dio. The glock state machine will take care of
-	 * everything else.
-	 *
-	 * If in fact the cached glock state (gl->gl_state) is deferred (CW) in
-	 * the first place, mapping->nr_pages will always be zero.
-	 */
-	if (mapping->nrpages) {
-		loff_t lstart = offset & ~(PAGE_SIZE - 1);
-		loff_t len = iov_iter_count(iter);
-		loff_t end = PAGE_ALIGN(offset + len) - 1;
-
-		rv = 0;
-		if (len == 0)
-			goto out;
-		if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
-			unmap_shared_mapping_range(ip->i_inode.i_mapping, offset, len);
-		rv = filemap_write_and_wait_range(mapping, lstart, end);
-		if (rv)
-			goto out;
-		if (iov_iter_rw(iter) == WRITE)
-			truncate_inode_pages_range(mapping, lstart, end);
-	}
-
-	rv = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
-				  gfs2_get_block_direct, NULL, NULL, 0);
-out:
-	gfs2_glock_dq(&gh);
-out_uninit:
-	gfs2_holder_uninit(&gh);
-	return rv;
-}
-
 /**
  * gfs2_releasepage - free the metadata associated with a page
  * @page: the page that's being released
@@ -1194,7 +1098,7 @@ static const struct address_space_operations gfs2_writeback_aops = {
 	.bmap = gfs2_bmap,
 	.invalidatepage = gfs2_invalidatepage,
 	.releasepage = gfs2_releasepage,
-	.direct_IO = gfs2_direct_IO,
+	.direct_IO = noop_direct_IO,
 	.migratepage = buffer_migrate_page,
 	.is_partially_uptodate = block_is_partially_uptodate,
 	.error_remove_page = generic_error_remove_page,
@@ -1211,7 +1115,7 @@ static const struct address_space_operations gfs2_ordered_aops = {
 	.bmap = gfs2_bmap,
 	.invalidatepage = gfs2_invalidatepage,
 	.releasepage = gfs2_releasepage,
-	.direct_IO = gfs2_direct_IO,
+	.direct_IO = noop_direct_IO,
 	.migratepage = buffer_migrate_page,
 	.is_partially_uptodate = block_is_partially_uptodate,
 	.error_remove_page = generic_error_remove_page,
diff --git a/fs/gfs2/bmap.c b/fs/gfs2/bmap.c
index b70ef98a9637..3978b8cb3178 100644
--- a/fs/gfs2/bmap.c
+++ b/fs/gfs2/bmap.c
@@ -906,6 +906,10 @@ static int gfs2_iomap_get(struct inode *inode, loff_t pos, loff_t length,
 			iomap->length = size - pos;
 	} else if (flags & IOMAP_WRITE) {
 		u64 size;
+
+		if (flags & IOMAP_DIRECT)
+			goto out;
+
 		size = gfs2_alloc_size(inode, mp, len) << inode->i_blkbits;
 		if (size < iomap->length)
 			iomap->length = size;
@@ -1080,7 +1084,14 @@ static int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length,
 
 	trace_gfs2_iomap_start(ip, pos, length, flags);
 	if (flags & IOMAP_WRITE) {
-		ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap);
+		if (flags & IOMAP_DIRECT) {
+			ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
+			release_metapath(&mp);
+			if (iomap->type != IOMAP_MAPPED)
+				ret = -ENOTBLK;
+		} else {
+			ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap);
+		}
 	} else {
 		ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
 		release_metapath(&mp);
@@ -1097,7 +1108,7 @@ static int gfs2_iomap_end(struct inode *inode, loff_t pos, loff_t length,
 	struct gfs2_trans *tr = current->journal_info;
 	struct buffer_head *dibh;
 
-	if (!(flags & IOMAP_WRITE))
+	if ((flags & (IOMAP_WRITE | IOMAP_DIRECT)) != IOMAP_WRITE)
 		return 0;
 
 	dibh = iomap->private;
diff --git a/fs/gfs2/file.c b/fs/gfs2/file.c
index 16dd395479a5..8de92708f18b 100644
--- a/fs/gfs2/file.c
+++ b/fs/gfs2/file.c
@@ -690,6 +690,91 @@ static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
 	return ret ? ret : ret1;
 }
 
+static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to)
+{
+	struct file *file = iocb->ki_filp;
+	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
+	size_t count = iov_iter_count(to);
+	struct gfs2_holder gh;
+	ssize_t ret;
+
+	if (!count)
+		return 0; /* skip atime */
+
+	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
+	ret = gfs2_glock_nq(&gh);
+	if (ret)
+		goto out_uninit;
+
+	/* fall back to buffered I/O for stuffed files */
+	ret = -ENOTBLK;
+	if (gfs2_is_stuffed(ip))
+		goto out;
+
+	ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL);
+
+out:
+	gfs2_glock_dq(&gh);
+out_uninit:
+	gfs2_holder_uninit(&gh);
+	return ret;
+}
+
+static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
+{
+	struct file *file = iocb->ki_filp;
+	struct inode *inode = file->f_mapping->host;
+	struct gfs2_inode *ip = GFS2_I(inode);
+	size_t len = iov_iter_count(from);
+	loff_t offset = iocb->ki_pos;
+	struct gfs2_holder gh;
+	ssize_t ret;
+
+	/*
+	 * Deferred lock, even if its a write, since we do no allocation on
+	 * this path. All we need to change is the atime, and this lock mode
+	 * ensures that other nodes have flushed their buffered read caches
+	 * (i.e. their page cache entries for this inode). We do not,
+	 * unfortunately, have the option of only flushing a range like the
+	 * VFS does.
+	 */
+	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
+	ret = gfs2_glock_nq(&gh);
+	if (ret)
+		goto out_uninit;
+
+	/* Silently fall back to buffered I/O for stuffed files */
+	if (gfs2_is_stuffed(ip))
+		goto out;
+
+	/* Silently fall back to buffered I/O when writing beyond EOF */
+	if (offset + len > i_size_read(&ip->i_inode))
+		goto out;
+
+	ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL);
+
+out:
+	gfs2_glock_dq(&gh);
+out_uninit:
+	gfs2_holder_uninit(&gh);
+	return ret;
+}
+
+static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
+{
+	ssize_t ret;
+
+	if (iocb->ki_flags & IOCB_DIRECT) {
+		ret = gfs2_file_direct_read(iocb, to);
+		if (likely(ret != -ENOTBLK))
+			goto out;
+		iocb->ki_flags &= ~IOCB_DIRECT;
+	}
+	ret = generic_file_read_iter(iocb, to);
+out:
+	return ret;
+}
+
 /**
  * gfs2_file_write_iter - Perform a write to a file
  * @iocb: The io context
@@ -707,7 +792,7 @@ static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 	struct file *file = iocb->ki_filp;
 	struct inode *inode = file_inode(file);
 	struct gfs2_inode *ip = GFS2_I(inode);
-	ssize_t ret;
+	ssize_t written = 0, ret;
 
 	ret = gfs2_rsqa_alloc(ip);
 	if (ret)
@@ -724,9 +809,6 @@ static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 		gfs2_glock_dq_uninit(&gh);
 	}
 
-	if (iocb->ki_flags & IOCB_DIRECT)
-		return generic_file_write_iter(iocb, from);
-
 	inode_lock(inode);
 	ret = generic_write_checks(iocb, from);
 	if (ret <= 0)
@@ -743,19 +825,55 @@ static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 	if (ret)
 		goto out2;
 
-	ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
+	if (iocb->ki_flags & IOCB_DIRECT) {
+		struct address_space *mapping = file->f_mapping;
+		loff_t pos, endbyte;
+		ssize_t buffered;
+
+		written = gfs2_file_direct_write(iocb, from);
+		if (written < 0 || !iov_iter_count(from))
+			goto out2;
+
+		ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
+		if (unlikely(ret < 0))
+			goto out2;
+		buffered = ret;
+
+		/*
+		 * We need to ensure that the page cache pages are written to
+		 * disk and invalidated to preserve the expected O_DIRECT
+		 * semantics.
+		 */
+		pos = iocb->ki_pos;
+		endbyte = pos + buffered - 1;
+		ret = filemap_write_and_wait_range(mapping, pos, endbyte);
+		if (!ret) {
+			iocb->ki_pos += buffered;
+			written += buffered;
+			invalidate_mapping_pages(mapping,
+						 pos >> PAGE_SHIFT,
+						 endbyte >> PAGE_SHIFT);
+		} else {
+			/*
+			 * We don't know how much we wrote, so just return
+			 * the number of bytes which were direct-written
+			 */
+		}
+	} else {
+		ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
+		if (likely(ret > 0))
+			iocb->ki_pos += ret;
+	}
 
 out2:
 	current->backing_dev_info = NULL;
 out:
 	inode_unlock(inode);
 	if (likely(ret > 0)) {
-		iocb->ki_pos += ret;
-
 		/* Handle various SYNC-type writes */
 		ret = generic_write_sync(iocb, ret);
 	}
-	return ret;
+	return written ? written : ret;
 }
 
 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
@@ -1157,7 +1275,7 @@ static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
 
 const struct file_operations gfs2_file_fops = {
 	.llseek		= gfs2_llseek,
-	.read_iter	= generic_file_read_iter,
+	.read_iter	= gfs2_file_read_iter,
 	.write_iter	= gfs2_file_write_iter,
 	.unlocked_ioctl	= gfs2_ioctl,
 	.mmap		= gfs2_mmap,
@@ -1187,7 +1305,7 @@ const struct file_operations gfs2_dir_fops = {
 
 const struct file_operations gfs2_file_fops_nolock = {
 	.llseek		= gfs2_llseek,
-	.read_iter	= generic_file_read_iter,
+	.read_iter	= gfs2_file_read_iter,
 	.write_iter	= gfs2_file_write_iter,
 	.unlocked_ioctl	= gfs2_ioctl,
 	.mmap		= gfs2_mmap,
-- 
2.17.1

[Cluster-devel] [PATCH v9 4/5] gfs2: iomap direct I/O support

[Andreas Gruenbacher]

The page unmapping previously done in gfs2_direct_IO is now done
generically in iomap_dio_rw.

Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
---
 fs/gfs2/aops.c | 100 +----------------------------------
 fs/gfs2/bmap.c |  15 +++++-
 fs/gfs2/file.c | 138 +++++++++++++++++++++++++++++++++++++++++++++----
 3 files changed, 143 insertions(+), 110 deletions(-)

diff --git a/fs/gfs2/aops.c b/fs/gfs2/aops.c
index ecfbca9c88ff..1054cc4a96db 100644
--- a/fs/gfs2/aops.c
+++ b/fs/gfs2/aops.c
@@ -84,12 +84,6 @@ static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
 	return 0;
 }
 
-static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
-				 struct buffer_head *bh_result, int create)
-{
-	return gfs2_block_map(inode, lblock, bh_result, 0);
-}
-
 /**
  * gfs2_writepage_common - Common bits of writepage
  * @page: The page to be written
@@ -1024,96 +1018,6 @@ static void gfs2_invalidatepage(struct page *page, unsigned int offset,
 		try_to_release_page(page, 0);
 }
 
-/**
- * gfs2_ok_for_dio - check that dio is valid on this file
- * @ip: The inode
- * @offset: The offset at which we are reading or writing
- *
- * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
- *          1 (to accept the i/o request)
- */
-static int gfs2_ok_for_dio(struct gfs2_inode *ip, loff_t offset)
-{
-	/*
-	 * Should we return an error here? I can't see that O_DIRECT for
-	 * a stuffed file makes any sense. For now we'll silently fall
-	 * back to buffered I/O
-	 */
-	if (gfs2_is_stuffed(ip))
-		return 0;
-
-	if (offset >= i_size_read(&ip->i_inode))
-		return 0;
-	return 1;
-}
-
-
-
-static ssize_t gfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
-{
-	struct file *file = iocb->ki_filp;
-	struct inode *inode = file->f_mapping->host;
-	struct address_space *mapping = inode->i_mapping;
-	struct gfs2_inode *ip = GFS2_I(inode);
-	loff_t offset = iocb->ki_pos;
-	struct gfs2_holder gh;
-	int rv;
-
-	/*
-	 * Deferred lock, even if its a write, since we do no allocation
-	 * on this path. All we need change is atime, and this lock mode
-	 * ensures that other nodes have flushed their buffered read caches
-	 * (i.e. their page cache entries for this inode). We do not,
-	 * unfortunately have the option of only flushing a range like
-	 * the VFS does.
-	 */
-	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
-	rv = gfs2_glock_nq(&gh);
-	if (rv)
-		goto out_uninit;
-	rv = gfs2_ok_for_dio(ip, offset);
-	if (rv != 1)
-		goto out; /* dio not valid, fall back to buffered i/o */
-
-	/*
-	 * Now since we are holding a deferred (CW) lock at this point, you
-	 * might be wondering why this is ever needed. There is a case however
-	 * where we've granted a deferred local lock against a cached exclusive
-	 * glock. That is ok provided all granted local locks are deferred, but
-	 * it also means that it is possible to encounter pages which are
-	 * cached and possibly also mapped. So here we check for that and sort
-	 * them out ahead of the dio. The glock state machine will take care of
-	 * everything else.
-	 *
-	 * If in fact the cached glock state (gl->gl_state) is deferred (CW) in
-	 * the first place, mapping->nr_pages will always be zero.
-	 */
-	if (mapping->nrpages) {
-		loff_t lstart = offset & ~(PAGE_SIZE - 1);
-		loff_t len = iov_iter_count(iter);
-		loff_t end = PAGE_ALIGN(offset + len) - 1;
-
-		rv = 0;
-		if (len == 0)
-			goto out;
-		if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags))
-			unmap_shared_mapping_range(ip->i_inode.i_mapping, offset, len);
-		rv = filemap_write_and_wait_range(mapping, lstart, end);
-		if (rv)
-			goto out;
-		if (iov_iter_rw(iter) == WRITE)
-			truncate_inode_pages_range(mapping, lstart, end);
-	}
-
-	rv = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter,
-				  gfs2_get_block_direct, NULL, NULL, 0);
-out:
-	gfs2_glock_dq(&gh);
-out_uninit:
-	gfs2_holder_uninit(&gh);
-	return rv;
-}
-
 /**
  * gfs2_releasepage - free the metadata associated with a page
  * @page: the page that's being released
@@ -1194,7 +1098,7 @@ static const struct address_space_operations gfs2_writeback_aops = {
 	.bmap = gfs2_bmap,
 	.invalidatepage = gfs2_invalidatepage,
 	.releasepage = gfs2_releasepage,
-	.direct_IO = gfs2_direct_IO,
+	.direct_IO = noop_direct_IO,
 	.migratepage = buffer_migrate_page,
 	.is_partially_uptodate = block_is_partially_uptodate,
 	.error_remove_page = generic_error_remove_page,
@@ -1211,7 +1115,7 @@ static const struct address_space_operations gfs2_ordered_aops = {
 	.bmap = gfs2_bmap,
 	.invalidatepage = gfs2_invalidatepage,
 	.releasepage = gfs2_releasepage,
-	.direct_IO = gfs2_direct_IO,
+	.direct_IO = noop_direct_IO,
 	.migratepage = buffer_migrate_page,
 	.is_partially_uptodate = block_is_partially_uptodate,
 	.error_remove_page = generic_error_remove_page,
diff --git a/fs/gfs2/bmap.c b/fs/gfs2/bmap.c
index b70ef98a9637..3978b8cb3178 100644
--- a/fs/gfs2/bmap.c
+++ b/fs/gfs2/bmap.c
@@ -906,6 +906,10 @@ static int gfs2_iomap_get(struct inode *inode, loff_t pos, loff_t length,
 			iomap->length = size - pos;
 	} else if (flags & IOMAP_WRITE) {
 		u64 size;
+
+		if (flags & IOMAP_DIRECT)
+			goto out;
+
 		size = gfs2_alloc_size(inode, mp, len) << inode->i_blkbits;
 		if (size < iomap->length)
 			iomap->length = size;
@@ -1080,7 +1084,14 @@ static int gfs2_iomap_begin(struct inode *inode, loff_t pos, loff_t length,
 
 	trace_gfs2_iomap_start(ip, pos, length, flags);
 	if (flags & IOMAP_WRITE) {
-		ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap);
+		if (flags & IOMAP_DIRECT) {
+			ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
+			release_metapath(&mp);
+			if (iomap->type != IOMAP_MAPPED)
+				ret = -ENOTBLK;
+		} else {
+			ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap);
+		}
 	} else {
 		ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
 		release_metapath(&mp);
@@ -1097,7 +1108,7 @@ static int gfs2_iomap_end(struct inode *inode, loff_t pos, loff_t length,
 	struct gfs2_trans *tr = current->journal_info;
 	struct buffer_head *dibh;
 
-	if (!(flags & IOMAP_WRITE))
+	if ((flags & (IOMAP_WRITE | IOMAP_DIRECT)) != IOMAP_WRITE)
 		return 0;
 
 	dibh = iomap->private;
diff --git a/fs/gfs2/file.c b/fs/gfs2/file.c
index 16dd395479a5..8de92708f18b 100644
--- a/fs/gfs2/file.c
+++ b/fs/gfs2/file.c
@@ -690,6 +690,91 @@ static int gfs2_fsync(struct file *file, loff_t start, loff_t end,
 	return ret ? ret : ret1;
 }
 
+static ssize_t gfs2_file_direct_read(struct kiocb *iocb, struct iov_iter *to)
+{
+	struct file *file = iocb->ki_filp;
+	struct gfs2_inode *ip = GFS2_I(file->f_mapping->host);
+	size_t count = iov_iter_count(to);
+	struct gfs2_holder gh;
+	ssize_t ret;
+
+	if (!count)
+		return 0; /* skip atime */
+
+	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
+	ret = gfs2_glock_nq(&gh);
+	if (ret)
+		goto out_uninit;
+
+	/* fall back to buffered I/O for stuffed files */
+	ret = -ENOTBLK;
+	if (gfs2_is_stuffed(ip))
+		goto out;
+
+	ret = iomap_dio_rw(iocb, to, &gfs2_iomap_ops, NULL);
+
+out:
+	gfs2_glock_dq(&gh);
+out_uninit:
+	gfs2_holder_uninit(&gh);
+	return ret;
+}
+
+static ssize_t gfs2_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
+{
+	struct file *file = iocb->ki_filp;
+	struct inode *inode = file->f_mapping->host;
+	struct gfs2_inode *ip = GFS2_I(inode);
+	size_t len = iov_iter_count(from);
+	loff_t offset = iocb->ki_pos;
+	struct gfs2_holder gh;
+	ssize_t ret;
+
+	/*
+	 * Deferred lock, even if its a write, since we do no allocation on
+	 * this path. All we need to change is the atime, and this lock mode
+	 * ensures that other nodes have flushed their buffered read caches
+	 * (i.e. their page cache entries for this inode). We do not,
+	 * unfortunately, have the option of only flushing a range like the
+	 * VFS does.
+	 */
+	gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh);
+	ret = gfs2_glock_nq(&gh);
+	if (ret)
+		goto out_uninit;
+
+	/* Silently fall back to buffered I/O for stuffed files */
+	if (gfs2_is_stuffed(ip))
+		goto out;
+
+	/* Silently fall back to buffered I/O when writing beyond EOF */
+	if (offset + len > i_size_read(&ip->i_inode))
+		goto out;
+
+	ret = iomap_dio_rw(iocb, from, &gfs2_iomap_ops, NULL);
+
+out:
+	gfs2_glock_dq(&gh);
+out_uninit:
+	gfs2_holder_uninit(&gh);
+	return ret;
+}
+
+static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
+{
+	ssize_t ret;
+
+	if (iocb->ki_flags & IOCB_DIRECT) {
+		ret = gfs2_file_direct_read(iocb, to);
+		if (likely(ret != -ENOTBLK))
+			goto out;
+		iocb->ki_flags &= ~IOCB_DIRECT;
+	}
+	ret = generic_file_read_iter(iocb, to);
+out:
+	return ret;
+}
+
 /**
  * gfs2_file_write_iter - Perform a write to a file
  * @iocb: The io context
@@ -707,7 +792,7 @@ static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 	struct file *file = iocb->ki_filp;
 	struct inode *inode = file_inode(file);
 	struct gfs2_inode *ip = GFS2_I(inode);
-	ssize_t ret;
+	ssize_t written = 0, ret;
 
 	ret = gfs2_rsqa_alloc(ip);
 	if (ret)
@@ -724,9 +809,6 @@ static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 		gfs2_glock_dq_uninit(&gh);
 	}
 
-	if (iocb->ki_flags & IOCB_DIRECT)
-		return generic_file_write_iter(iocb, from);
-
 	inode_lock(inode);
 	ret = generic_write_checks(iocb, from);
 	if (ret <= 0)
@@ -743,19 +825,55 @@ static ssize_t gfs2_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 	if (ret)
 		goto out2;
 
-	ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
+	if (iocb->ki_flags & IOCB_DIRECT) {
+		struct address_space *mapping = file->f_mapping;
+		loff_t pos, endbyte;
+		ssize_t buffered;
+
+		written = gfs2_file_direct_write(iocb, from);
+		if (written < 0 || !iov_iter_count(from))
+			goto out2;
+
+		ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
+		if (unlikely(ret < 0))
+			goto out2;
+		buffered = ret;
+
+		/*
+		 * We need to ensure that the page cache pages are written to
+		 * disk and invalidated to preserve the expected O_DIRECT
+		 * semantics.
+		 */
+		pos = iocb->ki_pos;
+		endbyte = pos + buffered - 1;
+		ret = filemap_write_and_wait_range(mapping, pos, endbyte);
+		if (!ret) {
+			iocb->ki_pos += buffered;
+			written += buffered;
+			invalidate_mapping_pages(mapping,
+						 pos >> PAGE_SHIFT,
+						 endbyte >> PAGE_SHIFT);
+		} else {
+			/*
+			 * We don't know how much we wrote, so just return
+			 * the number of bytes which were direct-written
+			 */
+		}
+	} else {
+		ret = iomap_file_buffered_write(iocb, from, &gfs2_iomap_ops);
+		if (likely(ret > 0))
+			iocb->ki_pos += ret;
+	}
 
 out2:
 	current->backing_dev_info = NULL;
 out:
 	inode_unlock(inode);
 	if (likely(ret > 0)) {
-		iocb->ki_pos += ret;
-
 		/* Handle various SYNC-type writes */
 		ret = generic_write_sync(iocb, ret);
 	}
-	return ret;
+	return written ? written : ret;
 }
 
 static int fallocate_chunk(struct inode *inode, loff_t offset, loff_t len,
@@ -1157,7 +1275,7 @@ static int gfs2_flock(struct file *file, int cmd, struct file_lock *fl)
 
 const struct file_operations gfs2_file_fops = {
 	.llseek		= gfs2_llseek,
-	.read_iter	= generic_file_read_iter,
+	.read_iter	= gfs2_file_read_iter,
 	.write_iter	= gfs2_file_write_iter,
 	.unlocked_ioctl	= gfs2_ioctl,
 	.mmap		= gfs2_mmap,
@@ -1187,7 +1305,7 @@ const struct file_operations gfs2_dir_fops = {
 
 const struct file_operations gfs2_file_fops_nolock = {
 	.llseek		= gfs2_llseek,
-	.read_iter	= generic_file_read_iter,
+	.read_iter	= gfs2_file_read_iter,
 	.write_iter	= gfs2_file_write_iter,
 	.unlocked_ioctl	= gfs2_ioctl,
 	.mmap		= gfs2_mmap,
-- 
2.17.1

Re: [PATCH v9 4/5] gfs2: iomap direct I/O support

[Christoph Hellwig]

>  	} else if (flags & IOMAP_WRITE) {
>  		u64 size;
> +
> +		if (flags & IOMAP_DIRECT)
> +			goto out;
> +

Maybe add a comment here on why you don't allow block allocations for
direct I/O.

> +		if (flags & IOMAP_DIRECT) {
> +			ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
> +			release_metapath(&mp);
> +			if (iomap->type != IOMAP_MAPPED)
> +				ret = -ENOTBLK;
> +		} else {
> +			ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap);
> +		}

A couple too long lines.

>  	} else {
>  		ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
>  		release_metapath(&mp);

But shouldn't the direct I/O code try to reuse this part anyway?

E.g. something like:

	if ((flags & (IOMAP_WRITE | IOMAP_DIRECT)) == IOMAP_WRITE)) {
		ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap);
	} else {
 		ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
  		release_metapath(&mp);
		if ((flags & IOMAP_WRITE) && iomap->type != IOMAP_MAPPED)
			ret = -ENOTBLK;

> +	/* fall back to buffered I/O for stuffed files */
> +	ret = -ENOTBLK;
> +	if (gfs2_is_stuffed(ip))
> +		goto out;

I think we can handle stuffed files in the direct I/O code trivially
by copying out the inline data in the iomap.  It would be great to
just handle this instead of adding fallbacks.

> +	/* Silently fall back to buffered I/O for stuffed files */
> +	if (gfs2_is_stuffed(ip))
> +		goto out;

Same here.

> +static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
> +{
> +	ssize_t ret;
> +
> +	if (iocb->ki_flags & IOCB_DIRECT) {
> +		ret = gfs2_file_direct_read(iocb, to);
> +		if (likely(ret != -ENOTBLK))
> +			goto out;

			return ret;

> +		iocb->ki_flags &= ~IOCB_DIRECT;
> +	}
> +	ret = generic_file_read_iter(iocb, to);
> +out:
> +	return ret;

	return generic_file_read_iter(iocb, to);

[Cluster-devel] [PATCH v9 4/5] gfs2: iomap direct I/O support

[Christoph Hellwig]

>  	} else if (flags & IOMAP_WRITE) {
>  		u64 size;
> +
> +		if (flags & IOMAP_DIRECT)
> +			goto out;
> +

Maybe add a comment here on why you don't allow block allocations for
direct I/O.

> +		if (flags & IOMAP_DIRECT) {
> +			ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
> +			release_metapath(&mp);
> +			if (iomap->type != IOMAP_MAPPED)
> +				ret = -ENOTBLK;
> +		} else {
> +			ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap);
> +		}

A couple too long lines.

>  	} else {
>  		ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
>  		release_metapath(&mp);

But shouldn't the direct I/O code try to reuse this part anyway?

E.g. something like:

	if ((flags & (IOMAP_WRITE | IOMAP_DIRECT)) == IOMAP_WRITE)) {
		ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap);
	} else {
 		ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
  		release_metapath(&mp);
		if ((flags & IOMAP_WRITE) && iomap->type != IOMAP_MAPPED)
			ret = -ENOTBLK;

> +	/* fall back to buffered I/O for stuffed files */
> +	ret = -ENOTBLK;
> +	if (gfs2_is_stuffed(ip))
> +		goto out;

I think we can handle stuffed files in the direct I/O code trivially
by copying out the inline data in the iomap.  It would be great to
just handle this instead of adding fallbacks.

> +	/* Silently fall back to buffered I/O for stuffed files */
> +	if (gfs2_is_stuffed(ip))
> +		goto out;

Same here.

> +static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
> +{
> +	ssize_t ret;
> +
> +	if (iocb->ki_flags & IOCB_DIRECT) {
> +		ret = gfs2_file_direct_read(iocb, to);
> +		if (likely(ret != -ENOTBLK))
> +			goto out;

			return ret;

> +		iocb->ki_flags &= ~IOCB_DIRECT;
> +	}
> +	ret = generic_file_read_iter(iocb, to);
> +out:
> +	return ret;

	return generic_file_read_iter(iocb, to);

Re: [PATCH v9 4/5] gfs2: iomap direct I/O support

[Andreas Gruenbacher]

On 19 June 2018 at 08:42, Christoph Hellwig <hch@lst.de> wrote:
>>       } else if (flags & IOMAP_WRITE) {
>>               u64 size;
>> +
>> +             if (flags & IOMAP_DIRECT)
>> +                     goto out;
>> +
>
> Maybe add a comment here on why you don't allow block allocations for
> direct I/O.
>
>> +             if (flags & IOMAP_DIRECT) {
>> +                     ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
>> +                     release_metapath(&mp);
>> +                     if (iomap->type != IOMAP_MAPPED)
>> +                             ret = -ENOTBLK;
>> +             } else {
>> +                     ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap);
>> +             }
>
> A couple too long lines.
>
>>       } else {
>>               ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
>>               release_metapath(&mp);
>
> But shouldn't the direct I/O code try to reuse this part anyway?
>
> E.g. something like:
>
>         if ((flags & (IOMAP_WRITE | IOMAP_DIRECT)) == IOMAP_WRITE)) {
>                 ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap);
>         } else {
>                 ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
>                 release_metapath(&mp);
>                 if ((flags & IOMAP_WRITE) && iomap->type != IOMAP_MAPPED)
>                         ret = -ENOTBLK;
>
>> +     /* fall back to buffered I/O for stuffed files */
>> +     ret = -ENOTBLK;
>> +     if (gfs2_is_stuffed(ip))
>> +             goto out;
>
> I think we can handle stuffed files in the direct I/O code trivially
> by copying out the inline data in the iomap.  It would be great to
> just handle this instead of adding fallbacks.

On gfs2, direct I/O uses a "deferred lock", which is a special kind of
shared lock. We can probably do inline reads under it, but not inline
writes, so falling back to buffered I/O seems inevitable in that case
at least.

>> +     /* Silently fall back to buffered I/O for stuffed files */
>> +     if (gfs2_is_stuffed(ip))
>> +             goto out;
>
> Same here.
>
>> +static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
>> +{
>> +     ssize_t ret;
>> +
>> +     if (iocb->ki_flags & IOCB_DIRECT) {
>> +             ret = gfs2_file_direct_read(iocb, to);
>> +             if (likely(ret != -ENOTBLK))
>> +                     goto out;
>
>                         return ret;
>
>> +             iocb->ki_flags &= ~IOCB_DIRECT;
>> +     }
>> +     ret = generic_file_read_iter(iocb, to);
>> +out:
>> +     return ret;
>
>         return generic_file_read_iter(iocb, to);

Andreas

[Cluster-devel] [PATCH v9 4/5] gfs2: iomap direct I/O support

[Andreas Gruenbacher]

On 19 June 2018 at 08:42, Christoph Hellwig <hch@lst.de> wrote:
>>       } else if (flags & IOMAP_WRITE) {
>>               u64 size;
>> +
>> +             if (flags & IOMAP_DIRECT)
>> +                     goto out;
>> +
>
> Maybe add a comment here on why you don't allow block allocations for
> direct I/O.
>
>> +             if (flags & IOMAP_DIRECT) {
>> +                     ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
>> +                     release_metapath(&mp);
>> +                     if (iomap->type != IOMAP_MAPPED)
>> +                             ret = -ENOTBLK;
>> +             } else {
>> +                     ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap);
>> +             }
>
> A couple too long lines.
>
>>       } else {
>>               ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
>>               release_metapath(&mp);
>
> But shouldn't the direct I/O code try to reuse this part anyway?
>
> E.g. something like:
>
>         if ((flags & (IOMAP_WRITE | IOMAP_DIRECT)) == IOMAP_WRITE)) {
>                 ret = gfs2_iomap_begin_write(inode, pos, length, flags, iomap);
>         } else {
>                 ret = gfs2_iomap_get(inode, pos, length, flags, iomap, &mp);
>                 release_metapath(&mp);
>                 if ((flags & IOMAP_WRITE) && iomap->type != IOMAP_MAPPED)
>                         ret = -ENOTBLK;
>
>> +     /* fall back to buffered I/O for stuffed files */
>> +     ret = -ENOTBLK;
>> +     if (gfs2_is_stuffed(ip))
>> +             goto out;
>
> I think we can handle stuffed files in the direct I/O code trivially
> by copying out the inline data in the iomap.  It would be great to
> just handle this instead of adding fallbacks.

On gfs2, direct I/O uses a "deferred lock", which is a special kind of
shared lock. We can probably do inline reads under it, but not inline
writes, so falling back to buffered I/O seems inevitable in that case
at least.

>> +     /* Silently fall back to buffered I/O for stuffed files */
>> +     if (gfs2_is_stuffed(ip))
>> +             goto out;
>
> Same here.
>
>> +static ssize_t gfs2_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
>> +{
>> +     ssize_t ret;
>> +
>> +     if (iocb->ki_flags & IOCB_DIRECT) {
>> +             ret = gfs2_file_direct_read(iocb, to);
>> +             if (likely(ret != -ENOTBLK))
>> +                     goto out;
>
>                         return ret;
>
>> +             iocb->ki_flags &= ~IOCB_DIRECT;
>> +     }
>> +     ret = generic_file_read_iter(iocb, to);
>> +out:
>> +     return ret;
>
>         return generic_file_read_iter(iocb, to);

Andreas

[PATCH v9 5/5] gfs2: Remove gfs2_write_{begin,end}

[Andreas Gruenbacher]

Now that generic_file_write_iter is no longer used, there are no
remaining users of these address space operations.

Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
---
 fs/gfs2/aops.c | 213 -------------------------------------------------
 1 file changed, 213 deletions(-)

diff --git a/fs/gfs2/aops.c b/fs/gfs2/aops.c
index 1054cc4a96db..cc80fd71f3dd 100644
--- a/fs/gfs2/aops.c
+++ b/fs/gfs2/aops.c
@@ -639,135 +639,6 @@ static int gfs2_readpages(struct file *file, struct address_space *mapping,
 	return ret;
 }
 
-/**
- * gfs2_write_begin - Begin to write to a file
- * @file: The file to write to
- * @mapping: The mapping in which to write
- * @pos: The file offset at which to start writing
- * @len: Length of the write
- * @flags: Various flags
- * @pagep: Pointer to return the page
- * @fsdata: Pointer to return fs data (unused by GFS2)
- *
- * Returns: errno
- */
-
-static int gfs2_write_begin(struct file *file, struct address_space *mapping,
-			    loff_t pos, unsigned len, unsigned flags,
-			    struct page **pagep, void **fsdata)
-{
-	struct gfs2_inode *ip = GFS2_I(mapping->host);
-	struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
-	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
-	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
-	unsigned requested = 0;
-	int alloc_required;
-	int error = 0;
-	pgoff_t index = pos >> PAGE_SHIFT;
-	unsigned from = pos & (PAGE_SIZE - 1);
-	struct page *page;
-
-	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
-	error = gfs2_glock_nq(&ip->i_gh);
-	if (unlikely(error))
-		goto out_uninit;
-	if (&ip->i_inode == sdp->sd_rindex) {
-		error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
-					   GL_NOCACHE, &m_ip->i_gh);
-		if (unlikely(error)) {
-			gfs2_glock_dq(&ip->i_gh);
-			goto out_uninit;
-		}
-	}
-
-	alloc_required = gfs2_write_alloc_required(ip, pos, len);
-
-	if (alloc_required || gfs2_is_jdata(ip))
-		gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
-
-	if (alloc_required) {
-		struct gfs2_alloc_parms ap = { .aflags = 0, };
-		requested = data_blocks + ind_blocks;
-		ap.target = requested;
-		error = gfs2_quota_lock_check(ip, &ap);
-		if (error)
-			goto out_unlock;
-
-		error = gfs2_inplace_reserve(ip, &ap);
-		if (error)
-			goto out_qunlock;
-	}
-
-	rblocks = RES_DINODE + ind_blocks;
-	if (gfs2_is_jdata(ip))
-		rblocks += data_blocks ? data_blocks : 1;
-	if (ind_blocks || data_blocks)
-		rblocks += RES_STATFS + RES_QUOTA;
-	if (&ip->i_inode == sdp->sd_rindex)
-		rblocks += 2 * RES_STATFS;
-	if (alloc_required)
-		rblocks += gfs2_rg_blocks(ip, requested);
-
-	error = gfs2_trans_begin(sdp, rblocks,
-				 PAGE_SIZE/sdp->sd_sb.sb_bsize);
-	if (error)
-		goto out_trans_fail;
-
-	error = -ENOMEM;
-	flags |= AOP_FLAG_NOFS;
-	page = grab_cache_page_write_begin(mapping, index, flags);
-	*pagep = page;
-	if (unlikely(!page))
-		goto out_endtrans;
-
-	if (gfs2_is_stuffed(ip)) {
-		error = 0;
-		if (pos + len > gfs2_max_stuffed_size(ip)) {
-			error = gfs2_unstuff_dinode(ip, page);
-			if (error == 0)
-				goto prepare_write;
-		} else if (!PageUptodate(page)) {
-			error = stuffed_readpage(ip, page);
-		}
-		goto out;
-	}
-
-prepare_write:
-	error = __block_write_begin(page, from, len, gfs2_block_map);
-out:
-	if (error == 0)
-		return 0;
-
-	unlock_page(page);
-	put_page(page);
-
-	gfs2_trans_end(sdp);
-	if (alloc_required) {
-		gfs2_inplace_release(ip);
-		if (pos + len > ip->i_inode.i_size)
-			gfs2_trim_blocks(&ip->i_inode);
-	}
-	goto out_qunlock;
-
-out_endtrans:
-	gfs2_trans_end(sdp);
-out_trans_fail:
-	if (alloc_required)
-		gfs2_inplace_release(ip);
-out_qunlock:
-	if (alloc_required)
-		gfs2_quota_unlock(ip);
-out_unlock:
-	if (&ip->i_inode == sdp->sd_rindex) {
-		gfs2_glock_dq(&m_ip->i_gh);
-		gfs2_holder_uninit(&m_ip->i_gh);
-	}
-	gfs2_glock_dq(&ip->i_gh);
-out_uninit:
-	gfs2_holder_uninit(&ip->i_gh);
-	return error;
-}
-
 /**
  * adjust_fs_space - Adjusts the free space available due to gfs2_grow
  * @inode: the rindex inode
@@ -848,84 +719,6 @@ int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
 	return copied;
 }
 
-/**
- * gfs2_write_end
- * @file: The file to write to
- * @mapping: The address space to write to
- * @pos: The file position
- * @len: The length of the data
- * @copied: How much was actually copied by the VFS
- * @page: The page that has been written
- * @fsdata: The fsdata (unused in GFS2)
- *
- * The main write_end function for GFS2. We just put our locking around the VFS
- * provided functions.
- *
- * Returns: copied bytes or errno
- */
-
-static int gfs2_write_end(struct file *file, struct address_space *mapping,
-			  loff_t pos, unsigned len, unsigned copied,
-			  struct page *page, void *fsdata)
-{
-	struct inode *inode = page->mapping->host;
-	struct gfs2_inode *ip = GFS2_I(inode);
-	struct gfs2_sbd *sdp = GFS2_SB(inode);
-	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
-	struct buffer_head *dibh;
-	int ret;
-	struct gfs2_trans *tr = current->journal_info;
-	BUG_ON(!tr);
-
-	BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
-
-	ret = gfs2_meta_inode_buffer(ip, &dibh);
-	if (unlikely(ret))
-		goto out;
-
-	if (gfs2_is_stuffed(ip)) {
-		ret = gfs2_stuffed_write_end(inode, dibh, pos, copied, page);
-		page = NULL;
-		goto out2;
-	}
-
-	if (gfs2_is_jdata(ip))
-		gfs2_page_add_databufs(ip, page, pos & ~PAGE_MASK, len);
-	else
-		gfs2_ordered_add_inode(ip);
-
-	ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
-	page = NULL;
-	if (tr->tr_num_buf_new)
-		__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
-	else
-		gfs2_trans_add_meta(ip->i_gl, dibh);
-
-out2:
-	if (inode == sdp->sd_rindex) {
-		adjust_fs_space(inode);
-		sdp->sd_rindex_uptodate = 0;
-	}
-
-	brelse(dibh);
-out:
-	if (page) {
-		unlock_page(page);
-		put_page(page);
-	}
-	gfs2_trans_end(sdp);
-	gfs2_inplace_release(ip);
-	if (ip->i_qadata && ip->i_qadata->qa_qd_num)
-		gfs2_quota_unlock(ip);
-	if (inode == sdp->sd_rindex) {
-		gfs2_glock_dq(&m_ip->i_gh);
-		gfs2_holder_uninit(&m_ip->i_gh);
-	}
-	gfs2_glock_dq(&ip->i_gh);
-	gfs2_holder_uninit(&ip->i_gh);
-	return ret;
-}
-
 /**
  * jdata_set_page_dirty - Page dirtying function
  * @page: The page to dirty
@@ -1093,8 +886,6 @@ static const struct address_space_operations gfs2_writeback_aops = {
 	.writepages = gfs2_writepages,
 	.readpage = gfs2_readpage,
 	.readpages = gfs2_readpages,
-	.write_begin = gfs2_write_begin,
-	.write_end = gfs2_write_end,
 	.bmap = gfs2_bmap,
 	.invalidatepage = gfs2_invalidatepage,
 	.releasepage = gfs2_releasepage,
@@ -1109,8 +900,6 @@ static const struct address_space_operations gfs2_ordered_aops = {
 	.writepages = gfs2_writepages,
 	.readpage = gfs2_readpage,
 	.readpages = gfs2_readpages,
-	.write_begin = gfs2_write_begin,
-	.write_end = gfs2_write_end,
 	.set_page_dirty = __set_page_dirty_buffers,
 	.bmap = gfs2_bmap,
 	.invalidatepage = gfs2_invalidatepage,
@@ -1126,8 +915,6 @@ static const struct address_space_operations gfs2_jdata_aops = {
 	.writepages = gfs2_jdata_writepages,
 	.readpage = gfs2_readpage,
 	.readpages = gfs2_readpages,
-	.write_begin = gfs2_write_begin,
-	.write_end = gfs2_write_end,
 	.set_page_dirty = jdata_set_page_dirty,
 	.bmap = gfs2_bmap,
 	.invalidatepage = gfs2_invalidatepage,
-- 
2.17.1

[Cluster-devel] [PATCH v9 5/5] gfs2: Remove gfs2_write_{begin, end}

[Andreas Gruenbacher]

Now that generic_file_write_iter is no longer used, there are no
remaining users of these address space operations.

Signed-off-by: Andreas Gruenbacher <agruenba@redhat.com>
---
 fs/gfs2/aops.c | 213 -------------------------------------------------
 1 file changed, 213 deletions(-)

diff --git a/fs/gfs2/aops.c b/fs/gfs2/aops.c
index 1054cc4a96db..cc80fd71f3dd 100644
--- a/fs/gfs2/aops.c
+++ b/fs/gfs2/aops.c
@@ -639,135 +639,6 @@ static int gfs2_readpages(struct file *file, struct address_space *mapping,
 	return ret;
 }
 
-/**
- * gfs2_write_begin - Begin to write to a file
- * @file: The file to write to
- * @mapping: The mapping in which to write
- * @pos: The file offset at which to start writing
- * @len: Length of the write
- * @flags: Various flags
- * @pagep: Pointer to return the page
- * @fsdata: Pointer to return fs data (unused by GFS2)
- *
- * Returns: errno
- */
-
-static int gfs2_write_begin(struct file *file, struct address_space *mapping,
-			    loff_t pos, unsigned len, unsigned flags,
-			    struct page **pagep, void **fsdata)
-{
-	struct gfs2_inode *ip = GFS2_I(mapping->host);
-	struct gfs2_sbd *sdp = GFS2_SB(mapping->host);
-	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
-	unsigned int data_blocks = 0, ind_blocks = 0, rblocks;
-	unsigned requested = 0;
-	int alloc_required;
-	int error = 0;
-	pgoff_t index = pos >> PAGE_SHIFT;
-	unsigned from = pos & (PAGE_SIZE - 1);
-	struct page *page;
-
-	gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh);
-	error = gfs2_glock_nq(&ip->i_gh);
-	if (unlikely(error))
-		goto out_uninit;
-	if (&ip->i_inode == sdp->sd_rindex) {
-		error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE,
-					   GL_NOCACHE, &m_ip->i_gh);
-		if (unlikely(error)) {
-			gfs2_glock_dq(&ip->i_gh);
-			goto out_uninit;
-		}
-	}
-
-	alloc_required = gfs2_write_alloc_required(ip, pos, len);
-
-	if (alloc_required || gfs2_is_jdata(ip))
-		gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks);
-
-	if (alloc_required) {
-		struct gfs2_alloc_parms ap = { .aflags = 0, };
-		requested = data_blocks + ind_blocks;
-		ap.target = requested;
-		error = gfs2_quota_lock_check(ip, &ap);
-		if (error)
-			goto out_unlock;
-
-		error = gfs2_inplace_reserve(ip, &ap);
-		if (error)
-			goto out_qunlock;
-	}
-
-	rblocks = RES_DINODE + ind_blocks;
-	if (gfs2_is_jdata(ip))
-		rblocks += data_blocks ? data_blocks : 1;
-	if (ind_blocks || data_blocks)
-		rblocks += RES_STATFS + RES_QUOTA;
-	if (&ip->i_inode == sdp->sd_rindex)
-		rblocks += 2 * RES_STATFS;
-	if (alloc_required)
-		rblocks += gfs2_rg_blocks(ip, requested);
-
-	error = gfs2_trans_begin(sdp, rblocks,
-				 PAGE_SIZE/sdp->sd_sb.sb_bsize);
-	if (error)
-		goto out_trans_fail;
-
-	error = -ENOMEM;
-	flags |= AOP_FLAG_NOFS;
-	page = grab_cache_page_write_begin(mapping, index, flags);
-	*pagep = page;
-	if (unlikely(!page))
-		goto out_endtrans;
-
-	if (gfs2_is_stuffed(ip)) {
-		error = 0;
-		if (pos + len > gfs2_max_stuffed_size(ip)) {
-			error = gfs2_unstuff_dinode(ip, page);
-			if (error == 0)
-				goto prepare_write;
-		} else if (!PageUptodate(page)) {
-			error = stuffed_readpage(ip, page);
-		}
-		goto out;
-	}
-
-prepare_write:
-	error = __block_write_begin(page, from, len, gfs2_block_map);
-out:
-	if (error == 0)
-		return 0;
-
-	unlock_page(page);
-	put_page(page);
-
-	gfs2_trans_end(sdp);
-	if (alloc_required) {
-		gfs2_inplace_release(ip);
-		if (pos + len > ip->i_inode.i_size)
-			gfs2_trim_blocks(&ip->i_inode);
-	}
-	goto out_qunlock;
-
-out_endtrans:
-	gfs2_trans_end(sdp);
-out_trans_fail:
-	if (alloc_required)
-		gfs2_inplace_release(ip);
-out_qunlock:
-	if (alloc_required)
-		gfs2_quota_unlock(ip);
-out_unlock:
-	if (&ip->i_inode == sdp->sd_rindex) {
-		gfs2_glock_dq(&m_ip->i_gh);
-		gfs2_holder_uninit(&m_ip->i_gh);
-	}
-	gfs2_glock_dq(&ip->i_gh);
-out_uninit:
-	gfs2_holder_uninit(&ip->i_gh);
-	return error;
-}
-
 /**
  * adjust_fs_space - Adjusts the free space available due to gfs2_grow
  * @inode: the rindex inode
@@ -848,84 +719,6 @@ int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh,
 	return copied;
 }
 
-/**
- * gfs2_write_end
- * @file: The file to write to
- * @mapping: The address space to write to
- * @pos: The file position
- * @len: The length of the data
- * @copied: How much was actually copied by the VFS
- * @page: The page that has been written
- * @fsdata: The fsdata (unused in GFS2)
- *
- * The main write_end function for GFS2. We just put our locking around the VFS
- * provided functions.
- *
- * Returns: copied bytes or errno
- */
-
-static int gfs2_write_end(struct file *file, struct address_space *mapping,
-			  loff_t pos, unsigned len, unsigned copied,
-			  struct page *page, void *fsdata)
-{
-	struct inode *inode = page->mapping->host;
-	struct gfs2_inode *ip = GFS2_I(inode);
-	struct gfs2_sbd *sdp = GFS2_SB(inode);
-	struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode);
-	struct buffer_head *dibh;
-	int ret;
-	struct gfs2_trans *tr = current->journal_info;
-	BUG_ON(!tr);
-
-	BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL);
-
-	ret = gfs2_meta_inode_buffer(ip, &dibh);
-	if (unlikely(ret))
-		goto out;
-
-	if (gfs2_is_stuffed(ip)) {
-		ret = gfs2_stuffed_write_end(inode, dibh, pos, copied, page);
-		page = NULL;
-		goto out2;
-	}
-
-	if (gfs2_is_jdata(ip))
-		gfs2_page_add_databufs(ip, page, pos & ~PAGE_MASK, len);
-	else
-		gfs2_ordered_add_inode(ip);
-
-	ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
-	page = NULL;
-	if (tr->tr_num_buf_new)
-		__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
-	else
-		gfs2_trans_add_meta(ip->i_gl, dibh);
-
-out2:
-	if (inode == sdp->sd_rindex) {
-		adjust_fs_space(inode);
-		sdp->sd_rindex_uptodate = 0;
-	}
-
-	brelse(dibh);
-out:
-	if (page) {
-		unlock_page(page);
-		put_page(page);
-	}
-	gfs2_trans_end(sdp);
-	gfs2_inplace_release(ip);
-	if (ip->i_qadata && ip->i_qadata->qa_qd_num)
-		gfs2_quota_unlock(ip);
-	if (inode == sdp->sd_rindex) {
-		gfs2_glock_dq(&m_ip->i_gh);
-		gfs2_holder_uninit(&m_ip->i_gh);
-	}
-	gfs2_glock_dq(&ip->i_gh);
-	gfs2_holder_uninit(&ip->i_gh);
-	return ret;
-}
-
 /**
  * jdata_set_page_dirty - Page dirtying function
  * @page: The page to dirty
@@ -1093,8 +886,6 @@ static const struct address_space_operations gfs2_writeback_aops = {
 	.writepages = gfs2_writepages,
 	.readpage = gfs2_readpage,
 	.readpages = gfs2_readpages,
-	.write_begin = gfs2_write_begin,
-	.write_end = gfs2_write_end,
 	.bmap = gfs2_bmap,
 	.invalidatepage = gfs2_invalidatepage,
 	.releasepage = gfs2_releasepage,
@@ -1109,8 +900,6 @@ static const struct address_space_operations gfs2_ordered_aops = {
 	.writepages = gfs2_writepages,
 	.readpage = gfs2_readpage,
 	.readpages = gfs2_readpages,
-	.write_begin = gfs2_write_begin,
-	.write_end = gfs2_write_end,
 	.set_page_dirty = __set_page_dirty_buffers,
 	.bmap = gfs2_bmap,
 	.invalidatepage = gfs2_invalidatepage,
@@ -1126,8 +915,6 @@ static const struct address_space_operations gfs2_jdata_aops = {
 	.writepages = gfs2_jdata_writepages,
 	.readpage = gfs2_readpage,
 	.readpages = gfs2_readpages,
-	.write_begin = gfs2_write_begin,
-	.write_end = gfs2_write_end,
 	.set_page_dirty = jdata_set_page_dirty,
 	.bmap = gfs2_bmap,
 	.invalidatepage = gfs2_invalidatepage,
-- 
2.17.1