nvdimm.lists.linux.dev archive mirror
 help / color / mirror / Atom feed
* [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path
       [not found] <OSBPR01MB2920A2BCD568364C1363AFA6F4369@OSBPR01MB2920.jpnprd01.prod.outlook.com>
@ 2021-06-15  7:21 ` Shiyang Ruan
  2021-06-24  8:49   ` ruansy.fnst
  0 siblings, 1 reply; 9+ messages in thread
From: Shiyang Ruan @ 2021-06-15  7:21 UTC (permalink / raw)
  To: darrick.wong
  Cc: ruansy.fnst, dan.j.williams, david, djwong, hch, linux-fsdevel,
	linux-kernel, nvdimm, linux-xfs, rgoldwyn, viro, willy

Hi Darrick,

Since other patches looks good, I post this RFC patch singly to hot-fix the
problem in xfs_dax_write_iomap_ops->iomap_end() of v6 that the error code was
ingored. I will split this in two patches(changes in iomap and xfs
respectively) in next formal version if it looks ok.

====

Introduce a new interface called "iomap_post_actor()" in iomap_ops.  And call it
between ->actor() and ->iomap_end().  It is mean to handle the error code
returned from ->actor().  In this patchset, it is used to remap or cancel the
CoW extents according to the error code.

Signed-off-by: Shiyang Ruan <ruansy.fnst@fujitsu.com>
---
 fs/dax.c               | 27 ++++++++++++++++++---------
 fs/iomap/apply.c       |  4 ++++
 fs/xfs/xfs_bmap_util.c |  3 +--
 fs/xfs/xfs_file.c      |  5 +++--
 fs/xfs/xfs_iomap.c     | 33 ++++++++++++++++++++++++++++++++-
 fs/xfs/xfs_iomap.h     | 24 ++++++++++++++++++++++++
 fs/xfs/xfs_iops.c      |  7 +++----
 fs/xfs/xfs_reflink.c   |  3 +--
 include/linux/iomap.h  |  8 ++++++++
 9 files changed, 94 insertions(+), 20 deletions(-)

diff --git a/fs/dax.c b/fs/dax.c
index 93f16210847b..0740c2610b6f 100644
--- a/fs/dax.c
+++ b/fs/dax.c
@@ -1537,7 +1537,7 @@ static vm_fault_t dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
 	struct iomap iomap = { .type = IOMAP_HOLE };
 	struct iomap srcmap = { .type = IOMAP_HOLE };
 	unsigned flags = IOMAP_FAULT;
-	int error;
+	int error, copied = PAGE_SIZE;
 	bool write = vmf->flags & FAULT_FLAG_WRITE;
 	vm_fault_t ret = 0, major = 0;
 	void *entry;
@@ -1598,7 +1598,7 @@ static vm_fault_t dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
 	ret = dax_fault_actor(vmf, pfnp, &xas, &entry, false, flags,
 			      &iomap, &srcmap);
 	if (ret == VM_FAULT_SIGBUS)
-		goto finish_iomap;
+		goto finish_iomap_actor;
 
 	/* read/write MAPPED, CoW UNWRITTEN */
 	if (iomap.flags & IOMAP_F_NEW) {
@@ -1607,10 +1607,16 @@ static vm_fault_t dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
 		major = VM_FAULT_MAJOR;
 	}
 
+ finish_iomap_actor:
+	if (ops->iomap_post_actor) {
+		if (ret & VM_FAULT_ERROR)
+			copied = 0;
+		ops->iomap_post_actor(inode, pos, PMD_SIZE, copied, flags,
+				      &iomap, &srcmap);
+	}
+
 finish_iomap:
 	if (ops->iomap_end) {
-		int copied = PAGE_SIZE;
-
 		if (ret & VM_FAULT_ERROR)
 			copied = 0;
 		/*
@@ -1677,7 +1683,7 @@ static vm_fault_t dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp,
 	pgoff_t max_pgoff;
 	void *entry;
 	loff_t pos;
-	int error;
+	int error, copied = PMD_SIZE;
 
 	/*
 	 * Check whether offset isn't beyond end of file now. Caller is
@@ -1736,12 +1742,15 @@ static vm_fault_t dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp,
 	ret = dax_fault_actor(vmf, pfnp, &xas, &entry, true, flags,
 			      &iomap, &srcmap);
 
+	if (ret == VM_FAULT_FALLBACK)
+		copied = 0;
+	if (ops->iomap_post_actor) {
+		ops->iomap_post_actor(inode, pos, PMD_SIZE, copied, flags,
+				      &iomap, &srcmap);
+	}
+
 finish_iomap:
 	if (ops->iomap_end) {
-		int copied = PMD_SIZE;
-
-		if (ret == VM_FAULT_FALLBACK)
-			copied = 0;
 		/*
 		 * The fault is done by now and there's no way back (other
 		 * thread may be already happily using PMD we have installed).
diff --git a/fs/iomap/apply.c b/fs/iomap/apply.c
index 0493da5286ad..26a54ded184f 100644
--- a/fs/iomap/apply.c
+++ b/fs/iomap/apply.c
@@ -84,6 +84,10 @@ iomap_apply(struct inode *inode, loff_t pos, loff_t length, unsigned flags,
 	written = actor(inode, pos, length, data, &iomap,
 			srcmap.type != IOMAP_HOLE ? &srcmap : &iomap);
 
+	if (ops->iomap_post_actor) {
+		written = ops->iomap_post_actor(inode, pos, length, written,
+						flags, &iomap, &srcmap);
+	}
 out:
 	/*
 	 * Now the data has been copied, commit the range we've copied.  This
diff --git a/fs/xfs/xfs_bmap_util.c b/fs/xfs/xfs_bmap_util.c
index a5e9d7d34023..2a36dc93ff27 100644
--- a/fs/xfs/xfs_bmap_util.c
+++ b/fs/xfs/xfs_bmap_util.c
@@ -965,8 +965,7 @@ xfs_free_file_space(
 		return 0;
 	if (offset + len > XFS_ISIZE(ip))
 		len = XFS_ISIZE(ip) - offset;
-	error = iomap_zero_range(VFS_I(ip), offset, len, NULL,
-			&xfs_buffered_write_iomap_ops);
+	error = xfs_iomap_zero_range(ip, offset, len, NULL);
 	if (error)
 		return error;
 
diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
index 396ef36dcd0a..89406ec6741b 100644
--- a/fs/xfs/xfs_file.c
+++ b/fs/xfs/xfs_file.c
@@ -684,11 +684,12 @@ xfs_file_dax_write(
 	pos = iocb->ki_pos;
 
 	trace_xfs_file_dax_write(iocb, from);
-	ret = dax_iomap_rw(iocb, from, &xfs_direct_write_iomap_ops);
+	ret = dax_iomap_rw(iocb, from, &xfs_dax_write_iomap_ops);
 	if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
 		i_size_write(inode, iocb->ki_pos);
 		error = xfs_setfilesize(ip, pos, ret);
 	}
+
 out:
 	if (iolock)
 		xfs_iunlock(ip, iolock);
@@ -1309,7 +1310,7 @@ __xfs_filemap_fault(
 
 		ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL,
 				(write_fault && !vmf->cow_page) ?
-				 &xfs_direct_write_iomap_ops :
+				 &xfs_dax_write_iomap_ops :
 				 &xfs_read_iomap_ops);
 		if (ret & VM_FAULT_NEEDDSYNC)
 			ret = dax_finish_sync_fault(vmf, pe_size, pfn);
diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
index d154f42e2dc6..2f322e2f8544 100644
--- a/fs/xfs/xfs_iomap.c
+++ b/fs/xfs/xfs_iomap.c
@@ -761,7 +761,8 @@ xfs_direct_write_iomap_begin(
 
 		/* may drop and re-acquire the ilock */
 		error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
-				&lockmode, flags & IOMAP_DIRECT);
+				&lockmode,
+				(flags & IOMAP_DIRECT) || IS_DAX(inode));
 		if (error)
 			goto out_unlock;
 		if (shared)
@@ -854,6 +855,36 @@ const struct iomap_ops xfs_direct_write_iomap_ops = {
 	.iomap_begin		= xfs_direct_write_iomap_begin,
 };
 
+static int
+xfs_dax_write_iomap_post_actor(
+	struct inode		*inode,
+	loff_t			pos,
+	loff_t			length,
+	ssize_t			written,
+	unsigned int		flags,
+	struct iomap		*iomap,
+	struct iomap		*srcmap)
+{
+	int			error = 0;
+	struct xfs_inode	*ip = XFS_I(inode);
+	bool			cow = xfs_is_cow_inode(ip);
+
+	if (written <= 0) {
+		if (cow)
+			xfs_reflink_cancel_cow_range(ip, pos, length, true);
+		return written;
+	}
+
+	if (cow)
+		error = xfs_reflink_end_cow(ip, pos, written);
+	return error ?: written;
+}
+
+const struct iomap_ops xfs_dax_write_iomap_ops = {
+	.iomap_begin		= xfs_direct_write_iomap_begin,
+	.iomap_post_actor	= xfs_dax_write_iomap_post_actor,
+};
+
 static int
 xfs_buffered_write_iomap_begin(
 	struct inode		*inode,
diff --git a/fs/xfs/xfs_iomap.h b/fs/xfs/xfs_iomap.h
index 7d3703556d0e..fbacf638ab21 100644
--- a/fs/xfs/xfs_iomap.h
+++ b/fs/xfs/xfs_iomap.h
@@ -42,8 +42,32 @@ xfs_aligned_fsb_count(
 
 extern const struct iomap_ops xfs_buffered_write_iomap_ops;
 extern const struct iomap_ops xfs_direct_write_iomap_ops;
+extern const struct iomap_ops xfs_dax_write_iomap_ops;
 extern const struct iomap_ops xfs_read_iomap_ops;
 extern const struct iomap_ops xfs_seek_iomap_ops;
 extern const struct iomap_ops xfs_xattr_iomap_ops;
 
+static inline int
+xfs_iomap_zero_range(
+	struct xfs_inode	*ip,
+	loff_t			offset,
+	loff_t			len,
+	bool			*did_zero)
+{
+	return iomap_zero_range(VFS_I(ip), offset, len, did_zero,
+			IS_DAX(VFS_I(ip)) ? &xfs_dax_write_iomap_ops
+					  : &xfs_buffered_write_iomap_ops);
+}
+
+static inline int
+xfs_iomap_truncate_page(
+	struct xfs_inode	*ip,
+	loff_t			pos,
+	bool			*did_zero)
+{
+	return iomap_truncate_page(VFS_I(ip), pos, did_zero,
+			IS_DAX(VFS_I(ip)) ? &xfs_dax_write_iomap_ops
+					  : &xfs_buffered_write_iomap_ops);
+}
+
 #endif /* __XFS_IOMAP_H__*/
diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c
index dfe24b7f26e5..6d936c3e1a6e 100644
--- a/fs/xfs/xfs_iops.c
+++ b/fs/xfs/xfs_iops.c
@@ -911,8 +911,8 @@ xfs_setattr_size(
 	 */
 	if (newsize > oldsize) {
 		trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
-		error = iomap_zero_range(inode, oldsize, newsize - oldsize,
-				&did_zeroing, &xfs_buffered_write_iomap_ops);
+		error = xfs_iomap_zero_range(ip, oldsize, newsize - oldsize,
+				&did_zeroing);
 	} else {
 		/*
 		 * iomap won't detect a dirty page over an unwritten block (or a
@@ -924,8 +924,7 @@ xfs_setattr_size(
 						     newsize);
 		if (error)
 			return error;
-		error = iomap_truncate_page(inode, newsize, &did_zeroing,
-				&xfs_buffered_write_iomap_ops);
+		error = xfs_iomap_truncate_page(ip, newsize, &did_zeroing);
 	}
 
 	if (error)
diff --git a/fs/xfs/xfs_reflink.c b/fs/xfs/xfs_reflink.c
index d25434f93235..9a780948dbd0 100644
--- a/fs/xfs/xfs_reflink.c
+++ b/fs/xfs/xfs_reflink.c
@@ -1266,8 +1266,7 @@ xfs_reflink_zero_posteof(
 		return 0;
 
 	trace_xfs_zero_eof(ip, isize, pos - isize);
-	return iomap_zero_range(VFS_I(ip), isize, pos - isize, NULL,
-			&xfs_buffered_write_iomap_ops);
+	return xfs_iomap_zero_range(ip, isize, pos - isize, NULL);
 }
 
 /*
diff --git a/include/linux/iomap.h b/include/linux/iomap.h
index 95562f863ad0..58f2e1c78018 100644
--- a/include/linux/iomap.h
+++ b/include/linux/iomap.h
@@ -135,6 +135,14 @@ struct iomap_ops {
 			unsigned flags, struct iomap *iomap,
 			struct iomap *srcmap);
 
+	/*
+	 * Handle the error code from actor(). Do the finishing jobs for extra
+	 * operations, such as CoW, according to whether written is negative.
+	 */
+	int (*iomap_post_actor)(struct inode *inode, loff_t pos, loff_t length,
+			ssize_t written, unsigned flags, struct iomap *iomap,
+			struct iomap *srcmap);
+
 	/*
 	 * Commit and/or unreserve space previous allocated using iomap_begin.
 	 * Written indicates the length of the successful write operation which
-- 
2.31.1




^ permalink raw reply related	[flat|nested] 9+ messages in thread

* RE: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path
  2021-06-15  7:21 ` [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path Shiyang Ruan
@ 2021-06-24  8:49   ` ruansy.fnst
  2021-06-25 22:18     ` Darrick J. Wong
  0 siblings, 1 reply; 9+ messages in thread
From: ruansy.fnst @ 2021-06-24  8:49 UTC (permalink / raw)
  To: Darrick J. Wong
  Cc: dan.j.williams, david, djwong, hch, linux-fsdevel, linux-kernel,
	nvdimm, linux-xfs, rgoldwyn, viro, willy

Hi Darrick,

Do you have any comment on this?


--
Thanks,
Ruan.

> -----Original Message-----
> From: Shiyang Ruan <ruansy.fnst@fujitsu.com>
> Subject: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path
> 
> Hi Darrick,
> 
> Since other patches looks good, I post this RFC patch singly to hot-fix the
> problem in xfs_dax_write_iomap_ops->iomap_end() of v6 that the error code
> was ingored. I will split this in two patches(changes in iomap and xfs
> respectively) in next formal version if it looks ok.
> 
> ====
> 
> Introduce a new interface called "iomap_post_actor()" in iomap_ops.  And call
> it between ->actor() and ->iomap_end().  It is mean to handle the error code
> returned from ->actor().  In this patchset, it is used to remap or cancel the
> CoW extents according to the error code.
> 
> Signed-off-by: Shiyang Ruan <ruansy.fnst@fujitsu.com>
> ---
>  fs/dax.c               | 27 ++++++++++++++++++---------
>  fs/iomap/apply.c       |  4 ++++
>  fs/xfs/xfs_bmap_util.c |  3 +--
>  fs/xfs/xfs_file.c      |  5 +++--
>  fs/xfs/xfs_iomap.c     | 33 ++++++++++++++++++++++++++++++++-
>  fs/xfs/xfs_iomap.h     | 24 ++++++++++++++++++++++++
>  fs/xfs/xfs_iops.c      |  7 +++----
>  fs/xfs/xfs_reflink.c   |  3 +--
>  include/linux/iomap.h  |  8 ++++++++
>  9 files changed, 94 insertions(+), 20 deletions(-)
> 
> diff --git a/fs/dax.c b/fs/dax.c
> index 93f16210847b..0740c2610b6f 100644
> --- a/fs/dax.c
> +++ b/fs/dax.c
> @@ -1537,7 +1537,7 @@ static vm_fault_t dax_iomap_pte_fault(struct
> vm_fault *vmf, pfn_t *pfnp,
>  	struct iomap iomap = { .type = IOMAP_HOLE };
>  	struct iomap srcmap = { .type = IOMAP_HOLE };
>  	unsigned flags = IOMAP_FAULT;
> -	int error;
> +	int error, copied = PAGE_SIZE;
>  	bool write = vmf->flags & FAULT_FLAG_WRITE;
>  	vm_fault_t ret = 0, major = 0;
>  	void *entry;
> @@ -1598,7 +1598,7 @@ static vm_fault_t dax_iomap_pte_fault(struct
> vm_fault *vmf, pfn_t *pfnp,
>  	ret = dax_fault_actor(vmf, pfnp, &xas, &entry, false, flags,
>  			      &iomap, &srcmap);
>  	if (ret == VM_FAULT_SIGBUS)
> -		goto finish_iomap;
> +		goto finish_iomap_actor;
> 
>  	/* read/write MAPPED, CoW UNWRITTEN */
>  	if (iomap.flags & IOMAP_F_NEW) {
> @@ -1607,10 +1607,16 @@ static vm_fault_t dax_iomap_pte_fault(struct
> vm_fault *vmf, pfn_t *pfnp,
>  		major = VM_FAULT_MAJOR;
>  	}
> 
> + finish_iomap_actor:
> +	if (ops->iomap_post_actor) {
> +		if (ret & VM_FAULT_ERROR)
> +			copied = 0;
> +		ops->iomap_post_actor(inode, pos, PMD_SIZE, copied, flags,
> +				      &iomap, &srcmap);
> +	}
> +
>  finish_iomap:
>  	if (ops->iomap_end) {
> -		int copied = PAGE_SIZE;
> -
>  		if (ret & VM_FAULT_ERROR)
>  			copied = 0;
>  		/*
> @@ -1677,7 +1683,7 @@ static vm_fault_t dax_iomap_pmd_fault(struct
> vm_fault *vmf, pfn_t *pfnp,
>  	pgoff_t max_pgoff;
>  	void *entry;
>  	loff_t pos;
> -	int error;
> +	int error, copied = PMD_SIZE;
> 
>  	/*
>  	 * Check whether offset isn't beyond end of file now. Caller is @@ -1736,12
> +1742,15 @@ static vm_fault_t dax_iomap_pmd_fault(struct vm_fault *vmf,
> pfn_t *pfnp,
>  	ret = dax_fault_actor(vmf, pfnp, &xas, &entry, true, flags,
>  			      &iomap, &srcmap);
> 
> +	if (ret == VM_FAULT_FALLBACK)
> +		copied = 0;
> +	if (ops->iomap_post_actor) {
> +		ops->iomap_post_actor(inode, pos, PMD_SIZE, copied, flags,
> +				      &iomap, &srcmap);
> +	}
> +
>  finish_iomap:
>  	if (ops->iomap_end) {
> -		int copied = PMD_SIZE;
> -
> -		if (ret == VM_FAULT_FALLBACK)
> -			copied = 0;
>  		/*
>  		 * The fault is done by now and there's no way back (other
>  		 * thread may be already happily using PMD we have installed).
> diff --git a/fs/iomap/apply.c b/fs/iomap/apply.c index
> 0493da5286ad..26a54ded184f 100644
> --- a/fs/iomap/apply.c
> +++ b/fs/iomap/apply.c
> @@ -84,6 +84,10 @@ iomap_apply(struct inode *inode, loff_t pos, loff_t length,
> unsigned flags,
>  	written = actor(inode, pos, length, data, &iomap,
>  			srcmap.type != IOMAP_HOLE ? &srcmap : &iomap);
> 
> +	if (ops->iomap_post_actor) {
> +		written = ops->iomap_post_actor(inode, pos, length, written,
> +						flags, &iomap, &srcmap);
> +	}
>  out:
>  	/*
>  	 * Now the data has been copied, commit the range we've copied.  This
> diff --git a/fs/xfs/xfs_bmap_util.c b/fs/xfs/xfs_bmap_util.c index
> a5e9d7d34023..2a36dc93ff27 100644
> --- a/fs/xfs/xfs_bmap_util.c
> +++ b/fs/xfs/xfs_bmap_util.c
> @@ -965,8 +965,7 @@ xfs_free_file_space(
>  		return 0;
>  	if (offset + len > XFS_ISIZE(ip))
>  		len = XFS_ISIZE(ip) - offset;
> -	error = iomap_zero_range(VFS_I(ip), offset, len, NULL,
> -			&xfs_buffered_write_iomap_ops);
> +	error = xfs_iomap_zero_range(ip, offset, len, NULL);
>  	if (error)
>  		return error;
> 
> diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c index 396ef36dcd0a..89406ec6741b
> 100644
> --- a/fs/xfs/xfs_file.c
> +++ b/fs/xfs/xfs_file.c
> @@ -684,11 +684,12 @@ xfs_file_dax_write(
>  	pos = iocb->ki_pos;
> 
>  	trace_xfs_file_dax_write(iocb, from);
> -	ret = dax_iomap_rw(iocb, from, &xfs_direct_write_iomap_ops);
> +	ret = dax_iomap_rw(iocb, from, &xfs_dax_write_iomap_ops);
>  	if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
>  		i_size_write(inode, iocb->ki_pos);
>  		error = xfs_setfilesize(ip, pos, ret);
>  	}
> +
>  out:
>  	if (iolock)
>  		xfs_iunlock(ip, iolock);
> @@ -1309,7 +1310,7 @@ __xfs_filemap_fault(
> 
>  		ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL,
>  				(write_fault && !vmf->cow_page) ?
> -				 &xfs_direct_write_iomap_ops :
> +				 &xfs_dax_write_iomap_ops :
>  				 &xfs_read_iomap_ops);
>  		if (ret & VM_FAULT_NEEDDSYNC)
>  			ret = dax_finish_sync_fault(vmf, pe_size, pfn); diff --git
> a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c index d154f42e2dc6..2f322e2f8544
> 100644
> --- a/fs/xfs/xfs_iomap.c
> +++ b/fs/xfs/xfs_iomap.c
> @@ -761,7 +761,8 @@ xfs_direct_write_iomap_begin(
> 
>  		/* may drop and re-acquire the ilock */
>  		error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
> -				&lockmode, flags & IOMAP_DIRECT);
> +				&lockmode,
> +				(flags & IOMAP_DIRECT) || IS_DAX(inode));
>  		if (error)
>  			goto out_unlock;
>  		if (shared)
> @@ -854,6 +855,36 @@ const struct iomap_ops xfs_direct_write_iomap_ops =
> {
>  	.iomap_begin		= xfs_direct_write_iomap_begin,
>  };
> 
> +static int
> +xfs_dax_write_iomap_post_actor(
> +	struct inode		*inode,
> +	loff_t			pos,
> +	loff_t			length,
> +	ssize_t			written,
> +	unsigned int		flags,
> +	struct iomap		*iomap,
> +	struct iomap		*srcmap)
> +{
> +	int			error = 0;
> +	struct xfs_inode	*ip = XFS_I(inode);
> +	bool			cow = xfs_is_cow_inode(ip);
> +
> +	if (written <= 0) {
> +		if (cow)
> +			xfs_reflink_cancel_cow_range(ip, pos, length, true);
> +		return written;
> +	}
> +
> +	if (cow)
> +		error = xfs_reflink_end_cow(ip, pos, written);
> +	return error ?: written;
> +}
> +
> +const struct iomap_ops xfs_dax_write_iomap_ops = {
> +	.iomap_begin		= xfs_direct_write_iomap_begin,
> +	.iomap_post_actor	= xfs_dax_write_iomap_post_actor,
> +};
> +
>  static int
>  xfs_buffered_write_iomap_begin(
>  	struct inode		*inode,
> diff --git a/fs/xfs/xfs_iomap.h b/fs/xfs/xfs_iomap.h index
> 7d3703556d0e..fbacf638ab21 100644
> --- a/fs/xfs/xfs_iomap.h
> +++ b/fs/xfs/xfs_iomap.h
> @@ -42,8 +42,32 @@ xfs_aligned_fsb_count(
> 
>  extern const struct iomap_ops xfs_buffered_write_iomap_ops;  extern const
> struct iomap_ops xfs_direct_write_iomap_ops;
> +extern const struct iomap_ops xfs_dax_write_iomap_ops;
>  extern const struct iomap_ops xfs_read_iomap_ops;  extern const struct
> iomap_ops xfs_seek_iomap_ops;  extern const struct iomap_ops
> xfs_xattr_iomap_ops;
> 
> +static inline int
> +xfs_iomap_zero_range(
> +	struct xfs_inode	*ip,
> +	loff_t			offset,
> +	loff_t			len,
> +	bool			*did_zero)
> +{
> +	return iomap_zero_range(VFS_I(ip), offset, len, did_zero,
> +			IS_DAX(VFS_I(ip)) ? &xfs_dax_write_iomap_ops
> +					  : &xfs_buffered_write_iomap_ops); }
> +
> +static inline int
> +xfs_iomap_truncate_page(
> +	struct xfs_inode	*ip,
> +	loff_t			pos,
> +	bool			*did_zero)
> +{
> +	return iomap_truncate_page(VFS_I(ip), pos, did_zero,
> +			IS_DAX(VFS_I(ip)) ? &xfs_dax_write_iomap_ops
> +					  : &xfs_buffered_write_iomap_ops); }
> +
>  #endif /* __XFS_IOMAP_H__*/
> diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c index
> dfe24b7f26e5..6d936c3e1a6e 100644
> --- a/fs/xfs/xfs_iops.c
> +++ b/fs/xfs/xfs_iops.c
> @@ -911,8 +911,8 @@ xfs_setattr_size(
>  	 */
>  	if (newsize > oldsize) {
>  		trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
> -		error = iomap_zero_range(inode, oldsize, newsize - oldsize,
> -				&did_zeroing, &xfs_buffered_write_iomap_ops);
> +		error = xfs_iomap_zero_range(ip, oldsize, newsize - oldsize,
> +				&did_zeroing);
>  	} else {
>  		/*
>  		 * iomap won't detect a dirty page over an unwritten block (or a @@
> -924,8 +924,7 @@ xfs_setattr_size(
>  						     newsize);
>  		if (error)
>  			return error;
> -		error = iomap_truncate_page(inode, newsize, &did_zeroing,
> -				&xfs_buffered_write_iomap_ops);
> +		error = xfs_iomap_truncate_page(ip, newsize, &did_zeroing);
>  	}
> 
>  	if (error)
> diff --git a/fs/xfs/xfs_reflink.c b/fs/xfs/xfs_reflink.c index
> d25434f93235..9a780948dbd0 100644
> --- a/fs/xfs/xfs_reflink.c
> +++ b/fs/xfs/xfs_reflink.c
> @@ -1266,8 +1266,7 @@ xfs_reflink_zero_posteof(
>  		return 0;
> 
>  	trace_xfs_zero_eof(ip, isize, pos - isize);
> -	return iomap_zero_range(VFS_I(ip), isize, pos - isize, NULL,
> -			&xfs_buffered_write_iomap_ops);
> +	return xfs_iomap_zero_range(ip, isize, pos - isize, NULL);
>  }
> 
>  /*
> diff --git a/include/linux/iomap.h b/include/linux/iomap.h index
> 95562f863ad0..58f2e1c78018 100644
> --- a/include/linux/iomap.h
> +++ b/include/linux/iomap.h
> @@ -135,6 +135,14 @@ struct iomap_ops {
>  			unsigned flags, struct iomap *iomap,
>  			struct iomap *srcmap);
> 
> +	/*
> +	 * Handle the error code from actor(). Do the finishing jobs for extra
> +	 * operations, such as CoW, according to whether written is negative.
> +	 */
> +	int (*iomap_post_actor)(struct inode *inode, loff_t pos, loff_t length,
> +			ssize_t written, unsigned flags, struct iomap *iomap,
> +			struct iomap *srcmap);
> +
>  	/*
>  	 * Commit and/or unreserve space previous allocated using iomap_begin.
>  	 * Written indicates the length of the successful write operation which
> --
> 2.31.1


^ permalink raw reply	[flat|nested] 9+ messages in thread

* Re: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path
  2021-06-24  8:49   ` ruansy.fnst
@ 2021-06-25 22:18     ` Darrick J. Wong
  2021-06-28  2:55       ` ruansy.fnst
  0 siblings, 1 reply; 9+ messages in thread
From: Darrick J. Wong @ 2021-06-25 22:18 UTC (permalink / raw)
  To: ruansy.fnst
  Cc: dan.j.williams, david, hch, linux-fsdevel, linux-kernel, nvdimm,
	linux-xfs, rgoldwyn, viro, willy

On Thu, Jun 24, 2021 at 08:49:17AM +0000, ruansy.fnst@fujitsu.com wrote:
> Hi Darrick,
> 
> Do you have any comment on this?

Sorry, was on vacation.

> Thanks,
> Ruan.
> 
> > -----Original Message-----
> > From: Shiyang Ruan <ruansy.fnst@fujitsu.com>
> > Subject: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path
> > 
> > Hi Darrick,
> > 
> > Since other patches looks good, I post this RFC patch singly to hot-fix the
> > problem in xfs_dax_write_iomap_ops->iomap_end() of v6 that the error code
> > was ingored. I will split this in two patches(changes in iomap and xfs
> > respectively) in next formal version if it looks ok.
> > 
> > ====
> > 
> > Introduce a new interface called "iomap_post_actor()" in iomap_ops.  And call
> > it between ->actor() and ->iomap_end().  It is mean to handle the error code
> > returned from ->actor().  In this patchset, it is used to remap or cancel the
> > CoW extents according to the error code.
> > 
> > Signed-off-by: Shiyang Ruan <ruansy.fnst@fujitsu.com>
> > ---
> >  fs/dax.c               | 27 ++++++++++++++++++---------
> >  fs/iomap/apply.c       |  4 ++++
> >  fs/xfs/xfs_bmap_util.c |  3 +--
> >  fs/xfs/xfs_file.c      |  5 +++--
> >  fs/xfs/xfs_iomap.c     | 33 ++++++++++++++++++++++++++++++++-
> >  fs/xfs/xfs_iomap.h     | 24 ++++++++++++++++++++++++
> >  fs/xfs/xfs_iops.c      |  7 +++----
> >  fs/xfs/xfs_reflink.c   |  3 +--
> >  include/linux/iomap.h  |  8 ++++++++
> >  9 files changed, 94 insertions(+), 20 deletions(-)
> > 
> > diff --git a/fs/dax.c b/fs/dax.c
> > index 93f16210847b..0740c2610b6f 100644
> > --- a/fs/dax.c
> > +++ b/fs/dax.c
> > @@ -1537,7 +1537,7 @@ static vm_fault_t dax_iomap_pte_fault(struct
> > vm_fault *vmf, pfn_t *pfnp,
> >  	struct iomap iomap = { .type = IOMAP_HOLE };
> >  	struct iomap srcmap = { .type = IOMAP_HOLE };
> >  	unsigned flags = IOMAP_FAULT;
> > -	int error;
> > +	int error, copied = PAGE_SIZE;
> >  	bool write = vmf->flags & FAULT_FLAG_WRITE;
> >  	vm_fault_t ret = 0, major = 0;
> >  	void *entry;
> > @@ -1598,7 +1598,7 @@ static vm_fault_t dax_iomap_pte_fault(struct
> > vm_fault *vmf, pfn_t *pfnp,
> >  	ret = dax_fault_actor(vmf, pfnp, &xas, &entry, false, flags,
> >  			      &iomap, &srcmap);
> >  	if (ret == VM_FAULT_SIGBUS)
> > -		goto finish_iomap;
> > +		goto finish_iomap_actor;
> > 
> >  	/* read/write MAPPED, CoW UNWRITTEN */
> >  	if (iomap.flags & IOMAP_F_NEW) {
> > @@ -1607,10 +1607,16 @@ static vm_fault_t dax_iomap_pte_fault(struct
> > vm_fault *vmf, pfn_t *pfnp,
> >  		major = VM_FAULT_MAJOR;
> >  	}
> > 
> > + finish_iomap_actor:
> > +	if (ops->iomap_post_actor) {
> > +		if (ret & VM_FAULT_ERROR)
> > +			copied = 0;
> > +		ops->iomap_post_actor(inode, pos, PMD_SIZE, copied, flags,
> > +				      &iomap, &srcmap);
> > +	}
> > +
> >  finish_iomap:
> >  	if (ops->iomap_end) {
> > -		int copied = PAGE_SIZE;
> > -
> >  		if (ret & VM_FAULT_ERROR)
> >  			copied = 0;
> >  		/*
> > @@ -1677,7 +1683,7 @@ static vm_fault_t dax_iomap_pmd_fault(struct
> > vm_fault *vmf, pfn_t *pfnp,
> >  	pgoff_t max_pgoff;
> >  	void *entry;
> >  	loff_t pos;
> > -	int error;
> > +	int error, copied = PMD_SIZE;
> > 
> >  	/*
> >  	 * Check whether offset isn't beyond end of file now. Caller is @@ -1736,12
> > +1742,15 @@ static vm_fault_t dax_iomap_pmd_fault(struct vm_fault *vmf,
> > pfn_t *pfnp,
> >  	ret = dax_fault_actor(vmf, pfnp, &xas, &entry, true, flags,
> >  			      &iomap, &srcmap);
> > 
> > +	if (ret == VM_FAULT_FALLBACK)
> > +		copied = 0;
> > +	if (ops->iomap_post_actor) {
> > +		ops->iomap_post_actor(inode, pos, PMD_SIZE, copied, flags,
> > +				      &iomap, &srcmap);
> > +	}
> > +
> >  finish_iomap:
> >  	if (ops->iomap_end) {
> > -		int copied = PMD_SIZE;
> > -
> > -		if (ret == VM_FAULT_FALLBACK)
> > -			copied = 0;
> >  		/*
> >  		 * The fault is done by now and there's no way back (other
> >  		 * thread may be already happily using PMD we have installed).
> > diff --git a/fs/iomap/apply.c b/fs/iomap/apply.c index
> > 0493da5286ad..26a54ded184f 100644
> > --- a/fs/iomap/apply.c
> > +++ b/fs/iomap/apply.c
> > @@ -84,6 +84,10 @@ iomap_apply(struct inode *inode, loff_t pos, loff_t length,
> > unsigned flags,
> >  	written = actor(inode, pos, length, data, &iomap,
> >  			srcmap.type != IOMAP_HOLE ? &srcmap : &iomap);
> > 
> > +	if (ops->iomap_post_actor) {
> > +		written = ops->iomap_post_actor(inode, pos, length, written,
> > +						flags, &iomap, &srcmap);

How many operations actually need an iomap_post_actor?  It's just the
dax ones, right?  Which is ... iomap_truncate_page, iomap_zero_range,
dax_iomap_fault, and dax_iomap_rw, right?  We don't need a post_actor
for other iomap functionality (like FIEMAP, SEEK_DATA/SEEK_HOLE, etc.)
so adding a new function pointer for all operations feels a bit
overbroad.

I had imagined that you'd create a struct dax_iomap_ops to wrap all the
extra functionality that you need for dax operations:

struct dax_iomap_ops {
	struct iomap_ops	iomap_ops;

	int			(*end_io)(inode, pos, length...);
};

And alter the four functions that you need to take the special
dax_iomap_ops.  I guess the downside is that this makes
iomap_truncate_page and iomap_zero_range more complicated, but maybe
it's just time to split those into DAX-specific versions.  Then we'd be
rid of the cross-links betwee fs/iomap/buffered-io.c and fs/dax.c.

> > +	}
> >  out:
> >  	/*
> >  	 * Now the data has been copied, commit the range we've copied.  This
> > diff --git a/fs/xfs/xfs_bmap_util.c b/fs/xfs/xfs_bmap_util.c index
> > a5e9d7d34023..2a36dc93ff27 100644
> > --- a/fs/xfs/xfs_bmap_util.c
> > +++ b/fs/xfs/xfs_bmap_util.c
> > @@ -965,8 +965,7 @@ xfs_free_file_space(
> >  		return 0;
> >  	if (offset + len > XFS_ISIZE(ip))
> >  		len = XFS_ISIZE(ip) - offset;
> > -	error = iomap_zero_range(VFS_I(ip), offset, len, NULL,
> > -			&xfs_buffered_write_iomap_ops);
> > +	error = xfs_iomap_zero_range(ip, offset, len, NULL);
> >  	if (error)
> >  		return error;
> > 
> > diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c index 396ef36dcd0a..89406ec6741b
> > 100644
> > --- a/fs/xfs/xfs_file.c
> > +++ b/fs/xfs/xfs_file.c
> > @@ -684,11 +684,12 @@ xfs_file_dax_write(
> >  	pos = iocb->ki_pos;
> > 
> >  	trace_xfs_file_dax_write(iocb, from);
> > -	ret = dax_iomap_rw(iocb, from, &xfs_direct_write_iomap_ops);
> > +	ret = dax_iomap_rw(iocb, from, &xfs_dax_write_iomap_ops);
> >  	if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
> >  		i_size_write(inode, iocb->ki_pos);
> >  		error = xfs_setfilesize(ip, pos, ret);
> >  	}
> > +
> >  out:
> >  	if (iolock)
> >  		xfs_iunlock(ip, iolock);
> > @@ -1309,7 +1310,7 @@ __xfs_filemap_fault(
> > 
> >  		ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL,
> >  				(write_fault && !vmf->cow_page) ?
> > -				 &xfs_direct_write_iomap_ops :
> > +				 &xfs_dax_write_iomap_ops :
> >  				 &xfs_read_iomap_ops);
> >  		if (ret & VM_FAULT_NEEDDSYNC)
> >  			ret = dax_finish_sync_fault(vmf, pe_size, pfn); diff --git
> > a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c index d154f42e2dc6..2f322e2f8544
> > 100644
> > --- a/fs/xfs/xfs_iomap.c
> > +++ b/fs/xfs/xfs_iomap.c
> > @@ -761,7 +761,8 @@ xfs_direct_write_iomap_begin(
> > 
> >  		/* may drop and re-acquire the ilock */
> >  		error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
> > -				&lockmode, flags & IOMAP_DIRECT);
> > +				&lockmode,
> > +				(flags & IOMAP_DIRECT) || IS_DAX(inode));
> >  		if (error)
> >  			goto out_unlock;
> >  		if (shared)
> > @@ -854,6 +855,36 @@ const struct iomap_ops xfs_direct_write_iomap_ops =
> > {
> >  	.iomap_begin		= xfs_direct_write_iomap_begin,
> >  };
> > 
> > +static int
> > +xfs_dax_write_iomap_post_actor(
> > +	struct inode		*inode,
> > +	loff_t			pos,
> > +	loff_t			length,
> > +	ssize_t			written,
> > +	unsigned int		flags,
> > +	struct iomap		*iomap,
> > +	struct iomap		*srcmap)
> > +{
> > +	int			error = 0;
> > +	struct xfs_inode	*ip = XFS_I(inode);
> > +	bool			cow = xfs_is_cow_inode(ip);
> > +
> > +	if (written <= 0) {
> > +		if (cow)
> > +			xfs_reflink_cancel_cow_range(ip, pos, length, true);
> > +		return written;
> > +	}
> > +
> > +	if (cow)
> > +		error = xfs_reflink_end_cow(ip, pos, written);
> > +	return error ?: written;
> > +}

This is pretty much the same as what xfs_dio_write_end_io does, right?

I had imagined that you'd change the function signatures to drop the
iocb so that you could reuse this code instead of creating a whole new
callback.

Ah well.  Can I send you some prep patches to clean up some of the weird
iomap code as a preparation series for this?

--D

> > +
> > +const struct iomap_ops xfs_dax_write_iomap_ops = {
> > +	.iomap_begin		= xfs_direct_write_iomap_begin,
> > +	.iomap_post_actor	= xfs_dax_write_iomap_post_actor,
> > +};
> > +
> >  static int
> >  xfs_buffered_write_iomap_begin(
> >  	struct inode		*inode,
> > diff --git a/fs/xfs/xfs_iomap.h b/fs/xfs/xfs_iomap.h index
> > 7d3703556d0e..fbacf638ab21 100644
> > --- a/fs/xfs/xfs_iomap.h
> > +++ b/fs/xfs/xfs_iomap.h
> > @@ -42,8 +42,32 @@ xfs_aligned_fsb_count(
> > 
> >  extern const struct iomap_ops xfs_buffered_write_iomap_ops;  extern const
> > struct iomap_ops xfs_direct_write_iomap_ops;
> > +extern const struct iomap_ops xfs_dax_write_iomap_ops;
> >  extern const struct iomap_ops xfs_read_iomap_ops;  extern const struct
> > iomap_ops xfs_seek_iomap_ops;  extern const struct iomap_ops
> > xfs_xattr_iomap_ops;
> > 
> > +static inline int
> > +xfs_iomap_zero_range(
> > +	struct xfs_inode	*ip,
> > +	loff_t			offset,
> > +	loff_t			len,
> > +	bool			*did_zero)
> > +{
> > +	return iomap_zero_range(VFS_I(ip), offset, len, did_zero,
> > +			IS_DAX(VFS_I(ip)) ? &xfs_dax_write_iomap_ops
> > +					  : &xfs_buffered_write_iomap_ops); }
> > +
> > +static inline int
> > +xfs_iomap_truncate_page(
> > +	struct xfs_inode	*ip,
> > +	loff_t			pos,
> > +	bool			*did_zero)
> > +{
> > +	return iomap_truncate_page(VFS_I(ip), pos, did_zero,
> > +			IS_DAX(VFS_I(ip)) ? &xfs_dax_write_iomap_ops
> > +					  : &xfs_buffered_write_iomap_ops); }
> > +
> >  #endif /* __XFS_IOMAP_H__*/
> > diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c index
> > dfe24b7f26e5..6d936c3e1a6e 100644
> > --- a/fs/xfs/xfs_iops.c
> > +++ b/fs/xfs/xfs_iops.c
> > @@ -911,8 +911,8 @@ xfs_setattr_size(
> >  	 */
> >  	if (newsize > oldsize) {
> >  		trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
> > -		error = iomap_zero_range(inode, oldsize, newsize - oldsize,
> > -				&did_zeroing, &xfs_buffered_write_iomap_ops);
> > +		error = xfs_iomap_zero_range(ip, oldsize, newsize - oldsize,
> > +				&did_zeroing);
> >  	} else {
> >  		/*
> >  		 * iomap won't detect a dirty page over an unwritten block (or a @@
> > -924,8 +924,7 @@ xfs_setattr_size(
> >  						     newsize);
> >  		if (error)
> >  			return error;
> > -		error = iomap_truncate_page(inode, newsize, &did_zeroing,
> > -				&xfs_buffered_write_iomap_ops);
> > +		error = xfs_iomap_truncate_page(ip, newsize, &did_zeroing);
> >  	}
> > 
> >  	if (error)
> > diff --git a/fs/xfs/xfs_reflink.c b/fs/xfs/xfs_reflink.c index
> > d25434f93235..9a780948dbd0 100644
> > --- a/fs/xfs/xfs_reflink.c
> > +++ b/fs/xfs/xfs_reflink.c
> > @@ -1266,8 +1266,7 @@ xfs_reflink_zero_posteof(
> >  		return 0;
> > 
> >  	trace_xfs_zero_eof(ip, isize, pos - isize);
> > -	return iomap_zero_range(VFS_I(ip), isize, pos - isize, NULL,
> > -			&xfs_buffered_write_iomap_ops);
> > +	return xfs_iomap_zero_range(ip, isize, pos - isize, NULL);
> >  }
> > 
> >  /*
> > diff --git a/include/linux/iomap.h b/include/linux/iomap.h index
> > 95562f863ad0..58f2e1c78018 100644
> > --- a/include/linux/iomap.h
> > +++ b/include/linux/iomap.h
> > @@ -135,6 +135,14 @@ struct iomap_ops {
> >  			unsigned flags, struct iomap *iomap,
> >  			struct iomap *srcmap);
> > 
> > +	/*
> > +	 * Handle the error code from actor(). Do the finishing jobs for extra
> > +	 * operations, such as CoW, according to whether written is negative.
> > +	 */
> > +	int (*iomap_post_actor)(struct inode *inode, loff_t pos, loff_t length,
> > +			ssize_t written, unsigned flags, struct iomap *iomap,
> > +			struct iomap *srcmap);
> > +
> >  	/*
> >  	 * Commit and/or unreserve space previous allocated using iomap_begin.
> >  	 * Written indicates the length of the successful write operation which
> > --
> > 2.31.1
> 

^ permalink raw reply	[flat|nested] 9+ messages in thread

* RE: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path
  2021-06-25 22:18     ` Darrick J. Wong
@ 2021-06-28  2:55       ` ruansy.fnst
  2021-06-28  5:09         ` Darrick J. Wong
  2021-07-09 12:36         ` [PATCH v6.2 6/7] dax: Introduce dax_iomap_ops for end of reflink Shiyang Ruan
  0 siblings, 2 replies; 9+ messages in thread
From: ruansy.fnst @ 2021-06-28  2:55 UTC (permalink / raw)
  To: Darrick J. Wong
  Cc: dan.j.williams, david, hch, linux-fsdevel, linux-kernel, nvdimm,
	linux-xfs, rgoldwyn, viro, willy

> -----Original Message-----
> Subject: Re: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path
> 
> On Thu, Jun 24, 2021 at 08:49:17AM +0000, ruansy.fnst@fujitsu.com wrote:
> > Hi Darrick,
> >
> > Do you have any comment on this?
> 
> Sorry, was on vacation.
> 
> > Thanks,
> > Ruan.
> >
> > > -----Original Message-----
> > > From: Shiyang Ruan <ruansy.fnst@fujitsu.com>
> > > Subject: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path
> > >
> > > Hi Darrick,
> > >
> > > Since other patches looks good, I post this RFC patch singly to
> > > hot-fix the problem in xfs_dax_write_iomap_ops->iomap_end() of v6
> > > that the error code was ingored. I will split this in two
> > > patches(changes in iomap and xfs
> > > respectively) in next formal version if it looks ok.
> > >
> > > ====
> > >
> > > Introduce a new interface called "iomap_post_actor()" in iomap_ops.
> > > And call it between ->actor() and ->iomap_end().  It is mean to
> > > handle the error code returned from ->actor().  In this patchset, it
> > > is used to remap or cancel the CoW extents according to the error code.
> > >
> > > Signed-off-by: Shiyang Ruan <ruansy.fnst@fujitsu.com>
> > > ---
> > >  fs/dax.c               | 27 ++++++++++++++++++---------
> > >  fs/iomap/apply.c       |  4 ++++
> > >  fs/xfs/xfs_bmap_util.c |  3 +--
> > >  fs/xfs/xfs_file.c      |  5 +++--
> > >  fs/xfs/xfs_iomap.c     | 33 ++++++++++++++++++++++++++++++++-
> > >  fs/xfs/xfs_iomap.h     | 24 ++++++++++++++++++++++++
> > >  fs/xfs/xfs_iops.c      |  7 +++----
> > >  fs/xfs/xfs_reflink.c   |  3 +--
> > >  include/linux/iomap.h  |  8 ++++++++
> > >  9 files changed, 94 insertions(+), 20 deletions(-)
> > >
> > > diff --git a/fs/dax.c b/fs/dax.c
> > > index 93f16210847b..0740c2610b6f 100644
> > > --- a/fs/dax.c
> > > +++ b/fs/dax.c
> > > @@ -1537,7 +1537,7 @@ static vm_fault_t dax_iomap_pte_fault(struct
> > > vm_fault *vmf, pfn_t *pfnp,
> > >  	struct iomap iomap = { .type = IOMAP_HOLE };
> > >  	struct iomap srcmap = { .type = IOMAP_HOLE };
> > >  	unsigned flags = IOMAP_FAULT;
> > > -	int error;
> > > +	int error, copied = PAGE_SIZE;
> > >  	bool write = vmf->flags & FAULT_FLAG_WRITE;
> > >  	vm_fault_t ret = 0, major = 0;
> > >  	void *entry;
> > > @@ -1598,7 +1598,7 @@ static vm_fault_t dax_iomap_pte_fault(struct
> > > vm_fault *vmf, pfn_t *pfnp,
> > >  	ret = dax_fault_actor(vmf, pfnp, &xas, &entry, false, flags,
> > >  			      &iomap, &srcmap);
> > >  	if (ret == VM_FAULT_SIGBUS)
> > > -		goto finish_iomap;
> > > +		goto finish_iomap_actor;
> > >
> > >  	/* read/write MAPPED, CoW UNWRITTEN */
> > >  	if (iomap.flags & IOMAP_F_NEW) {
> > > @@ -1607,10 +1607,16 @@ static vm_fault_t dax_iomap_pte_fault(struct
> > > vm_fault *vmf, pfn_t *pfnp,
> > >  		major = VM_FAULT_MAJOR;
> > >  	}
> > >
> > > + finish_iomap_actor:
> > > +	if (ops->iomap_post_actor) {
> > > +		if (ret & VM_FAULT_ERROR)
> > > +			copied = 0;
> > > +		ops->iomap_post_actor(inode, pos, PMD_SIZE, copied, flags,
> > > +				      &iomap, &srcmap);
> > > +	}
> > > +
> > >  finish_iomap:
> > >  	if (ops->iomap_end) {
> > > -		int copied = PAGE_SIZE;
> > > -
> > >  		if (ret & VM_FAULT_ERROR)
> > >  			copied = 0;
> > >  		/*
> > > @@ -1677,7 +1683,7 @@ static vm_fault_t dax_iomap_pmd_fault(struct
> > > vm_fault *vmf, pfn_t *pfnp,
> > >  	pgoff_t max_pgoff;
> > >  	void *entry;
> > >  	loff_t pos;
> > > -	int error;
> > > +	int error, copied = PMD_SIZE;
> > >
> > >  	/*
> > >  	 * Check whether offset isn't beyond end of file now. Caller is @@
> > > -1736,12
> > > +1742,15 @@ static vm_fault_t dax_iomap_pmd_fault(struct vm_fault
> > > +*vmf,
> > > pfn_t *pfnp,
> > >  	ret = dax_fault_actor(vmf, pfnp, &xas, &entry, true, flags,
> > >  			      &iomap, &srcmap);
> > >
> > > +	if (ret == VM_FAULT_FALLBACK)
> > > +		copied = 0;
> > > +	if (ops->iomap_post_actor) {
> > > +		ops->iomap_post_actor(inode, pos, PMD_SIZE, copied, flags,
> > > +				      &iomap, &srcmap);
> > > +	}
> > > +
> > >  finish_iomap:
> > >  	if (ops->iomap_end) {
> > > -		int copied = PMD_SIZE;
> > > -
> > > -		if (ret == VM_FAULT_FALLBACK)
> > > -			copied = 0;
> > >  		/*
> > >  		 * The fault is done by now and there's no way back (other
> > >  		 * thread may be already happily using PMD we have installed).
> > > diff --git a/fs/iomap/apply.c b/fs/iomap/apply.c index
> > > 0493da5286ad..26a54ded184f 100644
> > > --- a/fs/iomap/apply.c
> > > +++ b/fs/iomap/apply.c
> > > @@ -84,6 +84,10 @@ iomap_apply(struct inode *inode, loff_t pos,
> > > loff_t length, unsigned flags,
> > >  	written = actor(inode, pos, length, data, &iomap,
> > >  			srcmap.type != IOMAP_HOLE ? &srcmap : &iomap);
> > >
> > > +	if (ops->iomap_post_actor) {
> > > +		written = ops->iomap_post_actor(inode, pos, length, written,
> > > +						flags, &iomap, &srcmap);
> 
> How many operations actually need an iomap_post_actor?  It's just the dax
> ones, right?  Which is ... iomap_truncate_page, iomap_zero_range,
> dax_iomap_fault, and dax_iomap_rw, right?  We don't need a post_actor for
> other iomap functionality (like FIEMAP, SEEK_DATA/SEEK_HOLE, etc.) so adding
> a new function pointer for all operations feels a bit overbroad.

Yes.

> 
> I had imagined that you'd create a struct dax_iomap_ops to wrap all the extra
> functionality that you need for dax operations:
> 
> struct dax_iomap_ops {
> 	struct iomap_ops	iomap_ops;
> 
> 	int			(*end_io)(inode, pos, length...);
> };
> 
> And alter the four functions that you need to take the special dax_iomap_ops.
> I guess the downside is that this makes iomap_truncate_page and
> iomap_zero_range more complicated, but maybe it's just time to split those into
> DAX-specific versions.  Then we'd be rid of the cross-links betwee
> fs/iomap/buffered-io.c and fs/dax.c.

This seems to be a better solution.  I'll try in this way.  Thanks for your guidance.

> 
> > > +	}
> > >  out:
> > >  	/*
> > >  	 * Now the data has been copied, commit the range we've copied.
> > > This diff --git a/fs/xfs/xfs_bmap_util.c b/fs/xfs/xfs_bmap_util.c
> > > index
> > > a5e9d7d34023..2a36dc93ff27 100644
> > > --- a/fs/xfs/xfs_bmap_util.c
> > > +++ b/fs/xfs/xfs_bmap_util.c
> > > @@ -965,8 +965,7 @@ xfs_free_file_space(
> > >  		return 0;
> > >  	if (offset + len > XFS_ISIZE(ip))
> > >  		len = XFS_ISIZE(ip) - offset;
> > > -	error = iomap_zero_range(VFS_I(ip), offset, len, NULL,
> > > -			&xfs_buffered_write_iomap_ops);
> > > +	error = xfs_iomap_zero_range(ip, offset, len, NULL);
> > >  	if (error)
> > >  		return error;
> > >
> > > diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c index
> > > 396ef36dcd0a..89406ec6741b
> > > 100644
> > > --- a/fs/xfs/xfs_file.c
> > > +++ b/fs/xfs/xfs_file.c
> > > @@ -684,11 +684,12 @@ xfs_file_dax_write(
> > >  	pos = iocb->ki_pos;
> > >
> > >  	trace_xfs_file_dax_write(iocb, from);
> > > -	ret = dax_iomap_rw(iocb, from, &xfs_direct_write_iomap_ops);
> > > +	ret = dax_iomap_rw(iocb, from, &xfs_dax_write_iomap_ops);
> > >  	if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
> > >  		i_size_write(inode, iocb->ki_pos);
> > >  		error = xfs_setfilesize(ip, pos, ret);
> > >  	}
> > > +
> > >  out:
> > >  	if (iolock)
> > >  		xfs_iunlock(ip, iolock);
> > > @@ -1309,7 +1310,7 @@ __xfs_filemap_fault(
> > >
> > >  		ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL,
> > >  				(write_fault && !vmf->cow_page) ?
> > > -				 &xfs_direct_write_iomap_ops :
> > > +				 &xfs_dax_write_iomap_ops :
> > >  				 &xfs_read_iomap_ops);
> > >  		if (ret & VM_FAULT_NEEDDSYNC)
> > >  			ret = dax_finish_sync_fault(vmf, pe_size, pfn); diff --git
> > > a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c index
> > > d154f42e2dc6..2f322e2f8544
> > > 100644
> > > --- a/fs/xfs/xfs_iomap.c
> > > +++ b/fs/xfs/xfs_iomap.c
> > > @@ -761,7 +761,8 @@ xfs_direct_write_iomap_begin(
> > >
> > >  		/* may drop and re-acquire the ilock */
> > >  		error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
> > > -				&lockmode, flags & IOMAP_DIRECT);
> > > +				&lockmode,
> > > +				(flags & IOMAP_DIRECT) || IS_DAX(inode));
> > >  		if (error)
> > >  			goto out_unlock;
> > >  		if (shared)
> > > @@ -854,6 +855,36 @@ const struct iomap_ops
> > > xfs_direct_write_iomap_ops = {
> > >  	.iomap_begin		= xfs_direct_write_iomap_begin,
> > >  };
> > >
> > > +static int
> > > +xfs_dax_write_iomap_post_actor(
> > > +	struct inode		*inode,
> > > +	loff_t			pos,
> > > +	loff_t			length,
> > > +	ssize_t			written,
> > > +	unsigned int		flags,
> > > +	struct iomap		*iomap,
> > > +	struct iomap		*srcmap)
> > > +{
> > > +	int			error = 0;
> > > +	struct xfs_inode	*ip = XFS_I(inode);
> > > +	bool			cow = xfs_is_cow_inode(ip);
> > > +
> > > +	if (written <= 0) {
> > > +		if (cow)
> > > +			xfs_reflink_cancel_cow_range(ip, pos, length, true);
> > > +		return written;
> > > +	}
> > > +
> > > +	if (cow)
> > > +		error = xfs_reflink_end_cow(ip, pos, written);
> > > +	return error ?: written;
> > > +}
> 
> This is pretty much the same as what xfs_dio_write_end_io does, right?

It just handles the end part of CoW here.
xfs_dio_write_end_io() also updates file size, which is only needed in write() but not in page fault.  And the update file size work is done in xfs_dax_file_write(), it's fine, no need to modify it.

> 
> I had imagined that you'd change the function signatures to drop the iocb so
> that you could reuse this code instead of creating a whole new callback.
> 
> Ah well.  Can I send you some prep patches to clean up some of the weird
> iomap code as a preparation series for this?

Sure.  Thanks.


--
Ruan.

> 
> --D
> 
> > > +
> > > +const struct iomap_ops xfs_dax_write_iomap_ops = {
> > > +	.iomap_begin		= xfs_direct_write_iomap_begin,
> > > +	.iomap_post_actor	= xfs_dax_write_iomap_post_actor,
> > > +};
> > > +
> > >  static int
> > >  xfs_buffered_write_iomap_begin(
> > >  	struct inode		*inode,
> > > diff --git a/fs/xfs/xfs_iomap.h b/fs/xfs/xfs_iomap.h index
> > > 7d3703556d0e..fbacf638ab21 100644
> > > --- a/fs/xfs/xfs_iomap.h
> > > +++ b/fs/xfs/xfs_iomap.h
> > > @@ -42,8 +42,32 @@ xfs_aligned_fsb_count(
> > >
> > >  extern const struct iomap_ops xfs_buffered_write_iomap_ops;  extern
> > > const struct iomap_ops xfs_direct_write_iomap_ops;
> > > +extern const struct iomap_ops xfs_dax_write_iomap_ops;
> > >  extern const struct iomap_ops xfs_read_iomap_ops;  extern const
> > > struct iomap_ops xfs_seek_iomap_ops;  extern const struct iomap_ops
> > > xfs_xattr_iomap_ops;
> > >
> > > +static inline int
> > > +xfs_iomap_zero_range(
> > > +	struct xfs_inode	*ip,
> > > +	loff_t			offset,
> > > +	loff_t			len,
> > > +	bool			*did_zero)
> > > +{
> > > +	return iomap_zero_range(VFS_I(ip), offset, len, did_zero,
> > > +			IS_DAX(VFS_I(ip)) ? &xfs_dax_write_iomap_ops
> > > +					  : &xfs_buffered_write_iomap_ops); }
> > > +
> > > +static inline int
> > > +xfs_iomap_truncate_page(
> > > +	struct xfs_inode	*ip,
> > > +	loff_t			pos,
> > > +	bool			*did_zero)
> > > +{
> > > +	return iomap_truncate_page(VFS_I(ip), pos, did_zero,
> > > +			IS_DAX(VFS_I(ip)) ? &xfs_dax_write_iomap_ops
> > > +					  : &xfs_buffered_write_iomap_ops); }
> > > +
> > >  #endif /* __XFS_IOMAP_H__*/
> > > diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c index
> > > dfe24b7f26e5..6d936c3e1a6e 100644
> > > --- a/fs/xfs/xfs_iops.c
> > > +++ b/fs/xfs/xfs_iops.c
> > > @@ -911,8 +911,8 @@ xfs_setattr_size(
> > >  	 */
> > >  	if (newsize > oldsize) {
> > >  		trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
> > > -		error = iomap_zero_range(inode, oldsize, newsize - oldsize,
> > > -				&did_zeroing, &xfs_buffered_write_iomap_ops);
> > > +		error = xfs_iomap_zero_range(ip, oldsize, newsize - oldsize,
> > > +				&did_zeroing);
> > >  	} else {
> > >  		/*
> > >  		 * iomap won't detect a dirty page over an unwritten block (or a
> > > @@
> > > -924,8 +924,7 @@ xfs_setattr_size(
> > >  						     newsize);
> > >  		if (error)
> > >  			return error;
> > > -		error = iomap_truncate_page(inode, newsize, &did_zeroing,
> > > -				&xfs_buffered_write_iomap_ops);
> > > +		error = xfs_iomap_truncate_page(ip, newsize, &did_zeroing);
> > >  	}
> > >
> > >  	if (error)
> > > diff --git a/fs/xfs/xfs_reflink.c b/fs/xfs/xfs_reflink.c index
> > > d25434f93235..9a780948dbd0 100644
> > > --- a/fs/xfs/xfs_reflink.c
> > > +++ b/fs/xfs/xfs_reflink.c
> > > @@ -1266,8 +1266,7 @@ xfs_reflink_zero_posteof(
> > >  		return 0;
> > >
> > >  	trace_xfs_zero_eof(ip, isize, pos - isize);
> > > -	return iomap_zero_range(VFS_I(ip), isize, pos - isize, NULL,
> > > -			&xfs_buffered_write_iomap_ops);
> > > +	return xfs_iomap_zero_range(ip, isize, pos - isize, NULL);
> > >  }
> > >
> > >  /*
> > > diff --git a/include/linux/iomap.h b/include/linux/iomap.h index
> > > 95562f863ad0..58f2e1c78018 100644
> > > --- a/include/linux/iomap.h
> > > +++ b/include/linux/iomap.h
> > > @@ -135,6 +135,14 @@ struct iomap_ops {
> > >  			unsigned flags, struct iomap *iomap,
> > >  			struct iomap *srcmap);
> > >
> > > +	/*
> > > +	 * Handle the error code from actor(). Do the finishing jobs for extra
> > > +	 * operations, such as CoW, according to whether written is negative.
> > > +	 */
> > > +	int (*iomap_post_actor)(struct inode *inode, loff_t pos, loff_t length,
> > > +			ssize_t written, unsigned flags, struct iomap *iomap,
> > > +			struct iomap *srcmap);
> > > +
> > >  	/*
> > >  	 * Commit and/or unreserve space previous allocated using
> iomap_begin.
> > >  	 * Written indicates the length of the successful write operation
> > > which
> > > --
> > > 2.31.1
> >

^ permalink raw reply	[flat|nested] 9+ messages in thread

* Re: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path
  2021-06-28  2:55       ` ruansy.fnst
@ 2021-06-28  5:09         ` Darrick J. Wong
  2021-06-29 11:25           ` ruansy.fnst
  2021-07-08 23:16           ` Dave Chinner
  2021-07-09 12:36         ` [PATCH v6.2 6/7] dax: Introduce dax_iomap_ops for end of reflink Shiyang Ruan
  1 sibling, 2 replies; 9+ messages in thread
From: Darrick J. Wong @ 2021-06-28  5:09 UTC (permalink / raw)
  To: ruansy.fnst
  Cc: dan.j.williams, david, hch, linux-fsdevel, linux-kernel, nvdimm,
	linux-xfs, rgoldwyn, viro, willy

On Mon, Jun 28, 2021 at 02:55:03AM +0000, ruansy.fnst@fujitsu.com wrote:
> > -----Original Message-----
> > Subject: Re: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path
> > 
> > On Thu, Jun 24, 2021 at 08:49:17AM +0000, ruansy.fnst@fujitsu.com wrote:
> > > Hi Darrick,
> > >
> > > Do you have any comment on this?
> > 
> > Sorry, was on vacation.
> > 
> > > Thanks,
> > > Ruan.
> > >
> > > > -----Original Message-----
> > > > From: Shiyang Ruan <ruansy.fnst@fujitsu.com>
> > > > Subject: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path
> > > >
> > > > Hi Darrick,
> > > >
> > > > Since other patches looks good, I post this RFC patch singly to
> > > > hot-fix the problem in xfs_dax_write_iomap_ops->iomap_end() of v6
> > > > that the error code was ingored. I will split this in two
> > > > patches(changes in iomap and xfs
> > > > respectively) in next formal version if it looks ok.
> > > >
> > > > ====
> > > >
> > > > Introduce a new interface called "iomap_post_actor()" in iomap_ops.
> > > > And call it between ->actor() and ->iomap_end().  It is mean to
> > > > handle the error code returned from ->actor().  In this patchset, it
> > > > is used to remap or cancel the CoW extents according to the error code.
> > > >
> > > > Signed-off-by: Shiyang Ruan <ruansy.fnst@fujitsu.com>
> > > > ---
> > > >  fs/dax.c               | 27 ++++++++++++++++++---------
> > > >  fs/iomap/apply.c       |  4 ++++
> > > >  fs/xfs/xfs_bmap_util.c |  3 +--
> > > >  fs/xfs/xfs_file.c      |  5 +++--
> > > >  fs/xfs/xfs_iomap.c     | 33 ++++++++++++++++++++++++++++++++-
> > > >  fs/xfs/xfs_iomap.h     | 24 ++++++++++++++++++++++++
> > > >  fs/xfs/xfs_iops.c      |  7 +++----
> > > >  fs/xfs/xfs_reflink.c   |  3 +--
> > > >  include/linux/iomap.h  |  8 ++++++++
> > > >  9 files changed, 94 insertions(+), 20 deletions(-)
> > > >
> > > > diff --git a/fs/dax.c b/fs/dax.c
> > > > index 93f16210847b..0740c2610b6f 100644
> > > > --- a/fs/dax.c
> > > > +++ b/fs/dax.c
> > > > @@ -1537,7 +1537,7 @@ static vm_fault_t dax_iomap_pte_fault(struct
> > > > vm_fault *vmf, pfn_t *pfnp,
> > > >  	struct iomap iomap = { .type = IOMAP_HOLE };
> > > >  	struct iomap srcmap = { .type = IOMAP_HOLE };
> > > >  	unsigned flags = IOMAP_FAULT;
> > > > -	int error;
> > > > +	int error, copied = PAGE_SIZE;
> > > >  	bool write = vmf->flags & FAULT_FLAG_WRITE;
> > > >  	vm_fault_t ret = 0, major = 0;
> > > >  	void *entry;
> > > > @@ -1598,7 +1598,7 @@ static vm_fault_t dax_iomap_pte_fault(struct
> > > > vm_fault *vmf, pfn_t *pfnp,
> > > >  	ret = dax_fault_actor(vmf, pfnp, &xas, &entry, false, flags,
> > > >  			      &iomap, &srcmap);
> > > >  	if (ret == VM_FAULT_SIGBUS)
> > > > -		goto finish_iomap;
> > > > +		goto finish_iomap_actor;
> > > >
> > > >  	/* read/write MAPPED, CoW UNWRITTEN */
> > > >  	if (iomap.flags & IOMAP_F_NEW) {
> > > > @@ -1607,10 +1607,16 @@ static vm_fault_t dax_iomap_pte_fault(struct
> > > > vm_fault *vmf, pfn_t *pfnp,
> > > >  		major = VM_FAULT_MAJOR;
> > > >  	}
> > > >
> > > > + finish_iomap_actor:
> > > > +	if (ops->iomap_post_actor) {
> > > > +		if (ret & VM_FAULT_ERROR)
> > > > +			copied = 0;
> > > > +		ops->iomap_post_actor(inode, pos, PMD_SIZE, copied, flags,
> > > > +				      &iomap, &srcmap);
> > > > +	}
> > > > +
> > > >  finish_iomap:
> > > >  	if (ops->iomap_end) {
> > > > -		int copied = PAGE_SIZE;
> > > > -
> > > >  		if (ret & VM_FAULT_ERROR)
> > > >  			copied = 0;
> > > >  		/*
> > > > @@ -1677,7 +1683,7 @@ static vm_fault_t dax_iomap_pmd_fault(struct
> > > > vm_fault *vmf, pfn_t *pfnp,
> > > >  	pgoff_t max_pgoff;
> > > >  	void *entry;
> > > >  	loff_t pos;
> > > > -	int error;
> > > > +	int error, copied = PMD_SIZE;
> > > >
> > > >  	/*
> > > >  	 * Check whether offset isn't beyond end of file now. Caller is @@
> > > > -1736,12
> > > > +1742,15 @@ static vm_fault_t dax_iomap_pmd_fault(struct vm_fault
> > > > +*vmf,
> > > > pfn_t *pfnp,
> > > >  	ret = dax_fault_actor(vmf, pfnp, &xas, &entry, true, flags,
> > > >  			      &iomap, &srcmap);
> > > >
> > > > +	if (ret == VM_FAULT_FALLBACK)
> > > > +		copied = 0;
> > > > +	if (ops->iomap_post_actor) {
> > > > +		ops->iomap_post_actor(inode, pos, PMD_SIZE, copied, flags,
> > > > +				      &iomap, &srcmap);
> > > > +	}
> > > > +
> > > >  finish_iomap:
> > > >  	if (ops->iomap_end) {
> > > > -		int copied = PMD_SIZE;
> > > > -
> > > > -		if (ret == VM_FAULT_FALLBACK)
> > > > -			copied = 0;
> > > >  		/*
> > > >  		 * The fault is done by now and there's no way back (other
> > > >  		 * thread may be already happily using PMD we have installed).
> > > > diff --git a/fs/iomap/apply.c b/fs/iomap/apply.c index
> > > > 0493da5286ad..26a54ded184f 100644
> > > > --- a/fs/iomap/apply.c
> > > > +++ b/fs/iomap/apply.c
> > > > @@ -84,6 +84,10 @@ iomap_apply(struct inode *inode, loff_t pos,
> > > > loff_t length, unsigned flags,
> > > >  	written = actor(inode, pos, length, data, &iomap,
> > > >  			srcmap.type != IOMAP_HOLE ? &srcmap : &iomap);
> > > >
> > > > +	if (ops->iomap_post_actor) {
> > > > +		written = ops->iomap_post_actor(inode, pos, length, written,
> > > > +						flags, &iomap, &srcmap);
> > 
> > How many operations actually need an iomap_post_actor?  It's just the dax
> > ones, right?  Which is ... iomap_truncate_page, iomap_zero_range,
> > dax_iomap_fault, and dax_iomap_rw, right?  We don't need a post_actor for
> > other iomap functionality (like FIEMAP, SEEK_DATA/SEEK_HOLE, etc.) so adding
> > a new function pointer for all operations feels a bit overbroad.
> 
> Yes.
> 
> > 
> > I had imagined that you'd create a struct dax_iomap_ops to wrap all the extra
> > functionality that you need for dax operations:
> > 
> > struct dax_iomap_ops {
> > 	struct iomap_ops	iomap_ops;
> > 
> > 	int			(*end_io)(inode, pos, length...);
> > };
> > 
> > And alter the four functions that you need to take the special dax_iomap_ops.
> > I guess the downside is that this makes iomap_truncate_page and
> > iomap_zero_range more complicated, but maybe it's just time to split those into
> > DAX-specific versions.  Then we'd be rid of the cross-links betwee
> > fs/iomap/buffered-io.c and fs/dax.c.
> 
> This seems to be a better solution.  I'll try in this way.  Thanks for your guidance.

I started writing on Friday a patchset to apply this style cleanup both
to the directio and dax paths.  The cleanups were pretty straightforward
until I started reading the dax code paths again and realized that file
writes still have the weird behavior of mapping extents into a file,
zeroing them, then issuing the actual write to the extent.  IOWs, a
double-write to avoid exposing stale contents if crash.

Apparently the reason for this was that dax (at least 6 years ago) had
no concept paralleling the page lock, so it was necessary to do that to
avoid page fault handlers racing to map pfns into the file mapping?
That would seem to prevent us from doing the more standard behavior of
allocate unwritten, write data, convert mapping... but is that still the
case?  Or can we get rid of this bad quirk?

--D

> 
> > 
> > > > +	}
> > > >  out:
> > > >  	/*
> > > >  	 * Now the data has been copied, commit the range we've copied.
> > > > This diff --git a/fs/xfs/xfs_bmap_util.c b/fs/xfs/xfs_bmap_util.c
> > > > index
> > > > a5e9d7d34023..2a36dc93ff27 100644
> > > > --- a/fs/xfs/xfs_bmap_util.c
> > > > +++ b/fs/xfs/xfs_bmap_util.c
> > > > @@ -965,8 +965,7 @@ xfs_free_file_space(
> > > >  		return 0;
> > > >  	if (offset + len > XFS_ISIZE(ip))
> > > >  		len = XFS_ISIZE(ip) - offset;
> > > > -	error = iomap_zero_range(VFS_I(ip), offset, len, NULL,
> > > > -			&xfs_buffered_write_iomap_ops);
> > > > +	error = xfs_iomap_zero_range(ip, offset, len, NULL);
> > > >  	if (error)
> > > >  		return error;
> > > >
> > > > diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c index
> > > > 396ef36dcd0a..89406ec6741b
> > > > 100644
> > > > --- a/fs/xfs/xfs_file.c
> > > > +++ b/fs/xfs/xfs_file.c
> > > > @@ -684,11 +684,12 @@ xfs_file_dax_write(
> > > >  	pos = iocb->ki_pos;
> > > >
> > > >  	trace_xfs_file_dax_write(iocb, from);
> > > > -	ret = dax_iomap_rw(iocb, from, &xfs_direct_write_iomap_ops);
> > > > +	ret = dax_iomap_rw(iocb, from, &xfs_dax_write_iomap_ops);
> > > >  	if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
> > > >  		i_size_write(inode, iocb->ki_pos);
> > > >  		error = xfs_setfilesize(ip, pos, ret);
> > > >  	}
> > > > +
> > > >  out:
> > > >  	if (iolock)
> > > >  		xfs_iunlock(ip, iolock);
> > > > @@ -1309,7 +1310,7 @@ __xfs_filemap_fault(
> > > >
> > > >  		ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL,
> > > >  				(write_fault && !vmf->cow_page) ?
> > > > -				 &xfs_direct_write_iomap_ops :
> > > > +				 &xfs_dax_write_iomap_ops :
> > > >  				 &xfs_read_iomap_ops);
> > > >  		if (ret & VM_FAULT_NEEDDSYNC)
> > > >  			ret = dax_finish_sync_fault(vmf, pe_size, pfn); diff --git
> > > > a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c index
> > > > d154f42e2dc6..2f322e2f8544
> > > > 100644
> > > > --- a/fs/xfs/xfs_iomap.c
> > > > +++ b/fs/xfs/xfs_iomap.c
> > > > @@ -761,7 +761,8 @@ xfs_direct_write_iomap_begin(
> > > >
> > > >  		/* may drop and re-acquire the ilock */
> > > >  		error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
> > > > -				&lockmode, flags & IOMAP_DIRECT);
> > > > +				&lockmode,
> > > > +				(flags & IOMAP_DIRECT) || IS_DAX(inode));
> > > >  		if (error)
> > > >  			goto out_unlock;
> > > >  		if (shared)
> > > > @@ -854,6 +855,36 @@ const struct iomap_ops
> > > > xfs_direct_write_iomap_ops = {
> > > >  	.iomap_begin		= xfs_direct_write_iomap_begin,
> > > >  };
> > > >
> > > > +static int
> > > > +xfs_dax_write_iomap_post_actor(
> > > > +	struct inode		*inode,
> > > > +	loff_t			pos,
> > > > +	loff_t			length,
> > > > +	ssize_t			written,
> > > > +	unsigned int		flags,
> > > > +	struct iomap		*iomap,
> > > > +	struct iomap		*srcmap)
> > > > +{
> > > > +	int			error = 0;
> > > > +	struct xfs_inode	*ip = XFS_I(inode);
> > > > +	bool			cow = xfs_is_cow_inode(ip);
> > > > +
> > > > +	if (written <= 0) {
> > > > +		if (cow)
> > > > +			xfs_reflink_cancel_cow_range(ip, pos, length, true);
> > > > +		return written;
> > > > +	}
> > > > +
> > > > +	if (cow)
> > > > +		error = xfs_reflink_end_cow(ip, pos, written);
> > > > +	return error ?: written;
> > > > +}
> > 
> > This is pretty much the same as what xfs_dio_write_end_io does, right?
> 
> It just handles the end part of CoW here.
> xfs_dio_write_end_io() also updates file size, which is only needed in write() but not in page fault.  And the update file size work is done in xfs_dax_file_write(), it's fine, no need to modify it.
> 
> > 
> > I had imagined that you'd change the function signatures to drop the iocb so
> > that you could reuse this code instead of creating a whole new callback.
> > 
> > Ah well.  Can I send you some prep patches to clean up some of the weird
> > iomap code as a preparation series for this?
> 
> Sure.  Thanks.
> 
> 
> --
> Ruan.
> 
> > 
> > --D
> > 
> > > > +
> > > > +const struct iomap_ops xfs_dax_write_iomap_ops = {
> > > > +	.iomap_begin		= xfs_direct_write_iomap_begin,
> > > > +	.iomap_post_actor	= xfs_dax_write_iomap_post_actor,
> > > > +};
> > > > +
> > > >  static int
> > > >  xfs_buffered_write_iomap_begin(
> > > >  	struct inode		*inode,
> > > > diff --git a/fs/xfs/xfs_iomap.h b/fs/xfs/xfs_iomap.h index
> > > > 7d3703556d0e..fbacf638ab21 100644
> > > > --- a/fs/xfs/xfs_iomap.h
> > > > +++ b/fs/xfs/xfs_iomap.h
> > > > @@ -42,8 +42,32 @@ xfs_aligned_fsb_count(
> > > >
> > > >  extern const struct iomap_ops xfs_buffered_write_iomap_ops;  extern
> > > > const struct iomap_ops xfs_direct_write_iomap_ops;
> > > > +extern const struct iomap_ops xfs_dax_write_iomap_ops;
> > > >  extern const struct iomap_ops xfs_read_iomap_ops;  extern const
> > > > struct iomap_ops xfs_seek_iomap_ops;  extern const struct iomap_ops
> > > > xfs_xattr_iomap_ops;
> > > >
> > > > +static inline int
> > > > +xfs_iomap_zero_range(
> > > > +	struct xfs_inode	*ip,
> > > > +	loff_t			offset,
> > > > +	loff_t			len,
> > > > +	bool			*did_zero)
> > > > +{
> > > > +	return iomap_zero_range(VFS_I(ip), offset, len, did_zero,
> > > > +			IS_DAX(VFS_I(ip)) ? &xfs_dax_write_iomap_ops
> > > > +					  : &xfs_buffered_write_iomap_ops); }
> > > > +
> > > > +static inline int
> > > > +xfs_iomap_truncate_page(
> > > > +	struct xfs_inode	*ip,
> > > > +	loff_t			pos,
> > > > +	bool			*did_zero)
> > > > +{
> > > > +	return iomap_truncate_page(VFS_I(ip), pos, did_zero,
> > > > +			IS_DAX(VFS_I(ip)) ? &xfs_dax_write_iomap_ops
> > > > +					  : &xfs_buffered_write_iomap_ops); }
> > > > +
> > > >  #endif /* __XFS_IOMAP_H__*/
> > > > diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c index
> > > > dfe24b7f26e5..6d936c3e1a6e 100644
> > > > --- a/fs/xfs/xfs_iops.c
> > > > +++ b/fs/xfs/xfs_iops.c
> > > > @@ -911,8 +911,8 @@ xfs_setattr_size(
> > > >  	 */
> > > >  	if (newsize > oldsize) {
> > > >  		trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
> > > > -		error = iomap_zero_range(inode, oldsize, newsize - oldsize,
> > > > -				&did_zeroing, &xfs_buffered_write_iomap_ops);
> > > > +		error = xfs_iomap_zero_range(ip, oldsize, newsize - oldsize,
> > > > +				&did_zeroing);
> > > >  	} else {
> > > >  		/*
> > > >  		 * iomap won't detect a dirty page over an unwritten block (or a
> > > > @@
> > > > -924,8 +924,7 @@ xfs_setattr_size(
> > > >  						     newsize);
> > > >  		if (error)
> > > >  			return error;
> > > > -		error = iomap_truncate_page(inode, newsize, &did_zeroing,
> > > > -				&xfs_buffered_write_iomap_ops);
> > > > +		error = xfs_iomap_truncate_page(ip, newsize, &did_zeroing);
> > > >  	}
> > > >
> > > >  	if (error)
> > > > diff --git a/fs/xfs/xfs_reflink.c b/fs/xfs/xfs_reflink.c index
> > > > d25434f93235..9a780948dbd0 100644
> > > > --- a/fs/xfs/xfs_reflink.c
> > > > +++ b/fs/xfs/xfs_reflink.c
> > > > @@ -1266,8 +1266,7 @@ xfs_reflink_zero_posteof(
> > > >  		return 0;
> > > >
> > > >  	trace_xfs_zero_eof(ip, isize, pos - isize);
> > > > -	return iomap_zero_range(VFS_I(ip), isize, pos - isize, NULL,
> > > > -			&xfs_buffered_write_iomap_ops);
> > > > +	return xfs_iomap_zero_range(ip, isize, pos - isize, NULL);
> > > >  }
> > > >
> > > >  /*
> > > > diff --git a/include/linux/iomap.h b/include/linux/iomap.h index
> > > > 95562f863ad0..58f2e1c78018 100644
> > > > --- a/include/linux/iomap.h
> > > > +++ b/include/linux/iomap.h
> > > > @@ -135,6 +135,14 @@ struct iomap_ops {
> > > >  			unsigned flags, struct iomap *iomap,
> > > >  			struct iomap *srcmap);
> > > >
> > > > +	/*
> > > > +	 * Handle the error code from actor(). Do the finishing jobs for extra
> > > > +	 * operations, such as CoW, according to whether written is negative.
> > > > +	 */
> > > > +	int (*iomap_post_actor)(struct inode *inode, loff_t pos, loff_t length,
> > > > +			ssize_t written, unsigned flags, struct iomap *iomap,
> > > > +			struct iomap *srcmap);
> > > > +
> > > >  	/*
> > > >  	 * Commit and/or unreserve space previous allocated using
> > iomap_begin.
> > > >  	 * Written indicates the length of the successful write operation
> > > > which
> > > > --
> > > > 2.31.1
> > >

^ permalink raw reply	[flat|nested] 9+ messages in thread

* RE: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path
  2021-06-28  5:09         ` Darrick J. Wong
@ 2021-06-29 11:25           ` ruansy.fnst
  2021-06-29 21:01             ` Darrick J. Wong
  2021-07-08 23:16           ` Dave Chinner
  1 sibling, 1 reply; 9+ messages in thread
From: ruansy.fnst @ 2021-06-29 11:25 UTC (permalink / raw)
  To: Darrick J. Wong
  Cc: dan.j.williams, david, hch, linux-fsdevel, linux-kernel, nvdimm,
	linux-xfs, rgoldwyn, viro, willy



> -----Original Message-----
> From: Darrick J. Wong <djwong@kernel.org>
> Subject: Re: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path
> 
> On Mon, Jun 28, 2021 at 02:55:03AM +0000, ruansy.fnst@fujitsu.com wrote:
> > > -----Original Message-----
> > > Subject: Re: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write()
> > > path
> > >
> > > On Thu, Jun 24, 2021 at 08:49:17AM +0000, ruansy.fnst@fujitsu.com wrote:
> > > > Hi Darrick,
> > > >
> > > > Do you have any comment on this?
> > >
> > > Sorry, was on vacation.
> > >
> > > > Thanks,
> > > > Ruan.
> > > >
> > > > > -----Original Message-----
> > > > > From: Shiyang Ruan <ruansy.fnst@fujitsu.com>
> > > > > Subject: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write()
> > > > > path
> > > > >
> > > > > Hi Darrick,
> > > > >
> > > > > Since other patches looks good, I post this RFC patch singly to
> > > > > hot-fix the problem in xfs_dax_write_iomap_ops->iomap_end() of
> > > > > v6 that the error code was ingored. I will split this in two
> > > > > patches(changes in iomap and xfs
> > > > > respectively) in next formal version if it looks ok.
> > > > >
> > > > > ====
> > > > >
> > > > > Introduce a new interface called "iomap_post_actor()" in iomap_ops.
> > > > > And call it between ->actor() and ->iomap_end().  It is mean to
> > > > > handle the error code returned from ->actor().  In this
> > > > > patchset, it is used to remap or cancel the CoW extents according to the
> error code.
> > > > >
> > > > > Signed-off-by: Shiyang Ruan <ruansy.fnst@fujitsu.com>
> > > > > ---
> > > > >  fs/dax.c               | 27 ++++++++++++++++++---------
> > > > >  fs/iomap/apply.c       |  4 ++++
> > > > >  fs/xfs/xfs_bmap_util.c |  3 +--
> > > > >  fs/xfs/xfs_file.c      |  5 +++--
> > > > >  fs/xfs/xfs_iomap.c     | 33 ++++++++++++++++++++++++++++++++-
> > > > >  fs/xfs/xfs_iomap.h     | 24 ++++++++++++++++++++++++
> > > > >  fs/xfs/xfs_iops.c      |  7 +++----
> > > > >  fs/xfs/xfs_reflink.c   |  3 +--
> > > > >  include/linux/iomap.h  |  8 ++++++++
> > > > >  9 files changed, 94 insertions(+), 20 deletions(-)
> > > > >
> > > > > diff --git a/fs/dax.c b/fs/dax.c index
> > > > > 93f16210847b..0740c2610b6f 100644
> > > > > --- a/fs/dax.c
> > > > > +++ b/fs/dax.c
> > > > > @@ -1537,7 +1537,7 @@ static vm_fault_t
> > > > > dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
> > > > >  	struct iomap iomap = { .type = IOMAP_HOLE };
> > > > >  	struct iomap srcmap = { .type = IOMAP_HOLE };
> > > > >  	unsigned flags = IOMAP_FAULT;
> > > > > -	int error;
> > > > > +	int error, copied = PAGE_SIZE;
> > > > >  	bool write = vmf->flags & FAULT_FLAG_WRITE;
> > > > >  	vm_fault_t ret = 0, major = 0;
> > > > >  	void *entry;
> > > > > @@ -1598,7 +1598,7 @@ static vm_fault_t
> > > > > dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
> > > > >  	ret = dax_fault_actor(vmf, pfnp, &xas, &entry, false, flags,
> > > > >  			      &iomap, &srcmap);
> > > > >  	if (ret == VM_FAULT_SIGBUS)
> > > > > -		goto finish_iomap;
> > > > > +		goto finish_iomap_actor;
> > > > >
> > > > >  	/* read/write MAPPED, CoW UNWRITTEN */
> > > > >  	if (iomap.flags & IOMAP_F_NEW) { @@ -1607,10 +1607,16 @@
> > > > > static vm_fault_t dax_iomap_pte_fault(struct vm_fault *vmf,
> > > > > pfn_t *pfnp,
> > > > >  		major = VM_FAULT_MAJOR;
> > > > >  	}
> > > > >
> > > > > + finish_iomap_actor:
> > > > > +	if (ops->iomap_post_actor) {
> > > > > +		if (ret & VM_FAULT_ERROR)
> > > > > +			copied = 0;
> > > > > +		ops->iomap_post_actor(inode, pos, PMD_SIZE, copied, flags,
> > > > > +				      &iomap, &srcmap);
> > > > > +	}
> > > > > +
> > > > >  finish_iomap:
> > > > >  	if (ops->iomap_end) {
> > > > > -		int copied = PAGE_SIZE;
> > > > > -
> > > > >  		if (ret & VM_FAULT_ERROR)
> > > > >  			copied = 0;
> > > > >  		/*
> > > > > @@ -1677,7 +1683,7 @@ static vm_fault_t
> > > > > dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp,
> > > > >  	pgoff_t max_pgoff;
> > > > >  	void *entry;
> > > > >  	loff_t pos;
> > > > > -	int error;
> > > > > +	int error, copied = PMD_SIZE;
> > > > >
> > > > >  	/*
> > > > >  	 * Check whether offset isn't beyond end of file now. Caller
> > > > > is @@
> > > > > -1736,12
> > > > > +1742,15 @@ static vm_fault_t dax_iomap_pmd_fault(struct
> > > > > +vm_fault *vmf,
> > > > > pfn_t *pfnp,
> > > > >  	ret = dax_fault_actor(vmf, pfnp, &xas, &entry, true, flags,
> > > > >  			      &iomap, &srcmap);
> > > > >
> > > > > +	if (ret == VM_FAULT_FALLBACK)
> > > > > +		copied = 0;
> > > > > +	if (ops->iomap_post_actor) {
> > > > > +		ops->iomap_post_actor(inode, pos, PMD_SIZE, copied, flags,
> > > > > +				      &iomap, &srcmap);
> > > > > +	}
> > > > > +
> > > > >  finish_iomap:
> > > > >  	if (ops->iomap_end) {
> > > > > -		int copied = PMD_SIZE;
> > > > > -
> > > > > -		if (ret == VM_FAULT_FALLBACK)
> > > > > -			copied = 0;
> > > > >  		/*
> > > > >  		 * The fault is done by now and there's no way back (other
> > > > >  		 * thread may be already happily using PMD we have installed).
> > > > > diff --git a/fs/iomap/apply.c b/fs/iomap/apply.c index
> > > > > 0493da5286ad..26a54ded184f 100644
> > > > > --- a/fs/iomap/apply.c
> > > > > +++ b/fs/iomap/apply.c
> > > > > @@ -84,6 +84,10 @@ iomap_apply(struct inode *inode, loff_t pos,
> > > > > loff_t length, unsigned flags,
> > > > >  	written = actor(inode, pos, length, data, &iomap,
> > > > >  			srcmap.type != IOMAP_HOLE ? &srcmap : &iomap);
> > > > >
> > > > > +	if (ops->iomap_post_actor) {
> > > > > +		written = ops->iomap_post_actor(inode, pos, length, written,
> > > > > +						flags, &iomap, &srcmap);
> > >
> > > How many operations actually need an iomap_post_actor?  It's just
> > > the dax ones, right?  Which is ... iomap_truncate_page,
> > > iomap_zero_range, dax_iomap_fault, and dax_iomap_rw, right?  We
> > > don't need a post_actor for other iomap functionality (like FIEMAP,
> > > SEEK_DATA/SEEK_HOLE, etc.) so adding a new function pointer for all
> operations feels a bit overbroad.
> >
> > Yes.
> >
> > >
> > > I had imagined that you'd create a struct dax_iomap_ops to wrap all
> > > the extra functionality that you need for dax operations:
> > >
> > > struct dax_iomap_ops {
> > > 	struct iomap_ops	iomap_ops;
> > >
> > > 	int			(*end_io)(inode, pos, length...);
> > > };
> > >
> > > And alter the four functions that you need to take the special
> dax_iomap_ops.
> > > I guess the downside is that this makes iomap_truncate_page and
> > > iomap_zero_range more complicated, but maybe it's just time to split
> > > those into DAX-specific versions.  Then we'd be rid of the
> > > cross-links betwee fs/iomap/buffered-io.c and fs/dax.c.
> >
> > This seems to be a better solution.  I'll try in this way.  Thanks for your
> guidance.
> 
> I started writing on Friday a patchset to apply this style cleanup both to the
> directio and dax paths.  The cleanups were pretty straightforward until I
> started reading the dax code paths again and realized that file writes still have
> the weird behavior of mapping extents into a file, zeroing them, then issuing the
> actual write to the extent.  IOWs, a double-write to avoid exposing stale
> contents if crash.

The current code seems not zeroing an unwritten extent when writing in fsdax mode?  Just allocate unwritten extents in filesystem, and then write data in fsdax.

> 
> Apparently the reason for this was that dax (at least 6 years ago) had no
> concept paralleling the page lock, so it was necessary to do that to avoid page
> fault handlers racing to map pfns into the file mapping?
> That would seem to prevent us from doing the more standard behavior of
> allocate unwritten, write data, convert mapping... but is that still the case?  Or
> can we get rid of this bad quirk?

I am not sure about this...


--
Thanks,
Ruan.

> 
> --D
> 
> >
> > >
> > > > > +	}
> > > > >  out:
> > > > >  	/*
> > > > >  	 * Now the data has been copied, commit the range we've copied.
> > > > > This diff --git a/fs/xfs/xfs_bmap_util.c
> > > > > b/fs/xfs/xfs_bmap_util.c index
> > > > > a5e9d7d34023..2a36dc93ff27 100644
> > > > > --- a/fs/xfs/xfs_bmap_util.c
> > > > > +++ b/fs/xfs/xfs_bmap_util.c
> > > > > @@ -965,8 +965,7 @@ xfs_free_file_space(
> > > > >  		return 0;
> > > > >  	if (offset + len > XFS_ISIZE(ip))
> > > > >  		len = XFS_ISIZE(ip) - offset;
> > > > > -	error = iomap_zero_range(VFS_I(ip), offset, len, NULL,
> > > > > -			&xfs_buffered_write_iomap_ops);
> > > > > +	error = xfs_iomap_zero_range(ip, offset, len, NULL);
> > > > >  	if (error)
> > > > >  		return error;
> > > > >
> > > > > diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c index
> > > > > 396ef36dcd0a..89406ec6741b
> > > > > 100644
> > > > > --- a/fs/xfs/xfs_file.c
> > > > > +++ b/fs/xfs/xfs_file.c
> > > > > @@ -684,11 +684,12 @@ xfs_file_dax_write(
> > > > >  	pos = iocb->ki_pos;
> > > > >
> > > > >  	trace_xfs_file_dax_write(iocb, from);
> > > > > -	ret = dax_iomap_rw(iocb, from, &xfs_direct_write_iomap_ops);
> > > > > +	ret = dax_iomap_rw(iocb, from, &xfs_dax_write_iomap_ops);
> > > > >  	if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
> > > > >  		i_size_write(inode, iocb->ki_pos);
> > > > >  		error = xfs_setfilesize(ip, pos, ret);
> > > > >  	}
> > > > > +
> > > > >  out:
> > > > >  	if (iolock)
> > > > >  		xfs_iunlock(ip, iolock);
> > > > > @@ -1309,7 +1310,7 @@ __xfs_filemap_fault(
> > > > >
> > > > >  		ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL,
> > > > >  				(write_fault && !vmf->cow_page) ?
> > > > > -				 &xfs_direct_write_iomap_ops :
> > > > > +				 &xfs_dax_write_iomap_ops :
> > > > >  				 &xfs_read_iomap_ops);
> > > > >  		if (ret & VM_FAULT_NEEDDSYNC)
> > > > >  			ret = dax_finish_sync_fault(vmf, pe_size, pfn); diff --git
> > > > > a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c index
> > > > > d154f42e2dc6..2f322e2f8544
> > > > > 100644
> > > > > --- a/fs/xfs/xfs_iomap.c
> > > > > +++ b/fs/xfs/xfs_iomap.c
> > > > > @@ -761,7 +761,8 @@ xfs_direct_write_iomap_begin(
> > > > >
> > > > >  		/* may drop and re-acquire the ilock */
> > > > >  		error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
> > > > > -				&lockmode, flags & IOMAP_DIRECT);
> > > > > +				&lockmode,
> > > > > +				(flags & IOMAP_DIRECT) || IS_DAX(inode));
> > > > >  		if (error)
> > > > >  			goto out_unlock;
> > > > >  		if (shared)
> > > > > @@ -854,6 +855,36 @@ const struct iomap_ops
> > > > > xfs_direct_write_iomap_ops = {
> > > > >  	.iomap_begin		= xfs_direct_write_iomap_begin,
> > > > >  };
> > > > >
> > > > > +static int
> > > > > +xfs_dax_write_iomap_post_actor(
> > > > > +	struct inode		*inode,
> > > > > +	loff_t			pos,
> > > > > +	loff_t			length,
> > > > > +	ssize_t			written,
> > > > > +	unsigned int		flags,
> > > > > +	struct iomap		*iomap,
> > > > > +	struct iomap		*srcmap)
> > > > > +{
> > > > > +	int			error = 0;
> > > > > +	struct xfs_inode	*ip = XFS_I(inode);
> > > > > +	bool			cow = xfs_is_cow_inode(ip);
> > > > > +
> > > > > +	if (written <= 0) {
> > > > > +		if (cow)
> > > > > +			xfs_reflink_cancel_cow_range(ip, pos, length, true);
> > > > > +		return written;
> > > > > +	}
> > > > > +
> > > > > +	if (cow)
> > > > > +		error = xfs_reflink_end_cow(ip, pos, written);
> > > > > +	return error ?: written;
> > > > > +}
> > >
> > > This is pretty much the same as what xfs_dio_write_end_io does, right?
> >
> > It just handles the end part of CoW here.
> > xfs_dio_write_end_io() also updates file size, which is only needed in write()
> but not in page fault.  And the update file size work is done in
> xfs_dax_file_write(), it's fine, no need to modify it.
> >
> > >
> > > I had imagined that you'd change the function signatures to drop the
> > > iocb so that you could reuse this code instead of creating a whole new
> callback.
> > >
> > > Ah well.  Can I send you some prep patches to clean up some of the
> > > weird iomap code as a preparation series for this?
> >
> > Sure.  Thanks.
> >
> >
> > --
> > Ruan.
> >
> > >
> > > --D
> > >
> > > > > +
> > > > > +const struct iomap_ops xfs_dax_write_iomap_ops = {
> > > > > +	.iomap_begin		= xfs_direct_write_iomap_begin,
> > > > > +	.iomap_post_actor	= xfs_dax_write_iomap_post_actor,
> > > > > +};
> > > > > +
> > > > >  static int
> > > > >  xfs_buffered_write_iomap_begin(
> > > > >  	struct inode		*inode,
> > > > > diff --git a/fs/xfs/xfs_iomap.h b/fs/xfs/xfs_iomap.h index
> > > > > 7d3703556d0e..fbacf638ab21 100644
> > > > > --- a/fs/xfs/xfs_iomap.h
> > > > > +++ b/fs/xfs/xfs_iomap.h
> > > > > @@ -42,8 +42,32 @@ xfs_aligned_fsb_count(
> > > > >
> > > > >  extern const struct iomap_ops xfs_buffered_write_iomap_ops;
> > > > > extern const struct iomap_ops xfs_direct_write_iomap_ops;
> > > > > +extern const struct iomap_ops xfs_dax_write_iomap_ops;
> > > > >  extern const struct iomap_ops xfs_read_iomap_ops;  extern const
> > > > > struct iomap_ops xfs_seek_iomap_ops;  extern const struct
> > > > > iomap_ops xfs_xattr_iomap_ops;
> > > > >
> > > > > +static inline int
> > > > > +xfs_iomap_zero_range(
> > > > > +	struct xfs_inode	*ip,
> > > > > +	loff_t			offset,
> > > > > +	loff_t			len,
> > > > > +	bool			*did_zero)
> > > > > +{
> > > > > +	return iomap_zero_range(VFS_I(ip), offset, len, did_zero,
> > > > > +			IS_DAX(VFS_I(ip)) ? &xfs_dax_write_iomap_ops
> > > > > +					  : &xfs_buffered_write_iomap_ops); }
> > > > > +
> > > > > +static inline int
> > > > > +xfs_iomap_truncate_page(
> > > > > +	struct xfs_inode	*ip,
> > > > > +	loff_t			pos,
> > > > > +	bool			*did_zero)
> > > > > +{
> > > > > +	return iomap_truncate_page(VFS_I(ip), pos, did_zero,
> > > > > +			IS_DAX(VFS_I(ip)) ? &xfs_dax_write_iomap_ops
> > > > > +					  : &xfs_buffered_write_iomap_ops); }
> > > > > +
> > > > >  #endif /* __XFS_IOMAP_H__*/
> > > > > diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c index
> > > > > dfe24b7f26e5..6d936c3e1a6e 100644
> > > > > --- a/fs/xfs/xfs_iops.c
> > > > > +++ b/fs/xfs/xfs_iops.c
> > > > > @@ -911,8 +911,8 @@ xfs_setattr_size(
> > > > >  	 */
> > > > >  	if (newsize > oldsize) {
> > > > >  		trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
> > > > > -		error = iomap_zero_range(inode, oldsize, newsize - oldsize,
> > > > > -				&did_zeroing, &xfs_buffered_write_iomap_ops);
> > > > > +		error = xfs_iomap_zero_range(ip, oldsize, newsize - oldsize,
> > > > > +				&did_zeroing);
> > > > >  	} else {
> > > > >  		/*
> > > > >  		 * iomap won't detect a dirty page over an unwritten block
> > > > > (or a @@
> > > > > -924,8 +924,7 @@ xfs_setattr_size(
> > > > >  						     newsize);
> > > > >  		if (error)
> > > > >  			return error;
> > > > > -		error = iomap_truncate_page(inode, newsize, &did_zeroing,
> > > > > -				&xfs_buffered_write_iomap_ops);
> > > > > +		error = xfs_iomap_truncate_page(ip, newsize, &did_zeroing);
> > > > >  	}
> > > > >
> > > > >  	if (error)
> > > > > diff --git a/fs/xfs/xfs_reflink.c b/fs/xfs/xfs_reflink.c index
> > > > > d25434f93235..9a780948dbd0 100644
> > > > > --- a/fs/xfs/xfs_reflink.c
> > > > > +++ b/fs/xfs/xfs_reflink.c
> > > > > @@ -1266,8 +1266,7 @@ xfs_reflink_zero_posteof(
> > > > >  		return 0;
> > > > >
> > > > >  	trace_xfs_zero_eof(ip, isize, pos - isize);
> > > > > -	return iomap_zero_range(VFS_I(ip), isize, pos - isize, NULL,
> > > > > -			&xfs_buffered_write_iomap_ops);
> > > > > +	return xfs_iomap_zero_range(ip, isize, pos - isize, NULL);
> > > > >  }
> > > > >
> > > > >  /*
> > > > > diff --git a/include/linux/iomap.h b/include/linux/iomap.h index
> > > > > 95562f863ad0..58f2e1c78018 100644
> > > > > --- a/include/linux/iomap.h
> > > > > +++ b/include/linux/iomap.h
> > > > > @@ -135,6 +135,14 @@ struct iomap_ops {
> > > > >  			unsigned flags, struct iomap *iomap,
> > > > >  			struct iomap *srcmap);
> > > > >
> > > > > +	/*
> > > > > +	 * Handle the error code from actor(). Do the finishing jobs for extra
> > > > > +	 * operations, such as CoW, according to whether written is negative.
> > > > > +	 */
> > > > > +	int (*iomap_post_actor)(struct inode *inode, loff_t pos, loff_t length,
> > > > > +			ssize_t written, unsigned flags, struct iomap *iomap,
> > > > > +			struct iomap *srcmap);
> > > > > +
> > > > >  	/*
> > > > >  	 * Commit and/or unreserve space previous allocated using
> > > iomap_begin.
> > > > >  	 * Written indicates the length of the successful write
> > > > > operation which
> > > > > --
> > > > > 2.31.1
> > > >

^ permalink raw reply	[flat|nested] 9+ messages in thread

* Re: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path
  2021-06-29 11:25           ` ruansy.fnst
@ 2021-06-29 21:01             ` Darrick J. Wong
  0 siblings, 0 replies; 9+ messages in thread
From: Darrick J. Wong @ 2021-06-29 21:01 UTC (permalink / raw)
  To: ruansy.fnst
  Cc: dan.j.williams, david, hch, linux-fsdevel, linux-kernel, nvdimm,
	linux-xfs, rgoldwyn, viro, willy

On Tue, Jun 29, 2021 at 11:25:37AM +0000, ruansy.fnst@fujitsu.com wrote:
> 
> 
> > -----Original Message-----
> > From: Darrick J. Wong <djwong@kernel.org>
> > Subject: Re: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path
> > 
> > On Mon, Jun 28, 2021 at 02:55:03AM +0000, ruansy.fnst@fujitsu.com wrote:
> > > > -----Original Message-----
> > > > Subject: Re: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write()
> > > > path
> > > >
> > > > On Thu, Jun 24, 2021 at 08:49:17AM +0000, ruansy.fnst@fujitsu.com wrote:
> > > > > Hi Darrick,
> > > > >
> > > > > Do you have any comment on this?
> > > >
> > > > Sorry, was on vacation.
> > > >
> > > > > Thanks,
> > > > > Ruan.
> > > > >
> > > > > > -----Original Message-----
> > > > > > From: Shiyang Ruan <ruansy.fnst@fujitsu.com>
> > > > > > Subject: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write()
> > > > > > path
> > > > > >
> > > > > > Hi Darrick,
> > > > > >
> > > > > > Since other patches looks good, I post this RFC patch singly to
> > > > > > hot-fix the problem in xfs_dax_write_iomap_ops->iomap_end() of
> > > > > > v6 that the error code was ingored. I will split this in two
> > > > > > patches(changes in iomap and xfs
> > > > > > respectively) in next formal version if it looks ok.
> > > > > >
> > > > > > ====
> > > > > >
> > > > > > Introduce a new interface called "iomap_post_actor()" in iomap_ops.
> > > > > > And call it between ->actor() and ->iomap_end().  It is mean to
> > > > > > handle the error code returned from ->actor().  In this
> > > > > > patchset, it is used to remap or cancel the CoW extents according to the
> > error code.
> > > > > >
> > > > > > Signed-off-by: Shiyang Ruan <ruansy.fnst@fujitsu.com>
> > > > > > ---
> > > > > >  fs/dax.c               | 27 ++++++++++++++++++---------
> > > > > >  fs/iomap/apply.c       |  4 ++++
> > > > > >  fs/xfs/xfs_bmap_util.c |  3 +--
> > > > > >  fs/xfs/xfs_file.c      |  5 +++--
> > > > > >  fs/xfs/xfs_iomap.c     | 33 ++++++++++++++++++++++++++++++++-
> > > > > >  fs/xfs/xfs_iomap.h     | 24 ++++++++++++++++++++++++
> > > > > >  fs/xfs/xfs_iops.c      |  7 +++----
> > > > > >  fs/xfs/xfs_reflink.c   |  3 +--
> > > > > >  include/linux/iomap.h  |  8 ++++++++
> > > > > >  9 files changed, 94 insertions(+), 20 deletions(-)
> > > > > >
> > > > > > diff --git a/fs/dax.c b/fs/dax.c index
> > > > > > 93f16210847b..0740c2610b6f 100644
> > > > > > --- a/fs/dax.c
> > > > > > +++ b/fs/dax.c
> > > > > > @@ -1537,7 +1537,7 @@ static vm_fault_t
> > > > > > dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
> > > > > >  	struct iomap iomap = { .type = IOMAP_HOLE };
> > > > > >  	struct iomap srcmap = { .type = IOMAP_HOLE };
> > > > > >  	unsigned flags = IOMAP_FAULT;
> > > > > > -	int error;
> > > > > > +	int error, copied = PAGE_SIZE;
> > > > > >  	bool write = vmf->flags & FAULT_FLAG_WRITE;
> > > > > >  	vm_fault_t ret = 0, major = 0;
> > > > > >  	void *entry;
> > > > > > @@ -1598,7 +1598,7 @@ static vm_fault_t
> > > > > > dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
> > > > > >  	ret = dax_fault_actor(vmf, pfnp, &xas, &entry, false, flags,
> > > > > >  			      &iomap, &srcmap);
> > > > > >  	if (ret == VM_FAULT_SIGBUS)
> > > > > > -		goto finish_iomap;
> > > > > > +		goto finish_iomap_actor;
> > > > > >
> > > > > >  	/* read/write MAPPED, CoW UNWRITTEN */
> > > > > >  	if (iomap.flags & IOMAP_F_NEW) { @@ -1607,10 +1607,16 @@
> > > > > > static vm_fault_t dax_iomap_pte_fault(struct vm_fault *vmf,
> > > > > > pfn_t *pfnp,
> > > > > >  		major = VM_FAULT_MAJOR;
> > > > > >  	}
> > > > > >
> > > > > > + finish_iomap_actor:
> > > > > > +	if (ops->iomap_post_actor) {
> > > > > > +		if (ret & VM_FAULT_ERROR)
> > > > > > +			copied = 0;
> > > > > > +		ops->iomap_post_actor(inode, pos, PMD_SIZE, copied, flags,
> > > > > > +				      &iomap, &srcmap);
> > > > > > +	}
> > > > > > +
> > > > > >  finish_iomap:
> > > > > >  	if (ops->iomap_end) {
> > > > > > -		int copied = PAGE_SIZE;
> > > > > > -
> > > > > >  		if (ret & VM_FAULT_ERROR)
> > > > > >  			copied = 0;
> > > > > >  		/*
> > > > > > @@ -1677,7 +1683,7 @@ static vm_fault_t
> > > > > > dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp,
> > > > > >  	pgoff_t max_pgoff;
> > > > > >  	void *entry;
> > > > > >  	loff_t pos;
> > > > > > -	int error;
> > > > > > +	int error, copied = PMD_SIZE;
> > > > > >
> > > > > >  	/*
> > > > > >  	 * Check whether offset isn't beyond end of file now. Caller
> > > > > > is @@
> > > > > > -1736,12
> > > > > > +1742,15 @@ static vm_fault_t dax_iomap_pmd_fault(struct
> > > > > > +vm_fault *vmf,
> > > > > > pfn_t *pfnp,
> > > > > >  	ret = dax_fault_actor(vmf, pfnp, &xas, &entry, true, flags,
> > > > > >  			      &iomap, &srcmap);
> > > > > >
> > > > > > +	if (ret == VM_FAULT_FALLBACK)
> > > > > > +		copied = 0;
> > > > > > +	if (ops->iomap_post_actor) {
> > > > > > +		ops->iomap_post_actor(inode, pos, PMD_SIZE, copied, flags,
> > > > > > +				      &iomap, &srcmap);
> > > > > > +	}
> > > > > > +
> > > > > >  finish_iomap:
> > > > > >  	if (ops->iomap_end) {
> > > > > > -		int copied = PMD_SIZE;
> > > > > > -
> > > > > > -		if (ret == VM_FAULT_FALLBACK)
> > > > > > -			copied = 0;
> > > > > >  		/*
> > > > > >  		 * The fault is done by now and there's no way back (other
> > > > > >  		 * thread may be already happily using PMD we have installed).
> > > > > > diff --git a/fs/iomap/apply.c b/fs/iomap/apply.c index
> > > > > > 0493da5286ad..26a54ded184f 100644
> > > > > > --- a/fs/iomap/apply.c
> > > > > > +++ b/fs/iomap/apply.c
> > > > > > @@ -84,6 +84,10 @@ iomap_apply(struct inode *inode, loff_t pos,
> > > > > > loff_t length, unsigned flags,
> > > > > >  	written = actor(inode, pos, length, data, &iomap,
> > > > > >  			srcmap.type != IOMAP_HOLE ? &srcmap : &iomap);
> > > > > >
> > > > > > +	if (ops->iomap_post_actor) {
> > > > > > +		written = ops->iomap_post_actor(inode, pos, length, written,
> > > > > > +						flags, &iomap, &srcmap);
> > > >
> > > > How many operations actually need an iomap_post_actor?  It's just
> > > > the dax ones, right?  Which is ... iomap_truncate_page,
> > > > iomap_zero_range, dax_iomap_fault, and dax_iomap_rw, right?  We
> > > > don't need a post_actor for other iomap functionality (like FIEMAP,
> > > > SEEK_DATA/SEEK_HOLE, etc.) so adding a new function pointer for all
> > operations feels a bit overbroad.
> > >
> > > Yes.
> > >
> > > >
> > > > I had imagined that you'd create a struct dax_iomap_ops to wrap all
> > > > the extra functionality that you need for dax operations:
> > > >
> > > > struct dax_iomap_ops {
> > > > 	struct iomap_ops	iomap_ops;
> > > >
> > > > 	int			(*end_io)(inode, pos, length...);
> > > > };
> > > >
> > > > And alter the four functions that you need to take the special
> > dax_iomap_ops.
> > > > I guess the downside is that this makes iomap_truncate_page and
> > > > iomap_zero_range more complicated, but maybe it's just time to split
> > > > those into DAX-specific versions.  Then we'd be rid of the
> > > > cross-links betwee fs/iomap/buffered-io.c and fs/dax.c.
> > >
> > > This seems to be a better solution.  I'll try in this way.  Thanks for your
> > guidance.
> > 
> > I started writing on Friday a patchset to apply this style cleanup both to the
> > directio and dax paths.  The cleanups were pretty straightforward until I
> > started reading the dax code paths again and realized that file writes still have
> > the weird behavior of mapping extents into a file, zeroing them, then issuing the
> > actual write to the extent.  IOWs, a double-write to avoid exposing stale
> > contents if crash.
> 
> The current code seems not zeroing an unwritten extent when writing in
> fsdax mode?  Just allocate unwritten extents in filesystem, and then
> write data in fsdax.

That's not what it does.  See xfs_iomap_write_direct:

	/*
	 * For DAX, we do not allocate unwritten extents, but instead we
	 * zero the block before we commit the transaction.  Ideally
	 * we'd like to do this outside the transaction context, but if
	 * we commit and then crash we may not have zeroed the blocks
	 * and this will be exposed on recovery of the allocation. Hence
	 * we must zero before commit.
	 *
	 * Further, if we are mapping unwritten extents here, we need to
	 * zero and convert them to written so that we don't need an
	 * unwritten extent callback for DAX. This also means that we
	 * need to be able to dip into the reserve block pool for bmbt
	 * block allocation if there is no space left but we need to do
	 * unwritten extent conversion.
	 */
	if (IS_DAX(VFS_I(ip))) {
		bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
		if (imap->br_state == XFS_EXT_UNWRITTEN) {
			force = true;
			nr_exts = XFS_IEXT_WRITE_UNWRITTEN_CNT;
			dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
		}
	}

Originally added in:

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=1ca191576fc862b4766f58e41aa362b28a7c1866

Of course, that was six years ago when the mm folks were still arguing
about whether they'd have struct page or pfns or some combination of the
two for DAX.  I'm not sure if those limitations still exist, or if they
quietly disappeared and xfs/ext4/ext2 haven't noticed.  I think I see
that we store pfns in i_pages along with a lock bit, but I don't know if
that lock bit is sufficient to prevent races between page faults.

Hence my question below:
> 
> > 
> > Apparently the reason for this was that dax (at least 6 years ago) had no
> > concept paralleling the page lock, so it was necessary to do that to avoid page
> > fault handlers racing to map pfns into the file mapping?
> > That would seem to prevent us from doing the more standard behavior of
> > allocate unwritten, write data, convert mapping... but is that still the case?  Or
> > can we get rid of this bad quirk?
> 
> I am not sure about this...

Me neither.

--D

> 
> 
> --
> Thanks,
> Ruan.
> 
> > 
> > --D
> > 
> > >
> > > >
> > > > > > +	}
> > > > > >  out:
> > > > > >  	/*
> > > > > >  	 * Now the data has been copied, commit the range we've copied.
> > > > > > This diff --git a/fs/xfs/xfs_bmap_util.c
> > > > > > b/fs/xfs/xfs_bmap_util.c index
> > > > > > a5e9d7d34023..2a36dc93ff27 100644
> > > > > > --- a/fs/xfs/xfs_bmap_util.c
> > > > > > +++ b/fs/xfs/xfs_bmap_util.c
> > > > > > @@ -965,8 +965,7 @@ xfs_free_file_space(
> > > > > >  		return 0;
> > > > > >  	if (offset + len > XFS_ISIZE(ip))
> > > > > >  		len = XFS_ISIZE(ip) - offset;
> > > > > > -	error = iomap_zero_range(VFS_I(ip), offset, len, NULL,
> > > > > > -			&xfs_buffered_write_iomap_ops);
> > > > > > +	error = xfs_iomap_zero_range(ip, offset, len, NULL);
> > > > > >  	if (error)
> > > > > >  		return error;
> > > > > >
> > > > > > diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c index
> > > > > > 396ef36dcd0a..89406ec6741b
> > > > > > 100644
> > > > > > --- a/fs/xfs/xfs_file.c
> > > > > > +++ b/fs/xfs/xfs_file.c
> > > > > > @@ -684,11 +684,12 @@ xfs_file_dax_write(
> > > > > >  	pos = iocb->ki_pos;
> > > > > >
> > > > > >  	trace_xfs_file_dax_write(iocb, from);
> > > > > > -	ret = dax_iomap_rw(iocb, from, &xfs_direct_write_iomap_ops);
> > > > > > +	ret = dax_iomap_rw(iocb, from, &xfs_dax_write_iomap_ops);
> > > > > >  	if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
> > > > > >  		i_size_write(inode, iocb->ki_pos);
> > > > > >  		error = xfs_setfilesize(ip, pos, ret);
> > > > > >  	}
> > > > > > +
> > > > > >  out:
> > > > > >  	if (iolock)
> > > > > >  		xfs_iunlock(ip, iolock);
> > > > > > @@ -1309,7 +1310,7 @@ __xfs_filemap_fault(
> > > > > >
> > > > > >  		ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL,
> > > > > >  				(write_fault && !vmf->cow_page) ?
> > > > > > -				 &xfs_direct_write_iomap_ops :
> > > > > > +				 &xfs_dax_write_iomap_ops :
> > > > > >  				 &xfs_read_iomap_ops);
> > > > > >  		if (ret & VM_FAULT_NEEDDSYNC)
> > > > > >  			ret = dax_finish_sync_fault(vmf, pe_size, pfn); diff --git
> > > > > > a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c index
> > > > > > d154f42e2dc6..2f322e2f8544
> > > > > > 100644
> > > > > > --- a/fs/xfs/xfs_iomap.c
> > > > > > +++ b/fs/xfs/xfs_iomap.c
> > > > > > @@ -761,7 +761,8 @@ xfs_direct_write_iomap_begin(
> > > > > >
> > > > > >  		/* may drop and re-acquire the ilock */
> > > > > >  		error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
> > > > > > -				&lockmode, flags & IOMAP_DIRECT);
> > > > > > +				&lockmode,
> > > > > > +				(flags & IOMAP_DIRECT) || IS_DAX(inode));
> > > > > >  		if (error)
> > > > > >  			goto out_unlock;
> > > > > >  		if (shared)
> > > > > > @@ -854,6 +855,36 @@ const struct iomap_ops
> > > > > > xfs_direct_write_iomap_ops = {
> > > > > >  	.iomap_begin		= xfs_direct_write_iomap_begin,
> > > > > >  };
> > > > > >
> > > > > > +static int
> > > > > > +xfs_dax_write_iomap_post_actor(
> > > > > > +	struct inode		*inode,
> > > > > > +	loff_t			pos,
> > > > > > +	loff_t			length,
> > > > > > +	ssize_t			written,
> > > > > > +	unsigned int		flags,
> > > > > > +	struct iomap		*iomap,
> > > > > > +	struct iomap		*srcmap)
> > > > > > +{
> > > > > > +	int			error = 0;
> > > > > > +	struct xfs_inode	*ip = XFS_I(inode);
> > > > > > +	bool			cow = xfs_is_cow_inode(ip);
> > > > > > +
> > > > > > +	if (written <= 0) {
> > > > > > +		if (cow)
> > > > > > +			xfs_reflink_cancel_cow_range(ip, pos, length, true);
> > > > > > +		return written;
> > > > > > +	}
> > > > > > +
> > > > > > +	if (cow)
> > > > > > +		error = xfs_reflink_end_cow(ip, pos, written);
> > > > > > +	return error ?: written;
> > > > > > +}
> > > >
> > > > This is pretty much the same as what xfs_dio_write_end_io does, right?
> > >
> > > It just handles the end part of CoW here.
> > > xfs_dio_write_end_io() also updates file size, which is only needed in write()
> > but not in page fault.  And the update file size work is done in
> > xfs_dax_file_write(), it's fine, no need to modify it.
> > >
> > > >
> > > > I had imagined that you'd change the function signatures to drop the
> > > > iocb so that you could reuse this code instead of creating a whole new
> > callback.
> > > >
> > > > Ah well.  Can I send you some prep patches to clean up some of the
> > > > weird iomap code as a preparation series for this?
> > >
> > > Sure.  Thanks.
> > >
> > >
> > > --
> > > Ruan.
> > >
> > > >
> > > > --D
> > > >
> > > > > > +
> > > > > > +const struct iomap_ops xfs_dax_write_iomap_ops = {
> > > > > > +	.iomap_begin		= xfs_direct_write_iomap_begin,
> > > > > > +	.iomap_post_actor	= xfs_dax_write_iomap_post_actor,
> > > > > > +};
> > > > > > +
> > > > > >  static int
> > > > > >  xfs_buffered_write_iomap_begin(
> > > > > >  	struct inode		*inode,
> > > > > > diff --git a/fs/xfs/xfs_iomap.h b/fs/xfs/xfs_iomap.h index
> > > > > > 7d3703556d0e..fbacf638ab21 100644
> > > > > > --- a/fs/xfs/xfs_iomap.h
> > > > > > +++ b/fs/xfs/xfs_iomap.h
> > > > > > @@ -42,8 +42,32 @@ xfs_aligned_fsb_count(
> > > > > >
> > > > > >  extern const struct iomap_ops xfs_buffered_write_iomap_ops;
> > > > > > extern const struct iomap_ops xfs_direct_write_iomap_ops;
> > > > > > +extern const struct iomap_ops xfs_dax_write_iomap_ops;
> > > > > >  extern const struct iomap_ops xfs_read_iomap_ops;  extern const
> > > > > > struct iomap_ops xfs_seek_iomap_ops;  extern const struct
> > > > > > iomap_ops xfs_xattr_iomap_ops;
> > > > > >
> > > > > > +static inline int
> > > > > > +xfs_iomap_zero_range(
> > > > > > +	struct xfs_inode	*ip,
> > > > > > +	loff_t			offset,
> > > > > > +	loff_t			len,
> > > > > > +	bool			*did_zero)
> > > > > > +{
> > > > > > +	return iomap_zero_range(VFS_I(ip), offset, len, did_zero,
> > > > > > +			IS_DAX(VFS_I(ip)) ? &xfs_dax_write_iomap_ops
> > > > > > +					  : &xfs_buffered_write_iomap_ops); }
> > > > > > +
> > > > > > +static inline int
> > > > > > +xfs_iomap_truncate_page(
> > > > > > +	struct xfs_inode	*ip,
> > > > > > +	loff_t			pos,
> > > > > > +	bool			*did_zero)
> > > > > > +{
> > > > > > +	return iomap_truncate_page(VFS_I(ip), pos, did_zero,
> > > > > > +			IS_DAX(VFS_I(ip)) ? &xfs_dax_write_iomap_ops
> > > > > > +					  : &xfs_buffered_write_iomap_ops); }
> > > > > > +
> > > > > >  #endif /* __XFS_IOMAP_H__*/
> > > > > > diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c index
> > > > > > dfe24b7f26e5..6d936c3e1a6e 100644
> > > > > > --- a/fs/xfs/xfs_iops.c
> > > > > > +++ b/fs/xfs/xfs_iops.c
> > > > > > @@ -911,8 +911,8 @@ xfs_setattr_size(
> > > > > >  	 */
> > > > > >  	if (newsize > oldsize) {
> > > > > >  		trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
> > > > > > -		error = iomap_zero_range(inode, oldsize, newsize - oldsize,
> > > > > > -				&did_zeroing, &xfs_buffered_write_iomap_ops);
> > > > > > +		error = xfs_iomap_zero_range(ip, oldsize, newsize - oldsize,
> > > > > > +				&did_zeroing);
> > > > > >  	} else {
> > > > > >  		/*
> > > > > >  		 * iomap won't detect a dirty page over an unwritten block
> > > > > > (or a @@
> > > > > > -924,8 +924,7 @@ xfs_setattr_size(
> > > > > >  						     newsize);
> > > > > >  		if (error)
> > > > > >  			return error;
> > > > > > -		error = iomap_truncate_page(inode, newsize, &did_zeroing,
> > > > > > -				&xfs_buffered_write_iomap_ops);
> > > > > > +		error = xfs_iomap_truncate_page(ip, newsize, &did_zeroing);
> > > > > >  	}
> > > > > >
> > > > > >  	if (error)
> > > > > > diff --git a/fs/xfs/xfs_reflink.c b/fs/xfs/xfs_reflink.c index
> > > > > > d25434f93235..9a780948dbd0 100644
> > > > > > --- a/fs/xfs/xfs_reflink.c
> > > > > > +++ b/fs/xfs/xfs_reflink.c
> > > > > > @@ -1266,8 +1266,7 @@ xfs_reflink_zero_posteof(
> > > > > >  		return 0;
> > > > > >
> > > > > >  	trace_xfs_zero_eof(ip, isize, pos - isize);
> > > > > > -	return iomap_zero_range(VFS_I(ip), isize, pos - isize, NULL,
> > > > > > -			&xfs_buffered_write_iomap_ops);
> > > > > > +	return xfs_iomap_zero_range(ip, isize, pos - isize, NULL);
> > > > > >  }
> > > > > >
> > > > > >  /*
> > > > > > diff --git a/include/linux/iomap.h b/include/linux/iomap.h index
> > > > > > 95562f863ad0..58f2e1c78018 100644
> > > > > > --- a/include/linux/iomap.h
> > > > > > +++ b/include/linux/iomap.h
> > > > > > @@ -135,6 +135,14 @@ struct iomap_ops {
> > > > > >  			unsigned flags, struct iomap *iomap,
> > > > > >  			struct iomap *srcmap);
> > > > > >
> > > > > > +	/*
> > > > > > +	 * Handle the error code from actor(). Do the finishing jobs for extra
> > > > > > +	 * operations, such as CoW, according to whether written is negative.
> > > > > > +	 */
> > > > > > +	int (*iomap_post_actor)(struct inode *inode, loff_t pos, loff_t length,
> > > > > > +			ssize_t written, unsigned flags, struct iomap *iomap,
> > > > > > +			struct iomap *srcmap);
> > > > > > +
> > > > > >  	/*
> > > > > >  	 * Commit and/or unreserve space previous allocated using
> > > > iomap_begin.
> > > > > >  	 * Written indicates the length of the successful write
> > > > > > operation which
> > > > > > --
> > > > > > 2.31.1
> > > > >

^ permalink raw reply	[flat|nested] 9+ messages in thread

* Re: [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path
  2021-06-28  5:09         ` Darrick J. Wong
  2021-06-29 11:25           ` ruansy.fnst
@ 2021-07-08 23:16           ` Dave Chinner
  1 sibling, 0 replies; 9+ messages in thread
From: Dave Chinner @ 2021-07-08 23:16 UTC (permalink / raw)
  To: Darrick J. Wong
  Cc: ruansy.fnst, dan.j.williams, hch, linux-fsdevel, linux-kernel,
	nvdimm, linux-xfs, rgoldwyn, viro, willy

On Sun, Jun 27, 2021 at 10:09:19PM -0700, Darrick J. Wong wrote:
> > > I had imagined that you'd create a struct dax_iomap_ops to wrap all the extra
> > > functionality that you need for dax operations:
> > > 
> > > struct dax_iomap_ops {
> > > 	struct iomap_ops	iomap_ops;
> > > 
> > > 	int			(*end_io)(inode, pos, length...);
> > > };
> > > 
> > > And alter the four functions that you need to take the special dax_iomap_ops.
> > > I guess the downside is that this makes iomap_truncate_page and
> > > iomap_zero_range more complicated, but maybe it's just time to split those into
> > > DAX-specific versions.  Then we'd be rid of the cross-links betwee
> > > fs/iomap/buffered-io.c and fs/dax.c.
> > 
> > This seems to be a better solution.  I'll try in this way.  Thanks for your guidance.
> 
> I started writing on Friday a patchset to apply this style cleanup both
> to the directio and dax paths.  The cleanups were pretty straightforward
> until I started reading the dax code paths again and realized that file
> writes still have the weird behavior of mapping extents into a file,
> zeroing them, then issuing the actual write to the extent.  IOWs, a
> double-write to avoid exposing stale contents if crash.
> 
> Apparently the reason for this was that dax (at least 6 years ago) had
> no concept paralleling the page lock, so it was necessary to do that to
> avoid page fault handlers racing to map pfns into the file mapping?
> That would seem to prevent us from doing the more standard behavior of
> allocate unwritten, write data, convert mapping... but is that still the
> case?  Or can we get rid of this bad quirk?

Yeah, so that was the deciding factor in getting rid of unwritten
extent allocation in DAX similar to the DIO path. However, we were
already considering getting rid of it for another reason: write
performance.

That is, doing two extent tree manipulation transactions per write
is way more expensive than the double memory write for small IOs.
IIRC, for small writes (4kB) the double memroy write version we now
have was 2-3x faster than the {unwritten allocation, write, convert}
algorithm we had originally.

I don't think we want to go back to the unwritten allocation
behaviour - it sucked when it was first done because all DAX write
IO is synchronous, and it will still suck now because DAX writes are
still synchronous. What we really want to do here is copy the data
into the new extent before we commit the allocation transaction....

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

^ permalink raw reply	[flat|nested] 9+ messages in thread

* [PATCH v6.2 6/7] dax: Introduce dax_iomap_ops for end of reflink
  2021-06-28  2:55       ` ruansy.fnst
  2021-06-28  5:09         ` Darrick J. Wong
@ 2021-07-09 12:36         ` Shiyang Ruan
  1 sibling, 0 replies; 9+ messages in thread
From: Shiyang Ruan @ 2021-07-09 12:36 UTC (permalink / raw)
  To: darrick.wong
  Cc: ruansy.fnst, dan.j.williams, david, djwong, hch, linux-fsdevel,
	linux-kernel, nvdimm, linux-xfs, rgoldwyn, viro, willy

After writing data, reflink requires end operations to remap those new
allocated extents.  The current ->iomap_end() ignores the error code
returned from ->actor(), so we need to introduce this dax_iomap_ops and
change the dax_iomap_* interfaces to do this job.

- the dax_iomap_ops contains the original struct iomap_ops and fsdax
    specific ->actor_end(), which is for the end operations of reflink
- also introduce dax specific zero_range, truncate_page
- create new dax_iomap_ops for ext2 and ext4

Then enable fsdax and reflink together in xfs.

Signed-off-by: Shiyang Ruan <ruansy.fnst@fujitsu.com>
---
 fs/dax.c               | 105 +++++++++++++++++++++++++++++++++++------
 fs/ext2/ext2.h         |   3 ++
 fs/ext2/file.c         |   6 +--
 fs/ext2/inode.c        |  11 ++++-
 fs/ext4/ext4.h         |   3 ++
 fs/ext4/file.c         |   6 +--
 fs/ext4/inode.c        |  13 ++++-
 fs/iomap/buffered-io.c |   6 +--
 fs/xfs/xfs_bmap_util.c |   3 +-
 fs/xfs/xfs_file.c      |   8 ++--
 fs/xfs/xfs_iomap.c     |  36 +++++++++++++-
 fs/xfs/xfs_iomap.h     |  33 +++++++++++++
 fs/xfs/xfs_iops.c      |   7 ++-
 fs/xfs/xfs_reflink.c   |   3 +-
 include/linux/dax.h    |  28 +++++++++--
 include/linux/iomap.h  |   2 +
 16 files changed, 228 insertions(+), 45 deletions(-)

diff --git a/fs/dax.c b/fs/dax.c
index 93f16210847b..9285ea796668 100644
--- a/fs/dax.c
+++ b/fs/dax.c
@@ -1240,7 +1240,7 @@ s64 dax_iomap_zero(loff_t pos, u64 length, struct iomap *iomap,
 }
 
 static loff_t
-dax_iomap_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
+__dax_iomap_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
 		struct iomap *iomap, struct iomap *srcmap)
 {
 	struct block_device *bdev = iomap->bdev;
@@ -1344,11 +1344,25 @@ dax_iomap_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
 	return done ? done : ret;
 }
 
+static loff_t
+dax_iomap_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
+		struct iomap *iomap, struct iomap *srcmap)
+{
+	struct dax_iomap_data *idata = data;
+	loff_t ret = __dax_iomap_actor(inode, pos, length, idata->data,
+					iomap, srcmap);
+
+	if (idata->ops->actor_end)
+		ret = idata->ops->actor_end(inode, pos, length, ret);
+
+	return ret;
+}
+
 /**
  * dax_iomap_rw - Perform I/O to a DAX file
  * @iocb:	The control block for this I/O
  * @iter:	The addresses to do I/O from or to
- * @ops:	iomap ops passed from the file system
+ * @ops:	dax iomap ops passed from the file system
  *
  * This function performs read and write operations to directly mapped
  * persistent memory.  The callers needs to take care of read/write exclusion
@@ -1356,12 +1370,13 @@ dax_iomap_actor(struct inode *inode, loff_t pos, loff_t length, void *data,
  */
 ssize_t
 dax_iomap_rw(struct kiocb *iocb, struct iov_iter *iter,
-		const struct iomap_ops *ops)
+		const struct dax_iomap_ops *ops)
 {
 	struct address_space *mapping = iocb->ki_filp->f_mapping;
 	struct inode *inode = mapping->host;
 	loff_t pos = iocb->ki_pos, ret = 0, done = 0;
 	unsigned flags = 0;
+	struct dax_iomap_data data = { iter, ops };
 
 	if (iov_iter_rw(iter) == WRITE) {
 		lockdep_assert_held_write(&inode->i_rwsem);
@@ -1374,8 +1389,8 @@ dax_iomap_rw(struct kiocb *iocb, struct iov_iter *iter,
 		flags |= IOMAP_NOWAIT;
 
 	while (iov_iter_count(iter)) {
-		ret = iomap_apply(inode, pos, iov_iter_count(iter), flags, ops,
-				iter, dax_iomap_actor);
+		ret = iomap_apply(inode, pos, iov_iter_count(iter), flags,
+				  &ops->iomap_ops, &data, dax_iomap_actor);
 		if (ret <= 0)
 			break;
 		pos += ret;
@@ -1387,6 +1402,55 @@ dax_iomap_rw(struct kiocb *iocb, struct iov_iter *iter,
 }
 EXPORT_SYMBOL_GPL(dax_iomap_rw);
 
+static loff_t
+dax_iomap_zero_range_actor(struct inode *inode, loff_t pos, loff_t length,
+		void *data, struct iomap *iomap, struct iomap *srcmap)
+{
+	struct dax_iomap_data *idata = data;
+	loff_t ret = iomap_zero_range_actor(inode, pos, length, idata->data,
+					    iomap, srcmap);
+
+	if (idata->ops->actor_end)
+		ret = idata->ops->actor_end(inode, pos, length, ret);
+
+	return ret;
+}
+
+int
+dax_iomap_zero_range(struct inode *inode, loff_t pos, loff_t len,
+		bool *did_zero, const struct dax_iomap_ops *ops)
+{
+	struct dax_iomap_data data = { did_zero, ops };
+	loff_t ret;
+
+	while (len > 0) {
+		ret = iomap_apply(inode, pos, len, IOMAP_ZERO, &ops->iomap_ops,
+				  &data, dax_iomap_zero_range_actor);
+		if (ret <= 0)
+			return ret;
+
+		pos += ret;
+		len -= ret;
+	}
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(dax_iomap_zero_range);
+
+int
+dax_iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero,
+		const struct dax_iomap_ops *ops)
+{
+	unsigned int blocksize = i_blocksize(inode);
+	unsigned int off = pos & (blocksize - 1);
+
+	/* Block boundary? Nothing to do */
+	if (!off)
+		return 0;
+	return dax_iomap_zero_range(inode, pos, blocksize - off, did_zero, ops);
+}
+EXPORT_SYMBOL_GPL(dax_iomap_truncate_page);
+
 static vm_fault_t dax_fault_return(int error)
 {
 	if (error == 0)
@@ -1527,7 +1591,7 @@ static vm_fault_t dax_fault_actor(struct vm_fault *vmf, pfn_t *pfnp,
 }
 
 static vm_fault_t dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
-			       int *iomap_errp, const struct iomap_ops *ops)
+		int *iomap_errp, const struct dax_iomap_ops *dops)
 {
 	struct vm_area_struct *vma = vmf->vma;
 	struct address_space *mapping = vma->vm_file->f_mapping;
@@ -1536,8 +1600,9 @@ static vm_fault_t dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
 	loff_t pos = (loff_t)vmf->pgoff << PAGE_SHIFT;
 	struct iomap iomap = { .type = IOMAP_HOLE };
 	struct iomap srcmap = { .type = IOMAP_HOLE };
+	const struct iomap_ops *ops = &dops->iomap_ops;
 	unsigned flags = IOMAP_FAULT;
-	int error;
+	int error, copied = PAGE_SIZE;
 	bool write = vmf->flags & FAULT_FLAG_WRITE;
 	vm_fault_t ret = 0, major = 0;
 	void *entry;
@@ -1598,7 +1663,7 @@ static vm_fault_t dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
 	ret = dax_fault_actor(vmf, pfnp, &xas, &entry, false, flags,
 			      &iomap, &srcmap);
 	if (ret == VM_FAULT_SIGBUS)
-		goto finish_iomap;
+		goto finish_iomap_actor_end;
 
 	/* read/write MAPPED, CoW UNWRITTEN */
 	if (iomap.flags & IOMAP_F_NEW) {
@@ -1607,10 +1672,15 @@ static vm_fault_t dax_iomap_pte_fault(struct vm_fault *vmf, pfn_t *pfnp,
 		major = VM_FAULT_MAJOR;
 	}
 
+finish_iomap_actor_end:
+	if (dops->actor_end) {
+		if (ret & VM_FAULT_ERROR)
+			copied = 0;
+		dops->actor_end(inode, pos, PMD_SIZE, copied);
+	}
+
 finish_iomap:
 	if (ops->iomap_end) {
-		int copied = PAGE_SIZE;
-
 		if (ret & VM_FAULT_ERROR)
 			copied = 0;
 		/*
@@ -1663,7 +1733,7 @@ static bool dax_fault_check_fallback(struct vm_fault *vmf, struct xa_state *xas,
 }
 
 static vm_fault_t dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp,
-			       const struct iomap_ops *ops)
+		const struct dax_iomap_ops *dops)
 {
 	struct vm_area_struct *vma = vmf->vma;
 	struct address_space *mapping = vma->vm_file->f_mapping;
@@ -1674,10 +1744,11 @@ static vm_fault_t dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp,
 	vm_fault_t ret = VM_FAULT_FALLBACK;
 	struct iomap iomap = { .type = IOMAP_HOLE };
 	struct iomap srcmap = { .type = IOMAP_HOLE };
+	const struct iomap_ops *ops = &dops->iomap_ops;
 	pgoff_t max_pgoff;
 	void *entry;
 	loff_t pos;
-	int error;
+	int error, copied = PMD_SIZE;
 
 	/*
 	 * Check whether offset isn't beyond end of file now. Caller is
@@ -1736,10 +1807,14 @@ static vm_fault_t dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp,
 	ret = dax_fault_actor(vmf, pfnp, &xas, &entry, true, flags,
 			      &iomap, &srcmap);
 
+	if (dops->actor_end) {
+		if (ret == VM_FAULT_FALLBACK)
+			copied = 0;
+		dops->actor_end(inode, pos, PMD_SIZE, copied);
+	}
+
 finish_iomap:
 	if (ops->iomap_end) {
-		int copied = PMD_SIZE;
-
 		if (ret == VM_FAULT_FALLBACK)
 			copied = 0;
 		/*
@@ -1783,7 +1858,7 @@ static vm_fault_t dax_iomap_pmd_fault(struct vm_fault *vmf, pfn_t *pfnp,
  * successfully.
  */
 vm_fault_t dax_iomap_fault(struct vm_fault *vmf, enum page_entry_size pe_size,
-		    pfn_t *pfnp, int *iomap_errp, const struct iomap_ops *ops)
+		pfn_t *pfnp, int *iomap_errp, const struct dax_iomap_ops *ops)
 {
 	switch (pe_size) {
 	case PE_SIZE_PTE:
diff --git a/fs/ext2/ext2.h b/fs/ext2/ext2.h
index b0a694820cb7..765269804f83 100644
--- a/fs/ext2/ext2.h
+++ b/fs/ext2/ext2.h
@@ -806,6 +806,9 @@ extern void ext2_set_file_ops(struct inode *inode);
 extern const struct address_space_operations ext2_aops;
 extern const struct address_space_operations ext2_nobh_aops;
 extern const struct iomap_ops ext2_iomap_ops;
+#ifdef CONFIG_FS_DAX
+extern const struct dax_iomap_ops ext2_dax_iomap_ops;
+#endif
 
 /* namei.c */
 extern const struct inode_operations ext2_dir_inode_operations;
diff --git a/fs/ext2/file.c b/fs/ext2/file.c
index f98466acc672..d5dd82111128 100644
--- a/fs/ext2/file.c
+++ b/fs/ext2/file.c
@@ -39,7 +39,7 @@ static ssize_t ext2_dax_read_iter(struct kiocb *iocb, struct iov_iter *to)
 		return 0; /* skip atime */
 
 	inode_lock_shared(inode);
-	ret = dax_iomap_rw(iocb, to, &ext2_iomap_ops);
+	ret = dax_iomap_rw(iocb, to, &ext2_dax_iomap_ops);
 	inode_unlock_shared(inode);
 
 	file_accessed(iocb->ki_filp);
@@ -63,7 +63,7 @@ static ssize_t ext2_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
 	if (ret)
 		goto out_unlock;
 
-	ret = dax_iomap_rw(iocb, from, &ext2_iomap_ops);
+	ret = dax_iomap_rw(iocb, from, &ext2_dax_iomap_ops);
 	if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
 		i_size_write(inode, iocb->ki_pos);
 		mark_inode_dirty(inode);
@@ -102,7 +102,7 @@ static vm_fault_t ext2_dax_fault(struct vm_fault *vmf)
 	}
 	down_read(&ei->dax_sem);
 
-	ret = dax_iomap_fault(vmf, PE_SIZE_PTE, NULL, NULL, &ext2_iomap_ops);
+	ret = dax_iomap_fault(vmf, PE_SIZE_PTE, NULL, NULL, &ext2_dax_iomap_ops);
 
 	up_read(&ei->dax_sem);
 	if (write)
diff --git a/fs/ext2/inode.c b/fs/ext2/inode.c
index 68178b2234bd..a94744bbf82f 100644
--- a/fs/ext2/inode.c
+++ b/fs/ext2/inode.c
@@ -852,6 +852,13 @@ const struct iomap_ops ext2_iomap_ops = {
 	.iomap_begin		= ext2_iomap_begin,
 	.iomap_end		= ext2_iomap_end,
 };
+
+const struct dax_iomap_ops ext2_dax_iomap_ops = {
+	.iomap_ops	= {
+		.iomap_begin	= ext2_iomap_begin,
+		.iomap_end	= ext2_iomap_end,
+	},
+};
 #else
 /* Define empty ops for !CONFIG_FS_DAX case to avoid ugly ifdefs */
 const struct iomap_ops ext2_iomap_ops;
@@ -1294,9 +1301,9 @@ static int ext2_setsize(struct inode *inode, loff_t newsize)
 	inode_dio_wait(inode);
 
 	if (IS_DAX(inode)) {
-		error = iomap_zero_range(inode, newsize,
+		error = dax_iomap_zero_range(inode, newsize,
 					 PAGE_ALIGN(newsize) - newsize, NULL,
-					 &ext2_iomap_ops);
+					 &ext2_dax_iomap_ops);
 	} else if (test_opt(inode->i_sb, NOBH))
 		error = nobh_truncate_page(inode->i_mapping,
 				newsize, ext2_get_block);
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 37002663d521..b4e6df93dd82 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -3773,6 +3773,9 @@ static inline void ext4_clear_io_unwritten_flag(ext4_io_end_t *io_end)
 }
 
 extern const struct iomap_ops ext4_iomap_ops;
+#ifdef CONFIG_FS_DAX
+extern const struct dax_iomap_ops ext4_dax_iomap_ops;
+#endif
 extern const struct iomap_ops ext4_iomap_overwrite_ops;
 extern const struct iomap_ops ext4_iomap_report_ops;
 
diff --git a/fs/ext4/file.c b/fs/ext4/file.c
index 816dedcbd541..a7a3497429ca 100644
--- a/fs/ext4/file.c
+++ b/fs/ext4/file.c
@@ -102,7 +102,7 @@ static ssize_t ext4_dax_read_iter(struct kiocb *iocb, struct iov_iter *to)
 		/* Fallback to buffered IO in case we cannot support DAX */
 		return generic_file_read_iter(iocb, to);
 	}
-	ret = dax_iomap_rw(iocb, to, &ext4_iomap_ops);
+	ret = dax_iomap_rw(iocb, to, &ext4_dax_iomap_ops);
 	inode_unlock_shared(inode);
 
 	file_accessed(iocb->ki_filp);
@@ -650,7 +650,7 @@ ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
 		ext4_journal_stop(handle);
 	}
 
-	ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
+	ret = dax_iomap_rw(iocb, from, &ext4_dax_iomap_ops);
 
 	if (extend)
 		ret = ext4_handle_inode_extension(inode, offset, ret, count);
@@ -721,7 +721,7 @@ static vm_fault_t ext4_dax_huge_fault(struct vm_fault *vmf,
 	} else {
 		down_read(&EXT4_I(inode)->i_mmap_sem);
 	}
-	result = dax_iomap_fault(vmf, pe_size, &pfn, &error, &ext4_iomap_ops);
+	result = dax_iomap_fault(vmf, pe_size, &pfn, &error, &ext4_dax_iomap_ops);
 	if (write) {
 		ext4_journal_stop(handle);
 
diff --git a/fs/ext4/inode.c b/fs/ext4/inode.c
index fe6045a46599..2310f5cc6cd5 100644
--- a/fs/ext4/inode.c
+++ b/fs/ext4/inode.c
@@ -3523,6 +3523,15 @@ const struct iomap_ops ext4_iomap_ops = {
 	.iomap_end		= ext4_iomap_end,
 };
 
+#ifdef CONFIG_FS_DAX
+const struct dax_iomap_ops ext4_dax_iomap_ops = {
+	.iomap_ops		= {
+		.iomap_begin = ext4_iomap_begin,
+		.iomap_end   = ext4_iomap_end,
+	},
+};
+#endif
+
 const struct iomap_ops ext4_iomap_overwrite_ops = {
 	.iomap_begin		= ext4_iomap_overwrite_begin,
 	.iomap_end		= ext4_iomap_end,
@@ -3840,8 +3849,8 @@ static int ext4_block_zero_page_range(handle_t *handle,
 		length = max;
 
 	if (IS_DAX(inode)) {
-		return iomap_zero_range(inode, from, length, NULL,
-					&ext4_iomap_ops);
+		return dax_iomap_zero_range(inode, from, length, NULL,
+					    &ext4_dax_iomap_ops);
 	}
 	return __ext4_block_zero_page_range(handle, mapping, from, length);
 }
diff --git a/fs/iomap/buffered-io.c b/fs/iomap/buffered-io.c
index fdaac4ba9b9d..32c6b2ab6251 100644
--- a/fs/iomap/buffered-io.c
+++ b/fs/iomap/buffered-io.c
@@ -918,9 +918,9 @@ static s64 iomap_zero(struct inode *inode, loff_t pos, u64 length,
 	return iomap_write_end(inode, pos, bytes, bytes, page, iomap, srcmap);
 }
 
-static loff_t iomap_zero_range_actor(struct inode *inode, loff_t pos,
-		loff_t length, void *data, struct iomap *iomap,
-		struct iomap *srcmap)
+loff_t
+iomap_zero_range_actor(struct inode *inode, loff_t pos, loff_t length,
+		void *data, struct iomap *iomap, struct iomap *srcmap)
 {
 	bool *did_zero = data;
 	loff_t written = 0;
diff --git a/fs/xfs/xfs_bmap_util.c b/fs/xfs/xfs_bmap_util.c
index 0936f3a96fe6..4b0744b5a75f 100644
--- a/fs/xfs/xfs_bmap_util.c
+++ b/fs/xfs/xfs_bmap_util.c
@@ -1009,8 +1009,7 @@ xfs_free_file_space(
 		return 0;
 	if (offset + len > XFS_ISIZE(ip))
 		len = XFS_ISIZE(ip) - offset;
-	error = iomap_zero_range(VFS_I(ip), offset, len, NULL,
-			&xfs_buffered_write_iomap_ops);
+	error = xfs_iomap_zero_range(ip, offset, len, NULL);
 	if (error)
 		return error;
 
diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
index 396ef36dcd0a..9bca68872242 100644
--- a/fs/xfs/xfs_file.c
+++ b/fs/xfs/xfs_file.c
@@ -281,7 +281,7 @@ xfs_file_dax_read(
 	ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
 	if (ret)
 		return ret;
-	ret = dax_iomap_rw(iocb, to, &xfs_read_iomap_ops);
+	ret = dax_iomap_rw(iocb, to, &xfs_dax_read_iomap_ops);
 	xfs_iunlock(ip, XFS_IOLOCK_SHARED);
 
 	file_accessed(iocb->ki_filp);
@@ -684,7 +684,7 @@ xfs_file_dax_write(
 	pos = iocb->ki_pos;
 
 	trace_xfs_file_dax_write(iocb, from);
-	ret = dax_iomap_rw(iocb, from, &xfs_direct_write_iomap_ops);
+	ret = dax_iomap_rw(iocb, from, &xfs_dax_write_iomap_ops);
 	if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
 		i_size_write(inode, iocb->ki_pos);
 		error = xfs_setfilesize(ip, pos, ret);
@@ -1309,8 +1309,8 @@ __xfs_filemap_fault(
 
 		ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL,
 				(write_fault && !vmf->cow_page) ?
-				 &xfs_direct_write_iomap_ops :
-				 &xfs_read_iomap_ops);
+				 &xfs_dax_write_iomap_ops :
+				 &xfs_dax_read_iomap_ops);
 		if (ret & VM_FAULT_NEEDDSYNC)
 			ret = dax_finish_sync_fault(vmf, pe_size, pfn);
 	} else {
diff --git a/fs/xfs/xfs_iomap.c b/fs/xfs/xfs_iomap.c
index d154f42e2dc6..48004cf28a88 100644
--- a/fs/xfs/xfs_iomap.c
+++ b/fs/xfs/xfs_iomap.c
@@ -761,7 +761,8 @@ xfs_direct_write_iomap_begin(
 
 		/* may drop and re-acquire the ilock */
 		error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
-				&lockmode, flags & IOMAP_DIRECT);
+				&lockmode,
+				(flags & IOMAP_DIRECT) || IS_DAX(inode));
 		if (error)
 			goto out_unlock;
 		if (shared)
@@ -854,6 +855,33 @@ const struct iomap_ops xfs_direct_write_iomap_ops = {
 	.iomap_begin		= xfs_direct_write_iomap_begin,
 };
 
+static int
+xfs_dax_write_iomap_actor_end(
+	struct inode		*inode,
+	loff_t			pos,
+	loff_t			length,
+	ssize_t			written)
+{
+	int			error = 0;
+	struct xfs_inode	*ip = XFS_I(inode);
+	bool			cow = xfs_is_cow_inode(ip);
+
+	if (cow) {
+		if (written <= 0)
+			xfs_reflink_cancel_cow_range(ip, pos, length, true);
+		else
+			error = xfs_reflink_end_cow(ip, pos, written);
+	}
+	return error ?: written;
+}
+
+const struct dax_iomap_ops xfs_dax_write_iomap_ops = {
+	.iomap_ops 		= {
+		.iomap_begin = xfs_direct_write_iomap_begin,
+	},
+	.actor_end		= xfs_dax_write_iomap_actor_end,
+};
+
 static int
 xfs_buffered_write_iomap_begin(
 	struct inode		*inode,
@@ -1184,6 +1212,12 @@ const struct iomap_ops xfs_read_iomap_ops = {
 	.iomap_begin		= xfs_read_iomap_begin,
 };
 
+const struct dax_iomap_ops xfs_dax_read_iomap_ops = {
+	.iomap_ops		= {
+		.iomap_begin = xfs_read_iomap_begin,
+	},
+};
+
 static int
 xfs_seek_iomap_begin(
 	struct inode		*inode,
diff --git a/fs/xfs/xfs_iomap.h b/fs/xfs/xfs_iomap.h
index 7d3703556d0e..5eacb5d8ca88 100644
--- a/fs/xfs/xfs_iomap.h
+++ b/fs/xfs/xfs_iomap.h
@@ -45,5 +45,38 @@ extern const struct iomap_ops xfs_direct_write_iomap_ops;
 extern const struct iomap_ops xfs_read_iomap_ops;
 extern const struct iomap_ops xfs_seek_iomap_ops;
 extern const struct iomap_ops xfs_xattr_iomap_ops;
+extern const struct dax_iomap_ops xfs_dax_write_iomap_ops;
+extern const struct dax_iomap_ops xfs_dax_read_iomap_ops;
+
+static inline int
+xfs_iomap_zero_range(
+	struct xfs_inode	*ip,
+	loff_t			pos,
+	loff_t			len,
+	bool			*did_zero)
+{
+	struct inode		*inode = VFS_I(ip);
+
+	return IS_DAX(inode)
+			? dax_iomap_zero_range(inode, pos, len, did_zero,
+					       &xfs_dax_write_iomap_ops)
+			: iomap_zero_range(inode, pos, len, did_zero,
+					       &xfs_buffered_write_iomap_ops);
+}
+
+static inline int
+xfs_iomap_truncate_page(
+	struct xfs_inode	*ip,
+	loff_t			pos,
+	bool			*did_zero)
+{
+	struct inode		*inode = VFS_I(ip);
+
+	return IS_DAX(inode)
+			? dax_iomap_truncate_page(inode, pos, did_zero,
+					       &xfs_dax_write_iomap_ops)
+			: iomap_truncate_page(inode, pos, did_zero,
+					       &xfs_buffered_write_iomap_ops);
+}
 
 #endif /* __XFS_IOMAP_H__*/
diff --git a/fs/xfs/xfs_iops.c b/fs/xfs/xfs_iops.c
index dfe24b7f26e5..6d936c3e1a6e 100644
--- a/fs/xfs/xfs_iops.c
+++ b/fs/xfs/xfs_iops.c
@@ -911,8 +911,8 @@ xfs_setattr_size(
 	 */
 	if (newsize > oldsize) {
 		trace_xfs_zero_eof(ip, oldsize, newsize - oldsize);
-		error = iomap_zero_range(inode, oldsize, newsize - oldsize,
-				&did_zeroing, &xfs_buffered_write_iomap_ops);
+		error = xfs_iomap_zero_range(ip, oldsize, newsize - oldsize,
+				&did_zeroing);
 	} else {
 		/*
 		 * iomap won't detect a dirty page over an unwritten block (or a
@@ -924,8 +924,7 @@ xfs_setattr_size(
 						     newsize);
 		if (error)
 			return error;
-		error = iomap_truncate_page(inode, newsize, &did_zeroing,
-				&xfs_buffered_write_iomap_ops);
+		error = xfs_iomap_truncate_page(ip, newsize, &did_zeroing);
 	}
 
 	if (error)
diff --git a/fs/xfs/xfs_reflink.c b/fs/xfs/xfs_reflink.c
index d25434f93235..9a780948dbd0 100644
--- a/fs/xfs/xfs_reflink.c
+++ b/fs/xfs/xfs_reflink.c
@@ -1266,8 +1266,7 @@ xfs_reflink_zero_posteof(
 		return 0;
 
 	trace_xfs_zero_eof(ip, isize, pos - isize);
-	return iomap_zero_range(VFS_I(ip), isize, pos - isize, NULL,
-			&xfs_buffered_write_iomap_ops);
+	return xfs_iomap_zero_range(ip, isize, pos - isize, NULL);
 }
 
 /*
diff --git a/include/linux/dax.h b/include/linux/dax.h
index 106d1f033a78..64393f6e96cf 100644
--- a/include/linux/dax.h
+++ b/include/linux/dax.h
@@ -3,6 +3,7 @@
 #define _LINUX_DAX_H
 
 #include <linux/fs.h>
+#include <linux/iomap.h>
 #include <linux/mm.h>
 #include <linux/radix-tree.h>
 
@@ -11,8 +12,6 @@
 
 typedef unsigned long dax_entry_t;
 
-struct iomap_ops;
-struct iomap;
 struct dax_device;
 struct dax_operations {
 	/*
@@ -38,6 +37,23 @@ struct dax_operations {
 	int (*zero_page_range)(struct dax_device *, pgoff_t, size_t);
 };
 
+struct dax_iomap_ops {
+	/* the original iomap ops */
+	struct iomap_ops iomap_ops;
+	/*
+	 * actor_end: accept error code returned from ->actor(), deal with it
+	 * before ->iomap_end()
+	 */
+	int (*actor_end)(struct inode *, loff_t, loff_t, ssize_t);
+};
+
+/* dax iomap specific data, in order to call ->actor_end() in ->actor() */
+struct dax_iomap_data {
+	/* the original data pointer */
+	void *data;
+	const struct dax_iomap_ops *ops;
+};
+
 extern struct attribute_group dax_attribute_group;
 
 #if IS_ENABLED(CONFIG_DAX)
@@ -229,14 +245,18 @@ int dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
 void dax_flush(struct dax_device *dax_dev, void *addr, size_t size);
 
 ssize_t dax_iomap_rw(struct kiocb *iocb, struct iov_iter *iter,
-		const struct iomap_ops *ops);
+		const struct dax_iomap_ops *ops);
 vm_fault_t dax_iomap_fault(struct vm_fault *vmf, enum page_entry_size pe_size,
-		    pfn_t *pfnp, int *errp, const struct iomap_ops *ops);
+		pfn_t *pfnp, int *errp, const struct dax_iomap_ops *ops);
 vm_fault_t dax_finish_sync_fault(struct vm_fault *vmf,
 		enum page_entry_size pe_size, pfn_t pfn);
 int dax_delete_mapping_entry(struct address_space *mapping, pgoff_t index);
 int dax_invalidate_mapping_entry_sync(struct address_space *mapping,
 				      pgoff_t index);
+int dax_iomap_zero_range(struct inode *inode, loff_t pos, loff_t len,
+		bool *did_zero, const struct dax_iomap_ops *ops);
+int dax_iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero,
+		const struct dax_iomap_ops *ops);
 s64 dax_iomap_zero(loff_t pos, u64 length, struct iomap *iomap,
 		struct iomap *srcmap);
 int dax_dedupe_file_range_compare(struct inode *src, loff_t srcoff,
diff --git a/include/linux/iomap.h b/include/linux/iomap.h
index 95562f863ad0..05437fbf5f68 100644
--- a/include/linux/iomap.h
+++ b/include/linux/iomap.h
@@ -178,6 +178,8 @@ int iomap_migrate_page(struct address_space *mapping, struct page *newpage,
 #endif
 int iomap_file_unshare(struct inode *inode, loff_t pos, loff_t len,
 		const struct iomap_ops *ops);
+loff_t iomap_zero_range_actor(struct inode *inode, loff_t pos, loff_t length,
+		void *data, struct iomap *iomap, struct iomap *srcmap);
 int iomap_zero_range(struct inode *inode, loff_t pos, loff_t len,
 		bool *did_zero, const struct iomap_ops *ops);
 int iomap_truncate_page(struct inode *inode, loff_t pos, bool *did_zero,
-- 
2.32.0




^ permalink raw reply related	[flat|nested] 9+ messages in thread

end of thread, other threads:[~2021-07-09 12:37 UTC | newest]

Thread overview: 9+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
     [not found] <OSBPR01MB2920A2BCD568364C1363AFA6F4369@OSBPR01MB2920.jpnprd01.prod.outlook.com>
2021-06-15  7:21 ` [PATCH v6.1 6/7] fs/xfs: Handle CoW for fsdax write() path Shiyang Ruan
2021-06-24  8:49   ` ruansy.fnst
2021-06-25 22:18     ` Darrick J. Wong
2021-06-28  2:55       ` ruansy.fnst
2021-06-28  5:09         ` Darrick J. Wong
2021-06-29 11:25           ` ruansy.fnst
2021-06-29 21:01             ` Darrick J. Wong
2021-07-08 23:16           ` Dave Chinner
2021-07-09 12:36         ` [PATCH v6.2 6/7] dax: Introduce dax_iomap_ops for end of reflink Shiyang Ruan

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).