Skip to contentSkip to conversation

ext2: preserve IOMAP_F_DIRTY for extending direct writes

PATCH In review Hungyu Lin 3 participants 4 messages Oct 29 – Nov 4

Subsystems (2)
lore
Share & tools
lore

Hungyu Lin

author
#0 lore raw

PATCH 1/1 ext2: preserve IOMAP_F_DIRTY for extending direct writes

ext2_iomap_begin() sets IOMAP_F_DIRTY for writes that may extend i_size, but clears the flag after ext2_get_blocks(). When the write uses an existing mapped block, iomap can then use FUA and skip generic_write_sync(), even though ext2_dio_write_end_io() updates the inode size. An O_DSYNC write can therefore return without persisting the new size.

For example, on a FUA-capable device, a 512-byte O_DIRECT|O_DSYNC write at offset 512 into a fsynced one-byte file returns successfully and extends the in-memory size to 1024, while an immediate read of the on-disk inode still shows size 1. An explicit fsync persists the new size.

Initialize iomap->flags before setting IOMAP_F_DIRTY so that the flag survives and the required metadata synchronization takes place.

Fixes fb5de4358e1a ("ext2: Move direct-io to use iomap")
---
 fs/ext2/inode.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/ext2/inode.c +1 -1
index 12ac4cfe1500..ba23cadb157b 100644
@@ -805,6 +805,7 @@ static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 	    (first_block << blkbits) < i_size_read(inode))
 		create = 0;
 
+	iomap->flags = 0;
 	/*
 	 * Writes that span EOF might trigger an IO size update on completion,
 	 * so consider them to be dirty for the purposes of O_DSYNC even if
@@ -818,7 +819,6 @@ static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 	if (ret < 0)
 		return ret;
 
-	iomap->flags = 0;
 	iomap->offset = (u64)first_block << blkbits;
 	iomap->bdev = inode->i_sb->s_bdev;
 
Signature · 1 line
2.43.0
#1 lore raw
Sashiko has reviewed this patch and found no issues.
Quoted 13 lines

ext2_iomap_begin() sets IOMAP_F_DIRTY for writes that may extend i_size, but clears the flag after ext2_get_blocks(). When the write uses an existing mapped block, iomap can then use FUA and skip generic_write_sync(), even though ext2_dio_write_end_io() updates the inode size. An O_DSYNC write can therefore return without persisting the new size.

For example, on a FUA-capable device, a 512-byte O_DIRECT|O_DSYNC write at offset 512 into a fsynced one-byte file returns successfully and extends the in-memory size to 1024, while an immediate read of the on-disk inode still shows size 1. An explicit fsync persists the new size.

Quoted 6 lines

Initialize iomap->flags before setting IOMAP_F_DIRTY so that the flag survives and the required metadata synchronization takes place.

Fixes fb5de4358e1a ("ext2: Move direct-io to use iomap")

Sashiko has reviewed this patch and found no issues.

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
#2 lore raw
In reply to Hungyu Lin
Quoted 13 lines · Hungyu Lin

ext2_iomap_begin() sets IOMAP_F_DIRTY for writes that may extend i_size, but clears the flag after ext2_get_blocks(). When the write uses an existing mapped block, iomap can then use FUA and skip generic_write_sync(), even though ext2_dio_write_end_io() updates the inode size. An O_DSYNC write can therefore return without persisting the new size.

For example, on a FUA-capable device, a 512-byte O_DIRECT|O_DSYNC write at offset 512 into a fsynced one-byte file returns successfully and extends the in-memory size to 1024, while an immediate read of the on-disk inode still shows size 1. An explicit fsync persists the new size.

Quoted 6 lines

Initialize iomap->flags before setting IOMAP_F_DIRTY so that the flag survives and the required metadata synchronization takes place.

Fixes fb5de4358e1a ("ext2: Move direct-io to use iomap")

Thanks for the fix. I've just move the setting of the IOMAP_F_DIRTY bit instead of moving the initialization. Functionally this is equivalent and the iomap modification is better localized that way (the resulting patch is attached).

Honza

Quoted 27 lines
---
 fs/ext2/inode.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
fs/ext2/inode.c +1 -1
index 12ac4cfe1500..ba23cadb157b 100644
@@ -805,6 +805,7 @@ static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 	    (first_block << blkbits) < i_size_read(inode))
 		create = 0;
 
+	iomap->flags = 0;
 	/*
 	 * Writes that span EOF might trigger an IO size update on completion,
 	 * so consider them to be dirty for the purposes of O_DSYNC even if
@@ -818,7 +819,6 @@ static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 	if (ret < 0)
 		return ret;
 
-	iomap->flags = 0;
 	iomap->offset = (u64)first_block << blkbits;
 	iomap->bdev = inode->i_sb->s_bdev;
 
Signature · 1 line
2.43.0
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR

From 6fda7e6fba044590ce5d7bae6eebd437df1f5b92 Mon Sep 17 00:00:00 2001
From: Hungyu Lin <[email protected]>
Date: Thu, 29 Oct 2026 01:36:16 +0000
Subject: [PATCH] ext2: preserve IOMAP_F_DIRTY for extending direct writes

ext2_iomap_begin() sets IOMAP_F_DIRTY for writes that may extend i_size, but clears the flag after ext2_get_blocks(). When the write uses an existing mapped block, iomap can then use FUA and skip generic_write_sync(), even though ext2_dio_write_end_io() updates the inode size. An O_DSYNC write can therefore return without persisting the new size.

For example, on a FUA-capable device, a 512-byte O_DIRECT|O_DSYNC write at offset 512 into a fsynced one-byte file returns successfully and extends the in-memory size to 1024, while an immediate read of the on-disk inode still shows size 1. An explicit fsync persists the new size.

Move setting of IOMAP_F_DIRTY after the initialization.

Fixes fb5de4358e1a ("ext2: Move direct-io to use iomap")
---
 fs/ext2/inode.c | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)
fs/ext2/inode.c +6 -7
index 12ac4cfe1500..51250717c70b 100644
@@ -805,6 +805,12 @@ static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 	    (first_block << blkbits) < i_size_read(inode))
 		create = 0;
 
+	ret = ext2_get_blocks(inode, first_block, max_blocks,
+			&bno, &new, &boundary, create);
+	if (ret < 0)
+		return ret;
+
+	iomap->flags = 0;
 	/*
 	 * Writes that span EOF might trigger an IO size update on completion,
 	 * so consider them to be dirty for the purposes of O_DSYNC even if
@@ -812,13 +818,6 @@ static int ext2_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
 	 */
 	if ((flags & IOMAP_WRITE) && offset + length > i_size_read(inode))
 		iomap->flags |= IOMAP_F_DIRTY;
-
-	ret = ext2_get_blocks(inode, first_block, max_blocks,
-			&bno, &new, &boundary, create);
-	if (ret < 0)
-		return ret;
-
-	iomap->flags = 0;
 	iomap->offset = (u64)first_block << blkbits;
 	iomap->bdev = inode->i_sb->s_bdev;
 
Signature · 1 line
2.53.0
author
#3 lore raw
Quoted 4 lines · Jan Kara

Thanks for the fix. I've just move the setting of the IOMAP_F_DIRTY bit instead of moving the initialization. Functionally this is equivalent and the iomap modification is better localized that way (the resulting patch is attached).

Thanks for the adjustment! That looks good to me.

Hungyu