Skip to contentSkip to conversation

ext4: fix i_version handling in ext4

PATCH In review Jeff Layton 3 participants 9 messages Aug 16, 2022 – Aug 17, 2022

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ext4: fix i_version handling in ext4

ext4 currently updates the i_version counter when the atime is updated during a read. This is less than ideal as it can cause unnecessary cache invalidations with NFSv4. The increment in ext4_mark_iloc_dirty is also problematic since it can also corrupt the i_version counter for ea_inodes.

We aren't bumping the file times in ext4_mark_iloc_dirty, so changing the i_version there seems wrong, and is the cause of both problems. Remove that callsite and add increments to the setattr and setxattr codepaths (at the same time that we update the ctime). The i_version bump that already happens during timestamp updates should take care of the rest.

Cc Lukas Czerner <[email protected]>
Cc Jan Kara <[email protected]>
Cc Christian Brauner <[email protected]>
---
 fs/ext4/inode.c | 10 +++++-----
 fs/ext4/xattr.c |  2 ++
 2 files changed, 7 insertions(+), 5 deletions(-)

I think this patch should probably supersede Lukas' patch entitled:

ext4: don't increase iversion counter for ea_inodes

This will also mean that we'll need to respin the patch to turn on the i_version counter unconditionally in ext4 (though that should be trivial).

fs/ext4/inode.c +5 -5
index 601214453c3a..a70921df89a5 100644
@@ -5342,6 +5342,7 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 	int error, rc = 0;
 	int orphan = 0;
 	const unsigned int ia_valid = attr->ia_valid;
+	bool inc_ivers = IS_IVERSION(inode);
 
 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 		return -EIO;
@@ -5425,8 +5426,8 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 			return -EINVAL;
 		}
 
-		if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
-			inode_inc_iversion(inode);
+		if (attr->ia_size == inode->i_size)
+			inc_ivers = false;
 
 		if (shrink) {
 			if (ext4_should_order_data(inode)) {
@@ -5528,6 +5529,8 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 	}
 
 	if (!error) {
+		if (inc_ivers)
+			inode_inc_iversion(inode);
 		setattr_copy(mnt_userns, inode, attr);
 		mark_inode_dirty(inode);
 	}
@@ -5731,9 +5734,6 @@ int ext4_mark_iloc_dirty(handle_t *handle,
 	}
 	ext4_fc_track_inode(handle, inode);
 
-	if (IS_I_VERSION(inode))
-		inode_inc_iversion(inode);
-
 	/* the do_update_inode consumes one bh->b_count */
 	get_bh(iloc->bh);
 
fs/ext4/xattr.c +2 -0
index 533216e80fa2..4d84919d1c9c 100644
@@ -2412,6 +2412,8 @@ ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
 	if (!error) {
 		ext4_xattr_update_super_block(handle, inode->i_sb);
 		inode->i_ctime = current_time(inode);
+		if (IS_IVERSION(inode))
+			inode_inc_iversion(inode);
 		if (!value)
 			no_expand = 0;
 		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
Signature · 1 line
2.37.2
#1 lore raw
Quoted 13 lines · Jeff Layton

ext4 currently updates the i_version counter when the atime is updated during a read. This is less than ideal as it can cause unnecessary cache invalidations with NFSv4. The increment in ext4_mark_iloc_dirty is also problematic since it can also corrupt the i_version counter for ea_inodes.

We aren't bumping the file times in ext4_mark_iloc_dirty, so changing the i_version there seems wrong, and is the cause of both problems. Remove that callsite and add increments to the setattr and setxattr codepaths (at the same time that we update the ctime). The i_version bump that already happens during timestamp updates should take care of the rest.

Quoted 5 lines
Cc Lukas Czerner <[email protected]>
Cc Jan Kara <[email protected]>
Cc Christian Brauner <[email protected]>

---

Seems good to me. But it seems that the xfs patch you sent does have inode_inc_version() right after setattr_copy() as well. So I wonder if we couldn't just try and move inode_inc_version() into setattr_copy() itself.

Quoted 70 lines
 fs/ext4/inode.c | 10 +++++-----
 fs/ext4/xattr.c |  2 ++
 2 files changed, 7 insertions(+), 5 deletions(-)

I think this patch should probably supersede Lukas' patch entitled:

ext4: don't increase iversion counter for ea_inodes

This will also mean that we'll need to respin the patch to turn on the i_version counter unconditionally in ext4 (though that should be trivial).

fs/ext4/inode.c +5 -5
index 601214453c3a..a70921df89a5 100644
@@ -5342,6 +5342,7 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 	int error, rc = 0;
 	int orphan = 0;
 	const unsigned int ia_valid = attr->ia_valid;
+	bool inc_ivers = IS_IVERSION(inode);
 
 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 		return -EIO;
@@ -5425,8 +5426,8 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 			return -EINVAL;
 		}
 
-		if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
-			inode_inc_iversion(inode);
+		if (attr->ia_size == inode->i_size)
+			inc_ivers = false;
 
 		if (shrink) {
 			if (ext4_should_order_data(inode)) {
@@ -5528,6 +5529,8 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 	}
 
 	if (!error) {
+		if (inc_ivers)
+			inode_inc_iversion(inode);
 		setattr_copy(mnt_userns, inode, attr);
 		mark_inode_dirty(inode);
 	}
@@ -5731,9 +5734,6 @@ int ext4_mark_iloc_dirty(handle_t *handle,
 	}
 	ext4_fc_track_inode(handle, inode);
 
-	if (IS_I_VERSION(inode))
-		inode_inc_iversion(inode);
-
 	/* the do_update_inode consumes one bh->b_count */
 	get_bh(iloc->bh);
 
fs/ext4/xattr.c +2 -0
index 533216e80fa2..4d84919d1c9c 100644
@@ -2412,6 +2412,8 @@ ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
 	if (!error) {
 		ext4_xattr_update_super_block(handle, inode->i_sb);
 		inode->i_ctime = current_time(inode);
+		if (IS_IVERSION(inode))
+			inode_inc_iversion(inode);
 		if (!value)
 			no_expand = 0;
 		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
Signature · 1 line
2.37.2
#2 lore raw
Quoted 20 lines · Christian Brauner
Quoted 13 lines · Jeff Layton

ext4 currently updates the i_version counter when the atime is updated during a read. This is less than ideal as it can cause unnecessary cache invalidations with NFSv4. The increment in ext4_mark_iloc_dirty is also problematic since it can also corrupt the i_version counter for ea_inodes.

We aren't bumping the file times in ext4_mark_iloc_dirty, so changing the i_version there seems wrong, and is the cause of both problems. Remove that callsite and add increments to the setattr and setxattr codepaths (at the same time that we update the ctime). The i_version bump that already happens during timestamp updates should take care of the rest.

Quoted 5 lines
Cc Lukas Czerner <[email protected]>
Cc Jan Kara <[email protected]>
Cc Christian Brauner <[email protected]>

---

Quoted 5 lines

Seems good to me. But it seems that the xfs patch you sent does have inode_inc_version() right after setattr_copy() as well. So I wonder if we couldn't just try and move inode_inc_version() into setattr_copy() itself.

We probably could, but setattr_copy has a lot of callers and most filesystems don't need this. Also, there are some cases where we don't want to update the i_version after a setattr.

In particular, if you do a truncate and the size doesn't change, then you really don't want to update the timestamps (and therefore the i_version shouldn't change either).

Quoted 70 lines
Quoted 70 lines
 fs/ext4/inode.c | 10 +++++-----
 fs/ext4/xattr.c |  2 ++
 2 files changed, 7 insertions(+), 5 deletions(-)

I think this patch should probably supersede Lukas' patch entitled:

ext4: don't increase iversion counter for ea_inodes

This will also mean that we'll need to respin the patch to turn on the i_version counter unconditionally in ext4 (though that should be trivial).

fs/ext4/inode.c +5 -5
index 601214453c3a..a70921df89a5 100644
@@ -5342,6 +5342,7 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 	int error, rc = 0;
 	int orphan = 0;
 	const unsigned int ia_valid = attr->ia_valid;
+	bool inc_ivers = IS_IVERSION(inode);
 
 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 		return -EIO;
@@ -5425,8 +5426,8 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 			return -EINVAL;
 		}
 
-		if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
-			inode_inc_iversion(inode);
+		if (attr->ia_size == inode->i_size)
+			inc_ivers = false;
 
 		if (shrink) {
 			if (ext4_should_order_data(inode)) {
@@ -5528,6 +5529,8 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 	}
 
 	if (!error) {
+		if (inc_ivers)
+			inode_inc_iversion(inode);
 		setattr_copy(mnt_userns, inode, attr);
 		mark_inode_dirty(inode);
 	}
@@ -5731,9 +5734,6 @@ int ext4_mark_iloc_dirty(handle_t *handle,
 	}
 	ext4_fc_track_inode(handle, inode);
 
-	if (IS_I_VERSION(inode))
-		inode_inc_iversion(inode);
-
 	/* the do_update_inode consumes one bh->b_count */
 	get_bh(iloc->bh);
 
fs/ext4/xattr.c +2 -0
index 533216e80fa2..4d84919d1c9c 100644
@@ -2412,6 +2412,8 @@ ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
 	if (!error) {
 		ext4_xattr_update_super_block(handle, inode->i_sb);
 		inode->i_ctime = current_time(inode);
+		if (IS_IVERSION(inode))
+			inode_inc_iversion(inode);
 		if (!value)
 			no_expand = 0;
 		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
Signature · 1 line
2.37.2
Signature · 1 line
Jeff Layton <[email protected]>
maintainer Reviewed-by
#3 lore raw
Quoted 27 lines · Jeff Layton
Quoted 20 lines · Christian Brauner
Quoted 13 lines · Jeff Layton

ext4 currently updates the i_version counter when the atime is updated during a read. This is less than ideal as it can cause unnecessary cache invalidations with NFSv4. The increment in ext4_mark_iloc_dirty is also problematic since it can also corrupt the i_version counter for ea_inodes.

We aren't bumping the file times in ext4_mark_iloc_dirty, so changing the i_version there seems wrong, and is the cause of both problems. Remove that callsite and add increments to the setattr and setxattr codepaths (at the same time that we update the ctime). The i_version bump that already happens during timestamp updates should take care of the rest.

Quoted 5 lines
Cc Lukas Czerner <[email protected]>
Cc Jan Kara <[email protected]>
Cc Christian Brauner <[email protected]>

---

Quoted 5 lines

Seems good to me. But it seems that the xfs patch you sent does have inode_inc_version() right after setattr_copy() as well. So I wonder if we couldn't just try and move inode_inc_version() into setattr_copy() itself.

Quoted 7 lines

We probably could, but setattr_copy has a lot of callers and most filesystems don't need this. Also, there are some cases where we don't want to update the i_version after a setattr.

In particular, if you do a truncate and the size doesn't change, then you really don't want to update the timestamps (and therefore the i_version shouldn't change either).

We could probably all handle that with some massaging but I'm also fine with doing it just for ext4 and xfs if these are the only ones where this is relevant:

reviewer Reviewed-by on v1 1/1
#4 lore raw
In reply to Jeff Layton
Quoted 13 lines · Jeff Layton

ext4 currently updates the i_version counter when the atime is updated during a read. This is less than ideal as it can cause unnecessary cache invalidations with NFSv4. The increment in ext4_mark_iloc_dirty is also problematic since it can also corrupt the i_version counter for ea_inodes.

We aren't bumping the file times in ext4_mark_iloc_dirty, so changing the i_version there seems wrong, and is the cause of both problems. Remove that callsite and add increments to the setattr and setxattr codepaths (at the same time that we update the ctime). The i_version bump that already happens during timestamp updates should take care of the rest.

Quoted 4 lines
Cc Lukas Czerner <[email protected]>
Cc Jan Kara <[email protected]>
Cc Christian Brauner <[email protected]>

After some verification (which was not completely trivial e.g. for directories) I agree all cases should be covered. Feel free to add:

Honza

Quoted 71 lines
---
 fs/ext4/inode.c | 10 +++++-----
 fs/ext4/xattr.c |  2 ++
 2 files changed, 7 insertions(+), 5 deletions(-)

I think this patch should probably supersede Lukas' patch entitled:

ext4: don't increase iversion counter for ea_inodes

This will also mean that we'll need to respin the patch to turn on the i_version counter unconditionally in ext4 (though that should be trivial).

fs/ext4/inode.c +5 -5
index 601214453c3a..a70921df89a5 100644
@@ -5342,6 +5342,7 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 	int error, rc = 0;
 	int orphan = 0;
 	const unsigned int ia_valid = attr->ia_valid;
+	bool inc_ivers = IS_IVERSION(inode);
 
 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 		return -EIO;
@@ -5425,8 +5426,8 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 			return -EINVAL;
 		}
 
-		if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
-			inode_inc_iversion(inode);
+		if (attr->ia_size == inode->i_size)
+			inc_ivers = false;
 
 		if (shrink) {
 			if (ext4_should_order_data(inode)) {
@@ -5528,6 +5529,8 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 	}
 
 	if (!error) {
+		if (inc_ivers)
+			inode_inc_iversion(inode);
 		setattr_copy(mnt_userns, inode, attr);
 		mark_inode_dirty(inode);
 	}
@@ -5731,9 +5734,6 @@ int ext4_mark_iloc_dirty(handle_t *handle,
 	}
 	ext4_fc_track_inode(handle, inode);
 
-	if (IS_I_VERSION(inode))
-		inode_inc_iversion(inode);
-
 	/* the do_update_inode consumes one bh->b_count */
 	get_bh(iloc->bh);
 
fs/ext4/xattr.c +2 -0
index 533216e80fa2..4d84919d1c9c 100644
@@ -2412,6 +2412,8 @@ ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
 	if (!error) {
 		ext4_xattr_update_super_block(handle, inode->i_sb);
 		inode->i_ctime = current_time(inode);
+		if (IS_IVERSION(inode))
+			inode_inc_iversion(inode);
 		if (!value)
 			no_expand = 0;
 		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
Signature · 1 line
2.37.2
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#5 lore raw
Quoted 19 lines · Jan Kara
Quoted 13 lines · Jeff Layton

ext4 currently updates the i_version counter when the atime is updated during a read. This is less than ideal as it can cause unnecessary cache invalidations with NFSv4. The increment in ext4_mark_iloc_dirty is also problematic since it can also corrupt the i_version counter for ea_inodes.

We aren't bumping the file times in ext4_mark_iloc_dirty, so changing the i_version there seems wrong, and is the cause of both problems. Remove that callsite and add increments to the setattr and setxattr codepaths (at the same time that we update the ctime). The i_version bump that already happens during timestamp updates should take care of the rest.

Quoted 4 lines
Cc Lukas Czerner <[email protected]>
Cc Jan Kara <[email protected]>
Cc Christian Brauner <[email protected]>
Quoted 7 lines

After some verification (which was not completely trivial e.g. for directories) I agree all cases should be covered. Feel free to add:

Honza

Thanks.

I think this covers the typical cases, but there are some places I missed:

The setacl codepath, for one, and there are a number of places that set the ctime explicitly for hole punching and the like. I'm planning to send a v2 once I do a bit more testing. I'll hold off on adding your Reviewed-by just yet, since the final patch may be quite a bit different.

Quoted 71 lines
Quoted 71 lines
---
 fs/ext4/inode.c | 10 +++++-----
 fs/ext4/xattr.c |  2 ++
 2 files changed, 7 insertions(+), 5 deletions(-)

I think this patch should probably supersede Lukas' patch entitled:

ext4: don't increase iversion counter for ea_inodes

This will also mean that we'll need to respin the patch to turn on the i_version counter unconditionally in ext4 (though that should be trivial).

fs/ext4/inode.c +5 -5
index 601214453c3a..a70921df89a5 100644
@@ -5342,6 +5342,7 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 	int error, rc = 0;
 	int orphan = 0;
 	const unsigned int ia_valid = attr->ia_valid;
+	bool inc_ivers = IS_IVERSION(inode);
 
 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 		return -EIO;
@@ -5425,8 +5426,8 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 			return -EINVAL;
 		}
 
-		if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
-			inode_inc_iversion(inode);
+		if (attr->ia_size == inode->i_size)
+			inc_ivers = false;
 
 		if (shrink) {
 			if (ext4_should_order_data(inode)) {
@@ -5528,6 +5529,8 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 	}
 
 	if (!error) {
+		if (inc_ivers)
+			inode_inc_iversion(inode);
 		setattr_copy(mnt_userns, inode, attr);
 		mark_inode_dirty(inode);
 	}
@@ -5731,9 +5734,6 @@ int ext4_mark_iloc_dirty(handle_t *handle,
 	}
 	ext4_fc_track_inode(handle, inode);
 
-	if (IS_I_VERSION(inode))
-		inode_inc_iversion(inode);
-
 	/* the do_update_inode consumes one bh->b_count */
 	get_bh(iloc->bh);
 
fs/ext4/xattr.c +2 -0
index 533216e80fa2..4d84919d1c9c 100644
@@ -2412,6 +2412,8 @@ ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
 	if (!error) {
 		ext4_xattr_update_super_block(handle, inode->i_sb);
 		inode->i_ctime = current_time(inode);
+		if (IS_IVERSION(inode))
+			inode_inc_iversion(inode);
 		if (!value)
 			no_expand = 0;
 		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
Signature · 1 line
2.37.2
Signature · 1 line
Jeff Layton <[email protected]>
reviewer
#6 lore raw
Quoted 28 lines · Jeff Layton
Quoted 19 lines · Jan Kara
Quoted 13 lines · Jeff Layton

ext4 currently updates the i_version counter when the atime is updated during a read. This is less than ideal as it can cause unnecessary cache invalidations with NFSv4. The increment in ext4_mark_iloc_dirty is also problematic since it can also corrupt the i_version counter for ea_inodes.

We aren't bumping the file times in ext4_mark_iloc_dirty, so changing the i_version there seems wrong, and is the cause of both problems. Remove that callsite and add increments to the setattr and setxattr codepaths (at the same time that we update the ctime). The i_version bump that already happens during timestamp updates should take care of the rest.

Quoted 4 lines
Cc Lukas Czerner <[email protected]>
Cc Jan Kara <[email protected]>
Cc Christian Brauner <[email protected]>
Quoted 7 lines

After some verification (which was not completely trivial e.g. for directories) I agree all cases should be covered. Feel free to add:

Honza

Quoted 7 lines

Thanks.

I think this covers the typical cases, but there are some places I missed:

The setacl codepath, for one, and there are a number of places that set the ctime explicitly for hole punching and the like.

Hum, why is setacl() not covered by your change to ext4_xattr_set_handle()? ext4_set_acl() ends up calling it... I have checked hole punching (whole ext4_fallocate()) and it seems to be incrementing iversion where needed.

Quoted 4 lines

I'm planning to
send a v2 once I do a bit more testing. I'll hold off on adding your Reviewed-by just yet, since the final patch may be quite a bit different.

OK, thanks!

Honza

Quoted 76 lines
Quoted 73 lines
Quoted 73 lines

---
 fs/ext4/inode.c | 10 +++++-----
 fs/ext4/xattr.c |  2 ++
 2 files changed, 7 insertions(+), 5 deletions(-)

I think this patch should probably supersede Lukas' patch entitled:

ext4: don't increase iversion counter for ea_inodes

This will also mean that we'll need to respin the patch to turn on the i_version counter unconditionally in ext4 (though that should be trivial).

fs/ext4/inode.c +5 -5
index 601214453c3a..a70921df89a5 100644
@@ -5342,6 +5342,7 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 	int error, rc = 0;
 	int orphan = 0;
 	const unsigned int ia_valid = attr->ia_valid;
+	bool inc_ivers = IS_IVERSION(inode);
 
 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 		return -EIO;
@@ -5425,8 +5426,8 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 			return -EINVAL;
 		}
 
-		if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
-			inode_inc_iversion(inode);
+		if (attr->ia_size == inode->i_size)
+			inc_ivers = false;
 
 		if (shrink) {
 			if (ext4_should_order_data(inode)) {
@@ -5528,6 +5529,8 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 	}
 
 	if (!error) {
+		if (inc_ivers)
+			inode_inc_iversion(inode);
 		setattr_copy(mnt_userns, inode, attr);
 		mark_inode_dirty(inode);
 	}
@@ -5731,9 +5734,6 @@ int ext4_mark_iloc_dirty(handle_t *handle,
 	}
 	ext4_fc_track_inode(handle, inode);
 
-	if (IS_I_VERSION(inode))
-		inode_inc_iversion(inode);
-
 	/* the do_update_inode consumes one bh->b_count */
 	get_bh(iloc->bh);
 
fs/ext4/xattr.c +2 -0
index 533216e80fa2..4d84919d1c9c 100644
@@ -2412,6 +2412,8 @@ ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
 	if (!error) {
 		ext4_xattr_update_super_block(handle, inode->i_sb);
 		inode->i_ctime = current_time(inode);
+		if (IS_IVERSION(inode))
+			inode_inc_iversion(inode);
 		if (!value)
 			no_expand = 0;
 		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
Signature · 1 line
2.37.2
Signature · 1 line
Jeff Layton <[email protected]>
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
#7 lore raw
Quoted 34 lines · Jan Kara
Quoted 33 lines · Jeff Layton
Quoted 19 lines · Jan Kara
Quoted 13 lines · Jeff Layton

ext4 currently updates the i_version counter when the atime is updated during a read. This is less than ideal as it can cause unnecessary cache invalidations with NFSv4. The increment in ext4_mark_iloc_dirty is also problematic since it can also corrupt the i_version counter for ea_inodes.

We aren't bumping the file times in ext4_mark_iloc_dirty, so changing the i_version there seems wrong, and is the cause of both problems. Remove that callsite and add increments to the setattr and setxattr codepaths (at the same time that we update the ctime). The i_version bump that already happens during timestamp updates should take care of the rest.

Quoted 4 lines
Cc Lukas Czerner <[email protected]>
Cc Jan Kara <[email protected]>
Cc Christian Brauner <[email protected]>
Quoted 7 lines

After some verification (which was not completely trivial e.g. for directories) I agree all cases should be covered. Feel free to add:

Honza

Thanks.

I think this covers the typical cases, but there are some places I missed:

Quoted 7 lines
Quoted 2 lines

The setacl codepath, for one, and there are a number of places that set the ctime explicitly for hole punching and the like.

Hum, why is setacl() not covered by your change to ext4_xattr_set_handle()? ext4_set_acl() ends up calling it... I have checked hole punching (whole ext4_fallocate()) and it seems to be incrementing iversion where needed.

Oh, ok! I mostly noticed the places I was "missing" by inspection. It's possible that I don't need those changes after all. If you think this patch is sufficient, then I'll plan to just go with this one.

Quoted 85 lines
Quoted 4 lines

I'm planning to
send a v2 once I do a bit more testing. I'll hold off on adding your Reviewed-by just yet, since the final patch may be quite a bit different.

OK, thanks!

Honza

Quoted 76 lines
Quoted 73 lines
Quoted 73 lines

---
 fs/ext4/inode.c | 10 +++++-----
 fs/ext4/xattr.c |  2 ++
 2 files changed, 7 insertions(+), 5 deletions(-)

I think this patch should probably supersede Lukas' patch entitled:

ext4: don't increase iversion counter for ea_inodes

This will also mean that we'll need to respin the patch to turn on the i_version counter unconditionally in ext4 (though that should be trivial).

fs/ext4/inode.c +5 -5
index 601214453c3a..a70921df89a5 100644
@@ -5342,6 +5342,7 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 	int error, rc = 0;
 	int orphan = 0;
 	const unsigned int ia_valid = attr->ia_valid;
+	bool inc_ivers = IS_IVERSION(inode);
 
 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 		return -EIO;
@@ -5425,8 +5426,8 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 			return -EINVAL;
 		}
 
-		if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
-			inode_inc_iversion(inode);
+		if (attr->ia_size == inode->i_size)
+			inc_ivers = false;
 
 		if (shrink) {
 			if (ext4_should_order_data(inode)) {
@@ -5528,6 +5529,8 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 	}
 
 	if (!error) {
+		if (inc_ivers)
+			inode_inc_iversion(inode);
 		setattr_copy(mnt_userns, inode, attr);
 		mark_inode_dirty(inode);
 	}
@@ -5731,9 +5734,6 @@ int ext4_mark_iloc_dirty(handle_t *handle,
 	}
 	ext4_fc_track_inode(handle, inode);
 
-	if (IS_I_VERSION(inode))
-		inode_inc_iversion(inode);
-
 	/* the do_update_inode consumes one bh->b_count */
 	get_bh(iloc->bh);
 
fs/ext4/xattr.c +2 -0
index 533216e80fa2..4d84919d1c9c 100644
@@ -2412,6 +2412,8 @@ ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
 	if (!error) {
 		ext4_xattr_update_super_block(handle, inode->i_sb);
 		inode->i_ctime = current_time(inode);
+		if (IS_IVERSION(inode))
+			inode_inc_iversion(inode);
 		if (!value)
 			no_expand = 0;
 		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
Signature · 1 line
2.37.2
Signature · 1 line
Jeff Layton <[email protected]>
Signature · 1 line
Jeff Layton <[email protected]>
reviewer
#8 lore raw
Quoted 38 lines · Jeff Layton
Quoted 37 lines · Jan Kara
Quoted 35 lines · Jeff Layton
Quoted 19 lines · Jan Kara
Quoted 13 lines · Jeff Layton

ext4 currently updates the i_version counter when the atime is updated during a read. This is less than ideal as it can cause unnecessary cache invalidations with NFSv4. The increment in ext4_mark_iloc_dirty is also problematic since it can also corrupt the i_version counter for ea_inodes.

We aren't bumping the file times in ext4_mark_iloc_dirty, so changing the i_version there seems wrong, and is the cause of both problems. Remove that callsite and add increments to the setattr and setxattr codepaths (at the same time that we update the ctime). The i_version bump that already happens during timestamp updates should take care of the rest.

Quoted 4 lines
Cc Lukas Czerner <[email protected]>
Cc Jan Kara <[email protected]>
Cc Christian Brauner <[email protected]>
Quoted 7 lines

After some verification (which was not completely trivial e.g. for directories) I agree all cases should be covered. Feel free to add:

Honza

Thanks.

I think this covers the typical cases, but there are some places I missed:

The setacl codepath, for one, and there are a number of places that set the ctime explicitly for hole punching and the like.

Quoted 8 lines
Quoted 4 lines

Hum, why is setacl() not covered by your change to ext4_xattr_set_handle()? ext4_set_acl() ends up calling it... I have checked hole punching (whole ext4_fallocate()) and it seems to be incrementing iversion where needed.

Oh, ok! I mostly noticed the places I was "missing" by inspection. It's possible that I don't need those changes after all. If you think this patch is sufficient, then I'll plan to just go with this one.

So I did some more grepping and I think we are missing i_version increment in ext4 ioctl handling. In particular stuff like swap_inode_boot_loader(), ext4_ioctl_setflags(), ext4_ioctl_setproject(), and EXT4_IOC_SETVERSION handling in __ext4_ioctl() need i_version increment. Similarly for defrag ioctl implemented in ext4_move_extents() (that even seems to miss a ctime update).

Honza

Quoted 77 lines
Quoted 74 lines
Quoted 74 lines
Quoted 71 lines
Quoted 71 lines
---
 fs/ext4/inode.c | 10 +++++-----
 fs/ext4/xattr.c |  2 ++
 2 files changed, 7 insertions(+), 5 deletions(-)

I think this patch should probably supersede Lukas' patch entitled:

ext4: don't increase iversion counter for ea_inodes

This will also mean that we'll need to respin the patch to turn on the i_version counter unconditionally in ext4 (though that should be trivial).

fs/ext4/inode.c +5 -5
index 601214453c3a..a70921df89a5 100644
@@ -5342,6 +5342,7 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 	int error, rc = 0;
 	int orphan = 0;
 	const unsigned int ia_valid = attr->ia_valid;
+	bool inc_ivers = IS_IVERSION(inode);
 
 	if (unlikely(ext4_forced_shutdown(EXT4_SB(inode->i_sb))))
 		return -EIO;
@@ -5425,8 +5426,8 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 			return -EINVAL;
 		}
 
-		if (IS_I_VERSION(inode) && attr->ia_size != inode->i_size)
-			inode_inc_iversion(inode);
+		if (attr->ia_size == inode->i_size)
+			inc_ivers = false;
 
 		if (shrink) {
 			if (ext4_should_order_data(inode)) {
@@ -5528,6 +5529,8 @@ int ext4_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
 	}
 
 	if (!error) {
+		if (inc_ivers)
+			inode_inc_iversion(inode);
 		setattr_copy(mnt_userns, inode, attr);
 		mark_inode_dirty(inode);
 	}
@@ -5731,9 +5734,6 @@ int ext4_mark_iloc_dirty(handle_t *handle,
 	}
 	ext4_fc_track_inode(handle, inode);
 
-	if (IS_I_VERSION(inode))
-		inode_inc_iversion(inode);
-
 	/* the do_update_inode consumes one bh->b_count */
 	get_bh(iloc->bh);
 
fs/ext4/xattr.c +2 -0
index 533216e80fa2..4d84919d1c9c 100644
@@ -2412,6 +2412,8 @@ ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index,
 	if (!error) {
 		ext4_xattr_update_super_block(handle, inode->i_sb);
 		inode->i_ctime = current_time(inode);
+		if (IS_IVERSION(inode))
+			inode_inc_iversion(inode);
 		if (!value)
 			no_expand = 0;
 		error = ext4_mark_iloc_dirty(handle, inode, &is.iloc);
Signature · 1 line
2.37.2
Signature · 1 line
Jeff Layton <[email protected]>
Signature · 1 line
Jeff Layton <[email protected]>
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR