[PATCH] ext4: sync the directory inode in ext4_sync_parent()

[PATCH] ext4: sync the directory inode in ext4_sync_parent()

[Curt Wohlgemuth]

ext4 has taken the stance that, in the absence of a journal,
when an fsync/fdatasync of an inode is done, the parent
directory should be sync'ed if this inode entry is new.
ext4_sync_parent(), which implements this, does indeed sync
the dirent pages for parent directories, but it does not
sync the directory *inode*.  This patch fixes this.

I tested this using a power fail test, which panics a
machine running a file server getting requests from a
client.  Without this patch, on about every other test run,
the server is missing many, many files that had been synced.
With this patch, on > 6 runs, I see zero files being lost.

Signed-off-by: Curt Wohlgemuth <curtw@google.com>
---
 fs/ext4/fsync.c |   11 ++++++++++-
 1 files changed, 10 insertions(+), 1 deletions(-)

diff --git a/fs/ext4/fsync.c b/fs/ext4/fsync.c
index 7f74019..a276348 100644
--- a/fs/ext4/fsync.c
+++ b/fs/ext4/fsync.c
@@ -128,6 +128,7 @@ extern int ext4_flush_completed_IO(struct inode *inode)
 static void ext4_sync_parent(struct inode *inode)
 {
 	struct dentry *dentry = NULL;
+	int ret;
 
 	while (inode && ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
 		ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
@@ -136,7 +137,15 @@ static void ext4_sync_parent(struct inode *inode)
 		if (!dentry || !dentry->d_parent || !dentry->d_parent->d_inode)
 			break;
 		inode = dentry->d_parent->d_inode;
-		sync_mapping_buffers(inode->i_mapping);
+		ret = sync_mapping_buffers(inode->i_mapping);
+		if (! ret) {
+			struct writeback_control wbc = {
+				.sync_mode = WB_SYNC_ALL,
+				.nr_to_write = 0, /* metadata-only; caller
+						     takes care of data */
+			};
+			(void)sync_inode(inode, &wbc);
+		}
 	}
 }
 
-- 
1.7.3.1

Re: [PATCH] ext4: sync the directory inode in ext4_sync_parent()

[Eric Sandeen]

On 4/5/11 12:27 PM, Curt Wohlgemuth wrote:
> ext4 has taken the stance that, in the absence of a journal,
> when an fsync/fdatasync of an inode is done, the parent
> directory should be sync'ed if this inode entry is new.
> ext4_sync_parent(), which implements this, does indeed sync
> the dirent pages for parent directories, but it does not
> sync the directory *inode*.  This patch fixes this.
> 
> I tested this using a power fail test, which panics a
> machine running a file server getting requests from a
> client.  Without this patch, on about every other test run,
> the server is missing many, many files that had been synced.
> With this patch, on > 6 runs, I see zero files being lost.
> 
> Signed-off-by: Curt Wohlgemuth <curtw@google.com>
> ---
>  fs/ext4/fsync.c |   11 ++++++++++-
>  1 files changed, 10 insertions(+), 1 deletions(-)
> 
> diff --git a/fs/ext4/fsync.c b/fs/ext4/fsync.c
> index 7f74019..a276348 100644
> --- a/fs/ext4/fsync.c
> +++ b/fs/ext4/fsync.c
> @@ -128,6 +128,7 @@ extern int ext4_flush_completed_IO(struct inode *inode)
>  static void ext4_sync_parent(struct inode *inode)
>  {
>  	struct dentry *dentry = NULL;
> +	int ret;
>  
>  	while (inode && ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
>  		ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
> @@ -136,7 +137,15 @@ static void ext4_sync_parent(struct inode *inode)
>  		if (!dentry || !dentry->d_parent || !dentry->d_parent->d_inode)
>  			break;
>  		inode = dentry->d_parent->d_inode;
> -		sync_mapping_buffers(inode->i_mapping);
> +		ret = sync_mapping_buffers(inode->i_mapping);
> +		if (! ret) {
> +			struct writeback_control wbc = {
> +				.sync_mode = WB_SYNC_ALL,
> +				.nr_to_write = 0, /* metadata-only; caller
> +						     takes care of data */
> +			};
> +			(void)sync_inode(inode, &wbc);

I know ext4_sync_parent was a void already, but why don't we send errors back up through ext4_sync_file?

-Eric

> +		}
>  	}
>  }
>  

Re: [PATCH] ext4: sync the directory inode in ext4_sync_parent()

[Curt Wohlgemuth]

Hi Eric:

On Tue, Apr 5, 2011 at 12:35 PM, Eric Sandeen <sandeen@redhat.com> wrote:
> On 4/5/11 12:27 PM, Curt Wohlgemuth wrote:
>> ext4 has taken the stance that, in the absence of a journal,
>> when an fsync/fdatasync of an inode is done, the parent
>> directory should be sync'ed if this inode entry is new.
>> ext4_sync_parent(), which implements this, does indeed sync
>> the dirent pages for parent directories, but it does not
>> sync the directory *inode*.  This patch fixes this.
>>
>> I tested this using a power fail test, which panics a
>> machine running a file server getting requests from a
>> client.  Without this patch, on about every other test run,
>> the server is missing many, many files that had been synced.
>> With this patch, on > 6 runs, I see zero files being lost.
>>
>> Signed-off-by: Curt Wohlgemuth <curtw@google.com>
>> ---
>>  fs/ext4/fsync.c |   11 ++++++++++-
>>  1 files changed, 10 insertions(+), 1 deletions(-)
>>
>> diff --git a/fs/ext4/fsync.c b/fs/ext4/fsync.c
>> index 7f74019..a276348 100644
>> --- a/fs/ext4/fsync.c
>> +++ b/fs/ext4/fsync.c
>> @@ -128,6 +128,7 @@ extern int ext4_flush_completed_IO(struct inode *inode)
>>  static void ext4_sync_parent(struct inode *inode)
>>  {
>>       struct dentry *dentry = NULL;
>> +     int ret;
>>
>>       while (inode && ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
>>               ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
>> @@ -136,7 +137,15 @@ static void ext4_sync_parent(struct inode *inode)
>>               if (!dentry || !dentry->d_parent || !dentry->d_parent->d_inode)
>>                       break;
>>               inode = dentry->d_parent->d_inode;
>> -             sync_mapping_buffers(inode->i_mapping);
>> +             ret = sync_mapping_buffers(inode->i_mapping);
>> +             if (! ret) {
>> +                     struct writeback_control wbc = {
>> +                             .sync_mode = WB_SYNC_ALL,
>> +                             .nr_to_write = 0, /* metadata-only; caller
>> +                                                  takes care of data */
>> +                     };
>> +                     (void)sync_inode(inode, &wbc);
>
> I know ext4_sync_parent was a void already, but why don't we send errors back up through ext4_sync_file?

Well, you could argue that a failure to sync the parent shouldn't
cause the inode's fsync() to fail, but I probably wouldn't :-) .  I'll
resend a patch that implements this.

Thanks,
Curt

>
> -Eric
>
>> +             }
>>       }
>>  }
>>
>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] ext4: sync the directory inode in ext4_sync_parent()

[Eric Sandeen]

On 4/5/11 12:43 PM, Curt Wohlgemuth wrote:
> Well, you could argue that a failure to sync the parent shouldn't
> cause the inode's fsync() to fail, but I probably wouldn't :-) .  I'll
> resend a patch that implements this.
> 
> Thanks,
> Curt

It might be a valid argument, since we're doing it "just to be nice" - but if you're getting EIOs you'd probably rather know sooner than later...

-Eric

[PATCH] ext4: sync the directory inode in ext4_sync_parent()

[Curt Wohlgemuth]

ext4 has taken the stance that, in the absence of a journal,
when an fsync/fdatasync of an inode is done, the parent
directory should be sync'ed if this inode entry is new.
ext4_sync_parent(), which implements this, does indeed sync
the dirent pages for parent directories, but it does not
sync the directory *inode*.  This patch fixes this.

Also now return error status from ext4_sync_parent().

I tested this using a power fail test, which panics a
machine running a file server getting requests from a
client.  Without this patch, on about every other test run,
the server is missing many, many files that had been synced.
With this patch, on > 6 runs, I see zero files being lost.

Signed-off-by: Curt Wohlgemuth <curtw@google.com>
---
 fs/ext4/fsync.c |   18 +++++++++++++++---
 1 files changed, 15 insertions(+), 3 deletions(-)

diff --git a/fs/ext4/fsync.c b/fs/ext4/fsync.c
index 7f74019..bb0a4e7 100644
--- a/fs/ext4/fsync.c
+++ b/fs/ext4/fsync.c
@@ -125,9 +125,11 @@ extern int ext4_flush_completed_IO(struct inode *inode)
  * the parent directory's parent as well, and so on recursively, if
  * they are also freshly created.
  */
-static void ext4_sync_parent(struct inode *inode)
+static int ext4_sync_parent(struct inode *inode)
 {
 	struct dentry *dentry = NULL;
+	int ret = 0;
+	struct writeback_control wbc;
 
 	while (inode && ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
 		ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
@@ -136,8 +138,18 @@ static void ext4_sync_parent(struct inode *inode)
 		if (!dentry || !dentry->d_parent || !dentry->d_parent->d_inode)
 			break;
 		inode = dentry->d_parent->d_inode;
-		sync_mapping_buffers(inode->i_mapping);
+		ret = sync_mapping_buffers(inode->i_mapping);
+		if (ret)
+			goto out;
+		memset(&wbc, 0, sizeof(wbc));
+		wbc.sync_mode = WB_SYNC_ALL;
+		wbc.nr_to_write = 0;         /* only write out the inode */
+		ret = sync_inode(inode, &wbc);
+		if (ret)
+			goto out;
 	}
+out:
+	return ret;
 }
 
 /*
@@ -176,7 +188,7 @@ int ext4_sync_file(struct file *file, int datasync)
 	if (!journal) {
 		ret = generic_file_fsync(file, datasync);
 		if (!ret && !list_empty(&inode->i_dentry))
-			ext4_sync_parent(inode);
+			ret = ext4_sync_parent(inode);
 		goto out;
 	}
 
-- 
1.7.3.1

Re: [PATCH] ext4: sync the directory inode in ext4_sync_parent()

[Ted Ts'o]

Hi Curt,

Thanks, accepted into the tree.  I made a few tiny stylistic changes
(mainly by replacing "goto out" with "break").

       		     	 	       	 - Ted

>From 8429a2e121372395f25d80a486ef065e77139ebb Mon Sep 17 00:00:00 2001
From: Curt Wohlgemuth <curtw@google.com>
Date: Wed, 6 Apr 2011 16:57:09 -0400
Subject: [PATCH] ext4: sync the directory inode in ext4_sync_parent()

ext4 has taken the stance that, in the absence of a journal,
when an fsync/fdatasync of an inode is done, the parent
directory should be sync'ed if this inode entry is new.
ext4_sync_parent(), which implements this, does indeed sync
the dirent pages for parent directories, but it does not
sync the directory *inode*.  This patch fixes this.

Also now return error status from ext4_sync_parent().

I tested this using a power fail test, which panics a
machine running a file server getting requests from a
client.  Without this patch, on about every other test run,
the server is missing many, many files that had been synced.
With this patch, on > 6 runs, I see zero files being lost.

Signed-off-by: Curt Wohlgemuth <curtw@google.com>
Signed-off-by: "Theodore Ts'o" <tytso@mit.edu>
---
 fs/ext4/fsync.c |   17 ++++++++++++++---
 1 files changed, 14 insertions(+), 3 deletions(-)

diff --git a/fs/ext4/fsync.c b/fs/ext4/fsync.c
index 7f74019..b1f9b5f 100644
--- a/fs/ext4/fsync.c
+++ b/fs/ext4/fsync.c
@@ -125,9 +125,11 @@ extern int ext4_flush_completed_IO(struct inode *inode)
  * the parent directory's parent as well, and so on recursively, if
  * they are also freshly created.
  */
-static void ext4_sync_parent(struct inode *inode)
+static int ext4_sync_parent(struct inode *inode)
 {
+	struct writeback_control wbc;
 	struct dentry *dentry = NULL;
+	int ret = 0;
 
 	while (inode && ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
 		ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
@@ -136,8 +138,17 @@ static void ext4_sync_parent(struct inode *inode)
 		if (!dentry || !dentry->d_parent || !dentry->d_parent->d_inode)
 			break;
 		inode = dentry->d_parent->d_inode;
-		sync_mapping_buffers(inode->i_mapping);
+		ret = sync_mapping_buffers(inode->i_mapping);
+		if (ret)
+			break;
+		memset(&wbc, 0, sizeof(wbc));
+		wbc.sync_mode = WB_SYNC_ALL;
+		wbc.nr_to_write = 0;         /* only write out the inode */
+		ret = sync_inode(inode, &wbc);
+		if (ret)
+			break;
 	}
+	return ret;
 }
 
 /*
@@ -176,7 +187,7 @@ int ext4_sync_file(struct file *file, int datasync)
 	if (!journal) {
 		ret = generic_file_fsync(file, datasync);
 		if (!ret && !list_empty(&inode->i_dentry))
-			ext4_sync_parent(inode);
+			ret = ext4_sync_parent(inode);
 		goto out;
 	}
 
-- 
1.7.3.1