[PATCH] xfs: lseek SEEK_DATA/SEEK_HOLE consolidation

[PATCH] xfs: lseek SEEK_DATA/SEEK_HOLE consolidation

[Jeff Liu]

From: Jie Liu <jeff.liu@oracle.com>

Consolidate lseek(2) SEEK_DATA/SEEK_HOLE according to the
implementation of VFS lseek_execute():
- if end up with a negative offset, return EINVAL if file
  is not huge.
- if end up with an offset larger than s_maxbytes, return
  EINVAL as well.
- reset file version to 0 if end up with an offset that is
  not equal to the current file offset.

Signed-off-by: Jie Liu <jeff.liu@oracle.com>

---
 fs/xfs/xfs_file.c |   26 ++++++++++++++++++++++++--
 1 file changed, 24 insertions(+), 2 deletions(-)

diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
index a5f2042..dc42751 100644
--- a/fs/xfs/xfs_file.c
+++ b/fs/xfs/xfs_file.c
@@ -1270,8 +1270,19 @@ xfs_seek_data(
 	}
 
 out:
-	if (offset != file->f_pos)
+	if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) {
+		error = EINVAL;
+		goto out_unlock;
+	}
+	if (offset > inode->i_sb->s_maxbytes) {
+		error = EINVAL;
+		goto out_unlock;
+	}
+
+	if (offset != file->f_pos) {
 		file->f_pos = offset;
+		file->f_version = 0;
+	}
 
 out_unlock:
 	xfs_iunlock_map_shared(ip, lock);
@@ -1372,6 +1383,15 @@ xfs_seek_hole(
 	}
 
 out:
+	if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) {
+		error = EINVAL;
+		goto out_unlock;
+	}
+	if (offset > inode->i_sb->s_maxbytes) {
+		error = EINVAL;
+		goto out_unlock;
+	}
+
 	/*
 	 * At this point, we must have found a hole.  However, the returned
 	 * offset may be bigger than the file size as it may be aligned to
@@ -1379,8 +1399,10 @@ out:
 	 * situation in particular.
 	 */
 	offset = min_t(loff_t, offset, isize);
-	if (offset != file->f_pos)
+	if (offset != file->f_pos) {
 		file->f_pos = offset;
+		file->f_version = 0;
+	}
 
 out_unlock:
 	xfs_iunlock_map_shared(ip, lock);
-- 
1.7.9.5

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: [PATCH] xfs: lseek SEEK_DATA/SEEK_HOLE consolidation

[Dave Chinner]

On Fri, Jun 21, 2013 at 09:27:58PM +0800, Jeff Liu wrote:
> From: Jie Liu <jeff.liu@oracle.com>
> 
> Consolidate lseek(2) SEEK_DATA/SEEK_HOLE according to the
> implementation of VFS lseek_execute():
> - if end up with a negative offset, return EINVAL if file
>   is not huge.
> - if end up with an offset larger than s_maxbytes, return
>   EINVAL as well.
> - reset file version to 0 if end up with an offset that is
>   not equal to the current file offset.
> 
> Signed-off-by: Jie Liu <jeff.liu@oracle.com>
> 
> ---
>  fs/xfs/xfs_file.c |   26 ++++++++++++++++++++++++--
>  1 file changed, 24 insertions(+), 2 deletions(-)
> 
> diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
> index a5f2042..dc42751 100644
> --- a/fs/xfs/xfs_file.c
> +++ b/fs/xfs/xfs_file.c
> @@ -1270,8 +1270,19 @@ xfs_seek_data(
>  	}
>  
>  out:
> -	if (offset != file->f_pos)
> +	if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) {
> +		error = EINVAL;
> +		goto out_unlock;
> +	}
> +	if (offset > inode->i_sb->s_maxbytes) {
> +		error = EINVAL;
> +		goto out_unlock;
> +	}
> +
> +	if (offset != file->f_pos) {
>  		file->f_pos = offset;
> +		file->f_version = 0;
> +	}

Hi Jeff, why are you copy-n-pasting this code from lseek_execute()
rather than making lseek_execute() an exported function and calling
that directly? 

>  
>  out_unlock:
>  	xfs_iunlock_map_shared(ip, lock);
> @@ -1372,6 +1383,15 @@ xfs_seek_hole(
>  	}
>  
>  out:
> +	if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) {
> +		error = EINVAL;
> +		goto out_unlock;
> +	}
> +	if (offset > inode->i_sb->s_maxbytes) {
> +		error = EINVAL;
> +		goto out_unlock;
> +	}

These checks belong after we truncated offset to isize, don't they?
And that would make both of these functions simply require a call to
lseek_execute(), yes?

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: [PATCH] xfs: lseek SEEK_DATA/SEEK_HOLE consolidation

[Jeff Liu]

Hi Dave,
On 06/24/2013 07:13 AM, Dave Chinner wrote:

> On Fri, Jun 21, 2013 at 09:27:58PM +0800, Jeff Liu wrote:
>> From: Jie Liu <jeff.liu@oracle.com>
>>
>> Consolidate lseek(2) SEEK_DATA/SEEK_HOLE according to the
>> implementation of VFS lseek_execute():
>> - if end up with a negative offset, return EINVAL if file
>>   is not huge.
>> - if end up with an offset larger than s_maxbytes, return
>>   EINVAL as well.
>> - reset file version to 0 if end up with an offset that is
>>   not equal to the current file offset.
>>
>> Signed-off-by: Jie Liu <jeff.liu@oracle.com>
>>
>> ---
>>  fs/xfs/xfs_file.c |   26 ++++++++++++++++++++++++--
>>  1 file changed, 24 insertions(+), 2 deletions(-)
>>
>> diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
>> index a5f2042..dc42751 100644
>> --- a/fs/xfs/xfs_file.c
>> +++ b/fs/xfs/xfs_file.c
>> @@ -1270,8 +1270,19 @@ xfs_seek_data(
>>  	}
>>  
>>  out:
>> -	if (offset != file->f_pos)
>> +	if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) {
>> +		error = EINVAL;
>> +		goto out_unlock;
>> +	}
>> +	if (offset > inode->i_sb->s_maxbytes) {
>> +		error = EINVAL;
>> +		goto out_unlock;
>> +	}
>> +
>> +	if (offset != file->f_pos) {
>>  		file->f_pos = offset;
>> +		file->f_version = 0;
>> +	}
> 
> Hi Jeff, why are you copy-n-pasting this code from lseek_execute()
> rather than making lseek_execute() an exported function and calling
> that directly? 

I found other file systems are implemented in this way.  But I should
consider how to make it better rather than following others in this
situation.

> 
>>  
>>  out_unlock:
>>  	xfs_iunlock_map_shared(ip, lock);
>> @@ -1372,6 +1383,15 @@ xfs_seek_hole(
>>  	}
>>  
>>  out:
>> +	if (offset < 0 && !(file->f_mode & FMODE_UNSIGNED_OFFSET)) {
>> +		error = EINVAL;
>> +		goto out_unlock;
>> +	}
>> +	if (offset > inode->i_sb->s_maxbytes) {
>> +		error = EINVAL;
>> +		goto out_unlock;
>> +	}
> 
> These checks belong after we truncated offset to isize, don't they?

They do.

> And that would make both of these functions simply require a call to
> lseek_execute(), yes?

Yep, I'll post a patch set to export this call and propagate it to other
file systems as well.

Thanks,
-Jeff

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs