[PATCH 1/2] btrfs: btrfs_defrag_root() doesn't support any option

[PATCH 1/2] btrfs: btrfs_defrag_root() doesn't support any option

[Anand Jain]

Both BTRFS_IOC_DEFRAG and BTRFS_IOC_DEFRAG_RANGE call the same
function- btrfs_ioctl_defrag(), however BTRFS_IOC_DEFRAG does
not support any argument, so check that and return not supported
if provided. This has valid impact at the user end, in the cli below
  btrfs filesystem defrag -clzo /btrfs
the -c option (or similarly other) is only applicable to the file
defrag, which is either provided in the argument or when used along
with the -r option, however as of now the above cli does not report
any error. This patch will fix it.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
---
 fs/btrfs/ioctl.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index 7acbd2cf6192..bd1bb7fcea8a 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -2608,6 +2608,10 @@ static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
 			ret = -EPERM;
 			goto out;
 		}
+		if (argp) {
+			ret = -EOPNOTSUPP;
+			goto out;
+		}
 		ret = btrfs_defrag_root(root);
 		if (ret)
 			goto out;
-- 
2.10.0

[PATCH 2/2 RFC] btrfs: btrfs_defrag_root() doesn't defrag extent root tree

[Anand Jain]

Since btrfs_defrag_leaves() does not support extent_root,
remove its corresponding call. The user can use the file
based defrag to defrag extents as of now.

Signed-off-by: Anand Jain <anand.jain@oracle.com>
---
 fs/btrfs/ioctl.c | 3 ---
 1 file changed, 3 deletions(-)

diff --git a/fs/btrfs/ioctl.c b/fs/btrfs/ioctl.c
index bd1bb7fcea8a..112456baffe9 100644
--- a/fs/btrfs/ioctl.c
+++ b/fs/btrfs/ioctl.c
@@ -2613,9 +2613,6 @@ static int btrfs_ioctl_defrag(struct file *file, void __user *argp)
 			goto out;
 		}
 		ret = btrfs_defrag_root(root);
-		if (ret)
-			goto out;
-		ret = btrfs_defrag_root(root->fs_info->extent_root);
 		break;
 	case S_IFREG:
 		if (!(file->f_mode & FMODE_WRITE)) {
-- 
2.10.0

Re: [PATCH 1/2] btrfs: btrfs_defrag_root() doesn't support any option

[David Sterba]

On Wed, Dec 21, 2016 at 03:42:07PM +0800, Anand Jain wrote:
> Both BTRFS_IOC_DEFRAG and BTRFS_IOC_DEFRAG_RANGE call the same
> function- btrfs_ioctl_defrag(), however BTRFS_IOC_DEFRAG does
> not support any argument, so check that and return not supported
> if provided. This has valid impact at the user end, in the cli below
>   btrfs filesystem defrag -clzo /btrfs
> the -c option (or similarly other) is only applicable to the file
> defrag, which is either provided in the argument or when used along
> with the -r option, however as of now the above cli does not report
> any error. This patch will fix it.

I think it would be better to do the sanitization on the userspace side.
In this case combination of a compression option and bare directory
would not lead to the expected result.

IMO it's deciding between

  "pass any arguments to kernel" - here returning EOPNOTSUPP would be
  quite unclear what went wrong

vs.

  "validate as much arguments as we can before passing to kernel" - and
  we'll know early that the combination is not valid, also that we want
  to deprecate it etc.

I vote for #2 as we have enough information to give an informative error
message.

Re: [PATCH 2/2 RFC] btrfs: btrfs_defrag_root() doesn't defrag extent root tree

[David Sterba]

On Wed, Dec 21, 2016 at 03:42:08PM +0800, Anand Jain wrote:
> Since btrfs_defrag_leaves() does not support extent_root,
> remove its corresponding call. The user can use the file
> based defrag to defrag extents as of now.
> 
> Signed-off-by: Anand Jain <anand.jain@oracle.com>

Reviewed-by: David Sterba <dsterba@suse.com>

Oh right, btrfs_defrag_leaves even explicitly checks for extent_root, so
this has never worked anyway, safe to remove.

Re: [PATCH 1/2] btrfs: btrfs_defrag_root() doesn't support any option

[Anand Jain]

On 01/04/17 00:21, David Sterba wrote:
> On Wed, Dec 21, 2016 at 03:42:07PM +0800, Anand Jain wrote:
>> Both BTRFS_IOC_DEFRAG and BTRFS_IOC_DEFRAG_RANGE call the same
>> function- btrfs_ioctl_defrag(), however BTRFS_IOC_DEFRAG does
>> not support any argument, so check that and return not supported
>> if provided. This has valid impact at the user end, in the cli below
>>   btrfs filesystem defrag -clzo /btrfs
>> the -c option (or similarly other) is only applicable to the file
>> defrag, which is either provided in the argument or when used along
>> with the -r option, however as of now the above cli does not report
>> any error. This patch will fix it.
>
> I think it would be better to do the sanitization on the userspace side.
 >
> In this case combination of a compression option and bare directory
> would not lead to the expected result.

  Hmm.. I am confused on the expected result here, what should it be ?
  My thinking, compression option on the directory is not supported.
  That's because the kernel calls btrfs_defrag_root() for dir as arg,
  which does not accept any option.

  In the same context if either -r option is used (then it picks up
  individual files as arg) or if a file is used as arg, then kernel
  calls btrfs_defrag_files() which then the options makes sense.

> IMO it's deciding between
>
>   "pass any arguments to kernel" - here returning EOPNOTSUPP would be
>   quite unclear what went wrong
>
> vs.
>
>   "validate as much arguments as we can before passing to kernel" - and
>   we'll know early that the combination is not valid, also that we want
>   to deprecate it etc.

  btrfs-progs is one of the tool. If API is tested with a test program
  it will still can be shown as failing.

  The kernel problem statement is:
    - ioctl("/dir", BTRFS_IOC_DEFRAG, options)
    - ioctl("/dir", BTRFS_IOC_DEFRAG_RANGE, options)
  the above ioctl usage provides false impression that options have
  worked, but the kernel does not use the options at all.

  Its a different story if btrfs_defrag_root() would support options
  in the long term ?

> I vote for #2 as we have enough information to give an informative error
> message.

Thanks, Anand