Re: [PATCH 2/3] simplify xfs_setattr

[Timothy Shimmin <tes@sgi.com>]

Christoph Hellwig wrote:
> On Wed, Jul 09, 2008 at 06:58:21PM +1000, Timothy Shimmin wrote:
>>> -	if (ia_valid & (ATTR_MTIME_SET | ATTR_ATIME_SET))
>>> -		flags |= ATTR_UTIME;
>>> -#ifdef ATTR_NO_BLOCK
>>> -	if ((ia_valid & ATTR_NO_BLOCK))
>>> -		flags |= ATTR_NONBLOCK;
>>> -#endif
>>> -
>> So this code looks different.
>> We are now dropping the flags. Why is that?
>> Presumably because we were mapping ia_valid's:
>>    ATTR_MTIME_SET or ATTR_ATIME_SET --> ATTR_UTIME
>>    ATTR_NO_BLOCK -> ATTR_NONBLOCK
>> But now we pass ATTR_?TIME_SET and ATTR_NO_BLOCK straight thru.
>> So previously we didn't map them onto va_mask bits but as separate flags
>> instead.
> 
> Yeah, not that ATTR_NO_BLOCK doesn't actually exist in any tree I have
> access to, and thus it's not actually handled in the new xfs_setattr.
> 

Hmm...confusing re NO_BLOCK.

1 linux-2.6/xfs_vnode.h <global>           228 #define ATTR_NONBLOCK 0x100
2 linux-2.6/xfs_ioctl.c xfs_ioc_space      686 attr_flags |= ATTR_NONBLOCK;
3 linux-2.6/xfs_ioctl.c xfs_ioc_fssetxattr 899 attr_flags |= ATTR_NONBLOCK;
4 linux-2.6/xfs_ioctl.c xfs_ioc_setxflags  951 attr_flags |= ATTR_NONBLOCK;
5 linux-2.6/xfs_iops.c  xfs_vn_setattr     703 flags |= ATTR_NONBLOCK;

1 xfs/xfs_dmapi.h    <global>     169 #define AT_DELAY_FLAG(f) ((f&ATTR_NONBLOCK) ? DM_FLAGS_NDELAY : 0)
2 xfs/xfs_vnodeops.c xfs_setattr          200 int dmflags = AT_DELAY_FLAG(flags) | DM_SEM_FLAG_WR;
3 xfs/xfs_vnodeops.c xfs_setattr          757 0, 0, AT_DELAY_FLAG(flags));
4 xfs/xfs_vnodeops.c xfs_free_file_space 3536 AT_DELAY_FLAG(attr_flags), NULL);

So it's just used for dmapi stuff as it only looks like it is tested by
AT_DELAY_FLAG().
So for xfs_setattr it will be expecting it coming in as flags param
and then testing it via AT_DELAY_FLAG.
Hmmm...doesn't seem so good. So looks like it was handled by
xfs_setattr and still is.
So are we going to be doing the right thing now?
If it could just never be set then what is the point of using
AT_DELAY_FLAG() in xfs_setattr?


>> So we get rid of the test for XFS_AT_NOSET.
>> where:
>> #define XFS_AT_NOSET    (XFS_AT_NLINK|XFS_AT_RDEV|XFS_AT_FSID|XFS_AT_NODEID|\
>>                 XFS_AT_TYPE|XFS_AT_BLKSIZE|XFS_AT_NBLOCKS|XFS_AT_VCODE|\
>>                 XFS_AT_NEXTENTS|XFS_AT_ANEXTENTS|XFS_AT_GENCOUNT)
>>
>> I can't see anywhere we set any of these.
>> Presumably out of the xattr calls.
>> Some left over from IRIX I guess.
> 
> Probably.  Note that linux uses the ATTR_ flags only for ->setattr so
> there are per defintion none that can't be set.
> 
Which is nice.

>> #define XFS_AT_UPDATIME         0x00010000
>> #define XFS_AT_UPDMTIME         0x00020000
>> #define XFS_AT_UPDCTIME         0x00040000
>> 3 more not supported by vfs ATTR_* macros.
>> I can't see where we set any of these.
>> So no loss there I guess.
>> Presumably they were just for IRIX.
> 
> It's an IRIX leftover.  I will submit a patch to introduce something
> similar to Linux for 2.6.27, that's why I'd like these patches in for
> 2.6.26 so that I have a clean base to start from.
Ok.

--Tim