[PATCH] xfs_db: fix the setting of unaligned directory fields

[Mark Tinguely]

On 02/08/14 02:30, Dave Chinner wrote:
> On Fri, Feb 07, 2014 at 05:13:21PM -0600, Mark Tinguely wrote:
>> On 02/07/14 16:33, Dave Chinner wrote:
>>> On Fri, Feb 07, 2014 at 04:03:42PM -0600, Mark Tinguely wrote:
>>>> @@ -526,16 +527,20 @@ convert_arg(
>>>>   		 */
>>>>   		*value = strtoll(arg, NULL, 0);
>>>
>>> If we are touching this code, the return value here should be error
>>> checked.
>>>
>>> xfs_db>   write u3.bmx[0].startblock 3rgfdw
>>> u3.bmx[0].startblock = 52776558133248
>>
>> hmm, It should stop at 3. I will take a look.
>
> Even if it stops at 3, that's still wrong because it's failed to
> process the entire user input....
>
>>> xfs_db>   write u3.bmx[0].startblock x3rgfdw
>>> u3.bmx[0].startblock = 0
>>> xfs_db>
>>>
>>> i.e. it accepts garbage rather than erroring out.
>>
>> as does all the other writes ...
>> xfs_db>   write core.nblocks x3rgfdw
>> core.nblocks = 0
>>
>> Fixing convert_arg() is beyond the scope of just this patch.
>
> Sure, but I'm not asking you to fix all of convert_args in this
> patch, just asking you to do a complete job of fixing the bitval
> input processing in this patch.
>
> But, seeing as you've raised that convert_args() has other broken
> paths, can you also write new patches to address those issues? It
> won't take you long while all this code is fresh in your mind, and
> if you do it now it won't get dropped on the floor until somebody
> else hits it a couple of years down the track...
>

It needs to be split up. write_string() needs string inputs, 
write_struct() need numeric inputs.  Who uses the UUID-style hex blocks? 
It feels like a black hole of time.

It goes on my 'off the clock' to do list.

--Mark.

--Mark.