On 25 Jan 2021 at 19:47, Eric Sandeen wrote:
> On 1/25/21 3:58 AM, Chandan Babu R wrote:
>> The first argument passed to qsort() in fsrfs() is an array of "struct
>> xfs_bulkstat". Hence the two arguments to the cmp() function must be
>> interpreted as being of type "struct xfs_bulkstat *" as against "struct
>> xfs_bstat *" that is being used to currently typecast them.
>>
>> Signed-off-by: Chandan Babu R <chandanrlinux@xxxxxxxxx>
>
> Yikes. Broken since 5.3.0, and the structures have different sizes and
> different bs_extents offsets. :(
>
> Fixes: 4cca629d6 ("misc: convert xfrog_bulkstat functions to have v5 semantics")
> Reviewed-by: Eric Sandeen <sandeen@xxxxxxxxxx>
>
> At least it's only affecting the whole-fs defragment which is generally not
> recommended, but is still available and does get used.
>
> I wonder if it explains this bug report:
>
> Jan 07 20:52:44 <Tharn> hey, quick question... the first time I ran xfs_fsr last night, it ran for 2 hours and looking at the console log, it ended with a lot of "start inode=0" repeating
With my limited understanding of xfs_fsr code, looks like the bug reporter saw
the logs corresponding to the default (i.e. 10) number of passes that xfs_fsr
executes on the inodes of the filesystem. Sorry, I couldn't be of much help in
this case.
>
>> ---
>> fsr/xfs_fsr.c | 5 ++---
>> 1 file changed, 2 insertions(+), 3 deletions(-)
>>
>> diff --git a/fsr/xfs_fsr.c b/fsr/xfs_fsr.c
>> index 77a10a1d..635e4c70 100644
>> --- a/fsr/xfs_fsr.c
>> +++ b/fsr/xfs_fsr.c
>> @@ -702,9 +702,8 @@ out0:
>> int
>> cmp(const void *s1, const void *s2)
>> {
>> - return( ((struct xfs_bstat *)s2)->bs_extents -
>> - ((struct xfs_bstat *)s1)->bs_extents);
>> -
>> + return( ((struct xfs_bulkstat *)s2)->bs_extents -
>> + ((struct xfs_bulkstat *)s1)->bs_extents);
>> }
>>
>> /*
>>
--
chandan
