Re: [PATCH 1/2] xfsprogs: xfs_fsr: Interpret arguments of qsort's compare function correctly

[Eric Sandeen <sandeen@sandeen.net>]

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@gmail.com>

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@redhat.com>

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

> ---
>  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);
>  }
>  
>  /*
>