xfs: Use for_each_perag() to iterate all available AGs

  for_each_perag_wrap() doesn't expect 0 as 2nd arg.
To iterate all the available AGs, just use for_each_perag() instead.

Signed-off-by: Ryosuke Yasuoka <ryasuoka@redhat.com>
---
 fs/xfs/xfs_filestream.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
  

On 4/4/23 3:47 AM, Ryosuke Yasuoka wrote:
> for_each_perag_wrap() doesn't expect 0 as 2nd arg.
> To iterate all the available AGs, just use for_each_perag() instead.

Can you explain what goes wrong if it is zero? Is there a test for this?

If it's a general problem, what if the other 2 callers pass in the variable
start_agno with a value of 0?

(I may well be missing something obvious as those macros are a bit dense)

Thanks,
-Eric
 
> Signed-off-by: Ryosuke Yasuoka <ryasuoka@redhat.com>
> ---
>  fs/xfs/xfs_filestream.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/xfs/xfs_filestream.c b/fs/xfs/xfs_filestream.c
> index 22c13933c8f8..48f43c340c58 100644
> --- a/fs/xfs/xfs_filestream.c
> +++ b/fs/xfs/xfs_filestream.c
> @@ -151,7 +151,7 @@ xfs_filestream_pick_ag(
>  		 * grab.
>  		 */
>  		if (!max_pag) {
> -			for_each_perag_wrap(args->mp, 0, start_agno, args->pag)
> +			for_each_perag(args->mp, start_agno, args->pag)
>  				break;
>  			atomic_inc(&args->pag->pagf_fstrms);
>  			*longest = 0;
  

On Tue, Apr 04, 2023 at 05:47:01PM +0900, Ryosuke Yasuoka wrote:
> for_each_perag_wrap() doesn't expect 0 as 2nd arg.
> To iterate all the available AGs, just use for_each_perag() instead.

Thanks, Ryosuke-san. IIUC, this is a fix for the recent sysbot
reported filestreams oops regression?

Can you include the context of the failure it reported (i.e. the
trace from the oops), and the 'reported-by' tag for the syzbot
report?

It should probably also include a 'Fixes: bd4f5d09cc93 ("xfs:
refactor the filestreams allocator pick functions")' tag as well.

> Signed-off-by: Ryosuke Yasuoka <ryasuoka@redhat.com>
> ---
>  fs/xfs/xfs_filestream.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/xfs/xfs_filestream.c b/fs/xfs/xfs_filestream.c
> index 22c13933c8f8..48f43c340c58 100644
> --- a/fs/xfs/xfs_filestream.c
> +++ b/fs/xfs/xfs_filestream.c
> @@ -151,7 +151,7 @@ xfs_filestream_pick_ag(
>  		 * grab.
>  		 */
>  		if (!max_pag) {
> -			for_each_perag_wrap(args->mp, 0, start_agno, args->pag)
> +			for_each_perag(args->mp, start_agno, args->pag)
>  				break;

While this will definitely avoid the oops, I don't think it is quite
right. If we want to iterate all AGs, then we should be starting the
iteration at AG 0, not start_agno. i.e.

+			for_each_perag(args->mp, 0, args->pag)

Cheers,

Dave.
  

On Wed, Apr 05, 2023 at 05:04:14PM +0900, Ryosuke Yasuoka wrote:
> Dave,
> 
> Thank you for reviewing my requests.
> 
> > > for_each_perag_wrap() doesn't expect 0 as 2nd arg.
> > > To iterate all the available AGs, just use for_each_perag() instead.
> >
> > Thanks, Ryosuke-san. IIUC, this is a fix for the recent sysbot
> > reported filestreams oops regression?
> >
> > Can you include the context of the failure it reported (i.e. the
> > trace from the oops), and the 'reported-by' tag for the syzbot
> > report?
> >
> > It should probably also include a 'Fixes: bd4f5d09cc93 ("xfs:
> > refactor the filestreams allocator pick functions")' tag as well.
> 
> No. my request is in the same code area where syzbot bug was reported,
> but it might not be relevant. A kernel applying my patch got the same Oops.
> 
> I'm indeed checking the syzbot's bug and I realized that this small bug fix
> is not related to it based on my tests. Thus I sent the patch
> as a separate one.
> 
> > While this will definitely avoid the oops, I don't think it is quite
> > right. If we want to iterate all AGs, then we should be starting the
> > iteration at AG 0, not start_agno. i.e.
> >
> > +                       for_each_perag(args->mp, 0, args->pag)
> 
> I agree with your proposal because it is more direct.
> However, as the current for_each_perag() macro always assigns 0 to (agno),
> it will cause compilation errors.

Yup, I didn't compile test my suggestion - i just quickly wrote it
down to demonstrate what I was thinking. I expect that you have
understood that using for_each_perag() was what I was suggesting is
used, not that the sample code I wrote is exactly correct. IOWs,

		for_each_perag(args->mp, start_agno, args->pag)

would have worked, even though the code does not do what it looks
like it should from the context of start_agno. Which means this
would be better:

		start_agno = 0;
		for_each_perag_from(args->mp, start_agno, args->pag)

because it directly documents the value we are iterating from.

> Although I haven't checked other callers deeply, we should modify
> the macro as follows:
> 
>  #define for_each_perag(mp, agno, pag) \
> -   (agno) = 0; \
>   for_each_perag_from((mp), (agno), (pag))

That is not correct, either. agno needs to be a variable - it is
the loop agno counter that tracks the iteration.

Cheers,

Dave.
  

Eric,

I failed to reply to you since I got some mistakes.
Let me re-send my reply just in case.

Thank you for reviewing my requests.

> Can you explain what goes wrong if it is zero? Is there a test for this?
>
> If it's a general problem, what if the other 2 callers pass in the variable
> start_agno with a value of 0?

Sorry I couldn't prepare any tests to confirm what happens if it is zero
because it is a kind of general problem.

IIUC, passing zero to for_each_perag_wrap() is not problematic.

As the comment describes, this macro iterates all AG from start_agno through
wrap_agno, then wrap to restart_agno, and then iterates again toward to
start_agno - 1. It looks like some issue occurs when start_agno is zero.
However, for_each_perag_wrap() actually won't wrap if start_agno is zero.

static inline struct xfs_perag *
xfs_perag_next_wrap(
struct xfs_perag *pag,
xfs_agnumber_t *agno,
xfs_agnumber_t stop_agno,
xfs_agnumber_t restart_agno,
xfs_agnumber_t wrap_agno)
{
struct xfs_mount *mp = pag->pag_mount;

*agno = pag->pag_agno + 1;
xfs_perag_rele(pag);
while (*agno != stop_agno) {
if (*agno >= wrap_agno) {
if (restart_agno >= stop_agno) <<<--- HERE
break;
*agno = restart_agno;
}

pag = xfs_perag_grab(mp, *agno);
if (pag)
return pag;
(*agno)++;
}
return NULL;
}

/*
* Iterate all AGs from start_agno through wrap_agno, then restart_agno through
* (start_agno - 1).
*/
#define for_each_perag_wrap_range(mp, start_agno, restart_agno,
wrap_agno, agno, pag) \
for ((agno) = (start_agno), (pag) = xfs_perag_grab((mp), (agno)); \
(pag) != NULL; \
(pag) = xfs_perag_next_wrap((pag), &(agno), (start_agno), \
(restart_agno), (wrap_agno)))
...
#define for_each_perag_wrap_at(mp, start_agno, wrap_agno, agno, pag) \
for_each_perag_wrap_range((mp), (start_agno), 0, (wrap_agno), (agno), (pag))
...
#define for_each_perag_wrap(mp, start_agno, agno, pag) \
for_each_perag_wrap_at((mp), (start_agno), (mp)->m_sb.sb_agcount, \
(agno), (pag))

OTOH, since we have already a for_each_perag() macro, which just iterates all AG
from 0 and doesn't wrap, I think it is simpler to use for_earch_perag().

Regards,
Ryosuke
  

Dave,

> On Thu, Apr 6, 2023 at 8:04 AM Dave Chinner <david@fromorbit.com> wrote:
> >
> > On Wed, Apr 05, 2023 at 05:04:14PM +0900, Ryosuke Yasuoka wrote:
> > > Dave,
> > >
> > > Thank you for reviewing my requests.
> > >
> > > > > for_each_perag_wrap() doesn't expect 0 as 2nd arg.
> > > > > To iterate all the available AGs, just use for_each_perag() instead.
> > > >
> > > > Thanks, Ryosuke-san. IIUC, this is a fix for the recent sysbot
> > > > reported filestreams oops regression?
> > > >
> > > > Can you include the context of the failure it reported (i.e. the
> > > > trace from the oops), and the 'reported-by' tag for the syzbot
> > > > report?
> > > >
> > > > It should probably also include a 'Fixes: bd4f5d09cc93 ("xfs:
> > > > refactor the filestreams allocator pick functions")' tag as well.
> > >
> > > No. my request is in the same code area where syzbot bug was reported,
> > > but it might not be relevant. A kernel applying my patch got the same Oops.
> > >
> > > I'm indeed checking the syzbot's bug and I realized that this small bug fix
> > > is not related to it based on my tests. Thus I sent the patch
> > > as a separate one.
> > >
> > > > While this will definitely avoid the oops, I don't think it is quite
> > > > right. If we want to iterate all AGs, then we should be starting the
> > > > iteration at AG 0, not start_agno. i.e.
> > > >
> > > > +                       for_each_perag(args->mp, 0, args->pag)
> > >
> > > I agree with your proposal because it is more direct.
> > > However, as the current for_each_perag() macro always assigns 0 to (agno),
> > > it will cause compilation errors.
> >
> > Yup, I didn't compile test my suggestion - i just quickly wrote it
> > down to demonstrate what I was thinking. I expect that you have
> > understood that using for_each_perag() was what I was suggesting is
> > used, not that the sample code I wrote is exactly correct. IOWs,
> >
> >                 for_each_perag(args->mp, start_agno, args->pag)
> >
> > would have worked, even though the code does not do what it looks
> > like it should from the context of start_agno. Which means this
> > would be better:
> >
> >                 start_agno = 0;
> >                 for_each_perag_from(args->mp, start_agno, args->pag)
> >
> > because it directly documents the value we are iterating from.

OK. I'll update my patch, run a compile test, and then send again as a
v2 another thread


Thank you for reviewing.
Ryosuke
  

On 4/6/23 11:03 AM, Ryosuke Yasuoka wrote:
> Eric,
> 
> I failed to reply to you since I got some mistakes.
> Let me re-send my reply just in case.
> 
> Thank you for reviewing my requests.
> 
>> Can you explain what goes wrong if it is zero? Is there a test for this?
>>
>> If it's a general problem, what if the other 2 callers pass in the variable
>> start_agno with a value of 0?
> Sorry I couldn't prepare any tests to confirm what happens if it is zero
> because it is a kind of general problem.
> 
> IIUC, passing zero to for_each_perag_wrap() is not problematic.

...

> OTOH, since we have already a for_each_perag() macro, which just iterates all AG
> from 0 and doesn't wrap, I think it is simpler to use for_earch_perag().
> 
> Regards,
> Ryosuke

Ok - I couldn't tell from the original email if this was a bugfix or a 
cleanup, and wanted to be sure.

Thanks!
-Eric