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(-)

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;
-- 
2.39.2

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.
-- 
Dave Chinner
david@fromorbit.com

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;

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

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