[PATCH] afs: Fix potential thrashing in afs writeback

[Marc Dionne]

On Thu, Mar 10, 2022 at 11:47 AM David Howells <dhowells at redhat.com> wrote:
>
> In afs_writepages_region(), if the dirty page we find is undergoing
> writeback or write to cache, but the sync_mode is WB_SYNC_NONE, we go round
> the loop trying the same page again and again with no pausing or waiting
> unless and until another thread manages to clear the writeback and fscache
> flags.
>
> Fix this with three measures:
>
>  (1) Advance the start to after the page we found.
>
>  (2) Break out of the loop and return if rescheduling is requested.
>
>  (3) Arbitrarily give up after a maximum of 5 skips.
>
> Fixes: 31143d5d515e ("AFS: implement basic file write support")
> Reported-by: Marc Dionne <marc.dionne at auristor.com>
> Signed-off-by: David Howells <dhowells at redhat.com>
> ---
>
>  fs/afs/write.c |    9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
>
> diff --git a/fs/afs/write.c b/fs/afs/write.c
> index 85c9056ba9fb..bd0201f4939a 100644
> --- a/fs/afs/write.c
> +++ b/fs/afs/write.c
> @@ -701,7 +701,7 @@ static int afs_writepages_region(struct address_space *mapping,
>         struct folio *folio;
>         struct page *head_page;
>         ssize_t ret;
> -       int n;
> +       int n, skips = 0;
>
>         _enter("%llx,%llx,", start, end);
>
> @@ -752,8 +752,15 @@ static int afs_writepages_region(struct address_space *mapping,
>  #ifdef CONFIG_AFS_FSCACHE
>                                 folio_wait_fscache(folio);
>  #endif
> +                       } else {
> +                               start += folio_size(folio);
>                         }
>                         folio_put(folio);
> +                       if (wbc->sync_mode == WB_SYNC_NONE) {
> +                               if (skips >= 5 || need_resched())
> +                                       break;
> +                               skips++;
> +                       }
>                         continue;
>                 }

Looks good in testing, the problem would show up in xfstests generic/285.

Tested-by: Marc Dionne <marc.dionne at auristor.com>
Acked-by: Marc Dionne <marc.dionne at auristor.com>

Marc