[Linux-cachefs] [PATCH] FS-Cache: Fix handling of an attempt to store a page that is now beyond EOF

[Milosz Tanski]

Hey guys it looks like this patch got lost in the shuffle. I checked
out a recent kernel and learned this wasn't applied the hard way (by
running into it).

Can this please get applied to master as it fixes a real issue?

Thanks,
- Milosz

On Mon, Nov 11, 2013 at 5:18 PM, David Howells <dhowells at redhat.com> wrote:
> Fix the handling of an attempt to store a page that is now beyond EOF.  This
> may happen, for example, if the page got pushed for storage before the netfs
> file got truncated on the server.  In such a case, we should just remove the
> excessive pages from the cookie->stores radix tree and wake up the waiter.
>
> This can be seen in /proc/fs/fscache/stats on this line:
>
>         Stores : ops=350 run=1895 pgs=1545 rxd=1727 olm=9
>
> where olm=N has N > 0.
>
> Reported-by: Milosz Tanski <milosz at adfin.com>
> Signed-off-by: David Howells <dhowells at redhat.com>
> Acked-by: Milosz Tanski <milosz at adfin.com>
> ---
>
>  fs/fscache/page.c |   37 ++++++++++++++++++++++++++++++++++++-
>  1 file changed, 36 insertions(+), 1 deletion(-)
>
> diff --git a/fs/fscache/page.c b/fs/fscache/page.c
> index 7f5c658af755..b4730cf52aec 100644
> --- a/fs/fscache/page.c
> +++ b/fs/fscache/page.c
> @@ -755,6 +755,7 @@ static void fscache_write_op(struct fscache_operation *_op)
>         struct fscache_object *object = op->op.object;
>         struct fscache_cookie *cookie;
>         struct page *page;
> +       pgoff_t index;
>         unsigned n;
>         void *results[1];
>         int ret;
> @@ -803,7 +804,7 @@ static void fscache_write_op(struct fscache_operation *_op)
>         _debug("gang %d [%lx]", n, page->index);
>         if (page->index > op->store_limit) {
>                 fscache_stat(&fscache_n_store_pages_over_limit);
> -               goto superseded;
> +               goto page_beyond_limit;
>         }
>
>         radix_tree_tag_set(&cookie->stores, page->index,
> @@ -829,6 +830,40 @@ static void fscache_write_op(struct fscache_operation *_op)
>         _leave("");
>         return;
>
> +page_beyond_limit:
> +       spin_unlock(&object->lock);
> +
> +page_beyond_limit_unlocked:
> +       /* pages that are now beyond the end of the storage object must have
> +        * their pending storage records cleared.
> +        */
> +       index = page->index;
> +       radix_tree_tag_clear(&cookie->stores, page->index,
> +                            FSCACHE_COOKIE_PENDING_TAG);
> +       if (!radix_tree_tag_get(&cookie->stores, page->index,
> +                               FSCACHE_COOKIE_STORING_TAG)) {
> +               fscache_stat(&fscache_n_store_radix_deletes);
> +               radix_tree_delete(&cookie->stores, page->index);
> +               page_cache_release(page);
> +       }
> +       if (!need_resched()) {
> +               n = radix_tree_gang_lookup_tag(&cookie->stores, results,
> +                                              index + 1, 1,
> +                                              FSCACHE_COOKIE_PENDING_TAG);
> +               if (n == 1) {
> +                       page = results[0];
> +                       goto page_beyond_limit_unlocked;
> +               }
> +               spin_unlock(&cookie->stores_lock);
> +               wake_up_bit(&cookie->flags, 0);
> +       } else {
> +               spin_unlock(&cookie->stores_lock);
> +       }
> +
> +       fscache_enqueue_operation(&op->op);
> +       _leave("");
> +       return;
> +
>  superseded:
>         /* this writer is going away and there aren't any more things to
>          * write */
>



-- 
Milosz Tanski
CTO
10 East 53rd Street, 37th floor
New York, NY 10022

p: 646-253-9055
e: milosz at adfin.com