Ignore zfs_arc_shrinker_limit in direct reclaim mode

zfs_arc_shrinker_limit (default: 10000) avoids ARC collapse
due to excessive memory reclaim. However, when the kernel is
in direct reclaim mode (ie: low on memory), limiting ARC reclaim
increases OOM risk. This is especially true on system without
(or with inadequate) swap.

This patch ignores zfs_arc_shrinker_limit when the kernel is in
direct reclaim mode, avoiding most OOM. It also restores
"echo 3 > /proc/sys/vm/drop_caches" ability to correctly drop
(almost) all ARC.

Signed-off-by: Gionatan Danti <g.danti@assyoma.it>

man/man4/zfs.4

@@ -821,6 +821,7 @@ This is a limit on how many pages the ARC shrinker makes available for
 eviction in response to one page allocation attempt.
 Note that in practice, the kernel's shrinker can ask us to evict
 up to about four times this for one allocation attempt.
+To reduce OOM risk, this limit is applied for kswapd reclaims only.
 .Pp
 The default limit of
 .Sy 10000 Pq in practice, Em 160 MiB No per allocation attempt with 4 KiB pages

module/os/linux/zfs/arc_os.c

@@ -201,9 +201,9 @@ arc_shrinker_count(struct shrinker *shrink, struct shrink_control *sc)
 	 * See also the comment above zfs_arc_shrinker_limit.
 	 */
 	int64_t can_free = btop(arc_evictable_memory());
-	int64_t limit = zfs_arc_shrinker_limit != 0 ?
-	    zfs_arc_shrinker_limit : INT64_MAX;
-	return (MIN(can_free, limit));
+	if (current_is_kswapd() && zfs_arc_shrinker_limit)
+		can_free = MIN(can_free, zfs_arc_shrinker_limit);
+	return (can_free);
 }
 
 static unsigned long