LU-25: Use LDLM_ASYNC with ldlm_cancel_lru to avoid blocking.

Have ldlm_cli_pool_shrink() call ldlm_cancel_lru() with the LDLM_ASYNC
option to have the blocking callback handled in another thread.  This
will avoid randomly blocking for excessively long times when the shrinker
is called.

Change-Id: Ide705df2640a2a96be554ffc452930a2e180fc20
Signed-off-by: Christopher J. Morrone <morrone2@llnl.gov>
Reviewed-on: http://review.whamcloud.com/277
Tested-by: Hudson
Reviewed-by: Oleg Drokin <green@whamcloud.com>

diff --git a/lustre/ldlm/ldlm_pool.c b/lustre/ldlm/ldlm_pool.c
index 6d3cf0d..1219b39 100644 (file)
--- a/lustre/ldlm/ldlm_pool.c
+++ b/lustre/ldlm/ldlm_pool.c
@@ -551,7 +551,7 @@ static int ldlm_cli_pool_shrink(struct ldlm_pool *pl,
         cfs_spin_unlock(&ns->ns_lock);
         
         if (nr) {
-                canceled = ldlm_cancel_lru(ns, nr, LDLM_SYNC, 
+                canceled = ldlm_cancel_lru(ns, nr, LDLM_ASYNC,
                                            LDLM_CANCEL_SHRINK);
         }
 #ifdef __KERNEL__