X-Git-Url: https://git.whamcloud.com/?a=blobdiff_plain;f=lustre%2Fmdt%2Fmdt_io.c;h=5924aa197194374175376129eb59f6edd1c1cd54;hb=a54ecd2c2d964e2ae226fca5b043b5462a34eb7b;hp=929563b7178b1555104b4c5223453205c856f4dc;hpb=3c75d2522786a2a86db450a44da4e764d5ea8f30;p=fs%2Flustre-release.git

diff --git a/lustre/mdt/mdt_io.c b/lustre/mdt/mdt_io.c
index 929563b..5924aa1 100644
--- a/lustre/mdt/mdt_io.c
+++ b/lustre/mdt/mdt_io.c
@@ -63,7 +63,7 @@ static void mdt_dom_resource_prolong(struct ldlm_prolong_args *arg)
 	ENTRY;
 
 	res = ldlm_resource_get(arg->lpa_export->exp_obd->obd_namespace, NULL,
-				&arg->lpa_resid, LDLM_EXTENT, 0);
+				&arg->lpa_resid, LDLM_IBITS, 0);
 	if (IS_ERR(res)) {
 		CDEBUG(D_DLMTRACE,
 		       "Failed to get resource for resid %llu/%llu\n",
@@ -76,7 +76,11 @@ static void mdt_dom_resource_prolong(struct ldlm_prolong_args *arg)
 		if (ldlm_has_dom(lock)) {
 			LDLM_DEBUG(lock, "DOM lock to prolong ");
 			ldlm_lock_prolong_one(lock, arg);
-			break;
+			/* only one PW or EX lock can be granted,
+			 * no need to continue search
+			 */
+			if (lock->l_granted_mode & (LCK_PW | LCK_EX))
+				break;
 		}
 	}
 	unlock_res(res);
@@ -137,7 +141,7 @@ static int mdt_rw_hpreq_lock_match(struct ptlrpc_request *req,
 		RETURN(0);
 
 	/* a bulk write can only hold a reference on a PW extent lock. */
-	mode = LCK_PW;
+	mode = LCK_PW | LCK_GROUP;
 	if (opc == OST_READ)
 		/* whereas a bulk read can be protected by either a PR or PW
 		 * extent lock */
@@ -177,7 +181,7 @@ static int mdt_rw_hpreq_check(struct ptlrpc_request *req)
 	LASSERT(rnb != NULL);
 	LASSERT(!(rnb->rnb_flags & OBD_BRW_SRVLOCK));
 
-	pa.lpa_mode = LCK_PW;
+	pa.lpa_mode = LCK_PW | LCK_GROUP;
 	if (opc == OST_READ)
 		pa.lpa_mode |= LCK_PR;
 
@@ -265,7 +269,7 @@ static int mdt_punch_hpreq_lock_match(struct ptlrpc_request *req,
 	if (!fid_res_name_eq(&oa->o_oi.oi_fid, &lock->l_resource->lr_name))
 		RETURN(0);
 
-	if (!(lock->l_granted_mode & LCK_PW))
+	if (!(lock->l_granted_mode & (LCK_PW | LCK_GROUP)))
 		RETURN(0);
 
 	RETURN(1);
@@ -303,14 +307,13 @@ static int mdt_punch_hpreq_check(struct ptlrpc_request *req)
 	LASSERT(!(oa->o_valid & OBD_MD_FLFLAGS &&
 		  oa->o_flags & OBD_FL_SRVLOCK));
 
-	pa.lpa_mode = LCK_PW;
+	pa.lpa_mode = LCK_PW | LCK_GROUP;
 
 	CDEBUG(D_DLMTRACE, "%s: refresh DOM lock for "DFID"\n",
 	       tgt_name(tsi->tsi_tgt), PFID(&tsi->tsi_fid));
 
 	mdt_prolong_dom_lock(tsi, &pa);
 
-
 	if (pa.lpa_blocks_cnt > 0) {
 		CDEBUG(D_DLMTRACE,
 		       "%s: refreshed %u locks timeout for req %p.\n",
@@ -595,7 +598,7 @@ static int mdt_commitrw_write(const struct lu_env *env, struct obd_export *exp,
 	struct thandle *th;
 	int rc = 0;
 	int retries = 0;
-	int i;
+	int i, restart = 0;
 
 	ENTRY;
 
@@ -655,8 +658,10 @@ retry:
 
 	dt_write_lock(env, dob, 0);
 	rc = dt_write_commit(env, dob, lnb, niocount, th, oa->o_size);
-	if (rc)
+	if (rc) {
+		restart = th->th_restart_tran;
 		GOTO(unlock, rc);
+	}
 
 	if (la->la_valid) {
 		rc = dt_attr_set(env, dob, la, th);
@@ -680,13 +685,23 @@ out_stop:
 			granted = 0;
 	}
 
-	th->th_result = rc;
+	th->th_result = restart ? 0 : rc;
 	dt_trans_stop(env, dt, th);
 	if (rc == -ENOSPC && retries++ < 3) {
 		CDEBUG(D_INODE, "retry after force commit, retries:%d\n",
 		       retries);
 		goto retry;
 	}
+	if (restart) {
+		retries++;
+		restart = 0;
+		if (retries % 10000 == 0)
+			CERROR("%s: restart IO write too many times: %d\n",
+			       exp->exp_obd->obd_name, retries);
+		CDEBUG(D_INODE, "retry transaction, retries:%d\n",
+		       retries);
+		goto retry;
+	}
 
 out:
 	dt_bufs_put(env, dob, lnb, niocount);