b=23174 Lustre inode size is not coherent across nodes.

 Update lvbo from disk when AST fails with EINVAL.

 i=vitaly.fertman
 i=andrew.perepechko

diff --git a/lustre/ldlm/ldlm_lockd.c b/lustre/ldlm/ldlm_lockd.c
index 81075ee..9000c5b 100644 (file)
--- a/lustre/ldlm/ldlm_lockd.c
+++ b/lustre/ldlm/ldlm_lockd.c
@@ -606,19 +606,29 @@ static int ldlm_handle_ast_error(struct ldlm_lock *lock,
                         ldlm_failed_ast(lock, rc, ast_type);
                 }
         } else if (rc) {
                         ldlm_failed_ast(lock, rc, ast_type);
                 }
         } else if (rc) {

-                if (rc == -EINVAL)
+                if (rc == -EINVAL) {
+                        struct ldlm_resource *res = lock->l_resource;

                         LDLM_DEBUG(lock, "client (nid %s) returned %d"
                                " from %s AST - normal race",
                                libcfs_nid2str(peer.nid),
                                req->rq_repmsg ?
                                lustre_msg_get_status(req->rq_repmsg) : -1,
                                ast_type);
                         LDLM_DEBUG(lock, "client (nid %s) returned %d"
                                " from %s AST - normal race",
                                libcfs_nid2str(peer.nid),
                                req->rq_repmsg ?
                                lustre_msg_get_status(req->rq_repmsg) : -1,
                                ast_type);

-                else
+                        if (res) {
+                                /* update lvbo to return proper attributes.
+                                 * see bug 23174 */
+                                ldlm_resource_getref(res);
+                                ldlm_res_lvbo_update(res, NULL, 1);
+                                ldlm_resource_putref(res);
+                        }
+
+                } else {

                         LDLM_ERROR(lock, "client (nid %s) returned %d "
                                    "from %s AST", libcfs_nid2str(peer.nid),
                                    (req->rq_repmsg != NULL) ?
                                    lustre_msg_get_status(req->rq_repmsg) : 0,
                                    ast_type);
                         LDLM_ERROR(lock, "client (nid %s) returned %d "
                                    "from %s AST", libcfs_nid2str(peer.nid),
                                    (req->rq_repmsg != NULL) ?
                                    lustre_msg_get_status(req->rq_repmsg) : 0,
                                    ast_type);

+                }

                 ldlm_lock_cancel(lock);
                 /* Server-side AST functions are called from ldlm_reprocess_all,
                  * which needs to be told to please restart its reprocessing. */
                 ldlm_lock_cancel(lock);
                 /* Server-side AST functions are called from ldlm_reprocess_all,
                  * which needs to be told to please restart its reprocessing. */


@@ -641,11 +651,6 @@ static int ldlm_cb_interpret(const struct lu_env *env,
         lock = req->rq_async_args.pointer_arg[1];
         LASSERT(lock != NULL);
         if (rc != 0) {
         lock = req->rq_async_args.pointer_arg[1];
         LASSERT(lock != NULL);
         if (rc != 0) {

-                /* If client canceled the lock but the cancel has not
-                 * been received yet, we need to update lvbo to have the
-                 * proper attributes cached. */
-                if (rc == -EINVAL && arg->type == LDLM_BL_CALLBACK)
-                        ldlm_res_lvbo_update(lock->l_resource, NULL, 1);

                 rc = ldlm_handle_ast_error(lock, req, rc,
                                            arg->type == LDLM_BL_CALLBACK
                                            ? "blocking" : "completion");
                 rc = ldlm_handle_ast_error(lock, req, rc,
                                            arg->type == LDLM_BL_CALLBACK
                                            ? "blocking" : "completion");

diff --git a/lustre/obdfilter/filter.c b/lustre/obdfilter/filter.c
index 3570a30..6f55d0e 100644 (file)
--- a/lustre/obdfilter/filter.c
+++ b/lustre/obdfilter/filter.c
@@ -1797,13 +1797,6 @@ static int filter_intent_policy(struct ldlm_namespace *ns,
 
         LASSERTF(l->l_glimpse_ast != NULL, "l == %p", l);
         rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */
 
         LASSERTF(l->l_glimpse_ast != NULL, "l == %p", l);
         rc = l->l_glimpse_ast(l, NULL); /* this will update the LVB */

-        /* Update the LVB from disk if the AST failed (this is a legal race) */
-        /*
-         * XXX nikita: situation when ldlm_server_glimpse_ast() failed before
-         * sending ast is not handled. This can result in lost client writes.
-         */
-        if (rc != 0)
-                ldlm_res_lvbo_update(res, NULL, 1);

 
         lock_res(res);
         *reply_lvb = *res_lvb;
 
         lock_res(res);
         *reply_lvb = *res_lvb;