b=23127 osc lock granted fix

[fs/lustre-release.git] / lustre / osc / osc_request.c

diff --git a/lustre/osc/osc_request.c b/lustre/osc/osc_request.c
index 12edc7e..2ba487b 100644 (file)
--- a/lustre/osc/osc_request.c
+++ b/lustre/osc/osc_request.c
@@ -1087,11 +1087,21 @@ static void osc_init_grant(struct client_obd *cli, struct obd_connect_data *ocd)
                 cli->cl_avail_grant = ocd->ocd_grant;
         else
                 cli->cl_avail_grant = ocd->ocd_grant - cli->cl_dirty;
+
+        if (cli->cl_avail_grant < 0) {
+                CWARN("%s: available grant < 0, the OSS is probably not running"
+                      " with patch from bug20278 (%ld) \n",
+                      cli->cl_import->imp_obd->obd_name, cli->cl_avail_grant);
+                /* workaround for 1.6 servers which do not have 
+                 * the patch from bug20278 */
+                cli->cl_avail_grant = ocd->ocd_grant;
+        }
+
         client_obd_list_unlock(&cli->cl_loi_list_lock);
 
-        CDEBUG(D_CACHE, "setting cl_avail_grant: %ld cl_lost_grant: %ld \n",
+        CDEBUG(D_CACHE, "%s, setting cl_avail_grant: %ld cl_lost_grant: %ld \n",
+               cli->cl_import->imp_obd->obd_name,
                cli->cl_avail_grant, cli->cl_lost_grant);
-        LASSERT(cli->cl_avail_grant >= 0);
 
         if (ocd->ocd_connect_flags & OBD_CONNECT_GRANT_SHRINK &&
             cfs_list_empty(&cli->cl_grant_shrink_list))
@@ -1177,7 +1187,7 @@ static inline int can_merge_pages(struct brw_page *p1, struct brw_page *p2)
 {
         if (p1->flag != p2->flag) {
                 unsigned mask = ~(OBD_BRW_FROM_GRANT|
-                                  OBD_BRW_NOCACHE|OBD_BRW_SYNC);
+                                  OBD_BRW_NOCACHE|OBD_BRW_SYNC|OBD_BRW_ASYNC);
 
                 /* warn if we try to combine flags that we don't know to be
                  * safe to combine */
@@ -1427,6 +1437,10 @@ static int check_write_checksum(struct obdo *oa, const lnet_process_id_t *peer,
                 return 0;
         }
 
+        /* If this is mmaped file - it can be changed at any time */
+        if (oa->o_valid & OBD_MD_FLFLAGS && oa->o_flags & OBD_FL_MMAP)
+                return 1;
+
         if (oa->o_valid & OBD_MD_FLFLAGS)
                 cksum_type = cksum_type_unpack(oa->o_flags);
         else
@@ -2167,9 +2181,20 @@ static int brw_interpret(const struct lu_env *env,
         rc = osc_brw_fini_request(req, rc);
         CDEBUG(D_INODE, "request %p aa %p rc %d\n", req, aa, rc);
         if (osc_recoverable_error(rc)) {
-                rc = osc_brw_redo_request(req, aa);
-                if (rc == 0)
-                        RETURN(0);
+                /* Only retry once for mmaped files since the mmaped page
+                 * might be modified at anytime. We have to retry at least
+                 * once in case there WAS really a corruption of the page
+                 * on the network, that was not caused by mmap() modifying
+                 * the page. Bug11742 */
+                if ((rc == -EAGAIN) && (aa->aa_resends > 0) &&
+                    aa->aa_oa->o_valid & OBD_MD_FLFLAGS &&
+                    aa->aa_oa->o_flags & OBD_FL_MMAP) {
+                        rc = 0;
+                } else {
+                        rc = osc_brw_redo_request(req, aa);
+                        if (rc == 0)
+                                RETURN(0);
+                }
         }
 
         if (aa->aa_ocapa) {
@@ -2201,7 +2226,7 @@ static int brw_interpret(const struct lu_env *env,
                 }
                 OBDO_FREE(aa->aa_oa);
         } else { /* from async_internal() */
-                int i;
+                obd_count i;
                 for (i = 0; i < aa->aa_page_count; i++)
                         osc_release_write_grant(aa->aa_cli, aa->aa_ppga[i], 1);
 
@@ -2214,6 +2239,7 @@ static int brw_interpret(const struct lu_env *env,
         if (!async)
                 cl_req_completion(env, aa->aa_clerq, rc);
         osc_release_ppga(aa->aa_ppga, aa->aa_page_count);
+
         RETURN(rc);
 }
 
@@ -3201,6 +3227,9 @@ static int osc_enqueue_interpret(const struct lu_env *env,
          * osc_enqueue_fini(). */
         ldlm_lock_addref(&handle, mode);
 
+        /* Let CP AST to grant the lock first. */
+        OBD_FAIL_TIMEOUT(OBD_FAIL_OSC_CP_ENQ_RACE, 1);
+
         /* Complete obtaining the lock procedure. */
         rc = ldlm_cli_enqueue_fini(aa->oa_exp, req, aa->oa_ei->ei_type, 1,
                                    mode, aa->oa_flags, aa->oa_lvb,