Whamcloud - gitweb
LU-243 async lov_sync() operation
[fs/lustre-release.git] / lustre / ost / ost_handler.c
index 945ad73..e039671 100644 (file)
@@ -26,7 +26,7 @@
  * GPL HEADER END
  */
 /*
- * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  */
 /*
@@ -72,7 +72,7 @@ CFS_MODULE_PARM(oss_num_create_threads, "i", int, 0444,
 /**
  * Do not return server-side uid/gid to remote client
  */
-static void ost_drop_id(struct obd_export *exp, struct  obdo *oa)
+static void ost_drop_id(struct obd_export *exp, struct obdo *oa)
 {
         if (exp_connect_rmtclient(exp)) {
                 oa->o_uid = -1;
@@ -249,8 +249,9 @@ static void ost_lock_put(struct obd_export *exp,
 static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
 {
         struct ost_body *body, *repbody;
-        struct obd_info oinfo = { { { 0 } } };
+        struct obd_info *oinfo;
         struct lustre_handle lh = { 0 };
+        struct lustre_capa *capa = NULL;
         int rc;
         ENTRY;
 
@@ -266,24 +267,31 @@ static int ost_getattr(struct obd_export *exp, struct ptlrpc_request *req)
         if (rc)
                 RETURN(rc);
 
-        repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
-        repbody->oa = body->oa;
-
         rc = ost_lock_get(exp, &body->oa, 0, OBD_OBJECT_EOF, &lh, LCK_PR, 0);
         if (rc)
                 RETURN(rc);
 
-        oinfo.oi_oa = &repbody->oa;
-        if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) {
-                oinfo.oi_capa = req_capsule_client_get(&req->rq_pill,
-                                                       &RMF_CAPA1);
-                if (oinfo.oi_capa == NULL) {
+        if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
+                capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
+                if (capa == NULL) {
                         CERROR("Missing capability for OST GETATTR");
                         RETURN (-EFAULT);
                 }
         }
 
-        req->rq_status = obd_getattr(exp, &oinfo);
+        OBD_ALLOC_PTR(oinfo);
+        if (!oinfo)
+                RETURN(-ENOMEM);
+        oinfo->oi_oa = &body->oa;
+        oinfo->oi_capa = capa;
+
+        req->rq_status = obd_getattr(exp, oinfo);
+
+        OBD_FREE_PTR(oinfo);
+
+        repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
+        repbody->oa = body->oa;
+
         ost_lock_put(exp, &lh, LCK_PR);
 
         ost_drop_id(exp, &repbody->oa);
@@ -303,7 +311,8 @@ static int ost_statfs(struct ptlrpc_request *req)
         osfs = req_capsule_server_get(&req->rq_pill, &RMF_OBD_STATFS);
 
         req->rq_status = obd_statfs(req->rq_export->exp_obd, osfs,
-                                    cfs_time_current_64() - CFS_HZ, 0);
+                                    cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
+                                    0);
         if (req->rq_status != 0)
                 CERROR("ost: statfs failed: rc %d\n", req->rq_status);
 
@@ -330,8 +339,8 @@ static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
                 RETURN(rc);
 
         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
-        memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
-        oti->oti_logcookies = &repbody->oa.o_lcookie;
+        repbody->oa = body->oa;
+        oti->oti_logcookies = &body->oa.o_lcookie;
 
         req->rq_status = obd_create(exp, &repbody->oa, NULL, oti);
         //obd_log_cancel(conn, NULL, 1, oti->oti_logcookies, 0);
@@ -341,7 +350,6 @@ static int ost_create(struct obd_export *exp, struct ptlrpc_request *req,
 static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
                      struct obd_trans_info *oti)
 {
-        struct obd_info oinfo = { { { 0 } } };
         struct ost_body *body, *repbody;
         int rc, flags = 0;
         struct lustre_handle lh = {0,};
@@ -358,11 +366,7 @@ static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
         if (rc)
                 RETURN(rc);
 
-        oinfo.oi_oa = &body->oa;
-        oinfo.oi_policy.l_extent.start = oinfo.oi_oa->o_size;
-        oinfo.oi_policy.l_extent.end = oinfo.oi_oa->o_blocks;
-
-        if ((oinfo.oi_oa->o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
+        if ((body->oa.o_valid & (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS)) !=
             (OBD_MD_FLSIZE | OBD_MD_FLBLOCKS))
                 RETURN(-EPROTO);
 
@@ -372,34 +376,48 @@ static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
 
         /* standard truncate optimization: if file body is completely
          * destroyed, don't send data back to the server. */
-        if (oinfo.oi_oa->o_size == 0)
+        if (body->oa.o_size == 0)
                 flags |= LDLM_AST_DISCARD_DATA;
 
-        repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
-        rc = ost_lock_get(exp, oinfo.oi_oa, oinfo.oi_oa->o_size,
-                          oinfo.oi_oa->o_blocks, &lh, LCK_PW, flags);
+        rc = ost_lock_get(exp, &body->oa, body->oa.o_size, body->oa.o_blocks,
+                          &lh, LCK_PW, flags);
         if (rc == 0) {
-                if (oinfo.oi_oa->o_valid & OBD_MD_FLFLAGS &&
-                    oinfo.oi_oa->o_flags == OBD_FL_SRVLOCK)
+                struct obd_info *oinfo;
+                struct lustre_capa *capa = NULL;
+
+                if (body->oa.o_valid & OBD_MD_FLFLAGS &&
+                    body->oa.o_flags == OBD_FL_SRVLOCK)
                         /*
                          * If OBD_FL_SRVLOCK is the only bit set in
                          * ->o_flags, clear OBD_MD_FLFLAGS to avoid falling
                          * through filter_setattr() to filter_iocontrol().
                          */
-                        oinfo.oi_oa->o_valid &= ~OBD_MD_FLFLAGS;
+                        body->oa.o_valid &= ~OBD_MD_FLFLAGS;
 
-                if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) {
-                        oinfo.oi_capa = req_capsule_client_get(&req->rq_pill,
-                                                               &RMF_CAPA1);
-                        if (oinfo.oi_capa == NULL) {
+                if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
+                        capa = req_capsule_client_get(&req->rq_pill,
+                                                      &RMF_CAPA1);
+                        if (capa == NULL) {
                                 CERROR("Missing capability for OST PUNCH");
                                 RETURN (-EFAULT);
                         }
                 }
-                req->rq_status = obd_punch(exp, &oinfo, oti, NULL);
+
+                OBD_ALLOC_PTR(oinfo);
+                if (!oinfo)
+                        RETURN(-ENOMEM);
+                oinfo->oi_oa = &body->oa;
+                oinfo->oi_policy.l_extent.start = oinfo->oi_oa->o_size;
+                oinfo->oi_policy.l_extent.end = oinfo->oi_oa->o_blocks;
+                oinfo->oi_capa = capa;
+
+                req->rq_status = obd_punch(exp, oinfo, oti, NULL);
+                OBD_FREE_PTR(oinfo);
                 ost_lock_put(exp, &lh, LCK_PW);
         }
-        repbody->oa = *oinfo.oi_oa;
+
+        repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
+        repbody->oa = body->oa;
         ost_drop_id(exp, &repbody->oa);
         RETURN(rc);
 }
@@ -407,6 +425,7 @@ static int ost_punch(struct obd_export *exp, struct ptlrpc_request *req,
 static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req)
 {
         struct ost_body *body, *repbody;
+        struct obd_info *oinfo;
         struct lustre_capa *capa = NULL;
         int rc;
         ENTRY;
@@ -431,10 +450,18 @@ static int ost_sync(struct obd_export *exp, struct ptlrpc_request *req)
         if (rc)
                 RETURN(rc);
 
+        OBD_ALLOC_PTR(oinfo);
+        if (!oinfo)
+                RETURN(-ENOMEM);
+
+        oinfo->oi_oa = &body->oa;
+        oinfo->oi_capa = capa;
+        req->rq_status = obd_sync(exp, oinfo, body->oa.o_size,
+                                  body->oa.o_blocks, NULL);
+        OBD_FREE_PTR(oinfo);
+
         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
-        memcpy(&repbody->oa, &body->oa, sizeof(body->oa));
-        req->rq_status = obd_sync(exp, &repbody->oa, NULL, repbody->oa.o_size,
-                                  repbody->oa.o_blocks, capa);
+        repbody->oa = body->oa;
         ost_drop_id(exp, &repbody->oa);
         RETURN(0);
 }
@@ -443,8 +470,9 @@ static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
                        struct obd_trans_info *oti)
 {
         struct ost_body *body, *repbody;
+        struct obd_info *oinfo;
+        struct lustre_capa *capa = NULL;
         int rc;
-        struct obd_info oinfo = { { { 0 } } };
         ENTRY;
 
         body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
@@ -459,19 +487,26 @@ static int ost_setattr(struct obd_export *exp, struct ptlrpc_request *req,
         if (rc)
                 RETURN(rc);
 
-        repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
-        repbody->oa = body->oa;
-
-        oinfo.oi_oa = &repbody->oa;
-        if (oinfo.oi_oa->o_valid & OBD_MD_FLOSSCAPA) {
-                oinfo.oi_capa = req_capsule_client_get(&req->rq_pill,
-                                                       &RMF_CAPA1);
-                if (oinfo.oi_capa == NULL) {
+        if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
+                capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
+                if (capa == NULL) {
                         CERROR("Missing capability for OST SETATTR");
                         RETURN (-EFAULT);
                 }
         }
-        req->rq_status = obd_setattr(exp, &oinfo, oti);
+
+        OBD_ALLOC_PTR(oinfo);
+        if (!oinfo)
+                RETURN(-ENOMEM);
+        oinfo->oi_oa = &body->oa;
+        oinfo->oi_capa = capa;
+
+        req->rq_status = obd_setattr(exp, oinfo, oti);
+
+        OBD_FREE_PTR(oinfo);
+
+        repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
+        repbody->oa = body->oa;
         ost_drop_id(exp, &repbody->oa);
         RETURN(0);
 }
@@ -822,7 +857,7 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
         desc = ptlrpc_prep_bulk_exp(req, npages,
                                      BULK_PUT_SOURCE, OST_BULK_PORTAL);
         if (desc == NULL)
-                GOTO(out_lock, rc = -ENOMEM);
+                GOTO(out_commitrw, rc = -ENOMEM);
 
         if (!lustre_handle_is_used(&lockh))
                 /* no needs to try to prolong lock if server is asked
@@ -945,6 +980,7 @@ static int ost_brw_read(struct ptlrpc_request *req, struct obd_trans_info *oti)
                 no_reply = rc != 0;
         }
 
+out_commitrw:
         /* Must commit after prep above in all cases */
         rc = obd_commitrw(OBD_BRW_READ, exp, &body->oa, 1, ioo,
                           remote_nb, npages, local_nb, oti, rc);
@@ -1004,7 +1040,7 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
         int rc, i, j;
         obd_count                client_cksum = 0, server_cksum = 0;
         cksum_type_t             cksum_type = OBD_CKSUM_CRC32;
-        int                      no_reply = 0;
+        int                      no_reply = 0, mmap = 0;
         __u32                    o_uid = 0, o_gid = 0;
         struct ost_thread_local_cache *tls;
         ENTRY;
@@ -1036,10 +1072,6 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
         if (body == NULL)
                 GOTO(out, rc = -EFAULT);
 
-        if ((body->oa.o_flags & OBD_BRW_MEMALLOC) &&
-            (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
-                libcfs_memory_pressure_set();
-
         objcount = req_capsule_get_size(&req->rq_pill, &RMF_OBD_IOOBJ,
                                         RCL_CLIENT) / sizeof(*ioo);
         ioo = req_capsule_client_get(&req->rq_pill, &RMF_OBD_IOOBJ);
@@ -1063,6 +1095,10 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
             &RMF_NIOBUF_REMOTE, RCL_CLIENT) / sizeof(*remote_nb)))
                 GOTO(out, rc = -EFAULT);
 
+        if ((remote_nb[0].flags & OBD_BRW_MEMALLOC) &&
+            (exp->exp_connection->c_peer.nid == exp->exp_connection->c_self))
+                cfs_memory_pressure_set();
+
         if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
                 capa = req_capsule_client_get(&req->rq_pill, &RMF_CAPA1);
                 if (capa == NULL) {
@@ -1114,6 +1150,8 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
                 if (body->oa.o_valid & OBD_MD_FLFLAGS)
                         cksum_type = cksum_type_unpack(body->oa.o_flags);
         }
+        if (body->oa.o_valid & OBD_MD_FLFLAGS && body->oa.o_flags & OBD_FL_MMAP)
+                mmap = 1;
 
         /* Because we already sync grant info with client when reconnect,
          * grant info will be cleared for resent req, then fed_grant and
@@ -1136,7 +1174,7 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
         desc = ptlrpc_prep_bulk_exp(req, npages,
                                      BULK_GET_SINK, OST_BULK_PORTAL);
         if (desc == NULL)
-                GOTO(out_lock, rc = -ENOMEM);
+                GOTO(skip_transfer, rc = -ENOMEM);
 
         /* NB Having prepped, we must commit... */
 
@@ -1203,6 +1241,7 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
         }
         no_reply = rc != 0;
 
+skip_transfer:
         repbody = req_capsule_server_get(&req->rq_pill, &RMF_OST_BODY);
         memcpy(&repbody->oa, &body->oa, sizeof(repbody->oa));
 
@@ -1215,8 +1254,9 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
                 repbody->oa.o_cksum = server_cksum;
                 cksum_counter++;
                 if (unlikely(client_cksum != server_cksum)) {
-                        CERROR("client csum %x, server csum %x\n",
-                               client_cksum, server_cksum);
+                        CDEBUG_LIMIT(mmap ? D_INFO : D_ERROR,
+                                     "client csum %x, server csum %x\n",
+                                     client_cksum, server_cksum);
                         cksum_counter = 0;
                 } else if ((cksum_counter & (-cksum_counter)) == cksum_counter){
                         CDEBUG(D_INFO, "Checksum %u from %s OK: %x\n",
@@ -1246,7 +1286,7 @@ static int ost_brw_write(struct ptlrpc_request *req, struct obd_trans_info *oti)
          */
         repbody->oa.o_valid &= ~(OBD_MD_FLMTIME | OBD_MD_FLATIME);
 
-        if (unlikely(client_cksum != server_cksum && rc == 0)) {
+        if (unlikely(client_cksum != server_cksum && rc == 0 &&  !mmap)) {
                 int  new_cksum = ost_checksum_bulk(desc, OST_WRITE, cksum_type);
                 char *msg;
                 char *via;
@@ -1319,14 +1359,6 @@ out_bulk:
         if (desc)
                 ptlrpc_free_bulk(desc);
 out:
-       /* XXX: don't send reply if obd rdonly mode, this can cause data loss
-        * on client, see bug 22190. Remove this when async bulk will be done.
-        * Meanwhile, if this is umount then don't reply anything. */
-        if (req->rq_export->exp_obd->obd_no_transno) {
-                no_reply = req->rq_export->exp_obd->obd_stopping;
-                rc = -EIO;
-        }
-
         if (rc == 0) {
                 oti_to_request(oti, req);
                 target_committed_to_req(req);
@@ -1346,7 +1378,7 @@ out:
                       exp->exp_connection->c_remote_uuid.uuid,
                       libcfs_id2str(req->rq_peer));
         }
-        libcfs_memory_pressure_clr();
+        cfs_memory_pressure_clr();
         RETURN(rc);
 }
 
@@ -1532,7 +1564,7 @@ static int ost_handle_quota_adjust_qunit(struct ptlrpc_request *req)
                 GOTO(out, rc);
 
         repoqa = req_capsule_server_get(&req->rq_pill, &RMF_QUOTA_ADJUST_QUNIT);
-        req->rq_status = obd_quota_adjust_qunit(req->rq_export, oqaq, qctxt);
+        req->rq_status = obd_quota_adjust_qunit(req->rq_export, oqaq, qctxt, NULL);
         *repoqa = *oqaq;
 
  out:
@@ -1727,6 +1759,56 @@ static int ost_connect_check_sptlrpc(struct ptlrpc_request *req)
         return rc;
 }
 
+/* Ensure that data and metadata are synced to the disk when lock is cancelled
+ * (if requested) */
+int ost_blocking_ast(struct ldlm_lock *lock,
+                             struct ldlm_lock_desc *desc,
+                             void *data, int flag)
+{
+        __u32 sync_lock_cancel = 0;
+        __u32 len = sizeof(sync_lock_cancel);
+        int rc = 0;
+        ENTRY;
+
+        rc = obd_get_info(lock->l_export, sizeof(KEY_SYNC_LOCK_CANCEL),
+                          KEY_SYNC_LOCK_CANCEL, &len, &sync_lock_cancel, NULL);
+
+        if (!rc && flag == LDLM_CB_CANCELING &&
+            (lock->l_granted_mode & (LCK_PW|LCK_GROUP)) &&
+            (sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
+             (sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL &&
+              lock->l_flags & LDLM_FL_CBPENDING))) {
+                struct obd_info *oinfo;
+                struct obdo *oa;
+                int rc;
+
+                OBD_ALLOC_PTR(oinfo);
+                if (!oinfo)
+                        RETURN(-ENOMEM);
+                OBDO_ALLOC(oa);
+                if (!oa) {
+                        OBD_FREE_PTR(oa);
+                        RETURN(-ENOMEM);
+                }
+                oa->o_id = lock->l_resource->lr_name.name[0];
+                oa->o_seq = lock->l_resource->lr_name.name[1];
+                oa->o_valid = OBD_MD_FLID|OBD_MD_FLGROUP;
+                oinfo->oi_oa = oa;
+
+                rc = obd_sync(lock->l_export, oinfo,
+                              lock->l_policy_data.l_extent.start,
+                              lock->l_policy_data.l_extent.end, NULL);
+                if (rc)
+                        CERROR("Error %d syncing data on lock cancel\n", rc);
+
+                OBDO_FREE(oa);
+                OBD_FREE_PTR(oinfo);
+        }
+
+        rc = ldlm_server_blocking_ast(lock, desc, data, flag);
+        RETURN(rc);
+}
+
 static int ost_filter_recovery_request(struct ptlrpc_request *req,
                                        struct obd_device *obd, int *process)
 {
@@ -2362,7 +2444,7 @@ int ost_handle(struct ptlrpc_request *req)
                 if (OBD_FAIL_CHECK(OBD_FAIL_LDLM_ENQUEUE))
                         RETURN(0);
                 rc = ldlm_handle_enqueue(req, ldlm_server_completion_ast,
-                                         ldlm_server_blocking_ast,
+                                         ost_blocking_ast,
                                          ldlm_server_glimpse_ast);
                 fail = OBD_FAIL_OST_LDLM_REPLY_NET;
                 break;
@@ -2508,7 +2590,7 @@ static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
                 GOTO(out_lprocfs, rc = -ENOMEM);
         }
 
-        rc = ptlrpc_start_threads(obd, ost->ost_service);
+        rc = ptlrpc_start_threads(ost->ost_service);
         if (rc)
                 GOTO(out_service, rc = -EINVAL);
 
@@ -2537,7 +2619,7 @@ static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
                 GOTO(out_service, rc = -ENOMEM);
         }
 
-        rc = ptlrpc_start_threads(obd, ost->ost_create_service);
+        rc = ptlrpc_start_threads(ost->ost_create_service);
         if (rc)
                 GOTO(out_create, rc = -EINVAL);
 
@@ -2557,7 +2639,7 @@ static int ost_setup(struct obd_device *obd, struct lustre_cfg* lcfg)
         ost->ost_io_service->srv_init = ost_thread_init;
         ost->ost_io_service->srv_done = ost_thread_done;
         ost->ost_io_service->srv_cpu_affinity = 1;
-        rc = ptlrpc_start_threads(obd, ost->ost_io_service);
+        rc = ptlrpc_start_threads(ost->ost_io_service);
         if (rc)
                 GOTO(out_io, rc = -EINVAL);
 
@@ -2587,13 +2669,9 @@ static int ost_cleanup(struct obd_device *obd)
 
         ping_evictor_stop();
 
-        cfs_spin_lock_bh(&obd->obd_processing_task_lock);
-        if (obd->obd_recovering) {
-                target_cancel_recovery_timer(obd);
-                obd->obd_recovering = 0;
-        }
-        cfs_spin_unlock_bh(&obd->obd_processing_task_lock);
-
+        /* there is no recovery for OST OBD, all recovery is controlled by
+         * obdfilter OBD */
+        LASSERT(obd->obd_recovering == 0);
         cfs_down(&ost->ost_health_sem);
         ptlrpc_unregister_service(ost->ost_service);
         ptlrpc_unregister_service(ost->ost_create_service);