Whamcloud - gitweb
LU-5319 tests: testcases for multiple modify RPCs feature
[fs/lustre-release.git] / lustre / target / tgt_handler.c
index bab3ff2..6b32fca 100644 (file)
@@ -21,7 +21,7 @@
  * GPL HEADER END
  */
 /*
- * Copyright (c) 2011, 2012, Intel Corporation.
+ * Copyright (c) 2011, 2014, Intel Corporation.
  */
 /*
  * lustre/target/tgt_handler.c
 
 #define DEBUG_SUBSYSTEM S_CLASS
 
+#include <linux/user_namespace.h>
+#ifdef HAVE_UIDGID_HEADER
+# include <linux/uidgid.h>
+#endif
+
 #include <obd.h>
 #include <obd_class.h>
 #include <obd_cksum.h>
 #include <md_object.h>
 #include <lustre_lfsck.h>
+#include <lustre_nodemap.h>
 
 #include "tgt_internal.h"
 
@@ -96,12 +102,7 @@ static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
        if (!IS_ERR(obj)) {
                if ((flags & HABEO_CORPUS) && !lu_object_exists(obj)) {
                        lu_object_put(tsi->tsi_env, obj);
-                       /* for capability renew ENOENT will be handled in
-                        * mdt_renew_capa */
-                       if (body->mbo_valid & OBD_MD_FLOSSCAPA)
-                               rc = 0;
-                       else
-                               rc = -ENOENT;
+                       rc = -ENOENT;
                } else {
                        tsi->tsi_corpus = obj;
                        rc = 0;
@@ -133,8 +134,8 @@ static int tgt_mdt_body_unpack(struct tgt_session_info *tsi, __u32 flags)
 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa)
 {
        struct ost_id   *oi     = &oa->o_oi;
-       obd_seq          seq    = ostid_seq(oi);
-       obd_id           id     = ostid_id(oi);
+       u64              seq    = ostid_seq(oi);
+       u64              id     = ostid_id(oi);
        int              rc;
        ENTRY;
 
@@ -241,7 +242,7 @@ static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
 {
        struct ost_body         *body;
        struct req_capsule      *pill = tsi->tsi_pill;
-       struct lustre_capa      *capa;
+       struct lu_nodemap       *nodemap;
        int                      rc;
 
        ENTRY;
@@ -254,14 +255,14 @@ static int tgt_ost_body_unpack(struct tgt_session_info *tsi, __u32 flags)
        if (rc)
                RETURN(rc);
 
-       if (body->oa.o_valid & OBD_MD_FLOSSCAPA) {
-               capa = req_capsule_client_get(pill, &RMF_CAPA1);
-               if (capa == NULL) {
-                       CERROR("%s: OSSCAPA flag is set without capability\n",
-                              tgt_name(tsi->tsi_tgt));
-                       RETURN(-EFAULT);
-               }
-       }
+       nodemap = tsi->tsi_exp->exp_target_data.ted_nodemap;
+
+       body->oa.o_uid = nodemap_map_id(nodemap, NODEMAP_UID,
+                                       NODEMAP_CLIENT_TO_FS,
+                                       body->oa.o_uid);
+       body->oa.o_gid = nodemap_map_id(nodemap, NODEMAP_GID,
+                                       NODEMAP_CLIENT_TO_FS,
+                                       body->oa.o_gid);
 
        tsi->tsi_ost_body = body;
        tsi->tsi_fid = body->oa.o_oi.oi_fid;
@@ -373,9 +374,24 @@ static int tgt_handle_request0(struct tgt_session_info *tsi,
 {
        int      serious = 0;
        int      rc;
+       __u32    opc = lustre_msg_get_opc(req->rq_reqmsg);
 
        ENTRY;
 
+
+       /* When dealing with sec context requests, no export is associated yet,
+        * because these requests are sent before *_CONNECT requests.
+        * A NULL req->rq_export means the normal *_common_slice handlers will
+        * not be called, because there is no reference to the target.
+        * So deal with them by hand and jump directly to target_send_reply().
+        */
+       switch (opc) {
+       case SEC_CTX_INIT:
+       case SEC_CTX_INIT_CONT:
+       case SEC_CTX_FINI:
+               GOTO(out, rc = 0);
+       }
+
        /*
         * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
         * to put same checks into handlers like mdt_close(), mdt_reint(),
@@ -389,6 +405,9 @@ static int tgt_handle_request0(struct tgt_session_info *tsi,
         */
        if (OBD_FAIL_CHECK_ORSET(h->th_fail_id, OBD_FAIL_ONCE))
                RETURN(0);
+       if (unlikely(lustre_msg_get_opc(req->rq_reqmsg) == MDS_REINT &&
+                    OBD_FAIL_CHECK(OBD_FAIL_MDS_REINT_MULTI_NET)))
+               RETURN(0);
 
        rc = tgt_request_preprocess(tsi, h, req);
        /* pack reply if reply format is fixed */
@@ -443,6 +462,7 @@ static int tgt_handle_request0(struct tgt_session_info *tsi,
        if (likely(rc == 0 && req->rq_export))
                target_committed_to_req(req);
 
+out:
        target_send_reply(req, rc, tsi->tsi_reply_fail_id);
        RETURN(0);
 }
@@ -457,13 +477,13 @@ static int tgt_filter_recovery_request(struct ptlrpc_request *req,
                *process = 1;
                RETURN(0);
        case MDS_CLOSE:
-       case MDS_DONE_WRITING:
        case MDS_SYNC: /* used in unmounting */
        case OBD_PING:
        case MDS_REINT:
        case OUT_UPDATE:
        case SEQ_QUERY:
        case FLD_QUERY:
+       case FLD_READ:
        case LDLM_ENQUEUE:
        case OST_CREATE:
        case OST_DESTROY:
@@ -471,6 +491,9 @@ static int tgt_filter_recovery_request(struct ptlrpc_request *req,
        case OST_SETATTR:
        case OST_SYNC:
        case OST_WRITE:
+       case MDS_HSM_PROGRESS:
+       case MDS_HSM_STATE_SET:
+       case MDS_HSM_REQUEST:
                *process = target_queue_recovery_request(req, obd);
                RETURN(0);
 
@@ -487,7 +510,7 @@ static int tgt_filter_recovery_request(struct ptlrpc_request *req,
  *       -ve: abort immediately with the given error code;
  *         0: send reply with error code in req->rq_status;
  */
-int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
+static int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
 {
        ENTRY;
 
@@ -506,11 +529,11 @@ int tgt_handle_recovery(struct ptlrpc_request *req, int reply_fail_id)
 
        /* sanity check: if the xid matches, the request must be marked as a
         * resent or replayed */
-       if (req_xid_is_last(req)) {
+       if (req_can_reconstruct(req, NULL)) {
                if (!(lustre_msg_get_flags(req->rq_reqmsg) &
                      (MSG_RESENT | MSG_REPLAY))) {
                        DEBUG_REQ(D_WARNING, req, "rq_xid "LPU64" matches "
-                                 "last_xid, expected REPLAY or RESENT flag "
+                                 "saved xid, expected REPLAY or RESENT flag "
                                  "(%x)", req->rq_xid,
                                  lustre_msg_get_flags(req->rq_reqmsg));
                        req->rq_status = -ENOTCONN;
@@ -553,6 +576,11 @@ static struct tgt_handler *tgt_handler_find_check(struct ptlrpc_request *req)
        ENTRY;
 
        tgt = class_exp2tgt(req->rq_export);
+       if (unlikely(tgt == NULL)) {
+               DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
+                         class_exp2obd(req->rq_export)->obd_name);
+               RETURN(ERR_PTR(-EINVAL));
+       }
 
        for (s = tgt->lut_slice; s->tos_hs != NULL; s++)
                if (s->tos_opc_start <= opc && opc < s->tos_opc_end)
@@ -612,6 +640,14 @@ int tgt_request_handle(struct ptlrpc_request *req)
        }
 
        if (unlikely(!class_connected_export(req->rq_export))) {
+               if (opc == SEC_CTX_INIT || opc == SEC_CTX_INIT_CONT ||
+                   opc == SEC_CTX_FINI) {
+                       /* sec context initialization has to be handled
+                        * by hand in tgt_handle_request0() */
+                       tsi->tsi_reply_fail_id = OBD_FAIL_SEC_CTX_INIT_NET;
+                       h = NULL;
+                       GOTO(handle_recov, rc = 0);
+               }
                CDEBUG(D_HA, "operation %d on unconnected OST from %s\n",
                       opc, libcfs_id2str(req->rq_peer));
                req->rq_status = -ENOTCONN;
@@ -626,9 +662,46 @@ int tgt_request_handle(struct ptlrpc_request *req)
        else
                tsi->tsi_jobid = NULL;
 
+       if (tgt == NULL) {
+               DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
+                         class_exp2obd(req->rq_export)->obd_name);
+               req->rq_status = -EINVAL;
+               rc = ptlrpc_error(req);
+               GOTO(out, rc);
+       }
+
+       /* check request's xid is consistent with export's last_xid */
+       if (req->rq_export != NULL) {
+               __u64 last_xid = lustre_msg_get_last_xid(req->rq_reqmsg);
+               if (last_xid != 0)
+                       req->rq_export->exp_last_xid = last_xid;
+               if (req->rq_xid == 0 ||
+                   req->rq_xid <= req->rq_export->exp_last_xid) {
+                       DEBUG_REQ(D_ERROR, req,
+                                 "Unexpected xid %llx vs. last_xid %llx\n",
+                                 req->rq_xid, req->rq_export->exp_last_xid);
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 7, 93, 0)
+                       LBUG();
+#endif
+                       req->rq_status = -EPROTO;
+                       rc = ptlrpc_error(req);
+                       GOTO(out, rc);
+               }
+       }
+
        request_fail_id = tgt->lut_request_fail_id;
        tsi->tsi_reply_fail_id = tgt->lut_reply_fail_id;
 
+       /* try to release in-memory reply data */
+       if (tgt_is_multimodrpcs_client(req->rq_export)) {
+               tgt_handle_received_xid(req->rq_export,
+                               lustre_msg_get_last_xid(req->rq_reqmsg));
+               if (!(lustre_msg_get_flags(req->rq_reqmsg) &
+                     (MSG_RESENT | MSG_REPLAY)))
+                       tgt_handle_tag(req->rq_export,
+                                      lustre_msg_get_tag(req->rq_reqmsg));
+       }
+
        h = tgt_handler_find_check(req);
        if (IS_ERR(h)) {
                req->rq_status = PTR_ERR(h);
@@ -636,6 +709,9 @@ int tgt_request_handle(struct ptlrpc_request *req)
                GOTO(out, rc);
        }
 
+       LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
+                h->th_opc, opc);
+
        if (CFS_FAIL_CHECK_ORSET(request_fail_id, CFS_FAIL_ONCE))
                GOTO(out, rc = 0);
 
@@ -649,10 +725,9 @@ int tgt_request_handle(struct ptlrpc_request *req)
                GOTO(out, rc);
        }
 
+handle_recov:
        rc = tgt_handle_recovery(req, tsi->tsi_reply_fail_id);
        if (likely(rc == 1)) {
-               LASSERTF(h->th_opc == opc, "opcode mismatch %d != %d\n",
-                        h->th_opc, opc);
                rc = tgt_handle_request0(tsi, h, req);
                if (rc)
                        GOTO(out, rc);
@@ -720,19 +795,16 @@ EXPORT_SYMBOL(tgt_counter_incr);
 static inline void tgt_init_sec_none(struct obd_connect_data *reply)
 {
        reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
-                                     OBD_CONNECT_RMT_CLIENT_FORCE |
-                                     OBD_CONNECT_MDS_CAPA |
-                                     OBD_CONNECT_OSS_CAPA);
+                                     OBD_CONNECT_RMT_CLIENT_FORCE);
 }
 
 static int tgt_init_sec_level(struct ptlrpc_request *req)
 {
        struct lu_target        *tgt = class_exp2tgt(req->rq_export);
-       char                    *client = libcfs_nid2str(req->rq_peer.nid);
+       char                    *client;
        struct obd_connect_data *data, *reply;
        int                      rc = 0;
        bool                     remote;
-
        ENTRY;
 
        data = req_capsule_client_get(&req->rq_pill, &RMF_CONNECT_DATA);
@@ -746,6 +818,13 @@ static int tgt_init_sec_level(struct ptlrpc_request *req)
                RETURN(0);
        }
 
+       if (unlikely(tgt == NULL)) {
+               DEBUG_REQ(D_ERROR, req, "%s: No target for connected export\n",
+                         class_exp2obd(req->rq_export)->obd_name);
+               RETURN(-EINVAL);
+       }
+
+       client = libcfs_nid2str(req->rq_peer.nid);
        /* no GSS support case */
        if (!req->rq_auth_gss) {
                if (tgt->lut_sec_level > LUSTRE_SEC_NONE) {
@@ -788,15 +867,7 @@ static int tgt_init_sec_level(struct ptlrpc_request *req)
                       "as remote by default.\n", client, tgt_name(tgt));
        }
 
-       if (remote) {
-               if (!tgt->lut_oss_capa) {
-                       CDEBUG(D_SEC,
-                              "client %s -> target %s is set as remote,"
-                              " but OSS capabilities are not enabled: %d.\n",
-                              client, tgt_name(tgt), tgt->lut_oss_capa);
-                       RETURN(-EACCES);
-               }
-       } else {
+       if (remote == 0) {
                if (!uid_valid(make_kuid(&init_user_ns, req->rq_auth_uid))) {
                        CDEBUG(D_SEC, "client %s -> target %s: user is not "
                               "authenticated!\n", client, tgt_name(tgt));
@@ -825,10 +896,8 @@ static int tgt_init_sec_level(struct ptlrpc_request *req)
                        break;
                reply->ocd_connect_flags &= ~(OBD_CONNECT_RMT_CLIENT |
                                              OBD_CONNECT_RMT_CLIENT_FORCE);
-               if (!tgt->lut_oss_capa)
-                       reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
-               if (!tgt->lut_mds_capa)
-                       reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
+               reply->ocd_connect_flags &= ~OBD_CONNECT_OSS_CAPA;
+               reply->ocd_connect_flags &= ~OBD_CONNECT_MDS_CAPA;
                break;
        default:
                RETURN(-EINVAL);
@@ -865,6 +934,16 @@ int tgt_connect_check_sptlrpc(struct ptlrpc_request *req, struct obd_export *exp
                spin_lock(&exp->exp_lock);
                exp->exp_sp_peer = req->rq_sp_from;
                exp->exp_flvr = flvr;
+
+               /* when on mgs, if no restriction is set, or if client
+                * is loopback, allow any flavor */
+               if ((strcmp(exp->exp_obd->obd_type->typ_name,
+                          LUSTRE_MGS_NAME) == 0) &&
+                    (exp->exp_flvr.sf_rpc == SPTLRPC_FLVR_NULL ||
+                     LNET_NETTYP(LNET_NIDNET(exp->exp_connection->c_peer.nid))
+                     == LOLND))
+                       exp->exp_flvr.sf_rpc = SPTLRPC_FLVR_ANY;
+
                if (exp->exp_flvr.sf_rpc != SPTLRPC_FLVR_ANY &&
                    exp->exp_flvr.sf_rpc != req->rq_flvr.sf_rpc) {
                        CERROR("%s: unauthorized rpc flavor %x from %s, "
@@ -895,6 +974,11 @@ int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial)
        struct sptlrpc_rule_set  tmp_rset;
        int                      rc;
 
+       if (unlikely(tgt == NULL)) {
+               CERROR("No target passed");
+               return -EINVAL;
+       }
+
        sptlrpc_rule_set_init(&tmp_rset);
        rc = sptlrpc_conf_target_get_rules(tgt->lut_obd, &tmp_rset, initial);
        if (rc) {
@@ -986,13 +1070,6 @@ int tgt_obd_log_cancel(struct tgt_session_info *tsi)
 {
        return err_serious(-EOPNOTSUPP);
 }
-EXPORT_SYMBOL(tgt_obd_log_cancel);
-
-int tgt_obd_qc_callback(struct tgt_session_info *tsi)
-{
-       return err_serious(-EOPNOTSUPP);
-}
-EXPORT_SYMBOL(tgt_obd_qc_callback);
 
 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
 {
@@ -1008,8 +1085,11 @@ int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
 
        ENTRY;
 
-       desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1, BULK_PUT_SOURCE,
-                                   MDS_BULK_PORTAL);
+       desc = ptlrpc_prep_bulk_exp(req, rdpg->rp_npages, 1,
+                                   PTLRPC_BULK_PUT_SOURCE |
+                                       PTLRPC_BULK_BUF_KIOV,
+                                   MDS_BULK_PORTAL,
+                                   &ptlrpc_bulk_kiov_pin_ops);
        if (desc == NULL)
                RETURN(-ENOMEM);
 
@@ -1021,12 +1101,13 @@ int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob)
        for (i = 0, tmpcount = nob; i < rdpg->rp_npages && tmpcount > 0;
             i++, tmpcount -= tmpsize) {
                tmpsize = min_t(int, tmpcount, PAGE_CACHE_SIZE);
-               ptlrpc_prep_bulk_page_pin(desc, rdpg->rp_pages[i], 0, tmpsize);
+               desc->bd_frag_ops->add_kiov_frag(desc, rdpg->rp_pages[i], 0,
+                                                tmpsize);
        }
 
        LASSERT(desc->bd_nob == nob);
        rc = target_bulk_io(exp, desc, lwi);
-       ptlrpc_free_bulk_pin(desc);
+       ptlrpc_free_bulk(desc);
        RETURN(rc);
 }
 EXPORT_SYMBOL(tgt_sendpage);
@@ -1034,7 +1115,7 @@ EXPORT_SYMBOL(tgt_sendpage);
 /*
  * OBD_IDX_READ handler
  */
-int tgt_obd_idx_read(struct tgt_session_info *tsi)
+static int tgt_obd_idx_read(struct tgt_session_info *tsi)
 {
        struct tgt_thread_info  *tti = tgt_th_info(tsi->tsi_env);
        struct lu_rdpg          *rdpg = &tti->tti_u.rdpg.tti_rdpg;
@@ -1115,12 +1196,10 @@ out:
        }
        return rc;
 }
-EXPORT_SYMBOL(tgt_obd_idx_read);
 
 struct tgt_handler tgt_obd_handlers[] = {
 TGT_OBD_HDL    (0,     OBD_PING,               tgt_obd_ping),
 TGT_OBD_HDL_VAR(0,     OBD_LOG_CANCEL,         tgt_obd_log_cancel),
-TGT_OBD_HDL_VAR(0,     OBD_QC_CALLBACK,        tgt_obd_qc_callback),
 TGT_OBD_HDL    (0,     OBD_IDX_READ,           tgt_obd_idx_read)
 };
 EXPORT_SYMBOL(tgt_obd_handlers);
@@ -1149,8 +1228,8 @@ EXPORT_SYMBOL(tgt_sync);
 
 /* Ensure that data and metadata are synced to the disk when lock is cancelled
  * (if requested) */
-int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
-                    void *data, int flag)
+static int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
+                           void *data, int flag)
 {
        struct lu_env            env;
        struct lu_target        *tgt;
@@ -1162,6 +1241,12 @@ int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
 
        tgt = class_exp2tgt(lock->l_export);
 
+       if (unlikely(tgt == NULL)) {
+               CDEBUG(D_ERROR, "%s: No target for connected export\n",
+                      class_exp2obd(lock->l_export)->obd_name);
+               RETURN(-EINVAL);
+       }
+
        if (flag == LDLM_CB_CANCELING &&
            (lock->l_granted_mode & (LCK_PW | LCK_GROUP)) &&
            (tgt->lut_sync_lock_cancel == ALWAYS_SYNC_ON_CANCEL ||
@@ -1206,7 +1291,7 @@ err_env:
        RETURN(rc);
 }
 
-struct ldlm_callback_suite tgt_dlm_cbs = {
+static struct ldlm_callback_suite tgt_dlm_cbs = {
        .lcs_completion = ldlm_server_completion_ast,
        .lcs_blocking   = tgt_blocking_ast,
        .lcs_glimpse    = ldlm_server_glimpse_ast
@@ -1228,6 +1313,20 @@ int tgt_enqueue(struct tgt_session_info *tsi)
        if (rc)
                RETURN(err_serious(rc));
 
+       switch (LUT_FAIL_CLASS(tsi->tsi_reply_fail_id)) {
+       case LUT_FAIL_MDT:
+               tsi->tsi_reply_fail_id = OBD_FAIL_MDS_LDLM_REPLY_NET;
+               break;
+       case LUT_FAIL_OST:
+               tsi->tsi_reply_fail_id = OBD_FAIL_OST_LDLM_REPLY_NET;
+               break;
+       case LUT_FAIL_MGT:
+               tsi->tsi_reply_fail_id = OBD_FAIL_MGS_LDLM_REPLY_NET;
+               break;
+       default:
+               tsi->tsi_reply_fail_id = OBD_FAIL_LDLM_REPLY;
+               break;
+       }
        RETURN(req->rq_status);
 }
 EXPORT_SYMBOL(tgt_enqueue);
@@ -1245,19 +1344,16 @@ int tgt_convert(struct tgt_session_info *tsi)
 
        RETURN(req->rq_status);
 }
-EXPORT_SYMBOL(tgt_convert);
 
 int tgt_bl_callback(struct tgt_session_info *tsi)
 {
        return err_serious(-EOPNOTSUPP);
 }
-EXPORT_SYMBOL(tgt_bl_callback);
 
 int tgt_cp_callback(struct tgt_session_info *tsi)
 {
        return err_serious(-EOPNOTSUPP);
 }
-EXPORT_SYMBOL(tgt_cp_callback);
 
 /* generic LDLM target handler */
 struct tgt_handler tgt_dlm_handlers[] = {
@@ -1306,7 +1402,6 @@ int tgt_llog_destroy(struct tgt_session_info *tsi)
 
        RETURN(rc);
 }
-EXPORT_SYMBOL(tgt_llog_destroy);
 
 int tgt_llog_read_header(struct tgt_session_info *tsi)
 {
@@ -1355,29 +1450,15 @@ TGT_LLOG_HDL_VAR(0,     LLOG_ORIGIN_HANDLE_CLOSE,       tgt_llog_close),
 };
 EXPORT_SYMBOL(tgt_llog_handlers);
 
-/*
- * sec context handlers
- */
-/* XXX: Implement based on mdt_sec_ctx_handle()? */
-int tgt_sec_ctx_handle(struct tgt_session_info *tsi)
-{
-       return 0;
-}
-
-struct tgt_handler tgt_sec_ctx_handlers[] = {
-TGT_SEC_HDL_VAR(0,     SEC_CTX_INIT,           tgt_sec_ctx_handle),
-TGT_SEC_HDL_VAR(0,     SEC_CTX_INIT_CONT,      tgt_sec_ctx_handle),
-TGT_SEC_HDL_VAR(0,     SEC_CTX_FINI,           tgt_sec_ctx_handle),
-};
-EXPORT_SYMBOL(tgt_sec_ctx_handlers);
-
 int (*tgt_lfsck_in_notify)(const struct lu_env *env,
                           struct dt_device *key,
-                          struct lfsck_request *lr) = NULL;
+                          struct lfsck_request *lr,
+                          struct thandle *th) = NULL;
 
 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
                                                struct dt_device *,
-                                               struct lfsck_request *))
+                                               struct lfsck_request *,
+                                               struct thandle *))
 {
        tgt_lfsck_in_notify = notify;
 }
@@ -1408,7 +1489,7 @@ static int tgt_handle_lfsck_notify(struct tgt_session_info *tsi)
        if (lr == NULL)
                RETURN(-EPROTO);
 
-       rc = tgt_lfsck_in_notify(env, key, lr);
+       rc = tgt_lfsck_in_notify(env, key, lr, NULL);
 
        RETURN(rc);
 }
@@ -1417,7 +1498,7 @@ static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
 {
        struct lfsck_request    *request;
        struct lfsck_reply      *reply;
-       int                      rc      = 0;
+       int                      rc;
        ENTRY;
 
        request = req_capsule_client_get(tsi->tsi_pill, &RMF_LFSCK_REQUEST);
@@ -1428,12 +1509,10 @@ static int tgt_handle_lfsck_query(struct tgt_session_info *tsi)
        if (reply == NULL)
                RETURN(-ENOMEM);
 
-       reply->lr_status = tgt_lfsck_query(tsi->tsi_env,
-                                          tsi->tsi_tgt->lut_bottom, request);
-       if (reply->lr_status < 0)
-               rc = reply->lr_status;
+       rc = tgt_lfsck_query(tsi->tsi_env, tsi->tsi_tgt->lut_bottom, request);
+       reply->lr_status = rc;
 
-       RETURN(rc);
+       RETURN(rc < 0 ? rc : 0);
 }
 
 struct tgt_handler tgt_lfsck_handlers[] = {
@@ -1503,7 +1582,7 @@ int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
        LASSERT(!lustre_handle_is_used(lh));
 
        policy.l_extent.gid = 0;
-       policy.l_extent.start = start & CFS_PAGE_MASK;
+       policy.l_extent.start = start & PAGE_MASK;
 
        /*
         * If ->o_blocks is EOF it means "lock till the end of the file".
@@ -1512,7 +1591,7 @@ int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
        if (end == OBD_OBJECT_EOF || end < start)
                policy.l_extent.end = OBD_OBJECT_EOF;
        else
-               policy.l_extent.end = end | ~CFS_PAGE_MASK;
+               policy.l_extent.end = end | ~PAGE_MASK;
 
        rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_EXTENT, &policy, mode,
                                    flags, ldlm_blocking_ast,
@@ -1554,7 +1633,6 @@ int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
                               nb[nrbufs - 1].rnb_len - 1,
                               lh, mode, &flags));
 }
-EXPORT_SYMBOL(tgt_brw_lock);
 
 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
                    struct lustre_handle *lh, int mode)
@@ -1570,7 +1648,6 @@ void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
                tgt_extent_unlock(lh, mode);
        EXIT;
 }
-EXPORT_SYMBOL(tgt_brw_unlock);
 
 static __u32 tgt_checksum_bulk(struct lu_target *tgt,
                               struct ptlrpc_bulk_desc *desc, int opc,
@@ -1582,6 +1659,8 @@ static __u32 tgt_checksum_bulk(struct lu_target *tgt,
        unsigned char                   cfs_alg = cksum_obd2cfs(cksum_type);
        __u32                           cksum;
 
+       LASSERT(ptlrpc_is_bulk_desc_kiov(desc->bd_type));
+
        hdesc = cfs_crypto_hash_init(cfs_alg, NULL, 0);
        if (IS_ERR(hdesc)) {
                CERROR("%s: unable to initialize checksum hash %s\n",
@@ -1595,10 +1674,11 @@ static __u32 tgt_checksum_bulk(struct lu_target *tgt,
                 * simulate a client->OST data error */
                if (i == 0 && opc == OST_WRITE &&
                    OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_RECEIVE)) {
-                       int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
-                       int len = desc->bd_iov[i].kiov_len;
+                       int off = BD_GET_KIOV(desc, i).kiov_offset &
+                               ~PAGE_MASK;
+                       int len = BD_GET_KIOV(desc, i).kiov_len;
                        struct page *np = tgt_page_to_corrupt;
-                       char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
+                       char *ptr = kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
 
                        if (np) {
                                char *ptr2 = kmap(np) + off;
@@ -1606,24 +1686,28 @@ static __u32 tgt_checksum_bulk(struct lu_target *tgt,
                                memcpy(ptr2, ptr, len);
                                memcpy(ptr2, "bad3", min(4, len));
                                kunmap(np);
-                               desc->bd_iov[i].kiov_page = np;
+                               BD_GET_KIOV(desc, i).kiov_page = np;
                        } else {
                                CERROR("%s: can't alloc page for corruption\n",
                                       tgt_name(tgt));
                        }
                }
-               cfs_crypto_hash_update_page(hdesc, desc->bd_iov[i].kiov_page,
-                                 desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK,
-                                 desc->bd_iov[i].kiov_len);
+               cfs_crypto_hash_update_page(hdesc,
+                                 BD_GET_KIOV(desc, i).kiov_page,
+                                 BD_GET_KIOV(desc, i).kiov_offset &
+                                       ~PAGE_MASK,
+                                 BD_GET_KIOV(desc, i).kiov_len);
 
                 /* corrupt the data after we compute the checksum, to
                 * simulate an OST->client data error */
                if (i == 0 && opc == OST_READ &&
                    OBD_FAIL_CHECK(OBD_FAIL_OST_CHECKSUM_SEND)) {
-                       int off = desc->bd_iov[i].kiov_offset & ~CFS_PAGE_MASK;
-                       int len = desc->bd_iov[i].kiov_len;
+                       int off = BD_GET_KIOV(desc, i).kiov_offset
+                         & ~PAGE_MASK;
+                       int len = BD_GET_KIOV(desc, i).kiov_len;
                        struct page *np = tgt_page_to_corrupt;
-                       char *ptr = kmap(desc->bd_iov[i].kiov_page) + off;
+                       char *ptr =
+                         kmap(BD_GET_KIOV(desc, i).kiov_page) + off;
 
                        if (np) {
                                char *ptr2 = kmap(np) + off;
@@ -1631,7 +1715,7 @@ static __u32 tgt_checksum_bulk(struct lu_target *tgt,
                                memcpy(ptr2, ptr, len);
                                memcpy(ptr2, "bad4", min(4, len));
                                kunmap(np);
-                               desc->bd_iov[i].kiov_page = np;
+                               BD_GET_KIOV(desc, i).kiov_page = np;
                        } else {
                                CERROR("%s: can't alloc page for corruption\n",
                                       tgt_name(tgt));
@@ -1656,8 +1740,7 @@ int tgt_brw_read(struct tgt_session_info *tsi)
        struct ost_body         *body, *repbody;
        struct l_wait_info       lwi;
        struct lustre_handle     lockh = { 0 };
-       int                      niocount, npages, nob = 0, rc, i;
-       int                      no_reply = 0;
+       int                      npages, nob = 0, rc, i, no_reply = 0;
        struct tgt_thread_big_cache *tbc = req->rq_svc_thread->t_data;
 
        ENTRY;
@@ -1700,7 +1783,6 @@ int tgt_brw_read(struct tgt_session_info *tsi)
        ioo = req_capsule_client_get(tsi->tsi_pill, &RMF_OBD_IOOBJ);
        LASSERT(ioo != NULL); /* must exists after tgt_ost_body_unpack */
 
-       niocount = ioo->ioo_bufcnt;
        remote_nb = req_capsule_client_get(&req->rq_pill, &RMF_NIOBUF_REMOTE);
        LASSERT(remote_nb != NULL); /* must exists after tgt_ost_body_unpack */
 
@@ -1731,12 +1813,15 @@ int tgt_brw_read(struct tgt_session_info *tsi)
 
        npages = PTLRPC_MAX_BRW_PAGES;
        rc = obd_preprw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1,
-                       ioo, remote_nb, &npages, local_nb, NULL, BYPASS_CAPA);
+                       ioo, remote_nb, &npages, local_nb);
        if (rc != 0)
                GOTO(out_lock, rc);
 
        desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
-                                   BULK_PUT_SOURCE, OST_BULK_PORTAL);
+                                   PTLRPC_BULK_PUT_SOURCE |
+                                       PTLRPC_BULK_BUF_KIOV,
+                                   OST_BULK_PORTAL,
+                                   &ptlrpc_bulk_kiov_nopin_ops);
        if (desc == NULL)
                GOTO(out_commitrw, rc = -ENOMEM);
 
@@ -1752,9 +1837,10 @@ int tgt_brw_read(struct tgt_session_info *tsi)
                nob += page_rc;
                if (page_rc != 0) { /* some data! */
                        LASSERT(local_nb[i].lnb_page != NULL);
-                       ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
-                                                   local_nb[i].lnb_page_offset,
-                                                   page_rc);
+                       desc->bd_frag_ops->add_kiov_frag
+                         (desc, local_nb[i].lnb_page,
+                          local_nb[i].lnb_page_offset,
+                          page_rc);
                }
 
                if (page_rc != local_nb[i].lnb_len) { /* short read */
@@ -1790,16 +1876,15 @@ int tgt_brw_read(struct tgt_session_info *tsi)
 
 out_commitrw:
        /* Must commit after prep above in all cases */
-       rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp,
-                         &repbody->oa, 1, ioo, remote_nb, npages, local_nb,
-                         NULL, rc);
+       rc = obd_commitrw(tsi->tsi_env, OBD_BRW_READ, exp, &repbody->oa, 1, ioo,
+                         remote_nb, npages, local_nb, rc);
        if (rc == 0)
                tgt_drop_id(exp, &repbody->oa);
 out_lock:
        tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PR);
 
        if (desc && !CFS_FAIL_PRECHECK(OBD_FAIL_PTLRPC_CLIENT_BULK_CB2))
-               ptlrpc_free_bulk_nopin(desc);
+               ptlrpc_free_bulk(desc);
 
        LASSERT(rc <= 0);
        if (rc == 0) {
@@ -1827,7 +1912,7 @@ out_lock:
                lwi1 = LWI_TIMEOUT_INTR(cfs_time_seconds(3), NULL, NULL, NULL);
                l_wait_event(waitq, 0, &lwi1);
                target_bulk_io(exp, desc, &lwi);
-               ptlrpc_free_bulk_nopin(desc);
+               ptlrpc_free_bulk(desc);
        }
 
        RETURN(rc);
@@ -1837,20 +1922,18 @@ EXPORT_SYMBOL(tgt_brw_read);
 static void tgt_warn_on_cksum(struct ptlrpc_request *req,
                              struct ptlrpc_bulk_desc *desc,
                              struct niobuf_local *local_nb, int npages,
-                             obd_count client_cksum, obd_count server_cksum,
+                             u32 client_cksum, u32 server_cksum,
                              bool mmap)
 {
        struct obd_export *exp = req->rq_export;
        struct ost_body *body;
-       char *router;
-       char *via;
+       char *router = "";
+       char *via = "";
 
        body = req_capsule_client_get(&req->rq_pill, &RMF_OST_BODY);
        LASSERT(body != NULL);
 
-       if (req->rq_peer.nid == desc->bd_sender) {
-               via = router = "";
-       } else {
+       if (req->rq_peer.nid != desc->bd_sender) {
                via = " via ";
                router = libcfs_nid2str(desc->bd_sender);
        }
@@ -1999,21 +2082,23 @@ int tgt_brw_write(struct tgt_session_info *tsi)
 
        npages = PTLRPC_MAX_BRW_PAGES;
        rc = obd_preprw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
-                       objcount, ioo, remote_nb, &npages, local_nb, NULL,
-                       BYPASS_CAPA);
+                       objcount, ioo, remote_nb, &npages, local_nb);
        if (rc < 0)
                GOTO(out_lock, rc);
 
        desc = ptlrpc_prep_bulk_exp(req, npages, ioobj_max_brw_get(ioo),
-                                   BULK_GET_SINK, OST_BULK_PORTAL);
+                                   PTLRPC_BULK_GET_SINK | PTLRPC_BULK_BUF_KIOV,
+                                   OST_BULK_PORTAL,
+                                   &ptlrpc_bulk_kiov_nopin_ops);
        if (desc == NULL)
                GOTO(skip_transfer, rc = -ENOMEM);
 
        /* NB Having prepped, we must commit... */
        for (i = 0; i < npages; i++)
-               ptlrpc_prep_bulk_page_nopin(desc, local_nb[i].lnb_page,
-                                           local_nb[i].lnb_page_offset,
-                                           local_nb[i].lnb_len);
+               desc->bd_frag_ops->add_kiov_frag(desc,
+                                                local_nb[i].lnb_page,
+                                                local_nb[i].lnb_page_offset,
+                                                local_nb[i].lnb_len);
 
        rc = sptlrpc_svc_prep_bulk(req, desc);
        if (rc != 0)
@@ -2054,8 +2139,7 @@ skip_transfer:
 
        /* Must commit after prep above in all cases */
        rc = obd_commitrw(tsi->tsi_env, OBD_BRW_WRITE, exp, &repbody->oa,
-                         objcount, ioo, remote_nb, npages, local_nb, NULL,
-                         rc);
+                         objcount, ioo, remote_nb, npages, local_nb, rc);
        if (rc == -ENOTCONN)
                /* quota acquire process has been given up because
                 * either the client has been evicted or the client
@@ -2095,7 +2179,7 @@ skip_transfer:
 out_lock:
        tgt_brw_unlock(ioo, remote_nb, &lockh, LCK_PW);
        if (desc)
-               ptlrpc_free_bulk_nopin(desc);
+               ptlrpc_free_bulk(desc);
 out:
        if (no_reply) {
                req->rq_no_reply = 1;
@@ -2111,3 +2195,44 @@ out:
        RETURN(rc);
 }
 EXPORT_SYMBOL(tgt_brw_write);
+
+/* Check if request can be reconstructed from saved reply data
+ * A copy of the reply data is returned in @trd if the pointer is not NULL
+ */
+bool req_can_reconstruct(struct ptlrpc_request *req,
+                        struct tg_reply_data *trd)
+{
+       struct tg_export_data *ted = &req->rq_export->exp_target_data;
+       struct lsd_client_data *lcd = ted->ted_lcd;
+       bool found;
+
+       if (tgt_is_multimodrpcs_client(req->rq_export))
+               return tgt_lookup_reply(req, trd);
+
+       mutex_lock(&ted->ted_lcd_lock);
+       found = req->rq_xid == lcd->lcd_last_xid ||
+               req->rq_xid == lcd->lcd_last_close_xid;
+
+       if (found && trd != NULL) {
+               if (lustre_msg_get_opc(req->rq_reqmsg) == MDS_CLOSE) {
+                       trd->trd_reply.lrd_xid = lcd->lcd_last_close_xid;
+                       trd->trd_reply.lrd_transno =
+                                               lcd->lcd_last_close_transno;
+                       trd->trd_reply.lrd_result = lcd->lcd_last_close_result;
+               } else {
+                       trd->trd_reply.lrd_xid = lcd->lcd_last_xid;
+                       trd->trd_reply.lrd_transno = lcd->lcd_last_transno;
+                       trd->trd_reply.lrd_result = lcd->lcd_last_result;
+                       trd->trd_reply.lrd_data = lcd->lcd_last_data;
+                       trd->trd_pre_versions[0] = lcd->lcd_pre_versions[0];
+                       trd->trd_pre_versions[1] = lcd->lcd_pre_versions[1];
+                       trd->trd_pre_versions[2] = lcd->lcd_pre_versions[2];
+                       trd->trd_pre_versions[3] = lcd->lcd_pre_versions[3];
+               }
+       }
+       mutex_unlock(&ted->ted_lcd_lock);
+
+       return found;
+}
+EXPORT_SYMBOL(req_can_reconstruct);
+