Whamcloud - gitweb
branch: HEAD
[fs/lustre-release.git] / lustre / mdt / mdt_handler.c
index d9437f7..6cc0c6c 100644 (file)
@@ -165,6 +165,8 @@ static struct mdt_opc_slice mdt_fld_handlers[];
 static struct mdt_device *mdt_dev(struct lu_device *d);
 static int mdt_regular_handle(struct ptlrpc_request *req);
 static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags);
+static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
+                        struct getinfo_fid2path *fp);
 
 static const struct lu_object_operations mdt_obj_ops;
 
@@ -349,7 +351,7 @@ static int mdt_statfs(struct mdt_thread_info *info)
         /* This will trigger a watchdog timeout */
         OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_STATFS_LCW_SLEEP,
                          (MDT_SERVICE_WATCHDOG_FACTOR *
-                          at_get(&svc->srv_at_estimate) / 1000) + 1);
+                          at_get(&svc->srv_at_estimate)) + 1);
 
         rc = mdt_check_ucred(info);
         if (rc)
@@ -521,6 +523,8 @@ static int mdt_getattr_internal(struct mdt_thread_info *info,
                         CERROR("readlink failed: %d\n", rc);
                         rc = -EFAULT;
                 } else {
+                        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_READLINK_EPROTO))
+                                 rc -= 2;
                         repbody->valid |= OBD_MD_LINKNAME;
                         repbody->eadatasize = rc;
                         /* NULL terminate */
@@ -953,6 +957,7 @@ static int mdt_getattr_name_lock(struct mdt_thread_info *info,
 relock:
                 ma = &info->mti_attr;
 
+                OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_RESEND, obd_timeout*2);
                 mdt_lock_handle_init(lhc);
                 mdt_lock_reg_init(lhc, LCK_PR);
 
@@ -998,7 +1003,7 @@ relock:
 
                         /* Debugging code. */
                         res_id = &lock->l_resource->lr_name;
-                        LDLM_DEBUG(lock, "Returning lock to client\n");
+                        LDLM_DEBUG(lock, "Returning lock to client");
                         LASSERTF(fid_res_name_eq(mdt_object_fid(child),
                                                  &lock->l_resource->lr_name),
                                  "Lock res_id: %lu/%lu/%lu, Fid: "DFID".\n",
@@ -1068,12 +1073,15 @@ static int lu_device_is_mdt(struct lu_device *d)
         return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
 }
 
+static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
+                         void *karg, void *uarg);
+
 static int mdt_set_info(struct mdt_thread_info *info)
 {
         struct ptlrpc_request *req = mdt_info_req(info);
         char *key;
-        __u32 *val;
-        int keylen, rc = 0;
+        void *val;
+        int keylen, vallen, rc = 0;
         ENTRY;
 
         rc = req_capsule_server_pack(info->mti_pill);
@@ -1095,19 +1103,40 @@ static int mdt_set_info(struct mdt_thread_info *info)
                 RETURN(-EFAULT);
         }
 
-        if (!KEY_IS(KEY_READ_ONLY))
-                RETURN(-EINVAL);
+        vallen = req_capsule_get_size(info->mti_pill, &RMF_SETINFO_VAL,
+                                      RCL_CLIENT);
 
-        req->rq_status = 0;
-        lustre_msg_set_status(req->rq_repmsg, 0);
+        /* Swab any part of val you need to here */
+        if (KEY_IS(KEY_READ_ONLY)) {
+                req->rq_status = 0;
+                lustre_msg_set_status(req->rq_repmsg, 0);
 
-        spin_lock(&req->rq_export->exp_lock);
-        if (*val)
-                req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
-        else
-                req->rq_export->exp_connect_flags &= ~OBD_CONNECT_RDONLY;
-        spin_unlock(&req->rq_export->exp_lock);
+                spin_lock(&req->rq_export->exp_lock);
+                if (*(__u32 *)val)
+                        req->rq_export->exp_connect_flags |= OBD_CONNECT_RDONLY;
+                else
+                        req->rq_export->exp_connect_flags &=~OBD_CONNECT_RDONLY;
+                spin_unlock(&req->rq_export->exp_lock);
+
+        } else if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
+                struct changelog_setinfo *cs =
+                        (struct changelog_setinfo *)val;
+                if (vallen != sizeof(*cs)) {
+                        CERROR("Bad changelog_clear setinfo size %d\n", vallen);
+                        RETURN(-EINVAL);
+                }
+                if (ptlrpc_req_need_swab(req)) {
+                        __swab64s(&cs->cs_recno);
+                        __swab32s(&cs->cs_id);
+                }
+
+                rc = mdt_iocontrol(OBD_IOC_CHANGELOG_CLEAR, info->mti_exp,
+                                   vallen, val, NULL);
+                lustre_msg_set_status(req->rq_repmsg, rc);
 
+        } else {
+                RETURN(-EINVAL);
+        }
         RETURN(0);
 }
 
@@ -1168,6 +1197,10 @@ static int mdt_sendpage(struct mdt_thread_info *info,
         }
 
         LASSERT(desc->bd_nob == rdpg->rp_count);
+        rc = sptlrpc_svc_wrap_bulk(req, desc);
+        if (rc)
+                GOTO(free_desc, rc);
+
         rc = ptlrpc_start_bulk_transfer(desc);
         if (rc)
                 GOTO(free_desc, rc);
@@ -1327,6 +1360,9 @@ static int mdt_writepage(struct mdt_thread_info *info)
         ptlrpc_prep_bulk_page(desc, page, (int)reqbody->size,
                               (int)reqbody->nlink);
 
+        rc = sptlrpc_svc_prep_bulk(req, desc);
+        if (rc != 0)
+                GOTO(cleanup_page, rc);
         /*
          * Check if client was evicted while we were doing i/o before touching
          * network.
@@ -1407,6 +1443,8 @@ static int mdt_readpage(struct mdt_thread_info *info)
                        rdpg->rp_hash, reqbody->size);
                 RETURN(-EFAULT);
         }
+
+        rdpg->rp_attrs = reqbody->mode;
         rdpg->rp_count  = reqbody->nlink;
         rdpg->rp_npages = (rdpg->rp_count + CFS_PAGE_SIZE - 1)>>CFS_PAGE_SHIFT;
         OBD_ALLOC(rdpg->rp_pages, rdpg->rp_npages * sizeof rdpg->rp_pages[0]);
@@ -1481,6 +1519,8 @@ static int mdt_reint_internal(struct mdt_thread_info *info,
                 GOTO(out_shrink, rc = err_serious(rc));
         }
 
+        OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_REINT_DELAY, 10);
+
         /* for replay no cookkie / lmm need, because client have this already */
         if (info->mti_spec.no_create == 1)  {
                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
@@ -1656,9 +1696,6 @@ static int mdt_quotacheck_handle(struct mdt_thread_info *info)
         int rc;
         ENTRY;
 
-        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_QUOTACHECK_NET))
-                RETURN(0);
-
         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
         if (oqctl == NULL)
                 RETURN(-EPROTO);
@@ -1686,9 +1723,6 @@ static int mdt_quotactl_handle(struct mdt_thread_info *info)
         int id, rc;
         ENTRY;
 
-        if (OBD_FAIL_CHECK(OBD_FAIL_MDS_QUOTACTL_NET))
-                RETURN(0);
-
         oqctl = req_capsule_client_get(pill, &RMF_OBD_QUOTACTL);
         if (oqctl == NULL)
                 RETURN(-EPROTO);
@@ -1729,10 +1763,10 @@ static int mdt_quotactl_handle(struct mdt_thread_info *info)
 
         switch (oqctl->qc_cmd) {
         case Q_QUOTAON:
-                rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type, id);
+                rc = mqo->mqo_on(info->mti_env, next, oqctl->qc_type);
                 break;
         case Q_QUOTAOFF:
-                rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type, id);
+                rc = mqo->mqo_off(info->mti_env, next, oqctl->qc_type);
                 break;
         case Q_SETINFO:
                 rc = mqo->mqo_setinfo(info->mti_env, next, oqctl->qc_type, id,
@@ -1775,6 +1809,7 @@ static int mdt_quotactl_handle(struct mdt_thread_info *info)
 }
 #endif
 
+
 /*
  * OBD PING and other handlers.
  */
@@ -1803,6 +1838,101 @@ static int mdt_obd_qc_callback(struct mdt_thread_info *info)
 
 
 /*
+ * LLOG handlers.
+ */
+
+/** clone llog ctxt from child (mdd)
+ * This allows remote llog (replicator) access.
+ * We can either pass all llog RPCs (eg mdt_llog_create) on to child where the
+ * context was originally set up, or we can handle them directly.
+ * I choose the latter, but that means I need any llog
+ * contexts set up by child to be accessable by the mdt.  So we clone the
+ * context into our context list here.
+ */
+static int mdt_llog_ctxt_clone(const struct lu_env *env, struct mdt_device *mdt,
+                               int idx)
+{
+        struct md_device  *next = mdt->mdt_child;
+        struct llog_ctxt *ctxt;
+        int rc;
+
+        if (!llog_ctxt_null(mdt2obd_dev(mdt), idx))
+                return 0;
+
+        rc = next->md_ops->mdo_llog_ctxt_get(env, next, idx, (void **)&ctxt);
+        if (rc || ctxt == NULL) {
+                CERROR("Can't get mdd ctxt %d\n", rc);
+                return rc;
+        }
+
+        rc = llog_group_set_ctxt(&mdt2obd_dev(mdt)->obd_olg, ctxt, idx);
+        if (rc)
+                CERROR("Can't set mdt ctxt %d\n", rc);
+
+        return rc;
+}
+
+static int mdt_llog_ctxt_unclone(const struct lu_env *env,
+                                 struct mdt_device *mdt, int idx)
+{
+        struct llog_ctxt *ctxt;
+
+        ctxt = llog_get_context(mdt2obd_dev(mdt), idx);
+        if (ctxt == NULL)
+                return 0;
+        /* Put once for the get we just did, and once for the clone */
+        llog_ctxt_put(ctxt);
+        llog_ctxt_put(ctxt);
+        return 0;
+}
+
+static int mdt_llog_create(struct mdt_thread_info *info)
+{
+        int rc;
+
+        req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_CREATE);
+        rc = llog_origin_handle_create(mdt_info_req(info));
+        return (rc < 0 ? err_serious(rc) : rc);
+}
+
+static int mdt_llog_destroy(struct mdt_thread_info *info)
+{
+        int rc;
+
+        req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_DESTROY);
+        rc = llog_origin_handle_destroy(mdt_info_req(info));
+        return (rc < 0 ? err_serious(rc) : rc);
+}
+
+static int mdt_llog_read_header(struct mdt_thread_info *info)
+{
+        int rc;
+
+        req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_READ_HEADER);
+        rc = llog_origin_handle_read_header(mdt_info_req(info));
+        return (rc < 0 ? err_serious(rc) : rc);
+}
+
+static int mdt_llog_next_block(struct mdt_thread_info *info)
+{
+        int rc;
+
+        req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_NEXT_BLOCK);
+        rc = llog_origin_handle_next_block(mdt_info_req(info));
+        return (rc < 0 ? err_serious(rc) : rc);
+}
+
+static int mdt_llog_prev_block(struct mdt_thread_info *info)
+{
+        int rc;
+
+        req_capsule_set(info->mti_pill, &RQF_LLOG_ORIGIN_HANDLE_PREV_BLOCK);
+        rc = llog_origin_handle_prev_block(mdt_info_req(info));
+        return (rc < 0 ? err_serious(rc) : rc);
+}
+
+
+/*
  * DLM handlers.
  */
 static struct ldlm_callback_suite cbs = {
@@ -2167,7 +2297,7 @@ void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
  *
  * \param info thread info object
  * \param o mdt object
- * \param h mdt lock handle referencing regular and PDO locks
+ * \param lh mdt lock handle referencing regular and PDO locks
  * \param decref force immediate lock releasing
  */
 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *o,
@@ -2222,7 +2352,9 @@ static struct mdt_handler *mdt_handler_find(__u32 opc,
                 if (s->mos_opc_start <= opc && opc < s->mos_opc_end) {
                         h = s->mos_hs + (opc - s->mos_opc_start);
                         if (likely(h->mh_opc != 0))
-                                LASSERT(h->mh_opc == opc);
+                                LASSERTF(h->mh_opc == opc,
+                                         "opcode mismatch %d != %d\n",
+                                         h->mh_opc, opc);
                         else
                                 h = NULL; /* unsupported opc */
                         break;
@@ -2326,6 +2458,7 @@ static int mdt_unpack_req_pack_rep(struct mdt_thread_info *info, __u32 flags)
                 struct mdt_device *mdt = info->mti_mdt;
 
                 /* Pack reply. */
+
                 if (req_capsule_has_field(pill, &RMF_MDT_MD, RCL_SERVER))
                         req_capsule_set_size(pill, &RMF_MDT_MD, RCL_SERVER,
                                              mdt->mdt_max_mdsize);
@@ -2370,7 +2503,7 @@ static int mdt_req_handle(struct mdt_thread_info *info,
          * Checking for various OBD_FAIL_$PREF_$OPC_NET codes. _Do_ not try
          * to put same checks into handlers like mdt_close(), mdt_reint(),
          * etc., without talking to mdt authors first. Checking same thing
-         * there again is useless and returning 0 error wihtout packing reply
+         * there again is useless and returning 0 error without packing reply
          * is buggy! Handlers either pack reply or return error.
          *
          * We return 0 here and do not send any reply in order to emulate
@@ -2458,7 +2591,7 @@ static int mdt_req_handle(struct mdt_thread_info *info,
         }
 
         /* If we're DISCONNECTing, the mdt_export_data is already freed */
-        if (likely(rc == 0 && h->mh_opc != MDS_DISCONNECT))
+        if (likely(rc == 0 && req->rq_export && h->mh_opc != MDS_DISCONNECT))
                 target_committed_to_req(req);
 
         if (unlikely((lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY) &&
@@ -2525,6 +2658,10 @@ static void mdt_thread_info_init(struct ptlrpc_request *req,
 
         info->mti_fail_id = OBD_FAIL_MDS_ALL_REPLY_NET;
         info->mti_transno = lustre_msg_get_transno(req->rq_reqmsg);
+        info->mti_mos[0] = NULL;
+        info->mti_mos[1] = NULL;
+        info->mti_mos[2] = NULL;
+        info->mti_mos[3] = NULL;
 
         memset(&info->mti_attr, 0, sizeof(info->mti_attr));
         info->mti_body = NULL;
@@ -2537,6 +2674,7 @@ static void mdt_thread_info_init(struct ptlrpc_request *req,
 
         /* To not check for split by default. */
         info->mti_spec.sp_ck_split = 0;
+        info->mti_spec.no_create = 0;
 }
 
 static void mdt_thread_info_fini(struct mdt_thread_info *info)
@@ -2545,6 +2683,11 @@ static void mdt_thread_info_fini(struct mdt_thread_info *info)
 
         req_capsule_fini(info->mti_pill);
         if (info->mti_object != NULL) {
+                /*
+                 * freeing an object may lead to OSD level transaction, do not
+                 * let it mess with MDT. bz19385.
+                 */
+                info->mti_no_need_trans = 1;
                 mdt_object_put(info->mti_env, info->mti_object);
                 info->mti_object = NULL;
         }
@@ -2678,6 +2821,7 @@ static int mdt_msg_check_version(struct lustre_msg *msg)
         switch (lustre_msg_get_opc(msg)) {
         case MDS_CONNECT:
         case MDS_DISCONNECT:
+        case MDS_SET_INFO:
         case OBD_PING:
         case SEC_CTX_INIT:
         case SEC_CTX_INIT_CONT:
@@ -2703,7 +2847,7 @@ static int mdt_msg_check_version(struct lustre_msg *msg)
         case MDS_SYNC:
         case MDS_GETXATTR:
         case MDS_SETXATTR:
-        case MDS_SET_INFO:
+        case MDS_GET_INFO:
         case MDS_QUOTACHECK:
         case MDS_QUOTACTL:
         case QUOTA_DQACQ:
@@ -2775,7 +2919,8 @@ static int mdt_handle0(struct ptlrpc_request *req,
                         if (likely(h != NULL)) {
                                 rc = mdt_req_handle(info, h, req);
                         } else {
-                                CERROR("The unsupported opc: 0x%x\n", lustre_msg_get_opc(msg) );
+                                CERROR("The unsupported opc: 0x%x\n",
+                                       lustre_msg_get_opc(msg) );
                                 req->rq_status = -ENOTSUPP;
                                 rc = ptlrpc_error(req);
                                 RETURN(rc);
@@ -3024,7 +3169,7 @@ int mdt_intent_lock_replace(struct mdt_thread_info *info,
                 new_lock->l_writers--;
         }
 
-        new_lock->l_export = class_export_get(req->rq_export);
+        new_lock->l_export = class_export_lock_get(req->rq_export);
         new_lock->l_blocking_ast = lock->l_blocking_ast;
         new_lock->l_completion_ast = lock->l_completion_ast;
         new_lock->l_remote_handle = lock->l_remote_handle;
@@ -3216,14 +3361,15 @@ static int mdt_intent_reint(enum mdt_it_code opcode,
         }
         rep->lock_policy_res2 = clear_serious(rc);
 
-        lhc->mlh_reg_lh.cookie = 0ull;
-        if (rc == -ENOTCONN || rc == -ENODEV) {
+        if (rc == -ENOTCONN || rc == -ENODEV ||
+            rc == -EOVERFLOW) { /**< if VBR failure then return error */
                 /*
                  * If it is the disconnect error (ENODEV & ENOCONN), the error
                  * will be returned by rq_status, and client at ptlrpc layer
                  * will detect this, then disconnect, reconnect the import
                  * immediately, instead of impacting the following the rpc.
                  */
+                lhc->mlh_reg_lh.cookie = 0ull;
                 RETURN(rc);
         } else {
                 /*
@@ -3234,7 +3380,14 @@ static int mdt_intent_reint(enum mdt_it_code opcode,
                   * FIXME: when open lock is finished, that should be
                   * checked here.
                   */
-                RETURN(ELDLM_LOCK_ABORTED);
+                if (lustre_handle_is_used(&lhc->mlh_reg_lh)) {
+                        rep->lock_policy_res2 = 0;
+                        rc = mdt_intent_lock_replace(info, lockp, NULL, lhc, flags);
+                        RETURN(rc);
+                } else {
+                        lhc->mlh_reg_lh.cookie = 0ull;
+                        RETURN(ELDLM_LOCK_ABORTED);
+                }
         }
 }
 
@@ -3373,7 +3526,7 @@ static void mdt_seq_adjust(const struct lu_env *env,
                           struct mdt_device *m, int lost)
 {
         struct md_site *ms = mdt_md_site(m);
-        struct lu_range out;
+        struct lu_seq_range out;
         ENTRY;
 
         LASSERT(ms && ms->ms_server_seq);
@@ -3439,6 +3592,7 @@ static int mdt_seq_init(const struct lu_env *env,
                 rc = seq_server_init(ms->ms_control_seq,
                                      m->mdt_bottom, uuid,
                                      LUSTRE_SEQ_CONTROLLER,
+                                     ms,
                                      env);
 
                 if (rc)
@@ -3480,6 +3634,7 @@ static int mdt_seq_init(const struct lu_env *env,
         rc = seq_server_init(ms->ms_server_seq,
                              m->mdt_bottom, uuid,
                              LUSTRE_SEQ_SERVER,
+                             ms,
                              env);
         if (rc)
                 GOTO(out_seq_fini, rc = -ENOMEM);
@@ -3634,7 +3789,8 @@ static int mdt_fld_init(const struct lu_env *env,
                 RETURN(rc = -ENOMEM);
 
         rc = fld_server_init(ms->ms_server_fld,
-                             m->mdt_bottom, uuid, env);
+                             m->mdt_bottom, uuid,
+                             env, ms->ms_node_id);
         if (rc) {
                 OBD_FREE_PTR(ms->ms_server_fld);
                 ms->ms_server_fld = NULL;
@@ -4111,7 +4267,7 @@ static int mdt_obd_llog_setup(struct obd_device *obd,
 
         obd->obd_fsops = fsfilt_get_ops(MT_STR(lsi->lsi_ldd));
         if (IS_ERR(obd->obd_fsops))
-                return (int) PTR_ERR(obd->obd_fsops);
+                return PTR_ERR(obd->obd_fsops);
 
         rc = fsfilt_setup(obd, lsi->lsi_srv_mnt->mnt_sb);
         if (rc) {
@@ -4127,7 +4283,7 @@ static int mdt_obd_llog_setup(struct obd_device *obd,
         rc = llog_setup(obd, &obd->obd_olg, LLOG_CONFIG_ORIG_CTXT, obd,
                         0, NULL, &llog_lvfs_ops);
         if (rc) {
-                CERROR("llog setup failed: %d\n", rc);
+                CERROR("llog_setup() failed: %d\n", rc);
                 fsfilt_put_ops(obd->obd_fsops);
         }
 
@@ -4142,77 +4298,74 @@ static void mdt_obd_llog_cleanup(struct obd_device *obd)
         if (ctxt)
                 llog_cleanup(ctxt);
 
-        if (obd->obd_fsops)
+        if (obd->obd_fsops) {
                 fsfilt_put_ops(obd->obd_fsops);
+                obd->obd_fsops = NULL;
+        }
 }
 
 static void mdt_fini(const struct lu_env *env, struct mdt_device *m)
 {
-        struct md_device *next = m->mdt_child;
-        struct lu_device *d    = &m->mdt_md_dev.md_lu_dev;
-        struct lu_site   *ls   = d->ld_site;
+        struct md_device  *next = m->mdt_child;
+        struct lu_device  *d    = &m->mdt_md_dev.md_lu_dev;
+        struct lu_site    *ls   = d->ld_site;
         struct obd_device *obd = mdt2obd_dev(m);
-        int             waited = 0;
         ENTRY;
 
-        /* At this point, obd exports might still be on the "obd_zombie_exports"
-         * list, and obd_zombie_impexp_thread() is trying to destroy them.
-         * We wait a little bit until all exports (except the self-export)
-         * have been destroyed, because the whole mdt stack might be accessed
-         * in mdt_destroy_export(). This will not be a long time, maybe one or
-         * two seconds are enough. This is not a problem while umounting.
-         *
-         * The three references that should be remaining are the
-         * obd_self_export and the attach and setup references.
-         */
-        while (atomic_read(&obd->obd_refcount) > 3) {
-                cfs_schedule_timeout(CFS_TASK_UNINT, cfs_time_seconds(1));
-                ++waited;
-                if (waited > 5 && IS_PO2(waited))
-                        LCONSOLE_WARN("Waiting for obd_zombie_impexp_thread "
-                                      "more than %d seconds to destroy all "
-                                      "the exports. The current obd refcount ="
-                                      " %d. Is it stuck there?\n",
-                                      waited, atomic_read(&obd->obd_refcount));
-        }
+        target_recovery_fini(obd);
 
         ping_evictor_stop();
 
-        target_recovery_fini(obd);
         mdt_stop_ptlrpc_service(m);
+        mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
         mdt_obd_llog_cleanup(obd);
+        obd_exports_barrier(obd);
         obd_zombie_barrier();
 #ifdef HAVE_QUOTA_SUPPORT
         next->md_ops->mdo_quota.mqo_cleanup(env, next);
 #endif
+        lut_fini(env, &m->mdt_lut);
         mdt_fs_cleanup(env, m);
         upcall_cache_cleanup(m->mdt_identity_cache);
         m->mdt_identity_cache = NULL;
 
         if (m->mdt_namespace != NULL) {
-                ldlm_namespace_free(m->mdt_namespace, NULL, d->ld_obd->obd_force);
+                ldlm_namespace_free(m->mdt_namespace, NULL,
+                                    d->ld_obd->obd_force);
                 d->ld_obd->obd_namespace = m->mdt_namespace = NULL;
         }
 
+        cfs_free_nidlist(&m->mdt_nosquash_nids);
+        if (m->mdt_nosquash_str) {
+                OBD_FREE(m->mdt_nosquash_str, m->mdt_nosquash_strlen);
+                m->mdt_nosquash_str = NULL;
+                m->mdt_nosquash_strlen = 0;
+        }
+
         mdt_seq_fini(env, m);
         mdt_seq_fini_cli(m);
         mdt_fld_fini(env, m);
-        mdt_procfs_fini(m);
-        lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
-        lprocfs_free_per_client_stats(obd);
-        lprocfs_free_obd_stats(obd);
-        ptlrpc_lprocfs_unregister_obd(d->ld_obd);
-        lprocfs_obd_cleanup(d->ld_obd);
-
         sptlrpc_rule_set_free(&m->mdt_sptlrpc_rset);
 
         next->md_ops->mdo_init_capa_ctxt(env, next, 0, 0, 0, NULL);
         cfs_timer_disarm(&m->mdt_ck_timer);
         mdt_ck_thread_stop(m);
 
-        /* finish the stack */
+        /*
+         * Finish the stack
+         */
         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
 
+        mdt_procfs_fini(m);
+        if (obd->obd_proc_exports_entry) {
+                lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
+                obd->obd_proc_exports_entry = NULL;
+        }
+        lprocfs_free_per_client_stats(obd);
+        lprocfs_free_obd_stats(obd);
+        ptlrpc_lprocfs_unregister_obd(obd);
+        lprocfs_obd_cleanup(obd);
+
         if (ls) {
                 struct md_site *mite;
 
@@ -4339,6 +4492,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
 
         m->mdt_max_mdsize = MAX_MD_SIZE;
         m->mdt_max_cookiesize = sizeof(struct llog_cookie);
+        m->mdt_som_conf = 0;
 
         m->mdt_opts.mo_user_xattr = 0;
         m->mdt_opts.mo_acl = 0;
@@ -4368,6 +4522,12 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
         m->mdt_capa_timeout = CAPA_TIMEOUT;
         m->mdt_capa_alg = CAPA_HMAC_ALG_SHA1;
         m->mdt_ck_timeout = CAPA_KEY_TIMEOUT;
+        m->mdt_squash_uid = 0;
+        m->mdt_squash_gid = 0;
+        CFS_INIT_LIST_HEAD(&m->mdt_nosquash_nids);
+        m->mdt_nosquash_str = NULL;
+        m->mdt_nosquash_strlen = 0;
+        init_rwsem(&m->mdt_squash_sem);
 
         spin_lock_init(&m->mdt_client_bitmap_lock);
 
@@ -4435,10 +4595,14 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
                 GOTO(err_fini_proc, rc);
         }
 
-        rc = mdt_fld_init(env, obd->obd_name, m);
+        rc = lut_init(env, &m->mdt_lut, obd, m->mdt_bottom);
         if (rc)
                 GOTO(err_fini_stack, rc);
 
+        rc = mdt_fld_init(env, obd->obd_name, m);
+        if (rc)
+                GOTO(err_lut, rc);
+
         rc = mdt_seq_init(env, obd->obd_name, m);
         if (rc)
                 GOTO(err_fini_fld, rc);
@@ -4482,6 +4646,10 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
         if (rc)
                 GOTO(err_fs_cleanup, rc);
 
+        rc = mdt_llog_ctxt_clone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
+        if (rc)
+                GOTO(err_llog_cleanup, rc);
+
         mdt_adapt_sptlrpc_conf(obd, 1);
 
 #ifdef HAVE_QUOTA_SUPPORT
@@ -4494,7 +4662,7 @@ static int mdt_init0(const struct lu_env *env, struct mdt_device *m,
         server_put_mount_2(dev, lmi->lmi_mnt);
         lmi = NULL;
 
-        target_recovery_init(obd, mdt_recovery_handle);
+        target_recovery_init(&m->mdt_lut, mdt_recovery_handle);
 
         rc = mdt_start_ptlrpc_service(m);
         if (rc)
@@ -4526,8 +4694,9 @@ err_recovery:
         target_recovery_fini(obd);
 #ifdef HAVE_QUOTA_SUPPORT
         next->md_ops->mdo_quota.mqo_cleanup(env, next);
-err_llog_cleanup:
 #endif
+err_llog_cleanup:
+        mdt_llog_ctxt_unclone(env, m, LLOG_CHANGELOG_ORIG_CTXT);
         mdt_obd_llog_cleanup(obd);
 err_fs_cleanup:
         mdt_fs_cleanup(env, m);
@@ -4543,12 +4712,16 @@ err_fini_seq:
         mdt_seq_fini(env, m);
 err_fini_fld:
         mdt_fld_fini(env, m);
+err_lut:
+        lut_fini(env, &m->mdt_lut);
 err_fini_stack:
         mdt_stack_fini(env, m, md2lu_dev(m->mdt_child));
 err_fini_proc:
         mdt_procfs_fini(m);
-        if (obd->obd_proc_exports_entry)
+        if (obd->obd_proc_exports_entry) {
                 lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
+                obd->obd_proc_exports_entry = NULL;
+        }
         ptlrpc_lprocfs_unregister_obd(obd);
         lprocfs_obd_cleanup(obd);
 err_fini_site:
@@ -4556,7 +4729,7 @@ err_fini_site:
 err_free_site:
         OBD_FREE_PTR(mite);
 err_lmi:
-        if (lmi) 
+        if (lmi)
                 server_put_mount_2(dev, lmi->lmi_mnt);
         return (rc);
 }
@@ -4641,7 +4814,7 @@ static struct lu_object *mdt_object_alloc(const struct lu_env *env,
 }
 
 static int mdt_object_init(const struct lu_env *env, struct lu_object *o,
-                           const struct lu_object_conf *_)
+                           const struct lu_object_conf *unused)
 {
         struct mdt_device *d = mdt_dev(o->lo_dev);
         struct lu_device  *under;
@@ -4678,6 +4851,16 @@ static void mdt_object_free(const struct lu_env *env, struct lu_object *o)
         EXIT;
 }
 
+static int mdt_object_print(const struct lu_env *env, void *cookie,
+                            lu_printer_t p, const struct lu_object *o)
+{
+        struct mdt_object *mdto = mdt_obj((struct lu_object *)o);
+        return (*p)(env, cookie, LUSTRE_MDT_NAME"-object@%p(ioepoch=%llu "
+                    "flags=%llx, epochcount=%d, writecount=%d)",
+                    mdto, mdto->mot_ioepoch, mdto->mot_flags,
+                    mdto->mot_epochcount, mdto->mot_writecount);
+}
+
 static const struct lu_device_operations mdt_lu_ops = {
         .ldo_object_alloc   = mdt_object_alloc,
         .ldo_process_config = mdt_process_config,
@@ -4685,7 +4868,8 @@ static const struct lu_device_operations mdt_lu_ops = {
 
 static const struct lu_object_operations mdt_obj_ops = {
         .loo_object_init    = mdt_object_init,
-        .loo_object_free    = mdt_object_free
+        .loo_object_free    = mdt_object_free,
+        .loo_object_print   = mdt_object_print
 };
 
 static int mdt_obd_set_info_async(struct obd_export *exp,
@@ -4729,6 +4913,9 @@ static int mdt_connect_internal(struct obd_export *exp,
                 if (!mdt->mdt_opts.mo_user_xattr)
                         data->ocd_connect_flags &= ~OBD_CONNECT_XATTR;
 
+                if (!mdt->mdt_som_conf)
+                        data->ocd_connect_flags &= ~OBD_CONNECT_SOM;
+
                 spin_lock(&exp->exp_lock);
                 exp->exp_connect_flags = data->ocd_connect_flags;
                 spin_unlock(&exp->exp_lock);
@@ -4751,6 +4938,14 @@ static int mdt_connect_internal(struct obd_export *exp,
                 return -EBADE;
         }
 
+        if (mdt->mdt_som_conf &&
+            !(exp->exp_connect_flags & OBD_CONNECT_MDS_MDS) &&
+            !(exp->exp_connect_flags & OBD_CONNECT_SOM)) {
+                CWARN("%s: MDS has SOM enabled, but client does not support "
+                      "it\n", mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name);
+                return -EBADE;
+        }
+
         return 0;
 }
 
@@ -4800,39 +4995,40 @@ static int mdt_connect_check_sptlrpc(struct mdt_device *mdt,
 
 /* mds_connect copy */
 static int mdt_obd_connect(const struct lu_env *env,
-                           struct lustre_handle *conn, struct obd_device *obd,
+                           struct obd_export **exp, struct obd_device *obd,
                            struct obd_uuid *cluuid,
                            struct obd_connect_data *data,
                            void *localdata)
 {
         struct mdt_thread_info *info;
         struct lsd_client_data *lcd;
-        struct obd_export      *exp;
+        struct obd_export      *lexp;
+        struct lustre_handle    conn = { 0 };
         struct mdt_device      *mdt;
         struct ptlrpc_request  *req;
         int                     rc;
         ENTRY;
 
         LASSERT(env != NULL);
-        if (!conn || !obd || !cluuid)
+        if (!exp || !obd || !cluuid)
                 RETURN(-EINVAL);
 
         info = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
         req = info->mti_pill->rc_req;
         mdt = mdt_dev(obd->obd_lu_dev);
 
-        rc = class_connect(conn, obd, cluuid);
+        rc = class_connect(&conn, obd, cluuid);
         if (rc)
                 RETURN(rc);
 
-        exp = class_conn2export(conn);
-        LASSERT(exp != NULL);
+        lexp = class_conn2export(&conn);
+        LASSERT(lexp != NULL);
 
-        rc = mdt_connect_check_sptlrpc(mdt, exp, req);
+        rc = mdt_connect_check_sptlrpc(mdt, lexp, req);
         if (rc)
                 GOTO(out, rc);
 
-        rc = mdt_connect_internal(exp, mdt, data);
+        rc = mdt_connect_internal(lexp, mdt, data);
         if (rc == 0) {
                 OBD_ALLOC_PTR(lcd);
                 if (lcd != NULL) {
@@ -4840,15 +5036,15 @@ static int mdt_obd_connect(const struct lu_env *env,
                         mti = lu_context_key_get(&env->le_ctx,
                                                  &mdt_thread_key);
                         LASSERT(mti != NULL);
-                        mti->mti_exp = exp;
+                        mti->mti_exp = lexp;
                         memcpy(lcd->lcd_uuid, cluuid, sizeof lcd->lcd_uuid);
-                        exp->exp_mdt_data.med_lcd = lcd;
+                        lexp->exp_mdt_data.med_lcd = lcd;
                         rc = mdt_client_new(env, mdt);
                         if (rc != 0) {
                                 OBD_FREE_PTR(lcd);
-                                exp->exp_mdt_data.med_lcd = NULL;
+                                lexp->exp_mdt_data.med_lcd = NULL;
                         } else {
-                                mdt_export_stats_init(obd, exp, localdata);
+                                mdt_export_stats_init(obd, lexp, localdata);
                         }
                 } else
                         rc = -ENOMEM;
@@ -4856,9 +5052,9 @@ static int mdt_obd_connect(const struct lu_env *env,
 
 out:
         if (rc != 0)
-                class_disconnect(exp);
+                class_disconnect(lexp);
         else
-                class_export_put(exp);
+                *exp = lexp;
 
         RETURN(rc);
 }
@@ -4892,6 +5088,83 @@ static int mdt_obd_reconnect(const struct lu_env *env,
 
         RETURN(rc);
 }
+static int mdt_mfd_cleanup(struct obd_export *exp)
+{
+        struct mdt_export_data *med = &exp->exp_mdt_data;
+        struct obd_device      *obd = exp->exp_obd;
+        struct mdt_device      *mdt;
+        struct mdt_thread_info *info;
+        struct lu_env           env;
+        CFS_LIST_HEAD(closing_list);
+        struct mdt_file_data *mfd, *n;
+        int rc = 0;
+        ENTRY;
+
+        spin_lock(&med->med_open_lock);
+        while (!list_empty(&med->med_open_head)) {
+                struct list_head *tmp = med->med_open_head.next;
+                mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
+
+                /* Remove mfd handle so it can't be found again.
+                 * We are consuming the mfd_list reference here. */
+                class_handle_unhash(&mfd->mfd_handle);
+                list_move_tail(&mfd->mfd_list, &closing_list);
+        }
+        spin_unlock(&med->med_open_lock);
+        mdt = mdt_dev(obd->obd_lu_dev);
+        LASSERT(mdt != NULL);
+
+        rc = lu_env_init(&env, LCT_MD_THREAD);
+        if (rc)
+                RETURN(rc);
+
+        info = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
+        LASSERT(info != NULL);
+        memset(info, 0, sizeof *info);
+        info->mti_env = &env;
+        info->mti_mdt = mdt;
+        info->mti_exp = exp;
+
+        if (!list_empty(&closing_list)) {
+                struct md_attr *ma = &info->mti_attr;
+                int lmm_size;
+                int cookie_size;
+
+                lmm_size = mdt->mdt_max_mdsize;
+                OBD_ALLOC(ma->ma_lmm, lmm_size);
+                if (ma->ma_lmm == NULL)
+                        GOTO(out_lmm, rc = -ENOMEM);
+
+                cookie_size = mdt->mdt_max_cookiesize;
+                OBD_ALLOC(ma->ma_cookie, cookie_size);
+                if (ma->ma_cookie == NULL)
+                        GOTO(out_cookie, rc = -ENOMEM);
+
+                /* Close any open files (which may also cause orphan unlinking). */
+                list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
+                        list_del_init(&mfd->mfd_list);
+                        memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
+                        ma->ma_lmm_size = lmm_size;
+                        ma->ma_cookie_size = cookie_size;
+                        ma->ma_need = 0;
+                        /* It is not for setattr, just tell MDD to send
+                         * DESTROY RPC to OSS if needed */
+                        ma->ma_attr_flags = MDS_CLOSE_CLEANUP;
+                        ma->ma_valid = MA_FLAGS;
+                        mdt_mfd_close(info, mfd);
+                }
+                info->mti_mdt = NULL;
+                OBD_FREE(ma->ma_cookie, cookie_size);
+                ma->ma_cookie = NULL;
+out_cookie:
+                OBD_FREE(ma->ma_lmm, lmm_size);
+                ma->ma_lmm = NULL;
+        }
+out_lmm:
+        lu_env_fini(&env);
+
+        RETURN(rc);
+}
 
 static int mdt_obd_disconnect(struct obd_export *exp)
 {
@@ -4920,11 +5193,13 @@ static int mdt_obd_disconnect(struct obd_export *exp)
 
                 spin_lock(&svc->srv_lock);
                 list_del_init(&rs->rs_exp_list);
+                spin_lock(&rs->rs_lock);
                 ptlrpc_schedule_difficult_reply(rs);
+                spin_unlock(&rs->rs_lock);
                 spin_unlock(&svc->srv_lock);
         }
         spin_unlock(&exp->exp_lock);
-
+        rc = mdt_mfd_cleanup(exp);
         class_export_put(exp);
         RETURN(rc);
 }
@@ -4956,11 +5231,6 @@ static int mdt_destroy_export(struct obd_export *export)
         struct mdt_device      *mdt;
         struct mdt_thread_info *info;
         struct lu_env           env;
-        struct md_attr         *ma;
-        int lmm_size;
-        int cookie_size;
-        CFS_LIST_HEAD(closing_list);
-        struct mdt_file_data *mfd, *n;
         int rc = 0;
         ENTRY;
 
@@ -4971,6 +5241,8 @@ static int mdt_destroy_export(struct obd_export *export)
         target_destroy_export(export);
         ldlm_destroy_export(export);
 
+        LASSERT(list_empty(&export->exp_outstanding_replies));
+        LASSERT(list_empty(&med->med_open_head));
         if (obd_uuid_equals(&export->exp_client_uuid, &obd->obd_uuid))
                 RETURN(0);
 
@@ -4985,62 +5257,12 @@ static int mdt_destroy_export(struct obd_export *export)
         LASSERT(info != NULL);
         memset(info, 0, sizeof *info);
         info->mti_env = &env;
-        info->mti_mdt = mdt;
         info->mti_exp = export;
-
-        ma = &info->mti_attr;
-        lmm_size = ma->ma_lmm_size = mdt->mdt_max_mdsize;
-        cookie_size = ma->ma_cookie_size = mdt->mdt_max_cookiesize;
-        OBD_ALLOC(ma->ma_lmm, lmm_size);
-        OBD_ALLOC(ma->ma_cookie, cookie_size);
-
-        if (ma->ma_lmm == NULL || ma->ma_cookie == NULL)
-                GOTO(out, rc = -ENOMEM);
-        ma->ma_need = MA_LOV | MA_COOKIE;
-        ma->ma_valid = 0;
-        /* Close any open files (which may also cause orphan unlinking). */
-        spin_lock(&med->med_open_lock);
-        while (!list_empty(&med->med_open_head)) {
-                struct list_head *tmp = med->med_open_head.next;
-                mfd = list_entry(tmp, struct mdt_file_data, mfd_list);
-
-                /* Remove mfd handle so it can't be found again.
-                 * We are consuming the mfd_list reference here. */
-                class_handle_unhash(&mfd->mfd_handle);
-                list_move_tail(&mfd->mfd_list, &closing_list);
-        }
-        spin_unlock(&med->med_open_lock);
-
-        list_for_each_entry_safe(mfd, n, &closing_list, mfd_list) {
-                list_del_init(&mfd->mfd_list);
-                mdt_mfd_close(info, mfd);
-                /* TODO: if we close the unlinked file,
-                 * we need to remove its objects from OST */
-                memset(&ma->ma_attr, 0, sizeof(ma->ma_attr));
-                spin_lock(&med->med_open_lock);
-                ma->ma_lmm_size = lmm_size;
-                ma->ma_cookie_size = cookie_size;
-                ma->ma_need = MA_LOV | MA_COOKIE;
-                ma->ma_valid = 0;
-                spin_unlock(&med->med_open_lock);
-        }
-
         info->mti_mdt = NULL;
         mdt_client_del(&env, mdt);
 
-        EXIT;
-out:
-        if (lmm_size) {
-                OBD_FREE(ma->ma_lmm, lmm_size);
-                ma->ma_lmm = NULL;
-        }
-        if (cookie_size) {
-                OBD_FREE(ma->ma_cookie, cookie_size);
-                ma->ma_cookie = NULL;
-        }
         lu_env_fini(&env);
-
-        return rc;
+        RETURN(rc);
 }
 
 static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
@@ -5058,7 +5280,7 @@ static void mdt_allow_cli(struct mdt_device *m, unsigned int flag)
 }
 
 static int mdt_upcall(const struct lu_env *env, struct md_device *md,
-                      enum md_upcall_event ev)
+                      enum md_upcall_event ev, void *data)
 {
         struct mdt_device *m = mdt_dev(&md->md_lu_dev);
         struct md_device  *next  = m->mdt_child;
@@ -5074,10 +5296,8 @@ static int mdt_upcall(const struct lu_env *env, struct md_device *md,
                         CDEBUG(D_INFO, "get max mdsize %d max cookiesize %d\n",
                                      m->mdt_max_mdsize, m->mdt_max_cookiesize);
                         mdt_allow_cli(m, CONFIG_SYNC);
-#ifdef HAVE_QUOTA_SUPPORT
-                        if (md->md_lu_dev.ld_obd->obd_recovering == 0)
-                                next->md_ops->mdo_quota.mqo_recovery(env, next);
-#endif
+                        if (data)
+                                (*(__u64 *)data) = m->mdt_mount_count;
                         break;
                 case MD_NO_TRANS:
                         mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
@@ -5088,6 +5308,13 @@ static int mdt_upcall(const struct lu_env *env, struct md_device *md,
                         /* Check that MDT is not yet configured */
                         LASSERT(!m->mdt_fl_cfglog);
                         break;
+#ifdef HAVE_QUOTA_SUPPORT
+                case MD_LOV_QUOTA:
+                        if (md->md_lu_dev.ld_obd->obd_recovering == 0 &&
+                            likely(md->md_lu_dev.ld_obd->obd_stopping == 0))
+                                next->md_ops->mdo_quota.mqo_recovery(env, next);
+                        break;
+#endif
                 default:
                         CERROR("invalid event\n");
                         rc = -EINVAL;
@@ -5122,11 +5349,183 @@ static int mdt_obd_notify(struct obd_device *host,
         RETURN(0);
 }
 
+static int mdt_rpc_fid2path(struct mdt_thread_info *info, void *key,
+                            void *val, int vallen)
+{
+        struct mdt_device *mdt = mdt_dev(info->mti_exp->exp_obd->obd_lu_dev);
+        struct getinfo_fid2path *fpout, *fpin;
+        int rc = 0;
+
+        fpin = key + size_round(sizeof(KEY_FID2PATH));
+        fpout = val;
+
+        if (ptlrpc_req_need_swab(info->mti_pill->rc_req))
+                lustre_swab_fid2path(fpin);
+
+        memcpy(fpout, fpin, sizeof(*fpin));
+        if (fpout->gf_pathlen != vallen - sizeof(*fpin))
+                RETURN(-EINVAL);
+
+        rc = mdt_fid2path(info->mti_env, mdt, fpout);
+        RETURN(rc);
+}
+
+static int mdt_fid2path(const struct lu_env *env, struct mdt_device *mdt,
+                        struct getinfo_fid2path *fp)
+{
+        struct mdt_object *obj;
+        int    rc;
+        ENTRY;
+
+        CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
+               PFID(&fp->gf_fid), fp->gf_recno, fp->gf_linkno);
+
+        if (!fid_is_sane(&fp->gf_fid))
+                RETURN(-EINVAL);
+
+        obj = mdt_object_find(env, mdt, &fp->gf_fid);
+        if (obj == NULL || IS_ERR(obj)) {
+                CDEBUG(D_IOCTL, "no object "DFID": %ld\n",PFID(&fp->gf_fid),
+                       PTR_ERR(obj));
+                RETURN(-EINVAL);
+        }
+
+        rc = lu_object_exists(&obj->mot_obj.mo_lu);
+        if (rc <= 0) {
+                if (rc == -1)
+                        rc = -EREMOTE;
+                else
+                        rc = -ENOENT;
+                mdt_object_put(env, obj);
+                CDEBUG(D_IOCTL, "nonlocal object "DFID": %d\n",
+                       PFID(&fp->gf_fid), rc);
+                RETURN(rc);
+        }
+
+        rc = mo_path(env, md_object_next(&obj->mot_obj), fp->gf_path,
+                     fp->gf_pathlen, &fp->gf_recno, &fp->gf_linkno);
+        mdt_object_put(env, obj);
+
+        RETURN(rc);
+}
+
+static int mdt_get_info(struct mdt_thread_info *info)
+{
+        struct ptlrpc_request *req = mdt_info_req(info);
+        char *key;
+        int keylen;
+        __u32 *vallen;
+        void *valout;
+        int rc;
+        ENTRY;
+
+        key = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_KEY);
+        if (key == NULL) {
+                CDEBUG(D_IOCTL, "No GETINFO key");
+                RETURN(-EFAULT);
+        }
+        keylen = req_capsule_get_size(info->mti_pill, &RMF_GETINFO_KEY,
+                                      RCL_CLIENT);
+
+        vallen = req_capsule_client_get(info->mti_pill, &RMF_GETINFO_VALLEN);
+        if (vallen == NULL) {
+                CDEBUG(D_IOCTL, "Unable to get RMF_GETINFO_VALLEN buffer");
+                RETURN(-EFAULT);
+        }
+
+        req_capsule_set_size(info->mti_pill, &RMF_GETINFO_VAL, RCL_SERVER,
+                             *vallen);
+        rc = req_capsule_server_pack(info->mti_pill);
+        valout = req_capsule_server_get(info->mti_pill, &RMF_GETINFO_VAL);
+        if (valout == NULL) {
+                CDEBUG(D_IOCTL, "Unable to get get-info RPC out buffer");
+                RETURN(-EFAULT);
+        }
+
+        if (KEY_IS(KEY_FID2PATH))
+                rc = mdt_rpc_fid2path(info, key, valout, *vallen);
+        else
+                rc = -EINVAL;
+
+        lustre_msg_set_status(req->rq_repmsg, rc);
+
+        RETURN(rc);
+}
+
+/* Pass the ioc down */
+static int mdt_ioc_child(struct lu_env *env, struct mdt_device *mdt,
+                         unsigned int cmd, int len, void *data)
+{
+        struct lu_context ioctl_session;
+        struct md_device *next = mdt->mdt_child;
+        int rc;
+        ENTRY;
+
+        rc = lu_context_init(&ioctl_session, LCT_SESSION);
+        if (rc)
+                RETURN(rc);
+        ioctl_session.lc_thread = (struct ptlrpc_thread *)cfs_current();
+        lu_context_enter(&ioctl_session);
+        env->le_ses = &ioctl_session;
+
+        LASSERT(next->md_ops->mdo_iocontrol);
+        rc = next->md_ops->mdo_iocontrol(env, next, cmd, len, data);
+
+        lu_context_exit(&ioctl_session);
+        lu_context_fini(&ioctl_session);
+        RETURN(rc);
+}
+
+static int mdt_ioc_version_get(struct mdt_thread_info *mti, void *karg)
+{
+        struct obd_ioctl_data *data = karg;
+        struct lu_fid *fid = (struct lu_fid *)data->ioc_inlbuf1;
+        __u64 version;
+        struct mdt_object *obj;
+        struct mdt_lock_handle  *lh;
+        int rc;
+        ENTRY;
+        CDEBUG(D_IOCTL, "getting version for "DFID"\n", PFID(fid));
+        if (!fid_is_sane(fid))
+                RETURN(-EINVAL);
+
+        lh = &mti->mti_lh[MDT_LH_PARENT];
+        mdt_lock_reg_init(lh, LCK_CR);
+
+        obj = mdt_object_find_lock(mti, fid, lh, MDS_INODELOCK_UPDATE);
+        if (IS_ERR(obj))
+                RETURN(PTR_ERR(obj));
+
+        rc = mdt_object_exists(obj);
+        if (rc < 0) {
+                rc = -EREMOTE;
+                /**
+                 * before calling version get the correct MDS should be
+                 * fid, this is error to find remote object here
+                 */
+                CERROR("nonlocal object "DFID"\n", PFID(fid));
+        } else if (rc == 0) {
+                rc = -ENOENT;
+                CDEBUG(D_IOCTL, "no such object: "DFID"\n", PFID(fid));
+        } else {
+                version = mo_version_get(mti->mti_env, mdt_object_child(obj));
+                if (version < 0) {
+                        rc = (int)version;
+                } else {
+                        *(__u64 *)data->ioc_inlbuf2 = version;
+                        rc = 0;
+                }
+        }
+        mdt_object_unlock_put(mti, obj, lh, 1);
+        RETURN(rc);
+}
+
+/* ioctls on obd dev */
 static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                          void *karg, void *uarg)
 {
         struct lu_env      env;
-        struct obd_device *obd= exp->exp_obd;
+        struct obd_device *obd = exp->exp_obd;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
         struct dt_device  *dt = mdt->mdt_bottom;
         int rc;
@@ -5149,6 +5548,22 @@ static int mdt_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                 target_stop_recovery_thread(obd);
                 rc = 0;
                 break;
+        case OBD_IOC_CHANGELOG_REG:
+        case OBD_IOC_CHANGELOG_DEREG:
+        case OBD_IOC_CHANGELOG_CLEAR:
+                rc = mdt_ioc_child(&env, mdt, cmd, len, karg);
+                break;
+        case OBD_IOC_GET_OBJ_VERSION: {
+                struct mdt_thread_info *mti;
+                mti = lu_context_key_get(&env.le_ctx, &mdt_thread_key);
+                memset(mti, 0, sizeof *mti);
+                mti->mti_env = &env;
+                mti->mti_mdt = mdt;
+                mti->mti_exp = exp;
+
+                rc = mdt_ioc_version_get(mti, karg);
+                break;
+        }
         default:
                 CERROR("Not supported cmd = %d for device %s\n",
                        cmd, obd->obd_name);
@@ -5175,7 +5590,8 @@ int mdt_postrecov(const struct lu_env *env, struct mdt_device *mdt)
 
         rc = ld->ld_ops->ldo_recovery_complete(env, ld);
 #ifdef HAVE_QUOTA_SUPPORT
-        next->md_ops->mdo_quota.mqo_recovery(env, next);
+        if (likely(obd->obd_stopping == 0))
+                next->md_ops->mdo_quota.mqo_recovery(env, next);
 #endif
         RETURN(rc);
 }
@@ -5193,6 +5609,62 @@ int mdt_obd_postrecov(struct obd_device *obd)
         return rc;
 }
 
+/**
+ * Send a copytool req to a client
+ * Note this sends a request RPC from a server (MDT) to a client (MDC),
+ * backwards of normal comms.
+ */
+int mdt_hsm_copytool_send(struct obd_export *exp)
+{
+        struct lnl_hdr *lh;
+        struct hsm_action_list *hal;
+        struct hsm_action_item *hai;
+        int rc, len;
+        ENTRY;
+
+        CWARN("%s: writing to mdc at %s\n", exp->exp_obd->obd_name,
+              libcfs_nid2str(exp->exp_connection->c_peer.nid));
+
+        len = sizeof(*lh) + sizeof(*hal) + MTI_NAME_MAXLEN +
+                /* for mockup below */ 2 * size_round(sizeof(*hai));
+        OBD_ALLOC(lh, len);
+        if (lh == NULL)
+                RETURN(-ENOMEM);
+
+        lh->lnl_magic = LNL_MAGIC;
+        lh->lnl_transport = LNL_TRANSPORT_HSM;
+        lh->lnl_msgtype = HMT_ACTION_LIST;
+        lh->lnl_msglen = len;
+
+        hal = (struct hsm_action_list *)(lh + 1);
+        hal->hal_version = HAL_VERSION;
+        hal->hal_archive_num = 1;
+        obd_uuid2fsname(hal->hal_fsname, exp->exp_obd->obd_name,
+                        MTI_NAME_MAXLEN);
+
+        /* mock up an action list */
+        hal->hal_count = 2;
+        hai = hai_zero(hal);
+        hai->hai_action = HSMA_ARCHIVE;
+        hai->hai_fid.f_oid = 0xA00A;
+        hai->hai_len = sizeof(*hai);
+        hai = hai_next(hai);
+        hai->hai_action = HSMA_RESTORE;
+        hai->hai_fid.f_oid = 0xB00B;
+        hai->hai_len = sizeof(*hai);
+
+        /* Uses the ldlm reverse import; this rpc will be seen by
+          the ldlm_callback_handler */
+        rc = target_set_info_rpc(exp->exp_imp_reverse, LDLM_SET_INFO,
+                                 sizeof(KEY_HSM_COPYTOOL_SEND),
+                                 KEY_HSM_COPYTOOL_SEND,
+                                 len, lh, NULL);
+
+        OBD_FREE(lh, len);
+
+        RETURN(rc);
+}
+
 static struct obd_ops mdt_obd_device_ops = {
         .o_owner          = THIS_MODULE,
         .o_set_info_async = mdt_obd_set_info_async,
@@ -5272,7 +5744,7 @@ struct md_ucred *mdt_ucred(const struct mdt_thread_info *info)
 }
 
 /**
- * Enable/disable COS.
+ * Enable/disable COS (Commit On Sharing).
  *
  * Set/Clear the COS flag in mdt options.
  *
@@ -5296,9 +5768,9 @@ void mdt_enable_cos(struct mdt_device *mdt, int val)
 }
 
 /**
- * Check COS status.
+ * Check COS (Commit On Sharing) status.
  *
- * Return COS flag status/
+ * Return COS flag status.
  *
  * \param mdt mdt device
  */
@@ -5353,6 +5825,7 @@ static int __init mdt_mod_init(void)
 
 static void __exit mdt_mod_exit(void)
 {
+        llo_local_obj_unregister(&mdt_last_recv);
         class_unregister_type(LUSTRE_MDT_NAME);
 }
 
@@ -5395,7 +5868,9 @@ static void __exit mdt_mod_exit(void)
 static struct mdt_handler mdt_mds_ops[] = {
 DEF_MDT_HNDL_F(0,                         CONNECT,      mdt_connect),
 DEF_MDT_HNDL_F(0,                         DISCONNECT,   mdt_disconnect),
-DEF_MDT_HNDL_F(0,                         SET_INFO,     mdt_set_info),
+DEF_MDT_HNDL  (0,                         SET_INFO,     mdt_set_info,
+                                                             &RQF_OBD_SET_INFO),
+DEF_MDT_HNDL_F(0,                         GET_INFO,     mdt_get_info),
 DEF_MDT_HNDL_F(0           |HABEO_REFERO, GETSTATUS,    mdt_getstatus),
 DEF_MDT_HNDL_F(HABEO_CORPUS,              GETATTR,      mdt_getattr),
 DEF_MDT_HNDL_F(HABEO_CORPUS|HABEO_REFERO, GETATTR_NAME, mdt_getattr_name),
@@ -5435,7 +5910,19 @@ static struct mdt_handler mdt_dlm_ops[] = {
         DEF_DLM_HNDL_0(0,            CP_CALLBACK,    mdt_cp_callback)
 };
 
+#define DEF_LLOG_HNDL(flags, name, fn)                   \
+        DEF_HNDL(LLOG, ORIGIN_HANDLE_CREATE, _NET, flags, name, fn, NULL)
+
 static struct mdt_handler mdt_llog_ops[] = {
+        DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CREATE,      mdt_llog_create),
+        DEF_LLOG_HNDL(0, ORIGIN_HANDLE_NEXT_BLOCK,  mdt_llog_next_block),
+        DEF_LLOG_HNDL(0, ORIGIN_HANDLE_READ_HEADER, mdt_llog_read_header),
+        DEF_LLOG_HNDL(0, ORIGIN_HANDLE_WRITE_REC,   NULL),
+        DEF_LLOG_HNDL(0, ORIGIN_HANDLE_CLOSE,       NULL),
+        DEF_LLOG_HNDL(0, ORIGIN_CONNECT,            NULL),
+        DEF_LLOG_HNDL(0, CATINFO,                   NULL),
+        DEF_LLOG_HNDL(0, ORIGIN_HANDLE_PREV_BLOCK,  mdt_llog_prev_block),
+        DEF_LLOG_HNDL(0, ORIGIN_HANDLE_DESTROY,     mdt_llog_destroy),
 };
 
 #define DEF_SEC_CTX_HNDL(name, fn)                      \