Whamcloud - gitweb
LU-3866 hsm: permission checks on HSM operations
[fs/lustre-release.git] / lustre / mdc / mdc_request.c
index d2fda85..ab9a375 100644 (file)
@@ -1,6 +1,4 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
- *
+/*
  * GPL HEADER START
  *
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  * 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.
+ *
+ * Copyright (c) 2011, 2013, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
  * Lustre is a trademark of Sun Microsystems, Inc.
  */
 
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
 #define DEBUG_SUBSYSTEM S_MDC
 
 #ifdef __KERNEL__
 # include <linux/pagemap.h>
 # include <linux/miscdevice.h>
 # include <linux/init.h>
+# include <linux/utsname.h>
 #else
 # include <liblustre.h>
 #endif
 
 #include <lustre_acl.h>
 #include <obd_class.h>
-#include <lustre_dlm.h>
 #include <lustre_fid.h>
-#include <md_object.h>
 #include <lprocfs_status.h>
 #include <lustre_param.h>
+#include <lustre_log.h>
+
 #include "mdc_internal.h"
-#include <lustre/lustre_idl.h>
 
 #define REQUEST_MINOR 244
 
-static quota_interface_t *quota_interface;
-extern quota_interface_t mdc_quota_interface;
+struct mdc_renew_capa_args {
+        struct obd_capa        *ra_oc;
+        renew_capa_cb_t         ra_cb;
+};
 
 static int mdc_cleanup(struct obd_device *obd);
 
@@ -88,6 +87,24 @@ int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
         }
 }
 
+static inline int mdc_queue_wait(struct ptlrpc_request *req)
+{
+       struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
+       int rc;
+
+       /* mdc_enter_request() ensures that this client has no more
+        * than cl_max_rpcs_in_flight RPCs simultaneously inf light
+        * against an MDT. */
+       rc = mdc_enter_request(cli);
+       if (rc != 0)
+               return rc;
+
+       rc = ptlrpc_queue_wait(req);
+       mdc_exit_request(cli);
+
+       return rc;
+}
+
 /* Helper that implements most of mdc_getstatus and signal_completed_replay. */
 /* XXX this should become mdc_get_info("key"), sending MDS_GET_INFO RPC */
 static int send_getstatus(struct obd_import *imp, struct lu_fid *rootfid,
@@ -209,6 +226,11 @@ int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
         int                    rc;
         ENTRY;
 
+       /* Single MDS without an LMV case */
+       if (op_data->op_flags & MF_GET_MDT_IDX) {
+               op_data->op_mds = 0;
+               RETURN(0);
+       }
         *request = NULL;
         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
         if (req == NULL)
@@ -346,27 +368,45 @@ static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
                                      input_size);
         }
 
-        rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
-        if (rc) {
-                ptlrpc_request_free(req);
-                RETURN(rc);
-        }
+       /* Flush local XATTR locks to get rid of a possible cancel RPC */
+       if (opcode == MDS_REINT && fid_is_sane(fid) &&
+           exp->exp_connect_data.ocd_ibits_known & MDS_INODELOCK_XATTR) {
+               CFS_LIST_HEAD(cancels);
+               int count;
+
+               /* Without that packing would fail */
+               if (input_size == 0)
+                       req_capsule_set_size(&req->rq_pill, &RMF_EADATA,
+                                            RCL_CLIENT, 0);
+
+               count = mdc_resource_get_unused(exp, fid,
+                                               &cancels, LCK_EX,
+                                               MDS_INODELOCK_XATTR);
+
+               rc = mdc_prep_elc_req(exp, req, MDS_REINT, &cancels, count);
+               if (rc) {
+                       ptlrpc_request_free(req);
+                       RETURN(rc);
+               }
+       } else {
+               rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, opcode);
+               if (rc) {
+                       ptlrpc_request_free(req);
+                       RETURN(rc);
+               }
+       }
 
         if (opcode == MDS_REINT) {
                 struct mdt_rec_setxattr *rec;
 
                 CLASSERT(sizeof(struct mdt_rec_setxattr) ==
                          sizeof(struct mdt_rec_reint));
-                rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
-                rec->sx_opcode = REINT_SETXATTR;
-                /* TODO:
-                 *  cfs_curproc_fs{u,g}id() should replace
-                 *  current->fs{u,g}id for portability.
-                 */
-                rec->sx_fsuid  = cfs_curproc_fsuid();
-                rec->sx_fsgid  = cfs_curproc_fsgid();
-                rec->sx_cap    = cfs_curproc_cap_pack();
-                rec->sx_suppgid1 = suppgid;
+               rec = req_capsule_client_get(&req->rq_pill, &RMF_REC_REINT);
+               rec->sx_opcode = REINT_SETXATTR;
+               rec->sx_fsuid  = current_fsuid();
+               rec->sx_fsgid  = current_fsgid();
+               rec->sx_cap    = cfs_curproc_cap_pack();
+               rec->sx_suppgid1 = suppgid;
                 rec->sx_suppgid2 = -1;
                 rec->sx_fid    = *fid;
                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
@@ -449,7 +489,7 @@ static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
         if (!buf)
                 RETURN(-EPROTO);
 
-        acl = posix_acl_from_xattr(buf, body->aclsize);
+        acl = posix_acl_from_xattr(&init_user_ns, buf, body->aclsize);
         if (IS_ERR(acl)) {
                 rc = PTR_ERR(acl);
                 CERROR("convert xattr to acl: %d\n", rc);
@@ -693,11 +733,11 @@ void mdc_commit_open(struct ptlrpc_request *req)
          * be put along with freeing \var mod.
          */
         ptlrpc_request_addref(req);
-        cfs_spin_lock(&req->rq_lock);
-        req->rq_committed = 1;
-        cfs_spin_unlock(&req->rq_lock);
-        req->rq_cb_data = NULL;
-        obd_mod_put(mod);
+       spin_lock(&req->rq_lock);
+       req->rq_committed = 1;
+       spin_unlock(&req->rq_lock);
+       req->rq_cb_data = NULL;
+       obd_mod_put(mod);
 }
 
 int mdc_set_open_replay_data(struct obd_export *exp,
@@ -738,13 +778,13 @@ int mdc_set_open_replay_data(struct obd_export *exp,
                 obd_mod_get(mod);
                 obd_mod_get(mod);
 
-                cfs_spin_lock(&open_req->rq_lock);
-                och->och_mod = mod;
-                mod->mod_och = och;
-                mod->mod_open_req = open_req;
-                open_req->rq_cb_data = mod;
-                open_req->rq_commit_cb = mdc_commit_open;
-                cfs_spin_unlock(&open_req->rq_lock);
+               spin_lock(&open_req->rq_lock);
+               och->och_mod = mod;
+               mod->mod_och = och;
+               mod->mod_open_req = open_req;
+               open_req->rq_cb_data = mod;
+               open_req->rq_commit_cb = mdc_commit_open;
+               spin_unlock(&open_req->rq_lock);
         }
 
         rec->cr_fid2 = body->fid1;
@@ -767,7 +807,14 @@ int mdc_clear_open_replay_data(struct obd_export *exp,
         struct md_open_data *mod = och->och_mod;
         ENTRY;
 
-        LASSERT(mod != LP_POISON && mod != NULL);
+        /**
+         * It is possible to not have \var mod in a case of eviction between
+         * lookup and ll_file_open().
+         **/
+        if (mod == NULL)
+                RETURN(0);
+
+        LASSERT(mod != LP_POISON);
 
         mod->mod_och = NULL;
         och->och_mod = NULL;
@@ -795,13 +842,29 @@ static void mdc_close_handle_reply(struct ptlrpc_request *req,
 int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
               struct md_open_data *mod, struct ptlrpc_request **request)
 {
-        struct obd_device     *obd = class_exp2obd(exp);
-        struct ptlrpc_request *req;
-        int                    rc;
-        ENTRY;
-
-        *request = NULL;
-        req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_CLOSE);
+       struct obd_device     *obd = class_exp2obd(exp);
+       struct ptlrpc_request *req;
+       struct req_format     *req_fmt;
+       int                    rc;
+       int                    saved_rc = 0;
+       ENTRY;
+
+       req_fmt = &RQF_MDS_CLOSE;
+       if (op_data->op_bias & MDS_HSM_RELEASE) {
+               req_fmt = &RQF_MDS_RELEASE_CLOSE;
+
+               /* allocate a FID for volatile file */
+               rc = mdc_fid_alloc(exp, &op_data->op_fid2, op_data);
+               if (rc < 0) {
+                       CERROR("%s: "DFID" failed to allocate FID: %d\n",
+                              obd->obd_name, PFID(&op_data->op_fid1), rc);
+                       /* save the errcode and proceed to close */
+                       saved_rc = rc;
+               }
+       }
+
+       *request = NULL;
+       req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
         if (req == NULL)
                 RETURN(-ENOMEM);
 
@@ -830,9 +893,9 @@ int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched open");
                 /* We no longer want to preserve this open for replay even
                  * though the open was committed. b=3632, b=3633 */
-                cfs_spin_lock(&mod->mod_open_req->rq_lock);
-                mod->mod_open_req->rq_replay = 0;
-                cfs_spin_unlock(&mod->mod_open_req->rq_lock);
+               spin_lock(&mod->mod_open_req->rq_lock);
+               mod->mod_open_req->rq_replay = 0;
+               spin_unlock(&mod->mod_open_req->rq_lock);
         } else {
                  CDEBUG(D_HA, "couldn't find open req; expecting close error\n");
         }
@@ -875,6 +938,7 @@ int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
                  * exists and return no error in that case
                  */
                 if (mod) {
+                        DEBUG_REQ(D_HA, req, "Reset ESTALE = %d", rc);
                         LASSERT(mod->mod_open_req != NULL);
                         if (mod->mod_open_req->rq_committed)
                                 rc = 0;
@@ -890,7 +954,7 @@ int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
         }
         *request = req;
         mdc_close_handle_reply(req, op_data, rc);
-        RETURN(rc);
+        RETURN(rc < 0 ? rc : saved_rc);
 }
 
 int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
@@ -922,9 +986,9 @@ int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
                 DEBUG_REQ(D_HA, mod->mod_open_req, "matched setattr");
                 /* We no longer want to preserve this setattr for replay even
                  * though the open was committed. b=3632, b=3633 */
-                cfs_spin_lock(&mod->mod_open_req->rq_lock);
-                mod->mod_open_req->rq_replay = 0;
-                cfs_spin_unlock(&mod->mod_open_req->rq_lock);
+               spin_lock(&mod->mod_open_req->rq_lock);
+               mod->mod_open_req->rq_replay = 0;
+               spin_unlock(&mod->mod_open_req->rq_lock);
         }
 
         mdc_close_pack(req, op_data);
@@ -985,9 +1049,9 @@ int mdc_sendpage(struct obd_export *exp, const struct lu_fid *fid,
         req->rq_request_portal = MDS_READPAGE_PORTAL;
         ptlrpc_at_set_req_timeout(req);
 
-        desc = ptlrpc_prep_bulk_imp(req, 1, BULK_GET_SOURCE, MDS_BULK_PORTAL);
-        if (desc == NULL)
-                GOTO(out, rc = -ENOMEM);
+       desc = ptlrpc_prep_bulk_imp(req, 1, 1,BULK_GET_SOURCE, MDS_BULK_PORTAL);
+       if (desc == NULL)
+               GOTO(out, rc = -ENOMEM);
 
         /* NB req now owns desc and will free it when it gets freed. */
         ptlrpc_prep_bulk_page(desc, (struct page *)page, 0, offset);
@@ -1006,21 +1070,27 @@ out:
 EXPORT_SYMBOL(mdc_sendpage);
 #endif
 
-int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
-                 struct obd_capa *oc, __u64 offset, struct page *page,
-                 struct ptlrpc_request **request)
+int mdc_readpage(struct obd_export *exp, struct md_op_data *op_data,
+                struct page **pages, struct ptlrpc_request **request)
 {
-        struct ptlrpc_request   *req;
-        struct ptlrpc_bulk_desc *desc;
-        int                      rc;
-        ENTRY;
-
-        *request = NULL;
+       struct ptlrpc_request   *req;
+       struct ptlrpc_bulk_desc *desc;
+       int                      i;
+       wait_queue_head_t        waitq;
+       int                      resends = 0;
+       struct l_wait_info       lwi;
+       int                      rc;
+       ENTRY;
+
+       *request = NULL;
+       init_waitqueue_head(&waitq);
+
+restart_bulk:
         req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_READPAGE);
         if (req == NULL)
                 RETURN(-ENOMEM);
 
-        mdc_set_capa_size(req, &RMF_CAPA1, oc);
+        mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
 
         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
         if (rc) {
@@ -1031,21 +1101,37 @@ int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
         req->rq_request_portal = MDS_READPAGE_PORTAL;
         ptlrpc_at_set_req_timeout(req);
 
-        desc = ptlrpc_prep_bulk_imp(req, 1, BULK_PUT_SINK, MDS_BULK_PORTAL);
+       desc = ptlrpc_prep_bulk_imp(req, op_data->op_npages, 1, BULK_PUT_SINK,
+                                   MDS_BULK_PORTAL);
         if (desc == NULL) {
                 ptlrpc_request_free(req);
                 RETURN(-ENOMEM);
         }
 
         /* NB req now owns desc and will free it when it gets freed */
-        ptlrpc_prep_bulk_page(desc, page, 0, CFS_PAGE_SIZE);
-        mdc_readdir_pack(req, offset, CFS_PAGE_SIZE, fid, oc);
+        for (i = 0; i < op_data->op_npages; i++)
+               ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
+
+        mdc_readdir_pack(req, op_data->op_offset,
+                        PAGE_CACHE_SIZE * op_data->op_npages,
+                         &op_data->op_fid1, op_data->op_capa1);
 
         ptlrpc_request_set_replen(req);
         rc = ptlrpc_queue_wait(req);
         if (rc) {
                 ptlrpc_req_finished(req);
-                RETURN(rc);
+                if (rc != -ETIMEDOUT)
+                        RETURN(rc);
+
+                resends++;
+                if (!client_should_resend(resends, &exp->exp_obd->u.cli)) {
+                        CERROR("too many resend retries, returning error\n");
+                        RETURN(-EIO);
+                }
+                lwi = LWI_TIMEOUT_INTR(cfs_time_seconds(resends), NULL, NULL, NULL);
+                l_wait_event(waitq, 0, &lwi);
+
+                goto restart_bulk;
         }
 
         rc = sptlrpc_cli_unwrap_bulk_read(req, req->rq_bulk,
@@ -1055,9 +1141,10 @@ int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
                 RETURN(rc);
         }
 
-        if (req->rq_bulk->bd_nob_transferred != CFS_PAGE_SIZE) {
+        if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
                 CERROR("Unexpected # bytes transferred: %d (%ld expected)\n",
-                        req->rq_bulk->bd_nob_transferred, CFS_PAGE_SIZE);
+                        req->rq_bulk->bd_nob_transferred,
+                       PAGE_CACHE_SIZE * op_data->op_npages);
                 ptlrpc_req_finished(req);
                 RETURN(-EPROTO);
         }
@@ -1066,9 +1153,11 @@ int mdc_readpage(struct obd_export *exp, const struct lu_fid *fid,
         RETURN(0);
 }
 
-static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
+static int mdc_statfs(const struct lu_env *env,
+                      struct obd_export *exp, struct obd_statfs *osfs,
                       __u64 max_age, __u32 flags)
 {
+        struct obd_device     *obd = class_exp2obd(exp);
         struct ptlrpc_request *req;
         struct obd_statfs     *msfs;
         struct obd_import     *imp = NULL;
@@ -1079,10 +1168,10 @@ static int mdc_statfs(struct obd_device *obd, struct obd_statfs *osfs,
          * Since the request might also come from lprocfs, so we need
          * sync this with client_disconnect_export Bug15684
          */
-        cfs_down_read(&obd->u.cli.cl_sem);
+       down_read(&obd->u.cli.cl_sem);
         if (obd->u.cli.cl_import)
                 imp = class_import_get(obd->u.cli.cl_import);
-        cfs_up_read(&obd->u.cli.cl_sem);
+       up_read(&obd->u.cli.cl_sem);
         if (!imp)
                 RETURN(-ENODEV);
 
@@ -1148,9 +1237,9 @@ static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
         /* Val is struct getinfo_fid2path result plus path */
         vallen = sizeof(*gf) + gf->gf_pathlen;
 
-        rc = obd_get_info(exp, keylen, key, &vallen, gf, NULL);
-        if (rc)
-                GOTO(out, rc);
+       rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
+       if (rc != 0 && rc != -EREMOTE)
+               GOTO(out, rc);
 
         if (vallen <= sizeof(*gf))
                 GOTO(out, rc = -EPROTO);
@@ -1165,61 +1254,337 @@ out:
         return rc;
 }
 
-static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
+static int mdc_ioc_hsm_progress(struct obd_export *exp,
+                               struct hsm_progress_kernel *hpk)
 {
-        struct kuc_hdr *lh = (struct kuc_hdr *)buf;
+       struct obd_import               *imp = class_exp2cliimp(exp);
+       struct hsm_progress_kernel      *req_hpk;
+       struct ptlrpc_request           *req;
+       int                              rc;
+       ENTRY;
+
+       req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_PROGRESS,
+                                       LUSTRE_MDS_VERSION, MDS_HSM_PROGRESS);
+       if (req == NULL)
+               GOTO(out, rc = -ENOMEM);
+
+       mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
 
-        LASSERT(len <= CR_MAXSIZE);
+       /* Copy hsm_progress struct */
+       req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
+       if (req_hpk == NULL)
+               GOTO(out, rc = -EPROTO);
 
-        lh->kuc_magic = KUC_MAGIC;
-        lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
-        lh->kuc_flags = flags;
-        lh->kuc_msgtype = CL_RECORD;
-        lh->kuc_msglen = len;
-        return lh;
+       *req_hpk = *hpk;
+       req_hpk->hpk_errval = lustre_errno_hton(hpk->hpk_errval);
+
+       ptlrpc_request_set_replen(req);
+
+       rc = mdc_queue_wait(req);
+       GOTO(out, rc);
+out:
+       ptlrpc_req_finished(req);
+       return rc;
 }
 
-#define D_CHANGELOG 0
+static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
+{
+       __u32                   *archive_mask;
+       struct ptlrpc_request   *req;
+       int                      rc;
+       ENTRY;
 
-struct changelog_show {
-        __u64       cs_startrec;
-        __u32       cs_flags;
-        cfs_file_t *cs_fp;
-        char       *cs_buf;
-        struct obd_device *cs_obd;
-};
+       req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_REGISTER,
+                                       LUSTRE_MDS_VERSION,
+                                       MDS_HSM_CT_REGISTER);
+       if (req == NULL)
+               GOTO(out, rc = -ENOMEM);
+
+       mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
+
+       /* Copy hsm_progress struct */
+       archive_mask = req_capsule_client_get(&req->rq_pill,
+                                             &RMF_MDS_HSM_ARCHIVE);
+       if (archive_mask == NULL)
+               GOTO(out, rc = -EPROTO);
+
+       *archive_mask = archives;
 
-static int changelog_show_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
-                             void *data)
+       ptlrpc_request_set_replen(req);
+
+       rc = mdc_queue_wait(req);
+       GOTO(out, rc);
+out:
+       ptlrpc_req_finished(req);
+       return rc;
+}
+
+static int mdc_ioc_hsm_current_action(struct obd_export *exp,
+                                     struct md_op_data *op_data)
 {
-        struct changelog_show *cs = data;
-        struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
-        struct kuc_hdr *lh;
-        int len, rc;
-        ENTRY;
+       struct hsm_current_action       *hca = op_data->op_data;
+       struct hsm_current_action       *req_hca;
+       struct ptlrpc_request           *req;
+       int                              rc;
+       ENTRY;
 
-        if ((rec->cr_hdr.lrh_type != CHANGELOG_REC) ||
-            (rec->cr.cr_type >= CL_LAST)) {
-                CERROR("Not a changelog rec %d/%d\n", rec->cr_hdr.lrh_type,
-                       rec->cr.cr_type);
-                RETURN(-EINVAL);
-        }
+       req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+                                  &RQF_MDS_HSM_ACTION);
+       if (req == NULL)
+               RETURN(-ENOMEM);
 
-        if (rec->cr.cr_index < cs->cs_startrec) {
-                /* Skip entries earlier than what we are interested in */
-                CDEBUG(D_CHANGELOG, "rec="LPU64" start="LPU64"\n",
-                       rec->cr.cr_index, cs->cs_startrec);
-                RETURN(0);
-        }
+       mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
+
+       rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_ACTION);
+       if (rc) {
+               ptlrpc_request_free(req);
+               RETURN(rc);
+       }
+
+       mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
+                     OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
+
+       ptlrpc_request_set_replen(req);
+
+       rc = mdc_queue_wait(req);
+       if (rc)
+               GOTO(out, rc);
+
+       req_hca = req_capsule_server_get(&req->rq_pill,
+                                        &RMF_MDS_HSM_CURRENT_ACTION);
+       if (req_hca == NULL)
+               GOTO(out, rc = -EPROTO);
+
+       *hca = *req_hca;
+
+       EXIT;
+out:
+       ptlrpc_req_finished(req);
+       return rc;
+}
+
+static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
+{
+       struct ptlrpc_request   *req;
+       int                      rc;
+       ENTRY;
+
+       req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_HSM_CT_UNREGISTER,
+                                       LUSTRE_MDS_VERSION,
+                                       MDS_HSM_CT_UNREGISTER);
+       if (req == NULL)
+               GOTO(out, rc = -ENOMEM);
+
+       mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
+
+       ptlrpc_request_set_replen(req);
+
+       rc = mdc_queue_wait(req);
+       GOTO(out, rc);
+out:
+       ptlrpc_req_finished(req);
+       return rc;
+}
 
-        CDEBUG(D_CHANGELOG, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID
-               " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
-               changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
-               rec->cr.cr_flags & CLF_FLAGMASK,
-               PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
-               rec->cr.cr_namelen, rec->cr.cr_name);
+static int mdc_ioc_hsm_state_get(struct obd_export *exp,
+                                struct md_op_data *op_data)
+{
+       struct hsm_user_state   *hus = op_data->op_data;
+       struct hsm_user_state   *req_hus;
+       struct ptlrpc_request   *req;
+       int                      rc;
+       ENTRY;
+
+       req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+                                  &RQF_MDS_HSM_STATE_GET);
+       if (req == NULL)
+               RETURN(-ENOMEM);
+
+       mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
+
+       rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_GET);
+       if (rc != 0) {
+               ptlrpc_request_free(req);
+               RETURN(rc);
+       }
+
+       mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
+                     OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
+
+       ptlrpc_request_set_replen(req);
+
+       rc = mdc_queue_wait(req);
+       if (rc)
+               GOTO(out, rc);
+
+       req_hus = req_capsule_server_get(&req->rq_pill, &RMF_HSM_USER_STATE);
+       if (req_hus == NULL)
+               GOTO(out, rc = -EPROTO);
+
+       *hus = *req_hus;
+
+       EXIT;
+out:
+       ptlrpc_req_finished(req);
+       return rc;
+}
+
+static int mdc_ioc_hsm_state_set(struct obd_export *exp,
+                                struct md_op_data *op_data)
+{
+       struct hsm_state_set    *hss = op_data->op_data;
+       struct hsm_state_set    *req_hss;
+       struct ptlrpc_request   *req;
+       int                      rc;
+       ENTRY;
+
+       req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+                                  &RQF_MDS_HSM_STATE_SET);
+       if (req == NULL)
+               RETURN(-ENOMEM);
+
+       mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
+
+       rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_STATE_SET);
+       if (rc) {
+               ptlrpc_request_free(req);
+               RETURN(rc);
+       }
+
+       mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
+                     OBD_MD_FLRMTPERM, 0, op_data->op_suppgids[0], 0);
+
+       /* Copy states */
+       req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
+       if (req_hss == NULL)
+               GOTO(out, rc = -EPROTO);
+       *req_hss = *hss;
+
+       ptlrpc_request_set_replen(req);
+
+       rc = mdc_queue_wait(req);
+       GOTO(out, rc);
 
-        len = sizeof(*lh) + sizeof(rec->cr) + rec->cr.cr_namelen;
+       EXIT;
+out:
+       ptlrpc_req_finished(req);
+       return rc;
+}
+
+static int mdc_ioc_hsm_request(struct obd_export *exp,
+                              struct hsm_user_request *hur)
+{
+       struct obd_import       *imp = class_exp2cliimp(exp);
+       struct ptlrpc_request   *req;
+       struct hsm_request      *req_hr;
+       struct hsm_user_item    *req_hui;
+       char                    *req_opaque;
+       int                      rc;
+       ENTRY;
+
+       req = ptlrpc_request_alloc(imp, &RQF_MDS_HSM_REQUEST);
+       if (req == NULL)
+               GOTO(out, rc = -ENOMEM);
+
+       req_capsule_set_size(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM, RCL_CLIENT,
+                            hur->hur_request.hr_itemcount
+                            * sizeof(struct hsm_user_item));
+       req_capsule_set_size(&req->rq_pill, &RMF_GENERIC_DATA, RCL_CLIENT,
+                            hur->hur_request.hr_data_len);
+
+       rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_HSM_REQUEST);
+       if (rc) {
+               ptlrpc_request_free(req);
+               RETURN(rc);
+       }
+
+       mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
+
+       /* Copy hsm_request struct */
+       req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
+       if (req_hr == NULL)
+               GOTO(out, rc = -EPROTO);
+       *req_hr = hur->hur_request;
+
+       /* Copy hsm_user_item structs */
+       req_hui = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_USER_ITEM);
+       if (req_hui == NULL)
+               GOTO(out, rc = -EPROTO);
+       memcpy(req_hui, hur->hur_user_item,
+              hur->hur_request.hr_itemcount * sizeof(struct hsm_user_item));
+
+       /* Copy opaque field */
+       req_opaque = req_capsule_client_get(&req->rq_pill, &RMF_GENERIC_DATA);
+       if (req_opaque == NULL)
+               GOTO(out, rc = -EPROTO);
+       memcpy(req_opaque, hur_data(hur), hur->hur_request.hr_data_len);
+
+       ptlrpc_request_set_replen(req);
+
+       rc = mdc_queue_wait(req);
+       GOTO(out, rc);
+
+out:
+       ptlrpc_req_finished(req);
+       return rc;
+}
+
+static struct kuc_hdr *changelog_kuc_hdr(char *buf, int len, int flags)
+{
+       struct kuc_hdr *lh = (struct kuc_hdr *)buf;
+
+       LASSERT(len <= KUC_CHANGELOG_MSG_MAXSIZE);
+
+       lh->kuc_magic = KUC_MAGIC;
+       lh->kuc_transport = KUC_TRANSPORT_CHANGELOG;
+       lh->kuc_flags = flags;
+       lh->kuc_msgtype = CL_RECORD;
+       lh->kuc_msglen = len;
+       return lh;
+}
+
+#define D_CHANGELOG 0
+
+struct changelog_show {
+       __u64           cs_startrec;
+       __u32           cs_flags;
+       struct file     *cs_fp;
+       char            *cs_buf;
+       struct obd_device *cs_obd;
+};
+
+static int changelog_kkuc_cb(const struct lu_env *env, struct llog_handle *llh,
+                            struct llog_rec_hdr *hdr, void *data)
+{
+       struct changelog_show *cs = data;
+       struct llog_changelog_rec *rec = (struct llog_changelog_rec *)hdr;
+       struct kuc_hdr *lh;
+       int len, rc;
+       ENTRY;
+
+       if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
+               rc = -EINVAL;
+               CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
+                      cs->cs_obd->obd_name, rec->cr_hdr.lrh_type,
+                      rec->cr.cr_type, rc);
+               RETURN(rc);
+       }
+
+       if (rec->cr.cr_index < cs->cs_startrec) {
+               /* Skip entries earlier than what we are interested in */
+               CDEBUG(D_CHANGELOG, "rec="LPU64" start="LPU64"\n",
+                      rec->cr.cr_index, cs->cs_startrec);
+               RETURN(0);
+       }
+
+       CDEBUG(D_CHANGELOG, LPU64" %02d%-5s "LPU64" 0x%x t="DFID" p="DFID
+               " %.*s\n", rec->cr.cr_index, rec->cr.cr_type,
+               changelog_type2str(rec->cr.cr_type), rec->cr.cr_time,
+               rec->cr.cr_flags & CLF_FLAGMASK,
+               PFID(&rec->cr.cr_tfid), PFID(&rec->cr.cr_pfid),
+               rec->cr.cr_namelen, changelog_rec_name(&rec->cr));
+
+       len = sizeof(*lh) + changelog_rec_size(&rec->cr) + rec->cr.cr_namelen;
 
         /* Set up the message */
         lh = changelog_kuc_hdr(cs->cs_buf, len, cs->cs_flags);
@@ -1233,37 +1598,37 @@ static int changelog_show_cb(struct llog_handle *llh, struct llog_rec_hdr *hdr,
 
 static int mdc_changelog_send_thread(void *csdata)
 {
-        struct changelog_show *cs = csdata;
-        struct llog_ctxt *ctxt = NULL;
-        struct llog_handle *llh = NULL;
-        struct kuc_hdr *kuch;
-        int rc;
+       struct changelog_show *cs = csdata;
+       struct llog_ctxt *ctxt = NULL;
+       struct llog_handle *llh = NULL;
+       struct kuc_hdr *kuch;
+       int rc;
 
-        CDEBUG(D_CHANGELOG, "changelog to fp=%p start "LPU64"\n",
-               cs->cs_fp, cs->cs_startrec);
+       CDEBUG(D_CHANGELOG, "changelog to fp=%p start "LPU64"\n",
+              cs->cs_fp, cs->cs_startrec);
 
-        OBD_ALLOC(cs->cs_buf, CR_MAXSIZE);
-        if (cs->cs_buf == NULL)
-                GOTO(out, rc = -ENOMEM);
+       OBD_ALLOC(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
+       if (cs->cs_buf == NULL)
+               GOTO(out, rc = -ENOMEM);
 
         /* Set up the remote catalog handle */
         ctxt = llog_get_context(cs->cs_obd, LLOG_CHANGELOG_REPL_CTXT);
         if (ctxt == NULL)
                 GOTO(out, rc = -ENOENT);
-        rc = llog_create(ctxt, &llh, NULL, CHANGELOG_CATALOG);
-        if (rc) {
-                CERROR("llog_create() failed %d\n", rc);
-                GOTO(out, rc);
-        }
-        rc = llog_init_handle(llh, LLOG_F_IS_CAT, NULL);
-        if (rc) {
-                CERROR("llog_init_handle failed %d\n", rc);
-                GOTO(out, rc);
-        }
-
-        /* We need the pipe fd open, so llog_process can't daemonize */
-        rc = llog_cat_process_flags(llh, changelog_show_cb, cs,
-                                    LLOG_FLAG_NODEAMON, 0, 0);
+       rc = llog_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
+                      LLOG_OPEN_EXISTS);
+       if (rc) {
+               CERROR("%s: fail to open changelog catalog: rc = %d\n",
+                      cs->cs_obd->obd_name, rc);
+               GOTO(out, rc);
+       }
+       rc = llog_init_handle(NULL, llh, LLOG_F_IS_CAT, NULL);
+       if (rc) {
+               CERROR("llog_init_handle failed %d\n", rc);
+               GOTO(out, rc);
+       }
+
+       rc = llog_cat_process(NULL, llh, changelog_kkuc_cb, cs, 0, 0);
 
         /* Send EOF no matter what our result */
         if ((kuch = changelog_kuc_hdr(cs->cs_buf, sizeof(*kuch),
@@ -1273,15 +1638,15 @@ static int mdc_changelog_send_thread(void *csdata)
         }
 
 out:
-        cfs_put_file(cs->cs_fp);
-        if (llh)
-                llog_cat_put(llh);
+       fput(cs->cs_fp);
+       if (llh)
+               llog_cat_close(NULL, llh);
         if (ctxt)
                 llog_ctxt_put(ctxt);
-        if (cs->cs_buf)
-                OBD_FREE(cs->cs_buf, CR_MAXSIZE);
-        OBD_FREE_PTR(cs);
-        return rc;
+       if (cs->cs_buf)
+               OBD_FREE(cs->cs_buf, KUC_CHANGELOG_MSG_MAXSIZE);
+       OBD_FREE_PTR(cs);
+       return rc;
 }
 
 static int mdc_ioc_changelog_send(struct obd_device *obd,
@@ -1295,20 +1660,22 @@ static int mdc_ioc_changelog_send(struct obd_device *obd,
         if (!cs)
                 return -ENOMEM;
 
-        cs->cs_obd = obd;
-        cs->cs_startrec = icc->icc_recno;
-        /* matching cfs_put_file in mdc_changelog_send_thread */
-        cs->cs_fp = cfs_get_fd(icc->icc_id);
-        cs->cs_flags = icc->icc_flags;
-
-        /* New thread because we should return to user app before
-           writing into our pipe */
-        rc = cfs_kernel_thread(mdc_changelog_send_thread, cs,
-                               CLONE_VM | CLONE_FILES);
-        if (rc >= 0) {
-                CDEBUG(D_CHANGELOG, "start changelog thread: %d\n", rc);
-                return 0;
-        }
+       cs->cs_obd = obd;
+       cs->cs_startrec = icc->icc_recno;
+       /* matching fput in mdc_changelog_send_thread */
+       cs->cs_fp = fget(icc->icc_id);
+       cs->cs_flags = icc->icc_flags;
+
+       /*
+        * New thread because we should return to user app before
+        * writing into our pipe
+        */
+       rc = PTR_ERR(kthread_run(mdc_changelog_send_thread, cs,
+                                "mdc_clg_send_thread"));
+       if (!IS_ERR_VALUE(rc)) {
+               CDEBUG(D_CHANGELOG, "start changelog thread\n");
+               return 0;
+       }
 
         CERROR("Failed to start changelog thread: %d\n", rc);
         OBD_FREE_PTR(cs);
@@ -1318,6 +1685,147 @@ static int mdc_ioc_changelog_send(struct obd_device *obd,
 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
                                 struct lustre_kernelcomm *lk);
 
+static int mdc_quotacheck(struct obd_device *unused, struct obd_export *exp,
+                          struct obd_quotactl *oqctl)
+{
+        struct client_obd       *cli = &exp->exp_obd->u.cli;
+        struct ptlrpc_request   *req;
+        struct obd_quotactl     *body;
+        int                      rc;
+        ENTRY;
+
+        req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
+                                        &RQF_MDS_QUOTACHECK, LUSTRE_MDS_VERSION,
+                                        MDS_QUOTACHECK);
+        if (req == NULL)
+                RETURN(-ENOMEM);
+
+        body = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
+        *body = *oqctl;
+
+        ptlrpc_request_set_replen(req);
+
+        /* the next poll will find -ENODATA, that means quotacheck is
+         * going on */
+        cli->cl_qchk_stat = -ENODATA;
+        rc = ptlrpc_queue_wait(req);
+        if (rc)
+                cli->cl_qchk_stat = rc;
+        ptlrpc_req_finished(req);
+        RETURN(rc);
+}
+
+static int mdc_quota_poll_check(struct obd_export *exp,
+                                struct if_quotacheck *qchk)
+{
+        struct client_obd *cli = &exp->exp_obd->u.cli;
+        int rc;
+        ENTRY;
+
+        qchk->obd_uuid = cli->cl_target_uuid;
+        memcpy(qchk->obd_type, LUSTRE_MDS_NAME, strlen(LUSTRE_MDS_NAME));
+
+        rc = cli->cl_qchk_stat;
+        /* the client is not the previous one */
+        if (rc == CL_NOT_QUOTACHECKED)
+                rc = -EINTR;
+        RETURN(rc);
+}
+
+static int mdc_quotactl(struct obd_device *unused, struct obd_export *exp,
+                        struct obd_quotactl *oqctl)
+{
+        struct ptlrpc_request   *req;
+        struct obd_quotactl     *oqc;
+        int                      rc;
+        ENTRY;
+
+        req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp),
+                                        &RQF_MDS_QUOTACTL, LUSTRE_MDS_VERSION,
+                                        MDS_QUOTACTL);
+        if (req == NULL)
+                RETURN(-ENOMEM);
+
+        oqc = req_capsule_client_get(&req->rq_pill, &RMF_OBD_QUOTACTL);
+        *oqc = *oqctl;
+
+        ptlrpc_request_set_replen(req);
+        ptlrpc_at_set_req_timeout(req);
+        req->rq_no_resend = 1;
+
+        rc = ptlrpc_queue_wait(req);
+        if (rc)
+                CERROR("ptlrpc_queue_wait failed, rc: %d\n", rc);
+
+        if (req->rq_repmsg &&
+            (oqc = req_capsule_server_get(&req->rq_pill, &RMF_OBD_QUOTACTL))) {
+                *oqctl = *oqc;
+        } else if (!rc) {
+                CERROR ("Can't unpack obd_quotactl\n");
+                rc = -EPROTO;
+        }
+        ptlrpc_req_finished(req);
+
+        RETURN(rc);
+}
+
+static int mdc_ioc_swap_layouts(struct obd_export *exp,
+                               struct md_op_data *op_data)
+{
+       CFS_LIST_HEAD(cancels);
+       struct ptlrpc_request   *req;
+       int                      rc, count;
+       struct mdc_swap_layouts *msl, *payload;
+       ENTRY;
+
+       msl = op_data->op_data;
+
+       /* When the MDT will get the MDS_SWAP_LAYOUTS RPC the
+        * first thing it will do is to cancel the 2 layout
+        * locks hold by this client.
+        * So the client must cancel its layout locks on the 2 fids
+        * with the request RPC to avoid extra RPC round trips
+        */
+       count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
+                                       LCK_CR, MDS_INODELOCK_LAYOUT);
+       count += mdc_resource_get_unused(exp, &op_data->op_fid2, &cancels,
+                                        LCK_CR, MDS_INODELOCK_LAYOUT);
+
+       req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+                                  &RQF_MDS_SWAP_LAYOUTS);
+       if (req == NULL) {
+               ldlm_lock_list_put(&cancels, l_bl_ast, count);
+               RETURN(-ENOMEM);
+       }
+
+       mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
+       mdc_set_capa_size(req, &RMF_CAPA2, op_data->op_capa2);
+
+       rc = mdc_prep_elc_req(exp, req, MDS_SWAP_LAYOUTS, &cancels, count);
+       if (rc) {
+               ptlrpc_request_free(req);
+               RETURN(rc);
+       }
+
+       mdc_swap_layouts_pack(req, op_data);
+
+       payload = req_capsule_client_get(&req->rq_pill, &RMF_SWAP_LAYOUTS);
+       LASSERT(payload);
+
+       *payload = *msl;
+
+       ptlrpc_request_set_replen(req);
+
+       rc = ptlrpc_queue_wait(req);
+       if (rc)
+               GOTO(out, rc);
+       EXIT;
+
+out:
+       ptlrpc_req_finished(req);
+       return rc;
+}
+
 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                          void *karg, void *uarg)
 {
@@ -1328,14 +1836,11 @@ static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
         int rc;
         ENTRY;
 
-        if (!cfs_try_module_get(THIS_MODULE)) {
-                CERROR("Can't get module. Is it alive?");
-                return -EINVAL;
-        }
+       if (!try_module_get(THIS_MODULE)) {
+               CERROR("Can't get module. Is it alive?");
+               return -EINVAL;
+       }
         switch (cmd) {
-        case LL_IOC_HSM_CT_START:
-                rc = mdc_ioc_hsm_ct_start(exp, karg);
-                GOTO(out, rc);
         case OBD_IOC_CHANGELOG_SEND:
                 rc = mdc_ioc_changelog_send(obd, karg);
                 GOTO(out, rc);
@@ -1343,17 +1848,37 @@ static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                 struct ioc_changelog *icc = karg;
                 struct changelog_setinfo cs =
                         {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
-                rc = obd_set_info_async(exp, strlen(KEY_CHANGELOG_CLEAR),
+                rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
                                         KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
                                         NULL);
                 GOTO(out, rc);
         }
-        case OBD_IOC_FID2PATH: {
-                rc = mdc_ioc_fid2path(exp, karg);
-                GOTO(out, rc);
-        }
+       case OBD_IOC_FID2PATH:
+               rc = mdc_ioc_fid2path(exp, karg);
+               GOTO(out, rc);
+       case LL_IOC_HSM_CT_START:
+               rc = mdc_ioc_hsm_ct_start(exp, karg);
+               /* ignore if it was already registered on this MDS. */
+               if (rc == -EEXIST)
+                       rc = 0;
+               GOTO(out, rc);
+       case LL_IOC_HSM_PROGRESS:
+               rc = mdc_ioc_hsm_progress(exp, karg);
+               GOTO(out, rc);
+       case LL_IOC_HSM_STATE_GET:
+               rc = mdc_ioc_hsm_state_get(exp, karg);
+               GOTO(out, rc);
+       case LL_IOC_HSM_STATE_SET:
+               rc = mdc_ioc_hsm_state_set(exp, karg);
+               GOTO(out, rc);
+       case LL_IOC_HSM_ACTION:
+               rc = mdc_ioc_hsm_current_action(exp, karg);
+               GOTO(out, rc);
+       case LL_IOC_HSM_REQUEST:
+               rc = mdc_ioc_hsm_request(exp, karg);
+               GOTO(out, rc);
         case OBD_IOC_CLIENT_RECOVER:
-                rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1);
+                rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
                 if (rc < 0)
                         GOTO(out, rc);
                 GOTO(out, rc = 0);
@@ -1362,7 +1887,8 @@ static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                 GOTO(out, rc);
         case OBD_IOC_PARSE: {
                 ctxt = llog_get_context(exp->exp_obd, LLOG_CONFIG_REPL_CTXT);
-                rc = class_config_parse_llog(ctxt, data->ioc_inlbuf1, NULL);
+               rc = class_config_parse_llog(NULL, ctxt, data->ioc_inlbuf1,
+                                            NULL);
                 llog_ctxt_put(ctxt);
                 GOTO(out, rc);
         }
@@ -1370,23 +1896,23 @@ static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
         case OBD_IOC_LLOG_INFO:
         case OBD_IOC_LLOG_PRINT: {
                 ctxt = llog_get_context(obd, LLOG_CONFIG_REPL_CTXT);
-                rc = llog_ioctl(ctxt, cmd, data);
+               rc = llog_ioctl(NULL, ctxt, cmd, data);
                 llog_ctxt_put(ctxt);
                 GOTO(out, rc);
         }
 #endif
         case OBD_IOC_POLL_QUOTACHECK:
-                rc = lquota_poll_check(quota_interface, exp,
-                                       (struct if_quotacheck *)karg);
+                rc = mdc_quota_poll_check(exp, (struct if_quotacheck *)karg);
                 GOTO(out, rc);
         case OBD_IOC_PING_TARGET:
                 rc = ptlrpc_obd_ping(obd);
                 GOTO(out, rc);
         /*
-         * Normally IOC_OBD_STATFS iocontrol is handled by LMV instead of MDC.
-         * But when the cluster is upgraded from 1.8, there'd be no LMV layer
-         * thus we might be called here. Eventually this code should be removed.
-         * bz20731.
+         * Normally IOC_OBD_STATFS, OBD_IOC_QUOTACTL iocontrol are handled by
+         * LMV instead of MDC. But when the cluster is upgraded from 1.8,
+         * there'd be no LMV layer thus we might be called here. Eventually
+         * this code should be removed.
+         * bz20731, LU-592.
          */
         case IOC_OBD_STATFS: {
                 struct obd_statfs stat_buf = {0};
@@ -1394,28 +1920,61 @@ static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                 if (*((__u32 *) data->ioc_inlbuf2) != 0)
                         GOTO(out, rc = -ENODEV);
 
-                rc = mdc_statfs(obd, &stat_buf,
-                                cfs_time_current_64() - CFS_HZ, 0);
-                if (rc != 0)
-                        GOTO(out, rc);
-
-                if (cfs_copy_to_user(data->ioc_pbuf1, &stat_buf,
-                                     data->ioc_plen1))
-                        GOTO(out, rc = -EFAULT);
-                if (cfs_copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
-                                     data->ioc_plen2))
+                /* copy UUID */
+               if (copy_to_user(data->ioc_pbuf2, obd2cli_tgt(obd),
+                                min((int)data->ioc_plen2,
+                                    (int)sizeof(struct obd_uuid))))
+                       GOTO(out, rc = -EFAULT);
+
+               rc = mdc_statfs(NULL, obd->obd_self_export, &stat_buf,
+                               cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
+                               0);
+               if (rc != 0)
+                       GOTO(out, rc);
+
+               if (copy_to_user(data->ioc_pbuf1, &stat_buf,
+                                     min((int) data->ioc_plen1,
+                                         (int) sizeof(stat_buf))))
                         GOTO(out, rc = -EFAULT);
 
                 GOTO(out, rc = 0);
         }
-        default:
-                CERROR("mdc_ioctl(): unrecognised ioctl %#x\n", cmd);
-                GOTO(out, rc = -ENOTTY);
-        }
+       case OBD_IOC_QUOTACTL: {
+               struct if_quotactl *qctl = karg;
+               struct obd_quotactl *oqctl;
+
+               OBD_ALLOC_PTR(oqctl);
+               if (oqctl == NULL)
+                       GOTO(out, rc = -ENOMEM);
+
+               QCTL_COPY(oqctl, qctl);
+               rc = obd_quotactl(exp, oqctl);
+               if (rc == 0) {
+                       QCTL_COPY(qctl, oqctl);
+                       qctl->qc_valid = QC_MDTIDX;
+                       qctl->obd_uuid = obd->u.cli.cl_target_uuid;
+               }
+
+               OBD_FREE_PTR(oqctl);
+               GOTO(out, rc);
+       }
+       case LL_IOC_GET_CONNECT_FLAGS:
+               if (copy_to_user(uarg, exp_connect_flags_ptr(exp),
+                                sizeof(*exp_connect_flags_ptr(exp))))
+                       GOTO(out, rc = -EFAULT);
+
+               GOTO(out, rc = 0);
+       case LL_IOC_LOV_SWAP_LAYOUTS:
+               rc = mdc_ioc_swap_layouts(exp, karg);
+               GOTO(out, rc);
+       default:
+               CERROR("unrecognised ioctl: cmd = %#x\n", cmd);
+               GOTO(out, rc = -ENOTTY);
+       }
 out:
-        cfs_module_put(THIS_MODULE);
+       module_put(THIS_MODULE);
 
-        return rc;
+       return rc;
 }
 
 int mdc_get_info_rpc(struct obd_export *exp,
@@ -1452,74 +2011,79 @@ int mdc_get_info_rpc(struct obd_export *exp,
                              RCL_SERVER, vallen);
         ptlrpc_request_set_replen(req);
 
-        rc = ptlrpc_queue_wait(req);
-        if (rc == 0) {
-                tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
-                memcpy(val, tmp, vallen);
-                if (ptlrpc_rep_need_swab(req)) {
-                        if (KEY_IS(KEY_FID2PATH)) {
-                                lustre_swab_fid2path(val);
-                        }
-                }
-        }
-        ptlrpc_req_finished(req);
-
-        RETURN(rc);
+       rc = ptlrpc_queue_wait(req);
+       /* -EREMOTE means the get_info result is partial, and it needs to
+        * continue on another MDT, see fid2path part in lmv_iocontrol */
+       if (rc == 0 || rc == -EREMOTE) {
+               tmp = req_capsule_server_get(&req->rq_pill, &RMF_GETINFO_VAL);
+               memcpy(val, tmp, vallen);
+               if (ptlrpc_rep_need_swab(req)) {
+                       if (KEY_IS(KEY_FID2PATH))
+                               lustre_swab_fid2path(val);
+               }
+       }
+       ptlrpc_req_finished(req);
+
+       RETURN(rc);
 }
 
 static void lustre_swab_hai(struct hsm_action_item *h)
 {
-        __swab32s(&h->hai_len);
-        __swab32s(&h->hai_action);
-        lustre_swab_lu_fid(&h->hai_fid);
-        __swab64s(&h->hai_cookie);
-        __swab64s(&h->hai_extent_start);
-        __swab64s(&h->hai_extent_end);
-        __swab64s(&h->hai_gid);
+       __swab32s(&h->hai_len);
+       __swab32s(&h->hai_action);
+       lustre_swab_lu_fid(&h->hai_fid);
+       lustre_swab_lu_fid(&h->hai_dfid);
+       __swab64s(&h->hai_cookie);
+       __swab64s(&h->hai_extent.offset);
+       __swab64s(&h->hai_extent.length);
+       __swab64s(&h->hai_gid);
 }
 
 static void lustre_swab_hal(struct hsm_action_list *h)
 {
-        struct hsm_action_item *hai;
-        int i;
+       struct hsm_action_item  *hai;
+       int                      i;
+
+       __swab32s(&h->hal_version);
+       __swab32s(&h->hal_count);
+       __swab32s(&h->hal_archive_id);
+       __swab64s(&h->hal_flags);
+       hai = hai_first(h);
+       for (i = 0; i < h->hal_count; i++, hai = hai_next(hai))
+               lustre_swab_hai(hai);
+}
 
-        __swab32s(&h->hal_version);
-        __swab32s(&h->hal_count);
-        __swab32s(&h->hal_archive_num);
-        hai = hai_zero(h);
-        for (i = 0; i < h->hal_count; i++) {
-                lustre_swab_hai(hai);
-                hai = hai_next(hai);
-        }
+static void lustre_swab_kuch(struct kuc_hdr *l)
+{
+        __swab16s(&l->kuc_magic);
+        /* __u8 l->kuc_transport */
+        __swab16s(&l->kuc_msgtype);
+        __swab16s(&l->kuc_msglen);
 }
 
 static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
-                                struct lustre_kernelcomm *lk)
+                               struct lustre_kernelcomm *lk)
 {
-        int rc = 0;
-
-        if (lk->lk_group != KUC_GRP_HSM) {
-                CERROR("Bad copytool group %d\n", lk->lk_group);
-                return -EINVAL;
-        }
-
-        CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
-               lk->lk_uid, lk->lk_group, lk->lk_flags);
-
-        if (lk->lk_flags & LK_FLG_STOP)
-                rc = libcfs_kkuc_group_rem(lk->lk_uid,lk->lk_group);
-        else {
-                cfs_file_t *fp = cfs_get_fd(lk->lk_wfd);
-                rc = libcfs_kkuc_group_add(fp, lk->lk_uid,lk->lk_group);
-                if (rc && fp)
-                        cfs_put_file(fp);
-        }
-
-        /* lk_data is archive number mask */
-        /* TODO: register archive num with mdt so coordinator can choose
-           correct agent. */
-
-        return rc;
+       struct obd_import  *imp = class_exp2cliimp(exp);
+       __u32               archive = lk->lk_data;
+       int                 rc = 0;
+
+       if (lk->lk_group != KUC_GRP_HSM) {
+               CERROR("Bad copytool group %d\n", lk->lk_group);
+               return -EINVAL;
+       }
+
+       CDEBUG(D_HSM, "CT start r%d w%d u%d g%d f%#x\n", lk->lk_rfd, lk->lk_wfd,
+              lk->lk_uid, lk->lk_group, lk->lk_flags);
+
+       if (lk->lk_flags & LK_FLG_STOP) {
+               /* Unregister with the coordinator */
+               rc = mdc_ioc_hsm_ct_unregister(imp);
+       } else {
+               rc = mdc_ioc_hsm_ct_register(imp, archive);
+       }
+
+       return rc;
 }
 
 /**
@@ -1529,79 +2093,97 @@ static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
  */
 static int mdc_hsm_copytool_send(int len, void *val)
 {
-        struct kuc_hdr *lh = (struct kuc_hdr *)val;
-        struct hsm_action_list *hal = (struct hsm_action_list *)(lh + 1);
-        int rc;
-        ENTRY;
+       struct kuc_hdr          *lh = (struct kuc_hdr *)val;
+       struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
+       int                      rc;
+       ENTRY;
+
+       if (len < sizeof(*lh) + sizeof(*hal)) {
+               CERROR("Short HSM message %d < %d\n", len,
+                      (int) (sizeof(*lh) + sizeof(*hal)));
+               RETURN(-EPROTO);
+       }
+       if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
+               lustre_swab_kuch(lh);
+               lustre_swab_hal(hal);
+       } else if (lh->kuc_magic != KUC_MAGIC) {
+               CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
+               RETURN(-EPROTO);
+       }
+
+       CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d "
+              "on %s\n",
+              lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
+              lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
+
+       /* Broadcast to HSM listeners */
+       rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
+
+       RETURN(rc);
+}
 
-        if (len < sizeof(*lh) + sizeof(*hal)) {
-                CERROR("Short HSM message %d < %d\n", len,
-                      (int) (sizeof(*lh) + sizeof(*hal)));
-                RETURN(-EPROTO);
-        }
-        if (lh->kuc_magic == __swab16(KUC_MAGIC)) {
-                lustre_swab_kuch(lh);
-                lustre_swab_hal(hal);
-        } else if (lh->kuc_magic != KUC_MAGIC) {
-                CERROR("Bad magic %x!=%x\n", lh->kuc_magic, KUC_MAGIC);
-                RETURN(-EPROTO);
-        }
+/**
+ * callback function passed to kuc for re-registering each HSM copytool
+ * running on MDC, after MDT shutdown/recovery.
+ * @param data copytool registration data
+ * @param cb_arg callback argument (obd_import)
+ */
+static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
+{
+       struct kkuc_ct_data     *kcd = data;
+       struct obd_import       *imp = (struct obd_import *)cb_arg;
+       int                      rc;
 
-        CDEBUG(D_HSM, " Received message mg=%x t=%d m=%d l=%d actions=%d\n",
-               lh->kuc_magic, lh->kuc_transport, lh->kuc_msgtype,
-               lh->kuc_msglen, hal->hal_count);
+       if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
+               return -EPROTO;
 
-        /* Broadcast to HSM listeners */
-        rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
+       if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
+               return 0;
 
-        RETURN(rc);
+       CDEBUG(D_HA, "%s: recover copytool registration to MDT (archive=%#x)\n",
+              imp->imp_obd->obd_name, kcd->kcd_archive);
+       rc = mdc_ioc_hsm_ct_register(imp, kcd->kcd_archive);
+
+       /* ignore error if the copytool is already registered */
+       return (rc == -EEXIST) ? 0 : rc;
 }
 
-int mdc_set_info_async(struct obd_export *exp,
-                       obd_count keylen, void *key,
-                       obd_count vallen, void *val,
-                       struct ptlrpc_request_set *set)
+/**
+ * Re-establish all kuc contexts with MDT
+ * after MDT shutdown/recovery.
+ */
+static int mdc_kuc_reregister(struct obd_import *imp)
 {
-        struct obd_import *imp = class_exp2cliimp(exp);
-        int                rc = -EINVAL;
-        ENTRY;
-
-        if (KEY_IS(KEY_INIT_RECOV)) {
-                if (vallen != sizeof(int))
-                        RETURN(-EINVAL);
-                cfs_spin_lock(&imp->imp_lock);
-                imp->imp_initial_recov = *(int *)val;
-                cfs_spin_unlock(&imp->imp_lock);
-                CDEBUG(D_HA, "%s: set imp_initial_recov = %d\n",
-                       exp->exp_obd->obd_name, imp->imp_initial_recov);
-                RETURN(0);
-        }
-        /* Turn off initial_recov after we try all backup servers once */
-        if (KEY_IS(KEY_INIT_RECOV_BACKUP)) {
-                if (vallen != sizeof(int))
-                        RETURN(-EINVAL);
-                cfs_spin_lock(&imp->imp_lock);
-                imp->imp_initial_recov_bk = *(int *)val;
-                if (imp->imp_initial_recov_bk)
-                        imp->imp_initial_recov = 1;
-                cfs_spin_unlock(&imp->imp_lock);
-                CDEBUG(D_HA, "%s: set imp_initial_recov_bk = %d\n",
-                       exp->exp_obd->obd_name, imp->imp_initial_recov_bk);
-                RETURN(0);
-        }
-        if (KEY_IS(KEY_READ_ONLY)) {
-                if (vallen != sizeof(int))
-                        RETURN(-EINVAL);
+       /* re-register HSM agents */
+       return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
+                                        (void *)imp);
+}
 
-                cfs_spin_lock(&imp->imp_lock);
-                if (*((int *)val)) {
-                        imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
-                        imp->imp_connect_data.ocd_connect_flags |= OBD_CONNECT_RDONLY;
-                } else {
-                        imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
-                        imp->imp_connect_data.ocd_connect_flags &= ~OBD_CONNECT_RDONLY;
-                }
-                cfs_spin_unlock(&imp->imp_lock);
+int mdc_set_info_async(const struct lu_env *env,
+                      struct obd_export *exp,
+                      obd_count keylen, void *key,
+                      obd_count vallen, void *val,
+                      struct ptlrpc_request_set *set)
+{
+       struct obd_import       *imp = class_exp2cliimp(exp);
+       int                      rc;
+       ENTRY;
+
+       if (KEY_IS(KEY_READ_ONLY)) {
+               if (vallen != sizeof(int))
+                       RETURN(-EINVAL);
+
+               spin_lock(&imp->imp_lock);
+               if (*((int *)val)) {
+                       imp->imp_connect_flags_orig |= OBD_CONNECT_RDONLY;
+                       imp->imp_connect_data.ocd_connect_flags |=
+                                                       OBD_CONNECT_RDONLY;
+               } else {
+                       imp->imp_connect_flags_orig &= ~OBD_CONNECT_RDONLY;
+                       imp->imp_connect_data.ocd_connect_flags &=
+                                                       ~OBD_CONNECT_RDONLY;
+               }
+               spin_unlock(&imp->imp_lock);
 
                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
                                        keylen, key, vallen, val, set);
@@ -1615,15 +2197,6 @@ int mdc_set_info_async(struct obd_export *exp,
                 sptlrpc_import_flush_my_ctx(imp);
                 RETURN(0);
         }
-        if (KEY_IS(KEY_MDS_CONN)) {
-                /* mds-mds import */
-                cfs_spin_lock(&imp->imp_lock);
-                imp->imp_server_timeout = 1;
-                cfs_spin_unlock(&imp->imp_lock);
-                imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
-                CDEBUG(D_OTHER, "%s: timeout / 2\n", exp->exp_obd->obd_name);
-                RETURN(0);
-        }
         if (KEY_IS(KEY_CHANGELOG_CLEAR)) {
                 rc = do_set_info_async(imp, MDS_SET_INFO, LUSTRE_MDS_VERSION,
                                        keylen, key, vallen, val, set);
@@ -1634,11 +2207,13 @@ int mdc_set_info_async(struct obd_export *exp,
                 RETURN(rc);
         }
 
-        RETURN(rc);
+       CERROR("Unknown key %s\n", (char *)key);
+       RETURN(-EINVAL);
 }
 
-int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
-                 __u32 *vallen, void *val, struct lov_stripe_md *lsm)
+int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
+                 __u32 keylen, void *key, __u32 *vallen, void *val,
+                 struct lov_stripe_md *lsm)
 {
         int rc = -EINVAL;
 
@@ -1653,8 +2228,7 @@ int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
                 max_easize = val;
                 *max_easize = exp->exp_obd->u.cli.cl_max_mds_easize;
                 RETURN(0);
-        }
-        if (KEY_IS(KEY_CONN_DATA)) {
+        } else if (KEY_IS(KEY_CONN_DATA)) {
                 struct obd_import *imp = class_exp2cliimp(exp);
                 struct obd_connect_data *data = val;
 
@@ -1663,6 +2237,9 @@ int mdc_get_info(struct obd_export *exp, __u32 keylen, void *key,
 
                 *data = imp->imp_connect_data;
                 RETURN(0);
+        } else if (KEY_IS(KEY_TGT_COUNT)) {
+                *((int *)val) = 1;
+                RETURN(0);
         }
 
         rc = mdc_get_info_rpc(exp, keylen, key, *vallen, val);
@@ -1757,8 +2334,8 @@ static int mdc_unpin(struct obd_export *exp, struct obd_client_handle *handle,
         RETURN(rc);
 }
 
-int mdc_sync(struct obd_export *exp, const struct lu_fid *fid,
-             struct obd_capa *oc, struct ptlrpc_request **request)
+int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
+             struct obd_capa *oc, struct ptlrpc_request **request)
 {
         struct ptlrpc_request *req;
         int                    rc;
@@ -1809,12 +2386,9 @@ static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
                 /*
                  * Flush current sequence to make client obtain new one
                  * from server in case of disconnect/reconnect.
-                 * If range is already empty then no need to flush it.
                  */
-                if (cli->cl_seq != NULL &&
-                    !range_is_exhausted(&cli->cl_seq->lcs_space)) {
+                if (cli->cl_seq != NULL)
                         seq_client_flush(cli->cl_seq);
-                }
 
                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
                 break;
@@ -1826,14 +2400,18 @@ static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
 
                 break;
         }
-        case IMP_EVENT_ACTIVE: {
-                rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
-                break;
-        }
+       case IMP_EVENT_ACTIVE:
+               rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
+               /* redo the kuc registration after reconnecting */
+               if (rc == 0)
+                       rc = mdc_kuc_reregister(imp);
+               break;
         case IMP_EVENT_OCD:
                 rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD, NULL);
                 break;
-
+        case IMP_EVENT_DEACTIVATE:
+        case IMP_EVENT_ACTIVATE:
+                break;
         default:
                 CERROR("Unknown import event %x\n", event);
                 LBUG();
@@ -1841,99 +2419,70 @@ static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
         RETURN(rc);
 }
 
-static int mdc_fid_init(struct obd_export *exp)
+int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
+                  struct md_op_data *op_data)
 {
         struct client_obd *cli = &exp->exp_obd->u.cli;
-        char *prefix;
-        int rc;
+        struct lu_client_seq *seq = cli->cl_seq;
         ENTRY;
-
-        OBD_ALLOC_PTR(cli->cl_seq);
-        if (cli->cl_seq == NULL)
-                RETURN(-ENOMEM);
-
-        OBD_ALLOC(prefix, MAX_OBD_NAME + 5);
-        if (prefix == NULL)
-                GOTO(out_free_seq, rc = -ENOMEM);
-
-        snprintf(prefix, MAX_OBD_NAME + 5, "srv-%s",
-                 exp->exp_obd->obd_name);
-
-        /* Init client side sequence-manager */
-        rc = seq_client_init(cli->cl_seq, exp,
-                             LUSTRE_SEQ_METADATA,
-                             prefix, NULL);
-        OBD_FREE(prefix, MAX_OBD_NAME + 5);
-        if (rc)
-                GOTO(out_free_seq, rc);
-
-        RETURN(rc);
-out_free_seq:
-        OBD_FREE_PTR(cli->cl_seq);
-        cli->cl_seq = NULL;
-        return rc;
+        RETURN(seq_client_alloc_fid(NULL, seq, fid));
 }
 
-static int mdc_fid_fini(struct obd_export *exp)
-{
+struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
         struct client_obd *cli = &exp->exp_obd->u.cli;
-        ENTRY;
-
-        if (cli->cl_seq != NULL) {
-                seq_client_fini(cli->cl_seq);
-                OBD_FREE_PTR(cli->cl_seq);
-                cli->cl_seq = NULL;
-        }
-
-        RETURN(0);
+        return &cli->cl_target_uuid;
 }
 
-int mdc_fid_alloc(struct obd_export *exp, struct lu_fid *fid,
-                  struct md_op_data *op_data)
+/**
+ * Determine whether the lock can be canceled before replaying it during
+ * recovery, non zero value will be return if the lock can be canceled,
+ * or zero returned for not
+ */
+static int mdc_cancel_for_recovery(struct ldlm_lock *lock)
 {
-        struct client_obd *cli = &exp->exp_obd->u.cli;
-        struct lu_client_seq *seq = cli->cl_seq;
-        ENTRY;
-        RETURN(seq_client_alloc_fid(seq, fid));
+        if (lock->l_resource->lr_type != LDLM_IBITS)
+                RETURN(0);
+
+        /* FIXME: if we ever get into a situation where there are too many
+         * opened files with open locks on a single node, then we really
+         * should replay these open locks to reget it */
+        if (lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_OPEN)
+                RETURN(0);
+
+        RETURN(1);
 }
 
-/* XXX This method is used only to clear current fid seq
- * once fld/mds insert failed */
-static int mdc_fid_delete(struct obd_export *exp, const struct lu_fid *fid)
+static int mdc_resource_inode_free(struct ldlm_resource *res)
 {
-        struct client_obd *cli = &exp->exp_obd->u.cli;
+       if (res->lr_lvb_inode)
+               res->lr_lvb_inode = NULL;
 
-        seq_client_flush(cli->cl_seq);
-        return 0;
+       return 0;
 }
 
-struct obd_uuid *mdc_get_uuid(struct obd_export *exp) {
-        struct client_obd *cli = &exp->exp_obd->u.cli;
-        return &cli->cl_target_uuid;
-}
+struct ldlm_valblock_ops inode_lvbo = {
+       lvbo_free: mdc_resource_inode_free
+};
 
 static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
 {
-        struct client_obd *cli = &obd->u.cli;
-        struct lprocfs_static_vars lvars = { 0 };
-        int rc;
-        ENTRY;
+       struct client_obd               *cli = &obd->u.cli;
+       struct lprocfs_static_vars      lvars = { 0 };
+       int                             rc;
+       ENTRY;
 
         OBD_ALLOC(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
         if (!cli->cl_rpc_lock)
                 RETURN(-ENOMEM);
         mdc_init_rpc_lock(cli->cl_rpc_lock);
 
-        ptlrpcd_addref();
-
-        OBD_ALLOC(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
-        if (!cli->cl_setattr_lock)
-                GOTO(err_rpc_lock, rc = -ENOMEM);
-        mdc_init_rpc_lock(cli->cl_setattr_lock);
+       rc = ptlrpcd_addref();
+       if (rc < 0)
+               GOTO(err_rpc_lock, rc);
 
         OBD_ALLOC(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
         if (!cli->cl_close_lock)
-                GOTO(err_setattr_lock, rc = -ENOMEM);
+                GOTO(err_ptlrpcd_decref, rc = -ENOMEM);
         mdc_init_rpc_lock(cli->cl_close_lock);
 
         rc = client_obd_setup(obd, cfg);
@@ -1941,9 +2490,14 @@ static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
                 GOTO(err_close_lock, rc);
         lprocfs_mdc_init_vars(&lvars);
         lprocfs_obd_setup(obd, lvars.obd_vars);
+       lprocfs_alloc_md_stats(obd, 0);
         sptlrpc_lprocfs_cliobd_attach(obd);
         ptlrpc_lprocfs_register_obd(obd);
 
+        ns_register_cancel(obd->obd_namespace, mdc_cancel_for_recovery);
+
+       obd->obd_namespace->ns_lvbo = &inode_lvbo;
+
         rc = obd_llog_init(obd, &obd->obd_olg, obd, NULL);
         if (rc) {
                 mdc_cleanup(obd);
@@ -1954,11 +2508,10 @@ static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
 
 err_close_lock:
         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
-err_setattr_lock:
-        OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
+err_ptlrpcd_decref:
+        ptlrpcd_decref();
 err_rpc_lock:
         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
-        ptlrpcd_decref();
         RETURN(rc);
 }
 
@@ -1994,22 +2547,15 @@ static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
         case OBD_CLEANUP_EARLY:
                 break;
         case OBD_CLEANUP_EXPORTS:
-                /* Failsafe, ok if racy */
-                if (obd->obd_type->typ_refcnt <= 1)
-                        libcfs_kkuc_group_rem(0, KUC_GRP_HSM);
-
-                /* If we set up but never connected, the
-                   client import will not have been cleaned. */
-                if (obd->u.cli.cl_import) {
-                        struct obd_import *imp;
-                        cfs_down_write(&obd->u.cli.cl_sem);
-                        imp = obd->u.cli.cl_import;
-                        CERROR("client import never connected\n");
-                        ptlrpc_invalidate_import(imp);
-                        class_destroy_import(imp);
-                        cfs_up_write(&obd->u.cli.cl_sem);
-                        obd->u.cli.cl_import = NULL;
-                }
+               /* Failsafe, ok if racy */
+               if (obd->obd_type->typ_refcnt <= 1)
+                       libcfs_kkuc_group_rem(0, KUC_GRP_HSM, NULL);
+
+                obd_cleanup_client_import(obd);
+                ptlrpc_lprocfs_unregister_obd(obd);
+                lprocfs_obd_cleanup(obd);
+               lprocfs_free_md_stats(obd);
+
                 rc = obd_llog_finish(obd, 0);
                 if (rc != 0)
                         CERROR("failed to cleanup llogging subsystems\n");
@@ -2023,11 +2569,8 @@ static int mdc_cleanup(struct obd_device *obd)
         struct client_obd *cli = &obd->u.cli;
 
         OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
-        OBD_FREE(cli->cl_setattr_lock, sizeof (*cli->cl_setattr_lock));
         OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
 
-        ptlrpc_lprocfs_unregister_obd(obd);
-        lprocfs_obd_cleanup(obd);
         ptlrpcd_decref();
 
         return client_obd_cleanup(obd);
@@ -2037,47 +2580,36 @@ static int mdc_cleanup(struct obd_device *obd)
 static int mdc_llog_init(struct obd_device *obd, struct obd_llog_group *olg,
                          struct obd_device *tgt, int *index)
 {
-        struct llog_ctxt *ctxt;
-        int rc;
-        ENTRY;
+       struct llog_ctxt        *ctxt;
+       int                      rc;
 
-        LASSERT(olg == &obd->obd_olg);
+       ENTRY;
 
-        rc = llog_setup(obd, olg, LLOG_LOVEA_REPL_CTXT, tgt, 0, NULL,
-                        &llog_client_ops);
-        if (rc)
-                RETURN(rc);
+       LASSERT(olg == &obd->obd_olg);
 
-        ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
-        llog_initiator_connect(ctxt);
-        llog_ctxt_put(ctxt);
+       rc = llog_setup(NULL, obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt,
+                       &llog_client_ops);
+       if (rc)
+               RETURN(rc);
 
-        rc = llog_setup(obd, olg, LLOG_CHANGELOG_REPL_CTXT, tgt, 0, NULL,
-                        &llog_client_ops);
-        if (rc == 0) {
-                ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
-                llog_initiator_connect(ctxt);
-                llog_ctxt_put(ctxt);
-        }
+       ctxt = llog_group_get_ctxt(olg, LLOG_CHANGELOG_REPL_CTXT);
+       llog_initiator_connect(ctxt);
+       llog_ctxt_put(ctxt);
 
-        RETURN(rc);
+       RETURN(0);
 }
 
 static int mdc_llog_finish(struct obd_device *obd, int count)
 {
-        struct llog_ctxt *ctxt;
-        int rc = 0;
-        ENTRY;
+       struct llog_ctxt *ctxt;
 
-        ctxt = llog_get_context(obd, LLOG_LOVEA_REPL_CTXT);
-        if (ctxt)
-                rc = llog_cleanup(ctxt);
+       ENTRY;
 
-        ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
-        if (ctxt)
-                rc = llog_cleanup(ctxt);
+       ctxt = llog_get_context(obd, LLOG_CHANGELOG_REPL_CTXT);
+       if (ctxt)
+               llog_cleanup(NULL, ctxt);
 
-        RETURN(rc);
+       RETURN(0);
 }
 
 static int mdc_process_config(struct obd_device *obd, obd_count len, void *buf)
@@ -2139,11 +2671,10 @@ int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
 }
 
 static int mdc_interpret_renew_capa(const struct lu_env *env,
-                                    struct ptlrpc_request *req, void *unused,
+                                    struct ptlrpc_request *req, void *args,
                                     int status)
 {
-        struct obd_capa *oc = req->rq_async_args.pointer_arg[0];
-        renew_capa_cb_t cb = req->rq_async_args.pointer_arg[1];
+        struct mdc_renew_capa_args *ra = args;
         struct mdt_body *body = NULL;
         struct lustre_capa *capa;
         ENTRY;
@@ -2163,7 +2694,7 @@ static int mdc_interpret_renew_capa(const struct lu_env *env,
                 GOTO(out, capa = ERR_PTR(-EFAULT));
         EXIT;
 out:
-        cb(oc, capa);
+        ra->ra_cb(ra->ra_oc, capa);
         return 0;
 }
 
@@ -2171,6 +2702,7 @@ static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
                           renew_capa_cb_t cb)
 {
         struct ptlrpc_request *req;
+        struct mdc_renew_capa_args *ra;
         ENTRY;
 
         req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
@@ -2184,34 +2716,15 @@ static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
         mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
         ptlrpc_request_set_replen(req);
 
-        req->rq_async_args.pointer_arg[0] = oc;
-        req->rq_async_args.pointer_arg[1] = cb;
+        CLASSERT(sizeof(*ra) <= sizeof(req->rq_async_args));
+        ra = ptlrpc_req_async_args(req);
+        ra->ra_oc = oc;
+        ra->ra_cb = cb;
         req->rq_interpret_reply = mdc_interpret_renew_capa;
-        ptlrpcd_add_req(req, PSCOPE_OTHER);
+        ptlrpcd_add_req(req, PDL_POLICY_LOCAL, -1);
         RETURN(0);
 }
 
-static int mdc_connect(const struct lu_env *env,
-                       struct obd_export **exp,
-                       struct obd_device *obd, struct obd_uuid *cluuid,
-                       struct obd_connect_data *data,
-                       void *localdata)
-{
-        struct obd_import *imp = obd->u.cli.cl_import;
-
-        /* mds-mds import features */
-        if (data && (data->ocd_connect_flags & OBD_CONNECT_MDS_MDS)) {
-                cfs_spin_lock(&imp->imp_lock);
-                imp->imp_server_timeout = 1;
-                cfs_spin_unlock(&imp->imp_lock);
-                imp->imp_client->cli_request_portal = MDS_MDS_PORTAL;
-                CDEBUG(D_OTHER, "%s: Set 'mds' portal and timeout\n",
-                       obd->obd_name);
-        }
-
-        return client_connect_import(env, exp, obd, cluuid, data, NULL);
-}
-
 struct obd_ops mdc_obd_ops = {
         .o_owner            = THIS_MODULE,
         .o_setup            = mdc_setup,
@@ -2219,28 +2732,29 @@ struct obd_ops mdc_obd_ops = {
         .o_cleanup          = mdc_cleanup,
         .o_add_conn         = client_import_add_conn,
         .o_del_conn         = client_import_del_conn,
-        .o_connect          = mdc_connect,
+        .o_connect          = client_connect_import,
         .o_disconnect       = client_disconnect_export,
         .o_iocontrol        = mdc_iocontrol,
         .o_set_info_async   = mdc_set_info_async,
         .o_statfs           = mdc_statfs,
         .o_pin              = mdc_pin,
         .o_unpin            = mdc_unpin,
-        .o_fid_init         = mdc_fid_init,
-        .o_fid_fini         = mdc_fid_fini,
+       .o_fid_init         = client_fid_init,
+       .o_fid_fini         = client_fid_fini,
         .o_fid_alloc        = mdc_fid_alloc,
-        .o_fid_delete       = mdc_fid_delete,
         .o_import_event     = mdc_import_event,
         .o_llog_init        = mdc_llog_init,
         .o_llog_finish      = mdc_llog_finish,
         .o_get_info         = mdc_get_info,
         .o_process_config   = mdc_process_config,
         .o_get_uuid         = mdc_get_uuid,
+        .o_quotactl         = mdc_quotactl,
+        .o_quotacheck       = mdc_quotacheck
 };
 
 struct md_ops mdc_md_ops = {
         .m_getstatus        = mdc_getstatus,
-        .m_change_cbdata    = mdc_change_cbdata,
+        .m_null_inode      = mdc_null_inode,
         .m_find_cbdata      = mdc_find_cbdata,
         .m_close            = mdc_close,
         .m_create           = mdc_create,
@@ -2255,7 +2769,7 @@ struct md_ops mdc_md_ops = {
         .m_setattr          = mdc_setattr,
         .m_setxattr         = mdc_setxattr,
         .m_getxattr         = mdc_getxattr,
-        .m_sync             = mdc_sync,
+       .m_fsync            = mdc_fsync,
         .m_readpage         = mdc_readpage,
         .m_unlink           = mdc_unlink,
         .m_cancel_unused    = mdc_cancel_unused,
@@ -2279,24 +2793,14 @@ int __init mdc_init(void)
         struct lprocfs_static_vars lvars = { 0 };
         lprocfs_mdc_init_vars(&lvars);
 
-        cfs_request_module("lquota");
-        quota_interface = PORTAL_SYMBOL_GET(mdc_quota_interface);
-        init_obd_quota_ops(quota_interface, &mdc_obd_ops);
-
         rc = class_register_type(&mdc_obd_ops, &mdc_md_ops, lvars.module_vars,
                                  LUSTRE_MDC_NAME, NULL);
-        if (rc && quota_interface)
-                PORTAL_SYMBOL_PUT(mdc_quota_interface);
-
         RETURN(rc);
 }
 
 #ifdef __KERNEL__
 static void /*__exit*/ mdc_exit(void)
 {
-        if (quota_interface)
-                PORTAL_SYMBOL_PUT(mdc_quota_interface);
-
         class_unregister_type(LUSTRE_MDC_NAME);
 }