Whamcloud - gitweb
LU-8066 mdc: move mdc-specific procfs files to sysfs
[fs/lustre-release.git] / lustre / mdc / mdc_request.c
index ffd5d57..daf1e21 100644 (file)
  *
  * You should have received a copy of the GNU General Public License
  * version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
  *
  * GPL HEADER END
  */
@@ -27,7 +23,7 @@
  * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2011, 2013, Intel Corporation.
+ * Copyright (c) 2011, 2017, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
 
 #define DEBUG_SUBSYSTEM S_MDC
 
+#include <linux/init.h>
+#include <linux/kthread.h>
 #include <linux/module.h>
 #include <linux/pagemap.h>
-#include <linux/miscdevice.h>
-#include <linux/init.h>
+#include <linux/user_namespace.h>
 #include <linux/utsname.h>
+#ifdef HAVE_UIDGID_HEADER
+# include <linux/uidgid.h>
+#endif
 
+#include <lustre_errno.h>
+
+#include <cl_object.h>
+#include <llog_swab.h>
+#include <lprocfs_status.h>
 #include <lustre_acl.h>
-#include <lustre_ioctl.h>
-#include <obd_class.h>
-#include <lustre_lmv.h>
 #include <lustre_fid.h>
-#include <lprocfs_status.h>
-#include <lustre_param.h>
+#include <uapi/linux/lustre/lustre_ioctl.h>
+#include <lustre_kernelcomm.h>
+#include <lustre_lmv.h>
 #include <lustre_log.h>
-#include <cl_object.h>
+#include <uapi/linux/lustre/lustre_param.h>
+#include <lustre_swab.h>
+#include <obd_class.h>
+#include <lustre_osc.h>
 
 #include "mdc_internal.h"
 
 #define REQUEST_MINOR 244
 
-struct mdc_renew_capa_args {
-        struct obd_capa        *ra_oc;
-        renew_capa_cb_t         ra_cb;
-};
-
 static int mdc_cleanup(struct obd_device *obd);
 
-static int mdc_unpack_capa(struct obd_export *exp, struct ptlrpc_request *req,
-                          const struct req_msg_field *field,
-                          struct obd_capa **oc)
-{
-        struct lustre_capa *capa;
-        struct obd_capa *c;
-        ENTRY;
-
-        /* swabbed already in mdc_enqueue */
-        capa = req_capsule_server_get(&req->rq_pill, field);
-        if (capa == NULL)
-                RETURN(-EPROTO);
-
-        c = alloc_capa(CAPA_SITE_CLIENT);
-        if (IS_ERR(c)) {
-                CDEBUG(D_INFO, "alloc capa failed!\n");
-                RETURN(PTR_ERR(c));
-        } else {
-                c->c_capa = *capa;
-                *oc = c;
-                RETURN(0);
-        }
-}
-
 static inline int mdc_queue_wait(struct ptlrpc_request *req)
 {
        struct client_obd *cli = &req->rq_import->imp_obd->u.cli;
@@ -105,56 +82,72 @@ static inline int mdc_queue_wait(struct ptlrpc_request *req)
        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,
-                          struct obd_capa **pc, int level, int msg_flags)
+/*
+ * Send MDS_GET_ROOT RPC to fetch root FID.
+ *
+ * If \a fileset is not NULL it should contain a subdirectory off
+ * the ROOT/ directory to be mounted on the client. Return the FID
+ * of the subdirectory to the client to mount onto its mountpoint.
+ *
+ * \param[in]  imp     MDC import
+ * \param[in]  fileset fileset name, which could be NULL
+ * \param[out] rootfid root FID of this mountpoint
+ * \param[out] pc      root capa will be unpacked and saved in this pointer
+ *
+ * \retval     0 on success, negative errno on failure
+ */
+static int mdc_get_root(struct obd_export *exp, const char *fileset,
+                        struct lu_fid *rootfid)
 {
-        struct ptlrpc_request *req;
-        struct mdt_body       *body;
-        int                    rc;
-        ENTRY;
-
-        req = ptlrpc_request_alloc_pack(imp, &RQF_MDS_GETSTATUS,
-                                        LUSTRE_MDS_VERSION, MDS_GETSTATUS);
-        if (req == NULL)
-                RETURN(-ENOMEM);
+       struct ptlrpc_request   *req;
+       struct mdt_body         *body;
+       int                      rc;
 
-        mdc_pack_body(req, NULL, NULL, 0, 0, -1, 0);
-        lustre_msg_add_flags(req->rq_reqmsg, msg_flags);
-        req->rq_send_state = level;
+       ENTRY;
 
-        ptlrpc_request_set_replen(req);
+       if (fileset && !(exp_connect_flags(exp) & OBD_CONNECT_SUBTREE))
+               RETURN(-ENOTSUPP);
 
-        rc = ptlrpc_queue_wait(req);
-        if (rc)
-                GOTO(out, rc);
+       req = ptlrpc_request_alloc(class_exp2cliimp(exp),
+                               &RQF_MDS_GET_ROOT);
+       if (req == NULL)
+               RETURN(-ENOMEM);
 
-        body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
-        if (body == NULL)
-                GOTO(out, rc = -EPROTO);
+       if (fileset != NULL)
+               req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
+                                    strlen(fileset) + 1);
+       rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_ROOT);
+       if (rc) {
+               ptlrpc_request_free(req);
+               RETURN(rc);
+       }
+       mdc_pack_body(req, NULL, 0, 0, -1, 0);
+       if (fileset != NULL) {
+               char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
 
-       if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
-               rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, pc);
-               if (rc)
-                       GOTO(out, rc);
+               memcpy(name, fileset, strlen(fileset));
        }
+       lustre_msg_add_flags(req->rq_reqmsg, LUSTRE_IMP_FULL);
+       req->rq_send_state = LUSTRE_IMP_FULL;
+
+       ptlrpc_request_set_replen(req);
+
+       rc = ptlrpc_queue_wait(req);
+       if (rc)
+               GOTO(out, rc);
+
+       body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
+       if (body == NULL)
+               GOTO(out, rc = -EPROTO);
 
        *rootfid = body->mbo_fid1;
-       CDEBUG(D_NET, "root fid="DFID", last_committed="LPU64"\n",
+       CDEBUG(D_NET, "root fid="DFID", last_committed=%llu\n",
               PFID(rootfid), lustre_msg_get_last_committed(req->rq_repmsg));
-        EXIT;
+       EXIT;
 out:
-        ptlrpc_req_finished(req);
-        return rc;
-}
+       ptlrpc_req_finished(req);
 
-/* This should be mdc_get_info("rootfid") */
-static int mdc_getstatus(struct obd_export *exp, struct lu_fid *rootfid,
-                        struct obd_capa **pc)
-{
-        return send_getstatus(class_exp2cliimp(exp), rootfid, pc,
-                              LUSTRE_IMP_FULL, 0);
+       return rc;
 }
 
 /*
@@ -196,85 +189,85 @@ static int mdc_getattr_common(struct obd_export *exp,
                        RETURN(-EPROTO);
        }
 
-       if (body->mbo_valid & OBD_MD_FLRMTPERM) {
-                struct mdt_remote_perm *perm;
-
-                LASSERT(client_is_remote(exp));
-                perm = req_capsule_server_swab_get(pill, &RMF_ACL,
-                                                lustre_swab_mdt_remote_perm);
-                if (perm == NULL)
-                        RETURN(-EPROTO);
-        }
-
-       if (body->mbo_valid & OBD_MD_FLMDSCAPA) {
-                struct lustre_capa *capa;
-                capa = req_capsule_server_get(pill, &RMF_CAPA1);
-                if (capa == NULL)
-                        RETURN(-EPROTO);
-        }
-
         RETURN(0);
 }
 
+static void mdc_reset_acl_req(struct ptlrpc_request *req)
+{
+       spin_lock(&req->rq_early_free_lock);
+       sptlrpc_cli_free_repbuf(req);
+       req->rq_repbuf = NULL;
+       req->rq_repbuf_len = 0;
+       req->rq_repdata = NULL;
+       req->rq_reqdata_len = 0;
+       spin_unlock(&req->rq_early_free_lock);
+}
+
 static int mdc_getattr(struct obd_export *exp, struct md_op_data *op_data,
                       struct ptlrpc_request **request)
 {
-        struct ptlrpc_request *req;
-        int                    rc;
-        ENTRY;
+       struct ptlrpc_request *req;
+       struct obd_import *imp = class_exp2cliimp(exp);
+       __u32 acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
+       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);
+
+       *request = NULL;
+       req = ptlrpc_request_alloc(imp, &RQF_MDS_GETATTR);
         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_GETATTR);
         if (rc) {
                 ptlrpc_request_free(req);
                 RETURN(rc);
         }
 
-        mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
-                      op_data->op_valid, op_data->op_mode, -1, 0);
+again:
+       mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
+                     op_data->op_mode, -1, 0);
+       req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
+       req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
+                            op_data->op_mode);
+       ptlrpc_request_set_replen(req);
 
-        req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
-                             op_data->op_mode);
-        if (op_data->op_valid & OBD_MD_FLRMTPERM) {
-                LASSERT(client_is_remote(exp));
-                req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
-                                     sizeof(struct mdt_remote_perm));
-        }
-        ptlrpc_request_set_replen(req);
+       rc = mdc_getattr_common(exp, req);
+       if (rc) {
+               if (rc == -ERANGE &&
+                   acl_bufsize != imp->imp_connect_data.ocd_max_easize) {
+                       acl_bufsize = imp->imp_connect_data.ocd_max_easize;
+                       mdc_reset_acl_req(req);
+                       goto again;
+               }
 
-        rc = mdc_getattr_common(exp, req);
-        if (rc)
-                ptlrpc_req_finished(req);
-        else
-                *request = req;
-        RETURN(rc);
+               ptlrpc_req_finished(req);
+       } else {
+               *request = req;
+       }
+
+       RETURN(rc);
 }
 
 static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
                            struct ptlrpc_request **request)
 {
-        struct ptlrpc_request *req;
-        int                    rc;
-        ENTRY;
+       struct ptlrpc_request *req;
+       struct obd_import *imp = class_exp2cliimp(exp);
+       __u32 acl_bufsize = LUSTRE_POSIX_ACL_MAX_SIZE_OLD;
+       int rc;
+       ENTRY;
 
-        *request = NULL;
-        req = ptlrpc_request_alloc(class_exp2cliimp(exp),
-                                   &RQF_MDS_GETATTR_NAME);
+       *request = NULL;
+       req = ptlrpc_request_alloc(imp, &RQF_MDS_GETATTR_NAME);
         if (req == NULL)
                 RETURN(-ENOMEM);
 
-        mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
         req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
                              op_data->op_namelen + 1);
 
@@ -284,10 +277,6 @@ static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
                 RETURN(rc);
         }
 
-        mdc_pack_body(req, &op_data->op_fid1, op_data->op_capa1,
-                      op_data->op_valid, op_data->op_mode,
-                      op_data->op_suppgids[0], 0);
-
         if (op_data->op_name) {
                 char *name = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
                 LASSERT(strnlen(op_data->op_name, op_data->op_namelen) ==
@@ -295,24 +284,36 @@ static int mdc_getattr_name(struct obd_export *exp, struct md_op_data *op_data,
                 memcpy(name, op_data->op_name, op_data->op_namelen);
         }
 
-        req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
-                             op_data->op_mode);
-        ptlrpc_request_set_replen(req);
+again:
+       mdc_pack_body(req, &op_data->op_fid1, op_data->op_valid,
+                     op_data->op_mode, op_data->op_suppgids[0], 0);
+       req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
+                            op_data->op_mode);
+       req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER, acl_bufsize);
+       ptlrpc_request_set_replen(req);
 
-        rc = mdc_getattr_common(exp, req);
-        if (rc)
-                ptlrpc_req_finished(req);
-        else
-                *request = req;
-        RETURN(rc);
+       rc = mdc_getattr_common(exp, req);
+       if (rc) {
+               if (rc == -ERANGE &&
+                   acl_bufsize != imp->imp_connect_data.ocd_max_easize) {
+                       acl_bufsize = imp->imp_connect_data.ocd_max_easize;
+                       mdc_reset_acl_req(req);
+                       goto again;
+               }
+
+               ptlrpc_req_finished(req);
+       } else {
+               *request = req;
+       }
+
+       RETURN(rc);
 }
 
 static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
-                            const struct lu_fid *fid,
-                            struct obd_capa *oc, int opcode, obd_valid valid,
-                            const char *xattr_name, const char *input,
-                            int input_size, int output_size, int flags,
-                            __u32 suppgid, struct ptlrpc_request **request)
+                           const struct lu_fid *fid, int opcode, u64 valid,
+                           const char *xattr_name, const char *input,
+                           int input_size, int output_size, int flags,
+                           __u32 suppgid, struct ptlrpc_request **request)
 {
         struct ptlrpc_request *req;
         int   xattr_namelen = 0;
@@ -325,7 +326,6 @@ static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
         if (req == NULL)
                 RETURN(-ENOMEM);
 
-        mdc_set_capa_size(req, &RMF_CAPA1, oc);
         if (xattr_name) {
                 xattr_namelen = strlen(xattr_name) + 1;
                 req_capsule_set_size(&req->rq_pill, &RMF_NAME, RCL_CLIENT,
@@ -365,11 +365,11 @@ static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
                }
        }
 
-        if (opcode == MDS_REINT) {
-                struct mdt_rec_setxattr *rec;
+       if (opcode == MDS_REINT) {
+               struct mdt_rec_setxattr *rec;
 
-                CLASSERT(sizeof(struct mdt_rec_setxattr) ==
-                         sizeof(struct mdt_rec_reint));
+               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;
                rec->sx_fsuid  = from_kuid(&init_user_ns, current_fsuid());
@@ -379,14 +379,12 @@ static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
                 rec->sx_suppgid2 = -1;
                 rec->sx_fid    = *fid;
                 rec->sx_valid  = valid | OBD_MD_FLCTIME;
-                rec->sx_time   = cfs_time_current_sec();
+               rec->sx_time   = ktime_get_real_seconds();
                 rec->sx_size   = output_size;
                 rec->sx_flags  = flags;
-
-                mdc_pack_capa(req, &RMF_CAPA1, oc);
-        } else {
-                mdc_pack_body(req, fid, oc, valid, output_size, suppgid, flags);
-        }
+       } else {
+               mdc_pack_body(req, fid, valid, output_size, suppgid, flags);
+       }
 
         if (xattr_name) {
                 tmp = req_capsule_client_get(&req->rq_pill, &RMF_NAME);
@@ -404,12 +402,12 @@ static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 
         /* make rpc */
         if (opcode == MDS_REINT)
-                mdc_get_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
+               mdc_get_mod_rpc_slot(req, NULL);
 
         rc = ptlrpc_queue_wait(req);
 
-        if (opcode == MDS_REINT)
-                mdc_put_rpc_lock(exp->exp_obd->u.cli.cl_rpc_lock, NULL);
+       if (opcode == MDS_REINT)
+               mdc_put_mod_rpc_slot(req, NULL);
 
         if (rc)
                 ptlrpc_req_finished(req);
@@ -419,28 +417,30 @@ static int mdc_xattr_common(struct obd_export *exp,const struct req_format *fmt,
 }
 
 static int mdc_setxattr(struct obd_export *exp, const struct lu_fid *fid,
-                       struct obd_capa *oc, obd_valid valid,
-                       const char *xattr_name,
-                       const char *input, int input_size, int output_size,
-                       int flags, __u32 suppgid,
-                       struct ptlrpc_request **request)
+                       u64 obd_md_valid, const char *name,
+                       const void *value, size_t value_size,
+                       unsigned int xattr_flags, u32 suppgid,
+                       struct ptlrpc_request **req)
 {
-        return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
-                                fid, oc, MDS_REINT, valid, xattr_name,
-                                input, input_size, output_size, flags,
-                                suppgid, request);
+       LASSERT(obd_md_valid == OBD_MD_FLXATTR ||
+               obd_md_valid == OBD_MD_FLXATTRRM);
+
+       return mdc_xattr_common(exp, &RQF_MDS_REINT_SETXATTR,
+                               fid, MDS_REINT, obd_md_valid, name,
+                               value, value_size, 0, xattr_flags, suppgid,
+                               req);
 }
 
 static int mdc_getxattr(struct obd_export *exp, const struct lu_fid *fid,
-                       struct obd_capa *oc, obd_valid valid,
-                       const char *xattr_name,
-                       const char *input, int input_size, int output_size,
-                       int flags, struct ptlrpc_request **request)
+                       u64 obd_md_valid, const char *name, size_t buf_size,
+                       struct ptlrpc_request **req)
 {
-        return mdc_xattr_common(exp, &RQF_MDS_GETXATTR,
-                                fid, oc, MDS_GETXATTR, valid, xattr_name,
-                                input, input_size, output_size, flags,
-                                -1, request);
+       LASSERT(obd_md_valid == OBD_MD_FLXATTR ||
+               obd_md_valid == OBD_MD_FLXATTRLS);
+
+       return mdc_xattr_common(exp, &RQF_MDS_GETXATTR, fid, MDS_GETXATTR,
+                               obd_md_valid, name, NULL, 0, buf_size, 0, -1,
+                               req);
 }
 
 #ifdef CONFIG_FS_POSIX_ACL
@@ -470,7 +470,7 @@ static int mdc_unpack_acl(struct ptlrpc_request *req, struct lustre_md *md)
                 RETURN(rc);
         }
 
-        rc = posix_acl_valid(acl);
+        rc = posix_acl_valid(&init_user_ns, acl);
         if (rc) {
                 CERROR("validate acl: %d\n", rc);
                 posix_acl_release(acl);
@@ -499,9 +499,6 @@ int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
         LASSERT(md->body != NULL);
 
        if (md->body->mbo_valid & OBD_MD_FLEASIZE) {
-               int lmmsize;
-               struct lov_mds_md *lmm;
-
                if (!S_ISREG(md->body->mbo_mode)) {
                        CDEBUG(D_INFO, "OBD_MD_FLEASIZE set, should be a "
                               "regular file, but is not\n");
@@ -514,25 +511,15 @@ int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
                        GOTO(out, rc = -EPROTO);
                }
 
-               lmmsize = md->body->mbo_eadatasize;
-                lmm = req_capsule_server_sized_get(pill, &RMF_MDT_MD, lmmsize);
-                if (!lmm)
-                        GOTO(out, rc = -EPROTO);
-
-                rc = obd_unpackmd(dt_exp, &md->lsm, lmm, lmmsize);
-                if (rc < 0)
-                        GOTO(out, rc);
-
-               if (rc < (typeof(rc))sizeof(*md->lsm)) {
-                        CDEBUG(D_INFO, "lsm size too small: "
-                               "rc < sizeof (*md->lsm) (%d < %d)\n",
-                               rc, (int)sizeof(*md->lsm));
-                        GOTO(out, rc = -EPROTO);
-                }
-
+               md->layout.lb_len = md->body->mbo_eadatasize;
+               md->layout.lb_buf = req_capsule_server_sized_get(pill,
+                                                       &RMF_MDT_MD,
+                                                       md->layout.lb_len);
+               if (md->layout.lb_buf == NULL)
+                       GOTO(out, rc = -EPROTO);
        } else if (md->body->mbo_valid & OBD_MD_FLDIREA) {
-               int lmvsize;
-               struct lov_mds_md *lmv;
+               const union lmv_mds_md *lmv;
+               size_t lmv_size;
 
                if (!S_ISDIR(md->body->mbo_mode)) {
                        CDEBUG(D_INFO, "OBD_MD_FLDIREA set, should be a "
@@ -540,21 +527,20 @@ int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
                        GOTO(out, rc = -EPROTO);
                }
 
-               if (md->body->mbo_eadatasize == 0) {
+               lmv_size = md->body->mbo_eadatasize;
+               if (lmv_size == 0) {
                        CDEBUG(D_INFO, "OBD_MD_FLDIREA is set, "
                               "but eadatasize 0\n");
                        RETURN(-EPROTO);
                }
 
                if (md->body->mbo_valid & OBD_MD_MEA) {
-                       lmvsize = md->body->mbo_eadatasize;
                        lmv = req_capsule_server_sized_get(pill, &RMF_MDT_MD,
-                                                          lmvsize);
-                       if (!lmv)
+                                                          lmv_size);
+                       if (lmv == NULL)
                                GOTO(out, rc = -EPROTO);
 
-                       rc = obd_unpackmd(md_exp, (void *)&md->lmv, lmv,
-                                         lmvsize);
+                       rc = md_unpackmd(md_exp, &md->lmv, lmv, lmv_size);
                        if (rc < 0)
                                GOTO(out, rc);
 
@@ -568,14 +554,7 @@ int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
         }
         rc = 0;
 
-       if (md->body->mbo_valid & OBD_MD_FLRMTPERM) {
-               /* remote permission */
-               LASSERT(client_is_remote(exp));
-               md->remote_perm = req_capsule_server_swab_get(pill, &RMF_ACL,
-                                               lustre_swab_mdt_remote_perm);
-               if (!md->remote_perm)
-                       GOTO(out, rc = -EPROTO);
-       } else if (md->body->mbo_valid & OBD_MD_FLACL) {
+       if (md->body->mbo_valid & OBD_MD_FLACL) {
                /* for ACL, it's possible that FLACL is set but aclsize is zero.
                 * only when aclsize != 0 there's an actual segment for ACL
                 * in reply buffer.
@@ -590,40 +569,13 @@ int mdc_get_lustre_md(struct obd_export *exp, struct ptlrpc_request *req,
 #endif
                 }
         }
-       if (md->body->mbo_valid & OBD_MD_FLMDSCAPA) {
-                struct obd_capa *oc = NULL;
-
-                rc = mdc_unpack_capa(NULL, req, &RMF_CAPA1, &oc);
-                if (rc)
-                        GOTO(out, rc);
-                md->mds_capa = oc;
-        }
-
-       if (md->body->mbo_valid & OBD_MD_FLOSSCAPA) {
-                struct obd_capa *oc = NULL;
-
-                rc = mdc_unpack_capa(NULL, req, &RMF_CAPA2, &oc);
-                if (rc)
-                        GOTO(out, rc);
-                md->oss_capa = oc;
-        }
 
         EXIT;
 out:
         if (rc) {
-                if (md->oss_capa) {
-                        capa_put(md->oss_capa);
-                        md->oss_capa = NULL;
-                }
-                if (md->mds_capa) {
-                        capa_put(md->mds_capa);
-                        md->mds_capa = NULL;
-                }
 #ifdef CONFIG_FS_POSIX_ACL
                 posix_acl_release(md->posix_acl);
 #endif
-                if (md->lsm)
-                        obd_free_memmd(dt_exp, &md->lsm);
         }
         return rc;
 }
@@ -634,10 +586,6 @@ int mdc_free_lustre_md(struct obd_export *exp, struct lustre_md *md)
         RETURN(0);
 }
 
-/**
- * Handles both OPEN and SETATTR RPCs for OPEN-CLOSE and SETATTR-DONE_WRITING
- * RPC chains.
- */
 void mdc_replay_open(struct ptlrpc_request *req)
 {
         struct md_open_data *mod = req->rq_cb_data;
@@ -657,34 +605,43 @@ void mdc_replay_open(struct ptlrpc_request *req)
         body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
         LASSERT(body != NULL);
 
-        och = mod->mod_och;
-        if (och != NULL) {
-                struct lustre_handle *file_fh;
+       spin_lock(&req->rq_lock);
+       och = mod->mod_och;
+       if (och && och->och_fh.cookie)
+               req->rq_early_free_repbuf = 1;
+       else
+               req->rq_early_free_repbuf = 0;
+       spin_unlock(&req->rq_lock);
+
+       if (req->rq_early_free_repbuf) {
+               struct lustre_handle *file_fh;
 
-                LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
+               LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
 
-                file_fh = &och->och_fh;
-               CDEBUG(D_HA, "updating handle from "LPX64" to "LPX64"\n",
+               file_fh = &och->och_fh;
+               CDEBUG(D_HA, "updating handle from %#llx to %#llx\n",
                       file_fh->cookie, body->mbo_handle.cookie);
                old = *file_fh;
                *file_fh = body->mbo_handle;
-        }
-        close_req = mod->mod_close_req;
-        if (close_req != NULL) {
-                __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
-                struct mdt_ioepoch *epoch;
-
-                LASSERT(opc == MDS_CLOSE || opc == MDS_DONE_WRITING);
-                epoch = req_capsule_client_get(&close_req->rq_pill,
-                                               &RMF_MDT_EPOCH);
-                LASSERT(epoch);
-
-                if (och != NULL)
-                        LASSERT(!memcmp(&old, &epoch->handle, sizeof(old)));
-                DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
-               epoch->handle = body->mbo_handle;
-        }
-        EXIT;
+       }
+
+       close_req = mod->mod_close_req;
+       if (close_req) {
+               __u32 opc = lustre_msg_get_opc(close_req->rq_reqmsg);
+               struct mdt_ioepoch *epoch;
+
+               LASSERT(opc == MDS_CLOSE);
+               epoch = req_capsule_client_get(&close_req->rq_pill,
+                                              &RMF_MDT_EPOCH);
+               LASSERT(epoch);
+
+               if (req->rq_early_free_repbuf)
+                       LASSERT(!memcmp(&old, &epoch->mio_handle, sizeof(old)));
+
+               DEBUG_REQ(D_HA, close_req, "updating close body with new fh");
+               epoch->mio_handle = body->mbo_handle;
+       }
+       EXIT;
 }
 
 void mdc_commit_open(struct ptlrpc_request *req)
@@ -722,7 +679,7 @@ int mdc_set_open_replay_data(struct obd_export *exp,
        struct md_open_data     *mod;
        struct mdt_rec_create   *rec;
        struct mdt_body         *body;
-       struct ptlrpc_request   *open_req = it->d.lustre.it_data;
+       struct ptlrpc_request   *open_req = it->it_request;
        struct obd_import       *imp = open_req->rq_import;
        ENTRY;
 
@@ -762,11 +719,11 @@ int mdc_set_open_replay_data(struct obd_export *exp,
                mod->mod_open_req = open_req;
                open_req->rq_cb_data = mod;
                open_req->rq_commit_cb = mdc_commit_open;
+               open_req->rq_early_free_repbuf = 1;
                spin_unlock(&open_req->rq_lock);
         }
 
        rec->cr_fid2 = body->mbo_fid1;
-       rec->cr_ioepoch = body->mbo_ioepoch;
        rec->cr_old_handle.cookie = body->mbo_handle.cookie;
        open_req->rq_replay_cb = mdc_replay_open;
        if (!fid_is_sane(&body->mbo_fid1)) {
@@ -787,9 +744,15 @@ static void mdc_free_open(struct md_open_data *mod)
            imp_connect_disp_stripe(mod->mod_open_req->rq_import))
                committed = 1;
 
-       LASSERT(mod->mod_open_req->rq_replay == 0);
+       /**
+        * No reason to asssert here if the open request has
+        * rq_replay == 1. It means that mdc_close failed, and
+        * close request wasn`t sent. It is not fatal to client.
+        * The worst thing is eviction if the client gets open lock
+        **/
 
-       DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request\n");
+       DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "free open request rq_replay"
+                 "= %d\n", mod->mod_open_req->rq_replay);
 
        ptlrpc_request_committed(mod->mod_open_req, committed);
        if (mod->mod_close_req)
@@ -809,8 +772,14 @@ int mdc_clear_open_replay_data(struct obd_export *exp,
         if (mod == NULL)
                 RETURN(0);
 
-        LASSERT(mod != LP_POISON);
+       LASSERT(mod != LP_POISON);
        LASSERT(mod->mod_open_req != NULL);
+
+       spin_lock(&mod->mod_open_req->rq_lock);
+       if (mod->mod_och)
+               mod->mod_och->och_fh.cookie = 0;
+       mod->mod_open_req->rq_early_free_repbuf = 0;
+       spin_unlock(&mod->mod_open_req->rq_lock);
        mdc_free_open(mod);
 
         mod->mod_och = NULL;
@@ -820,58 +789,88 @@ int mdc_clear_open_replay_data(struct obd_export *exp,
         RETURN(0);
 }
 
-/* Prepares the request for the replay by the given reply */
-static void mdc_close_handle_reply(struct ptlrpc_request *req,
-                                   struct md_op_data *op_data, int rc) {
-        struct mdt_body  *repbody;
-        struct mdt_ioepoch *epoch;
-
-        if (req && rc == -EAGAIN) {
-                repbody = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
-                epoch = req_capsule_client_get(&req->rq_pill, &RMF_MDT_EPOCH);
-
-                epoch->flags |= MF_SOM_AU;
-               if (repbody->mbo_valid & OBD_MD_FLGETATTRLOCK)
-                        op_data->op_flags |= MF_GETATTR_LOCK;
-        }
-}
-
 static 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;
        struct req_format     *req_fmt;
+       size_t                 u32_count = 0;
        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(NULL, 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;
+       CDEBUG(D_INODE, "%s: "DFID" file closed with intent: %x\n",
+              exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
+              op_data->op_bias);
+
+       if (op_data->op_bias & MDS_CLOSE_INTENT) {
+               req_fmt = &RQF_MDS_CLOSE_INTENT;
+               if (op_data->op_bias & MDS_HSM_RELEASE) {
+                       /* allocate a FID for volatile file */
+                       rc = mdc_fid_alloc(NULL, exp, &op_data->op_fid2,
+                                          op_data);
+                       if (rc < 0) {
+                               CERROR("%s: "DFID" allocating FID: rc = %d\n",
+                                      obd->obd_name, PFID(&op_data->op_fid1),
+                                      rc);
+                               /* save the errcode and proceed to close */
+                               saved_rc = rc;
+                       }
+               }
+               if (op_data->op_bias & MDS_CLOSE_RESYNC_DONE) {
+                       size_t count = op_data->op_data_size / sizeof(__u32);
+
+                       if (count > INLINE_RESYNC_ARRAY_SIZE)
+                               u32_count = count;
                }
+       } else {
+               req_fmt = &RQF_MDS_CLOSE;
        }
 
        *request = NULL;
-       req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
-        if (req == NULL)
-                RETURN(-ENOMEM);
+       if (OBD_FAIL_CHECK(OBD_FAIL_MDC_CLOSE))
+               req = NULL;
+       else
+               req = ptlrpc_request_alloc(class_exp2cliimp(exp), req_fmt);
+
+       /* Ensure that this close's handle is fixed up during replay. */
+       if (likely(mod != NULL)) {
+               LASSERTF(mod->mod_open_req != NULL &&
+                        mod->mod_open_req->rq_type != LI_POISON,
+                        "POISONED open %p!\n", mod->mod_open_req);
+
+               mod->mod_close_req = req;
+
+               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 */
+               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");
+       }
+       if (req == NULL) {
+               /**
+                * TODO: repeat close after errors
+                */
+               CWARN("%s: close of FID "DFID" failed, file reference will be "
+                     "dropped when this client unmounts or is evicted\n",
+                     obd->obd_name, PFID(&op_data->op_fid1));
+               GOTO(out, rc = -ENOMEM);
+       }
 
-        mdc_set_capa_size(req, &RMF_CAPA1, op_data->op_capa1);
+       if (u32_count > 0)
+               req_capsule_set_size(&req->rq_pill, &RMF_U32, RCL_CLIENT,
+                                    u32_count * sizeof(__u32));
 
-        rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
-        if (rc) {
-                ptlrpc_request_free(req);
-                RETURN(rc);
-        }
+       rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_CLOSE);
+       if (rc) {
+               ptlrpc_request_free(req);
+               req = NULL;
+               GOTO(out, rc);
+       }
 
         /* To avoid a livelock (bug 7034), we need to send CLOSE RPCs to a
          * portal whose threads are not taking any DLM locks and are therefore
@@ -879,36 +878,17 @@ static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
         req->rq_request_portal = MDS_READPAGE_PORTAL;
         ptlrpc_at_set_req_timeout(req);
 
-        /* Ensure that this close's handle is fixed up during replay. */
-        if (likely(mod != NULL)) {
-                LASSERTF(mod->mod_open_req != NULL &&
-                         mod->mod_open_req->rq_type != LI_POISON,
-                         "POISONED open %p!\n", mod->mod_open_req);
-
-                mod->mod_close_req = req;
-
-                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 */
-               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");
-        }
 
         mdc_close_pack(req, op_data);
 
        req_capsule_set_size(&req->rq_pill, &RMF_MDT_MD, RCL_SERVER,
                             obd->u.cli.cl_default_mds_easize);
-       req_capsule_set_size(&req->rq_pill, &RMF_LOGCOOKIES, RCL_SERVER,
-                            obd->u.cli.cl_default_mds_cookiesize);
 
         ptlrpc_request_set_replen(req);
 
-        mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
-        rc = ptlrpc_queue_wait(req);
-        mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
+       mdc_get_mod_rpc_slot(req, NULL);
+       rc = ptlrpc_queue_wait(req);
+       mdc_put_mod_rpc_slot(req, NULL);
 
         if (req->rq_repmsg == NULL) {
                 CDEBUG(D_RPCTRACE, "request failed to send: %p, %d\n", req,
@@ -942,6 +922,7 @@ static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
                 }
         }
 
+out:
         if (mod) {
                 if (rc != 0)
                         mod->mod_close_req = NULL;
@@ -950,83 +931,12 @@ static int mdc_close(struct obd_export *exp, struct md_op_data *op_data,
                 obd_mod_put(mod);
         }
         *request = req;
-        mdc_close_handle_reply(req, op_data, rc);
-        RETURN(rc < 0 ? rc : saved_rc);
-}
-
-static int mdc_done_writing(struct obd_export *exp, struct md_op_data *op_data,
-                           struct md_open_data *mod)
-{
-        struct obd_device     *obd = class_exp2obd(exp);
-        struct ptlrpc_request *req;
-        int                    rc;
-        ENTRY;
-
-        req = ptlrpc_request_alloc(class_exp2cliimp(exp),
-                                   &RQF_MDS_DONE_WRITING);
-        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_DONE_WRITING);
-        if (rc) {
-                ptlrpc_request_free(req);
-                RETURN(rc);
-        }
-
-        if (mod != NULL) {
-                LASSERTF(mod->mod_open_req != NULL &&
-                         mod->mod_open_req->rq_type != LI_POISON,
-                         "POISONED setattr %p!\n", mod->mod_open_req);
-
-                mod->mod_close_req = req;
-                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 */
-               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);
-        ptlrpc_request_set_replen(req);
-
-        mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
-        rc = ptlrpc_queue_wait(req);
-        mdc_put_rpc_lock(obd->u.cli.cl_close_lock, NULL);
-
-        if (rc == -ESTALE) {
-                /**
-                 * it can be allowed error after 3633 if open or setattr were
-                 * committed and server failed before close was sent.
-                 * Let's check if mod exists and return no error in that case
-                 */
-                if (mod) {
-                        LASSERT(mod->mod_open_req != NULL);
-                        if (mod->mod_open_req->rq_committed)
-                                rc = 0;
-                }
-        }
-
-        if (mod) {
-                if (rc != 0)
-                        mod->mod_close_req = NULL;
-               LASSERT(mod->mod_open_req != NULL);
-               mdc_free_open(mod);
-
-                /* Since now, mod is accessed through setattr req only,
-                 * thus DW req does not keep a reference on mod anymore. */
-                obd_mod_put(mod);
-        }
 
-        mdc_close_handle_reply(req, op_data, rc);
-        ptlrpc_req_finished(req);
-        RETURN(rc);
+       RETURN(rc < 0 ? rc : saved_rc);
 }
 
 static int mdc_getpage(struct obd_export *exp, const struct lu_fid *fid,
-                      __u64 offset, struct obd_capa *oc,
-                      struct page **pages, int npages,
+                      u64 offset, struct page **pages, int npages,
                       struct ptlrpc_request **request)
 {
        struct ptlrpc_request   *req;
@@ -1046,8 +956,6 @@ restart_bulk:
        if (req == NULL)
                RETURN(-ENOMEM);
 
-       mdc_set_capa_size(req, &RMF_CAPA1, oc);
-
        rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_READPAGE);
        if (rc) {
                ptlrpc_request_free(req);
@@ -1057,18 +965,21 @@ restart_bulk:
        req->rq_request_portal = MDS_READPAGE_PORTAL;
        ptlrpc_at_set_req_timeout(req);
 
-       desc = ptlrpc_prep_bulk_imp(req, npages, 1, BULK_PUT_SINK,
-                                   MDS_BULK_PORTAL);
+       desc = ptlrpc_prep_bulk_imp(req, npages, 1,
+                                   PTLRPC_BULK_PUT_SINK | PTLRPC_BULK_BUF_KIOV,
+                                   MDS_BULK_PORTAL,
+                                   &ptlrpc_bulk_kiov_pin_ops);
        if (desc == NULL) {
-               ptlrpc_request_free(req);
+               ptlrpc_req_finished(req);
                RETURN(-ENOMEM);
        }
 
        /* NB req now owns desc and will free it when it gets freed */
        for (i = 0; i < npages; i++)
-               ptlrpc_prep_bulk_page_pin(desc, pages[i], 0, PAGE_CACHE_SIZE);
+               desc->bd_frag_ops->add_kiov_frag(desc, pages[i], 0,
+                                                PAGE_SIZE);
 
-       mdc_readdir_pack(req, offset, PAGE_CACHE_SIZE * npages, fid, oc);
+       mdc_readdir_pack(req, offset, PAGE_SIZE * npages, fid);
 
        ptlrpc_request_set_replen(req);
        rc = ptlrpc_queue_wait(req);
@@ -1100,7 +1011,7 @@ restart_bulk:
        if (req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK) {
                CERROR("%s: unexpected bytes transferred: %d (%ld expected)\n",
                       exp->exp_obd->obd_name, req->rq_bulk->bd_nob_transferred,
-                      PAGE_CACHE_SIZE * npages);
+                      PAGE_SIZE * npages);
                ptlrpc_req_finished(req);
                RETURN(-EPROTO);
        }
@@ -1117,7 +1028,7 @@ static void mdc_release_page(struct page *page, int remove)
                        truncate_complete_page(page->mapping, page);
                unlock_page(page);
        }
-       page_cache_release(page);
+       put_page(page);
 }
 
 static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
@@ -1138,7 +1049,7 @@ static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
        if (found > 0 && !radix_tree_exceptional_entry(page)) {
                struct lu_dirpage *dp;
 
-               page_cache_get(page);
+               get_page(page);
                spin_unlock_irq(&mapping->tree_lock);
                /*
                 * In contrast to find_lock_page() we are sure that directory
@@ -1162,11 +1073,11 @@ static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
                        if (unlikely(*start == 1 && *hash == 0))
                                *hash = *start;
                        else
-                               LASSERTF(*start <= *hash, "start = "LPX64
-                                        ",end = "LPX64",hash = "LPX64"\n",
+                               LASSERTF(*start <= *hash, "start = %#llx"
+                                        ",end = %#llx,hash = %#llx\n",
                                         *start, *end, *hash);
-                       CDEBUG(D_VFSTRACE, "offset %lx ["LPX64" "LPX64"],"
-                             " hash "LPX64"\n", offset, *start, *end, *hash);
+                       CDEBUG(D_VFSTRACE, "offset %lx [%#llx %#llx],"
+                             " hash %#llx\n", offset, *start, *end, *hash);
                        if (*hash > *end) {
                                kunmap(page);
                                mdc_release_page(page, 0);
@@ -1184,7 +1095,7 @@ static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
                                page = NULL;
                        }
                } else {
-                       page_cache_release(page);
+                       put_page(page);
                        page = ERR_PTR(-EIO);
                }
        } else {
@@ -1207,36 +1118,36 @@ static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
  *  |        |
  * .|--------v-------   -----.
  * |s|e|f|p|ent|ent| ... |ent|
- * '--|--------------   -----'   Each CFS_PAGE contains a single
+ * '--|--------------   -----'   Each PAGE contains a single
  *    '------.                   lu_dirpage.
  * .---------v-------   -----.
  * |s|e|f|p|ent| 0 | ... | 0 |
  * '-----------------   -----'
  *
- * However, on hosts where the native VM page size (PAGE_CACHE_SIZE) is
+ * However, on hosts where the native VM page size (PAGE_SIZE) is
  * larger than LU_PAGE_SIZE, a single host page may contain multiple
  * lu_dirpages. After reading the lu_dirpages from the MDS, the
  * ldp_hash_end of the first lu_dirpage refers to the one immediately
- * after it in the same CFS_PAGE (arrows simplified for brevity, but
+ * after it in the same PAGE (arrows simplified for brevity, but
  * in general e0==s1, e1==s2, etc.):
  *
  * .--------------------   -----.
  * |s0|e0|f0|p|ent|ent| ... |ent|
  * |---v----------------   -----|
  * |s1|e1|f1|p|ent|ent| ... |ent|
- * |---v----------------   -----|  Here, each CFS_PAGE contains
+ * |---v----------------   -----|  Here, each PAGE contains
  *             ...                 multiple lu_dirpages.
  * |---v----------------   -----|
  * |s'|e'|f'|p|ent|ent| ... |ent|
  * '---|----------------   -----'
  *     v
  * .----------------------------.
- * |        next CFS_PAGE       |
+ * |        next PAGE           |
  *
  * This structure is transformed into a single logical lu_dirpage as follows:
  *
  * - Replace e0 with e' so the request for the next lu_dirpage gets the page
- *   labeled 'next CFS_PAGE'.
+ *   labeled 'next PAGE'.
  *
  * - Copy the LDF_COLLIDE flag from f' to f0 to correctly reflect whether
  *   a hash collision with the next page exists.
@@ -1244,7 +1155,7 @@ static struct page *mdc_page_locate(struct address_space *mapping, __u64 *hash,
  * - Adjust the lde_reclen of the ending entry of each lu_dirpage to span
  *   to the first entry of the next lu_dirpage.
  */
-#if PAGE_CACHE_SIZE > LU_PAGE_SIZE
+#if PAGE_SIZE > LU_PAGE_SIZE
 static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
 {
        int i;
@@ -1265,8 +1176,8 @@ static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
                        /* Advance dp to next lu_dirpage. */
                        dp = (struct lu_dirpage *)((char *)dp + LU_PAGE_SIZE);
 
-                       /* Check if we've reached the end of the CFS_PAGE. */
-                       if (!((unsigned long)dp & ~CFS_PAGE_MASK))
+                       /* Check if we've reached the end of the PAGE. */
+                       if (!((unsigned long)dp & ~PAGE_MASK))
                                break;
 
                        /* Save the hash and flags of this lu_dirpage. */
@@ -1295,7 +1206,7 @@ static void mdc_adjust_dirpages(struct page **pages, int cfs_pgs, int lu_pgs)
 }
 #else
 #define mdc_adjust_dirpages(pages, cfs_pgs, lu_pgs) do {} while (0)
-#endif /* PAGE_CACHE_SIZE > LU_PAGE_SIZE */
+#endif /* PAGE_SIZE > LU_PAGE_SIZE */
 
 /* parameters for readdir page */
 struct readpage_param {
@@ -1310,7 +1221,7 @@ struct readpage_param {
 static inline void delete_from_page_cache(struct page *page)
 {
        remove_from_page_cache(page);
-       page_cache_release(page);
+       put_page(page);
 }
 #endif
 
@@ -1321,27 +1232,27 @@ static inline void delete_from_page_cache(struct page *page)
  * a header lu_dirpage which describes the start/end hash, and whether this
  * page is empty (contains no dir entry) or hash collide with next page.
  * After client receives reply, several pages will be integrated into dir page
- * in CFS_PAGE_SIZE (if CFS_PAGE_SIZE greater than LU_PAGE_SIZE), and the
+ * in PAGE_SIZE (if PAGE_SIZE greater than LU_PAGE_SIZE), and the
  * lu_dirpage for this integrated page will be adjusted.
  **/
 static int mdc_read_page_remote(void *data, struct page *page0)
 {
-       struct readpage_param   *rp = data;
-       struct page             **page_pool;
-       struct page             *page;
-       struct lu_dirpage       *dp;
-       int                     rd_pgs = 0; /* number of pages read actually */
-       int                     npages;
-       struct md_op_data       *op_data = rp->rp_mod;
-       struct ptlrpc_request   *req;
-       int                     max_pages = op_data->op_max_pages;
-       struct inode            *inode;
-       struct lu_fid           *fid;
-       int                     i;
-       int                     rc;
-       ENTRY;
-
-       LASSERT(max_pages > 0 && max_pages <= PTLRPC_MAX_BRW_PAGES);
+       struct readpage_param *rp = data;
+       struct page **page_pool;
+       struct page *page;
+       struct lu_dirpage *dp;
+       struct md_op_data *op_data = rp->rp_mod;
+       struct ptlrpc_request *req;
+       int max_pages;
+       struct inode *inode;
+       struct lu_fid *fid;
+       int rd_pgs = 0; /* number of pages actually read */
+       int npages;
+       int i;
+       int rc;
+       ENTRY;
+
+       max_pages = rp->rp_exp->exp_obd->u.cli.cl_max_pages_per_rpc;
        inode = op_data->op_data;
        fid = &op_data->op_fid1;
        LASSERT(inode != NULL);
@@ -1355,24 +1266,23 @@ static int mdc_read_page_remote(void *data, struct page *page0)
        }
 
        for (npages = 1; npages < max_pages; npages++) {
-               page = page_cache_alloc_cold(inode->i_mapping);
+               page = __page_cache_alloc(mapping_gfp_mask(inode->i_mapping)
+                                         | __GFP_COLD);
                if (page == NULL)
                        break;
                page_pool[npages] = page;
        }
 
-       rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, op_data->op_capa1,
-                        page_pool, npages, &req);
+       rc = mdc_getpage(rp->rp_exp, fid, rp->rp_off, page_pool, npages, &req);
        if (rc < 0) {
                /* page0 is special, which was added into page cache early */
                delete_from_page_cache(page0);
        } else {
                int lu_pgs;
 
-               rd_pgs = (req->rq_bulk->bd_nob_transferred +
-                           PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
-               lu_pgs = req->rq_bulk->bd_nob_transferred >>
-                                                       LU_PAGE_SHIFT;
+               rd_pgs = (req->rq_bulk->bd_nob_transferred + PAGE_SIZE - 1) >>
+                       PAGE_SHIFT;
+               lu_pgs = req->rq_bulk->bd_nob_transferred >> LU_PAGE_SHIFT;
                LASSERT(!(req->rq_bulk->bd_nob_transferred & ~LU_PAGE_MASK));
 
                CDEBUG(D_INODE, "read %d(%d) pages\n", rd_pgs, lu_pgs);
@@ -1393,7 +1303,7 @@ static int mdc_read_page_remote(void *data, struct page *page0)
                page = page_pool[i];
 
                if (rc < 0 || i >= rd_pgs) {
-                       page_cache_release(page);
+                       put_page(page);
                        continue;
                }
 
@@ -1413,7 +1323,7 @@ static int mdc_read_page_remote(void *data, struct page *page0)
                else
                        CDEBUG(D_VFSTRACE, "page %lu add to page cache failed:"
                               " rc = %d\n", offset, ret);
-               page_cache_release(page);
+               put_page(page);
        }
 
        if (page_pool != &page0)
@@ -1472,14 +1382,15 @@ static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
        }
 
        rc = 0;
-       mdc_set_lock_data(exp, &it.d.lustre.it_lock_handle, dir, NULL);
+       lockh.cookie = it.it_lock_handle;
+       mdc_set_lock_data(exp, &lockh, dir, NULL);
 
        rp_param.rp_off = hash_offset;
        rp_param.rp_hash64 = op_data->op_cli_flags & CLI_HASH64;
        page = mdc_page_locate(mapping, &rp_param.rp_off, &start, &end,
                               rp_param.rp_hash64);
        if (IS_ERR(page)) {
-               CERROR("%s: dir page locate: "DFID" at "LPU64": rc %ld\n",
+               CERROR("%s: dir page locate: "DFID" at %llu: rc %ld\n",
                       exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
                       rp_param.rp_off, PTR_ERR(page));
                GOTO(out_unlock, rc = PTR_ERR(page));
@@ -1508,7 +1419,7 @@ static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
                                            rp_param.rp_hash64),
                               mdc_read_page_remote, &rp_param);
        if (IS_ERR(page)) {
-               CDEBUG(D_INFO, "%s: read cache page: "DFID" at "LPU64": %ld\n",
+               CDEBUG(D_INFO, "%s: read cache page: "DFID" at %llu: %ld\n",
                       exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
                       rp_param.rp_off, PTR_ERR(page));
                GOTO(out_unlock, rc = PTR_ERR(page));
@@ -1517,7 +1428,7 @@ static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
        wait_on_page_locked(page);
        (void)kmap(page);
        if (!PageUptodate(page)) {
-               CERROR("%s: page not updated: "DFID" at "LPU64": rc %d\n",
+               CERROR("%s: page not updated: "DFID" at %llu: rc %d\n",
                       exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
                       rp_param.rp_off, -5);
                goto fail;
@@ -1525,7 +1436,7 @@ static int mdc_read_page(struct obd_export *exp, struct md_op_data *op_data,
        if (!PageChecked(page))
                SetPageChecked(page);
        if (PageError(page)) {
-               CERROR("%s: page error: "DFID" at "LPU64": rc %d\n",
+               CERROR("%s: page error: "DFID" at %llu: rc %d\n",
                       exp->exp_obd->obd_name, PFID(&op_data->op_fid1),
                       rp_param.rp_off, -5);
                goto fail;
@@ -1546,8 +1457,8 @@ hash_collision:
                LASSERT(start == rp_param.rp_off);
                CWARN("Page-wide hash collision: %#lx\n", (unsigned long)end);
 #if BITS_PER_LONG == 32
-               CWARN("Real page-wide hash collision at ["LPU64" "LPU64"] with "
-                     "hash "LPU64"\n", le64_to_cpu(dp->ldp_hash_start),
+               CWARN("Real page-wide hash collision at [%llu %llu] with "
+                     "hash %llu\n", le64_to_cpu(dp->ldp_hash_start),
                      le64_to_cpu(dp->ldp_hash_end), hash_offset);
 #endif
 
@@ -1560,9 +1471,7 @@ hash_collision:
        }
        *ppage = page;
 out_unlock:
-       lockh.cookie = it.d.lustre.it_lock_handle;
-       ldlm_lock_decref(&lockh, it.d.lustre.it_lock_mode);
-       it.d.lustre.it_lock_handle = 0;
+       ldlm_lock_decref(&lockh, it.it_lock_mode);
        return rc;
 fail:
        kunmap(page);
@@ -1571,10 +1480,9 @@ fail:
        goto out_unlock;
 }
 
-
 static int mdc_statfs(const struct lu_env *env,
                       struct obd_export *exp, struct obd_statfs *osfs,
-                      __u64 max_age, __u32 flags)
+                     time64_t max_age, __u32 flags)
 {
         struct obd_device     *obd = class_exp2obd(exp);
         struct ptlrpc_request *req;
@@ -1630,47 +1538,52 @@ output:
 
 static int mdc_ioc_fid2path(struct obd_export *exp, struct getinfo_fid2path *gf)
 {
-        __u32 keylen, vallen;
-        void *key;
-        int rc;
+       __u32 keylen, vallen;
+       void *key;
+       int rc;
 
-        if (gf->gf_pathlen > PATH_MAX)
-                RETURN(-ENAMETOOLONG);
-        if (gf->gf_pathlen < 2)
-                RETURN(-EOVERFLOW);
+       if (gf->gf_pathlen > PATH_MAX)
+               RETURN(-ENAMETOOLONG);
+       if (gf->gf_pathlen < 2)
+               RETURN(-EOVERFLOW);
 
-        /* Key is KEY_FID2PATH + getinfo_fid2path description */
-        keylen = cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf);
-        OBD_ALLOC(key, keylen);
-        if (key == NULL)
-                RETURN(-ENOMEM);
-        memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
-        memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
-
-        CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n",
-               PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
+       /* Key is KEY_FID2PATH + getinfo_fid2path description */
+       keylen = cfs_size_round(sizeof(KEY_FID2PATH) + sizeof(*gf) +
+                               sizeof(struct lu_fid));
+       OBD_ALLOC(key, keylen);
+       if (key == NULL)
+               RETURN(-ENOMEM);
+       memcpy(key, KEY_FID2PATH, sizeof(KEY_FID2PATH));
+       memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)), gf, sizeof(*gf));
+       memcpy(key + cfs_size_round(sizeof(KEY_FID2PATH)) + sizeof(*gf),
+              gf->gf_u.gf_root_fid, sizeof(struct lu_fid));
+       CDEBUG(D_IOCTL, "path get "DFID" from %llu #%d\n",
+              PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno);
 
-        if (!fid_is_sane(&gf->gf_fid))
-                GOTO(out, rc = -EINVAL);
+       if (!fid_is_sane(&gf->gf_fid))
+               GOTO(out, rc = -EINVAL);
 
         /* Val is struct getinfo_fid2path result plus path */
         vallen = sizeof(*gf) + gf->gf_pathlen;
 
-       rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf, NULL);
+       rc = obd_get_info(NULL, exp, keylen, key, &vallen, gf);
        if (rc != 0 && rc != -EREMOTE)
                GOTO(out, rc);
 
-        if (vallen <= sizeof(*gf))
-                GOTO(out, rc = -EPROTO);
-        else if (vallen > sizeof(*gf) + gf->gf_pathlen)
-                GOTO(out, rc = -EOVERFLOW);
+       if (vallen <= sizeof(*gf))
+               GOTO(out, rc = -EPROTO);
+       if (vallen > sizeof(*gf) + gf->gf_pathlen)
+               GOTO(out, rc = -EOVERFLOW);
 
-        CDEBUG(D_IOCTL, "path get "DFID" from "LPU64" #%d\n%s\n",
-               PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno, gf->gf_path);
+       CDEBUG(D_IOCTL, "path got "DFID" from %llu #%d: %s\n",
+              PFID(&gf->gf_fid), gf->gf_recno, gf->gf_linkno,
+              gf->gf_pathlen < 512 ? gf->gf_u.gf_path :
+              /* only log the last 512 characters of the path */
+              gf->gf_u.gf_path + gf->gf_pathlen - 512);
 
 out:
-        OBD_FREE(key, keylen);
-        return rc;
+       OBD_FREE(key, keylen);
+       return rc;
 }
 
 static int mdc_ioc_hsm_progress(struct obd_export *exp,
@@ -1687,7 +1600,7 @@ static int mdc_ioc_hsm_progress(struct obd_export *exp,
        if (req == NULL)
                GOTO(out, rc = -ENOMEM);
 
-       mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
+       mdc_pack_body(req, NULL, 0, 0, -1, 0);
 
        /* Copy hsm_progress struct */
        req_hpk = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_PROGRESS);
@@ -1699,7 +1612,10 @@ static int mdc_ioc_hsm_progress(struct obd_export *exp,
 
        ptlrpc_request_set_replen(req);
 
-       rc = mdc_queue_wait(req);
+       mdc_get_mod_rpc_slot(req, NULL);
+       rc = ptlrpc_queue_wait(req);
+       mdc_put_mod_rpc_slot(req, NULL);
+
        GOTO(out, rc);
 out:
        ptlrpc_req_finished(req);
@@ -1719,7 +1635,7 @@ static int mdc_ioc_hsm_ct_register(struct obd_import *imp, __u32 archives)
        if (req == NULL)
                GOTO(out, rc = -ENOMEM);
 
-       mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
+       mdc_pack_body(req, NULL, 0, 0, -1, 0);
 
        /* Copy hsm_progress struct */
        archive_mask = req_capsule_client_get(&req->rq_pill,
@@ -1752,16 +1668,14 @@ static int mdc_ioc_hsm_current_action(struct obd_export *exp,
        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_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);
+       mdc_pack_body(req, &op_data->op_fid1, 0, 0,
+                     op_data->op_suppgids[0], 0);
 
        ptlrpc_request_set_replen(req);
 
@@ -1794,7 +1708,7 @@ static int mdc_ioc_hsm_ct_unregister(struct obd_import *imp)
        if (req == NULL)
                GOTO(out, rc = -ENOMEM);
 
-       mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
+       mdc_pack_body(req, NULL, 0, 0, -1, 0);
 
        ptlrpc_request_set_replen(req);
 
@@ -1819,16 +1733,14 @@ static int mdc_ioc_hsm_state_get(struct obd_export *exp,
        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);
+       mdc_pack_body(req, &op_data->op_fid1, 0, 0,
+                     op_data->op_suppgids[0], 0);
 
        ptlrpc_request_set_replen(req);
 
@@ -1862,16 +1774,14 @@ static int mdc_ioc_hsm_state_set(struct obd_export *exp,
        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);
+       mdc_pack_body(req, &op_data->op_fid1, 0, 0,
+                     op_data->op_suppgids[0], 0);
 
        /* Copy states */
        req_hss = req_capsule_client_get(&req->rq_pill, &RMF_HSM_STATE_SET);
@@ -1881,10 +1791,11 @@ static int mdc_ioc_hsm_state_set(struct obd_export *exp,
 
        ptlrpc_request_set_replen(req);
 
-       rc = mdc_queue_wait(req);
-       GOTO(out, rc);
+       mdc_get_mod_rpc_slot(req, NULL);
+       rc = ptlrpc_queue_wait(req);
+       mdc_put_mod_rpc_slot(req, NULL);
 
-       EXIT;
+       GOTO(out, rc);
 out:
        ptlrpc_req_finished(req);
        return rc;
@@ -1917,7 +1828,7 @@ static int mdc_ioc_hsm_request(struct obd_export *exp,
                RETURN(rc);
        }
 
-       mdc_pack_body(req, NULL, NULL, OBD_MD_FLRMTPERM, 0, -1, 0);
+       mdc_pack_body(req, NULL, 0, 0, -1, 0);
 
        /* Copy hsm_request struct */
        req_hr = req_capsule_client_get(&req->rq_pill, &RMF_MDS_HSM_REQUEST);
@@ -1940,7 +1851,10 @@ static int mdc_ioc_hsm_request(struct obd_export *exp,
 
        ptlrpc_request_set_replen(req);
 
-       rc = mdc_queue_wait(req);
+       mdc_get_mod_rpc_slot(req, NULL);
+       rc = ptlrpc_queue_wait(req);
+       mdc_put_mod_rpc_slot(req, NULL);
+
        GOTO(out, rc);
 
 out:
@@ -1948,258 +1862,43 @@ out:
        return rc;
 }
 
-static struct kuc_hdr *changelog_kuc_hdr(char *buf, size_t len, __u32 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;
-}
-
-struct changelog_show {
-       __u64                            cs_startrec;
-       enum changelog_send_flag         cs_flags;
-       struct file                     *cs_fp;
-       char                            *cs_buf;
-       struct obd_device               *cs_obd;
-};
-
-static inline char *cs_obd_name(struct changelog_show *cs)
-{
-       return cs->cs_obd->obd_name;
-}
-
-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;
-       size_t                           len;
-       int                              rc;
-       ENTRY;
-
-       if (rec->cr_hdr.lrh_type != CHANGELOG_REC) {
-               rc = -EINVAL;
-               CERROR("%s: not a changelog rec %x/%d: rc = %d\n",
-                      cs_obd_name(cs), 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_HSM, "rec="LPU64" start="LPU64"\n",
-                      rec->cr.cr_index, cs->cs_startrec);
-               RETURN(0);
-       }
-
-       CDEBUG(D_HSM, 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);
-        memcpy(lh + 1, &rec->cr, len - sizeof(*lh));
-
-        rc = libcfs_kkuc_msg_put(cs->cs_fp, lh);
-       CDEBUG(D_HSM, "kucmsg fp %p len %zu rc %d\n", cs->cs_fp, len, rc);
-
-        RETURN(rc);
-}
-
-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;
-       enum llog_flag           flags = LLOG_F_IS_CAT;
-       int                      rc;
-
-       CDEBUG(D_HSM, "changelog to fp=%p start "LPU64"\n",
-              cs->cs_fp, cs->cs_startrec);
-
-       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_open(NULL, ctxt, &llh, NULL, CHANGELOG_CATALOG,
-                      LLOG_OPEN_EXISTS);
-       if (rc) {
-               CERROR("%s: fail to open changelog catalog: rc = %d\n",
-                      cs_obd_name(cs), rc);
-               GOTO(out, rc);
-       }
-
-       if (cs->cs_flags & CHANGELOG_FLAG_JOBID)
-               flags |= LLOG_F_EXT_JOBID;
-
-       rc = llog_init_handle(NULL, llh, flags, 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),
-                                      cs->cs_flags))) {
-                kuch->kuc_msgtype = CL_EOF;
-                libcfs_kkuc_msg_put(cs->cs_fp, kuch);
-        }
-
-out:
-       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, KUC_CHANGELOG_MSG_MAXSIZE);
-       OBD_FREE_PTR(cs);
-       return rc;
-}
-
-static int mdc_ioc_changelog_send(struct obd_device *obd,
-                                  struct ioc_changelog *icc)
-{
-       struct changelog_show *cs;
-       struct task_struct *task;
-       int rc;
-
-        /* Freed in mdc_changelog_send_thread */
-        OBD_ALLOC_PTR(cs);
-        if (!cs)
-                return -ENOMEM;
-
-       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
-        */
-       task = kthread_run(mdc_changelog_send_thread, cs,
-                          "mdc_clg_send_thread");
-       if (IS_ERR(task)) {
-               rc = PTR_ERR(task);
-               CERROR("%s: cannot start changelog thread: rc = %d\n",
-                      cs_obd_name(cs), rc);
-               OBD_FREE_PTR(cs);
-       } else {
-               rc = 0;
-               CDEBUG(D_HSM, "%s: started changelog thread\n",
-                      cs_obd_name(cs));
-       }
-
-       return rc;
-}
-
 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;
+       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);
+       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;
+       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;
+       ptlrpc_request_set_replen(req);
+       ptlrpc_at_set_req_timeout(req);
 
-        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);
+       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);
+       RETURN(rc);
 }
 
 static int mdc_ioc_swap_layouts(struct obd_export *exp,
@@ -2215,9 +1914,9 @@ static int mdc_ioc_swap_layouts(struct obd_export *exp,
 
        /* 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.
+        * locks held 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
+        * with the request RPC to avoid extra RPC round trips.
         */
        count = mdc_resource_get_unused(exp, &op_data->op_fid1, &cancels,
                                        LCK_EX, MDS_INODELOCK_LAYOUT |
@@ -2233,9 +1932,6 @@ static int mdc_ioc_swap_layouts(struct obd_export *exp,
                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);
@@ -2264,30 +1960,18 @@ out:
 static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
                         void *karg, void __user *uarg)
 {
-        struct obd_device *obd = exp->exp_obd;
-        struct obd_ioctl_data *data = karg;
-        struct obd_import *imp = obd->u.cli.cl_import;
-        int rc;
-        ENTRY;
+       struct obd_device *obd = exp->exp_obd;
+       struct obd_ioctl_data *data = karg;
+       struct obd_import *imp = obd->u.cli.cl_import;
+       int rc;
+       ENTRY;
 
        if (!try_module_get(THIS_MODULE)) {
                CERROR("%s: cannot get module '%s'\n", obd->obd_name,
                       module_name(THIS_MODULE));
                return -EINVAL;
        }
-        switch (cmd) {
-        case OBD_IOC_CHANGELOG_SEND:
-                rc = mdc_ioc_changelog_send(obd, karg);
-                GOTO(out, rc);
-        case OBD_IOC_CHANGELOG_CLEAR: {
-                struct ioc_changelog *icc = karg;
-                struct changelog_setinfo cs =
-                        {.cs_recno = icc->icc_recno, .cs_id = icc->icc_id};
-                rc = obd_set_info_async(NULL, exp, strlen(KEY_CHANGELOG_CLEAR),
-                                        KEY_CHANGELOG_CLEAR, sizeof(cs), &cs,
-                                        NULL);
-                GOTO(out, rc);
-        }
+       switch (cmd) {
        case OBD_IOC_FID2PATH:
                rc = mdc_ioc_fid2path(exp, karg);
                GOTO(out, rc);
@@ -2312,52 +1996,49 @@ static int mdc_iocontrol(unsigned int cmd, struct obd_export *exp, int len,
        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, 0);
-                if (rc < 0)
-                        GOTO(out, rc);
-                GOTO(out, rc = 0);
-        case IOC_OSC_SET_ACTIVE:
-                rc = ptlrpc_set_import_active(imp, data->ioc_offset);
-                GOTO(out, rc);
-        case OBD_IOC_POLL_QUOTACHECK:
-                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, 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};
+       case OBD_IOC_CLIENT_RECOVER:
+               rc = ptlrpc_recover_import(imp, data->ioc_inlbuf1, 0);
+               if (rc < 0)
+                       GOTO(out, rc);
+               GOTO(out, rc = 0);
+       case IOC_OSC_SET_ACTIVE:
+               rc = ptlrpc_set_import_active(imp, data->ioc_offset);
+               GOTO(out, rc);
+       case OBD_IOC_PING_TARGET:
+               rc = ptlrpc_obd_ping(obd);
+               GOTO(out, rc);
+       /*
+        * 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};
 
-                if (*((__u32 *) data->ioc_inlbuf2) != 0)
-                        GOTO(out, rc = -ENODEV);
+               if (*((__u32 *) data->ioc_inlbuf2) != 0)
+                       GOTO(out, rc = -ENODEV);
 
-                /* copy UUID */
+               /* 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),
+                               ktime_get_seconds() - 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);
+                                    min((int) data->ioc_plen1,
+                                        (int) sizeof(stat_buf))))
+                       GOTO(out, rc = -EFAULT);
 
-                GOTO(out, rc = 0);
-        }
+               GOTO(out, rc = 0);
+       }
        case OBD_IOC_QUOTACTL: {
                struct if_quotactl *qctl = karg;
                struct obd_quotactl *oqctl;
@@ -2397,8 +2078,8 @@ out:
 }
 
 static int mdc_get_info_rpc(struct obd_export *exp,
-                           obd_count keylen, void *key,
-                           int vallen, void *val)
+                           u32 keylen, void *key,
+                           u32 vallen, void *val)
 {
         struct obd_import      *imp = class_exp2cliimp(exp);
         struct ptlrpc_request  *req;
@@ -2413,7 +2094,7 @@ static int mdc_get_info_rpc(struct obd_export *exp,
         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_KEY,
                              RCL_CLIENT, keylen);
         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VALLEN,
-                             RCL_CLIENT, sizeof(__u32));
+                            RCL_CLIENT, sizeof(vallen));
 
         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GET_INFO);
         if (rc) {
@@ -2424,7 +2105,7 @@ static int mdc_get_info_rpc(struct obd_export *exp,
         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_KEY);
         memcpy(tmp, key, keylen);
         tmp = req_capsule_client_get(&req->rq_pill, &RMF_GETINFO_VALLEN);
-        memcpy(tmp, &vallen, sizeof(__u32));
+       memcpy(tmp, &vallen, sizeof(vallen));
 
         req_capsule_set_size(&req->rq_pill, &RMF_GETINFO_VAL,
                              RCL_SERVER, vallen);
@@ -2510,7 +2191,8 @@ static int mdc_ioc_hsm_ct_start(struct obd_export *exp,
  * @param val KUC message (kuc_hdr + hsm_action_list)
  * @param len total length of message
  */
-static int mdc_hsm_copytool_send(size_t len, void *val)
+static int mdc_hsm_copytool_send(const struct obd_uuid *uuid,
+                                size_t len, void *val)
 {
        struct kuc_hdr          *lh = (struct kuc_hdr *)val;
        struct hsm_action_list  *hal = (struct hsm_action_list *)(lh + 1);
@@ -2536,7 +2218,7 @@ static int mdc_hsm_copytool_send(size_t len, void *val)
               lh->kuc_msglen, hal->hal_count, hal->hal_fsname);
 
        /* Broadcast to HSM listeners */
-       rc = libcfs_kkuc_group_put(KUC_GRP_HSM, lh);
+       rc = libcfs_kkuc_group_put(uuid, KUC_GRP_HSM, lh);
 
        RETURN(rc);
 }
@@ -2556,9 +2238,6 @@ static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
        if (kcd == NULL || kcd->kcd_magic != KKUC_CT_DATA_MAGIC)
                return -EPROTO;
 
-       if (!obd_uuid_equals(&kcd->kcd_uuid, &imp->imp_obd->obd_uuid))
-               return 0;
-
        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);
@@ -2574,14 +2253,14 @@ static int mdc_hsm_ct_reregister(void *data, void *cb_arg)
 static int mdc_kuc_reregister(struct obd_import *imp)
 {
        /* re-register HSM agents */
-       return libcfs_kkuc_group_foreach(KUC_GRP_HSM, mdc_hsm_ct_reregister,
-                                        (void *)imp);
+       return libcfs_kkuc_group_foreach(&imp->imp_obd->obd_uuid, KUC_GRP_HSM,
+                                        mdc_hsm_ct_reregister, imp);
 }
 
 static int mdc_set_info_async(const struct lu_env *env,
                              struct obd_export *exp,
-                             obd_count keylen, void *key,
-                             obd_count vallen, void *val,
+                             u32 keylen, void *key,
+                             u32 vallen, void *val,
                              struct ptlrpc_request_set *set)
 {
        struct obd_import       *imp = class_exp2cliimp(exp);
@@ -2608,31 +2287,30 @@ static int mdc_set_info_async(const struct lu_env *env,
                                        keylen, key, vallen, val, set);
                 RETURN(rc);
         }
-        if (KEY_IS(KEY_SPTLRPC_CONF)) {
-                sptlrpc_conf_client_adapt(exp->exp_obd);
-                RETURN(0);
-        }
-        if (KEY_IS(KEY_FLUSH_CTX)) {
-                sptlrpc_import_flush_my_ctx(imp);
-                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);
                 RETURN(rc);
         }
         if (KEY_IS(KEY_HSM_COPYTOOL_SEND)) {
-                rc = mdc_hsm_copytool_send(vallen, val);
+               rc = mdc_hsm_copytool_send(&imp->imp_obd->obd_uuid, vallen,
+                                          val);
                 RETURN(rc);
         }
 
-       CERROR("Unknown key %s\n", (char *)key);
-       RETURN(-EINVAL);
+       if (KEY_IS(KEY_DEFAULT_EASIZE)) {
+               __u32 *default_easize = val;
+
+               exp->exp_obd->u.cli.cl_default_mds_easize = *default_easize;
+               RETURN(0);
+       }
+
+       rc = osc_set_info_async(env, exp, keylen, key, vallen, val, set);
+       RETURN(rc);
 }
 
 static 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)
+                       __u32 keylen, void *key, __u32 *vallen, void *val)
 {
        int rc = -EINVAL;
 
@@ -2655,26 +2333,6 @@ static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
                default_easize = val;
                *default_easize = exp->exp_obd->u.cli.cl_default_mds_easize;
                RETURN(0);
-       } else if (KEY_IS(KEY_MAX_COOKIESIZE)) {
-               __u32 mdsize, *max_cookiesize;
-
-               if (*vallen != sizeof(int))
-                       RETURN(-EINVAL);
-               mdsize = *(int *)val;
-               if (mdsize > exp->exp_obd->u.cli.cl_max_mds_cookiesize)
-                       exp->exp_obd->u.cli.cl_max_mds_cookiesize = mdsize;
-               max_cookiesize = val;
-               *max_cookiesize = exp->exp_obd->u.cli.cl_max_mds_cookiesize;
-               RETURN(0);
-       } else if (KEY_IS(KEY_DEFAULT_COOKIESIZE)) {
-               __u32 *default_cookiesize;
-
-               if (*vallen != sizeof(int))
-                       RETURN(-EINVAL);
-               default_cookiesize = val;
-               *default_cookiesize =
-                       exp->exp_obd->u.cli.cl_default_mds_cookiesize;
-               RETURN(0);
         } else if (KEY_IS(KEY_CONN_DATA)) {
                 struct obd_import *imp = class_exp2cliimp(exp);
                 struct obd_connect_data *data = val;
@@ -2695,7 +2353,7 @@ static int mdc_get_info(const struct lu_env *env, struct obd_export *exp,
 }
 
 static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
-                    struct obd_capa *oc, struct ptlrpc_request **request)
+                    struct ptlrpc_request **request)
 {
         struct ptlrpc_request *req;
         int                    rc;
@@ -2706,15 +2364,13 @@ static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
         if (req == NULL)
                 RETURN(-ENOMEM);
 
-        mdc_set_capa_size(req, &RMF_CAPA1, oc);
-
         rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_SYNC);
         if (rc) {
                 ptlrpc_request_free(req);
                 RETURN(rc);
         }
 
-        mdc_pack_body(req, fid, oc, 0, 0, -1, 0);
+       mdc_pack_body(req, fid, 0, 0, -1, 0);
 
         ptlrpc_request_set_replen(req);
 
@@ -2727,65 +2383,95 @@ static int mdc_fsync(struct obd_export *exp, const struct lu_fid *fid,
 }
 
 static int mdc_import_event(struct obd_device *obd, struct obd_import *imp,
-                            enum obd_import_event event)
+                           enum obd_import_event event)
 {
-        int rc = 0;
-
-        LASSERT(imp->imp_obd == obd);
-
-        switch (event) {
-        case IMP_EVENT_DISCON: {
-#if 0
-                /* XXX Pass event up to OBDs stack. used only for FLD now */
-                rc = obd_notify_observer(obd, obd, OBD_NOTIFY_DISCON, NULL);
-#endif
-                break;
-        }
-        case IMP_EVENT_INACTIVE: {
-                struct client_obd *cli = &obd->u.cli;
-                /*
-                 * Flush current sequence to make client obtain new one
-                 * from server in case of disconnect/reconnect.
-                 */
-                if (cli->cl_seq != NULL)
-                        seq_client_flush(cli->cl_seq);
+       struct client_obd *cli = &obd->u.cli;
+       int rc = 0;
 
-                rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE, NULL);
-                break;
-        }
-        case IMP_EVENT_INVALIDATE: {
-                struct ldlm_namespace *ns = obd->obd_namespace;
+       LASSERT(imp->imp_obd == obd);
 
-                ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
+       switch (event) {
+       case IMP_EVENT_DISCON:
+               spin_lock(&cli->cl_loi_list_lock);
+               cli->cl_avail_grant = 0;
+               cli->cl_lost_grant = 0;
+               spin_unlock(&cli->cl_loi_list_lock);
+               break;
+       case IMP_EVENT_INACTIVE:
+               /*
+                * Flush current sequence to make client obtain new one
+                * from server in case of disconnect/reconnect.
+                */
+               down_read(&cli->cl_seq_rwsem);
+               if (cli->cl_seq)
+                       seq_client_flush(cli->cl_seq);
+               up_read(&cli->cl_seq_rwsem);
 
-                break;
-        }
+               rc = obd_notify_observer(obd, obd, OBD_NOTIFY_INACTIVE);
+               break;
+       case IMP_EVENT_INVALIDATE: {
+               struct ldlm_namespace *ns = obd->obd_namespace;
+               struct lu_env *env;
+               __u16 refcheck;
+
+               ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
+
+               env = cl_env_get(&refcheck);
+               if (!IS_ERR(env)) {
+                       /* Reset grants. All pages go to failing rpcs due to
+                        * the invalid import.
+                        */
+                       osc_io_unplug(env, cli, NULL);
+
+                       cfs_hash_for_each_nolock(ns->ns_rs_hash,
+                                                osc_ldlm_resource_invalidate,
+                                                env, 0);
+                       cl_env_put(env, &refcheck);
+                       ldlm_namespace_cleanup(ns, LDLM_FL_LOCAL_ONLY);
+               } else {
+                       rc = PTR_ERR(env);
+               }
+               break;
+       }
        case IMP_EVENT_ACTIVE:
-               rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE, NULL);
+               rc = obd_notify_observer(obd, obd, OBD_NOTIFY_ACTIVE);
                /* 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();
-        }
-        RETURN(rc);
+       case IMP_EVENT_OCD: {
+               struct obd_connect_data *ocd = &imp->imp_connect_data;
+
+               if (OCD_HAS_FLAG(ocd, GRANT))
+                       osc_init_grant(cli, ocd);
+
+               rc = obd_notify_observer(obd, obd, OBD_NOTIFY_OCD);
+               break;
+       }
+       case IMP_EVENT_DEACTIVATE:
+       case IMP_EVENT_ACTIVATE:
+               break;
+       default:
+               CERROR("Unknown import event %x\n", event);
+               LBUG();
+       }
+       RETURN(rc);
 }
 
 int mdc_fid_alloc(const struct lu_env *env, struct obd_export *exp,
                  struct lu_fid *fid, struct md_op_data *op_data)
 {
        struct client_obd *cli = &exp->exp_obd->u.cli;
-       struct lu_client_seq *seq = cli->cl_seq;
+       int rc = -EIO;
+
        ENTRY;
-       RETURN(seq_client_alloc_fid(env, seq, fid));
+
+       down_read(&cli->cl_seq_rwsem);
+       if (cli->cl_seq)
+               rc = seq_client_alloc_fid(env, cli->cl_seq, fid);
+       up_read(&cli->cl_seq_rwsem);
+
+       RETURN(rc);
 }
 
 static struct obd_uuid *mdc_get_uuid(struct obd_export *exp)
@@ -2858,36 +2544,19 @@ static void mdc_llog_finish(struct obd_device *obd)
        EXIT;
 }
 
-static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
+int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
 {
-       struct client_obd               *cli = &obd->u.cli;
-       int                             rc;
-       ENTRY;
+       int rc;
 
-        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);
+       ENTRY;
 
-       rc = ptlrpcd_addref();
+       rc = osc_setup_common(obd, cfg);
        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_ptlrpcd_decref, rc = -ENOMEM);
-        mdc_init_rpc_lock(cli->cl_close_lock);
+               RETURN(rc);
 
-        rc = client_obd_setup(obd, cfg);
-        if (rc)
-                GOTO(err_close_lock, rc);
-#ifdef LPROCFS
-       obd->obd_vars = lprocfs_mdc_obd_vars;
-       lprocfs_obd_setup(obd);
-       lprocfs_alloc_md_stats(obd, 0);
-#endif
-       sptlrpc_lprocfs_cliobd_attach(obd);
-       ptlrpc_lprocfs_register_obd(obd);
+       rc = mdc_tunables_init(obd);
+       if (rc)
+               GOTO(err_osc_cleanup, rc);
 
        ns_register_cancel(obd->obd_namespace, mdc_cancel_weight);
 
@@ -2895,32 +2564,40 @@ static int mdc_setup(struct obd_device *obd, struct lustre_cfg *cfg)
 
        rc = mdc_llog_init(obd);
         if (rc) {
-                mdc_cleanup(obd);
-                CERROR("failed to setup llogging subsystems\n");
+                CERROR("%s: failed to setup llogging subsystems: rc = %d\n",
+                      obd->obd_name, rc);
+               GOTO(err_llog_cleanup, rc);
         }
 
-        RETURN(rc);
+       rc = mdc_changelog_cdev_init(obd);
+       if (rc) {
+               CERROR("%s: failed to setup changelog char device: rc = %d\n",
+                      obd->obd_name, rc);
+               GOTO(err_changelog_cleanup, rc);
+       }
 
-err_close_lock:
-        OBD_FREE(cli->cl_close_lock, sizeof (*cli->cl_close_lock));
-err_ptlrpcd_decref:
-        ptlrpcd_decref();
-err_rpc_lock:
-        OBD_FREE(cli->cl_rpc_lock, sizeof (*cli->cl_rpc_lock));
-        RETURN(rc);
+       RETURN(rc);
+
+err_changelog_cleanup:
+       mdc_llog_finish(obd);
+err_llog_cleanup:
+       lprocfs_free_md_stats(obd);
+       ptlrpc_lprocfs_unregister_obd(obd);
+err_osc_cleanup:
+       osc_cleanup_common(obd);
+       return rc;
 }
 
-/* Initialize the default and maximum LOV EA and cookie sizes.  This allows
+/* Initialize the default and maximum LOV EA sizes.  This allows
  * us to make MDS RPCs with large enough reply buffers to hold a default
- * sized EA and cookie without having to calculate this (via a call into the
+ * sized EA without having to calculate this (via a call into the
  * LOV + OSCs) each time we make an RPC.  The maximum size is also tracked
  * but not used to avoid wastefully vmalloc()'ing large reply buffers when
  * a large number of stripes is possible.  If a larger reply buffer is
  * required it will be reallocated in the ptlrpc layer due to overflow.
  */
 static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
-                           __u32 def_easize, __u32 cookiesize,
-                           __u32 def_cookiesize)
+                           __u32 def_easize)
 {
        struct obd_device *obd = exp->exp_obd;
        struct client_obd *cli = &obd->u.cli;
@@ -2932,150 +2609,35 @@ static int mdc_init_ea_size(struct obd_export *exp, __u32 easize,
        if (cli->cl_default_mds_easize < def_easize)
                cli->cl_default_mds_easize = def_easize;
 
-       if (cli->cl_max_mds_cookiesize < cookiesize)
-               cli->cl_max_mds_cookiesize = cookiesize;
-
-       if (cli->cl_default_mds_cookiesize < def_cookiesize)
-               cli->cl_default_mds_cookiesize = def_cookiesize;
-
        RETURN(0);
 }
 
-static int mdc_precleanup(struct obd_device *obd, enum obd_cleanup_stage stage)
+static int mdc_precleanup(struct obd_device *obd)
 {
-        int rc = 0;
-        ENTRY;
-
-        switch (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, NULL);
-
-                obd_cleanup_client_import(obd);
-                ptlrpc_lprocfs_unregister_obd(obd);
-                lprocfs_obd_cleanup(obd);
-               lprocfs_free_md_stats(obd);
-               mdc_llog_finish(obd);
-                break;
-        }
-        RETURN(rc);
-}
-
-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_close_lock, sizeof (*cli->cl_close_lock));
-
-        ptlrpcd_decref();
-
-        return client_obd_cleanup(obd);
-}
-
-static int mdc_process_config(struct obd_device *obd, size_t len, void *buf)
-{
-        struct lustre_cfg *lcfg = buf;
-       int rc = class_process_proc_param(PARAM_MDC, obd->obd_vars, lcfg, obd);
-       return (rc > 0 ? 0: rc);
-}
-
-
-/* get remote permission for current user on fid */
-static int mdc_get_remote_perm(struct obd_export *exp, const struct lu_fid *fid,
-                              struct obd_capa *oc, __u32 suppgid,
-                              struct ptlrpc_request **request)
-{
-        struct ptlrpc_request  *req;
-        int                    rc;
-        ENTRY;
-
-        LASSERT(client_is_remote(exp));
-
-        *request = NULL;
-        req = ptlrpc_request_alloc(class_exp2cliimp(exp), &RQF_MDS_GETATTR);
-        if (req == NULL)
-                RETURN(-ENOMEM);
-
-        mdc_set_capa_size(req, &RMF_CAPA1, oc);
-
-        rc = ptlrpc_request_pack(req, LUSTRE_MDS_VERSION, MDS_GETATTR);
-        if (rc) {
-                ptlrpc_request_free(req);
-                RETURN(rc);
-        }
-
-        mdc_pack_body(req, fid, oc, OBD_MD_FLRMTPERM, 0, suppgid, 0);
-
-        req_capsule_set_size(&req->rq_pill, &RMF_ACL, RCL_SERVER,
-                             sizeof(struct mdt_remote_perm));
+       ENTRY;
 
-        ptlrpc_request_set_replen(req);
+       osc_precleanup_common(obd);
+       mdc_changelog_cdev_finish(obd);
 
-        rc = ptlrpc_queue_wait(req);
-        if (rc)
-                ptlrpc_req_finished(req);
-        else
-                *request = req;
-        RETURN(rc);
+       obd_cleanup_client_import(obd);
+       ptlrpc_lprocfs_unregister_obd(obd);
+       lprocfs_free_md_stats(obd);
+       mdc_llog_finish(obd);
+       RETURN(0);
 }
 
-static int mdc_interpret_renew_capa(const struct lu_env *env,
-                                    struct ptlrpc_request *req, void *args,
-                                    int status)
+static int mdc_cleanup(struct obd_device *obd)
 {
-        struct mdc_renew_capa_args *ra = args;
-        struct mdt_body *body = NULL;
-        struct lustre_capa *capa;
-        ENTRY;
-
-        if (status)
-                GOTO(out, capa = ERR_PTR(status));
-
-        body = req_capsule_server_get(&req->rq_pill, &RMF_MDT_BODY);
-        if (body == NULL)
-                GOTO(out, capa = ERR_PTR(-EFAULT));
-
-       if ((body->mbo_valid & OBD_MD_FLOSSCAPA) == 0)
-                GOTO(out, capa = ERR_PTR(-ENOENT));
-
-        capa = req_capsule_server_get(&req->rq_pill, &RMF_CAPA2);
-        if (!capa)
-                GOTO(out, capa = ERR_PTR(-EFAULT));
-        EXIT;
-out:
-        ra->ra_cb(ra->ra_oc, capa);
-        return 0;
+       return osc_cleanup_common(obd);
 }
 
-static int mdc_renew_capa(struct obd_export *exp, struct obd_capa *oc,
-                          renew_capa_cb_t cb)
+int mdc_process_config(struct obd_device *obd, size_t len, void *buf)
 {
-        struct ptlrpc_request *req;
-        struct mdc_renew_capa_args *ra;
-        ENTRY;
-
-        req = ptlrpc_request_alloc_pack(class_exp2cliimp(exp), &RQF_MDS_GETATTR,
-                                        LUSTRE_MDS_VERSION, MDS_GETATTR);
-        if (req == NULL)
-                RETURN(-ENOMEM);
+       struct lustre_cfg *lcfg = buf;
+       size_t count  = class_modify_config(lcfg, PARAM_MDC,
+                                           &obd->obd_kset.kobj);
 
-        /* NB, OBD_MD_FLOSSCAPA is set here, but it doesn't necessarily mean the
-         * capa to renew is oss capa.
-         */
-        mdc_pack_body(req, &oc->c_capa.lc_fid, oc, OBD_MD_FLOSSCAPA, 0, -1, 0);
-        ptlrpc_request_set_replen(req);
-
-        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, PDL_POLICY_LOCAL, -1);
-        RETURN(0);
+       return count > 0 ? 0 : count;
 }
 
 static struct obd_ops mdc_obd_ops = {
@@ -3086,7 +2648,8 @@ static struct obd_ops mdc_obd_ops = {
         .o_add_conn         = client_import_add_conn,
         .o_del_conn         = client_import_del_conn,
         .o_connect          = client_connect_import,
-        .o_disconnect       = client_disconnect_export,
+       .o_reconnect        = osc_reconnect,
+       .o_disconnect       = osc_disconnect,
         .o_iocontrol        = mdc_iocontrol,
         .o_set_info_async   = mdc_set_info_async,
         .o_statfs           = mdc_statfs,
@@ -3098,16 +2661,13 @@ static struct obd_ops mdc_obd_ops = {
         .o_process_config   = mdc_process_config,
         .o_get_uuid         = mdc_get_uuid,
         .o_quotactl         = mdc_quotactl,
-        .o_quotacheck       = mdc_quotacheck
 };
 
 static struct md_ops mdc_md_ops = {
-        .m_getstatus        = mdc_getstatus,
+       .m_get_root         = mdc_get_root,
         .m_null_inode      = mdc_null_inode,
-        .m_find_cbdata      = mdc_find_cbdata,
         .m_close            = mdc_close,
         .m_create           = mdc_create,
-        .m_done_writing     = mdc_done_writing,
         .m_enqueue          = mdc_enqueue,
         .m_getattr          = mdc_getattr,
         .m_getattr_name     = mdc_getattr_name,
@@ -3118,6 +2678,7 @@ static struct md_ops mdc_md_ops = {
         .m_setxattr         = mdc_setxattr,
         .m_getxattr         = mdc_getxattr,
        .m_fsync                = mdc_fsync,
+       .m_file_resync          = mdc_file_resync,
        .m_read_page            = mdc_read_page,
         .m_unlink           = mdc_unlink,
         .m_cancel_unused    = mdc_cancel_unused,
@@ -3128,9 +2689,6 @@ static struct md_ops mdc_md_ops = {
         .m_free_lustre_md   = mdc_free_lustre_md,
         .m_set_open_replay_data = mdc_set_open_replay_data,
         .m_clear_open_replay_data = mdc_clear_open_replay_data,
-        .m_renew_capa       = mdc_renew_capa,
-        .m_unpack_capa      = mdc_unpack_capa,
-        .m_get_remote_perm  = mdc_get_remote_perm,
         .m_intent_getattr_async = mdc_intent_getattr_async,
         .m_revalidate_lock      = mdc_revalidate_lock
 };
@@ -3138,16 +2696,17 @@ static struct md_ops mdc_md_ops = {
 static int __init mdc_init(void)
 {
        return class_register_type(&mdc_obd_ops, &mdc_md_ops, true, NULL,
-                                  LUSTRE_MDC_NAME, NULL);
+                                  LUSTRE_MDC_NAME, &mdc_device_type);
 }
 
-static void /*__exit*/ mdc_exit(void)
+static void __exit mdc_exit(void)
 {
         class_unregister_type(LUSTRE_MDC_NAME);
 }
 
-MODULE_AUTHOR("Sun Microsystems, Inc. <http://www.lustre.org/>");
+MODULE_AUTHOR("OpenSFS, Inc. <http://www.lustre.org/>");
 MODULE_DESCRIPTION("Lustre Metadata Client");
+MODULE_VERSION(LUSTRE_VERSION_STRING);
 MODULE_LICENSE("GPL");
 
 module_init(mdc_init);