Whamcloud - gitweb
LU-1303 lod: introduce lod device
[fs/lustre-release.git] / lustre / mdt / mdt_lproc.c
index d2517d3..26d4692 100644 (file)
@@ -1,39 +1,48 @@
-/* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
- * vim:expandtab:shiftwidth=8:tabstop=8:
+/*
+ * GPL HEADER START
+ *
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 only,
+ * as published by the Free Software Foundation.
  *
- * Copyright (C) 2004-2006 Cluster File Systems, Inc.
- *   Author: Lai Siyao <lsy@clusterfs.com>
- *   Author: Fan Yong <fanyong@clusterfs.com>
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License version 2 for more details (a copy is included
+ * in the LICENSE file that accompanied this code).
  *
- *   This file is part of the Lustre file system, http://www.lustre.org
- *   Lustre is a trademark of Cluster File Systems, Inc.
+ * 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
  *
- *   You may have signed or agreed to another license before downloading
- *   this software.  If so, you are bound by the terms and conditions
- *   of that agreement, and the following does not apply to you.  See the
- *   LICENSE file included with this distribution for more information.
+ * 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.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Use is subject to license terms.
  *
- *   If you did not agree to a different license, then this copy of Lustre
- *   is open source software; you can redistribute it and/or modify it
- *   under the terms of version 2 of the GNU General Public License as
- *   published by the Free Software Foundation.
+ * Copyright (c) 2011, 2012, Whamcloud, Inc.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
  *
- *   In either case, Lustre is distributed in the hope that it will be
- *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
- *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- *   license text for more details.
+ * lustre/mdt/mdt_lproc.c
  *
+ * Author: Lai Siyao <lsy@clusterfs.com>
+ * Author: Fan Yong <fanyong@clusterfs.com>
  */
 
-#ifndef EXPORT_SYMTAB
-# define EXPORT_SYMTAB
-#endif
 #define DEBUG_SUBSYSTEM S_MDS
 
 #include <linux/version.h>
-#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
 #include <asm/statfs.h>
-#endif
 
 #include <linux/module.h>
 
 #include <lprocfs_status.h>
 #include <lu_time.h>
 #include "mdt_internal.h"
+#include <lnet/lib-lnet.h>
 
+enum {
+        LPROC_MDT_NR
+};
 static const char *mdt_proc_names[LPROC_MDT_NR] = {
 };
 
+/**
+ * The rename stats output would be YAML formats, like
+ * rename_stats:
+ * - snapshot_time: 1234567890.123456
+ * - same_dir:
+ *     4kB: { samples: 1230, pct: 33, cum_pct: 45 }
+ *     8kB: { samples: 1242, pct: 33, cum_pct: 78 }
+ *     16kB: { samples: 132, pct: 3, cum_pct: 81 }
+ * - crossdir_src:
+ *     4kB: { samples: 123, pct: 33, cum_pct: 45 }
+ *     8kB: { samples: 124, pct: 33, cum_pct: 78 }
+ *     16kB: { samples: 12, pct: 3, cum_pct: 81 }
+ * - crossdir_tgt:
+ *     4kB: { samples: 123, pct: 33, cum_pct: 45 }
+ *     8kB: { samples: 124, pct: 33, cum_pct: 78 }
+ *     16kB: { samples: 12, pct: 3, cum_pct: 81 }
+ **/
+
+#define pct(a, b) (b ? a * 100 / b : 0)
+
+static void display_rename_stats(struct seq_file *seq, char *name,
+                                 struct obd_histogram *hist)
+{
+        unsigned long tot, t, cum = 0;
+        int i;
+
+        tot = lprocfs_oh_sum(hist);
+        if (tot > 0)
+                seq_printf(seq, "- %-15s\n", name);
+        /* dir size start from 4K, start i from 10(2^10) here */
+        for (i = 0; i < OBD_HIST_MAX; i++) {
+                t = hist->oh_buckets[i];
+                cum += t;
+                if (cum == 0)
+                        continue;
+
+                if (i < 10)
+                        seq_printf(seq, "%6s%d%s", " ", 1<< i, "bytes:");
+                else if (i < 20)
+                        seq_printf(seq, "%6s%d%s", " ", 1<<(i-10), "KB:");
+                else
+                        seq_printf(seq, "%6s%d%s", " ", 1<<(i-20), "MB:");
+
+                seq_printf(seq, " { sample: %3lu, pct: %3lu, cum_pct: %3lu }\n",
+                           t, pct(t, tot), pct(cum, tot));
+
+                if (cum == tot)
+                        break;
+        }
+}
+
+static void rename_stats_show(struct seq_file *seq,
+                              struct rename_stats *rename_stats)
+{
+        struct timeval now;
+
+        /* this sampling races with updates */
+        do_gettimeofday(&now);
+        seq_printf(seq, "rename_stats:\n");
+        seq_printf(seq, "- %-15s %lu.%lu\n", "snapshot_time:",
+                   now.tv_sec, now.tv_usec);
+
+        display_rename_stats(seq, "same_dir",
+                             &rename_stats->hist[RENAME_SAMEDIR_SIZE]);
+        display_rename_stats(seq, "crossdir_src",
+                             &rename_stats->hist[RENAME_CROSSDIR_SRC_SIZE]);
+        display_rename_stats(seq, "crossdir_tgt",
+                             &rename_stats->hist[RENAME_CROSSDIR_TGT_SIZE]);
+}
+
+#undef pct
+
+static int mdt_rename_stats_seq_show(struct seq_file *seq, void *v)
+{
+        struct mdt_device *mdt = seq->private;
+
+        rename_stats_show(seq, &mdt->mdt_rename_stats);
+
+        return 0;
+}
+
+static ssize_t mdt_rename_stats_seq_write(struct file *file, const char *buf,
+                                          size_t len, loff_t *off)
+{
+        struct seq_file *seq = file->private_data;
+        struct mdt_device *mdt = seq->private;
+        int i;
+
+        for (i = 0; i < RENAME_LAST; i++)
+                lprocfs_oh_clear(&mdt->mdt_rename_stats.hist[i]);
+
+        return len;
+}
+
+LPROC_SEQ_FOPS(mdt_rename_stats);
+
+static int lproc_mdt_attach_rename_seqstat(struct mdt_device *mdt)
+{
+        struct lu_device *ld = &mdt->mdt_md_dev.md_lu_dev;
+        struct obd_device *obd = ld->ld_obd;
+        int i;
+
+        for (i = 0; i < RENAME_LAST; i++)
+                spin_lock_init(&mdt->mdt_rename_stats.hist[i].oh_lock);
+
+        return lprocfs_obd_seq_create(obd, "rename_stats", 0444,
+                                      &mdt_rename_stats_fops, mdt);
+}
+
+void mdt_rename_counter_tally(struct mdt_thread_info *info,
+                             struct mdt_device *mdt,
+                             struct ptlrpc_request *req,
+                             struct mdt_object *src,
+                             struct mdt_object *tgt)
+{
+        struct md_attr *ma = &info->mti_attr;
+        struct rename_stats *rstats = &mdt->mdt_rename_stats;
+        int rc;
+
+        ma->ma_need = MA_INODE;
+        ma->ma_valid = 0;
+        rc = mo_attr_get(info->mti_env, mdt_object_child(src), ma);
+        if (rc) {
+                CERROR("%s: "DFID" attr_get, rc = %d\n",
+                      req->rq_export->exp_obd->obd_name,
+                      PFID(mdt_object_fid(src)), rc);
+                return;
+        }
+
+        if (src == tgt) {
+               mdt_counter_incr(req, LPROC_MDT_SAMEDIR_RENAME);
+                lprocfs_oh_tally_log2(&rstats->hist[RENAME_SAMEDIR_SIZE],
+                                      (unsigned int)ma->ma_attr.la_size);
+                return;
+        }
+
+       mdt_counter_incr(req, LPROC_MDT_CROSSDIR_RENAME);
+        lprocfs_oh_tally_log2(&rstats->hist[RENAME_CROSSDIR_SRC_SIZE],
+                              (unsigned int)ma->ma_attr.la_size);
+
+        ma->ma_need = MA_INODE;
+        ma->ma_valid = 0;
+        rc = mo_attr_get(info->mti_env, mdt_object_child(tgt), ma);
+        if (rc) {
+                CERROR("%s: "DFID" attr_get, rc = %d\n",
+                      req->rq_export->exp_obd->obd_name,
+                      PFID(mdt_object_fid(tgt)), rc);
+                return;
+        }
+
+        lprocfs_oh_tally_log2(&rstats->hist[RENAME_CROSSDIR_TGT_SIZE],
+                              (unsigned int)ma->ma_attr.la_size);
+}
+
 int mdt_procfs_init(struct mdt_device *mdt, const char *name)
 {
         struct lu_device *ld = &mdt->mdt_md_dev.md_lu_dev;
-        int result;
+        struct obd_device *obd = ld->ld_obd;
+        struct lprocfs_static_vars lvars;
+        int rc;
         ENTRY;
 
         LASSERT(name != NULL);
-        mdt->mdt_proc_entry = ld->ld_obd->obd_proc_entry;
+
+        lprocfs_mdt_init_vars(&lvars);
+        rc = lprocfs_obd_setup(obd, lvars.obd_vars);
+        if (rc) {
+                CERROR("Can't init lprocfs, rc %d\n", rc);
+                return rc;
+        }
+        ptlrpc_lprocfs_register_obd(obd);
+
+        mdt->mdt_proc_entry = obd->obd_proc_entry;
         LASSERT(mdt->mdt_proc_entry != NULL);
 
-        result = lu_time_init(&mdt->mdt_stats, mdt->mdt_proc_entry,
-                              mdt_proc_names, ARRAY_SIZE(mdt_proc_names));
-        if (result == 0)
-                result = lu_time_named_init(&ld->ld_site->ls_time_stats,
-                                            "site_time", mdt->mdt_proc_entry,
-                                            lu_time_names,
-                                            ARRAY_SIZE(lu_time_names));
-        RETURN(result);
+        rc = lu_time_init(&mdt->mdt_stats, mdt->mdt_proc_entry,
+                          mdt_proc_names, ARRAY_SIZE(mdt_proc_names));
+        if (rc == 0)
+                rc = lu_time_named_init(&ld->ld_site->ls_time_stats,
+                                        "site_time", mdt->mdt_proc_entry,
+                                         lu_time_names,
+                                         ARRAY_SIZE(lu_time_names));
+        if (rc)
+                return rc;
+
+        obd->obd_proc_exports_entry = proc_mkdir("exports",
+                                                 obd->obd_proc_entry);
+        if (obd->obd_proc_exports_entry)
+                lprocfs_add_simple(obd->obd_proc_exports_entry,
+                                   "clear", lprocfs_nid_stats_clear_read,
+                                   lprocfs_nid_stats_clear_write, obd, NULL);
+        rc = lprocfs_alloc_md_stats(obd, LPROC_MDT_LAST);
+       if (rc)
+               return rc;
+       mdt_stats_counter_init(obd->md_stats);
+
+       rc = lprocfs_job_stats_init(obd, LPROC_MDT_LAST,
+                                   mdt_stats_counter_init);
+
+        rc = lproc_mdt_attach_rename_seqstat(mdt);
+        if (rc)
+                CERROR("%s: MDT can not create rename stats rc = %d\n",
+                       obd->obd_name, rc);
+
+        RETURN(rc);
 }
 
 int mdt_procfs_fini(struct mdt_device *mdt)
 {
         struct lu_device *ld = &mdt->mdt_md_dev.md_lu_dev;
-        lu_time_fini(&ld->ld_site->ls_time_stats);
-        lu_time_fini(&mdt->mdt_stats);
-        mdt->mdt_proc_entry = NULL;
+        struct obd_device *obd = ld->ld_obd;
+
+       lprocfs_job_stats_fini(obd);
+
+        if (obd->obd_proc_exports_entry) {
+                lprocfs_remove_proc_entry("clear", obd->obd_proc_exports_entry);
+                obd->obd_proc_exports_entry = NULL;
+        }
+        lprocfs_free_per_client_stats(obd);
+        lprocfs_obd_cleanup(obd);
+        ptlrpc_lprocfs_unregister_obd(obd);
+        if (mdt->mdt_proc_entry) {
+                lu_time_fini(&ld->ld_site->ls_time_stats);
+                lu_time_fini(&mdt->mdt_stats);
+                mdt->mdt_proc_entry = NULL;
+        }
+        lprocfs_free_md_stats(obd);
+        lprocfs_free_obd_stats(obd);
+
         RETURN(0);
 }
 
@@ -104,8 +320,8 @@ static int lprocfs_rd_identity_expire(char *page, char **start, off_t off,
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
 
         *eof = 1;
-        return snprintf(page, count, "%lu\n",
-                        mdt->mdt_identity_cache->uc_entry_expire / HZ);
+        return snprintf(page, count, "%u\n",
+                        mdt->mdt_identity_cache->uc_entry_expire);
 }
 
 static int lprocfs_wr_identity_expire(struct file *file, const char *buffer,
@@ -119,7 +335,7 @@ static int lprocfs_wr_identity_expire(struct file *file, const char *buffer,
         if (rc)
                 return rc;
 
-        mdt->mdt_identity_cache->uc_entry_expire = val * HZ;
+        mdt->mdt_identity_cache->uc_entry_expire = val;
         return count;
 }
 
@@ -131,8 +347,8 @@ static int lprocfs_rd_identity_acquire_expire(char *page, char **start,
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
 
         *eof = 1;
-        return snprintf(page, count, "%lu\n",
-                        mdt->mdt_identity_cache->uc_acquire_expire / HZ);
+        return snprintf(page, count, "%u\n",
+                        mdt->mdt_identity_cache->uc_acquire_expire);
 }
 
 static int lprocfs_wr_identity_acquire_expire(struct file *file,
@@ -148,7 +364,7 @@ static int lprocfs_wr_identity_acquire_expire(struct file *file,
         if (rc)
                 return rc;
 
-        mdt->mdt_identity_cache->uc_acquire_expire = val * HZ;
+        mdt->mdt_identity_cache->uc_acquire_expire = val;
         return count;
 }
 
@@ -157,10 +373,14 @@ static int lprocfs_rd_identity_upcall(char *page, char **start, off_t off,
 {
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
+        struct upcall_cache *hash = mdt->mdt_identity_cache;
+        int len;
 
         *eof = 1;
-        return snprintf(page, count, "%s\n",
-                        mdt->mdt_identity_cache->uc_upcall);
+        cfs_read_lock(&hash->uc_upcall_rwlock);
+        len = snprintf(page, count, "%s\n", hash->uc_upcall);
+        cfs_read_unlock(&hash->uc_upcall_rwlock);
+        return len;
 }
 
 static int lprocfs_wr_identity_upcall(struct file *file, const char *buffer,
@@ -169,26 +389,40 @@ static int lprocfs_wr_identity_upcall(struct file *file, const char *buffer,
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
         struct upcall_cache *hash = mdt->mdt_identity_cache;
-        char kernbuf[UC_CACHE_UPCALL_MAXPATH] = { '\0' };
+        int rc;
+        char *kernbuf;
 
         if (count >= UC_CACHE_UPCALL_MAXPATH) {
                 CERROR("%s: identity upcall too long\n", obd->obd_name);
                 return -EINVAL;
         }
-
-        if (copy_from_user(kernbuf, buffer,
-                           min(count, UC_CACHE_UPCALL_MAXPATH - 1)))
-                return -EFAULT;
+        OBD_ALLOC(kernbuf, count + 1);
+        if (kernbuf == NULL)
+                GOTO(failed, rc = -ENOMEM);
+        if (cfs_copy_from_user(kernbuf, buffer, count))
+                GOTO(failed, rc = -EFAULT);
 
         /* Remove any extraneous bits from the upcall (e.g. linefeeds) */
+        cfs_write_lock(&hash->uc_upcall_rwlock);
         sscanf(kernbuf, "%s", hash->uc_upcall);
+        cfs_write_unlock(&hash->uc_upcall_rwlock);
 
         if (strcmp(hash->uc_name, obd->obd_name) != 0)
                 CWARN("%s: write to upcall name %s\n",
                       obd->obd_name, hash->uc_upcall);
+
+        if (strcmp(hash->uc_upcall, "NONE") == 0 && mdt->mdt_opts.mo_acl)
+                CWARN("%s: disable \"identity_upcall\" with ACL enabled maybe "
+                      "cause unexpected \"EACCESS\"\n", obd->obd_name);
+
         CWARN("%s: identity upcall set to %s\n", obd->obd_name, hash->uc_upcall);
+        OBD_FREE(kernbuf, count + 1);
+        RETURN(count);
 
-        return count;
+ failed:
+        if (kernbuf)
+                OBD_FREE(kernbuf, count + 1);
+        RETURN(rc);
 }
 
 static int lprocfs_wr_identity_flush(struct file *file, const char *buffer,
@@ -211,52 +445,51 @@ static int lprocfs_wr_identity_info(struct file *file, const char *buffer,
 {
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
-        struct identity_downcall_data sparam, *param = &sparam;
-        int size = 0, rc = 0;
+        struct identity_downcall_data *param;
+        int size = sizeof(*param), rc, checked = 0;
 
-        if (count < sizeof(*param)) {
-                CERROR("%s: invalid data size %lu\n", obd->obd_name, count);
-                return count;
+again:
+        if (count < size) {
+                CERROR("%s: invalid data count = %lu, size = %d\n",
+                       obd->obd_name, count, size);
+                return -EINVAL;
         }
 
-        if (copy_from_user(&sparam, buffer, sizeof(sparam))) {
+        OBD_ALLOC(param, size);
+        if (param == NULL)
+                return -ENOMEM;
+
+        if (cfs_copy_from_user(param, buffer, size)) {
                 CERROR("%s: bad identity data\n", obd->obd_name);
                 GOTO(out, rc = -EFAULT);
         }
 
-        if (sparam.idd_magic != IDENTITY_DOWNCALL_MAGIC) {
-                CERROR("%s: MDS identity downcall bad params\n", obd->obd_name);
-                GOTO(out, rc = -EINVAL);
-        }
+        if (checked == 0) {
+                checked = 1;
+                if (param->idd_magic != IDENTITY_DOWNCALL_MAGIC) {
+                        CERROR("%s: MDS identity downcall bad params\n",
+                               obd->obd_name);
+                        GOTO(out, rc = -EINVAL);
+                }
 
-        if (sparam.idd_nperms > N_SETXID_PERMS_MAX) {
-                CERROR("%s: perm count %d more than maximum %d\n",
-                       obd->obd_name, sparam.idd_nperms, N_SETXID_PERMS_MAX);
-                GOTO(out, rc = -EINVAL);
-        }
+                if (param->idd_nperms > N_PERMS_MAX) {
+                        CERROR("%s: perm count %d more than maximum %d\n",
+                               obd->obd_name, param->idd_nperms, N_PERMS_MAX);
+                        GOTO(out, rc = -EINVAL);
+                }
 
-        if (sparam.idd_ngroups > NGROUPS_MAX) {
-                CERROR("%s: group count %d more than maximum %d\n",
-                       obd->obd_name, sparam.idd_ngroups, NGROUPS_MAX);
-                GOTO(out, rc = -EINVAL);
-        }
+                if (param->idd_ngroups > NGROUPS_MAX) {
+                        CERROR("%s: group count %d more than maximum %d\n",
+                               obd->obd_name, param->idd_ngroups, NGROUPS_MAX);
+                        GOTO(out, rc = -EINVAL);
+                }
 
-        if (sparam.idd_ngroups) {
-                size = offsetof(struct identity_downcall_data,
-                                idd_groups[sparam.idd_ngroups]);
-                OBD_ALLOC(param, size);
-                if (!param) {
-                        CERROR("%s: fail to alloc %d bytes for uid %u"
-                               " with %d groups\n", obd->obd_name, size,
-                               sparam.idd_uid, sparam.idd_ngroups);
-                        param = &sparam;
-                        param->idd_ngroups = 0;
-                } else if (copy_from_user(param, buffer, size)) {
-                        CERROR("%s: uid %u bad supplementary group data\n",
-                               obd->obd_name, sparam.idd_uid);
+                if (param->idd_ngroups) {
+                        rc = param->idd_ngroups; /* save idd_ngroups */
                         OBD_FREE(param, size);
-                        param = &sparam;
-                        param->idd_ngroups = 0;
+                        size = offsetof(struct identity_downcall_data,
+                                        idd_groups[rc]);
+                        goto again;
                 }
         }
 
@@ -264,171 +497,179 @@ static int lprocfs_wr_identity_info(struct file *file, const char *buffer,
                                    param->idd_uid, param);
 
 out:
-        if (param && (param != &sparam))
+        if (param != NULL)
                 OBD_FREE(param, size);
 
-        return rc ?: count;
+        return rc ? rc : count;
 }
 
-static int lprocfs_rd_rmtacl_expire(char *page, char **start, off_t off,
-                                    int count, int *eof, void *data)
+/* for debug only */
+static int lprocfs_rd_capa(char *page, char **start, off_t off,
+                           int count, int *eof, void *data)
 {
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
 
-        *eof = 1;
-        return snprintf(page, count, "%lu\n",
-                        mdt->mdt_rmtacl_cache->uc_entry_expire / HZ);
+        return snprintf(page, count, "capability on: %s %s\n",
+                        mdt->mdt_opts.mo_oss_capa ? "oss" : "",
+                        mdt->mdt_opts.mo_mds_capa ? "mds" : "");
 }
 
-static int lprocfs_wr_rmtacl_expire(struct file *file, const char *buffer,
-                                    unsigned long count, void *data)
+static int lprocfs_wr_capa(struct file *file, const char *buffer,
+                           unsigned long count, void *data)
 {
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
-        int rc, val;
+        int val, rc;
 
         rc = lprocfs_write_helper(buffer, count, &val);
         if (rc)
                 return rc;
 
-        mdt->mdt_rmtacl_cache->uc_entry_expire = val * HZ;
+        if (val < 0 || val > 3) {
+                CERROR("invalid capability mode, only 0/2/3 is accepted.\n"
+                       " 0:  disable fid capability\n"
+                       " 2:  enable MDS fid capability\n"
+                       " 3:  enable both MDS and OSS fid capability\n");
+                return -EINVAL;
+        }
+
+        /* OSS fid capability needs enable both MDS and OSS fid capability on
+         * MDS */
+        if (val == 1) {
+                CERROR("can't enable OSS fid capability only, you should use "
+                       "'3' to enable both MDS and OSS fid capability.\n");
+                return -EINVAL;
+        }
+
+        mdt->mdt_opts.mo_oss_capa = (val & 0x1);
+        mdt->mdt_opts.mo_mds_capa = !!(val & 0x2);
+        mdt->mdt_capa_conf = 1;
+        LCONSOLE_INFO("MDS %s %s MDS fid capability.\n",
+                      obd->obd_name,
+                      mdt->mdt_opts.mo_mds_capa ? "enabled" : "disabled");
+        LCONSOLE_INFO("MDS %s %s OSS fid capability.\n",
+                      obd->obd_name,
+                      mdt->mdt_opts.mo_oss_capa ? "enabled" : "disabled");
         return count;
 }
 
-static int lprocfs_rd_rmtacl_acquire_expire(char *page, char **start,
-                                            off_t off, int count, int *eof,
-                                            void *data)
+static int lprocfs_rd_capa_count(char *page, char **start, off_t off,
+                                 int count, int *eof, void *data)
+{
+        return snprintf(page, count, "%d %d\n",
+                        capa_count[CAPA_SITE_CLIENT],
+                        capa_count[CAPA_SITE_SERVER]);
+}
+
+static int lprocfs_rd_site_stats(char *page, char **start, off_t off,
+                                 int count, int *eof, void *data)
 {
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
 
-        *eof = 1;
-        return snprintf(page, count, "%lu\n",
-                        mdt->mdt_rmtacl_cache->uc_acquire_expire / HZ);
+        return lu_site_stats_print(mdt_lu_site(mdt), page, count);
 }
 
-static int lprocfs_wr_rmtacl_acquire_expire(struct file *file,
-                                            const char *buffer,
-                                            unsigned long count,
-                                            void *data)
+static int lprocfs_rd_capa_timeout(char *page, char **start, off_t off,
+                                   int count, int *eof, void *data)
 {
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
-        int rc, val;
+
+        return snprintf(page, count, "%lu\n", mdt->mdt_capa_timeout);
+}
+
+static int lprocfs_wr_capa_timeout(struct file *file, const char *buffer,
+                                   unsigned long count, void *data)
+{
+        struct obd_device *obd = data;
+        struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
+        int val, rc;
 
         rc = lprocfs_write_helper(buffer, count, &val);
         if (rc)
                 return rc;
 
-        mdt->mdt_rmtacl_cache->uc_acquire_expire = val * HZ;
+        mdt->mdt_capa_timeout = (unsigned long)val;
+        mdt->mdt_capa_conf = 1;
         return count;
 }
 
-static int lprocfs_rd_rmtacl_upcall(char *page, char **start, off_t off,
-                                      int count, int *eof, void *data)
+static int lprocfs_rd_ck_timeout(char *page, char **start, off_t off, int count,
+                                 int *eof, void *data)
 {
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
 
-        *eof = 1;
-        return snprintf(page, count, "%s\n",
-                        mdt->mdt_rmtacl_cache->uc_upcall);
+        return snprintf(page, count, "%lu\n", mdt->mdt_ck_timeout);
 }
 
-static int lprocfs_wr_rmtacl_upcall(struct file *file, const char *buffer,
-                                      unsigned long count, void *data)
+static int lprocfs_wr_ck_timeout(struct file *file, const char *buffer,
+                                 unsigned long count, void *data)
 {
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
-        struct upcall_cache *hash = mdt->mdt_rmtacl_cache;
-        char kernbuf[UC_CACHE_UPCALL_MAXPATH] = { '\0' };
-
-        if (count >= UC_CACHE_UPCALL_MAXPATH) {
-                CERROR("%s: remote ACL upcall too long\n", obd->obd_name);
-                return -EINVAL;
-        }
-
-        if (copy_from_user(kernbuf, buffer,
-                           min(count, UC_CACHE_UPCALL_MAXPATH - 1)))
-                return -EFAULT;
-
-        /* Remove any extraneous bits from the upcall (e.g. linefeeds) */
-        sscanf(kernbuf, "%s", hash->uc_upcall);
+        int val, rc;
 
-        if (strcmp(hash->uc_name, obd->obd_name) != 0)
-                CWARN("%s: write to upcall name %s\n",
-                      obd->obd_name, hash->uc_upcall);
-        CWARN("%s: remote ACL upcall set to %s\n", obd->obd_name, hash->uc_upcall);
+        rc = lprocfs_write_helper(buffer, count, &val);
+        if (rc)
+                return rc;
 
+        mdt->mdt_ck_timeout = (unsigned long)val;
+        mdt->mdt_capa_conf = 1;
         return count;
 }
 
-static int lprocfs_wr_rmtacl_info(struct file *file, const char *buffer,
-                                  unsigned long count, void *data)
-{
-        struct obd_device *obd = data;
-        struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
-        struct rmtacl_downcall_data sparam, *param = &sparam;
-        int size = 0, rc = 0;
+#define BUFLEN (UUID_MAX + 4)
 
-        if (count < sizeof(*param)) {
-                CERROR("%s: invalid data size %lu\n", obd->obd_name, count);
-                return count;
-        }
+static int lprocfs_mdt_wr_evict_client(struct file *file, const char *buffer,
+                                       unsigned long count, void *data)
+{
+        char *kbuf;
+        char *tmpbuf;
 
-        if (copy_from_user(&sparam, buffer, sizeof(sparam))) {
-                CERROR("%s: bad remote acl data\n", obd->obd_name);
-                GOTO(out, rc = -EFAULT);
-        }
+        OBD_ALLOC(kbuf, BUFLEN);
+        if (kbuf == NULL)
+                return -ENOMEM;
 
-        if (sparam.add_magic != RMTACL_DOWNCALL_MAGIC) {
-                CERROR("%s: MDT remote acl downcall bad params\n", obd->obd_name);
-                GOTO(out, rc = -EINVAL);
+        /*
+         * OBD_ALLOC() will zero kbuf, but we only copy BUFLEN - 1
+         * bytes into kbuf, to ensure that the string is NUL-terminated.
+         * UUID_MAX should include a trailing NUL already.
+         */
+        if (cfs_copy_from_user(kbuf, buffer,
+                               min_t(unsigned long, BUFLEN - 1, count))) {
+                count = -EFAULT;
+                goto out;
         }
+        tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, BUFLEN - 1, count));
 
-        if (sparam.add_buflen) {
-                size = offsetof(struct rmtacl_downcall_data,
-                                add_buf[sparam.add_buflen]);
-                OBD_ALLOC(param, size);
-                if (!param) {
-                        CERROR("%s: fail to alloc %d bytes for ino "LPU64"\n",
-                               obd->obd_name, size, sparam.add_key);
-                        param = &sparam;
-                        param->add_buflen = 0;
-                } else if (copy_from_user(param, buffer, size)) {
-                        CERROR("%s: ino "LPU64" bad remote acl data\n",
-                               obd->obd_name, sparam.add_key);
-                        OBD_FREE(param, size);
-                        param = &sparam;
-                        param->add_buflen = 0;
-                }
+        if (strncmp(tmpbuf, "nid:", 4) != 0) {
+                count = lprocfs_wr_evict_client(file, buffer, count, data);
+                goto out;
         }
 
-        rc = upcall_cache_downcall(mdt->mdt_rmtacl_cache, 0, param->add_key,
-                                   param);
+        CERROR("NOT implement evict client by nid %s\n", tmpbuf);
 
 out:
-        if (param && (param != &sparam))
-                OBD_FREE(param, size);
-
-        return rc ?: count;
+        OBD_FREE(kbuf, BUFLEN);
+        return count;
 }
 
-static int lprocfs_rd_rootsquash_uid(char *page, char **start, off_t off,
-                                 int count, int *eof, void *data)
+#undef BUFLEN
+
+static int lprocfs_rd_sec_level(char *page, char **start, off_t off,
+                                int count, int *eof, void *data)
 {
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
-        struct rootsquash_info *rsi = mdt->mdt_rootsquash_info;
 
-        *eof = 1;
-        return snprintf(page, count, "%u\n",
-                        rsi ? rsi->rsi_uid : 0);
+        return snprintf(page, count, "%d\n", mdt->mdt_sec_level);
 }
 
-static int lprocfs_wr_rootsquash_uid(struct file *file, const char *buffer,
-                                     unsigned long count, void *data)
+static int lprocfs_wr_sec_level(struct file *file, const char *buffer,
+                                unsigned long count, void *data)
 {
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
@@ -438,29 +679,30 @@ static int lprocfs_wr_rootsquash_uid(struct file *file, const char *buffer,
         if (rc)
                 return rc;
 
-        if (!mdt->mdt_rootsquash_info)
-                OBD_ALLOC_PTR(mdt->mdt_rootsquash_info);
-        if (!mdt->mdt_rootsquash_info)
-                return -ENOMEM;
+        if (val > LUSTRE_SEC_ALL || val < LUSTRE_SEC_NONE)
+                return -EINVAL;
 
-        mdt->mdt_rootsquash_info->rsi_uid = val;
+        if (val == LUSTRE_SEC_SPECIFY) {
+                CWARN("security level %d will be supported in future.\n",
+                      LUSTRE_SEC_SPECIFY);
+                return -EINVAL;
+        }
+
+        mdt->mdt_sec_level = val;
         return count;
 }
 
-static int lprocfs_rd_rootsquash_gid(char *page, char **start, off_t off,
-                                 int count, int *eof, void *data)
+static int lprocfs_rd_cos(char *page, char **start, off_t off,
+                              int count, int *eof, void *data)
 {
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
-        struct rootsquash_info *rsi = mdt->mdt_rootsquash_info;
 
-        *eof = 1;
-        return snprintf(page, count, "%u\n",
-                        rsi ? rsi->rsi_gid : 0);
+        return snprintf(page, count, "%u\n", mdt_cos_is_enabled(mdt));
 }
 
-static int lprocfs_wr_rootsquash_gid(struct file *file, const char *buffer,
-                                     unsigned long count, void *data)
+static int lprocfs_wr_cos(struct file *file, const char *buffer,
+                                  unsigned long count, void *data)
 {
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
@@ -469,252 +711,275 @@ static int lprocfs_wr_rootsquash_gid(struct file *file, const char *buffer,
         rc = lprocfs_write_helper(buffer, count, &val);
         if (rc)
                 return rc;
-
-        if (!mdt->mdt_rootsquash_info)
-                OBD_ALLOC_PTR(mdt->mdt_rootsquash_info);
-        if (!mdt->mdt_rootsquash_info)
-                return -ENOMEM;
-
-        mdt->mdt_rootsquash_info->rsi_gid = val;
+        mdt_enable_cos(mdt, val);
         return count;
 }
 
-static int lprocfs_rd_nosquash_nids(char *page, char **start, off_t off,
-                                       int count, int *eof, void *data)
+static int lprocfs_rd_root_squash(char *page, char **start, off_t off,
+                                  int count, int *eof, void *data)
 {
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
-        struct rootsquash_info *rsi = mdt->mdt_rootsquash_info;
-        int i, ret;
-
-        ret = snprintf(page, count, "rootsquash skip list:\n");
-        for (i = 0; rsi && (i < rsi->rsi_n_nosquash_nids); i++) {
-                ret += snprintf(page + ret, count - ret, "%s\n",
-                                libcfs_nid2str(rsi->rsi_nosquash_nids[i]));
-        }
 
-        *eof = 1;
-        return ret;
+        return snprintf(page, count, "%u:%u\n", mdt->mdt_squash_uid,
+                        mdt->mdt_squash_gid);
 }
 
-static inline void remove_newline(char *str)
+static int safe_strtoul(const char *str, char **endp, unsigned long *res)
 {
-        int len = strlen(str);
-
-        if (str[len - 1] == '\n')
-                str[len - 1] = '\0';
-}
+        char n[24];
 
-/* XXX: This macro is copied from lnet/libcfs/nidstring.c */
-#define LNET_NIDSTR_SIZE   32      /* size of each one (see below for usage) */
+        *res = simple_strtoul(str, endp, 0);
+        if (str == *endp)
+                return 1;
 
-static void do_process_nosquash_nids(struct mdt_device *m, char *buf)
-{
-        struct rootsquash_info *rsi = m->mdt_rootsquash_info;
-        char str[LNET_NIDSTR_SIZE], *end;
-        lnet_nid_t nid;
-
-        LASSERT(rsi);
-        rsi->rsi_n_nosquash_nids = 0;
-        while (rsi->rsi_n_nosquash_nids < N_NOSQUASH_NIDS) {
-                end = strchr(buf, ',');
-                memset(str, 0, sizeof(str));
-                if (end)
-                        strncpy(str, buf, min_t(int, sizeof(str), end - buf));
-                else
-                        strncpy(str, buf, min_t(int, sizeof(str), strlen(buf)));
-
-                if (!strcmp(str, "*")) {
-                        nid = LNET_NID_ANY;
-                } else {
-                        nid = libcfs_str2nid(str);
-                        if (nid == LNET_NID_ANY)
-                                goto ignore;
-                }
-                rsi->rsi_nosquash_nids[rsi->rsi_n_nosquash_nids++] = nid;
-ignore:
-                if (!end || (*(end + 1) == 0))
-                        return;
-                buf = end + 1;
-        }
+        sprintf(n, "%lu", *res);
+        if (strncmp(n, str, *endp - str))
+                /* overflow */
+                return 1;
+        return 0;
 }
 
-static int lprocfs_wr_nosquash_nids(struct file *file, const char *buffer,
-                                       unsigned long count, void *data)
+static int lprocfs_wr_root_squash(struct file *file, const char *buffer,
+                                  unsigned long count, void *data)
 {
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
-        char skips[LNET_NIDSTR_SIZE * N_NOSQUASH_NIDS] = "";
-        unsigned long size = sizeof(skips);
+        int rc;
+        char kernbuf[50], *tmp, *end, *errmsg;
+        unsigned long uid, gid;
+        int nouid, nogid;
+        ENTRY;
 
-        if (count > size) {
-                CERROR("parameter exceeds max limit %lu\n", size);
-                return -EINVAL;
+        if (count >= sizeof(kernbuf)) {
+                errmsg = "string too long";
+                GOTO(failed, rc = -EINVAL);
+        }
+        if (cfs_copy_from_user(kernbuf, buffer, count)) {
+                errmsg = "bad address";
+                GOTO(failed, rc = -EFAULT);
         }
+        kernbuf[count] = '\0';
 
-        if (copy_from_user(skips, buffer, min(size, count)))
-                return -EFAULT;
+        nouid = nogid = 0;
+        if (safe_strtoul(buffer, &tmp, &uid)) {
+                uid = mdt->mdt_squash_uid;
+                nouid = 1;
+        }
 
-        if (!mdt->mdt_rootsquash_info)
-                OBD_ALLOC_PTR(mdt->mdt_rootsquash_info);
-        if (!mdt->mdt_rootsquash_info)
-                return -ENOMEM;
+        /* skip ':' */
+        if (*tmp == ':') {
+                tmp++;
+                if (safe_strtoul(tmp, &end, &gid)) {
+                        gid = mdt->mdt_squash_gid;
+                        nogid = 1;
+                }
+        } else {
+                gid = mdt->mdt_squash_gid;
+                nogid = 1;
+        }
 
-        remove_newline(skips);
-        do_process_nosquash_nids(mdt, skips);
-        return count;
+        mdt->mdt_squash_uid = uid;
+        mdt->mdt_squash_gid = gid;
+
+        if (nouid && nogid) {
+                errmsg = "needs uid:gid format";
+                GOTO(failed, rc = -EINVAL);
+        }
+
+        LCONSOLE_INFO("%s: root_squash is set to %u:%u\n",
+                      obd->obd_name,
+                      mdt->mdt_squash_uid,  mdt->mdt_squash_gid);
+        RETURN(count);
+
+ failed:
+        CWARN("%s: failed to set root_squash to \"%s\", %s: rc %d\n",
+              obd->obd_name, buffer, errmsg, rc);
+        RETURN(rc);
 }
 
-/* for debug only */
-static int lprocfs_rd_capa(char *page, char **start, off_t off,
-                           int count, int *eof, void *data)
+static int lprocfs_rd_nosquash_nids(char *page, char **start, off_t off,
+                                    int count, int *eof, void *data)
 {
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
 
-        return snprintf(page, count, "capability on: %s %s\n",
-                        mdt->mdt_opts.mo_oss_capa ? "oss" : "",
-                        mdt->mdt_opts.mo_mds_capa ? "mds" : "");
+        if (mdt->mdt_nosquash_str)
+                return snprintf(page, count, "%s\n", mdt->mdt_nosquash_str);
+        return snprintf(page, count, "NONE\n");
 }
 
-static int lprocfs_wr_capa(struct file *file, const char *buffer,
-                           unsigned long count, void *data)
+static int lprocfs_wr_nosquash_nids(struct file *file, const char *buffer,
+                                    unsigned long count, void *data)
 {
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
-        int val, rc;
-
-        rc = lprocfs_write_helper(buffer, count, &val);
-        if (rc)
-                return rc;
+        int rc;
+        char *kernbuf, *errmsg;
+        cfs_list_t tmp;
+        ENTRY;
 
-        if (val < 0 || val > 3) {
-                CERROR("invalid capability mode, only 0/2/3 is accepted.\n"
-                       " 0:  disable fid capability\n"
-                       " 2:  enable MDS fid capability\n"
-                       " 3:  enable both MDS and OSS fid capability\n");
-                return -EINVAL;
+        OBD_ALLOC(kernbuf, count + 1);
+        if (kernbuf == NULL) {
+                errmsg = "no memory";
+                GOTO(failed, rc = -ENOMEM);
         }
-
-        /* OSS fid capability needs enable both MDS and OSS fid capability on
-         * MDS */
-        if (val == 1) {
-                CERROR("can't enable OSS fid capability only, you should use "
-                       "'3' to enable both MDS and OSS fid capability.\n");
-                return -EINVAL;
+        if (cfs_copy_from_user(kernbuf, buffer, count)) {
+                errmsg = "bad address";
+                GOTO(failed, rc = -EFAULT);
+        }
+        kernbuf[count] = '\0';
+
+        if (!strcmp(kernbuf, "NONE") || !strcmp(kernbuf, "clear")) {
+                /* empty string is special case */
+                cfs_down_write(&mdt->mdt_squash_sem);
+                if (!cfs_list_empty(&mdt->mdt_nosquash_nids)) {
+                        cfs_free_nidlist(&mdt->mdt_nosquash_nids);
+                        OBD_FREE(mdt->mdt_nosquash_str,
+                                 mdt->mdt_nosquash_strlen);
+                        mdt->mdt_nosquash_str = NULL;
+                        mdt->mdt_nosquash_strlen = 0;
+                }
+                cfs_up_write(&mdt->mdt_squash_sem);
+                LCONSOLE_INFO("%s: nosquash_nids is cleared\n",
+                              obd->obd_name);
+                OBD_FREE(kernbuf, count + 1);
+                RETURN(count);
         }
 
-        mdt->mdt_opts.mo_oss_capa = (val & 0x1);
-        mdt->mdt_opts.mo_mds_capa = !!(val & 0x2);
-        mdt->mdt_capa_conf = 1;
-        LCONSOLE_INFO("MDS %s %s MDS fid capability.\n",
-                      obd->obd_name,
-                      mdt->mdt_opts.mo_mds_capa ? "enabled" : "disabled");
-        LCONSOLE_INFO("MDS %s %s OSS fid capability.\n",
-                      obd->obd_name,
-                      mdt->mdt_opts.mo_oss_capa ? "enabled" : "disabled");
-        return count;
-}
+        CFS_INIT_LIST_HEAD(&tmp);
+        if (cfs_parse_nidlist(kernbuf, count, &tmp) <= 0) {
+                errmsg = "can't parse";
+                GOTO(failed, rc = -EINVAL);
+        }
 
-static int lprocfs_rd_capa_count(char *page, char **start, off_t off,
-                                 int count, int *eof, void *data)
-{
-        return snprintf(page, count, "%d %d\n",
-                        capa_count[CAPA_SITE_CLIENT],
-                        capa_count[CAPA_SITE_SERVER]);
+        cfs_down_write(&mdt->mdt_squash_sem);
+        if (!cfs_list_empty(&mdt->mdt_nosquash_nids)) {
+                cfs_free_nidlist(&mdt->mdt_nosquash_nids);
+                OBD_FREE(mdt->mdt_nosquash_str, mdt->mdt_nosquash_strlen);
+        }
+        mdt->mdt_nosquash_str = kernbuf;
+        mdt->mdt_nosquash_strlen = count + 1;
+        cfs_list_splice(&tmp, &mdt->mdt_nosquash_nids);
+
+        LCONSOLE_INFO("%s: nosquash_nids is set to %s\n",
+                      obd->obd_name, kernbuf);
+        cfs_up_write(&mdt->mdt_squash_sem);
+        RETURN(count);
+
+ failed:
+        CWARN("%s: failed to set nosquash_nids to \"%s\", %s: rc %d\n",
+              obd->obd_name, kernbuf, errmsg, rc);
+        if (kernbuf)
+                OBD_FREE(kernbuf, count + 1);
+        RETURN(rc);
 }
 
-static int lprocfs_rd_site_stats(char *page, char **start, off_t off,
-                                 int count, int *eof, void *data)
+static int lprocfs_rd_mdt_som(char *page, char **start, off_t off,
+                              int count, int *eof, void *data)
 {
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
-        struct lu_site    *s   = mdt->mdt_md_dev.md_lu_dev.ld_site;
-        int i;
-        int populated;
 
-        /*
-         * How many hash buckets are not-empty? Don't bother with locks: it's
-         * an estimation anyway.
-         */
-        for (i = 0, populated = 0; i < s->ls_hash_size; i++)
-                populated += !hlist_empty(&s->ls_hash[i]);
-
-        return snprintf(page, count, "%d %d %d/%d %d %d %d %d %d %d\n",
-                        s->ls_total,
-                        s->ls_busy,
-                        populated,
-                        s->ls_hash_size,
-                        s->ls_stats.s_created,
-                        s->ls_stats.s_cache_hit,
-                        s->ls_stats.s_cache_miss,
-                        s->ls_stats.s_cache_check,
-                        s->ls_stats.s_cache_race,
-                        s->ls_stats.s_lru_purged);
+        return snprintf(page, count, "%sabled\n",
+                        mdt->mdt_som_conf ? "en" : "dis");
 }
 
-static int lprocfs_rd_capa_timeout(char *page, char **start, off_t off,
-                                   int count, int *eof, void *data)
+static int lprocfs_wr_mdt_som(struct file *file, const char *buffer,
+                              unsigned long count, void *data)
 {
+        struct obd_export *exp;
         struct obd_device *obd = data;
         struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
+        char kernbuf[16];
+        unsigned long val = 0;
 
-        return snprintf(page, count, "%lu\n", mdt->mdt_capa_timeout);
-}
+        if (count > (sizeof(kernbuf) - 1))
+                return -EINVAL;
 
-static int lprocfs_wr_capa_timeout(struct file *file, const char *buffer,
-                                   unsigned long count, void *data)
-{
-        struct obd_device *obd = data;
-        struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
-        int val, rc;
+        if (cfs_copy_from_user(kernbuf, buffer, count))
+                return -EFAULT;
 
-        rc = lprocfs_write_helper(buffer, count, &val);
-        if (rc)
-                return rc;
+        kernbuf[count] = '\0';
 
-        mdt->mdt_capa_timeout = (unsigned long)val;
-        mdt->mdt_capa_conf = 1;
-        return count;
-}
+        if (!strcmp(kernbuf, "enabled"))
+                val = 1;
+        else if (strcmp(kernbuf, "disabled"))
+                return -EINVAL;
 
-static int lprocfs_rd_ck_timeout(char *page, char **start, off_t off, int count,
-                                 int *eof, void *data)
-{
-        struct obd_device *obd = data;
-        struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
+        if (mdt->mdt_som_conf == val)
+                return count;
 
-        return snprintf(page, count, "%lu\n", mdt->mdt_ck_timeout);
-}
+        if (!obd->obd_process_conf) {
+                CERROR("Temporary SOM change is not supported, use lctl "
+                       "conf_param for permanent setting\n");
+                return count;
+        }
 
-static int lprocfs_wr_ck_timeout(struct file *file, const char *buffer,
-                                 unsigned long count, void *data)
-{
-        struct obd_device *obd = data;
-        struct mdt_device *mdt = mdt_dev(obd->obd_lu_dev);
-        int val, rc;
+        /* 1 stands for self export. */
+        cfs_list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
+                if (exp == obd->obd_self_export)
+                        continue;
+                if (exp->exp_connect_flags & OBD_CONNECT_MDS_MDS)
+                        continue;
+                /* Some clients are already connected, skip the change */
+                LCONSOLE_INFO("%s is already connected, SOM will be %s on "
+                              "the next mount\n", exp->exp_client_uuid.uuid,
+                              val ? "enabled" : "disabled");
+                return count;
+        }
 
-        rc = lprocfs_write_helper(buffer, count, &val);
-        if (rc)
-                return rc;
+        mdt->mdt_som_conf = val;
+        LCONSOLE_INFO("Enabling SOM\n");
 
-        mdt->mdt_ck_timeout = (unsigned long)val;
-        mdt->mdt_capa_conf = 1;
         return count;
 }
 
-static int lprocfs_mdt_wr_evict_client(struct file *file, const char *buffer,
-                                       unsigned long count, void *data)
+/* Temporary; for testing purposes only */
+static int lprocfs_mdt_wr_mdc(struct file *file, const char *buffer,
+                              unsigned long count, void *data)
 {
-        char tmpbuf[sizeof(struct obd_uuid)];
+        struct obd_device *obd = data;
+        struct obd_export *exp = NULL;
+        struct obd_uuid   *uuid;
+        char              *kbuf;
+        char              *tmpbuf;
 
-        sscanf(buffer, "%40s", tmpbuf);
+        OBD_ALLOC(kbuf, UUID_MAX);
+        if (kbuf == NULL)
+                return -ENOMEM;
 
-        if (strncmp(tmpbuf, "nid:", 4) != 0)
-                return lprocfs_wr_evict_client(file, buffer, count, data);
+        /*
+         * OBD_ALLOC() will zero kbuf, but we only copy UUID_MAX - 1
+         * bytes into kbuf, to ensure that the string is NUL-terminated.
+         * UUID_MAX should include a trailing NUL already.
+         */
+        if (cfs_copy_from_user(kbuf, buffer,
+                               min_t(unsigned long, UUID_MAX - 1, count))) {
+                count = -EFAULT;
+                goto out;
+        }
+        tmpbuf = cfs_firststr(kbuf, min_t(unsigned long, UUID_MAX - 1, count));
 
-        CERROR("NOT implement evict client by nid %s\n", tmpbuf);
+        OBD_ALLOC(uuid, UUID_MAX);
+        if (uuid == NULL) {
+                count = -ENOMEM;
+                goto out;
+        }
+
+        obd_str2uuid(uuid, tmpbuf);
+        exp = cfs_hash_lookup(obd->obd_uuid_hash, uuid);
+        if (exp == NULL) {
+                CERROR("%s: no export %s found\n",
+                       obd->obd_name, obd_uuid2str(uuid));
+        } else {
+                mdt_hsm_copytool_send(exp);
+                class_export_put(exp);
+        }
 
+        OBD_FREE(uuid, UUID_MAX);
+out:
+        OBD_FREE(kbuf, UUID_MAX);
         return count;
 }
 
@@ -730,19 +995,6 @@ static struct lprocfs_vars lprocfs_mdt_obd_vars[] = {
                                         lprocfs_wr_identity_upcall,         0 },
         { "identity_flush",             0, lprocfs_wr_identity_flush,       0 },
         { "identity_info",              0, lprocfs_wr_identity_info,        0 },
-        { "rmtacl_expire",              lprocfs_rd_rmtacl_expire,
-                                        lprocfs_wr_rmtacl_expire,           0 },
-        { "rmtacl_acquire_expire",      lprocfs_rd_rmtacl_acquire_expire,
-                                        lprocfs_wr_rmtacl_acquire_expire,   0 },
-        { "rmtacl_upcall",              lprocfs_rd_rmtacl_upcall,
-                                        lprocfs_wr_rmtacl_upcall,           0 },
-        { "rmtacl_info",                0, lprocfs_wr_rmtacl_info,          0 },
-        { "rootsquash_uid",             lprocfs_rd_rootsquash_uid,
-                                        lprocfs_wr_rootsquash_uid,          0 },
-        { "rootsquash_gid",             lprocfs_rd_rootsquash_gid,
-                                        lprocfs_wr_rootsquash_gid,          0 },
-        { "nosquash_nids",              lprocfs_rd_nosquash_nids,
-                                        lprocfs_wr_nosquash_nids,           0 },
         { "capa",                       lprocfs_rd_capa,
                                         lprocfs_wr_capa,                    0 },
         { "capa_timeout",               lprocfs_rd_capa_timeout,
@@ -752,6 +1004,22 @@ static struct lprocfs_vars lprocfs_mdt_obd_vars[] = {
         { "capa_count",                 lprocfs_rd_capa_count,           0, 0 },
         { "site_stats",                 lprocfs_rd_site_stats,           0, 0 },
         { "evict_client",               0, lprocfs_mdt_wr_evict_client,     0 },
+        { "hash_stats",                 lprocfs_obd_rd_hash,    0, 0 },
+        { "sec_level",                  lprocfs_rd_sec_level,
+                                        lprocfs_wr_sec_level,               0 },
+        { "commit_on_sharing",          lprocfs_rd_cos, lprocfs_wr_cos, 0 },
+        { "root_squash",                lprocfs_rd_root_squash,
+                                        lprocfs_wr_root_squash,             0 },
+        { "nosquash_nids",              lprocfs_rd_nosquash_nids,
+                                        lprocfs_wr_nosquash_nids,           0 },
+        { "som",                        lprocfs_rd_mdt_som,
+                                        lprocfs_wr_mdt_som, 0 },
+        { "mdccomm",                    0, lprocfs_mdt_wr_mdc,              0 },
+        { "instance",                   lprocfs_target_rd_instance,         0 },
+        { "ir_factor",                  lprocfs_obd_rd_ir_factor,
+                                        lprocfs_obd_wr_ir_factor,           0 },
+       { "job_cleanup_interval",       lprocfs_rd_job_interval,
+                                       lprocfs_wr_job_interval, 0 },
         { 0 }
 };
 
@@ -760,4 +1028,45 @@ static struct lprocfs_vars lprocfs_mdt_module_vars[] = {
         { 0 }
 };
 
-LPROCFS_INIT_VARS(mdt, lprocfs_mdt_module_vars, lprocfs_mdt_obd_vars);
+void lprocfs_mdt_init_vars(struct lprocfs_static_vars *lvars)
+{
+    lvars->module_vars  = lprocfs_mdt_module_vars;
+    lvars->obd_vars     = lprocfs_mdt_obd_vars;
+}
+
+void mdt_counter_incr(struct ptlrpc_request *req, int opcode)
+{
+       struct obd_export *exp = req->rq_export;
+
+       if (exp->exp_obd && exp->exp_obd->md_stats)
+               lprocfs_counter_incr(exp->exp_obd->md_stats, opcode);
+       if (exp->exp_nid_stats && exp->exp_nid_stats->nid_stats != NULL)
+               lprocfs_counter_incr(exp->exp_nid_stats->nid_stats, opcode);
+       if (exp->exp_obd && exp->exp_obd->u.obt.obt_jobstats.ojs_hash &&
+           (exp->exp_connect_flags & OBD_CONNECT_JOBSTATS))
+               lprocfs_job_stats_log(exp->exp_obd,
+                                     lustre_msg_get_jobid(req->rq_reqmsg),
+                                     opcode, 1);
+}
+
+void mdt_stats_counter_init(struct lprocfs_stats *stats)
+{
+        lprocfs_counter_init(stats, LPROC_MDT_OPEN, 0, "open", "reqs");
+        lprocfs_counter_init(stats, LPROC_MDT_CLOSE, 0, "close", "reqs");
+        lprocfs_counter_init(stats, LPROC_MDT_MKNOD, 0, "mknod", "reqs");
+        lprocfs_counter_init(stats, LPROC_MDT_LINK, 0, "link", "reqs");
+        lprocfs_counter_init(stats, LPROC_MDT_UNLINK, 0, "unlink", "reqs");
+        lprocfs_counter_init(stats, LPROC_MDT_MKDIR, 0, "mkdir", "reqs");
+        lprocfs_counter_init(stats, LPROC_MDT_RMDIR, 0, "rmdir", "reqs");
+        lprocfs_counter_init(stats, LPROC_MDT_RENAME, 0, "rename", "reqs");
+        lprocfs_counter_init(stats, LPROC_MDT_GETATTR, 0, "getattr", "reqs");
+        lprocfs_counter_init(stats, LPROC_MDT_SETATTR, 0, "setattr", "reqs");
+        lprocfs_counter_init(stats, LPROC_MDT_GETXATTR, 0, "getxattr", "reqs");
+        lprocfs_counter_init(stats, LPROC_MDT_SETXATTR, 0, "setxattr", "reqs");
+        lprocfs_counter_init(stats, LPROC_MDT_STATFS, 0, "statfs", "reqs");
+        lprocfs_counter_init(stats, LPROC_MDT_SYNC, 0, "sync", "reqs");
+        lprocfs_counter_init(stats, LPROC_MDT_SAMEDIR_RENAME, 0,
+                             "samedir_rename", "reqs");
+        lprocfs_counter_init(stats, LPROC_MDT_CROSSDIR_RENAME, 0,
+                             "crossdir_rename", "reqs");
+}