Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / staging / lustre / lustre / obdclass / obd_mount.c
diff --git a/kernel/drivers/staging/lustre/lustre/obdclass/obd_mount.c b/kernel/drivers/staging/lustre/lustre/obdclass/obd_mount.c
new file mode 100644 (file)
index 0000000..3437b2e
--- /dev/null
@@ -0,0 +1,1319 @@
+/*
+ * 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.
+ *
+ * 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).
+ *
+ * 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.
+ *
+ * GPL HEADER END
+ */
+/*
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
+ * Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2012, Intel Corporation.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
+ *
+ * lustre/obdclass/obd_mount.c
+ *
+ * Client mount routines
+ *
+ * Author: Nathan Rutman <nathan@clusterfs.com>
+ */
+
+
+#define DEBUG_SUBSYSTEM S_CLASS
+#define D_MOUNT (D_SUPER|D_CONFIG/*|D_WARNING */)
+#define PRINT_CMD CDEBUG
+
+#include "../include/obd.h"
+#include "../include/linux/lustre_compat25.h"
+#include "../include/obd_class.h"
+#include "../include/lustre/lustre_user.h"
+#include "../include/lustre_log.h"
+#include "../include/lustre_disk.h"
+#include "../include/lustre_param.h"
+
+static int (*client_fill_super)(struct super_block *sb,
+                               struct vfsmount *mnt);
+
+static void (*kill_super_cb)(struct super_block *sb);
+
+/**************** config llog ********************/
+
+/** Get a config log from the MGS and process it.
+ * This func is called for both clients and servers.
+ * Continue to process new statements appended to the logs
+ * (whenever the config lock is revoked) until lustre_end_log
+ * is called.
+ * @param sb The superblock is used by the MGC to write to the local copy of
+ *   the config log
+ * @param logname The name of the llog to replicate from the MGS
+ * @param cfg Since the same mgc may be used to follow multiple config logs
+ *   (e.g. ost1, ost2, client), the config_llog_instance keeps the state for
+ *   this log, and is added to the mgc's list of logs to follow.
+ */
+int lustre_process_log(struct super_block *sb, char *logname,
+                    struct config_llog_instance *cfg)
+{
+       struct lustre_cfg *lcfg;
+       struct lustre_cfg_bufs *bufs;
+       struct lustre_sb_info *lsi = s2lsi(sb);
+       struct obd_device *mgc = lsi->lsi_mgc;
+       int rc;
+
+       LASSERT(mgc);
+       LASSERT(cfg);
+
+       OBD_ALLOC_PTR(bufs);
+       if (bufs == NULL)
+               return -ENOMEM;
+
+       /* mgc_process_config */
+       lustre_cfg_bufs_reset(bufs, mgc->obd_name);
+       lustre_cfg_bufs_set_string(bufs, 1, logname);
+       lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg));
+       lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb));
+       lcfg = lustre_cfg_new(LCFG_LOG_START, bufs);
+       rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
+       lustre_cfg_free(lcfg);
+
+       OBD_FREE_PTR(bufs);
+
+       if (rc == -EINVAL)
+               LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s' failed from the MGS (%d).  Make sure this client and the MGS are running compatible versions of Lustre.\n",
+                                  mgc->obd_name, logname, rc);
+
+       if (rc)
+               LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' failed (%d). This may be the result of communication errors between this node and the MGS, a bad configuration, or other errors. See the syslog for more information.\n",
+                                  mgc->obd_name, logname,
+                                  rc);
+
+       /* class_obd_list(); */
+       return rc;
+}
+EXPORT_SYMBOL(lustre_process_log);
+
+/* Stop watching this config log for updates */
+int lustre_end_log(struct super_block *sb, char *logname,
+                      struct config_llog_instance *cfg)
+{
+       struct lustre_cfg *lcfg;
+       struct lustre_cfg_bufs bufs;
+       struct lustre_sb_info *lsi = s2lsi(sb);
+       struct obd_device *mgc = lsi->lsi_mgc;
+       int rc;
+
+       if (!mgc)
+               return -ENOENT;
+
+       /* mgc_process_config */
+       lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
+       lustre_cfg_bufs_set_string(&bufs, 1, logname);
+       if (cfg)
+               lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
+       lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
+       rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
+       lustre_cfg_free(lcfg);
+       return rc;
+}
+EXPORT_SYMBOL(lustre_end_log);
+
+/**************** obd start *******************/
+
+/** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from
+ * lctl (and do for echo cli/srv.
+ */
+int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
+           char *s1, char *s2, char *s3, char *s4)
+{
+       struct lustre_cfg_bufs bufs;
+       struct lustre_cfg    * lcfg = NULL;
+       int rc;
+
+       CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
+              cmd, s1, s2, s3, s4);
+
+       lustre_cfg_bufs_reset(&bufs, cfgname);
+       if (s1)
+               lustre_cfg_bufs_set_string(&bufs, 1, s1);
+       if (s2)
+               lustre_cfg_bufs_set_string(&bufs, 2, s2);
+       if (s3)
+               lustre_cfg_bufs_set_string(&bufs, 3, s3);
+       if (s4)
+               lustre_cfg_bufs_set_string(&bufs, 4, s4);
+
+       lcfg = lustre_cfg_new(cmd, &bufs);
+       lcfg->lcfg_nid = nid;
+       rc = class_process_config(lcfg);
+       lustre_cfg_free(lcfg);
+       return rc;
+}
+EXPORT_SYMBOL(do_lcfg);
+
+/** Call class_attach and class_setup.  These methods in turn call
+ * obd type-specific methods.
+ */
+int lustre_start_simple(char *obdname, char *type, char *uuid,
+                       char *s1, char *s2, char *s3, char *s4)
+{
+       int rc;
+       CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
+
+       rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, NULL, NULL);
+       if (rc) {
+               CERROR("%s attach error %d\n", obdname, rc);
+               return rc;
+       }
+       rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, s3, s4);
+       if (rc) {
+               CERROR("%s setup error %d\n", obdname, rc);
+               do_lcfg(obdname, 0, LCFG_DETACH, NULL, NULL, NULL, NULL);
+       }
+       return rc;
+}
+
+DEFINE_MUTEX(mgc_start_lock);
+
+/** Set up a mgc obd to process startup logs
+ *
+ * \param sb [in] super block of the mgc obd
+ *
+ * \retval 0 success, otherwise error code
+ */
+int lustre_start_mgc(struct super_block *sb)
+{
+       struct obd_connect_data *data = NULL;
+       struct lustre_sb_info *lsi = s2lsi(sb);
+       struct obd_device *obd;
+       struct obd_export *exp;
+       struct obd_uuid *uuid;
+       class_uuid_t uuidc;
+       lnet_nid_t nid;
+       char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
+       char *ptr;
+       int rc = 0, i = 0, j, len;
+
+       LASSERT(lsi->lsi_lmd);
+
+       /* Find the first non-lo MGS nid for our MGC name */
+       if (IS_SERVER(lsi)) {
+               /* mount -o mgsnode=nid */
+               ptr = lsi->lsi_lmd->lmd_mgs;
+               if (lsi->lsi_lmd->lmd_mgs &&
+                   (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) {
+                       i++;
+               } else if (IS_MGS(lsi)) {
+                       lnet_process_id_t id;
+                       while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
+                               if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
+                                       continue;
+                               nid = id.nid;
+                               i++;
+                               break;
+                       }
+               }
+       } else { /* client */
+               /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
+               ptr = lsi->lsi_lmd->lmd_dev;
+               if (class_parse_nid(ptr, &nid, &ptr) == 0)
+                       i++;
+       }
+       if (i == 0) {
+               CERROR("No valid MGS nids found.\n");
+               return -EINVAL;
+       }
+
+       mutex_lock(&mgc_start_lock);
+
+       len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
+       OBD_ALLOC(mgcname, len);
+       OBD_ALLOC(niduuid, len + 2);
+       if (!mgcname || !niduuid) {
+               rc = -ENOMEM;
+               goto out_free;
+       }
+       sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
+
+       mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
+
+       OBD_ALLOC_PTR(data);
+       if (data == NULL) {
+               rc = -ENOMEM;
+               goto out_free;
+       }
+
+       obd = class_name2obd(mgcname);
+       if (obd && !obd->obd_stopping) {
+               int recov_bk;
+
+               rc = obd_set_info_async(NULL, obd->obd_self_export,
+                                       strlen(KEY_MGSSEC), KEY_MGSSEC,
+                                       strlen(mgssec), mgssec, NULL);
+               if (rc)
+                       goto out_free;
+
+               /* Re-using an existing MGC */
+               atomic_inc(&obd->u.cli.cl_mgc_refcount);
+
+               /* IR compatibility check, only for clients */
+               if (lmd_is_client(lsi->lsi_lmd)) {
+                       int has_ir;
+                       int vallen = sizeof(*data);
+                       __u32 *flags = &lsi->lsi_lmd->lmd_flags;
+
+                       rc = obd_get_info(NULL, obd->obd_self_export,
+                                         strlen(KEY_CONN_DATA), KEY_CONN_DATA,
+                                         &vallen, data, NULL);
+                       LASSERT(rc == 0);
+                       has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
+                       if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
+                               /* LMD_FLG_NOIR is for test purpose only */
+                               LCONSOLE_WARN(
+                                       "Trying to mount a client with IR setting not compatible with current mgc. Force to use current mgc setting that is IR %s.\n",
+                                       has_ir ? "enabled" : "disabled");
+                               if (has_ir)
+                                       *flags &= ~LMD_FLG_NOIR;
+                               else
+                                       *flags |= LMD_FLG_NOIR;
+                       }
+               }
+
+               recov_bk = 0;
+               /* If we are restarting the MGS, don't try to keep the MGC's
+                  old connection, or registration will fail. */
+               if (IS_MGS(lsi)) {
+                       CDEBUG(D_MOUNT, "New MGS with live MGC\n");
+                       recov_bk = 1;
+               }
+
+               /* Try all connections, but only once (again).
+                  We don't want to block another target from starting
+                  (using its local copy of the log), but we do want to connect
+                  if at all possible. */
+               recov_bk++;
+               CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,
+                      recov_bk);
+               rc = obd_set_info_async(NULL, obd->obd_self_export,
+                                       sizeof(KEY_INIT_RECOV_BACKUP),
+                                       KEY_INIT_RECOV_BACKUP,
+                                       sizeof(recov_bk), &recov_bk, NULL);
+               rc = 0;
+               goto out;
+       }
+
+       CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
+
+       /* Add the primary nids for the MGS */
+       i = 0;
+       sprintf(niduuid, "%s_%x", mgcname, i);
+       if (IS_SERVER(lsi)) {
+               ptr = lsi->lsi_lmd->lmd_mgs;
+               if (IS_MGS(lsi)) {
+                       /* Use local nids (including LO) */
+                       lnet_process_id_t id;
+                       while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
+                               rc = do_lcfg(mgcname, id.nid,
+                                            LCFG_ADD_UUID, niduuid,
+                                            NULL, NULL, NULL);
+                       }
+               } else {
+                       /* Use mgsnode= nids */
+                       /* mount -o mgsnode=nid */
+                       if (lsi->lsi_lmd->lmd_mgs) {
+                               ptr = lsi->lsi_lmd->lmd_mgs;
+                       } else if (class_find_param(ptr, PARAM_MGSNODE,
+                                                   &ptr) != 0) {
+                               CERROR("No MGS nids given.\n");
+                               rc = -EINVAL;
+                               goto out_free;
+                       }
+                       while (class_parse_nid(ptr, &nid, &ptr) == 0) {
+                               rc = do_lcfg(mgcname, nid,
+                                            LCFG_ADD_UUID, niduuid,
+                                            NULL, NULL, NULL);
+                               i++;
+                       }
+               }
+       } else { /* client */
+               /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
+               ptr = lsi->lsi_lmd->lmd_dev;
+               while (class_parse_nid(ptr, &nid, &ptr) == 0) {
+                       rc = do_lcfg(mgcname, nid,
+                                    LCFG_ADD_UUID, niduuid, NULL, NULL, NULL);
+                       i++;
+                       /* Stop at the first failover nid */
+                       if (*ptr == ':')
+                               break;
+               }
+       }
+       if (i == 0) {
+               CERROR("No valid MGS nids found.\n");
+               rc = -EINVAL;
+               goto out_free;
+       }
+       lsi->lsi_lmd->lmd_mgs_failnodes = 1;
+
+       /* Random uuid for MGC allows easier reconnects */
+       OBD_ALLOC_PTR(uuid);
+       if (!uuid) {
+               rc = -ENOMEM;
+               goto out_free;
+       }
+
+       ll_generate_random_uuid(uuidc);
+       class_uuid_unparse(uuidc, uuid);
+
+       /* Start the MGC */
+       rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
+                                (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
+                                niduuid, NULL, NULL);
+       OBD_FREE_PTR(uuid);
+       if (rc)
+               goto out_free;
+
+       /* Add any failover MGS nids */
+       i = 1;
+       while (ptr && ((*ptr == ':' ||
+              class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) {
+               /* New failover node */
+               sprintf(niduuid, "%s_%x", mgcname, i);
+               j = 0;
+               while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
+                       j++;
+                       rc = do_lcfg(mgcname, nid,
+                                    LCFG_ADD_UUID, niduuid, NULL, NULL, NULL);
+                       if (*ptr == ':')
+                               break;
+               }
+               if (j > 0) {
+                       rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
+                                    niduuid, NULL, NULL, NULL);
+                       i++;
+               } else {
+                       /* at ":/fsname" */
+                       break;
+               }
+       }
+       lsi->lsi_lmd->lmd_mgs_failnodes = i;
+
+       obd = class_name2obd(mgcname);
+       if (!obd) {
+               CERROR("Can't find mgcobd %s\n", mgcname);
+               rc = -ENOTCONN;
+               goto out_free;
+       }
+
+       rc = obd_set_info_async(NULL, obd->obd_self_export,
+                               strlen(KEY_MGSSEC), KEY_MGSSEC,
+                               strlen(mgssec), mgssec, NULL);
+       if (rc)
+               goto out_free;
+
+       /* Keep a refcount of servers/clients who started with "mount",
+          so we know when we can get rid of the mgc. */
+       atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
+
+       /* We connect to the MGS at setup, and don't disconnect until cleanup */
+       data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
+                                 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV |
+                                 OBD_CONNECT_LVB_TYPE;
+
+#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
+       data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB;
+#else
+#warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
+#endif
+
+       if (lmd_is_client(lsi->lsi_lmd) &&
+           lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
+               data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
+       data->ocd_version = LUSTRE_VERSION_CODE;
+       rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
+       if (rc) {
+               CERROR("connect failed %d\n", rc);
+               goto out;
+       }
+
+       obd->u.cli.cl_mgc_mgsexp = exp;
+
+out:
+       /* Keep the mgc info in the sb. Note that many lsi's can point
+          to the same mgc.*/
+       lsi->lsi_mgc = obd;
+out_free:
+       mutex_unlock(&mgc_start_lock);
+
+       if (data)
+               OBD_FREE_PTR(data);
+       if (mgcname)
+               OBD_FREE(mgcname, len);
+       if (niduuid)
+               OBD_FREE(niduuid, len + 2);
+       return rc;
+}
+
+static int lustre_stop_mgc(struct super_block *sb)
+{
+       struct lustre_sb_info *lsi = s2lsi(sb);
+       struct obd_device *obd;
+       char *niduuid = NULL, *ptr = NULL;
+       int i, rc = 0, len = 0;
+
+       if (!lsi)
+               return -ENOENT;
+       obd = lsi->lsi_mgc;
+       if (!obd)
+               return -ENOENT;
+       lsi->lsi_mgc = NULL;
+
+       mutex_lock(&mgc_start_lock);
+       LASSERT(atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
+       if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
+               /* This is not fatal, every client that stops
+                  will call in here. */
+               CDEBUG(D_MOUNT, "mgc still has %d references.\n",
+                      atomic_read(&obd->u.cli.cl_mgc_refcount));
+               rc = -EBUSY;
+               goto out;
+       }
+
+       /* The MGC has no recoverable data in any case.
+        * force shutdown set in umount_begin */
+       obd->obd_no_recov = 1;
+
+       if (obd->u.cli.cl_mgc_mgsexp) {
+               /* An error is not fatal, if we are unable to send the
+                  disconnect mgs ping evictor cleans up the export */
+               rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
+               if (rc)
+                       CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
+       }
+
+       /* Save the obdname for cleaning the nid uuids, which are
+          obdname_XX */
+       len = strlen(obd->obd_name) + 6;
+       OBD_ALLOC(niduuid, len);
+       if (niduuid) {
+               strcpy(niduuid, obd->obd_name);
+               ptr = niduuid + strlen(niduuid);
+       }
+
+       rc = class_manual_cleanup(obd);
+       if (rc)
+               goto out;
+
+       /* Clean the nid uuids */
+       if (!niduuid) {
+               rc = -ENOMEM;
+               goto out;
+       }
+
+       for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
+               sprintf(ptr, "_%x", i);
+               rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
+                            niduuid, NULL, NULL, NULL);
+               if (rc)
+                       CERROR("del MDC UUID %s failed: rc = %d\n",
+                              niduuid, rc);
+       }
+out:
+       if (niduuid)
+               OBD_FREE(niduuid, len);
+
+       /* class_import_put will get rid of the additional connections */
+       mutex_unlock(&mgc_start_lock);
+       return rc;
+}
+
+/***************** lustre superblock **************/
+
+struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
+{
+       struct lustre_sb_info *lsi;
+
+       OBD_ALLOC_PTR(lsi);
+       if (!lsi)
+               return NULL;
+       OBD_ALLOC_PTR(lsi->lsi_lmd);
+       if (!lsi->lsi_lmd) {
+               OBD_FREE_PTR(lsi);
+               return NULL;
+       }
+
+       lsi->lsi_lmd->lmd_exclude_count = 0;
+       lsi->lsi_lmd->lmd_recovery_time_soft = 0;
+       lsi->lsi_lmd->lmd_recovery_time_hard = 0;
+       s2lsi_nocast(sb) = lsi;
+       /* we take 1 extra ref for our setup */
+       atomic_set(&lsi->lsi_mounts, 1);
+
+       /* Default umount style */
+       lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
+
+       return lsi;
+}
+
+static int lustre_free_lsi(struct super_block *sb)
+{
+       struct lustre_sb_info *lsi = s2lsi(sb);
+
+       LASSERT(lsi != NULL);
+       CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
+
+       /* someone didn't call server_put_mount. */
+       LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
+
+       if (lsi->lsi_lmd != NULL) {
+               if (lsi->lsi_lmd->lmd_dev != NULL)
+                       OBD_FREE(lsi->lsi_lmd->lmd_dev,
+                                strlen(lsi->lsi_lmd->lmd_dev) + 1);
+               if (lsi->lsi_lmd->lmd_profile != NULL)
+                       OBD_FREE(lsi->lsi_lmd->lmd_profile,
+                                strlen(lsi->lsi_lmd->lmd_profile) + 1);
+               if (lsi->lsi_lmd->lmd_mgssec != NULL)
+                       OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
+                                strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
+               if (lsi->lsi_lmd->lmd_opts != NULL)
+                       OBD_FREE(lsi->lsi_lmd->lmd_opts,
+                                strlen(lsi->lsi_lmd->lmd_opts) + 1);
+               if (lsi->lsi_lmd->lmd_exclude_count)
+                       OBD_FREE(lsi->lsi_lmd->lmd_exclude,
+                                sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
+                                lsi->lsi_lmd->lmd_exclude_count);
+               if (lsi->lsi_lmd->lmd_mgs != NULL)
+                       OBD_FREE(lsi->lsi_lmd->lmd_mgs,
+                                strlen(lsi->lsi_lmd->lmd_mgs) + 1);
+               if (lsi->lsi_lmd->lmd_osd_type != NULL)
+                       OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
+                                strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
+               if (lsi->lsi_lmd->lmd_params != NULL)
+                       OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
+
+               OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
+       }
+
+       LASSERT(lsi->lsi_llsbi == NULL);
+       OBD_FREE(lsi, sizeof(*lsi));
+       s2lsi_nocast(sb) = NULL;
+
+       return 0;
+}
+
+/* The lsi has one reference for every server that is using the disk -
+   e.g. MDT, MGS, and potentially MGC */
+int lustre_put_lsi(struct super_block *sb)
+{
+       struct lustre_sb_info *lsi = s2lsi(sb);
+
+       LASSERT(lsi != NULL);
+
+       CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
+       if (atomic_dec_and_test(&lsi->lsi_mounts)) {
+               if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
+                       lu_device_put(&lsi->lsi_dt_dev->dd_lu_dev);
+                       lsi->lsi_osd_exp->exp_obd->obd_lvfs_ctxt.dt = NULL;
+                       lsi->lsi_dt_dev = NULL;
+                       obd_disconnect(lsi->lsi_osd_exp);
+                       /* wait till OSD is gone */
+                       obd_zombie_barrier();
+               }
+               lustre_free_lsi(sb);
+               return 1;
+       }
+       return 0;
+}
+
+/*** SERVER NAME ***
+ * <FSNAME><SEPARATOR><TYPE><INDEX>
+ * FSNAME is between 1 and 8 characters (inclusive).
+ *     Excluded characters are '/' and ':'
+ * SEPARATOR is either ':' or '-'
+ * TYPE: "OST", "MDT", etc.
+ * INDEX: Hex representation of the index
+ */
+
+/** Get the fsname ("lustre") from the server name ("lustre-OST003F").
+ * @param [in] svname server name including type and index
+ * @param [out] fsname Buffer to copy filesystem name prefix into.
+ *  Must have at least 'strlen(fsname) + 1' chars.
+ * @param [out] endptr if endptr isn't NULL it is set to end of fsname
+ * rc < 0  on error
+ */
+int server_name2fsname(const char *svname, char *fsname, const char **endptr)
+{
+       const char *dash;
+
+       dash = svname + strnlen(svname, 8); /* max fsname length is 8 */
+       for (; dash > svname && *dash != '-' && *dash != ':'; dash--)
+               ;
+       if (dash == svname)
+               return -EINVAL;
+
+       if (fsname != NULL) {
+               strncpy(fsname, svname, dash - svname);
+               fsname[dash - svname] = '\0';
+       }
+
+       if (endptr != NULL)
+               *endptr = dash;
+
+       return 0;
+}
+EXPORT_SYMBOL(server_name2fsname);
+
+/**
+ * Get service name (svname) from string
+ * rc < 0 on error
+ * if endptr isn't NULL it is set to end of fsname *
+ */
+int server_name2svname(const char *label, char *svname, const char **endptr,
+                      size_t svsize)
+{
+       int rc;
+       const char *dash;
+
+       /* We use server_name2fsname() just for parsing */
+       rc = server_name2fsname(label, NULL, &dash);
+       if (rc != 0)
+               return rc;
+
+       if (endptr != NULL)
+               *endptr = dash;
+
+       if (strlcpy(svname, dash + 1, svsize) >= svsize)
+               return -E2BIG;
+
+       return 0;
+}
+EXPORT_SYMBOL(server_name2svname);
+
+
+/* Get the index from the obd name.
+   rc = server type, or
+   rc < 0  on error
+   if endptr isn't NULL it is set to end of name */
+int server_name2index(const char *svname, __u32 *idx, const char **endptr)
+{
+       unsigned long index;
+       int rc;
+       const char *dash;
+
+       /* We use server_name2fsname() just for parsing */
+       rc = server_name2fsname(svname, NULL, &dash);
+       if (rc != 0)
+               return rc;
+
+       dash++;
+
+       if (strncmp(dash, "MDT", 3) == 0)
+               rc = LDD_F_SV_TYPE_MDT;
+       else if (strncmp(dash, "OST", 3) == 0)
+               rc = LDD_F_SV_TYPE_OST;
+       else
+               return -EINVAL;
+
+       dash += 3;
+
+       if (strncmp(dash, "all", 3) == 0) {
+               if (endptr != NULL)
+                       *endptr = dash + 3;
+               return rc | LDD_F_SV_ALL;
+       }
+
+       index = simple_strtoul(dash, (char **)endptr, 16);
+       if (idx != NULL)
+               *idx = index;
+
+       /* Account for -mdc after index that is possible when specifying mdt */
+       if (endptr != NULL && strncmp(LUSTRE_MDC_NAME, *endptr + 1,
+                                     sizeof(LUSTRE_MDC_NAME)-1) == 0)
+               *endptr += sizeof(LUSTRE_MDC_NAME);
+
+       return rc;
+}
+EXPORT_SYMBOL(server_name2index);
+
+/*************** mount common between server and client ***************/
+
+/* Common umount */
+int lustre_common_put_super(struct super_block *sb)
+{
+       int rc;
+
+       CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
+
+       /* Drop a ref to the MGC */
+       rc = lustre_stop_mgc(sb);
+       if (rc && (rc != -ENOENT)) {
+               if (rc != -EBUSY) {
+                       CERROR("Can't stop MGC: %d\n", rc);
+                       return rc;
+               }
+               /* BUSY just means that there's some other obd that
+                  needs the mgc.  Let him clean it up. */
+               CDEBUG(D_MOUNT, "MGC still in use\n");
+       }
+       /* Drop a ref to the mounted disk */
+       lustre_put_lsi(sb);
+       lu_types_stop();
+       return rc;
+}
+EXPORT_SYMBOL(lustre_common_put_super);
+
+static void lmd_print(struct lustre_mount_data *lmd)
+{
+       int i;
+
+       PRINT_CMD(D_MOUNT, "  mount data:\n");
+       if (lmd_is_client(lmd))
+               PRINT_CMD(D_MOUNT, "profile: %s\n", lmd->lmd_profile);
+       PRINT_CMD(D_MOUNT, "device:  %s\n", lmd->lmd_dev);
+       PRINT_CMD(D_MOUNT, "flags:   %x\n", lmd->lmd_flags);
+
+       if (lmd->lmd_opts)
+               PRINT_CMD(D_MOUNT, "options: %s\n", lmd->lmd_opts);
+
+       if (lmd->lmd_recovery_time_soft)
+               PRINT_CMD(D_MOUNT, "recovery time soft: %d\n",
+                         lmd->lmd_recovery_time_soft);
+
+       if (lmd->lmd_recovery_time_hard)
+               PRINT_CMD(D_MOUNT, "recovery time hard: %d\n",
+                         lmd->lmd_recovery_time_hard);
+
+       for (i = 0; i < lmd->lmd_exclude_count; i++) {
+               PRINT_CMD(D_MOUNT, "exclude %d:  OST%04x\n", i,
+                         lmd->lmd_exclude[i]);
+       }
+}
+
+/* Is this server on the exclusion list */
+int lustre_check_exclusion(struct super_block *sb, char *svname)
+{
+       struct lustre_sb_info *lsi = s2lsi(sb);
+       struct lustre_mount_data *lmd = lsi->lsi_lmd;
+       __u32 index;
+       int i, rc;
+
+       rc = server_name2index(svname, &index, NULL);
+       if (rc != LDD_F_SV_TYPE_OST)
+               /* Only exclude OSTs */
+               return 0;
+
+       CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
+              index, lmd->lmd_exclude_count, lmd->lmd_dev);
+
+       for (i = 0; i < lmd->lmd_exclude_count; i++) {
+               if (index == lmd->lmd_exclude[i]) {
+                       CWARN("Excluding %s (on exclusion list)\n", svname);
+                       return 1;
+               }
+       }
+       return 0;
+}
+
+/* mount -v  -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
+static int lmd_make_exclusion(struct lustre_mount_data *lmd, const char *ptr)
+{
+       const char *s1 = ptr, *s2;
+       __u32 index, *exclude_list;
+       int rc = 0, devmax;
+
+       /* The shortest an ost name can be is 8 chars: -OST0000.
+          We don't actually know the fsname at this time, so in fact
+          a user could specify any fsname. */
+       devmax = strlen(ptr) / 8 + 1;
+
+       /* temp storage until we figure out how many we have */
+       OBD_ALLOC(exclude_list, sizeof(index) * devmax);
+       if (!exclude_list)
+               return -ENOMEM;
+
+       /* we enter this fn pointing at the '=' */
+       while (*s1 && *s1 != ' ' && *s1 != ',') {
+               s1++;
+               rc = server_name2index(s1, &index, &s2);
+               if (rc < 0) {
+                       CERROR("Can't parse server name '%s': rc = %d\n",
+                              s1, rc);
+                       break;
+               }
+               if (rc == LDD_F_SV_TYPE_OST)
+                       exclude_list[lmd->lmd_exclude_count++] = index;
+               else
+                       CDEBUG(D_MOUNT, "ignoring exclude %.*s: type = %#x\n",
+                              (uint)(s2-s1), s1, rc);
+               s1 = s2;
+               /* now we are pointing at ':' (next exclude)
+                  or ',' (end of excludes) */
+               if (lmd->lmd_exclude_count >= devmax)
+                       break;
+       }
+       if (rc >= 0) /* non-err */
+               rc = 0;
+
+       if (lmd->lmd_exclude_count) {
+               /* permanent, freed in lustre_free_lsi */
+               OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
+                         lmd->lmd_exclude_count);
+               if (lmd->lmd_exclude) {
+                       memcpy(lmd->lmd_exclude, exclude_list,
+                              sizeof(index) * lmd->lmd_exclude_count);
+               } else {
+                       rc = -ENOMEM;
+                       lmd->lmd_exclude_count = 0;
+               }
+       }
+       OBD_FREE(exclude_list, sizeof(index) * devmax);
+       return rc;
+}
+
+static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
+{
+       char   *tail;
+       int     length;
+
+       if (lmd->lmd_mgssec != NULL) {
+               OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
+               lmd->lmd_mgssec = NULL;
+       }
+
+       tail = strchr(ptr, ',');
+       if (tail == NULL)
+               length = strlen(ptr);
+       else
+               length = tail - ptr;
+
+       OBD_ALLOC(lmd->lmd_mgssec, length + 1);
+       if (lmd->lmd_mgssec == NULL)
+               return -ENOMEM;
+
+       memcpy(lmd->lmd_mgssec, ptr, length);
+       lmd->lmd_mgssec[length] = '\0';
+       return 0;
+}
+
+static int lmd_parse_string(char **handle, char *ptr)
+{
+       char   *tail;
+       int     length;
+
+       if ((handle == NULL) || (ptr == NULL))
+               return -EINVAL;
+
+       if (*handle != NULL) {
+               OBD_FREE(*handle, strlen(*handle) + 1);
+               *handle = NULL;
+       }
+
+       tail = strchr(ptr, ',');
+       if (tail == NULL)
+               length = strlen(ptr);
+       else
+               length = tail - ptr;
+
+       OBD_ALLOC(*handle, length + 1);
+       if (*handle == NULL)
+               return -ENOMEM;
+
+       memcpy(*handle, ptr, length);
+       (*handle)[length] = '\0';
+
+       return 0;
+}
+
+/* Collect multiple values for mgsnid specifiers */
+static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
+{
+       lnet_nid_t nid;
+       char *tail = *ptr;
+       char *mgsnid;
+       int   length;
+       int   oldlen = 0;
+
+       /* Find end of nidlist */
+       while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
+       length = tail - *ptr;
+       if (length == 0) {
+               LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
+               return -EINVAL;
+       }
+
+       if (lmd->lmd_mgs != NULL)
+               oldlen = strlen(lmd->lmd_mgs) + 1;
+
+       OBD_ALLOC(mgsnid, oldlen + length + 1);
+       if (mgsnid == NULL)
+               return -ENOMEM;
+
+       if (lmd->lmd_mgs != NULL) {
+               /* Multiple mgsnid= are taken to mean failover locations */
+               memcpy(mgsnid, lmd->lmd_mgs, oldlen);
+               mgsnid[oldlen - 1] = ':';
+               OBD_FREE(lmd->lmd_mgs, oldlen);
+       }
+       memcpy(mgsnid + oldlen, *ptr, length);
+       mgsnid[oldlen + length] = '\0';
+       lmd->lmd_mgs = mgsnid;
+       *ptr = tail;
+
+       return 0;
+}
+
+/** Parse mount line options
+ * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
+ * dev is passed as device=uml1:/lustre by mount.lustre
+ */
+static int lmd_parse(char *options, struct lustre_mount_data *lmd)
+{
+       char *s1, *s2, *devname = NULL;
+       struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
+       int rc = 0;
+
+       LASSERT(lmd);
+       if (!options) {
+               LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that /sbin/mount.lustre is installed.\n");
+               return -EINVAL;
+       }
+
+       /* Options should be a string - try to detect old lmd data */
+       if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
+               LCONSOLE_ERROR_MSG(0x163, "You're using an old version of /sbin/mount.lustre.  Please install version %s\n",
+                                  LUSTRE_VERSION_STRING);
+               return -EINVAL;
+       }
+       lmd->lmd_magic = LMD_MAGIC;
+
+       OBD_ALLOC(lmd->lmd_params, 4096);
+       if (lmd->lmd_params == NULL)
+               return -ENOMEM;
+       lmd->lmd_params[0] = '\0';
+
+       /* Set default flags here */
+
+       s1 = options;
+       while (*s1) {
+               int clear = 0;
+               int time_min = OBD_RECOVERY_TIME_MIN;
+
+               /* Skip whitespace and extra commas */
+               while (*s1 == ' ' || *s1 == ',')
+                       s1++;
+
+               /* Client options are parsed in ll_options: eg. flock,
+                  user_xattr, acl */
+
+               /* Parse non-ldiskfs options here. Rather than modifying
+                  ldiskfs, we just zero these out here */
+               if (strncmp(s1, "abort_recov", 11) == 0) {
+                       lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
+                       clear++;
+               } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
+                       lmd->lmd_recovery_time_soft = max_t(int,
+                               simple_strtoul(s1 + 19, NULL, 10), time_min);
+                       clear++;
+               } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
+                       lmd->lmd_recovery_time_hard = max_t(int,
+                               simple_strtoul(s1 + 19, NULL, 10), time_min);
+                       clear++;
+               } else if (strncmp(s1, "noir", 4) == 0) {
+                       lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
+                       clear++;
+               } else if (strncmp(s1, "nosvc", 5) == 0) {
+                       lmd->lmd_flags |= LMD_FLG_NOSVC;
+                       clear++;
+               } else if (strncmp(s1, "nomgs", 5) == 0) {
+                       lmd->lmd_flags |= LMD_FLG_NOMGS;
+                       clear++;
+               } else if (strncmp(s1, "noscrub", 7) == 0) {
+                       lmd->lmd_flags |= LMD_FLG_NOSCRUB;
+                       clear++;
+               } else if (strncmp(s1, PARAM_MGSNODE,
+                                  sizeof(PARAM_MGSNODE) - 1) == 0) {
+                       s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
+                       /* Assume the next mount opt is the first
+                          invalid nid we get to. */
+                       rc = lmd_parse_mgs(lmd, &s2);
+                       if (rc)
+                               goto invalid;
+                       clear++;
+               } else if (strncmp(s1, "writeconf", 9) == 0) {
+                       lmd->lmd_flags |= LMD_FLG_WRITECONF;
+                       clear++;
+               } else if (strncmp(s1, "update", 6) == 0) {
+                       lmd->lmd_flags |= LMD_FLG_UPDATE;
+                       clear++;
+               } else if (strncmp(s1, "virgin", 6) == 0) {
+                       lmd->lmd_flags |= LMD_FLG_VIRGIN;
+                       clear++;
+               } else if (strncmp(s1, "noprimnode", 10) == 0) {
+                       lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
+                       clear++;
+               } else if (strncmp(s1, "mgssec=", 7) == 0) {
+                       rc = lmd_parse_mgssec(lmd, s1 + 7);
+                       if (rc)
+                               goto invalid;
+                       clear++;
+               /* ost exclusion list */
+               } else if (strncmp(s1, "exclude=", 8) == 0) {
+                       rc = lmd_make_exclusion(lmd, s1 + 7);
+                       if (rc)
+                               goto invalid;
+                       clear++;
+               } else if (strncmp(s1, "mgs", 3) == 0) {
+                       /* We are an MGS */
+                       lmd->lmd_flags |= LMD_FLG_MGS;
+                       clear++;
+               } else if (strncmp(s1, "svname=", 7) == 0) {
+                       rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
+                       if (rc)
+                               goto invalid;
+                       clear++;
+               } else if (strncmp(s1, "param=", 6) == 0) {
+                       int length;
+                       char *tail = strchr(s1 + 6, ',');
+                       if (tail == NULL)
+                               length = strlen(s1);
+                       else
+                               length = tail - s1;
+                       length -= 6;
+                       strncat(lmd->lmd_params, s1 + 6, length);
+                       strcat(lmd->lmd_params, " ");
+                       clear++;
+               } else if (strncmp(s1, "osd=", 4) == 0) {
+                       rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
+                       if (rc)
+                               goto invalid;
+                       clear++;
+               }
+               /* Linux 2.4 doesn't pass the device, so we stuck it at the
+                  end of the options. */
+               else if (strncmp(s1, "device=", 7) == 0) {
+                       devname = s1 + 7;
+                       /* terminate options right before device.  device
+                          must be the last one. */
+                       *s1 = '\0';
+                       break;
+               }
+
+               /* Find next opt */
+               s2 = strchr(s1, ',');
+               if (s2 == NULL) {
+                       if (clear)
+                               *s1 = '\0';
+                       break;
+               }
+               s2++;
+               if (clear)
+                       memmove(s1, s2, strlen(s2) + 1);
+               else
+                       s1 = s2;
+       }
+
+       if (!devname) {
+               LCONSOLE_ERROR_MSG(0x164, "Can't find the device name (need mount option 'device=...')\n");
+               goto invalid;
+       }
+
+       s1 = strstr(devname, ":/");
+       if (s1) {
+               ++s1;
+               lmd->lmd_flags |= LMD_FLG_CLIENT;
+               /* Remove leading /s from fsname */
+               while (*++s1 == '/') ;
+               /* Freed in lustre_free_lsi */
+               OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
+               if (!lmd->lmd_profile)
+                       return -ENOMEM;
+               sprintf(lmd->lmd_profile, "%s-client", s1);
+       }
+
+       /* Freed in lustre_free_lsi */
+       OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
+       if (!lmd->lmd_dev)
+               return -ENOMEM;
+       strcpy(lmd->lmd_dev, devname);
+
+       /* Save mount options */
+       s1 = options + strlen(options) - 1;
+       while (s1 >= options && (*s1 == ',' || *s1 == ' '))
+               *s1-- = 0;
+       if (*options != 0) {
+               /* Freed in lustre_free_lsi */
+               OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
+               if (!lmd->lmd_opts)
+                       return -ENOMEM;
+               strcpy(lmd->lmd_opts, options);
+       }
+
+       lmd_print(lmd);
+       lmd->lmd_magic = LMD_MAGIC;
+
+       return rc;
+
+invalid:
+       CERROR("Bad mount options %s\n", options);
+       return -EINVAL;
+}
+
+struct lustre_mount_data2 {
+       void *lmd2_data;
+       struct vfsmount *lmd2_mnt;
+};
+
+/** This is the entry point for the mount call into Lustre.
+ * This is called when a server or client is mounted,
+ * and this is where we start setting things up.
+ * @param data Mount options (e.g. -o flock,abort_recov)
+ */
+int lustre_fill_super(struct super_block *sb, void *data, int silent)
+{
+       struct lustre_mount_data *lmd;
+       struct lustre_mount_data2 *lmd2 = data;
+       struct lustre_sb_info *lsi;
+       int rc;
+
+       CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
+
+       lsi = lustre_init_lsi(sb);
+       if (!lsi)
+               return -ENOMEM;
+       lmd = lsi->lsi_lmd;
+
+       /*
+        * Disable lockdep during mount, because mount locking patterns are
+        * `special'.
+        */
+       lockdep_off();
+
+       /*
+        * LU-639: the obd cleanup of last mount may not finish yet, wait here.
+        */
+       obd_zombie_barrier();
+
+       /* Figure out the lmd from the mount options */
+       if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
+               lustre_put_lsi(sb);
+               rc = -EINVAL;
+               goto out;
+       }
+
+       if (lmd_is_client(lmd)) {
+               CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
+               if (client_fill_super == NULL)
+                       request_module("lustre");
+               if (client_fill_super == NULL) {
+                       LCONSOLE_ERROR_MSG(0x165, "Nothing registered for client mount! Is the 'lustre' module loaded?\n");
+                       lustre_put_lsi(sb);
+                       rc = -ENODEV;
+               } else {
+                       rc = lustre_start_mgc(sb);
+                       if (rc) {
+                               lustre_put_lsi(sb);
+                               goto out;
+                       }
+                       /* Connect and start */
+                       /* (should always be ll_fill_super) */
+                       rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
+                       /* c_f_s will call lustre_common_put_super on failure */
+               }
+       } else {
+               CERROR("This is client-side-only module, cannot handle server mount.\n");
+               rc = -EINVAL;
+       }
+
+       /* If error happens in fill_super() call, @lsi will be killed there.
+        * This is why we do not put it here. */
+       goto out;
+out:
+       if (rc) {
+               CERROR("Unable to mount %s (%d)\n",
+                      s2lsi(sb) ? lmd->lmd_dev : "", rc);
+       } else {
+               CDEBUG(D_SUPER, "Mount %s complete\n",
+                      lmd->lmd_dev);
+       }
+       lockdep_on();
+       return rc;
+}
+
+
+/* We can't call ll_fill_super by name because it lives in a module that
+   must be loaded after this one. */
+void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
+                                                 struct vfsmount *mnt))
+{
+       client_fill_super = cfs;
+}
+EXPORT_SYMBOL(lustre_register_client_fill_super);
+
+void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
+{
+       kill_super_cb = cfs;
+}
+EXPORT_SYMBOL(lustre_register_kill_super_cb);
+
+/***************** FS registration ******************/
+struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
+                               const char *devname, void *data)
+{
+       struct lustre_mount_data2 lmd2 = {
+               .lmd2_data = data,
+               .lmd2_mnt = NULL
+       };
+
+       return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
+}
+
+static void lustre_kill_super(struct super_block *sb)
+{
+       struct lustre_sb_info *lsi = s2lsi(sb);
+
+       if (kill_super_cb && lsi && !IS_SERVER(lsi))
+               (*kill_super_cb)(sb);
+
+       kill_anon_super(sb);
+}
+
+/** Register the "lustre" fs type
+ */
+struct file_system_type lustre_fs_type = {
+       .owner  = THIS_MODULE,
+       .name    = "lustre",
+       .mount  = lustre_mount,
+       .kill_sb      = lustre_kill_super,
+       .fs_flags     = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
+                       FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE,
+};
+MODULE_ALIAS_FS("lustre");
+
+int lustre_register_fs(void)
+{
+       return register_filesystem(&lustre_fs_type);
+}
+
+int lustre_unregister_fs(void)
+{
+       return unregister_filesystem(&lustre_fs_type);
+}