Whamcloud - gitweb
LU-9183 llite: handle xattr with the xattr_handler infrastructure
[fs/lustre-release.git] / lustre / obdclass / obd_mount_server.c
index 0cefea2..8a4883a 100644 (file)
  *
  * You should have received a copy of the GNU General Public License
  * version 2 along with this program; If not, see
- * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
- *
- * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
- * CA 95054 USA or visit www.sun.com if you need additional information or
- * have any questions.
+ * http://www.gnu.org/licenses/gpl-2.0.html
  *
  * GPL HEADER END
  */
@@ -27,7 +23,7 @@
  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  *
- * Copyright (c) 2013, Intel Corporation.
+ * Copyright (c) 2013, 2016, Intel Corporation.
  */
 /*
  * This file is part of Lustre, http://www.lustre.org/
 #define PRINT_CMD CDEBUG
 #define PRINT_MASK (D_SUPER | D_CONFIG)
 
-#include <obd.h>
-#include <obd_class.h>
-#include <lustre/lustre_user.h>
+#include <linux/types.h>
+#include <linux/selinux.h>
+#include <linux/statfs.h>
 #include <linux/version.h>
-#include <lustre_ioctl.h>
-#include <lustre_log.h>
-#include <lustre_disk.h>
-#include <lustre_param.h>
 #ifdef HAVE_KERNEL_LOCKED
 #include <linux/smp_lock.h>
 #endif
-#ifdef HAVE_SELINUX_IS_ENABLED
-#include <linux/selinux.h>
-#endif
+
+#include <lustre/lustre_idl.h>
+#include <lustre/lustre_user.h>
+
+#include <llog_swab.h>
+#include <lustre_disk.h>
+#include <uapi/linux/lustre_ioctl.h>
+#include <lustre_log.h>
+#include <uapi/linux/lustre_param.h>
+#include <obd.h>
+#include <obd_class.h>
 
 /*********** mount lookup *********/
 
-DEFINE_MUTEX(lustre_mount_info_lock);
-static CFS_LIST_HEAD(server_mount_info_list);
+static DEFINE_MUTEX(lustre_mount_info_lock);
+static LIST_HEAD(server_mount_info_list);
 
 static struct lustre_mount_info *server_find_mount(const char *name)
 {
-       cfs_list_t *tmp;
+       struct list_head *tmp;
        struct lustre_mount_info *lmi;
        ENTRY;
 
-       cfs_list_for_each(tmp, &server_mount_info_list) {
-               lmi = cfs_list_entry(tmp, struct lustre_mount_info,
-                                    lmi_list_chain);
+       list_for_each(tmp, &server_mount_info_list) {
+               lmi = list_entry(tmp, struct lustre_mount_info,
+                                lmi_list_chain);
                if (strcmp(name, lmi->lmi_name) == 0)
                        RETURN(lmi);
        }
@@ -113,7 +113,7 @@ static int server_register_mount(const char *name, struct super_block *sb)
        }
        lmi->lmi_name = name_cp;
        lmi->lmi_sb = sb;
-       cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list);
+       list_add(&lmi->lmi_list_chain, &server_mount_info_list);
 
        mutex_unlock(&lustre_mount_info_lock);
 
@@ -139,7 +139,7 @@ static int server_deregister_mount(const char *name)
        CDEBUG(D_MOUNT, "deregister mount %p from %s\n", lmi->lmi_sb, name);
 
        OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
-       cfs_list_del(&lmi->lmi_list_chain);
+       list_del(&lmi->lmi_list_chain);
        OBD_FREE(lmi, sizeof(*lmi));
        mutex_unlock(&lustre_mount_info_lock);
 
@@ -239,8 +239,8 @@ static int server_start_mgs(struct super_block *sb)
 
        if (!rc) {
                rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
-                                        LUSTRE_MGS_OBDNAME, 0, 0,
-                                        lsi->lsi_osd_obdname, 0);
+                                        LUSTRE_MGS_OBDNAME, NULL, NULL,
+                                        lsi->lsi_osd_obdname, NULL);
                /* server_deregister_mount() is not called previously, for lsi
                 * and other stuff can't be freed cleanly when mgs calls
                 * server_put_mount() in error handling case (see b=17758),
@@ -262,8 +262,14 @@ static int server_stop_mgs(struct super_block *sb)
 {
        struct obd_device *obd;
        int rc;
+       struct lustre_mount_info *lmi;
        ENTRY;
 
+       /* Do not stop MGS if this device is not the running MGT */
+       lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
+       if (lmi != NULL && lmi->lmi_sb != sb)
+               RETURN(0);
+
        CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
 
        /* There better be only one MGS */
@@ -281,7 +287,8 @@ static int server_stop_mgs(struct super_block *sb)
 
 /* Since there's only one mgc per node, we have to change it's fs to get
    access to the right disk. */
-static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
+static int server_mgc_set_fs(const struct lu_env *env,
+                            struct obd_device *mgc, struct super_block *sb)
 {
        struct lustre_sb_info *lsi = s2lsi(sb);
        int rc;
@@ -290,7 +297,7 @@ static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
        CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
 
        /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
-       rc = obd_set_info_async(NULL, mgc->obd_self_export,
+       rc = obd_set_info_async(env, mgc->obd_self_export,
                                sizeof(KEY_SET_FS), KEY_SET_FS,
                                sizeof(*sb), sb, NULL);
        if (rc != 0)
@@ -299,14 +306,15 @@ static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
        RETURN(rc);
 }
 
-static int server_mgc_clear_fs(struct obd_device *mgc)
+static int server_mgc_clear_fs(const struct lu_env *env,
+                              struct obd_device *mgc)
 {
        int rc;
        ENTRY;
 
        CDEBUG(D_MOUNT, "Unassign mgc disk\n");
 
-       rc = obd_set_info_async(NULL, mgc->obd_self_export,
+       rc = obd_set_info_async(env, mgc->obd_self_export,
                                sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
                                0, NULL, NULL);
        RETURN(rc);
@@ -374,14 +382,26 @@ cleanup:
 }
 EXPORT_SYMBOL(tgt_name2lwp_name);
 
-static CFS_LIST_HEAD(lwp_register_list);
-DEFINE_MUTEX(lwp_register_list_lock);
+static LIST_HEAD(lwp_register_list);
+static DEFINE_SPINLOCK(lwp_register_list_lock);
+
+static void lustre_put_lwp_item(struct lwp_register_item *lri)
+{
+       if (atomic_dec_and_test(&lri->lri_ref)) {
+               LASSERT(list_empty(&lri->lri_list));
+
+               if (*lri->lri_exp != NULL)
+                       class_export_put(*lri->lri_exp);
+               OBD_FREE_PTR(lri);
+       }
+}
 
 int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp,
                             register_lwp_cb cb_func, void *cb_data)
 {
        struct obd_device        *lwp;
        struct lwp_register_item *lri;
+       bool cb = false;
        ENTRY;
 
        LASSERTF(strlen(lwpname) < MTI_NAME_MAXLEN, "lwpname is too long %s\n",
@@ -392,15 +412,12 @@ int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp,
        if (lri == NULL)
                RETURN(-ENOMEM);
 
-       mutex_lock(&lwp_register_list_lock);
-
        lwp = class_name2obd(lwpname);
        if (lwp != NULL && lwp->obd_set_up == 1) {
                struct obd_uuid *uuid;
 
                OBD_ALLOC_PTR(uuid);
                if (uuid == NULL) {
-                       mutex_unlock(&lwp_register_list_lock);
                        OBD_FREE_PTR(lri);
                        RETURN(-ENOMEM);
                }
@@ -413,32 +430,59 @@ int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp,
        lri->lri_exp = exp;
        lri->lri_cb_func = cb_func;
        lri->lri_cb_data = cb_data;
-       CFS_INIT_LIST_HEAD(&lri->lri_list);
-       cfs_list_add(&lri->lri_list, &lwp_register_list);
+       INIT_LIST_HEAD(&lri->lri_list);
+       /*
+        * Initialize the lri_ref at 2, one will be released before
+        * current function returned via lustre_put_lwp_item(), the
+        * other will be released in lustre_deregister_lwp_item().
+        */
+       atomic_set(&lri->lri_ref, 2);
+
+       spin_lock(&lwp_register_list_lock);
+       list_add(&lri->lri_list, &lwp_register_list);
+       if (*exp != NULL)
+               cb = true;
+       spin_unlock(&lwp_register_list_lock);
 
-       if (*exp != NULL && cb_func != NULL)
+       if (cb && cb_func != NULL)
                cb_func(cb_data);
+       lustre_put_lwp_item(lri);
 
-       mutex_unlock(&lwp_register_list_lock);
        RETURN(0);
 }
 EXPORT_SYMBOL(lustre_register_lwp_item);
 
 void lustre_deregister_lwp_item(struct obd_export **exp)
 {
-       struct lwp_register_item *lri, *tmp;
+       struct lwp_register_item *lri;
+       bool removed = false;
+       int repeat = 0;
 
-       mutex_lock(&lwp_register_list_lock);
-       cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
+       spin_lock(&lwp_register_list_lock);
+       list_for_each_entry(lri, &lwp_register_list, lri_list) {
                if (exp == lri->lri_exp) {
-                       if (*exp)
-                               class_export_put(*exp);
-                       cfs_list_del(&lri->lri_list);
-                       OBD_FREE_PTR(lri);
+                       list_del_init(&lri->lri_list);
+                       removed = true;
                        break;
                }
        }
-       mutex_unlock(&lwp_register_list_lock);
+       spin_unlock(&lwp_register_list_lock);
+
+       if (!removed)
+               return;
+
+       /* See lustre_notify_lwp_list(), in some extreme race conditions,
+        * the notify callback could be still on the fly, we need to wait
+        * for the callback done before moving on to free the data used
+        * by callback. */
+       while (atomic_read(&lri->lri_ref) > 1) {
+               CDEBUG(D_MOUNT, "lri reference count %u, repeat: %d\n",
+                      atomic_read(&lri->lri_ref), repeat);
+               repeat++;
+               set_current_state(TASK_INTERRUPTIBLE);
+               schedule_timeout(cfs_time_seconds(1));
+       }
+       lustre_put_lwp_item(lri);
 }
 EXPORT_SYMBOL(lustre_deregister_lwp_item);
 
@@ -469,7 +513,7 @@ struct obd_export *lustre_find_lwp_by_index(const char *dev, __u32 idx)
        list_for_each_entry(lwp, &lsi->lsi_lwp_list, obd_lwp_list) {
                char *ptr = strstr(lwp->obd_name, lwp_name);
 
-               if (ptr != NULL) {
+               if (ptr != NULL && lwp->obd_lwp_export != NULL) {
                        exp = class_export_get(lwp->obd_lwp_export);
                        break;
                }
@@ -483,23 +527,37 @@ err_lmi:
 }
 EXPORT_SYMBOL(lustre_find_lwp_by_index);
 
-static void lustre_notify_lwp_list(struct obd_export *exp)
+void lustre_notify_lwp_list(struct obd_export *exp)
 {
-       struct lwp_register_item *lri, *tmp;
+       struct lwp_register_item *lri;
        LASSERT(exp != NULL);
 
-       mutex_lock(&lwp_register_list_lock);
-       cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
+again:
+       spin_lock(&lwp_register_list_lock);
+       list_for_each_entry(lri, &lwp_register_list, lri_list) {
                if (strcmp(exp->exp_obd->obd_name, lri->lri_name))
                        continue;
                if (*lri->lri_exp != NULL)
                        continue;
                *lri->lri_exp = class_export_get(exp);
-               if (lri->lri_cb_func != NULL)
-                       lri->lri_cb_func(lri->lri_cb_data);
+               if (lri->lri_cb_func == NULL)
+                       continue;
+               atomic_inc(&lri->lri_ref);
+               spin_unlock(&lwp_register_list_lock);
+
+               lri->lri_cb_func(lri->lri_cb_data);
+               lustre_put_lwp_item(lri);
+
+               /* Others may have changed the list after we unlock, we have
+                * to rescan the list from the beginning. Usually, the list
+                * 'lwp_register_list' is very short, and there is 'guard'
+                * lri::lri_exp that will prevent the callback to be done
+                * repeatedly. So rescanning the list has no problem. */
+               goto again;
        }
-       mutex_unlock(&lwp_register_list_lock);
+       spin_unlock(&lwp_register_list_lock);
 }
+EXPORT_SYMBOL(lustre_notify_lwp_list);
 
 static int lustre_lwp_connect(struct obd_device *lwp)
 {
@@ -530,7 +588,8 @@ static int lustre_lwp_connect(struct obd_device *lwp)
        data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
                OBD_CONNECT_AT | OBD_CONNECT_LRU_RESIZE |
                OBD_CONNECT_FULL20 | OBD_CONNECT_LVB_TYPE |
-               OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_LFSCK;
+               OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_LFSCK |
+               OBD_CONNECT_BULK_MBITS;
        OBD_ALLOC_PTR(uuid);
        if (uuid == NULL)
                GOTO(out, rc = -ENOMEM);
@@ -550,7 +609,6 @@ static int lustre_lwp_connect(struct obd_device *lwp)
                if (unlikely(lwp->obd_lwp_export != NULL))
                        class_export_put(lwp->obd_lwp_export);
                lwp->obd_lwp_export = class_export_get(exp);
-               lustre_notify_lwp_list(exp);
        }
 
        GOTO(out, rc);
@@ -606,7 +664,7 @@ static int lustre_lwp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi,
        sprintf(lwpuuid, "%s_UUID", lwpname);
        rc = lustre_start_simple(lwpname, LUSTRE_LWP_NAME,
                                 lwpuuid, lustre_cfg_string(lcfg, 1),
-                                0, 0, 0);
+                                NULL, NULL, NULL);
        if (rc) {
                CERROR("%s: setup up failed: rc %d\n", lwpname, rc);
                GOTO(out, rc);
@@ -697,15 +755,18 @@ static int lustre_lwp_add_conn(struct lustre_cfg *cfg,
        lustre_cfg_bufs_set_string(bufs, 1,
                                   lustre_cfg_string(cfg, 1));
 
-       lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
-       if (lcfg == NULL)
+       OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
+       if (!lcfg)
                GOTO(out_cfg, rc = -ENOMEM);
+       lustre_cfg_init(lcfg, LCFG_ADD_CONN, bufs);
+
        rc = class_add_conn(lwp, lcfg);
        if (rc)
                CERROR("%s: can't add conn: rc = %d\n", lwpname, rc);
 
-       if (lcfg != NULL)
-               lustre_cfg_free(lcfg);
+       if (lcfg)
+               OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount,
+                                             lcfg->lcfg_buflens));
 out_cfg:
        if (bufs != NULL)
                OBD_FREE_PTR(bufs);
@@ -737,7 +798,7 @@ static int client_lwp_config_process(const struct lu_env *env,
                                     struct llog_handle *handle,
                                     struct llog_rec_hdr *rec, void *data)
 {
-       struct config_llog_instance *clli = data;
+       struct config_llog_instance *cfg = data;
        int                          cfg_len = rec->lrh_len;
        char                        *cfg_buf = (char *) (rec + 1);
        struct lustre_cfg           *lcfg = NULL;
@@ -751,8 +812,9 @@ static int client_lwp_config_process(const struct lu_env *env,
                RETURN(-EINVAL);
        }
 
-       LASSERT(clli->cfg_sb != NULL);
-       lsi = s2lsi(clli->cfg_sb);
+       if (cfg->cfg_sb == NULL)
+               GOTO(out, rc = -EINVAL);
+       lsi = s2lsi(cfg->cfg_sb);
 
        lcfg = (struct lustre_cfg *)cfg_buf;
        if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
@@ -774,35 +836,35 @@ static int client_lwp_config_process(const struct lu_env *env,
                    marker->cm_flags & CM_EXCLUDE)
                        GOTO(out, rc = 0);
 
-               if (!tgt_is_mdt(marker->cm_tgtname, &clli->cfg_lwp_idx))
+               if (!tgt_is_mdt(marker->cm_tgtname, &cfg->cfg_lwp_idx))
                        GOTO(out, rc = 0);
 
-               if (IS_MDT(lsi) && clli->cfg_lwp_idx != 0)
+               if (IS_MDT(lsi) && cfg->cfg_lwp_idx != 0)
                        GOTO(out, rc = 0);
 
                if (!strncmp(marker->cm_comment, "add mdc", 7) ||
                    !strncmp(marker->cm_comment, "add failnid", 11)) {
                        if (marker->cm_flags & CM_START) {
-                               clli->cfg_flags = CFG_F_MARKER;
+                               cfg->cfg_flags = CFG_F_MARKER;
                                /* This hack is to differentiate the
                                 * ADD_UUID is come from "add mdc" record
                                 * or from "add failnid" record. */
                                if (!strncmp(marker->cm_comment,
                                             "add failnid", 11))
-                                       clli->cfg_flags |= CFG_F_SKIP;
+                                       cfg->cfg_flags |= CFG_F_SKIP;
                        } else if (marker->cm_flags & CM_END) {
-                               clli->cfg_flags = 0;
+                               cfg->cfg_flags = 0;
                        }
                }
                break;
        }
        case LCFG_ADD_UUID: {
-               if (clli->cfg_flags == CFG_F_MARKER) {
-                       rc = lustre_lwp_setup(lcfg, lsi, clli->cfg_lwp_idx);
+               if (cfg->cfg_flags == CFG_F_MARKER) {
+                       rc = lustre_lwp_setup(lcfg, lsi, cfg->cfg_lwp_idx);
                        /* XXX: process only the first nid as
                         * we don't need another instance of lwp */
-                       clli->cfg_flags |= CFG_F_SKIP;
-               } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
+                       cfg->cfg_flags |= CFG_F_SKIP;
+               } else if (cfg->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
                        rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
                                            lcfg->lcfg_nid);
                        if (rc)
@@ -874,7 +936,7 @@ static int lustre_disconnect_lwp(struct super_block *sb)
                /* end log first */
                cfg->cfg_instance = sb;
                rc = lustre_end_log(sb, logname, cfg);
-               if (rc != 0)
+               if (rc != 0 && rc != -ENOENT)
                        GOTO(out, rc);
 
                lsi->lsi_lwp_started = 0;
@@ -894,15 +956,18 @@ static int lustre_disconnect_lwp(struct super_block *sb)
 
                lustre_cfg_bufs_reset(bufs, lwp->obd_name);
                lustre_cfg_bufs_set_string(bufs, 1, NULL);
-               lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
-               if (lcfg == NULL)
+               OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount,
+                                              bufs->lcfg_buflen));
+               if (!lcfg)
                        GOTO(out, rc = -ENOMEM);
+               lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
 
                /* Disconnect import first. NULL is passed for the '@env',
                 * since it will not be used. */
                rc = lwp->obd_lu_dev->ld_ops->ldo_process_config(NULL,
                                                        lwp->obd_lu_dev, lcfg);
-               lustre_cfg_free(lcfg);
+               OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount,
+                                             lcfg->lcfg_buflens));
                if (rc != 0 && rc != -ETIMEDOUT) {
                        CERROR("%s: fail to disconnect LWP: rc = %d\n",
                               lwp->obd_name, rc);
@@ -983,8 +1048,8 @@ static int lustre_start_lwp(struct super_block *sb)
        cfg->cfg_callback = client_lwp_config_process;
        cfg->cfg_instance = sb;
        rc = lustre_process_log(sb, logname, cfg);
-       if (rc == 0)
-               lsi->lsi_lwp_started = 1;
+       /* need to remove config llog from mgc */
+       lsi->lsi_lwp_started = 1;
 
        GOTO(out, rc);
 
@@ -996,7 +1061,7 @@ out:
        return rc;
 }
 
-DEFINE_MUTEX(server_start_lock);
+static DEFINE_MUTEX(server_start_lock);
 
 /* Stop MDS/OSS if nobody is using them */
 static int server_stop_servers(int lsiflags)
@@ -1044,13 +1109,12 @@ int server_mti_print(const char *title, struct mgs_target_info *mti)
                  mti->mti_config_ver, mti->mti_flags);
        return 0;
 }
-EXPORT_SYMBOL(server_mti_print);
 
 /* Generate data for registration */
 static int server_lsi2mti(struct lustre_sb_info *lsi,
                          struct mgs_target_info *mti)
 {
-       lnet_process_id_t id;
+       struct lnet_process_id id;
        int rc, i = 0;
        int cplen = 0;
        ENTRY;
@@ -1088,6 +1152,13 @@ static int server_lsi2mti(struct lustre_sb_info *lsi,
                }
        }
 
+       if (mti->mti_nid_count == 0) {
+               CERROR("Failed to get NID for server %s, please check whether "
+                      "the target is specifed with improper --servicenode or "
+                      "--network options.\n", mti->mti_svname);
+               RETURN(-EINVAL);
+       }
+
        mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
        mti->mti_config_ver = 0;
 
@@ -1161,8 +1232,8 @@ static int server_register_target(struct lustre_sb_info *lsi)
                                "rc = %d. Is the MGS running?\n",
                                lsi->lsi_svname, rc);
                } else {
-                       CERROR("%s: error registering with the MGS: rc = %d "
-                              "(not fatal)\n", lsi->lsi_svname, rc);
+                       CDEBUG(D_HA, "%s: error registering with the MGS: "
+                              "rc = %d (not fatal)\n", lsi->lsi_svname, rc);
                        /* reset the error code for non-fatal error. */
                        rc = 0;
                }
@@ -1227,7 +1298,7 @@ static int server_start_targets(struct super_block *sb)
        struct obd_device *obd;
        struct lustre_sb_info *lsi = s2lsi(sb);
        struct config_llog_instance cfg;
-       struct lu_env env;
+       struct lu_env mgc_env;
        struct lu_device *dev;
        int rc;
        ENTRY;
@@ -1242,7 +1313,7 @@ static int server_start_targets(struct super_block *sb)
                        rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
                                                 LUSTRE_MDS_NAME,
                                                 LUSTRE_MDS_OBDNAME"_uuid",
-                                                0, 0, 0, 0);
+                                                NULL, NULL, NULL, NULL);
                        if (rc) {
                                mutex_unlock(&server_start_lock);
                                CERROR("failed to start MDS: %d\n", rc);
@@ -1261,7 +1332,7 @@ static int server_start_targets(struct super_block *sb)
                        rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
                                                 LUSTRE_OSS_NAME,
                                                 LUSTRE_OSS_OBDNAME"_uuid",
-                                                0, 0, 0, 0);
+                                                NULL, NULL, NULL, NULL);
                        if (rc) {
                                mutex_unlock(&server_start_lock);
                                CERROR("failed to start OSS: %d\n", rc);
@@ -1271,11 +1342,15 @@ static int server_start_targets(struct super_block *sb)
                mutex_unlock(&server_start_lock);
        }
 
+       rc = lu_env_init(&mgc_env, LCT_MG_THREAD);
+       if (rc != 0)
+               GOTO(out_stop_service, rc);
+
        /* Set the mgc fs to our server disk.  This allows the MGC to
         * read and write configs locally, in case it can't talk to the MGS. */
-       rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
+       rc = server_mgc_set_fs(&mgc_env, lsi->lsi_mgc, sb);
        if (rc)
-               GOTO(out_stop_service, rc);
+               GOTO(out_env, rc);
 
        /* Register with MGS */
        rc = server_register_target(lsi);
@@ -1291,6 +1366,7 @@ static int server_start_targets(struct super_block *sb)
        /* Start targets using the llog named for the target */
        memset(&cfg, 0, sizeof(cfg));
        cfg.cfg_callback = class_config_llog_handler;
+       cfg.cfg_sub_clds = CONFIG_SUB_SERVER;
        rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
        if (rc) {
                CERROR("failed to start server %s: %d\n",
@@ -1321,12 +1397,11 @@ static int server_start_targets(struct super_block *sb)
        /* calculate recovery timeout, do it after lustre_process_log */
        server_calc_timeout(lsi, obd);
 
-       /* log has been fully processed */
-       obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
-
        /* log has been fully processed, let clients connect */
        dev = obd->obd_lu_dev;
        if (dev && dev->ld_ops->ldo_prepare) {
+               struct lu_env env;
+
                rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
                if (rc == 0) {
                        struct lu_context  session_ctx;
@@ -1354,8 +1429,9 @@ static int server_start_targets(struct super_block *sb)
 
 out_mgc:
        /* Release the mgc fs for others to use */
-       server_mgc_clear_fs(lsi->lsi_mgc);
-
+       server_mgc_clear_fs(&mgc_env, lsi->lsi_mgc);
+out_env:
+       lu_env_fini(&mgc_env);
 out_stop_service:
        if (rc != 0)
                server_stop_servers(lsi->lsi_flags);
@@ -1365,6 +1441,8 @@ out_stop_service:
 
 static int lsi_prepare(struct lustre_sb_info *lsi)
 {
+       const char *osd_type;
+       const char *fstype;
        __u32 index;
        int rc;
        ENTRY;
@@ -1378,28 +1456,26 @@ static int lsi_prepare(struct lustre_sb_info *lsi)
                RETURN(-EINVAL);
        }
 
-       if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
-               RETURN(-ENAMETOOLONG);
-
-       strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
-
        /* Determine osd type */
-       if (lsi->lsi_lmd->lmd_osd_type != NULL) {
-               if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
-                   sizeof(lsi->lsi_osd_type))
-                       RETURN(-ENAMETOOLONG);
-
-               strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
+       if (lsi->lsi_lmd->lmd_osd_type == NULL) {
+               osd_type = LUSTRE_OSD_LDISKFS_NAME;
+               fstype = "ldiskfs";
        } else {
-               strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
+               osd_type = lsi->lsi_lmd->lmd_osd_type;
+               fstype = lsi->lsi_lmd->lmd_osd_type;
        }
 
+       if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname) ||
+           strlen(osd_type) >= sizeof(lsi->lsi_osd_type) ||
+           strlen(fstype) >= sizeof(lsi->lsi_fstype))
+               RETURN(-ENAMETOOLONG);
+
+       strlcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile,
+               sizeof(lsi->lsi_svname));
+       strlcpy(lsi->lsi_osd_type, osd_type, sizeof(lsi->lsi_osd_type));
        /* XXX: a temp. solution for components using ldiskfs
         *      to be removed in one of the subsequent patches */
-       if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs"))
-               strcpy(lsi->lsi_fstype, "ldiskfs");
-       else
-               strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
+       strlcpy(lsi->lsi_fstype, fstype, sizeof(lsi->lsi_fstype));
 
        /* Determine server type */
        rc = server_name2index(lsi->lsi_svname, &index, NULL);
@@ -1478,9 +1554,13 @@ static void server_put_super(struct super_block *sb)
                   If there are any setup/cleanup errors, save the lov
                   name for safety cleanup later. */
                lprof = class_get_profile(lsi->lsi_svname);
-               if (lprof && lprof->lp_dt) {
-                       OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
-                       strcpy(extraname, lprof->lp_dt);
+               if (lprof != NULL) {
+                       if (lprof->lp_dt != NULL) {
+                               OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
+                               strncpy(extraname, lprof->lp_dt,
+                                       strlen(lprof->lp_dt) + 1);
+                       }
+                       class_put_profile(lprof);
                }
 
                obd = class_name2obd(lsi->lsi_svname);
@@ -1596,6 +1676,7 @@ static struct super_operations server_ops = {
 /*
  * Xattr support for Lustre servers
  */
+#ifdef HAVE_IOP_XATTR
 static ssize_t lustre_getxattr(struct dentry *dentry, const char *name,
                                void *buffer, size_t size)
 {
@@ -1609,6 +1690,7 @@ static int lustre_setxattr(struct dentry *dentry, const char *name,
 {
        return -EOPNOTSUPP;
 }
+#endif
 
 static ssize_t lustre_listxattr(struct dentry *d_entry, char *name,
                                size_t size)
@@ -1616,9 +1698,11 @@ static ssize_t lustre_listxattr(struct dentry *d_entry, char *name,
        return -EOPNOTSUPP;
 }
 
-const struct inode_operations server_inode_operations = {
+static const struct inode_operations server_inode_operations = {
+#ifdef HAVE_IOP_XATTR
        .setxattr       = lustre_setxattr,
        .getxattr       = lustre_getxattr,
+#endif
        .listxattr      = lustre_listxattr,
 };
 
@@ -1627,7 +1711,7 @@ const struct inode_operations server_inode_operations = {
 
 static int server_fill_super_common(struct super_block *sb)
 {
-       struct inode *root = 0;
+       struct inode *root = NULL;
        ENTRY;
 
        CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
@@ -1688,10 +1772,20 @@ static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
                        GOTO(out, rc);
                obd = class_name2obd(lsi->lsi_osd_obdname);
                LASSERT(obd);
+       } else {
+               CDEBUG(D_MOUNT, "%s already started\n", lsi->lsi_osd_obdname);
+               /* but continue setup to allow special case of MDT and internal
+                * MGT being started separately. */
+               if (!((IS_MGS(lsi) && (lsi->lsi_lmd->lmd_flags &
+                                     LMD_FLG_NOMGS)) ||
+                    (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags &
+                                     LMD_FLG_NOSVC))))
+                       RETURN(-EALREADY);
        }
 
        rc = obd_connect(NULL, &lsi->lsi_osd_exp,
                         obd, &obd->obd_uuid, NULL, NULL);
+
        if (rc) {
                obd->obd_force = 1;
                class_manual_cleanup(obd);
@@ -1724,6 +1818,9 @@ int server_fill_super(struct super_block *sb)
        int rc;
        ENTRY;
 
+       /* to simulate target mount race */
+       OBD_RACE(OBD_FAIL_TGT_MOUNT_RACE);
+
        rc = lsi_prepare(lsi);
        if (rc)
                RETURN(rc);
@@ -1829,28 +1926,25 @@ void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
 
        if (has_ir) {
                int new_soft = soft;
-               int new_hard = hard;
 
                /* adjust timeout value by imperative recovery */
-
                new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
-               new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
-
                /* make sure the timeout is not too short */
                new_soft = max(min, new_soft);
-               new_hard = max(new_soft, new_hard);
 
                LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
                              "window shrunk from %d-%d down to %d-%d\n",
-                             obd->obd_name, soft, hard, new_soft, new_hard);
+                             obd->obd_name, soft, hard, new_soft, hard);
 
                soft = new_soft;
-               hard = new_hard;
+       } else {
+               LCONSOLE_INFO("%s: Imperative Recovery not enabled, recovery "
+                             "window %d-%d\n", obd->obd_name, soft, hard);
        }
 
        /* we're done */
-       obd->obd_recovery_timeout   = max(obd->obd_recovery_timeout, soft);
+       obd->obd_recovery_timeout = max_t(time64_t, obd->obd_recovery_timeout,
+                                         soft);
        obd->obd_recovery_time_hard = hard;
        obd->obd_recovery_ir_factor = factor;
 }
-EXPORT_SYMBOL(server_calc_timeout);