Whamcloud - gitweb
Branch HEAD
[fs/lustre-release.git] / lustre / obdclass / obd_config.c
index 646e7fe..a6917f1 100644 (file)
@@ -1,28 +1,41 @@
 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
  * vim:expandtab:shiftwidth=8:tabstop=8:
  *
- *  Copyright (c) 2001-2006 Cluster File Systems, Inc.
+ * GPL HEADER START
  *
- *   This file is part of the Lustre file system, http://www.lustre.org
- *   Lustre is a trademark of Cluster File Systems, Inc.
+ * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
- *   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.
+ * 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.
  *
- *   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.
+ * 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).
  *
- *   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.
+ * 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
  *
- * Config API
+ * 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  2008 Sun Microsystems, Inc. All rights reserved
+ * Use is subject to license terms.
+ */
+/*
+ * This file is part of Lustre, http://www.lustre.org/
+ * Lustre is a trademark of Sun Microsystems, Inc.
+ *
+ * lustre/obdclass/obd_config.c
  *
+ * Config API
  */
 
 #define DEBUG_SUBSYSTEM S_CLASS
@@ -40,8 +53,9 @@
 #include <lustre_param.h>
 #include <class_hash.h>
 
-extern struct lustre_hash_operations uuid_hash_operations;
-extern struct lustre_hash_operations nid_hash_operations;
+static lustre_hash_ops_t uuid_hash_ops;
+static lustre_hash_ops_t nid_hash_ops;
+static lustre_hash_ops_t nid_stat_hash_ops;
 
 /*********** string parsing utils *********/
 
@@ -50,35 +64,101 @@ int class_find_param(char *buf, char *key, char **valp)
 {
         char *ptr;
 
-        if (!buf) 
+        if (!buf)
                 return 1;
 
-        if ((ptr = strstr(buf, key)) == NULL) 
+        if ((ptr = strstr(buf, key)) == NULL)
                 return 1;
 
-        if (valp) 
+        if (valp)
                 *valp = ptr + strlen(key);
-        
+
         return 0;
 }
 
+/**
+ * Finds a parameter in \a params and copies it to \a copy.
+ *
+ * Leading spaces are skipped. Next space or end of string is the
+ * parameter terminator with the exception that spaces inside single or double
+ * quotes get included into a parameter. The parameter is copied into \a copy
+ * which has to be allocated big enough by a caller, quotes are stripped in
+ * the copy and the copy is terminated by 0.
+ *
+ * On return \a params is set to next parameter or to NULL if last
+ * parameter is returned.
+ *
+ * \retval 0 if parameter is returned in \a copy
+ * \retval 1 otherwise
+ * \retval -EINVAL if unbalanced quota is found
+ */
+int class_get_next_param(char **params, char *copy)
+{
+        char *q1, *q2, *str;
+        int len;
+
+        str = *params;
+        while (*str == ' ')
+                str++;
+
+        if (*str == '\0') {
+                *params = NULL;
+                return 1;
+        }
+
+        while (1) {
+                q1 = strpbrk(str, " '\"");
+                if (q1 == NULL) {
+                        len = strlen(str);
+                        memcpy(copy, str, len);
+                        copy[len] = '\0';
+                        *params = NULL;
+                        return 0;
+                }
+                len = q1 - str;
+                if (*q1 == ' ') {
+                        memcpy(copy, str, len);
+                        copy[len] = '\0';
+                        *params = str + len;
+                        return 0;
+                }
+
+                memcpy(copy, str, len);
+                copy += len;
+
+                /* search for the matching closing quote */
+                str = q1 + 1;
+                q2 = strchr(str, *q1);
+                if (q2 == NULL) {
+                        CERROR("Unbalanced quota in parameters: \"%s\"\n",
+                               *params);
+                        return -EINVAL;
+                }
+                len = q2 - str;
+                memcpy(copy, str, len);
+                copy += len;
+                str = q2 + 1;
+        }
+        return 1;
+}
+
 /* returns 0 if this is the first key in the buffer, else 1.
    valp points to first char after key. */
 int class_match_param(char *buf, char *key, char **valp)
 {
-        if (!buf) 
+        if (!buf)
                 return 1;
 
-        if (memcmp(buf, key, strlen(key)) != 0) 
+        if (memcmp(buf, key, strlen(key)) != 0)
                 return 1;
 
-        if (valp) 
+        if (valp)
                 *valp = buf + strlen(key);
-        
+
         return 0;
 }
 
-/* 0 is good nid, 
+/* 0 is good nid,
    1 not found
    < 0 error
    endh is set to next separator */
@@ -86,16 +166,16 @@ int class_parse_nid(char *buf, lnet_nid_t *nid, char **endh)
 {
         char tmp, *endp;
 
-        if (!buf) 
+        if (!buf)
                 return 1;
-        while (*buf == ',' || *buf == ':') 
+        while (*buf == ',' || *buf == ':')
                 buf++;
-        if (*buf == ' ' || *buf == '/' || *buf == '\0') 
+        if (*buf == ' ' || *buf == '/' || *buf == '\0')
                 return 1;
 
         /* nid separators or end of nids */
         endp = strpbrk(buf, ",: /");
-        if (endp == NULL) 
+        if (endp == NULL)
                 endp = buf + strlen(buf);
 
         tmp = *endp;
@@ -108,20 +188,22 @@ int class_parse_nid(char *buf, lnet_nid_t *nid, char **endh)
         }
         *endp = tmp;
 
-        if (endh) 
+        if (endh)
                 *endh = endp;
         CDEBUG(D_INFO, "Nid %s\n", libcfs_nid2str(*nid));
         return 0;
 }
 
 EXPORT_SYMBOL(class_find_param);
+EXPORT_SYMBOL(class_get_next_param);
 EXPORT_SYMBOL(class_match_param);
 EXPORT_SYMBOL(class_parse_nid);
 
 /********************** class fns **********************/
 
-/* Create a new device and set the type, name and uuid.  If
- * successful, the new device can be accessed by either name or uuid.
+/**
+ * Create a new device and set the type, name and uuid.  If successful, the new
+ * device can be accessed by either name or uuid.
  */
 int class_attach(struct lustre_cfg *lcfg)
 {
@@ -162,17 +244,23 @@ int class_attach(struct lustre_cfg *lcfg)
         }
         LASSERTF(obd != NULL, "Cannot get obd device %s of type %s\n",
                  name, typename);
-        LASSERTF(obd->obd_magic == OBD_DEVICE_MAGIC, 
+        LASSERTF(obd->obd_magic == OBD_DEVICE_MAGIC,
                  "obd %p obd_magic %08X != %08X\n",
                  obd, obd->obd_magic, OBD_DEVICE_MAGIC);
-        LASSERTF(strncmp(obd->obd_name, name, strlen(name)) == 0, "%p obd_name %s != %s\n",
-                 obd, obd->obd_name, name);
+        LASSERTF(strncmp(obd->obd_name, name, strlen(name)) == 0,
+                 "%p obd_name %s != %s\n", obd, obd->obd_name, name);
+
+        rwlock_init(&obd->obd_pool_lock);
+        obd->obd_pool_limit = 0;
+        obd->obd_pool_slv = 0;
 
         CFS_INIT_LIST_HEAD(&obd->obd_exports);
+        CFS_INIT_LIST_HEAD(&obd->obd_delayed_exports);
         CFS_INIT_LIST_HEAD(&obd->obd_exports_timed);
+        CFS_INIT_LIST_HEAD(&obd->obd_nid_stats);
+        spin_lock_init(&obd->obd_nid_lock);
         spin_lock_init(&obd->obd_dev_lock);
         sema_init(&obd->obd_dev_sem, 1);
-        sema_init(&obd->obd_proc_exp_sem, 1);
         spin_lock_init(&obd->obd_osfs_lock);
         /* obd->obd_osfs_age must be set to a value in the distant
          * past to guarantee a fresh statfs is fetched on mount. */
@@ -188,13 +276,12 @@ int class_attach(struct lustre_cfg *lcfg)
         CFS_INIT_LIST_HEAD(&obd->obd_lock_replay_queue);
         CFS_INIT_LIST_HEAD(&obd->obd_final_req_queue);
 
-        spin_lock_init(&obd->obd_uncommitted_replies_lock);
-        CFS_INIT_LIST_HEAD(&obd->obd_uncommitted_replies);
+        llog_group_init(&obd->obd_olg, FILTER_GROUP_LLOG);
 
         len = strlen(uuid);
         if (len >= sizeof(obd->obd_uuid)) {
-                CERROR("uuid must be < "LPSZ" bytes long\n",
-                       sizeof(obd->obd_uuid));
+                CERROR("uuid must be < %d bytes long\n",
+                       (int)sizeof(obd->obd_uuid));
                 GOTO(out, rc = -EINVAL);
         }
         memcpy(obd->obd_uuid.uuid, uuid, len);
@@ -210,6 +297,8 @@ int class_attach(struct lustre_cfg *lcfg)
         spin_lock(&obd->obd_dev_lock);
         atomic_set(&obd->obd_refcount, 1);
         spin_unlock(&obd->obd_dev_lock);
+        lu_ref_init(&obd->obd_reference);
+        lu_ref_add(&obd->obd_reference, "attach", obd);
 
         obd->obd_attached = 1;
         CDEBUG(D_IOCTL, "OBD: dev %d attached type %s with refcount %d\n",
@@ -229,9 +318,11 @@ int class_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
         ENTRY;
 
         LASSERT(obd != NULL);
-        LASSERTF(obd == class_num2obd(obd->obd_minor), "obd %p != obd_devs[%d] %p\n", 
+        LASSERTF(obd == class_num2obd(obd->obd_minor),
+                 "obd %p != obd_devs[%d] %p\n",
                  obd, obd->obd_minor, class_num2obd(obd->obd_minor));
-        LASSERTF(obd->obd_magic == OBD_DEVICE_MAGIC, "obd %p obd_magic %08x != %08x\n", 
+        LASSERTF(obd->obd_magic == OBD_DEVICE_MAGIC,
+                 "obd %p obd_magic %08x != %08x\n",
                  obd, obd->obd_magic, OBD_DEVICE_MAGIC);
 
         /* have we attached a type to this device? */
@@ -257,24 +348,39 @@ int class_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
         /* just leave this on forever.  I can't use obd_set_up here because
            other fns check that status, and we're not actually set up yet. */
         obd->obd_starting = 1;
+        obd->obd_uuid_hash = NULL;
+        obd->obd_nid_hash = NULL;
+        obd->obd_nid_stats_hash = NULL;
         spin_unlock(&obd->obd_dev_lock);
 
-        /* create an uuid-export hash body */
-        err = lustre_hash_init(&obd->obd_uuid_hash_body, "UUID_HASH", 
-                               128, &uuid_hash_operations);
-        if (err) {
-                GOTO(err_hash, err);
-        }
-
-        /* create a nid-export hash body */
-        err = lustre_hash_init(&obd->obd_nid_hash_body, "NID_HASH", 
-                               128, &nid_hash_operations);
-        if (err)
-                GOTO(err_hash, err);
+        /* create an uuid-export lustre hash */
+        obd->obd_uuid_hash = lustre_hash_init("UUID_HASH",
+                                              HASH_UUID_CUR_BITS,
+                                              HASH_UUID_MAX_BITS,
+                                              &uuid_hash_ops, 0);
+        if (!obd->obd_uuid_hash)
+                GOTO(err_hash, err = -ENOMEM);
+
+        /* create a nid-export lustre hash */
+        obd->obd_nid_hash = lustre_hash_init("NID_HASH",
+                                             HASH_NID_CUR_BITS,
+                                             HASH_NID_MAX_BITS,
+                                             &nid_hash_ops, 0);
+        if (!obd->obd_nid_hash)
+                GOTO(err_hash, err = -ENOMEM);
+
+        /* create a nid-stats lustre hash */
+        obd->obd_nid_stats_hash = lustre_hash_init("NID_STATS",
+                                                   HASH_NID_STATS_CUR_BITS,
+                                                   HASH_NID_STATS_MAX_BITS,
+                                                   &nid_stat_hash_ops, 0);
+        if (!obd->obd_nid_stats_hash)
+                GOTO(err_hash, err = -ENOMEM);
 
         exp = class_new_export(obd, &obd->obd_uuid);
         if (IS_ERR(exp))
-                RETURN(PTR_ERR(exp));
+                GOTO(err_hash, err = PTR_ERR(exp));
+
         obd->obd_self_export = exp;
         list_del_init(&exp->exp_obd_chain_timed);
         class_export_put(exp);
@@ -284,26 +390,37 @@ int class_setup(struct obd_device *obd, struct lustre_cfg *lcfg)
                 GOTO(err_exp, err);
 
         obd->obd_set_up = 1;
-        
+
         spin_lock(&obd->obd_dev_lock);
         /* cleanup drops this */
-        class_incref(obd);
+        class_incref(obd, "setup", obd);
         spin_unlock(&obd->obd_dev_lock);
 
         CDEBUG(D_IOCTL, "finished setup of obd %s (uuid %s)\n",
                obd->obd_name, obd->obd_uuid.uuid);
 
         RETURN(0);
-
 err_exp:
-        class_unlink_export(obd->obd_self_export);
-        obd->obd_self_export = NULL;
+        if (obd->obd_self_export) {
+                class_unlink_export(obd->obd_self_export);
+                obd->obd_self_export = NULL;
+        }
 err_hash:
-        lustre_hash_exit(&obd->obd_uuid_hash_body);
-        lustre_hash_exit(&obd->obd_nid_hash_body);
+        if (obd->obd_uuid_hash) {
+                lustre_hash_exit(obd->obd_uuid_hash);
+                obd->obd_uuid_hash = NULL;
+        }
+        if (obd->obd_nid_hash) {
+                lustre_hash_exit(obd->obd_nid_hash);
+                obd->obd_nid_hash = NULL;
+        }
+        if (obd->obd_nid_stats_hash) {
+                lustre_hash_exit(obd->obd_nid_stats_hash);
+                obd->obd_nid_stats_hash = NULL;
+        }
         obd->obd_starting = 0;
         CERROR("setup %s failed (%d)\n", obd->obd_name, err);
-        RETURN(err);
+        return err;
 }
 
 int class_detach(struct obd_device *obd, struct lustre_cfg *lcfg)
@@ -327,29 +444,32 @@ int class_detach(struct obd_device *obd, struct lustre_cfg *lcfg)
         CDEBUG(D_IOCTL, "detach on obd %s (uuid %s)\n",
                obd->obd_name, obd->obd_uuid.uuid);
 
-        class_decref(obd);
-        
+        class_decref(obd, "attach", obd);
+
         /* not strictly necessary, but cleans up eagerly */
         obd_zombie_impexp_cull();
-        
+
         RETURN(0);
 }
 
 static void dump_exports(struct obd_device *obd)
 {
-        struct obd_export *exp, *n;
+        struct obd_export *exp;
 
-        list_for_each_entry_safe(exp, n, &obd->obd_exports, exp_obd_chain) {
+        spin_lock(&obd->obd_dev_lock);
+        list_for_each_entry(exp, &obd->obd_exports, exp_obd_chain) {
                 struct ptlrpc_reply_state *rs;
                 struct ptlrpc_reply_state *first_reply = NULL;
                 int                        nreplies = 0;
 
+                spin_lock(&exp->exp_lock);
                 list_for_each_entry (rs, &exp->exp_outstanding_replies,
                                      rs_exp_list) {
                         if (nreplies == 0)
                                 first_reply = rs;
                         nreplies++;
                 }
+                spin_unlock(&exp->exp_lock);
 
                 CDEBUG(D_IOCTL, "%s: %p %s %s %d %d %d: %p %s\n",
                        obd->obd_name, exp, exp->exp_client_uuid.uuid,
@@ -358,6 +478,7 @@ static void dump_exports(struct obd_device *obd)
                        exp->exp_failed, nreplies, first_reply,
                        nreplies > 3 ? "..." : "");
         }
+        spin_unlock(&obd->obd_dev_lock);
 }
 
 int class_cleanup(struct obd_device *obd, struct lustre_cfg *lcfg)
@@ -395,33 +516,27 @@ int class_cleanup(struct obd_device *obd, struct lustre_cfg *lcfg)
                                 obd->obd_fail = 1;
                                 obd->obd_no_transno = 1;
                                 obd->obd_no_recov = 1;
-                                /* Set the obd readonly if we can */
-                                if (OBP(obd, iocontrol))
+                                if (OBP(obd, iocontrol)) {
+                                        obd_iocontrol(OBD_IOC_SYNC,
+                                                      obd->obd_self_export,
+                                                      0, NULL, NULL);
+                                        /* Set the obd readonly if we can */
                                         obd_iocontrol(OBD_IOC_SET_READONLY,
                                                       obd->obd_self_export,
                                                       0, NULL, NULL);
+                                }
                                 break;
                         default:
-                                CERROR("unrecognised flag '%c'\n",
-                                       *flag);
+                                CERROR("Unrecognised flag '%c'\n", *flag);
                         }
         }
 
+        LASSERT(obd->obd_self_export);
+
         /* The three references that should be remaining are the
          * obd_self_export and the attach and setup references. */
         if (atomic_read(&obd->obd_refcount) > 3) {
-#if 0           /* We should never fail to cleanup with mountconf */ 
-                if (!(obd->obd_fail || obd->obd_force)) {
-                        CERROR("OBD %s is still busy with %d references\n"
-                               "You should stop active file system users,"
-                               " or use the --force option to cleanup.\n",
-                               obd->obd_name, atomic_read(&obd->obd_refcount));
-                        dump_exports(obd);
-                        /* Allow a failed cleanup to try again. */
-                        obd->obd_stopping = 0;
-                }
-#endif
-                /* refcounf - 3 might be the number of real exports 
+                /* refcounf - 3 might be the number of real exports
                    (excluding self export). But class_incref is called
                    by other things as well, so don't count on it. */
                 CDEBUG(D_IOCTL, "%s: forcing exports to disconnect: %d\n",
@@ -429,28 +544,40 @@ int class_cleanup(struct obd_device *obd, struct lustre_cfg *lcfg)
                 dump_exports(obd);
                 class_disconnect_exports(obd);
         }
-        LASSERT(obd->obd_self_export);
 
         /* destroy an uuid-export hash body */
-        lustre_hash_exit(&obd->obd_uuid_hash_body);
+        if (obd->obd_uuid_hash) {
+                lustre_hash_exit(obd->obd_uuid_hash);
+                obd->obd_uuid_hash = NULL;
+        }
 
         /* destroy a nid-export hash body */
-        lustre_hash_exit(&obd->obd_nid_hash_body);
+        if (obd->obd_nid_hash) {
+                lustre_hash_exit(obd->obd_nid_hash);
+                obd->obd_nid_hash = NULL;
+        }
 
-        /* Precleanup stage 1, we must make sure all exports (other than the
-           self-export) get destroyed. */
+        /* destroy a nid-stats hash body */
+        if (obd->obd_nid_stats_hash) {
+                lustre_hash_exit(obd->obd_nid_stats_hash);
+                obd->obd_nid_stats_hash = NULL;
+        }
+
+        /* Precleanup, we must make sure all exports get destroyed. */
         err = obd_precleanup(obd, OBD_CLEANUP_EXPORTS);
         if (err)
                 CERROR("Precleanup %s returned %d\n",
                        obd->obd_name, err);
-
-        class_decref(obd);
+        class_decref(obd, "setup", obd);
         obd->obd_set_up = 0;
         RETURN(0);
 }
 
-struct obd_device *class_incref(struct obd_device *obd)
+struct obd_device *class_incref(struct obd_device *obd,
+                                const char *scope, const void *source)
 {
+        LASSERT(!obd->obd_stopping);
+        lu_ref_add_atomic(&obd->obd_reference, scope, source);
         atomic_inc(&obd->obd_refcount);
         CDEBUG(D_INFO, "incref %s (%p) now %d\n", obd->obd_name, obd,
                atomic_read(&obd->obd_refcount));
@@ -458,7 +585,7 @@ struct obd_device *class_incref(struct obd_device *obd)
         return obd;
 }
 
-void class_decref(struct obd_device *obd)
+void class_decref(struct obd_device *obd, const char *scope, const void *source)
 {
         int err;
         int refs;
@@ -467,25 +594,16 @@ void class_decref(struct obd_device *obd)
         atomic_dec(&obd->obd_refcount);
         refs = atomic_read(&obd->obd_refcount);
         spin_unlock(&obd->obd_dev_lock);
+        lu_ref_del(&obd->obd_reference, scope, source);
 
         CDEBUG(D_INFO, "Decref %s (%p) now %d\n", obd->obd_name, obd, refs);
 
         if ((refs == 1) && obd->obd_stopping) {
-                /* All exports (other than the self-export) have been
-                   destroyed; there should be no more in-progress ops
-                   by this point.*/
-                /* if we're not stopping, we didn't finish setup */
-                /* Precleanup stage 2,  do other type-specific
-                   cleanup requiring the self-export. */
-                err = obd_precleanup(obd, OBD_CLEANUP_SELF_EXP);
-                if (err)
-                        CERROR("Precleanup %s returned %d\n",
-                               obd->obd_name, err);
-                
+                /* All exports have been destroyed; there should
+                   be no more in-progress ops by this point.*/
+
                 spin_lock(&obd->obd_self_export->exp_lock);
-                obd->obd_self_export->exp_flags |=
-                        (obd->obd_fail ? OBD_OPT_FAILOVER : 0) |
-                        (obd->obd_force ? OBD_OPT_FORCE : 0);
+                obd->obd_self_export->exp_flags |= exp_flags_from_obd(obd);
                 spin_unlock(&obd->obd_self_export->exp_lock);
 
                 /* note that we'll recurse into class_decref again */
@@ -505,7 +623,7 @@ void class_decref(struct obd_device *obd)
                                        obd->obd_name, err);
                 }
                 if (OBP(obd, detach)) {
-                        err = OBP(obd,detach)(obd);
+                        err = OBP(obd, detach)(obd);
                         if (err)
                                 CERROR("Detach returned %d\n", err);
                 }
@@ -633,7 +751,7 @@ out:
                 OBD_FREE(lprof->lp_dt, osclen);
         if (lprof->lp_profile)
                 OBD_FREE(lprof->lp_profile, proflen);
-        OBD_FREE(lprof, sizeof(*lprof));        
+        OBD_FREE(lprof, sizeof(*lprof));
         RETURN(err);
 }
 
@@ -673,6 +791,28 @@ void class_del_profiles(void)
         EXIT;
 }
 
+static int class_set_global(char *ptr, int val) {
+        ENTRY;
+
+        if (class_match_param(ptr, PARAM_AT_MIN, NULL) == 0)
+            at_min = val;
+        else if (class_match_param(ptr, PARAM_AT_MAX, NULL) == 0)
+                at_max = val;
+        else if (class_match_param(ptr, PARAM_AT_EXTRA, NULL) == 0)
+                at_extra = val;
+        else if (class_match_param(ptr, PARAM_AT_EARLY_MARGIN, NULL) == 0)
+                at_early_margin = val;
+        else if (class_match_param(ptr, PARAM_AT_HISTORY, NULL) == 0)
+                at_history = val;
+        else
+                RETURN(-EINVAL);
+
+        CDEBUG(D_IOCTL, "global %s = %d\n", ptr, val);
+
+        RETURN(0);
+}
+
+
 /* We can't call ll_process_config directly because it lives in a module that
    must be loaded after this one. */
 static int (*client_process_config)(struct lustre_cfg *lcfg) = NULL;
@@ -738,7 +878,15 @@ int class_process_config(struct lustre_cfg *lcfg)
                 CDEBUG(D_IOCTL, "changing lustre timeout from %d to %d\n",
                        obd_timeout, lcfg->lcfg_num);
                 obd_timeout = max(lcfg->lcfg_num, 1U);
-                obd_health_check_timeout = HEALTH_CHECK_TIMEOUT;
+                GOTO(out, err = 0);
+        }
+        case LCFG_SET_LDLM_TIMEOUT: {
+                CDEBUG(D_IOCTL, "changing lustre ldlm_timeout from %d to %d\n",
+                       ldlm_timeout, lcfg->lcfg_num);
+                ldlm_timeout = max(lcfg->lcfg_num, 1U);
+                if (ldlm_timeout >= obd_timeout)
+                        ldlm_timeout = max(obd_timeout / 3, 1U);
+
                 GOTO(out, err = 0);
         }
         case LCFG_SET_UPCALL: {
@@ -754,13 +902,24 @@ int class_process_config(struct lustre_cfg *lcfg)
                 GOTO(out, err = 0);
         }
         case LCFG_PARAM: {
+                char *tmp;
                 /* llite has no obd */
-                if ((class_match_param(lustre_cfg_string(lcfg, 1), 
+                if ((class_match_param(lustre_cfg_string(lcfg, 1),
                                        PARAM_LLITE, 0) == 0) &&
                     client_process_config) {
                         err = (*client_process_config)(lcfg);
                         GOTO(out, err);
+                } else if ((class_match_param(lustre_cfg_string(lcfg, 1),
+                                              PARAM_SYS, &tmp) == 0)) {
+                        /* Global param settings */
+                        err = class_set_global(tmp, lcfg->lcfg_num);
+                        /* Note that since LCFG_PARAM is LCFG_REQUIRED, new
+                           unknown globals would cause config to fail */
+                        if (err)
+                                CWARN("Ignoring unknown param %s\n", tmp);
+                        GOTO(out, 0);
                 }
+
                 /* Fall through */
                 break;
         }
@@ -799,6 +958,28 @@ int class_process_config(struct lustre_cfg *lcfg)
                 err = class_del_conn(obd, lcfg);
                 GOTO(out, err = 0);
         }
+        case LCFG_POOL_NEW: {
+                err = obd_pool_new(obd, lustre_cfg_string(lcfg, 2));
+                GOTO(out, err = 0);
+                break;
+        }
+        case LCFG_POOL_ADD: {
+                err = obd_pool_add(obd, lustre_cfg_string(lcfg, 2),
+                                   lustre_cfg_string(lcfg, 3));
+                GOTO(out, err = 0);
+                break;
+        }
+        case LCFG_POOL_REM: {
+                err = obd_pool_rem(obd, lustre_cfg_string(lcfg, 2),
+                                   lustre_cfg_string(lcfg, 3));
+                GOTO(out, err = 0);
+                break;
+        }
+        case LCFG_POOL_DEL: {
+                err = obd_pool_del(obd, lustre_cfg_string(lcfg, 2));
+                GOTO(out, err = 0);
+                break;
+        }
         default: {
                 err = obd_process_config(obd, sizeof(*lcfg), lcfg);
                 GOTO(out, err);
@@ -807,14 +988,14 @@ int class_process_config(struct lustre_cfg *lcfg)
         }
 out:
         if ((err < 0) && !(lcfg->lcfg_command & LCFG_REQUIRED)) {
-                CWARN("Ignoring error %d on optional command %#x\n", err, 
+                CWARN("Ignoring error %d on optional command %#x\n", err,
                       lcfg->lcfg_command);
                 err = 0;
         }
         return err;
 }
 
-int class_process_proc_param(char *prefix, struct lprocfs_vars *lvars, 
+int class_process_proc_param(char *prefix, struct lprocfs_vars *lvars,
                              struct lustre_cfg *lcfg, void *data)
 {
 #ifdef __KERNEL__
@@ -823,6 +1004,7 @@ int class_process_proc_param(char *prefix, struct lprocfs_vars *lvars,
         int i, keylen, vallen;
         int matched = 0, j = 0;
         int rc = 0;
+        int skip = 0;
         ENTRY;
 
         if (lcfg->lcfg_command != LCFG_PARAM) {
@@ -839,9 +1021,8 @@ int class_process_proc_param(char *prefix, struct lprocfs_vars *lvars,
                 class_match_param(key, prefix, &key);
                 sval = strchr(key, '=');
                 if (!sval || (*(sval + 1) == 0)) {
-                        CERROR("Can't parse param %s\n", key);
-                        rc = -EINVAL;
-                        /* continue parsing other params */
+                        CERROR("Can't parse param %s (missing '=')\n", key);
+                        /* rc = -EINVAL;        continue parsing other params */
                         continue;
                 }
                 keylen = sval - key;
@@ -864,28 +1045,35 @@ int class_process_proc_param(char *prefix, struct lprocfs_vars *lvars,
                                                                vallen, data);
                                         set_fs(oldfs);
                                 }
-                                if (rc < 0) 
-                                        CERROR("writing proc entry %s err %d\n", 
-                                               var->name, rc);
                                 break;
                         }
                         j++;
-                }    
+                }
                 if (!matched) {
+                        /* If the prefix doesn't match, return error so we
+                           can pass it down the stack */
+                        if (strnchr(key, keylen, '.'))
+                            RETURN(-ENOSYS);
                         CERROR("%s: unknown param %s\n",
                                (char *)lustre_cfg_string(lcfg, 0), key);
-                        rc = -EINVAL;
-                        /* continue parsing other params */
+                        /* rc = -EINVAL;        continue parsing other params */
+                        skip++;
+                } else if (rc < 0) {
+                        CERROR("writing proc entry %s err %d\n",
+                               var->name, rc);
+                        rc = 0;
                 } else {
-                        LCONSOLE_INFO("%s.%.*s: set parameter %.*s=%s\n", 
-                                      (char *)lustre_cfg_string(lcfg, 0),
+                        LCONSOLE_INFO("%s.%.*s: set parameter %.*s=%s\n",
+                                      lustre_cfg_string(lcfg, 0),
                                       (int)strlen(prefix) - 1, prefix,
                                       (int)(sval - key - 1), key, sval);
                 }
         }
-        
-        if (rc > 0) 
+
+        if (rc > 0)
                 rc = 0;
+        if (!rc && skip)
+                rc = skip;
         RETURN(rc);
 #else
         CDEBUG(D_CONFIG, "liblustre can't process params.\n");
@@ -920,11 +1108,13 @@ static int class_config_llog_handler(struct llog_handle * handle,
                 struct lustre_cfg_bufs bufs;
                 char *inst_name = NULL;
                 int inst_len = 0;
-                int inst = 0;
+                int inst = 0, swab = 0;
 
                 lcfg = (struct lustre_cfg *)cfg_buf;
-                if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION))
+                if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
                         lustre_swab_lustre_cfg(lcfg);
+                        swab = 1;
+                }
 
                 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
                 if (rc)
@@ -933,18 +1123,21 @@ static int class_config_llog_handler(struct llog_handle * handle,
                 /* Figure out config state info */
                 if (lcfg->lcfg_command == LCFG_MARKER) {
                         struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
+                        lustre_swab_cfg_marker(marker, swab,
+                                               LUSTRE_CFG_BUFLEN(lcfg, 1));
                         CDEBUG(D_CONFIG, "Marker, inst_flg=%#x mark_flg=%#x\n",
                                clli->cfg_flags, marker->cm_flags);
                         if (marker->cm_flags & CM_START) {
                                 /* all previous flags off */
                                 clli->cfg_flags = CFG_F_MARKER;
-                                if (marker->cm_flags & CM_SKIP) { 
+                                if (marker->cm_flags & CM_SKIP) {
                                         clli->cfg_flags |= CFG_F_SKIP;
                                         CDEBUG(D_CONFIG, "SKIP #%d\n",
                                                marker->cm_step);
                                 } else if ((marker->cm_flags & CM_EXCLUDE) ||
-                                           lustre_check_exclusion(clli->cfg_sb, 
-                                                          marker->cm_tgtname)) {
+                                           (clli->cfg_sb &&
+                                            lustre_check_exclusion(clli->cfg_sb,
+                                                         marker->cm_tgtname))) {
                                         clli->cfg_flags |= CFG_F_EXCLUDE;
                                         CDEBUG(D_CONFIG, "EXCLUDE %d\n",
                                                marker->cm_step);
@@ -953,10 +1146,10 @@ static int class_config_llog_handler(struct llog_handle * handle,
                                 clli->cfg_flags = 0;
                         }
                 }
-                /* A config command without a start marker before it is 
+                /* A config command without a start marker before it is
                    illegal (post 146) */
                 if (!(clli->cfg_flags & CFG_F_COMPAT146) &&
-                    !(clli->cfg_flags & CFG_F_MARKER) && 
+                    !(clli->cfg_flags & CFG_F_MARKER) &&
                     (lcfg->lcfg_command != LCFG_MARKER)) {
                         CWARN("Config not inside markers, ignoring! "
                               "(inst: %s, uuid: %s, flags: %#x)\n",
@@ -972,14 +1165,37 @@ static int class_config_llog_handler(struct llog_handle * handle,
                         break;
                 }
 
-                if ((clli->cfg_flags & CFG_F_EXCLUDE) && 
+                /*
+                 * For interoperability between 1.8 and 2.0,
+                 * rename "mds" obd device type to "mdt".
+                 */
+                {
+                        char *typename = lustre_cfg_string(lcfg, 1);
+                        char *index = lustre_cfg_string(lcfg, 2);
+
+                        if ((lcfg->lcfg_command == LCFG_ATTACH && typename &&
+                             strcmp(typename, "mds") == 0)) {
+                                CWARN("For 1.8 interoperability, rename obd "
+                                       "type from mds to mdt\n");
+                                typename[2] = 't';
+                        }
+                        if ((lcfg->lcfg_command == LCFG_SETUP && index &&
+                             strcmp(index, "type") == 0)) {
+                                CDEBUG(D_INFO, "For 1.8 interoperability, "
+                                       "set this index to '0'\n");
+                                index[0] = '0';
+                                index[1] = 0;
+                        }
+                }
+
+                if ((clli->cfg_flags & CFG_F_EXCLUDE) &&
                     (lcfg->lcfg_command == LCFG_LOV_ADD_OBD))
                         /* Add inactive instead */
                         lcfg->lcfg_command = LCFG_LOV_ADD_INA;
 
                 lustre_cfg_bufs_init(&bufs, lcfg);
 
-                if (clli && clli->cfg_instance && 
+                if (clli && clli->cfg_instance &&
                     LUSTRE_CFG_BUFLEN(lcfg, 0) > 0){
                         inst = 1;
                         inst_len = LUSTRE_CFG_BUFLEN(lcfg, 0) +
@@ -997,11 +1213,27 @@ static int class_config_llog_handler(struct llog_handle * handle,
 
                 /* we override the llog's uuid for clients, to insure they
                 are unique */
-                if (clli && clli->cfg_instance && 
+                if (clli && clli->cfg_instance &&
                     lcfg->lcfg_command == LCFG_ATTACH) {
                         lustre_cfg_bufs_set_string(&bufs, 2,
                                                    clli->cfg_uuid.uuid);
                 }
+                /*
+                 * sptlrpc config record, we expect 2 data segments:
+                 *  [0]: fs_name/target_name,
+                 *  [1]: rule string
+                 * moving them to index [1] and [2], and insert MGC's
+                 * obdname at index [0].
+                 */
+                if (clli && clli->cfg_instance == NULL &&
+                    lcfg->lcfg_command == LCFG_SPTLRPC_CONF) {
+                        lustre_cfg_bufs_set(&bufs, 2, bufs.lcfg_buf[1],
+                                            bufs.lcfg_buflen[1]);
+                        lustre_cfg_bufs_set(&bufs, 1, bufs.lcfg_buf[0],
+                                            bufs.lcfg_buflen[0]);
+                        lustre_cfg_bufs_set_string(&bufs, 0,
+                                                   clli->cfg_obdname);
+                }
 
                 lcfg_new = lustre_cfg_new(lcfg->lcfg_command, &bufs);
 
@@ -1064,16 +1296,16 @@ int class_config_parse_llog(struct llog_ctxt *ctxt, char *name,
 
         /* continue processing from where we last stopped to end-of-log */
         if (cfg)
-                cd.first_idx = cfg->cfg_last_idx;
-        cd.last_idx = 0;
+                cd.lpcd_first_idx = cfg->cfg_last_idx;
+        cd.lpcd_last_idx = 0;
 
         rc = llog_process(llh, class_config_llog_handler, cfg, &cd);
 
-        CDEBUG(D_CONFIG, "Processed log %s gen %d-%d (rc=%d)\n", name, 
-               cd.first_idx + 1, cd.last_idx, rc);
+        CDEBUG(D_CONFIG, "Processed log %s gen %d-%d (rc=%d)\n", name,
+               cd.lpcd_first_idx + 1, cd.lpcd_last_idx, rc);
 
         if (cfg)
-                cfg->cfg_last_idx = cd.last_idx;
+                cfg->cfg_last_idx = cd.lpcd_last_idx;
 
 parse_out:
         rc2 = llog_close(llh);
@@ -1174,10 +1406,10 @@ parse_out:
 /* Cleanup and detach */
 int class_manual_cleanup(struct obd_device *obd)
 {
-        struct lustre_cfg *lcfg;
-        struct lustre_cfg_bufs bufs;
-        int rc;
-        char flags[3]="";
+        char                    flags[3] = "";
+        struct lustre_cfg      *lcfg;
+        struct lustre_cfg_bufs  bufs;
+        int                     rc;
         ENTRY;
 
         if (!obd) {
@@ -1196,6 +1428,8 @@ int class_manual_cleanup(struct obd_device *obd)
         lustre_cfg_bufs_reset(&bufs, obd->obd_name);
         lustre_cfg_bufs_set_string(&bufs, 1, flags);
         lcfg = lustre_cfg_new(LCFG_CLEANUP, &bufs);
+        if (!lcfg)
+                RETURN(-ENOMEM);
 
         rc = class_process_config(lcfg);
         if (rc) {
@@ -1213,3 +1447,187 @@ out:
         RETURN(rc);
 }
 
+/*
+ * uuid<->export lustre hash operations
+ */
+
+static unsigned
+uuid_hash(lustre_hash_t *lh,  void *key, unsigned mask)
+{
+        return lh_djb2_hash(((struct obd_uuid *)key)->uuid,
+                            sizeof(((struct obd_uuid *)key)->uuid), mask);
+}
+
+static void *
+uuid_key(struct hlist_node *hnode)
+{
+        struct obd_export *exp;
+
+        exp = hlist_entry(hnode, struct obd_export, exp_uuid_hash);
+
+        RETURN(&exp->exp_client_uuid);
+}
+
+/*
+ * NOTE: It is impossible to find an export that is in failed
+ *       state with this function
+ */
+static int
+uuid_compare(void *key, struct hlist_node *hnode)
+{
+        struct obd_export *exp;
+
+        LASSERT(key);
+        exp = hlist_entry(hnode, struct obd_export, exp_uuid_hash);
+
+        RETURN(obd_uuid_equals((struct obd_uuid *)key,&exp->exp_client_uuid) &&
+               !exp->exp_failed);
+}
+
+static void *
+uuid_export_get(struct hlist_node *hnode)
+{
+        struct obd_export *exp;
+
+        exp = hlist_entry(hnode, struct obd_export, exp_uuid_hash);
+        class_export_get(exp);
+
+        RETURN(exp);
+}
+
+static void *
+uuid_export_put(struct hlist_node *hnode)
+{
+        struct obd_export *exp;
+
+        exp = hlist_entry(hnode, struct obd_export, exp_uuid_hash);
+        class_export_put(exp);
+
+        RETURN(exp);
+}
+
+static lustre_hash_ops_t uuid_hash_ops = {
+        .lh_hash    = uuid_hash,
+        .lh_key     = uuid_key,
+        .lh_compare = uuid_compare,
+        .lh_get     = uuid_export_get,
+        .lh_put     = uuid_export_put,
+};
+
+
+/*
+ * nid<->export hash operations
+ */
+
+static unsigned
+nid_hash(lustre_hash_t *lh,  void *key, unsigned mask)
+{
+        return lh_djb2_hash(key, sizeof(lnet_nid_t), mask);
+}
+
+static void *
+nid_key(struct hlist_node *hnode)
+{
+        struct obd_export *exp;
+
+        exp = hlist_entry(hnode, struct obd_export, exp_nid_hash);
+
+        RETURN(&exp->exp_connection->c_peer.nid);
+}
+
+/*
+ * NOTE: It is impossible to find an export that is in failed
+ *       state with this function
+ */
+static int
+nid_compare(void *key, struct hlist_node *hnode)
+{
+        struct obd_export *exp;
+
+        LASSERT(key);
+        exp = hlist_entry(hnode, struct obd_export, exp_nid_hash);
+
+        RETURN(exp->exp_connection->c_peer.nid == *(lnet_nid_t *)key &&
+               !exp->exp_failed);
+}
+
+static void *
+nid_export_get(struct hlist_node *hnode)
+{
+        struct obd_export *exp;
+
+        exp = hlist_entry(hnode, struct obd_export, exp_nid_hash);
+        class_export_get(exp);
+
+        RETURN(exp);
+}
+
+static void *
+nid_export_put(struct hlist_node *hnode)
+{
+        struct obd_export *exp;
+
+        exp = hlist_entry(hnode, struct obd_export, exp_nid_hash);
+        class_export_put(exp);
+
+        RETURN(exp);
+}
+
+static lustre_hash_ops_t nid_hash_ops = {
+        .lh_hash    = nid_hash,
+        .lh_key     = nid_key,
+        .lh_compare = nid_compare,
+        .lh_get     = nid_export_get,
+        .lh_put     = nid_export_put,
+};
+
+
+/*
+ * nid<->nidstats hash operations
+ */
+
+static void *
+nidstats_key(struct hlist_node *hnode)
+{
+        struct nid_stat *ns;
+
+        ns = hlist_entry(hnode, struct nid_stat, nid_hash);
+
+        RETURN(&ns->nid);
+}
+
+static int
+nidstats_compare(void *key, struct hlist_node *hnode)
+{
+        RETURN(*(lnet_nid_t *)nidstats_key(hnode) == *(lnet_nid_t *)key);
+}
+
+static void *
+nidstats_get(struct hlist_node *hnode)
+{
+        struct nid_stat *ns;
+
+        ns = hlist_entry(hnode, struct nid_stat, nid_hash);
+        nidstat_getref(ns);
+
+        RETURN(ns);
+}
+
+static void *
+nidstats_put(struct hlist_node *hnode)
+{
+        struct nid_stat *ns;
+
+        ns = hlist_entry(hnode, struct nid_stat, nid_hash);
+        nidstat_putref(ns);
+
+        RETURN(ns);
+}
+
+static lustre_hash_ops_t nid_stat_hash_ops = {
+        .lh_hash    = nid_hash,
+        .lh_key     = nidstats_key,
+        .lh_compare = nidstats_compare,
+        .lh_get     = nidstats_get,
+        .lh_put     = nidstats_put,
+};