Whamcloud - gitweb
Revert "b=22464 add writeconf as mount option (includes bz 23228)"
[fs/lustre-release.git] / lustre / obdclass / obd_mount.c
index 04d012c..8531464 100644 (file)
@@ -26,7 +26,7 @@
  * GPL HEADER END
  */
 /*
- * Copyright  2008 Sun Microsystems, Inc. All rights reserved
+ * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  * Use is subject to license terms.
  */
 /*
@@ -61,17 +61,18 @@ static void (*kill_super_cb)(struct super_block *sb) = NULL;
 
 /*********** mount lookup *********/
 
-DECLARE_MUTEX(lustre_mount_info_lock);
+CFS_DECLARE_MUTEX(lustre_mount_info_lock);
 static CFS_LIST_HEAD(server_mount_info_list);
 
 static struct lustre_mount_info *server_find_mount(const char *name)
 {
-        struct list_head *tmp;
+        cfs_list_t *tmp;
         struct lustre_mount_info *lmi;
         ENTRY;
 
-        list_for_each(tmp, &server_mount_info_list) {
-                lmi = list_entry(tmp, struct lustre_mount_info, lmi_list_chain);
+        cfs_list_for_each(tmp, &server_mount_info_list) {
+                lmi = cfs_list_entry(tmp, struct lustre_mount_info,
+                                     lmi_list_chain);
                 if (strcmp(name, lmi->lmi_name) == 0)
                         RETURN(lmi);
         }
@@ -101,10 +102,10 @@ static int server_register_mount(const char *name, struct super_block *sb,
         }
         strcpy(name_cp, name);
 
-        down(&lustre_mount_info_lock);
+        cfs_down(&lustre_mount_info_lock);
 
         if (server_find_mount(name)) {
-                up(&lustre_mount_info_lock);
+                cfs_up(&lustre_mount_info_lock);
                 OBD_FREE(lmi, sizeof(*lmi));
                 OBD_FREE(name_cp, strlen(name) + 1);
                 CERROR("Already registered %s\n", name);
@@ -113,12 +114,12 @@ static int server_register_mount(const char *name, struct super_block *sb,
         lmi->lmi_name = name_cp;
         lmi->lmi_sb = sb;
         lmi->lmi_mnt = mnt;
-        list_add(&lmi->lmi_list_chain, &server_mount_info_list);
+        cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list);
 
-        up(&lustre_mount_info_lock);
+        cfs_up(&lustre_mount_info_lock);
 
         CDEBUG(D_MOUNT, "reg_mnt %p from %s, vfscount=%d\n",
-               lmi->lmi_mnt, name, atomic_read(&lmi->lmi_mnt->mnt_count));
+               lmi->lmi_mnt, name, cfs_atomic_read(&lmi->lmi_mnt->mnt_count));
 
         RETURN(0);
 }
@@ -129,21 +130,21 @@ static int server_deregister_mount(const char *name)
         struct lustre_mount_info *lmi;
         ENTRY;
 
-        down(&lustre_mount_info_lock);
+        cfs_down(&lustre_mount_info_lock);
         lmi = server_find_mount(name);
         if (!lmi) {
-                up(&lustre_mount_info_lock);
+                cfs_up(&lustre_mount_info_lock);
                 CERROR("%s not registered\n", name);
                 RETURN(-ENOENT);
         }
 
         CDEBUG(D_MOUNT, "dereg_mnt %p from %s, vfscount=%d\n",
-               lmi->lmi_mnt, name, atomic_read(&lmi->lmi_mnt->mnt_count));
+               lmi->lmi_mnt, name, cfs_atomic_read(&lmi->lmi_mnt->mnt_count));
 
         OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
-        list_del(&lmi->lmi_list_chain);
+        cfs_list_del(&lmi->lmi_list_chain);
         OBD_FREE(lmi, sizeof(*lmi));
-        up(&lustre_mount_info_lock);
+        cfs_up(&lustre_mount_info_lock);
 
         RETURN(0);
 }
@@ -157,20 +158,20 @@ struct lustre_mount_info *server_get_mount(const char *name)
         struct lustre_sb_info *lsi;
         ENTRY;
 
-        down(&lustre_mount_info_lock);
+        cfs_down(&lustre_mount_info_lock);
         lmi = server_find_mount(name);
-        up(&lustre_mount_info_lock);
+        cfs_up(&lustre_mount_info_lock);
         if (!lmi) {
                 CERROR("Can't find mount for %s\n", name);
                 RETURN(NULL);
         }
         lsi = s2lsi(lmi->lmi_sb);
         mntget(lmi->lmi_mnt);
-        atomic_inc(&lsi->lsi_mounts);
+        cfs_atomic_inc(&lsi->lsi_mounts);
 
         CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d, vfscount=%d\n",
-               lmi->lmi_mnt, name, atomic_read(&lsi->lsi_mounts),
-               atomic_read(&lmi->lmi_mnt->mnt_count));
+               lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts),
+               cfs_atomic_read(&lmi->lmi_mnt->mnt_count));
 
         RETURN(lmi);
 }
@@ -185,9 +186,9 @@ struct lustre_mount_info *server_get_mount_2(const char *name)
         struct lustre_mount_info *lmi;
         ENTRY;
 
-        down(&lustre_mount_info_lock);
+        cfs_down(&lustre_mount_info_lock);
         lmi = server_find_mount(name);
-        up(&lustre_mount_info_lock);
+        cfs_up(&lustre_mount_info_lock);
         if (!lmi)
                 CERROR("Can't find mount for %s\n", name);
 
@@ -197,9 +198,9 @@ struct lustre_mount_info *server_get_mount_2(const char *name)
 static void unlock_mntput(struct vfsmount *mnt)
 {
         if (kernel_locked()) {
-                unlock_kernel();
+                cfs_unlock_kernel();
                 mntput(mnt);
-                lock_kernel();
+                cfs_lock_kernel();
         } else {
                 mntput(mnt);
         }
@@ -218,9 +219,9 @@ int server_put_mount(const char *name, struct vfsmount *mnt)
         /* This might be the last one, can't deref after this */
         unlock_mntput(mnt);
 
-        down(&lustre_mount_info_lock);
+        cfs_down(&lustre_mount_info_lock);
         lmi = server_find_mount(name);
-        up(&lustre_mount_info_lock);
+        cfs_up(&lustre_mount_info_lock);
         if (!lmi) {
                 CERROR("Can't find mount for %s\n", name);
                 RETURN(-ENOENT);
@@ -229,7 +230,7 @@ int server_put_mount(const char *name, struct vfsmount *mnt)
         LASSERT(lmi->lmi_mnt == mnt);
 
         CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d, vfscount=%d\n",
-               lmi->lmi_mnt, name, atomic_read(&lsi->lsi_mounts), count);
+               lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts), count);
 
         if (lustre_put_lsi(lmi->lmi_sb)) {
                 CDEBUG(D_MOUNT, "Last put of mnt %p from %s, vfscount=%d\n",
@@ -293,8 +294,8 @@ static int ldd_parse(struct lvfs_run_ctxt *mount_ctxt,
         len = i_size_read(file->f_dentry->d_inode);
         CDEBUG(D_MOUNT, "Have %s, size %lu\n", MOUNT_DATA_FILE, len);
         if (len != sizeof(*ldd)) {
-                CERROR("disk data size does not match: see %lu expect "LPSZ"\n",
-                       len, sizeof(*ldd));
+                CERROR("disk data size does not match: see %lu expect %u\n",
+                       len, (int)sizeof(*ldd));
                 GOTO(out_close, rc = -EINVAL);
         }
 
@@ -376,11 +377,18 @@ out:
 
 /**************** 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. */
+/** 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)
 {
@@ -450,6 +458,9 @@ int lustre_end_log(struct super_block *sb, char *logname,
 
 /**************** 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)
 {
@@ -477,6 +488,9 @@ int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
         return(rc);
 }
 
+/** Call class_attach and class_setup.  These methods in turn call
+ * obd type-specific methods.
+ */
 static int lustre_start_simple(char *obdname, char *type, char *uuid,
                                char *s1, char *s2)
 {
@@ -557,7 +571,7 @@ static int server_stop_mgs(struct super_block *sb)
         RETURN(rc);
 }
 
-DECLARE_MUTEX(mgc_start_lock);
+CFS_DECLARE_MUTEX(mgc_start_lock);
 
 /** Set up a mgc obd to process startup logs
  *
@@ -619,7 +633,7 @@ static int lustre_start_mgc(struct super_block *sb)
 
         mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
 
-        mutex_down(&mgc_start_lock);
+        cfs_mutex_down(&mgc_start_lock);
 
         obd = class_name2obd(mgcname);
         if (obd && !obd->obd_stopping) {
@@ -630,7 +644,7 @@ static int lustre_start_mgc(struct super_block *sb)
                         GOTO(out_free, rc);
 
                 /* Re-using an existing MGC */
-                atomic_inc(&obd->u.cli.cl_mgc_refcount);
+                cfs_atomic_inc(&obd->u.cli.cl_mgc_refcount);
 
                 recov_bk = 0;
                 /* If we are restarting the MGS, don't try to keep the MGC's
@@ -749,7 +763,7 @@ static int lustre_start_mgc(struct super_block *sb)
 
         /* 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);
+        cfs_atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
 
         /* Try all connections, but only once. */
         recov_bk = 1;
@@ -765,7 +779,7 @@ static int lustre_start_mgc(struct super_block *sb)
         if (data == NULL)
                 GOTO(out, rc = -ENOMEM);
         data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID |
-                                  OBD_CONNECT_AT;
+                                  OBD_CONNECT_AT | OBD_CONNECT_FULL20;
         data->ocd_version = LUSTRE_VERSION_CODE;
         rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
         OBD_FREE_PTR(data);
@@ -781,7 +795,7 @@ out:
            to the same mgc.*/
         lsi->lsi_mgc = obd;
 out_free:
-        mutex_up(&mgc_start_lock);
+        cfs_mutex_up(&mgc_start_lock);
 
         if (mgcname)
                 OBD_FREE(mgcname, len);
@@ -805,13 +819,13 @@ static int lustre_stop_mgc(struct super_block *sb)
                 RETURN(-ENOENT);
         lsi->lsi_mgc = NULL;
 
-        mutex_down(&mgc_start_lock);
-        LASSERT(atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
-        if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
+        cfs_mutex_down(&mgc_start_lock);
+        LASSERT(cfs_atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
+        if (!cfs_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));
+                       cfs_atomic_read(&obd->u.cli.cl_mgc_refcount));
                 GOTO(out, rc = -EBUSY);
         }
 
@@ -857,7 +871,7 @@ out:
                 OBD_FREE(niduuid, len);
 
         /* class_import_put will get rid of the additional connections */
-        mutex_up(&mgc_start_lock);
+        cfs_mutex_up(&mgc_start_lock);
         RETURN(rc);
 }
 
@@ -895,7 +909,7 @@ static int server_mgc_clear_fs(struct obd_device *mgc)
         RETURN(rc);
 }
 
-DECLARE_MUTEX(server_start_lock);
+CFS_DECLARE_MUTEX(server_start_lock);
 
 /* Stop MDS/OSS if nobody is using them */
 static int server_stop_servers(int lddflags, int lsiflags)
@@ -905,7 +919,7 @@ static int server_stop_servers(int lddflags, int lsiflags)
         int rc = 0;
         ENTRY;
 
-        mutex_down(&server_start_lock);
+        cfs_mutex_down(&server_start_lock);
 
         /* Either an MDT or an OST or neither  */
         /* if this was an MDT, and there are no more MDT's, clean up the MDS */
@@ -929,7 +943,7 @@ static int server_stop_servers(int lddflags, int lsiflags)
                         rc = err;
         }
 
-        mutex_up(&server_start_lock);
+        cfs_mutex_up(&server_start_lock);
 
         RETURN(rc);
 }
@@ -950,7 +964,7 @@ static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
         struct lustre_sb_info    *lsi = s2lsi(sb);
         struct lustre_disk_data  *ldd = lsi->lsi_ldd;
         lnet_process_id_t         id;
-        int i = 0;
+        int                       i = 0;
         ENTRY;
 
         if (!(lsi->lsi_flags & LSI_SERVER))
@@ -965,6 +979,14 @@ static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
         while (LNetGetId(i++, &id) != -ENOENT) {
                 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
                         continue;
+
+                if (class_find_param(ldd->ldd_params,
+                                     PARAM_NETWORK, NULL) == 0 &&
+                    !class_match_net(ldd->ldd_params, id.nid)) {
+                        /* can't match specified network */
+                        continue;
+                }
+
                 mti->mti_nids[mti->mti_nid_count] = id.nid;
                 mti->mti_nid_count++;
                 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
@@ -1057,7 +1079,8 @@ out:
         RETURN(rc);
 }
 
-/* Start targets */
+/** Start server targets: MDTs and OSTs
+ */
 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
 {
         struct obd_device *obd;
@@ -1072,7 +1095,7 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
         /* If we're an MDT, make sure the global MDS is running */
         if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
                 /* make sure the MDS is started */
-                mutex_down(&server_start_lock);
+                cfs_mutex_down(&server_start_lock);
                 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
                 if (!obd) {
                         rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
@@ -1081,19 +1104,19 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
                                                  LUSTRE_MDS_OBDNAME"_uuid",
                                                  0, 0);
                         if (rc) {
-                                mutex_up(&server_start_lock);
+                                cfs_mutex_up(&server_start_lock);
                                 CERROR("failed to start MDS: %d\n", rc);
                                 RETURN(rc);
                         }
                 }
-                mutex_up(&server_start_lock);
+                cfs_mutex_up(&server_start_lock);
         }
 #endif
 
         /* If we're an OST, make sure the global OSS is running */
         if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
                 /* make sure OSS is started */
-                mutex_down(&server_start_lock);
+                cfs_mutex_down(&server_start_lock);
                 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
                 if (!obd) {
                         rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
@@ -1101,16 +1124,16 @@ static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
                                                  LUSTRE_OSS_OBDNAME"_uuid",
                                                  0, 0);
                         if (rc) {
-                                mutex_up(&server_start_lock);
+                                cfs_mutex_up(&server_start_lock);
                                 CERROR("failed to start OSS: %d\n", rc);
                                 RETURN(rc);
                         }
                 }
-                mutex_up(&server_start_lock);
+                cfs_mutex_up(&server_start_lock);
         }
 
-        /* Set the mgc fs to our server disk.  This allows the MGC
-           to read and write configs locally. */
+        /* 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);
         if (rc)
                 RETURN(rc);
@@ -1198,9 +1221,11 @@ struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
         }
 
         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);
+        cfs_atomic_set(&lsi->lsi_mounts, 1);
 
         /* Default umount style */
         lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
@@ -1217,7 +1242,7 @@ static int lustre_free_lsi(struct super_block *sb)
         CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
 
         /* someone didn't call server_put_mount. */
-        LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
+        LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0);
 
         if (lsi->lsi_ldd != NULL)
                 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
@@ -1258,8 +1283,8 @@ static int lustre_put_lsi(struct super_block *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)) {
+        CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts));
+        if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) {
                 lustre_free_lsi(sb);
                 RETURN(1);
         }
@@ -1268,7 +1293,11 @@ static int lustre_put_lsi(struct super_block *sb)
 
 /*************** server mount ******************/
 
-/* Kernel mount using mount options in MOUNT_DATA_FILE */
+/** Kernel mount using mount options in MOUNT_DATA_FILE.
+ * Since this file lives on the disk, we pre-mount using a common
+ * type, read the file, then re-mount using the type specified in the
+ * file.
+ */
 static struct vfsmount *server_kernel_mount(struct super_block *sb)
 {
         struct lvfs_run_ctxt mount_ctxt;
@@ -1373,30 +1402,46 @@ out_free:
         RETURN(ERR_PTR(rc));
 }
 
+/** Wait here forever until the mount refcount is 0 before completing umount,
+ * else we risk dereferencing a null pointer.
+ * LNET may take e.g. 165s before killing zombies.
+ */
 static void server_wait_finished(struct vfsmount *mnt)
 {
-        wait_queue_head_t   waitq;
-        struct l_wait_info  lwi;
-        int                 retries = 330;
-
-        init_waitqueue_head(&waitq);
-
-        while ((atomic_read(&mnt->mnt_count) > 1) && (retries > 0)) {
-                LCONSOLE_WARN("Mount still busy with %d refs, waiting for "
-                              "%d secs...\n",
-                              atomic_read(&mnt->mnt_count), retries);
-
-                /* Wait for a bit */
-                retries -= 5;
-                lwi = LWI_TIMEOUT(5 * HZ, NULL, NULL);
-                l_wait_event(waitq, 0, &lwi);
-        }
-        if (atomic_read(&mnt->mnt_count) > 1) {
-                CERROR("Mount %p is still busy (%d refs), giving up.\n",
-                       mnt, atomic_read(&mnt->mnt_count));
-        }
+       cfs_waitq_t             waitq;
+       int                     rc, waited = 0;
+       cfs_sigset_t            blocked;
+
+       cfs_waitq_init(&waitq);
+
+       while (atomic_read(&mnt->mnt_count) > 1) {
+               if (waited && (waited % 30 == 0))
+                       LCONSOLE_WARN("Mount still busy with %d refs after "
+                                      "%d secs.\n",
+                                      atomic_read(&mnt->mnt_count),
+                                      waited);
+               /* Cannot use l_event_wait() for an interruptible sleep. */
+               waited += 3;
+               blocked = l_w_e_set_sigs(sigmask(SIGKILL));
+               cfs_waitq_wait_event_interruptible_timeout(
+                       waitq,
+                       (atomic_read(&mnt->mnt_count) == 1),
+                       cfs_time_seconds(3),
+                       rc);
+               cfs_block_sigs(blocked);
+               if (rc < 0) {
+                       LCONSOLE_EMERG("Danger: interrupted umount %s with "
+                                      "%d refs!\n",
+                                     mnt->mnt_devname,
+                                      atomic_read(&mnt->mnt_count));
+                       break;
+               }
+
+       }
 }
 
+/** Start the shutdown of servers at umount.
+ */
 static void server_put_super(struct super_block *sb)
 {
         struct lustre_sb_info *lsi = s2lsi(sb);
@@ -1414,6 +1459,8 @@ static void server_put_super(struct super_block *sb)
         OBD_ALLOC(tmpname, tmpname_sz);
         memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
         CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
+        if (IS_MDT(lsi->lsi_ldd) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
+                snprintf(tmpname, tmpname_sz, "MGS");
 
         /* Stop the target */
         if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
@@ -1486,6 +1533,8 @@ static void server_put_super(struct super_block *sb)
         EXIT;
 }
 
+/** Called only for 'umount -f'
+ */
 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
 {
@@ -1514,10 +1563,10 @@ static void server_umount_begin(struct super_block *sb)
 }
 
 #ifndef HAVE_STATFS_DENTRY_PARAM
-static int server_statfs (struct super_block *sb, struct kstatfs *buf)
+static int server_statfs (struct super_block *sb, cfs_kstatfs_t *buf)
 {
 #else
-static int server_statfs (struct dentry *dentry, struct kstatfs *buf)
+static int server_statfs (struct dentry *dentry, cfs_kstatfs_t *buf)
 {
         struct super_block *sb = dentry->d_sb;
 #endif
@@ -1548,6 +1597,9 @@ static int server_statfs (struct dentry *dentry, struct kstatfs *buf)
         RETURN(0);
 }
 
+/** The operations we support directly on the superblock:
+ * mount, umount, and df.
+ */
 static struct super_operations server_ops =
 {
         .put_super      = server_put_super,
@@ -1555,7 +1607,7 @@ static struct super_operations server_ops =
         .statfs         = server_statfs,
 };
 
-#define log2(n) ffz(~(n))
+#define log2(n) cfs_ffz(~(n))
 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
 
 static int server_fill_super_common(struct super_block *sb)
@@ -1593,6 +1645,11 @@ static int server_fill_super_common(struct super_block *sb)
         RETURN(0);
 }
 
+/** Fill in the superblock info for a Lustre server.
+ * Mount the device with the correct options.
+ * Read the on-disk config file.
+ * Start the services.
+ */
 static int server_fill_super(struct super_block *sb)
 {
         struct lustre_sb_info *lsi = s2lsi(sb);
@@ -1633,6 +1690,7 @@ static int server_fill_super(struct super_block *sb)
                         GOTO(out_mnt, rc);
         }
 
+        /* Start MGC before servers */
         rc = lustre_start_mgc(sb);
         if (rc)
                 GOTO(out_mnt, rc);
@@ -1660,9 +1718,6 @@ static int server_fill_super(struct super_block *sb)
         if (rc)
                 GOTO(out_mnt, rc);
 
-        LCONSOLE_WARN("Server %s on device %s has started\n",
-                      lsi->lsi_ldd->ldd_svname, lsi->lsi_lmd->lmd_dev);
-
         RETURN(0);
 out_mnt:
         /* We jump here in case of failure while starting targets or MGS.
@@ -1684,6 +1739,15 @@ int server_name2index(char *svname, __u32 *idx, char **endptr)
         if (!dash)
                 return(-EINVAL);
 
+        /* intepret <fsname>-MDTXXXXX-mdc as mdt, the better way is to pass
+         * in the fsname, then determine the server index */
+        if (!strcmp(LUSTRE_MDC_NAME, dash + 1)) {
+                dash--;
+                for (; dash > svname && *dash != '-'; dash--);
+                if (dash == svname)
+                        return(-EINVAL);
+        }
+
         if (strncmp(dash + 1, "MDT", 3) == 0)
                 rc = LDD_F_SV_TYPE_MDT;
         else if (strncmp(dash + 1, "OST", 3) == 0)
@@ -1725,7 +1789,6 @@ int lustre_common_put_super(struct super_block *sb)
         RETURN(rc);
 }
 
-#if 0
 static void lmd_print(struct lustre_mount_data *lmd)
 {
         int i;
@@ -1735,14 +1798,23 @@ static void lmd_print(struct lustre_mount_data *lmd)
                 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
         PRINT_CMD(PRINT_MASK, "device:  %s\n", lmd->lmd_dev);
         PRINT_CMD(PRINT_MASK, "flags:   %x\n", lmd->lmd_flags);
+
         if (lmd->lmd_opts)
                 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
+
+        if (lmd->lmd_recovery_time_soft)
+                PRINT_CMD(PRINT_MASK, "recovery time soft: %d\n",
+                          lmd->lmd_recovery_time_soft);
+
+        if (lmd->lmd_recovery_time_hard)
+                PRINT_CMD(PRINT_MASK, "recovery time hard: %d\n",
+                          lmd->lmd_recovery_time_hard);
+
         for (i = 0; i < lmd->lmd_exclude_count; i++) {
                 PRINT_CMD(PRINT_MASK, "exclude %d:  OST%04x\n", i,
                           lmd->lmd_exclude[i]);
         }
 }
-#endif
 
 /* Is this server on the exclusion list */
 int lustre_check_exclusion(struct super_block *sb, char *svname)
@@ -1850,7 +1922,10 @@ static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
         return 0;
 }
 
-/* mount -v -t lustre uml1:uml2:/lustre-client /mnt/lustre */
+/** 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;
@@ -1879,6 +1954,9 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd)
         s1 = options;
         while (*s1) {
                 int clear = 0;
+                int time_min = 2 * (CONNECTION_SWITCH_MAX +
+                               2 * INITIAL_CONNECT_TIMEOUT);
+
                 /* Skip whitespace and extra commas */
                 while (*s1 == ' ' || *s1 == ',')
                         s1++;
@@ -1891,6 +1969,14 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd)
                 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, "nosvc", 5) == 0) {
                         lmd->lmd_flags |= LMD_FLG_NOSVC;
                         clear++;
@@ -1970,6 +2056,7 @@ static int lmd_parse(char *options, struct lustre_mount_data *lmd)
                 strcpy(lmd->lmd_opts, options);
         }
 
+        lmd_print(lmd);
         lmd->lmd_magic = LMD_MAGIC;
 
         RETURN(rc);
@@ -1980,7 +2067,11 @@ invalid:
 }
 
 
-/* Common mount */
+/** 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;
@@ -1999,7 +2090,7 @@ int lustre_fill_super(struct super_block *sb, void *data, int silent)
          * Disable lockdep during mount, because mount locking patterns are
          * `special'.
          */
-        lockdep_off();
+        cfs_lockdep_off();
 
         /* Figure out the lmd from the mount options */
         if (lmd_parse((char *)data, lmd)) {
@@ -2047,7 +2138,7 @@ out:
                 CDEBUG(D_SUPER, "Mount %s complete\n",
                        lmd->lmd_dev);
         }
-        lockdep_on();
+        cfs_lockdep_on();
         return rc;
 }
 
@@ -2091,12 +2182,17 @@ void lustre_kill_super(struct super_block *sb)
         kill_anon_super(sb);
 }
 
+/** Register the "lustre" fs type
+ */
 struct file_system_type lustre_fs_type = {
         .owner        = THIS_MODULE,
         .name         = "lustre",
         .get_sb       = lustre_get_sb,
         .kill_sb      = lustre_kill_super,
         .fs_flags     = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
+#ifdef FS_HAS_FIEMAP
+                        FS_HAS_FIEMAP |
+#endif
                         LL_RENAME_DOES_D_MOVE,
 };