4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/obdclass/obd_mount.c
38 * Client/server mount routines
40 * Author: Nathan Rutman <nathan@clusterfs.com>
44 #define DEBUG_SUBSYSTEM S_CLASS
45 #define D_MOUNT D_SUPER|D_CONFIG /*|D_WARNING */
46 #define PRINT_CMD CDEBUG
47 #define PRINT_MASK D_SUPER|D_CONFIG
51 #include <lustre_fsfilt.h>
52 #include <obd_class.h>
53 #include <lustre/lustre_user.h>
54 #include <linux/version.h>
55 #include <lustre_log.h>
56 #include <lustre_disk.h>
57 #include <lustre_param.h>
58 #ifdef HAVE_KERNEL_LOCKED
59 #include <linux/smp_lock.h>
62 static int (*client_fill_super)(struct super_block *sb,
63 struct vfsmount *mnt) = NULL;
64 static void (*kill_super_cb)(struct super_block *sb) = NULL;
66 /*********** mount lookup *********/
68 CFS_DEFINE_MUTEX(lustre_mount_info_lock);
69 static CFS_LIST_HEAD(server_mount_info_list);
71 static struct lustre_mount_info *server_find_mount(const char *name)
74 struct lustre_mount_info *lmi;
77 cfs_list_for_each(tmp, &server_mount_info_list) {
78 lmi = cfs_list_entry(tmp, struct lustre_mount_info,
80 if (strcmp(name, lmi->lmi_name) == 0)
86 /* we must register an obd for a mount before we call the setup routine.
87 *_setup will call lustre_get_mount to get the mnt struct
88 by obd_name, since we can't pass the pointer to setup. */
89 static int server_register_mount(const char *name, struct super_block *sb,
92 struct lustre_mount_info *lmi;
98 OBD_ALLOC(lmi, sizeof(*lmi));
101 OBD_ALLOC(name_cp, strlen(name) + 1);
103 OBD_FREE(lmi, sizeof(*lmi));
106 strcpy(name_cp, name);
108 cfs_mutex_lock(&lustre_mount_info_lock);
110 if (server_find_mount(name)) {
111 cfs_mutex_unlock(&lustre_mount_info_lock);
112 OBD_FREE(lmi, sizeof(*lmi));
113 OBD_FREE(name_cp, strlen(name) + 1);
114 CERROR("Already registered %s\n", name);
117 lmi->lmi_name = name_cp;
120 cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list);
122 cfs_mutex_unlock(&lustre_mount_info_lock);
124 CDEBUG(D_MOUNT, "reg_mnt %p from %s\n", lmi->lmi_mnt, name);
129 /* when an obd no longer needs a mount */
130 static int server_deregister_mount(const char *name)
132 struct lustre_mount_info *lmi;
135 cfs_mutex_lock(&lustre_mount_info_lock);
136 lmi = server_find_mount(name);
138 cfs_mutex_unlock(&lustre_mount_info_lock);
139 CERROR("%s not registered\n", name);
143 CDEBUG(D_MOUNT, "dereg_mnt %p from %s\n", lmi->lmi_mnt, name);
145 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
146 cfs_list_del(&lmi->lmi_list_chain);
147 OBD_FREE(lmi, sizeof(*lmi));
148 cfs_mutex_unlock(&lustre_mount_info_lock);
153 /* obd's look up a registered mount using their obdname. This is just
154 for initial obd setup to find the mount struct. It should not be
155 called every time you want to mntget. */
156 struct lustre_mount_info *server_get_mount(const char *name)
158 struct lustre_mount_info *lmi;
159 struct lustre_sb_info *lsi;
162 cfs_mutex_lock(&lustre_mount_info_lock);
163 lmi = server_find_mount(name);
164 cfs_mutex_unlock(&lustre_mount_info_lock);
166 CERROR("Can't find mount for %s\n", name);
169 lsi = s2lsi(lmi->lmi_sb);
171 cfs_atomic_inc(&lsi->lsi_mounts);
173 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d\n", lmi->lmi_mnt,
174 name, cfs_atomic_read(&lsi->lsi_mounts));
178 EXPORT_SYMBOL(server_get_mount);
181 * Used by mdt to get mount_info from obdname.
182 * There are no blocking when using the mount_info.
183 * Do not use server_get_mount for this purpose.
185 struct lustre_mount_info *server_get_mount_2(const char *name)
187 struct lustre_mount_info *lmi;
190 cfs_mutex_lock(&lustre_mount_info_lock);
191 lmi = server_find_mount(name);
192 cfs_mutex_unlock(&lustre_mount_info_lock);
194 CERROR("Can't find mount for %s\n", name);
198 EXPORT_SYMBOL(server_get_mount_2);
200 static int lustre_put_lsi(struct super_block *sb);
202 /* to be called from obd_cleanup methods */
203 int server_put_mount(const char *name, struct vfsmount *mnt)
205 struct lustre_mount_info *lmi;
206 struct lustre_sb_info *lsi;
209 cfs_mutex_lock(&lustre_mount_info_lock);
210 lmi = server_find_mount(name);
211 cfs_mutex_unlock(&lustre_mount_info_lock);
213 CERROR("Can't find mount for %s\n", name);
216 lsi = s2lsi(lmi->lmi_sb);
218 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d\n",
219 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts));
221 if (lustre_put_lsi(lmi->lmi_sb))
222 CDEBUG(D_MOUNT, "Last put of mnt %p from %s\n",
225 /* this obd should never need the mount again */
226 server_deregister_mount(name);
230 EXPORT_SYMBOL(server_put_mount);
232 /* Corresponding to server_get_mount_2 */
233 int server_put_mount_2(const char *name, struct vfsmount *mnt)
238 EXPORT_SYMBOL(server_put_mount_2);
240 /**************** config llog ********************/
242 /** Get a config log from the MGS and process it.
243 * This func is called for both clients and servers.
244 * Continue to process new statements appended to the logs
245 * (whenever the config lock is revoked) until lustre_end_log
247 * @param sb The superblock is used by the MGC to write to the local copy of
249 * @param logname The name of the llog to replicate from the MGS
250 * @param cfg Since the same mgc may be used to follow multiple config logs
251 * (e.g. ost1, ost2, client), the config_llog_instance keeps the state for
252 * this log, and is added to the mgc's list of logs to follow.
254 int lustre_process_log(struct super_block *sb, char *logname,
255 struct config_llog_instance *cfg)
257 struct lustre_cfg *lcfg;
258 struct lustre_cfg_bufs *bufs;
259 struct lustre_sb_info *lsi = s2lsi(sb);
260 struct obd_device *mgc = lsi->lsi_mgc;
271 /* mgc_process_config */
272 lustre_cfg_bufs_reset(bufs, mgc->obd_name);
273 lustre_cfg_bufs_set_string(bufs, 1, logname);
274 lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg));
275 lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb));
276 lcfg = lustre_cfg_new(LCFG_LOG_START, bufs);
277 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
278 lustre_cfg_free(lcfg);
283 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
284 "failed from the MGS (%d). Make sure this "
285 "client and the MGS are running compatible "
286 "versions of Lustre.\n",
287 mgc->obd_name, logname, rc);
290 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
291 "failed (%d). This may be the result of "
292 "communication errors between this node and "
293 "the MGS, a bad configuration, or other "
294 "errors. See the syslog for more "
295 "information.\n", mgc->obd_name, logname,
298 /* class_obd_list(); */
301 EXPORT_SYMBOL(lustre_process_log);
303 /* Stop watching this config log for updates */
304 int lustre_end_log(struct super_block *sb, char *logname,
305 struct config_llog_instance *cfg)
307 struct lustre_cfg *lcfg;
308 struct lustre_cfg_bufs bufs;
309 struct lustre_sb_info *lsi = s2lsi(sb);
310 struct obd_device *mgc = lsi->lsi_mgc;
317 /* mgc_process_config */
318 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
319 lustre_cfg_bufs_set_string(&bufs, 1, logname);
321 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
322 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
323 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
324 lustre_cfg_free(lcfg);
327 EXPORT_SYMBOL(lustre_end_log);
329 /**************** obd start *******************/
331 /** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from
332 * lctl (and do for echo cli/srv.
334 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
335 char *s1, char *s2, char *s3, char *s4)
337 struct lustre_cfg_bufs bufs;
338 struct lustre_cfg * lcfg = NULL;
341 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
342 cmd, s1, s2, s3, s4);
344 lustre_cfg_bufs_reset(&bufs, cfgname);
346 lustre_cfg_bufs_set_string(&bufs, 1, s1);
348 lustre_cfg_bufs_set_string(&bufs, 2, s2);
350 lustre_cfg_bufs_set_string(&bufs, 3, s3);
352 lustre_cfg_bufs_set_string(&bufs, 4, s4);
354 lcfg = lustre_cfg_new(cmd, &bufs);
355 lcfg->lcfg_nid = nid;
356 rc = class_process_config(lcfg);
357 lustre_cfg_free(lcfg);
360 EXPORT_SYMBOL(do_lcfg);
362 /** Call class_attach and class_setup. These methods in turn call
363 * obd type-specific methods.
365 static int lustre_start_simple(char *obdname, char *type, char *uuid,
366 char *s1, char *s2, char *s3, char *s4)
369 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
371 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
373 CERROR("%s attach error %d\n", obdname, rc);
376 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, s3, s4);
378 CERROR("%s setup error %d\n", obdname, rc);
379 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
384 /* Set up a MGS to serve startup logs */
385 static int server_start_mgs(struct super_block *sb)
387 struct lustre_sb_info *lsi = s2lsi(sb);
388 struct vfsmount *mnt = lsi->lsi_srv_mnt;
389 struct lustre_mount_info *lmi;
393 /* It is impossible to have more than 1 MGS per node, since
394 MGC wouldn't know which to connect to */
395 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
397 lsi = s2lsi(lmi->lmi_sb);
398 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
403 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
405 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
408 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
409 LUSTRE_MGS_OBDNAME, 0, 0,
410 lsi->lsi_osd_obdname, 0);
411 /* Do NOT call server_deregister_mount() here. This leads to
412 * inability cleanup cleanly and free lsi and other stuff when
413 * mgs calls server_put_mount() in error handling case. -umka */
417 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
418 "Is the 'mgs' module loaded?\n",
419 LUSTRE_MGS_OBDNAME, rc);
423 static int server_stop_mgs(struct super_block *sb)
425 struct obd_device *obd;
429 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
431 /* There better be only one MGS */
432 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
434 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
438 /* The MGS should always stop when we say so */
440 rc = class_manual_cleanup(obd);
444 CFS_DEFINE_MUTEX(mgc_start_lock);
446 /** Set up a mgc obd to process startup logs
448 * \param sb [in] super block of the mgc obd
450 * \retval 0 success, otherwise error code
452 static int lustre_start_mgc(struct super_block *sb)
454 struct obd_connect_data *data = NULL;
455 struct lustre_sb_info *lsi = s2lsi(sb);
456 struct obd_device *obd;
457 struct obd_export *exp;
458 struct obd_uuid *uuid;
461 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
464 int rc = 0, i = 0, j, len;
467 LASSERT(lsi->lsi_lmd);
469 /* Find the first non-lo MGS nid for our MGC name */
470 if (IS_SERVER(lsi)) {
471 /* mount -o mgsnode=nid */
472 ptr = lsi->lsi_lmd->lmd_mgs;
473 if (lsi->lsi_lmd->lmd_mgs &&
474 (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) {
476 } else if (IS_MGS(lsi)) {
477 lnet_process_id_t id;
478 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
479 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
486 } else { /* client */
487 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
488 ptr = lsi->lsi_lmd->lmd_dev;
489 if (class_parse_nid(ptr, &nid, &ptr) == 0)
493 CERROR("No valid MGS nids found.\n");
497 cfs_mutex_lock(&mgc_start_lock);
499 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
500 OBD_ALLOC(mgcname, len);
501 OBD_ALLOC(niduuid, len + 2);
502 if (!mgcname || !niduuid)
503 GOTO(out_free, rc = -ENOMEM);
504 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
506 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
510 GOTO(out_free, rc = -ENOMEM);
512 obd = class_name2obd(mgcname);
513 if (obd && !obd->obd_stopping) {
514 rc = obd_set_info_async(NULL, obd->obd_self_export,
515 strlen(KEY_MGSSEC), KEY_MGSSEC,
516 strlen(mgssec), mgssec, NULL);
520 /* Re-using an existing MGC */
521 cfs_atomic_inc(&obd->u.cli.cl_mgc_refcount);
523 /* IR compatibility check, only for clients */
524 if (lmd_is_client(lsi->lsi_lmd)) {
526 int vallen = sizeof(*data);
527 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
529 rc = obd_get_info(NULL, obd->obd_self_export,
530 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
531 &vallen, data, NULL);
533 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
534 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
535 /* LMD_FLG_NOIR is for test purpose only */
537 "Trying to mount a client with IR setting "
538 "not compatible with current mgc. "
539 "Force to use current mgc setting that is "
541 has_ir ? "enabled" : "disabled");
543 *flags &= ~LMD_FLG_NOIR;
545 *flags |= LMD_FLG_NOIR;
550 /* If we are restarting the MGS, don't try to keep the MGC's
551 old connection, or registration will fail. */
553 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
557 /* Try all connections, but only once (again).
558 We don't want to block another target from starting
559 (using its local copy of the log), but we do want to connect
560 if at all possible. */
562 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
563 rc = obd_set_info_async(NULL, obd->obd_self_export,
564 sizeof(KEY_INIT_RECOV_BACKUP),
565 KEY_INIT_RECOV_BACKUP,
566 sizeof(recov_bk), &recov_bk, NULL);
570 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
572 /* Add the primary nids for the MGS */
574 sprintf(niduuid, "%s_%x", mgcname, i);
575 if (IS_SERVER(lsi)) {
576 ptr = lsi->lsi_lmd->lmd_mgs;
578 /* Use local nids (including LO) */
579 lnet_process_id_t id;
580 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
581 rc = do_lcfg(mgcname, id.nid,
582 LCFG_ADD_UUID, niduuid, 0,0,0);
585 /* Use mgsnode= nids */
586 /* mount -o mgsnode=nid */
587 if (lsi->lsi_lmd->lmd_mgs) {
588 ptr = lsi->lsi_lmd->lmd_mgs;
589 } else if (class_find_param(ptr, PARAM_MGSNODE,
591 CERROR("No MGS nids given.\n");
592 GOTO(out_free, rc = -EINVAL);
594 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
595 rc = do_lcfg(mgcname, nid,
596 LCFG_ADD_UUID, niduuid, 0,0,0);
600 } else { /* client */
601 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
602 ptr = lsi->lsi_lmd->lmd_dev;
603 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
604 rc = do_lcfg(mgcname, nid,
605 LCFG_ADD_UUID, niduuid, 0,0,0);
607 /* Stop at the first failover nid */
613 CERROR("No valid MGS nids found.\n");
614 GOTO(out_free, rc = -EINVAL);
616 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
618 /* Random uuid for MGC allows easier reconnects */
620 ll_generate_random_uuid(uuidc);
621 class_uuid_unparse(uuidc, uuid);
624 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
625 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
631 /* Add any failover MGS nids */
633 while (ptr && ((*ptr == ':' ||
634 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) {
635 /* New failover node */
636 sprintf(niduuid, "%s_%x", mgcname, i);
638 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
640 rc = do_lcfg(mgcname, nid,
641 LCFG_ADD_UUID, niduuid, 0,0,0);
646 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
654 lsi->lsi_lmd->lmd_mgs_failnodes = i;
656 obd = class_name2obd(mgcname);
658 CERROR("Can't find mgcobd %s\n", mgcname);
659 GOTO(out_free, rc = -ENOTCONN);
662 rc = obd_set_info_async(NULL, obd->obd_self_export,
663 strlen(KEY_MGSSEC), KEY_MGSSEC,
664 strlen(mgssec), mgssec, NULL);
668 /* Keep a refcount of servers/clients who started with "mount",
669 so we know when we can get rid of the mgc. */
670 cfs_atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
672 /* Try all connections, but only once. */
674 rc = obd_set_info_async(NULL, obd->obd_self_export,
675 sizeof(KEY_INIT_RECOV_BACKUP),
676 KEY_INIT_RECOV_BACKUP,
677 sizeof(recov_bk), &recov_bk, NULL);
680 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
682 /* We connect to the MGS at setup, and don't disconnect until cleanup */
683 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
684 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV;
686 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
687 data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB;
689 #warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
692 if (lmd_is_client(lsi->lsi_lmd) &&
693 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
694 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
695 data->ocd_version = LUSTRE_VERSION_CODE;
696 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
698 CERROR("connect failed %d\n", rc);
702 obd->u.cli.cl_mgc_mgsexp = exp;
705 /* Keep the mgc info in the sb. Note that many lsi's can point
709 cfs_mutex_unlock(&mgc_start_lock);
714 OBD_FREE(mgcname, len);
716 OBD_FREE(niduuid, len + 2);
720 static int lustre_stop_mgc(struct super_block *sb)
722 struct lustre_sb_info *lsi = s2lsi(sb);
723 struct obd_device *obd;
724 char *niduuid = 0, *ptr = 0;
725 int i, rc = 0, len = 0;
735 cfs_mutex_lock(&mgc_start_lock);
736 LASSERT(cfs_atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
737 if (!cfs_atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
738 /* This is not fatal, every client that stops
739 will call in here. */
740 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
741 cfs_atomic_read(&obd->u.cli.cl_mgc_refcount));
742 GOTO(out, rc = -EBUSY);
745 /* The MGC has no recoverable data in any case.
746 * force shotdown set in umount_begin */
747 obd->obd_no_recov = 1;
749 if (obd->u.cli.cl_mgc_mgsexp) {
750 /* An error is not fatal, if we are unable to send the
751 disconnect mgs ping evictor cleans up the export */
752 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
754 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
757 /* Save the obdname for cleaning the nid uuids, which are
759 len = strlen(obd->obd_name) + 6;
760 OBD_ALLOC(niduuid, len);
762 strcpy(niduuid, obd->obd_name);
763 ptr = niduuid + strlen(niduuid);
766 rc = class_manual_cleanup(obd);
770 /* Clean the nid uuids */
772 GOTO(out, rc = -ENOMEM);
774 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
775 sprintf(ptr, "_%x", i);
776 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
779 CERROR("del MDC UUID %s failed: rc = %d\n",
784 OBD_FREE(niduuid, len);
786 /* class_import_put will get rid of the additional connections */
787 cfs_mutex_unlock(&mgc_start_lock);
791 /* Since there's only one mgc per node, we have to change it's fs to get
792 access to the right disk. */
793 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
795 struct lustre_sb_info *lsi = s2lsi(sb);
799 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
801 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
802 rc = obd_set_info_async(NULL, mgc->obd_self_export,
803 sizeof(KEY_SET_FS), KEY_SET_FS,
804 sizeof(*sb), sb, NULL);
806 CERROR("can't set_fs %d\n", rc);
812 static int server_mgc_clear_fs(struct obd_device *mgc)
817 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
819 rc = obd_set_info_async(NULL, mgc->obd_self_export,
820 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
825 CFS_DEFINE_MUTEX(server_start_lock);
827 /* Stop MDS/OSS if nobody is using them */
828 static int server_stop_servers(int lsiflags)
830 struct obd_device *obd = NULL;
831 struct obd_type *type = NULL;
835 cfs_mutex_lock(&server_start_lock);
837 /* Either an MDT or an OST or neither */
838 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
839 if ((lsiflags & LDD_F_SV_TYPE_MDT) &&
840 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
841 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
842 type = class_search_type(LUSTRE_MDS_NAME);
844 /* if this was an OST, and there are no more OST's, clean up the OSS */
845 if ((lsiflags & LDD_F_SV_TYPE_OST) &&
846 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
847 type = class_search_type(LUSTRE_OST_NAME);
850 if (obd && (!type || !type->typ_refcnt)) {
853 /* obd_fail doesn't mean much on a server obd */
854 err = class_manual_cleanup(obd);
859 cfs_mutex_unlock(&server_start_lock);
864 int server_mti_print(char *title, struct mgs_target_info *mti)
866 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
867 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
868 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
869 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
870 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
871 mti->mti_config_ver, mti->mti_flags);
875 /** Get the fsname ("lustre") from the server name ("lustre-OST003F").
876 * @param [in] svname server name including type and index
877 * @param [out] fsname Buffer to copy filesystem name prefix into.
878 * Must have at least 'strlen(fsname) + 1' chars.
879 * @param [out] endptr if endptr isn't NULL it is set to end of fsname
882 static int server_name2fsname(char *svname, char *fsname, char **endptr)
886 p = strstr(svname, "-OST");
888 p = strstr(svname, "-MDT");
893 strncpy(fsname, svname, p - svname);
894 fsname[p - svname] = '\0';
904 * Get service name (svname) from string
906 * if endptr isn't NULL it is set to end of fsname *
908 int server_name2svname(char *label, char *svname, char **endptr)
913 /* We use server_name2fsname() just for parsing */
914 rc = server_name2fsname(label, NULL, &dash);
921 strncpy(svname, dash + 1, MTI_NAME_MAXLEN);
925 EXPORT_SYMBOL(server_name2svname);
928 /* Get the index from the obd name.
931 if endptr isn't NULL it is set to end of name */
932 int server_name2index(char *svname, __u32 *idx, char **endptr)
938 /* We use server_name2fsname() just for parsing */
939 rc = server_name2fsname(svname, NULL, &dash);
948 if (strncmp(dash, "MDT", 3) == 0)
949 rc = LDD_F_SV_TYPE_MDT;
950 else if (strncmp(dash, "OST", 3) == 0)
951 rc = LDD_F_SV_TYPE_OST;
957 if (strcmp(dash, "all") == 0)
958 return rc | LDD_F_SV_ALL;
960 index = simple_strtoul(dash, endptr, 16);
965 EXPORT_SYMBOL(server_name2index);
967 /* Generate data for registration */
968 static int server_lsi2mti(struct lustre_sb_info *lsi,
969 struct mgs_target_info *mti)
971 lnet_process_id_t id;
978 strncpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname));
980 mti->mti_nid_count = 0;
981 while (LNetGetId(i++, &id) != -ENOENT) {
982 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
985 /* server use --servicenode param, only allow specified
986 * nids be registered */
987 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
988 class_match_nid(lsi->lsi_lmd->lmd_params,
989 PARAM_FAILNODE, id.nid) < 1)
992 /* match specified network */
993 if (!class_match_net(lsi->lsi_lmd->lmd_params,
994 PARAM_NETWORK, LNET_NIDNET(id.nid)))
997 mti->mti_nids[mti->mti_nid_count] = id.nid;
998 mti->mti_nid_count++;
999 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1000 CWARN("Only using first %d nids for %s\n",
1001 mti->mti_nid_count, mti->mti_svname);
1006 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1007 mti->mti_config_ver = 0;
1009 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1013 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1016 /* Orion requires index to be set */
1017 LASSERT(!(rc & LDD_F_NEED_INDEX));
1018 /* keep only LDD flags */
1019 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1020 mti->mti_flags |= LDD_F_UPDATE;
1021 strncpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1022 sizeof(mti->mti_params));
1026 /* Register an old or new target with the MGS. If needed MGS will construct
1027 startup logs and assign index */
1028 static int server_register_target(struct lustre_sb_info *lsi)
1030 struct obd_device *mgc = lsi->lsi_mgc;
1031 struct mgs_target_info *mti = NULL;
1038 if (!IS_SERVER(lsi))
1045 rc = server_lsi2mti(lsi, mti);
1049 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1050 mti->mti_svname, mti->mti_fsname,
1051 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1054 /* if write_conf is true, the registration must succeed */
1055 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1056 mti->mti_flags |= LDD_F_OPC_REG;
1058 /* Register the target */
1059 /* FIXME use mgc_process_config instead */
1060 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1061 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
1062 sizeof(*mti), mti, NULL);
1064 if (mti->mti_flags & LDD_F_ERROR) {
1065 LCONSOLE_ERROR_MSG(0x160,
1066 "The MGS is refusing to allow this "
1067 "server (%s) to start. Please see messages"
1068 " on the MGS node.\n", lsi->lsi_svname);
1069 } else if (writeconf) {
1070 LCONSOLE_ERROR_MSG(0x15f,
1071 "Communication to the MGS return error %d. "
1072 "Is the MGS running?\n", rc);
1074 CERROR("Cannot talk to the MGS: %d, not fatal\n", rc);
1075 /* reset the error code for non-fatal error. */
1088 * Notify the MGS that this target is ready.
1089 * Used by IR - if the MGS receives this message, it will notify clients.
1091 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1093 struct lustre_sb_info *lsi = s2lsi(sb);
1094 struct obd_device *mgc = lsi->lsi_mgc;
1095 struct mgs_target_info *mti = NULL;
1101 if (!(IS_SERVER(lsi)))
1107 rc = server_lsi2mti(lsi, mti);
1111 mti->mti_instance = obd->u.obt.obt_instance;
1112 mti->mti_flags |= LDD_F_OPC_READY;
1114 /* FIXME use mgc_process_config instead */
1115 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1116 sizeof(KEY_REGISTER_TARGET),
1117 KEY_REGISTER_TARGET,
1118 sizeof(*mti), mti, NULL);
1120 /* Imperative recovery: if the mgs informs us to use IR? */
1121 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1122 (mti->mti_flags & LDD_F_IR_CAPABLE))
1123 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1132 /** Start server targets: MDTs and OSTs
1134 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1136 struct obd_device *obd;
1137 struct lustre_sb_info *lsi = s2lsi(sb);
1138 struct config_llog_instance cfg;
1140 struct lu_device *dev;
1144 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1147 /* If we're an MDT, make sure the global MDS is running */
1148 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1149 /* make sure the MDS is started */
1150 cfs_mutex_lock(&server_start_lock);
1151 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1153 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1154 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1156 LUSTRE_MDS_OBDNAME"_uuid",
1159 cfs_mutex_unlock(&server_start_lock);
1160 CERROR("failed to start MDS: %d\n", rc);
1164 cfs_mutex_unlock(&server_start_lock);
1168 /* If we're an OST, make sure the global OSS is running */
1170 /* make sure OSS is started */
1171 cfs_mutex_lock(&server_start_lock);
1172 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1174 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1176 LUSTRE_OSS_OBDNAME"_uuid",
1179 cfs_mutex_unlock(&server_start_lock);
1180 CERROR("failed to start OSS: %d\n", rc);
1184 cfs_mutex_unlock(&server_start_lock);
1187 /* Set the mgc fs to our server disk. This allows the MGC to
1188 * read and write configs locally, in case it can't talk to the MGS. */
1189 if (lsi->lsi_srv_mnt) {
1190 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1195 /* Register with MGS */
1196 rc = server_register_target(lsi);
1200 /* Let the target look up the mount using the target's name
1201 (we can't pass the sb or mnt through class_process_config.) */
1202 rc = server_register_mount(lsi->lsi_svname, sb, mnt);
1206 /* Start targets using the llog named for the target */
1207 memset(&cfg, 0, sizeof(cfg));
1208 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1210 CERROR("failed to start server %s: %d\n",
1211 lsi->lsi_svname, rc);
1212 /* Do NOT call server_deregister_mount() here. This makes it
1213 * impossible to find mount later in cleanup time and leaves
1214 * @lsi and othder stuff leaked. -umka */
1219 /* Release the mgc fs for others to use */
1220 if (lsi->lsi_srv_mnt)
1221 server_mgc_clear_fs(lsi->lsi_mgc);
1224 obd = class_name2obd(lsi->lsi_svname);
1226 CERROR("no server named %s was started\n",
1231 server_notify_target(sb, obd);
1233 /* calculate recovery timeout, do it after lustre_process_log */
1234 server_calc_timeout(lsi, obd);
1236 /* log has been fully processed */
1237 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1239 /* log has been fully processed, let clients connect */
1240 dev = obd->obd_lu_dev;
1241 if (dev && dev->ld_ops->ldo_prepare) {
1242 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1244 struct lu_context session_ctx;
1246 lu_context_init(&session_ctx, LCT_SESSION);
1247 session_ctx.lc_thread = NULL;
1248 lu_context_enter(&session_ctx);
1249 env.le_ses = &session_ctx;
1251 dev->ld_ops->ldo_prepare(&env, NULL, dev);
1254 lu_context_exit(&session_ctx);
1255 lu_context_fini(&session_ctx);
1259 /* abort recovery only on the complete stack:
1260 * many devices can be involved */
1261 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1262 (OBP(obd, iocontrol))) {
1263 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1264 obd->obd_self_export, 0, NULL, NULL);
1271 /***************** lustre superblock **************/
1273 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1275 struct lustre_sb_info *lsi;
1281 OBD_ALLOC_PTR(lsi->lsi_lmd);
1282 if (!lsi->lsi_lmd) {
1287 lsi->lsi_lmd->lmd_exclude_count = 0;
1288 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
1289 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
1290 s2lsi_nocast(sb) = lsi;
1291 /* we take 1 extra ref for our setup */
1292 cfs_atomic_set(&lsi->lsi_mounts, 1);
1294 /* Default umount style */
1295 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1300 static int lustre_free_lsi(struct super_block *sb)
1302 struct lustre_sb_info *lsi = s2lsi(sb);
1305 LASSERT(lsi != NULL);
1306 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
1308 /* someone didn't call server_put_mount. */
1309 LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0);
1311 if (lsi->lsi_lmd != NULL) {
1312 if (lsi->lsi_lmd->lmd_dev != NULL)
1313 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1314 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1315 if (lsi->lsi_lmd->lmd_profile != NULL)
1316 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1317 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1318 if (lsi->lsi_lmd->lmd_mgssec != NULL)
1319 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
1320 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
1321 if (lsi->lsi_lmd->lmd_opts != NULL)
1322 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1323 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1324 if (lsi->lsi_lmd->lmd_exclude_count)
1325 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1326 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1327 lsi->lsi_lmd->lmd_exclude_count);
1328 if (lsi->lsi_lmd->lmd_mgs != NULL)
1329 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
1330 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
1331 if (lsi->lsi_lmd->lmd_osd_type != NULL)
1332 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
1333 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
1334 if (lsi->lsi_lmd->lmd_params != NULL)
1335 OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
1337 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1340 LASSERT(lsi->lsi_llsbi == NULL);
1341 OBD_FREE(lsi, sizeof(*lsi));
1342 s2lsi_nocast(sb) = NULL;
1347 /* The lsi has one reference for every server that is using the disk -
1348 e.g. MDT, MGS, and potentially MGC */
1349 static int lustre_put_lsi(struct super_block *sb)
1351 struct lustre_sb_info *lsi = s2lsi(sb);
1354 LASSERT(lsi != NULL);
1356 CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts));
1357 if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) {
1358 if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
1359 obd_disconnect(lsi->lsi_osd_exp);
1360 /* wait till OSD is gone */
1361 obd_zombie_barrier();
1363 lustre_free_lsi(sb);
1369 static int lsi_prepare(struct lustre_sb_info *lsi)
1376 LASSERT(lsi->lsi_lmd);
1378 /* The server name is given as a mount line option */
1379 if (lsi->lsi_lmd->lmd_profile == NULL) {
1380 LCONSOLE_ERROR("Can't determine server name\n");
1384 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
1385 RETURN(-ENAMETOOLONG);
1387 strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
1389 /* Determine osd type */
1390 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
1391 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
1392 sizeof(lsi->lsi_osd_type))
1393 RETURN(-ENAMETOOLONG);
1395 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
1397 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
1400 /* XXX: a temp. solution for components using fsfilt
1401 * to be removed in one of the subsequent patches */
1402 if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs")) {
1403 strcpy(lsi->lsi_fstype, "ldiskfs");
1405 strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
1408 /* Determine server type */
1409 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1411 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
1412 /* Assume we're a bare MGS */
1414 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
1416 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
1421 lsi->lsi_flags |= rc;
1423 /* Add mount line flags that used to be in ldd:
1424 * writeconf, mgs, iam, anything else?
1426 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
1427 LDD_F_WRITECONF : 0;
1428 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
1430 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
1431 LDD_F_SV_TYPE_MGS : 0;
1432 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_IAM) ?
1434 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
1435 LDD_F_NO_PRIMNODE : 0;
1440 /*************** server mount ******************/
1442 /** Start the shutdown of servers at umount.
1444 static void server_put_super(struct super_block *sb)
1446 struct lustre_sb_info *lsi = s2lsi(sb);
1447 struct obd_device *obd;
1448 char *tmpname, *extraname = NULL;
1450 int lsiflags = lsi->lsi_flags;
1453 LASSERT(IS_SERVER(lsi));
1455 tmpname_sz = strlen(lsi->lsi_svname) + 1;
1456 OBD_ALLOC(tmpname, tmpname_sz);
1457 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
1458 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1459 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1460 snprintf(tmpname, tmpname_sz, "MGS");
1462 /* Stop the target */
1463 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1464 (IS_MDT(lsi) || IS_OST(lsi))) {
1465 struct lustre_profile *lprof = NULL;
1467 /* tell the mgc to drop the config log */
1468 lustre_end_log(sb, lsi->lsi_svname, NULL);
1470 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1471 If there are any setup/cleanup errors, save the lov
1472 name for safety cleanup later. */
1473 lprof = class_get_profile(lsi->lsi_svname);
1474 if (lprof && lprof->lp_dt) {
1475 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1476 strcpy(extraname, lprof->lp_dt);
1479 obd = class_name2obd(lsi->lsi_svname);
1481 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1482 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1484 /* We can't seem to give an error return code
1485 * to .put_super, so we better make sure we clean up! */
1487 class_manual_cleanup(obd);
1489 CERROR("no obd %s\n", lsi->lsi_svname);
1490 server_deregister_mount(lsi->lsi_svname);
1494 /* If they wanted the mgs to stop separately from the mdt, they
1495 should have put it on a different device. */
1497 /* if MDS start with --nomgs, don't stop MGS then */
1498 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1499 server_stop_mgs(sb);
1502 /* Clean the mgc and sb */
1503 lustre_common_put_super(sb);
1505 /* wait till all in-progress cleanups are done
1506 * specifically we're interested in ofd cleanup
1508 obd_zombie_barrier();
1510 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1511 until the target is really gone so that our type refcount check
1513 server_stop_servers(lsiflags);
1515 /* In case of startup or cleanup err, stop related obds */
1517 obd = class_name2obd(extraname);
1519 CWARN("Cleaning orphaned obd %s\n", extraname);
1521 class_manual_cleanup(obd);
1523 OBD_FREE(extraname, strlen(extraname) + 1);
1526 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1527 OBD_FREE(tmpname, tmpname_sz);
1531 /** Called only for 'umount -f'
1533 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1534 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1536 struct super_block *sb = vfsmnt->mnt_sb;
1538 static void server_umount_begin(struct super_block *sb)
1541 struct lustre_sb_info *lsi = s2lsi(sb);
1544 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1545 if (!(flags & MNT_FORCE)) {
1551 CDEBUG(D_MOUNT, "umount -f\n");
1552 /* umount = failover
1554 no third way to do non-force, non-failover */
1555 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1559 static int server_statfs (struct dentry *dentry, cfs_kstatfs_t *buf)
1561 struct super_block *sb = dentry->d_sb;
1562 struct lustre_sb_info *lsi = s2lsi(sb);
1563 struct obd_statfs statfs;
1567 if (lsi->lsi_dt_dev) {
1568 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
1570 statfs_unpack(buf, &statfs);
1571 buf->f_type = sb->s_magic;
1577 buf->f_type = sb->s_magic;
1578 buf->f_bsize = sb->s_blocksize;
1584 buf->f_namelen = NAME_MAX;
1588 /** The operations we support directly on the superblock:
1589 * mount, umount, and df.
1591 static struct super_operations server_ops =
1593 .put_super = server_put_super,
1594 .umount_begin = server_umount_begin, /* umount -f */
1595 .statfs = server_statfs,
1598 #define log2(n) cfs_ffz(~(n))
1599 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1601 static int server_fill_super_common(struct super_block *sb)
1603 struct inode *root = 0;
1606 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1608 sb->s_blocksize = 4096;
1609 sb->s_blocksize_bits = log2(sb->s_blocksize);
1610 sb->s_magic = LUSTRE_SUPER_MAGIC;
1611 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1612 sb->s_flags |= MS_RDONLY;
1613 sb->s_op = &server_ops;
1615 root = new_inode(sb);
1617 CERROR("Can't make root inode\n");
1621 /* returns -EIO for every operation */
1622 /* make_bad_inode(root); -- badness - can't umount */
1623 /* apparently we need to be a directory for the mount to finish */
1624 root->i_mode = S_IFDIR;
1626 sb->s_root = d_alloc_root(root);
1628 CERROR("Can't make root dentry\n");
1636 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
1638 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1639 struct obd_device *obd;
1640 struct dt_device_param p;
1646 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
1647 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
1649 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
1650 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
1651 strcat(lsi->lsi_osd_uuid, "_UUID");
1652 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
1654 obd = class_name2obd(lsi->lsi_osd_obdname);
1656 rc = lustre_start_simple(lsi->lsi_osd_obdname,
1658 lsi->lsi_osd_uuid, lmd->lmd_dev,
1659 flagstr, lsi->lsi_lmd->lmd_opts,
1663 obd = class_name2obd(lsi->lsi_osd_obdname);
1667 rc = obd_connect(NULL, &lsi->lsi_osd_exp, obd, &obd->obd_uuid, NULL, NULL);
1670 class_manual_cleanup(obd);
1671 lsi->lsi_dt_dev = NULL;
1674 /* XXX: to keep support old components relying on lsi_srv_mnt
1675 * we get this info from OSD just started */
1676 LASSERT(obd->obd_lu_dev);
1677 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
1678 LASSERT(lsi->lsi_dt_dev);
1680 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
1682 lsi->lsi_srv_mnt = p.ddp_mnt;
1688 /** Fill in the superblock info for a Lustre server.
1689 * Mount the device with the correct options.
1690 * Read the on-disk config file.
1691 * Start the services.
1693 static int server_fill_super(struct super_block *sb)
1695 struct lustre_sb_info *lsi = s2lsi(sb);
1699 rc = lsi_prepare(lsi);
1703 /* Start low level OSD */
1704 rc = osd_start(lsi, sb->s_flags);
1706 CERROR("Unable to start osd on %s: %d\n",
1707 lsi->lsi_lmd->lmd_dev, rc);
1712 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
1713 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
1715 if (class_name2obd(lsi->lsi_svname)) {
1716 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1717 "running. Double-mount may have compromised"
1718 " the disk journal.\n",
1724 /* Start MGS before MGC */
1725 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
1726 rc = server_start_mgs(sb);
1731 /* Start MGC before servers */
1732 rc = lustre_start_mgc(sb);
1736 /* Set up all obd devices for service */
1737 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1738 (IS_OST(lsi) || IS_MDT(lsi))) {
1739 rc = server_start_targets(sb, lsi->lsi_srv_mnt);
1741 CERROR("Unable to start targets: %d\n", rc);
1744 /* FIXME overmount client here,
1745 or can we just start a client log and client_fill_super on this sb?
1746 We need to make sure server_put_super gets called too - ll_put_super
1747 calls lustre_common_put_super; check there for LSI_SERVER flag,
1749 Probably should start client from new thread so we can return.
1750 Client will not finish until all servers are connected.
1751 Note - MGS-only server does NOT get a client, since there is no
1752 lustre fs associated - the MGS is for all lustre fs's */
1755 rc = server_fill_super_common(sb);
1761 /* We jump here in case of failure while starting targets or MGS.
1762 * In this case we can't just put @mnt and have to do real cleanup
1763 * with stoping targets, etc. */
1764 server_put_super(sb);
1769 * Calculate timeout value for a target.
1771 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1773 struct lustre_mount_data *lmd;
1777 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
1778 int min = OBD_RECOVERY_TIME_MIN;
1780 LASSERT(IS_SERVER(lsi));
1784 soft = lmd->lmd_recovery_time_soft;
1785 hard = lmd->lmd_recovery_time_hard;
1786 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1787 obd->obd_no_ir = !has_ir;
1791 soft = OBD_RECOVERY_TIME_SOFT;
1793 hard = OBD_RECOVERY_TIME_HARD;
1795 /* target may have ir_factor configured. */
1796 factor = OBD_IR_FACTOR_DEFAULT;
1797 if (obd->obd_recovery_ir_factor)
1798 factor = obd->obd_recovery_ir_factor;
1801 int new_soft = soft;
1802 int new_hard = hard;
1804 /* adjust timeout value by imperative recovery */
1806 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1807 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1809 /* make sure the timeout is not too short */
1810 new_soft = max(min, new_soft);
1811 new_hard = max(new_soft, new_hard);
1813 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1814 "window shrunk from %d-%d down to %d-%d\n",
1815 obd->obd_name, soft, hard, new_soft, new_hard);
1822 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1823 obd->obd_recovery_time_hard = hard;
1824 obd->obd_recovery_ir_factor = factor;
1826 EXPORT_SYMBOL(server_calc_timeout);
1828 /*************** mount common betweeen server and client ***************/
1831 int lustre_common_put_super(struct super_block *sb)
1836 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1838 /* Drop a ref to the MGC */
1839 rc = lustre_stop_mgc(sb);
1840 if (rc && (rc != -ENOENT)) {
1842 CERROR("Can't stop MGC: %d\n", rc);
1845 /* BUSY just means that there's some other obd that
1846 needs the mgc. Let him clean it up. */
1847 CDEBUG(D_MOUNT, "MGC still in use\n");
1849 /* Drop a ref to the mounted disk */
1854 EXPORT_SYMBOL(lustre_common_put_super);
1856 static void lmd_print(struct lustre_mount_data *lmd)
1860 PRINT_CMD(PRINT_MASK, " mount data:\n");
1861 if (lmd_is_client(lmd))
1862 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
1863 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
1864 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
1867 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
1869 if (lmd->lmd_recovery_time_soft)
1870 PRINT_CMD(PRINT_MASK, "recovery time soft: %d\n",
1871 lmd->lmd_recovery_time_soft);
1873 if (lmd->lmd_recovery_time_hard)
1874 PRINT_CMD(PRINT_MASK, "recovery time hard: %d\n",
1875 lmd->lmd_recovery_time_hard);
1877 for (i = 0; i < lmd->lmd_exclude_count; i++) {
1878 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
1879 lmd->lmd_exclude[i]);
1883 /* Is this server on the exclusion list */
1884 int lustre_check_exclusion(struct super_block *sb, char *svname)
1886 struct lustre_sb_info *lsi = s2lsi(sb);
1887 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1892 rc = server_name2index(svname, &index, NULL);
1893 if (rc != LDD_F_SV_TYPE_OST)
1894 /* Only exclude OSTs */
1897 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
1898 index, lmd->lmd_exclude_count, lmd->lmd_dev);
1900 for(i = 0; i < lmd->lmd_exclude_count; i++) {
1901 if (index == lmd->lmd_exclude[i]) {
1902 CWARN("Excluding %s (on exclusion list)\n", svname);
1909 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
1910 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
1912 char *s1 = ptr, *s2;
1913 __u32 index, *exclude_list;
1917 /* The shortest an ost name can be is 8 chars: -OST0000.
1918 We don't actually know the fsname at this time, so in fact
1919 a user could specify any fsname. */
1920 devmax = strlen(ptr) / 8 + 1;
1922 /* temp storage until we figure out how many we have */
1923 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
1927 /* we enter this fn pointing at the '=' */
1928 while (*s1 && *s1 != ' ' && *s1 != ',') {
1930 rc = server_name2index(s1, &index, &s2);
1932 CERROR("Can't parse server name '%s'\n", s1);
1935 if (rc == LDD_F_SV_TYPE_OST)
1936 exclude_list[lmd->lmd_exclude_count++] = index;
1938 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
1940 /* now we are pointing at ':' (next exclude)
1941 or ',' (end of excludes) */
1942 if (lmd->lmd_exclude_count >= devmax)
1945 if (rc >= 0) /* non-err */
1948 if (lmd->lmd_exclude_count) {
1949 /* permanent, freed in lustre_free_lsi */
1950 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
1951 lmd->lmd_exclude_count);
1952 if (lmd->lmd_exclude) {
1953 memcpy(lmd->lmd_exclude, exclude_list,
1954 sizeof(index) * lmd->lmd_exclude_count);
1957 lmd->lmd_exclude_count = 0;
1960 OBD_FREE(exclude_list, sizeof(index) * devmax);
1964 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
1969 if (lmd->lmd_mgssec != NULL) {
1970 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
1971 lmd->lmd_mgssec = NULL;
1974 tail = strchr(ptr, ',');
1976 length = strlen(ptr);
1978 length = tail - ptr;
1980 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
1981 if (lmd->lmd_mgssec == NULL)
1984 memcpy(lmd->lmd_mgssec, ptr, length);
1985 lmd->lmd_mgssec[length] = '\0';
1989 static int lmd_parse_string(char **handle, char *ptr)
1994 if ((handle == NULL) || (ptr == NULL))
1997 if (*handle != NULL) {
1998 OBD_FREE(*handle, strlen(*handle) + 1);
2002 tail = strchr(ptr, ',');
2004 length = strlen(ptr);
2006 length = tail - ptr;
2008 OBD_ALLOC(*handle, length + 1);
2009 if (*handle == NULL)
2012 memcpy(*handle, ptr, length);
2013 (*handle)[length] = '\0';
2018 /* Collect multiple values for mgsnid specifiers */
2019 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
2027 /* Find end of nidlist */
2028 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
2029 length = tail - *ptr;
2031 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
2035 if (lmd->lmd_mgs != NULL)
2036 oldlen = strlen(lmd->lmd_mgs) + 1;
2038 OBD_ALLOC(mgsnid, oldlen + length + 1);
2042 if (lmd->lmd_mgs != NULL) {
2043 /* Multiple mgsnid= are taken to mean failover locations */
2044 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
2045 mgsnid[oldlen - 1] = ':';
2046 OBD_FREE(lmd->lmd_mgs, oldlen);
2048 memcpy(mgsnid + oldlen, *ptr, length);
2049 mgsnid[oldlen + length] = '\0';
2050 lmd->lmd_mgs = mgsnid;
2056 /** Parse mount line options
2057 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
2058 * dev is passed as device=uml1:/lustre by mount.lustre
2060 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
2062 char *s1, *s2, *devname = NULL;
2063 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
2069 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
2070 "/sbin/mount.lustre is installed.\n");
2074 /* Options should be a string - try to detect old lmd data */
2075 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
2076 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
2077 "/sbin/mount.lustre. Please install "
2078 "version %s\n", LUSTRE_VERSION_STRING);
2081 lmd->lmd_magic = LMD_MAGIC;
2083 OBD_ALLOC(lmd->lmd_params, 4096);
2084 if (lmd->lmd_params == NULL)
2086 lmd->lmd_params[0] = '\0';
2088 /* Set default flags here */
2093 int time_min = OBD_RECOVERY_TIME_MIN;
2095 /* Skip whitespace and extra commas */
2096 while (*s1 == ' ' || *s1 == ',')
2099 /* Client options are parsed in ll_options: eg. flock,
2102 /* Parse non-ldiskfs options here. Rather than modifying
2103 ldiskfs, we just zero these out here */
2104 if (strncmp(s1, "abort_recov", 11) == 0) {
2105 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
2107 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
2108 lmd->lmd_recovery_time_soft = max_t(int,
2109 simple_strtoul(s1 + 19, NULL, 10), time_min);
2111 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
2112 lmd->lmd_recovery_time_hard = max_t(int,
2113 simple_strtoul(s1 + 19, NULL, 10), time_min);
2115 } else if (strncmp(s1, "noir", 4) == 0) {
2116 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
2118 } else if (strncmp(s1, "nosvc", 5) == 0) {
2119 lmd->lmd_flags |= LMD_FLG_NOSVC;
2121 } else if (strncmp(s1, "nomgs", 5) == 0) {
2122 lmd->lmd_flags |= LMD_FLG_NOMGS;
2124 } else if (strncmp(s1, "noscrub", 7) == 0) {
2125 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
2127 } else if (strncmp(s1, PARAM_MGSNODE,
2128 sizeof(PARAM_MGSNODE) - 1) == 0) {
2129 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
2130 /* Assume the next mount opt is the first
2131 invalid nid we get to. */
2132 rc = lmd_parse_mgs(lmd, &s2);
2136 } else if (strncmp(s1, "writeconf", 9) == 0) {
2137 lmd->lmd_flags |= LMD_FLG_WRITECONF;
2139 } else if (strncmp(s1, "virgin", 6) == 0) {
2140 lmd->lmd_flags |= LMD_FLG_VIRGIN;
2142 } else if (strncmp(s1, "noprimnode", 10) == 0) {
2143 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
2145 } else if (strncmp(s1, "mgssec=", 7) == 0) {
2146 rc = lmd_parse_mgssec(lmd, s1 + 7);
2150 /* ost exclusion list */
2151 } else if (strncmp(s1, "exclude=", 8) == 0) {
2152 rc = lmd_make_exclusion(lmd, s1 + 7);
2156 } else if (strncmp(s1, "mgs", 3) == 0) {
2158 lmd->lmd_flags |= LMD_FLG_MGS;
2160 } else if (strncmp(s1, "svname=", 7) == 0) {
2161 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
2165 } else if (strncmp(s1, "param=", 6) == 0) {
2167 char *tail = strchr(s1 + 6, ',');
2169 length = strlen(s1);
2173 strncat(lmd->lmd_params, s1 + 6, length);
2174 strcat(lmd->lmd_params, " ");
2176 } else if (strncmp(s1, "osd=", 4) == 0) {
2177 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
2182 /* Linux 2.4 doesn't pass the device, so we stuck it at the
2183 end of the options. */
2184 else if (strncmp(s1, "device=", 7) == 0) {
2186 /* terminate options right before device. device
2187 must be the last one. */
2193 s2 = strchr(s1, ',');
2201 memmove(s1, s2, strlen(s2) + 1);
2207 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
2208 "(need mount option 'device=...')\n");
2212 s1 = strstr(devname, ":/");
2215 lmd->lmd_flags |= LMD_FLG_CLIENT;
2216 /* Remove leading /s from fsname */
2217 while (*++s1 == '/') ;
2218 /* Freed in lustre_free_lsi */
2219 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
2220 if (!lmd->lmd_profile)
2222 sprintf(lmd->lmd_profile, "%s-client", s1);
2225 /* Freed in lustre_free_lsi */
2226 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
2229 strcpy(lmd->lmd_dev, devname);
2231 /* Save mount options */
2232 s1 = options + strlen(options) - 1;
2233 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
2235 if (*options != 0) {
2236 /* Freed in lustre_free_lsi */
2237 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
2240 strcpy(lmd->lmd_opts, options);
2244 lmd->lmd_magic = LMD_MAGIC;
2249 CERROR("Bad mount options %s\n", options);
2253 struct lustre_mount_data2 {
2255 struct vfsmount *lmd2_mnt;
2258 /** This is the entry point for the mount call into Lustre.
2259 * This is called when a server or client is mounted,
2260 * and this is where we start setting things up.
2261 * @param data Mount options (e.g. -o flock,abort_recov)
2263 int lustre_fill_super(struct super_block *sb, void *data, int silent)
2265 struct lustre_mount_data *lmd;
2266 struct lustre_mount_data2 *lmd2 = data;
2267 struct lustre_sb_info *lsi;
2271 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
2273 lsi = lustre_init_lsi(sb);
2279 * Disable lockdep during mount, because mount locking patterns are
2285 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
2287 obd_zombie_barrier();
2289 /* Figure out the lmd from the mount options */
2290 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
2292 GOTO(out, rc = -EINVAL);
2295 if (lmd_is_client(lmd)) {
2296 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
2297 if (!client_fill_super) {
2298 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
2299 "client mount! Is the 'lustre' "
2300 "module loaded?\n");
2304 rc = lustre_start_mgc(sb);
2309 /* Connect and start */
2310 /* (should always be ll_fill_super) */
2311 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
2312 /* c_f_s will call lustre_common_put_super on failure */
2315 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
2316 rc = server_fill_super(sb);
2317 /* s_f_s calls lustre_start_mgc after the mount because we need
2318 the MGS nids which are stored on disk. Plus, we may
2319 need to start the MGS first. */
2320 /* s_f_s will call server_put_super on failure */
2323 /* If error happens in fill_super() call, @lsi will be killed there.
2324 * This is why we do not put it here. */
2328 CERROR("Unable to mount %s (%d)\n",
2329 s2lsi(sb) ? lmd->lmd_dev : "", rc);
2331 CDEBUG(D_SUPER, "Mount %s complete\n",
2339 /* We can't call ll_fill_super by name because it lives in a module that
2340 must be loaded after this one. */
2341 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
2342 struct vfsmount *mnt))
2344 client_fill_super = cfs;
2346 EXPORT_SYMBOL(lustre_register_client_fill_super);
2348 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
2350 kill_super_cb = cfs;
2352 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2354 /***************** FS registration ******************/
2355 #ifdef HAVE_FSTYPE_MOUNT
2356 struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
2357 const char *devname, void *data)
2359 struct lustre_mount_data2 lmd2 = { data, NULL };
2361 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
2364 int lustre_get_sb(struct file_system_type *fs_type, int flags,
2365 const char *devname, void * data, struct vfsmount *mnt)
2367 struct lustre_mount_data2 lmd2 = { data, mnt };
2369 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
2373 void lustre_kill_super(struct super_block *sb)
2375 struct lustre_sb_info *lsi = s2lsi(sb);
2377 if (kill_super_cb && lsi && !IS_SERVER(lsi))
2378 (*kill_super_cb)(sb);
2380 kill_anon_super(sb);
2383 /** Register the "lustre" fs type
2385 struct file_system_type lustre_fs_type = {
2386 .owner = THIS_MODULE,
2388 #ifdef HAVE_FSTYPE_MOUNT
2389 .mount = lustre_mount,
2391 .get_sb = lustre_get_sb,
2393 .kill_sb = lustre_kill_super,
2394 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
2395 #ifdef FS_HAS_FIEMAP
2398 LL_RENAME_DOES_D_MOVE,
2401 int lustre_register_fs(void)
2403 return register_filesystem(&lustre_fs_type);
2406 int lustre_unregister_fs(void)
2408 return unregister_filesystem(&lustre_fs_type);
2411 EXPORT_SYMBOL(server_mti_print);