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;
394 /* It is impossible to have more than 1 MGS per node, since
395 MGC wouldn't know which to connect to */
396 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
398 lsi = s2lsi(lmi->lmi_sb);
399 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
404 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
406 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
409 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
410 LUSTRE_MGS_OBDNAME, 0, 0,
411 lsi->lsi_osd_obdname, 0);
412 /* Do NOT call server_deregister_mount() here. This leads to
413 * inability cleanup cleanly and free lsi and other stuff when
414 * mgs calls server_put_mount() in error handling case. -umka */
418 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
419 "Is the 'mgs' module loaded?\n",
420 LUSTRE_MGS_OBDNAME, rc);
424 static int server_stop_mgs(struct super_block *sb)
426 struct obd_device *obd;
430 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
432 /* There better be only one MGS */
433 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
435 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
439 /* The MGS should always stop when we say so */
441 rc = class_manual_cleanup(obd);
445 CFS_DEFINE_MUTEX(mgc_start_lock);
447 /** Set up a mgc obd to process startup logs
449 * \param sb [in] super block of the mgc obd
451 * \retval 0 success, otherwise error code
453 static int lustre_start_mgc(struct super_block *sb)
455 struct obd_connect_data *data = NULL;
456 struct lustre_sb_info *lsi = s2lsi(sb);
457 struct obd_device *obd;
458 struct obd_export *exp;
459 struct obd_uuid *uuid;
462 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
465 int rc = 0, i = 0, j, len;
468 LASSERT(lsi->lsi_lmd);
470 /* Find the first non-lo MGS nid for our MGC name */
471 if (IS_SERVER(lsi)) {
472 /* mount -o mgsnode=nid */
473 ptr = lsi->lsi_lmd->lmd_mgs;
474 if (lsi->lsi_lmd->lmd_mgs &&
475 (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) {
477 } else if (IS_MGS(lsi)) {
478 lnet_process_id_t id;
479 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
480 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
487 } else { /* client */
488 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
489 ptr = lsi->lsi_lmd->lmd_dev;
490 if (class_parse_nid(ptr, &nid, &ptr) == 0)
494 CERROR("No valid MGS nids found.\n");
498 cfs_mutex_lock(&mgc_start_lock);
500 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
501 OBD_ALLOC(mgcname, len);
502 OBD_ALLOC(niduuid, len + 2);
503 if (!mgcname || !niduuid)
504 GOTO(out_free, rc = -ENOMEM);
505 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
507 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
511 GOTO(out_free, rc = -ENOMEM);
513 obd = class_name2obd(mgcname);
514 if (obd && !obd->obd_stopping) {
515 rc = obd_set_info_async(NULL, obd->obd_self_export,
516 strlen(KEY_MGSSEC), KEY_MGSSEC,
517 strlen(mgssec), mgssec, NULL);
521 /* Re-using an existing MGC */
522 cfs_atomic_inc(&obd->u.cli.cl_mgc_refcount);
524 /* IR compatibility check, only for clients */
525 if (lmd_is_client(lsi->lsi_lmd)) {
527 int vallen = sizeof(*data);
528 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
530 rc = obd_get_info(NULL, obd->obd_self_export,
531 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
532 &vallen, data, NULL);
534 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
535 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
536 /* LMD_FLG_NOIR is for test purpose only */
538 "Trying to mount a client with IR setting "
539 "not compatible with current mgc. "
540 "Force to use current mgc setting that is "
542 has_ir ? "enabled" : "disabled");
544 *flags &= ~LMD_FLG_NOIR;
546 *flags |= LMD_FLG_NOIR;
551 /* If we are restarting the MGS, don't try to keep the MGC's
552 old connection, or registration will fail. */
554 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
558 /* Try all connections, but only once (again).
559 We don't want to block another target from starting
560 (using its local copy of the log), but we do want to connect
561 if at all possible. */
563 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
564 rc = obd_set_info_async(NULL, obd->obd_self_export,
565 sizeof(KEY_INIT_RECOV_BACKUP),
566 KEY_INIT_RECOV_BACKUP,
567 sizeof(recov_bk), &recov_bk, NULL);
571 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
573 /* Add the primary nids for the MGS */
575 sprintf(niduuid, "%s_%x", mgcname, i);
576 if (IS_SERVER(lsi)) {
577 ptr = lsi->lsi_lmd->lmd_mgs;
579 /* Use local nids (including LO) */
580 lnet_process_id_t id;
581 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
582 rc = do_lcfg(mgcname, id.nid,
583 LCFG_ADD_UUID, niduuid, 0,0,0);
586 /* Use mgsnode= nids */
587 /* mount -o mgsnode=nid */
588 if (lsi->lsi_lmd->lmd_mgs) {
589 ptr = lsi->lsi_lmd->lmd_mgs;
590 } else if (class_find_param(ptr, PARAM_MGSNODE,
592 CERROR("No MGS nids given.\n");
593 GOTO(out_free, rc = -EINVAL);
595 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
596 rc = do_lcfg(mgcname, nid,
597 LCFG_ADD_UUID, niduuid, 0,0,0);
601 } else { /* client */
602 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
603 ptr = lsi->lsi_lmd->lmd_dev;
604 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
605 rc = do_lcfg(mgcname, nid,
606 LCFG_ADD_UUID, niduuid, 0,0,0);
608 /* Stop at the first failover nid */
614 CERROR("No valid MGS nids found.\n");
615 GOTO(out_free, rc = -EINVAL);
617 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
619 /* Random uuid for MGC allows easier reconnects */
621 ll_generate_random_uuid(uuidc);
622 class_uuid_unparse(uuidc, uuid);
625 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
626 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
632 /* Add any failover MGS nids */
634 while (ptr && ((*ptr == ':' ||
635 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) {
636 /* New failover node */
637 sprintf(niduuid, "%s_%x", mgcname, i);
639 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
641 rc = do_lcfg(mgcname, nid,
642 LCFG_ADD_UUID, niduuid, 0,0,0);
647 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
655 lsi->lsi_lmd->lmd_mgs_failnodes = i;
657 obd = class_name2obd(mgcname);
659 CERROR("Can't find mgcobd %s\n", mgcname);
660 GOTO(out_free, rc = -ENOTCONN);
663 rc = obd_set_info_async(NULL, obd->obd_self_export,
664 strlen(KEY_MGSSEC), KEY_MGSSEC,
665 strlen(mgssec), mgssec, NULL);
669 /* Keep a refcount of servers/clients who started with "mount",
670 so we know when we can get rid of the mgc. */
671 cfs_atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
673 /* Try all connections, but only once. */
675 rc = obd_set_info_async(NULL, obd->obd_self_export,
676 sizeof(KEY_INIT_RECOV_BACKUP),
677 KEY_INIT_RECOV_BACKUP,
678 sizeof(recov_bk), &recov_bk, NULL);
681 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
683 /* We connect to the MGS at setup, and don't disconnect until cleanup */
684 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
685 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV;
687 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
688 data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB;
690 #warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
693 if (lmd_is_client(lsi->lsi_lmd) &&
694 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
695 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
696 data->ocd_version = LUSTRE_VERSION_CODE;
697 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
699 CERROR("connect failed %d\n", rc);
703 obd->u.cli.cl_mgc_mgsexp = exp;
706 /* Keep the mgc info in the sb. Note that many lsi's can point
710 cfs_mutex_unlock(&mgc_start_lock);
715 OBD_FREE(mgcname, len);
717 OBD_FREE(niduuid, len + 2);
721 static int lustre_stop_mgc(struct super_block *sb)
723 struct lustre_sb_info *lsi = s2lsi(sb);
724 struct obd_device *obd;
725 char *niduuid = 0, *ptr = 0;
726 int i, rc = 0, len = 0;
736 cfs_mutex_lock(&mgc_start_lock);
737 LASSERT(cfs_atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
738 if (!cfs_atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
739 /* This is not fatal, every client that stops
740 will call in here. */
741 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
742 cfs_atomic_read(&obd->u.cli.cl_mgc_refcount));
743 GOTO(out, rc = -EBUSY);
746 /* The MGC has no recoverable data in any case.
747 * force shotdown set in umount_begin */
748 obd->obd_no_recov = 1;
750 if (obd->u.cli.cl_mgc_mgsexp) {
751 /* An error is not fatal, if we are unable to send the
752 disconnect mgs ping evictor cleans up the export */
753 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
755 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
758 /* Save the obdname for cleaning the nid uuids, which are
760 len = strlen(obd->obd_name) + 6;
761 OBD_ALLOC(niduuid, len);
763 strcpy(niduuid, obd->obd_name);
764 ptr = niduuid + strlen(niduuid);
767 rc = class_manual_cleanup(obd);
771 /* Clean the nid uuids */
773 GOTO(out, rc = -ENOMEM);
775 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
776 sprintf(ptr, "_%x", i);
777 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
780 CERROR("del MDC UUID %s failed: rc = %d\n",
785 OBD_FREE(niduuid, len);
787 /* class_import_put will get rid of the additional connections */
788 cfs_mutex_unlock(&mgc_start_lock);
792 /* Since there's only one mgc per node, we have to change it's fs to get
793 access to the right disk. */
794 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
796 struct lustre_sb_info *lsi = s2lsi(sb);
800 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
802 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
803 rc = obd_set_info_async(NULL, mgc->obd_self_export,
804 sizeof(KEY_SET_FS), KEY_SET_FS,
805 sizeof(*sb), sb, NULL);
807 CERROR("can't set_fs %d\n", rc);
813 static int server_mgc_clear_fs(struct obd_device *mgc)
818 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
820 rc = obd_set_info_async(NULL, mgc->obd_self_export,
821 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
826 CFS_DEFINE_MUTEX(server_start_lock);
828 /* Stop MDS/OSS if nobody is using them */
829 static int server_stop_servers(int lsiflags)
831 struct obd_device *obd = NULL;
832 struct obd_type *type = NULL;
836 cfs_mutex_lock(&server_start_lock);
838 /* Either an MDT or an OST or neither */
839 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
840 if ((lsiflags & LDD_F_SV_TYPE_MDT) &&
841 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
842 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
843 type = class_search_type(LUSTRE_MDS_NAME);
845 /* if this was an OST, and there are no more OST's, clean up the OSS */
846 if ((lsiflags & LDD_F_SV_TYPE_OST) &&
847 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
848 type = class_search_type(LUSTRE_OST_NAME);
851 if (obd && (!type || !type->typ_refcnt)) {
854 /* obd_fail doesn't mean much on a server obd */
855 err = class_manual_cleanup(obd);
860 cfs_mutex_unlock(&server_start_lock);
865 int server_mti_print(char *title, struct mgs_target_info *mti)
867 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
868 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
869 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
870 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
871 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
872 mti->mti_config_ver, mti->mti_flags);
876 /** Get the fsname ("lustre") from the server name ("lustre-OST003F").
877 * @param [in] svname server name including type and index
878 * @param [out] fsname Buffer to copy filesystem name prefix into.
879 * Must have at least 'strlen(fsname) + 1' chars.
880 * @param [out] endptr if endptr isn't NULL it is set to end of fsname
883 static int server_name2fsname(char *svname, char *fsname, char **endptr)
887 p = strstr(svname, "-OST");
889 p = strstr(svname, "-MDT");
894 strncpy(fsname, svname, p - svname);
895 fsname[p - svname] = '\0';
905 * Get service name (svname) from string
907 * if endptr isn't NULL it is set to end of fsname *
909 int server_name2svname(char *label, char *svname, char **endptr)
914 /* We use server_name2fsname() just for parsing */
915 rc = server_name2fsname(label, NULL, &dash);
922 strncpy(svname, dash + 1, MTI_NAME_MAXLEN);
926 EXPORT_SYMBOL(server_name2svname);
929 /* Get the index from the obd name.
932 if endptr isn't NULL it is set to end of name */
933 int server_name2index(char *svname, __u32 *idx, char **endptr)
939 /* We use server_name2fsname() just for parsing */
940 rc = server_name2fsname(svname, NULL, &dash);
949 if (strncmp(dash, "MDT", 3) == 0)
950 rc = LDD_F_SV_TYPE_MDT;
951 else if (strncmp(dash, "OST", 3) == 0)
952 rc = LDD_F_SV_TYPE_OST;
958 if (strcmp(dash, "all") == 0)
959 return rc | LDD_F_SV_ALL;
961 index = simple_strtoul(dash, endptr, 16);
966 EXPORT_SYMBOL(server_name2index);
968 /* Generate data for registration */
969 static int server_lsi2mti(struct lustre_sb_info *lsi,
970 struct mgs_target_info *mti)
972 lnet_process_id_t id;
979 strncpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname));
981 mti->mti_nid_count = 0;
982 while (LNetGetId(i++, &id) != -ENOENT) {
983 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
986 /* server use --servicenode param, only allow specified
987 * nids be registered */
988 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
989 class_match_nid(lsi->lsi_lmd->lmd_params,
990 PARAM_FAILNODE, id.nid) < 1)
993 /* match specified network */
994 if (!class_match_net(lsi->lsi_lmd->lmd_params,
995 PARAM_NETWORK, LNET_NIDNET(id.nid)))
998 mti->mti_nids[mti->mti_nid_count] = id.nid;
999 mti->mti_nid_count++;
1000 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1001 CWARN("Only using first %d nids for %s\n",
1002 mti->mti_nid_count, mti->mti_svname);
1007 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1008 mti->mti_config_ver = 0;
1010 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1014 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1017 /* Orion requires index to be set */
1018 LASSERT(!(rc & LDD_F_NEED_INDEX));
1019 /* keep only LDD flags */
1020 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1021 mti->mti_flags |= LDD_F_UPDATE;
1022 strncpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1023 sizeof(mti->mti_params));
1027 /* Register an old or new target with the MGS. If needed MGS will construct
1028 startup logs and assign index */
1029 static int server_register_target(struct lustre_sb_info *lsi)
1031 struct obd_device *mgc = lsi->lsi_mgc;
1032 struct mgs_target_info *mti = NULL;
1039 if (!IS_SERVER(lsi))
1046 rc = server_lsi2mti(lsi, mti);
1050 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1051 mti->mti_svname, mti->mti_fsname,
1052 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1055 /* if write_conf is true, the registration must succeed */
1056 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1057 mti->mti_flags |= LDD_F_OPC_REG;
1059 /* Register the target */
1060 /* FIXME use mgc_process_config instead */
1061 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1062 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
1063 sizeof(*mti), mti, NULL);
1065 if (mti->mti_flags & LDD_F_ERROR) {
1066 LCONSOLE_ERROR_MSG(0x160,
1067 "The MGS is refusing to allow this "
1068 "server (%s) to start. Please see messages"
1069 " on the MGS node.\n", lsi->lsi_svname);
1070 } else if (writeconf) {
1071 LCONSOLE_ERROR_MSG(0x15f,
1072 "Communication to the MGS return error %d. "
1073 "Is the MGS running?\n", rc);
1075 CERROR("Cannot talk to the MGS: %d, not fatal\n", rc);
1076 /* reset the error code for non-fatal error. */
1089 * Notify the MGS that this target is ready.
1090 * Used by IR - if the MGS receives this message, it will notify clients.
1092 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1094 struct lustre_sb_info *lsi = s2lsi(sb);
1095 struct obd_device *mgc = lsi->lsi_mgc;
1096 struct mgs_target_info *mti = NULL;
1102 if (!(IS_SERVER(lsi)))
1108 rc = server_lsi2mti(lsi, mti);
1112 mti->mti_instance = obd->u.obt.obt_instance;
1113 mti->mti_flags |= LDD_F_OPC_READY;
1115 /* FIXME use mgc_process_config instead */
1116 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1117 sizeof(KEY_REGISTER_TARGET),
1118 KEY_REGISTER_TARGET,
1119 sizeof(*mti), mti, NULL);
1121 /* Imperative recovery: if the mgs informs us to use IR? */
1122 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1123 (mti->mti_flags & LDD_F_IR_CAPABLE))
1124 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1133 /** Start server targets: MDTs and OSTs
1135 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1137 struct obd_device *obd;
1138 struct lustre_sb_info *lsi = s2lsi(sb);
1139 struct config_llog_instance cfg;
1141 struct lu_device *dev;
1145 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1148 /* If we're an MDT, make sure the global MDS is running */
1149 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1150 /* make sure the MDS is started */
1151 cfs_mutex_lock(&server_start_lock);
1152 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1154 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1155 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1157 LUSTRE_MDS_OBDNAME"_uuid",
1160 cfs_mutex_unlock(&server_start_lock);
1161 CERROR("failed to start MDS: %d\n", rc);
1165 cfs_mutex_unlock(&server_start_lock);
1169 /* If we're an OST, make sure the global OSS is running */
1171 /* make sure OSS is started */
1172 cfs_mutex_lock(&server_start_lock);
1173 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1175 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1177 LUSTRE_OSS_OBDNAME"_uuid",
1180 cfs_mutex_unlock(&server_start_lock);
1181 CERROR("failed to start OSS: %d\n", rc);
1185 cfs_mutex_unlock(&server_start_lock);
1188 /* Set the mgc fs to our server disk. This allows the MGC to
1189 * read and write configs locally, in case it can't talk to the MGS. */
1190 if (lsi->lsi_srv_mnt) {
1191 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1196 /* Register with MGS */
1197 rc = server_register_target(lsi);
1201 /* Let the target look up the mount using the target's name
1202 (we can't pass the sb or mnt through class_process_config.) */
1203 rc = server_register_mount(lsi->lsi_svname, sb, mnt);
1207 /* Start targets using the llog named for the target */
1208 memset(&cfg, 0, sizeof(cfg));
1209 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1211 CERROR("failed to start server %s: %d\n",
1212 lsi->lsi_svname, rc);
1213 /* Do NOT call server_deregister_mount() here. This makes it
1214 * impossible to find mount later in cleanup time and leaves
1215 * @lsi and othder stuff leaked. -umka */
1220 /* Release the mgc fs for others to use */
1221 if (lsi->lsi_srv_mnt)
1222 server_mgc_clear_fs(lsi->lsi_mgc);
1225 obd = class_name2obd(lsi->lsi_svname);
1227 CERROR("no server named %s was started\n",
1232 server_notify_target(sb, obd);
1234 /* calculate recovery timeout, do it after lustre_process_log */
1235 server_calc_timeout(lsi, obd);
1237 /* log has been fully processed */
1238 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1240 /* log has been fully processed, let clients connect */
1241 dev = obd->obd_lu_dev;
1242 if (dev && dev->ld_ops->ldo_prepare) {
1243 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1245 struct lu_context session_ctx;
1247 lu_context_init(&session_ctx, LCT_SESSION);
1248 session_ctx.lc_thread = NULL;
1249 lu_context_enter(&session_ctx);
1250 env.le_ses = &session_ctx;
1252 dev->ld_ops->ldo_prepare(&env, NULL, dev);
1255 lu_context_exit(&session_ctx);
1256 lu_context_fini(&session_ctx);
1260 /* abort recovery only on the complete stack:
1261 * many devices can be involved */
1262 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1263 (OBP(obd, iocontrol))) {
1264 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1265 obd->obd_self_export, 0, NULL, NULL);
1272 /***************** lustre superblock **************/
1274 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1276 struct lustre_sb_info *lsi;
1282 OBD_ALLOC_PTR(lsi->lsi_lmd);
1283 if (!lsi->lsi_lmd) {
1288 lsi->lsi_lmd->lmd_exclude_count = 0;
1289 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
1290 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
1291 s2lsi_nocast(sb) = lsi;
1292 /* we take 1 extra ref for our setup */
1293 cfs_atomic_set(&lsi->lsi_mounts, 1);
1295 /* Default umount style */
1296 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1301 static int lustre_free_lsi(struct super_block *sb)
1303 struct lustre_sb_info *lsi = s2lsi(sb);
1306 LASSERT(lsi != NULL);
1307 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
1309 /* someone didn't call server_put_mount. */
1310 LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0);
1312 if (lsi->lsi_lmd != NULL) {
1313 if (lsi->lsi_lmd->lmd_dev != NULL)
1314 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1315 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1316 if (lsi->lsi_lmd->lmd_profile != NULL)
1317 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1318 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1319 if (lsi->lsi_lmd->lmd_mgssec != NULL)
1320 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
1321 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
1322 if (lsi->lsi_lmd->lmd_opts != NULL)
1323 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1324 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1325 if (lsi->lsi_lmd->lmd_exclude_count)
1326 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1327 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1328 lsi->lsi_lmd->lmd_exclude_count);
1329 if (lsi->lsi_lmd->lmd_mgs != NULL)
1330 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
1331 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
1332 if (lsi->lsi_lmd->lmd_osd_type != NULL)
1333 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
1334 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
1335 if (lsi->lsi_lmd->lmd_params != NULL)
1336 OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
1338 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1341 LASSERT(lsi->lsi_llsbi == NULL);
1342 OBD_FREE(lsi, sizeof(*lsi));
1343 s2lsi_nocast(sb) = NULL;
1348 /* The lsi has one reference for every server that is using the disk -
1349 e.g. MDT, MGS, and potentially MGC */
1350 static int lustre_put_lsi(struct super_block *sb)
1352 struct lustre_sb_info *lsi = s2lsi(sb);
1355 LASSERT(lsi != NULL);
1357 CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts));
1358 if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) {
1359 if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
1360 obd_disconnect(lsi->lsi_osd_exp);
1361 /* wait till OSD is gone */
1362 obd_zombie_barrier();
1364 lustre_free_lsi(sb);
1370 static int lsi_prepare(struct lustre_sb_info *lsi)
1377 LASSERT(lsi->lsi_lmd);
1379 /* The server name is given as a mount line option */
1380 if (lsi->lsi_lmd->lmd_profile == NULL) {
1381 LCONSOLE_ERROR("Can't determine server name\n");
1385 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
1386 RETURN(-ENAMETOOLONG);
1388 strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
1390 /* Determine osd type */
1391 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
1392 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
1393 sizeof(lsi->lsi_osd_type))
1394 RETURN(-ENAMETOOLONG);
1396 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
1398 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
1401 /* XXX: a temp. solution for components using fsfilt
1402 * to be removed in one of the subsequent patches */
1403 if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs")) {
1404 strcpy(lsi->lsi_fstype, "ldiskfs");
1406 strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
1409 /* Determine server type */
1410 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1412 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
1413 /* Assume we're a bare MGS */
1415 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
1417 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
1422 lsi->lsi_flags |= rc;
1424 /* Add mount line flags that used to be in ldd:
1425 * writeconf, mgs, iam, anything else?
1427 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
1428 LDD_F_WRITECONF : 0;
1429 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
1431 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
1432 LDD_F_SV_TYPE_MGS : 0;
1433 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_IAM) ?
1435 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
1436 LDD_F_NO_PRIMNODE : 0;
1441 /*************** server mount ******************/
1443 /** Start the shutdown of servers at umount.
1445 static void server_put_super(struct super_block *sb)
1447 struct lustre_sb_info *lsi = s2lsi(sb);
1448 struct obd_device *obd;
1449 char *tmpname, *extraname = NULL;
1451 int lsiflags = lsi->lsi_flags;
1454 LASSERT(IS_SERVER(lsi));
1456 tmpname_sz = strlen(lsi->lsi_svname) + 1;
1457 OBD_ALLOC(tmpname, tmpname_sz);
1458 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
1459 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1460 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1461 snprintf(tmpname, tmpname_sz, "MGS");
1463 /* Stop the target */
1464 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1465 (IS_MDT(lsi) || IS_OST(lsi))) {
1466 struct lustre_profile *lprof = NULL;
1468 /* tell the mgc to drop the config log */
1469 lustre_end_log(sb, lsi->lsi_svname, NULL);
1471 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1472 If there are any setup/cleanup errors, save the lov
1473 name for safety cleanup later. */
1474 lprof = class_get_profile(lsi->lsi_svname);
1475 if (lprof && lprof->lp_dt) {
1476 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1477 strcpy(extraname, lprof->lp_dt);
1480 obd = class_name2obd(lsi->lsi_svname);
1482 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1483 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1485 /* We can't seem to give an error return code
1486 * to .put_super, so we better make sure we clean up! */
1488 class_manual_cleanup(obd);
1490 CERROR("no obd %s\n", lsi->lsi_svname);
1491 server_deregister_mount(lsi->lsi_svname);
1495 /* If they wanted the mgs to stop separately from the mdt, they
1496 should have put it on a different device. */
1498 /* if MDS start with --nomgs, don't stop MGS then */
1499 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1500 server_stop_mgs(sb);
1503 /* Clean the mgc and sb */
1504 lustre_common_put_super(sb);
1506 /* wait till all in-progress cleanups are done
1507 * specifically we're interested in ofd cleanup
1509 obd_zombie_barrier();
1511 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1512 until the target is really gone so that our type refcount check
1514 server_stop_servers(lsiflags);
1516 /* In case of startup or cleanup err, stop related obds */
1518 obd = class_name2obd(extraname);
1520 CWARN("Cleaning orphaned obd %s\n", extraname);
1522 class_manual_cleanup(obd);
1524 OBD_FREE(extraname, strlen(extraname) + 1);
1527 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1528 OBD_FREE(tmpname, tmpname_sz);
1532 /** Called only for 'umount -f'
1534 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1535 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1537 struct super_block *sb = vfsmnt->mnt_sb;
1539 static void server_umount_begin(struct super_block *sb)
1542 struct lustre_sb_info *lsi = s2lsi(sb);
1545 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1546 if (!(flags & MNT_FORCE)) {
1552 CDEBUG(D_MOUNT, "umount -f\n");
1553 /* umount = failover
1555 no third way to do non-force, non-failover */
1556 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1560 static int server_statfs (struct dentry *dentry, cfs_kstatfs_t *buf)
1562 struct super_block *sb = dentry->d_sb;
1563 struct lustre_sb_info *lsi = s2lsi(sb);
1564 struct obd_statfs statfs;
1568 if (lsi->lsi_dt_dev) {
1569 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
1571 statfs_unpack(buf, &statfs);
1572 buf->f_type = sb->s_magic;
1578 buf->f_type = sb->s_magic;
1579 buf->f_bsize = sb->s_blocksize;
1585 buf->f_namelen = NAME_MAX;
1589 /** The operations we support directly on the superblock:
1590 * mount, umount, and df.
1592 static struct super_operations server_ops =
1594 .put_super = server_put_super,
1595 .umount_begin = server_umount_begin, /* umount -f */
1596 .statfs = server_statfs,
1599 #define log2(n) cfs_ffz(~(n))
1600 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1602 static int server_fill_super_common(struct super_block *sb)
1604 struct inode *root = 0;
1607 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1609 sb->s_blocksize = 4096;
1610 sb->s_blocksize_bits = log2(sb->s_blocksize);
1611 sb->s_magic = LUSTRE_SUPER_MAGIC;
1612 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1613 sb->s_flags |= MS_RDONLY;
1614 sb->s_op = &server_ops;
1616 root = new_inode(sb);
1618 CERROR("Can't make root inode\n");
1622 /* returns -EIO for every operation */
1623 /* make_bad_inode(root); -- badness - can't umount */
1624 /* apparently we need to be a directory for the mount to finish */
1625 root->i_mode = S_IFDIR;
1627 sb->s_root = d_alloc_root(root);
1629 CERROR("Can't make root dentry\n");
1637 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
1639 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1640 struct obd_device *obd;
1641 struct dt_device_param p;
1647 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
1648 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
1650 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
1651 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
1652 strcat(lsi->lsi_osd_uuid, "_UUID");
1653 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
1655 obd = class_name2obd(lsi->lsi_osd_obdname);
1657 rc = lustre_start_simple(lsi->lsi_osd_obdname,
1659 lsi->lsi_osd_uuid, lmd->lmd_dev,
1660 flagstr, lsi->lsi_lmd->lmd_opts,
1664 obd = class_name2obd(lsi->lsi_osd_obdname);
1668 rc = obd_connect(NULL, &lsi->lsi_osd_exp, obd, &obd->obd_uuid, NULL, NULL);
1671 class_manual_cleanup(obd);
1672 lsi->lsi_dt_dev = NULL;
1675 /* XXX: to keep support old components relying on lsi_srv_mnt
1676 * we get this info from OSD just started */
1677 LASSERT(obd->obd_lu_dev);
1678 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
1679 LASSERT(lsi->lsi_dt_dev);
1681 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
1683 lsi->lsi_srv_mnt = p.ddp_mnt;
1689 /** Fill in the superblock info for a Lustre server.
1690 * Mount the device with the correct options.
1691 * Read the on-disk config file.
1692 * Start the services.
1694 static int server_fill_super(struct super_block *sb)
1696 struct lustre_sb_info *lsi = s2lsi(sb);
1700 rc = lsi_prepare(lsi);
1704 /* Start low level OSD */
1705 rc = osd_start(lsi, sb->s_flags);
1707 CERROR("Unable to start osd on %s: %d\n",
1708 lsi->lsi_lmd->lmd_dev, rc);
1713 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
1714 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
1716 if (class_name2obd(lsi->lsi_svname)) {
1717 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1718 "running. Double-mount may have compromised"
1719 " the disk journal.\n",
1725 /* Start MGS before MGC */
1726 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
1727 rc = server_start_mgs(sb);
1732 /* Start MGC before servers */
1733 rc = lustre_start_mgc(sb);
1737 /* Set up all obd devices for service */
1738 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1739 (IS_OST(lsi) || IS_MDT(lsi))) {
1740 rc = server_start_targets(sb, lsi->lsi_srv_mnt);
1742 CERROR("Unable to start targets: %d\n", rc);
1745 /* FIXME overmount client here,
1746 or can we just start a client log and client_fill_super on this sb?
1747 We need to make sure server_put_super gets called too - ll_put_super
1748 calls lustre_common_put_super; check there for LSI_SERVER flag,
1750 Probably should start client from new thread so we can return.
1751 Client will not finish until all servers are connected.
1752 Note - MGS-only server does NOT get a client, since there is no
1753 lustre fs associated - the MGS is for all lustre fs's */
1756 rc = server_fill_super_common(sb);
1762 /* We jump here in case of failure while starting targets or MGS.
1763 * In this case we can't just put @mnt and have to do real cleanup
1764 * with stoping targets, etc. */
1765 server_put_super(sb);
1770 * Calculate timeout value for a target.
1772 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1774 struct lustre_mount_data *lmd;
1778 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
1779 int min = OBD_RECOVERY_TIME_MIN;
1781 LASSERT(IS_SERVER(lsi));
1785 soft = lmd->lmd_recovery_time_soft;
1786 hard = lmd->lmd_recovery_time_hard;
1787 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1788 obd->obd_no_ir = !has_ir;
1792 soft = OBD_RECOVERY_TIME_SOFT;
1794 hard = OBD_RECOVERY_TIME_HARD;
1796 /* target may have ir_factor configured. */
1797 factor = OBD_IR_FACTOR_DEFAULT;
1798 if (obd->obd_recovery_ir_factor)
1799 factor = obd->obd_recovery_ir_factor;
1802 int new_soft = soft;
1803 int new_hard = hard;
1805 /* adjust timeout value by imperative recovery */
1807 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1808 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1810 /* make sure the timeout is not too short */
1811 new_soft = max(min, new_soft);
1812 new_hard = max(new_soft, new_hard);
1814 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1815 "window shrunk from %d-%d down to %d-%d\n",
1816 obd->obd_name, soft, hard, new_soft, new_hard);
1823 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1824 obd->obd_recovery_time_hard = hard;
1825 obd->obd_recovery_ir_factor = factor;
1827 EXPORT_SYMBOL(server_calc_timeout);
1829 /*************** mount common betweeen server and client ***************/
1832 int lustre_common_put_super(struct super_block *sb)
1837 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1839 /* Drop a ref to the MGC */
1840 rc = lustre_stop_mgc(sb);
1841 if (rc && (rc != -ENOENT)) {
1843 CERROR("Can't stop MGC: %d\n", rc);
1846 /* BUSY just means that there's some other obd that
1847 needs the mgc. Let him clean it up. */
1848 CDEBUG(D_MOUNT, "MGC still in use\n");
1850 /* Drop a ref to the mounted disk */
1855 EXPORT_SYMBOL(lustre_common_put_super);
1857 static void lmd_print(struct lustre_mount_data *lmd)
1861 PRINT_CMD(PRINT_MASK, " mount data:\n");
1862 if (lmd_is_client(lmd))
1863 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
1864 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
1865 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
1868 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
1870 if (lmd->lmd_recovery_time_soft)
1871 PRINT_CMD(PRINT_MASK, "recovery time soft: %d\n",
1872 lmd->lmd_recovery_time_soft);
1874 if (lmd->lmd_recovery_time_hard)
1875 PRINT_CMD(PRINT_MASK, "recovery time hard: %d\n",
1876 lmd->lmd_recovery_time_hard);
1878 for (i = 0; i < lmd->lmd_exclude_count; i++) {
1879 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
1880 lmd->lmd_exclude[i]);
1884 /* Is this server on the exclusion list */
1885 int lustre_check_exclusion(struct super_block *sb, char *svname)
1887 struct lustre_sb_info *lsi = s2lsi(sb);
1888 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1893 rc = server_name2index(svname, &index, NULL);
1894 if (rc != LDD_F_SV_TYPE_OST)
1895 /* Only exclude OSTs */
1898 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
1899 index, lmd->lmd_exclude_count, lmd->lmd_dev);
1901 for(i = 0; i < lmd->lmd_exclude_count; i++) {
1902 if (index == lmd->lmd_exclude[i]) {
1903 CWARN("Excluding %s (on exclusion list)\n", svname);
1910 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
1911 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
1913 char *s1 = ptr, *s2;
1914 __u32 index, *exclude_list;
1918 /* The shortest an ost name can be is 8 chars: -OST0000.
1919 We don't actually know the fsname at this time, so in fact
1920 a user could specify any fsname. */
1921 devmax = strlen(ptr) / 8 + 1;
1923 /* temp storage until we figure out how many we have */
1924 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
1928 /* we enter this fn pointing at the '=' */
1929 while (*s1 && *s1 != ' ' && *s1 != ',') {
1931 rc = server_name2index(s1, &index, &s2);
1933 CERROR("Can't parse server name '%s'\n", s1);
1936 if (rc == LDD_F_SV_TYPE_OST)
1937 exclude_list[lmd->lmd_exclude_count++] = index;
1939 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
1941 /* now we are pointing at ':' (next exclude)
1942 or ',' (end of excludes) */
1943 if (lmd->lmd_exclude_count >= devmax)
1946 if (rc >= 0) /* non-err */
1949 if (lmd->lmd_exclude_count) {
1950 /* permanent, freed in lustre_free_lsi */
1951 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
1952 lmd->lmd_exclude_count);
1953 if (lmd->lmd_exclude) {
1954 memcpy(lmd->lmd_exclude, exclude_list,
1955 sizeof(index) * lmd->lmd_exclude_count);
1958 lmd->lmd_exclude_count = 0;
1961 OBD_FREE(exclude_list, sizeof(index) * devmax);
1965 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
1970 if (lmd->lmd_mgssec != NULL) {
1971 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
1972 lmd->lmd_mgssec = NULL;
1975 tail = strchr(ptr, ',');
1977 length = strlen(ptr);
1979 length = tail - ptr;
1981 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
1982 if (lmd->lmd_mgssec == NULL)
1985 memcpy(lmd->lmd_mgssec, ptr, length);
1986 lmd->lmd_mgssec[length] = '\0';
1990 static int lmd_parse_string(char **handle, char *ptr)
1995 if ((handle == NULL) || (ptr == NULL))
1998 if (*handle != NULL) {
1999 OBD_FREE(*handle, strlen(*handle) + 1);
2003 tail = strchr(ptr, ',');
2005 length = strlen(ptr);
2007 length = tail - ptr;
2009 OBD_ALLOC(*handle, length + 1);
2010 if (*handle == NULL)
2013 memcpy(*handle, ptr, length);
2014 (*handle)[length] = '\0';
2019 /* Collect multiple values for mgsnid specifiers */
2020 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
2028 /* Find end of nidlist */
2029 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
2030 length = tail - *ptr;
2032 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
2036 if (lmd->lmd_mgs != NULL)
2037 oldlen = strlen(lmd->lmd_mgs) + 1;
2039 OBD_ALLOC(mgsnid, oldlen + length + 1);
2043 if (lmd->lmd_mgs != NULL) {
2044 /* Multiple mgsnid= are taken to mean failover locations */
2045 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
2046 mgsnid[oldlen - 1] = ':';
2047 OBD_FREE(lmd->lmd_mgs, oldlen);
2049 memcpy(mgsnid + oldlen, *ptr, length);
2050 mgsnid[oldlen + length] = '\0';
2051 lmd->lmd_mgs = mgsnid;
2057 /** Parse mount line options
2058 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
2059 * dev is passed as device=uml1:/lustre by mount.lustre
2061 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
2063 char *s1, *s2, *devname = NULL;
2064 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
2070 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
2071 "/sbin/mount.lustre is installed.\n");
2075 /* Options should be a string - try to detect old lmd data */
2076 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
2077 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
2078 "/sbin/mount.lustre. Please install "
2079 "version %s\n", LUSTRE_VERSION_STRING);
2082 lmd->lmd_magic = LMD_MAGIC;
2084 OBD_ALLOC(lmd->lmd_params, 4096);
2085 if (lmd->lmd_params == NULL)
2087 lmd->lmd_params[0] = '\0';
2089 /* Set default flags here */
2094 int time_min = OBD_RECOVERY_TIME_MIN;
2096 /* Skip whitespace and extra commas */
2097 while (*s1 == ' ' || *s1 == ',')
2100 /* Client options are parsed in ll_options: eg. flock,
2103 /* Parse non-ldiskfs options here. Rather than modifying
2104 ldiskfs, we just zero these out here */
2105 if (strncmp(s1, "abort_recov", 11) == 0) {
2106 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
2108 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
2109 lmd->lmd_recovery_time_soft = max_t(int,
2110 simple_strtoul(s1 + 19, NULL, 10), time_min);
2112 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
2113 lmd->lmd_recovery_time_hard = max_t(int,
2114 simple_strtoul(s1 + 19, NULL, 10), time_min);
2116 } else if (strncmp(s1, "noir", 4) == 0) {
2117 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
2119 } else if (strncmp(s1, "nosvc", 5) == 0) {
2120 lmd->lmd_flags |= LMD_FLG_NOSVC;
2122 } else if (strncmp(s1, "nomgs", 5) == 0) {
2123 lmd->lmd_flags |= LMD_FLG_NOMGS;
2125 } else if (strncmp(s1, "noscrub", 7) == 0) {
2126 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
2128 } else if (strncmp(s1, PARAM_MGSNODE,
2129 sizeof(PARAM_MGSNODE) - 1) == 0) {
2130 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
2131 /* Assume the next mount opt is the first
2132 invalid nid we get to. */
2133 rc = lmd_parse_mgs(lmd, &s2);
2137 } else if (strncmp(s1, "writeconf", 9) == 0) {
2138 lmd->lmd_flags |= LMD_FLG_WRITECONF;
2140 } else if (strncmp(s1, "virgin", 6) == 0) {
2141 lmd->lmd_flags |= LMD_FLG_VIRGIN;
2143 } else if (strncmp(s1, "noprimnode", 10) == 0) {
2144 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
2146 } else if (strncmp(s1, "mgssec=", 7) == 0) {
2147 rc = lmd_parse_mgssec(lmd, s1 + 7);
2151 /* ost exclusion list */
2152 } else if (strncmp(s1, "exclude=", 8) == 0) {
2153 rc = lmd_make_exclusion(lmd, s1 + 7);
2157 } else if (strncmp(s1, "mgs", 3) == 0) {
2159 lmd->lmd_flags |= LMD_FLG_MGS;
2161 } else if (strncmp(s1, "svname=", 7) == 0) {
2162 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
2166 } else if (strncmp(s1, "param=", 6) == 0) {
2168 char *tail = strchr(s1 + 6, ',');
2170 length = strlen(s1);
2174 strncat(lmd->lmd_params, s1 + 6, length);
2175 strcat(lmd->lmd_params, " ");
2177 } else if (strncmp(s1, "osd=", 4) == 0) {
2178 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
2183 /* Linux 2.4 doesn't pass the device, so we stuck it at the
2184 end of the options. */
2185 else if (strncmp(s1, "device=", 7) == 0) {
2187 /* terminate options right before device. device
2188 must be the last one. */
2194 s2 = strchr(s1, ',');
2202 memmove(s1, s2, strlen(s2) + 1);
2208 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
2209 "(need mount option 'device=...')\n");
2213 s1 = strstr(devname, ":/");
2216 lmd->lmd_flags |= LMD_FLG_CLIENT;
2217 /* Remove leading /s from fsname */
2218 while (*++s1 == '/') ;
2219 /* Freed in lustre_free_lsi */
2220 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
2221 if (!lmd->lmd_profile)
2223 sprintf(lmd->lmd_profile, "%s-client", s1);
2226 /* Freed in lustre_free_lsi */
2227 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
2230 strcpy(lmd->lmd_dev, devname);
2232 /* Save mount options */
2233 s1 = options + strlen(options) - 1;
2234 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
2236 if (*options != 0) {
2237 /* Freed in lustre_free_lsi */
2238 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
2241 strcpy(lmd->lmd_opts, options);
2245 lmd->lmd_magic = LMD_MAGIC;
2250 CERROR("Bad mount options %s\n", options);
2254 struct lustre_mount_data2 {
2256 struct vfsmount *lmd2_mnt;
2259 /** This is the entry point for the mount call into Lustre.
2260 * This is called when a server or client is mounted,
2261 * and this is where we start setting things up.
2262 * @param data Mount options (e.g. -o flock,abort_recov)
2264 int lustre_fill_super(struct super_block *sb, void *data, int silent)
2266 struct lustre_mount_data *lmd;
2267 struct lustre_mount_data2 *lmd2 = data;
2268 struct lustre_sb_info *lsi;
2272 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
2274 lsi = lustre_init_lsi(sb);
2280 * Disable lockdep during mount, because mount locking patterns are
2286 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
2288 obd_zombie_barrier();
2290 /* Figure out the lmd from the mount options */
2291 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
2293 GOTO(out, rc = -EINVAL);
2296 if (lmd_is_client(lmd)) {
2297 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
2298 if (!client_fill_super) {
2299 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
2300 "client mount! Is the 'lustre' "
2301 "module loaded?\n");
2305 rc = lustre_start_mgc(sb);
2310 /* Connect and start */
2311 /* (should always be ll_fill_super) */
2312 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
2313 /* c_f_s will call lustre_common_put_super on failure */
2316 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
2317 rc = server_fill_super(sb);
2318 /* s_f_s calls lustre_start_mgc after the mount because we need
2319 the MGS nids which are stored on disk. Plus, we may
2320 need to start the MGS first. */
2321 /* s_f_s will call server_put_super on failure */
2324 /* If error happens in fill_super() call, @lsi will be killed there.
2325 * This is why we do not put it here. */
2329 CERROR("Unable to mount %s (%d)\n",
2330 s2lsi(sb) ? lmd->lmd_dev : "", rc);
2332 CDEBUG(D_SUPER, "Mount %s complete\n",
2340 /* We can't call ll_fill_super by name because it lives in a module that
2341 must be loaded after this one. */
2342 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
2343 struct vfsmount *mnt))
2345 client_fill_super = cfs;
2347 EXPORT_SYMBOL(lustre_register_client_fill_super);
2349 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
2351 kill_super_cb = cfs;
2353 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2355 /***************** FS registration ******************/
2356 #ifdef HAVE_FSTYPE_MOUNT
2357 struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
2358 const char *devname, void *data)
2360 struct lustre_mount_data2 lmd2 = { data, NULL };
2362 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
2365 int lustre_get_sb(struct file_system_type *fs_type, int flags,
2366 const char *devname, void * data, struct vfsmount *mnt)
2368 struct lustre_mount_data2 lmd2 = { data, mnt };
2370 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
2374 void lustre_kill_super(struct super_block *sb)
2376 struct lustre_sb_info *lsi = s2lsi(sb);
2378 if (kill_super_cb && lsi && !IS_SERVER(lsi))
2379 (*kill_super_cb)(sb);
2381 kill_anon_super(sb);
2384 /** Register the "lustre" fs type
2386 struct file_system_type lustre_fs_type = {
2387 .owner = THIS_MODULE,
2389 #ifdef HAVE_FSTYPE_MOUNT
2390 .mount = lustre_mount,
2392 .get_sb = lustre_get_sb,
2394 .kill_sb = lustre_kill_super,
2395 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
2396 #ifdef FS_HAS_FIEMAP
2399 LL_RENAME_DOES_D_MOVE,
2402 int lustre_register_fs(void)
2404 return register_filesystem(&lustre_fs_type);
2407 int lustre_unregister_fs(void)
2409 return unregister_filesystem(&lustre_fs_type);
2412 EXPORT_SYMBOL(server_mti_print);