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, vfscount=%d\n",
126 lmi->lmi_mnt ? mnt_get_count(lmi->lmi_mnt) : -1);
131 /* when an obd no longer needs a mount */
132 static int server_deregister_mount(const char *name)
134 struct lustre_mount_info *lmi;
137 cfs_mutex_lock(&lustre_mount_info_lock);
138 lmi = server_find_mount(name);
140 cfs_mutex_unlock(&lustre_mount_info_lock);
141 CERROR("%s not registered\n", name);
145 CDEBUG(D_MOUNT, "dereg_mnt %p from %s, vfscount=%d\n",
147 lmi->lmi_mnt ? mnt_get_count(lmi->lmi_mnt) : -1);
149 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
150 cfs_list_del(&lmi->lmi_list_chain);
151 OBD_FREE(lmi, sizeof(*lmi));
152 cfs_mutex_unlock(&lustre_mount_info_lock);
157 /* obd's look up a registered mount using their obdname. This is just
158 for initial obd setup to find the mount struct. It should not be
159 called every time you want to mntget. */
160 struct lustre_mount_info *server_get_mount(const char *name)
162 struct lustre_mount_info *lmi;
163 struct lustre_sb_info *lsi;
166 cfs_mutex_lock(&lustre_mount_info_lock);
167 lmi = server_find_mount(name);
168 cfs_mutex_unlock(&lustre_mount_info_lock);
170 CERROR("Can't find mount for %s\n", name);
173 lsi = s2lsi(lmi->lmi_sb);
176 mntget(lmi->lmi_mnt);
177 cfs_atomic_inc(&lsi->lsi_mounts);
179 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d, vfscount=%d\n",
180 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts),
181 lmi->lmi_mnt ? mnt_get_count(lmi->lmi_mnt) - 1 : -1);
185 EXPORT_SYMBOL(server_get_mount);
188 * Used by mdt to get mount_info from obdname.
189 * There are no blocking when using the mount_info.
190 * Do not use server_get_mount for this purpose.
192 struct lustre_mount_info *server_get_mount_2(const char *name)
194 struct lustre_mount_info *lmi;
197 cfs_mutex_lock(&lustre_mount_info_lock);
198 lmi = server_find_mount(name);
199 cfs_mutex_unlock(&lustre_mount_info_lock);
201 CERROR("Can't find mount for %s\n", name);
205 EXPORT_SYMBOL(server_get_mount_2);
207 static void unlock_mntput(struct vfsmount *mnt)
209 #ifdef HAVE_KERNEL_LOCKED
210 /* for kernel < 2.6.37 */
211 if (kernel_locked()) {
223 static int lustre_put_lsi(struct super_block *sb);
225 /* to be called from obd_cleanup methods */
226 int server_put_mount(const char *name, struct vfsmount *mnt)
228 struct lustre_mount_info *lmi;
229 struct lustre_sb_info *lsi;
233 /* This might be the last one, can't deref after this */
235 count = mnt_get_count(mnt) - 1;
239 cfs_mutex_lock(&lustre_mount_info_lock);
240 lmi = server_find_mount(name);
241 cfs_mutex_unlock(&lustre_mount_info_lock);
243 CERROR("Can't find mount for %s\n", name);
246 lsi = s2lsi(lmi->lmi_sb);
247 LASSERT(lmi->lmi_mnt == mnt);
249 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d, vfscount=%d\n",
250 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts), count);
252 if (lustre_put_lsi(lmi->lmi_sb)) {
253 CDEBUG(D_MOUNT, "Last put of mnt %p from %s, vfscount=%d\n",
254 lmi->lmi_mnt, name, count);
255 /* last mount is the One True Mount */
257 CERROR("%s: mount busy, vfscount=%d!\n", name, count);
260 /* this obd should never need the mount again */
261 server_deregister_mount(name);
265 EXPORT_SYMBOL(server_put_mount);
267 /* Corresponding to server_get_mount_2 */
268 int server_put_mount_2(const char *name, struct vfsmount *mnt)
273 EXPORT_SYMBOL(server_put_mount_2);
275 /**************** config llog ********************/
277 /** Get a config log from the MGS and process it.
278 * This func is called for both clients and servers.
279 * Continue to process new statements appended to the logs
280 * (whenever the config lock is revoked) until lustre_end_log
282 * @param sb The superblock is used by the MGC to write to the local copy of
284 * @param logname The name of the llog to replicate from the MGS
285 * @param cfg Since the same mgc may be used to follow multiple config logs
286 * (e.g. ost1, ost2, client), the config_llog_instance keeps the state for
287 * this log, and is added to the mgc's list of logs to follow.
289 int lustre_process_log(struct super_block *sb, char *logname,
290 struct config_llog_instance *cfg)
292 struct lustre_cfg *lcfg;
293 struct lustre_cfg_bufs *bufs;
294 struct lustre_sb_info *lsi = s2lsi(sb);
295 struct obd_device *mgc = lsi->lsi_mgc;
306 /* mgc_process_config */
307 lustre_cfg_bufs_reset(bufs, mgc->obd_name);
308 lustre_cfg_bufs_set_string(bufs, 1, logname);
309 lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg));
310 lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb));
311 lcfg = lustre_cfg_new(LCFG_LOG_START, bufs);
312 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
313 lustre_cfg_free(lcfg);
318 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
319 "failed from the MGS (%d). Make sure this "
320 "client and the MGS are running compatible "
321 "versions of Lustre.\n",
322 mgc->obd_name, logname, rc);
325 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
326 "failed (%d). This may be the result of "
327 "communication errors between this node and "
328 "the MGS, a bad configuration, or other "
329 "errors. See the syslog for more "
330 "information.\n", mgc->obd_name, logname,
333 /* class_obd_list(); */
336 EXPORT_SYMBOL(lustre_process_log);
338 /* Stop watching this config log for updates */
339 int lustre_end_log(struct super_block *sb, char *logname,
340 struct config_llog_instance *cfg)
342 struct lustre_cfg *lcfg;
343 struct lustre_cfg_bufs bufs;
344 struct lustre_sb_info *lsi = s2lsi(sb);
345 struct obd_device *mgc = lsi->lsi_mgc;
352 /* mgc_process_config */
353 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
354 lustre_cfg_bufs_set_string(&bufs, 1, logname);
356 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
357 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
358 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
359 lustre_cfg_free(lcfg);
362 EXPORT_SYMBOL(lustre_end_log);
364 /**************** obd start *******************/
366 /** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from
367 * lctl (and do for echo cli/srv.
369 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
370 char *s1, char *s2, char *s3, char *s4)
372 struct lustre_cfg_bufs bufs;
373 struct lustre_cfg * lcfg = NULL;
376 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
377 cmd, s1, s2, s3, s4);
379 lustre_cfg_bufs_reset(&bufs, cfgname);
381 lustre_cfg_bufs_set_string(&bufs, 1, s1);
383 lustre_cfg_bufs_set_string(&bufs, 2, s2);
385 lustre_cfg_bufs_set_string(&bufs, 3, s3);
387 lustre_cfg_bufs_set_string(&bufs, 4, s4);
389 lcfg = lustre_cfg_new(cmd, &bufs);
390 lcfg->lcfg_nid = nid;
391 rc = class_process_config(lcfg);
392 lustre_cfg_free(lcfg);
395 EXPORT_SYMBOL(do_lcfg);
397 /** Call class_attach and class_setup. These methods in turn call
398 * obd type-specific methods.
400 static int lustre_start_simple(char *obdname, char *type, char *uuid,
404 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
406 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
408 CERROR("%s attach error %d\n", obdname, rc);
411 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, 0, 0);
413 CERROR("%s setup error %d\n", obdname, rc);
414 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
419 /* Set up a MGS to serve startup logs */
420 static int server_start_mgs(struct super_block *sb)
422 struct lustre_sb_info *lsi = s2lsi(sb);
423 struct vfsmount *mnt = lsi->lsi_srv_mnt;
424 struct lustre_mount_info *lmi;
429 /* It is impossible to have more than 1 MGS per node, since
430 MGC wouldn't know which to connect to */
431 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
433 lsi = s2lsi(lmi->lmi_sb);
434 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
436 lsi->lsi_ldd->ldd_svname);
440 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
442 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
445 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
446 LUSTRE_MGS_OBDNAME, 0, 0);
447 /* Do NOT call server_deregister_mount() here. This leads to
448 * inability cleanup cleanly and free lsi and other stuff when
449 * mgs calls server_put_mount() in error handling case. -umka */
453 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
454 "Is the 'mgs' module loaded?\n",
455 LUSTRE_MGS_OBDNAME, rc);
459 static int server_stop_mgs(struct super_block *sb)
461 struct obd_device *obd;
465 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
467 /* There better be only one MGS */
468 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
470 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
474 /* The MGS should always stop when we say so */
476 rc = class_manual_cleanup(obd);
480 CFS_DEFINE_MUTEX(mgc_start_lock);
482 /** Set up a mgc obd to process startup logs
484 * \param sb [in] super block of the mgc obd
486 * \retval 0 success, otherwise error code
488 static int lustre_start_mgc(struct super_block *sb)
490 struct obd_connect_data *data = NULL;
491 struct lustre_sb_info *lsi = s2lsi(sb);
492 struct obd_device *obd;
493 struct obd_export *exp;
494 struct obd_uuid *uuid;
497 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
500 int rc = 0, i = 0, j, len;
503 LASSERT(lsi->lsi_lmd);
505 /* Find the first non-lo MGS nid for our MGC name */
506 if (lsi->lsi_flags & LSI_SERVER) {
507 ptr = lsi->lsi_ldd->ldd_params;
508 /* mount -o mgsnode=nid */
509 if (lsi->lsi_lmd->lmd_mgs &&
510 (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) {
512 /* Use mgsnode= nids */
513 } else if ((class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0) &&
514 (class_parse_nid(ptr, &nid, &ptr) == 0)) {
516 } else if (IS_MGS(lsi->lsi_ldd)) {
517 lnet_process_id_t id;
518 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
519 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
526 } else { /* client */
527 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
528 ptr = lsi->lsi_lmd->lmd_dev;
529 if (class_parse_nid(ptr, &nid, &ptr) == 0)
533 CERROR("No valid MGS nids found.\n");
537 cfs_mutex_lock(&mgc_start_lock);
539 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
540 OBD_ALLOC(mgcname, len);
541 OBD_ALLOC(niduuid, len + 2);
542 if (!mgcname || !niduuid)
543 GOTO(out_free, rc = -ENOMEM);
544 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
546 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
550 GOTO(out_free, rc = -ENOMEM);
552 obd = class_name2obd(mgcname);
553 if (obd && !obd->obd_stopping) {
554 rc = obd_set_info_async(NULL, obd->obd_self_export,
555 strlen(KEY_MGSSEC), KEY_MGSSEC,
556 strlen(mgssec), mgssec, NULL);
560 /* Re-using an existing MGC */
561 cfs_atomic_inc(&obd->u.cli.cl_mgc_refcount);
563 /* IR compatibility check, only for clients */
564 if (lmd_is_client(lsi->lsi_lmd)) {
566 int vallen = sizeof(*data);
567 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
569 rc = obd_get_info(NULL, obd->obd_self_export,
570 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
571 &vallen, data, NULL);
573 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
574 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
575 /* LMD_FLG_NOIR is for test purpose only */
577 "Trying to mount a client with IR setting "
578 "not compatible with current mgc. "
579 "Force to use current mgc setting that is "
581 has_ir ? "enabled" : "disabled");
583 *flags &= ~LMD_FLG_NOIR;
585 *flags |= LMD_FLG_NOIR;
590 /* If we are restarting the MGS, don't try to keep the MGC's
591 old connection, or registration will fail. */
592 if ((lsi->lsi_flags & LSI_SERVER) && IS_MGS(lsi->lsi_ldd)) {
593 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
597 /* Try all connections, but only once (again).
598 We don't want to block another target from starting
599 (using its local copy of the log), but we do want to connect
600 if at all possible. */
602 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
603 rc = obd_set_info_async(NULL, obd->obd_self_export,
604 sizeof(KEY_INIT_RECOV_BACKUP),
605 KEY_INIT_RECOV_BACKUP,
606 sizeof(recov_bk), &recov_bk, NULL);
610 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
612 /* Add the primary nids for the MGS */
614 sprintf(niduuid, "%s_%x", mgcname, i);
615 if (lsi->lsi_flags & LSI_SERVER) {
616 ptr = lsi->lsi_ldd->ldd_params;
617 if (IS_MGS(lsi->lsi_ldd)) {
618 /* Use local nids (including LO) */
619 lnet_process_id_t id;
620 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
621 rc = do_lcfg(mgcname, id.nid,
622 LCFG_ADD_UUID, niduuid, 0,0,0);
625 /* Use mgsnode= nids */
626 /* mount -o mgsnode=nid */
627 if (lsi->lsi_lmd->lmd_mgs) {
628 ptr = lsi->lsi_lmd->lmd_mgs;
629 } else if (class_find_param(ptr, PARAM_MGSNODE,
631 CERROR("No MGS nids given.\n");
632 GOTO(out_free, rc = -EINVAL);
634 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
635 rc = do_lcfg(mgcname, nid,
636 LCFG_ADD_UUID, niduuid, 0,0,0);
640 } else { /* client */
641 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
642 ptr = lsi->lsi_lmd->lmd_dev;
643 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
644 rc = do_lcfg(mgcname, nid,
645 LCFG_ADD_UUID, niduuid, 0,0,0);
647 /* Stop at the first failover nid */
653 CERROR("No valid MGS nids found.\n");
654 GOTO(out_free, rc = -EINVAL);
656 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
658 /* Random uuid for MGC allows easier reconnects */
660 ll_generate_random_uuid(uuidc);
661 class_uuid_unparse(uuidc, uuid);
664 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
665 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
671 /* Add any failover MGS nids */
673 while ((*ptr == ':' ||
674 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0)) {
675 /* New failover node */
676 sprintf(niduuid, "%s_%x", mgcname, i);
678 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
680 rc = do_lcfg(mgcname, nid,
681 LCFG_ADD_UUID, niduuid, 0,0,0);
686 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
694 lsi->lsi_lmd->lmd_mgs_failnodes = i;
696 obd = class_name2obd(mgcname);
698 CERROR("Can't find mgcobd %s\n", mgcname);
699 GOTO(out_free, rc = -ENOTCONN);
702 rc = obd_set_info_async(NULL, obd->obd_self_export,
703 strlen(KEY_MGSSEC), KEY_MGSSEC,
704 strlen(mgssec), mgssec, NULL);
708 /* Keep a refcount of servers/clients who started with "mount",
709 so we know when we can get rid of the mgc. */
710 cfs_atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
712 /* Try all connections, but only once. */
714 rc = obd_set_info_async(NULL, obd->obd_self_export,
715 sizeof(KEY_INIT_RECOV_BACKUP),
716 KEY_INIT_RECOV_BACKUP,
717 sizeof(recov_bk), &recov_bk, NULL);
720 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
722 /* We connect to the MGS at setup, and don't disconnect until cleanup */
723 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
724 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV |
725 OBD_CONNECT_MNE_SWAB;
726 if (lmd_is_client(lsi->lsi_lmd) &&
727 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
728 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
729 data->ocd_version = LUSTRE_VERSION_CODE;
730 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
732 CERROR("connect failed %d\n", rc);
736 obd->u.cli.cl_mgc_mgsexp = exp;
739 /* Keep the mgc info in the sb. Note that many lsi's can point
743 cfs_mutex_unlock(&mgc_start_lock);
748 OBD_FREE(mgcname, len);
750 OBD_FREE(niduuid, len + 2);
754 static int lustre_stop_mgc(struct super_block *sb)
756 struct lustre_sb_info *lsi = s2lsi(sb);
757 struct obd_device *obd;
758 char *niduuid = 0, *ptr = 0;
759 int i, rc = 0, len = 0;
769 cfs_mutex_lock(&mgc_start_lock);
770 LASSERT(cfs_atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
771 if (!cfs_atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
772 /* This is not fatal, every client that stops
773 will call in here. */
774 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
775 cfs_atomic_read(&obd->u.cli.cl_mgc_refcount));
776 GOTO(out, rc = -EBUSY);
779 /* The MGC has no recoverable data in any case.
780 * force shotdown set in umount_begin */
781 obd->obd_no_recov = 1;
783 if (obd->u.cli.cl_mgc_mgsexp) {
784 /* An error is not fatal, if we are unable to send the
785 disconnect mgs ping evictor cleans up the export */
786 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
788 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
791 /* Save the obdname for cleaning the nid uuids, which are
793 len = strlen(obd->obd_name) + 6;
794 OBD_ALLOC(niduuid, len);
796 strcpy(niduuid, obd->obd_name);
797 ptr = niduuid + strlen(niduuid);
800 rc = class_manual_cleanup(obd);
804 /* Clean the nid uuids */
806 GOTO(out, rc = -ENOMEM);
808 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
809 sprintf(ptr, "_%x", i);
810 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
813 CERROR("del MDC UUID %s failed: rc = %d\n",
818 OBD_FREE(niduuid, len);
820 /* class_import_put will get rid of the additional connections */
821 cfs_mutex_unlock(&mgc_start_lock);
825 /* Since there's only one mgc per node, we have to change it's fs to get
826 access to the right disk. */
827 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
829 struct lustre_sb_info *lsi = s2lsi(sb);
833 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
835 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
836 rc = obd_set_info_async(NULL, mgc->obd_self_export,
837 sizeof(KEY_SET_FS), KEY_SET_FS,
838 sizeof(*sb), sb, NULL);
840 CERROR("can't set_fs %d\n", rc);
846 static int server_mgc_clear_fs(struct obd_device *mgc)
851 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
853 rc = obd_set_info_async(NULL, mgc->obd_self_export,
854 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
859 CFS_DEFINE_MUTEX(server_start_lock);
861 /* Stop MDS/OSS if nobody is using them */
862 static int server_stop_servers(int lddflags, int lsiflags)
864 struct obd_device *obd = NULL;
865 struct obd_type *type = NULL;
869 cfs_mutex_lock(&server_start_lock);
871 /* Either an MDT or an OST or neither */
872 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
873 if ((lddflags & LDD_F_SV_TYPE_MDT) &&
874 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
875 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
876 type = class_search_type(LUSTRE_MDS_NAME);
878 /* if this was an OST, and there are no more OST's, clean up the OSS */
879 if ((lddflags & LDD_F_SV_TYPE_OST) &&
880 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
881 type = class_search_type(LUSTRE_OST_NAME);
884 if (obd && (!type || !type->typ_refcnt)) {
887 /* obd_fail doesn't mean much on a server obd */
888 err = class_manual_cleanup(obd);
893 cfs_mutex_unlock(&server_start_lock);
898 int server_mti_print(char *title, struct mgs_target_info *mti)
900 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
901 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
902 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
903 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
904 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
905 mti->mti_config_ver, mti->mti_flags);
908 EXPORT_SYMBOL(server_mti_print);
910 static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
912 struct lustre_sb_info *lsi = s2lsi(sb);
913 struct lustre_disk_data *ldd = lsi->lsi_ldd;
914 lnet_process_id_t id;
918 if (!(lsi->lsi_flags & LSI_SERVER))
921 strncpy(mti->mti_fsname, ldd->ldd_fsname,
922 sizeof(mti->mti_fsname));
923 strncpy(mti->mti_svname, ldd->ldd_svname,
924 sizeof(mti->mti_svname));
926 mti->mti_nid_count = 0;
927 while (LNetGetId(i++, &id) != -ENOENT) {
928 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
931 /* server use --servicenode param, only allow specified
932 * nids be registered */
933 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
934 class_match_nid(lsi->lsi_lmd->lmd_params,
935 PARAM_FAILNODE, id.nid) < 1)
938 /* match specified network */
939 if (!class_match_net(lsi->lsi_lmd->lmd_params,
940 PARAM_NETWORK, LNET_NIDNET(id.nid)))
943 mti->mti_nids[mti->mti_nid_count] = id.nid;
944 mti->mti_nid_count++;
945 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
946 CWARN("Only using first %d nids for %s\n",
947 mti->mti_nid_count, mti->mti_svname);
952 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
953 mti->mti_config_ver = 0;
954 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF)
955 ldd->ldd_flags |= LDD_F_WRITECONF;
956 mti->mti_flags = ldd->ldd_flags;
957 mti->mti_stripe_index = ldd->ldd_svindex;
958 memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid));
959 if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) {
960 CERROR("params too big for mti\n");
963 strncpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
964 sizeof(mti->mti_params));
968 /* Register an old or new target with the MGS. If needed MGS will construct
969 startup logs and assign index */
970 int server_register_target(struct super_block *sb)
972 struct lustre_sb_info *lsi = s2lsi(sb);
973 struct obd_device *mgc = lsi->lsi_mgc;
974 struct lustre_disk_data *ldd = lsi->lsi_ldd;
975 struct mgs_target_info *mti = NULL;
982 if (!(lsi->lsi_flags & LSI_SERVER))
988 rc = server_sb2mti(sb, mti);
992 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
993 mti->mti_svname, mti->mti_fsname,
994 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
997 /* if write_conf is true, the registration must succeed */
998 writeconf = !!(ldd->ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
999 mti->mti_flags |= LDD_F_OPC_REG;
1001 /* Register the target */
1002 /* FIXME use mgc_process_config instead */
1003 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1004 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
1005 sizeof(*mti), mti, NULL);
1007 if (mti->mti_flags & LDD_F_ERROR) {
1008 LCONSOLE_ERROR_MSG(0x160,
1009 "The MGS is refusing to allow this "
1010 "server (%s) to start. Please see messages"
1011 " on the MGS node.\n", ldd->ldd_svname);
1012 } else if (writeconf) {
1013 LCONSOLE_ERROR_MSG(0x15f,
1014 "Communication to the MGS return error %d. "
1015 "Is the MGS running?\n", rc);
1017 CERROR("Cannot talk to the MGS: %d, not fatal\n", rc);
1018 /* reset the error code for non-fatal error. */
1029 EXPORT_SYMBOL(server_register_target);
1032 * Notify the MGS that this target is ready.
1033 * Used by IR - if the MGS receives this message, it will notify clients.
1035 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1037 struct lustre_sb_info *lsi = s2lsi(sb);
1038 struct obd_device *mgc = lsi->lsi_mgc;
1039 struct mgs_target_info *mti = NULL;
1045 if (!(lsi->lsi_flags & LSI_SERVER))
1051 rc = server_sb2mti(sb, mti);
1055 mti->mti_instance = obd->u.obt.obt_instance;
1056 mti->mti_flags |= LDD_F_OPC_READY;
1058 /* FIXME use mgc_process_config instead */
1059 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1060 sizeof(KEY_REGISTER_TARGET),
1061 KEY_REGISTER_TARGET,
1062 sizeof(*mti), mti, NULL);
1064 /* Imperative recovery: if the mgs informs us to use IR? */
1065 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1066 (mti->mti_flags & LDD_F_IR_CAPABLE))
1067 lsi->lsi_flags |= LSI_IR_CAPABLE;
1076 /** Start server targets: MDTs and OSTs
1078 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1080 struct obd_device *obd;
1081 struct lustre_sb_info *lsi = s2lsi(sb);
1082 struct config_llog_instance cfg;
1086 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname);
1089 /* If we're an MDT, make sure the global MDS is running */
1090 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1091 /* make sure the MDS is started */
1092 cfs_mutex_lock(&server_start_lock);
1093 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1095 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1096 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1098 LUSTRE_MDS_OBDNAME"_uuid",
1101 cfs_mutex_unlock(&server_start_lock);
1102 CERROR("failed to start MDS: %d\n", rc);
1106 cfs_mutex_unlock(&server_start_lock);
1110 /* If we're an OST, make sure the global OSS is running */
1111 if (IS_OST(lsi->lsi_ldd)) {
1112 /* make sure OSS is started */
1113 cfs_mutex_lock(&server_start_lock);
1114 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1116 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1118 LUSTRE_OSS_OBDNAME"_uuid",
1121 cfs_mutex_unlock(&server_start_lock);
1122 CERROR("failed to start OSS: %d\n", rc);
1126 cfs_mutex_unlock(&server_start_lock);
1129 /* Set the mgc fs to our server disk. This allows the MGC to
1130 * read and write configs locally, in case it can't talk to the MGS. */
1131 if (lsi->lsi_srv_mnt) {
1132 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1137 /* Register with MGS */
1138 rc = server_register_target(sb);
1142 /* Let the target look up the mount using the target's name
1143 (we can't pass the sb or mnt through class_process_config.) */
1144 rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
1148 /* Start targets using the llog named for the target */
1149 memset(&cfg, 0, sizeof(cfg));
1150 rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg);
1152 CERROR("failed to start server %s: %d\n",
1153 lsi->lsi_ldd->ldd_svname, rc);
1154 /* Do NOT call server_deregister_mount() here. This makes it
1155 * impossible to find mount later in cleanup time and leaves
1156 * @lsi and othder stuff leaked. -umka */
1161 /* Release the mgc fs for others to use */
1162 if (lsi->lsi_srv_mnt)
1163 server_mgc_clear_fs(lsi->lsi_mgc);
1166 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1168 CERROR("no server named %s was started\n",
1169 lsi->lsi_ldd->ldd_svname);
1173 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1174 (OBP(obd, iocontrol))) {
1175 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1176 obd->obd_self_export, 0, NULL, NULL);
1179 server_notify_target(sb, obd);
1181 /* calculate recovery timeout, do it after lustre_process_log */
1182 server_calc_timeout(lsi, obd);
1184 /* log has been fully processed */
1185 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1191 /***************** lustre superblock **************/
1193 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1195 struct lustre_sb_info *lsi;
1201 OBD_ALLOC_PTR(lsi->lsi_lmd);
1202 if (!lsi->lsi_lmd) {
1207 lsi->lsi_lmd->lmd_exclude_count = 0;
1208 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
1209 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
1210 s2lsi_nocast(sb) = lsi;
1211 /* we take 1 extra ref for our setup */
1212 cfs_atomic_set(&lsi->lsi_mounts, 1);
1214 /* Default umount style */
1215 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1220 static int lustre_free_lsi(struct super_block *sb)
1222 struct lustre_sb_info *lsi = s2lsi(sb);
1225 LASSERT(lsi != NULL);
1226 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
1228 /* someone didn't call server_put_mount. */
1229 LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0);
1231 if (lsi->lsi_ldd != NULL)
1232 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
1234 if (lsi->lsi_lmd != NULL) {
1235 if (lsi->lsi_lmd->lmd_dev != NULL)
1236 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1237 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1238 if (lsi->lsi_lmd->lmd_profile != NULL)
1239 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1240 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1241 if (lsi->lsi_lmd->lmd_mgssec != NULL)
1242 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
1243 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
1244 if (lsi->lsi_lmd->lmd_opts != NULL)
1245 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1246 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1247 if (lsi->lsi_lmd->lmd_exclude_count)
1248 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1249 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1250 lsi->lsi_lmd->lmd_exclude_count);
1251 if (lsi->lsi_lmd->lmd_mgs != NULL)
1252 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
1253 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
1254 if (lsi->lsi_lmd->lmd_osd_type != NULL)
1255 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
1256 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
1257 if (lsi->lsi_lmd->lmd_params != NULL)
1258 OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
1260 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1263 LASSERT(lsi->lsi_llsbi == NULL);
1264 OBD_FREE(lsi, sizeof(*lsi));
1265 s2lsi_nocast(sb) = NULL;
1270 /* The lsi has one reference for every server that is using the disk -
1271 e.g. MDT, MGS, and potentially MGC */
1272 static int lustre_put_lsi(struct super_block *sb)
1274 struct lustre_sb_info *lsi = s2lsi(sb);
1277 LASSERT(lsi != NULL);
1279 CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts));
1280 if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) {
1281 lustre_free_lsi(sb);
1287 static int lsi_prepare(struct lustre_sb_info *lsi)
1289 struct lustre_disk_data *ldd;
1295 LASSERT(lsi->lsi_lmd);
1297 OBD_ALLOC(ldd, sizeof(*ldd));
1301 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_NAME);
1302 ldd->ldd_mount_type = LDD_MT_LDISKFS;
1304 /* The server name is given as a mount line option */
1305 if (lsi->lsi_lmd->lmd_profile == NULL) {
1306 LCONSOLE_ERROR("Can't determine server name\n");
1307 GOTO(err, rc = -EINVAL);
1310 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(ldd->ldd_svname))
1311 GOTO(err, rc = -ENAMETOOLONG);
1313 strcpy(ldd->ldd_svname, lsi->lsi_lmd->lmd_profile);
1315 /* Determine osd type */
1316 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
1317 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
1318 sizeof(lsi->lsi_osd_type))
1319 GOTO(err, rc = -ENAMETOOLONG);
1321 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
1324 if ((p = strstr(ldd->ldd_svname, "-OST"))) {
1325 ldd->ldd_flags = LDD_F_SV_TYPE_OST;
1326 } else if ((p = strstr(ldd->ldd_svname, "-MDT"))) {
1327 ldd->ldd_flags = LDD_F_SV_TYPE_MDT;
1329 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
1331 GOTO(err, rc = -EINVAL);
1334 len = p - ldd->ldd_svname;
1335 if (len >= MTI_NAME_MAXLEN)
1336 GOTO(err, rc = -ENAMETOOLONG);
1337 memcpy(ldd->ldd_fsname, ldd->ldd_svname, len);
1338 ldd->ldd_fsname[len] = '\0';
1340 ldd->ldd_svindex = simple_strtoul(p + 4, NULL, 16);
1341 ldd->ldd_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
1342 LDD_F_WRITECONF : 0;
1343 ldd->ldd_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
1344 LDD_F_VIRGIN | LDD_F_UPDATE : 0;
1345 ldd->ldd_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
1346 LDD_F_SV_TYPE_MGS : 0;
1347 ldd->ldd_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
1348 LDD_F_NO_PRIMNODE : 0;
1352 /* Add mount line flags that used to be in ldd:
1353 * writeconf, mgs, iam, anything else?
1356 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
1357 LDD_F_WRITECONF : 0;
1358 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
1359 LDD_F_SV_TYPE_MGS : 0;
1360 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_IAM) ?
1362 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
1363 LDD_F_NO_PRIMNODE : 0;
1369 OBD_FREE(ldd, sizeof(*ldd));
1373 /*************** server mount ******************/
1375 /** Kernel mount using mount options in MOUNT_DATA_FILE.
1376 * Since this file lives on the disk, we pre-mount using a common
1377 * type, read the file, then re-mount using the type specified in the
1380 static struct vfsmount *server_kernel_mount(struct super_block *sb)
1382 struct lustre_sb_info *lsi = s2lsi(sb);
1383 struct lustre_disk_data *ldd;
1384 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1385 struct vfsmount *mnt;
1386 struct file_system_type *type;
1387 char *options = NULL;
1388 unsigned long page, s_flags;
1389 struct page *__page;
1394 rc = lsi_prepare(lsi);
1396 RETURN(ERR_PTR(rc));
1398 if (strcmp(lmd->lmd_osd_type, "osd-ldiskfs") == 0) {
1399 /* with ldiskfs we're still mounting in the kernel space */
1400 OBD_FREE(lmd->lmd_osd_type,
1401 strlen(lmd->lmd_osd_type) + 1);
1402 lmd->lmd_osd_type = NULL;
1404 /* non-ldiskfs backends (zfs) do mounting internally */
1410 /* In the past, we have always used flags = 0.
1411 Note ext3/ldiskfs can't be mounted ro. */
1412 s_flags = sb->s_flags;
1414 /* allocate memory for options */
1415 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
1417 GOTO(out_free, rc = -ENOMEM);
1418 page = (unsigned long)cfs_page_address(__page);
1419 options = (char *)page;
1420 memset(options, 0, CFS_PAGE_SIZE);
1422 /* Glom up mount options */
1423 memset(options, 0, CFS_PAGE_SIZE);
1424 strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2);
1426 len = CFS_PAGE_SIZE - strlen(options) - 2;
1428 strcat(options, ",");
1429 strncat(options, "no_mbcache", len);
1431 /* Add in any mount-line options */
1432 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) {
1433 len = CFS_PAGE_SIZE - strlen(options) - 2;
1434 strcat(options, ",");
1435 strncat(options, lmd->lmd_opts, len);
1438 /* Special permanent mount flags */
1440 s_flags |= MS_NOATIME | MS_NODIRATIME;
1442 CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
1443 MT_STR(ldd), lmd->lmd_dev, options);
1444 type = get_fs_type(MT_STR(ldd));
1446 CERROR("get_fs_type failed\n");
1447 GOTO(out_free, rc = -ENODEV);
1449 mnt = vfs_kern_mount(type, s_flags, lmd->lmd_dev, (void *)options);
1450 cfs_module_put(type->owner);
1453 CERROR("vfs_kern_mount failed: rc = %d\n", rc);
1457 if (lmd->lmd_flags & LMD_FLG_ABORT_RECOV)
1458 simple_truncate(mnt->mnt_sb->s_root, mnt, LAST_RCVD,
1461 OBD_PAGE_FREE(__page);
1462 lsi->lsi_ldd = ldd; /* freed at lsi cleanup */
1463 CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
1468 OBD_PAGE_FREE(__page);
1469 OBD_FREE(ldd, sizeof(*ldd));
1470 lsi->lsi_ldd = NULL;
1471 RETURN(ERR_PTR(rc));
1474 /** Wait here forever until the mount refcount is 0 before completing umount,
1475 * else we risk dereferencing a null pointer.
1476 * LNET may take e.g. 165s before killing zombies.
1478 static void server_wait_finished(struct vfsmount *mnt)
1482 cfs_sigset_t blocked;
1485 cfs_waitq_init(&waitq);
1486 cfs_waitq_wait_event_interruptible_timeout(waitq, 0,
1487 cfs_time_seconds(3), rc);
1492 cfs_waitq_init(&waitq);
1494 while (mnt_get_count(mnt) > 1) {
1495 if (waited && (waited % 30 == 0))
1496 LCONSOLE_WARN("Mount still busy with %d refs after "
1500 /* Cannot use l_event_wait() for an interruptible sleep. */
1502 blocked = cfs_block_sigsinv(sigmask(SIGKILL));
1503 cfs_waitq_wait_event_interruptible_timeout(
1505 (mnt_get_count(mnt) == 1),
1506 cfs_time_seconds(3),
1508 cfs_restore_sigs(blocked);
1510 LCONSOLE_EMERG("Danger: interrupted umount %s with "
1511 "%d refs!\n", mnt_get_devname(mnt),
1512 mnt_get_count(mnt));
1519 /** Start the shutdown of servers at umount.
1521 static void server_put_super(struct super_block *sb)
1523 struct lustre_sb_info *lsi = s2lsi(sb);
1524 struct obd_device *obd;
1525 struct vfsmount *mnt = lsi->lsi_srv_mnt;
1526 char *tmpname, *extraname = NULL;
1528 int lddflags = lsi->lsi_ldd->ldd_flags;
1529 int lsiflags = lsi->lsi_flags;
1532 LASSERT(lsiflags & LSI_SERVER);
1534 tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1;
1535 OBD_ALLOC(tmpname, tmpname_sz);
1536 memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
1537 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1538 if (IS_MDT(lsi->lsi_ldd) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1539 snprintf(tmpname, tmpname_sz, "MGS");
1541 /* Stop the target */
1542 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1543 (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd))) {
1544 struct lustre_profile *lprof = NULL;
1546 /* tell the mgc to drop the config log */
1547 lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL);
1549 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1550 If there are any setup/cleanup errors, save the lov
1551 name for safety cleanup later. */
1552 lprof = class_get_profile(lsi->lsi_ldd->ldd_svname);
1553 if (lprof && lprof->lp_dt) {
1554 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1555 strcpy(extraname, lprof->lp_dt);
1558 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1560 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1561 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1563 /* We can't seem to give an error return code
1564 * to .put_super, so we better make sure we clean up! */
1566 class_manual_cleanup(obd);
1568 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
1569 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1573 /* If they wanted the mgs to stop separately from the mdt, they
1574 should have put it on a different device. */
1575 if (IS_MGS(lsi->lsi_ldd)) {
1576 /* if MDS start with --nomgs, don't stop MGS then */
1577 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1578 server_stop_mgs(sb);
1581 /* Clean the mgc and sb */
1582 lustre_common_put_super(sb);
1584 /* Wait for the targets to really clean up - can't exit (and let the
1585 sb get destroyed) while the mount is still in use */
1586 server_wait_finished(mnt);
1588 /* drop the One True Mount */
1592 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1593 until the target is really gone so that our type refcount check
1595 server_stop_servers(lddflags, lsiflags);
1597 /* In case of startup or cleanup err, stop related obds */
1599 obd = class_name2obd(extraname);
1601 CWARN("Cleaning orphaned obd %s\n", extraname);
1603 class_manual_cleanup(obd);
1605 OBD_FREE(extraname, strlen(extraname) + 1);
1608 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1609 OBD_FREE(tmpname, tmpname_sz);
1613 /** Called only for 'umount -f'
1615 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1616 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1618 struct super_block *sb = vfsmnt->mnt_sb;
1620 static void server_umount_begin(struct super_block *sb)
1623 struct lustre_sb_info *lsi = s2lsi(sb);
1626 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1627 if (!(flags & MNT_FORCE)) {
1633 CDEBUG(D_MOUNT, "umount -f\n");
1634 /* umount = failover
1636 no third way to do non-force, non-failover */
1637 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1638 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1642 static int server_statfs (struct dentry *dentry, cfs_kstatfs_t *buf)
1644 struct super_block *sb = dentry->d_sb;
1645 struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt;
1648 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) {
1649 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf);
1651 buf->f_type = sb->s_magic;
1657 buf->f_type = sb->s_magic;
1658 buf->f_bsize = sb->s_blocksize;
1664 buf->f_namelen = NAME_MAX;
1668 /** The operations we support directly on the superblock:
1669 * mount, umount, and df.
1671 static struct super_operations server_ops =
1673 .put_super = server_put_super,
1674 .umount_begin = server_umount_begin, /* umount -f */
1675 .statfs = server_statfs,
1678 #define log2(n) cfs_ffz(~(n))
1679 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1681 static int server_fill_super_common(struct super_block *sb)
1683 struct inode *root = 0;
1686 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1688 sb->s_blocksize = 4096;
1689 sb->s_blocksize_bits = log2(sb->s_blocksize);
1690 sb->s_magic = LUSTRE_SUPER_MAGIC;
1691 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1692 sb->s_flags |= MS_RDONLY;
1693 sb->s_op = &server_ops;
1695 root = new_inode(sb);
1697 CERROR("Can't make root inode\n");
1701 /* returns -EIO for every operation */
1702 /* make_bad_inode(root); -- badness - can't umount */
1703 /* apparently we need to be a directory for the mount to finish */
1704 root->i_mode = S_IFDIR;
1706 sb->s_root = d_alloc_root(root);
1708 CERROR("Can't make root dentry\n");
1716 /** Fill in the superblock info for a Lustre server.
1717 * Mount the device with the correct options.
1718 * Read the on-disk config file.
1719 * Start the services.
1721 static int server_fill_super(struct super_block *sb)
1723 struct lustre_sb_info *lsi = s2lsi(sb);
1724 struct vfsmount *mnt;
1728 /* the One True Mount */
1729 mnt = server_kernel_mount(sb);
1732 CERROR("Unable to mount device %s: %d\n",
1733 lsi->lsi_lmd->lmd_dev, rc);
1737 lsi->lsi_srv_mnt = mnt;
1739 LASSERT(lsi->lsi_ldd);
1740 CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n",
1741 lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname,
1742 lsi->lsi_lmd->lmd_dev);
1744 if (class_name2obd(lsi->lsi_ldd->ldd_svname)) {
1745 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1746 "running. Double-mount may have compromised"
1747 " the disk journal.\n",
1748 lsi->lsi_ldd->ldd_svname);
1754 /* Start MGS before MGC */
1755 if (IS_MGS(lsi->lsi_ldd) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
1756 rc = server_start_mgs(sb);
1761 /* Start MGC before servers */
1762 rc = lustre_start_mgc(sb);
1766 /* Set up all obd devices for service */
1767 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1768 (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) {
1769 rc = server_start_targets(sb, mnt);
1771 CERROR("Unable to start targets: %d\n", rc);
1774 /* FIXME overmount client here,
1775 or can we just start a client log and client_fill_super on this sb?
1776 We need to make sure server_put_super gets called too - ll_put_super
1777 calls lustre_common_put_super; check there for LSI_SERVER flag,
1779 Probably should start client from new thread so we can return.
1780 Client will not finish until all servers are connected.
1781 Note - MGS-only server does NOT get a client, since there is no
1782 lustre fs associated - the MGS is for all lustre fs's */
1785 rc = server_fill_super_common(sb);
1791 /* We jump here in case of failure while starting targets or MGS.
1792 * In this case we can't just put @mnt and have to do real cleanup
1793 * with stoping targets, etc. */
1794 server_put_super(sb);
1798 /* Get the index from the obd name.
1799 rc = server type, or
1801 if endptr isn't NULL it is set to end of name */
1802 int server_name2index(char *svname, __u32 *idx, char **endptr)
1804 unsigned long index;
1806 char *dash = strrchr(svname, '-');
1810 /* intepret <fsname>-MDTXXXXX-mdc as mdt, the better way is to pass
1811 * in the fsname, then determine the server index */
1812 if (!strcmp(LUSTRE_MDC_NAME, dash + 1)) {
1814 for (; dash > svname && *dash != '-'; dash--);
1819 if (strncmp(dash + 1, "MDT", 3) == 0)
1820 rc = LDD_F_SV_TYPE_MDT;
1821 else if (strncmp(dash + 1, "OST", 3) == 0)
1822 rc = LDD_F_SV_TYPE_OST;
1825 if (strcmp(dash + 4, "all") == 0)
1826 return rc | LDD_F_SV_ALL;
1828 index = simple_strtoul(dash + 4, endptr, 16);
1832 EXPORT_SYMBOL(server_name2index);
1835 * Calculate timeout value for a target.
1837 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1839 struct lustre_mount_data *lmd;
1843 bool has_ir = !!(lsi->lsi_flags & LSI_IR_CAPABLE);
1844 int min = OBD_RECOVERY_TIME_MIN;
1846 LASSERT(lsi->lsi_flags & LSI_SERVER);
1850 soft = lmd->lmd_recovery_time_soft;
1851 hard = lmd->lmd_recovery_time_hard;
1852 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1853 obd->obd_no_ir = !has_ir;
1857 soft = OBD_RECOVERY_TIME_SOFT;
1859 hard = OBD_RECOVERY_TIME_HARD;
1861 /* target may have ir_factor configured. */
1862 factor = OBD_IR_FACTOR_DEFAULT;
1863 if (obd->obd_recovery_ir_factor)
1864 factor = obd->obd_recovery_ir_factor;
1867 int new_soft = soft;
1868 int new_hard = hard;
1870 /* adjust timeout value by imperative recovery */
1872 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1873 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1875 /* make sure the timeout is not too short */
1876 new_soft = max(min, new_soft);
1877 new_hard = max(new_soft, new_hard);
1879 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1880 "window shrunk from %d-%d down to %d-%d\n",
1881 obd->obd_name, soft, hard, new_soft, new_hard);
1888 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1889 obd->obd_recovery_time_hard = hard;
1890 obd->obd_recovery_ir_factor = factor;
1892 EXPORT_SYMBOL(server_calc_timeout);
1894 /*************** mount common betweeen server and client ***************/
1897 int lustre_common_put_super(struct super_block *sb)
1902 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1904 /* Drop a ref to the MGC */
1905 rc = lustre_stop_mgc(sb);
1906 if (rc && (rc != -ENOENT)) {
1908 CERROR("Can't stop MGC: %d\n", rc);
1911 /* BUSY just means that there's some other obd that
1912 needs the mgc. Let him clean it up. */
1913 CDEBUG(D_MOUNT, "MGC still in use\n");
1915 /* Drop a ref to the mounted disk */
1920 EXPORT_SYMBOL(lustre_common_put_super);
1922 static void lmd_print(struct lustre_mount_data *lmd)
1926 PRINT_CMD(PRINT_MASK, " mount data:\n");
1927 if (lmd_is_client(lmd))
1928 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
1929 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
1930 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
1933 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
1935 if (lmd->lmd_recovery_time_soft)
1936 PRINT_CMD(PRINT_MASK, "recovery time soft: %d\n",
1937 lmd->lmd_recovery_time_soft);
1939 if (lmd->lmd_recovery_time_hard)
1940 PRINT_CMD(PRINT_MASK, "recovery time hard: %d\n",
1941 lmd->lmd_recovery_time_hard);
1943 for (i = 0; i < lmd->lmd_exclude_count; i++) {
1944 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
1945 lmd->lmd_exclude[i]);
1949 /* Is this server on the exclusion list */
1950 int lustre_check_exclusion(struct super_block *sb, char *svname)
1952 struct lustre_sb_info *lsi = s2lsi(sb);
1953 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1958 rc = server_name2index(svname, &index, NULL);
1959 if (rc != LDD_F_SV_TYPE_OST)
1960 /* Only exclude OSTs */
1963 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
1964 index, lmd->lmd_exclude_count, lmd->lmd_dev);
1966 for(i = 0; i < lmd->lmd_exclude_count; i++) {
1967 if (index == lmd->lmd_exclude[i]) {
1968 CWARN("Excluding %s (on exclusion list)\n", svname);
1975 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
1976 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
1978 char *s1 = ptr, *s2;
1979 __u32 index, *exclude_list;
1983 /* The shortest an ost name can be is 8 chars: -OST0000.
1984 We don't actually know the fsname at this time, so in fact
1985 a user could specify any fsname. */
1986 devmax = strlen(ptr) / 8 + 1;
1988 /* temp storage until we figure out how many we have */
1989 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
1993 /* we enter this fn pointing at the '=' */
1994 while (*s1 && *s1 != ' ' && *s1 != ',') {
1996 rc = server_name2index(s1, &index, &s2);
1998 CERROR("Can't parse server name '%s'\n", s1);
2001 if (rc == LDD_F_SV_TYPE_OST)
2002 exclude_list[lmd->lmd_exclude_count++] = index;
2004 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
2006 /* now we are pointing at ':' (next exclude)
2007 or ',' (end of excludes) */
2008 if (lmd->lmd_exclude_count >= devmax)
2011 if (rc >= 0) /* non-err */
2014 if (lmd->lmd_exclude_count) {
2015 /* permanent, freed in lustre_free_lsi */
2016 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
2017 lmd->lmd_exclude_count);
2018 if (lmd->lmd_exclude) {
2019 memcpy(lmd->lmd_exclude, exclude_list,
2020 sizeof(index) * lmd->lmd_exclude_count);
2023 lmd->lmd_exclude_count = 0;
2026 OBD_FREE(exclude_list, sizeof(index) * devmax);
2030 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
2035 if (lmd->lmd_mgssec != NULL) {
2036 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
2037 lmd->lmd_mgssec = NULL;
2040 tail = strchr(ptr, ',');
2042 length = strlen(ptr);
2044 length = tail - ptr;
2046 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
2047 if (lmd->lmd_mgssec == NULL)
2050 memcpy(lmd->lmd_mgssec, ptr, length);
2051 lmd->lmd_mgssec[length] = '\0';
2055 static int lmd_parse_string(char **handle, char *ptr)
2060 if ((handle == NULL) || (ptr == NULL))
2063 if (*handle != NULL) {
2064 OBD_FREE(*handle, strlen(*handle) + 1);
2068 tail = strchr(ptr, ',');
2070 length = strlen(ptr);
2072 length = tail - ptr;
2074 OBD_ALLOC(*handle, length + 1);
2075 if (*handle == NULL)
2078 memcpy(*handle, ptr, length);
2079 (*handle)[length] = '\0';
2084 /* Collect multiple values for mgsnid specifiers */
2085 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
2093 /* Find end of nidlist */
2094 while (class_parse_nid(tail, &nid, &tail) == 0) {}
2095 length = tail - *ptr;
2097 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
2101 if (lmd->lmd_mgs != NULL)
2102 oldlen = strlen(lmd->lmd_mgs) + 1;
2104 OBD_ALLOC(mgsnid, oldlen + length + 1);
2108 if (lmd->lmd_mgs != NULL) {
2109 /* Multiple mgsnid= are taken to mean failover locations */
2110 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
2111 mgsnid[oldlen - 1] = ':';
2112 OBD_FREE(lmd->lmd_mgs, oldlen);
2114 memcpy(mgsnid + oldlen, *ptr, length);
2115 mgsnid[oldlen + length] = '\0';
2116 lmd->lmd_mgs = mgsnid;
2122 /** Parse mount line options
2123 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
2124 * dev is passed as device=uml1:/lustre by mount.lustre
2126 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
2128 char *s1, *s2, *devname = NULL;
2129 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
2135 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
2136 "/sbin/mount.lustre is installed.\n");
2140 /* Options should be a string - try to detect old lmd data */
2141 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
2142 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
2143 "/sbin/mount.lustre. Please install "
2144 "version %s\n", LUSTRE_VERSION_STRING);
2147 lmd->lmd_magic = LMD_MAGIC;
2149 OBD_ALLOC(lmd->lmd_params, 4096);
2150 if (lmd->lmd_params == NULL)
2152 lmd->lmd_params[0] = '\0';
2154 /* Set default flags here */
2159 int time_min = OBD_RECOVERY_TIME_MIN;
2161 /* Skip whitespace and extra commas */
2162 while (*s1 == ' ' || *s1 == ',')
2165 /* Client options are parsed in ll_options: eg. flock,
2168 /* Parse non-ldiskfs options here. Rather than modifying
2169 ldiskfs, we just zero these out here */
2170 if (strncmp(s1, "abort_recov", 11) == 0) {
2171 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
2173 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
2174 lmd->lmd_recovery_time_soft = max_t(int,
2175 simple_strtoul(s1 + 19, NULL, 10), time_min);
2177 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
2178 lmd->lmd_recovery_time_hard = max_t(int,
2179 simple_strtoul(s1 + 19, NULL, 10), time_min);
2181 } else if (strncmp(s1, "noir", 4) == 0) {
2182 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
2184 } else if (strncmp(s1, "nosvc", 5) == 0) {
2185 lmd->lmd_flags |= LMD_FLG_NOSVC;
2187 } else if (strncmp(s1, "nomgs", 5) == 0) {
2188 lmd->lmd_flags |= LMD_FLG_NOMGS;
2190 } else if (strncmp(s1, "noscrub", 7) == 0) {
2191 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
2193 } else if (strncmp(s1, PARAM_MGSNODE,
2194 sizeof(PARAM_MGSNODE) - 1) == 0) {
2195 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
2196 /* Assume the next mount opt is the first
2197 invalid nid we get to. */
2198 rc = lmd_parse_mgs(lmd, &s2);
2202 } else if (strncmp(s1, "writeconf", 9) == 0) {
2203 lmd->lmd_flags |= LMD_FLG_WRITECONF;
2205 } else if (strncmp(s1, "virgin", 6) == 0) {
2206 lmd->lmd_flags |= LMD_FLG_VIRGIN;
2208 } else if (strncmp(s1, "noprimnode", 10) == 0) {
2209 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
2211 } else if (strncmp(s1, "mgssec=", 7) == 0) {
2212 rc = lmd_parse_mgssec(lmd, s1 + 7);
2216 /* ost exclusion list */
2217 } else if (strncmp(s1, "exclude=", 8) == 0) {
2218 rc = lmd_make_exclusion(lmd, s1 + 7);
2222 } else if (strncmp(s1, "mgs", 3) == 0) {
2224 lmd->lmd_flags |= LMD_FLG_MGS;
2226 } else if (strncmp(s1, "svname=", 7) == 0) {
2227 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
2231 } else if (strncmp(s1, "param=", 6) == 0) {
2233 char *tail = strchr(s1 + 6, ',');
2235 length = strlen(s1);
2239 strncat(lmd->lmd_params, s1 + 6, length);
2240 strcat(lmd->lmd_params, " ");
2242 } else if (strncmp(s1, "osd=", 4) == 0) {
2243 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
2248 /* Linux 2.4 doesn't pass the device, so we stuck it at the
2249 end of the options. */
2250 else if (strncmp(s1, "device=", 7) == 0) {
2252 /* terminate options right before device. device
2253 must be the last one. */
2259 s2 = strchr(s1, ',');
2267 memmove(s1, s2, strlen(s2) + 1);
2273 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
2274 "(need mount option 'device=...')\n");
2278 s1 = strstr(devname, ":/");
2281 lmd->lmd_flags |= LMD_FLG_CLIENT;
2282 /* Remove leading /s from fsname */
2283 while (*++s1 == '/') ;
2284 /* Freed in lustre_free_lsi */
2285 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
2286 if (!lmd->lmd_profile)
2288 sprintf(lmd->lmd_profile, "%s-client", s1);
2291 /* Freed in lustre_free_lsi */
2292 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
2295 strcpy(lmd->lmd_dev, devname);
2297 /* Save mount options */
2298 s1 = options + strlen(options) - 1;
2299 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
2301 if (*options != 0) {
2302 /* Freed in lustre_free_lsi */
2303 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
2306 strcpy(lmd->lmd_opts, options);
2310 lmd->lmd_magic = LMD_MAGIC;
2315 CERROR("Bad mount options %s\n", options);
2319 struct lustre_mount_data2 {
2321 struct vfsmount *lmd2_mnt;
2324 /** This is the entry point for the mount call into Lustre.
2325 * This is called when a server or client is mounted,
2326 * and this is where we start setting things up.
2327 * @param data Mount options (e.g. -o flock,abort_recov)
2329 int lustre_fill_super(struct super_block *sb, void *data, int silent)
2331 struct lustre_mount_data *lmd;
2332 struct lustre_mount_data2 *lmd2 = data;
2333 struct lustre_sb_info *lsi;
2337 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
2339 lsi = lustre_init_lsi(sb);
2345 * Disable lockdep during mount, because mount locking patterns are
2351 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
2353 obd_zombie_barrier();
2355 /* Figure out the lmd from the mount options */
2356 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
2358 GOTO(out, rc = -EINVAL);
2361 if (lmd_is_client(lmd)) {
2362 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
2363 if (!client_fill_super) {
2364 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
2365 "client mount! Is the 'lustre' "
2366 "module loaded?\n");
2370 rc = lustre_start_mgc(sb);
2375 /* Connect and start */
2376 /* (should always be ll_fill_super) */
2377 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
2378 /* c_f_s will call lustre_common_put_super on failure */
2381 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
2382 lsi->lsi_flags |= LSI_SERVER;
2383 rc = server_fill_super(sb);
2384 /* s_f_s calls lustre_start_mgc after the mount because we need
2385 the MGS nids which are stored on disk. Plus, we may
2386 need to start the MGS first. */
2387 /* s_f_s will call server_put_super on failure */
2390 /* If error happens in fill_super() call, @lsi will be killed there.
2391 * This is why we do not put it here. */
2395 CERROR("Unable to mount %s (%d)\n",
2396 s2lsi(sb) ? lmd->lmd_dev : "", rc);
2398 CDEBUG(D_SUPER, "Mount %s complete\n",
2406 /* We can't call ll_fill_super by name because it lives in a module that
2407 must be loaded after this one. */
2408 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
2409 struct vfsmount *mnt))
2411 client_fill_super = cfs;
2413 EXPORT_SYMBOL(lustre_register_client_fill_super);
2415 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
2417 kill_super_cb = cfs;
2419 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2421 /***************** FS registration ******************/
2422 #ifdef HAVE_FSTYPE_MOUNT
2423 struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
2424 const char *devname, void *data)
2426 struct lustre_mount_data2 lmd2 = { data, NULL };
2428 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
2431 int lustre_get_sb(struct file_system_type *fs_type, int flags,
2432 const char *devname, void * data, struct vfsmount *mnt)
2434 struct lustre_mount_data2 lmd2 = { data, mnt };
2436 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
2440 void lustre_kill_super(struct super_block *sb)
2442 struct lustre_sb_info *lsi = s2lsi(sb);
2444 if (kill_super_cb && lsi && !(lsi->lsi_flags & LSI_SERVER))
2445 (*kill_super_cb)(sb);
2447 kill_anon_super(sb);
2450 /** Register the "lustre" fs type
2452 struct file_system_type lustre_fs_type = {
2453 .owner = THIS_MODULE,
2455 #ifdef HAVE_FSTYPE_MOUNT
2456 .mount = lustre_mount,
2458 .get_sb = lustre_get_sb,
2460 .kill_sb = lustre_kill_super,
2461 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
2462 #ifdef FS_HAS_FIEMAP
2465 LL_RENAME_DOES_D_MOVE,
2468 int lustre_register_fs(void)
2470 return register_filesystem(&lustre_fs_type);
2473 int lustre_unregister_fs(void)
2475 return unregister_filesystem(&lustre_fs_type);