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, Intel Corporation.
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 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
48 #include "../include/obd.h"
49 #include "../include/linux/lustre_compat25.h"
50 #include "../include/obd_class.h"
51 #include "../include/lustre/lustre_user.h"
52 #include "../include/lustre_log.h"
53 #include "../include/lustre_disk.h"
54 #include "../include/lustre_param.h"
56 static int (*client_fill_super
)(struct super_block
*sb
,
57 struct vfsmount
*mnt
);
59 static void (*kill_super_cb
)(struct super_block
*sb
);
61 /**************** config llog ********************/
63 /** Get a config log from the MGS and process it.
64 * This func is called for both clients and servers.
65 * Continue to process new statements appended to the logs
66 * (whenever the config lock is revoked) until lustre_end_log
68 * @param sb The superblock is used by the MGC to write to the local copy of
70 * @param logname The name of the llog to replicate from the MGS
71 * @param cfg Since the same mgc may be used to follow multiple config logs
72 * (e.g. ost1, ost2, client), the config_llog_instance keeps the state for
73 * this log, and is added to the mgc's list of logs to follow.
75 int lustre_process_log(struct super_block
*sb
, char *logname
,
76 struct config_llog_instance
*cfg
)
78 struct lustre_cfg
*lcfg
;
79 struct lustre_cfg_bufs
*bufs
;
80 struct lustre_sb_info
*lsi
= s2lsi(sb
);
81 struct obd_device
*mgc
= lsi
->lsi_mgc
;
87 bufs
= kzalloc(sizeof(*bufs
), GFP_NOFS
);
91 /* mgc_process_config */
92 lustre_cfg_bufs_reset(bufs
, mgc
->obd_name
);
93 lustre_cfg_bufs_set_string(bufs
, 1, logname
);
94 lustre_cfg_bufs_set(bufs
, 2, cfg
, sizeof(*cfg
));
95 lustre_cfg_bufs_set(bufs
, 3, &sb
, sizeof(sb
));
96 lcfg
= lustre_cfg_new(LCFG_LOG_START
, bufs
);
97 rc
= obd_process_config(mgc
, sizeof(*lcfg
), lcfg
);
98 lustre_cfg_free(lcfg
);
103 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s' failed from the MGS (%d). Make sure this client and the MGS are running compatible versions of Lustre.\n",
104 mgc
->obd_name
, logname
, rc
);
107 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' failed (%d). This may be the result of communication errors between this node and the MGS, a bad configuration, or other errors. See the syslog for more information.\n",
108 mgc
->obd_name
, logname
,
111 /* class_obd_list(); */
114 EXPORT_SYMBOL(lustre_process_log
);
116 /* Stop watching this config log for updates */
117 int lustre_end_log(struct super_block
*sb
, char *logname
,
118 struct config_llog_instance
*cfg
)
120 struct lustre_cfg
*lcfg
;
121 struct lustre_cfg_bufs bufs
;
122 struct lustre_sb_info
*lsi
= s2lsi(sb
);
123 struct obd_device
*mgc
= lsi
->lsi_mgc
;
129 /* mgc_process_config */
130 lustre_cfg_bufs_reset(&bufs
, mgc
->obd_name
);
131 lustre_cfg_bufs_set_string(&bufs
, 1, logname
);
133 lustre_cfg_bufs_set(&bufs
, 2, cfg
, sizeof(*cfg
));
134 lcfg
= lustre_cfg_new(LCFG_LOG_END
, &bufs
);
135 rc
= obd_process_config(mgc
, sizeof(*lcfg
), lcfg
);
136 lustre_cfg_free(lcfg
);
139 EXPORT_SYMBOL(lustre_end_log
);
141 /**************** obd start *******************/
143 /** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from
144 * lctl (and do for echo cli/srv.
146 int do_lcfg(char *cfgname
, lnet_nid_t nid
, int cmd
,
147 char *s1
, char *s2
, char *s3
, char *s4
)
149 struct lustre_cfg_bufs bufs
;
150 struct lustre_cfg
*lcfg
= NULL
;
153 CDEBUG(D_TRACE
, "lcfg %s %#x %s %s %s %s\n", cfgname
,
154 cmd
, s1
, s2
, s3
, s4
);
156 lustre_cfg_bufs_reset(&bufs
, cfgname
);
158 lustre_cfg_bufs_set_string(&bufs
, 1, s1
);
160 lustre_cfg_bufs_set_string(&bufs
, 2, s2
);
162 lustre_cfg_bufs_set_string(&bufs
, 3, s3
);
164 lustre_cfg_bufs_set_string(&bufs
, 4, s4
);
166 lcfg
= lustre_cfg_new(cmd
, &bufs
);
167 lcfg
->lcfg_nid
= nid
;
168 rc
= class_process_config(lcfg
);
169 lustre_cfg_free(lcfg
);
172 EXPORT_SYMBOL(do_lcfg
);
174 /** Call class_attach and class_setup. These methods in turn call
175 * obd type-specific methods.
177 int lustre_start_simple(char *obdname
, char *type
, char *uuid
,
178 char *s1
, char *s2
, char *s3
, char *s4
)
181 CDEBUG(D_MOUNT
, "Starting obd %s (typ=%s)\n", obdname
, type
);
183 rc
= do_lcfg(obdname
, 0, LCFG_ATTACH
, type
, uuid
, NULL
, NULL
);
185 CERROR("%s attach error %d\n", obdname
, rc
);
188 rc
= do_lcfg(obdname
, 0, LCFG_SETUP
, s1
, s2
, s3
, s4
);
190 CERROR("%s setup error %d\n", obdname
, rc
);
191 do_lcfg(obdname
, 0, LCFG_DETACH
, NULL
, NULL
, NULL
, NULL
);
196 DEFINE_MUTEX(mgc_start_lock
);
198 /** Set up a mgc obd to process startup logs
200 * \param sb [in] super block of the mgc obd
202 * \retval 0 success, otherwise error code
204 int lustre_start_mgc(struct super_block
*sb
)
206 struct obd_connect_data
*data
= NULL
;
207 struct lustre_sb_info
*lsi
= s2lsi(sb
);
208 struct obd_device
*obd
;
209 struct obd_export
*exp
;
210 struct obd_uuid
*uuid
;
213 char *mgcname
= NULL
, *niduuid
= NULL
, *mgssec
= NULL
;
215 int rc
= 0, i
= 0, j
, len
;
217 LASSERT(lsi
->lsi_lmd
);
219 /* Find the first non-lo MGS nid for our MGC name */
220 if (IS_SERVER(lsi
)) {
221 /* mount -o mgsnode=nid */
222 ptr
= lsi
->lsi_lmd
->lmd_mgs
;
223 if (lsi
->lsi_lmd
->lmd_mgs
&&
224 (class_parse_nid(lsi
->lsi_lmd
->lmd_mgs
, &nid
, &ptr
) == 0)) {
226 } else if (IS_MGS(lsi
)) {
227 lnet_process_id_t id
;
228 while ((rc
= LNetGetId(i
++, &id
)) != -ENOENT
) {
229 if (LNET_NETTYP(LNET_NIDNET(id
.nid
)) == LOLND
)
236 } else { /* client */
237 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
238 ptr
= lsi
->lsi_lmd
->lmd_dev
;
239 if (class_parse_nid(ptr
, &nid
, &ptr
) == 0)
243 CERROR("No valid MGS nids found.\n");
247 mutex_lock(&mgc_start_lock
);
249 len
= strlen(LUSTRE_MGC_OBDNAME
) + strlen(libcfs_nid2str(nid
)) + 1;
250 mgcname
= kzalloc(len
, GFP_NOFS
);
251 niduuid
= kzalloc(len
+ 2, GFP_NOFS
);
252 if (!mgcname
|| !niduuid
) {
256 sprintf(mgcname
, "%s%s", LUSTRE_MGC_OBDNAME
, libcfs_nid2str(nid
));
258 mgssec
= lsi
->lsi_lmd
->lmd_mgssec
? lsi
->lsi_lmd
->lmd_mgssec
: "";
260 data
= kzalloc(sizeof(*data
), GFP_NOFS
);
266 obd
= class_name2obd(mgcname
);
267 if (obd
&& !obd
->obd_stopping
) {
270 rc
= obd_set_info_async(NULL
, obd
->obd_self_export
,
271 strlen(KEY_MGSSEC
), KEY_MGSSEC
,
272 strlen(mgssec
), mgssec
, NULL
);
276 /* Re-using an existing MGC */
277 atomic_inc(&obd
->u
.cli
.cl_mgc_refcount
);
279 /* IR compatibility check, only for clients */
280 if (lmd_is_client(lsi
->lsi_lmd
)) {
282 int vallen
= sizeof(*data
);
283 __u32
*flags
= &lsi
->lsi_lmd
->lmd_flags
;
285 rc
= obd_get_info(NULL
, obd
->obd_self_export
,
286 strlen(KEY_CONN_DATA
), KEY_CONN_DATA
,
287 &vallen
, data
, NULL
);
289 has_ir
= OCD_HAS_FLAG(data
, IMP_RECOV
);
290 if (has_ir
^ !(*flags
& LMD_FLG_NOIR
)) {
291 /* LMD_FLG_NOIR is for test purpose only */
293 "Trying to mount a client with IR setting not compatible with current mgc. Force to use current mgc setting that is IR %s.\n",
294 has_ir
? "enabled" : "disabled");
296 *flags
&= ~LMD_FLG_NOIR
;
298 *flags
|= LMD_FLG_NOIR
;
303 /* If we are restarting the MGS, don't try to keep the MGC's
304 old connection, or registration will fail. */
306 CDEBUG(D_MOUNT
, "New MGS with live MGC\n");
310 /* Try all connections, but only once (again).
311 We don't want to block another target from starting
312 (using its local copy of the log), but we do want to connect
313 if at all possible. */
315 CDEBUG(D_MOUNT
, "%s: Set MGC reconnect %d\n", mgcname
,
317 rc
= obd_set_info_async(NULL
, obd
->obd_self_export
,
318 sizeof(KEY_INIT_RECOV_BACKUP
),
319 KEY_INIT_RECOV_BACKUP
,
320 sizeof(recov_bk
), &recov_bk
, NULL
);
325 CDEBUG(D_MOUNT
, "Start MGC '%s'\n", mgcname
);
327 /* Add the primary nids for the MGS */
329 sprintf(niduuid
, "%s_%x", mgcname
, i
);
330 if (IS_SERVER(lsi
)) {
331 ptr
= lsi
->lsi_lmd
->lmd_mgs
;
333 /* Use local nids (including LO) */
334 lnet_process_id_t id
;
335 while ((rc
= LNetGetId(i
++, &id
)) != -ENOENT
) {
336 rc
= do_lcfg(mgcname
, id
.nid
,
337 LCFG_ADD_UUID
, niduuid
,
341 /* Use mgsnode= nids */
342 /* mount -o mgsnode=nid */
343 if (lsi
->lsi_lmd
->lmd_mgs
) {
344 ptr
= lsi
->lsi_lmd
->lmd_mgs
;
345 } else if (class_find_param(ptr
, PARAM_MGSNODE
,
347 CERROR("No MGS nids given.\n");
351 while (class_parse_nid(ptr
, &nid
, &ptr
) == 0) {
352 rc
= do_lcfg(mgcname
, nid
,
353 LCFG_ADD_UUID
, niduuid
,
358 } else { /* client */
359 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
360 ptr
= lsi
->lsi_lmd
->lmd_dev
;
361 while (class_parse_nid(ptr
, &nid
, &ptr
) == 0) {
362 rc
= do_lcfg(mgcname
, nid
,
363 LCFG_ADD_UUID
, niduuid
, NULL
, NULL
, NULL
);
365 /* Stop at the first failover nid */
371 CERROR("No valid MGS nids found.\n");
375 lsi
->lsi_lmd
->lmd_mgs_failnodes
= 1;
377 /* Random uuid for MGC allows easier reconnects */
378 uuid
= kzalloc(sizeof(*uuid
), GFP_NOFS
);
384 ll_generate_random_uuid(uuidc
);
385 class_uuid_unparse(uuidc
, uuid
);
388 rc
= lustre_start_simple(mgcname
, LUSTRE_MGC_NAME
,
389 (char *)uuid
->uuid
, LUSTRE_MGS_OBDNAME
,
390 niduuid
, NULL
, NULL
);
395 /* Add any failover MGS nids */
397 while (ptr
&& ((*ptr
== ':' ||
398 class_find_param(ptr
, PARAM_MGSNODE
, &ptr
) == 0))) {
399 /* New failover node */
400 sprintf(niduuid
, "%s_%x", mgcname
, i
);
402 while (class_parse_nid_quiet(ptr
, &nid
, &ptr
) == 0) {
404 rc
= do_lcfg(mgcname
, nid
,
405 LCFG_ADD_UUID
, niduuid
, NULL
, NULL
, NULL
);
410 rc
= do_lcfg(mgcname
, 0, LCFG_ADD_CONN
,
411 niduuid
, NULL
, NULL
, NULL
);
418 lsi
->lsi_lmd
->lmd_mgs_failnodes
= i
;
420 obd
= class_name2obd(mgcname
);
422 CERROR("Can't find mgcobd %s\n", mgcname
);
427 rc
= obd_set_info_async(NULL
, obd
->obd_self_export
,
428 strlen(KEY_MGSSEC
), KEY_MGSSEC
,
429 strlen(mgssec
), mgssec
, NULL
);
433 /* Keep a refcount of servers/clients who started with "mount",
434 so we know when we can get rid of the mgc. */
435 atomic_set(&obd
->u
.cli
.cl_mgc_refcount
, 1);
437 /* We connect to the MGS at setup, and don't disconnect until cleanup */
438 data
->ocd_connect_flags
= OBD_CONNECT_VERSION
| OBD_CONNECT_AT
|
439 OBD_CONNECT_FULL20
| OBD_CONNECT_IMP_RECOV
|
440 OBD_CONNECT_LVB_TYPE
;
442 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
443 data
->ocd_connect_flags
|= OBD_CONNECT_MNE_SWAB
;
445 #warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
448 if (lmd_is_client(lsi
->lsi_lmd
) &&
449 lsi
->lsi_lmd
->lmd_flags
& LMD_FLG_NOIR
)
450 data
->ocd_connect_flags
&= ~OBD_CONNECT_IMP_RECOV
;
451 data
->ocd_version
= LUSTRE_VERSION_CODE
;
452 rc
= obd_connect(NULL
, &exp
, obd
, &(obd
->obd_uuid
), data
, NULL
);
454 CERROR("connect failed %d\n", rc
);
458 obd
->u
.cli
.cl_mgc_mgsexp
= exp
;
461 /* Keep the mgc info in the sb. Note that many lsi's can point
465 mutex_unlock(&mgc_start_lock
);
473 static int lustre_stop_mgc(struct super_block
*sb
)
475 struct lustre_sb_info
*lsi
= s2lsi(sb
);
476 struct obd_device
*obd
;
477 char *niduuid
= NULL
, *ptr
= NULL
;
478 int i
, rc
= 0, len
= 0;
487 mutex_lock(&mgc_start_lock
);
488 LASSERT(atomic_read(&obd
->u
.cli
.cl_mgc_refcount
) > 0);
489 if (!atomic_dec_and_test(&obd
->u
.cli
.cl_mgc_refcount
)) {
490 /* This is not fatal, every client that stops
491 will call in here. */
492 CDEBUG(D_MOUNT
, "mgc still has %d references.\n",
493 atomic_read(&obd
->u
.cli
.cl_mgc_refcount
));
498 /* The MGC has no recoverable data in any case.
499 * force shutdown set in umount_begin */
500 obd
->obd_no_recov
= 1;
502 if (obd
->u
.cli
.cl_mgc_mgsexp
) {
503 /* An error is not fatal, if we are unable to send the
504 disconnect mgs ping evictor cleans up the export */
505 rc
= obd_disconnect(obd
->u
.cli
.cl_mgc_mgsexp
);
507 CDEBUG(D_MOUNT
, "disconnect failed %d\n", rc
);
510 /* Save the obdname for cleaning the nid uuids, which are
512 len
= strlen(obd
->obd_name
) + 6;
513 niduuid
= kzalloc(len
, GFP_NOFS
);
515 strcpy(niduuid
, obd
->obd_name
);
516 ptr
= niduuid
+ strlen(niduuid
);
519 rc
= class_manual_cleanup(obd
);
523 /* Clean the nid uuids */
529 for (i
= 0; i
< lsi
->lsi_lmd
->lmd_mgs_failnodes
; i
++) {
530 sprintf(ptr
, "_%x", i
);
531 rc
= do_lcfg(LUSTRE_MGC_OBDNAME
, 0, LCFG_DEL_UUID
,
532 niduuid
, NULL
, NULL
, NULL
);
534 CERROR("del MDC UUID %s failed: rc = %d\n",
540 /* class_import_put will get rid of the additional connections */
541 mutex_unlock(&mgc_start_lock
);
545 /***************** lustre superblock **************/
547 struct lustre_sb_info
*lustre_init_lsi(struct super_block
*sb
)
549 struct lustre_sb_info
*lsi
;
551 lsi
= kzalloc(sizeof(*lsi
), GFP_NOFS
);
554 lsi
->lsi_lmd
= kzalloc(sizeof(*lsi
->lsi_lmd
), GFP_NOFS
);
560 lsi
->lsi_lmd
->lmd_exclude_count
= 0;
561 lsi
->lsi_lmd
->lmd_recovery_time_soft
= 0;
562 lsi
->lsi_lmd
->lmd_recovery_time_hard
= 0;
563 s2lsi_nocast(sb
) = lsi
;
564 /* we take 1 extra ref for our setup */
565 atomic_set(&lsi
->lsi_mounts
, 1);
567 /* Default umount style */
568 lsi
->lsi_flags
= LSI_UMOUNT_FAILOVER
;
573 static int lustre_free_lsi(struct super_block
*sb
)
575 struct lustre_sb_info
*lsi
= s2lsi(sb
);
577 LASSERT(lsi
!= NULL
);
578 CDEBUG(D_MOUNT
, "Freeing lsi %p\n", lsi
);
580 /* someone didn't call server_put_mount. */
581 LASSERT(atomic_read(&lsi
->lsi_mounts
) == 0);
583 if (lsi
->lsi_lmd
!= NULL
) {
584 kfree(lsi
->lsi_lmd
->lmd_dev
);
585 kfree(lsi
->lsi_lmd
->lmd_profile
);
586 kfree(lsi
->lsi_lmd
->lmd_mgssec
);
587 kfree(lsi
->lsi_lmd
->lmd_opts
);
588 if (lsi
->lsi_lmd
->lmd_exclude_count
)
589 kfree(lsi
->lsi_lmd
->lmd_exclude
);
590 kfree(lsi
->lsi_lmd
->lmd_mgs
);
591 kfree(lsi
->lsi_lmd
->lmd_osd_type
);
592 kfree(lsi
->lsi_lmd
->lmd_params
);
597 LASSERT(lsi
->lsi_llsbi
== NULL
);
599 s2lsi_nocast(sb
) = NULL
;
604 /* The lsi has one reference for every server that is using the disk -
605 e.g. MDT, MGS, and potentially MGC */
606 int lustre_put_lsi(struct super_block
*sb
)
608 struct lustre_sb_info
*lsi
= s2lsi(sb
);
610 LASSERT(lsi
!= NULL
);
612 CDEBUG(D_MOUNT
, "put %p %d\n", sb
, atomic_read(&lsi
->lsi_mounts
));
613 if (atomic_dec_and_test(&lsi
->lsi_mounts
)) {
614 if (IS_SERVER(lsi
) && lsi
->lsi_osd_exp
) {
615 lu_device_put(&lsi
->lsi_dt_dev
->dd_lu_dev
);
616 lsi
->lsi_osd_exp
->exp_obd
->obd_lvfs_ctxt
.dt
= NULL
;
617 lsi
->lsi_dt_dev
= NULL
;
618 obd_disconnect(lsi
->lsi_osd_exp
);
619 /* wait till OSD is gone */
620 obd_zombie_barrier();
629 * <FSNAME><SEPARATOR><TYPE><INDEX>
630 * FSNAME is between 1 and 8 characters (inclusive).
631 * Excluded characters are '/' and ':'
632 * SEPARATOR is either ':' or '-'
633 * TYPE: "OST", "MDT", etc.
634 * INDEX: Hex representation of the index
637 /** Get the fsname ("lustre") from the server name ("lustre-OST003F").
638 * @param [in] svname server name including type and index
639 * @param [out] fsname Buffer to copy filesystem name prefix into.
640 * Must have at least 'strlen(fsname) + 1' chars.
641 * @param [out] endptr if endptr isn't NULL it is set to end of fsname
644 int server_name2fsname(const char *svname
, char *fsname
, const char **endptr
)
648 dash
= svname
+ strnlen(svname
, 8); /* max fsname length is 8 */
649 for (; dash
> svname
&& *dash
!= '-' && *dash
!= ':'; dash
--)
654 if (fsname
!= NULL
) {
655 strncpy(fsname
, svname
, dash
- svname
);
656 fsname
[dash
- svname
] = '\0';
664 EXPORT_SYMBOL(server_name2fsname
);
667 * Get service name (svname) from string
669 * if endptr isn't NULL it is set to end of fsname *
671 int server_name2svname(const char *label
, char *svname
, const char **endptr
,
677 /* We use server_name2fsname() just for parsing */
678 rc
= server_name2fsname(label
, NULL
, &dash
);
685 if (strlcpy(svname
, dash
+ 1, svsize
) >= svsize
)
690 EXPORT_SYMBOL(server_name2svname
);
693 /* Get the index from the obd name.
696 if endptr isn't NULL it is set to end of name */
697 int server_name2index(const char *svname
, __u32
*idx
, const char **endptr
)
703 /* We use server_name2fsname() just for parsing */
704 rc
= server_name2fsname(svname
, NULL
, &dash
);
710 if (strncmp(dash
, "MDT", 3) == 0)
711 rc
= LDD_F_SV_TYPE_MDT
;
712 else if (strncmp(dash
, "OST", 3) == 0)
713 rc
= LDD_F_SV_TYPE_OST
;
719 if (strncmp(dash
, "all", 3) == 0) {
722 return rc
| LDD_F_SV_ALL
;
725 index
= simple_strtoul(dash
, (char **)endptr
, 16);
729 /* Account for -mdc after index that is possible when specifying mdt */
730 if (endptr
!= NULL
&& strncmp(LUSTRE_MDC_NAME
, *endptr
+ 1,
731 sizeof(LUSTRE_MDC_NAME
)-1) == 0)
732 *endptr
+= sizeof(LUSTRE_MDC_NAME
);
736 EXPORT_SYMBOL(server_name2index
);
738 /*************** mount common between server and client ***************/
741 int lustre_common_put_super(struct super_block
*sb
)
745 CDEBUG(D_MOUNT
, "dropping sb %p\n", sb
);
747 /* Drop a ref to the MGC */
748 rc
= lustre_stop_mgc(sb
);
749 if (rc
&& (rc
!= -ENOENT
)) {
751 CERROR("Can't stop MGC: %d\n", rc
);
754 /* BUSY just means that there's some other obd that
755 needs the mgc. Let him clean it up. */
756 CDEBUG(D_MOUNT
, "MGC still in use\n");
758 /* Drop a ref to the mounted disk */
763 EXPORT_SYMBOL(lustre_common_put_super
);
765 static void lmd_print(struct lustre_mount_data
*lmd
)
769 PRINT_CMD(D_MOUNT
, " mount data:\n");
770 if (lmd_is_client(lmd
))
771 PRINT_CMD(D_MOUNT
, "profile: %s\n", lmd
->lmd_profile
);
772 PRINT_CMD(D_MOUNT
, "device: %s\n", lmd
->lmd_dev
);
773 PRINT_CMD(D_MOUNT
, "flags: %x\n", lmd
->lmd_flags
);
776 PRINT_CMD(D_MOUNT
, "options: %s\n", lmd
->lmd_opts
);
778 if (lmd
->lmd_recovery_time_soft
)
779 PRINT_CMD(D_MOUNT
, "recovery time soft: %d\n",
780 lmd
->lmd_recovery_time_soft
);
782 if (lmd
->lmd_recovery_time_hard
)
783 PRINT_CMD(D_MOUNT
, "recovery time hard: %d\n",
784 lmd
->lmd_recovery_time_hard
);
786 for (i
= 0; i
< lmd
->lmd_exclude_count
; i
++) {
787 PRINT_CMD(D_MOUNT
, "exclude %d: OST%04x\n", i
,
788 lmd
->lmd_exclude
[i
]);
792 /* Is this server on the exclusion list */
793 int lustre_check_exclusion(struct super_block
*sb
, char *svname
)
795 struct lustre_sb_info
*lsi
= s2lsi(sb
);
796 struct lustre_mount_data
*lmd
= lsi
->lsi_lmd
;
800 rc
= server_name2index(svname
, &index
, NULL
);
801 if (rc
!= LDD_F_SV_TYPE_OST
)
802 /* Only exclude OSTs */
805 CDEBUG(D_MOUNT
, "Check exclusion %s (%d) in %d of %s\n", svname
,
806 index
, lmd
->lmd_exclude_count
, lmd
->lmd_dev
);
808 for (i
= 0; i
< lmd
->lmd_exclude_count
; i
++) {
809 if (index
== lmd
->lmd_exclude
[i
]) {
810 CWARN("Excluding %s (on exclusion list)\n", svname
);
817 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
818 static int lmd_make_exclusion(struct lustre_mount_data
*lmd
, const char *ptr
)
820 const char *s1
= ptr
, *s2
;
821 __u32 index
, *exclude_list
;
824 /* The shortest an ost name can be is 8 chars: -OST0000.
825 We don't actually know the fsname at this time, so in fact
826 a user could specify any fsname. */
827 devmax
= strlen(ptr
) / 8 + 1;
829 /* temp storage until we figure out how many we have */
830 exclude_list
= kcalloc(devmax
, sizeof(index
), GFP_NOFS
);
834 /* we enter this fn pointing at the '=' */
835 while (*s1
&& *s1
!= ' ' && *s1
!= ',') {
837 rc
= server_name2index(s1
, &index
, &s2
);
839 CERROR("Can't parse server name '%s': rc = %d\n",
843 if (rc
== LDD_F_SV_TYPE_OST
)
844 exclude_list
[lmd
->lmd_exclude_count
++] = index
;
846 CDEBUG(D_MOUNT
, "ignoring exclude %.*s: type = %#x\n",
847 (uint
)(s2
-s1
), s1
, rc
);
849 /* now we are pointing at ':' (next exclude)
850 or ',' (end of excludes) */
851 if (lmd
->lmd_exclude_count
>= devmax
)
854 if (rc
>= 0) /* non-err */
857 if (lmd
->lmd_exclude_count
) {
858 /* permanent, freed in lustre_free_lsi */
859 lmd
->lmd_exclude
= kcalloc(lmd
->lmd_exclude_count
,
860 sizeof(index
), GFP_NOFS
);
861 if (lmd
->lmd_exclude
) {
862 memcpy(lmd
->lmd_exclude
, exclude_list
,
863 sizeof(index
) * lmd
->lmd_exclude_count
);
866 lmd
->lmd_exclude_count
= 0;
873 static int lmd_parse_mgssec(struct lustre_mount_data
*lmd
, char *ptr
)
878 kfree(lmd
->lmd_mgssec
);
879 lmd
->lmd_mgssec
= NULL
;
881 tail
= strchr(ptr
, ',');
883 length
= strlen(ptr
);
887 lmd
->lmd_mgssec
= kzalloc(length
+ 1, GFP_NOFS
);
888 if (lmd
->lmd_mgssec
== NULL
)
891 memcpy(lmd
->lmd_mgssec
, ptr
, length
);
892 lmd
->lmd_mgssec
[length
] = '\0';
896 static int lmd_parse_string(char **handle
, char *ptr
)
901 if ((handle
== NULL
) || (ptr
== NULL
))
907 tail
= strchr(ptr
, ',');
909 length
= strlen(ptr
);
913 *handle
= kzalloc(length
+ 1, GFP_NOFS
);
917 memcpy(*handle
, ptr
, length
);
918 (*handle
)[length
] = '\0';
923 /* Collect multiple values for mgsnid specifiers */
924 static int lmd_parse_mgs(struct lustre_mount_data
*lmd
, char **ptr
)
932 /* Find end of nidlist */
933 while (class_parse_nid_quiet(tail
, &nid
, &tail
) == 0) {}
934 length
= tail
- *ptr
;
936 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr
);
940 if (lmd
->lmd_mgs
!= NULL
)
941 oldlen
= strlen(lmd
->lmd_mgs
) + 1;
943 mgsnid
= kzalloc(oldlen
+ length
+ 1, GFP_NOFS
);
947 if (lmd
->lmd_mgs
!= NULL
) {
948 /* Multiple mgsnid= are taken to mean failover locations */
949 memcpy(mgsnid
, lmd
->lmd_mgs
, oldlen
);
950 mgsnid
[oldlen
- 1] = ':';
953 memcpy(mgsnid
+ oldlen
, *ptr
, length
);
954 mgsnid
[oldlen
+ length
] = '\0';
955 lmd
->lmd_mgs
= mgsnid
;
961 /** Parse mount line options
962 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
963 * dev is passed as device=uml1:/lustre by mount.lustre
965 static int lmd_parse(char *options
, struct lustre_mount_data
*lmd
)
967 char *s1
, *s2
, *devname
= NULL
;
968 struct lustre_mount_data
*raw
= (struct lustre_mount_data
*)options
;
973 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that /sbin/mount.lustre is installed.\n");
977 /* Options should be a string - try to detect old lmd data */
978 if ((raw
->lmd_magic
& 0xffffff00) == (LMD_MAGIC
& 0xffffff00)) {
979 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of /sbin/mount.lustre. Please install version %s\n",
980 LUSTRE_VERSION_STRING
);
983 lmd
->lmd_magic
= LMD_MAGIC
;
985 lmd
->lmd_params
= kzalloc(4096, GFP_NOFS
);
986 if (lmd
->lmd_params
== NULL
)
988 lmd
->lmd_params
[0] = '\0';
990 /* Set default flags here */
995 int time_min
= OBD_RECOVERY_TIME_MIN
;
997 /* Skip whitespace and extra commas */
998 while (*s1
== ' ' || *s1
== ',')
1001 /* Client options are parsed in ll_options: eg. flock,
1004 /* Parse non-ldiskfs options here. Rather than modifying
1005 ldiskfs, we just zero these out here */
1006 if (strncmp(s1
, "abort_recov", 11) == 0) {
1007 lmd
->lmd_flags
|= LMD_FLG_ABORT_RECOV
;
1009 } else if (strncmp(s1
, "recovery_time_soft=", 19) == 0) {
1010 lmd
->lmd_recovery_time_soft
= max_t(int,
1011 simple_strtoul(s1
+ 19, NULL
, 10), time_min
);
1013 } else if (strncmp(s1
, "recovery_time_hard=", 19) == 0) {
1014 lmd
->lmd_recovery_time_hard
= max_t(int,
1015 simple_strtoul(s1
+ 19, NULL
, 10), time_min
);
1017 } else if (strncmp(s1
, "noir", 4) == 0) {
1018 lmd
->lmd_flags
|= LMD_FLG_NOIR
; /* test purpose only. */
1020 } else if (strncmp(s1
, "nosvc", 5) == 0) {
1021 lmd
->lmd_flags
|= LMD_FLG_NOSVC
;
1023 } else if (strncmp(s1
, "nomgs", 5) == 0) {
1024 lmd
->lmd_flags
|= LMD_FLG_NOMGS
;
1026 } else if (strncmp(s1
, "noscrub", 7) == 0) {
1027 lmd
->lmd_flags
|= LMD_FLG_NOSCRUB
;
1029 } else if (strncmp(s1
, PARAM_MGSNODE
,
1030 sizeof(PARAM_MGSNODE
) - 1) == 0) {
1031 s2
= s1
+ sizeof(PARAM_MGSNODE
) - 1;
1032 /* Assume the next mount opt is the first
1033 invalid nid we get to. */
1034 rc
= lmd_parse_mgs(lmd
, &s2
);
1038 } else if (strncmp(s1
, "writeconf", 9) == 0) {
1039 lmd
->lmd_flags
|= LMD_FLG_WRITECONF
;
1041 } else if (strncmp(s1
, "update", 6) == 0) {
1042 lmd
->lmd_flags
|= LMD_FLG_UPDATE
;
1044 } else if (strncmp(s1
, "virgin", 6) == 0) {
1045 lmd
->lmd_flags
|= LMD_FLG_VIRGIN
;
1047 } else if (strncmp(s1
, "noprimnode", 10) == 0) {
1048 lmd
->lmd_flags
|= LMD_FLG_NO_PRIMNODE
;
1050 } else if (strncmp(s1
, "mgssec=", 7) == 0) {
1051 rc
= lmd_parse_mgssec(lmd
, s1
+ 7);
1055 /* ost exclusion list */
1056 } else if (strncmp(s1
, "exclude=", 8) == 0) {
1057 rc
= lmd_make_exclusion(lmd
, s1
+ 7);
1061 } else if (strncmp(s1
, "mgs", 3) == 0) {
1063 lmd
->lmd_flags
|= LMD_FLG_MGS
;
1065 } else if (strncmp(s1
, "svname=", 7) == 0) {
1066 rc
= lmd_parse_string(&lmd
->lmd_profile
, s1
+ 7);
1070 } else if (strncmp(s1
, "param=", 6) == 0) {
1072 char *tail
= strchr(s1
+ 6, ',');
1074 length
= strlen(s1
);
1078 strncat(lmd
->lmd_params
, s1
+ 6, length
);
1079 strcat(lmd
->lmd_params
, " ");
1081 } else if (strncmp(s1
, "osd=", 4) == 0) {
1082 rc
= lmd_parse_string(&lmd
->lmd_osd_type
, s1
+ 4);
1087 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1088 end of the options. */
1089 else if (strncmp(s1
, "device=", 7) == 0) {
1091 /* terminate options right before device. device
1092 must be the last one. */
1098 s2
= strchr(s1
, ',');
1106 memmove(s1
, s2
, strlen(s2
) + 1);
1112 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name (need mount option 'device=...')\n");
1116 s1
= strstr(devname
, ":/");
1119 lmd
->lmd_flags
|= LMD_FLG_CLIENT
;
1120 /* Remove leading /s from fsname */
1121 while (*++s1
== '/') ;
1122 /* Freed in lustre_free_lsi */
1123 lmd
->lmd_profile
= kzalloc(strlen(s1
) + 8, GFP_NOFS
);
1124 if (!lmd
->lmd_profile
)
1126 sprintf(lmd
->lmd_profile
, "%s-client", s1
);
1129 /* Freed in lustre_free_lsi */
1130 lmd
->lmd_dev
= kzalloc(strlen(devname
) + 1, GFP_NOFS
);
1133 strcpy(lmd
->lmd_dev
, devname
);
1135 /* Save mount options */
1136 s1
= options
+ strlen(options
) - 1;
1137 while (s1
>= options
&& (*s1
== ',' || *s1
== ' '))
1139 if (*options
!= 0) {
1140 /* Freed in lustre_free_lsi */
1141 lmd
->lmd_opts
= kzalloc(strlen(options
) + 1, GFP_NOFS
);
1144 strcpy(lmd
->lmd_opts
, options
);
1148 lmd
->lmd_magic
= LMD_MAGIC
;
1153 CERROR("Bad mount options %s\n", options
);
1157 struct lustre_mount_data2
{
1159 struct vfsmount
*lmd2_mnt
;
1162 /** This is the entry point for the mount call into Lustre.
1163 * This is called when a server or client is mounted,
1164 * and this is where we start setting things up.
1165 * @param data Mount options (e.g. -o flock,abort_recov)
1167 int lustre_fill_super(struct super_block
*sb
, void *data
, int silent
)
1169 struct lustre_mount_data
*lmd
;
1170 struct lustre_mount_data2
*lmd2
= data
;
1171 struct lustre_sb_info
*lsi
;
1174 CDEBUG(D_MOUNT
|D_VFSTRACE
, "VFS Op: sb %p\n", sb
);
1176 lsi
= lustre_init_lsi(sb
);
1182 * Disable lockdep during mount, because mount locking patterns are
1188 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
1190 obd_zombie_barrier();
1192 /* Figure out the lmd from the mount options */
1193 if (lmd_parse((char *)(lmd2
->lmd2_data
), lmd
)) {
1199 if (lmd_is_client(lmd
)) {
1200 CDEBUG(D_MOUNT
, "Mounting client %s\n", lmd
->lmd_profile
);
1201 if (client_fill_super
== NULL
)
1202 request_module("lustre");
1203 if (client_fill_super
== NULL
) {
1204 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for client mount! Is the 'lustre' module loaded?\n");
1208 rc
= lustre_start_mgc(sb
);
1213 /* Connect and start */
1214 /* (should always be ll_fill_super) */
1215 rc
= (*client_fill_super
)(sb
, lmd2
->lmd2_mnt
);
1216 /* c_f_s will call lustre_common_put_super on failure */
1219 CERROR("This is client-side-only module, cannot handle server mount.\n");
1223 /* If error happens in fill_super() call, @lsi will be killed there.
1224 * This is why we do not put it here. */
1228 CERROR("Unable to mount %s (%d)\n",
1229 s2lsi(sb
) ? lmd
->lmd_dev
: "", rc
);
1231 CDEBUG(D_SUPER
, "Mount %s complete\n",
1239 /* We can't call ll_fill_super by name because it lives in a module that
1240 must be loaded after this one. */
1241 void lustre_register_client_fill_super(int (*cfs
)(struct super_block
*sb
,
1242 struct vfsmount
*mnt
))
1244 client_fill_super
= cfs
;
1246 EXPORT_SYMBOL(lustre_register_client_fill_super
);
1248 void lustre_register_kill_super_cb(void (*cfs
)(struct super_block
*sb
))
1250 kill_super_cb
= cfs
;
1252 EXPORT_SYMBOL(lustre_register_kill_super_cb
);
1254 /***************** FS registration ******************/
1255 struct dentry
*lustre_mount(struct file_system_type
*fs_type
, int flags
,
1256 const char *devname
, void *data
)
1258 struct lustre_mount_data2 lmd2
= {
1263 return mount_nodev(fs_type
, flags
, &lmd2
, lustre_fill_super
);
1266 static void lustre_kill_super(struct super_block
*sb
)
1268 struct lustre_sb_info
*lsi
= s2lsi(sb
);
1270 if (kill_super_cb
&& lsi
&& !IS_SERVER(lsi
))
1271 (*kill_super_cb
)(sb
);
1273 kill_anon_super(sb
);
1276 /** Register the "lustre" fs type
1278 struct file_system_type lustre_fs_type
= {
1279 .owner
= THIS_MODULE
,
1281 .mount
= lustre_mount
,
1282 .kill_sb
= lustre_kill_super
,
1283 .fs_flags
= FS_BINARY_MOUNTDATA
| FS_REQUIRES_DEV
|
1284 FS_HAS_FIEMAP
| FS_RENAME_DOES_D_MOVE
,
1286 MODULE_ALIAS_FS("lustre");
1288 int lustre_register_fs(void)
1290 return register_filesystem(&lustre_fs_type
);
1293 int lustre_unregister_fs(void)
1295 return unregister_filesystem(&lustre_fs_type
);