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, 2013, 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
49 #include <obd_class.h>
50 #include <lustre/lustre_user.h>
51 #include <linux/version.h>
52 #include <lustre_log.h>
53 #include <lustre_disk.h>
54 #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;
92 /* mgc_process_config */
93 lustre_cfg_bufs_reset(bufs, mgc->obd_name);
94 lustre_cfg_bufs_set_string(bufs, 1, logname);
95 lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg));
96 lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb));
97 lcfg = lustre_cfg_new(LCFG_LOG_START, bufs);
99 GOTO(out, rc = -ENOMEM);
100 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
101 lustre_cfg_free(lcfg);
106 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
107 "failed from the MGS (%d). Make sure this "
108 "client and the MGS are running compatible "
109 "versions of Lustre.\n",
110 mgc->obd_name, logname, rc);
112 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
113 "failed (%d). This may be the result of "
114 "communication errors between this node and "
115 "the MGS, a bad configuration, or other "
116 "errors. See the syslog for more "
117 "information.\n", mgc->obd_name, logname,
120 /* class_obd_list(); */
123 EXPORT_SYMBOL(lustre_process_log);
125 /* Stop watching this config log for updates */
126 int lustre_end_log(struct super_block *sb, char *logname,
127 struct config_llog_instance *cfg)
129 struct lustre_cfg *lcfg;
130 struct lustre_cfg_bufs bufs;
131 struct lustre_sb_info *lsi = s2lsi(sb);
132 struct obd_device *mgc = lsi->lsi_mgc;
139 /* mgc_process_config */
140 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
141 lustre_cfg_bufs_set_string(&bufs, 1, logname);
143 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
144 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
147 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
148 lustre_cfg_free(lcfg);
151 EXPORT_SYMBOL(lustre_end_log);
153 /**************** obd start *******************/
155 /** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from
156 * lctl (and do for echo cli/srv.
158 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
159 char *s1, char *s2, char *s3, char *s4)
161 struct lustre_cfg_bufs bufs;
162 struct lustre_cfg * lcfg = NULL;
165 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
166 cmd, s1, s2, s3, s4);
168 lustre_cfg_bufs_reset(&bufs, cfgname);
170 lustre_cfg_bufs_set_string(&bufs, 1, s1);
172 lustre_cfg_bufs_set_string(&bufs, 2, s2);
174 lustre_cfg_bufs_set_string(&bufs, 3, s3);
176 lustre_cfg_bufs_set_string(&bufs, 4, s4);
178 lcfg = lustre_cfg_new(cmd, &bufs);
181 lcfg->lcfg_nid = nid;
182 rc = class_process_config(lcfg);
183 lustre_cfg_free(lcfg);
186 EXPORT_SYMBOL(do_lcfg);
188 /** Call class_attach and class_setup. These methods in turn call
189 * obd type-specific methods.
191 int lustre_start_simple(char *obdname, char *type, char *uuid,
192 char *s1, char *s2, char *s3, char *s4)
195 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
197 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
199 CERROR("%s attach error %d\n", obdname, rc);
202 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, s3, s4);
204 CERROR("%s setup error %d\n", obdname, rc);
205 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
210 DEFINE_MUTEX(mgc_start_lock);
212 /** Set up a mgc obd to process startup logs
214 * \param sb [in] super block of the mgc obd
216 * \retval 0 success, otherwise error code
218 int lustre_start_mgc(struct super_block *sb)
220 struct obd_connect_data *data = NULL;
221 struct lustre_sb_info *lsi = s2lsi(sb);
222 struct obd_device *obd;
223 struct obd_export *exp;
224 struct obd_uuid *uuid;
227 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
229 int rc = 0, i = 0, j, len;
232 LASSERT(lsi->lsi_lmd);
234 /* Find the first non-lo MGS nid for our MGC name */
235 if (IS_SERVER(lsi)) {
236 /* mount -o mgsnode=nid */
237 ptr = lsi->lsi_lmd->lmd_mgs;
238 if (lsi->lsi_lmd->lmd_mgs &&
239 (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) {
241 } else if (IS_MGS(lsi)) {
242 lnet_process_id_t id;
243 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
244 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
251 } else { /* client */
252 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
253 ptr = lsi->lsi_lmd->lmd_dev;
254 if (class_parse_nid(ptr, &nid, &ptr) == 0)
258 CERROR("No valid MGS nids found.\n");
262 mutex_lock(&mgc_start_lock);
264 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
265 OBD_ALLOC(mgcname, len);
266 OBD_ALLOC(niduuid, len + 2);
267 if (!mgcname || !niduuid)
268 GOTO(out_free, rc = -ENOMEM);
269 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
271 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
275 GOTO(out_free, rc = -ENOMEM);
277 obd = class_name2obd(mgcname);
278 if (obd && !obd->obd_stopping) {
281 rc = obd_set_info_async(NULL, obd->obd_self_export,
282 strlen(KEY_MGSSEC), KEY_MGSSEC,
283 strlen(mgssec), mgssec, NULL);
287 /* Re-using an existing MGC */
288 atomic_inc(&obd->u.cli.cl_mgc_refcount);
290 /* IR compatibility check, only for clients */
291 if (lmd_is_client(lsi->lsi_lmd)) {
293 int vallen = sizeof(*data);
294 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
296 rc = obd_get_info(NULL, obd->obd_self_export,
297 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
298 &vallen, data, NULL);
300 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
301 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
302 /* LMD_FLG_NOIR is for test purpose only */
304 "Trying to mount a client with IR setting "
305 "not compatible with current mgc. "
306 "Force to use current mgc setting that is "
308 has_ir ? "enabled" : "disabled");
310 *flags &= ~LMD_FLG_NOIR;
312 *flags |= LMD_FLG_NOIR;
317 /* If we are restarting the MGS, don't try to keep the MGC's
318 old connection, or registration will fail. */
320 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
324 /* Try all connections, but only once (again).
325 We don't want to block another target from starting
326 (using its local copy of the log), but we do want to connect
327 if at all possible. */
329 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
330 rc = obd_set_info_async(NULL, obd->obd_self_export,
331 sizeof(KEY_INIT_RECOV_BACKUP),
332 KEY_INIT_RECOV_BACKUP,
333 sizeof(recov_bk), &recov_bk, NULL);
337 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
339 /* Add the primary nids for the MGS */
341 sprintf(niduuid, "%s_%x", mgcname, i);
342 if (IS_SERVER(lsi)) {
343 ptr = lsi->lsi_lmd->lmd_mgs;
344 CDEBUG(D_MOUNT, "mgs nids %s.\n", ptr);
346 /* Use local nids (including LO) */
347 lnet_process_id_t id;
348 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
349 rc = do_lcfg(mgcname, id.nid, LCFG_ADD_UUID,
353 /* Use mgsnode= nids */
354 /* mount -o mgsnode=nid */
355 if (lsi->lsi_lmd->lmd_mgs) {
356 ptr = lsi->lsi_lmd->lmd_mgs;
357 } else if (class_find_param(ptr, PARAM_MGSNODE,
359 CERROR("No MGS nids given.\n");
360 GOTO(out_free, rc = -EINVAL);
364 * Here we only take the first mgsnid as its primary
365 * serving mgs node, the rest mgsnid will be taken as
366 * failover mgs node, otherwise they would be takens
367 * as multiple nids of a single mgs node.
369 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
370 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
378 } else { /* client */
379 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
380 ptr = lsi->lsi_lmd->lmd_dev;
381 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
382 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
386 /* Stop at the first failover nid */
392 CERROR("No valid MGS nids found.\n");
393 GOTO(out_free, rc = -EINVAL);
395 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
397 /* Random uuid for MGC allows easier reconnects */
399 ll_generate_random_uuid(uuidc);
400 class_uuid_unparse(uuidc, uuid);
403 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
404 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
410 /* Add any failover MGS nids */
412 while (ptr && ((*ptr == ':' ||
413 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) {
414 /* New failover node */
415 sprintf(niduuid, "%s_%x", mgcname, i);
417 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
418 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
426 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
435 lsi->lsi_lmd->lmd_mgs_failnodes = i;
437 obd = class_name2obd(mgcname);
439 CERROR("Can't find mgcobd %s\n", mgcname);
440 GOTO(out_free, rc = -ENOTCONN);
443 rc = obd_set_info_async(NULL, obd->obd_self_export,
444 strlen(KEY_MGSSEC), KEY_MGSSEC,
445 strlen(mgssec), mgssec, NULL);
449 /* Keep a refcount of servers/clients who started with "mount",
450 so we know when we can get rid of the mgc. */
451 atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
453 /* We connect to the MGS at setup, and don't disconnect until cleanup */
454 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
455 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV |
456 OBD_CONNECT_LVB_TYPE;
458 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
459 data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB;
461 #warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
464 if (lmd_is_client(lsi->lsi_lmd) &&
465 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
466 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
467 data->ocd_version = LUSTRE_VERSION_CODE;
468 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
470 CERROR("connect failed %d\n", rc);
474 obd->u.cli.cl_mgc_mgsexp = exp;
477 /* Keep the mgc info in the sb. Note that many lsi's can point
481 mutex_unlock(&mgc_start_lock);
486 OBD_FREE(mgcname, len);
488 OBD_FREE(niduuid, len + 2);
492 static int lustre_stop_mgc(struct super_block *sb)
494 struct lustre_sb_info *lsi = s2lsi(sb);
495 struct obd_device *obd;
496 char *niduuid = 0, *ptr = 0;
497 int i, rc = 0, len = 0;
507 mutex_lock(&mgc_start_lock);
508 LASSERT(atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
509 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
510 /* This is not fatal, every client that stops
511 will call in here. */
512 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
513 atomic_read(&obd->u.cli.cl_mgc_refcount));
514 GOTO(out, rc = -EBUSY);
517 /* The MGC has no recoverable data in any case.
518 * force shotdown set in umount_begin */
519 obd->obd_no_recov = 1;
521 if (obd->u.cli.cl_mgc_mgsexp) {
522 /* An error is not fatal, if we are unable to send the
523 disconnect mgs ping evictor cleans up the export */
524 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
526 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
529 /* Save the obdname for cleaning the nid uuids, which are
531 len = strlen(obd->obd_name) + 6;
532 OBD_ALLOC(niduuid, len);
534 strcpy(niduuid, obd->obd_name);
535 ptr = niduuid + strlen(niduuid);
538 rc = class_manual_cleanup(obd);
542 /* Clean the nid uuids */
544 GOTO(out, rc = -ENOMEM);
546 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
547 sprintf(ptr, "_%x", i);
548 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
551 CERROR("del MDC UUID %s failed: rc = %d\n",
556 OBD_FREE(niduuid, len);
558 /* class_import_put will get rid of the additional connections */
559 mutex_unlock(&mgc_start_lock);
563 /***************** lustre superblock **************/
565 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
567 struct lustre_sb_info *lsi;
573 OBD_ALLOC_PTR(lsi->lsi_lmd);
579 lsi->lsi_lmd->lmd_exclude_count = 0;
580 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
581 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
582 s2lsi_nocast(sb) = lsi;
583 /* we take 1 extra ref for our setup */
584 atomic_set(&lsi->lsi_mounts, 1);
586 /* Default umount style */
587 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
588 INIT_LIST_HEAD(&lsi->lsi_lwp_list);
589 spin_lock_init(&lsi->lsi_lwp_lock);
594 static int lustre_free_lsi(struct super_block *sb)
596 struct lustre_sb_info *lsi = s2lsi(sb);
599 LASSERT(lsi != NULL);
600 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
602 /* someone didn't call server_put_mount. */
603 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
605 if (lsi->lsi_lmd != NULL) {
606 if (lsi->lsi_lmd->lmd_dev != NULL)
607 OBD_FREE(lsi->lsi_lmd->lmd_dev,
608 strlen(lsi->lsi_lmd->lmd_dev) + 1);
609 if (lsi->lsi_lmd->lmd_profile != NULL)
610 OBD_FREE(lsi->lsi_lmd->lmd_profile,
611 strlen(lsi->lsi_lmd->lmd_profile) + 1);
612 if (lsi->lsi_lmd->lmd_mgssec != NULL)
613 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
614 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
615 if (lsi->lsi_lmd->lmd_opts != NULL)
616 OBD_FREE(lsi->lsi_lmd->lmd_opts,
617 strlen(lsi->lsi_lmd->lmd_opts) + 1);
618 if (lsi->lsi_lmd->lmd_exclude_count)
619 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
620 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
621 lsi->lsi_lmd->lmd_exclude_count);
622 if (lsi->lsi_lmd->lmd_mgs != NULL)
623 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
624 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
625 if (lsi->lsi_lmd->lmd_osd_type != NULL)
626 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
627 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
628 if (lsi->lsi_lmd->lmd_params != NULL)
629 OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
631 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
634 LASSERT(lsi->lsi_llsbi == NULL);
635 OBD_FREE(lsi, sizeof(*lsi));
636 s2lsi_nocast(sb) = NULL;
641 /* The lsi has one reference for every server that is using the disk -
642 e.g. MDT, MGS, and potentially MGC */
643 int lustre_put_lsi(struct super_block *sb)
645 struct lustre_sb_info *lsi = s2lsi(sb);
648 LASSERT(lsi != NULL);
650 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
651 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
652 if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
653 lu_device_put(&lsi->lsi_dt_dev->dd_lu_dev);
654 lsi->lsi_osd_exp->exp_obd->obd_lvfs_ctxt.dt = NULL;
655 lsi->lsi_dt_dev = NULL;
656 obd_disconnect(lsi->lsi_osd_exp);
657 /* wait till OSD is gone */
658 obd_zombie_barrier();
667 * <FSNAME><SEPERATOR><TYPE><INDEX>
668 * FSNAME is between 1 and 8 characters (inclusive).
669 * Excluded characters are '/' and ':'
670 * SEPERATOR is either ':' or '-'
671 * TYPE: "OST", "MDT", etc.
672 * INDEX: Hex representation of the index
675 /** Get the fsname ("lustre") from the server name ("lustre-OST003F").
676 * @param [in] svname server name including type and index
677 * @param [out] fsname Buffer to copy filesystem name prefix into.
678 * Must have at least 'strlen(fsname) + 1' chars.
679 * @param [out] endptr if endptr isn't NULL it is set to end of fsname
682 int server_name2fsname(const char *svname, char *fsname, const char **endptr)
686 dash = svname + strnlen(svname, 8); /* max fsname length is 8 */
687 for (; dash > svname && *dash != '-' && *dash != ':'; dash--)
692 if (fsname != NULL) {
693 strncpy(fsname, svname, dash - svname);
694 fsname[dash - svname] = '\0';
702 EXPORT_SYMBOL(server_name2fsname);
705 * Get service name (svname) from string
707 * if endptr isn't NULL it is set to end of fsname *
709 int server_name2svname(const char *label, char *svname, const char **endptr,
715 /* We use server_name2fsname() just for parsing */
716 rc = server_name2fsname(label, NULL, &dash);
723 if (strlcpy(svname, dash + 1, svsize) >= svsize)
728 EXPORT_SYMBOL(server_name2svname);
731 * check server name is OST.
733 int server_name_is_ost(const char *svname)
738 /* We use server_name2fsname() just for parsing */
739 rc = server_name2fsname(svname, NULL, &dash);
745 if (strncmp(dash, "OST", 3) == 0)
749 EXPORT_SYMBOL(server_name_is_ost);
752 * Get the index from the target name MDTXXXX/OSTXXXX
753 * rc = server type, or rc < 0 on error
755 int target_name2index(const char *tgtname, __u32 *idx, const char **endptr)
757 const char *dash = tgtname;
761 if (strncmp(dash, "MDT", 3) == 0)
762 rc = LDD_F_SV_TYPE_MDT;
763 else if (strncmp(dash, "OST", 3) == 0)
764 rc = LDD_F_SV_TYPE_OST;
770 if (strncmp(dash, "all", 3) == 0) {
773 return rc | LDD_F_SV_ALL;
776 index = simple_strtoul(dash, (char **)endptr, 16);
781 EXPORT_SYMBOL(target_name2index);
783 /* Get the index from the obd name.
786 if endptr isn't NULL it is set to end of name */
787 int server_name2index(const char *svname, __u32 *idx, const char **endptr)
792 /* We use server_name2fsname() just for parsing */
793 rc = server_name2fsname(svname, NULL, &dash);
798 rc = target_name2index(dash, idx, endptr);
802 /* Account for -mdc after index that is possible when specifying mdt */
803 if (endptr != NULL && strncmp(LUSTRE_MDC_NAME, *endptr + 1,
804 sizeof(LUSTRE_MDC_NAME)-1) == 0)
805 *endptr += sizeof(LUSTRE_MDC_NAME);
809 EXPORT_SYMBOL(server_name2index);
811 /*************** mount common betweeen server and client ***************/
814 int lustre_common_put_super(struct super_block *sb)
819 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
821 /* Drop a ref to the MGC */
822 rc = lustre_stop_mgc(sb);
823 if (rc && (rc != -ENOENT)) {
825 CERROR("Can't stop MGC: %d\n", rc);
828 /* BUSY just means that there's some other obd that
829 needs the mgc. Let him clean it up. */
830 CDEBUG(D_MOUNT, "MGC still in use\n");
832 /* Drop a ref to the mounted disk */
837 EXPORT_SYMBOL(lustre_common_put_super);
839 static void lmd_print(struct lustre_mount_data *lmd)
843 PRINT_CMD(D_MOUNT, " mount data:\n");
844 if (lmd_is_client(lmd))
845 PRINT_CMD(D_MOUNT, "profile: %s\n", lmd->lmd_profile);
846 PRINT_CMD(D_MOUNT, "device: %s\n", lmd->lmd_dev);
847 PRINT_CMD(D_MOUNT, "flags: %x\n", lmd->lmd_flags);
850 PRINT_CMD(D_MOUNT, "options: %s\n", lmd->lmd_opts);
852 if (lmd->lmd_recovery_time_soft)
853 PRINT_CMD(D_MOUNT, "recovery time soft: %d\n",
854 lmd->lmd_recovery_time_soft);
856 if (lmd->lmd_recovery_time_hard)
857 PRINT_CMD(D_MOUNT, "recovery time hard: %d\n",
858 lmd->lmd_recovery_time_hard);
860 for (i = 0; i < lmd->lmd_exclude_count; i++) {
861 PRINT_CMD(D_MOUNT, "exclude %d: OST%04x\n", i,
862 lmd->lmd_exclude[i]);
866 /* Is this server on the exclusion list */
867 int lustre_check_exclusion(struct super_block *sb, char *svname)
869 struct lustre_sb_info *lsi = s2lsi(sb);
870 struct lustre_mount_data *lmd = lsi->lsi_lmd;
875 rc = server_name2index(svname, &index, NULL);
876 if (rc != LDD_F_SV_TYPE_OST)
877 /* Only exclude OSTs */
880 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
881 index, lmd->lmd_exclude_count, lmd->lmd_dev);
883 for(i = 0; i < lmd->lmd_exclude_count; i++) {
884 if (index == lmd->lmd_exclude[i]) {
885 CWARN("Excluding %s (on exclusion list)\n", svname);
892 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
893 static int lmd_make_exclusion(struct lustre_mount_data *lmd, const char *ptr)
895 const char *s1 = ptr, *s2;
901 /* The shortest an ost name can be is 8 chars: -OST0000.
902 We don't actually know the fsname at this time, so in fact
903 a user could specify any fsname. */
904 devmax = strlen(ptr) / 8 + 1;
906 /* temp storage until we figure out how many we have */
907 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
911 /* we enter this fn pointing at the '=' */
912 while (*s1 && *s1 != ' ' && *s1 != ',') {
914 rc = server_name2index(s1, &index, &s2);
916 CERROR("Can't parse server name '%s': rc = %d\n",
920 if (rc == LDD_F_SV_TYPE_OST)
921 exclude_list[lmd->lmd_exclude_count++] = index;
923 CDEBUG(D_MOUNT, "ignoring exclude %.*s: type = %#x\n",
924 (uint)(s2-s1), s1, rc);
926 /* now we are pointing at ':' (next exclude)
927 or ',' (end of excludes) */
928 if (lmd->lmd_exclude_count >= devmax)
931 if (rc >= 0) /* non-err */
934 if (lmd->lmd_exclude_count) {
935 /* permanent, freed in lustre_free_lsi */
936 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
937 lmd->lmd_exclude_count);
938 if (lmd->lmd_exclude) {
939 memcpy(lmd->lmd_exclude, exclude_list,
940 sizeof(index) * lmd->lmd_exclude_count);
943 lmd->lmd_exclude_count = 0;
946 OBD_FREE(exclude_list, sizeof(index) * devmax);
950 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
955 if (lmd->lmd_mgssec != NULL) {
956 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
957 lmd->lmd_mgssec = NULL;
960 tail = strchr(ptr, ',');
962 length = strlen(ptr);
966 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
967 if (lmd->lmd_mgssec == NULL)
970 memcpy(lmd->lmd_mgssec, ptr, length);
971 lmd->lmd_mgssec[length] = '\0';
975 static int lmd_parse_string(char **handle, char *ptr)
980 if ((handle == NULL) || (ptr == NULL))
983 if (*handle != NULL) {
984 OBD_FREE(*handle, strlen(*handle) + 1);
988 tail = strchr(ptr, ',');
990 length = strlen(ptr);
994 OBD_ALLOC(*handle, length + 1);
998 memcpy(*handle, ptr, length);
999 (*handle)[length] = '\0';
1004 /* Collect multiple values for mgsnid specifiers */
1005 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
1013 /* Find end of nidlist */
1014 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
1015 length = tail - *ptr;
1017 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
1021 if (lmd->lmd_mgs != NULL)
1022 oldlen = strlen(lmd->lmd_mgs) + 1;
1024 OBD_ALLOC(mgsnid, oldlen + length + 1);
1028 if (lmd->lmd_mgs != NULL) {
1029 /* Multiple mgsnid= are taken to mean failover locations */
1030 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
1031 mgsnid[oldlen - 1] = ':';
1032 OBD_FREE(lmd->lmd_mgs, oldlen);
1034 memcpy(mgsnid + oldlen, *ptr, length);
1035 mgsnid[oldlen + length] = '\0';
1036 lmd->lmd_mgs = mgsnid;
1042 /** Parse mount line options
1043 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
1044 * dev is passed as device=uml1:/lustre by mount.lustre
1046 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1048 char *s1, *s2, *s3, *devname = NULL;
1049 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1055 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1056 "/sbin/mount.lustre is installed.\n");
1060 /* Options should be a string - try to detect old lmd data */
1061 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1062 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1063 "/sbin/mount.lustre. Please install "
1064 "version %s\n", LUSTRE_VERSION_STRING);
1067 lmd->lmd_magic = LMD_MAGIC;
1069 OBD_ALLOC(lmd->lmd_params, 4096);
1070 if (lmd->lmd_params == NULL)
1072 lmd->lmd_params[0] = '\0';
1074 /* Set default flags here */
1079 int time_min = OBD_RECOVERY_TIME_MIN;
1081 /* Skip whitespace and extra commas */
1082 while (*s1 == ' ' || *s1 == ',')
1086 /* Client options are parsed in ll_options: eg. flock,
1089 /* Parse non-ldiskfs options here. Rather than modifying
1090 ldiskfs, we just zero these out here */
1091 if (strncmp(s1, "abort_recov", 11) == 0) {
1092 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1094 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
1095 lmd->lmd_recovery_time_soft = max_t(int,
1096 simple_strtoul(s1 + 19, NULL, 10), time_min);
1098 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
1099 lmd->lmd_recovery_time_hard = max_t(int,
1100 simple_strtoul(s1 + 19, NULL, 10), time_min);
1102 } else if (strncmp(s1, "noir", 4) == 0) {
1103 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
1105 } else if (strncmp(s1, "nosvc", 5) == 0) {
1106 lmd->lmd_flags |= LMD_FLG_NOSVC;
1108 } else if (strncmp(s1, "nomgs", 5) == 0) {
1109 lmd->lmd_flags |= LMD_FLG_NOMGS;
1111 } else if (strncmp(s1, "noscrub", 7) == 0) {
1112 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
1114 } else if (strncmp(s1, PARAM_MGSNODE,
1115 sizeof(PARAM_MGSNODE) - 1) == 0) {
1116 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
1117 /* Assume the next mount opt is the first
1118 invalid nid we get to. */
1119 rc = lmd_parse_mgs(lmd, &s2);
1124 } else if (strncmp(s1, "writeconf", 9) == 0) {
1125 lmd->lmd_flags |= LMD_FLG_WRITECONF;
1127 } else if (strncmp(s1, "update", 6) == 0) {
1128 lmd->lmd_flags |= LMD_FLG_UPDATE;
1130 } else if (strncmp(s1, "virgin", 6) == 0) {
1131 lmd->lmd_flags |= LMD_FLG_VIRGIN;
1133 } else if (strncmp(s1, "noprimnode", 10) == 0) {
1134 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
1136 } else if (strncmp(s1, "mgssec=", 7) == 0) {
1137 rc = lmd_parse_mgssec(lmd, s1 + 7);
1141 /* ost exclusion list */
1142 } else if (strncmp(s1, "exclude=", 8) == 0) {
1143 rc = lmd_make_exclusion(lmd, s1 + 7);
1147 } else if (strncmp(s1, "mgs", 3) == 0) {
1149 lmd->lmd_flags |= LMD_FLG_MGS;
1151 } else if (strncmp(s1, "svname=", 7) == 0) {
1152 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
1156 } else if (strncmp(s1, "param=", 6) == 0) {
1158 char *tail = strchr(s1 + 6, ',');
1160 length = strlen(s1);
1163 char *param_str = tail + 1;
1164 int supplementary = 1;
1166 while (class_parse_nid_quiet(param_str, &nid,
1170 length = param_str - s1 - supplementary;
1173 strncat(lmd->lmd_params, s1 + 6, length);
1174 strcat(lmd->lmd_params, " ");
1175 s3 = s1 + 6 + length;
1177 } else if (strncmp(s1, "osd=", 4) == 0) {
1178 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
1183 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1184 end of the options. */
1185 else if (strncmp(s1, "device=", 7) == 0) {
1187 /* terminate options right before device. device
1188 must be the last one. */
1194 s2 = strchr(s3, ',');
1202 memmove(s1, s2, strlen(s2) + 1);
1208 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1209 "(need mount option 'device=...')\n");
1213 s1 = strstr(devname, ":/");
1216 lmd->lmd_flags |= LMD_FLG_CLIENT;
1217 /* Remove leading /s from fsname */
1218 while (*++s1 == '/') ;
1219 /* Freed in lustre_free_lsi */
1220 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1221 if (!lmd->lmd_profile)
1223 sprintf(lmd->lmd_profile, "%s-client", s1);
1226 /* Freed in lustre_free_lsi */
1227 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1230 strcpy(lmd->lmd_dev, devname);
1232 /* Save mount options */
1233 s1 = options + strlen(options) - 1;
1234 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1236 if (*options != 0) {
1237 /* Freed in lustre_free_lsi */
1238 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1241 strcpy(lmd->lmd_opts, options);
1245 lmd->lmd_magic = LMD_MAGIC;
1250 CERROR("Bad mount options %s\n", options);
1254 struct lustre_mount_data2 {
1256 struct vfsmount *lmd2_mnt;
1259 /** This is the entry point for the mount call into Lustre.
1260 * This is called when a server or client is mounted,
1261 * and this is where we start setting things up.
1262 * @param data Mount options (e.g. -o flock,abort_recov)
1264 int lustre_fill_super(struct super_block *sb, void *data, int silent)
1266 struct lustre_mount_data *lmd;
1267 struct lustre_mount_data2 *lmd2 = data;
1268 struct lustre_sb_info *lsi;
1272 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1274 lsi = lustre_init_lsi(sb);
1280 * Disable lockdep during mount, because mount locking patterns are
1286 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
1288 obd_zombie_barrier();
1290 /* Figure out the lmd from the mount options */
1291 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
1293 GOTO(out, rc = -EINVAL);
1296 if (lmd_is_client(lmd)) {
1297 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1298 if (client_fill_super == NULL)
1299 request_module("lustre");
1300 if (client_fill_super == NULL) {
1301 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1302 "client mount! Is the 'lustre' "
1303 "module loaded?\n");
1307 rc = lustre_start_mgc(sb);
1312 /* Connect and start */
1313 /* (should always be ll_fill_super) */
1314 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
1315 /* c_f_s will call lustre_common_put_super on failure */
1318 #ifdef HAVE_SERVER_SUPPORT
1319 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1320 rc = server_fill_super(sb);
1321 /* s_f_s calls lustre_start_mgc after the mount because we need
1322 the MGS nids which are stored on disk. Plus, we may
1323 need to start the MGS first. */
1324 /* s_f_s will call server_put_super on failure */
1326 CERROR("This is client-side-only module, "
1327 "cannot handle server mount.\n");
1332 /* If error happens in fill_super() call, @lsi will be killed there.
1333 * This is why we do not put it here. */
1337 CERROR("Unable to mount %s (%d)\n",
1338 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1340 CDEBUG(D_SUPER, "Mount %s complete\n",
1348 /* We can't call ll_fill_super by name because it lives in a module that
1349 must be loaded after this one. */
1350 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
1351 struct vfsmount *mnt))
1353 client_fill_super = cfs;
1355 EXPORT_SYMBOL(lustre_register_client_fill_super);
1357 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
1359 kill_super_cb = cfs;
1361 EXPORT_SYMBOL(lustre_register_kill_super_cb);
1363 /***************** FS registration ******************/
1364 #ifdef HAVE_FSTYPE_MOUNT
1365 struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
1366 const char *devname, void *data)
1368 struct lustre_mount_data2 lmd2 = { data, NULL };
1370 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
1373 int lustre_get_sb(struct file_system_type *fs_type, int flags,
1374 const char *devname, void * data, struct vfsmount *mnt)
1376 struct lustre_mount_data2 lmd2 = { data, mnt };
1378 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
1382 void lustre_kill_super(struct super_block *sb)
1384 struct lustre_sb_info *lsi = s2lsi(sb);
1386 if (kill_super_cb && lsi && !IS_SERVER(lsi))
1387 (*kill_super_cb)(sb);
1389 kill_anon_super(sb);
1392 /** Register the "lustre" fs type
1394 struct file_system_type lustre_fs_type = {
1395 .owner = THIS_MODULE,
1397 #ifdef HAVE_FSTYPE_MOUNT
1398 .mount = lustre_mount,
1400 .get_sb = lustre_get_sb,
1402 .kill_sb = lustre_kill_super,
1403 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
1404 FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE,
1406 MODULE_ALIAS_FS("lustre");
1408 int lustre_register_fs(void)
1410 return register_filesystem(&lustre_fs_type);
1413 int lustre_unregister_fs(void)
1415 return unregister_filesystem(&lustre_fs_type);