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, 2014, 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, NULL, NULL);
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, NULL, NULL, NULL, NULL);
210 static 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,
350 niduuid, NULL, NULL, NULL);
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,
371 niduuid, NULL, NULL, NULL);
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,
383 niduuid, NULL, NULL, NULL);
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,
405 niduuid, NULL, NULL);
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,
419 niduuid, NULL, NULL, NULL);
426 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
427 niduuid, NULL, NULL, NULL);
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, 0, 53, 0)
459 data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB;
462 if (lmd_is_client(lsi->lsi_lmd) &&
463 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
464 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
465 data->ocd_version = LUSTRE_VERSION_CODE;
466 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
468 CERROR("connect failed %d\n", rc);
472 obd->u.cli.cl_mgc_mgsexp = exp;
475 /* Keep the mgc info in the sb. Note that many lsi's can point
479 mutex_unlock(&mgc_start_lock);
484 OBD_FREE(mgcname, len);
486 OBD_FREE(niduuid, len + 2);
490 static int lustre_stop_mgc(struct super_block *sb)
492 struct lustre_sb_info *lsi = s2lsi(sb);
493 struct obd_device *obd;
494 char *niduuid = NULL, *ptr = NULL;
495 int i, rc = 0, len = 0;
505 mutex_lock(&mgc_start_lock);
506 LASSERT(atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
507 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
508 /* This is not fatal, every client that stops
509 will call in here. */
510 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
511 atomic_read(&obd->u.cli.cl_mgc_refcount));
512 GOTO(out, rc = -EBUSY);
515 /* The MGC has no recoverable data in any case.
516 * force shotdown set in umount_begin */
517 obd->obd_no_recov = 1;
519 if (obd->u.cli.cl_mgc_mgsexp) {
520 /* An error is not fatal, if we are unable to send the
521 disconnect mgs ping evictor cleans up the export */
522 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
524 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
527 /* Save the obdname for cleaning the nid uuids, which are
529 len = strlen(obd->obd_name) + 6;
530 OBD_ALLOC(niduuid, len);
532 strcpy(niduuid, obd->obd_name);
533 ptr = niduuid + strlen(niduuid);
536 rc = class_manual_cleanup(obd);
540 /* Clean the nid uuids */
542 GOTO(out, rc = -ENOMEM);
544 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
545 sprintf(ptr, "_%x", i);
546 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
547 niduuid, NULL, NULL, NULL);
549 CERROR("del MDC UUID %s failed: rc = %d\n",
554 OBD_FREE(niduuid, len);
556 /* class_import_put will get rid of the additional connections */
557 mutex_unlock(&mgc_start_lock);
561 /***************** lustre superblock **************/
563 static struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
565 struct lustre_sb_info *lsi;
571 OBD_ALLOC_PTR(lsi->lsi_lmd);
577 lsi->lsi_lmd->lmd_exclude_count = 0;
578 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
579 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
580 s2lsi_nocast(sb) = lsi;
581 /* we take 1 extra ref for our setup */
582 atomic_set(&lsi->lsi_mounts, 1);
584 /* Default umount style */
585 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
586 INIT_LIST_HEAD(&lsi->lsi_lwp_list);
587 spin_lock_init(&lsi->lsi_lwp_lock);
592 static int lustre_free_lsi(struct super_block *sb)
594 struct lustre_sb_info *lsi = s2lsi(sb);
597 LASSERT(lsi != NULL);
598 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
600 /* someone didn't call server_put_mount. */
601 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
603 if (lsi->lsi_lmd != NULL) {
604 if (lsi->lsi_lmd->lmd_dev != NULL)
605 OBD_FREE(lsi->lsi_lmd->lmd_dev,
606 strlen(lsi->lsi_lmd->lmd_dev) + 1);
607 if (lsi->lsi_lmd->lmd_profile != NULL)
608 OBD_FREE(lsi->lsi_lmd->lmd_profile,
609 strlen(lsi->lsi_lmd->lmd_profile) + 1);
610 if (lsi->lsi_lmd->lmd_mgssec != NULL)
611 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
612 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
613 if (lsi->lsi_lmd->lmd_opts != NULL)
614 OBD_FREE(lsi->lsi_lmd->lmd_opts,
615 strlen(lsi->lsi_lmd->lmd_opts) + 1);
616 if (lsi->lsi_lmd->lmd_exclude_count)
617 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
618 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
619 lsi->lsi_lmd->lmd_exclude_count);
620 if (lsi->lsi_lmd->lmd_mgs != NULL)
621 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
622 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
623 if (lsi->lsi_lmd->lmd_osd_type != NULL)
624 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
625 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
626 if (lsi->lsi_lmd->lmd_params != NULL)
627 OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
629 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
632 LASSERT(lsi->lsi_llsbi == NULL);
633 OBD_FREE(lsi, sizeof(*lsi));
634 s2lsi_nocast(sb) = NULL;
639 /* The lsi has one reference for every server that is using the disk -
640 e.g. MDT, MGS, and potentially MGC */
641 int lustre_put_lsi(struct super_block *sb)
643 struct lustre_sb_info *lsi = s2lsi(sb);
646 LASSERT(lsi != NULL);
648 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
649 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
650 if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
651 lu_device_put(&lsi->lsi_dt_dev->dd_lu_dev);
652 lsi->lsi_osd_exp->exp_obd->obd_lvfs_ctxt.dt = NULL;
653 lsi->lsi_dt_dev = NULL;
654 obd_disconnect(lsi->lsi_osd_exp);
655 /* wait till OSD is gone */
656 obd_zombie_barrier();
665 * <FSNAME><SEPERATOR><TYPE><INDEX>
666 * FSNAME is between 1 and 8 characters (inclusive).
667 * Excluded characters are '/' and ':'
668 * SEPERATOR is either ':' or '-'
669 * TYPE: "OST", "MDT", etc.
670 * INDEX: Hex representation of the index
673 /** Get the fsname ("lustre") from the server name ("lustre-OST003F").
674 * @param [in] svname server name including type and index
675 * @param [out] fsname Buffer to copy filesystem name prefix into.
676 * Must have at least 'strlen(fsname) + 1' chars.
677 * @param [out] endptr if endptr isn't NULL it is set to end of fsname
680 int server_name2fsname(const char *svname, char *fsname, const char **endptr)
684 dash = svname + strnlen(svname, 8); /* max fsname length is 8 */
685 for (; dash > svname && *dash != '-' && *dash != ':'; dash--)
690 if (fsname != NULL) {
691 strncpy(fsname, svname, dash - svname);
692 fsname[dash - svname] = '\0';
700 EXPORT_SYMBOL(server_name2fsname);
703 * Get service name (svname) from string
705 * if endptr isn't NULL it is set to end of fsname *
707 int server_name2svname(const char *label, char *svname, const char **endptr,
713 /* We use server_name2fsname() just for parsing */
714 rc = server_name2fsname(label, NULL, &dash);
721 if (strlcpy(svname, dash + 1, svsize) >= svsize)
726 EXPORT_SYMBOL(server_name2svname);
729 * check server name is OST.
731 int server_name_is_ost(const char *svname)
736 /* We use server_name2fsname() just for parsing */
737 rc = server_name2fsname(svname, NULL, &dash);
743 if (strncmp(dash, "OST", 3) == 0)
747 EXPORT_SYMBOL(server_name_is_ost);
750 * Get the index from the target name MDTXXXX/OSTXXXX
751 * rc = server type, or rc < 0 on error
753 int target_name2index(const char *tgtname, __u32 *idx, const char **endptr)
755 const char *dash = tgtname;
759 if (strncmp(dash, "MDT", 3) == 0)
760 rc = LDD_F_SV_TYPE_MDT;
761 else if (strncmp(dash, "OST", 3) == 0)
762 rc = LDD_F_SV_TYPE_OST;
768 if (strncmp(dash, "all", 3) == 0) {
771 return rc | LDD_F_SV_ALL;
774 index = simple_strtoul(dash, (char **)endptr, 16);
779 EXPORT_SYMBOL(target_name2index);
781 /* Get the index from the obd name.
784 if endptr isn't NULL it is set to end of name */
785 int server_name2index(const char *svname, __u32 *idx, const char **endptr)
790 /* We use server_name2fsname() just for parsing */
791 rc = server_name2fsname(svname, NULL, &dash);
796 rc = target_name2index(dash, idx, endptr);
800 /* Account for -mdc after index that is possible when specifying mdt */
801 if (endptr != NULL && strncmp(LUSTRE_MDC_NAME, *endptr + 1,
802 sizeof(LUSTRE_MDC_NAME)-1) == 0)
803 *endptr += sizeof(LUSTRE_MDC_NAME);
807 EXPORT_SYMBOL(server_name2index);
809 /*************** mount common betweeen server and client ***************/
812 int lustre_common_put_super(struct super_block *sb)
817 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
819 /* Drop a ref to the MGC */
820 rc = lustre_stop_mgc(sb);
821 if (rc && (rc != -ENOENT)) {
823 CERROR("Can't stop MGC: %d\n", rc);
826 /* BUSY just means that there's some other obd that
827 needs the mgc. Let him clean it up. */
828 CDEBUG(D_MOUNT, "MGC still in use\n");
830 /* Drop a ref to the mounted disk */
835 EXPORT_SYMBOL(lustre_common_put_super);
837 static void lmd_print(struct lustre_mount_data *lmd)
841 PRINT_CMD(D_MOUNT, " mount data:\n");
842 if (lmd_is_client(lmd))
843 PRINT_CMD(D_MOUNT, "profile: %s\n", lmd->lmd_profile);
844 PRINT_CMD(D_MOUNT, "device: %s\n", lmd->lmd_dev);
845 PRINT_CMD(D_MOUNT, "flags: %x\n", lmd->lmd_flags);
848 PRINT_CMD(D_MOUNT, "options: %s\n", lmd->lmd_opts);
850 if (lmd->lmd_recovery_time_soft)
851 PRINT_CMD(D_MOUNT, "recovery time soft: %d\n",
852 lmd->lmd_recovery_time_soft);
854 if (lmd->lmd_recovery_time_hard)
855 PRINT_CMD(D_MOUNT, "recovery time hard: %d\n",
856 lmd->lmd_recovery_time_hard);
858 for (i = 0; i < lmd->lmd_exclude_count; i++) {
859 PRINT_CMD(D_MOUNT, "exclude %d: OST%04x\n", i,
860 lmd->lmd_exclude[i]);
864 /* Is this server on the exclusion list */
865 int lustre_check_exclusion(struct super_block *sb, char *svname)
867 struct lustre_sb_info *lsi = s2lsi(sb);
868 struct lustre_mount_data *lmd = lsi->lsi_lmd;
873 rc = server_name2index(svname, &index, NULL);
874 if (rc != LDD_F_SV_TYPE_OST)
875 /* Only exclude OSTs */
878 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
879 index, lmd->lmd_exclude_count, lmd->lmd_dev);
881 for(i = 0; i < lmd->lmd_exclude_count; i++) {
882 if (index == lmd->lmd_exclude[i]) {
883 CWARN("Excluding %s (on exclusion list)\n", svname);
890 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
891 static int lmd_make_exclusion(struct lustre_mount_data *lmd, const char *ptr)
893 const char *s1 = ptr, *s2;
899 /* The shortest an ost name can be is 8 chars: -OST0000.
900 We don't actually know the fsname at this time, so in fact
901 a user could specify any fsname. */
902 devmax = strlen(ptr) / 8 + 1;
904 /* temp storage until we figure out how many we have */
905 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
909 /* we enter this fn pointing at the '=' */
910 while (*s1 && *s1 != ' ' && *s1 != ',') {
912 rc = server_name2index(s1, &index, &s2);
914 CERROR("Can't parse server name '%s': rc = %d\n",
918 if (rc == LDD_F_SV_TYPE_OST)
919 exclude_list[lmd->lmd_exclude_count++] = index;
921 CDEBUG(D_MOUNT, "ignoring exclude %.*s: type = %#x\n",
922 (uint)(s2-s1), s1, rc);
924 /* now we are pointing at ':' (next exclude)
925 or ',' (end of excludes) */
926 if (lmd->lmd_exclude_count >= devmax)
929 if (rc >= 0) /* non-err */
932 if (lmd->lmd_exclude_count) {
933 /* permanent, freed in lustre_free_lsi */
934 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
935 lmd->lmd_exclude_count);
936 if (lmd->lmd_exclude) {
937 memcpy(lmd->lmd_exclude, exclude_list,
938 sizeof(index) * lmd->lmd_exclude_count);
941 lmd->lmd_exclude_count = 0;
944 OBD_FREE(exclude_list, sizeof(index) * devmax);
948 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
953 if (lmd->lmd_mgssec != NULL) {
954 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
955 lmd->lmd_mgssec = NULL;
958 tail = strchr(ptr, ',');
960 length = strlen(ptr);
964 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
965 if (lmd->lmd_mgssec == NULL)
968 memcpy(lmd->lmd_mgssec, ptr, length);
969 lmd->lmd_mgssec[length] = '\0';
973 static int lmd_parse_string(char **handle, char *ptr)
978 if ((handle == NULL) || (ptr == NULL))
981 if (*handle != NULL) {
982 OBD_FREE(*handle, strlen(*handle) + 1);
986 tail = strchr(ptr, ',');
988 length = strlen(ptr);
992 OBD_ALLOC(*handle, length + 1);
996 memcpy(*handle, ptr, length);
997 (*handle)[length] = '\0';
1002 /* Collect multiple values for mgsnid specifiers */
1003 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
1011 /* Find end of nidlist */
1012 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
1013 length = tail - *ptr;
1015 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
1019 if (lmd->lmd_mgs != NULL)
1020 oldlen = strlen(lmd->lmd_mgs) + 1;
1022 OBD_ALLOC(mgsnid, oldlen + length + 1);
1026 if (lmd->lmd_mgs != NULL) {
1027 /* Multiple mgsnid= are taken to mean failover locations */
1028 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
1029 mgsnid[oldlen - 1] = ':';
1030 OBD_FREE(lmd->lmd_mgs, oldlen);
1032 memcpy(mgsnid + oldlen, *ptr, length);
1033 mgsnid[oldlen + length] = '\0';
1034 lmd->lmd_mgs = mgsnid;
1040 /** Parse mount line options
1041 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
1042 * dev is passed as device=uml1:/lustre by mount.lustre
1044 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1046 char *s1, *s2, *s3, *devname = NULL;
1047 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1053 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1054 "/sbin/mount.lustre is installed.\n");
1058 /* Options should be a string - try to detect old lmd data */
1059 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1060 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1061 "/sbin/mount.lustre. Please install "
1062 "version %s\n", LUSTRE_VERSION_STRING);
1065 lmd->lmd_magic = LMD_MAGIC;
1067 OBD_ALLOC(lmd->lmd_params, LMD_PARAMS_MAXLEN);
1068 if (lmd->lmd_params == NULL)
1070 lmd->lmd_params[0] = '\0';
1072 /* Set default flags here */
1077 int time_min = OBD_RECOVERY_TIME_MIN;
1079 /* Skip whitespace and extra commas */
1080 while (*s1 == ' ' || *s1 == ',')
1084 /* Client options are parsed in ll_options: eg. flock,
1087 /* Parse non-ldiskfs options here. Rather than modifying
1088 ldiskfs, we just zero these out here */
1089 if (strncmp(s1, "abort_recov", 11) == 0) {
1090 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1092 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
1093 lmd->lmd_recovery_time_soft = max_t(int,
1094 simple_strtoul(s1 + 19, NULL, 10), time_min);
1096 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
1097 lmd->lmd_recovery_time_hard = max_t(int,
1098 simple_strtoul(s1 + 19, NULL, 10), time_min);
1100 } else if (strncmp(s1, "noir", 4) == 0) {
1101 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
1103 } else if (strncmp(s1, "nosvc", 5) == 0) {
1104 lmd->lmd_flags |= LMD_FLG_NOSVC;
1106 } else if (strncmp(s1, "nomgs", 5) == 0) {
1107 lmd->lmd_flags |= LMD_FLG_NOMGS;
1109 } else if (strncmp(s1, "noscrub", 7) == 0) {
1110 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
1112 } else if (strncmp(s1, PARAM_MGSNODE,
1113 sizeof(PARAM_MGSNODE) - 1) == 0) {
1114 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
1115 /* Assume the next mount opt is the first
1116 invalid nid we get to. */
1117 rc = lmd_parse_mgs(lmd, &s2);
1122 } else if (strncmp(s1, "writeconf", 9) == 0) {
1123 lmd->lmd_flags |= LMD_FLG_WRITECONF;
1125 } else if (strncmp(s1, "update", 6) == 0) {
1126 lmd->lmd_flags |= LMD_FLG_UPDATE;
1128 } else if (strncmp(s1, "virgin", 6) == 0) {
1129 lmd->lmd_flags |= LMD_FLG_VIRGIN;
1131 } else if (strncmp(s1, "noprimnode", 10) == 0) {
1132 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
1134 } else if (strncmp(s1, "mgssec=", 7) == 0) {
1135 rc = lmd_parse_mgssec(lmd, s1 + 7);
1139 /* ost exclusion list */
1140 } else if (strncmp(s1, "exclude=", 8) == 0) {
1141 rc = lmd_make_exclusion(lmd, s1 + 7);
1145 } else if (strncmp(s1, "mgs", 3) == 0) {
1147 lmd->lmd_flags |= LMD_FLG_MGS;
1149 } else if (strncmp(s1, "svname=", 7) == 0) {
1150 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
1154 } else if (strncmp(s1, "param=", 6) == 0) {
1155 size_t length, params_length;
1156 char *tail = strchr(s1 + 6, ',');
1158 length = strlen(s1);
1161 char *param_str = tail + 1;
1162 int supplementary = 1;
1164 while (class_parse_nid_quiet(param_str, &nid,
1168 length = param_str - s1 - supplementary;
1171 params_length = strlen(lmd->lmd_params);
1172 if (params_length + length + 1 >= LMD_PARAMS_MAXLEN)
1174 strncat(lmd->lmd_params, s1 + 6, length);
1175 lmd->lmd_params[params_length + length] = '\0';
1176 strlcat(lmd->lmd_params, " ", LMD_PARAMS_MAXLEN);
1177 s3 = s1 + 6 + length;
1179 } else if (strncmp(s1, "osd=", 4) == 0) {
1180 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
1185 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1186 end of the options. */
1187 else if (strncmp(s1, "device=", 7) == 0) {
1189 /* terminate options right before device. device
1190 must be the last one. */
1196 s2 = strchr(s3, ',');
1204 memmove(s1, s2, strlen(s2) + 1);
1210 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1211 "(need mount option 'device=...')\n");
1215 s1 = strstr(devname, ":/");
1218 lmd->lmd_flags |= LMD_FLG_CLIENT;
1219 /* Remove leading /s from fsname */
1220 while (*++s1 == '/') ;
1221 /* Freed in lustre_free_lsi */
1222 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1223 if (!lmd->lmd_profile)
1225 sprintf(lmd->lmd_profile, "%s-client", s1);
1228 /* Freed in lustre_free_lsi */
1229 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1232 strcpy(lmd->lmd_dev, devname);
1234 /* Save mount options */
1235 s1 = options + strlen(options) - 1;
1236 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1238 if (*options != 0) {
1239 /* Freed in lustre_free_lsi */
1240 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1243 strcpy(lmd->lmd_opts, options);
1247 lmd->lmd_magic = LMD_MAGIC;
1252 CERROR("Bad mount options %s\n", options);
1256 struct lustre_mount_data2 {
1258 struct vfsmount *lmd2_mnt;
1261 /** This is the entry point for the mount call into Lustre.
1262 * This is called when a server or client is mounted,
1263 * and this is where we start setting things up.
1264 * @param data Mount options (e.g. -o flock,abort_recov)
1266 static int lustre_fill_super(struct super_block *sb, void *data, int silent)
1268 struct lustre_mount_data *lmd;
1269 struct lustre_mount_data2 *lmd2 = data;
1270 struct lustre_sb_info *lsi;
1274 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1276 lsi = lustre_init_lsi(sb);
1282 * Disable lockdep during mount, because mount locking patterns are
1288 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
1290 obd_zombie_barrier();
1292 /* Figure out the lmd from the mount options */
1293 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
1295 GOTO(out, rc = -EINVAL);
1298 if (lmd_is_client(lmd)) {
1299 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1300 if (client_fill_super == NULL)
1301 request_module("lustre");
1302 if (client_fill_super == NULL) {
1303 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1304 "client mount! Is the 'lustre' "
1305 "module loaded?\n");
1309 rc = lustre_start_mgc(sb);
1314 /* Connect and start */
1315 /* (should always be ll_fill_super) */
1316 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
1317 /* c_f_s will call lustre_common_put_super on failure */
1320 #ifdef HAVE_SERVER_SUPPORT
1321 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1322 rc = server_fill_super(sb);
1323 /* s_f_s calls lustre_start_mgc after the mount because we need
1324 the MGS nids which are stored on disk. Plus, we may
1325 need to start the MGS first. */
1326 /* s_f_s will call server_put_super on failure */
1328 CERROR("This is client-side-only module, "
1329 "cannot handle server mount.\n");
1334 /* If error happens in fill_super() call, @lsi will be killed there.
1335 * This is why we do not put it here. */
1339 CERROR("Unable to mount %s (%d)\n",
1340 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1342 CDEBUG(D_SUPER, "Mount %s complete\n",
1350 /* We can't call ll_fill_super by name because it lives in a module that
1351 must be loaded after this one. */
1352 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
1353 struct vfsmount *mnt))
1355 client_fill_super = cfs;
1357 EXPORT_SYMBOL(lustre_register_client_fill_super);
1359 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
1361 kill_super_cb = cfs;
1363 EXPORT_SYMBOL(lustre_register_kill_super_cb);
1365 /***************** FS registration ******************/
1366 #ifdef HAVE_FSTYPE_MOUNT
1367 static struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
1368 const char *devname, void *data)
1370 struct lustre_mount_data2 lmd2 = { data, NULL };
1372 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
1375 static int lustre_get_sb(struct file_system_type *fs_type, int flags,
1376 const char *devname, void *data, struct vfsmount *mnt)
1378 struct lustre_mount_data2 lmd2 = { data, mnt };
1380 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
1384 static void lustre_kill_super(struct super_block *sb)
1386 struct lustre_sb_info *lsi = s2lsi(sb);
1388 if (kill_super_cb && lsi && !IS_SERVER(lsi))
1389 (*kill_super_cb)(sb);
1391 kill_anon_super(sb);
1394 /** Register the "lustre" fs type
1396 static struct file_system_type lustre_fs_type = {
1397 .owner = THIS_MODULE,
1399 #ifdef HAVE_FSTYPE_MOUNT
1400 .mount = lustre_mount,
1402 .get_sb = lustre_get_sb,
1404 .kill_sb = lustre_kill_super,
1405 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
1406 FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE,
1408 MODULE_ALIAS_FS("lustre");
1410 int lustre_register_fs(void)
1412 return register_filesystem(&lustre_fs_type);
1415 int lustre_unregister_fs(void)
1417 return unregister_filesystem(&lustre_fs_type);