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);
187 /** Call class_attach and class_setup. These methods in turn call
188 * obd type-specific methods.
190 int lustre_start_simple(char *obdname, char *type, char *uuid,
191 char *s1, char *s2, char *s3, char *s4)
194 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
196 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, NULL, NULL);
198 CERROR("%s attach error %d\n", obdname, rc);
201 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, s3, s4);
203 CERROR("%s setup error %d\n", obdname, rc);
204 do_lcfg(obdname, 0, LCFG_DETACH, NULL, NULL, NULL, NULL);
209 static DEFINE_MUTEX(mgc_start_lock);
211 /** Set up a mgc obd to process startup logs
213 * \param sb [in] super block of the mgc obd
215 * \retval 0 success, otherwise error code
217 int lustre_start_mgc(struct super_block *sb)
219 struct obd_connect_data *data = NULL;
220 struct lustre_sb_info *lsi = s2lsi(sb);
221 struct obd_device *obd;
222 struct obd_export *exp;
223 struct obd_uuid *uuid;
226 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
228 int rc = 0, i = 0, j, len;
231 LASSERT(lsi->lsi_lmd);
233 /* Find the first non-lo MGS nid for our MGC name */
234 if (IS_SERVER(lsi)) {
235 /* mount -o mgsnode=nid */
236 ptr = lsi->lsi_lmd->lmd_mgs;
237 if (lsi->lsi_lmd->lmd_mgs &&
238 (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) {
240 } else if (IS_MGS(lsi)) {
241 lnet_process_id_t id;
242 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
243 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
250 } else { /* client */
251 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
252 ptr = lsi->lsi_lmd->lmd_dev;
253 if (class_parse_nid(ptr, &nid, &ptr) == 0)
257 CERROR("No valid MGS nids found.\n");
261 mutex_lock(&mgc_start_lock);
263 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
264 OBD_ALLOC(mgcname, len);
265 OBD_ALLOC(niduuid, len + 2);
266 if (!mgcname || !niduuid)
267 GOTO(out_free, rc = -ENOMEM);
268 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
270 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
274 GOTO(out_free, rc = -ENOMEM);
276 obd = class_name2obd(mgcname);
277 if (obd && !obd->obd_stopping) {
280 rc = obd_set_info_async(NULL, obd->obd_self_export,
281 strlen(KEY_MGSSEC), KEY_MGSSEC,
282 strlen(mgssec), mgssec, NULL);
286 /* Re-using an existing MGC */
287 atomic_inc(&obd->u.cli.cl_mgc_refcount);
289 /* IR compatibility check, only for clients */
290 if (lmd_is_client(lsi->lsi_lmd)) {
292 int vallen = sizeof(*data);
293 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
295 rc = obd_get_info(NULL, obd->obd_self_export,
296 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
297 &vallen, data, NULL);
299 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
300 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
301 /* LMD_FLG_NOIR is for test purpose only */
303 "Trying to mount a client with IR setting "
304 "not compatible with current mgc. "
305 "Force to use current mgc setting that is "
307 has_ir ? "enabled" : "disabled");
309 *flags &= ~LMD_FLG_NOIR;
311 *flags |= LMD_FLG_NOIR;
316 /* If we are restarting the MGS, don't try to keep the MGC's
317 old connection, or registration will fail. */
319 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
323 /* Try all connections, but only once (again).
324 We don't want to block another target from starting
325 (using its local copy of the log), but we do want to connect
326 if at all possible. */
328 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
329 rc = obd_set_info_async(NULL, obd->obd_self_export,
330 sizeof(KEY_INIT_RECOV_BACKUP),
331 KEY_INIT_RECOV_BACKUP,
332 sizeof(recov_bk), &recov_bk, NULL);
336 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
338 /* Add the primary nids for the MGS */
340 sprintf(niduuid, "%s_%x", mgcname, i);
341 if (IS_SERVER(lsi)) {
342 ptr = lsi->lsi_lmd->lmd_mgs;
343 CDEBUG(D_MOUNT, "mgs nids %s.\n", ptr);
345 /* Use local nids (including LO) */
346 lnet_process_id_t id;
347 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
348 rc = do_lcfg(mgcname, id.nid, LCFG_ADD_UUID,
349 niduuid, NULL, NULL, NULL);
352 /* Use mgsnode= nids */
353 /* mount -o mgsnode=nid */
354 if (lsi->lsi_lmd->lmd_mgs) {
355 ptr = lsi->lsi_lmd->lmd_mgs;
356 } else if (class_find_param(ptr, PARAM_MGSNODE,
358 CERROR("No MGS nids given.\n");
359 GOTO(out_free, rc = -EINVAL);
363 * Here we only take the first mgsnid as its primary
364 * serving mgs node, the rest mgsnid will be taken as
365 * failover mgs node, otherwise they would be takens
366 * as multiple nids of a single mgs node.
368 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
369 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
370 niduuid, NULL, NULL, NULL);
377 } else { /* client */
378 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
379 ptr = lsi->lsi_lmd->lmd_dev;
380 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
381 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
382 niduuid, NULL, NULL, NULL);
385 /* Stop at the first failover nid */
391 CERROR("No valid MGS nids found.\n");
392 GOTO(out_free, rc = -EINVAL);
394 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
396 /* Random uuid for MGC allows easier reconnects */
398 ll_generate_random_uuid(uuidc);
399 class_uuid_unparse(uuidc, uuid);
402 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
403 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
404 niduuid, NULL, NULL);
409 /* Add any failover MGS nids */
411 while (ptr && ((*ptr == ':' ||
412 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) {
413 /* New failover node */
414 sprintf(niduuid, "%s_%x", mgcname, i);
416 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
417 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
418 niduuid, NULL, NULL, NULL);
425 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
426 niduuid, NULL, NULL, NULL);
434 lsi->lsi_lmd->lmd_mgs_failnodes = i;
436 obd = class_name2obd(mgcname);
438 CERROR("Can't find mgcobd %s\n", mgcname);
439 GOTO(out_free, rc = -ENOTCONN);
442 rc = obd_set_info_async(NULL, obd->obd_self_export,
443 strlen(KEY_MGSSEC), KEY_MGSSEC,
444 strlen(mgssec), mgssec, NULL);
448 /* Keep a refcount of servers/clients who started with "mount",
449 so we know when we can get rid of the mgc. */
450 atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
452 /* We connect to the MGS at setup, and don't disconnect until cleanup */
453 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
454 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV |
455 OBD_CONNECT_LVB_TYPE;
457 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
458 data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB;
461 if (lmd_is_client(lsi->lsi_lmd) &&
462 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
463 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
464 data->ocd_version = LUSTRE_VERSION_CODE;
465 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
467 CERROR("connect failed %d\n", rc);
471 obd->u.cli.cl_mgc_mgsexp = exp;
474 /* Keep the mgc info in the sb. Note that many lsi's can point
478 mutex_unlock(&mgc_start_lock);
483 OBD_FREE(mgcname, len);
485 OBD_FREE(niduuid, len + 2);
489 static int lustre_stop_mgc(struct super_block *sb)
491 struct lustre_sb_info *lsi = s2lsi(sb);
492 struct obd_device *obd;
493 char *niduuid = NULL, *ptr = NULL;
494 int i, rc = 0, len = 0;
504 mutex_lock(&mgc_start_lock);
505 LASSERT(atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
506 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
507 /* This is not fatal, every client that stops
508 will call in here. */
509 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
510 atomic_read(&obd->u.cli.cl_mgc_refcount));
511 GOTO(out, rc = -EBUSY);
514 /* The MGC has no recoverable data in any case.
515 * force shotdown set in umount_begin */
516 obd->obd_no_recov = 1;
518 if (obd->u.cli.cl_mgc_mgsexp) {
519 /* An error is not fatal, if we are unable to send the
520 disconnect mgs ping evictor cleans up the export */
521 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
523 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
526 /* Save the obdname for cleaning the nid uuids, which are
528 len = strlen(obd->obd_name) + 6;
529 OBD_ALLOC(niduuid, len);
531 strcpy(niduuid, obd->obd_name);
532 ptr = niduuid + strlen(niduuid);
535 rc = class_manual_cleanup(obd);
539 /* Clean the nid uuids */
541 GOTO(out, rc = -ENOMEM);
543 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
544 sprintf(ptr, "_%x", i);
545 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
546 niduuid, NULL, NULL, NULL);
548 CERROR("del MDC UUID %s failed: rc = %d\n",
553 OBD_FREE(niduuid, len);
555 /* class_import_put will get rid of the additional connections */
556 mutex_unlock(&mgc_start_lock);
560 /***************** lustre superblock **************/
562 static struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
564 struct lustre_sb_info *lsi;
570 OBD_ALLOC_PTR(lsi->lsi_lmd);
576 lsi->lsi_lmd->lmd_exclude_count = 0;
577 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
578 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
579 s2lsi_nocast(sb) = lsi;
580 /* we take 1 extra ref for our setup */
581 atomic_set(&lsi->lsi_mounts, 1);
583 /* Default umount style */
584 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
585 INIT_LIST_HEAD(&lsi->lsi_lwp_list);
586 spin_lock_init(&lsi->lsi_lwp_lock);
591 static int lustre_free_lsi(struct super_block *sb)
593 struct lustre_sb_info *lsi = s2lsi(sb);
596 LASSERT(lsi != NULL);
597 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
599 /* someone didn't call server_put_mount. */
600 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
602 if (lsi->lsi_lmd != NULL) {
603 if (lsi->lsi_lmd->lmd_dev != NULL)
604 OBD_FREE(lsi->lsi_lmd->lmd_dev,
605 strlen(lsi->lsi_lmd->lmd_dev) + 1);
606 if (lsi->lsi_lmd->lmd_profile != NULL)
607 OBD_FREE(lsi->lsi_lmd->lmd_profile,
608 strlen(lsi->lsi_lmd->lmd_profile) + 1);
609 if (lsi->lsi_lmd->lmd_mgssec != NULL)
610 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
611 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
612 if (lsi->lsi_lmd->lmd_opts != NULL)
613 OBD_FREE(lsi->lsi_lmd->lmd_opts,
614 strlen(lsi->lsi_lmd->lmd_opts) + 1);
615 if (lsi->lsi_lmd->lmd_exclude_count)
616 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
617 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
618 lsi->lsi_lmd->lmd_exclude_count);
619 if (lsi->lsi_lmd->lmd_mgs != NULL)
620 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
621 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
622 if (lsi->lsi_lmd->lmd_osd_type != NULL)
623 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
624 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
625 if (lsi->lsi_lmd->lmd_params != NULL)
626 OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
628 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
631 LASSERT(lsi->lsi_llsbi == NULL);
632 OBD_FREE(lsi, sizeof(*lsi));
633 s2lsi_nocast(sb) = NULL;
638 /* The lsi has one reference for every server that is using the disk -
639 e.g. MDT, MGS, and potentially MGC */
640 int lustre_put_lsi(struct super_block *sb)
642 struct lustre_sb_info *lsi = s2lsi(sb);
645 LASSERT(lsi != NULL);
647 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
648 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
649 if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
650 lu_device_put(&lsi->lsi_dt_dev->dd_lu_dev);
651 lsi->lsi_osd_exp->exp_obd->obd_lvfs_ctxt.dt = NULL;
652 lsi->lsi_dt_dev = NULL;
653 obd_disconnect(lsi->lsi_osd_exp);
654 /* wait till OSD is gone */
655 obd_zombie_barrier();
664 * <FSNAME><SEPERATOR><TYPE><INDEX>
665 * FSNAME is between 1 and 8 characters (inclusive).
666 * Excluded characters are '/' and ':'
667 * SEPERATOR is either ':' or '-'
668 * TYPE: "OST", "MDT", etc.
669 * INDEX: Hex representation of the index
672 /** Get the fsname ("lustre") from the server name ("lustre-OST003F").
673 * @param [in] svname server name including type and index
674 * @param [out] fsname Buffer to copy filesystem name prefix into.
675 * Must have at least 'strlen(fsname) + 1' chars.
676 * @param [out] endptr if endptr isn't NULL it is set to end of fsname
679 int server_name2fsname(const char *svname, char *fsname, const char **endptr)
683 dash = svname + strnlen(svname, 8); /* max fsname length is 8 */
684 for (; dash > svname && *dash != '-' && *dash != ':'; dash--)
689 if (fsname != NULL) {
690 strncpy(fsname, svname, dash - svname);
691 fsname[dash - svname] = '\0';
699 EXPORT_SYMBOL(server_name2fsname);
702 * Get service name (svname) from string
704 * if endptr isn't NULL it is set to end of fsname *
706 int server_name2svname(const char *label, char *svname, const char **endptr,
712 /* We use server_name2fsname() just for parsing */
713 rc = server_name2fsname(label, NULL, &dash);
720 if (strlcpy(svname, dash + 1, svsize) >= svsize)
725 EXPORT_SYMBOL(server_name2svname);
728 * check server name is OST.
730 int server_name_is_ost(const char *svname)
735 /* We use server_name2fsname() just for parsing */
736 rc = server_name2fsname(svname, NULL, &dash);
742 if (strncmp(dash, "OST", 3) == 0)
746 EXPORT_SYMBOL(server_name_is_ost);
749 * Get the index from the target name MDTXXXX/OSTXXXX
750 * rc = server type, or rc < 0 on error
752 int target_name2index(const char *tgtname, __u32 *idx, const char **endptr)
754 const char *dash = tgtname;
758 if (strncmp(dash, "MDT", 3) == 0)
759 rc = LDD_F_SV_TYPE_MDT;
760 else if (strncmp(dash, "OST", 3) == 0)
761 rc = LDD_F_SV_TYPE_OST;
767 if (strncmp(dash, "all", 3) == 0) {
770 return rc | LDD_F_SV_ALL;
773 index = simple_strtoul(dash, (char **)endptr, 16);
778 EXPORT_SYMBOL(target_name2index);
780 /* Get the index from the obd name.
783 if endptr isn't NULL it is set to end of name */
784 int server_name2index(const char *svname, __u32 *idx, const char **endptr)
789 /* We use server_name2fsname() just for parsing */
790 rc = server_name2fsname(svname, NULL, &dash);
795 rc = target_name2index(dash, idx, endptr);
799 /* Account for -mdc after index that is possible when specifying mdt */
800 if (endptr != NULL && strncmp(LUSTRE_MDC_NAME, *endptr + 1,
801 sizeof(LUSTRE_MDC_NAME)-1) == 0)
802 *endptr += sizeof(LUSTRE_MDC_NAME);
806 EXPORT_SYMBOL(server_name2index);
808 /*************** mount common betweeen server and client ***************/
811 int lustre_common_put_super(struct super_block *sb)
816 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
818 /* Drop a ref to the MGC */
819 rc = lustre_stop_mgc(sb);
820 if (rc && (rc != -ENOENT)) {
822 CERROR("Can't stop MGC: %d\n", rc);
825 /* BUSY just means that there's some other obd that
826 needs the mgc. Let him clean it up. */
827 CDEBUG(D_MOUNT, "MGC still in use\n");
829 /* Drop a ref to the mounted disk */
834 EXPORT_SYMBOL(lustre_common_put_super);
836 static void lmd_print(struct lustre_mount_data *lmd)
840 PRINT_CMD(D_MOUNT, " mount data:\n");
841 if (lmd_is_client(lmd))
842 PRINT_CMD(D_MOUNT, "profile: %s\n", lmd->lmd_profile);
843 PRINT_CMD(D_MOUNT, "device: %s\n", lmd->lmd_dev);
844 PRINT_CMD(D_MOUNT, "flags: %x\n", lmd->lmd_flags);
847 PRINT_CMD(D_MOUNT, "options: %s\n", lmd->lmd_opts);
849 if (lmd->lmd_recovery_time_soft)
850 PRINT_CMD(D_MOUNT, "recovery time soft: %d\n",
851 lmd->lmd_recovery_time_soft);
853 if (lmd->lmd_recovery_time_hard)
854 PRINT_CMD(D_MOUNT, "recovery time hard: %d\n",
855 lmd->lmd_recovery_time_hard);
857 for (i = 0; i < lmd->lmd_exclude_count; i++) {
858 PRINT_CMD(D_MOUNT, "exclude %d: OST%04x\n", i,
859 lmd->lmd_exclude[i]);
863 /* Is this server on the exclusion list */
864 int lustre_check_exclusion(struct super_block *sb, char *svname)
866 struct lustre_sb_info *lsi = s2lsi(sb);
867 struct lustre_mount_data *lmd = lsi->lsi_lmd;
872 rc = server_name2index(svname, &index, NULL);
873 if (rc != LDD_F_SV_TYPE_OST)
874 /* Only exclude OSTs */
877 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
878 index, lmd->lmd_exclude_count, lmd->lmd_dev);
880 for(i = 0; i < lmd->lmd_exclude_count; i++) {
881 if (index == lmd->lmd_exclude[i]) {
882 CWARN("Excluding %s (on exclusion list)\n", svname);
889 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
890 static int lmd_make_exclusion(struct lustre_mount_data *lmd, const char *ptr)
892 const char *s1 = ptr, *s2;
898 /* The shortest an ost name can be is 8 chars: -OST0000.
899 We don't actually know the fsname at this time, so in fact
900 a user could specify any fsname. */
901 devmax = strlen(ptr) / 8 + 1;
903 /* temp storage until we figure out how many we have */
904 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
908 /* we enter this fn pointing at the '=' */
909 while (*s1 && *s1 != ' ' && *s1 != ',') {
911 rc = server_name2index(s1, &index, &s2);
913 CERROR("Can't parse server name '%s': rc = %d\n",
917 if (rc == LDD_F_SV_TYPE_OST)
918 exclude_list[lmd->lmd_exclude_count++] = index;
920 CDEBUG(D_MOUNT, "ignoring exclude %.*s: type = %#x\n",
921 (uint)(s2-s1), s1, rc);
923 /* now we are pointing at ':' (next exclude)
924 or ',' (end of excludes) */
925 if (lmd->lmd_exclude_count >= devmax)
928 if (rc >= 0) /* non-err */
931 if (lmd->lmd_exclude_count) {
932 /* permanent, freed in lustre_free_lsi */
933 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
934 lmd->lmd_exclude_count);
935 if (lmd->lmd_exclude) {
936 memcpy(lmd->lmd_exclude, exclude_list,
937 sizeof(index) * lmd->lmd_exclude_count);
940 lmd->lmd_exclude_count = 0;
943 OBD_FREE(exclude_list, sizeof(index) * devmax);
947 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
952 if (lmd->lmd_mgssec != NULL) {
953 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
954 lmd->lmd_mgssec = NULL;
957 tail = strchr(ptr, ',');
959 length = strlen(ptr);
963 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
964 if (lmd->lmd_mgssec == NULL)
967 memcpy(lmd->lmd_mgssec, ptr, length);
968 lmd->lmd_mgssec[length] = '\0';
972 static int lmd_parse_string(char **handle, char *ptr)
977 if ((handle == NULL) || (ptr == NULL))
980 if (*handle != NULL) {
981 OBD_FREE(*handle, strlen(*handle) + 1);
985 tail = strchr(ptr, ',');
987 length = strlen(ptr);
991 OBD_ALLOC(*handle, length + 1);
995 memcpy(*handle, ptr, length);
996 (*handle)[length] = '\0';
1001 /* Collect multiple values for mgsnid specifiers */
1002 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
1010 /* Find end of nidlist */
1011 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
1012 length = tail - *ptr;
1014 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
1018 if (lmd->lmd_mgs != NULL)
1019 oldlen = strlen(lmd->lmd_mgs) + 1;
1021 OBD_ALLOC(mgsnid, oldlen + length + 1);
1025 if (lmd->lmd_mgs != NULL) {
1026 /* Multiple mgsnid= are taken to mean failover locations */
1027 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
1028 mgsnid[oldlen - 1] = ':';
1029 OBD_FREE(lmd->lmd_mgs, oldlen);
1031 memcpy(mgsnid + oldlen, *ptr, length);
1032 mgsnid[oldlen + length] = '\0';
1033 lmd->lmd_mgs = mgsnid;
1039 /** Parse mount line options
1040 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
1041 * dev is passed as device=uml1:/lustre by mount.lustre
1043 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1045 char *s1, *s2, *s3, *devname = NULL;
1046 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1052 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1053 "/sbin/mount.lustre is installed.\n");
1057 /* Options should be a string - try to detect old lmd data */
1058 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1059 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1060 "/sbin/mount.lustre. Please install "
1061 "version %s\n", LUSTRE_VERSION_STRING);
1064 lmd->lmd_magic = LMD_MAGIC;
1066 OBD_ALLOC(lmd->lmd_params, LMD_PARAMS_MAXLEN);
1067 if (lmd->lmd_params == NULL)
1069 lmd->lmd_params[0] = '\0';
1071 /* Set default flags here */
1076 int time_min = OBD_RECOVERY_TIME_MIN;
1078 /* Skip whitespace and extra commas */
1079 while (*s1 == ' ' || *s1 == ',')
1083 /* Client options are parsed in ll_options: eg. flock,
1086 /* Parse non-ldiskfs options here. Rather than modifying
1087 ldiskfs, we just zero these out here */
1088 if (strncmp(s1, "abort_recov", 11) == 0) {
1089 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1091 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
1092 lmd->lmd_recovery_time_soft = max_t(int,
1093 simple_strtoul(s1 + 19, NULL, 10), time_min);
1095 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
1096 lmd->lmd_recovery_time_hard = max_t(int,
1097 simple_strtoul(s1 + 19, NULL, 10), time_min);
1099 } else if (strncmp(s1, "noir", 4) == 0) {
1100 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
1102 } else if (strncmp(s1, "nosvc", 5) == 0) {
1103 lmd->lmd_flags |= LMD_FLG_NOSVC;
1105 } else if (strncmp(s1, "nomgs", 5) == 0) {
1106 lmd->lmd_flags |= LMD_FLG_NOMGS;
1108 } else if (strncmp(s1, "noscrub", 7) == 0) {
1109 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
1111 } else if (strncmp(s1, PARAM_MGSNODE,
1112 sizeof(PARAM_MGSNODE) - 1) == 0) {
1113 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
1114 /* Assume the next mount opt is the first
1115 invalid nid we get to. */
1116 rc = lmd_parse_mgs(lmd, &s2);
1121 } else if (strncmp(s1, "writeconf", 9) == 0) {
1122 lmd->lmd_flags |= LMD_FLG_WRITECONF;
1124 } else if (strncmp(s1, "update", 6) == 0) {
1125 lmd->lmd_flags |= LMD_FLG_UPDATE;
1127 } else if (strncmp(s1, "virgin", 6) == 0) {
1128 lmd->lmd_flags |= LMD_FLG_VIRGIN;
1130 } else if (strncmp(s1, "noprimnode", 10) == 0) {
1131 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
1133 } else if (strncmp(s1, "mgssec=", 7) == 0) {
1134 rc = lmd_parse_mgssec(lmd, s1 + 7);
1138 /* ost exclusion list */
1139 } else if (strncmp(s1, "exclude=", 8) == 0) {
1140 rc = lmd_make_exclusion(lmd, s1 + 7);
1144 } else if (strncmp(s1, "mgs", 3) == 0) {
1146 lmd->lmd_flags |= LMD_FLG_MGS;
1148 } else if (strncmp(s1, "svname=", 7) == 0) {
1149 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
1153 } else if (strncmp(s1, "param=", 6) == 0) {
1154 size_t length, params_length;
1155 char *tail = strchr(s1 + 6, ',');
1157 length = strlen(s1);
1160 char *param_str = tail + 1;
1161 int supplementary = 1;
1163 while (class_parse_nid_quiet(param_str, &nid,
1167 length = param_str - s1 - supplementary;
1170 params_length = strlen(lmd->lmd_params);
1171 if (params_length + length + 1 >= LMD_PARAMS_MAXLEN)
1173 strncat(lmd->lmd_params, s1 + 6, length);
1174 lmd->lmd_params[params_length + length] = '\0';
1175 strlcat(lmd->lmd_params, " ", LMD_PARAMS_MAXLEN);
1176 s3 = s1 + 6 + length;
1178 } else if (strncmp(s1, "osd=", 4) == 0) {
1179 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
1184 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1185 end of the options. */
1186 else if (strncmp(s1, "device=", 7) == 0) {
1188 /* terminate options right before device. device
1189 must be the last one. */
1195 s2 = strchr(s3, ',');
1203 memmove(s1, s2, strlen(s2) + 1);
1209 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1210 "(need mount option 'device=...')\n");
1214 s1 = strstr(devname, ":/");
1217 lmd->lmd_flags |= LMD_FLG_CLIENT;
1218 /* Remove leading /s from fsname */
1219 while (*++s1 == '/') ;
1220 /* Freed in lustre_free_lsi */
1221 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1222 if (!lmd->lmd_profile)
1224 sprintf(lmd->lmd_profile, "%s-client", s1);
1227 /* Freed in lustre_free_lsi */
1228 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1231 strcpy(lmd->lmd_dev, devname);
1233 /* Save mount options */
1234 s1 = options + strlen(options) - 1;
1235 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1237 if (*options != 0) {
1238 /* Freed in lustre_free_lsi */
1239 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1242 strcpy(lmd->lmd_opts, options);
1246 lmd->lmd_magic = LMD_MAGIC;
1251 CERROR("Bad mount options %s\n", options);
1255 struct lustre_mount_data2 {
1257 struct vfsmount *lmd2_mnt;
1260 /** This is the entry point for the mount call into Lustre.
1261 * This is called when a server or client is mounted,
1262 * and this is where we start setting things up.
1263 * @param data Mount options (e.g. -o flock,abort_recov)
1265 static int lustre_fill_super(struct super_block *sb, void *data, int silent)
1267 struct lustre_mount_data *lmd;
1268 struct lustre_mount_data2 *lmd2 = data;
1269 struct lustre_sb_info *lsi;
1273 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1275 lsi = lustre_init_lsi(sb);
1281 * Disable lockdep during mount, because mount locking patterns are
1287 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
1289 obd_zombie_barrier();
1291 /* Figure out the lmd from the mount options */
1292 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
1294 GOTO(out, rc = -EINVAL);
1297 if (lmd_is_client(lmd)) {
1298 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1299 if (client_fill_super == NULL)
1300 request_module("lustre");
1301 if (client_fill_super == NULL) {
1302 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1303 "client mount! Is the 'lustre' "
1304 "module loaded?\n");
1308 rc = lustre_start_mgc(sb);
1313 /* Connect and start */
1314 /* (should always be ll_fill_super) */
1315 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
1316 /* c_f_s will call lustre_common_put_super on failure */
1319 #ifdef HAVE_SERVER_SUPPORT
1320 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1321 rc = server_fill_super(sb);
1322 /* s_f_s calls lustre_start_mgc after the mount because we need
1323 the MGS nids which are stored on disk. Plus, we may
1324 need to start the MGS first. */
1325 /* s_f_s will call server_put_super on failure */
1327 CERROR("This is client-side-only module, "
1328 "cannot handle server mount.\n");
1333 /* If error happens in fill_super() call, @lsi will be killed there.
1334 * This is why we do not put it here. */
1338 CERROR("Unable to mount %s (%d)\n",
1339 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1341 CDEBUG(D_SUPER, "Mount %s complete\n",
1349 /* We can't call ll_fill_super by name because it lives in a module that
1350 must be loaded after this one. */
1351 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
1352 struct vfsmount *mnt))
1354 client_fill_super = cfs;
1356 EXPORT_SYMBOL(lustre_register_client_fill_super);
1358 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
1360 kill_super_cb = cfs;
1362 EXPORT_SYMBOL(lustre_register_kill_super_cb);
1364 /***************** FS registration ******************/
1365 #ifdef HAVE_FSTYPE_MOUNT
1366 static struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
1367 const char *devname, void *data)
1369 struct lustre_mount_data2 lmd2 = { data, NULL };
1371 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
1374 static int lustre_get_sb(struct file_system_type *fs_type, int flags,
1375 const char *devname, void *data, struct vfsmount *mnt)
1377 struct lustre_mount_data2 lmd2 = { data, mnt };
1379 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
1383 static void lustre_kill_super(struct super_block *sb)
1385 struct lustre_sb_info *lsi = s2lsi(sb);
1387 if (kill_super_cb && lsi && !IS_SERVER(lsi))
1388 (*kill_super_cb)(sb);
1390 kill_anon_super(sb);
1393 /** Register the "lustre" fs type
1395 static struct file_system_type lustre_fs_type = {
1396 .owner = THIS_MODULE,
1398 #ifdef HAVE_FSTYPE_MOUNT
1399 .mount = lustre_mount,
1401 .get_sb = lustre_get_sb,
1403 .kill_sb = lustre_kill_super,
1404 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
1405 FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE,
1407 MODULE_ALIAS_FS("lustre");
1409 int lustre_register_fs(void)
1411 return register_filesystem(&lustre_fs_type);
1414 int lustre_unregister_fs(void)
1416 return unregister_filesystem(&lustre_fs_type);