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, 2015, 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 nidstr[LNET_NIDSTR_SIZE];
227 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
229 int rc = 0, i = 0, j;
233 LASSERT(lsi->lsi_lmd);
235 /* Find the first non-lo MGS nid for our MGC name */
236 if (IS_SERVER(lsi)) {
237 /* mount -o mgsnode=nid */
238 ptr = lsi->lsi_lmd->lmd_mgs;
239 if (lsi->lsi_lmd->lmd_mgs &&
240 (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) {
242 } else if (IS_MGS(lsi)) {
243 lnet_process_id_t id;
244 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
245 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
252 } else { /* client */
253 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
254 ptr = lsi->lsi_lmd->lmd_dev;
255 if (class_parse_nid(ptr, &nid, &ptr) == 0)
259 CERROR("No valid MGS nids found.\n");
263 mutex_lock(&mgc_start_lock);
265 libcfs_nid2str_r(nid, nidstr, sizeof(nidstr));
266 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(nidstr) + 1;
267 OBD_ALLOC(mgcname, len);
268 OBD_ALLOC(niduuid, len + 2);
269 if (mgcname == NULL || niduuid == NULL)
270 GOTO(out_free, rc = -ENOMEM);
271 snprintf(mgcname, len, "%s%s", LUSTRE_MGC_OBDNAME, nidstr);
273 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
277 GOTO(out_free, rc = -ENOMEM);
279 obd = class_name2obd(mgcname);
280 if (obd && !obd->obd_stopping) {
283 rc = obd_set_info_async(NULL, obd->obd_self_export,
284 strlen(KEY_MGSSEC), KEY_MGSSEC,
285 strlen(mgssec), mgssec, NULL);
289 /* Re-using an existing MGC */
290 atomic_inc(&obd->u.cli.cl_mgc_refcount);
292 /* IR compatibility check, only for clients */
293 if (lmd_is_client(lsi->lsi_lmd)) {
295 int vallen = sizeof(*data);
296 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
298 rc = obd_get_info(NULL, obd->obd_self_export,
299 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
302 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
303 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
304 /* LMD_FLG_NOIR is for test purpose only */
306 "Trying to mount a client with IR setting "
307 "not compatible with current mgc. "
308 "Force to use current mgc setting that is "
310 has_ir ? "enabled" : "disabled");
312 *flags &= ~LMD_FLG_NOIR;
314 *flags |= LMD_FLG_NOIR;
319 /* If we are restarting the MGS, don't try to keep the MGC's
320 old connection, or registration will fail. */
322 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
326 /* Try all connections, but only once (again).
327 We don't want to block another target from starting
328 (using its local copy of the log), but we do want to connect
329 if at all possible. */
331 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
332 rc = obd_set_info_async(NULL, obd->obd_self_export,
333 sizeof(KEY_INIT_RECOV_BACKUP),
334 KEY_INIT_RECOV_BACKUP,
335 sizeof(recov_bk), &recov_bk, NULL);
339 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
341 /* Add the primary nids for the MGS */
343 snprintf(niduuid, len + 2, "%s_%x", mgcname, i);
344 if (IS_SERVER(lsi)) {
345 ptr = lsi->lsi_lmd->lmd_mgs;
346 CDEBUG(D_MOUNT, "mgs nids %s.\n", ptr);
348 /* Use local nids (including LO) */
349 lnet_process_id_t id;
350 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
351 rc = do_lcfg(mgcname, id.nid, LCFG_ADD_UUID,
352 niduuid, NULL, NULL, NULL);
355 /* Use mgsnode= nids */
356 /* mount -o mgsnode=nid */
357 if (lsi->lsi_lmd->lmd_mgs) {
358 ptr = lsi->lsi_lmd->lmd_mgs;
359 } else if (class_find_param(ptr, PARAM_MGSNODE,
361 CERROR("No MGS nids given.\n");
362 GOTO(out_free, rc = -EINVAL);
366 * Here we only take the first mgsnid as its primary
367 * serving mgs node, the rest mgsnid will be taken as
368 * failover mgs node, otherwise they would be takens
369 * as multiple nids of a single mgs node.
371 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
372 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
373 niduuid, NULL, NULL, NULL);
380 } else { /* client */
381 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
382 ptr = lsi->lsi_lmd->lmd_dev;
383 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
384 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
385 niduuid, NULL, NULL, NULL);
388 /* Stop at the first failover nid */
394 CERROR("No valid MGS nids found.\n");
395 GOTO(out_free, rc = -EINVAL);
397 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
399 /* Random uuid for MGC allows easier reconnects */
401 ll_generate_random_uuid(uuidc);
402 class_uuid_unparse(uuidc, uuid);
405 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
406 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
407 niduuid, NULL, NULL);
412 /* Add any failover MGS nids */
414 while (ptr && ((*ptr == ':' ||
415 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) {
416 /* New failover node */
417 sprintf(niduuid, "%s_%x", mgcname, i);
419 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
420 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
421 niduuid, NULL, NULL, NULL);
428 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
429 niduuid, NULL, NULL, NULL);
437 lsi->lsi_lmd->lmd_mgs_failnodes = i;
439 obd = class_name2obd(mgcname);
441 CERROR("Can't find mgcobd %s\n", mgcname);
442 GOTO(out_free, rc = -ENOTCONN);
445 rc = obd_set_info_async(NULL, obd->obd_self_export,
446 strlen(KEY_MGSSEC), KEY_MGSSEC,
447 strlen(mgssec), mgssec, NULL);
451 /* Keep a refcount of servers/clients who started with "mount",
452 so we know when we can get rid of the mgc. */
453 atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
455 /* We connect to the MGS at setup, and don't disconnect until cleanup */
456 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
457 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV |
458 OBD_CONNECT_LVB_TYPE | OBD_CONNECT_BULK_MBITS;
460 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
461 data->ocd_connect_flags |= OBD_CONNECT_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 = NULL, *ptr = NULL;
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,
549 niduuid, NULL, NULL, NULL);
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 static 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_fileset != NULL)
613 OBD_FREE(lsi->lsi_lmd->lmd_fileset,
614 strlen(lsi->lsi_lmd->lmd_fileset) + 1);
615 if (lsi->lsi_lmd->lmd_mgssec != NULL)
616 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
617 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
618 if (lsi->lsi_lmd->lmd_opts != NULL)
619 OBD_FREE(lsi->lsi_lmd->lmd_opts,
620 strlen(lsi->lsi_lmd->lmd_opts) + 1);
621 if (lsi->lsi_lmd->lmd_exclude_count)
622 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
623 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
624 lsi->lsi_lmd->lmd_exclude_count);
625 if (lsi->lsi_lmd->lmd_mgs != NULL)
626 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
627 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
628 if (lsi->lsi_lmd->lmd_osd_type != NULL)
629 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
630 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
631 if (lsi->lsi_lmd->lmd_params != NULL)
632 OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
634 OBD_FREE_PTR(lsi->lsi_lmd);
637 LASSERT(lsi->lsi_llsbi == NULL);
639 s2lsi_nocast(sb) = NULL;
644 /* The lsi has one reference for every server that is using the disk -
645 e.g. MDT, MGS, and potentially MGC */
646 int lustre_put_lsi(struct super_block *sb)
648 struct lustre_sb_info *lsi = s2lsi(sb);
651 LASSERT(lsi != NULL);
653 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
654 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
655 if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
656 lu_device_put(&lsi->lsi_dt_dev->dd_lu_dev);
657 lsi->lsi_osd_exp->exp_obd->obd_lvfs_ctxt.dt = NULL;
658 lsi->lsi_dt_dev = NULL;
659 obd_disconnect(lsi->lsi_osd_exp);
660 /* wait till OSD is gone */
661 obd_zombie_barrier();
670 * <FSNAME><SEPARATOR><TYPE><INDEX>
671 * FSNAME is between 1 and 8 characters (inclusive).
672 * Excluded characters are '/' and ':'
673 * SEPARATOR is either ':' or '-'
674 * TYPE: "OST", "MDT", etc.
675 * INDEX: Hex representation of the index
678 /** Get the fsname ("lustre") from the server name ("lustre-OST003F").
679 * @param [in] svname server name including type and index
680 * @param [out] fsname Buffer to copy filesystem name prefix into.
681 * Must have at least 'strlen(fsname) + 1' chars.
682 * @param [out] endptr if endptr isn't NULL it is set to end of fsname
685 int server_name2fsname(const char *svname, char *fsname, const char **endptr)
689 dash = svname + strnlen(svname, 8); /* max fsname length is 8 */
690 for (; dash > svname && *dash != '-' && *dash != ':'; dash--)
695 if (fsname != NULL) {
696 strncpy(fsname, svname, dash - svname);
697 fsname[dash - svname] = '\0';
705 EXPORT_SYMBOL(server_name2fsname);
708 * Get service name (svname) from string
710 * if endptr isn't NULL it is set to end of fsname *
712 int server_name2svname(const char *label, char *svname, const char **endptr,
718 /* We use server_name2fsname() just for parsing */
719 rc = server_name2fsname(label, NULL, &dash);
726 if (strlcpy(svname, dash + 1, svsize) >= svsize)
731 EXPORT_SYMBOL(server_name2svname);
734 * check server name is OST.
736 int server_name_is_ost(const char *svname)
741 /* We use server_name2fsname() just for parsing */
742 rc = server_name2fsname(svname, NULL, &dash);
748 if (strncmp(dash, "OST", 3) == 0)
752 EXPORT_SYMBOL(server_name_is_ost);
755 * Get the index from the target name MDTXXXX/OSTXXXX
756 * rc = server type, or rc < 0 on error
758 int target_name2index(const char *tgtname, __u32 *idx, const char **endptr)
760 const char *dash = tgtname;
764 if (strncmp(dash, "MDT", 3) == 0)
765 rc = LDD_F_SV_TYPE_MDT;
766 else if (strncmp(dash, "OST", 3) == 0)
767 rc = LDD_F_SV_TYPE_OST;
773 if (strncmp(dash, "all", 3) == 0) {
776 return rc | LDD_F_SV_ALL;
779 index = simple_strtoul(dash, (char **)endptr, 16);
784 EXPORT_SYMBOL(target_name2index);
786 /* Get the index from the obd name.
789 if endptr isn't NULL it is set to end of name */
790 int server_name2index(const char *svname, __u32 *idx, const char **endptr)
795 /* We use server_name2fsname() just for parsing */
796 rc = server_name2fsname(svname, NULL, &dash);
801 rc = target_name2index(dash, idx, endptr);
805 /* Account for -mdc after index that is possible when specifying mdt */
806 if (endptr != NULL && strncmp(LUSTRE_MDC_NAME, *endptr + 1,
807 sizeof(LUSTRE_MDC_NAME)-1) == 0)
808 *endptr += sizeof(LUSTRE_MDC_NAME);
812 EXPORT_SYMBOL(server_name2index);
814 /*************** mount common betweeen server and client ***************/
817 int lustre_common_put_super(struct super_block *sb)
822 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
824 /* Drop a ref to the MGC */
825 rc = lustre_stop_mgc(sb);
826 if (rc && (rc != -ENOENT)) {
828 CERROR("Can't stop MGC: %d\n", rc);
831 /* BUSY just means that there's some other obd that
832 needs the mgc. Let him clean it up. */
833 CDEBUG(D_MOUNT, "MGC still in use\n");
835 /* Drop a ref to the mounted disk */
840 EXPORT_SYMBOL(lustre_common_put_super);
842 static void lmd_print(struct lustre_mount_data *lmd)
846 PRINT_CMD(D_MOUNT, " mount data:\n");
847 if (lmd_is_client(lmd))
848 PRINT_CMD(D_MOUNT, "profile: %s\n", lmd->lmd_profile);
849 PRINT_CMD(D_MOUNT, "device: %s\n", lmd->lmd_dev);
850 PRINT_CMD(D_MOUNT, "flags: %x\n", lmd->lmd_flags);
853 PRINT_CMD(D_MOUNT, "options: %s\n", lmd->lmd_opts);
855 if (lmd->lmd_recovery_time_soft)
856 PRINT_CMD(D_MOUNT, "recovery time soft: %d\n",
857 lmd->lmd_recovery_time_soft);
859 if (lmd->lmd_recovery_time_hard)
860 PRINT_CMD(D_MOUNT, "recovery time hard: %d\n",
861 lmd->lmd_recovery_time_hard);
863 for (i = 0; i < lmd->lmd_exclude_count; i++) {
864 PRINT_CMD(D_MOUNT, "exclude %d: OST%04x\n", i,
865 lmd->lmd_exclude[i]);
869 /* Is this server on the exclusion list */
870 int lustre_check_exclusion(struct super_block *sb, char *svname)
872 struct lustre_sb_info *lsi = s2lsi(sb);
873 struct lustre_mount_data *lmd = lsi->lsi_lmd;
878 rc = server_name2index(svname, &index, NULL);
879 if (rc != LDD_F_SV_TYPE_OST)
880 /* Only exclude OSTs */
883 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
884 index, lmd->lmd_exclude_count, lmd->lmd_dev);
886 for(i = 0; i < lmd->lmd_exclude_count; i++) {
887 if (index == lmd->lmd_exclude[i]) {
888 CWARN("Excluding %s (on exclusion list)\n", svname);
895 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
896 static int lmd_make_exclusion(struct lustre_mount_data *lmd, const char *ptr)
898 const char *s1 = ptr, *s2;
904 /* The shortest an ost name can be is 8 chars: -OST0000.
905 We don't actually know the fsname at this time, so in fact
906 a user could specify any fsname. */
907 devmax = strlen(ptr) / 8 + 1;
909 /* temp storage until we figure out how many we have */
910 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
914 /* we enter this fn pointing at the '=' */
915 while (*s1 && *s1 != ' ' && *s1 != ',') {
917 rc = server_name2index(s1, &index, &s2);
919 CERROR("Can't parse server name '%s': rc = %d\n",
923 if (rc == LDD_F_SV_TYPE_OST)
924 exclude_list[lmd->lmd_exclude_count++] = index;
926 CDEBUG(D_MOUNT, "ignoring exclude %.*s: type = %#x\n",
927 (uint)(s2-s1), s1, rc);
929 /* now we are pointing at ':' (next exclude)
930 or ',' (end of excludes) */
931 if (lmd->lmd_exclude_count >= devmax)
934 if (rc >= 0) /* non-err */
937 if (lmd->lmd_exclude_count) {
938 /* permanent, freed in lustre_free_lsi */
939 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
940 lmd->lmd_exclude_count);
941 if (lmd->lmd_exclude) {
942 memcpy(lmd->lmd_exclude, exclude_list,
943 sizeof(index) * lmd->lmd_exclude_count);
946 lmd->lmd_exclude_count = 0;
949 OBD_FREE(exclude_list, sizeof(index) * devmax);
953 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
958 if (lmd->lmd_mgssec != NULL) {
959 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
960 lmd->lmd_mgssec = NULL;
963 tail = strchr(ptr, ',');
965 length = strlen(ptr);
969 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
970 if (lmd->lmd_mgssec == NULL)
973 memcpy(lmd->lmd_mgssec, ptr, length);
974 lmd->lmd_mgssec[length] = '\0';
978 static int lmd_parse_string(char **handle, char *ptr)
983 if ((handle == NULL) || (ptr == NULL))
986 if (*handle != NULL) {
987 OBD_FREE(*handle, strlen(*handle) + 1);
991 tail = strchr(ptr, ',');
993 length = strlen(ptr);
997 OBD_ALLOC(*handle, length + 1);
1001 memcpy(*handle, ptr, length);
1002 (*handle)[length] = '\0';
1007 /* Collect multiple values for mgsnid specifiers */
1008 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
1016 /* Find end of nidlist */
1017 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
1018 length = tail - *ptr;
1020 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
1024 if (lmd->lmd_mgs != NULL)
1025 oldlen = strlen(lmd->lmd_mgs) + 1;
1027 OBD_ALLOC(mgsnid, oldlen + length + 1);
1031 if (lmd->lmd_mgs != NULL) {
1032 /* Multiple mgsnid= are taken to mean failover locations */
1033 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
1034 mgsnid[oldlen - 1] = ':';
1035 OBD_FREE(lmd->lmd_mgs, oldlen);
1037 memcpy(mgsnid + oldlen, *ptr, length);
1038 mgsnid[oldlen + length] = '\0';
1039 lmd->lmd_mgs = mgsnid;
1046 * Find the first comma delimiter from the specified \a buf and make \a *endh
1047 * point to the string starting with the comma. The commas in expression list
1048 * [...] will be skipped.
1050 * \param[in] buf a comma-separated string
1051 * \param[in] endh a pointer to a pointer that will point to the string
1052 * starting with the comma
1054 * \retval 0 if comma delimiter is found
1055 * \retval 1 if comma delimiter is not found
1057 static int lmd_find_comma(char *buf, char **endh)
1065 while (*c != '\0') {
1071 if (*c == ',' && skip == 0) {
1084 * Find the first valid string delimited by comma from the specified \a buf
1085 # and parse it to see whether it's a valid nid list. If yes, \a *endh will
1086 * point to the next string starting with the comma.
1088 * \param[in] buf a comma-separated string
1089 * \param[in] endh a pointer to a pointer that will point to the string
1090 * starting with the comma
1092 * \retval 0 if the string is a valid nid list
1093 * \retval 1 if the string is not a valid nid list
1095 static int lmd_parse_nidlist(char *buf, char **endh)
1097 struct list_head nidlist;
1104 while (*buf == ',' || *buf == ':')
1106 if (*buf == ' ' || *buf == '/' || *buf == '\0')
1109 if (lmd_find_comma(buf, &endp) != 0)
1110 endp = buf + strlen(buf);
1115 INIT_LIST_HEAD(&nidlist);
1116 if (cfs_parse_nidlist(buf, strlen(buf), &nidlist) <= 0)
1118 cfs_free_nidlist(&nidlist);
1128 /** Parse mount line options
1129 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
1130 * dev is passed as device=uml1:/lustre by mount.lustre
1132 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1134 char *s1, *s2, *s3, *devname = NULL;
1135 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1141 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1142 "/sbin/mount.lustre is installed.\n");
1146 /* Options should be a string - try to detect old lmd data */
1147 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1148 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1149 "/sbin/mount.lustre. Please install "
1150 "version %s\n", LUSTRE_VERSION_STRING);
1153 lmd->lmd_magic = LMD_MAGIC;
1155 OBD_ALLOC(lmd->lmd_params, LMD_PARAMS_MAXLEN);
1156 if (lmd->lmd_params == NULL)
1158 lmd->lmd_params[0] = '\0';
1160 /* Set default flags here */
1165 int time_min = OBD_RECOVERY_TIME_MIN;
1167 /* Skip whitespace and extra commas */
1168 while (*s1 == ' ' || *s1 == ',')
1172 /* Client options are parsed in ll_options: eg. flock,
1175 /* Parse non-ldiskfs options here. Rather than modifying
1176 ldiskfs, we just zero these out here */
1177 if (strncmp(s1, "abort_recov", 11) == 0) {
1178 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1180 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
1181 lmd->lmd_recovery_time_soft = max_t(int,
1182 simple_strtoul(s1 + 19, NULL, 10), time_min);
1184 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
1185 lmd->lmd_recovery_time_hard = max_t(int,
1186 simple_strtoul(s1 + 19, NULL, 10), time_min);
1188 } else if (strncmp(s1, "noir", 4) == 0) {
1189 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
1191 } else if (strncmp(s1, "nosvc", 5) == 0) {
1192 lmd->lmd_flags |= LMD_FLG_NOSVC;
1194 } else if (strncmp(s1, "nomgs", 5) == 0) {
1195 lmd->lmd_flags |= LMD_FLG_NOMGS;
1197 } else if (strncmp(s1, "noscrub", 7) == 0) {
1198 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
1200 } else if (strncmp(s1, "skip_lfsck", 10) == 0) {
1201 lmd->lmd_flags |= LMD_FLG_SKIP_LFSCK;
1203 } else if (strncmp(s1, PARAM_MGSNODE,
1204 sizeof(PARAM_MGSNODE) - 1) == 0) {
1205 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
1206 /* Assume the next mount opt is the first
1207 invalid nid we get to. */
1208 rc = lmd_parse_mgs(lmd, &s2);
1213 } else if (strncmp(s1, "writeconf", 9) == 0) {
1214 lmd->lmd_flags |= LMD_FLG_WRITECONF;
1216 } else if (strncmp(s1, "update", 6) == 0) {
1217 lmd->lmd_flags |= LMD_FLG_UPDATE;
1219 } else if (strncmp(s1, "virgin", 6) == 0) {
1220 lmd->lmd_flags |= LMD_FLG_VIRGIN;
1222 } else if (strncmp(s1, "noprimnode", 10) == 0) {
1223 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
1225 } else if (strncmp(s1, "mgssec=", 7) == 0) {
1226 rc = lmd_parse_mgssec(lmd, s1 + 7);
1230 /* ost exclusion list */
1231 } else if (strncmp(s1, "exclude=", 8) == 0) {
1232 rc = lmd_make_exclusion(lmd, s1 + 7);
1236 } else if (strncmp(s1, "mgs", 3) == 0) {
1238 lmd->lmd_flags |= LMD_FLG_MGS;
1240 } else if (strncmp(s1, "svname=", 7) == 0) {
1241 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
1245 } else if (strncmp(s1, "param=", 6) == 0) {
1246 size_t length, params_length;
1248 if (lmd_find_comma(s1 + 6, &tail) != 0)
1249 length = strlen(s1);
1251 char *param_str = tail + 1;
1252 int supplementary = 1;
1253 while (lmd_parse_nidlist(param_str,
1257 length = param_str - s1 - supplementary;
1260 params_length = strlen(lmd->lmd_params);
1261 if (params_length + length + 1 >= LMD_PARAMS_MAXLEN)
1263 strncat(lmd->lmd_params, s1 + 6, length);
1264 lmd->lmd_params[params_length + length] = '\0';
1265 strlcat(lmd->lmd_params, " ", LMD_PARAMS_MAXLEN);
1266 s3 = s1 + 6 + length;
1268 } else if (strncmp(s1, "osd=", 4) == 0) {
1269 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
1274 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1275 end of the options. */
1276 else if (strncmp(s1, "device=", 7) == 0) {
1278 /* terminate options right before device. device
1279 must be the last one. */
1285 s2 = strchr(s3, ',');
1293 memmove(s1, s2, strlen(s2) + 1);
1299 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1300 "(need mount option 'device=...')\n");
1304 s1 = strstr(devname, ":/");
1307 lmd->lmd_flags |= LMD_FLG_CLIENT;
1308 /* Remove leading /s from fsname */
1309 while (*++s1 == '/')
1312 while (*s2 != '/' && *s2 != '\0')
1314 /* Freed in lustre_free_lsi */
1315 OBD_ALLOC(lmd->lmd_profile, s2 - s1 + 8);
1316 if (!lmd->lmd_profile)
1319 strncat(lmd->lmd_profile, s1, s2 - s1);
1320 strncat(lmd->lmd_profile, "-client", 7);
1323 s2 = s1 + strlen(s1) - 1;
1324 /* Remove padding /s from fileset */
1328 OBD_ALLOC(lmd->lmd_fileset, s2 - s1 + 2);
1329 if (lmd->lmd_fileset == NULL) {
1330 OBD_FREE(lmd->lmd_profile,
1331 strlen(lmd->lmd_profile) + 1);
1334 strncat(lmd->lmd_fileset, s1, s2 - s1 + 1);
1338 /* Freed in lustre_free_lsi */
1339 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1342 strcpy(lmd->lmd_dev, devname);
1344 /* Save mount options */
1345 s1 = options + strlen(options) - 1;
1346 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1348 if (*options != 0) {
1349 /* Freed in lustre_free_lsi */
1350 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1353 strcpy(lmd->lmd_opts, options);
1357 lmd->lmd_magic = LMD_MAGIC;
1362 CERROR("Bad mount options %s\n", options);
1366 struct lustre_mount_data2 {
1368 struct vfsmount *lmd2_mnt;
1371 /** This is the entry point for the mount call into Lustre.
1372 * This is called when a server or client is mounted,
1373 * and this is where we start setting things up.
1374 * @param data Mount options (e.g. -o flock,abort_recov)
1376 static int lustre_fill_super(struct super_block *sb, void *data, int silent)
1378 struct lustre_mount_data *lmd;
1379 struct lustre_mount_data2 *lmd2 = data;
1380 struct lustre_sb_info *lsi;
1384 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1386 lsi = lustre_init_lsi(sb);
1392 * Disable lockdep during mount, because mount locking patterns are
1398 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
1400 obd_zombie_barrier();
1402 /* Figure out the lmd from the mount options */
1403 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
1405 GOTO(out, rc = -EINVAL);
1408 if (lmd_is_client(lmd)) {
1409 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1410 if (client_fill_super == NULL)
1411 request_module("lustre");
1412 if (client_fill_super == NULL) {
1413 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1414 "client mount! Is the 'lustre' "
1415 "module loaded?\n");
1419 rc = lustre_start_mgc(sb);
1424 /* Connect and start */
1425 /* (should always be ll_fill_super) */
1426 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
1427 /* c_f_s will call lustre_common_put_super on failure */
1430 #ifdef HAVE_SERVER_SUPPORT
1431 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1432 rc = server_fill_super(sb);
1433 /* s_f_s calls lustre_start_mgc after the mount because we need
1434 the MGS nids which are stored on disk. Plus, we may
1435 need to start the MGS first. */
1436 /* s_f_s will call server_put_super on failure */
1438 CERROR("This is client-side-only module, "
1439 "cannot handle server mount.\n");
1444 /* If error happens in fill_super() call, @lsi will be killed there.
1445 * This is why we do not put it here. */
1449 CERROR("Unable to mount %s (%d)\n",
1450 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1452 CDEBUG(D_SUPER, "Mount %s complete\n",
1460 /* We can't call ll_fill_super by name because it lives in a module that
1461 must be loaded after this one. */
1462 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
1463 struct vfsmount *mnt))
1465 client_fill_super = cfs;
1467 EXPORT_SYMBOL(lustre_register_client_fill_super);
1469 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
1471 kill_super_cb = cfs;
1473 EXPORT_SYMBOL(lustre_register_kill_super_cb);
1475 /***************** FS registration ******************/
1476 #ifdef HAVE_FSTYPE_MOUNT
1477 static struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
1478 const char *devname, void *data)
1480 struct lustre_mount_data2 lmd2 = { data, NULL };
1482 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
1485 static int lustre_get_sb(struct file_system_type *fs_type, int flags,
1486 const char *devname, void *data, struct vfsmount *mnt)
1488 struct lustre_mount_data2 lmd2 = { data, mnt };
1490 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
1494 static void lustre_kill_super(struct super_block *sb)
1496 struct lustre_sb_info *lsi = s2lsi(sb);
1498 if (kill_super_cb && lsi && !IS_SERVER(lsi))
1499 (*kill_super_cb)(sb);
1501 kill_anon_super(sb);
1504 /** Register the "lustre" fs type
1506 static struct file_system_type lustre_fs_type = {
1507 .owner = THIS_MODULE,
1509 #ifdef HAVE_FSTYPE_MOUNT
1510 .mount = lustre_mount,
1512 .get_sb = lustre_get_sb,
1514 .kill_sb = lustre_kill_super,
1515 .fs_flags = FS_REQUIRES_DEV | FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE,
1517 MODULE_ALIAS_FS("lustre");
1519 int lustre_register_fs(void)
1521 return register_filesystem(&lustre_fs_type);
1524 int lustre_unregister_fs(void)
1526 return unregister_filesystem(&lustre_fs_type);