4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/obdclass/obd_mount.c
38 * Client mount routines
40 * Author: Nathan Rutman <nathan@clusterfs.com>
44 #define DEBUG_SUBSYSTEM S_CLASS
45 #define D_MOUNT (D_SUPER|D_CONFIG/*|D_WARNING */)
46 #define PRINT_CMD CDEBUG
50 #include <obd_class.h>
51 #include <lustre/lustre_user.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;
91 /* mgc_process_config */
92 lustre_cfg_bufs_reset(bufs, mgc->obd_name);
93 lustre_cfg_bufs_set_string(bufs, 1, logname);
94 lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg));
95 lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb));
96 lcfg = lustre_cfg_new(LCFG_LOG_START, bufs);
97 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
98 lustre_cfg_free(lcfg);
103 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
104 "failed from the MGS (%d). Make sure this "
105 "client and the MGS are running compatible "
106 "versions of Lustre.\n",
107 mgc->obd_name, logname, rc);
110 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
111 "failed (%d). This may be the result of "
112 "communication errors between this node and "
113 "the MGS, a bad configuration, or other "
114 "errors. See the syslog for more "
115 "information.\n", mgc->obd_name, logname,
118 /* class_obd_list(); */
121 EXPORT_SYMBOL(lustre_process_log);
123 /* Stop watching this config log for updates */
124 int lustre_end_log(struct super_block *sb, char *logname,
125 struct config_llog_instance *cfg)
127 struct lustre_cfg *lcfg;
128 struct lustre_cfg_bufs bufs;
129 struct lustre_sb_info *lsi = s2lsi(sb);
130 struct obd_device *mgc = lsi->lsi_mgc;
136 /* mgc_process_config */
137 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
138 lustre_cfg_bufs_set_string(&bufs, 1, logname);
140 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
141 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
142 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
143 lustre_cfg_free(lcfg);
146 EXPORT_SYMBOL(lustre_end_log);
148 /**************** obd start *******************/
150 /** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from
151 * lctl (and do for echo cli/srv.
153 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
154 char *s1, char *s2, char *s3, char *s4)
156 struct lustre_cfg_bufs bufs;
157 struct lustre_cfg * lcfg = NULL;
160 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
161 cmd, s1, s2, s3, s4);
163 lustre_cfg_bufs_reset(&bufs, cfgname);
165 lustre_cfg_bufs_set_string(&bufs, 1, s1);
167 lustre_cfg_bufs_set_string(&bufs, 2, s2);
169 lustre_cfg_bufs_set_string(&bufs, 3, s3);
171 lustre_cfg_bufs_set_string(&bufs, 4, s4);
173 lcfg = lustre_cfg_new(cmd, &bufs);
174 lcfg->lcfg_nid = nid;
175 rc = class_process_config(lcfg);
176 lustre_cfg_free(lcfg);
179 EXPORT_SYMBOL(do_lcfg);
181 /** Call class_attach and class_setup. These methods in turn call
182 * obd type-specific methods.
184 int lustre_start_simple(char *obdname, char *type, char *uuid,
185 char *s1, char *s2, char *s3, char *s4)
188 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
190 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
192 CERROR("%s attach error %d\n", obdname, rc);
195 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, s3, s4);
197 CERROR("%s setup error %d\n", obdname, rc);
198 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
203 DEFINE_MUTEX(mgc_start_lock);
205 /** Set up a mgc obd to process startup logs
207 * \param sb [in] super block of the mgc obd
209 * \retval 0 success, otherwise error code
211 int lustre_start_mgc(struct super_block *sb)
213 struct obd_connect_data *data = NULL;
214 struct lustre_sb_info *lsi = s2lsi(sb);
215 struct obd_device *obd;
216 struct obd_export *exp;
217 struct obd_uuid *uuid;
220 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
223 int rc = 0, i = 0, j, len;
225 LASSERT(lsi->lsi_lmd);
227 /* Find the first non-lo MGS nid for our MGC name */
228 if (IS_SERVER(lsi)) {
229 /* mount -o mgsnode=nid */
230 ptr = lsi->lsi_lmd->lmd_mgs;
231 if (lsi->lsi_lmd->lmd_mgs &&
232 (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) {
234 } else if (IS_MGS(lsi)) {
235 lnet_process_id_t id;
236 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
237 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
244 } else { /* client */
245 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
246 ptr = lsi->lsi_lmd->lmd_dev;
247 if (class_parse_nid(ptr, &nid, &ptr) == 0)
251 CERROR("No valid MGS nids found.\n");
255 mutex_lock(&mgc_start_lock);
257 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
258 OBD_ALLOC(mgcname, len);
259 OBD_ALLOC(niduuid, len + 2);
260 if (!mgcname || !niduuid)
261 GOTO(out_free, rc = -ENOMEM);
262 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
264 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
268 GOTO(out_free, rc = -ENOMEM);
270 obd = class_name2obd(mgcname);
271 if (obd && !obd->obd_stopping) {
272 rc = obd_set_info_async(NULL, obd->obd_self_export,
273 strlen(KEY_MGSSEC), KEY_MGSSEC,
274 strlen(mgssec), mgssec, NULL);
278 /* Re-using an existing MGC */
279 atomic_inc(&obd->u.cli.cl_mgc_refcount);
281 /* IR compatibility check, only for clients */
282 if (lmd_is_client(lsi->lsi_lmd)) {
284 int vallen = sizeof(*data);
285 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
287 rc = obd_get_info(NULL, obd->obd_self_export,
288 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
289 &vallen, data, NULL);
291 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
292 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
293 /* LMD_FLG_NOIR is for test purpose only */
295 "Trying to mount a client with IR setting "
296 "not compatible with current mgc. "
297 "Force to use current mgc setting that is "
299 has_ir ? "enabled" : "disabled");
301 *flags &= ~LMD_FLG_NOIR;
303 *flags |= LMD_FLG_NOIR;
308 /* If we are restarting the MGS, don't try to keep the MGC's
309 old connection, or registration will fail. */
311 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
315 /* Try all connections, but only once (again).
316 We don't want to block another target from starting
317 (using its local copy of the log), but we do want to connect
318 if at all possible. */
320 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
321 rc = obd_set_info_async(NULL, obd->obd_self_export,
322 sizeof(KEY_INIT_RECOV_BACKUP),
323 KEY_INIT_RECOV_BACKUP,
324 sizeof(recov_bk), &recov_bk, NULL);
328 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
330 /* Add the primary nids for the MGS */
332 sprintf(niduuid, "%s_%x", mgcname, i);
333 if (IS_SERVER(lsi)) {
334 ptr = lsi->lsi_lmd->lmd_mgs;
336 /* Use local nids (including LO) */
337 lnet_process_id_t id;
338 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
339 rc = do_lcfg(mgcname, id.nid,
340 LCFG_ADD_UUID, niduuid, 0,0,0);
343 /* Use mgsnode= nids */
344 /* mount -o mgsnode=nid */
345 if (lsi->lsi_lmd->lmd_mgs) {
346 ptr = lsi->lsi_lmd->lmd_mgs;
347 } else if (class_find_param(ptr, PARAM_MGSNODE,
349 CERROR("No MGS nids given.\n");
350 GOTO(out_free, rc = -EINVAL);
352 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
353 rc = do_lcfg(mgcname, nid,
354 LCFG_ADD_UUID, niduuid, 0,0,0);
358 } else { /* client */
359 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
360 ptr = lsi->lsi_lmd->lmd_dev;
361 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
362 rc = do_lcfg(mgcname, nid,
363 LCFG_ADD_UUID, niduuid, 0,0,0);
365 /* Stop at the first failover nid */
371 CERROR("No valid MGS nids found.\n");
372 GOTO(out_free, rc = -EINVAL);
374 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
376 /* Random uuid for MGC allows easier reconnects */
378 ll_generate_random_uuid(uuidc);
379 class_uuid_unparse(uuidc, uuid);
382 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
383 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
389 /* Add any failover MGS nids */
391 while (ptr && ((*ptr == ':' ||
392 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) {
393 /* New failover node */
394 sprintf(niduuid, "%s_%x", mgcname, i);
396 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
398 rc = do_lcfg(mgcname, nid,
399 LCFG_ADD_UUID, niduuid, 0,0,0);
404 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
412 lsi->lsi_lmd->lmd_mgs_failnodes = i;
414 obd = class_name2obd(mgcname);
416 CERROR("Can't find mgcobd %s\n", mgcname);
417 GOTO(out_free, rc = -ENOTCONN);
420 rc = obd_set_info_async(NULL, obd->obd_self_export,
421 strlen(KEY_MGSSEC), KEY_MGSSEC,
422 strlen(mgssec), mgssec, NULL);
426 /* Keep a refcount of servers/clients who started with "mount",
427 so we know when we can get rid of the mgc. */
428 atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
430 /* Try all connections, but only once. */
432 rc = obd_set_info_async(NULL, obd->obd_self_export,
433 sizeof(KEY_INIT_RECOV_BACKUP),
434 KEY_INIT_RECOV_BACKUP,
435 sizeof(recov_bk), &recov_bk, NULL);
438 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
440 /* We connect to the MGS at setup, and don't disconnect until cleanup */
441 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
442 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV |
443 OBD_CONNECT_LVB_TYPE;
445 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
446 data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB;
448 #warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
451 if (lmd_is_client(lsi->lsi_lmd) &&
452 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
453 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
454 data->ocd_version = LUSTRE_VERSION_CODE;
455 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
457 CERROR("connect failed %d\n", rc);
461 obd->u.cli.cl_mgc_mgsexp = exp;
464 /* Keep the mgc info in the sb. Note that many lsi's can point
468 mutex_unlock(&mgc_start_lock);
473 OBD_FREE(mgcname, len);
475 OBD_FREE(niduuid, len + 2);
479 static int lustre_stop_mgc(struct super_block *sb)
481 struct lustre_sb_info *lsi = s2lsi(sb);
482 struct obd_device *obd;
483 char *niduuid = 0, *ptr = 0;
484 int i, rc = 0, len = 0;
493 mutex_lock(&mgc_start_lock);
494 LASSERT(atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
495 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
496 /* This is not fatal, every client that stops
497 will call in here. */
498 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
499 atomic_read(&obd->u.cli.cl_mgc_refcount));
500 GOTO(out, rc = -EBUSY);
503 /* The MGC has no recoverable data in any case.
504 * force shotdown set in umount_begin */
505 obd->obd_no_recov = 1;
507 if (obd->u.cli.cl_mgc_mgsexp) {
508 /* An error is not fatal, if we are unable to send the
509 disconnect mgs ping evictor cleans up the export */
510 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
512 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
515 /* Save the obdname for cleaning the nid uuids, which are
517 len = strlen(obd->obd_name) + 6;
518 OBD_ALLOC(niduuid, len);
520 strcpy(niduuid, obd->obd_name);
521 ptr = niduuid + strlen(niduuid);
524 rc = class_manual_cleanup(obd);
528 /* Clean the nid uuids */
530 GOTO(out, rc = -ENOMEM);
532 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
533 sprintf(ptr, "_%x", i);
534 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
537 CERROR("del MDC UUID %s failed: rc = %d\n",
542 OBD_FREE(niduuid, len);
544 /* class_import_put will get rid of the additional connections */
545 mutex_unlock(&mgc_start_lock);
549 /***************** lustre superblock **************/
551 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
553 struct lustre_sb_info *lsi;
558 OBD_ALLOC_PTR(lsi->lsi_lmd);
564 lsi->lsi_lmd->lmd_exclude_count = 0;
565 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
566 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
567 s2lsi_nocast(sb) = lsi;
568 /* we take 1 extra ref for our setup */
569 atomic_set(&lsi->lsi_mounts, 1);
571 /* Default umount style */
572 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
577 static int lustre_free_lsi(struct super_block *sb)
579 struct lustre_sb_info *lsi = s2lsi(sb);
581 LASSERT(lsi != NULL);
582 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
584 /* someone didn't call server_put_mount. */
585 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
587 if (lsi->lsi_lmd != NULL) {
588 if (lsi->lsi_lmd->lmd_dev != NULL)
589 OBD_FREE(lsi->lsi_lmd->lmd_dev,
590 strlen(lsi->lsi_lmd->lmd_dev) + 1);
591 if (lsi->lsi_lmd->lmd_profile != NULL)
592 OBD_FREE(lsi->lsi_lmd->lmd_profile,
593 strlen(lsi->lsi_lmd->lmd_profile) + 1);
594 if (lsi->lsi_lmd->lmd_mgssec != NULL)
595 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
596 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
597 if (lsi->lsi_lmd->lmd_opts != NULL)
598 OBD_FREE(lsi->lsi_lmd->lmd_opts,
599 strlen(lsi->lsi_lmd->lmd_opts) + 1);
600 if (lsi->lsi_lmd->lmd_exclude_count)
601 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
602 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
603 lsi->lsi_lmd->lmd_exclude_count);
604 if (lsi->lsi_lmd->lmd_mgs != NULL)
605 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
606 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
607 if (lsi->lsi_lmd->lmd_osd_type != NULL)
608 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
609 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
610 if (lsi->lsi_lmd->lmd_params != NULL)
611 OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
613 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
616 LASSERT(lsi->lsi_llsbi == NULL);
617 OBD_FREE(lsi, sizeof(*lsi));
618 s2lsi_nocast(sb) = NULL;
623 /* The lsi has one reference for every server that is using the disk -
624 e.g. MDT, MGS, and potentially MGC */
625 int lustre_put_lsi(struct super_block *sb)
627 struct lustre_sb_info *lsi = s2lsi(sb);
629 LASSERT(lsi != NULL);
631 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
632 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
633 if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
634 lu_device_put(&lsi->lsi_dt_dev->dd_lu_dev);
635 lsi->lsi_osd_exp->exp_obd->obd_lvfs_ctxt.dt = NULL;
636 lsi->lsi_dt_dev = NULL;
637 obd_disconnect(lsi->lsi_osd_exp);
638 /* wait till OSD is gone */
639 obd_zombie_barrier();
648 * <FSNAME><SEPERATOR><TYPE><INDEX>
649 * FSNAME is between 1 and 8 characters (inclusive).
650 * Excluded characters are '/' and ':'
651 * SEPERATOR is either ':' or '-'
652 * TYPE: "OST", "MDT", etc.
653 * INDEX: Hex representation of the index
656 /** Get the fsname ("lustre") from the server name ("lustre-OST003F").
657 * @param [in] svname server name including type and index
658 * @param [out] fsname Buffer to copy filesystem name prefix into.
659 * Must have at least 'strlen(fsname) + 1' chars.
660 * @param [out] endptr if endptr isn't NULL it is set to end of fsname
663 int server_name2fsname(const char *svname, char *fsname, const char **endptr)
667 dash = svname + strnlen(svname, 8); /* max fsname length is 8 */
668 for (; dash > svname && *dash != '-' && *dash != ':'; dash--)
673 if (fsname != NULL) {
674 strncpy(fsname, svname, dash - svname);
675 fsname[dash - svname] = '\0';
683 EXPORT_SYMBOL(server_name2fsname);
686 * Get service name (svname) from string
688 * if endptr isn't NULL it is set to end of fsname *
690 int server_name2svname(const char *label, char *svname, const char **endptr,
696 /* We use server_name2fsname() just for parsing */
697 rc = server_name2fsname(label, NULL, &dash);
704 if (strlcpy(svname, dash + 1, svsize) >= svsize)
709 EXPORT_SYMBOL(server_name2svname);
712 /* Get the index from the obd name.
715 if endptr isn't NULL it is set to end of name */
716 int server_name2index(const char *svname, __u32 *idx, const char **endptr)
722 /* We use server_name2fsname() just for parsing */
723 rc = server_name2fsname(svname, NULL, &dash);
729 if (strncmp(dash, "MDT", 3) == 0)
730 rc = LDD_F_SV_TYPE_MDT;
731 else if (strncmp(dash, "OST", 3) == 0)
732 rc = LDD_F_SV_TYPE_OST;
738 if (strncmp(dash, "all", 3) == 0) {
741 return rc | LDD_F_SV_ALL;
744 index = simple_strtoul(dash, (char **)endptr, 16);
748 /* Account for -mdc after index that is possible when specifying mdt */
749 if (endptr != NULL && strncmp(LUSTRE_MDC_NAME, *endptr + 1,
750 sizeof(LUSTRE_MDC_NAME)-1) == 0)
751 *endptr += sizeof(LUSTRE_MDC_NAME);
755 EXPORT_SYMBOL(server_name2index);
757 /*************** mount common betweeen server and client ***************/
760 int lustre_common_put_super(struct super_block *sb)
764 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
766 /* Drop a ref to the MGC */
767 rc = lustre_stop_mgc(sb);
768 if (rc && (rc != -ENOENT)) {
770 CERROR("Can't stop MGC: %d\n", rc);
773 /* BUSY just means that there's some other obd that
774 needs the mgc. Let him clean it up. */
775 CDEBUG(D_MOUNT, "MGC still in use\n");
777 /* Drop a ref to the mounted disk */
782 EXPORT_SYMBOL(lustre_common_put_super);
784 static void lmd_print(struct lustre_mount_data *lmd)
788 PRINT_CMD(D_MOUNT, " mount data:\n");
789 if (lmd_is_client(lmd))
790 PRINT_CMD(D_MOUNT, "profile: %s\n", lmd->lmd_profile);
791 PRINT_CMD(D_MOUNT, "device: %s\n", lmd->lmd_dev);
792 PRINT_CMD(D_MOUNT, "flags: %x\n", lmd->lmd_flags);
795 PRINT_CMD(D_MOUNT, "options: %s\n", lmd->lmd_opts);
797 if (lmd->lmd_recovery_time_soft)
798 PRINT_CMD(D_MOUNT, "recovery time soft: %d\n",
799 lmd->lmd_recovery_time_soft);
801 if (lmd->lmd_recovery_time_hard)
802 PRINT_CMD(D_MOUNT, "recovery time hard: %d\n",
803 lmd->lmd_recovery_time_hard);
805 for (i = 0; i < lmd->lmd_exclude_count; i++) {
806 PRINT_CMD(D_MOUNT, "exclude %d: OST%04x\n", i,
807 lmd->lmd_exclude[i]);
811 /* Is this server on the exclusion list */
812 int lustre_check_exclusion(struct super_block *sb, char *svname)
814 struct lustre_sb_info *lsi = s2lsi(sb);
815 struct lustre_mount_data *lmd = lsi->lsi_lmd;
819 rc = server_name2index(svname, &index, NULL);
820 if (rc != LDD_F_SV_TYPE_OST)
821 /* Only exclude OSTs */
824 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
825 index, lmd->lmd_exclude_count, lmd->lmd_dev);
827 for(i = 0; i < lmd->lmd_exclude_count; i++) {
828 if (index == lmd->lmd_exclude[i]) {
829 CWARN("Excluding %s (on exclusion list)\n", svname);
836 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
837 static int lmd_make_exclusion(struct lustre_mount_data *lmd, const char *ptr)
839 const char *s1 = ptr, *s2;
840 __u32 index, *exclude_list;
843 /* The shortest an ost name can be is 8 chars: -OST0000.
844 We don't actually know the fsname at this time, so in fact
845 a user could specify any fsname. */
846 devmax = strlen(ptr) / 8 + 1;
848 /* temp storage until we figure out how many we have */
849 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
853 /* we enter this fn pointing at the '=' */
854 while (*s1 && *s1 != ' ' && *s1 != ',') {
856 rc = server_name2index(s1, &index, &s2);
858 CERROR("Can't parse server name '%s': rc = %d\n",
862 if (rc == LDD_F_SV_TYPE_OST)
863 exclude_list[lmd->lmd_exclude_count++] = index;
865 CDEBUG(D_MOUNT, "ignoring exclude %.*s: type = %#x\n",
866 (uint)(s2-s1), s1, rc);
868 /* now we are pointing at ':' (next exclude)
869 or ',' (end of excludes) */
870 if (lmd->lmd_exclude_count >= devmax)
873 if (rc >= 0) /* non-err */
876 if (lmd->lmd_exclude_count) {
877 /* permanent, freed in lustre_free_lsi */
878 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
879 lmd->lmd_exclude_count);
880 if (lmd->lmd_exclude) {
881 memcpy(lmd->lmd_exclude, exclude_list,
882 sizeof(index) * lmd->lmd_exclude_count);
885 lmd->lmd_exclude_count = 0;
888 OBD_FREE(exclude_list, sizeof(index) * devmax);
892 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
897 if (lmd->lmd_mgssec != NULL) {
898 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
899 lmd->lmd_mgssec = NULL;
902 tail = strchr(ptr, ',');
904 length = strlen(ptr);
908 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
909 if (lmd->lmd_mgssec == NULL)
912 memcpy(lmd->lmd_mgssec, ptr, length);
913 lmd->lmd_mgssec[length] = '\0';
917 static int lmd_parse_string(char **handle, char *ptr)
922 if ((handle == NULL) || (ptr == NULL))
925 if (*handle != NULL) {
926 OBD_FREE(*handle, strlen(*handle) + 1);
930 tail = strchr(ptr, ',');
932 length = strlen(ptr);
936 OBD_ALLOC(*handle, length + 1);
940 memcpy(*handle, ptr, length);
941 (*handle)[length] = '\0';
946 /* Collect multiple values for mgsnid specifiers */
947 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
955 /* Find end of nidlist */
956 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
957 length = tail - *ptr;
959 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
963 if (lmd->lmd_mgs != NULL)
964 oldlen = strlen(lmd->lmd_mgs) + 1;
966 OBD_ALLOC(mgsnid, oldlen + length + 1);
970 if (lmd->lmd_mgs != NULL) {
971 /* Multiple mgsnid= are taken to mean failover locations */
972 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
973 mgsnid[oldlen - 1] = ':';
974 OBD_FREE(lmd->lmd_mgs, oldlen);
976 memcpy(mgsnid + oldlen, *ptr, length);
977 mgsnid[oldlen + length] = '\0';
978 lmd->lmd_mgs = mgsnid;
984 /** Parse mount line options
985 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
986 * dev is passed as device=uml1:/lustre by mount.lustre
988 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
990 char *s1, *s2, *devname = NULL;
991 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
996 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
997 "/sbin/mount.lustre is installed.\n");
1001 /* Options should be a string - try to detect old lmd data */
1002 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1003 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1004 "/sbin/mount.lustre. Please install "
1005 "version %s\n", LUSTRE_VERSION_STRING);
1008 lmd->lmd_magic = LMD_MAGIC;
1010 OBD_ALLOC(lmd->lmd_params, 4096);
1011 if (lmd->lmd_params == NULL)
1013 lmd->lmd_params[0] = '\0';
1015 /* Set default flags here */
1020 int time_min = OBD_RECOVERY_TIME_MIN;
1022 /* Skip whitespace and extra commas */
1023 while (*s1 == ' ' || *s1 == ',')
1026 /* Client options are parsed in ll_options: eg. flock,
1029 /* Parse non-ldiskfs options here. Rather than modifying
1030 ldiskfs, we just zero these out here */
1031 if (strncmp(s1, "abort_recov", 11) == 0) {
1032 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1034 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
1035 lmd->lmd_recovery_time_soft = max_t(int,
1036 simple_strtoul(s1 + 19, NULL, 10), time_min);
1038 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
1039 lmd->lmd_recovery_time_hard = max_t(int,
1040 simple_strtoul(s1 + 19, NULL, 10), time_min);
1042 } else if (strncmp(s1, "noir", 4) == 0) {
1043 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
1045 } else if (strncmp(s1, "nosvc", 5) == 0) {
1046 lmd->lmd_flags |= LMD_FLG_NOSVC;
1048 } else if (strncmp(s1, "nomgs", 5) == 0) {
1049 lmd->lmd_flags |= LMD_FLG_NOMGS;
1051 } else if (strncmp(s1, "noscrub", 7) == 0) {
1052 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
1054 } else if (strncmp(s1, PARAM_MGSNODE,
1055 sizeof(PARAM_MGSNODE) - 1) == 0) {
1056 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
1057 /* Assume the next mount opt is the first
1058 invalid nid we get to. */
1059 rc = lmd_parse_mgs(lmd, &s2);
1063 } else if (strncmp(s1, "writeconf", 9) == 0) {
1064 lmd->lmd_flags |= LMD_FLG_WRITECONF;
1066 } else if (strncmp(s1, "update", 6) == 0) {
1067 lmd->lmd_flags |= LMD_FLG_UPDATE;
1069 } else if (strncmp(s1, "virgin", 6) == 0) {
1070 lmd->lmd_flags |= LMD_FLG_VIRGIN;
1072 } else if (strncmp(s1, "noprimnode", 10) == 0) {
1073 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
1075 } else if (strncmp(s1, "mgssec=", 7) == 0) {
1076 rc = lmd_parse_mgssec(lmd, s1 + 7);
1080 /* ost exclusion list */
1081 } else if (strncmp(s1, "exclude=", 8) == 0) {
1082 rc = lmd_make_exclusion(lmd, s1 + 7);
1086 } else if (strncmp(s1, "mgs", 3) == 0) {
1088 lmd->lmd_flags |= LMD_FLG_MGS;
1090 } else if (strncmp(s1, "svname=", 7) == 0) {
1091 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
1095 } else if (strncmp(s1, "param=", 6) == 0) {
1097 char *tail = strchr(s1 + 6, ',');
1099 length = strlen(s1);
1103 strncat(lmd->lmd_params, s1 + 6, length);
1104 strcat(lmd->lmd_params, " ");
1106 } else if (strncmp(s1, "osd=", 4) == 0) {
1107 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
1112 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1113 end of the options. */
1114 else if (strncmp(s1, "device=", 7) == 0) {
1116 /* terminate options right before device. device
1117 must be the last one. */
1123 s2 = strchr(s1, ',');
1131 memmove(s1, s2, strlen(s2) + 1);
1137 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1138 "(need mount option 'device=...')\n");
1142 s1 = strstr(devname, ":/");
1145 lmd->lmd_flags |= LMD_FLG_CLIENT;
1146 /* Remove leading /s from fsname */
1147 while (*++s1 == '/') ;
1148 /* Freed in lustre_free_lsi */
1149 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1150 if (!lmd->lmd_profile)
1152 sprintf(lmd->lmd_profile, "%s-client", s1);
1155 /* Freed in lustre_free_lsi */
1156 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1159 strcpy(lmd->lmd_dev, devname);
1161 /* Save mount options */
1162 s1 = options + strlen(options) - 1;
1163 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1165 if (*options != 0) {
1166 /* Freed in lustre_free_lsi */
1167 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1170 strcpy(lmd->lmd_opts, options);
1174 lmd->lmd_magic = LMD_MAGIC;
1179 CERROR("Bad mount options %s\n", options);
1183 struct lustre_mount_data2 {
1185 struct vfsmount *lmd2_mnt;
1188 /** This is the entry point for the mount call into Lustre.
1189 * This is called when a server or client is mounted,
1190 * and this is where we start setting things up.
1191 * @param data Mount options (e.g. -o flock,abort_recov)
1193 int lustre_fill_super(struct super_block *sb, void *data, int silent)
1195 struct lustre_mount_data *lmd;
1196 struct lustre_mount_data2 *lmd2 = data;
1197 struct lustre_sb_info *lsi;
1200 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1202 lsi = lustre_init_lsi(sb);
1208 * Disable lockdep during mount, because mount locking patterns are
1214 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
1216 obd_zombie_barrier();
1218 /* Figure out the lmd from the mount options */
1219 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
1221 GOTO(out, rc = -EINVAL);
1224 if (lmd_is_client(lmd)) {
1225 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1226 if (!client_fill_super) {
1227 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1228 "client mount! Is the 'lustre' "
1229 "module loaded?\n");
1233 rc = lustre_start_mgc(sb);
1238 /* Connect and start */
1239 /* (should always be ll_fill_super) */
1240 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
1241 /* c_f_s will call lustre_common_put_super on failure */
1244 CERROR("This is client-side-only module, "
1245 "cannot handle server mount.\n");
1249 /* If error happens in fill_super() call, @lsi will be killed there.
1250 * This is why we do not put it here. */
1254 CERROR("Unable to mount %s (%d)\n",
1255 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1257 CDEBUG(D_SUPER, "Mount %s complete\n",
1265 /* We can't call ll_fill_super by name because it lives in a module that
1266 must be loaded after this one. */
1267 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
1268 struct vfsmount *mnt))
1270 client_fill_super = cfs;
1272 EXPORT_SYMBOL(lustre_register_client_fill_super);
1274 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
1276 kill_super_cb = cfs;
1278 EXPORT_SYMBOL(lustre_register_kill_super_cb);
1280 /***************** FS registration ******************/
1281 struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
1282 const char *devname, void *data)
1284 struct lustre_mount_data2 lmd2 = { data, NULL };
1286 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
1289 void lustre_kill_super(struct super_block *sb)
1291 struct lustre_sb_info *lsi = s2lsi(sb);
1293 if (kill_super_cb && lsi && !IS_SERVER(lsi))
1294 (*kill_super_cb)(sb);
1296 kill_anon_super(sb);
1299 /** Register the "lustre" fs type
1301 struct file_system_type lustre_fs_type = {
1302 .owner = THIS_MODULE,
1304 .mount = lustre_mount,
1305 .kill_sb = lustre_kill_super,
1306 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
1307 FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE,
1310 int lustre_register_fs(void)
1312 return register_filesystem(&lustre_fs_type);
1315 int lustre_unregister_fs(void)
1317 return unregister_filesystem(&lustre_fs_type);