4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/obdclass/obd_mount.c
38 * Client mount routines
40 * Author: Nathan Rutman <nathan@clusterfs.com>
44 #define DEBUG_SUBSYSTEM S_CLASS
45 #define D_MOUNT (D_SUPER|D_CONFIG/*|D_WARNING */)
46 #define PRINT_CMD CDEBUG
50 #include <obd_class.h>
51 #include <lustre/lustre_user.h>
52 #include <linux/version.h>
53 #include <lustre_log.h>
54 #include <lustre_disk.h>
55 #include <lustre_param.h>
57 static int (*client_fill_super)(struct super_block *sb,
58 struct vfsmount *mnt);
60 static void (*kill_super_cb)(struct super_block *sb);
62 /**************** config llog ********************/
64 /** Get a config log from the MGS and process it.
65 * This func is called for both clients and servers.
66 * Continue to process new statements appended to the logs
67 * (whenever the config lock is revoked) until lustre_end_log
69 * @param sb The superblock is used by the MGC to write to the local copy of
71 * @param logname The name of the llog to replicate from the MGS
72 * @param cfg Since the same mgc may be used to follow multiple config logs
73 * (e.g. ost1, ost2, client), the config_llog_instance keeps the state for
74 * this log, and is added to the mgc's list of logs to follow.
76 int lustre_process_log(struct super_block *sb, char *logname,
77 struct config_llog_instance *cfg)
79 struct lustre_cfg *lcfg;
80 struct lustre_cfg_bufs *bufs;
81 struct lustre_sb_info *lsi = s2lsi(sb);
82 struct obd_device *mgc = lsi->lsi_mgc;
93 /* mgc_process_config */
94 lustre_cfg_bufs_reset(bufs, mgc->obd_name);
95 lustre_cfg_bufs_set_string(bufs, 1, logname);
96 lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg));
97 lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb));
98 lcfg = lustre_cfg_new(LCFG_LOG_START, bufs);
99 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
100 lustre_cfg_free(lcfg);
105 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
106 "failed from the MGS (%d). Make sure this "
107 "client and the MGS are running compatible "
108 "versions of Lustre.\n",
109 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);
145 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
146 lustre_cfg_free(lcfg);
149 EXPORT_SYMBOL(lustre_end_log);
151 /**************** obd start *******************/
153 /** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from
154 * lctl (and do for echo cli/srv.
156 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
157 char *s1, char *s2, char *s3, char *s4)
159 struct lustre_cfg_bufs bufs;
160 struct lustre_cfg * lcfg = NULL;
163 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
164 cmd, s1, s2, s3, s4);
166 lustre_cfg_bufs_reset(&bufs, cfgname);
168 lustre_cfg_bufs_set_string(&bufs, 1, s1);
170 lustre_cfg_bufs_set_string(&bufs, 2, s2);
172 lustre_cfg_bufs_set_string(&bufs, 3, s3);
174 lustre_cfg_bufs_set_string(&bufs, 4, s4);
176 lcfg = lustre_cfg_new(cmd, &bufs);
177 lcfg->lcfg_nid = nid;
178 rc = class_process_config(lcfg);
179 lustre_cfg_free(lcfg);
182 EXPORT_SYMBOL(do_lcfg);
184 /** Call class_attach and class_setup. These methods in turn call
185 * obd type-specific methods.
187 int lustre_start_simple(char *obdname, char *type, char *uuid,
188 char *s1, char *s2, char *s3, char *s4)
191 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
193 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
195 CERROR("%s attach error %d\n", obdname, rc);
198 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, s3, s4);
200 CERROR("%s setup error %d\n", obdname, rc);
201 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
206 DEFINE_MUTEX(mgc_start_lock);
208 /** Set up a mgc obd to process startup logs
210 * \param sb [in] super block of the mgc obd
212 * \retval 0 success, otherwise error code
214 int lustre_start_mgc(struct super_block *sb)
216 struct obd_connect_data *data = NULL;
217 struct lustre_sb_info *lsi = s2lsi(sb);
218 struct obd_device *obd;
219 struct obd_export *exp;
220 struct obd_uuid *uuid;
223 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
226 int rc = 0, i = 0, j, len;
229 LASSERT(lsi->lsi_lmd);
231 /* Find the first non-lo MGS nid for our MGC name */
232 if (IS_SERVER(lsi)) {
233 /* mount -o mgsnode=nid */
234 ptr = lsi->lsi_lmd->lmd_mgs;
235 if (lsi->lsi_lmd->lmd_mgs &&
236 (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) {
238 } else if (IS_MGS(lsi)) {
239 lnet_process_id_t id;
240 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
241 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
248 } else { /* client */
249 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
250 ptr = lsi->lsi_lmd->lmd_dev;
251 if (class_parse_nid(ptr, &nid, &ptr) == 0)
255 CERROR("No valid MGS nids found.\n");
259 mutex_lock(&mgc_start_lock);
261 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
262 OBD_ALLOC(mgcname, len);
263 OBD_ALLOC(niduuid, len + 2);
264 if (!mgcname || !niduuid)
265 GOTO(out_free, rc = -ENOMEM);
266 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
268 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
272 GOTO(out_free, rc = -ENOMEM);
274 obd = class_name2obd(mgcname);
275 if (obd && !obd->obd_stopping) {
276 rc = obd_set_info_async(NULL, obd->obd_self_export,
277 strlen(KEY_MGSSEC), KEY_MGSSEC,
278 strlen(mgssec), mgssec, NULL);
282 /* Re-using an existing MGC */
283 cfs_atomic_inc(&obd->u.cli.cl_mgc_refcount);
285 /* IR compatibility check, only for clients */
286 if (lmd_is_client(lsi->lsi_lmd)) {
288 int vallen = sizeof(*data);
289 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
291 rc = obd_get_info(NULL, obd->obd_self_export,
292 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
293 &vallen, data, NULL);
295 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
296 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
297 /* LMD_FLG_NOIR is for test purpose only */
299 "Trying to mount a client with IR setting "
300 "not compatible with current mgc. "
301 "Force to use current mgc setting that is "
303 has_ir ? "enabled" : "disabled");
305 *flags &= ~LMD_FLG_NOIR;
307 *flags |= LMD_FLG_NOIR;
312 /* If we are restarting the MGS, don't try to keep the MGC's
313 old connection, or registration will fail. */
315 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
319 /* Try all connections, but only once (again).
320 We don't want to block another target from starting
321 (using its local copy of the log), but we do want to connect
322 if at all possible. */
324 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
325 rc = obd_set_info_async(NULL, obd->obd_self_export,
326 sizeof(KEY_INIT_RECOV_BACKUP),
327 KEY_INIT_RECOV_BACKUP,
328 sizeof(recov_bk), &recov_bk, NULL);
332 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
334 /* Add the primary nids for the MGS */
336 sprintf(niduuid, "%s_%x", mgcname, i);
337 if (IS_SERVER(lsi)) {
338 ptr = lsi->lsi_lmd->lmd_mgs;
340 /* Use local nids (including LO) */
341 lnet_process_id_t id;
342 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
343 rc = do_lcfg(mgcname, id.nid,
344 LCFG_ADD_UUID, niduuid, 0,0,0);
347 /* Use mgsnode= nids */
348 /* mount -o mgsnode=nid */
349 if (lsi->lsi_lmd->lmd_mgs) {
350 ptr = lsi->lsi_lmd->lmd_mgs;
351 } else if (class_find_param(ptr, PARAM_MGSNODE,
353 CERROR("No MGS nids given.\n");
354 GOTO(out_free, rc = -EINVAL);
356 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
357 rc = do_lcfg(mgcname, nid,
358 LCFG_ADD_UUID, niduuid, 0,0,0);
362 } else { /* client */
363 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
364 ptr = lsi->lsi_lmd->lmd_dev;
365 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
366 rc = do_lcfg(mgcname, nid,
367 LCFG_ADD_UUID, niduuid, 0,0,0);
369 /* Stop at the first failover nid */
375 CERROR("No valid MGS nids found.\n");
376 GOTO(out_free, rc = -EINVAL);
378 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
380 /* Random uuid for MGC allows easier reconnects */
382 ll_generate_random_uuid(uuidc);
383 class_uuid_unparse(uuidc, uuid);
386 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
387 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
393 /* Add any failover MGS nids */
395 while (ptr && ((*ptr == ':' ||
396 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) {
397 /* New failover node */
398 sprintf(niduuid, "%s_%x", mgcname, i);
400 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
402 rc = do_lcfg(mgcname, nid,
403 LCFG_ADD_UUID, niduuid, 0,0,0);
408 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
416 lsi->lsi_lmd->lmd_mgs_failnodes = i;
418 obd = class_name2obd(mgcname);
420 CERROR("Can't find mgcobd %s\n", mgcname);
421 GOTO(out_free, rc = -ENOTCONN);
424 rc = obd_set_info_async(NULL, obd->obd_self_export,
425 strlen(KEY_MGSSEC), KEY_MGSSEC,
426 strlen(mgssec), mgssec, NULL);
430 /* Keep a refcount of servers/clients who started with "mount",
431 so we know when we can get rid of the mgc. */
432 cfs_atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
434 /* Try all connections, but only once. */
436 rc = obd_set_info_async(NULL, obd->obd_self_export,
437 sizeof(KEY_INIT_RECOV_BACKUP),
438 KEY_INIT_RECOV_BACKUP,
439 sizeof(recov_bk), &recov_bk, NULL);
442 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
444 /* We connect to the MGS at setup, and don't disconnect until cleanup */
445 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
446 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV |
447 OBD_CONNECT_LVB_TYPE;
449 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
450 data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB;
452 #warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
455 if (lmd_is_client(lsi->lsi_lmd) &&
456 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
457 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
458 data->ocd_version = LUSTRE_VERSION_CODE;
459 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
461 CERROR("connect failed %d\n", rc);
465 obd->u.cli.cl_mgc_mgsexp = exp;
468 /* Keep the mgc info in the sb. Note that many lsi's can point
472 mutex_unlock(&mgc_start_lock);
477 OBD_FREE(mgcname, len);
479 OBD_FREE(niduuid, len + 2);
483 static int lustre_stop_mgc(struct super_block *sb)
485 struct lustre_sb_info *lsi = s2lsi(sb);
486 struct obd_device *obd;
487 char *niduuid = 0, *ptr = 0;
488 int i, rc = 0, len = 0;
498 mutex_lock(&mgc_start_lock);
499 LASSERT(cfs_atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
500 if (!cfs_atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
501 /* This is not fatal, every client that stops
502 will call in here. */
503 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
504 cfs_atomic_read(&obd->u.cli.cl_mgc_refcount));
505 GOTO(out, rc = -EBUSY);
508 /* The MGC has no recoverable data in any case.
509 * force shotdown set in umount_begin */
510 obd->obd_no_recov = 1;
512 if (obd->u.cli.cl_mgc_mgsexp) {
513 /* An error is not fatal, if we are unable to send the
514 disconnect mgs ping evictor cleans up the export */
515 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
517 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
520 /* Save the obdname for cleaning the nid uuids, which are
522 len = strlen(obd->obd_name) + 6;
523 OBD_ALLOC(niduuid, len);
525 strcpy(niduuid, obd->obd_name);
526 ptr = niduuid + strlen(niduuid);
529 rc = class_manual_cleanup(obd);
533 /* Clean the nid uuids */
535 GOTO(out, rc = -ENOMEM);
537 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
538 sprintf(ptr, "_%x", i);
539 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
542 CERROR("del MDC UUID %s failed: rc = %d\n",
547 OBD_FREE(niduuid, len);
549 /* class_import_put will get rid of the additional connections */
550 mutex_unlock(&mgc_start_lock);
554 /***************** lustre superblock **************/
556 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
558 struct lustre_sb_info *lsi;
564 OBD_ALLOC_PTR(lsi->lsi_lmd);
570 lsi->lsi_lmd->lmd_exclude_count = 0;
571 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
572 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
573 s2lsi_nocast(sb) = lsi;
574 /* we take 1 extra ref for our setup */
575 cfs_atomic_set(&lsi->lsi_mounts, 1);
577 /* Default umount style */
578 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
583 static int lustre_free_lsi(struct super_block *sb)
585 struct lustre_sb_info *lsi = s2lsi(sb);
588 LASSERT(lsi != NULL);
589 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
591 /* someone didn't call server_put_mount. */
592 LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0);
594 if (lsi->lsi_lmd != NULL) {
595 if (lsi->lsi_lmd->lmd_dev != NULL)
596 OBD_FREE(lsi->lsi_lmd->lmd_dev,
597 strlen(lsi->lsi_lmd->lmd_dev) + 1);
598 if (lsi->lsi_lmd->lmd_profile != NULL)
599 OBD_FREE(lsi->lsi_lmd->lmd_profile,
600 strlen(lsi->lsi_lmd->lmd_profile) + 1);
601 if (lsi->lsi_lmd->lmd_mgssec != NULL)
602 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
603 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
604 if (lsi->lsi_lmd->lmd_opts != NULL)
605 OBD_FREE(lsi->lsi_lmd->lmd_opts,
606 strlen(lsi->lsi_lmd->lmd_opts) + 1);
607 if (lsi->lsi_lmd->lmd_exclude_count)
608 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
609 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
610 lsi->lsi_lmd->lmd_exclude_count);
611 if (lsi->lsi_lmd->lmd_mgs != NULL)
612 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
613 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
614 if (lsi->lsi_lmd->lmd_osd_type != NULL)
615 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
616 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
617 if (lsi->lsi_lmd->lmd_params != NULL)
618 OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
620 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
623 LASSERT(lsi->lsi_llsbi == NULL);
624 OBD_FREE(lsi, sizeof(*lsi));
625 s2lsi_nocast(sb) = NULL;
630 /* The lsi has one reference for every server that is using the disk -
631 e.g. MDT, MGS, and potentially MGC */
632 int lustre_put_lsi(struct super_block *sb)
634 struct lustre_sb_info *lsi = s2lsi(sb);
637 LASSERT(lsi != NULL);
639 CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts));
640 if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) {
641 if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
642 lu_device_put(&lsi->lsi_dt_dev->dd_lu_dev);
643 lsi->lsi_osd_exp->exp_obd->obd_lvfs_ctxt.dt = NULL;
644 lsi->lsi_dt_dev = NULL;
645 obd_disconnect(lsi->lsi_osd_exp);
646 /* wait till OSD is gone */
647 obd_zombie_barrier();
656 * <FSNAME><SEPERATOR><TYPE><INDEX>
657 * FSNAME is between 1 and 8 characters (inclusive).
658 * Excluded characters are '/' and ':'
659 * SEPERATOR is either ':' or '-'
660 * TYPE: "OST", "MDT", etc.
661 * INDEX: Hex representation of the index
664 /** Get the fsname ("lustre") from the server name ("lustre-OST003F").
665 * @param [in] svname server name including type and index
666 * @param [out] fsname Buffer to copy filesystem name prefix into.
667 * Must have at least 'strlen(fsname) + 1' chars.
668 * @param [out] endptr if endptr isn't NULL it is set to end of fsname
671 int server_name2fsname(const char *svname, char *fsname, const char **endptr)
675 dash = svname + strnlen(svname, 8); /* max fsname length is 8 */
676 for (; dash > svname && *dash != '-' && *dash != ':'; dash--)
681 if (fsname != NULL) {
682 strncpy(fsname, svname, dash - svname);
683 fsname[dash - svname] = '\0';
691 EXPORT_SYMBOL(server_name2fsname);
694 * Get service name (svname) from string
696 * if endptr isn't NULL it is set to end of fsname *
698 int server_name2svname(const char *label, char *svname, const char **endptr,
704 /* We use server_name2fsname() just for parsing */
705 rc = server_name2fsname(label, NULL, &dash);
712 if (strlcpy(svname, dash + 1, svsize) >= svsize)
717 EXPORT_SYMBOL(server_name2svname);
720 * check server name is OST.
722 int server_name_is_ost(const char *svname)
727 /* We use server_name2fsname() just for parsing */
728 rc = server_name2fsname(svname, NULL, &dash);
734 if (strncmp(dash, "OST", 3) == 0)
738 EXPORT_SYMBOL(server_name_is_ost);
740 /* Get the index from the obd name.
743 if endptr isn't NULL it is set to end of name */
744 int server_name2index(const char *svname, __u32 *idx, const char **endptr)
750 /* We use server_name2fsname() just for parsing */
751 rc = server_name2fsname(svname, NULL, &dash);
757 if (strncmp(dash, "MDT", 3) == 0)
758 rc = LDD_F_SV_TYPE_MDT;
759 else if (strncmp(dash, "OST", 3) == 0)
760 rc = LDD_F_SV_TYPE_OST;
766 if (strncmp(dash, "all", 3) == 0) {
769 return rc | LDD_F_SV_ALL;
772 index = simple_strtoul(dash, (char **)endptr, 16);
776 /* Account for -mdc after index that is possible when specifying mdt */
777 if (endptr != NULL && strncmp(LUSTRE_MDC_NAME, *endptr + 1,
778 sizeof(LUSTRE_MDC_NAME)-1) == 0)
779 *endptr += sizeof(LUSTRE_MDC_NAME);
783 EXPORT_SYMBOL(server_name2index);
785 /*************** mount common betweeen server and client ***************/
788 int lustre_common_put_super(struct super_block *sb)
793 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
795 /* Drop a ref to the MGC */
796 rc = lustre_stop_mgc(sb);
797 if (rc && (rc != -ENOENT)) {
799 CERROR("Can't stop MGC: %d\n", rc);
802 /* BUSY just means that there's some other obd that
803 needs the mgc. Let him clean it up. */
804 CDEBUG(D_MOUNT, "MGC still in use\n");
806 /* Drop a ref to the mounted disk */
811 EXPORT_SYMBOL(lustre_common_put_super);
813 static void lmd_print(struct lustre_mount_data *lmd)
817 PRINT_CMD(D_MOUNT, " mount data:\n");
818 if (lmd_is_client(lmd))
819 PRINT_CMD(D_MOUNT, "profile: %s\n", lmd->lmd_profile);
820 PRINT_CMD(D_MOUNT, "device: %s\n", lmd->lmd_dev);
821 PRINT_CMD(D_MOUNT, "flags: %x\n", lmd->lmd_flags);
824 PRINT_CMD(D_MOUNT, "options: %s\n", lmd->lmd_opts);
826 if (lmd->lmd_recovery_time_soft)
827 PRINT_CMD(D_MOUNT, "recovery time soft: %d\n",
828 lmd->lmd_recovery_time_soft);
830 if (lmd->lmd_recovery_time_hard)
831 PRINT_CMD(D_MOUNT, "recovery time hard: %d\n",
832 lmd->lmd_recovery_time_hard);
834 for (i = 0; i < lmd->lmd_exclude_count; i++) {
835 PRINT_CMD(D_MOUNT, "exclude %d: OST%04x\n", i,
836 lmd->lmd_exclude[i]);
840 /* Is this server on the exclusion list */
841 int lustre_check_exclusion(struct super_block *sb, char *svname)
843 struct lustre_sb_info *lsi = s2lsi(sb);
844 struct lustre_mount_data *lmd = lsi->lsi_lmd;
849 rc = server_name2index(svname, &index, NULL);
850 if (rc != LDD_F_SV_TYPE_OST)
851 /* Only exclude OSTs */
854 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
855 index, lmd->lmd_exclude_count, lmd->lmd_dev);
857 for(i = 0; i < lmd->lmd_exclude_count; i++) {
858 if (index == lmd->lmd_exclude[i]) {
859 CWARN("Excluding %s (on exclusion list)\n", svname);
866 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
867 static int lmd_make_exclusion(struct lustre_mount_data *lmd, const char *ptr)
869 const char *s1 = ptr, *s2;
870 __u32 index, *exclude_list;
874 /* The shortest an ost name can be is 8 chars: -OST0000.
875 We don't actually know the fsname at this time, so in fact
876 a user could specify any fsname. */
877 devmax = strlen(ptr) / 8 + 1;
879 /* temp storage until we figure out how many we have */
880 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
884 /* we enter this fn pointing at the '=' */
885 while (*s1 && *s1 != ' ' && *s1 != ',') {
887 rc = server_name2index(s1, &index, &s2);
889 CERROR("Can't parse server name '%s': rc = %d\n",
893 if (rc == LDD_F_SV_TYPE_OST)
894 exclude_list[lmd->lmd_exclude_count++] = index;
896 CDEBUG(D_MOUNT, "ignoring exclude %.*s: type = %#x\n",
897 (uint)(s2-s1), s1, rc);
899 /* now we are pointing at ':' (next exclude)
900 or ',' (end of excludes) */
901 if (lmd->lmd_exclude_count >= devmax)
904 if (rc >= 0) /* non-err */
907 if (lmd->lmd_exclude_count) {
908 /* permanent, freed in lustre_free_lsi */
909 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
910 lmd->lmd_exclude_count);
911 if (lmd->lmd_exclude) {
912 memcpy(lmd->lmd_exclude, exclude_list,
913 sizeof(index) * lmd->lmd_exclude_count);
916 lmd->lmd_exclude_count = 0;
919 OBD_FREE(exclude_list, sizeof(index) * devmax);
923 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
928 if (lmd->lmd_mgssec != NULL) {
929 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
930 lmd->lmd_mgssec = NULL;
933 tail = strchr(ptr, ',');
935 length = strlen(ptr);
939 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
940 if (lmd->lmd_mgssec == NULL)
943 memcpy(lmd->lmd_mgssec, ptr, length);
944 lmd->lmd_mgssec[length] = '\0';
948 static int lmd_parse_string(char **handle, char *ptr)
953 if ((handle == NULL) || (ptr == NULL))
956 if (*handle != NULL) {
957 OBD_FREE(*handle, strlen(*handle) + 1);
961 tail = strchr(ptr, ',');
963 length = strlen(ptr);
967 OBD_ALLOC(*handle, length + 1);
971 memcpy(*handle, ptr, length);
972 (*handle)[length] = '\0';
977 /* Collect multiple values for mgsnid specifiers */
978 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
986 /* Find end of nidlist */
987 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
988 length = tail - *ptr;
990 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
994 if (lmd->lmd_mgs != NULL)
995 oldlen = strlen(lmd->lmd_mgs) + 1;
997 OBD_ALLOC(mgsnid, oldlen + length + 1);
1001 if (lmd->lmd_mgs != NULL) {
1002 /* Multiple mgsnid= are taken to mean failover locations */
1003 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
1004 mgsnid[oldlen - 1] = ':';
1005 OBD_FREE(lmd->lmd_mgs, oldlen);
1007 memcpy(mgsnid + oldlen, *ptr, length);
1008 mgsnid[oldlen + length] = '\0';
1009 lmd->lmd_mgs = mgsnid;
1015 /** Parse mount line options
1016 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
1017 * dev is passed as device=uml1:/lustre by mount.lustre
1019 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1021 char *s1, *s2, *devname = NULL;
1022 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1028 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1029 "/sbin/mount.lustre is installed.\n");
1033 /* Options should be a string - try to detect old lmd data */
1034 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1035 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1036 "/sbin/mount.lustre. Please install "
1037 "version %s\n", LUSTRE_VERSION_STRING);
1040 lmd->lmd_magic = LMD_MAGIC;
1042 OBD_ALLOC(lmd->lmd_params, 4096);
1043 if (lmd->lmd_params == NULL)
1045 lmd->lmd_params[0] = '\0';
1047 /* Set default flags here */
1052 int time_min = OBD_RECOVERY_TIME_MIN;
1054 /* Skip whitespace and extra commas */
1055 while (*s1 == ' ' || *s1 == ',')
1058 /* Client options are parsed in ll_options: eg. flock,
1061 /* Parse non-ldiskfs options here. Rather than modifying
1062 ldiskfs, we just zero these out here */
1063 if (strncmp(s1, "abort_recov", 11) == 0) {
1064 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1066 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
1067 lmd->lmd_recovery_time_soft = max_t(int,
1068 simple_strtoul(s1 + 19, NULL, 10), time_min);
1070 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
1071 lmd->lmd_recovery_time_hard = max_t(int,
1072 simple_strtoul(s1 + 19, NULL, 10), time_min);
1074 } else if (strncmp(s1, "noir", 4) == 0) {
1075 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
1077 } else if (strncmp(s1, "nosvc", 5) == 0) {
1078 lmd->lmd_flags |= LMD_FLG_NOSVC;
1080 } else if (strncmp(s1, "nomgs", 5) == 0) {
1081 lmd->lmd_flags |= LMD_FLG_NOMGS;
1083 } else if (strncmp(s1, "noscrub", 7) == 0) {
1084 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
1086 } else if (strncmp(s1, PARAM_MGSNODE,
1087 sizeof(PARAM_MGSNODE) - 1) == 0) {
1088 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
1089 /* Assume the next mount opt is the first
1090 invalid nid we get to. */
1091 rc = lmd_parse_mgs(lmd, &s2);
1095 } else if (strncmp(s1, "writeconf", 9) == 0) {
1096 lmd->lmd_flags |= LMD_FLG_WRITECONF;
1098 } else if (strncmp(s1, "update", 6) == 0) {
1099 lmd->lmd_flags |= LMD_FLG_UPDATE;
1101 } else if (strncmp(s1, "virgin", 6) == 0) {
1102 lmd->lmd_flags |= LMD_FLG_VIRGIN;
1104 } else if (strncmp(s1, "noprimnode", 10) == 0) {
1105 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
1107 } else if (strncmp(s1, "mgssec=", 7) == 0) {
1108 rc = lmd_parse_mgssec(lmd, s1 + 7);
1112 /* ost exclusion list */
1113 } else if (strncmp(s1, "exclude=", 8) == 0) {
1114 rc = lmd_make_exclusion(lmd, s1 + 7);
1118 } else if (strncmp(s1, "mgs", 3) == 0) {
1120 lmd->lmd_flags |= LMD_FLG_MGS;
1122 } else if (strncmp(s1, "svname=", 7) == 0) {
1123 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
1127 } else if (strncmp(s1, "param=", 6) == 0) {
1129 char *tail = strchr(s1 + 6, ',');
1131 length = strlen(s1);
1135 strncat(lmd->lmd_params, s1 + 6, length);
1136 strcat(lmd->lmd_params, " ");
1138 } else if (strncmp(s1, "osd=", 4) == 0) {
1139 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
1144 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1145 end of the options. */
1146 else if (strncmp(s1, "device=", 7) == 0) {
1148 /* terminate options right before device. device
1149 must be the last one. */
1155 s2 = strchr(s1, ',');
1163 memmove(s1, s2, strlen(s2) + 1);
1169 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1170 "(need mount option 'device=...')\n");
1174 s1 = strstr(devname, ":/");
1177 lmd->lmd_flags |= LMD_FLG_CLIENT;
1178 /* Remove leading /s from fsname */
1179 while (*++s1 == '/') ;
1180 /* Freed in lustre_free_lsi */
1181 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1182 if (!lmd->lmd_profile)
1184 sprintf(lmd->lmd_profile, "%s-client", s1);
1187 /* Freed in lustre_free_lsi */
1188 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1191 strcpy(lmd->lmd_dev, devname);
1193 /* Save mount options */
1194 s1 = options + strlen(options) - 1;
1195 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1197 if (*options != 0) {
1198 /* Freed in lustre_free_lsi */
1199 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1202 strcpy(lmd->lmd_opts, options);
1206 lmd->lmd_magic = LMD_MAGIC;
1211 CERROR("Bad mount options %s\n", options);
1215 struct lustre_mount_data2 {
1217 struct vfsmount *lmd2_mnt;
1220 /** This is the entry point for the mount call into Lustre.
1221 * This is called when a server or client is mounted,
1222 * and this is where we start setting things up.
1223 * @param data Mount options (e.g. -o flock,abort_recov)
1225 int lustre_fill_super(struct super_block *sb, void *data, int silent)
1227 struct lustre_mount_data *lmd;
1228 struct lustre_mount_data2 *lmd2 = data;
1229 struct lustre_sb_info *lsi;
1233 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1235 lsi = lustre_init_lsi(sb);
1241 * Disable lockdep during mount, because mount locking patterns are
1247 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
1249 obd_zombie_barrier();
1251 /* Figure out the lmd from the mount options */
1252 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
1254 GOTO(out, rc = -EINVAL);
1257 if (lmd_is_client(lmd)) {
1258 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1259 if (!client_fill_super) {
1260 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1261 "client mount! Is the 'lustre' "
1262 "module loaded?\n");
1266 rc = lustre_start_mgc(sb);
1271 /* Connect and start */
1272 /* (should always be ll_fill_super) */
1273 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
1274 /* c_f_s will call lustre_common_put_super on failure */
1277 #ifdef HAVE_SERVER_SUPPORT
1278 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1279 rc = server_fill_super(sb);
1280 /* s_f_s calls lustre_start_mgc after the mount because we need
1281 the MGS nids which are stored on disk. Plus, we may
1282 need to start the MGS first. */
1283 /* s_f_s will call server_put_super on failure */
1285 CERROR("This is client-side-only module, "
1286 "cannot handle server mount.\n");
1291 /* If error happens in fill_super() call, @lsi will be killed there.
1292 * This is why we do not put it here. */
1296 CERROR("Unable to mount %s (%d)\n",
1297 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1299 CDEBUG(D_SUPER, "Mount %s complete\n",
1307 /* We can't call ll_fill_super by name because it lives in a module that
1308 must be loaded after this one. */
1309 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
1310 struct vfsmount *mnt))
1312 client_fill_super = cfs;
1314 EXPORT_SYMBOL(lustre_register_client_fill_super);
1316 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
1318 kill_super_cb = cfs;
1320 EXPORT_SYMBOL(lustre_register_kill_super_cb);
1322 /***************** FS registration ******************/
1323 #ifdef HAVE_FSTYPE_MOUNT
1324 struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
1325 const char *devname, void *data)
1327 struct lustre_mount_data2 lmd2 = { data, NULL };
1329 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
1332 int lustre_get_sb(struct file_system_type *fs_type, int flags,
1333 const char *devname, void * data, struct vfsmount *mnt)
1335 struct lustre_mount_data2 lmd2 = { data, mnt };
1337 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
1341 void lustre_kill_super(struct super_block *sb)
1343 struct lustre_sb_info *lsi = s2lsi(sb);
1345 if (kill_super_cb && lsi && !IS_SERVER(lsi))
1346 (*kill_super_cb)(sb);
1348 kill_anon_super(sb);
1351 /** Register the "lustre" fs type
1353 struct file_system_type lustre_fs_type = {
1354 .owner = THIS_MODULE,
1356 #ifdef HAVE_FSTYPE_MOUNT
1357 .mount = lustre_mount,
1359 .get_sb = lustre_get_sb,
1361 .kill_sb = lustre_kill_super,
1362 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
1363 FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE,
1366 int lustre_register_fs(void)
1368 return register_filesystem(&lustre_fs_type);
1371 int lustre_unregister_fs(void)
1373 return unregister_filesystem(&lustre_fs_type);