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;
339 CDEBUG(D_MOUNT, "mgs nids %s.\n", ptr);
341 /* Use local nids (including LO) */
342 lnet_process_id_t id;
343 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
344 rc = do_lcfg(mgcname, id.nid, LCFG_ADD_UUID,
348 /* Use mgsnode= nids */
349 /* mount -o mgsnode=nid */
350 if (lsi->lsi_lmd->lmd_mgs) {
351 ptr = lsi->lsi_lmd->lmd_mgs;
352 } else if (class_find_param(ptr, PARAM_MGSNODE,
354 CERROR("No MGS nids given.\n");
355 GOTO(out_free, rc = -EINVAL);
359 * Here we only take the first mgsnid as its primary
360 * serving mgs node, the rest mgsnid will be taken as
361 * failover mgs node, otherwise they would be takens
362 * as multiple nids of a single mgs node.
364 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
365 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
373 } else { /* client */
374 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
375 ptr = lsi->lsi_lmd->lmd_dev;
376 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
377 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
381 /* Stop at the first failover nid */
387 CERROR("No valid MGS nids found.\n");
388 GOTO(out_free, rc = -EINVAL);
390 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
392 /* Random uuid for MGC allows easier reconnects */
394 ll_generate_random_uuid(uuidc);
395 class_uuid_unparse(uuidc, uuid);
398 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
399 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
405 /* Add any failover MGS nids */
407 while (ptr && ((*ptr == ':' ||
408 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) {
409 /* New failover node */
410 sprintf(niduuid, "%s_%x", mgcname, i);
412 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
413 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
421 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
430 lsi->lsi_lmd->lmd_mgs_failnodes = i;
432 obd = class_name2obd(mgcname);
434 CERROR("Can't find mgcobd %s\n", mgcname);
435 GOTO(out_free, rc = -ENOTCONN);
438 rc = obd_set_info_async(NULL, obd->obd_self_export,
439 strlen(KEY_MGSSEC), KEY_MGSSEC,
440 strlen(mgssec), mgssec, NULL);
444 /* Keep a refcount of servers/clients who started with "mount",
445 so we know when we can get rid of the mgc. */
446 cfs_atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
448 /* Try all connections, but only once. */
450 rc = obd_set_info_async(NULL, obd->obd_self_export,
451 sizeof(KEY_INIT_RECOV_BACKUP),
452 KEY_INIT_RECOV_BACKUP,
453 sizeof(recov_bk), &recov_bk, NULL);
456 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
458 /* We connect to the MGS at setup, and don't disconnect until cleanup */
459 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
460 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV |
461 OBD_CONNECT_LVB_TYPE;
463 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
464 data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB;
466 #warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
469 if (lmd_is_client(lsi->lsi_lmd) &&
470 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
471 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
472 data->ocd_version = LUSTRE_VERSION_CODE;
473 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
475 CERROR("connect failed %d\n", rc);
479 obd->u.cli.cl_mgc_mgsexp = exp;
482 /* Keep the mgc info in the sb. Note that many lsi's can point
486 mutex_unlock(&mgc_start_lock);
491 OBD_FREE(mgcname, len);
493 OBD_FREE(niduuid, len + 2);
497 static int lustre_stop_mgc(struct super_block *sb)
499 struct lustre_sb_info *lsi = s2lsi(sb);
500 struct obd_device *obd;
501 char *niduuid = 0, *ptr = 0;
502 int i, rc = 0, len = 0;
512 mutex_lock(&mgc_start_lock);
513 LASSERT(cfs_atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
514 if (!cfs_atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
515 /* This is not fatal, every client that stops
516 will call in here. */
517 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
518 cfs_atomic_read(&obd->u.cli.cl_mgc_refcount));
519 GOTO(out, rc = -EBUSY);
522 /* The MGC has no recoverable data in any case.
523 * force shotdown set in umount_begin */
524 obd->obd_no_recov = 1;
526 if (obd->u.cli.cl_mgc_mgsexp) {
527 /* An error is not fatal, if we are unable to send the
528 disconnect mgs ping evictor cleans up the export */
529 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
531 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
534 /* Save the obdname for cleaning the nid uuids, which are
536 len = strlen(obd->obd_name) + 6;
537 OBD_ALLOC(niduuid, len);
539 strcpy(niduuid, obd->obd_name);
540 ptr = niduuid + strlen(niduuid);
543 rc = class_manual_cleanup(obd);
547 /* Clean the nid uuids */
549 GOTO(out, rc = -ENOMEM);
551 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
552 sprintf(ptr, "_%x", i);
553 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
556 CERROR("del MDC UUID %s failed: rc = %d\n",
561 OBD_FREE(niduuid, len);
563 /* class_import_put will get rid of the additional connections */
564 mutex_unlock(&mgc_start_lock);
568 /***************** lustre superblock **************/
570 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
572 struct lustre_sb_info *lsi;
578 OBD_ALLOC_PTR(lsi->lsi_lmd);
584 lsi->lsi_lmd->lmd_exclude_count = 0;
585 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
586 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
587 s2lsi_nocast(sb) = lsi;
588 /* we take 1 extra ref for our setup */
589 cfs_atomic_set(&lsi->lsi_mounts, 1);
591 /* Default umount style */
592 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
597 static int lustre_free_lsi(struct super_block *sb)
599 struct lustre_sb_info *lsi = s2lsi(sb);
602 LASSERT(lsi != NULL);
603 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
605 /* someone didn't call server_put_mount. */
606 LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0);
608 if (lsi->lsi_lmd != NULL) {
609 if (lsi->lsi_lmd->lmd_dev != NULL)
610 OBD_FREE(lsi->lsi_lmd->lmd_dev,
611 strlen(lsi->lsi_lmd->lmd_dev) + 1);
612 if (lsi->lsi_lmd->lmd_profile != NULL)
613 OBD_FREE(lsi->lsi_lmd->lmd_profile,
614 strlen(lsi->lsi_lmd->lmd_profile) + 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(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
637 LASSERT(lsi->lsi_llsbi == NULL);
638 OBD_FREE(lsi, sizeof(*lsi));
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, cfs_atomic_read(&lsi->lsi_mounts));
654 if (cfs_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><SEPERATOR><TYPE><INDEX>
671 * FSNAME is between 1 and 8 characters (inclusive).
672 * Excluded characters are '/' and ':'
673 * SEPERATOR 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);
754 /* Get the index from the obd name.
757 if endptr isn't NULL it is set to end of name */
758 int server_name2index(const char *svname, __u32 *idx, const char **endptr)
764 /* We use server_name2fsname() just for parsing */
765 rc = server_name2fsname(svname, NULL, &dash);
771 if (strncmp(dash, "MDT", 3) == 0)
772 rc = LDD_F_SV_TYPE_MDT;
773 else if (strncmp(dash, "OST", 3) == 0)
774 rc = LDD_F_SV_TYPE_OST;
780 if (strncmp(dash, "all", 3) == 0) {
783 return rc | LDD_F_SV_ALL;
786 index = simple_strtoul(dash, (char **)endptr, 16);
790 /* Account for -mdc after index that is possible when specifying mdt */
791 if (endptr != NULL && strncmp(LUSTRE_MDC_NAME, *endptr + 1,
792 sizeof(LUSTRE_MDC_NAME)-1) == 0)
793 *endptr += sizeof(LUSTRE_MDC_NAME);
797 EXPORT_SYMBOL(server_name2index);
799 /*************** mount common betweeen server and client ***************/
802 int lustre_common_put_super(struct super_block *sb)
807 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
809 /* Drop a ref to the MGC */
810 rc = lustre_stop_mgc(sb);
811 if (rc && (rc != -ENOENT)) {
813 CERROR("Can't stop MGC: %d\n", rc);
816 /* BUSY just means that there's some other obd that
817 needs the mgc. Let him clean it up. */
818 CDEBUG(D_MOUNT, "MGC still in use\n");
820 /* Drop a ref to the mounted disk */
825 EXPORT_SYMBOL(lustre_common_put_super);
827 static void lmd_print(struct lustre_mount_data *lmd)
831 PRINT_CMD(D_MOUNT, " mount data:\n");
832 if (lmd_is_client(lmd))
833 PRINT_CMD(D_MOUNT, "profile: %s\n", lmd->lmd_profile);
834 PRINT_CMD(D_MOUNT, "device: %s\n", lmd->lmd_dev);
835 PRINT_CMD(D_MOUNT, "flags: %x\n", lmd->lmd_flags);
838 PRINT_CMD(D_MOUNT, "options: %s\n", lmd->lmd_opts);
840 if (lmd->lmd_recovery_time_soft)
841 PRINT_CMD(D_MOUNT, "recovery time soft: %d\n",
842 lmd->lmd_recovery_time_soft);
844 if (lmd->lmd_recovery_time_hard)
845 PRINT_CMD(D_MOUNT, "recovery time hard: %d\n",
846 lmd->lmd_recovery_time_hard);
848 for (i = 0; i < lmd->lmd_exclude_count; i++) {
849 PRINT_CMD(D_MOUNT, "exclude %d: OST%04x\n", i,
850 lmd->lmd_exclude[i]);
854 /* Is this server on the exclusion list */
855 int lustre_check_exclusion(struct super_block *sb, char *svname)
857 struct lustre_sb_info *lsi = s2lsi(sb);
858 struct lustre_mount_data *lmd = lsi->lsi_lmd;
863 rc = server_name2index(svname, &index, NULL);
864 if (rc != LDD_F_SV_TYPE_OST)
865 /* Only exclude OSTs */
868 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
869 index, lmd->lmd_exclude_count, lmd->lmd_dev);
871 for(i = 0; i < lmd->lmd_exclude_count; i++) {
872 if (index == lmd->lmd_exclude[i]) {
873 CWARN("Excluding %s (on exclusion list)\n", svname);
880 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
881 static int lmd_make_exclusion(struct lustre_mount_data *lmd, const char *ptr)
883 const char *s1 = ptr, *s2;
884 __u32 index, *exclude_list;
888 /* The shortest an ost name can be is 8 chars: -OST0000.
889 We don't actually know the fsname at this time, so in fact
890 a user could specify any fsname. */
891 devmax = strlen(ptr) / 8 + 1;
893 /* temp storage until we figure out how many we have */
894 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
898 /* we enter this fn pointing at the '=' */
899 while (*s1 && *s1 != ' ' && *s1 != ',') {
901 rc = server_name2index(s1, &index, &s2);
903 CERROR("Can't parse server name '%s': rc = %d\n",
907 if (rc == LDD_F_SV_TYPE_OST)
908 exclude_list[lmd->lmd_exclude_count++] = index;
910 CDEBUG(D_MOUNT, "ignoring exclude %.*s: type = %#x\n",
911 (uint)(s2-s1), s1, rc);
913 /* now we are pointing at ':' (next exclude)
914 or ',' (end of excludes) */
915 if (lmd->lmd_exclude_count >= devmax)
918 if (rc >= 0) /* non-err */
921 if (lmd->lmd_exclude_count) {
922 /* permanent, freed in lustre_free_lsi */
923 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
924 lmd->lmd_exclude_count);
925 if (lmd->lmd_exclude) {
926 memcpy(lmd->lmd_exclude, exclude_list,
927 sizeof(index) * lmd->lmd_exclude_count);
930 lmd->lmd_exclude_count = 0;
933 OBD_FREE(exclude_list, sizeof(index) * devmax);
937 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
942 if (lmd->lmd_mgssec != NULL) {
943 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
944 lmd->lmd_mgssec = NULL;
947 tail = strchr(ptr, ',');
949 length = strlen(ptr);
953 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
954 if (lmd->lmd_mgssec == NULL)
957 memcpy(lmd->lmd_mgssec, ptr, length);
958 lmd->lmd_mgssec[length] = '\0';
962 static int lmd_parse_string(char **handle, char *ptr)
967 if ((handle == NULL) || (ptr == NULL))
970 if (*handle != NULL) {
971 OBD_FREE(*handle, strlen(*handle) + 1);
975 tail = strchr(ptr, ',');
977 length = strlen(ptr);
981 OBD_ALLOC(*handle, length + 1);
985 memcpy(*handle, ptr, length);
986 (*handle)[length] = '\0';
991 /* Collect multiple values for mgsnid specifiers */
992 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
1000 /* Find end of nidlist */
1001 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
1002 length = tail - *ptr;
1004 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
1008 if (lmd->lmd_mgs != NULL)
1009 oldlen = strlen(lmd->lmd_mgs) + 1;
1011 OBD_ALLOC(mgsnid, oldlen + length + 1);
1015 if (lmd->lmd_mgs != NULL) {
1016 /* Multiple mgsnid= are taken to mean failover locations */
1017 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
1018 mgsnid[oldlen - 1] = ':';
1019 OBD_FREE(lmd->lmd_mgs, oldlen);
1021 memcpy(mgsnid + oldlen, *ptr, length);
1022 mgsnid[oldlen + length] = '\0';
1023 lmd->lmd_mgs = mgsnid;
1029 /** Parse mount line options
1030 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
1031 * dev is passed as device=uml1:/lustre by mount.lustre
1033 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1035 char *s1, *s2, *devname = NULL;
1036 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1042 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1043 "/sbin/mount.lustre is installed.\n");
1047 /* Options should be a string - try to detect old lmd data */
1048 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1049 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1050 "/sbin/mount.lustre. Please install "
1051 "version %s\n", LUSTRE_VERSION_STRING);
1054 lmd->lmd_magic = LMD_MAGIC;
1056 OBD_ALLOC(lmd->lmd_params, 4096);
1057 if (lmd->lmd_params == NULL)
1059 lmd->lmd_params[0] = '\0';
1061 /* Set default flags here */
1066 int time_min = OBD_RECOVERY_TIME_MIN;
1068 /* Skip whitespace and extra commas */
1069 while (*s1 == ' ' || *s1 == ',')
1072 /* Client options are parsed in ll_options: eg. flock,
1075 /* Parse non-ldiskfs options here. Rather than modifying
1076 ldiskfs, we just zero these out here */
1077 if (strncmp(s1, "abort_recov", 11) == 0) {
1078 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1080 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
1081 lmd->lmd_recovery_time_soft = max_t(int,
1082 simple_strtoul(s1 + 19, NULL, 10), time_min);
1084 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
1085 lmd->lmd_recovery_time_hard = max_t(int,
1086 simple_strtoul(s1 + 19, NULL, 10), time_min);
1088 } else if (strncmp(s1, "noir", 4) == 0) {
1089 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
1091 } else if (strncmp(s1, "nosvc", 5) == 0) {
1092 lmd->lmd_flags |= LMD_FLG_NOSVC;
1094 } else if (strncmp(s1, "nomgs", 5) == 0) {
1095 lmd->lmd_flags |= LMD_FLG_NOMGS;
1097 } else if (strncmp(s1, "noscrub", 7) == 0) {
1098 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
1100 } else if (strncmp(s1, PARAM_MGSNODE,
1101 sizeof(PARAM_MGSNODE) - 1) == 0) {
1102 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
1103 /* Assume the next mount opt is the first
1104 invalid nid we get to. */
1105 rc = lmd_parse_mgs(lmd, &s2);
1109 } else if (strncmp(s1, "writeconf", 9) == 0) {
1110 lmd->lmd_flags |= LMD_FLG_WRITECONF;
1112 } else if (strncmp(s1, "update", 6) == 0) {
1113 lmd->lmd_flags |= LMD_FLG_UPDATE;
1115 } else if (strncmp(s1, "virgin", 6) == 0) {
1116 lmd->lmd_flags |= LMD_FLG_VIRGIN;
1118 } else if (strncmp(s1, "noprimnode", 10) == 0) {
1119 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
1121 } else if (strncmp(s1, "mgssec=", 7) == 0) {
1122 rc = lmd_parse_mgssec(lmd, s1 + 7);
1126 /* ost exclusion list */
1127 } else if (strncmp(s1, "exclude=", 8) == 0) {
1128 rc = lmd_make_exclusion(lmd, s1 + 7);
1132 } else if (strncmp(s1, "mgs", 3) == 0) {
1134 lmd->lmd_flags |= LMD_FLG_MGS;
1136 } else if (strncmp(s1, "svname=", 7) == 0) {
1137 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
1141 } else if (strncmp(s1, "param=", 6) == 0) {
1143 char *tail = strchr(s1 + 6, ',');
1145 length = strlen(s1);
1149 strncat(lmd->lmd_params, s1 + 6, length);
1150 strcat(lmd->lmd_params, " ");
1152 } else if (strncmp(s1, "osd=", 4) == 0) {
1153 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
1158 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1159 end of the options. */
1160 else if (strncmp(s1, "device=", 7) == 0) {
1162 /* terminate options right before device. device
1163 must be the last one. */
1169 s2 = strchr(s1, ',');
1177 memmove(s1, s2, strlen(s2) + 1);
1183 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1184 "(need mount option 'device=...')\n");
1188 s1 = strstr(devname, ":/");
1191 lmd->lmd_flags |= LMD_FLG_CLIENT;
1192 /* Remove leading /s from fsname */
1193 while (*++s1 == '/') ;
1194 /* Freed in lustre_free_lsi */
1195 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1196 if (!lmd->lmd_profile)
1198 sprintf(lmd->lmd_profile, "%s-client", s1);
1201 /* Freed in lustre_free_lsi */
1202 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1205 strcpy(lmd->lmd_dev, devname);
1207 /* Save mount options */
1208 s1 = options + strlen(options) - 1;
1209 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1211 if (*options != 0) {
1212 /* Freed in lustre_free_lsi */
1213 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1216 strcpy(lmd->lmd_opts, options);
1220 lmd->lmd_magic = LMD_MAGIC;
1225 CERROR("Bad mount options %s\n", options);
1229 struct lustre_mount_data2 {
1231 struct vfsmount *lmd2_mnt;
1234 /** This is the entry point for the mount call into Lustre.
1235 * This is called when a server or client is mounted,
1236 * and this is where we start setting things up.
1237 * @param data Mount options (e.g. -o flock,abort_recov)
1239 int lustre_fill_super(struct super_block *sb, void *data, int silent)
1241 struct lustre_mount_data *lmd;
1242 struct lustre_mount_data2 *lmd2 = data;
1243 struct lustre_sb_info *lsi;
1247 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1249 lsi = lustre_init_lsi(sb);
1255 * Disable lockdep during mount, because mount locking patterns are
1261 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
1263 obd_zombie_barrier();
1265 /* Figure out the lmd from the mount options */
1266 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
1268 GOTO(out, rc = -EINVAL);
1271 if (lmd_is_client(lmd)) {
1272 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1273 if (!client_fill_super) {
1274 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1275 "client mount! Is the 'lustre' "
1276 "module loaded?\n");
1280 rc = lustre_start_mgc(sb);
1285 /* Connect and start */
1286 /* (should always be ll_fill_super) */
1287 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
1288 /* c_f_s will call lustre_common_put_super on failure */
1291 #ifdef HAVE_SERVER_SUPPORT
1292 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1293 rc = server_fill_super(sb);
1294 /* s_f_s calls lustre_start_mgc after the mount because we need
1295 the MGS nids which are stored on disk. Plus, we may
1296 need to start the MGS first. */
1297 /* s_f_s will call server_put_super on failure */
1299 CERROR("This is client-side-only module, "
1300 "cannot handle server mount.\n");
1305 /* If error happens in fill_super() call, @lsi will be killed there.
1306 * This is why we do not put it here. */
1310 CERROR("Unable to mount %s (%d)\n",
1311 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1313 CDEBUG(D_SUPER, "Mount %s complete\n",
1321 /* We can't call ll_fill_super by name because it lives in a module that
1322 must be loaded after this one. */
1323 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
1324 struct vfsmount *mnt))
1326 client_fill_super = cfs;
1328 EXPORT_SYMBOL(lustre_register_client_fill_super);
1330 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
1332 kill_super_cb = cfs;
1334 EXPORT_SYMBOL(lustre_register_kill_super_cb);
1336 /***************** FS registration ******************/
1337 #ifdef HAVE_FSTYPE_MOUNT
1338 struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
1339 const char *devname, void *data)
1341 struct lustre_mount_data2 lmd2 = { data, NULL };
1343 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
1346 int lustre_get_sb(struct file_system_type *fs_type, int flags,
1347 const char *devname, void * data, struct vfsmount *mnt)
1349 struct lustre_mount_data2 lmd2 = { data, mnt };
1351 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
1355 void lustre_kill_super(struct super_block *sb)
1357 struct lustre_sb_info *lsi = s2lsi(sb);
1359 if (kill_super_cb && lsi && !IS_SERVER(lsi))
1360 (*kill_super_cb)(sb);
1362 kill_anon_super(sb);
1365 /** Register the "lustre" fs type
1367 struct file_system_type lustre_fs_type = {
1368 .owner = THIS_MODULE,
1370 #ifdef HAVE_FSTYPE_MOUNT
1371 .mount = lustre_mount,
1373 .get_sb = lustre_get_sb,
1375 .kill_sb = lustre_kill_super,
1376 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
1377 FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE,
1380 int lustre_register_fs(void)
1382 return register_filesystem(&lustre_fs_type);
1385 int lustre_unregister_fs(void)
1387 return unregister_filesystem(&lustre_fs_type);