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
49 #include <obd_class.h>
50 #include <lustre/lustre_user.h>
51 #include <linux/version.h>
52 #include <lustre_log.h>
53 #include <lustre_disk.h>
54 #include <lustre_param.h>
56 static int (*client_fill_super)(struct super_block *sb,
57 struct vfsmount *mnt);
59 static void (*kill_super_cb)(struct super_block *sb);
61 /**************** config llog ********************/
63 /** Get a config log from the MGS and process it.
64 * This func is called for both clients and servers.
65 * Continue to process new statements appended to the logs
66 * (whenever the config lock is revoked) until lustre_end_log
68 * @param sb The superblock is used by the MGC to write to the local copy of
70 * @param logname The name of the llog to replicate from the MGS
71 * @param cfg Since the same mgc may be used to follow multiple config logs
72 * (e.g. ost1, ost2, client), the config_llog_instance keeps the state for
73 * this log, and is added to the mgc's list of logs to follow.
75 int lustre_process_log(struct super_block *sb, char *logname,
76 struct config_llog_instance *cfg)
78 struct lustre_cfg *lcfg;
79 struct lustre_cfg_bufs *bufs;
80 struct lustre_sb_info *lsi = s2lsi(sb);
81 struct obd_device *mgc = lsi->lsi_mgc;
92 /* mgc_process_config */
93 lustre_cfg_bufs_reset(bufs, mgc->obd_name);
94 lustre_cfg_bufs_set_string(bufs, 1, logname);
95 lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg));
96 lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb));
97 lcfg = lustre_cfg_new(LCFG_LOG_START, bufs);
99 GOTO(out, rc = -ENOMEM);
100 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
101 lustre_cfg_free(lcfg);
106 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
107 "failed from the MGS (%d). Make sure this "
108 "client and the MGS are running compatible "
109 "versions of Lustre.\n",
110 mgc->obd_name, logname, rc);
112 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
113 "failed (%d). This may be the result of "
114 "communication errors between this node and "
115 "the MGS, a bad configuration, or other "
116 "errors. See the syslog for more "
117 "information.\n", mgc->obd_name, logname,
120 /* class_obd_list(); */
123 EXPORT_SYMBOL(lustre_process_log);
125 /* Stop watching this config log for updates */
126 int lustre_end_log(struct super_block *sb, char *logname,
127 struct config_llog_instance *cfg)
129 struct lustre_cfg *lcfg;
130 struct lustre_cfg_bufs bufs;
131 struct lustre_sb_info *lsi = s2lsi(sb);
132 struct obd_device *mgc = lsi->lsi_mgc;
139 /* mgc_process_config */
140 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
141 lustre_cfg_bufs_set_string(&bufs, 1, logname);
143 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
144 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
147 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
148 lustre_cfg_free(lcfg);
151 EXPORT_SYMBOL(lustre_end_log);
153 /**************** obd start *******************/
155 /** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from
156 * lctl (and do for echo cli/srv.
158 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
159 char *s1, char *s2, char *s3, char *s4)
161 struct lustre_cfg_bufs bufs;
162 struct lustre_cfg * lcfg = NULL;
165 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
166 cmd, s1, s2, s3, s4);
168 lustre_cfg_bufs_reset(&bufs, cfgname);
170 lustre_cfg_bufs_set_string(&bufs, 1, s1);
172 lustre_cfg_bufs_set_string(&bufs, 2, s2);
174 lustre_cfg_bufs_set_string(&bufs, 3, s3);
176 lustre_cfg_bufs_set_string(&bufs, 4, s4);
178 lcfg = lustre_cfg_new(cmd, &bufs);
181 lcfg->lcfg_nid = nid;
182 rc = class_process_config(lcfg);
183 lustre_cfg_free(lcfg);
186 EXPORT_SYMBOL(do_lcfg);
188 /** Call class_attach and class_setup. These methods in turn call
189 * obd type-specific methods.
191 int lustre_start_simple(char *obdname, char *type, char *uuid,
192 char *s1, char *s2, char *s3, char *s4)
195 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
197 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
199 CERROR("%s attach error %d\n", obdname, rc);
202 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, s3, s4);
204 CERROR("%s setup error %d\n", obdname, rc);
205 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
210 DEFINE_MUTEX(mgc_start_lock);
212 /** Set up a mgc obd to process startup logs
214 * \param sb [in] super block of the mgc obd
216 * \retval 0 success, otherwise error code
218 int lustre_start_mgc(struct super_block *sb)
220 struct obd_connect_data *data = NULL;
221 struct lustre_sb_info *lsi = s2lsi(sb);
222 struct obd_device *obd;
223 struct obd_export *exp;
224 struct obd_uuid *uuid;
227 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
230 int rc = 0, i = 0, j, len;
233 LASSERT(lsi->lsi_lmd);
235 /* Find the first non-lo MGS nid for our MGC name */
236 if (IS_SERVER(lsi)) {
237 /* mount -o mgsnode=nid */
238 ptr = lsi->lsi_lmd->lmd_mgs;
239 if (lsi->lsi_lmd->lmd_mgs &&
240 (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) {
242 } else if (IS_MGS(lsi)) {
243 lnet_process_id_t id;
244 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
245 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
252 } else { /* client */
253 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
254 ptr = lsi->lsi_lmd->lmd_dev;
255 if (class_parse_nid(ptr, &nid, &ptr) == 0)
259 CERROR("No valid MGS nids found.\n");
263 mutex_lock(&mgc_start_lock);
265 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
266 OBD_ALLOC(mgcname, len);
267 OBD_ALLOC(niduuid, len + 2);
268 if (!mgcname || !niduuid)
269 GOTO(out_free, rc = -ENOMEM);
270 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
272 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
276 GOTO(out_free, rc = -ENOMEM);
278 obd = class_name2obd(mgcname);
279 if (obd && !obd->obd_stopping) {
280 rc = obd_set_info_async(NULL, obd->obd_self_export,
281 strlen(KEY_MGSSEC), KEY_MGSSEC,
282 strlen(mgssec), mgssec, NULL);
286 /* Re-using an existing MGC */
287 atomic_inc(&obd->u.cli.cl_mgc_refcount);
289 /* IR compatibility check, only for clients */
290 if (lmd_is_client(lsi->lsi_lmd)) {
292 int vallen = sizeof(*data);
293 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
295 rc = obd_get_info(NULL, obd->obd_self_export,
296 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
297 &vallen, data, NULL);
299 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
300 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
301 /* LMD_FLG_NOIR is for test purpose only */
303 "Trying to mount a client with IR setting "
304 "not compatible with current mgc. "
305 "Force to use current mgc setting that is "
307 has_ir ? "enabled" : "disabled");
309 *flags &= ~LMD_FLG_NOIR;
311 *flags |= LMD_FLG_NOIR;
316 /* If we are restarting the MGS, don't try to keep the MGC's
317 old connection, or registration will fail. */
319 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
323 /* Try all connections, but only once (again).
324 We don't want to block another target from starting
325 (using its local copy of the log), but we do want to connect
326 if at all possible. */
328 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
329 rc = obd_set_info_async(NULL, obd->obd_self_export,
330 sizeof(KEY_INIT_RECOV_BACKUP),
331 KEY_INIT_RECOV_BACKUP,
332 sizeof(recov_bk), &recov_bk, NULL);
336 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
338 /* Add the primary nids for the MGS */
340 sprintf(niduuid, "%s_%x", mgcname, i);
341 if (IS_SERVER(lsi)) {
342 ptr = lsi->lsi_lmd->lmd_mgs;
343 CDEBUG(D_MOUNT, "mgs nids %s.\n", ptr);
345 /* Use local nids (including LO) */
346 lnet_process_id_t id;
347 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
348 rc = do_lcfg(mgcname, id.nid, LCFG_ADD_UUID,
352 /* Use mgsnode= nids */
353 /* mount -o mgsnode=nid */
354 if (lsi->lsi_lmd->lmd_mgs) {
355 ptr = lsi->lsi_lmd->lmd_mgs;
356 } else if (class_find_param(ptr, PARAM_MGSNODE,
358 CERROR("No MGS nids given.\n");
359 GOTO(out_free, rc = -EINVAL);
363 * Here we only take the first mgsnid as its primary
364 * serving mgs node, the rest mgsnid will be taken as
365 * failover mgs node, otherwise they would be takens
366 * as multiple nids of a single mgs node.
368 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
369 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
377 } else { /* client */
378 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
379 ptr = lsi->lsi_lmd->lmd_dev;
380 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
381 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
385 /* Stop at the first failover nid */
391 CERROR("No valid MGS nids found.\n");
392 GOTO(out_free, rc = -EINVAL);
394 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
396 /* Random uuid for MGC allows easier reconnects */
398 ll_generate_random_uuid(uuidc);
399 class_uuid_unparse(uuidc, uuid);
402 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
403 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
409 /* Add any failover MGS nids */
411 while (ptr && ((*ptr == ':' ||
412 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) {
413 /* New failover node */
414 sprintf(niduuid, "%s_%x", mgcname, i);
416 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
417 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
425 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
434 lsi->lsi_lmd->lmd_mgs_failnodes = i;
436 obd = class_name2obd(mgcname);
438 CERROR("Can't find mgcobd %s\n", mgcname);
439 GOTO(out_free, rc = -ENOTCONN);
442 rc = obd_set_info_async(NULL, obd->obd_self_export,
443 strlen(KEY_MGSSEC), KEY_MGSSEC,
444 strlen(mgssec), mgssec, NULL);
448 /* Keep a refcount of servers/clients who started with "mount",
449 so we know when we can get rid of the mgc. */
450 atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
452 /* Try all connections, but only once. */
454 rc = obd_set_info_async(NULL, obd->obd_self_export,
455 sizeof(KEY_INIT_RECOV_BACKUP),
456 KEY_INIT_RECOV_BACKUP,
457 sizeof(recov_bk), &recov_bk, NULL);
460 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
462 /* We connect to the MGS at setup, and don't disconnect until cleanup */
463 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
464 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV |
465 OBD_CONNECT_LVB_TYPE;
467 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
468 data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB;
470 #warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
473 if (lmd_is_client(lsi->lsi_lmd) &&
474 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
475 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
476 data->ocd_version = LUSTRE_VERSION_CODE;
477 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
479 CERROR("connect failed %d\n", rc);
483 obd->u.cli.cl_mgc_mgsexp = exp;
486 /* Keep the mgc info in the sb. Note that many lsi's can point
490 mutex_unlock(&mgc_start_lock);
495 OBD_FREE(mgcname, len);
497 OBD_FREE(niduuid, len + 2);
501 static int lustre_stop_mgc(struct super_block *sb)
503 struct lustre_sb_info *lsi = s2lsi(sb);
504 struct obd_device *obd;
505 char *niduuid = 0, *ptr = 0;
506 int i, rc = 0, len = 0;
516 mutex_lock(&mgc_start_lock);
517 LASSERT(atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
518 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
519 /* This is not fatal, every client that stops
520 will call in here. */
521 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
522 atomic_read(&obd->u.cli.cl_mgc_refcount));
523 GOTO(out, rc = -EBUSY);
526 /* The MGC has no recoverable data in any case.
527 * force shotdown set in umount_begin */
528 obd->obd_no_recov = 1;
530 if (obd->u.cli.cl_mgc_mgsexp) {
531 /* An error is not fatal, if we are unable to send the
532 disconnect mgs ping evictor cleans up the export */
533 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
535 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
538 /* Save the obdname for cleaning the nid uuids, which are
540 len = strlen(obd->obd_name) + 6;
541 OBD_ALLOC(niduuid, len);
543 strcpy(niduuid, obd->obd_name);
544 ptr = niduuid + strlen(niduuid);
547 rc = class_manual_cleanup(obd);
551 /* Clean the nid uuids */
553 GOTO(out, rc = -ENOMEM);
555 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
556 sprintf(ptr, "_%x", i);
557 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
560 CERROR("del MDC UUID %s failed: rc = %d\n",
565 OBD_FREE(niduuid, len);
567 /* class_import_put will get rid of the additional connections */
568 mutex_unlock(&mgc_start_lock);
572 /***************** lustre superblock **************/
574 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
576 struct lustre_sb_info *lsi;
582 OBD_ALLOC_PTR(lsi->lsi_lmd);
588 lsi->lsi_lmd->lmd_exclude_count = 0;
589 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
590 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
591 s2lsi_nocast(sb) = lsi;
592 /* we take 1 extra ref for our setup */
593 atomic_set(&lsi->lsi_mounts, 1);
595 /* Default umount style */
596 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
597 INIT_LIST_HEAD(&lsi->lsi_lwp_list);
598 spin_lock_init(&lsi->lsi_lwp_lock);
603 static int lustre_free_lsi(struct super_block *sb)
605 struct lustre_sb_info *lsi = s2lsi(sb);
608 LASSERT(lsi != NULL);
609 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
611 /* someone didn't call server_put_mount. */
612 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
614 if (lsi->lsi_lmd != NULL) {
615 if (lsi->lsi_lmd->lmd_dev != NULL)
616 OBD_FREE(lsi->lsi_lmd->lmd_dev,
617 strlen(lsi->lsi_lmd->lmd_dev) + 1);
618 if (lsi->lsi_lmd->lmd_profile != NULL)
619 OBD_FREE(lsi->lsi_lmd->lmd_profile,
620 strlen(lsi->lsi_lmd->lmd_profile) + 1);
621 if (lsi->lsi_lmd->lmd_mgssec != NULL)
622 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
623 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
624 if (lsi->lsi_lmd->lmd_opts != NULL)
625 OBD_FREE(lsi->lsi_lmd->lmd_opts,
626 strlen(lsi->lsi_lmd->lmd_opts) + 1);
627 if (lsi->lsi_lmd->lmd_exclude_count)
628 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
629 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
630 lsi->lsi_lmd->lmd_exclude_count);
631 if (lsi->lsi_lmd->lmd_mgs != NULL)
632 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
633 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
634 if (lsi->lsi_lmd->lmd_osd_type != NULL)
635 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
636 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
637 if (lsi->lsi_lmd->lmd_params != NULL)
638 OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
640 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
643 LASSERT(lsi->lsi_llsbi == NULL);
644 OBD_FREE(lsi, sizeof(*lsi));
645 s2lsi_nocast(sb) = NULL;
650 /* The lsi has one reference for every server that is using the disk -
651 e.g. MDT, MGS, and potentially MGC */
652 int lustre_put_lsi(struct super_block *sb)
654 struct lustre_sb_info *lsi = s2lsi(sb);
657 LASSERT(lsi != NULL);
659 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
660 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
661 if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
662 lu_device_put(&lsi->lsi_dt_dev->dd_lu_dev);
663 lsi->lsi_osd_exp->exp_obd->obd_lvfs_ctxt.dt = NULL;
664 lsi->lsi_dt_dev = NULL;
665 obd_disconnect(lsi->lsi_osd_exp);
666 /* wait till OSD is gone */
667 obd_zombie_barrier();
676 * <FSNAME><SEPERATOR><TYPE><INDEX>
677 * FSNAME is between 1 and 8 characters (inclusive).
678 * Excluded characters are '/' and ':'
679 * SEPERATOR is either ':' or '-'
680 * TYPE: "OST", "MDT", etc.
681 * INDEX: Hex representation of the index
684 /** Get the fsname ("lustre") from the server name ("lustre-OST003F").
685 * @param [in] svname server name including type and index
686 * @param [out] fsname Buffer to copy filesystem name prefix into.
687 * Must have at least 'strlen(fsname) + 1' chars.
688 * @param [out] endptr if endptr isn't NULL it is set to end of fsname
691 int server_name2fsname(const char *svname, char *fsname, const char **endptr)
695 dash = svname + strnlen(svname, 8); /* max fsname length is 8 */
696 for (; dash > svname && *dash != '-' && *dash != ':'; dash--)
701 if (fsname != NULL) {
702 strncpy(fsname, svname, dash - svname);
703 fsname[dash - svname] = '\0';
711 EXPORT_SYMBOL(server_name2fsname);
714 * Get service name (svname) from string
716 * if endptr isn't NULL it is set to end of fsname *
718 int server_name2svname(const char *label, char *svname, const char **endptr,
724 /* We use server_name2fsname() just for parsing */
725 rc = server_name2fsname(label, NULL, &dash);
732 if (strlcpy(svname, dash + 1, svsize) >= svsize)
737 EXPORT_SYMBOL(server_name2svname);
740 * check server name is OST.
742 int server_name_is_ost(const char *svname)
747 /* We use server_name2fsname() just for parsing */
748 rc = server_name2fsname(svname, NULL, &dash);
754 if (strncmp(dash, "OST", 3) == 0)
758 EXPORT_SYMBOL(server_name_is_ost);
761 * Get the index from the target name MDTXXXX/OSTXXXX
762 * rc = server type, or rc < 0 on error
764 int target_name2index(const char *tgtname, __u32 *idx, const char **endptr)
766 const char *dash = tgtname;
770 if (strncmp(dash, "MDT", 3) == 0)
771 rc = LDD_F_SV_TYPE_MDT;
772 else if (strncmp(dash, "OST", 3) == 0)
773 rc = LDD_F_SV_TYPE_OST;
779 if (strncmp(dash, "all", 3) == 0) {
782 return rc | LDD_F_SV_ALL;
785 index = simple_strtoul(dash, (char **)endptr, 16);
790 EXPORT_SYMBOL(target_name2index);
792 /* Get the index from the obd name.
795 if endptr isn't NULL it is set to end of name */
796 int server_name2index(const char *svname, __u32 *idx, const char **endptr)
801 /* We use server_name2fsname() just for parsing */
802 rc = server_name2fsname(svname, NULL, &dash);
807 rc = target_name2index(dash, idx, endptr);
811 /* Account for -mdc after index that is possible when specifying mdt */
812 if (endptr != NULL && strncmp(LUSTRE_MDC_NAME, *endptr + 1,
813 sizeof(LUSTRE_MDC_NAME)-1) == 0)
814 *endptr += sizeof(LUSTRE_MDC_NAME);
818 EXPORT_SYMBOL(server_name2index);
820 /*************** mount common betweeen server and client ***************/
823 int lustre_common_put_super(struct super_block *sb)
828 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
830 /* Drop a ref to the MGC */
831 rc = lustre_stop_mgc(sb);
832 if (rc && (rc != -ENOENT)) {
834 CERROR("Can't stop MGC: %d\n", rc);
837 /* BUSY just means that there's some other obd that
838 needs the mgc. Let him clean it up. */
839 CDEBUG(D_MOUNT, "MGC still in use\n");
841 /* Drop a ref to the mounted disk */
846 EXPORT_SYMBOL(lustre_common_put_super);
848 static void lmd_print(struct lustre_mount_data *lmd)
852 PRINT_CMD(D_MOUNT, " mount data:\n");
853 if (lmd_is_client(lmd))
854 PRINT_CMD(D_MOUNT, "profile: %s\n", lmd->lmd_profile);
855 PRINT_CMD(D_MOUNT, "device: %s\n", lmd->lmd_dev);
856 PRINT_CMD(D_MOUNT, "flags: %x\n", lmd->lmd_flags);
859 PRINT_CMD(D_MOUNT, "options: %s\n", lmd->lmd_opts);
861 if (lmd->lmd_recovery_time_soft)
862 PRINT_CMD(D_MOUNT, "recovery time soft: %d\n",
863 lmd->lmd_recovery_time_soft);
865 if (lmd->lmd_recovery_time_hard)
866 PRINT_CMD(D_MOUNT, "recovery time hard: %d\n",
867 lmd->lmd_recovery_time_hard);
869 for (i = 0; i < lmd->lmd_exclude_count; i++) {
870 PRINT_CMD(D_MOUNT, "exclude %d: OST%04x\n", i,
871 lmd->lmd_exclude[i]);
875 /* Is this server on the exclusion list */
876 int lustre_check_exclusion(struct super_block *sb, char *svname)
878 struct lustre_sb_info *lsi = s2lsi(sb);
879 struct lustre_mount_data *lmd = lsi->lsi_lmd;
884 rc = server_name2index(svname, &index, NULL);
885 if (rc != LDD_F_SV_TYPE_OST)
886 /* Only exclude OSTs */
889 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
890 index, lmd->lmd_exclude_count, lmd->lmd_dev);
892 for(i = 0; i < lmd->lmd_exclude_count; i++) {
893 if (index == lmd->lmd_exclude[i]) {
894 CWARN("Excluding %s (on exclusion list)\n", svname);
901 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
902 static int lmd_make_exclusion(struct lustre_mount_data *lmd, const char *ptr)
904 const char *s1 = ptr, *s2;
905 __u32 index, *exclude_list;
909 /* The shortest an ost name can be is 8 chars: -OST0000.
910 We don't actually know the fsname at this time, so in fact
911 a user could specify any fsname. */
912 devmax = strlen(ptr) / 8 + 1;
914 /* temp storage until we figure out how many we have */
915 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
919 /* we enter this fn pointing at the '=' */
920 while (*s1 && *s1 != ' ' && *s1 != ',') {
922 rc = server_name2index(s1, &index, &s2);
924 CERROR("Can't parse server name '%s': rc = %d\n",
928 if (rc == LDD_F_SV_TYPE_OST)
929 exclude_list[lmd->lmd_exclude_count++] = index;
931 CDEBUG(D_MOUNT, "ignoring exclude %.*s: type = %#x\n",
932 (uint)(s2-s1), s1, rc);
934 /* now we are pointing at ':' (next exclude)
935 or ',' (end of excludes) */
936 if (lmd->lmd_exclude_count >= devmax)
939 if (rc >= 0) /* non-err */
942 if (lmd->lmd_exclude_count) {
943 /* permanent, freed in lustre_free_lsi */
944 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
945 lmd->lmd_exclude_count);
946 if (lmd->lmd_exclude) {
947 memcpy(lmd->lmd_exclude, exclude_list,
948 sizeof(index) * lmd->lmd_exclude_count);
951 lmd->lmd_exclude_count = 0;
954 OBD_FREE(exclude_list, sizeof(index) * devmax);
958 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
963 if (lmd->lmd_mgssec != NULL) {
964 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
965 lmd->lmd_mgssec = NULL;
968 tail = strchr(ptr, ',');
970 length = strlen(ptr);
974 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
975 if (lmd->lmd_mgssec == NULL)
978 memcpy(lmd->lmd_mgssec, ptr, length);
979 lmd->lmd_mgssec[length] = '\0';
983 static int lmd_parse_string(char **handle, char *ptr)
988 if ((handle == NULL) || (ptr == NULL))
991 if (*handle != NULL) {
992 OBD_FREE(*handle, strlen(*handle) + 1);
996 tail = strchr(ptr, ',');
998 length = strlen(ptr);
1000 length = tail - ptr;
1002 OBD_ALLOC(*handle, length + 1);
1003 if (*handle == NULL)
1006 memcpy(*handle, ptr, length);
1007 (*handle)[length] = '\0';
1012 /* Collect multiple values for mgsnid specifiers */
1013 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
1021 /* Find end of nidlist */
1022 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
1023 length = tail - *ptr;
1025 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
1029 if (lmd->lmd_mgs != NULL)
1030 oldlen = strlen(lmd->lmd_mgs) + 1;
1032 OBD_ALLOC(mgsnid, oldlen + length + 1);
1036 if (lmd->lmd_mgs != NULL) {
1037 /* Multiple mgsnid= are taken to mean failover locations */
1038 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
1039 mgsnid[oldlen - 1] = ':';
1040 OBD_FREE(lmd->lmd_mgs, oldlen);
1042 memcpy(mgsnid + oldlen, *ptr, length);
1043 mgsnid[oldlen + length] = '\0';
1044 lmd->lmd_mgs = mgsnid;
1050 /** Parse mount line options
1051 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
1052 * dev is passed as device=uml1:/lustre by mount.lustre
1054 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1056 char *s1, *s2, *s3, *devname = NULL;
1057 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1063 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1064 "/sbin/mount.lustre is installed.\n");
1068 /* Options should be a string - try to detect old lmd data */
1069 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1070 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1071 "/sbin/mount.lustre. Please install "
1072 "version %s\n", LUSTRE_VERSION_STRING);
1075 lmd->lmd_magic = LMD_MAGIC;
1077 OBD_ALLOC(lmd->lmd_params, 4096);
1078 if (lmd->lmd_params == NULL)
1080 lmd->lmd_params[0] = '\0';
1082 /* Set default flags here */
1087 int time_min = OBD_RECOVERY_TIME_MIN;
1089 /* Skip whitespace and extra commas */
1090 while (*s1 == ' ' || *s1 == ',')
1094 /* Client options are parsed in ll_options: eg. flock,
1097 /* Parse non-ldiskfs options here. Rather than modifying
1098 ldiskfs, we just zero these out here */
1099 if (strncmp(s1, "abort_recov", 11) == 0) {
1100 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1102 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
1103 lmd->lmd_recovery_time_soft = max_t(int,
1104 simple_strtoul(s1 + 19, NULL, 10), time_min);
1106 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
1107 lmd->lmd_recovery_time_hard = max_t(int,
1108 simple_strtoul(s1 + 19, NULL, 10), time_min);
1110 } else if (strncmp(s1, "noir", 4) == 0) {
1111 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
1113 } else if (strncmp(s1, "nosvc", 5) == 0) {
1114 lmd->lmd_flags |= LMD_FLG_NOSVC;
1116 } else if (strncmp(s1, "nomgs", 5) == 0) {
1117 lmd->lmd_flags |= LMD_FLG_NOMGS;
1119 } else if (strncmp(s1, "noscrub", 7) == 0) {
1120 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
1122 } else if (strncmp(s1, PARAM_MGSNODE,
1123 sizeof(PARAM_MGSNODE) - 1) == 0) {
1124 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
1125 /* Assume the next mount opt is the first
1126 invalid nid we get to. */
1127 rc = lmd_parse_mgs(lmd, &s2);
1132 } else if (strncmp(s1, "writeconf", 9) == 0) {
1133 lmd->lmd_flags |= LMD_FLG_WRITECONF;
1135 } else if (strncmp(s1, "update", 6) == 0) {
1136 lmd->lmd_flags |= LMD_FLG_UPDATE;
1138 } else if (strncmp(s1, "virgin", 6) == 0) {
1139 lmd->lmd_flags |= LMD_FLG_VIRGIN;
1141 } else if (strncmp(s1, "noprimnode", 10) == 0) {
1142 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
1144 } else if (strncmp(s1, "mgssec=", 7) == 0) {
1145 rc = lmd_parse_mgssec(lmd, s1 + 7);
1149 /* ost exclusion list */
1150 } else if (strncmp(s1, "exclude=", 8) == 0) {
1151 rc = lmd_make_exclusion(lmd, s1 + 7);
1155 } else if (strncmp(s1, "mgs", 3) == 0) {
1157 lmd->lmd_flags |= LMD_FLG_MGS;
1159 } else if (strncmp(s1, "svname=", 7) == 0) {
1160 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
1164 } else if (strncmp(s1, "param=", 6) == 0) {
1166 char *tail = strchr(s1 + 6, ',');
1168 length = strlen(s1);
1171 char *param_str = tail + 1;
1172 int supplementary = 1;
1174 while (class_parse_nid_quiet(param_str, &nid,
1178 length = param_str - s1 - supplementary;
1181 strncat(lmd->lmd_params, s1 + 6, length);
1182 strcat(lmd->lmd_params, " ");
1183 s3 = s1 + 6 + length;
1185 } else if (strncmp(s1, "osd=", 4) == 0) {
1186 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
1191 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1192 end of the options. */
1193 else if (strncmp(s1, "device=", 7) == 0) {
1195 /* terminate options right before device. device
1196 must be the last one. */
1202 s2 = strchr(s3, ',');
1210 memmove(s1, s2, strlen(s2) + 1);
1216 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1217 "(need mount option 'device=...')\n");
1221 s1 = strstr(devname, ":/");
1224 lmd->lmd_flags |= LMD_FLG_CLIENT;
1225 /* Remove leading /s from fsname */
1226 while (*++s1 == '/') ;
1227 /* Freed in lustre_free_lsi */
1228 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1229 if (!lmd->lmd_profile)
1231 sprintf(lmd->lmd_profile, "%s-client", s1);
1234 /* Freed in lustre_free_lsi */
1235 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1238 strcpy(lmd->lmd_dev, devname);
1240 /* Save mount options */
1241 s1 = options + strlen(options) - 1;
1242 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1244 if (*options != 0) {
1245 /* Freed in lustre_free_lsi */
1246 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1249 strcpy(lmd->lmd_opts, options);
1253 lmd->lmd_magic = LMD_MAGIC;
1258 CERROR("Bad mount options %s\n", options);
1262 struct lustre_mount_data2 {
1264 struct vfsmount *lmd2_mnt;
1267 /** This is the entry point for the mount call into Lustre.
1268 * This is called when a server or client is mounted,
1269 * and this is where we start setting things up.
1270 * @param data Mount options (e.g. -o flock,abort_recov)
1272 int lustre_fill_super(struct super_block *sb, void *data, int silent)
1274 struct lustre_mount_data *lmd;
1275 struct lustre_mount_data2 *lmd2 = data;
1276 struct lustre_sb_info *lsi;
1280 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1282 lsi = lustre_init_lsi(sb);
1288 * Disable lockdep during mount, because mount locking patterns are
1294 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
1296 obd_zombie_barrier();
1298 /* Figure out the lmd from the mount options */
1299 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
1301 GOTO(out, rc = -EINVAL);
1304 if (lmd_is_client(lmd)) {
1305 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1306 if (!client_fill_super) {
1307 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1308 "client mount! Is the 'lustre' "
1309 "module loaded?\n");
1313 rc = lustre_start_mgc(sb);
1318 /* Connect and start */
1319 /* (should always be ll_fill_super) */
1320 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
1321 /* c_f_s will call lustre_common_put_super on failure */
1324 #ifdef HAVE_SERVER_SUPPORT
1325 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1326 rc = server_fill_super(sb);
1327 /* s_f_s calls lustre_start_mgc after the mount because we need
1328 the MGS nids which are stored on disk. Plus, we may
1329 need to start the MGS first. */
1330 /* s_f_s will call server_put_super on failure */
1332 CERROR("This is client-side-only module, "
1333 "cannot handle server mount.\n");
1338 /* If error happens in fill_super() call, @lsi will be killed there.
1339 * This is why we do not put it here. */
1343 CERROR("Unable to mount %s (%d)\n",
1344 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1346 CDEBUG(D_SUPER, "Mount %s complete\n",
1354 /* We can't call ll_fill_super by name because it lives in a module that
1355 must be loaded after this one. */
1356 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
1357 struct vfsmount *mnt))
1359 client_fill_super = cfs;
1361 EXPORT_SYMBOL(lustre_register_client_fill_super);
1363 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
1365 kill_super_cb = cfs;
1367 EXPORT_SYMBOL(lustre_register_kill_super_cb);
1369 /***************** FS registration ******************/
1370 #ifdef HAVE_FSTYPE_MOUNT
1371 struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
1372 const char *devname, void *data)
1374 struct lustre_mount_data2 lmd2 = { data, NULL };
1376 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
1379 int lustre_get_sb(struct file_system_type *fs_type, int flags,
1380 const char *devname, void * data, struct vfsmount *mnt)
1382 struct lustre_mount_data2 lmd2 = { data, mnt };
1384 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
1388 void lustre_kill_super(struct super_block *sb)
1390 struct lustre_sb_info *lsi = s2lsi(sb);
1392 if (kill_super_cb && lsi && !IS_SERVER(lsi))
1393 (*kill_super_cb)(sb);
1395 kill_anon_super(sb);
1398 /** Register the "lustre" fs type
1400 struct file_system_type lustre_fs_type = {
1401 .owner = THIS_MODULE,
1403 #ifdef HAVE_FSTYPE_MOUNT
1404 .mount = lustre_mount,
1406 .get_sb = lustre_get_sb,
1408 .kill_sb = lustre_kill_super,
1409 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
1410 FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE,
1413 int lustre_register_fs(void)
1415 return register_filesystem(&lustre_fs_type);
1418 int lustre_unregister_fs(void)
1420 return unregister_filesystem(&lustre_fs_type);