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);
98 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
99 lustre_cfg_free(lcfg);
104 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
105 "failed from the MGS (%d). Make sure this "
106 "client and the MGS are running compatible "
107 "versions of Lustre.\n",
108 mgc->obd_name, logname, rc);
110 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
111 "failed (%d). This may be the result of "
112 "communication errors between this node and "
113 "the MGS, a bad configuration, or other "
114 "errors. See the syslog for more "
115 "information.\n", mgc->obd_name, logname,
118 /* class_obd_list(); */
121 EXPORT_SYMBOL(lustre_process_log);
123 /* Stop watching this config log for updates */
124 int lustre_end_log(struct super_block *sb, char *logname,
125 struct config_llog_instance *cfg)
127 struct lustre_cfg *lcfg;
128 struct lustre_cfg_bufs bufs;
129 struct lustre_sb_info *lsi = s2lsi(sb);
130 struct obd_device *mgc = lsi->lsi_mgc;
137 /* mgc_process_config */
138 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
139 lustre_cfg_bufs_set_string(&bufs, 1, logname);
141 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
142 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
143 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
144 lustre_cfg_free(lcfg);
147 EXPORT_SYMBOL(lustre_end_log);
149 /**************** obd start *******************/
151 /** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from
152 * lctl (and do for echo cli/srv.
154 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
155 char *s1, char *s2, char *s3, char *s4)
157 struct lustre_cfg_bufs bufs;
158 struct lustre_cfg * lcfg = NULL;
161 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
162 cmd, s1, s2, s3, s4);
164 lustre_cfg_bufs_reset(&bufs, cfgname);
166 lustre_cfg_bufs_set_string(&bufs, 1, s1);
168 lustre_cfg_bufs_set_string(&bufs, 2, s2);
170 lustre_cfg_bufs_set_string(&bufs, 3, s3);
172 lustre_cfg_bufs_set_string(&bufs, 4, s4);
174 lcfg = lustre_cfg_new(cmd, &bufs);
175 lcfg->lcfg_nid = nid;
176 rc = class_process_config(lcfg);
177 lustre_cfg_free(lcfg);
180 EXPORT_SYMBOL(do_lcfg);
182 /** Call class_attach and class_setup. These methods in turn call
183 * obd type-specific methods.
185 int lustre_start_simple(char *obdname, char *type, char *uuid,
186 char *s1, char *s2, char *s3, char *s4)
189 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
191 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
193 CERROR("%s attach error %d\n", obdname, rc);
196 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, s3, s4);
198 CERROR("%s setup error %d\n", obdname, rc);
199 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
204 DEFINE_MUTEX(mgc_start_lock);
206 /** Set up a mgc obd to process startup logs
208 * \param sb [in] super block of the mgc obd
210 * \retval 0 success, otherwise error code
212 int lustre_start_mgc(struct super_block *sb)
214 struct obd_connect_data *data = NULL;
215 struct lustre_sb_info *lsi = s2lsi(sb);
216 struct obd_device *obd;
217 struct obd_export *exp;
218 struct obd_uuid *uuid;
221 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
224 int rc = 0, i = 0, j, len;
227 LASSERT(lsi->lsi_lmd);
229 /* Find the first non-lo MGS nid for our MGC name */
230 if (IS_SERVER(lsi)) {
231 /* mount -o mgsnode=nid */
232 ptr = lsi->lsi_lmd->lmd_mgs;
233 if (lsi->lsi_lmd->lmd_mgs &&
234 (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) {
236 } else if (IS_MGS(lsi)) {
237 lnet_process_id_t id;
238 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
239 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
246 } else { /* client */
247 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
248 ptr = lsi->lsi_lmd->lmd_dev;
249 if (class_parse_nid(ptr, &nid, &ptr) == 0)
253 CERROR("No valid MGS nids found.\n");
257 mutex_lock(&mgc_start_lock);
259 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
260 OBD_ALLOC(mgcname, len);
261 OBD_ALLOC(niduuid, len + 2);
262 if (!mgcname || !niduuid)
263 GOTO(out_free, rc = -ENOMEM);
264 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
266 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
270 GOTO(out_free, rc = -ENOMEM);
272 obd = class_name2obd(mgcname);
273 if (obd && !obd->obd_stopping) {
274 rc = obd_set_info_async(NULL, obd->obd_self_export,
275 strlen(KEY_MGSSEC), KEY_MGSSEC,
276 strlen(mgssec), mgssec, NULL);
280 /* Re-using an existing MGC */
281 atomic_inc(&obd->u.cli.cl_mgc_refcount);
283 /* IR compatibility check, only for clients */
284 if (lmd_is_client(lsi->lsi_lmd)) {
286 int vallen = sizeof(*data);
287 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
289 rc = obd_get_info(NULL, obd->obd_self_export,
290 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
291 &vallen, data, NULL);
293 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
294 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
295 /* LMD_FLG_NOIR is for test purpose only */
297 "Trying to mount a client with IR setting "
298 "not compatible with current mgc. "
299 "Force to use current mgc setting that is "
301 has_ir ? "enabled" : "disabled");
303 *flags &= ~LMD_FLG_NOIR;
305 *flags |= LMD_FLG_NOIR;
310 /* If we are restarting the MGS, don't try to keep the MGC's
311 old connection, or registration will fail. */
313 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
317 /* Try all connections, but only once (again).
318 We don't want to block another target from starting
319 (using its local copy of the log), but we do want to connect
320 if at all possible. */
322 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
323 rc = obd_set_info_async(NULL, obd->obd_self_export,
324 sizeof(KEY_INIT_RECOV_BACKUP),
325 KEY_INIT_RECOV_BACKUP,
326 sizeof(recov_bk), &recov_bk, NULL);
330 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
332 /* Add the primary nids for the MGS */
334 sprintf(niduuid, "%s_%x", mgcname, i);
335 if (IS_SERVER(lsi)) {
336 ptr = lsi->lsi_lmd->lmd_mgs;
337 CDEBUG(D_MOUNT, "mgs nids %s.\n", ptr);
339 /* Use local nids (including LO) */
340 lnet_process_id_t id;
341 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
342 rc = do_lcfg(mgcname, id.nid, LCFG_ADD_UUID,
346 /* Use mgsnode= nids */
347 /* mount -o mgsnode=nid */
348 if (lsi->lsi_lmd->lmd_mgs) {
349 ptr = lsi->lsi_lmd->lmd_mgs;
350 } else if (class_find_param(ptr, PARAM_MGSNODE,
352 CERROR("No MGS nids given.\n");
353 GOTO(out_free, rc = -EINVAL);
357 * Here we only take the first mgsnid as its primary
358 * serving mgs node, the rest mgsnid will be taken as
359 * failover mgs node, otherwise they would be takens
360 * as multiple nids of a single mgs node.
362 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
363 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
371 } else { /* client */
372 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
373 ptr = lsi->lsi_lmd->lmd_dev;
374 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
375 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
379 /* Stop at the first failover nid */
385 CERROR("No valid MGS nids found.\n");
386 GOTO(out_free, rc = -EINVAL);
388 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
390 /* Random uuid for MGC allows easier reconnects */
392 ll_generate_random_uuid(uuidc);
393 class_uuid_unparse(uuidc, uuid);
396 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
397 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
403 /* Add any failover MGS nids */
405 while (ptr && ((*ptr == ':' ||
406 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) {
407 /* New failover node */
408 sprintf(niduuid, "%s_%x", mgcname, i);
410 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
411 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
419 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
428 lsi->lsi_lmd->lmd_mgs_failnodes = i;
430 obd = class_name2obd(mgcname);
432 CERROR("Can't find mgcobd %s\n", mgcname);
433 GOTO(out_free, rc = -ENOTCONN);
436 rc = obd_set_info_async(NULL, obd->obd_self_export,
437 strlen(KEY_MGSSEC), KEY_MGSSEC,
438 strlen(mgssec), mgssec, NULL);
442 /* Keep a refcount of servers/clients who started with "mount",
443 so we know when we can get rid of the mgc. */
444 atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
446 /* Try all connections, but only once. */
448 rc = obd_set_info_async(NULL, obd->obd_self_export,
449 sizeof(KEY_INIT_RECOV_BACKUP),
450 KEY_INIT_RECOV_BACKUP,
451 sizeof(recov_bk), &recov_bk, NULL);
454 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
456 /* We connect to the MGS at setup, and don't disconnect until cleanup */
457 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
458 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV |
459 OBD_CONNECT_LVB_TYPE;
461 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
462 data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB;
464 #warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
467 if (lmd_is_client(lsi->lsi_lmd) &&
468 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
469 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
470 data->ocd_version = LUSTRE_VERSION_CODE;
471 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
473 CERROR("connect failed %d\n", rc);
477 obd->u.cli.cl_mgc_mgsexp = exp;
480 /* Keep the mgc info in the sb. Note that many lsi's can point
484 mutex_unlock(&mgc_start_lock);
489 OBD_FREE(mgcname, len);
491 OBD_FREE(niduuid, len + 2);
495 static int lustre_stop_mgc(struct super_block *sb)
497 struct lustre_sb_info *lsi = s2lsi(sb);
498 struct obd_device *obd;
499 char *niduuid = 0, *ptr = 0;
500 int i, rc = 0, len = 0;
510 mutex_lock(&mgc_start_lock);
511 LASSERT(atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
512 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
513 /* This is not fatal, every client that stops
514 will call in here. */
515 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
516 atomic_read(&obd->u.cli.cl_mgc_refcount));
517 GOTO(out, rc = -EBUSY);
520 /* The MGC has no recoverable data in any case.
521 * force shotdown set in umount_begin */
522 obd->obd_no_recov = 1;
524 if (obd->u.cli.cl_mgc_mgsexp) {
525 /* An error is not fatal, if we are unable to send the
526 disconnect mgs ping evictor cleans up the export */
527 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
529 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
532 /* Save the obdname for cleaning the nid uuids, which are
534 len = strlen(obd->obd_name) + 6;
535 OBD_ALLOC(niduuid, len);
537 strcpy(niduuid, obd->obd_name);
538 ptr = niduuid + strlen(niduuid);
541 rc = class_manual_cleanup(obd);
545 /* Clean the nid uuids */
547 GOTO(out, rc = -ENOMEM);
549 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
550 sprintf(ptr, "_%x", i);
551 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
554 CERROR("del MDC UUID %s failed: rc = %d\n",
559 OBD_FREE(niduuid, len);
561 /* class_import_put will get rid of the additional connections */
562 mutex_unlock(&mgc_start_lock);
566 /***************** lustre superblock **************/
568 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
570 struct lustre_sb_info *lsi;
576 OBD_ALLOC_PTR(lsi->lsi_lmd);
582 lsi->lsi_lmd->lmd_exclude_count = 0;
583 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
584 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
585 s2lsi_nocast(sb) = lsi;
586 /* we take 1 extra ref for our setup */
587 atomic_set(&lsi->lsi_mounts, 1);
589 /* Default umount style */
590 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
595 static int lustre_free_lsi(struct super_block *sb)
597 struct lustre_sb_info *lsi = s2lsi(sb);
600 LASSERT(lsi != NULL);
601 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
603 /* someone didn't call server_put_mount. */
604 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
606 if (lsi->lsi_lmd != NULL) {
607 if (lsi->lsi_lmd->lmd_dev != NULL)
608 OBD_FREE(lsi->lsi_lmd->lmd_dev,
609 strlen(lsi->lsi_lmd->lmd_dev) + 1);
610 if (lsi->lsi_lmd->lmd_profile != NULL)
611 OBD_FREE(lsi->lsi_lmd->lmd_profile,
612 strlen(lsi->lsi_lmd->lmd_profile) + 1);
613 if (lsi->lsi_lmd->lmd_mgssec != NULL)
614 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
615 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
616 if (lsi->lsi_lmd->lmd_opts != NULL)
617 OBD_FREE(lsi->lsi_lmd->lmd_opts,
618 strlen(lsi->lsi_lmd->lmd_opts) + 1);
619 if (lsi->lsi_lmd->lmd_exclude_count)
620 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
621 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
622 lsi->lsi_lmd->lmd_exclude_count);
623 if (lsi->lsi_lmd->lmd_mgs != NULL)
624 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
625 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
626 if (lsi->lsi_lmd->lmd_osd_type != NULL)
627 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
628 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
629 if (lsi->lsi_lmd->lmd_params != NULL)
630 OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
632 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
635 LASSERT(lsi->lsi_llsbi == NULL);
636 OBD_FREE(lsi, sizeof(*lsi));
637 s2lsi_nocast(sb) = NULL;
642 /* The lsi has one reference for every server that is using the disk -
643 e.g. MDT, MGS, and potentially MGC */
644 int lustre_put_lsi(struct super_block *sb)
646 struct lustre_sb_info *lsi = s2lsi(sb);
649 LASSERT(lsi != NULL);
651 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
652 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
653 if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
654 lu_device_put(&lsi->lsi_dt_dev->dd_lu_dev);
655 lsi->lsi_osd_exp->exp_obd->obd_lvfs_ctxt.dt = NULL;
656 lsi->lsi_dt_dev = NULL;
657 obd_disconnect(lsi->lsi_osd_exp);
658 /* wait till OSD is gone */
659 obd_zombie_barrier();
668 * <FSNAME><SEPERATOR><TYPE><INDEX>
669 * FSNAME is between 1 and 8 characters (inclusive).
670 * Excluded characters are '/' and ':'
671 * SEPERATOR is either ':' or '-'
672 * TYPE: "OST", "MDT", etc.
673 * INDEX: Hex representation of the index
676 /** Get the fsname ("lustre") from the server name ("lustre-OST003F").
677 * @param [in] svname server name including type and index
678 * @param [out] fsname Buffer to copy filesystem name prefix into.
679 * Must have at least 'strlen(fsname) + 1' chars.
680 * @param [out] endptr if endptr isn't NULL it is set to end of fsname
683 int server_name2fsname(const char *svname, char *fsname, const char **endptr)
687 dash = svname + strnlen(svname, 8); /* max fsname length is 8 */
688 for (; dash > svname && *dash != '-' && *dash != ':'; dash--)
693 if (fsname != NULL) {
694 strncpy(fsname, svname, dash - svname);
695 fsname[dash - svname] = '\0';
703 EXPORT_SYMBOL(server_name2fsname);
706 * Get service name (svname) from string
708 * if endptr isn't NULL it is set to end of fsname *
710 int server_name2svname(const char *label, char *svname, const char **endptr,
716 /* We use server_name2fsname() just for parsing */
717 rc = server_name2fsname(label, NULL, &dash);
724 if (strlcpy(svname, dash + 1, svsize) >= svsize)
729 EXPORT_SYMBOL(server_name2svname);
732 * check server name is OST.
734 int server_name_is_ost(const char *svname)
739 /* We use server_name2fsname() just for parsing */
740 rc = server_name2fsname(svname, NULL, &dash);
746 if (strncmp(dash, "OST", 3) == 0)
750 EXPORT_SYMBOL(server_name_is_ost);
753 * Get the index from the target name MDTXXXX/OSTXXXX
754 * rc = server type, or rc < 0 on error
756 int target_name2index(const char *tgtname, __u32 *idx, const char **endptr)
758 const char *dash = tgtname;
762 if (strncmp(dash, "MDT", 3) == 0)
763 rc = LDD_F_SV_TYPE_MDT;
764 else if (strncmp(dash, "OST", 3) == 0)
765 rc = LDD_F_SV_TYPE_OST;
771 if (strncmp(dash, "all", 3) == 0) {
774 return rc | LDD_F_SV_ALL;
777 index = simple_strtoul(dash, (char **)endptr, 16);
782 EXPORT_SYMBOL(target_name2index);
784 /* Get the index from the obd name.
787 if endptr isn't NULL it is set to end of name */
788 int server_name2index(const char *svname, __u32 *idx, const char **endptr)
793 /* We use server_name2fsname() just for parsing */
794 rc = server_name2fsname(svname, NULL, &dash);
799 rc = target_name2index(dash, idx, endptr);
803 /* Account for -mdc after index that is possible when specifying mdt */
804 if (endptr != NULL && strncmp(LUSTRE_MDC_NAME, *endptr + 1,
805 sizeof(LUSTRE_MDC_NAME)-1) == 0)
806 *endptr += sizeof(LUSTRE_MDC_NAME);
810 EXPORT_SYMBOL(server_name2index);
812 /*************** mount common betweeen server and client ***************/
815 int lustre_common_put_super(struct super_block *sb)
820 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
822 /* Drop a ref to the MGC */
823 rc = lustre_stop_mgc(sb);
824 if (rc && (rc != -ENOENT)) {
826 CERROR("Can't stop MGC: %d\n", rc);
829 /* BUSY just means that there's some other obd that
830 needs the mgc. Let him clean it up. */
831 CDEBUG(D_MOUNT, "MGC still in use\n");
833 /* Drop a ref to the mounted disk */
838 EXPORT_SYMBOL(lustre_common_put_super);
840 static void lmd_print(struct lustre_mount_data *lmd)
844 PRINT_CMD(D_MOUNT, " mount data:\n");
845 if (lmd_is_client(lmd))
846 PRINT_CMD(D_MOUNT, "profile: %s\n", lmd->lmd_profile);
847 PRINT_CMD(D_MOUNT, "device: %s\n", lmd->lmd_dev);
848 PRINT_CMD(D_MOUNT, "flags: %x\n", lmd->lmd_flags);
851 PRINT_CMD(D_MOUNT, "options: %s\n", lmd->lmd_opts);
853 if (lmd->lmd_recovery_time_soft)
854 PRINT_CMD(D_MOUNT, "recovery time soft: %d\n",
855 lmd->lmd_recovery_time_soft);
857 if (lmd->lmd_recovery_time_hard)
858 PRINT_CMD(D_MOUNT, "recovery time hard: %d\n",
859 lmd->lmd_recovery_time_hard);
861 for (i = 0; i < lmd->lmd_exclude_count; i++) {
862 PRINT_CMD(D_MOUNT, "exclude %d: OST%04x\n", i,
863 lmd->lmd_exclude[i]);
867 /* Is this server on the exclusion list */
868 int lustre_check_exclusion(struct super_block *sb, char *svname)
870 struct lustre_sb_info *lsi = s2lsi(sb);
871 struct lustre_mount_data *lmd = lsi->lsi_lmd;
876 rc = server_name2index(svname, &index, NULL);
877 if (rc != LDD_F_SV_TYPE_OST)
878 /* Only exclude OSTs */
881 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
882 index, lmd->lmd_exclude_count, lmd->lmd_dev);
884 for(i = 0; i < lmd->lmd_exclude_count; i++) {
885 if (index == lmd->lmd_exclude[i]) {
886 CWARN("Excluding %s (on exclusion list)\n", svname);
893 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
894 static int lmd_make_exclusion(struct lustre_mount_data *lmd, const char *ptr)
896 const char *s1 = ptr, *s2;
897 __u32 index, *exclude_list;
901 /* The shortest an ost name can be is 8 chars: -OST0000.
902 We don't actually know the fsname at this time, so in fact
903 a user could specify any fsname. */
904 devmax = strlen(ptr) / 8 + 1;
906 /* temp storage until we figure out how many we have */
907 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
911 /* we enter this fn pointing at the '=' */
912 while (*s1 && *s1 != ' ' && *s1 != ',') {
914 rc = server_name2index(s1, &index, &s2);
916 CERROR("Can't parse server name '%s': rc = %d\n",
920 if (rc == LDD_F_SV_TYPE_OST)
921 exclude_list[lmd->lmd_exclude_count++] = index;
923 CDEBUG(D_MOUNT, "ignoring exclude %.*s: type = %#x\n",
924 (uint)(s2-s1), s1, rc);
926 /* now we are pointing at ':' (next exclude)
927 or ',' (end of excludes) */
928 if (lmd->lmd_exclude_count >= devmax)
931 if (rc >= 0) /* non-err */
934 if (lmd->lmd_exclude_count) {
935 /* permanent, freed in lustre_free_lsi */
936 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
937 lmd->lmd_exclude_count);
938 if (lmd->lmd_exclude) {
939 memcpy(lmd->lmd_exclude, exclude_list,
940 sizeof(index) * lmd->lmd_exclude_count);
943 lmd->lmd_exclude_count = 0;
946 OBD_FREE(exclude_list, sizeof(index) * devmax);
950 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
955 if (lmd->lmd_mgssec != NULL) {
956 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
957 lmd->lmd_mgssec = NULL;
960 tail = strchr(ptr, ',');
962 length = strlen(ptr);
966 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
967 if (lmd->lmd_mgssec == NULL)
970 memcpy(lmd->lmd_mgssec, ptr, length);
971 lmd->lmd_mgssec[length] = '\0';
975 static int lmd_parse_string(char **handle, char *ptr)
980 if ((handle == NULL) || (ptr == NULL))
983 if (*handle != NULL) {
984 OBD_FREE(*handle, strlen(*handle) + 1);
988 tail = strchr(ptr, ',');
990 length = strlen(ptr);
994 OBD_ALLOC(*handle, length + 1);
998 memcpy(*handle, ptr, length);
999 (*handle)[length] = '\0';
1004 /* Collect multiple values for mgsnid specifiers */
1005 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
1013 /* Find end of nidlist */
1014 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
1015 length = tail - *ptr;
1017 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
1021 if (lmd->lmd_mgs != NULL)
1022 oldlen = strlen(lmd->lmd_mgs) + 1;
1024 OBD_ALLOC(mgsnid, oldlen + length + 1);
1028 if (lmd->lmd_mgs != NULL) {
1029 /* Multiple mgsnid= are taken to mean failover locations */
1030 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
1031 mgsnid[oldlen - 1] = ':';
1032 OBD_FREE(lmd->lmd_mgs, oldlen);
1034 memcpy(mgsnid + oldlen, *ptr, length);
1035 mgsnid[oldlen + length] = '\0';
1036 lmd->lmd_mgs = mgsnid;
1042 /** Parse mount line options
1043 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
1044 * dev is passed as device=uml1:/lustre by mount.lustre
1046 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1048 char *s1, *s2, *devname = NULL;
1049 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1055 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1056 "/sbin/mount.lustre is installed.\n");
1060 /* Options should be a string - try to detect old lmd data */
1061 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1062 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1063 "/sbin/mount.lustre. Please install "
1064 "version %s\n", LUSTRE_VERSION_STRING);
1067 lmd->lmd_magic = LMD_MAGIC;
1069 OBD_ALLOC(lmd->lmd_params, 4096);
1070 if (lmd->lmd_params == NULL)
1072 lmd->lmd_params[0] = '\0';
1074 /* Set default flags here */
1079 int time_min = OBD_RECOVERY_TIME_MIN;
1081 /* Skip whitespace and extra commas */
1082 while (*s1 == ' ' || *s1 == ',')
1085 /* Client options are parsed in ll_options: eg. flock,
1088 /* Parse non-ldiskfs options here. Rather than modifying
1089 ldiskfs, we just zero these out here */
1090 if (strncmp(s1, "abort_recov", 11) == 0) {
1091 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1093 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
1094 lmd->lmd_recovery_time_soft = max_t(int,
1095 simple_strtoul(s1 + 19, NULL, 10), time_min);
1097 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
1098 lmd->lmd_recovery_time_hard = max_t(int,
1099 simple_strtoul(s1 + 19, NULL, 10), time_min);
1101 } else if (strncmp(s1, "noir", 4) == 0) {
1102 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
1104 } else if (strncmp(s1, "nosvc", 5) == 0) {
1105 lmd->lmd_flags |= LMD_FLG_NOSVC;
1107 } else if (strncmp(s1, "nomgs", 5) == 0) {
1108 lmd->lmd_flags |= LMD_FLG_NOMGS;
1110 } else if (strncmp(s1, "noscrub", 7) == 0) {
1111 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
1113 } else if (strncmp(s1, PARAM_MGSNODE,
1114 sizeof(PARAM_MGSNODE) - 1) == 0) {
1115 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
1116 /* Assume the next mount opt is the first
1117 invalid nid we get to. */
1118 rc = lmd_parse_mgs(lmd, &s2);
1122 } else if (strncmp(s1, "writeconf", 9) == 0) {
1123 lmd->lmd_flags |= LMD_FLG_WRITECONF;
1125 } else if (strncmp(s1, "update", 6) == 0) {
1126 lmd->lmd_flags |= LMD_FLG_UPDATE;
1128 } else if (strncmp(s1, "virgin", 6) == 0) {
1129 lmd->lmd_flags |= LMD_FLG_VIRGIN;
1131 } else if (strncmp(s1, "noprimnode", 10) == 0) {
1132 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
1134 } else if (strncmp(s1, "mgssec=", 7) == 0) {
1135 rc = lmd_parse_mgssec(lmd, s1 + 7);
1139 /* ost exclusion list */
1140 } else if (strncmp(s1, "exclude=", 8) == 0) {
1141 rc = lmd_make_exclusion(lmd, s1 + 7);
1145 } else if (strncmp(s1, "mgs", 3) == 0) {
1147 lmd->lmd_flags |= LMD_FLG_MGS;
1149 } else if (strncmp(s1, "svname=", 7) == 0) {
1150 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
1154 } else if (strncmp(s1, "param=", 6) == 0) {
1156 char *tail = strchr(s1 + 6, ',');
1158 length = strlen(s1);
1162 strncat(lmd->lmd_params, s1 + 6, length);
1163 strcat(lmd->lmd_params, " ");
1165 } else if (strncmp(s1, "osd=", 4) == 0) {
1166 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
1171 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1172 end of the options. */
1173 else if (strncmp(s1, "device=", 7) == 0) {
1175 /* terminate options right before device. device
1176 must be the last one. */
1182 s2 = strchr(s1, ',');
1190 memmove(s1, s2, strlen(s2) + 1);
1196 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1197 "(need mount option 'device=...')\n");
1201 s1 = strstr(devname, ":/");
1204 lmd->lmd_flags |= LMD_FLG_CLIENT;
1205 /* Remove leading /s from fsname */
1206 while (*++s1 == '/') ;
1207 /* Freed in lustre_free_lsi */
1208 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1209 if (!lmd->lmd_profile)
1211 sprintf(lmd->lmd_profile, "%s-client", s1);
1214 /* Freed in lustre_free_lsi */
1215 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1218 strcpy(lmd->lmd_dev, devname);
1220 /* Save mount options */
1221 s1 = options + strlen(options) - 1;
1222 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1224 if (*options != 0) {
1225 /* Freed in lustre_free_lsi */
1226 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1229 strcpy(lmd->lmd_opts, options);
1233 lmd->lmd_magic = LMD_MAGIC;
1238 CERROR("Bad mount options %s\n", options);
1242 struct lustre_mount_data2 {
1244 struct vfsmount *lmd2_mnt;
1247 /** This is the entry point for the mount call into Lustre.
1248 * This is called when a server or client is mounted,
1249 * and this is where we start setting things up.
1250 * @param data Mount options (e.g. -o flock,abort_recov)
1252 int lustre_fill_super(struct super_block *sb, void *data, int silent)
1254 struct lustre_mount_data *lmd;
1255 struct lustre_mount_data2 *lmd2 = data;
1256 struct lustre_sb_info *lsi;
1260 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1262 lsi = lustre_init_lsi(sb);
1268 * Disable lockdep during mount, because mount locking patterns are
1274 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
1276 obd_zombie_barrier();
1278 /* Figure out the lmd from the mount options */
1279 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
1281 GOTO(out, rc = -EINVAL);
1284 if (lmd_is_client(lmd)) {
1285 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1286 if (!client_fill_super) {
1287 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1288 "client mount! Is the 'lustre' "
1289 "module loaded?\n");
1293 rc = lustre_start_mgc(sb);
1298 /* Connect and start */
1299 /* (should always be ll_fill_super) */
1300 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
1301 /* c_f_s will call lustre_common_put_super on failure */
1304 #ifdef HAVE_SERVER_SUPPORT
1305 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1306 rc = server_fill_super(sb);
1307 /* s_f_s calls lustre_start_mgc after the mount because we need
1308 the MGS nids which are stored on disk. Plus, we may
1309 need to start the MGS first. */
1310 /* s_f_s will call server_put_super on failure */
1312 CERROR("This is client-side-only module, "
1313 "cannot handle server mount.\n");
1318 /* If error happens in fill_super() call, @lsi will be killed there.
1319 * This is why we do not put it here. */
1323 CERROR("Unable to mount %s (%d)\n",
1324 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1326 CDEBUG(D_SUPER, "Mount %s complete\n",
1334 /* We can't call ll_fill_super by name because it lives in a module that
1335 must be loaded after this one. */
1336 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
1337 struct vfsmount *mnt))
1339 client_fill_super = cfs;
1341 EXPORT_SYMBOL(lustre_register_client_fill_super);
1343 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
1345 kill_super_cb = cfs;
1347 EXPORT_SYMBOL(lustre_register_kill_super_cb);
1349 /***************** FS registration ******************/
1350 #ifdef HAVE_FSTYPE_MOUNT
1351 struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
1352 const char *devname, void *data)
1354 struct lustre_mount_data2 lmd2 = { data, NULL };
1356 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
1359 int lustre_get_sb(struct file_system_type *fs_type, int flags,
1360 const char *devname, void * data, struct vfsmount *mnt)
1362 struct lustre_mount_data2 lmd2 = { data, mnt };
1364 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
1368 void lustre_kill_super(struct super_block *sb)
1370 struct lustre_sb_info *lsi = s2lsi(sb);
1372 if (kill_super_cb && lsi && !IS_SERVER(lsi))
1373 (*kill_super_cb)(sb);
1375 kill_anon_super(sb);
1378 /** Register the "lustre" fs type
1380 struct file_system_type lustre_fs_type = {
1381 .owner = THIS_MODULE,
1383 #ifdef HAVE_FSTYPE_MOUNT
1384 .mount = lustre_mount,
1386 .get_sb = lustre_get_sb,
1388 .kill_sb = lustre_kill_super,
1389 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
1390 FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE,
1393 int lustre_register_fs(void)
1395 return register_filesystem(&lustre_fs_type);
1398 int lustre_unregister_fs(void)
1400 return unregister_filesystem(&lustre_fs_type);