4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
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/server 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
47 #define PRINT_MASK D_SUPER|D_CONFIG
51 #include <lustre_fsfilt.h>
52 #include <obd_class.h>
53 #include <lustre/lustre_user.h>
54 #include <linux/version.h>
55 #include <lustre_log.h>
56 #include <lustre_disk.h>
57 #include <lustre_param.h>
59 static int (*client_fill_super)(struct super_block *sb,
60 struct vfsmount *mnt) = NULL;
61 static void (*kill_super_cb)(struct super_block *sb) = NULL;
63 /*********** mount lookup *********/
65 CFS_DEFINE_MUTEX(lustre_mount_info_lock);
66 static CFS_LIST_HEAD(server_mount_info_list);
68 static struct lustre_mount_info *server_find_mount(const char *name)
71 struct lustre_mount_info *lmi;
74 cfs_list_for_each(tmp, &server_mount_info_list) {
75 lmi = cfs_list_entry(tmp, struct lustre_mount_info,
77 if (strcmp(name, lmi->lmi_name) == 0)
83 /* we must register an obd for a mount before we call the setup routine.
84 *_setup will call lustre_get_mount to get the mnt struct
85 by obd_name, since we can't pass the pointer to setup. */
86 static int server_register_mount(const char *name, struct super_block *sb,
89 struct lustre_mount_info *lmi;
96 OBD_ALLOC(lmi, sizeof(*lmi));
99 OBD_ALLOC(name_cp, strlen(name) + 1);
101 OBD_FREE(lmi, sizeof(*lmi));
104 strcpy(name_cp, name);
106 cfs_mutex_lock(&lustre_mount_info_lock);
108 if (server_find_mount(name)) {
109 cfs_mutex_unlock(&lustre_mount_info_lock);
110 OBD_FREE(lmi, sizeof(*lmi));
111 OBD_FREE(name_cp, strlen(name) + 1);
112 CERROR("Already registered %s\n", name);
115 lmi->lmi_name = name_cp;
118 cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list);
120 cfs_mutex_unlock(&lustre_mount_info_lock);
122 CDEBUG(D_MOUNT, "reg_mnt %p from %s, vfscount=%d\n",
123 lmi->lmi_mnt, name, mnt_get_count(lmi->lmi_mnt));
128 /* when an obd no longer needs a mount */
129 static int server_deregister_mount(const char *name)
131 struct lustre_mount_info *lmi;
134 cfs_mutex_lock(&lustre_mount_info_lock);
135 lmi = server_find_mount(name);
137 cfs_mutex_unlock(&lustre_mount_info_lock);
138 CERROR("%s not registered\n", name);
142 CDEBUG(D_MOUNT, "dereg_mnt %p from %s, vfscount=%d\n",
143 lmi->lmi_mnt, name, mnt_get_count(lmi->lmi_mnt));
145 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
146 cfs_list_del(&lmi->lmi_list_chain);
147 OBD_FREE(lmi, sizeof(*lmi));
148 cfs_mutex_unlock(&lustre_mount_info_lock);
153 /* obd's look up a registered mount using their obdname. This is just
154 for initial obd setup to find the mount struct. It should not be
155 called every time you want to mntget. */
156 struct lustre_mount_info *server_get_mount(const char *name)
158 struct lustre_mount_info *lmi;
159 struct lustre_sb_info *lsi;
162 cfs_mutex_lock(&lustre_mount_info_lock);
163 lmi = server_find_mount(name);
164 cfs_mutex_unlock(&lustre_mount_info_lock);
166 CERROR("Can't find mount for %s\n", name);
169 lsi = s2lsi(lmi->lmi_sb);
170 mntget(lmi->lmi_mnt);
171 cfs_atomic_inc(&lsi->lsi_mounts);
173 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d, vfscount=%d\n",
174 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts),
175 mnt_get_count(lmi->lmi_mnt));
181 * Used by mdt to get mount_info from obdname.
182 * There are no blocking when using the mount_info.
183 * Do not use server_get_mount for this purpose.
185 struct lustre_mount_info *server_get_mount_2(const char *name)
187 struct lustre_mount_info *lmi;
190 cfs_mutex_lock(&lustre_mount_info_lock);
191 lmi = server_find_mount(name);
192 cfs_mutex_unlock(&lustre_mount_info_lock);
194 CERROR("Can't find mount for %s\n", name);
199 static void unlock_mntput(struct vfsmount *mnt)
201 #if HAVE_KERNEL_LOCKED
202 /* for kernel < 2.6.37 */
203 if (kernel_locked()) {
213 static int lustre_put_lsi(struct super_block *sb);
215 /* to be called from obd_cleanup methods */
216 int server_put_mount(const char *name, struct vfsmount *mnt)
218 struct lustre_mount_info *lmi;
219 struct lustre_sb_info *lsi;
220 int count = mnt_get_count(mnt) - 1;
223 /* This might be the last one, can't deref after this */
226 cfs_mutex_lock(&lustre_mount_info_lock);
227 lmi = server_find_mount(name);
228 cfs_mutex_unlock(&lustre_mount_info_lock);
230 CERROR("Can't find mount for %s\n", name);
233 lsi = s2lsi(lmi->lmi_sb);
234 LASSERT(lmi->lmi_mnt == mnt);
236 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d, vfscount=%d\n",
237 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts), count);
239 if (lustre_put_lsi(lmi->lmi_sb)) {
240 CDEBUG(D_MOUNT, "Last put of mnt %p from %s, vfscount=%d\n",
241 lmi->lmi_mnt, name, count);
242 /* last mount is the One True Mount */
244 CERROR("%s: mount busy, vfscount=%d!\n", name, count);
247 /* this obd should never need the mount again */
248 server_deregister_mount(name);
253 /* Corresponding to server_get_mount_2 */
254 int server_put_mount_2(const char *name, struct vfsmount *mnt)
260 /******* mount helper utilities *********/
263 static void ldd_print(struct lustre_disk_data *ldd)
265 PRINT_CMD(PRINT_MASK, " disk data:\n");
266 PRINT_CMD(PRINT_MASK, "server: %s\n", ldd->ldd_svname);
267 PRINT_CMD(PRINT_MASK, "uuid: %s\n", (char *)ldd->ldd_uuid);
268 PRINT_CMD(PRINT_MASK, "fs: %s\n", ldd->ldd_fsname);
269 PRINT_CMD(PRINT_MASK, "index: %04x\n", ldd->ldd_svindex);
270 PRINT_CMD(PRINT_MASK, "config: %d\n", ldd->ldd_config_ver);
271 PRINT_CMD(PRINT_MASK, "flags: %#x\n", ldd->ldd_flags);
272 PRINT_CMD(PRINT_MASK, "diskfs: %s\n", MT_STR(ldd));
273 PRINT_CMD(PRINT_MASK, "options: %s\n", ldd->ldd_mount_opts);
274 PRINT_CMD(PRINT_MASK, "params: %s\n", ldd->ldd_params);
275 PRINT_CMD(PRINT_MASK, "comment: %s\n", ldd->ldd_userdata);
279 static int ldd_parse(struct lvfs_run_ctxt *mount_ctxt,
280 struct lustre_disk_data *ldd)
282 struct lvfs_run_ctxt saved;
289 push_ctxt(&saved, mount_ctxt, NULL);
291 file = filp_open(MOUNT_DATA_FILE, O_RDONLY, 0644);
294 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
298 len = i_size_read(file->f_dentry->d_inode);
299 CDEBUG(D_MOUNT, "Have %s, size %lu\n", MOUNT_DATA_FILE, len);
300 if (len != sizeof(*ldd)) {
301 CERROR("disk data size does not match: see %lu expect %u\n",
302 len, (int)sizeof(*ldd));
303 GOTO(out_close, rc = -EINVAL);
306 rc = lustre_fread(file, ldd, len, &off);
308 CERROR("error reading %s: read %d of %lu\n",
309 MOUNT_DATA_FILE, rc, len);
310 GOTO(out_close, rc = -EINVAL);
314 if (ldd->ldd_magic != LDD_MAGIC) {
315 /* FIXME add swabbing support */
316 CERROR("Bad magic in %s: %x!=%x\n", MOUNT_DATA_FILE,
317 ldd->ldd_magic, LDD_MAGIC);
318 GOTO(out_close, rc = -EINVAL);
321 if (ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP) {
322 CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
324 ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP);
325 GOTO(out_close, rc = -EINVAL);
327 if (ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP) {
328 CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
330 ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP);
331 /* Do something like remount filesystem read-only */
332 GOTO(out_close, rc = -EINVAL);
338 pop_ctxt(&saved, mount_ctxt, NULL);
342 static int ldd_write(struct lvfs_run_ctxt *mount_ctxt,
343 struct lustre_disk_data *ldd)
345 struct lvfs_run_ctxt saved;
348 unsigned long len = sizeof(struct lustre_disk_data);
352 LASSERT(ldd->ldd_magic == LDD_MAGIC);
354 ldd->ldd_config_ver++;
356 push_ctxt(&saved, mount_ctxt, NULL);
358 file = filp_open(MOUNT_DATA_FILE, O_RDWR|O_SYNC, 0644);
361 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
365 rc = lustre_fwrite(file, ldd, len, &off);
367 CERROR("error writing %s: read %d of %lu\n",
368 MOUNT_DATA_FILE, rc, len);
369 GOTO(out_close, rc = -EINVAL);
377 pop_ctxt(&saved, mount_ctxt, NULL);
382 /**************** config llog ********************/
384 /** Get a config log from the MGS and process it.
385 * This func is called for both clients and servers.
386 * Continue to process new statements appended to the logs
387 * (whenever the config lock is revoked) until lustre_end_log
389 * @param sb The superblock is used by the MGC to write to the local copy of
391 * @param logname The name of the llog to replicate from the MGS
392 * @param cfg Since the same mgc may be used to follow multiple config logs
393 * (e.g. ost1, ost2, client), the config_llog_instance keeps the state for
394 * this log, and is added to the mgc's list of logs to follow.
396 int lustre_process_log(struct super_block *sb, char *logname,
397 struct config_llog_instance *cfg)
399 struct lustre_cfg *lcfg;
400 struct lustre_cfg_bufs *bufs;
401 struct lustre_sb_info *lsi = s2lsi(sb);
402 struct obd_device *mgc = lsi->lsi_mgc;
413 /* mgc_process_config */
414 lustre_cfg_bufs_reset(bufs, mgc->obd_name);
415 lustre_cfg_bufs_set_string(bufs, 1, logname);
416 lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg));
417 lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb));
418 lcfg = lustre_cfg_new(LCFG_LOG_START, bufs);
419 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
420 lustre_cfg_free(lcfg);
425 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
426 "failed from the MGS (%d). Make sure this "
427 "client and the MGS are running compatible "
428 "versions of Lustre.\n",
429 mgc->obd_name, logname, rc);
432 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
433 "failed (%d). This may be the result of "
434 "communication errors between this node and "
435 "the MGS, a bad configuration, or other "
436 "errors. See the syslog for more "
437 "information.\n", mgc->obd_name, logname,
440 /* class_obd_list(); */
444 /* Stop watching this config log for updates */
445 int lustre_end_log(struct super_block *sb, char *logname,
446 struct config_llog_instance *cfg)
448 struct lustre_cfg *lcfg;
449 struct lustre_cfg_bufs bufs;
450 struct lustre_sb_info *lsi = s2lsi(sb);
451 struct obd_device *mgc = lsi->lsi_mgc;
458 /* mgc_process_config */
459 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
460 lustre_cfg_bufs_set_string(&bufs, 1, logname);
462 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
463 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
464 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
465 lustre_cfg_free(lcfg);
469 /**************** obd start *******************/
471 /** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from
472 * lctl (and do for echo cli/srv.
474 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
475 char *s1, char *s2, char *s3, char *s4)
477 struct lustre_cfg_bufs bufs;
478 struct lustre_cfg * lcfg = NULL;
481 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
482 cmd, s1, s2, s3, s4);
484 lustre_cfg_bufs_reset(&bufs, cfgname);
486 lustre_cfg_bufs_set_string(&bufs, 1, s1);
488 lustre_cfg_bufs_set_string(&bufs, 2, s2);
490 lustre_cfg_bufs_set_string(&bufs, 3, s3);
492 lustre_cfg_bufs_set_string(&bufs, 4, s4);
494 lcfg = lustre_cfg_new(cmd, &bufs);
495 lcfg->lcfg_nid = nid;
496 rc = class_process_config(lcfg);
497 lustre_cfg_free(lcfg);
501 /** Call class_attach and class_setup. These methods in turn call
502 * obd type-specific methods.
504 static int lustre_start_simple(char *obdname, char *type, char *uuid,
508 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
510 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
512 CERROR("%s attach error %d\n", obdname, rc);
515 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, 0, 0);
517 CERROR("%s setup error %d\n", obdname, rc);
518 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
523 /* Set up a MGS to serve startup logs */
524 static int server_start_mgs(struct super_block *sb)
526 struct lustre_sb_info *lsi = s2lsi(sb);
527 struct vfsmount *mnt = lsi->lsi_srv_mnt;
528 struct lustre_mount_info *lmi;
533 /* It is impossible to have more than 1 MGS per node, since
534 MGC wouldn't know which to connect to */
535 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
537 lsi = s2lsi(lmi->lmi_sb);
538 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
540 lsi->lsi_ldd->ldd_svname);
544 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
546 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
549 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
550 LUSTRE_MGS_OBDNAME, 0, 0);
551 /* Do NOT call server_deregister_mount() here. This leads to
552 * inability cleanup cleanly and free lsi and other stuff when
553 * mgs calls server_put_mount() in error handling case. -umka */
557 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
558 "Is the 'mgs' module loaded?\n",
559 LUSTRE_MGS_OBDNAME, rc);
563 static int server_stop_mgs(struct super_block *sb)
565 struct obd_device *obd;
569 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
571 /* There better be only one MGS */
572 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
574 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
578 /* The MGS should always stop when we say so */
580 rc = class_manual_cleanup(obd);
584 CFS_DEFINE_MUTEX(mgc_start_lock);
586 /** Set up a mgc obd to process startup logs
588 * \param sb [in] super block of the mgc obd
590 * \retval 0 success, otherwise error code
592 static int lustre_start_mgc(struct super_block *sb)
594 struct obd_connect_data *data = NULL;
595 struct lustre_sb_info *lsi = s2lsi(sb);
596 struct obd_device *obd;
597 struct obd_export *exp;
598 struct obd_uuid *uuid;
601 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
604 int rc = 0, i = 0, j, len;
607 LASSERT(lsi->lsi_lmd);
609 /* Find the first non-lo MGS nid for our MGC name */
610 if (lsi->lsi_flags & LSI_SERVER) {
611 ptr = lsi->lsi_ldd->ldd_params;
612 /* Use mgsnode= nids */
613 if ((class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0) &&
614 (class_parse_nid(ptr, &nid, &ptr) == 0)) {
616 } else if (IS_MGS(lsi->lsi_ldd)) {
617 lnet_process_id_t id;
618 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
619 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
626 } else { /* client */
627 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
628 ptr = lsi->lsi_lmd->lmd_dev;
629 if (class_parse_nid(ptr, &nid, &ptr) == 0)
633 CERROR("No valid MGS nids found.\n");
637 cfs_mutex_lock(&mgc_start_lock);
639 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
640 OBD_ALLOC(mgcname, len);
641 OBD_ALLOC(niduuid, len + 2);
642 if (!mgcname || !niduuid)
643 GOTO(out_free, rc = -ENOMEM);
644 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
646 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
650 GOTO(out_free, rc = -ENOMEM);
652 obd = class_name2obd(mgcname);
653 if (obd && !obd->obd_stopping) {
654 rc = obd_set_info_async(NULL, obd->obd_self_export,
655 strlen(KEY_MGSSEC), KEY_MGSSEC,
656 strlen(mgssec), mgssec, NULL);
660 /* Re-using an existing MGC */
661 cfs_atomic_inc(&obd->u.cli.cl_mgc_refcount);
663 /* IR compatibility check, only for clients */
664 if (lmd_is_client(lsi->lsi_lmd)) {
666 int vallen = sizeof(*data);
667 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
669 rc = obd_get_info(NULL, obd->obd_self_export,
670 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
671 &vallen, data, NULL);
673 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
674 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
675 /* LMD_FLG_NOIR is for test purpose only */
677 "Trying to mount a client with IR setting "
678 "not compatible with current mgc. "
679 "Force to use current mgc setting that is "
681 has_ir ? "enabled" : "disabled");
683 *flags &= ~LMD_FLG_NOIR;
685 *flags |= LMD_FLG_NOIR;
690 /* If we are restarting the MGS, don't try to keep the MGC's
691 old connection, or registration will fail. */
692 if ((lsi->lsi_flags & LSI_SERVER) && IS_MGS(lsi->lsi_ldd)) {
693 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
697 /* Try all connections, but only once (again).
698 We don't want to block another target from starting
699 (using its local copy of the log), but we do want to connect
700 if at all possible. */
702 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
703 rc = obd_set_info_async(NULL, obd->obd_self_export,
704 sizeof(KEY_INIT_RECOV_BACKUP),
705 KEY_INIT_RECOV_BACKUP,
706 sizeof(recov_bk), &recov_bk, NULL);
710 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
712 /* Add the primary nids for the MGS */
714 sprintf(niduuid, "%s_%x", mgcname, i);
715 if (lsi->lsi_flags & LSI_SERVER) {
716 ptr = lsi->lsi_ldd->ldd_params;
717 if (IS_MGS(lsi->lsi_ldd)) {
718 /* Use local nids (including LO) */
719 lnet_process_id_t id;
720 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
721 rc = do_lcfg(mgcname, id.nid,
722 LCFG_ADD_UUID, niduuid, 0,0,0);
725 /* Use mgsnode= nids */
726 if (class_find_param(ptr, PARAM_MGSNODE, &ptr) != 0) {
727 CERROR("No MGS nids given.\n");
728 GOTO(out_free, rc = -EINVAL);
730 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
731 rc = do_lcfg(mgcname, nid,
732 LCFG_ADD_UUID, niduuid, 0,0,0);
736 } else { /* client */
737 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
738 ptr = lsi->lsi_lmd->lmd_dev;
739 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
740 rc = do_lcfg(mgcname, nid,
741 LCFG_ADD_UUID, niduuid, 0,0,0);
743 /* Stop at the first failover nid */
749 CERROR("No valid MGS nids found.\n");
750 GOTO(out_free, rc = -EINVAL);
752 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
754 /* Random uuid for MGC allows easier reconnects */
756 ll_generate_random_uuid(uuidc);
757 class_uuid_unparse(uuidc, uuid);
760 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
761 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
767 /* Add any failover MGS nids */
769 while ((*ptr == ':' ||
770 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0)) {
771 /* New failover node */
772 sprintf(niduuid, "%s_%x", mgcname, i);
774 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
776 rc = do_lcfg(mgcname, nid,
777 LCFG_ADD_UUID, niduuid, 0,0,0);
782 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
790 lsi->lsi_lmd->lmd_mgs_failnodes = i;
792 obd = class_name2obd(mgcname);
794 CERROR("Can't find mgcobd %s\n", mgcname);
795 GOTO(out_free, rc = -ENOTCONN);
798 rc = obd_set_info_async(NULL, obd->obd_self_export,
799 strlen(KEY_MGSSEC), KEY_MGSSEC,
800 strlen(mgssec), mgssec, NULL);
804 /* Keep a refcount of servers/clients who started with "mount",
805 so we know when we can get rid of the mgc. */
806 cfs_atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
808 /* Try all connections, but only once. */
810 rc = obd_set_info_async(NULL, obd->obd_self_export,
811 sizeof(KEY_INIT_RECOV_BACKUP),
812 KEY_INIT_RECOV_BACKUP,
813 sizeof(recov_bk), &recov_bk, NULL);
816 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
817 /* We connect to the MGS at setup, and don't disconnect until cleanup */
818 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID |
819 OBD_CONNECT_AT | OBD_CONNECT_FULL20 |
820 OBD_CONNECT_IMP_RECOV;
821 if (lmd_is_client(lsi->lsi_lmd) &&
822 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
823 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
824 data->ocd_version = LUSTRE_VERSION_CODE;
825 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
827 CERROR("connect failed %d\n", rc);
831 obd->u.cli.cl_mgc_mgsexp = exp;
834 /* Keep the mgc info in the sb. Note that many lsi's can point
838 cfs_mutex_unlock(&mgc_start_lock);
843 OBD_FREE(mgcname, len);
845 OBD_FREE(niduuid, len + 2);
849 static int lustre_stop_mgc(struct super_block *sb)
851 struct lustre_sb_info *lsi = s2lsi(sb);
852 struct obd_device *obd;
853 char *niduuid = 0, *ptr = 0;
854 int i, rc = 0, len = 0;
864 cfs_mutex_lock(&mgc_start_lock);
865 LASSERT(cfs_atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
866 if (!cfs_atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
867 /* This is not fatal, every client that stops
868 will call in here. */
869 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
870 cfs_atomic_read(&obd->u.cli.cl_mgc_refcount));
871 GOTO(out, rc = -EBUSY);
874 /* The MGC has no recoverable data in any case.
875 * force shotdown set in umount_begin */
876 obd->obd_no_recov = 1;
878 if (obd->u.cli.cl_mgc_mgsexp) {
879 /* An error is not fatal, if we are unable to send the
880 disconnect mgs ping evictor cleans up the export */
881 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
883 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
886 /* Save the obdname for cleaning the nid uuids, which are
888 len = strlen(obd->obd_name) + 6;
889 OBD_ALLOC(niduuid, len);
891 strcpy(niduuid, obd->obd_name);
892 ptr = niduuid + strlen(niduuid);
895 rc = class_manual_cleanup(obd);
899 /* Clean the nid uuids */
901 GOTO(out, rc = -ENOMEM);
903 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
904 sprintf(ptr, "_%x", i);
905 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
908 CERROR("del MDC UUID %s failed: rc = %d\n",
913 OBD_FREE(niduuid, len);
915 /* class_import_put will get rid of the additional connections */
916 cfs_mutex_unlock(&mgc_start_lock);
920 /* Since there's only one mgc per node, we have to change it's fs to get
921 access to the right disk. */
922 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
924 struct lustre_sb_info *lsi = s2lsi(sb);
928 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
930 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
931 rc = obd_set_info_async(NULL, mgc->obd_self_export,
932 sizeof(KEY_SET_FS), KEY_SET_FS,
933 sizeof(*sb), sb, NULL);
935 CERROR("can't set_fs %d\n", rc);
941 static int server_mgc_clear_fs(struct obd_device *mgc)
946 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
948 rc = obd_set_info_async(NULL, mgc->obd_self_export,
949 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
954 CFS_DEFINE_MUTEX(server_start_lock);
956 /* Stop MDS/OSS if nobody is using them */
957 static int server_stop_servers(int lddflags, int lsiflags)
959 struct obd_device *obd = NULL;
960 struct obd_type *type = NULL;
964 cfs_mutex_lock(&server_start_lock);
966 /* Either an MDT or an OST or neither */
967 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
968 if ((lddflags & LDD_F_SV_TYPE_MDT) &&
969 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
970 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
971 type = class_search_type(LUSTRE_MDS_NAME);
973 /* if this was an OST, and there are no more OST's, clean up the OSS */
974 if ((lddflags & LDD_F_SV_TYPE_OST) &&
975 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
976 type = class_search_type(LUSTRE_OST_NAME);
979 if (obd && (!type || !type->typ_refcnt)) {
982 /* obd_fail doesn't mean much on a server obd */
983 err = class_manual_cleanup(obd);
988 cfs_mutex_unlock(&server_start_lock);
993 int server_mti_print(char *title, struct mgs_target_info *mti)
995 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
996 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
997 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
998 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
999 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
1000 mti->mti_config_ver, mti->mti_flags);
1004 static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
1006 struct lustre_sb_info *lsi = s2lsi(sb);
1007 struct lustre_disk_data *ldd = lsi->lsi_ldd;
1008 lnet_process_id_t id;
1012 if (!(lsi->lsi_flags & LSI_SERVER))
1015 strncpy(mti->mti_fsname, ldd->ldd_fsname,
1016 sizeof(mti->mti_fsname));
1017 strncpy(mti->mti_svname, ldd->ldd_svname,
1018 sizeof(mti->mti_svname));
1020 mti->mti_nid_count = 0;
1021 while (LNetGetId(i++, &id) != -ENOENT) {
1022 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1025 /* server use --servicenode param, only allow specified
1026 * nids be registered */
1027 if ((ldd->ldd_flags & LDD_F_NO_PRIMNODE) != 0 &&
1028 class_match_nid(ldd->ldd_params,
1029 PARAM_FAILNODE, id.nid) < 1)
1032 /* match specified network */
1033 if (!class_match_net(ldd->ldd_params,
1034 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1037 mti->mti_nids[mti->mti_nid_count] = id.nid;
1038 mti->mti_nid_count++;
1039 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1040 CWARN("Only using first %d nids for %s\n",
1041 mti->mti_nid_count, mti->mti_svname);
1046 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1047 mti->mti_config_ver = 0;
1048 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF)
1049 ldd->ldd_flags |= LDD_F_WRITECONF;
1050 mti->mti_flags = ldd->ldd_flags;
1051 mti->mti_stripe_index = ldd->ldd_svindex;
1052 memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid));
1053 if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) {
1054 CERROR("params too big for mti\n");
1057 memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params));
1061 /* Register an old or new target with the MGS. If needed MGS will construct
1062 startup logs and assign index */
1063 int server_register_target(struct super_block *sb)
1065 struct lustre_sb_info *lsi = s2lsi(sb);
1066 struct obd_device *mgc = lsi->lsi_mgc;
1067 struct lustre_disk_data *ldd = lsi->lsi_ldd;
1068 struct mgs_target_info *mti = NULL;
1075 if (!(lsi->lsi_flags & LSI_SERVER))
1081 rc = server_sb2mti(sb, mti);
1085 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1086 mti->mti_svname, mti->mti_fsname,
1087 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1090 /* if write_conf is true, the registration must succeed */
1091 writeconf = !!(ldd->ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1092 mti->mti_flags |= LDD_F_OPC_REG;
1094 /* Register the target */
1095 /* FIXME use mgc_process_config instead */
1096 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1097 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
1098 sizeof(*mti), mti, NULL);
1100 if (mti->mti_flags & LDD_F_ERROR) {
1101 LCONSOLE_ERROR_MSG(0x160,
1102 "The MGS is refusing to allow this "
1103 "server (%s) to start. Please see messages"
1104 " on the MGS node.\n", ldd->ldd_svname);
1105 } else if (writeconf) {
1106 LCONSOLE_ERROR_MSG(0x15f,
1107 "Communication to the MGS return error %d. "
1108 "Is the MGS running?\n", rc);
1110 CERROR("Cannot talk to the MGS: %d, not fatal\n", rc);
1111 /* reset the error code for non-fatal error. */
1117 /* Always update our flags */
1118 ldd->ldd_flags = mti->mti_flags & LDD_F_ONDISK_MASK;
1120 /* If this flag is set, it means the MGS wants us to change our
1121 on-disk data. (So far this means just the index.) */
1122 if (mti->mti_flags & LDD_F_REWRITE_LDD) {
1125 CDEBUG(D_MOUNT, "Changing on-disk index from %#x to %#x "
1126 "for %s\n", ldd->ldd_svindex, mti->mti_stripe_index,
1128 ldd->ldd_svindex = mti->mti_stripe_index;
1129 strncpy(ldd->ldd_svname, mti->mti_svname,
1130 sizeof(ldd->ldd_svname));
1131 /* or ldd_make_sv_name(ldd); */
1132 ldd_write(&mgc->obd_lvfs_ctxt, ldd);
1133 err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb,
1136 CERROR("Label set error %d\n", err);
1137 label = fsfilt_get_label(mgc, lsi->lsi_srv_mnt->mnt_sb);
1139 CDEBUG(D_MOUNT, "Disk label changed to %s\n", label);
1141 /* Flush the new ldd to disk */
1142 fsfilt_sync(mgc, lsi->lsi_srv_mnt->mnt_sb);
1152 * Notify the MGS that this target is ready.
1153 * Used by IR - if the MGS receives this message, it will notify clients.
1155 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1157 struct lustre_sb_info *lsi = s2lsi(sb);
1158 struct obd_device *mgc = lsi->lsi_mgc;
1159 struct mgs_target_info *mti = NULL;
1165 if (!(lsi->lsi_flags & LSI_SERVER))
1171 rc = server_sb2mti(sb, mti);
1175 mti->mti_instance = obd->u.obt.obt_instance;
1176 mti->mti_flags |= LDD_F_OPC_READY;
1178 /* FIXME use mgc_process_config instead */
1179 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1180 sizeof(KEY_REGISTER_TARGET),
1181 KEY_REGISTER_TARGET,
1182 sizeof(*mti), mti, NULL);
1184 /* Imperative recovery: if the mgs informs us to use IR? */
1185 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1186 (mti->mti_flags & LDD_F_IR_CAPABLE))
1187 lsi->lsi_flags |= LSI_IR_CAPABLE;
1196 /** Start server targets: MDTs and OSTs
1198 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1200 struct obd_device *obd;
1201 struct lustre_sb_info *lsi = s2lsi(sb);
1202 struct config_llog_instance cfg;
1206 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname);
1209 /* If we're an MDT, make sure the global MDS is running */
1210 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1211 /* make sure the MDS is started */
1212 cfs_mutex_lock(&server_start_lock);
1213 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1215 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1216 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1218 LUSTRE_MDS_OBDNAME"_uuid",
1221 cfs_mutex_unlock(&server_start_lock);
1222 CERROR("failed to start MDS: %d\n", rc);
1226 cfs_mutex_unlock(&server_start_lock);
1230 /* If we're an OST, make sure the global OSS is running */
1231 if (IS_OST(lsi->lsi_ldd)) {
1232 /* make sure OSS is started */
1233 cfs_mutex_lock(&server_start_lock);
1234 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1236 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1238 LUSTRE_OSS_OBDNAME"_uuid",
1241 cfs_mutex_unlock(&server_start_lock);
1242 CERROR("failed to start OSS: %d\n", rc);
1246 cfs_mutex_unlock(&server_start_lock);
1249 /* Set the mgc fs to our server disk. This allows the MGC to
1250 * read and write configs locally, in case it can't talk to the MGS. */
1251 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1255 /* Register with MGS */
1256 rc = server_register_target(sb);
1260 /* Let the target look up the mount using the target's name
1261 (we can't pass the sb or mnt through class_process_config.) */
1262 rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
1266 /* Start targets using the llog named for the target */
1267 memset(&cfg, 0, sizeof(cfg));
1268 rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg);
1270 CERROR("failed to start server %s: %d\n",
1271 lsi->lsi_ldd->ldd_svname, rc);
1272 /* Do NOT call server_deregister_mount() here. This makes it
1273 * impossible to find mount later in cleanup time and leaves
1274 * @lsi and othder stuff leaked. -umka */
1279 /* Release the mgc fs for others to use */
1280 server_mgc_clear_fs(lsi->lsi_mgc);
1283 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1285 CERROR("no server named %s was started\n",
1286 lsi->lsi_ldd->ldd_svname);
1290 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1291 (OBP(obd, iocontrol))) {
1292 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1293 obd->obd_self_export, 0, NULL, NULL);
1296 server_notify_target(sb, obd);
1298 /* calculate recovery timeout, do it after lustre_process_log */
1299 server_calc_timeout(lsi, obd);
1301 /* log has been fully processed */
1302 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1308 /***************** lustre superblock **************/
1310 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1312 struct lustre_sb_info *lsi;
1318 OBD_ALLOC_PTR(lsi->lsi_lmd);
1319 if (!lsi->lsi_lmd) {
1324 lsi->lsi_lmd->lmd_exclude_count = 0;
1325 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
1326 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
1327 s2lsi_nocast(sb) = lsi;
1328 /* we take 1 extra ref for our setup */
1329 cfs_atomic_set(&lsi->lsi_mounts, 1);
1331 /* Default umount style */
1332 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1337 static int lustre_free_lsi(struct super_block *sb)
1339 struct lustre_sb_info *lsi = s2lsi(sb);
1342 LASSERT(lsi != NULL);
1343 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
1345 /* someone didn't call server_put_mount. */
1346 LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0);
1348 if (lsi->lsi_ldd != NULL)
1349 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
1351 if (lsi->lsi_lmd != NULL) {
1352 if (lsi->lsi_lmd->lmd_dev != NULL)
1353 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1354 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1355 if (lsi->lsi_lmd->lmd_profile != NULL)
1356 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1357 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1358 if (lsi->lsi_lmd->lmd_mgssec != NULL)
1359 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
1360 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
1361 if (lsi->lsi_lmd->lmd_opts != NULL)
1362 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1363 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1364 if (lsi->lsi_lmd->lmd_exclude_count)
1365 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1366 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1367 lsi->lsi_lmd->lmd_exclude_count);
1368 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1371 LASSERT(lsi->lsi_llsbi == NULL);
1372 OBD_FREE(lsi, sizeof(*lsi));
1373 s2lsi_nocast(sb) = NULL;
1378 /* The lsi has one reference for every server that is using the disk -
1379 e.g. MDT, MGS, and potentially MGC */
1380 static int lustre_put_lsi(struct super_block *sb)
1382 struct lustre_sb_info *lsi = s2lsi(sb);
1385 LASSERT(lsi != NULL);
1387 CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts));
1388 if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) {
1389 lustre_free_lsi(sb);
1395 /*************** server mount ******************/
1397 /** Kernel mount using mount options in MOUNT_DATA_FILE.
1398 * Since this file lives on the disk, we pre-mount using a common
1399 * type, read the file, then re-mount using the type specified in the
1402 static struct vfsmount *server_kernel_mount(struct super_block *sb)
1404 struct lvfs_run_ctxt mount_ctxt;
1405 struct lustre_sb_info *lsi = s2lsi(sb);
1406 struct lustre_disk_data *ldd;
1407 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1408 struct vfsmount *mnt;
1409 struct file_system_type *type;
1410 char *options = NULL;
1411 unsigned long page, s_flags;
1412 struct page *__page;
1417 OBD_ALLOC(ldd, sizeof(*ldd));
1419 RETURN(ERR_PTR(-ENOMEM));
1421 /* In the past, we have always used flags = 0.
1422 Note ext3/ldiskfs can't be mounted ro. */
1423 s_flags = sb->s_flags;
1425 /* allocate memory for options */
1426 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
1428 GOTO(out_free, rc = -ENOMEM);
1429 page = (unsigned long)cfs_page_address(__page);
1430 options = (char *)page;
1431 memset(options, 0, CFS_PAGE_SIZE);
1433 /* mount-line options must be added for pre-mount because it may
1434 * contain mount options such as journal_dev which are required
1435 * to mount successfuly the underlying filesystem */
1436 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0))
1437 strncat(options, lmd->lmd_opts, CFS_PAGE_SIZE - 1);
1439 /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */
1440 CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev);
1441 type = get_fs_type("ldiskfs");
1443 CERROR("premount failed: cannot find ldiskfs module\n");
1444 GOTO(out_free, rc = -ENODEV);
1446 mnt = vfs_kern_mount(type, s_flags, lmd->lmd_dev, (void *)options);
1447 cfs_module_put(type->owner);
1450 CERROR("premount %s:%#lx ldiskfs failed: %d "
1451 "Is the ldiskfs module available?\n",
1452 lmd->lmd_dev, s_flags, rc );
1456 OBD_SET_CTXT_MAGIC(&mount_ctxt);
1457 mount_ctxt.pwdmnt = mnt;
1458 mount_ctxt.pwd = mnt->mnt_root;
1459 mount_ctxt.fs = get_ds();
1461 rc = ldd_parse(&mount_ctxt, ldd);
1465 CERROR("premount parse options failed: rc = %d\n", rc);
1469 /* Done with our pre-mount, now do the real mount. */
1471 /* Glom up mount options */
1472 memset(options, 0, CFS_PAGE_SIZE);
1473 strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2);
1475 len = CFS_PAGE_SIZE - strlen(options) - 2;
1477 strcat(options, ",");
1478 strncat(options, "no_mbcache", len);
1480 /* Add in any mount-line options */
1481 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) {
1482 len = CFS_PAGE_SIZE - strlen(options) - 2;
1483 strcat(options, ",");
1484 strncat(options, lmd->lmd_opts, len);
1487 /* Special permanent mount flags */
1489 s_flags |= MS_NOATIME | MS_NODIRATIME;
1491 CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
1492 MT_STR(ldd), lmd->lmd_dev, options);
1493 type = get_fs_type(MT_STR(ldd));
1495 CERROR("get_fs_type failed\n");
1496 GOTO(out_free, rc = -ENODEV);
1498 mnt = vfs_kern_mount(type, s_flags, lmd->lmd_dev, (void *)options);
1499 cfs_module_put(type->owner);
1502 CERROR("vfs_kern_mount failed: rc = %d\n", rc);
1506 if (lmd->lmd_flags & LMD_FLG_ABORT_RECOV)
1507 simple_truncate(mnt->mnt_sb->s_root, mnt, LAST_RCVD,
1510 OBD_PAGE_FREE(__page);
1511 lsi->lsi_ldd = ldd; /* freed at lsi cleanup */
1512 CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
1517 OBD_PAGE_FREE(__page);
1518 OBD_FREE(ldd, sizeof(*ldd));
1519 lsi->lsi_ldd = NULL;
1520 RETURN(ERR_PTR(rc));
1523 /** Wait here forever until the mount refcount is 0 before completing umount,
1524 * else we risk dereferencing a null pointer.
1525 * LNET may take e.g. 165s before killing zombies.
1527 static void server_wait_finished(struct vfsmount *mnt)
1531 cfs_sigset_t blocked;
1533 cfs_waitq_init(&waitq);
1535 while (mnt_get_count(mnt) > 1) {
1536 if (waited && (waited % 30 == 0))
1537 LCONSOLE_WARN("Mount still busy with %d refs after "
1541 /* Cannot use l_event_wait() for an interruptible sleep. */
1543 blocked = cfs_block_sigsinv(sigmask(SIGKILL));
1544 cfs_waitq_wait_event_interruptible_timeout(
1546 (mnt_get_count(mnt) == 1),
1547 cfs_time_seconds(3),
1549 cfs_block_sigs(blocked);
1551 LCONSOLE_EMERG("Danger: interrupted umount %s with "
1552 "%d refs!\n", mnt->mnt_devname,
1553 mnt_get_count(mnt));
1560 /** Start the shutdown of servers at umount.
1562 static void server_put_super(struct super_block *sb)
1564 struct lustre_sb_info *lsi = s2lsi(sb);
1565 struct obd_device *obd;
1566 struct vfsmount *mnt = lsi->lsi_srv_mnt;
1567 char *tmpname, *extraname = NULL;
1569 int lddflags = lsi->lsi_ldd->ldd_flags;
1570 int lsiflags = lsi->lsi_flags;
1573 LASSERT(lsiflags & LSI_SERVER);
1575 tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1;
1576 OBD_ALLOC(tmpname, tmpname_sz);
1577 memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
1578 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1579 if (IS_MDT(lsi->lsi_ldd) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1580 snprintf(tmpname, tmpname_sz, "MGS");
1582 /* Stop the target */
1583 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1584 (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd))) {
1585 struct lustre_profile *lprof = NULL;
1587 /* tell the mgc to drop the config log */
1588 lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL);
1590 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1591 If there are any setup/cleanup errors, save the lov
1592 name for safety cleanup later. */
1593 lprof = class_get_profile(lsi->lsi_ldd->ldd_svname);
1594 if (lprof && lprof->lp_dt) {
1595 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1596 strcpy(extraname, lprof->lp_dt);
1599 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1601 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1602 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1604 /* We can't seem to give an error return code
1605 * to .put_super, so we better make sure we clean up! */
1607 class_manual_cleanup(obd);
1609 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
1610 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1614 /* If they wanted the mgs to stop separately from the mdt, they
1615 should have put it on a different device. */
1616 if (IS_MGS(lsi->lsi_ldd)) {
1617 /* if MDS start with --nomgs, don't stop MGS then */
1618 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1621 OBD_ALLOC(logname, MGS_PARAM_MAXLEN);
1623 LCONSOLE_WARN("Stopping mgs failed %d, please "
1624 "try again.", -ENOMEM);
1627 strcpy(logname, lsi->lsi_ldd->ldd_fsname);
1628 strcat(logname, "-params");
1629 /* tell the mgc to drop parameter config log */
1630 lustre_end_log(sb, logname, NULL);
1631 OBD_FREE(logname, MGS_PARAM_MAXLEN);
1633 server_stop_mgs(sb);
1637 /* Clean the mgc and sb */
1638 lustre_common_put_super(sb);
1640 /* Wait for the targets to really clean up - can't exit (and let the
1641 sb get destroyed) while the mount is still in use */
1642 server_wait_finished(mnt);
1644 /* drop the One True Mount */
1647 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1648 until the target is really gone so that our type refcount check
1650 server_stop_servers(lddflags, lsiflags);
1652 /* In case of startup or cleanup err, stop related obds */
1654 obd = class_name2obd(extraname);
1656 CWARN("Cleaning orphaned obd %s\n", extraname);
1658 class_manual_cleanup(obd);
1660 OBD_FREE(extraname, strlen(extraname) + 1);
1663 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1664 OBD_FREE(tmpname, tmpname_sz);
1668 /** Called only for 'umount -f'
1670 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1671 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1673 struct super_block *sb = vfsmnt->mnt_sb;
1675 static void server_umount_begin(struct super_block *sb)
1678 struct lustre_sb_info *lsi = s2lsi(sb);
1681 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1682 if (!(flags & MNT_FORCE)) {
1688 CDEBUG(D_MOUNT, "umount -f\n");
1689 /* umount = failover
1691 no third way to do non-force, non-failover */
1692 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1693 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1697 #ifndef HAVE_STATFS_DENTRY_PARAM
1698 static int server_statfs (struct super_block *sb, cfs_kstatfs_t *buf)
1701 static int server_statfs (struct dentry *dentry, cfs_kstatfs_t *buf)
1703 struct super_block *sb = dentry->d_sb;
1705 struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt;
1708 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) {
1709 #ifdef HAVE_STATFS_DENTRY_PARAM
1710 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf);
1712 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_sb, buf);
1715 buf->f_type = sb->s_magic;
1721 buf->f_type = sb->s_magic;
1722 buf->f_bsize = sb->s_blocksize;
1728 buf->f_namelen = NAME_MAX;
1732 /** The operations we support directly on the superblock:
1733 * mount, umount, and df.
1735 static struct super_operations server_ops =
1737 .put_super = server_put_super,
1738 .umount_begin = server_umount_begin, /* umount -f */
1739 .statfs = server_statfs,
1742 #define log2(n) cfs_ffz(~(n))
1743 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1745 static int server_fill_super_common(struct super_block *sb)
1747 struct inode *root = 0;
1750 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1752 sb->s_blocksize = 4096;
1753 sb->s_blocksize_bits = log2(sb->s_blocksize);
1754 sb->s_magic = LUSTRE_SUPER_MAGIC;
1755 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1756 sb->s_flags |= MS_RDONLY;
1757 sb->s_op = &server_ops;
1759 root = new_inode(sb);
1761 CERROR("Can't make root inode\n");
1765 /* returns -EIO for every operation */
1766 /* make_bad_inode(root); -- badness - can't umount */
1767 /* apparently we need to be a directory for the mount to finish */
1768 root->i_mode = S_IFDIR;
1770 sb->s_root = d_alloc_root(root);
1772 CERROR("Can't make root dentry\n");
1780 /** Fill in the superblock info for a Lustre server.
1781 * Mount the device with the correct options.
1782 * Read the on-disk config file.
1783 * Start the services.
1785 static int server_fill_super(struct super_block *sb)
1787 struct lustre_sb_info *lsi = s2lsi(sb);
1788 struct vfsmount *mnt;
1792 /* the One True Mount */
1793 mnt = server_kernel_mount(sb);
1796 CERROR("Unable to mount device %s: %d\n",
1797 lsi->lsi_lmd->lmd_dev, rc);
1801 lsi->lsi_srv_mnt = mnt;
1803 LASSERT(lsi->lsi_ldd);
1804 CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n",
1805 lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname,
1806 lsi->lsi_lmd->lmd_dev);
1808 if (class_name2obd(lsi->lsi_ldd->ldd_svname)) {
1809 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1810 "running. Double-mount may have compromised"
1811 " the disk journal.\n",
1812 lsi->lsi_ldd->ldd_svname);
1818 /* Start MGS before MGC */
1819 if (IS_MGS(lsi->lsi_ldd) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
1820 rc = server_start_mgs(sb);
1825 /* Start MGC before servers */
1826 rc = lustre_start_mgc(sb);
1830 /* Set up all obd devices for service */
1831 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1832 (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) {
1833 rc = server_start_targets(sb, mnt);
1835 CERROR("Unable to start targets: %d\n", rc);
1838 /* FIXME overmount client here,
1839 or can we just start a client log and client_fill_super on this sb?
1840 We need to make sure server_put_super gets called too - ll_put_super
1841 calls lustre_common_put_super; check there for LSI_SERVER flag,
1843 Probably should start client from new thread so we can return.
1844 Client will not finish until all servers are connected.
1845 Note - MGS-only server does NOT get a client, since there is no
1846 lustre fs associated - the MGS is for all lustre fs's */
1847 } else if (IS_MGS(lsi->lsi_ldd) &&
1848 !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
1849 struct config_llog_instance cfg;
1852 OBD_ALLOC(logname, MGS_PARAM_MAXLEN);
1853 if (logname == NULL)
1854 GOTO(out_mnt, rc = -ENOMEM);
1855 strcpy(logname, lsi->lsi_ldd->ldd_fsname);
1856 strcat(logname, "-params");
1858 memset(&cfg, 0, sizeof(cfg));
1859 rc = lustre_process_log(sb, logname, &cfg);
1860 OBD_FREE(logname, MGS_PARAM_MAXLEN);
1862 CERROR("failed to process parameters %s: %d\n",
1868 rc = server_fill_super_common(sb);
1874 /* We jump here in case of failure while starting targets or MGS.
1875 * In this case we can't just put @mnt and have to do real cleanup
1876 * with stoping targets, etc. */
1877 server_put_super(sb);
1881 /* Get the index from the obd name.
1882 rc = server type, or
1884 if endptr isn't NULL it is set to end of name */
1885 int server_name2index(char *svname, __u32 *idx, char **endptr)
1887 unsigned long index;
1889 char *dash = strrchr(svname, '-');
1893 /* intepret <fsname>-MDTXXXXX-mdc as mdt, the better way is to pass
1894 * in the fsname, then determine the server index */
1895 if (!strcmp(LUSTRE_MDC_NAME, dash + 1)) {
1897 for (; dash > svname && *dash != '-'; dash--);
1902 if (strncmp(dash + 1, "MDT", 3) == 0)
1903 rc = LDD_F_SV_TYPE_MDT;
1904 else if (strncmp(dash + 1, "OST", 3) == 0)
1905 rc = LDD_F_SV_TYPE_OST;
1908 if (strcmp(dash + 4, "all") == 0)
1909 return rc | LDD_F_SV_ALL;
1911 index = simple_strtoul(dash + 4, endptr, 16);
1917 * Calculate timeout value for a target.
1919 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1921 struct lustre_mount_data *lmd;
1925 bool has_ir = !!(lsi->lsi_flags & LSI_IR_CAPABLE);
1926 int min = OBD_RECOVERY_TIME_MIN;
1928 LASSERT(lsi->lsi_flags & LSI_SERVER);
1932 soft = lmd->lmd_recovery_time_soft;
1933 hard = lmd->lmd_recovery_time_hard;
1934 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1935 obd->obd_no_ir = !has_ir;
1939 soft = OBD_RECOVERY_TIME_SOFT;
1941 hard = OBD_RECOVERY_TIME_HARD;
1943 /* target may have ir_factor configured. */
1944 factor = OBD_IR_FACTOR_DEFAULT;
1945 if (obd->obd_recovery_ir_factor)
1946 factor = obd->obd_recovery_ir_factor;
1949 int new_soft = soft;
1950 int new_hard = hard;
1952 /* adjust timeout value by imperative recovery */
1954 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1955 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1957 /* make sure the timeout is not too short */
1958 new_soft = max(min, new_soft);
1959 new_hard = max(new_soft, new_hard);
1961 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1962 "window shrunk from %d-%d down to %d-%d\n",
1963 obd->obd_name, soft, hard, new_soft, new_hard);
1970 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1971 obd->obd_recovery_time_hard = hard;
1972 obd->obd_recovery_ir_factor = factor;
1974 EXPORT_SYMBOL(server_calc_timeout);
1976 /*************** mount common betweeen server and client ***************/
1979 int lustre_common_put_super(struct super_block *sb)
1984 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1986 /* Drop a ref to the MGC */
1987 rc = lustre_stop_mgc(sb);
1988 if (rc && (rc != -ENOENT)) {
1990 CERROR("Can't stop MGC: %d\n", rc);
1993 /* BUSY just means that there's some other obd that
1994 needs the mgc. Let him clean it up. */
1995 CDEBUG(D_MOUNT, "MGC still in use\n");
1997 /* Drop a ref to the mounted disk */
2003 static void lmd_print(struct lustre_mount_data *lmd)
2007 PRINT_CMD(PRINT_MASK, " mount data:\n");
2008 if (lmd_is_client(lmd))
2009 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
2010 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
2011 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
2014 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
2016 if (lmd->lmd_recovery_time_soft)
2017 PRINT_CMD(PRINT_MASK, "recovery time soft: %d\n",
2018 lmd->lmd_recovery_time_soft);
2020 if (lmd->lmd_recovery_time_hard)
2021 PRINT_CMD(PRINT_MASK, "recovery time hard: %d\n",
2022 lmd->lmd_recovery_time_hard);
2024 for (i = 0; i < lmd->lmd_exclude_count; i++) {
2025 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
2026 lmd->lmd_exclude[i]);
2030 /* Is this server on the exclusion list */
2031 int lustre_check_exclusion(struct super_block *sb, char *svname)
2033 struct lustre_sb_info *lsi = s2lsi(sb);
2034 struct lustre_mount_data *lmd = lsi->lsi_lmd;
2039 rc = server_name2index(svname, &index, NULL);
2040 if (rc != LDD_F_SV_TYPE_OST)
2041 /* Only exclude OSTs */
2044 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
2045 index, lmd->lmd_exclude_count, lmd->lmd_dev);
2047 for(i = 0; i < lmd->lmd_exclude_count; i++) {
2048 if (index == lmd->lmd_exclude[i]) {
2049 CWARN("Excluding %s (on exclusion list)\n", svname);
2056 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
2057 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
2059 char *s1 = ptr, *s2;
2060 __u32 index, *exclude_list;
2064 /* The shortest an ost name can be is 8 chars: -OST0000.
2065 We don't actually know the fsname at this time, so in fact
2066 a user could specify any fsname. */
2067 devmax = strlen(ptr) / 8 + 1;
2069 /* temp storage until we figure out how many we have */
2070 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
2074 /* we enter this fn pointing at the '=' */
2075 while (*s1 && *s1 != ' ' && *s1 != ',') {
2077 rc = server_name2index(s1, &index, &s2);
2079 CERROR("Can't parse server name '%s'\n", s1);
2082 if (rc == LDD_F_SV_TYPE_OST)
2083 exclude_list[lmd->lmd_exclude_count++] = index;
2085 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
2087 /* now we are pointing at ':' (next exclude)
2088 or ',' (end of excludes) */
2089 if (lmd->lmd_exclude_count >= devmax)
2092 if (rc >= 0) /* non-err */
2095 if (lmd->lmd_exclude_count) {
2096 /* permanent, freed in lustre_free_lsi */
2097 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
2098 lmd->lmd_exclude_count);
2099 if (lmd->lmd_exclude) {
2100 memcpy(lmd->lmd_exclude, exclude_list,
2101 sizeof(index) * lmd->lmd_exclude_count);
2104 lmd->lmd_exclude_count = 0;
2107 OBD_FREE(exclude_list, sizeof(index) * devmax);
2111 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
2116 if (lmd->lmd_mgssec != NULL) {
2117 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
2118 lmd->lmd_mgssec = NULL;
2121 tail = strchr(ptr, ',');
2123 length = strlen(ptr);
2125 length = tail - ptr;
2127 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
2128 if (lmd->lmd_mgssec == NULL)
2131 memcpy(lmd->lmd_mgssec, ptr, length);
2132 lmd->lmd_mgssec[length] = '\0';
2136 /** Parse mount line options
2137 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
2138 * dev is passed as device=uml1:/lustre by mount.lustre
2140 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
2142 char *s1, *s2, *devname = NULL;
2143 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
2149 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
2150 "/sbin/mount.lustre is installed.\n");
2154 /* Options should be a string - try to detect old lmd data */
2155 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
2156 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
2157 "/sbin/mount.lustre. Please install "
2158 "version %s\n", LUSTRE_VERSION_STRING);
2161 lmd->lmd_magic = LMD_MAGIC;
2163 /* Set default flags here */
2168 int time_min = OBD_RECOVERY_TIME_MIN;
2170 /* Skip whitespace and extra commas */
2171 while (*s1 == ' ' || *s1 == ',')
2174 /* Client options are parsed in ll_options: eg. flock,
2177 /* Parse non-ldiskfs options here. Rather than modifying
2178 ldiskfs, we just zero these out here */
2179 if (strncmp(s1, "abort_recov", 11) == 0) {
2180 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
2182 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
2183 lmd->lmd_recovery_time_soft = max_t(int,
2184 simple_strtoul(s1 + 19, NULL, 10), time_min);
2186 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
2187 lmd->lmd_recovery_time_hard = max_t(int,
2188 simple_strtoul(s1 + 19, NULL, 10), time_min);
2190 } else if (strncmp(s1, "noir", 4) == 0) {
2191 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
2193 } else if (strncmp(s1, "nosvc", 5) == 0) {
2194 lmd->lmd_flags |= LMD_FLG_NOSVC;
2196 } else if (strncmp(s1, "nomgs", 5) == 0) {
2197 lmd->lmd_flags |= LMD_FLG_NOMGS;
2199 } else if (strncmp(s1, "writeconf", 9) == 0) {
2200 lmd->lmd_flags |= LMD_FLG_WRITECONF;
2202 } else if (strncmp(s1, "mgssec=", 7) == 0) {
2203 rc = lmd_parse_mgssec(lmd, s1 + 7);
2207 /* ost exclusion list */
2208 } else if (strncmp(s1, "exclude=", 8) == 0) {
2209 rc = lmd_make_exclusion(lmd, s1 + 7);
2214 /* Linux 2.4 doesn't pass the device, so we stuck it at the
2215 end of the options. */
2216 else if (strncmp(s1, "device=", 7) == 0) {
2218 /* terminate options right before device. device
2219 must be the last one. */
2225 s2 = strchr(s1, ',');
2233 memmove(s1, s2, strlen(s2) + 1);
2239 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
2240 "(need mount option 'device=...')\n");
2244 s1 = strstr(devname, ":/");
2247 lmd->lmd_flags |= LMD_FLG_CLIENT;
2248 /* Remove leading /s from fsname */
2249 while (*++s1 == '/') ;
2250 /* Freed in lustre_free_lsi */
2251 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
2252 if (!lmd->lmd_profile)
2254 sprintf(lmd->lmd_profile, "%s-client", s1);
2257 /* Freed in lustre_free_lsi */
2258 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
2261 strcpy(lmd->lmd_dev, devname);
2263 /* Save mount options */
2264 s1 = options + strlen(options) - 1;
2265 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
2267 if (*options != 0) {
2268 /* Freed in lustre_free_lsi */
2269 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
2272 strcpy(lmd->lmd_opts, options);
2276 lmd->lmd_magic = LMD_MAGIC;
2281 CERROR("Bad mount options %s\n", options);
2285 struct lustre_mount_data2 {
2287 struct vfsmount *lmd2_mnt;
2290 /** This is the entry point for the mount call into Lustre.
2291 * This is called when a server or client is mounted,
2292 * and this is where we start setting things up.
2293 * @param data Mount options (e.g. -o flock,abort_recov)
2295 int lustre_fill_super(struct super_block *sb, void *data, int silent)
2297 struct lustre_mount_data *lmd;
2298 struct lustre_mount_data2 *lmd2 = data;
2299 struct lustre_sb_info *lsi;
2303 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
2305 lsi = lustre_init_lsi(sb);
2311 * Disable lockdep during mount, because mount locking patterns are
2317 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
2319 obd_zombie_barrier();
2321 /* Figure out the lmd from the mount options */
2322 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
2324 GOTO(out, rc = -EINVAL);
2327 if (lmd_is_client(lmd)) {
2328 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
2329 if (!client_fill_super) {
2330 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
2331 "client mount! Is the 'lustre' "
2332 "module loaded?\n");
2336 rc = lustre_start_mgc(sb);
2341 /* Connect and start */
2342 /* (should always be ll_fill_super) */
2343 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
2344 /* c_f_s will call lustre_common_put_super on failure */
2347 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
2348 lsi->lsi_flags |= LSI_SERVER;
2349 rc = server_fill_super(sb);
2350 /* s_f_s calls lustre_start_mgc after the mount because we need
2351 the MGS nids which are stored on disk. Plus, we may
2352 need to start the MGS first. */
2353 /* s_f_s will call server_put_super on failure */
2356 /* If error happens in fill_super() call, @lsi will be killed there.
2357 * This is why we do not put it here. */
2361 CERROR("Unable to mount %s (%d)\n",
2362 s2lsi(sb) ? lmd->lmd_dev : "", rc);
2364 CDEBUG(D_SUPER, "Mount %s complete\n",
2372 /* We can't call ll_fill_super by name because it lives in a module that
2373 must be loaded after this one. */
2374 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
2375 struct vfsmount *mnt))
2377 client_fill_super = cfs;
2380 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
2382 kill_super_cb = cfs;
2385 /***************** FS registration ******************/
2387 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))
2388 struct super_block * lustre_get_sb(struct file_system_type *fs_type, int flags,
2389 const char *devname, void * data)
2391 return get_sb_nodev(fs_type, flags, data, lustre_fill_super);
2394 int lustre_get_sb(struct file_system_type *fs_type, int flags,
2395 const char *devname, void * data, struct vfsmount *mnt)
2397 struct lustre_mount_data2 lmd2 = {data, mnt};
2399 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
2403 void lustre_kill_super(struct super_block *sb)
2405 struct lustre_sb_info *lsi = s2lsi(sb);
2407 if (kill_super_cb && lsi && !(lsi->lsi_flags & LSI_SERVER))
2408 (*kill_super_cb)(sb);
2410 kill_anon_super(sb);
2413 /** Register the "lustre" fs type
2415 struct file_system_type lustre_fs_type = {
2416 .owner = THIS_MODULE,
2418 .get_sb = lustre_get_sb,
2419 .kill_sb = lustre_kill_super,
2420 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
2421 #ifdef FS_HAS_FIEMAP
2424 LL_RENAME_DOES_D_MOVE,
2427 int lustre_register_fs(void)
2429 return register_filesystem(&lustre_fs_type);
2432 int lustre_unregister_fs(void)
2434 return unregister_filesystem(&lustre_fs_type);
2437 EXPORT_SYMBOL(lustre_register_client_fill_super);
2438 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2439 EXPORT_SYMBOL(lustre_common_put_super);
2440 EXPORT_SYMBOL(lustre_process_log);
2441 EXPORT_SYMBOL(lustre_end_log);
2442 EXPORT_SYMBOL(server_get_mount);
2443 EXPORT_SYMBOL(server_get_mount_2);
2444 EXPORT_SYMBOL(server_put_mount);
2445 EXPORT_SYMBOL(server_put_mount_2);
2446 EXPORT_SYMBOL(server_register_target);
2447 EXPORT_SYMBOL(server_name2index);
2448 EXPORT_SYMBOL(server_mti_print);
2449 EXPORT_SYMBOL(do_lcfg);