1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
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) = NULL;
60 static void (*kill_super_cb)(struct super_block *sb) = NULL;
62 /*********** mount lookup *********/
64 DECLARE_MUTEX(lustre_mount_info_lock);
65 static CFS_LIST_HEAD(server_mount_info_list);
67 static struct lustre_mount_info *server_find_mount(const char *name)
69 struct list_head *tmp;
70 struct lustre_mount_info *lmi;
73 list_for_each(tmp, &server_mount_info_list) {
74 lmi = list_entry(tmp, struct lustre_mount_info, lmi_list_chain);
75 if (strcmp(name, lmi->lmi_name) == 0)
81 /* we must register an obd for a mount before we call the setup routine.
82 *_setup will call lustre_get_mount to get the mnt struct
83 by obd_name, since we can't pass the pointer to setup. */
84 static int server_register_mount(const char *name, struct super_block *sb,
87 struct lustre_mount_info *lmi;
94 OBD_ALLOC(lmi, sizeof(*lmi));
97 OBD_ALLOC(name_cp, strlen(name) + 1);
99 OBD_FREE(lmi, sizeof(*lmi));
102 strcpy(name_cp, name);
104 down(&lustre_mount_info_lock);
106 if (server_find_mount(name)) {
107 up(&lustre_mount_info_lock);
108 OBD_FREE(lmi, sizeof(*lmi));
109 OBD_FREE(name_cp, strlen(name) + 1);
110 CERROR("Already registered %s\n", name);
113 lmi->lmi_name = name_cp;
116 list_add(&lmi->lmi_list_chain, &server_mount_info_list);
118 up(&lustre_mount_info_lock);
120 CDEBUG(D_MOUNT, "reg_mnt %p from %s, vfscount=%d\n",
121 lmi->lmi_mnt, name, atomic_read(&lmi->lmi_mnt->mnt_count));
126 /* when an obd no longer needs a mount */
127 static int server_deregister_mount(const char *name)
129 struct lustre_mount_info *lmi;
132 down(&lustre_mount_info_lock);
133 lmi = server_find_mount(name);
135 up(&lustre_mount_info_lock);
136 CERROR("%s not registered\n", name);
140 CDEBUG(D_MOUNT, "dereg_mnt %p from %s, vfscount=%d\n",
141 lmi->lmi_mnt, name, atomic_read(&lmi->lmi_mnt->mnt_count));
143 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
144 list_del(&lmi->lmi_list_chain);
145 OBD_FREE(lmi, sizeof(*lmi));
146 up(&lustre_mount_info_lock);
151 /* obd's look up a registered mount using their obdname. This is just
152 for initial obd setup to find the mount struct. It should not be
153 called every time you want to mntget. */
154 struct lustre_mount_info *server_get_mount(const char *name)
156 struct lustre_mount_info *lmi;
157 struct lustre_sb_info *lsi;
160 down(&lustre_mount_info_lock);
161 lmi = server_find_mount(name);
162 up(&lustre_mount_info_lock);
164 CERROR("Can't find mount for %s\n", name);
167 lsi = s2lsi(lmi->lmi_sb);
168 mntget(lmi->lmi_mnt);
169 atomic_inc(&lsi->lsi_mounts);
171 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d, vfscount=%d\n",
172 lmi->lmi_mnt, name, atomic_read(&lsi->lsi_mounts),
173 atomic_read(&lmi->lmi_mnt->mnt_count));
179 * Used by mdt to get mount_info from obdname.
180 * There are no blocking when using the mount_info.
181 * Do not use server_get_mount for this purpose.
183 struct lustre_mount_info *server_get_mount_2(const char *name)
185 struct lustre_mount_info *lmi;
188 down(&lustre_mount_info_lock);
189 lmi = server_find_mount(name);
190 up(&lustre_mount_info_lock);
192 CERROR("Can't find mount for %s\n", name);
197 static void unlock_mntput(struct vfsmount *mnt)
199 if (kernel_locked()) {
208 static int lustre_put_lsi(struct super_block *sb);
210 /* to be called from obd_cleanup methods */
211 int server_put_mount(const char *name, struct vfsmount *mnt)
213 struct lustre_mount_info *lmi;
214 struct lustre_sb_info *lsi;
215 int count = atomic_read(&mnt->mnt_count) - 1;
218 /* This might be the last one, can't deref after this */
221 down(&lustre_mount_info_lock);
222 lmi = server_find_mount(name);
223 up(&lustre_mount_info_lock);
225 CERROR("Can't find mount for %s\n", name);
228 lsi = s2lsi(lmi->lmi_sb);
229 LASSERT(lmi->lmi_mnt == mnt);
231 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d, vfscount=%d\n",
232 lmi->lmi_mnt, name, atomic_read(&lsi->lsi_mounts), count);
234 if (lustre_put_lsi(lmi->lmi_sb)) {
235 CDEBUG(D_MOUNT, "Last put of mnt %p from %s, vfscount=%d\n",
236 lmi->lmi_mnt, name, count);
237 /* last mount is the One True Mount */
239 CERROR("%s: mount busy, vfscount=%d!\n", name, count);
242 /* this obd should never need the mount again */
243 server_deregister_mount(name);
248 /* Corresponding to server_get_mount_2 */
249 int server_put_mount_2(const char *name, struct vfsmount *mnt)
255 /******* mount helper utilities *********/
258 static void ldd_print(struct lustre_disk_data *ldd)
260 PRINT_CMD(PRINT_MASK, " disk data:\n");
261 PRINT_CMD(PRINT_MASK, "server: %s\n", ldd->ldd_svname);
262 PRINT_CMD(PRINT_MASK, "uuid: %s\n", (char *)ldd->ldd_uuid);
263 PRINT_CMD(PRINT_MASK, "fs: %s\n", ldd->ldd_fsname);
264 PRINT_CMD(PRINT_MASK, "index: %04x\n", ldd->ldd_svindex);
265 PRINT_CMD(PRINT_MASK, "config: %d\n", ldd->ldd_config_ver);
266 PRINT_CMD(PRINT_MASK, "flags: %#x\n", ldd->ldd_flags);
267 PRINT_CMD(PRINT_MASK, "diskfs: %s\n", MT_STR(ldd));
268 PRINT_CMD(PRINT_MASK, "options: %s\n", ldd->ldd_mount_opts);
269 PRINT_CMD(PRINT_MASK, "params: %s\n", ldd->ldd_params);
270 PRINT_CMD(PRINT_MASK, "comment: %s\n", ldd->ldd_userdata);
274 static int ldd_parse(struct lvfs_run_ctxt *mount_ctxt,
275 struct lustre_disk_data *ldd)
277 struct lvfs_run_ctxt saved;
284 push_ctxt(&saved, mount_ctxt, NULL);
286 file = filp_open(MOUNT_DATA_FILE, O_RDONLY, 0644);
289 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
293 len = i_size_read(file->f_dentry->d_inode);
294 CDEBUG(D_MOUNT, "Have %s, size %lu\n", MOUNT_DATA_FILE, len);
295 if (len != sizeof(*ldd)) {
296 CERROR("disk data size does not match: see %lu expect "LPSZ"\n",
298 GOTO(out_close, rc = -EINVAL);
301 rc = lustre_fread(file, ldd, len, &off);
303 CERROR("error reading %s: read %d of %lu\n",
304 MOUNT_DATA_FILE, rc, len);
305 GOTO(out_close, rc = -EINVAL);
309 if (ldd->ldd_magic != LDD_MAGIC) {
310 /* FIXME add swabbing support */
311 CERROR("Bad magic in %s: %x!=%x\n", MOUNT_DATA_FILE,
312 ldd->ldd_magic, LDD_MAGIC);
313 GOTO(out_close, rc = -EINVAL);
316 if (ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP) {
317 CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
319 ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP);
320 GOTO(out_close, rc = -EINVAL);
322 if (ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP) {
323 CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
325 ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP);
326 /* Do something like remount filesystem read-only */
327 GOTO(out_close, rc = -EINVAL);
333 pop_ctxt(&saved, mount_ctxt, NULL);
337 static int ldd_write(struct lvfs_run_ctxt *mount_ctxt,
338 struct lustre_disk_data *ldd)
340 struct lvfs_run_ctxt saved;
343 unsigned long len = sizeof(struct lustre_disk_data);
347 LASSERT(ldd->ldd_magic == LDD_MAGIC);
349 ldd->ldd_config_ver++;
351 push_ctxt(&saved, mount_ctxt, NULL);
353 file = filp_open(MOUNT_DATA_FILE, O_RDWR, 0644);
356 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
360 rc = lustre_fwrite(file, ldd, len, &off);
362 CERROR("error writing %s: read %d of %lu\n",
363 MOUNT_DATA_FILE, rc, len);
364 GOTO(out_close, rc = -EINVAL);
372 pop_ctxt(&saved, mount_ctxt, NULL);
377 /**************** config llog ********************/
379 /* Get a config log from the MGS and process it.
380 This func is called for both clients and servers.
381 Continue to process new statements appended to the logs
382 (whenever the config lock is revoked) until lustre_end_log
384 int lustre_process_log(struct super_block *sb, char *logname,
385 struct config_llog_instance *cfg)
387 struct lustre_cfg *lcfg;
388 struct lustre_cfg_bufs bufs;
389 struct lustre_sb_info *lsi = s2lsi(sb);
390 struct obd_device *mgc = lsi->lsi_mgc;
397 /* mgc_process_config */
398 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
399 lustre_cfg_bufs_set_string(&bufs, 1, logname);
400 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
401 lustre_cfg_bufs_set(&bufs, 3, &sb, sizeof(sb));
402 lcfg = lustre_cfg_new(LCFG_LOG_START, &bufs);
403 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
404 lustre_cfg_free(lcfg);
407 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
408 "failed from the MGS (%d). Make sure this "
409 "client and the MGS are running compatible "
410 "versions of Lustre.\n",
411 mgc->obd_name, logname, rc);
414 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
415 "failed (%d). This may be the result of "
416 "communication errors between this node and "
417 "the MGS, a bad configuration, or other "
418 "errors. See the syslog for more "
419 "information.\n", mgc->obd_name, logname,
422 /* class_obd_list(); */
426 /* Stop watching this config log for updates */
427 int lustre_end_log(struct super_block *sb, char *logname,
428 struct config_llog_instance *cfg)
430 struct lustre_cfg *lcfg;
431 struct lustre_cfg_bufs bufs;
432 struct lustre_sb_info *lsi = s2lsi(sb);
433 struct obd_device *mgc = lsi->lsi_mgc;
440 /* mgc_process_config */
441 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
442 lustre_cfg_bufs_set_string(&bufs, 1, logname);
444 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
445 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
446 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
447 lustre_cfg_free(lcfg);
451 /**************** obd start *******************/
453 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
454 char *s1, char *s2, char *s3, char *s4)
456 struct lustre_cfg_bufs bufs;
457 struct lustre_cfg * lcfg = NULL;
460 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
461 cmd, s1, s2, s3, s4);
463 lustre_cfg_bufs_reset(&bufs, cfgname);
465 lustre_cfg_bufs_set_string(&bufs, 1, s1);
467 lustre_cfg_bufs_set_string(&bufs, 2, s2);
469 lustre_cfg_bufs_set_string(&bufs, 3, s3);
471 lustre_cfg_bufs_set_string(&bufs, 4, s4);
473 lcfg = lustre_cfg_new(cmd, &bufs);
474 lcfg->lcfg_nid = nid;
475 rc = class_process_config(lcfg);
476 lustre_cfg_free(lcfg);
480 static int lustre_start_simple(char *obdname, char *type, char *uuid,
484 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
486 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
488 CERROR("%s attach error %d\n", obdname, rc);
491 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, 0, 0);
493 CERROR("%s setup error %d\n", obdname, rc);
494 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
499 /* Set up a MGS to serve startup logs */
500 static int server_start_mgs(struct super_block *sb)
502 struct lustre_sb_info *lsi = s2lsi(sb);
503 struct vfsmount *mnt = lsi->lsi_srv_mnt;
504 struct lustre_mount_info *lmi;
509 /* It is impossible to have more than 1 MGS per node, since
510 MGC wouldn't know which to connect to */
511 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
513 lsi = s2lsi(lmi->lmi_sb);
514 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
516 lsi->lsi_ldd->ldd_svname);
520 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
522 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
525 ((rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
526 LUSTRE_MGS_OBDNAME, 0, 0))))
527 server_deregister_mount(LUSTRE_MGS_OBDNAME);
530 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
531 "Is the 'mgs' module loaded?\n",
532 LUSTRE_MGS_OBDNAME, rc);
536 static int server_stop_mgs(struct super_block *sb)
538 struct obd_device *obd;
542 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
544 /* There better be only one MGS */
545 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
547 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
551 /* The MGS should always stop when we say so */
553 rc = class_manual_cleanup(obd);
557 DECLARE_MUTEX(mgc_start_lock);
559 /** Set up a mgc obd to process startup logs
561 * \param sb [in] super block of the mgc obd
563 * \retval 0 success, otherwise error code
565 static int lustre_start_mgc(struct super_block *sb)
567 struct lustre_handle mgc_conn = {0, };
568 struct obd_connect_data *data = NULL;
569 struct lustre_sb_info *lsi = s2lsi(sb);
570 struct obd_device *obd;
571 struct obd_export *exp;
572 struct obd_uuid *uuid;
575 char *mgcname, *niduuid;
578 int rc = 0, i = 0, j, len;
581 LASSERT(lsi->lsi_lmd);
583 /* Find the first non-lo MGS nid for our MGC name */
584 if (lsi->lsi_flags & LSI_SERVER) {
585 ptr = lsi->lsi_ldd->ldd_params;
586 /* Use mgsnode= nids */
587 if ((class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0) &&
588 (class_parse_nid(ptr, &nid, &ptr) == 0)) {
590 } else if (IS_MGS(lsi->lsi_ldd)) {
591 lnet_process_id_t id;
592 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
593 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
600 } else { /* client */
601 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
602 ptr = lsi->lsi_lmd->lmd_dev;
603 if (class_parse_nid(ptr, &nid, &ptr) == 0)
607 CERROR("No valid MGS nids found.\n");
611 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
612 OBD_ALLOC(mgcname, len);
613 OBD_ALLOC(niduuid, len + 2);
614 if (!mgcname || !niduuid)
615 GOTO(out_free, rc = -ENOMEM);
616 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
618 mutex_down(&mgc_start_lock);
620 obd = class_name2obd(mgcname);
621 if (obd && !obd->obd_stopping) {
622 /* Re-using an existing MGC */
623 atomic_inc(&obd->u.cli.cl_mgc_refcount);
626 /* If we are restarting the MGS, don't try to keep the MGC's
627 old connection, or registration will fail. */
628 if ((lsi->lsi_flags & LSI_SERVER) && IS_MGS(lsi->lsi_ldd)) {
629 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
633 /* Try all connections, but only once (again).
634 We don't want to block another target from starting
635 (using its local copy of the log), but we do want to connect
636 if at all possible. */
638 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
639 rc = obd_set_info_async(obd->obd_self_export,
640 sizeof(KEY_INIT_RECOV_BACKUP),
641 KEY_INIT_RECOV_BACKUP,
642 sizeof(recov_bk), &recov_bk, NULL);
646 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
648 /* Add the primary nids for the MGS */
650 sprintf(niduuid, "%s_%x", mgcname, i);
651 if (lsi->lsi_flags & LSI_SERVER) {
652 ptr = lsi->lsi_ldd->ldd_params;
653 if (IS_MGS(lsi->lsi_ldd)) {
654 /* Use local nids (including LO) */
655 lnet_process_id_t id;
656 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
657 rc = do_lcfg(mgcname, id.nid,
658 LCFG_ADD_UUID, niduuid, 0,0,0);
661 /* Use mgsnode= nids */
662 if (class_find_param(ptr, PARAM_MGSNODE, &ptr) != 0) {
663 CERROR("No MGS nids given.\n");
664 GOTO(out_free, rc = -EINVAL);
666 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
667 rc = do_lcfg(mgcname, nid,
668 LCFG_ADD_UUID, niduuid, 0,0,0);
672 } else { /* client */
673 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
674 ptr = lsi->lsi_lmd->lmd_dev;
675 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
676 rc = do_lcfg(mgcname, nid,
677 LCFG_ADD_UUID, niduuid, 0,0,0);
679 /* Stop at the first failover nid */
685 CERROR("No valid MGS nids found.\n");
686 GOTO(out_free, rc = -EINVAL);
688 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
690 /* Random uuid for MGC allows easier reconnects */
692 ll_generate_random_uuid(uuidc);
693 class_uuid_unparse(uuidc, uuid);
696 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
697 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
703 /* Add any failover MGS nids */
705 while ((*ptr == ':' ||
706 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0)) {
707 /* New failover node */
708 sprintf(niduuid, "%s_%x", mgcname, i);
710 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
712 rc = do_lcfg(mgcname, nid,
713 LCFG_ADD_UUID, niduuid, 0,0,0);
718 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
726 lsi->lsi_lmd->lmd_mgs_failnodes = i;
728 obd = class_name2obd(mgcname);
730 CERROR("Can't find mgcobd %s\n", mgcname);
731 GOTO(out_free, rc = -ENOTCONN);
734 /* Keep a refcount of servers/clients who started with "mount",
735 so we know when we can get rid of the mgc. */
736 atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
738 /* Try all connections, but only once. */
740 rc = obd_set_info_async(obd->obd_self_export,
741 sizeof(KEY_INIT_RECOV_BACKUP),
742 KEY_INIT_RECOV_BACKUP,
743 sizeof(recov_bk), &recov_bk, NULL);
746 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
747 /* We connect to the MGS at setup, and don't disconnect until cleanup */
750 GOTO(out, rc = -ENOMEM);
751 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID |
753 data->ocd_version = LUSTRE_VERSION_CODE;
754 rc = obd_connect(NULL, &mgc_conn, obd, &(obd->obd_uuid), data, NULL);
757 CERROR("connect failed %d\n", rc);
761 exp = class_conn2export(&mgc_conn);
762 obd->u.cli.cl_mgc_mgsexp = exp;
765 /* Keep the mgc info in the sb. Note that many lsi's can point
769 mutex_up(&mgc_start_lock);
772 OBD_FREE(mgcname, len);
774 OBD_FREE(niduuid, len + 2);
778 static int lustre_stop_mgc(struct super_block *sb)
780 struct lustre_sb_info *lsi = s2lsi(sb);
781 struct obd_device *obd;
782 char *niduuid = 0, *ptr = 0;
783 int i, rc = 0, len = 0;
793 mutex_down(&mgc_start_lock);
794 LASSERT(atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
795 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
796 /* This is not fatal, every client that stops
797 will call in here. */
798 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
799 atomic_read(&obd->u.cli.cl_mgc_refcount));
800 GOTO(out, rc = -EBUSY);
803 /* The MGC has no recoverable data in any case.
804 * force shotdown set in umount_begin */
805 obd->obd_no_recov = 1;
807 if (obd->u.cli.cl_mgc_mgsexp) {
808 /* An error is not fatal, if we are unable to send the
809 disconnect mgs ping evictor cleans up the export */
810 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
812 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
815 /* Save the obdname for cleaning the nid uuids, which are
817 len = strlen(obd->obd_name) + 6;
818 OBD_ALLOC(niduuid, len);
820 strcpy(niduuid, obd->obd_name);
821 ptr = niduuid + strlen(niduuid);
824 rc = class_manual_cleanup(obd);
828 /* Clean the nid uuids */
830 GOTO(out, rc = -ENOMEM);
832 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
833 sprintf(ptr, "_%x", i);
834 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
837 CERROR("del MDC UUID %s failed: rc = %d\n",
842 OBD_FREE(niduuid, len);
844 /* class_import_put will get rid of the additional connections */
845 mutex_up(&mgc_start_lock);
849 /* Since there's only one mgc per node, we have to change it's fs to get
850 access to the right disk. */
851 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
853 struct lustre_sb_info *lsi = s2lsi(sb);
857 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
859 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
860 rc = obd_set_info_async(mgc->obd_self_export,
861 sizeof(KEY_SET_FS), KEY_SET_FS,
862 sizeof(*sb), sb, NULL);
864 CERROR("can't set_fs %d\n", rc);
870 static int server_mgc_clear_fs(struct obd_device *mgc)
875 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
877 rc = obd_set_info_async(mgc->obd_self_export,
878 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
883 DECLARE_MUTEX(server_start_lock);
885 /* Stop MDS/OSS if nobody is using them */
886 static int server_stop_servers(int lddflags, int lsiflags)
888 struct obd_device *obd = NULL;
889 struct obd_type *type = NULL;
893 mutex_down(&server_start_lock);
895 /* Either an MDT or an OST or neither */
896 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
897 if ((lddflags & LDD_F_SV_TYPE_MDT) &&
898 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
899 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
900 type = class_search_type(LUSTRE_MDS_NAME);
902 /* if this was an OST, and there are no more OST's, clean up the OSS */
903 if ((lddflags & LDD_F_SV_TYPE_OST) &&
904 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
905 type = class_search_type(LUSTRE_OST_NAME);
908 if (obd && (!type || !type->typ_refcnt)) {
911 /* obd_fail doesn't mean much on a server obd */
912 err = class_manual_cleanup(obd);
917 mutex_up(&server_start_lock);
922 int server_mti_print(char *title, struct mgs_target_info *mti)
924 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
925 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
926 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
927 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
928 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
929 mti->mti_config_ver, mti->mti_flags);
933 static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
935 struct lustre_sb_info *lsi = s2lsi(sb);
936 struct lustre_disk_data *ldd = lsi->lsi_ldd;
937 lnet_process_id_t id;
941 if (!(lsi->lsi_flags & LSI_SERVER))
944 strncpy(mti->mti_fsname, ldd->ldd_fsname,
945 sizeof(mti->mti_fsname));
946 strncpy(mti->mti_svname, ldd->ldd_svname,
947 sizeof(mti->mti_svname));
949 mti->mti_nid_count = 0;
950 while (LNetGetId(i++, &id) != -ENOENT) {
951 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
953 mti->mti_nids[mti->mti_nid_count] = id.nid;
954 mti->mti_nid_count++;
955 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
956 CWARN("Only using first %d nids for %s\n",
957 mti->mti_nid_count, mti->mti_svname);
962 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
963 mti->mti_config_ver = 0;
964 mti->mti_flags = ldd->ldd_flags;
965 mti->mti_stripe_index = ldd->ldd_svindex;
966 memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid));
967 if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) {
968 CERROR("params too big for mti\n");
971 memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params));
975 /* Register an old or new target with the MGS. If needed MGS will construct
976 startup logs and assign index */
977 int server_register_target(struct super_block *sb)
979 struct lustre_sb_info *lsi = s2lsi(sb);
980 struct obd_device *mgc = lsi->lsi_mgc;
981 struct lustre_disk_data *ldd = lsi->lsi_ldd;
982 struct mgs_target_info *mti = NULL;
988 if (!(lsi->lsi_flags & LSI_SERVER))
994 rc = server_sb2mti(sb, mti);
998 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
999 mti->mti_svname, mti->mti_fsname,
1000 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1003 /* Register the target */
1004 /* FIXME use mgc_process_config instead */
1005 rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
1006 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
1007 sizeof(*mti), mti, NULL);
1011 /* Always update our flags */
1012 ldd->ldd_flags = mti->mti_flags & ~LDD_F_REWRITE_LDD;
1014 /* If this flag is set, it means the MGS wants us to change our
1015 on-disk data. (So far this means just the index.) */
1016 if (mti->mti_flags & LDD_F_REWRITE_LDD) {
1019 CDEBUG(D_MOUNT, "Changing on-disk index from %#x to %#x "
1020 "for %s\n", ldd->ldd_svindex, mti->mti_stripe_index,
1022 ldd->ldd_svindex = mti->mti_stripe_index;
1023 strncpy(ldd->ldd_svname, mti->mti_svname,
1024 sizeof(ldd->ldd_svname));
1025 /* or ldd_make_sv_name(ldd); */
1026 ldd_write(&mgc->obd_lvfs_ctxt, ldd);
1027 err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb,
1030 CERROR("Label set error %d\n", err);
1031 label = fsfilt_get_label(mgc, lsi->lsi_srv_mnt->mnt_sb);
1033 CDEBUG(D_MOUNT, "Disk label changed to %s\n", label);
1035 /* Flush the new ldd to disk */
1036 fsfilt_sync(mgc, lsi->lsi_srv_mnt->mnt_sb);
1046 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1048 struct obd_device *obd;
1049 struct lustre_sb_info *lsi = s2lsi(sb);
1050 struct config_llog_instance cfg;
1054 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname);
1057 /* If we're an MDT, make sure the global MDS is running */
1058 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1059 /* make sure the MDS is started */
1060 mutex_down(&server_start_lock);
1061 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1063 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1064 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1066 LUSTRE_MDS_OBDNAME"_uuid",
1069 mutex_up(&server_start_lock);
1070 CERROR("failed to start MDS: %d\n", rc);
1074 mutex_up(&server_start_lock);
1078 /* If we're an OST, make sure the global OSS is running */
1079 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
1080 /* make sure OSS is started */
1081 mutex_down(&server_start_lock);
1082 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1084 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1086 LUSTRE_OSS_OBDNAME"_uuid",
1089 mutex_up(&server_start_lock);
1090 CERROR("failed to start OSS: %d\n", rc);
1094 mutex_up(&server_start_lock);
1097 /* Set the mgc fs to our server disk. This allows the MGC
1098 to read and write configs locally. */
1099 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1103 /* Register with MGS */
1104 rc = server_register_target(sb);
1105 if (rc && (lsi->lsi_ldd->ldd_flags &
1106 (LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_UPGRADE14))){
1107 CERROR("Required registration failed for %s: %d\n",
1108 lsi->lsi_ldd->ldd_svname, rc);
1110 LCONSOLE_ERROR_MSG(0x15f, "Communication error with "
1111 "the MGS. Is the MGS running?\n");
1115 if (rc == -EINVAL) {
1116 LCONSOLE_ERROR_MSG(0x160, "The MGS is refusing to allow this "
1117 "server (%s) to start. Please see messages"
1118 " on the MGS node.\n",
1119 lsi->lsi_ldd->ldd_svname);
1122 /* non-fatal error of registeration with MGS */
1124 CDEBUG(D_MOUNT, "Cannot register with MGS: %d\n", rc);
1126 /* Let the target look up the mount using the target's name
1127 (we can't pass the sb or mnt through class_process_config.) */
1128 rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
1132 /* Start targets using the llog named for the target */
1133 memset(&cfg, 0, sizeof(cfg));
1134 rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg);
1136 CERROR("failed to start server %s: %d\n",
1137 lsi->lsi_ldd->ldd_svname, rc);
1142 /* Release the mgc fs for others to use */
1143 server_mgc_clear_fs(lsi->lsi_mgc);
1146 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1148 CERROR("no server named %s was started\n",
1149 lsi->lsi_ldd->ldd_svname);
1153 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1154 (OBP(obd, iocontrol))) {
1155 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1156 obd->obd_self_export, 0, NULL, NULL);
1159 /* log has been fully processed */
1160 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1166 /***************** lustre superblock **************/
1168 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1170 struct lustre_sb_info *lsi;
1176 OBD_ALLOC_PTR(lsi->lsi_lmd);
1177 if (!lsi->lsi_lmd) {
1182 lsi->lsi_lmd->lmd_exclude_count = 0;
1183 s2lsi_nocast(sb) = lsi;
1184 /* we take 1 extra ref for our setup */
1185 atomic_set(&lsi->lsi_mounts, 1);
1187 /* Default umount style */
1188 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1193 static int lustre_free_lsi(struct super_block *sb)
1195 struct lustre_sb_info *lsi = s2lsi(sb);
1198 LASSERT(lsi != NULL);
1199 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
1201 /* someone didn't call server_put_mount. */
1202 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
1204 if (lsi->lsi_ldd != NULL)
1205 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
1207 if (lsi->lsi_lmd != NULL) {
1208 if (lsi->lsi_lmd->lmd_dev != NULL)
1209 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1210 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1211 if (lsi->lsi_lmd->lmd_profile != NULL)
1212 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1213 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1214 if (lsi->lsi_lmd->lmd_opts != NULL)
1215 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1216 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1217 if (lsi->lsi_lmd->lmd_exclude_count)
1218 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1219 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1220 lsi->lsi_lmd->lmd_exclude_count);
1221 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1224 LASSERT(lsi->lsi_llsbi == NULL);
1225 OBD_FREE(lsi, sizeof(*lsi));
1226 s2lsi_nocast(sb) = NULL;
1231 /* The lsi has one reference for every server that is using the disk -
1232 e.g. MDT, MGS, and potentially MGC */
1233 static int lustre_put_lsi(struct super_block *sb)
1235 struct lustre_sb_info *lsi = s2lsi(sb);
1238 LASSERT(lsi != NULL);
1240 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
1241 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
1242 lustre_free_lsi(sb);
1248 /*************** server mount ******************/
1250 /* Kernel mount using mount options in MOUNT_DATA_FILE */
1251 static struct vfsmount *server_kernel_mount(struct super_block *sb)
1253 struct lvfs_run_ctxt mount_ctxt;
1254 struct lustre_sb_info *lsi = s2lsi(sb);
1255 struct lustre_disk_data *ldd;
1256 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1257 struct vfsmount *mnt;
1258 char *options = NULL;
1259 unsigned long page, s_flags;
1260 struct page *__page;
1264 OBD_ALLOC(ldd, sizeof(*ldd));
1266 RETURN(ERR_PTR(-ENOMEM));
1268 /* In the past, we have always used flags = 0.
1269 Note ext3/ldiskfs can't be mounted ro. */
1270 s_flags = sb->s_flags;
1272 /* allocate memory for options */
1273 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
1275 GOTO(out_free, rc = -ENOMEM);
1276 page = (unsigned long)cfs_page_address(__page);
1277 options = (char *)page;
1278 memset(options, 0, CFS_PAGE_SIZE);
1280 /* mount-line options must be added for pre-mount because it may
1281 * contain mount options such as journal_dev which are required
1282 * to mount successfuly the underlying filesystem */
1283 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0))
1284 strncat(options, lmd->lmd_opts, CFS_PAGE_SIZE - 1);
1286 /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */
1287 CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev);
1288 mnt = ll_kern_mount("ldiskfs", s_flags, lmd->lmd_dev, (void *)options);
1291 CERROR("premount %s:%#lx ldiskfs failed: %d "
1292 "Is the ldiskfs module available?\n",
1293 lmd->lmd_dev, s_flags, rc );
1297 OBD_SET_CTXT_MAGIC(&mount_ctxt);
1298 mount_ctxt.pwdmnt = mnt;
1299 mount_ctxt.pwd = mnt->mnt_root;
1300 mount_ctxt.fs = get_ds();
1302 rc = ldd_parse(&mount_ctxt, ldd);
1306 CERROR("premount parse options failed: rc = %d\n", rc);
1310 /* Done with our pre-mount, now do the real mount. */
1312 /* Glom up mount options */
1313 memset(options, 0, CFS_PAGE_SIZE);
1314 strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2);
1316 /* Add in any mount-line options */
1317 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) {
1318 int len = CFS_PAGE_SIZE - strlen(options) - 2;
1320 strcat(options, ",");
1321 strncat(options, lmd->lmd_opts, len);
1324 /* Special permanent mount flags */
1326 s_flags |= MS_NOATIME | MS_NODIRATIME;
1328 CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
1329 MT_STR(ldd), lmd->lmd_dev, options);
1330 mnt = ll_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev,
1334 CERROR("ll_kern_mount failed: rc = %d\n", rc);
1338 OBD_PAGE_FREE(__page);
1339 lsi->lsi_ldd = ldd; /* freed at lsi cleanup */
1340 CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
1345 OBD_PAGE_FREE(__page);
1346 OBD_FREE(ldd, sizeof(*ldd));
1347 lsi->lsi_ldd = NULL;
1348 RETURN(ERR_PTR(rc));
1351 static void server_wait_finished(struct vfsmount *mnt)
1353 wait_queue_head_t waitq;
1354 struct l_wait_info lwi;
1357 init_waitqueue_head(&waitq);
1359 while ((atomic_read(&mnt->mnt_count) > 1) && (retries > 0)) {
1360 LCONSOLE_WARN("Mount still busy with %d refs, waiting for "
1362 atomic_read(&mnt->mnt_count), retries);
1364 /* Wait for a bit */
1366 lwi = LWI_TIMEOUT(5 * HZ, NULL, NULL);
1367 l_wait_event(waitq, 0, &lwi);
1369 if (atomic_read(&mnt->mnt_count) > 1) {
1370 CERROR("Mount %p is still busy (%d refs), giving up.\n",
1371 mnt, atomic_read(&mnt->mnt_count));
1375 static void server_put_super(struct super_block *sb)
1377 struct lustre_sb_info *lsi = s2lsi(sb);
1378 struct obd_device *obd;
1379 struct vfsmount *mnt = lsi->lsi_srv_mnt;
1380 char *tmpname, *extraname = NULL;
1382 int lddflags = lsi->lsi_ldd->ldd_flags;
1383 int lsiflags = lsi->lsi_flags;
1386 LASSERT(lsiflags & LSI_SERVER);
1388 tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1;
1389 OBD_ALLOC(tmpname, tmpname_sz);
1390 memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
1391 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1393 /* Stop the target */
1394 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1395 (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd))) {
1396 struct lustre_profile *lprof = NULL;
1398 /* tell the mgc to drop the config log */
1399 lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL);
1401 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1402 If there are any setup/cleanup errors, save the lov
1403 name for safety cleanup later. */
1404 lprof = class_get_profile(lsi->lsi_ldd->ldd_svname);
1405 if (lprof && lprof->lp_dt) {
1406 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1407 strcpy(extraname, lprof->lp_dt);
1410 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1412 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1413 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1415 /* We can't seem to give an error return code
1416 * to .put_super, so we better make sure we clean up! */
1418 class_manual_cleanup(obd);
1420 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
1421 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1425 /* If they wanted the mgs to stop separately from the mdt, they
1426 should have put it on a different device. */
1427 if (IS_MGS(lsi->lsi_ldd)) {
1428 /* if MDS start with --nomgs, don't stop MGS then */
1429 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1430 server_stop_mgs(sb);
1433 /* Clean the mgc and sb */
1434 lustre_common_put_super(sb);
1436 /* Wait for the targets to really clean up - can't exit (and let the
1437 sb get destroyed) while the mount is still in use */
1438 server_wait_finished(mnt);
1440 /* drop the One True Mount */
1443 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1444 until the target is really gone so that our type refcount check
1446 server_stop_servers(lddflags, lsiflags);
1448 /* In case of startup or cleanup err, stop related obds */
1450 obd = class_name2obd(extraname);
1452 CWARN("Cleaning orphaned obd %s\n", extraname);
1454 class_manual_cleanup(obd);
1456 OBD_FREE(extraname, strlen(extraname) + 1);
1459 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1460 OBD_FREE(tmpname, tmpname_sz);
1464 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1465 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1467 struct super_block *sb = vfsmnt->mnt_sb;
1469 static void server_umount_begin(struct super_block *sb)
1472 struct lustre_sb_info *lsi = s2lsi(sb);
1475 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1476 if (!(flags & MNT_FORCE)) {
1482 CDEBUG(D_MOUNT, "umount -f\n");
1483 /* umount = failover
1485 no third way to do non-force, non-failover */
1486 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1487 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1491 #ifndef HAVE_STATFS_DENTRY_PARAM
1492 static int server_statfs (struct super_block *sb, struct kstatfs *buf)
1495 static int server_statfs (struct dentry *dentry, struct kstatfs *buf)
1497 struct super_block *sb = dentry->d_sb;
1499 struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt;
1502 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) {
1503 #ifdef HAVE_STATFS_DENTRY_PARAM
1504 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf);
1506 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_sb, buf);
1509 buf->f_type = sb->s_magic;
1515 buf->f_type = sb->s_magic;
1516 buf->f_bsize = sb->s_blocksize;
1522 buf->f_namelen = NAME_MAX;
1526 static struct super_operations server_ops =
1528 .put_super = server_put_super,
1529 .umount_begin = server_umount_begin, /* umount -f */
1530 .statfs = server_statfs,
1533 #define log2(n) ffz(~(n))
1534 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1536 static int server_fill_super_common(struct super_block *sb)
1538 struct inode *root = 0;
1541 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1543 sb->s_blocksize = 4096;
1544 sb->s_blocksize_bits = log2(sb->s_blocksize);
1545 sb->s_magic = LUSTRE_SUPER_MAGIC;
1546 sb->s_maxbytes = 0; //PAGE_CACHE_MAXBYTES;
1547 sb->s_flags |= MS_RDONLY;
1548 sb->s_op = &server_ops;
1550 root = new_inode(sb);
1552 CERROR("Can't make root inode\n");
1556 /* returns -EIO for every operation */
1557 /* make_bad_inode(root); -- badness - can't umount */
1558 /* apparently we need to be a directory for the mount to finish */
1559 root->i_mode = S_IFDIR;
1561 sb->s_root = d_alloc_root(root);
1563 CERROR("Can't make root dentry\n");
1571 static int server_fill_super(struct super_block *sb)
1573 struct lustre_sb_info *lsi = s2lsi(sb);
1574 struct vfsmount *mnt;
1578 /* the One True Mount */
1579 mnt = server_kernel_mount(sb);
1582 CERROR("Unable to mount device %s: %d\n",
1583 lsi->lsi_lmd->lmd_dev, rc);
1587 lsi->lsi_srv_mnt = mnt;
1589 LASSERT(lsi->lsi_ldd);
1590 CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n",
1591 lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname,
1592 lsi->lsi_lmd->lmd_dev);
1594 if (class_name2obd(lsi->lsi_ldd->ldd_svname)) {
1595 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1596 "running. Double-mount may have compromised"
1597 " the disk journal.\n",
1598 lsi->lsi_ldd->ldd_svname);
1604 /* Start MGS before MGC */
1605 if (IS_MGS(lsi->lsi_ldd) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1606 rc = server_start_mgs(sb);
1611 rc = lustre_start_mgc(sb);
1615 /* Set up all obd devices for service */
1616 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1617 (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) {
1618 rc = server_start_targets(sb, mnt);
1620 CERROR("Unable to start targets: %d\n", rc);
1623 /* FIXME overmount client here,
1624 or can we just start a client log and client_fill_super on this sb?
1625 We need to make sure server_put_super gets called too - ll_put_super
1626 calls lustre_common_put_super; check there for LSI_SERVER flag,
1628 Probably should start client from new thread so we can return.
1629 Client will not finish until all servers are connected.
1630 Note - MGS-only server does NOT get a client, since there is no
1631 lustre fs associated - the MGS is for all lustre fs's */
1634 rc = server_fill_super_common(sb);
1638 LCONSOLE_WARN("Server %s on device %s has started\n",
1639 lsi->lsi_ldd->ldd_svname, lsi->lsi_lmd->lmd_dev);
1643 /* We jump here in case of failure while starting targets or MGS.
1644 * In this case we can't just put @mnt and have to do real cleanup
1645 * with stoping targets, etc. */
1646 server_put_super(sb);
1650 /* Get the index from the obd name.
1651 rc = server type, or
1653 if endptr isn't NULL it is set to end of name */
1654 int server_name2index(char *svname, __u32 *idx, char **endptr)
1656 unsigned long index;
1658 char *dash = strchr(svname, '-');
1662 if (strncmp(dash + 1, "MDT", 3) == 0)
1663 rc = LDD_F_SV_TYPE_MDT;
1664 else if (strncmp(dash + 1, "OST", 3) == 0)
1665 rc = LDD_F_SV_TYPE_OST;
1669 index = simple_strtoul(dash + 4, endptr, 16);
1674 /*************** mount common betweeen server and client ***************/
1677 int lustre_common_put_super(struct super_block *sb)
1682 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1684 /* Drop a ref to the MGC */
1685 rc = lustre_stop_mgc(sb);
1686 if (rc && (rc != -ENOENT)) {
1688 CERROR("Can't stop MGC: %d\n", rc);
1691 /* BUSY just means that there's some other obd that
1692 needs the mgc. Let him clean it up. */
1693 CDEBUG(D_MOUNT, "MGC still in use\n");
1695 /* Drop a ref to the mounted disk */
1702 static void lmd_print(struct lustre_mount_data *lmd)
1706 PRINT_CMD(PRINT_MASK, " mount data:\n");
1707 if (lmd_is_client(lmd))
1708 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
1709 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
1710 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
1712 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
1713 for (i = 0; i < lmd->lmd_exclude_count; i++) {
1714 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
1715 lmd->lmd_exclude[i]);
1720 /* Is this server on the exclusion list */
1721 int lustre_check_exclusion(struct super_block *sb, char *svname)
1723 struct lustre_sb_info *lsi = s2lsi(sb);
1724 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1729 rc = server_name2index(svname, &index, NULL);
1730 if (rc != LDD_F_SV_TYPE_OST)
1731 /* Only exclude OSTs */
1734 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
1735 index, lmd->lmd_exclude_count, lmd->lmd_dev);
1737 for(i = 0; i < lmd->lmd_exclude_count; i++) {
1738 if (index == lmd->lmd_exclude[i]) {
1739 CWARN("Excluding %s (on exclusion list)\n", svname);
1746 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
1747 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
1749 char *s1 = ptr, *s2;
1750 __u32 index, *exclude_list;
1754 /* The shortest an ost name can be is 8 chars: -OST0000.
1755 We don't actually know the fsname at this time, so in fact
1756 a user could specify any fsname. */
1757 devmax = strlen(ptr) / 8 + 1;
1759 /* temp storage until we figure out how many we have */
1760 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
1764 /* we enter this fn pointing at the '=' */
1765 while (*s1 && *s1 != ' ' && *s1 != ',') {
1767 rc = server_name2index(s1, &index, &s2);
1769 CERROR("Can't parse server name '%s'\n", s1);
1772 if (rc == LDD_F_SV_TYPE_OST)
1773 exclude_list[lmd->lmd_exclude_count++] = index;
1775 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
1777 /* now we are pointing at ':' (next exclude)
1778 or ',' (end of excludes) */
1779 if (lmd->lmd_exclude_count >= devmax)
1782 if (rc >= 0) /* non-err */
1785 if (lmd->lmd_exclude_count) {
1786 /* permanent, freed in lustre_free_lsi */
1787 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
1788 lmd->lmd_exclude_count);
1789 if (lmd->lmd_exclude) {
1790 memcpy(lmd->lmd_exclude, exclude_list,
1791 sizeof(index) * lmd->lmd_exclude_count);
1794 lmd->lmd_exclude_count = 0;
1797 OBD_FREE(exclude_list, sizeof(index) * devmax);
1801 /* mount -v -t lustre uml1:uml2:/lustre-client /mnt/lustre */
1802 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1804 char *s1, *s2, *devname = NULL;
1805 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1811 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1812 "/sbin/mount.lustre is installed.\n");
1816 /* Options should be a string - try to detect old lmd data */
1817 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1818 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1819 "/sbin/mount.lustre. Please install "
1820 "version %s\n", LUSTRE_VERSION_STRING);
1823 lmd->lmd_magic = LMD_MAGIC;
1825 /* Set default flags here */
1830 /* Skip whitespace and extra commas */
1831 while (*s1 == ' ' || *s1 == ',')
1834 /* Client options are parsed in ll_options: eg. flock,
1837 /* Parse non-ldiskfs options here. Rather than modifying
1838 ldiskfs, we just zero these out here */
1839 if (strncmp(s1, "abort_recov", 11) == 0) {
1840 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1842 } else if (strncmp(s1, "nosvc", 5) == 0) {
1843 lmd->lmd_flags |= LMD_FLG_NOSVC;
1845 } else if (strncmp(s1, "nomgs", 5) == 0) {
1846 lmd->lmd_flags |= LMD_FLG_NOMGS;
1848 /* ost exclusion list */
1849 } else if (strncmp(s1, "exclude=", 8) == 0) {
1850 rc = lmd_make_exclusion(lmd, s1 + 7);
1855 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1856 end of the options. */
1857 else if (strncmp(s1, "device=", 7) == 0) {
1859 /* terminate options right before device. device
1860 must be the last one. */
1866 s2 = strchr(s1, ',');
1874 memmove(s1, s2, strlen(s2) + 1);
1880 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1881 "(need mount option 'device=...')\n");
1885 s1 = strstr(devname, ":/");
1888 lmd->lmd_flags = LMD_FLG_CLIENT;
1889 /* Remove leading /s from fsname */
1890 while (*++s1 == '/') ;
1891 /* Freed in lustre_free_lsi */
1892 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1893 if (!lmd->lmd_profile)
1895 sprintf(lmd->lmd_profile, "%s-client", s1);
1898 /* Freed in lustre_free_lsi */
1899 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1902 strcpy(lmd->lmd_dev, devname);
1904 /* Save mount options */
1905 s1 = options + strlen(options) - 1;
1906 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1908 if (*options != 0) {
1909 /* Freed in lustre_free_lsi */
1910 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1913 strcpy(lmd->lmd_opts, options);
1916 lmd->lmd_magic = LMD_MAGIC;
1921 CERROR("Bad mount options %s\n", options);
1927 int lustre_fill_super(struct super_block *sb, void *data, int silent)
1929 struct lustre_mount_data *lmd;
1930 struct lustre_sb_info *lsi;
1934 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1936 lsi = lustre_init_lsi(sb);
1942 * Disable lockdep during mount, because mount locking patterns are
1947 /* Figure out the lmd from the mount options */
1948 if (lmd_parse((char *)data, lmd)) {
1950 GOTO(out, rc = -EINVAL);
1953 if (lmd_is_client(lmd)) {
1954 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1955 if (!client_fill_super) {
1956 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1957 "client mount! Is the 'lustre' "
1958 "module loaded?\n");
1962 rc = lustre_start_mgc(sb);
1967 /* Connect and start */
1968 /* (should always be ll_fill_super) */
1969 rc = (*client_fill_super)(sb);
1970 /* c_f_s will call lustre_common_put_super on failure */
1973 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1974 lsi->lsi_flags |= LSI_SERVER;
1975 rc = server_fill_super(sb);
1976 /* s_f_s calls lustre_start_mgc after the mount because we need
1977 the MGS nids which are stored on disk. Plus, we may
1978 need to start the MGS first. */
1979 /* s_f_s will call server_put_super on failure */
1982 /* If error happens in fill_super() call, @lsi will be killed there.
1983 * This is why we do not put it here. */
1987 CERROR("Unable to mount %s (%d)\n",
1988 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1990 CDEBUG(D_SUPER, "Mount %s complete\n",
1998 /* We can't call ll_fill_super by name because it lives in a module that
1999 must be loaded after this one. */
2000 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb))
2002 client_fill_super = cfs;
2005 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
2007 kill_super_cb = cfs;
2010 /***************** FS registration ******************/
2012 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))
2013 struct super_block * lustre_get_sb(struct file_system_type *fs_type,
2014 int flags, const char *devname, void * data)
2016 return get_sb_nodev(fs_type, flags, data, lustre_fill_super);
2019 int lustre_get_sb(struct file_system_type *fs_type,
2020 int flags, const char *devname, void * data,
2021 struct vfsmount *mnt)
2023 return get_sb_nodev(fs_type, flags, data, lustre_fill_super, mnt);
2027 void lustre_kill_super(struct super_block *sb)
2029 struct lustre_sb_info *lsi = s2lsi(sb);
2031 if (kill_super_cb && lsi && !(lsi->lsi_flags & LSI_SERVER))
2032 (*kill_super_cb)(sb);
2034 kill_anon_super(sb);
2037 struct file_system_type lustre_fs_type = {
2038 .owner = THIS_MODULE,
2040 .get_sb = lustre_get_sb,
2041 .kill_sb = lustre_kill_super,
2042 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
2043 LL_RENAME_DOES_D_MOVE,
2046 int lustre_register_fs(void)
2048 return register_filesystem(&lustre_fs_type);
2051 int lustre_unregister_fs(void)
2053 return unregister_filesystem(&lustre_fs_type);
2056 EXPORT_SYMBOL(lustre_register_client_fill_super);
2057 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2058 EXPORT_SYMBOL(lustre_common_put_super);
2059 EXPORT_SYMBOL(lustre_process_log);
2060 EXPORT_SYMBOL(lustre_end_log);
2061 EXPORT_SYMBOL(server_get_mount);
2062 EXPORT_SYMBOL(server_get_mount_2);
2063 EXPORT_SYMBOL(server_put_mount);
2064 EXPORT_SYMBOL(server_put_mount_2);
2065 EXPORT_SYMBOL(server_register_target);
2066 EXPORT_SYMBOL(server_name2index);
2067 EXPORT_SYMBOL(server_mti_print);
2068 EXPORT_SYMBOL(do_lcfg);