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 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
795 /* This is not fatal, every client that stops
796 will call in here. */
797 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
798 atomic_read(&obd->u.cli.cl_mgc_refcount));
799 GOTO(out, rc = -EBUSY);
802 /* The MGC has no recoverable data in any case.
803 * force shotdown set in umount_begin */
804 obd->obd_no_recov = 1;
806 if (obd->u.cli.cl_mgc_mgsexp) {
807 /* An error is not fatal, if we are unable to send the
808 disconnect mgs ping evictor cleans up the export */
809 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
811 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
814 /* Save the obdname for cleaning the nid uuids, which are
816 len = strlen(obd->obd_name) + 6;
817 OBD_ALLOC(niduuid, len);
819 strcpy(niduuid, obd->obd_name);
820 ptr = niduuid + strlen(niduuid);
823 rc = class_manual_cleanup(obd);
827 /* Clean the nid uuids */
830 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
831 sprintf(ptr, "_%x", i);
832 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
835 CERROR("del MDC UUID %s failed: rc = %d\n",
838 OBD_FREE(niduuid, len);
839 /* class_import_put will get rid of the additional connections */
842 mutex_up(&mgc_start_lock);
846 /* Since there's only one mgc per node, we have to change it's fs to get
847 access to the right disk. */
848 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
850 struct lustre_sb_info *lsi = s2lsi(sb);
854 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
856 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
857 rc = obd_set_info_async(mgc->obd_self_export,
858 sizeof(KEY_SET_FS), KEY_SET_FS,
859 sizeof(*sb), sb, NULL);
861 CERROR("can't set_fs %d\n", rc);
867 static int server_mgc_clear_fs(struct obd_device *mgc)
872 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
874 rc = obd_set_info_async(mgc->obd_self_export,
875 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
880 DECLARE_MUTEX(server_start_lock);
882 /* Stop MDS/OSS if nobody is using them */
883 static int server_stop_servers(int lddflags, int lsiflags)
885 struct obd_device *obd = NULL;
886 struct obd_type *type = NULL;
890 mutex_down(&server_start_lock);
892 /* Either an MDT or an OST or neither */
893 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
894 if ((lddflags & LDD_F_SV_TYPE_MDT) &&
895 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
896 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
897 type = class_search_type(LUSTRE_MDS_NAME);
899 /* if this was an OST, and there are no more OST's, clean up the OSS */
900 if ((lddflags & LDD_F_SV_TYPE_OST) &&
901 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
902 type = class_search_type(LUSTRE_OST_NAME);
905 if (obd && (!type || !type->typ_refcnt)) {
908 /* obd_fail doesn't mean much on a server obd */
909 err = class_manual_cleanup(obd);
914 mutex_up(&server_start_lock);
919 int server_mti_print(char *title, struct mgs_target_info *mti)
921 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
922 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
923 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
924 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
925 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
926 mti->mti_config_ver, mti->mti_flags);
930 static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
932 struct lustre_sb_info *lsi = s2lsi(sb);
933 struct lustre_disk_data *ldd = lsi->lsi_ldd;
934 lnet_process_id_t id;
938 if (!(lsi->lsi_flags & LSI_SERVER))
941 strncpy(mti->mti_fsname, ldd->ldd_fsname,
942 sizeof(mti->mti_fsname));
943 strncpy(mti->mti_svname, ldd->ldd_svname,
944 sizeof(mti->mti_svname));
946 mti->mti_nid_count = 0;
947 while (LNetGetId(i++, &id) != -ENOENT) {
948 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
950 mti->mti_nids[mti->mti_nid_count] = id.nid;
951 mti->mti_nid_count++;
952 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
953 CWARN("Only using first %d nids for %s\n",
954 mti->mti_nid_count, mti->mti_svname);
959 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
960 mti->mti_config_ver = 0;
961 mti->mti_flags = ldd->ldd_flags;
962 mti->mti_stripe_index = ldd->ldd_svindex;
963 memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid));
964 if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) {
965 CERROR("params too big for mti\n");
968 memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params));
972 /* Register an old or new target with the MGS. If needed MGS will construct
973 startup logs and assign index */
974 int server_register_target(struct super_block *sb)
976 struct lustre_sb_info *lsi = s2lsi(sb);
977 struct obd_device *mgc = lsi->lsi_mgc;
978 struct lustre_disk_data *ldd = lsi->lsi_ldd;
979 struct mgs_target_info *mti = NULL;
985 if (!(lsi->lsi_flags & LSI_SERVER))
991 rc = server_sb2mti(sb, mti);
995 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
996 mti->mti_svname, mti->mti_fsname,
997 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1000 /* Register the target */
1001 /* FIXME use mgc_process_config instead */
1002 rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
1003 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
1004 sizeof(*mti), mti, NULL);
1008 /* Always update our flags */
1009 ldd->ldd_flags = mti->mti_flags & ~LDD_F_REWRITE_LDD;
1011 /* If this flag is set, it means the MGS wants us to change our
1012 on-disk data. (So far this means just the index.) */
1013 if (mti->mti_flags & LDD_F_REWRITE_LDD) {
1016 CDEBUG(D_MOUNT, "Changing on-disk index from %#x to %#x "
1017 "for %s\n", ldd->ldd_svindex, mti->mti_stripe_index,
1019 ldd->ldd_svindex = mti->mti_stripe_index;
1020 strncpy(ldd->ldd_svname, mti->mti_svname,
1021 sizeof(ldd->ldd_svname));
1022 /* or ldd_make_sv_name(ldd); */
1023 ldd_write(&mgc->obd_lvfs_ctxt, ldd);
1024 err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb,
1027 CERROR("Label set error %d\n", err);
1028 label = fsfilt_get_label(mgc, lsi->lsi_srv_mnt->mnt_sb);
1030 CDEBUG(D_MOUNT, "Disk label changed to %s\n", label);
1032 /* Flush the new ldd to disk */
1033 fsfilt_sync(mgc, lsi->lsi_srv_mnt->mnt_sb);
1043 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1045 struct obd_device *obd;
1046 struct lustre_sb_info *lsi = s2lsi(sb);
1047 struct config_llog_instance cfg;
1051 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname);
1054 /* If we're an MDT, make sure the global MDS is running */
1055 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1056 /* make sure the MDS is started */
1057 mutex_down(&server_start_lock);
1058 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1060 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1061 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1063 LUSTRE_MDS_OBDNAME"_uuid",
1066 mutex_up(&server_start_lock);
1067 CERROR("failed to start MDS: %d\n", rc);
1071 mutex_up(&server_start_lock);
1075 /* If we're an OST, make sure the global OSS is running */
1076 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
1077 /* make sure OSS is started */
1078 mutex_down(&server_start_lock);
1079 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1081 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1083 LUSTRE_OSS_OBDNAME"_uuid",
1086 mutex_up(&server_start_lock);
1087 CERROR("failed to start OSS: %d\n", rc);
1091 mutex_up(&server_start_lock);
1094 /* Set the mgc fs to our server disk. This allows the MGC
1095 to read and write configs locally. */
1096 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1100 /* Register with MGS */
1101 rc = server_register_target(sb);
1102 if (rc && (lsi->lsi_ldd->ldd_flags &
1103 (LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_UPGRADE14))){
1104 CERROR("Required registration failed for %s: %d\n",
1105 lsi->lsi_ldd->ldd_svname, rc);
1107 LCONSOLE_ERROR_MSG(0x15f, "Communication error with "
1108 "the MGS. Is the MGS running?\n");
1112 if (rc == -EINVAL) {
1113 LCONSOLE_ERROR_MSG(0x160, "The MGS is refusing to allow this "
1114 "server (%s) to start. Please see messages"
1115 " on the MGS node.\n",
1116 lsi->lsi_ldd->ldd_svname);
1119 /* non-fatal error of registeration with MGS */
1121 CDEBUG(D_MOUNT, "Cannot register with MGS: %d\n", rc);
1123 /* Let the target look up the mount using the target's name
1124 (we can't pass the sb or mnt through class_process_config.) */
1125 rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
1129 /* Start targets using the llog named for the target */
1130 memset(&cfg, 0, sizeof(cfg));
1131 rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg);
1133 CERROR("failed to start server %s: %d\n",
1134 lsi->lsi_ldd->ldd_svname, rc);
1139 /* Release the mgc fs for others to use */
1140 server_mgc_clear_fs(lsi->lsi_mgc);
1143 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1145 CERROR("no server named %s was started\n",
1146 lsi->lsi_ldd->ldd_svname);
1150 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1151 (OBP(obd, iocontrol))) {
1152 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1153 obd->obd_self_export, 0, NULL, NULL);
1156 /* log has been fully processed */
1157 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1163 /***************** lustre superblock **************/
1165 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1167 struct lustre_sb_info *lsi;
1173 OBD_ALLOC_PTR(lsi->lsi_lmd);
1174 if (!lsi->lsi_lmd) {
1179 lsi->lsi_lmd->lmd_exclude_count = 0;
1180 s2lsi_nocast(sb) = lsi;
1181 /* we take 1 extra ref for our setup */
1182 atomic_set(&lsi->lsi_mounts, 1);
1184 /* Default umount style */
1185 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1190 static int lustre_free_lsi(struct super_block *sb)
1192 struct lustre_sb_info *lsi = s2lsi(sb);
1195 LASSERT(lsi != NULL);
1196 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
1198 /* someone didn't call server_put_mount. */
1199 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
1201 if (lsi->lsi_ldd != NULL)
1202 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
1204 if (lsi->lsi_lmd != NULL) {
1205 if (lsi->lsi_lmd->lmd_dev != NULL)
1206 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1207 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1208 if (lsi->lsi_lmd->lmd_profile != NULL)
1209 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1210 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1211 if (lsi->lsi_lmd->lmd_opts != NULL)
1212 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1213 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1214 if (lsi->lsi_lmd->lmd_exclude_count)
1215 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1216 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1217 lsi->lsi_lmd->lmd_exclude_count);
1218 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1221 LASSERT(lsi->lsi_llsbi == NULL);
1222 OBD_FREE(lsi, sizeof(*lsi));
1223 s2lsi_nocast(sb) = NULL;
1228 /* The lsi has one reference for every server that is using the disk -
1229 e.g. MDT, MGS, and potentially MGC */
1230 static int lustre_put_lsi(struct super_block *sb)
1232 struct lustre_sb_info *lsi = s2lsi(sb);
1235 LASSERT(lsi != NULL);
1237 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
1238 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
1239 lustre_free_lsi(sb);
1245 /*************** server mount ******************/
1247 /* Kernel mount using mount options in MOUNT_DATA_FILE */
1248 static struct vfsmount *server_kernel_mount(struct super_block *sb)
1250 struct lvfs_run_ctxt mount_ctxt;
1251 struct lustre_sb_info *lsi = s2lsi(sb);
1252 struct lustre_disk_data *ldd;
1253 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1254 struct vfsmount *mnt;
1255 char *options = NULL;
1256 unsigned long page, s_flags;
1257 struct page *__page;
1261 OBD_ALLOC(ldd, sizeof(*ldd));
1263 RETURN(ERR_PTR(-ENOMEM));
1265 /* In the past, we have always used flags = 0.
1266 Note ext3/ldiskfs can't be mounted ro. */
1267 s_flags = sb->s_flags;
1269 /* allocate memory for options */
1270 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
1272 GOTO(out_free, rc = -ENOMEM);
1273 page = (unsigned long)cfs_page_address(__page);
1274 options = (char *)page;
1275 memset(options, 0, CFS_PAGE_SIZE);
1277 /* mount-line options must be added for pre-mount because it may
1278 * contain mount options such as journal_dev which are required
1279 * to mount successfuly the underlying filesystem */
1280 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0))
1281 strncat(options, lmd->lmd_opts, CFS_PAGE_SIZE - 1);
1283 /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */
1284 CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev);
1285 mnt = ll_kern_mount("ldiskfs", s_flags, lmd->lmd_dev, (void *)options);
1288 CERROR("premount %s:%#lx ldiskfs failed: %d "
1289 "Is the ldiskfs module available?\n",
1290 lmd->lmd_dev, s_flags, rc );
1294 OBD_SET_CTXT_MAGIC(&mount_ctxt);
1295 mount_ctxt.pwdmnt = mnt;
1296 mount_ctxt.pwd = mnt->mnt_root;
1297 mount_ctxt.fs = get_ds();
1299 rc = ldd_parse(&mount_ctxt, ldd);
1303 CERROR("premount parse options failed: rc = %d\n", rc);
1307 /* Done with our pre-mount, now do the real mount. */
1309 /* Glom up mount options */
1310 memset(options, 0, CFS_PAGE_SIZE);
1311 strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2);
1313 /* Add in any mount-line options */
1314 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) {
1315 int len = CFS_PAGE_SIZE - strlen(options) - 2;
1317 strcat(options, ",");
1318 strncat(options, lmd->lmd_opts, len);
1321 /* Special permanent mount flags */
1323 s_flags |= MS_NOATIME | MS_NODIRATIME;
1325 CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
1326 MT_STR(ldd), lmd->lmd_dev, options);
1327 mnt = ll_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev,
1331 CERROR("ll_kern_mount failed: rc = %d\n", rc);
1335 OBD_PAGE_FREE(__page);
1336 lsi->lsi_ldd = ldd; /* freed at lsi cleanup */
1337 CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
1342 OBD_PAGE_FREE(__page);
1343 OBD_FREE(ldd, sizeof(*ldd));
1344 lsi->lsi_ldd = NULL;
1345 RETURN(ERR_PTR(rc));
1348 static void server_wait_finished(struct vfsmount *mnt)
1350 wait_queue_head_t waitq;
1351 struct l_wait_info lwi;
1354 init_waitqueue_head(&waitq);
1356 while ((atomic_read(&mnt->mnt_count) > 1) && (retries > 0)) {
1357 LCONSOLE_WARN("Mount still busy with %d refs, waiting for "
1359 atomic_read(&mnt->mnt_count), retries);
1361 /* Wait for a bit */
1363 lwi = LWI_TIMEOUT(5 * HZ, NULL, NULL);
1364 l_wait_event(waitq, 0, &lwi);
1366 if (atomic_read(&mnt->mnt_count) > 1) {
1367 CERROR("Mount %p is still busy (%d refs), giving up.\n",
1368 mnt, atomic_read(&mnt->mnt_count));
1372 static void server_put_super(struct super_block *sb)
1374 struct lustre_sb_info *lsi = s2lsi(sb);
1375 struct obd_device *obd;
1376 struct vfsmount *mnt = lsi->lsi_srv_mnt;
1377 char *tmpname, *extraname = NULL;
1379 int lddflags = lsi->lsi_ldd->ldd_flags;
1380 int lsiflags = lsi->lsi_flags;
1384 LASSERT(lsiflags & LSI_SERVER);
1386 tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1;
1387 OBD_ALLOC(tmpname, tmpname_sz);
1388 memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
1389 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1391 /* Stop the target */
1392 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1393 (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd))) {
1394 struct lustre_profile *lprof = NULL;
1396 /* tell the mgc to drop the config log */
1397 lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL);
1399 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1400 If there are any setup/cleanup errors, save the lov
1401 name for safety cleanup later. */
1402 lprof = class_get_profile(lsi->lsi_ldd->ldd_svname);
1403 if (lprof && lprof->lp_dt) {
1404 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1405 strcpy(extraname, lprof->lp_dt);
1408 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1410 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1411 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1413 /* We can't seem to give an error return code
1414 * to .put_super, so we better make sure we clean up! */
1416 class_manual_cleanup(obd);
1418 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
1419 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1423 /* If they wanted the mgs to stop separately from the mdt, they
1424 should have put it on a different device. */
1425 if (IS_MGS(lsi->lsi_ldd)) {
1426 /* stop the mgc before the mgs so the connection gets cleaned
1428 lustre_stop_mgc(sb);
1429 /* if MDS start with --nomgs, don't stop MGS then */
1430 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1431 server_stop_mgs(sb);
1434 /* Clean the mgc and sb */
1435 rc = lustre_common_put_super(sb);
1436 /* FIXME how can I report a failure to umount? */
1438 /* Wait for the targets to really clean up - can't exit (and let the
1439 sb get destroyed) while the mount is still in use */
1440 server_wait_finished(mnt);
1442 /* drop the One True Mount */
1445 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1446 until the target is really gone so that our type refcount check
1448 server_stop_servers(lddflags, lsiflags);
1450 /* In case of startup or cleanup err, stop related obds */
1452 obd = class_name2obd(extraname);
1454 CWARN("Cleaning orphaned obd %s\n", extraname);
1456 class_manual_cleanup(obd);
1458 OBD_FREE(extraname, strlen(extraname) + 1);
1461 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1462 OBD_FREE(tmpname, tmpname_sz);
1466 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1467 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1469 struct super_block *sb = vfsmnt->mnt_sb;
1471 static void server_umount_begin(struct super_block *sb)
1474 struct lustre_sb_info *lsi = s2lsi(sb);
1477 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1478 if (!(flags & MNT_FORCE)) {
1484 CDEBUG(D_MOUNT, "umount -f\n");
1485 /* umount = failover
1487 no third way to do non-force, non-failover */
1488 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1489 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1493 #ifndef HAVE_STATFS_DENTRY_PARAM
1494 static int server_statfs (struct super_block *sb, struct kstatfs *buf)
1497 static int server_statfs (struct dentry *dentry, struct kstatfs *buf)
1499 struct super_block *sb = dentry->d_sb;
1501 struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt;
1504 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) {
1505 #ifdef HAVE_STATFS_DENTRY_PARAM
1506 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf);
1508 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_sb, buf);
1511 buf->f_type = sb->s_magic;
1517 buf->f_type = sb->s_magic;
1518 buf->f_bsize = sb->s_blocksize;
1524 buf->f_namelen = NAME_MAX;
1528 static struct super_operations server_ops =
1530 .put_super = server_put_super,
1531 .umount_begin = server_umount_begin, /* umount -f */
1532 .statfs = server_statfs,
1535 #define log2(n) ffz(~(n))
1536 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1538 static int server_fill_super_common(struct super_block *sb)
1540 struct inode *root = 0;
1543 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1545 sb->s_blocksize = 4096;
1546 sb->s_blocksize_bits = log2(sb->s_blocksize);
1547 sb->s_magic = LUSTRE_SUPER_MAGIC;
1548 sb->s_maxbytes = 0; //PAGE_CACHE_MAXBYTES;
1549 sb->s_flags |= MS_RDONLY;
1550 sb->s_op = &server_ops;
1552 root = new_inode(sb);
1554 CERROR("Can't make root inode\n");
1558 /* returns -EIO for every operation */
1559 /* make_bad_inode(root); -- badness - can't umount */
1560 /* apparently we need to be a directory for the mount to finish */
1561 root->i_mode = S_IFDIR;
1563 sb->s_root = d_alloc_root(root);
1565 CERROR("Can't make root dentry\n");
1573 static int server_fill_super(struct super_block *sb)
1575 struct lustre_sb_info *lsi = s2lsi(sb);
1576 struct vfsmount *mnt;
1580 /* the One True Mount */
1581 mnt = server_kernel_mount(sb);
1584 CERROR("Unable to mount device %s: %d\n",
1585 lsi->lsi_lmd->lmd_dev, rc);
1589 lsi->lsi_srv_mnt = mnt;
1591 LASSERT(lsi->lsi_ldd);
1592 CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n",
1593 lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname,
1594 lsi->lsi_lmd->lmd_dev);
1596 if (class_name2obd(lsi->lsi_ldd->ldd_svname)) {
1597 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1598 "running. Double-mount may have compromised"
1599 " the disk journal.\n",
1600 lsi->lsi_ldd->ldd_svname);
1606 /* Start MGS before MGC */
1607 if (IS_MGS(lsi->lsi_ldd) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1608 rc = server_start_mgs(sb);
1613 rc = lustre_start_mgc(sb);
1617 /* Set up all obd devices for service */
1618 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1619 (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) {
1620 rc = server_start_targets(sb, mnt);
1622 CERROR("Unable to start targets: %d\n", rc);
1625 /* FIXME overmount client here,
1626 or can we just start a client log and client_fill_super on this sb?
1627 We need to make sure server_put_super gets called too - ll_put_super
1628 calls lustre_common_put_super; check there for LSI_SERVER flag,
1630 Probably should start client from new thread so we can return.
1631 Client will not finish until all servers are connected.
1632 Note - MGS-only server does NOT get a client, since there is no
1633 lustre fs associated - the MGS is for all lustre fs's */
1636 rc = server_fill_super_common(sb);
1640 LCONSOLE_WARN("Server %s on device %s has started\n",
1641 lsi->lsi_ldd->ldd_svname, lsi->lsi_lmd->lmd_dev);
1645 /* We jump here in case of failure while starting targets or MGS.
1646 * In this case we can't just put @mnt and have to do real cleanup
1647 * with stoping targets, etc. */
1648 server_put_super(sb);
1652 /* Get the index from the obd name.
1653 rc = server type, or
1655 if endptr isn't NULL it is set to end of name */
1656 int server_name2index(char *svname, __u32 *idx, char **endptr)
1658 unsigned long index;
1660 char *dash = strchr(svname, '-');
1664 if (strncmp(dash + 1, "MDT", 3) == 0)
1665 rc = LDD_F_SV_TYPE_MDT;
1666 else if (strncmp(dash + 1, "OST", 3) == 0)
1667 rc = LDD_F_SV_TYPE_OST;
1671 index = simple_strtoul(dash + 4, endptr, 16);
1676 /*************** mount common betweeen server and client ***************/
1679 int lustre_common_put_super(struct super_block *sb)
1684 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1686 /* Drop a ref to the MGC */
1687 rc = lustre_stop_mgc(sb);
1688 if (rc && (rc != -ENOENT)) {
1690 CERROR("Can't stop MGC: %d\n", rc);
1693 /* BUSY just means that there's some other obd that
1694 needs the mgc. Let him clean it up. */
1695 CDEBUG(D_MOUNT, "MGC still in use\n");
1697 /* Drop a ref to the mounted disk */
1704 static void lmd_print(struct lustre_mount_data *lmd)
1708 PRINT_CMD(PRINT_MASK, " mount data:\n");
1709 if (lmd_is_client(lmd))
1710 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
1711 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
1712 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
1714 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
1715 for (i = 0; i < lmd->lmd_exclude_count; i++) {
1716 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
1717 lmd->lmd_exclude[i]);
1722 /* Is this server on the exclusion list */
1723 int lustre_check_exclusion(struct super_block *sb, char *svname)
1725 struct lustre_sb_info *lsi = s2lsi(sb);
1726 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1731 rc = server_name2index(svname, &index, NULL);
1732 if (rc != LDD_F_SV_TYPE_OST)
1733 /* Only exclude OSTs */
1736 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
1737 index, lmd->lmd_exclude_count, lmd->lmd_dev);
1739 for(i = 0; i < lmd->lmd_exclude_count; i++) {
1740 if (index == lmd->lmd_exclude[i]) {
1741 CWARN("Excluding %s (on exclusion list)\n", svname);
1748 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
1749 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
1751 char *s1 = ptr, *s2;
1752 __u32 index, *exclude_list;
1756 /* The shortest an ost name can be is 8 chars: -OST0000.
1757 We don't actually know the fsname at this time, so in fact
1758 a user could specify any fsname. */
1759 devmax = strlen(ptr) / 8 + 1;
1761 /* temp storage until we figure out how many we have */
1762 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
1766 /* we enter this fn pointing at the '=' */
1767 while (*s1 && *s1 != ' ' && *s1 != ',') {
1769 rc = server_name2index(s1, &index, &s2);
1771 CERROR("Can't parse server name '%s'\n", s1);
1774 if (rc == LDD_F_SV_TYPE_OST)
1775 exclude_list[lmd->lmd_exclude_count++] = index;
1777 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
1779 /* now we are pointing at ':' (next exclude)
1780 or ',' (end of excludes) */
1781 if (lmd->lmd_exclude_count >= devmax)
1784 if (rc >= 0) /* non-err */
1787 if (lmd->lmd_exclude_count) {
1788 /* permanent, freed in lustre_free_lsi */
1789 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
1790 lmd->lmd_exclude_count);
1791 if (lmd->lmd_exclude) {
1792 memcpy(lmd->lmd_exclude, exclude_list,
1793 sizeof(index) * lmd->lmd_exclude_count);
1796 lmd->lmd_exclude_count = 0;
1799 OBD_FREE(exclude_list, sizeof(index) * devmax);
1803 /* mount -v -t lustre uml1:uml2:/lustre-client /mnt/lustre */
1804 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1806 char *s1, *s2, *devname = NULL;
1807 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1813 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1814 "/sbin/mount.lustre is installed.\n");
1818 /* Options should be a string - try to detect old lmd data */
1819 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1820 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1821 "/sbin/mount.lustre. Please install "
1822 "version %s\n", LUSTRE_VERSION_STRING);
1825 lmd->lmd_magic = LMD_MAGIC;
1827 /* Set default flags here */
1832 /* Skip whitespace and extra commas */
1833 while (*s1 == ' ' || *s1 == ',')
1836 /* Client options are parsed in ll_options: eg. flock,
1839 /* Parse non-ldiskfs options here. Rather than modifying
1840 ldiskfs, we just zero these out here */
1841 if (strncmp(s1, "abort_recov", 11) == 0) {
1842 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1844 } else if (strncmp(s1, "nosvc", 5) == 0) {
1845 lmd->lmd_flags |= LMD_FLG_NOSVC;
1847 } else if (strncmp(s1, "nomgs", 5) == 0) {
1848 lmd->lmd_flags |= LMD_FLG_NOMGS;
1850 /* ost exclusion list */
1851 } else if (strncmp(s1, "exclude=", 8) == 0) {
1852 rc = lmd_make_exclusion(lmd, s1 + 7);
1857 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1858 end of the options. */
1859 else if (strncmp(s1, "device=", 7) == 0) {
1861 /* terminate options right before device. device
1862 must be the last one. */
1868 s2 = strchr(s1, ',');
1876 memmove(s1, s2, strlen(s2) + 1);
1882 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1883 "(need mount option 'device=...')\n");
1887 s1 = strstr(devname, ":/");
1890 lmd->lmd_flags = LMD_FLG_CLIENT;
1891 /* Remove leading /s from fsname */
1892 while (*++s1 == '/') ;
1893 /* Freed in lustre_free_lsi */
1894 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1895 if (!lmd->lmd_profile)
1897 sprintf(lmd->lmd_profile, "%s-client", s1);
1900 /* Freed in lustre_free_lsi */
1901 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1904 strcpy(lmd->lmd_dev, devname);
1906 /* Save mount options */
1907 s1 = options + strlen(options) - 1;
1908 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1910 if (*options != 0) {
1911 /* Freed in lustre_free_lsi */
1912 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1915 strcpy(lmd->lmd_opts, options);
1918 lmd->lmd_magic = LMD_MAGIC;
1923 CERROR("Bad mount options %s\n", options);
1929 int lustre_fill_super(struct super_block *sb, void *data, int silent)
1931 struct lustre_mount_data *lmd;
1932 struct lustre_sb_info *lsi;
1936 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1938 lsi = lustre_init_lsi(sb);
1943 /* Figure out the lmd from the mount options */
1944 if (lmd_parse((char *)data, lmd)) {
1946 GOTO(out, rc = -EINVAL);
1949 if (lmd_is_client(lmd)) {
1950 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1951 if (!client_fill_super) {
1952 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1953 "client mount! Is the 'lustre' "
1954 "module loaded?\n");
1958 rc = lustre_start_mgc(sb);
1963 /* Connect and start */
1964 /* (should always be ll_fill_super) */
1965 rc = (*client_fill_super)(sb);
1966 /* c_f_s will call lustre_common_put_super on failure */
1969 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1970 lsi->lsi_flags |= LSI_SERVER;
1971 rc = server_fill_super(sb);
1972 /* s_f_s calls lustre_start_mgc after the mount because we need
1973 the MGS nids which are stored on disk. Plus, we may
1974 need to start the MGS first. */
1975 /* s_f_s will call server_put_super on failure */
1978 /* If error happens in fill_super() call, @lsi will be killed there.
1979 * This is why we do not put it here. */
1983 CERROR("Unable to mount %s (%d)\n",
1984 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1986 CDEBUG(D_SUPER, "Mount %s complete\n",
1993 /* We can't call ll_fill_super by name because it lives in a module that
1994 must be loaded after this one. */
1995 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb))
1997 client_fill_super = cfs;
2000 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
2002 kill_super_cb = cfs;
2005 /***************** FS registration ******************/
2007 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))
2008 struct super_block * lustre_get_sb(struct file_system_type *fs_type,
2009 int flags, const char *devname, void * data)
2011 return get_sb_nodev(fs_type, flags, data, lustre_fill_super);
2014 int lustre_get_sb(struct file_system_type *fs_type,
2015 int flags, const char *devname, void * data,
2016 struct vfsmount *mnt)
2018 return get_sb_nodev(fs_type, flags, data, lustre_fill_super, mnt);
2022 void lustre_kill_super(struct super_block *sb)
2024 struct lustre_sb_info *lsi = s2lsi(sb);
2026 if (kill_super_cb && lsi && !(lsi->lsi_flags & LSI_SERVER))
2027 (*kill_super_cb)(sb);
2029 kill_anon_super(sb);
2032 struct file_system_type lustre_fs_type = {
2033 .owner = THIS_MODULE,
2035 .get_sb = lustre_get_sb,
2036 .kill_sb = lustre_kill_super,
2037 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
2038 LL_RENAME_DOES_D_MOVE,
2041 int lustre_register_fs(void)
2043 return register_filesystem(&lustre_fs_type);
2046 int lustre_unregister_fs(void)
2048 return unregister_filesystem(&lustre_fs_type);
2051 EXPORT_SYMBOL(lustre_register_client_fill_super);
2052 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2053 EXPORT_SYMBOL(lustre_common_put_super);
2054 EXPORT_SYMBOL(lustre_process_log);
2055 EXPORT_SYMBOL(lustre_end_log);
2056 EXPORT_SYMBOL(server_get_mount);
2057 EXPORT_SYMBOL(server_get_mount_2);
2058 EXPORT_SYMBOL(server_put_mount);
2059 EXPORT_SYMBOL(server_put_mount_2);
2060 EXPORT_SYMBOL(server_register_target);
2061 EXPORT_SYMBOL(server_name2index);
2062 EXPORT_SYMBOL(server_mti_print);
2063 EXPORT_SYMBOL(do_lcfg);