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, *mgssec;
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 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
620 mutex_down(&mgc_start_lock);
622 obd = class_name2obd(mgcname);
623 if (obd && !obd->obd_stopping) {
624 rc = obd_set_info_async(obd->obd_self_export,
625 strlen(KEY_MGSSEC), KEY_MGSSEC,
626 strlen(mgssec), mgssec, NULL);
630 /* Re-using an existing MGC */
631 atomic_inc(&obd->u.cli.cl_mgc_refcount);
634 /* If we are restarting the MGS, don't try to keep the MGC's
635 old connection, or registration will fail. */
636 if ((lsi->lsi_flags & LSI_SERVER) && IS_MGS(lsi->lsi_ldd)) {
637 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
641 /* Try all connections, but only once (again).
642 We don't want to block another target from starting
643 (using its local copy of the log), but we do want to connect
644 if at all possible. */
646 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
647 rc = obd_set_info_async(obd->obd_self_export,
648 sizeof(KEY_INIT_RECOV_BACKUP),
649 KEY_INIT_RECOV_BACKUP,
650 sizeof(recov_bk), &recov_bk, NULL);
654 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
656 /* Add the primary nids for the MGS */
658 sprintf(niduuid, "%s_%x", mgcname, i);
659 if (lsi->lsi_flags & LSI_SERVER) {
660 ptr = lsi->lsi_ldd->ldd_params;
661 if (IS_MGS(lsi->lsi_ldd)) {
662 /* Use local nids (including LO) */
663 lnet_process_id_t id;
664 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
665 rc = do_lcfg(mgcname, id.nid,
666 LCFG_ADD_UUID, niduuid, 0,0,0);
669 /* Use mgsnode= nids */
670 if (class_find_param(ptr, PARAM_MGSNODE, &ptr) != 0) {
671 CERROR("No MGS nids given.\n");
672 GOTO(out_free, rc = -EINVAL);
674 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
675 rc = do_lcfg(mgcname, nid,
676 LCFG_ADD_UUID, niduuid, 0,0,0);
680 } else { /* client */
681 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
682 ptr = lsi->lsi_lmd->lmd_dev;
683 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
684 rc = do_lcfg(mgcname, nid,
685 LCFG_ADD_UUID, niduuid, 0,0,0);
687 /* Stop at the first failover nid */
693 CERROR("No valid MGS nids found.\n");
694 GOTO(out_free, rc = -EINVAL);
696 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
698 /* Random uuid for MGC allows easier reconnects */
700 ll_generate_random_uuid(uuidc);
701 class_uuid_unparse(uuidc, uuid);
704 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
705 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
711 /* Add any failover MGS nids */
713 while ((*ptr == ':' ||
714 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0)) {
715 /* New failover node */
716 sprintf(niduuid, "%s_%x", mgcname, i);
718 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
720 rc = do_lcfg(mgcname, nid,
721 LCFG_ADD_UUID, niduuid, 0,0,0);
726 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
734 lsi->lsi_lmd->lmd_mgs_failnodes = i;
736 obd = class_name2obd(mgcname);
738 CERROR("Can't find mgcobd %s\n", mgcname);
739 GOTO(out_free, rc = -ENOTCONN);
742 rc = obd_set_info_async(obd->obd_self_export,
743 strlen(KEY_MGSSEC), KEY_MGSSEC,
744 strlen(mgssec), mgssec, NULL);
748 /* Keep a refcount of servers/clients who started with "mount",
749 so we know when we can get rid of the mgc. */
750 atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
752 /* Try all connections, but only once. */
754 rc = obd_set_info_async(obd->obd_self_export,
755 sizeof(KEY_INIT_RECOV_BACKUP),
756 KEY_INIT_RECOV_BACKUP,
757 sizeof(recov_bk), &recov_bk, NULL);
760 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
761 /* We connect to the MGS at setup, and don't disconnect until cleanup */
764 GOTO(out, rc = -ENOMEM);
765 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID |
767 data->ocd_version = LUSTRE_VERSION_CODE;
768 rc = obd_connect(NULL, &mgc_conn, obd, &(obd->obd_uuid), data, NULL);
771 CERROR("connect failed %d\n", rc);
775 exp = class_conn2export(&mgc_conn);
776 obd->u.cli.cl_mgc_mgsexp = exp;
779 /* Keep the mgc info in the sb. Note that many lsi's can point
783 mutex_up(&mgc_start_lock);
786 OBD_FREE(mgcname, len);
788 OBD_FREE(niduuid, len + 2);
792 static int lustre_stop_mgc(struct super_block *sb)
794 struct lustre_sb_info *lsi = s2lsi(sb);
795 struct obd_device *obd;
796 char *niduuid = 0, *ptr = 0;
797 int i, rc = 0, len = 0;
807 mutex_down(&mgc_start_lock);
808 LASSERT(atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
809 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
810 /* This is not fatal, every client that stops
811 will call in here. */
812 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
813 atomic_read(&obd->u.cli.cl_mgc_refcount));
814 GOTO(out, rc = -EBUSY);
817 /* The MGC has no recoverable data in any case.
818 * force shotdown set in umount_begin */
819 obd->obd_no_recov = 1;
821 if (obd->u.cli.cl_mgc_mgsexp) {
822 /* An error is not fatal, if we are unable to send the
823 disconnect mgs ping evictor cleans up the export */
824 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
826 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
829 /* Save the obdname for cleaning the nid uuids, which are
831 len = strlen(obd->obd_name) + 6;
832 OBD_ALLOC(niduuid, len);
834 strcpy(niduuid, obd->obd_name);
835 ptr = niduuid + strlen(niduuid);
838 rc = class_manual_cleanup(obd);
842 /* Clean the nid uuids */
844 GOTO(out, rc = -ENOMEM);
846 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
847 sprintf(ptr, "_%x", i);
848 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
851 CERROR("del MDC UUID %s failed: rc = %d\n",
856 OBD_FREE(niduuid, len);
858 /* class_import_put will get rid of the additional connections */
859 mutex_up(&mgc_start_lock);
863 /* Since there's only one mgc per node, we have to change it's fs to get
864 access to the right disk. */
865 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
867 struct lustre_sb_info *lsi = s2lsi(sb);
871 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
873 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
874 rc = obd_set_info_async(mgc->obd_self_export,
875 sizeof(KEY_SET_FS), KEY_SET_FS,
876 sizeof(*sb), sb, NULL);
878 CERROR("can't set_fs %d\n", rc);
884 static int server_mgc_clear_fs(struct obd_device *mgc)
889 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
891 rc = obd_set_info_async(mgc->obd_self_export,
892 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
897 DECLARE_MUTEX(server_start_lock);
899 /* Stop MDS/OSS if nobody is using them */
900 static int server_stop_servers(int lddflags, int lsiflags)
902 struct obd_device *obd = NULL;
903 struct obd_type *type = NULL;
907 mutex_down(&server_start_lock);
909 /* Either an MDT or an OST or neither */
910 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
911 if ((lddflags & LDD_F_SV_TYPE_MDT) &&
912 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
913 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
914 type = class_search_type(LUSTRE_MDS_NAME);
916 /* if this was an OST, and there are no more OST's, clean up the OSS */
917 if ((lddflags & LDD_F_SV_TYPE_OST) &&
918 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
919 type = class_search_type(LUSTRE_OST_NAME);
922 if (obd && (!type || !type->typ_refcnt)) {
925 /* obd_fail doesn't mean much on a server obd */
926 err = class_manual_cleanup(obd);
931 mutex_up(&server_start_lock);
936 int server_mti_print(char *title, struct mgs_target_info *mti)
938 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
939 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
940 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
941 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
942 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
943 mti->mti_config_ver, mti->mti_flags);
947 static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
949 struct lustre_sb_info *lsi = s2lsi(sb);
950 struct lustre_disk_data *ldd = lsi->lsi_ldd;
951 lnet_process_id_t id;
955 if (!(lsi->lsi_flags & LSI_SERVER))
958 strncpy(mti->mti_fsname, ldd->ldd_fsname,
959 sizeof(mti->mti_fsname));
960 strncpy(mti->mti_svname, ldd->ldd_svname,
961 sizeof(mti->mti_svname));
963 mti->mti_nid_count = 0;
964 while (LNetGetId(i++, &id) != -ENOENT) {
965 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
967 mti->mti_nids[mti->mti_nid_count] = id.nid;
968 mti->mti_nid_count++;
969 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
970 CWARN("Only using first %d nids for %s\n",
971 mti->mti_nid_count, mti->mti_svname);
976 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
977 mti->mti_config_ver = 0;
978 mti->mti_flags = ldd->ldd_flags;
979 mti->mti_stripe_index = ldd->ldd_svindex;
980 memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid));
981 if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) {
982 CERROR("params too big for mti\n");
985 memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params));
989 /* Register an old or new target with the MGS. If needed MGS will construct
990 startup logs and assign index */
991 int server_register_target(struct super_block *sb)
993 struct lustre_sb_info *lsi = s2lsi(sb);
994 struct obd_device *mgc = lsi->lsi_mgc;
995 struct lustre_disk_data *ldd = lsi->lsi_ldd;
996 struct mgs_target_info *mti = NULL;
1002 if (!(lsi->lsi_flags & LSI_SERVER))
1008 rc = server_sb2mti(sb, mti);
1012 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1013 mti->mti_svname, mti->mti_fsname,
1014 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1017 /* Register the target */
1018 /* FIXME use mgc_process_config instead */
1019 rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
1020 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
1021 sizeof(*mti), mti, NULL);
1025 /* Always update our flags */
1026 ldd->ldd_flags = mti->mti_flags & ~LDD_F_REWRITE_LDD;
1028 /* If this flag is set, it means the MGS wants us to change our
1029 on-disk data. (So far this means just the index.) */
1030 if (mti->mti_flags & LDD_F_REWRITE_LDD) {
1033 CDEBUG(D_MOUNT, "Changing on-disk index from %#x to %#x "
1034 "for %s\n", ldd->ldd_svindex, mti->mti_stripe_index,
1036 ldd->ldd_svindex = mti->mti_stripe_index;
1037 strncpy(ldd->ldd_svname, mti->mti_svname,
1038 sizeof(ldd->ldd_svname));
1039 /* or ldd_make_sv_name(ldd); */
1040 ldd_write(&mgc->obd_lvfs_ctxt, ldd);
1041 err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb,
1044 CERROR("Label set error %d\n", err);
1045 label = fsfilt_get_label(mgc, lsi->lsi_srv_mnt->mnt_sb);
1047 CDEBUG(D_MOUNT, "Disk label changed to %s\n", label);
1049 /* Flush the new ldd to disk */
1050 fsfilt_sync(mgc, lsi->lsi_srv_mnt->mnt_sb);
1060 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1062 struct obd_device *obd;
1063 struct lustre_sb_info *lsi = s2lsi(sb);
1064 struct config_llog_instance cfg;
1068 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname);
1071 /* If we're an MDT, make sure the global MDS is running */
1072 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1073 /* make sure the MDS is started */
1074 mutex_down(&server_start_lock);
1075 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1077 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1078 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1080 LUSTRE_MDS_OBDNAME"_uuid",
1083 mutex_up(&server_start_lock);
1084 CERROR("failed to start MDS: %d\n", rc);
1088 mutex_up(&server_start_lock);
1092 /* If we're an OST, make sure the global OSS is running */
1093 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
1094 /* make sure OSS is started */
1095 mutex_down(&server_start_lock);
1096 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1098 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1100 LUSTRE_OSS_OBDNAME"_uuid",
1103 mutex_up(&server_start_lock);
1104 CERROR("failed to start OSS: %d\n", rc);
1108 mutex_up(&server_start_lock);
1111 /* Set the mgc fs to our server disk. This allows the MGC
1112 to read and write configs locally. */
1113 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1117 /* Register with MGS */
1118 rc = server_register_target(sb);
1119 if (rc && (lsi->lsi_ldd->ldd_flags &
1120 (LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_UPGRADE14))){
1121 CERROR("Required registration failed for %s: %d\n",
1122 lsi->lsi_ldd->ldd_svname, rc);
1124 LCONSOLE_ERROR_MSG(0x15f, "Communication error with "
1125 "the MGS. Is the MGS running?\n");
1129 if (rc == -EINVAL) {
1130 LCONSOLE_ERROR_MSG(0x160, "The MGS is refusing to allow this "
1131 "server (%s) to start. Please see messages"
1132 " on the MGS node.\n",
1133 lsi->lsi_ldd->ldd_svname);
1136 /* non-fatal error of registeration with MGS */
1138 CDEBUG(D_MOUNT, "Cannot register with MGS: %d\n", rc);
1140 /* Let the target look up the mount using the target's name
1141 (we can't pass the sb or mnt through class_process_config.) */
1142 rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
1146 /* Start targets using the llog named for the target */
1147 memset(&cfg, 0, sizeof(cfg));
1148 rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg);
1150 CERROR("failed to start server %s: %d\n",
1151 lsi->lsi_ldd->ldd_svname, rc);
1152 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1157 /* Release the mgc fs for others to use */
1158 server_mgc_clear_fs(lsi->lsi_mgc);
1161 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1163 CERROR("no server named %s was started\n",
1164 lsi->lsi_ldd->ldd_svname);
1168 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1169 (OBP(obd, iocontrol))) {
1170 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1171 obd->obd_self_export, 0, NULL, NULL);
1174 /* log has been fully processed */
1175 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1181 /***************** lustre superblock **************/
1183 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1185 struct lustre_sb_info *lsi;
1191 OBD_ALLOC_PTR(lsi->lsi_lmd);
1192 if (!lsi->lsi_lmd) {
1197 lsi->lsi_lmd->lmd_exclude_count = 0;
1198 s2lsi_nocast(sb) = lsi;
1199 /* we take 1 extra ref for our setup */
1200 atomic_set(&lsi->lsi_mounts, 1);
1202 /* Default umount style */
1203 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1208 static int lustre_free_lsi(struct super_block *sb)
1210 struct lustre_sb_info *lsi = s2lsi(sb);
1213 LASSERT(lsi != NULL);
1214 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
1216 /* someone didn't call server_put_mount. */
1217 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
1219 if (lsi->lsi_ldd != NULL)
1220 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
1222 if (lsi->lsi_lmd != NULL) {
1223 if (lsi->lsi_lmd->lmd_dev != NULL)
1224 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1225 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1226 if (lsi->lsi_lmd->lmd_profile != NULL)
1227 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1228 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1229 if (lsi->lsi_lmd->lmd_mgssec != NULL)
1230 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
1231 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
1232 if (lsi->lsi_lmd->lmd_opts != NULL)
1233 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1234 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1235 if (lsi->lsi_lmd->lmd_exclude_count)
1236 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1237 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1238 lsi->lsi_lmd->lmd_exclude_count);
1239 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1242 LASSERT(lsi->lsi_llsbi == NULL);
1243 OBD_FREE(lsi, sizeof(*lsi));
1244 s2lsi_nocast(sb) = NULL;
1249 /* The lsi has one reference for every server that is using the disk -
1250 e.g. MDT, MGS, and potentially MGC */
1251 static int lustre_put_lsi(struct super_block *sb)
1253 struct lustre_sb_info *lsi = s2lsi(sb);
1256 LASSERT(lsi != NULL);
1258 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
1259 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
1260 lustre_free_lsi(sb);
1266 /*************** server mount ******************/
1268 /* Kernel mount using mount options in MOUNT_DATA_FILE */
1269 static struct vfsmount *server_kernel_mount(struct super_block *sb)
1271 struct lvfs_run_ctxt mount_ctxt;
1272 struct lustre_sb_info *lsi = s2lsi(sb);
1273 struct lustre_disk_data *ldd;
1274 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1275 struct vfsmount *mnt;
1276 char *options = NULL;
1277 unsigned long page, s_flags;
1278 struct page *__page;
1282 OBD_ALLOC(ldd, sizeof(*ldd));
1284 RETURN(ERR_PTR(-ENOMEM));
1286 /* In the past, we have always used flags = 0.
1287 Note ext3/ldiskfs can't be mounted ro. */
1288 s_flags = sb->s_flags;
1290 /* allocate memory for options */
1291 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
1293 GOTO(out_free, rc = -ENOMEM);
1294 page = (unsigned long)cfs_page_address(__page);
1295 options = (char *)page;
1296 memset(options, 0, CFS_PAGE_SIZE);
1298 /* mount-line options must be added for pre-mount because it may
1299 * contain mount options such as journal_dev which are required
1300 * to mount successfuly the underlying filesystem */
1301 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0))
1302 strncat(options, lmd->lmd_opts, CFS_PAGE_SIZE - 1);
1304 /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */
1305 CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev);
1306 mnt = ll_kern_mount("ldiskfs", s_flags, lmd->lmd_dev, (void *)options);
1309 CERROR("premount %s:%#lx ldiskfs failed: %d "
1310 "Is the ldiskfs module available?\n",
1311 lmd->lmd_dev, s_flags, rc );
1315 OBD_SET_CTXT_MAGIC(&mount_ctxt);
1316 mount_ctxt.pwdmnt = mnt;
1317 mount_ctxt.pwd = mnt->mnt_root;
1318 mount_ctxt.fs = get_ds();
1320 rc = ldd_parse(&mount_ctxt, ldd);
1324 CERROR("premount parse options failed: rc = %d\n", rc);
1328 /* Done with our pre-mount, now do the real mount. */
1330 /* Glom up mount options */
1331 memset(options, 0, CFS_PAGE_SIZE);
1332 strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2);
1334 /* Add in any mount-line options */
1335 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) {
1336 int len = CFS_PAGE_SIZE - strlen(options) - 2;
1338 strcat(options, ",");
1339 strncat(options, lmd->lmd_opts, len);
1342 /* Special permanent mount flags */
1344 s_flags |= MS_NOATIME | MS_NODIRATIME;
1346 CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
1347 MT_STR(ldd), lmd->lmd_dev, options);
1348 mnt = ll_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev,
1352 CERROR("ll_kern_mount failed: rc = %d\n", rc);
1356 OBD_PAGE_FREE(__page);
1357 lsi->lsi_ldd = ldd; /* freed at lsi cleanup */
1358 CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
1363 OBD_PAGE_FREE(__page);
1364 OBD_FREE(ldd, sizeof(*ldd));
1365 lsi->lsi_ldd = NULL;
1366 RETURN(ERR_PTR(rc));
1369 static void server_wait_finished(struct vfsmount *mnt)
1371 wait_queue_head_t waitq;
1372 struct l_wait_info lwi;
1375 init_waitqueue_head(&waitq);
1377 while ((atomic_read(&mnt->mnt_count) > 1) && (retries > 0)) {
1378 LCONSOLE_WARN("Mount still busy with %d refs, waiting for "
1380 atomic_read(&mnt->mnt_count), retries);
1382 /* Wait for a bit */
1384 lwi = LWI_TIMEOUT(5 * HZ, NULL, NULL);
1385 l_wait_event(waitq, 0, &lwi);
1387 if (atomic_read(&mnt->mnt_count) > 1) {
1388 CERROR("Mount %p is still busy (%d refs), giving up.\n",
1389 mnt, atomic_read(&mnt->mnt_count));
1393 static void server_put_super(struct super_block *sb)
1395 struct lustre_sb_info *lsi = s2lsi(sb);
1396 struct obd_device *obd;
1397 struct vfsmount *mnt = lsi->lsi_srv_mnt;
1398 char *tmpname, *extraname = NULL;
1400 int lddflags = lsi->lsi_ldd->ldd_flags;
1401 int lsiflags = lsi->lsi_flags;
1404 LASSERT(lsiflags & LSI_SERVER);
1406 tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1;
1407 OBD_ALLOC(tmpname, tmpname_sz);
1408 memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
1409 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1411 /* Stop the target */
1412 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1413 (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd))) {
1414 struct lustre_profile *lprof = NULL;
1416 /* tell the mgc to drop the config log */
1417 lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL);
1419 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1420 If there are any setup/cleanup errors, save the lov
1421 name for safety cleanup later. */
1422 lprof = class_get_profile(lsi->lsi_ldd->ldd_svname);
1423 if (lprof && lprof->lp_dt) {
1424 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1425 strcpy(extraname, lprof->lp_dt);
1428 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1430 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1431 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1433 /* We can't seem to give an error return code
1434 * to .put_super, so we better make sure we clean up! */
1436 class_manual_cleanup(obd);
1438 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
1439 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1443 /* If they wanted the mgs to stop separately from the mdt, they
1444 should have put it on a different device. */
1445 if (IS_MGS(lsi->lsi_ldd)) {
1446 /* if MDS start with --nomgs, don't stop MGS then */
1447 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1448 server_stop_mgs(sb);
1451 /* Clean the mgc and sb */
1452 lustre_common_put_super(sb);
1454 /* Wait for the targets to really clean up - can't exit (and let the
1455 sb get destroyed) while the mount is still in use */
1456 server_wait_finished(mnt);
1458 /* drop the One True Mount */
1461 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1462 until the target is really gone so that our type refcount check
1464 server_stop_servers(lddflags, lsiflags);
1466 /* In case of startup or cleanup err, stop related obds */
1468 obd = class_name2obd(extraname);
1470 CWARN("Cleaning orphaned obd %s\n", extraname);
1472 class_manual_cleanup(obd);
1474 OBD_FREE(extraname, strlen(extraname) + 1);
1477 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1478 OBD_FREE(tmpname, tmpname_sz);
1482 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1483 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1485 struct super_block *sb = vfsmnt->mnt_sb;
1487 static void server_umount_begin(struct super_block *sb)
1490 struct lustre_sb_info *lsi = s2lsi(sb);
1493 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1494 if (!(flags & MNT_FORCE)) {
1500 CDEBUG(D_MOUNT, "umount -f\n");
1501 /* umount = failover
1503 no third way to do non-force, non-failover */
1504 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1505 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1509 #ifndef HAVE_STATFS_DENTRY_PARAM
1510 static int server_statfs (struct super_block *sb, struct kstatfs *buf)
1513 static int server_statfs (struct dentry *dentry, struct kstatfs *buf)
1515 struct super_block *sb = dentry->d_sb;
1517 struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt;
1520 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) {
1521 #ifdef HAVE_STATFS_DENTRY_PARAM
1522 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf);
1524 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_sb, buf);
1527 buf->f_type = sb->s_magic;
1533 buf->f_type = sb->s_magic;
1534 buf->f_bsize = sb->s_blocksize;
1540 buf->f_namelen = NAME_MAX;
1544 static struct super_operations server_ops =
1546 .put_super = server_put_super,
1547 .umount_begin = server_umount_begin, /* umount -f */
1548 .statfs = server_statfs,
1551 #define log2(n) ffz(~(n))
1552 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1554 static int server_fill_super_common(struct super_block *sb)
1556 struct inode *root = 0;
1559 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1561 sb->s_blocksize = 4096;
1562 sb->s_blocksize_bits = log2(sb->s_blocksize);
1563 sb->s_magic = LUSTRE_SUPER_MAGIC;
1564 sb->s_maxbytes = 0; //PAGE_CACHE_MAXBYTES;
1565 sb->s_flags |= MS_RDONLY;
1566 sb->s_op = &server_ops;
1568 root = new_inode(sb);
1570 CERROR("Can't make root inode\n");
1574 /* returns -EIO for every operation */
1575 /* make_bad_inode(root); -- badness - can't umount */
1576 /* apparently we need to be a directory for the mount to finish */
1577 root->i_mode = S_IFDIR;
1579 sb->s_root = d_alloc_root(root);
1581 CERROR("Can't make root dentry\n");
1589 static int server_fill_super(struct super_block *sb)
1591 struct lustre_sb_info *lsi = s2lsi(sb);
1592 struct vfsmount *mnt;
1596 /* the One True Mount */
1597 mnt = server_kernel_mount(sb);
1600 CERROR("Unable to mount device %s: %d\n",
1601 lsi->lsi_lmd->lmd_dev, rc);
1605 lsi->lsi_srv_mnt = mnt;
1607 LASSERT(lsi->lsi_ldd);
1608 CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n",
1609 lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname,
1610 lsi->lsi_lmd->lmd_dev);
1612 if (class_name2obd(lsi->lsi_ldd->ldd_svname)) {
1613 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1614 "running. Double-mount may have compromised"
1615 " the disk journal.\n",
1616 lsi->lsi_ldd->ldd_svname);
1622 /* Start MGS before MGC */
1623 if (IS_MGS(lsi->lsi_ldd) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
1624 rc = server_start_mgs(sb);
1629 rc = lustre_start_mgc(sb);
1633 /* Set up all obd devices for service */
1634 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1635 (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) {
1636 rc = server_start_targets(sb, mnt);
1638 CERROR("Unable to start targets: %d\n", rc);
1641 /* FIXME overmount client here,
1642 or can we just start a client log and client_fill_super on this sb?
1643 We need to make sure server_put_super gets called too - ll_put_super
1644 calls lustre_common_put_super; check there for LSI_SERVER flag,
1646 Probably should start client from new thread so we can return.
1647 Client will not finish until all servers are connected.
1648 Note - MGS-only server does NOT get a client, since there is no
1649 lustre fs associated - the MGS is for all lustre fs's */
1652 rc = server_fill_super_common(sb);
1656 LCONSOLE_WARN("Server %s on device %s has started\n",
1657 lsi->lsi_ldd->ldd_svname, lsi->lsi_lmd->lmd_dev);
1661 /* We jump here in case of failure while starting targets or MGS.
1662 * In this case we can't just put @mnt and have to do real cleanup
1663 * with stoping targets, etc. */
1664 server_put_super(sb);
1668 /* Get the index from the obd name.
1669 rc = server type, or
1671 if endptr isn't NULL it is set to end of name */
1672 int server_name2index(char *svname, __u32 *idx, char **endptr)
1674 unsigned long index;
1676 char *dash = strchr(svname, '-');
1680 if (strncmp(dash + 1, "MDT", 3) == 0)
1681 rc = LDD_F_SV_TYPE_MDT;
1682 else if (strncmp(dash + 1, "OST", 3) == 0)
1683 rc = LDD_F_SV_TYPE_OST;
1687 index = simple_strtoul(dash + 4, endptr, 16);
1692 /*************** mount common betweeen server and client ***************/
1695 int lustre_common_put_super(struct super_block *sb)
1700 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1702 /* Drop a ref to the MGC */
1703 rc = lustre_stop_mgc(sb);
1704 if (rc && (rc != -ENOENT)) {
1706 CERROR("Can't stop MGC: %d\n", rc);
1709 /* BUSY just means that there's some other obd that
1710 needs the mgc. Let him clean it up. */
1711 CDEBUG(D_MOUNT, "MGC still in use\n");
1713 /* Drop a ref to the mounted disk */
1720 static void lmd_print(struct lustre_mount_data *lmd)
1724 PRINT_CMD(PRINT_MASK, " mount data:\n");
1725 if (lmd_is_client(lmd))
1726 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
1727 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
1728 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
1730 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
1731 for (i = 0; i < lmd->lmd_exclude_count; i++) {
1732 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
1733 lmd->lmd_exclude[i]);
1738 /* Is this server on the exclusion list */
1739 int lustre_check_exclusion(struct super_block *sb, char *svname)
1741 struct lustre_sb_info *lsi = s2lsi(sb);
1742 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1747 rc = server_name2index(svname, &index, NULL);
1748 if (rc != LDD_F_SV_TYPE_OST)
1749 /* Only exclude OSTs */
1752 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
1753 index, lmd->lmd_exclude_count, lmd->lmd_dev);
1755 for(i = 0; i < lmd->lmd_exclude_count; i++) {
1756 if (index == lmd->lmd_exclude[i]) {
1757 CWARN("Excluding %s (on exclusion list)\n", svname);
1764 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
1765 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
1767 char *s1 = ptr, *s2;
1768 __u32 index, *exclude_list;
1772 /* The shortest an ost name can be is 8 chars: -OST0000.
1773 We don't actually know the fsname at this time, so in fact
1774 a user could specify any fsname. */
1775 devmax = strlen(ptr) / 8 + 1;
1777 /* temp storage until we figure out how many we have */
1778 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
1782 /* we enter this fn pointing at the '=' */
1783 while (*s1 && *s1 != ' ' && *s1 != ',') {
1785 rc = server_name2index(s1, &index, &s2);
1787 CERROR("Can't parse server name '%s'\n", s1);
1790 if (rc == LDD_F_SV_TYPE_OST)
1791 exclude_list[lmd->lmd_exclude_count++] = index;
1793 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
1795 /* now we are pointing at ':' (next exclude)
1796 or ',' (end of excludes) */
1797 if (lmd->lmd_exclude_count >= devmax)
1800 if (rc >= 0) /* non-err */
1803 if (lmd->lmd_exclude_count) {
1804 /* permanent, freed in lustre_free_lsi */
1805 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
1806 lmd->lmd_exclude_count);
1807 if (lmd->lmd_exclude) {
1808 memcpy(lmd->lmd_exclude, exclude_list,
1809 sizeof(index) * lmd->lmd_exclude_count);
1812 lmd->lmd_exclude_count = 0;
1815 OBD_FREE(exclude_list, sizeof(index) * devmax);
1819 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
1824 if (lmd->lmd_mgssec != NULL) {
1825 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
1826 lmd->lmd_mgssec = NULL;
1829 tail = strchr(ptr, ',');
1831 length = strlen(ptr);
1833 length = tail - ptr;
1835 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
1836 if (lmd->lmd_mgssec == NULL)
1839 memcpy(lmd->lmd_mgssec, ptr, length);
1840 lmd->lmd_mgssec[length] = '\0';
1844 /* mount -v -t lustre uml1:uml2:/lustre-client /mnt/lustre */
1845 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1847 char *s1, *s2, *devname = NULL;
1848 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1854 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1855 "/sbin/mount.lustre is installed.\n");
1859 /* Options should be a string - try to detect old lmd data */
1860 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1861 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1862 "/sbin/mount.lustre. Please install "
1863 "version %s\n", LUSTRE_VERSION_STRING);
1866 lmd->lmd_magic = LMD_MAGIC;
1868 /* Set default flags here */
1873 /* Skip whitespace and extra commas */
1874 while (*s1 == ' ' || *s1 == ',')
1877 /* Client options are parsed in ll_options: eg. flock,
1880 /* Parse non-ldiskfs options here. Rather than modifying
1881 ldiskfs, we just zero these out here */
1882 if (strncmp(s1, "abort_recov", 11) == 0) {
1883 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1885 } else if (strncmp(s1, "nosvc", 5) == 0) {
1886 lmd->lmd_flags |= LMD_FLG_NOSVC;
1888 } else if (strncmp(s1, "nomgs", 5) == 0) {
1889 lmd->lmd_flags |= LMD_FLG_NOMGS;
1891 } else if (strncmp(s1, "mgssec=", 7) == 0) {
1892 rc = lmd_parse_mgssec(lmd, s1 + 7);
1896 /* ost exclusion list */
1897 } else if (strncmp(s1, "exclude=", 8) == 0) {
1898 rc = lmd_make_exclusion(lmd, s1 + 7);
1903 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1904 end of the options. */
1905 else if (strncmp(s1, "device=", 7) == 0) {
1907 /* terminate options right before device. device
1908 must be the last one. */
1914 s2 = strchr(s1, ',');
1922 memmove(s1, s2, strlen(s2) + 1);
1928 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1929 "(need mount option 'device=...')\n");
1933 s1 = strstr(devname, ":/");
1936 lmd->lmd_flags = LMD_FLG_CLIENT;
1937 /* Remove leading /s from fsname */
1938 while (*++s1 == '/') ;
1939 /* Freed in lustre_free_lsi */
1940 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1941 if (!lmd->lmd_profile)
1943 sprintf(lmd->lmd_profile, "%s-client", s1);
1946 /* Freed in lustre_free_lsi */
1947 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1950 strcpy(lmd->lmd_dev, devname);
1952 /* Save mount options */
1953 s1 = options + strlen(options) - 1;
1954 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1956 if (*options != 0) {
1957 /* Freed in lustre_free_lsi */
1958 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1961 strcpy(lmd->lmd_opts, options);
1964 lmd->lmd_magic = LMD_MAGIC;
1969 CERROR("Bad mount options %s\n", options);
1975 int lustre_fill_super(struct super_block *sb, void *data, int silent)
1977 struct lustre_mount_data *lmd;
1978 struct lustre_sb_info *lsi;
1982 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1984 lsi = lustre_init_lsi(sb);
1990 * Disable lockdep during mount, because mount locking patterns are
1995 /* Figure out the lmd from the mount options */
1996 if (lmd_parse((char *)data, lmd)) {
1998 GOTO(out, rc = -EINVAL);
2001 if (lmd_is_client(lmd)) {
2002 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
2003 if (!client_fill_super) {
2004 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
2005 "client mount! Is the 'lustre' "
2006 "module loaded?\n");
2010 rc = lustre_start_mgc(sb);
2015 /* Connect and start */
2016 /* (should always be ll_fill_super) */
2017 rc = (*client_fill_super)(sb);
2018 /* c_f_s will call lustre_common_put_super on failure */
2021 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
2022 lsi->lsi_flags |= LSI_SERVER;
2023 rc = server_fill_super(sb);
2024 /* s_f_s calls lustre_start_mgc after the mount because we need
2025 the MGS nids which are stored on disk. Plus, we may
2026 need to start the MGS first. */
2027 /* s_f_s will call server_put_super on failure */
2030 /* If error happens in fill_super() call, @lsi will be killed there.
2031 * This is why we do not put it here. */
2035 CERROR("Unable to mount %s (%d)\n",
2036 s2lsi(sb) ? lmd->lmd_dev : "", rc);
2038 CDEBUG(D_SUPER, "Mount %s complete\n",
2046 /* We can't call ll_fill_super by name because it lives in a module that
2047 must be loaded after this one. */
2048 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb))
2050 client_fill_super = cfs;
2053 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
2055 kill_super_cb = cfs;
2058 /***************** FS registration ******************/
2060 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))
2061 struct super_block * lustre_get_sb(struct file_system_type *fs_type,
2062 int flags, const char *devname, void * data)
2064 return get_sb_nodev(fs_type, flags, data, lustre_fill_super);
2067 int lustre_get_sb(struct file_system_type *fs_type,
2068 int flags, const char *devname, void * data,
2069 struct vfsmount *mnt)
2071 return get_sb_nodev(fs_type, flags, data, lustre_fill_super, mnt);
2075 void lustre_kill_super(struct super_block *sb)
2077 struct lustre_sb_info *lsi = s2lsi(sb);
2079 if (kill_super_cb && lsi && !(lsi->lsi_flags & LSI_SERVER))
2080 (*kill_super_cb)(sb);
2082 kill_anon_super(sb);
2085 struct file_system_type lustre_fs_type = {
2086 .owner = THIS_MODULE,
2088 .get_sb = lustre_get_sb,
2089 .kill_sb = lustre_kill_super,
2090 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
2091 LL_RENAME_DOES_D_MOVE,
2094 int lustre_register_fs(void)
2096 return register_filesystem(&lustre_fs_type);
2099 int lustre_unregister_fs(void)
2101 return unregister_filesystem(&lustre_fs_type);
2104 EXPORT_SYMBOL(lustre_register_client_fill_super);
2105 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2106 EXPORT_SYMBOL(lustre_common_put_super);
2107 EXPORT_SYMBOL(lustre_process_log);
2108 EXPORT_SYMBOL(lustre_end_log);
2109 EXPORT_SYMBOL(server_get_mount);
2110 EXPORT_SYMBOL(server_get_mount_2);
2111 EXPORT_SYMBOL(server_put_mount);
2112 EXPORT_SYMBOL(server_put_mount_2);
2113 EXPORT_SYMBOL(server_register_target);
2114 EXPORT_SYMBOL(server_name2index);
2115 EXPORT_SYMBOL(server_mti_print);
2116 EXPORT_SYMBOL(do_lcfg);