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 /* Do NOT call server_deregister_mount() here. This leads to
528 * inability cleanup cleanly and free lsi and other stuff when
529 * mgs calls server_put_mount() in error handling case. -umka */
533 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
534 "Is the 'mgs' module loaded?\n",
535 LUSTRE_MGS_OBDNAME, rc);
539 static int server_stop_mgs(struct super_block *sb)
541 struct obd_device *obd;
545 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
547 /* There better be only one MGS */
548 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
550 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
554 /* The MGS should always stop when we say so */
556 rc = class_manual_cleanup(obd);
560 DECLARE_MUTEX(mgc_start_lock);
562 /** Set up a mgc obd to process startup logs
564 * \param sb [in] super block of the mgc obd
566 * \retval 0 success, otherwise error code
568 static int lustre_start_mgc(struct super_block *sb)
570 struct lustre_handle mgc_conn = {0, };
571 struct obd_connect_data *data = NULL;
572 struct lustre_sb_info *lsi = s2lsi(sb);
573 struct obd_device *obd;
574 struct obd_export *exp;
575 struct obd_uuid *uuid;
578 char *mgcname, *niduuid, *mgssec;
581 int rc = 0, i = 0, j, len;
584 LASSERT(lsi->lsi_lmd);
586 /* Find the first non-lo MGS nid for our MGC name */
587 if (lsi->lsi_flags & LSI_SERVER) {
588 ptr = lsi->lsi_ldd->ldd_params;
589 /* Use mgsnode= nids */
590 if ((class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0) &&
591 (class_parse_nid(ptr, &nid, &ptr) == 0)) {
593 } else if (IS_MGS(lsi->lsi_ldd)) {
594 lnet_process_id_t id;
595 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
596 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
603 } else { /* client */
604 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
605 ptr = lsi->lsi_lmd->lmd_dev;
606 if (class_parse_nid(ptr, &nid, &ptr) == 0)
610 CERROR("No valid MGS nids found.\n");
614 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
615 OBD_ALLOC(mgcname, len);
616 OBD_ALLOC(niduuid, len + 2);
617 if (!mgcname || !niduuid)
618 GOTO(out_free, rc = -ENOMEM);
619 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
621 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
623 mutex_down(&mgc_start_lock);
625 obd = class_name2obd(mgcname);
626 if (obd && !obd->obd_stopping) {
627 rc = obd_set_info_async(obd->obd_self_export,
628 strlen(KEY_MGSSEC), KEY_MGSSEC,
629 strlen(mgssec), mgssec, NULL);
633 /* Re-using an existing MGC */
634 atomic_inc(&obd->u.cli.cl_mgc_refcount);
637 /* If we are restarting the MGS, don't try to keep the MGC's
638 old connection, or registration will fail. */
639 if ((lsi->lsi_flags & LSI_SERVER) && IS_MGS(lsi->lsi_ldd)) {
640 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
644 /* Try all connections, but only once (again).
645 We don't want to block another target from starting
646 (using its local copy of the log), but we do want to connect
647 if at all possible. */
649 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
650 rc = obd_set_info_async(obd->obd_self_export,
651 sizeof(KEY_INIT_RECOV_BACKUP),
652 KEY_INIT_RECOV_BACKUP,
653 sizeof(recov_bk), &recov_bk, NULL);
657 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
659 /* Add the primary nids for the MGS */
661 sprintf(niduuid, "%s_%x", mgcname, i);
662 if (lsi->lsi_flags & LSI_SERVER) {
663 ptr = lsi->lsi_ldd->ldd_params;
664 if (IS_MGS(lsi->lsi_ldd)) {
665 /* Use local nids (including LO) */
666 lnet_process_id_t id;
667 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
668 rc = do_lcfg(mgcname, id.nid,
669 LCFG_ADD_UUID, niduuid, 0,0,0);
672 /* Use mgsnode= nids */
673 if (class_find_param(ptr, PARAM_MGSNODE, &ptr) != 0) {
674 CERROR("No MGS nids given.\n");
675 GOTO(out_free, rc = -EINVAL);
677 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
678 rc = do_lcfg(mgcname, nid,
679 LCFG_ADD_UUID, niduuid, 0,0,0);
683 } else { /* client */
684 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
685 ptr = lsi->lsi_lmd->lmd_dev;
686 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
687 rc = do_lcfg(mgcname, nid,
688 LCFG_ADD_UUID, niduuid, 0,0,0);
690 /* Stop at the first failover nid */
696 CERROR("No valid MGS nids found.\n");
697 GOTO(out_free, rc = -EINVAL);
699 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
701 /* Random uuid for MGC allows easier reconnects */
703 ll_generate_random_uuid(uuidc);
704 class_uuid_unparse(uuidc, uuid);
707 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
708 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
714 /* Add any failover MGS nids */
716 while ((*ptr == ':' ||
717 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0)) {
718 /* New failover node */
719 sprintf(niduuid, "%s_%x", mgcname, i);
721 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
723 rc = do_lcfg(mgcname, nid,
724 LCFG_ADD_UUID, niduuid, 0,0,0);
729 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
737 lsi->lsi_lmd->lmd_mgs_failnodes = i;
739 obd = class_name2obd(mgcname);
741 CERROR("Can't find mgcobd %s\n", mgcname);
742 GOTO(out_free, rc = -ENOTCONN);
745 rc = obd_set_info_async(obd->obd_self_export,
746 strlen(KEY_MGSSEC), KEY_MGSSEC,
747 strlen(mgssec), mgssec, NULL);
751 /* Keep a refcount of servers/clients who started with "mount",
752 so we know when we can get rid of the mgc. */
753 atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
755 /* Try all connections, but only once. */
757 rc = obd_set_info_async(obd->obd_self_export,
758 sizeof(KEY_INIT_RECOV_BACKUP),
759 KEY_INIT_RECOV_BACKUP,
760 sizeof(recov_bk), &recov_bk, NULL);
763 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
764 /* We connect to the MGS at setup, and don't disconnect until cleanup */
767 GOTO(out, rc = -ENOMEM);
768 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID |
770 data->ocd_version = LUSTRE_VERSION_CODE;
771 rc = obd_connect(NULL, &mgc_conn, obd, &(obd->obd_uuid), data, NULL);
774 CERROR("connect failed %d\n", rc);
778 exp = class_conn2export(&mgc_conn);
779 obd->u.cli.cl_mgc_mgsexp = exp;
782 /* Keep the mgc info in the sb. Note that many lsi's can point
786 mutex_up(&mgc_start_lock);
789 OBD_FREE(mgcname, len);
791 OBD_FREE(niduuid, len + 2);
795 static int lustre_stop_mgc(struct super_block *sb)
797 struct lustre_sb_info *lsi = s2lsi(sb);
798 struct obd_device *obd;
799 char *niduuid = 0, *ptr = 0;
800 int i, rc = 0, len = 0;
810 mutex_down(&mgc_start_lock);
811 LASSERT(atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
812 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
813 /* This is not fatal, every client that stops
814 will call in here. */
815 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
816 atomic_read(&obd->u.cli.cl_mgc_refcount));
817 GOTO(out, rc = -EBUSY);
820 /* The MGC has no recoverable data in any case.
821 * force shotdown set in umount_begin */
822 obd->obd_no_recov = 1;
824 if (obd->u.cli.cl_mgc_mgsexp) {
825 /* An error is not fatal, if we are unable to send the
826 disconnect mgs ping evictor cleans up the export */
827 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
829 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
832 /* Save the obdname for cleaning the nid uuids, which are
834 len = strlen(obd->obd_name) + 6;
835 OBD_ALLOC(niduuid, len);
837 strcpy(niduuid, obd->obd_name);
838 ptr = niduuid + strlen(niduuid);
841 rc = class_manual_cleanup(obd);
845 /* Clean the nid uuids */
847 GOTO(out, rc = -ENOMEM);
849 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
850 sprintf(ptr, "_%x", i);
851 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
854 CERROR("del MDC UUID %s failed: rc = %d\n",
859 OBD_FREE(niduuid, len);
861 /* class_import_put will get rid of the additional connections */
862 mutex_up(&mgc_start_lock);
866 /* Since there's only one mgc per node, we have to change it's fs to get
867 access to the right disk. */
868 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
870 struct lustre_sb_info *lsi = s2lsi(sb);
874 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
876 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
877 rc = obd_set_info_async(mgc->obd_self_export,
878 sizeof(KEY_SET_FS), KEY_SET_FS,
879 sizeof(*sb), sb, NULL);
881 CERROR("can't set_fs %d\n", rc);
887 static int server_mgc_clear_fs(struct obd_device *mgc)
892 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
894 rc = obd_set_info_async(mgc->obd_self_export,
895 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
900 DECLARE_MUTEX(server_start_lock);
902 /* Stop MDS/OSS if nobody is using them */
903 static int server_stop_servers(int lddflags, int lsiflags)
905 struct obd_device *obd = NULL;
906 struct obd_type *type = NULL;
910 mutex_down(&server_start_lock);
912 /* Either an MDT or an OST or neither */
913 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
914 if ((lddflags & LDD_F_SV_TYPE_MDT) &&
915 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
916 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
917 type = class_search_type(LUSTRE_MDS_NAME);
919 /* if this was an OST, and there are no more OST's, clean up the OSS */
920 if ((lddflags & LDD_F_SV_TYPE_OST) &&
921 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
922 type = class_search_type(LUSTRE_OST_NAME);
925 if (obd && (!type || !type->typ_refcnt)) {
928 /* obd_fail doesn't mean much on a server obd */
929 err = class_manual_cleanup(obd);
934 mutex_up(&server_start_lock);
939 int server_mti_print(char *title, struct mgs_target_info *mti)
941 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
942 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
943 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
944 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
945 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
946 mti->mti_config_ver, mti->mti_flags);
950 static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
952 struct lustre_sb_info *lsi = s2lsi(sb);
953 struct lustre_disk_data *ldd = lsi->lsi_ldd;
954 lnet_process_id_t id;
958 if (!(lsi->lsi_flags & LSI_SERVER))
961 strncpy(mti->mti_fsname, ldd->ldd_fsname,
962 sizeof(mti->mti_fsname));
963 strncpy(mti->mti_svname, ldd->ldd_svname,
964 sizeof(mti->mti_svname));
966 mti->mti_nid_count = 0;
967 while (LNetGetId(i++, &id) != -ENOENT) {
968 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
970 mti->mti_nids[mti->mti_nid_count] = id.nid;
971 mti->mti_nid_count++;
972 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
973 CWARN("Only using first %d nids for %s\n",
974 mti->mti_nid_count, mti->mti_svname);
979 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
980 mti->mti_config_ver = 0;
981 mti->mti_flags = ldd->ldd_flags;
982 mti->mti_stripe_index = ldd->ldd_svindex;
983 memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid));
984 if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) {
985 CERROR("params too big for mti\n");
988 memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params));
992 /* Register an old or new target with the MGS. If needed MGS will construct
993 startup logs and assign index */
994 int server_register_target(struct super_block *sb)
996 struct lustre_sb_info *lsi = s2lsi(sb);
997 struct obd_device *mgc = lsi->lsi_mgc;
998 struct lustre_disk_data *ldd = lsi->lsi_ldd;
999 struct mgs_target_info *mti = NULL;
1005 if (!(lsi->lsi_flags & LSI_SERVER))
1011 rc = server_sb2mti(sb, mti);
1015 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1016 mti->mti_svname, mti->mti_fsname,
1017 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1020 /* Register the target */
1021 /* FIXME use mgc_process_config instead */
1022 rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
1023 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
1024 sizeof(*mti), mti, NULL);
1028 /* Always update our flags */
1029 ldd->ldd_flags = mti->mti_flags & ~LDD_F_REWRITE_LDD;
1031 /* If this flag is set, it means the MGS wants us to change our
1032 on-disk data. (So far this means just the index.) */
1033 if (mti->mti_flags & LDD_F_REWRITE_LDD) {
1036 CDEBUG(D_MOUNT, "Changing on-disk index from %#x to %#x "
1037 "for %s\n", ldd->ldd_svindex, mti->mti_stripe_index,
1039 ldd->ldd_svindex = mti->mti_stripe_index;
1040 strncpy(ldd->ldd_svname, mti->mti_svname,
1041 sizeof(ldd->ldd_svname));
1042 /* or ldd_make_sv_name(ldd); */
1043 ldd_write(&mgc->obd_lvfs_ctxt, ldd);
1044 err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb,
1047 CERROR("Label set error %d\n", err);
1048 label = fsfilt_get_label(mgc, lsi->lsi_srv_mnt->mnt_sb);
1050 CDEBUG(D_MOUNT, "Disk label changed to %s\n", label);
1052 /* Flush the new ldd to disk */
1053 fsfilt_sync(mgc, lsi->lsi_srv_mnt->mnt_sb);
1063 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1065 struct obd_device *obd;
1066 struct lustre_sb_info *lsi = s2lsi(sb);
1067 struct config_llog_instance cfg;
1071 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname);
1074 /* If we're an MDT, make sure the global MDS is running */
1075 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1076 /* make sure the MDS is started */
1077 mutex_down(&server_start_lock);
1078 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1080 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1081 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1083 LUSTRE_MDS_OBDNAME"_uuid",
1086 mutex_up(&server_start_lock);
1087 CERROR("failed to start MDS: %d\n", rc);
1091 mutex_up(&server_start_lock);
1095 /* If we're an OST, make sure the global OSS is running */
1096 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
1097 /* make sure OSS is started */
1098 mutex_down(&server_start_lock);
1099 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1101 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1103 LUSTRE_OSS_OBDNAME"_uuid",
1106 mutex_up(&server_start_lock);
1107 CERROR("failed to start OSS: %d\n", rc);
1111 mutex_up(&server_start_lock);
1114 /* Set the mgc fs to our server disk. This allows the MGC
1115 to read and write configs locally. */
1116 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1120 /* Register with MGS */
1121 rc = server_register_target(sb);
1122 if (rc && (lsi->lsi_ldd->ldd_flags &
1123 (LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_UPGRADE14))){
1124 CERROR("Required registration failed for %s: %d\n",
1125 lsi->lsi_ldd->ldd_svname, rc);
1127 LCONSOLE_ERROR_MSG(0x15f, "Communication error with "
1128 "the MGS. Is the MGS running?\n");
1132 if (rc == -EINVAL) {
1133 LCONSOLE_ERROR_MSG(0x160, "The MGS is refusing to allow this "
1134 "server (%s) to start. Please see messages"
1135 " on the MGS node.\n",
1136 lsi->lsi_ldd->ldd_svname);
1139 /* non-fatal error of registeration with MGS */
1141 CDEBUG(D_MOUNT, "Cannot register with MGS: %d\n", rc);
1143 /* Let the target look up the mount using the target's name
1144 (we can't pass the sb or mnt through class_process_config.) */
1145 rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
1149 /* Start targets using the llog named for the target */
1150 memset(&cfg, 0, sizeof(cfg));
1151 rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg);
1153 CERROR("failed to start server %s: %d\n",
1154 lsi->lsi_ldd->ldd_svname, rc);
1155 /* Do NOT call server_deregister_mount() here. This makes it
1156 * impossible to find mount later in cleanup time and leaves
1157 * @lsi and othder stuff leaked. -umka */
1162 /* Release the mgc fs for others to use */
1163 server_mgc_clear_fs(lsi->lsi_mgc);
1166 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1168 CERROR("no server named %s was started\n",
1169 lsi->lsi_ldd->ldd_svname);
1173 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1174 (OBP(obd, iocontrol))) {
1175 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1176 obd->obd_self_export, 0, NULL, NULL);
1179 /* log has been fully processed */
1180 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1186 /***************** lustre superblock **************/
1188 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1190 struct lustre_sb_info *lsi;
1196 OBD_ALLOC_PTR(lsi->lsi_lmd);
1197 if (!lsi->lsi_lmd) {
1202 lsi->lsi_lmd->lmd_exclude_count = 0;
1203 s2lsi_nocast(sb) = lsi;
1204 /* we take 1 extra ref for our setup */
1205 atomic_set(&lsi->lsi_mounts, 1);
1207 /* Default umount style */
1208 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1213 static int lustre_free_lsi(struct super_block *sb)
1215 struct lustre_sb_info *lsi = s2lsi(sb);
1218 LASSERT(lsi != NULL);
1219 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
1221 /* someone didn't call server_put_mount. */
1222 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
1224 if (lsi->lsi_ldd != NULL)
1225 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
1227 if (lsi->lsi_lmd != NULL) {
1228 if (lsi->lsi_lmd->lmd_dev != NULL)
1229 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1230 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1231 if (lsi->lsi_lmd->lmd_profile != NULL)
1232 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1233 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1234 if (lsi->lsi_lmd->lmd_mgssec != NULL)
1235 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
1236 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
1237 if (lsi->lsi_lmd->lmd_opts != NULL)
1238 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1239 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1240 if (lsi->lsi_lmd->lmd_exclude_count)
1241 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1242 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1243 lsi->lsi_lmd->lmd_exclude_count);
1244 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1247 LASSERT(lsi->lsi_llsbi == NULL);
1248 OBD_FREE(lsi, sizeof(*lsi));
1249 s2lsi_nocast(sb) = NULL;
1254 /* The lsi has one reference for every server that is using the disk -
1255 e.g. MDT, MGS, and potentially MGC */
1256 static int lustre_put_lsi(struct super_block *sb)
1258 struct lustre_sb_info *lsi = s2lsi(sb);
1261 LASSERT(lsi != NULL);
1263 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
1264 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
1265 lustre_free_lsi(sb);
1271 /*************** server mount ******************/
1273 /* Kernel mount using mount options in MOUNT_DATA_FILE */
1274 static struct vfsmount *server_kernel_mount(struct super_block *sb)
1276 struct lvfs_run_ctxt mount_ctxt;
1277 struct lustre_sb_info *lsi = s2lsi(sb);
1278 struct lustre_disk_data *ldd;
1279 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1280 struct vfsmount *mnt;
1281 char *options = NULL;
1282 unsigned long page, s_flags;
1283 struct page *__page;
1287 OBD_ALLOC(ldd, sizeof(*ldd));
1289 RETURN(ERR_PTR(-ENOMEM));
1291 /* In the past, we have always used flags = 0.
1292 Note ext3/ldiskfs can't be mounted ro. */
1293 s_flags = sb->s_flags;
1295 /* allocate memory for options */
1296 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
1298 GOTO(out_free, rc = -ENOMEM);
1299 page = (unsigned long)cfs_page_address(__page);
1300 options = (char *)page;
1301 memset(options, 0, CFS_PAGE_SIZE);
1303 /* mount-line options must be added for pre-mount because it may
1304 * contain mount options such as journal_dev which are required
1305 * to mount successfuly the underlying filesystem */
1306 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0))
1307 strncat(options, lmd->lmd_opts, CFS_PAGE_SIZE - 1);
1309 /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */
1310 CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev);
1311 mnt = ll_kern_mount("ldiskfs", s_flags, lmd->lmd_dev, (void *)options);
1314 CERROR("premount %s:%#lx ldiskfs failed: %d "
1315 "Is the ldiskfs module available?\n",
1316 lmd->lmd_dev, s_flags, rc );
1320 OBD_SET_CTXT_MAGIC(&mount_ctxt);
1321 mount_ctxt.pwdmnt = mnt;
1322 mount_ctxt.pwd = mnt->mnt_root;
1323 mount_ctxt.fs = get_ds();
1325 rc = ldd_parse(&mount_ctxt, ldd);
1329 CERROR("premount parse options failed: rc = %d\n", rc);
1333 /* Done with our pre-mount, now do the real mount. */
1335 /* Glom up mount options */
1336 memset(options, 0, CFS_PAGE_SIZE);
1337 strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2);
1339 /* Add in any mount-line options */
1340 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) {
1341 int len = CFS_PAGE_SIZE - strlen(options) - 2;
1343 strcat(options, ",");
1344 strncat(options, lmd->lmd_opts, len);
1347 /* Special permanent mount flags */
1349 s_flags |= MS_NOATIME | MS_NODIRATIME;
1351 CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
1352 MT_STR(ldd), lmd->lmd_dev, options);
1353 mnt = ll_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev,
1357 CERROR("ll_kern_mount failed: rc = %d\n", rc);
1361 OBD_PAGE_FREE(__page);
1362 lsi->lsi_ldd = ldd; /* freed at lsi cleanup */
1363 CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
1368 OBD_PAGE_FREE(__page);
1369 OBD_FREE(ldd, sizeof(*ldd));
1370 lsi->lsi_ldd = NULL;
1371 RETURN(ERR_PTR(rc));
1374 static void server_wait_finished(struct vfsmount *mnt)
1376 wait_queue_head_t waitq;
1377 struct l_wait_info lwi;
1380 init_waitqueue_head(&waitq);
1382 while ((atomic_read(&mnt->mnt_count) > 1) && (retries > 0)) {
1383 LCONSOLE_WARN("Mount still busy with %d refs, waiting for "
1385 atomic_read(&mnt->mnt_count), retries);
1387 /* Wait for a bit */
1389 lwi = LWI_TIMEOUT(5 * HZ, NULL, NULL);
1390 l_wait_event(waitq, 0, &lwi);
1392 if (atomic_read(&mnt->mnt_count) > 1) {
1393 CERROR("Mount %p is still busy (%d refs), giving up.\n",
1394 mnt, atomic_read(&mnt->mnt_count));
1398 static void server_put_super(struct super_block *sb)
1400 struct lustre_sb_info *lsi = s2lsi(sb);
1401 struct obd_device *obd;
1402 struct vfsmount *mnt = lsi->lsi_srv_mnt;
1403 char *tmpname, *extraname = NULL;
1405 int lddflags = lsi->lsi_ldd->ldd_flags;
1406 int lsiflags = lsi->lsi_flags;
1409 LASSERT(lsiflags & LSI_SERVER);
1411 tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1;
1412 OBD_ALLOC(tmpname, tmpname_sz);
1413 memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
1414 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1416 /* Stop the target */
1417 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1418 (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd))) {
1419 struct lustre_profile *lprof = NULL;
1421 /* tell the mgc to drop the config log */
1422 lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL);
1424 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1425 If there are any setup/cleanup errors, save the lov
1426 name for safety cleanup later. */
1427 lprof = class_get_profile(lsi->lsi_ldd->ldd_svname);
1428 if (lprof && lprof->lp_dt) {
1429 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1430 strcpy(extraname, lprof->lp_dt);
1433 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1435 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1436 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1438 /* We can't seem to give an error return code
1439 * to .put_super, so we better make sure we clean up! */
1441 class_manual_cleanup(obd);
1443 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
1444 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1448 /* If they wanted the mgs to stop separately from the mdt, they
1449 should have put it on a different device. */
1450 if (IS_MGS(lsi->lsi_ldd)) {
1451 /* if MDS start with --nomgs, don't stop MGS then */
1452 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1453 server_stop_mgs(sb);
1456 /* Clean the mgc and sb */
1457 lustre_common_put_super(sb);
1459 /* Wait for the targets to really clean up - can't exit (and let the
1460 sb get destroyed) while the mount is still in use */
1461 server_wait_finished(mnt);
1463 /* drop the One True Mount */
1466 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1467 until the target is really gone so that our type refcount check
1469 server_stop_servers(lddflags, lsiflags);
1471 /* In case of startup or cleanup err, stop related obds */
1473 obd = class_name2obd(extraname);
1475 CWARN("Cleaning orphaned obd %s\n", extraname);
1477 class_manual_cleanup(obd);
1479 OBD_FREE(extraname, strlen(extraname) + 1);
1482 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1483 OBD_FREE(tmpname, tmpname_sz);
1487 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1488 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1490 struct super_block *sb = vfsmnt->mnt_sb;
1492 static void server_umount_begin(struct super_block *sb)
1495 struct lustre_sb_info *lsi = s2lsi(sb);
1498 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1499 if (!(flags & MNT_FORCE)) {
1505 CDEBUG(D_MOUNT, "umount -f\n");
1506 /* umount = failover
1508 no third way to do non-force, non-failover */
1509 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1510 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1514 #ifndef HAVE_STATFS_DENTRY_PARAM
1515 static int server_statfs (struct super_block *sb, struct kstatfs *buf)
1518 static int server_statfs (struct dentry *dentry, struct kstatfs *buf)
1520 struct super_block *sb = dentry->d_sb;
1522 struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt;
1525 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) {
1526 #ifdef HAVE_STATFS_DENTRY_PARAM
1527 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf);
1529 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_sb, buf);
1532 buf->f_type = sb->s_magic;
1538 buf->f_type = sb->s_magic;
1539 buf->f_bsize = sb->s_blocksize;
1545 buf->f_namelen = NAME_MAX;
1549 static struct super_operations server_ops =
1551 .put_super = server_put_super,
1552 .umount_begin = server_umount_begin, /* umount -f */
1553 .statfs = server_statfs,
1556 #define log2(n) ffz(~(n))
1557 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1559 static int server_fill_super_common(struct super_block *sb)
1561 struct inode *root = 0;
1564 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1566 sb->s_blocksize = 4096;
1567 sb->s_blocksize_bits = log2(sb->s_blocksize);
1568 sb->s_magic = LUSTRE_SUPER_MAGIC;
1569 sb->s_maxbytes = 0; //PAGE_CACHE_MAXBYTES;
1570 sb->s_flags |= MS_RDONLY;
1571 sb->s_op = &server_ops;
1573 root = new_inode(sb);
1575 CERROR("Can't make root inode\n");
1579 /* returns -EIO for every operation */
1580 /* make_bad_inode(root); -- badness - can't umount */
1581 /* apparently we need to be a directory for the mount to finish */
1582 root->i_mode = S_IFDIR;
1584 sb->s_root = d_alloc_root(root);
1586 CERROR("Can't make root dentry\n");
1594 static int server_fill_super(struct super_block *sb)
1596 struct lustre_sb_info *lsi = s2lsi(sb);
1597 struct vfsmount *mnt;
1601 /* the One True Mount */
1602 mnt = server_kernel_mount(sb);
1605 CERROR("Unable to mount device %s: %d\n",
1606 lsi->lsi_lmd->lmd_dev, rc);
1610 lsi->lsi_srv_mnt = mnt;
1612 LASSERT(lsi->lsi_ldd);
1613 CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n",
1614 lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname,
1615 lsi->lsi_lmd->lmd_dev);
1617 if (class_name2obd(lsi->lsi_ldd->ldd_svname)) {
1618 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1619 "running. Double-mount may have compromised"
1620 " the disk journal.\n",
1621 lsi->lsi_ldd->ldd_svname);
1627 /* Start MGS before MGC */
1628 if (IS_MGS(lsi->lsi_ldd) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
1629 rc = server_start_mgs(sb);
1634 rc = lustre_start_mgc(sb);
1638 /* Set up all obd devices for service */
1639 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1640 (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) {
1641 rc = server_start_targets(sb, mnt);
1643 CERROR("Unable to start targets: %d\n", rc);
1646 /* FIXME overmount client here,
1647 or can we just start a client log and client_fill_super on this sb?
1648 We need to make sure server_put_super gets called too - ll_put_super
1649 calls lustre_common_put_super; check there for LSI_SERVER flag,
1651 Probably should start client from new thread so we can return.
1652 Client will not finish until all servers are connected.
1653 Note - MGS-only server does NOT get a client, since there is no
1654 lustre fs associated - the MGS is for all lustre fs's */
1657 rc = server_fill_super_common(sb);
1661 LCONSOLE_WARN("Server %s on device %s has started\n",
1662 lsi->lsi_ldd->ldd_svname, lsi->lsi_lmd->lmd_dev);
1666 /* We jump here in case of failure while starting targets or MGS.
1667 * In this case we can't just put @mnt and have to do real cleanup
1668 * with stoping targets, etc. */
1669 server_put_super(sb);
1673 /* Get the index from the obd name.
1674 rc = server type, or
1676 if endptr isn't NULL it is set to end of name */
1677 int server_name2index(char *svname, __u32 *idx, char **endptr)
1679 unsigned long index;
1681 char *dash = strrchr(svname, '-');
1685 if (strncmp(dash + 1, "MDT", 3) == 0)
1686 rc = LDD_F_SV_TYPE_MDT;
1687 else if (strncmp(dash + 1, "OST", 3) == 0)
1688 rc = LDD_F_SV_TYPE_OST;
1691 if (strcmp(dash + 4, "all") == 0)
1692 return rc | LDD_F_SV_ALL;
1694 index = simple_strtoul(dash + 4, endptr, 16);
1699 /*************** mount common betweeen server and client ***************/
1702 int lustre_common_put_super(struct super_block *sb)
1707 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1709 /* Drop a ref to the MGC */
1710 rc = lustre_stop_mgc(sb);
1711 if (rc && (rc != -ENOENT)) {
1713 CERROR("Can't stop MGC: %d\n", rc);
1716 /* BUSY just means that there's some other obd that
1717 needs the mgc. Let him clean it up. */
1718 CDEBUG(D_MOUNT, "MGC still in use\n");
1720 /* Drop a ref to the mounted disk */
1727 static void lmd_print(struct lustre_mount_data *lmd)
1731 PRINT_CMD(PRINT_MASK, " mount data:\n");
1732 if (lmd_is_client(lmd))
1733 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
1734 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
1735 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
1737 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
1738 for (i = 0; i < lmd->lmd_exclude_count; i++) {
1739 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
1740 lmd->lmd_exclude[i]);
1745 /* Is this server on the exclusion list */
1746 int lustre_check_exclusion(struct super_block *sb, char *svname)
1748 struct lustre_sb_info *lsi = s2lsi(sb);
1749 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1754 rc = server_name2index(svname, &index, NULL);
1755 if (rc != LDD_F_SV_TYPE_OST)
1756 /* Only exclude OSTs */
1759 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
1760 index, lmd->lmd_exclude_count, lmd->lmd_dev);
1762 for(i = 0; i < lmd->lmd_exclude_count; i++) {
1763 if (index == lmd->lmd_exclude[i]) {
1764 CWARN("Excluding %s (on exclusion list)\n", svname);
1771 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
1772 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
1774 char *s1 = ptr, *s2;
1775 __u32 index, *exclude_list;
1779 /* The shortest an ost name can be is 8 chars: -OST0000.
1780 We don't actually know the fsname at this time, so in fact
1781 a user could specify any fsname. */
1782 devmax = strlen(ptr) / 8 + 1;
1784 /* temp storage until we figure out how many we have */
1785 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
1789 /* we enter this fn pointing at the '=' */
1790 while (*s1 && *s1 != ' ' && *s1 != ',') {
1792 rc = server_name2index(s1, &index, &s2);
1794 CERROR("Can't parse server name '%s'\n", s1);
1797 if (rc == LDD_F_SV_TYPE_OST)
1798 exclude_list[lmd->lmd_exclude_count++] = index;
1800 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
1802 /* now we are pointing at ':' (next exclude)
1803 or ',' (end of excludes) */
1804 if (lmd->lmd_exclude_count >= devmax)
1807 if (rc >= 0) /* non-err */
1810 if (lmd->lmd_exclude_count) {
1811 /* permanent, freed in lustre_free_lsi */
1812 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
1813 lmd->lmd_exclude_count);
1814 if (lmd->lmd_exclude) {
1815 memcpy(lmd->lmd_exclude, exclude_list,
1816 sizeof(index) * lmd->lmd_exclude_count);
1819 lmd->lmd_exclude_count = 0;
1822 OBD_FREE(exclude_list, sizeof(index) * devmax);
1826 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
1831 if (lmd->lmd_mgssec != NULL) {
1832 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
1833 lmd->lmd_mgssec = NULL;
1836 tail = strchr(ptr, ',');
1838 length = strlen(ptr);
1840 length = tail - ptr;
1842 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
1843 if (lmd->lmd_mgssec == NULL)
1846 memcpy(lmd->lmd_mgssec, ptr, length);
1847 lmd->lmd_mgssec[length] = '\0';
1851 /* mount -v -t lustre uml1:uml2:/lustre-client /mnt/lustre */
1852 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1854 char *s1, *s2, *devname = NULL;
1855 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1861 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1862 "/sbin/mount.lustre is installed.\n");
1866 /* Options should be a string - try to detect old lmd data */
1867 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1868 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1869 "/sbin/mount.lustre. Please install "
1870 "version %s\n", LUSTRE_VERSION_STRING);
1873 lmd->lmd_magic = LMD_MAGIC;
1875 /* Set default flags here */
1880 /* Skip whitespace and extra commas */
1881 while (*s1 == ' ' || *s1 == ',')
1884 /* Client options are parsed in ll_options: eg. flock,
1887 /* Parse non-ldiskfs options here. Rather than modifying
1888 ldiskfs, we just zero these out here */
1889 if (strncmp(s1, "abort_recov", 11) == 0) {
1890 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1892 } else if (strncmp(s1, "nosvc", 5) == 0) {
1893 lmd->lmd_flags |= LMD_FLG_NOSVC;
1895 } else if (strncmp(s1, "nomgs", 5) == 0) {
1896 lmd->lmd_flags |= LMD_FLG_NOMGS;
1898 } else if (strncmp(s1, "mgssec=", 7) == 0) {
1899 rc = lmd_parse_mgssec(lmd, s1 + 7);
1903 /* ost exclusion list */
1904 } else if (strncmp(s1, "exclude=", 8) == 0) {
1905 rc = lmd_make_exclusion(lmd, s1 + 7);
1910 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1911 end of the options. */
1912 else if (strncmp(s1, "device=", 7) == 0) {
1914 /* terminate options right before device. device
1915 must be the last one. */
1921 s2 = strchr(s1, ',');
1929 memmove(s1, s2, strlen(s2) + 1);
1935 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1936 "(need mount option 'device=...')\n");
1940 s1 = strstr(devname, ":/");
1943 lmd->lmd_flags = LMD_FLG_CLIENT;
1944 /* Remove leading /s from fsname */
1945 while (*++s1 == '/') ;
1946 /* Freed in lustre_free_lsi */
1947 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1948 if (!lmd->lmd_profile)
1950 sprintf(lmd->lmd_profile, "%s-client", s1);
1953 /* Freed in lustre_free_lsi */
1954 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1957 strcpy(lmd->lmd_dev, devname);
1959 /* Save mount options */
1960 s1 = options + strlen(options) - 1;
1961 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1963 if (*options != 0) {
1964 /* Freed in lustre_free_lsi */
1965 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1968 strcpy(lmd->lmd_opts, options);
1971 lmd->lmd_magic = LMD_MAGIC;
1976 CERROR("Bad mount options %s\n", options);
1982 int lustre_fill_super(struct super_block *sb, void *data, int silent)
1984 struct lustre_mount_data *lmd;
1985 struct lustre_sb_info *lsi;
1989 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1991 lsi = lustre_init_lsi(sb);
1997 * Disable lockdep during mount, because mount locking patterns are
2002 /* Figure out the lmd from the mount options */
2003 if (lmd_parse((char *)data, lmd)) {
2005 GOTO(out, rc = -EINVAL);
2008 if (lmd_is_client(lmd)) {
2009 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
2010 if (!client_fill_super) {
2011 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
2012 "client mount! Is the 'lustre' "
2013 "module loaded?\n");
2017 rc = lustre_start_mgc(sb);
2022 /* Connect and start */
2023 /* (should always be ll_fill_super) */
2024 rc = (*client_fill_super)(sb);
2025 /* c_f_s will call lustre_common_put_super on failure */
2028 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
2029 lsi->lsi_flags |= LSI_SERVER;
2030 rc = server_fill_super(sb);
2031 /* s_f_s calls lustre_start_mgc after the mount because we need
2032 the MGS nids which are stored on disk. Plus, we may
2033 need to start the MGS first. */
2034 /* s_f_s will call server_put_super on failure */
2037 /* If error happens in fill_super() call, @lsi will be killed there.
2038 * This is why we do not put it here. */
2042 CERROR("Unable to mount %s (%d)\n",
2043 s2lsi(sb) ? lmd->lmd_dev : "", rc);
2045 CDEBUG(D_SUPER, "Mount %s complete\n",
2053 /* We can't call ll_fill_super by name because it lives in a module that
2054 must be loaded after this one. */
2055 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb))
2057 client_fill_super = cfs;
2060 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
2062 kill_super_cb = cfs;
2065 /***************** FS registration ******************/
2067 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))
2068 struct super_block * lustre_get_sb(struct file_system_type *fs_type,
2069 int flags, const char *devname, void * data)
2071 return get_sb_nodev(fs_type, flags, data, lustre_fill_super);
2074 int lustre_get_sb(struct file_system_type *fs_type,
2075 int flags, const char *devname, void * data,
2076 struct vfsmount *mnt)
2078 return get_sb_nodev(fs_type, flags, data, lustre_fill_super, mnt);
2082 void lustre_kill_super(struct super_block *sb)
2084 struct lustre_sb_info *lsi = s2lsi(sb);
2086 if (kill_super_cb && lsi && !(lsi->lsi_flags & LSI_SERVER))
2087 (*kill_super_cb)(sb);
2089 kill_anon_super(sb);
2092 struct file_system_type lustre_fs_type = {
2093 .owner = THIS_MODULE,
2095 .get_sb = lustre_get_sb,
2096 .kill_sb = lustre_kill_super,
2097 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
2098 LL_RENAME_DOES_D_MOVE,
2101 int lustre_register_fs(void)
2103 return register_filesystem(&lustre_fs_type);
2106 int lustre_unregister_fs(void)
2108 return unregister_filesystem(&lustre_fs_type);
2111 EXPORT_SYMBOL(lustre_register_client_fill_super);
2112 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2113 EXPORT_SYMBOL(lustre_common_put_super);
2114 EXPORT_SYMBOL(lustre_process_log);
2115 EXPORT_SYMBOL(lustre_end_log);
2116 EXPORT_SYMBOL(server_get_mount);
2117 EXPORT_SYMBOL(server_get_mount_2);
2118 EXPORT_SYMBOL(server_put_mount);
2119 EXPORT_SYMBOL(server_put_mount_2);
2120 EXPORT_SYMBOL(server_register_target);
2121 EXPORT_SYMBOL(server_name2index);
2122 EXPORT_SYMBOL(server_mti_print);
2123 EXPORT_SYMBOL(do_lcfg);