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 %u\n",
297 len, (int)sizeof(*ldd));
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 obd_connect_data *data = NULL;
571 struct lustre_sb_info *lsi = s2lsi(sb);
572 struct obd_device *obd;
573 struct obd_export *exp;
574 struct obd_uuid *uuid;
577 char *mgcname, *niduuid, *mgssec;
580 int rc = 0, i = 0, j, len;
583 LASSERT(lsi->lsi_lmd);
585 /* Find the first non-lo MGS nid for our MGC name */
586 if (lsi->lsi_flags & LSI_SERVER) {
587 ptr = lsi->lsi_ldd->ldd_params;
588 /* Use mgsnode= nids */
589 if ((class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0) &&
590 (class_parse_nid(ptr, &nid, &ptr) == 0)) {
592 } else if (IS_MGS(lsi->lsi_ldd)) {
593 lnet_process_id_t id;
594 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
595 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
602 } else { /* client */
603 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
604 ptr = lsi->lsi_lmd->lmd_dev;
605 if (class_parse_nid(ptr, &nid, &ptr) == 0)
609 CERROR("No valid MGS nids found.\n");
613 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
614 OBD_ALLOC(mgcname, len);
615 OBD_ALLOC(niduuid, len + 2);
616 if (!mgcname || !niduuid)
617 GOTO(out_free, rc = -ENOMEM);
618 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
620 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
622 mutex_down(&mgc_start_lock);
624 obd = class_name2obd(mgcname);
625 if (obd && !obd->obd_stopping) {
626 rc = obd_set_info_async(obd->obd_self_export,
627 strlen(KEY_MGSSEC), KEY_MGSSEC,
628 strlen(mgssec), mgssec, NULL);
632 /* Re-using an existing MGC */
633 atomic_inc(&obd->u.cli.cl_mgc_refcount);
636 /* If we are restarting the MGS, don't try to keep the MGC's
637 old connection, or registration will fail. */
638 if ((lsi->lsi_flags & LSI_SERVER) && IS_MGS(lsi->lsi_ldd)) {
639 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
643 /* Try all connections, but only once (again).
644 We don't want to block another target from starting
645 (using its local copy of the log), but we do want to connect
646 if at all possible. */
648 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
649 rc = obd_set_info_async(obd->obd_self_export,
650 sizeof(KEY_INIT_RECOV_BACKUP),
651 KEY_INIT_RECOV_BACKUP,
652 sizeof(recov_bk), &recov_bk, NULL);
656 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
658 /* Add the primary nids for the MGS */
660 sprintf(niduuid, "%s_%x", mgcname, i);
661 if (lsi->lsi_flags & LSI_SERVER) {
662 ptr = lsi->lsi_ldd->ldd_params;
663 if (IS_MGS(lsi->lsi_ldd)) {
664 /* Use local nids (including LO) */
665 lnet_process_id_t id;
666 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
667 rc = do_lcfg(mgcname, id.nid,
668 LCFG_ADD_UUID, niduuid, 0,0,0);
671 /* Use mgsnode= nids */
672 if (class_find_param(ptr, PARAM_MGSNODE, &ptr) != 0) {
673 CERROR("No MGS nids given.\n");
674 GOTO(out_free, rc = -EINVAL);
676 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
677 rc = do_lcfg(mgcname, nid,
678 LCFG_ADD_UUID, niduuid, 0,0,0);
682 } else { /* client */
683 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
684 ptr = lsi->lsi_lmd->lmd_dev;
685 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
686 rc = do_lcfg(mgcname, nid,
687 LCFG_ADD_UUID, niduuid, 0,0,0);
689 /* Stop at the first failover nid */
695 CERROR("No valid MGS nids found.\n");
696 GOTO(out_free, rc = -EINVAL);
698 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
700 /* Random uuid for MGC allows easier reconnects */
702 ll_generate_random_uuid(uuidc);
703 class_uuid_unparse(uuidc, uuid);
706 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
707 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
713 /* Add any failover MGS nids */
715 while ((*ptr == ':' ||
716 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0)) {
717 /* New failover node */
718 sprintf(niduuid, "%s_%x", mgcname, i);
720 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
722 rc = do_lcfg(mgcname, nid,
723 LCFG_ADD_UUID, niduuid, 0,0,0);
728 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
736 lsi->lsi_lmd->lmd_mgs_failnodes = i;
738 obd = class_name2obd(mgcname);
740 CERROR("Can't find mgcobd %s\n", mgcname);
741 GOTO(out_free, rc = -ENOTCONN);
744 rc = obd_set_info_async(obd->obd_self_export,
745 strlen(KEY_MGSSEC), KEY_MGSSEC,
746 strlen(mgssec), mgssec, NULL);
750 /* Keep a refcount of servers/clients who started with "mount",
751 so we know when we can get rid of the mgc. */
752 atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
754 /* Try all connections, but only once. */
756 rc = obd_set_info_async(obd->obd_self_export,
757 sizeof(KEY_INIT_RECOV_BACKUP),
758 KEY_INIT_RECOV_BACKUP,
759 sizeof(recov_bk), &recov_bk, NULL);
762 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
763 /* We connect to the MGS at setup, and don't disconnect until cleanup */
766 GOTO(out, rc = -ENOMEM);
767 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID |
769 data->ocd_version = LUSTRE_VERSION_CODE;
770 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
773 CERROR("connect failed %d\n", rc);
777 obd->u.cli.cl_mgc_mgsexp = exp;
780 /* Keep the mgc info in the sb. Note that many lsi's can point
784 mutex_up(&mgc_start_lock);
787 OBD_FREE(mgcname, len);
789 OBD_FREE(niduuid, len + 2);
793 static int lustre_stop_mgc(struct super_block *sb)
795 struct lustre_sb_info *lsi = s2lsi(sb);
796 struct obd_device *obd;
797 char *niduuid = 0, *ptr = 0;
798 int i, rc = 0, len = 0;
808 mutex_down(&mgc_start_lock);
809 LASSERT(atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
810 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
811 /* This is not fatal, every client that stops
812 will call in here. */
813 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
814 atomic_read(&obd->u.cli.cl_mgc_refcount));
815 GOTO(out, rc = -EBUSY);
818 /* The MGC has no recoverable data in any case.
819 * force shotdown set in umount_begin */
820 obd->obd_no_recov = 1;
822 if (obd->u.cli.cl_mgc_mgsexp) {
823 /* An error is not fatal, if we are unable to send the
824 disconnect mgs ping evictor cleans up the export */
825 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
827 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
830 /* Save the obdname for cleaning the nid uuids, which are
832 len = strlen(obd->obd_name) + 6;
833 OBD_ALLOC(niduuid, len);
835 strcpy(niduuid, obd->obd_name);
836 ptr = niduuid + strlen(niduuid);
839 rc = class_manual_cleanup(obd);
843 /* Clean the nid uuids */
845 GOTO(out, rc = -ENOMEM);
847 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
848 sprintf(ptr, "_%x", i);
849 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
852 CERROR("del MDC UUID %s failed: rc = %d\n",
857 OBD_FREE(niduuid, len);
859 /* class_import_put will get rid of the additional connections */
860 mutex_up(&mgc_start_lock);
864 /* Since there's only one mgc per node, we have to change it's fs to get
865 access to the right disk. */
866 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
868 struct lustre_sb_info *lsi = s2lsi(sb);
872 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
874 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
875 rc = obd_set_info_async(mgc->obd_self_export,
876 sizeof(KEY_SET_FS), KEY_SET_FS,
877 sizeof(*sb), sb, NULL);
879 CERROR("can't set_fs %d\n", rc);
885 static int server_mgc_clear_fs(struct obd_device *mgc)
890 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
892 rc = obd_set_info_async(mgc->obd_self_export,
893 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
898 DECLARE_MUTEX(server_start_lock);
900 /* Stop MDS/OSS if nobody is using them */
901 static int server_stop_servers(int lddflags, int lsiflags)
903 struct obd_device *obd = NULL;
904 struct obd_type *type = NULL;
908 mutex_down(&server_start_lock);
910 /* Either an MDT or an OST or neither */
911 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
912 if ((lddflags & LDD_F_SV_TYPE_MDT) &&
913 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
914 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
915 type = class_search_type(LUSTRE_MDS_NAME);
917 /* if this was an OST, and there are no more OST's, clean up the OSS */
918 if ((lddflags & LDD_F_SV_TYPE_OST) &&
919 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
920 type = class_search_type(LUSTRE_OST_NAME);
923 if (obd && (!type || !type->typ_refcnt)) {
926 /* obd_fail doesn't mean much on a server obd */
927 err = class_manual_cleanup(obd);
932 mutex_up(&server_start_lock);
937 int server_mti_print(char *title, struct mgs_target_info *mti)
939 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
940 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
941 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
942 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
943 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
944 mti->mti_config_ver, mti->mti_flags);
948 static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
950 struct lustre_sb_info *lsi = s2lsi(sb);
951 struct lustre_disk_data *ldd = lsi->lsi_ldd;
952 lnet_process_id_t id;
956 if (!(lsi->lsi_flags & LSI_SERVER))
959 strncpy(mti->mti_fsname, ldd->ldd_fsname,
960 sizeof(mti->mti_fsname));
961 strncpy(mti->mti_svname, ldd->ldd_svname,
962 sizeof(mti->mti_svname));
964 mti->mti_nid_count = 0;
965 while (LNetGetId(i++, &id) != -ENOENT) {
966 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
968 mti->mti_nids[mti->mti_nid_count] = id.nid;
969 mti->mti_nid_count++;
970 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
971 CWARN("Only using first %d nids for %s\n",
972 mti->mti_nid_count, mti->mti_svname);
977 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
978 mti->mti_config_ver = 0;
979 mti->mti_flags = ldd->ldd_flags;
980 mti->mti_stripe_index = ldd->ldd_svindex;
981 memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid));
982 if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) {
983 CERROR("params too big for mti\n");
986 memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params));
990 /* Register an old or new target with the MGS. If needed MGS will construct
991 startup logs and assign index */
992 int server_register_target(struct super_block *sb)
994 struct lustre_sb_info *lsi = s2lsi(sb);
995 struct obd_device *mgc = lsi->lsi_mgc;
996 struct lustre_disk_data *ldd = lsi->lsi_ldd;
997 struct mgs_target_info *mti = NULL;
1003 if (!(lsi->lsi_flags & LSI_SERVER))
1009 rc = server_sb2mti(sb, mti);
1013 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1014 mti->mti_svname, mti->mti_fsname,
1015 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1018 /* Register the target */
1019 /* FIXME use mgc_process_config instead */
1020 rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
1021 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
1022 sizeof(*mti), mti, NULL);
1026 /* Always update our flags */
1027 ldd->ldd_flags = mti->mti_flags & ~LDD_F_REWRITE_LDD;
1029 /* If this flag is set, it means the MGS wants us to change our
1030 on-disk data. (So far this means just the index.) */
1031 if (mti->mti_flags & LDD_F_REWRITE_LDD) {
1034 CDEBUG(D_MOUNT, "Changing on-disk index from %#x to %#x "
1035 "for %s\n", ldd->ldd_svindex, mti->mti_stripe_index,
1037 ldd->ldd_svindex = mti->mti_stripe_index;
1038 strncpy(ldd->ldd_svname, mti->mti_svname,
1039 sizeof(ldd->ldd_svname));
1040 /* or ldd_make_sv_name(ldd); */
1041 ldd_write(&mgc->obd_lvfs_ctxt, ldd);
1042 err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb,
1045 CERROR("Label set error %d\n", err);
1046 label = fsfilt_get_label(mgc, lsi->lsi_srv_mnt->mnt_sb);
1048 CDEBUG(D_MOUNT, "Disk label changed to %s\n", label);
1050 /* Flush the new ldd to disk */
1051 fsfilt_sync(mgc, lsi->lsi_srv_mnt->mnt_sb);
1061 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1063 struct obd_device *obd;
1064 struct lustre_sb_info *lsi = s2lsi(sb);
1065 struct config_llog_instance cfg;
1069 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname);
1072 /* If we're an MDT, make sure the global MDS is running */
1073 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1074 /* make sure the MDS is started */
1075 mutex_down(&server_start_lock);
1076 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1078 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1079 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1081 LUSTRE_MDS_OBDNAME"_uuid",
1084 mutex_up(&server_start_lock);
1085 CERROR("failed to start MDS: %d\n", rc);
1089 mutex_up(&server_start_lock);
1093 /* If we're an OST, make sure the global OSS is running */
1094 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
1095 /* make sure OSS is started */
1096 mutex_down(&server_start_lock);
1097 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1099 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1101 LUSTRE_OSS_OBDNAME"_uuid",
1104 mutex_up(&server_start_lock);
1105 CERROR("failed to start OSS: %d\n", rc);
1109 mutex_up(&server_start_lock);
1112 /* Set the mgc fs to our server disk. This allows the MGC
1113 to read and write configs locally. */
1114 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1118 /* Register with MGS */
1119 rc = server_register_target(sb);
1120 if (rc && (lsi->lsi_ldd->ldd_flags &
1121 (LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_UPGRADE14))){
1122 CERROR("Required registration failed for %s: %d\n",
1123 lsi->lsi_ldd->ldd_svname, rc);
1125 LCONSOLE_ERROR_MSG(0x15f, "Communication error with "
1126 "the MGS. Is the MGS running?\n");
1130 if (rc == -EINVAL) {
1131 LCONSOLE_ERROR_MSG(0x160, "The MGS is refusing to allow this "
1132 "server (%s) to start. Please see messages"
1133 " on the MGS node.\n",
1134 lsi->lsi_ldd->ldd_svname);
1137 /* non-fatal error of registeration with MGS */
1139 CDEBUG(D_MOUNT, "Cannot register with MGS: %d\n", rc);
1141 /* Let the target look up the mount using the target's name
1142 (we can't pass the sb or mnt through class_process_config.) */
1143 rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
1147 /* Start targets using the llog named for the target */
1148 memset(&cfg, 0, sizeof(cfg));
1149 rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg);
1151 CERROR("failed to start server %s: %d\n",
1152 lsi->lsi_ldd->ldd_svname, rc);
1153 /* Do NOT call server_deregister_mount() here. This makes it
1154 * impossible to find mount later in cleanup time and leaves
1155 * @lsi and othder stuff leaked. -umka */
1160 /* Release the mgc fs for others to use */
1161 server_mgc_clear_fs(lsi->lsi_mgc);
1164 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1166 CERROR("no server named %s was started\n",
1167 lsi->lsi_ldd->ldd_svname);
1171 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1172 (OBP(obd, iocontrol))) {
1173 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1174 obd->obd_self_export, 0, NULL, NULL);
1177 /* log has been fully processed */
1178 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1184 /***************** lustre superblock **************/
1186 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1188 struct lustre_sb_info *lsi;
1194 OBD_ALLOC_PTR(lsi->lsi_lmd);
1195 if (!lsi->lsi_lmd) {
1200 lsi->lsi_lmd->lmd_exclude_count = 0;
1201 s2lsi_nocast(sb) = lsi;
1202 /* we take 1 extra ref for our setup */
1203 atomic_set(&lsi->lsi_mounts, 1);
1205 /* Default umount style */
1206 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1211 static int lustre_free_lsi(struct super_block *sb)
1213 struct lustre_sb_info *lsi = s2lsi(sb);
1216 LASSERT(lsi != NULL);
1217 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
1219 /* someone didn't call server_put_mount. */
1220 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
1222 if (lsi->lsi_ldd != NULL)
1223 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
1225 if (lsi->lsi_lmd != NULL) {
1226 if (lsi->lsi_lmd->lmd_dev != NULL)
1227 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1228 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1229 if (lsi->lsi_lmd->lmd_profile != NULL)
1230 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1231 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1232 if (lsi->lsi_lmd->lmd_mgssec != NULL)
1233 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
1234 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
1235 if (lsi->lsi_lmd->lmd_opts != NULL)
1236 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1237 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1238 if (lsi->lsi_lmd->lmd_exclude_count)
1239 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1240 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1241 lsi->lsi_lmd->lmd_exclude_count);
1242 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1245 LASSERT(lsi->lsi_llsbi == NULL);
1246 OBD_FREE(lsi, sizeof(*lsi));
1247 s2lsi_nocast(sb) = NULL;
1252 /* The lsi has one reference for every server that is using the disk -
1253 e.g. MDT, MGS, and potentially MGC */
1254 static int lustre_put_lsi(struct super_block *sb)
1256 struct lustre_sb_info *lsi = s2lsi(sb);
1259 LASSERT(lsi != NULL);
1261 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
1262 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
1263 lustre_free_lsi(sb);
1269 /*************** server mount ******************/
1271 /* Kernel mount using mount options in MOUNT_DATA_FILE */
1272 static struct vfsmount *server_kernel_mount(struct super_block *sb)
1274 struct lvfs_run_ctxt mount_ctxt;
1275 struct lustre_sb_info *lsi = s2lsi(sb);
1276 struct lustre_disk_data *ldd;
1277 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1278 struct vfsmount *mnt;
1279 char *options = NULL;
1280 unsigned long page, s_flags;
1281 struct page *__page;
1285 OBD_ALLOC(ldd, sizeof(*ldd));
1287 RETURN(ERR_PTR(-ENOMEM));
1289 /* In the past, we have always used flags = 0.
1290 Note ext3/ldiskfs can't be mounted ro. */
1291 s_flags = sb->s_flags;
1293 /* allocate memory for options */
1294 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
1296 GOTO(out_free, rc = -ENOMEM);
1297 page = (unsigned long)cfs_page_address(__page);
1298 options = (char *)page;
1299 memset(options, 0, CFS_PAGE_SIZE);
1301 /* mount-line options must be added for pre-mount because it may
1302 * contain mount options such as journal_dev which are required
1303 * to mount successfuly the underlying filesystem */
1304 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0))
1305 strncat(options, lmd->lmd_opts, CFS_PAGE_SIZE - 1);
1307 /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */
1308 CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev);
1309 mnt = ll_kern_mount("ldiskfs", s_flags, lmd->lmd_dev, (void *)options);
1312 CERROR("premount %s:%#lx ldiskfs failed: %d "
1313 "Is the ldiskfs module available?\n",
1314 lmd->lmd_dev, s_flags, rc );
1318 OBD_SET_CTXT_MAGIC(&mount_ctxt);
1319 mount_ctxt.pwdmnt = mnt;
1320 mount_ctxt.pwd = mnt->mnt_root;
1321 mount_ctxt.fs = get_ds();
1323 rc = ldd_parse(&mount_ctxt, ldd);
1327 CERROR("premount parse options failed: rc = %d\n", rc);
1331 /* Done with our pre-mount, now do the real mount. */
1333 /* Glom up mount options */
1334 memset(options, 0, CFS_PAGE_SIZE);
1335 strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2);
1337 /* Add in any mount-line options */
1338 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) {
1339 int len = CFS_PAGE_SIZE - strlen(options) - 2;
1341 strcat(options, ",");
1342 strncat(options, lmd->lmd_opts, len);
1345 /* Special permanent mount flags */
1347 s_flags |= MS_NOATIME | MS_NODIRATIME;
1349 CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
1350 MT_STR(ldd), lmd->lmd_dev, options);
1351 mnt = ll_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev,
1355 CERROR("ll_kern_mount failed: rc = %d\n", rc);
1359 if (lmd->lmd_flags & LMD_FLG_ABORT_RECOV)
1360 simple_truncate(mnt->mnt_sb->s_root, mnt, LAST_RCVD,
1363 OBD_PAGE_FREE(__page);
1364 lsi->lsi_ldd = ldd; /* freed at lsi cleanup */
1365 CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
1370 OBD_PAGE_FREE(__page);
1371 OBD_FREE(ldd, sizeof(*ldd));
1372 lsi->lsi_ldd = NULL;
1373 RETURN(ERR_PTR(rc));
1376 /* Wait here forever until the mount refcount is 0 before completing umount,
1377 * else we risk dereferencing a null pointer.
1378 * LNET may take e.g. 165s before killing zombies.
1380 static void server_wait_finished(struct vfsmount *mnt)
1384 cfs_sigset_t blocked;
1386 cfs_waitq_init(&waitq);
1388 while (cfs_atomic_read(&mnt->mnt_count) > 1) {
1389 if (waited && (waited % 30 == 0))
1390 LCONSOLE_WARN("Mount still busy with %d refs after "
1392 atomic_read(&mnt->mnt_count),
1394 /* Cannot use l_event_wait() for an interruptible sleep. */
1396 blocked = l_w_e_set_sigs(sigmask(SIGKILL));
1397 cfs_waitq_wait_event_interruptible_timeout(
1399 (cfs_atomic_read(&mnt->mnt_count) == 1),
1400 cfs_time_seconds(3),
1402 cfs_block_sigs(blocked);
1404 LCONSOLE_EMERG("Danger: interrupted umount %s with "
1407 atomic_read(&mnt->mnt_count));
1414 static void server_put_super(struct super_block *sb)
1416 struct lustre_sb_info *lsi = s2lsi(sb);
1417 struct obd_device *obd;
1418 struct vfsmount *mnt = lsi->lsi_srv_mnt;
1419 char *tmpname, *extraname = NULL;
1421 int lddflags = lsi->lsi_ldd->ldd_flags;
1422 int lsiflags = lsi->lsi_flags;
1425 LASSERT(lsiflags & LSI_SERVER);
1427 tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1;
1428 OBD_ALLOC(tmpname, tmpname_sz);
1429 memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
1430 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1431 if (IS_MDT(lsi->lsi_ldd) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1432 snprintf(tmpname, tmpname_sz, "MGS");
1434 /* Stop the target */
1435 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1436 (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd))) {
1437 struct lustre_profile *lprof = NULL;
1439 /* tell the mgc to drop the config log */
1440 lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL);
1442 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1443 If there are any setup/cleanup errors, save the lov
1444 name for safety cleanup later. */
1445 lprof = class_get_profile(lsi->lsi_ldd->ldd_svname);
1446 if (lprof && lprof->lp_dt) {
1447 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1448 strcpy(extraname, lprof->lp_dt);
1451 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1453 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1454 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1456 /* We can't seem to give an error return code
1457 * to .put_super, so we better make sure we clean up! */
1459 class_manual_cleanup(obd);
1461 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
1462 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1466 /* If they wanted the mgs to stop separately from the mdt, they
1467 should have put it on a different device. */
1468 if (IS_MGS(lsi->lsi_ldd)) {
1469 /* if MDS start with --nomgs, don't stop MGS then */
1470 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1471 server_stop_mgs(sb);
1474 /* Clean the mgc and sb */
1475 lustre_common_put_super(sb);
1477 /* Wait for the targets to really clean up - can't exit (and let the
1478 sb get destroyed) while the mount is still in use */
1479 server_wait_finished(mnt);
1481 /* drop the One True Mount */
1484 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1485 until the target is really gone so that our type refcount check
1487 server_stop_servers(lddflags, lsiflags);
1489 /* In case of startup or cleanup err, stop related obds */
1491 obd = class_name2obd(extraname);
1493 CWARN("Cleaning orphaned obd %s\n", extraname);
1495 class_manual_cleanup(obd);
1497 OBD_FREE(extraname, strlen(extraname) + 1);
1500 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1501 OBD_FREE(tmpname, tmpname_sz);
1505 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1506 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1508 struct super_block *sb = vfsmnt->mnt_sb;
1510 static void server_umount_begin(struct super_block *sb)
1513 struct lustre_sb_info *lsi = s2lsi(sb);
1516 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1517 if (!(flags & MNT_FORCE)) {
1523 CDEBUG(D_MOUNT, "umount -f\n");
1524 /* umount = failover
1526 no third way to do non-force, non-failover */
1527 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1528 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1532 #ifndef HAVE_STATFS_DENTRY_PARAM
1533 static int server_statfs (struct super_block *sb, struct kstatfs *buf)
1536 static int server_statfs (struct dentry *dentry, struct kstatfs *buf)
1538 struct super_block *sb = dentry->d_sb;
1540 struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt;
1543 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) {
1544 #ifdef HAVE_STATFS_DENTRY_PARAM
1545 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf);
1547 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_sb, buf);
1550 buf->f_type = sb->s_magic;
1556 buf->f_type = sb->s_magic;
1557 buf->f_bsize = sb->s_blocksize;
1563 buf->f_namelen = NAME_MAX;
1567 static struct super_operations server_ops =
1569 .put_super = server_put_super,
1570 .umount_begin = server_umount_begin, /* umount -f */
1571 .statfs = server_statfs,
1574 #define log2(n) ffz(~(n))
1575 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1577 static int server_fill_super_common(struct super_block *sb)
1579 struct inode *root = 0;
1582 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1584 sb->s_blocksize = 4096;
1585 sb->s_blocksize_bits = log2(sb->s_blocksize);
1586 sb->s_magic = LUSTRE_SUPER_MAGIC;
1587 sb->s_maxbytes = 0; //PAGE_CACHE_MAXBYTES;
1588 sb->s_flags |= MS_RDONLY;
1589 sb->s_op = &server_ops;
1591 root = new_inode(sb);
1593 CERROR("Can't make root inode\n");
1597 /* returns -EIO for every operation */
1598 /* make_bad_inode(root); -- badness - can't umount */
1599 /* apparently we need to be a directory for the mount to finish */
1600 root->i_mode = S_IFDIR;
1602 sb->s_root = d_alloc_root(root);
1604 CERROR("Can't make root dentry\n");
1612 static int server_fill_super(struct super_block *sb)
1614 struct lustre_sb_info *lsi = s2lsi(sb);
1615 struct vfsmount *mnt;
1619 /* the One True Mount */
1620 mnt = server_kernel_mount(sb);
1623 CERROR("Unable to mount device %s: %d\n",
1624 lsi->lsi_lmd->lmd_dev, rc);
1628 lsi->lsi_srv_mnt = mnt;
1630 LASSERT(lsi->lsi_ldd);
1631 CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n",
1632 lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname,
1633 lsi->lsi_lmd->lmd_dev);
1635 if (class_name2obd(lsi->lsi_ldd->ldd_svname)) {
1636 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1637 "running. Double-mount may have compromised"
1638 " the disk journal.\n",
1639 lsi->lsi_ldd->ldd_svname);
1645 /* Start MGS before MGC */
1646 if (IS_MGS(lsi->lsi_ldd) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
1647 rc = server_start_mgs(sb);
1652 rc = lustre_start_mgc(sb);
1656 /* Set up all obd devices for service */
1657 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1658 (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) {
1659 rc = server_start_targets(sb, mnt);
1661 CERROR("Unable to start targets: %d\n", rc);
1664 /* FIXME overmount client here,
1665 or can we just start a client log and client_fill_super on this sb?
1666 We need to make sure server_put_super gets called too - ll_put_super
1667 calls lustre_common_put_super; check there for LSI_SERVER flag,
1669 Probably should start client from new thread so we can return.
1670 Client will not finish until all servers are connected.
1671 Note - MGS-only server does NOT get a client, since there is no
1672 lustre fs associated - the MGS is for all lustre fs's */
1675 rc = server_fill_super_common(sb);
1679 LCONSOLE_WARN("Server %s on device %s has started\n",
1680 ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1681 (IS_MDT(lsi->lsi_ldd))) ? "MGS" : lsi->lsi_ldd->ldd_svname,
1682 lsi->lsi_lmd->lmd_dev);
1686 /* We jump here in case of failure while starting targets or MGS.
1687 * In this case we can't just put @mnt and have to do real cleanup
1688 * with stoping targets, etc. */
1689 server_put_super(sb);
1693 /* Get the index from the obd name.
1694 rc = server type, or
1696 if endptr isn't NULL it is set to end of name */
1697 int server_name2index(char *svname, __u32 *idx, char **endptr)
1699 unsigned long index;
1701 char *dash = strrchr(svname, '-');
1705 /* intepret <fsname>-MDTXXXXX-mdc as mdt, the better way is to pass
1706 * in the fsname, then determine the server index */
1707 if (!strcmp(LUSTRE_MDC_NAME, dash + 1)) {
1709 for (; dash > svname && *dash != '-'; dash--);
1714 if (strncmp(dash + 1, "MDT", 3) == 0)
1715 rc = LDD_F_SV_TYPE_MDT;
1716 else if (strncmp(dash + 1, "OST", 3) == 0)
1717 rc = LDD_F_SV_TYPE_OST;
1720 if (strcmp(dash + 4, "all") == 0)
1721 return rc | LDD_F_SV_ALL;
1723 index = simple_strtoul(dash + 4, endptr, 16);
1728 /*************** mount common betweeen server and client ***************/
1731 int lustre_common_put_super(struct super_block *sb)
1736 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1738 /* Drop a ref to the MGC */
1739 rc = lustre_stop_mgc(sb);
1740 if (rc && (rc != -ENOENT)) {
1742 CERROR("Can't stop MGC: %d\n", rc);
1745 /* BUSY just means that there's some other obd that
1746 needs the mgc. Let him clean it up. */
1747 CDEBUG(D_MOUNT, "MGC still in use\n");
1749 /* Drop a ref to the mounted disk */
1756 static void lmd_print(struct lustre_mount_data *lmd)
1760 PRINT_CMD(PRINT_MASK, " mount data:\n");
1761 if (lmd_is_client(lmd))
1762 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
1763 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
1764 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
1766 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
1767 for (i = 0; i < lmd->lmd_exclude_count; i++) {
1768 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
1769 lmd->lmd_exclude[i]);
1774 /* Is this server on the exclusion list */
1775 int lustre_check_exclusion(struct super_block *sb, char *svname)
1777 struct lustre_sb_info *lsi = s2lsi(sb);
1778 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1783 rc = server_name2index(svname, &index, NULL);
1784 if (rc != LDD_F_SV_TYPE_OST)
1785 /* Only exclude OSTs */
1788 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
1789 index, lmd->lmd_exclude_count, lmd->lmd_dev);
1791 for(i = 0; i < lmd->lmd_exclude_count; i++) {
1792 if (index == lmd->lmd_exclude[i]) {
1793 CWARN("Excluding %s (on exclusion list)\n", svname);
1800 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
1801 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
1803 char *s1 = ptr, *s2;
1804 __u32 index, *exclude_list;
1808 /* The shortest an ost name can be is 8 chars: -OST0000.
1809 We don't actually know the fsname at this time, so in fact
1810 a user could specify any fsname. */
1811 devmax = strlen(ptr) / 8 + 1;
1813 /* temp storage until we figure out how many we have */
1814 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
1818 /* we enter this fn pointing at the '=' */
1819 while (*s1 && *s1 != ' ' && *s1 != ',') {
1821 rc = server_name2index(s1, &index, &s2);
1823 CERROR("Can't parse server name '%s'\n", s1);
1826 if (rc == LDD_F_SV_TYPE_OST)
1827 exclude_list[lmd->lmd_exclude_count++] = index;
1829 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
1831 /* now we are pointing at ':' (next exclude)
1832 or ',' (end of excludes) */
1833 if (lmd->lmd_exclude_count >= devmax)
1836 if (rc >= 0) /* non-err */
1839 if (lmd->lmd_exclude_count) {
1840 /* permanent, freed in lustre_free_lsi */
1841 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
1842 lmd->lmd_exclude_count);
1843 if (lmd->lmd_exclude) {
1844 memcpy(lmd->lmd_exclude, exclude_list,
1845 sizeof(index) * lmd->lmd_exclude_count);
1848 lmd->lmd_exclude_count = 0;
1851 OBD_FREE(exclude_list, sizeof(index) * devmax);
1855 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
1860 if (lmd->lmd_mgssec != NULL) {
1861 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
1862 lmd->lmd_mgssec = NULL;
1865 tail = strchr(ptr, ',');
1867 length = strlen(ptr);
1869 length = tail - ptr;
1871 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
1872 if (lmd->lmd_mgssec == NULL)
1875 memcpy(lmd->lmd_mgssec, ptr, length);
1876 lmd->lmd_mgssec[length] = '\0';
1880 /* mount -v -t lustre uml1:uml2:/lustre-client /mnt/lustre */
1881 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1883 char *s1, *s2, *devname = NULL;
1884 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1890 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1891 "/sbin/mount.lustre is installed.\n");
1895 /* Options should be a string - try to detect old lmd data */
1896 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1897 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1898 "/sbin/mount.lustre. Please install "
1899 "version %s\n", LUSTRE_VERSION_STRING);
1902 lmd->lmd_magic = LMD_MAGIC;
1904 /* Set default flags here */
1909 /* Skip whitespace and extra commas */
1910 while (*s1 == ' ' || *s1 == ',')
1913 /* Client options are parsed in ll_options: eg. flock,
1916 /* Parse non-ldiskfs options here. Rather than modifying
1917 ldiskfs, we just zero these out here */
1918 if (strncmp(s1, "abort_recov", 11) == 0) {
1919 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1921 } else if (strncmp(s1, "nosvc", 5) == 0) {
1922 lmd->lmd_flags |= LMD_FLG_NOSVC;
1924 } else if (strncmp(s1, "nomgs", 5) == 0) {
1925 lmd->lmd_flags |= LMD_FLG_NOMGS;
1927 } else if (strncmp(s1, "mgssec=", 7) == 0) {
1928 rc = lmd_parse_mgssec(lmd, s1 + 7);
1932 /* ost exclusion list */
1933 } else if (strncmp(s1, "exclude=", 8) == 0) {
1934 rc = lmd_make_exclusion(lmd, s1 + 7);
1939 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1940 end of the options. */
1941 else if (strncmp(s1, "device=", 7) == 0) {
1943 /* terminate options right before device. device
1944 must be the last one. */
1950 s2 = strchr(s1, ',');
1958 memmove(s1, s2, strlen(s2) + 1);
1964 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1965 "(need mount option 'device=...')\n");
1969 s1 = strstr(devname, ":/");
1972 lmd->lmd_flags = LMD_FLG_CLIENT;
1973 /* Remove leading /s from fsname */
1974 while (*++s1 == '/') ;
1975 /* Freed in lustre_free_lsi */
1976 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1977 if (!lmd->lmd_profile)
1979 sprintf(lmd->lmd_profile, "%s-client", s1);
1982 /* Freed in lustre_free_lsi */
1983 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1986 strcpy(lmd->lmd_dev, devname);
1988 /* Save mount options */
1989 s1 = options + strlen(options) - 1;
1990 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1992 if (*options != 0) {
1993 /* Freed in lustre_free_lsi */
1994 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1997 strcpy(lmd->lmd_opts, options);
2000 lmd->lmd_magic = LMD_MAGIC;
2005 CERROR("Bad mount options %s\n", options);
2011 int lustre_fill_super(struct super_block *sb, void *data, int silent)
2013 struct lustre_mount_data *lmd;
2014 struct lustre_sb_info *lsi;
2018 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
2020 lsi = lustre_init_lsi(sb);
2026 * Disable lockdep during mount, because mount locking patterns are
2031 /* Figure out the lmd from the mount options */
2032 if (lmd_parse((char *)data, lmd)) {
2034 GOTO(out, rc = -EINVAL);
2037 if (lmd_is_client(lmd)) {
2038 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
2039 if (!client_fill_super) {
2040 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
2041 "client mount! Is the 'lustre' "
2042 "module loaded?\n");
2046 rc = lustre_start_mgc(sb);
2051 /* Connect and start */
2052 /* (should always be ll_fill_super) */
2053 rc = (*client_fill_super)(sb);
2054 /* c_f_s will call lustre_common_put_super on failure */
2057 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
2058 lsi->lsi_flags |= LSI_SERVER;
2059 rc = server_fill_super(sb);
2060 /* s_f_s calls lustre_start_mgc after the mount because we need
2061 the MGS nids which are stored on disk. Plus, we may
2062 need to start the MGS first. */
2063 /* s_f_s will call server_put_super on failure */
2066 /* If error happens in fill_super() call, @lsi will be killed there.
2067 * This is why we do not put it here. */
2071 CERROR("Unable to mount %s (%d)\n",
2072 s2lsi(sb) ? lmd->lmd_dev : "", rc);
2074 CDEBUG(D_SUPER, "Mount %s complete\n",
2082 /* We can't call ll_fill_super by name because it lives in a module that
2083 must be loaded after this one. */
2084 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb))
2086 client_fill_super = cfs;
2089 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
2091 kill_super_cb = cfs;
2094 /***************** FS registration ******************/
2096 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))
2097 struct super_block * lustre_get_sb(struct file_system_type *fs_type,
2098 int flags, const char *devname, void * data)
2100 return get_sb_nodev(fs_type, flags, data, lustre_fill_super);
2103 int lustre_get_sb(struct file_system_type *fs_type,
2104 int flags, const char *devname, void * data,
2105 struct vfsmount *mnt)
2107 return get_sb_nodev(fs_type, flags, data, lustre_fill_super, mnt);
2111 void lustre_kill_super(struct super_block *sb)
2113 struct lustre_sb_info *lsi = s2lsi(sb);
2115 if (kill_super_cb && lsi && !(lsi->lsi_flags & LSI_SERVER))
2116 (*kill_super_cb)(sb);
2118 kill_anon_super(sb);
2121 struct file_system_type lustre_fs_type = {
2122 .owner = THIS_MODULE,
2124 .get_sb = lustre_get_sb,
2125 .kill_sb = lustre_kill_super,
2126 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
2127 #ifdef HAVE_LINUX_FIEMAP_H
2130 LL_RENAME_DOES_D_MOVE,
2133 int lustre_register_fs(void)
2135 return register_filesystem(&lustre_fs_type);
2138 int lustre_unregister_fs(void)
2140 return unregister_filesystem(&lustre_fs_type);
2143 EXPORT_SYMBOL(lustre_register_client_fill_super);
2144 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2145 EXPORT_SYMBOL(lustre_common_put_super);
2146 EXPORT_SYMBOL(lustre_process_log);
2147 EXPORT_SYMBOL(lustre_end_log);
2148 EXPORT_SYMBOL(server_get_mount);
2149 EXPORT_SYMBOL(server_get_mount_2);
2150 EXPORT_SYMBOL(server_put_mount);
2151 EXPORT_SYMBOL(server_put_mount_2);
2152 EXPORT_SYMBOL(server_register_target);
2153 EXPORT_SYMBOL(server_name2index);
2154 EXPORT_SYMBOL(server_mti_print);
2155 EXPORT_SYMBOL(do_lcfg);