1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * lustre/obdclass/obd_mount.c
5 * Client/server mount routines
7 * Copyright (c) 2006 Cluster File Systems, Inc.
8 * Author: Nathan Rutman <nathan@clusterfs.com>
10 * This file is part of Lustre, http://www.lustre.org/
12 * Lustre is free software; you can redistribute it and/or
13 * modify it under the terms of version 2 of the GNU General Public
14 * License as published by the Free Software Foundation.
16 * Lustre is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with Lustre; if not, write to the Free Software
23 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
27 #define DEBUG_SUBSYSTEM S_CLASS
28 #define D_MOUNT D_SUPER|D_CONFIG /*|D_WARNING */
29 #define PRINT_CMD CDEBUG
30 #define PRINT_MASK D_SUPER|D_CONFIG
34 #include <lustre_fsfilt.h>
35 #include <obd_class.h>
36 #include <lustre/lustre_user.h>
37 #include <linux/version.h>
38 #include <lustre_log.h>
39 #include <lustre_disk.h>
40 #include <lustre_param.h>
42 static int (*client_fill_super)(struct super_block *sb) = NULL;
43 static void (*kill_super_cb)(struct super_block *sb) = NULL;
45 /*********** mount lookup *********/
47 DECLARE_MUTEX(lustre_mount_info_lock);
48 struct list_head server_mount_info_list = LIST_HEAD_INIT(server_mount_info_list);
50 static struct lustre_mount_info *server_find_mount(char *name)
52 struct list_head *tmp;
53 struct lustre_mount_info *lmi;
56 list_for_each(tmp, &server_mount_info_list) {
57 lmi = list_entry(tmp, struct lustre_mount_info, lmi_list_chain);
58 if (strcmp(name, lmi->lmi_name) == 0)
64 /* we must register an obd for a mount before we call the setup routine.
65 *_setup will call lustre_get_mount to get the mnt struct
66 by obd_name, since we can't pass the pointer to setup. */
67 static int server_register_mount(char *name, struct super_block *sb,
70 struct lustre_mount_info *lmi;
77 OBD_ALLOC(lmi, sizeof(*lmi));
80 OBD_ALLOC(name_cp, strlen(name) + 1);
82 OBD_FREE(lmi, sizeof(*lmi));
85 strcpy(name_cp, name);
87 down(&lustre_mount_info_lock);
89 if (server_find_mount(name)) {
90 up(&lustre_mount_info_lock);
91 OBD_FREE(lmi, sizeof(*lmi));
92 OBD_FREE(name_cp, strlen(name) + 1);
93 CERROR("Already registered %s\n", name);
96 lmi->lmi_name = name_cp;
99 list_add(&lmi->lmi_list_chain, &server_mount_info_list);
101 up(&lustre_mount_info_lock);
103 CDEBUG(D_MOUNT, "reg_mnt %p from %s, vfscount=%d\n",
104 lmi->lmi_mnt, name, atomic_read(&lmi->lmi_mnt->mnt_count));
109 /* when an obd no longer needs a mount */
110 static int server_deregister_mount(char *name)
112 struct lustre_mount_info *lmi;
115 down(&lustre_mount_info_lock);
116 lmi = server_find_mount(name);
118 up(&lustre_mount_info_lock);
119 CERROR("%s not registered\n", name);
123 CDEBUG(D_MOUNT, "dereg_mnt %p from %s, vfscount=%d\n",
124 lmi->lmi_mnt, name, atomic_read(&lmi->lmi_mnt->mnt_count));
126 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
127 list_del(&lmi->lmi_list_chain);
128 OBD_FREE(lmi, sizeof(*lmi));
129 up(&lustre_mount_info_lock);
134 /* obd's look up a registered mount using their obdname. This is just
135 for initial obd setup to find the mount struct. It should not be
136 called every time you want to mntget. */
137 struct lustre_mount_info *server_get_mount(char *name)
139 struct lustre_mount_info *lmi;
140 struct lustre_sb_info *lsi;
143 down(&lustre_mount_info_lock);
144 lmi = server_find_mount(name);
145 up(&lustre_mount_info_lock);
147 CERROR("Can't find mount for %s\n", name);
150 lsi = s2lsi(lmi->lmi_sb);
151 mntget(lmi->lmi_mnt);
152 atomic_inc(&lsi->lsi_mounts);
154 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d, vfscount=%d\n",
155 lmi->lmi_mnt, name, atomic_read(&lsi->lsi_mounts),
156 atomic_read(&lmi->lmi_mnt->mnt_count));
161 static void unlock_mntput(struct vfsmount *mnt)
163 if (kernel_locked()) {
172 static int lustre_put_lsi(struct super_block *sb);
174 /* to be called from obd_cleanup methods */
175 int server_put_mount(char *name, struct vfsmount *mnt)
177 struct lustre_mount_info *lmi;
178 struct lustre_sb_info *lsi;
179 int count = atomic_read(&mnt->mnt_count) - 1;
182 /* This might be the last one, can't deref after this */
185 down(&lustre_mount_info_lock);
186 lmi = server_find_mount(name);
187 up(&lustre_mount_info_lock);
189 CERROR("Can't find mount for %s\n", name);
192 LASSERT(lmi->lmi_mnt == mnt);
193 /* Note if lsi is poisoned, it's probably because server_wait_finished
194 didn't wait long enough and the sb was freed at the umount. */
195 lsi = s2lsi(lmi->lmi_sb);
197 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d, vfscount=%d\n",
198 lmi->lmi_mnt, name, atomic_read(&lsi->lsi_mounts), count);
200 if (lustre_put_lsi(lmi->lmi_sb)) {
201 CDEBUG(D_MOUNT, "Last put of mnt %p from %s, vfscount=%d\n",
202 lmi->lmi_mnt, name, count);
203 /* last mount is the One True Mount */
205 CERROR("%s: mount busy, vfscount=%d!\n", name, count);
208 /* this obd should never need the mount again */
209 server_deregister_mount(name);
215 /******* mount helper utilities *********/
217 static void ldd_print(struct lustre_disk_data *ldd)
219 PRINT_CMD(PRINT_MASK, " disk data:\n");
220 PRINT_CMD(PRINT_MASK, "server: %s\n", ldd->ldd_svname);
221 PRINT_CMD(PRINT_MASK, "uuid: %s\n", (char *)ldd->ldd_uuid);
222 PRINT_CMD(PRINT_MASK, "fs: %s\n", ldd->ldd_fsname);
223 PRINT_CMD(PRINT_MASK, "index: %04x\n", ldd->ldd_svindex);
224 PRINT_CMD(PRINT_MASK, "config: %d\n", ldd->ldd_config_ver);
225 PRINT_CMD(PRINT_MASK, "flags: %#x\n", ldd->ldd_flags);
226 PRINT_CMD(PRINT_MASK, "diskfs: %s\n", MT_STR(ldd));
227 PRINT_CMD(PRINT_MASK, "options: %s\n", ldd->ldd_mount_opts);
228 PRINT_CMD(PRINT_MASK, "params: %s\n", ldd->ldd_params);
229 PRINT_CMD(PRINT_MASK, "comment: %s\n", ldd->ldd_userdata);
232 static int ldd_parse(struct lvfs_run_ctxt *mount_ctxt,
233 struct lustre_disk_data *ldd)
235 struct lvfs_run_ctxt saved;
242 push_ctxt(&saved, mount_ctxt, NULL);
244 file = filp_open(MOUNT_DATA_FILE, O_RDONLY, 0644);
247 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
251 len = i_size_read(file->f_dentry->d_inode);
252 CDEBUG(D_MOUNT, "Have %s, size %lu\n", MOUNT_DATA_FILE, len);
253 if (len != sizeof(*ldd)) {
254 CERROR("disk data size does not match: see %lu expect "LPSZ"\n",
256 GOTO(out_close, rc = -EINVAL);
259 rc = lustre_fread(file, ldd, len, &off);
261 CERROR("error reading %s: read %d of %lu\n",
262 MOUNT_DATA_FILE, rc, len);
263 GOTO(out_close, rc = -EINVAL);
267 if (ldd->ldd_magic != LDD_MAGIC) {
268 /* FIXME add swabbing support */
269 CERROR("Bad magic in %s: %x!=%x\n", MOUNT_DATA_FILE,
270 ldd->ldd_magic, LDD_MAGIC);
271 GOTO(out_close, rc = -EINVAL);
274 if (ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP) {
275 CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
277 ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP);
278 GOTO(out_close, rc = -EINVAL);
280 if (ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP) {
281 CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
283 ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP);
284 /* Do something like remount filesystem read-only */
285 GOTO(out_close, rc = -EINVAL);
293 pop_ctxt(&saved, mount_ctxt, NULL);
297 static int ldd_write(struct lvfs_run_ctxt *mount_ctxt,
298 struct lustre_disk_data *ldd)
300 struct lvfs_run_ctxt saved;
303 unsigned long len = sizeof(struct lustre_disk_data);
307 LASSERT(ldd->ldd_magic == LDD_MAGIC);
309 ldd->ldd_config_ver++;
311 push_ctxt(&saved, mount_ctxt, NULL);
313 file = filp_open(MOUNT_DATA_FILE, O_RDWR, 0644);
316 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
320 rc = lustre_fwrite(file, ldd, len, &off);
322 CERROR("error writing %s: read %d of %lu\n",
323 MOUNT_DATA_FILE, rc, len);
324 GOTO(out_close, rc = -EINVAL);
333 pop_ctxt(&saved, mount_ctxt, NULL);
338 /**************** config llog ********************/
340 /* Get a config log from the MGS and process it.
341 This func is called for both clients and servers.
342 Continue to process new statements appended to the logs
343 (whenever the config lock is revoked) until lustre_end_log
345 int lustre_process_log(struct super_block *sb, char *logname,
346 struct config_llog_instance *cfg)
348 struct lustre_cfg *lcfg;
349 struct lustre_cfg_bufs bufs;
350 struct lustre_sb_info *lsi = s2lsi(sb);
351 struct obd_device *mgc = lsi->lsi_mgc;
358 /* mgc_process_config */
359 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
360 lustre_cfg_bufs_set_string(&bufs, 1, logname);
361 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
362 lustre_cfg_bufs_set(&bufs, 3, &sb, sizeof(sb));
363 lcfg = lustre_cfg_new(LCFG_LOG_START, &bufs);
364 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
365 lustre_cfg_free(lcfg);
368 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s' "
369 "failed (%d). Make sure this client and "
370 "the MGS are running compatible versions of "
372 mgc->obd_name, logname, rc);
375 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
376 "failed (%d). This may be the result of "
377 "communication errors between this node and "
378 "the MGS, a bad configuration, or other "
379 "errors. See the syslog for more "
380 "information.\n", mgc->obd_name, logname,
383 /* class_obd_list(); */
387 /* Stop watching this config log for updates */
388 int lustre_end_log(struct super_block *sb, char *logname,
389 struct config_llog_instance *cfg)
391 struct lustre_cfg *lcfg;
392 struct lustre_cfg_bufs bufs;
393 struct lustre_sb_info *lsi = s2lsi(sb);
394 struct obd_device *mgc = lsi->lsi_mgc;
401 /* mgc_process_config */
402 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
403 lustre_cfg_bufs_set_string(&bufs, 1, logname);
405 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
406 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
407 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
408 lustre_cfg_free(lcfg);
412 /**************** obd start *******************/
414 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
415 char *s1, char *s2, char *s3, char *s4)
417 struct lustre_cfg_bufs bufs;
418 struct lustre_cfg * lcfg = NULL;
421 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
422 cmd, s1, s2, s3, s4);
424 lustre_cfg_bufs_reset(&bufs, cfgname);
426 lustre_cfg_bufs_set_string(&bufs, 1, s1);
428 lustre_cfg_bufs_set_string(&bufs, 2, s2);
430 lustre_cfg_bufs_set_string(&bufs, 3, s3);
432 lustre_cfg_bufs_set_string(&bufs, 4, s4);
434 lcfg = lustre_cfg_new(cmd, &bufs);
435 lcfg->lcfg_nid = nid;
436 rc = class_process_config(lcfg);
437 lustre_cfg_free(lcfg);
441 static int lustre_start_simple(char *obdname, char *type, char *uuid,
445 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
447 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
449 CERROR("%s attach error %d\n", obdname, rc);
452 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, 0, 0);
454 CERROR("%s setup error %d\n", obdname, rc);
455 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
460 /* Set up a MGS to serve startup logs */
461 static int server_start_mgs(struct super_block *sb)
463 struct lustre_sb_info *lsi = s2lsi(sb);
464 struct vfsmount *mnt = lsi->lsi_srv_mnt;
465 struct lustre_mount_info *lmi;
470 /* It is impossible to have more than 1 MGS per node, since
471 MGC wouldn't know which to connect to */
472 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
474 lsi = s2lsi(lmi->lmi_sb);
475 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started "
476 "from server %s\n", lsi->lsi_ldd->ldd_svname);
480 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
482 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
485 ((rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
486 LUSTRE_MGS_OBDNAME, 0, 0))))
487 server_deregister_mount(LUSTRE_MGS_OBDNAME);
490 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). Is "
491 "the 'mgs' module loaded?\n",
492 LUSTRE_MGS_OBDNAME, rc);
497 static int server_stop_mgs(struct super_block *sb)
499 struct obd_device *obd;
503 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
505 /* There better be only one MGS */
506 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
508 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
512 /* The MGS should always stop when we say so */
514 rc = class_manual_cleanup(obd);
518 DECLARE_MUTEX(mgc_start_lock);
520 /* Set up a mgcobd to process startup logs */
521 static int lustre_start_mgc(struct super_block *sb)
523 struct lustre_handle mgc_conn = {0, };
524 struct obd_connect_data *data = NULL;
525 struct lustre_sb_info *lsi = s2lsi(sb);
526 struct obd_device *obd;
527 struct obd_export *exp;
528 struct obd_uuid *uuid;
531 char *mgcname, *niduuid;
534 int rc = 0, i = 0, j, len;
537 LASSERT(lsi->lsi_lmd);
539 /* Find the first non-lo MGS nid for our MGC name */
540 if (lsi->lsi_flags & LSI_SERVER) {
541 ptr = lsi->lsi_ldd->ldd_params;
542 /* Use mgsnode= nids */
543 if ((class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0) &&
544 (class_parse_nid(ptr, &nid, &ptr) == 0)) {
546 } else if (IS_MGS(lsi->lsi_ldd)) {
547 lnet_process_id_t id;
548 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
549 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
556 } else { /* client */
557 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
558 ptr = lsi->lsi_lmd->lmd_dev;
559 if (class_parse_nid(ptr, &nid, &ptr) == 0)
563 CERROR("No valid MGS nids found.\n");
567 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
568 OBD_ALLOC(mgcname, len);
569 OBD_ALLOC(niduuid, len + 2);
570 if (!mgcname || !niduuid)
571 GOTO(out_free, rc = -ENOMEM);
572 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
574 mutex_down(&mgc_start_lock);
576 obd = class_name2obd(mgcname);
578 /* Re-using an existing MGC */
579 atomic_inc(&obd->u.cli.cl_mgc_refcount);
582 /* If we are restarting the MGS, don't try to keep the MGC's
583 old connection, or registration will fail. */
584 if ((lsi->lsi_flags & LSI_SERVER) && IS_MGS(lsi->lsi_ldd)) {
585 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
589 /* Try all connections, but only once (again).
590 We don't want to block another target from starting
591 (using its local copy of the log), but we do want to connect
592 if at all possible. */
594 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
595 rc = obd_set_info_async(obd->obd_self_export,
596 strlen(KEY_INIT_RECOV_BACKUP),
597 KEY_INIT_RECOV_BACKUP,
598 sizeof(recov_bk), &recov_bk, NULL);
602 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
604 /* Add the primary nids for the MGS */
606 sprintf(niduuid, "%s_%x", mgcname, i);
607 if (lsi->lsi_flags & LSI_SERVER) {
608 ptr = lsi->lsi_ldd->ldd_params;
609 if (IS_MGS(lsi->lsi_ldd)) {
610 /* Use local nids (including LO) */
611 lnet_process_id_t id;
612 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
613 rc = do_lcfg(mgcname, id.nid,
614 LCFG_ADD_UUID, niduuid, 0,0,0);
617 /* Use mgsnode= nids */
618 if (class_find_param(ptr, PARAM_MGSNODE, &ptr) != 0) {
619 CERROR("No MGS nids given.\n");
620 GOTO(out_free, rc = -EINVAL);
622 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
623 rc = do_lcfg(mgcname, nid,
624 LCFG_ADD_UUID, niduuid, 0,0,0);
628 } else { /* client */
629 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
630 ptr = lsi->lsi_lmd->lmd_dev;
631 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
632 rc = do_lcfg(mgcname, nid,
633 LCFG_ADD_UUID, niduuid, 0,0,0);
635 /* Stop at the first failover nid */
641 CERROR("No valid MGS nids found.\n");
642 GOTO(out_free, rc = -EINVAL);
644 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
646 /* Random uuid for MGC allows easier reconnects */
648 ll_generate_random_uuid(uuidc);
649 class_uuid_unparse(uuidc, uuid);
652 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
653 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
659 /* Add any failover MGS nids */
661 while ((*ptr == ':' ||
662 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0)) {
663 /* New failover node */
664 sprintf(niduuid, "%s_%x", mgcname, i);
666 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
668 rc = do_lcfg(mgcname, nid,
669 LCFG_ADD_UUID, niduuid, 0,0,0);
674 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
682 lsi->lsi_lmd->lmd_mgs_failnodes = i;
684 obd = class_name2obd(mgcname);
686 CERROR("Can't find mgcobd %s\n", mgcname);
687 GOTO(out_free, rc = -ENOTCONN);
690 /* Keep a refcount of servers/clients who started with "mount",
691 so we know when we can get rid of the mgc. */
692 atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
694 /* Try all connections, but only once. */
696 rc = obd_set_info_async(obd->obd_self_export,
697 strlen(KEY_INIT_RECOV_BACKUP),
698 KEY_INIT_RECOV_BACKUP,
699 sizeof(recov_bk), &recov_bk, NULL);
702 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
706 GOTO(out, rc = -ENOMEM);
707 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT;
708 data->ocd_version = LUSTRE_VERSION_CODE;
709 /* We connect to the MGS at setup, and don't disconnect until cleanup */
710 rc = obd_connect(&mgc_conn, obd, &(obd->obd_uuid), data, NULL);
713 CERROR("connect failed %d\n", rc);
717 exp = class_conn2export(&mgc_conn);
718 obd->u.cli.cl_mgc_mgsexp = exp;
721 /* Keep the mgc info in the sb. Note that many lsi's can point
725 mutex_up(&mgc_start_lock);
728 OBD_FREE(mgcname, len);
730 OBD_FREE(niduuid, len + 2);
734 static int lustre_stop_mgc(struct super_block *sb)
736 struct lustre_sb_info *lsi = s2lsi(sb);
737 struct obd_device *obd;
738 char *niduuid = 0, *ptr = 0;
739 int i, rc = 0, len = 0;
749 mutex_down(&mgc_start_lock);
750 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
751 /* This is not fatal, every client that stops
752 will call in here. */
753 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
754 atomic_read(&obd->u.cli.cl_mgc_refcount));
755 GOTO(out, rc = -EBUSY);
758 /* The MGC has no recoverable data in any case.
759 * force shotdown set in umount_begin */
760 obd->obd_no_recov = 1;
762 if (obd->u.cli.cl_mgc_mgsexp) {
763 /* An error is not fatal, if we are unable to send the
764 disconnect mgs ping evictor cleans up the export */
765 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
767 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
770 /* Save the obdname for cleaning the nid uuids, which are
772 len = strlen(obd->obd_name) + 6;
773 OBD_ALLOC(niduuid, len);
775 strcpy(niduuid, obd->obd_name);
776 ptr = niduuid + strlen(niduuid);
779 rc = class_manual_cleanup(obd);
783 /* Clean the nid uuids */
785 GOTO(out, rc = -ENOMEM);
787 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
788 sprintf(ptr, "_%x", i);
789 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
792 CERROR("del MDC UUID %s failed: rc = %d\n",
797 OBD_FREE(niduuid, len);
799 /* class_import_put will get rid of the additional connections */
800 mutex_up(&mgc_start_lock);
804 /* Since there's only one mgc per node, we have to change it's fs to get
805 access to the right disk. */
806 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
808 struct lustre_sb_info *lsi = s2lsi(sb);
812 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
814 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
815 rc = obd_set_info_async(mgc->obd_self_export,
816 strlen("set_fs"), "set_fs",
817 sizeof(*sb), sb, NULL);
819 CERROR("can't set_fs %d\n", rc);
825 static int server_mgc_clear_fs(struct obd_device *mgc)
830 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
832 rc = obd_set_info_async(mgc->obd_self_export,
833 strlen("clear_fs"), "clear_fs",
838 DECLARE_MUTEX(server_start_lock);
840 /* Stop MDS/OSS if nobody is using them */
841 static int server_stop_servers(int lddflags, int lsiflags)
843 struct obd_device *obd = NULL;
844 struct obd_type *type = NULL;
848 mutex_down(&server_start_lock);
850 /* Either an MDT or an OST or neither */
851 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
852 if ((lddflags & LDD_F_SV_TYPE_MDT) &&
853 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
854 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
855 type = class_search_type(LUSTRE_MDS_NAME);
857 /* if this was an OST, and there are no more OST's, clean up the OSS */
858 if ((lddflags & LDD_F_SV_TYPE_OST) &&
859 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
860 type = class_search_type(LUSTRE_OST_NAME);
863 if (obd && (!type || !type->typ_refcnt)) {
866 /* obd_fail doesn't mean much on a server obd */
867 err = class_manual_cleanup(obd);
872 mutex_up(&server_start_lock);
877 int server_mti_print(char *title, struct mgs_target_info *mti)
879 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
880 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
881 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
882 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
883 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
884 mti->mti_config_ver, mti->mti_flags);
888 static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
890 struct lustre_sb_info *lsi = s2lsi(sb);
891 struct lustre_disk_data *ldd = lsi->lsi_ldd;
892 lnet_process_id_t id;
896 if (!(lsi->lsi_flags & LSI_SERVER))
899 strncpy(mti->mti_fsname, ldd->ldd_fsname,
900 sizeof(mti->mti_fsname));
901 strncpy(mti->mti_svname, ldd->ldd_svname,
902 sizeof(mti->mti_svname));
904 mti->mti_nid_count = 0;
905 while (LNetGetId(i++, &id) != -ENOENT) {
906 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
908 mti->mti_nids[mti->mti_nid_count] = id.nid;
909 mti->mti_nid_count++;
910 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
911 CWARN("Only using first %d nids for %s\n",
912 mti->mti_nid_count, mti->mti_svname);
917 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
918 mti->mti_config_ver = 0;
919 mti->mti_flags = ldd->ldd_flags;
920 mti->mti_stripe_index = ldd->ldd_svindex;
921 memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid));
922 if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) {
923 CERROR("params too big for mti\n");
926 memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params));
930 /* Register an old or new target with the MGS. If needed MGS will construct
931 startup logs and assign index */
932 int server_register_target(struct super_block *sb)
934 struct lustre_sb_info *lsi = s2lsi(sb);
935 struct obd_device *mgc = lsi->lsi_mgc;
936 struct lustre_disk_data *ldd = lsi->lsi_ldd;
937 struct mgs_target_info *mti = NULL;
943 if (!(lsi->lsi_flags & LSI_SERVER))
949 rc = server_sb2mti(sb, mti);
953 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
954 mti->mti_svname, mti->mti_fsname,
955 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
958 /* Register the target */
959 /* FIXME use mgc_process_config instead */
960 rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
961 strlen("register_target"), "register_target",
962 sizeof(*mti), mti, NULL);
966 /* Always update our flags */
967 ldd->ldd_flags = mti->mti_flags & ~LDD_F_REWRITE_LDD;
969 /* If this flag is set, it means the MGS wants us to change our
970 on-disk data. (So far this means just the index.) */
971 if (mti->mti_flags & LDD_F_REWRITE_LDD) {
974 CDEBUG(D_MOUNT, "Changing on-disk index from %#x to %#x "
975 "for %s\n", ldd->ldd_svindex, mti->mti_stripe_index,
977 ldd->ldd_svindex = mti->mti_stripe_index;
978 strncpy(ldd->ldd_svname, mti->mti_svname,
979 sizeof(ldd->ldd_svname));
980 /* or ldd_make_sv_name(ldd); */
981 ldd_write(&mgc->obd_lvfs_ctxt, ldd);
983 err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb,
986 CERROR("Label set error %d\n", err);
987 label = fsfilt_get_label(mgc, lsi->lsi_srv_mnt->mnt_sb);
989 CDEBUG(D_MOUNT, "Disk label changed to %s\n", label);
991 /* Flush the new ldd to disk */
992 fsfilt_sync(mgc, lsi->lsi_srv_mnt->mnt_sb);
1002 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1004 struct obd_device *obd;
1005 struct lustre_sb_info *lsi = s2lsi(sb);
1006 struct config_llog_instance cfg;
1010 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname);
1012 /* If we're an MDT, make sure the global MDS is running */
1013 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1014 /* make sure the MDS is started */
1015 mutex_down(&server_start_lock);
1016 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1018 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1019 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1021 LUSTRE_MDS_OBDNAME"_uuid",
1024 mutex_up(&server_start_lock);
1025 CERROR("failed to start MDS: %d\n", rc);
1029 mutex_up(&server_start_lock);
1032 /* If we're an OST, make sure the global OSS is running */
1033 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
1034 /* make sure OSS is started */
1035 mutex_down(&server_start_lock);
1036 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1038 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1040 LUSTRE_OSS_OBDNAME"_uuid",
1043 mutex_up(&server_start_lock);
1044 CERROR("failed to start OSS: %d\n", rc);
1048 mutex_up(&server_start_lock);
1051 /* Set the mgc fs to our server disk. This allows the MGC
1052 to read and write configs locally. */
1053 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1057 /* Register with MGS */
1058 rc = server_register_target(sb);
1059 if (rc && (lsi->lsi_ldd->ldd_flags &
1060 (LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_UPGRADE14))){
1061 CERROR("Required registration failed for %s: %d\n",
1062 lsi->lsi_ldd->ldd_svname, rc);
1064 LCONSOLE_ERROR_MSG(0x15f, "Communication error with the"
1065 " MGS. Is the MGS running?\n");
1069 if (rc == -EINVAL) {
1070 LCONSOLE_ERROR_MSG(0x160, "The MGS is refusing to allow this "
1071 "server (%s) to start. Please see messages "
1072 "on the MGS node.\n",
1073 lsi->lsi_ldd->ldd_svname);
1076 /* non-fatal error of registeration with MGS */
1078 CDEBUG(D_MOUNT, "Cannot register with MGS: %d\n", rc);
1080 /* Let the target look up the mount using the target's name
1081 (we can't pass the sb or mnt through class_process_config.) */
1082 rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
1086 /* Start targets using the llog named for the target */
1087 memset(&cfg, 0, sizeof(cfg));
1088 rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg);
1090 CERROR("failed to start server %s: %d\n",
1091 lsi->lsi_ldd->ldd_svname, rc);
1096 /* Release the mgc fs for others to use */
1097 server_mgc_clear_fs(lsi->lsi_mgc);
1100 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1102 CERROR("no server named %s was started\n",
1103 lsi->lsi_ldd->ldd_svname);
1107 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1108 (OBP(obd, iocontrol))) {
1109 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1110 obd->obd_self_export, 0, NULL, NULL);
1113 /* log has been fully processed */
1114 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1120 /***************** lustre superblock **************/
1122 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1124 struct lustre_sb_info *lsi = NULL;
1127 OBD_ALLOC(lsi, sizeof(*lsi));
1130 OBD_ALLOC(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1131 if (!lsi->lsi_lmd) {
1132 OBD_FREE(lsi, sizeof(*lsi));
1136 lsi->lsi_lmd->lmd_exclude_count = 0;
1137 s2lsi_nocast(sb) = lsi;
1138 /* we take 1 extra ref for our setup */
1139 atomic_set(&lsi->lsi_mounts, 1);
1141 /* Default umount style */
1142 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1146 static int lustre_free_lsi(struct super_block *sb)
1148 struct lustre_sb_info *lsi = s2lsi(sb);
1154 CDEBUG(D_MOUNT, "Freeing lsi\n");
1156 /* someone didn't call server_put_mount. */
1157 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
1159 if (lsi->lsi_ldd != NULL)
1160 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
1162 if (lsi->lsi_lmd != NULL) {
1163 if (lsi->lsi_lmd->lmd_dev != NULL)
1164 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1165 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1166 if (lsi->lsi_lmd->lmd_profile != NULL)
1167 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1168 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1169 if (lsi->lsi_lmd->lmd_opts != NULL)
1170 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1171 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1172 if (lsi->lsi_lmd->lmd_exclude_count)
1173 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1174 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1175 lsi->lsi_lmd->lmd_exclude_count);
1176 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1179 LASSERT(lsi->lsi_llsbi == NULL);
1181 OBD_FREE(lsi, sizeof(*lsi));
1182 s2lsi_nocast(sb) = NULL;
1187 /* The lsi has one reference for every server that is using the disk -
1188 e.g. MDT, MGS, and potentially MGC */
1189 static int lustre_put_lsi(struct super_block *sb)
1191 struct lustre_sb_info *lsi = s2lsi(sb);
1196 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
1198 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
1199 lustre_free_lsi(sb);
1205 /*************** server mount ******************/
1207 /* Kernel mount using mount options in MOUNT_DATA_FILE */
1208 static struct vfsmount *server_kernel_mount(struct super_block *sb)
1210 struct lvfs_run_ctxt mount_ctxt;
1211 struct lustre_sb_info *lsi = s2lsi(sb);
1212 struct lustre_disk_data *ldd;
1213 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1214 struct vfsmount *mnt;
1215 char *options = NULL;
1216 unsigned long page, s_flags;
1217 struct page *__page;
1221 OBD_ALLOC(ldd, sizeof(*ldd));
1223 RETURN(ERR_PTR(-ENOMEM));
1225 /* In the past, we have always used flags = 0.
1226 Note ext3/ldiskfs can't be mounted ro. */
1227 s_flags = sb->s_flags;
1229 /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */
1230 CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev);
1231 mnt = ll_kern_mount("ldiskfs", s_flags, lmd->lmd_dev, 0);
1234 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1235 /* 2.6 kernels: if ldiskfs fails, try ldiskfs2 */
1236 mnt = ll_kern_mount("ldiskfs2", s_flags, lmd->lmd_dev, 0);
1238 int rc2 = PTR_ERR(mnt);
1239 CERROR("premount %s:%#lx ldiskfs failed: %d, ldiskfs2 "
1240 "failed: %d. Is the ldiskfs module available?\n",
1241 lmd->lmd_dev, s_flags, rc, rc2);
1245 /* 2.4 kernels: if ldiskfs fails, try ext3 */
1246 mnt = ll_kern_mount("ext3", s_flags, lmd->lmd_dev, 0);
1249 CERROR("premount ext3 failed: rc = %d\n", rc);
1255 OBD_SET_CTXT_MAGIC(&mount_ctxt);
1256 mount_ctxt.pwdmnt = mnt;
1257 mount_ctxt.pwd = mnt->mnt_root;
1258 mount_ctxt.fs = get_ds();
1260 rc = ldd_parse(&mount_ctxt, ldd);
1264 CERROR("premount parse options failed: rc = %d\n", rc);
1268 /* Done with our pre-mount, now do the real mount. */
1270 /* Glom up mount options */
1271 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
1273 GOTO(out_free, rc = -ENOMEM);
1274 page = (unsigned long)cfs_page_address(__page);
1276 options = (char *)page;
1277 memset(options, 0, CFS_PAGE_SIZE);
1278 strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2);
1280 /* Add in any mount-line options */
1281 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) {
1282 int len = CFS_PAGE_SIZE - strlen(options) - 2;
1284 strcat(options, ",");
1285 strncat(options, lmd->lmd_opts, len);
1288 /* Special permanent mount flags */
1290 s_flags |= MS_NOATIME | MS_NODIRATIME;
1292 CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
1293 MT_STR(ldd), lmd->lmd_dev, options);
1294 mnt = ll_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev,
1296 OBD_PAGE_FREE(__page);
1299 CERROR("ll_kern_mount failed: rc = %d\n", rc);
1303 lsi->lsi_ldd = ldd; /* freed at lsi cleanup */
1304 CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
1308 OBD_FREE(ldd, sizeof(*ldd));
1309 lsi->lsi_ldd = NULL;
1310 RETURN(ERR_PTR(rc));
1313 static void server_wait_finished(struct vfsmount *mnt)
1315 wait_queue_head_t waitq;
1316 struct l_wait_info lwi;
1319 init_waitqueue_head(&waitq);
1321 while ((atomic_read(&mnt->mnt_count) > 1) && (retries > 0)) {
1322 LCONSOLE_WARN("Mount still busy with %d refs, waiting for "
1324 atomic_read(&mnt->mnt_count), retries);
1326 /* Wait for a bit */
1328 lwi = LWI_TIMEOUT(5 * HZ, NULL, NULL);
1329 l_wait_event(waitq, 0, &lwi);
1331 if (atomic_read(&mnt->mnt_count) > 1) {
1332 CERROR("Mount %p is still busy (%d refs), giving up.\n",
1333 mnt, atomic_read(&mnt->mnt_count));
1337 static void server_put_super(struct super_block *sb)
1339 struct lustre_sb_info *lsi = s2lsi(sb);
1340 struct obd_device *obd;
1341 struct vfsmount *mnt = lsi->lsi_srv_mnt;
1342 char *tmpname, *extraname = NULL;
1344 int lddflags = lsi->lsi_ldd->ldd_flags;
1345 int lsiflags = lsi->lsi_flags;
1349 LASSERT(lsiflags & LSI_SERVER);
1351 tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1;
1352 OBD_ALLOC(tmpname, tmpname_sz);
1353 memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
1354 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1356 /* Stop the target */
1357 if (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd)) {
1358 struct lustre_profile *lprof = NULL;
1360 /* tell the mgc to drop the config log */
1361 lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL);
1363 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1364 If there are any setup/cleanup errors, save the lov
1365 name for safety cleanup later. */
1366 lprof = class_get_profile(lsi->lsi_ldd->ldd_svname);
1367 if (lprof && lprof->lp_osc) {
1368 OBD_ALLOC(extraname, strlen(lprof->lp_osc) + 1);
1369 strcpy(extraname, lprof->lp_osc);
1372 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1374 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1375 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1377 /* We can't seem to give an error return code
1378 * to .put_super, so we better make sure we clean up! */
1380 class_manual_cleanup(obd);
1382 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
1383 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1388 /* If they wanted the mgs to stop separately from the mdt, they
1389 should have put it on a different device. */
1390 if (IS_MGS(lsi->lsi_ldd)) {
1391 /* stop the mgc before the mgs so the connection gets cleaned
1393 lustre_stop_mgc(sb);
1394 server_stop_mgs(sb);
1397 /* Clean the mgc and sb */
1398 rc = lustre_common_put_super(sb);
1399 /* FIXME how can I report a failure to umount? */
1401 /* Wait for the targets to really clean up - can't exit (and let the
1402 sb get destroyed) while the mount is still in use */
1403 server_wait_finished(mnt);
1405 /* drop the One True Mount */
1408 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1409 until the target is really gone so that our type refcount check
1411 server_stop_servers(lddflags, lsiflags);
1413 /* In case of startup or cleanup err, stop related obds */
1415 obd = class_name2obd(extraname);
1417 CWARN("Cleaning orphaned obd %s\n", extraname);
1419 class_manual_cleanup(obd);
1421 OBD_FREE(extraname, strlen(extraname) + 1);
1424 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1425 OBD_FREE(tmpname, tmpname_sz);
1428 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1429 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1431 struct super_block *sb = vfsmnt->mnt_sb;
1433 static void server_umount_begin(struct super_block *sb)
1436 struct lustre_sb_info *lsi = s2lsi(sb);
1439 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1440 if (!(flags & MNT_FORCE)) {
1446 CDEBUG(D_MOUNT, "umount -f\n");
1447 /* umount = failover
1449 no third way to do non-force, non-failover */
1450 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1451 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1455 #ifndef HAVE_STATFS_DENTRY_PARAM
1456 static int server_statfs (struct super_block *sb, struct kstatfs *buf)
1459 static int server_statfs (struct dentry *dentry, struct kstatfs *buf)
1461 struct super_block *sb = dentry->d_sb;
1463 struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt;
1466 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) {
1467 #ifdef HAVE_STATFS_DENTRY_PARAM
1468 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf);
1470 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_sb, buf);
1473 buf->f_type = sb->s_magic;
1479 buf->f_type = sb->s_magic;
1480 buf->f_bsize = sb->s_blocksize;
1486 buf->f_namelen = NAME_MAX;
1490 static int server_show_options(struct seq_file *seq, struct vfsmount *vfsmnt)
1492 struct vfsmount *mnt = s2lsi(vfsmnt->mnt_sb)->lsi_srv_mnt;
1495 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->show_options) {
1496 int rc = mnt->mnt_sb->s_op->show_options(seq, mnt);
1502 static struct super_operations server_ops =
1504 .put_super = server_put_super,
1505 .umount_begin = server_umount_begin, /* umount -f */
1506 .statfs = server_statfs,
1507 .show_options = server_show_options,
1510 #define log2(n) ffz(~(n))
1511 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1513 static int server_fill_super_common(struct super_block *sb)
1515 struct inode *root = 0;
1518 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1520 sb->s_blocksize = 4096;
1521 sb->s_blocksize_bits = log2(sb->s_blocksize);
1522 sb->s_magic = LUSTRE_SUPER_MAGIC;
1523 sb->s_maxbytes = 0; //PAGE_CACHE_MAXBYTES;
1524 sb->s_flags |= MS_RDONLY;
1525 sb->s_op = &server_ops;
1527 root = new_inode(sb);
1529 CERROR("Can't make root inode\n");
1533 /* returns -EIO for every operation */
1534 /* make_bad_inode(root); -- badness - can't umount */
1535 /* apparently we need to be a directory for the mount to finish */
1536 root->i_mode = S_IFDIR;
1538 sb->s_root = d_alloc_root(root);
1540 CERROR("Can't make root dentry\n");
1548 static int server_fill_super(struct super_block *sb)
1550 struct lustre_sb_info *lsi = s2lsi(sb);
1551 struct vfsmount *mnt;
1555 /* the One True Mount */
1556 mnt = server_kernel_mount(sb);
1559 CERROR("Unable to mount device %s: %d\n",
1560 lsi->lsi_lmd->lmd_dev, rc);
1564 lsi->lsi_srv_mnt = mnt;
1566 LASSERT(lsi->lsi_ldd);
1567 CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n",
1568 lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname,
1569 lsi->lsi_lmd->lmd_dev);
1571 if (class_name2obd(lsi->lsi_ldd->ldd_svname)) {
1572 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1573 "running. Double-mount may have compromised "
1574 "the disk journal.\n",
1575 lsi->lsi_ldd->ldd_svname);
1578 GOTO(out, rc = -EALREADY);
1581 /* start MGS before MGC */
1582 if (IS_MGS(lsi->lsi_ldd)) {
1583 rc = server_start_mgs(sb);
1588 rc = lustre_start_mgc(sb);
1592 /* Set up all obd devices for service */
1593 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1594 (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) {
1595 rc = server_start_targets(sb, mnt);
1597 CERROR("Unable to start targets: %d\n", rc);
1600 /* FIXME overmount client here,
1601 or can we just start a client log and client_fill_super on this sb?
1602 We need to make sure server_put_super gets called too - ll_put_super
1603 calls lustre_common_put_super; check there for LSI_SERVER flag,
1605 Probably should start client from new thread so we can return.
1606 Client will not finish until all servers are connected.
1607 Note - MGS-only server does NOT get a client, since there is no
1608 lustre fs associated - the MGS is for all lustre fs's */
1611 rc = server_fill_super_common(sb);
1615 LCONSOLE_WARN("Server %s on device %s has started\n",
1616 lsi->lsi_ldd->ldd_svname, lsi->lsi_lmd->lmd_dev);
1621 server_put_super(sb);
1626 /* Get the index from the obd name.
1627 rc = server type, or
1629 if endptr isn't NULL it is set to end of name */
1630 int server_name2index(char *svname, __u32 *idx, char **endptr)
1632 unsigned long index;
1634 char *dash = strchr(svname, '-');
1638 if (strncmp(dash + 1, "MDT", 3) == 0)
1639 rc = LDD_F_SV_TYPE_MDT;
1640 else if (strncmp(dash + 1, "OST", 3) == 0)
1641 rc = LDD_F_SV_TYPE_OST;
1645 index = simple_strtoul(dash + 4, endptr, 16);
1650 /*************** mount common betweeen server and client ***************/
1653 int lustre_common_put_super(struct super_block *sb)
1658 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1660 /* Drop a ref to the MGC */
1661 rc = lustre_stop_mgc(sb);
1662 if (rc && (rc != -ENOENT)) {
1664 CERROR("Can't stop MGC: %d\n", rc);
1667 /* BUSY just means that there's some other obd that
1668 needs the mgc. Let him clean it up. */
1669 CDEBUG(D_MOUNT, "MGC still in use\n");
1671 /* Drop a ref to the mounted disk */
1676 static void lmd_print(struct lustre_mount_data *lmd)
1680 PRINT_CMD(PRINT_MASK, " mount data:\n");
1681 if (lmd_is_client(lmd))
1682 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
1683 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
1684 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
1686 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
1687 for (i = 0; i < lmd->lmd_exclude_count; i++) {
1688 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
1689 lmd->lmd_exclude[i]);
1693 /* Is this server on the exclusion list */
1694 int lustre_check_exclusion(struct super_block *sb, char *svname)
1696 struct lustre_sb_info *lsi = s2lsi(sb);
1697 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1702 rc = server_name2index(svname, &index, NULL);
1703 if (rc != LDD_F_SV_TYPE_OST)
1704 /* Only exclude OSTs */
1707 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
1708 index, lmd->lmd_exclude_count, lmd->lmd_dev);
1710 for(i = 0; i < lmd->lmd_exclude_count; i++) {
1711 if (index == lmd->lmd_exclude[i]) {
1712 CWARN("Excluding %s (on exclusion list)\n", svname);
1719 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
1720 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
1722 char *s1 = ptr, *s2;
1723 __u32 index, *exclude_list;
1727 /* The shortest an ost name can be is 8 chars: -OST0000.
1728 We don't actually know the fsname at this time, so in fact
1729 a user could specify any fsname. */
1730 devmax = strlen(ptr) / 8 + 1;
1732 /* temp storage until we figure out how many we have */
1733 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
1737 /* we enter this fn pointing at the '=' */
1738 while (*s1 && *s1 != ' ' && *s1 != ',') {
1740 rc = server_name2index(s1, &index, &s2);
1742 CERROR("Can't parse server name '%s'\n", s1);
1745 if (rc == LDD_F_SV_TYPE_OST)
1746 exclude_list[lmd->lmd_exclude_count++] = index;
1748 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
1750 /* now we are pointing at ':' (next exclude)
1751 or ',' (end of excludes) */
1753 if (lmd->lmd_exclude_count >= devmax)
1756 if (rc >= 0) /* non-err */
1759 if (lmd->lmd_exclude_count) {
1760 /* permanent, freed in lustre_free_lsi */
1761 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
1762 lmd->lmd_exclude_count);
1763 if (lmd->lmd_exclude) {
1764 memcpy(lmd->lmd_exclude, exclude_list,
1765 sizeof(index) * lmd->lmd_exclude_count);
1768 lmd->lmd_exclude_count = 0;
1771 OBD_FREE(exclude_list, sizeof(index) * devmax);
1775 /* mount -v -t lustre uml1:uml2:/lustre-client /mnt/lustre */
1776 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1778 char *s1, *s2, *devname = NULL;
1779 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1785 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1786 "/sbin/mount.lustre is installed.\n");
1790 /* Options should be a string - try to detect old lmd data */
1791 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1792 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1793 "/sbin/mount.lustre. Please install "
1794 "version %s\n", LUSTRE_VERSION_STRING);
1797 lmd->lmd_magic = LMD_MAGIC;
1799 /* Set default flags here */
1804 /* Skip whitespace and extra commas */
1805 while (*s1 == ' ' || *s1 == ',')
1808 /* Client options are parsed in ll_options: eg. flock,
1811 /* Parse non-ldiskfs options here. Rather than modifying
1812 ldiskfs, we just zero these out here */
1813 if (strncmp(s1, "abort_recov", 11) == 0) {
1814 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1816 } else if (strncmp(s1, "nosvc", 5) == 0) {
1817 lmd->lmd_flags |= LMD_FLG_NOSVC;
1819 /* ost exclusion list */
1820 } else if (strncmp(s1, "exclude=", 8) == 0) {
1821 rc = lmd_make_exclusion(lmd, s1 + 7);
1827 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1828 end of the options. */
1829 else if (strncmp(s1, "device=", 7) == 0) {
1831 /* terminate options right before device. device
1832 must be the last one. */
1838 s2 = strchr(s1, ',');
1846 memmove(s1, s2, strlen(s2) + 1);
1852 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1853 "(need mount option 'device=...')\n");
1857 s1 = strrchr(devname, ':');
1859 lmd->lmd_flags = LMD_FLG_CLIENT;
1860 /* Remove leading /s from fsname */
1861 while (*++s1 == '/') ;
1862 /* Freed in lustre_free_lsi */
1863 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1864 if (!lmd->lmd_profile)
1866 sprintf(lmd->lmd_profile, "%s-client", s1);
1869 /* Freed in lustre_free_lsi */
1870 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1873 strcpy(lmd->lmd_dev, devname);
1875 /* Save mount options */
1876 s1 = options + strlen(options) - 1;
1877 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1879 if (*options != 0) {
1880 /* Freed in lustre_free_lsi */
1881 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1884 strcpy(lmd->lmd_opts, options);
1887 lmd->lmd_magic = LMD_MAGIC;
1893 CERROR("Bad mount options %s\n", options);
1899 int lustre_fill_super(struct super_block *sb, void *data, int silent)
1901 struct lustre_mount_data *lmd;
1902 struct lustre_sb_info *lsi;
1906 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1908 lsi = lustre_init_lsi(sb);
1913 /* Figure out the lmd from the mount options */
1914 if (lmd_parse((char *)data, lmd)) {
1919 if (lmd_is_client(lmd)) {
1920 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1921 if (!client_fill_super) {
1922 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1923 "client mount! Is the 'lustre' "
1924 "module loaded?\n");
1927 rc = lustre_start_mgc(sb);
1929 lustre_stop_mgc(sb);
1932 /* Connect and start */
1933 /* (should always be ll_fill_super) */
1934 rc = (*client_fill_super)(sb);
1935 /* c_f_s will call lustre_common_put_super on failure */
1939 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1940 lsi->lsi_flags |= LSI_SERVER;
1941 rc = server_fill_super(sb);
1942 /* s_f_s calls lustre_start_mgc after the mount because we need
1943 the MGS nids which are stored on disk. Plus, we may
1944 need to start the MGS first. */
1945 /* s_f_s will call server_put_super on failure */
1950 CERROR("Unable to mount %s (%d)\n",
1951 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1953 CDEBUG(D_SUPER, "mount %s complete\n", lmd->lmd_dev);
1959 /* We can't call ll_fill_super by name because it lives in a module that
1960 must be loaded after this one. */
1961 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb))
1963 client_fill_super = cfs;
1966 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
1968 kill_super_cb = cfs;
1971 /***************** FS registration ******************/
1973 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1975 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))
1976 struct super_block * lustre_get_sb(struct file_system_type *fs_type,
1977 int flags, const char *devname, void * data)
1979 /* calls back in fill super */
1980 /* we could append devname= onto options (*data) here,
1981 but 2.4 doesn't get devname. So we do it in mount_lustre.c */
1982 return get_sb_nodev(fs_type, flags, data, lustre_fill_super);
1985 int lustre_get_sb(struct file_system_type *fs_type,
1986 int flags, const char *devname, void * data,
1987 struct vfsmount *mnt)
1989 /* calls back in fill super */
1990 /* we could append devname= onto options (*data) here,
1991 but 2.4 doesn't get devname. So we do it in mount_lustre.c */
1992 return get_sb_nodev(fs_type, flags, data, lustre_fill_super, mnt);
1996 void lustre_kill_super(struct super_block *sb)
1998 struct lustre_sb_info *lsi = s2lsi(sb);
2000 if (kill_super_cb && lsi && !(lsi->lsi_flags & LSI_SERVER))
2001 (*kill_super_cb)(sb);
2003 kill_anon_super(sb);
2006 struct file_system_type lustre_fs_type = {
2007 .owner = THIS_MODULE,
2009 .get_sb = lustre_get_sb,
2010 .kill_sb = lustre_kill_super,
2011 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV,
2016 static struct super_block *lustre_read_super(struct super_block *sb,
2017 void *data, int silent)
2022 rc = lustre_fill_super(sb, data, silent);
2028 static struct file_system_type lustre_fs_type = {
2029 .owner = THIS_MODULE,
2031 .fs_flags = FS_NFSEXP_FSID,
2032 .read_super = lustre_read_super,
2036 int lustre_register_fs(void)
2038 return register_filesystem(&lustre_fs_type);
2041 int lustre_unregister_fs(void)
2043 return unregister_filesystem(&lustre_fs_type);
2046 EXPORT_SYMBOL(lustre_register_client_fill_super);
2047 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2048 EXPORT_SYMBOL(lustre_common_put_super);
2049 EXPORT_SYMBOL(lustre_process_log);
2050 EXPORT_SYMBOL(lustre_end_log);
2051 EXPORT_SYMBOL(server_get_mount);
2052 EXPORT_SYMBOL(server_put_mount);
2053 EXPORT_SYMBOL(server_register_target);
2054 EXPORT_SYMBOL(server_name2index);
2055 EXPORT_SYMBOL(server_mti_print);
2056 EXPORT_SYMBOL(do_lcfg);