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;
44 /*********** mount lookup *********/
46 DECLARE_MUTEX(lustre_mount_info_lock);
47 struct list_head server_mount_info_list = LIST_HEAD_INIT(server_mount_info_list);
49 static struct lustre_mount_info *server_find_mount(char *name)
51 struct list_head *tmp;
52 struct lustre_mount_info *lmi;
55 list_for_each(tmp, &server_mount_info_list) {
56 lmi = list_entry(tmp, struct lustre_mount_info, lmi_list_chain);
57 if (strcmp(name, lmi->lmi_name) == 0)
63 /* we must register an obd for a mount before we call the setup routine.
64 *_setup will call lustre_get_mount to get the mnt struct
65 by obd_name, since we can't pass the pointer to setup. */
66 static int server_register_mount(char *name, struct super_block *sb,
69 struct lustre_mount_info *lmi;
76 OBD_ALLOC(lmi, sizeof(*lmi));
79 OBD_ALLOC(name_cp, strlen(name) + 1);
81 OBD_FREE(lmi, sizeof(*lmi));
84 strcpy(name_cp, name);
86 down(&lustre_mount_info_lock);
88 if (server_find_mount(name)) {
89 up(&lustre_mount_info_lock);
90 OBD_FREE(lmi, sizeof(*lmi));
91 OBD_FREE(name_cp, strlen(name) + 1);
92 CERROR("Already registered %s\n", name);
95 lmi->lmi_name = name_cp;
98 list_add(&lmi->lmi_list_chain, &server_mount_info_list);
100 up(&lustre_mount_info_lock);
102 CDEBUG(D_MOUNT, "reg_mnt %p from %s, vfscount=%d\n",
103 lmi->lmi_mnt, name, atomic_read(&lmi->lmi_mnt->mnt_count));
108 /* when an obd no longer needs a mount */
109 static int server_deregister_mount(char *name)
111 struct lustre_mount_info *lmi;
114 down(&lustre_mount_info_lock);
115 lmi = server_find_mount(name);
117 up(&lustre_mount_info_lock);
118 CERROR("%s not registered\n", name);
122 CDEBUG(D_MOUNT, "dereg_mnt %p from %s, vfscount=%d\n",
123 lmi->lmi_mnt, name, atomic_read(&lmi->lmi_mnt->mnt_count));
125 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
126 list_del(&lmi->lmi_list_chain);
127 OBD_FREE(lmi, sizeof(*lmi));
128 up(&lustre_mount_info_lock);
133 /* obd's look up a registered mount using their obdname. This is just
134 for initial obd setup to find the mount struct. It should not be
135 called every time you want to mntget. */
136 struct lustre_mount_info *server_get_mount(char *name)
138 struct lustre_mount_info *lmi;
139 struct lustre_sb_info *lsi;
142 down(&lustre_mount_info_lock);
143 lmi = server_find_mount(name);
144 up(&lustre_mount_info_lock);
146 CERROR("Can't find mount for %s\n", name);
149 lsi = s2lsi(lmi->lmi_sb);
150 mntget(lmi->lmi_mnt);
151 atomic_inc(&lsi->lsi_mounts);
153 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d, vfscount=%d\n",
154 lmi->lmi_mnt, name, atomic_read(&lsi->lsi_mounts),
155 atomic_read(&lmi->lmi_mnt->mnt_count));
160 static void unlock_mntput(struct vfsmount *mnt)
162 if (kernel_locked()) {
171 static int lustre_put_lsi(struct super_block *sb);
173 /* to be called from obd_cleanup methods */
174 int server_put_mount(char *name, struct vfsmount *mnt)
176 struct lustre_mount_info *lmi;
177 struct lustre_sb_info *lsi;
180 down(&lustre_mount_info_lock);
181 lmi = server_find_mount(name);
182 up(&lustre_mount_info_lock);
184 CERROR("Can't find mount for %s\n", name);
187 lsi = s2lsi(lmi->lmi_sb);
188 LASSERT(lmi->lmi_mnt == mnt);
189 unlock_mntput(lmi->lmi_mnt);
191 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d, vfscount=%d\n",
192 lmi->lmi_mnt, name, atomic_read(&lsi->lsi_mounts),
193 atomic_read(&lmi->lmi_mnt->mnt_count));
195 if (lustre_put_lsi(lmi->lmi_sb)) {
196 CDEBUG(D_MOUNT, "Last put of mnt %p from %s, vfscount=%d\n",
198 atomic_read(&lmi->lmi_mnt->mnt_count));
199 /* last mount is the One True Mount */
200 if (atomic_read(&lmi->lmi_mnt->mnt_count) > 1)
201 CERROR("%s: mount busy, vfscount=%d!\n", name,
202 atomic_read(&lmi->lmi_mnt->mnt_count));
205 /* this obd should never need the mount again */
206 server_deregister_mount(name);
212 /******* mount helper utilities *********/
214 static void ldd_print(struct lustre_disk_data *ldd)
216 PRINT_CMD(PRINT_MASK, " disk data:\n");
217 PRINT_CMD(PRINT_MASK, "server: %s\n", ldd->ldd_svname);
218 PRINT_CMD(PRINT_MASK, "uuid: %s\n", (char *)ldd->ldd_uuid);
219 PRINT_CMD(PRINT_MASK, "fs: %s\n", ldd->ldd_fsname);
220 PRINT_CMD(PRINT_MASK, "index: %04x\n", ldd->ldd_svindex);
221 PRINT_CMD(PRINT_MASK, "config: %d\n", ldd->ldd_config_ver);
222 PRINT_CMD(PRINT_MASK, "flags: %#x\n", ldd->ldd_flags);
223 PRINT_CMD(PRINT_MASK, "diskfs: %s\n", MT_STR(ldd));
224 PRINT_CMD(PRINT_MASK, "options: %s\n", ldd->ldd_mount_opts);
225 PRINT_CMD(PRINT_MASK, "params: %s\n", ldd->ldd_params);
226 PRINT_CMD(PRINT_MASK, "comment: %s\n", ldd->ldd_userdata);
229 static int ldd_parse(struct lvfs_run_ctxt *mount_ctxt,
230 struct lustre_disk_data *ldd)
232 struct lvfs_run_ctxt saved;
239 push_ctxt(&saved, mount_ctxt, NULL);
241 file = filp_open(MOUNT_DATA_FILE, O_RDONLY, 0644);
244 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
248 len = file->f_dentry->d_inode->i_size;
249 CDEBUG(D_MOUNT, "Have %s, size %lu\n", MOUNT_DATA_FILE, len);
250 if (len != sizeof(*ldd)) {
251 CERROR("disk data size does not match: see %lu expect "LPSZ"\n",
253 GOTO(out_close, rc = -EINVAL);
256 rc = lustre_fread(file, ldd, len, &off);
258 CERROR("error reading %s: read %d of %lu\n",
259 MOUNT_DATA_FILE, rc, len);
260 GOTO(out_close, rc = -EINVAL);
264 if (ldd->ldd_magic != LDD_MAGIC) {
265 /* FIXME add swabbing support */
266 CERROR("Bad magic in %s: %x!=%x\n", MOUNT_DATA_FILE,
267 ldd->ldd_magic, LDD_MAGIC);
268 GOTO(out_close, rc = -EINVAL);
271 if (ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP) {
272 CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
274 ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP);
275 GOTO(out_close, rc = -EINVAL);
277 if (ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP) {
278 CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
280 ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP);
281 /* Do something like remount filesystem read-only */
282 GOTO(out_close, rc = -EINVAL);
290 pop_ctxt(&saved, mount_ctxt, NULL);
294 static int ldd_write(struct lvfs_run_ctxt *mount_ctxt,
295 struct lustre_disk_data *ldd)
297 struct lvfs_run_ctxt saved;
300 unsigned long len = sizeof(struct lustre_disk_data);
304 LASSERT(ldd->ldd_magic == LDD_MAGIC);
306 ldd->ldd_config_ver++;
308 push_ctxt(&saved, mount_ctxt, NULL);
310 file = filp_open(MOUNT_DATA_FILE, O_RDWR, 0644);
313 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
317 rc = lustre_fwrite(file, ldd, len, &off);
319 CERROR("error writing %s: read %d of %lu\n",
320 MOUNT_DATA_FILE, rc, len);
321 GOTO(out_close, rc = -EINVAL);
330 pop_ctxt(&saved, mount_ctxt, NULL);
335 /**************** config llog ********************/
337 /* Get a config log from the MGS and process it.
338 This func is called for both clients and servers.
339 Continue to process new statements appended to the logs
340 (whenever the config lock is revoked) until lustre_end_log
342 int lustre_process_log(struct super_block *sb, char *logname,
343 struct config_llog_instance *cfg)
345 struct lustre_cfg *lcfg;
346 struct lustre_cfg_bufs bufs;
347 struct lustre_sb_info *lsi = s2lsi(sb);
348 struct obd_device *mgc = lsi->lsi_mgc;
355 /* mgc_process_config */
356 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
357 lustre_cfg_bufs_set_string(&bufs, 1, logname);
358 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
359 lustre_cfg_bufs_set(&bufs, 3, &sb, sizeof(sb));
360 lcfg = lustre_cfg_new(LCFG_LOG_START, &bufs);
361 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
362 lustre_cfg_free(lcfg);
365 LCONSOLE_ERROR("%s: The configuration from log '%s' failed "
366 "(%d). Make sure this client and "
367 "the MGS are running compatible versions of "
369 mgc->obd_name, logname, rc);
372 LCONSOLE_ERROR("%s: The configuration from log '%s' failed "
373 "(%d). This may be the result of "
374 "communication errors between this node and "
375 "the MGS, a bad configuration, or other errors."
376 " See the syslog for more information.\n",
377 mgc->obd_name, logname, rc);
379 /* class_obd_list(); */
383 /* Stop watching this config log for updates */
384 int lustre_end_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);
401 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
402 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
403 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
404 lustre_cfg_free(lcfg);
408 /**************** obd start *******************/
410 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
411 char *s1, char *s2, char *s3, char *s4)
413 struct lustre_cfg_bufs bufs;
414 struct lustre_cfg * lcfg = NULL;
417 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
418 cmd, s1, s2, s3, s4);
420 lustre_cfg_bufs_reset(&bufs, cfgname);
422 lustre_cfg_bufs_set_string(&bufs, 1, s1);
424 lustre_cfg_bufs_set_string(&bufs, 2, s2);
426 lustre_cfg_bufs_set_string(&bufs, 3, s3);
428 lustre_cfg_bufs_set_string(&bufs, 4, s4);
430 lcfg = lustre_cfg_new(cmd, &bufs);
431 lcfg->lcfg_nid = nid;
432 rc = class_process_config(lcfg);
433 lustre_cfg_free(lcfg);
437 static int lustre_start_simple(char *obdname, char *type, char *uuid,
441 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
443 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
445 CERROR("%s attach error %d\n", obdname, rc);
448 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, 0, 0);
450 CERROR("%s setup error %d\n", obdname, rc);
451 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
456 /* Set up a MGS to serve startup logs */
457 static int server_start_mgs(struct super_block *sb)
459 struct lustre_sb_info *lsi = s2lsi(sb);
460 struct vfsmount *mnt = lsi->lsi_srv_mnt;
461 struct lustre_mount_info *lmi;
466 /* It is impossible to have more than 1 MGS per node, since
467 MGC wouldn't know which to connect to */
468 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
470 lsi = s2lsi(lmi->lmi_sb);
471 LCONSOLE_ERROR("The MGS service was already started from "
472 "server %s\n", lsi->lsi_ldd->ldd_svname);
476 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
478 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
481 ((rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
482 LUSTRE_MGS_OBDNAME, 0, 0))))
483 server_deregister_mount(LUSTRE_MGS_OBDNAME);
486 LCONSOLE_ERROR("Failed to start MGS '%s' (%d). Is the 'mgs' "
487 "module loaded?\n", LUSTRE_MGS_OBDNAME, rc);
492 static int server_stop_mgs(struct super_block *sb)
494 struct obd_device *obd;
498 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
500 /* There better be only one MGS */
501 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
503 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
507 /* The MGS should always stop when we say so */
509 rc = class_manual_cleanup(obd);
513 DECLARE_MUTEX(mgc_start_lock);
515 /* Set up a mgcobd to process startup logs */
516 static int lustre_start_mgc(struct super_block *sb)
518 struct lustre_handle mgc_conn = {0, };
519 struct obd_connect_data ocd = { 0 };
520 struct lustre_sb_info *lsi = s2lsi(sb);
521 struct obd_device *obd;
522 struct obd_export *exp;
523 struct obd_uuid *uuid;
526 char *mgcname, *niduuid;
529 int rc = 0, i = 0, j, len;
532 LASSERT(lsi->lsi_lmd);
534 /* Find the first non-lo MGS nid for our MGC name */
535 if (lsi->lsi_flags & LSI_SERVER) {
536 ptr = lsi->lsi_ldd->ldd_params;
537 /* Use mgsnode= nids */
538 if ((class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0) &&
539 (class_parse_nid(ptr, &nid, &ptr) == 0)) {
541 } else if (IS_MGS(lsi->lsi_ldd)) {
542 lnet_process_id_t id;
543 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
544 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
551 } else { /* client */
552 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
553 ptr = lsi->lsi_lmd->lmd_dev;
554 if (class_parse_nid(ptr, &nid, &ptr) == 0)
558 CERROR("No valid MGS nids found.\n");
562 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
563 OBD_ALLOC(mgcname, len);
564 OBD_ALLOC(niduuid, len + 2);
565 if (!mgcname || !niduuid)
566 GOTO(out_free, rc = -ENOMEM);
567 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
569 mutex_down(&mgc_start_lock);
571 obd = class_name2obd(mgcname);
573 /* Re-using an existing MGC */
574 atomic_inc(&obd->u.cli.cl_mgc_refcount);
577 /* If we are restarting the MGS, don't try to keep the MGC's
578 old connection, or registration will fail. */
579 if ((lsi->lsi_flags & LSI_SERVER) && IS_MGS(lsi->lsi_ldd)) {
580 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
584 /* Try all connections, but only once (again).
585 We don't want to block another target from starting
586 (using its local copy of the log), but we do want to connect
587 if at all possible. */
589 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
590 rc = obd_set_info_async(obd->obd_self_export,
591 strlen(KEY_INIT_RECOV_BACKUP),
592 KEY_INIT_RECOV_BACKUP,
593 sizeof(recov_bk), &recov_bk, NULL);
597 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
599 /* Add the primary nids for the MGS */
601 sprintf(niduuid, "%s_%x", mgcname, i);
602 if (lsi->lsi_flags & LSI_SERVER) {
603 ptr = lsi->lsi_ldd->ldd_params;
604 if (IS_MGS(lsi->lsi_ldd)) {
605 /* Use local nids (including LO) */
606 lnet_process_id_t id;
607 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
608 rc = do_lcfg(mgcname, id.nid,
609 LCFG_ADD_UUID, niduuid, 0,0,0);
612 /* Use mgsnode= nids */
613 if (class_find_param(ptr, PARAM_MGSNODE, &ptr) != 0) {
614 CERROR("No MGS nids given.\n");
615 GOTO(out_free, rc = -EINVAL);
617 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
618 rc = do_lcfg(mgcname, nid,
619 LCFG_ADD_UUID, niduuid, 0,0,0);
623 } else { /* client */
624 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
625 ptr = lsi->lsi_lmd->lmd_dev;
626 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
627 rc = do_lcfg(mgcname, nid,
628 LCFG_ADD_UUID, niduuid, 0,0,0);
630 /* Stop at the first failover nid */
636 CERROR("No valid MGS nids found.\n");
637 GOTO(out_free, rc = -EINVAL);
639 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
641 /* Random uuid for MGC allows easier reconnects */
643 class_generate_random_uuid(uuidc);
644 class_uuid_unparse(uuidc, uuid);
647 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
648 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
654 /* Add any failover MGS nids */
656 while ((*ptr == ':' ||
657 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0)) {
658 /* New failover node */
659 sprintf(niduuid, "%s_%x", mgcname, i);
661 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
663 rc = do_lcfg(mgcname, nid,
664 LCFG_ADD_UUID, niduuid, 0,0,0);
669 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
677 lsi->lsi_lmd->lmd_mgs_failnodes = i;
679 obd = class_name2obd(mgcname);
681 CERROR("Can't find mgcobd %s\n", mgcname);
682 GOTO(out_free, rc = -ENOTCONN);
685 /* Keep a refcount of servers/clients who started with "mount",
686 so we know when we can get rid of the mgc. */
687 atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
689 /* Try all connections, but only once. */
691 rc = obd_set_info_async(obd->obd_self_export,
692 strlen(KEY_INIT_RECOV_BACKUP),
693 KEY_INIT_RECOV_BACKUP,
694 sizeof(recov_bk), &recov_bk, NULL);
697 CERROR("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
699 /* We connect to the MGS at setup, and don't disconnect until cleanup */
700 rc = obd_connect(&mgc_conn, obd, &(obd->obd_uuid), &ocd);
702 CERROR("connect failed %d\n", rc);
706 exp = class_conn2export(&mgc_conn);
707 obd->u.cli.cl_mgc_mgsexp = exp;
710 /* Keep the mgc info in the sb. Note that many lsi's can point
714 mutex_up(&mgc_start_lock);
717 OBD_FREE(mgcname, len);
719 OBD_FREE(niduuid, len + 2);
723 static int lustre_stop_mgc(struct super_block *sb)
725 struct lustre_sb_info *lsi = s2lsi(sb);
726 struct obd_device *obd;
727 char *niduuid, *ptr = 0;
738 mutex_down(&mgc_start_lock);
739 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
740 /* This is not fatal, every client that stops
741 will call in here. */
742 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
743 atomic_read(&obd->u.cli.cl_mgc_refcount));
744 GOTO(out, rc = -EBUSY);
747 /* MGC must always stop */
749 /* client_disconnect_export uses the no_recov flag to decide whether it
750 should disconnect or just invalidate. (The MGC has no
751 recoverable data in any case.) */
752 obd->obd_no_recov = 1;
754 if (obd->u.cli.cl_mgc_mgsexp)
755 obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
757 /* Save the obdname for cleaning the nid uuids, which are
759 len = strlen(obd->obd_name) + 6;
760 OBD_ALLOC(niduuid, len);
762 strcpy(niduuid, obd->obd_name);
763 ptr = niduuid + strlen(niduuid);
766 rc = class_manual_cleanup(obd);
770 /* Clean the nid uuids */
773 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
774 sprintf(ptr, "_%x", i);
775 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
778 CERROR("del MDC UUID %s failed: rc = %d\n",
781 OBD_FREE(niduuid, len);
782 /* class_import_put will get rid of the additional connections */
785 mutex_up(&mgc_start_lock);
789 /* Since there's only one mgc per node, we have to change it's fs to get
790 access to the right disk. */
791 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
793 struct lustre_sb_info *lsi = s2lsi(sb);
797 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
799 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
800 rc = obd_set_info_async(mgc->obd_self_export,
801 strlen("set_fs"), "set_fs",
802 sizeof(*sb), sb, NULL);
804 CERROR("can't set_fs %d\n", rc);
810 static int server_mgc_clear_fs(struct obd_device *mgc)
815 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
817 rc = obd_set_info_async(mgc->obd_self_export,
818 strlen("clear_fs"), "clear_fs",
823 DECLARE_MUTEX(server_start_lock);
825 /* Stop MDS/OSS if nobody is using them */
826 static int server_stop_servers(int lddflags, int lsiflags)
828 struct obd_device *obd = NULL;
829 struct obd_type *type = NULL;
833 mutex_down(&server_start_lock);
835 /* Either an MDT or an OST or neither */
836 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
837 if ((lddflags & LDD_F_SV_TYPE_MDT) &&
838 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
839 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
840 type = class_search_type(LUSTRE_MDS_NAME);
842 /* if this was an OST, and there are no more OST's, clean up the OSS */
843 if ((lddflags & LDD_F_SV_TYPE_OST) &&
844 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
845 type = class_search_type(LUSTRE_OST_NAME);
848 if (obd && (!type || !type->typ_refcnt)) {
851 /* obd_fail doesn't mean much on a server obd */
852 err = class_manual_cleanup(obd);
857 mutex_up(&server_start_lock);
862 int server_mti_print(char *title, struct mgs_target_info *mti)
864 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
865 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
866 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
867 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
868 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
869 mti->mti_config_ver, mti->mti_flags);
873 static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
875 struct lustre_sb_info *lsi = s2lsi(sb);
876 struct lustre_disk_data *ldd = lsi->lsi_ldd;
877 lnet_process_id_t id;
881 if (!(lsi->lsi_flags & LSI_SERVER))
884 strncpy(mti->mti_fsname, ldd->ldd_fsname,
885 sizeof(mti->mti_fsname));
886 strncpy(mti->mti_svname, ldd->ldd_svname,
887 sizeof(mti->mti_svname));
889 mti->mti_nid_count = 0;
890 while (LNetGetId(i++, &id) != -ENOENT) {
891 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
893 mti->mti_nids[mti->mti_nid_count] = id.nid;
894 mti->mti_nid_count++;
895 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
896 CWARN("Only using first %d nids for %s\n",
897 mti->mti_nid_count, mti->mti_svname);
902 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
903 mti->mti_config_ver = 0;
904 mti->mti_flags = ldd->ldd_flags;
905 mti->mti_stripe_index = ldd->ldd_svindex;
906 memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid));
907 if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) {
908 CERROR("params too big for mti\n");
911 memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params));
915 /* Register an old or new target with the MGS. If needed MGS will construct
916 startup logs and assign index */
917 int server_register_target(struct super_block *sb)
919 struct lustre_sb_info *lsi = s2lsi(sb);
920 struct obd_device *mgc = lsi->lsi_mgc;
921 struct lustre_disk_data *ldd = lsi->lsi_ldd;
922 struct mgs_target_info *mti = NULL;
928 if (!(lsi->lsi_flags & LSI_SERVER))
934 rc = server_sb2mti(sb, mti);
938 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
939 mti->mti_svname, mti->mti_fsname,
940 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
943 /* Register the target */
944 /* FIXME use mgc_process_config instead */
945 rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
946 strlen("register_target"), "register_target",
947 sizeof(*mti), mti, NULL);
949 CERROR("registration with the MGS failed (%d)\n", rc);
953 /* Always update our flags */
954 ldd->ldd_flags = mti->mti_flags & ~LDD_F_REWRITE_LDD;
956 /* If this flag is set, it means the MGS wants us to change our
957 on-disk data. (So far this means just the index.) */
958 if (mti->mti_flags & LDD_F_REWRITE_LDD) {
961 CDEBUG(D_MOUNT, "Changing on-disk index from %#x to %#x "
962 "for %s\n", ldd->ldd_svindex, mti->mti_stripe_index,
964 ldd->ldd_svindex = mti->mti_stripe_index;
965 strncpy(ldd->ldd_svname, mti->mti_svname,
966 sizeof(ldd->ldd_svname));
967 /* or ldd_make_sv_name(ldd); */
968 ldd_write(&mgc->obd_lvfs_ctxt, ldd);
970 err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb,
973 CERROR("Label set error %d\n", err);
974 label = fsfilt_get_label(mgc, lsi->lsi_srv_mnt->mnt_sb);
976 CDEBUG(D_MOUNT, "Disk label changed to %s\n", label);
978 /* Flush the new ldd to disk */
979 fsfilt_sync(mgc, lsi->lsi_srv_mnt->mnt_sb);
989 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
991 struct obd_device *obd;
992 struct lustre_sb_info *lsi = s2lsi(sb);
993 struct config_llog_instance cfg;
997 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname);
999 /* If we're an MDT, make sure the global MDS is running */
1000 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1001 /* make sure the MDS is started */
1002 mutex_down(&server_start_lock);
1003 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1005 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1006 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1008 LUSTRE_MDS_OBDNAME"_uuid",
1011 mutex_up(&server_start_lock);
1012 CERROR("failed to start MDS: %d\n", rc);
1016 mutex_up(&server_start_lock);
1019 /* If we're an OST, make sure the global OSS is running */
1020 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
1021 /* make sure OSS is started */
1022 mutex_down(&server_start_lock);
1023 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1025 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1027 LUSTRE_OSS_OBDNAME"_uuid",
1030 mutex_up(&server_start_lock);
1031 CERROR("failed to start OSS: %d\n", rc);
1035 mutex_up(&server_start_lock);
1038 /* Set the mgc fs to our server disk. This allows the MGC
1039 to read and write configs locally. */
1040 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1044 /* Register with MGS */
1045 rc = server_register_target(sb);
1046 if (rc && (lsi->lsi_ldd->ldd_flags &
1047 (LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_UPGRADE14))){
1048 CERROR("Required registration failed for %s: %d\n",
1049 lsi->lsi_ldd->ldd_svname, rc);
1051 LCONSOLE_ERROR("Communication error with the MGS. Is "
1052 "the MGS running?\n");
1056 if (rc == -EINVAL) {
1057 LCONSOLE_ERROR("The MGS is refusing to allow this server (%s) "
1058 "to start. Please see messages on the MGS node."
1059 "\n", lsi->lsi_ldd->ldd_svname);
1063 /* Let the target look up the mount using the target's name
1064 (we can't pass the sb or mnt through class_process_config.) */
1065 rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
1069 /* Start targets using the llog named for the target */
1070 memset(&cfg, 0, sizeof(cfg));
1071 rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg);
1073 CERROR("failed to start server %s: %d\n",
1074 lsi->lsi_ldd->ldd_svname, rc);
1079 /* Release the mgc fs for others to use */
1080 server_mgc_clear_fs(lsi->lsi_mgc);
1083 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1085 CERROR("no server named %s was started\n",
1086 lsi->lsi_ldd->ldd_svname);
1090 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1091 (OBP(obd, iocontrol))) {
1092 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1093 obd->obd_self_export, 0, NULL, NULL);
1100 /***************** lustre superblock **************/
1102 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1104 struct lustre_sb_info *lsi = NULL;
1107 OBD_ALLOC(lsi, sizeof(*lsi));
1110 OBD_ALLOC(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1111 if (!lsi->lsi_lmd) {
1112 OBD_FREE(lsi, sizeof(*lsi));
1116 lsi->lsi_lmd->lmd_exclude_count = 0;
1117 s2lsi_nocast(sb) = lsi;
1118 /* we take 1 extra ref for our setup */
1119 atomic_set(&lsi->lsi_mounts, 1);
1121 /* Default umount style */
1122 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1126 static int lustre_free_lsi(struct super_block *sb)
1128 struct lustre_sb_info *lsi = s2lsi(sb);
1134 CDEBUG(D_MOUNT, "Freeing lsi\n");
1136 /* someone didn't call server_put_mount. */
1137 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
1139 if (lsi->lsi_ldd != NULL)
1140 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
1142 if (lsi->lsi_lmd != NULL) {
1143 if (lsi->lsi_lmd->lmd_dev != NULL)
1144 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1145 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1146 if (lsi->lsi_lmd->lmd_profile != NULL)
1147 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1148 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1149 if (lsi->lsi_lmd->lmd_opts != NULL)
1150 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1151 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1152 if (lsi->lsi_lmd->lmd_exclude_count)
1153 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1154 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1155 lsi->lsi_lmd->lmd_exclude_count);
1156 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1159 LASSERT(lsi->lsi_llsbi == NULL);
1161 OBD_FREE(lsi, sizeof(*lsi));
1162 s2lsi_nocast(sb) = NULL;
1167 /* The lsi has one reference for every server that is using the disk -
1168 e.g. MDT, MGS, and potentially MGC */
1169 static int lustre_put_lsi(struct super_block *sb)
1171 struct lustre_sb_info *lsi = s2lsi(sb);
1176 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
1178 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
1179 lustre_free_lsi(sb);
1185 /*************** server mount ******************/
1187 /* Kernel mount using mount options in MOUNT_DATA_FILE */
1188 static struct vfsmount *server_kernel_mount(struct super_block *sb)
1190 struct lvfs_run_ctxt mount_ctxt;
1191 struct lustre_sb_info *lsi = s2lsi(sb);
1192 struct lustre_disk_data *ldd;
1193 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1194 struct vfsmount *mnt;
1195 char *options = NULL;
1196 unsigned long page, s_flags;
1200 OBD_ALLOC(ldd, sizeof(*ldd));
1202 RETURN(ERR_PTR(-ENOMEM));
1204 /* In the past, we have always used flags = 0.
1205 Note ext3/ldiskfs can't be mounted ro. */
1206 s_flags = sb->s_flags;
1208 /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */
1209 CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev);
1210 mnt = ll_kern_mount("ldiskfs", s_flags, lmd->lmd_dev, 0);
1213 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1214 CERROR("premount %s:%#lx ldiskfs failed (%d), is the ldiskfs "
1215 "module available?\n", lmd->lmd_dev, s_flags, rc);
1218 /* If ldisk fails, try ext3 */
1219 mnt = ll_kern_mount("ext3", s_flags, lmd->lmd_dev, 0);
1222 CERROR("premount ext3 failed: rc = %d\n", rc);
1228 OBD_SET_CTXT_MAGIC(&mount_ctxt);
1229 mount_ctxt.pwdmnt = mnt;
1230 mount_ctxt.pwd = mnt->mnt_root;
1231 mount_ctxt.fs = get_ds();
1233 rc = ldd_parse(&mount_ctxt, ldd);
1237 CERROR("premount parse options failed: rc = %d\n", rc);
1241 /* Done with our pre-mount, now do the real mount. */
1243 /* Glom up mount options */
1244 page = __get_free_page(GFP_KERNEL);
1246 GOTO(out_free, rc = -ENOMEM);
1248 options = (char *)page;
1249 memset(options, 0, CFS_PAGE_SIZE);
1250 strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2);
1252 /* Add in any mount-line options */
1253 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) {
1254 int len = CFS_PAGE_SIZE - strlen(options) - 2;
1256 strcat(options, ",");
1257 strncat(options, lmd->lmd_opts, len);
1260 /* Special permanent mount flags */
1262 s_flags |= MS_NOATIME | MS_NODIRATIME;
1264 CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
1265 MT_STR(ldd), lmd->lmd_dev, options);
1266 mnt = ll_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev,
1271 CERROR("ll_kern_mount failed: rc = %d\n", rc);
1275 lsi->lsi_ldd = ldd; /* freed at lsi cleanup */
1276 CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
1280 OBD_FREE(ldd, sizeof(*ldd));
1281 lsi->lsi_ldd = NULL;
1282 RETURN(ERR_PTR(rc));
1285 static void server_wait_finished(struct vfsmount *mnt)
1287 wait_queue_head_t waitq;
1288 struct l_wait_info lwi;
1291 init_waitqueue_head(&waitq);
1293 while ((atomic_read(&mnt->mnt_count) > 1) && (retries > 0)) {
1294 LCONSOLE_WARN("Mount still busy with %d refs, waiting for "
1296 atomic_read(&mnt->mnt_count), retries);
1298 /* Wait for a bit */
1300 lwi = LWI_TIMEOUT(5 * HZ, NULL, NULL);
1301 l_wait_event(waitq, 0, &lwi);
1303 if (atomic_read(&mnt->mnt_count) > 1) {
1304 CERROR("Mount %p is still busy (%d refs), giving up.\n",
1305 mnt, atomic_read(&mnt->mnt_count));
1309 static void server_put_super(struct super_block *sb)
1311 struct lustre_sb_info *lsi = s2lsi(sb);
1312 struct obd_device *obd;
1313 struct vfsmount *mnt = lsi->lsi_srv_mnt;
1314 lvfs_sbdev_type save_dev;
1315 char *tmpname, *extraname = NULL;
1317 int lddflags = lsi->lsi_ldd->ldd_flags;
1318 int lsiflags = lsi->lsi_flags;
1322 LASSERT(lsiflags & LSI_SERVER);
1324 tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1;
1325 OBD_ALLOC(tmpname, tmpname_sz);
1326 memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
1327 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1329 /* Stop the target */
1330 if (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd)) {
1331 struct lustre_profile *lprof = NULL;
1333 /* tell the mgc to drop the config log */
1334 lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL);
1336 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1337 If there are any setup/cleanup errors, save the lov
1338 name for safety cleanup later. */
1339 lprof = class_get_profile(lsi->lsi_ldd->ldd_svname);
1340 if (lprof && lprof->lp_osc) {
1341 OBD_ALLOC(extraname, strlen(lprof->lp_osc) + 1);
1342 strcpy(extraname, lprof->lp_osc);
1345 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1347 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1348 if (lsi->lsi_flags & LSI_UMOUNT_FORCE)
1350 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1352 /* We can't seem to give an error return code
1353 to .put_super, so we better make sure we clean up! */
1355 class_manual_cleanup(obd);
1357 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
1358 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1363 /* If they wanted the mgs to stop separately from the mdt, they
1364 should have put it on a different device. */
1365 if (IS_MGS(lsi->lsi_ldd)) {
1366 /* stop the mgc before the mgs so the connection gets cleaned
1368 lustre_stop_mgc(sb);
1369 server_stop_mgs(sb);
1372 save_dev = lvfs_sbdev(sb);
1374 /* Clean the mgc and sb */
1375 rc = lustre_common_put_super(sb);
1376 /* FIXME how can I report a failure to umount? */
1378 /* Wait for the targets to really clean up - can't exit (and let the
1379 sb get destroyed) while the mount is still in use */
1380 server_wait_finished(mnt);
1382 /* drop the One True Mount */
1384 #ifndef LUSTRE_PATCHLESS
1385 lvfs_clear_rdonly(save_dev);
1388 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1389 until the target is really gone so that our type refcount check
1391 server_stop_servers(lddflags, lsiflags);
1393 /* In case of startup or cleanup err, stop related obds */
1395 obd = class_name2obd(extraname);
1397 CWARN("Cleaning orphaned obd %s\n", extraname);
1399 class_manual_cleanup(obd);
1401 OBD_FREE(extraname, strlen(extraname) + 1);
1404 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1405 OBD_FREE(tmpname, tmpname_sz);
1408 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1409 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1411 struct super_block *sb = vfsmnt->mnt_sb;
1413 static void server_umount_begin(struct super_block *sb)
1416 struct lustre_sb_info *lsi = s2lsi(sb);
1419 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1420 if (!(flags & MNT_FORCE)) {
1426 CDEBUG(D_MOUNT, "umount -f\n");
1427 /* umount = failover
1429 no third way to do non-force, non-failover */
1430 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1431 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1435 #ifndef HAVE_STATFS_DENTRY_PARAM
1436 static int server_statfs (struct super_block *sb, struct kstatfs *buf)
1439 static int server_statfs (struct dentry *dentry, struct kstatfs *buf)
1441 struct super_block *sb = dentry->d_sb;
1443 struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt;
1446 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) {
1447 #ifdef HAVE_STATFS_DENTRY_PARAM
1448 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf);
1450 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_sb, buf);
1453 buf->f_type = sb->s_magic;
1459 buf->f_type = sb->s_magic;
1460 buf->f_bsize = sb->s_blocksize;
1466 buf->f_namelen = NAME_MAX;
1470 static struct super_operations server_ops =
1472 .put_super = server_put_super,
1473 .umount_begin = server_umount_begin, /* umount -f */
1474 .statfs = server_statfs,
1477 #define log2(n) ffz(~(n))
1478 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1480 static int server_fill_super_common(struct super_block *sb)
1482 struct inode *root = 0;
1485 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1487 sb->s_blocksize = 4096;
1488 sb->s_blocksize_bits = log2(sb->s_blocksize);
1489 sb->s_magic = LUSTRE_SUPER_MAGIC;
1490 sb->s_maxbytes = 0; //PAGE_CACHE_MAXBYTES;
1491 sb->s_flags |= MS_RDONLY;
1492 sb->s_op = &server_ops;
1494 root = new_inode(sb);
1496 CERROR("Can't make root inode\n");
1500 /* returns -EIO for every operation */
1501 /* make_bad_inode(root); -- badness - can't umount */
1502 /* apparently we need to be a directory for the mount to finish */
1503 root->i_mode = S_IFDIR;
1505 sb->s_root = d_alloc_root(root);
1507 CERROR("Can't make root dentry\n");
1515 static int server_fill_super(struct super_block *sb)
1517 struct lustre_sb_info *lsi = s2lsi(sb);
1518 struct vfsmount *mnt;
1522 /* the One True Mount */
1523 mnt = server_kernel_mount(sb);
1526 CERROR("Unable to mount device %s: %d\n",
1527 lsi->lsi_lmd->lmd_dev, rc);
1531 lsi->lsi_srv_mnt = mnt;
1533 LASSERT(lsi->lsi_ldd);
1534 CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n",
1535 lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname,
1536 lsi->lsi_lmd->lmd_dev);
1538 if (class_name2obd(lsi->lsi_ldd->ldd_svname)) {
1539 LCONSOLE_ERROR("The target named %s is already running. "
1540 "Double-mount may have compromised the disk "
1541 "journal.\n", lsi->lsi_ldd->ldd_svname);
1544 GOTO(out, rc = -EALREADY);
1547 /* start MGS before MGC */
1548 if (IS_MGS(lsi->lsi_ldd)) {
1549 rc = server_start_mgs(sb);
1554 rc = lustre_start_mgc(sb);
1558 /* Set up all obd devices for service */
1559 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1560 (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) {
1561 rc = server_start_targets(sb, mnt);
1563 CERROR("Unable to start targets: %d\n", rc);
1566 /* FIXME overmount client here,
1567 or can we just start a client log and client_fill_super on this sb?
1568 We need to make sure server_put_super gets called too - ll_put_super
1569 calls lustre_common_put_super; check there for LSI_SERVER flag,
1571 Probably should start client from new thread so we can return.
1572 Client will not finish until all servers are connected.
1573 Note - MGS-only server does NOT get a client, since there is no
1574 lustre fs associated - the MGS is for all lustre fs's */
1577 rc = server_fill_super_common(sb);
1581 LCONSOLE_WARN("Server %s on device %s has started\n",
1582 lsi->lsi_ldd->ldd_svname, lsi->lsi_lmd->lmd_dev);
1587 server_put_super(sb);
1592 /* Get the index from the obd name.
1593 rc = server type, or
1595 if endptr isn't NULL it is set to end of name */
1596 int server_name2index(char *svname, __u32 *idx, char **endptr)
1598 unsigned long index;
1600 char *dash = strchr(svname, '-');
1604 if (strncmp(dash + 1, "MDT", 3) == 0)
1605 rc = LDD_F_SV_TYPE_MDT;
1606 else if (strncmp(dash + 1, "OST", 3) == 0)
1607 rc = LDD_F_SV_TYPE_OST;
1611 index = simple_strtoul(dash + 4, endptr, 16);
1616 /*************** mount common betweeen server and client ***************/
1619 int lustre_common_put_super(struct super_block *sb)
1624 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1626 /* Drop a ref to the MGC */
1627 rc = lustre_stop_mgc(sb);
1628 if (rc && (rc != -ENOENT)) {
1630 CERROR("Can't stop MGC: %d\n", rc);
1633 /* BUSY just means that there's some other obd that
1634 needs the mgc. Let him clean it up. */
1635 CDEBUG(D_MOUNT, "MGC still in use\n");
1637 /* Drop a ref to the mounted disk */
1642 static void lmd_print(struct lustre_mount_data *lmd)
1646 PRINT_CMD(PRINT_MASK, " mount data:\n");
1647 if (lmd_is_client(lmd))
1648 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
1649 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
1650 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
1652 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
1653 for (i = 0; i < lmd->lmd_exclude_count; i++) {
1654 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
1655 lmd->lmd_exclude[i]);
1659 /* Is this server on the exclusion list */
1660 int lustre_check_exclusion(struct super_block *sb, char *svname)
1662 struct lustre_sb_info *lsi = s2lsi(sb);
1663 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1668 rc = server_name2index(svname, &index, NULL);
1669 if (rc != LDD_F_SV_TYPE_OST)
1670 /* Only exclude OSTs */
1673 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
1674 index, lmd->lmd_exclude_count, lmd->lmd_dev);
1676 for(i = 0; i < lmd->lmd_exclude_count; i++) {
1677 if (index == lmd->lmd_exclude[i]) {
1678 CWARN("Excluding %s (on exclusion list)\n", svname);
1685 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
1686 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
1688 char *s1 = ptr, *s2;
1689 __u32 index, *exclude_list;
1693 /* The shortest an ost name can be is 8 chars: -OST0000.
1694 We don't actually know the fsname at this time, so in fact
1695 a user could specify any fsname. */
1696 devmax = strlen(ptr) / 8 + 1;
1698 /* temp storage until we figure out how many we have */
1699 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
1703 /* we enter this fn pointing at the '=' */
1704 while (*s1 && *s1 != ' ' && *s1 != ',') {
1706 rc = server_name2index(s1, &index, &s2);
1708 CERROR("Can't parse server name '%s'\n", s1);
1711 if (rc == LDD_F_SV_TYPE_OST)
1712 exclude_list[lmd->lmd_exclude_count++] = index;
1714 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
1716 /* now we are pointing at ':' (next exclude)
1717 or ',' (end of excludes) */
1719 if (lmd->lmd_exclude_count >= devmax)
1722 if (rc >= 0) /* non-err */
1725 if (lmd->lmd_exclude_count) {
1726 /* permanent, freed in lustre_free_lsi */
1727 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
1728 lmd->lmd_exclude_count);
1729 if (lmd->lmd_exclude) {
1730 memcpy(lmd->lmd_exclude, exclude_list,
1731 sizeof(index) * lmd->lmd_exclude_count);
1734 lmd->lmd_exclude_count = 0;
1737 OBD_FREE(exclude_list, sizeof(index) * devmax);
1741 /* mount -v -t lustre uml1:uml2:/lustre-client /mnt/lustre */
1742 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1744 char *s1, *s2, *devname = NULL;
1745 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1751 LCONSOLE_ERROR("Missing mount data: check that "
1752 "/sbin/mount.lustre is installed.\n");
1756 /* Options should be a string - try to detect old lmd data */
1757 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1758 LCONSOLE_ERROR("You're using an old version of "
1759 "/sbin/mount.lustre. Please install version "
1760 "%s\n", LUSTRE_VERSION_STRING);
1763 lmd->lmd_magic = LMD_MAGIC;
1765 /* Set default flags here */
1770 /* Skip whitespace and extra commas */
1771 while (*s1 == ' ' || *s1 == ',')
1774 /* Client options are parsed in ll_options: eg. flock,
1777 /* Parse non-ldiskfs options here. Rather than modifying
1778 ldiskfs, we just zero these out here */
1779 if (strncmp(s1, "abort_recov", 11) == 0) {
1780 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1782 } else if (strncmp(s1, "nosvc", 5) == 0) {
1783 lmd->lmd_flags |= LMD_FLG_NOSVC;
1785 /* ost exclusion list */
1786 } else if (strncmp(s1, "exclude=", 8) == 0) {
1787 rc = lmd_make_exclusion(lmd, s1 + 7);
1793 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1794 end of the options. */
1795 else if (strncmp(s1, "device=", 7) == 0) {
1797 /* terminate options right before device. device
1798 must be the last one. */
1804 s2 = strchr(s1, ',');
1812 memmove(s1, s2, strlen(s2) + 1);
1818 LCONSOLE_ERROR("Can't find the device name "
1819 "(need mount option 'device=...')\n");
1823 s1 = strrchr(devname, ':');
1825 lmd->lmd_flags = LMD_FLG_CLIENT;
1826 /* Remove leading /s from fsname */
1827 while (*++s1 == '/') ;
1828 /* Freed in lustre_free_lsi */
1829 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1830 if (!lmd->lmd_profile)
1832 sprintf(lmd->lmd_profile, "%s-client", s1);
1835 /* Freed in lustre_free_lsi */
1836 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1839 strcpy(lmd->lmd_dev, devname);
1841 /* Save mount options */
1842 s1 = options + strlen(options) - 1;
1843 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1845 if (*options != 0) {
1846 /* Freed in lustre_free_lsi */
1847 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1850 strcpy(lmd->lmd_opts, options);
1853 lmd->lmd_magic = LMD_MAGIC;
1859 CERROR("Bad mount options %s\n", options);
1865 int lustre_fill_super(struct super_block *sb, void *data, int silent)
1867 struct lustre_mount_data *lmd;
1868 struct lustre_sb_info *lsi;
1872 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1874 lsi = lustre_init_lsi(sb);
1879 /* Figure out the lmd from the mount options */
1880 if (lmd_parse((char *)data, lmd)) {
1885 if (lmd_is_client(lmd)) {
1886 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1887 if (!client_fill_super) {
1888 LCONSOLE_ERROR("Nothing registered for client mount!"
1889 " Is the 'lustre' module loaded?\n");
1892 rc = lustre_start_mgc(sb);
1894 lustre_stop_mgc(sb);
1897 /* Connect and start */
1898 /* (should always be ll_fill_super) */
1899 rc = (*client_fill_super)(sb);
1900 /* c_f_s will call lustre_common_put_super on failure */
1904 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1905 lsi->lsi_flags |= LSI_SERVER;
1906 rc = server_fill_super(sb);
1907 /* s_f_s calls lustre_start_mgc after the mount because we need
1908 the MGS nids which are stored on disk. Plus, we may
1909 need to start the MGS first. */
1910 /* s_f_s will call server_put_super on failure */
1915 CERROR("Unable to mount %s (%d)\n",
1916 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1918 CDEBUG(D_SUPER, "mount %s complete\n", lmd->lmd_dev);
1924 /* We can't call ll_fill_super by name because it lives in a module that
1925 must be loaded after this one. */
1926 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb))
1928 client_fill_super = cfs;
1931 /***************** FS registration ******************/
1933 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1935 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))
1936 struct super_block * lustre_get_sb(struct file_system_type *fs_type,
1937 int flags, const char *devname, void * data)
1939 /* calls back in fill super */
1940 /* we could append devname= onto options (*data) here,
1941 but 2.4 doesn't get devname. So we do it in mount_lustre.c */
1942 return get_sb_nodev(fs_type, flags, data, lustre_fill_super);
1945 int lustre_get_sb(struct file_system_type *fs_type,
1946 int flags, const char *devname, void * data,
1947 struct vfsmount *mnt)
1949 /* calls back in fill super */
1950 /* we could append devname= onto options (*data) here,
1951 but 2.4 doesn't get devname. So we do it in mount_lustre.c */
1952 return get_sb_nodev(fs_type, flags, data, lustre_fill_super, mnt);
1956 struct file_system_type lustre_fs_type = {
1957 .owner = THIS_MODULE,
1959 .get_sb = lustre_get_sb,
1960 .kill_sb = kill_anon_super,
1961 .fs_flags = FS_BINARY_MOUNTDATA,
1966 static struct super_block *lustre_read_super(struct super_block *sb,
1967 void *data, int silent)
1972 rc = lustre_fill_super(sb, data, silent);
1978 static struct file_system_type lustre_fs_type = {
1979 .owner = THIS_MODULE,
1981 .fs_flags = FS_NFSEXP_FSID,
1982 .read_super = lustre_read_super,
1986 int lustre_register_fs(void)
1988 return register_filesystem(&lustre_fs_type);
1991 int lustre_unregister_fs(void)
1993 return unregister_filesystem(&lustre_fs_type);
1996 EXPORT_SYMBOL(lustre_register_client_fill_super);
1997 EXPORT_SYMBOL(lustre_common_put_super);
1998 EXPORT_SYMBOL(lustre_process_log);
1999 EXPORT_SYMBOL(lustre_end_log);
2000 EXPORT_SYMBOL(server_get_mount);
2001 EXPORT_SYMBOL(server_put_mount);
2002 EXPORT_SYMBOL(server_register_target);
2003 EXPORT_SYMBOL(server_name2index);
2004 EXPORT_SYMBOL(server_mti_print);
2005 EXPORT_SYMBOL(do_lcfg);