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(const 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(const 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(const 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(const 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));
162 * Used by mdt to get mount_info from obdname.
163 * There are no blocking when using the mount_info.
164 * Do not use server_get_mount for this purpose.
166 struct lustre_mount_info *server_get_mount_2(const char *name)
168 struct lustre_mount_info *lmi;
171 down(&lustre_mount_info_lock);
172 lmi = server_find_mount(name);
173 up(&lustre_mount_info_lock);
175 CERROR("Can't find mount for %s\n", name);
180 static void unlock_mntput(struct vfsmount *mnt)
182 if (kernel_locked()) {
191 static int lustre_put_lsi(struct super_block *sb);
193 /* to be called from obd_cleanup methods */
194 int server_put_mount(const char *name, struct vfsmount *mnt)
196 struct lustre_mount_info *lmi;
197 struct lustre_sb_info *lsi;
198 int count = atomic_read(&mnt->mnt_count) - 1;
201 /* This might be the last one, can't deref after this */
204 down(&lustre_mount_info_lock);
205 lmi = server_find_mount(name);
206 up(&lustre_mount_info_lock);
208 CERROR("Can't find mount for %s\n", name);
211 lsi = s2lsi(lmi->lmi_sb);
212 LASSERT(lmi->lmi_mnt == mnt);
214 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d, vfscount=%d\n",
215 lmi->lmi_mnt, name, atomic_read(&lsi->lsi_mounts), count);
217 if (lustre_put_lsi(lmi->lmi_sb)) {
218 CDEBUG(D_MOUNT, "Last put of mnt %p from %s, vfscount=%d\n",
219 lmi->lmi_mnt, name, count);
220 /* last mount is the One True Mount */
222 CERROR("%s: mount busy, vfscount=%d!\n", name, count);
225 /* this obd should never need the mount again */
226 server_deregister_mount(name);
231 /* Corresponding to server_get_mount_2 */
232 int server_put_mount_2(const char *name, struct vfsmount *mnt)
238 /******* mount helper utilities *********/
241 static void ldd_print(struct lustre_disk_data *ldd)
243 PRINT_CMD(PRINT_MASK, " disk data:\n");
244 PRINT_CMD(PRINT_MASK, "server: %s\n", ldd->ldd_svname);
245 PRINT_CMD(PRINT_MASK, "uuid: %s\n", (char *)ldd->ldd_uuid);
246 PRINT_CMD(PRINT_MASK, "fs: %s\n", ldd->ldd_fsname);
247 PRINT_CMD(PRINT_MASK, "index: %04x\n", ldd->ldd_svindex);
248 PRINT_CMD(PRINT_MASK, "config: %d\n", ldd->ldd_config_ver);
249 PRINT_CMD(PRINT_MASK, "flags: %#x\n", ldd->ldd_flags);
250 PRINT_CMD(PRINT_MASK, "diskfs: %s\n", MT_STR(ldd));
251 PRINT_CMD(PRINT_MASK, "options: %s\n", ldd->ldd_mount_opts);
252 PRINT_CMD(PRINT_MASK, "params: %s\n", ldd->ldd_params);
253 PRINT_CMD(PRINT_MASK, "comment: %s\n", ldd->ldd_userdata);
257 static int ldd_parse(struct lvfs_run_ctxt *mount_ctxt,
258 struct lustre_disk_data *ldd)
260 struct lvfs_run_ctxt saved;
267 push_ctxt(&saved, mount_ctxt, NULL);
269 file = filp_open(MOUNT_DATA_FILE, O_RDONLY, 0644);
272 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
276 len = i_size_read(file->f_dentry->d_inode);
277 CDEBUG(D_MOUNT, "Have %s, size %lu\n", MOUNT_DATA_FILE, len);
278 if (len != sizeof(*ldd)) {
279 CERROR("disk data size does not match: see %lu expect "LPSZ"\n",
281 GOTO(out_close, rc = -EINVAL);
284 rc = lustre_fread(file, ldd, len, &off);
286 CERROR("error reading %s: read %d of %lu\n",
287 MOUNT_DATA_FILE, rc, len);
288 GOTO(out_close, rc = -EINVAL);
292 if (ldd->ldd_magic != LDD_MAGIC) {
293 /* FIXME add swabbing support */
294 CERROR("Bad magic in %s: %x!=%x\n", MOUNT_DATA_FILE,
295 ldd->ldd_magic, LDD_MAGIC);
296 GOTO(out_close, rc = -EINVAL);
299 if (ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP) {
300 CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
302 ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP);
303 GOTO(out_close, rc = -EINVAL);
305 if (ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP) {
306 CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
308 ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP);
309 /* Do something like remount filesystem read-only */
310 GOTO(out_close, rc = -EINVAL);
316 pop_ctxt(&saved, mount_ctxt, NULL);
320 static int ldd_write(struct lvfs_run_ctxt *mount_ctxt,
321 struct lustre_disk_data *ldd)
323 struct lvfs_run_ctxt saved;
326 unsigned long len = sizeof(struct lustre_disk_data);
330 LASSERT(ldd->ldd_magic == LDD_MAGIC);
332 ldd->ldd_config_ver++;
334 push_ctxt(&saved, mount_ctxt, NULL);
336 file = filp_open(MOUNT_DATA_FILE, O_RDWR, 0644);
339 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
343 rc = lustre_fwrite(file, ldd, len, &off);
345 CERROR("error writing %s: read %d of %lu\n",
346 MOUNT_DATA_FILE, rc, len);
347 GOTO(out_close, rc = -EINVAL);
355 pop_ctxt(&saved, mount_ctxt, NULL);
360 /**************** config llog ********************/
362 /* Get a config log from the MGS and process it.
363 This func is called for both clients and servers.
364 Continue to process new statements appended to the logs
365 (whenever the config lock is revoked) until lustre_end_log
367 int lustre_process_log(struct super_block *sb, char *logname,
368 struct config_llog_instance *cfg)
370 struct lustre_cfg *lcfg;
371 struct lustre_cfg_bufs bufs;
372 struct lustre_sb_info *lsi = s2lsi(sb);
373 struct obd_device *mgc = lsi->lsi_mgc;
380 /* mgc_process_config */
381 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
382 lustre_cfg_bufs_set_string(&bufs, 1, logname);
383 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
384 lustre_cfg_bufs_set(&bufs, 3, &sb, sizeof(sb));
385 lcfg = lustre_cfg_new(LCFG_LOG_START, &bufs);
386 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
387 lustre_cfg_free(lcfg);
390 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
391 "failed from the MGS (%d). Make sure this "
392 "client and the MGS are running compatible "
393 "versions of Lustre.\n",
394 mgc->obd_name, logname, rc);
397 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
398 "failed (%d). This may be the result of "
399 "communication errors between this node and "
400 "the MGS, a bad configuration, or other "
401 "errors. See the syslog for more "
402 "information.\n", mgc->obd_name, logname,
405 /* class_obd_list(); */
409 /* Stop watching this config log for updates */
410 int lustre_end_log(struct super_block *sb, char *logname,
411 struct config_llog_instance *cfg)
413 struct lustre_cfg *lcfg;
414 struct lustre_cfg_bufs bufs;
415 struct lustre_sb_info *lsi = s2lsi(sb);
416 struct obd_device *mgc = lsi->lsi_mgc;
423 /* mgc_process_config */
424 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
425 lustre_cfg_bufs_set_string(&bufs, 1, logname);
427 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
428 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
429 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
430 lustre_cfg_free(lcfg);
434 /**************** obd start *******************/
436 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
437 char *s1, char *s2, char *s3, char *s4)
439 struct lustre_cfg_bufs bufs;
440 struct lustre_cfg * lcfg = NULL;
443 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
444 cmd, s1, s2, s3, s4);
446 lustre_cfg_bufs_reset(&bufs, cfgname);
448 lustre_cfg_bufs_set_string(&bufs, 1, s1);
450 lustre_cfg_bufs_set_string(&bufs, 2, s2);
452 lustre_cfg_bufs_set_string(&bufs, 3, s3);
454 lustre_cfg_bufs_set_string(&bufs, 4, s4);
456 lcfg = lustre_cfg_new(cmd, &bufs);
457 lcfg->lcfg_nid = nid;
458 rc = class_process_config(lcfg);
459 lustre_cfg_free(lcfg);
463 static int lustre_start_simple(char *obdname, char *type, char *uuid,
467 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
469 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
471 CERROR("%s attach error %d\n", obdname, rc);
474 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, 0, 0);
476 CERROR("%s setup error %d\n", obdname, rc);
477 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
482 /* Set up a MGS to serve startup logs */
483 static int server_start_mgs(struct super_block *sb)
485 struct lustre_sb_info *lsi = s2lsi(sb);
486 struct vfsmount *mnt = lsi->lsi_srv_mnt;
487 struct lustre_mount_info *lmi;
492 /* It is impossible to have more than 1 MGS per node, since
493 MGC wouldn't know which to connect to */
494 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
496 lsi = s2lsi(lmi->lmi_sb);
497 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
499 lsi->lsi_ldd->ldd_svname);
503 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
505 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
508 ((rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
509 LUSTRE_MGS_OBDNAME, 0, 0))))
510 server_deregister_mount(LUSTRE_MGS_OBDNAME);
513 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
514 "Is the 'mgs' module loaded?\n",
515 LUSTRE_MGS_OBDNAME, rc);
519 static int server_stop_mgs(struct super_block *sb)
521 struct obd_device *obd;
525 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
527 /* There better be only one MGS */
528 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
530 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
534 /* The MGS should always stop when we say so */
536 rc = class_manual_cleanup(obd);
540 DECLARE_MUTEX(mgc_start_lock);
542 /* Set up a mgcobd to process startup logs */
543 static int lustre_start_mgc(struct super_block *sb)
545 struct lustre_handle mgc_conn = {0, };
546 struct obd_connect_data ocd = { 0 };
547 struct lustre_sb_info *lsi = s2lsi(sb);
548 struct obd_device *obd;
549 struct obd_export *exp;
550 struct obd_uuid *uuid;
553 char *mgcname, *niduuid;
556 int rc = 0, i = 0, j, len;
559 LASSERT(lsi->lsi_lmd);
561 /* Find the first non-lo MGS nid for our MGC name */
562 if (lsi->lsi_flags & LSI_SERVER) {
563 ptr = lsi->lsi_ldd->ldd_params;
564 /* Use mgsnode= nids */
565 if ((class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0) &&
566 (class_parse_nid(ptr, &nid, &ptr) == 0)) {
568 } else if (IS_MGS(lsi->lsi_ldd)) {
569 lnet_process_id_t id;
570 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
571 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
578 } else { /* client */
579 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
580 ptr = lsi->lsi_lmd->lmd_dev;
581 if (class_parse_nid(ptr, &nid, &ptr) == 0)
585 CERROR("No valid MGS nids found.\n");
589 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
590 OBD_ALLOC(mgcname, len);
591 OBD_ALLOC(niduuid, len + 2);
592 if (!mgcname || !niduuid)
593 GOTO(out_free, rc = -ENOMEM);
594 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
596 mutex_down(&mgc_start_lock);
598 obd = class_name2obd(mgcname);
600 /* Re-using an existing MGC */
601 atomic_inc(&obd->u.cli.cl_mgc_refcount);
604 /* If we are restarting the MGS, don't try to keep the MGC's
605 old connection, or registration will fail. */
606 if ((lsi->lsi_flags & LSI_SERVER) && IS_MGS(lsi->lsi_ldd)) {
607 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
611 /* Try all connections, but only once (again).
612 We don't want to block another target from starting
613 (using its local copy of the log), but we do want to connect
614 if at all possible. */
616 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
617 rc = obd_set_info_async(obd->obd_self_export,
618 strlen(KEY_INIT_RECOV_BACKUP),
619 KEY_INIT_RECOV_BACKUP,
620 sizeof(recov_bk), &recov_bk, NULL);
624 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
626 /* Add the primary nids for the MGS */
628 sprintf(niduuid, "%s_%x", mgcname, i);
629 if (lsi->lsi_flags & LSI_SERVER) {
630 ptr = lsi->lsi_ldd->ldd_params;
631 if (IS_MGS(lsi->lsi_ldd)) {
632 /* Use local nids (including LO) */
633 lnet_process_id_t id;
634 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
635 rc = do_lcfg(mgcname, id.nid,
636 LCFG_ADD_UUID, niduuid, 0,0,0);
639 /* Use mgsnode= nids */
640 if (class_find_param(ptr, PARAM_MGSNODE, &ptr) != 0) {
641 CERROR("No MGS nids given.\n");
642 GOTO(out_free, rc = -EINVAL);
644 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
645 rc = do_lcfg(mgcname, nid,
646 LCFG_ADD_UUID, niduuid, 0,0,0);
650 } else { /* client */
651 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
652 ptr = lsi->lsi_lmd->lmd_dev;
653 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
654 rc = do_lcfg(mgcname, nid,
655 LCFG_ADD_UUID, niduuid, 0,0,0);
657 /* Stop at the first failover nid */
663 CERROR("No valid MGS nids found.\n");
664 GOTO(out_free, rc = -EINVAL);
666 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
668 /* Random uuid for MGC allows easier reconnects */
670 ll_generate_random_uuid(uuidc);
671 class_uuid_unparse(uuidc, uuid);
674 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
675 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
681 /* Add any failover MGS nids */
683 while ((*ptr == ':' ||
684 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0)) {
685 /* New failover node */
686 sprintf(niduuid, "%s_%x", mgcname, i);
688 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
690 rc = do_lcfg(mgcname, nid,
691 LCFG_ADD_UUID, niduuid, 0,0,0);
696 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
704 lsi->lsi_lmd->lmd_mgs_failnodes = i;
706 obd = class_name2obd(mgcname);
708 CERROR("Can't find mgcobd %s\n", mgcname);
709 GOTO(out_free, rc = -ENOTCONN);
712 /* Keep a refcount of servers/clients who started with "mount",
713 so we know when we can get rid of the mgc. */
714 atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
716 /* Try all connections, but only once. */
718 rc = obd_set_info_async(obd->obd_self_export,
719 strlen(KEY_INIT_RECOV_BACKUP),
720 KEY_INIT_RECOV_BACKUP,
721 sizeof(recov_bk), &recov_bk, NULL);
724 CERROR("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
725 /* We connect to the MGS at setup, and don't disconnect until cleanup */
726 rc = obd_connect(NULL, &mgc_conn, obd, &(obd->obd_uuid), &ocd);
728 CERROR("connect failed %d\n", rc);
732 exp = class_conn2export(&mgc_conn);
733 obd->u.cli.cl_mgc_mgsexp = exp;
736 /* Keep the mgc info in the sb. Note that many lsi's can point
740 mutex_up(&mgc_start_lock);
743 OBD_FREE(mgcname, len);
745 OBD_FREE(niduuid, len + 2);
749 static int lustre_stop_mgc(struct super_block *sb)
751 struct lustre_sb_info *lsi = s2lsi(sb);
752 struct obd_device *obd;
753 char *niduuid = 0, *ptr = 0;
754 int i, rc = 0, len = 0;
764 mutex_down(&mgc_start_lock);
765 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
766 /* This is not fatal, every client that stops
767 will call in here. */
768 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
769 atomic_read(&obd->u.cli.cl_mgc_refcount));
770 GOTO(out, rc = -EBUSY);
773 /* The MGC has no recoverable data in any case.
774 * force shotdown set in umount_begin */
775 obd->obd_no_recov = 1;
777 if (obd->u.cli.cl_mgc_mgsexp)
778 obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
780 /* Save the obdname for cleaning the nid uuids, which are
782 len = strlen(obd->obd_name) + 6;
783 OBD_ALLOC(niduuid, len);
785 strcpy(niduuid, obd->obd_name);
786 ptr = niduuid + strlen(niduuid);
789 rc = class_manual_cleanup(obd);
793 /* Clean the nid uuids */
796 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
797 sprintf(ptr, "_%x", i);
798 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
801 CERROR("del MDC UUID %s failed: rc = %d\n",
804 OBD_FREE(niduuid, len);
805 /* class_import_put will get rid of the additional connections */
808 mutex_up(&mgc_start_lock);
812 /* Since there's only one mgc per node, we have to change it's fs to get
813 access to the right disk. */
814 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
816 struct lustre_sb_info *lsi = s2lsi(sb);
820 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
822 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
823 rc = obd_set_info_async(mgc->obd_self_export,
824 strlen("set_fs"), "set_fs",
825 sizeof(*sb), sb, NULL);
827 CERROR("can't set_fs %d\n", rc);
833 static int server_mgc_clear_fs(struct obd_device *mgc)
838 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
840 rc = obd_set_info_async(mgc->obd_self_export,
841 strlen("clear_fs"), "clear_fs",
846 DECLARE_MUTEX(server_start_lock);
848 /* Stop MDS/OSS if nobody is using them */
849 static int server_stop_servers(int lddflags, int lsiflags)
851 struct obd_device *obd = NULL;
852 struct obd_type *type = NULL;
856 mutex_down(&server_start_lock);
858 /* Either an MDT or an OST or neither */
859 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
860 if ((lddflags & LDD_F_SV_TYPE_MDT) &&
861 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
862 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
863 type = class_search_type(LUSTRE_MDS_NAME);
865 /* if this was an OST, and there are no more OST's, clean up the OSS */
866 if ((lddflags & LDD_F_SV_TYPE_OST) &&
867 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
868 type = class_search_type(LUSTRE_OST_NAME);
871 if (obd && (!type || !type->typ_refcnt)) {
874 /* obd_fail doesn't mean much on a server obd */
875 err = class_manual_cleanup(obd);
880 mutex_up(&server_start_lock);
885 int server_mti_print(char *title, struct mgs_target_info *mti)
887 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
888 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
889 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
890 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
891 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
892 mti->mti_config_ver, mti->mti_flags);
896 static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
898 struct lustre_sb_info *lsi = s2lsi(sb);
899 struct lustre_disk_data *ldd = lsi->lsi_ldd;
900 lnet_process_id_t id;
904 if (!(lsi->lsi_flags & LSI_SERVER))
907 strncpy(mti->mti_fsname, ldd->ldd_fsname,
908 sizeof(mti->mti_fsname));
909 strncpy(mti->mti_svname, ldd->ldd_svname,
910 sizeof(mti->mti_svname));
912 mti->mti_nid_count = 0;
913 while (LNetGetId(i++, &id) != -ENOENT) {
914 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
916 mti->mti_nids[mti->mti_nid_count] = id.nid;
917 mti->mti_nid_count++;
918 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
919 CWARN("Only using first %d nids for %s\n",
920 mti->mti_nid_count, mti->mti_svname);
925 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
926 mti->mti_config_ver = 0;
927 mti->mti_flags = ldd->ldd_flags;
928 mti->mti_stripe_index = ldd->ldd_svindex;
929 memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid));
930 if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) {
931 CERROR("params too big for mti\n");
934 memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params));
938 /* Register an old or new target with the MGS. If needed MGS will construct
939 startup logs and assign index */
940 int server_register_target(struct super_block *sb)
942 struct lustre_sb_info *lsi = s2lsi(sb);
943 struct obd_device *mgc = lsi->lsi_mgc;
944 struct lustre_disk_data *ldd = lsi->lsi_ldd;
945 struct mgs_target_info *mti = NULL;
951 if (!(lsi->lsi_flags & LSI_SERVER))
957 rc = server_sb2mti(sb, mti);
961 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
962 mti->mti_svname, mti->mti_fsname,
963 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
966 /* Register the target */
967 /* FIXME use mgc_process_config instead */
968 rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
969 strlen("register_target"), "register_target",
970 sizeof(*mti), mti, NULL);
972 CERROR("registration with the MGS failed (%d)\n", rc);
976 /* Always update our flags */
977 ldd->ldd_flags = mti->mti_flags & ~LDD_F_REWRITE_LDD;
979 /* If this flag is set, it means the MGS wants us to change our
980 on-disk data. (So far this means just the index.) */
981 if (mti->mti_flags & LDD_F_REWRITE_LDD) {
984 CDEBUG(D_MOUNT, "Changing on-disk index from %#x to %#x "
985 "for %s\n", ldd->ldd_svindex, mti->mti_stripe_index,
987 ldd->ldd_svindex = mti->mti_stripe_index;
988 strncpy(ldd->ldd_svname, mti->mti_svname,
989 sizeof(ldd->ldd_svname));
990 /* or ldd_make_sv_name(ldd); */
991 ldd_write(&mgc->obd_lvfs_ctxt, ldd);
992 err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb,
995 CERROR("Label set error %d\n", err);
996 label = fsfilt_get_label(mgc, lsi->lsi_srv_mnt->mnt_sb);
998 CDEBUG(D_MOUNT, "Disk label changed to %s\n", label);
1000 /* Flush the new ldd to disk */
1001 fsfilt_sync(mgc, lsi->lsi_srv_mnt->mnt_sb);
1011 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1013 struct obd_device *obd;
1014 struct lustre_sb_info *lsi = s2lsi(sb);
1015 struct config_llog_instance cfg;
1019 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname);
1022 /* If we're an MDT, make sure the global MDS is running */
1023 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1024 /* make sure the MDS is started */
1025 mutex_down(&server_start_lock);
1026 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1028 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1029 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1031 LUSTRE_MDS_OBDNAME"_uuid",
1034 mutex_up(&server_start_lock);
1035 CERROR("failed to start MDS: %d\n", rc);
1039 mutex_up(&server_start_lock);
1043 /* If we're an OST, make sure the global OSS is running */
1044 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
1045 /* make sure OSS is started */
1046 mutex_down(&server_start_lock);
1047 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1049 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1051 LUSTRE_OSS_OBDNAME"_uuid",
1054 mutex_up(&server_start_lock);
1055 CERROR("failed to start OSS: %d\n", rc);
1059 mutex_up(&server_start_lock);
1062 /* Set the mgc fs to our server disk. This allows the MGC
1063 to read and write configs locally. */
1064 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1068 /* Register with MGS */
1069 rc = server_register_target(sb);
1070 if (rc && (lsi->lsi_ldd->ldd_flags &
1071 (LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_UPGRADE14))){
1072 CERROR("Required registration failed for %s: %d\n",
1073 lsi->lsi_ldd->ldd_svname, rc);
1075 LCONSOLE_ERROR_MSG(0x15f, "Communication error with "
1076 "the MGS. Is the MGS running?\n");
1080 if (rc == -EINVAL) {
1081 LCONSOLE_ERROR_MSG(0x160, "The MGS is refusing to allow this "
1082 "server (%s) to start. Please see messages"
1083 " on the MGS node.\n",
1084 lsi->lsi_ldd->ldd_svname);
1088 /* Let the target look up the mount using the target's name
1089 (we can't pass the sb or mnt through class_process_config.) */
1090 rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
1094 /* Start targets using the llog named for the target */
1095 memset(&cfg, 0, sizeof(cfg));
1096 rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg);
1098 CERROR("failed to start server %s: %d\n",
1099 lsi->lsi_ldd->ldd_svname, rc);
1104 /* Release the mgc fs for others to use */
1105 server_mgc_clear_fs(lsi->lsi_mgc);
1108 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1110 CERROR("no server named %s was started\n",
1111 lsi->lsi_ldd->ldd_svname);
1115 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1116 (OBP(obd, iocontrol))) {
1117 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1118 obd->obd_self_export, 0, NULL, NULL);
1121 /* log has been fully processed */
1122 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1128 /***************** lustre superblock **************/
1130 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1132 struct lustre_sb_info *lsi = NULL;
1135 OBD_ALLOC(lsi, sizeof(*lsi));
1138 OBD_ALLOC(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1139 if (!lsi->lsi_lmd) {
1140 OBD_FREE(lsi, sizeof(*lsi));
1144 lsi->lsi_lmd->lmd_exclude_count = 0;
1145 s2lsi_nocast(sb) = lsi;
1146 /* we take 1 extra ref for our setup */
1147 atomic_set(&lsi->lsi_mounts, 1);
1149 /* Default umount style */
1150 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1155 static int lustre_free_lsi(struct super_block *sb)
1157 struct lustre_sb_info *lsi = s2lsi(sb);
1163 CDEBUG(D_MOUNT, "Freeing lsi\n");
1165 /* someone didn't call server_put_mount. */
1166 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
1168 if (lsi->lsi_ldd != NULL)
1169 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
1171 if (lsi->lsi_lmd != NULL) {
1172 if (lsi->lsi_lmd->lmd_dev != NULL)
1173 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1174 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1175 if (lsi->lsi_lmd->lmd_profile != NULL)
1176 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1177 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1178 if (lsi->lsi_lmd->lmd_opts != NULL)
1179 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1180 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1181 if (lsi->lsi_lmd->lmd_exclude_count)
1182 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1183 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1184 lsi->lsi_lmd->lmd_exclude_count);
1185 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1188 LASSERT(lsi->lsi_llsbi == NULL);
1189 OBD_FREE(lsi, sizeof(*lsi));
1190 s2lsi_nocast(sb) = NULL;
1195 /* The lsi has one reference for every server that is using the disk -
1196 e.g. MDT, MGS, and potentially MGC */
1197 static int lustre_put_lsi(struct super_block *sb)
1199 struct lustre_sb_info *lsi = s2lsi(sb);
1204 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
1206 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
1207 lustre_free_lsi(sb);
1213 /*************** server mount ******************/
1215 /* Kernel mount using mount options in MOUNT_DATA_FILE */
1216 static struct vfsmount *server_kernel_mount(struct super_block *sb)
1218 struct lvfs_run_ctxt mount_ctxt;
1219 struct lustre_sb_info *lsi = s2lsi(sb);
1220 struct lustre_disk_data *ldd;
1221 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1222 struct vfsmount *mnt;
1223 char *options = NULL;
1224 unsigned long page, s_flags;
1225 struct page *__page;
1229 OBD_ALLOC(ldd, sizeof(*ldd));
1231 RETURN(ERR_PTR(-ENOMEM));
1233 /* In the past, we have always used flags = 0.
1234 Note ext3/ldiskfs can't be mounted ro. */
1235 s_flags = sb->s_flags;
1237 /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */
1238 CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev);
1239 mnt = ll_kern_mount("ldiskfs", s_flags, lmd->lmd_dev, 0);
1242 CERROR("premount %s:%#lx ldiskfs failed: %d "
1243 "Is the ldiskfs module available?\n",
1244 lmd->lmd_dev, s_flags, rc );
1248 OBD_SET_CTXT_MAGIC(&mount_ctxt);
1249 mount_ctxt.pwdmnt = mnt;
1250 mount_ctxt.pwd = mnt->mnt_root;
1251 mount_ctxt.fs = get_ds();
1253 rc = ldd_parse(&mount_ctxt, ldd);
1257 CERROR("premount parse options failed: rc = %d\n", rc);
1261 /* Done with our pre-mount, now do the real mount. */
1263 /* Glom up mount options */
1264 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
1266 GOTO(out_free, rc = -ENOMEM);
1267 page = (unsigned long)cfs_page_address(__page);
1269 options = (char *)page;
1270 memset(options, 0, CFS_PAGE_SIZE);
1271 strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2);
1273 /* Add in any mount-line options */
1274 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) {
1275 int len = CFS_PAGE_SIZE - strlen(options) - 2;
1277 strcat(options, ",");
1278 strncat(options, lmd->lmd_opts, len);
1281 /* Special permanent mount flags */
1283 s_flags |= MS_NOATIME | MS_NODIRATIME;
1285 CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
1286 MT_STR(ldd), lmd->lmd_dev, options);
1287 mnt = ll_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev,
1289 OBD_PAGE_FREE(__page);
1292 CERROR("ll_kern_mount failed: rc = %d\n", rc);
1296 lsi->lsi_ldd = ldd; /* freed at lsi cleanup */
1297 CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
1301 OBD_FREE(ldd, sizeof(*ldd));
1302 lsi->lsi_ldd = NULL;
1303 RETURN(ERR_PTR(rc));
1306 static void server_wait_finished(struct vfsmount *mnt)
1308 wait_queue_head_t waitq;
1309 struct l_wait_info lwi;
1312 init_waitqueue_head(&waitq);
1314 while ((atomic_read(&mnt->mnt_count) > 1) && (retries > 0)) {
1315 LCONSOLE_WARN("Mount still busy with %d refs, waiting for "
1317 atomic_read(&mnt->mnt_count), retries);
1319 /* Wait for a bit */
1321 lwi = LWI_TIMEOUT(5 * HZ, NULL, NULL);
1322 l_wait_event(waitq, 0, &lwi);
1324 if (atomic_read(&mnt->mnt_count) > 1) {
1325 CERROR("Mount %p is still busy (%d refs), giving up.\n",
1326 mnt, atomic_read(&mnt->mnt_count));
1330 static void server_put_super(struct super_block *sb)
1332 struct lustre_sb_info *lsi = s2lsi(sb);
1333 struct obd_device *obd;
1334 struct vfsmount *mnt = lsi->lsi_srv_mnt;
1335 char *tmpname, *extraname = NULL;
1337 int lddflags = lsi->lsi_ldd->ldd_flags;
1338 int lsiflags = lsi->lsi_flags;
1342 LASSERT(lsiflags & LSI_SERVER);
1344 tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1;
1345 OBD_ALLOC(tmpname, tmpname_sz);
1346 memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
1347 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1349 /* Stop the target */
1350 if (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd)) {
1351 struct lustre_profile *lprof = NULL;
1353 /* tell the mgc to drop the config log */
1354 lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL);
1356 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1357 If there are any setup/cleanup errors, save the lov
1358 name for safety cleanup later. */
1359 lprof = class_get_profile(lsi->lsi_ldd->ldd_svname);
1360 if (lprof && lprof->lp_dt) {
1361 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1362 strcpy(extraname, lprof->lp_dt);
1365 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1367 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1368 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1370 /* We can't seem to give an error return code
1371 * to .put_super, so we better make sure we clean up! */
1373 class_manual_cleanup(obd);
1375 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
1376 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1380 /* If they wanted the mgs to stop separately from the mdt, they
1381 should have put it on a different device. */
1382 if (IS_MGS(lsi->lsi_ldd)) {
1383 /* stop the mgc before the mgs so the connection gets cleaned
1385 lustre_stop_mgc(sb);
1386 server_stop_mgs(sb);
1389 /* Clean the mgc and sb */
1390 rc = lustre_common_put_super(sb);
1391 /* FIXME how can I report a failure to umount? */
1393 /* Wait for the targets to really clean up - can't exit (and let the
1394 sb get destroyed) while the mount is still in use */
1395 server_wait_finished(mnt);
1397 /* drop the One True Mount */
1400 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1401 until the target is really gone so that our type refcount check
1403 server_stop_servers(lddflags, lsiflags);
1405 /* In case of startup or cleanup err, stop related obds */
1407 obd = class_name2obd(extraname);
1409 CWARN("Cleaning orphaned obd %s\n", extraname);
1411 class_manual_cleanup(obd);
1413 OBD_FREE(extraname, strlen(extraname) + 1);
1416 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1417 OBD_FREE(tmpname, tmpname_sz);
1421 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1422 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1424 struct super_block *sb = vfsmnt->mnt_sb;
1426 static void server_umount_begin(struct super_block *sb)
1429 struct lustre_sb_info *lsi = s2lsi(sb);
1432 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1433 if (!(flags & MNT_FORCE)) {
1439 CDEBUG(D_MOUNT, "umount -f\n");
1440 /* umount = failover
1442 no third way to do non-force, non-failover */
1443 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1444 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1448 #ifndef HAVE_STATFS_DENTRY_PARAM
1449 static int server_statfs (struct super_block *sb, struct kstatfs *buf)
1452 static int server_statfs (struct dentry *dentry, struct kstatfs *buf)
1454 struct super_block *sb = dentry->d_sb;
1456 struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt;
1459 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) {
1460 #ifdef HAVE_STATFS_DENTRY_PARAM
1461 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf);
1463 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_sb, buf);
1466 buf->f_type = sb->s_magic;
1472 buf->f_type = sb->s_magic;
1473 buf->f_bsize = sb->s_blocksize;
1479 buf->f_namelen = NAME_MAX;
1483 static struct super_operations server_ops =
1485 .put_super = server_put_super,
1486 .umount_begin = server_umount_begin, /* umount -f */
1487 .statfs = server_statfs,
1490 #define log2(n) ffz(~(n))
1491 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1493 static int server_fill_super_common(struct super_block *sb)
1495 struct inode *root = 0;
1498 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1500 sb->s_blocksize = 4096;
1501 sb->s_blocksize_bits = log2(sb->s_blocksize);
1502 sb->s_magic = LUSTRE_SUPER_MAGIC;
1503 sb->s_maxbytes = 0; //PAGE_CACHE_MAXBYTES;
1504 sb->s_flags |= MS_RDONLY;
1505 sb->s_op = &server_ops;
1507 root = new_inode(sb);
1509 CERROR("Can't make root inode\n");
1513 /* returns -EIO for every operation */
1514 /* make_bad_inode(root); -- badness - can't umount */
1515 /* apparently we need to be a directory for the mount to finish */
1516 root->i_mode = S_IFDIR;
1518 sb->s_root = d_alloc_root(root);
1520 CERROR("Can't make root dentry\n");
1528 static int server_fill_super(struct super_block *sb)
1530 struct lustre_sb_info *lsi = s2lsi(sb);
1531 struct vfsmount *mnt;
1535 /* the One True Mount */
1536 mnt = server_kernel_mount(sb);
1539 CERROR("Unable to mount device %s: %d\n",
1540 lsi->lsi_lmd->lmd_dev, rc);
1544 lsi->lsi_srv_mnt = mnt;
1546 LASSERT(lsi->lsi_ldd);
1547 CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n",
1548 lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname,
1549 lsi->lsi_lmd->lmd_dev);
1551 if (class_name2obd(lsi->lsi_ldd->ldd_svname)) {
1552 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1553 "running. Double-mount may have compromised"
1554 " the disk journal.\n",
1555 lsi->lsi_ldd->ldd_svname);
1558 GOTO(out, rc = -EALREADY);
1561 /* start MGS before MGC */
1562 if (IS_MGS(lsi->lsi_ldd)) {
1563 rc = server_start_mgs(sb);
1568 rc = lustre_start_mgc(sb);
1572 /* Set up all obd devices for service */
1573 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1574 (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) {
1575 rc = server_start_targets(sb, mnt);
1577 CERROR("Unable to start targets: %d\n", rc);
1580 /* FIXME overmount client here,
1581 or can we just start a client log and client_fill_super on this sb?
1582 We need to make sure server_put_super gets called too - ll_put_super
1583 calls lustre_common_put_super; check there for LSI_SERVER flag,
1585 Probably should start client from new thread so we can return.
1586 Client will not finish until all servers are connected.
1587 Note - MGS-only server does NOT get a client, since there is no
1588 lustre fs associated - the MGS is for all lustre fs's */
1591 rc = server_fill_super_common(sb);
1595 LCONSOLE_WARN("Server %s on device %s has started\n",
1596 lsi->lsi_ldd->ldd_svname, lsi->lsi_lmd->lmd_dev);
1601 server_put_super(sb);
1606 /* Get the index from the obd name.
1607 rc = server type, or
1609 if endptr isn't NULL it is set to end of name */
1610 int server_name2index(char *svname, __u32 *idx, char **endptr)
1612 unsigned long index;
1614 char *dash = strchr(svname, '-');
1618 if (strncmp(dash + 1, "MDT", 3) == 0)
1619 rc = LDD_F_SV_TYPE_MDT;
1620 else if (strncmp(dash + 1, "OST", 3) == 0)
1621 rc = LDD_F_SV_TYPE_OST;
1625 index = simple_strtoul(dash + 4, endptr, 16);
1630 /*************** mount common betweeen server and client ***************/
1633 int lustre_common_put_super(struct super_block *sb)
1638 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1640 /* Drop a ref to the MGC */
1641 rc = lustre_stop_mgc(sb);
1642 if (rc && (rc != -ENOENT)) {
1644 CERROR("Can't stop MGC: %d\n", rc);
1647 /* BUSY just means that there's some other obd that
1648 needs the mgc. Let him clean it up. */
1649 CDEBUG(D_MOUNT, "MGC still in use\n");
1651 /* Drop a ref to the mounted disk */
1657 static void lmd_print(struct lustre_mount_data *lmd)
1661 PRINT_CMD(PRINT_MASK, " mount data:\n");
1662 if (lmd_is_client(lmd))
1663 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
1664 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
1665 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
1667 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
1668 for (i = 0; i < lmd->lmd_exclude_count; i++) {
1669 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
1670 lmd->lmd_exclude[i]);
1675 /* Is this server on the exclusion list */
1676 int lustre_check_exclusion(struct super_block *sb, char *svname)
1678 struct lustre_sb_info *lsi = s2lsi(sb);
1679 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1684 rc = server_name2index(svname, &index, NULL);
1685 if (rc != LDD_F_SV_TYPE_OST)
1686 /* Only exclude OSTs */
1689 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
1690 index, lmd->lmd_exclude_count, lmd->lmd_dev);
1692 for(i = 0; i < lmd->lmd_exclude_count; i++) {
1693 if (index == lmd->lmd_exclude[i]) {
1694 CWARN("Excluding %s (on exclusion list)\n", svname);
1701 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
1702 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
1704 char *s1 = ptr, *s2;
1705 __u32 index, *exclude_list;
1709 /* The shortest an ost name can be is 8 chars: -OST0000.
1710 We don't actually know the fsname at this time, so in fact
1711 a user could specify any fsname. */
1712 devmax = strlen(ptr) / 8 + 1;
1714 /* temp storage until we figure out how many we have */
1715 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
1719 /* we enter this fn pointing at the '=' */
1720 while (*s1 && *s1 != ' ' && *s1 != ',') {
1722 rc = server_name2index(s1, &index, &s2);
1724 CERROR("Can't parse server name '%s'\n", s1);
1727 if (rc == LDD_F_SV_TYPE_OST)
1728 exclude_list[lmd->lmd_exclude_count++] = index;
1730 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
1732 /* now we are pointing at ':' (next exclude)
1733 or ',' (end of excludes) */
1734 if (lmd->lmd_exclude_count >= devmax)
1737 if (rc >= 0) /* non-err */
1740 if (lmd->lmd_exclude_count) {
1741 /* permanent, freed in lustre_free_lsi */
1742 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
1743 lmd->lmd_exclude_count);
1744 if (lmd->lmd_exclude) {
1745 memcpy(lmd->lmd_exclude, exclude_list,
1746 sizeof(index) * lmd->lmd_exclude_count);
1749 lmd->lmd_exclude_count = 0;
1752 OBD_FREE(exclude_list, sizeof(index) * devmax);
1756 /* mount -v -t lustre uml1:uml2:/lustre-client /mnt/lustre */
1757 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1759 char *s1, *s2, *devname = NULL;
1760 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1766 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1767 "/sbin/mount.lustre is installed.\n");
1771 /* Options should be a string - try to detect old lmd data */
1772 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1773 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1774 "/sbin/mount.lustre. Please install "
1775 "version %s\n", LUSTRE_VERSION_STRING);
1778 lmd->lmd_magic = LMD_MAGIC;
1780 /* Set default flags here */
1785 /* Skip whitespace and extra commas */
1786 while (*s1 == ' ' || *s1 == ',')
1789 /* Client options are parsed in ll_options: eg. flock,
1792 /* Parse non-ldiskfs options here. Rather than modifying
1793 ldiskfs, we just zero these out here */
1794 if (strncmp(s1, "abort_recov", 11) == 0) {
1795 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1797 } else if (strncmp(s1, "nosvc", 5) == 0) {
1798 lmd->lmd_flags |= LMD_FLG_NOSVC;
1800 /* ost exclusion list */
1801 } else if (strncmp(s1, "exclude=", 8) == 0) {
1802 rc = lmd_make_exclusion(lmd, s1 + 7);
1807 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1808 end of the options. */
1809 else if (strncmp(s1, "device=", 7) == 0) {
1811 /* terminate options right before device. device
1812 must be the last one. */
1818 s2 = strchr(s1, ',');
1826 memmove(s1, s2, strlen(s2) + 1);
1832 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1833 "(need mount option 'device=...')\n");
1837 s1 = strrchr(devname, ':');
1839 lmd->lmd_flags = LMD_FLG_CLIENT;
1840 /* Remove leading /s from fsname */
1841 while (*++s1 == '/') ;
1842 /* Freed in lustre_free_lsi */
1843 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1844 if (!lmd->lmd_profile)
1846 sprintf(lmd->lmd_profile, "%s-client", s1);
1849 /* Freed in lustre_free_lsi */
1850 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1853 strcpy(lmd->lmd_dev, devname);
1855 /* Save mount options */
1856 s1 = options + strlen(options) - 1;
1857 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1859 if (*options != 0) {
1860 /* Freed in lustre_free_lsi */
1861 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1864 strcpy(lmd->lmd_opts, options);
1867 lmd->lmd_magic = LMD_MAGIC;
1872 CERROR("Bad mount options %s\n", options);
1878 int lustre_fill_super(struct super_block *sb, void *data, int silent)
1880 struct lustre_mount_data *lmd;
1881 struct lustre_sb_info *lsi;
1885 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1887 lsi = lustre_init_lsi(sb);
1892 /* Figure out the lmd from the mount options */
1893 if (lmd_parse((char *)data, lmd)) {
1898 if (lmd_is_client(lmd)) {
1899 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1900 if (!client_fill_super) {
1901 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1902 "client mount! Is the 'lustre' "
1903 "module loaded?\n");
1906 rc = lustre_start_mgc(sb);
1908 lustre_stop_mgc(sb);
1911 /* Connect and start */
1912 /* (should always be ll_fill_super) */
1913 rc = (*client_fill_super)(sb);
1914 /* c_f_s will call lustre_common_put_super on failure */
1917 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1918 lsi->lsi_flags |= LSI_SERVER;
1919 rc = server_fill_super(sb);
1920 /* s_f_s calls lustre_start_mgc after the mount because we need
1921 the MGS nids which are stored on disk. Plus, we may
1922 need to start the MGS first. */
1923 /* s_f_s will call server_put_super on failure */
1928 CERROR("Unable to mount %s (%d)\n",
1929 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1931 CDEBUG(D_SUPER, "mount %s complete\n", lmd->lmd_dev);
1937 /* We can't call ll_fill_super by name because it lives in a module that
1938 must be loaded after this one. */
1939 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb))
1941 client_fill_super = cfs;
1944 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
1946 kill_super_cb = cfs;
1949 /***************** FS registration ******************/
1951 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))
1952 struct super_block * lustre_get_sb(struct file_system_type *fs_type,
1953 int flags, const char *devname, void * data)
1955 /* calls back in fill super */
1956 /* we could append devname= onto options (*data) here,
1957 but 2.4 doesn't get devname. So we do it in mount_lustre.c */
1958 return get_sb_nodev(fs_type, flags, data, lustre_fill_super);
1961 int lustre_get_sb(struct file_system_type *fs_type,
1962 int flags, const char *devname, void * data,
1963 struct vfsmount *mnt)
1965 /* calls back in fill super */
1966 /* we could append devname= onto options (*data) here,
1967 but 2.4 doesn't get devname. So we do it in mount_lustre.c */
1968 return get_sb_nodev(fs_type, flags, data, lustre_fill_super, mnt);
1972 void lustre_kill_super(struct super_block *sb)
1974 struct lustre_sb_info *lsi = s2lsi(sb);
1976 if (kill_super_cb && lsi &&(lsi->lsi_flags & LSI_SERVER))
1977 (*kill_super_cb)(sb);
1979 kill_anon_super(sb);
1982 struct file_system_type lustre_fs_type = {
1983 .owner = THIS_MODULE,
1985 .get_sb = lustre_get_sb,
1986 .kill_sb = lustre_kill_super,
1987 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV,
1990 int lustre_register_fs(void)
1992 return register_filesystem(&lustre_fs_type);
1995 int lustre_unregister_fs(void)
1997 return unregister_filesystem(&lustre_fs_type);
2000 EXPORT_SYMBOL(lustre_register_client_fill_super);
2001 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2002 EXPORT_SYMBOL(lustre_common_put_super);
2003 EXPORT_SYMBOL(lustre_process_log);
2004 EXPORT_SYMBOL(lustre_end_log);
2005 EXPORT_SYMBOL(server_get_mount);
2006 EXPORT_SYMBOL(server_get_mount_2);
2007 EXPORT_SYMBOL(server_put_mount);
2008 EXPORT_SYMBOL(server_put_mount_2);
2009 EXPORT_SYMBOL(server_register_target);
2010 EXPORT_SYMBOL(server_name2index);
2011 EXPORT_SYMBOL(server_mti_print);
2012 EXPORT_SYMBOL(do_lcfg);