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 static CFS_LIST_HEAD(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 *data = NULL;
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 sizeof(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 sizeof(KEY_INIT_RECOV_BACKUP),
720 KEY_INIT_RECOV_BACKUP,
721 sizeof(recov_bk), &recov_bk, NULL);
724 CWARN("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 */
728 GOTO(out, rc = -ENOMEM);
729 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID |
731 data->ocd_version = LUSTRE_VERSION_CODE;
732 rc = obd_connect(NULL, &mgc_conn, obd, &(obd->obd_uuid), data, NULL);
735 CERROR("connect failed %d\n", rc);
739 exp = class_conn2export(&mgc_conn);
740 obd->u.cli.cl_mgc_mgsexp = exp;
743 /* Keep the mgc info in the sb. Note that many lsi's can point
747 mutex_up(&mgc_start_lock);
750 OBD_FREE(mgcname, len);
752 OBD_FREE(niduuid, len + 2);
756 static int lustre_stop_mgc(struct super_block *sb)
758 struct lustre_sb_info *lsi = s2lsi(sb);
759 struct obd_device *obd;
760 char *niduuid = 0, *ptr = 0;
761 int i, rc = 0, len = 0;
771 mutex_down(&mgc_start_lock);
772 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
773 /* This is not fatal, every client that stops
774 will call in here. */
775 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
776 atomic_read(&obd->u.cli.cl_mgc_refcount));
777 GOTO(out, rc = -EBUSY);
780 /* The MGC has no recoverable data in any case.
781 * force shotdown set in umount_begin */
782 obd->obd_no_recov = 1;
784 if (obd->u.cli.cl_mgc_mgsexp) {
785 /* An error is not fatal, if we are unable to send the
786 disconnect mgs ping evictor cleans up the export */
787 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
789 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
792 /* Save the obdname for cleaning the nid uuids, which are
794 len = strlen(obd->obd_name) + 6;
795 OBD_ALLOC(niduuid, len);
797 strcpy(niduuid, obd->obd_name);
798 ptr = niduuid + strlen(niduuid);
801 rc = class_manual_cleanup(obd);
805 /* Clean the nid uuids */
808 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
809 sprintf(ptr, "_%x", i);
810 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
813 CERROR("del MDC UUID %s failed: rc = %d\n",
816 OBD_FREE(niduuid, len);
817 /* class_import_put will get rid of the additional connections */
820 mutex_up(&mgc_start_lock);
824 /* Since there's only one mgc per node, we have to change it's fs to get
825 access to the right disk. */
826 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
828 struct lustre_sb_info *lsi = s2lsi(sb);
832 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
834 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
835 rc = obd_set_info_async(mgc->obd_self_export,
836 sizeof(KEY_SET_FS), KEY_SET_FS,
837 sizeof(*sb), sb, NULL);
839 CERROR("can't set_fs %d\n", rc);
845 static int server_mgc_clear_fs(struct obd_device *mgc)
850 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
852 rc = obd_set_info_async(mgc->obd_self_export,
853 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
858 DECLARE_MUTEX(server_start_lock);
860 /* Stop MDS/OSS if nobody is using them */
861 static int server_stop_servers(int lddflags, int lsiflags)
863 struct obd_device *obd = NULL;
864 struct obd_type *type = NULL;
868 mutex_down(&server_start_lock);
870 /* Either an MDT or an OST or neither */
871 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
872 if ((lddflags & LDD_F_SV_TYPE_MDT) &&
873 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
874 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
875 type = class_search_type(LUSTRE_MDS_NAME);
877 /* if this was an OST, and there are no more OST's, clean up the OSS */
878 if ((lddflags & LDD_F_SV_TYPE_OST) &&
879 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
880 type = class_search_type(LUSTRE_OST_NAME);
883 if (obd && (!type || !type->typ_refcnt)) {
886 /* obd_fail doesn't mean much on a server obd */
887 err = class_manual_cleanup(obd);
892 mutex_up(&server_start_lock);
897 int server_mti_print(char *title, struct mgs_target_info *mti)
899 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
900 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
901 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
902 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
903 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
904 mti->mti_config_ver, mti->mti_flags);
908 static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
910 struct lustre_sb_info *lsi = s2lsi(sb);
911 struct lustre_disk_data *ldd = lsi->lsi_ldd;
912 lnet_process_id_t id;
916 if (!(lsi->lsi_flags & LSI_SERVER))
919 strncpy(mti->mti_fsname, ldd->ldd_fsname,
920 sizeof(mti->mti_fsname));
921 strncpy(mti->mti_svname, ldd->ldd_svname,
922 sizeof(mti->mti_svname));
924 mti->mti_nid_count = 0;
925 while (LNetGetId(i++, &id) != -ENOENT) {
926 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
928 mti->mti_nids[mti->mti_nid_count] = id.nid;
929 mti->mti_nid_count++;
930 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
931 CWARN("Only using first %d nids for %s\n",
932 mti->mti_nid_count, mti->mti_svname);
937 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
938 mti->mti_config_ver = 0;
939 mti->mti_flags = ldd->ldd_flags;
940 mti->mti_stripe_index = ldd->ldd_svindex;
941 memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid));
942 if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) {
943 CERROR("params too big for mti\n");
946 memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params));
950 /* Register an old or new target with the MGS. If needed MGS will construct
951 startup logs and assign index */
952 int server_register_target(struct super_block *sb)
954 struct lustre_sb_info *lsi = s2lsi(sb);
955 struct obd_device *mgc = lsi->lsi_mgc;
956 struct lustre_disk_data *ldd = lsi->lsi_ldd;
957 struct mgs_target_info *mti = NULL;
963 if (!(lsi->lsi_flags & LSI_SERVER))
969 rc = server_sb2mti(sb, mti);
973 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
974 mti->mti_svname, mti->mti_fsname,
975 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
978 /* Register the target */
979 /* FIXME use mgc_process_config instead */
980 rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
981 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
982 sizeof(*mti), mti, NULL);
986 /* Always update our flags */
987 ldd->ldd_flags = mti->mti_flags & ~LDD_F_REWRITE_LDD;
989 /* If this flag is set, it means the MGS wants us to change our
990 on-disk data. (So far this means just the index.) */
991 if (mti->mti_flags & LDD_F_REWRITE_LDD) {
994 CDEBUG(D_MOUNT, "Changing on-disk index from %#x to %#x "
995 "for %s\n", ldd->ldd_svindex, mti->mti_stripe_index,
997 ldd->ldd_svindex = mti->mti_stripe_index;
998 strncpy(ldd->ldd_svname, mti->mti_svname,
999 sizeof(ldd->ldd_svname));
1000 /* or ldd_make_sv_name(ldd); */
1001 ldd_write(&mgc->obd_lvfs_ctxt, ldd);
1002 err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb,
1005 CERROR("Label set error %d\n", err);
1006 label = fsfilt_get_label(mgc, lsi->lsi_srv_mnt->mnt_sb);
1008 CDEBUG(D_MOUNT, "Disk label changed to %s\n", label);
1010 /* Flush the new ldd to disk */
1011 fsfilt_sync(mgc, lsi->lsi_srv_mnt->mnt_sb);
1021 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1023 struct obd_device *obd;
1024 struct lustre_sb_info *lsi = s2lsi(sb);
1025 struct config_llog_instance cfg;
1029 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname);
1032 /* If we're an MDT, make sure the global MDS is running */
1033 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1034 /* make sure the MDS is started */
1035 mutex_down(&server_start_lock);
1036 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1038 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1039 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1041 LUSTRE_MDS_OBDNAME"_uuid",
1044 mutex_up(&server_start_lock);
1045 CERROR("failed to start MDS: %d\n", rc);
1049 mutex_up(&server_start_lock);
1053 /* If we're an OST, make sure the global OSS is running */
1054 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
1055 /* make sure OSS is started */
1056 mutex_down(&server_start_lock);
1057 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1059 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1061 LUSTRE_OSS_OBDNAME"_uuid",
1064 mutex_up(&server_start_lock);
1065 CERROR("failed to start OSS: %d\n", rc);
1069 mutex_up(&server_start_lock);
1072 /* Set the mgc fs to our server disk. This allows the MGC
1073 to read and write configs locally. */
1074 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1078 /* Register with MGS */
1079 rc = server_register_target(sb);
1080 if (rc && (lsi->lsi_ldd->ldd_flags &
1081 (LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_UPGRADE14))){
1082 CERROR("Required registration failed for %s: %d\n",
1083 lsi->lsi_ldd->ldd_svname, rc);
1085 LCONSOLE_ERROR_MSG(0x15f, "Communication error with "
1086 "the MGS. Is the MGS running?\n");
1090 if (rc == -EINVAL) {
1091 LCONSOLE_ERROR_MSG(0x160, "The MGS is refusing to allow this "
1092 "server (%s) to start. Please see messages"
1093 " on the MGS node.\n",
1094 lsi->lsi_ldd->ldd_svname);
1097 /* non-fatal error of registeration with MGS */
1099 CDEBUG(D_MOUNT, "Cannot register with MGS: %d\n", rc);
1101 /* Let the target look up the mount using the target's name
1102 (we can't pass the sb or mnt through class_process_config.) */
1103 rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
1107 /* Start targets using the llog named for the target */
1108 memset(&cfg, 0, sizeof(cfg));
1109 rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg);
1111 CERROR("failed to start server %s: %d\n",
1112 lsi->lsi_ldd->ldd_svname, rc);
1117 /* Release the mgc fs for others to use */
1118 server_mgc_clear_fs(lsi->lsi_mgc);
1121 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1123 CERROR("no server named %s was started\n",
1124 lsi->lsi_ldd->ldd_svname);
1128 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1129 (OBP(obd, iocontrol))) {
1130 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1131 obd->obd_self_export, 0, NULL, NULL);
1134 /* log has been fully processed */
1135 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1141 /***************** lustre superblock **************/
1143 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1145 struct lustre_sb_info *lsi = NULL;
1148 OBD_ALLOC(lsi, sizeof(*lsi));
1151 OBD_ALLOC(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1152 if (!lsi->lsi_lmd) {
1153 OBD_FREE(lsi, sizeof(*lsi));
1157 lsi->lsi_lmd->lmd_exclude_count = 0;
1158 s2lsi_nocast(sb) = lsi;
1159 /* we take 1 extra ref for our setup */
1160 atomic_set(&lsi->lsi_mounts, 1);
1162 /* Default umount style */
1163 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1168 static int lustre_free_lsi(struct super_block *sb)
1170 struct lustre_sb_info *lsi = s2lsi(sb);
1176 CDEBUG(D_MOUNT, "Freeing lsi\n");
1178 /* someone didn't call server_put_mount. */
1179 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
1181 if (lsi->lsi_ldd != NULL)
1182 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
1184 if (lsi->lsi_lmd != NULL) {
1185 if (lsi->lsi_lmd->lmd_dev != NULL)
1186 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1187 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1188 if (lsi->lsi_lmd->lmd_profile != NULL)
1189 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1190 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1191 if (lsi->lsi_lmd->lmd_opts != NULL)
1192 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1193 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1194 if (lsi->lsi_lmd->lmd_exclude_count)
1195 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1196 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1197 lsi->lsi_lmd->lmd_exclude_count);
1198 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1201 LASSERT(lsi->lsi_llsbi == NULL);
1202 OBD_FREE(lsi, sizeof(*lsi));
1203 s2lsi_nocast(sb) = NULL;
1208 /* The lsi has one reference for every server that is using the disk -
1209 e.g. MDT, MGS, and potentially MGC */
1210 static int lustre_put_lsi(struct super_block *sb)
1212 struct lustre_sb_info *lsi = s2lsi(sb);
1217 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
1219 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
1220 lustre_free_lsi(sb);
1226 /*************** server mount ******************/
1228 /* Kernel mount using mount options in MOUNT_DATA_FILE */
1229 static struct vfsmount *server_kernel_mount(struct super_block *sb)
1231 struct lvfs_run_ctxt mount_ctxt;
1232 struct lustre_sb_info *lsi = s2lsi(sb);
1233 struct lustre_disk_data *ldd;
1234 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1235 struct vfsmount *mnt;
1236 char *options = NULL;
1237 unsigned long page, s_flags;
1238 struct page *__page;
1242 OBD_ALLOC(ldd, sizeof(*ldd));
1244 RETURN(ERR_PTR(-ENOMEM));
1246 /* In the past, we have always used flags = 0.
1247 Note ext3/ldiskfs can't be mounted ro. */
1248 s_flags = sb->s_flags;
1250 /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */
1251 CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev);
1252 mnt = ll_kern_mount("ldiskfs", s_flags, lmd->lmd_dev, 0);
1255 CERROR("premount %s:%#lx ldiskfs failed: %d "
1256 "Is the ldiskfs module available?\n",
1257 lmd->lmd_dev, s_flags, rc );
1261 OBD_SET_CTXT_MAGIC(&mount_ctxt);
1262 mount_ctxt.pwdmnt = mnt;
1263 mount_ctxt.pwd = mnt->mnt_root;
1264 mount_ctxt.fs = get_ds();
1266 rc = ldd_parse(&mount_ctxt, ldd);
1270 CERROR("premount parse options failed: rc = %d\n", rc);
1274 /* Done with our pre-mount, now do the real mount. */
1276 /* Glom up mount options */
1277 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
1279 GOTO(out_free, rc = -ENOMEM);
1280 page = (unsigned long)cfs_page_address(__page);
1282 options = (char *)page;
1283 memset(options, 0, CFS_PAGE_SIZE);
1284 strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2);
1286 /* Add in any mount-line options */
1287 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) {
1288 int len = CFS_PAGE_SIZE - strlen(options) - 2;
1290 strcat(options, ",");
1291 strncat(options, lmd->lmd_opts, len);
1294 /* Special permanent mount flags */
1296 s_flags |= MS_NOATIME | MS_NODIRATIME;
1298 CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
1299 MT_STR(ldd), lmd->lmd_dev, options);
1300 mnt = ll_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev,
1302 OBD_PAGE_FREE(__page);
1305 CERROR("ll_kern_mount failed: rc = %d\n", rc);
1309 lsi->lsi_ldd = ldd; /* freed at lsi cleanup */
1310 CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
1314 OBD_FREE(ldd, sizeof(*ldd));
1315 lsi->lsi_ldd = NULL;
1316 RETURN(ERR_PTR(rc));
1319 static void server_wait_finished(struct vfsmount *mnt)
1321 wait_queue_head_t waitq;
1322 struct l_wait_info lwi;
1325 init_waitqueue_head(&waitq);
1327 while ((atomic_read(&mnt->mnt_count) > 1) && (retries > 0)) {
1328 LCONSOLE_WARN("Mount still busy with %d refs, waiting for "
1330 atomic_read(&mnt->mnt_count), retries);
1332 /* Wait for a bit */
1334 lwi = LWI_TIMEOUT(5 * HZ, NULL, NULL);
1335 l_wait_event(waitq, 0, &lwi);
1337 if (atomic_read(&mnt->mnt_count) > 1) {
1338 CERROR("Mount %p is still busy (%d refs), giving up.\n",
1339 mnt, atomic_read(&mnt->mnt_count));
1343 static void server_put_super(struct super_block *sb)
1345 struct lustre_sb_info *lsi = s2lsi(sb);
1346 struct obd_device *obd;
1347 struct vfsmount *mnt = lsi->lsi_srv_mnt;
1348 char *tmpname, *extraname = NULL;
1350 int lddflags = lsi->lsi_ldd->ldd_flags;
1351 int lsiflags = lsi->lsi_flags;
1355 LASSERT(lsiflags & LSI_SERVER);
1357 tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1;
1358 OBD_ALLOC(tmpname, tmpname_sz);
1359 memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
1360 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1362 /* Stop the target */
1363 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1364 (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd))) {
1365 struct lustre_profile *lprof = NULL;
1367 /* tell the mgc to drop the config log */
1368 lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL);
1370 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1371 If there are any setup/cleanup errors, save the lov
1372 name for safety cleanup later. */
1373 lprof = class_get_profile(lsi->lsi_ldd->ldd_svname);
1374 if (lprof && lprof->lp_dt) {
1375 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1376 strcpy(extraname, lprof->lp_dt);
1379 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1381 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1382 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1384 /* We can't seem to give an error return code
1385 * to .put_super, so we better make sure we clean up! */
1387 class_manual_cleanup(obd);
1389 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
1390 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1394 /* If they wanted the mgs to stop separately from the mdt, they
1395 should have put it on a different device. */
1396 if (IS_MGS(lsi->lsi_ldd)) {
1397 /* stop the mgc before the mgs so the connection gets cleaned
1399 lustre_stop_mgc(sb);
1400 /* if MDS start with --nomgs, don't stop MGS then */
1401 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1402 server_stop_mgs(sb);
1405 /* Clean the mgc and sb */
1406 rc = lustre_common_put_super(sb);
1407 /* FIXME how can I report a failure to umount? */
1409 /* Wait for the targets to really clean up - can't exit (and let the
1410 sb get destroyed) while the mount is still in use */
1411 server_wait_finished(mnt);
1413 /* drop the One True Mount */
1416 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1417 until the target is really gone so that our type refcount check
1419 server_stop_servers(lddflags, lsiflags);
1421 /* In case of startup or cleanup err, stop related obds */
1423 obd = class_name2obd(extraname);
1425 CWARN("Cleaning orphaned obd %s\n", extraname);
1427 class_manual_cleanup(obd);
1429 OBD_FREE(extraname, strlen(extraname) + 1);
1432 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1433 OBD_FREE(tmpname, tmpname_sz);
1437 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1438 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1440 struct super_block *sb = vfsmnt->mnt_sb;
1442 static void server_umount_begin(struct super_block *sb)
1445 struct lustre_sb_info *lsi = s2lsi(sb);
1448 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1449 if (!(flags & MNT_FORCE)) {
1455 CDEBUG(D_MOUNT, "umount -f\n");
1456 /* umount = failover
1458 no third way to do non-force, non-failover */
1459 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1460 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1464 #ifndef HAVE_STATFS_DENTRY_PARAM
1465 static int server_statfs (struct super_block *sb, struct kstatfs *buf)
1468 static int server_statfs (struct dentry *dentry, struct kstatfs *buf)
1470 struct super_block *sb = dentry->d_sb;
1472 struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt;
1475 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) {
1476 #ifdef HAVE_STATFS_DENTRY_PARAM
1477 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf);
1479 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_sb, buf);
1482 buf->f_type = sb->s_magic;
1488 buf->f_type = sb->s_magic;
1489 buf->f_bsize = sb->s_blocksize;
1495 buf->f_namelen = NAME_MAX;
1499 static struct super_operations server_ops =
1501 .put_super = server_put_super,
1502 .umount_begin = server_umount_begin, /* umount -f */
1503 .statfs = server_statfs,
1506 #define log2(n) ffz(~(n))
1507 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1509 static int server_fill_super_common(struct super_block *sb)
1511 struct inode *root = 0;
1514 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1516 sb->s_blocksize = 4096;
1517 sb->s_blocksize_bits = log2(sb->s_blocksize);
1518 sb->s_magic = LUSTRE_SUPER_MAGIC;
1519 sb->s_maxbytes = 0; //PAGE_CACHE_MAXBYTES;
1520 sb->s_flags |= MS_RDONLY;
1521 sb->s_op = &server_ops;
1523 root = new_inode(sb);
1525 CERROR("Can't make root inode\n");
1529 /* returns -EIO for every operation */
1530 /* make_bad_inode(root); -- badness - can't umount */
1531 /* apparently we need to be a directory for the mount to finish */
1532 root->i_mode = S_IFDIR;
1534 sb->s_root = d_alloc_root(root);
1536 CERROR("Can't make root dentry\n");
1544 static int server_fill_super(struct super_block *sb)
1546 struct lustre_sb_info *lsi = s2lsi(sb);
1547 struct vfsmount *mnt;
1551 /* the One True Mount */
1552 mnt = server_kernel_mount(sb);
1555 CERROR("Unable to mount device %s: %d\n",
1556 lsi->lsi_lmd->lmd_dev, rc);
1560 lsi->lsi_srv_mnt = mnt;
1562 LASSERT(lsi->lsi_ldd);
1563 CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n",
1564 lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname,
1565 lsi->lsi_lmd->lmd_dev);
1567 if (class_name2obd(lsi->lsi_ldd->ldd_svname)) {
1568 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1569 "running. Double-mount may have compromised"
1570 " the disk journal.\n",
1571 lsi->lsi_ldd->ldd_svname);
1574 GOTO(out, rc = -EALREADY);
1577 /* start MGS before MGC */
1578 if (IS_MGS(lsi->lsi_ldd) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1579 rc = server_start_mgs(sb);
1584 rc = lustre_start_mgc(sb);
1588 /* Set up all obd devices for service */
1589 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1590 (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) {
1591 rc = server_start_targets(sb, mnt);
1593 CERROR("Unable to start targets: %d\n", rc);
1596 /* FIXME overmount client here,
1597 or can we just start a client log and client_fill_super on this sb?
1598 We need to make sure server_put_super gets called too - ll_put_super
1599 calls lustre_common_put_super; check there for LSI_SERVER flag,
1601 Probably should start client from new thread so we can return.
1602 Client will not finish until all servers are connected.
1603 Note - MGS-only server does NOT get a client, since there is no
1604 lustre fs associated - the MGS is for all lustre fs's */
1607 rc = server_fill_super_common(sb);
1611 LCONSOLE_WARN("Server %s on device %s has started\n",
1612 lsi->lsi_ldd->ldd_svname, lsi->lsi_lmd->lmd_dev);
1617 server_put_super(sb);
1622 /* Get the index from the obd name.
1623 rc = server type, or
1625 if endptr isn't NULL it is set to end of name */
1626 int server_name2index(char *svname, __u32 *idx, char **endptr)
1628 unsigned long index;
1630 char *dash = strchr(svname, '-');
1634 if (strncmp(dash + 1, "MDT", 3) == 0)
1635 rc = LDD_F_SV_TYPE_MDT;
1636 else if (strncmp(dash + 1, "OST", 3) == 0)
1637 rc = LDD_F_SV_TYPE_OST;
1641 index = simple_strtoul(dash + 4, endptr, 16);
1646 /*************** mount common betweeen server and client ***************/
1649 int lustre_common_put_super(struct super_block *sb)
1654 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1656 /* Drop a ref to the MGC */
1657 rc = lustre_stop_mgc(sb);
1658 if (rc && (rc != -ENOENT)) {
1660 CERROR("Can't stop MGC: %d\n", rc);
1663 /* BUSY just means that there's some other obd that
1664 needs the mgc. Let him clean it up. */
1665 CDEBUG(D_MOUNT, "MGC still in use\n");
1667 /* Drop a ref to the mounted disk */
1673 static void lmd_print(struct lustre_mount_data *lmd)
1677 PRINT_CMD(PRINT_MASK, " mount data:\n");
1678 if (lmd_is_client(lmd))
1679 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
1680 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
1681 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
1683 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
1684 for (i = 0; i < lmd->lmd_exclude_count; i++) {
1685 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
1686 lmd->lmd_exclude[i]);
1691 /* Is this server on the exclusion list */
1692 int lustre_check_exclusion(struct super_block *sb, char *svname)
1694 struct lustre_sb_info *lsi = s2lsi(sb);
1695 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1700 rc = server_name2index(svname, &index, NULL);
1701 if (rc != LDD_F_SV_TYPE_OST)
1702 /* Only exclude OSTs */
1705 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
1706 index, lmd->lmd_exclude_count, lmd->lmd_dev);
1708 for(i = 0; i < lmd->lmd_exclude_count; i++) {
1709 if (index == lmd->lmd_exclude[i]) {
1710 CWARN("Excluding %s (on exclusion list)\n", svname);
1717 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
1718 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
1720 char *s1 = ptr, *s2;
1721 __u32 index, *exclude_list;
1725 /* The shortest an ost name can be is 8 chars: -OST0000.
1726 We don't actually know the fsname at this time, so in fact
1727 a user could specify any fsname. */
1728 devmax = strlen(ptr) / 8 + 1;
1730 /* temp storage until we figure out how many we have */
1731 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
1735 /* we enter this fn pointing at the '=' */
1736 while (*s1 && *s1 != ' ' && *s1 != ',') {
1738 rc = server_name2index(s1, &index, &s2);
1740 CERROR("Can't parse server name '%s'\n", s1);
1743 if (rc == LDD_F_SV_TYPE_OST)
1744 exclude_list[lmd->lmd_exclude_count++] = index;
1746 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
1748 /* now we are pointing at ':' (next exclude)
1749 or ',' (end of excludes) */
1750 if (lmd->lmd_exclude_count >= devmax)
1753 if (rc >= 0) /* non-err */
1756 if (lmd->lmd_exclude_count) {
1757 /* permanent, freed in lustre_free_lsi */
1758 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
1759 lmd->lmd_exclude_count);
1760 if (lmd->lmd_exclude) {
1761 memcpy(lmd->lmd_exclude, exclude_list,
1762 sizeof(index) * lmd->lmd_exclude_count);
1765 lmd->lmd_exclude_count = 0;
1768 OBD_FREE(exclude_list, sizeof(index) * devmax);
1772 /* mount -v -t lustre uml1:uml2:/lustre-client /mnt/lustre */
1773 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1775 char *s1, *s2, *devname = NULL;
1776 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1782 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1783 "/sbin/mount.lustre is installed.\n");
1787 /* Options should be a string - try to detect old lmd data */
1788 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1789 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1790 "/sbin/mount.lustre. Please install "
1791 "version %s\n", LUSTRE_VERSION_STRING);
1794 lmd->lmd_magic = LMD_MAGIC;
1796 /* Set default flags here */
1801 /* Skip whitespace and extra commas */
1802 while (*s1 == ' ' || *s1 == ',')
1805 /* Client options are parsed in ll_options: eg. flock,
1808 /* Parse non-ldiskfs options here. Rather than modifying
1809 ldiskfs, we just zero these out here */
1810 if (strncmp(s1, "abort_recov", 11) == 0) {
1811 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1813 } else if (strncmp(s1, "nosvc", 5) == 0) {
1814 lmd->lmd_flags |= LMD_FLG_NOSVC;
1816 } else if (strncmp(s1, "nomgs", 5) == 0) {
1817 lmd->lmd_flags |= LMD_FLG_NOMGS;
1819 /* ost exclusion list */
1820 } else if (strncmp(s1, "exclude=", 8) == 0) {
1821 rc = lmd_make_exclusion(lmd, s1 + 7);
1826 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1827 end of the options. */
1828 else if (strncmp(s1, "device=", 7) == 0) {
1830 /* terminate options right before device. device
1831 must be the last one. */
1837 s2 = strchr(s1, ',');
1845 memmove(s1, s2, strlen(s2) + 1);
1851 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1852 "(need mount option 'device=...')\n");
1856 s1 = strrchr(devname, ':');
1858 lmd->lmd_flags = LMD_FLG_CLIENT;
1859 /* Remove leading /s from fsname */
1860 while (*++s1 == '/') ;
1861 /* Freed in lustre_free_lsi */
1862 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1863 if (!lmd->lmd_profile)
1865 sprintf(lmd->lmd_profile, "%s-client", s1);
1868 /* Freed in lustre_free_lsi */
1869 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1872 strcpy(lmd->lmd_dev, devname);
1874 /* Save mount options */
1875 s1 = options + strlen(options) - 1;
1876 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1878 if (*options != 0) {
1879 /* Freed in lustre_free_lsi */
1880 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1883 strcpy(lmd->lmd_opts, options);
1886 lmd->lmd_magic = LMD_MAGIC;
1891 CERROR("Bad mount options %s\n", options);
1897 int lustre_fill_super(struct super_block *sb, void *data, int silent)
1899 struct lustre_mount_data *lmd;
1900 struct lustre_sb_info *lsi;
1904 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1906 lsi = lustre_init_lsi(sb);
1911 /* Figure out the lmd from the mount options */
1912 if (lmd_parse((char *)data, lmd)) {
1917 if (lmd_is_client(lmd)) {
1918 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1919 if (!client_fill_super) {
1920 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1921 "client mount! Is the 'lustre' "
1922 "module loaded?\n");
1925 rc = lustre_start_mgc(sb);
1927 lustre_stop_mgc(sb);
1930 /* Connect and start */
1931 /* (should always be ll_fill_super) */
1932 rc = (*client_fill_super)(sb);
1933 /* c_f_s will call lustre_common_put_super on failure */
1936 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1937 lsi->lsi_flags |= LSI_SERVER;
1938 rc = server_fill_super(sb);
1939 /* s_f_s calls lustre_start_mgc after the mount because we need
1940 the MGS nids which are stored on disk. Plus, we may
1941 need to start the MGS first. */
1942 /* s_f_s will call server_put_super on failure */
1947 CERROR("Unable to mount %s (%d)\n",
1948 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1950 CDEBUG(D_SUPER, "mount %s complete\n", lmd->lmd_dev);
1956 /* We can't call ll_fill_super by name because it lives in a module that
1957 must be loaded after this one. */
1958 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb))
1960 client_fill_super = cfs;
1963 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
1965 kill_super_cb = cfs;
1968 /***************** FS registration ******************/
1970 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))
1971 struct super_block * lustre_get_sb(struct file_system_type *fs_type,
1972 int flags, const char *devname, void * data)
1974 /* calls back in fill super */
1975 /* we could append devname= onto options (*data) here,
1976 but 2.4 doesn't get devname. So we do it in mount_lustre.c */
1977 return get_sb_nodev(fs_type, flags, data, lustre_fill_super);
1980 int lustre_get_sb(struct file_system_type *fs_type,
1981 int flags, const char *devname, void * data,
1982 struct vfsmount *mnt)
1984 /* calls back in fill super */
1985 /* we could append devname= onto options (*data) here,
1986 but 2.4 doesn't get devname. So we do it in mount_lustre.c */
1987 return get_sb_nodev(fs_type, flags, data, lustre_fill_super, mnt);
1991 void lustre_kill_super(struct super_block *sb)
1993 struct lustre_sb_info *lsi = s2lsi(sb);
1995 if (kill_super_cb && lsi &&(lsi->lsi_flags & LSI_SERVER))
1996 (*kill_super_cb)(sb);
1998 kill_anon_super(sb);
2001 struct file_system_type lustre_fs_type = {
2002 .owner = THIS_MODULE,
2004 .get_sb = lustre_get_sb,
2005 .kill_sb = lustre_kill_super,
2006 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
2007 LL_RENAME_DOES_D_MOVE,
2010 int lustre_register_fs(void)
2012 return register_filesystem(&lustre_fs_type);
2015 int lustre_unregister_fs(void)
2017 return unregister_filesystem(&lustre_fs_type);
2020 EXPORT_SYMBOL(lustre_register_client_fill_super);
2021 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2022 EXPORT_SYMBOL(lustre_common_put_super);
2023 EXPORT_SYMBOL(lustre_process_log);
2024 EXPORT_SYMBOL(lustre_end_log);
2025 EXPORT_SYMBOL(server_get_mount);
2026 EXPORT_SYMBOL(server_get_mount_2);
2027 EXPORT_SYMBOL(server_put_mount);
2028 EXPORT_SYMBOL(server_put_mount_2);
2029 EXPORT_SYMBOL(server_register_target);
2030 EXPORT_SYMBOL(server_name2index);
2031 EXPORT_SYMBOL(server_mti_print);
2032 EXPORT_SYMBOL(do_lcfg);