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 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 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 */
727 ocd.ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID;
728 ocd.ocd_version = LUSTRE_VERSION_CODE;
730 rc = obd_connect(NULL, &mgc_conn, obd, &(obd->obd_uuid), &ocd, NULL);
732 CERROR("connect failed %d\n", rc);
736 exp = class_conn2export(&mgc_conn);
737 obd->u.cli.cl_mgc_mgsexp = exp;
740 /* Keep the mgc info in the sb. Note that many lsi's can point
744 mutex_up(&mgc_start_lock);
747 OBD_FREE(mgcname, len);
749 OBD_FREE(niduuid, len + 2);
753 static int lustre_stop_mgc(struct super_block *sb)
755 struct lustre_sb_info *lsi = s2lsi(sb);
756 struct obd_device *obd;
757 char *niduuid = 0, *ptr = 0;
758 int i, rc = 0, len = 0;
768 mutex_down(&mgc_start_lock);
769 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
770 /* This is not fatal, every client that stops
771 will call in here. */
772 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
773 atomic_read(&obd->u.cli.cl_mgc_refcount));
774 GOTO(out, rc = -EBUSY);
777 /* The MGC has no recoverable data in any case.
778 * force shotdown set in umount_begin */
779 obd->obd_no_recov = 1;
781 if (obd->u.cli.cl_mgc_mgsexp) {
782 /* An error is not fatal, if we are unable to send the
783 disconnect mgs ping evictor cleans up the export */
784 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
786 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
789 /* Save the obdname for cleaning the nid uuids, which are
791 len = strlen(obd->obd_name) + 6;
792 OBD_ALLOC(niduuid, len);
794 strcpy(niduuid, obd->obd_name);
795 ptr = niduuid + strlen(niduuid);
798 rc = class_manual_cleanup(obd);
802 /* Clean the nid uuids */
805 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
806 sprintf(ptr, "_%x", i);
807 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
810 CERROR("del MDC UUID %s failed: rc = %d\n",
813 OBD_FREE(niduuid, len);
814 /* class_import_put will get rid of the additional connections */
817 mutex_up(&mgc_start_lock);
821 /* Since there's only one mgc per node, we have to change it's fs to get
822 access to the right disk. */
823 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
825 struct lustre_sb_info *lsi = s2lsi(sb);
829 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
831 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
832 rc = obd_set_info_async(mgc->obd_self_export,
833 strlen("set_fs"), "set_fs",
834 sizeof(*sb), sb, NULL);
836 CERROR("can't set_fs %d\n", rc);
842 static int server_mgc_clear_fs(struct obd_device *mgc)
847 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
849 rc = obd_set_info_async(mgc->obd_self_export,
850 strlen("clear_fs"), "clear_fs",
855 DECLARE_MUTEX(server_start_lock);
857 /* Stop MDS/OSS if nobody is using them */
858 static int server_stop_servers(int lddflags, int lsiflags)
860 struct obd_device *obd = NULL;
861 struct obd_type *type = NULL;
865 mutex_down(&server_start_lock);
867 /* Either an MDT or an OST or neither */
868 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
869 if ((lddflags & LDD_F_SV_TYPE_MDT) &&
870 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
871 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
872 type = class_search_type(LUSTRE_MDS_NAME);
874 /* if this was an OST, and there are no more OST's, clean up the OSS */
875 if ((lddflags & LDD_F_SV_TYPE_OST) &&
876 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
877 type = class_search_type(LUSTRE_OST_NAME);
880 if (obd && (!type || !type->typ_refcnt)) {
883 /* obd_fail doesn't mean much on a server obd */
884 err = class_manual_cleanup(obd);
889 mutex_up(&server_start_lock);
894 int server_mti_print(char *title, struct mgs_target_info *mti)
896 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
897 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
898 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
899 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
900 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
901 mti->mti_config_ver, mti->mti_flags);
905 static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
907 struct lustre_sb_info *lsi = s2lsi(sb);
908 struct lustre_disk_data *ldd = lsi->lsi_ldd;
909 lnet_process_id_t id;
913 if (!(lsi->lsi_flags & LSI_SERVER))
916 strncpy(mti->mti_fsname, ldd->ldd_fsname,
917 sizeof(mti->mti_fsname));
918 strncpy(mti->mti_svname, ldd->ldd_svname,
919 sizeof(mti->mti_svname));
921 mti->mti_nid_count = 0;
922 while (LNetGetId(i++, &id) != -ENOENT) {
923 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
925 mti->mti_nids[mti->mti_nid_count] = id.nid;
926 mti->mti_nid_count++;
927 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
928 CWARN("Only using first %d nids for %s\n",
929 mti->mti_nid_count, mti->mti_svname);
934 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
935 mti->mti_config_ver = 0;
936 mti->mti_flags = ldd->ldd_flags;
937 mti->mti_stripe_index = ldd->ldd_svindex;
938 memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid));
939 if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) {
940 CERROR("params too big for mti\n");
943 memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params));
947 /* Register an old or new target with the MGS. If needed MGS will construct
948 startup logs and assign index */
949 int server_register_target(struct super_block *sb)
951 struct lustre_sb_info *lsi = s2lsi(sb);
952 struct obd_device *mgc = lsi->lsi_mgc;
953 struct lustre_disk_data *ldd = lsi->lsi_ldd;
954 struct mgs_target_info *mti = NULL;
960 if (!(lsi->lsi_flags & LSI_SERVER))
966 rc = server_sb2mti(sb, mti);
970 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
971 mti->mti_svname, mti->mti_fsname,
972 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
975 /* Register the target */
976 /* FIXME use mgc_process_config instead */
977 rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
978 strlen("register_target"), "register_target",
979 sizeof(*mti), mti, NULL);
983 /* Always update our flags */
984 ldd->ldd_flags = mti->mti_flags & ~LDD_F_REWRITE_LDD;
986 /* If this flag is set, it means the MGS wants us to change our
987 on-disk data. (So far this means just the index.) */
988 if (mti->mti_flags & LDD_F_REWRITE_LDD) {
991 CDEBUG(D_MOUNT, "Changing on-disk index from %#x to %#x "
992 "for %s\n", ldd->ldd_svindex, mti->mti_stripe_index,
994 ldd->ldd_svindex = mti->mti_stripe_index;
995 strncpy(ldd->ldd_svname, mti->mti_svname,
996 sizeof(ldd->ldd_svname));
997 /* or ldd_make_sv_name(ldd); */
998 ldd_write(&mgc->obd_lvfs_ctxt, ldd);
999 err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb,
1002 CERROR("Label set error %d\n", err);
1003 label = fsfilt_get_label(mgc, lsi->lsi_srv_mnt->mnt_sb);
1005 CDEBUG(D_MOUNT, "Disk label changed to %s\n", label);
1007 /* Flush the new ldd to disk */
1008 fsfilt_sync(mgc, lsi->lsi_srv_mnt->mnt_sb);
1018 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1020 struct obd_device *obd;
1021 struct lustre_sb_info *lsi = s2lsi(sb);
1022 struct config_llog_instance cfg;
1026 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname);
1029 /* If we're an MDT, make sure the global MDS is running */
1030 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1031 /* make sure the MDS is started */
1032 mutex_down(&server_start_lock);
1033 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1035 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1036 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1038 LUSTRE_MDS_OBDNAME"_uuid",
1041 mutex_up(&server_start_lock);
1042 CERROR("failed to start MDS: %d\n", rc);
1046 mutex_up(&server_start_lock);
1050 /* If we're an OST, make sure the global OSS is running */
1051 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
1052 /* make sure OSS is started */
1053 mutex_down(&server_start_lock);
1054 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1056 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1058 LUSTRE_OSS_OBDNAME"_uuid",
1061 mutex_up(&server_start_lock);
1062 CERROR("failed to start OSS: %d\n", rc);
1066 mutex_up(&server_start_lock);
1069 /* Set the mgc fs to our server disk. This allows the MGC
1070 to read and write configs locally. */
1071 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1075 /* Register with MGS */
1076 rc = server_register_target(sb);
1077 if (rc && (lsi->lsi_ldd->ldd_flags &
1078 (LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_UPGRADE14))){
1079 CERROR("Required registration failed for %s: %d\n",
1080 lsi->lsi_ldd->ldd_svname, rc);
1082 LCONSOLE_ERROR_MSG(0x15f, "Communication error with "
1083 "the MGS. Is the MGS running?\n");
1087 if (rc == -EINVAL) {
1088 LCONSOLE_ERROR_MSG(0x160, "The MGS is refusing to allow this "
1089 "server (%s) to start. Please see messages"
1090 " on the MGS node.\n",
1091 lsi->lsi_ldd->ldd_svname);
1094 /* non-fatal error of registeration with MGS */
1096 CDEBUG(D_MOUNT, "Cannot register with MGS: %d\n", rc);
1098 /* Let the target look up the mount using the target's name
1099 (we can't pass the sb or mnt through class_process_config.) */
1100 rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
1104 /* Start targets using the llog named for the target */
1105 memset(&cfg, 0, sizeof(cfg));
1106 rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg);
1108 CERROR("failed to start server %s: %d\n",
1109 lsi->lsi_ldd->ldd_svname, rc);
1114 /* Release the mgc fs for others to use */
1115 server_mgc_clear_fs(lsi->lsi_mgc);
1118 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1120 CERROR("no server named %s was started\n",
1121 lsi->lsi_ldd->ldd_svname);
1125 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1126 (OBP(obd, iocontrol))) {
1127 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1128 obd->obd_self_export, 0, NULL, NULL);
1131 /* log has been fully processed */
1132 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1138 /***************** lustre superblock **************/
1140 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1142 struct lustre_sb_info *lsi = NULL;
1145 OBD_ALLOC(lsi, sizeof(*lsi));
1148 OBD_ALLOC(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1149 if (!lsi->lsi_lmd) {
1150 OBD_FREE(lsi, sizeof(*lsi));
1154 lsi->lsi_lmd->lmd_exclude_count = 0;
1155 s2lsi_nocast(sb) = lsi;
1156 /* we take 1 extra ref for our setup */
1157 atomic_set(&lsi->lsi_mounts, 1);
1159 /* Default umount style */
1160 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1165 static int lustre_free_lsi(struct super_block *sb)
1167 struct lustre_sb_info *lsi = s2lsi(sb);
1173 CDEBUG(D_MOUNT, "Freeing lsi\n");
1175 /* someone didn't call server_put_mount. */
1176 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
1178 if (lsi->lsi_ldd != NULL)
1179 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
1181 if (lsi->lsi_lmd != NULL) {
1182 if (lsi->lsi_lmd->lmd_dev != NULL)
1183 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1184 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1185 if (lsi->lsi_lmd->lmd_profile != NULL)
1186 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1187 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1188 if (lsi->lsi_lmd->lmd_opts != NULL)
1189 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1190 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1191 if (lsi->lsi_lmd->lmd_exclude_count)
1192 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1193 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1194 lsi->lsi_lmd->lmd_exclude_count);
1195 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1198 LASSERT(lsi->lsi_llsbi == NULL);
1199 OBD_FREE(lsi, sizeof(*lsi));
1200 s2lsi_nocast(sb) = NULL;
1205 /* The lsi has one reference for every server that is using the disk -
1206 e.g. MDT, MGS, and potentially MGC */
1207 static int lustre_put_lsi(struct super_block *sb)
1209 struct lustre_sb_info *lsi = s2lsi(sb);
1214 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
1216 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
1217 lustre_free_lsi(sb);
1223 /*************** server mount ******************/
1225 /* Kernel mount using mount options in MOUNT_DATA_FILE */
1226 static struct vfsmount *server_kernel_mount(struct super_block *sb)
1228 struct lvfs_run_ctxt mount_ctxt;
1229 struct lustre_sb_info *lsi = s2lsi(sb);
1230 struct lustre_disk_data *ldd;
1231 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1232 struct vfsmount *mnt;
1233 char *options = NULL;
1234 unsigned long page, s_flags;
1235 struct page *__page;
1239 OBD_ALLOC(ldd, sizeof(*ldd));
1241 RETURN(ERR_PTR(-ENOMEM));
1243 /* In the past, we have always used flags = 0.
1244 Note ext3/ldiskfs can't be mounted ro. */
1245 s_flags = sb->s_flags;
1247 /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */
1248 CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev);
1249 mnt = ll_kern_mount("ldiskfs", s_flags, lmd->lmd_dev, 0);
1252 CERROR("premount %s:%#lx ldiskfs failed: %d "
1253 "Is the ldiskfs module available?\n",
1254 lmd->lmd_dev, s_flags, rc );
1258 OBD_SET_CTXT_MAGIC(&mount_ctxt);
1259 mount_ctxt.pwdmnt = mnt;
1260 mount_ctxt.pwd = mnt->mnt_root;
1261 mount_ctxt.fs = get_ds();
1263 rc = ldd_parse(&mount_ctxt, ldd);
1267 CERROR("premount parse options failed: rc = %d\n", rc);
1271 /* Done with our pre-mount, now do the real mount. */
1273 /* Glom up mount options */
1274 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
1276 GOTO(out_free, rc = -ENOMEM);
1277 page = (unsigned long)cfs_page_address(__page);
1279 options = (char *)page;
1280 memset(options, 0, CFS_PAGE_SIZE);
1281 strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2);
1283 /* Add in any mount-line options */
1284 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) {
1285 int len = CFS_PAGE_SIZE - strlen(options) - 2;
1287 strcat(options, ",");
1288 strncat(options, lmd->lmd_opts, len);
1291 /* Special permanent mount flags */
1293 s_flags |= MS_NOATIME | MS_NODIRATIME;
1295 CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
1296 MT_STR(ldd), lmd->lmd_dev, options);
1297 mnt = ll_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev,
1299 OBD_PAGE_FREE(__page);
1302 CERROR("ll_kern_mount failed: rc = %d\n", rc);
1306 lsi->lsi_ldd = ldd; /* freed at lsi cleanup */
1307 CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
1311 OBD_FREE(ldd, sizeof(*ldd));
1312 lsi->lsi_ldd = NULL;
1313 RETURN(ERR_PTR(rc));
1316 static void server_wait_finished(struct vfsmount *mnt)
1318 wait_queue_head_t waitq;
1319 struct l_wait_info lwi;
1322 init_waitqueue_head(&waitq);
1324 while ((atomic_read(&mnt->mnt_count) > 1) && (retries > 0)) {
1325 LCONSOLE_WARN("Mount still busy with %d refs, waiting for "
1327 atomic_read(&mnt->mnt_count), retries);
1329 /* Wait for a bit */
1331 lwi = LWI_TIMEOUT(5 * HZ, NULL, NULL);
1332 l_wait_event(waitq, 0, &lwi);
1334 if (atomic_read(&mnt->mnt_count) > 1) {
1335 CERROR("Mount %p is still busy (%d refs), giving up.\n",
1336 mnt, atomic_read(&mnt->mnt_count));
1340 static void server_put_super(struct super_block *sb)
1342 struct lustre_sb_info *lsi = s2lsi(sb);
1343 struct obd_device *obd;
1344 struct vfsmount *mnt = lsi->lsi_srv_mnt;
1345 char *tmpname, *extraname = NULL;
1347 int lddflags = lsi->lsi_ldd->ldd_flags;
1348 int lsiflags = lsi->lsi_flags;
1352 LASSERT(lsiflags & LSI_SERVER);
1354 tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1;
1355 OBD_ALLOC(tmpname, tmpname_sz);
1356 memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
1357 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1359 /* Stop the target */
1360 if (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd)) {
1361 struct lustre_profile *lprof = NULL;
1363 /* tell the mgc to drop the config log */
1364 lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL);
1366 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1367 If there are any setup/cleanup errors, save the lov
1368 name for safety cleanup later. */
1369 lprof = class_get_profile(lsi->lsi_ldd->ldd_svname);
1370 if (lprof && lprof->lp_dt) {
1371 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1372 strcpy(extraname, lprof->lp_dt);
1375 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1377 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1378 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1380 /* We can't seem to give an error return code
1381 * to .put_super, so we better make sure we clean up! */
1383 class_manual_cleanup(obd);
1385 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
1386 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1390 /* If they wanted the mgs to stop separately from the mdt, they
1391 should have put it on a different device. */
1392 if (IS_MGS(lsi->lsi_ldd)) {
1393 /* stop the mgc before the mgs so the connection gets cleaned
1395 lustre_stop_mgc(sb);
1396 server_stop_mgs(sb);
1399 /* Clean the mgc and sb */
1400 rc = lustre_common_put_super(sb);
1401 /* FIXME how can I report a failure to umount? */
1403 /* Wait for the targets to really clean up - can't exit (and let the
1404 sb get destroyed) while the mount is still in use */
1405 server_wait_finished(mnt);
1407 /* drop the One True Mount */
1410 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1411 until the target is really gone so that our type refcount check
1413 server_stop_servers(lddflags, lsiflags);
1415 /* In case of startup or cleanup err, stop related obds */
1417 obd = class_name2obd(extraname);
1419 CWARN("Cleaning orphaned obd %s\n", extraname);
1421 class_manual_cleanup(obd);
1423 OBD_FREE(extraname, strlen(extraname) + 1);
1426 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1427 OBD_FREE(tmpname, tmpname_sz);
1431 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1432 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1434 struct super_block *sb = vfsmnt->mnt_sb;
1436 static void server_umount_begin(struct super_block *sb)
1439 struct lustre_sb_info *lsi = s2lsi(sb);
1442 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1443 if (!(flags & MNT_FORCE)) {
1449 CDEBUG(D_MOUNT, "umount -f\n");
1450 /* umount = failover
1452 no third way to do non-force, non-failover */
1453 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1454 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1458 #ifndef HAVE_STATFS_DENTRY_PARAM
1459 static int server_statfs (struct super_block *sb, struct kstatfs *buf)
1462 static int server_statfs (struct dentry *dentry, struct kstatfs *buf)
1464 struct super_block *sb = dentry->d_sb;
1466 struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt;
1469 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) {
1470 #ifdef HAVE_STATFS_DENTRY_PARAM
1471 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf);
1473 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_sb, buf);
1476 buf->f_type = sb->s_magic;
1482 buf->f_type = sb->s_magic;
1483 buf->f_bsize = sb->s_blocksize;
1489 buf->f_namelen = NAME_MAX;
1493 static struct super_operations server_ops =
1495 .put_super = server_put_super,
1496 .umount_begin = server_umount_begin, /* umount -f */
1497 .statfs = server_statfs,
1500 #define log2(n) ffz(~(n))
1501 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1503 static int server_fill_super_common(struct super_block *sb)
1505 struct inode *root = 0;
1508 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1510 sb->s_blocksize = 4096;
1511 sb->s_blocksize_bits = log2(sb->s_blocksize);
1512 sb->s_magic = LUSTRE_SUPER_MAGIC;
1513 sb->s_maxbytes = 0; //PAGE_CACHE_MAXBYTES;
1514 sb->s_flags |= MS_RDONLY;
1515 sb->s_op = &server_ops;
1517 root = new_inode(sb);
1519 CERROR("Can't make root inode\n");
1523 /* returns -EIO for every operation */
1524 /* make_bad_inode(root); -- badness - can't umount */
1525 /* apparently we need to be a directory for the mount to finish */
1526 root->i_mode = S_IFDIR;
1528 sb->s_root = d_alloc_root(root);
1530 CERROR("Can't make root dentry\n");
1538 static int server_fill_super(struct super_block *sb)
1540 struct lustre_sb_info *lsi = s2lsi(sb);
1541 struct vfsmount *mnt;
1545 /* the One True Mount */
1546 mnt = server_kernel_mount(sb);
1549 CERROR("Unable to mount device %s: %d\n",
1550 lsi->lsi_lmd->lmd_dev, rc);
1554 lsi->lsi_srv_mnt = mnt;
1556 LASSERT(lsi->lsi_ldd);
1557 CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n",
1558 lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname,
1559 lsi->lsi_lmd->lmd_dev);
1561 if (class_name2obd(lsi->lsi_ldd->ldd_svname)) {
1562 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1563 "running. Double-mount may have compromised"
1564 " the disk journal.\n",
1565 lsi->lsi_ldd->ldd_svname);
1568 GOTO(out, rc = -EALREADY);
1571 /* start MGS before MGC */
1572 if (IS_MGS(lsi->lsi_ldd)) {
1573 rc = server_start_mgs(sb);
1578 rc = lustre_start_mgc(sb);
1582 /* Set up all obd devices for service */
1583 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1584 (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) {
1585 rc = server_start_targets(sb, mnt);
1587 CERROR("Unable to start targets: %d\n", rc);
1590 /* FIXME overmount client here,
1591 or can we just start a client log and client_fill_super on this sb?
1592 We need to make sure server_put_super gets called too - ll_put_super
1593 calls lustre_common_put_super; check there for LSI_SERVER flag,
1595 Probably should start client from new thread so we can return.
1596 Client will not finish until all servers are connected.
1597 Note - MGS-only server does NOT get a client, since there is no
1598 lustre fs associated - the MGS is for all lustre fs's */
1601 rc = server_fill_super_common(sb);
1605 LCONSOLE_WARN("Server %s on device %s has started\n",
1606 lsi->lsi_ldd->ldd_svname, lsi->lsi_lmd->lmd_dev);
1611 server_put_super(sb);
1616 /* Get the index from the obd name.
1617 rc = server type, or
1619 if endptr isn't NULL it is set to end of name */
1620 int server_name2index(char *svname, __u32 *idx, char **endptr)
1622 unsigned long index;
1624 char *dash = strchr(svname, '-');
1628 if (strncmp(dash + 1, "MDT", 3) == 0)
1629 rc = LDD_F_SV_TYPE_MDT;
1630 else if (strncmp(dash + 1, "OST", 3) == 0)
1631 rc = LDD_F_SV_TYPE_OST;
1635 index = simple_strtoul(dash + 4, endptr, 16);
1640 /*************** mount common betweeen server and client ***************/
1643 int lustre_common_put_super(struct super_block *sb)
1648 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1650 /* Drop a ref to the MGC */
1651 rc = lustre_stop_mgc(sb);
1652 if (rc && (rc != -ENOENT)) {
1654 CERROR("Can't stop MGC: %d\n", rc);
1657 /* BUSY just means that there's some other obd that
1658 needs the mgc. Let him clean it up. */
1659 CDEBUG(D_MOUNT, "MGC still in use\n");
1661 /* Drop a ref to the mounted disk */
1667 static void lmd_print(struct lustre_mount_data *lmd)
1671 PRINT_CMD(PRINT_MASK, " mount data:\n");
1672 if (lmd_is_client(lmd))
1673 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
1674 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
1675 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
1677 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
1678 for (i = 0; i < lmd->lmd_exclude_count; i++) {
1679 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
1680 lmd->lmd_exclude[i]);
1685 /* Is this server on the exclusion list */
1686 int lustre_check_exclusion(struct super_block *sb, char *svname)
1688 struct lustre_sb_info *lsi = s2lsi(sb);
1689 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1694 rc = server_name2index(svname, &index, NULL);
1695 if (rc != LDD_F_SV_TYPE_OST)
1696 /* Only exclude OSTs */
1699 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
1700 index, lmd->lmd_exclude_count, lmd->lmd_dev);
1702 for(i = 0; i < lmd->lmd_exclude_count; i++) {
1703 if (index == lmd->lmd_exclude[i]) {
1704 CWARN("Excluding %s (on exclusion list)\n", svname);
1711 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
1712 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
1714 char *s1 = ptr, *s2;
1715 __u32 index, *exclude_list;
1719 /* The shortest an ost name can be is 8 chars: -OST0000.
1720 We don't actually know the fsname at this time, so in fact
1721 a user could specify any fsname. */
1722 devmax = strlen(ptr) / 8 + 1;
1724 /* temp storage until we figure out how many we have */
1725 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
1729 /* we enter this fn pointing at the '=' */
1730 while (*s1 && *s1 != ' ' && *s1 != ',') {
1732 rc = server_name2index(s1, &index, &s2);
1734 CERROR("Can't parse server name '%s'\n", s1);
1737 if (rc == LDD_F_SV_TYPE_OST)
1738 exclude_list[lmd->lmd_exclude_count++] = index;
1740 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
1742 /* now we are pointing at ':' (next exclude)
1743 or ',' (end of excludes) */
1744 if (lmd->lmd_exclude_count >= devmax)
1747 if (rc >= 0) /* non-err */
1750 if (lmd->lmd_exclude_count) {
1751 /* permanent, freed in lustre_free_lsi */
1752 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
1753 lmd->lmd_exclude_count);
1754 if (lmd->lmd_exclude) {
1755 memcpy(lmd->lmd_exclude, exclude_list,
1756 sizeof(index) * lmd->lmd_exclude_count);
1759 lmd->lmd_exclude_count = 0;
1762 OBD_FREE(exclude_list, sizeof(index) * devmax);
1766 /* mount -v -t lustre uml1:uml2:/lustre-client /mnt/lustre */
1767 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1769 char *s1, *s2, *devname = NULL;
1770 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1776 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1777 "/sbin/mount.lustre is installed.\n");
1781 /* Options should be a string - try to detect old lmd data */
1782 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1783 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1784 "/sbin/mount.lustre. Please install "
1785 "version %s\n", LUSTRE_VERSION_STRING);
1788 lmd->lmd_magic = LMD_MAGIC;
1790 /* Set default flags here */
1795 /* Skip whitespace and extra commas */
1796 while (*s1 == ' ' || *s1 == ',')
1799 /* Client options are parsed in ll_options: eg. flock,
1802 /* Parse non-ldiskfs options here. Rather than modifying
1803 ldiskfs, we just zero these out here */
1804 if (strncmp(s1, "abort_recov", 11) == 0) {
1805 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1807 } else if (strncmp(s1, "nosvc", 5) == 0) {
1808 lmd->lmd_flags |= LMD_FLG_NOSVC;
1810 /* ost exclusion list */
1811 } else if (strncmp(s1, "exclude=", 8) == 0) {
1812 rc = lmd_make_exclusion(lmd, s1 + 7);
1817 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1818 end of the options. */
1819 else if (strncmp(s1, "device=", 7) == 0) {
1821 /* terminate options right before device. device
1822 must be the last one. */
1828 s2 = strchr(s1, ',');
1836 memmove(s1, s2, strlen(s2) + 1);
1842 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1843 "(need mount option 'device=...')\n");
1847 s1 = strrchr(devname, ':');
1849 lmd->lmd_flags = LMD_FLG_CLIENT;
1850 /* Remove leading /s from fsname */
1851 while (*++s1 == '/') ;
1852 /* Freed in lustre_free_lsi */
1853 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1854 if (!lmd->lmd_profile)
1856 sprintf(lmd->lmd_profile, "%s-client", s1);
1859 /* Freed in lustre_free_lsi */
1860 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1863 strcpy(lmd->lmd_dev, devname);
1865 /* Save mount options */
1866 s1 = options + strlen(options) - 1;
1867 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1869 if (*options != 0) {
1870 /* Freed in lustre_free_lsi */
1871 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1874 strcpy(lmd->lmd_opts, options);
1877 lmd->lmd_magic = LMD_MAGIC;
1882 CERROR("Bad mount options %s\n", options);
1888 int lustre_fill_super(struct super_block *sb, void *data, int silent)
1890 struct lustre_mount_data *lmd;
1891 struct lustre_sb_info *lsi;
1895 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1897 lsi = lustre_init_lsi(sb);
1902 /* Figure out the lmd from the mount options */
1903 if (lmd_parse((char *)data, lmd)) {
1908 if (lmd_is_client(lmd)) {
1909 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1910 if (!client_fill_super) {
1911 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1912 "client mount! Is the 'lustre' "
1913 "module loaded?\n");
1916 rc = lustre_start_mgc(sb);
1918 lustre_stop_mgc(sb);
1921 /* Connect and start */
1922 /* (should always be ll_fill_super) */
1923 rc = (*client_fill_super)(sb);
1924 /* c_f_s will call lustre_common_put_super on failure */
1927 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1928 lsi->lsi_flags |= LSI_SERVER;
1929 rc = server_fill_super(sb);
1930 /* s_f_s calls lustre_start_mgc after the mount because we need
1931 the MGS nids which are stored on disk. Plus, we may
1932 need to start the MGS first. */
1933 /* s_f_s will call server_put_super on failure */
1938 CERROR("Unable to mount %s (%d)\n",
1939 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1941 CDEBUG(D_SUPER, "mount %s complete\n", lmd->lmd_dev);
1947 /* We can't call ll_fill_super by name because it lives in a module that
1948 must be loaded after this one. */
1949 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb))
1951 client_fill_super = cfs;
1954 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
1956 kill_super_cb = cfs;
1959 /***************** FS registration ******************/
1961 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))
1962 struct super_block * lustre_get_sb(struct file_system_type *fs_type,
1963 int flags, const char *devname, void * data)
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);
1971 int lustre_get_sb(struct file_system_type *fs_type,
1972 int flags, const char *devname, void * data,
1973 struct vfsmount *mnt)
1975 /* calls back in fill super */
1976 /* we could append devname= onto options (*data) here,
1977 but 2.4 doesn't get devname. So we do it in mount_lustre.c */
1978 return get_sb_nodev(fs_type, flags, data, lustre_fill_super, mnt);
1982 void lustre_kill_super(struct super_block *sb)
1984 struct lustre_sb_info *lsi = s2lsi(sb);
1986 if (kill_super_cb && lsi &&(lsi->lsi_flags & LSI_SERVER))
1987 (*kill_super_cb)(sb);
1989 kill_anon_super(sb);
1992 struct file_system_type lustre_fs_type = {
1993 .owner = THIS_MODULE,
1995 .get_sb = lustre_get_sb,
1996 .kill_sb = lustre_kill_super,
1997 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV,
2000 int lustre_register_fs(void)
2002 return register_filesystem(&lustre_fs_type);
2005 int lustre_unregister_fs(void)
2007 return unregister_filesystem(&lustre_fs_type);
2010 EXPORT_SYMBOL(lustre_register_client_fill_super);
2011 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2012 EXPORT_SYMBOL(lustre_common_put_super);
2013 EXPORT_SYMBOL(lustre_process_log);
2014 EXPORT_SYMBOL(lustre_end_log);
2015 EXPORT_SYMBOL(server_get_mount);
2016 EXPORT_SYMBOL(server_get_mount_2);
2017 EXPORT_SYMBOL(server_put_mount);
2018 EXPORT_SYMBOL(server_put_mount_2);
2019 EXPORT_SYMBOL(server_register_target);
2020 EXPORT_SYMBOL(server_name2index);
2021 EXPORT_SYMBOL(server_mti_print);
2022 EXPORT_SYMBOL(do_lcfg);