1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2011, 2012, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/obdclass/obd_mount.c
40 * Client/server mount routines
42 * Author: Nathan Rutman <nathan@clusterfs.com>
46 #define DEBUG_SUBSYSTEM S_CLASS
47 #define D_MOUNT D_SUPER|D_CONFIG /*|D_WARNING */
48 #define PRINT_CMD CDEBUG
49 #define PRINT_MASK D_SUPER|D_CONFIG
53 #include <lustre_fsfilt.h>
54 #include <obd_class.h>
55 #include <lustre/lustre_user.h>
56 #include <linux/version.h>
57 #include <lustre_log.h>
58 #include <lustre_disk.h>
59 #include <lustre_param.h>
61 static int (*client_fill_super)(struct super_block *sb,
62 struct vfsmount *mnt) = NULL;
63 static void (*kill_super_cb)(struct super_block *sb) = NULL;
65 /*********** mount lookup *********/
67 CFS_DECLARE_MUTEX(lustre_mount_info_lock);
68 static CFS_LIST_HEAD(server_mount_info_list);
70 static struct lustre_mount_info *server_find_mount(const char *name)
73 struct lustre_mount_info *lmi;
76 cfs_list_for_each(tmp, &server_mount_info_list) {
77 lmi = cfs_list_entry(tmp, struct lustre_mount_info,
79 if (strcmp(name, lmi->lmi_name) == 0)
85 /* we must register an obd for a mount before we call the setup routine.
86 *_setup will call lustre_get_mount to get the mnt struct
87 by obd_name, since we can't pass the pointer to setup. */
88 static int server_register_mount(const char *name, struct super_block *sb,
91 struct lustre_mount_info *lmi;
98 OBD_ALLOC(lmi, sizeof(*lmi));
101 OBD_ALLOC(name_cp, strlen(name) + 1);
103 OBD_FREE(lmi, sizeof(*lmi));
106 strcpy(name_cp, name);
108 cfs_down(&lustre_mount_info_lock);
110 if (server_find_mount(name)) {
111 cfs_up(&lustre_mount_info_lock);
112 OBD_FREE(lmi, sizeof(*lmi));
113 OBD_FREE(name_cp, strlen(name) + 1);
114 CERROR("Already registered %s\n", name);
117 lmi->lmi_name = name_cp;
120 cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list);
122 cfs_up(&lustre_mount_info_lock);
124 CDEBUG(D_MOUNT, "reg_mnt %p from %s, vfscount=%d\n",
125 lmi->lmi_mnt, name, mnt_get_count(lmi->lmi_mnt));
130 /* when an obd no longer needs a mount */
131 static int server_deregister_mount(const char *name)
133 struct lustre_mount_info *lmi;
136 cfs_down(&lustre_mount_info_lock);
137 lmi = server_find_mount(name);
139 cfs_up(&lustre_mount_info_lock);
140 CERROR("%s not registered\n", name);
144 CDEBUG(D_MOUNT, "dereg_mnt %p from %s, vfscount=%d\n",
145 lmi->lmi_mnt, name, mnt_get_count(lmi->lmi_mnt));
147 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
148 cfs_list_del(&lmi->lmi_list_chain);
149 OBD_FREE(lmi, sizeof(*lmi));
150 cfs_up(&lustre_mount_info_lock);
155 /* obd's look up a registered mount using their obdname. This is just
156 for initial obd setup to find the mount struct. It should not be
157 called every time you want to mntget. */
158 struct lustre_mount_info *server_get_mount(const char *name)
160 struct lustre_mount_info *lmi;
161 struct lustre_sb_info *lsi;
164 cfs_down(&lustre_mount_info_lock);
165 lmi = server_find_mount(name);
166 cfs_up(&lustre_mount_info_lock);
168 CERROR("Can't find mount for %s\n", name);
171 lsi = s2lsi(lmi->lmi_sb);
172 mntget(lmi->lmi_mnt);
173 cfs_atomic_inc(&lsi->lsi_mounts);
175 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d, vfscount=%d\n",
176 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts),
177 mnt_get_count(lmi->lmi_mnt));
183 * Used by mdt to get mount_info from obdname.
184 * There are no blocking when using the mount_info.
185 * Do not use server_get_mount for this purpose.
187 struct lustre_mount_info *server_get_mount_2(const char *name)
189 struct lustre_mount_info *lmi;
192 cfs_down(&lustre_mount_info_lock);
193 lmi = server_find_mount(name);
194 cfs_up(&lustre_mount_info_lock);
196 CERROR("Can't find mount for %s\n", name);
201 static void unlock_mntput(struct vfsmount *mnt)
203 if (kernel_locked()) {
212 static int lustre_put_lsi(struct super_block *sb);
214 /* to be called from obd_cleanup methods */
215 int server_put_mount(const char *name, struct vfsmount *mnt)
217 struct lustre_mount_info *lmi;
218 struct lustre_sb_info *lsi;
219 int count = mnt_get_count(mnt) - 1;
222 /* This might be the last one, can't deref after this */
225 cfs_down(&lustre_mount_info_lock);
226 lmi = server_find_mount(name);
227 cfs_up(&lustre_mount_info_lock);
229 CERROR("Can't find mount for %s\n", name);
232 lsi = s2lsi(lmi->lmi_sb);
233 LASSERT(lmi->lmi_mnt == mnt);
235 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d, vfscount=%d\n",
236 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts), count);
238 if (lustre_put_lsi(lmi->lmi_sb)) {
239 CDEBUG(D_MOUNT, "Last put of mnt %p from %s, vfscount=%d\n",
240 lmi->lmi_mnt, name, count);
241 /* last mount is the One True Mount */
243 CERROR("%s: mount busy, vfscount=%d!\n", name, count);
246 /* this obd should never need the mount again */
247 server_deregister_mount(name);
252 /* Corresponding to server_get_mount_2 */
253 int server_put_mount_2(const char *name, struct vfsmount *mnt)
259 /******* mount helper utilities *********/
262 static void ldd_print(struct lustre_disk_data *ldd)
264 PRINT_CMD(PRINT_MASK, " disk data:\n");
265 PRINT_CMD(PRINT_MASK, "server: %s\n", ldd->ldd_svname);
266 PRINT_CMD(PRINT_MASK, "uuid: %s\n", (char *)ldd->ldd_uuid);
267 PRINT_CMD(PRINT_MASK, "fs: %s\n", ldd->ldd_fsname);
268 PRINT_CMD(PRINT_MASK, "index: %04x\n", ldd->ldd_svindex);
269 PRINT_CMD(PRINT_MASK, "config: %d\n", ldd->ldd_config_ver);
270 PRINT_CMD(PRINT_MASK, "flags: %#x\n", ldd->ldd_flags);
271 PRINT_CMD(PRINT_MASK, "diskfs: %s\n", MT_STR(ldd));
272 PRINT_CMD(PRINT_MASK, "options: %s\n", ldd->ldd_mount_opts);
273 PRINT_CMD(PRINT_MASK, "params: %s\n", ldd->ldd_params);
274 PRINT_CMD(PRINT_MASK, "comment: %s\n", ldd->ldd_userdata);
278 static int ldd_parse(struct lvfs_run_ctxt *mount_ctxt,
279 struct lustre_disk_data *ldd)
281 struct lvfs_run_ctxt saved;
288 push_ctxt(&saved, mount_ctxt, NULL);
290 file = filp_open(MOUNT_DATA_FILE, O_RDONLY, 0644);
293 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
297 len = i_size_read(file->f_dentry->d_inode);
298 CDEBUG(D_MOUNT, "Have %s, size %lu\n", MOUNT_DATA_FILE, len);
299 if (len != sizeof(*ldd)) {
300 CERROR("disk data size does not match: see %lu expect %u\n",
301 len, (int)sizeof(*ldd));
302 GOTO(out_close, rc = -EINVAL);
305 rc = lustre_fread(file, ldd, len, &off);
307 CERROR("error reading %s: read %d of %lu\n",
308 MOUNT_DATA_FILE, rc, len);
309 GOTO(out_close, rc = -EINVAL);
313 if (ldd->ldd_magic != LDD_MAGIC) {
314 /* FIXME add swabbing support */
315 CERROR("Bad magic in %s: %x!=%x\n", MOUNT_DATA_FILE,
316 ldd->ldd_magic, LDD_MAGIC);
317 GOTO(out_close, rc = -EINVAL);
320 if (ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP) {
321 CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
323 ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP);
324 GOTO(out_close, rc = -EINVAL);
326 if (ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP) {
327 CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
329 ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP);
330 /* Do something like remount filesystem read-only */
331 GOTO(out_close, rc = -EINVAL);
337 pop_ctxt(&saved, mount_ctxt, NULL);
341 static int ldd_write(struct lvfs_run_ctxt *mount_ctxt,
342 struct lustre_disk_data *ldd)
344 struct lvfs_run_ctxt saved;
347 unsigned long len = sizeof(struct lustre_disk_data);
351 LASSERT(ldd->ldd_magic == LDD_MAGIC);
353 ldd->ldd_config_ver++;
355 push_ctxt(&saved, mount_ctxt, NULL);
357 file = filp_open(MOUNT_DATA_FILE, O_RDWR|O_SYNC, 0644);
360 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
364 rc = lustre_fwrite(file, ldd, len, &off);
366 CERROR("error writing %s: read %d of %lu\n",
367 MOUNT_DATA_FILE, rc, len);
368 GOTO(out_close, rc = -EINVAL);
376 pop_ctxt(&saved, mount_ctxt, NULL);
381 /**************** config llog ********************/
383 /** Get a config log from the MGS and process it.
384 * This func is called for both clients and servers.
385 * Continue to process new statements appended to the logs
386 * (whenever the config lock is revoked) until lustre_end_log
388 * @param sb The superblock is used by the MGC to write to the local copy of
390 * @param logname The name of the llog to replicate from the MGS
391 * @param cfg Since the same mgc may be used to follow multiple config logs
392 * (e.g. ost1, ost2, client), the config_llog_instance keeps the state for
393 * this log, and is added to the mgc's list of logs to follow.
395 int lustre_process_log(struct super_block *sb, char *logname,
396 struct config_llog_instance *cfg)
398 struct lustre_cfg *lcfg;
399 struct lustre_cfg_bufs *bufs;
400 struct lustre_sb_info *lsi = s2lsi(sb);
401 struct obd_device *mgc = lsi->lsi_mgc;
412 /* mgc_process_config */
413 lustre_cfg_bufs_reset(bufs, mgc->obd_name);
414 lustre_cfg_bufs_set_string(bufs, 1, logname);
415 lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg));
416 lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb));
417 lcfg = lustre_cfg_new(LCFG_LOG_START, bufs);
418 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
419 lustre_cfg_free(lcfg);
424 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
425 "failed from the MGS (%d). Make sure this "
426 "client and the MGS are running compatible "
427 "versions of Lustre.\n",
428 mgc->obd_name, logname, rc);
431 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
432 "failed (%d). This may be the result of "
433 "communication errors between this node and "
434 "the MGS, a bad configuration, or other "
435 "errors. See the syslog for more "
436 "information.\n", mgc->obd_name, logname,
439 /* class_obd_list(); */
443 /* Stop watching this config log for updates */
444 int lustre_end_log(struct super_block *sb, char *logname,
445 struct config_llog_instance *cfg)
447 struct lustre_cfg *lcfg;
448 struct lustre_cfg_bufs bufs;
449 struct lustre_sb_info *lsi = s2lsi(sb);
450 struct obd_device *mgc = lsi->lsi_mgc;
457 /* mgc_process_config */
458 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
459 lustre_cfg_bufs_set_string(&bufs, 1, logname);
461 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
462 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
463 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
464 lustre_cfg_free(lcfg);
468 /**************** obd start *******************/
470 /** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from
471 * lctl (and do for echo cli/srv.
473 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
474 char *s1, char *s2, char *s3, char *s4)
476 struct lustre_cfg_bufs bufs;
477 struct lustre_cfg * lcfg = NULL;
480 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
481 cmd, s1, s2, s3, s4);
483 lustre_cfg_bufs_reset(&bufs, cfgname);
485 lustre_cfg_bufs_set_string(&bufs, 1, s1);
487 lustre_cfg_bufs_set_string(&bufs, 2, s2);
489 lustre_cfg_bufs_set_string(&bufs, 3, s3);
491 lustre_cfg_bufs_set_string(&bufs, 4, s4);
493 lcfg = lustre_cfg_new(cmd, &bufs);
494 lcfg->lcfg_nid = nid;
495 rc = class_process_config(lcfg);
496 lustre_cfg_free(lcfg);
500 /** Call class_attach and class_setup. These methods in turn call
501 * obd type-specific methods.
503 static int lustre_start_simple(char *obdname, char *type, char *uuid,
507 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
509 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
511 CERROR("%s attach error %d\n", obdname, rc);
514 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, 0, 0);
516 CERROR("%s setup error %d\n", obdname, rc);
517 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
522 /* Set up a MGS to serve startup logs */
523 static int server_start_mgs(struct super_block *sb)
525 struct lustre_sb_info *lsi = s2lsi(sb);
526 struct vfsmount *mnt = lsi->lsi_srv_mnt;
527 struct lustre_mount_info *lmi;
532 /* It is impossible to have more than 1 MGS per node, since
533 MGC wouldn't know which to connect to */
534 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
536 lsi = s2lsi(lmi->lmi_sb);
537 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
539 lsi->lsi_ldd->ldd_svname);
543 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
545 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
548 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
549 LUSTRE_MGS_OBDNAME, 0, 0);
550 /* Do NOT call server_deregister_mount() here. This leads to
551 * inability cleanup cleanly and free lsi and other stuff when
552 * mgs calls server_put_mount() in error handling case. -umka */
556 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
557 "Is the 'mgs' module loaded?\n",
558 LUSTRE_MGS_OBDNAME, rc);
562 static int server_stop_mgs(struct super_block *sb)
564 struct obd_device *obd;
568 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
570 /* There better be only one MGS */
571 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
573 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
577 /* The MGS should always stop when we say so */
579 rc = class_manual_cleanup(obd);
583 CFS_DECLARE_MUTEX(mgc_start_lock);
585 /** Set up a mgc obd to process startup logs
587 * \param sb [in] super block of the mgc obd
589 * \retval 0 success, otherwise error code
591 static int lustre_start_mgc(struct super_block *sb)
593 struct obd_connect_data *data = NULL;
594 struct lustre_sb_info *lsi = s2lsi(sb);
595 struct obd_device *obd;
596 struct obd_export *exp;
597 struct obd_uuid *uuid;
600 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
603 int rc = 0, i = 0, j, len;
606 LASSERT(lsi->lsi_lmd);
608 /* Find the first non-lo MGS nid for our MGC name */
609 if (lsi->lsi_flags & LSI_SERVER) {
610 ptr = lsi->lsi_ldd->ldd_params;
611 /* Use mgsnode= nids */
612 if ((class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0) &&
613 (class_parse_nid(ptr, &nid, &ptr) == 0)) {
615 } else if (IS_MGS(lsi->lsi_ldd)) {
616 lnet_process_id_t id;
617 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
618 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
625 } else { /* client */
626 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
627 ptr = lsi->lsi_lmd->lmd_dev;
628 if (class_parse_nid(ptr, &nid, &ptr) == 0)
632 CERROR("No valid MGS nids found.\n");
636 cfs_mutex_down(&mgc_start_lock);
638 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
639 OBD_ALLOC(mgcname, len);
640 OBD_ALLOC(niduuid, len + 2);
641 if (!mgcname || !niduuid)
642 GOTO(out_free, rc = -ENOMEM);
643 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
645 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
649 GOTO(out_free, rc = -ENOMEM);
651 obd = class_name2obd(mgcname);
652 if (obd && !obd->obd_stopping) {
653 rc = obd_set_info_async(obd->obd_self_export,
654 strlen(KEY_MGSSEC), KEY_MGSSEC,
655 strlen(mgssec), mgssec, NULL);
659 /* Re-using an existing MGC */
660 cfs_atomic_inc(&obd->u.cli.cl_mgc_refcount);
662 /* IR compatibility check, only for clients */
663 if (lmd_is_client(lsi->lsi_lmd)) {
665 int vallen = sizeof(*data);
666 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
668 rc = obd_get_info(obd->obd_self_export,
669 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
670 &vallen, data, NULL);
672 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
673 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
674 /* LMD_FLG_NOIR is for test purpose only */
676 "Trying to mount a client with IR setting "
677 "not compatible with current mgc. "
678 "Force to use current mgc setting that is "
680 has_ir ? "enabled" : "disabled");
682 *flags &= ~LMD_FLG_NOIR;
684 *flags |= LMD_FLG_NOIR;
689 /* If we are restarting the MGS, don't try to keep the MGC's
690 old connection, or registration will fail. */
691 if ((lsi->lsi_flags & LSI_SERVER) && IS_MGS(lsi->lsi_ldd)) {
692 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
696 /* Try all connections, but only once (again).
697 We don't want to block another target from starting
698 (using its local copy of the log), but we do want to connect
699 if at all possible. */
701 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
702 rc = obd_set_info_async(obd->obd_self_export,
703 sizeof(KEY_INIT_RECOV_BACKUP),
704 KEY_INIT_RECOV_BACKUP,
705 sizeof(recov_bk), &recov_bk, NULL);
709 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
711 /* Add the primary nids for the MGS */
713 sprintf(niduuid, "%s_%x", mgcname, i);
714 if (lsi->lsi_flags & LSI_SERVER) {
715 ptr = lsi->lsi_ldd->ldd_params;
716 if (IS_MGS(lsi->lsi_ldd)) {
717 /* Use local nids (including LO) */
718 lnet_process_id_t id;
719 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
720 rc = do_lcfg(mgcname, id.nid,
721 LCFG_ADD_UUID, niduuid, 0,0,0);
724 /* Use mgsnode= nids */
725 if (class_find_param(ptr, PARAM_MGSNODE, &ptr) != 0) {
726 CERROR("No MGS nids given.\n");
727 GOTO(out_free, rc = -EINVAL);
729 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
730 rc = do_lcfg(mgcname, nid,
731 LCFG_ADD_UUID, niduuid, 0,0,0);
735 } else { /* client */
736 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
737 ptr = lsi->lsi_lmd->lmd_dev;
738 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
739 rc = do_lcfg(mgcname, nid,
740 LCFG_ADD_UUID, niduuid, 0,0,0);
742 /* Stop at the first failover nid */
748 CERROR("No valid MGS nids found.\n");
749 GOTO(out_free, rc = -EINVAL);
751 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
753 /* Random uuid for MGC allows easier reconnects */
755 ll_generate_random_uuid(uuidc);
756 class_uuid_unparse(uuidc, uuid);
759 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
760 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
766 /* Add any failover MGS nids */
768 while ((*ptr == ':' ||
769 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0)) {
770 /* New failover node */
771 sprintf(niduuid, "%s_%x", mgcname, i);
773 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
775 rc = do_lcfg(mgcname, nid,
776 LCFG_ADD_UUID, niduuid, 0,0,0);
781 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
789 lsi->lsi_lmd->lmd_mgs_failnodes = i;
791 obd = class_name2obd(mgcname);
793 CERROR("Can't find mgcobd %s\n", mgcname);
794 GOTO(out_free, rc = -ENOTCONN);
797 rc = obd_set_info_async(obd->obd_self_export,
798 strlen(KEY_MGSSEC), KEY_MGSSEC,
799 strlen(mgssec), mgssec, NULL);
803 /* Keep a refcount of servers/clients who started with "mount",
804 so we know when we can get rid of the mgc. */
805 cfs_atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
807 /* Try all connections, but only once. */
809 rc = obd_set_info_async(obd->obd_self_export,
810 sizeof(KEY_INIT_RECOV_BACKUP),
811 KEY_INIT_RECOV_BACKUP,
812 sizeof(recov_bk), &recov_bk, NULL);
815 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
816 /* We connect to the MGS at setup, and don't disconnect until cleanup */
817 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID |
818 OBD_CONNECT_AT | OBD_CONNECT_FULL20 |
819 OBD_CONNECT_IMP_RECOV;
820 if (lmd_is_client(lsi->lsi_lmd) &&
821 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
822 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
823 data->ocd_version = LUSTRE_VERSION_CODE;
824 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
826 CERROR("connect failed %d\n", rc);
830 obd->u.cli.cl_mgc_mgsexp = exp;
833 /* Keep the mgc info in the sb. Note that many lsi's can point
837 cfs_mutex_up(&mgc_start_lock);
842 OBD_FREE(mgcname, len);
844 OBD_FREE(niduuid, len + 2);
848 static int lustre_stop_mgc(struct super_block *sb)
850 struct lustre_sb_info *lsi = s2lsi(sb);
851 struct obd_device *obd;
852 char *niduuid = 0, *ptr = 0;
853 int i, rc = 0, len = 0;
863 cfs_mutex_down(&mgc_start_lock);
864 LASSERT(cfs_atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
865 if (!cfs_atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
866 /* This is not fatal, every client that stops
867 will call in here. */
868 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
869 cfs_atomic_read(&obd->u.cli.cl_mgc_refcount));
870 GOTO(out, rc = -EBUSY);
873 /* The MGC has no recoverable data in any case.
874 * force shotdown set in umount_begin */
875 obd->obd_no_recov = 1;
877 if (obd->u.cli.cl_mgc_mgsexp) {
878 /* An error is not fatal, if we are unable to send the
879 disconnect mgs ping evictor cleans up the export */
880 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
882 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
885 /* Save the obdname for cleaning the nid uuids, which are
887 len = strlen(obd->obd_name) + 6;
888 OBD_ALLOC(niduuid, len);
890 strcpy(niduuid, obd->obd_name);
891 ptr = niduuid + strlen(niduuid);
894 rc = class_manual_cleanup(obd);
898 /* Clean the nid uuids */
900 GOTO(out, rc = -ENOMEM);
902 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
903 sprintf(ptr, "_%x", i);
904 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
907 CERROR("del MDC UUID %s failed: rc = %d\n",
912 OBD_FREE(niduuid, len);
914 /* class_import_put will get rid of the additional connections */
915 cfs_mutex_up(&mgc_start_lock);
919 /* Since there's only one mgc per node, we have to change it's fs to get
920 access to the right disk. */
921 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
923 struct lustre_sb_info *lsi = s2lsi(sb);
927 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
929 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
930 rc = obd_set_info_async(mgc->obd_self_export,
931 sizeof(KEY_SET_FS), KEY_SET_FS,
932 sizeof(*sb), sb, NULL);
934 CERROR("can't set_fs %d\n", rc);
940 static int server_mgc_clear_fs(struct obd_device *mgc)
945 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
947 rc = obd_set_info_async(mgc->obd_self_export,
948 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
953 CFS_DECLARE_MUTEX(server_start_lock);
955 /* Stop MDS/OSS if nobody is using them */
956 static int server_stop_servers(int lddflags, int lsiflags)
958 struct obd_device *obd = NULL;
959 struct obd_type *type = NULL;
963 cfs_mutex_down(&server_start_lock);
965 /* Either an MDT or an OST or neither */
966 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
967 if ((lddflags & LDD_F_SV_TYPE_MDT) &&
968 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
969 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
970 type = class_search_type(LUSTRE_MDS_NAME);
972 /* if this was an OST, and there are no more OST's, clean up the OSS */
973 if ((lddflags & LDD_F_SV_TYPE_OST) &&
974 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
975 type = class_search_type(LUSTRE_OST_NAME);
978 if (obd && (!type || !type->typ_refcnt)) {
981 /* obd_fail doesn't mean much on a server obd */
982 err = class_manual_cleanup(obd);
987 cfs_mutex_up(&server_start_lock);
992 int server_mti_print(char *title, struct mgs_target_info *mti)
994 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
995 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
996 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
997 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
998 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
999 mti->mti_config_ver, mti->mti_flags);
1003 static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
1005 struct lustre_sb_info *lsi = s2lsi(sb);
1006 struct lustre_disk_data *ldd = lsi->lsi_ldd;
1007 lnet_process_id_t id;
1011 if (!(lsi->lsi_flags & LSI_SERVER))
1014 strncpy(mti->mti_fsname, ldd->ldd_fsname,
1015 sizeof(mti->mti_fsname));
1016 strncpy(mti->mti_svname, ldd->ldd_svname,
1017 sizeof(mti->mti_svname));
1019 mti->mti_nid_count = 0;
1020 while (LNetGetId(i++, &id) != -ENOENT) {
1021 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1024 /* server use --servicenode param, only allow specified
1025 * nids be registered */
1026 if ((ldd->ldd_flags & LDD_F_NO_PRIMNODE) != 0 &&
1027 class_match_nid(ldd->ldd_params,
1028 PARAM_FAILNODE, id.nid) < 1)
1031 /* match specified network */
1032 if (!class_match_net(ldd->ldd_params,
1033 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1036 mti->mti_nids[mti->mti_nid_count] = id.nid;
1037 mti->mti_nid_count++;
1038 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1039 CWARN("Only using first %d nids for %s\n",
1040 mti->mti_nid_count, mti->mti_svname);
1045 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1046 mti->mti_config_ver = 0;
1047 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF)
1048 ldd->ldd_flags |= LDD_F_WRITECONF;
1049 mti->mti_flags = ldd->ldd_flags;
1050 mti->mti_stripe_index = ldd->ldd_svindex;
1051 memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid));
1052 if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) {
1053 CERROR("params too big for mti\n");
1056 memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params));
1060 /* Register an old or new target with the MGS. If needed MGS will construct
1061 startup logs and assign index */
1062 int server_register_target(struct super_block *sb)
1064 struct lustre_sb_info *lsi = s2lsi(sb);
1065 struct obd_device *mgc = lsi->lsi_mgc;
1066 struct lustre_disk_data *ldd = lsi->lsi_ldd;
1067 struct mgs_target_info *mti = NULL;
1074 if (!(lsi->lsi_flags & LSI_SERVER))
1080 rc = server_sb2mti(sb, mti);
1084 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1085 mti->mti_svname, mti->mti_fsname,
1086 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1089 /* if write_conf is true, the registration must succeed */
1090 writeconf = !!(ldd->ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1091 mti->mti_flags |= LDD_F_OPC_REG;
1093 /* Register the target */
1094 /* FIXME use mgc_process_config instead */
1095 rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
1096 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
1097 sizeof(*mti), mti, NULL);
1099 if (mti->mti_flags & LDD_F_ERROR) {
1100 LCONSOLE_ERROR_MSG(0x160,
1101 "The MGS is refusing to allow this "
1102 "server (%s) to start. Please see messages"
1103 " on the MGS node.\n", ldd->ldd_svname);
1104 } else if (writeconf) {
1105 LCONSOLE_ERROR_MSG(0x15f,
1106 "Communication to the MGS return error %d. "
1107 "Is the MGS running?\n", rc);
1109 CERROR("Cannot talk to the MGS: %d, not fatal\n", rc);
1110 /* reset the error code for non-fatal error. */
1116 /* Always update our flags */
1117 ldd->ldd_flags = mti->mti_flags & LDD_F_ONDISK_MASK;
1119 /* If this flag is set, it means the MGS wants us to change our
1120 on-disk data. (So far this means just the index.) */
1121 if (mti->mti_flags & LDD_F_REWRITE_LDD) {
1124 CDEBUG(D_MOUNT, "Changing on-disk index from %#x to %#x "
1125 "for %s\n", ldd->ldd_svindex, mti->mti_stripe_index,
1127 ldd->ldd_svindex = mti->mti_stripe_index;
1128 strncpy(ldd->ldd_svname, mti->mti_svname,
1129 sizeof(ldd->ldd_svname));
1130 /* or ldd_make_sv_name(ldd); */
1131 ldd_write(&mgc->obd_lvfs_ctxt, ldd);
1132 err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb,
1135 CERROR("Label set error %d\n", err);
1136 label = fsfilt_get_label(mgc, lsi->lsi_srv_mnt->mnt_sb);
1138 CDEBUG(D_MOUNT, "Disk label changed to %s\n", label);
1140 /* Flush the new ldd to disk */
1141 fsfilt_sync(mgc, lsi->lsi_srv_mnt->mnt_sb);
1151 * Notify the MGS that this target is ready.
1152 * Used by IR - if the MGS receives this message, it will notify clients.
1154 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1156 struct lustre_sb_info *lsi = s2lsi(sb);
1157 struct obd_device *mgc = lsi->lsi_mgc;
1158 struct mgs_target_info *mti = NULL;
1164 if (!(lsi->lsi_flags & LSI_SERVER))
1170 rc = server_sb2mti(sb, mti);
1174 mti->mti_instance = obd->u.obt.obt_instance;
1175 mti->mti_flags |= LDD_F_OPC_READY;
1177 /* FIXME use mgc_process_config instead */
1178 rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
1179 sizeof(KEY_REGISTER_TARGET),
1180 KEY_REGISTER_TARGET,
1181 sizeof(*mti), mti, NULL);
1183 /* Imperative recovery: if the mgs informs us to use IR? */
1184 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1185 (mti->mti_flags & LDD_F_IR_CAPABLE))
1186 lsi->lsi_flags |= LSI_IR_CAPABLE;
1195 /** Start server targets: MDTs and OSTs
1197 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1199 struct obd_device *obd;
1200 struct lustre_sb_info *lsi = s2lsi(sb);
1201 struct config_llog_instance cfg;
1205 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname);
1208 /* If we're an MDT, make sure the global MDS is running */
1209 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1210 /* make sure the MDS is started */
1211 cfs_mutex_down(&server_start_lock);
1212 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1214 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1215 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1217 LUSTRE_MDS_OBDNAME"_uuid",
1220 cfs_mutex_up(&server_start_lock);
1221 CERROR("failed to start MDS: %d\n", rc);
1225 cfs_mutex_up(&server_start_lock);
1229 /* If we're an OST, make sure the global OSS is running */
1230 if (IS_OST(lsi->lsi_ldd)) {
1231 /* make sure OSS is started */
1232 cfs_mutex_down(&server_start_lock);
1233 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1235 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1237 LUSTRE_OSS_OBDNAME"_uuid",
1240 cfs_mutex_up(&server_start_lock);
1241 CERROR("failed to start OSS: %d\n", rc);
1245 cfs_mutex_up(&server_start_lock);
1248 /* Set the mgc fs to our server disk. This allows the MGC to
1249 * read and write configs locally, in case it can't talk to the MGS. */
1250 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1254 /* Register with MGS */
1255 rc = server_register_target(sb);
1259 /* Let the target look up the mount using the target's name
1260 (we can't pass the sb or mnt through class_process_config.) */
1261 rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
1265 /* Start targets using the llog named for the target */
1266 memset(&cfg, 0, sizeof(cfg));
1267 rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg);
1269 CERROR("failed to start server %s: %d\n",
1270 lsi->lsi_ldd->ldd_svname, rc);
1271 /* Do NOT call server_deregister_mount() here. This makes it
1272 * impossible to find mount later in cleanup time and leaves
1273 * @lsi and othder stuff leaked. -umka */
1278 /* Release the mgc fs for others to use */
1279 server_mgc_clear_fs(lsi->lsi_mgc);
1282 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1284 CERROR("no server named %s was started\n",
1285 lsi->lsi_ldd->ldd_svname);
1289 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1290 (OBP(obd, iocontrol))) {
1291 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1292 obd->obd_self_export, 0, NULL, NULL);
1295 server_notify_target(sb, obd);
1297 /* calculate recovery timeout, do it after lustre_process_log */
1298 server_calc_timeout(lsi, obd);
1300 /* log has been fully processed */
1301 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1307 /***************** lustre superblock **************/
1309 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1311 struct lustre_sb_info *lsi;
1317 OBD_ALLOC_PTR(lsi->lsi_lmd);
1318 if (!lsi->lsi_lmd) {
1323 lsi->lsi_lmd->lmd_exclude_count = 0;
1324 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
1325 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
1326 s2lsi_nocast(sb) = lsi;
1327 /* we take 1 extra ref for our setup */
1328 cfs_atomic_set(&lsi->lsi_mounts, 1);
1330 /* Default umount style */
1331 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1336 static int lustre_free_lsi(struct super_block *sb)
1338 struct lustre_sb_info *lsi = s2lsi(sb);
1341 LASSERT(lsi != NULL);
1342 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
1344 /* someone didn't call server_put_mount. */
1345 LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0);
1347 if (lsi->lsi_ldd != NULL)
1348 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
1350 if (lsi->lsi_lmd != NULL) {
1351 if (lsi->lsi_lmd->lmd_dev != NULL)
1352 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1353 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1354 if (lsi->lsi_lmd->lmd_profile != NULL)
1355 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1356 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1357 if (lsi->lsi_lmd->lmd_mgssec != NULL)
1358 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
1359 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
1360 if (lsi->lsi_lmd->lmd_opts != NULL)
1361 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1362 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1363 if (lsi->lsi_lmd->lmd_exclude_count)
1364 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1365 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1366 lsi->lsi_lmd->lmd_exclude_count);
1367 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1370 LASSERT(lsi->lsi_llsbi == NULL);
1371 OBD_FREE(lsi, sizeof(*lsi));
1372 s2lsi_nocast(sb) = NULL;
1377 /* The lsi has one reference for every server that is using the disk -
1378 e.g. MDT, MGS, and potentially MGC */
1379 static int lustre_put_lsi(struct super_block *sb)
1381 struct lustre_sb_info *lsi = s2lsi(sb);
1384 LASSERT(lsi != NULL);
1386 CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts));
1387 if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) {
1388 lustre_free_lsi(sb);
1394 /*************** server mount ******************/
1396 /** Kernel mount using mount options in MOUNT_DATA_FILE.
1397 * Since this file lives on the disk, we pre-mount using a common
1398 * type, read the file, then re-mount using the type specified in the
1401 static struct vfsmount *server_kernel_mount(struct super_block *sb)
1403 struct lvfs_run_ctxt mount_ctxt;
1404 struct lustre_sb_info *lsi = s2lsi(sb);
1405 struct lustre_disk_data *ldd;
1406 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1407 struct vfsmount *mnt;
1408 char *options = NULL;
1409 unsigned long page, s_flags;
1410 struct page *__page;
1415 OBD_ALLOC(ldd, sizeof(*ldd));
1417 RETURN(ERR_PTR(-ENOMEM));
1419 /* In the past, we have always used flags = 0.
1420 Note ext3/ldiskfs can't be mounted ro. */
1421 s_flags = sb->s_flags;
1423 /* allocate memory for options */
1424 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
1426 GOTO(out_free, rc = -ENOMEM);
1427 page = (unsigned long)cfs_page_address(__page);
1428 options = (char *)page;
1429 memset(options, 0, CFS_PAGE_SIZE);
1431 /* mount-line options must be added for pre-mount because it may
1432 * contain mount options such as journal_dev which are required
1433 * to mount successfuly the underlying filesystem */
1434 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0))
1435 strncat(options, lmd->lmd_opts, CFS_PAGE_SIZE - 1);
1437 /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */
1438 CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev);
1439 mnt = ll_kern_mount("ldiskfs", s_flags, lmd->lmd_dev, (void *)options);
1442 CERROR("premount %s:%#lx ldiskfs failed: %d "
1443 "Is the ldiskfs module available?\n",
1444 lmd->lmd_dev, s_flags, rc );
1448 OBD_SET_CTXT_MAGIC(&mount_ctxt);
1449 mount_ctxt.pwdmnt = mnt;
1450 mount_ctxt.pwd = mnt->mnt_root;
1451 mount_ctxt.fs = get_ds();
1453 rc = ldd_parse(&mount_ctxt, ldd);
1457 CERROR("premount parse options failed: rc = %d\n", rc);
1461 /* Done with our pre-mount, now do the real mount. */
1463 /* Glom up mount options */
1464 memset(options, 0, CFS_PAGE_SIZE);
1465 strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2);
1467 len = CFS_PAGE_SIZE - strlen(options) - 2;
1469 strcat(options, ",");
1470 strncat(options, "no_mbcache", len);
1472 /* Add in any mount-line options */
1473 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) {
1474 len = CFS_PAGE_SIZE - strlen(options) - 2;
1475 strcat(options, ",");
1476 strncat(options, lmd->lmd_opts, len);
1479 /* Special permanent mount flags */
1481 s_flags |= MS_NOATIME | MS_NODIRATIME;
1483 CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
1484 MT_STR(ldd), lmd->lmd_dev, options);
1485 mnt = ll_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev,
1489 CERROR("ll_kern_mount failed: rc = %d\n", rc);
1493 if (lmd->lmd_flags & LMD_FLG_ABORT_RECOV)
1494 simple_truncate(mnt->mnt_sb->s_root, mnt, LAST_RCVD,
1497 OBD_PAGE_FREE(__page);
1498 lsi->lsi_ldd = ldd; /* freed at lsi cleanup */
1499 CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
1504 OBD_PAGE_FREE(__page);
1505 OBD_FREE(ldd, sizeof(*ldd));
1506 lsi->lsi_ldd = NULL;
1507 RETURN(ERR_PTR(rc));
1510 /** Wait here forever until the mount refcount is 0 before completing umount,
1511 * else we risk dereferencing a null pointer.
1512 * LNET may take e.g. 165s before killing zombies.
1514 static void server_wait_finished(struct vfsmount *mnt)
1518 cfs_sigset_t blocked;
1520 cfs_waitq_init(&waitq);
1522 while (mnt_get_count(mnt) > 1) {
1523 if (waited && (waited % 30 == 0))
1524 LCONSOLE_WARN("Mount still busy with %d refs after "
1528 /* Cannot use l_event_wait() for an interruptible sleep. */
1530 blocked = cfs_block_sigsinv(sigmask(SIGKILL));
1531 cfs_waitq_wait_event_interruptible_timeout(
1533 (mnt_get_count(mnt) == 1),
1534 cfs_time_seconds(3),
1536 cfs_block_sigs(blocked);
1538 LCONSOLE_EMERG("Danger: interrupted umount %s with "
1539 "%d refs!\n", mnt->mnt_devname,
1540 mnt_get_count(mnt));
1547 /** Start the shutdown of servers at umount.
1549 static void server_put_super(struct super_block *sb)
1551 struct lustre_sb_info *lsi = s2lsi(sb);
1552 struct obd_device *obd;
1553 struct vfsmount *mnt = lsi->lsi_srv_mnt;
1554 char *tmpname, *extraname = NULL;
1556 int lddflags = lsi->lsi_ldd->ldd_flags;
1557 int lsiflags = lsi->lsi_flags;
1560 LASSERT(lsiflags & LSI_SERVER);
1562 tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1;
1563 OBD_ALLOC(tmpname, tmpname_sz);
1564 memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
1565 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1566 if (IS_MDT(lsi->lsi_ldd) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1567 snprintf(tmpname, tmpname_sz, "MGS");
1569 /* Stop the target */
1570 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1571 (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd))) {
1572 struct lustre_profile *lprof = NULL;
1574 /* tell the mgc to drop the config log */
1575 lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL);
1577 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1578 If there are any setup/cleanup errors, save the lov
1579 name for safety cleanup later. */
1580 lprof = class_get_profile(lsi->lsi_ldd->ldd_svname);
1581 if (lprof && lprof->lp_dt) {
1582 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1583 strcpy(extraname, lprof->lp_dt);
1586 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1588 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1589 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1591 /* We can't seem to give an error return code
1592 * to .put_super, so we better make sure we clean up! */
1594 class_manual_cleanup(obd);
1596 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
1597 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1601 /* If they wanted the mgs to stop separately from the mdt, they
1602 should have put it on a different device. */
1603 if (IS_MGS(lsi->lsi_ldd)) {
1604 /* if MDS start with --nomgs, don't stop MGS then */
1605 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1608 OBD_ALLOC(logname, MGS_PARAM_MAXLEN);
1610 LCONSOLE_WARN("Stopping mgs failed %d, please "
1611 "try again.", -ENOMEM);
1614 strcpy(logname, lsi->lsi_ldd->ldd_fsname);
1615 strcat(logname, "-params");
1616 /* tell the mgc to drop parameter config log */
1617 lustre_end_log(sb, logname, NULL);
1618 OBD_FREE(logname, MGS_PARAM_MAXLEN);
1620 server_stop_mgs(sb);
1624 /* Clean the mgc and sb */
1625 lustre_common_put_super(sb);
1627 /* Wait for the targets to really clean up - can't exit (and let the
1628 sb get destroyed) while the mount is still in use */
1629 server_wait_finished(mnt);
1631 /* drop the One True Mount */
1634 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1635 until the target is really gone so that our type refcount check
1637 server_stop_servers(lddflags, lsiflags);
1639 /* In case of startup or cleanup err, stop related obds */
1641 obd = class_name2obd(extraname);
1643 CWARN("Cleaning orphaned obd %s\n", extraname);
1645 class_manual_cleanup(obd);
1647 OBD_FREE(extraname, strlen(extraname) + 1);
1650 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1651 OBD_FREE(tmpname, tmpname_sz);
1655 /** Called only for 'umount -f'
1657 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1658 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1660 struct super_block *sb = vfsmnt->mnt_sb;
1662 static void server_umount_begin(struct super_block *sb)
1665 struct lustre_sb_info *lsi = s2lsi(sb);
1668 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1669 if (!(flags & MNT_FORCE)) {
1675 CDEBUG(D_MOUNT, "umount -f\n");
1676 /* umount = failover
1678 no third way to do non-force, non-failover */
1679 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1680 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1684 #ifndef HAVE_STATFS_DENTRY_PARAM
1685 static int server_statfs (struct super_block *sb, cfs_kstatfs_t *buf)
1688 static int server_statfs (struct dentry *dentry, cfs_kstatfs_t *buf)
1690 struct super_block *sb = dentry->d_sb;
1692 struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt;
1695 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) {
1696 #ifdef HAVE_STATFS_DENTRY_PARAM
1697 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf);
1699 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_sb, buf);
1702 buf->f_type = sb->s_magic;
1708 buf->f_type = sb->s_magic;
1709 buf->f_bsize = sb->s_blocksize;
1715 buf->f_namelen = NAME_MAX;
1719 /** The operations we support directly on the superblock:
1720 * mount, umount, and df.
1722 static struct super_operations server_ops =
1724 .put_super = server_put_super,
1725 .umount_begin = server_umount_begin, /* umount -f */
1726 .statfs = server_statfs,
1729 #define log2(n) cfs_ffz(~(n))
1730 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1732 static int server_fill_super_common(struct super_block *sb)
1734 struct inode *root = 0;
1737 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1739 sb->s_blocksize = 4096;
1740 sb->s_blocksize_bits = log2(sb->s_blocksize);
1741 sb->s_magic = LUSTRE_SUPER_MAGIC;
1742 sb->s_maxbytes = 0; //PAGE_CACHE_MAXBYTES;
1743 sb->s_flags |= MS_RDONLY;
1744 sb->s_op = &server_ops;
1746 root = new_inode(sb);
1748 CERROR("Can't make root inode\n");
1752 /* returns -EIO for every operation */
1753 /* make_bad_inode(root); -- badness - can't umount */
1754 /* apparently we need to be a directory for the mount to finish */
1755 root->i_mode = S_IFDIR;
1757 sb->s_root = d_alloc_root(root);
1759 CERROR("Can't make root dentry\n");
1767 /** Fill in the superblock info for a Lustre server.
1768 * Mount the device with the correct options.
1769 * Read the on-disk config file.
1770 * Start the services.
1772 static int server_fill_super(struct super_block *sb)
1774 struct lustre_sb_info *lsi = s2lsi(sb);
1775 struct vfsmount *mnt;
1779 /* the One True Mount */
1780 mnt = server_kernel_mount(sb);
1783 CERROR("Unable to mount device %s: %d\n",
1784 lsi->lsi_lmd->lmd_dev, rc);
1788 lsi->lsi_srv_mnt = mnt;
1790 LASSERT(lsi->lsi_ldd);
1791 CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n",
1792 lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname,
1793 lsi->lsi_lmd->lmd_dev);
1795 if (class_name2obd(lsi->lsi_ldd->ldd_svname)) {
1796 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1797 "running. Double-mount may have compromised"
1798 " the disk journal.\n",
1799 lsi->lsi_ldd->ldd_svname);
1805 /* Start MGS before MGC */
1806 if (IS_MGS(lsi->lsi_ldd) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
1807 rc = server_start_mgs(sb);
1812 /* Start MGC before servers */
1813 rc = lustre_start_mgc(sb);
1817 /* Set up all obd devices for service */
1818 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1819 (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) {
1820 rc = server_start_targets(sb, mnt);
1822 CERROR("Unable to start targets: %d\n", rc);
1825 /* FIXME overmount client here,
1826 or can we just start a client log and client_fill_super on this sb?
1827 We need to make sure server_put_super gets called too - ll_put_super
1828 calls lustre_common_put_super; check there for LSI_SERVER flag,
1830 Probably should start client from new thread so we can return.
1831 Client will not finish until all servers are connected.
1832 Note - MGS-only server does NOT get a client, since there is no
1833 lustre fs associated - the MGS is for all lustre fs's */
1834 } else if (IS_MGS(lsi->lsi_ldd) &&
1835 !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
1836 struct config_llog_instance cfg;
1839 OBD_ALLOC(logname, MGS_PARAM_MAXLEN);
1840 if (logname == NULL)
1841 GOTO(out_mnt, rc = -ENOMEM);
1842 strcpy(logname, lsi->lsi_ldd->ldd_fsname);
1843 strcat(logname, "-params");
1845 memset(&cfg, 0, sizeof(cfg));
1846 rc = lustre_process_log(sb, logname, &cfg);
1847 OBD_FREE(logname, MGS_PARAM_MAXLEN);
1849 CERROR("failed to process parameters %s: %d\n",
1855 rc = server_fill_super_common(sb);
1861 /* We jump here in case of failure while starting targets or MGS.
1862 * In this case we can't just put @mnt and have to do real cleanup
1863 * with stoping targets, etc. */
1864 server_put_super(sb);
1868 /* Get the index from the obd name.
1869 rc = server type, or
1871 if endptr isn't NULL it is set to end of name */
1872 int server_name2index(char *svname, __u32 *idx, char **endptr)
1874 unsigned long index;
1876 char *dash = strrchr(svname, '-');
1880 /* intepret <fsname>-MDTXXXXX-mdc as mdt, the better way is to pass
1881 * in the fsname, then determine the server index */
1882 if (!strcmp(LUSTRE_MDC_NAME, dash + 1)) {
1884 for (; dash > svname && *dash != '-'; dash--);
1889 if (strncmp(dash + 1, "MDT", 3) == 0)
1890 rc = LDD_F_SV_TYPE_MDT;
1891 else if (strncmp(dash + 1, "OST", 3) == 0)
1892 rc = LDD_F_SV_TYPE_OST;
1895 if (strcmp(dash + 4, "all") == 0)
1896 return rc | LDD_F_SV_ALL;
1898 index = simple_strtoul(dash + 4, endptr, 16);
1904 * Calculate timeout value for a target.
1906 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1908 struct lustre_mount_data *lmd;
1912 bool has_ir = !!(lsi->lsi_flags & LSI_IR_CAPABLE);
1913 int min = OBD_RECOVERY_TIME_MIN;
1915 LASSERT(lsi->lsi_flags & LSI_SERVER);
1919 soft = lmd->lmd_recovery_time_soft;
1920 hard = lmd->lmd_recovery_time_hard;
1921 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1922 obd->obd_no_ir = !has_ir;
1926 soft = OBD_RECOVERY_TIME_SOFT;
1928 hard = OBD_RECOVERY_TIME_HARD;
1930 /* target may have ir_factor configured. */
1931 factor = OBD_IR_FACTOR_DEFAULT;
1932 if (obd->obd_recovery_ir_factor)
1933 factor = obd->obd_recovery_ir_factor;
1936 int new_soft = soft;
1937 int new_hard = hard;
1939 /* adjust timeout value by imperative recovery */
1941 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1942 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1944 /* make sure the timeout is not too short */
1945 new_soft = max(min, new_soft);
1946 new_hard = max(new_soft, new_hard);
1948 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1949 "window shrunk from %d-%d down to %d-%d\n",
1950 obd->obd_name, soft, hard, new_soft, new_hard);
1957 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1958 obd->obd_recovery_time_hard = hard;
1959 obd->obd_recovery_ir_factor = factor;
1961 EXPORT_SYMBOL(server_calc_timeout);
1963 /*************** mount common betweeen server and client ***************/
1966 int lustre_common_put_super(struct super_block *sb)
1971 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1973 /* Drop a ref to the MGC */
1974 rc = lustre_stop_mgc(sb);
1975 if (rc && (rc != -ENOENT)) {
1977 CERROR("Can't stop MGC: %d\n", rc);
1980 /* BUSY just means that there's some other obd that
1981 needs the mgc. Let him clean it up. */
1982 CDEBUG(D_MOUNT, "MGC still in use\n");
1984 /* Drop a ref to the mounted disk */
1990 static void lmd_print(struct lustre_mount_data *lmd)
1994 PRINT_CMD(PRINT_MASK, " mount data:\n");
1995 if (lmd_is_client(lmd))
1996 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
1997 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
1998 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
2001 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
2003 if (lmd->lmd_recovery_time_soft)
2004 PRINT_CMD(PRINT_MASK, "recovery time soft: %d\n",
2005 lmd->lmd_recovery_time_soft);
2007 if (lmd->lmd_recovery_time_hard)
2008 PRINT_CMD(PRINT_MASK, "recovery time hard: %d\n",
2009 lmd->lmd_recovery_time_hard);
2011 for (i = 0; i < lmd->lmd_exclude_count; i++) {
2012 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
2013 lmd->lmd_exclude[i]);
2017 /* Is this server on the exclusion list */
2018 int lustre_check_exclusion(struct super_block *sb, char *svname)
2020 struct lustre_sb_info *lsi = s2lsi(sb);
2021 struct lustre_mount_data *lmd = lsi->lsi_lmd;
2026 rc = server_name2index(svname, &index, NULL);
2027 if (rc != LDD_F_SV_TYPE_OST)
2028 /* Only exclude OSTs */
2031 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
2032 index, lmd->lmd_exclude_count, lmd->lmd_dev);
2034 for(i = 0; i < lmd->lmd_exclude_count; i++) {
2035 if (index == lmd->lmd_exclude[i]) {
2036 CWARN("Excluding %s (on exclusion list)\n", svname);
2043 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
2044 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
2046 char *s1 = ptr, *s2;
2047 __u32 index, *exclude_list;
2051 /* The shortest an ost name can be is 8 chars: -OST0000.
2052 We don't actually know the fsname at this time, so in fact
2053 a user could specify any fsname. */
2054 devmax = strlen(ptr) / 8 + 1;
2056 /* temp storage until we figure out how many we have */
2057 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
2061 /* we enter this fn pointing at the '=' */
2062 while (*s1 && *s1 != ' ' && *s1 != ',') {
2064 rc = server_name2index(s1, &index, &s2);
2066 CERROR("Can't parse server name '%s'\n", s1);
2069 if (rc == LDD_F_SV_TYPE_OST)
2070 exclude_list[lmd->lmd_exclude_count++] = index;
2072 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
2074 /* now we are pointing at ':' (next exclude)
2075 or ',' (end of excludes) */
2076 if (lmd->lmd_exclude_count >= devmax)
2079 if (rc >= 0) /* non-err */
2082 if (lmd->lmd_exclude_count) {
2083 /* permanent, freed in lustre_free_lsi */
2084 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
2085 lmd->lmd_exclude_count);
2086 if (lmd->lmd_exclude) {
2087 memcpy(lmd->lmd_exclude, exclude_list,
2088 sizeof(index) * lmd->lmd_exclude_count);
2091 lmd->lmd_exclude_count = 0;
2094 OBD_FREE(exclude_list, sizeof(index) * devmax);
2098 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
2103 if (lmd->lmd_mgssec != NULL) {
2104 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
2105 lmd->lmd_mgssec = NULL;
2108 tail = strchr(ptr, ',');
2110 length = strlen(ptr);
2112 length = tail - ptr;
2114 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
2115 if (lmd->lmd_mgssec == NULL)
2118 memcpy(lmd->lmd_mgssec, ptr, length);
2119 lmd->lmd_mgssec[length] = '\0';
2123 /** Parse mount line options
2124 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
2125 * dev is passed as device=uml1:/lustre by mount.lustre
2127 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
2129 char *s1, *s2, *devname = NULL;
2130 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
2136 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
2137 "/sbin/mount.lustre is installed.\n");
2141 /* Options should be a string - try to detect old lmd data */
2142 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
2143 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
2144 "/sbin/mount.lustre. Please install "
2145 "version %s\n", LUSTRE_VERSION_STRING);
2148 lmd->lmd_magic = LMD_MAGIC;
2150 /* Set default flags here */
2155 int time_min = OBD_RECOVERY_TIME_MIN;
2157 /* Skip whitespace and extra commas */
2158 while (*s1 == ' ' || *s1 == ',')
2161 /* Client options are parsed in ll_options: eg. flock,
2164 /* Parse non-ldiskfs options here. Rather than modifying
2165 ldiskfs, we just zero these out here */
2166 if (strncmp(s1, "abort_recov", 11) == 0) {
2167 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
2169 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
2170 lmd->lmd_recovery_time_soft = max_t(int,
2171 simple_strtoul(s1 + 19, NULL, 10), time_min);
2173 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
2174 lmd->lmd_recovery_time_hard = max_t(int,
2175 simple_strtoul(s1 + 19, NULL, 10), time_min);
2177 } else if (strncmp(s1, "noir", 4) == 0) {
2178 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
2180 } else if (strncmp(s1, "nosvc", 5) == 0) {
2181 lmd->lmd_flags |= LMD_FLG_NOSVC;
2183 } else if (strncmp(s1, "nomgs", 5) == 0) {
2184 lmd->lmd_flags |= LMD_FLG_NOMGS;
2186 } else if (strncmp(s1, "writeconf", 9) == 0) {
2187 lmd->lmd_flags |= LMD_FLG_WRITECONF;
2189 } else if (strncmp(s1, "mgssec=", 7) == 0) {
2190 rc = lmd_parse_mgssec(lmd, s1 + 7);
2194 /* ost exclusion list */
2195 } else if (strncmp(s1, "exclude=", 8) == 0) {
2196 rc = lmd_make_exclusion(lmd, s1 + 7);
2201 /* Linux 2.4 doesn't pass the device, so we stuck it at the
2202 end of the options. */
2203 else if (strncmp(s1, "device=", 7) == 0) {
2205 /* terminate options right before device. device
2206 must be the last one. */
2212 s2 = strchr(s1, ',');
2220 memmove(s1, s2, strlen(s2) + 1);
2226 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
2227 "(need mount option 'device=...')\n");
2231 s1 = strstr(devname, ":/");
2234 lmd->lmd_flags |= LMD_FLG_CLIENT;
2235 /* Remove leading /s from fsname */
2236 while (*++s1 == '/') ;
2237 /* Freed in lustre_free_lsi */
2238 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
2239 if (!lmd->lmd_profile)
2241 sprintf(lmd->lmd_profile, "%s-client", s1);
2244 /* Freed in lustre_free_lsi */
2245 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
2248 strcpy(lmd->lmd_dev, devname);
2250 /* Save mount options */
2251 s1 = options + strlen(options) - 1;
2252 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
2254 if (*options != 0) {
2255 /* Freed in lustre_free_lsi */
2256 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
2259 strcpy(lmd->lmd_opts, options);
2263 lmd->lmd_magic = LMD_MAGIC;
2268 CERROR("Bad mount options %s\n", options);
2272 struct lustre_mount_data2 {
2274 struct vfsmount *lmd2_mnt;
2277 /** This is the entry point for the mount call into Lustre.
2278 * This is called when a server or client is mounted,
2279 * and this is where we start setting things up.
2280 * @param data Mount options (e.g. -o flock,abort_recov)
2282 int lustre_fill_super(struct super_block *sb, void *data, int silent)
2284 struct lustre_mount_data *lmd;
2285 struct lustre_mount_data2 *lmd2 = data;
2286 struct lustre_sb_info *lsi;
2290 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
2292 lsi = lustre_init_lsi(sb);
2298 * Disable lockdep during mount, because mount locking patterns are
2304 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
2306 obd_zombie_barrier();
2308 /* Figure out the lmd from the mount options */
2309 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
2311 GOTO(out, rc = -EINVAL);
2314 if (lmd_is_client(lmd)) {
2315 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
2316 if (!client_fill_super) {
2317 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
2318 "client mount! Is the 'lustre' "
2319 "module loaded?\n");
2323 rc = lustre_start_mgc(sb);
2328 /* Connect and start */
2329 /* (should always be ll_fill_super) */
2330 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
2331 /* c_f_s will call lustre_common_put_super on failure */
2334 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
2335 lsi->lsi_flags |= LSI_SERVER;
2336 rc = server_fill_super(sb);
2337 /* s_f_s calls lustre_start_mgc after the mount because we need
2338 the MGS nids which are stored on disk. Plus, we may
2339 need to start the MGS first. */
2340 /* s_f_s will call server_put_super on failure */
2343 /* If error happens in fill_super() call, @lsi will be killed there.
2344 * This is why we do not put it here. */
2348 CERROR("Unable to mount %s (%d)\n",
2349 s2lsi(sb) ? lmd->lmd_dev : "", rc);
2351 CDEBUG(D_SUPER, "Mount %s complete\n",
2359 /* We can't call ll_fill_super by name because it lives in a module that
2360 must be loaded after this one. */
2361 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
2362 struct vfsmount *mnt))
2364 client_fill_super = cfs;
2367 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
2369 kill_super_cb = cfs;
2372 /***************** FS registration ******************/
2374 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))
2375 struct super_block * lustre_get_sb(struct file_system_type *fs_type, int flags,
2376 const char *devname, void * data)
2378 return get_sb_nodev(fs_type, flags, data, lustre_fill_super);
2381 int lustre_get_sb(struct file_system_type *fs_type, int flags,
2382 const char *devname, void * data, struct vfsmount *mnt)
2384 struct lustre_mount_data2 lmd2 = {data, mnt};
2386 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
2390 void lustre_kill_super(struct super_block *sb)
2392 struct lustre_sb_info *lsi = s2lsi(sb);
2394 if (kill_super_cb && lsi && !(lsi->lsi_flags & LSI_SERVER))
2395 (*kill_super_cb)(sb);
2397 kill_anon_super(sb);
2400 /** Register the "lustre" fs type
2402 struct file_system_type lustre_fs_type = {
2403 .owner = THIS_MODULE,
2405 .get_sb = lustre_get_sb,
2406 .kill_sb = lustre_kill_super,
2407 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
2408 #ifdef FS_HAS_FIEMAP
2411 LL_RENAME_DOES_D_MOVE,
2414 int lustre_register_fs(void)
2416 return register_filesystem(&lustre_fs_type);
2419 int lustre_unregister_fs(void)
2421 return unregister_filesystem(&lustre_fs_type);
2424 EXPORT_SYMBOL(lustre_register_client_fill_super);
2425 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2426 EXPORT_SYMBOL(lustre_common_put_super);
2427 EXPORT_SYMBOL(lustre_process_log);
2428 EXPORT_SYMBOL(lustre_end_log);
2429 EXPORT_SYMBOL(server_get_mount);
2430 EXPORT_SYMBOL(server_get_mount_2);
2431 EXPORT_SYMBOL(server_put_mount);
2432 EXPORT_SYMBOL(server_put_mount_2);
2433 EXPORT_SYMBOL(server_register_target);
2434 EXPORT_SYMBOL(server_name2index);
2435 EXPORT_SYMBOL(server_mti_print);
2436 EXPORT_SYMBOL(do_lcfg);