4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/obdclass/obd_mount.c
38 * Client/server mount routines
40 * Author: Nathan Rutman <nathan@clusterfs.com>
44 #define DEBUG_SUBSYSTEM S_CLASS
45 #define D_MOUNT D_SUPER|D_CONFIG /*|D_WARNING */
46 #define PRINT_CMD CDEBUG
47 #define PRINT_MASK D_SUPER|D_CONFIG
51 #include <lustre_fsfilt.h>
52 #include <obd_class.h>
53 #include <lustre/lustre_user.h>
54 #include <linux/version.h>
55 #include <lustre_log.h>
56 #include <lustre_disk.h>
57 #include <lustre_param.h>
59 static int (*client_fill_super)(struct super_block *sb,
60 struct vfsmount *mnt) = NULL;
61 static void (*kill_super_cb)(struct super_block *sb) = NULL;
63 /*********** mount lookup *********/
65 CFS_DEFINE_MUTEX(lustre_mount_info_lock);
66 static CFS_LIST_HEAD(server_mount_info_list);
68 static struct lustre_mount_info *server_find_mount(const char *name)
71 struct lustre_mount_info *lmi;
74 cfs_list_for_each(tmp, &server_mount_info_list) {
75 lmi = cfs_list_entry(tmp, struct lustre_mount_info,
77 if (strcmp(name, lmi->lmi_name) == 0)
83 /* we must register an obd for a mount before we call the setup routine.
84 *_setup will call lustre_get_mount to get the mnt struct
85 by obd_name, since we can't pass the pointer to setup. */
86 static int server_register_mount(const char *name, struct super_block *sb,
89 struct lustre_mount_info *lmi;
96 OBD_ALLOC(lmi, sizeof(*lmi));
99 OBD_ALLOC(name_cp, strlen(name) + 1);
101 OBD_FREE(lmi, sizeof(*lmi));
104 strcpy(name_cp, name);
106 cfs_mutex_lock(&lustre_mount_info_lock);
108 if (server_find_mount(name)) {
109 cfs_mutex_unlock(&lustre_mount_info_lock);
110 OBD_FREE(lmi, sizeof(*lmi));
111 OBD_FREE(name_cp, strlen(name) + 1);
112 CERROR("Already registered %s\n", name);
115 lmi->lmi_name = name_cp;
118 cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list);
120 cfs_mutex_unlock(&lustre_mount_info_lock);
122 CDEBUG(D_MOUNT, "reg_mnt %p from %s, vfscount=%d\n",
123 lmi->lmi_mnt, name, mnt_get_count(lmi->lmi_mnt));
128 /* when an obd no longer needs a mount */
129 static int server_deregister_mount(const char *name)
131 struct lustre_mount_info *lmi;
134 cfs_mutex_lock(&lustre_mount_info_lock);
135 lmi = server_find_mount(name);
137 cfs_mutex_unlock(&lustre_mount_info_lock);
138 CERROR("%s not registered\n", name);
142 CDEBUG(D_MOUNT, "dereg_mnt %p from %s, vfscount=%d\n",
143 lmi->lmi_mnt, name, mnt_get_count(lmi->lmi_mnt));
145 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
146 cfs_list_del(&lmi->lmi_list_chain);
147 OBD_FREE(lmi, sizeof(*lmi));
148 cfs_mutex_unlock(&lustre_mount_info_lock);
153 /* obd's look up a registered mount using their obdname. This is just
154 for initial obd setup to find the mount struct. It should not be
155 called every time you want to mntget. */
156 struct lustre_mount_info *server_get_mount(const char *name)
158 struct lustre_mount_info *lmi;
159 struct lustre_sb_info *lsi;
162 cfs_mutex_lock(&lustre_mount_info_lock);
163 lmi = server_find_mount(name);
164 cfs_mutex_unlock(&lustre_mount_info_lock);
166 CERROR("Can't find mount for %s\n", name);
169 lsi = s2lsi(lmi->lmi_sb);
170 mntget(lmi->lmi_mnt);
171 cfs_atomic_inc(&lsi->lsi_mounts);
173 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d, vfscount=%d\n",
174 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts),
175 mnt_get_count(lmi->lmi_mnt));
181 * Used by mdt to get mount_info from obdname.
182 * There are no blocking when using the mount_info.
183 * Do not use server_get_mount for this purpose.
185 struct lustre_mount_info *server_get_mount_2(const char *name)
187 struct lustre_mount_info *lmi;
190 cfs_mutex_lock(&lustre_mount_info_lock);
191 lmi = server_find_mount(name);
192 cfs_mutex_unlock(&lustre_mount_info_lock);
194 CERROR("Can't find mount for %s\n", name);
199 static void unlock_mntput(struct vfsmount *mnt)
201 if (kernel_locked()) {
210 static int lustre_put_lsi(struct super_block *sb);
212 /* to be called from obd_cleanup methods */
213 int server_put_mount(const char *name, struct vfsmount *mnt)
215 struct lustre_mount_info *lmi;
216 struct lustre_sb_info *lsi;
217 int count = mnt_get_count(mnt) - 1;
220 /* This might be the last one, can't deref after this */
223 cfs_mutex_lock(&lustre_mount_info_lock);
224 lmi = server_find_mount(name);
225 cfs_mutex_unlock(&lustre_mount_info_lock);
227 CERROR("Can't find mount for %s\n", name);
230 lsi = s2lsi(lmi->lmi_sb);
231 LASSERT(lmi->lmi_mnt == mnt);
233 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d, vfscount=%d\n",
234 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts), count);
236 if (lustre_put_lsi(lmi->lmi_sb)) {
237 CDEBUG(D_MOUNT, "Last put of mnt %p from %s, vfscount=%d\n",
238 lmi->lmi_mnt, name, count);
239 /* last mount is the One True Mount */
241 CERROR("%s: mount busy, vfscount=%d!\n", name, count);
244 /* this obd should never need the mount again */
245 server_deregister_mount(name);
250 /* Corresponding to server_get_mount_2 */
251 int server_put_mount_2(const char *name, struct vfsmount *mnt)
257 /******* mount helper utilities *********/
260 static void ldd_print(struct lustre_disk_data *ldd)
262 PRINT_CMD(PRINT_MASK, " disk data:\n");
263 PRINT_CMD(PRINT_MASK, "server: %s\n", ldd->ldd_svname);
264 PRINT_CMD(PRINT_MASK, "uuid: %s\n", (char *)ldd->ldd_uuid);
265 PRINT_CMD(PRINT_MASK, "fs: %s\n", ldd->ldd_fsname);
266 PRINT_CMD(PRINT_MASK, "index: %04x\n", ldd->ldd_svindex);
267 PRINT_CMD(PRINT_MASK, "config: %d\n", ldd->ldd_config_ver);
268 PRINT_CMD(PRINT_MASK, "flags: %#x\n", ldd->ldd_flags);
269 PRINT_CMD(PRINT_MASK, "diskfs: %s\n", MT_STR(ldd));
270 PRINT_CMD(PRINT_MASK, "options: %s\n", ldd->ldd_mount_opts);
271 PRINT_CMD(PRINT_MASK, "params: %s\n", ldd->ldd_params);
272 PRINT_CMD(PRINT_MASK, "comment: %s\n", ldd->ldd_userdata);
276 static int ldd_parse(struct lvfs_run_ctxt *mount_ctxt,
277 struct lustre_disk_data *ldd)
279 struct lvfs_run_ctxt saved;
286 push_ctxt(&saved, mount_ctxt, NULL);
288 file = filp_open(MOUNT_DATA_FILE, O_RDONLY, 0644);
291 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
295 len = i_size_read(file->f_dentry->d_inode);
296 CDEBUG(D_MOUNT, "Have %s, size %lu\n", MOUNT_DATA_FILE, len);
297 if (len != sizeof(*ldd)) {
298 CERROR("disk data size does not match: see %lu expect %u\n",
299 len, (int)sizeof(*ldd));
300 GOTO(out_close, rc = -EINVAL);
303 rc = lustre_fread(file, ldd, len, &off);
305 CERROR("error reading %s: read %d of %lu\n",
306 MOUNT_DATA_FILE, rc, len);
307 GOTO(out_close, rc = -EINVAL);
311 if (ldd->ldd_magic != LDD_MAGIC) {
312 /* FIXME add swabbing support */
313 CERROR("Bad magic in %s: %x!=%x\n", MOUNT_DATA_FILE,
314 ldd->ldd_magic, LDD_MAGIC);
315 GOTO(out_close, rc = -EINVAL);
318 if (ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP) {
319 CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
321 ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP);
322 GOTO(out_close, rc = -EINVAL);
324 if (ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP) {
325 CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
327 ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP);
328 /* Do something like remount filesystem read-only */
329 GOTO(out_close, rc = -EINVAL);
335 pop_ctxt(&saved, mount_ctxt, NULL);
339 static int ldd_write(struct lvfs_run_ctxt *mount_ctxt,
340 struct lustre_disk_data *ldd)
342 struct lvfs_run_ctxt saved;
345 unsigned long len = sizeof(struct lustre_disk_data);
349 LASSERT(ldd->ldd_magic == LDD_MAGIC);
351 ldd->ldd_config_ver++;
353 push_ctxt(&saved, mount_ctxt, NULL);
355 file = filp_open(MOUNT_DATA_FILE, O_RDWR|O_SYNC, 0644);
358 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
362 rc = lustre_fwrite(file, ldd, len, &off);
364 CERROR("error writing %s: read %d of %lu\n",
365 MOUNT_DATA_FILE, rc, len);
366 GOTO(out_close, rc = -EINVAL);
374 pop_ctxt(&saved, mount_ctxt, NULL);
379 /**************** config llog ********************/
381 /** Get a config log from the MGS and process it.
382 * This func is called for both clients and servers.
383 * Continue to process new statements appended to the logs
384 * (whenever the config lock is revoked) until lustre_end_log
386 * @param sb The superblock is used by the MGC to write to the local copy of
388 * @param logname The name of the llog to replicate from the MGS
389 * @param cfg Since the same mgc may be used to follow multiple config logs
390 * (e.g. ost1, ost2, client), the config_llog_instance keeps the state for
391 * this log, and is added to the mgc's list of logs to follow.
393 int lustre_process_log(struct super_block *sb, char *logname,
394 struct config_llog_instance *cfg)
396 struct lustre_cfg *lcfg;
397 struct lustre_cfg_bufs *bufs;
398 struct lustre_sb_info *lsi = s2lsi(sb);
399 struct obd_device *mgc = lsi->lsi_mgc;
410 /* mgc_process_config */
411 lustre_cfg_bufs_reset(bufs, mgc->obd_name);
412 lustre_cfg_bufs_set_string(bufs, 1, logname);
413 lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg));
414 lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb));
415 lcfg = lustre_cfg_new(LCFG_LOG_START, bufs);
416 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
417 lustre_cfg_free(lcfg);
422 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
423 "failed from the MGS (%d). Make sure this "
424 "client and the MGS are running compatible "
425 "versions of Lustre.\n",
426 mgc->obd_name, logname, rc);
429 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
430 "failed (%d). This may be the result of "
431 "communication errors between this node and "
432 "the MGS, a bad configuration, or other "
433 "errors. See the syslog for more "
434 "information.\n", mgc->obd_name, logname,
437 /* class_obd_list(); */
441 /* Stop watching this config log for updates */
442 int lustre_end_log(struct super_block *sb, char *logname,
443 struct config_llog_instance *cfg)
445 struct lustre_cfg *lcfg;
446 struct lustre_cfg_bufs bufs;
447 struct lustre_sb_info *lsi = s2lsi(sb);
448 struct obd_device *mgc = lsi->lsi_mgc;
455 /* mgc_process_config */
456 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
457 lustre_cfg_bufs_set_string(&bufs, 1, logname);
459 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
460 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
461 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
462 lustre_cfg_free(lcfg);
466 /**************** obd start *******************/
468 /** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from
469 * lctl (and do for echo cli/srv.
471 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
472 char *s1, char *s2, char *s3, char *s4)
474 struct lustre_cfg_bufs bufs;
475 struct lustre_cfg * lcfg = NULL;
478 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
479 cmd, s1, s2, s3, s4);
481 lustre_cfg_bufs_reset(&bufs, cfgname);
483 lustre_cfg_bufs_set_string(&bufs, 1, s1);
485 lustre_cfg_bufs_set_string(&bufs, 2, s2);
487 lustre_cfg_bufs_set_string(&bufs, 3, s3);
489 lustre_cfg_bufs_set_string(&bufs, 4, s4);
491 lcfg = lustre_cfg_new(cmd, &bufs);
492 lcfg->lcfg_nid = nid;
493 rc = class_process_config(lcfg);
494 lustre_cfg_free(lcfg);
498 /** Call class_attach and class_setup. These methods in turn call
499 * obd type-specific methods.
501 static int lustre_start_simple(char *obdname, char *type, char *uuid,
505 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
507 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
509 CERROR("%s attach error %d\n", obdname, rc);
512 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, 0, 0);
514 CERROR("%s setup error %d\n", obdname, rc);
515 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
520 /* Set up a MGS to serve startup logs */
521 static int server_start_mgs(struct super_block *sb)
523 struct lustre_sb_info *lsi = s2lsi(sb);
524 struct vfsmount *mnt = lsi->lsi_srv_mnt;
525 struct lustre_mount_info *lmi;
530 /* It is impossible to have more than 1 MGS per node, since
531 MGC wouldn't know which to connect to */
532 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
534 lsi = s2lsi(lmi->lmi_sb);
535 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
537 lsi->lsi_ldd->ldd_svname);
541 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
543 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
546 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
547 LUSTRE_MGS_OBDNAME, 0, 0);
548 /* Do NOT call server_deregister_mount() here. This leads to
549 * inability cleanup cleanly and free lsi and other stuff when
550 * mgs calls server_put_mount() in error handling case. -umka */
554 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
555 "Is the 'mgs' module loaded?\n",
556 LUSTRE_MGS_OBDNAME, rc);
560 static int server_stop_mgs(struct super_block *sb)
562 struct obd_device *obd;
566 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
568 /* There better be only one MGS */
569 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
571 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
575 /* The MGS should always stop when we say so */
577 rc = class_manual_cleanup(obd);
581 CFS_DEFINE_MUTEX(mgc_start_lock);
583 /** Set up a mgc obd to process startup logs
585 * \param sb [in] super block of the mgc obd
587 * \retval 0 success, otherwise error code
589 static int lustre_start_mgc(struct super_block *sb)
591 struct obd_connect_data *data = NULL;
592 struct lustre_sb_info *lsi = s2lsi(sb);
593 struct obd_device *obd;
594 struct obd_export *exp;
595 struct obd_uuid *uuid;
598 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
601 int rc = 0, i = 0, j, len;
604 LASSERT(lsi->lsi_lmd);
606 /* Find the first non-lo MGS nid for our MGC name */
607 if (lsi->lsi_flags & LSI_SERVER) {
608 ptr = lsi->lsi_ldd->ldd_params;
609 /* Use mgsnode= nids */
610 if ((class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0) &&
611 (class_parse_nid(ptr, &nid, &ptr) == 0)) {
613 } else if (IS_MGS(lsi->lsi_ldd)) {
614 lnet_process_id_t id;
615 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
616 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
623 } else { /* client */
624 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
625 ptr = lsi->lsi_lmd->lmd_dev;
626 if (class_parse_nid(ptr, &nid, &ptr) == 0)
630 CERROR("No valid MGS nids found.\n");
634 cfs_mutex_lock(&mgc_start_lock);
636 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
637 OBD_ALLOC(mgcname, len);
638 OBD_ALLOC(niduuid, len + 2);
639 if (!mgcname || !niduuid)
640 GOTO(out_free, rc = -ENOMEM);
641 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
643 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
647 GOTO(out_free, rc = -ENOMEM);
649 obd = class_name2obd(mgcname);
650 if (obd && !obd->obd_stopping) {
651 rc = obd_set_info_async(NULL, obd->obd_self_export,
652 strlen(KEY_MGSSEC), KEY_MGSSEC,
653 strlen(mgssec), mgssec, NULL);
657 /* Re-using an existing MGC */
658 cfs_atomic_inc(&obd->u.cli.cl_mgc_refcount);
660 /* IR compatibility check, only for clients */
661 if (lmd_is_client(lsi->lsi_lmd)) {
663 int vallen = sizeof(*data);
664 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
666 rc = obd_get_info(NULL, obd->obd_self_export,
667 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
668 &vallen, data, NULL);
670 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
671 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
672 /* LMD_FLG_NOIR is for test purpose only */
674 "Trying to mount a client with IR setting "
675 "not compatible with current mgc. "
676 "Force to use current mgc setting that is "
678 has_ir ? "enabled" : "disabled");
680 *flags &= ~LMD_FLG_NOIR;
682 *flags |= LMD_FLG_NOIR;
687 /* If we are restarting the MGS, don't try to keep the MGC's
688 old connection, or registration will fail. */
689 if ((lsi->lsi_flags & LSI_SERVER) && IS_MGS(lsi->lsi_ldd)) {
690 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
694 /* Try all connections, but only once (again).
695 We don't want to block another target from starting
696 (using its local copy of the log), but we do want to connect
697 if at all possible. */
699 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
700 rc = obd_set_info_async(NULL, obd->obd_self_export,
701 sizeof(KEY_INIT_RECOV_BACKUP),
702 KEY_INIT_RECOV_BACKUP,
703 sizeof(recov_bk), &recov_bk, NULL);
707 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
709 /* Add the primary nids for the MGS */
711 sprintf(niduuid, "%s_%x", mgcname, i);
712 if (lsi->lsi_flags & LSI_SERVER) {
713 ptr = lsi->lsi_ldd->ldd_params;
714 if (IS_MGS(lsi->lsi_ldd)) {
715 /* Use local nids (including LO) */
716 lnet_process_id_t id;
717 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
718 rc = do_lcfg(mgcname, id.nid,
719 LCFG_ADD_UUID, niduuid, 0,0,0);
722 /* Use mgsnode= nids */
723 if (class_find_param(ptr, PARAM_MGSNODE, &ptr) != 0) {
724 CERROR("No MGS nids given.\n");
725 GOTO(out_free, rc = -EINVAL);
727 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
728 rc = do_lcfg(mgcname, nid,
729 LCFG_ADD_UUID, niduuid, 0,0,0);
733 } else { /* client */
734 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
735 ptr = lsi->lsi_lmd->lmd_dev;
736 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
737 rc = do_lcfg(mgcname, nid,
738 LCFG_ADD_UUID, niduuid, 0,0,0);
740 /* Stop at the first failover nid */
746 CERROR("No valid MGS nids found.\n");
747 GOTO(out_free, rc = -EINVAL);
749 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
751 /* Random uuid for MGC allows easier reconnects */
753 ll_generate_random_uuid(uuidc);
754 class_uuid_unparse(uuidc, uuid);
757 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
758 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
764 /* Add any failover MGS nids */
766 while ((*ptr == ':' ||
767 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0)) {
768 /* New failover node */
769 sprintf(niduuid, "%s_%x", mgcname, i);
771 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
773 rc = do_lcfg(mgcname, nid,
774 LCFG_ADD_UUID, niduuid, 0,0,0);
779 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
787 lsi->lsi_lmd->lmd_mgs_failnodes = i;
789 obd = class_name2obd(mgcname);
791 CERROR("Can't find mgcobd %s\n", mgcname);
792 GOTO(out_free, rc = -ENOTCONN);
795 rc = obd_set_info_async(NULL, obd->obd_self_export,
796 strlen(KEY_MGSSEC), KEY_MGSSEC,
797 strlen(mgssec), mgssec, NULL);
801 /* Keep a refcount of servers/clients who started with "mount",
802 so we know when we can get rid of the mgc. */
803 cfs_atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
805 /* Try all connections, but only once. */
807 rc = obd_set_info_async(NULL, obd->obd_self_export,
808 sizeof(KEY_INIT_RECOV_BACKUP),
809 KEY_INIT_RECOV_BACKUP,
810 sizeof(recov_bk), &recov_bk, NULL);
813 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
814 /* We connect to the MGS at setup, and don't disconnect until cleanup */
815 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID |
816 OBD_CONNECT_AT | OBD_CONNECT_FULL20 |
817 OBD_CONNECT_IMP_RECOV;
818 if (lmd_is_client(lsi->lsi_lmd) &&
819 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
820 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
821 data->ocd_version = LUSTRE_VERSION_CODE;
822 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
824 CERROR("connect failed %d\n", rc);
828 obd->u.cli.cl_mgc_mgsexp = exp;
831 /* Keep the mgc info in the sb. Note that many lsi's can point
835 cfs_mutex_unlock(&mgc_start_lock);
840 OBD_FREE(mgcname, len);
842 OBD_FREE(niduuid, len + 2);
846 static int lustre_stop_mgc(struct super_block *sb)
848 struct lustre_sb_info *lsi = s2lsi(sb);
849 struct obd_device *obd;
850 char *niduuid = 0, *ptr = 0;
851 int i, rc = 0, len = 0;
861 cfs_mutex_lock(&mgc_start_lock);
862 LASSERT(cfs_atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
863 if (!cfs_atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
864 /* This is not fatal, every client that stops
865 will call in here. */
866 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
867 cfs_atomic_read(&obd->u.cli.cl_mgc_refcount));
868 GOTO(out, rc = -EBUSY);
871 /* The MGC has no recoverable data in any case.
872 * force shotdown set in umount_begin */
873 obd->obd_no_recov = 1;
875 if (obd->u.cli.cl_mgc_mgsexp) {
876 /* An error is not fatal, if we are unable to send the
877 disconnect mgs ping evictor cleans up the export */
878 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
880 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
883 /* Save the obdname for cleaning the nid uuids, which are
885 len = strlen(obd->obd_name) + 6;
886 OBD_ALLOC(niduuid, len);
888 strcpy(niduuid, obd->obd_name);
889 ptr = niduuid + strlen(niduuid);
892 rc = class_manual_cleanup(obd);
896 /* Clean the nid uuids */
898 GOTO(out, rc = -ENOMEM);
900 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
901 sprintf(ptr, "_%x", i);
902 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
905 CERROR("del MDC UUID %s failed: rc = %d\n",
910 OBD_FREE(niduuid, len);
912 /* class_import_put will get rid of the additional connections */
913 cfs_mutex_unlock(&mgc_start_lock);
917 /* Since there's only one mgc per node, we have to change it's fs to get
918 access to the right disk. */
919 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
921 struct lustre_sb_info *lsi = s2lsi(sb);
925 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
927 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
928 rc = obd_set_info_async(NULL, mgc->obd_self_export,
929 sizeof(KEY_SET_FS), KEY_SET_FS,
930 sizeof(*sb), sb, NULL);
932 CERROR("can't set_fs %d\n", rc);
938 static int server_mgc_clear_fs(struct obd_device *mgc)
943 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
945 rc = obd_set_info_async(NULL, mgc->obd_self_export,
946 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
951 CFS_DEFINE_MUTEX(server_start_lock);
953 /* Stop MDS/OSS if nobody is using them */
954 static int server_stop_servers(int lddflags, int lsiflags)
956 struct obd_device *obd = NULL;
957 struct obd_type *type = NULL;
961 cfs_mutex_lock(&server_start_lock);
963 /* Either an MDT or an OST or neither */
964 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
965 if ((lddflags & LDD_F_SV_TYPE_MDT) &&
966 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
967 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
968 type = class_search_type(LUSTRE_MDS_NAME);
970 /* if this was an OST, and there are no more OST's, clean up the OSS */
971 if ((lddflags & LDD_F_SV_TYPE_OST) &&
972 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
973 type = class_search_type(LUSTRE_OST_NAME);
976 if (obd && (!type || !type->typ_refcnt)) {
979 /* obd_fail doesn't mean much on a server obd */
980 err = class_manual_cleanup(obd);
985 cfs_mutex_unlock(&server_start_lock);
990 int server_mti_print(char *title, struct mgs_target_info *mti)
992 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
993 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
994 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
995 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
996 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
997 mti->mti_config_ver, mti->mti_flags);
1001 static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
1003 struct lustre_sb_info *lsi = s2lsi(sb);
1004 struct lustre_disk_data *ldd = lsi->lsi_ldd;
1005 lnet_process_id_t id;
1009 if (!(lsi->lsi_flags & LSI_SERVER))
1012 strncpy(mti->mti_fsname, ldd->ldd_fsname,
1013 sizeof(mti->mti_fsname));
1014 strncpy(mti->mti_svname, ldd->ldd_svname,
1015 sizeof(mti->mti_svname));
1017 mti->mti_nid_count = 0;
1018 while (LNetGetId(i++, &id) != -ENOENT) {
1019 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1022 /* server use --servicenode param, only allow specified
1023 * nids be registered */
1024 if ((ldd->ldd_flags & LDD_F_NO_PRIMNODE) != 0 &&
1025 class_match_nid(ldd->ldd_params,
1026 PARAM_FAILNODE, id.nid) < 1)
1029 /* match specified network */
1030 if (!class_match_net(ldd->ldd_params,
1031 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1034 mti->mti_nids[mti->mti_nid_count] = id.nid;
1035 mti->mti_nid_count++;
1036 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1037 CWARN("Only using first %d nids for %s\n",
1038 mti->mti_nid_count, mti->mti_svname);
1043 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1044 mti->mti_config_ver = 0;
1045 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF)
1046 ldd->ldd_flags |= LDD_F_WRITECONF;
1047 mti->mti_flags = ldd->ldd_flags;
1048 mti->mti_stripe_index = ldd->ldd_svindex;
1049 memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid));
1050 if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) {
1051 CERROR("params too big for mti\n");
1054 memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params));
1058 /* Register an old or new target with the MGS. If needed MGS will construct
1059 startup logs and assign index */
1060 int server_register_target(struct super_block *sb)
1062 struct lustre_sb_info *lsi = s2lsi(sb);
1063 struct obd_device *mgc = lsi->lsi_mgc;
1064 struct lustre_disk_data *ldd = lsi->lsi_ldd;
1065 struct mgs_target_info *mti = NULL;
1072 if (!(lsi->lsi_flags & LSI_SERVER))
1078 rc = server_sb2mti(sb, mti);
1082 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1083 mti->mti_svname, mti->mti_fsname,
1084 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1087 /* if write_conf is true, the registration must succeed */
1088 writeconf = !!(ldd->ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1089 mti->mti_flags |= LDD_F_OPC_REG;
1091 /* Register the target */
1092 /* FIXME use mgc_process_config instead */
1093 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1094 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
1095 sizeof(*mti), mti, NULL);
1097 if (mti->mti_flags & LDD_F_ERROR) {
1098 LCONSOLE_ERROR_MSG(0x160,
1099 "The MGS is refusing to allow this "
1100 "server (%s) to start. Please see messages"
1101 " on the MGS node.\n", ldd->ldd_svname);
1102 } else if (writeconf) {
1103 LCONSOLE_ERROR_MSG(0x15f,
1104 "Communication to the MGS return error %d. "
1105 "Is the MGS running?\n", rc);
1107 CERROR("Cannot talk to the MGS: %d, not fatal\n", rc);
1108 /* reset the error code for non-fatal error. */
1114 /* Always update our flags */
1115 ldd->ldd_flags = mti->mti_flags & LDD_F_ONDISK_MASK;
1117 /* If this flag is set, it means the MGS wants us to change our
1118 on-disk data. (So far this means just the index.) */
1119 if (mti->mti_flags & LDD_F_REWRITE_LDD) {
1122 CDEBUG(D_MOUNT, "Changing on-disk index from %#x to %#x "
1123 "for %s\n", ldd->ldd_svindex, mti->mti_stripe_index,
1125 ldd->ldd_svindex = mti->mti_stripe_index;
1126 strncpy(ldd->ldd_svname, mti->mti_svname,
1127 sizeof(ldd->ldd_svname));
1128 /* or ldd_make_sv_name(ldd); */
1129 ldd_write(&mgc->obd_lvfs_ctxt, ldd);
1130 err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb,
1133 CERROR("Label set error %d\n", err);
1134 label = fsfilt_get_label(mgc, lsi->lsi_srv_mnt->mnt_sb);
1136 CDEBUG(D_MOUNT, "Disk label changed to %s\n", label);
1138 /* Flush the new ldd to disk */
1139 fsfilt_sync(mgc, lsi->lsi_srv_mnt->mnt_sb);
1149 * Notify the MGS that this target is ready.
1150 * Used by IR - if the MGS receives this message, it will notify clients.
1152 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1154 struct lustre_sb_info *lsi = s2lsi(sb);
1155 struct obd_device *mgc = lsi->lsi_mgc;
1156 struct mgs_target_info *mti = NULL;
1162 if (!(lsi->lsi_flags & LSI_SERVER))
1168 rc = server_sb2mti(sb, mti);
1172 mti->mti_instance = obd->u.obt.obt_instance;
1173 mti->mti_flags |= LDD_F_OPC_READY;
1175 /* FIXME use mgc_process_config instead */
1176 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1177 sizeof(KEY_REGISTER_TARGET),
1178 KEY_REGISTER_TARGET,
1179 sizeof(*mti), mti, NULL);
1181 /* Imperative recovery: if the mgs informs us to use IR? */
1182 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1183 (mti->mti_flags & LDD_F_IR_CAPABLE))
1184 lsi->lsi_flags |= LSI_IR_CAPABLE;
1193 /** Start server targets: MDTs and OSTs
1195 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1197 struct obd_device *obd;
1198 struct lustre_sb_info *lsi = s2lsi(sb);
1199 struct config_llog_instance cfg;
1203 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname);
1206 /* If we're an MDT, make sure the global MDS is running */
1207 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1208 /* make sure the MDS is started */
1209 cfs_mutex_lock(&server_start_lock);
1210 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1212 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1213 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1215 LUSTRE_MDS_OBDNAME"_uuid",
1218 cfs_mutex_unlock(&server_start_lock);
1219 CERROR("failed to start MDS: %d\n", rc);
1223 cfs_mutex_unlock(&server_start_lock);
1227 /* If we're an OST, make sure the global OSS is running */
1228 if (IS_OST(lsi->lsi_ldd)) {
1229 /* make sure OSS is started */
1230 cfs_mutex_lock(&server_start_lock);
1231 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1233 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1235 LUSTRE_OSS_OBDNAME"_uuid",
1238 cfs_mutex_unlock(&server_start_lock);
1239 CERROR("failed to start OSS: %d\n", rc);
1243 cfs_mutex_unlock(&server_start_lock);
1246 /* Set the mgc fs to our server disk. This allows the MGC to
1247 * read and write configs locally, in case it can't talk to the MGS. */
1248 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1252 /* Register with MGS */
1253 rc = server_register_target(sb);
1257 /* Let the target look up the mount using the target's name
1258 (we can't pass the sb or mnt through class_process_config.) */
1259 rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
1263 /* Start targets using the llog named for the target */
1264 memset(&cfg, 0, sizeof(cfg));
1265 rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg);
1267 CERROR("failed to start server %s: %d\n",
1268 lsi->lsi_ldd->ldd_svname, rc);
1269 /* Do NOT call server_deregister_mount() here. This makes it
1270 * impossible to find mount later in cleanup time and leaves
1271 * @lsi and othder stuff leaked. -umka */
1276 /* Release the mgc fs for others to use */
1277 server_mgc_clear_fs(lsi->lsi_mgc);
1280 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1282 CERROR("no server named %s was started\n",
1283 lsi->lsi_ldd->ldd_svname);
1287 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1288 (OBP(obd, iocontrol))) {
1289 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1290 obd->obd_self_export, 0, NULL, NULL);
1293 server_notify_target(sb, obd);
1295 /* calculate recovery timeout, do it after lustre_process_log */
1296 server_calc_timeout(lsi, obd);
1298 /* log has been fully processed */
1299 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1305 /***************** lustre superblock **************/
1307 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1309 struct lustre_sb_info *lsi;
1315 OBD_ALLOC_PTR(lsi->lsi_lmd);
1316 if (!lsi->lsi_lmd) {
1321 lsi->lsi_lmd->lmd_exclude_count = 0;
1322 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
1323 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
1324 s2lsi_nocast(sb) = lsi;
1325 /* we take 1 extra ref for our setup */
1326 cfs_atomic_set(&lsi->lsi_mounts, 1);
1328 /* Default umount style */
1329 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1334 static int lustre_free_lsi(struct super_block *sb)
1336 struct lustre_sb_info *lsi = s2lsi(sb);
1339 LASSERT(lsi != NULL);
1340 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
1342 /* someone didn't call server_put_mount. */
1343 LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0);
1345 if (lsi->lsi_ldd != NULL)
1346 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
1348 if (lsi->lsi_lmd != NULL) {
1349 if (lsi->lsi_lmd->lmd_dev != NULL)
1350 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1351 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1352 if (lsi->lsi_lmd->lmd_profile != NULL)
1353 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1354 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1355 if (lsi->lsi_lmd->lmd_mgssec != NULL)
1356 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
1357 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
1358 if (lsi->lsi_lmd->lmd_opts != NULL)
1359 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1360 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1361 if (lsi->lsi_lmd->lmd_exclude_count)
1362 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1363 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1364 lsi->lsi_lmd->lmd_exclude_count);
1365 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1368 LASSERT(lsi->lsi_llsbi == NULL);
1369 OBD_FREE(lsi, sizeof(*lsi));
1370 s2lsi_nocast(sb) = NULL;
1375 /* The lsi has one reference for every server that is using the disk -
1376 e.g. MDT, MGS, and potentially MGC */
1377 static int lustre_put_lsi(struct super_block *sb)
1379 struct lustre_sb_info *lsi = s2lsi(sb);
1382 LASSERT(lsi != NULL);
1384 CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts));
1385 if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) {
1386 lustre_free_lsi(sb);
1392 /*************** server mount ******************/
1394 /** Kernel mount using mount options in MOUNT_DATA_FILE.
1395 * Since this file lives on the disk, we pre-mount using a common
1396 * type, read the file, then re-mount using the type specified in the
1399 static struct vfsmount *server_kernel_mount(struct super_block *sb)
1401 struct lvfs_run_ctxt mount_ctxt;
1402 struct lustre_sb_info *lsi = s2lsi(sb);
1403 struct lustre_disk_data *ldd;
1404 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1405 struct vfsmount *mnt;
1406 struct file_system_type *type;
1407 char *options = NULL;
1408 unsigned long page, s_flags;
1409 struct page *__page;
1414 OBD_ALLOC(ldd, sizeof(*ldd));
1416 RETURN(ERR_PTR(-ENOMEM));
1418 /* In the past, we have always used flags = 0.
1419 Note ext3/ldiskfs can't be mounted ro. */
1420 s_flags = sb->s_flags;
1422 /* allocate memory for options */
1423 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
1425 GOTO(out_free, rc = -ENOMEM);
1426 page = (unsigned long)cfs_page_address(__page);
1427 options = (char *)page;
1428 memset(options, 0, CFS_PAGE_SIZE);
1430 /* mount-line options must be added for pre-mount because it may
1431 * contain mount options such as journal_dev which are required
1432 * to mount successfuly the underlying filesystem */
1433 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0))
1434 strncat(options, lmd->lmd_opts, CFS_PAGE_SIZE - 1);
1436 /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */
1437 CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev);
1438 type = get_fs_type("ldiskfs");
1440 CERROR("premount failed: cannot find ldiskfs module\n");
1441 GOTO(out_free, rc = -ENODEV);
1443 mnt = vfs_kern_mount(type, s_flags, lmd->lmd_dev, (void *)options);
1444 cfs_module_put(type->owner);
1447 CERROR("premount %s:%#lx ldiskfs failed: %d "
1448 "Is the ldiskfs module available?\n",
1449 lmd->lmd_dev, s_flags, rc );
1453 OBD_SET_CTXT_MAGIC(&mount_ctxt);
1454 mount_ctxt.pwdmnt = mnt;
1455 mount_ctxt.pwd = mnt->mnt_root;
1456 mount_ctxt.fs = get_ds();
1458 rc = ldd_parse(&mount_ctxt, ldd);
1462 CERROR("premount parse options failed: rc = %d\n", rc);
1466 /* Done with our pre-mount, now do the real mount. */
1468 /* Glom up mount options */
1469 memset(options, 0, CFS_PAGE_SIZE);
1470 strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2);
1472 len = CFS_PAGE_SIZE - strlen(options) - 2;
1474 strcat(options, ",");
1475 strncat(options, "no_mbcache", len);
1477 /* Add in any mount-line options */
1478 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) {
1479 len = CFS_PAGE_SIZE - strlen(options) - 2;
1480 strcat(options, ",");
1481 strncat(options, lmd->lmd_opts, len);
1484 /* Special permanent mount flags */
1486 s_flags |= MS_NOATIME | MS_NODIRATIME;
1488 CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
1489 MT_STR(ldd), lmd->lmd_dev, options);
1490 type = get_fs_type(MT_STR(ldd));
1492 CERROR("get_fs_type failed\n");
1493 GOTO(out_free, rc = -ENODEV);
1495 mnt = vfs_kern_mount(type, s_flags, lmd->lmd_dev, (void *)options);
1496 cfs_module_put(type->owner);
1499 CERROR("vfs_kern_mount failed: rc = %d\n", rc);
1503 if (lmd->lmd_flags & LMD_FLG_ABORT_RECOV)
1504 simple_truncate(mnt->mnt_sb->s_root, mnt, LAST_RCVD,
1507 OBD_PAGE_FREE(__page);
1508 lsi->lsi_ldd = ldd; /* freed at lsi cleanup */
1509 CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
1514 OBD_PAGE_FREE(__page);
1515 OBD_FREE(ldd, sizeof(*ldd));
1516 lsi->lsi_ldd = NULL;
1517 RETURN(ERR_PTR(rc));
1520 /** Wait here forever until the mount refcount is 0 before completing umount,
1521 * else we risk dereferencing a null pointer.
1522 * LNET may take e.g. 165s before killing zombies.
1524 static void server_wait_finished(struct vfsmount *mnt)
1528 cfs_sigset_t blocked;
1530 cfs_waitq_init(&waitq);
1532 while (mnt_get_count(mnt) > 1) {
1533 if (waited && (waited % 30 == 0))
1534 LCONSOLE_WARN("Mount still busy with %d refs after "
1538 /* Cannot use l_event_wait() for an interruptible sleep. */
1540 blocked = cfs_block_sigsinv(sigmask(SIGKILL));
1541 cfs_waitq_wait_event_interruptible_timeout(
1543 (mnt_get_count(mnt) == 1),
1544 cfs_time_seconds(3),
1546 cfs_block_sigs(blocked);
1548 LCONSOLE_EMERG("Danger: interrupted umount %s with "
1549 "%d refs!\n", mnt->mnt_devname,
1550 mnt_get_count(mnt));
1557 /** Start the shutdown of servers at umount.
1559 static void server_put_super(struct super_block *sb)
1561 struct lustre_sb_info *lsi = s2lsi(sb);
1562 struct obd_device *obd;
1563 struct vfsmount *mnt = lsi->lsi_srv_mnt;
1564 char *tmpname, *extraname = NULL;
1566 int lddflags = lsi->lsi_ldd->ldd_flags;
1567 int lsiflags = lsi->lsi_flags;
1570 LASSERT(lsiflags & LSI_SERVER);
1572 tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1;
1573 OBD_ALLOC(tmpname, tmpname_sz);
1574 memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
1575 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1576 if (IS_MDT(lsi->lsi_ldd) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1577 snprintf(tmpname, tmpname_sz, "MGS");
1579 /* Stop the target */
1580 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1581 (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd))) {
1582 struct lustre_profile *lprof = NULL;
1584 /* tell the mgc to drop the config log */
1585 lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL);
1587 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1588 If there are any setup/cleanup errors, save the lov
1589 name for safety cleanup later. */
1590 lprof = class_get_profile(lsi->lsi_ldd->ldd_svname);
1591 if (lprof && lprof->lp_dt) {
1592 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1593 strcpy(extraname, lprof->lp_dt);
1596 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1598 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1599 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1601 /* We can't seem to give an error return code
1602 * to .put_super, so we better make sure we clean up! */
1604 class_manual_cleanup(obd);
1606 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
1607 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1611 /* If they wanted the mgs to stop separately from the mdt, they
1612 should have put it on a different device. */
1613 if (IS_MGS(lsi->lsi_ldd)) {
1614 /* if MDS start with --nomgs, don't stop MGS then */
1615 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1618 OBD_ALLOC(logname, MGS_PARAM_MAXLEN);
1620 LCONSOLE_WARN("Stopping mgs failed %d, please "
1621 "try again.", -ENOMEM);
1624 strcpy(logname, lsi->lsi_ldd->ldd_fsname);
1625 strcat(logname, "-params");
1626 /* tell the mgc to drop parameter config log */
1627 lustre_end_log(sb, logname, NULL);
1628 OBD_FREE(logname, MGS_PARAM_MAXLEN);
1630 server_stop_mgs(sb);
1634 /* Clean the mgc and sb */
1635 lustre_common_put_super(sb);
1637 /* Wait for the targets to really clean up - can't exit (and let the
1638 sb get destroyed) while the mount is still in use */
1639 server_wait_finished(mnt);
1641 /* drop the One True Mount */
1644 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1645 until the target is really gone so that our type refcount check
1647 server_stop_servers(lddflags, lsiflags);
1649 /* In case of startup or cleanup err, stop related obds */
1651 obd = class_name2obd(extraname);
1653 CWARN("Cleaning orphaned obd %s\n", extraname);
1655 class_manual_cleanup(obd);
1657 OBD_FREE(extraname, strlen(extraname) + 1);
1660 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1661 OBD_FREE(tmpname, tmpname_sz);
1665 /** Called only for 'umount -f'
1667 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1668 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1670 struct super_block *sb = vfsmnt->mnt_sb;
1672 static void server_umount_begin(struct super_block *sb)
1675 struct lustre_sb_info *lsi = s2lsi(sb);
1678 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1679 if (!(flags & MNT_FORCE)) {
1685 CDEBUG(D_MOUNT, "umount -f\n");
1686 /* umount = failover
1688 no third way to do non-force, non-failover */
1689 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1690 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1694 #ifndef HAVE_STATFS_DENTRY_PARAM
1695 static int server_statfs (struct super_block *sb, cfs_kstatfs_t *buf)
1698 static int server_statfs (struct dentry *dentry, cfs_kstatfs_t *buf)
1700 struct super_block *sb = dentry->d_sb;
1702 struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt;
1705 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) {
1706 #ifdef HAVE_STATFS_DENTRY_PARAM
1707 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf);
1709 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_sb, buf);
1712 buf->f_type = sb->s_magic;
1718 buf->f_type = sb->s_magic;
1719 buf->f_bsize = sb->s_blocksize;
1725 buf->f_namelen = NAME_MAX;
1729 /** The operations we support directly on the superblock:
1730 * mount, umount, and df.
1732 static struct super_operations server_ops =
1734 .put_super = server_put_super,
1735 .umount_begin = server_umount_begin, /* umount -f */
1736 .statfs = server_statfs,
1739 #define log2(n) cfs_ffz(~(n))
1740 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1742 static int server_fill_super_common(struct super_block *sb)
1744 struct inode *root = 0;
1747 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1749 sb->s_blocksize = 4096;
1750 sb->s_blocksize_bits = log2(sb->s_blocksize);
1751 sb->s_magic = LUSTRE_SUPER_MAGIC;
1752 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1753 sb->s_flags |= MS_RDONLY;
1754 sb->s_op = &server_ops;
1756 root = new_inode(sb);
1758 CERROR("Can't make root inode\n");
1762 /* returns -EIO for every operation */
1763 /* make_bad_inode(root); -- badness - can't umount */
1764 /* apparently we need to be a directory for the mount to finish */
1765 root->i_mode = S_IFDIR;
1767 sb->s_root = d_alloc_root(root);
1769 CERROR("Can't make root dentry\n");
1777 /** Fill in the superblock info for a Lustre server.
1778 * Mount the device with the correct options.
1779 * Read the on-disk config file.
1780 * Start the services.
1782 static int server_fill_super(struct super_block *sb)
1784 struct lustre_sb_info *lsi = s2lsi(sb);
1785 struct vfsmount *mnt;
1789 /* the One True Mount */
1790 mnt = server_kernel_mount(sb);
1793 CERROR("Unable to mount device %s: %d\n",
1794 lsi->lsi_lmd->lmd_dev, rc);
1798 lsi->lsi_srv_mnt = mnt;
1800 LASSERT(lsi->lsi_ldd);
1801 CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n",
1802 lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname,
1803 lsi->lsi_lmd->lmd_dev);
1805 if (class_name2obd(lsi->lsi_ldd->ldd_svname)) {
1806 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1807 "running. Double-mount may have compromised"
1808 " the disk journal.\n",
1809 lsi->lsi_ldd->ldd_svname);
1815 /* Start MGS before MGC */
1816 if (IS_MGS(lsi->lsi_ldd) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
1817 rc = server_start_mgs(sb);
1822 /* Start MGC before servers */
1823 rc = lustre_start_mgc(sb);
1827 /* Set up all obd devices for service */
1828 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1829 (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) {
1830 rc = server_start_targets(sb, mnt);
1832 CERROR("Unable to start targets: %d\n", rc);
1835 /* FIXME overmount client here,
1836 or can we just start a client log and client_fill_super on this sb?
1837 We need to make sure server_put_super gets called too - ll_put_super
1838 calls lustre_common_put_super; check there for LSI_SERVER flag,
1840 Probably should start client from new thread so we can return.
1841 Client will not finish until all servers are connected.
1842 Note - MGS-only server does NOT get a client, since there is no
1843 lustre fs associated - the MGS is for all lustre fs's */
1844 } else if (IS_MGS(lsi->lsi_ldd) &&
1845 !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
1846 struct config_llog_instance cfg;
1849 OBD_ALLOC(logname, MGS_PARAM_MAXLEN);
1850 if (logname == NULL)
1851 GOTO(out_mnt, rc = -ENOMEM);
1852 strcpy(logname, lsi->lsi_ldd->ldd_fsname);
1853 strcat(logname, "-params");
1855 memset(&cfg, 0, sizeof(cfg));
1856 rc = lustre_process_log(sb, logname, &cfg);
1857 OBD_FREE(logname, MGS_PARAM_MAXLEN);
1859 CERROR("failed to process parameters %s: %d\n",
1865 rc = server_fill_super_common(sb);
1871 /* We jump here in case of failure while starting targets or MGS.
1872 * In this case we can't just put @mnt and have to do real cleanup
1873 * with stoping targets, etc. */
1874 server_put_super(sb);
1878 /* Get the index from the obd name.
1879 rc = server type, or
1881 if endptr isn't NULL it is set to end of name */
1882 int server_name2index(char *svname, __u32 *idx, char **endptr)
1884 unsigned long index;
1886 char *dash = strrchr(svname, '-');
1890 /* intepret <fsname>-MDTXXXXX-mdc as mdt, the better way is to pass
1891 * in the fsname, then determine the server index */
1892 if (!strcmp(LUSTRE_MDC_NAME, dash + 1)) {
1894 for (; dash > svname && *dash != '-'; dash--);
1899 if (strncmp(dash + 1, "MDT", 3) == 0)
1900 rc = LDD_F_SV_TYPE_MDT;
1901 else if (strncmp(dash + 1, "OST", 3) == 0)
1902 rc = LDD_F_SV_TYPE_OST;
1905 if (strcmp(dash + 4, "all") == 0)
1906 return rc | LDD_F_SV_ALL;
1908 index = simple_strtoul(dash + 4, endptr, 16);
1914 * Calculate timeout value for a target.
1916 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1918 struct lustre_mount_data *lmd;
1922 bool has_ir = !!(lsi->lsi_flags & LSI_IR_CAPABLE);
1923 int min = OBD_RECOVERY_TIME_MIN;
1925 LASSERT(lsi->lsi_flags & LSI_SERVER);
1929 soft = lmd->lmd_recovery_time_soft;
1930 hard = lmd->lmd_recovery_time_hard;
1931 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1932 obd->obd_no_ir = !has_ir;
1936 soft = OBD_RECOVERY_TIME_SOFT;
1938 hard = OBD_RECOVERY_TIME_HARD;
1940 /* target may have ir_factor configured. */
1941 factor = OBD_IR_FACTOR_DEFAULT;
1942 if (obd->obd_recovery_ir_factor)
1943 factor = obd->obd_recovery_ir_factor;
1946 int new_soft = soft;
1947 int new_hard = hard;
1949 /* adjust timeout value by imperative recovery */
1951 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1952 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1954 /* make sure the timeout is not too short */
1955 new_soft = max(min, new_soft);
1956 new_hard = max(new_soft, new_hard);
1958 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1959 "window shrunk from %d-%d down to %d-%d\n",
1960 obd->obd_name, soft, hard, new_soft, new_hard);
1967 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1968 obd->obd_recovery_time_hard = hard;
1969 obd->obd_recovery_ir_factor = factor;
1971 EXPORT_SYMBOL(server_calc_timeout);
1973 /*************** mount common betweeen server and client ***************/
1976 int lustre_common_put_super(struct super_block *sb)
1981 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1983 /* Drop a ref to the MGC */
1984 rc = lustre_stop_mgc(sb);
1985 if (rc && (rc != -ENOENT)) {
1987 CERROR("Can't stop MGC: %d\n", rc);
1990 /* BUSY just means that there's some other obd that
1991 needs the mgc. Let him clean it up. */
1992 CDEBUG(D_MOUNT, "MGC still in use\n");
1994 /* Drop a ref to the mounted disk */
2000 static void lmd_print(struct lustre_mount_data *lmd)
2004 PRINT_CMD(PRINT_MASK, " mount data:\n");
2005 if (lmd_is_client(lmd))
2006 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
2007 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
2008 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
2011 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
2013 if (lmd->lmd_recovery_time_soft)
2014 PRINT_CMD(PRINT_MASK, "recovery time soft: %d\n",
2015 lmd->lmd_recovery_time_soft);
2017 if (lmd->lmd_recovery_time_hard)
2018 PRINT_CMD(PRINT_MASK, "recovery time hard: %d\n",
2019 lmd->lmd_recovery_time_hard);
2021 for (i = 0; i < lmd->lmd_exclude_count; i++) {
2022 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
2023 lmd->lmd_exclude[i]);
2027 /* Is this server on the exclusion list */
2028 int lustre_check_exclusion(struct super_block *sb, char *svname)
2030 struct lustre_sb_info *lsi = s2lsi(sb);
2031 struct lustre_mount_data *lmd = lsi->lsi_lmd;
2036 rc = server_name2index(svname, &index, NULL);
2037 if (rc != LDD_F_SV_TYPE_OST)
2038 /* Only exclude OSTs */
2041 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
2042 index, lmd->lmd_exclude_count, lmd->lmd_dev);
2044 for(i = 0; i < lmd->lmd_exclude_count; i++) {
2045 if (index == lmd->lmd_exclude[i]) {
2046 CWARN("Excluding %s (on exclusion list)\n", svname);
2053 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
2054 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
2056 char *s1 = ptr, *s2;
2057 __u32 index, *exclude_list;
2061 /* The shortest an ost name can be is 8 chars: -OST0000.
2062 We don't actually know the fsname at this time, so in fact
2063 a user could specify any fsname. */
2064 devmax = strlen(ptr) / 8 + 1;
2066 /* temp storage until we figure out how many we have */
2067 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
2071 /* we enter this fn pointing at the '=' */
2072 while (*s1 && *s1 != ' ' && *s1 != ',') {
2074 rc = server_name2index(s1, &index, &s2);
2076 CERROR("Can't parse server name '%s'\n", s1);
2079 if (rc == LDD_F_SV_TYPE_OST)
2080 exclude_list[lmd->lmd_exclude_count++] = index;
2082 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
2084 /* now we are pointing at ':' (next exclude)
2085 or ',' (end of excludes) */
2086 if (lmd->lmd_exclude_count >= devmax)
2089 if (rc >= 0) /* non-err */
2092 if (lmd->lmd_exclude_count) {
2093 /* permanent, freed in lustre_free_lsi */
2094 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
2095 lmd->lmd_exclude_count);
2096 if (lmd->lmd_exclude) {
2097 memcpy(lmd->lmd_exclude, exclude_list,
2098 sizeof(index) * lmd->lmd_exclude_count);
2101 lmd->lmd_exclude_count = 0;
2104 OBD_FREE(exclude_list, sizeof(index) * devmax);
2108 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
2113 if (lmd->lmd_mgssec != NULL) {
2114 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
2115 lmd->lmd_mgssec = NULL;
2118 tail = strchr(ptr, ',');
2120 length = strlen(ptr);
2122 length = tail - ptr;
2124 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
2125 if (lmd->lmd_mgssec == NULL)
2128 memcpy(lmd->lmd_mgssec, ptr, length);
2129 lmd->lmd_mgssec[length] = '\0';
2133 /** Parse mount line options
2134 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
2135 * dev is passed as device=uml1:/lustre by mount.lustre
2137 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
2139 char *s1, *s2, *devname = NULL;
2140 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
2146 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
2147 "/sbin/mount.lustre is installed.\n");
2151 /* Options should be a string - try to detect old lmd data */
2152 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
2153 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
2154 "/sbin/mount.lustre. Please install "
2155 "version %s\n", LUSTRE_VERSION_STRING);
2158 lmd->lmd_magic = LMD_MAGIC;
2160 /* Set default flags here */
2165 int time_min = OBD_RECOVERY_TIME_MIN;
2167 /* Skip whitespace and extra commas */
2168 while (*s1 == ' ' || *s1 == ',')
2171 /* Client options are parsed in ll_options: eg. flock,
2174 /* Parse non-ldiskfs options here. Rather than modifying
2175 ldiskfs, we just zero these out here */
2176 if (strncmp(s1, "abort_recov", 11) == 0) {
2177 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
2179 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
2180 lmd->lmd_recovery_time_soft = max_t(int,
2181 simple_strtoul(s1 + 19, NULL, 10), time_min);
2183 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
2184 lmd->lmd_recovery_time_hard = max_t(int,
2185 simple_strtoul(s1 + 19, NULL, 10), time_min);
2187 } else if (strncmp(s1, "noir", 4) == 0) {
2188 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
2190 } else if (strncmp(s1, "nosvc", 5) == 0) {
2191 lmd->lmd_flags |= LMD_FLG_NOSVC;
2193 } else if (strncmp(s1, "nomgs", 5) == 0) {
2194 lmd->lmd_flags |= LMD_FLG_NOMGS;
2196 } else if (strncmp(s1, "noscrub", 7) == 0) {
2197 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
2199 } else if (strncmp(s1, "writeconf", 9) == 0) {
2200 lmd->lmd_flags |= LMD_FLG_WRITECONF;
2202 } else if (strncmp(s1, "mgssec=", 7) == 0) {
2203 rc = lmd_parse_mgssec(lmd, s1 + 7);
2207 /* ost exclusion list */
2208 } else if (strncmp(s1, "exclude=", 8) == 0) {
2209 rc = lmd_make_exclusion(lmd, s1 + 7);
2214 /* Linux 2.4 doesn't pass the device, so we stuck it at the
2215 end of the options. */
2216 else if (strncmp(s1, "device=", 7) == 0) {
2218 /* terminate options right before device. device
2219 must be the last one. */
2225 s2 = strchr(s1, ',');
2233 memmove(s1, s2, strlen(s2) + 1);
2239 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
2240 "(need mount option 'device=...')\n");
2244 s1 = strstr(devname, ":/");
2247 lmd->lmd_flags |= LMD_FLG_CLIENT;
2248 /* Remove leading /s from fsname */
2249 while (*++s1 == '/') ;
2250 /* Freed in lustre_free_lsi */
2251 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
2252 if (!lmd->lmd_profile)
2254 sprintf(lmd->lmd_profile, "%s-client", s1);
2257 /* Freed in lustre_free_lsi */
2258 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
2261 strcpy(lmd->lmd_dev, devname);
2263 /* Save mount options */
2264 s1 = options + strlen(options) - 1;
2265 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
2267 if (*options != 0) {
2268 /* Freed in lustre_free_lsi */
2269 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
2272 strcpy(lmd->lmd_opts, options);
2276 lmd->lmd_magic = LMD_MAGIC;
2281 CERROR("Bad mount options %s\n", options);
2285 struct lustre_mount_data2 {
2287 struct vfsmount *lmd2_mnt;
2290 /** This is the entry point for the mount call into Lustre.
2291 * This is called when a server or client is mounted,
2292 * and this is where we start setting things up.
2293 * @param data Mount options (e.g. -o flock,abort_recov)
2295 int lustre_fill_super(struct super_block *sb, void *data, int silent)
2297 struct lustre_mount_data *lmd;
2298 struct lustre_mount_data2 *lmd2 = data;
2299 struct lustre_sb_info *lsi;
2303 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
2305 lsi = lustre_init_lsi(sb);
2311 * Disable lockdep during mount, because mount locking patterns are
2317 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
2319 obd_zombie_barrier();
2321 /* Figure out the lmd from the mount options */
2322 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
2324 GOTO(out, rc = -EINVAL);
2327 if (lmd_is_client(lmd)) {
2328 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
2329 if (!client_fill_super) {
2330 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
2331 "client mount! Is the 'lustre' "
2332 "module loaded?\n");
2336 rc = lustre_start_mgc(sb);
2341 /* Connect and start */
2342 /* (should always be ll_fill_super) */
2343 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
2344 /* c_f_s will call lustre_common_put_super on failure */
2347 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
2348 lsi->lsi_flags |= LSI_SERVER;
2349 rc = server_fill_super(sb);
2350 /* s_f_s calls lustre_start_mgc after the mount because we need
2351 the MGS nids which are stored on disk. Plus, we may
2352 need to start the MGS first. */
2353 /* s_f_s will call server_put_super on failure */
2356 /* If error happens in fill_super() call, @lsi will be killed there.
2357 * This is why we do not put it here. */
2361 CERROR("Unable to mount %s (%d)\n",
2362 s2lsi(sb) ? lmd->lmd_dev : "", rc);
2364 CDEBUG(D_SUPER, "Mount %s complete\n",
2372 /* We can't call ll_fill_super by name because it lives in a module that
2373 must be loaded after this one. */
2374 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
2375 struct vfsmount *mnt))
2377 client_fill_super = cfs;
2380 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
2382 kill_super_cb = cfs;
2385 /***************** FS registration ******************/
2387 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))
2388 struct super_block * lustre_get_sb(struct file_system_type *fs_type, int flags,
2389 const char *devname, void * data)
2391 return get_sb_nodev(fs_type, flags, data, lustre_fill_super);
2394 int lustre_get_sb(struct file_system_type *fs_type, int flags,
2395 const char *devname, void * data, struct vfsmount *mnt)
2397 struct lustre_mount_data2 lmd2 = {data, mnt};
2399 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
2403 void lustre_kill_super(struct super_block *sb)
2405 struct lustre_sb_info *lsi = s2lsi(sb);
2407 if (kill_super_cb && lsi && !(lsi->lsi_flags & LSI_SERVER))
2408 (*kill_super_cb)(sb);
2410 kill_anon_super(sb);
2413 /** Register the "lustre" fs type
2415 struct file_system_type lustre_fs_type = {
2416 .owner = THIS_MODULE,
2418 .get_sb = lustre_get_sb,
2419 .kill_sb = lustre_kill_super,
2420 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
2421 #ifdef FS_HAS_FIEMAP
2424 LL_RENAME_DOES_D_MOVE,
2427 int lustre_register_fs(void)
2429 return register_filesystem(&lustre_fs_type);
2432 int lustre_unregister_fs(void)
2434 return unregister_filesystem(&lustre_fs_type);
2437 EXPORT_SYMBOL(lustre_register_client_fill_super);
2438 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2439 EXPORT_SYMBOL(lustre_common_put_super);
2440 EXPORT_SYMBOL(lustre_process_log);
2441 EXPORT_SYMBOL(lustre_end_log);
2442 EXPORT_SYMBOL(server_get_mount);
2443 EXPORT_SYMBOL(server_get_mount_2);
2444 EXPORT_SYMBOL(server_put_mount);
2445 EXPORT_SYMBOL(server_put_mount_2);
2446 EXPORT_SYMBOL(server_register_target);
2447 EXPORT_SYMBOL(server_name2index);
2448 EXPORT_SYMBOL(server_mti_print);
2449 EXPORT_SYMBOL(do_lcfg);