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;
95 OBD_ALLOC(lmi, sizeof(*lmi));
98 OBD_ALLOC(name_cp, strlen(name) + 1);
100 OBD_FREE(lmi, sizeof(*lmi));
103 strcpy(name_cp, name);
105 cfs_mutex_lock(&lustre_mount_info_lock);
107 if (server_find_mount(name)) {
108 cfs_mutex_unlock(&lustre_mount_info_lock);
109 OBD_FREE(lmi, sizeof(*lmi));
110 OBD_FREE(name_cp, strlen(name) + 1);
111 CERROR("Already registered %s\n", name);
114 lmi->lmi_name = name_cp;
117 cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list);
119 cfs_mutex_unlock(&lustre_mount_info_lock);
121 CDEBUG(D_MOUNT, "reg_mnt %p from %s, vfscount=%d\n",
123 lmi->lmi_mnt ? mnt_get_count(lmi->lmi_mnt) : -1);
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",
144 lmi->lmi_mnt ? mnt_get_count(lmi->lmi_mnt) : -1);
146 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
147 cfs_list_del(&lmi->lmi_list_chain);
148 OBD_FREE(lmi, sizeof(*lmi));
149 cfs_mutex_unlock(&lustre_mount_info_lock);
154 /* obd's look up a registered mount using their obdname. This is just
155 for initial obd setup to find the mount struct. It should not be
156 called every time you want to mntget. */
157 struct lustre_mount_info *server_get_mount(const char *name)
159 struct lustre_mount_info *lmi;
160 struct lustre_sb_info *lsi;
163 cfs_mutex_lock(&lustre_mount_info_lock);
164 lmi = server_find_mount(name);
165 cfs_mutex_unlock(&lustre_mount_info_lock);
167 CERROR("Can't find mount for %s\n", name);
170 lsi = s2lsi(lmi->lmi_sb);
173 mntget(lmi->lmi_mnt);
174 cfs_atomic_inc(&lsi->lsi_mounts);
176 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d, vfscount=%d\n",
177 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts),
178 lmi->lmi_mnt ? mnt_get_count(lmi->lmi_mnt) - 1 : -1);
184 * Used by mdt to get mount_info from obdname.
185 * There are no blocking when using the mount_info.
186 * Do not use server_get_mount for this purpose.
188 struct lustre_mount_info *server_get_mount_2(const char *name)
190 struct lustre_mount_info *lmi;
193 cfs_mutex_lock(&lustre_mount_info_lock);
194 lmi = server_find_mount(name);
195 cfs_mutex_unlock(&lustre_mount_info_lock);
197 CERROR("Can't find mount for %s\n", name);
202 static void unlock_mntput(struct vfsmount *mnt)
204 #ifdef HAVE_KERNEL_LOCKED
205 /* for kernel < 2.6.37 */
206 if (kernel_locked()) {
218 static int lustre_put_lsi(struct super_block *sb);
220 /* to be called from obd_cleanup methods */
221 int server_put_mount(const char *name, struct vfsmount *mnt)
223 struct lustre_mount_info *lmi;
224 struct lustre_sb_info *lsi;
228 /* This might be the last one, can't deref after this */
230 count = mnt_get_count(mnt) - 1;
234 cfs_mutex_lock(&lustre_mount_info_lock);
235 lmi = server_find_mount(name);
236 cfs_mutex_unlock(&lustre_mount_info_lock);
238 CERROR("Can't find mount for %s\n", name);
241 lsi = s2lsi(lmi->lmi_sb);
242 LASSERT(lmi->lmi_mnt == mnt);
244 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d, vfscount=%d\n",
245 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts), count);
247 if (lustre_put_lsi(lmi->lmi_sb)) {
248 CDEBUG(D_MOUNT, "Last put of mnt %p from %s, vfscount=%d\n",
249 lmi->lmi_mnt, name, count);
250 /* last mount is the One True Mount */
252 CERROR("%s: mount busy, vfscount=%d!\n", name, count);
255 /* this obd should never need the mount again */
256 server_deregister_mount(name);
261 /* Corresponding to server_get_mount_2 */
262 int server_put_mount_2(const char *name, struct vfsmount *mnt)
268 /******* mount helper utilities *********/
271 static void ldd_print(struct lustre_disk_data *ldd)
273 PRINT_CMD(PRINT_MASK, " disk data:\n");
274 PRINT_CMD(PRINT_MASK, "server: %s\n", ldd->ldd_svname);
275 PRINT_CMD(PRINT_MASK, "uuid: %s\n", (char *)ldd->ldd_uuid);
276 PRINT_CMD(PRINT_MASK, "fs: %s\n", ldd->ldd_fsname);
277 PRINT_CMD(PRINT_MASK, "index: %04x\n", ldd->ldd_svindex);
278 PRINT_CMD(PRINT_MASK, "config: %d\n", ldd->ldd_config_ver);
279 PRINT_CMD(PRINT_MASK, "flags: %#x\n", ldd->ldd_flags);
280 PRINT_CMD(PRINT_MASK, "diskfs: %s\n", MT_STR(ldd));
281 PRINT_CMD(PRINT_MASK, "options: %s\n", ldd->ldd_mount_opts);
282 PRINT_CMD(PRINT_MASK, "params: %s\n", ldd->ldd_params);
283 PRINT_CMD(PRINT_MASK, "comment: %s\n", ldd->ldd_userdata);
287 static int ldd_parse(struct lvfs_run_ctxt *mount_ctxt,
288 struct lustre_disk_data *ldd)
290 struct lvfs_run_ctxt saved;
297 push_ctxt(&saved, mount_ctxt, NULL);
299 file = filp_open(MOUNT_DATA_FILE, O_RDONLY, 0644);
302 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
306 len = i_size_read(file->f_dentry->d_inode);
307 CDEBUG(D_MOUNT, "Have %s, size %lu\n", MOUNT_DATA_FILE, len);
308 if (len != sizeof(*ldd)) {
309 CERROR("disk data size does not match: see %lu expect %u\n",
310 len, (int)sizeof(*ldd));
311 GOTO(out_close, rc = -EINVAL);
314 rc = lustre_fread(file, ldd, len, &off);
316 CERROR("error reading %s: read %d of %lu\n",
317 MOUNT_DATA_FILE, rc, len);
318 GOTO(out_close, rc = -EINVAL);
322 if (ldd->ldd_magic != LDD_MAGIC) {
323 /* FIXME add swabbing support */
324 CERROR("Bad magic in %s: %x!=%x\n", MOUNT_DATA_FILE,
325 ldd->ldd_magic, LDD_MAGIC);
326 GOTO(out_close, rc = -EINVAL);
329 if (ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP) {
330 CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
332 ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP);
333 GOTO(out_close, rc = -EINVAL);
335 if (ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP) {
336 CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
338 ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP);
339 /* Do something like remount filesystem read-only */
340 GOTO(out_close, rc = -EINVAL);
346 pop_ctxt(&saved, mount_ctxt, NULL);
350 static int ldd_write(struct lvfs_run_ctxt *mount_ctxt,
351 struct lustre_disk_data *ldd)
353 struct lvfs_run_ctxt saved;
356 unsigned long len = sizeof(struct lustre_disk_data);
360 if (ldd->ldd_magic == 0)
363 LASSERT(ldd->ldd_magic == LDD_MAGIC);
365 ldd->ldd_config_ver++;
367 push_ctxt(&saved, mount_ctxt, NULL);
369 file = filp_open(MOUNT_DATA_FILE, O_RDWR|O_SYNC, 0644);
372 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
376 rc = lustre_fwrite(file, ldd, len, &off);
378 CERROR("error writing %s: read %d of %lu\n",
379 MOUNT_DATA_FILE, rc, len);
380 GOTO(out_close, rc = -EINVAL);
388 pop_ctxt(&saved, mount_ctxt, NULL);
393 /**************** config llog ********************/
395 /** Get a config log from the MGS and process it.
396 * This func is called for both clients and servers.
397 * Continue to process new statements appended to the logs
398 * (whenever the config lock is revoked) until lustre_end_log
400 * @param sb The superblock is used by the MGC to write to the local copy of
402 * @param logname The name of the llog to replicate from the MGS
403 * @param cfg Since the same mgc may be used to follow multiple config logs
404 * (e.g. ost1, ost2, client), the config_llog_instance keeps the state for
405 * this log, and is added to the mgc's list of logs to follow.
407 int lustre_process_log(struct super_block *sb, char *logname,
408 struct config_llog_instance *cfg)
410 struct lustre_cfg *lcfg;
411 struct lustre_cfg_bufs *bufs;
412 struct lustre_sb_info *lsi = s2lsi(sb);
413 struct obd_device *mgc = lsi->lsi_mgc;
424 /* mgc_process_config */
425 lustre_cfg_bufs_reset(bufs, mgc->obd_name);
426 lustre_cfg_bufs_set_string(bufs, 1, logname);
427 lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg));
428 lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb));
429 lcfg = lustre_cfg_new(LCFG_LOG_START, bufs);
430 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
431 lustre_cfg_free(lcfg);
436 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
437 "failed from the MGS (%d). Make sure this "
438 "client and the MGS are running compatible "
439 "versions of Lustre.\n",
440 mgc->obd_name, logname, rc);
443 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
444 "failed (%d). This may be the result of "
445 "communication errors between this node and "
446 "the MGS, a bad configuration, or other "
447 "errors. See the syslog for more "
448 "information.\n", mgc->obd_name, logname,
451 /* class_obd_list(); */
455 /* Stop watching this config log for updates */
456 int lustre_end_log(struct super_block *sb, char *logname,
457 struct config_llog_instance *cfg)
459 struct lustre_cfg *lcfg;
460 struct lustre_cfg_bufs bufs;
461 struct lustre_sb_info *lsi = s2lsi(sb);
462 struct obd_device *mgc = lsi->lsi_mgc;
469 /* mgc_process_config */
470 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
471 lustre_cfg_bufs_set_string(&bufs, 1, logname);
473 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
474 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
475 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
476 lustre_cfg_free(lcfg);
480 /**************** obd start *******************/
482 /** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from
483 * lctl (and do for echo cli/srv.
485 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
486 char *s1, char *s2, char *s3, char *s4)
488 struct lustre_cfg_bufs bufs;
489 struct lustre_cfg * lcfg = NULL;
492 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
493 cmd, s1, s2, s3, s4);
495 lustre_cfg_bufs_reset(&bufs, cfgname);
497 lustre_cfg_bufs_set_string(&bufs, 1, s1);
499 lustre_cfg_bufs_set_string(&bufs, 2, s2);
501 lustre_cfg_bufs_set_string(&bufs, 3, s3);
503 lustre_cfg_bufs_set_string(&bufs, 4, s4);
505 lcfg = lustre_cfg_new(cmd, &bufs);
506 lcfg->lcfg_nid = nid;
507 rc = class_process_config(lcfg);
508 lustre_cfg_free(lcfg);
512 /** Call class_attach and class_setup. These methods in turn call
513 * obd type-specific methods.
515 static int lustre_start_simple(char *obdname, char *type, char *uuid,
519 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
521 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
523 CERROR("%s attach error %d\n", obdname, rc);
526 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, 0, 0);
528 CERROR("%s setup error %d\n", obdname, rc);
529 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
534 /* Set up a MGS to serve startup logs */
535 static int server_start_mgs(struct super_block *sb)
537 struct lustre_sb_info *lsi = s2lsi(sb);
538 struct vfsmount *mnt = lsi->lsi_srv_mnt;
539 struct lustre_mount_info *lmi;
544 /* It is impossible to have more than 1 MGS per node, since
545 MGC wouldn't know which to connect to */
546 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
548 lsi = s2lsi(lmi->lmi_sb);
549 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
551 lsi->lsi_ldd->ldd_svname);
555 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
557 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
560 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
561 LUSTRE_MGS_OBDNAME, 0, 0);
562 /* Do NOT call server_deregister_mount() here. This leads to
563 * inability cleanup cleanly and free lsi and other stuff when
564 * mgs calls server_put_mount() in error handling case. -umka */
568 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
569 "Is the 'mgs' module loaded?\n",
570 LUSTRE_MGS_OBDNAME, rc);
574 static int server_stop_mgs(struct super_block *sb)
576 struct obd_device *obd;
580 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
582 /* There better be only one MGS */
583 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
585 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
589 /* The MGS should always stop when we say so */
591 rc = class_manual_cleanup(obd);
595 CFS_DEFINE_MUTEX(mgc_start_lock);
597 /** Set up a mgc obd to process startup logs
599 * \param sb [in] super block of the mgc obd
601 * \retval 0 success, otherwise error code
603 static int lustre_start_mgc(struct super_block *sb)
605 struct obd_connect_data *data = NULL;
606 struct lustre_sb_info *lsi = s2lsi(sb);
607 struct obd_device *obd;
608 struct obd_export *exp;
609 struct obd_uuid *uuid;
612 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
615 int rc = 0, i = 0, j, len;
618 LASSERT(lsi->lsi_lmd);
620 /* Find the first non-lo MGS nid for our MGC name */
621 if (lsi->lsi_flags & LSI_SERVER) {
622 ptr = lsi->lsi_ldd->ldd_params;
623 /* mount -o mgsnode=nid */
624 if (lsi->lsi_lmd->lmd_mgs &&
625 (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) {
627 /* Use mgsnode= nids */
628 } else if ((class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0) &&
629 (class_parse_nid(ptr, &nid, &ptr) == 0)) {
631 } else if (IS_MGS(lsi->lsi_ldd)) {
632 lnet_process_id_t id;
633 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
634 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
641 } else { /* client */
642 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
643 ptr = lsi->lsi_lmd->lmd_dev;
644 if (class_parse_nid(ptr, &nid, &ptr) == 0)
648 CERROR("No valid MGS nids found.\n");
652 cfs_mutex_lock(&mgc_start_lock);
654 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
655 OBD_ALLOC(mgcname, len);
656 OBD_ALLOC(niduuid, len + 2);
657 if (!mgcname || !niduuid)
658 GOTO(out_free, rc = -ENOMEM);
659 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
661 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
665 GOTO(out_free, rc = -ENOMEM);
667 obd = class_name2obd(mgcname);
668 if (obd && !obd->obd_stopping) {
669 rc = obd_set_info_async(NULL, obd->obd_self_export,
670 strlen(KEY_MGSSEC), KEY_MGSSEC,
671 strlen(mgssec), mgssec, NULL);
675 /* Re-using an existing MGC */
676 cfs_atomic_inc(&obd->u.cli.cl_mgc_refcount);
678 /* IR compatibility check, only for clients */
679 if (lmd_is_client(lsi->lsi_lmd)) {
681 int vallen = sizeof(*data);
682 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
684 rc = obd_get_info(NULL, obd->obd_self_export,
685 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
686 &vallen, data, NULL);
688 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
689 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
690 /* LMD_FLG_NOIR is for test purpose only */
692 "Trying to mount a client with IR setting "
693 "not compatible with current mgc. "
694 "Force to use current mgc setting that is "
696 has_ir ? "enabled" : "disabled");
698 *flags &= ~LMD_FLG_NOIR;
700 *flags |= LMD_FLG_NOIR;
705 /* If we are restarting the MGS, don't try to keep the MGC's
706 old connection, or registration will fail. */
707 if ((lsi->lsi_flags & LSI_SERVER) && IS_MGS(lsi->lsi_ldd)) {
708 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
712 /* Try all connections, but only once (again).
713 We don't want to block another target from starting
714 (using its local copy of the log), but we do want to connect
715 if at all possible. */
717 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
718 rc = obd_set_info_async(NULL, obd->obd_self_export,
719 sizeof(KEY_INIT_RECOV_BACKUP),
720 KEY_INIT_RECOV_BACKUP,
721 sizeof(recov_bk), &recov_bk, NULL);
725 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
727 /* Add the primary nids for the MGS */
729 sprintf(niduuid, "%s_%x", mgcname, i);
730 if (lsi->lsi_flags & LSI_SERVER) {
731 ptr = lsi->lsi_ldd->ldd_params;
732 if (IS_MGS(lsi->lsi_ldd)) {
733 /* Use local nids (including LO) */
734 lnet_process_id_t id;
735 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
736 rc = do_lcfg(mgcname, id.nid,
737 LCFG_ADD_UUID, niduuid, 0,0,0);
740 /* Use mgsnode= nids */
741 /* mount -o mgsnode=nid */
742 if (lsi->lsi_lmd->lmd_mgs) {
743 ptr = lsi->lsi_lmd->lmd_mgs;
744 } else if (class_find_param(ptr, PARAM_MGSNODE,
746 CERROR("No MGS nids given.\n");
747 GOTO(out_free, rc = -EINVAL);
749 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
750 rc = do_lcfg(mgcname, nid,
751 LCFG_ADD_UUID, niduuid, 0,0,0);
755 } else { /* client */
756 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
757 ptr = lsi->lsi_lmd->lmd_dev;
758 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
759 rc = do_lcfg(mgcname, nid,
760 LCFG_ADD_UUID, niduuid, 0,0,0);
762 /* Stop at the first failover nid */
768 CERROR("No valid MGS nids found.\n");
769 GOTO(out_free, rc = -EINVAL);
771 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
773 /* Random uuid for MGC allows easier reconnects */
775 ll_generate_random_uuid(uuidc);
776 class_uuid_unparse(uuidc, uuid);
779 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
780 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
786 /* Add any failover MGS nids */
788 while ((*ptr == ':' ||
789 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0)) {
790 /* New failover node */
791 sprintf(niduuid, "%s_%x", mgcname, i);
793 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
795 rc = do_lcfg(mgcname, nid,
796 LCFG_ADD_UUID, niduuid, 0,0,0);
801 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
809 lsi->lsi_lmd->lmd_mgs_failnodes = i;
811 obd = class_name2obd(mgcname);
813 CERROR("Can't find mgcobd %s\n", mgcname);
814 GOTO(out_free, rc = -ENOTCONN);
817 rc = obd_set_info_async(NULL, obd->obd_self_export,
818 strlen(KEY_MGSSEC), KEY_MGSSEC,
819 strlen(mgssec), mgssec, NULL);
823 /* Keep a refcount of servers/clients who started with "mount",
824 so we know when we can get rid of the mgc. */
825 cfs_atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
827 /* Try all connections, but only once. */
829 rc = obd_set_info_async(NULL, obd->obd_self_export,
830 sizeof(KEY_INIT_RECOV_BACKUP),
831 KEY_INIT_RECOV_BACKUP,
832 sizeof(recov_bk), &recov_bk, NULL);
835 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
836 /* We connect to the MGS at setup, and don't disconnect until cleanup */
837 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID |
838 OBD_CONNECT_AT | OBD_CONNECT_FULL20 |
839 OBD_CONNECT_IMP_RECOV;
840 if (lmd_is_client(lsi->lsi_lmd) &&
841 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
842 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
843 data->ocd_version = LUSTRE_VERSION_CODE;
844 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
846 CERROR("connect failed %d\n", rc);
850 obd->u.cli.cl_mgc_mgsexp = exp;
853 /* Keep the mgc info in the sb. Note that many lsi's can point
857 cfs_mutex_unlock(&mgc_start_lock);
862 OBD_FREE(mgcname, len);
864 OBD_FREE(niduuid, len + 2);
868 static int lustre_stop_mgc(struct super_block *sb)
870 struct lustre_sb_info *lsi = s2lsi(sb);
871 struct obd_device *obd;
872 char *niduuid = 0, *ptr = 0;
873 int i, rc = 0, len = 0;
883 cfs_mutex_lock(&mgc_start_lock);
884 LASSERT(cfs_atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
885 if (!cfs_atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
886 /* This is not fatal, every client that stops
887 will call in here. */
888 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
889 cfs_atomic_read(&obd->u.cli.cl_mgc_refcount));
890 GOTO(out, rc = -EBUSY);
893 /* The MGC has no recoverable data in any case.
894 * force shotdown set in umount_begin */
895 obd->obd_no_recov = 1;
897 if (obd->u.cli.cl_mgc_mgsexp) {
898 /* An error is not fatal, if we are unable to send the
899 disconnect mgs ping evictor cleans up the export */
900 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
902 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
905 /* Save the obdname for cleaning the nid uuids, which are
907 len = strlen(obd->obd_name) + 6;
908 OBD_ALLOC(niduuid, len);
910 strcpy(niduuid, obd->obd_name);
911 ptr = niduuid + strlen(niduuid);
914 rc = class_manual_cleanup(obd);
918 /* Clean the nid uuids */
920 GOTO(out, rc = -ENOMEM);
922 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
923 sprintf(ptr, "_%x", i);
924 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
927 CERROR("del MDC UUID %s failed: rc = %d\n",
932 OBD_FREE(niduuid, len);
934 /* class_import_put will get rid of the additional connections */
935 cfs_mutex_unlock(&mgc_start_lock);
939 /* Since there's only one mgc per node, we have to change it's fs to get
940 access to the right disk. */
941 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
943 struct lustre_sb_info *lsi = s2lsi(sb);
947 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
949 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
950 rc = obd_set_info_async(NULL, mgc->obd_self_export,
951 sizeof(KEY_SET_FS), KEY_SET_FS,
952 sizeof(*sb), sb, NULL);
954 CERROR("can't set_fs %d\n", rc);
960 static int server_mgc_clear_fs(struct obd_device *mgc)
965 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
967 rc = obd_set_info_async(NULL, mgc->obd_self_export,
968 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
973 CFS_DEFINE_MUTEX(server_start_lock);
975 /* Stop MDS/OSS if nobody is using them */
976 static int server_stop_servers(int lddflags, int lsiflags)
978 struct obd_device *obd = NULL;
979 struct obd_type *type = NULL;
983 cfs_mutex_lock(&server_start_lock);
985 /* Either an MDT or an OST or neither */
986 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
987 if ((lddflags & LDD_F_SV_TYPE_MDT) &&
988 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
989 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
990 type = class_search_type(LUSTRE_MDS_NAME);
992 /* if this was an OST, and there are no more OST's, clean up the OSS */
993 if ((lddflags & LDD_F_SV_TYPE_OST) &&
994 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
995 type = class_search_type(LUSTRE_OST_NAME);
998 if (obd && (!type || !type->typ_refcnt)) {
1001 /* obd_fail doesn't mean much on a server obd */
1002 err = class_manual_cleanup(obd);
1007 cfs_mutex_unlock(&server_start_lock);
1012 int server_mti_print(char *title, struct mgs_target_info *mti)
1014 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
1015 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
1016 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
1017 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
1018 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
1019 mti->mti_config_ver, mti->mti_flags);
1023 static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
1025 struct lustre_sb_info *lsi = s2lsi(sb);
1026 struct lustre_disk_data *ldd = lsi->lsi_ldd;
1027 lnet_process_id_t id;
1031 if (!(lsi->lsi_flags & LSI_SERVER))
1034 strncpy(mti->mti_fsname, ldd->ldd_fsname,
1035 sizeof(mti->mti_fsname));
1036 strncpy(mti->mti_svname, ldd->ldd_svname,
1037 sizeof(mti->mti_svname));
1039 mti->mti_nid_count = 0;
1040 while (LNetGetId(i++, &id) != -ENOENT) {
1041 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1044 /* server use --servicenode param, only allow specified
1045 * nids be registered */
1046 if ((ldd->ldd_flags & LDD_F_NO_PRIMNODE) != 0 &&
1047 class_match_nid(ldd->ldd_params,
1048 PARAM_FAILNODE, id.nid) < 1)
1051 /* match specified network */
1052 if (!class_match_net(ldd->ldd_params,
1053 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1056 mti->mti_nids[mti->mti_nid_count] = id.nid;
1057 mti->mti_nid_count++;
1058 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1059 CWARN("Only using first %d nids for %s\n",
1060 mti->mti_nid_count, mti->mti_svname);
1065 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1066 mti->mti_config_ver = 0;
1067 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF)
1068 ldd->ldd_flags |= LDD_F_WRITECONF;
1069 mti->mti_flags = ldd->ldd_flags;
1070 mti->mti_stripe_index = ldd->ldd_svindex;
1071 memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid));
1072 if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) {
1073 CERROR("params too big for mti\n");
1076 memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params));
1080 /* Register an old or new target with the MGS. If needed MGS will construct
1081 startup logs and assign index */
1082 int server_register_target(struct super_block *sb)
1084 struct lustre_sb_info *lsi = s2lsi(sb);
1085 struct obd_device *mgc = lsi->lsi_mgc;
1086 struct lustre_disk_data *ldd = lsi->lsi_ldd;
1087 struct mgs_target_info *mti = NULL;
1094 if (!(lsi->lsi_flags & LSI_SERVER))
1100 rc = server_sb2mti(sb, mti);
1104 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1105 mti->mti_svname, mti->mti_fsname,
1106 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1109 /* if write_conf is true, the registration must succeed */
1110 writeconf = !!(ldd->ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1111 mti->mti_flags |= LDD_F_OPC_REG;
1113 /* Register the target */
1114 /* FIXME use mgc_process_config instead */
1115 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1116 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
1117 sizeof(*mti), mti, NULL);
1119 if (mti->mti_flags & LDD_F_ERROR) {
1120 LCONSOLE_ERROR_MSG(0x160,
1121 "The MGS is refusing to allow this "
1122 "server (%s) to start. Please see messages"
1123 " on the MGS node.\n", ldd->ldd_svname);
1124 } else if (writeconf) {
1125 LCONSOLE_ERROR_MSG(0x15f,
1126 "Communication to the MGS return error %d. "
1127 "Is the MGS running?\n", rc);
1129 CERROR("Cannot talk to the MGS: %d, not fatal\n", rc);
1130 /* reset the error code for non-fatal error. */
1136 /* Always update our flags */
1137 ldd->ldd_flags = mti->mti_flags & LDD_F_ONDISK_MASK;
1139 /* If this flag is set, it means the MGS wants us to change our
1140 on-disk data. (So far this means just the index.) */
1141 if (mti->mti_flags & LDD_F_REWRITE_LDD) {
1144 CDEBUG(D_MOUNT, "Changing on-disk index from %#x to %#x "
1145 "for %s\n", ldd->ldd_svindex, mti->mti_stripe_index,
1147 ldd->ldd_svindex = mti->mti_stripe_index;
1148 strncpy(ldd->ldd_svname, mti->mti_svname,
1149 sizeof(ldd->ldd_svname));
1150 /* or ldd_make_sv_name(ldd); */
1151 ldd_write(&mgc->obd_lvfs_ctxt, ldd);
1152 if (lsi->lsi_lmd->lmd_osd_type)
1154 err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb,
1157 CERROR("Label set error %d\n", err);
1158 label = fsfilt_get_label(mgc, lsi->lsi_srv_mnt->mnt_sb);
1160 CDEBUG(D_MOUNT, "Disk label changed to %s\n", label);
1162 /* Flush the new ldd to disk */
1163 fsfilt_sync(mgc, lsi->lsi_srv_mnt->mnt_sb);
1173 * Notify the MGS that this target is ready.
1174 * Used by IR - if the MGS receives this message, it will notify clients.
1176 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1178 struct lustre_sb_info *lsi = s2lsi(sb);
1179 struct obd_device *mgc = lsi->lsi_mgc;
1180 struct mgs_target_info *mti = NULL;
1186 if (!(lsi->lsi_flags & LSI_SERVER))
1192 rc = server_sb2mti(sb, mti);
1196 mti->mti_instance = obd->u.obt.obt_instance;
1197 mti->mti_flags |= LDD_F_OPC_READY;
1199 /* FIXME use mgc_process_config instead */
1200 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1201 sizeof(KEY_REGISTER_TARGET),
1202 KEY_REGISTER_TARGET,
1203 sizeof(*mti), mti, NULL);
1205 /* Imperative recovery: if the mgs informs us to use IR? */
1206 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1207 (mti->mti_flags & LDD_F_IR_CAPABLE))
1208 lsi->lsi_flags |= LSI_IR_CAPABLE;
1217 /** Start server targets: MDTs and OSTs
1219 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1221 struct obd_device *obd;
1222 struct lustre_sb_info *lsi = s2lsi(sb);
1223 struct config_llog_instance cfg;
1227 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname);
1230 /* If we're an MDT, make sure the global MDS is running */
1231 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1232 /* make sure the MDS is started */
1233 cfs_mutex_lock(&server_start_lock);
1234 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1236 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1237 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1239 LUSTRE_MDS_OBDNAME"_uuid",
1242 cfs_mutex_unlock(&server_start_lock);
1243 CERROR("failed to start MDS: %d\n", rc);
1247 cfs_mutex_unlock(&server_start_lock);
1251 /* If we're an OST, make sure the global OSS is running */
1252 if (IS_OST(lsi->lsi_ldd)) {
1253 /* make sure OSS is started */
1254 cfs_mutex_lock(&server_start_lock);
1255 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1257 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1259 LUSTRE_OSS_OBDNAME"_uuid",
1262 cfs_mutex_unlock(&server_start_lock);
1263 CERROR("failed to start OSS: %d\n", rc);
1267 cfs_mutex_unlock(&server_start_lock);
1270 /* Set the mgc fs to our server disk. This allows the MGC to
1271 * read and write configs locally, in case it can't talk to the MGS. */
1272 if (lsi->lsi_lmd->lmd_osd_type == NULL) {
1273 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1278 /* Register with MGS */
1279 rc = server_register_target(sb);
1283 /* Let the target look up the mount using the target's name
1284 (we can't pass the sb or mnt through class_process_config.) */
1285 rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
1289 /* Start targets using the llog named for the target */
1290 memset(&cfg, 0, sizeof(cfg));
1291 rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg);
1293 CERROR("failed to start server %s: %d\n",
1294 lsi->lsi_ldd->ldd_svname, rc);
1295 /* Do NOT call server_deregister_mount() here. This makes it
1296 * impossible to find mount later in cleanup time and leaves
1297 * @lsi and othder stuff leaked. -umka */
1302 /* Release the mgc fs for others to use */
1303 if (lsi->lsi_lmd->lmd_osd_type == NULL)
1304 server_mgc_clear_fs(lsi->lsi_mgc);
1307 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1309 CERROR("no server named %s was started\n",
1310 lsi->lsi_ldd->ldd_svname);
1314 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1315 (OBP(obd, iocontrol))) {
1316 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1317 obd->obd_self_export, 0, NULL, NULL);
1320 server_notify_target(sb, obd);
1322 /* calculate recovery timeout, do it after lustre_process_log */
1323 server_calc_timeout(lsi, obd);
1325 /* log has been fully processed */
1326 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1332 /***************** lustre superblock **************/
1334 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1336 struct lustre_sb_info *lsi;
1342 OBD_ALLOC_PTR(lsi->lsi_lmd);
1343 if (!lsi->lsi_lmd) {
1348 lsi->lsi_lmd->lmd_exclude_count = 0;
1349 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
1350 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
1351 s2lsi_nocast(sb) = lsi;
1352 /* we take 1 extra ref for our setup */
1353 cfs_atomic_set(&lsi->lsi_mounts, 1);
1355 /* Default umount style */
1356 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1361 static int lustre_free_lsi(struct super_block *sb)
1363 struct lustre_sb_info *lsi = s2lsi(sb);
1366 LASSERT(lsi != NULL);
1367 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
1369 /* someone didn't call server_put_mount. */
1370 LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0);
1372 if (lsi->lsi_ldd != NULL)
1373 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
1375 if (lsi->lsi_lmd != NULL) {
1376 if (lsi->lsi_lmd->lmd_dev != NULL)
1377 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1378 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1379 if (lsi->lsi_lmd->lmd_profile != NULL)
1380 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1381 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1382 if (lsi->lsi_lmd->lmd_mgssec != NULL)
1383 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
1384 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
1385 if (lsi->lsi_lmd->lmd_opts != NULL)
1386 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1387 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1388 if (lsi->lsi_lmd->lmd_exclude_count)
1389 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1390 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1391 lsi->lsi_lmd->lmd_exclude_count);
1392 if (lsi->lsi_lmd->lmd_mgs != NULL)
1393 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
1394 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
1395 if (lsi->lsi_lmd->lmd_osd_type != NULL)
1396 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
1397 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
1399 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1402 LASSERT(lsi->lsi_llsbi == NULL);
1403 OBD_FREE(lsi, sizeof(*lsi));
1404 s2lsi_nocast(sb) = NULL;
1409 /* The lsi has one reference for every server that is using the disk -
1410 e.g. MDT, MGS, and potentially MGC */
1411 static int lustre_put_lsi(struct super_block *sb)
1413 struct lustre_sb_info *lsi = s2lsi(sb);
1416 LASSERT(lsi != NULL);
1418 CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts));
1419 if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) {
1420 lustre_free_lsi(sb);
1426 static int lsi_prepare(struct lustre_sb_info *lsi)
1428 struct lustre_disk_data *ldd;
1434 LASSERT(lsi->lsi_lmd);
1436 OBD_ALLOC(ldd, sizeof(*ldd));
1439 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_NAME);
1441 /* The server name is given as a mount line option */
1442 if (lsi->lsi_lmd->lmd_profile == NULL) {
1443 LCONSOLE_ERROR("Can't determine server name\n");
1447 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(ldd->ldd_svname))
1448 RETURN(-ENAMETOOLONG);
1450 strcpy(ldd->ldd_svname, lsi->lsi_lmd->lmd_profile);
1451 strcpy(ldd->ldd_fsname, "lustre");
1453 /* Determine osd type */
1454 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
1455 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
1456 sizeof(lsi->lsi_osd_type))
1457 RETURN (-ENAMETOOLONG);
1459 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
1462 /* Determine server type */
1463 rc = server_name2index(ldd->ldd_svname, &index, NULL);
1465 if (0 /*lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS*/) {
1466 /* Assume we're a bare MGS */
1468 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
1470 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
1475 ldd->ldd_svindex = index;
1476 //lsi->lsi_flags |= rc;
1477 ldd->ldd_flags = rc | LDD_F_WRITECONF;
1481 /* Add mount line flags that used to be in ldd:
1482 * writeconf, mgs, iam, anything else?
1485 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
1486 LDD_F_WRITECONF : 0;
1487 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
1488 LDD_F_SV_TYPE_MGS : 0;
1489 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_IAM) ?
1491 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
1492 LDD_F_NO_PRIMNODE : 0;
1498 /*************** server mount ******************/
1500 /** Kernel mount using mount options in MOUNT_DATA_FILE.
1501 * Since this file lives on the disk, we pre-mount using a common
1502 * type, read the file, then re-mount using the type specified in the
1505 static struct vfsmount *server_kernel_mount(struct super_block *sb)
1507 struct lvfs_run_ctxt mount_ctxt;
1508 struct lustre_sb_info *lsi = s2lsi(sb);
1509 struct lustre_disk_data *ldd;
1510 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1511 struct vfsmount *mnt;
1512 struct file_system_type *type;
1513 char *options = NULL;
1514 unsigned long page, s_flags;
1515 struct page *__page;
1520 if (lsi->lsi_lmd->lmd_osd_type) {
1521 rc = lsi_prepare(lsi);
1522 RETURN(ERR_PTR(rc));
1525 OBD_ALLOC(ldd, sizeof(*ldd));
1527 RETURN(ERR_PTR(-ENOMEM));
1528 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_NAME);
1530 /* In the past, we have always used flags = 0.
1531 Note ext3/ldiskfs can't be mounted ro. */
1532 s_flags = sb->s_flags;
1534 /* allocate memory for options */
1535 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
1537 GOTO(out_free, rc = -ENOMEM);
1538 page = (unsigned long)cfs_page_address(__page);
1539 options = (char *)page;
1540 memset(options, 0, CFS_PAGE_SIZE);
1542 /* mount-line options must be added for pre-mount because it may
1543 * contain mount options such as journal_dev which are required
1544 * to mount successfuly the underlying filesystem */
1545 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0))
1546 strncat(options, lmd->lmd_opts, CFS_PAGE_SIZE - 1);
1548 /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */
1549 CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev);
1550 type = get_fs_type("ldiskfs");
1552 CERROR("premount failed: cannot find ldiskfs module\n");
1553 GOTO(out_free, rc = -ENODEV);
1555 mnt = vfs_kern_mount(type, s_flags, lmd->lmd_dev, (void *)options);
1556 cfs_module_put(type->owner);
1559 CERROR("premount %s:%#lx ldiskfs failed: %d "
1560 "Is the ldiskfs module available?\n",
1561 lmd->lmd_dev, s_flags, rc );
1565 OBD_SET_CTXT_MAGIC(&mount_ctxt);
1566 mount_ctxt.pwdmnt = mnt;
1567 mount_ctxt.pwd = mnt->mnt_root;
1568 mount_ctxt.fs = get_ds();
1570 rc = ldd_parse(&mount_ctxt, ldd);
1574 CERROR("premount parse options failed: rc = %d\n", rc);
1578 /* Done with our pre-mount, now do the real mount. */
1580 /* Glom up mount options */
1581 memset(options, 0, CFS_PAGE_SIZE);
1582 strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2);
1584 len = CFS_PAGE_SIZE - strlen(options) - 2;
1586 strcat(options, ",");
1587 strncat(options, "no_mbcache", len);
1589 /* Add in any mount-line options */
1590 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) {
1591 len = CFS_PAGE_SIZE - strlen(options) - 2;
1592 strcat(options, ",");
1593 strncat(options, lmd->lmd_opts, len);
1596 /* Special permanent mount flags */
1598 s_flags |= MS_NOATIME | MS_NODIRATIME;
1600 CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
1601 MT_STR(ldd), lmd->lmd_dev, options);
1602 type = get_fs_type(MT_STR(ldd));
1604 CERROR("get_fs_type failed\n");
1605 GOTO(out_free, rc = -ENODEV);
1607 mnt = vfs_kern_mount(type, s_flags, lmd->lmd_dev, (void *)options);
1608 cfs_module_put(type->owner);
1611 CERROR("vfs_kern_mount failed: rc = %d\n", rc);
1615 if (lmd->lmd_flags & LMD_FLG_ABORT_RECOV)
1616 simple_truncate(mnt->mnt_sb->s_root, mnt, LAST_RCVD,
1619 OBD_PAGE_FREE(__page);
1620 lsi->lsi_ldd = ldd; /* freed at lsi cleanup */
1621 CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
1626 OBD_PAGE_FREE(__page);
1627 OBD_FREE(ldd, sizeof(*ldd));
1628 lsi->lsi_ldd = NULL;
1629 RETURN(ERR_PTR(rc));
1632 /** Wait here forever until the mount refcount is 0 before completing umount,
1633 * else we risk dereferencing a null pointer.
1634 * LNET may take e.g. 165s before killing zombies.
1636 static void server_wait_finished(struct vfsmount *mnt)
1640 cfs_sigset_t blocked;
1643 cfs_waitq_init(&waitq);
1644 cfs_waitq_wait_event_interruptible_timeout(waitq, 0,
1645 cfs_time_seconds(3), rc);
1650 cfs_waitq_init(&waitq);
1652 while (mnt_get_count(mnt) > 1) {
1653 if (waited && (waited % 30 == 0))
1654 LCONSOLE_WARN("Mount still busy with %d refs after "
1658 /* Cannot use l_event_wait() for an interruptible sleep. */
1660 blocked = cfs_block_sigsinv(sigmask(SIGKILL));
1661 cfs_waitq_wait_event_interruptible_timeout(
1663 (mnt_get_count(mnt) == 1),
1664 cfs_time_seconds(3),
1666 cfs_restore_sigs(blocked);
1668 LCONSOLE_EMERG("Danger: interrupted umount %s with "
1669 "%d refs!\n", mnt->mnt_devname,
1670 mnt_get_count(mnt));
1677 /** Start the shutdown of servers at umount.
1679 static void server_put_super(struct super_block *sb)
1681 struct lustre_sb_info *lsi = s2lsi(sb);
1682 struct obd_device *obd;
1683 struct vfsmount *mnt = lsi->lsi_srv_mnt;
1684 char *tmpname, *extraname = NULL;
1686 int lddflags = lsi->lsi_ldd->ldd_flags;
1687 int lsiflags = lsi->lsi_flags;
1690 LASSERT(lsiflags & LSI_SERVER);
1692 tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1;
1693 OBD_ALLOC(tmpname, tmpname_sz);
1694 memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
1695 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1696 if (IS_MDT(lsi->lsi_ldd) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1697 snprintf(tmpname, tmpname_sz, "MGS");
1699 /* Stop the target */
1700 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1701 (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd))) {
1702 struct lustre_profile *lprof = NULL;
1704 /* tell the mgc to drop the config log */
1705 lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL);
1707 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1708 If there are any setup/cleanup errors, save the lov
1709 name for safety cleanup later. */
1710 lprof = class_get_profile(lsi->lsi_ldd->ldd_svname);
1711 if (lprof && lprof->lp_dt) {
1712 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1713 strcpy(extraname, lprof->lp_dt);
1716 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1718 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1719 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1721 /* We can't seem to give an error return code
1722 * to .put_super, so we better make sure we clean up! */
1724 class_manual_cleanup(obd);
1726 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
1727 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1731 /* If they wanted the mgs to stop separately from the mdt, they
1732 should have put it on a different device. */
1733 if (IS_MGS(lsi->lsi_ldd)) {
1734 /* if MDS start with --nomgs, don't stop MGS then */
1735 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1736 server_stop_mgs(sb);
1739 /* Clean the mgc and sb */
1740 lustre_common_put_super(sb);
1742 /* Wait for the targets to really clean up - can't exit (and let the
1743 sb get destroyed) while the mount is still in use */
1744 server_wait_finished(mnt);
1746 /* drop the One True Mount */
1750 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1751 until the target is really gone so that our type refcount check
1753 server_stop_servers(lddflags, lsiflags);
1755 /* In case of startup or cleanup err, stop related obds */
1757 obd = class_name2obd(extraname);
1759 CWARN("Cleaning orphaned obd %s\n", extraname);
1761 class_manual_cleanup(obd);
1763 OBD_FREE(extraname, strlen(extraname) + 1);
1766 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1767 OBD_FREE(tmpname, tmpname_sz);
1771 /** Called only for 'umount -f'
1773 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1774 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1776 struct super_block *sb = vfsmnt->mnt_sb;
1778 static void server_umount_begin(struct super_block *sb)
1781 struct lustre_sb_info *lsi = s2lsi(sb);
1784 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1785 if (!(flags & MNT_FORCE)) {
1791 CDEBUG(D_MOUNT, "umount -f\n");
1792 /* umount = failover
1794 no third way to do non-force, non-failover */
1795 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1796 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1800 #ifndef HAVE_STATFS_DENTRY_PARAM
1801 static int server_statfs (struct super_block *sb, cfs_kstatfs_t *buf)
1804 static int server_statfs (struct dentry *dentry, cfs_kstatfs_t *buf)
1806 struct super_block *sb = dentry->d_sb;
1808 struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt;
1811 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) {
1812 #ifdef HAVE_STATFS_DENTRY_PARAM
1813 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf);
1815 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_sb, buf);
1818 buf->f_type = sb->s_magic;
1824 buf->f_type = sb->s_magic;
1825 buf->f_bsize = sb->s_blocksize;
1831 buf->f_namelen = NAME_MAX;
1835 /** The operations we support directly on the superblock:
1836 * mount, umount, and df.
1838 static struct super_operations server_ops =
1840 .put_super = server_put_super,
1841 .umount_begin = server_umount_begin, /* umount -f */
1842 .statfs = server_statfs,
1845 #define log2(n) cfs_ffz(~(n))
1846 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1848 static int server_fill_super_common(struct super_block *sb)
1850 struct inode *root = 0;
1853 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1855 sb->s_blocksize = 4096;
1856 sb->s_blocksize_bits = log2(sb->s_blocksize);
1857 sb->s_magic = LUSTRE_SUPER_MAGIC;
1858 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1859 sb->s_flags |= MS_RDONLY;
1860 sb->s_op = &server_ops;
1862 root = new_inode(sb);
1864 CERROR("Can't make root inode\n");
1868 /* returns -EIO for every operation */
1869 /* make_bad_inode(root); -- badness - can't umount */
1870 /* apparently we need to be a directory for the mount to finish */
1871 root->i_mode = S_IFDIR;
1873 sb->s_root = d_alloc_root(root);
1875 CERROR("Can't make root dentry\n");
1883 /** Fill in the superblock info for a Lustre server.
1884 * Mount the device with the correct options.
1885 * Read the on-disk config file.
1886 * Start the services.
1888 static int server_fill_super(struct super_block *sb)
1890 struct lustre_sb_info *lsi = s2lsi(sb);
1891 struct vfsmount *mnt;
1895 /* the One True Mount */
1896 mnt = server_kernel_mount(sb);
1899 CERROR("Unable to mount device %s: %d\n",
1900 lsi->lsi_lmd->lmd_dev, rc);
1904 lsi->lsi_srv_mnt = mnt;
1906 LASSERT(lsi->lsi_ldd);
1907 CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n",
1908 lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname,
1909 lsi->lsi_lmd->lmd_dev);
1911 if (class_name2obd(lsi->lsi_ldd->ldd_svname)) {
1912 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1913 "running. Double-mount may have compromised"
1914 " the disk journal.\n",
1915 lsi->lsi_ldd->ldd_svname);
1921 /* Start MGS before MGC */
1922 if (IS_MGS(lsi->lsi_ldd) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
1923 rc = server_start_mgs(sb);
1928 /* Start MGC before servers */
1929 rc = lustre_start_mgc(sb);
1933 /* Set up all obd devices for service */
1934 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1935 (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) {
1936 rc = server_start_targets(sb, mnt);
1938 CERROR("Unable to start targets: %d\n", rc);
1941 /* FIXME overmount client here,
1942 or can we just start a client log and client_fill_super on this sb?
1943 We need to make sure server_put_super gets called too - ll_put_super
1944 calls lustre_common_put_super; check there for LSI_SERVER flag,
1946 Probably should start client from new thread so we can return.
1947 Client will not finish until all servers are connected.
1948 Note - MGS-only server does NOT get a client, since there is no
1949 lustre fs associated - the MGS is for all lustre fs's */
1952 rc = server_fill_super_common(sb);
1958 /* We jump here in case of failure while starting targets or MGS.
1959 * In this case we can't just put @mnt and have to do real cleanup
1960 * with stoping targets, etc. */
1961 server_put_super(sb);
1965 /* Get the index from the obd name.
1966 rc = server type, or
1968 if endptr isn't NULL it is set to end of name */
1969 int server_name2index(char *svname, __u32 *idx, char **endptr)
1971 unsigned long index;
1973 char *dash = strrchr(svname, '-');
1977 /* intepret <fsname>-MDTXXXXX-mdc as mdt, the better way is to pass
1978 * in the fsname, then determine the server index */
1979 if (!strcmp(LUSTRE_MDC_NAME, dash + 1)) {
1981 for (; dash > svname && *dash != '-'; dash--);
1986 if (strncmp(dash + 1, "MDT", 3) == 0)
1987 rc = LDD_F_SV_TYPE_MDT;
1988 else if (strncmp(dash + 1, "OST", 3) == 0)
1989 rc = LDD_F_SV_TYPE_OST;
1992 if (strcmp(dash + 4, "all") == 0)
1993 return rc | LDD_F_SV_ALL;
1995 index = simple_strtoul(dash + 4, endptr, 16);
2001 * Calculate timeout value for a target.
2003 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
2005 struct lustre_mount_data *lmd;
2009 bool has_ir = !!(lsi->lsi_flags & LSI_IR_CAPABLE);
2010 int min = OBD_RECOVERY_TIME_MIN;
2012 LASSERT(lsi->lsi_flags & LSI_SERVER);
2016 soft = lmd->lmd_recovery_time_soft;
2017 hard = lmd->lmd_recovery_time_hard;
2018 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
2019 obd->obd_no_ir = !has_ir;
2023 soft = OBD_RECOVERY_TIME_SOFT;
2025 hard = OBD_RECOVERY_TIME_HARD;
2027 /* target may have ir_factor configured. */
2028 factor = OBD_IR_FACTOR_DEFAULT;
2029 if (obd->obd_recovery_ir_factor)
2030 factor = obd->obd_recovery_ir_factor;
2033 int new_soft = soft;
2034 int new_hard = hard;
2036 /* adjust timeout value by imperative recovery */
2038 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
2039 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
2041 /* make sure the timeout is not too short */
2042 new_soft = max(min, new_soft);
2043 new_hard = max(new_soft, new_hard);
2045 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
2046 "window shrunk from %d-%d down to %d-%d\n",
2047 obd->obd_name, soft, hard, new_soft, new_hard);
2054 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
2055 obd->obd_recovery_time_hard = hard;
2056 obd->obd_recovery_ir_factor = factor;
2058 EXPORT_SYMBOL(server_calc_timeout);
2060 /*************** mount common betweeen server and client ***************/
2063 int lustre_common_put_super(struct super_block *sb)
2068 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
2070 /* Drop a ref to the MGC */
2071 rc = lustre_stop_mgc(sb);
2072 if (rc && (rc != -ENOENT)) {
2074 CERROR("Can't stop MGC: %d\n", rc);
2077 /* BUSY just means that there's some other obd that
2078 needs the mgc. Let him clean it up. */
2079 CDEBUG(D_MOUNT, "MGC still in use\n");
2081 /* Drop a ref to the mounted disk */
2087 static void lmd_print(struct lustre_mount_data *lmd)
2091 PRINT_CMD(PRINT_MASK, " mount data:\n");
2092 if (lmd_is_client(lmd))
2093 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
2094 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
2095 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
2098 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
2100 if (lmd->lmd_recovery_time_soft)
2101 PRINT_CMD(PRINT_MASK, "recovery time soft: %d\n",
2102 lmd->lmd_recovery_time_soft);
2104 if (lmd->lmd_recovery_time_hard)
2105 PRINT_CMD(PRINT_MASK, "recovery time hard: %d\n",
2106 lmd->lmd_recovery_time_hard);
2108 for (i = 0; i < lmd->lmd_exclude_count; i++) {
2109 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
2110 lmd->lmd_exclude[i]);
2114 /* Is this server on the exclusion list */
2115 int lustre_check_exclusion(struct super_block *sb, char *svname)
2117 struct lustre_sb_info *lsi = s2lsi(sb);
2118 struct lustre_mount_data *lmd = lsi->lsi_lmd;
2123 rc = server_name2index(svname, &index, NULL);
2124 if (rc != LDD_F_SV_TYPE_OST)
2125 /* Only exclude OSTs */
2128 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
2129 index, lmd->lmd_exclude_count, lmd->lmd_dev);
2131 for(i = 0; i < lmd->lmd_exclude_count; i++) {
2132 if (index == lmd->lmd_exclude[i]) {
2133 CWARN("Excluding %s (on exclusion list)\n", svname);
2140 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
2141 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
2143 char *s1 = ptr, *s2;
2144 __u32 index, *exclude_list;
2148 /* The shortest an ost name can be is 8 chars: -OST0000.
2149 We don't actually know the fsname at this time, so in fact
2150 a user could specify any fsname. */
2151 devmax = strlen(ptr) / 8 + 1;
2153 /* temp storage until we figure out how many we have */
2154 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
2158 /* we enter this fn pointing at the '=' */
2159 while (*s1 && *s1 != ' ' && *s1 != ',') {
2161 rc = server_name2index(s1, &index, &s2);
2163 CERROR("Can't parse server name '%s'\n", s1);
2166 if (rc == LDD_F_SV_TYPE_OST)
2167 exclude_list[lmd->lmd_exclude_count++] = index;
2169 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
2171 /* now we are pointing at ':' (next exclude)
2172 or ',' (end of excludes) */
2173 if (lmd->lmd_exclude_count >= devmax)
2176 if (rc >= 0) /* non-err */
2179 if (lmd->lmd_exclude_count) {
2180 /* permanent, freed in lustre_free_lsi */
2181 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
2182 lmd->lmd_exclude_count);
2183 if (lmd->lmd_exclude) {
2184 memcpy(lmd->lmd_exclude, exclude_list,
2185 sizeof(index) * lmd->lmd_exclude_count);
2188 lmd->lmd_exclude_count = 0;
2191 OBD_FREE(exclude_list, sizeof(index) * devmax);
2195 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
2200 if (lmd->lmd_mgssec != NULL) {
2201 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
2202 lmd->lmd_mgssec = NULL;
2205 tail = strchr(ptr, ',');
2207 length = strlen(ptr);
2209 length = tail - ptr;
2211 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
2212 if (lmd->lmd_mgssec == NULL)
2215 memcpy(lmd->lmd_mgssec, ptr, length);
2216 lmd->lmd_mgssec[length] = '\0';
2220 static int lmd_parse_string(char **handle, char *ptr)
2225 if ((handle == NULL) || (ptr == NULL))
2228 if (*handle != NULL) {
2229 OBD_FREE(*handle, strlen(*handle) + 1);
2233 tail = strchr(ptr, ',');
2235 length = strlen(ptr);
2237 length = tail - ptr;
2239 OBD_ALLOC(*handle, length + 1);
2240 if (*handle == NULL)
2243 memcpy(*handle, ptr, length);
2244 (*handle)[length] = '\0';
2249 /* Collect multiple values for mgsnid specifiers */
2250 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
2258 /* Find end of nidlist */
2259 while (class_parse_nid(tail, &nid, &tail) == 0) {}
2260 length = tail - *ptr;
2262 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
2266 if (lmd->lmd_mgs != NULL)
2267 oldlen = strlen(lmd->lmd_mgs) + 1;
2269 OBD_ALLOC(mgsnid, oldlen + length + 1);
2273 if (lmd->lmd_mgs != NULL) {
2274 /* Multiple mgsnid= are taken to mean failover locations */
2275 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
2276 mgsnid[oldlen - 1] = ':';
2277 OBD_FREE(lmd->lmd_mgs, oldlen);
2279 memcpy(mgsnid + oldlen, *ptr, length);
2280 mgsnid[oldlen + length] = '\0';
2281 lmd->lmd_mgs = mgsnid;
2287 /** Parse mount line options
2288 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
2289 * dev is passed as device=uml1:/lustre by mount.lustre
2291 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
2293 char *s1, *s2, *devname = NULL;
2294 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
2300 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
2301 "/sbin/mount.lustre is installed.\n");
2305 /* Options should be a string - try to detect old lmd data */
2306 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
2307 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
2308 "/sbin/mount.lustre. Please install "
2309 "version %s\n", LUSTRE_VERSION_STRING);
2312 lmd->lmd_magic = LMD_MAGIC;
2314 /* Set default flags here */
2319 int time_min = OBD_RECOVERY_TIME_MIN;
2321 /* Skip whitespace and extra commas */
2322 while (*s1 == ' ' || *s1 == ',')
2325 /* Client options are parsed in ll_options: eg. flock,
2328 /* Parse non-ldiskfs options here. Rather than modifying
2329 ldiskfs, we just zero these out here */
2330 if (strncmp(s1, "abort_recov", 11) == 0) {
2331 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
2333 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
2334 lmd->lmd_recovery_time_soft = max_t(int,
2335 simple_strtoul(s1 + 19, NULL, 10), time_min);
2337 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
2338 lmd->lmd_recovery_time_hard = max_t(int,
2339 simple_strtoul(s1 + 19, NULL, 10), time_min);
2341 } else if (strncmp(s1, "noir", 4) == 0) {
2342 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
2344 } else if (strncmp(s1, "nosvc", 5) == 0) {
2345 lmd->lmd_flags |= LMD_FLG_NOSVC;
2347 } else if (strncmp(s1, "nomgs", 5) == 0) {
2348 lmd->lmd_flags |= LMD_FLG_NOMGS;
2350 } else if (strncmp(s1, "noscrub", 7) == 0) {
2351 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
2353 } else if (strncmp(s1, PARAM_MGSNODE,
2354 sizeof(PARAM_MGSNODE) - 1) == 0) {
2355 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
2356 /* Assume the next mount opt is the first
2357 invalid nid we get to. */
2358 rc = lmd_parse_mgs(lmd, &s2);
2362 } else if (strncmp(s1, "writeconf", 9) == 0) {
2363 lmd->lmd_flags |= LMD_FLG_WRITECONF;
2365 } else if (strncmp(s1, "mgssec=", 7) == 0) {
2366 rc = lmd_parse_mgssec(lmd, s1 + 7);
2370 /* ost exclusion list */
2371 } else if (strncmp(s1, "exclude=", 8) == 0) {
2372 rc = lmd_make_exclusion(lmd, s1 + 7);
2376 } else if (strncmp(s1, "svname=", 7) == 0) {
2377 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
2381 } else if (strncmp(s1, "osd=", 4) == 0) {
2382 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
2385 /* with ldiskfs we're still doing ldd parsing
2386 * in the kernel space */
2387 if (!strcmp(lmd->lmd_osd_type, "osd-ldiskfs")) {
2388 OBD_FREE(lmd->lmd_osd_type,
2389 strlen(lmd->lmd_osd_type) + 1);
2390 lmd->lmd_osd_type = NULL;
2394 /* Linux 2.4 doesn't pass the device, so we stuck it at the
2395 end of the options. */
2396 else if (strncmp(s1, "device=", 7) == 0) {
2398 /* terminate options right before device. device
2399 must be the last one. */
2405 s2 = strchr(s1, ',');
2413 memmove(s1, s2, strlen(s2) + 1);
2419 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
2420 "(need mount option 'device=...')\n");
2424 s1 = strstr(devname, ":/");
2427 lmd->lmd_flags |= LMD_FLG_CLIENT;
2428 /* Remove leading /s from fsname */
2429 while (*++s1 == '/') ;
2430 /* Freed in lustre_free_lsi */
2431 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
2432 if (!lmd->lmd_profile)
2434 sprintf(lmd->lmd_profile, "%s-client", s1);
2437 /* Freed in lustre_free_lsi */
2438 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
2441 strcpy(lmd->lmd_dev, devname);
2443 /* Save mount options */
2444 s1 = options + strlen(options) - 1;
2445 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
2447 if (*options != 0) {
2448 /* Freed in lustre_free_lsi */
2449 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
2452 strcpy(lmd->lmd_opts, options);
2456 lmd->lmd_magic = LMD_MAGIC;
2461 CERROR("Bad mount options %s\n", options);
2465 struct lustre_mount_data2 {
2467 struct vfsmount *lmd2_mnt;
2470 /** This is the entry point for the mount call into Lustre.
2471 * This is called when a server or client is mounted,
2472 * and this is where we start setting things up.
2473 * @param data Mount options (e.g. -o flock,abort_recov)
2475 int lustre_fill_super(struct super_block *sb, void *data, int silent)
2477 struct lustre_mount_data *lmd;
2478 struct lustre_mount_data2 *lmd2 = data;
2479 struct lustre_sb_info *lsi;
2483 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
2485 lsi = lustre_init_lsi(sb);
2491 * Disable lockdep during mount, because mount locking patterns are
2497 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
2499 obd_zombie_barrier();
2501 /* Figure out the lmd from the mount options */
2502 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
2504 GOTO(out, rc = -EINVAL);
2507 if (lmd_is_client(lmd)) {
2508 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
2509 if (!client_fill_super) {
2510 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
2511 "client mount! Is the 'lustre' "
2512 "module loaded?\n");
2516 rc = lustre_start_mgc(sb);
2521 /* Connect and start */
2522 /* (should always be ll_fill_super) */
2523 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
2524 /* c_f_s will call lustre_common_put_super on failure */
2527 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
2528 lsi->lsi_flags |= LSI_SERVER;
2529 rc = server_fill_super(sb);
2530 /* s_f_s calls lustre_start_mgc after the mount because we need
2531 the MGS nids which are stored on disk. Plus, we may
2532 need to start the MGS first. */
2533 /* s_f_s will call server_put_super on failure */
2536 /* If error happens in fill_super() call, @lsi will be killed there.
2537 * This is why we do not put it here. */
2541 CERROR("Unable to mount %s (%d)\n",
2542 s2lsi(sb) ? lmd->lmd_dev : "", rc);
2544 CDEBUG(D_SUPER, "Mount %s complete\n",
2552 /* We can't call ll_fill_super by name because it lives in a module that
2553 must be loaded after this one. */
2554 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
2555 struct vfsmount *mnt))
2557 client_fill_super = cfs;
2560 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
2562 kill_super_cb = cfs;
2565 /***************** FS registration ******************/
2567 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))
2568 struct super_block * lustre_get_sb(struct file_system_type *fs_type, int flags,
2569 const char *devname, void * data)
2571 return get_sb_nodev(fs_type, flags, data, lustre_fill_super);
2574 int lustre_get_sb(struct file_system_type *fs_type, int flags,
2575 const char *devname, void * data, struct vfsmount *mnt)
2577 struct lustre_mount_data2 lmd2 = {data, mnt};
2579 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
2583 void lustre_kill_super(struct super_block *sb)
2585 struct lustre_sb_info *lsi = s2lsi(sb);
2587 if (kill_super_cb && lsi && !(lsi->lsi_flags & LSI_SERVER))
2588 (*kill_super_cb)(sb);
2590 kill_anon_super(sb);
2593 /** Register the "lustre" fs type
2595 struct file_system_type lustre_fs_type = {
2596 .owner = THIS_MODULE,
2598 .get_sb = lustre_get_sb,
2599 .kill_sb = lustre_kill_super,
2600 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
2601 #ifdef FS_HAS_FIEMAP
2604 LL_RENAME_DOES_D_MOVE,
2607 int lustre_register_fs(void)
2609 return register_filesystem(&lustre_fs_type);
2612 int lustre_unregister_fs(void)
2614 return unregister_filesystem(&lustre_fs_type);
2617 EXPORT_SYMBOL(lustre_register_client_fill_super);
2618 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2619 EXPORT_SYMBOL(lustre_common_put_super);
2620 EXPORT_SYMBOL(lustre_process_log);
2621 EXPORT_SYMBOL(lustre_end_log);
2622 EXPORT_SYMBOL(server_get_mount);
2623 EXPORT_SYMBOL(server_get_mount_2);
2624 EXPORT_SYMBOL(server_put_mount);
2625 EXPORT_SYMBOL(server_put_mount_2);
2626 EXPORT_SYMBOL(server_register_target);
2627 EXPORT_SYMBOL(server_name2index);
2628 EXPORT_SYMBOL(server_mti_print);
2629 EXPORT_SYMBOL(do_lcfg);