1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * lustre/obdclass/obd_mount.c
5 * Client/server mount routines
7 * Copyright (c) 2006 Cluster File Systems, Inc.
8 * Author: Nathan Rutman <nathan@clusterfs.com>
10 * This file is part of Lustre, http://www.lustre.org/
12 * Lustre is free software; you can redistribute it and/or
13 * modify it under the terms of version 2 of the GNU General Public
14 * License as published by the Free Software Foundation.
16 * Lustre is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
19 * GNU General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with Lustre; if not, write to the Free Software
23 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
27 #define DEBUG_SUBSYSTEM S_CLASS
28 #define D_MOUNT D_SUPER|D_CONFIG /*|D_WARNING */
29 #define PRINT_CMD CDEBUG
30 #define PRINT_MASK D_SUPER|D_CONFIG
34 #include <lustre_fsfilt.h>
35 #include <obd_class.h>
36 #include <lustre/lustre_user.h>
37 #include <linux/version.h>
38 #include <lustre_log.h>
39 #include <lustre_disk.h>
40 #include <lustre_param.h>
42 static int (*client_fill_super)(struct super_block *sb) = NULL;
43 static void (*kill_super_cb)(struct super_block *sb) = NULL;
45 /*********** mount lookup *********/
47 DECLARE_MUTEX(lustre_mount_info_lock);
48 struct list_head server_mount_info_list = LIST_HEAD_INIT(server_mount_info_list);
50 static struct lustre_mount_info *server_find_mount(const char *name)
52 struct list_head *tmp;
53 struct lustre_mount_info *lmi;
56 list_for_each(tmp, &server_mount_info_list) {
57 lmi = list_entry(tmp, struct lustre_mount_info, lmi_list_chain);
58 if (strcmp(name, lmi->lmi_name) == 0)
64 /* we must register an obd for a mount before we call the setup routine.
65 *_setup will call lustre_get_mount to get the mnt struct
66 by obd_name, since we can't pass the pointer to setup. */
67 static int server_register_mount(const char *name, struct super_block *sb,
70 struct lustre_mount_info *lmi;
77 OBD_ALLOC(lmi, sizeof(*lmi));
80 OBD_ALLOC(name_cp, strlen(name) + 1);
82 OBD_FREE(lmi, sizeof(*lmi));
85 strcpy(name_cp, name);
87 down(&lustre_mount_info_lock);
89 if (server_find_mount(name)) {
90 up(&lustre_mount_info_lock);
91 OBD_FREE(lmi, sizeof(*lmi));
92 OBD_FREE(name_cp, strlen(name) + 1);
93 CERROR("Already registered %s\n", name);
96 lmi->lmi_name = name_cp;
99 list_add(&lmi->lmi_list_chain, &server_mount_info_list);
101 up(&lustre_mount_info_lock);
103 CDEBUG(D_MOUNT, "reg_mnt %p from %s, vfscount=%d\n",
104 lmi->lmi_mnt, name, atomic_read(&lmi->lmi_mnt->mnt_count));
109 /* when an obd no longer needs a mount */
110 static int server_deregister_mount(const char *name)
112 struct lustre_mount_info *lmi;
115 down(&lustre_mount_info_lock);
116 lmi = server_find_mount(name);
118 up(&lustre_mount_info_lock);
119 CERROR("%s not registered\n", name);
123 CDEBUG(D_MOUNT, "dereg_mnt %p from %s, vfscount=%d\n",
124 lmi->lmi_mnt, name, atomic_read(&lmi->lmi_mnt->mnt_count));
126 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
127 list_del(&lmi->lmi_list_chain);
128 OBD_FREE(lmi, sizeof(*lmi));
129 up(&lustre_mount_info_lock);
134 /* obd's look up a registered mount using their obdname. This is just
135 for initial obd setup to find the mount struct. It should not be
136 called every time you want to mntget. */
137 struct lustre_mount_info *server_get_mount(const char *name)
139 struct lustre_mount_info *lmi;
140 struct lustre_sb_info *lsi;
143 down(&lustre_mount_info_lock);
144 lmi = server_find_mount(name);
145 up(&lustre_mount_info_lock);
147 CERROR("Can't find mount for %s\n", name);
150 lsi = s2lsi(lmi->lmi_sb);
151 mntget(lmi->lmi_mnt);
152 atomic_inc(&lsi->lsi_mounts);
154 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d, vfscount=%d\n",
155 lmi->lmi_mnt, name, atomic_read(&lsi->lsi_mounts),
156 atomic_read(&lmi->lmi_mnt->mnt_count));
162 * Used by mdt to get mount_info from obdname.
163 * There are no blocking when using the mount_info.
164 * Do not use server_get_mount for this purpose.
166 struct lustre_mount_info *server_get_mount_2(const char *name)
168 struct lustre_mount_info *lmi;
171 down(&lustre_mount_info_lock);
172 lmi = server_find_mount(name);
173 up(&lustre_mount_info_lock);
175 CERROR("Can't find mount for %s\n", name);
180 static void unlock_mntput(struct vfsmount *mnt)
182 if (kernel_locked()) {
191 static int lustre_put_lsi(struct super_block *sb);
193 /* to be called from obd_cleanup methods */
194 int server_put_mount(const char *name, struct vfsmount *mnt)
196 struct lustre_mount_info *lmi;
197 struct lustre_sb_info *lsi;
198 int count = atomic_read(&mnt->mnt_count) - 1;
201 /* This might be the last one, can't deref after this */
204 down(&lustre_mount_info_lock);
205 lmi = server_find_mount(name);
206 up(&lustre_mount_info_lock);
208 CERROR("Can't find mount for %s\n", name);
211 lsi = s2lsi(lmi->lmi_sb);
212 LASSERT(lmi->lmi_mnt == mnt);
214 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d, vfscount=%d\n",
215 lmi->lmi_mnt, name, atomic_read(&lsi->lsi_mounts), count);
217 if (lustre_put_lsi(lmi->lmi_sb)) {
218 CDEBUG(D_MOUNT, "Last put of mnt %p from %s, vfscount=%d\n",
219 lmi->lmi_mnt, name, count);
220 /* last mount is the One True Mount */
222 CERROR("%s: mount busy, vfscount=%d!\n", name, count);
225 /* this obd should never need the mount again */
226 server_deregister_mount(name);
231 /* Corresponding to server_get_mount_2 */
232 int server_put_mount_2(const char *name, struct vfsmount *mnt)
238 /******* mount helper utilities *********/
241 static void ldd_print(struct lustre_disk_data *ldd)
243 PRINT_CMD(PRINT_MASK, " disk data:\n");
244 PRINT_CMD(PRINT_MASK, "server: %s\n", ldd->ldd_svname);
245 PRINT_CMD(PRINT_MASK, "uuid: %s\n", (char *)ldd->ldd_uuid);
246 PRINT_CMD(PRINT_MASK, "fs: %s\n", ldd->ldd_fsname);
247 PRINT_CMD(PRINT_MASK, "index: %04x\n", ldd->ldd_svindex);
248 PRINT_CMD(PRINT_MASK, "config: %d\n", ldd->ldd_config_ver);
249 PRINT_CMD(PRINT_MASK, "flags: %#x\n", ldd->ldd_flags);
250 PRINT_CMD(PRINT_MASK, "diskfs: %s\n", MT_STR(ldd));
251 PRINT_CMD(PRINT_MASK, "options: %s\n", ldd->ldd_mount_opts);
252 PRINT_CMD(PRINT_MASK, "params: %s\n", ldd->ldd_params);
253 PRINT_CMD(PRINT_MASK, "comment: %s\n", ldd->ldd_userdata);
257 static int ldd_parse(struct lvfs_run_ctxt *mount_ctxt,
258 struct lustre_disk_data *ldd)
260 struct lvfs_run_ctxt saved;
267 push_ctxt(&saved, mount_ctxt, NULL);
269 file = filp_open(MOUNT_DATA_FILE, O_RDONLY, 0644);
272 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
276 len = i_size_read(file->f_dentry->d_inode);
277 CDEBUG(D_MOUNT, "Have %s, size %lu\n", MOUNT_DATA_FILE, len);
278 if (len != sizeof(*ldd)) {
279 CERROR("disk data size does not match: see %lu expect "LPSZ"\n",
281 GOTO(out_close, rc = -EINVAL);
284 rc = lustre_fread(file, ldd, len, &off);
286 CERROR("error reading %s: read %d of %lu\n",
287 MOUNT_DATA_FILE, rc, len);
288 GOTO(out_close, rc = -EINVAL);
292 if (ldd->ldd_magic != LDD_MAGIC) {
293 /* FIXME add swabbing support */
294 CERROR("Bad magic in %s: %x!=%x\n", MOUNT_DATA_FILE,
295 ldd->ldd_magic, LDD_MAGIC);
296 GOTO(out_close, rc = -EINVAL);
299 if (ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP) {
300 CERROR("%s: unsupported incompat filesystem feature(s) %x\n",
302 ldd->ldd_feature_incompat & ~LDD_INCOMPAT_SUPP);
303 GOTO(out_close, rc = -EINVAL);
305 if (ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP) {
306 CERROR("%s: unsupported read-only filesystem feature(s) %x\n",
308 ldd->ldd_feature_rocompat & ~LDD_ROCOMPAT_SUPP);
309 /* Do something like remount filesystem read-only */
310 GOTO(out_close, rc = -EINVAL);
316 pop_ctxt(&saved, mount_ctxt, NULL);
320 static int ldd_write(struct lvfs_run_ctxt *mount_ctxt,
321 struct lustre_disk_data *ldd)
323 struct lvfs_run_ctxt saved;
326 unsigned long len = sizeof(struct lustre_disk_data);
330 LASSERT(ldd->ldd_magic == LDD_MAGIC);
332 ldd->ldd_config_ver++;
334 push_ctxt(&saved, mount_ctxt, NULL);
336 file = filp_open(MOUNT_DATA_FILE, O_RDWR, 0644);
339 CERROR("cannot open %s: rc = %d\n", MOUNT_DATA_FILE, rc);
343 rc = lustre_fwrite(file, ldd, len, &off);
345 CERROR("error writing %s: read %d of %lu\n",
346 MOUNT_DATA_FILE, rc, len);
347 GOTO(out_close, rc = -EINVAL);
355 pop_ctxt(&saved, mount_ctxt, NULL);
360 /**************** config llog ********************/
362 /* Get a config log from the MGS and process it.
363 This func is called for both clients and servers.
364 Continue to process new statements appended to the logs
365 (whenever the config lock is revoked) until lustre_end_log
367 int lustre_process_log(struct super_block *sb, char *logname,
368 struct config_llog_instance *cfg)
370 struct lustre_cfg *lcfg;
371 struct lustre_cfg_bufs bufs;
372 struct lustre_sb_info *lsi = s2lsi(sb);
373 struct obd_device *mgc = lsi->lsi_mgc;
380 /* mgc_process_config */
381 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
382 lustre_cfg_bufs_set_string(&bufs, 1, logname);
383 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
384 lustre_cfg_bufs_set(&bufs, 3, &sb, sizeof(sb));
385 lcfg = lustre_cfg_new(LCFG_LOG_START, &bufs);
386 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
387 lustre_cfg_free(lcfg);
390 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
391 "failed from the MGS (%d). Make sure this "
392 "client and the MGS are running compatible "
393 "versions of Lustre.\n",
394 mgc->obd_name, logname, rc);
397 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
398 "failed (%d). This may be the result of "
399 "communication errors between this node and "
400 "the MGS, a bad configuration, or other "
401 "errors. See the syslog for more "
402 "information.\n", mgc->obd_name, logname,
405 /* class_obd_list(); */
409 /* Stop watching this config log for updates */
410 int lustre_end_log(struct super_block *sb, char *logname,
411 struct config_llog_instance *cfg)
413 struct lustre_cfg *lcfg;
414 struct lustre_cfg_bufs bufs;
415 struct lustre_sb_info *lsi = s2lsi(sb);
416 struct obd_device *mgc = lsi->lsi_mgc;
423 /* mgc_process_config */
424 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
425 lustre_cfg_bufs_set_string(&bufs, 1, logname);
427 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
428 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
429 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
430 lustre_cfg_free(lcfg);
434 /**************** obd start *******************/
436 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
437 char *s1, char *s2, char *s3, char *s4)
439 struct lustre_cfg_bufs bufs;
440 struct lustre_cfg * lcfg = NULL;
443 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
444 cmd, s1, s2, s3, s4);
446 lustre_cfg_bufs_reset(&bufs, cfgname);
448 lustre_cfg_bufs_set_string(&bufs, 1, s1);
450 lustre_cfg_bufs_set_string(&bufs, 2, s2);
452 lustre_cfg_bufs_set_string(&bufs, 3, s3);
454 lustre_cfg_bufs_set_string(&bufs, 4, s4);
456 lcfg = lustre_cfg_new(cmd, &bufs);
457 lcfg->lcfg_nid = nid;
458 rc = class_process_config(lcfg);
459 lustre_cfg_free(lcfg);
463 static int lustre_start_simple(char *obdname, char *type, char *uuid,
467 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
469 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
471 CERROR("%s attach error %d\n", obdname, rc);
474 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, 0, 0);
476 CERROR("%s setup error %d\n", obdname, rc);
477 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
482 /* Set up a MGS to serve startup logs */
483 static int server_start_mgs(struct super_block *sb)
485 struct lustre_sb_info *lsi = s2lsi(sb);
486 struct vfsmount *mnt = lsi->lsi_srv_mnt;
487 struct lustre_mount_info *lmi;
492 /* It is impossible to have more than 1 MGS per node, since
493 MGC wouldn't know which to connect to */
494 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
496 lsi = s2lsi(lmi->lmi_sb);
497 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
499 lsi->lsi_ldd->ldd_svname);
503 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
505 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
508 ((rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
509 LUSTRE_MGS_OBDNAME, 0, 0))))
510 server_deregister_mount(LUSTRE_MGS_OBDNAME);
513 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
514 "Is the 'mgs' module loaded?\n",
515 LUSTRE_MGS_OBDNAME, rc);
519 static int server_stop_mgs(struct super_block *sb)
521 struct obd_device *obd;
525 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
527 /* There better be only one MGS */
528 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
530 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
534 /* The MGS should always stop when we say so */
536 rc = class_manual_cleanup(obd);
540 DECLARE_MUTEX(mgc_start_lock);
542 /* Set up a mgcobd to process startup logs */
543 static int lustre_start_mgc(struct super_block *sb)
545 struct lustre_handle mgc_conn = {0, };
546 struct obd_connect_data ocd = { 0 };
547 struct lustre_sb_info *lsi = s2lsi(sb);
548 struct obd_device *obd;
549 struct obd_export *exp;
550 struct obd_uuid *uuid;
553 char *mgcname, *niduuid;
556 int rc = 0, i = 0, j, len;
559 LASSERT(lsi->lsi_lmd);
561 /* Find the first non-lo MGS nid for our MGC name */
562 if (lsi->lsi_flags & LSI_SERVER) {
563 ptr = lsi->lsi_ldd->ldd_params;
564 /* Use mgsnode= nids */
565 if ((class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0) &&
566 (class_parse_nid(ptr, &nid, &ptr) == 0)) {
568 } else if (IS_MGS(lsi->lsi_ldd)) {
569 lnet_process_id_t id;
570 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
571 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
578 } else { /* client */
579 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
580 ptr = lsi->lsi_lmd->lmd_dev;
581 if (class_parse_nid(ptr, &nid, &ptr) == 0)
585 CERROR("No valid MGS nids found.\n");
589 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
590 OBD_ALLOC(mgcname, len);
591 OBD_ALLOC(niduuid, len + 2);
592 if (!mgcname || !niduuid)
593 GOTO(out_free, rc = -ENOMEM);
594 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
596 mutex_down(&mgc_start_lock);
598 obd = class_name2obd(mgcname);
600 /* Re-using an existing MGC */
601 atomic_inc(&obd->u.cli.cl_mgc_refcount);
604 /* If we are restarting the MGS, don't try to keep the MGC's
605 old connection, or registration will fail. */
606 if ((lsi->lsi_flags & LSI_SERVER) && IS_MGS(lsi->lsi_ldd)) {
607 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
611 /* Try all connections, but only once (again).
612 We don't want to block another target from starting
613 (using its local copy of the log), but we do want to connect
614 if at all possible. */
616 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
617 rc = obd_set_info_async(obd->obd_self_export,
618 strlen(KEY_INIT_RECOV_BACKUP),
619 KEY_INIT_RECOV_BACKUP,
620 sizeof(recov_bk), &recov_bk, NULL);
624 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
626 /* Add the primary nids for the MGS */
628 sprintf(niduuid, "%s_%x", mgcname, i);
629 if (lsi->lsi_flags & LSI_SERVER) {
630 ptr = lsi->lsi_ldd->ldd_params;
631 if (IS_MGS(lsi->lsi_ldd)) {
632 /* Use local nids (including LO) */
633 lnet_process_id_t id;
634 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
635 rc = do_lcfg(mgcname, id.nid,
636 LCFG_ADD_UUID, niduuid, 0,0,0);
639 /* Use mgsnode= nids */
640 if (class_find_param(ptr, PARAM_MGSNODE, &ptr) != 0) {
641 CERROR("No MGS nids given.\n");
642 GOTO(out_free, rc = -EINVAL);
644 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
645 rc = do_lcfg(mgcname, nid,
646 LCFG_ADD_UUID, niduuid, 0,0,0);
650 } else { /* client */
651 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
652 ptr = lsi->lsi_lmd->lmd_dev;
653 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
654 rc = do_lcfg(mgcname, nid,
655 LCFG_ADD_UUID, niduuid, 0,0,0);
657 /* Stop at the first failover nid */
663 CERROR("No valid MGS nids found.\n");
664 GOTO(out_free, rc = -EINVAL);
666 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
668 /* Random uuid for MGC allows easier reconnects */
670 ll_generate_random_uuid(uuidc);
671 class_uuid_unparse(uuidc, uuid);
674 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
675 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
681 /* Add any failover MGS nids */
683 while ((*ptr == ':' ||
684 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0)) {
685 /* New failover node */
686 sprintf(niduuid, "%s_%x", mgcname, i);
688 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
690 rc = do_lcfg(mgcname, nid,
691 LCFG_ADD_UUID, niduuid, 0,0,0);
696 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
704 lsi->lsi_lmd->lmd_mgs_failnodes = i;
706 obd = class_name2obd(mgcname);
708 CERROR("Can't find mgcobd %s\n", mgcname);
709 GOTO(out_free, rc = -ENOTCONN);
712 /* Keep a refcount of servers/clients who started with "mount",
713 so we know when we can get rid of the mgc. */
714 atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
716 /* Try all connections, but only once. */
718 rc = obd_set_info_async(obd->obd_self_export,
719 strlen(KEY_INIT_RECOV_BACKUP),
720 KEY_INIT_RECOV_BACKUP,
721 sizeof(recov_bk), &recov_bk, NULL);
724 CERROR("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
725 /* We connect to the MGS at setup, and don't disconnect until cleanup */
727 ocd.ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_FID;
728 ocd.ocd_version = LUSTRE_VERSION_CODE;
730 rc = obd_connect(NULL, &mgc_conn, obd, &(obd->obd_uuid), &ocd);
732 CERROR("connect failed %d\n", rc);
736 exp = class_conn2export(&mgc_conn);
737 obd->u.cli.cl_mgc_mgsexp = exp;
740 /* Keep the mgc info in the sb. Note that many lsi's can point
744 mutex_up(&mgc_start_lock);
747 OBD_FREE(mgcname, len);
749 OBD_FREE(niduuid, len + 2);
753 static int lustre_stop_mgc(struct super_block *sb)
755 struct lustre_sb_info *lsi = s2lsi(sb);
756 struct obd_device *obd;
757 char *niduuid = 0, *ptr = 0;
758 int i, rc = 0, len = 0;
768 mutex_down(&mgc_start_lock);
769 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
770 /* This is not fatal, every client that stops
771 will call in here. */
772 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
773 atomic_read(&obd->u.cli.cl_mgc_refcount));
774 GOTO(out, rc = -EBUSY);
777 /* The MGC has no recoverable data in any case.
778 * force shotdown set in umount_begin */
779 obd->obd_no_recov = 1;
781 if (obd->u.cli.cl_mgc_mgsexp)
782 obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
784 /* Save the obdname for cleaning the nid uuids, which are
786 len = strlen(obd->obd_name) + 6;
787 OBD_ALLOC(niduuid, len);
789 strcpy(niduuid, obd->obd_name);
790 ptr = niduuid + strlen(niduuid);
793 rc = class_manual_cleanup(obd);
797 /* Clean the nid uuids */
800 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
801 sprintf(ptr, "_%x", i);
802 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
805 CERROR("del MDC UUID %s failed: rc = %d\n",
808 OBD_FREE(niduuid, len);
809 /* class_import_put will get rid of the additional connections */
812 mutex_up(&mgc_start_lock);
816 /* Since there's only one mgc per node, we have to change it's fs to get
817 access to the right disk. */
818 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
820 struct lustre_sb_info *lsi = s2lsi(sb);
824 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
826 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
827 rc = obd_set_info_async(mgc->obd_self_export,
828 strlen("set_fs"), "set_fs",
829 sizeof(*sb), sb, NULL);
831 CERROR("can't set_fs %d\n", rc);
837 static int server_mgc_clear_fs(struct obd_device *mgc)
842 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
844 rc = obd_set_info_async(mgc->obd_self_export,
845 strlen("clear_fs"), "clear_fs",
850 DECLARE_MUTEX(server_start_lock);
852 /* Stop MDS/OSS if nobody is using them */
853 static int server_stop_servers(int lddflags, int lsiflags)
855 struct obd_device *obd = NULL;
856 struct obd_type *type = NULL;
860 mutex_down(&server_start_lock);
862 /* Either an MDT or an OST or neither */
863 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
864 if ((lddflags & LDD_F_SV_TYPE_MDT) &&
865 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
866 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
867 type = class_search_type(LUSTRE_MDS_NAME);
869 /* if this was an OST, and there are no more OST's, clean up the OSS */
870 if ((lddflags & LDD_F_SV_TYPE_OST) &&
871 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
872 type = class_search_type(LUSTRE_OST_NAME);
875 if (obd && (!type || !type->typ_refcnt)) {
878 /* obd_fail doesn't mean much on a server obd */
879 err = class_manual_cleanup(obd);
884 mutex_up(&server_start_lock);
889 int server_mti_print(char *title, struct mgs_target_info *mti)
891 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
892 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
893 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
894 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
895 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
896 mti->mti_config_ver, mti->mti_flags);
900 static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
902 struct lustre_sb_info *lsi = s2lsi(sb);
903 struct lustre_disk_data *ldd = lsi->lsi_ldd;
904 lnet_process_id_t id;
908 if (!(lsi->lsi_flags & LSI_SERVER))
911 strncpy(mti->mti_fsname, ldd->ldd_fsname,
912 sizeof(mti->mti_fsname));
913 strncpy(mti->mti_svname, ldd->ldd_svname,
914 sizeof(mti->mti_svname));
916 mti->mti_nid_count = 0;
917 while (LNetGetId(i++, &id) != -ENOENT) {
918 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
920 mti->mti_nids[mti->mti_nid_count] = id.nid;
921 mti->mti_nid_count++;
922 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
923 CWARN("Only using first %d nids for %s\n",
924 mti->mti_nid_count, mti->mti_svname);
929 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
930 mti->mti_config_ver = 0;
931 mti->mti_flags = ldd->ldd_flags;
932 mti->mti_stripe_index = ldd->ldd_svindex;
933 memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid));
934 if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) {
935 CERROR("params too big for mti\n");
938 memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params));
942 /* Register an old or new target with the MGS. If needed MGS will construct
943 startup logs and assign index */
944 int server_register_target(struct super_block *sb)
946 struct lustre_sb_info *lsi = s2lsi(sb);
947 struct obd_device *mgc = lsi->lsi_mgc;
948 struct lustre_disk_data *ldd = lsi->lsi_ldd;
949 struct mgs_target_info *mti = NULL;
955 if (!(lsi->lsi_flags & LSI_SERVER))
961 rc = server_sb2mti(sb, mti);
965 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
966 mti->mti_svname, mti->mti_fsname,
967 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
970 /* Register the target */
971 /* FIXME use mgc_process_config instead */
972 rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
973 strlen("register_target"), "register_target",
974 sizeof(*mti), mti, NULL);
976 CERROR("registration with the MGS failed (%d)\n", rc);
980 /* Always update our flags */
981 ldd->ldd_flags = mti->mti_flags & ~LDD_F_REWRITE_LDD;
983 /* If this flag is set, it means the MGS wants us to change our
984 on-disk data. (So far this means just the index.) */
985 if (mti->mti_flags & LDD_F_REWRITE_LDD) {
988 CDEBUG(D_MOUNT, "Changing on-disk index from %#x to %#x "
989 "for %s\n", ldd->ldd_svindex, mti->mti_stripe_index,
991 ldd->ldd_svindex = mti->mti_stripe_index;
992 strncpy(ldd->ldd_svname, mti->mti_svname,
993 sizeof(ldd->ldd_svname));
994 /* or ldd_make_sv_name(ldd); */
995 ldd_write(&mgc->obd_lvfs_ctxt, ldd);
996 err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb,
999 CERROR("Label set error %d\n", err);
1000 label = fsfilt_get_label(mgc, lsi->lsi_srv_mnt->mnt_sb);
1002 CDEBUG(D_MOUNT, "Disk label changed to %s\n", label);
1004 /* Flush the new ldd to disk */
1005 fsfilt_sync(mgc, lsi->lsi_srv_mnt->mnt_sb);
1015 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1017 struct obd_device *obd;
1018 struct lustre_sb_info *lsi = s2lsi(sb);
1019 struct config_llog_instance cfg;
1023 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname);
1026 /* If we're an MDT, make sure the global MDS is running */
1027 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1028 /* make sure the MDS is started */
1029 mutex_down(&server_start_lock);
1030 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1032 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1033 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1035 LUSTRE_MDS_OBDNAME"_uuid",
1038 mutex_up(&server_start_lock);
1039 CERROR("failed to start MDS: %d\n", rc);
1043 mutex_up(&server_start_lock);
1047 /* If we're an OST, make sure the global OSS is running */
1048 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
1049 /* make sure OSS is started */
1050 mutex_down(&server_start_lock);
1051 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1053 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1055 LUSTRE_OSS_OBDNAME"_uuid",
1058 mutex_up(&server_start_lock);
1059 CERROR("failed to start OSS: %d\n", rc);
1063 mutex_up(&server_start_lock);
1066 /* Set the mgc fs to our server disk. This allows the MGC
1067 to read and write configs locally. */
1068 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1072 /* Register with MGS */
1073 rc = server_register_target(sb);
1074 if (rc && (lsi->lsi_ldd->ldd_flags &
1075 (LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_UPGRADE14))){
1076 CERROR("Required registration failed for %s: %d\n",
1077 lsi->lsi_ldd->ldd_svname, rc);
1079 LCONSOLE_ERROR_MSG(0x15f, "Communication error with "
1080 "the MGS. Is the MGS running?\n");
1084 if (rc == -EINVAL) {
1085 LCONSOLE_ERROR_MSG(0x160, "The MGS is refusing to allow this "
1086 "server (%s) to start. Please see messages"
1087 " on the MGS node.\n",
1088 lsi->lsi_ldd->ldd_svname);
1092 /* Let the target look up the mount using the target's name
1093 (we can't pass the sb or mnt through class_process_config.) */
1094 rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
1098 /* Start targets using the llog named for the target */
1099 memset(&cfg, 0, sizeof(cfg));
1100 rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg);
1102 CERROR("failed to start server %s: %d\n",
1103 lsi->lsi_ldd->ldd_svname, rc);
1108 /* Release the mgc fs for others to use */
1109 server_mgc_clear_fs(lsi->lsi_mgc);
1112 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1114 CERROR("no server named %s was started\n",
1115 lsi->lsi_ldd->ldd_svname);
1119 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1120 (OBP(obd, iocontrol))) {
1121 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1122 obd->obd_self_export, 0, NULL, NULL);
1125 /* log has been fully processed */
1126 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1132 /***************** lustre superblock **************/
1134 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1136 struct lustre_sb_info *lsi = NULL;
1139 OBD_ALLOC(lsi, sizeof(*lsi));
1142 OBD_ALLOC(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1143 if (!lsi->lsi_lmd) {
1144 OBD_FREE(lsi, sizeof(*lsi));
1148 lsi->lsi_lmd->lmd_exclude_count = 0;
1149 s2lsi_nocast(sb) = lsi;
1150 /* we take 1 extra ref for our setup */
1151 atomic_set(&lsi->lsi_mounts, 1);
1153 /* Default umount style */
1154 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1159 static int lustre_free_lsi(struct super_block *sb)
1161 struct lustre_sb_info *lsi = s2lsi(sb);
1167 CDEBUG(D_MOUNT, "Freeing lsi\n");
1169 /* someone didn't call server_put_mount. */
1170 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
1172 if (lsi->lsi_ldd != NULL)
1173 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
1175 if (lsi->lsi_lmd != NULL) {
1176 if (lsi->lsi_lmd->lmd_dev != NULL)
1177 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1178 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1179 if (lsi->lsi_lmd->lmd_profile != NULL)
1180 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1181 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1182 if (lsi->lsi_lmd->lmd_opts != NULL)
1183 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1184 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1185 if (lsi->lsi_lmd->lmd_exclude_count)
1186 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1187 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1188 lsi->lsi_lmd->lmd_exclude_count);
1189 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1192 LASSERT(lsi->lsi_llsbi == NULL);
1193 OBD_FREE(lsi, sizeof(*lsi));
1194 s2lsi_nocast(sb) = NULL;
1199 /* The lsi has one reference for every server that is using the disk -
1200 e.g. MDT, MGS, and potentially MGC */
1201 static int lustre_put_lsi(struct super_block *sb)
1203 struct lustre_sb_info *lsi = s2lsi(sb);
1208 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
1210 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
1211 lustre_free_lsi(sb);
1217 /*************** server mount ******************/
1219 /* Kernel mount using mount options in MOUNT_DATA_FILE */
1220 static struct vfsmount *server_kernel_mount(struct super_block *sb)
1222 struct lvfs_run_ctxt mount_ctxt;
1223 struct lustre_sb_info *lsi = s2lsi(sb);
1224 struct lustre_disk_data *ldd;
1225 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1226 struct vfsmount *mnt;
1227 char *options = NULL;
1228 unsigned long page, s_flags;
1229 struct page *__page;
1233 OBD_ALLOC(ldd, sizeof(*ldd));
1235 RETURN(ERR_PTR(-ENOMEM));
1237 /* In the past, we have always used flags = 0.
1238 Note ext3/ldiskfs can't be mounted ro. */
1239 s_flags = sb->s_flags;
1241 /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */
1242 CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev);
1243 mnt = ll_kern_mount("ldiskfs", s_flags, lmd->lmd_dev, 0);
1246 CERROR("premount %s:%#lx ldiskfs failed: %d "
1247 "Is the ldiskfs module available?\n",
1248 lmd->lmd_dev, s_flags, rc );
1252 OBD_SET_CTXT_MAGIC(&mount_ctxt);
1253 mount_ctxt.pwdmnt = mnt;
1254 mount_ctxt.pwd = mnt->mnt_root;
1255 mount_ctxt.fs = get_ds();
1257 rc = ldd_parse(&mount_ctxt, ldd);
1261 CERROR("premount parse options failed: rc = %d\n", rc);
1265 /* Done with our pre-mount, now do the real mount. */
1267 /* Glom up mount options */
1268 OBD_PAGE_ALLOC(__page, CFS_ALLOC_STD);
1270 GOTO(out_free, rc = -ENOMEM);
1271 page = (unsigned long)cfs_page_address(__page);
1273 options = (char *)page;
1274 memset(options, 0, CFS_PAGE_SIZE);
1275 strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2);
1277 /* Add in any mount-line options */
1278 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) {
1279 int len = CFS_PAGE_SIZE - strlen(options) - 2;
1281 strcat(options, ",");
1282 strncat(options, lmd->lmd_opts, len);
1285 /* Special permanent mount flags */
1287 s_flags |= MS_NOATIME | MS_NODIRATIME;
1289 CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
1290 MT_STR(ldd), lmd->lmd_dev, options);
1291 mnt = ll_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev,
1293 OBD_PAGE_FREE(__page);
1296 CERROR("ll_kern_mount failed: rc = %d\n", rc);
1300 lsi->lsi_ldd = ldd; /* freed at lsi cleanup */
1301 CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
1305 OBD_FREE(ldd, sizeof(*ldd));
1306 lsi->lsi_ldd = NULL;
1307 RETURN(ERR_PTR(rc));
1310 static void server_wait_finished(struct vfsmount *mnt)
1312 wait_queue_head_t waitq;
1313 struct l_wait_info lwi;
1316 init_waitqueue_head(&waitq);
1318 while ((atomic_read(&mnt->mnt_count) > 1) && (retries > 0)) {
1319 LCONSOLE_WARN("Mount still busy with %d refs, waiting for "
1321 atomic_read(&mnt->mnt_count), retries);
1323 /* Wait for a bit */
1325 lwi = LWI_TIMEOUT(5 * HZ, NULL, NULL);
1326 l_wait_event(waitq, 0, &lwi);
1328 if (atomic_read(&mnt->mnt_count) > 1) {
1329 CERROR("Mount %p is still busy (%d refs), giving up.\n",
1330 mnt, atomic_read(&mnt->mnt_count));
1334 static void server_put_super(struct super_block *sb)
1336 struct lustre_sb_info *lsi = s2lsi(sb);
1337 struct obd_device *obd;
1338 struct vfsmount *mnt = lsi->lsi_srv_mnt;
1339 char *tmpname, *extraname = NULL;
1341 int lddflags = lsi->lsi_ldd->ldd_flags;
1342 int lsiflags = lsi->lsi_flags;
1346 LASSERT(lsiflags & LSI_SERVER);
1348 tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1;
1349 OBD_ALLOC(tmpname, tmpname_sz);
1350 memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
1351 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1353 /* Stop the target */
1354 if (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd)) {
1355 struct lustre_profile *lprof = NULL;
1357 /* tell the mgc to drop the config log */
1358 lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL);
1360 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1361 If there are any setup/cleanup errors, save the lov
1362 name for safety cleanup later. */
1363 lprof = class_get_profile(lsi->lsi_ldd->ldd_svname);
1364 if (lprof && lprof->lp_dt) {
1365 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1366 strcpy(extraname, lprof->lp_dt);
1369 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1371 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1372 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1374 /* We can't seem to give an error return code
1375 * to .put_super, so we better make sure we clean up! */
1377 class_manual_cleanup(obd);
1379 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
1380 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1384 /* If they wanted the mgs to stop separately from the mdt, they
1385 should have put it on a different device. */
1386 if (IS_MGS(lsi->lsi_ldd)) {
1387 /* stop the mgc before the mgs so the connection gets cleaned
1389 lustre_stop_mgc(sb);
1390 server_stop_mgs(sb);
1393 /* Clean the mgc and sb */
1394 rc = lustre_common_put_super(sb);
1395 /* FIXME how can I report a failure to umount? */
1397 /* Wait for the targets to really clean up - can't exit (and let the
1398 sb get destroyed) while the mount is still in use */
1399 server_wait_finished(mnt);
1401 /* drop the One True Mount */
1404 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1405 until the target is really gone so that our type refcount check
1407 server_stop_servers(lddflags, lsiflags);
1409 /* In case of startup or cleanup err, stop related obds */
1411 obd = class_name2obd(extraname);
1413 CWARN("Cleaning orphaned obd %s\n", extraname);
1415 class_manual_cleanup(obd);
1417 OBD_FREE(extraname, strlen(extraname) + 1);
1420 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1421 OBD_FREE(tmpname, tmpname_sz);
1425 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1426 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1428 struct super_block *sb = vfsmnt->mnt_sb;
1430 static void server_umount_begin(struct super_block *sb)
1433 struct lustre_sb_info *lsi = s2lsi(sb);
1436 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1437 if (!(flags & MNT_FORCE)) {
1443 CDEBUG(D_MOUNT, "umount -f\n");
1444 /* umount = failover
1446 no third way to do non-force, non-failover */
1447 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1448 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1452 #ifndef HAVE_STATFS_DENTRY_PARAM
1453 static int server_statfs (struct super_block *sb, struct kstatfs *buf)
1456 static int server_statfs (struct dentry *dentry, struct kstatfs *buf)
1458 struct super_block *sb = dentry->d_sb;
1460 struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt;
1463 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) {
1464 #ifdef HAVE_STATFS_DENTRY_PARAM
1465 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf);
1467 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_sb, buf);
1470 buf->f_type = sb->s_magic;
1476 buf->f_type = sb->s_magic;
1477 buf->f_bsize = sb->s_blocksize;
1483 buf->f_namelen = NAME_MAX;
1487 static struct super_operations server_ops =
1489 .put_super = server_put_super,
1490 .umount_begin = server_umount_begin, /* umount -f */
1491 .statfs = server_statfs,
1494 #define log2(n) ffz(~(n))
1495 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1497 static int server_fill_super_common(struct super_block *sb)
1499 struct inode *root = 0;
1502 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1504 sb->s_blocksize = 4096;
1505 sb->s_blocksize_bits = log2(sb->s_blocksize);
1506 sb->s_magic = LUSTRE_SUPER_MAGIC;
1507 sb->s_maxbytes = 0; //PAGE_CACHE_MAXBYTES;
1508 sb->s_flags |= MS_RDONLY;
1509 sb->s_op = &server_ops;
1511 root = new_inode(sb);
1513 CERROR("Can't make root inode\n");
1517 /* returns -EIO for every operation */
1518 /* make_bad_inode(root); -- badness - can't umount */
1519 /* apparently we need to be a directory for the mount to finish */
1520 root->i_mode = S_IFDIR;
1522 sb->s_root = d_alloc_root(root);
1524 CERROR("Can't make root dentry\n");
1532 static int server_fill_super(struct super_block *sb)
1534 struct lustre_sb_info *lsi = s2lsi(sb);
1535 struct vfsmount *mnt;
1539 /* the One True Mount */
1540 mnt = server_kernel_mount(sb);
1543 CERROR("Unable to mount device %s: %d\n",
1544 lsi->lsi_lmd->lmd_dev, rc);
1548 lsi->lsi_srv_mnt = mnt;
1550 LASSERT(lsi->lsi_ldd);
1551 CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n",
1552 lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname,
1553 lsi->lsi_lmd->lmd_dev);
1555 if (class_name2obd(lsi->lsi_ldd->ldd_svname)) {
1556 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1557 "running. Double-mount may have compromised"
1558 " the disk journal.\n",
1559 lsi->lsi_ldd->ldd_svname);
1562 GOTO(out, rc = -EALREADY);
1565 /* start MGS before MGC */
1566 if (IS_MGS(lsi->lsi_ldd)) {
1567 rc = server_start_mgs(sb);
1572 rc = lustre_start_mgc(sb);
1576 /* Set up all obd devices for service */
1577 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1578 (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) {
1579 rc = server_start_targets(sb, mnt);
1581 CERROR("Unable to start targets: %d\n", rc);
1584 /* FIXME overmount client here,
1585 or can we just start a client log and client_fill_super on this sb?
1586 We need to make sure server_put_super gets called too - ll_put_super
1587 calls lustre_common_put_super; check there for LSI_SERVER flag,
1589 Probably should start client from new thread so we can return.
1590 Client will not finish until all servers are connected.
1591 Note - MGS-only server does NOT get a client, since there is no
1592 lustre fs associated - the MGS is for all lustre fs's */
1595 rc = server_fill_super_common(sb);
1599 LCONSOLE_WARN("Server %s on device %s has started\n",
1600 lsi->lsi_ldd->ldd_svname, lsi->lsi_lmd->lmd_dev);
1605 server_put_super(sb);
1610 /* Get the index from the obd name.
1611 rc = server type, or
1613 if endptr isn't NULL it is set to end of name */
1614 int server_name2index(char *svname, __u32 *idx, char **endptr)
1616 unsigned long index;
1618 char *dash = strchr(svname, '-');
1622 if (strncmp(dash + 1, "MDT", 3) == 0)
1623 rc = LDD_F_SV_TYPE_MDT;
1624 else if (strncmp(dash + 1, "OST", 3) == 0)
1625 rc = LDD_F_SV_TYPE_OST;
1629 index = simple_strtoul(dash + 4, endptr, 16);
1634 /*************** mount common betweeen server and client ***************/
1637 int lustre_common_put_super(struct super_block *sb)
1642 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1644 /* Drop a ref to the MGC */
1645 rc = lustre_stop_mgc(sb);
1646 if (rc && (rc != -ENOENT)) {
1648 CERROR("Can't stop MGC: %d\n", rc);
1651 /* BUSY just means that there's some other obd that
1652 needs the mgc. Let him clean it up. */
1653 CDEBUG(D_MOUNT, "MGC still in use\n");
1655 /* Drop a ref to the mounted disk */
1661 static void lmd_print(struct lustre_mount_data *lmd)
1665 PRINT_CMD(PRINT_MASK, " mount data:\n");
1666 if (lmd_is_client(lmd))
1667 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
1668 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
1669 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
1671 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
1672 for (i = 0; i < lmd->lmd_exclude_count; i++) {
1673 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
1674 lmd->lmd_exclude[i]);
1679 /* Is this server on the exclusion list */
1680 int lustre_check_exclusion(struct super_block *sb, char *svname)
1682 struct lustre_sb_info *lsi = s2lsi(sb);
1683 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1688 rc = server_name2index(svname, &index, NULL);
1689 if (rc != LDD_F_SV_TYPE_OST)
1690 /* Only exclude OSTs */
1693 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
1694 index, lmd->lmd_exclude_count, lmd->lmd_dev);
1696 for(i = 0; i < lmd->lmd_exclude_count; i++) {
1697 if (index == lmd->lmd_exclude[i]) {
1698 CWARN("Excluding %s (on exclusion list)\n", svname);
1705 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
1706 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
1708 char *s1 = ptr, *s2;
1709 __u32 index, *exclude_list;
1713 /* The shortest an ost name can be is 8 chars: -OST0000.
1714 We don't actually know the fsname at this time, so in fact
1715 a user could specify any fsname. */
1716 devmax = strlen(ptr) / 8 + 1;
1718 /* temp storage until we figure out how many we have */
1719 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
1723 /* we enter this fn pointing at the '=' */
1724 while (*s1 && *s1 != ' ' && *s1 != ',') {
1726 rc = server_name2index(s1, &index, &s2);
1728 CERROR("Can't parse server name '%s'\n", s1);
1731 if (rc == LDD_F_SV_TYPE_OST)
1732 exclude_list[lmd->lmd_exclude_count++] = index;
1734 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
1736 /* now we are pointing at ':' (next exclude)
1737 or ',' (end of excludes) */
1738 if (lmd->lmd_exclude_count >= devmax)
1741 if (rc >= 0) /* non-err */
1744 if (lmd->lmd_exclude_count) {
1745 /* permanent, freed in lustre_free_lsi */
1746 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
1747 lmd->lmd_exclude_count);
1748 if (lmd->lmd_exclude) {
1749 memcpy(lmd->lmd_exclude, exclude_list,
1750 sizeof(index) * lmd->lmd_exclude_count);
1753 lmd->lmd_exclude_count = 0;
1756 OBD_FREE(exclude_list, sizeof(index) * devmax);
1760 /* mount -v -t lustre uml1:uml2:/lustre-client /mnt/lustre */
1761 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1763 char *s1, *s2, *devname = NULL;
1764 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1770 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1771 "/sbin/mount.lustre is installed.\n");
1775 /* Options should be a string - try to detect old lmd data */
1776 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1777 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1778 "/sbin/mount.lustre. Please install "
1779 "version %s\n", LUSTRE_VERSION_STRING);
1782 lmd->lmd_magic = LMD_MAGIC;
1784 /* Set default flags here */
1789 /* Skip whitespace and extra commas */
1790 while (*s1 == ' ' || *s1 == ',')
1793 /* Client options are parsed in ll_options: eg. flock,
1796 /* Parse non-ldiskfs options here. Rather than modifying
1797 ldiskfs, we just zero these out here */
1798 if (strncmp(s1, "abort_recov", 11) == 0) {
1799 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1801 } else if (strncmp(s1, "nosvc", 5) == 0) {
1802 lmd->lmd_flags |= LMD_FLG_NOSVC;
1804 /* ost exclusion list */
1805 } else if (strncmp(s1, "exclude=", 8) == 0) {
1806 rc = lmd_make_exclusion(lmd, s1 + 7);
1811 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1812 end of the options. */
1813 else if (strncmp(s1, "device=", 7) == 0) {
1815 /* terminate options right before device. device
1816 must be the last one. */
1822 s2 = strchr(s1, ',');
1830 memmove(s1, s2, strlen(s2) + 1);
1836 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1837 "(need mount option 'device=...')\n");
1841 s1 = strrchr(devname, ':');
1843 lmd->lmd_flags = LMD_FLG_CLIENT;
1844 /* Remove leading /s from fsname */
1845 while (*++s1 == '/') ;
1846 /* Freed in lustre_free_lsi */
1847 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1848 if (!lmd->lmd_profile)
1850 sprintf(lmd->lmd_profile, "%s-client", s1);
1853 /* Freed in lustre_free_lsi */
1854 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1857 strcpy(lmd->lmd_dev, devname);
1859 /* Save mount options */
1860 s1 = options + strlen(options) - 1;
1861 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1863 if (*options != 0) {
1864 /* Freed in lustre_free_lsi */
1865 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1868 strcpy(lmd->lmd_opts, options);
1871 lmd->lmd_magic = LMD_MAGIC;
1876 CERROR("Bad mount options %s\n", options);
1882 int lustre_fill_super(struct super_block *sb, void *data, int silent)
1884 struct lustre_mount_data *lmd;
1885 struct lustre_sb_info *lsi;
1889 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1891 lsi = lustre_init_lsi(sb);
1896 /* Figure out the lmd from the mount options */
1897 if (lmd_parse((char *)data, lmd)) {
1902 if (lmd_is_client(lmd)) {
1903 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1904 if (!client_fill_super) {
1905 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1906 "client mount! Is the 'lustre' "
1907 "module loaded?\n");
1910 rc = lustre_start_mgc(sb);
1912 lustre_stop_mgc(sb);
1915 /* Connect and start */
1916 /* (should always be ll_fill_super) */
1917 rc = (*client_fill_super)(sb);
1918 /* c_f_s will call lustre_common_put_super on failure */
1921 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1922 lsi->lsi_flags |= LSI_SERVER;
1923 rc = server_fill_super(sb);
1924 /* s_f_s calls lustre_start_mgc after the mount because we need
1925 the MGS nids which are stored on disk. Plus, we may
1926 need to start the MGS first. */
1927 /* s_f_s will call server_put_super on failure */
1932 CERROR("Unable to mount %s (%d)\n",
1933 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1935 CDEBUG(D_SUPER, "mount %s complete\n", lmd->lmd_dev);
1941 /* We can't call ll_fill_super by name because it lives in a module that
1942 must be loaded after this one. */
1943 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb))
1945 client_fill_super = cfs;
1948 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
1950 kill_super_cb = cfs;
1953 /***************** FS registration ******************/
1955 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))
1956 struct super_block * lustre_get_sb(struct file_system_type *fs_type,
1957 int flags, const char *devname, void * data)
1959 /* calls back in fill super */
1960 /* we could append devname= onto options (*data) here,
1961 but 2.4 doesn't get devname. So we do it in mount_lustre.c */
1962 return get_sb_nodev(fs_type, flags, data, lustre_fill_super);
1965 int lustre_get_sb(struct file_system_type *fs_type,
1966 int flags, const char *devname, void * data,
1967 struct vfsmount *mnt)
1969 /* calls back in fill super */
1970 /* we could append devname= onto options (*data) here,
1971 but 2.4 doesn't get devname. So we do it in mount_lustre.c */
1972 return get_sb_nodev(fs_type, flags, data, lustre_fill_super, mnt);
1976 void lustre_kill_super(struct super_block *sb)
1978 struct lustre_sb_info *lsi = s2lsi(sb);
1980 if (kill_super_cb && lsi &&(lsi->lsi_flags & LSI_SERVER))
1981 (*kill_super_cb)(sb);
1983 kill_anon_super(sb);
1986 struct file_system_type lustre_fs_type = {
1987 .owner = THIS_MODULE,
1989 .get_sb = lustre_get_sb,
1990 .kill_sb = lustre_kill_super,
1991 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV,
1994 int lustre_register_fs(void)
1996 return register_filesystem(&lustre_fs_type);
1999 int lustre_unregister_fs(void)
2001 return unregister_filesystem(&lustre_fs_type);
2004 EXPORT_SYMBOL(lustre_register_client_fill_super);
2005 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2006 EXPORT_SYMBOL(lustre_common_put_super);
2007 EXPORT_SYMBOL(lustre_process_log);
2008 EXPORT_SYMBOL(lustre_end_log);
2009 EXPORT_SYMBOL(server_get_mount);
2010 EXPORT_SYMBOL(server_get_mount_2);
2011 EXPORT_SYMBOL(server_put_mount);
2012 EXPORT_SYMBOL(server_put_mount_2);
2013 EXPORT_SYMBOL(server_register_target);
2014 EXPORT_SYMBOL(server_name2index);
2015 EXPORT_SYMBOL(server_mti_print);
2016 EXPORT_SYMBOL(do_lcfg);