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 */
726 rc = obd_connect(NULL, &mgc_conn, obd, &(obd->obd_uuid), &ocd);
728 CERROR("connect failed %d\n", rc);
732 exp = class_conn2export(&mgc_conn);
733 obd->u.cli.cl_mgc_mgsexp = exp;
736 /* Keep the mgc info in the sb. Note that many lsi's can point
740 mutex_up(&mgc_start_lock);
743 OBD_FREE(mgcname, len);
745 OBD_FREE(niduuid, len + 2);
749 static int lustre_stop_mgc(struct super_block *sb)
751 struct lustre_sb_info *lsi = s2lsi(sb);
752 struct obd_device *obd;
753 char *niduuid, *ptr = 0;
764 mutex_down(&mgc_start_lock);
765 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
766 /* This is not fatal, every client that stops
767 will call in here. */
768 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
769 atomic_read(&obd->u.cli.cl_mgc_refcount));
770 GOTO(out, rc = -EBUSY);
773 /* MGC must always stop */
775 /* client_disconnect_export uses the no_recov flag to decide whether it
776 should disconnect or just invalidate. (The MGC has no
777 recoverable data in any case.) */
778 obd->obd_no_recov = 1;
780 if (obd->u.cli.cl_mgc_mgsexp)
781 obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
783 /* Save the obdname for cleaning the nid uuids, which are
785 len = strlen(obd->obd_name) + 6;
786 OBD_ALLOC(niduuid, len);
788 strcpy(niduuid, obd->obd_name);
789 ptr = niduuid + strlen(niduuid);
792 rc = class_manual_cleanup(obd);
796 /* Clean the nid uuids */
799 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
800 sprintf(ptr, "_%x", i);
801 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
804 CERROR("del MDC UUID %s failed: rc = %d\n",
807 OBD_FREE(niduuid, len);
808 /* class_import_put will get rid of the additional connections */
811 mutex_up(&mgc_start_lock);
815 /* Since there's only one mgc per node, we have to change it's fs to get
816 access to the right disk. */
817 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
819 struct lustre_sb_info *lsi = s2lsi(sb);
823 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
825 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
826 rc = obd_set_info_async(mgc->obd_self_export,
827 strlen("set_fs"), "set_fs",
828 sizeof(*sb), sb, NULL);
830 CERROR("can't set_fs %d\n", rc);
836 static int server_mgc_clear_fs(struct obd_device *mgc)
841 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
843 rc = obd_set_info_async(mgc->obd_self_export,
844 strlen("clear_fs"), "clear_fs",
849 DECLARE_MUTEX(server_start_lock);
851 /* Stop MDS/OSS if nobody is using them */
852 static int server_stop_servers(int lddflags, int lsiflags)
854 struct obd_device *obd = NULL;
855 struct obd_type *type = NULL;
859 mutex_down(&server_start_lock);
861 /* Either an MDT or an OST or neither */
862 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
863 if ((lddflags & LDD_F_SV_TYPE_MDT) &&
864 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
865 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
866 type = class_search_type(LUSTRE_MDS_NAME);
868 /* if this was an OST, and there are no more OST's, clean up the OSS */
869 if ((lddflags & LDD_F_SV_TYPE_OST) &&
870 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
871 type = class_search_type(LUSTRE_OST_NAME);
874 if (obd && (!type || !type->typ_refcnt)) {
877 /* obd_fail doesn't mean much on a server obd */
878 err = class_manual_cleanup(obd);
883 mutex_up(&server_start_lock);
888 int server_mti_print(char *title, struct mgs_target_info *mti)
890 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
891 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
892 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
893 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
894 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
895 mti->mti_config_ver, mti->mti_flags);
900 int mti_set_sec_opts(struct mgs_target_info *mti, struct lustre_mount_data *lmd)
904 if (lmd->lmd_sec_mdt == NULL && lmd->lmd_sec_cli == NULL) {
905 /* just let on-disk params do its work. but we have an
906 * assumption that any changes of on-disk data by tune2fs
907 * should lead to server rewrite log.
912 /* filter out existing sec options */
913 s1 = mti->mti_params;
920 if (strncmp(s1, PARAM_SEC_RPC_MDT,
921 sizeof(PARAM_SEC_RPC_MDT) - 1) == 0 ||
922 strncmp(s1, PARAM_SEC_RPC_CLI,
923 sizeof(PARAM_SEC_RPC_CLI) - 1) == 0)
928 s2 = strchr(s1, ' ');
936 memmove(s1, s2, strlen(s2) + 1);
941 /* append sec options from lmd */
942 /* FIXME add flag LDD_F_UPDATE after mountconf start supporting
945 if (lmd->lmd_sec_mdt) {
946 if (strlen(mti->mti_params) + strlen(lmd->lmd_sec_mdt) +
947 sizeof(PARAM_SEC_RPC_MDT) + 1 >= sizeof(mti->mti_params)) {
948 CERROR("security params too big for mti\n");
951 strcat(mti->mti_params, " "PARAM_SEC_RPC_MDT);
952 strcat(mti->mti_params, lmd->lmd_sec_mdt);
953 //mti->mti_flags |= LDD_F_UPDATE;
955 if (lmd->lmd_sec_cli) {
956 if (strlen(mti->mti_params) + strlen(lmd->lmd_sec_cli) +
957 sizeof(PARAM_SEC_RPC_CLI) + 2 > sizeof(mti->mti_params)) {
958 CERROR("security params too big for mti\n");
961 strcat(mti->mti_params, " "PARAM_SEC_RPC_CLI);
962 strcat(mti->mti_params, lmd->lmd_sec_cli);
963 //mti->mti_flags |= LDD_F_UPDATE;
969 static int server_sb2mti(struct super_block *sb, struct mgs_target_info *mti)
971 struct lustre_sb_info *lsi = s2lsi(sb);
972 struct lustre_disk_data *ldd = lsi->lsi_ldd;
973 struct lustre_mount_data *lmd = lsi->lsi_lmd;
974 lnet_process_id_t id;
978 if (!(lsi->lsi_flags & LSI_SERVER))
981 strncpy(mti->mti_fsname, ldd->ldd_fsname,
982 sizeof(mti->mti_fsname));
983 strncpy(mti->mti_svname, ldd->ldd_svname,
984 sizeof(mti->mti_svname));
986 mti->mti_nid_count = 0;
987 while (LNetGetId(i++, &id) != -ENOENT) {
988 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
990 mti->mti_nids[mti->mti_nid_count] = id.nid;
991 mti->mti_nid_count++;
992 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
993 CWARN("Only using first %d nids for %s\n",
994 mti->mti_nid_count, mti->mti_svname);
999 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1000 mti->mti_config_ver = 0;
1001 mti->mti_flags = ldd->ldd_flags;
1002 mti->mti_stripe_index = ldd->ldd_svindex;
1003 memcpy(mti->mti_uuid, ldd->ldd_uuid, sizeof(mti->mti_uuid));
1004 if (strlen(ldd->ldd_params) > sizeof(mti->mti_params)) {
1005 CERROR("params too big for mti\n");
1008 memcpy(mti->mti_params, ldd->ldd_params, sizeof(mti->mti_params));
1010 RETURN(mti_set_sec_opts(mti, lmd));
1013 /* Register an old or new target with the MGS. If needed MGS will construct
1014 startup logs and assign index */
1015 int server_register_target(struct super_block *sb)
1017 struct lustre_sb_info *lsi = s2lsi(sb);
1018 struct obd_device *mgc = lsi->lsi_mgc;
1019 struct lustre_disk_data *ldd = lsi->lsi_ldd;
1020 struct mgs_target_info *mti = NULL;
1026 if (!(lsi->lsi_flags & LSI_SERVER))
1032 rc = server_sb2mti(sb, mti);
1036 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1037 mti->mti_svname, mti->mti_fsname,
1038 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1041 /* Register the target */
1042 /* FIXME use mgc_process_config instead */
1043 rc = obd_set_info_async(mgc->u.cli.cl_mgc_mgsexp,
1044 strlen("register_target"), "register_target",
1045 sizeof(*mti), mti, NULL);
1047 CERROR("registration with the MGS failed (%d)\n", rc);
1051 /* Always update our flags */
1052 ldd->ldd_flags = mti->mti_flags & ~LDD_F_REWRITE_LDD;
1054 /* If this flag is set, it means the MGS wants us to change our
1055 on-disk data. (So far this means just the index.) */
1056 if (mti->mti_flags & LDD_F_REWRITE_LDD) {
1059 CDEBUG(D_MOUNT, "Changing on-disk index from %#x to %#x "
1060 "for %s\n", ldd->ldd_svindex, mti->mti_stripe_index,
1062 ldd->ldd_svindex = mti->mti_stripe_index;
1063 strncpy(ldd->ldd_svname, mti->mti_svname,
1064 sizeof(ldd->ldd_svname));
1065 /* or ldd_make_sv_name(ldd); */
1066 ldd_write(&mgc->obd_lvfs_ctxt, ldd);
1067 err = fsfilt_set_label(mgc, lsi->lsi_srv_mnt->mnt_sb,
1070 CERROR("Label set error %d\n", err);
1071 label = fsfilt_get_label(mgc, lsi->lsi_srv_mnt->mnt_sb);
1073 CDEBUG(D_MOUNT, "Disk label changed to %s\n", label);
1075 /* Flush the new ldd to disk */
1076 fsfilt_sync(mgc, lsi->lsi_srv_mnt->mnt_sb);
1086 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1088 struct obd_device *obd;
1089 struct lustre_sb_info *lsi = s2lsi(sb);
1090 struct config_llog_instance cfg;
1094 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_ldd->ldd_svname);
1097 /* If we're an MDT, make sure the global MDS is running */
1098 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1099 /* make sure the MDS is started */
1100 mutex_down(&server_start_lock);
1101 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1103 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1104 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1106 LUSTRE_MDS_OBDNAME"_uuid",
1109 mutex_up(&server_start_lock);
1110 CERROR("failed to start MDS: %d\n", rc);
1114 mutex_up(&server_start_lock);
1118 /* If we're an OST, make sure the global OSS is running */
1119 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_OST) {
1120 /* make sure OSS is started */
1121 mutex_down(&server_start_lock);
1122 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1124 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1126 LUSTRE_OSS_OBDNAME"_uuid",
1129 mutex_up(&server_start_lock);
1130 CERROR("failed to start OSS: %d\n", rc);
1134 mutex_up(&server_start_lock);
1137 /* Set the mgc fs to our server disk. This allows the MGC
1138 to read and write configs locally. */
1139 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1143 /* Register with MGS */
1144 rc = server_register_target(sb);
1145 if (rc && (lsi->lsi_ldd->ldd_flags &
1146 (LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_UPGRADE14))){
1147 CERROR("Required registration failed for %s: %d\n",
1148 lsi->lsi_ldd->ldd_svname, rc);
1150 LCONSOLE_ERROR_MSG(0x15f, "Communication error with "
1151 "the MGS. Is the MGS running?\n");
1155 if (rc == -EINVAL) {
1156 LCONSOLE_ERROR_MSG(0x160, "The MGS is refusing to allow this "
1157 "server (%s) to start. Please see messages"
1158 " on the MGS node.\n",
1159 lsi->lsi_ldd->ldd_svname);
1163 /* Let the target look up the mount using the target's name
1164 (we can't pass the sb or mnt through class_process_config.) */
1165 rc = server_register_mount(lsi->lsi_ldd->ldd_svname, sb, mnt);
1169 /* Start targets using the llog named for the target */
1170 memset(&cfg, 0, sizeof(cfg));
1171 rc = lustre_process_log(sb, lsi->lsi_ldd->ldd_svname, &cfg);
1173 CERROR("failed to start server %s: %d\n",
1174 lsi->lsi_ldd->ldd_svname, rc);
1179 /* Release the mgc fs for others to use */
1180 server_mgc_clear_fs(lsi->lsi_mgc);
1183 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1185 CERROR("no server named %s was started\n",
1186 lsi->lsi_ldd->ldd_svname);
1190 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1191 (OBP(obd, iocontrol))) {
1192 obd_iocontrol(OBD_IOC_ABORT_RECOVERY,
1193 obd->obd_self_export, 0, NULL, NULL);
1196 /* log has been fully processed */
1197 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1203 /***************** lustre superblock **************/
1205 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1207 struct lustre_sb_info *lsi = NULL;
1210 OBD_ALLOC(lsi, sizeof(*lsi));
1213 OBD_ALLOC(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1214 if (!lsi->lsi_lmd) {
1215 OBD_FREE(lsi, sizeof(*lsi));
1219 lsi->lsi_lmd->lmd_exclude_count = 0;
1220 s2lsi_nocast(sb) = lsi;
1221 /* we take 1 extra ref for our setup */
1222 atomic_set(&lsi->lsi_mounts, 1);
1224 /* Default umount style */
1225 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1227 lsi->lsi_lmd->lmd_nllu = NOBODY_UID;
1228 lsi->lsi_lmd->lmd_nllg = NOBODY_GID;
1232 static int lustre_free_lsi(struct super_block *sb)
1234 struct lustre_sb_info *lsi = s2lsi(sb);
1240 CDEBUG(D_MOUNT, "Freeing lsi\n");
1242 /* someone didn't call server_put_mount. */
1243 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
1245 if (lsi->lsi_ldd != NULL)
1246 OBD_FREE(lsi->lsi_ldd, sizeof(*lsi->lsi_ldd));
1248 if (lsi->lsi_lmd != NULL) {
1249 if (lsi->lsi_lmd->lmd_dev != NULL)
1250 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1251 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1252 if (lsi->lsi_lmd->lmd_profile != NULL)
1253 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1254 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1255 if (lsi->lsi_lmd->lmd_sec_mdt != NULL)
1256 OBD_FREE(lsi->lsi_lmd->lmd_sec_mdt,
1257 strlen(lsi->lsi_lmd->lmd_sec_mdt) + 1);
1258 if (lsi->lsi_lmd->lmd_sec_cli != NULL)
1259 OBD_FREE(lsi->lsi_lmd->lmd_sec_cli,
1260 strlen(lsi->lsi_lmd->lmd_sec_cli) + 1);
1261 if (lsi->lsi_lmd->lmd_opts != NULL)
1262 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1263 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1264 if (lsi->lsi_lmd->lmd_exclude_count)
1265 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1266 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1267 lsi->lsi_lmd->lmd_exclude_count);
1268 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1271 LASSERT(lsi->lsi_llsbi == NULL);
1272 OBD_FREE(lsi, sizeof(*lsi));
1273 s2lsi_nocast(sb) = NULL;
1278 /* The lsi has one reference for every server that is using the disk -
1279 e.g. MDT, MGS, and potentially MGC */
1280 static int lustre_put_lsi(struct super_block *sb)
1282 struct lustre_sb_info *lsi = s2lsi(sb);
1287 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
1289 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
1290 lustre_free_lsi(sb);
1296 /*************** server mount ******************/
1298 /* Kernel mount using mount options in MOUNT_DATA_FILE */
1299 static struct vfsmount *server_kernel_mount(struct super_block *sb)
1301 struct lvfs_run_ctxt mount_ctxt;
1302 struct lustre_sb_info *lsi = s2lsi(sb);
1303 struct lustre_disk_data *ldd;
1304 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1305 struct vfsmount *mnt;
1306 char *options = NULL;
1307 unsigned long page, s_flags;
1311 OBD_ALLOC(ldd, sizeof(*ldd));
1313 RETURN(ERR_PTR(-ENOMEM));
1315 /* In the past, we have always used flags = 0.
1316 Note ext3/ldiskfs can't be mounted ro. */
1317 s_flags = sb->s_flags;
1319 /* Pre-mount ldiskfs to read the MOUNT_DATA_FILE */
1320 CDEBUG(D_MOUNT, "Pre-mount ldiskfs %s\n", lmd->lmd_dev);
1321 mnt = ll_kern_mount("ldiskfs", s_flags, lmd->lmd_dev, 0);
1324 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
1325 /* 2.6 kernels: if ldiskfs fails, try ldiskfs2 */
1326 mnt = ll_kern_mount("ldiskfs2", s_flags, lmd->lmd_dev, 0);
1328 int rc2 = PTR_ERR(mnt);
1329 CERROR("premount %s:%#lx ldiskfs failed: %d, ldiskfs2 "
1330 "failed: %d. Is the ldiskfs module available?\n",
1331 lmd->lmd_dev, s_flags, rc, rc2);
1335 /* 2.4 kernels: if ldiskfs fails, try ext3 */
1336 mnt = ll_kern_mount("ext3", s_flags, lmd->lmd_dev, 0);
1339 CERROR("premount ext3 failed: rc = %d\n", rc);
1345 OBD_SET_CTXT_MAGIC(&mount_ctxt);
1346 mount_ctxt.pwdmnt = mnt;
1347 mount_ctxt.pwd = mnt->mnt_root;
1348 mount_ctxt.fs = get_ds();
1350 rc = ldd_parse(&mount_ctxt, ldd);
1354 CERROR("premount parse options failed: rc = %d\n", rc);
1358 /* Done with our pre-mount, now do the real mount. */
1360 /* Glom up mount options */
1361 page = __get_free_page(GFP_KERNEL);
1363 GOTO(out_free, rc = -ENOMEM);
1365 options = (char *)page;
1366 memset(options, 0, CFS_PAGE_SIZE);
1367 strncpy(options, ldd->ldd_mount_opts, CFS_PAGE_SIZE - 2);
1369 /* Add in any mount-line options */
1370 if (lmd->lmd_opts && (*(lmd->lmd_opts) != 0)) {
1371 int len = CFS_PAGE_SIZE - strlen(options) - 2;
1373 strcat(options, ",");
1374 strncat(options, lmd->lmd_opts, len);
1377 /* Special permanent mount flags */
1379 s_flags |= MS_NOATIME | MS_NODIRATIME;
1381 CDEBUG(D_MOUNT, "kern_mount: %s %s %s\n",
1382 MT_STR(ldd), lmd->lmd_dev, options);
1383 mnt = ll_kern_mount(MT_STR(ldd), s_flags, lmd->lmd_dev,
1388 CERROR("ll_kern_mount failed: rc = %d\n", rc);
1392 lsi->lsi_ldd = ldd; /* freed at lsi cleanup */
1393 CDEBUG(D_SUPER, "%s: mnt = %p\n", lmd->lmd_dev, mnt);
1397 OBD_FREE(ldd, sizeof(*ldd));
1398 lsi->lsi_ldd = NULL;
1399 RETURN(ERR_PTR(rc));
1402 static void server_wait_finished(struct vfsmount *mnt)
1404 wait_queue_head_t waitq;
1405 struct l_wait_info lwi;
1408 init_waitqueue_head(&waitq);
1410 while ((atomic_read(&mnt->mnt_count) > 1) && (retries > 0)) {
1411 LCONSOLE_WARN("Mount still busy with %d refs, waiting for "
1413 atomic_read(&mnt->mnt_count), retries);
1415 /* Wait for a bit */
1417 lwi = LWI_TIMEOUT(5 * HZ, NULL, NULL);
1418 l_wait_event(waitq, 0, &lwi);
1420 if (atomic_read(&mnt->mnt_count) > 1) {
1421 CERROR("Mount %p is still busy (%d refs), giving up.\n",
1422 mnt, atomic_read(&mnt->mnt_count));
1426 static void server_put_super(struct super_block *sb)
1428 struct lustre_sb_info *lsi = s2lsi(sb);
1429 struct obd_device *obd;
1430 struct vfsmount *mnt = lsi->lsi_srv_mnt;
1431 char *tmpname, *extraname = NULL;
1433 int lddflags = lsi->lsi_ldd->ldd_flags;
1434 int lsiflags = lsi->lsi_flags;
1438 LASSERT(lsiflags & LSI_SERVER);
1440 tmpname_sz = strlen(lsi->lsi_ldd->ldd_svname) + 1;
1441 OBD_ALLOC(tmpname, tmpname_sz);
1442 memcpy(tmpname, lsi->lsi_ldd->ldd_svname, tmpname_sz);
1443 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1445 /* Stop the target */
1446 if (IS_MDT(lsi->lsi_ldd) || IS_OST(lsi->lsi_ldd)) {
1447 struct lustre_profile *lprof = NULL;
1449 /* tell the mgc to drop the config log */
1450 lustre_end_log(sb, lsi->lsi_ldd->ldd_svname, NULL);
1452 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1453 If there are any setup/cleanup errors, save the lov
1454 name for safety cleanup later. */
1455 lprof = class_get_profile(lsi->lsi_ldd->ldd_svname);
1456 if (lprof && lprof->lp_dt) {
1457 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1458 strcpy(extraname, lprof->lp_dt);
1461 obd = class_name2obd(lsi->lsi_ldd->ldd_svname);
1463 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1464 if (lsi->lsi_flags & LSI_UMOUNT_FORCE)
1466 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
1468 /* We can't seem to give an error return code
1469 to .put_super, so we better make sure we clean up! */
1471 class_manual_cleanup(obd);
1473 CERROR("no obd %s\n", lsi->lsi_ldd->ldd_svname);
1474 server_deregister_mount(lsi->lsi_ldd->ldd_svname);
1478 /* If they wanted the mgs to stop separately from the mdt, they
1479 should have put it on a different device. */
1480 if (IS_MGS(lsi->lsi_ldd)) {
1481 /* stop the mgc before the mgs so the connection gets cleaned
1483 lustre_stop_mgc(sb);
1484 server_stop_mgs(sb);
1487 /* Clean the mgc and sb */
1488 rc = lustre_common_put_super(sb);
1489 /* FIXME how can I report a failure to umount? */
1491 /* Wait for the targets to really clean up - can't exit (and let the
1492 sb get destroyed) while the mount is still in use */
1493 server_wait_finished(mnt);
1495 /* drop the One True Mount */
1498 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1499 until the target is really gone so that our type refcount check
1501 server_stop_servers(lddflags, lsiflags);
1503 /* In case of startup or cleanup err, stop related obds */
1505 obd = class_name2obd(extraname);
1507 CWARN("Cleaning orphaned obd %s\n", extraname);
1509 class_manual_cleanup(obd);
1511 OBD_FREE(extraname, strlen(extraname) + 1);
1514 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1515 OBD_FREE(tmpname, tmpname_sz);
1519 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1520 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
1522 struct super_block *sb = vfsmnt->mnt_sb;
1524 static void server_umount_begin(struct super_block *sb)
1527 struct lustre_sb_info *lsi = s2lsi(sb);
1530 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
1531 if (!(flags & MNT_FORCE)) {
1537 CDEBUG(D_MOUNT, "umount -f\n");
1538 /* umount = failover
1540 no third way to do non-force, non-failover */
1541 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1542 lsi->lsi_flags |= LSI_UMOUNT_FORCE;
1546 #ifndef HAVE_STATFS_DENTRY_PARAM
1547 static int server_statfs (struct super_block *sb, struct kstatfs *buf)
1550 static int server_statfs (struct dentry *dentry, struct kstatfs *buf)
1552 struct super_block *sb = dentry->d_sb;
1554 struct vfsmount *mnt = s2lsi(sb)->lsi_srv_mnt;
1557 if (mnt && mnt->mnt_sb && mnt->mnt_sb->s_op->statfs) {
1558 #ifdef HAVE_STATFS_DENTRY_PARAM
1559 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_root, buf);
1561 int rc = mnt->mnt_sb->s_op->statfs(mnt->mnt_sb, buf);
1564 buf->f_type = sb->s_magic;
1570 buf->f_type = sb->s_magic;
1571 buf->f_bsize = sb->s_blocksize;
1577 buf->f_namelen = NAME_MAX;
1581 static struct super_operations server_ops =
1583 .put_super = server_put_super,
1584 .umount_begin = server_umount_begin, /* umount -f */
1585 .statfs = server_statfs,
1588 #define log2(n) ffz(~(n))
1589 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1591 static int server_fill_super_common(struct super_block *sb)
1593 struct inode *root = 0;
1596 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1598 sb->s_blocksize = 4096;
1599 sb->s_blocksize_bits = log2(sb->s_blocksize);
1600 sb->s_magic = LUSTRE_SUPER_MAGIC;
1601 sb->s_maxbytes = 0; //PAGE_CACHE_MAXBYTES;
1602 sb->s_flags |= MS_RDONLY;
1603 sb->s_op = &server_ops;
1605 root = new_inode(sb);
1607 CERROR("Can't make root inode\n");
1611 /* returns -EIO for every operation */
1612 /* make_bad_inode(root); -- badness - can't umount */
1613 /* apparently we need to be a directory for the mount to finish */
1614 root->i_mode = S_IFDIR;
1616 sb->s_root = d_alloc_root(root);
1618 CERROR("Can't make root dentry\n");
1626 static int server_fill_super(struct super_block *sb)
1628 struct lustre_sb_info *lsi = s2lsi(sb);
1629 struct vfsmount *mnt;
1633 /* the One True Mount */
1634 mnt = server_kernel_mount(sb);
1637 CERROR("Unable to mount device %s: %d\n",
1638 lsi->lsi_lmd->lmd_dev, rc);
1642 lsi->lsi_srv_mnt = mnt;
1644 LASSERT(lsi->lsi_ldd);
1645 CDEBUG(D_MOUNT, "Found service %s for fs '%s' on device %s\n",
1646 lsi->lsi_ldd->ldd_svname, lsi->lsi_ldd->ldd_fsname,
1647 lsi->lsi_lmd->lmd_dev);
1649 if (class_name2obd(lsi->lsi_ldd->ldd_svname)) {
1650 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1651 "running. Double-mount may have compromised"
1652 " the disk journal.\n",
1653 lsi->lsi_ldd->ldd_svname);
1656 GOTO(out, rc = -EALREADY);
1659 /* start MGS before MGC */
1660 if (IS_MGS(lsi->lsi_ldd)) {
1661 rc = server_start_mgs(sb);
1666 rc = lustre_start_mgc(sb);
1670 /* Set up all obd devices for service */
1671 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1672 (IS_OST(lsi->lsi_ldd) || IS_MDT(lsi->lsi_ldd))) {
1673 rc = server_start_targets(sb, mnt);
1675 CERROR("Unable to start targets: %d\n", rc);
1678 /* FIXME overmount client here,
1679 or can we just start a client log and client_fill_super on this sb?
1680 We need to make sure server_put_super gets called too - ll_put_super
1681 calls lustre_common_put_super; check there for LSI_SERVER flag,
1683 Probably should start client from new thread so we can return.
1684 Client will not finish until all servers are connected.
1685 Note - MGS-only server does NOT get a client, since there is no
1686 lustre fs associated - the MGS is for all lustre fs's */
1689 rc = server_fill_super_common(sb);
1693 LCONSOLE_WARN("Server %s on device %s has started\n",
1694 lsi->lsi_ldd->ldd_svname, lsi->lsi_lmd->lmd_dev);
1699 server_put_super(sb);
1704 /* Get the index from the obd name.
1705 rc = server type, or
1707 if endptr isn't NULL it is set to end of name */
1708 int server_name2index(char *svname, __u32 *idx, char **endptr)
1710 unsigned long index;
1712 char *dash = strchr(svname, '-');
1716 if (strncmp(dash + 1, "MDT", 3) == 0)
1717 rc = LDD_F_SV_TYPE_MDT;
1718 else if (strncmp(dash + 1, "OST", 3) == 0)
1719 rc = LDD_F_SV_TYPE_OST;
1723 index = simple_strtoul(dash + 4, endptr, 16);
1728 /*************** mount common betweeen server and client ***************/
1731 int lustre_common_put_super(struct super_block *sb)
1736 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
1738 /* Drop a ref to the MGC */
1739 rc = lustre_stop_mgc(sb);
1740 if (rc && (rc != -ENOENT)) {
1742 CERROR("Can't stop MGC: %d\n", rc);
1745 /* BUSY just means that there's some other obd that
1746 needs the mgc. Let him clean it up. */
1747 CDEBUG(D_MOUNT, "MGC still in use\n");
1749 /* Drop a ref to the mounted disk */
1755 static void lmd_print(struct lustre_mount_data *lmd)
1759 PRINT_CMD(PRINT_MASK, " mount data:\n");
1760 if (lmd_is_client(lmd))
1761 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
1762 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
1763 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
1764 if (lmd->lmd_sec_mdt)
1765 PRINT_CMD(PRINT_MASK, "sec_mdt: %s\n", lmd->lmd_sec_mdt);
1766 if (lmd->lmd_sec_cli)
1767 PRINT_CMD(PRINT_MASK, "sec_cli: %s\n", lmd->lmd_sec_cli);
1769 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
1770 for (i = 0; i < lmd->lmd_exclude_count; i++) {
1771 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
1772 lmd->lmd_exclude[i]);
1777 /* Is this server on the exclusion list */
1778 int lustre_check_exclusion(struct super_block *sb, char *svname)
1780 struct lustre_sb_info *lsi = s2lsi(sb);
1781 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1786 rc = server_name2index(svname, &index, NULL);
1787 if (rc != LDD_F_SV_TYPE_OST)
1788 /* Only exclude OSTs */
1791 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
1792 index, lmd->lmd_exclude_count, lmd->lmd_dev);
1794 for(i = 0; i < lmd->lmd_exclude_count; i++) {
1795 if (index == lmd->lmd_exclude[i]) {
1796 CWARN("Excluding %s (on exclusion list)\n", svname);
1803 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
1804 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
1806 char *s1 = ptr, *s2;
1807 __u32 index, *exclude_list;
1811 /* The shortest an ost name can be is 8 chars: -OST0000.
1812 We don't actually know the fsname at this time, so in fact
1813 a user could specify any fsname. */
1814 devmax = strlen(ptr) / 8 + 1;
1816 /* temp storage until we figure out how many we have */
1817 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
1821 /* we enter this fn pointing at the '=' */
1822 while (*s1 && *s1 != ' ' && *s1 != ',') {
1824 rc = server_name2index(s1, &index, &s2);
1826 CERROR("Can't parse server name '%s'\n", s1);
1829 if (rc == LDD_F_SV_TYPE_OST)
1830 exclude_list[lmd->lmd_exclude_count++] = index;
1832 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
1834 /* now we are pointing at ':' (next exclude)
1835 or ',' (end of excludes) */
1836 if (lmd->lmd_exclude_count >= devmax)
1839 if (rc >= 0) /* non-err */
1842 if (lmd->lmd_exclude_count) {
1843 /* permanent, freed in lustre_free_lsi */
1844 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
1845 lmd->lmd_exclude_count);
1846 if (lmd->lmd_exclude) {
1847 memcpy(lmd->lmd_exclude, exclude_list,
1848 sizeof(index) * lmd->lmd_exclude_count);
1851 lmd->lmd_exclude_count = 0;
1854 OBD_FREE(exclude_list, sizeof(index) * devmax);
1859 int lmd_set_sec_opts(char **set, char *opts, int length)
1862 OBD_FREE(*set, strlen(*set) + 1);
1864 OBD_ALLOC(*set, length + 1);
1868 memcpy(*set, opts, length);
1869 (*set)[length] = '\0';
1875 int lmd_parse_sec_opts(struct lustre_mount_data *lmd, char *ptr)
1881 /* check peer name */
1882 if (strncmp(ptr, "sec_mdt=", 8) == 0) {
1883 set = &lmd->lmd_sec_mdt;
1885 } else if (strncmp(ptr, "sec_cli=", 8) == 0) {
1886 set = &lmd->lmd_sec_cli;
1888 } else if (strncmp(ptr, "sec=", 4) == 0) {
1889 /* leave 'set' be null */
1892 CERROR("invalid security options: %s\n", ptr);
1896 tail = strchr(ptr, ',');
1898 length = strlen(ptr);
1900 length = tail - ptr;
1903 if (lmd_set_sec_opts(set, ptr, length))
1906 if (lmd->lmd_sec_mdt == NULL &&
1907 lmd_set_sec_opts(&lmd->lmd_sec_mdt, ptr, length))
1910 if (lmd->lmd_sec_cli == NULL &&
1911 lmd_set_sec_opts(&lmd->lmd_sec_cli, ptr, length))
1918 /* mount -v -t lustre uml1:uml2:/lustre-client /mnt/lustre */
1919 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1921 char *s1, *s2, *devname = NULL;
1922 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1928 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1929 "/sbin/mount.lustre is installed.\n");
1933 /* Options should be a string - try to detect old lmd data */
1934 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1935 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1936 "/sbin/mount.lustre. Please install "
1937 "version %s\n", LUSTRE_VERSION_STRING);
1940 lmd->lmd_magic = LMD_MAGIC;
1942 /* Set default flags here */
1947 /* Skip whitespace and extra commas */
1948 while (*s1 == ' ' || *s1 == ',')
1951 /* Client options are parsed in ll_options: eg. flock,
1954 /* Parse non-ldiskfs options here. Rather than modifying
1955 ldiskfs, we just zero these out here */
1956 if (strncmp(s1, "abort_recov", 11) == 0) {
1957 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1959 } else if (strncmp(s1, "nosvc", 5) == 0) {
1960 lmd->lmd_flags |= LMD_FLG_NOSVC;
1962 /* ost exclusion list */
1963 } else if (strncmp(s1, "exclude=", 8) == 0) {
1964 rc = lmd_make_exclusion(lmd, s1 + 7);
1968 } else if (strncmp(s1, "nllu=", 5) == 0) {
1969 lmd->lmd_nllu = simple_strtoul(s1 + 5, NULL, 10);
1971 } else if (strncmp(s1, "nllg=", 5) == 0) {
1972 lmd->lmd_nllg = simple_strtoul(s1 + 5, NULL, 10);
1974 } else if (strncmp(s1, "sec", 3) == 0) {
1975 rc = lmd_parse_sec_opts(lmd, s1);
1980 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1981 end of the options. */
1982 else if (strncmp(s1, "device=", 7) == 0) {
1984 /* terminate options right before device. device
1985 must be the last one. */
1991 s2 = strchr(s1, ',');
1999 memmove(s1, s2, strlen(s2) + 1);
2005 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
2006 "(need mount option 'device=...')\n");
2010 s1 = strrchr(devname, ':');
2012 lmd->lmd_flags = LMD_FLG_CLIENT;
2013 /* Remove leading /s from fsname */
2014 while (*++s1 == '/') ;
2015 /* Freed in lustre_free_lsi */
2016 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
2017 if (!lmd->lmd_profile)
2019 sprintf(lmd->lmd_profile, "%s-client", s1);
2022 /* Freed in lustre_free_lsi */
2023 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
2026 strcpy(lmd->lmd_dev, devname);
2028 /* Save mount options */
2029 s1 = options + strlen(options) - 1;
2030 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
2032 if (*options != 0) {
2033 /* Freed in lustre_free_lsi */
2034 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
2037 strcpy(lmd->lmd_opts, options);
2040 lmd->lmd_magic = LMD_MAGIC;
2045 CERROR("Bad mount options %s\n", options);
2051 int lustre_fill_super(struct super_block *sb, void *data, int silent)
2053 struct lustre_mount_data *lmd;
2054 struct lustre_sb_info *lsi;
2058 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
2060 lsi = lustre_init_lsi(sb);
2065 /* Figure out the lmd from the mount options */
2066 if (lmd_parse((char *)data, lmd)) {
2071 if (lmd_is_client(lmd)) {
2072 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
2073 if (!client_fill_super) {
2074 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
2075 "client mount! Is the 'lustre' "
2076 "module loaded?\n");
2079 rc = lustre_start_mgc(sb);
2081 lustre_stop_mgc(sb);
2084 /* Connect and start */
2085 /* (should always be ll_fill_super) */
2086 rc = (*client_fill_super)(sb);
2087 /* c_f_s will call lustre_common_put_super on failure */
2090 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
2091 lsi->lsi_flags |= LSI_SERVER;
2092 rc = server_fill_super(sb);
2093 /* s_f_s calls lustre_start_mgc after the mount because we need
2094 the MGS nids which are stored on disk. Plus, we may
2095 need to start the MGS first. */
2096 /* s_f_s will call server_put_super on failure */
2101 CERROR("Unable to mount %s (%d)\n",
2102 s2lsi(sb) ? lmd->lmd_dev : "", rc);
2104 CDEBUG(D_SUPER, "mount %s complete\n", lmd->lmd_dev);
2110 /* We can't call ll_fill_super by name because it lives in a module that
2111 must be loaded after this one. */
2112 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb))
2114 client_fill_super = cfs;
2117 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
2119 kill_super_cb = cfs;
2122 /***************** FS registration ******************/
2124 #if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
2126 #if (LINUX_VERSION_CODE < KERNEL_VERSION(2,6,18))
2127 struct super_block * lustre_get_sb(struct file_system_type *fs_type,
2128 int flags, const char *devname, void * data)
2130 /* calls back in fill super */
2131 /* we could append devname= onto options (*data) here,
2132 but 2.4 doesn't get devname. So we do it in mount_lustre.c */
2133 return get_sb_nodev(fs_type, flags, data, lustre_fill_super);
2136 int lustre_get_sb(struct file_system_type *fs_type,
2137 int flags, const char *devname, void * data,
2138 struct vfsmount *mnt)
2140 /* calls back in fill super */
2141 /* we could append devname= onto options (*data) here,
2142 but 2.4 doesn't get devname. So we do it in mount_lustre.c */
2143 return get_sb_nodev(fs_type, flags, data, lustre_fill_super, mnt);
2147 void lustre_kill_super(struct super_block *sb)
2149 struct lustre_sb_info *lsi = s2lsi(sb);
2151 if (kill_super_cb && lsi &&(lsi->lsi_flags & LSI_SERVER))
2152 (*kill_super_cb)(sb);
2154 kill_anon_super(sb);
2157 struct file_system_type lustre_fs_type = {
2158 .owner = THIS_MODULE,
2160 .get_sb = lustre_get_sb,
2161 .kill_sb = lustre_kill_super,
2162 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV,
2167 static struct super_block *lustre_read_super(struct super_block *sb,
2168 void *data, int silent)
2173 rc = lustre_fill_super(sb, data, silent);
2179 static struct file_system_type lustre_fs_type = {
2180 .owner = THIS_MODULE,
2182 .fs_flags = FS_NFSEXP_FSID,
2183 .read_super = lustre_read_super,
2187 int lustre_register_fs(void)
2189 return register_filesystem(&lustre_fs_type);
2192 int lustre_unregister_fs(void)
2194 return unregister_filesystem(&lustre_fs_type);
2197 EXPORT_SYMBOL(lustre_register_client_fill_super);
2198 EXPORT_SYMBOL(lustre_register_kill_super_cb);
2199 EXPORT_SYMBOL(lustre_common_put_super);
2200 EXPORT_SYMBOL(lustre_process_log);
2201 EXPORT_SYMBOL(lustre_end_log);
2202 EXPORT_SYMBOL(server_get_mount);
2203 EXPORT_SYMBOL(server_get_mount_2);
2204 EXPORT_SYMBOL(server_put_mount);
2205 EXPORT_SYMBOL(server_put_mount_2);
2206 EXPORT_SYMBOL(server_register_target);
2207 EXPORT_SYMBOL(server_name2index);
2208 EXPORT_SYMBOL(server_mti_print);
2209 EXPORT_SYMBOL(do_lcfg);