4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/obdclass/obd_mount.c
38 * Client/server mount routines
40 * Author: Nathan Rutman <nathan@clusterfs.com>
44 #define DEBUG_SUBSYSTEM S_CLASS
45 #define D_MOUNT D_SUPER|D_CONFIG /*|D_WARNING */
46 #define PRINT_CMD CDEBUG
47 #define PRINT_MASK D_SUPER|D_CONFIG
51 #include <lustre_fsfilt.h>
52 #include <obd_class.h>
53 #include <lustre/lustre_user.h>
54 #include <linux/version.h>
55 #include <lustre_log.h>
56 #include <lustre_disk.h>
57 #include <lustre_param.h>
58 #ifdef HAVE_KERNEL_LOCKED
59 #include <linux/smp_lock.h>
62 static int (*client_fill_super)(struct super_block *sb,
63 struct vfsmount *mnt) = NULL;
64 static void (*kill_super_cb)(struct super_block *sb) = NULL;
66 /*********** mount lookup *********/
68 DEFINE_MUTEX(lustre_mount_info_lock);
69 static CFS_LIST_HEAD(server_mount_info_list);
71 static struct lustre_mount_info *server_find_mount(const char *name)
74 struct lustre_mount_info *lmi;
77 cfs_list_for_each(tmp, &server_mount_info_list) {
78 lmi = cfs_list_entry(tmp, struct lustre_mount_info,
80 if (strcmp(name, lmi->lmi_name) == 0)
86 /* we must register an obd for a mount before we call the setup routine.
87 *_setup will call lustre_get_mount to get the mnt struct
88 by obd_name, since we can't pass the pointer to setup. */
89 static int server_register_mount(const char *name, struct super_block *sb,
92 struct lustre_mount_info *lmi;
98 OBD_ALLOC(lmi, sizeof(*lmi));
101 OBD_ALLOC(name_cp, strlen(name) + 1);
103 OBD_FREE(lmi, sizeof(*lmi));
106 strcpy(name_cp, name);
108 mutex_lock(&lustre_mount_info_lock);
110 if (server_find_mount(name)) {
111 mutex_unlock(&lustre_mount_info_lock);
112 OBD_FREE(lmi, sizeof(*lmi));
113 OBD_FREE(name_cp, strlen(name) + 1);
114 CERROR("Already registered %s\n", name);
117 lmi->lmi_name = name_cp;
120 cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list);
122 mutex_unlock(&lustre_mount_info_lock);
124 CDEBUG(D_MOUNT, "reg_mnt %p from %s\n", lmi->lmi_mnt, name);
129 /* when an obd no longer needs a mount */
130 static int server_deregister_mount(const char *name)
132 struct lustre_mount_info *lmi;
135 mutex_lock(&lustre_mount_info_lock);
136 lmi = server_find_mount(name);
138 mutex_unlock(&lustre_mount_info_lock);
139 CERROR("%s not registered\n", name);
143 CDEBUG(D_MOUNT, "dereg_mnt %p from %s\n", lmi->lmi_mnt, name);
145 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
146 cfs_list_del(&lmi->lmi_list_chain);
147 OBD_FREE(lmi, sizeof(*lmi));
148 mutex_unlock(&lustre_mount_info_lock);
153 /* obd's look up a registered mount using their obdname. This is just
154 for initial obd setup to find the mount struct. It should not be
155 called every time you want to mntget. */
156 struct lustre_mount_info *server_get_mount(const char *name)
158 struct lustre_mount_info *lmi;
159 struct lustre_sb_info *lsi;
162 mutex_lock(&lustre_mount_info_lock);
163 lmi = server_find_mount(name);
164 mutex_unlock(&lustre_mount_info_lock);
166 CERROR("Can't find mount for %s\n", name);
169 lsi = s2lsi(lmi->lmi_sb);
171 cfs_atomic_inc(&lsi->lsi_mounts);
173 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d\n", lmi->lmi_mnt,
174 name, cfs_atomic_read(&lsi->lsi_mounts));
178 EXPORT_SYMBOL(server_get_mount);
181 * Used by mdt to get mount_info from obdname.
182 * There are no blocking when using the mount_info.
183 * Do not use server_get_mount for this purpose.
185 struct lustre_mount_info *server_get_mount_2(const char *name)
187 struct lustre_mount_info *lmi;
190 mutex_lock(&lustre_mount_info_lock);
191 lmi = server_find_mount(name);
192 mutex_unlock(&lustre_mount_info_lock);
194 CERROR("Can't find mount for %s\n", name);
198 EXPORT_SYMBOL(server_get_mount_2);
200 static int lustre_put_lsi(struct super_block *sb);
202 /* to be called from obd_cleanup methods */
203 int server_put_mount(const char *name, struct vfsmount *mnt)
205 struct lustre_mount_info *lmi;
206 struct lustre_sb_info *lsi;
209 mutex_lock(&lustre_mount_info_lock);
210 lmi = server_find_mount(name);
211 mutex_unlock(&lustre_mount_info_lock);
213 CERROR("Can't find mount for %s\n", name);
216 lsi = s2lsi(lmi->lmi_sb);
218 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d\n",
219 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts));
221 if (lustre_put_lsi(lmi->lmi_sb))
222 CDEBUG(D_MOUNT, "Last put of mnt %p from %s\n",
225 /* this obd should never need the mount again */
226 server_deregister_mount(name);
230 EXPORT_SYMBOL(server_put_mount);
232 /* Corresponding to server_get_mount_2 */
233 int server_put_mount_2(const char *name, struct vfsmount *mnt)
238 EXPORT_SYMBOL(server_put_mount_2);
240 /**************** config llog ********************/
242 /** Get a config log from the MGS and process it.
243 * This func is called for both clients and servers.
244 * Continue to process new statements appended to the logs
245 * (whenever the config lock is revoked) until lustre_end_log
247 * @param sb The superblock is used by the MGC to write to the local copy of
249 * @param logname The name of the llog to replicate from the MGS
250 * @param cfg Since the same mgc may be used to follow multiple config logs
251 * (e.g. ost1, ost2, client), the config_llog_instance keeps the state for
252 * this log, and is added to the mgc's list of logs to follow.
254 int lustre_process_log(struct super_block *sb, char *logname,
255 struct config_llog_instance *cfg)
257 struct lustre_cfg *lcfg;
258 struct lustre_cfg_bufs *bufs;
259 struct lustre_sb_info *lsi = s2lsi(sb);
260 struct obd_device *mgc = lsi->lsi_mgc;
271 /* mgc_process_config */
272 lustre_cfg_bufs_reset(bufs, mgc->obd_name);
273 lustre_cfg_bufs_set_string(bufs, 1, logname);
274 lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg));
275 lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb));
276 lcfg = lustre_cfg_new(LCFG_LOG_START, bufs);
277 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
278 lustre_cfg_free(lcfg);
283 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
284 "failed from the MGS (%d). Make sure this "
285 "client and the MGS are running compatible "
286 "versions of Lustre.\n",
287 mgc->obd_name, logname, rc);
290 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
291 "failed (%d). This may be the result of "
292 "communication errors between this node and "
293 "the MGS, a bad configuration, or other "
294 "errors. See the syslog for more "
295 "information.\n", mgc->obd_name, logname,
298 /* class_obd_list(); */
301 EXPORT_SYMBOL(lustre_process_log);
303 /* Stop watching this config log for updates */
304 int lustre_end_log(struct super_block *sb, char *logname,
305 struct config_llog_instance *cfg)
307 struct lustre_cfg *lcfg;
308 struct lustre_cfg_bufs bufs;
309 struct lustre_sb_info *lsi = s2lsi(sb);
310 struct obd_device *mgc = lsi->lsi_mgc;
317 /* mgc_process_config */
318 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
319 lustre_cfg_bufs_set_string(&bufs, 1, logname);
321 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
322 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
323 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
324 lustre_cfg_free(lcfg);
327 EXPORT_SYMBOL(lustre_end_log);
329 /**************** obd start *******************/
331 /** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from
332 * lctl (and do for echo cli/srv.
334 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
335 char *s1, char *s2, char *s3, char *s4)
337 struct lustre_cfg_bufs bufs;
338 struct lustre_cfg * lcfg = NULL;
341 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
342 cmd, s1, s2, s3, s4);
344 lustre_cfg_bufs_reset(&bufs, cfgname);
346 lustre_cfg_bufs_set_string(&bufs, 1, s1);
348 lustre_cfg_bufs_set_string(&bufs, 2, s2);
350 lustre_cfg_bufs_set_string(&bufs, 3, s3);
352 lustre_cfg_bufs_set_string(&bufs, 4, s4);
354 lcfg = lustre_cfg_new(cmd, &bufs);
355 lcfg->lcfg_nid = nid;
356 rc = class_process_config(lcfg);
357 lustre_cfg_free(lcfg);
360 EXPORT_SYMBOL(do_lcfg);
362 /** Call class_attach and class_setup. These methods in turn call
363 * obd type-specific methods.
365 static int lustre_start_simple(char *obdname, char *type, char *uuid,
366 char *s1, char *s2, char *s3, char *s4)
369 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
371 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
373 CERROR("%s attach error %d\n", obdname, rc);
376 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, s3, s4);
378 CERROR("%s setup error %d\n", obdname, rc);
379 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
384 /* Set up a MGS to serve startup logs */
385 static int server_start_mgs(struct super_block *sb)
387 struct lustre_sb_info *lsi = s2lsi(sb);
388 struct vfsmount *mnt = lsi->lsi_srv_mnt;
389 struct lustre_mount_info *lmi;
393 /* It is impossible to have more than 1 MGS per node, since
394 MGC wouldn't know which to connect to */
395 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
397 lsi = s2lsi(lmi->lmi_sb);
398 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
403 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
405 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
408 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
409 LUSTRE_MGS_OBDNAME, 0, 0,
410 lsi->lsi_osd_obdname, 0);
411 /* Do NOT call server_deregister_mount() here. This leads to
412 * inability cleanup cleanly and free lsi and other stuff when
413 * mgs calls server_put_mount() in error handling case. -umka */
417 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
418 "Is the 'mgs' module loaded?\n",
419 LUSTRE_MGS_OBDNAME, rc);
423 static int server_stop_mgs(struct super_block *sb)
425 struct obd_device *obd;
429 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
431 /* There better be only one MGS */
432 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
434 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
438 /* The MGS should always stop when we say so */
440 rc = class_manual_cleanup(obd);
444 DEFINE_MUTEX(mgc_start_lock);
446 /** Set up a mgc obd to process startup logs
448 * \param sb [in] super block of the mgc obd
450 * \retval 0 success, otherwise error code
452 static int lustre_start_mgc(struct super_block *sb)
454 struct obd_connect_data *data = NULL;
455 struct lustre_sb_info *lsi = s2lsi(sb);
456 struct obd_device *obd;
457 struct obd_export *exp;
458 struct obd_uuid *uuid;
461 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
464 int rc = 0, i = 0, j, len;
467 LASSERT(lsi->lsi_lmd);
469 /* Find the first non-lo MGS nid for our MGC name */
470 if (IS_SERVER(lsi)) {
471 /* mount -o mgsnode=nid */
472 ptr = lsi->lsi_lmd->lmd_mgs;
473 if (lsi->lsi_lmd->lmd_mgs &&
474 (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) {
476 } else if (IS_MGS(lsi)) {
477 lnet_process_id_t id;
478 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
479 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
486 } else { /* client */
487 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
488 ptr = lsi->lsi_lmd->lmd_dev;
489 if (class_parse_nid(ptr, &nid, &ptr) == 0)
493 CERROR("No valid MGS nids found.\n");
497 mutex_lock(&mgc_start_lock);
499 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
500 OBD_ALLOC(mgcname, len);
501 OBD_ALLOC(niduuid, len + 2);
502 if (!mgcname || !niduuid)
503 GOTO(out_free, rc = -ENOMEM);
504 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
506 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
510 GOTO(out_free, rc = -ENOMEM);
512 obd = class_name2obd(mgcname);
513 if (obd && !obd->obd_stopping) {
514 rc = obd_set_info_async(NULL, obd->obd_self_export,
515 strlen(KEY_MGSSEC), KEY_MGSSEC,
516 strlen(mgssec), mgssec, NULL);
520 /* Re-using an existing MGC */
521 cfs_atomic_inc(&obd->u.cli.cl_mgc_refcount);
523 /* IR compatibility check, only for clients */
524 if (lmd_is_client(lsi->lsi_lmd)) {
526 int vallen = sizeof(*data);
527 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
529 rc = obd_get_info(NULL, obd->obd_self_export,
530 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
531 &vallen, data, NULL);
533 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
534 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
535 /* LMD_FLG_NOIR is for test purpose only */
537 "Trying to mount a client with IR setting "
538 "not compatible with current mgc. "
539 "Force to use current mgc setting that is "
541 has_ir ? "enabled" : "disabled");
543 *flags &= ~LMD_FLG_NOIR;
545 *flags |= LMD_FLG_NOIR;
550 /* If we are restarting the MGS, don't try to keep the MGC's
551 old connection, or registration will fail. */
553 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
557 /* Try all connections, but only once (again).
558 We don't want to block another target from starting
559 (using its local copy of the log), but we do want to connect
560 if at all possible. */
562 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
563 rc = obd_set_info_async(NULL, obd->obd_self_export,
564 sizeof(KEY_INIT_RECOV_BACKUP),
565 KEY_INIT_RECOV_BACKUP,
566 sizeof(recov_bk), &recov_bk, NULL);
570 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
572 /* Add the primary nids for the MGS */
574 sprintf(niduuid, "%s_%x", mgcname, i);
575 if (IS_SERVER(lsi)) {
576 ptr = lsi->lsi_lmd->lmd_mgs;
578 /* Use local nids (including LO) */
579 lnet_process_id_t id;
580 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
581 rc = do_lcfg(mgcname, id.nid,
582 LCFG_ADD_UUID, niduuid, 0,0,0);
585 /* Use mgsnode= nids */
586 /* mount -o mgsnode=nid */
587 if (lsi->lsi_lmd->lmd_mgs) {
588 ptr = lsi->lsi_lmd->lmd_mgs;
589 } else if (class_find_param(ptr, PARAM_MGSNODE,
591 CERROR("No MGS nids given.\n");
592 GOTO(out_free, rc = -EINVAL);
594 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
595 rc = do_lcfg(mgcname, nid,
596 LCFG_ADD_UUID, niduuid, 0,0,0);
600 } else { /* client */
601 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
602 ptr = lsi->lsi_lmd->lmd_dev;
603 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
604 rc = do_lcfg(mgcname, nid,
605 LCFG_ADD_UUID, niduuid, 0,0,0);
607 /* Stop at the first failover nid */
613 CERROR("No valid MGS nids found.\n");
614 GOTO(out_free, rc = -EINVAL);
616 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
618 /* Random uuid for MGC allows easier reconnects */
620 ll_generate_random_uuid(uuidc);
621 class_uuid_unparse(uuidc, uuid);
624 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
625 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
631 /* Add any failover MGS nids */
633 while (ptr && ((*ptr == ':' ||
634 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) {
635 /* New failover node */
636 sprintf(niduuid, "%s_%x", mgcname, i);
638 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
640 rc = do_lcfg(mgcname, nid,
641 LCFG_ADD_UUID, niduuid, 0,0,0);
646 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
654 lsi->lsi_lmd->lmd_mgs_failnodes = i;
656 obd = class_name2obd(mgcname);
658 CERROR("Can't find mgcobd %s\n", mgcname);
659 GOTO(out_free, rc = -ENOTCONN);
662 rc = obd_set_info_async(NULL, obd->obd_self_export,
663 strlen(KEY_MGSSEC), KEY_MGSSEC,
664 strlen(mgssec), mgssec, NULL);
668 /* Keep a refcount of servers/clients who started with "mount",
669 so we know when we can get rid of the mgc. */
670 cfs_atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
672 /* Try all connections, but only once. */
674 rc = obd_set_info_async(NULL, obd->obd_self_export,
675 sizeof(KEY_INIT_RECOV_BACKUP),
676 KEY_INIT_RECOV_BACKUP,
677 sizeof(recov_bk), &recov_bk, NULL);
680 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
682 /* We connect to the MGS at setup, and don't disconnect until cleanup */
683 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
684 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV |
685 OBD_CONNECT_LVB_TYPE;
687 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
688 data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB;
690 #warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
693 if (lmd_is_client(lsi->lsi_lmd) &&
694 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
695 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
696 data->ocd_version = LUSTRE_VERSION_CODE;
697 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
699 CERROR("connect failed %d\n", rc);
703 obd->u.cli.cl_mgc_mgsexp = exp;
706 /* Keep the mgc info in the sb. Note that many lsi's can point
710 mutex_unlock(&mgc_start_lock);
715 OBD_FREE(mgcname, len);
717 OBD_FREE(niduuid, len + 2);
721 static int lustre_stop_mgc(struct super_block *sb)
723 struct lustre_sb_info *lsi = s2lsi(sb);
724 struct obd_device *obd;
725 char *niduuid = 0, *ptr = 0;
726 int i, rc = 0, len = 0;
736 mutex_lock(&mgc_start_lock);
737 LASSERT(cfs_atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
738 if (!cfs_atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
739 /* This is not fatal, every client that stops
740 will call in here. */
741 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
742 cfs_atomic_read(&obd->u.cli.cl_mgc_refcount));
743 GOTO(out, rc = -EBUSY);
746 /* The MGC has no recoverable data in any case.
747 * force shotdown set in umount_begin */
748 obd->obd_no_recov = 1;
750 if (obd->u.cli.cl_mgc_mgsexp) {
751 /* An error is not fatal, if we are unable to send the
752 disconnect mgs ping evictor cleans up the export */
753 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
755 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
758 /* Save the obdname for cleaning the nid uuids, which are
760 len = strlen(obd->obd_name) + 6;
761 OBD_ALLOC(niduuid, len);
763 strcpy(niduuid, obd->obd_name);
764 ptr = niduuid + strlen(niduuid);
767 rc = class_manual_cleanup(obd);
771 /* Clean the nid uuids */
773 GOTO(out, rc = -ENOMEM);
775 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
776 sprintf(ptr, "_%x", i);
777 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
780 CERROR("del MDC UUID %s failed: rc = %d\n",
785 OBD_FREE(niduuid, len);
787 /* class_import_put will get rid of the additional connections */
788 mutex_unlock(&mgc_start_lock);
792 /* Since there's only one mgc per node, we have to change it's fs to get
793 access to the right disk. */
794 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
796 struct lustre_sb_info *lsi = s2lsi(sb);
800 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
802 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
803 rc = obd_set_info_async(NULL, mgc->obd_self_export,
804 sizeof(KEY_SET_FS), KEY_SET_FS,
805 sizeof(*sb), sb, NULL);
807 CERROR("can't set_fs %d\n", rc);
813 static int server_mgc_clear_fs(struct obd_device *mgc)
818 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
820 rc = obd_set_info_async(NULL, mgc->obd_self_export,
821 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
826 /** Get the fsname ("lustre") from the server name ("lustre-OST003F").
827 * @param [in] svname server name including type and index
828 * @param [out] fsname Buffer to copy filesystem name prefix into.
829 * Must have at least 'strlen(fsname) + 1' chars.
830 * @param [out] endptr if endptr isn't NULL it is set to end of fsname
833 int server_name2fsname(char *svname, char *fsname, char **endptr)
835 char *dash = strrchr(svname, '-');
837 dash = strrchr(svname, ':');
842 /* interpret <fsname>-MDTXXXXX-mdc as mdt, the better way is to pass
843 * in the fsname, then determine the server index */
844 if (!strcmp(LUSTRE_MDC_NAME, dash + 1)) {
846 for (; dash > svname && *dash != '-' && *dash != ':'; dash--)
852 if (fsname != NULL) {
853 strncpy(fsname, svname, dash - svname);
854 fsname[dash - svname] = '\0';
862 EXPORT_SYMBOL(server_name2fsname);
864 static int is_mdc_device(char *devname)
867 ptr = strrchr(devname, '-');
868 if (ptr != NULL && strcmp(ptr, "-mdc") == 0)
873 static int inline tgt_is_mdt0(char *tgtname)
878 type = server_name2index(tgtname, &idx, NULL);
879 if (type != LDD_F_SV_TYPE_MDT)
882 return (idx == 0) ? 1 :0;
885 static int inline is_mdc_for_mdt0(char *devname)
889 if (!is_mdc_device(devname))
892 ptr = strrchr(devname, '-');
897 if (tgt_is_mdt0(devname)) {
906 * Convert OST/MDT name(fsname-OSTxxxx) to an osp name
907 * (fsname-MDT0000-osp-OSTxxxx), which will be used to
908 * communicate with MDT0 for this target.
910 int tgt_name2ospname(char *svname, char *ospname)
917 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
921 rc = server_name2fsname(svname, fsname, &tgt);
923 CERROR("%s change fsname error: rc %d\n", svname, rc);
927 if (*tgt != '-' && *tgt != ':') {
928 CERROR("%s wrong svname name!\n", svname);
929 GOTO(cleanup, rc = -EINVAL);
933 if (strncmp(tgt, "OST", 3) != 0 && strncmp(tgt, "MDT", 3) != 0) {
934 CERROR("%s is not an OST or MDT target!\n", svname);
935 GOTO(cleanup, rc = -EINVAL);
937 sprintf(ospname, "%s-MDT0000-%s-%s", fsname, LUSTRE_OSP_NAME, tgt);
940 OBD_FREE(fsname, MTI_NAME_MAXLEN);
943 EXPORT_SYMBOL(tgt_name2ospname);
945 static CFS_LIST_HEAD(osp_register_list);
946 DEFINE_MUTEX(osp_register_list_lock);
948 int lustre_register_osp_item(char *ospname, struct obd_export **exp,
949 register_osp_cb cb_func, void *cb_data)
951 struct obd_device *osp;
952 struct osp_register_item *ori;
955 LASSERTF(strlen(ospname) < MTI_NAME_MAXLEN, "ospname is too long %s\n",
957 LASSERT(exp != NULL && *exp == NULL);
963 mutex_lock(&osp_register_list_lock);
965 osp = class_name2obd(ospname);
966 if (osp != NULL && osp->obd_set_up == 1) {
967 struct obd_uuid *uuid;
971 mutex_unlock(&osp_register_list_lock);
975 memcpy(uuid->uuid, ospname, strlen(ospname));
976 *exp = cfs_hash_lookup(osp->obd_uuid_hash, uuid);
980 memcpy(ori->ori_name, ospname, strlen(ospname));
982 ori->ori_cb_func = cb_func;
983 ori->ori_cb_data = cb_data;
984 CFS_INIT_LIST_HEAD(&ori->ori_list);
985 cfs_list_add(&ori->ori_list, &osp_register_list);
987 if (*exp != NULL && cb_func != NULL)
990 mutex_unlock(&osp_register_list_lock);
993 EXPORT_SYMBOL(lustre_register_osp_item);
995 void lustre_deregister_osp_item(struct obd_export **exp)
997 struct osp_register_item *ori, *tmp;
999 mutex_lock(&osp_register_list_lock);
1000 cfs_list_for_each_entry_safe(ori, tmp, &osp_register_list, ori_list) {
1001 if (exp == ori->ori_exp) {
1003 class_export_put(*exp);
1004 cfs_list_del(&ori->ori_list);
1009 mutex_unlock(&osp_register_list_lock);
1011 EXPORT_SYMBOL(lustre_deregister_osp_item);
1013 static void lustre_notify_osp_list(struct obd_export *exp)
1015 struct osp_register_item *ori, *tmp;
1016 LASSERT(exp != NULL);
1018 mutex_lock(&osp_register_list_lock);
1019 cfs_list_for_each_entry_safe(ori, tmp, &osp_register_list, ori_list) {
1020 if (strcmp(exp->exp_obd->obd_name, ori->ori_name))
1022 if (*ori->ori_exp != NULL)
1024 *ori->ori_exp = class_export_get(exp);
1025 if (ori->ori_cb_func != NULL)
1026 ori->ori_cb_func(ori->ori_cb_data);
1028 mutex_unlock(&osp_register_list_lock);
1031 static int lustre_osp_connect(struct obd_device *osp)
1034 struct lu_context session_ctx;
1035 struct obd_export *exp;
1036 struct obd_uuid *uuid = NULL;
1037 struct obd_connect_data *data = NULL;
1041 /* log has been fully processed, let clients connect */
1042 rc = lu_env_init(&env, osp->obd_lu_dev->ld_type->ldt_ctx_tags);
1046 lu_context_init(&session_ctx, LCT_SESSION);
1047 session_ctx.lc_thread = NULL;
1048 lu_context_enter(&session_ctx);
1049 env.le_ses = &session_ctx;
1051 OBD_ALLOC_PTR(data);
1053 GOTO(out, rc = -ENOMEM);
1055 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
1056 data->ocd_version = LUSTRE_VERSION_CODE;
1057 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
1058 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_IBITS |
1059 OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
1060 OBD_CONNECT_AT | OBD_CONNECT_FULL20 |
1061 OBD_CONNECT_LVB_TYPE |
1062 OBD_CONNECT_LIGHTWEIGHT;
1063 OBD_ALLOC_PTR(uuid);
1065 GOTO(out, rc = -ENOMEM);
1067 if (strlen(osp->obd_name) > sizeof(uuid->uuid)) {
1068 CERROR("%s: Too long osp name %s, max_size is %d\n",
1069 osp->obd_name, osp->obd_name, (int)sizeof(uuid->uuid));
1070 GOTO(out, rc = -EINVAL);
1072 /* Use osp name as the uuid, so we find the export by
1074 memcpy(uuid->uuid, osp->obd_name, strlen(osp->obd_name));
1075 rc = obd_connect(&env, &exp, osp, uuid, data, NULL);
1077 CERROR("%s: connect failed: rc = %d\n", osp->obd_name, rc);
1079 lustre_notify_osp_list(exp);
1088 lu_context_exit(&session_ctx);
1089 lu_context_fini(&session_ctx);
1095 * osp-on-ost is used by slaves (Non-MDT0 targets) to manage the connection
1098 * The OSTs will communicate with MDT0 by the connection established by the
1099 * osp-on-ost to get quota and fid sequence.
1102 static int lustre_osp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi)
1104 struct obd_connect_data *data = NULL;
1105 struct obd_device *obd;
1106 char *ospname = NULL;
1107 char *ospuuid = NULL;
1111 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
1114 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
1118 OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
1119 if (ospname == NULL)
1120 GOTO(out, rc = -ENOMEM);
1122 rc = tgt_name2ospname(lsi->lsi_svname, ospname);
1124 CERROR("%s change ospname error: rc %d\n",
1125 lsi->lsi_svname, rc);
1129 OBD_ALLOC(ospuuid, MTI_NAME_MAXLEN);
1130 if (ospuuid == NULL)
1131 GOTO(out, rc = -ENOMEM);
1133 sprintf(ospuuid, "%s_UUID", ospname);
1134 rc = lustre_start_simple(ospname, LUSTRE_OSP_NAME,
1135 ospuuid, lustre_cfg_string(lcfg, 1),
1138 CERROR("%s: setup up failed: rc %d\n", ospname, rc);
1142 obd = class_name2obd(ospname);
1143 LASSERT(obd != NULL);
1145 rc = lustre_osp_connect(obd);
1147 CERROR("%s: connect failed: rc = %d\n", ospname, rc);
1151 if (ospname != NULL)
1152 OBD_FREE(ospname, MTI_NAME_MAXLEN);
1153 if (ospuuid != NULL)
1154 OBD_FREE(ospuuid, MTI_NAME_MAXLEN);
1159 static int lustre_osp_add_conn(struct lustre_cfg *cfg,
1160 struct lustre_sb_info *lsi)
1162 struct lustre_cfg_bufs *bufs = NULL;
1163 struct lustre_cfg *lcfg = NULL;
1164 char *ospname = NULL;
1165 struct obd_device *osp;
1169 OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
1170 if (ospname == NULL)
1171 GOTO(out, rc = -ENOMEM);
1173 rc = tgt_name2ospname(lsi->lsi_svname, ospname);
1175 CERROR("%s change ospname error: rc %d\n",
1176 lsi->lsi_svname, rc);
1180 OBD_ALLOC_PTR(bufs);
1182 GOTO(out, rc = -ENOMEM);
1184 lustre_cfg_bufs_reset(bufs, ospname);
1185 lustre_cfg_bufs_set_string(bufs, 1,
1186 lustre_cfg_string(cfg, 1));
1188 lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
1190 osp = class_name2obd(ospname);
1192 CERROR("Can not find %s\n", ospname);
1193 GOTO(out, rc = -EINVAL);
1196 rc = class_add_conn(osp, lcfg);
1198 CERROR("%s: can't add conn: rc = %d\n", ospname, rc);
1204 lustre_cfg_free(lcfg);
1205 if (ospname != NULL)
1206 OBD_FREE(ospname, MTI_NAME_MAXLEN);
1212 * Retrieve MDT nids from the client log, then start the osp-on-ost device.
1213 * there are only two scenarios which would include mdt nid.
1215 * marker 5 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add mdc' xxx-
1216 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
1217 * attach 0:lustre-MDT0000-mdc 1:mdc 2:lustre-clilmv_UUID
1218 * setup 0:lustre-MDT0000-mdc 1:lustre-MDT0000_UUID 2:192.168.122.162@tcp
1219 * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
1220 * add_conn 0:lustre-MDT0000-mdc 1:192.168.172.1@tcp
1221 * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID xxxx
1222 * marker 5 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add mdc' xxxx-
1224 * marker 7 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
1225 * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
1226 * add_conn 0:lustre-MDT0000-mdc 1:192.168.122.2@tcp
1227 * marker 7 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
1229 static int client_osp_config_process(const struct lu_env *env,
1230 struct llog_handle *handle,
1231 struct llog_rec_hdr *rec, void *data)
1233 struct config_llog_instance *clli = data;
1234 int cfg_len = rec->lrh_len;
1235 char *cfg_buf = (char *) (rec + 1);
1236 struct lustre_cfg *lcfg = NULL;
1237 struct lustre_sb_info *lsi;
1238 int rc = 0, swab = 0;
1241 if (rec->lrh_type != OBD_CFG_REC) {
1242 CERROR("Unknown llog record type %#x encountered\n",
1247 LASSERT(clli->cfg_sb != NULL);
1248 lsi = s2lsi(clli->cfg_sb);
1250 lcfg = (struct lustre_cfg *)cfg_buf;
1251 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
1252 lustre_swab_lustre_cfg(lcfg);
1256 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
1260 switch (lcfg->lcfg_command) {
1262 struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
1264 lustre_swab_cfg_marker(marker, swab,
1265 LUSTRE_CFG_BUFLEN(lcfg, 1));
1266 if (marker->cm_flags & CM_SKIP ||
1267 marker->cm_flags & CM_EXCLUDE)
1270 if (!tgt_is_mdt0(marker->cm_tgtname))
1273 if(!strncmp(marker->cm_comment, "add mdc", 7) ||
1274 !strncmp(marker->cm_comment, "add failnid", 11)) {
1275 if (marker->cm_flags & CM_START) {
1276 clli->cfg_flags = CFG_F_MARKER;
1277 /* This hack is to differentiate the
1278 * ADD_UUID is come from "add mdc" record
1279 * or from "add failnid" record. */
1280 if (!strncmp(marker->cm_comment,
1282 clli->cfg_flags |= CFG_F_SKIP;
1283 } else if (marker->cm_flags & CM_END) {
1284 clli->cfg_flags = 0;
1289 case LCFG_ADD_UUID: {
1290 if (clli->cfg_flags == CFG_F_MARKER) {
1291 rc = lustre_osp_setup(lcfg, lsi);
1292 /* XXX: process only the first nid as
1293 * we don't need another instance of osp */
1294 clli->cfg_flags |= CFG_F_SKIP;
1295 } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
1296 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
1299 CERROR("%s: Fail to add uuid, rc:%d\n",
1300 lsi->lsi_svname, rc);
1304 case LCFG_ADD_CONN: {
1305 if (is_mdc_for_mdt0(lustre_cfg_string(lcfg, 0)))
1306 rc = lustre_osp_add_conn(lcfg, lsi);
1316 static int lustre_disconnect_osp(struct super_block *sb)
1318 struct lustre_sb_info *lsi = s2lsi(sb);
1319 struct obd_device *osp;
1320 char *ospname = NULL;
1321 char *logname = NULL;
1322 struct lustre_cfg *lcfg = NULL;
1323 struct lustre_cfg_bufs *bufs = NULL;
1324 struct config_llog_instance *cfg = NULL;
1328 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
1332 /* Only disconnect MDT0-osp-MDT0 here, other osp on MDT
1333 * will be disconnect during MDT stack cleanup.
1334 * FIXME: remove later when quota on DNE is finished */
1335 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1341 OBD_ALLOC(logname, MTI_NAME_MAXLEN);
1342 if (logname == NULL)
1345 OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
1346 if (ospname == NULL)
1347 GOTO(out, rc = -ENOMEM);
1349 rc = server_name2fsname(lsi->lsi_svname, ospname, NULL);
1351 CERROR("%s: get fsname error: %d\n",
1352 lsi->lsi_svname, rc);
1355 sprintf(logname, "%s-client", ospname);
1359 GOTO(out, rc = -ENOMEM);
1362 cfg->cfg_instance = sb;
1363 rc = lustre_end_log(sb, logname, cfg);
1365 CERROR("Can't end config log %s\n", ospname);
1369 rc = tgt_name2ospname(lsi->lsi_svname, ospname);
1371 CERROR("%s: get osp name error: %d\n",
1372 lsi->lsi_svname, rc);
1376 osp = class_name2obd(ospname);
1378 CERROR("Can't find osp-on-ost %s\n", ospname);
1379 GOTO(out, rc = -ENOENT);
1382 OBD_ALLOC_PTR(bufs);
1384 GOTO(out, rc = -ENOMEM);
1386 lustre_cfg_bufs_reset(bufs, osp->obd_name);
1387 lustre_cfg_bufs_set_string(bufs, 1, NULL);
1388 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
1390 GOTO(out, rc = -ENOMEM);
1392 /* Disconnect import first. NULL is passed for the '@env', since
1393 * it will not be used for the 'osp-on-ost'. (see osp_shutdown()) */
1394 rc = osp->obd_lu_dev->ld_ops->ldo_process_config(NULL, osp->obd_lu_dev,
1398 lustre_cfg_free(lcfg);
1404 OBD_FREE(ospname, MTI_NAME_MAXLEN);
1406 OBD_FREE(logname, MTI_NAME_MAXLEN);
1411 * Stop the osp(fsname-MDT0000-osp-OSTxxxx) or (fsname-MDT0000-osp-MDT0000) for an OST target.
1413 static int lustre_stop_osp(struct super_block *sb)
1415 struct lustre_sb_info *lsi = s2lsi(sb);
1416 struct obd_device *osp = NULL;
1417 char *ospname = NULL;
1421 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
1425 /* Only disconnect MDT0-osp-MDT0 here, other osp on MDT
1426 * will be disconnect during MDT stack cleanup.
1427 * FIXME: remove later when quota on DNE is finished */
1428 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1435 OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
1436 rc = tgt_name2ospname(lsi->lsi_svname, ospname);
1438 CERROR("%s get fsname error: rc %d\n",
1439 lsi->lsi_svname, rc);
1443 osp = class_name2obd(ospname);
1445 CERROR("Can not find osp-on-ost %s\n", ospname);
1446 GOTO(cleanup, rc = -ENOENT);
1450 rc = class_manual_cleanup(osp);
1453 if (ospname != NULL)
1454 OBD_FREE(ospname, MTI_NAME_MAXLEN);
1458 DEFINE_MUTEX(server_start_lock);
1460 /* Stop MDS/OSS if nobody is using them */
1461 static int server_stop_servers(int lsiflags)
1463 struct obd_device *obd = NULL;
1464 struct obd_type *type = NULL;
1468 mutex_lock(&server_start_lock);
1470 /* Either an MDT or an OST or neither */
1471 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
1472 if ((lsiflags & LDD_F_SV_TYPE_MDT) &&
1473 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
1474 type = class_search_type(LUSTRE_MDT_NAME);
1476 /* if this was an OST, and there are no more OST's, clean up the OSS */
1477 if ((lsiflags & LDD_F_SV_TYPE_OST) &&
1478 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
1479 type = class_search_type(LUSTRE_OST_NAME);
1482 if (obd && (!type || !type->typ_refcnt)) {
1485 /* obd_fail doesn't mean much on a server obd */
1486 err = class_manual_cleanup(obd);
1491 mutex_unlock(&server_start_lock);
1496 int server_mti_print(char *title, struct mgs_target_info *mti)
1498 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
1499 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
1500 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
1501 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
1502 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
1503 mti->mti_config_ver, mti->mti_flags);
1508 * Get service name (svname) from string
1510 * if endptr isn't NULL it is set to end of fsname *
1512 int server_name2svname(char *label, char *svname, char **endptr)
1517 /* We use server_name2fsname() just for parsing */
1518 rc = server_name2fsname(label, NULL, &dash);
1525 strncpy(svname, dash + 1, MTI_NAME_MAXLEN);
1529 EXPORT_SYMBOL(server_name2svname);
1532 /* Get the index from the obd name.
1533 rc = server type, or
1535 if endptr isn't NULL it is set to end of name */
1536 int server_name2index(char *svname, __u32 *idx, char **endptr)
1538 unsigned long index;
1542 /* We use server_name2fsname() just for parsing */
1543 rc = server_name2fsname(svname, NULL, &dash);
1552 if (strncmp(dash, "MDT", 3) == 0)
1553 rc = LDD_F_SV_TYPE_MDT;
1554 else if (strncmp(dash, "OST", 3) == 0)
1555 rc = LDD_F_SV_TYPE_OST;
1561 if (strcmp(dash, "all") == 0)
1562 return rc | LDD_F_SV_ALL;
1564 index = simple_strtoul(dash, endptr, 16);
1569 EXPORT_SYMBOL(server_name2index);
1571 /* Generate data for registration */
1572 static int server_lsi2mti(struct lustre_sb_info *lsi,
1573 struct mgs_target_info *mti)
1575 lnet_process_id_t id;
1579 if (!IS_SERVER(lsi))
1582 strncpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname));
1584 mti->mti_nid_count = 0;
1585 while (LNetGetId(i++, &id) != -ENOENT) {
1586 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1589 /* server use --servicenode param, only allow specified
1590 * nids be registered */
1591 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
1592 class_match_nid(lsi->lsi_lmd->lmd_params,
1593 PARAM_FAILNODE, id.nid) < 1)
1596 /* match specified network */
1597 if (!class_match_net(lsi->lsi_lmd->lmd_params,
1598 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1601 mti->mti_nids[mti->mti_nid_count] = id.nid;
1602 mti->mti_nid_count++;
1603 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1604 CWARN("Only using first %d nids for %s\n",
1605 mti->mti_nid_count, mti->mti_svname);
1610 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1611 mti->mti_config_ver = 0;
1613 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1617 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1620 /* Orion requires index to be set */
1621 LASSERT(!(rc & LDD_F_NEED_INDEX));
1622 /* keep only LDD flags */
1623 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1624 if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN))
1625 mti->mti_flags |= LDD_F_UPDATE;
1626 strncpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1627 sizeof(mti->mti_params));
1631 /* Register an old or new target with the MGS. If needed MGS will construct
1632 startup logs and assign index */
1633 static int server_register_target(struct lustre_sb_info *lsi)
1635 struct obd_device *mgc = lsi->lsi_mgc;
1636 struct mgs_target_info *mti = NULL;
1643 if (!IS_SERVER(lsi))
1650 rc = server_lsi2mti(lsi, mti);
1654 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1655 mti->mti_svname, mti->mti_fsname,
1656 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1659 /* if write_conf is true, the registration must succeed */
1660 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1661 mti->mti_flags |= LDD_F_OPC_REG;
1663 /* Register the target */
1664 /* FIXME use mgc_process_config instead */
1665 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1666 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
1667 sizeof(*mti), mti, NULL);
1669 if (mti->mti_flags & LDD_F_ERROR) {
1670 LCONSOLE_ERROR_MSG(0x160,
1671 "The MGS is refusing to allow this "
1672 "server (%s) to start. Please see messages"
1673 " on the MGS node.\n", lsi->lsi_svname);
1674 } else if (writeconf) {
1675 LCONSOLE_ERROR_MSG(0x15f,
1676 "Communication to the MGS return error %d. "
1677 "Is the MGS running?\n", rc);
1679 CERROR("Cannot talk to the MGS: %d, not fatal\n", rc);
1680 /* reset the error code for non-fatal error. */
1693 * Notify the MGS that this target is ready.
1694 * Used by IR - if the MGS receives this message, it will notify clients.
1696 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1698 struct lustre_sb_info *lsi = s2lsi(sb);
1699 struct obd_device *mgc = lsi->lsi_mgc;
1700 struct mgs_target_info *mti = NULL;
1706 if (!(IS_SERVER(lsi)))
1712 rc = server_lsi2mti(lsi, mti);
1716 mti->mti_instance = obd->u.obt.obt_instance;
1717 mti->mti_flags |= LDD_F_OPC_READY;
1719 /* FIXME use mgc_process_config instead */
1720 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1721 sizeof(KEY_REGISTER_TARGET),
1722 KEY_REGISTER_TARGET,
1723 sizeof(*mti), mti, NULL);
1725 /* Imperative recovery: if the mgs informs us to use IR? */
1726 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1727 (mti->mti_flags & LDD_F_IR_CAPABLE))
1728 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1738 * Start the osp(fsname-MDT0000-osp-OSTxxxx) for an OST target,
1739 * which would be used to communicate with MDT0 for quota and FID.
1741 static int lustre_start_osp(struct super_block *sb)
1743 struct lustre_sb_info *lsi = s2lsi(sb);
1744 struct config_llog_instance *cfg = NULL;
1745 struct obd_device *osp;
1746 char *ospname = NULL;
1747 char *logname = NULL;
1752 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
1753 OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
1754 OBD_ALLOC(logname, MTI_NAME_MAXLEN);
1755 if (ospname == NULL || logname == NULL)
1756 GOTO(cleanup, rc = -ENOMEM);
1758 rc = server_name2fsname(lsi->lsi_svname, ospname, &tgt);
1760 CERROR("%s change fsname error: rc %d\n",
1761 lsi->lsi_svname, rc);
1764 sprintf(logname, "%s-client", ospname);
1766 rc = tgt_name2ospname(lsi->lsi_svname, ospname);
1768 CERROR("%s change ospname error: rc %d\n",
1769 lsi->lsi_svname, rc);
1773 osp = class_name2obd(ospname);
1775 GOTO(cleanup, rc = 0);
1779 GOTO(cleanup, rc = -ENOMEM);
1781 cfg->cfg_callback = client_osp_config_process;
1782 cfg->cfg_instance = sb;
1784 rc = lustre_process_log(sb, logname, cfg);
1787 if (ospname != NULL)
1788 OBD_FREE(ospname, MTI_NAME_MAXLEN);
1789 if (logname != NULL)
1790 OBD_FREE(logname, MTI_NAME_MAXLEN);
1797 /** Start server targets: MDTs and OSTs
1799 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1801 struct obd_device *obd;
1802 struct lustre_sb_info *lsi = s2lsi(sb);
1803 struct config_llog_instance cfg;
1805 struct lu_device *dev;
1809 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1812 /* make sure the MDS is started */
1813 mutex_lock(&server_start_lock);
1814 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1816 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1818 LUSTRE_MDS_OBDNAME"_uuid",
1821 mutex_unlock(&server_start_lock);
1822 CERROR("failed to start MDS: %d\n", rc);
1826 mutex_unlock(&server_start_lock);
1829 /* If we're an OST, make sure the global OSS is running */
1831 /* make sure OSS is started */
1832 mutex_lock(&server_start_lock);
1833 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1835 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1837 LUSTRE_OSS_OBDNAME"_uuid",
1840 mutex_unlock(&server_start_lock);
1841 CERROR("failed to start OSS: %d\n", rc);
1845 mutex_unlock(&server_start_lock);
1848 /* Set the mgc fs to our server disk. This allows the MGC to
1849 * read and write configs locally, in case it can't talk to the MGS. */
1850 if (lsi->lsi_srv_mnt) {
1851 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1853 GOTO(out_stop_service, rc);
1856 /* Register with MGS */
1857 rc = server_register_target(lsi);
1861 /* Let the target look up the mount using the target's name
1862 (we can't pass the sb or mnt through class_process_config.) */
1863 rc = server_register_mount(lsi->lsi_svname, sb, mnt);
1867 /* Start targets using the llog named for the target */
1868 memset(&cfg, 0, sizeof(cfg));
1869 cfg.cfg_callback = class_config_llog_handler;
1870 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1872 CERROR("failed to start server %s: %d\n",
1873 lsi->lsi_svname, rc);
1874 /* Do NOT call server_deregister_mount() here. This makes it
1875 * impossible to find mount later in cleanup time and leaves
1876 * @lsi and othder stuff leaked. -umka */
1880 obd = class_name2obd(lsi->lsi_svname);
1882 CERROR("no server named %s was started\n", lsi->lsi_svname);
1883 GOTO(out_mgc, rc = -ENXIO);
1886 if (IS_OST(lsi) || IS_MDT(lsi)) {
1887 rc = lustre_start_osp(sb);
1889 CERROR("%s: failed to start OSP: %d\n",
1890 lsi->lsi_svname, rc);
1895 server_notify_target(sb, obd);
1897 /* calculate recovery timeout, do it after lustre_process_log */
1898 server_calc_timeout(lsi, obd);
1900 /* log has been fully processed */
1901 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1903 /* log has been fully processed, let clients connect */
1904 dev = obd->obd_lu_dev;
1905 if (dev && dev->ld_ops->ldo_prepare) {
1906 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1908 struct lu_context session_ctx;
1910 lu_context_init(&session_ctx, LCT_SESSION);
1911 session_ctx.lc_thread = NULL;
1912 lu_context_enter(&session_ctx);
1913 env.le_ses = &session_ctx;
1915 dev->ld_ops->ldo_prepare(&env, NULL, dev);
1918 lu_context_exit(&session_ctx);
1919 lu_context_fini(&session_ctx);
1923 /* abort recovery only on the complete stack:
1924 * many devices can be involved */
1925 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1926 (OBP(obd, iocontrol))) {
1927 obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
1932 /* Release the mgc fs for others to use */
1933 if (lsi->lsi_srv_mnt)
1934 server_mgc_clear_fs(lsi->lsi_mgc);
1938 server_stop_servers(lsi->lsi_flags);
1943 /***************** lustre superblock **************/
1945 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1947 struct lustre_sb_info *lsi;
1953 OBD_ALLOC_PTR(lsi->lsi_lmd);
1954 if (!lsi->lsi_lmd) {
1959 lsi->lsi_lmd->lmd_exclude_count = 0;
1960 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
1961 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
1962 s2lsi_nocast(sb) = lsi;
1963 /* we take 1 extra ref for our setup */
1964 cfs_atomic_set(&lsi->lsi_mounts, 1);
1966 /* Default umount style */
1967 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1972 static int lustre_free_lsi(struct super_block *sb)
1974 struct lustre_sb_info *lsi = s2lsi(sb);
1977 LASSERT(lsi != NULL);
1978 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
1980 /* someone didn't call server_put_mount. */
1981 LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0);
1983 if (lsi->lsi_lmd != NULL) {
1984 if (lsi->lsi_lmd->lmd_dev != NULL)
1985 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1986 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1987 if (lsi->lsi_lmd->lmd_profile != NULL)
1988 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1989 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1990 if (lsi->lsi_lmd->lmd_mgssec != NULL)
1991 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
1992 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
1993 if (lsi->lsi_lmd->lmd_opts != NULL)
1994 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1995 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1996 if (lsi->lsi_lmd->lmd_exclude_count)
1997 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1998 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1999 lsi->lsi_lmd->lmd_exclude_count);
2000 if (lsi->lsi_lmd->lmd_mgs != NULL)
2001 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
2002 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
2003 if (lsi->lsi_lmd->lmd_osd_type != NULL)
2004 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
2005 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
2006 if (lsi->lsi_lmd->lmd_params != NULL)
2007 OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
2009 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
2012 LASSERT(lsi->lsi_llsbi == NULL);
2013 OBD_FREE(lsi, sizeof(*lsi));
2014 s2lsi_nocast(sb) = NULL;
2019 /* The lsi has one reference for every server that is using the disk -
2020 e.g. MDT, MGS, and potentially MGC */
2021 static int lustre_put_lsi(struct super_block *sb)
2023 struct lustre_sb_info *lsi = s2lsi(sb);
2026 LASSERT(lsi != NULL);
2028 CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts));
2029 if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) {
2030 if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
2031 obd_disconnect(lsi->lsi_osd_exp);
2032 /* wait till OSD is gone */
2033 obd_zombie_barrier();
2035 lustre_free_lsi(sb);
2041 static int lsi_prepare(struct lustre_sb_info *lsi)
2048 LASSERT(lsi->lsi_lmd);
2050 /* The server name is given as a mount line option */
2051 if (lsi->lsi_lmd->lmd_profile == NULL) {
2052 LCONSOLE_ERROR("Can't determine server name\n");
2056 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
2057 RETURN(-ENAMETOOLONG);
2059 strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
2061 /* Determine osd type */
2062 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
2063 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
2064 sizeof(lsi->lsi_osd_type))
2065 RETURN(-ENAMETOOLONG);
2067 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
2069 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
2072 /* XXX: a temp. solution for components using fsfilt
2073 * to be removed in one of the subsequent patches */
2074 if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs")) {
2075 strcpy(lsi->lsi_fstype, "ldiskfs");
2077 strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
2080 /* Determine server type */
2081 rc = server_name2index(lsi->lsi_svname, &index, NULL);
2083 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
2084 /* Assume we're a bare MGS */
2086 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
2088 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
2093 lsi->lsi_flags |= rc;
2095 /* Add mount line flags that used to be in ldd:
2096 * writeconf, mgs, anything else?
2098 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
2099 LDD_F_WRITECONF : 0;
2100 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
2102 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
2103 LDD_F_SV_TYPE_MGS : 0;
2104 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
2105 LDD_F_NO_PRIMNODE : 0;
2110 /*************** server mount ******************/
2112 /** Start the shutdown of servers at umount.
2114 static void server_put_super(struct super_block *sb)
2116 struct lustre_sb_info *lsi = s2lsi(sb);
2117 struct obd_device *obd;
2118 char *tmpname, *extraname = NULL;
2120 int lsiflags = lsi->lsi_flags;
2123 LASSERT(IS_SERVER(lsi));
2125 tmpname_sz = strlen(lsi->lsi_svname) + 1;
2126 OBD_ALLOC(tmpname, tmpname_sz);
2127 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
2128 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
2129 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
2130 snprintf(tmpname, tmpname_sz, "MGS");
2132 /* disconnect the osp-on-ost first to drain off the inflight request */
2133 if (IS_OST(lsi) || IS_MDT(lsi)) {
2136 rc = lustre_disconnect_osp(sb);
2137 if (rc && rc != ETIMEDOUT)
2138 CERROR("%s: failed to disconnect osp-on-ost (rc=%d)!\n",
2142 /* Stop the target */
2143 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
2144 (IS_MDT(lsi) || IS_OST(lsi))) {
2145 struct lustre_profile *lprof = NULL;
2147 /* tell the mgc to drop the config log */
2148 lustre_end_log(sb, lsi->lsi_svname, NULL);
2150 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
2151 If there are any setup/cleanup errors, save the lov
2152 name for safety cleanup later. */
2153 lprof = class_get_profile(lsi->lsi_svname);
2154 if (lprof && lprof->lp_dt) {
2155 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
2156 strcpy(extraname, lprof->lp_dt);
2159 obd = class_name2obd(lsi->lsi_svname);
2161 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
2162 if (lsiflags & LSI_UMOUNT_FAILOVER)
2164 /* We can't seem to give an error return code
2165 * to .put_super, so we better make sure we clean up! */
2167 class_manual_cleanup(obd);
2169 CERROR("no obd %s\n", lsi->lsi_svname);
2170 server_deregister_mount(lsi->lsi_svname);
2174 /* If they wanted the mgs to stop separately from the mdt, they
2175 should have put it on a different device. */
2177 /* if MDS start with --nomgs, don't stop MGS then */
2178 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
2179 server_stop_mgs(sb);
2182 if (IS_OST(lsi) || IS_MDT(lsi)) {
2183 if (lustre_stop_osp(sb) < 0)
2184 CERROR("%s: Fail to stop osp-on-ost!\n", tmpname);
2187 /* Clean the mgc and sb */
2188 lustre_common_put_super(sb);
2190 /* wait till all in-progress cleanups are done
2191 * specifically we're interested in ofd cleanup
2193 obd_zombie_barrier();
2195 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
2196 until the target is really gone so that our type refcount check
2198 server_stop_servers(lsiflags);
2200 /* In case of startup or cleanup err, stop related obds */
2202 obd = class_name2obd(extraname);
2204 CWARN("Cleaning orphaned obd %s\n", extraname);
2206 class_manual_cleanup(obd);
2208 OBD_FREE(extraname, strlen(extraname) + 1);
2211 LCONSOLE_WARN("server umount %s complete\n", tmpname);
2212 OBD_FREE(tmpname, tmpname_sz);
2216 /** Called only for 'umount -f'
2218 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2219 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
2221 struct super_block *sb = vfsmnt->mnt_sb;
2223 static void server_umount_begin(struct super_block *sb)
2226 struct lustre_sb_info *lsi = s2lsi(sb);
2229 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2230 if (!(flags & MNT_FORCE)) {
2236 CDEBUG(D_MOUNT, "umount -f\n");
2237 /* umount = failover
2239 no third way to do non-force, non-failover */
2240 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
2244 static int server_statfs (struct dentry *dentry, cfs_kstatfs_t *buf)
2246 struct super_block *sb = dentry->d_sb;
2247 struct lustre_sb_info *lsi = s2lsi(sb);
2248 struct obd_statfs statfs;
2252 if (lsi->lsi_dt_dev) {
2253 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
2255 statfs_unpack(buf, &statfs);
2256 buf->f_type = sb->s_magic;
2262 buf->f_type = sb->s_magic;
2263 buf->f_bsize = sb->s_blocksize;
2269 buf->f_namelen = NAME_MAX;
2273 /** The operations we support directly on the superblock:
2274 * mount, umount, and df.
2276 static struct super_operations server_ops =
2278 .put_super = server_put_super,
2279 .umount_begin = server_umount_begin, /* umount -f */
2280 .statfs = server_statfs,
2283 #define log2(n) ffz(~(n))
2284 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
2286 static int server_fill_super_common(struct super_block *sb)
2288 struct inode *root = 0;
2291 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
2293 sb->s_blocksize = 4096;
2294 sb->s_blocksize_bits = log2(sb->s_blocksize);
2295 sb->s_magic = LUSTRE_SUPER_MAGIC;
2296 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
2297 sb->s_flags |= MS_RDONLY;
2298 sb->s_op = &server_ops;
2300 root = new_inode(sb);
2302 CERROR("Can't make root inode\n");
2306 /* returns -EIO for every operation */
2307 /* make_bad_inode(root); -- badness - can't umount */
2308 /* apparently we need to be a directory for the mount to finish */
2309 root->i_mode = S_IFDIR;
2311 sb->s_root = d_make_root(root);
2313 CERROR("%s: can't make root dentry\n", sb->s_id);
2320 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
2322 struct lustre_mount_data *lmd = lsi->lsi_lmd;
2323 struct obd_device *obd;
2324 struct dt_device_param p;
2330 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
2331 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
2333 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
2334 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
2335 strcat(lsi->lsi_osd_uuid, "_UUID");
2336 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
2338 obd = class_name2obd(lsi->lsi_osd_obdname);
2340 rc = lustre_start_simple(lsi->lsi_osd_obdname,
2342 lsi->lsi_osd_uuid, lmd->lmd_dev,
2343 flagstr, lsi->lsi_lmd->lmd_opts,
2347 obd = class_name2obd(lsi->lsi_osd_obdname);
2351 rc = obd_connect(NULL, &lsi->lsi_osd_exp, obd, &obd->obd_uuid, NULL, NULL);
2354 class_manual_cleanup(obd);
2355 lsi->lsi_dt_dev = NULL;
2358 /* XXX: to keep support old components relying on lsi_srv_mnt
2359 * we get this info from OSD just started */
2360 LASSERT(obd->obd_lu_dev);
2361 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
2362 LASSERT(lsi->lsi_dt_dev);
2364 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
2366 lsi->lsi_srv_mnt = p.ddp_mnt;
2372 /** Fill in the superblock info for a Lustre server.
2373 * Mount the device with the correct options.
2374 * Read the on-disk config file.
2375 * Start the services.
2377 static int server_fill_super(struct super_block *sb)
2379 struct lustre_sb_info *lsi = s2lsi(sb);
2383 rc = lsi_prepare(lsi);
2387 /* Start low level OSD */
2388 rc = osd_start(lsi, sb->s_flags);
2390 CERROR("Unable to start osd on %s: %d\n",
2391 lsi->lsi_lmd->lmd_dev, rc);
2396 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
2397 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
2399 if (class_name2obd(lsi->lsi_svname)) {
2400 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
2401 "running. Double-mount may have compromised"
2402 " the disk journal.\n",
2408 /* Start MGS before MGC */
2409 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
2410 rc = server_start_mgs(sb);
2415 /* Start MGC before servers */
2416 rc = lustre_start_mgc(sb);
2420 /* Set up all obd devices for service */
2421 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
2422 (IS_OST(lsi) || IS_MDT(lsi))) {
2423 rc = server_start_targets(sb, lsi->lsi_srv_mnt);
2425 CERROR("Unable to start targets: %d\n", rc);
2428 /* FIXME overmount client here,
2429 or can we just start a client log and client_fill_super on this sb?
2430 We need to make sure server_put_super gets called too - ll_put_super
2431 calls lustre_common_put_super; check there for LSI_SERVER flag,
2433 Probably should start client from new thread so we can return.
2434 Client will not finish until all servers are connected.
2435 Note - MGS-only server does NOT get a client, since there is no
2436 lustre fs associated - the MGS is for all lustre fs's */
2439 rc = server_fill_super_common(sb);
2445 /* We jump here in case of failure while starting targets or MGS.
2446 * In this case we can't just put @mnt and have to do real cleanup
2447 * with stoping targets, etc. */
2448 server_put_super(sb);
2453 * Calculate timeout value for a target.
2455 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
2457 struct lustre_mount_data *lmd;
2461 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
2462 int min = OBD_RECOVERY_TIME_MIN;
2464 LASSERT(IS_SERVER(lsi));
2468 soft = lmd->lmd_recovery_time_soft;
2469 hard = lmd->lmd_recovery_time_hard;
2470 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
2471 obd->obd_no_ir = !has_ir;
2475 soft = OBD_RECOVERY_TIME_SOFT;
2477 hard = OBD_RECOVERY_TIME_HARD;
2479 /* target may have ir_factor configured. */
2480 factor = OBD_IR_FACTOR_DEFAULT;
2481 if (obd->obd_recovery_ir_factor)
2482 factor = obd->obd_recovery_ir_factor;
2485 int new_soft = soft;
2486 int new_hard = hard;
2488 /* adjust timeout value by imperative recovery */
2490 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
2491 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
2493 /* make sure the timeout is not too short */
2494 new_soft = max(min, new_soft);
2495 new_hard = max(new_soft, new_hard);
2497 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
2498 "window shrunk from %d-%d down to %d-%d\n",
2499 obd->obd_name, soft, hard, new_soft, new_hard);
2506 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
2507 obd->obd_recovery_time_hard = hard;
2508 obd->obd_recovery_ir_factor = factor;
2510 EXPORT_SYMBOL(server_calc_timeout);
2512 /*************** mount common betweeen server and client ***************/
2515 int lustre_common_put_super(struct super_block *sb)
2520 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
2522 /* Drop a ref to the MGC */
2523 rc = lustre_stop_mgc(sb);
2524 if (rc && (rc != -ENOENT)) {
2526 CERROR("Can't stop MGC: %d\n", rc);
2529 /* BUSY just means that there's some other obd that
2530 needs the mgc. Let him clean it up. */
2531 CDEBUG(D_MOUNT, "MGC still in use\n");
2533 /* Drop a ref to the mounted disk */
2538 EXPORT_SYMBOL(lustre_common_put_super);
2540 static void lmd_print(struct lustre_mount_data *lmd)
2544 PRINT_CMD(PRINT_MASK, " mount data:\n");
2545 if (lmd_is_client(lmd))
2546 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
2547 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
2548 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
2551 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
2553 if (lmd->lmd_recovery_time_soft)
2554 PRINT_CMD(PRINT_MASK, "recovery time soft: %d\n",
2555 lmd->lmd_recovery_time_soft);
2557 if (lmd->lmd_recovery_time_hard)
2558 PRINT_CMD(PRINT_MASK, "recovery time hard: %d\n",
2559 lmd->lmd_recovery_time_hard);
2561 for (i = 0; i < lmd->lmd_exclude_count; i++) {
2562 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
2563 lmd->lmd_exclude[i]);
2567 /* Is this server on the exclusion list */
2568 int lustre_check_exclusion(struct super_block *sb, char *svname)
2570 struct lustre_sb_info *lsi = s2lsi(sb);
2571 struct lustre_mount_data *lmd = lsi->lsi_lmd;
2576 rc = server_name2index(svname, &index, NULL);
2577 if (rc != LDD_F_SV_TYPE_OST)
2578 /* Only exclude OSTs */
2581 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
2582 index, lmd->lmd_exclude_count, lmd->lmd_dev);
2584 for(i = 0; i < lmd->lmd_exclude_count; i++) {
2585 if (index == lmd->lmd_exclude[i]) {
2586 CWARN("Excluding %s (on exclusion list)\n", svname);
2593 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
2594 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
2596 char *s1 = ptr, *s2;
2597 __u32 index, *exclude_list;
2601 /* The shortest an ost name can be is 8 chars: -OST0000.
2602 We don't actually know the fsname at this time, so in fact
2603 a user could specify any fsname. */
2604 devmax = strlen(ptr) / 8 + 1;
2606 /* temp storage until we figure out how many we have */
2607 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
2611 /* we enter this fn pointing at the '=' */
2612 while (*s1 && *s1 != ' ' && *s1 != ',') {
2614 rc = server_name2index(s1, &index, &s2);
2616 CERROR("Can't parse server name '%s'\n", s1);
2619 if (rc == LDD_F_SV_TYPE_OST)
2620 exclude_list[lmd->lmd_exclude_count++] = index;
2622 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
2624 /* now we are pointing at ':' (next exclude)
2625 or ',' (end of excludes) */
2626 if (lmd->lmd_exclude_count >= devmax)
2629 if (rc >= 0) /* non-err */
2632 if (lmd->lmd_exclude_count) {
2633 /* permanent, freed in lustre_free_lsi */
2634 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
2635 lmd->lmd_exclude_count);
2636 if (lmd->lmd_exclude) {
2637 memcpy(lmd->lmd_exclude, exclude_list,
2638 sizeof(index) * lmd->lmd_exclude_count);
2641 lmd->lmd_exclude_count = 0;
2644 OBD_FREE(exclude_list, sizeof(index) * devmax);
2648 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
2653 if (lmd->lmd_mgssec != NULL) {
2654 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
2655 lmd->lmd_mgssec = NULL;
2658 tail = strchr(ptr, ',');
2660 length = strlen(ptr);
2662 length = tail - ptr;
2664 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
2665 if (lmd->lmd_mgssec == NULL)
2668 memcpy(lmd->lmd_mgssec, ptr, length);
2669 lmd->lmd_mgssec[length] = '\0';
2673 static int lmd_parse_string(char **handle, char *ptr)
2678 if ((handle == NULL) || (ptr == NULL))
2681 if (*handle != NULL) {
2682 OBD_FREE(*handle, strlen(*handle) + 1);
2686 tail = strchr(ptr, ',');
2688 length = strlen(ptr);
2690 length = tail - ptr;
2692 OBD_ALLOC(*handle, length + 1);
2693 if (*handle == NULL)
2696 memcpy(*handle, ptr, length);
2697 (*handle)[length] = '\0';
2702 /* Collect multiple values for mgsnid specifiers */
2703 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
2711 /* Find end of nidlist */
2712 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
2713 length = tail - *ptr;
2715 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
2719 if (lmd->lmd_mgs != NULL)
2720 oldlen = strlen(lmd->lmd_mgs) + 1;
2722 OBD_ALLOC(mgsnid, oldlen + length + 1);
2726 if (lmd->lmd_mgs != NULL) {
2727 /* Multiple mgsnid= are taken to mean failover locations */
2728 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
2729 mgsnid[oldlen - 1] = ':';
2730 OBD_FREE(lmd->lmd_mgs, oldlen);
2732 memcpy(mgsnid + oldlen, *ptr, length);
2733 mgsnid[oldlen + length] = '\0';
2734 lmd->lmd_mgs = mgsnid;
2740 /** Parse mount line options
2741 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
2742 * dev is passed as device=uml1:/lustre by mount.lustre
2744 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
2746 char *s1, *s2, *devname = NULL;
2747 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
2753 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
2754 "/sbin/mount.lustre is installed.\n");
2758 /* Options should be a string - try to detect old lmd data */
2759 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
2760 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
2761 "/sbin/mount.lustre. Please install "
2762 "version %s\n", LUSTRE_VERSION_STRING);
2765 lmd->lmd_magic = LMD_MAGIC;
2767 OBD_ALLOC(lmd->lmd_params, 4096);
2768 if (lmd->lmd_params == NULL)
2770 lmd->lmd_params[0] = '\0';
2772 /* Set default flags here */
2777 int time_min = OBD_RECOVERY_TIME_MIN;
2779 /* Skip whitespace and extra commas */
2780 while (*s1 == ' ' || *s1 == ',')
2783 /* Client options are parsed in ll_options: eg. flock,
2786 /* Parse non-ldiskfs options here. Rather than modifying
2787 ldiskfs, we just zero these out here */
2788 if (strncmp(s1, "abort_recov", 11) == 0) {
2789 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
2791 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
2792 lmd->lmd_recovery_time_soft = max_t(int,
2793 simple_strtoul(s1 + 19, NULL, 10), time_min);
2795 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
2796 lmd->lmd_recovery_time_hard = max_t(int,
2797 simple_strtoul(s1 + 19, NULL, 10), time_min);
2799 } else if (strncmp(s1, "noir", 4) == 0) {
2800 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
2802 } else if (strncmp(s1, "nosvc", 5) == 0) {
2803 lmd->lmd_flags |= LMD_FLG_NOSVC;
2805 } else if (strncmp(s1, "nomgs", 5) == 0) {
2806 lmd->lmd_flags |= LMD_FLG_NOMGS;
2808 } else if (strncmp(s1, "noscrub", 7) == 0) {
2809 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
2811 } else if (strncmp(s1, PARAM_MGSNODE,
2812 sizeof(PARAM_MGSNODE) - 1) == 0) {
2813 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
2814 /* Assume the next mount opt is the first
2815 invalid nid we get to. */
2816 rc = lmd_parse_mgs(lmd, &s2);
2820 } else if (strncmp(s1, "writeconf", 9) == 0) {
2821 lmd->lmd_flags |= LMD_FLG_WRITECONF;
2823 } else if (strncmp(s1, "virgin", 6) == 0) {
2824 lmd->lmd_flags |= LMD_FLG_VIRGIN;
2826 } else if (strncmp(s1, "noprimnode", 10) == 0) {
2827 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
2829 } else if (strncmp(s1, "mgssec=", 7) == 0) {
2830 rc = lmd_parse_mgssec(lmd, s1 + 7);
2834 /* ost exclusion list */
2835 } else if (strncmp(s1, "exclude=", 8) == 0) {
2836 rc = lmd_make_exclusion(lmd, s1 + 7);
2840 } else if (strncmp(s1, "mgs", 3) == 0) {
2842 lmd->lmd_flags |= LMD_FLG_MGS;
2844 } else if (strncmp(s1, "svname=", 7) == 0) {
2845 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
2849 } else if (strncmp(s1, "param=", 6) == 0) {
2851 char *tail = strchr(s1 + 6, ',');
2853 length = strlen(s1);
2857 strncat(lmd->lmd_params, s1 + 6, length);
2858 strcat(lmd->lmd_params, " ");
2860 } else if (strncmp(s1, "osd=", 4) == 0) {
2861 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
2866 /* Linux 2.4 doesn't pass the device, so we stuck it at the
2867 end of the options. */
2868 else if (strncmp(s1, "device=", 7) == 0) {
2870 /* terminate options right before device. device
2871 must be the last one. */
2877 s2 = strchr(s1, ',');
2885 memmove(s1, s2, strlen(s2) + 1);
2891 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
2892 "(need mount option 'device=...')\n");
2896 s1 = strstr(devname, ":/");
2899 lmd->lmd_flags |= LMD_FLG_CLIENT;
2900 /* Remove leading /s from fsname */
2901 while (*++s1 == '/') ;
2902 /* Freed in lustre_free_lsi */
2903 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
2904 if (!lmd->lmd_profile)
2906 sprintf(lmd->lmd_profile, "%s-client", s1);
2909 /* Freed in lustre_free_lsi */
2910 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
2913 strcpy(lmd->lmd_dev, devname);
2915 /* Save mount options */
2916 s1 = options + strlen(options) - 1;
2917 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
2919 if (*options != 0) {
2920 /* Freed in lustre_free_lsi */
2921 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
2924 strcpy(lmd->lmd_opts, options);
2928 lmd->lmd_magic = LMD_MAGIC;
2933 CERROR("Bad mount options %s\n", options);
2937 struct lustre_mount_data2 {
2939 struct vfsmount *lmd2_mnt;
2942 /** This is the entry point for the mount call into Lustre.
2943 * This is called when a server or client is mounted,
2944 * and this is where we start setting things up.
2945 * @param data Mount options (e.g. -o flock,abort_recov)
2947 int lustre_fill_super(struct super_block *sb, void *data, int silent)
2949 struct lustre_mount_data *lmd;
2950 struct lustre_mount_data2 *lmd2 = data;
2951 struct lustre_sb_info *lsi;
2955 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
2957 lsi = lustre_init_lsi(sb);
2963 * Disable lockdep during mount, because mount locking patterns are
2969 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
2971 obd_zombie_barrier();
2973 /* Figure out the lmd from the mount options */
2974 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
2976 GOTO(out, rc = -EINVAL);
2979 if (lmd_is_client(lmd)) {
2980 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
2981 if (!client_fill_super) {
2982 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
2983 "client mount! Is the 'lustre' "
2984 "module loaded?\n");
2988 rc = lustre_start_mgc(sb);
2993 /* Connect and start */
2994 /* (should always be ll_fill_super) */
2995 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
2996 /* c_f_s will call lustre_common_put_super on failure */
2999 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
3000 rc = server_fill_super(sb);
3001 /* s_f_s calls lustre_start_mgc after the mount because we need
3002 the MGS nids which are stored on disk. Plus, we may
3003 need to start the MGS first. */
3004 /* s_f_s will call server_put_super on failure */
3007 /* If error happens in fill_super() call, @lsi will be killed there.
3008 * This is why we do not put it here. */
3012 CERROR("Unable to mount %s (%d)\n",
3013 s2lsi(sb) ? lmd->lmd_dev : "", rc);
3015 CDEBUG(D_SUPER, "Mount %s complete\n",
3023 /* We can't call ll_fill_super by name because it lives in a module that
3024 must be loaded after this one. */
3025 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
3026 struct vfsmount *mnt))
3028 client_fill_super = cfs;
3030 EXPORT_SYMBOL(lustre_register_client_fill_super);
3032 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
3034 kill_super_cb = cfs;
3036 EXPORT_SYMBOL(lustre_register_kill_super_cb);
3038 /***************** FS registration ******************/
3039 #ifdef HAVE_FSTYPE_MOUNT
3040 struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
3041 const char *devname, void *data)
3043 struct lustre_mount_data2 lmd2 = { data, NULL };
3045 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
3048 int lustre_get_sb(struct file_system_type *fs_type, int flags,
3049 const char *devname, void * data, struct vfsmount *mnt)
3051 struct lustre_mount_data2 lmd2 = { data, mnt };
3053 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
3057 void lustre_kill_super(struct super_block *sb)
3059 struct lustre_sb_info *lsi = s2lsi(sb);
3061 if (kill_super_cb && lsi && !IS_SERVER(lsi))
3062 (*kill_super_cb)(sb);
3064 kill_anon_super(sb);
3067 /** Register the "lustre" fs type
3069 struct file_system_type lustre_fs_type = {
3070 .owner = THIS_MODULE,
3072 #ifdef HAVE_FSTYPE_MOUNT
3073 .mount = lustre_mount,
3075 .get_sb = lustre_get_sb,
3077 .kill_sb = lustre_kill_super,
3078 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
3079 FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE,
3082 int lustre_register_fs(void)
3084 return register_filesystem(&lustre_fs_type);
3087 int lustre_unregister_fs(void)
3089 return unregister_filesystem(&lustre_fs_type);
3092 EXPORT_SYMBOL(server_mti_print);