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);
974 memcpy(uuid->uuid, ospname, strlen(ospname));
975 *exp = cfs_hash_lookup(osp->obd_uuid_hash, uuid);
979 memcpy(ori->ori_name, ospname, strlen(ospname));
981 ori->ori_cb_func = cb_func;
982 ori->ori_cb_data = cb_data;
983 CFS_INIT_LIST_HEAD(&ori->ori_list);
984 cfs_list_add(&ori->ori_list, &osp_register_list);
986 if (*exp != NULL && cb_func != NULL)
989 mutex_unlock(&osp_register_list_lock);
992 EXPORT_SYMBOL(lustre_register_osp_item);
994 void lustre_deregister_osp_item(struct obd_export **exp)
996 struct osp_register_item *ori, *tmp;
998 mutex_lock(&osp_register_list_lock);
999 cfs_list_for_each_entry_safe(ori, tmp, &osp_register_list, ori_list) {
1000 if (exp == ori->ori_exp) {
1002 class_export_put(*exp);
1003 cfs_list_del(&ori->ori_list);
1008 mutex_unlock(&osp_register_list_lock);
1010 EXPORT_SYMBOL(lustre_deregister_osp_item);
1012 static void lustre_notify_osp_list(struct obd_export *exp)
1014 struct osp_register_item *ori, *tmp;
1015 LASSERT(exp != NULL);
1017 mutex_lock(&osp_register_list_lock);
1018 cfs_list_for_each_entry_safe(ori, tmp, &osp_register_list, ori_list) {
1019 if (strcmp(exp->exp_obd->obd_name, ori->ori_name))
1021 if (*ori->ori_exp != NULL)
1023 *ori->ori_exp = class_export_get(exp);
1024 if (ori->ori_cb_func != NULL)
1025 ori->ori_cb_func(ori->ori_cb_data);
1027 mutex_unlock(&osp_register_list_lock);
1030 static int lustre_osp_connect(struct obd_device *osp)
1033 struct lu_context session_ctx;
1034 struct obd_export *exp;
1035 struct obd_uuid *uuid = NULL;
1036 struct obd_connect_data *data = NULL;
1040 /* log has been fully processed, let clients connect */
1041 rc = lu_env_init(&env, osp->obd_lu_dev->ld_type->ldt_ctx_tags);
1045 lu_context_init(&session_ctx, LCT_SESSION);
1046 session_ctx.lc_thread = NULL;
1047 lu_context_enter(&session_ctx);
1048 env.le_ses = &session_ctx;
1050 OBD_ALLOC_PTR(data);
1052 GOTO(out, rc = -ENOMEM);
1054 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
1055 data->ocd_version = LUSTRE_VERSION_CODE;
1056 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
1057 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_IBITS |
1058 OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
1059 OBD_CONNECT_AT | OBD_CONNECT_FULL20 |
1060 OBD_CONNECT_LVB_TYPE |
1061 OBD_CONNECT_LIGHTWEIGHT;
1062 OBD_ALLOC_PTR(uuid);
1064 GOTO(out, rc = -ENOMEM);
1066 if (strlen(osp->obd_name) > sizeof(uuid->uuid)) {
1067 CERROR("%s: Too long osp name %s, max_size is %d\n",
1068 osp->obd_name, osp->obd_name, (int)sizeof(uuid->uuid));
1069 GOTO(out, rc = -EINVAL);
1071 /* Use osp name as the uuid, so we find the export by
1073 memcpy(uuid->uuid, osp->obd_name, strlen(osp->obd_name));
1074 rc = obd_connect(&env, &exp, osp, uuid, data, NULL);
1076 CERROR("%s: connect failed: rc = %d\n", osp->obd_name, rc);
1078 lustre_notify_osp_list(exp);
1087 lu_context_exit(&session_ctx);
1088 lu_context_fini(&session_ctx);
1094 * osp-on-ost is used by slaves (Non-MDT0 targets) to manage the connection
1097 * The OSTs will communicate with MDT0 by the connection established by the
1098 * osp-on-ost to get quota and fid sequence.
1101 static int lustre_osp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi)
1103 struct obd_connect_data *data = NULL;
1104 struct obd_device *obd;
1105 char *ospname = NULL;
1106 char *ospuuid = NULL;
1110 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
1113 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
1117 OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
1118 if (ospname == NULL)
1119 GOTO(out, rc = -ENOMEM);
1121 rc = tgt_name2ospname(lsi->lsi_svname, ospname);
1123 CERROR("%s change ospname error: rc %d\n",
1124 lsi->lsi_svname, rc);
1128 OBD_ALLOC(ospuuid, MTI_NAME_MAXLEN);
1129 if (ospuuid == NULL)
1130 GOTO(out, rc = -ENOMEM);
1132 sprintf(ospuuid, "%s_UUID", ospname);
1133 rc = lustre_start_simple(ospname, LUSTRE_OSP_NAME,
1134 ospuuid, lustre_cfg_string(lcfg, 1),
1137 CERROR("%s: setup up failed: rc %d\n", ospname, rc);
1141 obd = class_name2obd(ospname);
1142 LASSERT(obd != NULL);
1144 rc = lustre_osp_connect(obd);
1146 CERROR("%s: connect failed: rc = %d\n", ospname, rc);
1150 if (ospname != NULL)
1151 OBD_FREE(ospname, MTI_NAME_MAXLEN);
1152 if (ospuuid != NULL)
1153 OBD_FREE(ospuuid, MTI_NAME_MAXLEN);
1158 static int lustre_osp_add_conn(struct lustre_cfg *cfg,
1159 struct lustre_sb_info *lsi)
1161 struct lustre_cfg_bufs *bufs = NULL;
1162 struct lustre_cfg *lcfg = NULL;
1163 char *ospname = NULL;
1164 struct obd_device *osp;
1168 OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
1169 if (ospname == NULL)
1170 GOTO(out, rc = -ENOMEM);
1172 rc = tgt_name2ospname(lsi->lsi_svname, ospname);
1174 CERROR("%s change ospname error: rc %d\n",
1175 lsi->lsi_svname, rc);
1179 OBD_ALLOC_PTR(bufs);
1183 lustre_cfg_bufs_reset(bufs, ospname);
1184 lustre_cfg_bufs_set_string(bufs, 1,
1185 lustre_cfg_string(cfg, 1));
1187 lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
1189 osp = class_name2obd(ospname);
1191 CERROR("Can not find %s\n", ospname);
1192 GOTO(out, rc = -EINVAL);
1195 rc = class_add_conn(osp, lcfg);
1197 CERROR("%s: can't add conn: rc = %d\n", ospname, rc);
1203 lustre_cfg_free(lcfg);
1204 if (ospname != NULL)
1205 OBD_FREE(ospname, MTI_NAME_MAXLEN);
1211 * Retrieve MDT nids from the client log, then start the osp-on-ost device.
1212 * there are only two scenarios which would include mdt nid.
1214 * marker 5 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add mdc' xxx-
1215 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
1216 * attach 0:lustre-MDT0000-mdc 1:mdc 2:lustre-clilmv_UUID
1217 * setup 0:lustre-MDT0000-mdc 1:lustre-MDT0000_UUID 2:192.168.122.162@tcp
1218 * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
1219 * add_conn 0:lustre-MDT0000-mdc 1:192.168.172.1@tcp
1220 * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID xxxx
1221 * marker 5 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add mdc' xxxx-
1223 * marker 7 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
1224 * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
1225 * add_conn 0:lustre-MDT0000-mdc 1:192.168.122.2@tcp
1226 * marker 7 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
1228 static int client_osp_config_process(const struct lu_env *env,
1229 struct llog_handle *handle,
1230 struct llog_rec_hdr *rec, void *data)
1232 struct config_llog_instance *clli = data;
1233 int cfg_len = rec->lrh_len;
1234 char *cfg_buf = (char *) (rec + 1);
1235 struct lustre_cfg *lcfg = NULL;
1236 struct lustre_sb_info *lsi;
1237 int rc = 0, swab = 0;
1240 if (rec->lrh_type != OBD_CFG_REC) {
1241 CERROR("Unknown llog record type %#x encountered\n",
1246 LASSERT(clli->cfg_sb != NULL);
1247 lsi = s2lsi(clli->cfg_sb);
1249 lcfg = (struct lustre_cfg *)cfg_buf;
1250 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
1251 lustre_swab_lustre_cfg(lcfg);
1255 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
1259 switch (lcfg->lcfg_command) {
1261 struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
1263 lustre_swab_cfg_marker(marker, swab,
1264 LUSTRE_CFG_BUFLEN(lcfg, 1));
1265 if (marker->cm_flags & CM_SKIP ||
1266 marker->cm_flags & CM_EXCLUDE)
1269 if (!tgt_is_mdt0(marker->cm_tgtname))
1272 if(!strncmp(marker->cm_comment, "add mdc", 7) ||
1273 !strncmp(marker->cm_comment, "add failnid", 11)) {
1274 if (marker->cm_flags & CM_START) {
1275 clli->cfg_flags = CFG_F_MARKER;
1276 /* This hack is to differentiate the
1277 * ADD_UUID is come from "add mdc" record
1278 * or from "add failnid" record. */
1279 if (!strncmp(marker->cm_comment,
1281 clli->cfg_flags |= CFG_F_SKIP;
1282 } else if (marker->cm_flags & CM_END) {
1283 clli->cfg_flags = 0;
1288 case LCFG_ADD_UUID: {
1289 if (clli->cfg_flags == CFG_F_MARKER) {
1290 rc = lustre_osp_setup(lcfg, lsi);
1291 /* XXX: process only the first nid as
1292 * we don't need another instance of osp */
1293 clli->cfg_flags |= CFG_F_SKIP;
1294 } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
1295 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
1298 CERROR("%s: Fail to add uuid, rc:%d\n",
1299 lsi->lsi_svname, rc);
1303 case LCFG_ADD_CONN: {
1304 if (is_mdc_for_mdt0(lustre_cfg_string(lcfg, 0)))
1305 rc = lustre_osp_add_conn(lcfg, lsi);
1315 static int lustre_disconnect_osp(struct super_block *sb)
1317 struct lustre_sb_info *lsi = s2lsi(sb);
1318 struct obd_device *osp;
1319 char *ospname = NULL;
1320 char *logname = NULL;
1321 struct lustre_cfg *lcfg = NULL;
1322 struct lustre_cfg_bufs *bufs = NULL;
1323 struct config_llog_instance *cfg = NULL;
1327 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
1331 /* Only disconnect MDT0-osp-MDT0 here, other osp on MDT
1332 * will be disconnect during MDT stack cleanup.
1333 * FIXME: remove later when quota on DNE is finished */
1334 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1340 OBD_ALLOC(logname, MTI_NAME_MAXLEN);
1341 if (logname == NULL)
1344 OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
1345 if (ospname == NULL)
1346 GOTO(out, rc = -ENOMEM);
1348 rc = server_name2fsname(lsi->lsi_svname, ospname, NULL);
1350 CERROR("%s: get fsname error: %d\n",
1351 lsi->lsi_svname, rc);
1354 sprintf(logname, "%s-client", ospname);
1358 GOTO(out, rc = -ENOMEM);
1361 cfg->cfg_instance = sb;
1362 rc = lustre_end_log(sb, logname, cfg);
1364 CERROR("Can't end config log %s\n", ospname);
1368 rc = tgt_name2ospname(lsi->lsi_svname, ospname);
1370 CERROR("%s: get osp name error: %d\n",
1371 lsi->lsi_svname, rc);
1375 osp = class_name2obd(ospname);
1377 CERROR("Can't find osp-on-ost %s\n", ospname);
1378 GOTO(out, rc = -ENOENT);
1381 OBD_ALLOC_PTR(bufs);
1383 GOTO(out, rc = -ENOMEM);
1385 lustre_cfg_bufs_reset(bufs, osp->obd_name);
1386 lustre_cfg_bufs_set_string(bufs, 1, NULL);
1387 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
1389 GOTO(out, rc = -ENOMEM);
1391 /* Disconnect import first. NULL is passed for the '@env', since
1392 * it will not be used for the 'osp-on-ost'. (see osp_shutdown()) */
1393 rc = osp->obd_lu_dev->ld_ops->ldo_process_config(NULL, osp->obd_lu_dev,
1397 lustre_cfg_free(lcfg);
1403 OBD_FREE(ospname, MTI_NAME_MAXLEN);
1405 OBD_FREE(logname, MTI_NAME_MAXLEN);
1410 * Stop the osp(fsname-MDT0000-osp-OSTxxxx) or (fsname-MDT0000-osp-MDT0000) for an OST target.
1412 static int lustre_stop_osp(struct super_block *sb)
1414 struct lustre_sb_info *lsi = s2lsi(sb);
1415 struct obd_device *osp = NULL;
1416 char *ospname = NULL;
1420 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
1424 /* Only disconnect MDT0-osp-MDT0 here, other osp on MDT
1425 * will be disconnect during MDT stack cleanup.
1426 * FIXME: remove later when quota on DNE is finished */
1427 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1434 OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
1435 rc = tgt_name2ospname(lsi->lsi_svname, ospname);
1437 CERROR("%s get fsname error: rc %d\n",
1438 lsi->lsi_svname, rc);
1442 osp = class_name2obd(ospname);
1444 CERROR("Can not find osp-on-ost %s\n", ospname);
1445 GOTO(cleanup, rc = -ENOENT);
1449 rc = class_manual_cleanup(osp);
1452 if (ospname != NULL)
1453 OBD_FREE(ospname, MTI_NAME_MAXLEN);
1457 DEFINE_MUTEX(server_start_lock);
1459 /* Stop MDS/OSS if nobody is using them */
1460 static int server_stop_servers(int lsiflags)
1462 struct obd_device *obd = NULL;
1463 struct obd_type *type = NULL;
1467 mutex_lock(&server_start_lock);
1469 /* Either an MDT or an OST or neither */
1470 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
1471 if ((lsiflags & LDD_F_SV_TYPE_MDT) &&
1472 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
1473 type = class_search_type(LUSTRE_MDT_NAME);
1475 /* if this was an OST, and there are no more OST's, clean up the OSS */
1476 if ((lsiflags & LDD_F_SV_TYPE_OST) &&
1477 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
1478 type = class_search_type(LUSTRE_OST_NAME);
1481 if (obd && (!type || !type->typ_refcnt)) {
1484 /* obd_fail doesn't mean much on a server obd */
1485 err = class_manual_cleanup(obd);
1490 mutex_unlock(&server_start_lock);
1495 int server_mti_print(char *title, struct mgs_target_info *mti)
1497 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
1498 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
1499 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
1500 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
1501 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
1502 mti->mti_config_ver, mti->mti_flags);
1507 * Get service name (svname) from string
1509 * if endptr isn't NULL it is set to end of fsname *
1511 int server_name2svname(char *label, char *svname, char **endptr)
1516 /* We use server_name2fsname() just for parsing */
1517 rc = server_name2fsname(label, NULL, &dash);
1524 strncpy(svname, dash + 1, MTI_NAME_MAXLEN);
1528 EXPORT_SYMBOL(server_name2svname);
1531 /* Get the index from the obd name.
1532 rc = server type, or
1534 if endptr isn't NULL it is set to end of name */
1535 int server_name2index(char *svname, __u32 *idx, char **endptr)
1537 unsigned long index;
1541 /* We use server_name2fsname() just for parsing */
1542 rc = server_name2fsname(svname, NULL, &dash);
1551 if (strncmp(dash, "MDT", 3) == 0)
1552 rc = LDD_F_SV_TYPE_MDT;
1553 else if (strncmp(dash, "OST", 3) == 0)
1554 rc = LDD_F_SV_TYPE_OST;
1560 if (strcmp(dash, "all") == 0)
1561 return rc | LDD_F_SV_ALL;
1563 index = simple_strtoul(dash, endptr, 16);
1568 EXPORT_SYMBOL(server_name2index);
1570 /* Generate data for registration */
1571 static int server_lsi2mti(struct lustre_sb_info *lsi,
1572 struct mgs_target_info *mti)
1574 lnet_process_id_t id;
1578 if (!IS_SERVER(lsi))
1581 strncpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname));
1583 mti->mti_nid_count = 0;
1584 while (LNetGetId(i++, &id) != -ENOENT) {
1585 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1588 /* server use --servicenode param, only allow specified
1589 * nids be registered */
1590 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
1591 class_match_nid(lsi->lsi_lmd->lmd_params,
1592 PARAM_FAILNODE, id.nid) < 1)
1595 /* match specified network */
1596 if (!class_match_net(lsi->lsi_lmd->lmd_params,
1597 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1600 mti->mti_nids[mti->mti_nid_count] = id.nid;
1601 mti->mti_nid_count++;
1602 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1603 CWARN("Only using first %d nids for %s\n",
1604 mti->mti_nid_count, mti->mti_svname);
1609 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1610 mti->mti_config_ver = 0;
1612 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1616 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1619 /* Orion requires index to be set */
1620 LASSERT(!(rc & LDD_F_NEED_INDEX));
1621 /* keep only LDD flags */
1622 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1623 if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN))
1624 mti->mti_flags |= LDD_F_UPDATE;
1625 strncpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1626 sizeof(mti->mti_params));
1630 /* Register an old or new target with the MGS. If needed MGS will construct
1631 startup logs and assign index */
1632 static int server_register_target(struct lustre_sb_info *lsi)
1634 struct obd_device *mgc = lsi->lsi_mgc;
1635 struct mgs_target_info *mti = NULL;
1642 if (!IS_SERVER(lsi))
1649 rc = server_lsi2mti(lsi, mti);
1653 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1654 mti->mti_svname, mti->mti_fsname,
1655 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1658 /* if write_conf is true, the registration must succeed */
1659 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1660 mti->mti_flags |= LDD_F_OPC_REG;
1662 /* Register the target */
1663 /* FIXME use mgc_process_config instead */
1664 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1665 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
1666 sizeof(*mti), mti, NULL);
1668 if (mti->mti_flags & LDD_F_ERROR) {
1669 LCONSOLE_ERROR_MSG(0x160,
1670 "The MGS is refusing to allow this "
1671 "server (%s) to start. Please see messages"
1672 " on the MGS node.\n", lsi->lsi_svname);
1673 } else if (writeconf) {
1674 LCONSOLE_ERROR_MSG(0x15f,
1675 "Communication to the MGS return error %d. "
1676 "Is the MGS running?\n", rc);
1678 CERROR("Cannot talk to the MGS: %d, not fatal\n", rc);
1679 /* reset the error code for non-fatal error. */
1692 * Notify the MGS that this target is ready.
1693 * Used by IR - if the MGS receives this message, it will notify clients.
1695 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1697 struct lustre_sb_info *lsi = s2lsi(sb);
1698 struct obd_device *mgc = lsi->lsi_mgc;
1699 struct mgs_target_info *mti = NULL;
1705 if (!(IS_SERVER(lsi)))
1711 rc = server_lsi2mti(lsi, mti);
1715 mti->mti_instance = obd->u.obt.obt_instance;
1716 mti->mti_flags |= LDD_F_OPC_READY;
1718 /* FIXME use mgc_process_config instead */
1719 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1720 sizeof(KEY_REGISTER_TARGET),
1721 KEY_REGISTER_TARGET,
1722 sizeof(*mti), mti, NULL);
1724 /* Imperative recovery: if the mgs informs us to use IR? */
1725 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1726 (mti->mti_flags & LDD_F_IR_CAPABLE))
1727 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1737 * Start the osp(fsname-MDT0000-osp-OSTxxxx) for an OST target,
1738 * which would be used to communicate with MDT0 for quota and FID.
1740 static int lustre_start_osp(struct super_block *sb)
1742 struct lustre_sb_info *lsi = s2lsi(sb);
1743 struct config_llog_instance *cfg = NULL;
1744 struct obd_device *osp;
1745 char *ospname = NULL;
1746 char *logname = NULL;
1751 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
1752 OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
1753 OBD_ALLOC(logname, MTI_NAME_MAXLEN);
1754 if (ospname == NULL || logname == NULL)
1755 GOTO(cleanup, rc = -ENOMEM);
1757 rc = server_name2fsname(lsi->lsi_svname, ospname, &tgt);
1759 CERROR("%s change fsname error: rc %d\n",
1760 lsi->lsi_svname, rc);
1763 sprintf(logname, "%s-client", ospname);
1765 rc = tgt_name2ospname(lsi->lsi_svname, ospname);
1767 CERROR("%s change ospname error: rc %d\n",
1768 lsi->lsi_svname, rc);
1772 osp = class_name2obd(ospname);
1774 GOTO(cleanup, rc = 0);
1778 GOTO(cleanup, rc = -ENOMEM);
1780 cfg->cfg_callback = client_osp_config_process;
1781 cfg->cfg_instance = sb;
1783 rc = lustre_process_log(sb, logname, cfg);
1786 if (ospname != NULL)
1787 OBD_FREE(ospname, MTI_NAME_MAXLEN);
1788 if (logname != NULL)
1789 OBD_FREE(logname, MTI_NAME_MAXLEN);
1796 /** Start server targets: MDTs and OSTs
1798 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1800 struct obd_device *obd;
1801 struct lustre_sb_info *lsi = s2lsi(sb);
1802 struct config_llog_instance cfg;
1804 struct lu_device *dev;
1808 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1811 /* make sure the MDS is started */
1812 mutex_lock(&server_start_lock);
1813 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1815 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1817 LUSTRE_MDS_OBDNAME"_uuid",
1820 mutex_unlock(&server_start_lock);
1821 CERROR("failed to start MDS: %d\n", rc);
1825 mutex_unlock(&server_start_lock);
1828 /* If we're an OST, make sure the global OSS is running */
1830 /* make sure OSS is started */
1831 mutex_lock(&server_start_lock);
1832 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1834 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1836 LUSTRE_OSS_OBDNAME"_uuid",
1839 mutex_unlock(&server_start_lock);
1840 CERROR("failed to start OSS: %d\n", rc);
1844 mutex_unlock(&server_start_lock);
1847 /* Set the mgc fs to our server disk. This allows the MGC to
1848 * read and write configs locally, in case it can't talk to the MGS. */
1849 if (lsi->lsi_srv_mnt) {
1850 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1852 GOTO(out_stop_service, rc);
1855 /* Register with MGS */
1856 rc = server_register_target(lsi);
1860 /* Let the target look up the mount using the target's name
1861 (we can't pass the sb or mnt through class_process_config.) */
1862 rc = server_register_mount(lsi->lsi_svname, sb, mnt);
1866 /* Start targets using the llog named for the target */
1867 memset(&cfg, 0, sizeof(cfg));
1868 cfg.cfg_callback = class_config_llog_handler;
1869 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1871 CERROR("failed to start server %s: %d\n",
1872 lsi->lsi_svname, rc);
1873 /* Do NOT call server_deregister_mount() here. This makes it
1874 * impossible to find mount later in cleanup time and leaves
1875 * @lsi and othder stuff leaked. -umka */
1879 obd = class_name2obd(lsi->lsi_svname);
1881 CERROR("no server named %s was started\n", lsi->lsi_svname);
1882 GOTO(out_mgc, rc = -ENXIO);
1885 if (IS_OST(lsi) || IS_MDT(lsi)) {
1886 rc = lustre_start_osp(sb);
1888 CERROR("%s: failed to start OSP: %d\n",
1889 lsi->lsi_svname, rc);
1894 server_notify_target(sb, obd);
1896 /* calculate recovery timeout, do it after lustre_process_log */
1897 server_calc_timeout(lsi, obd);
1899 /* log has been fully processed */
1900 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1902 /* log has been fully processed, let clients connect */
1903 dev = obd->obd_lu_dev;
1904 if (dev && dev->ld_ops->ldo_prepare) {
1905 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1907 struct lu_context session_ctx;
1909 lu_context_init(&session_ctx, LCT_SESSION);
1910 session_ctx.lc_thread = NULL;
1911 lu_context_enter(&session_ctx);
1912 env.le_ses = &session_ctx;
1914 dev->ld_ops->ldo_prepare(&env, NULL, dev);
1917 lu_context_exit(&session_ctx);
1918 lu_context_fini(&session_ctx);
1922 /* abort recovery only on the complete stack:
1923 * many devices can be involved */
1924 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1925 (OBP(obd, iocontrol))) {
1926 obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
1931 /* Release the mgc fs for others to use */
1932 if (lsi->lsi_srv_mnt)
1933 server_mgc_clear_fs(lsi->lsi_mgc);
1937 server_stop_servers(lsi->lsi_flags);
1942 /***************** lustre superblock **************/
1944 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1946 struct lustre_sb_info *lsi;
1952 OBD_ALLOC_PTR(lsi->lsi_lmd);
1953 if (!lsi->lsi_lmd) {
1958 lsi->lsi_lmd->lmd_exclude_count = 0;
1959 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
1960 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
1961 s2lsi_nocast(sb) = lsi;
1962 /* we take 1 extra ref for our setup */
1963 cfs_atomic_set(&lsi->lsi_mounts, 1);
1965 /* Default umount style */
1966 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1971 static int lustre_free_lsi(struct super_block *sb)
1973 struct lustre_sb_info *lsi = s2lsi(sb);
1976 LASSERT(lsi != NULL);
1977 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
1979 /* someone didn't call server_put_mount. */
1980 LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0);
1982 if (lsi->lsi_lmd != NULL) {
1983 if (lsi->lsi_lmd->lmd_dev != NULL)
1984 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1985 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1986 if (lsi->lsi_lmd->lmd_profile != NULL)
1987 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1988 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1989 if (lsi->lsi_lmd->lmd_mgssec != NULL)
1990 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
1991 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
1992 if (lsi->lsi_lmd->lmd_opts != NULL)
1993 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1994 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1995 if (lsi->lsi_lmd->lmd_exclude_count)
1996 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1997 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1998 lsi->lsi_lmd->lmd_exclude_count);
1999 if (lsi->lsi_lmd->lmd_mgs != NULL)
2000 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
2001 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
2002 if (lsi->lsi_lmd->lmd_osd_type != NULL)
2003 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
2004 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
2005 if (lsi->lsi_lmd->lmd_params != NULL)
2006 OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
2008 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
2011 LASSERT(lsi->lsi_llsbi == NULL);
2012 OBD_FREE(lsi, sizeof(*lsi));
2013 s2lsi_nocast(sb) = NULL;
2018 /* The lsi has one reference for every server that is using the disk -
2019 e.g. MDT, MGS, and potentially MGC */
2020 static int lustre_put_lsi(struct super_block *sb)
2022 struct lustre_sb_info *lsi = s2lsi(sb);
2025 LASSERT(lsi != NULL);
2027 CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts));
2028 if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) {
2029 if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
2030 obd_disconnect(lsi->lsi_osd_exp);
2031 /* wait till OSD is gone */
2032 obd_zombie_barrier();
2034 lustre_free_lsi(sb);
2040 static int lsi_prepare(struct lustre_sb_info *lsi)
2047 LASSERT(lsi->lsi_lmd);
2049 /* The server name is given as a mount line option */
2050 if (lsi->lsi_lmd->lmd_profile == NULL) {
2051 LCONSOLE_ERROR("Can't determine server name\n");
2055 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
2056 RETURN(-ENAMETOOLONG);
2058 strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
2060 /* Determine osd type */
2061 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
2062 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
2063 sizeof(lsi->lsi_osd_type))
2064 RETURN(-ENAMETOOLONG);
2066 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
2068 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
2071 /* XXX: a temp. solution for components using fsfilt
2072 * to be removed in one of the subsequent patches */
2073 if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs")) {
2074 strcpy(lsi->lsi_fstype, "ldiskfs");
2076 strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
2079 /* Determine server type */
2080 rc = server_name2index(lsi->lsi_svname, &index, NULL);
2082 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
2083 /* Assume we're a bare MGS */
2085 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
2087 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
2092 lsi->lsi_flags |= rc;
2094 /* Add mount line flags that used to be in ldd:
2095 * writeconf, mgs, anything else?
2097 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
2098 LDD_F_WRITECONF : 0;
2099 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
2101 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
2102 LDD_F_SV_TYPE_MGS : 0;
2103 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
2104 LDD_F_NO_PRIMNODE : 0;
2109 /*************** server mount ******************/
2111 /** Start the shutdown of servers at umount.
2113 static void server_put_super(struct super_block *sb)
2115 struct lustre_sb_info *lsi = s2lsi(sb);
2116 struct obd_device *obd;
2117 char *tmpname, *extraname = NULL;
2119 int lsiflags = lsi->lsi_flags;
2122 LASSERT(IS_SERVER(lsi));
2124 tmpname_sz = strlen(lsi->lsi_svname) + 1;
2125 OBD_ALLOC(tmpname, tmpname_sz);
2126 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
2127 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
2128 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
2129 snprintf(tmpname, tmpname_sz, "MGS");
2131 /* disconnect the osp-on-ost first to drain off the inflight request */
2132 if (IS_OST(lsi) || IS_MDT(lsi)) {
2135 rc = lustre_disconnect_osp(sb);
2136 if (rc && rc != ETIMEDOUT)
2137 CERROR("%s: failed to disconnect osp-on-ost (rc=%d)!\n",
2141 /* Stop the target */
2142 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
2143 (IS_MDT(lsi) || IS_OST(lsi))) {
2144 struct lustre_profile *lprof = NULL;
2146 /* tell the mgc to drop the config log */
2147 lustre_end_log(sb, lsi->lsi_svname, NULL);
2149 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
2150 If there are any setup/cleanup errors, save the lov
2151 name for safety cleanup later. */
2152 lprof = class_get_profile(lsi->lsi_svname);
2153 if (lprof && lprof->lp_dt) {
2154 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
2155 strcpy(extraname, lprof->lp_dt);
2158 obd = class_name2obd(lsi->lsi_svname);
2160 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
2161 if (lsiflags & LSI_UMOUNT_FAILOVER)
2163 /* We can't seem to give an error return code
2164 * to .put_super, so we better make sure we clean up! */
2166 class_manual_cleanup(obd);
2168 CERROR("no obd %s\n", lsi->lsi_svname);
2169 server_deregister_mount(lsi->lsi_svname);
2173 /* If they wanted the mgs to stop separately from the mdt, they
2174 should have put it on a different device. */
2176 /* if MDS start with --nomgs, don't stop MGS then */
2177 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
2178 server_stop_mgs(sb);
2181 if (IS_OST(lsi) || IS_MDT(lsi)) {
2182 if (lustre_stop_osp(sb) < 0)
2183 CERROR("%s: Fail to stop osp-on-ost!\n", tmpname);
2186 /* Clean the mgc and sb */
2187 lustre_common_put_super(sb);
2189 /* wait till all in-progress cleanups are done
2190 * specifically we're interested in ofd cleanup
2192 obd_zombie_barrier();
2194 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
2195 until the target is really gone so that our type refcount check
2197 server_stop_servers(lsiflags);
2199 /* In case of startup or cleanup err, stop related obds */
2201 obd = class_name2obd(extraname);
2203 CWARN("Cleaning orphaned obd %s\n", extraname);
2205 class_manual_cleanup(obd);
2207 OBD_FREE(extraname, strlen(extraname) + 1);
2210 LCONSOLE_WARN("server umount %s complete\n", tmpname);
2211 OBD_FREE(tmpname, tmpname_sz);
2215 /** Called only for 'umount -f'
2217 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2218 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
2220 struct super_block *sb = vfsmnt->mnt_sb;
2222 static void server_umount_begin(struct super_block *sb)
2225 struct lustre_sb_info *lsi = s2lsi(sb);
2228 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2229 if (!(flags & MNT_FORCE)) {
2235 CDEBUG(D_MOUNT, "umount -f\n");
2236 /* umount = failover
2238 no third way to do non-force, non-failover */
2239 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
2243 static int server_statfs (struct dentry *dentry, cfs_kstatfs_t *buf)
2245 struct super_block *sb = dentry->d_sb;
2246 struct lustre_sb_info *lsi = s2lsi(sb);
2247 struct obd_statfs statfs;
2251 if (lsi->lsi_dt_dev) {
2252 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
2254 statfs_unpack(buf, &statfs);
2255 buf->f_type = sb->s_magic;
2261 buf->f_type = sb->s_magic;
2262 buf->f_bsize = sb->s_blocksize;
2268 buf->f_namelen = NAME_MAX;
2272 /** The operations we support directly on the superblock:
2273 * mount, umount, and df.
2275 static struct super_operations server_ops =
2277 .put_super = server_put_super,
2278 .umount_begin = server_umount_begin, /* umount -f */
2279 .statfs = server_statfs,
2282 #define log2(n) ffz(~(n))
2283 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
2285 static int server_fill_super_common(struct super_block *sb)
2287 struct inode *root = 0;
2290 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
2292 sb->s_blocksize = 4096;
2293 sb->s_blocksize_bits = log2(sb->s_blocksize);
2294 sb->s_magic = LUSTRE_SUPER_MAGIC;
2295 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
2296 sb->s_flags |= MS_RDONLY;
2297 sb->s_op = &server_ops;
2299 root = new_inode(sb);
2301 CERROR("Can't make root inode\n");
2305 /* returns -EIO for every operation */
2306 /* make_bad_inode(root); -- badness - can't umount */
2307 /* apparently we need to be a directory for the mount to finish */
2308 root->i_mode = S_IFDIR;
2310 sb->s_root = d_make_root(root);
2312 CERROR("%s: can't make root dentry\n", sb->s_id);
2319 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
2321 struct lustre_mount_data *lmd = lsi->lsi_lmd;
2322 struct obd_device *obd;
2323 struct dt_device_param p;
2329 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
2330 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
2332 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
2333 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
2334 strcat(lsi->lsi_osd_uuid, "_UUID");
2335 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
2337 obd = class_name2obd(lsi->lsi_osd_obdname);
2339 rc = lustre_start_simple(lsi->lsi_osd_obdname,
2341 lsi->lsi_osd_uuid, lmd->lmd_dev,
2342 flagstr, lsi->lsi_lmd->lmd_opts,
2346 obd = class_name2obd(lsi->lsi_osd_obdname);
2350 rc = obd_connect(NULL, &lsi->lsi_osd_exp, obd, &obd->obd_uuid, NULL, NULL);
2353 class_manual_cleanup(obd);
2354 lsi->lsi_dt_dev = NULL;
2357 /* XXX: to keep support old components relying on lsi_srv_mnt
2358 * we get this info from OSD just started */
2359 LASSERT(obd->obd_lu_dev);
2360 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
2361 LASSERT(lsi->lsi_dt_dev);
2363 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
2365 lsi->lsi_srv_mnt = p.ddp_mnt;
2371 /** Fill in the superblock info for a Lustre server.
2372 * Mount the device with the correct options.
2373 * Read the on-disk config file.
2374 * Start the services.
2376 static int server_fill_super(struct super_block *sb)
2378 struct lustre_sb_info *lsi = s2lsi(sb);
2382 rc = lsi_prepare(lsi);
2386 /* Start low level OSD */
2387 rc = osd_start(lsi, sb->s_flags);
2389 CERROR("Unable to start osd on %s: %d\n",
2390 lsi->lsi_lmd->lmd_dev, rc);
2395 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
2396 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
2398 if (class_name2obd(lsi->lsi_svname)) {
2399 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
2400 "running. Double-mount may have compromised"
2401 " the disk journal.\n",
2407 /* Start MGS before MGC */
2408 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
2409 rc = server_start_mgs(sb);
2414 /* Start MGC before servers */
2415 rc = lustre_start_mgc(sb);
2419 /* Set up all obd devices for service */
2420 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
2421 (IS_OST(lsi) || IS_MDT(lsi))) {
2422 rc = server_start_targets(sb, lsi->lsi_srv_mnt);
2424 CERROR("Unable to start targets: %d\n", rc);
2427 /* FIXME overmount client here,
2428 or can we just start a client log and client_fill_super on this sb?
2429 We need to make sure server_put_super gets called too - ll_put_super
2430 calls lustre_common_put_super; check there for LSI_SERVER flag,
2432 Probably should start client from new thread so we can return.
2433 Client will not finish until all servers are connected.
2434 Note - MGS-only server does NOT get a client, since there is no
2435 lustre fs associated - the MGS is for all lustre fs's */
2438 rc = server_fill_super_common(sb);
2444 /* We jump here in case of failure while starting targets or MGS.
2445 * In this case we can't just put @mnt and have to do real cleanup
2446 * with stoping targets, etc. */
2447 server_put_super(sb);
2452 * Calculate timeout value for a target.
2454 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
2456 struct lustre_mount_data *lmd;
2460 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
2461 int min = OBD_RECOVERY_TIME_MIN;
2463 LASSERT(IS_SERVER(lsi));
2467 soft = lmd->lmd_recovery_time_soft;
2468 hard = lmd->lmd_recovery_time_hard;
2469 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
2470 obd->obd_no_ir = !has_ir;
2474 soft = OBD_RECOVERY_TIME_SOFT;
2476 hard = OBD_RECOVERY_TIME_HARD;
2478 /* target may have ir_factor configured. */
2479 factor = OBD_IR_FACTOR_DEFAULT;
2480 if (obd->obd_recovery_ir_factor)
2481 factor = obd->obd_recovery_ir_factor;
2484 int new_soft = soft;
2485 int new_hard = hard;
2487 /* adjust timeout value by imperative recovery */
2489 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
2490 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
2492 /* make sure the timeout is not too short */
2493 new_soft = max(min, new_soft);
2494 new_hard = max(new_soft, new_hard);
2496 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
2497 "window shrunk from %d-%d down to %d-%d\n",
2498 obd->obd_name, soft, hard, new_soft, new_hard);
2505 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
2506 obd->obd_recovery_time_hard = hard;
2507 obd->obd_recovery_ir_factor = factor;
2509 EXPORT_SYMBOL(server_calc_timeout);
2511 /*************** mount common betweeen server and client ***************/
2514 int lustre_common_put_super(struct super_block *sb)
2519 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
2521 /* Drop a ref to the MGC */
2522 rc = lustre_stop_mgc(sb);
2523 if (rc && (rc != -ENOENT)) {
2525 CERROR("Can't stop MGC: %d\n", rc);
2528 /* BUSY just means that there's some other obd that
2529 needs the mgc. Let him clean it up. */
2530 CDEBUG(D_MOUNT, "MGC still in use\n");
2532 /* Drop a ref to the mounted disk */
2537 EXPORT_SYMBOL(lustre_common_put_super);
2539 static void lmd_print(struct lustre_mount_data *lmd)
2543 PRINT_CMD(PRINT_MASK, " mount data:\n");
2544 if (lmd_is_client(lmd))
2545 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
2546 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
2547 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
2550 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
2552 if (lmd->lmd_recovery_time_soft)
2553 PRINT_CMD(PRINT_MASK, "recovery time soft: %d\n",
2554 lmd->lmd_recovery_time_soft);
2556 if (lmd->lmd_recovery_time_hard)
2557 PRINT_CMD(PRINT_MASK, "recovery time hard: %d\n",
2558 lmd->lmd_recovery_time_hard);
2560 for (i = 0; i < lmd->lmd_exclude_count; i++) {
2561 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
2562 lmd->lmd_exclude[i]);
2566 /* Is this server on the exclusion list */
2567 int lustre_check_exclusion(struct super_block *sb, char *svname)
2569 struct lustre_sb_info *lsi = s2lsi(sb);
2570 struct lustre_mount_data *lmd = lsi->lsi_lmd;
2575 rc = server_name2index(svname, &index, NULL);
2576 if (rc != LDD_F_SV_TYPE_OST)
2577 /* Only exclude OSTs */
2580 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
2581 index, lmd->lmd_exclude_count, lmd->lmd_dev);
2583 for(i = 0; i < lmd->lmd_exclude_count; i++) {
2584 if (index == lmd->lmd_exclude[i]) {
2585 CWARN("Excluding %s (on exclusion list)\n", svname);
2592 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
2593 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
2595 char *s1 = ptr, *s2;
2596 __u32 index, *exclude_list;
2600 /* The shortest an ost name can be is 8 chars: -OST0000.
2601 We don't actually know the fsname at this time, so in fact
2602 a user could specify any fsname. */
2603 devmax = strlen(ptr) / 8 + 1;
2605 /* temp storage until we figure out how many we have */
2606 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
2610 /* we enter this fn pointing at the '=' */
2611 while (*s1 && *s1 != ' ' && *s1 != ',') {
2613 rc = server_name2index(s1, &index, &s2);
2615 CERROR("Can't parse server name '%s'\n", s1);
2618 if (rc == LDD_F_SV_TYPE_OST)
2619 exclude_list[lmd->lmd_exclude_count++] = index;
2621 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
2623 /* now we are pointing at ':' (next exclude)
2624 or ',' (end of excludes) */
2625 if (lmd->lmd_exclude_count >= devmax)
2628 if (rc >= 0) /* non-err */
2631 if (lmd->lmd_exclude_count) {
2632 /* permanent, freed in lustre_free_lsi */
2633 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
2634 lmd->lmd_exclude_count);
2635 if (lmd->lmd_exclude) {
2636 memcpy(lmd->lmd_exclude, exclude_list,
2637 sizeof(index) * lmd->lmd_exclude_count);
2640 lmd->lmd_exclude_count = 0;
2643 OBD_FREE(exclude_list, sizeof(index) * devmax);
2647 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
2652 if (lmd->lmd_mgssec != NULL) {
2653 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
2654 lmd->lmd_mgssec = NULL;
2657 tail = strchr(ptr, ',');
2659 length = strlen(ptr);
2661 length = tail - ptr;
2663 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
2664 if (lmd->lmd_mgssec == NULL)
2667 memcpy(lmd->lmd_mgssec, ptr, length);
2668 lmd->lmd_mgssec[length] = '\0';
2672 static int lmd_parse_string(char **handle, char *ptr)
2677 if ((handle == NULL) || (ptr == NULL))
2680 if (*handle != NULL) {
2681 OBD_FREE(*handle, strlen(*handle) + 1);
2685 tail = strchr(ptr, ',');
2687 length = strlen(ptr);
2689 length = tail - ptr;
2691 OBD_ALLOC(*handle, length + 1);
2692 if (*handle == NULL)
2695 memcpy(*handle, ptr, length);
2696 (*handle)[length] = '\0';
2701 /* Collect multiple values for mgsnid specifiers */
2702 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
2710 /* Find end of nidlist */
2711 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
2712 length = tail - *ptr;
2714 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
2718 if (lmd->lmd_mgs != NULL)
2719 oldlen = strlen(lmd->lmd_mgs) + 1;
2721 OBD_ALLOC(mgsnid, oldlen + length + 1);
2725 if (lmd->lmd_mgs != NULL) {
2726 /* Multiple mgsnid= are taken to mean failover locations */
2727 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
2728 mgsnid[oldlen - 1] = ':';
2729 OBD_FREE(lmd->lmd_mgs, oldlen);
2731 memcpy(mgsnid + oldlen, *ptr, length);
2732 mgsnid[oldlen + length] = '\0';
2733 lmd->lmd_mgs = mgsnid;
2739 /** Parse mount line options
2740 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
2741 * dev is passed as device=uml1:/lustre by mount.lustre
2743 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
2745 char *s1, *s2, *devname = NULL;
2746 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
2752 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
2753 "/sbin/mount.lustre is installed.\n");
2757 /* Options should be a string - try to detect old lmd data */
2758 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
2759 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
2760 "/sbin/mount.lustre. Please install "
2761 "version %s\n", LUSTRE_VERSION_STRING);
2764 lmd->lmd_magic = LMD_MAGIC;
2766 OBD_ALLOC(lmd->lmd_params, 4096);
2767 if (lmd->lmd_params == NULL)
2769 lmd->lmd_params[0] = '\0';
2771 /* Set default flags here */
2776 int time_min = OBD_RECOVERY_TIME_MIN;
2778 /* Skip whitespace and extra commas */
2779 while (*s1 == ' ' || *s1 == ',')
2782 /* Client options are parsed in ll_options: eg. flock,
2785 /* Parse non-ldiskfs options here. Rather than modifying
2786 ldiskfs, we just zero these out here */
2787 if (strncmp(s1, "abort_recov", 11) == 0) {
2788 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
2790 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
2791 lmd->lmd_recovery_time_soft = max_t(int,
2792 simple_strtoul(s1 + 19, NULL, 10), time_min);
2794 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
2795 lmd->lmd_recovery_time_hard = max_t(int,
2796 simple_strtoul(s1 + 19, NULL, 10), time_min);
2798 } else if (strncmp(s1, "noir", 4) == 0) {
2799 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
2801 } else if (strncmp(s1, "nosvc", 5) == 0) {
2802 lmd->lmd_flags |= LMD_FLG_NOSVC;
2804 } else if (strncmp(s1, "nomgs", 5) == 0) {
2805 lmd->lmd_flags |= LMD_FLG_NOMGS;
2807 } else if (strncmp(s1, "noscrub", 7) == 0) {
2808 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
2810 } else if (strncmp(s1, PARAM_MGSNODE,
2811 sizeof(PARAM_MGSNODE) - 1) == 0) {
2812 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
2813 /* Assume the next mount opt is the first
2814 invalid nid we get to. */
2815 rc = lmd_parse_mgs(lmd, &s2);
2819 } else if (strncmp(s1, "writeconf", 9) == 0) {
2820 lmd->lmd_flags |= LMD_FLG_WRITECONF;
2822 } else if (strncmp(s1, "virgin", 6) == 0) {
2823 lmd->lmd_flags |= LMD_FLG_VIRGIN;
2825 } else if (strncmp(s1, "noprimnode", 10) == 0) {
2826 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
2828 } else if (strncmp(s1, "mgssec=", 7) == 0) {
2829 rc = lmd_parse_mgssec(lmd, s1 + 7);
2833 /* ost exclusion list */
2834 } else if (strncmp(s1, "exclude=", 8) == 0) {
2835 rc = lmd_make_exclusion(lmd, s1 + 7);
2839 } else if (strncmp(s1, "mgs", 3) == 0) {
2841 lmd->lmd_flags |= LMD_FLG_MGS;
2843 } else if (strncmp(s1, "svname=", 7) == 0) {
2844 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
2848 } else if (strncmp(s1, "param=", 6) == 0) {
2850 char *tail = strchr(s1 + 6, ',');
2852 length = strlen(s1);
2856 strncat(lmd->lmd_params, s1 + 6, length);
2857 strcat(lmd->lmd_params, " ");
2859 } else if (strncmp(s1, "osd=", 4) == 0) {
2860 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
2865 /* Linux 2.4 doesn't pass the device, so we stuck it at the
2866 end of the options. */
2867 else if (strncmp(s1, "device=", 7) == 0) {
2869 /* terminate options right before device. device
2870 must be the last one. */
2876 s2 = strchr(s1, ',');
2884 memmove(s1, s2, strlen(s2) + 1);
2890 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
2891 "(need mount option 'device=...')\n");
2895 s1 = strstr(devname, ":/");
2898 lmd->lmd_flags |= LMD_FLG_CLIENT;
2899 /* Remove leading /s from fsname */
2900 while (*++s1 == '/') ;
2901 /* Freed in lustre_free_lsi */
2902 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
2903 if (!lmd->lmd_profile)
2905 sprintf(lmd->lmd_profile, "%s-client", s1);
2908 /* Freed in lustre_free_lsi */
2909 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
2912 strcpy(lmd->lmd_dev, devname);
2914 /* Save mount options */
2915 s1 = options + strlen(options) - 1;
2916 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
2918 if (*options != 0) {
2919 /* Freed in lustre_free_lsi */
2920 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
2923 strcpy(lmd->lmd_opts, options);
2927 lmd->lmd_magic = LMD_MAGIC;
2932 CERROR("Bad mount options %s\n", options);
2936 struct lustre_mount_data2 {
2938 struct vfsmount *lmd2_mnt;
2941 /** This is the entry point for the mount call into Lustre.
2942 * This is called when a server or client is mounted,
2943 * and this is where we start setting things up.
2944 * @param data Mount options (e.g. -o flock,abort_recov)
2946 int lustre_fill_super(struct super_block *sb, void *data, int silent)
2948 struct lustre_mount_data *lmd;
2949 struct lustre_mount_data2 *lmd2 = data;
2950 struct lustre_sb_info *lsi;
2954 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
2956 lsi = lustre_init_lsi(sb);
2962 * Disable lockdep during mount, because mount locking patterns are
2968 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
2970 obd_zombie_barrier();
2972 /* Figure out the lmd from the mount options */
2973 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
2975 GOTO(out, rc = -EINVAL);
2978 if (lmd_is_client(lmd)) {
2979 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
2980 if (!client_fill_super) {
2981 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
2982 "client mount! Is the 'lustre' "
2983 "module loaded?\n");
2987 rc = lustre_start_mgc(sb);
2992 /* Connect and start */
2993 /* (should always be ll_fill_super) */
2994 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
2995 /* c_f_s will call lustre_common_put_super on failure */
2998 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
2999 rc = server_fill_super(sb);
3000 /* s_f_s calls lustre_start_mgc after the mount because we need
3001 the MGS nids which are stored on disk. Plus, we may
3002 need to start the MGS first. */
3003 /* s_f_s will call server_put_super on failure */
3006 /* If error happens in fill_super() call, @lsi will be killed there.
3007 * This is why we do not put it here. */
3011 CERROR("Unable to mount %s (%d)\n",
3012 s2lsi(sb) ? lmd->lmd_dev : "", rc);
3014 CDEBUG(D_SUPER, "Mount %s complete\n",
3022 /* We can't call ll_fill_super by name because it lives in a module that
3023 must be loaded after this one. */
3024 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
3025 struct vfsmount *mnt))
3027 client_fill_super = cfs;
3029 EXPORT_SYMBOL(lustre_register_client_fill_super);
3031 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
3033 kill_super_cb = cfs;
3035 EXPORT_SYMBOL(lustre_register_kill_super_cb);
3037 /***************** FS registration ******************/
3038 #ifdef HAVE_FSTYPE_MOUNT
3039 struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
3040 const char *devname, void *data)
3042 struct lustre_mount_data2 lmd2 = { data, NULL };
3044 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
3047 int lustre_get_sb(struct file_system_type *fs_type, int flags,
3048 const char *devname, void * data, struct vfsmount *mnt)
3050 struct lustre_mount_data2 lmd2 = { data, mnt };
3052 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
3056 void lustre_kill_super(struct super_block *sb)
3058 struct lustre_sb_info *lsi = s2lsi(sb);
3060 if (kill_super_cb && lsi && !IS_SERVER(lsi))
3061 (*kill_super_cb)(sb);
3063 kill_anon_super(sb);
3066 /** Register the "lustre" fs type
3068 struct file_system_type lustre_fs_type = {
3069 .owner = THIS_MODULE,
3071 #ifdef HAVE_FSTYPE_MOUNT
3072 .mount = lustre_mount,
3074 .get_sb = lustre_get_sb,
3076 .kill_sb = lustre_kill_super,
3077 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
3078 FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE,
3081 int lustre_register_fs(void)
3083 return register_filesystem(&lustre_fs_type);
3086 int lustre_unregister_fs(void)
3088 return unregister_filesystem(&lustre_fs_type);
3091 EXPORT_SYMBOL(server_mti_print);