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 a lwp name
907 * (fsname-MDT0000-lwp-OSTxxxx)
909 int tgt_name2lwpname(char *svname, char *lwpname)
915 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
919 rc = server_name2fsname(svname, fsname, &tgt);
921 CERROR("%s: failed to get fsname from svname. %d\n",
926 if (*tgt != '-' && *tgt != ':') {
927 CERROR("%s: invalid svname name!\n", svname);
928 GOTO(cleanup, rc = -EINVAL);
932 if (strncmp(tgt, "OST", 3) != 0 && strncmp(tgt, "MDT", 3) != 0) {
933 CERROR("%s is not an OST or MDT target!\n", svname);
934 GOTO(cleanup, rc = -EINVAL);
936 sprintf(lwpname, "%s-MDT0000-%s-%s", fsname, LUSTRE_LWP_NAME, tgt);
939 OBD_FREE(fsname, MTI_NAME_MAXLEN);
942 EXPORT_SYMBOL(tgt_name2lwpname);
944 static CFS_LIST_HEAD(lwp_register_list);
945 DEFINE_MUTEX(lwp_register_list_lock);
947 int lustre_register_lwp_item(char *lwpname, struct obd_export **exp,
948 register_lwp_cb cb_func, void *cb_data)
950 struct obd_device *lwp;
951 struct lwp_register_item *lri;
954 LASSERTF(strlen(lwpname) < MTI_NAME_MAXLEN, "lwpname is too long %s\n",
956 LASSERT(exp != NULL && *exp == NULL);
962 mutex_lock(&lwp_register_list_lock);
964 lwp = class_name2obd(lwpname);
965 if (lwp != NULL && lwp->obd_set_up == 1) {
966 struct obd_uuid *uuid;
970 mutex_unlock(&lwp_register_list_lock);
974 memcpy(uuid->uuid, lwpname, strlen(lwpname));
975 *exp = cfs_hash_lookup(lwp->obd_uuid_hash, uuid);
979 memcpy(lri->lri_name, lwpname, strlen(lwpname));
981 lri->lri_cb_func = cb_func;
982 lri->lri_cb_data = cb_data;
983 CFS_INIT_LIST_HEAD(&lri->lri_list);
984 cfs_list_add(&lri->lri_list, &lwp_register_list);
986 if (*exp != NULL && cb_func != NULL)
989 mutex_unlock(&lwp_register_list_lock);
992 EXPORT_SYMBOL(lustre_register_lwp_item);
994 void lustre_deregister_lwp_item(struct obd_export **exp)
996 struct lwp_register_item *lri, *tmp;
998 mutex_lock(&lwp_register_list_lock);
999 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
1000 if (exp == lri->lri_exp) {
1002 class_export_put(*exp);
1003 cfs_list_del(&lri->lri_list);
1008 mutex_unlock(&lwp_register_list_lock);
1010 EXPORT_SYMBOL(lustre_deregister_lwp_item);
1012 static void lustre_notify_lwp_list(struct obd_export *exp)
1014 struct lwp_register_item *lri, *tmp;
1015 LASSERT(exp != NULL);
1017 mutex_lock(&lwp_register_list_lock);
1018 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
1019 if (strcmp(exp->exp_obd->obd_name, lri->lri_name))
1021 if (*lri->lri_exp != NULL)
1023 *lri->lri_exp = class_export_get(exp);
1024 if (lri->lri_cb_func != NULL)
1025 lri->lri_cb_func(lri->lri_cb_data);
1027 mutex_unlock(&lwp_register_list_lock);
1030 static int lustre_lwp_connect(struct obd_device *lwp)
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, lwp->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_connect_flags |= OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
1057 OBD_CONNECT_AT | OBD_CONNECT_LRU_RESIZE |
1058 OBD_CONNECT_FULL20 | OBD_CONNECT_LVB_TYPE |
1059 OBD_CONNECT_LIGHTWEIGHT;
1060 OBD_ALLOC_PTR(uuid);
1062 GOTO(out, rc = -ENOMEM);
1064 if (strlen(lwp->obd_name) > sizeof(uuid->uuid)) {
1065 CERROR("%s: Too long lwp name %s, max_size is %d\n",
1066 lwp->obd_name, lwp->obd_name, (int)sizeof(uuid->uuid));
1067 GOTO(out, rc = -EINVAL);
1070 /* Use lwp name as the uuid, so we find the export by lwp name later */
1071 memcpy(uuid->uuid, lwp->obd_name, strlen(lwp->obd_name));
1072 rc = obd_connect(&env, &exp, lwp, uuid, data, NULL);
1074 CERROR("%s: connect failed: rc = %d\n", lwp->obd_name, rc);
1076 lustre_notify_lwp_list(exp);
1085 lu_context_exit(&session_ctx);
1086 lu_context_fini(&session_ctx);
1092 * lwp is used by slaves (Non-MDT0 targets) to manage the connection
1095 static int lustre_lwp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi)
1097 struct obd_connect_data *data = NULL;
1098 struct obd_device *obd;
1099 char *lwpname = NULL;
1100 char *lwpuuid = NULL;
1104 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
1107 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
1111 OBD_ALLOC(lwpname, MTI_NAME_MAXLEN);
1112 if (lwpname == NULL)
1113 GOTO(out, rc = -ENOMEM);
1115 rc = tgt_name2lwpname(lsi->lsi_svname, lwpname);
1117 CERROR("%s: failed to generate lwp name. %d\n",
1118 lsi->lsi_svname, rc);
1122 OBD_ALLOC(lwpuuid, MTI_NAME_MAXLEN);
1123 if (lwpuuid == NULL)
1124 GOTO(out, rc = -ENOMEM);
1126 sprintf(lwpuuid, "%s_UUID", lwpname);
1127 rc = lustre_start_simple(lwpname, LUSTRE_LWP_NAME,
1128 lwpuuid, lustre_cfg_string(lcfg, 1),
1131 CERROR("%s: setup up failed: rc %d\n", lwpname, rc);
1135 obd = class_name2obd(lwpname);
1136 LASSERT(obd != NULL);
1138 rc = lustre_lwp_connect(obd);
1140 CERROR("%s: connect failed: rc = %d\n", lwpname, rc);
1144 if (lwpname != NULL)
1145 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
1146 if (lwpuuid != NULL)
1147 OBD_FREE(lwpuuid, MTI_NAME_MAXLEN);
1152 /* the caller is responsible for memory free */
1153 static struct obd_device *lustre_find_lwp(struct lustre_sb_info *lsi,
1154 char **lwpname, char **logname)
1156 struct obd_device *lwp;
1160 LASSERT(lwpname != NULL);
1161 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
1163 OBD_ALLOC(*lwpname, MTI_NAME_MAXLEN);
1164 if (*lwpname == NULL)
1165 RETURN(ERR_PTR(-ENOMEM));
1167 if (logname != NULL) {
1168 OBD_ALLOC(*logname, MTI_NAME_MAXLEN);
1169 if (*logname == NULL)
1170 GOTO(out, rc = -ENOMEM);
1171 rc = server_name2fsname(lsi->lsi_svname, *lwpname, NULL);
1173 CERROR("%s: failed to get fsname from svname. %d\n",
1174 lsi->lsi_svname, rc);
1175 GOTO(out, rc = -EINVAL);
1177 sprintf(*logname, "%s-client", *lwpname);
1180 rc = tgt_name2lwpname(lsi->lsi_svname, *lwpname);
1182 CERROR("%s: failed to generate lwp name. %d\n",
1183 lsi->lsi_svname, rc);
1184 GOTO(out, rc = -EINVAL);
1187 lwp = class_name2obd(*lwpname);
1191 if (*lwpname != NULL) {
1192 OBD_FREE(*lwpname, MTI_NAME_MAXLEN);
1195 if (logname != NULL && *logname != NULL) {
1196 OBD_FREE(*logname, MTI_NAME_MAXLEN);
1202 RETURN(lwp != NULL ? lwp : ERR_PTR(-ENOENT));
1205 static int lustre_lwp_add_conn(struct lustre_cfg *cfg,
1206 struct lustre_sb_info *lsi)
1208 struct lustre_cfg_bufs *bufs = NULL;
1209 struct lustre_cfg *lcfg = NULL;
1210 char *lwpname = NULL;
1211 struct obd_device *lwp;
1215 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
1217 CERROR("%s: can't find lwp device.\n", lsi->lsi_svname);
1218 GOTO(out, rc = PTR_ERR(lwp));
1220 LASSERT(lwpname != NULL);
1222 OBD_ALLOC_PTR(bufs);
1224 GOTO(out, rc = -ENOMEM);
1226 lustre_cfg_bufs_reset(bufs, lwpname);
1227 lustre_cfg_bufs_set_string(bufs, 1,
1228 lustre_cfg_string(cfg, 1));
1230 lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
1232 rc = class_add_conn(lwp, lcfg);
1234 CERROR("%s: can't add conn: rc = %d\n", lwpname, rc);
1240 lustre_cfg_free(lcfg);
1241 if (lwpname != NULL)
1242 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
1247 * Retrieve MDT nids from the client log, then start the lwp device.
1248 * there are only two scenarios which would include mdt nid.
1250 * marker 5 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add mdc' xxx-
1251 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
1252 * attach 0:lustre-MDT0000-mdc 1:mdc 2:lustre-clilmv_UUID
1253 * setup 0:lustre-MDT0000-mdc 1:lustre-MDT0000_UUID 2:192.168.122.162@tcp
1254 * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
1255 * add_conn 0:lustre-MDT0000-mdc 1:192.168.172.1@tcp
1256 * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID xxxx
1257 * marker 5 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add mdc' xxxx-
1259 * marker 7 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
1260 * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
1261 * add_conn 0:lustre-MDT0000-mdc 1:192.168.122.2@tcp
1262 * marker 7 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
1264 static int client_lwp_config_process(const struct lu_env *env,
1265 struct llog_handle *handle,
1266 struct llog_rec_hdr *rec, void *data)
1268 struct config_llog_instance *clli = data;
1269 int cfg_len = rec->lrh_len;
1270 char *cfg_buf = (char *) (rec + 1);
1271 struct lustre_cfg *lcfg = NULL;
1272 struct lustre_sb_info *lsi;
1273 int rc = 0, swab = 0;
1276 if (rec->lrh_type != OBD_CFG_REC) {
1277 CERROR("Unknown llog record type %#x encountered\n",
1282 LASSERT(clli->cfg_sb != NULL);
1283 lsi = s2lsi(clli->cfg_sb);
1285 lcfg = (struct lustre_cfg *)cfg_buf;
1286 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
1287 lustre_swab_lustre_cfg(lcfg);
1291 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
1295 switch (lcfg->lcfg_command) {
1297 struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
1299 lustre_swab_cfg_marker(marker, swab,
1300 LUSTRE_CFG_BUFLEN(lcfg, 1));
1301 if (marker->cm_flags & CM_SKIP ||
1302 marker->cm_flags & CM_EXCLUDE)
1305 if (!tgt_is_mdt0(marker->cm_tgtname))
1308 if(!strncmp(marker->cm_comment, "add mdc", 7) ||
1309 !strncmp(marker->cm_comment, "add failnid", 11)) {
1310 if (marker->cm_flags & CM_START) {
1311 clli->cfg_flags = CFG_F_MARKER;
1312 /* This hack is to differentiate the
1313 * ADD_UUID is come from "add mdc" record
1314 * or from "add failnid" record. */
1315 if (!strncmp(marker->cm_comment,
1317 clli->cfg_flags |= CFG_F_SKIP;
1318 } else if (marker->cm_flags & CM_END) {
1319 clli->cfg_flags = 0;
1324 case LCFG_ADD_UUID: {
1325 if (clli->cfg_flags == CFG_F_MARKER) {
1326 rc = lustre_lwp_setup(lcfg, lsi);
1327 /* XXX: process only the first nid as
1328 * we don't need another instance of lwp */
1329 clli->cfg_flags |= CFG_F_SKIP;
1330 } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
1331 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
1334 CERROR("%s: Fail to add uuid, rc:%d\n",
1335 lsi->lsi_svname, rc);
1339 case LCFG_ADD_CONN: {
1340 if (is_mdc_for_mdt0(lustre_cfg_string(lcfg, 0)))
1341 rc = lustre_lwp_add_conn(lcfg, lsi);
1351 static int lustre_disconnect_lwp(struct super_block *sb)
1353 struct lustre_sb_info *lsi = s2lsi(sb);
1354 struct obd_device *lwp;
1355 char *lwpname = NULL;
1356 char *logname = NULL;
1357 struct lustre_cfg *lcfg = NULL;
1358 struct lustre_cfg_bufs *bufs = NULL;
1359 struct config_llog_instance *cfg = NULL;
1363 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
1364 if (IS_ERR(lwp) && PTR_ERR(lwp) != -ENOENT)
1365 GOTO(out, rc = PTR_ERR(lwp));
1367 LASSERT(lwpname != NULL);
1368 LASSERT(logname != NULL);
1372 GOTO(out, rc = -ENOMEM);
1375 cfg->cfg_instance = sb;
1376 rc = lustre_end_log(sb, logname, cfg);
1378 CERROR("%s: Can't end config log %s.\n", lwpname, logname);
1382 if (PTR_ERR(lwp) == -ENOENT) {
1383 CDEBUG(D_CONFIG, "%s: lwp device wasn't started.\n",
1388 OBD_ALLOC_PTR(bufs);
1390 GOTO(out, rc = -ENOMEM);
1392 lustre_cfg_bufs_reset(bufs, lwp->obd_name);
1393 lustre_cfg_bufs_set_string(bufs, 1, NULL);
1394 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
1396 GOTO(out, rc = -ENOMEM);
1398 /* Disconnect import first. NULL is passed for the '@env', since
1399 * it will not be used. */
1400 rc = lwp->obd_lu_dev->ld_ops->ldo_process_config(NULL, lwp->obd_lu_dev,
1404 lustre_cfg_free(lcfg);
1410 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
1412 OBD_FREE(logname, MTI_NAME_MAXLEN);
1417 * Stop the lwp for an OST/MDT target.
1419 static int lustre_stop_lwp(struct super_block *sb)
1421 struct lustre_sb_info *lsi = s2lsi(sb);
1422 struct obd_device *lwp = NULL;
1423 char *lwpname = NULL;
1427 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
1429 CDEBUG(PTR_ERR(lwp) == -ENOENT ? D_CONFIG : D_ERROR,
1430 "%s: lwp wasn't started.\n", lsi->lsi_svname);
1435 rc = class_manual_cleanup(lwp);
1438 if (lwpname != NULL)
1439 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
1444 * Start the lwp(fsname-MDT0000-lwp-OSTxxxx) for an OST or MDT target,
1445 * which would be used to establish connection from OST to MDT0.
1447 static int lustre_start_lwp(struct super_block *sb)
1449 struct lustre_sb_info *lsi = s2lsi(sb);
1450 struct config_llog_instance *cfg = NULL;
1451 struct obd_device *lwp;
1452 char *lwpname = NULL;
1453 char *logname = NULL;
1457 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
1459 /* the lwp device already stared */
1460 if (lwp && !IS_ERR(lwp))
1463 if (PTR_ERR(lwp) != -ENOENT)
1464 GOTO(out, rc = PTR_ERR(lwp));
1466 LASSERT(lwpname != NULL);
1467 LASSERT(logname != NULL);
1471 GOTO(out, rc = -ENOMEM);
1473 cfg->cfg_callback = client_lwp_config_process;
1474 cfg->cfg_instance = sb;
1476 rc = lustre_process_log(sb, logname, cfg);
1478 if (lwpname != NULL)
1479 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
1480 if (logname != NULL)
1481 OBD_FREE(logname, MTI_NAME_MAXLEN);
1487 DEFINE_MUTEX(server_start_lock);
1489 /* Stop MDS/OSS if nobody is using them */
1490 static int server_stop_servers(int lsiflags)
1492 struct obd_device *obd = NULL;
1493 struct obd_type *type = NULL;
1497 mutex_lock(&server_start_lock);
1499 /* Either an MDT or an OST or neither */
1500 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
1501 if ((lsiflags & LDD_F_SV_TYPE_MDT) &&
1502 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
1503 type = class_search_type(LUSTRE_MDT_NAME);
1505 /* if this was an OST, and there are no more OST's, clean up the OSS */
1506 if ((lsiflags & LDD_F_SV_TYPE_OST) &&
1507 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
1508 type = class_search_type(LUSTRE_OST_NAME);
1511 if (obd && (!type || !type->typ_refcnt)) {
1514 /* obd_fail doesn't mean much on a server obd */
1515 err = class_manual_cleanup(obd);
1520 mutex_unlock(&server_start_lock);
1525 int server_mti_print(char *title, struct mgs_target_info *mti)
1527 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
1528 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
1529 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
1530 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
1531 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
1532 mti->mti_config_ver, mti->mti_flags);
1537 * Get service name (svname) from string
1539 * if endptr isn't NULL it is set to end of fsname *
1541 int server_name2svname(char *label, char *svname, char **endptr)
1546 /* We use server_name2fsname() just for parsing */
1547 rc = server_name2fsname(label, NULL, &dash);
1554 strncpy(svname, dash + 1, MTI_NAME_MAXLEN);
1558 EXPORT_SYMBOL(server_name2svname);
1561 /* Get the index from the obd name.
1562 rc = server type, or
1564 if endptr isn't NULL it is set to end of name */
1565 int server_name2index(char *svname, __u32 *idx, char **endptr)
1567 unsigned long index;
1571 /* We use server_name2fsname() just for parsing */
1572 rc = server_name2fsname(svname, NULL, &dash);
1581 if (strncmp(dash, "MDT", 3) == 0)
1582 rc = LDD_F_SV_TYPE_MDT;
1583 else if (strncmp(dash, "OST", 3) == 0)
1584 rc = LDD_F_SV_TYPE_OST;
1590 if (strcmp(dash, "all") == 0)
1591 return rc | LDD_F_SV_ALL;
1593 index = simple_strtoul(dash, endptr, 16);
1598 EXPORT_SYMBOL(server_name2index);
1600 /* Generate data for registration */
1601 static int server_lsi2mti(struct lustre_sb_info *lsi,
1602 struct mgs_target_info *mti)
1604 lnet_process_id_t id;
1608 if (!IS_SERVER(lsi))
1611 strncpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname));
1613 mti->mti_nid_count = 0;
1614 while (LNetGetId(i++, &id) != -ENOENT) {
1615 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1618 /* server use --servicenode param, only allow specified
1619 * nids be registered */
1620 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
1621 class_match_nid(lsi->lsi_lmd->lmd_params,
1622 PARAM_FAILNODE, id.nid) < 1)
1625 /* match specified network */
1626 if (!class_match_net(lsi->lsi_lmd->lmd_params,
1627 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1630 mti->mti_nids[mti->mti_nid_count] = id.nid;
1631 mti->mti_nid_count++;
1632 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1633 CWARN("Only using first %d nids for %s\n",
1634 mti->mti_nid_count, mti->mti_svname);
1639 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1640 mti->mti_config_ver = 0;
1642 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1646 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1649 /* Orion requires index to be set */
1650 LASSERT(!(rc & LDD_F_NEED_INDEX));
1651 /* keep only LDD flags */
1652 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1653 if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN))
1654 mti->mti_flags |= LDD_F_UPDATE;
1655 strncpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1656 sizeof(mti->mti_params));
1660 /* Register an old or new target with the MGS. If needed MGS will construct
1661 startup logs and assign index */
1662 static int server_register_target(struct lustre_sb_info *lsi)
1664 struct obd_device *mgc = lsi->lsi_mgc;
1665 struct mgs_target_info *mti = NULL;
1672 if (!IS_SERVER(lsi))
1679 rc = server_lsi2mti(lsi, mti);
1683 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1684 mti->mti_svname, mti->mti_fsname,
1685 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1688 /* if write_conf is true, the registration must succeed */
1689 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1690 mti->mti_flags |= LDD_F_OPC_REG;
1692 /* Register the target */
1693 /* FIXME use mgc_process_config instead */
1694 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1695 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
1696 sizeof(*mti), mti, NULL);
1698 if (mti->mti_flags & LDD_F_ERROR) {
1699 LCONSOLE_ERROR_MSG(0x160,
1700 "The MGS is refusing to allow this "
1701 "server (%s) to start. Please see messages"
1702 " on the MGS node.\n", lsi->lsi_svname);
1703 } else if (writeconf) {
1704 LCONSOLE_ERROR_MSG(0x15f,
1705 "Communication to the MGS return error %d. "
1706 "Is the MGS running?\n", rc);
1708 CERROR("Cannot talk to the MGS: %d, not fatal\n", rc);
1709 /* reset the error code for non-fatal error. */
1722 * Notify the MGS that this target is ready.
1723 * Used by IR - if the MGS receives this message, it will notify clients.
1725 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1727 struct lustre_sb_info *lsi = s2lsi(sb);
1728 struct obd_device *mgc = lsi->lsi_mgc;
1729 struct mgs_target_info *mti = NULL;
1735 if (!(IS_SERVER(lsi)))
1741 rc = server_lsi2mti(lsi, mti);
1745 mti->mti_instance = obd->u.obt.obt_instance;
1746 mti->mti_flags |= LDD_F_OPC_READY;
1748 /* FIXME use mgc_process_config instead */
1749 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1750 sizeof(KEY_REGISTER_TARGET),
1751 KEY_REGISTER_TARGET,
1752 sizeof(*mti), mti, NULL);
1754 /* Imperative recovery: if the mgs informs us to use IR? */
1755 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1756 (mti->mti_flags & LDD_F_IR_CAPABLE))
1757 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1766 /** Start server targets: MDTs and OSTs
1768 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1770 struct obd_device *obd;
1771 struct lustre_sb_info *lsi = s2lsi(sb);
1772 struct config_llog_instance cfg;
1774 struct lu_device *dev;
1778 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1781 /* make sure the MDS is started */
1782 mutex_lock(&server_start_lock);
1783 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1785 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1787 LUSTRE_MDS_OBDNAME"_uuid",
1790 mutex_unlock(&server_start_lock);
1791 CERROR("failed to start MDS: %d\n", rc);
1795 mutex_unlock(&server_start_lock);
1798 /* If we're an OST, make sure the global OSS is running */
1800 /* make sure OSS is started */
1801 mutex_lock(&server_start_lock);
1802 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1804 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1806 LUSTRE_OSS_OBDNAME"_uuid",
1809 mutex_unlock(&server_start_lock);
1810 CERROR("failed to start OSS: %d\n", rc);
1814 mutex_unlock(&server_start_lock);
1817 /* Set the mgc fs to our server disk. This allows the MGC to
1818 * read and write configs locally, in case it can't talk to the MGS. */
1819 if (lsi->lsi_srv_mnt) {
1820 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1822 GOTO(out_stop_service, rc);
1825 /* Register with MGS */
1826 rc = server_register_target(lsi);
1830 /* Let the target look up the mount using the target's name
1831 (we can't pass the sb or mnt through class_process_config.) */
1832 rc = server_register_mount(lsi->lsi_svname, sb, mnt);
1836 /* Start targets using the llog named for the target */
1837 memset(&cfg, 0, sizeof(cfg));
1838 cfg.cfg_callback = class_config_llog_handler;
1839 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1841 CERROR("failed to start server %s: %d\n",
1842 lsi->lsi_svname, rc);
1843 /* Do NOT call server_deregister_mount() here. This makes it
1844 * impossible to find mount later in cleanup time and leaves
1845 * @lsi and othder stuff leaked. -umka */
1849 obd = class_name2obd(lsi->lsi_svname);
1851 CERROR("no server named %s was started\n", lsi->lsi_svname);
1852 GOTO(out_mgc, rc = -ENXIO);
1855 if (IS_OST(lsi) || IS_MDT(lsi)) {
1856 rc = lustre_start_lwp(sb);
1858 CERROR("%s: failed to start LWP: %d\n",
1859 lsi->lsi_svname, rc);
1864 server_notify_target(sb, obd);
1866 /* calculate recovery timeout, do it after lustre_process_log */
1867 server_calc_timeout(lsi, obd);
1869 /* log has been fully processed */
1870 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1872 /* log has been fully processed, let clients connect */
1873 dev = obd->obd_lu_dev;
1874 if (dev && dev->ld_ops->ldo_prepare) {
1875 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1877 struct lu_context session_ctx;
1879 lu_context_init(&session_ctx, LCT_SESSION);
1880 session_ctx.lc_thread = NULL;
1881 lu_context_enter(&session_ctx);
1882 env.le_ses = &session_ctx;
1884 dev->ld_ops->ldo_prepare(&env, NULL, dev);
1887 lu_context_exit(&session_ctx);
1888 lu_context_fini(&session_ctx);
1892 /* abort recovery only on the complete stack:
1893 * many devices can be involved */
1894 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1895 (OBP(obd, iocontrol))) {
1896 obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
1901 /* Release the mgc fs for others to use */
1902 if (lsi->lsi_srv_mnt)
1903 server_mgc_clear_fs(lsi->lsi_mgc);
1907 server_stop_servers(lsi->lsi_flags);
1912 /***************** lustre superblock **************/
1914 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1916 struct lustre_sb_info *lsi;
1922 OBD_ALLOC_PTR(lsi->lsi_lmd);
1923 if (!lsi->lsi_lmd) {
1928 lsi->lsi_lmd->lmd_exclude_count = 0;
1929 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
1930 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
1931 s2lsi_nocast(sb) = lsi;
1932 /* we take 1 extra ref for our setup */
1933 cfs_atomic_set(&lsi->lsi_mounts, 1);
1935 /* Default umount style */
1936 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1941 static int lustre_free_lsi(struct super_block *sb)
1943 struct lustre_sb_info *lsi = s2lsi(sb);
1946 LASSERT(lsi != NULL);
1947 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
1949 /* someone didn't call server_put_mount. */
1950 LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0);
1952 if (lsi->lsi_lmd != NULL) {
1953 if (lsi->lsi_lmd->lmd_dev != NULL)
1954 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1955 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1956 if (lsi->lsi_lmd->lmd_profile != NULL)
1957 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1958 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1959 if (lsi->lsi_lmd->lmd_mgssec != NULL)
1960 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
1961 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
1962 if (lsi->lsi_lmd->lmd_opts != NULL)
1963 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1964 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1965 if (lsi->lsi_lmd->lmd_exclude_count)
1966 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1967 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1968 lsi->lsi_lmd->lmd_exclude_count);
1969 if (lsi->lsi_lmd->lmd_mgs != NULL)
1970 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
1971 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
1972 if (lsi->lsi_lmd->lmd_osd_type != NULL)
1973 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
1974 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
1975 if (lsi->lsi_lmd->lmd_params != NULL)
1976 OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
1978 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1981 LASSERT(lsi->lsi_llsbi == NULL);
1982 OBD_FREE(lsi, sizeof(*lsi));
1983 s2lsi_nocast(sb) = NULL;
1988 /* The lsi has one reference for every server that is using the disk -
1989 e.g. MDT, MGS, and potentially MGC */
1990 static int lustre_put_lsi(struct super_block *sb)
1992 struct lustre_sb_info *lsi = s2lsi(sb);
1995 LASSERT(lsi != NULL);
1997 CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts));
1998 if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) {
1999 if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
2000 obd_disconnect(lsi->lsi_osd_exp);
2001 /* wait till OSD is gone */
2002 obd_zombie_barrier();
2004 lustre_free_lsi(sb);
2010 static int lsi_prepare(struct lustre_sb_info *lsi)
2017 LASSERT(lsi->lsi_lmd);
2019 /* The server name is given as a mount line option */
2020 if (lsi->lsi_lmd->lmd_profile == NULL) {
2021 LCONSOLE_ERROR("Can't determine server name\n");
2025 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
2026 RETURN(-ENAMETOOLONG);
2028 strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
2030 /* Determine osd type */
2031 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
2032 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
2033 sizeof(lsi->lsi_osd_type))
2034 RETURN(-ENAMETOOLONG);
2036 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
2038 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
2041 /* XXX: a temp. solution for components using fsfilt
2042 * to be removed in one of the subsequent patches */
2043 if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs")) {
2044 strcpy(lsi->lsi_fstype, "ldiskfs");
2046 strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
2049 /* Determine server type */
2050 rc = server_name2index(lsi->lsi_svname, &index, NULL);
2052 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
2053 /* Assume we're a bare MGS */
2055 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
2057 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
2062 lsi->lsi_flags |= rc;
2064 /* Add mount line flags that used to be in ldd:
2065 * writeconf, mgs, anything else?
2067 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
2068 LDD_F_WRITECONF : 0;
2069 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
2071 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
2072 LDD_F_SV_TYPE_MGS : 0;
2073 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
2074 LDD_F_NO_PRIMNODE : 0;
2079 /*************** server mount ******************/
2081 /** Start the shutdown of servers at umount.
2083 static void server_put_super(struct super_block *sb)
2085 struct lustre_sb_info *lsi = s2lsi(sb);
2086 struct obd_device *obd;
2087 char *tmpname, *extraname = NULL;
2089 int lsiflags = lsi->lsi_flags;
2092 LASSERT(IS_SERVER(lsi));
2094 tmpname_sz = strlen(lsi->lsi_svname) + 1;
2095 OBD_ALLOC(tmpname, tmpname_sz);
2096 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
2097 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
2098 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
2099 snprintf(tmpname, tmpname_sz, "MGS");
2101 /* disconnect the lwp first to drain off the inflight request */
2102 if (IS_OST(lsi) || IS_MDT(lsi)) {
2105 rc = lustre_disconnect_lwp(sb);
2106 if (rc && rc != ETIMEDOUT)
2107 CERROR("%s: failed to disconnect lwp. (rc=%d)\n",
2111 /* Stop the target */
2112 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
2113 (IS_MDT(lsi) || IS_OST(lsi))) {
2114 struct lustre_profile *lprof = NULL;
2116 /* tell the mgc to drop the config log */
2117 lustre_end_log(sb, lsi->lsi_svname, NULL);
2119 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
2120 If there are any setup/cleanup errors, save the lov
2121 name for safety cleanup later. */
2122 lprof = class_get_profile(lsi->lsi_svname);
2123 if (lprof && lprof->lp_dt) {
2124 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
2125 strcpy(extraname, lprof->lp_dt);
2128 obd = class_name2obd(lsi->lsi_svname);
2130 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
2131 if (lsiflags & LSI_UMOUNT_FAILOVER)
2133 /* We can't seem to give an error return code
2134 * to .put_super, so we better make sure we clean up! */
2136 class_manual_cleanup(obd);
2138 CERROR("no obd %s\n", lsi->lsi_svname);
2139 server_deregister_mount(lsi->lsi_svname);
2143 /* If they wanted the mgs to stop separately from the mdt, they
2144 should have put it on a different device. */
2146 /* if MDS start with --nomgs, don't stop MGS then */
2147 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
2148 server_stop_mgs(sb);
2151 if (IS_OST(lsi) || IS_MDT(lsi)) {
2152 if (lustre_stop_lwp(sb) < 0)
2153 CERROR("%s: failed to stop lwp!\n", tmpname);
2156 /* Clean the mgc and sb */
2157 lustre_common_put_super(sb);
2159 /* wait till all in-progress cleanups are done
2160 * specifically we're interested in ofd cleanup
2162 obd_zombie_barrier();
2164 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
2165 until the target is really gone so that our type refcount check
2167 server_stop_servers(lsiflags);
2169 /* In case of startup or cleanup err, stop related obds */
2171 obd = class_name2obd(extraname);
2173 CWARN("Cleaning orphaned obd %s\n", extraname);
2175 class_manual_cleanup(obd);
2177 OBD_FREE(extraname, strlen(extraname) + 1);
2180 LCONSOLE_WARN("server umount %s complete\n", tmpname);
2181 OBD_FREE(tmpname, tmpname_sz);
2185 /** Called only for 'umount -f'
2187 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2188 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
2190 struct super_block *sb = vfsmnt->mnt_sb;
2192 static void server_umount_begin(struct super_block *sb)
2195 struct lustre_sb_info *lsi = s2lsi(sb);
2198 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2199 if (!(flags & MNT_FORCE)) {
2205 CDEBUG(D_MOUNT, "umount -f\n");
2206 /* umount = failover
2208 no third way to do non-force, non-failover */
2209 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
2213 static int server_statfs (struct dentry *dentry, cfs_kstatfs_t *buf)
2215 struct super_block *sb = dentry->d_sb;
2216 struct lustre_sb_info *lsi = s2lsi(sb);
2217 struct obd_statfs statfs;
2221 if (lsi->lsi_dt_dev) {
2222 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
2224 statfs_unpack(buf, &statfs);
2225 buf->f_type = sb->s_magic;
2231 buf->f_type = sb->s_magic;
2232 buf->f_bsize = sb->s_blocksize;
2238 buf->f_namelen = NAME_MAX;
2242 /** The operations we support directly on the superblock:
2243 * mount, umount, and df.
2245 static struct super_operations server_ops =
2247 .put_super = server_put_super,
2248 .umount_begin = server_umount_begin, /* umount -f */
2249 .statfs = server_statfs,
2252 #define log2(n) ffz(~(n))
2253 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
2255 static int server_fill_super_common(struct super_block *sb)
2257 struct inode *root = 0;
2260 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
2262 sb->s_blocksize = 4096;
2263 sb->s_blocksize_bits = log2(sb->s_blocksize);
2264 sb->s_magic = LUSTRE_SUPER_MAGIC;
2265 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
2266 sb->s_flags |= MS_RDONLY;
2267 sb->s_op = &server_ops;
2269 root = new_inode(sb);
2271 CERROR("Can't make root inode\n");
2275 /* returns -EIO for every operation */
2276 /* make_bad_inode(root); -- badness - can't umount */
2277 /* apparently we need to be a directory for the mount to finish */
2278 root->i_mode = S_IFDIR;
2280 sb->s_root = d_make_root(root);
2282 CERROR("%s: can't make root dentry\n", sb->s_id);
2289 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
2291 struct lustre_mount_data *lmd = lsi->lsi_lmd;
2292 struct obd_device *obd;
2293 struct dt_device_param p;
2299 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
2300 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
2302 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
2303 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
2304 strcat(lsi->lsi_osd_uuid, "_UUID");
2305 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
2307 obd = class_name2obd(lsi->lsi_osd_obdname);
2309 rc = lustre_start_simple(lsi->lsi_osd_obdname,
2311 lsi->lsi_osd_uuid, lmd->lmd_dev,
2312 flagstr, lsi->lsi_lmd->lmd_opts,
2316 obd = class_name2obd(lsi->lsi_osd_obdname);
2320 rc = obd_connect(NULL, &lsi->lsi_osd_exp, obd, &obd->obd_uuid, NULL, NULL);
2323 class_manual_cleanup(obd);
2324 lsi->lsi_dt_dev = NULL;
2327 /* XXX: to keep support old components relying on lsi_srv_mnt
2328 * we get this info from OSD just started */
2329 LASSERT(obd->obd_lu_dev);
2330 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
2331 LASSERT(lsi->lsi_dt_dev);
2333 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
2335 lsi->lsi_srv_mnt = p.ddp_mnt;
2341 /** Fill in the superblock info for a Lustre server.
2342 * Mount the device with the correct options.
2343 * Read the on-disk config file.
2344 * Start the services.
2346 static int server_fill_super(struct super_block *sb)
2348 struct lustre_sb_info *lsi = s2lsi(sb);
2352 rc = lsi_prepare(lsi);
2356 /* Start low level OSD */
2357 rc = osd_start(lsi, sb->s_flags);
2359 CERROR("Unable to start osd on %s: %d\n",
2360 lsi->lsi_lmd->lmd_dev, rc);
2365 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
2366 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
2368 if (class_name2obd(lsi->lsi_svname)) {
2369 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
2370 "running. Double-mount may have compromised"
2371 " the disk journal.\n",
2377 /* Start MGS before MGC */
2378 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
2379 rc = server_start_mgs(sb);
2384 /* Start MGC before servers */
2385 rc = lustre_start_mgc(sb);
2389 /* Set up all obd devices for service */
2390 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
2391 (IS_OST(lsi) || IS_MDT(lsi))) {
2392 rc = server_start_targets(sb, lsi->lsi_srv_mnt);
2394 CERROR("Unable to start targets: %d\n", rc);
2397 /* FIXME overmount client here,
2398 or can we just start a client log and client_fill_super on this sb?
2399 We need to make sure server_put_super gets called too - ll_put_super
2400 calls lustre_common_put_super; check there for LSI_SERVER flag,
2402 Probably should start client from new thread so we can return.
2403 Client will not finish until all servers are connected.
2404 Note - MGS-only server does NOT get a client, since there is no
2405 lustre fs associated - the MGS is for all lustre fs's */
2408 rc = server_fill_super_common(sb);
2414 /* We jump here in case of failure while starting targets or MGS.
2415 * In this case we can't just put @mnt and have to do real cleanup
2416 * with stoping targets, etc. */
2417 server_put_super(sb);
2422 * Calculate timeout value for a target.
2424 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
2426 struct lustre_mount_data *lmd;
2430 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
2431 int min = OBD_RECOVERY_TIME_MIN;
2433 LASSERT(IS_SERVER(lsi));
2437 soft = lmd->lmd_recovery_time_soft;
2438 hard = lmd->lmd_recovery_time_hard;
2439 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
2440 obd->obd_no_ir = !has_ir;
2444 soft = OBD_RECOVERY_TIME_SOFT;
2446 hard = OBD_RECOVERY_TIME_HARD;
2448 /* target may have ir_factor configured. */
2449 factor = OBD_IR_FACTOR_DEFAULT;
2450 if (obd->obd_recovery_ir_factor)
2451 factor = obd->obd_recovery_ir_factor;
2454 int new_soft = soft;
2455 int new_hard = hard;
2457 /* adjust timeout value by imperative recovery */
2459 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
2460 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
2462 /* make sure the timeout is not too short */
2463 new_soft = max(min, new_soft);
2464 new_hard = max(new_soft, new_hard);
2466 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
2467 "window shrunk from %d-%d down to %d-%d\n",
2468 obd->obd_name, soft, hard, new_soft, new_hard);
2475 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
2476 obd->obd_recovery_time_hard = hard;
2477 obd->obd_recovery_ir_factor = factor;
2479 EXPORT_SYMBOL(server_calc_timeout);
2481 /*************** mount common betweeen server and client ***************/
2484 int lustre_common_put_super(struct super_block *sb)
2489 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
2491 /* Drop a ref to the MGC */
2492 rc = lustre_stop_mgc(sb);
2493 if (rc && (rc != -ENOENT)) {
2495 CERROR("Can't stop MGC: %d\n", rc);
2498 /* BUSY just means that there's some other obd that
2499 needs the mgc. Let him clean it up. */
2500 CDEBUG(D_MOUNT, "MGC still in use\n");
2502 /* Drop a ref to the mounted disk */
2507 EXPORT_SYMBOL(lustre_common_put_super);
2509 static void lmd_print(struct lustre_mount_data *lmd)
2513 PRINT_CMD(PRINT_MASK, " mount data:\n");
2514 if (lmd_is_client(lmd))
2515 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
2516 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
2517 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
2520 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
2522 if (lmd->lmd_recovery_time_soft)
2523 PRINT_CMD(PRINT_MASK, "recovery time soft: %d\n",
2524 lmd->lmd_recovery_time_soft);
2526 if (lmd->lmd_recovery_time_hard)
2527 PRINT_CMD(PRINT_MASK, "recovery time hard: %d\n",
2528 lmd->lmd_recovery_time_hard);
2530 for (i = 0; i < lmd->lmd_exclude_count; i++) {
2531 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
2532 lmd->lmd_exclude[i]);
2536 /* Is this server on the exclusion list */
2537 int lustre_check_exclusion(struct super_block *sb, char *svname)
2539 struct lustre_sb_info *lsi = s2lsi(sb);
2540 struct lustre_mount_data *lmd = lsi->lsi_lmd;
2545 rc = server_name2index(svname, &index, NULL);
2546 if (rc != LDD_F_SV_TYPE_OST)
2547 /* Only exclude OSTs */
2550 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
2551 index, lmd->lmd_exclude_count, lmd->lmd_dev);
2553 for(i = 0; i < lmd->lmd_exclude_count; i++) {
2554 if (index == lmd->lmd_exclude[i]) {
2555 CWARN("Excluding %s (on exclusion list)\n", svname);
2562 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
2563 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
2565 char *s1 = ptr, *s2;
2566 __u32 index, *exclude_list;
2570 /* The shortest an ost name can be is 8 chars: -OST0000.
2571 We don't actually know the fsname at this time, so in fact
2572 a user could specify any fsname. */
2573 devmax = strlen(ptr) / 8 + 1;
2575 /* temp storage until we figure out how many we have */
2576 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
2580 /* we enter this fn pointing at the '=' */
2581 while (*s1 && *s1 != ' ' && *s1 != ',') {
2583 rc = server_name2index(s1, &index, &s2);
2585 CERROR("Can't parse server name '%s'\n", s1);
2588 if (rc == LDD_F_SV_TYPE_OST)
2589 exclude_list[lmd->lmd_exclude_count++] = index;
2591 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
2593 /* now we are pointing at ':' (next exclude)
2594 or ',' (end of excludes) */
2595 if (lmd->lmd_exclude_count >= devmax)
2598 if (rc >= 0) /* non-err */
2601 if (lmd->lmd_exclude_count) {
2602 /* permanent, freed in lustre_free_lsi */
2603 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
2604 lmd->lmd_exclude_count);
2605 if (lmd->lmd_exclude) {
2606 memcpy(lmd->lmd_exclude, exclude_list,
2607 sizeof(index) * lmd->lmd_exclude_count);
2610 lmd->lmd_exclude_count = 0;
2613 OBD_FREE(exclude_list, sizeof(index) * devmax);
2617 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
2622 if (lmd->lmd_mgssec != NULL) {
2623 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
2624 lmd->lmd_mgssec = NULL;
2627 tail = strchr(ptr, ',');
2629 length = strlen(ptr);
2631 length = tail - ptr;
2633 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
2634 if (lmd->lmd_mgssec == NULL)
2637 memcpy(lmd->lmd_mgssec, ptr, length);
2638 lmd->lmd_mgssec[length] = '\0';
2642 static int lmd_parse_string(char **handle, char *ptr)
2647 if ((handle == NULL) || (ptr == NULL))
2650 if (*handle != NULL) {
2651 OBD_FREE(*handle, strlen(*handle) + 1);
2655 tail = strchr(ptr, ',');
2657 length = strlen(ptr);
2659 length = tail - ptr;
2661 OBD_ALLOC(*handle, length + 1);
2662 if (*handle == NULL)
2665 memcpy(*handle, ptr, length);
2666 (*handle)[length] = '\0';
2671 /* Collect multiple values for mgsnid specifiers */
2672 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
2680 /* Find end of nidlist */
2681 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
2682 length = tail - *ptr;
2684 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
2688 if (lmd->lmd_mgs != NULL)
2689 oldlen = strlen(lmd->lmd_mgs) + 1;
2691 OBD_ALLOC(mgsnid, oldlen + length + 1);
2695 if (lmd->lmd_mgs != NULL) {
2696 /* Multiple mgsnid= are taken to mean failover locations */
2697 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
2698 mgsnid[oldlen - 1] = ':';
2699 OBD_FREE(lmd->lmd_mgs, oldlen);
2701 memcpy(mgsnid + oldlen, *ptr, length);
2702 mgsnid[oldlen + length] = '\0';
2703 lmd->lmd_mgs = mgsnid;
2709 /** Parse mount line options
2710 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
2711 * dev is passed as device=uml1:/lustre by mount.lustre
2713 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
2715 char *s1, *s2, *devname = NULL;
2716 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
2722 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
2723 "/sbin/mount.lustre is installed.\n");
2727 /* Options should be a string - try to detect old lmd data */
2728 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
2729 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
2730 "/sbin/mount.lustre. Please install "
2731 "version %s\n", LUSTRE_VERSION_STRING);
2734 lmd->lmd_magic = LMD_MAGIC;
2736 OBD_ALLOC(lmd->lmd_params, 4096);
2737 if (lmd->lmd_params == NULL)
2739 lmd->lmd_params[0] = '\0';
2741 /* Set default flags here */
2746 int time_min = OBD_RECOVERY_TIME_MIN;
2748 /* Skip whitespace and extra commas */
2749 while (*s1 == ' ' || *s1 == ',')
2752 /* Client options are parsed in ll_options: eg. flock,
2755 /* Parse non-ldiskfs options here. Rather than modifying
2756 ldiskfs, we just zero these out here */
2757 if (strncmp(s1, "abort_recov", 11) == 0) {
2758 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
2760 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
2761 lmd->lmd_recovery_time_soft = max_t(int,
2762 simple_strtoul(s1 + 19, NULL, 10), time_min);
2764 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
2765 lmd->lmd_recovery_time_hard = max_t(int,
2766 simple_strtoul(s1 + 19, NULL, 10), time_min);
2768 } else if (strncmp(s1, "noir", 4) == 0) {
2769 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
2771 } else if (strncmp(s1, "nosvc", 5) == 0) {
2772 lmd->lmd_flags |= LMD_FLG_NOSVC;
2774 } else if (strncmp(s1, "nomgs", 5) == 0) {
2775 lmd->lmd_flags |= LMD_FLG_NOMGS;
2777 } else if (strncmp(s1, "noscrub", 7) == 0) {
2778 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
2780 } else if (strncmp(s1, PARAM_MGSNODE,
2781 sizeof(PARAM_MGSNODE) - 1) == 0) {
2782 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
2783 /* Assume the next mount opt is the first
2784 invalid nid we get to. */
2785 rc = lmd_parse_mgs(lmd, &s2);
2789 } else if (strncmp(s1, "writeconf", 9) == 0) {
2790 lmd->lmd_flags |= LMD_FLG_WRITECONF;
2792 } else if (strncmp(s1, "virgin", 6) == 0) {
2793 lmd->lmd_flags |= LMD_FLG_VIRGIN;
2795 } else if (strncmp(s1, "noprimnode", 10) == 0) {
2796 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
2798 } else if (strncmp(s1, "mgssec=", 7) == 0) {
2799 rc = lmd_parse_mgssec(lmd, s1 + 7);
2803 /* ost exclusion list */
2804 } else if (strncmp(s1, "exclude=", 8) == 0) {
2805 rc = lmd_make_exclusion(lmd, s1 + 7);
2809 } else if (strncmp(s1, "mgs", 3) == 0) {
2811 lmd->lmd_flags |= LMD_FLG_MGS;
2813 } else if (strncmp(s1, "svname=", 7) == 0) {
2814 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
2818 } else if (strncmp(s1, "param=", 6) == 0) {
2820 char *tail = strchr(s1 + 6, ',');
2822 length = strlen(s1);
2826 strncat(lmd->lmd_params, s1 + 6, length);
2827 strcat(lmd->lmd_params, " ");
2829 } else if (strncmp(s1, "osd=", 4) == 0) {
2830 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
2835 /* Linux 2.4 doesn't pass the device, so we stuck it at the
2836 end of the options. */
2837 else if (strncmp(s1, "device=", 7) == 0) {
2839 /* terminate options right before device. device
2840 must be the last one. */
2846 s2 = strchr(s1, ',');
2854 memmove(s1, s2, strlen(s2) + 1);
2860 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
2861 "(need mount option 'device=...')\n");
2865 s1 = strstr(devname, ":/");
2868 lmd->lmd_flags |= LMD_FLG_CLIENT;
2869 /* Remove leading /s from fsname */
2870 while (*++s1 == '/') ;
2871 /* Freed in lustre_free_lsi */
2872 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
2873 if (!lmd->lmd_profile)
2875 sprintf(lmd->lmd_profile, "%s-client", s1);
2878 /* Freed in lustre_free_lsi */
2879 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
2882 strcpy(lmd->lmd_dev, devname);
2884 /* Save mount options */
2885 s1 = options + strlen(options) - 1;
2886 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
2888 if (*options != 0) {
2889 /* Freed in lustre_free_lsi */
2890 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
2893 strcpy(lmd->lmd_opts, options);
2897 lmd->lmd_magic = LMD_MAGIC;
2902 CERROR("Bad mount options %s\n", options);
2906 struct lustre_mount_data2 {
2908 struct vfsmount *lmd2_mnt;
2911 /** This is the entry point for the mount call into Lustre.
2912 * This is called when a server or client is mounted,
2913 * and this is where we start setting things up.
2914 * @param data Mount options (e.g. -o flock,abort_recov)
2916 int lustre_fill_super(struct super_block *sb, void *data, int silent)
2918 struct lustre_mount_data *lmd;
2919 struct lustre_mount_data2 *lmd2 = data;
2920 struct lustre_sb_info *lsi;
2924 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
2926 lsi = lustre_init_lsi(sb);
2932 * Disable lockdep during mount, because mount locking patterns are
2938 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
2940 obd_zombie_barrier();
2942 /* Figure out the lmd from the mount options */
2943 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
2945 GOTO(out, rc = -EINVAL);
2948 if (lmd_is_client(lmd)) {
2949 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
2950 if (!client_fill_super) {
2951 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
2952 "client mount! Is the 'lustre' "
2953 "module loaded?\n");
2957 rc = lustre_start_mgc(sb);
2962 /* Connect and start */
2963 /* (should always be ll_fill_super) */
2964 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
2965 /* c_f_s will call lustre_common_put_super on failure */
2968 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
2969 rc = server_fill_super(sb);
2970 /* s_f_s calls lustre_start_mgc after the mount because we need
2971 the MGS nids which are stored on disk. Plus, we may
2972 need to start the MGS first. */
2973 /* s_f_s will call server_put_super on failure */
2976 /* If error happens in fill_super() call, @lsi will be killed there.
2977 * This is why we do not put it here. */
2981 CERROR("Unable to mount %s (%d)\n",
2982 s2lsi(sb) ? lmd->lmd_dev : "", rc);
2984 CDEBUG(D_SUPER, "Mount %s complete\n",
2992 /* We can't call ll_fill_super by name because it lives in a module that
2993 must be loaded after this one. */
2994 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
2995 struct vfsmount *mnt))
2997 client_fill_super = cfs;
2999 EXPORT_SYMBOL(lustre_register_client_fill_super);
3001 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
3003 kill_super_cb = cfs;
3005 EXPORT_SYMBOL(lustre_register_kill_super_cb);
3007 /***************** FS registration ******************/
3008 #ifdef HAVE_FSTYPE_MOUNT
3009 struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
3010 const char *devname, void *data)
3012 struct lustre_mount_data2 lmd2 = { data, NULL };
3014 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
3017 int lustre_get_sb(struct file_system_type *fs_type, int flags,
3018 const char *devname, void * data, struct vfsmount *mnt)
3020 struct lustre_mount_data2 lmd2 = { data, mnt };
3022 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
3026 void lustre_kill_super(struct super_block *sb)
3028 struct lustre_sb_info *lsi = s2lsi(sb);
3030 if (kill_super_cb && lsi && !IS_SERVER(lsi))
3031 (*kill_super_cb)(sb);
3033 kill_anon_super(sb);
3036 /** Register the "lustre" fs type
3038 struct file_system_type lustre_fs_type = {
3039 .owner = THIS_MODULE,
3041 #ifdef HAVE_FSTYPE_MOUNT
3042 .mount = lustre_mount,
3044 .get_sb = lustre_get_sb,
3046 .kill_sb = lustre_kill_super,
3047 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
3048 FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE,
3051 int lustre_register_fs(void)
3053 return register_filesystem(&lustre_fs_type);
3056 int lustre_unregister_fs(void)
3058 return unregister_filesystem(&lustre_fs_type);
3061 EXPORT_SYMBOL(server_mti_print);