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) 2013, 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_server.c
38 * 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 <obd_class.h>
52 #include <lustre/lustre_user.h>
53 #include <linux/version.h>
54 #include <lustre_log.h>
55 #include <lustre_disk.h>
56 #include <lustre_param.h>
57 #ifdef HAVE_KERNEL_LOCKED
58 #include <linux/smp_lock.h>
61 /*********** mount lookup *********/
63 DEFINE_MUTEX(lustre_mount_info_lock);
64 static CFS_LIST_HEAD(server_mount_info_list);
66 static struct lustre_mount_info *server_find_mount(const char *name)
69 struct lustre_mount_info *lmi;
72 cfs_list_for_each(tmp, &server_mount_info_list) {
73 lmi = cfs_list_entry(tmp, struct lustre_mount_info,
75 if (strcmp(name, lmi->lmi_name) == 0)
81 /* we must register an obd for a mount before we call the setup routine.
82 *_setup will call lustre_get_mount to get the mnt struct
83 by obd_name, since we can't pass the pointer to setup. */
84 static int server_register_mount(const char *name, struct super_block *sb,
87 struct lustre_mount_info *lmi;
93 OBD_ALLOC(lmi, sizeof(*lmi));
96 OBD_ALLOC(name_cp, strlen(name) + 1);
98 OBD_FREE(lmi, sizeof(*lmi));
101 strcpy(name_cp, name);
103 mutex_lock(&lustre_mount_info_lock);
105 if (server_find_mount(name)) {
106 mutex_unlock(&lustre_mount_info_lock);
107 OBD_FREE(lmi, sizeof(*lmi));
108 OBD_FREE(name_cp, strlen(name) + 1);
109 CERROR("Already registered %s\n", name);
112 lmi->lmi_name = name_cp;
115 cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list);
117 mutex_unlock(&lustre_mount_info_lock);
119 CDEBUG(D_MOUNT, "reg_mnt %p from %s\n", lmi->lmi_mnt, name);
124 /* when an obd no longer needs a mount */
125 static int server_deregister_mount(const char *name)
127 struct lustre_mount_info *lmi;
130 mutex_lock(&lustre_mount_info_lock);
131 lmi = server_find_mount(name);
133 mutex_unlock(&lustre_mount_info_lock);
134 CERROR("%s not registered\n", name);
138 CDEBUG(D_MOUNT, "dereg_mnt %p from %s\n", lmi->lmi_mnt, name);
140 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
141 cfs_list_del(&lmi->lmi_list_chain);
142 OBD_FREE(lmi, sizeof(*lmi));
143 mutex_unlock(&lustre_mount_info_lock);
148 /* obd's look up a registered mount using their obdname. This is just
149 for initial obd setup to find the mount struct. It should not be
150 called every time you want to mntget. */
151 struct lustre_mount_info *server_get_mount(const char *name)
153 struct lustre_mount_info *lmi;
154 struct lustre_sb_info *lsi;
157 mutex_lock(&lustre_mount_info_lock);
158 lmi = server_find_mount(name);
159 mutex_unlock(&lustre_mount_info_lock);
161 CERROR("Can't find mount for %s\n", name);
164 lsi = s2lsi(lmi->lmi_sb);
166 cfs_atomic_inc(&lsi->lsi_mounts);
168 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d\n", lmi->lmi_mnt,
169 name, cfs_atomic_read(&lsi->lsi_mounts));
173 EXPORT_SYMBOL(server_get_mount);
176 * Used by mdt to get mount_info from obdname.
177 * There are no blocking when using the mount_info.
178 * Do not use server_get_mount for this purpose.
180 struct lustre_mount_info *server_get_mount_2(const char *name)
182 struct lustre_mount_info *lmi;
185 mutex_lock(&lustre_mount_info_lock);
186 lmi = server_find_mount(name);
187 mutex_unlock(&lustre_mount_info_lock);
189 CERROR("Can't find mount for %s\n", name);
193 EXPORT_SYMBOL(server_get_mount_2);
195 /* to be called from obd_cleanup methods */
196 int server_put_mount(const char *name, struct vfsmount *mnt)
198 struct lustre_mount_info *lmi;
199 struct lustre_sb_info *lsi;
202 mutex_lock(&lustre_mount_info_lock);
203 lmi = server_find_mount(name);
204 mutex_unlock(&lustre_mount_info_lock);
206 CERROR("Can't find mount for %s\n", name);
209 lsi = s2lsi(lmi->lmi_sb);
211 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d\n",
212 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts));
214 if (lustre_put_lsi(lmi->lmi_sb))
215 CDEBUG(D_MOUNT, "Last put of mnt %p from %s\n",
218 /* this obd should never need the mount again */
219 server_deregister_mount(name);
223 EXPORT_SYMBOL(server_put_mount);
225 /* Corresponding to server_get_mount_2 */
226 int server_put_mount_2(const char *name, struct vfsmount *mnt)
231 EXPORT_SYMBOL(server_put_mount_2);
233 /* Set up a MGS to serve startup logs */
234 static int server_start_mgs(struct super_block *sb)
236 struct lustre_sb_info *lsi = s2lsi(sb);
237 struct vfsmount *mnt = lsi->lsi_srv_mnt;
238 struct lustre_mount_info *lmi;
242 /* It is impossible to have more than 1 MGS per node, since
243 MGC wouldn't know which to connect to */
244 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
246 lsi = s2lsi(lmi->lmi_sb);
247 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
252 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
254 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
257 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
258 LUSTRE_MGS_OBDNAME, 0, 0,
259 lsi->lsi_osd_obdname, 0);
260 /* server_deregister_mount() is not called previously, for lsi
261 * and other stuff can't be freed cleanly when mgs calls
262 * server_put_mount() in error handling case (see b=17758),
263 * this problem is caused by a bug in mgs_init0, which forgot
264 * calling server_put_mount in error case. */
267 server_deregister_mount(LUSTRE_MGS_OBDNAME);
271 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
272 "Is the 'mgs' module loaded?\n",
273 LUSTRE_MGS_OBDNAME, rc);
277 static int server_stop_mgs(struct super_block *sb)
279 struct obd_device *obd;
283 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
285 /* There better be only one MGS */
286 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
288 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
292 /* The MGS should always stop when we say so */
294 rc = class_manual_cleanup(obd);
298 /* Since there's only one mgc per node, we have to change it's fs to get
299 access to the right disk. */
300 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
302 struct lustre_sb_info *lsi = s2lsi(sb);
306 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
308 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
309 rc = obd_set_info_async(NULL, mgc->obd_self_export,
310 sizeof(KEY_SET_FS), KEY_SET_FS,
311 sizeof(*sb), sb, NULL);
313 CERROR("can't set_fs %d\n", rc);
318 static int server_mgc_clear_fs(struct obd_device *mgc)
323 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
325 rc = obd_set_info_async(NULL, mgc->obd_self_export,
326 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
331 static int is_mdc_device(const char *devname)
335 ptr = strrchr(devname, '-');
336 if (ptr != NULL && strcmp(ptr, "-mdc") == 0)
342 static inline int tgt_is_mdt0(const char *tgtname)
347 type = server_name2index(tgtname, &idx, NULL);
348 if (type != LDD_F_SV_TYPE_MDT)
354 static inline int is_mdc_for_mdt0(const char *devname)
358 if (!is_mdc_device(devname))
361 ptr = strrchr(devname, '-');
366 if (tgt_is_mdt0(devname)) {
375 * Convert OST/MDT name(fsname-OSTxxxx) to a lwp name
376 * (fsname-MDT0000-lwp-OSTxxxx)
378 int tgt_name2lwpname(const char *svname, char *lwpname)
385 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
389 rc = server_name2fsname(svname, fsname, &tgt);
391 CERROR("%s: failed to get fsname from svname. %d\n",
396 if (*tgt != '-' && *tgt != ':') {
397 CERROR("%s: invalid svname name!\n", svname);
398 GOTO(cleanup, rc = -EINVAL);
402 if (strncmp(tgt, "OST", 3) != 0 && strncmp(tgt, "MDT", 3) != 0) {
403 CERROR("%s is not an OST or MDT target!\n", svname);
404 GOTO(cleanup, rc = -EINVAL);
406 sprintf(lwpname, "%s-MDT0000-%s-%s", fsname, LUSTRE_LWP_NAME, tgt);
409 OBD_FREE(fsname, MTI_NAME_MAXLEN);
412 EXPORT_SYMBOL(tgt_name2lwpname);
414 static CFS_LIST_HEAD(lwp_register_list);
415 DEFINE_MUTEX(lwp_register_list_lock);
417 int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp,
418 register_lwp_cb cb_func, void *cb_data)
420 struct obd_device *lwp;
421 struct lwp_register_item *lri;
424 LASSERTF(strlen(lwpname) < MTI_NAME_MAXLEN, "lwpname is too long %s\n",
426 LASSERT(exp != NULL && *exp == NULL);
432 mutex_lock(&lwp_register_list_lock);
434 lwp = class_name2obd(lwpname);
435 if (lwp != NULL && lwp->obd_set_up == 1) {
436 struct obd_uuid *uuid;
440 mutex_unlock(&lwp_register_list_lock);
444 memcpy(uuid->uuid, lwpname, strlen(lwpname));
445 *exp = cfs_hash_lookup(lwp->obd_uuid_hash, uuid);
449 memcpy(lri->lri_name, lwpname, strlen(lwpname));
451 lri->lri_cb_func = cb_func;
452 lri->lri_cb_data = cb_data;
453 CFS_INIT_LIST_HEAD(&lri->lri_list);
454 cfs_list_add(&lri->lri_list, &lwp_register_list);
456 if (*exp != NULL && cb_func != NULL)
459 mutex_unlock(&lwp_register_list_lock);
462 EXPORT_SYMBOL(lustre_register_lwp_item);
464 void lustre_deregister_lwp_item(struct obd_export **exp)
466 struct lwp_register_item *lri, *tmp;
468 mutex_lock(&lwp_register_list_lock);
469 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
470 if (exp == lri->lri_exp) {
472 class_export_put(*exp);
473 cfs_list_del(&lri->lri_list);
478 mutex_unlock(&lwp_register_list_lock);
480 EXPORT_SYMBOL(lustre_deregister_lwp_item);
482 static void lustre_notify_lwp_list(struct obd_export *exp)
484 struct lwp_register_item *lri, *tmp;
485 LASSERT(exp != NULL);
487 mutex_lock(&lwp_register_list_lock);
488 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
489 if (strcmp(exp->exp_obd->obd_name, lri->lri_name))
491 if (*lri->lri_exp != NULL)
493 *lri->lri_exp = class_export_get(exp);
494 if (lri->lri_cb_func != NULL)
495 lri->lri_cb_func(lri->lri_cb_data);
497 mutex_unlock(&lwp_register_list_lock);
500 static int lustre_lwp_connect(struct obd_device *lwp)
503 struct lu_context session_ctx;
504 struct obd_export *exp;
505 struct obd_uuid *uuid = NULL;
506 struct obd_connect_data *data = NULL;
510 /* log has been fully processed, let clients connect */
511 rc = lu_env_init(&env, lwp->obd_lu_dev->ld_type->ldt_ctx_tags);
515 lu_context_init(&session_ctx, LCT_SESSION);
516 session_ctx.lc_thread = NULL;
517 lu_context_enter(&session_ctx);
518 env.le_ses = &session_ctx;
522 GOTO(out, rc = -ENOMEM);
524 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
525 data->ocd_version = LUSTRE_VERSION_CODE;
526 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
527 OBD_CONNECT_AT | OBD_CONNECT_LRU_RESIZE |
528 OBD_CONNECT_FULL20 | OBD_CONNECT_LVB_TYPE |
529 OBD_CONNECT_LIGHTWEIGHT;
532 GOTO(out, rc = -ENOMEM);
534 if (strlen(lwp->obd_name) > sizeof(uuid->uuid)) {
535 CERROR("%s: Too long lwp name %s, max_size is %d\n",
536 lwp->obd_name, lwp->obd_name, (int)sizeof(uuid->uuid));
537 GOTO(out, rc = -EINVAL);
540 /* Use lwp name as the uuid, so we find the export by lwp name later */
541 memcpy(uuid->uuid, lwp->obd_name, strlen(lwp->obd_name));
542 rc = obd_connect(&env, &exp, lwp, uuid, data, NULL);
544 CERROR("%s: connect failed: rc = %d\n", lwp->obd_name, rc);
546 lustre_notify_lwp_list(exp);
555 lu_context_exit(&session_ctx);
556 lu_context_fini(&session_ctx);
562 * lwp is used by slaves (Non-MDT0 targets) to manage the connection
565 static int lustre_lwp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi)
567 struct obd_connect_data *data = NULL;
568 struct obd_device *obd;
569 char *lwpname = NULL;
570 char *lwpuuid = NULL;
574 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
577 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
581 OBD_ALLOC(lwpname, MTI_NAME_MAXLEN);
583 GOTO(out, rc = -ENOMEM);
585 rc = tgt_name2lwpname(lsi->lsi_svname, lwpname);
587 CERROR("%s: failed to generate lwp name. %d\n",
588 lsi->lsi_svname, rc);
592 OBD_ALLOC(lwpuuid, MTI_NAME_MAXLEN);
594 GOTO(out, rc = -ENOMEM);
596 sprintf(lwpuuid, "%s_UUID", lwpname);
597 rc = lustre_start_simple(lwpname, LUSTRE_LWP_NAME,
598 lwpuuid, lustre_cfg_string(lcfg, 1),
601 CERROR("%s: setup up failed: rc %d\n", lwpname, rc);
605 obd = class_name2obd(lwpname);
606 LASSERT(obd != NULL);
608 rc = lustre_lwp_connect(obd);
610 CERROR("%s: connect failed: rc = %d\n", lwpname, rc);
615 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
617 OBD_FREE(lwpuuid, MTI_NAME_MAXLEN);
622 /* the caller is responsible for memory free */
623 static struct obd_device *lustre_find_lwp(struct lustre_sb_info *lsi,
624 char **lwpname, char **logname)
626 struct obd_device *lwp;
630 LASSERT(lwpname != NULL);
631 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
633 OBD_ALLOC(*lwpname, MTI_NAME_MAXLEN);
634 if (*lwpname == NULL)
635 RETURN(ERR_PTR(-ENOMEM));
637 if (logname != NULL) {
638 OBD_ALLOC(*logname, MTI_NAME_MAXLEN);
639 if (*logname == NULL)
640 GOTO(out, rc = -ENOMEM);
641 rc = server_name2fsname(lsi->lsi_svname, *lwpname, NULL);
643 CERROR("%s: failed to get fsname from svname. %d\n",
644 lsi->lsi_svname, rc);
645 GOTO(out, rc = -EINVAL);
647 sprintf(*logname, "%s-client", *lwpname);
650 rc = tgt_name2lwpname(lsi->lsi_svname, *lwpname);
652 CERROR("%s: failed to generate lwp name. %d\n",
653 lsi->lsi_svname, rc);
654 GOTO(out, rc = -EINVAL);
657 lwp = class_name2obd(*lwpname);
661 if (*lwpname != NULL) {
662 OBD_FREE(*lwpname, MTI_NAME_MAXLEN);
665 if (logname != NULL && *logname != NULL) {
666 OBD_FREE(*logname, MTI_NAME_MAXLEN);
672 RETURN(lwp != NULL ? lwp : ERR_PTR(-ENOENT));
675 static int lustre_lwp_add_conn(struct lustre_cfg *cfg,
676 struct lustre_sb_info *lsi)
678 struct lustre_cfg_bufs *bufs = NULL;
679 struct lustre_cfg *lcfg = NULL;
680 char *lwpname = NULL;
681 struct obd_device *lwp;
685 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
687 CERROR("%s: can't find lwp device.\n", lsi->lsi_svname);
688 GOTO(out, rc = PTR_ERR(lwp));
690 LASSERT(lwpname != NULL);
694 GOTO(out, rc = -ENOMEM);
696 lustre_cfg_bufs_reset(bufs, lwpname);
697 lustre_cfg_bufs_set_string(bufs, 1,
698 lustre_cfg_string(cfg, 1));
700 lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
702 rc = class_add_conn(lwp, lcfg);
704 CERROR("%s: can't add conn: rc = %d\n", lwpname, rc);
710 lustre_cfg_free(lcfg);
712 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
717 * Retrieve MDT nids from the client log, then start the lwp device.
718 * there are only two scenarios which would include mdt nid.
720 * marker 5 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add mdc' xxx-
721 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
722 * attach 0:lustre-MDT0000-mdc 1:mdc 2:lustre-clilmv_UUID
723 * setup 0:lustre-MDT0000-mdc 1:lustre-MDT0000_UUID 2:192.168.122.162@tcp
724 * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
725 * add_conn 0:lustre-MDT0000-mdc 1:192.168.172.1@tcp
726 * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID xxxx
727 * marker 5 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add mdc' xxxx-
729 * marker 7 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
730 * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
731 * add_conn 0:lustre-MDT0000-mdc 1:192.168.122.2@tcp
732 * marker 7 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
734 static int client_lwp_config_process(const struct lu_env *env,
735 struct llog_handle *handle,
736 struct llog_rec_hdr *rec, void *data)
738 struct config_llog_instance *clli = data;
739 int cfg_len = rec->lrh_len;
740 char *cfg_buf = (char *) (rec + 1);
741 struct lustre_cfg *lcfg = NULL;
742 struct lustre_sb_info *lsi;
743 int rc = 0, swab = 0;
746 if (rec->lrh_type != OBD_CFG_REC) {
747 CERROR("Unknown llog record type %#x encountered\n",
752 LASSERT(clli->cfg_sb != NULL);
753 lsi = s2lsi(clli->cfg_sb);
755 lcfg = (struct lustre_cfg *)cfg_buf;
756 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
757 lustre_swab_lustre_cfg(lcfg);
761 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
765 switch (lcfg->lcfg_command) {
767 struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
769 lustre_swab_cfg_marker(marker, swab,
770 LUSTRE_CFG_BUFLEN(lcfg, 1));
771 if (marker->cm_flags & CM_SKIP ||
772 marker->cm_flags & CM_EXCLUDE)
775 if (!tgt_is_mdt0(marker->cm_tgtname))
778 if (!strncmp(marker->cm_comment, "add mdc", 7) ||
779 !strncmp(marker->cm_comment, "add failnid", 11)) {
780 if (marker->cm_flags & CM_START) {
781 clli->cfg_flags = CFG_F_MARKER;
782 /* This hack is to differentiate the
783 * ADD_UUID is come from "add mdc" record
784 * or from "add failnid" record. */
785 if (!strncmp(marker->cm_comment,
787 clli->cfg_flags |= CFG_F_SKIP;
788 } else if (marker->cm_flags & CM_END) {
794 case LCFG_ADD_UUID: {
795 if (clli->cfg_flags == CFG_F_MARKER) {
796 rc = lustre_lwp_setup(lcfg, lsi);
797 /* XXX: process only the first nid as
798 * we don't need another instance of lwp */
799 clli->cfg_flags |= CFG_F_SKIP;
800 } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
801 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
804 CERROR("%s: Fail to add uuid, rc:%d\n",
805 lsi->lsi_svname, rc);
809 case LCFG_ADD_CONN: {
810 if (is_mdc_for_mdt0(lustre_cfg_string(lcfg, 0)))
811 rc = lustre_lwp_add_conn(lcfg, lsi);
821 static int lustre_disconnect_lwp(struct super_block *sb)
823 struct lustre_sb_info *lsi = s2lsi(sb);
824 struct obd_device *lwp;
825 char *lwpname = NULL;
826 char *logname = NULL;
827 struct lustre_cfg *lcfg = NULL;
828 struct lustre_cfg_bufs *bufs = NULL;
829 struct config_llog_instance *cfg = NULL;
833 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
834 if (IS_ERR(lwp) && PTR_ERR(lwp) != -ENOENT)
835 GOTO(out, rc = PTR_ERR(lwp));
837 LASSERT(lwpname != NULL);
838 LASSERT(logname != NULL);
842 GOTO(out, rc = -ENOMEM);
845 cfg->cfg_instance = sb;
846 rc = lustre_end_log(sb, logname, cfg);
848 CERROR("%s: Can't end config log %s.\n", lwpname, logname);
852 if (PTR_ERR(lwp) == -ENOENT) {
853 CDEBUG(D_CONFIG, "%s: lwp device wasn't started.\n",
860 GOTO(out, rc = -ENOMEM);
862 lustre_cfg_bufs_reset(bufs, lwp->obd_name);
863 lustre_cfg_bufs_set_string(bufs, 1, NULL);
864 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
866 GOTO(out, rc = -ENOMEM);
868 /* Disconnect import first. NULL is passed for the '@env', since
869 * it will not be used. */
870 rc = lwp->obd_lu_dev->ld_ops->ldo_process_config(NULL, lwp->obd_lu_dev,
874 lustre_cfg_free(lcfg);
880 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
882 OBD_FREE(logname, MTI_NAME_MAXLEN);
887 * Stop the lwp for an OST/MDT target.
889 static int lustre_stop_lwp(struct super_block *sb)
891 struct lustre_sb_info *lsi = s2lsi(sb);
892 struct obd_device *lwp = NULL;
893 char *lwpname = NULL;
897 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
899 CDEBUG(PTR_ERR(lwp) == -ENOENT ? D_CONFIG : D_ERROR,
900 "%s: lwp wasn't started.\n", lsi->lsi_svname);
905 rc = class_manual_cleanup(lwp);
909 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
914 * Start the lwp(fsname-MDT0000-lwp-OSTxxxx) for an OST or MDT target,
915 * which would be used to establish connection from OST to MDT0.
917 static int lustre_start_lwp(struct super_block *sb)
919 struct lustre_sb_info *lsi = s2lsi(sb);
920 struct config_llog_instance *cfg = NULL;
921 struct obd_device *lwp;
922 char *lwpname = NULL;
923 char *logname = NULL;
927 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
929 /* the lwp device already stared */
930 if (lwp && !IS_ERR(lwp))
933 if (PTR_ERR(lwp) != -ENOENT)
934 GOTO(out, rc = PTR_ERR(lwp));
936 LASSERT(lwpname != NULL);
937 LASSERT(logname != NULL);
941 GOTO(out, rc = -ENOMEM);
943 cfg->cfg_callback = client_lwp_config_process;
944 cfg->cfg_instance = sb;
946 rc = lustre_process_log(sb, logname, cfg);
949 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
951 OBD_FREE(logname, MTI_NAME_MAXLEN);
957 DEFINE_MUTEX(server_start_lock);
959 /* Stop MDS/OSS if nobody is using them */
960 static int server_stop_servers(int lsiflags)
962 struct obd_device *obd = NULL;
963 struct obd_type *type = NULL;
967 mutex_lock(&server_start_lock);
969 /* Either an MDT or an OST or neither */
970 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
971 if (lsiflags & LDD_F_SV_TYPE_MDT) {
972 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
974 type = class_search_type(LUSTRE_MDT_NAME);
977 /* if this was an OST, and there are no more OST's, clean up the OSS */
978 if (lsiflags & LDD_F_SV_TYPE_OST) {
979 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
981 type = class_search_type(LUSTRE_OST_NAME);
984 if (obd != NULL && (type == NULL || type->typ_refcnt == 0)) {
988 /* obd_fail doesn't mean much on a server obd */
989 err = class_manual_cleanup(obd);
994 mutex_unlock(&server_start_lock);
999 int server_mti_print(const char *title, struct mgs_target_info *mti)
1001 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
1002 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
1003 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
1004 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
1005 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
1006 mti->mti_config_ver, mti->mti_flags);
1009 EXPORT_SYMBOL(server_mti_print);
1011 /* Generate data for registration */
1012 static int server_lsi2mti(struct lustre_sb_info *lsi,
1013 struct mgs_target_info *mti)
1015 lnet_process_id_t id;
1020 if (!IS_SERVER(lsi))
1023 if (strlcpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname))
1024 >= sizeof(mti->mti_svname))
1027 mti->mti_nid_count = 0;
1028 while (LNetGetId(i++, &id) != -ENOENT) {
1029 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1032 /* server use --servicenode param, only allow specified
1033 * nids be registered */
1034 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
1035 class_match_nid(lsi->lsi_lmd->lmd_params,
1036 PARAM_FAILNODE, id.nid) < 1)
1039 /* match specified network */
1040 if (!class_match_net(lsi->lsi_lmd->lmd_params,
1041 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1044 mti->mti_nids[mti->mti_nid_count] = id.nid;
1045 mti->mti_nid_count++;
1046 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1047 CWARN("Only using first %d nids for %s\n",
1048 mti->mti_nid_count, mti->mti_svname);
1053 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1054 mti->mti_config_ver = 0;
1056 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1060 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1063 /* Orion requires index to be set */
1064 LASSERT(!(rc & LDD_F_NEED_INDEX));
1065 /* keep only LDD flags */
1066 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1067 if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN))
1068 mti->mti_flags |= LDD_F_UPDATE;
1069 cplen = strlcpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1070 sizeof(mti->mti_params));
1071 if (cplen >= sizeof(mti->mti_params))
1076 /* Register an old or new target with the MGS. If needed MGS will construct
1077 startup logs and assign index */
1078 static int server_register_target(struct lustre_sb_info *lsi)
1080 struct obd_device *mgc = lsi->lsi_mgc;
1081 struct mgs_target_info *mti = NULL;
1088 if (!IS_SERVER(lsi))
1095 rc = server_lsi2mti(lsi, mti);
1099 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1100 mti->mti_svname, mti->mti_fsname,
1101 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1104 /* if write_conf is true, the registration must succeed */
1105 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1106 mti->mti_flags |= LDD_F_OPC_REG;
1108 /* Register the target */
1109 /* FIXME use mgc_process_config instead */
1110 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1111 sizeof(KEY_REGISTER_TARGET),
1112 KEY_REGISTER_TARGET,
1113 sizeof(*mti), mti, NULL);
1115 if (mti->mti_flags & LDD_F_ERROR) {
1116 LCONSOLE_ERROR_MSG(0x160,
1117 "%s: the MGS refuses to allow this server "
1118 "to start: rc = %d. Please see messages on "
1119 "the MGS.\n", lsi->lsi_svname, rc);
1120 } else if (writeconf) {
1121 LCONSOLE_ERROR_MSG(0x15f,
1122 "%s: cannot register this server with the MGS: "
1123 "rc = %d. Is the MGS running?\n",
1124 lsi->lsi_svname, rc);
1126 CERROR("%s: error registering with the MGS: rc = %d "
1127 "(not fatal)\n", lsi->lsi_svname, rc);
1128 /* reset the error code for non-fatal error. */
1141 * Notify the MGS that this target is ready.
1142 * Used by IR - if the MGS receives this message, it will notify clients.
1144 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1146 struct lustre_sb_info *lsi = s2lsi(sb);
1147 struct obd_device *mgc = lsi->lsi_mgc;
1148 struct mgs_target_info *mti = NULL;
1154 if (!(IS_SERVER(lsi)))
1160 rc = server_lsi2mti(lsi, mti);
1164 mti->mti_instance = obd->u.obt.obt_instance;
1165 mti->mti_flags |= LDD_F_OPC_READY;
1167 /* FIXME use mgc_process_config instead */
1168 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1169 sizeof(KEY_REGISTER_TARGET),
1170 KEY_REGISTER_TARGET,
1171 sizeof(*mti), mti, NULL);
1173 /* Imperative recovery: if the mgs informs us to use IR? */
1174 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1175 (mti->mti_flags & LDD_F_IR_CAPABLE))
1176 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1185 /** Start server targets: MDTs and OSTs
1187 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1189 struct obd_device *obd;
1190 struct lustre_sb_info *lsi = s2lsi(sb);
1191 struct config_llog_instance cfg;
1193 struct lu_device *dev;
1197 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1200 /* make sure the MDS is started */
1201 mutex_lock(&server_start_lock);
1202 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1204 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1206 LUSTRE_MDS_OBDNAME"_uuid",
1209 mutex_unlock(&server_start_lock);
1210 CERROR("failed to start MDS: %d\n", rc);
1214 mutex_unlock(&server_start_lock);
1217 /* If we're an OST, make sure the global OSS is running */
1219 /* make sure OSS is started */
1220 mutex_lock(&server_start_lock);
1221 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1223 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1225 LUSTRE_OSS_OBDNAME"_uuid",
1228 mutex_unlock(&server_start_lock);
1229 CERROR("failed to start OSS: %d\n", rc);
1233 mutex_unlock(&server_start_lock);
1236 /* Set the mgc fs to our server disk. This allows the MGC to
1237 * read and write configs locally, in case it can't talk to the MGS. */
1238 if (lsi->lsi_srv_mnt) {
1239 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1241 GOTO(out_stop_service, rc);
1244 /* Register with MGS */
1245 rc = server_register_target(lsi);
1249 /* Let the target look up the mount using the target's name
1250 (we can't pass the sb or mnt through class_process_config.) */
1251 rc = server_register_mount(lsi->lsi_svname, sb, mnt);
1255 /* Start targets using the llog named for the target */
1256 memset(&cfg, 0, sizeof(cfg));
1257 cfg.cfg_callback = class_config_llog_handler;
1258 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1260 CERROR("failed to start server %s: %d\n",
1261 lsi->lsi_svname, rc);
1262 /* Do NOT call server_deregister_mount() here. This makes it
1263 * impossible to find mount later in cleanup time and leaves
1264 * @lsi and othder stuff leaked. -umka */
1268 obd = class_name2obd(lsi->lsi_svname);
1270 CERROR("no server named %s was started\n", lsi->lsi_svname);
1271 GOTO(out_mgc, rc = -ENXIO);
1274 if (IS_OST(lsi) || IS_MDT(lsi)) {
1275 rc = lustre_start_lwp(sb);
1277 CERROR("%s: failed to start LWP: %d\n",
1278 lsi->lsi_svname, rc);
1283 server_notify_target(sb, obd);
1285 /* calculate recovery timeout, do it after lustre_process_log */
1286 server_calc_timeout(lsi, obd);
1288 /* log has been fully processed */
1289 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1291 /* log has been fully processed, let clients connect */
1292 dev = obd->obd_lu_dev;
1293 if (dev && dev->ld_ops->ldo_prepare) {
1294 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1296 struct lu_context session_ctx;
1298 lu_context_init(&session_ctx, LCT_SESSION);
1299 session_ctx.lc_thread = NULL;
1300 lu_context_enter(&session_ctx);
1301 env.le_ses = &session_ctx;
1303 rc = dev->ld_ops->ldo_prepare(&env, NULL, dev);
1306 lu_context_exit(&session_ctx);
1307 lu_context_fini(&session_ctx);
1311 /* abort recovery only on the complete stack:
1312 * many devices can be involved */
1313 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1314 (OBP(obd, iocontrol))) {
1315 obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
1320 /* Release the mgc fs for others to use */
1321 if (lsi->lsi_srv_mnt)
1322 server_mgc_clear_fs(lsi->lsi_mgc);
1326 server_stop_servers(lsi->lsi_flags);
1331 static int lsi_prepare(struct lustre_sb_info *lsi)
1338 LASSERT(lsi->lsi_lmd);
1340 /* The server name is given as a mount line option */
1341 if (lsi->lsi_lmd->lmd_profile == NULL) {
1342 LCONSOLE_ERROR("Can't determine server name\n");
1346 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
1347 RETURN(-ENAMETOOLONG);
1349 strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
1351 /* Determine osd type */
1352 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
1353 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
1354 sizeof(lsi->lsi_osd_type))
1355 RETURN(-ENAMETOOLONG);
1357 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
1359 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
1362 /* XXX: a temp. solution for components using fsfilt
1363 * to be removed in one of the subsequent patches */
1364 if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs"))
1365 strcpy(lsi->lsi_fstype, "ldiskfs");
1367 strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
1369 /* Determine server type */
1370 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1372 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
1373 /* Assume we're a bare MGS */
1375 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
1377 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
1382 lsi->lsi_flags |= rc;
1384 /* Add mount line flags that used to be in ldd:
1385 * writeconf, mgs, anything else?
1387 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
1388 LDD_F_WRITECONF : 0;
1389 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
1391 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_UPDATE) ?
1393 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
1394 LDD_F_SV_TYPE_MGS : 0;
1395 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
1396 LDD_F_NO_PRIMNODE : 0;
1401 /*************** server mount ******************/
1403 /** Start the shutdown of servers at umount.
1405 static void server_put_super(struct super_block *sb)
1407 struct lustre_sb_info *lsi = s2lsi(sb);
1408 struct obd_device *obd;
1409 char *tmpname, *extraname = NULL;
1411 int lsiflags = lsi->lsi_flags;
1414 LASSERT(IS_SERVER(lsi));
1416 tmpname_sz = strlen(lsi->lsi_svname) + 1;
1417 OBD_ALLOC(tmpname, tmpname_sz);
1418 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
1419 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1420 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1421 snprintf(tmpname, tmpname_sz, "MGS");
1423 /* disconnect the lwp first to drain off the inflight request */
1424 if (IS_OST(lsi) || IS_MDT(lsi)) {
1427 rc = lustre_disconnect_lwp(sb);
1428 if (rc && rc != ETIMEDOUT)
1429 CERROR("%s: failed to disconnect lwp. (rc=%d)\n",
1433 /* Stop the target */
1434 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1435 (IS_MDT(lsi) || IS_OST(lsi))) {
1436 struct lustre_profile *lprof = NULL;
1438 /* tell the mgc to drop the config log */
1439 lustre_end_log(sb, lsi->lsi_svname, NULL);
1441 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1442 If there are any setup/cleanup errors, save the lov
1443 name for safety cleanup later. */
1444 lprof = class_get_profile(lsi->lsi_svname);
1445 if (lprof && lprof->lp_dt) {
1446 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1447 strcpy(extraname, lprof->lp_dt);
1450 obd = class_name2obd(lsi->lsi_svname);
1452 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1453 if (lsiflags & LSI_UMOUNT_FAILOVER)
1455 /* We can't seem to give an error return code
1456 * to .put_super, so we better make sure we clean up! */
1458 class_manual_cleanup(obd);
1460 CERROR("no obd %s\n", lsi->lsi_svname);
1461 server_deregister_mount(lsi->lsi_svname);
1465 /* If they wanted the mgs to stop separately from the mdt, they
1466 should have put it on a different device. */
1468 /* if MDS start with --nomgs, don't stop MGS then */
1469 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1470 server_stop_mgs(sb);
1473 if (IS_OST(lsi) || IS_MDT(lsi)) {
1474 if (lustre_stop_lwp(sb) < 0)
1475 CERROR("%s: failed to stop lwp!\n", tmpname);
1478 /* Clean the mgc and sb */
1479 lustre_common_put_super(sb);
1481 /* wait till all in-progress cleanups are done
1482 * specifically we're interested in ofd cleanup
1484 obd_zombie_barrier();
1486 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1487 until the target is really gone so that our type refcount check
1489 server_stop_servers(lsiflags);
1491 /* In case of startup or cleanup err, stop related obds */
1493 obd = class_name2obd(extraname);
1495 CWARN("Cleaning orphaned obd %s\n", extraname);
1497 class_manual_cleanup(obd);
1499 OBD_FREE(extraname, strlen(extraname) + 1);
1502 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1503 OBD_FREE(tmpname, tmpname_sz);
1507 /** Called only for 'umount -f'
1509 static void server_umount_begin(struct super_block *sb)
1511 struct lustre_sb_info *lsi = s2lsi(sb);
1514 CDEBUG(D_MOUNT, "umount -f\n");
1515 /* umount = failover
1517 no third way to do non-force, non-failover */
1518 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1522 static int server_statfs(struct dentry *dentry, struct kstatfs *buf)
1524 struct super_block *sb = dentry->d_sb;
1525 struct lustre_sb_info *lsi = s2lsi(sb);
1526 struct obd_statfs statfs;
1530 if (lsi->lsi_dt_dev) {
1531 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
1533 statfs_unpack(buf, &statfs);
1534 buf->f_type = sb->s_magic;
1540 buf->f_type = sb->s_magic;
1541 buf->f_bsize = sb->s_blocksize;
1547 buf->f_namelen = NAME_MAX;
1551 /** The operations we support directly on the superblock:
1552 * mount, umount, and df.
1554 static struct super_operations server_ops = {
1555 .put_super = server_put_super,
1556 .umount_begin = server_umount_begin, /* umount -f */
1557 .statfs = server_statfs,
1560 #define log2(n) ffz(~(n))
1561 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1563 static int server_fill_super_common(struct super_block *sb)
1565 struct inode *root = 0;
1568 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1570 sb->s_blocksize = 4096;
1571 sb->s_blocksize_bits = log2(sb->s_blocksize);
1572 sb->s_magic = LUSTRE_SUPER_MAGIC;
1573 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1574 sb->s_flags |= MS_RDONLY;
1575 sb->s_op = &server_ops;
1577 root = new_inode(sb);
1579 CERROR("Can't make root inode\n");
1583 /* returns -EIO for every operation */
1584 /* make_bad_inode(root); -- badness - can't umount */
1585 /* apparently we need to be a directory for the mount to finish */
1586 root->i_mode = S_IFDIR;
1588 sb->s_root = d_make_root(root);
1590 CERROR("%s: can't make root dentry\n", sb->s_id);
1597 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
1599 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1600 struct obd_device *obd;
1601 struct dt_device_param p;
1607 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
1608 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
1610 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
1611 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
1612 strcat(lsi->lsi_osd_uuid, "_UUID");
1613 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
1615 obd = class_name2obd(lsi->lsi_osd_obdname);
1617 rc = lustre_start_simple(lsi->lsi_osd_obdname,
1619 lsi->lsi_osd_uuid, lmd->lmd_dev,
1620 flagstr, lsi->lsi_lmd->lmd_opts,
1624 obd = class_name2obd(lsi->lsi_osd_obdname);
1628 rc = obd_connect(NULL, &lsi->lsi_osd_exp,
1629 obd, &obd->obd_uuid, NULL, NULL);
1632 class_manual_cleanup(obd);
1633 lsi->lsi_dt_dev = NULL;
1636 /* XXX: to keep support old components relying on lsi_srv_mnt
1637 * we get this info from OSD just started */
1638 LASSERT(obd->obd_lu_dev);
1639 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
1640 LASSERT(lsi->lsi_dt_dev);
1642 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
1644 lsi->lsi_srv_mnt = p.ddp_mnt;
1650 /** Fill in the superblock info for a Lustre server.
1651 * Mount the device with the correct options.
1652 * Read the on-disk config file.
1653 * Start the services.
1655 int server_fill_super(struct super_block *sb)
1657 struct lustre_sb_info *lsi = s2lsi(sb);
1661 rc = lsi_prepare(lsi);
1665 /* Start low level OSD */
1666 rc = osd_start(lsi, sb->s_flags);
1668 CERROR("Unable to start osd on %s: %d\n",
1669 lsi->lsi_lmd->lmd_dev, rc);
1674 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
1675 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
1677 if (class_name2obd(lsi->lsi_svname)) {
1678 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1679 "running. Double-mount may have compromised"
1680 " the disk journal.\n",
1686 /* Start MGS before MGC */
1687 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1688 rc = server_start_mgs(sb);
1693 /* Start MGC before servers */
1694 rc = lustre_start_mgc(sb);
1698 /* Set up all obd devices for service */
1699 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1700 (IS_OST(lsi) || IS_MDT(lsi))) {
1701 rc = server_start_targets(sb, lsi->lsi_srv_mnt);
1703 CERROR("Unable to start targets: %d\n", rc);
1706 /* FIXME overmount client here, or can we just start a
1707 * client log and client_fill_super on this sb? We
1708 * need to make sure server_put_super gets called too
1709 * - ll_put_super calls lustre_common_put_super; check
1710 * there for LSI_SERVER flag, call s_p_s if so.
1712 * Probably should start client from new thread so we
1713 * can return. Client will not finish until all
1714 * servers are connected. Note - MGS-only server does
1715 * NOT get a client, since there is no lustre fs
1716 * associated - the MGS is for all lustre fs's */
1719 rc = server_fill_super_common(sb);
1725 /* We jump here in case of failure while starting targets or MGS.
1726 * In this case we can't just put @mnt and have to do real cleanup
1727 * with stoping targets, etc. */
1728 server_put_super(sb);
1733 * Calculate timeout value for a target.
1735 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1737 struct lustre_mount_data *lmd;
1741 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
1742 int min = OBD_RECOVERY_TIME_MIN;
1744 LASSERT(IS_SERVER(lsi));
1748 soft = lmd->lmd_recovery_time_soft;
1749 hard = lmd->lmd_recovery_time_hard;
1750 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1751 obd->obd_no_ir = !has_ir;
1755 soft = OBD_RECOVERY_TIME_SOFT;
1757 hard = OBD_RECOVERY_TIME_HARD;
1759 /* target may have ir_factor configured. */
1760 factor = OBD_IR_FACTOR_DEFAULT;
1761 if (obd->obd_recovery_ir_factor)
1762 factor = obd->obd_recovery_ir_factor;
1765 int new_soft = soft;
1766 int new_hard = hard;
1768 /* adjust timeout value by imperative recovery */
1770 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1771 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1773 /* make sure the timeout is not too short */
1774 new_soft = max(min, new_soft);
1775 new_hard = max(new_soft, new_hard);
1777 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1778 "window shrunk from %d-%d down to %d-%d\n",
1779 obd->obd_name, soft, hard, new_soft, new_hard);
1786 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1787 obd->obd_recovery_time_hard = hard;
1788 obd->obd_recovery_ir_factor = factor;
1790 EXPORT_SYMBOL(server_calc_timeout);