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>
60 #ifdef HAVE_SELINUX_IS_ENABLED
61 #include <linux/selinux.h>
64 /*********** mount lookup *********/
66 DEFINE_MUTEX(lustre_mount_info_lock);
67 static CFS_LIST_HEAD(server_mount_info_list);
69 static struct lustre_mount_info *server_find_mount(const char *name)
72 struct lustre_mount_info *lmi;
75 cfs_list_for_each(tmp, &server_mount_info_list) {
76 lmi = cfs_list_entry(tmp, struct lustre_mount_info,
78 if (strcmp(name, lmi->lmi_name) == 0)
84 /* we must register an obd for a mount before we call the setup routine.
85 *_setup will call lustre_get_mount to get the mnt struct
86 by obd_name, since we can't pass the pointer to setup. */
87 static int server_register_mount(const char *name, struct super_block *sb,
90 struct lustre_mount_info *lmi;
96 OBD_ALLOC(lmi, sizeof(*lmi));
99 OBD_ALLOC(name_cp, strlen(name) + 1);
101 OBD_FREE(lmi, sizeof(*lmi));
104 strcpy(name_cp, name);
106 mutex_lock(&lustre_mount_info_lock);
108 if (server_find_mount(name)) {
109 mutex_unlock(&lustre_mount_info_lock);
110 OBD_FREE(lmi, sizeof(*lmi));
111 OBD_FREE(name_cp, strlen(name) + 1);
112 CERROR("Already registered %s\n", name);
115 lmi->lmi_name = name_cp;
118 cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list);
120 mutex_unlock(&lustre_mount_info_lock);
122 CDEBUG(D_MOUNT, "reg_mnt %p from %s\n", lmi->lmi_mnt, name);
127 /* when an obd no longer needs a mount */
128 static int server_deregister_mount(const char *name)
130 struct lustre_mount_info *lmi;
133 mutex_lock(&lustre_mount_info_lock);
134 lmi = server_find_mount(name);
136 mutex_unlock(&lustre_mount_info_lock);
137 CERROR("%s not registered\n", name);
141 CDEBUG(D_MOUNT, "dereg_mnt %p from %s\n", lmi->lmi_mnt, name);
143 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
144 cfs_list_del(&lmi->lmi_list_chain);
145 OBD_FREE(lmi, sizeof(*lmi));
146 mutex_unlock(&lustre_mount_info_lock);
151 /* obd's look up a registered mount using their obdname. This is just
152 for initial obd setup to find the mount struct. It should not be
153 called every time you want to mntget. */
154 struct lustre_mount_info *server_get_mount(const char *name)
156 struct lustre_mount_info *lmi;
157 struct lustre_sb_info *lsi;
160 mutex_lock(&lustre_mount_info_lock);
161 lmi = server_find_mount(name);
162 mutex_unlock(&lustre_mount_info_lock);
164 CERROR("Can't find mount for %s\n", name);
167 lsi = s2lsi(lmi->lmi_sb);
169 cfs_atomic_inc(&lsi->lsi_mounts);
171 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d\n", lmi->lmi_mnt,
172 name, cfs_atomic_read(&lsi->lsi_mounts));
176 EXPORT_SYMBOL(server_get_mount);
179 * Used by mdt to get mount_info from obdname.
180 * There are no blocking when using the mount_info.
181 * Do not use server_get_mount for this purpose.
183 struct lustre_mount_info *server_get_mount_2(const char *name)
185 struct lustre_mount_info *lmi;
188 mutex_lock(&lustre_mount_info_lock);
189 lmi = server_find_mount(name);
190 mutex_unlock(&lustre_mount_info_lock);
192 CERROR("Can't find mount for %s\n", name);
196 EXPORT_SYMBOL(server_get_mount_2);
198 /* to be called from obd_cleanup methods */
199 int server_put_mount(const char *name, struct vfsmount *mnt)
201 struct lustre_mount_info *lmi;
202 struct lustre_sb_info *lsi;
205 mutex_lock(&lustre_mount_info_lock);
206 lmi = server_find_mount(name);
207 mutex_unlock(&lustre_mount_info_lock);
209 CERROR("Can't find mount for %s\n", name);
212 lsi = s2lsi(lmi->lmi_sb);
214 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d\n",
215 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts));
217 if (lustre_put_lsi(lmi->lmi_sb))
218 CDEBUG(D_MOUNT, "Last put of mnt %p from %s\n",
221 /* this obd should never need the mount again */
222 server_deregister_mount(name);
226 EXPORT_SYMBOL(server_put_mount);
228 /* Corresponding to server_get_mount_2 */
229 int server_put_mount_2(const char *name, struct vfsmount *mnt)
234 EXPORT_SYMBOL(server_put_mount_2);
236 /* Set up a MGS to serve startup logs */
237 static int server_start_mgs(struct super_block *sb)
239 struct lustre_sb_info *lsi = s2lsi(sb);
240 struct vfsmount *mnt = lsi->lsi_srv_mnt;
241 struct lustre_mount_info *lmi;
245 /* It is impossible to have more than 1 MGS per node, since
246 MGC wouldn't know which to connect to */
247 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
249 lsi = s2lsi(lmi->lmi_sb);
250 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
255 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
257 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
260 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
261 LUSTRE_MGS_OBDNAME, 0, 0,
262 lsi->lsi_osd_obdname, 0);
263 /* server_deregister_mount() is not called previously, for lsi
264 * and other stuff can't be freed cleanly when mgs calls
265 * server_put_mount() in error handling case (see b=17758),
266 * this problem is caused by a bug in mgs_init0, which forgot
267 * calling server_put_mount in error case. */
270 server_deregister_mount(LUSTRE_MGS_OBDNAME);
274 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
275 "Is the 'mgs' module loaded?\n",
276 LUSTRE_MGS_OBDNAME, rc);
280 static int server_stop_mgs(struct super_block *sb)
282 struct obd_device *obd;
286 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
288 /* There better be only one MGS */
289 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
291 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
295 /* The MGS should always stop when we say so */
297 rc = class_manual_cleanup(obd);
301 /* Since there's only one mgc per node, we have to change it's fs to get
302 access to the right disk. */
303 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
305 struct lustre_sb_info *lsi = s2lsi(sb);
309 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
311 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
312 rc = obd_set_info_async(NULL, mgc->obd_self_export,
313 sizeof(KEY_SET_FS), KEY_SET_FS,
314 sizeof(*sb), sb, NULL);
316 CERROR("can't set_fs %d\n", rc);
321 static int server_mgc_clear_fs(struct obd_device *mgc)
326 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
328 rc = obd_set_info_async(NULL, mgc->obd_self_export,
329 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
334 static int is_mdc_device(const char *devname)
338 ptr = strrchr(devname, '-');
339 if (ptr != NULL && strcmp(ptr, "-mdc") == 0)
345 static inline int tgt_is_mdt0(const char *tgtname)
350 type = server_name2index(tgtname, &idx, NULL);
351 if (type != LDD_F_SV_TYPE_MDT)
357 static inline int is_mdc_for_mdt0(const char *devname)
361 if (!is_mdc_device(devname))
364 ptr = strrchr(devname, '-');
369 if (tgt_is_mdt0(devname)) {
378 * Convert OST/MDT name(fsname-OSTxxxx) to a lwp name
379 * (fsname-MDT0000-lwp-OSTxxxx)
381 int tgt_name2lwpname(const char *svname, char *lwpname)
388 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
392 rc = server_name2fsname(svname, fsname, &tgt);
394 CERROR("%s: failed to get fsname from svname. %d\n",
399 if (*tgt != '-' && *tgt != ':') {
400 CERROR("%s: invalid svname name!\n", svname);
401 GOTO(cleanup, rc = -EINVAL);
405 if (strncmp(tgt, "OST", 3) != 0 && strncmp(tgt, "MDT", 3) != 0) {
406 CERROR("%s is not an OST or MDT target!\n", svname);
407 GOTO(cleanup, rc = -EINVAL);
409 sprintf(lwpname, "%s-MDT0000-%s-%s", fsname, LUSTRE_LWP_NAME, tgt);
412 OBD_FREE(fsname, MTI_NAME_MAXLEN);
415 EXPORT_SYMBOL(tgt_name2lwpname);
417 static CFS_LIST_HEAD(lwp_register_list);
418 DEFINE_MUTEX(lwp_register_list_lock);
420 int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp,
421 register_lwp_cb cb_func, void *cb_data)
423 struct obd_device *lwp;
424 struct lwp_register_item *lri;
427 LASSERTF(strlen(lwpname) < MTI_NAME_MAXLEN, "lwpname is too long %s\n",
429 LASSERT(exp != NULL && *exp == NULL);
435 mutex_lock(&lwp_register_list_lock);
437 lwp = class_name2obd(lwpname);
438 if (lwp != NULL && lwp->obd_set_up == 1) {
439 struct obd_uuid *uuid;
443 mutex_unlock(&lwp_register_list_lock);
447 memcpy(uuid->uuid, lwpname, strlen(lwpname));
448 *exp = cfs_hash_lookup(lwp->obd_uuid_hash, uuid);
452 memcpy(lri->lri_name, lwpname, strlen(lwpname));
454 lri->lri_cb_func = cb_func;
455 lri->lri_cb_data = cb_data;
456 CFS_INIT_LIST_HEAD(&lri->lri_list);
457 cfs_list_add(&lri->lri_list, &lwp_register_list);
459 if (*exp != NULL && cb_func != NULL)
462 mutex_unlock(&lwp_register_list_lock);
465 EXPORT_SYMBOL(lustre_register_lwp_item);
467 void lustre_deregister_lwp_item(struct obd_export **exp)
469 struct lwp_register_item *lri, *tmp;
471 mutex_lock(&lwp_register_list_lock);
472 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
473 if (exp == lri->lri_exp) {
475 class_export_put(*exp);
476 cfs_list_del(&lri->lri_list);
481 mutex_unlock(&lwp_register_list_lock);
483 EXPORT_SYMBOL(lustre_deregister_lwp_item);
485 static void lustre_notify_lwp_list(struct obd_export *exp)
487 struct lwp_register_item *lri, *tmp;
488 LASSERT(exp != NULL);
490 mutex_lock(&lwp_register_list_lock);
491 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
492 if (strcmp(exp->exp_obd->obd_name, lri->lri_name))
494 if (*lri->lri_exp != NULL)
496 *lri->lri_exp = class_export_get(exp);
497 if (lri->lri_cb_func != NULL)
498 lri->lri_cb_func(lri->lri_cb_data);
500 mutex_unlock(&lwp_register_list_lock);
503 static int lustre_lwp_connect(struct obd_device *lwp)
506 struct lu_context session_ctx;
507 struct obd_export *exp;
508 struct obd_uuid *uuid = NULL;
509 struct obd_connect_data *data = NULL;
513 /* log has been fully processed, let clients connect */
514 rc = lu_env_init(&env, lwp->obd_lu_dev->ld_type->ldt_ctx_tags);
518 lu_context_init(&session_ctx, LCT_SESSION);
519 session_ctx.lc_thread = NULL;
520 lu_context_enter(&session_ctx);
521 env.le_ses = &session_ctx;
525 GOTO(out, rc = -ENOMEM);
527 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
528 data->ocd_version = LUSTRE_VERSION_CODE;
529 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
530 OBD_CONNECT_AT | OBD_CONNECT_LRU_RESIZE |
531 OBD_CONNECT_FULL20 | OBD_CONNECT_LVB_TYPE |
532 OBD_CONNECT_LIGHTWEIGHT;
535 GOTO(out, rc = -ENOMEM);
537 if (strlen(lwp->obd_name) > sizeof(uuid->uuid)) {
538 CERROR("%s: Too long lwp name %s, max_size is %d\n",
539 lwp->obd_name, lwp->obd_name, (int)sizeof(uuid->uuid));
540 GOTO(out, rc = -EINVAL);
543 /* Use lwp name as the uuid, so we find the export by lwp name later */
544 memcpy(uuid->uuid, lwp->obd_name, strlen(lwp->obd_name));
545 rc = obd_connect(&env, &exp, lwp, uuid, data, NULL);
547 CERROR("%s: connect failed: rc = %d\n", lwp->obd_name, rc);
549 lustre_notify_lwp_list(exp);
558 lu_context_exit(&session_ctx);
559 lu_context_fini(&session_ctx);
565 * lwp is used by slaves (Non-MDT0 targets) to manage the connection
568 static int lustre_lwp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi)
570 struct obd_device *obd;
571 char *lwpname = NULL;
572 char *lwpuuid = NULL;
576 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
579 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
583 OBD_ALLOC(lwpname, MTI_NAME_MAXLEN);
585 GOTO(out, rc = -ENOMEM);
587 rc = tgt_name2lwpname(lsi->lsi_svname, lwpname);
589 CERROR("%s: failed to generate lwp name. %d\n",
590 lsi->lsi_svname, rc);
594 OBD_ALLOC(lwpuuid, MTI_NAME_MAXLEN);
596 GOTO(out, rc = -ENOMEM);
598 sprintf(lwpuuid, "%s_UUID", lwpname);
599 rc = lustre_start_simple(lwpname, LUSTRE_LWP_NAME,
600 lwpuuid, lustre_cfg_string(lcfg, 1),
603 CERROR("%s: setup up failed: rc %d\n", lwpname, rc);
607 obd = class_name2obd(lwpname);
608 LASSERT(obd != NULL);
610 rc = lustre_lwp_connect(obd);
612 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)) {
986 /* obd_fail doesn't mean much on a server obd */
987 rc = class_manual_cleanup(obd);
990 mutex_unlock(&server_start_lock);
995 int server_mti_print(const char *title, struct mgs_target_info *mti)
997 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
998 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
999 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
1000 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
1001 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
1002 mti->mti_config_ver, mti->mti_flags);
1005 EXPORT_SYMBOL(server_mti_print);
1007 /* Generate data for registration */
1008 static int server_lsi2mti(struct lustre_sb_info *lsi,
1009 struct mgs_target_info *mti)
1011 lnet_process_id_t id;
1016 if (!IS_SERVER(lsi))
1019 if (strlcpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname))
1020 >= sizeof(mti->mti_svname))
1023 mti->mti_nid_count = 0;
1024 while (LNetGetId(i++, &id) != -ENOENT) {
1025 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1028 /* server use --servicenode param, only allow specified
1029 * nids be registered */
1030 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
1031 class_match_nid(lsi->lsi_lmd->lmd_params,
1032 PARAM_FAILNODE, id.nid) < 1)
1035 /* match specified network */
1036 if (!class_match_net(lsi->lsi_lmd->lmd_params,
1037 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1040 mti->mti_nids[mti->mti_nid_count] = id.nid;
1041 mti->mti_nid_count++;
1042 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1043 CWARN("Only using first %d nids for %s\n",
1044 mti->mti_nid_count, mti->mti_svname);
1049 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1050 mti->mti_config_ver = 0;
1052 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1056 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1059 /* Orion requires index to be set */
1060 LASSERT(!(rc & LDD_F_NEED_INDEX));
1061 /* keep only LDD flags */
1062 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1063 if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN))
1064 mti->mti_flags |= LDD_F_UPDATE;
1065 cplen = strlcpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1066 sizeof(mti->mti_params));
1067 if (cplen >= sizeof(mti->mti_params))
1072 /* Register an old or new target with the MGS. If needed MGS will construct
1073 startup logs and assign index */
1074 static int server_register_target(struct lustre_sb_info *lsi)
1076 struct obd_device *mgc = lsi->lsi_mgc;
1077 struct mgs_target_info *mti = NULL;
1084 if (!IS_SERVER(lsi))
1091 rc = server_lsi2mti(lsi, mti);
1095 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1096 mti->mti_svname, mti->mti_fsname,
1097 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1100 /* if write_conf is true, the registration must succeed */
1101 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1102 mti->mti_flags |= LDD_F_OPC_REG;
1104 /* Register the target */
1105 /* FIXME use mgc_process_config instead */
1106 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1107 sizeof(KEY_REGISTER_TARGET),
1108 KEY_REGISTER_TARGET,
1109 sizeof(*mti), mti, NULL);
1111 if (mti->mti_flags & LDD_F_ERROR) {
1112 LCONSOLE_ERROR_MSG(0x160,
1113 "%s: the MGS refuses to allow this server "
1114 "to start: rc = %d. Please see messages on "
1115 "the MGS.\n", lsi->lsi_svname, rc);
1116 } else if (writeconf) {
1117 LCONSOLE_ERROR_MSG(0x15f,
1118 "%s: cannot register this server with the MGS: "
1119 "rc = %d. Is the MGS running?\n",
1120 lsi->lsi_svname, rc);
1122 CERROR("%s: error registering with the MGS: rc = %d "
1123 "(not fatal)\n", lsi->lsi_svname, rc);
1124 /* reset the error code for non-fatal error. */
1137 * Notify the MGS that this target is ready.
1138 * Used by IR - if the MGS receives this message, it will notify clients.
1140 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1142 struct lustre_sb_info *lsi = s2lsi(sb);
1143 struct obd_device *mgc = lsi->lsi_mgc;
1144 struct mgs_target_info *mti = NULL;
1150 if (!(IS_SERVER(lsi)))
1156 rc = server_lsi2mti(lsi, mti);
1160 mti->mti_instance = obd->u.obt.obt_instance;
1161 mti->mti_flags |= LDD_F_OPC_READY;
1163 /* FIXME use mgc_process_config instead */
1164 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1165 sizeof(KEY_REGISTER_TARGET),
1166 KEY_REGISTER_TARGET,
1167 sizeof(*mti), mti, NULL);
1169 /* Imperative recovery: if the mgs informs us to use IR? */
1170 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1171 (mti->mti_flags & LDD_F_IR_CAPABLE))
1172 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1181 /** Start server targets: MDTs and OSTs
1183 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1185 struct obd_device *obd;
1186 struct lustre_sb_info *lsi = s2lsi(sb);
1187 struct config_llog_instance cfg;
1189 struct lu_device *dev;
1193 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1196 /* make sure the MDS is started */
1197 mutex_lock(&server_start_lock);
1198 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1200 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1202 LUSTRE_MDS_OBDNAME"_uuid",
1205 mutex_unlock(&server_start_lock);
1206 CERROR("failed to start MDS: %d\n", rc);
1210 mutex_unlock(&server_start_lock);
1213 /* If we're an OST, make sure the global OSS is running */
1215 /* make sure OSS is started */
1216 mutex_lock(&server_start_lock);
1217 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1219 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1221 LUSTRE_OSS_OBDNAME"_uuid",
1224 mutex_unlock(&server_start_lock);
1225 CERROR("failed to start OSS: %d\n", rc);
1229 mutex_unlock(&server_start_lock);
1232 /* Set the mgc fs to our server disk. This allows the MGC to
1233 * read and write configs locally, in case it can't talk to the MGS. */
1234 if (lsi->lsi_srv_mnt) {
1235 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1237 GOTO(out_stop_service, rc);
1240 /* Register with MGS */
1241 rc = server_register_target(lsi);
1245 /* Let the target look up the mount using the target's name
1246 (we can't pass the sb or mnt through class_process_config.) */
1247 rc = server_register_mount(lsi->lsi_svname, sb, mnt);
1251 /* Start targets using the llog named for the target */
1252 memset(&cfg, 0, sizeof(cfg));
1253 cfg.cfg_callback = class_config_llog_handler;
1254 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1256 CERROR("failed to start server %s: %d\n",
1257 lsi->lsi_svname, rc);
1258 /* Do NOT call server_deregister_mount() here. This makes it
1259 * impossible to find mount later in cleanup time and leaves
1260 * @lsi and othder stuff leaked. -umka */
1264 obd = class_name2obd(lsi->lsi_svname);
1266 CERROR("no server named %s was started\n", lsi->lsi_svname);
1267 GOTO(out_mgc, rc = -ENXIO);
1270 if (IS_OST(lsi) || IS_MDT(lsi)) {
1271 rc = lustre_start_lwp(sb);
1273 CERROR("%s: failed to start LWP: %d\n",
1274 lsi->lsi_svname, rc);
1279 server_notify_target(sb, obd);
1281 /* calculate recovery timeout, do it after lustre_process_log */
1282 server_calc_timeout(lsi, obd);
1284 /* log has been fully processed */
1285 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1287 /* log has been fully processed, let clients connect */
1288 dev = obd->obd_lu_dev;
1289 if (dev && dev->ld_ops->ldo_prepare) {
1290 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1292 struct lu_context session_ctx;
1294 lu_context_init(&session_ctx, LCT_SESSION);
1295 session_ctx.lc_thread = NULL;
1296 lu_context_enter(&session_ctx);
1297 env.le_ses = &session_ctx;
1299 rc = dev->ld_ops->ldo_prepare(&env, NULL, dev);
1302 lu_context_exit(&session_ctx);
1303 lu_context_fini(&session_ctx);
1307 /* abort recovery only on the complete stack:
1308 * many devices can be involved */
1309 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1310 (OBP(obd, iocontrol))) {
1311 obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
1316 /* Release the mgc fs for others to use */
1317 if (lsi->lsi_srv_mnt)
1318 server_mgc_clear_fs(lsi->lsi_mgc);
1322 server_stop_servers(lsi->lsi_flags);
1327 static int lsi_prepare(struct lustre_sb_info *lsi)
1334 LASSERT(lsi->lsi_lmd);
1336 /* The server name is given as a mount line option */
1337 if (lsi->lsi_lmd->lmd_profile == NULL) {
1338 LCONSOLE_ERROR("Can't determine server name\n");
1342 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
1343 RETURN(-ENAMETOOLONG);
1345 strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
1347 /* Determine osd type */
1348 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
1349 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
1350 sizeof(lsi->lsi_osd_type))
1351 RETURN(-ENAMETOOLONG);
1353 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
1355 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
1358 /* XXX: a temp. solution for components using fsfilt
1359 * to be removed in one of the subsequent patches */
1360 if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs"))
1361 strcpy(lsi->lsi_fstype, "ldiskfs");
1363 strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
1365 /* Determine server type */
1366 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1368 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
1369 /* Assume we're a bare MGS */
1371 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
1373 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
1378 lsi->lsi_flags |= rc;
1380 /* Add mount line flags that used to be in ldd:
1381 * writeconf, mgs, anything else?
1383 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
1384 LDD_F_WRITECONF : 0;
1385 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
1387 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_UPDATE) ?
1389 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
1390 LDD_F_SV_TYPE_MGS : 0;
1391 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
1392 LDD_F_NO_PRIMNODE : 0;
1397 /*************** server mount ******************/
1399 /** Start the shutdown of servers at umount.
1401 static void server_put_super(struct super_block *sb)
1403 struct lustre_sb_info *lsi = s2lsi(sb);
1404 struct obd_device *obd;
1405 char *tmpname, *extraname = NULL;
1407 int lsiflags = lsi->lsi_flags;
1410 LASSERT(IS_SERVER(lsi));
1412 tmpname_sz = strlen(lsi->lsi_svname) + 1;
1413 OBD_ALLOC(tmpname, tmpname_sz);
1414 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
1415 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1416 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1417 snprintf(tmpname, tmpname_sz, "MGS");
1419 /* disconnect the lwp first to drain off the inflight request */
1420 if (IS_OST(lsi) || IS_MDT(lsi)) {
1423 rc = lustre_disconnect_lwp(sb);
1424 if (rc && rc != ETIMEDOUT)
1425 CERROR("%s: failed to disconnect lwp. (rc=%d)\n",
1429 /* Stop the target */
1430 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1431 (IS_MDT(lsi) || IS_OST(lsi))) {
1432 struct lustre_profile *lprof = NULL;
1434 /* tell the mgc to drop the config log */
1435 lustre_end_log(sb, lsi->lsi_svname, NULL);
1437 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1438 If there are any setup/cleanup errors, save the lov
1439 name for safety cleanup later. */
1440 lprof = class_get_profile(lsi->lsi_svname);
1441 if (lprof && lprof->lp_dt) {
1442 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1443 strcpy(extraname, lprof->lp_dt);
1446 obd = class_name2obd(lsi->lsi_svname);
1448 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1449 if (lsiflags & LSI_UMOUNT_FAILOVER)
1451 /* We can't seem to give an error return code
1452 * to .put_super, so we better make sure we clean up! */
1454 class_manual_cleanup(obd);
1456 CERROR("no obd %s\n", lsi->lsi_svname);
1457 server_deregister_mount(lsi->lsi_svname);
1461 /* If they wanted the mgs to stop separately from the mdt, they
1462 should have put it on a different device. */
1464 /* if MDS start with --nomgs, don't stop MGS then */
1465 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1466 server_stop_mgs(sb);
1469 if (IS_OST(lsi) || IS_MDT(lsi)) {
1470 if (lustre_stop_lwp(sb) < 0)
1471 CERROR("%s: failed to stop lwp!\n", tmpname);
1474 /* Clean the mgc and sb */
1475 lustre_common_put_super(sb);
1477 /* wait till all in-progress cleanups are done
1478 * specifically we're interested in ofd cleanup
1480 obd_zombie_barrier();
1482 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1483 until the target is really gone so that our type refcount check
1485 server_stop_servers(lsiflags);
1487 /* In case of startup or cleanup err, stop related obds */
1489 obd = class_name2obd(extraname);
1491 CWARN("Cleaning orphaned obd %s\n", extraname);
1493 class_manual_cleanup(obd);
1495 OBD_FREE(extraname, strlen(extraname) + 1);
1498 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1499 OBD_FREE(tmpname, tmpname_sz);
1503 /** Called only for 'umount -f'
1505 static void server_umount_begin(struct super_block *sb)
1507 struct lustre_sb_info *lsi = s2lsi(sb);
1510 CDEBUG(D_MOUNT, "umount -f\n");
1511 /* umount = failover
1513 no third way to do non-force, non-failover */
1514 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1518 static int server_statfs(struct dentry *dentry, struct kstatfs *buf)
1520 struct super_block *sb = dentry->d_sb;
1521 struct lustre_sb_info *lsi = s2lsi(sb);
1522 struct obd_statfs statfs;
1526 if (lsi->lsi_dt_dev) {
1527 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
1529 statfs_unpack(buf, &statfs);
1530 buf->f_type = sb->s_magic;
1536 buf->f_type = sb->s_magic;
1537 buf->f_bsize = sb->s_blocksize;
1543 buf->f_namelen = NAME_MAX;
1547 /** The operations we support directly on the superblock:
1548 * mount, umount, and df.
1550 static struct super_operations server_ops = {
1551 .put_super = server_put_super,
1552 .umount_begin = server_umount_begin, /* umount -f */
1553 .statfs = server_statfs,
1557 * Xattr support for Lustre servers
1559 static ssize_t lustre_getxattr(struct dentry *dentry, const char *name,
1560 void *buffer, size_t size)
1562 if (!selinux_is_enabled())
1567 static int lustre_setxattr(struct dentry *dentry, const char *name,
1568 const void *value, size_t size, int flags)
1573 static ssize_t lustre_listxattr(struct dentry *d_entry, char *name,
1579 const struct inode_operations server_inode_operations = {
1580 .setxattr = lustre_setxattr,
1581 .getxattr = lustre_getxattr,
1582 .listxattr = lustre_listxattr,
1585 #define log2(n) ffz(~(n))
1586 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1588 static int server_fill_super_common(struct super_block *sb)
1590 struct inode *root = 0;
1593 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1595 sb->s_blocksize = 4096;
1596 sb->s_blocksize_bits = log2(sb->s_blocksize);
1597 sb->s_magic = LUSTRE_SUPER_MAGIC;
1598 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1599 sb->s_flags |= MS_RDONLY;
1600 sb->s_op = &server_ops;
1602 root = new_inode(sb);
1604 CERROR("Can't make root inode\n");
1608 /* returns -EIO for every operation */
1609 /* make_bad_inode(root); -- badness - can't umount */
1610 /* apparently we need to be a directory for the mount to finish */
1611 root->i_mode = S_IFDIR;
1612 root->i_op = &server_inode_operations;
1613 sb->s_root = d_make_root(root);
1615 CERROR("%s: can't make root dentry\n", sb->s_id);
1622 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
1624 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1625 struct obd_device *obd;
1626 struct dt_device_param p;
1632 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
1633 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
1635 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
1636 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
1637 strcat(lsi->lsi_osd_uuid, "_UUID");
1638 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
1640 obd = class_name2obd(lsi->lsi_osd_obdname);
1642 rc = lustre_start_simple(lsi->lsi_osd_obdname,
1644 lsi->lsi_osd_uuid, lmd->lmd_dev,
1645 flagstr, lsi->lsi_lmd->lmd_opts,
1649 obd = class_name2obd(lsi->lsi_osd_obdname);
1653 rc = obd_connect(NULL, &lsi->lsi_osd_exp,
1654 obd, &obd->obd_uuid, NULL, NULL);
1657 class_manual_cleanup(obd);
1658 lsi->lsi_dt_dev = NULL;
1662 LASSERT(obd->obd_lu_dev);
1663 lu_device_get(obd->obd_lu_dev);
1664 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
1665 LASSERT(lsi->lsi_dt_dev);
1667 /* set disk context for llog usage */
1668 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1669 obd->obd_lvfs_ctxt.dt = lsi->lsi_dt_dev;
1671 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
1673 lsi->lsi_srv_mnt = p.ddp_mnt;
1679 /** Fill in the superblock info for a Lustre server.
1680 * Mount the device with the correct options.
1681 * Read the on-disk config file.
1682 * Start the services.
1684 int server_fill_super(struct super_block *sb)
1686 struct lustre_sb_info *lsi = s2lsi(sb);
1690 rc = lsi_prepare(lsi);
1694 /* Start low level OSD */
1695 rc = osd_start(lsi, sb->s_flags);
1697 CERROR("Unable to start osd on %s: %d\n",
1698 lsi->lsi_lmd->lmd_dev, rc);
1703 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
1704 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
1706 if (class_name2obd(lsi->lsi_svname)) {
1707 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1708 "running. Double-mount may have compromised"
1709 " the disk journal.\n",
1715 /* Start MGS before MGC */
1716 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1717 rc = server_start_mgs(sb);
1722 /* Start MGC before servers */
1723 rc = lustre_start_mgc(sb);
1727 /* Set up all obd devices for service */
1728 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1729 (IS_OST(lsi) || IS_MDT(lsi))) {
1730 rc = server_start_targets(sb, lsi->lsi_srv_mnt);
1732 CERROR("Unable to start targets: %d\n", rc);
1735 /* FIXME overmount client here, or can we just start a
1736 * client log and client_fill_super on this sb? We
1737 * need to make sure server_put_super gets called too
1738 * - ll_put_super calls lustre_common_put_super; check
1739 * there for LSI_SERVER flag, call s_p_s if so.
1741 * Probably should start client from new thread so we
1742 * can return. Client will not finish until all
1743 * servers are connected. Note - MGS-only server does
1744 * NOT get a client, since there is no lustre fs
1745 * associated - the MGS is for all lustre fs's */
1748 rc = server_fill_super_common(sb);
1754 /* We jump here in case of failure while starting targets or MGS.
1755 * In this case we can't just put @mnt and have to do real cleanup
1756 * with stoping targets, etc. */
1757 server_put_super(sb);
1762 * Calculate timeout value for a target.
1764 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1766 struct lustre_mount_data *lmd;
1770 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
1771 int min = OBD_RECOVERY_TIME_MIN;
1773 LASSERT(IS_SERVER(lsi));
1777 soft = lmd->lmd_recovery_time_soft;
1778 hard = lmd->lmd_recovery_time_hard;
1779 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1780 obd->obd_no_ir = !has_ir;
1784 soft = OBD_RECOVERY_TIME_SOFT;
1786 hard = OBD_RECOVERY_TIME_HARD;
1788 /* target may have ir_factor configured. */
1789 factor = OBD_IR_FACTOR_DEFAULT;
1790 if (obd->obd_recovery_ir_factor)
1791 factor = obd->obd_recovery_ir_factor;
1794 int new_soft = soft;
1795 int new_hard = hard;
1797 /* adjust timeout value by imperative recovery */
1799 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1800 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1802 /* make sure the timeout is not too short */
1803 new_soft = max(min, new_soft);
1804 new_hard = max(new_soft, new_hard);
1806 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1807 "window shrunk from %d-%d down to %d-%d\n",
1808 obd->obd_name, soft, hard, new_soft, new_hard);
1815 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1816 obd->obd_recovery_time_hard = hard;
1817 obd->obd_recovery_ir_factor = factor;
1819 EXPORT_SYMBOL(server_calc_timeout);