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)
89 struct lustre_mount_info *lmi;
95 OBD_ALLOC(lmi, sizeof(*lmi));
98 OBD_ALLOC(name_cp, strlen(name) + 1);
100 OBD_FREE(lmi, sizeof(*lmi));
103 strcpy(name_cp, name);
105 mutex_lock(&lustre_mount_info_lock);
107 if (server_find_mount(name)) {
108 mutex_unlock(&lustre_mount_info_lock);
109 OBD_FREE(lmi, sizeof(*lmi));
110 OBD_FREE(name_cp, strlen(name) + 1);
111 CERROR("Already registered %s\n", name);
114 lmi->lmi_name = name_cp;
116 cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list);
118 mutex_unlock(&lustre_mount_info_lock);
120 CDEBUG(D_MOUNT, "register mount %p from %s\n", sb, name);
125 /* when an obd no longer needs a mount */
126 static int server_deregister_mount(const char *name)
128 struct lustre_mount_info *lmi;
131 mutex_lock(&lustre_mount_info_lock);
132 lmi = server_find_mount(name);
134 mutex_unlock(&lustre_mount_info_lock);
135 CERROR("%s not registered\n", name);
139 CDEBUG(D_MOUNT, "deregister mount %p from %s\n", lmi->lmi_sb, name);
141 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
142 cfs_list_del(&lmi->lmi_list_chain);
143 OBD_FREE(lmi, sizeof(*lmi));
144 mutex_unlock(&lustre_mount_info_lock);
149 /* obd's look up a registered mount using their obdname. This is just
150 for initial obd setup to find the mount struct. It should not be
151 called every time you want to mntget. */
152 struct lustre_mount_info *server_get_mount(const char *name)
154 struct lustre_mount_info *lmi;
155 struct lustre_sb_info *lsi;
158 mutex_lock(&lustre_mount_info_lock);
159 lmi = server_find_mount(name);
160 mutex_unlock(&lustre_mount_info_lock);
162 CERROR("Can't find mount for %s\n", name);
165 lsi = s2lsi(lmi->lmi_sb);
167 cfs_atomic_inc(&lsi->lsi_mounts);
169 CDEBUG(D_MOUNT, "get mount %p from %s, refs=%d\n", lmi->lmi_sb,
170 name, cfs_atomic_read(&lsi->lsi_mounts));
174 EXPORT_SYMBOL(server_get_mount);
177 * Used by mdt to get mount_info from obdname.
178 * There are no blocking when using the mount_info.
179 * Do not use server_get_mount for this purpose.
181 struct lustre_mount_info *server_get_mount_2(const char *name)
183 struct lustre_mount_info *lmi;
186 mutex_lock(&lustre_mount_info_lock);
187 lmi = server_find_mount(name);
188 mutex_unlock(&lustre_mount_info_lock);
190 CERROR("Can't find mount for %s\n", name);
194 EXPORT_SYMBOL(server_get_mount_2);
196 /* to be called from obd_cleanup methods */
197 int server_put_mount(const char *name)
199 struct lustre_mount_info *lmi;
200 struct lustre_sb_info *lsi;
203 mutex_lock(&lustre_mount_info_lock);
204 lmi = server_find_mount(name);
205 mutex_unlock(&lustre_mount_info_lock);
207 CERROR("Can't find mount for %s\n", name);
210 lsi = s2lsi(lmi->lmi_sb);
212 CDEBUG(D_MOUNT, "put mount %p from %s, refs=%d\n",
213 lmi->lmi_sb, name, cfs_atomic_read(&lsi->lsi_mounts));
215 if (lustre_put_lsi(lmi->lmi_sb))
216 CDEBUG(D_MOUNT, "Last put of mount %p from %s\n",
219 /* this obd should never need the mount again */
220 server_deregister_mount(name);
224 EXPORT_SYMBOL(server_put_mount);
226 /* Corresponding to server_get_mount_2 */
227 int server_put_mount_2(const char *name, struct vfsmount *mnt)
232 EXPORT_SYMBOL(server_put_mount_2);
234 /* Set up a MGS to serve startup logs */
235 static int server_start_mgs(struct super_block *sb)
237 struct lustre_sb_info *lsi = s2lsi(sb);
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);
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_device *obd;
568 char *lwpname = NULL;
569 char *lwpuuid = NULL;
573 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
576 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
580 OBD_ALLOC(lwpname, MTI_NAME_MAXLEN);
582 GOTO(out, rc = -ENOMEM);
584 rc = tgt_name2lwpname(lsi->lsi_svname, lwpname);
586 CERROR("%s: failed to generate lwp name. %d\n",
587 lsi->lsi_svname, rc);
591 OBD_ALLOC(lwpuuid, MTI_NAME_MAXLEN);
593 GOTO(out, rc = -ENOMEM);
595 sprintf(lwpuuid, "%s_UUID", lwpname);
596 rc = lustre_start_simple(lwpname, LUSTRE_LWP_NAME,
597 lwpuuid, lustre_cfg_string(lcfg, 1),
600 CERROR("%s: setup up failed: rc %d\n", lwpname, rc);
604 obd = class_name2obd(lwpname);
605 LASSERT(obd != NULL);
607 rc = lustre_lwp_connect(obd);
609 CERROR("%s: connect failed: rc = %d\n", lwpname, rc);
612 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
614 OBD_FREE(lwpuuid, MTI_NAME_MAXLEN);
619 /* the caller is responsible for memory free */
620 static struct obd_device *lustre_find_lwp(struct lustre_sb_info *lsi,
621 char **lwpname, char **logname)
623 struct obd_device *lwp;
627 LASSERT(lwpname != NULL);
628 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
630 OBD_ALLOC(*lwpname, MTI_NAME_MAXLEN);
631 if (*lwpname == NULL)
632 RETURN(ERR_PTR(-ENOMEM));
634 if (logname != NULL) {
635 OBD_ALLOC(*logname, MTI_NAME_MAXLEN);
636 if (*logname == NULL)
637 GOTO(out, rc = -ENOMEM);
638 rc = server_name2fsname(lsi->lsi_svname, *lwpname, NULL);
640 CERROR("%s: failed to get fsname from svname. %d\n",
641 lsi->lsi_svname, rc);
642 GOTO(out, rc = -EINVAL);
644 sprintf(*logname, "%s-client", *lwpname);
647 rc = tgt_name2lwpname(lsi->lsi_svname, *lwpname);
649 CERROR("%s: failed to generate lwp name. %d\n",
650 lsi->lsi_svname, rc);
651 GOTO(out, rc = -EINVAL);
654 lwp = class_name2obd(*lwpname);
658 if (*lwpname != NULL) {
659 OBD_FREE(*lwpname, MTI_NAME_MAXLEN);
662 if (logname != NULL && *logname != NULL) {
663 OBD_FREE(*logname, MTI_NAME_MAXLEN);
669 RETURN(lwp != NULL ? lwp : ERR_PTR(-ENOENT));
672 static int lustre_lwp_add_conn(struct lustre_cfg *cfg,
673 struct lustre_sb_info *lsi)
675 struct lustre_cfg_bufs *bufs = NULL;
676 struct lustre_cfg *lcfg = NULL;
677 char *lwpname = NULL;
678 struct obd_device *lwp;
682 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
684 CERROR("%s: can't find lwp device.\n", lsi->lsi_svname);
685 GOTO(out, rc = PTR_ERR(lwp));
687 LASSERT(lwpname != NULL);
691 GOTO(out, rc = -ENOMEM);
693 lustre_cfg_bufs_reset(bufs, lwpname);
694 lustre_cfg_bufs_set_string(bufs, 1,
695 lustre_cfg_string(cfg, 1));
697 lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
699 rc = class_add_conn(lwp, lcfg);
701 CERROR("%s: can't add conn: rc = %d\n", lwpname, rc);
707 lustre_cfg_free(lcfg);
709 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
714 * Retrieve MDT nids from the client log, then start the lwp device.
715 * there are only two scenarios which would include mdt nid.
717 * marker 5 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add mdc' xxx-
718 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
719 * attach 0:lustre-MDT0000-mdc 1:mdc 2:lustre-clilmv_UUID
720 * setup 0:lustre-MDT0000-mdc 1:lustre-MDT0000_UUID 2:192.168.122.162@tcp
721 * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
722 * add_conn 0:lustre-MDT0000-mdc 1:192.168.172.1@tcp
723 * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID xxxx
724 * marker 5 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add mdc' xxxx-
726 * marker 7 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
727 * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
728 * add_conn 0:lustre-MDT0000-mdc 1:192.168.122.2@tcp
729 * marker 7 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
731 static int client_lwp_config_process(const struct lu_env *env,
732 struct llog_handle *handle,
733 struct llog_rec_hdr *rec, void *data)
735 struct config_llog_instance *clli = data;
736 int cfg_len = rec->lrh_len;
737 char *cfg_buf = (char *) (rec + 1);
738 struct lustre_cfg *lcfg = NULL;
739 struct lustre_sb_info *lsi;
740 int rc = 0, swab = 0;
743 if (rec->lrh_type != OBD_CFG_REC) {
744 CERROR("Unknown llog record type %#x encountered\n",
749 LASSERT(clli->cfg_sb != NULL);
750 lsi = s2lsi(clli->cfg_sb);
752 lcfg = (struct lustre_cfg *)cfg_buf;
753 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
754 lustre_swab_lustre_cfg(lcfg);
758 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
762 switch (lcfg->lcfg_command) {
764 struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
766 lustre_swab_cfg_marker(marker, swab,
767 LUSTRE_CFG_BUFLEN(lcfg, 1));
768 if (marker->cm_flags & CM_SKIP ||
769 marker->cm_flags & CM_EXCLUDE)
772 if (!tgt_is_mdt0(marker->cm_tgtname))
775 /* Don't try to connect old MDT server without LWP support,
776 * otherwise, the old MDT could regard this LWP client as
777 * a normal client and save the export on disk for recovery.
779 * This usually happen when rolling upgrade. LU-3929 */
780 if (marker->cm_vers < OBD_OCD_VERSION(2, 3, 60, 0))
783 if (!strncmp(marker->cm_comment, "add mdc", 7) ||
784 !strncmp(marker->cm_comment, "add failnid", 11)) {
785 if (marker->cm_flags & CM_START) {
786 clli->cfg_flags = CFG_F_MARKER;
787 /* This hack is to differentiate the
788 * ADD_UUID is come from "add mdc" record
789 * or from "add failnid" record. */
790 if (!strncmp(marker->cm_comment,
792 clli->cfg_flags |= CFG_F_SKIP;
793 } else if (marker->cm_flags & CM_END) {
799 case LCFG_ADD_UUID: {
800 if (clli->cfg_flags == CFG_F_MARKER) {
801 rc = lustre_lwp_setup(lcfg, lsi);
802 /* XXX: process only the first nid as
803 * we don't need another instance of lwp */
804 clli->cfg_flags |= CFG_F_SKIP;
805 } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
806 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
809 CERROR("%s: Fail to add uuid, rc:%d\n",
810 lsi->lsi_svname, rc);
814 case LCFG_ADD_CONN: {
815 if (is_mdc_for_mdt0(lustre_cfg_string(lcfg, 0)) &&
816 (clli->cfg_flags & CFG_F_MARKER) != 0) {
817 rc = lustre_lwp_add_conn(lcfg, lsi);
818 /* When the 'add mdc' record is old (< 2.3.60) but
819 * 'add failnid' record is new (>= 2.3.60), add
820 * connection should fail with -ENOENT since LWP
821 * device wasn't setup, we'd ignore such error. */
822 if (rc == -ENOENT && clli->cfg_flags & CFG_F_SKIP)
834 static int lustre_disconnect_lwp(struct super_block *sb)
836 struct lustre_sb_info *lsi = s2lsi(sb);
837 struct obd_device *lwp;
838 char *lwpname = NULL;
839 char *logname = NULL;
840 struct lustre_cfg *lcfg = NULL;
841 struct lustre_cfg_bufs *bufs = NULL;
842 struct config_llog_instance *cfg = NULL;
846 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
847 if (IS_ERR(lwp) && PTR_ERR(lwp) != -ENOENT)
848 GOTO(out, rc = PTR_ERR(lwp));
850 LASSERT(lwpname != NULL);
851 LASSERT(logname != NULL);
855 GOTO(out, rc = -ENOMEM);
858 cfg->cfg_instance = sb;
859 rc = lustre_end_log(sb, logname, cfg);
861 CERROR("%s: Can't end config log %s.\n", lwpname, logname);
865 if (PTR_ERR(lwp) == -ENOENT) {
866 CDEBUG(D_CONFIG, "%s: lwp device wasn't started.\n",
873 GOTO(out, rc = -ENOMEM);
875 lustre_cfg_bufs_reset(bufs, lwp->obd_name);
876 lustre_cfg_bufs_set_string(bufs, 1, NULL);
877 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
879 GOTO(out, rc = -ENOMEM);
881 /* Disconnect import first. NULL is passed for the '@env', since
882 * it will not be used. */
883 rc = lwp->obd_lu_dev->ld_ops->ldo_process_config(NULL, lwp->obd_lu_dev,
887 lustre_cfg_free(lcfg);
893 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
895 OBD_FREE(logname, MTI_NAME_MAXLEN);
900 * Stop the lwp for an OST/MDT target.
902 static int lustre_stop_lwp(struct super_block *sb)
904 struct lustre_sb_info *lsi = s2lsi(sb);
905 struct obd_device *lwp = NULL;
906 char *lwpname = NULL;
910 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
912 CDEBUG(PTR_ERR(lwp) == -ENOENT ? D_CONFIG : D_ERROR,
913 "%s: lwp wasn't started.\n", lsi->lsi_svname);
918 rc = class_manual_cleanup(lwp);
922 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
927 * Start the lwp(fsname-MDT0000-lwp-OSTxxxx) for an OST or MDT target,
928 * which would be used to establish connection from OST to MDT0.
930 static int lustre_start_lwp(struct super_block *sb)
932 struct lustre_sb_info *lsi = s2lsi(sb);
933 struct config_llog_instance *cfg = NULL;
934 struct obd_device *lwp;
935 char *lwpname = NULL;
936 char *logname = NULL;
940 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
942 /* the lwp device already stared */
943 if (lwp && !IS_ERR(lwp))
946 if (PTR_ERR(lwp) != -ENOENT)
947 GOTO(out, rc = PTR_ERR(lwp));
949 LASSERT(lwpname != NULL);
950 LASSERT(logname != NULL);
954 GOTO(out, rc = -ENOMEM);
956 cfg->cfg_callback = client_lwp_config_process;
957 cfg->cfg_instance = sb;
959 rc = lustre_process_log(sb, logname, cfg);
962 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
964 OBD_FREE(logname, MTI_NAME_MAXLEN);
970 DEFINE_MUTEX(server_start_lock);
972 /* Stop MDS/OSS if nobody is using them */
973 static int server_stop_servers(int lsiflags)
975 struct obd_device *obd = NULL;
976 struct obd_type *type = NULL;
980 mutex_lock(&server_start_lock);
982 /* Either an MDT or an OST or neither */
983 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
984 if (lsiflags & LDD_F_SV_TYPE_MDT) {
985 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
987 type = class_search_type(LUSTRE_MDT_NAME);
990 /* if this was an OST, and there are no more OST's, clean up the OSS */
991 if (lsiflags & LDD_F_SV_TYPE_OST) {
992 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
994 type = class_search_type(LUSTRE_OST_NAME);
997 if (obd != NULL && (type == NULL || type->typ_refcnt == 0)) {
999 /* obd_fail doesn't mean much on a server obd */
1000 rc = class_manual_cleanup(obd);
1003 mutex_unlock(&server_start_lock);
1008 int server_mti_print(const char *title, struct mgs_target_info *mti)
1010 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
1011 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
1012 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
1013 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
1014 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
1015 mti->mti_config_ver, mti->mti_flags);
1018 EXPORT_SYMBOL(server_mti_print);
1020 /* Generate data for registration */
1021 static int server_lsi2mti(struct lustre_sb_info *lsi,
1022 struct mgs_target_info *mti)
1024 lnet_process_id_t id;
1029 if (!IS_SERVER(lsi))
1032 if (strlcpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname))
1033 >= sizeof(mti->mti_svname))
1036 mti->mti_nid_count = 0;
1037 while (LNetGetId(i++, &id) != -ENOENT) {
1038 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1041 /* server use --servicenode param, only allow specified
1042 * nids be registered */
1043 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
1044 class_match_nid(lsi->lsi_lmd->lmd_params,
1045 PARAM_FAILNODE, id.nid) < 1)
1048 /* match specified network */
1049 if (!class_match_net(lsi->lsi_lmd->lmd_params,
1050 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1053 mti->mti_nids[mti->mti_nid_count] = id.nid;
1054 mti->mti_nid_count++;
1055 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1056 CWARN("Only using first %d nids for %s\n",
1057 mti->mti_nid_count, mti->mti_svname);
1062 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1063 mti->mti_config_ver = 0;
1065 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1069 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1072 /* Orion requires index to be set */
1073 LASSERT(!(rc & LDD_F_NEED_INDEX));
1074 /* keep only LDD flags */
1075 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1076 if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN))
1077 mti->mti_flags |= LDD_F_UPDATE;
1078 cplen = strlcpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1079 sizeof(mti->mti_params));
1080 if (cplen >= sizeof(mti->mti_params))
1085 /* Register an old or new target with the MGS. If needed MGS will construct
1086 startup logs and assign index */
1087 static int server_register_target(struct lustre_sb_info *lsi)
1089 struct obd_device *mgc = lsi->lsi_mgc;
1090 struct mgs_target_info *mti = NULL;
1097 if (!IS_SERVER(lsi))
1104 rc = server_lsi2mti(lsi, mti);
1108 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1109 mti->mti_svname, mti->mti_fsname,
1110 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1113 /* if write_conf is true, the registration must succeed */
1114 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1115 mti->mti_flags |= LDD_F_OPC_REG;
1117 /* Register the target */
1118 /* FIXME use mgc_process_config instead */
1119 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1120 sizeof(KEY_REGISTER_TARGET),
1121 KEY_REGISTER_TARGET,
1122 sizeof(*mti), mti, NULL);
1124 if (mti->mti_flags & LDD_F_ERROR) {
1125 LCONSOLE_ERROR_MSG(0x160,
1126 "%s: the MGS refuses to allow this server "
1127 "to start: rc = %d. Please see messages on "
1128 "the MGS.\n", lsi->lsi_svname, rc);
1129 } else if (writeconf) {
1130 LCONSOLE_ERROR_MSG(0x15f,
1131 "%s: cannot register this server with the MGS: "
1132 "rc = %d. Is the MGS running?\n",
1133 lsi->lsi_svname, rc);
1135 CERROR("%s: error registering with the MGS: rc = %d "
1136 "(not fatal)\n", lsi->lsi_svname, rc);
1137 /* reset the error code for non-fatal error. */
1150 * Notify the MGS that this target is ready.
1151 * Used by IR - if the MGS receives this message, it will notify clients.
1153 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1155 struct lustre_sb_info *lsi = s2lsi(sb);
1156 struct obd_device *mgc = lsi->lsi_mgc;
1157 struct mgs_target_info *mti = NULL;
1163 if (!(IS_SERVER(lsi)))
1169 rc = server_lsi2mti(lsi, mti);
1173 mti->mti_instance = obd->u.obt.obt_instance;
1174 mti->mti_flags |= LDD_F_OPC_READY;
1176 /* FIXME use mgc_process_config instead */
1177 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1178 sizeof(KEY_REGISTER_TARGET),
1179 KEY_REGISTER_TARGET,
1180 sizeof(*mti), mti, NULL);
1182 /* Imperative recovery: if the mgs informs us to use IR? */
1183 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1184 (mti->mti_flags & LDD_F_IR_CAPABLE))
1185 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1194 /** Start server targets: MDTs and OSTs
1196 static int server_start_targets(struct super_block *sb)
1198 struct obd_device *obd;
1199 struct lustre_sb_info *lsi = s2lsi(sb);
1200 struct config_llog_instance cfg;
1202 struct lu_device *dev;
1206 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1209 /* make sure the MDS is started */
1210 mutex_lock(&server_start_lock);
1211 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1213 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1215 LUSTRE_MDS_OBDNAME"_uuid",
1218 mutex_unlock(&server_start_lock);
1219 CERROR("failed to start MDS: %d\n", rc);
1223 mutex_unlock(&server_start_lock);
1226 /* If we're an OST, make sure the global OSS is running */
1228 /* make sure OSS is started */
1229 mutex_lock(&server_start_lock);
1230 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1232 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1234 LUSTRE_OSS_OBDNAME"_uuid",
1237 mutex_unlock(&server_start_lock);
1238 CERROR("failed to start OSS: %d\n", rc);
1242 mutex_unlock(&server_start_lock);
1245 /* Set the mgc fs to our server disk. This allows the MGC to
1246 * read and write configs locally, in case it can't talk to the MGS. */
1247 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1249 GOTO(out_stop_service, rc);
1251 /* Register with MGS */
1252 rc = server_register_target(lsi);
1256 /* Let the target look up the mount using the target's name
1257 (we can't pass the sb or mnt through class_process_config.) */
1258 rc = server_register_mount(lsi->lsi_svname, sb);
1262 /* Start targets using the llog named for the target */
1263 memset(&cfg, 0, sizeof(cfg));
1264 cfg.cfg_callback = class_config_llog_handler;
1265 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1267 CERROR("failed to start server %s: %d\n",
1268 lsi->lsi_svname, rc);
1269 /* Do NOT call server_deregister_mount() here. This makes it
1270 * impossible to find mount later in cleanup time and leaves
1271 * @lsi and othder stuff leaked. -umka */
1275 obd = class_name2obd(lsi->lsi_svname);
1277 CERROR("no server named %s was started\n", lsi->lsi_svname);
1278 GOTO(out_mgc, rc = -ENXIO);
1281 if (IS_OST(lsi) || IS_MDT(lsi)) {
1282 rc = lustre_start_lwp(sb);
1284 CERROR("%s: failed to start LWP: %d\n",
1285 lsi->lsi_svname, rc);
1290 server_notify_target(sb, obd);
1292 /* calculate recovery timeout, do it after lustre_process_log */
1293 server_calc_timeout(lsi, obd);
1295 /* log has been fully processed */
1296 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1298 /* log has been fully processed, let clients connect */
1299 dev = obd->obd_lu_dev;
1300 if (dev && dev->ld_ops->ldo_prepare) {
1301 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1303 struct lu_context session_ctx;
1305 lu_context_init(&session_ctx, LCT_SESSION);
1306 session_ctx.lc_thread = NULL;
1307 lu_context_enter(&session_ctx);
1308 env.le_ses = &session_ctx;
1310 rc = dev->ld_ops->ldo_prepare(&env, NULL, dev);
1313 lu_context_exit(&session_ctx);
1314 lu_context_fini(&session_ctx);
1318 /* abort recovery only on the complete stack:
1319 * many devices can be involved */
1320 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1321 (OBP(obd, iocontrol))) {
1322 obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
1327 /* Release the mgc fs for others to use */
1328 server_mgc_clear_fs(lsi->lsi_mgc);
1332 server_stop_servers(lsi->lsi_flags);
1337 static int lsi_prepare(struct lustre_sb_info *lsi)
1344 LASSERT(lsi->lsi_lmd);
1346 /* The server name is given as a mount line option */
1347 if (lsi->lsi_lmd->lmd_profile == NULL) {
1348 LCONSOLE_ERROR("Can't determine server name\n");
1352 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
1353 RETURN(-ENAMETOOLONG);
1355 strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
1357 /* Determine osd type */
1358 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
1359 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
1360 sizeof(lsi->lsi_osd_type))
1361 RETURN(-ENAMETOOLONG);
1363 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
1365 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
1368 /* XXX: a temp. solution for components using fsfilt
1369 * to be removed in one of the subsequent patches */
1370 if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs"))
1371 strcpy(lsi->lsi_fstype, "ldiskfs");
1373 strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
1375 /* Determine server type */
1376 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1378 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
1379 /* Assume we're a bare MGS */
1381 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
1383 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
1388 lsi->lsi_flags |= rc;
1390 /* Add mount line flags that used to be in ldd:
1391 * writeconf, mgs, anything else?
1393 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
1394 LDD_F_WRITECONF : 0;
1395 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
1397 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_UPDATE) ?
1399 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
1400 LDD_F_SV_TYPE_MGS : 0;
1401 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
1402 LDD_F_NO_PRIMNODE : 0;
1407 /*************** server mount ******************/
1409 /** Start the shutdown of servers at umount.
1411 static void server_put_super(struct super_block *sb)
1413 struct lustre_sb_info *lsi = s2lsi(sb);
1414 struct obd_device *obd;
1415 char *tmpname, *extraname = NULL;
1417 int lsiflags = lsi->lsi_flags;
1420 LASSERT(IS_SERVER(lsi));
1422 tmpname_sz = strlen(lsi->lsi_svname) + 1;
1423 OBD_ALLOC(tmpname, tmpname_sz);
1424 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
1425 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1426 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1427 snprintf(tmpname, tmpname_sz, "MGS");
1429 /* disconnect the lwp first to drain off the inflight request */
1430 if (IS_OST(lsi) || IS_MDT(lsi)) {
1433 rc = lustre_disconnect_lwp(sb);
1434 if (rc && rc != ETIMEDOUT)
1435 CERROR("%s: failed to disconnect lwp. (rc=%d)\n",
1439 /* Stop the target */
1440 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1441 (IS_MDT(lsi) || IS_OST(lsi))) {
1442 struct lustre_profile *lprof = NULL;
1444 /* tell the mgc to drop the config log */
1445 lustre_end_log(sb, lsi->lsi_svname, NULL);
1447 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1448 If there are any setup/cleanup errors, save the lov
1449 name for safety cleanup later. */
1450 lprof = class_get_profile(lsi->lsi_svname);
1451 if (lprof && lprof->lp_dt) {
1452 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1453 strcpy(extraname, lprof->lp_dt);
1456 obd = class_name2obd(lsi->lsi_svname);
1458 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1459 if (lsiflags & LSI_UMOUNT_FAILOVER)
1461 /* We can't seem to give an error return code
1462 * to .put_super, so we better make sure we clean up! */
1464 class_manual_cleanup(obd);
1466 CERROR("no obd %s\n", lsi->lsi_svname);
1467 server_deregister_mount(lsi->lsi_svname);
1471 /* If they wanted the mgs to stop separately from the mdt, they
1472 should have put it on a different device. */
1474 /* if MDS start with --nomgs, don't stop MGS then */
1475 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1476 server_stop_mgs(sb);
1479 if (IS_OST(lsi) || IS_MDT(lsi)) {
1480 if (lustre_stop_lwp(sb) < 0)
1481 CERROR("%s: failed to stop lwp!\n", tmpname);
1484 /* Clean the mgc and sb */
1485 lustre_common_put_super(sb);
1487 /* wait till all in-progress cleanups are done
1488 * specifically we're interested in ofd cleanup
1490 obd_zombie_barrier();
1492 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1493 until the target is really gone so that our type refcount check
1495 server_stop_servers(lsiflags);
1497 /* In case of startup or cleanup err, stop related obds */
1499 obd = class_name2obd(extraname);
1501 CWARN("Cleaning orphaned obd %s\n", extraname);
1503 class_manual_cleanup(obd);
1505 OBD_FREE(extraname, strlen(extraname) + 1);
1508 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1509 OBD_FREE(tmpname, tmpname_sz);
1513 /** Called only for 'umount -f'
1515 static void server_umount_begin(struct super_block *sb)
1517 struct lustre_sb_info *lsi = s2lsi(sb);
1520 CDEBUG(D_MOUNT, "umount -f\n");
1521 /* umount = failover
1523 no third way to do non-force, non-failover */
1524 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1528 static int server_statfs(struct dentry *dentry, struct kstatfs *buf)
1530 struct super_block *sb = dentry->d_sb;
1531 struct lustre_sb_info *lsi = s2lsi(sb);
1532 struct obd_statfs statfs;
1536 if (lsi->lsi_dt_dev) {
1537 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
1539 statfs_unpack(buf, &statfs);
1540 buf->f_type = sb->s_magic;
1546 buf->f_type = sb->s_magic;
1547 buf->f_bsize = sb->s_blocksize;
1553 buf->f_namelen = NAME_MAX;
1557 /** The operations we support directly on the superblock:
1558 * mount, umount, and df.
1560 static struct super_operations server_ops = {
1561 .put_super = server_put_super,
1562 .umount_begin = server_umount_begin, /* umount -f */
1563 .statfs = server_statfs,
1567 * Xattr support for Lustre servers
1569 static ssize_t lustre_getxattr(struct dentry *dentry, const char *name,
1570 void *buffer, size_t size)
1572 if (!selinux_is_enabled())
1577 static int lustre_setxattr(struct dentry *dentry, const char *name,
1578 const void *value, size_t size, int flags)
1583 static ssize_t lustre_listxattr(struct dentry *d_entry, char *name,
1589 const struct inode_operations server_inode_operations = {
1590 .setxattr = lustre_setxattr,
1591 .getxattr = lustre_getxattr,
1592 .listxattr = lustre_listxattr,
1595 #define log2(n) ffz(~(n))
1596 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1598 static int server_fill_super_common(struct super_block *sb)
1600 struct inode *root = 0;
1603 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1605 sb->s_blocksize = 4096;
1606 sb->s_blocksize_bits = log2(sb->s_blocksize);
1607 sb->s_magic = LUSTRE_SUPER_MAGIC;
1608 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1609 sb->s_flags |= MS_RDONLY;
1610 sb->s_op = &server_ops;
1612 root = new_inode(sb);
1614 CERROR("Can't make root inode\n");
1618 /* returns -EIO for every operation */
1619 /* make_bad_inode(root); -- badness - can't umount */
1620 /* apparently we need to be a directory for the mount to finish */
1621 root->i_mode = S_IFDIR;
1622 root->i_op = &server_inode_operations;
1623 sb->s_root = d_make_root(root);
1625 CERROR("%s: can't make root dentry\n", sb->s_id);
1632 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
1634 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1635 struct obd_device *obd;
1636 struct dt_device_param p;
1642 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
1643 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
1645 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
1646 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
1647 strcat(lsi->lsi_osd_uuid, "_UUID");
1648 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
1650 obd = class_name2obd(lsi->lsi_osd_obdname);
1652 rc = lustre_start_simple(lsi->lsi_osd_obdname,
1654 lsi->lsi_osd_uuid, lmd->lmd_dev,
1655 flagstr, lsi->lsi_lmd->lmd_opts,
1659 obd = class_name2obd(lsi->lsi_osd_obdname);
1663 rc = obd_connect(NULL, &lsi->lsi_osd_exp,
1664 obd, &obd->obd_uuid, NULL, NULL);
1667 class_manual_cleanup(obd);
1668 lsi->lsi_dt_dev = NULL;
1672 LASSERT(obd->obd_lu_dev);
1673 lu_device_get(obd->obd_lu_dev);
1674 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
1675 LASSERT(lsi->lsi_dt_dev);
1677 /* set disk context for llog usage */
1678 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1679 obd->obd_lvfs_ctxt.dt = lsi->lsi_dt_dev;
1681 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
1686 /** Fill in the superblock info for a Lustre server.
1687 * Mount the device with the correct options.
1688 * Read the on-disk config file.
1689 * Start the services.
1691 int server_fill_super(struct super_block *sb)
1693 struct lustre_sb_info *lsi = s2lsi(sb);
1697 rc = lsi_prepare(lsi);
1701 /* Start low level OSD */
1702 rc = osd_start(lsi, sb->s_flags);
1704 CERROR("Unable to start osd on %s: %d\n",
1705 lsi->lsi_lmd->lmd_dev, rc);
1710 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
1711 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
1713 if (class_name2obd(lsi->lsi_svname)) {
1714 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1715 "running. Double-mount may have compromised"
1716 " the disk journal.\n",
1722 /* Start MGS before MGC */
1723 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1724 rc = server_start_mgs(sb);
1729 /* Start MGC before servers */
1730 rc = lustre_start_mgc(sb);
1734 /* Set up all obd devices for service */
1735 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1736 (IS_OST(lsi) || IS_MDT(lsi))) {
1737 rc = server_start_targets(sb);
1739 CERROR("Unable to start targets: %d\n", rc);
1742 /* FIXME overmount client here, or can we just start a
1743 * client log and client_fill_super on this sb? We
1744 * need to make sure server_put_super gets called too
1745 * - ll_put_super calls lustre_common_put_super; check
1746 * there for LSI_SERVER flag, call s_p_s if so.
1748 * Probably should start client from new thread so we
1749 * can return. Client will not finish until all
1750 * servers are connected. Note - MGS-only server does
1751 * NOT get a client, since there is no lustre fs
1752 * associated - the MGS is for all lustre fs's */
1755 rc = server_fill_super_common(sb);
1761 /* We jump here in case of failure while starting targets or MGS.
1762 * In this case we can't just put @mnt and have to do real cleanup
1763 * with stoping targets, etc. */
1764 server_put_super(sb);
1769 * Calculate timeout value for a target.
1771 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1773 struct lustre_mount_data *lmd;
1777 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
1778 int min = OBD_RECOVERY_TIME_MIN;
1780 LASSERT(IS_SERVER(lsi));
1784 soft = lmd->lmd_recovery_time_soft;
1785 hard = lmd->lmd_recovery_time_hard;
1786 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1787 obd->obd_no_ir = !has_ir;
1791 soft = OBD_RECOVERY_TIME_SOFT;
1793 hard = OBD_RECOVERY_TIME_HARD;
1795 /* target may have ir_factor configured. */
1796 factor = OBD_IR_FACTOR_DEFAULT;
1797 if (obd->obd_recovery_ir_factor)
1798 factor = obd->obd_recovery_ir_factor;
1801 int new_soft = soft;
1802 int new_hard = hard;
1804 /* adjust timeout value by imperative recovery */
1806 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1807 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1809 /* make sure the timeout is not too short */
1810 new_soft = max(min, new_soft);
1811 new_hard = max(new_soft, new_hard);
1813 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1814 "window shrunk from %d-%d down to %d-%d\n",
1815 obd->obd_name, soft, hard, new_soft, new_hard);
1822 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1823 obd->obd_recovery_time_hard = hard;
1824 obd->obd_recovery_ir_factor = factor;
1826 EXPORT_SYMBOL(server_calc_timeout);