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)
50 #include <obd_class.h>
51 #include <lustre/lustre_user.h>
52 #include <linux/version.h>
53 #include <lustre_log.h>
54 #include <lustre_disk.h>
55 #include <lustre_param.h>
56 #ifdef HAVE_KERNEL_LOCKED
57 #include <linux/smp_lock.h>
59 #ifdef HAVE_SELINUX_IS_ENABLED
60 #include <linux/selinux.h>
63 /*********** mount lookup *********/
65 DEFINE_MUTEX(lustre_mount_info_lock);
66 static CFS_LIST_HEAD(server_mount_info_list);
68 static struct lustre_mount_info *server_find_mount(const char *name)
71 struct lustre_mount_info *lmi;
74 cfs_list_for_each(tmp, &server_mount_info_list) {
75 lmi = cfs_list_entry(tmp, struct lustre_mount_info,
77 if (strcmp(name, lmi->lmi_name) == 0)
83 /* we must register an obd for a mount before we call the setup routine.
84 *_setup will call lustre_get_mount to get the mnt struct
85 by obd_name, since we can't pass the pointer to setup. */
86 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;
117 cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list);
119 mutex_unlock(&lustre_mount_info_lock);
121 CDEBUG(D_MOUNT, "reg_mnt %p from %s\n", lmi->lmi_mnt, name);
126 /* when an obd no longer needs a mount */
127 static int server_deregister_mount(const char *name)
129 struct lustre_mount_info *lmi;
132 mutex_lock(&lustre_mount_info_lock);
133 lmi = server_find_mount(name);
135 mutex_unlock(&lustre_mount_info_lock);
136 CERROR("%s not registered\n", name);
140 CDEBUG(D_MOUNT, "dereg_mnt %p from %s\n", lmi->lmi_mnt, name);
142 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
143 cfs_list_del(&lmi->lmi_list_chain);
144 OBD_FREE(lmi, sizeof(*lmi));
145 mutex_unlock(&lustre_mount_info_lock);
150 /* obd's look up a registered mount using their obdname. This is just
151 for initial obd setup to find the mount struct. It should not be
152 called every time you want to mntget. */
153 struct lustre_mount_info *server_get_mount(const char *name)
155 struct lustre_mount_info *lmi;
156 struct lustre_sb_info *lsi;
159 mutex_lock(&lustre_mount_info_lock);
160 lmi = server_find_mount(name);
161 mutex_unlock(&lustre_mount_info_lock);
163 CERROR("Can't find mount for %s\n", name);
166 lsi = s2lsi(lmi->lmi_sb);
168 cfs_atomic_inc(&lsi->lsi_mounts);
170 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d\n", lmi->lmi_mnt,
171 name, cfs_atomic_read(&lsi->lsi_mounts));
175 EXPORT_SYMBOL(server_get_mount);
178 * Used by mdt to get mount_info from obdname.
179 * There are no blocking when using the mount_info.
180 * Do not use server_get_mount for this purpose.
182 struct lustre_mount_info *server_get_mount_2(const char *name)
184 struct lustre_mount_info *lmi;
187 mutex_lock(&lustre_mount_info_lock);
188 lmi = server_find_mount(name);
189 mutex_unlock(&lustre_mount_info_lock);
191 CERROR("Can't find mount for %s\n", name);
195 EXPORT_SYMBOL(server_get_mount_2);
197 /* to be called from obd_cleanup methods */
198 int server_put_mount(const char *name, struct vfsmount *mnt)
200 struct lustre_mount_info *lmi;
201 struct lustre_sb_info *lsi;
204 mutex_lock(&lustre_mount_info_lock);
205 lmi = server_find_mount(name);
206 mutex_unlock(&lustre_mount_info_lock);
208 CERROR("Can't find mount for %s\n", name);
211 lsi = s2lsi(lmi->lmi_sb);
213 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d\n",
214 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts));
216 if (lustre_put_lsi(lmi->lmi_sb))
217 CDEBUG(D_MOUNT, "Last put of mnt %p from %s\n",
220 /* this obd should never need the mount again */
221 server_deregister_mount(name);
225 EXPORT_SYMBOL(server_put_mount);
227 /* Corresponding to server_get_mount_2 */
228 int server_put_mount_2(const char *name, struct vfsmount *mnt)
233 EXPORT_SYMBOL(server_put_mount_2);
235 /* Set up a MGS to serve startup logs */
236 static int server_start_mgs(struct super_block *sb)
238 struct lustre_sb_info *lsi = s2lsi(sb);
239 struct vfsmount *mnt = lsi->lsi_srv_mnt;
240 struct lustre_mount_info *lmi;
244 /* It is impossible to have more than 1 MGS per node, since
245 MGC wouldn't know which to connect to */
246 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
248 lsi = s2lsi(lmi->lmi_sb);
249 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
254 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
256 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
259 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
260 LUSTRE_MGS_OBDNAME, 0, 0,
261 lsi->lsi_osd_obdname, 0);
262 /* server_deregister_mount() is not called previously, for lsi
263 * and other stuff can't be freed cleanly when mgs calls
264 * server_put_mount() in error handling case (see b=17758),
265 * this problem is caused by a bug in mgs_init0, which forgot
266 * calling server_put_mount in error case. */
269 server_deregister_mount(LUSTRE_MGS_OBDNAME);
273 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
274 "Is the 'mgs' module loaded?\n",
275 LUSTRE_MGS_OBDNAME, rc);
279 static int server_stop_mgs(struct super_block *sb)
281 struct obd_device *obd;
285 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
287 /* There better be only one MGS */
288 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
290 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
294 /* The MGS should always stop when we say so */
296 rc = class_manual_cleanup(obd);
300 /* Since there's only one mgc per node, we have to change it's fs to get
301 access to the right disk. */
302 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
304 struct lustre_sb_info *lsi = s2lsi(sb);
308 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
310 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
311 rc = obd_set_info_async(NULL, mgc->obd_self_export,
312 sizeof(KEY_SET_FS), KEY_SET_FS,
313 sizeof(*sb), sb, NULL);
315 CERROR("can't set_fs %d\n", rc);
320 static int server_mgc_clear_fs(struct obd_device *mgc)
325 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
327 rc = obd_set_info_async(NULL, mgc->obd_self_export,
328 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
333 static int is_mdc_device(const char *devname)
337 ptr = strrchr(devname, '-');
338 if (ptr != NULL && strcmp(ptr, "-mdc") == 0)
344 static inline int tgt_is_mdt0(const char *tgtname)
349 type = server_name2index(tgtname, &idx, NULL);
350 if (type != LDD_F_SV_TYPE_MDT)
356 static inline int is_mdc_for_mdt0(const char *devname)
360 if (!is_mdc_device(devname))
363 ptr = strrchr(devname, '-');
368 if (tgt_is_mdt0(devname)) {
377 * Convert OST/MDT name(fsname-OSTxxxx) to a lwp name
378 * (fsname-MDT0000-lwp-OSTxxxx)
380 int tgt_name2lwpname(const char *svname, char *lwpname)
387 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
391 rc = server_name2fsname(svname, fsname, &tgt);
393 CERROR("%s: failed to get fsname from svname. %d\n",
398 if (*tgt != '-' && *tgt != ':') {
399 CERROR("%s: invalid svname name!\n", svname);
400 GOTO(cleanup, rc = -EINVAL);
404 if (strncmp(tgt, "OST", 3) != 0 && strncmp(tgt, "MDT", 3) != 0) {
405 CERROR("%s is not an OST or MDT target!\n", svname);
406 GOTO(cleanup, rc = -EINVAL);
408 sprintf(lwpname, "%s-MDT0000-%s-%s", fsname, LUSTRE_LWP_NAME, tgt);
411 OBD_FREE(fsname, MTI_NAME_MAXLEN);
414 EXPORT_SYMBOL(tgt_name2lwpname);
416 static CFS_LIST_HEAD(lwp_register_list);
417 DEFINE_MUTEX(lwp_register_list_lock);
419 int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp,
420 register_lwp_cb cb_func, void *cb_data)
422 struct obd_device *lwp;
423 struct lwp_register_item *lri;
426 LASSERTF(strlen(lwpname) < MTI_NAME_MAXLEN, "lwpname is too long %s\n",
428 LASSERT(exp != NULL && *exp == NULL);
434 mutex_lock(&lwp_register_list_lock);
436 lwp = class_name2obd(lwpname);
437 if (lwp != NULL && lwp->obd_set_up == 1) {
438 struct obd_uuid *uuid;
442 mutex_unlock(&lwp_register_list_lock);
446 memcpy(uuid->uuid, lwpname, strlen(lwpname));
447 *exp = cfs_hash_lookup(lwp->obd_uuid_hash, uuid);
451 memcpy(lri->lri_name, lwpname, strlen(lwpname));
453 lri->lri_cb_func = cb_func;
454 lri->lri_cb_data = cb_data;
455 CFS_INIT_LIST_HEAD(&lri->lri_list);
456 cfs_list_add(&lri->lri_list, &lwp_register_list);
458 if (*exp != NULL && cb_func != NULL)
461 mutex_unlock(&lwp_register_list_lock);
464 EXPORT_SYMBOL(lustre_register_lwp_item);
466 void lustre_deregister_lwp_item(struct obd_export **exp)
468 struct lwp_register_item *lri, *tmp;
470 mutex_lock(&lwp_register_list_lock);
471 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
472 if (exp == lri->lri_exp) {
474 class_export_put(*exp);
475 cfs_list_del(&lri->lri_list);
480 mutex_unlock(&lwp_register_list_lock);
482 EXPORT_SYMBOL(lustre_deregister_lwp_item);
484 static void lustre_notify_lwp_list(struct obd_export *exp)
486 struct lwp_register_item *lri, *tmp;
487 LASSERT(exp != NULL);
489 mutex_lock(&lwp_register_list_lock);
490 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
491 if (strcmp(exp->exp_obd->obd_name, lri->lri_name))
493 if (*lri->lri_exp != NULL)
495 *lri->lri_exp = class_export_get(exp);
496 if (lri->lri_cb_func != NULL)
497 lri->lri_cb_func(lri->lri_cb_data);
499 mutex_unlock(&lwp_register_list_lock);
502 static int lustre_lwp_connect(struct obd_device *lwp)
505 struct lu_context session_ctx;
506 struct obd_export *exp;
507 struct obd_uuid *uuid = NULL;
508 struct obd_connect_data *data = NULL;
512 /* log has been fully processed, let clients connect */
513 rc = lu_env_init(&env, lwp->obd_lu_dev->ld_type->ldt_ctx_tags);
517 lu_context_init(&session_ctx, LCT_SESSION);
518 session_ctx.lc_thread = NULL;
519 lu_context_enter(&session_ctx);
520 env.le_ses = &session_ctx;
524 GOTO(out, rc = -ENOMEM);
526 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
527 data->ocd_version = LUSTRE_VERSION_CODE;
528 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
529 OBD_CONNECT_AT | OBD_CONNECT_LRU_RESIZE |
530 OBD_CONNECT_FULL20 | OBD_CONNECT_LVB_TYPE |
531 OBD_CONNECT_LIGHTWEIGHT;
534 GOTO(out, rc = -ENOMEM);
536 if (strlen(lwp->obd_name) > sizeof(uuid->uuid)) {
537 CERROR("%s: Too long lwp name %s, max_size is %d\n",
538 lwp->obd_name, lwp->obd_name, (int)sizeof(uuid->uuid));
539 GOTO(out, rc = -EINVAL);
542 /* Use lwp name as the uuid, so we find the export by lwp name later */
543 memcpy(uuid->uuid, lwp->obd_name, strlen(lwp->obd_name));
544 rc = obd_connect(&env, &exp, lwp, uuid, data, NULL);
546 CERROR("%s: connect failed: rc = %d\n", lwp->obd_name, rc);
548 lustre_notify_lwp_list(exp);
557 lu_context_exit(&session_ctx);
558 lu_context_fini(&session_ctx);
564 * lwp is used by slaves (Non-MDT0 targets) to manage the connection
567 static int lustre_lwp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi)
569 struct obd_device *obd;
570 char *lwpname = NULL;
571 char *lwpuuid = NULL;
575 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
578 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
582 OBD_ALLOC(lwpname, MTI_NAME_MAXLEN);
584 GOTO(out, rc = -ENOMEM);
586 rc = tgt_name2lwpname(lsi->lsi_svname, lwpname);
588 CERROR("%s: failed to generate lwp name. %d\n",
589 lsi->lsi_svname, rc);
593 OBD_ALLOC(lwpuuid, MTI_NAME_MAXLEN);
595 GOTO(out, rc = -ENOMEM);
597 sprintf(lwpuuid, "%s_UUID", lwpname);
598 rc = lustre_start_simple(lwpname, LUSTRE_LWP_NAME,
599 lwpuuid, lustre_cfg_string(lcfg, 1),
602 CERROR("%s: setup up failed: rc %d\n", lwpname, rc);
606 obd = class_name2obd(lwpname);
607 LASSERT(obd != NULL);
609 rc = lustre_lwp_connect(obd);
611 CERROR("%s: connect failed: rc = %d\n", lwpname, rc);
614 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
616 OBD_FREE(lwpuuid, MTI_NAME_MAXLEN);
621 /* the caller is responsible for memory free */
622 static struct obd_device *lustre_find_lwp(struct lustre_sb_info *lsi,
623 char **lwpname, char **logname)
625 struct obd_device *lwp;
629 LASSERT(lwpname != NULL);
630 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
632 OBD_ALLOC(*lwpname, MTI_NAME_MAXLEN);
633 if (*lwpname == NULL)
634 RETURN(ERR_PTR(-ENOMEM));
636 if (logname != NULL) {
637 OBD_ALLOC(*logname, MTI_NAME_MAXLEN);
638 if (*logname == NULL)
639 GOTO(out, rc = -ENOMEM);
640 rc = server_name2fsname(lsi->lsi_svname, *lwpname, NULL);
642 CERROR("%s: failed to get fsname from svname. %d\n",
643 lsi->lsi_svname, rc);
644 GOTO(out, rc = -EINVAL);
646 sprintf(*logname, "%s-client", *lwpname);
649 rc = tgt_name2lwpname(lsi->lsi_svname, *lwpname);
651 CERROR("%s: failed to generate lwp name. %d\n",
652 lsi->lsi_svname, rc);
653 GOTO(out, rc = -EINVAL);
656 lwp = class_name2obd(*lwpname);
660 if (*lwpname != NULL) {
661 OBD_FREE(*lwpname, MTI_NAME_MAXLEN);
664 if (logname != NULL && *logname != NULL) {
665 OBD_FREE(*logname, MTI_NAME_MAXLEN);
671 RETURN(lwp != NULL ? lwp : ERR_PTR(-ENOENT));
674 static int lustre_lwp_add_conn(struct lustre_cfg *cfg,
675 struct lustre_sb_info *lsi)
677 struct lustre_cfg_bufs *bufs = NULL;
678 struct lustre_cfg *lcfg = NULL;
679 char *lwpname = NULL;
680 struct obd_device *lwp;
684 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
686 CERROR("%s: can't find lwp device.\n", lsi->lsi_svname);
687 GOTO(out, rc = PTR_ERR(lwp));
689 LASSERT(lwpname != NULL);
693 GOTO(out, rc = -ENOMEM);
695 lustre_cfg_bufs_reset(bufs, lwpname);
696 lustre_cfg_bufs_set_string(bufs, 1,
697 lustre_cfg_string(cfg, 1));
699 lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
701 rc = class_add_conn(lwp, lcfg);
703 CERROR("%s: can't add conn: rc = %d\n", lwpname, rc);
709 lustre_cfg_free(lcfg);
711 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
716 * Retrieve MDT nids from the client log, then start the lwp device.
717 * there are only two scenarios which would include mdt nid.
719 * marker 5 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add mdc' xxx-
720 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
721 * attach 0:lustre-MDT0000-mdc 1:mdc 2:lustre-clilmv_UUID
722 * setup 0:lustre-MDT0000-mdc 1:lustre-MDT0000_UUID 2:192.168.122.162@tcp
723 * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
724 * add_conn 0:lustre-MDT0000-mdc 1:192.168.172.1@tcp
725 * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID xxxx
726 * marker 5 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add mdc' xxxx-
728 * marker 7 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
729 * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
730 * add_conn 0:lustre-MDT0000-mdc 1:192.168.122.2@tcp
731 * marker 7 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
733 static int client_lwp_config_process(const struct lu_env *env,
734 struct llog_handle *handle,
735 struct llog_rec_hdr *rec, void *data)
737 struct config_llog_instance *clli = data;
738 int cfg_len = rec->lrh_len;
739 char *cfg_buf = (char *) (rec + 1);
740 struct lustre_cfg *lcfg = NULL;
741 struct lustre_sb_info *lsi;
742 int rc = 0, swab = 0;
745 if (rec->lrh_type != OBD_CFG_REC) {
746 CERROR("Unknown llog record type %#x encountered\n",
751 LASSERT(clli->cfg_sb != NULL);
752 lsi = s2lsi(clli->cfg_sb);
754 lcfg = (struct lustre_cfg *)cfg_buf;
755 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
756 lustre_swab_lustre_cfg(lcfg);
760 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
764 switch (lcfg->lcfg_command) {
766 struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
768 lustre_swab_cfg_marker(marker, swab,
769 LUSTRE_CFG_BUFLEN(lcfg, 1));
770 if (marker->cm_flags & CM_SKIP ||
771 marker->cm_flags & CM_EXCLUDE)
774 if (!tgt_is_mdt0(marker->cm_tgtname))
777 if (!strncmp(marker->cm_comment, "add mdc", 7) ||
778 !strncmp(marker->cm_comment, "add failnid", 11)) {
779 if (marker->cm_flags & CM_START) {
780 clli->cfg_flags = CFG_F_MARKER;
781 /* This hack is to differentiate the
782 * ADD_UUID is come from "add mdc" record
783 * or from "add failnid" record. */
784 if (!strncmp(marker->cm_comment,
786 clli->cfg_flags |= CFG_F_SKIP;
787 } else if (marker->cm_flags & CM_END) {
793 case LCFG_ADD_UUID: {
794 if (clli->cfg_flags == CFG_F_MARKER) {
795 rc = lustre_lwp_setup(lcfg, lsi);
796 /* XXX: process only the first nid as
797 * we don't need another instance of lwp */
798 clli->cfg_flags |= CFG_F_SKIP;
799 } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
800 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
803 CERROR("%s: Fail to add uuid, rc:%d\n",
804 lsi->lsi_svname, rc);
808 case LCFG_ADD_CONN: {
809 if (is_mdc_for_mdt0(lustre_cfg_string(lcfg, 0)))
810 rc = lustre_lwp_add_conn(lcfg, lsi);
820 static int lustre_disconnect_lwp(struct super_block *sb)
822 struct lustre_sb_info *lsi = s2lsi(sb);
823 struct obd_device *lwp;
824 char *lwpname = NULL;
825 char *logname = NULL;
826 struct lustre_cfg *lcfg = NULL;
827 struct lustre_cfg_bufs *bufs = NULL;
828 struct config_llog_instance *cfg = NULL;
832 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
833 if (IS_ERR(lwp) && PTR_ERR(lwp) != -ENOENT)
834 GOTO(out, rc = PTR_ERR(lwp));
836 LASSERT(lwpname != NULL);
837 LASSERT(logname != NULL);
841 GOTO(out, rc = -ENOMEM);
844 cfg->cfg_instance = sb;
845 rc = lustre_end_log(sb, logname, cfg);
847 CERROR("%s: Can't end config log %s.\n", lwpname, logname);
851 if (PTR_ERR(lwp) == -ENOENT) {
852 CDEBUG(D_CONFIG, "%s: lwp device wasn't started.\n",
859 GOTO(out, rc = -ENOMEM);
861 lustre_cfg_bufs_reset(bufs, lwp->obd_name);
862 lustre_cfg_bufs_set_string(bufs, 1, NULL);
863 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
865 GOTO(out, rc = -ENOMEM);
867 /* Disconnect import first. NULL is passed for the '@env', since
868 * it will not be used. */
869 rc = lwp->obd_lu_dev->ld_ops->ldo_process_config(NULL, lwp->obd_lu_dev,
873 lustre_cfg_free(lcfg);
879 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
881 OBD_FREE(logname, MTI_NAME_MAXLEN);
886 * Stop the lwp for an OST/MDT target.
888 static int lustre_stop_lwp(struct super_block *sb)
890 struct lustre_sb_info *lsi = s2lsi(sb);
891 struct obd_device *lwp = NULL;
892 char *lwpname = NULL;
896 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
898 CDEBUG(PTR_ERR(lwp) == -ENOENT ? D_CONFIG : D_ERROR,
899 "%s: lwp wasn't started.\n", lsi->lsi_svname);
904 rc = class_manual_cleanup(lwp);
908 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
913 * Start the lwp(fsname-MDT0000-lwp-OSTxxxx) for an OST or MDT target,
914 * which would be used to establish connection from OST to MDT0.
916 static int lustre_start_lwp(struct super_block *sb)
918 struct lustre_sb_info *lsi = s2lsi(sb);
919 struct config_llog_instance *cfg = NULL;
920 struct obd_device *lwp;
921 char *lwpname = NULL;
922 char *logname = NULL;
926 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
928 /* the lwp device already stared */
929 if (lwp && !IS_ERR(lwp))
932 if (PTR_ERR(lwp) != -ENOENT)
933 GOTO(out, rc = PTR_ERR(lwp));
935 LASSERT(lwpname != NULL);
936 LASSERT(logname != NULL);
940 GOTO(out, rc = -ENOMEM);
942 cfg->cfg_callback = client_lwp_config_process;
943 cfg->cfg_instance = sb;
945 rc = lustre_process_log(sb, logname, cfg);
948 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
950 OBD_FREE(logname, MTI_NAME_MAXLEN);
956 DEFINE_MUTEX(server_start_lock);
958 /* Stop MDS/OSS if nobody is using them */
959 static int server_stop_servers(int lsiflags)
961 struct obd_device *obd = NULL;
962 struct obd_type *type = NULL;
966 mutex_lock(&server_start_lock);
968 /* Either an MDT or an OST or neither */
969 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
970 if (lsiflags & LDD_F_SV_TYPE_MDT) {
971 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
973 type = class_search_type(LUSTRE_MDT_NAME);
976 /* if this was an OST, and there are no more OST's, clean up the OSS */
977 if (lsiflags & LDD_F_SV_TYPE_OST) {
978 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
980 type = class_search_type(LUSTRE_OST_NAME);
983 if (obd != NULL && (type == NULL || type->typ_refcnt == 0)) {
985 /* obd_fail doesn't mean much on a server obd */
986 rc = class_manual_cleanup(obd);
989 mutex_unlock(&server_start_lock);
994 int server_mti_print(const char *title, struct mgs_target_info *mti)
996 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
997 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
998 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
999 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
1000 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
1001 mti->mti_config_ver, mti->mti_flags);
1004 EXPORT_SYMBOL(server_mti_print);
1006 /* Generate data for registration */
1007 static int server_lsi2mti(struct lustre_sb_info *lsi,
1008 struct mgs_target_info *mti)
1010 lnet_process_id_t id;
1015 if (!IS_SERVER(lsi))
1018 if (strlcpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname))
1019 >= sizeof(mti->mti_svname))
1022 mti->mti_nid_count = 0;
1023 while (LNetGetId(i++, &id) != -ENOENT) {
1024 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1027 /* server use --servicenode param, only allow specified
1028 * nids be registered */
1029 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
1030 class_match_nid(lsi->lsi_lmd->lmd_params,
1031 PARAM_FAILNODE, id.nid) < 1)
1034 /* match specified network */
1035 if (!class_match_net(lsi->lsi_lmd->lmd_params,
1036 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1039 mti->mti_nids[mti->mti_nid_count] = id.nid;
1040 mti->mti_nid_count++;
1041 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1042 CWARN("Only using first %d nids for %s\n",
1043 mti->mti_nid_count, mti->mti_svname);
1048 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1049 mti->mti_config_ver = 0;
1051 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1055 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1058 /* Orion requires index to be set */
1059 LASSERT(!(rc & LDD_F_NEED_INDEX));
1060 /* keep only LDD flags */
1061 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1062 if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN))
1063 mti->mti_flags |= LDD_F_UPDATE;
1064 cplen = strlcpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1065 sizeof(mti->mti_params));
1066 if (cplen >= sizeof(mti->mti_params))
1071 /* Register an old or new target with the MGS. If needed MGS will construct
1072 startup logs and assign index */
1073 static int server_register_target(struct lustre_sb_info *lsi)
1075 struct obd_device *mgc = lsi->lsi_mgc;
1076 struct mgs_target_info *mti = NULL;
1083 if (!IS_SERVER(lsi))
1090 rc = server_lsi2mti(lsi, mti);
1094 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1095 mti->mti_svname, mti->mti_fsname,
1096 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1099 /* if write_conf is true, the registration must succeed */
1100 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1101 mti->mti_flags |= LDD_F_OPC_REG;
1103 /* Register the target */
1104 /* FIXME use mgc_process_config instead */
1105 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1106 sizeof(KEY_REGISTER_TARGET),
1107 KEY_REGISTER_TARGET,
1108 sizeof(*mti), mti, NULL);
1110 if (mti->mti_flags & LDD_F_ERROR) {
1111 LCONSOLE_ERROR_MSG(0x160,
1112 "%s: the MGS refuses to allow this server "
1113 "to start: rc = %d. Please see messages on "
1114 "the MGS.\n", lsi->lsi_svname, rc);
1115 } else if (writeconf) {
1116 LCONSOLE_ERROR_MSG(0x15f,
1117 "%s: cannot register this server with the MGS: "
1118 "rc = %d. Is the MGS running?\n",
1119 lsi->lsi_svname, rc);
1121 CERROR("%s: error registering with the MGS: rc = %d "
1122 "(not fatal)\n", lsi->lsi_svname, rc);
1123 /* reset the error code for non-fatal error. */
1136 * Notify the MGS that this target is ready.
1137 * Used by IR - if the MGS receives this message, it will notify clients.
1139 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1141 struct lustre_sb_info *lsi = s2lsi(sb);
1142 struct obd_device *mgc = lsi->lsi_mgc;
1143 struct mgs_target_info *mti = NULL;
1149 if (!(IS_SERVER(lsi)))
1155 rc = server_lsi2mti(lsi, mti);
1159 mti->mti_instance = obd->u.obt.obt_instance;
1160 mti->mti_flags |= LDD_F_OPC_READY;
1162 /* FIXME use mgc_process_config instead */
1163 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1164 sizeof(KEY_REGISTER_TARGET),
1165 KEY_REGISTER_TARGET,
1166 sizeof(*mti), mti, NULL);
1168 /* Imperative recovery: if the mgs informs us to use IR? */
1169 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1170 (mti->mti_flags & LDD_F_IR_CAPABLE))
1171 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1180 /** Start server targets: MDTs and OSTs
1182 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1184 struct obd_device *obd;
1185 struct lustre_sb_info *lsi = s2lsi(sb);
1186 struct config_llog_instance cfg;
1188 struct lu_device *dev;
1192 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1195 /* make sure the MDS is started */
1196 mutex_lock(&server_start_lock);
1197 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1199 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1201 LUSTRE_MDS_OBDNAME"_uuid",
1204 mutex_unlock(&server_start_lock);
1205 CERROR("failed to start MDS: %d\n", rc);
1209 mutex_unlock(&server_start_lock);
1212 /* If we're an OST, make sure the global OSS is running */
1214 /* make sure OSS is started */
1215 mutex_lock(&server_start_lock);
1216 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1218 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1220 LUSTRE_OSS_OBDNAME"_uuid",
1223 mutex_unlock(&server_start_lock);
1224 CERROR("failed to start OSS: %d\n", rc);
1228 mutex_unlock(&server_start_lock);
1231 /* Set the mgc fs to our server disk. This allows the MGC to
1232 * read and write configs locally, in case it can't talk to the MGS. */
1233 if (lsi->lsi_srv_mnt) {
1234 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1236 GOTO(out_stop_service, rc);
1239 /* Register with MGS */
1240 rc = server_register_target(lsi);
1244 /* Let the target look up the mount using the target's name
1245 (we can't pass the sb or mnt through class_process_config.) */
1246 rc = server_register_mount(lsi->lsi_svname, sb, mnt);
1250 /* Start targets using the llog named for the target */
1251 memset(&cfg, 0, sizeof(cfg));
1252 cfg.cfg_callback = class_config_llog_handler;
1253 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1255 CERROR("failed to start server %s: %d\n",
1256 lsi->lsi_svname, rc);
1257 /* Do NOT call server_deregister_mount() here. This makes it
1258 * impossible to find mount later in cleanup time and leaves
1259 * @lsi and othder stuff leaked. -umka */
1263 obd = class_name2obd(lsi->lsi_svname);
1265 CERROR("no server named %s was started\n", lsi->lsi_svname);
1266 GOTO(out_mgc, rc = -ENXIO);
1269 if (IS_OST(lsi) || IS_MDT(lsi)) {
1270 rc = lustre_start_lwp(sb);
1272 CERROR("%s: failed to start LWP: %d\n",
1273 lsi->lsi_svname, rc);
1278 server_notify_target(sb, obd);
1280 /* calculate recovery timeout, do it after lustre_process_log */
1281 server_calc_timeout(lsi, obd);
1283 /* log has been fully processed */
1284 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1286 /* log has been fully processed, let clients connect */
1287 dev = obd->obd_lu_dev;
1288 if (dev && dev->ld_ops->ldo_prepare) {
1289 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1291 struct lu_context session_ctx;
1293 lu_context_init(&session_ctx, LCT_SESSION);
1294 session_ctx.lc_thread = NULL;
1295 lu_context_enter(&session_ctx);
1296 env.le_ses = &session_ctx;
1298 rc = dev->ld_ops->ldo_prepare(&env, NULL, dev);
1301 lu_context_exit(&session_ctx);
1302 lu_context_fini(&session_ctx);
1306 /* abort recovery only on the complete stack:
1307 * many devices can be involved */
1308 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1309 (OBP(obd, iocontrol))) {
1310 obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
1315 /* Release the mgc fs for others to use */
1316 if (lsi->lsi_srv_mnt)
1317 server_mgc_clear_fs(lsi->lsi_mgc);
1321 server_stop_servers(lsi->lsi_flags);
1326 static int lsi_prepare(struct lustre_sb_info *lsi)
1333 LASSERT(lsi->lsi_lmd);
1335 /* The server name is given as a mount line option */
1336 if (lsi->lsi_lmd->lmd_profile == NULL) {
1337 LCONSOLE_ERROR("Can't determine server name\n");
1341 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
1342 RETURN(-ENAMETOOLONG);
1344 strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
1346 /* Determine osd type */
1347 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
1348 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
1349 sizeof(lsi->lsi_osd_type))
1350 RETURN(-ENAMETOOLONG);
1352 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
1354 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
1357 /* XXX: a temp. solution for components using ldiskfs
1358 * to be removed in one of the subsequent patches */
1359 if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs"))
1360 strcpy(lsi->lsi_fstype, "ldiskfs");
1362 strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
1364 /* Determine server type */
1365 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1367 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
1368 /* Assume we're a bare MGS */
1370 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
1372 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
1377 lsi->lsi_flags |= rc;
1379 /* Add mount line flags that used to be in ldd:
1380 * writeconf, mgs, anything else?
1382 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
1383 LDD_F_WRITECONF : 0;
1384 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
1386 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_UPDATE) ?
1388 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
1389 LDD_F_SV_TYPE_MGS : 0;
1390 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
1391 LDD_F_NO_PRIMNODE : 0;
1396 /*************** server mount ******************/
1398 /** Start the shutdown of servers at umount.
1400 static void server_put_super(struct super_block *sb)
1402 struct lustre_sb_info *lsi = s2lsi(sb);
1403 struct obd_device *obd;
1404 char *tmpname, *extraname = NULL;
1406 int lsiflags = lsi->lsi_flags;
1409 LASSERT(IS_SERVER(lsi));
1411 tmpname_sz = strlen(lsi->lsi_svname) + 1;
1412 OBD_ALLOC(tmpname, tmpname_sz);
1413 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
1414 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1415 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1416 snprintf(tmpname, tmpname_sz, "MGS");
1418 /* disconnect the lwp first to drain off the inflight request */
1419 if (IS_OST(lsi) || IS_MDT(lsi)) {
1422 rc = lustre_disconnect_lwp(sb);
1423 if (rc && rc != ETIMEDOUT)
1424 CERROR("%s: failed to disconnect lwp. (rc=%d)\n",
1428 /* Stop the target */
1429 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1430 (IS_MDT(lsi) || IS_OST(lsi))) {
1431 struct lustre_profile *lprof = NULL;
1433 /* tell the mgc to drop the config log */
1434 lustre_end_log(sb, lsi->lsi_svname, NULL);
1436 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1437 If there are any setup/cleanup errors, save the lov
1438 name for safety cleanup later. */
1439 lprof = class_get_profile(lsi->lsi_svname);
1440 if (lprof && lprof->lp_dt) {
1441 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1442 strcpy(extraname, lprof->lp_dt);
1445 obd = class_name2obd(lsi->lsi_svname);
1447 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1448 if (lsiflags & LSI_UMOUNT_FAILOVER)
1450 /* We can't seem to give an error return code
1451 * to .put_super, so we better make sure we clean up! */
1453 class_manual_cleanup(obd);
1455 CERROR("no obd %s\n", lsi->lsi_svname);
1456 server_deregister_mount(lsi->lsi_svname);
1460 /* If they wanted the mgs to stop separately from the mdt, they
1461 should have put it on a different device. */
1463 /* if MDS start with --nomgs, don't stop MGS then */
1464 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1465 server_stop_mgs(sb);
1468 if (IS_OST(lsi) || IS_MDT(lsi)) {
1469 if (lustre_stop_lwp(sb) < 0)
1470 CERROR("%s: failed to stop lwp!\n", tmpname);
1473 /* Clean the mgc and sb */
1474 lustre_common_put_super(sb);
1476 /* wait till all in-progress cleanups are done
1477 * specifically we're interested in ofd cleanup
1479 obd_zombie_barrier();
1481 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1482 until the target is really gone so that our type refcount check
1484 server_stop_servers(lsiflags);
1486 /* In case of startup or cleanup err, stop related obds */
1488 obd = class_name2obd(extraname);
1490 CWARN("Cleaning orphaned obd %s\n", extraname);
1492 class_manual_cleanup(obd);
1494 OBD_FREE(extraname, strlen(extraname) + 1);
1497 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1498 OBD_FREE(tmpname, tmpname_sz);
1502 /** Called only for 'umount -f'
1504 static void server_umount_begin(struct super_block *sb)
1506 struct lustre_sb_info *lsi = s2lsi(sb);
1509 CDEBUG(D_MOUNT, "umount -f\n");
1510 /* umount = failover
1512 no third way to do non-force, non-failover */
1513 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1517 static int server_statfs(struct dentry *dentry, struct kstatfs *buf)
1519 struct super_block *sb = dentry->d_sb;
1520 struct lustre_sb_info *lsi = s2lsi(sb);
1521 struct obd_statfs statfs;
1525 if (lsi->lsi_dt_dev) {
1526 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
1528 statfs_unpack(buf, &statfs);
1529 buf->f_type = sb->s_magic;
1535 buf->f_type = sb->s_magic;
1536 buf->f_bsize = sb->s_blocksize;
1542 buf->f_namelen = NAME_MAX;
1546 /** The operations we support directly on the superblock:
1547 * mount, umount, and df.
1549 static struct super_operations server_ops = {
1550 .put_super = server_put_super,
1551 .umount_begin = server_umount_begin, /* umount -f */
1552 .statfs = server_statfs,
1556 * Xattr support for Lustre servers
1558 static ssize_t lustre_getxattr(struct dentry *dentry, const char *name,
1559 void *buffer, size_t size)
1561 if (!selinux_is_enabled())
1566 static int lustre_setxattr(struct dentry *dentry, const char *name,
1567 const void *value, size_t size, int flags)
1572 static ssize_t lustre_listxattr(struct dentry *d_entry, char *name,
1578 const struct inode_operations server_inode_operations = {
1579 .setxattr = lustre_setxattr,
1580 .getxattr = lustre_getxattr,
1581 .listxattr = lustre_listxattr,
1584 #define log2(n) ffz(~(n))
1585 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1587 static int server_fill_super_common(struct super_block *sb)
1589 struct inode *root = 0;
1592 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1594 sb->s_blocksize = 4096;
1595 sb->s_blocksize_bits = log2(sb->s_blocksize);
1596 sb->s_magic = LUSTRE_SUPER_MAGIC;
1597 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1598 sb->s_flags |= MS_RDONLY;
1599 sb->s_op = &server_ops;
1601 root = new_inode(sb);
1603 CERROR("Can't make root inode\n");
1607 /* returns -EIO for every operation */
1608 /* make_bad_inode(root); -- badness - can't umount */
1609 /* apparently we need to be a directory for the mount to finish */
1610 root->i_mode = S_IFDIR;
1611 root->i_op = &server_inode_operations;
1612 sb->s_root = d_make_root(root);
1614 CERROR("%s: can't make root dentry\n", sb->s_id);
1621 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
1623 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1624 struct obd_device *obd;
1625 struct dt_device_param p;
1631 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
1632 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
1634 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
1635 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
1636 strcat(lsi->lsi_osd_uuid, "_UUID");
1637 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
1639 obd = class_name2obd(lsi->lsi_osd_obdname);
1641 rc = lustre_start_simple(lsi->lsi_osd_obdname,
1643 lsi->lsi_osd_uuid, lmd->lmd_dev,
1644 flagstr, lsi->lsi_lmd->lmd_opts,
1648 obd = class_name2obd(lsi->lsi_osd_obdname);
1652 rc = obd_connect(NULL, &lsi->lsi_osd_exp,
1653 obd, &obd->obd_uuid, NULL, NULL);
1656 class_manual_cleanup(obd);
1657 lsi->lsi_dt_dev = NULL;
1661 LASSERT(obd->obd_lu_dev);
1662 lu_device_get(obd->obd_lu_dev);
1663 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
1664 LASSERT(lsi->lsi_dt_dev);
1666 /* set disk context for llog usage */
1667 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1668 obd->obd_lvfs_ctxt.dt = lsi->lsi_dt_dev;
1670 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
1672 lsi->lsi_srv_mnt = p.ddp_mnt;
1678 /** Fill in the superblock info for a Lustre server.
1679 * Mount the device with the correct options.
1680 * Read the on-disk config file.
1681 * Start the services.
1683 int server_fill_super(struct super_block *sb)
1685 struct lustre_sb_info *lsi = s2lsi(sb);
1689 rc = lsi_prepare(lsi);
1693 /* Start low level OSD */
1694 rc = osd_start(lsi, sb->s_flags);
1696 CERROR("Unable to start osd on %s: %d\n",
1697 lsi->lsi_lmd->lmd_dev, rc);
1702 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
1703 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
1705 if (class_name2obd(lsi->lsi_svname)) {
1706 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1707 "running. Double-mount may have compromised"
1708 " the disk journal.\n",
1714 /* Start MGS before MGC */
1715 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1716 rc = server_start_mgs(sb);
1721 /* Start MGC before servers */
1722 rc = lustre_start_mgc(sb);
1726 /* Set up all obd devices for service */
1727 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1728 (IS_OST(lsi) || IS_MDT(lsi))) {
1729 rc = server_start_targets(sb, lsi->lsi_srv_mnt);
1731 CERROR("Unable to start targets: %d\n", rc);
1734 /* FIXME overmount client here, or can we just start a
1735 * client log and client_fill_super on this sb? We
1736 * need to make sure server_put_super gets called too
1737 * - ll_put_super calls lustre_common_put_super; check
1738 * there for LSI_SERVER flag, call s_p_s if so.
1740 * Probably should start client from new thread so we
1741 * can return. Client will not finish until all
1742 * servers are connected. Note - MGS-only server does
1743 * NOT get a client, since there is no lustre fs
1744 * associated - the MGS is for all lustre fs's */
1747 rc = server_fill_super_common(sb);
1753 /* We jump here in case of failure while starting targets or MGS.
1754 * In this case we can't just put @mnt and have to do real cleanup
1755 * with stoping targets, etc. */
1756 server_put_super(sb);
1761 * Calculate timeout value for a target.
1763 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1765 struct lustre_mount_data *lmd;
1769 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
1770 int min = OBD_RECOVERY_TIME_MIN;
1772 LASSERT(IS_SERVER(lsi));
1776 soft = lmd->lmd_recovery_time_soft;
1777 hard = lmd->lmd_recovery_time_hard;
1778 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1779 obd->obd_no_ir = !has_ir;
1783 soft = OBD_RECOVERY_TIME_SOFT;
1785 hard = OBD_RECOVERY_TIME_HARD;
1787 /* target may have ir_factor configured. */
1788 factor = OBD_IR_FACTOR_DEFAULT;
1789 if (obd->obd_recovery_ir_factor)
1790 factor = obd->obd_recovery_ir_factor;
1793 int new_soft = soft;
1794 int new_hard = hard;
1796 /* adjust timeout value by imperative recovery */
1798 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1799 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1801 /* make sure the timeout is not too short */
1802 new_soft = max(min, new_soft);
1803 new_hard = max(new_soft, new_hard);
1805 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1806 "window shrunk from %d-%d down to %d-%d\n",
1807 obd->obd_name, soft, hard, new_soft, new_hard);
1814 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1815 obd->obd_recovery_time_hard = hard;
1816 obd->obd_recovery_ir_factor = factor;
1818 EXPORT_SYMBOL(server_calc_timeout);