4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/obdclass/obd_mount_server.c
38 * Server mount routines
40 * Author: Nathan Rutman <nathan@clusterfs.com>
44 #define DEBUG_SUBSYSTEM S_CLASS
45 #define D_MOUNT (D_SUPER | D_CONFIG /* | D_WARNING */)
46 #define PRINT_CMD CDEBUG
47 #define PRINT_MASK (D_SUPER | D_CONFIG)
51 #include <obd_class.h>
52 #include <lustre/lustre_user.h>
53 #include <linux/version.h>
54 #include <lustre_log.h>
55 #include <lustre_disk.h>
56 #include <lustre_param.h>
57 #ifdef HAVE_KERNEL_LOCKED
58 #include <linux/smp_lock.h>
61 /*********** mount lookup *********/
63 DEFINE_MUTEX(lustre_mount_info_lock);
64 static CFS_LIST_HEAD(server_mount_info_list);
66 static struct lustre_mount_info *server_find_mount(const char *name)
69 struct lustre_mount_info *lmi;
72 cfs_list_for_each(tmp, &server_mount_info_list) {
73 lmi = cfs_list_entry(tmp, struct lustre_mount_info,
75 if (strcmp(name, lmi->lmi_name) == 0)
81 /* we must register an obd for a mount before we call the setup routine.
82 *_setup will call lustre_get_mount to get the mnt struct
83 by obd_name, since we can't pass the pointer to setup. */
84 static int server_register_mount(const char *name, struct super_block *sb,
87 struct lustre_mount_info *lmi;
93 OBD_ALLOC(lmi, sizeof(*lmi));
96 OBD_ALLOC(name_cp, strlen(name) + 1);
98 OBD_FREE(lmi, sizeof(*lmi));
101 strcpy(name_cp, name);
103 mutex_lock(&lustre_mount_info_lock);
105 if (server_find_mount(name)) {
106 mutex_unlock(&lustre_mount_info_lock);
107 OBD_FREE(lmi, sizeof(*lmi));
108 OBD_FREE(name_cp, strlen(name) + 1);
109 CERROR("Already registered %s\n", name);
112 lmi->lmi_name = name_cp;
115 cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list);
117 mutex_unlock(&lustre_mount_info_lock);
119 CDEBUG(D_MOUNT, "reg_mnt %p from %s\n", lmi->lmi_mnt, name);
124 /* when an obd no longer needs a mount */
125 static int server_deregister_mount(const char *name)
127 struct lustre_mount_info *lmi;
130 mutex_lock(&lustre_mount_info_lock);
131 lmi = server_find_mount(name);
133 mutex_unlock(&lustre_mount_info_lock);
134 CERROR("%s not registered\n", name);
138 CDEBUG(D_MOUNT, "dereg_mnt %p from %s\n", lmi->lmi_mnt, name);
140 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
141 cfs_list_del(&lmi->lmi_list_chain);
142 OBD_FREE(lmi, sizeof(*lmi));
143 mutex_unlock(&lustre_mount_info_lock);
148 /* obd's look up a registered mount using their obdname. This is just
149 for initial obd setup to find the mount struct. It should not be
150 called every time you want to mntget. */
151 struct lustre_mount_info *server_get_mount(const char *name)
153 struct lustre_mount_info *lmi;
154 struct lustre_sb_info *lsi;
157 mutex_lock(&lustre_mount_info_lock);
158 lmi = server_find_mount(name);
159 mutex_unlock(&lustre_mount_info_lock);
161 CERROR("Can't find mount for %s\n", name);
164 lsi = s2lsi(lmi->lmi_sb);
166 cfs_atomic_inc(&lsi->lsi_mounts);
168 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d\n", lmi->lmi_mnt,
169 name, cfs_atomic_read(&lsi->lsi_mounts));
173 EXPORT_SYMBOL(server_get_mount);
176 * Used by mdt to get mount_info from obdname.
177 * There are no blocking when using the mount_info.
178 * Do not use server_get_mount for this purpose.
180 struct lustre_mount_info *server_get_mount_2(const char *name)
182 struct lustre_mount_info *lmi;
185 mutex_lock(&lustre_mount_info_lock);
186 lmi = server_find_mount(name);
187 mutex_unlock(&lustre_mount_info_lock);
189 CERROR("Can't find mount for %s\n", name);
193 EXPORT_SYMBOL(server_get_mount_2);
195 /* to be called from obd_cleanup methods */
196 int server_put_mount(const char *name, struct vfsmount *mnt)
198 struct lustre_mount_info *lmi;
199 struct lustre_sb_info *lsi;
202 mutex_lock(&lustre_mount_info_lock);
203 lmi = server_find_mount(name);
204 mutex_unlock(&lustre_mount_info_lock);
206 CERROR("Can't find mount for %s\n", name);
209 lsi = s2lsi(lmi->lmi_sb);
211 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d\n",
212 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts));
214 if (lustre_put_lsi(lmi->lmi_sb))
215 CDEBUG(D_MOUNT, "Last put of mnt %p from %s\n",
218 /* this obd should never need the mount again */
219 server_deregister_mount(name);
223 EXPORT_SYMBOL(server_put_mount);
225 /* Corresponding to server_get_mount_2 */
226 int server_put_mount_2(const char *name, struct vfsmount *mnt)
231 EXPORT_SYMBOL(server_put_mount_2);
233 /* Set up a MGS to serve startup logs */
234 static int server_start_mgs(struct super_block *sb)
236 struct lustre_sb_info *lsi = s2lsi(sb);
237 struct vfsmount *mnt = lsi->lsi_srv_mnt;
238 struct lustre_mount_info *lmi;
242 /* It is impossible to have more than 1 MGS per node, since
243 MGC wouldn't know which to connect to */
244 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
246 lsi = s2lsi(lmi->lmi_sb);
247 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
252 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
254 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
257 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
258 LUSTRE_MGS_OBDNAME, 0, 0,
259 lsi->lsi_osd_obdname, 0);
260 /* Do NOT call server_deregister_mount() here. This leads to
261 * inability cleanup cleanly and free lsi and other stuff when
262 * mgs calls server_put_mount() in error handling case. -umka */
266 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
267 "Is the 'mgs' module loaded?\n",
268 LUSTRE_MGS_OBDNAME, rc);
272 static int server_stop_mgs(struct super_block *sb)
274 struct obd_device *obd;
278 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
280 /* There better be only one MGS */
281 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
283 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
287 /* The MGS should always stop when we say so */
289 rc = class_manual_cleanup(obd);
293 /* Since there's only one mgc per node, we have to change it's fs to get
294 access to the right disk. */
295 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
297 struct lustre_sb_info *lsi = s2lsi(sb);
301 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
303 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
304 rc = obd_set_info_async(NULL, mgc->obd_self_export,
305 sizeof(KEY_SET_FS), KEY_SET_FS,
306 sizeof(*sb), sb, NULL);
308 CERROR("can't set_fs %d\n", rc);
313 static int server_mgc_clear_fs(struct obd_device *mgc)
318 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
320 rc = obd_set_info_async(NULL, mgc->obd_self_export,
321 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
326 static int is_mdc_device(const char *devname)
330 ptr = strrchr(devname, '-');
331 if (ptr != NULL && strcmp(ptr, "-mdc") == 0)
337 static inline int tgt_is_mdt0(const char *tgtname)
342 type = server_name2index(tgtname, &idx, NULL);
343 if (type != LDD_F_SV_TYPE_MDT)
349 static inline int is_mdc_for_mdt0(const char *devname)
353 if (!is_mdc_device(devname))
356 ptr = strrchr(devname, '-');
361 if (tgt_is_mdt0(devname)) {
370 * Convert OST/MDT name(fsname-OSTxxxx) to a lwp name
371 * (fsname-MDT0000-lwp-OSTxxxx)
373 int tgt_name2lwpname(const char *svname, char *lwpname)
380 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
384 rc = server_name2fsname(svname, fsname, &tgt);
386 CERROR("%s: failed to get fsname from svname. %d\n",
391 if (*tgt != '-' && *tgt != ':') {
392 CERROR("%s: invalid svname name!\n", svname);
393 GOTO(cleanup, rc = -EINVAL);
397 if (strncmp(tgt, "OST", 3) != 0 && strncmp(tgt, "MDT", 3) != 0) {
398 CERROR("%s is not an OST or MDT target!\n", svname);
399 GOTO(cleanup, rc = -EINVAL);
401 sprintf(lwpname, "%s-MDT0000-%s-%s", fsname, LUSTRE_LWP_NAME, tgt);
404 OBD_FREE(fsname, MTI_NAME_MAXLEN);
407 EXPORT_SYMBOL(tgt_name2lwpname);
409 static CFS_LIST_HEAD(lwp_register_list);
410 DEFINE_MUTEX(lwp_register_list_lock);
412 int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp,
413 register_lwp_cb cb_func, void *cb_data)
415 struct obd_device *lwp;
416 struct lwp_register_item *lri;
419 LASSERTF(strlen(lwpname) < MTI_NAME_MAXLEN, "lwpname is too long %s\n",
421 LASSERT(exp != NULL && *exp == NULL);
427 mutex_lock(&lwp_register_list_lock);
429 lwp = class_name2obd(lwpname);
430 if (lwp != NULL && lwp->obd_set_up == 1) {
431 struct obd_uuid *uuid;
435 mutex_unlock(&lwp_register_list_lock);
439 memcpy(uuid->uuid, lwpname, strlen(lwpname));
440 *exp = cfs_hash_lookup(lwp->obd_uuid_hash, uuid);
444 memcpy(lri->lri_name, lwpname, strlen(lwpname));
446 lri->lri_cb_func = cb_func;
447 lri->lri_cb_data = cb_data;
448 CFS_INIT_LIST_HEAD(&lri->lri_list);
449 cfs_list_add(&lri->lri_list, &lwp_register_list);
451 if (*exp != NULL && cb_func != NULL)
454 mutex_unlock(&lwp_register_list_lock);
457 EXPORT_SYMBOL(lustre_register_lwp_item);
459 void lustre_deregister_lwp_item(struct obd_export **exp)
461 struct lwp_register_item *lri, *tmp;
463 mutex_lock(&lwp_register_list_lock);
464 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
465 if (exp == lri->lri_exp) {
467 class_export_put(*exp);
468 cfs_list_del(&lri->lri_list);
473 mutex_unlock(&lwp_register_list_lock);
475 EXPORT_SYMBOL(lustre_deregister_lwp_item);
477 static void lustre_notify_lwp_list(struct obd_export *exp)
479 struct lwp_register_item *lri, *tmp;
480 LASSERT(exp != NULL);
482 mutex_lock(&lwp_register_list_lock);
483 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
484 if (strcmp(exp->exp_obd->obd_name, lri->lri_name))
486 if (*lri->lri_exp != NULL)
488 *lri->lri_exp = class_export_get(exp);
489 if (lri->lri_cb_func != NULL)
490 lri->lri_cb_func(lri->lri_cb_data);
492 mutex_unlock(&lwp_register_list_lock);
495 static int lustre_lwp_connect(struct obd_device *lwp)
498 struct lu_context session_ctx;
499 struct obd_export *exp;
500 struct obd_uuid *uuid = NULL;
501 struct obd_connect_data *data = NULL;
505 /* log has been fully processed, let clients connect */
506 rc = lu_env_init(&env, lwp->obd_lu_dev->ld_type->ldt_ctx_tags);
510 lu_context_init(&session_ctx, LCT_SESSION);
511 session_ctx.lc_thread = NULL;
512 lu_context_enter(&session_ctx);
513 env.le_ses = &session_ctx;
517 GOTO(out, rc = -ENOMEM);
519 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
520 data->ocd_version = LUSTRE_VERSION_CODE;
521 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
522 OBD_CONNECT_AT | OBD_CONNECT_LRU_RESIZE |
523 OBD_CONNECT_FULL20 | OBD_CONNECT_LVB_TYPE |
524 OBD_CONNECT_LIGHTWEIGHT;
527 GOTO(out, rc = -ENOMEM);
529 if (strlen(lwp->obd_name) > sizeof(uuid->uuid)) {
530 CERROR("%s: Too long lwp name %s, max_size is %d\n",
531 lwp->obd_name, lwp->obd_name, (int)sizeof(uuid->uuid));
532 GOTO(out, rc = -EINVAL);
535 /* Use lwp name as the uuid, so we find the export by lwp name later */
536 memcpy(uuid->uuid, lwp->obd_name, strlen(lwp->obd_name));
537 rc = obd_connect(&env, &exp, lwp, uuid, data, NULL);
539 CERROR("%s: connect failed: rc = %d\n", lwp->obd_name, rc);
541 lustre_notify_lwp_list(exp);
550 lu_context_exit(&session_ctx);
551 lu_context_fini(&session_ctx);
557 * lwp is used by slaves (Non-MDT0 targets) to manage the connection
560 static int lustre_lwp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi)
562 struct obd_connect_data *data = NULL;
563 struct obd_device *obd;
564 char *lwpname = NULL;
565 char *lwpuuid = NULL;
569 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
572 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
576 OBD_ALLOC(lwpname, MTI_NAME_MAXLEN);
578 GOTO(out, rc = -ENOMEM);
580 rc = tgt_name2lwpname(lsi->lsi_svname, lwpname);
582 CERROR("%s: failed to generate lwp name. %d\n",
583 lsi->lsi_svname, rc);
587 OBD_ALLOC(lwpuuid, MTI_NAME_MAXLEN);
589 GOTO(out, rc = -ENOMEM);
591 sprintf(lwpuuid, "%s_UUID", lwpname);
592 rc = lustre_start_simple(lwpname, LUSTRE_LWP_NAME,
593 lwpuuid, lustre_cfg_string(lcfg, 1),
596 CERROR("%s: setup up failed: rc %d\n", lwpname, rc);
600 obd = class_name2obd(lwpname);
601 LASSERT(obd != NULL);
603 rc = lustre_lwp_connect(obd);
605 CERROR("%s: connect failed: rc = %d\n", lwpname, rc);
610 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
612 OBD_FREE(lwpuuid, MTI_NAME_MAXLEN);
617 /* the caller is responsible for memory free */
618 static struct obd_device *lustre_find_lwp(struct lustre_sb_info *lsi,
619 char **lwpname, char **logname)
621 struct obd_device *lwp;
625 LASSERT(lwpname != NULL);
626 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
628 OBD_ALLOC(*lwpname, MTI_NAME_MAXLEN);
629 if (*lwpname == NULL)
630 RETURN(ERR_PTR(-ENOMEM));
632 if (logname != NULL) {
633 OBD_ALLOC(*logname, MTI_NAME_MAXLEN);
634 if (*logname == NULL)
635 GOTO(out, rc = -ENOMEM);
636 rc = server_name2fsname(lsi->lsi_svname, *lwpname, NULL);
638 CERROR("%s: failed to get fsname from svname. %d\n",
639 lsi->lsi_svname, rc);
640 GOTO(out, rc = -EINVAL);
642 sprintf(*logname, "%s-client", *lwpname);
645 rc = tgt_name2lwpname(lsi->lsi_svname, *lwpname);
647 CERROR("%s: failed to generate lwp name. %d\n",
648 lsi->lsi_svname, rc);
649 GOTO(out, rc = -EINVAL);
652 lwp = class_name2obd(*lwpname);
656 if (*lwpname != NULL) {
657 OBD_FREE(*lwpname, MTI_NAME_MAXLEN);
660 if (logname != NULL && *logname != NULL) {
661 OBD_FREE(*logname, MTI_NAME_MAXLEN);
667 RETURN(lwp != NULL ? lwp : ERR_PTR(-ENOENT));
670 static int lustre_lwp_add_conn(struct lustre_cfg *cfg,
671 struct lustre_sb_info *lsi)
673 struct lustre_cfg_bufs *bufs = NULL;
674 struct lustre_cfg *lcfg = NULL;
675 char *lwpname = NULL;
676 struct obd_device *lwp;
680 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
682 CERROR("%s: can't find lwp device.\n", lsi->lsi_svname);
683 GOTO(out, rc = PTR_ERR(lwp));
685 LASSERT(lwpname != NULL);
689 GOTO(out, rc = -ENOMEM);
691 lustre_cfg_bufs_reset(bufs, lwpname);
692 lustre_cfg_bufs_set_string(bufs, 1,
693 lustre_cfg_string(cfg, 1));
695 lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
697 rc = class_add_conn(lwp, lcfg);
699 CERROR("%s: can't add conn: rc = %d\n", lwpname, rc);
705 lustre_cfg_free(lcfg);
707 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
712 * Retrieve MDT nids from the client log, then start the lwp device.
713 * there are only two scenarios which would include mdt nid.
715 * marker 5 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add mdc' xxx-
716 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
717 * attach 0:lustre-MDT0000-mdc 1:mdc 2:lustre-clilmv_UUID
718 * setup 0:lustre-MDT0000-mdc 1:lustre-MDT0000_UUID 2:192.168.122.162@tcp
719 * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
720 * add_conn 0:lustre-MDT0000-mdc 1:192.168.172.1@tcp
721 * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID xxxx
722 * marker 5 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add mdc' xxxx-
724 * marker 7 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
725 * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
726 * add_conn 0:lustre-MDT0000-mdc 1:192.168.122.2@tcp
727 * marker 7 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
729 static int client_lwp_config_process(const struct lu_env *env,
730 struct llog_handle *handle,
731 struct llog_rec_hdr *rec, void *data)
733 struct config_llog_instance *clli = data;
734 int cfg_len = rec->lrh_len;
735 char *cfg_buf = (char *) (rec + 1);
736 struct lustre_cfg *lcfg = NULL;
737 struct lustre_sb_info *lsi;
738 int rc = 0, swab = 0;
741 if (rec->lrh_type != OBD_CFG_REC) {
742 CERROR("Unknown llog record type %#x encountered\n",
747 LASSERT(clli->cfg_sb != NULL);
748 lsi = s2lsi(clli->cfg_sb);
750 lcfg = (struct lustre_cfg *)cfg_buf;
751 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
752 lustre_swab_lustre_cfg(lcfg);
756 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
760 switch (lcfg->lcfg_command) {
762 struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
764 lustre_swab_cfg_marker(marker, swab,
765 LUSTRE_CFG_BUFLEN(lcfg, 1));
766 if (marker->cm_flags & CM_SKIP ||
767 marker->cm_flags & CM_EXCLUDE)
770 if (!tgt_is_mdt0(marker->cm_tgtname))
773 if (!strncmp(marker->cm_comment, "add mdc", 7) ||
774 !strncmp(marker->cm_comment, "add failnid", 11)) {
775 if (marker->cm_flags & CM_START) {
776 clli->cfg_flags = CFG_F_MARKER;
777 /* This hack is to differentiate the
778 * ADD_UUID is come from "add mdc" record
779 * or from "add failnid" record. */
780 if (!strncmp(marker->cm_comment,
782 clli->cfg_flags |= CFG_F_SKIP;
783 } else if (marker->cm_flags & CM_END) {
789 case LCFG_ADD_UUID: {
790 if (clli->cfg_flags == CFG_F_MARKER) {
791 rc = lustre_lwp_setup(lcfg, lsi);
792 /* XXX: process only the first nid as
793 * we don't need another instance of lwp */
794 clli->cfg_flags |= CFG_F_SKIP;
795 } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
796 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
799 CERROR("%s: Fail to add uuid, rc:%d\n",
800 lsi->lsi_svname, rc);
804 case LCFG_ADD_CONN: {
805 if (is_mdc_for_mdt0(lustre_cfg_string(lcfg, 0)))
806 rc = lustre_lwp_add_conn(lcfg, lsi);
816 static int lustre_disconnect_lwp(struct super_block *sb)
818 struct lustre_sb_info *lsi = s2lsi(sb);
819 struct obd_device *lwp;
820 char *lwpname = NULL;
821 char *logname = NULL;
822 struct lustre_cfg *lcfg = NULL;
823 struct lustre_cfg_bufs *bufs = NULL;
824 struct config_llog_instance *cfg = NULL;
828 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
829 if (IS_ERR(lwp) && PTR_ERR(lwp) != -ENOENT)
830 GOTO(out, rc = PTR_ERR(lwp));
832 LASSERT(lwpname != NULL);
833 LASSERT(logname != NULL);
837 GOTO(out, rc = -ENOMEM);
840 cfg->cfg_instance = sb;
841 rc = lustre_end_log(sb, logname, cfg);
843 CERROR("%s: Can't end config log %s.\n", lwpname, logname);
847 if (PTR_ERR(lwp) == -ENOENT) {
848 CDEBUG(D_CONFIG, "%s: lwp device wasn't started.\n",
855 GOTO(out, rc = -ENOMEM);
857 lustre_cfg_bufs_reset(bufs, lwp->obd_name);
858 lustre_cfg_bufs_set_string(bufs, 1, NULL);
859 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
861 GOTO(out, rc = -ENOMEM);
863 /* Disconnect import first. NULL is passed for the '@env', since
864 * it will not be used. */
865 rc = lwp->obd_lu_dev->ld_ops->ldo_process_config(NULL, lwp->obd_lu_dev,
869 lustre_cfg_free(lcfg);
875 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
877 OBD_FREE(logname, MTI_NAME_MAXLEN);
882 * Stop the lwp for an OST/MDT target.
884 static int lustre_stop_lwp(struct super_block *sb)
886 struct lustre_sb_info *lsi = s2lsi(sb);
887 struct obd_device *lwp = NULL;
888 char *lwpname = NULL;
892 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
894 CDEBUG(PTR_ERR(lwp) == -ENOENT ? D_CONFIG : D_ERROR,
895 "%s: lwp wasn't started.\n", lsi->lsi_svname);
900 rc = class_manual_cleanup(lwp);
904 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
909 * Start the lwp(fsname-MDT0000-lwp-OSTxxxx) for an OST or MDT target,
910 * which would be used to establish connection from OST to MDT0.
912 static int lustre_start_lwp(struct super_block *sb)
914 struct lustre_sb_info *lsi = s2lsi(sb);
915 struct config_llog_instance *cfg = NULL;
916 struct obd_device *lwp;
917 char *lwpname = NULL;
918 char *logname = NULL;
922 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
924 /* the lwp device already stared */
925 if (lwp && !IS_ERR(lwp))
928 if (PTR_ERR(lwp) != -ENOENT)
929 GOTO(out, rc = PTR_ERR(lwp));
931 LASSERT(lwpname != NULL);
932 LASSERT(logname != NULL);
936 GOTO(out, rc = -ENOMEM);
938 cfg->cfg_callback = client_lwp_config_process;
939 cfg->cfg_instance = sb;
941 rc = lustre_process_log(sb, logname, cfg);
944 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
946 OBD_FREE(logname, MTI_NAME_MAXLEN);
952 DEFINE_MUTEX(server_start_lock);
954 /* Stop MDS/OSS if nobody is using them */
955 static int server_stop_servers(int lsiflags)
957 struct obd_device *obd = NULL;
958 struct obd_type *type = NULL;
962 mutex_lock(&server_start_lock);
964 /* Either an MDT or an OST or neither */
965 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
966 if (lsiflags & LDD_F_SV_TYPE_MDT) {
967 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
969 type = class_search_type(LUSTRE_MDT_NAME);
972 /* if this was an OST, and there are no more OST's, clean up the OSS */
973 if (lsiflags & LDD_F_SV_TYPE_OST) {
974 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
976 type = class_search_type(LUSTRE_OST_NAME);
979 if (obd != NULL && (type == NULL || type->typ_refcnt == 0)) {
983 /* obd_fail doesn't mean much on a server obd */
984 err = 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 fsfilt
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,
1555 #define log2(n) ffz(~(n))
1556 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1558 static int server_fill_super_common(struct super_block *sb)
1560 struct inode *root = 0;
1563 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1565 sb->s_blocksize = 4096;
1566 sb->s_blocksize_bits = log2(sb->s_blocksize);
1567 sb->s_magic = LUSTRE_SUPER_MAGIC;
1568 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1569 sb->s_flags |= MS_RDONLY;
1570 sb->s_op = &server_ops;
1572 root = new_inode(sb);
1574 CERROR("Can't make root inode\n");
1578 /* returns -EIO for every operation */
1579 /* make_bad_inode(root); -- badness - can't umount */
1580 /* apparently we need to be a directory for the mount to finish */
1581 root->i_mode = S_IFDIR;
1583 sb->s_root = d_make_root(root);
1585 CERROR("%s: can't make root dentry\n", sb->s_id);
1592 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
1594 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1595 struct obd_device *obd;
1596 struct dt_device_param p;
1602 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
1603 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
1605 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
1606 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
1607 strcat(lsi->lsi_osd_uuid, "_UUID");
1608 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
1610 obd = class_name2obd(lsi->lsi_osd_obdname);
1612 rc = lustre_start_simple(lsi->lsi_osd_obdname,
1614 lsi->lsi_osd_uuid, lmd->lmd_dev,
1615 flagstr, lsi->lsi_lmd->lmd_opts,
1619 obd = class_name2obd(lsi->lsi_osd_obdname);
1623 rc = obd_connect(NULL, &lsi->lsi_osd_exp,
1624 obd, &obd->obd_uuid, NULL, NULL);
1627 class_manual_cleanup(obd);
1628 lsi->lsi_dt_dev = NULL;
1631 /* XXX: to keep support old components relying on lsi_srv_mnt
1632 * we get this info from OSD just started */
1633 LASSERT(obd->obd_lu_dev);
1634 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
1635 LASSERT(lsi->lsi_dt_dev);
1637 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
1639 lsi->lsi_srv_mnt = p.ddp_mnt;
1645 /** Fill in the superblock info for a Lustre server.
1646 * Mount the device with the correct options.
1647 * Read the on-disk config file.
1648 * Start the services.
1650 int server_fill_super(struct super_block *sb)
1652 struct lustre_sb_info *lsi = s2lsi(sb);
1656 rc = lsi_prepare(lsi);
1660 /* Start low level OSD */
1661 rc = osd_start(lsi, sb->s_flags);
1663 CERROR("Unable to start osd on %s: %d\n",
1664 lsi->lsi_lmd->lmd_dev, rc);
1669 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
1670 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
1672 if (class_name2obd(lsi->lsi_svname)) {
1673 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1674 "running. Double-mount may have compromised"
1675 " the disk journal.\n",
1681 /* Start MGS before MGC */
1682 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1683 rc = server_start_mgs(sb);
1688 /* Start MGC before servers */
1689 rc = lustre_start_mgc(sb);
1693 /* Set up all obd devices for service */
1694 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1695 (IS_OST(lsi) || IS_MDT(lsi))) {
1696 rc = server_start_targets(sb, lsi->lsi_srv_mnt);
1698 CERROR("Unable to start targets: %d\n", rc);
1701 /* FIXME overmount client here, or can we just start a
1702 * client log and client_fill_super on this sb? We
1703 * need to make sure server_put_super gets called too
1704 * - ll_put_super calls lustre_common_put_super; check
1705 * there for LSI_SERVER flag, call s_p_s if so.
1707 * Probably should start client from new thread so we
1708 * can return. Client will not finish until all
1709 * servers are connected. Note - MGS-only server does
1710 * NOT get a client, since there is no lustre fs
1711 * associated - the MGS is for all lustre fs's */
1714 rc = server_fill_super_common(sb);
1720 /* We jump here in case of failure while starting targets or MGS.
1721 * In this case we can't just put @mnt and have to do real cleanup
1722 * with stoping targets, etc. */
1723 server_put_super(sb);
1728 * Calculate timeout value for a target.
1730 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1732 struct lustre_mount_data *lmd;
1736 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
1737 int min = OBD_RECOVERY_TIME_MIN;
1739 LASSERT(IS_SERVER(lsi));
1743 soft = lmd->lmd_recovery_time_soft;
1744 hard = lmd->lmd_recovery_time_hard;
1745 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1746 obd->obd_no_ir = !has_ir;
1750 soft = OBD_RECOVERY_TIME_SOFT;
1752 hard = OBD_RECOVERY_TIME_HARD;
1754 /* target may have ir_factor configured. */
1755 factor = OBD_IR_FACTOR_DEFAULT;
1756 if (obd->obd_recovery_ir_factor)
1757 factor = obd->obd_recovery_ir_factor;
1760 int new_soft = soft;
1761 int new_hard = hard;
1763 /* adjust timeout value by imperative recovery */
1765 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1766 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1768 /* make sure the timeout is not too short */
1769 new_soft = max(min, new_soft);
1770 new_hard = max(new_soft, new_hard);
1772 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1773 "window shrunk from %d-%d down to %d-%d\n",
1774 obd->obd_name, soft, hard, new_soft, new_hard);
1781 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1782 obd->obd_recovery_time_hard = hard;
1783 obd->obd_recovery_ir_factor = factor;
1785 EXPORT_SYMBOL(server_calc_timeout);