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) 2011, 2012, 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 <lustre_fsfilt.h>
52 #include <obd_class.h>
53 #include <lustre/lustre_user.h>
54 #include <linux/version.h>
55 #include <lustre_log.h>
56 #include <lustre_disk.h>
57 #include <lustre_param.h>
59 /*********** mount lookup *********/
61 DEFINE_MUTEX(lustre_mount_info_lock);
62 static LIST_HEAD(server_mount_info_list);
64 static struct lustre_mount_info *server_find_mount(const char *name)
66 struct list_head *tmp;
67 struct lustre_mount_info *lmi;
70 list_for_each(tmp, &server_mount_info_list) {
71 lmi = list_entry(tmp, struct lustre_mount_info,
73 if (strcmp(name, lmi->lmi_name) == 0)
79 /* we must register an obd for a mount before we call the setup routine.
80 *_setup will call lustre_get_mount to get the mnt struct
81 by obd_name, since we can't pass the pointer to setup. */
82 static int server_register_mount(const char *name, struct super_block *sb,
85 struct lustre_mount_info *lmi;
91 OBD_ALLOC(lmi, sizeof(*lmi));
94 OBD_ALLOC(name_cp, strlen(name) + 1);
96 OBD_FREE(lmi, sizeof(*lmi));
99 strcpy(name_cp, name);
101 mutex_lock(&lustre_mount_info_lock);
103 if (server_find_mount(name)) {
104 mutex_unlock(&lustre_mount_info_lock);
105 OBD_FREE(lmi, sizeof(*lmi));
106 OBD_FREE(name_cp, strlen(name) + 1);
107 CERROR("Already registered %s\n", name);
110 lmi->lmi_name = name_cp;
113 list_add(&lmi->lmi_list_chain, &server_mount_info_list);
115 mutex_unlock(&lustre_mount_info_lock);
117 CDEBUG(D_MOUNT, "reg_mnt %p from %s\n", lmi->lmi_mnt, name);
122 /* when an obd no longer needs a mount */
123 static int server_deregister_mount(const char *name)
125 struct lustre_mount_info *lmi;
128 mutex_lock(&lustre_mount_info_lock);
129 lmi = server_find_mount(name);
131 mutex_unlock(&lustre_mount_info_lock);
132 CERROR("%s not registered\n", name);
136 CDEBUG(D_MOUNT, "dereg_mnt %p from %s\n", lmi->lmi_mnt, name);
138 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
139 list_del(&lmi->lmi_list_chain);
140 OBD_FREE(lmi, sizeof(*lmi));
141 mutex_unlock(&lustre_mount_info_lock);
146 /* obd's look up a registered mount using their obdname. This is just
147 for initial obd setup to find the mount struct. It should not be
148 called every time you want to mntget. */
149 struct lustre_mount_info *server_get_mount(const char *name)
151 struct lustre_mount_info *lmi;
152 struct lustre_sb_info *lsi;
155 mutex_lock(&lustre_mount_info_lock);
156 lmi = server_find_mount(name);
157 mutex_unlock(&lustre_mount_info_lock);
159 CERROR("Can't find mount for %s\n", name);
162 lsi = s2lsi(lmi->lmi_sb);
164 atomic_inc(&lsi->lsi_mounts);
166 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d\n", lmi->lmi_mnt,
167 name, atomic_read(&lsi->lsi_mounts));
171 EXPORT_SYMBOL(server_get_mount);
174 * Used by mdt to get mount_info from obdname.
175 * There are no blocking when using the mount_info.
176 * Do not use server_get_mount for this purpose.
178 struct lustre_mount_info *server_get_mount_2(const char *name)
180 struct lustre_mount_info *lmi;
183 mutex_lock(&lustre_mount_info_lock);
184 lmi = server_find_mount(name);
185 mutex_unlock(&lustre_mount_info_lock);
187 CERROR("Can't find mount for %s\n", name);
191 EXPORT_SYMBOL(server_get_mount_2);
193 /* to be called from obd_cleanup methods */
194 int server_put_mount(const char *name, struct vfsmount *mnt)
196 struct lustre_mount_info *lmi;
197 struct lustre_sb_info *lsi;
200 mutex_lock(&lustre_mount_info_lock);
201 lmi = server_find_mount(name);
202 mutex_unlock(&lustre_mount_info_lock);
204 CERROR("Can't find mount for %s\n", name);
207 lsi = s2lsi(lmi->lmi_sb);
209 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d\n",
210 lmi->lmi_mnt, name, atomic_read(&lsi->lsi_mounts));
212 if (lustre_put_lsi(lmi->lmi_sb))
213 CDEBUG(D_MOUNT, "Last put of mnt %p from %s\n",
216 /* this obd should never need the mount again */
217 server_deregister_mount(name);
221 EXPORT_SYMBOL(server_put_mount);
223 /* Corresponding to server_get_mount_2 */
224 int server_put_mount_2(const char *name, struct vfsmount *mnt)
229 EXPORT_SYMBOL(server_put_mount_2);
231 /* Set up a MGS to serve startup logs */
232 static int server_start_mgs(struct super_block *sb)
234 struct lustre_sb_info *lsi = s2lsi(sb);
235 struct vfsmount *mnt = lsi->lsi_srv_mnt;
236 struct lustre_mount_info *lmi;
240 /* It is impossible to have more than 1 MGS per node, since
241 MGC wouldn't know which to connect to */
242 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
244 lsi = s2lsi(lmi->lmi_sb);
245 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
250 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
252 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
255 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
256 LUSTRE_MGS_OBDNAME, 0, 0,
257 lsi->lsi_osd_obdname, 0);
258 /* Do NOT call server_deregister_mount() here. This leads to
259 * inability cleanup cleanly and free lsi and other stuff when
260 * mgs calls server_put_mount() in error handling case. -umka */
264 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
265 "Is the 'mgs' module loaded?\n",
266 LUSTRE_MGS_OBDNAME, rc);
270 static int server_stop_mgs(struct super_block *sb)
272 struct obd_device *obd;
276 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
278 /* There better be only one MGS */
279 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
281 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
285 /* The MGS should always stop when we say so */
287 rc = class_manual_cleanup(obd);
291 /* Since there's only one mgc per node, we have to change it's fs to get
292 access to the right disk. */
293 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
295 struct lustre_sb_info *lsi = s2lsi(sb);
299 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
301 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
302 rc = obd_set_info_async(NULL, mgc->obd_self_export,
303 sizeof(KEY_SET_FS), KEY_SET_FS,
304 sizeof(*sb), sb, NULL);
306 CERROR("can't set_fs %d\n", rc);
311 static int server_mgc_clear_fs(struct obd_device *mgc)
316 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
318 rc = obd_set_info_async(NULL, mgc->obd_self_export,
319 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
324 static int is_mdc_device(const char *devname)
328 ptr = strrchr(devname, '-');
329 if (ptr != NULL && strcmp(ptr, "-mdc") == 0)
335 static inline int tgt_is_mdt0(const char *tgtname)
340 type = server_name2index(tgtname, &idx, NULL);
341 if (type != LDD_F_SV_TYPE_MDT)
347 static inline int is_mdc_for_mdt0(const char *devname)
351 if (!is_mdc_device(devname))
354 ptr = strrchr(devname, '-');
359 if (tgt_is_mdt0(devname)) {
368 * Convert OST/MDT name(fsname-OSTxxxx) to a lwp name
369 * (fsname-MDT0000-lwp-OSTxxxx)
371 int tgt_name2lwpname(const char *svname, char *lwpname)
378 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
382 rc = server_name2fsname(svname, fsname, &tgt);
384 CERROR("%s: failed to get fsname from svname. %d\n",
389 if (*tgt != '-' && *tgt != ':') {
390 CERROR("%s: invalid svname name!\n", svname);
391 GOTO(cleanup, rc = -EINVAL);
395 if (strncmp(tgt, "OST", 3) != 0 && strncmp(tgt, "MDT", 3) != 0) {
396 CERROR("%s is not an OST or MDT target!\n", svname);
397 GOTO(cleanup, rc = -EINVAL);
399 sprintf(lwpname, "%s-MDT0000-%s-%s", fsname, LUSTRE_LWP_NAME, tgt);
402 OBD_FREE(fsname, MTI_NAME_MAXLEN);
405 EXPORT_SYMBOL(tgt_name2lwpname);
407 static LIST_HEAD(lwp_register_list);
408 DEFINE_MUTEX(lwp_register_list_lock);
410 int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp,
411 register_lwp_cb cb_func, void *cb_data)
413 struct obd_device *lwp;
414 struct lwp_register_item *lri;
417 LASSERTF(strlen(lwpname) < MTI_NAME_MAXLEN, "lwpname is too long %s\n",
419 LASSERT(exp != NULL && *exp == NULL);
425 mutex_lock(&lwp_register_list_lock);
427 lwp = class_name2obd(lwpname);
428 if (lwp != NULL && lwp->obd_set_up == 1) {
429 struct obd_uuid *uuid;
433 mutex_unlock(&lwp_register_list_lock);
437 memcpy(uuid->uuid, lwpname, strlen(lwpname));
438 *exp = cfs_hash_lookup(lwp->obd_uuid_hash, uuid);
442 memcpy(lri->lri_name, lwpname, strlen(lwpname));
444 lri->lri_cb_func = cb_func;
445 lri->lri_cb_data = cb_data;
446 INIT_LIST_HEAD(&lri->lri_list);
447 list_add(&lri->lri_list, &lwp_register_list);
449 if (*exp != NULL && cb_func != NULL)
452 mutex_unlock(&lwp_register_list_lock);
455 EXPORT_SYMBOL(lustre_register_lwp_item);
457 void lustre_deregister_lwp_item(struct obd_export **exp)
459 struct lwp_register_item *lri, *tmp;
461 mutex_lock(&lwp_register_list_lock);
462 list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
463 if (exp == lri->lri_exp) {
465 class_export_put(*exp);
466 list_del(&lri->lri_list);
471 mutex_unlock(&lwp_register_list_lock);
473 EXPORT_SYMBOL(lustre_deregister_lwp_item);
475 static void lustre_notify_lwp_list(struct obd_export *exp)
477 struct lwp_register_item *lri, *tmp;
478 LASSERT(exp != NULL);
480 mutex_lock(&lwp_register_list_lock);
481 list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
482 if (strcmp(exp->exp_obd->obd_name, lri->lri_name))
484 if (*lri->lri_exp != NULL)
486 *lri->lri_exp = class_export_get(exp);
487 if (lri->lri_cb_func != NULL)
488 lri->lri_cb_func(lri->lri_cb_data);
490 mutex_unlock(&lwp_register_list_lock);
493 static int lustre_lwp_connect(struct obd_device *lwp)
496 struct lu_context session_ctx;
497 struct obd_export *exp;
498 struct obd_uuid *uuid = NULL;
499 struct obd_connect_data *data = NULL;
503 /* log has been fully processed, let clients connect */
504 rc = lu_env_init(&env, lwp->obd_lu_dev->ld_type->ldt_ctx_tags);
508 lu_context_init(&session_ctx, LCT_SESSION);
509 session_ctx.lc_thread = NULL;
510 lu_context_enter(&session_ctx);
511 env.le_ses = &session_ctx;
515 GOTO(out, rc = -ENOMEM);
517 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
518 data->ocd_version = LUSTRE_VERSION_CODE;
519 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
520 OBD_CONNECT_AT | OBD_CONNECT_LRU_RESIZE |
521 OBD_CONNECT_FULL20 | OBD_CONNECT_LVB_TYPE |
522 OBD_CONNECT_LIGHTWEIGHT;
525 GOTO(out, rc = -ENOMEM);
527 if (strlen(lwp->obd_name) > sizeof(uuid->uuid)) {
528 CERROR("%s: Too long lwp name %s, max_size is %d\n",
529 lwp->obd_name, lwp->obd_name, (int)sizeof(uuid->uuid));
530 GOTO(out, rc = -EINVAL);
533 /* Use lwp name as the uuid, so we find the export by lwp name later */
534 memcpy(uuid->uuid, lwp->obd_name, strlen(lwp->obd_name));
535 rc = obd_connect(&env, &exp, lwp, uuid, data, NULL);
537 CERROR("%s: connect failed: rc = %d\n", lwp->obd_name, rc);
539 lustre_notify_lwp_list(exp);
548 lu_context_exit(&session_ctx);
549 lu_context_fini(&session_ctx);
555 * lwp is used by slaves (Non-MDT0 targets) to manage the connection
558 static int lustre_lwp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi)
560 struct obd_connect_data *data = NULL;
561 struct obd_device *obd;
562 char *lwpname = NULL;
563 char *lwpuuid = NULL;
567 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
570 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
574 OBD_ALLOC(lwpname, MTI_NAME_MAXLEN);
576 GOTO(out, rc = -ENOMEM);
578 rc = tgt_name2lwpname(lsi->lsi_svname, lwpname);
580 CERROR("%s: failed to generate lwp name. %d\n",
581 lsi->lsi_svname, rc);
585 OBD_ALLOC(lwpuuid, MTI_NAME_MAXLEN);
587 GOTO(out, rc = -ENOMEM);
589 sprintf(lwpuuid, "%s_UUID", lwpname);
590 rc = lustre_start_simple(lwpname, LUSTRE_LWP_NAME,
591 lwpuuid, lustre_cfg_string(lcfg, 1),
594 CERROR("%s: setup up failed: rc %d\n", lwpname, rc);
598 obd = class_name2obd(lwpname);
599 LASSERT(obd != NULL);
601 rc = lustre_lwp_connect(obd);
603 CERROR("%s: connect failed: rc = %d\n", lwpname, rc);
608 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
610 OBD_FREE(lwpuuid, MTI_NAME_MAXLEN);
615 /* the caller is responsible for memory free */
616 static struct obd_device *lustre_find_lwp(struct lustre_sb_info *lsi,
617 char **lwpname, char **logname)
619 struct obd_device *lwp;
623 LASSERT(lwpname != NULL);
624 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
626 OBD_ALLOC(*lwpname, MTI_NAME_MAXLEN);
627 if (*lwpname == NULL)
628 RETURN(ERR_PTR(-ENOMEM));
630 if (logname != NULL) {
631 OBD_ALLOC(*logname, MTI_NAME_MAXLEN);
632 if (*logname == NULL)
633 GOTO(out, rc = -ENOMEM);
634 rc = server_name2fsname(lsi->lsi_svname, *lwpname, NULL);
636 CERROR("%s: failed to get fsname from svname. %d\n",
637 lsi->lsi_svname, rc);
638 GOTO(out, rc = -EINVAL);
640 sprintf(*logname, "%s-client", *lwpname);
643 rc = tgt_name2lwpname(lsi->lsi_svname, *lwpname);
645 CERROR("%s: failed to generate lwp name. %d\n",
646 lsi->lsi_svname, rc);
647 GOTO(out, rc = -EINVAL);
650 lwp = class_name2obd(*lwpname);
654 if (*lwpname != NULL) {
655 OBD_FREE(*lwpname, MTI_NAME_MAXLEN);
658 if (logname != NULL && *logname != NULL) {
659 OBD_FREE(*logname, MTI_NAME_MAXLEN);
665 RETURN(lwp != NULL ? lwp : ERR_PTR(-ENOENT));
668 static int lustre_lwp_add_conn(struct lustre_cfg *cfg,
669 struct lustre_sb_info *lsi)
671 struct lustre_cfg_bufs *bufs = NULL;
672 struct lustre_cfg *lcfg = NULL;
673 char *lwpname = NULL;
674 struct obd_device *lwp;
678 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
680 CERROR("%s: can't find lwp device.\n", lsi->lsi_svname);
681 GOTO(out, rc = PTR_ERR(lwp));
683 LASSERT(lwpname != NULL);
687 GOTO(out, rc = -ENOMEM);
689 lustre_cfg_bufs_reset(bufs, lwpname);
690 lustre_cfg_bufs_set_string(bufs, 1,
691 lustre_cfg_string(cfg, 1));
693 lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
695 rc = class_add_conn(lwp, lcfg);
697 CERROR("%s: can't add conn: rc = %d\n", lwpname, rc);
703 lustre_cfg_free(lcfg);
705 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
710 * Retrieve MDT nids from the client log, then start the lwp device.
711 * there are only two scenarios which would include mdt nid.
713 * marker 5 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add mdc' xxx-
714 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
715 * attach 0:lustre-MDT0000-mdc 1:mdc 2:lustre-clilmv_UUID
716 * setup 0:lustre-MDT0000-mdc 1:lustre-MDT0000_UUID 2:192.168.122.162@tcp
717 * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
718 * add_conn 0:lustre-MDT0000-mdc 1:192.168.172.1@tcp
719 * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID xxxx
720 * marker 5 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add mdc' xxxx-
722 * marker 7 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
723 * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
724 * add_conn 0:lustre-MDT0000-mdc 1:192.168.122.2@tcp
725 * marker 7 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
727 static int client_lwp_config_process(const struct lu_env *env,
728 struct llog_handle *handle,
729 struct llog_rec_hdr *rec, void *data)
731 struct config_llog_instance *clli = data;
732 int cfg_len = rec->lrh_len;
733 char *cfg_buf = (char *) (rec + 1);
734 struct lustre_cfg *lcfg = NULL;
735 struct lustre_sb_info *lsi;
736 int rc = 0, swab = 0;
739 if (rec->lrh_type != OBD_CFG_REC) {
740 CERROR("Unknown llog record type %#x encountered\n",
745 LASSERT(clli->cfg_sb != NULL);
746 lsi = s2lsi(clli->cfg_sb);
748 lcfg = (struct lustre_cfg *)cfg_buf;
749 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
750 lustre_swab_lustre_cfg(lcfg);
754 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
758 switch (lcfg->lcfg_command) {
760 struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
762 lustre_swab_cfg_marker(marker, swab,
763 LUSTRE_CFG_BUFLEN(lcfg, 1));
764 if (marker->cm_flags & CM_SKIP ||
765 marker->cm_flags & CM_EXCLUDE)
768 if (!tgt_is_mdt0(marker->cm_tgtname))
771 if (!strncmp(marker->cm_comment, "add mdc", 7) ||
772 !strncmp(marker->cm_comment, "add failnid", 11)) {
773 if (marker->cm_flags & CM_START) {
774 clli->cfg_flags = CFG_F_MARKER;
775 /* This hack is to differentiate the
776 * ADD_UUID is come from "add mdc" record
777 * or from "add failnid" record. */
778 if (!strncmp(marker->cm_comment,
780 clli->cfg_flags |= CFG_F_SKIP;
781 } else if (marker->cm_flags & CM_END) {
787 case LCFG_ADD_UUID: {
788 if (clli->cfg_flags == CFG_F_MARKER) {
789 rc = lustre_lwp_setup(lcfg, lsi);
790 /* XXX: process only the first nid as
791 * we don't need another instance of lwp */
792 clli->cfg_flags |= CFG_F_SKIP;
793 } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
794 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
797 CERROR("%s: Fail to add uuid, rc:%d\n",
798 lsi->lsi_svname, rc);
802 case LCFG_ADD_CONN: {
803 if (is_mdc_for_mdt0(lustre_cfg_string(lcfg, 0)))
804 rc = lustre_lwp_add_conn(lcfg, lsi);
814 static int lustre_disconnect_lwp(struct super_block *sb)
816 struct lustre_sb_info *lsi = s2lsi(sb);
817 struct obd_device *lwp;
818 char *lwpname = NULL;
819 char *logname = NULL;
820 struct lustre_cfg *lcfg = NULL;
821 struct lustre_cfg_bufs *bufs = NULL;
822 struct config_llog_instance *cfg = NULL;
826 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
827 if (IS_ERR(lwp) && PTR_ERR(lwp) != -ENOENT)
828 GOTO(out, rc = PTR_ERR(lwp));
830 LASSERT(lwpname != NULL);
831 LASSERT(logname != NULL);
835 GOTO(out, rc = -ENOMEM);
838 cfg->cfg_instance = sb;
839 rc = lustre_end_log(sb, logname, cfg);
841 CERROR("%s: Can't end config log %s.\n", lwpname, logname);
845 if (PTR_ERR(lwp) == -ENOENT) {
846 CDEBUG(D_CONFIG, "%s: lwp device wasn't started.\n",
853 GOTO(out, rc = -ENOMEM);
855 lustre_cfg_bufs_reset(bufs, lwp->obd_name);
856 lustre_cfg_bufs_set_string(bufs, 1, NULL);
857 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
859 GOTO(out, rc = -ENOMEM);
861 /* Disconnect import first. NULL is passed for the '@env', since
862 * it will not be used. */
863 rc = lwp->obd_lu_dev->ld_ops->ldo_process_config(NULL, lwp->obd_lu_dev,
867 lustre_cfg_free(lcfg);
873 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
875 OBD_FREE(logname, MTI_NAME_MAXLEN);
880 * Stop the lwp for an OST/MDT target.
882 static int lustre_stop_lwp(struct super_block *sb)
884 struct lustre_sb_info *lsi = s2lsi(sb);
885 struct obd_device *lwp = NULL;
886 char *lwpname = NULL;
890 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
892 CDEBUG(PTR_ERR(lwp) == -ENOENT ? D_CONFIG : D_ERROR,
893 "%s: lwp wasn't started.\n", lsi->lsi_svname);
898 rc = class_manual_cleanup(lwp);
902 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
907 * Start the lwp(fsname-MDT0000-lwp-OSTxxxx) for an OST or MDT target,
908 * which would be used to establish connection from OST to MDT0.
910 static int lustre_start_lwp(struct super_block *sb)
912 struct lustre_sb_info *lsi = s2lsi(sb);
913 struct config_llog_instance *cfg = NULL;
914 struct obd_device *lwp;
915 char *lwpname = NULL;
916 char *logname = NULL;
920 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
922 /* the lwp device already stared */
923 if (lwp && !IS_ERR(lwp))
926 if (PTR_ERR(lwp) != -ENOENT)
927 GOTO(out, rc = PTR_ERR(lwp));
929 LASSERT(lwpname != NULL);
930 LASSERT(logname != NULL);
934 GOTO(out, rc = -ENOMEM);
936 cfg->cfg_callback = client_lwp_config_process;
937 cfg->cfg_instance = sb;
939 rc = lustre_process_log(sb, logname, cfg);
942 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
944 OBD_FREE(logname, MTI_NAME_MAXLEN);
950 DEFINE_MUTEX(server_start_lock);
952 /* Stop MDS/OSS if nobody is using them */
953 static int server_stop_servers(int lsiflags)
955 struct obd_device *obd = NULL;
956 struct obd_type *type = NULL;
960 mutex_lock(&server_start_lock);
962 /* Either an MDT or an OST or neither */
963 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
964 if (lsiflags & LDD_F_SV_TYPE_MDT) {
965 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
967 type = class_search_type(LUSTRE_MDT_NAME);
970 /* if this was an OST, and there are no more OST's, clean up the OSS */
971 if (lsiflags & LDD_F_SV_TYPE_OST) {
972 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
974 type = class_search_type(LUSTRE_OST_NAME);
977 if (obd != NULL && (type == NULL || type->typ_refcnt == 0)) {
981 /* obd_fail doesn't mean much on a server obd */
982 err = class_manual_cleanup(obd);
987 mutex_unlock(&server_start_lock);
992 int server_mti_print(const char *title, struct mgs_target_info *mti)
994 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
995 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
996 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
997 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
998 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
999 mti->mti_config_ver, mti->mti_flags);
1002 EXPORT_SYMBOL(server_mti_print);
1004 /* Generate data for registration */
1005 static int server_lsi2mti(struct lustre_sb_info *lsi,
1006 struct mgs_target_info *mti)
1008 lnet_process_id_t id;
1013 if (!IS_SERVER(lsi))
1016 if (strlcpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname))
1017 >= sizeof(mti->mti_svname))
1020 mti->mti_nid_count = 0;
1021 while (LNetGetId(i++, &id) != -ENOENT) {
1022 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1025 /* server use --servicenode param, only allow specified
1026 * nids be registered */
1027 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
1028 class_match_nid(lsi->lsi_lmd->lmd_params,
1029 PARAM_FAILNODE, id.nid) < 1)
1032 /* match specified network */
1033 if (!class_match_net(lsi->lsi_lmd->lmd_params,
1034 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1037 mti->mti_nids[mti->mti_nid_count] = id.nid;
1038 mti->mti_nid_count++;
1039 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1040 CWARN("Only using first %d nids for %s\n",
1041 mti->mti_nid_count, mti->mti_svname);
1046 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1047 mti->mti_config_ver = 0;
1049 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1053 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1056 /* Orion requires index to be set */
1057 LASSERT(!(rc & LDD_F_NEED_INDEX));
1058 /* keep only LDD flags */
1059 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1060 if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN))
1061 mti->mti_flags |= LDD_F_UPDATE;
1062 cplen = strlcpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1063 sizeof(mti->mti_params));
1064 if (cplen >= sizeof(mti->mti_params))
1069 /* Register an old or new target with the MGS. If needed MGS will construct
1070 startup logs and assign index */
1071 static int server_register_target(struct lustre_sb_info *lsi)
1073 struct obd_device *mgc = lsi->lsi_mgc;
1074 struct mgs_target_info *mti = NULL;
1081 if (!IS_SERVER(lsi))
1088 rc = server_lsi2mti(lsi, mti);
1092 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1093 mti->mti_svname, mti->mti_fsname,
1094 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1097 /* if write_conf is true, the registration must succeed */
1098 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1099 mti->mti_flags |= LDD_F_OPC_REG;
1101 /* Register the target */
1102 /* FIXME use mgc_process_config instead */
1103 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1104 sizeof(KEY_REGISTER_TARGET),
1105 KEY_REGISTER_TARGET,
1106 sizeof(*mti), mti, NULL);
1108 if (mti->mti_flags & LDD_F_ERROR) {
1109 LCONSOLE_ERROR_MSG(0x160,
1110 "%s: the MGS refuses to allow this server "
1111 "to start: rc = %d. Please see messages on "
1112 "the MGS.\n", lsi->lsi_svname, rc);
1113 } else if (writeconf) {
1114 LCONSOLE_ERROR_MSG(0x15f,
1115 "%s: cannot register this server with the MGS: "
1116 "rc = %d. Is the MGS running?\n",
1117 lsi->lsi_svname, rc);
1119 CERROR("%s: error registering with the MGS: rc = %d "
1120 "(not fatal)\n", lsi->lsi_svname, rc);
1121 /* reset the error code for non-fatal error. */
1134 * Notify the MGS that this target is ready.
1135 * Used by IR - if the MGS receives this message, it will notify clients.
1137 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1139 struct lustre_sb_info *lsi = s2lsi(sb);
1140 struct obd_device *mgc = lsi->lsi_mgc;
1141 struct mgs_target_info *mti = NULL;
1147 if (!(IS_SERVER(lsi)))
1153 rc = server_lsi2mti(lsi, mti);
1157 mti->mti_instance = obd->u.obt.obt_instance;
1158 mti->mti_flags |= LDD_F_OPC_READY;
1160 /* FIXME use mgc_process_config instead */
1161 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1162 sizeof(KEY_REGISTER_TARGET),
1163 KEY_REGISTER_TARGET,
1164 sizeof(*mti), mti, NULL);
1166 /* Imperative recovery: if the mgs informs us to use IR? */
1167 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1168 (mti->mti_flags & LDD_F_IR_CAPABLE))
1169 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1178 /** Start server targets: MDTs and OSTs
1180 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1182 struct obd_device *obd;
1183 struct lustre_sb_info *lsi = s2lsi(sb);
1184 struct config_llog_instance cfg;
1186 struct lu_device *dev;
1190 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1193 /* make sure the MDS is started */
1194 mutex_lock(&server_start_lock);
1195 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1197 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1199 LUSTRE_MDS_OBDNAME"_uuid",
1202 mutex_unlock(&server_start_lock);
1203 CERROR("failed to start MDS: %d\n", rc);
1207 mutex_unlock(&server_start_lock);
1210 /* If we're an OST, make sure the global OSS is running */
1212 /* make sure OSS is started */
1213 mutex_lock(&server_start_lock);
1214 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1216 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1218 LUSTRE_OSS_OBDNAME"_uuid",
1221 mutex_unlock(&server_start_lock);
1222 CERROR("failed to start OSS: %d\n", rc);
1226 mutex_unlock(&server_start_lock);
1229 /* Set the mgc fs to our server disk. This allows the MGC to
1230 * read and write configs locally, in case it can't talk to the MGS. */
1231 if (lsi->lsi_srv_mnt) {
1232 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1234 GOTO(out_stop_service, rc);
1237 /* Register with MGS */
1238 rc = server_register_target(lsi);
1242 /* Let the target look up the mount using the target's name
1243 (we can't pass the sb or mnt through class_process_config.) */
1244 rc = server_register_mount(lsi->lsi_svname, sb, mnt);
1248 /* Start targets using the llog named for the target */
1249 memset(&cfg, 0, sizeof(cfg));
1250 cfg.cfg_callback = class_config_llog_handler;
1251 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1253 CERROR("failed to start server %s: %d\n",
1254 lsi->lsi_svname, rc);
1255 /* Do NOT call server_deregister_mount() here. This makes it
1256 * impossible to find mount later in cleanup time and leaves
1257 * @lsi and othder stuff leaked. -umka */
1261 obd = class_name2obd(lsi->lsi_svname);
1263 CERROR("no server named %s was started\n", lsi->lsi_svname);
1264 GOTO(out_mgc, rc = -ENXIO);
1267 if (IS_OST(lsi) || IS_MDT(lsi)) {
1268 rc = lustre_start_lwp(sb);
1270 CERROR("%s: failed to start LWP: %d\n",
1271 lsi->lsi_svname, rc);
1276 server_notify_target(sb, obd);
1278 /* calculate recovery timeout, do it after lustre_process_log */
1279 server_calc_timeout(lsi, obd);
1281 /* log has been fully processed */
1282 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1284 /* log has been fully processed, let clients connect */
1285 dev = obd->obd_lu_dev;
1286 if (dev && dev->ld_ops->ldo_prepare) {
1287 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1289 struct lu_context session_ctx;
1291 lu_context_init(&session_ctx, LCT_SESSION);
1292 session_ctx.lc_thread = NULL;
1293 lu_context_enter(&session_ctx);
1294 env.le_ses = &session_ctx;
1296 rc = dev->ld_ops->ldo_prepare(&env, NULL, dev);
1299 lu_context_exit(&session_ctx);
1300 lu_context_fini(&session_ctx);
1304 /* abort recovery only on the complete stack:
1305 * many devices can be involved */
1306 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1307 (OBP(obd, iocontrol))) {
1308 obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
1313 /* Release the mgc fs for others to use */
1314 if (lsi->lsi_srv_mnt)
1315 server_mgc_clear_fs(lsi->lsi_mgc);
1319 server_stop_servers(lsi->lsi_flags);
1324 static int lsi_prepare(struct lustre_sb_info *lsi)
1331 LASSERT(lsi->lsi_lmd);
1333 /* The server name is given as a mount line option */
1334 if (lsi->lsi_lmd->lmd_profile == NULL) {
1335 LCONSOLE_ERROR("Can't determine server name\n");
1339 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
1340 RETURN(-ENAMETOOLONG);
1342 strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
1344 /* Determine osd type */
1345 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
1346 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
1347 sizeof(lsi->lsi_osd_type))
1348 RETURN(-ENAMETOOLONG);
1350 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
1352 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
1355 /* XXX: a temp. solution for components using fsfilt
1356 * to be removed in one of the subsequent patches */
1357 if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs"))
1358 strcpy(lsi->lsi_fstype, "ldiskfs");
1360 strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
1362 /* Determine server type */
1363 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1365 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
1366 /* Assume we're a bare MGS */
1368 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
1370 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
1375 lsi->lsi_flags |= rc;
1377 /* Add mount line flags that used to be in ldd:
1378 * writeconf, mgs, anything else?
1380 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
1381 LDD_F_WRITECONF : 0;
1382 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
1384 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_UPDATE) ?
1386 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
1387 LDD_F_SV_TYPE_MGS : 0;
1388 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
1389 LDD_F_NO_PRIMNODE : 0;
1394 /*************** server mount ******************/
1396 /** Start the shutdown of servers at umount.
1398 static void server_put_super(struct super_block *sb)
1400 struct lustre_sb_info *lsi = s2lsi(sb);
1401 struct obd_device *obd;
1402 char *tmpname, *extraname = NULL;
1404 int lsiflags = lsi->lsi_flags;
1407 LASSERT(IS_SERVER(lsi));
1409 tmpname_sz = strlen(lsi->lsi_svname) + 1;
1410 OBD_ALLOC(tmpname, tmpname_sz);
1411 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
1412 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1413 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1414 snprintf(tmpname, tmpname_sz, "MGS");
1416 /* disconnect the lwp first to drain off the inflight request */
1417 if (IS_OST(lsi) || IS_MDT(lsi)) {
1420 rc = lustre_disconnect_lwp(sb);
1421 if (rc && rc != ETIMEDOUT)
1422 CERROR("%s: failed to disconnect lwp. (rc=%d)\n",
1426 /* Stop the target */
1427 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1428 (IS_MDT(lsi) || IS_OST(lsi))) {
1429 struct lustre_profile *lprof = NULL;
1431 /* tell the mgc to drop the config log */
1432 lustre_end_log(sb, lsi->lsi_svname, NULL);
1434 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1435 If there are any setup/cleanup errors, save the lov
1436 name for safety cleanup later. */
1437 lprof = class_get_profile(lsi->lsi_svname);
1438 if (lprof && lprof->lp_dt) {
1439 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1440 strcpy(extraname, lprof->lp_dt);
1443 obd = class_name2obd(lsi->lsi_svname);
1445 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1446 if (lsiflags & LSI_UMOUNT_FAILOVER)
1448 /* We can't seem to give an error return code
1449 * to .put_super, so we better make sure we clean up! */
1451 class_manual_cleanup(obd);
1453 CERROR("no obd %s\n", lsi->lsi_svname);
1454 server_deregister_mount(lsi->lsi_svname);
1458 /* If they wanted the mgs to stop separately from the mdt, they
1459 should have put it on a different device. */
1461 /* if MDS start with --nomgs, don't stop MGS then */
1462 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1463 server_stop_mgs(sb);
1466 if (IS_OST(lsi) || IS_MDT(lsi)) {
1467 if (lustre_stop_lwp(sb) < 0)
1468 CERROR("%s: failed to stop lwp!\n", tmpname);
1471 /* Clean the mgc and sb */
1472 lustre_common_put_super(sb);
1474 /* wait till all in-progress cleanups are done
1475 * specifically we're interested in ofd cleanup
1477 obd_zombie_barrier();
1479 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1480 until the target is really gone so that our type refcount check
1482 server_stop_servers(lsiflags);
1484 /* In case of startup or cleanup err, stop related obds */
1486 obd = class_name2obd(extraname);
1488 CWARN("Cleaning orphaned obd %s\n", extraname);
1490 class_manual_cleanup(obd);
1492 OBD_FREE(extraname, strlen(extraname) + 1);
1495 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1496 OBD_FREE(tmpname, tmpname_sz);
1500 /** Called only for 'umount -f'
1502 static void server_umount_begin(struct super_block *sb)
1504 struct lustre_sb_info *lsi = s2lsi(sb);
1507 CDEBUG(D_MOUNT, "umount -f\n");
1508 /* umount = failover
1510 no third way to do non-force, non-failover */
1511 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1515 static int server_statfs(struct dentry *dentry, struct kstatfs *buf)
1517 struct super_block *sb = dentry->d_sb;
1518 struct lustre_sb_info *lsi = s2lsi(sb);
1519 struct obd_statfs statfs;
1523 if (lsi->lsi_dt_dev) {
1524 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
1526 statfs_unpack(buf, &statfs);
1527 buf->f_type = sb->s_magic;
1533 buf->f_type = sb->s_magic;
1534 buf->f_bsize = sb->s_blocksize;
1540 buf->f_namelen = NAME_MAX;
1544 /** The operations we support directly on the superblock:
1545 * mount, umount, and df.
1547 static struct super_operations server_ops = {
1548 .put_super = server_put_super,
1549 .umount_begin = server_umount_begin, /* umount -f */
1550 .statfs = server_statfs,
1553 #define log2(n) ffz(~(n))
1554 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1556 static int server_fill_super_common(struct super_block *sb)
1558 struct inode *root = 0;
1561 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1563 sb->s_blocksize = 4096;
1564 sb->s_blocksize_bits = log2(sb->s_blocksize);
1565 sb->s_magic = LUSTRE_SUPER_MAGIC;
1566 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1567 sb->s_flags |= MS_RDONLY;
1568 sb->s_op = &server_ops;
1570 root = new_inode(sb);
1572 CERROR("Can't make root inode\n");
1576 /* returns -EIO for every operation */
1577 /* make_bad_inode(root); -- badness - can't umount */
1578 /* apparently we need to be a directory for the mount to finish */
1579 root->i_mode = S_IFDIR;
1581 sb->s_root = d_make_root(root);
1583 CERROR("%s: can't make root dentry\n", sb->s_id);
1590 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
1592 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1593 struct obd_device *obd;
1594 struct dt_device_param p;
1600 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
1601 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
1603 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
1604 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
1605 strcat(lsi->lsi_osd_uuid, "_UUID");
1606 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
1608 obd = class_name2obd(lsi->lsi_osd_obdname);
1610 rc = lustre_start_simple(lsi->lsi_osd_obdname,
1612 lsi->lsi_osd_uuid, lmd->lmd_dev,
1613 flagstr, lsi->lsi_lmd->lmd_opts,
1617 obd = class_name2obd(lsi->lsi_osd_obdname);
1621 rc = obd_connect(NULL, &lsi->lsi_osd_exp,
1622 obd, &obd->obd_uuid, NULL, NULL);
1625 class_manual_cleanup(obd);
1626 lsi->lsi_dt_dev = NULL;
1629 /* XXX: to keep support old components relying on lsi_srv_mnt
1630 * we get this info from OSD just started */
1631 LASSERT(obd->obd_lu_dev);
1632 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
1633 LASSERT(lsi->lsi_dt_dev);
1635 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
1637 lsi->lsi_srv_mnt = p.ddp_mnt;
1643 /** Fill in the superblock info for a Lustre server.
1644 * Mount the device with the correct options.
1645 * Read the on-disk config file.
1646 * Start the services.
1648 int server_fill_super(struct super_block *sb)
1650 struct lustre_sb_info *lsi = s2lsi(sb);
1654 rc = lsi_prepare(lsi);
1658 /* Start low level OSD */
1659 rc = osd_start(lsi, sb->s_flags);
1661 CERROR("Unable to start osd on %s: %d\n",
1662 lsi->lsi_lmd->lmd_dev, rc);
1667 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
1668 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
1670 if (class_name2obd(lsi->lsi_svname)) {
1671 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1672 "running. Double-mount may have compromised"
1673 " the disk journal.\n",
1679 /* Start MGS before MGC */
1680 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1681 rc = server_start_mgs(sb);
1686 /* Start MGC before servers */
1687 rc = lustre_start_mgc(sb);
1691 /* Set up all obd devices for service */
1692 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1693 (IS_OST(lsi) || IS_MDT(lsi))) {
1694 rc = server_start_targets(sb, lsi->lsi_srv_mnt);
1696 CERROR("Unable to start targets: %d\n", rc);
1699 /* FIXME overmount client here, or can we just start a
1700 * client log and client_fill_super on this sb? We
1701 * need to make sure server_put_super gets called too
1702 * - ll_put_super calls lustre_common_put_super; check
1703 * there for LSI_SERVER flag, call s_p_s if so.
1705 * Probably should start client from new thread so we
1706 * can return. Client will not finish until all
1707 * servers are connected. Note - MGS-only server does
1708 * NOT get a client, since there is no lustre fs
1709 * associated - the MGS is for all lustre fs's */
1712 rc = server_fill_super_common(sb);
1718 /* We jump here in case of failure while starting targets or MGS.
1719 * In this case we can't just put @mnt and have to do real cleanup
1720 * with stoping targets, etc. */
1721 server_put_super(sb);
1726 * Calculate timeout value for a target.
1728 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1730 struct lustre_mount_data *lmd;
1734 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
1735 int min = OBD_RECOVERY_TIME_MIN;
1737 LASSERT(IS_SERVER(lsi));
1741 soft = lmd->lmd_recovery_time_soft;
1742 hard = lmd->lmd_recovery_time_hard;
1743 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1744 obd->obd_no_ir = !has_ir;
1748 soft = OBD_RECOVERY_TIME_SOFT;
1750 hard = OBD_RECOVERY_TIME_HARD;
1752 /* target may have ir_factor configured. */
1753 factor = OBD_IR_FACTOR_DEFAULT;
1754 if (obd->obd_recovery_ir_factor)
1755 factor = obd->obd_recovery_ir_factor;
1758 int new_soft = soft;
1759 int new_hard = hard;
1761 /* adjust timeout value by imperative recovery */
1763 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1764 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1766 /* make sure the timeout is not too short */
1767 new_soft = max(min, new_soft);
1768 new_hard = max(new_soft, new_hard);
1770 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1771 "window shrunk from %d-%d down to %d-%d\n",
1772 obd->obd_name, soft, hard, new_soft, new_hard);
1779 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1780 obd->obd_recovery_time_hard = hard;
1781 obd->obd_recovery_ir_factor = factor;
1783 EXPORT_SYMBOL(server_calc_timeout);