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)
88 struct lustre_mount_info *lmi;
94 OBD_ALLOC(lmi, sizeof(*lmi));
97 OBD_ALLOC(name_cp, strlen(name) + 1);
99 OBD_FREE(lmi, sizeof(*lmi));
102 strcpy(name_cp, name);
104 mutex_lock(&lustre_mount_info_lock);
106 if (server_find_mount(name)) {
107 mutex_unlock(&lustre_mount_info_lock);
108 OBD_FREE(lmi, sizeof(*lmi));
109 OBD_FREE(name_cp, strlen(name) + 1);
110 CERROR("Already registered %s\n", name);
113 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, "register mount %p from %s\n", sb, 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, "deregister mount %p from %s\n", lmi->lmi_sb, 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 atomic_inc(&lsi->lsi_mounts);
168 CDEBUG(D_MOUNT, "get mount %p from %s, refs=%d\n", lmi->lmi_sb,
169 name, 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)
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 mount %p from %s, refs=%d\n",
212 lmi->lmi_sb, name, atomic_read(&lsi->lsi_mounts));
214 if (lustre_put_lsi(lmi->lmi_sb))
215 CDEBUG(D_MOUNT, "Last put of mount %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 lustre_mount_info *lmi;
241 /* It is impossible to have more than 1 MGS per node, since
242 MGC wouldn't know which to connect to */
243 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
245 lsi = s2lsi(lmi->lmi_sb);
246 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
251 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
253 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb);
256 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
257 LUSTRE_MGS_OBDNAME, 0, 0,
258 lsi->lsi_osd_obdname, 0);
259 /* server_deregister_mount() is not called previously, for lsi
260 * and other stuff can't be freed cleanly when mgs calls
261 * server_put_mount() in error handling case (see b=17758),
262 * this problem is caused by a bug in mgs_init0, which forgot
263 * calling server_put_mount in error case. */
266 server_deregister_mount(LUSTRE_MGS_OBDNAME);
270 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
271 "Is the 'mgs' module loaded?\n",
272 LUSTRE_MGS_OBDNAME, rc);
276 static int server_stop_mgs(struct super_block *sb)
278 struct obd_device *obd;
282 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
284 /* There better be only one MGS */
285 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
287 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
291 /* The MGS should always stop when we say so */
293 rc = class_manual_cleanup(obd);
297 /* Since there's only one mgc per node, we have to change it's fs to get
298 access to the right disk. */
299 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
301 struct lustre_sb_info *lsi = s2lsi(sb);
305 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
307 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
308 rc = obd_set_info_async(NULL, mgc->obd_self_export,
309 sizeof(KEY_SET_FS), KEY_SET_FS,
310 sizeof(*sb), sb, NULL);
312 CERROR("can't set_fs %d\n", rc);
317 static int server_mgc_clear_fs(struct obd_device *mgc)
322 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
324 rc = obd_set_info_async(NULL, mgc->obd_self_export,
325 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
330 static int is_mdc_device(const char *devname)
334 ptr = strrchr(devname, '-');
335 if (ptr != NULL && strcmp(ptr, "-mdc") == 0)
341 static inline int tgt_is_mdt0(const char *tgtname)
346 type = server_name2index(tgtname, &idx, NULL);
347 if (type != LDD_F_SV_TYPE_MDT)
353 static inline int is_mdc_for_mdt0(const char *devname)
357 if (!is_mdc_device(devname))
360 ptr = strrchr(devname, '-');
365 if (tgt_is_mdt0(devname)) {
374 * Convert OST/MDT name(fsname-OSTxxxx) to a lwp name
375 * (fsname-MDT0000-lwp-OSTxxxx)
377 int tgt_name2lwpname(const char *svname, char *lwpname)
384 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
388 rc = server_name2fsname(svname, fsname, &tgt);
390 CERROR("%s: failed to get fsname from svname. %d\n",
395 if (*tgt != '-' && *tgt != ':') {
396 CERROR("%s: invalid svname name!\n", svname);
397 GOTO(cleanup, rc = -EINVAL);
401 if (strncmp(tgt, "OST", 3) != 0 && strncmp(tgt, "MDT", 3) != 0) {
402 CERROR("%s is not an OST or MDT target!\n", svname);
403 GOTO(cleanup, rc = -EINVAL);
405 sprintf(lwpname, "%s-MDT0000-%s-%s", fsname, LUSTRE_LWP_NAME, tgt);
408 OBD_FREE(fsname, MTI_NAME_MAXLEN);
411 EXPORT_SYMBOL(tgt_name2lwpname);
413 static CFS_LIST_HEAD(lwp_register_list);
414 DEFINE_MUTEX(lwp_register_list_lock);
416 int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp,
417 register_lwp_cb cb_func, void *cb_data)
419 struct obd_device *lwp;
420 struct lwp_register_item *lri;
423 LASSERTF(strlen(lwpname) < MTI_NAME_MAXLEN, "lwpname is too long %s\n",
425 LASSERT(exp != NULL && *exp == NULL);
431 mutex_lock(&lwp_register_list_lock);
433 lwp = class_name2obd(lwpname);
434 if (lwp != NULL && lwp->obd_set_up == 1) {
435 struct obd_uuid *uuid;
439 mutex_unlock(&lwp_register_list_lock);
443 memcpy(uuid->uuid, lwpname, strlen(lwpname));
444 *exp = cfs_hash_lookup(lwp->obd_uuid_hash, uuid);
448 memcpy(lri->lri_name, lwpname, strlen(lwpname));
450 lri->lri_cb_func = cb_func;
451 lri->lri_cb_data = cb_data;
452 CFS_INIT_LIST_HEAD(&lri->lri_list);
453 cfs_list_add(&lri->lri_list, &lwp_register_list);
455 if (*exp != NULL && cb_func != NULL)
458 mutex_unlock(&lwp_register_list_lock);
461 EXPORT_SYMBOL(lustre_register_lwp_item);
463 void lustre_deregister_lwp_item(struct obd_export **exp)
465 struct lwp_register_item *lri, *tmp;
467 mutex_lock(&lwp_register_list_lock);
468 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
469 if (exp == lri->lri_exp) {
471 class_export_put(*exp);
472 cfs_list_del(&lri->lri_list);
477 mutex_unlock(&lwp_register_list_lock);
479 EXPORT_SYMBOL(lustre_deregister_lwp_item);
481 static void lustre_notify_lwp_list(struct obd_export *exp)
483 struct lwp_register_item *lri, *tmp;
484 LASSERT(exp != NULL);
486 mutex_lock(&lwp_register_list_lock);
487 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
488 if (strcmp(exp->exp_obd->obd_name, lri->lri_name))
490 if (*lri->lri_exp != NULL)
492 *lri->lri_exp = class_export_get(exp);
493 if (lri->lri_cb_func != NULL)
494 lri->lri_cb_func(lri->lri_cb_data);
496 mutex_unlock(&lwp_register_list_lock);
499 static int lustre_lwp_connect(struct obd_device *lwp)
502 struct lu_context session_ctx;
503 struct obd_export *exp;
504 struct obd_uuid *uuid = NULL;
505 struct obd_connect_data *data = NULL;
509 /* log has been fully processed, let clients connect */
510 rc = lu_env_init(&env, lwp->obd_lu_dev->ld_type->ldt_ctx_tags);
514 lu_context_init(&session_ctx, LCT_SERVER_SESSION);
515 session_ctx.lc_thread = NULL;
516 lu_context_enter(&session_ctx);
517 env.le_ses = &session_ctx;
521 GOTO(out, rc = -ENOMEM);
523 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
524 data->ocd_version = LUSTRE_VERSION_CODE;
525 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
526 OBD_CONNECT_AT | OBD_CONNECT_LRU_RESIZE |
527 OBD_CONNECT_FULL20 | OBD_CONNECT_LVB_TYPE |
528 OBD_CONNECT_LIGHTWEIGHT;
531 GOTO(out, rc = -ENOMEM);
533 if (strlen(lwp->obd_name) > sizeof(uuid->uuid)) {
534 CERROR("%s: Too long lwp name %s, max_size is %d\n",
535 lwp->obd_name, lwp->obd_name, (int)sizeof(uuid->uuid));
536 GOTO(out, rc = -EINVAL);
539 /* Use lwp name as the uuid, so we find the export by lwp name later */
540 memcpy(uuid->uuid, lwp->obd_name, strlen(lwp->obd_name));
541 rc = obd_connect(&env, &exp, lwp, uuid, data, NULL);
543 CERROR("%s: connect failed: rc = %d\n", lwp->obd_name, rc);
545 lustre_notify_lwp_list(exp);
554 lu_context_exit(&session_ctx);
555 lu_context_fini(&session_ctx);
561 * lwp is used by slaves (Non-MDT0 targets) to manage the connection
564 static int lustre_lwp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi)
566 struct obd_device *obd;
567 char *lwpname = NULL;
568 char *lwpuuid = NULL;
572 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
575 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
579 OBD_ALLOC(lwpname, MTI_NAME_MAXLEN);
581 GOTO(out, rc = -ENOMEM);
583 rc = tgt_name2lwpname(lsi->lsi_svname, lwpname);
585 CERROR("%s: failed to generate lwp name. %d\n",
586 lsi->lsi_svname, rc);
590 OBD_ALLOC(lwpuuid, MTI_NAME_MAXLEN);
592 GOTO(out, rc = -ENOMEM);
594 sprintf(lwpuuid, "%s_UUID", lwpname);
595 rc = lustre_start_simple(lwpname, LUSTRE_LWP_NAME,
596 lwpuuid, lustre_cfg_string(lcfg, 1),
599 CERROR("%s: setup up failed: rc %d\n", lwpname, rc);
603 obd = class_name2obd(lwpname);
604 LASSERT(obd != NULL);
606 rc = lustre_lwp_connect(obd);
608 CERROR("%s: connect failed: rc = %d\n", lwpname, rc);
611 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
613 OBD_FREE(lwpuuid, MTI_NAME_MAXLEN);
618 /* the caller is responsible for memory free */
619 static struct obd_device *lustre_find_lwp(struct lustre_sb_info *lsi,
620 char **lwpname, char **logname)
622 struct obd_device *lwp;
626 LASSERT(lwpname != NULL);
627 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
629 OBD_ALLOC(*lwpname, MTI_NAME_MAXLEN);
630 if (*lwpname == NULL)
631 RETURN(ERR_PTR(-ENOMEM));
633 if (logname != NULL) {
634 OBD_ALLOC(*logname, MTI_NAME_MAXLEN);
635 if (*logname == NULL)
636 GOTO(out, rc = -ENOMEM);
637 rc = server_name2fsname(lsi->lsi_svname, *lwpname, NULL);
639 CERROR("%s: failed to get fsname from svname. %d\n",
640 lsi->lsi_svname, rc);
641 GOTO(out, rc = -EINVAL);
643 sprintf(*logname, "%s-client", *lwpname);
646 rc = tgt_name2lwpname(lsi->lsi_svname, *lwpname);
648 CERROR("%s: failed to generate lwp name. %d\n",
649 lsi->lsi_svname, rc);
650 GOTO(out, rc = -EINVAL);
653 lwp = class_name2obd(*lwpname);
657 if (*lwpname != NULL) {
658 OBD_FREE(*lwpname, MTI_NAME_MAXLEN);
661 if (logname != NULL && *logname != NULL) {
662 OBD_FREE(*logname, MTI_NAME_MAXLEN);
668 RETURN(lwp != NULL ? lwp : ERR_PTR(-ENOENT));
671 static int lustre_lwp_add_conn(struct lustre_cfg *cfg,
672 struct lustre_sb_info *lsi)
674 struct lustre_cfg_bufs *bufs = NULL;
675 struct lustre_cfg *lcfg = NULL;
676 char *lwpname = NULL;
677 struct obd_device *lwp;
681 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
683 CERROR("%s: can't find lwp device.\n", lsi->lsi_svname);
684 GOTO(out, rc = PTR_ERR(lwp));
686 LASSERT(lwpname != NULL);
690 GOTO(out, rc = -ENOMEM);
692 lustre_cfg_bufs_reset(bufs, lwpname);
693 lustre_cfg_bufs_set_string(bufs, 1,
694 lustre_cfg_string(cfg, 1));
696 lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
698 rc = class_add_conn(lwp, lcfg);
700 CERROR("%s: can't add conn: rc = %d\n", lwpname, rc);
706 lustre_cfg_free(lcfg);
708 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
713 * Retrieve MDT nids from the client log, then start the lwp device.
714 * there are only two scenarios which would include mdt nid.
716 * marker 5 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add mdc' xxx-
717 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
718 * attach 0:lustre-MDT0000-mdc 1:mdc 2:lustre-clilmv_UUID
719 * setup 0:lustre-MDT0000-mdc 1:lustre-MDT0000_UUID 2:192.168.122.162@tcp
720 * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
721 * add_conn 0:lustre-MDT0000-mdc 1:192.168.172.1@tcp
722 * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID xxxx
723 * marker 5 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add mdc' xxxx-
725 * marker 7 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
726 * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
727 * add_conn 0:lustre-MDT0000-mdc 1:192.168.122.2@tcp
728 * marker 7 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
730 static int client_lwp_config_process(const struct lu_env *env,
731 struct llog_handle *handle,
732 struct llog_rec_hdr *rec, void *data)
734 struct config_llog_instance *clli = data;
735 int cfg_len = rec->lrh_len;
736 char *cfg_buf = (char *) (rec + 1);
737 struct lustre_cfg *lcfg = NULL;
738 struct lustre_sb_info *lsi;
739 int rc = 0, swab = 0;
742 if (rec->lrh_type != OBD_CFG_REC) {
743 CERROR("Unknown llog record type %#x encountered\n",
748 LASSERT(clli->cfg_sb != NULL);
749 lsi = s2lsi(clli->cfg_sb);
751 lcfg = (struct lustre_cfg *)cfg_buf;
752 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
753 lustre_swab_lustre_cfg(lcfg);
757 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
761 switch (lcfg->lcfg_command) {
763 struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
765 lustre_swab_cfg_marker(marker, swab,
766 LUSTRE_CFG_BUFLEN(lcfg, 1));
767 if (marker->cm_flags & CM_SKIP ||
768 marker->cm_flags & CM_EXCLUDE)
771 if (!tgt_is_mdt0(marker->cm_tgtname))
774 if (!strncmp(marker->cm_comment, "add mdc", 7) ||
775 !strncmp(marker->cm_comment, "add failnid", 11)) {
776 if (marker->cm_flags & CM_START) {
777 clli->cfg_flags = CFG_F_MARKER;
778 /* This hack is to differentiate the
779 * ADD_UUID is come from "add mdc" record
780 * or from "add failnid" record. */
781 if (!strncmp(marker->cm_comment,
783 clli->cfg_flags |= CFG_F_SKIP;
784 } else if (marker->cm_flags & CM_END) {
790 case LCFG_ADD_UUID: {
791 if (clli->cfg_flags == CFG_F_MARKER) {
792 rc = lustre_lwp_setup(lcfg, lsi);
793 /* XXX: process only the first nid as
794 * we don't need another instance of lwp */
795 clli->cfg_flags |= CFG_F_SKIP;
796 } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
797 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
800 CERROR("%s: Fail to add uuid, rc:%d\n",
801 lsi->lsi_svname, rc);
805 case LCFG_ADD_CONN: {
806 if (is_mdc_for_mdt0(lustre_cfg_string(lcfg, 0)))
807 rc = lustre_lwp_add_conn(lcfg, lsi);
817 static int lustre_disconnect_lwp(struct super_block *sb)
819 struct lustre_sb_info *lsi = s2lsi(sb);
820 struct obd_device *lwp;
821 char *lwpname = NULL;
822 char *logname = NULL;
823 struct lustre_cfg *lcfg = NULL;
824 struct lustre_cfg_bufs *bufs = NULL;
825 struct config_llog_instance *cfg = NULL;
829 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
830 if (IS_ERR(lwp) && PTR_ERR(lwp) != -ENOENT)
831 GOTO(out, rc = PTR_ERR(lwp));
833 LASSERT(lwpname != NULL);
834 LASSERT(logname != NULL);
838 GOTO(out, rc = -ENOMEM);
841 cfg->cfg_instance = sb;
842 rc = lustre_end_log(sb, logname, cfg);
844 CERROR("%s: Can't end config log %s.\n", lwpname, logname);
848 if (PTR_ERR(lwp) == -ENOENT) {
849 CDEBUG(D_CONFIG, "%s: lwp device wasn't started.\n",
856 GOTO(out, rc = -ENOMEM);
858 lustre_cfg_bufs_reset(bufs, lwp->obd_name);
859 lustre_cfg_bufs_set_string(bufs, 1, NULL);
860 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
862 GOTO(out, rc = -ENOMEM);
864 /* Disconnect import first. NULL is passed for the '@env', since
865 * it will not be used. */
866 rc = lwp->obd_lu_dev->ld_ops->ldo_process_config(NULL, lwp->obd_lu_dev,
870 lustre_cfg_free(lcfg);
876 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
878 OBD_FREE(logname, MTI_NAME_MAXLEN);
883 * Stop the lwp for an OST/MDT target.
885 static int lustre_stop_lwp(struct super_block *sb)
887 struct lustre_sb_info *lsi = s2lsi(sb);
888 struct obd_device *lwp = NULL;
889 char *lwpname = NULL;
893 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
895 CDEBUG(PTR_ERR(lwp) == -ENOENT ? D_CONFIG : D_ERROR,
896 "%s: lwp wasn't started.\n", lsi->lsi_svname);
901 rc = class_manual_cleanup(lwp);
905 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
910 * Start the lwp(fsname-MDT0000-lwp-OSTxxxx) for an OST or MDT target,
911 * which would be used to establish connection from OST to MDT0.
913 static int lustre_start_lwp(struct super_block *sb)
915 struct lustre_sb_info *lsi = s2lsi(sb);
916 struct config_llog_instance *cfg = NULL;
917 struct obd_device *lwp;
918 char *lwpname = NULL;
919 char *logname = NULL;
923 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
925 /* the lwp device already stared */
926 if (lwp && !IS_ERR(lwp))
929 if (PTR_ERR(lwp) != -ENOENT)
930 GOTO(out, rc = PTR_ERR(lwp));
932 LASSERT(lwpname != NULL);
933 LASSERT(logname != NULL);
937 GOTO(out, rc = -ENOMEM);
939 cfg->cfg_callback = client_lwp_config_process;
940 cfg->cfg_instance = sb;
942 rc = lustre_process_log(sb, logname, cfg);
945 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
947 OBD_FREE(logname, MTI_NAME_MAXLEN);
953 DEFINE_MUTEX(server_start_lock);
955 /* Stop MDS/OSS if nobody is using them */
956 static int server_stop_servers(int lsiflags)
958 struct obd_device *obd = NULL;
959 struct obd_type *type = NULL;
963 mutex_lock(&server_start_lock);
965 /* Either an MDT or an OST or neither */
966 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
967 if (lsiflags & LDD_F_SV_TYPE_MDT) {
968 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
970 type = class_search_type(LUSTRE_MDT_NAME);
973 /* if this was an OST, and there are no more OST's, clean up the OSS */
974 if (lsiflags & LDD_F_SV_TYPE_OST) {
975 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
977 type = class_search_type(LUSTRE_OST_NAME);
980 if (obd != NULL && (type == NULL || type->typ_refcnt == 0)) {
982 /* obd_fail doesn't mean much on a server obd */
983 rc = class_manual_cleanup(obd);
986 mutex_unlock(&server_start_lock);
991 int server_mti_print(const char *title, struct mgs_target_info *mti)
993 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
994 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
995 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
996 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
997 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
998 mti->mti_config_ver, mti->mti_flags);
1001 EXPORT_SYMBOL(server_mti_print);
1003 /* Generate data for registration */
1004 static int server_lsi2mti(struct lustre_sb_info *lsi,
1005 struct mgs_target_info *mti)
1007 lnet_process_id_t id;
1012 if (!IS_SERVER(lsi))
1015 if (strlcpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname))
1016 >= sizeof(mti->mti_svname))
1019 mti->mti_nid_count = 0;
1020 while (LNetGetId(i++, &id) != -ENOENT) {
1021 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1024 /* server use --servicenode param, only allow specified
1025 * nids be registered */
1026 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
1027 class_match_nid(lsi->lsi_lmd->lmd_params,
1028 PARAM_FAILNODE, id.nid) < 1)
1031 /* match specified network */
1032 if (!class_match_net(lsi->lsi_lmd->lmd_params,
1033 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1036 mti->mti_nids[mti->mti_nid_count] = id.nid;
1037 mti->mti_nid_count++;
1038 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1039 CWARN("Only using first %d nids for %s\n",
1040 mti->mti_nid_count, mti->mti_svname);
1045 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1046 mti->mti_config_ver = 0;
1048 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1052 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1055 /* Orion requires index to be set */
1056 LASSERT(!(rc & LDD_F_NEED_INDEX));
1057 /* keep only LDD flags */
1058 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1059 if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN))
1060 mti->mti_flags |= LDD_F_UPDATE;
1061 cplen = strlcpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1062 sizeof(mti->mti_params));
1063 if (cplen >= sizeof(mti->mti_params))
1068 /* Register an old or new target with the MGS. If needed MGS will construct
1069 startup logs and assign index */
1070 static int server_register_target(struct lustre_sb_info *lsi)
1072 struct obd_device *mgc = lsi->lsi_mgc;
1073 struct mgs_target_info *mti = NULL;
1080 if (!IS_SERVER(lsi))
1087 rc = server_lsi2mti(lsi, mti);
1091 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1092 mti->mti_svname, mti->mti_fsname,
1093 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1096 /* if write_conf is true, the registration must succeed */
1097 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1098 mti->mti_flags |= LDD_F_OPC_REG;
1100 /* Register the target */
1101 /* FIXME use mgc_process_config instead */
1102 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1103 sizeof(KEY_REGISTER_TARGET),
1104 KEY_REGISTER_TARGET,
1105 sizeof(*mti), mti, NULL);
1107 if (mti->mti_flags & LDD_F_ERROR) {
1108 LCONSOLE_ERROR_MSG(0x160,
1109 "%s: the MGS refuses to allow this server "
1110 "to start: rc = %d. Please see messages on "
1111 "the MGS.\n", lsi->lsi_svname, rc);
1112 } else if (writeconf) {
1113 LCONSOLE_ERROR_MSG(0x15f,
1114 "%s: cannot register this server with the MGS: "
1115 "rc = %d. Is the MGS running?\n",
1116 lsi->lsi_svname, rc);
1118 CERROR("%s: error registering with the MGS: rc = %d "
1119 "(not fatal)\n", lsi->lsi_svname, rc);
1120 /* reset the error code for non-fatal error. */
1133 * Notify the MGS that this target is ready.
1134 * Used by IR - if the MGS receives this message, it will notify clients.
1136 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1138 struct lustre_sb_info *lsi = s2lsi(sb);
1139 struct obd_device *mgc = lsi->lsi_mgc;
1140 struct mgs_target_info *mti = NULL;
1146 if (!(IS_SERVER(lsi)))
1152 rc = server_lsi2mti(lsi, mti);
1156 mti->mti_instance = obd->u.obt.obt_instance;
1157 mti->mti_flags |= LDD_F_OPC_READY;
1159 /* FIXME use mgc_process_config instead */
1160 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1161 sizeof(KEY_REGISTER_TARGET),
1162 KEY_REGISTER_TARGET,
1163 sizeof(*mti), mti, NULL);
1165 /* Imperative recovery: if the mgs informs us to use IR? */
1166 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1167 (mti->mti_flags & LDD_F_IR_CAPABLE))
1168 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1177 /** Start server targets: MDTs and OSTs
1179 static int server_start_targets(struct super_block *sb)
1181 struct obd_device *obd;
1182 struct lustre_sb_info *lsi = s2lsi(sb);
1183 struct config_llog_instance cfg;
1185 struct lu_device *dev;
1189 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1192 /* make sure the MDS is started */
1193 mutex_lock(&server_start_lock);
1194 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1196 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1198 LUSTRE_MDS_OBDNAME"_uuid",
1201 mutex_unlock(&server_start_lock);
1202 CERROR("failed to start MDS: %d\n", rc);
1206 mutex_unlock(&server_start_lock);
1209 /* If we're an OST, make sure the global OSS is running */
1211 /* make sure OSS is started */
1212 mutex_lock(&server_start_lock);
1213 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1215 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1217 LUSTRE_OSS_OBDNAME"_uuid",
1220 mutex_unlock(&server_start_lock);
1221 CERROR("failed to start OSS: %d\n", rc);
1225 mutex_unlock(&server_start_lock);
1228 /* Set the mgc fs to our server disk. This allows the MGC to
1229 * read and write configs locally, in case it can't talk to the MGS. */
1230 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1232 GOTO(out_stop_service, rc);
1234 /* Register with MGS */
1235 rc = server_register_target(lsi);
1239 /* Let the target look up the mount using the target's name
1240 (we can't pass the sb or mnt through class_process_config.) */
1241 rc = server_register_mount(lsi->lsi_svname, sb);
1245 /* Start targets using the llog named for the target */
1246 memset(&cfg, 0, sizeof(cfg));
1247 cfg.cfg_callback = class_config_llog_handler;
1248 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1250 CERROR("failed to start server %s: %d\n",
1251 lsi->lsi_svname, rc);
1252 /* Do NOT call server_deregister_mount() here. This makes it
1253 * impossible to find mount later in cleanup time and leaves
1254 * @lsi and othder stuff leaked. -umka */
1258 obd = class_name2obd(lsi->lsi_svname);
1260 CERROR("no server named %s was started\n", lsi->lsi_svname);
1261 GOTO(out_mgc, rc = -ENXIO);
1264 if (IS_OST(lsi) || IS_MDT(lsi)) {
1265 rc = lustre_start_lwp(sb);
1267 CERROR("%s: failed to start LWP: %d\n",
1268 lsi->lsi_svname, rc);
1273 server_notify_target(sb, obd);
1275 /* calculate recovery timeout, do it after lustre_process_log */
1276 server_calc_timeout(lsi, obd);
1278 /* log has been fully processed */
1279 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1281 /* log has been fully processed, let clients connect */
1282 dev = obd->obd_lu_dev;
1283 if (dev && dev->ld_ops->ldo_prepare) {
1284 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1286 struct lu_context session_ctx;
1288 lu_context_init(&session_ctx, LCT_SERVER_SESSION);
1289 session_ctx.lc_thread = NULL;
1290 lu_context_enter(&session_ctx);
1291 env.le_ses = &session_ctx;
1293 rc = dev->ld_ops->ldo_prepare(&env, NULL, dev);
1296 lu_context_exit(&session_ctx);
1297 lu_context_fini(&session_ctx);
1301 /* abort recovery only on the complete stack:
1302 * many devices can be involved */
1303 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1304 (OBP(obd, iocontrol))) {
1305 obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
1310 /* Release the mgc fs for others to use */
1311 server_mgc_clear_fs(lsi->lsi_mgc);
1315 server_stop_servers(lsi->lsi_flags);
1320 static int lsi_prepare(struct lustre_sb_info *lsi)
1327 LASSERT(lsi->lsi_lmd);
1329 /* The server name is given as a mount line option */
1330 if (lsi->lsi_lmd->lmd_profile == NULL) {
1331 LCONSOLE_ERROR("Can't determine server name\n");
1335 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
1336 RETURN(-ENAMETOOLONG);
1338 strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
1340 /* Determine osd type */
1341 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
1342 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
1343 sizeof(lsi->lsi_osd_type))
1344 RETURN(-ENAMETOOLONG);
1346 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
1348 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
1351 /* XXX: a temp. solution for components using ldiskfs
1352 * to be removed in one of the subsequent patches */
1353 if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs"))
1354 strcpy(lsi->lsi_fstype, "ldiskfs");
1356 strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
1358 /* Determine server type */
1359 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1361 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
1362 /* Assume we're a bare MGS */
1364 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
1366 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
1371 lsi->lsi_flags |= rc;
1373 /* Add mount line flags that used to be in ldd:
1374 * writeconf, mgs, anything else?
1376 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
1377 LDD_F_WRITECONF : 0;
1378 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
1380 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_UPDATE) ?
1382 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
1383 LDD_F_SV_TYPE_MGS : 0;
1384 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
1385 LDD_F_NO_PRIMNODE : 0;
1390 /*************** server mount ******************/
1392 /** Start the shutdown of servers at umount.
1394 static void server_put_super(struct super_block *sb)
1396 struct lustre_sb_info *lsi = s2lsi(sb);
1397 struct obd_device *obd;
1398 char *tmpname, *extraname = NULL;
1400 int lsiflags = lsi->lsi_flags;
1403 LASSERT(IS_SERVER(lsi));
1405 tmpname_sz = strlen(lsi->lsi_svname) + 1;
1406 OBD_ALLOC(tmpname, tmpname_sz);
1407 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
1408 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1409 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1410 snprintf(tmpname, tmpname_sz, "MGS");
1412 /* disconnect the lwp first to drain off the inflight request */
1413 if (IS_OST(lsi) || IS_MDT(lsi)) {
1416 rc = lustre_disconnect_lwp(sb);
1417 if (rc && rc != ETIMEDOUT)
1418 CERROR("%s: failed to disconnect lwp. (rc=%d)\n",
1422 /* Stop the target */
1423 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1424 (IS_MDT(lsi) || IS_OST(lsi))) {
1425 struct lustre_profile *lprof = NULL;
1427 /* tell the mgc to drop the config log */
1428 lustre_end_log(sb, lsi->lsi_svname, NULL);
1430 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1431 If there are any setup/cleanup errors, save the lov
1432 name for safety cleanup later. */
1433 lprof = class_get_profile(lsi->lsi_svname);
1434 if (lprof && lprof->lp_dt) {
1435 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1436 strcpy(extraname, lprof->lp_dt);
1439 obd = class_name2obd(lsi->lsi_svname);
1441 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1442 if (lsiflags & LSI_UMOUNT_FAILOVER)
1444 /* We can't seem to give an error return code
1445 * to .put_super, so we better make sure we clean up! */
1447 class_manual_cleanup(obd);
1449 CERROR("no obd %s\n", lsi->lsi_svname);
1450 server_deregister_mount(lsi->lsi_svname);
1454 /* If they wanted the mgs to stop separately from the mdt, they
1455 should have put it on a different device. */
1457 /* if MDS start with --nomgs, don't stop MGS then */
1458 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1459 server_stop_mgs(sb);
1462 if (IS_OST(lsi) || IS_MDT(lsi)) {
1463 if (lustre_stop_lwp(sb) < 0)
1464 CERROR("%s: failed to stop lwp!\n", tmpname);
1467 /* Clean the mgc and sb */
1468 lustre_common_put_super(sb);
1470 /* wait till all in-progress cleanups are done
1471 * specifically we're interested in ofd cleanup
1473 obd_zombie_barrier();
1475 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1476 until the target is really gone so that our type refcount check
1478 server_stop_servers(lsiflags);
1480 /* In case of startup or cleanup err, stop related obds */
1482 obd = class_name2obd(extraname);
1484 CWARN("Cleaning orphaned obd %s\n", extraname);
1486 class_manual_cleanup(obd);
1488 OBD_FREE(extraname, strlen(extraname) + 1);
1491 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1492 OBD_FREE(tmpname, tmpname_sz);
1496 /** Called only for 'umount -f'
1498 static void server_umount_begin(struct super_block *sb)
1500 struct lustre_sb_info *lsi = s2lsi(sb);
1503 CDEBUG(D_MOUNT, "umount -f\n");
1504 /* umount = failover
1506 no third way to do non-force, non-failover */
1507 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1511 static int server_statfs(struct dentry *dentry, struct kstatfs *buf)
1513 struct super_block *sb = dentry->d_sb;
1514 struct lustre_sb_info *lsi = s2lsi(sb);
1515 struct obd_statfs statfs;
1519 if (lsi->lsi_dt_dev) {
1520 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
1522 statfs_unpack(buf, &statfs);
1523 buf->f_type = sb->s_magic;
1529 buf->f_type = sb->s_magic;
1530 buf->f_bsize = sb->s_blocksize;
1536 buf->f_namelen = NAME_MAX;
1540 /** The operations we support directly on the superblock:
1541 * mount, umount, and df.
1543 static struct super_operations server_ops = {
1544 .put_super = server_put_super,
1545 .umount_begin = server_umount_begin, /* umount -f */
1546 .statfs = server_statfs,
1550 * Xattr support for Lustre servers
1552 static ssize_t lustre_getxattr(struct dentry *dentry, const char *name,
1553 void *buffer, size_t size)
1555 if (!selinux_is_enabled())
1560 static int lustre_setxattr(struct dentry *dentry, const char *name,
1561 const void *value, size_t size, int flags)
1566 static ssize_t lustre_listxattr(struct dentry *d_entry, char *name,
1572 const struct inode_operations server_inode_operations = {
1573 .setxattr = lustre_setxattr,
1574 .getxattr = lustre_getxattr,
1575 .listxattr = lustre_listxattr,
1578 #define log2(n) ffz(~(n))
1579 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1581 static int server_fill_super_common(struct super_block *sb)
1583 struct inode *root = 0;
1586 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1588 sb->s_blocksize = 4096;
1589 sb->s_blocksize_bits = log2(sb->s_blocksize);
1590 sb->s_magic = LUSTRE_SUPER_MAGIC;
1591 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1592 sb->s_flags |= MS_RDONLY;
1593 sb->s_op = &server_ops;
1595 root = new_inode(sb);
1597 CERROR("Can't make root inode\n");
1601 /* returns -EIO for every operation */
1602 /* make_bad_inode(root); -- badness - can't umount */
1603 /* apparently we need to be a directory for the mount to finish */
1604 root->i_mode = S_IFDIR;
1605 root->i_op = &server_inode_operations;
1606 sb->s_root = d_make_root(root);
1608 CERROR("%s: can't make root dentry\n", sb->s_id);
1615 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
1617 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1618 struct obd_device *obd;
1619 struct dt_device_param p;
1625 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
1626 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
1628 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
1629 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
1630 strcat(lsi->lsi_osd_uuid, "_UUID");
1631 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
1633 obd = class_name2obd(lsi->lsi_osd_obdname);
1635 rc = lustre_start_simple(lsi->lsi_osd_obdname,
1637 lsi->lsi_osd_uuid, lmd->lmd_dev,
1638 flagstr, lsi->lsi_lmd->lmd_opts,
1642 obd = class_name2obd(lsi->lsi_osd_obdname);
1646 rc = obd_connect(NULL, &lsi->lsi_osd_exp,
1647 obd, &obd->obd_uuid, NULL, NULL);
1650 class_manual_cleanup(obd);
1651 lsi->lsi_dt_dev = NULL;
1655 LASSERT(obd->obd_lu_dev);
1656 lu_device_get(obd->obd_lu_dev);
1657 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
1658 LASSERT(lsi->lsi_dt_dev);
1660 /* set disk context for llog usage */
1661 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1662 obd->obd_lvfs_ctxt.dt = lsi->lsi_dt_dev;
1664 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
1669 /** Fill in the superblock info for a Lustre server.
1670 * Mount the device with the correct options.
1671 * Read the on-disk config file.
1672 * Start the services.
1674 int server_fill_super(struct super_block *sb)
1676 struct lustre_sb_info *lsi = s2lsi(sb);
1680 rc = lsi_prepare(lsi);
1684 /* Start low level OSD */
1685 rc = osd_start(lsi, sb->s_flags);
1687 CERROR("Unable to start osd on %s: %d\n",
1688 lsi->lsi_lmd->lmd_dev, rc);
1693 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
1694 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
1696 if (class_name2obd(lsi->lsi_svname)) {
1697 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1698 "running. Double-mount may have compromised"
1699 " the disk journal.\n",
1705 /* Start MGS before MGC */
1706 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1707 rc = server_start_mgs(sb);
1712 /* Start MGC before servers */
1713 rc = lustre_start_mgc(sb);
1717 /* Set up all obd devices for service */
1718 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1719 (IS_OST(lsi) || IS_MDT(lsi))) {
1720 rc = server_start_targets(sb);
1722 CERROR("Unable to start targets: %d\n", rc);
1725 /* FIXME overmount client here, or can we just start a
1726 * client log and client_fill_super on this sb? We
1727 * need to make sure server_put_super gets called too
1728 * - ll_put_super calls lustre_common_put_super; check
1729 * there for LSI_SERVER flag, call s_p_s if so.
1731 * Probably should start client from new thread so we
1732 * can return. Client will not finish until all
1733 * servers are connected. Note - MGS-only server does
1734 * NOT get a client, since there is no lustre fs
1735 * associated - the MGS is for all lustre fs's */
1738 rc = server_fill_super_common(sb);
1744 /* We jump here in case of failure while starting targets or MGS.
1745 * In this case we can't just put @mnt and have to do real cleanup
1746 * with stoping targets, etc. */
1747 server_put_super(sb);
1752 * Calculate timeout value for a target.
1754 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1756 struct lustre_mount_data *lmd;
1760 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
1761 int min = OBD_RECOVERY_TIME_MIN;
1763 LASSERT(IS_SERVER(lsi));
1767 soft = lmd->lmd_recovery_time_soft;
1768 hard = lmd->lmd_recovery_time_hard;
1769 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1770 obd->obd_no_ir = !has_ir;
1774 soft = OBD_RECOVERY_TIME_SOFT;
1776 hard = OBD_RECOVERY_TIME_HARD;
1778 /* target may have ir_factor configured. */
1779 factor = OBD_IR_FACTOR_DEFAULT;
1780 if (obd->obd_recovery_ir_factor)
1781 factor = obd->obd_recovery_ir_factor;
1784 int new_soft = soft;
1785 int new_hard = hard;
1787 /* adjust timeout value by imperative recovery */
1789 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1790 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1792 /* make sure the timeout is not too short */
1793 new_soft = max(min, new_soft);
1794 new_hard = max(new_soft, new_hard);
1796 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1797 "window shrunk from %d-%d down to %d-%d\n",
1798 obd->obd_name, soft, hard, new_soft, new_hard);
1805 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1806 obd->obd_recovery_time_hard = hard;
1807 obd->obd_recovery_ir_factor = factor;
1809 EXPORT_SYMBOL(server_calc_timeout);