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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2013, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
31 * lustre/obdclass/obd_mount_server.c
33 * Server mount routines
35 * Author: Nathan Rutman <nathan@clusterfs.com>
37 #define DEBUG_SUBSYSTEM S_CLASS
38 #define D_MOUNT (D_SUPER | D_CONFIG /* | D_WARNING */)
40 #include <linux/types.h>
41 #include <linux/generic-radix-tree.h>
42 #ifdef HAVE_LINUX_SELINUX_IS_ENABLED
43 #include <linux/selinux.h>
45 #include <linux/statfs.h>
46 #include <linux/version.h>
47 #include <linux/delay.h>
48 #include <linux/file.h>
50 #include <linux/fsmap.h>
53 #include <llog_swab.h>
54 #include <lustre_disk.h>
55 #include <uapi/linux/lustre/lustre_ioctl.h>
56 #include <lustre_log.h>
57 #include <uapi/linux/lustre/lustre_param.h>
59 #include <obd_class.h>
61 #include "tgt_internal.h"
63 /*********** mount lookup *********/
65 static DEFINE_MUTEX(lustre_mount_info_lock);
66 static LIST_HEAD(server_mount_info_list);
68 static struct lustre_mount_info *server_find_mount(const char *name)
70 struct list_head *tmp;
71 struct lustre_mount_info *lmi;
74 list_for_each(tmp, &server_mount_info_list) {
75 lmi = 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.
87 static int server_register_mount(const char *name, struct super_block *sb)
89 struct lustre_mount_info *lmi;
95 OBD_ALLOC(lmi, sizeof(*lmi));
98 OBD_ALLOC(name_cp, strlen(name) + 1);
100 OBD_FREE(lmi, sizeof(*lmi));
103 strcpy(name_cp, name);
105 mutex_lock(&lustre_mount_info_lock);
107 if (server_find_mount(name)) {
108 mutex_unlock(&lustre_mount_info_lock);
109 OBD_FREE(lmi, sizeof(*lmi));
110 OBD_FREE(name_cp, strlen(name) + 1);
111 CERROR("Already registered %s\n", name);
114 lmi->lmi_name = name_cp;
116 list_add(&lmi->lmi_list_chain, &server_mount_info_list);
118 mutex_unlock(&lustre_mount_info_lock);
120 CDEBUG(D_MOUNT, "register mount %p from %s\n", sb, name);
125 /* when an obd no longer needs a mount */
126 static int server_deregister_mount(const char *name)
128 struct lustre_mount_info *lmi;
131 mutex_lock(&lustre_mount_info_lock);
132 lmi = server_find_mount(name);
134 mutex_unlock(&lustre_mount_info_lock);
135 CERROR("%s not registered\n", name);
139 CDEBUG(D_MOUNT, "deregister mount %p from %s\n", lmi->lmi_sb, name);
141 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
142 list_del(&lmi->lmi_list_chain);
143 OBD_FREE(lmi, sizeof(*lmi));
144 mutex_unlock(&lustre_mount_info_lock);
146 CFS_RACE(OBD_FAIL_MDS_LLOG_UMOUNT_RACE);
150 /* obd's look up a registered mount using their obdname. This is just
151 * for initial obd setup to find the mount struct. It should not be
152 * called every time you want to mntget.
154 struct lustre_mount_info *server_get_mount(const char *name)
156 struct lustre_mount_info *lmi;
157 struct lustre_sb_info *lsi;
160 mutex_lock(&lustre_mount_info_lock);
161 lmi = server_find_mount(name);
162 mutex_unlock(&lustre_mount_info_lock);
164 CERROR("Can't find mount for %s\n", name);
167 lsi = s2lsi(lmi->lmi_sb);
169 kref_get(&lsi->lsi_mounts);
171 CDEBUG(D_MOUNT, "get mount %p from %s, refs=%d\n", lmi->lmi_sb,
172 name, kref_read(&lsi->lsi_mounts));
176 EXPORT_SYMBOL(server_get_mount);
179 * server_put_mount: to be called from obd_cleanup methods
181 * @dereg_mnt: 0 or 1 depending on whether the mount is to be deregistered or
184 * The caller decides whether server_deregister_mount() needs to be called or
185 * not. Calling of server_deregister_mount() does not depend on refcounting on
186 * lsi because we could have say the mgs and mds on the same node and we
187 * unmount the mds, then the ref on the lsi would still be non-zero but we
188 * would still want to deregister the mds mount.
190 int server_put_mount(const char *name, bool dereg_mnt)
192 struct lustre_mount_info *lmi;
193 struct lustre_sb_info *lsi;
196 mutex_lock(&lustre_mount_info_lock);
197 lmi = server_find_mount(name);
198 mutex_unlock(&lustre_mount_info_lock);
200 CERROR("Can't find mount for %s\n", name);
203 lsi = s2lsi(lmi->lmi_sb);
205 CDEBUG(D_MOUNT, "put mount %p from %s, refs=%d\n",
206 lmi->lmi_sb, name, kref_read(&lsi->lsi_mounts));
208 if (lustre_put_lsi(lmi->lmi_sb))
209 CDEBUG(D_MOUNT, "Last put of mount %p from %s\n",
213 /* this obd should never need the mount again */
214 server_deregister_mount(name);
218 EXPORT_SYMBOL(server_put_mount);
220 /* Set up a MGS to serve startup logs */
221 static int server_start_mgs(struct super_block *sb)
223 struct lustre_sb_info *lsi = s2lsi(sb);
224 struct lustre_mount_info *lmi;
228 /* It is impossible to have more than 1 MGS per node, since
229 * MGC wouldn't know which to connect to
231 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
233 lsi = s2lsi(lmi->lmi_sb);
234 LCONSOLE_ERROR_MSG(0x15d,
235 "The MGS service was already started from server\n");
239 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
241 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb);
243 GOTO(report_err, rc);
245 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
246 LUSTRE_MGS_OBDNAME, NULL, NULL,
247 lsi->lsi_osd_obdname, NULL);
248 /* server_deregister_mount() is not called previously, for lsi
249 * and other stuff can't be freed cleanly when mgs calls
250 * server_put_mount() in error handling case (see b=17758),
251 * this problem is caused by a bug in mgs_init0, which forgot
252 * calling server_put_mount in error case.
255 server_deregister_mount(LUSTRE_MGS_OBDNAME);
257 LCONSOLE_ERROR_MSG(0x15e,
258 "Failed to start MGS '%s' (%d). Is the 'mgs' module loaded?\n",
259 LUSTRE_MGS_OBDNAME, rc);
264 static int server_stop_mgs(struct super_block *sb)
266 struct obd_device *obd;
268 struct lustre_mount_info *lmi;
271 /* Do not stop MGS if this device is not the running MGT */
272 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
273 if (lmi && lmi->lmi_sb != sb)
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.
294 static int server_mgc_set_fs(const struct lu_env *env,
295 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(env, 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(const struct lu_env *env,
314 struct obd_device *mgc)
319 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
320 rc = obd_set_info_async(env, mgc->obd_self_export,
321 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
326 static inline bool is_mdc_device(const char *devname)
330 ptr = strrchr(devname, '-');
331 return ptr && strcmp(ptr, "-mdc") == 0;
334 static inline bool tgt_is_mdt(const char *tgtname, u32 *idx)
338 type = server_name2index(tgtname, idx, NULL);
340 return type == LDD_F_SV_TYPE_MDT;
344 * Convert OST/MDT name(fsname-{MDT,OST}xxxx) to a lwp name with the @idx:yyyy
345 * (fsname-MDTyyyy-lwp-{MDT,OST}xxxx)
347 int tgt_name2lwp_name(const char *tgt_name, char *lwp_name, int len, u32 idx)
354 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
358 rc = server_name2fsname(tgt_name, fsname, &tgt);
360 CERROR("%s: failed to get fsname from tgt_name: rc = %d\n",
365 if (*tgt != '-' && *tgt != ':') {
366 CERROR("%s: invalid tgt_name name!\n", tgt_name);
367 GOTO(cleanup, rc = -EINVAL);
371 if (strncmp(tgt, "OST", 3) != 0 && strncmp(tgt, "MDT", 3) != 0) {
372 CERROR("%s is not an OST or MDT target!\n", tgt_name);
373 GOTO(cleanup, rc = -EINVAL);
375 snprintf(lwp_name, len, "%s-MDT%04x-%s-%s",
376 fsname, idx, LUSTRE_LWP_NAME, tgt);
378 GOTO(cleanup, rc = 0);
382 OBD_FREE(fsname, MTI_NAME_MAXLEN);
386 EXPORT_SYMBOL(tgt_name2lwp_name);
388 static LIST_HEAD(lwp_register_list);
389 static DEFINE_SPINLOCK(lwp_register_list_lock);
391 static void lustre_put_lwp_item(struct lwp_register_item *lri)
393 if (atomic_dec_and_test(&lri->lri_ref)) {
394 LASSERT(list_empty(&lri->lri_list));
397 class_export_put(*lri->lri_exp);
402 int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp,
403 register_lwp_cb cb_func, void *cb_data)
405 struct obd_device *lwp;
406 struct lwp_register_item *lri;
410 LASSERTF(strlen(lwpname) < MTI_NAME_MAXLEN, "lwpname is too long %s\n",
412 LASSERT(exp && !*exp);
418 lwp = class_name2obd(lwpname);
419 if (lwp && lwp->obd_set_up == 1) {
420 struct obd_uuid *uuid;
427 memcpy(uuid->uuid, lwpname, strlen(lwpname));
428 *exp = obd_uuid_lookup(lwp, uuid);
432 memcpy(lri->lri_name, lwpname, strlen(lwpname));
434 lri->lri_cb_func = cb_func;
435 lri->lri_cb_data = cb_data;
436 INIT_LIST_HEAD(&lri->lri_list);
438 * Initialize the lri_ref at 2, one will be released before
439 * current function returned via lustre_put_lwp_item(), the
440 * other will be released in lustre_deregister_lwp_item().
442 atomic_set(&lri->lri_ref, 2);
444 spin_lock(&lwp_register_list_lock);
445 list_add(&lri->lri_list, &lwp_register_list);
448 spin_unlock(&lwp_register_list_lock);
452 lustre_put_lwp_item(lri);
456 EXPORT_SYMBOL(lustre_register_lwp_item);
458 void lustre_deregister_lwp_item(struct obd_export **exp)
460 struct lwp_register_item *lri;
461 bool removed = false;
464 spin_lock(&lwp_register_list_lock);
465 list_for_each_entry(lri, &lwp_register_list, lri_list) {
466 if (exp == lri->lri_exp) {
467 list_del_init(&lri->lri_list);
472 spin_unlock(&lwp_register_list_lock);
477 /* See lustre_notify_lwp_list(), in some extreme race conditions,
478 * the notify callback could be still on the fly, we need to wait
479 * for the callback done before moving on to free the data used
482 while (atomic_read(&lri->lri_ref) > 1) {
483 CDEBUG(D_MOUNT, "lri reference count %u, repeat: %d\n",
484 atomic_read(&lri->lri_ref), repeat);
486 schedule_timeout_interruptible(cfs_time_seconds(1));
488 lustre_put_lwp_item(lri);
490 EXPORT_SYMBOL(lustre_deregister_lwp_item);
492 struct obd_export *lustre_find_lwp_by_index(const char *dev, u32 idx)
494 struct lustre_mount_info *lmi;
495 struct lustre_sb_info *lsi;
496 struct obd_device *lwp;
497 struct obd_export *exp = NULL;
502 lmi = server_get_mount(dev);
506 lsi = s2lsi(lmi->lmi_sb);
507 rc = server_name2fsname(lsi->lsi_svname, fsname, NULL);
509 CERROR("%s: failed to get fsname: rc = %d\n",
510 lsi->lsi_svname, rc);
514 snprintf(lwp_name, sizeof(lwp_name), "%s-MDT%04x", fsname, idx);
515 mutex_lock(&lsi->lsi_lwp_mutex);
516 list_for_each_entry(lwp, &lsi->lsi_lwp_list, obd_lwp_list) {
517 char *ptr = strstr(lwp->obd_name, lwp_name);
519 if (ptr && lwp->obd_lwp_export) {
520 exp = class_export_get(lwp->obd_lwp_export);
524 mutex_unlock(&lsi->lsi_lwp_mutex);
527 server_put_mount(dev, false);
531 EXPORT_SYMBOL(lustre_find_lwp_by_index);
533 void lustre_notify_lwp_list(struct obd_export *exp)
535 struct lwp_register_item *lri;
539 spin_lock(&lwp_register_list_lock);
540 list_for_each_entry(lri, &lwp_register_list, lri_list) {
541 if (strcmp(exp->exp_obd->obd_name, lri->lri_name))
545 *lri->lri_exp = class_export_get(exp);
546 if (!lri->lri_cb_func)
548 atomic_inc(&lri->lri_ref);
549 spin_unlock(&lwp_register_list_lock);
551 lri->lri_cb_func(lri->lri_cb_data);
552 lustre_put_lwp_item(lri);
554 /* Others may have changed the list after we unlock, we have
555 * to rescan the list from the beginning. Usually, the list
556 * 'lwp_register_list' is very short, and there is 'guard'
557 * lri::lri_exp that will prevent the callback to be done
558 * repeatedly. So rescanning the list has no problem.
562 spin_unlock(&lwp_register_list_lock);
564 EXPORT_SYMBOL(lustre_notify_lwp_list);
566 static int lustre_lwp_connect(struct obd_device *lwp, bool is_mdt)
569 struct lu_context session_ctx;
570 struct obd_export *exp;
571 struct obd_uuid *uuid = NULL;
572 struct obd_connect_data *data = NULL;
576 /* log has been fully processed, let clients connect */
577 rc = lu_env_init(&env, lwp->obd_lu_dev->ld_type->ldt_ctx_tags);
581 lu_context_init(&session_ctx, LCT_SERVER_SESSION);
582 session_ctx.lc_thread = NULL;
583 lu_context_enter(&session_ctx);
584 env.le_ses = &session_ctx;
588 GOTO(out, rc = -ENOMEM);
590 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
591 data->ocd_version = LUSTRE_VERSION_CODE;
592 data->ocd_connect_flags |= OBD_CONNECT_FID | OBD_CONNECT_AT |
593 OBD_CONNECT_LRU_RESIZE | OBD_CONNECT_FULL20 |
594 OBD_CONNECT_LVB_TYPE | OBD_CONNECT_LIGHTWEIGHT |
595 OBD_CONNECT_LFSCK | OBD_CONNECT_BULK_MBITS;
598 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS;
602 GOTO(out, rc = -ENOMEM);
604 if (strlen(lwp->obd_name) > sizeof(uuid->uuid)) {
605 CERROR("%s: Too long lwp name %s, max_size is %d\n",
606 lwp->obd_name, lwp->obd_name, (int)sizeof(uuid->uuid));
607 GOTO(out, rc = -EINVAL);
610 /* Use lwp name as the uuid, so we find the export by lwp name later */
611 memcpy(uuid->uuid, lwp->obd_name, strlen(lwp->obd_name));
612 rc = obd_connect(&env, &exp, lwp, uuid, data, NULL);
614 CERROR("%s: connect failed: rc = %d\n", lwp->obd_name, rc);
616 if (unlikely(lwp->obd_lwp_export))
617 class_export_put(lwp->obd_lwp_export);
618 lwp->obd_lwp_export = class_export_get(exp);
630 lu_context_exit(&session_ctx);
631 lu_context_fini(&session_ctx);
637 * lwp is used by slaves (Non-MDT0 targets) to manage the connection to MDT0,
638 * or from the OSTx to MDTy.
640 static int lustre_lwp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi,
643 struct obd_device *obd;
644 char *lwpname = NULL;
645 char *lwpuuid = NULL;
651 lnet_nid4_to_nid(lcfg->lcfg_nid, &nid);
653 rc = libcfs_strnid(&nid, lustre_cfg_string(lcfg, 2));
657 rc = class_add_uuid(lustre_cfg_string(lcfg, 1), &nid);
659 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
663 OBD_ALLOC(lwpname, MTI_NAME_MAXLEN);
665 GOTO(out, rc = -ENOMEM);
667 rc = tgt_name2lwp_name(lsi->lsi_svname, lwpname, MTI_NAME_MAXLEN, idx);
669 CERROR("%s: failed to generate lwp name: rc = %d\n",
670 lsi->lsi_svname, rc);
674 OBD_ALLOC(lwpuuid, MTI_NAME_MAXLEN);
676 GOTO(out, rc = -ENOMEM);
678 sprintf(lwpuuid, "%s_UUID", lwpname);
679 rc = lustre_start_simple(lwpname, LUSTRE_LWP_NAME,
680 lwpuuid, lustre_cfg_string(lcfg, 1),
683 CERROR("%s: setup up failed: rc %d\n", lwpname, rc);
687 obd = class_name2obd(lwpname);
690 rc = lustre_lwp_connect(obd, strstr(lsi->lsi_svname, "-MDT") != NULL);
692 CERROR("%s: connect failed: rc = %d\n", lwpname, rc);
696 obd->u.cli.cl_max_mds_easize = MAX_MD_SIZE;
697 mutex_lock(&lsi->lsi_lwp_mutex);
698 list_add_tail(&obd->obd_lwp_list, &lsi->lsi_lwp_list);
699 mutex_unlock(&lsi->lsi_lwp_mutex);
702 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
704 OBD_FREE(lwpuuid, MTI_NAME_MAXLEN);
709 /* the caller is responsible for memory free */
710 static struct obd_device *lustre_find_lwp(struct lustre_sb_info *lsi,
711 char **lwpname, u32 idx)
713 struct obd_device *lwp;
718 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
720 OBD_ALLOC(*lwpname, MTI_NAME_MAXLEN);
721 if (*lwpname == NULL)
722 RETURN(ERR_PTR(-ENOMEM));
724 rc = tgt_name2lwp_name(lsi->lsi_svname, *lwpname, MTI_NAME_MAXLEN, idx);
726 CERROR("%s: failed to generate lwp name: rc = %d\n",
727 lsi->lsi_svname, rc);
728 GOTO(out, rc = -EINVAL);
731 lwp = class_name2obd(*lwpname);
735 if (*lwpname != NULL) {
736 OBD_FREE(*lwpname, MTI_NAME_MAXLEN);
742 RETURN(lwp ? lwp : ERR_PTR(-ENOENT));
745 static int lustre_lwp_add_conn(struct lustre_cfg *cfg,
746 struct lustre_sb_info *lsi, u32 idx)
748 struct lustre_cfg_bufs *bufs = NULL;
749 struct lustre_cfg *lcfg = NULL;
750 char *lwpname = NULL;
751 struct obd_device *lwp;
755 lwp = lustre_find_lwp(lsi, &lwpname, idx);
757 CERROR("%s: can't find lwp device.\n", lsi->lsi_svname);
758 GOTO(out, rc = PTR_ERR(lwp));
764 GOTO(out, rc = -ENOMEM);
766 lustre_cfg_bufs_reset(bufs, lwpname);
767 lustre_cfg_bufs_set_string(bufs, 1,
768 lustre_cfg_string(cfg, 1));
770 OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
772 GOTO(out_cfg, rc = -ENOMEM);
773 lustre_cfg_init(lcfg, LCFG_ADD_CONN, bufs);
775 rc = class_add_conn(lwp, lcfg);
777 CERROR("%s: can't add conn: rc = %d\n", lwpname, rc);
780 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount,
781 lcfg->lcfg_buflens));
787 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
792 * Retrieve MDT nids from the client log, then start the lwp device.
793 * there are only two scenarios which would include mdt nid.
795 * marker 5 (flags=0x01, v2.1.54.0) lustre-MDTyyyy 'add mdc' xxx-
796 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
797 * attach 0:lustre-MDTyyyy-mdc 1:mdc 2:lustre-clilmv_UUID
798 * setup 0:lustre-MDTyyyy-mdc 1:lustre-MDTyyyy_UUID 2:192.168.122.162@tcp
799 * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
800 * add_conn 0:lustre-MDTyyyy-mdc 1:192.168.172.1@tcp
801 * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDTyyyy_UUID xxxx
802 * marker 5 (flags=0x02, v2.1.54.0) lustre-MDTyyyy 'add mdc' xxxx-
804 * marker 7 (flags=0x01, v2.1.54.0) lustre-MDTyyyy 'add failnid' xxxx-
805 * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
806 * add_conn 0:lustre-MDTyyyy-mdc 1:192.168.122.2@tcp
807 * marker 7 (flags=0x02, v2.1.54.0) lustre-MDTyyyy 'add failnid' xxxx-
809 static int client_lwp_config_process(const struct lu_env *env,
810 struct llog_handle *handle,
811 struct llog_rec_hdr *rec, void *data)
813 struct config_llog_instance *cfg = data;
814 int cfg_len = rec->lrh_len;
815 char *cfg_buf = (char *) (rec + 1);
816 struct lustre_cfg *lcfg = NULL;
817 struct lustre_sb_info *lsi;
818 int rc = 0, swab = 0;
821 if (rec->lrh_type != OBD_CFG_REC) {
822 CERROR("Unknown llog record type %#x encountered\n",
828 GOTO(out, rc = -EINVAL);
829 lsi = s2lsi(cfg->cfg_sb);
831 lcfg = (struct lustre_cfg *)cfg_buf;
832 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
833 lustre_swab_lustre_cfg(lcfg);
837 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
841 switch (lcfg->lcfg_command) {
843 struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
845 lustre_swab_cfg_marker(marker, swab,
846 LUSTRE_CFG_BUFLEN(lcfg, 1));
847 if (marker->cm_flags & CM_SKIP ||
848 marker->cm_flags & CM_EXCLUDE)
851 if (!tgt_is_mdt(marker->cm_tgtname, &cfg->cfg_lwp_idx))
854 if (IS_MDT(lsi) && cfg->cfg_lwp_idx != 0)
857 if (!strncmp(marker->cm_comment, "add mdc", 7) ||
858 !strncmp(marker->cm_comment, "add failnid", 11)) {
859 if (marker->cm_flags & CM_START) {
860 cfg->cfg_flags = CFG_F_MARKER;
861 /* This hack is to differentiate the
862 * ADD_UUID is come from "add mdc" record
863 * or from "add failnid" record.
865 if (!strncmp(marker->cm_comment,
867 cfg->cfg_flags |= CFG_F_SKIP;
868 } else if (marker->cm_flags & CM_END) {
874 case LCFG_ADD_UUID: {
875 if (cfg->cfg_flags == CFG_F_MARKER) {
876 rc = lustre_lwp_setup(lcfg, lsi, cfg->cfg_lwp_idx);
877 /* XXX: process only the first nid as
878 * we don't need another instance of lwp
880 cfg->cfg_flags |= CFG_F_SKIP;
881 } else if (cfg->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
886 lnet_nid4_to_nid(lcfg->lcfg_nid, &nid);
888 rc = libcfs_strnid(&nid,
889 lustre_cfg_string(lcfg, 2));
891 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
894 CERROR("%s: Fail to add uuid, rc:%d\n",
895 lsi->lsi_svname, rc);
899 case LCFG_ADD_CONN: {
900 char *devname = lustre_cfg_string(lcfg, 0);
904 if (!is_mdc_device(devname))
907 if (!(cfg->cfg_flags & CFG_F_MARKER)) {
908 CDEBUG(D_CONFIG, "Skipping add_conn for %s, rec %d\n",
909 devname, rec->lrh_index);
913 /* add_conn should follow by add_uuid. This
914 * guarantee lwp device was created
916 if (!(cfg->cfg_flags & CFG_F_SKIP)) {
917 CWARN("Error at config for %s rec %d, add_conn should follow by add_uuid\n",
918 devname, rec->lrh_index);
921 ptr = strrchr(devname, '-');
926 if (!tgt_is_mdt(devname, &idx)) {
932 if (IS_MDT(lsi) && idx != 0)
935 rc = lustre_lwp_add_conn(lcfg, lsi, idx);
945 static int lustre_disconnect_lwp(struct super_block *sb)
947 struct lustre_sb_info *lsi = s2lsi(sb);
948 struct obd_device *lwp;
949 char *logname = NULL;
950 struct lustre_cfg_bufs *bufs = NULL;
951 struct config_llog_instance *cfg = NULL;
956 if (likely(lsi->lsi_lwp_started)) {
957 OBD_ALLOC(logname, MTI_NAME_MAXLEN);
961 rc = server_name2fsname(lsi->lsi_svname, logname, NULL);
963 CERROR("%s: failed to get fsname from svname: rc = %d\n",
964 lsi->lsi_svname, rc);
965 GOTO(out, rc = -EINVAL);
968 strcat(logname, "-client");
971 GOTO(out, rc = -ENOMEM);
974 cfg->cfg_instance = ll_get_cfg_instance(sb);
975 rc = lustre_end_log(sb, logname, cfg);
976 if (rc != 0 && rc != -ENOENT)
979 lsi->lsi_lwp_started = 0;
984 GOTO(out, rc = -ENOMEM);
986 mutex_lock(&lsi->lsi_lwp_mutex);
987 list_for_each_entry(lwp, &lsi->lsi_lwp_list, obd_lwp_list) {
988 struct lustre_cfg *lcfg;
990 if (likely(lwp->obd_lwp_export)) {
991 class_export_put(lwp->obd_lwp_export);
992 lwp->obd_lwp_export = NULL;
995 lustre_cfg_bufs_reset(bufs, lwp->obd_name);
996 lustre_cfg_bufs_set_string(bufs, 1, NULL);
997 OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount,
1003 lustre_cfg_init(lcfg, LCFG_CLEANUP, bufs);
1005 /* Disconnect import first. NULL is passed for the '@env',
1006 * since it will not be used.
1008 rc = lwp->obd_lu_dev->ld_ops->ldo_process_config(NULL,
1009 lwp->obd_lu_dev, lcfg);
1010 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount,
1011 lcfg->lcfg_buflens));
1012 if (rc != 0 && rc != -ETIMEDOUT && rc != -ENODEV &&
1013 rc != -ENOTCONN && rc != -ESHUTDOWN) {
1014 CERROR("%s: fail to disconnect LWP: rc = %d\n",
1019 mutex_unlock(&lsi->lsi_lwp_mutex);
1029 OBD_FREE(logname, MTI_NAME_MAXLEN);
1031 return rc1 != 0 ? rc1 : rc;
1035 * Stop the lwp for an OST/MDT target.
1037 static int lustre_stop_lwp(struct super_block *sb)
1039 struct lustre_sb_info *lsi = s2lsi(sb);
1040 struct obd_device *lwp;
1045 mutex_lock(&lsi->lsi_lwp_mutex);
1046 while (!list_empty(&lsi->lsi_lwp_list)) {
1047 lwp = list_first_entry(&lsi->lsi_lwp_list, struct obd_device,
1049 list_del_init(&lwp->obd_lwp_list);
1051 mutex_unlock(&lsi->lsi_lwp_mutex);
1053 rc = class_manual_cleanup(lwp);
1055 CERROR("%s: fail to stop LWP: rc = %d\n",
1059 mutex_lock(&lsi->lsi_lwp_mutex);
1061 mutex_unlock(&lsi->lsi_lwp_mutex);
1063 RETURN(rc1 != 0 ? rc1 : rc);
1067 * Start the lwp(fsname-MDTyyyy-lwp-{MDT,OST}xxxx) for a MDT/OST or MDT target.
1069 static int lustre_start_lwp(struct super_block *sb)
1071 struct lustre_sb_info *lsi = s2lsi(sb);
1072 struct config_llog_instance *cfg = NULL;
1077 if (unlikely(lsi->lsi_lwp_started))
1080 OBD_ALLOC(logname, MTI_NAME_MAXLEN);
1081 if (logname == NULL)
1084 rc = server_name2fsname(lsi->lsi_svname, logname, NULL);
1086 CERROR("%s: failed to get fsname from svname: rc = %d\n",
1087 lsi->lsi_svname, rc);
1088 GOTO(out, rc = -EINVAL);
1091 strcat(logname, "-client");
1094 GOTO(out, rc = -ENOMEM);
1096 cfg->cfg_callback = client_lwp_config_process;
1097 cfg->cfg_instance = ll_get_cfg_instance(sb);
1098 rc = lustre_process_log(sb, logname, cfg);
1099 /* need to remove config llog from mgc */
1100 lsi->lsi_lwp_started = 1;
1105 OBD_FREE(logname, MTI_NAME_MAXLEN);
1112 static DEFINE_MUTEX(server_start_lock);
1114 /* Stop MDS/OSS if nobody is using them */
1115 static int server_stop_servers(int lsiflags)
1117 struct obd_device *obd = NULL;
1118 struct obd_type *type = NULL;
1123 mutex_lock(&server_start_lock);
1125 /* Either an MDT or an OST or neither */
1126 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
1127 if (lsiflags & LDD_F_SV_TYPE_MDT) {
1128 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1129 type = class_search_type(LUSTRE_MDT_NAME);
1130 } else if (lsiflags & LDD_F_SV_TYPE_OST) {
1131 /* if this was an OST, and there are no more OST's, clean up the OSS */
1132 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1133 type = class_search_type(LUSTRE_OST_NAME);
1136 /* server_stop_servers is a pair of server_start_targets
1137 * Here we put type which was taken at server_start_targets.
1138 * If type is NULL then there is a wrong logic around type or
1141 LASSERTF(type, "Server flags %d, obd %s\n", lsiflags,
1142 obd ? obd->obd_name : "NULL");
1144 type_last = (atomic_read(&type->typ_refcnt) == 1);
1146 class_put_type(type);
1147 if (obd && type_last) {
1149 /* obd_fail doesn't mean much on a server obd */
1150 rc = class_manual_cleanup(obd);
1153 /* put reference taken by class_search_type */
1154 kobject_put(&type->typ_kobj);
1156 mutex_unlock(&server_start_lock);
1161 int server_mti_print(const char *title, struct mgs_target_info *mti)
1163 CDEBUG(D_MOUNT, "mti - %s\n", title);
1164 CDEBUG(D_MOUNT, "server: %s\n", mti->mti_svname);
1165 CDEBUG(D_MOUNT, "fs: %s\n", mti->mti_fsname);
1166 CDEBUG(D_MOUNT, "uuid: %s\n", mti->mti_uuid);
1167 CDEBUG(D_MOUNT, "ver: %d\n", mti->mti_config_ver);
1168 CDEBUG(D_MOUNT, "flags:\n");
1169 if (mti->mti_flags & LDD_F_SV_TYPE_MDT)
1170 CDEBUG(D_MOUNT, " LDD_F_SV_TYPE_MDT\n");
1171 if (mti->mti_flags & LDD_F_SV_TYPE_OST)
1172 CDEBUG(D_MOUNT, " LDD_F_SV_TYPE_OST\n");
1173 if (mti->mti_flags & LDD_F_SV_TYPE_MGS)
1174 CDEBUG(D_MOUNT, " LDD_F_SV_TYPE_MGS\n");
1175 if (mti->mti_flags & LDD_F_SV_ALL)
1176 CDEBUG(D_MOUNT, " LDD_F_SV_ALL\n");
1177 if (mti->mti_flags & LDD_F_NEED_INDEX)
1178 CDEBUG(D_MOUNT, " LDD_F_NEED_INDEX\n");
1179 if (mti->mti_flags & LDD_F_VIRGIN)
1180 CDEBUG(D_MOUNT, " LDD_F_VIRIGIN\n");
1181 if (mti->mti_flags & LDD_F_UPDATE)
1182 CDEBUG(D_MOUNT, " LDD_F_UPDATE\n");
1183 if (mti->mti_flags & LDD_F_REWRITE_LDD)
1184 CDEBUG(D_MOUNT, " LDD_F_REWRITE_LDD\n");
1185 if (mti->mti_flags & LDD_F_WRITECONF)
1186 CDEBUG(D_MOUNT, " LDD_F_WRITECONF\n");
1187 if (mti->mti_flags & LDD_F_PARAM)
1188 CDEBUG(D_MOUNT, " LDD_F_PARAM\n");
1189 if (mti->mti_flags & LDD_F_NO_PRIMNODE)
1190 CDEBUG(D_MOUNT, " LDD_F_NO_PRIMNODE\n");
1191 if (mti->mti_flags & LDD_F_IR_CAPABLE)
1192 CDEBUG(D_MOUNT, " LDD_F_IR_CAPABLE\n");
1193 if (mti->mti_flags & LDD_F_ERROR)
1194 CDEBUG(D_MOUNT, " LDD_F_ERROR\n");
1195 if (mti->mti_flags & LDD_F_PARAM2)
1196 CDEBUG(D_MOUNT, " LDD_F_PARAM2\n");
1197 if (mti->mti_flags & LDD_F_NO_LOCAL_LOGS)
1198 CDEBUG(D_MOUNT, " LDD_F_NO_LOCAL_LOGS\n");
1200 /* Upper 16 bits for target registering */
1201 if (target_supports_large_nid(mti))
1202 CDEBUG(D_MOUNT, " LDD_F_LARGE_NID\n");
1203 if (mti->mti_flags & LDD_F_OPC_REG)
1204 CDEBUG(D_MOUNT, " LDD_F_OPC_REG\n");
1205 if (mti->mti_flags & LDD_F_OPC_UNREG)
1206 CDEBUG(D_MOUNT, " LDD_F_OPC_UNREG\n");
1207 if (mti->mti_flags & LDD_F_OPC_READY)
1208 CDEBUG(D_MOUNT, " LDD_F_OPC_READY\n");
1212 EXPORT_SYMBOL(server_mti_print);
1214 /* Generate data for registration */
1215 static struct mgs_target_info *server_lsi2mti(struct lustre_sb_info *lsi)
1217 size_t len = offsetof(struct mgs_target_info, mti_nidlist);
1218 GENRADIX(struct lnet_processid) plist;
1219 struct lnet_processid id, *tmp;
1220 struct mgs_target_info *mti;
1221 bool large_nid = false;
1227 if (!IS_SERVER(lsi))
1228 RETURN(ERR_PTR(-EINVAL));
1230 if (exp_connect_flags2(lsi->lsi_mgc->u.cli.cl_mgc_mgsexp) &
1231 OBD_CONNECT2_LARGE_NID)
1234 genradix_init(&plist);
1236 while (LNetGetId(i++, &id, large_nid) != -ENOENT) {
1237 if (nid_is_lo0(&id.nid))
1240 /* server use --servicenode param, only allow specified
1241 * nids be registered
1243 if (test_bit(LMD_FLG_NO_PRIMNODE, lsi->lsi_lmd->lmd_flags) &&
1244 class_match_nid(lsi->lsi_lmd->lmd_params,
1245 PARAM_FAILNODE, &id.nid) < 1)
1248 if (!class_find_param(lsi->lsi_lmd->lmd_params,
1249 PARAM_NETWORK, NULL)) {
1250 if (LNetGetPeerDiscoveryStatus()) {
1251 CERROR("LNet Dynamic Peer Discovery is enabled"
1252 " on this node. 'network' option used in"
1253 " mkfs.lustre cannot be taken into"
1255 GOTO(free_list, mti = ERR_PTR(-EINVAL));
1259 /* match specified network */
1260 if (!class_match_net(lsi->lsi_lmd->lmd_params,
1261 PARAM_NETWORK, LNET_NID_NET(&id.nid)))
1264 tmp = genradix_ptr_alloc(&plist, nid_count++, GFP_KERNEL);
1266 GOTO(free_list, mti = ERR_PTR(-ENOMEM));
1269 len += LNET_NIDSTR_SIZE;
1273 if (nid_count == 0) {
1274 CERROR("Failed to get NID for server %s, please check whether the target is specifed with improper --servicenode or --network options.\n",
1276 GOTO(free_list, mti = ERR_PTR(-EINVAL));
1279 OBD_ALLOC(mti, len);
1281 GOTO(free_list, mti = ERR_PTR(-ENOMEM));
1283 if (strlcpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname))
1284 >= sizeof(mti->mti_svname))
1285 GOTO(free_mti, rc = -E2BIG);
1287 mti->mti_nid_count = nid_count;
1288 for (i = 0; i < mti->mti_nid_count; i++) {
1289 tmp = genradix_ptr(&plist, i);
1292 libcfs_nidstr_r(&tmp->nid, mti->mti_nidlist[i],
1293 sizeof(mti->mti_nidlist[i]));
1295 mti->mti_nids[i] = lnet_nid_to_nid4(&tmp->nid);
1297 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1298 mti->mti_config_ver = 0;
1300 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1304 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1308 /* Orion requires index to be set */
1309 LASSERT(!(rc & LDD_F_NEED_INDEX));
1310 /* keep only LDD flags */
1311 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1312 if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN))
1313 mti->mti_flags |= LDD_F_UPDATE;
1314 /* use NID strings instead */
1316 mti->mti_flags |= LDD_F_LARGE_NID;
1317 cplen = strlcpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1318 sizeof(mti->mti_params));
1319 if (cplen >= sizeof(mti->mti_params))
1327 genradix_free(&plist);
1332 /* Register an old or new target with the MGS. If needed MGS will construct
1333 * startup logs and assign index
1335 static int server_register_target(struct lustre_sb_info *lsi)
1337 struct obd_device *mgc = lsi->lsi_mgc;
1338 struct mgs_target_info *mti = NULL;
1339 size_t mti_len = sizeof(*mti);
1346 mti = server_lsi2mti(lsi);
1348 GOTO(out, rc = PTR_ERR(mti));
1350 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1351 mti->mti_svname, mti->mti_fsname, mti->mti_nidlist[0],
1352 mti->mti_stripe_index, mti->mti_flags);
1354 /* we cannot ignore registration failure if MGS logs must be updated. */
1355 must_succeed = !!(lsi->lsi_flags &
1356 (LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_WRITECONF |
1358 mti->mti_flags |= LDD_F_OPC_REG;
1359 if (target_supports_large_nid(mti))
1360 mti_len += mti->mti_nid_count * LNET_NIDSTR_SIZE;
1361 server_mti_print("server_register_target", mti);
1363 /* Register the target */
1364 /* FIXME use mgc_process_config instead */
1365 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1366 sizeof(KEY_REGISTER_TARGET),
1367 KEY_REGISTER_TARGET,
1368 mti_len, mti, NULL);
1370 if (mti->mti_flags & LDD_F_ERROR) {
1371 LCONSOLE_ERROR_MSG(0x160,
1372 "%s: the MGS refuses to allow this server to start: rc = %d. Please see messages on the MGS.\n",
1373 lsi->lsi_svname, rc);
1374 } else if (must_succeed) {
1375 if ((rc == -ESHUTDOWN || rc == -EIO) && ++tried < 5) {
1376 /* The connection with MGS is not established.
1377 * Try again after 2 seconds. Interruptable.
1379 schedule_timeout_interruptible(cfs_time_seconds(2));
1380 if (!signal_pending(current))
1384 LCONSOLE_ERROR_MSG(0x15f,
1385 "%s: cannot register this server with the MGS: rc = %d. Is the MGS running?\n",
1386 lsi->lsi_svname, rc);
1389 "%s: error registering with the MGS: rc = %d (not fatal)\n",
1390 lsi->lsi_svname, rc);
1391 /* reset the error code for non-fatal error. */
1396 OBD_FREE(mti, mti_len);
1402 * Notify the MGS that this target is ready.
1403 * Used by IR - if the MGS receives this message, it will notify clients.
1405 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1407 struct lustre_sb_info *lsi = s2lsi(sb);
1408 struct obd_device *mgc = lsi->lsi_mgc;
1409 struct mgs_target_info *mti = NULL;
1410 size_t mti_len = sizeof(*mti);
1415 mti = server_lsi2mti(lsi);
1417 GOTO(out, rc = PTR_ERR(mti));
1419 mti->mti_instance = obd2obt(obd)->obt_instance;
1420 mti->mti_flags |= LDD_F_OPC_READY;
1421 if (target_supports_large_nid(mti))
1422 mti_len += mti->mti_nid_count * LNET_NIDSTR_SIZE;
1423 server_mti_print("server_notify_target", mti);
1425 /* FIXME use mgc_process_config instead */
1426 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1427 sizeof(KEY_REGISTER_TARGET),
1428 KEY_REGISTER_TARGET,
1429 mti_len, mti, NULL);
1431 /* Imperative recovery: if the mgs informs us to use IR? */
1432 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1433 (mti->mti_flags & LDD_F_IR_CAPABLE))
1434 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1436 OBD_FREE(mti, mti_len);
1441 /* Start server targets: MDTs and OSTs */
1442 static int server_start_targets(struct super_block *sb)
1444 struct obd_device *obd;
1445 struct lustre_sb_info *lsi = s2lsi(sb);
1446 struct config_llog_instance cfg;
1447 struct lu_env mgc_env;
1448 struct lu_device *dev;
1449 char *name_service, *obd_name_service = NULL;
1450 struct obd_type *type = NULL;
1454 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1456 LASSERTF(IS_MDT(lsi) || IS_OST(lsi), "designed for MDT or OST only\n");
1459 obd_name_service = LUSTRE_MDS_OBDNAME;
1460 name_service = LUSTRE_MDS_NAME;
1462 obd_name_service = LUSTRE_OSS_OBDNAME;
1463 name_service = LUSTRE_OSS_NAME;
1466 /* make sure MDS/OSS is started */
1467 mutex_lock(&server_start_lock);
1468 obd = class_name2obd(obd_name_service);
1470 rc = lustre_start_simple(obd_name_service, name_service,
1472 LUSTRE_MDS_OBDNAME"_uuid" :
1473 LUSTRE_OSS_OBDNAME"_uuid"),
1474 NULL, NULL, NULL, NULL);
1476 mutex_unlock(&server_start_lock);
1477 CERROR("failed to start %s: %d\n",
1478 obd_name_service, rc);
1482 /* hold a type reference and put it at server_stop_servers */
1483 type = class_get_type(IS_MDT(lsi) ?
1484 LUSTRE_MDT_NAME : LUSTRE_OST_NAME);
1486 mutex_unlock(&server_start_lock);
1487 GOTO(out_stop_service, rc = -ENODEV);
1489 lsi->lsi_server_started = 1;
1490 mutex_unlock(&server_start_lock);
1491 if (CFS_FAIL_PRECHECK(OBD_FAIL_OBD_STOP_MDS_RACE) &&
1493 CFS_RACE(OBD_FAIL_OBD_STOP_MDS_RACE);
1494 msleep(2 * MSEC_PER_SEC);
1497 rc = lu_env_init(&mgc_env, LCT_MG_THREAD);
1499 GOTO(out_stop_service, rc);
1501 /* Set the mgc fs to our server disk. This allows the MGC to
1502 * read and write configs locally, in case it can't talk to the MGS.
1504 rc = server_mgc_set_fs(&mgc_env, lsi->lsi_mgc, sb);
1508 /* Register with MGS */
1509 rc = server_register_target(lsi);
1513 /* Let the target look up the mount using the target's name
1514 * (we can't pass the sb or mnt through class_process_config.)
1516 rc = server_register_mount(lsi->lsi_svname, sb);
1520 /* Start targets using the llog named for the target */
1521 memset(&cfg, 0, sizeof(cfg));
1522 cfg.cfg_callback = class_config_llog_handler;
1523 cfg.cfg_sub_clds = CONFIG_SUB_SERVER;
1524 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1526 CERROR("failed to start server %s: %d\n",
1527 lsi->lsi_svname, rc);
1528 /* Do NOT call server_deregister_mount() here. This makes it
1529 * impossible to find mount later in cleanup time and leaves
1530 * @lsi and othder stuff leaked. -umka
1535 obd = class_name2obd(lsi->lsi_svname);
1537 CERROR("no server named %s was started\n", lsi->lsi_svname);
1538 GOTO(out_mgc, rc = -ENXIO);
1541 if (IS_OST(lsi) || IS_MDT(lsi)) {
1542 rc = lustre_start_lwp(sb);
1544 CERROR("%s: failed to start LWP: %d\n",
1545 lsi->lsi_svname, rc);
1550 server_notify_target(sb, obd);
1552 /* calculate recovery timeout, do it after lustre_process_log */
1553 server_calc_timeout(lsi, obd);
1555 /* log has been fully processed, let clients connect */
1556 dev = obd->obd_lu_dev;
1557 if (dev && dev->ld_ops->ldo_prepare) {
1560 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1562 struct lu_context session_ctx;
1564 lu_context_init(&session_ctx, LCT_SERVER_SESSION);
1565 session_ctx.lc_thread = NULL;
1566 lu_context_enter(&session_ctx);
1567 env.le_ses = &session_ctx;
1569 rc = dev->ld_ops->ldo_prepare(&env, NULL, dev);
1572 lu_context_exit(&session_ctx);
1573 lu_context_fini(&session_ctx);
1577 /* abort recovery only on the complete stack:
1578 * many devices can be involved
1580 if ((test_bit(LMD_FLG_ABORT_RECOV, lsi->lsi_lmd->lmd_flags) ||
1581 (test_bit(LMD_FLG_ABORT_RECOV_MDT, lsi->lsi_lmd->lmd_flags))) &&
1582 (obd->obd_type->typ_dt_ops->o_iocontrol)) {
1583 struct obd_ioctl_data karg;
1585 if (test_bit(LMD_FLG_ABORT_RECOV, lsi->lsi_lmd->lmd_flags))
1586 karg.ioc_type = OBD_FLG_ABORT_RECOV_OST;
1588 karg.ioc_type = OBD_FLG_ABORT_RECOV_MDT;
1590 obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
1595 /* Release the mgc fs for others to use */
1596 server_mgc_clear_fs(&mgc_env, lsi->lsi_mgc);
1598 lu_env_fini(&mgc_env);
1600 /* in case of error upper function call
1601 * server_put_super->server_stop_servers()
1607 static int lsi_prepare(struct lustre_sb_info *lsi)
1609 const char *osd_type;
1616 LASSERT(lsi->lsi_lmd);
1618 /* The server name is given as a mount line option */
1619 if (!lsi->lsi_lmd->lmd_profile) {
1620 LCONSOLE_ERROR("Can't determine server name\n");
1624 /* Determine osd type */
1625 if (!lsi->lsi_lmd->lmd_osd_type) {
1626 osd_type = LUSTRE_OSD_LDISKFS_NAME;
1629 osd_type = lsi->lsi_lmd->lmd_osd_type;
1630 fstype = lsi->lsi_lmd->lmd_osd_type;
1633 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname) ||
1634 strlen(osd_type) >= sizeof(lsi->lsi_osd_type) ||
1635 strlen(fstype) >= sizeof(lsi->lsi_fstype))
1636 RETURN(-ENAMETOOLONG);
1638 strlcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile,
1639 sizeof(lsi->lsi_svname));
1640 strlcpy(lsi->lsi_osd_type, osd_type, sizeof(lsi->lsi_osd_type));
1641 /* XXX: a temp. solution for components using ldiskfs
1642 * to be removed in one of the subsequent patches
1644 strlcpy(lsi->lsi_fstype, fstype, sizeof(lsi->lsi_fstype));
1646 /* Determine server type */
1647 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1649 if (test_bit(LMD_FLG_MGS, lsi->lsi_lmd->lmd_flags)) {
1650 /* Assume we're a bare MGS */
1652 set_bit(LMD_FLG_NOSVC, lsi->lsi_lmd->lmd_flags);
1654 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
1659 lsi->lsi_flags |= rc;
1661 /* Add mount line flags that used to be in ldd:
1662 * writeconf, mgs, anything else?
1664 lsi->lsi_flags |= test_bit(LMD_FLG_WRITECONF, lsi->lsi_lmd->lmd_flags) ?
1665 LDD_F_WRITECONF : 0;
1666 lsi->lsi_flags |= test_bit(LMD_FLG_NO_LOCAL_LOGS, lsi->lsi_lmd->lmd_flags) ?
1667 LDD_F_NO_LOCAL_LOGS : 0;
1668 lsi->lsi_flags |= test_bit(LMD_FLG_VIRGIN, lsi->lsi_lmd->lmd_flags) ?
1670 lsi->lsi_flags |= test_bit(LMD_FLG_UPDATE, lsi->lsi_lmd->lmd_flags) ?
1672 lsi->lsi_flags |= test_bit(LMD_FLG_MGS, lsi->lsi_lmd->lmd_flags) ?
1673 LDD_F_SV_TYPE_MGS : 0;
1674 lsi->lsi_flags |= test_bit(LMD_FLG_NO_PRIMNODE, lsi->lsi_lmd->lmd_flags) ?
1675 LDD_F_NO_PRIMNODE : 0;
1680 /*************** server mount ******************/
1682 /** Start the shutdown of servers at umount.
1684 static void server_put_super(struct super_block *sb)
1686 struct lustre_sb_info *lsi = s2lsi(sb);
1687 struct obd_device *obd;
1688 char *tmpname, *extraname = NULL;
1690 int lsiflags = lsi->lsi_flags;
1691 bool stop_servers = lsi->lsi_server_started;
1694 LASSERT(IS_SERVER(lsi));
1696 tmpname_sz = strlen(lsi->lsi_svname) + 1;
1697 OBD_ALLOC(tmpname, tmpname_sz);
1698 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
1699 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1700 if (IS_MDT(lsi) && test_bit(LMD_FLG_NOSVC, lsi->lsi_lmd->lmd_flags))
1701 snprintf(tmpname, tmpname_sz, "MGS");
1703 /* disconnect the lwp first to drain off the inflight request */
1704 if (IS_OST(lsi) || IS_MDT(lsi)) {
1707 rc = lustre_disconnect_lwp(sb);
1708 if (rc != 0 && rc != -ETIMEDOUT && rc != -ENODEV &&
1709 rc != -ENOTCONN && rc != -ESHUTDOWN)
1710 CWARN("%s: failed to disconnect lwp: rc= %d\n",
1714 /* Stop the target */
1715 if (!test_bit(LMD_FLG_NOSVC, lsi->lsi_lmd->lmd_flags) &&
1716 (IS_MDT(lsi) || IS_OST(lsi))) {
1717 struct lustre_profile *lprof = NULL;
1719 /* tell the mgc to drop the config log */
1720 lustre_end_log(sb, lsi->lsi_svname, NULL);
1722 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1723 * If there are any setup/cleanup errors, save the lov
1724 * name for safety cleanup later.
1726 lprof = class_get_profile(lsi->lsi_svname);
1729 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1730 strncpy(extraname, lprof->lp_dt,
1731 strlen(lprof->lp_dt) + 1);
1733 class_put_profile(lprof);
1736 obd = class_name2obd(lsi->lsi_svname);
1738 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1739 if (lsiflags & LSI_UMOUNT_FAILOVER)
1741 /* We can't seem to give an error return code
1742 * to .put_super, so we better make sure we clean up!
1745 class_manual_cleanup(obd);
1746 if (CFS_FAIL_PRECHECK(OBD_FAIL_OBD_STOP_MDS_RACE)) {
1749 server_name2index(lsi->lsi_svname, &idx, NULL);
1750 /* sleeping for MDT0001 */
1752 CFS_RACE(OBD_FAIL_OBD_STOP_MDS_RACE);
1755 CERROR("no obd %s\n", lsi->lsi_svname);
1756 server_deregister_mount(lsi->lsi_svname);
1760 /* If they wanted the mgs to stop separately from the mdt, they
1761 * should have put it on a different device.
1763 lustre_stop_mgc(sb);
1765 /* if MDS start with --nomgs, don't stop MGS then */
1766 if (!test_bit(LMD_FLG_NOMGS, lsi->lsi_lmd->lmd_flags))
1767 server_stop_mgs(sb);
1770 if (IS_OST(lsi) || IS_MDT(lsi)) {
1771 if (lustre_stop_lwp(sb) < 0)
1772 CERROR("%s: failed to stop lwp!\n", tmpname);
1775 /* Drop a ref to the mounted disk */
1778 /* wait till all in-progress cleanups are done
1779 * specifically we're interested in ofd cleanup
1782 obd_zombie_barrier();
1784 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1785 * until the target is really gone so that our type refcount check
1789 server_stop_servers(lsiflags);
1791 /* In case of startup or cleanup err, stop related obds */
1793 obd = class_name2obd(extraname);
1795 CWARN("Cleaning orphaned obd %s\n", extraname);
1797 class_manual_cleanup(obd);
1799 OBD_FREE(extraname, strlen(extraname) + 1);
1802 LCONSOLE(D_WARNING, "server umount %s complete\n", tmpname);
1803 OBD_FREE(tmpname, tmpname_sz);
1807 /* Called only for 'umount -f' */
1808 static void server_umount_begin(struct super_block *sb)
1810 struct lustre_sb_info *lsi = s2lsi(sb);
1813 CDEBUG(D_MOUNT, "umount -f\n");
1814 /* umount = failover
1816 * no third way to do non-force, non-failover
1818 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1822 static int server_statfs(struct dentry *dentry, struct kstatfs *buf)
1824 struct super_block *sb = dentry->d_sb;
1825 struct lustre_sb_info *lsi = s2lsi(sb);
1826 struct obd_statfs statfs;
1830 if (lsi->lsi_dt_dev) {
1831 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
1833 statfs_unpack(buf, &statfs);
1834 buf->f_type = sb->s_magic;
1840 buf->f_type = sb->s_magic;
1841 buf->f_bsize = sb->s_blocksize;
1847 buf->f_namelen = NAME_MAX;
1851 static int server_show_options(struct seq_file *seq, struct dentry *dentry)
1853 struct lustre_sb_info *lsi;
1854 struct lustre_mount_data *lmd;
1855 struct obd_statfs osfs;
1856 struct super_block *sb;
1859 LASSERT(seq && dentry);
1860 lsi = s2lsi(dentry->d_sb);
1864 if (lsi->lsi_dt_dev) {
1865 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &osfs);
1867 /* Check FS State for OS_STATFS_READONLY
1868 * (Read only) flag. If it is not set then
1869 * toggle back the s_flag's SB_RDONLY bit.
1870 * The SB_RDONLY bit is always set for OST/MDT
1871 * during server prep (server_fill_super_common())
1874 * Also, if server is mounted with "rdonly_dev"
1875 * (LMD_FLG_DEV_RDONLY) then force flag to be 'ro'
1877 if (!test_bit(LMD_FLG_DEV_RDONLY, lmd->lmd_flags) &&
1878 !(osfs.os_state & OS_STATFS_READONLY))
1879 sb->s_flags &= ~SB_RDONLY;
1883 seq_printf(seq, ",svname=%s", lmd->lmd_profile);
1885 if (test_bit(LMD_FLG_ABORT_RECOV, lmd->lmd_flags))
1886 seq_puts(seq, ",abort_recov");
1888 if (test_bit(LMD_FLG_NOIR, lmd->lmd_flags))
1889 seq_puts(seq, ",noir");
1891 if (test_bit(LMD_FLG_NOSVC, lmd->lmd_flags))
1892 seq_puts(seq, ",nosvc");
1894 if (test_bit(LMD_FLG_NOMGS, lmd->lmd_flags))
1895 seq_puts(seq, ",nomgs");
1897 if (test_bit(LMD_FLG_NOSCRUB, lmd->lmd_flags))
1898 seq_puts(seq, ",noscrub");
1900 if (test_bit(LMD_FLG_SKIP_LFSCK, lmd->lmd_flags))
1901 seq_puts(seq, ",skip_lfsck");
1903 if (test_bit(LMD_FLG_DEV_RDONLY, lmd->lmd_flags))
1904 seq_puts(seq, ",rdonly_dev");
1906 if (test_bit(LMD_FLG_MGS, lmd->lmd_flags))
1907 seq_puts(seq, ",mgs");
1910 seq_printf(seq, ",mgsnode=%s", lmd->lmd_mgs);
1912 if (lmd->lmd_osd_type)
1913 seq_printf(seq, ",osd=%s", lmd->lmd_osd_type);
1915 if (lmd->lmd_opts) {
1917 seq_puts(seq, lmd->lmd_opts);
1923 /** The operations we support directly on the superblock:
1924 * mount, umount, and df.
1926 static const struct super_operations server_ops = {
1927 .put_super = server_put_super,
1928 .umount_begin = server_umount_begin, /* umount -f */
1929 .statfs = server_statfs,
1930 .show_options = server_show_options,
1933 #if defined(HAVE_USER_NAMESPACE_ARG)
1934 # define IDMAP_ARG idmap,
1937 # ifdef HAVE_INODEOPS_ENHANCED_GETATTR
1938 # define server_getattr(ns, path, st, rq, fl) server_getattr(path, st, rq, fl)
1943 * inode operations for Lustre server mountpoints
1945 #if defined(HAVE_USER_NAMESPACE_ARG) || defined(HAVE_INODEOPS_ENHANCED_GETATTR)
1946 static int server_getattr(struct mnt_idmap *idmap,
1947 const struct path *path, struct kstat *stat,
1948 u32 request_mask, unsigned int flags)
1950 struct inode *inode = d_inode(path->dentry);
1952 static int server_getattr(struct vfsmount *mnt, struct dentry *de,
1955 struct inode *inode = de->d_inode;
1957 struct lustre_sb_info *lsi = s2lsi(inode->i_sb);
1958 struct vfsmount *root_mnt;
1959 struct inode *root_inode;
1961 root_mnt = dt_mnt_get(lsi->lsi_dt_dev);
1962 if (IS_ERR(root_mnt))
1963 root_inode = igrab(inode);
1965 root_inode = igrab(root_mnt->mnt_sb->s_root->d_inode);
1969 CDEBUG(D_SUPER, "%s: root_inode from %s ino=%lu, dev=%x\n",
1970 lsi->lsi_svname, root_inode == inode ? "lsi" : "vfsmnt",
1971 root_inode->i_ino, root_inode->i_rdev);
1972 generic_fillattr(IDMAP_ARG RQMASK_ARG root_inode, stat);
1978 #ifdef HAVE_IOP_XATTR
1979 static ssize_t server_getxattr(struct dentry *dentry, const char *name,
1980 void *buffer, size_t size)
1982 if (!selinux_is_enabled())
1987 static int server_setxattr(struct dentry *dentry, const char *name,
1988 const void *value, size_t size, int flags)
1994 static ssize_t server_listxattr(struct dentry *d_entry, char *name,
2000 static bool is_cmd_supported(unsigned int cmd)
2002 CDEBUG(D_SUPER, "ioctl cmd=%x\n", cmd);
2007 case LL_IOC_RESIZE_FS:
2010 case FS_IOC_GETFSMAP:
2020 static long server_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
2022 struct lustre_sb_info *lsi = s2lsi(file_inode(filp)->i_sb);
2023 struct vfsmount *root_mnt;
2024 struct file *root_filp;
2025 struct inode *root_inode;
2028 if (!is_cmd_supported(cmd))
2031 root_mnt = dt_mnt_get(lsi->lsi_dt_dev);
2032 if (IS_ERR(root_mnt))
2035 root_inode = igrab(root_mnt->mnt_root->d_inode);
2039 root_filp = alloc_file_pseudo(root_inode, root_mnt, "/",
2040 O_RDWR | O_NOATIME, root_inode->i_fop);
2041 if (root_inode->i_fop && root_inode->i_fop->unlocked_ioctl)
2042 err = root_inode->i_fop->unlocked_ioctl(root_filp, cmd, arg);
2048 static const struct inode_operations server_inode_operations = {
2049 .getattr = server_getattr,
2050 #ifdef HAVE_IOP_XATTR
2051 .setxattr = server_setxattr,
2052 .getxattr = server_getxattr,
2054 .listxattr = server_listxattr,
2057 static const struct file_operations server_file_operations = {
2058 .unlocked_ioctl = server_ioctl,
2061 #define log2(n) ffz(~(n))
2062 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
2064 static int server_fill_super_common(struct super_block *sb)
2066 struct inode *root = NULL;
2069 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
2071 sb->s_blocksize = 4096;
2072 sb->s_blocksize_bits = log2(sb->s_blocksize);
2073 sb->s_magic = LUSTRE_SUPER_MAGIC;
2074 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
2075 sb->s_flags |= SB_RDONLY;
2076 sb->s_op = &server_ops;
2078 root = new_inode(sb);
2080 CERROR("Can't make root inode\n");
2084 /* returns -EIO for every operation */
2085 /* make_bad_inode(root); -- badness - can't umount */
2086 /* apparently we need to be a directory for the mount to finish */
2087 root->i_mode = S_IFDIR;
2088 root->i_op = &server_inode_operations;
2089 root->i_fop = &server_file_operations;
2090 sb->s_root = d_make_root(root);
2092 CERROR("%s: can't make root dentry\n", sb->s_id);
2099 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
2101 struct lustre_mount_data *lmd = lsi->lsi_lmd;
2102 struct obd_device *obd;
2103 struct dt_device_param p;
2104 char flagstr[20 + 1 + 10 + 1];
2110 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
2111 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
2113 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
2114 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
2115 strcat(lsi->lsi_osd_uuid, "_UUID");
2116 bitmap_to_arr32(&lmd_flags, lmd->lmd_flags, LMD_FLG_NUM_FLAGS);
2117 snprintf(flagstr, sizeof(flagstr), "%lu:%u", mflags, lmd_flags);
2119 obd = class_name2obd(lsi->lsi_osd_obdname);
2121 rc = lustre_start_simple(lsi->lsi_osd_obdname,
2123 lsi->lsi_osd_uuid, lmd->lmd_dev,
2124 flagstr, lsi->lsi_lmd->lmd_opts,
2128 obd = class_name2obd(lsi->lsi_osd_obdname);
2131 CDEBUG(D_MOUNT, "%s already started\n", lsi->lsi_osd_obdname);
2132 /* but continue setup to allow special case of MDT and internal
2133 * MGT being started separately.
2135 if (!((IS_MGS(lsi) &&
2136 test_bit(LMD_FLG_NOMGS, lsi->lsi_lmd->lmd_flags)) ||
2138 test_bit(LMD_FLG_NOSVC, lsi->lsi_lmd->lmd_flags))))
2142 rc = obd_connect(NULL, &lsi->lsi_osd_exp,
2143 obd, &obd->obd_uuid, NULL, NULL);
2147 class_manual_cleanup(obd);
2148 lsi->lsi_dt_dev = NULL;
2152 LASSERT(obd->obd_lu_dev);
2153 lu_device_get(obd->obd_lu_dev);
2154 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
2155 LASSERT(lsi->lsi_dt_dev);
2157 /* set disk context for llog usage */
2158 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
2159 obd->obd_lvfs_ctxt.dt = lsi->lsi_dt_dev;
2161 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
2166 /** Fill in the superblock info for a Lustre server.
2167 * Mount the device with the correct options.
2168 * Read the on-disk config file.
2169 * Start the services.
2171 int server_fill_super(struct super_block *sb)
2173 struct lustre_sb_info *lsi = s2lsi(sb);
2177 /* to simulate target mount race */
2178 CFS_RACE(OBD_FAIL_TGT_MOUNT_RACE);
2180 rc = lsi_prepare(lsi);
2186 /* Start low level OSD */
2187 rc = osd_start(lsi, sb->s_flags);
2189 CERROR("Unable to start osd on %s: %d\n",
2190 lsi->lsi_lmd->lmd_dev, rc);
2195 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
2196 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
2198 if (class_name2obd(lsi->lsi_svname)) {
2199 LCONSOLE_ERROR_MSG(0x161,
2200 "The target named %s is already running. Double-mount may have compromised the disk journal.\n",
2206 /* Start MGS before MGC */
2207 if (IS_MGS(lsi) && !test_bit(LMD_FLG_NOMGS, lsi->lsi_lmd->lmd_flags)) {
2208 rc = server_start_mgs(sb);
2213 /* Start MGC before servers */
2214 rc = lustre_start_mgc(sb);
2218 /* Set up all obd devices for service */
2219 if (!test_bit(LMD_FLG_NOSVC, lsi->lsi_lmd->lmd_flags) &&
2220 (IS_OST(lsi) || IS_MDT(lsi))) {
2221 rc = server_start_targets(sb);
2223 CERROR("Unable to start targets: %d\n", rc);
2226 /* FIXME overmount client here, or can we just start a
2227 * client log and client_fill_super on this sb? We
2228 * need to make sure server_put_super gets called too
2229 * - ll_put_super calls lustre_common_put_super; check
2230 * there for LSI_SERVER flag, call s_p_s if so.
2232 * Probably should start client from new thread so we
2233 * can return. Client will not finish until all
2234 * servers are connected. Note - MGS-only server does
2235 * NOT get a client, since there is no lustre fs
2236 * associated - the MGS is for all lustre fs's
2240 rc = server_fill_super_common(sb);
2246 /* We jump here in case of failure while starting targets or MGS.
2247 * In this case we can't just put @mnt and have to do real cleanup
2248 * with stoping targets, etc.
2250 server_put_super(sb);
2253 EXPORT_SYMBOL(server_fill_super);
2256 * Calculate timeout value for a target.
2258 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
2260 struct lustre_mount_data *lmd;
2264 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
2265 int min = OBD_RECOVERY_TIME_MIN;
2267 LASSERT(IS_SERVER(lsi));
2271 soft = lmd->lmd_recovery_time_soft;
2272 hard = lmd->lmd_recovery_time_hard;
2273 has_ir = has_ir && !test_bit(LMD_FLG_NOIR, lmd->lmd_flags);
2274 obd->obd_no_ir = !has_ir;
2278 soft = OBD_RECOVERY_TIME_SOFT;
2280 hard = OBD_RECOVERY_TIME_HARD;
2282 /* target may have ir_factor configured. */
2283 factor = OBD_IR_FACTOR_DEFAULT;
2284 if (obd->obd_recovery_ir_factor)
2285 factor = obd->obd_recovery_ir_factor;
2288 int new_soft = soft;
2290 /* adjust timeout value by imperative recovery */
2291 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
2292 /* make sure the timeout is not too short */
2293 new_soft = max(min, new_soft);
2295 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery window shrunk from %d-%d down to %d-%d\n",
2296 obd->obd_name, soft, hard, new_soft, hard);
2300 LCONSOLE_INFO("%s: Imperative Recovery not enabled, recovery window %d-%d\n",
2301 obd->obd_name, soft, hard);
2305 obd->obd_recovery_timeout = max_t(time64_t, obd->obd_recovery_timeout,
2307 obd->obd_recovery_time_hard = hard;
2308 obd->obd_recovery_ir_factor = factor;
2312 * This is the entry point for the mount call into Lustre.
2313 * This is called when a server target is mounted,
2314 * and this is where we start setting things up.
2315 * @param data Mount options (e.g. -o flock,abort_recov)
2317 static int lustre_tgt_fill_super(struct super_block *sb, void *lmd2_data,
2320 struct lustre_mount_data *lmd;
2321 struct lustre_sb_info *lsi;
2325 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
2327 lsi = lustre_init_lsi(sb);
2333 * Disable lockdep during mount, because mount locking patterns are
2339 * LU-639: the OBD cleanup of last mount may not finish yet, wait here.
2341 obd_zombie_barrier();
2343 /* Figure out the lmd from the mount options */
2344 if (lmd_parse(lmd2_data, lmd)) {
2346 GOTO(out, rc = -EINVAL);
2349 if (lmd_is_client(lmd)) {
2351 CERROR("%s: attempting to mount a client with -t lustre_tgt' which is only for server-side mounts: rc = %d\n",
2357 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
2358 rc = server_fill_super(sb);
2360 * server_fill_super calls lustre_start_mgc after the mount
2361 * because we need the MGS NIDs which are stored on disk.
2362 * Plus, we may need to start the MGS first.
2364 * server_fill_super will call server_put_super on failure
2366 * If error happens in fill_super() call, @lsi will be killed there.
2367 * This is why we do not put it here.
2371 CERROR("Unable to mount %s (%d)\n",
2372 s2lsi(sb) ? lmd->lmd_dev : "", rc);
2374 CDEBUG(D_SUPER, "Mount %s complete\n",
2381 /***************** FS registration ******************/
2382 static struct dentry *lustre_tgt_mount(struct file_system_type *fs_type,
2383 int flags, const char *devname,
2386 return mount_nodev(fs_type, flags, data, lustre_tgt_fill_super);
2389 /* Register the "lustre_tgt" fs type.
2391 * Right now this isn't any different than the normal "lustre" filesystem
2392 * type, but it is added so that there is some compatibility to allow
2393 * changing documentation and scripts to start using the "lustre_tgt" type
2394 * at mount time. That will simplify test interop, and in case of upgrades
2395 * that change to the new type and then need to roll back for some reason.
2397 * The long-term goal is to disentangle the client and server mount code.
2399 static struct file_system_type lustre_tgt_fstype = {
2400 .owner = THIS_MODULE,
2401 .name = "lustre_tgt",
2402 .mount = lustre_tgt_mount,
2403 .kill_sb = kill_anon_super,
2404 .fs_flags = FS_REQUIRES_DEV | FS_RENAME_DOES_D_MOVE,
2406 MODULE_ALIAS_FS("lustre_tgt");
2408 int lustre_tgt_register_fs(void)
2410 return register_filesystem(&lustre_tgt_fstype);
2413 void lustre_tgt_unregister_fs(void)
2415 unregister_filesystem(&lustre_tgt_fstype);