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 inline bool is_mdc_device(const char *devname)
334 ptr = strrchr(devname, '-');
335 return ptr != NULL && strcmp(ptr, "-mdc") == 0;
338 static inline bool tgt_is_mdt(const char *tgtname, __u32 *idx)
342 type = server_name2index(tgtname, idx, NULL);
344 return type == LDD_F_SV_TYPE_MDT;
348 * Convert OST/MDT name(fsname-{MDT,OST}xxxx) to a lwp name with the @idx:yyyy
349 * (fsname-MDTyyyy-lwp-{MDT,OST}xxxx)
351 int tgt_name2lwp_name(const char *tgt_name, char *lwp_name, int len, __u32 idx)
358 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
362 rc = server_name2fsname(tgt_name, fsname, &tgt);
364 CERROR("%s: failed to get fsname from tgt_name: rc = %d\n",
369 if (*tgt != '-' && *tgt != ':') {
370 CERROR("%s: invalid tgt_name name!\n", tgt_name);
371 GOTO(cleanup, rc = -EINVAL);
375 if (strncmp(tgt, "OST", 3) != 0 && strncmp(tgt, "MDT", 3) != 0) {
376 CERROR("%s is not an OST or MDT target!\n", tgt_name);
377 GOTO(cleanup, rc = -EINVAL);
379 snprintf(lwp_name, len, "%s-MDT%04x-%s-%s",
380 fsname, idx, LUSTRE_LWP_NAME, tgt);
382 GOTO(cleanup, rc = 0);
386 OBD_FREE(fsname, MTI_NAME_MAXLEN);
390 EXPORT_SYMBOL(tgt_name2lwp_name);
392 static CFS_LIST_HEAD(lwp_register_list);
393 DEFINE_MUTEX(lwp_register_list_lock);
395 int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp,
396 register_lwp_cb cb_func, void *cb_data)
398 struct obd_device *lwp;
399 struct lwp_register_item *lri;
402 LASSERTF(strlen(lwpname) < MTI_NAME_MAXLEN, "lwpname is too long %s\n",
404 LASSERT(exp != NULL && *exp == NULL);
410 mutex_lock(&lwp_register_list_lock);
412 lwp = class_name2obd(lwpname);
413 if (lwp != NULL && lwp->obd_set_up == 1) {
414 struct obd_uuid *uuid;
418 mutex_unlock(&lwp_register_list_lock);
422 memcpy(uuid->uuid, lwpname, strlen(lwpname));
423 *exp = cfs_hash_lookup(lwp->obd_uuid_hash, uuid);
427 memcpy(lri->lri_name, lwpname, strlen(lwpname));
429 lri->lri_cb_func = cb_func;
430 lri->lri_cb_data = cb_data;
431 CFS_INIT_LIST_HEAD(&lri->lri_list);
432 cfs_list_add(&lri->lri_list, &lwp_register_list);
434 if (*exp != NULL && cb_func != NULL)
437 mutex_unlock(&lwp_register_list_lock);
440 EXPORT_SYMBOL(lustre_register_lwp_item);
442 void lustre_deregister_lwp_item(struct obd_export **exp)
444 struct lwp_register_item *lri, *tmp;
446 mutex_lock(&lwp_register_list_lock);
447 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
448 if (exp == lri->lri_exp) {
450 class_export_put(*exp);
451 cfs_list_del(&lri->lri_list);
456 mutex_unlock(&lwp_register_list_lock);
458 EXPORT_SYMBOL(lustre_deregister_lwp_item);
460 struct obd_export *lustre_find_lwp_by_index(const char *dev, __u32 idx)
462 struct lustre_mount_info *lmi;
463 struct lustre_sb_info *lsi;
464 struct obd_device *lwp;
465 struct obd_export *exp = NULL;
470 lmi = server_get_mount_2(dev);
474 lsi = s2lsi(lmi->lmi_sb);
475 rc = server_name2fsname(lsi->lsi_svname, fsname, NULL);
477 CERROR("%s: failed to get fsname: rc = %d\n",
478 lsi->lsi_svname, rc);
482 snprintf(lwp_name, sizeof(lwp_name), "%s-MDT%04x", fsname, idx);
483 spin_lock(&lsi->lsi_lwp_lock);
484 list_for_each_entry(lwp, &lsi->lsi_lwp_list, obd_lwp_list) {
485 char *ptr = strstr(lwp->obd_name, lwp_name);
488 exp = class_export_get(lwp->obd_lwp_export);
492 spin_unlock(&lsi->lsi_lwp_lock);
496 EXPORT_SYMBOL(lustre_find_lwp_by_index);
498 static void lustre_notify_lwp_list(struct obd_export *exp)
500 struct lwp_register_item *lri, *tmp;
501 LASSERT(exp != NULL);
503 mutex_lock(&lwp_register_list_lock);
504 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
505 if (strcmp(exp->exp_obd->obd_name, lri->lri_name))
507 if (*lri->lri_exp != NULL)
509 *lri->lri_exp = class_export_get(exp);
510 if (lri->lri_cb_func != NULL)
511 lri->lri_cb_func(lri->lri_cb_data);
513 mutex_unlock(&lwp_register_list_lock);
516 static int lustre_lwp_connect(struct obd_device *lwp)
519 struct lu_context session_ctx;
520 struct obd_export *exp;
521 struct obd_uuid *uuid = NULL;
522 struct obd_connect_data *data = NULL;
526 /* log has been fully processed, let clients connect */
527 rc = lu_env_init(&env, lwp->obd_lu_dev->ld_type->ldt_ctx_tags);
531 lu_context_init(&session_ctx, LCT_SERVER_SESSION);
532 session_ctx.lc_thread = NULL;
533 lu_context_enter(&session_ctx);
534 env.le_ses = &session_ctx;
538 GOTO(out, rc = -ENOMEM);
540 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
541 data->ocd_version = LUSTRE_VERSION_CODE;
542 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
543 OBD_CONNECT_AT | OBD_CONNECT_LRU_RESIZE |
544 OBD_CONNECT_FULL20 | OBD_CONNECT_LVB_TYPE |
545 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_LFSCK;
548 GOTO(out, rc = -ENOMEM);
550 if (strlen(lwp->obd_name) > sizeof(uuid->uuid)) {
551 CERROR("%s: Too long lwp name %s, max_size is %d\n",
552 lwp->obd_name, lwp->obd_name, (int)sizeof(uuid->uuid));
553 GOTO(out, rc = -EINVAL);
556 /* Use lwp name as the uuid, so we find the export by lwp name later */
557 memcpy(uuid->uuid, lwp->obd_name, strlen(lwp->obd_name));
558 rc = obd_connect(&env, &exp, lwp, uuid, data, NULL);
560 CERROR("%s: connect failed: rc = %d\n", lwp->obd_name, rc);
562 if (unlikely(lwp->obd_lwp_export != NULL))
563 class_export_put(lwp->obd_lwp_export);
564 lwp->obd_lwp_export = class_export_get(exp);
565 lustre_notify_lwp_list(exp);
577 lu_context_exit(&session_ctx);
578 lu_context_fini(&session_ctx);
584 * lwp is used by slaves (Non-MDT0 targets) to manage the connection to MDT0,
585 * or from the OSTx to MDTy.
587 static int lustre_lwp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi,
590 struct obd_device *obd;
591 char *lwpname = NULL;
592 char *lwpuuid = NULL;
596 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
599 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
603 OBD_ALLOC(lwpname, MTI_NAME_MAXLEN);
605 GOTO(out, rc = -ENOMEM);
607 rc = tgt_name2lwp_name(lsi->lsi_svname, lwpname, MTI_NAME_MAXLEN, idx);
609 CERROR("%s: failed to generate lwp name: rc = %d\n",
610 lsi->lsi_svname, rc);
614 OBD_ALLOC(lwpuuid, MTI_NAME_MAXLEN);
616 GOTO(out, rc = -ENOMEM);
618 sprintf(lwpuuid, "%s_UUID", lwpname);
619 rc = lustre_start_simple(lwpname, LUSTRE_LWP_NAME,
620 lwpuuid, lustre_cfg_string(lcfg, 1),
623 CERROR("%s: setup up failed: rc %d\n", lwpname, rc);
627 obd = class_name2obd(lwpname);
628 LASSERT(obd != NULL);
630 rc = lustre_lwp_connect(obd);
632 obd->u.cli.cl_max_mds_easize = MAX_MD_SIZE;
633 spin_lock(&lsi->lsi_lwp_lock);
634 list_add_tail(&obd->obd_lwp_list, &lsi->lsi_lwp_list);
635 spin_unlock(&lsi->lsi_lwp_lock);
637 CERROR("%s: connect failed: rc = %d\n", lwpname, rc);
644 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
646 OBD_FREE(lwpuuid, MTI_NAME_MAXLEN);
651 /* the caller is responsible for memory free */
652 static struct obd_device *lustre_find_lwp(struct lustre_sb_info *lsi,
653 char **lwpname, __u32 idx)
655 struct obd_device *lwp;
659 LASSERT(lwpname != NULL);
660 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
662 OBD_ALLOC(*lwpname, MTI_NAME_MAXLEN);
663 if (*lwpname == NULL)
664 RETURN(ERR_PTR(-ENOMEM));
666 rc = tgt_name2lwp_name(lsi->lsi_svname, *lwpname, MTI_NAME_MAXLEN, idx);
668 CERROR("%s: failed to generate lwp name: rc = %d\n",
669 lsi->lsi_svname, rc);
670 GOTO(out, rc = -EINVAL);
673 lwp = class_name2obd(*lwpname);
677 if (*lwpname != NULL) {
678 OBD_FREE(*lwpname, MTI_NAME_MAXLEN);
684 RETURN(lwp != NULL ? lwp : ERR_PTR(-ENOENT));
687 static int lustre_lwp_add_conn(struct lustre_cfg *cfg,
688 struct lustre_sb_info *lsi, __u32 idx)
690 struct lustre_cfg_bufs *bufs = NULL;
691 struct lustre_cfg *lcfg = NULL;
692 char *lwpname = NULL;
693 struct obd_device *lwp;
697 lwp = lustre_find_lwp(lsi, &lwpname, idx);
699 CERROR("%s: can't find lwp device.\n", lsi->lsi_svname);
700 GOTO(out, rc = PTR_ERR(lwp));
702 LASSERT(lwpname != NULL);
706 GOTO(out, rc = -ENOMEM);
708 lustre_cfg_bufs_reset(bufs, lwpname);
709 lustre_cfg_bufs_set_string(bufs, 1,
710 lustre_cfg_string(cfg, 1));
712 lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
714 rc = class_add_conn(lwp, lcfg);
716 CERROR("%s: can't add conn: rc = %d\n", lwpname, rc);
722 lustre_cfg_free(lcfg);
724 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
729 * Retrieve MDT nids from the client log, then start the lwp device.
730 * there are only two scenarios which would include mdt nid.
732 * marker 5 (flags=0x01, v2.1.54.0) lustre-MDTyyyy 'add mdc' xxx-
733 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
734 * attach 0:lustre-MDTyyyy-mdc 1:mdc 2:lustre-clilmv_UUID
735 * setup 0:lustre-MDTyyyy-mdc 1:lustre-MDTyyyy_UUID 2:192.168.122.162@tcp
736 * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
737 * add_conn 0:lustre-MDTyyyy-mdc 1:192.168.172.1@tcp
738 * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDTyyyy_UUID xxxx
739 * marker 5 (flags=0x02, v2.1.54.0) lustre-MDTyyyy 'add mdc' xxxx-
741 * marker 7 (flags=0x01, v2.1.54.0) lustre-MDTyyyy 'add failnid' xxxx-
742 * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
743 * add_conn 0:lustre-MDTyyyy-mdc 1:192.168.122.2@tcp
744 * marker 7 (flags=0x02, v2.1.54.0) lustre-MDTyyyy 'add failnid' xxxx-
746 static int client_lwp_config_process(const struct lu_env *env,
747 struct llog_handle *handle,
748 struct llog_rec_hdr *rec, void *data)
750 struct config_llog_instance *clli = data;
751 int cfg_len = rec->lrh_len;
752 char *cfg_buf = (char *) (rec + 1);
753 struct lustre_cfg *lcfg = NULL;
754 struct lustre_sb_info *lsi;
755 int rc = 0, swab = 0;
758 if (rec->lrh_type != OBD_CFG_REC) {
759 CERROR("Unknown llog record type %#x encountered\n",
764 LASSERT(clli->cfg_sb != NULL);
765 lsi = s2lsi(clli->cfg_sb);
767 lcfg = (struct lustre_cfg *)cfg_buf;
768 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
769 lustre_swab_lustre_cfg(lcfg);
773 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
777 switch (lcfg->lcfg_command) {
779 struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
781 lustre_swab_cfg_marker(marker, swab,
782 LUSTRE_CFG_BUFLEN(lcfg, 1));
783 if (marker->cm_flags & CM_SKIP ||
784 marker->cm_flags & CM_EXCLUDE)
787 if (!tgt_is_mdt(marker->cm_tgtname, &clli->cfg_lwp_idx))
790 if (IS_MDT(lsi) && clli->cfg_lwp_idx != 0)
793 if (!strncmp(marker->cm_comment, "add mdc", 7) ||
794 !strncmp(marker->cm_comment, "add failnid", 11)) {
795 if (marker->cm_flags & CM_START) {
796 clli->cfg_flags = CFG_F_MARKER;
797 /* This hack is to differentiate the
798 * ADD_UUID is come from "add mdc" record
799 * or from "add failnid" record. */
800 if (!strncmp(marker->cm_comment,
802 clli->cfg_flags |= CFG_F_SKIP;
803 } else if (marker->cm_flags & CM_END) {
809 case LCFG_ADD_UUID: {
810 if (clli->cfg_flags == CFG_F_MARKER) {
811 rc = lustre_lwp_setup(lcfg, lsi, clli->cfg_lwp_idx);
812 /* XXX: process only the first nid as
813 * we don't need another instance of lwp */
814 clli->cfg_flags |= CFG_F_SKIP;
815 } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
816 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
819 CERROR("%s: Fail to add uuid, rc:%d\n",
820 lsi->lsi_svname, rc);
824 case LCFG_ADD_CONN: {
825 char *devname = lustre_cfg_string(lcfg, 0);
829 if (!is_mdc_device(devname))
832 ptr = strrchr(devname, '-');
837 if (!tgt_is_mdt(devname, &idx)) {
843 rc = lustre_lwp_add_conn(lcfg, lsi, idx);
853 static int lustre_disconnect_lwp(struct super_block *sb)
855 struct lustre_sb_info *lsi = s2lsi(sb);
856 struct obd_device *lwp;
857 char *logname = NULL;
858 struct lustre_cfg_bufs *bufs = NULL;
859 struct config_llog_instance *cfg = NULL;
864 if (likely(lsi->lsi_lwp_started)) {
865 OBD_ALLOC(logname, MTI_NAME_MAXLEN);
869 rc = server_name2fsname(lsi->lsi_svname, logname, NULL);
871 CERROR("%s: failed to get fsname from svname: "
872 "rc = %d\n", lsi->lsi_svname, rc);
873 GOTO(out, rc = -EINVAL);
876 strcat(logname, "-client");
879 GOTO(out, rc = -ENOMEM);
882 cfg->cfg_instance = sb;
883 rc = lustre_end_log(sb, logname, cfg);
887 lsi->lsi_lwp_started = 0;
892 GOTO(out, rc = -ENOMEM);
894 list_for_each_entry(lwp, &lsi->lsi_lwp_list, obd_lwp_list) {
895 struct lustre_cfg *lcfg;
897 if (likely(lwp->obd_lwp_export != NULL)) {
898 class_export_put(lwp->obd_lwp_export);
899 lwp->obd_lwp_export = NULL;
902 lustre_cfg_bufs_reset(bufs, lwp->obd_name);
903 lustre_cfg_bufs_set_string(bufs, 1, NULL);
904 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
906 GOTO(out, rc = -ENOMEM);
908 /* Disconnect import first. NULL is passed for the '@env',
909 * since it will not be used. */
910 rc = lwp->obd_lu_dev->ld_ops->ldo_process_config(NULL,
911 lwp->obd_lu_dev, lcfg);
912 lustre_cfg_free(lcfg);
913 if (rc != 0 && rc != -ETIMEDOUT) {
914 CERROR("%s: fail to disconnect LWP: rc = %d\n",
928 OBD_FREE(logname, MTI_NAME_MAXLEN);
930 return rc1 != 0 ? rc1 : rc;
934 * Stop the lwp for an OST/MDT target.
936 static int lustre_stop_lwp(struct super_block *sb)
938 struct lustre_sb_info *lsi = s2lsi(sb);
939 struct obd_device *lwp;
944 while (!list_empty(&lsi->lsi_lwp_list)) {
945 lwp = list_entry(lsi->lsi_lwp_list.next, struct obd_device,
947 list_del_init(&lwp->obd_lwp_list);
949 rc = class_manual_cleanup(lwp);
951 CERROR("%s: fail to stop LWP: rc = %d\n",
957 RETURN(rc1 != 0 ? rc1 : rc);
961 * Start the lwp(fsname-MDTyyyy-lwp-{MDT,OST}xxxx) for a MDT/OST or MDT target.
963 static int lustre_start_lwp(struct super_block *sb)
965 struct lustre_sb_info *lsi = s2lsi(sb);
966 struct config_llog_instance *cfg = NULL;
971 if (unlikely(lsi->lsi_lwp_started))
974 OBD_ALLOC(logname, MTI_NAME_MAXLEN);
978 rc = server_name2fsname(lsi->lsi_svname, logname, NULL);
980 CERROR("%s: failed to get fsname from svname: rc = %d\n",
981 lsi->lsi_svname, rc);
982 GOTO(out, rc = -EINVAL);
985 strcat(logname, "-client");
988 GOTO(out, rc = -ENOMEM);
990 cfg->cfg_callback = client_lwp_config_process;
991 cfg->cfg_instance = sb;
992 rc = lustre_process_log(sb, logname, cfg);
994 lsi->lsi_lwp_started = 1;
999 OBD_FREE(logname, MTI_NAME_MAXLEN);
1006 DEFINE_MUTEX(server_start_lock);
1008 /* Stop MDS/OSS if nobody is using them */
1009 static int server_stop_servers(int lsiflags)
1011 struct obd_device *obd = NULL;
1012 struct obd_type *type = NULL;
1016 mutex_lock(&server_start_lock);
1018 /* Either an MDT or an OST or neither */
1019 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
1020 if (lsiflags & LDD_F_SV_TYPE_MDT) {
1021 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1023 type = class_search_type(LUSTRE_MDT_NAME);
1026 /* if this was an OST, and there are no more OST's, clean up the OSS */
1027 if (lsiflags & LDD_F_SV_TYPE_OST) {
1028 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1030 type = class_search_type(LUSTRE_OST_NAME);
1033 if (obd != NULL && (type == NULL || type->typ_refcnt == 0)) {
1035 /* obd_fail doesn't mean much on a server obd */
1036 rc = class_manual_cleanup(obd);
1039 mutex_unlock(&server_start_lock);
1044 int server_mti_print(const char *title, struct mgs_target_info *mti)
1046 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
1047 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
1048 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
1049 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
1050 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
1051 mti->mti_config_ver, mti->mti_flags);
1054 EXPORT_SYMBOL(server_mti_print);
1056 /* Generate data for registration */
1057 static int server_lsi2mti(struct lustre_sb_info *lsi,
1058 struct mgs_target_info *mti)
1060 lnet_process_id_t id;
1065 if (!IS_SERVER(lsi))
1068 if (strlcpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname))
1069 >= sizeof(mti->mti_svname))
1072 mti->mti_nid_count = 0;
1073 while (LNetGetId(i++, &id) != -ENOENT) {
1074 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1077 /* server use --servicenode param, only allow specified
1078 * nids be registered */
1079 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
1080 class_match_nid(lsi->lsi_lmd->lmd_params,
1081 PARAM_FAILNODE, id.nid) < 1)
1084 /* match specified network */
1085 if (!class_match_net(lsi->lsi_lmd->lmd_params,
1086 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1089 mti->mti_nids[mti->mti_nid_count] = id.nid;
1090 mti->mti_nid_count++;
1091 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1092 CWARN("Only using first %d nids for %s\n",
1093 mti->mti_nid_count, mti->mti_svname);
1098 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1099 mti->mti_config_ver = 0;
1101 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1105 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1108 /* Orion requires index to be set */
1109 LASSERT(!(rc & LDD_F_NEED_INDEX));
1110 /* keep only LDD flags */
1111 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1112 if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN))
1113 mti->mti_flags |= LDD_F_UPDATE;
1114 cplen = strlcpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1115 sizeof(mti->mti_params));
1116 if (cplen >= sizeof(mti->mti_params))
1121 /* Register an old or new target with the MGS. If needed MGS will construct
1122 startup logs and assign index */
1123 static int server_register_target(struct lustre_sb_info *lsi)
1125 struct obd_device *mgc = lsi->lsi_mgc;
1126 struct mgs_target_info *mti = NULL;
1133 if (!IS_SERVER(lsi))
1140 rc = server_lsi2mti(lsi, mti);
1144 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1145 mti->mti_svname, mti->mti_fsname,
1146 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1149 /* if write_conf is true, the registration must succeed */
1150 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1151 mti->mti_flags |= LDD_F_OPC_REG;
1153 /* Register the target */
1154 /* FIXME use mgc_process_config instead */
1155 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1156 sizeof(KEY_REGISTER_TARGET),
1157 KEY_REGISTER_TARGET,
1158 sizeof(*mti), mti, NULL);
1160 if (mti->mti_flags & LDD_F_ERROR) {
1161 LCONSOLE_ERROR_MSG(0x160,
1162 "%s: the MGS refuses to allow this server "
1163 "to start: rc = %d. Please see messages on "
1164 "the MGS.\n", lsi->lsi_svname, rc);
1165 } else if (writeconf) {
1166 LCONSOLE_ERROR_MSG(0x15f,
1167 "%s: cannot register this server with the MGS: "
1168 "rc = %d. Is the MGS running?\n",
1169 lsi->lsi_svname, rc);
1171 CERROR("%s: error registering with the MGS: rc = %d "
1172 "(not fatal)\n", lsi->lsi_svname, rc);
1173 /* reset the error code for non-fatal error. */
1186 * Notify the MGS that this target is ready.
1187 * Used by IR - if the MGS receives this message, it will notify clients.
1189 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1191 struct lustre_sb_info *lsi = s2lsi(sb);
1192 struct obd_device *mgc = lsi->lsi_mgc;
1193 struct mgs_target_info *mti = NULL;
1199 if (!(IS_SERVER(lsi)))
1205 rc = server_lsi2mti(lsi, mti);
1209 mti->mti_instance = obd->u.obt.obt_instance;
1210 mti->mti_flags |= LDD_F_OPC_READY;
1212 /* FIXME use mgc_process_config instead */
1213 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1214 sizeof(KEY_REGISTER_TARGET),
1215 KEY_REGISTER_TARGET,
1216 sizeof(*mti), mti, NULL);
1218 /* Imperative recovery: if the mgs informs us to use IR? */
1219 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1220 (mti->mti_flags & LDD_F_IR_CAPABLE))
1221 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1230 /** Start server targets: MDTs and OSTs
1232 static int server_start_targets(struct super_block *sb)
1234 struct obd_device *obd;
1235 struct lustre_sb_info *lsi = s2lsi(sb);
1236 struct config_llog_instance cfg;
1238 struct lu_device *dev;
1242 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1245 /* make sure the MDS is started */
1246 mutex_lock(&server_start_lock);
1247 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1249 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1251 LUSTRE_MDS_OBDNAME"_uuid",
1254 mutex_unlock(&server_start_lock);
1255 CERROR("failed to start MDS: %d\n", rc);
1259 mutex_unlock(&server_start_lock);
1262 /* If we're an OST, make sure the global OSS is running */
1264 /* make sure OSS is started */
1265 mutex_lock(&server_start_lock);
1266 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1268 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1270 LUSTRE_OSS_OBDNAME"_uuid",
1273 mutex_unlock(&server_start_lock);
1274 CERROR("failed to start OSS: %d\n", rc);
1278 mutex_unlock(&server_start_lock);
1281 /* Set the mgc fs to our server disk. This allows the MGC to
1282 * read and write configs locally, in case it can't talk to the MGS. */
1283 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1285 GOTO(out_stop_service, rc);
1287 /* Register with MGS */
1288 rc = server_register_target(lsi);
1292 /* Let the target look up the mount using the target's name
1293 (we can't pass the sb or mnt through class_process_config.) */
1294 rc = server_register_mount(lsi->lsi_svname, sb);
1298 /* Start targets using the llog named for the target */
1299 memset(&cfg, 0, sizeof(cfg));
1300 cfg.cfg_callback = class_config_llog_handler;
1301 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1303 CERROR("failed to start server %s: %d\n",
1304 lsi->lsi_svname, rc);
1305 /* Do NOT call server_deregister_mount() here. This makes it
1306 * impossible to find mount later in cleanup time and leaves
1307 * @lsi and othder stuff leaked. -umka */
1311 obd = class_name2obd(lsi->lsi_svname);
1313 CERROR("no server named %s was started\n", lsi->lsi_svname);
1314 GOTO(out_mgc, rc = -ENXIO);
1317 if (IS_OST(lsi) || IS_MDT(lsi)) {
1318 rc = lustre_start_lwp(sb);
1320 CERROR("%s: failed to start LWP: %d\n",
1321 lsi->lsi_svname, rc);
1326 server_notify_target(sb, obd);
1328 /* calculate recovery timeout, do it after lustre_process_log */
1329 server_calc_timeout(lsi, obd);
1331 /* log has been fully processed */
1332 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1334 /* log has been fully processed, let clients connect */
1335 dev = obd->obd_lu_dev;
1336 if (dev && dev->ld_ops->ldo_prepare) {
1337 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1339 struct lu_context session_ctx;
1341 lu_context_init(&session_ctx, LCT_SERVER_SESSION);
1342 session_ctx.lc_thread = NULL;
1343 lu_context_enter(&session_ctx);
1344 env.le_ses = &session_ctx;
1346 rc = dev->ld_ops->ldo_prepare(&env, NULL, dev);
1349 lu_context_exit(&session_ctx);
1350 lu_context_fini(&session_ctx);
1354 /* abort recovery only on the complete stack:
1355 * many devices can be involved */
1356 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1357 (OBP(obd, iocontrol))) {
1358 obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
1363 /* Release the mgc fs for others to use */
1364 server_mgc_clear_fs(lsi->lsi_mgc);
1368 server_stop_servers(lsi->lsi_flags);
1373 static int lsi_prepare(struct lustre_sb_info *lsi)
1380 LASSERT(lsi->lsi_lmd);
1382 /* The server name is given as a mount line option */
1383 if (lsi->lsi_lmd->lmd_profile == NULL) {
1384 LCONSOLE_ERROR("Can't determine server name\n");
1388 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
1389 RETURN(-ENAMETOOLONG);
1391 strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
1393 /* Determine osd type */
1394 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
1395 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
1396 sizeof(lsi->lsi_osd_type))
1397 RETURN(-ENAMETOOLONG);
1399 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
1401 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
1404 /* XXX: a temp. solution for components using ldiskfs
1405 * to be removed in one of the subsequent patches */
1406 if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs"))
1407 strcpy(lsi->lsi_fstype, "ldiskfs");
1409 strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
1411 /* Determine server type */
1412 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1414 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
1415 /* Assume we're a bare MGS */
1417 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
1419 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
1424 lsi->lsi_flags |= rc;
1426 /* Add mount line flags that used to be in ldd:
1427 * writeconf, mgs, anything else?
1429 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
1430 LDD_F_WRITECONF : 0;
1431 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
1433 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_UPDATE) ?
1435 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
1436 LDD_F_SV_TYPE_MGS : 0;
1437 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
1438 LDD_F_NO_PRIMNODE : 0;
1443 /*************** server mount ******************/
1445 /** Start the shutdown of servers at umount.
1447 static void server_put_super(struct super_block *sb)
1449 struct lustre_sb_info *lsi = s2lsi(sb);
1450 struct obd_device *obd;
1451 char *tmpname, *extraname = NULL;
1453 int lsiflags = lsi->lsi_flags;
1456 LASSERT(IS_SERVER(lsi));
1458 tmpname_sz = strlen(lsi->lsi_svname) + 1;
1459 OBD_ALLOC(tmpname, tmpname_sz);
1460 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
1461 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1462 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1463 snprintf(tmpname, tmpname_sz, "MGS");
1465 /* disconnect the lwp first to drain off the inflight request */
1466 if (IS_OST(lsi) || IS_MDT(lsi)) {
1469 rc = lustre_disconnect_lwp(sb);
1470 if (rc != 0 && rc != -ETIMEDOUT &&
1471 rc != -ENOTCONN && rc != -ESHUTDOWN)
1472 CWARN("%s: failed to disconnect lwp: rc= %d\n",
1476 /* Stop the target */
1477 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1478 (IS_MDT(lsi) || IS_OST(lsi))) {
1479 struct lustre_profile *lprof = NULL;
1481 /* tell the mgc to drop the config log */
1482 lustre_end_log(sb, lsi->lsi_svname, NULL);
1484 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1485 If there are any setup/cleanup errors, save the lov
1486 name for safety cleanup later. */
1487 lprof = class_get_profile(lsi->lsi_svname);
1488 if (lprof && lprof->lp_dt) {
1489 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1490 strcpy(extraname, lprof->lp_dt);
1493 obd = class_name2obd(lsi->lsi_svname);
1495 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1496 if (lsiflags & LSI_UMOUNT_FAILOVER)
1498 /* We can't seem to give an error return code
1499 * to .put_super, so we better make sure we clean up! */
1501 class_manual_cleanup(obd);
1503 CERROR("no obd %s\n", lsi->lsi_svname);
1504 server_deregister_mount(lsi->lsi_svname);
1508 /* If they wanted the mgs to stop separately from the mdt, they
1509 should have put it on a different device. */
1511 /* if MDS start with --nomgs, don't stop MGS then */
1512 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1513 server_stop_mgs(sb);
1516 if (IS_OST(lsi) || IS_MDT(lsi)) {
1517 if (lustre_stop_lwp(sb) < 0)
1518 CERROR("%s: failed to stop lwp!\n", tmpname);
1521 /* Clean the mgc and sb */
1522 lustre_common_put_super(sb);
1524 /* wait till all in-progress cleanups are done
1525 * specifically we're interested in ofd cleanup
1527 obd_zombie_barrier();
1529 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1530 until the target is really gone so that our type refcount check
1532 server_stop_servers(lsiflags);
1534 /* In case of startup or cleanup err, stop related obds */
1536 obd = class_name2obd(extraname);
1538 CWARN("Cleaning orphaned obd %s\n", extraname);
1540 class_manual_cleanup(obd);
1542 OBD_FREE(extraname, strlen(extraname) + 1);
1545 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1546 OBD_FREE(tmpname, tmpname_sz);
1550 /** Called only for 'umount -f'
1552 static void server_umount_begin(struct super_block *sb)
1554 struct lustre_sb_info *lsi = s2lsi(sb);
1557 CDEBUG(D_MOUNT, "umount -f\n");
1558 /* umount = failover
1560 no third way to do non-force, non-failover */
1561 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1565 static int server_statfs(struct dentry *dentry, struct kstatfs *buf)
1567 struct super_block *sb = dentry->d_sb;
1568 struct lustre_sb_info *lsi = s2lsi(sb);
1569 struct obd_statfs statfs;
1573 if (lsi->lsi_dt_dev) {
1574 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
1576 statfs_unpack(buf, &statfs);
1577 buf->f_type = sb->s_magic;
1583 buf->f_type = sb->s_magic;
1584 buf->f_bsize = sb->s_blocksize;
1590 buf->f_namelen = NAME_MAX;
1594 /** The operations we support directly on the superblock:
1595 * mount, umount, and df.
1597 static struct super_operations server_ops = {
1598 .put_super = server_put_super,
1599 .umount_begin = server_umount_begin, /* umount -f */
1600 .statfs = server_statfs,
1604 * Xattr support for Lustre servers
1606 static ssize_t lustre_getxattr(struct dentry *dentry, const char *name,
1607 void *buffer, size_t size)
1609 if (!selinux_is_enabled())
1614 static int lustre_setxattr(struct dentry *dentry, const char *name,
1615 const void *value, size_t size, int flags)
1620 static ssize_t lustre_listxattr(struct dentry *d_entry, char *name,
1626 const struct inode_operations server_inode_operations = {
1627 .setxattr = lustre_setxattr,
1628 .getxattr = lustre_getxattr,
1629 .listxattr = lustre_listxattr,
1632 #define log2(n) ffz(~(n))
1633 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1635 static int server_fill_super_common(struct super_block *sb)
1637 struct inode *root = 0;
1640 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1642 sb->s_blocksize = 4096;
1643 sb->s_blocksize_bits = log2(sb->s_blocksize);
1644 sb->s_magic = LUSTRE_SUPER_MAGIC;
1645 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1646 sb->s_flags |= MS_RDONLY;
1647 sb->s_op = &server_ops;
1649 root = new_inode(sb);
1651 CERROR("Can't make root inode\n");
1655 /* returns -EIO for every operation */
1656 /* make_bad_inode(root); -- badness - can't umount */
1657 /* apparently we need to be a directory for the mount to finish */
1658 root->i_mode = S_IFDIR;
1659 root->i_op = &server_inode_operations;
1660 sb->s_root = d_make_root(root);
1662 CERROR("%s: can't make root dentry\n", sb->s_id);
1669 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
1671 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1672 struct obd_device *obd;
1673 struct dt_device_param p;
1679 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
1680 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
1682 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
1683 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
1684 strcat(lsi->lsi_osd_uuid, "_UUID");
1685 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
1687 obd = class_name2obd(lsi->lsi_osd_obdname);
1689 rc = lustre_start_simple(lsi->lsi_osd_obdname,
1691 lsi->lsi_osd_uuid, lmd->lmd_dev,
1692 flagstr, lsi->lsi_lmd->lmd_opts,
1696 obd = class_name2obd(lsi->lsi_osd_obdname);
1700 rc = obd_connect(NULL, &lsi->lsi_osd_exp,
1701 obd, &obd->obd_uuid, NULL, NULL);
1704 class_manual_cleanup(obd);
1705 lsi->lsi_dt_dev = NULL;
1709 LASSERT(obd->obd_lu_dev);
1710 lu_device_get(obd->obd_lu_dev);
1711 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
1712 LASSERT(lsi->lsi_dt_dev);
1714 /* set disk context for llog usage */
1715 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1716 obd->obd_lvfs_ctxt.dt = lsi->lsi_dt_dev;
1718 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
1723 /** Fill in the superblock info for a Lustre server.
1724 * Mount the device with the correct options.
1725 * Read the on-disk config file.
1726 * Start the services.
1728 int server_fill_super(struct super_block *sb)
1730 struct lustre_sb_info *lsi = s2lsi(sb);
1734 rc = lsi_prepare(lsi);
1738 /* Start low level OSD */
1739 rc = osd_start(lsi, sb->s_flags);
1741 CERROR("Unable to start osd on %s: %d\n",
1742 lsi->lsi_lmd->lmd_dev, rc);
1747 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
1748 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
1750 if (class_name2obd(lsi->lsi_svname)) {
1751 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1752 "running. Double-mount may have compromised"
1753 " the disk journal.\n",
1759 /* Start MGS before MGC */
1760 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1761 rc = server_start_mgs(sb);
1766 /* Start MGC before servers */
1767 rc = lustre_start_mgc(sb);
1771 /* Set up all obd devices for service */
1772 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1773 (IS_OST(lsi) || IS_MDT(lsi))) {
1774 rc = server_start_targets(sb);
1776 CERROR("Unable to start targets: %d\n", rc);
1779 /* FIXME overmount client here, or can we just start a
1780 * client log and client_fill_super on this sb? We
1781 * need to make sure server_put_super gets called too
1782 * - ll_put_super calls lustre_common_put_super; check
1783 * there for LSI_SERVER flag, call s_p_s if so.
1785 * Probably should start client from new thread so we
1786 * can return. Client will not finish until all
1787 * servers are connected. Note - MGS-only server does
1788 * NOT get a client, since there is no lustre fs
1789 * associated - the MGS is for all lustre fs's */
1792 rc = server_fill_super_common(sb);
1798 /* We jump here in case of failure while starting targets or MGS.
1799 * In this case we can't just put @mnt and have to do real cleanup
1800 * with stoping targets, etc. */
1801 server_put_super(sb);
1806 * Calculate timeout value for a target.
1808 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1810 struct lustre_mount_data *lmd;
1814 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
1815 int min = OBD_RECOVERY_TIME_MIN;
1817 LASSERT(IS_SERVER(lsi));
1821 soft = lmd->lmd_recovery_time_soft;
1822 hard = lmd->lmd_recovery_time_hard;
1823 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1824 obd->obd_no_ir = !has_ir;
1828 soft = OBD_RECOVERY_TIME_SOFT;
1830 hard = OBD_RECOVERY_TIME_HARD;
1832 /* target may have ir_factor configured. */
1833 factor = OBD_IR_FACTOR_DEFAULT;
1834 if (obd->obd_recovery_ir_factor)
1835 factor = obd->obd_recovery_ir_factor;
1838 int new_soft = soft;
1839 int new_hard = hard;
1841 /* adjust timeout value by imperative recovery */
1843 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1844 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1846 /* make sure the timeout is not too short */
1847 new_soft = max(min, new_soft);
1848 new_hard = max(new_soft, new_hard);
1850 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1851 "window shrunk from %d-%d down to %d-%d\n",
1852 obd->obd_name, soft, hard, new_soft, new_hard);
1859 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1860 obd->obd_recovery_time_hard = hard;
1861 obd->obd_recovery_ir_factor = factor;
1863 EXPORT_SYMBOL(server_calc_timeout);