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_ioctl.h>
54 #include <lustre_log.h>
55 #include <lustre_disk.h>
56 #include <lustre_param.h>
57 #ifdef HAVE_KERNEL_LOCKED
58 #include <linux/smp_lock.h>
60 #ifdef HAVE_SELINUX_IS_ENABLED
61 #include <linux/selinux.h>
64 /*********** mount lookup *********/
66 DEFINE_MUTEX(lustre_mount_info_lock);
67 static CFS_LIST_HEAD(server_mount_info_list);
69 static struct lustre_mount_info *server_find_mount(const char *name)
72 struct lustre_mount_info *lmi;
75 cfs_list_for_each(tmp, &server_mount_info_list) {
76 lmi = cfs_list_entry(tmp, struct lustre_mount_info,
78 if (strcmp(name, lmi->lmi_name) == 0)
84 /* we must register an obd for a mount before we call the setup routine.
85 *_setup will call lustre_get_mount to get the mnt struct
86 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 cfs_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 cfs_list_del(&lmi->lmi_list_chain);
143 OBD_FREE(lmi, sizeof(*lmi));
144 mutex_unlock(&lustre_mount_info_lock);
149 /* obd's look up a registered mount using their obdname. This is just
150 for initial obd setup to find the mount struct. It should not be
151 called every time you want to mntget. */
152 struct lustre_mount_info *server_get_mount(const char *name)
154 struct lustre_mount_info *lmi;
155 struct lustre_sb_info *lsi;
158 mutex_lock(&lustre_mount_info_lock);
159 lmi = server_find_mount(name);
160 mutex_unlock(&lustre_mount_info_lock);
162 CERROR("Can't find mount for %s\n", name);
165 lsi = s2lsi(lmi->lmi_sb);
167 atomic_inc(&lsi->lsi_mounts);
169 CDEBUG(D_MOUNT, "get mount %p from %s, refs=%d\n", lmi->lmi_sb,
170 name, atomic_read(&lsi->lsi_mounts));
174 EXPORT_SYMBOL(server_get_mount);
177 * server_put_mount: to be called from obd_cleanup methods
179 * @dereg_mnt: 0 or 1 depending on whether the mount is to be deregistered or
182 * The caller decides whether server_deregister_mount() needs to be called or
183 * not. Calling of server_deregister_mount() does not depend on refcounting on
184 * lsi because we could have say the mgs and mds on the same node and we
185 * unmount the mds, then the ref on the lsi would still be non-zero but we
186 * would still want to deregister the mds mount.
188 int server_put_mount(const char *name, bool dereg_mnt)
190 struct lustre_mount_info *lmi;
191 struct lustre_sb_info *lsi;
194 mutex_lock(&lustre_mount_info_lock);
195 lmi = server_find_mount(name);
196 mutex_unlock(&lustre_mount_info_lock);
198 CERROR("Can't find mount for %s\n", name);
201 lsi = s2lsi(lmi->lmi_sb);
203 CDEBUG(D_MOUNT, "put mount %p from %s, refs=%d\n",
204 lmi->lmi_sb, name, atomic_read(&lsi->lsi_mounts));
206 if (lustre_put_lsi(lmi->lmi_sb))
207 CDEBUG(D_MOUNT, "Last put of mount %p from %s\n",
211 /* this obd should never need the mount again */
212 server_deregister_mount(name);
216 EXPORT_SYMBOL(server_put_mount);
218 /* Set up a MGS to serve startup logs */
219 static int server_start_mgs(struct super_block *sb)
221 struct lustre_sb_info *lsi = s2lsi(sb);
222 struct lustre_mount_info *lmi;
226 /* It is impossible to have more than 1 MGS per node, since
227 MGC wouldn't know which to connect to */
228 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
230 lsi = s2lsi(lmi->lmi_sb);
231 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
236 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
238 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb);
241 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
242 LUSTRE_MGS_OBDNAME, 0, 0,
243 lsi->lsi_osd_obdname, 0);
244 /* server_deregister_mount() is not called previously, for lsi
245 * and other stuff can't be freed cleanly when mgs calls
246 * server_put_mount() in error handling case (see b=17758),
247 * this problem is caused by a bug in mgs_init0, which forgot
248 * calling server_put_mount in error case. */
251 server_deregister_mount(LUSTRE_MGS_OBDNAME);
255 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
256 "Is the 'mgs' module loaded?\n",
257 LUSTRE_MGS_OBDNAME, rc);
261 static int server_stop_mgs(struct super_block *sb)
263 struct obd_device *obd;
267 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
269 /* There better be only one MGS */
270 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
272 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
276 /* The MGS should always stop when we say so */
278 rc = class_manual_cleanup(obd);
282 /* Since there's only one mgc per node, we have to change it's fs to get
283 access to the right disk. */
284 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
286 struct lustre_sb_info *lsi = s2lsi(sb);
290 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
292 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
293 rc = obd_set_info_async(NULL, mgc->obd_self_export,
294 sizeof(KEY_SET_FS), KEY_SET_FS,
295 sizeof(*sb), sb, NULL);
297 CERROR("can't set_fs %d\n", rc);
302 static int server_mgc_clear_fs(struct obd_device *mgc)
307 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
309 rc = obd_set_info_async(NULL, mgc->obd_self_export,
310 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
315 static inline bool is_mdc_device(const char *devname)
319 ptr = strrchr(devname, '-');
320 return ptr != NULL && strcmp(ptr, "-mdc") == 0;
323 static inline bool tgt_is_mdt(const char *tgtname, __u32 *idx)
327 type = server_name2index(tgtname, idx, NULL);
329 return type == LDD_F_SV_TYPE_MDT;
333 * Convert OST/MDT name(fsname-{MDT,OST}xxxx) to a lwp name with the @idx:yyyy
334 * (fsname-MDTyyyy-lwp-{MDT,OST}xxxx)
336 int tgt_name2lwp_name(const char *tgt_name, char *lwp_name, int len, __u32 idx)
343 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
347 rc = server_name2fsname(tgt_name, fsname, &tgt);
349 CERROR("%s: failed to get fsname from tgt_name: rc = %d\n",
354 if (*tgt != '-' && *tgt != ':') {
355 CERROR("%s: invalid tgt_name name!\n", tgt_name);
356 GOTO(cleanup, rc = -EINVAL);
360 if (strncmp(tgt, "OST", 3) != 0 && strncmp(tgt, "MDT", 3) != 0) {
361 CERROR("%s is not an OST or MDT target!\n", tgt_name);
362 GOTO(cleanup, rc = -EINVAL);
364 snprintf(lwp_name, len, "%s-MDT%04x-%s-%s",
365 fsname, idx, LUSTRE_LWP_NAME, tgt);
367 GOTO(cleanup, rc = 0);
371 OBD_FREE(fsname, MTI_NAME_MAXLEN);
375 EXPORT_SYMBOL(tgt_name2lwp_name);
377 static CFS_LIST_HEAD(lwp_register_list);
378 DEFINE_MUTEX(lwp_register_list_lock);
380 int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp,
381 register_lwp_cb cb_func, void *cb_data)
383 struct obd_device *lwp;
384 struct lwp_register_item *lri;
387 LASSERTF(strlen(lwpname) < MTI_NAME_MAXLEN, "lwpname is too long %s\n",
389 LASSERT(exp != NULL && *exp == NULL);
395 mutex_lock(&lwp_register_list_lock);
397 lwp = class_name2obd(lwpname);
398 if (lwp != NULL && lwp->obd_set_up == 1) {
399 struct obd_uuid *uuid;
403 mutex_unlock(&lwp_register_list_lock);
407 memcpy(uuid->uuid, lwpname, strlen(lwpname));
408 *exp = cfs_hash_lookup(lwp->obd_uuid_hash, uuid);
412 memcpy(lri->lri_name, lwpname, strlen(lwpname));
414 lri->lri_cb_func = cb_func;
415 lri->lri_cb_data = cb_data;
416 CFS_INIT_LIST_HEAD(&lri->lri_list);
417 cfs_list_add(&lri->lri_list, &lwp_register_list);
419 if (*exp != NULL && cb_func != NULL)
422 mutex_unlock(&lwp_register_list_lock);
425 EXPORT_SYMBOL(lustre_register_lwp_item);
427 void lustre_deregister_lwp_item(struct obd_export **exp)
429 struct lwp_register_item *lri, *tmp;
431 mutex_lock(&lwp_register_list_lock);
432 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
433 if (exp == lri->lri_exp) {
435 class_export_put(*exp);
436 cfs_list_del(&lri->lri_list);
441 mutex_unlock(&lwp_register_list_lock);
443 EXPORT_SYMBOL(lustre_deregister_lwp_item);
445 struct obd_export *lustre_find_lwp_by_index(const char *dev, __u32 idx)
447 struct lustre_mount_info *lmi;
448 struct lustre_sb_info *lsi;
449 struct obd_device *lwp;
450 struct obd_export *exp = NULL;
455 lmi = server_get_mount(dev);
459 lsi = s2lsi(lmi->lmi_sb);
460 rc = server_name2fsname(lsi->lsi_svname, fsname, NULL);
462 CERROR("%s: failed to get fsname: rc = %d\n",
463 lsi->lsi_svname, rc);
467 snprintf(lwp_name, sizeof(lwp_name), "%s-MDT%04x", fsname, idx);
468 spin_lock(&lsi->lsi_lwp_lock);
469 list_for_each_entry(lwp, &lsi->lsi_lwp_list, obd_lwp_list) {
470 char *ptr = strstr(lwp->obd_name, lwp_name);
473 exp = class_export_get(lwp->obd_lwp_export);
477 spin_unlock(&lsi->lsi_lwp_lock);
480 server_put_mount(dev, false);
484 EXPORT_SYMBOL(lustre_find_lwp_by_index);
486 static void lustre_notify_lwp_list(struct obd_export *exp)
488 struct lwp_register_item *lri, *tmp;
489 LASSERT(exp != NULL);
491 mutex_lock(&lwp_register_list_lock);
492 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
493 if (strcmp(exp->exp_obd->obd_name, lri->lri_name))
495 if (*lri->lri_exp != NULL)
497 *lri->lri_exp = class_export_get(exp);
498 if (lri->lri_cb_func != NULL)
499 lri->lri_cb_func(lri->lri_cb_data);
501 mutex_unlock(&lwp_register_list_lock);
504 static int lustre_lwp_connect(struct obd_device *lwp)
507 struct lu_context session_ctx;
508 struct obd_export *exp;
509 struct obd_uuid *uuid = NULL;
510 struct obd_connect_data *data = NULL;
514 /* log has been fully processed, let clients connect */
515 rc = lu_env_init(&env, lwp->obd_lu_dev->ld_type->ldt_ctx_tags);
519 lu_context_init(&session_ctx, LCT_SERVER_SESSION);
520 session_ctx.lc_thread = NULL;
521 lu_context_enter(&session_ctx);
522 env.le_ses = &session_ctx;
526 GOTO(out, rc = -ENOMEM);
528 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
529 data->ocd_version = LUSTRE_VERSION_CODE;
530 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
531 OBD_CONNECT_AT | OBD_CONNECT_LRU_RESIZE |
532 OBD_CONNECT_FULL20 | OBD_CONNECT_LVB_TYPE |
533 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_LFSCK;
536 GOTO(out, rc = -ENOMEM);
538 if (strlen(lwp->obd_name) > sizeof(uuid->uuid)) {
539 CERROR("%s: Too long lwp name %s, max_size is %d\n",
540 lwp->obd_name, lwp->obd_name, (int)sizeof(uuid->uuid));
541 GOTO(out, rc = -EINVAL);
544 /* Use lwp name as the uuid, so we find the export by lwp name later */
545 memcpy(uuid->uuid, lwp->obd_name, strlen(lwp->obd_name));
546 rc = obd_connect(&env, &exp, lwp, uuid, data, NULL);
548 CERROR("%s: connect failed: rc = %d\n", lwp->obd_name, rc);
550 if (unlikely(lwp->obd_lwp_export != NULL))
551 class_export_put(lwp->obd_lwp_export);
552 lwp->obd_lwp_export = class_export_get(exp);
553 lustre_notify_lwp_list(exp);
565 lu_context_exit(&session_ctx);
566 lu_context_fini(&session_ctx);
572 * lwp is used by slaves (Non-MDT0 targets) to manage the connection to MDT0,
573 * or from the OSTx to MDTy.
575 static int lustre_lwp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi,
578 struct obd_device *obd;
579 char *lwpname = NULL;
580 char *lwpuuid = NULL;
584 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
587 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
591 OBD_ALLOC(lwpname, MTI_NAME_MAXLEN);
593 GOTO(out, rc = -ENOMEM);
595 rc = tgt_name2lwp_name(lsi->lsi_svname, lwpname, MTI_NAME_MAXLEN, idx);
597 CERROR("%s: failed to generate lwp name: rc = %d\n",
598 lsi->lsi_svname, rc);
602 OBD_ALLOC(lwpuuid, MTI_NAME_MAXLEN);
604 GOTO(out, rc = -ENOMEM);
606 sprintf(lwpuuid, "%s_UUID", lwpname);
607 rc = lustre_start_simple(lwpname, LUSTRE_LWP_NAME,
608 lwpuuid, lustre_cfg_string(lcfg, 1),
611 CERROR("%s: setup up failed: rc %d\n", lwpname, rc);
615 obd = class_name2obd(lwpname);
616 LASSERT(obd != NULL);
618 rc = lustre_lwp_connect(obd);
620 obd->u.cli.cl_max_mds_easize = MAX_MD_SIZE;
621 spin_lock(&lsi->lsi_lwp_lock);
622 list_add_tail(&obd->obd_lwp_list, &lsi->lsi_lwp_list);
623 spin_unlock(&lsi->lsi_lwp_lock);
625 CERROR("%s: connect failed: rc = %d\n", lwpname, rc);
632 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
634 OBD_FREE(lwpuuid, MTI_NAME_MAXLEN);
639 /* the caller is responsible for memory free */
640 static struct obd_device *lustre_find_lwp(struct lustre_sb_info *lsi,
641 char **lwpname, __u32 idx)
643 struct obd_device *lwp;
647 LASSERT(lwpname != NULL);
648 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
650 OBD_ALLOC(*lwpname, MTI_NAME_MAXLEN);
651 if (*lwpname == NULL)
652 RETURN(ERR_PTR(-ENOMEM));
654 rc = tgt_name2lwp_name(lsi->lsi_svname, *lwpname, MTI_NAME_MAXLEN, idx);
656 CERROR("%s: failed to generate lwp name: rc = %d\n",
657 lsi->lsi_svname, rc);
658 GOTO(out, rc = -EINVAL);
661 lwp = class_name2obd(*lwpname);
665 if (*lwpname != NULL) {
666 OBD_FREE(*lwpname, MTI_NAME_MAXLEN);
672 RETURN(lwp != NULL ? lwp : ERR_PTR(-ENOENT));
675 static int lustre_lwp_add_conn(struct lustre_cfg *cfg,
676 struct lustre_sb_info *lsi, __u32 idx)
678 struct lustre_cfg_bufs *bufs = NULL;
679 struct lustre_cfg *lcfg = NULL;
680 char *lwpname = NULL;
681 struct obd_device *lwp;
685 lwp = lustre_find_lwp(lsi, &lwpname, idx);
687 CERROR("%s: can't find lwp device.\n", lsi->lsi_svname);
688 GOTO(out, rc = PTR_ERR(lwp));
690 LASSERT(lwpname != NULL);
694 GOTO(out, rc = -ENOMEM);
696 lustre_cfg_bufs_reset(bufs, lwpname);
697 lustre_cfg_bufs_set_string(bufs, 1,
698 lustre_cfg_string(cfg, 1));
700 lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
702 GOTO(out_cfg, rc = -ENOMEM);
703 rc = class_add_conn(lwp, lcfg);
705 CERROR("%s: can't add conn: rc = %d\n", lwpname, rc);
708 lustre_cfg_free(lcfg);
714 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
719 * Retrieve MDT nids from the client log, then start the lwp device.
720 * there are only two scenarios which would include mdt nid.
722 * marker 5 (flags=0x01, v2.1.54.0) lustre-MDTyyyy 'add mdc' xxx-
723 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
724 * attach 0:lustre-MDTyyyy-mdc 1:mdc 2:lustre-clilmv_UUID
725 * setup 0:lustre-MDTyyyy-mdc 1:lustre-MDTyyyy_UUID 2:192.168.122.162@tcp
726 * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
727 * add_conn 0:lustre-MDTyyyy-mdc 1:192.168.172.1@tcp
728 * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDTyyyy_UUID xxxx
729 * marker 5 (flags=0x02, v2.1.54.0) lustre-MDTyyyy 'add mdc' xxxx-
731 * marker 7 (flags=0x01, v2.1.54.0) lustre-MDTyyyy 'add failnid' xxxx-
732 * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
733 * add_conn 0:lustre-MDTyyyy-mdc 1:192.168.122.2@tcp
734 * marker 7 (flags=0x02, v2.1.54.0) lustre-MDTyyyy 'add failnid' xxxx-
736 static int client_lwp_config_process(const struct lu_env *env,
737 struct llog_handle *handle,
738 struct llog_rec_hdr *rec, void *data)
740 struct config_llog_instance *clli = data;
741 int cfg_len = rec->lrh_len;
742 char *cfg_buf = (char *) (rec + 1);
743 struct lustre_cfg *lcfg = NULL;
744 struct lustre_sb_info *lsi;
745 int rc = 0, swab = 0;
748 if (rec->lrh_type != OBD_CFG_REC) {
749 CERROR("Unknown llog record type %#x encountered\n",
754 LASSERT(clli->cfg_sb != NULL);
755 lsi = s2lsi(clli->cfg_sb);
757 lcfg = (struct lustre_cfg *)cfg_buf;
758 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
759 lustre_swab_lustre_cfg(lcfg);
763 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
767 switch (lcfg->lcfg_command) {
769 struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
771 lustre_swab_cfg_marker(marker, swab,
772 LUSTRE_CFG_BUFLEN(lcfg, 1));
773 if (marker->cm_flags & CM_SKIP ||
774 marker->cm_flags & CM_EXCLUDE)
777 if (!tgt_is_mdt(marker->cm_tgtname, &clli->cfg_lwp_idx))
780 if (IS_MDT(lsi) && clli->cfg_lwp_idx != 0)
783 if (!strncmp(marker->cm_comment, "add mdc", 7) ||
784 !strncmp(marker->cm_comment, "add failnid", 11)) {
785 if (marker->cm_flags & CM_START) {
786 clli->cfg_flags = CFG_F_MARKER;
787 /* This hack is to differentiate the
788 * ADD_UUID is come from "add mdc" record
789 * or from "add failnid" record. */
790 if (!strncmp(marker->cm_comment,
792 clli->cfg_flags |= CFG_F_SKIP;
793 } else if (marker->cm_flags & CM_END) {
799 case LCFG_ADD_UUID: {
800 if (clli->cfg_flags == CFG_F_MARKER) {
801 rc = lustre_lwp_setup(lcfg, lsi, clli->cfg_lwp_idx);
802 /* XXX: process only the first nid as
803 * we don't need another instance of lwp */
804 clli->cfg_flags |= CFG_F_SKIP;
805 } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
806 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
809 CERROR("%s: Fail to add uuid, rc:%d\n",
810 lsi->lsi_svname, rc);
814 case LCFG_ADD_CONN: {
815 char *devname = lustre_cfg_string(lcfg, 0);
819 if (!is_mdc_device(devname))
822 ptr = strrchr(devname, '-');
827 if (!tgt_is_mdt(devname, &idx)) {
833 rc = lustre_lwp_add_conn(lcfg, lsi, idx);
843 static int lustre_disconnect_lwp(struct super_block *sb)
845 struct lustre_sb_info *lsi = s2lsi(sb);
846 struct obd_device *lwp;
847 char *logname = NULL;
848 struct lustre_cfg_bufs *bufs = NULL;
849 struct config_llog_instance *cfg = NULL;
854 if (likely(lsi->lsi_lwp_started)) {
855 OBD_ALLOC(logname, MTI_NAME_MAXLEN);
859 rc = server_name2fsname(lsi->lsi_svname, logname, NULL);
861 CERROR("%s: failed to get fsname from svname: "
862 "rc = %d\n", lsi->lsi_svname, rc);
863 GOTO(out, rc = -EINVAL);
866 strcat(logname, "-client");
869 GOTO(out, rc = -ENOMEM);
872 cfg->cfg_instance = sb;
873 rc = lustre_end_log(sb, logname, cfg);
877 lsi->lsi_lwp_started = 0;
882 GOTO(out, rc = -ENOMEM);
884 list_for_each_entry(lwp, &lsi->lsi_lwp_list, obd_lwp_list) {
885 struct lustre_cfg *lcfg;
887 if (likely(lwp->obd_lwp_export != NULL)) {
888 class_export_put(lwp->obd_lwp_export);
889 lwp->obd_lwp_export = NULL;
892 lustre_cfg_bufs_reset(bufs, lwp->obd_name);
893 lustre_cfg_bufs_set_string(bufs, 1, NULL);
894 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
896 GOTO(out, rc = -ENOMEM);
898 /* Disconnect import first. NULL is passed for the '@env',
899 * since it will not be used. */
900 rc = lwp->obd_lu_dev->ld_ops->ldo_process_config(NULL,
901 lwp->obd_lu_dev, lcfg);
902 lustre_cfg_free(lcfg);
903 if (rc != 0 && rc != -ETIMEDOUT) {
904 CERROR("%s: fail to disconnect LWP: rc = %d\n",
918 OBD_FREE(logname, MTI_NAME_MAXLEN);
920 return rc1 != 0 ? rc1 : rc;
924 * Stop the lwp for an OST/MDT target.
926 static int lustre_stop_lwp(struct super_block *sb)
928 struct lustre_sb_info *lsi = s2lsi(sb);
929 struct obd_device *lwp;
934 while (!list_empty(&lsi->lsi_lwp_list)) {
935 lwp = list_entry(lsi->lsi_lwp_list.next, struct obd_device,
937 list_del_init(&lwp->obd_lwp_list);
939 rc = class_manual_cleanup(lwp);
941 CERROR("%s: fail to stop LWP: rc = %d\n",
947 RETURN(rc1 != 0 ? rc1 : rc);
951 * Start the lwp(fsname-MDTyyyy-lwp-{MDT,OST}xxxx) for a MDT/OST or MDT target.
953 static int lustre_start_lwp(struct super_block *sb)
955 struct lustre_sb_info *lsi = s2lsi(sb);
956 struct config_llog_instance *cfg = NULL;
961 if (unlikely(lsi->lsi_lwp_started))
964 OBD_ALLOC(logname, MTI_NAME_MAXLEN);
968 rc = server_name2fsname(lsi->lsi_svname, logname, NULL);
970 CERROR("%s: failed to get fsname from svname: rc = %d\n",
971 lsi->lsi_svname, rc);
972 GOTO(out, rc = -EINVAL);
975 strcat(logname, "-client");
978 GOTO(out, rc = -ENOMEM);
980 cfg->cfg_callback = client_lwp_config_process;
981 cfg->cfg_instance = sb;
982 rc = lustre_process_log(sb, logname, cfg);
984 lsi->lsi_lwp_started = 1;
989 OBD_FREE(logname, MTI_NAME_MAXLEN);
996 DEFINE_MUTEX(server_start_lock);
998 /* Stop MDS/OSS if nobody is using them */
999 static int server_stop_servers(int lsiflags)
1001 struct obd_device *obd = NULL;
1002 struct obd_type *type = NULL;
1006 mutex_lock(&server_start_lock);
1008 /* Either an MDT or an OST or neither */
1009 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
1010 if (lsiflags & LDD_F_SV_TYPE_MDT) {
1011 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1013 type = class_search_type(LUSTRE_MDT_NAME);
1016 /* if this was an OST, and there are no more OST's, clean up the OSS */
1017 if (lsiflags & LDD_F_SV_TYPE_OST) {
1018 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1020 type = class_search_type(LUSTRE_OST_NAME);
1023 if (obd != NULL && (type == NULL || type->typ_refcnt == 0)) {
1025 /* obd_fail doesn't mean much on a server obd */
1026 rc = class_manual_cleanup(obd);
1029 mutex_unlock(&server_start_lock);
1034 int server_mti_print(const char *title, struct mgs_target_info *mti)
1036 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
1037 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
1038 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
1039 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
1040 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
1041 mti->mti_config_ver, mti->mti_flags);
1044 EXPORT_SYMBOL(server_mti_print);
1046 /* Generate data for registration */
1047 static int server_lsi2mti(struct lustre_sb_info *lsi,
1048 struct mgs_target_info *mti)
1050 lnet_process_id_t id;
1055 if (!IS_SERVER(lsi))
1058 if (strlcpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname))
1059 >= sizeof(mti->mti_svname))
1062 mti->mti_nid_count = 0;
1063 while (LNetGetId(i++, &id) != -ENOENT) {
1064 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1067 /* server use --servicenode param, only allow specified
1068 * nids be registered */
1069 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
1070 class_match_nid(lsi->lsi_lmd->lmd_params,
1071 PARAM_FAILNODE, id.nid) < 1)
1074 /* match specified network */
1075 if (!class_match_net(lsi->lsi_lmd->lmd_params,
1076 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1079 mti->mti_nids[mti->mti_nid_count] = id.nid;
1080 mti->mti_nid_count++;
1081 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1082 CWARN("Only using first %d nids for %s\n",
1083 mti->mti_nid_count, mti->mti_svname);
1088 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1089 mti->mti_config_ver = 0;
1091 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1095 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1098 /* Orion requires index to be set */
1099 LASSERT(!(rc & LDD_F_NEED_INDEX));
1100 /* keep only LDD flags */
1101 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1102 if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN))
1103 mti->mti_flags |= LDD_F_UPDATE;
1104 cplen = strlcpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1105 sizeof(mti->mti_params));
1106 if (cplen >= sizeof(mti->mti_params))
1111 /* Register an old or new target with the MGS. If needed MGS will construct
1112 startup logs and assign index */
1113 static int server_register_target(struct lustre_sb_info *lsi)
1115 struct obd_device *mgc = lsi->lsi_mgc;
1116 struct mgs_target_info *mti = NULL;
1123 if (!IS_SERVER(lsi))
1130 rc = server_lsi2mti(lsi, mti);
1134 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1135 mti->mti_svname, mti->mti_fsname,
1136 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1139 /* if write_conf is true, the registration must succeed */
1140 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1141 mti->mti_flags |= LDD_F_OPC_REG;
1143 /* Register the target */
1144 /* FIXME use mgc_process_config instead */
1145 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1146 sizeof(KEY_REGISTER_TARGET),
1147 KEY_REGISTER_TARGET,
1148 sizeof(*mti), mti, NULL);
1150 if (mti->mti_flags & LDD_F_ERROR) {
1151 LCONSOLE_ERROR_MSG(0x160,
1152 "%s: the MGS refuses to allow this server "
1153 "to start: rc = %d. Please see messages on "
1154 "the MGS.\n", lsi->lsi_svname, rc);
1155 } else if (writeconf) {
1156 LCONSOLE_ERROR_MSG(0x15f,
1157 "%s: cannot register this server with the MGS: "
1158 "rc = %d. Is the MGS running?\n",
1159 lsi->lsi_svname, rc);
1161 CERROR("%s: error registering with the MGS: rc = %d "
1162 "(not fatal)\n", lsi->lsi_svname, rc);
1163 /* reset the error code for non-fatal error. */
1176 * Notify the MGS that this target is ready.
1177 * Used by IR - if the MGS receives this message, it will notify clients.
1179 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1181 struct lustre_sb_info *lsi = s2lsi(sb);
1182 struct obd_device *mgc = lsi->lsi_mgc;
1183 struct mgs_target_info *mti = NULL;
1189 if (!(IS_SERVER(lsi)))
1195 rc = server_lsi2mti(lsi, mti);
1199 mti->mti_instance = obd->u.obt.obt_instance;
1200 mti->mti_flags |= LDD_F_OPC_READY;
1202 /* FIXME use mgc_process_config instead */
1203 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1204 sizeof(KEY_REGISTER_TARGET),
1205 KEY_REGISTER_TARGET,
1206 sizeof(*mti), mti, NULL);
1208 /* Imperative recovery: if the mgs informs us to use IR? */
1209 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1210 (mti->mti_flags & LDD_F_IR_CAPABLE))
1211 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1220 /** Start server targets: MDTs and OSTs
1222 static int server_start_targets(struct super_block *sb)
1224 struct obd_device *obd;
1225 struct lustre_sb_info *lsi = s2lsi(sb);
1226 struct config_llog_instance cfg;
1228 struct lu_device *dev;
1232 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1235 /* make sure the MDS is started */
1236 mutex_lock(&server_start_lock);
1237 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1239 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1241 LUSTRE_MDS_OBDNAME"_uuid",
1244 mutex_unlock(&server_start_lock);
1245 CERROR("failed to start MDS: %d\n", rc);
1249 mutex_unlock(&server_start_lock);
1252 /* If we're an OST, make sure the global OSS is running */
1254 /* make sure OSS is started */
1255 mutex_lock(&server_start_lock);
1256 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1258 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1260 LUSTRE_OSS_OBDNAME"_uuid",
1263 mutex_unlock(&server_start_lock);
1264 CERROR("failed to start OSS: %d\n", rc);
1268 mutex_unlock(&server_start_lock);
1271 /* Set the mgc fs to our server disk. This allows the MGC to
1272 * read and write configs locally, in case it can't talk to the MGS. */
1273 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1275 GOTO(out_stop_service, rc);
1277 /* Register with MGS */
1278 rc = server_register_target(lsi);
1282 /* Let the target look up the mount using the target's name
1283 (we can't pass the sb or mnt through class_process_config.) */
1284 rc = server_register_mount(lsi->lsi_svname, sb);
1288 /* Start targets using the llog named for the target */
1289 memset(&cfg, 0, sizeof(cfg));
1290 cfg.cfg_callback = class_config_llog_handler;
1291 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1293 CERROR("failed to start server %s: %d\n",
1294 lsi->lsi_svname, rc);
1295 /* Do NOT call server_deregister_mount() here. This makes it
1296 * impossible to find mount later in cleanup time and leaves
1297 * @lsi and othder stuff leaked. -umka */
1301 obd = class_name2obd(lsi->lsi_svname);
1303 CERROR("no server named %s was started\n", lsi->lsi_svname);
1304 GOTO(out_mgc, rc = -ENXIO);
1307 if (IS_OST(lsi) || IS_MDT(lsi)) {
1308 rc = lustre_start_lwp(sb);
1310 CERROR("%s: failed to start LWP: %d\n",
1311 lsi->lsi_svname, rc);
1316 server_notify_target(sb, obd);
1318 /* calculate recovery timeout, do it after lustre_process_log */
1319 server_calc_timeout(lsi, obd);
1321 /* log has been fully processed */
1322 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1324 /* log has been fully processed, let clients connect */
1325 dev = obd->obd_lu_dev;
1326 if (dev && dev->ld_ops->ldo_prepare) {
1327 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1329 struct lu_context session_ctx;
1331 lu_context_init(&session_ctx, LCT_SERVER_SESSION);
1332 session_ctx.lc_thread = NULL;
1333 lu_context_enter(&session_ctx);
1334 env.le_ses = &session_ctx;
1336 rc = dev->ld_ops->ldo_prepare(&env, NULL, dev);
1339 lu_context_exit(&session_ctx);
1340 lu_context_fini(&session_ctx);
1344 /* abort recovery only on the complete stack:
1345 * many devices can be involved */
1346 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1347 (OBP(obd, iocontrol))) {
1348 obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
1353 /* Release the mgc fs for others to use */
1354 server_mgc_clear_fs(lsi->lsi_mgc);
1358 server_stop_servers(lsi->lsi_flags);
1363 static int lsi_prepare(struct lustre_sb_info *lsi)
1370 LASSERT(lsi->lsi_lmd);
1372 /* The server name is given as a mount line option */
1373 if (lsi->lsi_lmd->lmd_profile == NULL) {
1374 LCONSOLE_ERROR("Can't determine server name\n");
1378 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
1379 RETURN(-ENAMETOOLONG);
1381 strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
1383 /* Determine osd type */
1384 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
1385 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
1386 sizeof(lsi->lsi_osd_type))
1387 RETURN(-ENAMETOOLONG);
1389 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
1391 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
1394 /* XXX: a temp. solution for components using ldiskfs
1395 * to be removed in one of the subsequent patches */
1396 if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs"))
1397 strcpy(lsi->lsi_fstype, "ldiskfs");
1399 strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
1401 /* Determine server type */
1402 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1404 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
1405 /* Assume we're a bare MGS */
1407 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
1409 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
1414 lsi->lsi_flags |= rc;
1416 /* Add mount line flags that used to be in ldd:
1417 * writeconf, mgs, anything else?
1419 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
1420 LDD_F_WRITECONF : 0;
1421 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
1423 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_UPDATE) ?
1425 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
1426 LDD_F_SV_TYPE_MGS : 0;
1427 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
1428 LDD_F_NO_PRIMNODE : 0;
1433 /*************** server mount ******************/
1435 /** Start the shutdown of servers at umount.
1437 static void server_put_super(struct super_block *sb)
1439 struct lustre_sb_info *lsi = s2lsi(sb);
1440 struct obd_device *obd;
1441 char *tmpname, *extraname = NULL;
1443 int lsiflags = lsi->lsi_flags;
1446 LASSERT(IS_SERVER(lsi));
1448 tmpname_sz = strlen(lsi->lsi_svname) + 1;
1449 OBD_ALLOC(tmpname, tmpname_sz);
1450 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
1451 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1452 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1453 snprintf(tmpname, tmpname_sz, "MGS");
1455 /* disconnect the lwp first to drain off the inflight request */
1456 if (IS_OST(lsi) || IS_MDT(lsi)) {
1459 rc = lustre_disconnect_lwp(sb);
1460 if (rc != 0 && rc != -ETIMEDOUT &&
1461 rc != -ENOTCONN && rc != -ESHUTDOWN)
1462 CWARN("%s: failed to disconnect lwp: rc= %d\n",
1466 /* Stop the target */
1467 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1468 (IS_MDT(lsi) || IS_OST(lsi))) {
1469 struct lustre_profile *lprof = NULL;
1471 /* tell the mgc to drop the config log */
1472 lustre_end_log(sb, lsi->lsi_svname, NULL);
1474 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1475 If there are any setup/cleanup errors, save the lov
1476 name for safety cleanup later. */
1477 lprof = class_get_profile(lsi->lsi_svname);
1478 if (lprof && lprof->lp_dt) {
1479 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1480 strcpy(extraname, lprof->lp_dt);
1483 obd = class_name2obd(lsi->lsi_svname);
1485 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1486 if (lsiflags & LSI_UMOUNT_FAILOVER)
1488 /* We can't seem to give an error return code
1489 * to .put_super, so we better make sure we clean up! */
1491 class_manual_cleanup(obd);
1493 CERROR("no obd %s\n", lsi->lsi_svname);
1494 server_deregister_mount(lsi->lsi_svname);
1498 /* If they wanted the mgs to stop separately from the mdt, they
1499 should have put it on a different device. */
1501 /* if MDS start with --nomgs, don't stop MGS then */
1502 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1503 server_stop_mgs(sb);
1506 if (IS_OST(lsi) || IS_MDT(lsi)) {
1507 if (lustre_stop_lwp(sb) < 0)
1508 CERROR("%s: failed to stop lwp!\n", tmpname);
1511 /* Clean the mgc and sb */
1512 lustre_common_put_super(sb);
1514 /* wait till all in-progress cleanups are done
1515 * specifically we're interested in ofd cleanup
1517 obd_zombie_barrier();
1519 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1520 until the target is really gone so that our type refcount check
1522 server_stop_servers(lsiflags);
1524 /* In case of startup or cleanup err, stop related obds */
1526 obd = class_name2obd(extraname);
1528 CWARN("Cleaning orphaned obd %s\n", extraname);
1530 class_manual_cleanup(obd);
1532 OBD_FREE(extraname, strlen(extraname) + 1);
1535 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1536 OBD_FREE(tmpname, tmpname_sz);
1540 /** Called only for 'umount -f'
1542 static void server_umount_begin(struct super_block *sb)
1544 struct lustre_sb_info *lsi = s2lsi(sb);
1547 CDEBUG(D_MOUNT, "umount -f\n");
1548 /* umount = failover
1550 no third way to do non-force, non-failover */
1551 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1555 static int server_statfs(struct dentry *dentry, struct kstatfs *buf)
1557 struct super_block *sb = dentry->d_sb;
1558 struct lustre_sb_info *lsi = s2lsi(sb);
1559 struct obd_statfs statfs;
1563 if (lsi->lsi_dt_dev) {
1564 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
1566 statfs_unpack(buf, &statfs);
1567 buf->f_type = sb->s_magic;
1573 buf->f_type = sb->s_magic;
1574 buf->f_bsize = sb->s_blocksize;
1580 buf->f_namelen = NAME_MAX;
1584 /** The operations we support directly on the superblock:
1585 * mount, umount, and df.
1587 static struct super_operations server_ops = {
1588 .put_super = server_put_super,
1589 .umount_begin = server_umount_begin, /* umount -f */
1590 .statfs = server_statfs,
1594 * Xattr support for Lustre servers
1596 static ssize_t lustre_getxattr(struct dentry *dentry, const char *name,
1597 void *buffer, size_t size)
1599 if (!selinux_is_enabled())
1604 static int lustre_setxattr(struct dentry *dentry, const char *name,
1605 const void *value, size_t size, int flags)
1610 static ssize_t lustre_listxattr(struct dentry *d_entry, char *name,
1616 const struct inode_operations server_inode_operations = {
1617 .setxattr = lustre_setxattr,
1618 .getxattr = lustre_getxattr,
1619 .listxattr = lustre_listxattr,
1622 #define log2(n) ffz(~(n))
1623 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1625 static int server_fill_super_common(struct super_block *sb)
1627 struct inode *root = 0;
1630 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1632 sb->s_blocksize = 4096;
1633 sb->s_blocksize_bits = log2(sb->s_blocksize);
1634 sb->s_magic = LUSTRE_SUPER_MAGIC;
1635 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1636 sb->s_flags |= MS_RDONLY;
1637 sb->s_op = &server_ops;
1639 root = new_inode(sb);
1641 CERROR("Can't make root inode\n");
1645 /* returns -EIO for every operation */
1646 /* make_bad_inode(root); -- badness - can't umount */
1647 /* apparently we need to be a directory for the mount to finish */
1648 root->i_mode = S_IFDIR;
1649 root->i_op = &server_inode_operations;
1650 sb->s_root = d_make_root(root);
1652 CERROR("%s: can't make root dentry\n", sb->s_id);
1659 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
1661 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1662 struct obd_device *obd;
1663 struct dt_device_param p;
1669 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
1670 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
1672 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
1673 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
1674 strcat(lsi->lsi_osd_uuid, "_UUID");
1675 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
1677 obd = class_name2obd(lsi->lsi_osd_obdname);
1679 rc = lustre_start_simple(lsi->lsi_osd_obdname,
1681 lsi->lsi_osd_uuid, lmd->lmd_dev,
1682 flagstr, lsi->lsi_lmd->lmd_opts,
1686 obd = class_name2obd(lsi->lsi_osd_obdname);
1690 rc = obd_connect(NULL, &lsi->lsi_osd_exp,
1691 obd, &obd->obd_uuid, NULL, NULL);
1694 class_manual_cleanup(obd);
1695 lsi->lsi_dt_dev = NULL;
1699 LASSERT(obd->obd_lu_dev);
1700 lu_device_get(obd->obd_lu_dev);
1701 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
1702 LASSERT(lsi->lsi_dt_dev);
1704 /* set disk context for llog usage */
1705 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1706 obd->obd_lvfs_ctxt.dt = lsi->lsi_dt_dev;
1708 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
1713 /** Fill in the superblock info for a Lustre server.
1714 * Mount the device with the correct options.
1715 * Read the on-disk config file.
1716 * Start the services.
1718 int server_fill_super(struct super_block *sb)
1720 struct lustre_sb_info *lsi = s2lsi(sb);
1724 rc = lsi_prepare(lsi);
1728 /* Start low level OSD */
1729 rc = osd_start(lsi, sb->s_flags);
1731 CERROR("Unable to start osd on %s: %d\n",
1732 lsi->lsi_lmd->lmd_dev, rc);
1737 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
1738 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
1740 if (class_name2obd(lsi->lsi_svname)) {
1741 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1742 "running. Double-mount may have compromised"
1743 " the disk journal.\n",
1749 /* Start MGS before MGC */
1750 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1751 rc = server_start_mgs(sb);
1756 /* Start MGC before servers */
1757 rc = lustre_start_mgc(sb);
1761 /* Set up all obd devices for service */
1762 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1763 (IS_OST(lsi) || IS_MDT(lsi))) {
1764 rc = server_start_targets(sb);
1766 CERROR("Unable to start targets: %d\n", rc);
1769 /* FIXME overmount client here, or can we just start a
1770 * client log and client_fill_super on this sb? We
1771 * need to make sure server_put_super gets called too
1772 * - ll_put_super calls lustre_common_put_super; check
1773 * there for LSI_SERVER flag, call s_p_s if so.
1775 * Probably should start client from new thread so we
1776 * can return. Client will not finish until all
1777 * servers are connected. Note - MGS-only server does
1778 * NOT get a client, since there is no lustre fs
1779 * associated - the MGS is for all lustre fs's */
1782 rc = server_fill_super_common(sb);
1788 /* We jump here in case of failure while starting targets or MGS.
1789 * In this case we can't just put @mnt and have to do real cleanup
1790 * with stoping targets, etc. */
1791 server_put_super(sb);
1796 * Calculate timeout value for a target.
1798 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1800 struct lustre_mount_data *lmd;
1804 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
1805 int min = OBD_RECOVERY_TIME_MIN;
1807 LASSERT(IS_SERVER(lsi));
1811 soft = lmd->lmd_recovery_time_soft;
1812 hard = lmd->lmd_recovery_time_hard;
1813 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1814 obd->obd_no_ir = !has_ir;
1818 soft = OBD_RECOVERY_TIME_SOFT;
1820 hard = OBD_RECOVERY_TIME_HARD;
1822 /* target may have ir_factor configured. */
1823 factor = OBD_IR_FACTOR_DEFAULT;
1824 if (obd->obd_recovery_ir_factor)
1825 factor = obd->obd_recovery_ir_factor;
1828 int new_soft = soft;
1829 int new_hard = hard;
1831 /* adjust timeout value by imperative recovery */
1833 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1834 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1836 /* make sure the timeout is not too short */
1837 new_soft = max(min, new_soft);
1838 new_hard = max(new_soft, new_hard);
1840 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1841 "window shrunk from %d-%d down to %d-%d\n",
1842 obd->obd_name, soft, hard, new_soft, new_hard);
1849 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1850 obd->obd_recovery_time_hard = hard;
1851 obd->obd_recovery_ir_factor = factor;
1853 EXPORT_SYMBOL(server_calc_timeout);