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, 2014, 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)
49 #include <linux/statfs.h>
51 #include <obd_class.h>
52 #include <lustre/lustre_user.h>
53 #include <linux/version.h>
54 #include <lustre_ioctl.h>
55 #include <lustre_log.h>
56 #include <lustre_disk.h>
57 #include <lustre_param.h>
58 #ifdef HAVE_KERNEL_LOCKED
59 #include <linux/smp_lock.h>
61 #ifdef HAVE_SELINUX_IS_ENABLED
62 #include <linux/selinux.h>
65 /*********** mount lookup *********/
67 static DEFINE_MUTEX(lustre_mount_info_lock);
68 static struct list_head server_mount_info_list =
69 LIST_HEAD_INIT(server_mount_info_list);
71 static struct lustre_mount_info *server_find_mount(const char *name)
73 struct list_head *tmp;
74 struct lustre_mount_info *lmi;
77 list_for_each(tmp, &server_mount_info_list) {
78 lmi = list_entry(tmp, struct lustre_mount_info,
80 if (strcmp(name, lmi->lmi_name) == 0)
86 /* we must register an obd for a mount before we call the setup routine.
87 *_setup will call lustre_get_mount to get the mnt struct
88 by obd_name, since we can't pass the pointer to setup. */
89 static int server_register_mount(const char *name, struct super_block *sb)
91 struct lustre_mount_info *lmi;
97 OBD_ALLOC(lmi, sizeof(*lmi));
100 OBD_ALLOC(name_cp, strlen(name) + 1);
102 OBD_FREE(lmi, sizeof(*lmi));
105 strcpy(name_cp, name);
107 mutex_lock(&lustre_mount_info_lock);
109 if (server_find_mount(name)) {
110 mutex_unlock(&lustre_mount_info_lock);
111 OBD_FREE(lmi, sizeof(*lmi));
112 OBD_FREE(name_cp, strlen(name) + 1);
113 CERROR("Already registered %s\n", name);
116 lmi->lmi_name = name_cp;
118 list_add(&lmi->lmi_list_chain, &server_mount_info_list);
120 mutex_unlock(&lustre_mount_info_lock);
122 CDEBUG(D_MOUNT, "register mount %p from %s\n", sb, name);
127 /* when an obd no longer needs a mount */
128 static int server_deregister_mount(const char *name)
130 struct lustre_mount_info *lmi;
133 mutex_lock(&lustre_mount_info_lock);
134 lmi = server_find_mount(name);
136 mutex_unlock(&lustre_mount_info_lock);
137 CERROR("%s not registered\n", name);
141 CDEBUG(D_MOUNT, "deregister mount %p from %s\n", lmi->lmi_sb, name);
143 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
144 list_del(&lmi->lmi_list_chain);
145 OBD_FREE(lmi, sizeof(*lmi));
146 mutex_unlock(&lustre_mount_info_lock);
151 /* obd's look up a registered mount using their obdname. This is just
152 for initial obd setup to find the mount struct. It should not be
153 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 atomic_inc(&lsi->lsi_mounts);
171 CDEBUG(D_MOUNT, "get mount %p from %s, refs=%d\n", lmi->lmi_sb,
172 name, atomic_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, atomic_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 */
230 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
232 lsi = s2lsi(lmi->lmi_sb);
233 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
238 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
240 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb);
243 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
244 LUSTRE_MGS_OBDNAME, NULL, NULL,
245 lsi->lsi_osd_obdname, NULL);
246 /* server_deregister_mount() is not called previously, for lsi
247 * and other stuff can't be freed cleanly when mgs calls
248 * server_put_mount() in error handling case (see b=17758),
249 * this problem is caused by a bug in mgs_init0, which forgot
250 * calling server_put_mount in error case. */
253 server_deregister_mount(LUSTRE_MGS_OBDNAME);
257 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
258 "Is the 'mgs' module loaded?\n",
259 LUSTRE_MGS_OBDNAME, rc);
263 static int server_stop_mgs(struct super_block *sb)
265 struct obd_device *obd;
267 struct lustre_mount_info *lmi;
270 /* Do not stop MGS if this device is not the running MGT */
271 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
272 if (lmi != NULL && lmi->lmi_sb != sb)
275 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
277 /* There better be only one MGS */
278 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
280 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
284 /* The MGS should always stop when we say so */
286 rc = class_manual_cleanup(obd);
290 /* Since there's only one mgc per node, we have to change it's fs to get
291 access to the right disk. */
292 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
294 struct lustre_sb_info *lsi = s2lsi(sb);
298 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
300 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
301 rc = obd_set_info_async(NULL, mgc->obd_self_export,
302 sizeof(KEY_SET_FS), KEY_SET_FS,
303 sizeof(*sb), sb, NULL);
305 CERROR("can't set_fs %d\n", rc);
310 static int server_mgc_clear_fs(struct obd_device *mgc)
315 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
317 rc = obd_set_info_async(NULL, mgc->obd_self_export,
318 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
323 static inline bool is_mdc_device(const char *devname)
327 ptr = strrchr(devname, '-');
328 return ptr != NULL && strcmp(ptr, "-mdc") == 0;
331 static inline bool tgt_is_mdt(const char *tgtname, __u32 *idx)
335 type = server_name2index(tgtname, idx, NULL);
337 return type == LDD_F_SV_TYPE_MDT;
341 * Convert OST/MDT name(fsname-{MDT,OST}xxxx) to a lwp name with the @idx:yyyy
342 * (fsname-MDTyyyy-lwp-{MDT,OST}xxxx)
344 int tgt_name2lwp_name(const char *tgt_name, char *lwp_name, int len, __u32 idx)
351 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
355 rc = server_name2fsname(tgt_name, fsname, &tgt);
357 CERROR("%s: failed to get fsname from tgt_name: rc = %d\n",
362 if (*tgt != '-' && *tgt != ':') {
363 CERROR("%s: invalid tgt_name name!\n", tgt_name);
364 GOTO(cleanup, rc = -EINVAL);
368 if (strncmp(tgt, "OST", 3) != 0 && strncmp(tgt, "MDT", 3) != 0) {
369 CERROR("%s is not an OST or MDT target!\n", tgt_name);
370 GOTO(cleanup, rc = -EINVAL);
372 snprintf(lwp_name, len, "%s-MDT%04x-%s-%s",
373 fsname, idx, LUSTRE_LWP_NAME, tgt);
375 GOTO(cleanup, rc = 0);
379 OBD_FREE(fsname, MTI_NAME_MAXLEN);
383 EXPORT_SYMBOL(tgt_name2lwp_name);
385 static struct list_head lwp_register_list =
386 LIST_HEAD_INIT(lwp_register_list);
387 static DEFINE_MUTEX(lwp_register_list_lock);
389 int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp,
390 register_lwp_cb cb_func, void *cb_data)
392 struct obd_device *lwp;
393 struct lwp_register_item *lri;
396 LASSERTF(strlen(lwpname) < MTI_NAME_MAXLEN, "lwpname is too long %s\n",
398 LASSERT(exp != NULL && *exp == NULL);
404 mutex_lock(&lwp_register_list_lock);
406 lwp = class_name2obd(lwpname);
407 if (lwp != NULL && lwp->obd_set_up == 1) {
408 struct obd_uuid *uuid;
412 mutex_unlock(&lwp_register_list_lock);
416 memcpy(uuid->uuid, lwpname, strlen(lwpname));
417 *exp = cfs_hash_lookup(lwp->obd_uuid_hash, uuid);
421 memcpy(lri->lri_name, lwpname, strlen(lwpname));
423 lri->lri_cb_func = cb_func;
424 lri->lri_cb_data = cb_data;
425 INIT_LIST_HEAD(&lri->lri_list);
426 list_add(&lri->lri_list, &lwp_register_list);
428 if (*exp != NULL && cb_func != NULL)
431 mutex_unlock(&lwp_register_list_lock);
434 EXPORT_SYMBOL(lustre_register_lwp_item);
436 void lustre_deregister_lwp_item(struct obd_export **exp)
438 struct lwp_register_item *lri, *tmp;
440 mutex_lock(&lwp_register_list_lock);
441 list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
442 if (exp == lri->lri_exp) {
444 class_export_put(*exp);
445 list_del(&lri->lri_list);
450 mutex_unlock(&lwp_register_list_lock);
452 EXPORT_SYMBOL(lustre_deregister_lwp_item);
454 struct obd_export *lustre_find_lwp_by_index(const char *dev, __u32 idx)
456 struct lustre_mount_info *lmi;
457 struct lustre_sb_info *lsi;
458 struct obd_device *lwp;
459 struct obd_export *exp = NULL;
464 lmi = server_get_mount(dev);
468 lsi = s2lsi(lmi->lmi_sb);
469 rc = server_name2fsname(lsi->lsi_svname, fsname, NULL);
471 CERROR("%s: failed to get fsname: rc = %d\n",
472 lsi->lsi_svname, rc);
476 snprintf(lwp_name, sizeof(lwp_name), "%s-MDT%04x", fsname, idx);
477 spin_lock(&lsi->lsi_lwp_lock);
478 list_for_each_entry(lwp, &lsi->lsi_lwp_list, obd_lwp_list) {
479 char *ptr = strstr(lwp->obd_name, lwp_name);
482 exp = class_export_get(lwp->obd_lwp_export);
486 spin_unlock(&lsi->lsi_lwp_lock);
489 server_put_mount(dev, false);
493 EXPORT_SYMBOL(lustre_find_lwp_by_index);
495 static void lustre_notify_lwp_list(struct obd_export *exp)
497 struct lwp_register_item *lri, *tmp;
498 LASSERT(exp != NULL);
500 mutex_lock(&lwp_register_list_lock);
501 list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
502 if (strcmp(exp->exp_obd->obd_name, lri->lri_name))
504 if (*lri->lri_exp != NULL)
506 *lri->lri_exp = class_export_get(exp);
507 if (lri->lri_cb_func != NULL)
508 lri->lri_cb_func(lri->lri_cb_data);
510 mutex_unlock(&lwp_register_list_lock);
513 static int lustre_lwp_connect(struct obd_device *lwp)
516 struct lu_context session_ctx;
517 struct obd_export *exp;
518 struct obd_uuid *uuid = NULL;
519 struct obd_connect_data *data = NULL;
523 /* log has been fully processed, let clients connect */
524 rc = lu_env_init(&env, lwp->obd_lu_dev->ld_type->ldt_ctx_tags);
528 lu_context_init(&session_ctx, LCT_SERVER_SESSION);
529 session_ctx.lc_thread = NULL;
530 lu_context_enter(&session_ctx);
531 env.le_ses = &session_ctx;
535 GOTO(out, rc = -ENOMEM);
537 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
538 data->ocd_version = LUSTRE_VERSION_CODE;
539 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
540 OBD_CONNECT_AT | OBD_CONNECT_LRU_RESIZE |
541 OBD_CONNECT_FULL20 | OBD_CONNECT_LVB_TYPE |
542 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_LFSCK;
545 GOTO(out, rc = -ENOMEM);
547 if (strlen(lwp->obd_name) > sizeof(uuid->uuid)) {
548 CERROR("%s: Too long lwp name %s, max_size is %d\n",
549 lwp->obd_name, lwp->obd_name, (int)sizeof(uuid->uuid));
550 GOTO(out, rc = -EINVAL);
553 /* Use lwp name as the uuid, so we find the export by lwp name later */
554 memcpy(uuid->uuid, lwp->obd_name, strlen(lwp->obd_name));
555 rc = obd_connect(&env, &exp, lwp, uuid, data, NULL);
557 CERROR("%s: connect failed: rc = %d\n", lwp->obd_name, rc);
559 if (unlikely(lwp->obd_lwp_export != NULL))
560 class_export_put(lwp->obd_lwp_export);
561 lwp->obd_lwp_export = class_export_get(exp);
562 lustre_notify_lwp_list(exp);
574 lu_context_exit(&session_ctx);
575 lu_context_fini(&session_ctx);
581 * lwp is used by slaves (Non-MDT0 targets) to manage the connection to MDT0,
582 * or from the OSTx to MDTy.
584 static int lustre_lwp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi,
587 struct obd_device *obd;
588 char *lwpname = NULL;
589 char *lwpuuid = NULL;
593 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
596 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
600 OBD_ALLOC(lwpname, MTI_NAME_MAXLEN);
602 GOTO(out, rc = -ENOMEM);
604 rc = tgt_name2lwp_name(lsi->lsi_svname, lwpname, MTI_NAME_MAXLEN, idx);
606 CERROR("%s: failed to generate lwp name: rc = %d\n",
607 lsi->lsi_svname, rc);
611 OBD_ALLOC(lwpuuid, MTI_NAME_MAXLEN);
613 GOTO(out, rc = -ENOMEM);
615 sprintf(lwpuuid, "%s_UUID", lwpname);
616 rc = lustre_start_simple(lwpname, LUSTRE_LWP_NAME,
617 lwpuuid, lustre_cfg_string(lcfg, 1),
620 CERROR("%s: setup up failed: rc %d\n", lwpname, rc);
624 obd = class_name2obd(lwpname);
625 LASSERT(obd != NULL);
627 rc = lustre_lwp_connect(obd);
629 obd->u.cli.cl_max_mds_easize = MAX_MD_SIZE;
630 spin_lock(&lsi->lsi_lwp_lock);
631 list_add_tail(&obd->obd_lwp_list, &lsi->lsi_lwp_list);
632 spin_unlock(&lsi->lsi_lwp_lock);
634 CERROR("%s: connect failed: rc = %d\n", lwpname, rc);
641 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
643 OBD_FREE(lwpuuid, MTI_NAME_MAXLEN);
648 /* the caller is responsible for memory free */
649 static struct obd_device *lustre_find_lwp(struct lustre_sb_info *lsi,
650 char **lwpname, __u32 idx)
652 struct obd_device *lwp;
656 LASSERT(lwpname != NULL);
657 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
659 OBD_ALLOC(*lwpname, MTI_NAME_MAXLEN);
660 if (*lwpname == NULL)
661 RETURN(ERR_PTR(-ENOMEM));
663 rc = tgt_name2lwp_name(lsi->lsi_svname, *lwpname, MTI_NAME_MAXLEN, idx);
665 CERROR("%s: failed to generate lwp name: rc = %d\n",
666 lsi->lsi_svname, rc);
667 GOTO(out, rc = -EINVAL);
670 lwp = class_name2obd(*lwpname);
674 if (*lwpname != NULL) {
675 OBD_FREE(*lwpname, MTI_NAME_MAXLEN);
681 RETURN(lwp != NULL ? lwp : ERR_PTR(-ENOENT));
684 static int lustre_lwp_add_conn(struct lustre_cfg *cfg,
685 struct lustre_sb_info *lsi, __u32 idx)
687 struct lustre_cfg_bufs *bufs = NULL;
688 struct lustre_cfg *lcfg = NULL;
689 char *lwpname = NULL;
690 struct obd_device *lwp;
694 lwp = lustre_find_lwp(lsi, &lwpname, idx);
696 CERROR("%s: can't find lwp device.\n", lsi->lsi_svname);
697 GOTO(out, rc = PTR_ERR(lwp));
699 LASSERT(lwpname != NULL);
703 GOTO(out, rc = -ENOMEM);
705 lustre_cfg_bufs_reset(bufs, lwpname);
706 lustre_cfg_bufs_set_string(bufs, 1,
707 lustre_cfg_string(cfg, 1));
709 lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
711 GOTO(out_cfg, rc = -ENOMEM);
712 rc = class_add_conn(lwp, lcfg);
714 CERROR("%s: can't add conn: rc = %d\n", lwpname, rc);
717 lustre_cfg_free(lcfg);
723 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
728 * Retrieve MDT nids from the client log, then start the lwp device.
729 * there are only two scenarios which would include mdt nid.
731 * marker 5 (flags=0x01, v2.1.54.0) lustre-MDTyyyy 'add mdc' xxx-
732 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
733 * attach 0:lustre-MDTyyyy-mdc 1:mdc 2:lustre-clilmv_UUID
734 * setup 0:lustre-MDTyyyy-mdc 1:lustre-MDTyyyy_UUID 2:192.168.122.162@tcp
735 * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
736 * add_conn 0:lustre-MDTyyyy-mdc 1:192.168.172.1@tcp
737 * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDTyyyy_UUID xxxx
738 * marker 5 (flags=0x02, v2.1.54.0) lustre-MDTyyyy 'add mdc' xxxx-
740 * marker 7 (flags=0x01, v2.1.54.0) lustre-MDTyyyy 'add failnid' xxxx-
741 * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
742 * add_conn 0:lustre-MDTyyyy-mdc 1:192.168.122.2@tcp
743 * marker 7 (flags=0x02, v2.1.54.0) lustre-MDTyyyy 'add failnid' xxxx-
745 static int client_lwp_config_process(const struct lu_env *env,
746 struct llog_handle *handle,
747 struct llog_rec_hdr *rec, void *data)
749 struct config_llog_instance *clli = data;
750 int cfg_len = rec->lrh_len;
751 char *cfg_buf = (char *) (rec + 1);
752 struct lustre_cfg *lcfg = NULL;
753 struct lustre_sb_info *lsi;
754 int rc = 0, swab = 0;
757 if (rec->lrh_type != OBD_CFG_REC) {
758 CERROR("Unknown llog record type %#x encountered\n",
763 LASSERT(clli->cfg_sb != NULL);
764 lsi = s2lsi(clli->cfg_sb);
766 lcfg = (struct lustre_cfg *)cfg_buf;
767 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
768 lustre_swab_lustre_cfg(lcfg);
772 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
776 switch (lcfg->lcfg_command) {
778 struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
780 lustre_swab_cfg_marker(marker, swab,
781 LUSTRE_CFG_BUFLEN(lcfg, 1));
782 if (marker->cm_flags & CM_SKIP ||
783 marker->cm_flags & CM_EXCLUDE)
786 if (!tgt_is_mdt(marker->cm_tgtname, &clli->cfg_lwp_idx))
789 if (IS_MDT(lsi) && clli->cfg_lwp_idx != 0)
792 if (!strncmp(marker->cm_comment, "add mdc", 7) ||
793 !strncmp(marker->cm_comment, "add failnid", 11)) {
794 if (marker->cm_flags & CM_START) {
795 clli->cfg_flags = CFG_F_MARKER;
796 /* This hack is to differentiate the
797 * ADD_UUID is come from "add mdc" record
798 * or from "add failnid" record. */
799 if (!strncmp(marker->cm_comment,
801 clli->cfg_flags |= CFG_F_SKIP;
802 } else if (marker->cm_flags & CM_END) {
808 case LCFG_ADD_UUID: {
809 if (clli->cfg_flags == CFG_F_MARKER) {
810 rc = lustre_lwp_setup(lcfg, lsi, clli->cfg_lwp_idx);
811 /* XXX: process only the first nid as
812 * we don't need another instance of lwp */
813 clli->cfg_flags |= CFG_F_SKIP;
814 } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
815 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
818 CERROR("%s: Fail to add uuid, rc:%d\n",
819 lsi->lsi_svname, rc);
823 case LCFG_ADD_CONN: {
824 char *devname = lustre_cfg_string(lcfg, 0);
828 if (!is_mdc_device(devname))
831 ptr = strrchr(devname, '-');
836 if (!tgt_is_mdt(devname, &idx)) {
842 if (IS_MDT(lsi) && idx != 0)
845 rc = lustre_lwp_add_conn(lcfg, lsi, idx);
855 static int lustre_disconnect_lwp(struct super_block *sb)
857 struct lustre_sb_info *lsi = s2lsi(sb);
858 struct obd_device *lwp;
859 char *logname = NULL;
860 struct lustre_cfg_bufs *bufs = NULL;
861 struct config_llog_instance *cfg = NULL;
866 if (likely(lsi->lsi_lwp_started)) {
867 OBD_ALLOC(logname, MTI_NAME_MAXLEN);
871 rc = server_name2fsname(lsi->lsi_svname, logname, NULL);
873 CERROR("%s: failed to get fsname from svname: "
874 "rc = %d\n", lsi->lsi_svname, rc);
875 GOTO(out, rc = -EINVAL);
878 strcat(logname, "-client");
881 GOTO(out, rc = -ENOMEM);
884 cfg->cfg_instance = sb;
885 rc = lustre_end_log(sb, logname, cfg);
889 lsi->lsi_lwp_started = 0;
894 GOTO(out, rc = -ENOMEM);
896 list_for_each_entry(lwp, &lsi->lsi_lwp_list, obd_lwp_list) {
897 struct lustre_cfg *lcfg;
899 if (likely(lwp->obd_lwp_export != NULL)) {
900 class_export_put(lwp->obd_lwp_export);
901 lwp->obd_lwp_export = NULL;
904 lustre_cfg_bufs_reset(bufs, lwp->obd_name);
905 lustre_cfg_bufs_set_string(bufs, 1, NULL);
906 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
908 GOTO(out, rc = -ENOMEM);
910 /* Disconnect import first. NULL is passed for the '@env',
911 * since it will not be used. */
912 rc = lwp->obd_lu_dev->ld_ops->ldo_process_config(NULL,
913 lwp->obd_lu_dev, lcfg);
914 lustre_cfg_free(lcfg);
915 if (rc != 0 && rc != -ETIMEDOUT) {
916 CERROR("%s: fail to disconnect LWP: rc = %d\n",
930 OBD_FREE(logname, MTI_NAME_MAXLEN);
932 return rc1 != 0 ? rc1 : rc;
936 * Stop the lwp for an OST/MDT target.
938 static int lustre_stop_lwp(struct super_block *sb)
940 struct lustre_sb_info *lsi = s2lsi(sb);
941 struct obd_device *lwp;
946 while (!list_empty(&lsi->lsi_lwp_list)) {
947 lwp = list_entry(lsi->lsi_lwp_list.next, struct obd_device,
949 list_del_init(&lwp->obd_lwp_list);
951 rc = class_manual_cleanup(lwp);
953 CERROR("%s: fail to stop LWP: rc = %d\n",
959 RETURN(rc1 != 0 ? rc1 : rc);
963 * Start the lwp(fsname-MDTyyyy-lwp-{MDT,OST}xxxx) for a MDT/OST or MDT target.
965 static int lustre_start_lwp(struct super_block *sb)
967 struct lustre_sb_info *lsi = s2lsi(sb);
968 struct config_llog_instance *cfg = NULL;
973 if (unlikely(lsi->lsi_lwp_started))
976 OBD_ALLOC(logname, MTI_NAME_MAXLEN);
980 rc = server_name2fsname(lsi->lsi_svname, logname, NULL);
982 CERROR("%s: failed to get fsname from svname: rc = %d\n",
983 lsi->lsi_svname, rc);
984 GOTO(out, rc = -EINVAL);
987 strcat(logname, "-client");
990 GOTO(out, rc = -ENOMEM);
992 cfg->cfg_callback = client_lwp_config_process;
993 cfg->cfg_instance = sb;
994 rc = lustre_process_log(sb, logname, cfg);
996 lsi->lsi_lwp_started = 1;
1001 OBD_FREE(logname, MTI_NAME_MAXLEN);
1008 static DEFINE_MUTEX(server_start_lock);
1010 /* Stop MDS/OSS if nobody is using them */
1011 static int server_stop_servers(int lsiflags)
1013 struct obd_device *obd = NULL;
1014 struct obd_type *type = NULL;
1018 mutex_lock(&server_start_lock);
1020 /* Either an MDT or an OST or neither */
1021 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
1022 if (lsiflags & LDD_F_SV_TYPE_MDT) {
1023 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1025 type = class_search_type(LUSTRE_MDT_NAME);
1028 /* if this was an OST, and there are no more OST's, clean up the OSS */
1029 if (lsiflags & LDD_F_SV_TYPE_OST) {
1030 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1032 type = class_search_type(LUSTRE_OST_NAME);
1035 if (obd != NULL && (type == NULL || type->typ_refcnt == 0)) {
1037 /* obd_fail doesn't mean much on a server obd */
1038 rc = class_manual_cleanup(obd);
1041 mutex_unlock(&server_start_lock);
1046 int server_mti_print(const char *title, struct mgs_target_info *mti)
1048 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
1049 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
1050 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
1051 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
1052 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
1053 mti->mti_config_ver, mti->mti_flags);
1057 /* Generate data for registration */
1058 static int server_lsi2mti(struct lustre_sb_info *lsi,
1059 struct mgs_target_info *mti)
1061 lnet_process_id_t id;
1066 if (!IS_SERVER(lsi))
1069 if (strlcpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname))
1070 >= sizeof(mti->mti_svname))
1073 mti->mti_nid_count = 0;
1074 while (LNetGetId(i++, &id) != -ENOENT) {
1075 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1078 /* server use --servicenode param, only allow specified
1079 * nids be registered */
1080 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
1081 class_match_nid(lsi->lsi_lmd->lmd_params,
1082 PARAM_FAILNODE, id.nid) < 1)
1085 /* match specified network */
1086 if (!class_match_net(lsi->lsi_lmd->lmd_params,
1087 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1090 mti->mti_nids[mti->mti_nid_count] = id.nid;
1091 mti->mti_nid_count++;
1092 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1093 CWARN("Only using first %d nids for %s\n",
1094 mti->mti_nid_count, mti->mti_svname);
1099 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1100 mti->mti_config_ver = 0;
1102 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1106 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1109 /* Orion requires index to be set */
1110 LASSERT(!(rc & LDD_F_NEED_INDEX));
1111 /* keep only LDD flags */
1112 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1113 if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN))
1114 mti->mti_flags |= LDD_F_UPDATE;
1115 cplen = strlcpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1116 sizeof(mti->mti_params));
1117 if (cplen >= sizeof(mti->mti_params))
1122 /* Register an old or new target with the MGS. If needed MGS will construct
1123 startup logs and assign index */
1124 static int server_register_target(struct lustre_sb_info *lsi)
1126 struct obd_device *mgc = lsi->lsi_mgc;
1127 struct mgs_target_info *mti = NULL;
1134 if (!IS_SERVER(lsi))
1141 rc = server_lsi2mti(lsi, mti);
1145 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1146 mti->mti_svname, mti->mti_fsname,
1147 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1150 /* if write_conf is true, the registration must succeed */
1151 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1152 mti->mti_flags |= LDD_F_OPC_REG;
1154 /* Register the target */
1155 /* FIXME use mgc_process_config instead */
1156 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1157 sizeof(KEY_REGISTER_TARGET),
1158 KEY_REGISTER_TARGET,
1159 sizeof(*mti), mti, NULL);
1161 if (mti->mti_flags & LDD_F_ERROR) {
1162 LCONSOLE_ERROR_MSG(0x160,
1163 "%s: the MGS refuses to allow this server "
1164 "to start: rc = %d. Please see messages on "
1165 "the MGS.\n", lsi->lsi_svname, rc);
1166 } else if (writeconf) {
1167 LCONSOLE_ERROR_MSG(0x15f,
1168 "%s: cannot register this server with the MGS: "
1169 "rc = %d. Is the MGS running?\n",
1170 lsi->lsi_svname, rc);
1172 CDEBUG(D_HA, "%s: error registering with the MGS: "
1173 "rc = %d (not fatal)\n", lsi->lsi_svname, rc);
1174 /* reset the error code for non-fatal error. */
1187 * Notify the MGS that this target is ready.
1188 * Used by IR - if the MGS receives this message, it will notify clients.
1190 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1192 struct lustre_sb_info *lsi = s2lsi(sb);
1193 struct obd_device *mgc = lsi->lsi_mgc;
1194 struct mgs_target_info *mti = NULL;
1200 if (!(IS_SERVER(lsi)))
1206 rc = server_lsi2mti(lsi, mti);
1210 mti->mti_instance = obd->u.obt.obt_instance;
1211 mti->mti_flags |= LDD_F_OPC_READY;
1213 /* FIXME use mgc_process_config instead */
1214 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1215 sizeof(KEY_REGISTER_TARGET),
1216 KEY_REGISTER_TARGET,
1217 sizeof(*mti), mti, NULL);
1219 /* Imperative recovery: if the mgs informs us to use IR? */
1220 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1221 (mti->mti_flags & LDD_F_IR_CAPABLE))
1222 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1231 /** Start server targets: MDTs and OSTs
1233 static int server_start_targets(struct super_block *sb)
1235 struct obd_device *obd;
1236 struct lustre_sb_info *lsi = s2lsi(sb);
1237 struct config_llog_instance cfg;
1239 struct lu_device *dev;
1243 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1246 /* make sure the MDS is started */
1247 mutex_lock(&server_start_lock);
1248 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1250 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1252 LUSTRE_MDS_OBDNAME"_uuid",
1253 NULL, NULL, NULL, NULL);
1255 mutex_unlock(&server_start_lock);
1256 CERROR("failed to start MDS: %d\n", rc);
1260 mutex_unlock(&server_start_lock);
1263 /* If we're an OST, make sure the global OSS is running */
1265 /* make sure OSS is started */
1266 mutex_lock(&server_start_lock);
1267 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1269 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1271 LUSTRE_OSS_OBDNAME"_uuid",
1272 NULL, NULL, NULL, NULL);
1274 mutex_unlock(&server_start_lock);
1275 CERROR("failed to start OSS: %d\n", rc);
1279 mutex_unlock(&server_start_lock);
1282 /* Set the mgc fs to our server disk. This allows the MGC to
1283 * read and write configs locally, in case it can't talk to the MGS. */
1284 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1286 GOTO(out_stop_service, rc);
1288 /* Register with MGS */
1289 rc = server_register_target(lsi);
1293 /* Let the target look up the mount using the target's name
1294 (we can't pass the sb or mnt through class_process_config.) */
1295 rc = server_register_mount(lsi->lsi_svname, sb);
1299 /* Start targets using the llog named for the target */
1300 memset(&cfg, 0, sizeof(cfg));
1301 cfg.cfg_callback = class_config_llog_handler;
1302 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1304 CERROR("failed to start server %s: %d\n",
1305 lsi->lsi_svname, rc);
1306 /* Do NOT call server_deregister_mount() here. This makes it
1307 * impossible to find mount later in cleanup time and leaves
1308 * @lsi and othder stuff leaked. -umka */
1312 obd = class_name2obd(lsi->lsi_svname);
1314 CERROR("no server named %s was started\n", lsi->lsi_svname);
1315 GOTO(out_mgc, rc = -ENXIO);
1318 if (IS_OST(lsi) || IS_MDT(lsi)) {
1319 rc = lustre_start_lwp(sb);
1321 CERROR("%s: failed to start LWP: %d\n",
1322 lsi->lsi_svname, rc);
1327 server_notify_target(sb, obd);
1329 /* calculate recovery timeout, do it after lustre_process_log */
1330 server_calc_timeout(lsi, obd);
1332 /* log has been fully processed */
1333 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1335 /* log has been fully processed, let clients connect */
1336 dev = obd->obd_lu_dev;
1337 if (dev && dev->ld_ops->ldo_prepare) {
1338 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1340 struct lu_context session_ctx;
1342 lu_context_init(&session_ctx, LCT_SERVER_SESSION);
1343 session_ctx.lc_thread = NULL;
1344 lu_context_enter(&session_ctx);
1345 env.le_ses = &session_ctx;
1347 rc = dev->ld_ops->ldo_prepare(&env, NULL, dev);
1350 lu_context_exit(&session_ctx);
1351 lu_context_fini(&session_ctx);
1355 /* abort recovery only on the complete stack:
1356 * many devices can be involved */
1357 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1358 (OBP(obd, iocontrol))) {
1359 obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
1364 /* Release the mgc fs for others to use */
1365 server_mgc_clear_fs(lsi->lsi_mgc);
1369 server_stop_servers(lsi->lsi_flags);
1374 static int lsi_prepare(struct lustre_sb_info *lsi)
1376 const char *osd_type;
1383 LASSERT(lsi->lsi_lmd);
1385 /* The server name is given as a mount line option */
1386 if (lsi->lsi_lmd->lmd_profile == NULL) {
1387 LCONSOLE_ERROR("Can't determine server name\n");
1391 /* Determine osd type */
1392 if (lsi->lsi_lmd->lmd_osd_type == NULL) {
1393 osd_type = LUSTRE_OSD_LDISKFS_NAME;
1396 osd_type = lsi->lsi_lmd->lmd_osd_type;
1397 fstype = lsi->lsi_lmd->lmd_osd_type;
1400 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname) ||
1401 strlen(osd_type) >= sizeof(lsi->lsi_osd_type) ||
1402 strlen(fstype) >= sizeof(lsi->lsi_fstype))
1403 RETURN(-ENAMETOOLONG);
1405 strlcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile,
1406 sizeof(lsi->lsi_svname));
1407 strlcpy(lsi->lsi_osd_type, osd_type, sizeof(lsi->lsi_osd_type));
1408 /* XXX: a temp. solution for components using ldiskfs
1409 * to be removed in one of the subsequent patches */
1410 strlcpy(lsi->lsi_fstype, fstype, sizeof(lsi->lsi_fstype));
1412 /* Determine server type */
1413 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1415 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
1416 /* Assume we're a bare MGS */
1418 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
1420 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
1425 lsi->lsi_flags |= rc;
1427 /* Add mount line flags that used to be in ldd:
1428 * writeconf, mgs, anything else?
1430 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
1431 LDD_F_WRITECONF : 0;
1432 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
1434 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_UPDATE) ?
1436 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
1437 LDD_F_SV_TYPE_MGS : 0;
1438 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
1439 LDD_F_NO_PRIMNODE : 0;
1444 /*************** server mount ******************/
1446 /** Start the shutdown of servers at umount.
1448 static void server_put_super(struct super_block *sb)
1450 struct lustre_sb_info *lsi = s2lsi(sb);
1451 struct obd_device *obd;
1452 char *tmpname, *extraname = NULL;
1454 int lsiflags = lsi->lsi_flags;
1457 LASSERT(IS_SERVER(lsi));
1459 tmpname_sz = strlen(lsi->lsi_svname) + 1;
1460 OBD_ALLOC(tmpname, tmpname_sz);
1461 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
1462 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1463 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1464 snprintf(tmpname, tmpname_sz, "MGS");
1466 /* disconnect the lwp first to drain off the inflight request */
1467 if (IS_OST(lsi) || IS_MDT(lsi)) {
1470 rc = lustre_disconnect_lwp(sb);
1471 if (rc != 0 && rc != -ETIMEDOUT &&
1472 rc != -ENOTCONN && rc != -ESHUTDOWN)
1473 CWARN("%s: failed to disconnect lwp: rc= %d\n",
1477 /* Stop the target */
1478 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1479 (IS_MDT(lsi) || IS_OST(lsi))) {
1480 struct lustre_profile *lprof = NULL;
1482 /* tell the mgc to drop the config log */
1483 lustre_end_log(sb, lsi->lsi_svname, NULL);
1485 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1486 If there are any setup/cleanup errors, save the lov
1487 name for safety cleanup later. */
1488 lprof = class_get_profile(lsi->lsi_svname);
1489 if (lprof && lprof->lp_dt) {
1490 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1491 strcpy(extraname, lprof->lp_dt);
1494 obd = class_name2obd(lsi->lsi_svname);
1496 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1497 if (lsiflags & LSI_UMOUNT_FAILOVER)
1499 /* We can't seem to give an error return code
1500 * to .put_super, so we better make sure we clean up! */
1502 class_manual_cleanup(obd);
1504 CERROR("no obd %s\n", lsi->lsi_svname);
1505 server_deregister_mount(lsi->lsi_svname);
1509 /* If they wanted the mgs to stop separately from the mdt, they
1510 should have put it on a different device. */
1512 /* if MDS start with --nomgs, don't stop MGS then */
1513 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1514 server_stop_mgs(sb);
1517 if (IS_OST(lsi) || IS_MDT(lsi)) {
1518 if (lustre_stop_lwp(sb) < 0)
1519 CERROR("%s: failed to stop lwp!\n", tmpname);
1522 /* Clean the mgc and sb */
1523 lustre_common_put_super(sb);
1525 /* wait till all in-progress cleanups are done
1526 * specifically we're interested in ofd cleanup
1528 obd_zombie_barrier();
1530 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1531 until the target is really gone so that our type refcount check
1533 server_stop_servers(lsiflags);
1535 /* In case of startup or cleanup err, stop related obds */
1537 obd = class_name2obd(extraname);
1539 CWARN("Cleaning orphaned obd %s\n", extraname);
1541 class_manual_cleanup(obd);
1543 OBD_FREE(extraname, strlen(extraname) + 1);
1546 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1547 OBD_FREE(tmpname, tmpname_sz);
1551 /** Called only for 'umount -f'
1553 static void server_umount_begin(struct super_block *sb)
1555 struct lustre_sb_info *lsi = s2lsi(sb);
1558 CDEBUG(D_MOUNT, "umount -f\n");
1559 /* umount = failover
1561 no third way to do non-force, non-failover */
1562 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1566 static int server_statfs(struct dentry *dentry, struct kstatfs *buf)
1568 struct super_block *sb = dentry->d_sb;
1569 struct lustre_sb_info *lsi = s2lsi(sb);
1570 struct obd_statfs statfs;
1574 if (lsi->lsi_dt_dev) {
1575 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
1577 statfs_unpack(buf, &statfs);
1578 buf->f_type = sb->s_magic;
1584 buf->f_type = sb->s_magic;
1585 buf->f_bsize = sb->s_blocksize;
1591 buf->f_namelen = NAME_MAX;
1595 /** The operations we support directly on the superblock:
1596 * mount, umount, and df.
1598 static struct super_operations server_ops = {
1599 .put_super = server_put_super,
1600 .umount_begin = server_umount_begin, /* umount -f */
1601 .statfs = server_statfs,
1605 * Xattr support for Lustre servers
1607 static ssize_t lustre_getxattr(struct dentry *dentry, const char *name,
1608 void *buffer, size_t size)
1610 if (!selinux_is_enabled())
1615 static int lustre_setxattr(struct dentry *dentry, const char *name,
1616 const void *value, size_t size, int flags)
1621 static ssize_t lustre_listxattr(struct dentry *d_entry, char *name,
1627 static const struct inode_operations server_inode_operations = {
1628 .setxattr = lustre_setxattr,
1629 .getxattr = lustre_getxattr,
1630 .listxattr = lustre_listxattr,
1633 #define log2(n) ffz(~(n))
1634 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1636 static int server_fill_super_common(struct super_block *sb)
1638 struct inode *root = NULL;
1641 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1643 sb->s_blocksize = 4096;
1644 sb->s_blocksize_bits = log2(sb->s_blocksize);
1645 sb->s_magic = LUSTRE_SUPER_MAGIC;
1646 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1647 sb->s_flags |= MS_RDONLY;
1648 sb->s_op = &server_ops;
1650 root = new_inode(sb);
1652 CERROR("Can't make root inode\n");
1656 /* returns -EIO for every operation */
1657 /* make_bad_inode(root); -- badness - can't umount */
1658 /* apparently we need to be a directory for the mount to finish */
1659 root->i_mode = S_IFDIR;
1660 root->i_op = &server_inode_operations;
1661 sb->s_root = d_make_root(root);
1663 CERROR("%s: can't make root dentry\n", sb->s_id);
1670 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
1672 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1673 struct obd_device *obd;
1674 struct dt_device_param p;
1680 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
1681 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
1683 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
1684 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
1685 strcat(lsi->lsi_osd_uuid, "_UUID");
1686 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
1688 obd = class_name2obd(lsi->lsi_osd_obdname);
1690 rc = lustre_start_simple(lsi->lsi_osd_obdname,
1692 lsi->lsi_osd_uuid, lmd->lmd_dev,
1693 flagstr, lsi->lsi_lmd->lmd_opts,
1697 obd = class_name2obd(lsi->lsi_osd_obdname);
1700 CDEBUG(D_MOUNT, "%s already started\n", lsi->lsi_osd_obdname);
1701 /* but continue setup to allow special case of MDT and internal
1702 * MGT being started separately. */
1703 if (!((IS_MGS(lsi) && (lsi->lsi_lmd->lmd_flags &
1705 (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags &
1710 rc = obd_connect(NULL, &lsi->lsi_osd_exp,
1711 obd, &obd->obd_uuid, NULL, NULL);
1713 OBD_FAIL_TIMEOUT(OBD_FAIL_TGT_DELAY_CONNECT, 10);
1717 class_manual_cleanup(obd);
1718 lsi->lsi_dt_dev = NULL;
1722 LASSERT(obd->obd_lu_dev);
1723 lu_device_get(obd->obd_lu_dev);
1724 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
1725 LASSERT(lsi->lsi_dt_dev);
1727 /* set disk context for llog usage */
1728 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1729 obd->obd_lvfs_ctxt.dt = lsi->lsi_dt_dev;
1731 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
1736 /** Fill in the superblock info for a Lustre server.
1737 * Mount the device with the correct options.
1738 * Read the on-disk config file.
1739 * Start the services.
1741 int server_fill_super(struct super_block *sb)
1743 struct lustre_sb_info *lsi = s2lsi(sb);
1747 rc = lsi_prepare(lsi);
1751 /* Start low level OSD */
1752 rc = osd_start(lsi, sb->s_flags);
1754 CERROR("Unable to start osd on %s: %d\n",
1755 lsi->lsi_lmd->lmd_dev, rc);
1760 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
1761 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
1763 if (class_name2obd(lsi->lsi_svname)) {
1764 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1765 "running. Double-mount may have compromised"
1766 " the disk journal.\n",
1772 /* Start MGS before MGC */
1773 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1774 rc = server_start_mgs(sb);
1779 /* Start MGC before servers */
1780 rc = lustre_start_mgc(sb);
1784 /* Set up all obd devices for service */
1785 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1786 (IS_OST(lsi) || IS_MDT(lsi))) {
1787 rc = server_start_targets(sb);
1789 CERROR("Unable to start targets: %d\n", rc);
1792 /* FIXME overmount client here, or can we just start a
1793 * client log and client_fill_super on this sb? We
1794 * need to make sure server_put_super gets called too
1795 * - ll_put_super calls lustre_common_put_super; check
1796 * there for LSI_SERVER flag, call s_p_s if so.
1798 * Probably should start client from new thread so we
1799 * can return. Client will not finish until all
1800 * servers are connected. Note - MGS-only server does
1801 * NOT get a client, since there is no lustre fs
1802 * associated - the MGS is for all lustre fs's */
1805 rc = server_fill_super_common(sb);
1811 /* We jump here in case of failure while starting targets or MGS.
1812 * In this case we can't just put @mnt and have to do real cleanup
1813 * with stoping targets, etc. */
1814 server_put_super(sb);
1819 * Calculate timeout value for a target.
1821 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1823 struct lustre_mount_data *lmd;
1827 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
1828 int min = OBD_RECOVERY_TIME_MIN;
1830 LASSERT(IS_SERVER(lsi));
1834 soft = lmd->lmd_recovery_time_soft;
1835 hard = lmd->lmd_recovery_time_hard;
1836 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1837 obd->obd_no_ir = !has_ir;
1841 soft = OBD_RECOVERY_TIME_SOFT;
1843 hard = OBD_RECOVERY_TIME_HARD;
1845 /* target may have ir_factor configured. */
1846 factor = OBD_IR_FACTOR_DEFAULT;
1847 if (obd->obd_recovery_ir_factor)
1848 factor = obd->obd_recovery_ir_factor;
1851 int new_soft = soft;
1852 int new_hard = hard;
1854 /* adjust timeout value by imperative recovery */
1856 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1857 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1859 /* make sure the timeout is not too short */
1860 new_soft = max(min, new_soft);
1861 new_hard = max(new_soft, new_hard);
1863 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1864 "window shrunk from %d-%d down to %d-%d\n",
1865 obd->obd_name, soft, hard, new_soft, new_hard);
1872 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1873 obd->obd_recovery_time_hard = hard;
1874 obd->obd_recovery_ir_factor = factor;