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 struct list_head server_mount_info_list =
68 LIST_HEAD_INIT(server_mount_info_list);
70 static struct lustre_mount_info *server_find_mount(const char *name)
72 struct list_head *tmp;
73 struct lustre_mount_info *lmi;
76 list_for_each(tmp, &server_mount_info_list) {
77 lmi = list_entry(tmp, struct lustre_mount_info,
79 if (strcmp(name, lmi->lmi_name) == 0)
85 /* we must register an obd for a mount before we call the setup routine.
86 *_setup will call lustre_get_mount to get the mnt struct
87 by obd_name, since we can't pass the pointer to setup. */
88 static int server_register_mount(const char *name, struct super_block *sb)
90 struct lustre_mount_info *lmi;
96 OBD_ALLOC(lmi, sizeof(*lmi));
99 OBD_ALLOC(name_cp, strlen(name) + 1);
101 OBD_FREE(lmi, sizeof(*lmi));
104 strcpy(name_cp, name);
106 mutex_lock(&lustre_mount_info_lock);
108 if (server_find_mount(name)) {
109 mutex_unlock(&lustre_mount_info_lock);
110 OBD_FREE(lmi, sizeof(*lmi));
111 OBD_FREE(name_cp, strlen(name) + 1);
112 CERROR("Already registered %s\n", name);
115 lmi->lmi_name = name_cp;
117 list_add(&lmi->lmi_list_chain, &server_mount_info_list);
119 mutex_unlock(&lustre_mount_info_lock);
121 CDEBUG(D_MOUNT, "register mount %p from %s\n", sb, name);
126 /* when an obd no longer needs a mount */
127 static int server_deregister_mount(const char *name)
129 struct lustre_mount_info *lmi;
132 mutex_lock(&lustre_mount_info_lock);
133 lmi = server_find_mount(name);
135 mutex_unlock(&lustre_mount_info_lock);
136 CERROR("%s not registered\n", name);
140 CDEBUG(D_MOUNT, "deregister mount %p from %s\n", lmi->lmi_sb, name);
142 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
143 list_del(&lmi->lmi_list_chain);
144 OBD_FREE(lmi, sizeof(*lmi));
145 mutex_unlock(&lustre_mount_info_lock);
150 /* obd's look up a registered mount using their obdname. This is just
151 for initial obd setup to find the mount struct. It should not be
152 called every time you want to mntget. */
153 struct lustre_mount_info *server_get_mount(const char *name)
155 struct lustre_mount_info *lmi;
156 struct lustre_sb_info *lsi;
159 mutex_lock(&lustre_mount_info_lock);
160 lmi = server_find_mount(name);
161 mutex_unlock(&lustre_mount_info_lock);
163 CERROR("Can't find mount for %s\n", name);
166 lsi = s2lsi(lmi->lmi_sb);
168 atomic_inc(&lsi->lsi_mounts);
170 CDEBUG(D_MOUNT, "get mount %p from %s, refs=%d\n", lmi->lmi_sb,
171 name, atomic_read(&lsi->lsi_mounts));
175 EXPORT_SYMBOL(server_get_mount);
178 * server_put_mount: to be called from obd_cleanup methods
180 * @dereg_mnt: 0 or 1 depending on whether the mount is to be deregistered or
183 * The caller decides whether server_deregister_mount() needs to be called or
184 * not. Calling of server_deregister_mount() does not depend on refcounting on
185 * lsi because we could have say the mgs and mds on the same node and we
186 * unmount the mds, then the ref on the lsi would still be non-zero but we
187 * would still want to deregister the mds mount.
189 int server_put_mount(const char *name, bool dereg_mnt)
191 struct lustre_mount_info *lmi;
192 struct lustre_sb_info *lsi;
195 mutex_lock(&lustre_mount_info_lock);
196 lmi = server_find_mount(name);
197 mutex_unlock(&lustre_mount_info_lock);
199 CERROR("Can't find mount for %s\n", name);
202 lsi = s2lsi(lmi->lmi_sb);
204 CDEBUG(D_MOUNT, "put mount %p from %s, refs=%d\n",
205 lmi->lmi_sb, name, atomic_read(&lsi->lsi_mounts));
207 if (lustre_put_lsi(lmi->lmi_sb))
208 CDEBUG(D_MOUNT, "Last put of mount %p from %s\n",
212 /* this obd should never need the mount again */
213 server_deregister_mount(name);
217 EXPORT_SYMBOL(server_put_mount);
219 /* Set up a MGS to serve startup logs */
220 static int server_start_mgs(struct super_block *sb)
222 struct lustre_sb_info *lsi = s2lsi(sb);
223 struct lustre_mount_info *lmi;
227 /* It is impossible to have more than 1 MGS per node, since
228 MGC wouldn't know which to connect to */
229 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
231 lsi = s2lsi(lmi->lmi_sb);
232 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
237 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
239 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb);
242 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
243 LUSTRE_MGS_OBDNAME, 0, 0,
244 lsi->lsi_osd_obdname, 0);
245 /* server_deregister_mount() is not called previously, for lsi
246 * and other stuff can't be freed cleanly when mgs calls
247 * server_put_mount() in error handling case (see b=17758),
248 * this problem is caused by a bug in mgs_init0, which forgot
249 * calling server_put_mount in error case. */
252 server_deregister_mount(LUSTRE_MGS_OBDNAME);
256 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
257 "Is the 'mgs' module loaded?\n",
258 LUSTRE_MGS_OBDNAME, rc);
262 static int server_stop_mgs(struct super_block *sb)
264 struct obd_device *obd;
268 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
270 /* There better be only one MGS */
271 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
273 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
277 /* The MGS should always stop when we say so */
279 rc = class_manual_cleanup(obd);
283 /* Since there's only one mgc per node, we have to change it's fs to get
284 access to the right disk. */
285 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
287 struct lustre_sb_info *lsi = s2lsi(sb);
291 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
293 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
294 rc = obd_set_info_async(NULL, mgc->obd_self_export,
295 sizeof(KEY_SET_FS), KEY_SET_FS,
296 sizeof(*sb), sb, NULL);
298 CERROR("can't set_fs %d\n", rc);
303 static int server_mgc_clear_fs(struct obd_device *mgc)
308 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
310 rc = obd_set_info_async(NULL, mgc->obd_self_export,
311 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
316 static inline bool is_mdc_device(const char *devname)
320 ptr = strrchr(devname, '-');
321 return ptr != NULL && strcmp(ptr, "-mdc") == 0;
324 static inline bool tgt_is_mdt(const char *tgtname, __u32 *idx)
328 type = server_name2index(tgtname, idx, NULL);
330 return type == LDD_F_SV_TYPE_MDT;
334 * Convert OST/MDT name(fsname-{MDT,OST}xxxx) to a lwp name with the @idx:yyyy
335 * (fsname-MDTyyyy-lwp-{MDT,OST}xxxx)
337 int tgt_name2lwp_name(const char *tgt_name, char *lwp_name, int len, __u32 idx)
344 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
348 rc = server_name2fsname(tgt_name, fsname, &tgt);
350 CERROR("%s: failed to get fsname from tgt_name: rc = %d\n",
355 if (*tgt != '-' && *tgt != ':') {
356 CERROR("%s: invalid tgt_name name!\n", tgt_name);
357 GOTO(cleanup, rc = -EINVAL);
361 if (strncmp(tgt, "OST", 3) != 0 && strncmp(tgt, "MDT", 3) != 0) {
362 CERROR("%s is not an OST or MDT target!\n", tgt_name);
363 GOTO(cleanup, rc = -EINVAL);
365 snprintf(lwp_name, len, "%s-MDT%04x-%s-%s",
366 fsname, idx, LUSTRE_LWP_NAME, tgt);
368 GOTO(cleanup, rc = 0);
372 OBD_FREE(fsname, MTI_NAME_MAXLEN);
376 EXPORT_SYMBOL(tgt_name2lwp_name);
378 static struct list_head lwp_register_list =
379 LIST_HEAD_INIT(lwp_register_list);
380 DEFINE_MUTEX(lwp_register_list_lock);
382 int lustre_register_lwp_item(const char *lwpname, struct obd_export **exp,
383 register_lwp_cb cb_func, void *cb_data)
385 struct obd_device *lwp;
386 struct lwp_register_item *lri;
389 LASSERTF(strlen(lwpname) < MTI_NAME_MAXLEN, "lwpname is too long %s\n",
391 LASSERT(exp != NULL && *exp == NULL);
397 mutex_lock(&lwp_register_list_lock);
399 lwp = class_name2obd(lwpname);
400 if (lwp != NULL && lwp->obd_set_up == 1) {
401 struct obd_uuid *uuid;
405 mutex_unlock(&lwp_register_list_lock);
409 memcpy(uuid->uuid, lwpname, strlen(lwpname));
410 *exp = cfs_hash_lookup(lwp->obd_uuid_hash, uuid);
414 memcpy(lri->lri_name, lwpname, strlen(lwpname));
416 lri->lri_cb_func = cb_func;
417 lri->lri_cb_data = cb_data;
418 INIT_LIST_HEAD(&lri->lri_list);
419 list_add(&lri->lri_list, &lwp_register_list);
421 if (*exp != NULL && cb_func != NULL)
424 mutex_unlock(&lwp_register_list_lock);
427 EXPORT_SYMBOL(lustre_register_lwp_item);
429 void lustre_deregister_lwp_item(struct obd_export **exp)
431 struct lwp_register_item *lri, *tmp;
433 mutex_lock(&lwp_register_list_lock);
434 list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
435 if (exp == lri->lri_exp) {
437 class_export_put(*exp);
438 list_del(&lri->lri_list);
443 mutex_unlock(&lwp_register_list_lock);
445 EXPORT_SYMBOL(lustre_deregister_lwp_item);
447 struct obd_export *lustre_find_lwp_by_index(const char *dev, __u32 idx)
449 struct lustre_mount_info *lmi;
450 struct lustre_sb_info *lsi;
451 struct obd_device *lwp;
452 struct obd_export *exp = NULL;
457 lmi = server_get_mount(dev);
461 lsi = s2lsi(lmi->lmi_sb);
462 rc = server_name2fsname(lsi->lsi_svname, fsname, NULL);
464 CERROR("%s: failed to get fsname: rc = %d\n",
465 lsi->lsi_svname, rc);
469 snprintf(lwp_name, sizeof(lwp_name), "%s-MDT%04x", fsname, idx);
470 spin_lock(&lsi->lsi_lwp_lock);
471 list_for_each_entry(lwp, &lsi->lsi_lwp_list, obd_lwp_list) {
472 char *ptr = strstr(lwp->obd_name, lwp_name);
475 exp = class_export_get(lwp->obd_lwp_export);
479 spin_unlock(&lsi->lsi_lwp_lock);
482 server_put_mount(dev, false);
486 EXPORT_SYMBOL(lustre_find_lwp_by_index);
488 static void lustre_notify_lwp_list(struct obd_export *exp)
490 struct lwp_register_item *lri, *tmp;
491 LASSERT(exp != NULL);
493 mutex_lock(&lwp_register_list_lock);
494 list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
495 if (strcmp(exp->exp_obd->obd_name, lri->lri_name))
497 if (*lri->lri_exp != NULL)
499 *lri->lri_exp = class_export_get(exp);
500 if (lri->lri_cb_func != NULL)
501 lri->lri_cb_func(lri->lri_cb_data);
503 mutex_unlock(&lwp_register_list_lock);
506 static int lustre_lwp_connect(struct obd_device *lwp)
509 struct lu_context session_ctx;
510 struct obd_export *exp;
511 struct obd_uuid *uuid = NULL;
512 struct obd_connect_data *data = NULL;
516 /* log has been fully processed, let clients connect */
517 rc = lu_env_init(&env, lwp->obd_lu_dev->ld_type->ldt_ctx_tags);
521 lu_context_init(&session_ctx, LCT_SERVER_SESSION);
522 session_ctx.lc_thread = NULL;
523 lu_context_enter(&session_ctx);
524 env.le_ses = &session_ctx;
528 GOTO(out, rc = -ENOMEM);
530 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
531 data->ocd_version = LUSTRE_VERSION_CODE;
532 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
533 OBD_CONNECT_AT | OBD_CONNECT_LRU_RESIZE |
534 OBD_CONNECT_FULL20 | OBD_CONNECT_LVB_TYPE |
535 OBD_CONNECT_LIGHTWEIGHT | OBD_CONNECT_LFSCK;
538 GOTO(out, rc = -ENOMEM);
540 if (strlen(lwp->obd_name) > sizeof(uuid->uuid)) {
541 CERROR("%s: Too long lwp name %s, max_size is %d\n",
542 lwp->obd_name, lwp->obd_name, (int)sizeof(uuid->uuid));
543 GOTO(out, rc = -EINVAL);
546 /* Use lwp name as the uuid, so we find the export by lwp name later */
547 memcpy(uuid->uuid, lwp->obd_name, strlen(lwp->obd_name));
548 rc = obd_connect(&env, &exp, lwp, uuid, data, NULL);
550 CERROR("%s: connect failed: rc = %d\n", lwp->obd_name, rc);
552 if (unlikely(lwp->obd_lwp_export != NULL))
553 class_export_put(lwp->obd_lwp_export);
554 lwp->obd_lwp_export = class_export_get(exp);
555 lustre_notify_lwp_list(exp);
567 lu_context_exit(&session_ctx);
568 lu_context_fini(&session_ctx);
574 * lwp is used by slaves (Non-MDT0 targets) to manage the connection to MDT0,
575 * or from the OSTx to MDTy.
577 static int lustre_lwp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi,
580 struct obd_device *obd;
581 char *lwpname = NULL;
582 char *lwpuuid = NULL;
586 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
589 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
593 OBD_ALLOC(lwpname, MTI_NAME_MAXLEN);
595 GOTO(out, rc = -ENOMEM);
597 rc = tgt_name2lwp_name(lsi->lsi_svname, lwpname, MTI_NAME_MAXLEN, idx);
599 CERROR("%s: failed to generate lwp name: rc = %d\n",
600 lsi->lsi_svname, rc);
604 OBD_ALLOC(lwpuuid, MTI_NAME_MAXLEN);
606 GOTO(out, rc = -ENOMEM);
608 sprintf(lwpuuid, "%s_UUID", lwpname);
609 rc = lustre_start_simple(lwpname, LUSTRE_LWP_NAME,
610 lwpuuid, lustre_cfg_string(lcfg, 1),
613 CERROR("%s: setup up failed: rc %d\n", lwpname, rc);
617 obd = class_name2obd(lwpname);
618 LASSERT(obd != NULL);
620 rc = lustre_lwp_connect(obd);
622 obd->u.cli.cl_max_mds_easize = MAX_MD_SIZE;
623 spin_lock(&lsi->lsi_lwp_lock);
624 list_add_tail(&obd->obd_lwp_list, &lsi->lsi_lwp_list);
625 spin_unlock(&lsi->lsi_lwp_lock);
627 CERROR("%s: connect failed: rc = %d\n", lwpname, rc);
634 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
636 OBD_FREE(lwpuuid, MTI_NAME_MAXLEN);
641 /* the caller is responsible for memory free */
642 static struct obd_device *lustre_find_lwp(struct lustre_sb_info *lsi,
643 char **lwpname, __u32 idx)
645 struct obd_device *lwp;
649 LASSERT(lwpname != NULL);
650 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
652 OBD_ALLOC(*lwpname, MTI_NAME_MAXLEN);
653 if (*lwpname == NULL)
654 RETURN(ERR_PTR(-ENOMEM));
656 rc = tgt_name2lwp_name(lsi->lsi_svname, *lwpname, MTI_NAME_MAXLEN, idx);
658 CERROR("%s: failed to generate lwp name: rc = %d\n",
659 lsi->lsi_svname, rc);
660 GOTO(out, rc = -EINVAL);
663 lwp = class_name2obd(*lwpname);
667 if (*lwpname != NULL) {
668 OBD_FREE(*lwpname, MTI_NAME_MAXLEN);
674 RETURN(lwp != NULL ? lwp : ERR_PTR(-ENOENT));
677 static int lustre_lwp_add_conn(struct lustre_cfg *cfg,
678 struct lustre_sb_info *lsi, __u32 idx)
680 struct lustre_cfg_bufs *bufs = NULL;
681 struct lustre_cfg *lcfg = NULL;
682 char *lwpname = NULL;
683 struct obd_device *lwp;
687 lwp = lustre_find_lwp(lsi, &lwpname, idx);
689 CERROR("%s: can't find lwp device.\n", lsi->lsi_svname);
690 GOTO(out, rc = PTR_ERR(lwp));
692 LASSERT(lwpname != NULL);
696 GOTO(out, rc = -ENOMEM);
698 lustre_cfg_bufs_reset(bufs, lwpname);
699 lustre_cfg_bufs_set_string(bufs, 1,
700 lustre_cfg_string(cfg, 1));
702 lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
704 GOTO(out_cfg, rc = -ENOMEM);
705 rc = class_add_conn(lwp, lcfg);
707 CERROR("%s: can't add conn: rc = %d\n", lwpname, rc);
710 lustre_cfg_free(lcfg);
716 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
721 * Retrieve MDT nids from the client log, then start the lwp device.
722 * there are only two scenarios which would include mdt nid.
724 * marker 5 (flags=0x01, v2.1.54.0) lustre-MDTyyyy 'add mdc' xxx-
725 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
726 * attach 0:lustre-MDTyyyy-mdc 1:mdc 2:lustre-clilmv_UUID
727 * setup 0:lustre-MDTyyyy-mdc 1:lustre-MDTyyyy_UUID 2:192.168.122.162@tcp
728 * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
729 * add_conn 0:lustre-MDTyyyy-mdc 1:192.168.172.1@tcp
730 * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDTyyyy_UUID xxxx
731 * marker 5 (flags=0x02, v2.1.54.0) lustre-MDTyyyy 'add mdc' xxxx-
733 * marker 7 (flags=0x01, v2.1.54.0) lustre-MDTyyyy 'add failnid' xxxx-
734 * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
735 * add_conn 0:lustre-MDTyyyy-mdc 1:192.168.122.2@tcp
736 * marker 7 (flags=0x02, v2.1.54.0) lustre-MDTyyyy 'add failnid' xxxx-
738 static int client_lwp_config_process(const struct lu_env *env,
739 struct llog_handle *handle,
740 struct llog_rec_hdr *rec, void *data)
742 struct config_llog_instance *clli = data;
743 int cfg_len = rec->lrh_len;
744 char *cfg_buf = (char *) (rec + 1);
745 struct lustre_cfg *lcfg = NULL;
746 struct lustre_sb_info *lsi;
747 int rc = 0, swab = 0;
750 if (rec->lrh_type != OBD_CFG_REC) {
751 CERROR("Unknown llog record type %#x encountered\n",
756 LASSERT(clli->cfg_sb != NULL);
757 lsi = s2lsi(clli->cfg_sb);
759 lcfg = (struct lustre_cfg *)cfg_buf;
760 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
761 lustre_swab_lustre_cfg(lcfg);
765 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
769 switch (lcfg->lcfg_command) {
771 struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
773 lustre_swab_cfg_marker(marker, swab,
774 LUSTRE_CFG_BUFLEN(lcfg, 1));
775 if (marker->cm_flags & CM_SKIP ||
776 marker->cm_flags & CM_EXCLUDE)
779 if (!tgt_is_mdt(marker->cm_tgtname, &clli->cfg_lwp_idx))
782 if (IS_MDT(lsi) && clli->cfg_lwp_idx != 0)
785 if (!strncmp(marker->cm_comment, "add mdc", 7) ||
786 !strncmp(marker->cm_comment, "add failnid", 11)) {
787 if (marker->cm_flags & CM_START) {
788 clli->cfg_flags = CFG_F_MARKER;
789 /* This hack is to differentiate the
790 * ADD_UUID is come from "add mdc" record
791 * or from "add failnid" record. */
792 if (!strncmp(marker->cm_comment,
794 clli->cfg_flags |= CFG_F_SKIP;
795 } else if (marker->cm_flags & CM_END) {
801 case LCFG_ADD_UUID: {
802 if (clli->cfg_flags == CFG_F_MARKER) {
803 rc = lustre_lwp_setup(lcfg, lsi, clli->cfg_lwp_idx);
804 /* XXX: process only the first nid as
805 * we don't need another instance of lwp */
806 clli->cfg_flags |= CFG_F_SKIP;
807 } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
808 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
811 CERROR("%s: Fail to add uuid, rc:%d\n",
812 lsi->lsi_svname, rc);
816 case LCFG_ADD_CONN: {
817 char *devname = lustre_cfg_string(lcfg, 0);
821 if (!is_mdc_device(devname))
824 ptr = strrchr(devname, '-');
829 if (!tgt_is_mdt(devname, &idx)) {
835 if (IS_MDT(lsi) && idx != 0)
838 rc = lustre_lwp_add_conn(lcfg, lsi, idx);
848 static int lustre_disconnect_lwp(struct super_block *sb)
850 struct lustre_sb_info *lsi = s2lsi(sb);
851 struct obd_device *lwp;
852 char *logname = NULL;
853 struct lustre_cfg_bufs *bufs = NULL;
854 struct config_llog_instance *cfg = NULL;
859 if (likely(lsi->lsi_lwp_started)) {
860 OBD_ALLOC(logname, MTI_NAME_MAXLEN);
864 rc = server_name2fsname(lsi->lsi_svname, logname, NULL);
866 CERROR("%s: failed to get fsname from svname: "
867 "rc = %d\n", lsi->lsi_svname, rc);
868 GOTO(out, rc = -EINVAL);
871 strcat(logname, "-client");
874 GOTO(out, rc = -ENOMEM);
877 cfg->cfg_instance = sb;
878 rc = lustre_end_log(sb, logname, cfg);
882 lsi->lsi_lwp_started = 0;
887 GOTO(out, rc = -ENOMEM);
889 list_for_each_entry(lwp, &lsi->lsi_lwp_list, obd_lwp_list) {
890 struct lustre_cfg *lcfg;
892 if (likely(lwp->obd_lwp_export != NULL)) {
893 class_export_put(lwp->obd_lwp_export);
894 lwp->obd_lwp_export = NULL;
897 lustre_cfg_bufs_reset(bufs, lwp->obd_name);
898 lustre_cfg_bufs_set_string(bufs, 1, NULL);
899 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
901 GOTO(out, rc = -ENOMEM);
903 /* Disconnect import first. NULL is passed for the '@env',
904 * since it will not be used. */
905 rc = lwp->obd_lu_dev->ld_ops->ldo_process_config(NULL,
906 lwp->obd_lu_dev, lcfg);
907 lustre_cfg_free(lcfg);
908 if (rc != 0 && rc != -ETIMEDOUT) {
909 CERROR("%s: fail to disconnect LWP: rc = %d\n",
923 OBD_FREE(logname, MTI_NAME_MAXLEN);
925 return rc1 != 0 ? rc1 : rc;
929 * Stop the lwp for an OST/MDT target.
931 static int lustre_stop_lwp(struct super_block *sb)
933 struct lustre_sb_info *lsi = s2lsi(sb);
934 struct obd_device *lwp;
939 while (!list_empty(&lsi->lsi_lwp_list)) {
940 lwp = list_entry(lsi->lsi_lwp_list.next, struct obd_device,
942 list_del_init(&lwp->obd_lwp_list);
944 rc = class_manual_cleanup(lwp);
946 CERROR("%s: fail to stop LWP: rc = %d\n",
952 RETURN(rc1 != 0 ? rc1 : rc);
956 * Start the lwp(fsname-MDTyyyy-lwp-{MDT,OST}xxxx) for a MDT/OST or MDT target.
958 static int lustre_start_lwp(struct super_block *sb)
960 struct lustre_sb_info *lsi = s2lsi(sb);
961 struct config_llog_instance *cfg = NULL;
966 if (unlikely(lsi->lsi_lwp_started))
969 OBD_ALLOC(logname, MTI_NAME_MAXLEN);
973 rc = server_name2fsname(lsi->lsi_svname, logname, NULL);
975 CERROR("%s: failed to get fsname from svname: rc = %d\n",
976 lsi->lsi_svname, rc);
977 GOTO(out, rc = -EINVAL);
980 strcat(logname, "-client");
983 GOTO(out, rc = -ENOMEM);
985 cfg->cfg_callback = client_lwp_config_process;
986 cfg->cfg_instance = sb;
987 rc = lustre_process_log(sb, logname, cfg);
989 lsi->lsi_lwp_started = 1;
994 OBD_FREE(logname, MTI_NAME_MAXLEN);
1001 DEFINE_MUTEX(server_start_lock);
1003 /* Stop MDS/OSS if nobody is using them */
1004 static int server_stop_servers(int lsiflags)
1006 struct obd_device *obd = NULL;
1007 struct obd_type *type = NULL;
1011 mutex_lock(&server_start_lock);
1013 /* Either an MDT or an OST or neither */
1014 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
1015 if (lsiflags & LDD_F_SV_TYPE_MDT) {
1016 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1018 type = class_search_type(LUSTRE_MDT_NAME);
1021 /* if this was an OST, and there are no more OST's, clean up the OSS */
1022 if (lsiflags & LDD_F_SV_TYPE_OST) {
1023 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1025 type = class_search_type(LUSTRE_OST_NAME);
1028 if (obd != NULL && (type == NULL || type->typ_refcnt == 0)) {
1030 /* obd_fail doesn't mean much on a server obd */
1031 rc = class_manual_cleanup(obd);
1034 mutex_unlock(&server_start_lock);
1039 int server_mti_print(const char *title, struct mgs_target_info *mti)
1041 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
1042 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
1043 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
1044 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
1045 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
1046 mti->mti_config_ver, mti->mti_flags);
1049 EXPORT_SYMBOL(server_mti_print);
1051 /* Generate data for registration */
1052 static int server_lsi2mti(struct lustre_sb_info *lsi,
1053 struct mgs_target_info *mti)
1055 lnet_process_id_t id;
1060 if (!IS_SERVER(lsi))
1063 if (strlcpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname))
1064 >= sizeof(mti->mti_svname))
1067 mti->mti_nid_count = 0;
1068 while (LNetGetId(i++, &id) != -ENOENT) {
1069 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1072 /* server use --servicenode param, only allow specified
1073 * nids be registered */
1074 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
1075 class_match_nid(lsi->lsi_lmd->lmd_params,
1076 PARAM_FAILNODE, id.nid) < 1)
1079 /* match specified network */
1080 if (!class_match_net(lsi->lsi_lmd->lmd_params,
1081 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1084 mti->mti_nids[mti->mti_nid_count] = id.nid;
1085 mti->mti_nid_count++;
1086 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1087 CWARN("Only using first %d nids for %s\n",
1088 mti->mti_nid_count, mti->mti_svname);
1093 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1094 mti->mti_config_ver = 0;
1096 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1100 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1103 /* Orion requires index to be set */
1104 LASSERT(!(rc & LDD_F_NEED_INDEX));
1105 /* keep only LDD flags */
1106 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1107 if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN))
1108 mti->mti_flags |= LDD_F_UPDATE;
1109 cplen = strlcpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1110 sizeof(mti->mti_params));
1111 if (cplen >= sizeof(mti->mti_params))
1116 /* Register an old or new target with the MGS. If needed MGS will construct
1117 startup logs and assign index */
1118 static int server_register_target(struct lustre_sb_info *lsi)
1120 struct obd_device *mgc = lsi->lsi_mgc;
1121 struct mgs_target_info *mti = NULL;
1128 if (!IS_SERVER(lsi))
1135 rc = server_lsi2mti(lsi, mti);
1139 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1140 mti->mti_svname, mti->mti_fsname,
1141 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1144 /* if write_conf is true, the registration must succeed */
1145 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1146 mti->mti_flags |= LDD_F_OPC_REG;
1148 /* Register the target */
1149 /* FIXME use mgc_process_config instead */
1150 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1151 sizeof(KEY_REGISTER_TARGET),
1152 KEY_REGISTER_TARGET,
1153 sizeof(*mti), mti, NULL);
1155 if (mti->mti_flags & LDD_F_ERROR) {
1156 LCONSOLE_ERROR_MSG(0x160,
1157 "%s: the MGS refuses to allow this server "
1158 "to start: rc = %d. Please see messages on "
1159 "the MGS.\n", lsi->lsi_svname, rc);
1160 } else if (writeconf) {
1161 LCONSOLE_ERROR_MSG(0x15f,
1162 "%s: cannot register this server with the MGS: "
1163 "rc = %d. Is the MGS running?\n",
1164 lsi->lsi_svname, rc);
1166 CERROR("%s: error registering with the MGS: rc = %d "
1167 "(not fatal)\n", lsi->lsi_svname, rc);
1168 /* reset the error code for non-fatal error. */
1181 * Notify the MGS that this target is ready.
1182 * Used by IR - if the MGS receives this message, it will notify clients.
1184 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1186 struct lustre_sb_info *lsi = s2lsi(sb);
1187 struct obd_device *mgc = lsi->lsi_mgc;
1188 struct mgs_target_info *mti = NULL;
1194 if (!(IS_SERVER(lsi)))
1200 rc = server_lsi2mti(lsi, mti);
1204 mti->mti_instance = obd->u.obt.obt_instance;
1205 mti->mti_flags |= LDD_F_OPC_READY;
1207 /* FIXME use mgc_process_config instead */
1208 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1209 sizeof(KEY_REGISTER_TARGET),
1210 KEY_REGISTER_TARGET,
1211 sizeof(*mti), mti, NULL);
1213 /* Imperative recovery: if the mgs informs us to use IR? */
1214 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1215 (mti->mti_flags & LDD_F_IR_CAPABLE))
1216 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1225 /** Start server targets: MDTs and OSTs
1227 static int server_start_targets(struct super_block *sb)
1229 struct obd_device *obd;
1230 struct lustre_sb_info *lsi = s2lsi(sb);
1231 struct config_llog_instance cfg;
1233 struct lu_device *dev;
1237 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1240 /* make sure the MDS is started */
1241 mutex_lock(&server_start_lock);
1242 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1244 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1246 LUSTRE_MDS_OBDNAME"_uuid",
1249 mutex_unlock(&server_start_lock);
1250 CERROR("failed to start MDS: %d\n", rc);
1254 mutex_unlock(&server_start_lock);
1257 /* If we're an OST, make sure the global OSS is running */
1259 /* make sure OSS is started */
1260 mutex_lock(&server_start_lock);
1261 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1263 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1265 LUSTRE_OSS_OBDNAME"_uuid",
1268 mutex_unlock(&server_start_lock);
1269 CERROR("failed to start OSS: %d\n", rc);
1273 mutex_unlock(&server_start_lock);
1276 /* Set the mgc fs to our server disk. This allows the MGC to
1277 * read and write configs locally, in case it can't talk to the MGS. */
1278 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1280 GOTO(out_stop_service, rc);
1282 /* Register with MGS */
1283 rc = server_register_target(lsi);
1287 /* Let the target look up the mount using the target's name
1288 (we can't pass the sb or mnt through class_process_config.) */
1289 rc = server_register_mount(lsi->lsi_svname, sb);
1293 /* Start targets using the llog named for the target */
1294 memset(&cfg, 0, sizeof(cfg));
1295 cfg.cfg_callback = class_config_llog_handler;
1296 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1298 CERROR("failed to start server %s: %d\n",
1299 lsi->lsi_svname, rc);
1300 /* Do NOT call server_deregister_mount() here. This makes it
1301 * impossible to find mount later in cleanup time and leaves
1302 * @lsi and othder stuff leaked. -umka */
1306 obd = class_name2obd(lsi->lsi_svname);
1308 CERROR("no server named %s was started\n", lsi->lsi_svname);
1309 GOTO(out_mgc, rc = -ENXIO);
1312 if (IS_OST(lsi) || IS_MDT(lsi)) {
1313 rc = lustre_start_lwp(sb);
1315 CERROR("%s: failed to start LWP: %d\n",
1316 lsi->lsi_svname, rc);
1321 server_notify_target(sb, obd);
1323 /* calculate recovery timeout, do it after lustre_process_log */
1324 server_calc_timeout(lsi, obd);
1326 /* log has been fully processed */
1327 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1329 /* log has been fully processed, let clients connect */
1330 dev = obd->obd_lu_dev;
1331 if (dev && dev->ld_ops->ldo_prepare) {
1332 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1334 struct lu_context session_ctx;
1336 lu_context_init(&session_ctx, LCT_SERVER_SESSION);
1337 session_ctx.lc_thread = NULL;
1338 lu_context_enter(&session_ctx);
1339 env.le_ses = &session_ctx;
1341 rc = dev->ld_ops->ldo_prepare(&env, NULL, dev);
1344 lu_context_exit(&session_ctx);
1345 lu_context_fini(&session_ctx);
1349 /* abort recovery only on the complete stack:
1350 * many devices can be involved */
1351 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1352 (OBP(obd, iocontrol))) {
1353 obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
1358 /* Release the mgc fs for others to use */
1359 server_mgc_clear_fs(lsi->lsi_mgc);
1363 server_stop_servers(lsi->lsi_flags);
1368 static int lsi_prepare(struct lustre_sb_info *lsi)
1375 LASSERT(lsi->lsi_lmd);
1377 /* The server name is given as a mount line option */
1378 if (lsi->lsi_lmd->lmd_profile == NULL) {
1379 LCONSOLE_ERROR("Can't determine server name\n");
1383 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
1384 RETURN(-ENAMETOOLONG);
1386 strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
1388 /* Determine osd type */
1389 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
1390 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
1391 sizeof(lsi->lsi_osd_type))
1392 RETURN(-ENAMETOOLONG);
1394 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
1396 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
1399 /* XXX: a temp. solution for components using ldiskfs
1400 * to be removed in one of the subsequent patches */
1401 if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs"))
1402 strcpy(lsi->lsi_fstype, "ldiskfs");
1404 strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
1406 /* Determine server type */
1407 rc = server_name2index(lsi->lsi_svname, &index, NULL);
1409 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
1410 /* Assume we're a bare MGS */
1412 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
1414 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
1419 lsi->lsi_flags |= rc;
1421 /* Add mount line flags that used to be in ldd:
1422 * writeconf, mgs, anything else?
1424 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
1425 LDD_F_WRITECONF : 0;
1426 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
1428 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_UPDATE) ?
1430 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
1431 LDD_F_SV_TYPE_MGS : 0;
1432 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
1433 LDD_F_NO_PRIMNODE : 0;
1438 /*************** server mount ******************/
1440 /** Start the shutdown of servers at umount.
1442 static void server_put_super(struct super_block *sb)
1444 struct lustre_sb_info *lsi = s2lsi(sb);
1445 struct obd_device *obd;
1446 char *tmpname, *extraname = NULL;
1448 int lsiflags = lsi->lsi_flags;
1451 LASSERT(IS_SERVER(lsi));
1453 tmpname_sz = strlen(lsi->lsi_svname) + 1;
1454 OBD_ALLOC(tmpname, tmpname_sz);
1455 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
1456 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
1457 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
1458 snprintf(tmpname, tmpname_sz, "MGS");
1460 /* disconnect the lwp first to drain off the inflight request */
1461 if (IS_OST(lsi) || IS_MDT(lsi)) {
1464 rc = lustre_disconnect_lwp(sb);
1465 if (rc != 0 && rc != -ETIMEDOUT &&
1466 rc != -ENOTCONN && rc != -ESHUTDOWN)
1467 CWARN("%s: failed to disconnect lwp: rc= %d\n",
1471 /* Stop the target */
1472 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1473 (IS_MDT(lsi) || IS_OST(lsi))) {
1474 struct lustre_profile *lprof = NULL;
1476 /* tell the mgc to drop the config log */
1477 lustre_end_log(sb, lsi->lsi_svname, NULL);
1479 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
1480 If there are any setup/cleanup errors, save the lov
1481 name for safety cleanup later. */
1482 lprof = class_get_profile(lsi->lsi_svname);
1483 if (lprof && lprof->lp_dt) {
1484 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
1485 strcpy(extraname, lprof->lp_dt);
1488 obd = class_name2obd(lsi->lsi_svname);
1490 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
1491 if (lsiflags & LSI_UMOUNT_FAILOVER)
1493 /* We can't seem to give an error return code
1494 * to .put_super, so we better make sure we clean up! */
1496 class_manual_cleanup(obd);
1498 CERROR("no obd %s\n", lsi->lsi_svname);
1499 server_deregister_mount(lsi->lsi_svname);
1503 /* If they wanted the mgs to stop separately from the mdt, they
1504 should have put it on a different device. */
1506 /* if MDS start with --nomgs, don't stop MGS then */
1507 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
1508 server_stop_mgs(sb);
1511 if (IS_OST(lsi) || IS_MDT(lsi)) {
1512 if (lustre_stop_lwp(sb) < 0)
1513 CERROR("%s: failed to stop lwp!\n", tmpname);
1516 /* Clean the mgc and sb */
1517 lustre_common_put_super(sb);
1519 /* wait till all in-progress cleanups are done
1520 * specifically we're interested in ofd cleanup
1522 obd_zombie_barrier();
1524 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
1525 until the target is really gone so that our type refcount check
1527 server_stop_servers(lsiflags);
1529 /* In case of startup or cleanup err, stop related obds */
1531 obd = class_name2obd(extraname);
1533 CWARN("Cleaning orphaned obd %s\n", extraname);
1535 class_manual_cleanup(obd);
1537 OBD_FREE(extraname, strlen(extraname) + 1);
1540 LCONSOLE_WARN("server umount %s complete\n", tmpname);
1541 OBD_FREE(tmpname, tmpname_sz);
1545 /** Called only for 'umount -f'
1547 static void server_umount_begin(struct super_block *sb)
1549 struct lustre_sb_info *lsi = s2lsi(sb);
1552 CDEBUG(D_MOUNT, "umount -f\n");
1553 /* umount = failover
1555 no third way to do non-force, non-failover */
1556 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
1560 static int server_statfs(struct dentry *dentry, struct kstatfs *buf)
1562 struct super_block *sb = dentry->d_sb;
1563 struct lustre_sb_info *lsi = s2lsi(sb);
1564 struct obd_statfs statfs;
1568 if (lsi->lsi_dt_dev) {
1569 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
1571 statfs_unpack(buf, &statfs);
1572 buf->f_type = sb->s_magic;
1578 buf->f_type = sb->s_magic;
1579 buf->f_bsize = sb->s_blocksize;
1585 buf->f_namelen = NAME_MAX;
1589 /** The operations we support directly on the superblock:
1590 * mount, umount, and df.
1592 static struct super_operations server_ops = {
1593 .put_super = server_put_super,
1594 .umount_begin = server_umount_begin, /* umount -f */
1595 .statfs = server_statfs,
1599 * Xattr support for Lustre servers
1601 static ssize_t lustre_getxattr(struct dentry *dentry, const char *name,
1602 void *buffer, size_t size)
1604 if (!selinux_is_enabled())
1609 static int lustre_setxattr(struct dentry *dentry, const char *name,
1610 const void *value, size_t size, int flags)
1615 static ssize_t lustre_listxattr(struct dentry *d_entry, char *name,
1621 const struct inode_operations server_inode_operations = {
1622 .setxattr = lustre_setxattr,
1623 .getxattr = lustre_getxattr,
1624 .listxattr = lustre_listxattr,
1627 #define log2(n) ffz(~(n))
1628 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
1630 static int server_fill_super_common(struct super_block *sb)
1632 struct inode *root = 0;
1635 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
1637 sb->s_blocksize = 4096;
1638 sb->s_blocksize_bits = log2(sb->s_blocksize);
1639 sb->s_magic = LUSTRE_SUPER_MAGIC;
1640 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
1641 sb->s_flags |= MS_RDONLY;
1642 sb->s_op = &server_ops;
1644 root = new_inode(sb);
1646 CERROR("Can't make root inode\n");
1650 /* returns -EIO for every operation */
1651 /* make_bad_inode(root); -- badness - can't umount */
1652 /* apparently we need to be a directory for the mount to finish */
1653 root->i_mode = S_IFDIR;
1654 root->i_op = &server_inode_operations;
1655 sb->s_root = d_make_root(root);
1657 CERROR("%s: can't make root dentry\n", sb->s_id);
1664 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
1666 struct lustre_mount_data *lmd = lsi->lsi_lmd;
1667 struct obd_device *obd;
1668 struct dt_device_param p;
1674 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
1675 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
1677 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
1678 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
1679 strcat(lsi->lsi_osd_uuid, "_UUID");
1680 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
1682 obd = class_name2obd(lsi->lsi_osd_obdname);
1684 rc = lustre_start_simple(lsi->lsi_osd_obdname,
1686 lsi->lsi_osd_uuid, lmd->lmd_dev,
1687 flagstr, lsi->lsi_lmd->lmd_opts,
1691 obd = class_name2obd(lsi->lsi_osd_obdname);
1695 rc = obd_connect(NULL, &lsi->lsi_osd_exp,
1696 obd, &obd->obd_uuid, NULL, NULL);
1699 class_manual_cleanup(obd);
1700 lsi->lsi_dt_dev = NULL;
1704 LASSERT(obd->obd_lu_dev);
1705 lu_device_get(obd->obd_lu_dev);
1706 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
1707 LASSERT(lsi->lsi_dt_dev);
1709 /* set disk context for llog usage */
1710 OBD_SET_CTXT_MAGIC(&obd->obd_lvfs_ctxt);
1711 obd->obd_lvfs_ctxt.dt = lsi->lsi_dt_dev;
1713 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
1718 /** Fill in the superblock info for a Lustre server.
1719 * Mount the device with the correct options.
1720 * Read the on-disk config file.
1721 * Start the services.
1723 int server_fill_super(struct super_block *sb)
1725 struct lustre_sb_info *lsi = s2lsi(sb);
1729 rc = lsi_prepare(lsi);
1733 /* Start low level OSD */
1734 rc = osd_start(lsi, sb->s_flags);
1736 CERROR("Unable to start osd on %s: %d\n",
1737 lsi->lsi_lmd->lmd_dev, rc);
1742 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
1743 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
1745 if (class_name2obd(lsi->lsi_svname)) {
1746 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
1747 "running. Double-mount may have compromised"
1748 " the disk journal.\n",
1754 /* Start MGS before MGC */
1755 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)) {
1756 rc = server_start_mgs(sb);
1761 /* Start MGC before servers */
1762 rc = lustre_start_mgc(sb);
1766 /* Set up all obd devices for service */
1767 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
1768 (IS_OST(lsi) || IS_MDT(lsi))) {
1769 rc = server_start_targets(sb);
1771 CERROR("Unable to start targets: %d\n", rc);
1774 /* FIXME overmount client here, or can we just start a
1775 * client log and client_fill_super on this sb? We
1776 * need to make sure server_put_super gets called too
1777 * - ll_put_super calls lustre_common_put_super; check
1778 * there for LSI_SERVER flag, call s_p_s if so.
1780 * Probably should start client from new thread so we
1781 * can return. Client will not finish until all
1782 * servers are connected. Note - MGS-only server does
1783 * NOT get a client, since there is no lustre fs
1784 * associated - the MGS is for all lustre fs's */
1787 rc = server_fill_super_common(sb);
1793 /* We jump here in case of failure while starting targets or MGS.
1794 * In this case we can't just put @mnt and have to do real cleanup
1795 * with stoping targets, etc. */
1796 server_put_super(sb);
1801 * Calculate timeout value for a target.
1803 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
1805 struct lustre_mount_data *lmd;
1809 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
1810 int min = OBD_RECOVERY_TIME_MIN;
1812 LASSERT(IS_SERVER(lsi));
1816 soft = lmd->lmd_recovery_time_soft;
1817 hard = lmd->lmd_recovery_time_hard;
1818 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
1819 obd->obd_no_ir = !has_ir;
1823 soft = OBD_RECOVERY_TIME_SOFT;
1825 hard = OBD_RECOVERY_TIME_HARD;
1827 /* target may have ir_factor configured. */
1828 factor = OBD_IR_FACTOR_DEFAULT;
1829 if (obd->obd_recovery_ir_factor)
1830 factor = obd->obd_recovery_ir_factor;
1833 int new_soft = soft;
1834 int new_hard = hard;
1836 /* adjust timeout value by imperative recovery */
1838 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
1839 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
1841 /* make sure the timeout is not too short */
1842 new_soft = max(min, new_soft);
1843 new_hard = max(new_soft, new_hard);
1845 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
1846 "window shrunk from %d-%d down to %d-%d\n",
1847 obd->obd_name, soft, hard, new_soft, new_hard);
1854 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
1855 obd->obd_recovery_time_hard = hard;
1856 obd->obd_recovery_ir_factor = factor;
1858 EXPORT_SYMBOL(server_calc_timeout);