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) 2011, 2012, 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.c
38 * Client/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
51 #include <lustre_fsfilt.h>
52 #include <obd_class.h>
53 #include <lustre/lustre_user.h>
54 #include <linux/version.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>
62 static int (*client_fill_super)(struct super_block *sb,
63 struct vfsmount *mnt) = NULL;
64 static void (*kill_super_cb)(struct super_block *sb) = NULL;
66 /*********** mount lookup *********/
68 DEFINE_MUTEX(lustre_mount_info_lock);
69 static CFS_LIST_HEAD(server_mount_info_list);
71 static struct lustre_mount_info *server_find_mount(const char *name)
74 struct lustre_mount_info *lmi;
77 cfs_list_for_each(tmp, &server_mount_info_list) {
78 lmi = cfs_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,
92 struct lustre_mount_info *lmi;
98 OBD_ALLOC(lmi, sizeof(*lmi));
101 OBD_ALLOC(name_cp, strlen(name) + 1);
103 OBD_FREE(lmi, sizeof(*lmi));
106 strcpy(name_cp, name);
108 mutex_lock(&lustre_mount_info_lock);
110 if (server_find_mount(name)) {
111 mutex_unlock(&lustre_mount_info_lock);
112 OBD_FREE(lmi, sizeof(*lmi));
113 OBD_FREE(name_cp, strlen(name) + 1);
114 CERROR("Already registered %s\n", name);
117 lmi->lmi_name = name_cp;
120 cfs_list_add(&lmi->lmi_list_chain, &server_mount_info_list);
122 mutex_unlock(&lustre_mount_info_lock);
124 CDEBUG(D_MOUNT, "reg_mnt %p from %s\n", lmi->lmi_mnt, name);
129 /* when an obd no longer needs a mount */
130 static int server_deregister_mount(const char *name)
132 struct lustre_mount_info *lmi;
135 mutex_lock(&lustre_mount_info_lock);
136 lmi = server_find_mount(name);
138 mutex_unlock(&lustre_mount_info_lock);
139 CERROR("%s not registered\n", name);
143 CDEBUG(D_MOUNT, "dereg_mnt %p from %s\n", lmi->lmi_mnt, name);
145 OBD_FREE(lmi->lmi_name, strlen(lmi->lmi_name) + 1);
146 cfs_list_del(&lmi->lmi_list_chain);
147 OBD_FREE(lmi, sizeof(*lmi));
148 mutex_unlock(&lustre_mount_info_lock);
153 /* obd's look up a registered mount using their obdname. This is just
154 for initial obd setup to find the mount struct. It should not be
155 called every time you want to mntget. */
156 struct lustre_mount_info *server_get_mount(const char *name)
158 struct lustre_mount_info *lmi;
159 struct lustre_sb_info *lsi;
162 mutex_lock(&lustre_mount_info_lock);
163 lmi = server_find_mount(name);
164 mutex_unlock(&lustre_mount_info_lock);
166 CERROR("Can't find mount for %s\n", name);
169 lsi = s2lsi(lmi->lmi_sb);
171 cfs_atomic_inc(&lsi->lsi_mounts);
173 CDEBUG(D_MOUNT, "get_mnt %p from %s, refs=%d\n", lmi->lmi_mnt,
174 name, cfs_atomic_read(&lsi->lsi_mounts));
178 EXPORT_SYMBOL(server_get_mount);
181 * Used by mdt to get mount_info from obdname.
182 * There are no blocking when using the mount_info.
183 * Do not use server_get_mount for this purpose.
185 struct lustre_mount_info *server_get_mount_2(const char *name)
187 struct lustre_mount_info *lmi;
190 mutex_lock(&lustre_mount_info_lock);
191 lmi = server_find_mount(name);
192 mutex_unlock(&lustre_mount_info_lock);
194 CERROR("Can't find mount for %s\n", name);
198 EXPORT_SYMBOL(server_get_mount_2);
200 static int lustre_put_lsi(struct super_block *sb);
202 /* to be called from obd_cleanup methods */
203 int server_put_mount(const char *name, struct vfsmount *mnt)
205 struct lustre_mount_info *lmi;
206 struct lustre_sb_info *lsi;
209 mutex_lock(&lustre_mount_info_lock);
210 lmi = server_find_mount(name);
211 mutex_unlock(&lustre_mount_info_lock);
213 CERROR("Can't find mount for %s\n", name);
216 lsi = s2lsi(lmi->lmi_sb);
218 CDEBUG(D_MOUNT, "put_mnt %p from %s, refs=%d\n",
219 lmi->lmi_mnt, name, cfs_atomic_read(&lsi->lsi_mounts));
221 if (lustre_put_lsi(lmi->lmi_sb))
222 CDEBUG(D_MOUNT, "Last put of mnt %p from %s\n",
225 /* this obd should never need the mount again */
226 server_deregister_mount(name);
230 EXPORT_SYMBOL(server_put_mount);
232 /* Corresponding to server_get_mount_2 */
233 int server_put_mount_2(const char *name, struct vfsmount *mnt)
238 EXPORT_SYMBOL(server_put_mount_2);
240 /**************** config llog ********************/
242 /** Get a config log from the MGS and process it.
243 * This func is called for both clients and servers.
244 * Continue to process new statements appended to the logs
245 * (whenever the config lock is revoked) until lustre_end_log
247 * @param sb The superblock is used by the MGC to write to the local copy of
249 * @param logname The name of the llog to replicate from the MGS
250 * @param cfg Since the same mgc may be used to follow multiple config logs
251 * (e.g. ost1, ost2, client), the config_llog_instance keeps the state for
252 * this log, and is added to the mgc's list of logs to follow.
254 int lustre_process_log(struct super_block *sb, char *logname,
255 struct config_llog_instance *cfg)
257 struct lustre_cfg *lcfg;
258 struct lustre_cfg_bufs *bufs;
259 struct lustre_sb_info *lsi = s2lsi(sb);
260 struct obd_device *mgc = lsi->lsi_mgc;
271 /* mgc_process_config */
272 lustre_cfg_bufs_reset(bufs, mgc->obd_name);
273 lustre_cfg_bufs_set_string(bufs, 1, logname);
274 lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg));
275 lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb));
276 lcfg = lustre_cfg_new(LCFG_LOG_START, bufs);
277 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
278 lustre_cfg_free(lcfg);
283 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
284 "failed from the MGS (%d). Make sure this "
285 "client and the MGS are running compatible "
286 "versions of Lustre.\n",
287 mgc->obd_name, logname, rc);
290 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
291 "failed (%d). This may be the result of "
292 "communication errors between this node and "
293 "the MGS, a bad configuration, or other "
294 "errors. See the syslog for more "
295 "information.\n", mgc->obd_name, logname,
298 /* class_obd_list(); */
301 EXPORT_SYMBOL(lustre_process_log);
303 /* Stop watching this config log for updates */
304 int lustre_end_log(struct super_block *sb, char *logname,
305 struct config_llog_instance *cfg)
307 struct lustre_cfg *lcfg;
308 struct lustre_cfg_bufs bufs;
309 struct lustre_sb_info *lsi = s2lsi(sb);
310 struct obd_device *mgc = lsi->lsi_mgc;
317 /* mgc_process_config */
318 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
319 lustre_cfg_bufs_set_string(&bufs, 1, logname);
321 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
322 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
323 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
324 lustre_cfg_free(lcfg);
327 EXPORT_SYMBOL(lustre_end_log);
329 /**************** obd start *******************/
331 /** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from
332 * lctl (and do for echo cli/srv.
334 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
335 char *s1, char *s2, char *s3, char *s4)
337 struct lustre_cfg_bufs bufs;
338 struct lustre_cfg * lcfg = NULL;
341 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
342 cmd, s1, s2, s3, s4);
344 lustre_cfg_bufs_reset(&bufs, cfgname);
346 lustre_cfg_bufs_set_string(&bufs, 1, s1);
348 lustre_cfg_bufs_set_string(&bufs, 2, s2);
350 lustre_cfg_bufs_set_string(&bufs, 3, s3);
352 lustre_cfg_bufs_set_string(&bufs, 4, s4);
354 lcfg = lustre_cfg_new(cmd, &bufs);
355 lcfg->lcfg_nid = nid;
356 rc = class_process_config(lcfg);
357 lustre_cfg_free(lcfg);
360 EXPORT_SYMBOL(do_lcfg);
362 /** Call class_attach and class_setup. These methods in turn call
363 * obd type-specific methods.
365 static int lustre_start_simple(char *obdname, char *type, char *uuid,
366 char *s1, char *s2, char *s3, char *s4)
369 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
371 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
373 CERROR("%s attach error %d\n", obdname, rc);
376 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, s3, s4);
378 CERROR("%s setup error %d\n", obdname, rc);
379 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
384 /* Set up a MGS to serve startup logs */
385 static int server_start_mgs(struct super_block *sb)
387 struct lustre_sb_info *lsi = s2lsi(sb);
388 struct vfsmount *mnt = lsi->lsi_srv_mnt;
389 struct lustre_mount_info *lmi;
393 /* It is impossible to have more than 1 MGS per node, since
394 MGC wouldn't know which to connect to */
395 lmi = server_find_mount(LUSTRE_MGS_OBDNAME);
397 lsi = s2lsi(lmi->lmi_sb);
398 LCONSOLE_ERROR_MSG(0x15d, "The MGS service was already started"
403 CDEBUG(D_CONFIG, "Start MGS service %s\n", LUSTRE_MGS_OBDNAME);
405 rc = server_register_mount(LUSTRE_MGS_OBDNAME, sb, mnt);
408 rc = lustre_start_simple(LUSTRE_MGS_OBDNAME, LUSTRE_MGS_NAME,
409 LUSTRE_MGS_OBDNAME, 0, 0,
410 lsi->lsi_osd_obdname, 0);
411 /* Do NOT call server_deregister_mount() here. This leads to
412 * inability cleanup cleanly and free lsi and other stuff when
413 * mgs calls server_put_mount() in error handling case. -umka */
417 LCONSOLE_ERROR_MSG(0x15e, "Failed to start MGS '%s' (%d). "
418 "Is the 'mgs' module loaded?\n",
419 LUSTRE_MGS_OBDNAME, rc);
423 static int server_stop_mgs(struct super_block *sb)
425 struct obd_device *obd;
429 CDEBUG(D_MOUNT, "Stop MGS service %s\n", LUSTRE_MGS_OBDNAME);
431 /* There better be only one MGS */
432 obd = class_name2obd(LUSTRE_MGS_OBDNAME);
434 CDEBUG(D_CONFIG, "mgs %s not running\n", LUSTRE_MGS_OBDNAME);
438 /* The MGS should always stop when we say so */
440 rc = class_manual_cleanup(obd);
444 DEFINE_MUTEX(mgc_start_lock);
446 /** Set up a mgc obd to process startup logs
448 * \param sb [in] super block of the mgc obd
450 * \retval 0 success, otherwise error code
452 static int lustre_start_mgc(struct super_block *sb)
454 struct obd_connect_data *data = NULL;
455 struct lustre_sb_info *lsi = s2lsi(sb);
456 struct obd_device *obd;
457 struct obd_export *exp;
458 struct obd_uuid *uuid;
461 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
464 int rc = 0, i = 0, j, len;
467 LASSERT(lsi->lsi_lmd);
469 /* Find the first non-lo MGS nid for our MGC name */
470 if (IS_SERVER(lsi)) {
471 /* mount -o mgsnode=nid */
472 ptr = lsi->lsi_lmd->lmd_mgs;
473 if (lsi->lsi_lmd->lmd_mgs &&
474 (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) {
476 } else if (IS_MGS(lsi)) {
477 lnet_process_id_t id;
478 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
479 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
486 } else { /* client */
487 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
488 ptr = lsi->lsi_lmd->lmd_dev;
489 if (class_parse_nid(ptr, &nid, &ptr) == 0)
493 CERROR("No valid MGS nids found.\n");
497 mutex_lock(&mgc_start_lock);
499 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
500 OBD_ALLOC(mgcname, len);
501 OBD_ALLOC(niduuid, len + 2);
502 if (!mgcname || !niduuid)
503 GOTO(out_free, rc = -ENOMEM);
504 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
506 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
510 GOTO(out_free, rc = -ENOMEM);
512 obd = class_name2obd(mgcname);
513 if (obd && !obd->obd_stopping) {
514 rc = obd_set_info_async(NULL, obd->obd_self_export,
515 strlen(KEY_MGSSEC), KEY_MGSSEC,
516 strlen(mgssec), mgssec, NULL);
520 /* Re-using an existing MGC */
521 cfs_atomic_inc(&obd->u.cli.cl_mgc_refcount);
523 /* IR compatibility check, only for clients */
524 if (lmd_is_client(lsi->lsi_lmd)) {
526 int vallen = sizeof(*data);
527 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
529 rc = obd_get_info(NULL, obd->obd_self_export,
530 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
531 &vallen, data, NULL);
533 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
534 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
535 /* LMD_FLG_NOIR is for test purpose only */
537 "Trying to mount a client with IR setting "
538 "not compatible with current mgc. "
539 "Force to use current mgc setting that is "
541 has_ir ? "enabled" : "disabled");
543 *flags &= ~LMD_FLG_NOIR;
545 *flags |= LMD_FLG_NOIR;
550 /* If we are restarting the MGS, don't try to keep the MGC's
551 old connection, or registration will fail. */
553 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
557 /* Try all connections, but only once (again).
558 We don't want to block another target from starting
559 (using its local copy of the log), but we do want to connect
560 if at all possible. */
562 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
563 rc = obd_set_info_async(NULL, obd->obd_self_export,
564 sizeof(KEY_INIT_RECOV_BACKUP),
565 KEY_INIT_RECOV_BACKUP,
566 sizeof(recov_bk), &recov_bk, NULL);
570 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
572 /* Add the primary nids for the MGS */
574 sprintf(niduuid, "%s_%x", mgcname, i);
575 if (IS_SERVER(lsi)) {
576 ptr = lsi->lsi_lmd->lmd_mgs;
578 /* Use local nids (including LO) */
579 lnet_process_id_t id;
580 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
581 rc = do_lcfg(mgcname, id.nid,
582 LCFG_ADD_UUID, niduuid, 0,0,0);
585 /* Use mgsnode= nids */
586 /* mount -o mgsnode=nid */
587 if (lsi->lsi_lmd->lmd_mgs) {
588 ptr = lsi->lsi_lmd->lmd_mgs;
589 } else if (class_find_param(ptr, PARAM_MGSNODE,
591 CERROR("No MGS nids given.\n");
592 GOTO(out_free, rc = -EINVAL);
594 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
595 rc = do_lcfg(mgcname, nid,
596 LCFG_ADD_UUID, niduuid, 0,0,0);
600 } else { /* client */
601 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
602 ptr = lsi->lsi_lmd->lmd_dev;
603 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
604 rc = do_lcfg(mgcname, nid,
605 LCFG_ADD_UUID, niduuid, 0,0,0);
607 /* Stop at the first failover nid */
613 CERROR("No valid MGS nids found.\n");
614 GOTO(out_free, rc = -EINVAL);
616 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
618 /* Random uuid for MGC allows easier reconnects */
620 ll_generate_random_uuid(uuidc);
621 class_uuid_unparse(uuidc, uuid);
624 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
625 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
631 /* Add any failover MGS nids */
633 while (ptr && ((*ptr == ':' ||
634 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) {
635 /* New failover node */
636 sprintf(niduuid, "%s_%x", mgcname, i);
638 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
640 rc = do_lcfg(mgcname, nid,
641 LCFG_ADD_UUID, niduuid, 0,0,0);
646 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
654 lsi->lsi_lmd->lmd_mgs_failnodes = i;
656 obd = class_name2obd(mgcname);
658 CERROR("Can't find mgcobd %s\n", mgcname);
659 GOTO(out_free, rc = -ENOTCONN);
662 rc = obd_set_info_async(NULL, obd->obd_self_export,
663 strlen(KEY_MGSSEC), KEY_MGSSEC,
664 strlen(mgssec), mgssec, NULL);
668 /* Keep a refcount of servers/clients who started with "mount",
669 so we know when we can get rid of the mgc. */
670 cfs_atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
672 /* Try all connections, but only once. */
674 rc = obd_set_info_async(NULL, obd->obd_self_export,
675 sizeof(KEY_INIT_RECOV_BACKUP),
676 KEY_INIT_RECOV_BACKUP,
677 sizeof(recov_bk), &recov_bk, NULL);
680 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
682 /* We connect to the MGS at setup, and don't disconnect until cleanup */
683 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
684 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV |
685 OBD_CONNECT_LVB_TYPE;
687 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
688 data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB;
690 #warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
693 if (lmd_is_client(lsi->lsi_lmd) &&
694 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
695 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
696 data->ocd_version = LUSTRE_VERSION_CODE;
697 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
699 CERROR("connect failed %d\n", rc);
703 obd->u.cli.cl_mgc_mgsexp = exp;
706 /* Keep the mgc info in the sb. Note that many lsi's can point
710 mutex_unlock(&mgc_start_lock);
715 OBD_FREE(mgcname, len);
717 OBD_FREE(niduuid, len + 2);
721 static int lustre_stop_mgc(struct super_block *sb)
723 struct lustre_sb_info *lsi = s2lsi(sb);
724 struct obd_device *obd;
725 char *niduuid = 0, *ptr = 0;
726 int i, rc = 0, len = 0;
736 mutex_lock(&mgc_start_lock);
737 LASSERT(cfs_atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
738 if (!cfs_atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
739 /* This is not fatal, every client that stops
740 will call in here. */
741 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
742 cfs_atomic_read(&obd->u.cli.cl_mgc_refcount));
743 GOTO(out, rc = -EBUSY);
746 /* The MGC has no recoverable data in any case.
747 * force shotdown set in umount_begin */
748 obd->obd_no_recov = 1;
750 if (obd->u.cli.cl_mgc_mgsexp) {
751 /* An error is not fatal, if we are unable to send the
752 disconnect mgs ping evictor cleans up the export */
753 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
755 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
758 /* Save the obdname for cleaning the nid uuids, which are
760 len = strlen(obd->obd_name) + 6;
761 OBD_ALLOC(niduuid, len);
763 strcpy(niduuid, obd->obd_name);
764 ptr = niduuid + strlen(niduuid);
767 rc = class_manual_cleanup(obd);
771 /* Clean the nid uuids */
773 GOTO(out, rc = -ENOMEM);
775 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
776 sprintf(ptr, "_%x", i);
777 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
780 CERROR("del MDC UUID %s failed: rc = %d\n",
785 OBD_FREE(niduuid, len);
787 /* class_import_put will get rid of the additional connections */
788 mutex_unlock(&mgc_start_lock);
792 /* Since there's only one mgc per node, we have to change it's fs to get
793 access to the right disk. */
794 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
796 struct lustre_sb_info *lsi = s2lsi(sb);
800 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
802 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
803 rc = obd_set_info_async(NULL, mgc->obd_self_export,
804 sizeof(KEY_SET_FS), KEY_SET_FS,
805 sizeof(*sb), sb, NULL);
807 CERROR("can't set_fs %d\n", rc);
813 static int server_mgc_clear_fs(struct obd_device *mgc)
818 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
820 rc = obd_set_info_async(NULL, mgc->obd_self_export,
821 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
826 /** Get the fsname ("lustre") from the server name ("lustre-OST003F").
827 * @param [in] svname server name including type and index
828 * @param [out] fsname Buffer to copy filesystem name prefix into.
829 * Must have at least 'strlen(fsname) + 1' chars.
830 * @param [out] endptr if endptr isn't NULL it is set to end of fsname
833 int server_name2fsname(char *svname, char *fsname, char **endptr)
835 char *dash = strrchr(svname, '-');
837 dash = strrchr(svname, ':');
842 /* interpret <fsname>-MDTXXXXX-mdc as mdt, the better way is to pass
843 * in the fsname, then determine the server index */
844 if (!strcmp(LUSTRE_MDC_NAME, dash + 1)) {
846 for (; dash > svname && *dash != '-' && *dash != ':'; dash--)
852 if (fsname != NULL) {
853 strncpy(fsname, svname, dash - svname);
854 fsname[dash - svname] = '\0';
862 EXPORT_SYMBOL(server_name2fsname);
864 static int is_mdc_device(char *devname)
867 ptr = strrchr(devname, '-');
868 if (ptr != NULL && strcmp(ptr, "-mdc") == 0)
873 static int inline tgt_is_mdt0(char *tgtname)
878 type = server_name2index(tgtname, &idx, NULL);
879 if (type != LDD_F_SV_TYPE_MDT)
882 return (idx == 0) ? 1 :0;
885 static int inline is_mdc_for_mdt0(char *devname)
889 if (!is_mdc_device(devname))
892 ptr = strrchr(devname, '-');
897 if (tgt_is_mdt0(devname)) {
906 * Convert OST/MDT name(fsname-OSTxxxx) to a lwp name
907 * (fsname-MDT0000-lwp-OSTxxxx)
909 int tgt_name2lwpname(char *svname, char *lwpname)
915 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
919 rc = server_name2fsname(svname, fsname, &tgt);
921 CERROR("%s: failed to get fsname from svname. %d\n",
926 if (*tgt != '-' && *tgt != ':') {
927 CERROR("%s: invalid svname name!\n", svname);
928 GOTO(cleanup, rc = -EINVAL);
932 if (strncmp(tgt, "OST", 3) != 0 && strncmp(tgt, "MDT", 3) != 0) {
933 CERROR("%s is not an OST or MDT target!\n", svname);
934 GOTO(cleanup, rc = -EINVAL);
936 sprintf(lwpname, "%s-MDT0000-%s-%s", fsname, LUSTRE_LWP_NAME, tgt);
939 OBD_FREE(fsname, MTI_NAME_MAXLEN);
942 EXPORT_SYMBOL(tgt_name2lwpname);
944 static CFS_LIST_HEAD(lwp_register_list);
945 DEFINE_MUTEX(lwp_register_list_lock);
947 int lustre_register_lwp_item(char *lwpname, struct obd_export **exp,
948 register_lwp_cb cb_func, void *cb_data)
950 struct obd_device *lwp;
951 struct lwp_register_item *lri;
954 LASSERTF(strlen(lwpname) < MTI_NAME_MAXLEN, "lwpname is too long %s\n",
956 LASSERT(exp != NULL && *exp == NULL);
962 mutex_lock(&lwp_register_list_lock);
964 lwp = class_name2obd(lwpname);
965 if (lwp != NULL && lwp->obd_set_up == 1) {
966 struct obd_uuid *uuid;
970 mutex_unlock(&lwp_register_list_lock);
974 memcpy(uuid->uuid, lwpname, strlen(lwpname));
975 *exp = cfs_hash_lookup(lwp->obd_uuid_hash, uuid);
979 memcpy(lri->lri_name, lwpname, strlen(lwpname));
981 lri->lri_cb_func = cb_func;
982 lri->lri_cb_data = cb_data;
983 CFS_INIT_LIST_HEAD(&lri->lri_list);
984 cfs_list_add(&lri->lri_list, &lwp_register_list);
986 if (*exp != NULL && cb_func != NULL)
989 mutex_unlock(&lwp_register_list_lock);
992 EXPORT_SYMBOL(lustre_register_lwp_item);
994 void lustre_deregister_lwp_item(struct obd_export **exp)
996 struct lwp_register_item *lri, *tmp;
998 mutex_lock(&lwp_register_list_lock);
999 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
1000 if (exp == lri->lri_exp) {
1002 class_export_put(*exp);
1003 cfs_list_del(&lri->lri_list);
1008 mutex_unlock(&lwp_register_list_lock);
1010 EXPORT_SYMBOL(lustre_deregister_lwp_item);
1012 static void lustre_notify_lwp_list(struct obd_export *exp)
1014 struct lwp_register_item *lri, *tmp;
1015 LASSERT(exp != NULL);
1017 mutex_lock(&lwp_register_list_lock);
1018 cfs_list_for_each_entry_safe(lri, tmp, &lwp_register_list, lri_list) {
1019 if (strcmp(exp->exp_obd->obd_name, lri->lri_name))
1021 if (*lri->lri_exp != NULL)
1023 *lri->lri_exp = class_export_get(exp);
1024 if (lri->lri_cb_func != NULL)
1025 lri->lri_cb_func(lri->lri_cb_data);
1027 mutex_unlock(&lwp_register_list_lock);
1030 static int lustre_lwp_connect(struct obd_device *lwp)
1033 struct lu_context session_ctx;
1034 struct obd_export *exp;
1035 struct obd_uuid *uuid = NULL;
1036 struct obd_connect_data *data = NULL;
1040 /* log has been fully processed, let clients connect */
1041 rc = lu_env_init(&env, lwp->obd_lu_dev->ld_type->ldt_ctx_tags);
1045 lu_context_init(&session_ctx, LCT_SESSION);
1046 session_ctx.lc_thread = NULL;
1047 lu_context_enter(&session_ctx);
1048 env.le_ses = &session_ctx;
1050 OBD_ALLOC_PTR(data);
1052 GOTO(out, rc = -ENOMEM);
1054 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
1055 data->ocd_version = LUSTRE_VERSION_CODE;
1056 data->ocd_connect_flags |= OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
1057 OBD_CONNECT_AT | OBD_CONNECT_LRU_RESIZE |
1058 OBD_CONNECT_FULL20 | OBD_CONNECT_LVB_TYPE |
1059 OBD_CONNECT_LIGHTWEIGHT |
1060 OBD_CONNECT_PINGLESS;
1061 OBD_ALLOC_PTR(uuid);
1063 GOTO(out, rc = -ENOMEM);
1065 if (strlen(lwp->obd_name) > sizeof(uuid->uuid)) {
1066 CERROR("%s: Too long lwp name %s, max_size is %d\n",
1067 lwp->obd_name, lwp->obd_name, (int)sizeof(uuid->uuid));
1068 GOTO(out, rc = -EINVAL);
1071 /* Use lwp name as the uuid, so we find the export by lwp name later */
1072 memcpy(uuid->uuid, lwp->obd_name, strlen(lwp->obd_name));
1073 rc = obd_connect(&env, &exp, lwp, uuid, data, NULL);
1075 CERROR("%s: connect failed: rc = %d\n", lwp->obd_name, rc);
1077 lustre_notify_lwp_list(exp);
1086 lu_context_exit(&session_ctx);
1087 lu_context_fini(&session_ctx);
1093 * lwp is used by slaves (Non-MDT0 targets) to manage the connection
1096 static int lustre_lwp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi)
1098 struct obd_connect_data *data = NULL;
1099 struct obd_device *obd;
1100 char *lwpname = NULL;
1101 char *lwpuuid = NULL;
1105 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
1108 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
1112 OBD_ALLOC(lwpname, MTI_NAME_MAXLEN);
1113 if (lwpname == NULL)
1114 GOTO(out, rc = -ENOMEM);
1116 rc = tgt_name2lwpname(lsi->lsi_svname, lwpname);
1118 CERROR("%s: failed to generate lwp name. %d\n",
1119 lsi->lsi_svname, rc);
1123 OBD_ALLOC(lwpuuid, MTI_NAME_MAXLEN);
1124 if (lwpuuid == NULL)
1125 GOTO(out, rc = -ENOMEM);
1127 sprintf(lwpuuid, "%s_UUID", lwpname);
1128 rc = lustre_start_simple(lwpname, LUSTRE_LWP_NAME,
1129 lwpuuid, lustre_cfg_string(lcfg, 1),
1132 CERROR("%s: setup up failed: rc %d\n", lwpname, rc);
1136 obd = class_name2obd(lwpname);
1137 LASSERT(obd != NULL);
1139 rc = lustre_lwp_connect(obd);
1141 CERROR("%s: connect failed: rc = %d\n", lwpname, rc);
1145 if (lwpname != NULL)
1146 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
1147 if (lwpuuid != NULL)
1148 OBD_FREE(lwpuuid, MTI_NAME_MAXLEN);
1153 /* the caller is responsible for memory free */
1154 static struct obd_device *lustre_find_lwp(struct lustre_sb_info *lsi,
1155 char **lwpname, char **logname)
1157 struct obd_device *lwp;
1161 LASSERT(lwpname != NULL);
1162 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
1164 OBD_ALLOC(*lwpname, MTI_NAME_MAXLEN);
1165 if (*lwpname == NULL)
1166 RETURN(ERR_PTR(-ENOMEM));
1168 if (logname != NULL) {
1169 OBD_ALLOC(*logname, MTI_NAME_MAXLEN);
1170 if (*logname == NULL)
1171 GOTO(out, rc = -ENOMEM);
1172 rc = server_name2fsname(lsi->lsi_svname, *lwpname, NULL);
1174 CERROR("%s: failed to get fsname from svname. %d\n",
1175 lsi->lsi_svname, rc);
1176 GOTO(out, rc = -EINVAL);
1178 sprintf(*logname, "%s-client", *lwpname);
1181 rc = tgt_name2lwpname(lsi->lsi_svname, *lwpname);
1183 CERROR("%s: failed to generate lwp name. %d\n",
1184 lsi->lsi_svname, rc);
1185 GOTO(out, rc = -EINVAL);
1188 lwp = class_name2obd(*lwpname);
1192 if (*lwpname != NULL) {
1193 OBD_FREE(*lwpname, MTI_NAME_MAXLEN);
1196 if (logname != NULL && *logname != NULL) {
1197 OBD_FREE(*logname, MTI_NAME_MAXLEN);
1203 RETURN(lwp != NULL ? lwp : ERR_PTR(-ENOENT));
1206 static int lustre_lwp_add_conn(struct lustre_cfg *cfg,
1207 struct lustre_sb_info *lsi)
1209 struct lustre_cfg_bufs *bufs = NULL;
1210 struct lustre_cfg *lcfg = NULL;
1211 char *lwpname = NULL;
1212 struct obd_device *lwp;
1216 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
1218 CERROR("%s: can't find lwp device.\n", lsi->lsi_svname);
1219 GOTO(out, rc = PTR_ERR(lwp));
1221 LASSERT(lwpname != NULL);
1223 OBD_ALLOC_PTR(bufs);
1225 GOTO(out, rc = -ENOMEM);
1227 lustre_cfg_bufs_reset(bufs, lwpname);
1228 lustre_cfg_bufs_set_string(bufs, 1,
1229 lustre_cfg_string(cfg, 1));
1231 lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
1233 rc = class_add_conn(lwp, lcfg);
1235 CERROR("%s: can't add conn: rc = %d\n", lwpname, rc);
1241 lustre_cfg_free(lcfg);
1242 if (lwpname != NULL)
1243 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
1248 * Retrieve MDT nids from the client log, then start the lwp device.
1249 * there are only two scenarios which would include mdt nid.
1251 * marker 5 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add mdc' xxx-
1252 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
1253 * attach 0:lustre-MDT0000-mdc 1:mdc 2:lustre-clilmv_UUID
1254 * setup 0:lustre-MDT0000-mdc 1:lustre-MDT0000_UUID 2:192.168.122.162@tcp
1255 * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
1256 * add_conn 0:lustre-MDT0000-mdc 1:192.168.172.1@tcp
1257 * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID xxxx
1258 * marker 5 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add mdc' xxxx-
1260 * marker 7 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
1261 * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
1262 * add_conn 0:lustre-MDT0000-mdc 1:192.168.122.2@tcp
1263 * marker 7 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
1265 static int client_lwp_config_process(const struct lu_env *env,
1266 struct llog_handle *handle,
1267 struct llog_rec_hdr *rec, void *data)
1269 struct config_llog_instance *clli = data;
1270 int cfg_len = rec->lrh_len;
1271 char *cfg_buf = (char *) (rec + 1);
1272 struct lustre_cfg *lcfg = NULL;
1273 struct lustre_sb_info *lsi;
1274 int rc = 0, swab = 0;
1277 if (rec->lrh_type != OBD_CFG_REC) {
1278 CERROR("Unknown llog record type %#x encountered\n",
1283 LASSERT(clli->cfg_sb != NULL);
1284 lsi = s2lsi(clli->cfg_sb);
1286 lcfg = (struct lustre_cfg *)cfg_buf;
1287 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
1288 lustre_swab_lustre_cfg(lcfg);
1292 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
1296 switch (lcfg->lcfg_command) {
1298 struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
1300 lustre_swab_cfg_marker(marker, swab,
1301 LUSTRE_CFG_BUFLEN(lcfg, 1));
1302 if (marker->cm_flags & CM_SKIP ||
1303 marker->cm_flags & CM_EXCLUDE)
1306 if (!tgt_is_mdt0(marker->cm_tgtname))
1309 if(!strncmp(marker->cm_comment, "add mdc", 7) ||
1310 !strncmp(marker->cm_comment, "add failnid", 11)) {
1311 if (marker->cm_flags & CM_START) {
1312 clli->cfg_flags = CFG_F_MARKER;
1313 /* This hack is to differentiate the
1314 * ADD_UUID is come from "add mdc" record
1315 * or from "add failnid" record. */
1316 if (!strncmp(marker->cm_comment,
1318 clli->cfg_flags |= CFG_F_SKIP;
1319 } else if (marker->cm_flags & CM_END) {
1320 clli->cfg_flags = 0;
1325 case LCFG_ADD_UUID: {
1326 if (clli->cfg_flags == CFG_F_MARKER) {
1327 rc = lustre_lwp_setup(lcfg, lsi);
1328 /* XXX: process only the first nid as
1329 * we don't need another instance of lwp */
1330 clli->cfg_flags |= CFG_F_SKIP;
1331 } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
1332 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
1335 CERROR("%s: Fail to add uuid, rc:%d\n",
1336 lsi->lsi_svname, rc);
1340 case LCFG_ADD_CONN: {
1341 if (is_mdc_for_mdt0(lustre_cfg_string(lcfg, 0)))
1342 rc = lustre_lwp_add_conn(lcfg, lsi);
1352 static int lustre_disconnect_lwp(struct super_block *sb)
1354 struct lustre_sb_info *lsi = s2lsi(sb);
1355 struct obd_device *lwp;
1356 char *lwpname = NULL;
1357 char *logname = NULL;
1358 struct lustre_cfg *lcfg = NULL;
1359 struct lustre_cfg_bufs *bufs = NULL;
1360 struct config_llog_instance *cfg = NULL;
1364 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
1365 if (IS_ERR(lwp) && PTR_ERR(lwp) != -ENOENT)
1366 GOTO(out, rc = PTR_ERR(lwp));
1368 LASSERT(lwpname != NULL);
1369 LASSERT(logname != NULL);
1373 GOTO(out, rc = -ENOMEM);
1376 cfg->cfg_instance = sb;
1377 rc = lustre_end_log(sb, logname, cfg);
1379 CERROR("%s: Can't end config log %s.\n", lwpname, logname);
1383 if (PTR_ERR(lwp) == -ENOENT) {
1384 CDEBUG(D_CONFIG, "%s: lwp device wasn't started.\n",
1389 OBD_ALLOC_PTR(bufs);
1391 GOTO(out, rc = -ENOMEM);
1393 lustre_cfg_bufs_reset(bufs, lwp->obd_name);
1394 lustre_cfg_bufs_set_string(bufs, 1, NULL);
1395 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
1397 GOTO(out, rc = -ENOMEM);
1399 /* Disconnect import first. NULL is passed for the '@env', since
1400 * it will not be used. */
1401 rc = lwp->obd_lu_dev->ld_ops->ldo_process_config(NULL, lwp->obd_lu_dev,
1405 lustre_cfg_free(lcfg);
1411 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
1413 OBD_FREE(logname, MTI_NAME_MAXLEN);
1418 * Stop the lwp for an OST/MDT target.
1420 static int lustre_stop_lwp(struct super_block *sb)
1422 struct lustre_sb_info *lsi = s2lsi(sb);
1423 struct obd_device *lwp = NULL;
1424 char *lwpname = NULL;
1428 lwp = lustre_find_lwp(lsi, &lwpname, NULL);
1430 CDEBUG(PTR_ERR(lwp) == -ENOENT ? D_CONFIG : D_ERROR,
1431 "%s: lwp wasn't started.\n", lsi->lsi_svname);
1436 rc = class_manual_cleanup(lwp);
1439 if (lwpname != NULL)
1440 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
1445 * Start the lwp(fsname-MDT0000-lwp-OSTxxxx) for an OST or MDT target,
1446 * which would be used to establish connection from OST to MDT0.
1448 static int lustre_start_lwp(struct super_block *sb)
1450 struct lustre_sb_info *lsi = s2lsi(sb);
1451 struct config_llog_instance *cfg = NULL;
1452 struct obd_device *lwp;
1453 char *lwpname = NULL;
1454 char *logname = NULL;
1458 lwp = lustre_find_lwp(lsi, &lwpname, &logname);
1460 /* the lwp device already stared */
1461 if (lwp && !IS_ERR(lwp))
1464 if (PTR_ERR(lwp) != -ENOENT)
1465 GOTO(out, rc = PTR_ERR(lwp));
1467 LASSERT(lwpname != NULL);
1468 LASSERT(logname != NULL);
1472 GOTO(out, rc = -ENOMEM);
1474 cfg->cfg_callback = client_lwp_config_process;
1475 cfg->cfg_instance = sb;
1477 rc = lustre_process_log(sb, logname, cfg);
1479 if (lwpname != NULL)
1480 OBD_FREE(lwpname, MTI_NAME_MAXLEN);
1481 if (logname != NULL)
1482 OBD_FREE(logname, MTI_NAME_MAXLEN);
1488 DEFINE_MUTEX(server_start_lock);
1490 /* Stop MDS/OSS if nobody is using them */
1491 static int server_stop_servers(int lsiflags)
1493 struct obd_device *obd = NULL;
1494 struct obd_type *type = NULL;
1498 mutex_lock(&server_start_lock);
1500 /* Either an MDT or an OST or neither */
1501 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
1502 if ((lsiflags & LDD_F_SV_TYPE_MDT) &&
1503 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
1504 type = class_search_type(LUSTRE_MDT_NAME);
1506 /* if this was an OST, and there are no more OST's, clean up the OSS */
1507 if ((lsiflags & LDD_F_SV_TYPE_OST) &&
1508 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
1509 type = class_search_type(LUSTRE_OST_NAME);
1512 if (obd && (!type || !type->typ_refcnt)) {
1515 /* obd_fail doesn't mean much on a server obd */
1516 err = class_manual_cleanup(obd);
1521 mutex_unlock(&server_start_lock);
1526 int server_mti_print(char *title, struct mgs_target_info *mti)
1528 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
1529 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
1530 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
1531 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
1532 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
1533 mti->mti_config_ver, mti->mti_flags);
1538 * Get service name (svname) from string
1540 * if endptr isn't NULL it is set to end of fsname *
1542 int server_name2svname(char *label, char *svname, char **endptr)
1547 /* We use server_name2fsname() just for parsing */
1548 rc = server_name2fsname(label, NULL, &dash);
1555 strncpy(svname, dash + 1, MTI_NAME_MAXLEN);
1559 EXPORT_SYMBOL(server_name2svname);
1562 /* Get the index from the obd name.
1563 rc = server type, or
1565 if endptr isn't NULL it is set to end of name */
1566 int server_name2index(char *svname, __u32 *idx, char **endptr)
1568 unsigned long index;
1572 /* We use server_name2fsname() just for parsing */
1573 rc = server_name2fsname(svname, NULL, &dash);
1582 if (strncmp(dash, "MDT", 3) == 0)
1583 rc = LDD_F_SV_TYPE_MDT;
1584 else if (strncmp(dash, "OST", 3) == 0)
1585 rc = LDD_F_SV_TYPE_OST;
1591 if (strcmp(dash, "all") == 0)
1592 return rc | LDD_F_SV_ALL;
1594 index = simple_strtoul(dash, endptr, 16);
1599 EXPORT_SYMBOL(server_name2index);
1601 /* Generate data for registration */
1602 static int server_lsi2mti(struct lustre_sb_info *lsi,
1603 struct mgs_target_info *mti)
1605 lnet_process_id_t id;
1609 if (!IS_SERVER(lsi))
1612 strncpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname));
1614 mti->mti_nid_count = 0;
1615 while (LNetGetId(i++, &id) != -ENOENT) {
1616 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1619 /* server use --servicenode param, only allow specified
1620 * nids be registered */
1621 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
1622 class_match_nid(lsi->lsi_lmd->lmd_params,
1623 PARAM_FAILNODE, id.nid) < 1)
1626 /* match specified network */
1627 if (!class_match_net(lsi->lsi_lmd->lmd_params,
1628 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1631 mti->mti_nids[mti->mti_nid_count] = id.nid;
1632 mti->mti_nid_count++;
1633 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1634 CWARN("Only using first %d nids for %s\n",
1635 mti->mti_nid_count, mti->mti_svname);
1640 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1641 mti->mti_config_ver = 0;
1643 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1647 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1650 /* Orion requires index to be set */
1651 LASSERT(!(rc & LDD_F_NEED_INDEX));
1652 /* keep only LDD flags */
1653 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1654 if (mti->mti_flags & (LDD_F_WRITECONF | LDD_F_VIRGIN))
1655 mti->mti_flags |= LDD_F_UPDATE;
1656 strncpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1657 sizeof(mti->mti_params));
1661 /* Register an old or new target with the MGS. If needed MGS will construct
1662 startup logs and assign index */
1663 static int server_register_target(struct lustre_sb_info *lsi)
1665 struct obd_device *mgc = lsi->lsi_mgc;
1666 struct mgs_target_info *mti = NULL;
1673 if (!IS_SERVER(lsi))
1680 rc = server_lsi2mti(lsi, mti);
1684 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1685 mti->mti_svname, mti->mti_fsname,
1686 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1689 /* if write_conf is true, the registration must succeed */
1690 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1691 mti->mti_flags |= LDD_F_OPC_REG;
1693 /* Register the target */
1694 /* FIXME use mgc_process_config instead */
1695 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1696 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
1697 sizeof(*mti), mti, NULL);
1699 if (mti->mti_flags & LDD_F_ERROR) {
1700 LCONSOLE_ERROR_MSG(0x160,
1701 "The MGS is refusing to allow this "
1702 "server (%s) to start. Please see messages"
1703 " on the MGS node.\n", lsi->lsi_svname);
1704 } else if (writeconf) {
1705 LCONSOLE_ERROR_MSG(0x15f,
1706 "Communication to the MGS return error %d. "
1707 "Is the MGS running?\n", rc);
1709 CERROR("Cannot talk to the MGS: %d, not fatal\n", rc);
1710 /* reset the error code for non-fatal error. */
1723 * Notify the MGS that this target is ready.
1724 * Used by IR - if the MGS receives this message, it will notify clients.
1726 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1728 struct lustre_sb_info *lsi = s2lsi(sb);
1729 struct obd_device *mgc = lsi->lsi_mgc;
1730 struct mgs_target_info *mti = NULL;
1736 if (!(IS_SERVER(lsi)))
1742 rc = server_lsi2mti(lsi, mti);
1746 mti->mti_instance = obd->u.obt.obt_instance;
1747 mti->mti_flags |= LDD_F_OPC_READY;
1749 /* FIXME use mgc_process_config instead */
1750 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1751 sizeof(KEY_REGISTER_TARGET),
1752 KEY_REGISTER_TARGET,
1753 sizeof(*mti), mti, NULL);
1755 /* Imperative recovery: if the mgs informs us to use IR? */
1756 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1757 (mti->mti_flags & LDD_F_IR_CAPABLE))
1758 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1767 /** Start server targets: MDTs and OSTs
1769 static int server_start_targets(struct super_block *sb, struct vfsmount *mnt)
1771 struct obd_device *obd;
1772 struct lustre_sb_info *lsi = s2lsi(sb);
1773 struct config_llog_instance cfg;
1775 struct lu_device *dev;
1779 CDEBUG(D_MOUNT, "starting target %s\n", lsi->lsi_svname);
1782 /* make sure the MDS is started */
1783 mutex_lock(&server_start_lock);
1784 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1786 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1788 LUSTRE_MDS_OBDNAME"_uuid",
1791 mutex_unlock(&server_start_lock);
1792 CERROR("failed to start MDS: %d\n", rc);
1796 mutex_unlock(&server_start_lock);
1799 /* If we're an OST, make sure the global OSS is running */
1801 /* make sure OSS is started */
1802 mutex_lock(&server_start_lock);
1803 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1805 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1807 LUSTRE_OSS_OBDNAME"_uuid",
1810 mutex_unlock(&server_start_lock);
1811 CERROR("failed to start OSS: %d\n", rc);
1815 mutex_unlock(&server_start_lock);
1818 /* Set the mgc fs to our server disk. This allows the MGC to
1819 * read and write configs locally, in case it can't talk to the MGS. */
1820 if (lsi->lsi_srv_mnt) {
1821 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1823 GOTO(out_stop_service, rc);
1826 /* Register with MGS */
1827 rc = server_register_target(lsi);
1831 /* Let the target look up the mount using the target's name
1832 (we can't pass the sb or mnt through class_process_config.) */
1833 rc = server_register_mount(lsi->lsi_svname, sb, mnt);
1837 /* Start targets using the llog named for the target */
1838 memset(&cfg, 0, sizeof(cfg));
1839 cfg.cfg_callback = class_config_llog_handler;
1840 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1842 CERROR("failed to start server %s: %d\n",
1843 lsi->lsi_svname, rc);
1844 /* Do NOT call server_deregister_mount() here. This makes it
1845 * impossible to find mount later in cleanup time and leaves
1846 * @lsi and othder stuff leaked. -umka */
1850 obd = class_name2obd(lsi->lsi_svname);
1852 CERROR("no server named %s was started\n", lsi->lsi_svname);
1853 GOTO(out_mgc, rc = -ENXIO);
1856 if (IS_OST(lsi) || IS_MDT(lsi)) {
1857 rc = lustre_start_lwp(sb);
1859 CERROR("%s: failed to start LWP: %d\n",
1860 lsi->lsi_svname, rc);
1865 server_notify_target(sb, obd);
1867 /* calculate recovery timeout, do it after lustre_process_log */
1868 server_calc_timeout(lsi, obd);
1870 /* log has been fully processed */
1871 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1873 /* log has been fully processed, let clients connect */
1874 dev = obd->obd_lu_dev;
1875 if (dev && dev->ld_ops->ldo_prepare) {
1876 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1878 struct lu_context session_ctx;
1880 lu_context_init(&session_ctx, LCT_SESSION);
1881 session_ctx.lc_thread = NULL;
1882 lu_context_enter(&session_ctx);
1883 env.le_ses = &session_ctx;
1885 dev->ld_ops->ldo_prepare(&env, NULL, dev);
1888 lu_context_exit(&session_ctx);
1889 lu_context_fini(&session_ctx);
1893 /* abort recovery only on the complete stack:
1894 * many devices can be involved */
1895 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1896 (OBP(obd, iocontrol))) {
1897 obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
1902 /* Release the mgc fs for others to use */
1903 if (lsi->lsi_srv_mnt)
1904 server_mgc_clear_fs(lsi->lsi_mgc);
1908 server_stop_servers(lsi->lsi_flags);
1913 /***************** lustre superblock **************/
1915 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
1917 struct lustre_sb_info *lsi;
1923 OBD_ALLOC_PTR(lsi->lsi_lmd);
1924 if (!lsi->lsi_lmd) {
1929 lsi->lsi_lmd->lmd_exclude_count = 0;
1930 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
1931 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
1932 s2lsi_nocast(sb) = lsi;
1933 /* we take 1 extra ref for our setup */
1934 cfs_atomic_set(&lsi->lsi_mounts, 1);
1936 /* Default umount style */
1937 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
1942 static int lustre_free_lsi(struct super_block *sb)
1944 struct lustre_sb_info *lsi = s2lsi(sb);
1947 LASSERT(lsi != NULL);
1948 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
1950 /* someone didn't call server_put_mount. */
1951 LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0);
1953 if (lsi->lsi_lmd != NULL) {
1954 if (lsi->lsi_lmd->lmd_dev != NULL)
1955 OBD_FREE(lsi->lsi_lmd->lmd_dev,
1956 strlen(lsi->lsi_lmd->lmd_dev) + 1);
1957 if (lsi->lsi_lmd->lmd_profile != NULL)
1958 OBD_FREE(lsi->lsi_lmd->lmd_profile,
1959 strlen(lsi->lsi_lmd->lmd_profile) + 1);
1960 if (lsi->lsi_lmd->lmd_mgssec != NULL)
1961 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
1962 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
1963 if (lsi->lsi_lmd->lmd_opts != NULL)
1964 OBD_FREE(lsi->lsi_lmd->lmd_opts,
1965 strlen(lsi->lsi_lmd->lmd_opts) + 1);
1966 if (lsi->lsi_lmd->lmd_exclude_count)
1967 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
1968 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
1969 lsi->lsi_lmd->lmd_exclude_count);
1970 if (lsi->lsi_lmd->lmd_mgs != NULL)
1971 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
1972 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
1973 if (lsi->lsi_lmd->lmd_osd_type != NULL)
1974 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
1975 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
1976 if (lsi->lsi_lmd->lmd_params != NULL)
1977 OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
1979 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
1982 LASSERT(lsi->lsi_llsbi == NULL);
1983 OBD_FREE(lsi, sizeof(*lsi));
1984 s2lsi_nocast(sb) = NULL;
1989 /* The lsi has one reference for every server that is using the disk -
1990 e.g. MDT, MGS, and potentially MGC */
1991 static int lustre_put_lsi(struct super_block *sb)
1993 struct lustre_sb_info *lsi = s2lsi(sb);
1996 LASSERT(lsi != NULL);
1998 CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts));
1999 if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) {
2000 if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
2001 obd_disconnect(lsi->lsi_osd_exp);
2002 /* wait till OSD is gone */
2003 obd_zombie_barrier();
2005 lustre_free_lsi(sb);
2011 static int lsi_prepare(struct lustre_sb_info *lsi)
2018 LASSERT(lsi->lsi_lmd);
2020 /* The server name is given as a mount line option */
2021 if (lsi->lsi_lmd->lmd_profile == NULL) {
2022 LCONSOLE_ERROR("Can't determine server name\n");
2026 if (strlen(lsi->lsi_lmd->lmd_profile) >= sizeof(lsi->lsi_svname))
2027 RETURN(-ENAMETOOLONG);
2029 strcpy(lsi->lsi_svname, lsi->lsi_lmd->lmd_profile);
2031 /* Determine osd type */
2032 if (lsi->lsi_lmd->lmd_osd_type != NULL) {
2033 if (strlen(lsi->lsi_lmd->lmd_osd_type) >=
2034 sizeof(lsi->lsi_osd_type))
2035 RETURN(-ENAMETOOLONG);
2037 strcpy(lsi->lsi_osd_type, lsi->lsi_lmd->lmd_osd_type);
2039 strcpy(lsi->lsi_osd_type, LUSTRE_OSD_LDISKFS_NAME);
2042 /* XXX: a temp. solution for components using fsfilt
2043 * to be removed in one of the subsequent patches */
2044 if (!strcmp(lsi->lsi_lmd->lmd_osd_type, "osd-ldiskfs")) {
2045 strcpy(lsi->lsi_fstype, "ldiskfs");
2047 strcpy(lsi->lsi_fstype, lsi->lsi_lmd->lmd_osd_type);
2050 /* Determine server type */
2051 rc = server_name2index(lsi->lsi_svname, &index, NULL);
2053 if (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) {
2054 /* Assume we're a bare MGS */
2056 lsi->lsi_lmd->lmd_flags |= LMD_FLG_NOSVC;
2058 LCONSOLE_ERROR("Can't determine server type of '%s'\n",
2063 lsi->lsi_flags |= rc;
2065 /* Add mount line flags that used to be in ldd:
2066 * writeconf, mgs, anything else?
2068 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_WRITECONF) ?
2069 LDD_F_WRITECONF : 0;
2070 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_VIRGIN) ?
2072 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_MGS) ?
2073 LDD_F_SV_TYPE_MGS : 0;
2074 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
2075 LDD_F_NO_PRIMNODE : 0;
2080 /*************** server mount ******************/
2082 /** Start the shutdown of servers at umount.
2084 static void server_put_super(struct super_block *sb)
2086 struct lustre_sb_info *lsi = s2lsi(sb);
2087 struct obd_device *obd;
2088 char *tmpname, *extraname = NULL;
2090 int lsiflags = lsi->lsi_flags;
2093 LASSERT(IS_SERVER(lsi));
2095 tmpname_sz = strlen(lsi->lsi_svname) + 1;
2096 OBD_ALLOC(tmpname, tmpname_sz);
2097 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
2098 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
2099 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
2100 snprintf(tmpname, tmpname_sz, "MGS");
2102 /* disconnect the lwp first to drain off the inflight request */
2103 if (IS_OST(lsi) || IS_MDT(lsi)) {
2106 rc = lustre_disconnect_lwp(sb);
2107 if (rc && rc != ETIMEDOUT)
2108 CERROR("%s: failed to disconnect lwp. (rc=%d)\n",
2112 /* Stop the target */
2113 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
2114 (IS_MDT(lsi) || IS_OST(lsi))) {
2115 struct lustre_profile *lprof = NULL;
2117 /* tell the mgc to drop the config log */
2118 lustre_end_log(sb, lsi->lsi_svname, NULL);
2120 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
2121 If there are any setup/cleanup errors, save the lov
2122 name for safety cleanup later. */
2123 lprof = class_get_profile(lsi->lsi_svname);
2124 if (lprof && lprof->lp_dt) {
2125 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
2126 strcpy(extraname, lprof->lp_dt);
2129 obd = class_name2obd(lsi->lsi_svname);
2131 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
2132 if (lsiflags & LSI_UMOUNT_FAILOVER)
2134 /* We can't seem to give an error return code
2135 * to .put_super, so we better make sure we clean up! */
2137 class_manual_cleanup(obd);
2139 CERROR("no obd %s\n", lsi->lsi_svname);
2140 server_deregister_mount(lsi->lsi_svname);
2144 /* If they wanted the mgs to stop separately from the mdt, they
2145 should have put it on a different device. */
2147 /* if MDS start with --nomgs, don't stop MGS then */
2148 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
2149 server_stop_mgs(sb);
2152 if (IS_OST(lsi) || IS_MDT(lsi)) {
2153 if (lustre_stop_lwp(sb) < 0)
2154 CERROR("%s: failed to stop lwp!\n", tmpname);
2157 /* Clean the mgc and sb */
2158 lustre_common_put_super(sb);
2160 /* wait till all in-progress cleanups are done
2161 * specifically we're interested in ofd cleanup
2163 obd_zombie_barrier();
2165 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
2166 until the target is really gone so that our type refcount check
2168 server_stop_servers(lsiflags);
2170 /* In case of startup or cleanup err, stop related obds */
2172 obd = class_name2obd(extraname);
2174 CWARN("Cleaning orphaned obd %s\n", extraname);
2176 class_manual_cleanup(obd);
2178 OBD_FREE(extraname, strlen(extraname) + 1);
2181 LCONSOLE_WARN("server umount %s complete\n", tmpname);
2182 OBD_FREE(tmpname, tmpname_sz);
2186 /** Called only for 'umount -f'
2188 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2189 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
2191 struct super_block *sb = vfsmnt->mnt_sb;
2193 static void server_umount_begin(struct super_block *sb)
2196 struct lustre_sb_info *lsi = s2lsi(sb);
2199 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2200 if (!(flags & MNT_FORCE)) {
2206 CDEBUG(D_MOUNT, "umount -f\n");
2207 /* umount = failover
2209 no third way to do non-force, non-failover */
2210 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
2214 static int server_statfs (struct dentry *dentry, cfs_kstatfs_t *buf)
2216 struct super_block *sb = dentry->d_sb;
2217 struct lustre_sb_info *lsi = s2lsi(sb);
2218 struct obd_statfs statfs;
2222 if (lsi->lsi_dt_dev) {
2223 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
2225 statfs_unpack(buf, &statfs);
2226 buf->f_type = sb->s_magic;
2232 buf->f_type = sb->s_magic;
2233 buf->f_bsize = sb->s_blocksize;
2239 buf->f_namelen = NAME_MAX;
2243 /** The operations we support directly on the superblock:
2244 * mount, umount, and df.
2246 static struct super_operations server_ops =
2248 .put_super = server_put_super,
2249 .umount_begin = server_umount_begin, /* umount -f */
2250 .statfs = server_statfs,
2253 #define log2(n) ffz(~(n))
2254 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
2256 static int server_fill_super_common(struct super_block *sb)
2258 struct inode *root = 0;
2261 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
2263 sb->s_blocksize = 4096;
2264 sb->s_blocksize_bits = log2(sb->s_blocksize);
2265 sb->s_magic = LUSTRE_SUPER_MAGIC;
2266 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
2267 sb->s_flags |= MS_RDONLY;
2268 sb->s_op = &server_ops;
2270 root = new_inode(sb);
2272 CERROR("Can't make root inode\n");
2276 /* returns -EIO for every operation */
2277 /* make_bad_inode(root); -- badness - can't umount */
2278 /* apparently we need to be a directory for the mount to finish */
2279 root->i_mode = S_IFDIR;
2281 sb->s_root = d_make_root(root);
2283 CERROR("%s: can't make root dentry\n", sb->s_id);
2290 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
2292 struct lustre_mount_data *lmd = lsi->lsi_lmd;
2293 struct obd_device *obd;
2294 struct dt_device_param p;
2300 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
2301 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
2303 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
2304 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
2305 strcat(lsi->lsi_osd_uuid, "_UUID");
2306 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
2308 obd = class_name2obd(lsi->lsi_osd_obdname);
2310 rc = lustre_start_simple(lsi->lsi_osd_obdname,
2312 lsi->lsi_osd_uuid, lmd->lmd_dev,
2313 flagstr, lsi->lsi_lmd->lmd_opts,
2317 obd = class_name2obd(lsi->lsi_osd_obdname);
2321 rc = obd_connect(NULL, &lsi->lsi_osd_exp, obd, &obd->obd_uuid, NULL, NULL);
2324 class_manual_cleanup(obd);
2325 lsi->lsi_dt_dev = NULL;
2328 /* XXX: to keep support old components relying on lsi_srv_mnt
2329 * we get this info from OSD just started */
2330 LASSERT(obd->obd_lu_dev);
2331 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
2332 LASSERT(lsi->lsi_dt_dev);
2334 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
2336 lsi->lsi_srv_mnt = p.ddp_mnt;
2342 /** Fill in the superblock info for a Lustre server.
2343 * Mount the device with the correct options.
2344 * Read the on-disk config file.
2345 * Start the services.
2347 static int server_fill_super(struct super_block *sb)
2349 struct lustre_sb_info *lsi = s2lsi(sb);
2353 rc = lsi_prepare(lsi);
2357 /* Start low level OSD */
2358 rc = osd_start(lsi, sb->s_flags);
2360 CERROR("Unable to start osd on %s: %d\n",
2361 lsi->lsi_lmd->lmd_dev, rc);
2366 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
2367 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
2369 if (class_name2obd(lsi->lsi_svname)) {
2370 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
2371 "running. Double-mount may have compromised"
2372 " the disk journal.\n",
2378 /* Start MGS before MGC */
2379 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
2380 rc = server_start_mgs(sb);
2385 /* Start MGC before servers */
2386 rc = lustre_start_mgc(sb);
2390 /* Set up all obd devices for service */
2391 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
2392 (IS_OST(lsi) || IS_MDT(lsi))) {
2393 rc = server_start_targets(sb, lsi->lsi_srv_mnt);
2395 CERROR("Unable to start targets: %d\n", rc);
2398 /* FIXME overmount client here,
2399 or can we just start a client log and client_fill_super on this sb?
2400 We need to make sure server_put_super gets called too - ll_put_super
2401 calls lustre_common_put_super; check there for LSI_SERVER flag,
2403 Probably should start client from new thread so we can return.
2404 Client will not finish until all servers are connected.
2405 Note - MGS-only server does NOT get a client, since there is no
2406 lustre fs associated - the MGS is for all lustre fs's */
2409 rc = server_fill_super_common(sb);
2415 /* We jump here in case of failure while starting targets or MGS.
2416 * In this case we can't just put @mnt and have to do real cleanup
2417 * with stoping targets, etc. */
2418 server_put_super(sb);
2423 * Calculate timeout value for a target.
2425 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
2427 struct lustre_mount_data *lmd;
2431 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
2432 int min = OBD_RECOVERY_TIME_MIN;
2434 LASSERT(IS_SERVER(lsi));
2438 soft = lmd->lmd_recovery_time_soft;
2439 hard = lmd->lmd_recovery_time_hard;
2440 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
2441 obd->obd_no_ir = !has_ir;
2445 soft = OBD_RECOVERY_TIME_SOFT;
2447 hard = OBD_RECOVERY_TIME_HARD;
2449 /* target may have ir_factor configured. */
2450 factor = OBD_IR_FACTOR_DEFAULT;
2451 if (obd->obd_recovery_ir_factor)
2452 factor = obd->obd_recovery_ir_factor;
2455 int new_soft = soft;
2456 int new_hard = hard;
2458 /* adjust timeout value by imperative recovery */
2460 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
2461 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
2463 /* make sure the timeout is not too short */
2464 new_soft = max(min, new_soft);
2465 new_hard = max(new_soft, new_hard);
2467 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
2468 "window shrunk from %d-%d down to %d-%d\n",
2469 obd->obd_name, soft, hard, new_soft, new_hard);
2476 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
2477 obd->obd_recovery_time_hard = hard;
2478 obd->obd_recovery_ir_factor = factor;
2480 EXPORT_SYMBOL(server_calc_timeout);
2482 /*************** mount common betweeen server and client ***************/
2485 int lustre_common_put_super(struct super_block *sb)
2490 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
2492 /* Drop a ref to the MGC */
2493 rc = lustre_stop_mgc(sb);
2494 if (rc && (rc != -ENOENT)) {
2496 CERROR("Can't stop MGC: %d\n", rc);
2499 /* BUSY just means that there's some other obd that
2500 needs the mgc. Let him clean it up. */
2501 CDEBUG(D_MOUNT, "MGC still in use\n");
2503 /* Drop a ref to the mounted disk */
2508 EXPORT_SYMBOL(lustre_common_put_super);
2510 static void lmd_print(struct lustre_mount_data *lmd)
2514 PRINT_CMD(PRINT_MASK, " mount data:\n");
2515 if (lmd_is_client(lmd))
2516 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
2517 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
2518 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
2521 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
2523 if (lmd->lmd_recovery_time_soft)
2524 PRINT_CMD(PRINT_MASK, "recovery time soft: %d\n",
2525 lmd->lmd_recovery_time_soft);
2527 if (lmd->lmd_recovery_time_hard)
2528 PRINT_CMD(PRINT_MASK, "recovery time hard: %d\n",
2529 lmd->lmd_recovery_time_hard);
2531 for (i = 0; i < lmd->lmd_exclude_count; i++) {
2532 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
2533 lmd->lmd_exclude[i]);
2537 /* Is this server on the exclusion list */
2538 int lustre_check_exclusion(struct super_block *sb, char *svname)
2540 struct lustre_sb_info *lsi = s2lsi(sb);
2541 struct lustre_mount_data *lmd = lsi->lsi_lmd;
2546 rc = server_name2index(svname, &index, NULL);
2547 if (rc != LDD_F_SV_TYPE_OST)
2548 /* Only exclude OSTs */
2551 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
2552 index, lmd->lmd_exclude_count, lmd->lmd_dev);
2554 for(i = 0; i < lmd->lmd_exclude_count; i++) {
2555 if (index == lmd->lmd_exclude[i]) {
2556 CWARN("Excluding %s (on exclusion list)\n", svname);
2563 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
2564 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
2566 char *s1 = ptr, *s2;
2567 __u32 index, *exclude_list;
2571 /* The shortest an ost name can be is 8 chars: -OST0000.
2572 We don't actually know the fsname at this time, so in fact
2573 a user could specify any fsname. */
2574 devmax = strlen(ptr) / 8 + 1;
2576 /* temp storage until we figure out how many we have */
2577 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
2581 /* we enter this fn pointing at the '=' */
2582 while (*s1 && *s1 != ' ' && *s1 != ',') {
2584 rc = server_name2index(s1, &index, &s2);
2586 CERROR("Can't parse server name '%s'\n", s1);
2589 if (rc == LDD_F_SV_TYPE_OST)
2590 exclude_list[lmd->lmd_exclude_count++] = index;
2592 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
2594 /* now we are pointing at ':' (next exclude)
2595 or ',' (end of excludes) */
2596 if (lmd->lmd_exclude_count >= devmax)
2599 if (rc >= 0) /* non-err */
2602 if (lmd->lmd_exclude_count) {
2603 /* permanent, freed in lustre_free_lsi */
2604 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
2605 lmd->lmd_exclude_count);
2606 if (lmd->lmd_exclude) {
2607 memcpy(lmd->lmd_exclude, exclude_list,
2608 sizeof(index) * lmd->lmd_exclude_count);
2611 lmd->lmd_exclude_count = 0;
2614 OBD_FREE(exclude_list, sizeof(index) * devmax);
2618 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
2623 if (lmd->lmd_mgssec != NULL) {
2624 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
2625 lmd->lmd_mgssec = NULL;
2628 tail = strchr(ptr, ',');
2630 length = strlen(ptr);
2632 length = tail - ptr;
2634 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
2635 if (lmd->lmd_mgssec == NULL)
2638 memcpy(lmd->lmd_mgssec, ptr, length);
2639 lmd->lmd_mgssec[length] = '\0';
2643 static int lmd_parse_string(char **handle, char *ptr)
2648 if ((handle == NULL) || (ptr == NULL))
2651 if (*handle != NULL) {
2652 OBD_FREE(*handle, strlen(*handle) + 1);
2656 tail = strchr(ptr, ',');
2658 length = strlen(ptr);
2660 length = tail - ptr;
2662 OBD_ALLOC(*handle, length + 1);
2663 if (*handle == NULL)
2666 memcpy(*handle, ptr, length);
2667 (*handle)[length] = '\0';
2672 /* Collect multiple values for mgsnid specifiers */
2673 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
2681 /* Find end of nidlist */
2682 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
2683 length = tail - *ptr;
2685 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
2689 if (lmd->lmd_mgs != NULL)
2690 oldlen = strlen(lmd->lmd_mgs) + 1;
2692 OBD_ALLOC(mgsnid, oldlen + length + 1);
2696 if (lmd->lmd_mgs != NULL) {
2697 /* Multiple mgsnid= are taken to mean failover locations */
2698 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
2699 mgsnid[oldlen - 1] = ':';
2700 OBD_FREE(lmd->lmd_mgs, oldlen);
2702 memcpy(mgsnid + oldlen, *ptr, length);
2703 mgsnid[oldlen + length] = '\0';
2704 lmd->lmd_mgs = mgsnid;
2710 /** Parse mount line options
2711 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
2712 * dev is passed as device=uml1:/lustre by mount.lustre
2714 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
2716 char *s1, *s2, *devname = NULL;
2717 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
2723 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
2724 "/sbin/mount.lustre is installed.\n");
2728 /* Options should be a string - try to detect old lmd data */
2729 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
2730 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
2731 "/sbin/mount.lustre. Please install "
2732 "version %s\n", LUSTRE_VERSION_STRING);
2735 lmd->lmd_magic = LMD_MAGIC;
2737 OBD_ALLOC(lmd->lmd_params, 4096);
2738 if (lmd->lmd_params == NULL)
2740 lmd->lmd_params[0] = '\0';
2742 /* Set default flags here */
2747 int time_min = OBD_RECOVERY_TIME_MIN;
2749 /* Skip whitespace and extra commas */
2750 while (*s1 == ' ' || *s1 == ',')
2753 /* Client options are parsed in ll_options: eg. flock,
2756 /* Parse non-ldiskfs options here. Rather than modifying
2757 ldiskfs, we just zero these out here */
2758 if (strncmp(s1, "abort_recov", 11) == 0) {
2759 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
2761 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
2762 lmd->lmd_recovery_time_soft = max_t(int,
2763 simple_strtoul(s1 + 19, NULL, 10), time_min);
2765 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
2766 lmd->lmd_recovery_time_hard = max_t(int,
2767 simple_strtoul(s1 + 19, NULL, 10), time_min);
2769 } else if (strncmp(s1, "noir", 4) == 0) {
2770 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
2772 } else if (strncmp(s1, "nosvc", 5) == 0) {
2773 lmd->lmd_flags |= LMD_FLG_NOSVC;
2775 } else if (strncmp(s1, "nomgs", 5) == 0) {
2776 lmd->lmd_flags |= LMD_FLG_NOMGS;
2778 } else if (strncmp(s1, "noscrub", 7) == 0) {
2779 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
2781 } else if (strncmp(s1, PARAM_MGSNODE,
2782 sizeof(PARAM_MGSNODE) - 1) == 0) {
2783 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
2784 /* Assume the next mount opt is the first
2785 invalid nid we get to. */
2786 rc = lmd_parse_mgs(lmd, &s2);
2790 } else if (strncmp(s1, "writeconf", 9) == 0) {
2791 lmd->lmd_flags |= LMD_FLG_WRITECONF;
2793 } else if (strncmp(s1, "virgin", 6) == 0) {
2794 lmd->lmd_flags |= LMD_FLG_VIRGIN;
2796 } else if (strncmp(s1, "noprimnode", 10) == 0) {
2797 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
2799 } else if (strncmp(s1, "mgssec=", 7) == 0) {
2800 rc = lmd_parse_mgssec(lmd, s1 + 7);
2804 /* ost exclusion list */
2805 } else if (strncmp(s1, "exclude=", 8) == 0) {
2806 rc = lmd_make_exclusion(lmd, s1 + 7);
2810 } else if (strncmp(s1, "mgs", 3) == 0) {
2812 lmd->lmd_flags |= LMD_FLG_MGS;
2814 } else if (strncmp(s1, "svname=", 7) == 0) {
2815 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
2819 } else if (strncmp(s1, "param=", 6) == 0) {
2821 char *tail = strchr(s1 + 6, ',');
2823 length = strlen(s1);
2827 strncat(lmd->lmd_params, s1 + 6, length);
2828 strcat(lmd->lmd_params, " ");
2830 } else if (strncmp(s1, "osd=", 4) == 0) {
2831 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
2836 /* Linux 2.4 doesn't pass the device, so we stuck it at the
2837 end of the options. */
2838 else if (strncmp(s1, "device=", 7) == 0) {
2840 /* terminate options right before device. device
2841 must be the last one. */
2847 s2 = strchr(s1, ',');
2855 memmove(s1, s2, strlen(s2) + 1);
2861 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
2862 "(need mount option 'device=...')\n");
2866 s1 = strstr(devname, ":/");
2869 lmd->lmd_flags |= LMD_FLG_CLIENT;
2870 /* Remove leading /s from fsname */
2871 while (*++s1 == '/') ;
2872 /* Freed in lustre_free_lsi */
2873 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
2874 if (!lmd->lmd_profile)
2876 sprintf(lmd->lmd_profile, "%s-client", s1);
2879 /* Freed in lustre_free_lsi */
2880 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
2883 strcpy(lmd->lmd_dev, devname);
2885 /* Save mount options */
2886 s1 = options + strlen(options) - 1;
2887 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
2889 if (*options != 0) {
2890 /* Freed in lustre_free_lsi */
2891 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
2894 strcpy(lmd->lmd_opts, options);
2898 lmd->lmd_magic = LMD_MAGIC;
2903 CERROR("Bad mount options %s\n", options);
2907 struct lustre_mount_data2 {
2909 struct vfsmount *lmd2_mnt;
2912 /** This is the entry point for the mount call into Lustre.
2913 * This is called when a server or client is mounted,
2914 * and this is where we start setting things up.
2915 * @param data Mount options (e.g. -o flock,abort_recov)
2917 int lustre_fill_super(struct super_block *sb, void *data, int silent)
2919 struct lustre_mount_data *lmd;
2920 struct lustre_mount_data2 *lmd2 = data;
2921 struct lustre_sb_info *lsi;
2925 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
2927 lsi = lustre_init_lsi(sb);
2933 * Disable lockdep during mount, because mount locking patterns are
2939 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
2941 obd_zombie_barrier();
2943 /* Figure out the lmd from the mount options */
2944 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
2946 GOTO(out, rc = -EINVAL);
2949 if (lmd_is_client(lmd)) {
2950 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
2951 if (!client_fill_super) {
2952 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
2953 "client mount! Is the 'lustre' "
2954 "module loaded?\n");
2958 rc = lustre_start_mgc(sb);
2963 /* Connect and start */
2964 /* (should always be ll_fill_super) */
2965 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
2966 /* c_f_s will call lustre_common_put_super on failure */
2969 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
2970 rc = server_fill_super(sb);
2971 /* s_f_s calls lustre_start_mgc after the mount because we need
2972 the MGS nids which are stored on disk. Plus, we may
2973 need to start the MGS first. */
2974 /* s_f_s will call server_put_super on failure */
2977 /* If error happens in fill_super() call, @lsi will be killed there.
2978 * This is why we do not put it here. */
2982 CERROR("Unable to mount %s (%d)\n",
2983 s2lsi(sb) ? lmd->lmd_dev : "", rc);
2985 CDEBUG(D_SUPER, "Mount %s complete\n",
2993 /* We can't call ll_fill_super by name because it lives in a module that
2994 must be loaded after this one. */
2995 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
2996 struct vfsmount *mnt))
2998 client_fill_super = cfs;
3000 EXPORT_SYMBOL(lustre_register_client_fill_super);
3002 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
3004 kill_super_cb = cfs;
3006 EXPORT_SYMBOL(lustre_register_kill_super_cb);
3008 /***************** FS registration ******************/
3009 #ifdef HAVE_FSTYPE_MOUNT
3010 struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
3011 const char *devname, void *data)
3013 struct lustre_mount_data2 lmd2 = { data, NULL };
3015 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
3018 int lustre_get_sb(struct file_system_type *fs_type, int flags,
3019 const char *devname, void * data, struct vfsmount *mnt)
3021 struct lustre_mount_data2 lmd2 = { data, mnt };
3023 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
3027 void lustre_kill_super(struct super_block *sb)
3029 struct lustre_sb_info *lsi = s2lsi(sb);
3031 if (kill_super_cb && lsi && !IS_SERVER(lsi))
3032 (*kill_super_cb)(sb);
3034 kill_anon_super(sb);
3037 /** Register the "lustre" fs type
3039 struct file_system_type lustre_fs_type = {
3040 .owner = THIS_MODULE,
3042 #ifdef HAVE_FSTYPE_MOUNT
3043 .mount = lustre_mount,
3045 .get_sb = lustre_get_sb,
3047 .kill_sb = lustre_kill_super,
3048 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
3049 FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE,
3052 int lustre_register_fs(void)
3054 return register_filesystem(&lustre_fs_type);
3057 int lustre_unregister_fs(void)
3059 return unregister_filesystem(&lustre_fs_type);
3062 EXPORT_SYMBOL(server_mti_print);