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, Whamcloud, Inc.
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 CFS_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 cfs_mutex_lock(&lustre_mount_info_lock);
110 if (server_find_mount(name)) {
111 cfs_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 cfs_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 cfs_mutex_lock(&lustre_mount_info_lock);
136 lmi = server_find_mount(name);
138 cfs_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 cfs_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 cfs_mutex_lock(&lustre_mount_info_lock);
163 lmi = server_find_mount(name);
164 cfs_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 cfs_mutex_lock(&lustre_mount_info_lock);
191 lmi = server_find_mount(name);
192 cfs_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 cfs_mutex_lock(&lustre_mount_info_lock);
210 lmi = server_find_mount(name);
211 cfs_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 CFS_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 cfs_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;
686 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
687 data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB;
689 #warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
692 if (lmd_is_client(lsi->lsi_lmd) &&
693 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
694 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
695 data->ocd_version = LUSTRE_VERSION_CODE;
696 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
698 CERROR("connect failed %d\n", rc);
702 obd->u.cli.cl_mgc_mgsexp = exp;
705 /* Keep the mgc info in the sb. Note that many lsi's can point
709 cfs_mutex_unlock(&mgc_start_lock);
714 OBD_FREE(mgcname, len);
716 OBD_FREE(niduuid, len + 2);
720 static int lustre_stop_mgc(struct super_block *sb)
722 struct lustre_sb_info *lsi = s2lsi(sb);
723 struct obd_device *obd;
724 char *niduuid = 0, *ptr = 0;
725 int i, rc = 0, len = 0;
735 cfs_mutex_lock(&mgc_start_lock);
736 LASSERT(cfs_atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
737 if (!cfs_atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
738 /* This is not fatal, every client that stops
739 will call in here. */
740 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
741 cfs_atomic_read(&obd->u.cli.cl_mgc_refcount));
742 GOTO(out, rc = -EBUSY);
745 /* The MGC has no recoverable data in any case.
746 * force shotdown set in umount_begin */
747 obd->obd_no_recov = 1;
749 if (obd->u.cli.cl_mgc_mgsexp) {
750 /* An error is not fatal, if we are unable to send the
751 disconnect mgs ping evictor cleans up the export */
752 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
754 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
757 /* Save the obdname for cleaning the nid uuids, which are
759 len = strlen(obd->obd_name) + 6;
760 OBD_ALLOC(niduuid, len);
762 strcpy(niduuid, obd->obd_name);
763 ptr = niduuid + strlen(niduuid);
766 rc = class_manual_cleanup(obd);
770 /* Clean the nid uuids */
772 GOTO(out, rc = -ENOMEM);
774 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
775 sprintf(ptr, "_%x", i);
776 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
779 CERROR("del MDC UUID %s failed: rc = %d\n",
784 OBD_FREE(niduuid, len);
786 /* class_import_put will get rid of the additional connections */
787 cfs_mutex_unlock(&mgc_start_lock);
791 /* Since there's only one mgc per node, we have to change it's fs to get
792 access to the right disk. */
793 static int server_mgc_set_fs(struct obd_device *mgc, struct super_block *sb)
795 struct lustre_sb_info *lsi = s2lsi(sb);
799 CDEBUG(D_MOUNT, "Set mgc disk for %s\n", lsi->lsi_lmd->lmd_dev);
801 /* cl_mgc_sem in mgc insures we sleep if the mgc_fs is busy */
802 rc = obd_set_info_async(NULL, mgc->obd_self_export,
803 sizeof(KEY_SET_FS), KEY_SET_FS,
804 sizeof(*sb), sb, NULL);
806 CERROR("can't set_fs %d\n", rc);
812 static int server_mgc_clear_fs(struct obd_device *mgc)
817 CDEBUG(D_MOUNT, "Unassign mgc disk\n");
819 rc = obd_set_info_async(NULL, mgc->obd_self_export,
820 sizeof(KEY_CLEAR_FS), KEY_CLEAR_FS,
825 /** Get the fsname ("lustre") from the server name ("lustre-OST003F").
826 * @param [in] svname server name including type and index
827 * @param [out] fsname Buffer to copy filesystem name prefix into.
828 * Must have at least 'strlen(fsname) + 1' chars.
829 * @param [out] endptr if endptr isn't NULL it is set to end of fsname
832 int server_name2fsname(char *svname, char *fsname, char **endptr)
834 char *dash = strrchr(svname, '-');
836 dash = strrchr(svname, ':');
841 /* interpret <fsname>-MDTXXXXX-mdc as mdt, the better way is to pass
842 * in the fsname, then determine the server index */
843 if (!strcmp(LUSTRE_MDC_NAME, dash + 1)) {
845 for (; dash > svname && *dash != '-' && *dash != ':'; dash--)
851 if (fsname != NULL) {
852 strncpy(fsname, svname, dash - svname);
853 fsname[dash - svname] = '\0';
861 EXPORT_SYMBOL(server_name2fsname);
863 static int is_mdc_device(char *devname)
866 ptr = strrchr(devname, '-');
867 if (ptr != NULL && strcmp(ptr, "-mdc") == 0)
872 static int inline tgt_is_mdt0(char *tgtname)
877 type = server_name2index(tgtname, &idx, NULL);
878 if (type != LDD_F_SV_TYPE_MDT)
881 return (idx == 0) ? 1 :0;
884 static int inline is_mdc_for_mdt0(char *devname)
888 if (!is_mdc_device(devname))
891 ptr = strrchr(devname, '-');
896 if (tgt_is_mdt0(devname)) {
905 * Convert OST/MDT name(fsname-OSTxxxx) to an osp name
906 * (fsname-MDT0000-osp-OSTxxxx), which will be used to
907 * communicate with MDT0 for this target.
909 int tgt_name2ospname(char *svname, char *ospname)
916 OBD_ALLOC(fsname, MTI_NAME_MAXLEN);
920 rc = server_name2fsname(svname, fsname, &tgt);
922 CERROR("%s change fsname error: rc %d\n", svname, rc);
926 if (*tgt != '-' && *tgt != ':') {
927 CERROR("%s wrong 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(ospname, "%s-MDT0000-%s-%s", fsname, LUSTRE_OSP_NAME, tgt);
939 OBD_FREE(fsname, MTI_NAME_MAXLEN);
942 EXPORT_SYMBOL(tgt_name2ospname);
944 static CFS_LIST_HEAD(osp_register_list);
945 CFS_DEFINE_MUTEX(osp_register_list_lock);
947 int lustre_register_osp_item(char *ospname, struct obd_export **exp,
948 register_osp_cb cb_func, void *cb_data)
950 struct obd_device *osp;
951 struct osp_register_item *ori;
954 LASSERTF(strlen(ospname) < MTI_NAME_MAXLEN, "ospname is too long %s\n",
956 LASSERT(exp != NULL && *exp == NULL);
962 cfs_mutex_lock(&osp_register_list_lock);
964 osp = class_name2obd(ospname);
965 if (osp != NULL && osp->obd_set_up == 1) {
966 struct obd_uuid *uuid;
970 cfs_mutex_unlock(&osp_register_list_lock);
973 memcpy(uuid->uuid, ospname, strlen(ospname));
974 *exp = cfs_hash_lookup(osp->obd_uuid_hash, uuid);
978 memcpy(ori->ori_name, ospname, strlen(ospname));
980 ori->ori_cb_func = cb_func;
981 ori->ori_cb_data = cb_data;
982 CFS_INIT_LIST_HEAD(&ori->ori_list);
983 cfs_list_add(&ori->ori_list, &osp_register_list);
985 if (*exp != NULL && cb_func != NULL)
988 cfs_mutex_unlock(&osp_register_list_lock);
991 EXPORT_SYMBOL(lustre_register_osp_item);
993 void lustre_deregister_osp_item(struct obd_export **exp)
995 struct osp_register_item *ori, *tmp;
997 cfs_mutex_lock(&osp_register_list_lock);
998 cfs_list_for_each_entry_safe(ori, tmp, &osp_register_list, ori_list) {
999 if (exp == ori->ori_exp) {
1001 class_export_put(*exp);
1002 cfs_list_del(&ori->ori_list);
1007 cfs_mutex_unlock(&osp_register_list_lock);
1009 EXPORT_SYMBOL(lustre_deregister_osp_item);
1011 static void lustre_notify_osp_list(struct obd_export *exp)
1013 struct osp_register_item *ori, *tmp;
1014 LASSERT(exp != NULL);
1016 cfs_mutex_lock(&osp_register_list_lock);
1017 cfs_list_for_each_entry_safe(ori, tmp, &osp_register_list, ori_list) {
1018 if (strcmp(exp->exp_obd->obd_name, ori->ori_name))
1020 if (*ori->ori_exp != NULL)
1022 *ori->ori_exp = class_export_get(exp);
1023 if (ori->ori_cb_func != NULL)
1024 ori->ori_cb_func(ori->ori_cb_data);
1026 cfs_mutex_unlock(&osp_register_list_lock);
1029 static int lustre_osp_connect(struct obd_device *osp)
1032 struct lu_context session_ctx;
1033 struct obd_export *exp;
1034 struct obd_uuid *uuid = NULL;
1035 struct obd_connect_data *data = NULL;
1039 /* log has been fully processed, let clients connect */
1040 rc = lu_env_init(&env, osp->obd_lu_dev->ld_type->ldt_ctx_tags);
1044 lu_context_init(&session_ctx, LCT_SESSION);
1045 session_ctx.lc_thread = NULL;
1046 lu_context_enter(&session_ctx);
1047 env.le_ses = &session_ctx;
1049 OBD_ALLOC_PTR(data);
1051 GOTO(out, rc = -ENOMEM);
1053 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_INDEX;
1054 data->ocd_version = LUSTRE_VERSION_CODE;
1055 data->ocd_ibits_known = MDS_INODELOCK_UPDATE;
1056 data->ocd_connect_flags |= OBD_CONNECT_ACL | OBD_CONNECT_IBITS |
1057 OBD_CONNECT_MDS_MDS | OBD_CONNECT_FID |
1058 OBD_CONNECT_AT | OBD_CONNECT_FULL20 |
1059 OBD_CONNECT_LIGHTWEIGHT;
1060 OBD_ALLOC_PTR(uuid);
1062 GOTO(out, rc = -ENOMEM);
1064 if (strlen(osp->obd_name) > sizeof(uuid->uuid)) {
1065 CERROR("%s: Too long osp name %s, max_size is %d\n",
1066 osp->obd_name, osp->obd_name, (int)sizeof(uuid->uuid));
1067 GOTO(out, rc = -EINVAL);
1069 /* Use osp name as the uuid, so we find the export by
1071 memcpy(uuid->uuid, osp->obd_name, strlen(osp->obd_name));
1072 rc = obd_connect(&env, &exp, osp, uuid, data, NULL);
1074 CERROR("%s: connect failed: rc = %d\n", osp->obd_name, rc);
1076 lustre_notify_osp_list(exp);
1085 lu_context_exit(&session_ctx);
1086 lu_context_fini(&session_ctx);
1092 * osp-on-ost is used by slaves (Non-MDT0 targets) to manage the connection
1095 * The OSTs will communicate with MDT0 by the connection established by the
1096 * osp-on-ost to get quota and fid sequence.
1099 static int lustre_osp_setup(struct lustre_cfg *lcfg, struct lustre_sb_info *lsi)
1101 struct obd_connect_data *data = NULL;
1102 struct obd_device *obd;
1103 char *ospname = NULL;
1104 char *ospuuid = NULL;
1108 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
1111 CERROR("%s: Can't add uuid: rc =%d\n", lsi->lsi_svname, rc);
1115 OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
1116 if (ospname == NULL)
1117 GOTO(out, rc = -ENOMEM);
1119 rc = tgt_name2ospname(lsi->lsi_svname, ospname);
1121 CERROR("%s change ospname error: rc %d\n",
1122 lsi->lsi_svname, rc);
1126 OBD_ALLOC(ospuuid, MTI_NAME_MAXLEN);
1127 if (ospuuid == NULL)
1128 GOTO(out, rc = -ENOMEM);
1130 sprintf(ospuuid, "%s_UUID", ospname);
1131 rc = lustre_start_simple(ospname, LUSTRE_OSP_NAME,
1132 ospuuid, lustre_cfg_string(lcfg, 1),
1135 CERROR("%s: setup up failed: rc %d\n", ospname, rc);
1139 obd = class_name2obd(ospname);
1140 LASSERT(obd != NULL);
1142 rc = lustre_osp_connect(obd);
1144 CERROR("%s: connect failed: rc = %d\n", ospname, rc);
1148 if (ospname != NULL)
1149 OBD_FREE(ospname, MTI_NAME_MAXLEN);
1150 if (ospuuid != NULL)
1151 OBD_FREE(ospuuid, MTI_NAME_MAXLEN);
1156 static int lustre_osp_add_conn(struct lustre_cfg *cfg,
1157 struct lustre_sb_info *lsi)
1159 struct lustre_cfg_bufs *bufs = NULL;
1160 struct lustre_cfg *lcfg = NULL;
1161 char *ospname = NULL;
1162 struct obd_device *osp;
1166 OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
1167 if (ospname == NULL)
1168 GOTO(out, rc = -ENOMEM);
1170 rc = tgt_name2ospname(lsi->lsi_svname, ospname);
1172 CERROR("%s change ospname error: rc %d\n",
1173 lsi->lsi_svname, rc);
1177 OBD_ALLOC_PTR(bufs);
1181 lustre_cfg_bufs_reset(bufs, ospname);
1182 lustre_cfg_bufs_set_string(bufs, 1,
1183 lustre_cfg_string(cfg, 1));
1185 lcfg = lustre_cfg_new(LCFG_ADD_CONN, bufs);
1187 osp = class_name2obd(ospname);
1189 CERROR("Can not find %s\n", ospname);
1190 GOTO(out, rc = -EINVAL);
1193 rc = class_add_conn(osp, lcfg);
1195 CERROR("%s: can't add conn: rc = %d\n", ospname, rc);
1201 lustre_cfg_free(lcfg);
1202 if (ospname != NULL)
1203 OBD_FREE(ospname, MTI_NAME_MAXLEN);
1209 * Retrieve MDT nids from the client log, then start the osp-on-ost device.
1210 * there are only two scenarios which would include mdt nid.
1212 * marker 5 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add mdc' xxx-
1213 * add_uuid nid=192.168.122.162@tcp(0x20000c0a87aa2) 0: 1:192.168.122.162@tcp
1214 * attach 0:lustre-MDT0000-mdc 1:mdc 2:lustre-clilmv_UUID
1215 * setup 0:lustre-MDT0000-mdc 1:lustre-MDT0000_UUID 2:192.168.122.162@tcp
1216 * add_uuid nid=192.168.172.1@tcp(0x20000c0a8ac01) 0: 1:192.168.172.1@tcp
1217 * add_conn 0:lustre-MDT0000-mdc 1:192.168.172.1@tcp
1218 * modify_mdc_tgts add 0:lustre-clilmv 1:lustre-MDT0000_UUID xxxx
1219 * marker 5 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add mdc' xxxx-
1221 * marker 7 (flags=0x01, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
1222 * add_uuid nid=192.168.122.2@tcp(0x20000c0a87a02) 0: 1:192.168.122.2@tcp
1223 * add_conn 0:lustre-MDT0000-mdc 1:192.168.122.2@tcp
1224 * marker 7 (flags=0x02, v2.1.54.0) lustre-MDT0000 'add failnid' xxxx-
1226 static int client_osp_config_process(const struct lu_env *env,
1227 struct llog_handle *handle,
1228 struct llog_rec_hdr *rec, void *data)
1230 struct config_llog_instance *clli = data;
1231 int cfg_len = rec->lrh_len;
1232 char *cfg_buf = (char *) (rec + 1);
1233 struct lustre_cfg *lcfg = NULL;
1234 struct lustre_sb_info *lsi;
1235 int rc = 0, swab = 0;
1238 if (rec->lrh_type != OBD_CFG_REC) {
1239 CERROR("Unknown llog record type %#x encountered\n",
1244 LASSERT(clli->cfg_sb != NULL);
1245 lsi = s2lsi(clli->cfg_sb);
1247 lcfg = (struct lustre_cfg *)cfg_buf;
1248 if (lcfg->lcfg_version == __swab32(LUSTRE_CFG_VERSION)) {
1249 lustre_swab_lustre_cfg(lcfg);
1253 rc = lustre_cfg_sanity_check(cfg_buf, cfg_len);
1257 switch (lcfg->lcfg_command) {
1259 struct cfg_marker *marker = lustre_cfg_buf(lcfg, 1);
1261 lustre_swab_cfg_marker(marker, swab,
1262 LUSTRE_CFG_BUFLEN(lcfg, 1));
1263 if (marker->cm_flags & CM_SKIP ||
1264 marker->cm_flags & CM_EXCLUDE)
1267 if (!tgt_is_mdt0(marker->cm_tgtname))
1270 if(!strncmp(marker->cm_comment, "add mdc", 7) ||
1271 !strncmp(marker->cm_comment, "add failnid", 11)) {
1272 if (marker->cm_flags & CM_START) {
1273 clli->cfg_flags = CFG_F_MARKER;
1274 /* This hack is to differentiate the
1275 * ADD_UUID is come from "add mdc" record
1276 * or from "add failnid" record. */
1277 if (!strncmp(marker->cm_comment,
1279 clli->cfg_flags |= CFG_F_SKIP;
1280 } else if (marker->cm_flags & CM_END) {
1281 clli->cfg_flags = 0;
1286 case LCFG_ADD_UUID: {
1287 if (clli->cfg_flags == CFG_F_MARKER) {
1288 rc = lustre_osp_setup(lcfg, lsi);
1289 } else if (clli->cfg_flags == (CFG_F_MARKER | CFG_F_SKIP)) {
1290 rc = class_add_uuid(lustre_cfg_string(lcfg, 1),
1293 CERROR("%s: Fail to add uuid, rc:%d\n",
1294 lsi->lsi_svname, rc);
1298 case LCFG_ADD_CONN: {
1299 if (is_mdc_for_mdt0(lustre_cfg_string(lcfg, 0)))
1300 rc = lustre_osp_add_conn(lcfg, lsi);
1310 static int lustre_disconnect_osp(struct super_block *sb)
1312 struct lustre_sb_info *lsi = s2lsi(sb);
1313 struct obd_device *osp;
1314 char *ospname = NULL;
1315 char *logname = NULL;
1316 struct lustre_cfg *lcfg = NULL;
1317 struct lustre_cfg_bufs *bufs = NULL;
1318 struct config_llog_instance *cfg = NULL;
1322 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
1323 OBD_ALLOC(logname, MTI_NAME_MAXLEN);
1324 if (logname == NULL)
1327 OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
1328 if (ospname == NULL)
1329 GOTO(out, rc = -ENOMEM);
1331 rc = server_name2fsname(lsi->lsi_svname, ospname, NULL);
1333 CERROR("%s: get fsname error: %d\n",
1334 lsi->lsi_svname, rc);
1337 sprintf(logname, "%s-client", ospname);
1341 GOTO(out, rc = -ENOMEM);
1344 cfg->cfg_instance = sb;
1345 rc = lustre_end_log(sb, logname, cfg);
1347 CERROR("Can't end config log %s\n", ospname);
1351 rc = tgt_name2ospname(lsi->lsi_svname, ospname);
1353 CERROR("%s: get osp name error: %d\n",
1354 lsi->lsi_svname, rc);
1358 osp = class_name2obd(ospname);
1360 CERROR("Can't find osp-on-ost %s\n", ospname);
1361 GOTO(out, rc = -ENOENT);
1364 OBD_ALLOC_PTR(bufs);
1366 GOTO(out, rc = -ENOMEM);
1368 lustre_cfg_bufs_reset(bufs, osp->obd_name);
1369 lustre_cfg_bufs_set_string(bufs, 1, NULL);
1370 lcfg = lustre_cfg_new(LCFG_CLEANUP, bufs);
1372 GOTO(out, rc = -ENOMEM);
1374 /* Disconnect import first. NULL is passed for the '@env', since
1375 * it will not be used for the 'osp-on-ost'. (see osp_shutdown()) */
1376 rc = osp->obd_lu_dev->ld_ops->ldo_process_config(NULL, osp->obd_lu_dev,
1380 lustre_cfg_free(lcfg);
1386 OBD_FREE(ospname, MTI_NAME_MAXLEN);
1388 OBD_FREE(logname, MTI_NAME_MAXLEN);
1393 * Stop the osp(fsname-MDT0000-osp-OSTxxxx) for an OST target.
1395 static int lustre_stop_osp(struct super_block *sb)
1397 struct lustre_sb_info *lsi = s2lsi(sb);
1398 struct obd_device *osp = NULL;
1399 char *ospname = NULL;
1403 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
1404 OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
1406 rc = tgt_name2ospname(lsi->lsi_svname, ospname);
1408 CERROR("%s get fsname error: rc %d\n",
1409 lsi->lsi_svname, rc);
1413 osp = class_name2obd(ospname);
1415 CERROR("Can not find osp-on-ost %s\n", ospname);
1416 GOTO(cleanup, rc = -ENOENT);
1420 rc = class_manual_cleanup(osp);
1423 if (ospname != NULL)
1424 OBD_FREE(ospname, MTI_NAME_MAXLEN);
1428 CFS_DEFINE_MUTEX(server_start_lock);
1430 /* Stop MDS/OSS if nobody is using them */
1431 static int server_stop_servers(int lsiflags)
1433 struct obd_device *obd = NULL;
1434 struct obd_type *type = NULL;
1438 cfs_mutex_lock(&server_start_lock);
1440 /* Either an MDT or an OST or neither */
1441 /* if this was an MDT, and there are no more MDT's, clean up the MDS */
1442 if ((lsiflags & LDD_F_SV_TYPE_MDT) &&
1443 (obd = class_name2obd(LUSTRE_MDS_OBDNAME))) {
1444 /*FIXME pre-rename, should eventually be LUSTRE_MDT_NAME*/
1445 type = class_search_type(LUSTRE_MDS_NAME);
1447 /* if this was an OST, and there are no more OST's, clean up the OSS */
1448 if ((lsiflags & LDD_F_SV_TYPE_OST) &&
1449 (obd = class_name2obd(LUSTRE_OSS_OBDNAME))) {
1450 type = class_search_type(LUSTRE_OST_NAME);
1453 if (obd && (!type || !type->typ_refcnt)) {
1456 /* obd_fail doesn't mean much on a server obd */
1457 err = class_manual_cleanup(obd);
1462 cfs_mutex_unlock(&server_start_lock);
1467 int server_mti_print(char *title, struct mgs_target_info *mti)
1469 PRINT_CMD(PRINT_MASK, "mti %s\n", title);
1470 PRINT_CMD(PRINT_MASK, "server: %s\n", mti->mti_svname);
1471 PRINT_CMD(PRINT_MASK, "fs: %s\n", mti->mti_fsname);
1472 PRINT_CMD(PRINT_MASK, "uuid: %s\n", mti->mti_uuid);
1473 PRINT_CMD(PRINT_MASK, "ver: %d flags: %#x\n",
1474 mti->mti_config_ver, mti->mti_flags);
1479 * Get service name (svname) from string
1481 * if endptr isn't NULL it is set to end of fsname *
1483 int server_name2svname(char *label, char *svname, char **endptr)
1488 /* We use server_name2fsname() just for parsing */
1489 rc = server_name2fsname(label, NULL, &dash);
1496 strncpy(svname, dash + 1, MTI_NAME_MAXLEN);
1500 EXPORT_SYMBOL(server_name2svname);
1503 /* Get the index from the obd name.
1504 rc = server type, or
1506 if endptr isn't NULL it is set to end of name */
1507 int server_name2index(char *svname, __u32 *idx, char **endptr)
1509 unsigned long index;
1513 /* We use server_name2fsname() just for parsing */
1514 rc = server_name2fsname(svname, NULL, &dash);
1523 if (strncmp(dash, "MDT", 3) == 0)
1524 rc = LDD_F_SV_TYPE_MDT;
1525 else if (strncmp(dash, "OST", 3) == 0)
1526 rc = LDD_F_SV_TYPE_OST;
1532 if (strcmp(dash, "all") == 0)
1533 return rc | LDD_F_SV_ALL;
1535 index = simple_strtoul(dash, endptr, 16);
1540 EXPORT_SYMBOL(server_name2index);
1542 /* Generate data for registration */
1543 static int server_lsi2mti(struct lustre_sb_info *lsi,
1544 struct mgs_target_info *mti)
1546 lnet_process_id_t id;
1550 if (!IS_SERVER(lsi))
1553 strncpy(mti->mti_svname, lsi->lsi_svname, sizeof(mti->mti_svname));
1555 mti->mti_nid_count = 0;
1556 while (LNetGetId(i++, &id) != -ENOENT) {
1557 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
1560 /* server use --servicenode param, only allow specified
1561 * nids be registered */
1562 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) != 0 &&
1563 class_match_nid(lsi->lsi_lmd->lmd_params,
1564 PARAM_FAILNODE, id.nid) < 1)
1567 /* match specified network */
1568 if (!class_match_net(lsi->lsi_lmd->lmd_params,
1569 PARAM_NETWORK, LNET_NIDNET(id.nid)))
1572 mti->mti_nids[mti->mti_nid_count] = id.nid;
1573 mti->mti_nid_count++;
1574 if (mti->mti_nid_count >= MTI_NIDS_MAX) {
1575 CWARN("Only using first %d nids for %s\n",
1576 mti->mti_nid_count, mti->mti_svname);
1581 mti->mti_lustre_ver = LUSTRE_VERSION_CODE;
1582 mti->mti_config_ver = 0;
1584 rc = server_name2fsname(lsi->lsi_svname, mti->mti_fsname, NULL);
1588 rc = server_name2index(lsi->lsi_svname, &mti->mti_stripe_index, NULL);
1591 /* Orion requires index to be set */
1592 LASSERT(!(rc & LDD_F_NEED_INDEX));
1593 /* keep only LDD flags */
1594 mti->mti_flags = lsi->lsi_flags & LDD_F_MASK;
1595 mti->mti_flags |= LDD_F_UPDATE;
1596 strncpy(mti->mti_params, lsi->lsi_lmd->lmd_params,
1597 sizeof(mti->mti_params));
1601 /* Register an old or new target with the MGS. If needed MGS will construct
1602 startup logs and assign index */
1603 static int server_register_target(struct lustre_sb_info *lsi)
1605 struct obd_device *mgc = lsi->lsi_mgc;
1606 struct mgs_target_info *mti = NULL;
1613 if (!IS_SERVER(lsi))
1620 rc = server_lsi2mti(lsi, mti);
1624 CDEBUG(D_MOUNT, "Registration %s, fs=%s, %s, index=%04x, flags=%#x\n",
1625 mti->mti_svname, mti->mti_fsname,
1626 libcfs_nid2str(mti->mti_nids[0]), mti->mti_stripe_index,
1629 /* if write_conf is true, the registration must succeed */
1630 writeconf = !!(lsi->lsi_flags & (LDD_F_NEED_INDEX | LDD_F_UPDATE));
1631 mti->mti_flags |= LDD_F_OPC_REG;
1633 /* Register the target */
1634 /* FIXME use mgc_process_config instead */
1635 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1636 sizeof(KEY_REGISTER_TARGET), KEY_REGISTER_TARGET,
1637 sizeof(*mti), mti, NULL);
1639 if (mti->mti_flags & LDD_F_ERROR) {
1640 LCONSOLE_ERROR_MSG(0x160,
1641 "The MGS is refusing to allow this "
1642 "server (%s) to start. Please see messages"
1643 " on the MGS node.\n", lsi->lsi_svname);
1644 } else if (writeconf) {
1645 LCONSOLE_ERROR_MSG(0x15f,
1646 "Communication to the MGS return error %d. "
1647 "Is the MGS running?\n", rc);
1649 CERROR("Cannot talk to the MGS: %d, not fatal\n", rc);
1650 /* reset the error code for non-fatal error. */
1663 * Notify the MGS that this target is ready.
1664 * Used by IR - if the MGS receives this message, it will notify clients.
1666 static int server_notify_target(struct super_block *sb, struct obd_device *obd)
1668 struct lustre_sb_info *lsi = s2lsi(sb);
1669 struct obd_device *mgc = lsi->lsi_mgc;
1670 struct mgs_target_info *mti = NULL;
1676 if (!(IS_SERVER(lsi)))
1682 rc = server_lsi2mti(lsi, mti);
1686 mti->mti_instance = obd->u.obt.obt_instance;
1687 mti->mti_flags |= LDD_F_OPC_READY;
1689 /* FIXME use mgc_process_config instead */
1690 rc = obd_set_info_async(NULL, mgc->u.cli.cl_mgc_mgsexp,
1691 sizeof(KEY_REGISTER_TARGET),
1692 KEY_REGISTER_TARGET,
1693 sizeof(*mti), mti, NULL);
1695 /* Imperative recovery: if the mgs informs us to use IR? */
1696 if (!rc && !(mti->mti_flags & LDD_F_ERROR) &&
1697 (mti->mti_flags & LDD_F_IR_CAPABLE))
1698 lsi->lsi_flags |= LDD_F_IR_CAPABLE;
1708 * Start the osp(fsname-MDT0000-osp-OSTxxxx) for an OST target,
1709 * which would be used to communicate with MDT0 for quota and FID.
1711 static int lustre_start_osp(struct super_block *sb)
1713 struct lustre_sb_info *lsi = s2lsi(sb);
1714 struct config_llog_instance *cfg = NULL;
1715 struct obd_device *osp;
1716 char *ospname = NULL;
1717 char *logname = NULL;
1722 LASSERT(IS_OST(lsi) || IS_MDT(lsi));
1723 OBD_ALLOC(ospname, MTI_NAME_MAXLEN);
1724 OBD_ALLOC(logname, MTI_NAME_MAXLEN);
1725 if (ospname == NULL || logname == NULL)
1726 GOTO(cleanup, rc = -ENOMEM);
1728 rc = server_name2fsname(lsi->lsi_svname, ospname, &tgt);
1730 CERROR("%s change fsname error: rc %d\n",
1731 lsi->lsi_svname, rc);
1734 sprintf(logname, "%s-client", ospname);
1736 rc = tgt_name2ospname(lsi->lsi_svname, ospname);
1738 CERROR("%s change ospname error: rc %d\n",
1739 lsi->lsi_svname, rc);
1743 osp = class_name2obd(ospname);
1745 GOTO(cleanup, rc = 0);
1749 GOTO(cleanup, rc = -ENOMEM);
1751 cfg->cfg_callback = client_osp_config_process;
1752 cfg->cfg_instance = sb;
1754 rc = lustre_process_log(sb, logname, cfg);
1757 if (ospname != NULL)
1758 OBD_FREE(ospname, MTI_NAME_MAXLEN);
1759 if (logname != NULL)
1760 OBD_FREE(logname, MTI_NAME_MAXLEN);
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 /* If we're an MDT, make sure the global MDS is running */
1783 if (lsi->lsi_ldd->ldd_flags & LDD_F_SV_TYPE_MDT) {
1784 /* make sure the MDS is started */
1785 cfs_mutex_lock(&server_start_lock);
1786 obd = class_name2obd(LUSTRE_MDS_OBDNAME);
1788 rc = lustre_start_simple(LUSTRE_MDS_OBDNAME,
1789 /* FIXME pre-rename, should eventually be LUSTRE_MDS_NAME */
1791 LUSTRE_MDS_OBDNAME"_uuid",
1794 cfs_mutex_unlock(&server_start_lock);
1795 CERROR("failed to start MDS: %d\n", rc);
1799 cfs_mutex_unlock(&server_start_lock);
1803 /* If we're an OST, make sure the global OSS is running */
1805 /* make sure OSS is started */
1806 cfs_mutex_lock(&server_start_lock);
1807 obd = class_name2obd(LUSTRE_OSS_OBDNAME);
1809 rc = lustre_start_simple(LUSTRE_OSS_OBDNAME,
1811 LUSTRE_OSS_OBDNAME"_uuid",
1814 cfs_mutex_unlock(&server_start_lock);
1815 CERROR("failed to start OSS: %d\n", rc);
1819 cfs_mutex_unlock(&server_start_lock);
1822 /* Set the mgc fs to our server disk. This allows the MGC to
1823 * read and write configs locally, in case it can't talk to the MGS. */
1824 if (lsi->lsi_srv_mnt) {
1825 rc = server_mgc_set_fs(lsi->lsi_mgc, sb);
1830 /* Register with MGS */
1831 rc = server_register_target(lsi);
1835 /* Let the target look up the mount using the target's name
1836 (we can't pass the sb or mnt through class_process_config.) */
1837 rc = server_register_mount(lsi->lsi_svname, sb, mnt);
1841 /* Start targets using the llog named for the target */
1842 memset(&cfg, 0, sizeof(cfg));
1843 cfg.cfg_callback = class_config_llog_handler;
1844 rc = lustre_process_log(sb, lsi->lsi_svname, &cfg);
1846 CERROR("failed to start server %s: %d\n",
1847 lsi->lsi_svname, rc);
1848 /* Do NOT call server_deregister_mount() here. This makes it
1849 * impossible to find mount later in cleanup time and leaves
1850 * @lsi and othder stuff leaked. -umka */
1854 obd = class_name2obd(lsi->lsi_svname);
1856 CERROR("no server named %s was started\n", lsi->lsi_svname);
1857 GOTO(out_mgc, rc = -ENXIO);
1860 if (IS_OST(lsi) || IS_MDT(lsi)) {
1861 rc = lustre_start_osp(sb);
1863 CERROR("%s: failed to start OSP: %d\n",
1864 lsi->lsi_svname, rc);
1869 server_notify_target(sb, obd);
1871 /* calculate recovery timeout, do it after lustre_process_log */
1872 server_calc_timeout(lsi, obd);
1874 /* log has been fully processed */
1875 obd_notify(obd, NULL, OBD_NOTIFY_CONFIG, (void *)CONFIG_LOG);
1877 /* log has been fully processed, let clients connect */
1878 dev = obd->obd_lu_dev;
1879 if (dev && dev->ld_ops->ldo_prepare) {
1880 rc = lu_env_init(&env, dev->ld_type->ldt_ctx_tags);
1882 struct lu_context session_ctx;
1884 lu_context_init(&session_ctx, LCT_SESSION);
1885 session_ctx.lc_thread = NULL;
1886 lu_context_enter(&session_ctx);
1887 env.le_ses = &session_ctx;
1889 dev->ld_ops->ldo_prepare(&env, NULL, dev);
1892 lu_context_exit(&session_ctx);
1893 lu_context_fini(&session_ctx);
1897 /* abort recovery only on the complete stack:
1898 * many devices can be involved */
1899 if ((lsi->lsi_lmd->lmd_flags & LMD_FLG_ABORT_RECOV) &&
1900 (OBP(obd, iocontrol))) {
1901 obd_iocontrol(OBD_IOC_ABORT_RECOVERY, obd->obd_self_export, 0,
1906 /* Release the mgc fs for others to use */
1907 if (lsi->lsi_srv_mnt)
1908 server_mgc_clear_fs(lsi->lsi_mgc);
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, iam, 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_IAM) ?
2076 lsi->lsi_flags |= (lsi->lsi_lmd->lmd_flags & LMD_FLG_NO_PRIMNODE) ?
2077 LDD_F_NO_PRIMNODE : 0;
2082 /*************** server mount ******************/
2084 /** Start the shutdown of servers at umount.
2086 static void server_put_super(struct super_block *sb)
2088 struct lustre_sb_info *lsi = s2lsi(sb);
2089 struct obd_device *obd;
2090 char *tmpname, *extraname = NULL;
2092 int lsiflags = lsi->lsi_flags;
2095 LASSERT(IS_SERVER(lsi));
2097 tmpname_sz = strlen(lsi->lsi_svname) + 1;
2098 OBD_ALLOC(tmpname, tmpname_sz);
2099 memcpy(tmpname, lsi->lsi_svname, tmpname_sz);
2100 CDEBUG(D_MOUNT, "server put_super %s\n", tmpname);
2101 if (IS_MDT(lsi) && (lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC))
2102 snprintf(tmpname, tmpname_sz, "MGS");
2104 /* disconnect the osp-on-ost first to drain off the inflight request */
2105 if (IS_OST(lsi) || IS_MDT(lsi)) {
2106 if (lustre_disconnect_osp(sb) < 0)
2107 CERROR("%s: Fail to disconnect osp-on-ost!\n", tmpname);
2110 /* Stop the target */
2111 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
2112 (IS_MDT(lsi) || IS_OST(lsi))) {
2113 struct lustre_profile *lprof = NULL;
2115 /* tell the mgc to drop the config log */
2116 lustre_end_log(sb, lsi->lsi_svname, NULL);
2118 /* COMPAT_146 - profile may get deleted in mgc_cleanup.
2119 If there are any setup/cleanup errors, save the lov
2120 name for safety cleanup later. */
2121 lprof = class_get_profile(lsi->lsi_svname);
2122 if (lprof && lprof->lp_dt) {
2123 OBD_ALLOC(extraname, strlen(lprof->lp_dt) + 1);
2124 strcpy(extraname, lprof->lp_dt);
2127 obd = class_name2obd(lsi->lsi_svname);
2129 CDEBUG(D_MOUNT, "stopping %s\n", obd->obd_name);
2130 if (lsi->lsi_flags & LSI_UMOUNT_FAILOVER)
2132 /* We can't seem to give an error return code
2133 * to .put_super, so we better make sure we clean up! */
2135 class_manual_cleanup(obd);
2137 CERROR("no obd %s\n", lsi->lsi_svname);
2138 server_deregister_mount(lsi->lsi_svname);
2142 /* If they wanted the mgs to stop separately from the mdt, they
2143 should have put it on a different device. */
2145 /* if MDS start with --nomgs, don't stop MGS then */
2146 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS))
2147 server_stop_mgs(sb);
2150 if (IS_OST(lsi) || IS_MDT(lsi)) {
2151 if (lustre_stop_osp(sb) < 0)
2152 CERROR("%s: Fail to stop osp-on-ost!\n", tmpname);
2155 /* Clean the mgc and sb */
2156 lustre_common_put_super(sb);
2158 /* wait till all in-progress cleanups are done
2159 * specifically we're interested in ofd cleanup
2161 obd_zombie_barrier();
2163 /* Stop the servers (MDS, OSS) if no longer needed. We must wait
2164 until the target is really gone so that our type refcount check
2166 server_stop_servers(lsiflags);
2168 /* In case of startup or cleanup err, stop related obds */
2170 obd = class_name2obd(extraname);
2172 CWARN("Cleaning orphaned obd %s\n", extraname);
2174 class_manual_cleanup(obd);
2176 OBD_FREE(extraname, strlen(extraname) + 1);
2179 LCONSOLE_WARN("server umount %s complete\n", tmpname);
2180 OBD_FREE(tmpname, tmpname_sz);
2184 /** Called only for 'umount -f'
2186 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2187 static void server_umount_begin(struct vfsmount *vfsmnt, int flags)
2189 struct super_block *sb = vfsmnt->mnt_sb;
2191 static void server_umount_begin(struct super_block *sb)
2194 struct lustre_sb_info *lsi = s2lsi(sb);
2197 #ifdef HAVE_UMOUNTBEGIN_VFSMOUNT
2198 if (!(flags & MNT_FORCE)) {
2204 CDEBUG(D_MOUNT, "umount -f\n");
2205 /* umount = failover
2207 no third way to do non-force, non-failover */
2208 lsi->lsi_flags &= ~LSI_UMOUNT_FAILOVER;
2212 static int server_statfs (struct dentry *dentry, cfs_kstatfs_t *buf)
2214 struct super_block *sb = dentry->d_sb;
2215 struct lustre_sb_info *lsi = s2lsi(sb);
2216 struct obd_statfs statfs;
2220 if (lsi->lsi_dt_dev) {
2221 rc = dt_statfs(NULL, lsi->lsi_dt_dev, &statfs);
2223 statfs_unpack(buf, &statfs);
2224 buf->f_type = sb->s_magic;
2230 buf->f_type = sb->s_magic;
2231 buf->f_bsize = sb->s_blocksize;
2237 buf->f_namelen = NAME_MAX;
2241 /** The operations we support directly on the superblock:
2242 * mount, umount, and df.
2244 static struct super_operations server_ops =
2246 .put_super = server_put_super,
2247 .umount_begin = server_umount_begin, /* umount -f */
2248 .statfs = server_statfs,
2251 #define log2(n) cfs_ffz(~(n))
2252 #define LUSTRE_SUPER_MAGIC 0x0BD00BD1
2254 static int server_fill_super_common(struct super_block *sb)
2256 struct inode *root = 0;
2259 CDEBUG(D_MOUNT, "Server sb, dev=%d\n", (int)sb->s_dev);
2261 sb->s_blocksize = 4096;
2262 sb->s_blocksize_bits = log2(sb->s_blocksize);
2263 sb->s_magic = LUSTRE_SUPER_MAGIC;
2264 sb->s_maxbytes = 0; /* we don't allow file IO on server mountpoints */
2265 sb->s_flags |= MS_RDONLY;
2266 sb->s_op = &server_ops;
2268 root = new_inode(sb);
2270 CERROR("Can't make root inode\n");
2274 /* returns -EIO for every operation */
2275 /* make_bad_inode(root); -- badness - can't umount */
2276 /* apparently we need to be a directory for the mount to finish */
2277 root->i_mode = S_IFDIR;
2279 sb->s_root = d_alloc_root(root);
2281 CERROR("Can't make root dentry\n");
2289 static int osd_start(struct lustre_sb_info *lsi, unsigned long mflags)
2291 struct lustre_mount_data *lmd = lsi->lsi_lmd;
2292 struct obd_device *obd;
2293 struct dt_device_param p;
2299 "Attempting to start %s, type=%s, lsifl=%x, mountfl=%lx\n",
2300 lsi->lsi_svname, lsi->lsi_osd_type, lsi->lsi_flags, mflags);
2302 sprintf(lsi->lsi_osd_obdname, "%s-osd", lsi->lsi_svname);
2303 strcpy(lsi->lsi_osd_uuid, lsi->lsi_osd_obdname);
2304 strcat(lsi->lsi_osd_uuid, "_UUID");
2305 sprintf(flagstr, "%lu:%lu", mflags, (unsigned long) lmd->lmd_flags);
2307 obd = class_name2obd(lsi->lsi_osd_obdname);
2309 rc = lustre_start_simple(lsi->lsi_osd_obdname,
2311 lsi->lsi_osd_uuid, lmd->lmd_dev,
2312 flagstr, lsi->lsi_lmd->lmd_opts,
2316 obd = class_name2obd(lsi->lsi_osd_obdname);
2320 rc = obd_connect(NULL, &lsi->lsi_osd_exp, obd, &obd->obd_uuid, NULL, NULL);
2323 class_manual_cleanup(obd);
2324 lsi->lsi_dt_dev = NULL;
2327 /* XXX: to keep support old components relying on lsi_srv_mnt
2328 * we get this info from OSD just started */
2329 LASSERT(obd->obd_lu_dev);
2330 lsi->lsi_dt_dev = lu2dt_dev(obd->obd_lu_dev);
2331 LASSERT(lsi->lsi_dt_dev);
2333 dt_conf_get(NULL, lsi->lsi_dt_dev, &p);
2335 lsi->lsi_srv_mnt = p.ddp_mnt;
2341 /** Fill in the superblock info for a Lustre server.
2342 * Mount the device with the correct options.
2343 * Read the on-disk config file.
2344 * Start the services.
2346 static int server_fill_super(struct super_block *sb)
2348 struct lustre_sb_info *lsi = s2lsi(sb);
2352 rc = lsi_prepare(lsi);
2356 /* Start low level OSD */
2357 rc = osd_start(lsi, sb->s_flags);
2359 CERROR("Unable to start osd on %s: %d\n",
2360 lsi->lsi_lmd->lmd_dev, rc);
2365 CDEBUG(D_MOUNT, "Found service %s on device %s\n",
2366 lsi->lsi_svname, lsi->lsi_lmd->lmd_dev);
2368 if (class_name2obd(lsi->lsi_svname)) {
2369 LCONSOLE_ERROR_MSG(0x161, "The target named %s is already "
2370 "running. Double-mount may have compromised"
2371 " the disk journal.\n",
2377 /* Start MGS before MGC */
2378 if (IS_MGS(lsi) && !(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOMGS)){
2379 rc = server_start_mgs(sb);
2384 /* Start MGC before servers */
2385 rc = lustre_start_mgc(sb);
2389 /* Set up all obd devices for service */
2390 if (!(lsi->lsi_lmd->lmd_flags & LMD_FLG_NOSVC) &&
2391 (IS_OST(lsi) || IS_MDT(lsi))) {
2392 rc = server_start_targets(sb, lsi->lsi_srv_mnt);
2394 CERROR("Unable to start targets: %d\n", rc);
2397 /* FIXME overmount client here,
2398 or can we just start a client log and client_fill_super on this sb?
2399 We need to make sure server_put_super gets called too - ll_put_super
2400 calls lustre_common_put_super; check there for LSI_SERVER flag,
2402 Probably should start client from new thread so we can return.
2403 Client will not finish until all servers are connected.
2404 Note - MGS-only server does NOT get a client, since there is no
2405 lustre fs associated - the MGS is for all lustre fs's */
2408 rc = server_fill_super_common(sb);
2414 /* We jump here in case of failure while starting targets or MGS.
2415 * In this case we can't just put @mnt and have to do real cleanup
2416 * with stoping targets, etc. */
2417 server_put_super(sb);
2422 * Calculate timeout value for a target.
2424 void server_calc_timeout(struct lustre_sb_info *lsi, struct obd_device *obd)
2426 struct lustre_mount_data *lmd;
2430 bool has_ir = !!(lsi->lsi_flags & LDD_F_IR_CAPABLE);
2431 int min = OBD_RECOVERY_TIME_MIN;
2433 LASSERT(IS_SERVER(lsi));
2437 soft = lmd->lmd_recovery_time_soft;
2438 hard = lmd->lmd_recovery_time_hard;
2439 has_ir = has_ir && !(lmd->lmd_flags & LMD_FLG_NOIR);
2440 obd->obd_no_ir = !has_ir;
2444 soft = OBD_RECOVERY_TIME_SOFT;
2446 hard = OBD_RECOVERY_TIME_HARD;
2448 /* target may have ir_factor configured. */
2449 factor = OBD_IR_FACTOR_DEFAULT;
2450 if (obd->obd_recovery_ir_factor)
2451 factor = obd->obd_recovery_ir_factor;
2454 int new_soft = soft;
2455 int new_hard = hard;
2457 /* adjust timeout value by imperative recovery */
2459 new_soft = (soft * factor) / OBD_IR_FACTOR_MAX;
2460 new_hard = (hard * factor) / OBD_IR_FACTOR_MAX;
2462 /* make sure the timeout is not too short */
2463 new_soft = max(min, new_soft);
2464 new_hard = max(new_soft, new_hard);
2466 LCONSOLE_INFO("%s: Imperative Recovery enabled, recovery "
2467 "window shrunk from %d-%d down to %d-%d\n",
2468 obd->obd_name, soft, hard, new_soft, new_hard);
2475 obd->obd_recovery_timeout = max(obd->obd_recovery_timeout, soft);
2476 obd->obd_recovery_time_hard = hard;
2477 obd->obd_recovery_ir_factor = factor;
2479 EXPORT_SYMBOL(server_calc_timeout);
2481 /*************** mount common betweeen server and client ***************/
2484 int lustre_common_put_super(struct super_block *sb)
2489 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
2491 /* Drop a ref to the MGC */
2492 rc = lustre_stop_mgc(sb);
2493 if (rc && (rc != -ENOENT)) {
2495 CERROR("Can't stop MGC: %d\n", rc);
2498 /* BUSY just means that there's some other obd that
2499 needs the mgc. Let him clean it up. */
2500 CDEBUG(D_MOUNT, "MGC still in use\n");
2502 /* Drop a ref to the mounted disk */
2507 EXPORT_SYMBOL(lustre_common_put_super);
2509 static void lmd_print(struct lustre_mount_data *lmd)
2513 PRINT_CMD(PRINT_MASK, " mount data:\n");
2514 if (lmd_is_client(lmd))
2515 PRINT_CMD(PRINT_MASK, "profile: %s\n", lmd->lmd_profile);
2516 PRINT_CMD(PRINT_MASK, "device: %s\n", lmd->lmd_dev);
2517 PRINT_CMD(PRINT_MASK, "flags: %x\n", lmd->lmd_flags);
2520 PRINT_CMD(PRINT_MASK, "options: %s\n", lmd->lmd_opts);
2522 if (lmd->lmd_recovery_time_soft)
2523 PRINT_CMD(PRINT_MASK, "recovery time soft: %d\n",
2524 lmd->lmd_recovery_time_soft);
2526 if (lmd->lmd_recovery_time_hard)
2527 PRINT_CMD(PRINT_MASK, "recovery time hard: %d\n",
2528 lmd->lmd_recovery_time_hard);
2530 for (i = 0; i < lmd->lmd_exclude_count; i++) {
2531 PRINT_CMD(PRINT_MASK, "exclude %d: OST%04x\n", i,
2532 lmd->lmd_exclude[i]);
2536 /* Is this server on the exclusion list */
2537 int lustre_check_exclusion(struct super_block *sb, char *svname)
2539 struct lustre_sb_info *lsi = s2lsi(sb);
2540 struct lustre_mount_data *lmd = lsi->lsi_lmd;
2545 rc = server_name2index(svname, &index, NULL);
2546 if (rc != LDD_F_SV_TYPE_OST)
2547 /* Only exclude OSTs */
2550 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
2551 index, lmd->lmd_exclude_count, lmd->lmd_dev);
2553 for(i = 0; i < lmd->lmd_exclude_count; i++) {
2554 if (index == lmd->lmd_exclude[i]) {
2555 CWARN("Excluding %s (on exclusion list)\n", svname);
2562 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
2563 static int lmd_make_exclusion(struct lustre_mount_data *lmd, char *ptr)
2565 char *s1 = ptr, *s2;
2566 __u32 index, *exclude_list;
2570 /* The shortest an ost name can be is 8 chars: -OST0000.
2571 We don't actually know the fsname at this time, so in fact
2572 a user could specify any fsname. */
2573 devmax = strlen(ptr) / 8 + 1;
2575 /* temp storage until we figure out how many we have */
2576 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
2580 /* we enter this fn pointing at the '=' */
2581 while (*s1 && *s1 != ' ' && *s1 != ',') {
2583 rc = server_name2index(s1, &index, &s2);
2585 CERROR("Can't parse server name '%s'\n", s1);
2588 if (rc == LDD_F_SV_TYPE_OST)
2589 exclude_list[lmd->lmd_exclude_count++] = index;
2591 CDEBUG(D_MOUNT, "ignoring exclude %.7s\n", s1);
2593 /* now we are pointing at ':' (next exclude)
2594 or ',' (end of excludes) */
2595 if (lmd->lmd_exclude_count >= devmax)
2598 if (rc >= 0) /* non-err */
2601 if (lmd->lmd_exclude_count) {
2602 /* permanent, freed in lustre_free_lsi */
2603 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
2604 lmd->lmd_exclude_count);
2605 if (lmd->lmd_exclude) {
2606 memcpy(lmd->lmd_exclude, exclude_list,
2607 sizeof(index) * lmd->lmd_exclude_count);
2610 lmd->lmd_exclude_count = 0;
2613 OBD_FREE(exclude_list, sizeof(index) * devmax);
2617 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
2622 if (lmd->lmd_mgssec != NULL) {
2623 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
2624 lmd->lmd_mgssec = NULL;
2627 tail = strchr(ptr, ',');
2629 length = strlen(ptr);
2631 length = tail - ptr;
2633 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
2634 if (lmd->lmd_mgssec == NULL)
2637 memcpy(lmd->lmd_mgssec, ptr, length);
2638 lmd->lmd_mgssec[length] = '\0';
2642 static int lmd_parse_string(char **handle, char *ptr)
2647 if ((handle == NULL) || (ptr == NULL))
2650 if (*handle != NULL) {
2651 OBD_FREE(*handle, strlen(*handle) + 1);
2655 tail = strchr(ptr, ',');
2657 length = strlen(ptr);
2659 length = tail - ptr;
2661 OBD_ALLOC(*handle, length + 1);
2662 if (*handle == NULL)
2665 memcpy(*handle, ptr, length);
2666 (*handle)[length] = '\0';
2671 /* Collect multiple values for mgsnid specifiers */
2672 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
2680 /* Find end of nidlist */
2681 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
2682 length = tail - *ptr;
2684 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
2688 if (lmd->lmd_mgs != NULL)
2689 oldlen = strlen(lmd->lmd_mgs) + 1;
2691 OBD_ALLOC(mgsnid, oldlen + length + 1);
2695 if (lmd->lmd_mgs != NULL) {
2696 /* Multiple mgsnid= are taken to mean failover locations */
2697 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
2698 mgsnid[oldlen - 1] = ':';
2699 OBD_FREE(lmd->lmd_mgs, oldlen);
2701 memcpy(mgsnid + oldlen, *ptr, length);
2702 mgsnid[oldlen + length] = '\0';
2703 lmd->lmd_mgs = mgsnid;
2709 /** Parse mount line options
2710 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
2711 * dev is passed as device=uml1:/lustre by mount.lustre
2713 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
2715 char *s1, *s2, *devname = NULL;
2716 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
2722 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
2723 "/sbin/mount.lustre is installed.\n");
2727 /* Options should be a string - try to detect old lmd data */
2728 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
2729 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
2730 "/sbin/mount.lustre. Please install "
2731 "version %s\n", LUSTRE_VERSION_STRING);
2734 lmd->lmd_magic = LMD_MAGIC;
2736 OBD_ALLOC(lmd->lmd_params, 4096);
2737 if (lmd->lmd_params == NULL)
2739 lmd->lmd_params[0] = '\0';
2741 /* Set default flags here */
2746 int time_min = OBD_RECOVERY_TIME_MIN;
2748 /* Skip whitespace and extra commas */
2749 while (*s1 == ' ' || *s1 == ',')
2752 /* Client options are parsed in ll_options: eg. flock,
2755 /* Parse non-ldiskfs options here. Rather than modifying
2756 ldiskfs, we just zero these out here */
2757 if (strncmp(s1, "abort_recov", 11) == 0) {
2758 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
2760 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
2761 lmd->lmd_recovery_time_soft = max_t(int,
2762 simple_strtoul(s1 + 19, NULL, 10), time_min);
2764 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
2765 lmd->lmd_recovery_time_hard = max_t(int,
2766 simple_strtoul(s1 + 19, NULL, 10), time_min);
2768 } else if (strncmp(s1, "noir", 4) == 0) {
2769 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
2771 } else if (strncmp(s1, "nosvc", 5) == 0) {
2772 lmd->lmd_flags |= LMD_FLG_NOSVC;
2774 } else if (strncmp(s1, "nomgs", 5) == 0) {
2775 lmd->lmd_flags |= LMD_FLG_NOMGS;
2777 } else if (strncmp(s1, "noscrub", 7) == 0) {
2778 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
2780 } else if (strncmp(s1, PARAM_MGSNODE,
2781 sizeof(PARAM_MGSNODE) - 1) == 0) {
2782 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
2783 /* Assume the next mount opt is the first
2784 invalid nid we get to. */
2785 rc = lmd_parse_mgs(lmd, &s2);
2789 } else if (strncmp(s1, "writeconf", 9) == 0) {
2790 lmd->lmd_flags |= LMD_FLG_WRITECONF;
2792 } else if (strncmp(s1, "virgin", 6) == 0) {
2793 lmd->lmd_flags |= LMD_FLG_VIRGIN;
2795 } else if (strncmp(s1, "noprimnode", 10) == 0) {
2796 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
2798 } else if (strncmp(s1, "mgssec=", 7) == 0) {
2799 rc = lmd_parse_mgssec(lmd, s1 + 7);
2803 /* ost exclusion list */
2804 } else if (strncmp(s1, "exclude=", 8) == 0) {
2805 rc = lmd_make_exclusion(lmd, s1 + 7);
2809 } else if (strncmp(s1, "mgs", 3) == 0) {
2811 lmd->lmd_flags |= LMD_FLG_MGS;
2813 } else if (strncmp(s1, "svname=", 7) == 0) {
2814 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
2818 } else if (strncmp(s1, "param=", 6) == 0) {
2820 char *tail = strchr(s1 + 6, ',');
2822 length = strlen(s1);
2826 strncat(lmd->lmd_params, s1 + 6, length);
2827 strcat(lmd->lmd_params, " ");
2829 } else if (strncmp(s1, "osd=", 4) == 0) {
2830 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
2835 /* Linux 2.4 doesn't pass the device, so we stuck it at the
2836 end of the options. */
2837 else if (strncmp(s1, "device=", 7) == 0) {
2839 /* terminate options right before device. device
2840 must be the last one. */
2846 s2 = strchr(s1, ',');
2854 memmove(s1, s2, strlen(s2) + 1);
2860 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
2861 "(need mount option 'device=...')\n");
2865 s1 = strstr(devname, ":/");
2868 lmd->lmd_flags |= LMD_FLG_CLIENT;
2869 /* Remove leading /s from fsname */
2870 while (*++s1 == '/') ;
2871 /* Freed in lustre_free_lsi */
2872 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
2873 if (!lmd->lmd_profile)
2875 sprintf(lmd->lmd_profile, "%s-client", s1);
2878 /* Freed in lustre_free_lsi */
2879 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
2882 strcpy(lmd->lmd_dev, devname);
2884 /* Save mount options */
2885 s1 = options + strlen(options) - 1;
2886 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
2888 if (*options != 0) {
2889 /* Freed in lustre_free_lsi */
2890 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
2893 strcpy(lmd->lmd_opts, options);
2897 lmd->lmd_magic = LMD_MAGIC;
2902 CERROR("Bad mount options %s\n", options);
2906 struct lustre_mount_data2 {
2908 struct vfsmount *lmd2_mnt;
2911 /** This is the entry point for the mount call into Lustre.
2912 * This is called when a server or client is mounted,
2913 * and this is where we start setting things up.
2914 * @param data Mount options (e.g. -o flock,abort_recov)
2916 int lustre_fill_super(struct super_block *sb, void *data, int silent)
2918 struct lustre_mount_data *lmd;
2919 struct lustre_mount_data2 *lmd2 = data;
2920 struct lustre_sb_info *lsi;
2924 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
2926 lsi = lustre_init_lsi(sb);
2932 * Disable lockdep during mount, because mount locking patterns are
2938 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
2940 obd_zombie_barrier();
2942 /* Figure out the lmd from the mount options */
2943 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
2945 GOTO(out, rc = -EINVAL);
2948 if (lmd_is_client(lmd)) {
2949 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
2950 if (!client_fill_super) {
2951 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
2952 "client mount! Is the 'lustre' "
2953 "module loaded?\n");
2957 rc = lustre_start_mgc(sb);
2962 /* Connect and start */
2963 /* (should always be ll_fill_super) */
2964 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
2965 /* c_f_s will call lustre_common_put_super on failure */
2968 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
2969 rc = server_fill_super(sb);
2970 /* s_f_s calls lustre_start_mgc after the mount because we need
2971 the MGS nids which are stored on disk. Plus, we may
2972 need to start the MGS first. */
2973 /* s_f_s will call server_put_super on failure */
2976 /* If error happens in fill_super() call, @lsi will be killed there.
2977 * This is why we do not put it here. */
2981 CERROR("Unable to mount %s (%d)\n",
2982 s2lsi(sb) ? lmd->lmd_dev : "", rc);
2984 CDEBUG(D_SUPER, "Mount %s complete\n",
2992 /* We can't call ll_fill_super by name because it lives in a module that
2993 must be loaded after this one. */
2994 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
2995 struct vfsmount *mnt))
2997 client_fill_super = cfs;
2999 EXPORT_SYMBOL(lustre_register_client_fill_super);
3001 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
3003 kill_super_cb = cfs;
3005 EXPORT_SYMBOL(lustre_register_kill_super_cb);
3007 /***************** FS registration ******************/
3008 #ifdef HAVE_FSTYPE_MOUNT
3009 struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
3010 const char *devname, void *data)
3012 struct lustre_mount_data2 lmd2 = { data, NULL };
3014 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
3017 int lustre_get_sb(struct file_system_type *fs_type, int flags,
3018 const char *devname, void * data, struct vfsmount *mnt)
3020 struct lustre_mount_data2 lmd2 = { data, mnt };
3022 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
3026 void lustre_kill_super(struct super_block *sb)
3028 struct lustre_sb_info *lsi = s2lsi(sb);
3030 if (kill_super_cb && lsi && !IS_SERVER(lsi))
3031 (*kill_super_cb)(sb);
3033 kill_anon_super(sb);
3036 /** Register the "lustre" fs type
3038 struct file_system_type lustre_fs_type = {
3039 .owner = THIS_MODULE,
3041 #ifdef HAVE_FSTYPE_MOUNT
3042 .mount = lustre_mount,
3044 .get_sb = lustre_get_sb,
3046 .kill_sb = lustre_kill_super,
3047 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
3048 #ifdef FS_HAS_FIEMAP
3051 LL_RENAME_DOES_D_MOVE,
3054 int lustre_register_fs(void)
3056 return register_filesystem(&lustre_fs_type);
3059 int lustre_unregister_fs(void)
3061 return unregister_filesystem(&lustre_fs_type);
3064 EXPORT_SYMBOL(server_mti_print);