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, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/obdclass/obd_mount.c
38 * Client 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
50 #include <lustre_fsfilt.h>
51 #include <obd_class.h>
52 #include <lustre/lustre_user.h>
53 #include <linux/version.h>
54 #include <lustre_log.h>
55 #include <lustre_disk.h>
56 #include <lustre_param.h>
58 static int (*client_fill_super)(struct super_block *sb,
59 struct vfsmount *mnt);
61 static void (*kill_super_cb)(struct super_block *sb);
63 /**************** config llog ********************/
65 /** Get a config log from the MGS and process it.
66 * This func is called for both clients and servers.
67 * Continue to process new statements appended to the logs
68 * (whenever the config lock is revoked) until lustre_end_log
70 * @param sb The superblock is used by the MGC to write to the local copy of
72 * @param logname The name of the llog to replicate from the MGS
73 * @param cfg Since the same mgc may be used to follow multiple config logs
74 * (e.g. ost1, ost2, client), the config_llog_instance keeps the state for
75 * this log, and is added to the mgc's list of logs to follow.
77 int lustre_process_log(struct super_block *sb, char *logname,
78 struct config_llog_instance *cfg)
80 struct lustre_cfg *lcfg;
81 struct lustre_cfg_bufs *bufs;
82 struct lustre_sb_info *lsi = s2lsi(sb);
83 struct obd_device *mgc = lsi->lsi_mgc;
94 /* mgc_process_config */
95 lustre_cfg_bufs_reset(bufs, mgc->obd_name);
96 lustre_cfg_bufs_set_string(bufs, 1, logname);
97 lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg));
98 lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb));
99 lcfg = lustre_cfg_new(LCFG_LOG_START, bufs);
100 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
101 lustre_cfg_free(lcfg);
106 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
107 "failed from the MGS (%d). Make sure this "
108 "client and the MGS are running compatible "
109 "versions of Lustre.\n",
110 mgc->obd_name, logname, rc);
113 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
114 "failed (%d). This may be the result of "
115 "communication errors between this node and "
116 "the MGS, a bad configuration, or other "
117 "errors. See the syslog for more "
118 "information.\n", mgc->obd_name, logname,
121 /* class_obd_list(); */
124 EXPORT_SYMBOL(lustre_process_log);
126 /* Stop watching this config log for updates */
127 int lustre_end_log(struct super_block *sb, char *logname,
128 struct config_llog_instance *cfg)
130 struct lustre_cfg *lcfg;
131 struct lustre_cfg_bufs bufs;
132 struct lustre_sb_info *lsi = s2lsi(sb);
133 struct obd_device *mgc = lsi->lsi_mgc;
140 /* mgc_process_config */
141 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
142 lustre_cfg_bufs_set_string(&bufs, 1, logname);
144 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
145 lcfg = lustre_cfg_new(LCFG_LOG_END, &bufs);
146 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
147 lustre_cfg_free(lcfg);
150 EXPORT_SYMBOL(lustre_end_log);
152 /**************** obd start *******************/
154 /** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from
155 * lctl (and do for echo cli/srv.
157 int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
158 char *s1, char *s2, char *s3, char *s4)
160 struct lustre_cfg_bufs bufs;
161 struct lustre_cfg * lcfg = NULL;
164 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
165 cmd, s1, s2, s3, s4);
167 lustre_cfg_bufs_reset(&bufs, cfgname);
169 lustre_cfg_bufs_set_string(&bufs, 1, s1);
171 lustre_cfg_bufs_set_string(&bufs, 2, s2);
173 lustre_cfg_bufs_set_string(&bufs, 3, s3);
175 lustre_cfg_bufs_set_string(&bufs, 4, s4);
177 lcfg = lustre_cfg_new(cmd, &bufs);
178 lcfg->lcfg_nid = nid;
179 rc = class_process_config(lcfg);
180 lustre_cfg_free(lcfg);
183 EXPORT_SYMBOL(do_lcfg);
185 /** Call class_attach and class_setup. These methods in turn call
186 * obd type-specific methods.
188 int lustre_start_simple(char *obdname, char *type, char *uuid,
189 char *s1, char *s2, char *s3, char *s4)
192 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
194 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, 0, 0);
196 CERROR("%s attach error %d\n", obdname, rc);
199 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, s3, s4);
201 CERROR("%s setup error %d\n", obdname, rc);
202 do_lcfg(obdname, 0, LCFG_DETACH, 0, 0, 0, 0);
207 DEFINE_MUTEX(mgc_start_lock);
209 /** Set up a mgc obd to process startup logs
211 * \param sb [in] super block of the mgc obd
213 * \retval 0 success, otherwise error code
215 int lustre_start_mgc(struct super_block *sb)
217 struct obd_connect_data *data = NULL;
218 struct lustre_sb_info *lsi = s2lsi(sb);
219 struct obd_device *obd;
220 struct obd_export *exp;
221 struct obd_uuid *uuid;
224 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
227 int rc = 0, i = 0, j, len;
230 LASSERT(lsi->lsi_lmd);
232 /* Find the first non-lo MGS nid for our MGC name */
233 if (IS_SERVER(lsi)) {
234 /* mount -o mgsnode=nid */
235 ptr = lsi->lsi_lmd->lmd_mgs;
236 if (lsi->lsi_lmd->lmd_mgs &&
237 (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) {
239 } else if (IS_MGS(lsi)) {
240 lnet_process_id_t id;
241 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
242 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
249 } else { /* client */
250 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
251 ptr = lsi->lsi_lmd->lmd_dev;
252 if (class_parse_nid(ptr, &nid, &ptr) == 0)
256 CERROR("No valid MGS nids found.\n");
260 mutex_lock(&mgc_start_lock);
262 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(libcfs_nid2str(nid)) + 1;
263 OBD_ALLOC(mgcname, len);
264 OBD_ALLOC(niduuid, len + 2);
265 if (!mgcname || !niduuid)
266 GOTO(out_free, rc = -ENOMEM);
267 sprintf(mgcname, "%s%s", LUSTRE_MGC_OBDNAME, libcfs_nid2str(nid));
269 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
273 GOTO(out_free, rc = -ENOMEM);
275 obd = class_name2obd(mgcname);
276 if (obd && !obd->obd_stopping) {
277 rc = obd_set_info_async(NULL, obd->obd_self_export,
278 strlen(KEY_MGSSEC), KEY_MGSSEC,
279 strlen(mgssec), mgssec, NULL);
283 /* Re-using an existing MGC */
284 cfs_atomic_inc(&obd->u.cli.cl_mgc_refcount);
286 /* IR compatibility check, only for clients */
287 if (lmd_is_client(lsi->lsi_lmd)) {
289 int vallen = sizeof(*data);
290 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
292 rc = obd_get_info(NULL, obd->obd_self_export,
293 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
294 &vallen, data, NULL);
296 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
297 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
298 /* LMD_FLG_NOIR is for test purpose only */
300 "Trying to mount a client with IR setting "
301 "not compatible with current mgc. "
302 "Force to use current mgc setting that is "
304 has_ir ? "enabled" : "disabled");
306 *flags &= ~LMD_FLG_NOIR;
308 *flags |= LMD_FLG_NOIR;
313 /* If we are restarting the MGS, don't try to keep the MGC's
314 old connection, or registration will fail. */
316 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
320 /* Try all connections, but only once (again).
321 We don't want to block another target from starting
322 (using its local copy of the log), but we do want to connect
323 if at all possible. */
325 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
326 rc = obd_set_info_async(NULL, obd->obd_self_export,
327 sizeof(KEY_INIT_RECOV_BACKUP),
328 KEY_INIT_RECOV_BACKUP,
329 sizeof(recov_bk), &recov_bk, NULL);
333 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
335 /* Add the primary nids for the MGS */
337 sprintf(niduuid, "%s_%x", mgcname, i);
338 if (IS_SERVER(lsi)) {
339 ptr = lsi->lsi_lmd->lmd_mgs;
341 /* Use local nids (including LO) */
342 lnet_process_id_t id;
343 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
344 rc = do_lcfg(mgcname, id.nid,
345 LCFG_ADD_UUID, niduuid, 0,0,0);
348 /* Use mgsnode= nids */
349 /* mount -o mgsnode=nid */
350 if (lsi->lsi_lmd->lmd_mgs) {
351 ptr = lsi->lsi_lmd->lmd_mgs;
352 } else if (class_find_param(ptr, PARAM_MGSNODE,
354 CERROR("No MGS nids given.\n");
355 GOTO(out_free, rc = -EINVAL);
357 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
358 rc = do_lcfg(mgcname, nid,
359 LCFG_ADD_UUID, niduuid, 0,0,0);
363 } else { /* client */
364 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
365 ptr = lsi->lsi_lmd->lmd_dev;
366 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
367 rc = do_lcfg(mgcname, nid,
368 LCFG_ADD_UUID, niduuid, 0,0,0);
370 /* Stop at the first failover nid */
376 CERROR("No valid MGS nids found.\n");
377 GOTO(out_free, rc = -EINVAL);
379 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
381 /* Random uuid for MGC allows easier reconnects */
383 ll_generate_random_uuid(uuidc);
384 class_uuid_unparse(uuidc, uuid);
387 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
388 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
394 /* Add any failover MGS nids */
396 while (ptr && ((*ptr == ':' ||
397 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) {
398 /* New failover node */
399 sprintf(niduuid, "%s_%x", mgcname, i);
401 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
403 rc = do_lcfg(mgcname, nid,
404 LCFG_ADD_UUID, niduuid, 0,0,0);
409 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
417 lsi->lsi_lmd->lmd_mgs_failnodes = i;
419 obd = class_name2obd(mgcname);
421 CERROR("Can't find mgcobd %s\n", mgcname);
422 GOTO(out_free, rc = -ENOTCONN);
425 rc = obd_set_info_async(NULL, obd->obd_self_export,
426 strlen(KEY_MGSSEC), KEY_MGSSEC,
427 strlen(mgssec), mgssec, NULL);
431 /* Keep a refcount of servers/clients who started with "mount",
432 so we know when we can get rid of the mgc. */
433 cfs_atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
435 /* Try all connections, but only once. */
437 rc = obd_set_info_async(NULL, obd->obd_self_export,
438 sizeof(KEY_INIT_RECOV_BACKUP),
439 KEY_INIT_RECOV_BACKUP,
440 sizeof(recov_bk), &recov_bk, NULL);
443 CWARN("can't set %s %d\n", KEY_INIT_RECOV_BACKUP, rc);
445 /* We connect to the MGS at setup, and don't disconnect until cleanup */
446 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
447 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV |
448 OBD_CONNECT_LVB_TYPE;
450 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 2, 50, 0)
451 data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB;
453 #warning "LU-1644: Remove old OBD_CONNECT_MNE_SWAB fixup and imp_need_mne_swab"
456 if (lmd_is_client(lsi->lsi_lmd) &&
457 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
458 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
459 data->ocd_version = LUSTRE_VERSION_CODE;
460 rc = obd_connect(NULL, &exp, obd, &(obd->obd_uuid), data, NULL);
462 CERROR("connect failed %d\n", rc);
466 obd->u.cli.cl_mgc_mgsexp = exp;
469 /* Keep the mgc info in the sb. Note that many lsi's can point
473 mutex_unlock(&mgc_start_lock);
478 OBD_FREE(mgcname, len);
480 OBD_FREE(niduuid, len + 2);
484 static int lustre_stop_mgc(struct super_block *sb)
486 struct lustre_sb_info *lsi = s2lsi(sb);
487 struct obd_device *obd;
488 char *niduuid = 0, *ptr = 0;
489 int i, rc = 0, len = 0;
499 mutex_lock(&mgc_start_lock);
500 LASSERT(cfs_atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
501 if (!cfs_atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
502 /* This is not fatal, every client that stops
503 will call in here. */
504 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
505 cfs_atomic_read(&obd->u.cli.cl_mgc_refcount));
506 GOTO(out, rc = -EBUSY);
509 /* The MGC has no recoverable data in any case.
510 * force shotdown set in umount_begin */
511 obd->obd_no_recov = 1;
513 if (obd->u.cli.cl_mgc_mgsexp) {
514 /* An error is not fatal, if we are unable to send the
515 disconnect mgs ping evictor cleans up the export */
516 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
518 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
521 /* Save the obdname for cleaning the nid uuids, which are
523 len = strlen(obd->obd_name) + 6;
524 OBD_ALLOC(niduuid, len);
526 strcpy(niduuid, obd->obd_name);
527 ptr = niduuid + strlen(niduuid);
530 rc = class_manual_cleanup(obd);
534 /* Clean the nid uuids */
536 GOTO(out, rc = -ENOMEM);
538 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
539 sprintf(ptr, "_%x", i);
540 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
543 CERROR("del MDC UUID %s failed: rc = %d\n",
548 OBD_FREE(niduuid, len);
550 /* class_import_put will get rid of the additional connections */
551 mutex_unlock(&mgc_start_lock);
555 /***************** lustre superblock **************/
557 struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
559 struct lustre_sb_info *lsi;
565 OBD_ALLOC_PTR(lsi->lsi_lmd);
571 lsi->lsi_lmd->lmd_exclude_count = 0;
572 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
573 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
574 s2lsi_nocast(sb) = lsi;
575 /* we take 1 extra ref for our setup */
576 cfs_atomic_set(&lsi->lsi_mounts, 1);
578 /* Default umount style */
579 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
584 static int lustre_free_lsi(struct super_block *sb)
586 struct lustre_sb_info *lsi = s2lsi(sb);
589 LASSERT(lsi != NULL);
590 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
592 /* someone didn't call server_put_mount. */
593 LASSERT(cfs_atomic_read(&lsi->lsi_mounts) == 0);
595 if (lsi->lsi_lmd != NULL) {
596 if (lsi->lsi_lmd->lmd_dev != NULL)
597 OBD_FREE(lsi->lsi_lmd->lmd_dev,
598 strlen(lsi->lsi_lmd->lmd_dev) + 1);
599 if (lsi->lsi_lmd->lmd_profile != NULL)
600 OBD_FREE(lsi->lsi_lmd->lmd_profile,
601 strlen(lsi->lsi_lmd->lmd_profile) + 1);
602 if (lsi->lsi_lmd->lmd_mgssec != NULL)
603 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
604 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
605 if (lsi->lsi_lmd->lmd_opts != NULL)
606 OBD_FREE(lsi->lsi_lmd->lmd_opts,
607 strlen(lsi->lsi_lmd->lmd_opts) + 1);
608 if (lsi->lsi_lmd->lmd_exclude_count)
609 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
610 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
611 lsi->lsi_lmd->lmd_exclude_count);
612 if (lsi->lsi_lmd->lmd_mgs != NULL)
613 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
614 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
615 if (lsi->lsi_lmd->lmd_osd_type != NULL)
616 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
617 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
618 if (lsi->lsi_lmd->lmd_params != NULL)
619 OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
621 OBD_FREE(lsi->lsi_lmd, sizeof(*lsi->lsi_lmd));
624 LASSERT(lsi->lsi_llsbi == NULL);
625 OBD_FREE(lsi, sizeof(*lsi));
626 s2lsi_nocast(sb) = NULL;
631 /* The lsi has one reference for every server that is using the disk -
632 e.g. MDT, MGS, and potentially MGC */
633 int lustre_put_lsi(struct super_block *sb)
635 struct lustre_sb_info *lsi = s2lsi(sb);
638 LASSERT(lsi != NULL);
640 CDEBUG(D_MOUNT, "put %p %d\n", sb, cfs_atomic_read(&lsi->lsi_mounts));
641 if (cfs_atomic_dec_and_test(&lsi->lsi_mounts)) {
642 if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
643 obd_disconnect(lsi->lsi_osd_exp);
644 /* wait till OSD is gone */
645 obd_zombie_barrier();
654 * <FSNAME><SEPERATOR><TYPE><INDEX>
655 * FSNAME is between 1 and 8 characters (inclusive).
656 * Excluded characters are '/' and ':'
657 * SEPERATOR is either ':' or '-'
658 * TYPE: "OST", "MDT", etc.
659 * INDEX: Hex representation of the index
662 /** Get the fsname ("lustre") from the server name ("lustre-OST003F").
663 * @param [in] svname server name including type and index
664 * @param [out] fsname Buffer to copy filesystem name prefix into.
665 * Must have at least 'strlen(fsname) + 1' chars.
666 * @param [out] endptr if endptr isn't NULL it is set to end of fsname
669 int server_name2fsname(const char *svname, char *fsname, const char **endptr)
673 dash = svname + strnlen(svname, 8); /* max fsname length is 8 */
674 for (; dash > svname && *dash != '-' && *dash != ':'; dash--)
679 if (fsname != NULL) {
680 strncpy(fsname, svname, dash - svname);
681 fsname[dash - svname] = '\0';
689 EXPORT_SYMBOL(server_name2fsname);
692 * Get service name (svname) from string
694 * if endptr isn't NULL it is set to end of fsname *
696 int server_name2svname(const char *label, char *svname, const char **endptr,
702 /* We use server_name2fsname() just for parsing */
703 rc = server_name2fsname(label, NULL, &dash);
710 if (strlcpy(svname, dash + 1, svsize) >= svsize)
715 EXPORT_SYMBOL(server_name2svname);
718 /* Get the index from the obd name.
721 if endptr isn't NULL it is set to end of name */
722 int server_name2index(const char *svname, __u32 *idx, const char **endptr)
728 /* We use server_name2fsname() just for parsing */
729 rc = server_name2fsname(svname, NULL, &dash);
735 if (strncmp(dash, "MDT", 3) == 0)
736 rc = LDD_F_SV_TYPE_MDT;
737 else if (strncmp(dash, "OST", 3) == 0)
738 rc = LDD_F_SV_TYPE_OST;
744 if (strncmp(dash, "all", 3) == 0) {
747 return rc | LDD_F_SV_ALL;
750 index = simple_strtoul(dash, (char **)endptr, 16);
754 /* Account for -mdc after index that is possible when specifying mdt */
755 if (endptr != NULL && strncmp(LUSTRE_MDC_NAME, *endptr + 1,
756 sizeof(LUSTRE_MDC_NAME)-1) == 0)
757 *endptr += sizeof(LUSTRE_MDC_NAME);
761 EXPORT_SYMBOL(server_name2index);
763 /*************** mount common betweeen server and client ***************/
766 int lustre_common_put_super(struct super_block *sb)
771 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
773 /* Drop a ref to the MGC */
774 rc = lustre_stop_mgc(sb);
775 if (rc && (rc != -ENOENT)) {
777 CERROR("Can't stop MGC: %d\n", rc);
780 /* BUSY just means that there's some other obd that
781 needs the mgc. Let him clean it up. */
782 CDEBUG(D_MOUNT, "MGC still in use\n");
784 /* Drop a ref to the mounted disk */
789 EXPORT_SYMBOL(lustre_common_put_super);
791 static void lmd_print(struct lustre_mount_data *lmd)
795 PRINT_CMD(D_MOUNT, " mount data:\n");
796 if (lmd_is_client(lmd))
797 PRINT_CMD(D_MOUNT, "profile: %s\n", lmd->lmd_profile);
798 PRINT_CMD(D_MOUNT, "device: %s\n", lmd->lmd_dev);
799 PRINT_CMD(D_MOUNT, "flags: %x\n", lmd->lmd_flags);
802 PRINT_CMD(D_MOUNT, "options: %s\n", lmd->lmd_opts);
804 if (lmd->lmd_recovery_time_soft)
805 PRINT_CMD(D_MOUNT, "recovery time soft: %d\n",
806 lmd->lmd_recovery_time_soft);
808 if (lmd->lmd_recovery_time_hard)
809 PRINT_CMD(D_MOUNT, "recovery time hard: %d\n",
810 lmd->lmd_recovery_time_hard);
812 for (i = 0; i < lmd->lmd_exclude_count; i++) {
813 PRINT_CMD(D_MOUNT, "exclude %d: OST%04x\n", i,
814 lmd->lmd_exclude[i]);
818 /* Is this server on the exclusion list */
819 int lustre_check_exclusion(struct super_block *sb, char *svname)
821 struct lustre_sb_info *lsi = s2lsi(sb);
822 struct lustre_mount_data *lmd = lsi->lsi_lmd;
827 rc = server_name2index(svname, &index, NULL);
828 if (rc != LDD_F_SV_TYPE_OST)
829 /* Only exclude OSTs */
832 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
833 index, lmd->lmd_exclude_count, lmd->lmd_dev);
835 for(i = 0; i < lmd->lmd_exclude_count; i++) {
836 if (index == lmd->lmd_exclude[i]) {
837 CWARN("Excluding %s (on exclusion list)\n", svname);
844 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
845 static int lmd_make_exclusion(struct lustre_mount_data *lmd, const char *ptr)
847 const char *s1 = ptr, *s2;
848 __u32 index, *exclude_list;
852 /* The shortest an ost name can be is 8 chars: -OST0000.
853 We don't actually know the fsname at this time, so in fact
854 a user could specify any fsname. */
855 devmax = strlen(ptr) / 8 + 1;
857 /* temp storage until we figure out how many we have */
858 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
862 /* we enter this fn pointing at the '=' */
863 while (*s1 && *s1 != ' ' && *s1 != ',') {
865 rc = server_name2index(s1, &index, &s2);
867 CERROR("Can't parse server name '%s': rc = %d\n",
871 if (rc == LDD_F_SV_TYPE_OST)
872 exclude_list[lmd->lmd_exclude_count++] = index;
874 CDEBUG(D_MOUNT, "ignoring exclude %.*s: type = %#x\n",
875 (uint)(s2-s1), s1, rc);
877 /* now we are pointing at ':' (next exclude)
878 or ',' (end of excludes) */
879 if (lmd->lmd_exclude_count >= devmax)
882 if (rc >= 0) /* non-err */
885 if (lmd->lmd_exclude_count) {
886 /* permanent, freed in lustre_free_lsi */
887 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
888 lmd->lmd_exclude_count);
889 if (lmd->lmd_exclude) {
890 memcpy(lmd->lmd_exclude, exclude_list,
891 sizeof(index) * lmd->lmd_exclude_count);
894 lmd->lmd_exclude_count = 0;
897 OBD_FREE(exclude_list, sizeof(index) * devmax);
901 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
906 if (lmd->lmd_mgssec != NULL) {
907 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
908 lmd->lmd_mgssec = NULL;
911 tail = strchr(ptr, ',');
913 length = strlen(ptr);
917 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
918 if (lmd->lmd_mgssec == NULL)
921 memcpy(lmd->lmd_mgssec, ptr, length);
922 lmd->lmd_mgssec[length] = '\0';
926 static int lmd_parse_string(char **handle, char *ptr)
931 if ((handle == NULL) || (ptr == NULL))
934 if (*handle != NULL) {
935 OBD_FREE(*handle, strlen(*handle) + 1);
939 tail = strchr(ptr, ',');
941 length = strlen(ptr);
945 OBD_ALLOC(*handle, length + 1);
949 memcpy(*handle, ptr, length);
950 (*handle)[length] = '\0';
955 /* Collect multiple values for mgsnid specifiers */
956 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
964 /* Find end of nidlist */
965 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
966 length = tail - *ptr;
968 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
972 if (lmd->lmd_mgs != NULL)
973 oldlen = strlen(lmd->lmd_mgs) + 1;
975 OBD_ALLOC(mgsnid, oldlen + length + 1);
979 if (lmd->lmd_mgs != NULL) {
980 /* Multiple mgsnid= are taken to mean failover locations */
981 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
982 mgsnid[oldlen - 1] = ':';
983 OBD_FREE(lmd->lmd_mgs, oldlen);
985 memcpy(mgsnid + oldlen, *ptr, length);
986 mgsnid[oldlen + length] = '\0';
987 lmd->lmd_mgs = mgsnid;
993 /** Parse mount line options
994 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
995 * dev is passed as device=uml1:/lustre by mount.lustre
997 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
999 char *s1, *s2, *devname = NULL;
1000 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1006 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1007 "/sbin/mount.lustre is installed.\n");
1011 /* Options should be a string - try to detect old lmd data */
1012 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1013 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1014 "/sbin/mount.lustre. Please install "
1015 "version %s\n", LUSTRE_VERSION_STRING);
1018 lmd->lmd_magic = LMD_MAGIC;
1020 OBD_ALLOC(lmd->lmd_params, 4096);
1021 if (lmd->lmd_params == NULL)
1023 lmd->lmd_params[0] = '\0';
1025 /* Set default flags here */
1030 int time_min = OBD_RECOVERY_TIME_MIN;
1032 /* Skip whitespace and extra commas */
1033 while (*s1 == ' ' || *s1 == ',')
1036 /* Client options are parsed in ll_options: eg. flock,
1039 /* Parse non-ldiskfs options here. Rather than modifying
1040 ldiskfs, we just zero these out here */
1041 if (strncmp(s1, "abort_recov", 11) == 0) {
1042 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1044 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
1045 lmd->lmd_recovery_time_soft = max_t(int,
1046 simple_strtoul(s1 + 19, NULL, 10), time_min);
1048 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
1049 lmd->lmd_recovery_time_hard = max_t(int,
1050 simple_strtoul(s1 + 19, NULL, 10), time_min);
1052 } else if (strncmp(s1, "noir", 4) == 0) {
1053 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
1055 } else if (strncmp(s1, "nosvc", 5) == 0) {
1056 lmd->lmd_flags |= LMD_FLG_NOSVC;
1058 } else if (strncmp(s1, "nomgs", 5) == 0) {
1059 lmd->lmd_flags |= LMD_FLG_NOMGS;
1061 } else if (strncmp(s1, "noscrub", 7) == 0) {
1062 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
1064 } else if (strncmp(s1, PARAM_MGSNODE,
1065 sizeof(PARAM_MGSNODE) - 1) == 0) {
1066 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
1067 /* Assume the next mount opt is the first
1068 invalid nid we get to. */
1069 rc = lmd_parse_mgs(lmd, &s2);
1073 } else if (strncmp(s1, "writeconf", 9) == 0) {
1074 lmd->lmd_flags |= LMD_FLG_WRITECONF;
1076 } else if (strncmp(s1, "update", 6) == 0) {
1077 lmd->lmd_flags |= LMD_FLG_UPDATE;
1079 } else if (strncmp(s1, "virgin", 6) == 0) {
1080 lmd->lmd_flags |= LMD_FLG_VIRGIN;
1082 } else if (strncmp(s1, "noprimnode", 10) == 0) {
1083 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
1085 } else if (strncmp(s1, "mgssec=", 7) == 0) {
1086 rc = lmd_parse_mgssec(lmd, s1 + 7);
1090 /* ost exclusion list */
1091 } else if (strncmp(s1, "exclude=", 8) == 0) {
1092 rc = lmd_make_exclusion(lmd, s1 + 7);
1096 } else if (strncmp(s1, "mgs", 3) == 0) {
1098 lmd->lmd_flags |= LMD_FLG_MGS;
1100 } else if (strncmp(s1, "svname=", 7) == 0) {
1101 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
1105 } else if (strncmp(s1, "param=", 6) == 0) {
1107 char *tail = strchr(s1 + 6, ',');
1109 length = strlen(s1);
1113 strncat(lmd->lmd_params, s1 + 6, length);
1114 strcat(lmd->lmd_params, " ");
1116 } else if (strncmp(s1, "osd=", 4) == 0) {
1117 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
1122 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1123 end of the options. */
1124 else if (strncmp(s1, "device=", 7) == 0) {
1126 /* terminate options right before device. device
1127 must be the last one. */
1133 s2 = strchr(s1, ',');
1141 memmove(s1, s2, strlen(s2) + 1);
1147 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1148 "(need mount option 'device=...')\n");
1152 s1 = strstr(devname, ":/");
1155 lmd->lmd_flags |= LMD_FLG_CLIENT;
1156 /* Remove leading /s from fsname */
1157 while (*++s1 == '/') ;
1158 /* Freed in lustre_free_lsi */
1159 OBD_ALLOC(lmd->lmd_profile, strlen(s1) + 8);
1160 if (!lmd->lmd_profile)
1162 sprintf(lmd->lmd_profile, "%s-client", s1);
1165 /* Freed in lustre_free_lsi */
1166 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1169 strcpy(lmd->lmd_dev, devname);
1171 /* Save mount options */
1172 s1 = options + strlen(options) - 1;
1173 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1175 if (*options != 0) {
1176 /* Freed in lustre_free_lsi */
1177 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1180 strcpy(lmd->lmd_opts, options);
1184 lmd->lmd_magic = LMD_MAGIC;
1189 CERROR("Bad mount options %s\n", options);
1193 struct lustre_mount_data2 {
1195 struct vfsmount *lmd2_mnt;
1198 /** This is the entry point for the mount call into Lustre.
1199 * This is called when a server or client is mounted,
1200 * and this is where we start setting things up.
1201 * @param data Mount options (e.g. -o flock,abort_recov)
1203 int lustre_fill_super(struct super_block *sb, void *data, int silent)
1205 struct lustre_mount_data *lmd;
1206 struct lustre_mount_data2 *lmd2 = data;
1207 struct lustre_sb_info *lsi;
1211 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1213 lsi = lustre_init_lsi(sb);
1219 * Disable lockdep during mount, because mount locking patterns are
1225 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
1227 obd_zombie_barrier();
1229 /* Figure out the lmd from the mount options */
1230 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
1232 GOTO(out, rc = -EINVAL);
1235 if (lmd_is_client(lmd)) {
1236 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1237 if (!client_fill_super) {
1238 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1239 "client mount! Is the 'lustre' "
1240 "module loaded?\n");
1244 rc = lustre_start_mgc(sb);
1249 /* Connect and start */
1250 /* (should always be ll_fill_super) */
1251 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
1252 /* c_f_s will call lustre_common_put_super on failure */
1255 #ifdef HAVE_SERVER_SUPPORT
1256 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1257 rc = server_fill_super(sb);
1258 /* s_f_s calls lustre_start_mgc after the mount because we need
1259 the MGS nids which are stored on disk. Plus, we may
1260 need to start the MGS first. */
1261 /* s_f_s will call server_put_super on failure */
1263 CERROR("This is client-side-only module, "
1264 "cannot handle server mount.\n");
1269 /* If error happens in fill_super() call, @lsi will be killed there.
1270 * This is why we do not put it here. */
1274 CERROR("Unable to mount %s (%d)\n",
1275 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1277 CDEBUG(D_SUPER, "Mount %s complete\n",
1285 /* We can't call ll_fill_super by name because it lives in a module that
1286 must be loaded after this one. */
1287 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
1288 struct vfsmount *mnt))
1290 client_fill_super = cfs;
1292 EXPORT_SYMBOL(lustre_register_client_fill_super);
1294 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
1296 kill_super_cb = cfs;
1298 EXPORT_SYMBOL(lustre_register_kill_super_cb);
1300 /***************** FS registration ******************/
1301 #ifdef HAVE_FSTYPE_MOUNT
1302 struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
1303 const char *devname, void *data)
1305 struct lustre_mount_data2 lmd2 = { data, NULL };
1307 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
1310 int lustre_get_sb(struct file_system_type *fs_type, int flags,
1311 const char *devname, void * data, struct vfsmount *mnt)
1313 struct lustre_mount_data2 lmd2 = { data, mnt };
1315 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
1319 void lustre_kill_super(struct super_block *sb)
1321 struct lustre_sb_info *lsi = s2lsi(sb);
1323 if (kill_super_cb && lsi && !IS_SERVER(lsi))
1324 (*kill_super_cb)(sb);
1326 kill_anon_super(sb);
1329 /** Register the "lustre" fs type
1331 struct file_system_type lustre_fs_type = {
1332 .owner = THIS_MODULE,
1334 #ifdef HAVE_FSTYPE_MOUNT
1335 .mount = lustre_mount,
1337 .get_sb = lustre_get_sb,
1339 .kill_sb = lustre_kill_super,
1340 .fs_flags = FS_BINARY_MOUNTDATA | FS_REQUIRES_DEV |
1341 FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE,
1344 int lustre_register_fs(void)
1346 return register_filesystem(&lustre_fs_type);
1349 int lustre_unregister_fs(void)
1351 return unregister_filesystem(&lustre_fs_type);