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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/obdclass/obd_mount.c
34 * Client mount routines
36 * Author: Nathan Rutman <nathan@clusterfs.com>
40 #define DEBUG_SUBSYSTEM S_CLASS
41 #define D_MOUNT (D_SUPER|D_CONFIG/*|D_WARNING */)
42 #define PRINT_CMD CDEBUG
45 #include <obd_class.h>
46 #include <linux/version.h>
47 #include <lustre_log.h>
48 #include <lustre_disk.h>
49 #include <uapi/linux/lustre/lustre_param.h>
51 static int (*client_fill_super)(struct super_block *sb,
52 struct vfsmount *mnt);
54 static void (*kill_super_cb)(struct super_block *sb);
56 /**************** config llog ********************/
58 /** Get a config log from the MGS and process it.
59 * This func is called for both clients and servers.
60 * Continue to process new statements appended to the logs
61 * (whenever the config lock is revoked) until lustre_end_log
63 * @param sb The superblock is used by the MGC to write to the local copy of
65 * @param logname The name of the llog to replicate from the MGS
66 * @param cfg Since the same mgc may be used to follow multiple config logs
67 * (e.g. ost1, ost2, client), the config_llog_instance keeps the state for
68 * this log, and is added to the mgc's list of logs to follow.
70 int lustre_process_log(struct super_block *sb, char *logname,
71 struct config_llog_instance *cfg)
73 struct lustre_cfg *lcfg;
74 struct lustre_cfg_bufs *bufs;
75 struct lustre_sb_info *lsi = s2lsi(sb);
76 struct obd_device *mgc = lsi->lsi_mgc;
87 /* mgc_process_config */
88 lustre_cfg_bufs_reset(bufs, mgc->obd_name);
89 lustre_cfg_bufs_set_string(bufs, 1, logname);
90 lustre_cfg_bufs_set(bufs, 2, cfg, sizeof(*cfg));
91 lustre_cfg_bufs_set(bufs, 3, &sb, sizeof(sb));
92 OBD_ALLOC(lcfg, lustre_cfg_len(bufs->lcfg_bufcount, bufs->lcfg_buflen));
94 GOTO(out, rc = -ENOMEM);
95 lustre_cfg_init(lcfg, LCFG_LOG_START, bufs);
97 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
98 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
103 LCONSOLE_ERROR_MSG(0x15b, "%s: The configuration from log '%s'"
104 "failed from the MGS (%d). Make sure this "
105 "client and the MGS are running compatible "
106 "versions of Lustre.\n",
107 mgc->obd_name, logname, rc);
109 LCONSOLE_ERROR_MSG(0x15c, "%s: The configuration from log '%s' "
110 "failed (%d). This may be the result of "
111 "communication errors between this node and "
112 "the MGS, a bad configuration, or other "
113 "errors. See the syslog for more "
114 "information.\n", mgc->obd_name, logname,
117 /* class_obd_list(); */
120 EXPORT_SYMBOL(lustre_process_log);
122 /* Stop watching this config log for updates */
123 int lustre_end_log(struct super_block *sb, char *logname,
124 struct config_llog_instance *cfg)
126 struct lustre_cfg *lcfg;
127 struct lustre_cfg_bufs bufs;
128 struct lustre_sb_info *lsi = s2lsi(sb);
129 struct obd_device *mgc = lsi->lsi_mgc;
136 /* mgc_process_config */
137 lustre_cfg_bufs_reset(&bufs, mgc->obd_name);
138 lustre_cfg_bufs_set_string(&bufs, 1, logname);
140 lustre_cfg_bufs_set(&bufs, 2, cfg, sizeof(*cfg));
141 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
144 lustre_cfg_init(lcfg, LCFG_LOG_END, &bufs);
145 rc = obd_process_config(mgc, sizeof(*lcfg), lcfg);
146 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
149 EXPORT_SYMBOL(lustre_end_log);
151 /**************** obd start *******************/
153 /** lustre_cfg_bufs are a holdover from 1.4; we can still set these up from
154 * lctl (and do for echo cli/srv.
156 static int do_lcfg(char *cfgname, lnet_nid_t nid, int cmd,
157 char *s1, char *s2, char *s3, char *s4)
159 struct lustre_cfg_bufs bufs;
160 struct lustre_cfg *lcfg = NULL;
163 CDEBUG(D_TRACE, "lcfg %s %#x %s %s %s %s\n", cfgname,
164 cmd, s1, s2, s3, s4);
166 lustre_cfg_bufs_reset(&bufs, cfgname);
168 lustre_cfg_bufs_set_string(&bufs, 1, s1);
170 lustre_cfg_bufs_set_string(&bufs, 2, s2);
172 lustre_cfg_bufs_set_string(&bufs, 3, s3);
174 lustre_cfg_bufs_set_string(&bufs, 4, s4);
176 OBD_ALLOC(lcfg, lustre_cfg_len(bufs.lcfg_bufcount, bufs.lcfg_buflen));
179 lustre_cfg_init(lcfg, cmd, &bufs);
180 lcfg->lcfg_nid = nid;
181 rc = class_process_config(lcfg);
182 OBD_FREE(lcfg, lustre_cfg_len(lcfg->lcfg_bufcount, lcfg->lcfg_buflens));
186 /** Call class_attach and class_setup. These methods in turn call
187 * obd type-specific methods.
189 int lustre_start_simple(char *obdname, char *type, char *uuid,
190 char *s1, char *s2, char *s3, char *s4)
193 CDEBUG(D_MOUNT, "Starting obd %s (typ=%s)\n", obdname, type);
195 rc = do_lcfg(obdname, 0, LCFG_ATTACH, type, uuid, NULL, NULL);
197 CERROR("%s attach error %d\n", obdname, rc);
200 rc = do_lcfg(obdname, 0, LCFG_SETUP, s1, s2, s3, s4);
202 CERROR("%s setup error %d\n", obdname, rc);
203 do_lcfg(obdname, 0, LCFG_DETACH, NULL, NULL, NULL, NULL);
208 static DEFINE_MUTEX(mgc_start_lock);
210 /** Set up a mgc obd to process startup logs
212 * \param sb [in] super block of the mgc obd
214 * \retval 0 success, otherwise error code
216 int lustre_start_mgc(struct super_block *sb)
218 struct obd_connect_data *data = NULL;
219 struct lustre_sb_info *lsi = s2lsi(sb);
220 struct obd_device *obd;
221 struct obd_export *exp;
222 struct obd_uuid *uuid = NULL;
225 char nidstr[LNET_NIDSTR_SIZE];
226 char *mgcname = NULL, *niduuid = NULL, *mgssec = NULL;
228 int rc = 0, i = 0, j;
232 LASSERT(lsi->lsi_lmd);
234 /* Find the first non-lo MGS nid for our MGC name */
235 if (IS_SERVER(lsi)) {
236 /* mount -o mgsnode=nid */
237 ptr = lsi->lsi_lmd->lmd_mgs;
238 if (lsi->lsi_lmd->lmd_mgs &&
239 (class_parse_nid(lsi->lsi_lmd->lmd_mgs, &nid, &ptr) == 0)) {
241 } else if (IS_MGS(lsi)) {
242 struct lnet_process_id id;
244 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
245 if (LNET_NETTYP(LNET_NIDNET(id.nid)) == LOLND)
252 } else { /* client */
253 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
254 ptr = lsi->lsi_lmd->lmd_dev;
255 if (class_parse_nid(ptr, &nid, &ptr) == 0)
259 CERROR("No valid MGS nids found.\n");
263 mutex_lock(&mgc_start_lock);
265 libcfs_nid2str_r(nid, nidstr, sizeof(nidstr));
266 len = strlen(LUSTRE_MGC_OBDNAME) + strlen(nidstr) + 1;
267 OBD_ALLOC(mgcname, len);
268 OBD_ALLOC(niduuid, len + 2);
269 if (mgcname == NULL || niduuid == NULL)
270 GOTO(out_free, rc = -ENOMEM);
271 snprintf(mgcname, len, "%s%s", LUSTRE_MGC_OBDNAME, nidstr);
273 mgssec = lsi->lsi_lmd->lmd_mgssec ? lsi->lsi_lmd->lmd_mgssec : "";
277 GOTO(out_free, rc = -ENOMEM);
279 obd = class_name2obd(mgcname);
280 if (obd && !obd->obd_stopping) {
283 rc = obd_set_info_async(NULL, obd->obd_self_export,
284 strlen(KEY_MGSSEC), KEY_MGSSEC,
285 strlen(mgssec), mgssec, NULL);
289 /* Re-using an existing MGC */
290 atomic_inc(&obd->u.cli.cl_mgc_refcount);
292 /* IR compatibility check, only for clients */
293 if (lmd_is_client(lsi->lsi_lmd)) {
295 int vallen = sizeof(*data);
296 __u32 *flags = &lsi->lsi_lmd->lmd_flags;
298 rc = obd_get_info(NULL, obd->obd_self_export,
299 strlen(KEY_CONN_DATA), KEY_CONN_DATA,
302 has_ir = OCD_HAS_FLAG(data, IMP_RECOV);
303 if (has_ir ^ !(*flags & LMD_FLG_NOIR)) {
304 /* LMD_FLG_NOIR is for test purpose only */
306 "Trying to mount a client with IR setting "
307 "not compatible with current mgc. "
308 "Force to use current mgc setting that is "
310 has_ir ? "enabled" : "disabled");
312 *flags &= ~LMD_FLG_NOIR;
314 *flags |= LMD_FLG_NOIR;
319 /* If we are restarting the MGS, don't try to keep the MGC's
320 old connection, or registration will fail. */
322 CDEBUG(D_MOUNT, "New MGS with live MGC\n");
326 /* Try all connections, but only once (again).
327 We don't want to block another target from starting
328 (using its local copy of the log), but we do want to connect
329 if at all possible. */
331 CDEBUG(D_MOUNT, "%s: Set MGC reconnect %d\n", mgcname,recov_bk);
332 rc = obd_set_info_async(NULL, obd->obd_self_export,
333 sizeof(KEY_INIT_RECOV_BACKUP),
334 KEY_INIT_RECOV_BACKUP,
335 sizeof(recov_bk), &recov_bk, NULL);
339 CDEBUG(D_MOUNT, "Start MGC '%s'\n", mgcname);
341 /* Add the primary nids for the MGS */
343 snprintf(niduuid, len + 2, "%s_%x", mgcname, i);
344 if (IS_SERVER(lsi)) {
345 ptr = lsi->lsi_lmd->lmd_mgs;
346 CDEBUG(D_MOUNT, "mgs nids %s.\n", ptr);
348 /* Use local nids (including LO) */
349 struct lnet_process_id id;
351 while ((rc = LNetGetId(i++, &id)) != -ENOENT) {
352 rc = do_lcfg(mgcname, id.nid, LCFG_ADD_UUID,
353 niduuid, NULL, NULL, NULL);
356 /* Use mgsnode= nids */
357 /* mount -o mgsnode=nid */
358 if (lsi->lsi_lmd->lmd_mgs) {
359 ptr = lsi->lsi_lmd->lmd_mgs;
360 } else if (class_find_param(ptr, PARAM_MGSNODE,
362 CERROR("No MGS nids given.\n");
363 GOTO(out_free, rc = -EINVAL);
366 * Add primary MGS nid(s).
367 * Multiple nids on one MGS node are separated
370 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
371 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
372 niduuid, NULL, NULL, NULL);
375 /* Stop at the first failover nid */
380 } else { /* client */
381 /* Use nids from mount line: uml1,1@elan:uml2,2@elan:/lustre */
382 ptr = lsi->lsi_lmd->lmd_dev;
383 while (class_parse_nid(ptr, &nid, &ptr) == 0) {
384 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
385 niduuid, NULL, NULL, NULL);
388 /* Stop at the first failover nid */
394 CERROR("No valid MGS nids found.\n");
395 GOTO(out_free, rc = -EINVAL);
397 lsi->lsi_lmd->lmd_mgs_failnodes = 1;
399 /* Random uuid for MGC allows easier reconnects */
402 GOTO(out_free, rc = -ENOMEM);
404 ll_generate_random_uuid(uuidc);
405 class_uuid_unparse(uuidc, uuid);
408 rc = lustre_start_simple(mgcname, LUSTRE_MGC_NAME,
409 (char *)uuid->uuid, LUSTRE_MGS_OBDNAME,
410 niduuid, NULL, NULL);
414 /* Add any failover MGS nids */
416 while (ptr && ((*ptr == ':' ||
417 class_find_param(ptr, PARAM_MGSNODE, &ptr) == 0))) {
418 /* New failover node */
419 sprintf(niduuid, "%s_%x", mgcname, i);
421 while (class_parse_nid_quiet(ptr, &nid, &ptr) == 0) {
422 rc = do_lcfg(mgcname, nid, LCFG_ADD_UUID,
423 niduuid, NULL, NULL, NULL);
430 rc = do_lcfg(mgcname, 0, LCFG_ADD_CONN,
431 niduuid, NULL, NULL, NULL);
439 lsi->lsi_lmd->lmd_mgs_failnodes = i;
441 obd = class_name2obd(mgcname);
443 CERROR("Can't find mgcobd %s\n", mgcname);
444 GOTO(out_free, rc = -ENOTCONN);
447 rc = obd_set_info_async(NULL, obd->obd_self_export,
448 strlen(KEY_MGSSEC), KEY_MGSSEC,
449 strlen(mgssec), mgssec, NULL);
453 /* Keep a refcount of servers/clients who started with "mount",
454 so we know when we can get rid of the mgc. */
455 atomic_set(&obd->u.cli.cl_mgc_refcount, 1);
457 /* We connect to the MGS at setup, and don't disconnect until cleanup */
458 data->ocd_connect_flags = OBD_CONNECT_VERSION | OBD_CONNECT_AT |
459 OBD_CONNECT_FULL20 | OBD_CONNECT_IMP_RECOV |
460 OBD_CONNECT_LVB_TYPE |
461 OBD_CONNECT_BULK_MBITS | OBD_CONNECT_BARRIER;
463 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 53, 0)
464 data->ocd_connect_flags |= OBD_CONNECT_MNE_SWAB;
467 if (lmd_is_client(lsi->lsi_lmd) &&
468 lsi->lsi_lmd->lmd_flags & LMD_FLG_NOIR)
469 data->ocd_connect_flags &= ~OBD_CONNECT_IMP_RECOV;
470 data->ocd_version = LUSTRE_VERSION_CODE;
471 rc = obd_connect(NULL, &exp, obd, uuid, data, NULL);
473 CERROR("connect failed %d\n", rc);
477 obd->u.cli.cl_mgc_mgsexp = exp;
480 /* Keep the mgc info in the sb. Note that many lsi's can point
484 mutex_unlock(&mgc_start_lock);
491 OBD_FREE(mgcname, len);
493 OBD_FREE(niduuid, len + 2);
497 static int lustre_stop_mgc(struct super_block *sb)
499 struct lustre_sb_info *lsi = s2lsi(sb);
500 struct obd_device *obd;
501 char *niduuid = NULL, *ptr = NULL;
502 int i, rc = 0, len = 0;
512 mutex_lock(&mgc_start_lock);
513 LASSERT(atomic_read(&obd->u.cli.cl_mgc_refcount) > 0);
514 if (!atomic_dec_and_test(&obd->u.cli.cl_mgc_refcount)) {
515 /* This is not fatal, every client that stops
516 will call in here. */
517 CDEBUG(D_MOUNT, "mgc still has %d references.\n",
518 atomic_read(&obd->u.cli.cl_mgc_refcount));
519 GOTO(out, rc = -EBUSY);
522 /* The MGC has no recoverable data in any case.
523 * force shotdown set in umount_begin */
524 obd->obd_no_recov = 1;
526 if (obd->u.cli.cl_mgc_mgsexp) {
527 /* An error is not fatal, if we are unable to send the
528 disconnect mgs ping evictor cleans up the export */
529 rc = obd_disconnect(obd->u.cli.cl_mgc_mgsexp);
531 CDEBUG(D_MOUNT, "disconnect failed %d\n", rc);
534 /* Save the obdname for cleaning the nid uuids, which are
536 len = strlen(obd->obd_name) + 6;
537 OBD_ALLOC(niduuid, len);
539 strcpy(niduuid, obd->obd_name);
540 ptr = niduuid + strlen(niduuid);
543 rc = class_manual_cleanup(obd);
547 /* Clean the nid uuids */
549 GOTO(out, rc = -ENOMEM);
551 for (i = 0; i < lsi->lsi_lmd->lmd_mgs_failnodes; i++) {
552 sprintf(ptr, "_%x", i);
553 rc = do_lcfg(LUSTRE_MGC_OBDNAME, 0, LCFG_DEL_UUID,
554 niduuid, NULL, NULL, NULL);
556 CERROR("del MDC UUID %s failed: rc = %d\n",
561 OBD_FREE(niduuid, len);
563 /* class_import_put will get rid of the additional connections */
564 mutex_unlock(&mgc_start_lock);
568 /***************** lustre superblock **************/
570 static struct lustre_sb_info *lustre_init_lsi(struct super_block *sb)
572 struct lustre_sb_info *lsi;
578 OBD_ALLOC_PTR(lsi->lsi_lmd);
584 lsi->lsi_lmd->lmd_exclude_count = 0;
585 lsi->lsi_lmd->lmd_recovery_time_soft = 0;
586 lsi->lsi_lmd->lmd_recovery_time_hard = 0;
587 s2lsi_nocast(sb) = lsi;
588 /* we take 1 extra ref for our setup */
589 atomic_set(&lsi->lsi_mounts, 1);
591 /* Default umount style */
592 lsi->lsi_flags = LSI_UMOUNT_FAILOVER;
593 INIT_LIST_HEAD(&lsi->lsi_lwp_list);
594 spin_lock_init(&lsi->lsi_lwp_lock);
599 static int lustre_free_lsi(struct super_block *sb)
601 struct lustre_sb_info *lsi = s2lsi(sb);
604 LASSERT(lsi != NULL);
605 CDEBUG(D_MOUNT, "Freeing lsi %p\n", lsi);
607 /* someone didn't call server_put_mount. */
608 LASSERT(atomic_read(&lsi->lsi_mounts) == 0);
610 if (lsi->lsi_lmd != NULL) {
611 if (lsi->lsi_lmd->lmd_dev != NULL)
612 OBD_FREE(lsi->lsi_lmd->lmd_dev,
613 strlen(lsi->lsi_lmd->lmd_dev) + 1);
614 if (lsi->lsi_lmd->lmd_profile != NULL)
615 OBD_FREE(lsi->lsi_lmd->lmd_profile,
616 strlen(lsi->lsi_lmd->lmd_profile) + 1);
617 if (lsi->lsi_lmd->lmd_fileset != NULL)
618 OBD_FREE(lsi->lsi_lmd->lmd_fileset,
619 strlen(lsi->lsi_lmd->lmd_fileset) + 1);
620 if (lsi->lsi_lmd->lmd_mgssec != NULL)
621 OBD_FREE(lsi->lsi_lmd->lmd_mgssec,
622 strlen(lsi->lsi_lmd->lmd_mgssec) + 1);
623 if (lsi->lsi_lmd->lmd_opts != NULL)
624 OBD_FREE(lsi->lsi_lmd->lmd_opts,
625 strlen(lsi->lsi_lmd->lmd_opts) + 1);
626 if (lsi->lsi_lmd->lmd_exclude_count)
627 OBD_FREE(lsi->lsi_lmd->lmd_exclude,
628 sizeof(lsi->lsi_lmd->lmd_exclude[0]) *
629 lsi->lsi_lmd->lmd_exclude_count);
630 if (lsi->lsi_lmd->lmd_mgs != NULL)
631 OBD_FREE(lsi->lsi_lmd->lmd_mgs,
632 strlen(lsi->lsi_lmd->lmd_mgs) + 1);
633 if (lsi->lsi_lmd->lmd_osd_type != NULL)
634 OBD_FREE(lsi->lsi_lmd->lmd_osd_type,
635 strlen(lsi->lsi_lmd->lmd_osd_type) + 1);
636 if (lsi->lsi_lmd->lmd_params != NULL)
637 OBD_FREE(lsi->lsi_lmd->lmd_params, 4096);
638 if (lsi->lsi_lmd->lmd_nidnet != NULL)
639 OBD_FREE(lsi->lsi_lmd->lmd_nidnet,
640 strlen(lsi->lsi_lmd->lmd_nidnet) + 1);
642 OBD_FREE_PTR(lsi->lsi_lmd);
645 LASSERT(lsi->lsi_llsbi == NULL);
647 s2lsi_nocast(sb) = NULL;
652 /* The lsi has one reference for every server that is using the disk -
653 e.g. MDT, MGS, and potentially MGC */
654 int lustre_put_lsi(struct super_block *sb)
656 struct lustre_sb_info *lsi = s2lsi(sb);
659 LASSERT(lsi != NULL);
661 CDEBUG(D_MOUNT, "put %p %d\n", sb, atomic_read(&lsi->lsi_mounts));
662 if (atomic_dec_and_test(&lsi->lsi_mounts)) {
663 if (IS_SERVER(lsi) && lsi->lsi_osd_exp) {
664 lu_device_put(&lsi->lsi_dt_dev->dd_lu_dev);
665 lsi->lsi_osd_exp->exp_obd->obd_lvfs_ctxt.dt = NULL;
666 lsi->lsi_dt_dev = NULL;
667 obd_disconnect(lsi->lsi_osd_exp);
668 /* wait till OSD is gone */
669 obd_zombie_barrier();
678 * <FSNAME><SEPARATOR><TYPE><INDEX>
679 * FSNAME is between 1 and 8 characters (inclusive).
680 * Excluded characters are '/' and ':'
681 * SEPARATOR is either ':' or '-'
682 * TYPE: "OST", "MDT", etc.
683 * INDEX: Hex representation of the index
686 /** Get the fsname ("lustre") from the server name ("lustre-OST003F").
687 * @param [in] svname server name including type and index
688 * @param [out] fsname Buffer to copy filesystem name prefix into.
689 * Must have at least 'strlen(fsname) + 1' chars.
690 * @param [out] endptr if endptr isn't NULL it is set to end of fsname
693 int server_name2fsname(const char *svname, char *fsname, const char **endptr)
697 dash = svname + strnlen(svname, LUSTRE_MAXFSNAME);
698 for (; dash > svname && *dash != '-' && *dash != ':'; dash--)
703 if (fsname != NULL) {
704 strncpy(fsname, svname, dash - svname);
705 fsname[dash - svname] = '\0';
713 EXPORT_SYMBOL(server_name2fsname);
716 * Get service name (svname) from string
718 * if endptr isn't NULL it is set to end of fsname *
720 int server_name2svname(const char *label, char *svname, const char **endptr,
726 /* We use server_name2fsname() just for parsing */
727 rc = server_name2fsname(label, NULL, &dash);
734 if (strlcpy(svname, dash + 1, svsize) >= svsize)
739 EXPORT_SYMBOL(server_name2svname);
742 * check server name is OST.
744 int server_name_is_ost(const char *svname)
749 /* We use server_name2fsname() just for parsing */
750 rc = server_name2fsname(svname, NULL, &dash);
756 if (strncmp(dash, "OST", 3) == 0)
760 EXPORT_SYMBOL(server_name_is_ost);
763 * Get the index from the target name MDTXXXX/OSTXXXX
764 * rc = server type, or rc < 0 on error
766 int target_name2index(const char *tgtname, __u32 *idx, const char **endptr)
768 const char *dash = tgtname;
772 if (strncmp(dash, "MDT", 3) == 0)
773 rc = LDD_F_SV_TYPE_MDT;
774 else if (strncmp(dash, "OST", 3) == 0)
775 rc = LDD_F_SV_TYPE_OST;
781 if (strncmp(dash, "all", 3) == 0) {
784 return rc | LDD_F_SV_ALL;
787 index = simple_strtoul(dash, (char **)endptr, 16);
792 EXPORT_SYMBOL(target_name2index);
794 /* Get the index from the obd name.
797 if endptr isn't NULL it is set to end of name */
798 int server_name2index(const char *svname, __u32 *idx, const char **endptr)
803 /* We use server_name2fsname() just for parsing */
804 rc = server_name2fsname(svname, NULL, &dash);
809 rc = target_name2index(dash, idx, endptr);
813 /* Account for -mdc after index that is possible when specifying mdt */
814 if (endptr != NULL && strncmp(LUSTRE_MDC_NAME, *endptr + 1,
815 sizeof(LUSTRE_MDC_NAME)-1) == 0)
816 *endptr += sizeof(LUSTRE_MDC_NAME);
820 EXPORT_SYMBOL(server_name2index);
822 /*************** mount common betweeen server and client ***************/
825 int lustre_common_put_super(struct super_block *sb)
830 CDEBUG(D_MOUNT, "dropping sb %p\n", sb);
832 /* Drop a ref to the MGC */
833 rc = lustre_stop_mgc(sb);
834 if (rc && (rc != -ENOENT)) {
836 CERROR("Can't stop MGC: %d\n", rc);
839 /* BUSY just means that there's some other obd that
840 needs the mgc. Let him clean it up. */
841 CDEBUG(D_MOUNT, "MGC still in use\n");
843 /* Drop a ref to the mounted disk */
848 EXPORT_SYMBOL(lustre_common_put_super);
850 static void lmd_print(struct lustre_mount_data *lmd)
854 PRINT_CMD(D_MOUNT, " mount data:\n");
855 if (lmd_is_client(lmd))
856 PRINT_CMD(D_MOUNT, "profile: %s\n", lmd->lmd_profile);
857 PRINT_CMD(D_MOUNT, "device: %s\n", lmd->lmd_dev);
858 PRINT_CMD(D_MOUNT, "flags: %x\n", lmd->lmd_flags);
861 PRINT_CMD(D_MOUNT, "options: %s\n", lmd->lmd_opts);
863 if (lmd->lmd_recovery_time_soft)
864 PRINT_CMD(D_MOUNT, "recovery time soft: %d\n",
865 lmd->lmd_recovery_time_soft);
867 if (lmd->lmd_recovery_time_hard)
868 PRINT_CMD(D_MOUNT, "recovery time hard: %d\n",
869 lmd->lmd_recovery_time_hard);
871 for (i = 0; i < lmd->lmd_exclude_count; i++) {
872 PRINT_CMD(D_MOUNT, "exclude %d: OST%04x\n", i,
873 lmd->lmd_exclude[i]);
877 /* Is this server on the exclusion list */
878 int lustre_check_exclusion(struct super_block *sb, char *svname)
880 struct lustre_sb_info *lsi = s2lsi(sb);
881 struct lustre_mount_data *lmd = lsi->lsi_lmd;
886 rc = server_name2index(svname, &index, NULL);
887 if (rc != LDD_F_SV_TYPE_OST)
888 /* Only exclude OSTs */
891 CDEBUG(D_MOUNT, "Check exclusion %s (%d) in %d of %s\n", svname,
892 index, lmd->lmd_exclude_count, lmd->lmd_dev);
894 for(i = 0; i < lmd->lmd_exclude_count; i++) {
895 if (index == lmd->lmd_exclude[i]) {
896 CWARN("Excluding %s (on exclusion list)\n", svname);
903 /* mount -v -o exclude=lustre-OST0001:lustre-OST0002 -t lustre ... */
904 static int lmd_make_exclusion(struct lustre_mount_data *lmd, const char *ptr)
906 const char *s1 = ptr, *s2;
912 /* The shortest an ost name can be is 8 chars: -OST0000.
913 We don't actually know the fsname at this time, so in fact
914 a user could specify any fsname. */
915 devmax = strlen(ptr) / 8 + 1;
917 /* temp storage until we figure out how many we have */
918 OBD_ALLOC(exclude_list, sizeof(index) * devmax);
922 /* we enter this fn pointing at the '=' */
923 while (*s1 && *s1 != ' ' && *s1 != ',') {
925 rc = server_name2index(s1, &index, &s2);
927 CERROR("Can't parse server name '%s': rc = %d\n",
931 if (rc == LDD_F_SV_TYPE_OST)
932 exclude_list[lmd->lmd_exclude_count++] = index;
934 CDEBUG(D_MOUNT, "ignoring exclude %.*s: type = %#x\n",
935 (uint)(s2-s1), s1, rc);
937 /* now we are pointing at ':' (next exclude)
938 or ',' (end of excludes) */
939 if (lmd->lmd_exclude_count >= devmax)
942 if (rc >= 0) /* non-err */
945 if (lmd->lmd_exclude_count) {
946 /* permanent, freed in lustre_free_lsi */
947 OBD_ALLOC(lmd->lmd_exclude, sizeof(index) *
948 lmd->lmd_exclude_count);
949 if (lmd->lmd_exclude) {
950 memcpy(lmd->lmd_exclude, exclude_list,
951 sizeof(index) * lmd->lmd_exclude_count);
954 lmd->lmd_exclude_count = 0;
957 OBD_FREE(exclude_list, sizeof(index) * devmax);
961 static int lmd_parse_mgssec(struct lustre_mount_data *lmd, char *ptr)
966 if (lmd->lmd_mgssec != NULL) {
967 OBD_FREE(lmd->lmd_mgssec, strlen(lmd->lmd_mgssec) + 1);
968 lmd->lmd_mgssec = NULL;
971 tail = strchr(ptr, ',');
973 length = strlen(ptr);
977 OBD_ALLOC(lmd->lmd_mgssec, length + 1);
978 if (lmd->lmd_mgssec == NULL)
981 memcpy(lmd->lmd_mgssec, ptr, length);
982 lmd->lmd_mgssec[length] = '\0';
986 static int lmd_parse_network(struct lustre_mount_data *lmd, char *ptr)
991 if (lmd->lmd_nidnet != NULL) {
992 OBD_FREE(lmd->lmd_nidnet, strlen(lmd->lmd_nidnet) + 1);
993 lmd->lmd_nidnet = NULL;
996 tail = strchr(ptr, ',');
998 length = strlen(ptr);
1000 length = tail - ptr;
1002 OBD_ALLOC(lmd->lmd_nidnet, length + 1);
1003 if (lmd->lmd_nidnet == NULL)
1006 memcpy(lmd->lmd_nidnet, ptr, length);
1007 lmd->lmd_nidnet[length] = '\0';
1011 static int lmd_parse_string(char **handle, char *ptr)
1016 if ((handle == NULL) || (ptr == NULL))
1019 if (*handle != NULL) {
1020 OBD_FREE(*handle, strlen(*handle) + 1);
1024 tail = strchr(ptr, ',');
1026 length = strlen(ptr);
1028 length = tail - ptr;
1030 OBD_ALLOC(*handle, length + 1);
1031 if (*handle == NULL)
1034 memcpy(*handle, ptr, length);
1035 (*handle)[length] = '\0';
1040 /* Collect multiple values for mgsnid specifiers */
1041 static int lmd_parse_mgs(struct lustre_mount_data *lmd, char **ptr)
1049 /* Find end of nidlist */
1050 while (class_parse_nid_quiet(tail, &nid, &tail) == 0) {}
1051 length = tail - *ptr;
1053 LCONSOLE_ERROR_MSG(0x159, "Can't parse NID '%s'\n", *ptr);
1057 if (lmd->lmd_mgs != NULL)
1058 oldlen = strlen(lmd->lmd_mgs) + 1;
1060 OBD_ALLOC(mgsnid, oldlen + length + 1);
1064 if (lmd->lmd_mgs != NULL) {
1065 /* Multiple mgsnid= are taken to mean failover locations */
1066 memcpy(mgsnid, lmd->lmd_mgs, oldlen);
1067 mgsnid[oldlen - 1] = ':';
1068 OBD_FREE(lmd->lmd_mgs, oldlen);
1070 memcpy(mgsnid + oldlen, *ptr, length);
1071 mgsnid[oldlen + length] = '\0';
1072 lmd->lmd_mgs = mgsnid;
1079 * Find the first delimiter (comma or colon) from the specified \a buf and
1080 * make \a *endh point to the string starting with the delimiter. The commas
1081 * in expression list [...] will be skipped.
1083 * \param[in] buf a delimiter-separated string
1084 * \param[in] endh a pointer to a pointer that will point to the string
1085 * starting with the delimiter
1087 * \retval 0 if delimiter is found
1088 * \retval 1 if delimiter is not found
1090 static int lmd_find_delimiter(char *buf, char **endh)
1098 while (*c != '\0') {
1104 if ((*c == ',' || *c == ':') && skip == 0) {
1117 * Find the first valid string delimited by comma or colon from the specified
1118 * \a buf and parse it to see whether it's a valid nid list. If yes, \a *endh
1119 * will point to the next string starting with the delimiter.
1121 * \param[in] buf a delimiter-separated string
1122 * \param[in] endh a pointer to a pointer that will point to the string
1123 * starting with the delimiter
1125 * \retval 0 if the string is a valid nid list
1126 * \retval 1 if the string is not a valid nid list
1128 static int lmd_parse_nidlist(char *buf, char **endh)
1130 struct list_head nidlist;
1137 while (*buf == ',' || *buf == ':')
1139 if (*buf == ' ' || *buf == '/' || *buf == '\0')
1142 if (lmd_find_delimiter(buf, &endp) != 0)
1143 endp = buf + strlen(buf);
1148 INIT_LIST_HEAD(&nidlist);
1149 if (cfs_parse_nidlist(buf, strlen(buf), &nidlist) <= 0)
1151 cfs_free_nidlist(&nidlist);
1161 /** Parse mount line options
1162 * e.g. mount -v -t lustre -o abort_recov uml1:uml2:/lustre-client /mnt/lustre
1163 * dev is passed as device=uml1:/lustre by mount.lustre
1165 static int lmd_parse(char *options, struct lustre_mount_data *lmd)
1167 char *s1, *s2, *devname = NULL;
1168 struct lustre_mount_data *raw = (struct lustre_mount_data *)options;
1174 LCONSOLE_ERROR_MSG(0x162, "Missing mount data: check that "
1175 "/sbin/mount.lustre is installed.\n");
1179 /* Options should be a string - try to detect old lmd data */
1180 if ((raw->lmd_magic & 0xffffff00) == (LMD_MAGIC & 0xffffff00)) {
1181 LCONSOLE_ERROR_MSG(0x163, "You're using an old version of "
1182 "/sbin/mount.lustre. Please install "
1183 "version %s\n", LUSTRE_VERSION_STRING);
1186 lmd->lmd_magic = LMD_MAGIC;
1188 OBD_ALLOC(lmd->lmd_params, LMD_PARAMS_MAXLEN);
1189 if (lmd->lmd_params == NULL)
1191 lmd->lmd_params[0] = '\0';
1193 /* Set default flags here */
1198 int time_min = OBD_RECOVERY_TIME_MIN;
1201 /* Skip whitespace and extra commas */
1202 while (*s1 == ' ' || *s1 == ',')
1206 /* Client options are parsed in ll_options: eg. flock,
1209 /* Parse non-ldiskfs options here. Rather than modifying
1210 ldiskfs, we just zero these out here */
1211 if (strncmp(s1, "abort_recov", 11) == 0) {
1212 lmd->lmd_flags |= LMD_FLG_ABORT_RECOV;
1214 } else if (strncmp(s1, "recovery_time_soft=", 19) == 0) {
1215 lmd->lmd_recovery_time_soft =
1216 max_t(int, simple_strtoul(s1 + 19, NULL, 10),
1219 } else if (strncmp(s1, "recovery_time_hard=", 19) == 0) {
1220 lmd->lmd_recovery_time_hard =
1221 max_t(int, simple_strtoul(s1 + 19, NULL, 10),
1224 } else if (strncmp(s1, "noir", 4) == 0) {
1225 lmd->lmd_flags |= LMD_FLG_NOIR; /* test purpose only. */
1227 } else if (strncmp(s1, "nosvc", 5) == 0) {
1228 lmd->lmd_flags |= LMD_FLG_NOSVC;
1230 } else if (strncmp(s1, "nomgs", 5) == 0) {
1231 lmd->lmd_flags |= LMD_FLG_NOMGS;
1233 } else if (strncmp(s1, "noscrub", 7) == 0) {
1234 lmd->lmd_flags |= LMD_FLG_NOSCRUB;
1236 } else if (strncmp(s1, "skip_lfsck", 10) == 0) {
1237 lmd->lmd_flags |= LMD_FLG_SKIP_LFSCK;
1239 } else if (strncmp(s1, "rdonly_dev", 10) == 0) {
1240 lmd->lmd_flags |= LMD_FLG_DEV_RDONLY;
1242 } else if (strncmp(s1, PARAM_MGSNODE,
1243 sizeof(PARAM_MGSNODE) - 1) == 0) {
1244 s2 = s1 + sizeof(PARAM_MGSNODE) - 1;
1245 /* Assume the next mount opt is the first
1246 invalid nid we get to. */
1247 rc = lmd_parse_mgs(lmd, &s2);
1252 } else if (strncmp(s1, "writeconf", 9) == 0) {
1253 lmd->lmd_flags |= LMD_FLG_WRITECONF;
1255 } else if (strncmp(s1, "update", 6) == 0) {
1256 lmd->lmd_flags |= LMD_FLG_UPDATE;
1258 } else if (strncmp(s1, "virgin", 6) == 0) {
1259 lmd->lmd_flags |= LMD_FLG_VIRGIN;
1261 } else if (strncmp(s1, "noprimnode", 10) == 0) {
1262 lmd->lmd_flags |= LMD_FLG_NO_PRIMNODE;
1264 } else if (strncmp(s1, "mgssec=", 7) == 0) {
1265 rc = lmd_parse_mgssec(lmd, s1 + 7);
1269 /* ost exclusion list */
1270 } else if (strncmp(s1, "exclude=", 8) == 0) {
1271 rc = lmd_make_exclusion(lmd, s1 + 7);
1275 } else if (strncmp(s1, "mgs", 3) == 0) {
1277 lmd->lmd_flags |= LMD_FLG_MGS;
1279 } else if (strncmp(s1, "svname=", 7) == 0) {
1280 rc = lmd_parse_string(&lmd->lmd_profile, s1 + 7);
1284 } else if (strncmp(s1, "param=", 6) == 0) {
1285 size_t length, params_length;
1287 if (lmd_find_delimiter(s1 + 6, &tail) != 0)
1288 length = strlen(s1);
1290 char *param_str = tail + 1;
1291 int supplementary = 1;
1292 while (lmd_parse_nidlist(param_str,
1296 length = param_str - s1 - supplementary;
1299 params_length = strlen(lmd->lmd_params);
1300 if (params_length + length + 1 >= LMD_PARAMS_MAXLEN)
1302 strncat(lmd->lmd_params, s1 + 6, length);
1303 lmd->lmd_params[params_length + length] = '\0';
1304 strlcat(lmd->lmd_params, " ", LMD_PARAMS_MAXLEN);
1305 s3 = s1 + 6 + length;
1307 } else if (strncmp(s1, "osd=", 4) == 0) {
1308 rc = lmd_parse_string(&lmd->lmd_osd_type, s1 + 4);
1313 /* Linux 2.4 doesn't pass the device, so we stuck it at the
1314 end of the options. */
1315 else if (strncmp(s1, "device=", 7) == 0) {
1317 /* terminate options right before device. device
1318 must be the last one. */
1321 } else if (strncmp(s1, "network=", 8) == 0) {
1322 rc = lmd_parse_network(lmd, s1 + 8);
1329 s2 = strchr(s3, ',');
1337 memmove(s1, s2, strlen(s2) + 1);
1343 LCONSOLE_ERROR_MSG(0x164, "Can't find the device name "
1344 "(need mount option 'device=...')\n");
1348 s1 = strstr(devname, ":/");
1351 lmd->lmd_flags |= LMD_FLG_CLIENT;
1352 /* Remove leading /s from fsname */
1353 while (*++s1 == '/')
1356 while (*s2 != '/' && *s2 != '\0')
1358 /* Freed in lustre_free_lsi */
1359 OBD_ALLOC(lmd->lmd_profile, s2 - s1 + 8);
1360 if (!lmd->lmd_profile)
1363 strncat(lmd->lmd_profile, s1, s2 - s1);
1364 strncat(lmd->lmd_profile, "-client", 7);
1367 s2 = s1 + strlen(s1) - 1;
1368 /* Remove padding /s from fileset */
1372 OBD_ALLOC(lmd->lmd_fileset, s2 - s1 + 2);
1373 if (lmd->lmd_fileset == NULL) {
1374 OBD_FREE(lmd->lmd_profile,
1375 strlen(lmd->lmd_profile) + 1);
1378 strncat(lmd->lmd_fileset, s1, s2 - s1 + 1);
1382 if (lmd->lmd_nidnet != NULL) {
1383 /* 'network=' mount option forbidden for server */
1384 OBD_FREE(lmd->lmd_nidnet, strlen(lmd->lmd_nidnet) + 1);
1385 lmd->lmd_nidnet = NULL;
1387 CERROR("%s: option 'network=' not allowed for Lustre "
1388 "servers: rc = %d\n", devname, rc);
1393 /* Freed in lustre_free_lsi */
1394 OBD_ALLOC(lmd->lmd_dev, strlen(devname) + 1);
1397 strncpy(lmd->lmd_dev, devname, strlen(devname)+1);
1399 /* Save mount options */
1400 s1 = options + strlen(options) - 1;
1401 while (s1 >= options && (*s1 == ',' || *s1 == ' '))
1403 if (*options != 0) {
1404 /* Freed in lustre_free_lsi */
1405 OBD_ALLOC(lmd->lmd_opts, strlen(options) + 1);
1408 strncpy(lmd->lmd_opts, options, strlen(options)+1);
1412 lmd->lmd_magic = LMD_MAGIC;
1417 CERROR("Bad mount options %s\n", options);
1421 struct lustre_mount_data2 {
1423 struct vfsmount *lmd2_mnt;
1426 /** This is the entry point for the mount call into Lustre.
1427 * This is called when a server or client is mounted,
1428 * and this is where we start setting things up.
1429 * @param data Mount options (e.g. -o flock,abort_recov)
1431 static int lustre_fill_super(struct super_block *sb, void *data, int silent)
1433 struct lustre_mount_data *lmd;
1434 struct lustre_mount_data2 *lmd2 = data;
1435 struct lustre_sb_info *lsi;
1439 CDEBUG(D_MOUNT|D_VFSTRACE, "VFS Op: sb %p\n", sb);
1441 lsi = lustre_init_lsi(sb);
1447 * Disable lockdep during mount, because mount locking patterns are
1453 * LU-639: the obd cleanup of last mount may not finish yet, wait here.
1455 obd_zombie_barrier();
1457 /* Figure out the lmd from the mount options */
1458 if (lmd_parse((char *)(lmd2->lmd2_data), lmd)) {
1460 GOTO(out, rc = -EINVAL);
1463 if (lmd_is_client(lmd)) {
1464 CDEBUG(D_MOUNT, "Mounting client %s\n", lmd->lmd_profile);
1465 if (client_fill_super == NULL)
1466 request_module("lustre");
1467 if (client_fill_super == NULL) {
1468 LCONSOLE_ERROR_MSG(0x165, "Nothing registered for "
1469 "client mount! Is the 'lustre' "
1470 "module loaded?\n");
1474 rc = lustre_start_mgc(sb);
1476 lustre_common_put_super(sb);
1479 /* Connect and start */
1480 /* (should always be ll_fill_super) */
1481 rc = (*client_fill_super)(sb, lmd2->lmd2_mnt);
1482 /* c_f_s will call lustre_common_put_super on failure */
1485 #ifdef HAVE_SERVER_SUPPORT
1486 CDEBUG(D_MOUNT, "Mounting server from %s\n", lmd->lmd_dev);
1487 rc = server_fill_super(sb);
1488 /* s_f_s calls lustre_start_mgc after the mount because we need
1489 the MGS nids which are stored on disk. Plus, we may
1490 need to start the MGS first. */
1491 /* s_f_s will call server_put_super on failure */
1493 CERROR("This is client-side-only module, "
1494 "cannot handle server mount.\n");
1499 /* If error happens in fill_super() call, @lsi will be killed there.
1500 * This is why we do not put it here. */
1504 CERROR("Unable to mount %s (%d)\n",
1505 s2lsi(sb) ? lmd->lmd_dev : "", rc);
1507 CDEBUG(D_SUPER, "Mount %s complete\n",
1515 /* We can't call ll_fill_super by name because it lives in a module that
1516 must be loaded after this one. */
1517 void lustre_register_client_fill_super(int (*cfs)(struct super_block *sb,
1518 struct vfsmount *mnt))
1520 client_fill_super = cfs;
1522 EXPORT_SYMBOL(lustre_register_client_fill_super);
1524 void lustre_register_kill_super_cb(void (*cfs)(struct super_block *sb))
1526 kill_super_cb = cfs;
1528 EXPORT_SYMBOL(lustre_register_kill_super_cb);
1530 /***************** FS registration ******************/
1531 #ifdef HAVE_FSTYPE_MOUNT
1532 static struct dentry *lustre_mount(struct file_system_type *fs_type, int flags,
1533 const char *devname, void *data)
1535 struct lustre_mount_data2 lmd2 = {
1539 return mount_nodev(fs_type, flags, &lmd2, lustre_fill_super);
1542 static int lustre_get_sb(struct file_system_type *fs_type, int flags,
1543 const char *devname, void *data, struct vfsmount *mnt)
1545 struct lustre_mount_data2 lmd2 = {
1550 return get_sb_nodev(fs_type, flags, &lmd2, lustre_fill_super, mnt);
1554 static void lustre_kill_super(struct super_block *sb)
1556 struct lustre_sb_info *lsi = s2lsi(sb);
1558 if (kill_super_cb && lsi && !IS_SERVER(lsi))
1559 (*kill_super_cb)(sb);
1561 kill_anon_super(sb);
1564 /** Register the "lustre" fs type
1566 static struct file_system_type lustre_fs_type = {
1567 .owner = THIS_MODULE,
1569 #ifdef HAVE_FSTYPE_MOUNT
1570 .mount = lustre_mount,
1572 .get_sb = lustre_get_sb,
1574 .kill_sb = lustre_kill_super,
1575 .fs_flags = FS_REQUIRES_DEV | FS_HAS_FIEMAP | FS_RENAME_DOES_D_MOVE,
1577 MODULE_ALIAS_FS("lustre");
1579 int lustre_register_fs(void)
1581 return register_filesystem(&lustre_fs_type);
1584 int lustre_unregister_fs(void)
1586 return unregister_filesystem(&lustre_fs_type);