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, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/utils/mkfs_lustre.c
38 * Author: Nathan Rutman <nathan@clusterfs.com>
41 /* This source file is compiled into both mkfs.lustre and tunefs.lustre */
45 #endif /* HAVE_CONFIG_H */
57 #include <sys/types.h>
59 #include <sys/mount.h>
60 #include <sys/utsname.h>
68 /* libcfs.h is not really needed here, but on SLES10/PPC, fs.h includes idr.h
69 * which requires BITS_PER_LONG to be defined */
70 #include <libcfs/libcfs.h>
72 #include <linux/fs.h> /* for BLKGETSIZE64 */
74 #include <linux/version.h>
76 #include <lustre_disk.h>
77 #include <lustre_param.h>
78 #include <lnet/lnetctl.h>
79 #include <lustre_ver.h>
80 #include "mount_utils.h"
88 static int print_only = 0;
89 static int upgrade_to_18 = 0;
91 #define FSLIST_LDISKFS "ldiskfs"
95 #define FSLIST_ZFS "|zfs"
97 #define FSLIST_ZFS "zfs"
101 #define FSLIST_ZFS ""
102 #endif /* HAVE_ZFS_OSD */
105 #error "no backing OSD types (ldiskfs or ZFS) are configured"
108 #define FSLIST FSLIST_LDISKFS FSLIST_ZFS
110 void usage(FILE *out)
112 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
113 fprintf(out, "usage: %s <target type> [--backfstype="FSLIST"] "
114 "--fsname=<filesystem name>\n"
115 "\t--index=<target index> [options] <device>\n", progname);
117 fprintf(out, "usage: %s <target type> --backfstype=zfs "
118 "--fsname=<filesystem name> [options]\n"
119 "\t<pool name>/<dataset name>\n"
120 "\t[[<vdev type>] <device> [<device> ...] [vdev type>] ...]\n",
124 "\t<device>:block device or file (e.g /dev/sda or /tmp/ost1)\n"
126 "\t<pool name>: name of ZFS pool where target is created "
128 "\t<dataset name>: name of new dataset, must be unique within "
130 "\t<vdev type>: type of vdev (mirror, raidz, raidz2, spare, "
135 "\t\t--mgs: configuration management service\n"
136 "\t\t--mdt: metadata storage, mutually exclusive with ost\n"
137 "\t\t--ost: object storage, mutually exclusive with mdt, mgs\n"
138 "\toptions (in order of popularity):\n"
139 "\t\t--index=#N: numerical target index (0..N)\n"
140 "\t\t\trequired for all targets other than the MGS\n"
141 "\t\t--fsname=<8_char_filesystem_name>: fs targets belong to\n"
142 "\t\t\trequired for all targets other than MGS\n"
143 "\t\t--mgsnode=<nid>[,<...>]: NID(s) of remote MGS\n"
144 "\t\t\trequired for all targets other than MGS\n"
145 "\t\t--mountfsoptions=<opts>: permanent mount options\n"
146 "\t\t--failnode=<nid>[,<...>]: NID(s) of backup failover node\n"
147 "\t\t\tmutually exclusive with --servicenode\n"
148 "\t\t--servicenode=<nid>[,<...>]: NID(s) of service partners\n"
149 "\t\t\ttreat nodes as equal service node, mutually exclusive "
151 "\t\t--param <key>=<value>: set a permanent parameter\n"
152 "\t\t\te.g. --param sys.timeout=40\n"
153 "\t\t\t --param lov.stripesize=2M\n"
154 "\t\t--network=<net>[,<...>]: restrict OST/MDT to network(s)\n"
156 "\t\t--backfstype=<fstype>: backing fs type (ext3, ldiskfs)\n"
157 "\t\t--device-size=#N(KB): device size for loop devices\n"
158 "\t\t--mkfsoptions=<opts>: format options\n"
159 "\t\t--reformat: overwrite an existing disk\n"
160 "\t\t--stripe-count-hint=#N: for optimizing MDT inode size\n"
162 "\t\t--erase-params: erase all old parameter settings\n"
163 "\t\t--nomgs: turn off MGS service on this MDT\n"
164 "\t\t--writeconf: erase all config logs for this fs.\n"
166 "\t\t--comment=<user comment>: arbitrary string (%d bytes)\n"
167 "\t\t--dryrun: report what we would do; don't write to disk\n"
168 "\t\t--verbose: e.g. show mkfs progress\n"
170 (int)sizeof(((struct lustre_disk_data *)0)->ldd_userdata));
174 /* ==================== Lustre config functions =============*/
176 void print_ldd(char *str, struct lustre_disk_data *ldd)
178 printf("\n %s:\n", str);
179 printf("Target: %s\n", ldd->ldd_svname);
180 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
181 printf("Index: unassigned\n");
183 printf("Index: %d\n", ldd->ldd_svindex);
184 if (ldd->ldd_uuid[0])
185 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
186 printf("Lustre FS: %s\n", ldd->ldd_fsname);
187 printf("Mount type: %s\n", MT_STR(ldd));
188 printf("Flags: %#x\n", ldd->ldd_flags);
189 printf(" (%s%s%s%s%s%s%s%s%s%s)\n",
190 IS_MDT(ldd) ? "MDT ":"",
191 IS_OST(ldd) ? "OST ":"",
192 IS_MGS(ldd) ? "MGS ":"",
193 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
194 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
195 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
196 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
197 ldd->ldd_flags & LDD_F_IAM_DIR ? "IAM_dir_format ":"",
198 ldd->ldd_flags & LDD_F_NO_PRIMNODE? "no_primnode ":"",
199 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
200 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
201 printf("Parameters:%s\n", ldd->ldd_params);
202 if (ldd->ldd_userdata[0])
203 printf("Comment: %s\n", ldd->ldd_userdata);
207 void set_defaults(struct mkfs_opts *mop)
209 mop->mo_ldd.ldd_magic = LDD_MAGIC;
210 mop->mo_ldd.ldd_config_ver = 1;
211 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
212 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
213 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
214 mop->mo_mgs_failnodes = 0;
215 mop->mo_stripe_count = 1;
216 mop->mo_pool_vdevs = NULL;
219 static inline void badopt(const char *opt, char *type)
221 fprintf(stderr, "%s: '--%s' only valid for %s\n",
222 progname, opt, type);
226 /* from mount_lustre */
227 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
228 #define MAXNIDSTR 1024
229 static char *convert_hostnames(char *s1)
231 char *converted, *s2 = 0, *c, *end, sep;
232 int left = MAXNIDSTR;
235 converted = malloc(left);
236 if (converted == NULL) {
240 end = s1 + strlen(s1);
242 while ((left > 0) && (s1 < end)) {
243 s2 = strpbrk(s1, ",:");
248 nid = libcfs_str2nid(s1);
251 if (nid == LNET_NID_ANY) {
252 fprintf(stderr, "%s: Can't parse NID '%s'\n",
257 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
258 strlen("127.0.0.1")) == 0) {
259 fprintf(stderr, "%s: The NID '%s' resolves to the "
260 "loopback address '%s'. Lustre requires a "
261 "non-loopback address.\n",
262 progname, s1, libcfs_nid2str(nid));
267 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
268 left = converted + MAXNIDSTR - c;
274 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
277 static struct option long_opt[] = {
278 {"iam-dir", 0, 0, 'a'},
279 {"backfstype", 1, 0, 'b'},
280 {"stripe-count-hint", 1, 0, 'c'},
281 {"comment", 1, 0, 'u'},
282 {"configdev", 1, 0, 'C'},
283 {"device-size", 1, 0, 'd'},
284 {"dryrun", 0, 0, 'n'},
285 {"erase-params", 0, 0, 'e'},
286 {"failnode", 1, 0, 'f'},
287 {"failover", 1, 0, 'f'},
290 {"index", 1, 0, 'i'},
291 {"mkfsoptions", 1, 0, 'k'},
292 {"mgsnode", 1, 0, 'm'},
293 {"mgsnid", 1, 0, 'm'},
295 {"fsname",1, 0, 'L'},
296 {"noformat", 0, 0, 'n'},
297 {"nomgs", 0, 0, 'N'},
298 {"mountfsoptions", 1, 0, 'o'},
300 {"param", 1, 0, 'p'},
301 {"print", 0, 0, 'n'},
302 {"quiet", 0, 0, 'q'},
303 {"reformat", 0, 0, 'r'},
304 {"servicenode", 1, 0, 's'},
305 {"verbose", 0, 0, 'v'},
306 {"writeconf", 0, 0, 'w'},
307 {"upgrade_to_18", 0, 0, 'U'},
308 {"network", 1, 0, 't'},
311 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqrs:t:Uu:vw";
314 int failnode_set = 0, servicenode_set = 0;
316 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
320 if (IS_MDT(&mop->mo_ldd))
321 mop->mo_ldd.ldd_flags |= LDD_F_IAM_DIR;
326 while (i < LDD_MT_LAST) {
327 if (strcmp(optarg, mt_str(i)) == 0) {
328 mop->mo_ldd.ldd_mount_type = i;
333 if (i == LDD_MT_LAST) {
334 fprintf(stderr, "%s: invalid backend filesystem"
335 " type %s\n", progname, optarg);
341 if (IS_MDT(&mop->mo_ldd)) {
342 int stripe_count = atol(optarg);
343 if (stripe_count <= 0) {
344 fprintf(stderr, "%s: bad stripe count "
345 "%d\n", progname, stripe_count);
348 mop->mo_stripe_count = stripe_count;
350 badopt(long_opt[longidx].name, "MDT");
354 case 'C': /* Configdev */
356 printf("Configdev not implemented\n");
359 mop->mo_device_sz = atol(optarg);
362 mop->mo_ldd.ldd_params[0] = '\0';
363 /* Must update the mgs logs */
364 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
370 if ((opt == 'f' && servicenode_set)
371 || (opt == 's' && failnode_set)) {
372 fprintf(stderr, "%s: %s cannot use with --%s\n",
373 progname, long_opt[longidx].name,
374 opt == 'f' ? "servicenode" : "failnode");
378 nids = convert_hostnames(optarg);
381 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
386 /* Must update the mgs logs */
387 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
391 mop->mo_ldd.ldd_flags |= LDD_F_NO_PRIMNODE;
394 mop->mo_flags |= MO_FAILOVER;
398 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
404 if (!(mop->mo_ldd.ldd_flags &
405 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
407 fprintf(stderr, "%s: cannot change the index of"
408 " a registered target\n", progname);
411 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
412 mop->mo_ldd.ldd_svindex = atol(optarg);
413 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
415 badopt(long_opt[longidx].name, "MDT,OST");
420 strscpy(mop->mo_mkfsopts, optarg,
421 sizeof(mop->mo_mkfsopts));
425 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
426 (!(mop->mo_ldd.ldd_flags &
427 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
428 LDD_F_WRITECONF)))) {
429 fprintf(stderr, "%s: cannot change the name of"
430 " a registered target\n", progname);
433 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
434 fprintf(stderr, "%s: filesystem name must be "
435 "1-8 chars\n", progname);
438 if ((tmp = strpbrk(optarg, "/:"))) {
439 fprintf(stderr, "%s: char '%c' not allowed in "
440 "filesystem name\n", progname, *tmp);
443 strscpy(mop->mo_ldd.ldd_fsname, optarg,
444 sizeof(mop->mo_ldd.ldd_fsname));
448 char *nids = convert_hostnames(optarg);
451 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
456 mop->mo_mgs_failnodes++;
460 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
466 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
472 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
475 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg);
478 /* Must update the mgs logs */
479 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
485 mop->mo_flags |= MO_FORCEFORMAT;
488 if (!IS_MDT(&mop->mo_ldd) && !IS_OST(&mop->mo_ldd)) {
489 badopt(long_opt[longidx].name, "MDT,OST");
496 rc = add_param(mop->mo_ldd.ldd_params,
497 PARAM_NETWORK, optarg);
500 /* Must update the mgs logs */
501 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
504 strscpy(mop->mo_ldd.ldd_userdata, optarg,
505 sizeof(mop->mo_ldd.ldd_userdata));
511 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
519 fprintf(stderr, "Unknown option '%c'\n", opt);
525 if (optind == argc) {
526 /* The user didn't specify device name */
528 fprintf(stderr, "Not enough arguments - device name or "
529 "pool/dataset name not specified.\n");
532 /* The device or pool/filesystem name */
533 strscpy(mop->mo_device, argv[optind], sizeof(mop->mo_device));
535 /* Followed by optional vdevs */
536 if (optind < argc - 1)
537 mop->mo_pool_vdevs = (char **) &argv[optind + 1];
543 int main(int argc, char *const argv[])
545 struct mkfs_opts mop;
546 struct lustre_disk_data *ldd;
547 char *mountopts = NULL;
548 char always_mountopts[512] = "";
549 char default_mountopts[512] = "";
553 if ((progname = strrchr(argv[0], '/')) != NULL)
558 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
563 memset(&mop, 0, sizeof(mop));
566 /* device is last arg */
567 strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
570 /* For tunefs, we must read in the old values before parsing any
573 /* Check whether the disk has already been formatted by mkfs.lustre */
574 ret = osd_is_lustre(mop.mo_device, &mount_type);
577 fprintf(stderr, "Device %s has not been formatted with "
578 "mkfs.lustre\n", mop.mo_device);
583 ret = osd_read_ldd(mop.mo_device, &mop.mo_ldd);
586 fprintf(stderr, "Failed to read previous Lustre data from %s "
587 "(%d)\n", mop.mo_device, ret);
590 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
591 mop.mo_mgs_failnodes++;
594 print_ldd("Read previous values", &(mop.mo_ldd));
601 ret = parse_opts(argc, argv, &mop, &mountopts);
607 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
609 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
614 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
616 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
621 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
622 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
624 fprintf(stderr, "Can't find the target index, "
625 "specify with --index\n");
631 * Comment out these 2 checks temporarily, since for multi-MDSes
632 * in single node only 1 mds node could have mgs service
634 if (IS_MDT(ldd) && !IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
635 verrprint("No management node specified, adding MGS to this "
637 ldd->ldd_flags |= LDD_F_SV_TYPE_MGS;
639 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
642 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
644 fprintf(stderr, "Must specify --mgsnode=\n");
649 if ((IS_MDT(ldd) || IS_OST(ldd)) && mop.mo_ldd.ldd_fsname[0] == '\0') {
651 fprintf(stderr, "Must specify --fsname for MDT/OST device\n");
656 /* These are the permanent mount options (always included) */
657 ret = osd_prepare_lustre(&mop,
658 default_mountopts, sizeof(default_mountopts),
659 always_mountopts, sizeof(always_mountopts));
662 fprintf(stderr, "unable to prepare backend (%d)\n", ret);
667 trim_mountfsoptions(mountopts);
668 (void)check_mountfsoptions(mountopts, default_mountopts, 1);
669 if (check_mountfsoptions(mountopts, always_mountopts, 0)) {
673 sprintf(ldd->ldd_mount_opts, "%s", mountopts);
676 if (ldd->ldd_mount_opts[0] == 0)
677 /* use the defaults unless old opts exist */
680 sprintf(ldd->ldd_mount_opts, "%s%s",
681 always_mountopts, default_mountopts);
682 trim_mountfsoptions(ldd->ldd_mount_opts);
686 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
687 ldd->ldd_fsname, ldd->ldd_svname);
690 print_ldd("Permanent disk data", ldd);
693 printf("exiting before disk write.\n");
697 if (check_mtab_entry(mop.mo_device, mop.mo_device, NULL, NULL))
700 /* Create the loopback file */
701 if (mop.mo_flags & MO_IS_LOOP) {
702 ret = access(mop.mo_device, F_OK);
705 #ifndef TUNEFS /* mkfs.lustre */
706 /* Reformat the loopback file */
707 if (ret || (mop.mo_flags & MO_FORCEFORMAT)) {
708 ret = loop_format(&mop);
714 ret = loop_setup(&mop);
717 fprintf(stderr, "Loop device setup for %s failed: %s\n",
718 mop.mo_device, strerror(ret));
723 #ifndef TUNEFS /* mkfs.lustre */
724 /* Check whether the disk has already been formatted by mkfs.lustre */
725 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
726 ret = osd_is_lustre(mop.mo_device, &mount_type);
729 fprintf(stderr, "Device %s was previously formatted "
730 "for lustre. Use --reformat to reformat it, "
731 "or tunefs.lustre to modify.\n",
737 /* Format the backing filesystem */
738 ret = osd_make_lustre(&mop);
741 fprintf(stderr, "mkfs failed %d\n", ret);
746 /* Write our config files */
747 ret = osd_write_ldd(&mop);
750 fprintf(stderr, "failed to write local files\n");
758 /* Fix any crazy return values from system() */
759 if (ret && ((ret & 255) == 0))
762 verrprint("%s: exiting with %d (%s)\n",
763 progname, ret, strerror(ret));