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;
93 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
94 fprintf(out, "usage: %s <target types> [options] <device>\n", progname);
96 "\t<device>:block device or file (e.g /dev/sda or /tmp/ost1)\n"
98 "\t\t--ost: object storage, mutually exclusive with mdt,mgs\n"
99 "\t\t--mdt: metadata storage, mutually exclusive with ost\n"
100 "\t\t--mgs: configuration management service - one per site\n"
101 "\toptions (in order of popularity):\n"
102 "\t\t--mgsnode=<nid>[,<...>] : NID(s) of a remote mgs node\n"
103 "\t\t\trequired for all targets other than the mgs node\n"
104 "\t\t--fsname=<filesystem_name> : default is 'lustre'\n"
105 "\t\t--failnode=<nid>[,<...>] : NID(s) of a failover partner\n"
106 "\t\t\tcannot be used with --servicenode\n"
107 "\t\t--servicenode=<nid>[,<...>] : NID(s) of all service partners\n"
108 "\t\t\ttreat all nodes as equal service node, cannot be used with --failnode\n"
109 "\t\t--param <key>=<value> : set a permanent parameter\n"
110 "\t\t\te.g. --param sys.timeout=40\n"
111 "\t\t\t --param lov.stripesize=2M\n"
112 "\t\t--index=#N : target index (i.e. ost index within lov)\n"
113 "\t\t--comment=<user comment>: arbitrary string (%d bytes)\n"
114 "\t\t--mountfsoptions=<opts> : permanent mount options\n"
115 "\t\t--network=<net>[,<...>] : restrict OST/MDT to network(s)\n"
117 "\t\t--backfstype=<fstype> : backing fs type (ext3, ldiskfs)\n"
118 "\t\t--device-size=#N(KB) : device size for loop devices\n"
119 "\t\t--mkfsoptions=<opts> : format options\n"
120 "\t\t--reformat: overwrite an existing disk\n"
121 "\t\t--stripe-count-hint=#N : for optimizing MDT inode size\n"
122 "\t\t--iam-dir: use IAM directory format, not ext3 compatible\n"
124 "\t\t--erase-params : erase all old parameter settings\n"
125 "\t\t--nomgs: turn off MGS service on this MDT\n"
126 "\t\t--writeconf: erase all config logs for this fs.\n"
128 "\t\t--dryrun: just report what we would do; "
129 "don't write to disk\n"
130 "\t\t--verbose : e.g. show mkfs progress\n"
132 (int)sizeof(((struct lustre_disk_data *)0)->ldd_userdata));
136 /* ==================== Lustre config functions =============*/
138 void print_ldd(char *str, struct lustre_disk_data *ldd)
140 printf("\n %s:\n", str);
141 printf("Target: %s\n", ldd->ldd_svname);
142 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
143 printf("Index: unassigned\n");
145 printf("Index: %d\n", ldd->ldd_svindex);
146 if (ldd->ldd_uuid[0])
147 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
148 printf("Lustre FS: %s\n", ldd->ldd_fsname);
149 printf("Mount type: %s\n", MT_STR(ldd));
150 printf("Flags: %#x\n", ldd->ldd_flags);
151 printf(" (%s%s%s%s%s%s%s%s%s%s)\n",
152 IS_MDT(ldd) ? "MDT ":"",
153 IS_OST(ldd) ? "OST ":"",
154 IS_MGS(ldd) ? "MGS ":"",
155 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
156 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
157 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
158 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
159 ldd->ldd_flags & LDD_F_IAM_DIR ? "IAM_dir_format ":"",
160 ldd->ldd_flags & LDD_F_NO_PRIMNODE? "no_primnode ":"",
161 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
162 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
163 printf("Parameters:%s\n", ldd->ldd_params);
164 if (ldd->ldd_userdata[0])
165 printf("Comment: %s\n", ldd->ldd_userdata);
169 void set_defaults(struct mkfs_opts *mop)
171 mop->mo_ldd.ldd_magic = LDD_MAGIC;
172 mop->mo_ldd.ldd_config_ver = 1;
173 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
174 mop->mo_mgs_failnodes = 0;
175 strcpy(mop->mo_ldd.ldd_fsname, "lustre");
176 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
178 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
179 mop->mo_stripe_count = 1;
182 static inline void badopt(const char *opt, char *type)
184 fprintf(stderr, "%s: '--%s' only valid for %s\n",
185 progname, opt, type);
189 /* from mount_lustre */
190 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
191 #define MAXNIDSTR 1024
192 static char *convert_hostnames(char *s1)
194 char *converted, *s2 = 0, *c, *end, sep;
195 int left = MAXNIDSTR;
198 converted = malloc(left);
199 if (converted == NULL) {
203 end = s1 + strlen(s1);
205 while ((left > 0) && (s1 < end)) {
206 s2 = strpbrk(s1, ",:");
211 nid = libcfs_str2nid(s1);
213 if (nid == LNET_NID_ANY) {
214 fprintf(stderr, "%s: Can't parse NID '%s'\n",
219 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
220 strlen("127.0.0.1")) == 0) {
221 fprintf(stderr, "%s: The NID '%s' resolves to the "
222 "loopback address '%s'. Lustre requires a "
223 "non-loopback address.\n",
224 progname, s1, libcfs_nid2str(nid));
229 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
230 left = converted + MAXNIDSTR - c;
236 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
239 static struct option long_opt[] = {
240 {"iam-dir", 0, 0, 'a'},
241 {"backfstype", 1, 0, 'b'},
242 {"stripe-count-hint", 1, 0, 'c'},
243 {"comment", 1, 0, 'u'},
244 {"configdev", 1, 0, 'C'},
245 {"device-size", 1, 0, 'd'},
246 {"dryrun", 0, 0, 'n'},
247 {"erase-params", 0, 0, 'e'},
248 {"failnode", 1, 0, 'f'},
249 {"failover", 1, 0, 'f'},
252 {"index", 1, 0, 'i'},
253 {"mkfsoptions", 1, 0, 'k'},
254 {"mgsnode", 1, 0, 'm'},
255 {"mgsnid", 1, 0, 'm'},
257 {"fsname",1, 0, 'L'},
258 {"noformat", 0, 0, 'n'},
259 {"nomgs", 0, 0, 'N'},
260 {"mountfsoptions", 1, 0, 'o'},
262 {"param", 1, 0, 'p'},
263 {"print", 0, 0, 'n'},
264 {"quiet", 0, 0, 'q'},
265 {"reformat", 0, 0, 'r'},
266 {"servicenode", 1, 0, 's'},
267 {"verbose", 0, 0, 'v'},
268 {"writeconf", 0, 0, 'w'},
269 {"upgrade_to_18", 0, 0, 'U'},
270 {"network", 1, 0, 't'},
273 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqrs:t:Uu:vw";
276 int failnode_set = 0, servicenode_set = 0;
278 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
282 if (IS_MDT(&mop->mo_ldd))
283 mop->mo_ldd.ldd_flags |= LDD_F_IAM_DIR;
288 while (i < LDD_MT_LAST) {
289 if (strcmp(optarg, mt_str(i)) == 0) {
290 mop->mo_ldd.ldd_mount_type = i;
298 if (IS_MDT(&mop->mo_ldd)) {
299 int stripe_count = atol(optarg);
300 if (stripe_count <= 0) {
301 fprintf(stderr, "%s: bad stripe count "
302 "%d\n", progname, stripe_count);
305 mop->mo_stripe_count = stripe_count;
307 badopt(long_opt[longidx].name, "MDT");
311 case 'C': /* Configdev */
313 printf("Configdev not implemented\n");
316 mop->mo_device_sz = atol(optarg);
319 mop->mo_ldd.ldd_params[0] = '\0';
320 /* Must update the mgs logs */
321 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
327 if ((opt == 'f' && servicenode_set)
328 || (opt == 's' && failnode_set)) {
329 fprintf(stderr, "%s: %s cannot use with --%s\n",
330 progname, long_opt[longidx].name,
331 opt == 'f' ? "servicenode" : "failnode");
335 nids = convert_hostnames(optarg);
338 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
343 /* Must update the mgs logs */
344 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
348 mop->mo_ldd.ldd_flags |= LDD_F_NO_PRIMNODE;
351 mop->mo_flags |= MO_FAILOVER;
355 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
361 if (!(mop->mo_ldd.ldd_flags &
362 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
364 fprintf(stderr, "%s: cannot change the index of"
365 " a registered target\n", progname);
368 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
369 mop->mo_ldd.ldd_svindex = atol(optarg);
370 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
372 badopt(long_opt[longidx].name, "MDT,OST");
377 strscpy(mop->mo_mkfsopts, optarg,
378 sizeof(mop->mo_mkfsopts));
382 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
383 (!(mop->mo_ldd.ldd_flags &
384 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
385 LDD_F_WRITECONF)))) {
386 fprintf(stderr, "%s: cannot change the name of"
387 " a registered target\n", progname);
390 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
391 fprintf(stderr, "%s: filesystem name must be "
392 "1-8 chars\n", progname);
395 if ((tmp = strpbrk(optarg, "/:"))) {
396 fprintf(stderr, "%s: char '%c' not allowed in "
397 "filesystem name\n", progname, *tmp);
400 strscpy(mop->mo_ldd.ldd_fsname, optarg,
401 sizeof(mop->mo_ldd.ldd_fsname));
405 char *nids = convert_hostnames(optarg);
408 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
413 mop->mo_mgs_failnodes++;
417 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
423 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
429 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
432 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg);
435 /* Must update the mgs logs */
436 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
442 mop->mo_flags |= MO_FORCEFORMAT;
445 if (!IS_MDT(&mop->mo_ldd) && !IS_OST(&mop->mo_ldd)) {
446 badopt(long_opt[longidx].name, "MDT,OST");
453 rc = add_param(mop->mo_ldd.ldd_params,
454 PARAM_NETWORK, optarg);
457 /* Must update the mgs logs */
458 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
461 strscpy(mop->mo_ldd.ldd_userdata, optarg,
462 sizeof(mop->mo_ldd.ldd_userdata));
468 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
476 fprintf(stderr, "Unknown option '%c'\n", opt);
482 /* Last arg is device */
483 if (optind != argc - 1) {
485 fprintf(stderr, "Bad argument: %s\n", argv[optind]);
489 /* single argument: <device> */
496 int main(int argc, char *const argv[])
498 struct mkfs_opts mop;
499 struct lustre_disk_data *ldd;
500 char *mountopts = NULL;
501 char always_mountopts[512] = "";
502 char default_mountopts[512] = "";
506 if ((progname = strrchr(argv[0], '/')) != NULL)
511 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
516 memset(&mop, 0, sizeof(mop));
519 /* device is last arg */
520 strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
523 /* For tunefs, we must read in the old values before parsing any
526 /* Check whether the disk has already been formatted by mkfs.lustre */
527 ret = osd_is_lustre(mop.mo_device, &mount_type);
530 fprintf(stderr, "Device %s has not been formatted with "
531 "mkfs.lustre\n", mop.mo_device);
536 ret = osd_read_ldd(mop.mo_device, &mop.mo_ldd);
539 fprintf(stderr, "Failed to read previous Lustre data from %s "
540 "(%d)\n", mop.mo_device, ret);
543 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
544 mop.mo_mgs_failnodes++;
547 print_ldd("Read previous values", &(mop.mo_ldd));
550 ret = parse_opts(argc, argv, &mop, &mountopts);
556 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
558 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
563 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
565 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
570 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
571 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
573 fprintf(stderr, "Can't find the target index, "
574 "specify with --index\n");
580 * Comment out these 2 checks temporarily, since for multi-MDSes
581 * in single node only 1 mds node could have mgs service
583 if (IS_MDT(ldd) && !IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
584 verrprint("No management node specified, adding MGS to this "
586 ldd->ldd_flags |= LDD_F_SV_TYPE_MGS;
588 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
591 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
593 fprintf(stderr, "Must specify --mgsnode=\n");
599 /* These are the permanent mount options (always included) */
600 ret = osd_prepare_lustre(&mop,
601 default_mountopts, sizeof(default_mountopts),
602 always_mountopts, sizeof(always_mountopts));
605 fprintf(stderr, "unable to prepare backend (%d)\n", ret);
610 trim_mountfsoptions(mountopts);
611 (void)check_mountfsoptions(mountopts, default_mountopts, 1);
612 if (check_mountfsoptions(mountopts, always_mountopts, 0)) {
616 sprintf(ldd->ldd_mount_opts, "%s", mountopts);
619 if (ldd->ldd_mount_opts[0] == 0)
620 /* use the defaults unless old opts exist */
623 sprintf(ldd->ldd_mount_opts, "%s%s",
624 always_mountopts, default_mountopts);
625 trim_mountfsoptions(ldd->ldd_mount_opts);
629 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
630 ldd->ldd_fsname, ldd->ldd_svname);
633 print_ldd("Permanent disk data", ldd);
636 printf("exiting before disk write.\n");
640 if (check_mtab_entry(mop.mo_device, mop.mo_device, NULL, NULL))
643 /* Create the loopback file */
644 if (mop.mo_flags & MO_IS_LOOP) {
645 ret = access(mop.mo_device, F_OK);
648 #ifndef TUNEFS /* mkfs.lustre */
649 /* Reformat the loopback file */
650 if (ret || (mop.mo_flags & MO_FORCEFORMAT)) {
651 ret = loop_format(&mop);
657 ret = loop_setup(&mop);
660 fprintf(stderr, "Loop device setup for %s failed: %s\n",
661 mop.mo_device, strerror(ret));
666 #ifndef TUNEFS /* mkfs.lustre */
667 /* Check whether the disk has already been formatted by mkfs.lustre */
668 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
669 ret = osd_is_lustre(mop.mo_device, &mount_type);
672 fprintf(stderr, "Device %s was previously formatted "
673 "for lustre. Use --reformat to reformat it, "
674 "or tunefs.lustre to modify.\n",
680 /* Format the backing filesystem */
681 ret = make_lustre_backfs(&mop);
684 fprintf(stderr, "mkfs failed %d\n", ret);
689 /* Write our config files */
690 ret = write_local_files(&mop);
693 fprintf(stderr, "failed to write local files\n");
700 /* Fix any crazy return values from system() */
701 if (ret && ((ret & 255) == 0))
704 verrprint("%s: exiting with %d (%s)\n",
705 progname, ret, strerror(ret));