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;
180 mop->mo_pool_vdevs = NULL;
183 static inline void badopt(const char *opt, char *type)
185 fprintf(stderr, "%s: '--%s' only valid for %s\n",
186 progname, opt, type);
190 /* from mount_lustre */
191 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
192 #define MAXNIDSTR 1024
193 static char *convert_hostnames(char *s1)
195 char *converted, *s2 = 0, *c, *end, sep;
196 int left = MAXNIDSTR;
199 converted = malloc(left);
200 if (converted == NULL) {
204 end = s1 + strlen(s1);
206 while ((left > 0) && (s1 < end)) {
207 s2 = strpbrk(s1, ",:");
212 nid = libcfs_str2nid(s1);
214 if (nid == LNET_NID_ANY) {
215 fprintf(stderr, "%s: Can't parse NID '%s'\n",
220 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
221 strlen("127.0.0.1")) == 0) {
222 fprintf(stderr, "%s: The NID '%s' resolves to the "
223 "loopback address '%s'. Lustre requires a "
224 "non-loopback address.\n",
225 progname, s1, libcfs_nid2str(nid));
230 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
231 left = converted + MAXNIDSTR - c;
237 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
240 static struct option long_opt[] = {
241 {"iam-dir", 0, 0, 'a'},
242 {"backfstype", 1, 0, 'b'},
243 {"stripe-count-hint", 1, 0, 'c'},
244 {"comment", 1, 0, 'u'},
245 {"configdev", 1, 0, 'C'},
246 {"device-size", 1, 0, 'd'},
247 {"dryrun", 0, 0, 'n'},
248 {"erase-params", 0, 0, 'e'},
249 {"failnode", 1, 0, 'f'},
250 {"failover", 1, 0, 'f'},
253 {"index", 1, 0, 'i'},
254 {"mkfsoptions", 1, 0, 'k'},
255 {"mgsnode", 1, 0, 'm'},
256 {"mgsnid", 1, 0, 'm'},
258 {"fsname",1, 0, 'L'},
259 {"noformat", 0, 0, 'n'},
260 {"nomgs", 0, 0, 'N'},
261 {"mountfsoptions", 1, 0, 'o'},
263 {"param", 1, 0, 'p'},
264 {"print", 0, 0, 'n'},
265 {"quiet", 0, 0, 'q'},
266 {"reformat", 0, 0, 'r'},
267 {"servicenode", 1, 0, 's'},
268 {"verbose", 0, 0, 'v'},
269 {"writeconf", 0, 0, 'w'},
270 {"upgrade_to_18", 0, 0, 'U'},
271 {"network", 1, 0, 't'},
274 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqrs:t:Uu:vw";
277 int failnode_set = 0, servicenode_set = 0;
279 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
283 if (IS_MDT(&mop->mo_ldd))
284 mop->mo_ldd.ldd_flags |= LDD_F_IAM_DIR;
289 while (i < LDD_MT_LAST) {
290 if (strcmp(optarg, mt_str(i)) == 0) {
291 mop->mo_ldd.ldd_mount_type = i;
299 if (IS_MDT(&mop->mo_ldd)) {
300 int stripe_count = atol(optarg);
301 if (stripe_count <= 0) {
302 fprintf(stderr, "%s: bad stripe count "
303 "%d\n", progname, stripe_count);
306 mop->mo_stripe_count = stripe_count;
308 badopt(long_opt[longidx].name, "MDT");
312 case 'C': /* Configdev */
314 printf("Configdev not implemented\n");
317 mop->mo_device_sz = atol(optarg);
320 mop->mo_ldd.ldd_params[0] = '\0';
321 /* Must update the mgs logs */
322 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
328 if ((opt == 'f' && servicenode_set)
329 || (opt == 's' && failnode_set)) {
330 fprintf(stderr, "%s: %s cannot use with --%s\n",
331 progname, long_opt[longidx].name,
332 opt == 'f' ? "servicenode" : "failnode");
336 nids = convert_hostnames(optarg);
339 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
344 /* Must update the mgs logs */
345 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
349 mop->mo_ldd.ldd_flags |= LDD_F_NO_PRIMNODE;
352 mop->mo_flags |= MO_FAILOVER;
356 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
362 if (!(mop->mo_ldd.ldd_flags &
363 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
365 fprintf(stderr, "%s: cannot change the index of"
366 " a registered target\n", progname);
369 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
370 mop->mo_ldd.ldd_svindex = atol(optarg);
371 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
373 badopt(long_opt[longidx].name, "MDT,OST");
378 strscpy(mop->mo_mkfsopts, optarg,
379 sizeof(mop->mo_mkfsopts));
383 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
384 (!(mop->mo_ldd.ldd_flags &
385 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
386 LDD_F_WRITECONF)))) {
387 fprintf(stderr, "%s: cannot change the name of"
388 " a registered target\n", progname);
391 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
392 fprintf(stderr, "%s: filesystem name must be "
393 "1-8 chars\n", progname);
396 if ((tmp = strpbrk(optarg, "/:"))) {
397 fprintf(stderr, "%s: char '%c' not allowed in "
398 "filesystem name\n", progname, *tmp);
401 strscpy(mop->mo_ldd.ldd_fsname, optarg,
402 sizeof(mop->mo_ldd.ldd_fsname));
406 char *nids = convert_hostnames(optarg);
409 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
414 mop->mo_mgs_failnodes++;
418 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
424 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
430 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
433 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg);
436 /* Must update the mgs logs */
437 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
443 mop->mo_flags |= MO_FORCEFORMAT;
446 if (!IS_MDT(&mop->mo_ldd) && !IS_OST(&mop->mo_ldd)) {
447 badopt(long_opt[longidx].name, "MDT,OST");
454 rc = add_param(mop->mo_ldd.ldd_params,
455 PARAM_NETWORK, optarg);
458 /* Must update the mgs logs */
459 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
462 strscpy(mop->mo_ldd.ldd_userdata, optarg,
463 sizeof(mop->mo_ldd.ldd_userdata));
469 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
477 fprintf(stderr, "Unknown option '%c'\n", opt);
483 if (optind == argc) {
484 /* The user didn't specify device name */
486 fprintf(stderr, "Not enough arguments - device name or "
487 "pool/dataset name not specified.\n");
490 /* The device or pool/filesystem name */
491 strscpy(mop->mo_device, argv[optind], sizeof(mop->mo_device));
493 /* Followed by optional vdevs */
494 if (optind < argc - 1)
495 mop->mo_pool_vdevs = (char **) &argv[optind + 1];
501 int main(int argc, char *const argv[])
503 struct mkfs_opts mop;
504 struct lustre_disk_data *ldd;
505 char *mountopts = NULL;
506 char always_mountopts[512] = "";
507 char default_mountopts[512] = "";
511 if ((progname = strrchr(argv[0], '/')) != NULL)
516 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
521 memset(&mop, 0, sizeof(mop));
524 /* device is last arg */
525 strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
528 /* For tunefs, we must read in the old values before parsing any
531 /* Check whether the disk has already been formatted by mkfs.lustre */
532 ret = osd_is_lustre(mop.mo_device, &mount_type);
535 fprintf(stderr, "Device %s has not been formatted with "
536 "mkfs.lustre\n", mop.mo_device);
541 ret = osd_read_ldd(mop.mo_device, &mop.mo_ldd);
544 fprintf(stderr, "Failed to read previous Lustre data from %s "
545 "(%d)\n", mop.mo_device, ret);
548 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
549 mop.mo_mgs_failnodes++;
552 print_ldd("Read previous values", &(mop.mo_ldd));
555 ret = parse_opts(argc, argv, &mop, &mountopts);
561 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
563 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
568 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
570 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
575 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
576 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
578 fprintf(stderr, "Can't find the target index, "
579 "specify with --index\n");
585 * Comment out these 2 checks temporarily, since for multi-MDSes
586 * in single node only 1 mds node could have mgs service
588 if (IS_MDT(ldd) && !IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
589 verrprint("No management node specified, adding MGS to this "
591 ldd->ldd_flags |= LDD_F_SV_TYPE_MGS;
593 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
596 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
598 fprintf(stderr, "Must specify --mgsnode=\n");
604 /* These are the permanent mount options (always included) */
605 ret = osd_prepare_lustre(&mop,
606 default_mountopts, sizeof(default_mountopts),
607 always_mountopts, sizeof(always_mountopts));
610 fprintf(stderr, "unable to prepare backend (%d)\n", ret);
615 trim_mountfsoptions(mountopts);
616 (void)check_mountfsoptions(mountopts, default_mountopts, 1);
617 if (check_mountfsoptions(mountopts, always_mountopts, 0)) {
621 sprintf(ldd->ldd_mount_opts, "%s", mountopts);
624 if (ldd->ldd_mount_opts[0] == 0)
625 /* use the defaults unless old opts exist */
628 sprintf(ldd->ldd_mount_opts, "%s%s",
629 always_mountopts, default_mountopts);
630 trim_mountfsoptions(ldd->ldd_mount_opts);
634 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
635 ldd->ldd_fsname, ldd->ldd_svname);
638 print_ldd("Permanent disk data", ldd);
641 printf("exiting before disk write.\n");
645 if (check_mtab_entry(mop.mo_device, mop.mo_device, NULL, NULL))
648 /* Create the loopback file */
649 if (mop.mo_flags & MO_IS_LOOP) {
650 ret = access(mop.mo_device, F_OK);
653 #ifndef TUNEFS /* mkfs.lustre */
654 /* Reformat the loopback file */
655 if (ret || (mop.mo_flags & MO_FORCEFORMAT)) {
656 ret = loop_format(&mop);
662 ret = loop_setup(&mop);
665 fprintf(stderr, "Loop device setup for %s failed: %s\n",
666 mop.mo_device, strerror(ret));
671 #ifndef TUNEFS /* mkfs.lustre */
672 /* Check whether the disk has already been formatted by mkfs.lustre */
673 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
674 ret = osd_is_lustre(mop.mo_device, &mount_type);
677 fprintf(stderr, "Device %s was previously formatted "
678 "for lustre. Use --reformat to reformat it, "
679 "or tunefs.lustre to modify.\n",
685 /* Format the backing filesystem */
686 ret = osd_make_lustre(&mop);
689 fprintf(stderr, "mkfs failed %d\n", ret);
694 /* Write our config files */
695 ret = osd_write_ldd(&mop);
698 fprintf(stderr, "failed to write local files\n");
705 /* Fix any crazy return values from system() */
706 if (ret && ((ret & 255) == 0))
709 verrprint("%s: exiting with %d (%s)\n",
710 progname, ret, strerror(ret));