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);
114 fprintf(out, "usage: %s <target types> [--backfstype=zfs] [options] "
115 "<pool name>/<dataset name> [[<vdev type>] <device> "
116 "[<device> ...] [[vdev type>] ...]]\n", progname);
119 fprintf(out, "usage: %s <target types> --backfstype="FSLIST" "
120 "[options] <device>\n", progname);
122 "\t<device>:block device or file (e.g /dev/sda or /tmp/ost1)\n"
124 "\t<pool name>: name of the ZFS pool where to create the "
125 "target (e.g. tank)\n"
126 "\t<dataset name>: name of the new dataset (e.g. ost1). The "
127 "dataset name must be unique within the ZFS pool\n"
128 "\t<vdev type>: type of vdev (mirror, raidz, raidz2, spare, "
133 "\t\t--ost: object storage, mutually exclusive with mdt,mgs\n"
134 "\t\t--mdt: metadata storage, mutually exclusive with ost\n"
135 "\t\t--mgs: configuration management service - one per site\n"
136 "\toptions (in order of popularity):\n"
137 "\t\t--mgsnode=<nid>[,<...>] : NID(s) of a remote mgs node\n"
138 "\t\t\trequired for all targets other than the mgs node\n"
139 "\t\t--fsname=<filesystem_name> : default is 'lustre'\n"
140 "\t\t--failnode=<nid>[,<...>] : NID(s) of a failover partner\n"
141 "\t\t\tcannot be used with --servicenode\n"
142 "\t\t--servicenode=<nid>[,<...>] : NID(s) of all service partners\n"
143 "\t\t\ttreat all nodes as equal service node, cannot be used with --failnode\n"
144 "\t\t--param <key>=<value> : set a permanent parameter\n"
145 "\t\t\te.g. --param sys.timeout=40\n"
146 "\t\t\t --param lov.stripesize=2M\n"
147 "\t\t--index=#N : target index (i.e. ost index within lov)\n"
148 "\t\t--comment=<user comment>: arbitrary string (%d bytes)\n"
149 "\t\t--mountfsoptions=<opts> : permanent mount options\n"
150 "\t\t--network=<net>[,<...>] : restrict OST/MDT to network(s)\n"
152 "\t\t--backfstype=<fstype> : backing fs type (ext3, ldiskfs)\n"
153 "\t\t--device-size=#N(KB) : device size for loop devices\n"
154 "\t\t--mkfsoptions=<opts> : format options\n"
155 "\t\t--reformat: overwrite an existing disk\n"
156 "\t\t--stripe-count-hint=#N : for optimizing MDT inode size\n"
157 "\t\t--iam-dir: use IAM directory format, not ext3 compatible\n"
159 "\t\t--erase-params : erase all old parameter settings\n"
160 "\t\t--nomgs: turn off MGS service on this MDT\n"
161 "\t\t--writeconf: erase all config logs for this fs.\n"
163 "\t\t--dryrun: just report what we would do; "
164 "don't write to disk\n"
165 "\t\t--verbose : e.g. show mkfs progress\n"
167 (int)sizeof(((struct lustre_disk_data *)0)->ldd_userdata));
171 /* ==================== Lustre config functions =============*/
173 void print_ldd(char *str, struct lustre_disk_data *ldd)
175 printf("\n %s:\n", str);
176 printf("Target: %s\n", ldd->ldd_svname);
177 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
178 printf("Index: unassigned\n");
180 printf("Index: %d\n", ldd->ldd_svindex);
181 if (ldd->ldd_uuid[0])
182 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
183 printf("Lustre FS: %s\n", ldd->ldd_fsname);
184 printf("Mount type: %s\n", MT_STR(ldd));
185 printf("Flags: %#x\n", ldd->ldd_flags);
186 printf(" (%s%s%s%s%s%s%s%s%s%s)\n",
187 IS_MDT(ldd) ? "MDT ":"",
188 IS_OST(ldd) ? "OST ":"",
189 IS_MGS(ldd) ? "MGS ":"",
190 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
191 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
192 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
193 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
194 ldd->ldd_flags & LDD_F_IAM_DIR ? "IAM_dir_format ":"",
195 ldd->ldd_flags & LDD_F_NO_PRIMNODE? "no_primnode ":"",
196 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
197 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
198 printf("Parameters:%s\n", ldd->ldd_params);
199 if (ldd->ldd_userdata[0])
200 printf("Comment: %s\n", ldd->ldd_userdata);
204 void set_defaults(struct mkfs_opts *mop)
206 mop->mo_ldd.ldd_magic = LDD_MAGIC;
207 mop->mo_ldd.ldd_config_ver = 1;
208 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
209 mop->mo_mgs_failnodes = 0;
210 strcpy(mop->mo_ldd.ldd_fsname, "lustre");
211 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
213 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
214 mop->mo_stripe_count = 1;
215 mop->mo_pool_vdevs = NULL;
218 static inline void badopt(const char *opt, char *type)
220 fprintf(stderr, "%s: '--%s' only valid for %s\n",
221 progname, opt, type);
225 /* from mount_lustre */
226 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
227 #define MAXNIDSTR 1024
228 static char *convert_hostnames(char *s1)
230 char *converted, *s2 = 0, *c, *end, sep;
231 int left = MAXNIDSTR;
234 converted = malloc(left);
235 if (converted == NULL) {
239 end = s1 + strlen(s1);
241 while ((left > 0) && (s1 < end)) {
242 s2 = strpbrk(s1, ",:");
247 nid = libcfs_str2nid(s1);
249 if (nid == LNET_NID_ANY) {
250 fprintf(stderr, "%s: Can't parse NID '%s'\n",
255 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
256 strlen("127.0.0.1")) == 0) {
257 fprintf(stderr, "%s: The NID '%s' resolves to the "
258 "loopback address '%s'. Lustre requires a "
259 "non-loopback address.\n",
260 progname, s1, libcfs_nid2str(nid));
265 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
266 left = converted + MAXNIDSTR - c;
272 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
275 static struct option long_opt[] = {
276 {"iam-dir", 0, 0, 'a'},
277 {"backfstype", 1, 0, 'b'},
278 {"stripe-count-hint", 1, 0, 'c'},
279 {"comment", 1, 0, 'u'},
280 {"configdev", 1, 0, 'C'},
281 {"device-size", 1, 0, 'd'},
282 {"dryrun", 0, 0, 'n'},
283 {"erase-params", 0, 0, 'e'},
284 {"failnode", 1, 0, 'f'},
285 {"failover", 1, 0, 'f'},
288 {"index", 1, 0, 'i'},
289 {"mkfsoptions", 1, 0, 'k'},
290 {"mgsnode", 1, 0, 'm'},
291 {"mgsnid", 1, 0, 'm'},
293 {"fsname",1, 0, 'L'},
294 {"noformat", 0, 0, 'n'},
295 {"nomgs", 0, 0, 'N'},
296 {"mountfsoptions", 1, 0, 'o'},
298 {"param", 1, 0, 'p'},
299 {"print", 0, 0, 'n'},
300 {"quiet", 0, 0, 'q'},
301 {"reformat", 0, 0, 'r'},
302 {"servicenode", 1, 0, 's'},
303 {"verbose", 0, 0, 'v'},
304 {"writeconf", 0, 0, 'w'},
305 {"upgrade_to_18", 0, 0, 'U'},
306 {"network", 1, 0, 't'},
309 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqrs:t:Uu:vw";
312 int failnode_set = 0, servicenode_set = 0;
314 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
318 if (IS_MDT(&mop->mo_ldd))
319 mop->mo_ldd.ldd_flags |= LDD_F_IAM_DIR;
324 while (i < LDD_MT_LAST) {
325 if (strcmp(optarg, mt_str(i)) == 0) {
326 mop->mo_ldd.ldd_mount_type = i;
331 if (i == LDD_MT_LAST) {
332 fprintf(stderr, "%s: invalid backend filesystem"
333 " type %s\n", progname, optarg);
339 if (IS_MDT(&mop->mo_ldd)) {
340 int stripe_count = atol(optarg);
341 if (stripe_count <= 0) {
342 fprintf(stderr, "%s: bad stripe count "
343 "%d\n", progname, stripe_count);
346 mop->mo_stripe_count = stripe_count;
348 badopt(long_opt[longidx].name, "MDT");
352 case 'C': /* Configdev */
354 printf("Configdev not implemented\n");
357 mop->mo_device_sz = atol(optarg);
360 mop->mo_ldd.ldd_params[0] = '\0';
361 /* Must update the mgs logs */
362 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
368 if ((opt == 'f' && servicenode_set)
369 || (opt == 's' && failnode_set)) {
370 fprintf(stderr, "%s: %s cannot use with --%s\n",
371 progname, long_opt[longidx].name,
372 opt == 'f' ? "servicenode" : "failnode");
376 nids = convert_hostnames(optarg);
379 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
384 /* Must update the mgs logs */
385 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
389 mop->mo_ldd.ldd_flags |= LDD_F_NO_PRIMNODE;
392 mop->mo_flags |= MO_FAILOVER;
396 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
402 if (!(mop->mo_ldd.ldd_flags &
403 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
405 fprintf(stderr, "%s: cannot change the index of"
406 " a registered target\n", progname);
409 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
410 mop->mo_ldd.ldd_svindex = atol(optarg);
411 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
413 badopt(long_opt[longidx].name, "MDT,OST");
418 strscpy(mop->mo_mkfsopts, optarg,
419 sizeof(mop->mo_mkfsopts));
423 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
424 (!(mop->mo_ldd.ldd_flags &
425 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
426 LDD_F_WRITECONF)))) {
427 fprintf(stderr, "%s: cannot change the name of"
428 " a registered target\n", progname);
431 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
432 fprintf(stderr, "%s: filesystem name must be "
433 "1-8 chars\n", progname);
436 if ((tmp = strpbrk(optarg, "/:"))) {
437 fprintf(stderr, "%s: char '%c' not allowed in "
438 "filesystem name\n", progname, *tmp);
441 strscpy(mop->mo_ldd.ldd_fsname, optarg,
442 sizeof(mop->mo_ldd.ldd_fsname));
446 char *nids = convert_hostnames(optarg);
449 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
454 mop->mo_mgs_failnodes++;
458 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
464 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
470 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
473 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg);
476 /* Must update the mgs logs */
477 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
483 mop->mo_flags |= MO_FORCEFORMAT;
486 if (!IS_MDT(&mop->mo_ldd) && !IS_OST(&mop->mo_ldd)) {
487 badopt(long_opt[longidx].name, "MDT,OST");
494 rc = add_param(mop->mo_ldd.ldd_params,
495 PARAM_NETWORK, optarg);
498 /* Must update the mgs logs */
499 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
502 strscpy(mop->mo_ldd.ldd_userdata, optarg,
503 sizeof(mop->mo_ldd.ldd_userdata));
509 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
517 fprintf(stderr, "Unknown option '%c'\n", opt);
523 if (optind == argc) {
524 /* The user didn't specify device name */
526 fprintf(stderr, "Not enough arguments - device name or "
527 "pool/dataset name not specified.\n");
530 /* The device or pool/filesystem name */
531 strscpy(mop->mo_device, argv[optind], sizeof(mop->mo_device));
533 /* Followed by optional vdevs */
534 if (optind < argc - 1)
535 mop->mo_pool_vdevs = (char **) &argv[optind + 1];
541 int main(int argc, char *const argv[])
543 struct mkfs_opts mop;
544 struct lustre_disk_data *ldd;
545 char *mountopts = NULL;
546 char always_mountopts[512] = "";
547 char default_mountopts[512] = "";
551 if ((progname = strrchr(argv[0], '/')) != NULL)
556 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
561 memset(&mop, 0, sizeof(mop));
564 /* device is last arg */
565 strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
568 /* For tunefs, we must read in the old values before parsing any
571 /* Check whether the disk has already been formatted by mkfs.lustre */
572 ret = osd_is_lustre(mop.mo_device, &mount_type);
575 fprintf(stderr, "Device %s has not been formatted with "
576 "mkfs.lustre\n", mop.mo_device);
581 ret = osd_read_ldd(mop.mo_device, &mop.mo_ldd);
584 fprintf(stderr, "Failed to read previous Lustre data from %s "
585 "(%d)\n", mop.mo_device, ret);
588 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
589 mop.mo_mgs_failnodes++;
592 print_ldd("Read previous values", &(mop.mo_ldd));
599 ret = parse_opts(argc, argv, &mop, &mountopts);
605 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
607 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
612 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
614 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
619 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
620 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
622 fprintf(stderr, "Can't find the target index, "
623 "specify with --index\n");
629 * Comment out these 2 checks temporarily, since for multi-MDSes
630 * in single node only 1 mds node could have mgs service
632 if (IS_MDT(ldd) && !IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
633 verrprint("No management node specified, adding MGS to this "
635 ldd->ldd_flags |= LDD_F_SV_TYPE_MGS;
637 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
640 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
642 fprintf(stderr, "Must specify --mgsnode=\n");
648 /* These are the permanent mount options (always included) */
649 ret = osd_prepare_lustre(&mop,
650 default_mountopts, sizeof(default_mountopts),
651 always_mountopts, sizeof(always_mountopts));
654 fprintf(stderr, "unable to prepare backend (%d)\n", ret);
659 trim_mountfsoptions(mountopts);
660 (void)check_mountfsoptions(mountopts, default_mountopts, 1);
661 if (check_mountfsoptions(mountopts, always_mountopts, 0)) {
665 sprintf(ldd->ldd_mount_opts, "%s", mountopts);
668 if (ldd->ldd_mount_opts[0] == 0)
669 /* use the defaults unless old opts exist */
672 sprintf(ldd->ldd_mount_opts, "%s%s",
673 always_mountopts, default_mountopts);
674 trim_mountfsoptions(ldd->ldd_mount_opts);
678 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
679 ldd->ldd_fsname, ldd->ldd_svname);
682 print_ldd("Permanent disk data", ldd);
685 printf("exiting before disk write.\n");
689 if (check_mtab_entry(mop.mo_device, mop.mo_device, NULL, NULL))
692 /* Create the loopback file */
693 if (mop.mo_flags & MO_IS_LOOP) {
694 ret = access(mop.mo_device, F_OK);
697 #ifndef TUNEFS /* mkfs.lustre */
698 /* Reformat the loopback file */
699 if (ret || (mop.mo_flags & MO_FORCEFORMAT)) {
700 ret = loop_format(&mop);
706 ret = loop_setup(&mop);
709 fprintf(stderr, "Loop device setup for %s failed: %s\n",
710 mop.mo_device, strerror(ret));
715 #ifndef TUNEFS /* mkfs.lustre */
716 /* Check whether the disk has already been formatted by mkfs.lustre */
717 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
718 ret = osd_is_lustre(mop.mo_device, &mount_type);
721 fprintf(stderr, "Device %s was previously formatted "
722 "for lustre. Use --reformat to reformat it, "
723 "or tunefs.lustre to modify.\n",
729 /* Format the backing filesystem */
730 ret = osd_make_lustre(&mop);
733 fprintf(stderr, "mkfs failed %d\n", ret);
738 /* Write our config files */
739 ret = osd_write_ldd(&mop);
742 fprintf(stderr, "failed to write local files\n");
750 /* Fix any crazy return values from system() */
751 if (ret && ((ret & 255) == 0))
754 verrprint("%s: exiting with %d (%s)\n",
755 progname, ret, strerror(ret));