1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
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"
86 #define MAX_LOOP_DEVICES 16
87 #define L_BLOCK_SIZE 4096
88 #define INDEX_UNASSIGNED 0xFFFF
89 #define MO_IS_LOOP 0x01
90 #define MO_FORCEFORMAT 0x02
92 /* used to describe the options to format the lustre disk, not persistent */
94 struct lustre_disk_data mo_ldd; /* to be written in MOUNT_DATA_FILE */
95 char mo_device[128]; /* disk device name */
96 char mo_loopdev[128]; /* in case a loop dev is needed */
97 char mo_mkfsopts[512]; /* options to the backing-store mkfs */
98 __u64 mo_device_sz; /* in KB */
101 int mo_mgs_failnodes;
106 static int print_only = 0;
107 static int failover = 0;
108 static int upgrade_to_18 = 0;
110 void usage(FILE *out)
112 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
113 fprintf(out, "usage: %s <target types> [options] <device>\n", progname);
115 "\t<device>:block device or file (e.g /dev/sda or /tmp/ost1)\n"
117 "\t\t--ost: object storage, mutually exclusive with mdt,mgs\n"
118 "\t\t--mdt: metadata storage, mutually exclusive with ost\n"
119 "\t\t--mgs: configuration management service - one per site\n"
120 "\toptions (in order of popularity):\n"
121 "\t\t--mgsnode=<nid>[,<...>] : NID(s) of a remote mgs node\n"
122 "\t\t\trequired for all targets other than the mgs node\n"
123 "\t\t--fsname=<filesystem_name> : default is 'lustre'\n"
124 "\t\t--failnode=<nid>[,<...>] : NID(s) of a failover partner\n"
125 "\t\t--param <key>=<value> : set a permanent parameter\n"
126 "\t\t\te.g. --param sys.timeout=40\n"
127 "\t\t\t --param lov.stripesize=2M\n"
128 "\t\t--index=#N : target index (i.e. ost index within lov)\n"
129 "\t\t--comment=<user comment>: arbitrary string (%d bytes)\n"
130 "\t\t--mountfsoptions=<opts> : permanent mount options\n"
131 "\t\t--network=<net>[,<...>] : restrict OST/MDT to network(s)\n"
133 "\t\t--backfstype=<fstype> : backing fs type (ext3, ldiskfs)\n"
134 "\t\t--device-size=#N(KB) : device size for loop devices\n"
135 "\t\t--mkfsoptions=<opts> : format options\n"
136 "\t\t--reformat: overwrite an existing disk\n"
137 "\t\t--stripe-count-hint=#N : for optimizing MDT inode size\n"
138 "\t\t--iam-dir: use IAM directory format, not ext3 compatible\n"
140 "\t\t--erase-params : erase all old parameter settings\n"
141 "\t\t--nomgs: turn off MGS service on this MDT\n"
142 "\t\t--writeconf: erase all config logs for this fs.\n"
144 "\t\t--dryrun: just report what we would do; "
145 "don't write to disk\n"
146 "\t\t--verbose : e.g. show mkfs progress\n"
148 (int)sizeof(((struct lustre_disk_data *)0)->ldd_userdata));
152 #define vprint if (verbose > 0) printf
153 #define verrprint if (verbose >= 0) printf
155 /*================ utility functions =====================*/
157 char *strscat(char *dst, char *src, int buflen) {
159 if (strlen(dst) + strlen(src) >= buflen) {
160 fprintf(stderr, "string buffer overflow (max %d): '%s' + '%s'"
161 "\n", buflen, dst, src);
164 return strcat(dst, src);
168 char *strscpy(char *dst, char *src, int buflen) {
170 return strscat(dst, src, buflen);
174 dev_major (unsigned long long int __dev)
176 return ((__dev >> 8) & 0xfff) | ((unsigned int) (__dev >> 32) & ~0xfff);
180 dev_minor (unsigned long long int __dev)
182 return (__dev & 0xff) | ((unsigned int) (__dev >> 12) & ~0xff);
187 static int version = 0;
191 char release[4] = "";
193 fd = open("/proc/sys/kernel/osrelease", O_RDONLY);
195 fprintf(stderr, "%s: Warning: Can't resolve kernel "
196 "version, assuming 2.6\n", progname);
198 if (read(fd, release, 4) < 0) {
199 fprintf(stderr, "reading from /proc/sys/kernel"
200 "/osrelease: %s\n", strerror(errno));
206 if (strncmp(release, "2.4.", 4) == 0)
214 static int check_mtab_entry(char *spec)
219 fp = setmntent(MOUNTED, "r");
223 while ((mnt = getmntent(fp)) != NULL) {
224 if (strcmp(mnt->mnt_fsname, spec) == 0) {
226 fprintf(stderr, "%s: according to %s %s is "
227 "already mounted on %s\n",
228 progname, MOUNTED, spec, mnt->mnt_dir);
237 /*============ disk dev functions ===================*/
239 /* Setup a file in the first unused loop_device */
240 int loop_setup(struct mkfs_opts *mop)
246 /* Figure out the loop device names */
247 if (!access("/dev/loop0", F_OK | R_OK)) {
248 strcpy(loop_base, "/dev/loop\0");
249 } else if (!access("/dev/loop/0", F_OK | R_OK)) {
250 strcpy(loop_base, "/dev/loop/\0");
252 fprintf(stderr, "%s: can't access loop devices\n", progname);
256 /* Find unused loop device */
257 for (i = 0; i < MAX_LOOP_DEVICES; i++) {
259 int cmdsz = sizeof(cmd);
261 sprintf(l_device, "%s%d", loop_base, i);
262 if (access(l_device, F_OK | R_OK))
264 snprintf(cmd, cmdsz, "losetup %s > /dev/null 2>&1", l_device);
267 /* losetup gets 1 (ret=256) for non-set-up device */
269 /* Set up a loopback device to our file */
270 snprintf(cmd, cmdsz, "losetup %s %s", l_device,
272 ret = run_command(cmd, cmdsz);
274 /* someone else picked up this loop device
278 fprintf(stderr, "%s: error %d on losetup: %s\n",
279 progname, ret, strerror(ret));
282 strscpy(mop->mo_loopdev, l_device,
283 sizeof(mop->mo_loopdev));
288 fprintf(stderr, "%s: out of loop devices!\n", progname);
292 int loop_cleanup(struct mkfs_opts *mop)
296 if ((mop->mo_flags & MO_IS_LOOP) && *mop->mo_loopdev) {
297 sprintf(cmd, "losetup -d %s", mop->mo_loopdev);
298 ret = run_command(cmd, sizeof(cmd));
303 /* Determine if a device is a block device (as opposed to a file) */
304 int is_block(char* devname)
309 ret = access(devname, F_OK);
312 ret = stat(devname, &st);
314 fprintf(stderr, "%s: cannot stat %s\n", progname, devname);
317 return S_ISBLK(st.st_mode);
320 __u64 get_device_size(char* device)
325 fd = open(device, O_RDONLY);
327 fprintf(stderr, "%s: cannot open %s: %s\n",
328 progname, device, strerror(errno));
333 /* size in bytes. bz5831 */
334 ret = ioctl(fd, BLKGETSIZE64, (void*)&size);
339 ret = ioctl(fd, BLKGETSIZE, (void*)&lsize);
340 size = (__u64)lsize * 512;
345 fprintf(stderr, "%s: size ioctl failed: %s\n",
346 progname, strerror(errno));
350 vprint("device size = "LPU64"MB\n", size >> 20);
351 /* return value in KB */
355 int loop_format(struct mkfs_opts *mop)
359 if (mop->mo_device_sz == 0) {
361 fprintf(stderr, "loop device requires a --device-size= "
366 ret = creat(mop->mo_device, S_IRUSR|S_IWUSR);
369 fprintf(stderr, "%s: Unable to create backing store: %d\n",
375 ret = truncate(mop->mo_device, mop->mo_device_sz * 1024);
378 fprintf(stderr, "%s: Unable to truncate backing store: %d\n",
385 /* Display the need for the latest e2fsprogs to be installed. make_backfs
386 * indicates if the caller is make_lustre_backfs() or not. */
387 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
389 static int msg_displayed;
392 fprintf(stderr, "WARNING: %s does not support %s "
393 "feature.\n\n", E2FSPROGS, feature);
399 fprintf(stderr, "WARNING: The %s package currently installed on "
400 "your system does not support \"%s\" feature.\n",
402 #if !(HAVE_LDISKFSPROGS)
403 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
404 "http://downloads.lustre.org/public/tools/e2fsprogs/\n"
405 "to enable this feature.\n");
408 fprintf(stderr, "Feature will not be enabled until %s"
409 "is updated and '%s -O %s %%{device}' "
410 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
413 /* Check whether the file exists in the device */
414 static int file_in_dev(char *file_name, char *dev_name)
417 char debugfs_cmd[256];
418 unsigned int inode_num;
421 /* Construct debugfs command line. */
422 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
423 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
424 DEBUGFS, file_name, dev_name);
426 fp = popen(debugfs_cmd, "r");
428 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
432 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
436 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd), fp);
439 fprintf(stderr, "%s", debugfs_cmd);
440 if (strstr(debugfs_cmd, "unsupported feature")) {
441 disp_old_e2fsprogs_msg("an unknown", 0);
450 /* Check whether the device has already been used with lustre */
451 static int is_lustre_target(struct mkfs_opts *mop)
455 vprint("checking for existing Lustre data: ");
457 if ((rc = file_in_dev(MOUNT_DATA_FILE, mop->mo_device))) {
459 (rc == 1) ? MOUNT_DATA_FILE : "extents");
460 /* in the -1 case, 'extents' means this really IS a lustre
465 if ((rc = file_in_dev(LAST_RCVD, mop->mo_device))) {
466 vprint("found %s\n", LAST_RCVD);
470 vprint("not found\n");
471 return 0; /* The device is not a lustre target. */
474 /* Check if a certain feature is supported by e2fsprogs.
475 * Firstly we try to use "debugfs supported_features" command to check if
476 * the feature is supported. If this fails we try to set this feature with
477 * mke2fs to check for its support. */
478 static int is_e2fsprogs_feature_supp(const char *feature)
480 static char supp_features[4096] = "";
483 char imgname[] = "/tmp/test-img-XXXXXX";
487 if (supp_features[0] == '\0') {
488 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
491 /* Using popen() instead of run_command() since debugfs does
492 * not return proper error code if command is not supported */
493 fp = popen(cmd, "r");
495 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
498 ret = fread(supp_features, 1, sizeof(supp_features), fp);
501 if (ret > 0 && strstr(supp_features,
502 strncmp(feature, "-O ", 3) ? feature : feature+3))
505 if ((fd = mkstemp(imgname)) < 0)
510 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
511 MKE2FS, feature, imgname);
512 /* run_command() displays the output of mke2fs when it fails for
513 * some feature, so use system() directly */
520 static void enable_default_ext4_features(struct mkfs_opts *mop)
522 if (IS_OST(&mop->mo_ldd))
523 strscat(mop->mo_mkfsopts, " -O extents,uninit_bg",
524 sizeof(mop->mo_mkfsopts));
525 else if (IS_MDT(&mop->mo_ldd))
526 strscat(mop->mo_mkfsopts, " -O dirdata,uninit_bg",
527 sizeof(mop->mo_mkfsopts));
529 strscat(mop->mo_mkfsopts, " -O uninit_bg",
530 sizeof(mop->mo_mkfsopts));
532 /* Multiple mount protection enabled only if failover node specified */
534 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
535 strscat(mop->mo_mkfsopts, ",mmp",
536 sizeof(mop->mo_mkfsopts));
538 disp_old_e2fsprogs_msg("mmp", 1);
541 /* Allow more than 65000 subdirectories */
542 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
543 strscat(mop->mo_mkfsopts,",dir_nlink",sizeof(mop->mo_mkfsopts));
545 #ifdef HAVE_EXT4_LDISKFS
546 /* The following options are only valid for ext4-based ldiskfs.
547 * If --backfstype=ext3 is specified, do not enable them. */
548 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
551 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
552 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
553 strscat(mop->mo_mkfsopts,",huge_file",sizeof(mop->mo_mkfsopts));
555 /* Cluster inode/block bitmaps and inode table for more efficient IO.
556 * Align the flex groups on a 1MB boundary for better performance.
557 * This -O feature needs to go last, since it adds an extra option. */
558 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
561 strscat(mop->mo_mkfsopts, ",flex_bg", sizeof(mop->mo_mkfsopts));
563 if (IS_OST(&mop->mo_ldd)) {
564 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
565 (1 << 20) / L_BLOCK_SIZE);
566 strscat(mop->mo_mkfsopts, tmp_buf,
567 sizeof(mop->mo_mkfsopts));
573 /* Build fs according to type */
574 int make_lustre_backfs(struct mkfs_opts *mop)
576 __u64 device_sz = mop->mo_device_sz, block_count = 0;
577 char mkfs_cmd[PATH_MAX];
582 if (!(mop->mo_flags & MO_IS_LOOP)) {
583 mop->mo_device_sz = get_device_size(mop->mo_device);
585 if (mop->mo_device_sz == 0)
588 /* Compare to real size */
589 if (device_sz == 0 || device_sz > mop->mo_device_sz)
590 device_sz = mop->mo_device_sz;
592 mop->mo_device_sz = device_sz;
595 if (mop->mo_device_sz != 0) {
596 if (mop->mo_device_sz < 8096){
597 fprintf(stderr, "%s: size of filesystem must be larger "
598 "than 8MB, but is set to %lldKB\n",
599 progname, (long long)mop->mo_device_sz);
602 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
603 /* If the LUN size is just over 2^32 blocks, limit the
604 * filesystem size to 2^32-1 blocks to avoid problems with
605 * ldiskfs/mkfs not handling this size. Bug 22906 */
606 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
607 block_count = 0xffffffffULL;
610 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
611 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
612 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
615 /* Journal size in MB */
616 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
617 /* Choose our own default journal size */
618 long journal_sz = 0, max_sz;
619 if (device_sz > 1024 * 1024) /* 1GB */
620 journal_sz = (device_sz / 102400) * 4;
621 /* cap journal size at 1GB */
622 if (journal_sz > 1024L)
625 max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
626 if (journal_sz > max_sz)
629 sprintf(buf, " -J size=%ld", journal_sz);
630 strscat(mop->mo_mkfsopts, buf,
631 sizeof(mop->mo_mkfsopts));
635 /* Inode size (for extended attributes). The LOV EA size is
636 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
637 * and we want some margin above that for ACLs, other EAs... */
638 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
639 if (IS_MDT(&mop->mo_ldd)) {
640 if (mop->mo_stripe_count > 72)
641 inode_size = 512; /* bz 7241 */
642 /* see also "-i" below for EA blocks */
643 else if (mop->mo_stripe_count > 32)
645 else if (mop->mo_stripe_count > 10)
649 } else if (IS_OST(&mop->mo_ldd)) {
650 /* We store MDS FID and OST objid in EA on OST
651 * we need to make inode bigger as well. */
655 if (inode_size > 0) {
656 sprintf(buf, " -I %ld", inode_size);
657 strscat(mop->mo_mkfsopts, buf,
658 sizeof(mop->mo_mkfsopts));
662 /* Bytes_per_inode: disk size / num inodes */
663 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
664 strstr(mop->mo_mkfsopts, "-N") == NULL) {
665 long bytes_per_inode = 0;
667 /* Allocate more inodes on MDT devices. There is
668 * no data stored on the MDT, and very little extra
669 * metadata beyond the inode. It could go down as
670 * low as 1024 bytes, but this is conservative.
671 * Account for external EA blocks for wide striping. */
672 if (IS_MDT(&mop->mo_ldd)) {
673 bytes_per_inode = inode_size + 1536;
675 if (mop->mo_stripe_count > 72) {
676 int extra = mop->mo_stripe_count * 24;
677 extra = ((extra - 1) | 4095) + 1;
678 bytes_per_inode += extra;
682 /* Allocate fewer inodes on large OST devices. Most
683 * filesystems can be much more aggressive than even
684 * this, but it is impossible to know in advance. */
685 if (IS_OST(&mop->mo_ldd)) {
686 /* OST > 8TB assume average file size 1MB */
687 if (device_sz >= (8ULL << 30))
688 bytes_per_inode = 1024 * 1024;
689 /* OST > 1TB assume average file size 256kB */
690 else if (device_sz >= (1ULL << 30))
691 bytes_per_inode = 256 * 1024;
692 /* OST > 100GB assume average file size 64kB,
693 * plus a bit so that inodes will fit into a
694 * 256x flex_bg without overflowing */
695 else if (device_sz >= (10ULL << 20))
696 bytes_per_inode = 69905;
700 if (bytes_per_inode > 0) {
701 sprintf(buf, " -i %ld", bytes_per_inode);
702 strscat(mop->mo_mkfsopts, buf,
703 sizeof(mop->mo_mkfsopts));
708 strscat(mop->mo_mkfsopts, " -q",
709 sizeof(mop->mo_mkfsopts));
712 if (strstr(mop->mo_mkfsopts, "-O") == NULL)
713 enable_default_ext4_features(mop);
715 /* In order to align the filesystem metadata on 1MB boundaries,
716 * give a resize value that will reserve a power-of-two group
717 * descriptor blocks, but leave one block for the superblock.
718 * Only useful for filesystems with < 2^32 blocks due to resize
720 if (IS_OST(&mop->mo_ldd) && mop->mo_device_sz > 100 * 1024 &&
721 mop->mo_device_sz / L_BLOCK_SIZE <= 0xffffffff) {
722 unsigned group_blocks = L_BLOCK_SIZE * 8;
723 unsigned desc_per_block = L_BLOCK_SIZE / 32;
724 unsigned resize_blks;
726 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
727 snprintf(buf, sizeof(buf)," -E resize=%u,",resize_blks);
729 strncpy(buf, " -E ", sizeof(buf));
732 /* Avoid zeroing out the full journal - speeds up mkfs */
733 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
734 strscat(buf, "lazy_journal_init,", sizeof(buf));
736 if (strlen(buf) > strlen(" -E "))
737 strscat(mop->mo_mkfsopts, buf,sizeof(mop->mo_mkfsopts));
739 /* Allow reformat of full devices (as opposed to
740 partitions.) We already checked for mounted dev. */
741 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
743 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
744 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
745 mop->mo_ldd.ldd_svname);
747 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
748 progname, mop->mo_ldd.ldd_mount_type,
749 MT_STR(&mop->mo_ldd));
753 /* For loop device format the dev, not the filename */
754 dev = mop->mo_device;
755 if (mop->mo_flags & MO_IS_LOOP)
756 dev = mop->mo_loopdev;
758 vprint("formatting backing filesystem %s on %s\n",
759 MT_STR(&mop->mo_ldd), dev);
760 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
761 vprint("\t4k blocks "LPU64"\n", block_count);
762 vprint("\toptions %s\n", mop->mo_mkfsopts);
764 /* mkfs_cmd's trailing space is important! */
765 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
766 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
767 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
768 if (block_count != 0) {
769 sprintf(buf, " "LPU64, block_count);
770 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
773 vprint("mkfs_cmd = %s\n", mkfs_cmd);
774 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
777 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
782 /* ==================== Lustre config functions =============*/
784 void print_ldd(char *str, struct lustre_disk_data *ldd)
786 printf("\n %s:\n", str);
787 printf("Target: %s\n", ldd->ldd_svname);
788 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
789 printf("Index: unassigned\n");
791 printf("Index: %d\n", ldd->ldd_svindex);
792 if (ldd->ldd_uuid[0])
793 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
794 printf("Lustre FS: %s\n", ldd->ldd_fsname);
795 printf("Mount type: %s\n", MT_STR(ldd));
796 printf("Flags: %#x\n", ldd->ldd_flags);
797 printf(" (%s%s%s%s%s%s%s%s%s)\n",
798 IS_MDT(ldd) ? "MDT ":"",
799 IS_OST(ldd) ? "OST ":"",
800 IS_MGS(ldd) ? "MGS ":"",
801 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
802 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
803 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
804 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
805 ldd->ldd_flags & LDD_F_IAM_DIR ? "IAM_dir_format ":"",
806 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
807 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
808 printf("Parameters:%s\n", ldd->ldd_params);
809 if (ldd->ldd_userdata[0])
810 printf("Comment: %s\n", ldd->ldd_userdata);
814 static int touch_file(char *filename)
818 if (filename == NULL) {
822 fd = open(filename, O_CREAT | O_TRUNC, 0600);
831 /* keep it less than LL_FID_NAMELEN */
832 #define DUMMY_FILE_NAME_LEN 25
833 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
835 /* Need to add these many entries to this directory to make HTREE dir. */
836 #define MIN_ENTRIES_REQ_FOR_HTREE ((L_BLOCK_SIZE / EXT3_DIRENT_SIZE))
838 static int add_dummy_files(char *dir)
840 char fpname[PATH_MAX];
844 for (i = 0; i < MIN_ENTRIES_REQ_FOR_HTREE; i++) {
845 snprintf(fpname, PATH_MAX, "%s/%0*d", dir,
846 DUMMY_FILE_NAME_LEN, i);
848 rc = touch_file(fpname);
849 if (rc && rc != -EEXIST) {
851 "%s: Can't create dummy file %s: %s\n",
852 progname, fpname , strerror(errno));
859 static int __l_mkdir(char * filepnm, int mode , struct mkfs_opts *mop)
863 ret = mkdir(filepnm, mode);
864 if (ret && ret != -EEXIST)
867 /* IAM mode supports ext3 directories of HTREE type only. So add dummy
868 * entries to new directory to create htree type of container for
870 if (mop->mo_ldd.ldd_flags & LDD_F_IAM_DIR)
871 return add_dummy_files(filepnm);
875 /* Write the server config files */
876 int write_local_files(struct mkfs_opts *mop)
878 char mntpt[] = "/tmp/mntXXXXXX";
885 /* Mount this device temporarily in order to write these files */
886 if (!mkdtemp(mntpt)) {
887 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
888 progname, mntpt, strerror(errno));
892 dev = mop->mo_device;
893 if (mop->mo_flags & MO_IS_LOOP)
894 dev = mop->mo_loopdev;
896 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
897 mop->mo_ldd.ldd_mount_opts);
899 fprintf(stderr, "%s: Unable to mount %s: %s\n",
900 progname, dev, strerror(errno));
902 if (errno == ENODEV) {
903 fprintf(stderr, "Is the %s module available?\n",
904 MT_STR(&mop->mo_ldd));
909 /* Set up initial directories */
910 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
911 ret = __l_mkdir(filepnm, 0777, mop);
912 if ((ret != 0) && (errno != EEXIST)) {
913 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
914 progname, filepnm, strerror(errno));
916 } else if (errno == EEXIST) {
920 /* Save the persistent mount data into a file. Lustre must pre-read
921 this file to get the real mount options. */
922 vprint("Writing %s\n", MOUNT_DATA_FILE);
923 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
924 filep = fopen(filepnm, "w");
926 fprintf(stderr, "%s: Unable to create %s file: %s\n",
927 progname, filepnm, strerror(errno));
930 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
931 if (num < 1 && ferror(filep)) {
932 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
933 progname, filepnm, strerror(errno));
939 /* Check for upgrade */
940 if ((mop->mo_ldd.ldd_flags & (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS))
941 == (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS)) {
944 int cmdsz = sizeof(cmd);
945 vprint("Copying old logs\n");
947 /* Copy the old client log to fsname-client */
948 sprintf(filepnm, "%s/%s/%s-client",
949 mntpt, MOUNT_CONFIGS_DIR, mop->mo_ldd.ldd_fsname);
950 snprintf(cmd, cmdsz, "cp %s/%s/client %s", mntpt, MDT_LOGS_DIR,
952 ret = run_command(cmd, cmdsz);
954 fprintf(stderr, "%s: Can't copy 1.4 config %s/client "
955 "(%d)\n", progname, MDT_LOGS_DIR, ret);
956 fprintf(stderr, "mount -t ldiskfs %s somewhere, "
957 "find the client log for fs %s and "
958 "copy it manually into %s/%s-client, "
961 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
962 mop->mo_ldd.ldd_fsname);
966 /* We need to use the old mdt log because otherwise mdt won't
967 have complete lov if old clients connect before all
969 /* Copy the old mdt log to fsname-MDT0000 (get old
970 name from mdt_UUID) */
972 strscpy(filepnm, (char *)mop->mo_ldd.ldd_uuid, sizeof(filepnm));
973 term = strstr(filepnm, "_UUID");
976 snprintf(cmd, cmdsz, "cp %s/%s/%s %s/%s/%s",
977 mntpt, MDT_LOGS_DIR, filepnm,
978 mntpt, MOUNT_CONFIGS_DIR,
979 mop->mo_ldd.ldd_svname);
980 ret = run_command(cmd, cmdsz);
983 fprintf(stderr, "%s: Can't copy 1.4 config %s/%s "
984 "(%d)\n", progname, MDT_LOGS_DIR, filepnm, ret);
985 fprintf(stderr, "mount -t ext3 %s somewhere, "
986 "find the MDT log for fs %s and "
987 "copy it manually into %s/%s, "
990 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
991 mop->mo_ldd.ldd_svname);
1005 int read_local_files(struct mkfs_opts *mop)
1007 char tmpdir[] = "/tmp/dirXXXXXX";
1013 int cmdsz = sizeof(cmd);
1015 /* Make a temporary directory to hold Lustre data files. */
1016 if (!mkdtemp(tmpdir)) {
1017 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
1018 progname, tmpdir, strerror(errno));
1022 dev = mop->mo_device;
1024 /* TODO: it's worth observing the get_mountdata() function that is
1025 in mount_utils.c for getting the mountdata out of the
1028 /* Construct debugfs command line. */
1029 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
1030 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
1032 ret = run_command(cmd, cmdsz);
1034 verrprint("%s: Unable to dump %s dir (%d)\n",
1035 progname, MOUNT_CONFIGS_DIR, ret);
1037 sprintf(filepnm, "%s/mountdata", tmpdir);
1038 filep = fopen(filepnm, "r");
1041 vprint("Reading %s\n", MOUNT_DATA_FILE);
1042 num_read = fread(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
1043 if (num_read < 1 && ferror(filep)) {
1044 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
1045 progname, filepnm, strerror(errno));
1050 /* Try to read pre-1.6 config from last_rcvd */
1051 struct lr_server_data lsd;
1052 verrprint("%s: Unable to read %d.%d config %s.\n",
1053 progname, LUSTRE_MAJOR, LUSTRE_MINOR, filepnm);
1055 verrprint("Trying 1.4 config from last_rcvd\n");
1056 sprintf(filepnm, "%s/%s", tmpdir, LAST_RCVD);
1058 /* Construct debugfs command line. */
1059 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s' %s",
1060 DEBUGFS, LAST_RCVD, filepnm, dev);
1062 ret = run_command(cmd, cmdsz);
1064 fprintf(stderr, "%s: Unable to dump %s file (%d)\n",
1065 progname, LAST_RCVD, ret);
1069 filep = fopen(filepnm, "r");
1071 fprintf(stderr, "%s: Unable to open %s: %s\n",
1072 progname, filepnm, strerror(errno));
1074 verrprint("Contents of %s:\n", tmpdir);
1076 snprintf(cmd, cmdsz, "ls -l %s/", tmpdir);
1077 run_command(cmd, cmdsz);
1078 verrprint("Contents of disk:\n");
1079 snprintf(cmd, cmdsz, "%s -c -R 'ls -l /' %s",
1081 run_command(cmd, cmdsz);
1085 vprint("Reading %s\n", LAST_RCVD);
1086 ret = fread(&lsd, 1, sizeof(lsd), filep);
1087 if (ret < sizeof(lsd)) {
1088 fprintf(stderr, "%s: Short read (%d of %d)\n",
1089 progname, ret, (int)sizeof(lsd));
1090 ret = ferror(filep);
1094 vprint("Feature compat=%x, incompat=%x\n",
1095 lsd.lsd_feature_compat, lsd.lsd_feature_incompat);
1097 if ((lsd.lsd_feature_compat & OBD_COMPAT_OST) ||
1098 (lsd.lsd_feature_incompat & OBD_INCOMPAT_OST)) {
1099 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
1100 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
1101 } else if ((lsd.lsd_feature_compat & OBD_COMPAT_MDT) ||
1102 (lsd.lsd_feature_incompat & OBD_INCOMPAT_MDT)) {
1103 /* We must co-locate so mgs can see old logs.
1104 If user doesn't want this, they can copy the old
1105 logs manually and re-tunefs. */
1106 mop->mo_ldd.ldd_flags =
1107 LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_MGS;
1108 mop->mo_ldd.ldd_svindex = lsd.lsd_mdt_index;
1110 /* If neither is set, we're pre-1.4.6, make a guess. */
1111 /* Construct debugfs command line. */
1112 snprintf(cmd, cmdsz, "%s -c -R 'rdump /%s %s' %s",
1113 DEBUGFS, MDT_LOGS_DIR, tmpdir, dev);
1114 run_command(cmd, cmdsz);
1116 sprintf(filepnm, "%s/%s", tmpdir, MDT_LOGS_DIR);
1117 if (lsd.lsd_ost_index > 0) {
1118 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
1119 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
1121 /* If there's a LOGS dir, it's an MDT */
1122 if ((ret = access(filepnm, F_OK)) == 0) {
1123 mop->mo_ldd.ldd_flags =
1126 /* Old MDT's are always index 0
1128 mop->mo_ldd.ldd_svindex = 0;
1130 /* The index may not be correct */
1131 mop->mo_ldd.ldd_flags =
1132 LDD_F_SV_TYPE_OST | LDD_F_NEED_INDEX;
1133 verrprint("OST with unknown index\n");
1139 memcpy(mop->mo_ldd.ldd_uuid, lsd.lsd_uuid,
1140 sizeof(mop->mo_ldd.ldd_uuid));
1141 mop->mo_ldd.ldd_flags |= LDD_F_UPGRADE14;
1143 /* end COMPAT_146 */
1148 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
1149 run_command(cmd, cmdsz);
1151 verrprint("Failed to read old data (%d)\n", ret);
1156 void set_defaults(struct mkfs_opts *mop)
1158 mop->mo_ldd.ldd_magic = LDD_MAGIC;
1159 mop->mo_ldd.ldd_config_ver = 1;
1160 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
1161 mop->mo_mgs_failnodes = 0;
1162 strcpy(mop->mo_ldd.ldd_fsname, "lustre");
1163 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
1165 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
1166 mop->mo_stripe_count = 1;
1169 static inline void badopt(const char *opt, char *type)
1171 fprintf(stderr, "%s: '--%s' only valid for %s\n",
1172 progname, opt, type);
1176 static int add_param(char *buf, char *key, char *val)
1178 int end = sizeof(((struct lustre_disk_data *)0)->ldd_params);
1179 int start = strlen(buf);
1183 keylen = strlen(key);
1184 if (start + 1 + keylen + strlen(val) >= end) {
1185 fprintf(stderr, "%s: params are too long-\n%s %s%s\n",
1186 progname, buf, key ? key : "", val);
1190 sprintf(buf + start, " %s%s", key ? key : "", val);
1194 /* from mount_lustre */
1195 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
1196 #define MAXNIDSTR 1024
1197 static char *convert_hostnames(char *s1)
1199 char *converted, *s2 = 0, *c, *end, sep;
1200 int left = MAXNIDSTR;
1203 converted = malloc(left);
1204 if (converted == NULL) {
1208 end = s1 + strlen(s1);
1210 while ((left > 0) && (s1 < end)) {
1211 s2 = strpbrk(s1, ",:");
1216 nid = libcfs_str2nid(s1);
1218 if (nid == LNET_NID_ANY) {
1219 fprintf(stderr, "%s: Can't parse NID '%s'\n",
1224 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
1225 strlen("127.0.0.1")) == 0) {
1226 fprintf(stderr, "%s: The NID '%s' resolves to the "
1227 "loopback address '%s'. Lustre requires a "
1228 "non-loopback address.\n",
1229 progname, s1, libcfs_nid2str(nid));
1234 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
1235 left = converted + MAXNIDSTR - c;
1241 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
1244 static struct option long_opt[] = {
1245 {"iam-dir", 0, 0, 'a'},
1246 {"backfstype", 1, 0, 'b'},
1247 {"stripe-count-hint", 1, 0, 'c'},
1248 {"comment", 1, 0, 'u'},
1249 {"configdev", 1, 0, 'C'},
1250 {"device-size", 1, 0, 'd'},
1251 {"dryrun", 0, 0, 'n'},
1252 {"erase-params", 0, 0, 'e'},
1253 {"failnode", 1, 0, 'f'},
1254 {"failover", 1, 0, 'f'},
1256 {"help", 0, 0, 'h'},
1257 {"index", 1, 0, 'i'},
1258 {"mkfsoptions", 1, 0, 'k'},
1259 {"mgsnode", 1, 0, 'm'},
1260 {"mgsnid", 1, 0, 'm'},
1262 {"fsname",1, 0, 'L'},
1263 {"noformat", 0, 0, 'n'},
1264 {"nomgs", 0, 0, 'N'},
1265 {"mountfsoptions", 1, 0, 'o'},
1267 {"param", 1, 0, 'p'},
1268 {"print", 0, 0, 'n'},
1269 {"quiet", 0, 0, 'q'},
1270 {"reformat", 0, 0, 'r'},
1271 {"verbose", 0, 0, 'v'},
1272 {"writeconf", 0, 0, 'w'},
1273 {"upgrade_to_18", 0, 0, 'U'},
1274 {"network", 1, 0, 't'},
1277 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqru:vw";
1281 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
1285 if (IS_MDT(&mop->mo_ldd))
1286 mop->mo_ldd.ldd_flags |= LDD_F_IAM_DIR;
1291 while (i < LDD_MT_LAST) {
1292 if (strcmp(optarg, mt_str(i)) == 0) {
1293 mop->mo_ldd.ldd_mount_type = i;
1301 if (IS_MDT(&mop->mo_ldd)) {
1302 int stripe_count = atol(optarg);
1303 if (stripe_count <= 0) {
1304 fprintf(stderr, "%s: bad stripe count "
1305 "%d\n", progname, stripe_count);
1308 mop->mo_stripe_count = stripe_count;
1310 badopt(long_opt[longidx].name, "MDT");
1314 case 'C': /* Configdev */
1316 printf("Configdev not implemented\n");
1319 mop->mo_device_sz = atol(optarg);
1322 mop->mo_ldd.ldd_params[0] = '\0';
1323 /* Must update the mgs logs */
1324 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1327 char *nids = convert_hostnames(optarg);
1330 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
1335 /* Must update the mgs logs */
1336 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1341 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
1347 if (!(mop->mo_ldd.ldd_flags &
1348 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1349 LDD_F_WRITECONF))) {
1350 fprintf(stderr, "%s: cannot change the index of"
1351 " a registered target\n", progname);
1354 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
1355 mop->mo_ldd.ldd_svindex = atol(optarg);
1356 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
1358 badopt(long_opt[longidx].name, "MDT,OST");
1363 strscpy(mop->mo_mkfsopts, optarg,
1364 sizeof(mop->mo_mkfsopts));
1368 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
1369 (!(mop->mo_ldd.ldd_flags &
1370 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1371 LDD_F_WRITECONF)))) {
1372 fprintf(stderr, "%s: cannot change the name of"
1373 " a registered target\n", progname);
1376 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
1377 fprintf(stderr, "%s: filesystem name must be "
1378 "1-8 chars\n", progname);
1381 if ((tmp = strpbrk(optarg, "/:"))) {
1382 fprintf(stderr, "%s: char '%c' not allowed in "
1383 "filesystem name\n", progname, *tmp);
1386 strscpy(mop->mo_ldd.ldd_fsname, optarg,
1387 sizeof(mop->mo_ldd.ldd_fsname));
1391 char *nids = convert_hostnames(optarg);
1394 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
1399 mop->mo_mgs_failnodes++;
1403 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
1409 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
1412 *mountopts = optarg;
1415 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
1418 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg);
1421 /* Must update the mgs logs */
1422 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1428 mop->mo_flags |= MO_FORCEFORMAT;
1431 if (!IS_MDT(&mop->mo_ldd) && !IS_OST(&mop->mo_ldd)) {
1432 badopt(long_opt[longidx].name, "MDT,OST");
1439 rc = add_param(mop->mo_ldd.ldd_params,
1440 PARAM_NETWORK, optarg);
1443 /* Must update the mgs logs */
1444 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1447 strscpy(mop->mo_ldd.ldd_userdata, optarg,
1448 sizeof(mop->mo_ldd.ldd_userdata));
1454 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
1462 fprintf(stderr, "Unknown option '%c'\n", opt);
1468 /* Last arg is device */
1469 if (optind != argc - 1) {
1471 fprintf(stderr, "Bad argument: %s\n", argv[optind]);
1475 /* single argument: <device> */
1482 /* Search for opt in mntlist, returning true if found.
1484 static int in_mntlist(char *opt, char *mntlist)
1486 char *ml, *mlp, *item, *ctx = NULL;
1488 if (!(ml = strdup(mntlist))) {
1489 fprintf(stderr, "%s: out of memory\n", progname);
1493 while ((item = strtok_r(mlp, ",", &ctx))) {
1494 if (!strcmp(opt, item))
1499 return (item != NULL);
1502 /* Issue a message on stderr for every item in wanted_mountopts that is not
1503 * present in mountopts. The justwarn boolean toggles between error and
1504 * warning message. Return an error count.
1506 static int check_mountfsoptions(char *mountopts, char *wanted_mountopts,
1509 char *ml, *mlp, *item, *ctx = NULL;
1512 if (!(ml = strdup(wanted_mountopts))) {
1513 fprintf(stderr, "%s: out of memory\n", progname);
1517 while ((item = strtok_r(mlp, ",", &ctx))) {
1518 if (!in_mntlist(item, mountopts)) {
1519 fprintf(stderr, "%s: %s mount option `%s' is missing\n",
1520 progname, justwarn ? "Warning: default"
1521 : "Error: mandatory", item);
1530 /* Trim embedded white space, leading and trailing commas from string s.
1532 static void trim_mountfsoptions(char *s)
1538 memmove(p, p + 1, strlen(p + 1) + 1);
1545 memmove(&s[0], &s[1], strlen(&s[1]) + 1);
1547 p = s + strlen(s) - 1;
1548 while (p >= s && *p == ',')
1552 int main(int argc, char *const argv[])
1554 struct mkfs_opts mop;
1555 struct lustre_disk_data *ldd;
1556 char *mountopts = NULL;
1557 char always_mountopts[512] = "";
1558 char default_mountopts[512] = "";
1561 if ((progname = strrchr(argv[0], '/')) != NULL)
1566 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
1571 memset(&mop, 0, sizeof(mop));
1574 /* device is last arg */
1575 strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
1577 /* Are we using a loop device? */
1578 ret = is_block(mop.mo_device);
1582 mop.mo_flags |= MO_IS_LOOP;
1585 /* For tunefs, we must read in the old values before parsing any
1588 /* Check whether the disk has already been formatted by mkfs.lustre */
1589 ret = is_lustre_target(&mop);
1592 fprintf(stderr, "Device %s has not been formatted with "
1593 "mkfs.lustre\n", mop.mo_device);
1598 ret = read_local_files(&mop);
1601 fprintf(stderr, "Failed to read previous Lustre data from %s "
1602 "(%d)\n", mop.mo_device, ret);
1605 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
1606 mop.mo_mgs_failnodes++;
1609 print_ldd("Read previous values", &(mop.mo_ldd));
1612 ret = parse_opts(argc, argv, &mop, &mountopts);
1618 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
1620 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
1625 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
1627 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
1632 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
1633 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
1635 fprintf(stderr, "Can't find the target index, "
1636 "specify with --index\n");
1642 * Comment out these 2 checks temporarily, since for multi-MDSes
1643 * in single node only 1 mds node could have mgs service
1645 if (IS_MDT(ldd) && !IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1646 verrprint("No management node specified, adding MGS to this "
1648 ldd->ldd_flags |= LDD_F_SV_TYPE_MGS;
1650 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1653 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
1655 fprintf(stderr, "Must specify --mgsnode=\n");
1661 /* These are the permanent mount options (always included) */
1662 switch (ldd->ldd_mount_type) {
1664 case LDD_MT_LDISKFS:
1665 case LDD_MT_LDISKFS2:
1666 strscat(default_mountopts, ",errors=remount-ro",
1667 sizeof(default_mountopts));
1668 if (IS_MDT(ldd) || IS_MGS(ldd))
1669 strscat(always_mountopts, ",user_xattr",
1670 sizeof(always_mountopts));
1671 /* NB: Files created while extents are enabled can only be read
1672 * if mounted using the ext4 or ldiskfs filesystem type. */
1674 (ldd->ldd_mount_type == LDD_MT_LDISKFS ||
1675 ldd->ldd_mount_type == LDD_MT_LDISKFS2)) {
1676 strscat(default_mountopts, ",extents,mballoc",
1677 sizeof(default_mountopts));
1682 fprintf(stderr, "unknown fs type %d '%s'\n",
1683 ldd->ldd_mount_type,
1690 trim_mountfsoptions(mountopts);
1691 (void)check_mountfsoptions(mountopts, default_mountopts, 1);
1692 if (check_mountfsoptions(mountopts, always_mountopts, 0)) {
1696 sprintf(ldd->ldd_mount_opts, "%s", mountopts);
1699 if (ldd->ldd_mount_opts[0] == 0)
1700 /* use the defaults unless old opts exist */
1703 sprintf(ldd->ldd_mount_opts, "%s%s",
1704 always_mountopts, default_mountopts);
1705 trim_mountfsoptions(ldd->ldd_mount_opts);
1709 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
1710 ldd->ldd_fsname, ldd->ldd_svname);
1713 print_ldd("Permanent disk data", ldd);
1716 printf("exiting before disk write.\n");
1720 if (check_mtab_entry(mop.mo_device))
1723 /* Create the loopback file */
1724 if (mop.mo_flags & MO_IS_LOOP) {
1725 ret = access(mop.mo_device, F_OK);
1728 #ifndef TUNEFS /* mkfs.lustre */
1729 /* Reformat the loopback file */
1730 if (ret || (mop.mo_flags & MO_FORCEFORMAT))
1731 ret = loop_format(&mop);
1734 ret = loop_setup(&mop);
1737 fprintf(stderr, "Loop device setup for %s failed: %s\n",
1738 mop.mo_device, strerror(ret));
1743 #ifndef TUNEFS /* mkfs.lustre */
1744 /* Check whether the disk has already been formatted by mkfs.lustre */
1745 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
1746 ret = is_lustre_target(&mop);
1749 fprintf(stderr, "Device %s was previously formatted "
1750 "for lustre. Use --reformat to reformat it, "
1751 "or tunefs.lustre to modify.\n",
1757 /* Format the backing filesystem */
1758 ret = make_lustre_backfs(&mop);
1761 fprintf(stderr, "mkfs failed %d\n", ret);
1766 /* Write our config files */
1767 ret = write_local_files(&mop);
1770 fprintf(stderr, "failed to write local files\n");
1777 /* Fix any crazy return values from system() */
1778 if (ret && ((ret & 255) == 0))
1781 verrprint("%s: exiting with %d (%s)\n",
1782 progname, ret, strerror(ret));