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\tcannot be used with --servicenode\n"
126 "\t\t--servicenode=<nid>[,<...>] : NID(s) of all service partners\n"
127 "\t\t\ttreat all nodes as equal service node, cannot be used with --failnode\n"
128 "\t\t--param <key>=<value> : set a permanent parameter\n"
129 "\t\t\te.g. --param sys.timeout=40\n"
130 "\t\t\t --param lov.stripesize=2M\n"
131 "\t\t--index=#N : target index (i.e. ost index within lov)\n"
132 "\t\t--comment=<user comment>: arbitrary string (%d bytes)\n"
133 "\t\t--mountfsoptions=<opts> : permanent mount options\n"
134 "\t\t--network=<net>[,<...>] : restrict OST/MDT to network(s)\n"
136 "\t\t--backfstype=<fstype> : backing fs type (ext3, ldiskfs)\n"
137 "\t\t--device-size=#N(KB) : device size for loop devices\n"
138 "\t\t--mkfsoptions=<opts> : format options\n"
139 "\t\t--reformat: overwrite an existing disk\n"
140 "\t\t--stripe-count-hint=#N : for optimizing MDT inode size\n"
141 "\t\t--iam-dir: use IAM directory format, not ext3 compatible\n"
143 "\t\t--erase-params : erase all old parameter settings\n"
144 "\t\t--nomgs: turn off MGS service on this MDT\n"
145 "\t\t--writeconf: erase all config logs for this fs.\n"
147 "\t\t--dryrun: just report what we would do; "
148 "don't write to disk\n"
149 "\t\t--verbose : e.g. show mkfs progress\n"
151 (int)sizeof(((struct lustre_disk_data *)0)->ldd_userdata));
155 #define vprint if (verbose > 0) printf
156 #define verrprint if (verbose >= 0) printf
158 /*================ utility functions =====================*/
160 char *strscat(char *dst, char *src, int buflen) {
162 if (strlen(dst) + strlen(src) >= buflen) {
163 fprintf(stderr, "string buffer overflow (max %d): '%s' + '%s'"
164 "\n", buflen, dst, src);
167 return strcat(dst, src);
171 char *strscpy(char *dst, char *src, int buflen) {
173 return strscat(dst, src, buflen);
177 dev_major (unsigned long long int __dev)
179 return ((__dev >> 8) & 0xfff) | ((unsigned int) (__dev >> 32) & ~0xfff);
183 dev_minor (unsigned long long int __dev)
185 return (__dev & 0xff) | ((unsigned int) (__dev >> 12) & ~0xff);
190 static int version = 0;
194 char release[4] = "";
196 fd = open("/proc/sys/kernel/osrelease", O_RDONLY);
198 fprintf(stderr, "%s: Warning: Can't resolve kernel "
199 "version, assuming 2.6\n", progname);
201 if (read(fd, release, 4) < 0) {
202 fprintf(stderr, "reading from /proc/sys/kernel"
203 "/osrelease: %s\n", strerror(errno));
209 if (strncmp(release, "2.4.", 4) == 0)
217 static int check_mtab_entry(char *spec)
222 fp = setmntent(MOUNTED, "r");
226 while ((mnt = getmntent(fp)) != NULL) {
227 if (strcmp(mnt->mnt_fsname, spec) == 0) {
229 fprintf(stderr, "%s: according to %s %s is "
230 "already mounted on %s\n",
231 progname, MOUNTED, spec, mnt->mnt_dir);
240 /*============ disk dev functions ===================*/
242 /* Setup a file in the first unused loop_device */
243 int loop_setup(struct mkfs_opts *mop)
249 /* Figure out the loop device names */
250 if (!access("/dev/loop0", F_OK | R_OK)) {
251 strcpy(loop_base, "/dev/loop\0");
252 } else if (!access("/dev/loop/0", F_OK | R_OK)) {
253 strcpy(loop_base, "/dev/loop/\0");
255 fprintf(stderr, "%s: can't access loop devices\n", progname);
259 /* Find unused loop device */
260 for (i = 0; i < MAX_LOOP_DEVICES; i++) {
262 int cmdsz = sizeof(cmd);
264 sprintf(l_device, "%s%d", loop_base, i);
265 if (access(l_device, F_OK | R_OK))
267 snprintf(cmd, cmdsz, "losetup %s > /dev/null 2>&1", l_device);
270 /* losetup gets 1 (ret=256) for non-set-up device */
272 /* Set up a loopback device to our file */
273 snprintf(cmd, cmdsz, "losetup %s %s", l_device,
275 ret = run_command(cmd, cmdsz);
277 /* someone else picked up this loop device
281 fprintf(stderr, "%s: error %d on losetup: %s\n",
282 progname, ret, strerror(ret));
285 strscpy(mop->mo_loopdev, l_device,
286 sizeof(mop->mo_loopdev));
291 fprintf(stderr, "%s: out of loop devices!\n", progname);
295 int loop_cleanup(struct mkfs_opts *mop)
299 if ((mop->mo_flags & MO_IS_LOOP) && *mop->mo_loopdev) {
300 sprintf(cmd, "losetup -d %s", mop->mo_loopdev);
301 ret = run_command(cmd, sizeof(cmd));
306 /* Determine if a device is a block device (as opposed to a file) */
307 int is_block(char* devname)
312 ret = access(devname, F_OK);
315 ret = stat(devname, &st);
317 fprintf(stderr, "%s: cannot stat %s\n", progname, devname);
320 return S_ISBLK(st.st_mode);
323 __u64 get_device_size(char* device)
328 fd = open(device, O_RDONLY);
330 fprintf(stderr, "%s: cannot open %s: %s\n",
331 progname, device, strerror(errno));
336 /* size in bytes. bz5831 */
337 ret = ioctl(fd, BLKGETSIZE64, (void*)&size);
342 ret = ioctl(fd, BLKGETSIZE, (void*)&lsize);
343 size = (__u64)lsize * 512;
348 fprintf(stderr, "%s: size ioctl failed: %s\n",
349 progname, strerror(errno));
353 vprint("device size = "LPU64"MB\n", size >> 20);
354 /* return value in KB */
358 int loop_format(struct mkfs_opts *mop)
362 if (mop->mo_device_sz == 0) {
364 fprintf(stderr, "loop device requires a --device-size= "
369 ret = creat(mop->mo_device, S_IRUSR|S_IWUSR);
372 fprintf(stderr, "%s: Unable to create backing store: %d\n",
378 ret = truncate(mop->mo_device, mop->mo_device_sz * 1024);
381 fprintf(stderr, "%s: Unable to truncate backing store: %d\n",
388 /* Display the need for the latest e2fsprogs to be installed. make_backfs
389 * indicates if the caller is make_lustre_backfs() or not. */
390 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
392 static int msg_displayed;
395 fprintf(stderr, "WARNING: %s does not support %s "
396 "feature.\n\n", E2FSPROGS, feature);
402 fprintf(stderr, "WARNING: The %s package currently installed on "
403 "your system does not support \"%s\" feature.\n",
405 #if !(HAVE_LDISKFSPROGS)
406 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
407 "http://downloads.whamcloud.com/public/e2fsprogs/latest/\n"
408 "to enable this feature.\n");
411 fprintf(stderr, "Feature will not be enabled until %s"
412 "is updated and '%s -O %s %%{device}' "
413 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
416 /* Check whether the file exists in the device */
417 static int file_in_dev(char *file_name, char *dev_name)
420 char debugfs_cmd[256];
421 unsigned int inode_num;
424 /* Construct debugfs command line. */
425 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
426 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
427 DEBUGFS, file_name, dev_name);
429 fp = popen(debugfs_cmd, "r");
431 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
435 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
439 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd), fp);
442 fprintf(stderr, "%s", debugfs_cmd);
443 if (strstr(debugfs_cmd, "unsupported feature")) {
444 disp_old_e2fsprogs_msg("an unknown", 0);
453 /* Check whether the device has already been used with lustre */
454 static int is_lustre_target(struct mkfs_opts *mop)
458 vprint("checking for existing Lustre data: ");
460 if ((rc = file_in_dev(MOUNT_DATA_FILE, mop->mo_device))) {
462 (rc == 1) ? MOUNT_DATA_FILE : "extents");
463 /* in the -1 case, 'extents' means this really IS a lustre
468 if ((rc = file_in_dev(LAST_RCVD, mop->mo_device))) {
469 vprint("found %s\n", LAST_RCVD);
473 vprint("not found\n");
474 return 0; /* The device is not a lustre target. */
477 /* Check if a certain feature is supported by e2fsprogs.
478 * Firstly we try to use "debugfs supported_features" command to check if
479 * the feature is supported. If this fails we try to set this feature with
480 * mke2fs to check for its support. */
481 static int is_e2fsprogs_feature_supp(const char *feature)
483 static char supp_features[4096] = "";
486 char imgname[] = "/tmp/test-img-XXXXXX";
490 if (supp_features[0] == '\0') {
491 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
494 /* Using popen() instead of run_command() since debugfs does
495 * not return proper error code if command is not supported */
496 fp = popen(cmd, "r");
498 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
501 ret = fread(supp_features, 1, sizeof(supp_features), fp);
504 if (ret > 0 && strstr(supp_features,
505 strncmp(feature, "-O ", 3) ? feature : feature+3))
508 if ((fd = mkstemp(imgname)) < 0)
513 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
514 MKE2FS, feature, imgname);
515 /* run_command() displays the output of mke2fs when it fails for
516 * some feature, so use system() directly */
524 * append_unique: append @key or @key=@val pair to @buf only if @key does not
526 * @buf: buffer to hold @key or @key=@val
527 * @prefix: prefix string before @key
529 * @val: value string if it's a @key=@val pair
531 static void append_unique(char *buf, char *prefix, char *key, char *val,
540 anchor = end = strstr(buf, key);
541 /* try to find exact match string in @buf */
542 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
545 if (anchor == NULL || strlen(key) != len ||
546 strncmp(anchor, key, len) != 0) {
548 strscat(buf, prefix, maxbuflen);
550 strscat(buf, key, maxbuflen);
552 strscat(buf, "=", maxbuflen);
553 strscat(buf, val, maxbuflen);
558 static void enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
559 size_t maxbuflen, int user_spec)
561 if (IS_OST(&mop->mo_ldd)) {
562 append_unique(anchor, user_spec ? "," : " -O ",
563 "extents", NULL, sizeof(mop->mo_mkfsopts));
564 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
565 } else if (IS_MDT(&mop->mo_ldd)) {
566 append_unique(anchor, user_spec ? "," : " -O ",
567 "dirdata", NULL, maxbuflen);
568 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
570 append_unique(anchor, user_spec ? "," : " -O ",
571 "uninit_bg", NULL, maxbuflen);
574 /* Multiple mount protection enabled only if failover node specified */
576 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
577 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
579 disp_old_e2fsprogs_msg("mmp", 1);
582 /* Allow more than 65000 subdirectories */
583 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
584 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
586 #ifdef HAVE_EXT4_LDISKFS
587 /* The following options are only valid for ext4-based ldiskfs.
588 * If --backfstype=ext3 is specified, do not enable them. */
589 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
592 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
593 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
594 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
596 /* Enable large block addresses if the LUN is over 2^32 blocks. */
597 if (mop->mo_device_sz / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
598 is_e2fsprogs_feature_supp("-O 64bit") == 0)
599 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
601 /* Cluster inode/block bitmaps and inode table for more efficient IO.
602 * Align the flex groups on a 1MB boundary for better performance. */
603 /* This -O feature needs to go last, since it adds the "-G" option. */
604 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
607 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
609 if (IS_OST(&mop->mo_ldd)) {
610 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
611 (1 << 20) / L_BLOCK_SIZE);
612 strscat(anchor, tmp_buf, maxbuflen);
615 /* Don't add any more "-O" options here, see last comment above */
620 * moveopts_to_end: find the option string, move remaining strings to
621 * where option string starts, and append the option
623 * @start: where the option string starts before the move
624 * RETURN: where the option string starts after the move
626 static char *moveopts_to_end(char *start)
631 /* skip whitespace before options */
636 /* find end of option characters */
637 while (*end != ' ' && *end != '\0')
641 strncpy(save, start, end - start);
642 save[end - start] = '\0';
644 /* move remaining options up front */
646 memmove(start, end, strlen(end));
647 *(start + strlen(end)) = '\0';
649 /* append the specified options */
650 if (*(start + strlen(start) - 1) != ' ')
652 idx = start + strlen(start);
658 /* Build fs according to type */
659 int make_lustre_backfs(struct mkfs_opts *mop)
661 __u64 device_sz = mop->mo_device_sz, block_count = 0;
662 char mkfs_cmd[PATH_MAX];
666 int ret = 0, ext_opts = 0;
669 if (!(mop->mo_flags & MO_IS_LOOP)) {
670 mop->mo_device_sz = get_device_size(mop->mo_device);
672 if (mop->mo_device_sz == 0)
675 /* Compare to real size */
676 if (device_sz == 0 || device_sz > mop->mo_device_sz)
677 device_sz = mop->mo_device_sz;
679 mop->mo_device_sz = device_sz;
682 if (mop->mo_device_sz != 0) {
683 if (mop->mo_device_sz < 8096){
684 fprintf(stderr, "%s: size of filesystem must be larger "
685 "than 8MB, but is set to %lldKB\n",
686 progname, (long long)mop->mo_device_sz);
689 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
690 /* If the LUN size is just over 2^32 blocks, limit the
691 * filesystem size to 2^32-1 blocks to avoid problems with
692 * ldiskfs/mkfs not handling this size. Bug 22906 */
693 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
694 block_count = 0xffffffffULL;
697 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
698 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
699 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
702 /* Journal size in MB */
703 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
704 /* Choose our own default journal size */
705 long journal_sz = 0, max_sz;
706 if (device_sz > 1024 * 1024) /* 1GB */
707 journal_sz = (device_sz / 102400) * 4;
708 /* cap journal size at 1GB */
709 if (journal_sz > 1024L)
712 max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
713 if (journal_sz > max_sz)
716 sprintf(buf, " -J size=%ld", journal_sz);
717 strscat(mop->mo_mkfsopts, buf,
718 sizeof(mop->mo_mkfsopts));
722 /* Inode size (for extended attributes). The LOV EA size is
723 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
724 * and we want some margin above that for ACLs, other EAs... */
725 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
726 if (IS_MDT(&mop->mo_ldd)) {
727 if (mop->mo_stripe_count > 72)
728 inode_size = 512; /* bz 7241 */
729 /* see also "-i" below for EA blocks */
730 else if (mop->mo_stripe_count > 32)
732 else if (mop->mo_stripe_count > 10)
736 } else if (IS_OST(&mop->mo_ldd)) {
737 /* We store MDS FID and OST objid in EA on OST
738 * we need to make inode bigger as well. */
742 if (inode_size > 0) {
743 sprintf(buf, " -I %ld", inode_size);
744 strscat(mop->mo_mkfsopts, buf,
745 sizeof(mop->mo_mkfsopts));
749 /* Bytes_per_inode: disk size / num inodes */
750 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
751 strstr(mop->mo_mkfsopts, "-N") == NULL) {
752 long bytes_per_inode = 0;
754 /* Allocate more inodes on MDT devices. There is
755 * no data stored on the MDT, and very little extra
756 * metadata beyond the inode. It could go down as
757 * low as 1024 bytes, but this is conservative.
758 * Account for external EA blocks for wide striping. */
759 if (IS_MDT(&mop->mo_ldd)) {
760 bytes_per_inode = inode_size + 1536;
762 if (mop->mo_stripe_count > 72) {
763 int extra = mop->mo_stripe_count * 24;
764 extra = ((extra - 1) | 4095) + 1;
765 bytes_per_inode += extra;
769 /* Allocate fewer inodes on large OST devices. Most
770 * filesystems can be much more aggressive than even
771 * this, but it is impossible to know in advance. */
772 if (IS_OST(&mop->mo_ldd)) {
773 /* OST > 16TB assume average file size 1MB */
774 if (device_sz > (16ULL << 30))
775 bytes_per_inode = 1024 * 1024;
776 /* OST > 4TB assume average file size 512kB */
777 else if (device_sz > (4ULL << 30))
778 bytes_per_inode = 512 * 1024;
779 /* OST > 1TB assume average file size 256kB */
780 else if (device_sz > (1ULL << 30))
781 bytes_per_inode = 256 * 1024;
782 /* OST > 10GB assume average file size 64kB,
783 * plus a bit so that inodes will fit into a
784 * 256x flex_bg without overflowing */
785 else if (device_sz > (10ULL << 20))
786 bytes_per_inode = 69905;
789 if (bytes_per_inode > 0) {
790 sprintf(buf, " -i %ld", bytes_per_inode);
791 strscat(mop->mo_mkfsopts, buf,
792 sizeof(mop->mo_mkfsopts));
797 strscat(mop->mo_mkfsopts, " -q",
798 sizeof(mop->mo_mkfsopts));
801 /* start handle -O mkfs options */
802 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
803 if (strstr(start + 2, "-O") != NULL) {
805 "%s: don't specify multiple -O options\n",
809 start = moveopts_to_end(start);
810 maxbuflen = sizeof(mop->mo_mkfsopts) -
811 (start - mop->mo_mkfsopts) - strlen(start);
812 enable_default_ext4_features(mop, start, maxbuflen, 1);
814 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
815 maxbuflen = sizeof(mop->mo_mkfsopts) -
816 strlen(mop->mo_mkfsopts);
817 enable_default_ext4_features(mop, start, maxbuflen, 0);
819 /* end handle -O mkfs options */
821 /* start handle -E mkfs options */
822 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
823 if (strstr(start + 2, "-E") != NULL) {
825 "%s: don't specify multiple -E options\n",
829 start = moveopts_to_end(start);
830 maxbuflen = sizeof(mop->mo_mkfsopts) -
831 (start - mop->mo_mkfsopts) - strlen(start);
834 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
835 maxbuflen = sizeof(mop->mo_mkfsopts) -
836 strlen(mop->mo_mkfsopts);
839 /* In order to align the filesystem metadata on 1MB boundaries,
840 * give a resize value that will reserve a power-of-two group
841 * descriptor blocks, but leave one block for the superblock.
842 * Only useful for filesystems with < 2^32 blocks due to resize
844 if (IS_OST(&mop->mo_ldd) && mop->mo_device_sz > 100 * 1024 &&
845 mop->mo_device_sz * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
846 unsigned group_blocks = L_BLOCK_SIZE * 8;
847 unsigned desc_per_block = L_BLOCK_SIZE / 32;
848 unsigned resize_blks;
850 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
851 snprintf(buf, sizeof(buf), "%u", resize_blks);
852 append_unique(start, ext_opts ? "," : " -E ",
853 "resize", buf, maxbuflen);
857 /* Avoid zeroing out the full journal - speeds up mkfs */
858 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
859 append_unique(start, ext_opts ? "," : " -E ",
860 "lazy_journal_init", NULL, maxbuflen);
861 /* end handle -E mkfs options */
863 /* Allow reformat of full devices (as opposed to
864 partitions.) We already checked for mounted dev. */
865 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
867 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
868 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
869 mop->mo_ldd.ldd_svname);
871 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
872 progname, mop->mo_ldd.ldd_mount_type,
873 MT_STR(&mop->mo_ldd));
877 /* For loop device format the dev, not the filename */
878 dev = mop->mo_device;
879 if (mop->mo_flags & MO_IS_LOOP)
880 dev = mop->mo_loopdev;
882 vprint("formatting backing filesystem %s on %s\n",
883 MT_STR(&mop->mo_ldd), dev);
884 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
885 vprint("\t4k blocks "LPU64"\n", block_count);
886 vprint("\toptions %s\n", mop->mo_mkfsopts);
888 /* mkfs_cmd's trailing space is important! */
889 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
890 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
891 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
892 if (block_count != 0) {
893 sprintf(buf, " "LPU64, block_count);
894 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
897 vprint("mkfs_cmd = %s\n", mkfs_cmd);
898 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
901 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
906 /* ==================== Lustre config functions =============*/
908 void print_ldd(char *str, struct lustre_disk_data *ldd)
910 printf("\n %s:\n", str);
911 printf("Target: %s\n", ldd->ldd_svname);
912 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
913 printf("Index: unassigned\n");
915 printf("Index: %d\n", ldd->ldd_svindex);
916 if (ldd->ldd_uuid[0])
917 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
918 printf("Lustre FS: %s\n", ldd->ldd_fsname);
919 printf("Mount type: %s\n", MT_STR(ldd));
920 printf("Flags: %#x\n", ldd->ldd_flags);
921 printf(" (%s%s%s%s%s%s%s%s%s%s)\n",
922 IS_MDT(ldd) ? "MDT ":"",
923 IS_OST(ldd) ? "OST ":"",
924 IS_MGS(ldd) ? "MGS ":"",
925 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
926 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
927 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
928 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
929 ldd->ldd_flags & LDD_F_IAM_DIR ? "IAM_dir_format ":"",
930 ldd->ldd_flags & LDD_F_NO_PRIMNODE? "no_primnode ":"",
931 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
932 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
933 printf("Parameters:%s\n", ldd->ldd_params);
934 if (ldd->ldd_userdata[0])
935 printf("Comment: %s\n", ldd->ldd_userdata);
939 static int touch_file(char *filename)
943 if (filename == NULL) {
947 fd = open(filename, O_CREAT | O_TRUNC, 0600);
956 /* keep it less than LL_FID_NAMELEN */
957 #define DUMMY_FILE_NAME_LEN 25
958 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
960 /* Need to add these many entries to this directory to make HTREE dir. */
961 #define MIN_ENTRIES_REQ_FOR_HTREE ((L_BLOCK_SIZE / EXT3_DIRENT_SIZE))
963 static int add_dummy_files(char *dir)
965 char fpname[PATH_MAX];
969 for (i = 0; i < MIN_ENTRIES_REQ_FOR_HTREE; i++) {
970 snprintf(fpname, PATH_MAX, "%s/%0*d", dir,
971 DUMMY_FILE_NAME_LEN, i);
973 rc = touch_file(fpname);
974 if (rc && rc != -EEXIST) {
976 "%s: Can't create dummy file %s: %s\n",
977 progname, fpname , strerror(errno));
984 static int __l_mkdir(char * filepnm, int mode , struct mkfs_opts *mop)
988 ret = mkdir(filepnm, mode);
989 if (ret && ret != -EEXIST)
992 /* IAM mode supports ext3 directories of HTREE type only. So add dummy
993 * entries to new directory to create htree type of container for
995 if (mop->mo_ldd.ldd_flags & LDD_F_IAM_DIR)
996 return add_dummy_files(filepnm);
1000 /* Write the server config files */
1001 int write_local_files(struct mkfs_opts *mop)
1003 char mntpt[] = "/tmp/mntXXXXXX";
1010 /* Mount this device temporarily in order to write these files */
1011 if (!mkdtemp(mntpt)) {
1012 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
1013 progname, mntpt, strerror(errno));
1017 dev = mop->mo_device;
1018 if (mop->mo_flags & MO_IS_LOOP)
1019 dev = mop->mo_loopdev;
1021 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
1022 mop->mo_ldd.ldd_mount_opts);
1024 fprintf(stderr, "%s: Unable to mount %s: %s\n",
1025 progname, dev, strerror(errno));
1027 if (errno == ENODEV) {
1028 fprintf(stderr, "Is the %s module available?\n",
1029 MT_STR(&mop->mo_ldd));
1034 /* Set up initial directories */
1035 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
1036 ret = __l_mkdir(filepnm, 0777, mop);
1037 if ((ret != 0) && (errno != EEXIST)) {
1038 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
1039 progname, filepnm, strerror(errno));
1041 } else if (errno == EEXIST) {
1045 /* Save the persistent mount data into a file. Lustre must pre-read
1046 this file to get the real mount options. */
1047 vprint("Writing %s\n", MOUNT_DATA_FILE);
1048 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
1049 filep = fopen(filepnm, "w");
1051 fprintf(stderr, "%s: Unable to create %s file: %s\n",
1052 progname, filepnm, strerror(errno));
1055 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
1056 if (num < 1 && ferror(filep)) {
1057 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
1058 progname, filepnm, strerror(errno));
1064 /* Check for upgrade */
1065 if ((mop->mo_ldd.ldd_flags & (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS))
1066 == (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS)) {
1069 int cmdsz = sizeof(cmd);
1070 vprint("Copying old logs\n");
1072 /* Copy the old client log to fsname-client */
1073 sprintf(filepnm, "%s/%s/%s-client",
1074 mntpt, MOUNT_CONFIGS_DIR, mop->mo_ldd.ldd_fsname);
1075 snprintf(cmd, cmdsz, "cp %s/%s/client %s", mntpt, MDT_LOGS_DIR,
1077 ret = run_command(cmd, cmdsz);
1079 fprintf(stderr, "%s: Can't copy 1.4 config %s/client "
1080 "(%d)\n", progname, MDT_LOGS_DIR, ret);
1081 fprintf(stderr, "mount -t ldiskfs %s somewhere, "
1082 "find the client log for fs %s and "
1083 "copy it manually into %s/%s-client, "
1086 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
1087 mop->mo_ldd.ldd_fsname);
1091 /* We need to use the old mdt log because otherwise mdt won't
1092 have complete lov if old clients connect before all
1094 /* Copy the old mdt log to fsname-MDT0000 (get old
1095 name from mdt_UUID) */
1097 strscpy(filepnm, (char *)mop->mo_ldd.ldd_uuid, sizeof(filepnm));
1098 term = strstr(filepnm, "_UUID");
1101 snprintf(cmd, cmdsz, "cp %s/%s/%s %s/%s/%s",
1102 mntpt, MDT_LOGS_DIR, filepnm,
1103 mntpt, MOUNT_CONFIGS_DIR,
1104 mop->mo_ldd.ldd_svname);
1105 ret = run_command(cmd, cmdsz);
1108 fprintf(stderr, "%s: Can't copy 1.4 config %s/%s "
1109 "(%d)\n", progname, MDT_LOGS_DIR, filepnm, ret);
1110 fprintf(stderr, "mount -t ext3 %s somewhere, "
1111 "find the MDT log for fs %s and "
1112 "copy it manually into %s/%s, "
1115 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
1116 mop->mo_ldd.ldd_svname);
1121 /* end COMPAT_146 */
1130 int read_local_files(struct mkfs_opts *mop)
1132 char tmpdir[] = "/tmp/dirXXXXXX";
1138 int cmdsz = sizeof(cmd);
1140 /* Make a temporary directory to hold Lustre data files. */
1141 if (!mkdtemp(tmpdir)) {
1142 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
1143 progname, tmpdir, strerror(errno));
1147 dev = mop->mo_device;
1149 /* TODO: it's worth observing the get_mountdata() function that is
1150 in mount_utils.c for getting the mountdata out of the
1153 /* Construct debugfs command line. */
1154 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
1155 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
1157 ret = run_command(cmd, cmdsz);
1159 verrprint("%s: Unable to dump %s dir (%d)\n",
1160 progname, MOUNT_CONFIGS_DIR, ret);
1162 sprintf(filepnm, "%s/mountdata", tmpdir);
1163 filep = fopen(filepnm, "r");
1166 vprint("Reading %s\n", MOUNT_DATA_FILE);
1167 num_read = fread(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
1168 if (num_read < 1 && ferror(filep)) {
1169 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
1170 progname, filepnm, strerror(errno));
1175 /* Try to read pre-1.6 config from last_rcvd */
1176 struct lr_server_data lsd;
1177 verrprint("%s: Unable to read %d.%d config %s.\n",
1178 progname, LUSTRE_MAJOR, LUSTRE_MINOR, filepnm);
1180 verrprint("Trying 1.4 config from last_rcvd\n");
1181 sprintf(filepnm, "%s/%s", tmpdir, LAST_RCVD);
1183 /* Construct debugfs command line. */
1184 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s' %s",
1185 DEBUGFS, LAST_RCVD, filepnm, dev);
1187 ret = run_command(cmd, cmdsz);
1189 fprintf(stderr, "%s: Unable to dump %s file (%d)\n",
1190 progname, LAST_RCVD, ret);
1194 filep = fopen(filepnm, "r");
1196 fprintf(stderr, "%s: Unable to open %s: %s\n",
1197 progname, filepnm, strerror(errno));
1199 verrprint("Contents of %s:\n", tmpdir);
1201 snprintf(cmd, cmdsz, "ls -l %s/", tmpdir);
1202 run_command(cmd, cmdsz);
1203 verrprint("Contents of disk:\n");
1204 snprintf(cmd, cmdsz, "%s -c -R 'ls -l /' %s",
1206 run_command(cmd, cmdsz);
1210 vprint("Reading %s\n", LAST_RCVD);
1211 ret = fread(&lsd, 1, sizeof(lsd), filep);
1212 if (ret < sizeof(lsd)) {
1213 fprintf(stderr, "%s: Short read (%d of %d)\n",
1214 progname, ret, (int)sizeof(lsd));
1215 ret = ferror(filep);
1219 vprint("Feature compat=%x, incompat=%x\n",
1220 lsd.lsd_feature_compat, lsd.lsd_feature_incompat);
1222 if ((lsd.lsd_feature_compat & OBD_COMPAT_OST) ||
1223 (lsd.lsd_feature_incompat & OBD_INCOMPAT_OST)) {
1224 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
1225 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
1226 } else if ((lsd.lsd_feature_compat & OBD_COMPAT_MDT) ||
1227 (lsd.lsd_feature_incompat & OBD_INCOMPAT_MDT)) {
1228 /* We must co-locate so mgs can see old logs.
1229 If user doesn't want this, they can copy the old
1230 logs manually and re-tunefs. */
1231 mop->mo_ldd.ldd_flags =
1232 LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_MGS;
1233 mop->mo_ldd.ldd_svindex = lsd.lsd_mdt_index;
1235 /* If neither is set, we're pre-1.4.6, make a guess. */
1236 /* Construct debugfs command line. */
1237 snprintf(cmd, cmdsz, "%s -c -R 'rdump /%s %s' %s",
1238 DEBUGFS, MDT_LOGS_DIR, tmpdir, dev);
1239 run_command(cmd, cmdsz);
1241 sprintf(filepnm, "%s/%s", tmpdir, MDT_LOGS_DIR);
1242 if (lsd.lsd_ost_index > 0) {
1243 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
1244 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
1246 /* If there's a LOGS dir, it's an MDT */
1247 if ((ret = access(filepnm, F_OK)) == 0) {
1248 mop->mo_ldd.ldd_flags =
1251 /* Old MDT's are always index 0
1253 mop->mo_ldd.ldd_svindex = 0;
1255 /* The index may not be correct */
1256 mop->mo_ldd.ldd_flags =
1257 LDD_F_SV_TYPE_OST | LDD_F_NEED_INDEX;
1258 verrprint("OST with unknown index\n");
1264 memcpy(mop->mo_ldd.ldd_uuid, lsd.lsd_uuid,
1265 sizeof(mop->mo_ldd.ldd_uuid));
1266 mop->mo_ldd.ldd_flags |= LDD_F_UPGRADE14;
1268 /* end COMPAT_146 */
1273 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
1274 run_command(cmd, cmdsz);
1276 verrprint("Failed to read old data (%d)\n", ret);
1281 void set_defaults(struct mkfs_opts *mop)
1283 mop->mo_ldd.ldd_magic = LDD_MAGIC;
1284 mop->mo_ldd.ldd_config_ver = 1;
1285 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
1286 mop->mo_mgs_failnodes = 0;
1287 strcpy(mop->mo_ldd.ldd_fsname, "lustre");
1288 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
1290 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
1291 mop->mo_stripe_count = 1;
1294 static inline void badopt(const char *opt, char *type)
1296 fprintf(stderr, "%s: '--%s' only valid for %s\n",
1297 progname, opt, type);
1301 static int add_param(char *buf, char *key, char *val)
1303 int end = sizeof(((struct lustre_disk_data *)0)->ldd_params);
1304 int start = strlen(buf);
1308 keylen = strlen(key);
1309 if (start + 1 + keylen + strlen(val) >= end) {
1310 fprintf(stderr, "%s: params are too long-\n%s %s%s\n",
1311 progname, buf, key ? key : "", val);
1315 sprintf(buf + start, " %s%s", key ? key : "", val);
1319 /* from mount_lustre */
1320 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
1321 #define MAXNIDSTR 1024
1322 static char *convert_hostnames(char *s1)
1324 char *converted, *s2 = 0, *c, *end, sep;
1325 int left = MAXNIDSTR;
1328 converted = malloc(left);
1329 if (converted == NULL) {
1333 end = s1 + strlen(s1);
1335 while ((left > 0) && (s1 < end)) {
1336 s2 = strpbrk(s1, ",:");
1341 nid = libcfs_str2nid(s1);
1343 if (nid == LNET_NID_ANY) {
1344 fprintf(stderr, "%s: Can't parse NID '%s'\n",
1349 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
1350 strlen("127.0.0.1")) == 0) {
1351 fprintf(stderr, "%s: The NID '%s' resolves to the "
1352 "loopback address '%s'. Lustre requires a "
1353 "non-loopback address.\n",
1354 progname, s1, libcfs_nid2str(nid));
1359 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
1360 left = converted + MAXNIDSTR - c;
1366 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
1369 static struct option long_opt[] = {
1370 {"iam-dir", 0, 0, 'a'},
1371 {"backfstype", 1, 0, 'b'},
1372 {"stripe-count-hint", 1, 0, 'c'},
1373 {"comment", 1, 0, 'u'},
1374 {"configdev", 1, 0, 'C'},
1375 {"device-size", 1, 0, 'd'},
1376 {"dryrun", 0, 0, 'n'},
1377 {"erase-params", 0, 0, 'e'},
1378 {"failnode", 1, 0, 'f'},
1379 {"failover", 1, 0, 'f'},
1381 {"help", 0, 0, 'h'},
1382 {"index", 1, 0, 'i'},
1383 {"mkfsoptions", 1, 0, 'k'},
1384 {"mgsnode", 1, 0, 'm'},
1385 {"mgsnid", 1, 0, 'm'},
1387 {"fsname",1, 0, 'L'},
1388 {"noformat", 0, 0, 'n'},
1389 {"nomgs", 0, 0, 'N'},
1390 {"mountfsoptions", 1, 0, 'o'},
1392 {"param", 1, 0, 'p'},
1393 {"print", 0, 0, 'n'},
1394 {"quiet", 0, 0, 'q'},
1395 {"reformat", 0, 0, 'r'},
1396 {"servicenode", 1, 0, 's'},
1397 {"verbose", 0, 0, 'v'},
1398 {"writeconf", 0, 0, 'w'},
1399 {"upgrade_to_18", 0, 0, 'U'},
1400 {"network", 1, 0, 't'},
1403 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqrs:t:Uu:vw";
1406 int failnode_set = 0, servicenode_set = 0;
1408 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
1412 if (IS_MDT(&mop->mo_ldd))
1413 mop->mo_ldd.ldd_flags |= LDD_F_IAM_DIR;
1418 while (i < LDD_MT_LAST) {
1419 if (strcmp(optarg, mt_str(i)) == 0) {
1420 mop->mo_ldd.ldd_mount_type = i;
1428 if (IS_MDT(&mop->mo_ldd)) {
1429 int stripe_count = atol(optarg);
1430 if (stripe_count <= 0) {
1431 fprintf(stderr, "%s: bad stripe count "
1432 "%d\n", progname, stripe_count);
1435 mop->mo_stripe_count = stripe_count;
1437 badopt(long_opt[longidx].name, "MDT");
1441 case 'C': /* Configdev */
1443 printf("Configdev not implemented\n");
1446 mop->mo_device_sz = atol(optarg);
1449 mop->mo_ldd.ldd_params[0] = '\0';
1450 /* Must update the mgs logs */
1451 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1457 if ((opt == 'f' && servicenode_set)
1458 || (opt == 's' && failnode_set)) {
1459 fprintf(stderr, "%s: %s cannot use with --%s\n",
1460 progname, long_opt[longidx].name,
1461 opt == 'f' ? "servicenode" : "failnode");
1465 nids = convert_hostnames(optarg);
1468 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
1473 /* Must update the mgs logs */
1474 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1478 mop->mo_ldd.ldd_flags |= LDD_F_NO_PRIMNODE;
1479 servicenode_set = 1;
1485 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
1491 if (!(mop->mo_ldd.ldd_flags &
1492 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1493 LDD_F_WRITECONF))) {
1494 fprintf(stderr, "%s: cannot change the index of"
1495 " a registered target\n", progname);
1498 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
1499 mop->mo_ldd.ldd_svindex = atol(optarg);
1500 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
1502 badopt(long_opt[longidx].name, "MDT,OST");
1507 strscpy(mop->mo_mkfsopts, optarg,
1508 sizeof(mop->mo_mkfsopts));
1512 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
1513 (!(mop->mo_ldd.ldd_flags &
1514 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1515 LDD_F_WRITECONF)))) {
1516 fprintf(stderr, "%s: cannot change the name of"
1517 " a registered target\n", progname);
1520 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
1521 fprintf(stderr, "%s: filesystem name must be "
1522 "1-8 chars\n", progname);
1525 if ((tmp = strpbrk(optarg, "/:"))) {
1526 fprintf(stderr, "%s: char '%c' not allowed in "
1527 "filesystem name\n", progname, *tmp);
1530 strscpy(mop->mo_ldd.ldd_fsname, optarg,
1531 sizeof(mop->mo_ldd.ldd_fsname));
1535 char *nids = convert_hostnames(optarg);
1538 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
1543 mop->mo_mgs_failnodes++;
1547 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
1553 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
1556 *mountopts = optarg;
1559 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
1562 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg);
1565 /* Must update the mgs logs */
1566 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1572 mop->mo_flags |= MO_FORCEFORMAT;
1575 if (!IS_MDT(&mop->mo_ldd) && !IS_OST(&mop->mo_ldd)) {
1576 badopt(long_opt[longidx].name, "MDT,OST");
1583 rc = add_param(mop->mo_ldd.ldd_params,
1584 PARAM_NETWORK, optarg);
1587 /* Must update the mgs logs */
1588 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1591 strscpy(mop->mo_ldd.ldd_userdata, optarg,
1592 sizeof(mop->mo_ldd.ldd_userdata));
1598 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
1606 fprintf(stderr, "Unknown option '%c'\n", opt);
1612 /* Last arg is device */
1613 if (optind != argc - 1) {
1615 fprintf(stderr, "Bad argument: %s\n", argv[optind]);
1619 /* single argument: <device> */
1626 /* Search for opt in mntlist, returning true if found.
1628 static int in_mntlist(char *opt, char *mntlist)
1630 char *ml, *mlp, *item, *ctx = NULL;
1632 if (!(ml = strdup(mntlist))) {
1633 fprintf(stderr, "%s: out of memory\n", progname);
1637 while ((item = strtok_r(mlp, ",", &ctx))) {
1638 if (!strcmp(opt, item))
1643 return (item != NULL);
1646 /* Issue a message on stderr for every item in wanted_mountopts that is not
1647 * present in mountopts. The justwarn boolean toggles between error and
1648 * warning message. Return an error count.
1650 static int check_mountfsoptions(char *mountopts, char *wanted_mountopts,
1653 char *ml, *mlp, *item, *ctx = NULL;
1656 if (!(ml = strdup(wanted_mountopts))) {
1657 fprintf(stderr, "%s: out of memory\n", progname);
1661 while ((item = strtok_r(mlp, ",", &ctx))) {
1662 if (!in_mntlist(item, mountopts)) {
1663 fprintf(stderr, "%s: %s mount option `%s' is missing\n",
1664 progname, justwarn ? "Warning: default"
1665 : "Error: mandatory", item);
1674 /* Trim embedded white space, leading and trailing commas from string s.
1676 static void trim_mountfsoptions(char *s)
1682 memmove(p, p + 1, strlen(p + 1) + 1);
1689 memmove(&s[0], &s[1], strlen(&s[1]) + 1);
1691 p = s + strlen(s) - 1;
1692 while (p >= s && *p == ',')
1696 int main(int argc, char *const argv[])
1698 struct mkfs_opts mop;
1699 struct lustre_disk_data *ldd;
1700 char *mountopts = NULL;
1701 char always_mountopts[512] = "";
1702 char default_mountopts[512] = "";
1705 if ((progname = strrchr(argv[0], '/')) != NULL)
1710 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
1715 memset(&mop, 0, sizeof(mop));
1718 /* device is last arg */
1719 strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
1721 /* Are we using a loop device? */
1722 ret = is_block(mop.mo_device);
1726 mop.mo_flags |= MO_IS_LOOP;
1729 /* For tunefs, we must read in the old values before parsing any
1732 /* Check whether the disk has already been formatted by mkfs.lustre */
1733 ret = is_lustre_target(&mop);
1736 fprintf(stderr, "Device %s has not been formatted with "
1737 "mkfs.lustre\n", mop.mo_device);
1742 ret = read_local_files(&mop);
1745 fprintf(stderr, "Failed to read previous Lustre data from %s "
1746 "(%d)\n", mop.mo_device, ret);
1749 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
1750 mop.mo_mgs_failnodes++;
1753 print_ldd("Read previous values", &(mop.mo_ldd));
1756 ret = parse_opts(argc, argv, &mop, &mountopts);
1762 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
1764 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
1769 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
1771 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
1776 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
1777 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
1779 fprintf(stderr, "Can't find the target index, "
1780 "specify with --index\n");
1786 * Comment out these 2 checks temporarily, since for multi-MDSes
1787 * in single node only 1 mds node could have mgs service
1789 if (IS_MDT(ldd) && !IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1790 verrprint("No management node specified, adding MGS to this "
1792 ldd->ldd_flags |= LDD_F_SV_TYPE_MGS;
1794 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1797 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
1799 fprintf(stderr, "Must specify --mgsnode=\n");
1805 /* These are the permanent mount options (always included) */
1806 switch (ldd->ldd_mount_type) {
1808 case LDD_MT_LDISKFS:
1809 case LDD_MT_LDISKFS2:
1810 strscat(default_mountopts, ",errors=remount-ro",
1811 sizeof(default_mountopts));
1812 if (IS_MDT(ldd) || IS_MGS(ldd))
1813 strscat(always_mountopts, ",user_xattr",
1814 sizeof(always_mountopts));
1815 /* NB: Files created while extents are enabled can only be read
1816 * if mounted using the ext4 or ldiskfs filesystem type. */
1818 (ldd->ldd_mount_type == LDD_MT_LDISKFS ||
1819 ldd->ldd_mount_type == LDD_MT_LDISKFS2)) {
1820 strscat(default_mountopts, ",extents,mballoc",
1821 sizeof(default_mountopts));
1826 fprintf(stderr, "unknown fs type %d '%s'\n",
1827 ldd->ldd_mount_type,
1834 trim_mountfsoptions(mountopts);
1835 (void)check_mountfsoptions(mountopts, default_mountopts, 1);
1836 if (check_mountfsoptions(mountopts, always_mountopts, 0)) {
1840 sprintf(ldd->ldd_mount_opts, "%s", mountopts);
1843 if (ldd->ldd_mount_opts[0] == 0)
1844 /* use the defaults unless old opts exist */
1847 sprintf(ldd->ldd_mount_opts, "%s%s",
1848 always_mountopts, default_mountopts);
1849 trim_mountfsoptions(ldd->ldd_mount_opts);
1853 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
1854 ldd->ldd_fsname, ldd->ldd_svname);
1857 print_ldd("Permanent disk data", ldd);
1860 printf("exiting before disk write.\n");
1864 if (check_mtab_entry(mop.mo_device))
1867 /* Create the loopback file */
1868 if (mop.mo_flags & MO_IS_LOOP) {
1869 ret = access(mop.mo_device, F_OK);
1872 #ifndef TUNEFS /* mkfs.lustre */
1873 /* Reformat the loopback file */
1874 if (ret || (mop.mo_flags & MO_FORCEFORMAT))
1875 ret = loop_format(&mop);
1878 ret = loop_setup(&mop);
1881 fprintf(stderr, "Loop device setup for %s failed: %s\n",
1882 mop.mo_device, strerror(ret));
1887 #ifndef TUNEFS /* mkfs.lustre */
1888 /* Check whether the disk has already been formatted by mkfs.lustre */
1889 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
1890 ret = is_lustre_target(&mop);
1893 fprintf(stderr, "Device %s was previously formatted "
1894 "for lustre. Use --reformat to reformat it, "
1895 "or tunefs.lustre to modify.\n",
1901 /* Format the backing filesystem */
1902 ret = make_lustre_backfs(&mop);
1905 fprintf(stderr, "mkfs failed %d\n", ret);
1910 /* Write our config files */
1911 ret = write_local_files(&mop);
1914 fprintf(stderr, "failed to write local files\n");
1921 /* Fix any crazy return values from system() */
1922 if (ret && ((ret & 255) == 0))
1925 verrprint("%s: exiting with %d (%s)\n",
1926 progname, ret, strerror(ret));