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.
32 * Copyright (c) 2011, Whamcloud, Inc.
35 * This file is part of Lustre, http://www.lustre.org/
36 * Lustre is a trademark of Sun Microsystems, Inc.
38 * lustre/utils/mkfs_lustre.c
40 * Author: Nathan Rutman <nathan@clusterfs.com>
43 /* This source file is compiled into both mkfs.lustre and tunefs.lustre */
47 #endif /* HAVE_CONFIG_H */
59 #include <sys/types.h>
61 #include <sys/mount.h>
62 #include <sys/utsname.h>
70 /* libcfs.h is not really needed here, but on SLES10/PPC, fs.h includes idr.h
71 * which requires BITS_PER_LONG to be defined */
72 #include <libcfs/libcfs.h>
74 #include <linux/fs.h> /* for BLKGETSIZE64 */
76 #include <linux/version.h>
78 #include <lustre_disk.h>
79 #include <lustre_param.h>
80 #include <lnet/lnetctl.h>
81 #include <lustre_ver.h>
82 #include "mount_utils.h"
88 #define MAX_LOOP_DEVICES 16
89 #define L_BLOCK_SIZE 4096
90 #define INDEX_UNASSIGNED 0xFFFF
91 #define MO_IS_LOOP 0x01
92 #define MO_FORCEFORMAT 0x02
94 /* used to describe the options to format the lustre disk, not persistent */
96 struct lustre_disk_data mo_ldd; /* to be written in MOUNT_DATA_FILE */
97 char mo_device[128]; /* disk device name */
98 char mo_loopdev[128]; /* in case a loop dev is needed */
99 char mo_mkfsopts[512]; /* options to the backing-store mkfs */
100 __u64 mo_device_sz; /* in KB */
103 int mo_mgs_failnodes;
108 static int print_only = 0;
109 static int failover = 0;
110 static int upgrade_to_18 = 0;
112 void usage(FILE *out)
114 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
115 fprintf(out, "usage: %s <target types> [options] <device>\n", progname);
117 "\t<device>:block device or file (e.g /dev/sda or /tmp/ost1)\n"
119 "\t\t--ost: object storage, mutually exclusive with mdt,mgs\n"
120 "\t\t--mdt: metadata storage, mutually exclusive with ost\n"
121 "\t\t--mgs: configuration management service - one per site\n"
122 "\toptions (in order of popularity):\n"
123 "\t\t--mgsnode=<nid>[,<...>] : NID(s) of a remote mgs node\n"
124 "\t\t\trequired for all targets other than the mgs node\n"
125 "\t\t--fsname=<filesystem_name> : default is 'lustre'\n"
126 "\t\t--failnode=<nid>[,<...>] : NID(s) of a failover partner\n"
127 "\t\t\tcannot be used with --servicenode\n"
128 "\t\t--servicenode=<nid>[,<...>] : NID(s) of all service partners\n"
129 "\t\t\ttreat all nodes as equal service node, cannot be used with --failnode\n"
130 "\t\t--param <key>=<value> : set a permanent parameter\n"
131 "\t\t\te.g. --param sys.timeout=40\n"
132 "\t\t\t --param lov.stripesize=2M\n"
133 "\t\t--index=#N : target index (i.e. ost index within lov)\n"
134 "\t\t--comment=<user comment>: arbitrary string (%d bytes)\n"
135 "\t\t--mountfsoptions=<opts> : permanent mount options\n"
136 "\t\t--network=<net>[,<...>] : restrict OST/MDT to network(s)\n"
138 "\t\t--backfstype=<fstype> : backing fs type (ext3, ldiskfs)\n"
139 "\t\t--device-size=#N(KB) : device size for loop devices\n"
140 "\t\t--mkfsoptions=<opts> : format options\n"
141 "\t\t--reformat: overwrite an existing disk\n"
142 "\t\t--stripe-count-hint=#N : for optimizing MDT inode size\n"
143 "\t\t--iam-dir: use IAM directory format, not ext3 compatible\n"
145 "\t\t--erase-params : erase all old parameter settings\n"
146 "\t\t--nomgs: turn off MGS service on this MDT\n"
147 "\t\t--writeconf: erase all config logs for this fs.\n"
149 "\t\t--dryrun: just report what we would do; "
150 "don't write to disk\n"
151 "\t\t--verbose : e.g. show mkfs progress\n"
153 (int)sizeof(((struct lustre_disk_data *)0)->ldd_userdata));
157 #define vprint if (verbose > 0) printf
158 #define verrprint if (verbose >= 0) printf
160 /*================ utility functions =====================*/
162 char *strscat(char *dst, char *src, int buflen) {
164 if (strlen(dst) + strlen(src) >= buflen) {
165 fprintf(stderr, "string buffer overflow (max %d): '%s' + '%s'"
166 "\n", buflen, dst, src);
169 return strcat(dst, src);
173 char *strscpy(char *dst, char *src, int buflen) {
175 return strscat(dst, src, buflen);
179 dev_major (unsigned long long int __dev)
181 return ((__dev >> 8) & 0xfff) | ((unsigned int) (__dev >> 32) & ~0xfff);
185 dev_minor (unsigned long long int __dev)
187 return (__dev & 0xff) | ((unsigned int) (__dev >> 12) & ~0xff);
192 static int version = 0;
196 char release[4] = "";
198 fd = open("/proc/sys/kernel/osrelease", O_RDONLY);
200 fprintf(stderr, "%s: Warning: Can't resolve kernel "
201 "version, assuming 2.6\n", progname);
203 if (read(fd, release, 4) < 0) {
204 fprintf(stderr, "reading from /proc/sys/kernel"
205 "/osrelease: %s\n", strerror(errno));
211 if (strncmp(release, "2.4.", 4) == 0)
219 static int check_mtab_entry(char *spec)
224 fp = setmntent(MOUNTED, "r");
228 while ((mnt = getmntent(fp)) != NULL) {
229 if (strcmp(mnt->mnt_fsname, spec) == 0) {
231 fprintf(stderr, "%s: according to %s %s is "
232 "already mounted on %s\n",
233 progname, MOUNTED, spec, mnt->mnt_dir);
242 /*============ disk dev functions ===================*/
244 /* Setup a file in the first unused loop_device */
245 int loop_setup(struct mkfs_opts *mop)
251 /* Figure out the loop device names */
252 if (!access("/dev/loop0", F_OK | R_OK)) {
253 strcpy(loop_base, "/dev/loop\0");
254 } else if (!access("/dev/loop/0", F_OK | R_OK)) {
255 strcpy(loop_base, "/dev/loop/\0");
257 fprintf(stderr, "%s: can't access loop devices\n", progname);
261 /* Find unused loop device */
262 for (i = 0; i < MAX_LOOP_DEVICES; i++) {
264 int cmdsz = sizeof(cmd);
266 sprintf(l_device, "%s%d", loop_base, i);
267 if (access(l_device, F_OK | R_OK))
269 snprintf(cmd, cmdsz, "losetup %s > /dev/null 2>&1", l_device);
272 /* losetup gets 1 (ret=256) for non-set-up device */
274 /* Set up a loopback device to our file */
275 snprintf(cmd, cmdsz, "losetup %s %s", l_device,
277 ret = run_command(cmd, cmdsz);
279 /* someone else picked up this loop device
283 fprintf(stderr, "%s: error %d on losetup: %s\n",
284 progname, ret, strerror(ret));
287 strscpy(mop->mo_loopdev, l_device,
288 sizeof(mop->mo_loopdev));
293 fprintf(stderr, "%s: out of loop devices!\n", progname);
297 int loop_cleanup(struct mkfs_opts *mop)
301 if ((mop->mo_flags & MO_IS_LOOP) && *mop->mo_loopdev) {
302 sprintf(cmd, "losetup -d %s", mop->mo_loopdev);
303 ret = run_command(cmd, sizeof(cmd));
308 /* Determine if a device is a block device (as opposed to a file) */
309 int is_block(char* devname)
314 ret = access(devname, F_OK);
317 ret = stat(devname, &st);
319 fprintf(stderr, "%s: cannot stat %s\n", progname, devname);
322 return S_ISBLK(st.st_mode);
325 __u64 get_device_size(char* device)
330 fd = open(device, O_RDONLY);
332 fprintf(stderr, "%s: cannot open %s: %s\n",
333 progname, device, strerror(errno));
338 /* size in bytes. bz5831 */
339 ret = ioctl(fd, BLKGETSIZE64, (void*)&size);
344 ret = ioctl(fd, BLKGETSIZE, (void*)&lsize);
345 size = (__u64)lsize * 512;
350 fprintf(stderr, "%s: size ioctl failed: %s\n",
351 progname, strerror(errno));
355 vprint("device size = "LPU64"MB\n", size >> 20);
356 /* return value in KB */
360 int loop_format(struct mkfs_opts *mop)
364 if (mop->mo_device_sz == 0) {
366 fprintf(stderr, "loop device requires a --device-size= "
371 ret = creat(mop->mo_device, S_IRUSR|S_IWUSR);
374 fprintf(stderr, "%s: Unable to create backing store: %d\n",
380 ret = truncate(mop->mo_device, mop->mo_device_sz * 1024);
383 fprintf(stderr, "%s: Unable to truncate backing store: %d\n",
390 /* Display the need for the latest e2fsprogs to be installed. make_backfs
391 * indicates if the caller is make_lustre_backfs() or not. */
392 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
394 static int msg_displayed;
397 fprintf(stderr, "WARNING: %s does not support %s "
398 "feature.\n\n", E2FSPROGS, feature);
404 fprintf(stderr, "WARNING: The %s package currently installed on "
405 "your system does not support \"%s\" feature.\n",
407 #if !(HAVE_LDISKFSPROGS)
408 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
409 "http://downloads.whamcloud.com/public/e2fsprogs/latest/\n"
410 "to enable this feature.\n");
413 fprintf(stderr, "Feature will not be enabled until %s"
414 "is updated and '%s -O %s %%{device}' "
415 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
418 /* Check whether the file exists in the device */
419 static int file_in_dev(char *file_name, char *dev_name)
422 char debugfs_cmd[256];
423 unsigned int inode_num;
426 /* Construct debugfs command line. */
427 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
428 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
429 DEBUGFS, file_name, dev_name);
431 fp = popen(debugfs_cmd, "r");
433 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
437 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
441 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd), fp);
444 fprintf(stderr, "%s", debugfs_cmd);
445 if (strstr(debugfs_cmd, "unsupported feature")) {
446 disp_old_e2fsprogs_msg("an unknown", 0);
455 /* Check whether the device has already been used with lustre */
456 static int is_lustre_target(struct mkfs_opts *mop)
460 vprint("checking for existing Lustre data: ");
462 if ((rc = file_in_dev(MOUNT_DATA_FILE, mop->mo_device))) {
464 (rc == 1) ? MOUNT_DATA_FILE : "extents");
465 /* in the -1 case, 'extents' means this really IS a lustre
470 if ((rc = file_in_dev(LAST_RCVD, mop->mo_device))) {
471 vprint("found %s\n", LAST_RCVD);
475 vprint("not found\n");
476 return 0; /* The device is not a lustre target. */
479 /* Check if a certain feature is supported by e2fsprogs.
480 * Firstly we try to use "debugfs supported_features" command to check if
481 * the feature is supported. If this fails we try to set this feature with
482 * mke2fs to check for its support. */
483 static int is_e2fsprogs_feature_supp(const char *feature)
485 static char supp_features[4096] = "";
488 char imgname[] = "/tmp/test-img-XXXXXX";
492 if (supp_features[0] == '\0') {
493 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
496 /* Using popen() instead of run_command() since debugfs does
497 * not return proper error code if command is not supported */
498 fp = popen(cmd, "r");
500 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
503 ret = fread(supp_features, 1, sizeof(supp_features), fp);
506 if (ret > 0 && strstr(supp_features,
507 strncmp(feature, "-O ", 3) ? feature : feature+3))
510 if ((fd = mkstemp(imgname)) < 0)
515 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
516 MKE2FS, feature, imgname);
517 /* run_command() displays the output of mke2fs when it fails for
518 * some feature, so use system() directly */
526 * append_unique: append @key or @key=@val pair to @buf only if @key does not
528 * @buf: buffer to hold @key or @key=@val
529 * @prefix: prefix string before @key
531 * @val: value string if it's a @key=@val pair
533 static void append_unique(char *buf, char *prefix, char *key, char *val,
542 anchor = end = strstr(buf, key);
543 /* try to find exact match string in @buf */
544 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
547 if (anchor == NULL || strlen(key) != len ||
548 strncmp(anchor, key, len) != 0) {
550 strscat(buf, prefix, maxbuflen);
552 strscat(buf, key, maxbuflen);
554 strscat(buf, "=", maxbuflen);
555 strscat(buf, val, maxbuflen);
560 static void enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
561 size_t maxbuflen, int user_spec)
563 if (IS_OST(&mop->mo_ldd)) {
564 append_unique(anchor, user_spec ? "," : " -O ",
565 "extents", NULL, sizeof(mop->mo_mkfsopts));
566 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
567 } else if (IS_MDT(&mop->mo_ldd)) {
568 append_unique(anchor, user_spec ? "," : " -O ",
569 "dirdata", NULL, maxbuflen);
570 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
572 append_unique(anchor, user_spec ? "," : " -O ",
573 "uninit_bg", NULL, maxbuflen);
576 /* Multiple mount protection enabled only if failover node specified */
578 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
579 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
581 disp_old_e2fsprogs_msg("mmp", 1);
584 /* Allow more than 65000 subdirectories */
585 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
586 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
588 #ifdef HAVE_EXT4_LDISKFS
589 /* The following options are only valid for ext4-based ldiskfs.
590 * If --backfstype=ext3 is specified, do not enable them. */
591 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
594 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
595 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
596 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
598 /* Enable large block addresses if the LUN is over 2^32 blocks. */
599 if (mop->mo_device_sz / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
600 is_e2fsprogs_feature_supp("-O 64bit") == 0)
601 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
603 /* Cluster inode/block bitmaps and inode table for more efficient IO.
604 * Align the flex groups on a 1MB boundary for better performance. */
605 /* This -O feature needs to go last, since it adds the "-G" option. */
606 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
609 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
611 if (IS_OST(&mop->mo_ldd)) {
612 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
613 (1 << 20) / L_BLOCK_SIZE);
614 strscat(anchor, tmp_buf, maxbuflen);
617 /* Don't add any more "-O" options here, see last comment above */
622 * moveopts_to_end: find the option string, move remaining strings to
623 * where option string starts, and append the option
625 * @start: where the option string starts before the move
626 * RETURN: where the option string starts after the move
628 static char *moveopts_to_end(char *start)
633 /* skip whitespace before options */
638 /* find end of option characters */
639 while (*end != ' ' && *end != '\0')
643 strncpy(save, start, end - start);
644 save[end - start] = '\0';
646 /* move remaining options up front */
648 memmove(start, end, strlen(end));
649 *(start + strlen(end)) = '\0';
651 /* append the specified options */
652 if (*(start + strlen(start) - 1) != ' ')
654 idx = start + strlen(start);
660 /* Build fs according to type */
661 int make_lustre_backfs(struct mkfs_opts *mop)
663 __u64 device_sz = mop->mo_device_sz, block_count = 0;
664 char mkfs_cmd[PATH_MAX];
668 int ret = 0, ext_opts = 0;
671 if (!(mop->mo_flags & MO_IS_LOOP)) {
672 mop->mo_device_sz = get_device_size(mop->mo_device);
674 if (mop->mo_device_sz == 0)
677 /* Compare to real size */
678 if (device_sz == 0 || device_sz > mop->mo_device_sz)
679 device_sz = mop->mo_device_sz;
681 mop->mo_device_sz = device_sz;
684 if (mop->mo_device_sz != 0) {
685 if (mop->mo_device_sz < 8096){
686 fprintf(stderr, "%s: size of filesystem must be larger "
687 "than 8MB, but is set to %lldKB\n",
688 progname, (long long)mop->mo_device_sz);
691 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
692 /* If the LUN size is just over 2^32 blocks, limit the
693 * filesystem size to 2^32-1 blocks to avoid problems with
694 * ldiskfs/mkfs not handling this size. Bug 22906 */
695 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
696 block_count = 0xffffffffULL;
699 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
700 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
701 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
704 /* Journal size in MB */
705 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
706 /* Choose our own default journal size */
707 long journal_sz = 0, max_sz;
708 if (device_sz > 1024 * 1024) /* 1GB */
709 journal_sz = (device_sz / 102400) * 4;
710 /* cap journal size at 1GB */
711 if (journal_sz > 1024L)
714 max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
715 if (journal_sz > max_sz)
718 sprintf(buf, " -J size=%ld", journal_sz);
719 strscat(mop->mo_mkfsopts, buf,
720 sizeof(mop->mo_mkfsopts));
724 /* Inode size (for extended attributes). The LOV EA size is
725 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
726 * and we want some margin above that for ACLs, other EAs... */
727 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
728 if (IS_MDT(&mop->mo_ldd)) {
729 if (mop->mo_stripe_count > 72)
730 inode_size = 512; /* bz 7241 */
731 /* see also "-i" below for EA blocks */
732 else if (mop->mo_stripe_count > 32)
734 else if (mop->mo_stripe_count > 10)
738 } else if (IS_OST(&mop->mo_ldd)) {
739 /* We store MDS FID and OST objid in EA on OST
740 * we need to make inode bigger as well. */
744 if (inode_size > 0) {
745 sprintf(buf, " -I %ld", inode_size);
746 strscat(mop->mo_mkfsopts, buf,
747 sizeof(mop->mo_mkfsopts));
751 /* Bytes_per_inode: disk size / num inodes */
752 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
753 strstr(mop->mo_mkfsopts, "-N") == NULL) {
754 long bytes_per_inode = 0;
756 /* Allocate more inodes on MDT devices. There is
757 * no data stored on the MDT, and very little extra
758 * metadata beyond the inode. It could go down as
759 * low as 1024 bytes, but this is conservative.
760 * Account for external EA blocks for wide striping. */
761 if (IS_MDT(&mop->mo_ldd)) {
762 bytes_per_inode = inode_size + 1536;
764 if (mop->mo_stripe_count > 72) {
765 int extra = mop->mo_stripe_count * 24;
766 extra = ((extra - 1) | 4095) + 1;
767 bytes_per_inode += extra;
771 /* Allocate fewer inodes on large OST devices. Most
772 * filesystems can be much more aggressive than even
773 * this, but it is impossible to know in advance. */
774 if (IS_OST(&mop->mo_ldd)) {
775 /* OST > 16TB assume average file size 1MB */
776 if (device_sz > (16ULL << 30))
777 bytes_per_inode = 1024 * 1024;
778 /* OST > 4TB assume average file size 512kB */
779 else if (device_sz > (4ULL << 30))
780 bytes_per_inode = 512 * 1024;
781 /* OST > 1TB assume average file size 256kB */
782 else if (device_sz > (1ULL << 30))
783 bytes_per_inode = 256 * 1024;
784 /* OST > 10GB assume average file size 64kB,
785 * plus a bit so that inodes will fit into a
786 * 256x flex_bg without overflowing */
787 else if (device_sz > (10ULL << 20))
788 bytes_per_inode = 69905;
791 if (bytes_per_inode > 0) {
792 sprintf(buf, " -i %ld", bytes_per_inode);
793 strscat(mop->mo_mkfsopts, buf,
794 sizeof(mop->mo_mkfsopts));
799 strscat(mop->mo_mkfsopts, " -q",
800 sizeof(mop->mo_mkfsopts));
803 /* start handle -O mkfs options */
804 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
805 if (strstr(start + 2, "-O") != NULL) {
807 "%s: don't specify multiple -O options\n",
811 start = moveopts_to_end(start);
812 maxbuflen = sizeof(mop->mo_mkfsopts) -
813 (start - mop->mo_mkfsopts) - strlen(start);
814 enable_default_ext4_features(mop, start, maxbuflen, 1);
816 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
817 maxbuflen = sizeof(mop->mo_mkfsopts) -
818 strlen(mop->mo_mkfsopts);
819 enable_default_ext4_features(mop, start, maxbuflen, 0);
821 /* end handle -O mkfs options */
823 /* start handle -E mkfs options */
824 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
825 if (strstr(start + 2, "-E") != NULL) {
827 "%s: don't specify multiple -E options\n",
831 start = moveopts_to_end(start);
832 maxbuflen = sizeof(mop->mo_mkfsopts) -
833 (start - mop->mo_mkfsopts) - strlen(start);
836 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
837 maxbuflen = sizeof(mop->mo_mkfsopts) -
838 strlen(mop->mo_mkfsopts);
841 /* In order to align the filesystem metadata on 1MB boundaries,
842 * give a resize value that will reserve a power-of-two group
843 * descriptor blocks, but leave one block for the superblock.
844 * Only useful for filesystems with < 2^32 blocks due to resize
846 if (IS_OST(&mop->mo_ldd) && mop->mo_device_sz > 100 * 1024 &&
847 mop->mo_device_sz * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
848 unsigned group_blocks = L_BLOCK_SIZE * 8;
849 unsigned desc_per_block = L_BLOCK_SIZE / 32;
850 unsigned resize_blks;
852 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
853 snprintf(buf, sizeof(buf), "%u", resize_blks);
854 append_unique(start, ext_opts ? "," : " -E ",
855 "resize", buf, maxbuflen);
859 /* Avoid zeroing out the full journal - speeds up mkfs */
860 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
861 append_unique(start, ext_opts ? "," : " -E ",
862 "lazy_journal_init", NULL, maxbuflen);
863 /* end handle -E mkfs options */
865 /* Allow reformat of full devices (as opposed to
866 partitions.) We already checked for mounted dev. */
867 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
869 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
870 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
871 mop->mo_ldd.ldd_svname);
873 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
874 progname, mop->mo_ldd.ldd_mount_type,
875 MT_STR(&mop->mo_ldd));
879 /* For loop device format the dev, not the filename */
880 dev = mop->mo_device;
881 if (mop->mo_flags & MO_IS_LOOP)
882 dev = mop->mo_loopdev;
884 vprint("formatting backing filesystem %s on %s\n",
885 MT_STR(&mop->mo_ldd), dev);
886 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
887 vprint("\t4k blocks "LPU64"\n", block_count);
888 vprint("\toptions %s\n", mop->mo_mkfsopts);
890 /* mkfs_cmd's trailing space is important! */
891 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
892 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
893 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
894 if (block_count != 0) {
895 sprintf(buf, " "LPU64, block_count);
896 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
899 vprint("mkfs_cmd = %s\n", mkfs_cmd);
900 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
903 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
908 /* ==================== Lustre config functions =============*/
910 void print_ldd(char *str, struct lustre_disk_data *ldd)
912 printf("\n %s:\n", str);
913 printf("Target: %s\n", ldd->ldd_svname);
914 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
915 printf("Index: unassigned\n");
917 printf("Index: %d\n", ldd->ldd_svindex);
918 if (ldd->ldd_uuid[0])
919 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
920 printf("Lustre FS: %s\n", ldd->ldd_fsname);
921 printf("Mount type: %s\n", MT_STR(ldd));
922 printf("Flags: %#x\n", ldd->ldd_flags);
923 printf(" (%s%s%s%s%s%s%s%s%s%s)\n",
924 IS_MDT(ldd) ? "MDT ":"",
925 IS_OST(ldd) ? "OST ":"",
926 IS_MGS(ldd) ? "MGS ":"",
927 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
928 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
929 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
930 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
931 ldd->ldd_flags & LDD_F_IAM_DIR ? "IAM_dir_format ":"",
932 ldd->ldd_flags & LDD_F_NO_PRIMNODE? "no_primnode ":"",
933 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
934 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
935 printf("Parameters:%s\n", ldd->ldd_params);
936 if (ldd->ldd_userdata[0])
937 printf("Comment: %s\n", ldd->ldd_userdata);
941 static int touch_file(char *filename)
945 if (filename == NULL) {
949 fd = open(filename, O_CREAT | O_TRUNC, 0600);
958 /* keep it less than LL_FID_NAMELEN */
959 #define DUMMY_FILE_NAME_LEN 25
960 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
962 /* Need to add these many entries to this directory to make HTREE dir. */
963 #define MIN_ENTRIES_REQ_FOR_HTREE ((L_BLOCK_SIZE / EXT3_DIRENT_SIZE))
965 static int add_dummy_files(char *dir)
967 char fpname[PATH_MAX];
971 for (i = 0; i < MIN_ENTRIES_REQ_FOR_HTREE; i++) {
972 snprintf(fpname, PATH_MAX, "%s/%0*d", dir,
973 DUMMY_FILE_NAME_LEN, i);
975 rc = touch_file(fpname);
976 if (rc && rc != -EEXIST) {
978 "%s: Can't create dummy file %s: %s\n",
979 progname, fpname , strerror(errno));
986 static int __l_mkdir(char * filepnm, int mode , struct mkfs_opts *mop)
990 ret = mkdir(filepnm, mode);
991 if (ret && ret != -EEXIST)
994 /* IAM mode supports ext3 directories of HTREE type only. So add dummy
995 * entries to new directory to create htree type of container for
997 if (mop->mo_ldd.ldd_flags & LDD_F_IAM_DIR)
998 return add_dummy_files(filepnm);
1002 /* Write the server config files */
1003 int write_local_files(struct mkfs_opts *mop)
1005 char mntpt[] = "/tmp/mntXXXXXX";
1012 /* Mount this device temporarily in order to write these files */
1013 if (!mkdtemp(mntpt)) {
1014 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
1015 progname, mntpt, strerror(errno));
1019 dev = mop->mo_device;
1020 if (mop->mo_flags & MO_IS_LOOP)
1021 dev = mop->mo_loopdev;
1023 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
1024 mop->mo_ldd.ldd_mount_opts);
1026 fprintf(stderr, "%s: Unable to mount %s: %s\n",
1027 progname, dev, strerror(errno));
1029 if (errno == ENODEV) {
1030 fprintf(stderr, "Is the %s module available?\n",
1031 MT_STR(&mop->mo_ldd));
1036 /* Set up initial directories */
1037 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
1038 ret = __l_mkdir(filepnm, 0777, mop);
1039 if ((ret != 0) && (errno != EEXIST)) {
1040 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
1041 progname, filepnm, strerror(errno));
1043 } else if (errno == EEXIST) {
1047 /* Save the persistent mount data into a file. Lustre must pre-read
1048 this file to get the real mount options. */
1049 vprint("Writing %s\n", MOUNT_DATA_FILE);
1050 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
1051 filep = fopen(filepnm, "w");
1053 fprintf(stderr, "%s: Unable to create %s file: %s\n",
1054 progname, filepnm, strerror(errno));
1057 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
1058 if (num < 1 && ferror(filep)) {
1059 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
1060 progname, filepnm, strerror(errno));
1066 /* Check for upgrade */
1067 if ((mop->mo_ldd.ldd_flags & (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS))
1068 == (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS)) {
1071 int cmdsz = sizeof(cmd);
1072 vprint("Copying old logs\n");
1074 /* Copy the old client log to fsname-client */
1075 sprintf(filepnm, "%s/%s/%s-client",
1076 mntpt, MOUNT_CONFIGS_DIR, mop->mo_ldd.ldd_fsname);
1077 snprintf(cmd, cmdsz, "cp %s/%s/client %s", mntpt, MDT_LOGS_DIR,
1079 ret = run_command(cmd, cmdsz);
1081 fprintf(stderr, "%s: Can't copy 1.4 config %s/client "
1082 "(%d)\n", progname, MDT_LOGS_DIR, ret);
1083 fprintf(stderr, "mount -t ldiskfs %s somewhere, "
1084 "find the client log for fs %s and "
1085 "copy it manually into %s/%s-client, "
1088 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
1089 mop->mo_ldd.ldd_fsname);
1093 /* We need to use the old mdt log because otherwise mdt won't
1094 have complete lov if old clients connect before all
1096 /* Copy the old mdt log to fsname-MDT0000 (get old
1097 name from mdt_UUID) */
1099 strscpy(filepnm, (char *)mop->mo_ldd.ldd_uuid, sizeof(filepnm));
1100 term = strstr(filepnm, "_UUID");
1103 snprintf(cmd, cmdsz, "cp %s/%s/%s %s/%s/%s",
1104 mntpt, MDT_LOGS_DIR, filepnm,
1105 mntpt, MOUNT_CONFIGS_DIR,
1106 mop->mo_ldd.ldd_svname);
1107 ret = run_command(cmd, cmdsz);
1110 fprintf(stderr, "%s: Can't copy 1.4 config %s/%s "
1111 "(%d)\n", progname, MDT_LOGS_DIR, filepnm, ret);
1112 fprintf(stderr, "mount -t ext3 %s somewhere, "
1113 "find the MDT log for fs %s and "
1114 "copy it manually into %s/%s, "
1117 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
1118 mop->mo_ldd.ldd_svname);
1123 /* end COMPAT_146 */
1132 int read_local_files(struct mkfs_opts *mop)
1134 char tmpdir[] = "/tmp/dirXXXXXX";
1140 int cmdsz = sizeof(cmd);
1142 /* Make a temporary directory to hold Lustre data files. */
1143 if (!mkdtemp(tmpdir)) {
1144 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
1145 progname, tmpdir, strerror(errno));
1149 dev = mop->mo_device;
1151 /* TODO: it's worth observing the get_mountdata() function that is
1152 in mount_utils.c for getting the mountdata out of the
1155 /* Construct debugfs command line. */
1156 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
1157 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
1159 ret = run_command(cmd, cmdsz);
1161 verrprint("%s: Unable to dump %s dir (%d)\n",
1162 progname, MOUNT_CONFIGS_DIR, ret);
1164 sprintf(filepnm, "%s/mountdata", tmpdir);
1165 filep = fopen(filepnm, "r");
1168 vprint("Reading %s\n", MOUNT_DATA_FILE);
1169 num_read = fread(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
1170 if (num_read < 1 && ferror(filep)) {
1171 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
1172 progname, filepnm, strerror(errno));
1177 /* Try to read pre-1.6 config from last_rcvd */
1178 struct lr_server_data lsd;
1179 verrprint("%s: Unable to read %d.%d config %s.\n",
1180 progname, LUSTRE_MAJOR, LUSTRE_MINOR, filepnm);
1182 verrprint("Trying 1.4 config from last_rcvd\n");
1183 sprintf(filepnm, "%s/%s", tmpdir, LAST_RCVD);
1185 /* Construct debugfs command line. */
1186 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s' %s",
1187 DEBUGFS, LAST_RCVD, filepnm, dev);
1189 ret = run_command(cmd, cmdsz);
1191 fprintf(stderr, "%s: Unable to dump %s file (%d)\n",
1192 progname, LAST_RCVD, ret);
1196 filep = fopen(filepnm, "r");
1198 fprintf(stderr, "%s: Unable to open %s: %s\n",
1199 progname, filepnm, strerror(errno));
1201 verrprint("Contents of %s:\n", tmpdir);
1203 snprintf(cmd, cmdsz, "ls -l %s/", tmpdir);
1204 run_command(cmd, cmdsz);
1205 verrprint("Contents of disk:\n");
1206 snprintf(cmd, cmdsz, "%s -c -R 'ls -l /' %s",
1208 run_command(cmd, cmdsz);
1212 vprint("Reading %s\n", LAST_RCVD);
1213 ret = fread(&lsd, 1, sizeof(lsd), filep);
1214 if (ret < sizeof(lsd)) {
1215 fprintf(stderr, "%s: Short read (%d of %d)\n",
1216 progname, ret, (int)sizeof(lsd));
1217 ret = ferror(filep);
1221 vprint("Feature compat=%x, incompat=%x\n",
1222 lsd.lsd_feature_compat, lsd.lsd_feature_incompat);
1224 if ((lsd.lsd_feature_compat & OBD_COMPAT_OST) ||
1225 (lsd.lsd_feature_incompat & OBD_INCOMPAT_OST)) {
1226 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
1227 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
1228 } else if ((lsd.lsd_feature_compat & OBD_COMPAT_MDT) ||
1229 (lsd.lsd_feature_incompat & OBD_INCOMPAT_MDT)) {
1230 /* We must co-locate so mgs can see old logs.
1231 If user doesn't want this, they can copy the old
1232 logs manually and re-tunefs. */
1233 mop->mo_ldd.ldd_flags =
1234 LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_MGS;
1235 mop->mo_ldd.ldd_svindex = lsd.lsd_mdt_index;
1237 /* If neither is set, we're pre-1.4.6, make a guess. */
1238 /* Construct debugfs command line. */
1239 snprintf(cmd, cmdsz, "%s -c -R 'rdump /%s %s' %s",
1240 DEBUGFS, MDT_LOGS_DIR, tmpdir, dev);
1241 run_command(cmd, cmdsz);
1243 sprintf(filepnm, "%s/%s", tmpdir, MDT_LOGS_DIR);
1244 if (lsd.lsd_ost_index > 0) {
1245 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
1246 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
1248 /* If there's a LOGS dir, it's an MDT */
1249 if ((ret = access(filepnm, F_OK)) == 0) {
1250 mop->mo_ldd.ldd_flags =
1253 /* Old MDT's are always index 0
1255 mop->mo_ldd.ldd_svindex = 0;
1257 /* The index may not be correct */
1258 mop->mo_ldd.ldd_flags =
1259 LDD_F_SV_TYPE_OST | LDD_F_NEED_INDEX;
1260 verrprint("OST with unknown index\n");
1266 memcpy(mop->mo_ldd.ldd_uuid, lsd.lsd_uuid,
1267 sizeof(mop->mo_ldd.ldd_uuid));
1268 mop->mo_ldd.ldd_flags |= LDD_F_UPGRADE14;
1270 /* end COMPAT_146 */
1275 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
1276 run_command(cmd, cmdsz);
1278 verrprint("Failed to read old data (%d)\n", ret);
1283 void set_defaults(struct mkfs_opts *mop)
1285 mop->mo_ldd.ldd_magic = LDD_MAGIC;
1286 mop->mo_ldd.ldd_config_ver = 1;
1287 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
1288 mop->mo_mgs_failnodes = 0;
1289 strcpy(mop->mo_ldd.ldd_fsname, "lustre");
1290 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
1292 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
1293 mop->mo_stripe_count = 1;
1296 static inline void badopt(const char *opt, char *type)
1298 fprintf(stderr, "%s: '--%s' only valid for %s\n",
1299 progname, opt, type);
1303 static int add_param(char *buf, char *key, char *val)
1305 int end = sizeof(((struct lustre_disk_data *)0)->ldd_params);
1306 int start = strlen(buf);
1310 keylen = strlen(key);
1311 if (start + 1 + keylen + strlen(val) >= end) {
1312 fprintf(stderr, "%s: params are too long-\n%s %s%s\n",
1313 progname, buf, key ? key : "", val);
1317 sprintf(buf + start, " %s%s", key ? key : "", val);
1321 /* from mount_lustre */
1322 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
1323 #define MAXNIDSTR 1024
1324 static char *convert_hostnames(char *s1)
1326 char *converted, *s2 = 0, *c, *end, sep;
1327 int left = MAXNIDSTR;
1330 converted = malloc(left);
1331 if (converted == NULL) {
1335 end = s1 + strlen(s1);
1337 while ((left > 0) && (s1 < end)) {
1338 s2 = strpbrk(s1, ",:");
1343 nid = libcfs_str2nid(s1);
1345 if (nid == LNET_NID_ANY) {
1346 fprintf(stderr, "%s: Can't parse NID '%s'\n",
1351 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
1352 strlen("127.0.0.1")) == 0) {
1353 fprintf(stderr, "%s: The NID '%s' resolves to the "
1354 "loopback address '%s'. Lustre requires a "
1355 "non-loopback address.\n",
1356 progname, s1, libcfs_nid2str(nid));
1361 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
1362 left = converted + MAXNIDSTR - c;
1368 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
1371 static struct option long_opt[] = {
1372 {"iam-dir", 0, 0, 'a'},
1373 {"backfstype", 1, 0, 'b'},
1374 {"stripe-count-hint", 1, 0, 'c'},
1375 {"comment", 1, 0, 'u'},
1376 {"configdev", 1, 0, 'C'},
1377 {"device-size", 1, 0, 'd'},
1378 {"dryrun", 0, 0, 'n'},
1379 {"erase-params", 0, 0, 'e'},
1380 {"failnode", 1, 0, 'f'},
1381 {"failover", 1, 0, 'f'},
1383 {"help", 0, 0, 'h'},
1384 {"index", 1, 0, 'i'},
1385 {"mkfsoptions", 1, 0, 'k'},
1386 {"mgsnode", 1, 0, 'm'},
1387 {"mgsnid", 1, 0, 'm'},
1389 {"fsname",1, 0, 'L'},
1390 {"noformat", 0, 0, 'n'},
1391 {"nomgs", 0, 0, 'N'},
1392 {"mountfsoptions", 1, 0, 'o'},
1394 {"param", 1, 0, 'p'},
1395 {"print", 0, 0, 'n'},
1396 {"quiet", 0, 0, 'q'},
1397 {"reformat", 0, 0, 'r'},
1398 {"servicenode", 1, 0, 's'},
1399 {"verbose", 0, 0, 'v'},
1400 {"writeconf", 0, 0, 'w'},
1401 {"upgrade_to_18", 0, 0, 'U'},
1402 {"network", 1, 0, 't'},
1405 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqrs:t:Uu:vw";
1408 int failnode_set = 0, servicenode_set = 0;
1410 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
1414 if (IS_MDT(&mop->mo_ldd))
1415 mop->mo_ldd.ldd_flags |= LDD_F_IAM_DIR;
1420 while (i < LDD_MT_LAST) {
1421 if (strcmp(optarg, mt_str(i)) == 0) {
1422 mop->mo_ldd.ldd_mount_type = i;
1430 if (IS_MDT(&mop->mo_ldd)) {
1431 int stripe_count = atol(optarg);
1432 if (stripe_count <= 0) {
1433 fprintf(stderr, "%s: bad stripe count "
1434 "%d\n", progname, stripe_count);
1437 mop->mo_stripe_count = stripe_count;
1439 badopt(long_opt[longidx].name, "MDT");
1443 case 'C': /* Configdev */
1445 printf("Configdev not implemented\n");
1448 mop->mo_device_sz = atol(optarg);
1451 mop->mo_ldd.ldd_params[0] = '\0';
1452 /* Must update the mgs logs */
1453 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1459 if ((opt == 'f' && servicenode_set)
1460 || (opt == 's' && failnode_set)) {
1461 fprintf(stderr, "%s: %s cannot use with --%s\n",
1462 progname, long_opt[longidx].name,
1463 opt == 'f' ? "servicenode" : "failnode");
1467 nids = convert_hostnames(optarg);
1470 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
1475 /* Must update the mgs logs */
1476 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1480 mop->mo_ldd.ldd_flags |= LDD_F_NO_PRIMNODE;
1481 servicenode_set = 1;
1487 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
1493 if (!(mop->mo_ldd.ldd_flags &
1494 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1495 LDD_F_WRITECONF))) {
1496 fprintf(stderr, "%s: cannot change the index of"
1497 " a registered target\n", progname);
1500 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
1501 mop->mo_ldd.ldd_svindex = atol(optarg);
1502 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
1504 badopt(long_opt[longidx].name, "MDT,OST");
1509 strscpy(mop->mo_mkfsopts, optarg,
1510 sizeof(mop->mo_mkfsopts));
1514 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
1515 (!(mop->mo_ldd.ldd_flags &
1516 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1517 LDD_F_WRITECONF)))) {
1518 fprintf(stderr, "%s: cannot change the name of"
1519 " a registered target\n", progname);
1522 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
1523 fprintf(stderr, "%s: filesystem name must be "
1524 "1-8 chars\n", progname);
1527 if ((tmp = strpbrk(optarg, "/:"))) {
1528 fprintf(stderr, "%s: char '%c' not allowed in "
1529 "filesystem name\n", progname, *tmp);
1532 strscpy(mop->mo_ldd.ldd_fsname, optarg,
1533 sizeof(mop->mo_ldd.ldd_fsname));
1537 char *nids = convert_hostnames(optarg);
1540 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
1545 mop->mo_mgs_failnodes++;
1549 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
1555 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
1558 *mountopts = optarg;
1561 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
1564 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg);
1567 /* Must update the mgs logs */
1568 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1574 mop->mo_flags |= MO_FORCEFORMAT;
1577 if (!IS_MDT(&mop->mo_ldd) && !IS_OST(&mop->mo_ldd)) {
1578 badopt(long_opt[longidx].name, "MDT,OST");
1585 rc = add_param(mop->mo_ldd.ldd_params,
1586 PARAM_NETWORK, optarg);
1589 /* Must update the mgs logs */
1590 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1593 strscpy(mop->mo_ldd.ldd_userdata, optarg,
1594 sizeof(mop->mo_ldd.ldd_userdata));
1600 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
1608 fprintf(stderr, "Unknown option '%c'\n", opt);
1614 /* Last arg is device */
1615 if (optind != argc - 1) {
1617 fprintf(stderr, "Bad argument: %s\n", argv[optind]);
1621 /* single argument: <device> */
1628 /* Search for opt in mntlist, returning true if found.
1630 static int in_mntlist(char *opt, char *mntlist)
1632 char *ml, *mlp, *item, *ctx = NULL;
1634 if (!(ml = strdup(mntlist))) {
1635 fprintf(stderr, "%s: out of memory\n", progname);
1639 while ((item = strtok_r(mlp, ",", &ctx))) {
1640 if (!strcmp(opt, item))
1645 return (item != NULL);
1648 /* Issue a message on stderr for every item in wanted_mountopts that is not
1649 * present in mountopts. The justwarn boolean toggles between error and
1650 * warning message. Return an error count.
1652 static int check_mountfsoptions(char *mountopts, char *wanted_mountopts,
1655 char *ml, *mlp, *item, *ctx = NULL;
1658 if (!(ml = strdup(wanted_mountopts))) {
1659 fprintf(stderr, "%s: out of memory\n", progname);
1663 while ((item = strtok_r(mlp, ",", &ctx))) {
1664 if (!in_mntlist(item, mountopts)) {
1665 fprintf(stderr, "%s: %s mount option `%s' is missing\n",
1666 progname, justwarn ? "Warning: default"
1667 : "Error: mandatory", item);
1676 /* Trim embedded white space, leading and trailing commas from string s.
1678 static void trim_mountfsoptions(char *s)
1684 memmove(p, p + 1, strlen(p + 1) + 1);
1691 memmove(&s[0], &s[1], strlen(&s[1]) + 1);
1693 p = s + strlen(s) - 1;
1694 while (p >= s && *p == ',')
1698 int main(int argc, char *const argv[])
1700 struct mkfs_opts mop;
1701 struct lustre_disk_data *ldd;
1702 char *mountopts = NULL;
1703 char always_mountopts[512] = "";
1704 char default_mountopts[512] = "";
1707 if ((progname = strrchr(argv[0], '/')) != NULL)
1712 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
1717 memset(&mop, 0, sizeof(mop));
1720 /* device is last arg */
1721 strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
1723 /* Are we using a loop device? */
1724 ret = is_block(mop.mo_device);
1728 mop.mo_flags |= MO_IS_LOOP;
1731 /* For tunefs, we must read in the old values before parsing any
1734 /* Check whether the disk has already been formatted by mkfs.lustre */
1735 ret = is_lustre_target(&mop);
1738 fprintf(stderr, "Device %s has not been formatted with "
1739 "mkfs.lustre\n", mop.mo_device);
1744 ret = read_local_files(&mop);
1747 fprintf(stderr, "Failed to read previous Lustre data from %s "
1748 "(%d)\n", mop.mo_device, ret);
1751 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
1752 mop.mo_mgs_failnodes++;
1755 print_ldd("Read previous values", &(mop.mo_ldd));
1758 ret = parse_opts(argc, argv, &mop, &mountopts);
1764 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
1766 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
1771 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
1773 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
1778 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
1779 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
1781 fprintf(stderr, "Can't find the target index, "
1782 "specify with --index\n");
1788 * Comment out these 2 checks temporarily, since for multi-MDSes
1789 * in single node only 1 mds node could have mgs service
1791 if (IS_MDT(ldd) && !IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1792 verrprint("No management node specified, adding MGS to this "
1794 ldd->ldd_flags |= LDD_F_SV_TYPE_MGS;
1796 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1799 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
1801 fprintf(stderr, "Must specify --mgsnode=\n");
1807 /* These are the permanent mount options (always included) */
1808 switch (ldd->ldd_mount_type) {
1810 case LDD_MT_LDISKFS:
1811 case LDD_MT_LDISKFS2:
1812 strscat(default_mountopts, ",errors=remount-ro",
1813 sizeof(default_mountopts));
1814 if (IS_MDT(ldd) || IS_MGS(ldd))
1815 strscat(always_mountopts, ",user_xattr",
1816 sizeof(always_mountopts));
1817 /* NB: Files created while extents are enabled can only be read
1818 * if mounted using the ext4 or ldiskfs filesystem type. */
1820 (ldd->ldd_mount_type == LDD_MT_LDISKFS ||
1821 ldd->ldd_mount_type == LDD_MT_LDISKFS2)) {
1822 strscat(default_mountopts, ",extents,mballoc",
1823 sizeof(default_mountopts));
1828 fprintf(stderr, "unknown fs type %d '%s'\n",
1829 ldd->ldd_mount_type,
1836 trim_mountfsoptions(mountopts);
1837 (void)check_mountfsoptions(mountopts, default_mountopts, 1);
1838 if (check_mountfsoptions(mountopts, always_mountopts, 0)) {
1842 sprintf(ldd->ldd_mount_opts, "%s", mountopts);
1845 if (ldd->ldd_mount_opts[0] == 0)
1846 /* use the defaults unless old opts exist */
1849 sprintf(ldd->ldd_mount_opts, "%s%s",
1850 always_mountopts, default_mountopts);
1851 trim_mountfsoptions(ldd->ldd_mount_opts);
1855 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
1856 ldd->ldd_fsname, ldd->ldd_svname);
1859 print_ldd("Permanent disk data", ldd);
1862 printf("exiting before disk write.\n");
1866 if (check_mtab_entry(mop.mo_device))
1869 /* Create the loopback file */
1870 if (mop.mo_flags & MO_IS_LOOP) {
1871 ret = access(mop.mo_device, F_OK);
1874 #ifndef TUNEFS /* mkfs.lustre */
1875 /* Reformat the loopback file */
1876 if (ret || (mop.mo_flags & MO_FORCEFORMAT))
1877 ret = loop_format(&mop);
1880 ret = loop_setup(&mop);
1883 fprintf(stderr, "Loop device setup for %s failed: %s\n",
1884 mop.mo_device, strerror(ret));
1889 #ifndef TUNEFS /* mkfs.lustre */
1890 /* Check whether the disk has already been formatted by mkfs.lustre */
1891 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
1892 ret = is_lustre_target(&mop);
1895 fprintf(stderr, "Device %s was previously formatted "
1896 "for lustre. Use --reformat to reformat it, "
1897 "or tunefs.lustre to modify.\n",
1903 /* Format the backing filesystem */
1904 ret = make_lustre_backfs(&mop);
1907 fprintf(stderr, "mkfs failed %d\n", ret);
1912 /* Write our config files */
1913 ret = write_local_files(&mop);
1916 fprintf(stderr, "failed to write local files\n");
1923 /* Fix any crazy return values from system() */
1924 if (ret && ((ret & 255) == 0))
1927 verrprint("%s: exiting with %d (%s)\n",
1928 progname, ret, strerror(ret));