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.
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
39 * lustre/utils/mkfs_lustre.c
41 * Author: Nathan Rutman <nathan@clusterfs.com>
44 /* This source file is compiled into both mkfs.lustre and tunefs.lustre */
48 #endif /* HAVE_CONFIG_H */
60 #include <sys/types.h>
62 #include <sys/mount.h>
63 #include <sys/utsname.h>
71 /* libcfs.h is not really needed here, but on SLES10/PPC, fs.h includes idr.h
72 * which requires BITS_PER_LONG to be defined */
73 #include <libcfs/libcfs.h>
75 #include <linux/fs.h> /* for BLKGETSIZE64 */
77 #include <linux/version.h>
79 #include <lustre_disk.h>
80 #include <lustre_param.h>
81 #include <lnet/lnetctl.h>
82 #include <lustre_ver.h>
83 #include "mount_utils.h"
89 #define MAX_LOOP_DEVICES 16
90 #define L_BLOCK_SIZE 4096
91 #define INDEX_UNASSIGNED 0xFFFF
92 #define MO_IS_LOOP 0x01
93 #define MO_FORCEFORMAT 0x02
95 /* used to describe the options to format the lustre disk, not persistent */
97 struct lustre_disk_data mo_ldd; /* to be written in MOUNT_DATA_FILE */
98 char mo_device[128]; /* disk device name */
99 char mo_loopdev[128]; /* in case a loop dev is needed */
100 char mo_mkfsopts[512]; /* options to the backing-store mkfs */
101 __u64 mo_device_sz; /* in KB */
104 int mo_mgs_failnodes;
109 static int print_only = 0;
110 static int failover = 0;
111 static int upgrade_to_18 = 0;
113 void usage(FILE *out)
115 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
116 fprintf(out, "usage: %s <target types> [options] <device>\n", progname);
118 "\t<device>:block device or file (e.g /dev/sda or /tmp/ost1)\n"
120 "\t\t--ost: object storage, mutually exclusive with mdt,mgs\n"
121 "\t\t--mdt: metadata storage, mutually exclusive with ost\n"
122 "\t\t--mgs: configuration management service - one per site\n"
123 "\toptions (in order of popularity):\n"
124 "\t\t--mgsnode=<nid>[,<...>] : NID(s) of a remote mgs node\n"
125 "\t\t\trequired for all targets other than the mgs node\n"
126 "\t\t--fsname=<filesystem_name> : default is 'lustre'\n"
127 "\t\t--failnode=<nid>[,<...>] : NID(s) of a failover partner\n"
128 "\t\t\tcannot be used with --servicenode\n"
129 "\t\t--servicenode=<nid>[,<...>] : NID(s) of all service partners\n"
130 "\t\t\ttreat all nodes as equal service node, cannot be used with --failnode\n"
131 "\t\t--param <key>=<value> : set a permanent parameter\n"
132 "\t\t\te.g. --param sys.timeout=40\n"
133 "\t\t\t --param lov.stripesize=2M\n"
134 "\t\t--index=#N : target index (i.e. ost index within lov)\n"
135 "\t\t--comment=<user comment>: arbitrary string (%d bytes)\n"
136 "\t\t--mountfsoptions=<opts> : permanent mount options\n"
137 "\t\t--network=<net>[,<...>] : restrict OST/MDT to network(s)\n"
139 "\t\t--backfstype=<fstype> : backing fs type (ext3, ldiskfs)\n"
140 "\t\t--device-size=#N(KB) : device size for loop devices\n"
141 "\t\t--mkfsoptions=<opts> : format options\n"
142 "\t\t--reformat: overwrite an existing disk\n"
143 "\t\t--stripe-count-hint=#N : for optimizing MDT inode size\n"
144 "\t\t--iam-dir: use IAM directory format, not ext3 compatible\n"
146 "\t\t--erase-params : erase all old parameter settings\n"
147 "\t\t--nomgs: turn off MGS service on this MDT\n"
148 "\t\t--writeconf: erase all config logs for this fs.\n"
150 "\t\t--dryrun: just report what we would do; "
151 "don't write to disk\n"
152 "\t\t--verbose : e.g. show mkfs progress\n"
154 (int)sizeof(((struct lustre_disk_data *)0)->ldd_userdata));
158 #define vprint if (verbose > 0) printf
159 #define verrprint if (verbose >= 0) printf
161 /*================ utility functions =====================*/
163 char *strscat(char *dst, char *src, int buflen) {
165 if (strlen(dst) + strlen(src) >= buflen) {
166 fprintf(stderr, "string buffer overflow (max %d): '%s' + '%s'"
167 "\n", buflen, dst, src);
170 return strcat(dst, src);
174 char *strscpy(char *dst, char *src, int buflen) {
176 return strscat(dst, src, buflen);
180 dev_major (unsigned long long int __dev)
182 return ((__dev >> 8) & 0xfff) | ((unsigned int) (__dev >> 32) & ~0xfff);
186 dev_minor (unsigned long long int __dev)
188 return (__dev & 0xff) | ((unsigned int) (__dev >> 12) & ~0xff);
193 static int version = 0;
197 char release[4] = "";
199 fd = open("/proc/sys/kernel/osrelease", O_RDONLY);
201 fprintf(stderr, "%s: Warning: Can't resolve kernel "
202 "version, assuming 2.6\n", progname);
204 if (read(fd, release, 4) < 0) {
205 fprintf(stderr, "reading from /proc/sys/kernel"
206 "/osrelease: %s\n", strerror(errno));
212 if (strncmp(release, "2.4.", 4) == 0)
220 static int check_mtab_entry(char *spec)
225 fp = setmntent(MOUNTED, "r");
229 while ((mnt = getmntent(fp)) != NULL) {
230 if (strcmp(mnt->mnt_fsname, spec) == 0) {
232 fprintf(stderr, "%s: according to %s %s is "
233 "already mounted on %s\n",
234 progname, MOUNTED, spec, mnt->mnt_dir);
243 /*============ disk dev functions ===================*/
245 /* Setup a file in the first unused loop_device */
246 int loop_setup(struct mkfs_opts *mop)
252 /* Figure out the loop device names */
253 if (!access("/dev/loop0", F_OK | R_OK)) {
254 strcpy(loop_base, "/dev/loop\0");
255 } else if (!access("/dev/loop/0", F_OK | R_OK)) {
256 strcpy(loop_base, "/dev/loop/\0");
258 fprintf(stderr, "%s: can't access loop devices\n", progname);
262 /* Find unused loop device */
263 for (i = 0; i < MAX_LOOP_DEVICES; i++) {
265 int cmdsz = sizeof(cmd);
267 sprintf(l_device, "%s%d", loop_base, i);
268 if (access(l_device, F_OK | R_OK))
270 snprintf(cmd, cmdsz, "losetup %s > /dev/null 2>&1", l_device);
273 /* losetup gets 1 (ret=256) for non-set-up device */
275 /* Set up a loopback device to our file */
276 snprintf(cmd, cmdsz, "losetup %s %s", l_device,
278 ret = run_command(cmd, cmdsz);
280 /* someone else picked up this loop device
284 fprintf(stderr, "%s: error %d on losetup: %s\n",
285 progname, ret, strerror(ret));
288 strscpy(mop->mo_loopdev, l_device,
289 sizeof(mop->mo_loopdev));
294 fprintf(stderr, "%s: out of loop devices!\n", progname);
298 int loop_cleanup(struct mkfs_opts *mop)
302 if ((mop->mo_flags & MO_IS_LOOP) && *mop->mo_loopdev) {
303 sprintf(cmd, "losetup -d %s", mop->mo_loopdev);
304 ret = run_command(cmd, sizeof(cmd));
309 /* Determine if a device is a block device (as opposed to a file) */
310 int is_block(char* devname)
315 ret = access(devname, F_OK);
318 ret = stat(devname, &st);
320 fprintf(stderr, "%s: cannot stat %s\n", progname, devname);
323 return S_ISBLK(st.st_mode);
326 __u64 get_device_size(char* device)
331 fd = open(device, O_RDONLY);
333 fprintf(stderr, "%s: cannot open %s: %s\n",
334 progname, device, strerror(errno));
339 /* size in bytes. bz5831 */
340 ret = ioctl(fd, BLKGETSIZE64, (void*)&size);
345 ret = ioctl(fd, BLKGETSIZE, (void*)&lsize);
346 size = (__u64)lsize * 512;
351 fprintf(stderr, "%s: size ioctl failed: %s\n",
352 progname, strerror(errno));
356 vprint("device size = "LPU64"MB\n", size >> 20);
357 /* return value in KB */
361 int loop_format(struct mkfs_opts *mop)
365 if (mop->mo_device_sz == 0) {
367 fprintf(stderr, "loop device requires a --device-size= "
372 ret = creat(mop->mo_device, S_IRUSR|S_IWUSR);
375 fprintf(stderr, "%s: Unable to create backing store: %d\n",
381 ret = truncate(mop->mo_device, mop->mo_device_sz * 1024);
384 fprintf(stderr, "%s: Unable to truncate backing store: %d\n",
391 /* Display the need for the latest e2fsprogs to be installed. make_backfs
392 * indicates if the caller is make_lustre_backfs() or not. */
393 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
395 static int msg_displayed;
398 fprintf(stderr, "WARNING: %s does not support %s "
399 "feature.\n\n", E2FSPROGS, feature);
405 fprintf(stderr, "WARNING: The %s package currently installed on "
406 "your system does not support \"%s\" feature.\n",
408 #if !(HAVE_LDISKFSPROGS)
409 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
410 "http://downloads.whamcloud.com/public/e2fsprogs/latest/\n"
411 "to enable this feature.\n");
414 fprintf(stderr, "Feature will not be enabled until %s"
415 "is updated and '%s -O %s %%{device}' "
416 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
419 /* Check whether the file exists in the device */
420 static int file_in_dev(char *file_name, char *dev_name)
423 char debugfs_cmd[256];
424 unsigned int inode_num;
427 /* Construct debugfs command line. */
428 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
429 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
430 DEBUGFS, file_name, dev_name);
432 fp = popen(debugfs_cmd, "r");
434 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
438 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
442 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd), fp);
445 fprintf(stderr, "%s", debugfs_cmd);
446 if (strstr(debugfs_cmd, "unsupported feature")) {
447 disp_old_e2fsprogs_msg("an unknown", 0);
456 /* Check whether the device has already been used with lustre */
457 static int is_lustre_target(struct mkfs_opts *mop)
461 vprint("checking for existing Lustre data: ");
463 if ((rc = file_in_dev(MOUNT_DATA_FILE, mop->mo_device))) {
465 (rc == 1) ? MOUNT_DATA_FILE : "extents");
466 /* in the -1 case, 'extents' means this really IS a lustre
471 if ((rc = file_in_dev(LAST_RCVD, mop->mo_device))) {
472 vprint("found %s\n", LAST_RCVD);
476 vprint("not found\n");
477 return 0; /* The device is not a lustre target. */
480 /* Check if a certain feature is supported by e2fsprogs.
481 * Firstly we try to use "debugfs supported_features" command to check if
482 * the feature is supported. If this fails we try to set this feature with
483 * mke2fs to check for its support. */
484 static int is_e2fsprogs_feature_supp(const char *feature)
486 static char supp_features[4096] = "";
489 char imgname[] = "/tmp/test-img-XXXXXX";
493 if (supp_features[0] == '\0') {
494 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
497 /* Using popen() instead of run_command() since debugfs does
498 * not return proper error code if command is not supported */
499 fp = popen(cmd, "r");
501 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
504 ret = fread(supp_features, 1, sizeof(supp_features), fp);
507 if (ret > 0 && strstr(supp_features,
508 strncmp(feature, "-O ", 3) ? feature : feature+3))
511 if ((fd = mkstemp(imgname)) < 0)
516 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
517 MKE2FS, feature, imgname);
518 /* run_command() displays the output of mke2fs when it fails for
519 * some feature, so use system() directly */
527 * append_unique: append @key or @key=@val pair to @buf only if @key does not
529 * @buf: buffer to hold @key or @key=@val
530 * @prefix: prefix string before @key
532 * @val: value string if it's a @key=@val pair
534 static void append_unique(char *buf, char *prefix, char *key, char *val,
543 anchor = end = strstr(buf, key);
544 /* try to find exact match string in @buf */
545 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
548 if (anchor == NULL || strlen(key) != len ||
549 strncmp(anchor, key, len) != 0) {
551 strscat(buf, prefix, maxbuflen);
553 strscat(buf, key, maxbuflen);
555 strscat(buf, "=", maxbuflen);
556 strscat(buf, val, maxbuflen);
561 static void enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
562 size_t maxbuflen, int user_spec)
564 if (IS_OST(&mop->mo_ldd)) {
565 append_unique(anchor, user_spec ? "," : " -O ",
566 "extents", NULL, sizeof(mop->mo_mkfsopts));
567 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
568 } else if (IS_MDT(&mop->mo_ldd)) {
569 append_unique(anchor, user_spec ? "," : " -O ",
570 "dirdata", NULL, maxbuflen);
571 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
573 append_unique(anchor, user_spec ? "," : " -O ",
574 "uninit_bg", NULL, maxbuflen);
577 /* Multiple mount protection enabled only if failover node specified */
579 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
580 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
582 disp_old_e2fsprogs_msg("mmp", 1);
585 /* Allow more than 65000 subdirectories */
586 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
587 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
589 #ifdef HAVE_EXT4_LDISKFS
590 /* The following options are only valid for ext4-based ldiskfs.
591 * If --backfstype=ext3 is specified, do not enable them. */
592 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
595 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
596 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
597 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
599 /* Enable large block addresses if the LUN is over 2^32 blocks. */
600 if (mop->mo_device_sz / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
601 is_e2fsprogs_feature_supp("-O 64bit") == 0)
602 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
604 /* Cluster inode/block bitmaps and inode table for more efficient IO.
605 * Align the flex groups on a 1MB boundary for better performance. */
606 /* This -O feature needs to go last, since it adds the "-G" option. */
607 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
610 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
612 if (IS_OST(&mop->mo_ldd)) {
613 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
614 (1 << 20) / L_BLOCK_SIZE);
615 strscat(anchor, tmp_buf, maxbuflen);
618 /* Don't add any more "-O" options here, see last comment above */
623 * moveopts_to_end: find the option string, move remaining strings to
624 * where option string starts, and append the option
626 * @start: where the option string starts before the move
627 * RETURN: where the option string starts after the move
629 static char *moveopts_to_end(char *start)
634 /* skip whitespace before options */
639 /* find end of option characters */
640 while (*end != ' ' && *end != '\0')
644 strncpy(save, start, end - start);
645 save[end - start] = '\0';
647 /* move remaining options up front */
649 memmove(start, end, strlen(end));
650 *(start + strlen(end)) = '\0';
652 /* append the specified options */
653 if (*(start + strlen(start) - 1) != ' ')
655 idx = start + strlen(start);
661 /* Build fs according to type */
662 int make_lustre_backfs(struct mkfs_opts *mop)
664 __u64 device_sz = mop->mo_device_sz, block_count = 0;
665 char mkfs_cmd[PATH_MAX];
669 int ret = 0, ext_opts = 0;
672 if (!(mop->mo_flags & MO_IS_LOOP)) {
673 mop->mo_device_sz = get_device_size(mop->mo_device);
675 if (mop->mo_device_sz == 0)
678 /* Compare to real size */
679 if (device_sz == 0 || device_sz > mop->mo_device_sz)
680 device_sz = mop->mo_device_sz;
682 mop->mo_device_sz = device_sz;
685 if (mop->mo_device_sz != 0) {
686 if (mop->mo_device_sz < 8096){
687 fprintf(stderr, "%s: size of filesystem must be larger "
688 "than 8MB, but is set to %lldKB\n",
689 progname, (long long)mop->mo_device_sz);
692 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
693 /* If the LUN size is just over 2^32 blocks, limit the
694 * filesystem size to 2^32-1 blocks to avoid problems with
695 * ldiskfs/mkfs not handling this size. Bug 22906 */
696 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
697 block_count = 0xffffffffULL;
700 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
701 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
702 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
705 /* Journal size in MB */
706 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
707 /* Choose our own default journal size */
708 long journal_sz = 0, max_sz;
709 if (device_sz > 1024 * 1024) /* 1GB */
710 journal_sz = (device_sz / 102400) * 4;
711 /* cap journal size at 1GB */
712 if (journal_sz > 1024L)
715 max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
716 if (journal_sz > max_sz)
719 sprintf(buf, " -J size=%ld", journal_sz);
720 strscat(mop->mo_mkfsopts, buf,
721 sizeof(mop->mo_mkfsopts));
725 /* Inode size (for extended attributes). The LOV EA size is
726 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
727 * and we want some margin above that for ACLs, other EAs... */
728 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
729 if (IS_MDT(&mop->mo_ldd)) {
730 if (mop->mo_stripe_count > 72)
731 inode_size = 512; /* bz 7241 */
732 /* see also "-i" below for EA blocks */
733 else if (mop->mo_stripe_count > 32)
735 else if (mop->mo_stripe_count > 10)
739 } else if (IS_OST(&mop->mo_ldd)) {
740 /* We store MDS FID and OST objid in EA on OST
741 * we need to make inode bigger as well. */
745 if (inode_size > 0) {
746 sprintf(buf, " -I %ld", inode_size);
747 strscat(mop->mo_mkfsopts, buf,
748 sizeof(mop->mo_mkfsopts));
752 /* Bytes_per_inode: disk size / num inodes */
753 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
754 strstr(mop->mo_mkfsopts, "-N") == NULL) {
755 long bytes_per_inode = 0;
757 /* Allocate more inodes on MDT devices. There is
758 * no data stored on the MDT, and very little extra
759 * metadata beyond the inode. It could go down as
760 * low as 1024 bytes, but this is conservative.
761 * Account for external EA blocks for wide striping. */
762 if (IS_MDT(&mop->mo_ldd)) {
763 bytes_per_inode = inode_size + 1536;
765 if (mop->mo_stripe_count > 72) {
766 int extra = mop->mo_stripe_count * 24;
767 extra = ((extra - 1) | 4095) + 1;
768 bytes_per_inode += extra;
772 /* Allocate fewer inodes on large OST devices. Most
773 * filesystems can be much more aggressive than even
774 * this, but it is impossible to know in advance. */
775 if (IS_OST(&mop->mo_ldd)) {
776 /* OST > 16TB assume average file size 1MB */
777 if (device_sz > (16ULL << 30))
778 bytes_per_inode = 1024 * 1024;
779 /* OST > 4TB assume average file size 512kB */
780 else if (device_sz > (4ULL << 30))
781 bytes_per_inode = 512 * 1024;
782 /* OST > 1TB assume average file size 256kB */
783 else if (device_sz > (1ULL << 30))
784 bytes_per_inode = 256 * 1024;
785 /* OST > 10GB assume average file size 64kB,
786 * plus a bit so that inodes will fit into a
787 * 256x flex_bg without overflowing */
788 else if (device_sz > (10ULL << 20))
789 bytes_per_inode = 69905;
792 if (bytes_per_inode > 0) {
793 sprintf(buf, " -i %ld", bytes_per_inode);
794 strscat(mop->mo_mkfsopts, buf,
795 sizeof(mop->mo_mkfsopts));
800 strscat(mop->mo_mkfsopts, " -q",
801 sizeof(mop->mo_mkfsopts));
804 /* start handle -O mkfs options */
805 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
806 if (strstr(start + 2, "-O") != NULL) {
808 "%s: don't specify multiple -O options\n",
812 start = moveopts_to_end(start);
813 maxbuflen = sizeof(mop->mo_mkfsopts) -
814 (start - mop->mo_mkfsopts) - strlen(start);
815 enable_default_ext4_features(mop, start, maxbuflen, 1);
817 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
818 maxbuflen = sizeof(mop->mo_mkfsopts) -
819 strlen(mop->mo_mkfsopts);
820 enable_default_ext4_features(mop, start, maxbuflen, 0);
822 /* end handle -O mkfs options */
824 /* start handle -E mkfs options */
825 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
826 if (strstr(start + 2, "-E") != NULL) {
828 "%s: don't specify multiple -E options\n",
832 start = moveopts_to_end(start);
833 maxbuflen = sizeof(mop->mo_mkfsopts) -
834 (start - mop->mo_mkfsopts) - strlen(start);
837 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
838 maxbuflen = sizeof(mop->mo_mkfsopts) -
839 strlen(mop->mo_mkfsopts);
842 /* In order to align the filesystem metadata on 1MB boundaries,
843 * give a resize value that will reserve a power-of-two group
844 * descriptor blocks, but leave one block for the superblock.
845 * Only useful for filesystems with < 2^32 blocks due to resize
847 if (IS_OST(&mop->mo_ldd) && mop->mo_device_sz > 100 * 1024 &&
848 mop->mo_device_sz * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
849 unsigned group_blocks = L_BLOCK_SIZE * 8;
850 unsigned desc_per_block = L_BLOCK_SIZE / 32;
851 unsigned resize_blks;
853 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
854 snprintf(buf, sizeof(buf), "%u", resize_blks);
855 append_unique(start, ext_opts ? "," : " -E ",
856 "resize", buf, maxbuflen);
860 /* Avoid zeroing out the full journal - speeds up mkfs */
861 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
862 append_unique(start, ext_opts ? "," : " -E ",
863 "lazy_journal_init", NULL, maxbuflen);
864 /* end handle -E mkfs options */
866 /* Allow reformat of full devices (as opposed to
867 partitions.) We already checked for mounted dev. */
868 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
870 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
871 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
872 mop->mo_ldd.ldd_svname);
874 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
875 progname, mop->mo_ldd.ldd_mount_type,
876 MT_STR(&mop->mo_ldd));
880 /* For loop device format the dev, not the filename */
881 dev = mop->mo_device;
882 if (mop->mo_flags & MO_IS_LOOP)
883 dev = mop->mo_loopdev;
885 vprint("formatting backing filesystem %s on %s\n",
886 MT_STR(&mop->mo_ldd), dev);
887 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
888 vprint("\t4k blocks "LPU64"\n", block_count);
889 vprint("\toptions %s\n", mop->mo_mkfsopts);
891 /* mkfs_cmd's trailing space is important! */
892 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
893 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
894 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
895 if (block_count != 0) {
896 sprintf(buf, " "LPU64, block_count);
897 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
900 vprint("mkfs_cmd = %s\n", mkfs_cmd);
901 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
904 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
909 /* ==================== Lustre config functions =============*/
911 void print_ldd(char *str, struct lustre_disk_data *ldd)
913 printf("\n %s:\n", str);
914 printf("Target: %s\n", ldd->ldd_svname);
915 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
916 printf("Index: unassigned\n");
918 printf("Index: %d\n", ldd->ldd_svindex);
919 if (ldd->ldd_uuid[0])
920 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
921 printf("Lustre FS: %s\n", ldd->ldd_fsname);
922 printf("Mount type: %s\n", MT_STR(ldd));
923 printf("Flags: %#x\n", ldd->ldd_flags);
924 printf(" (%s%s%s%s%s%s%s%s%s%s)\n",
925 IS_MDT(ldd) ? "MDT ":"",
926 IS_OST(ldd) ? "OST ":"",
927 IS_MGS(ldd) ? "MGS ":"",
928 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
929 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
930 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
931 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
932 ldd->ldd_flags & LDD_F_IAM_DIR ? "IAM_dir_format ":"",
933 ldd->ldd_flags & LDD_F_NO_PRIMNODE? "no_primnode ":"",
934 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
935 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
936 printf("Parameters:%s\n", ldd->ldd_params);
937 if (ldd->ldd_userdata[0])
938 printf("Comment: %s\n", ldd->ldd_userdata);
942 static int touch_file(char *filename)
946 if (filename == NULL) {
950 fd = open(filename, O_CREAT | O_TRUNC, 0600);
959 /* keep it less than LL_FID_NAMELEN */
960 #define DUMMY_FILE_NAME_LEN 25
961 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
963 /* Need to add these many entries to this directory to make HTREE dir. */
964 #define MIN_ENTRIES_REQ_FOR_HTREE ((L_BLOCK_SIZE / EXT3_DIRENT_SIZE))
966 static int add_dummy_files(char *dir)
968 char fpname[PATH_MAX];
972 for (i = 0; i < MIN_ENTRIES_REQ_FOR_HTREE; i++) {
973 snprintf(fpname, PATH_MAX, "%s/%0*d", dir,
974 DUMMY_FILE_NAME_LEN, i);
976 rc = touch_file(fpname);
977 if (rc && rc != -EEXIST) {
979 "%s: Can't create dummy file %s: %s\n",
980 progname, fpname , strerror(errno));
987 static int __l_mkdir(char * filepnm, int mode , struct mkfs_opts *mop)
991 ret = mkdir(filepnm, mode);
992 if (ret && ret != -EEXIST)
995 /* IAM mode supports ext3 directories of HTREE type only. So add dummy
996 * entries to new directory to create htree type of container for
998 if (mop->mo_ldd.ldd_flags & LDD_F_IAM_DIR)
999 return add_dummy_files(filepnm);
1003 /* Write the server config files */
1004 int write_local_files(struct mkfs_opts *mop)
1006 char mntpt[] = "/tmp/mntXXXXXX";
1013 /* Mount this device temporarily in order to write these files */
1014 if (!mkdtemp(mntpt)) {
1015 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
1016 progname, mntpt, strerror(errno));
1020 dev = mop->mo_device;
1021 if (mop->mo_flags & MO_IS_LOOP)
1022 dev = mop->mo_loopdev;
1024 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
1025 mop->mo_ldd.ldd_mount_opts);
1027 fprintf(stderr, "%s: Unable to mount %s: %s\n",
1028 progname, dev, strerror(errno));
1030 if (errno == ENODEV) {
1031 fprintf(stderr, "Is the %s module available?\n",
1032 MT_STR(&mop->mo_ldd));
1037 /* Set up initial directories */
1038 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
1039 ret = __l_mkdir(filepnm, 0777, mop);
1040 if ((ret != 0) && (errno != EEXIST)) {
1041 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
1042 progname, filepnm, strerror(errno));
1044 } else if (errno == EEXIST) {
1048 /* Save the persistent mount data into a file. Lustre must pre-read
1049 this file to get the real mount options. */
1050 vprint("Writing %s\n", MOUNT_DATA_FILE);
1051 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
1052 filep = fopen(filepnm, "w");
1054 fprintf(stderr, "%s: Unable to create %s file: %s\n",
1055 progname, filepnm, strerror(errno));
1058 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
1059 if (num < 1 && ferror(filep)) {
1060 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
1061 progname, filepnm, strerror(errno));
1067 /* Check for upgrade */
1068 if ((mop->mo_ldd.ldd_flags & (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS))
1069 == (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS)) {
1072 int cmdsz = sizeof(cmd);
1073 vprint("Copying old logs\n");
1075 /* Copy the old client log to fsname-client */
1076 sprintf(filepnm, "%s/%s/%s-client",
1077 mntpt, MOUNT_CONFIGS_DIR, mop->mo_ldd.ldd_fsname);
1078 snprintf(cmd, cmdsz, "cp %s/%s/client %s", mntpt, MDT_LOGS_DIR,
1080 ret = run_command(cmd, cmdsz);
1082 fprintf(stderr, "%s: Can't copy 1.4 config %s/client "
1083 "(%d)\n", progname, MDT_LOGS_DIR, ret);
1084 fprintf(stderr, "mount -t ldiskfs %s somewhere, "
1085 "find the client log for fs %s and "
1086 "copy it manually into %s/%s-client, "
1089 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
1090 mop->mo_ldd.ldd_fsname);
1094 /* We need to use the old mdt log because otherwise mdt won't
1095 have complete lov if old clients connect before all
1097 /* Copy the old mdt log to fsname-MDT0000 (get old
1098 name from mdt_UUID) */
1100 strscpy(filepnm, (char *)mop->mo_ldd.ldd_uuid, sizeof(filepnm));
1101 term = strstr(filepnm, "_UUID");
1104 snprintf(cmd, cmdsz, "cp %s/%s/%s %s/%s/%s",
1105 mntpt, MDT_LOGS_DIR, filepnm,
1106 mntpt, MOUNT_CONFIGS_DIR,
1107 mop->mo_ldd.ldd_svname);
1108 ret = run_command(cmd, cmdsz);
1111 fprintf(stderr, "%s: Can't copy 1.4 config %s/%s "
1112 "(%d)\n", progname, MDT_LOGS_DIR, filepnm, ret);
1113 fprintf(stderr, "mount -t ext3 %s somewhere, "
1114 "find the MDT log for fs %s and "
1115 "copy it manually into %s/%s, "
1118 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
1119 mop->mo_ldd.ldd_svname);
1124 /* end COMPAT_146 */
1133 int read_local_files(struct mkfs_opts *mop)
1135 char tmpdir[] = "/tmp/dirXXXXXX";
1141 int cmdsz = sizeof(cmd);
1143 /* Make a temporary directory to hold Lustre data files. */
1144 if (!mkdtemp(tmpdir)) {
1145 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
1146 progname, tmpdir, strerror(errno));
1150 dev = mop->mo_device;
1152 /* TODO: it's worth observing the get_mountdata() function that is
1153 in mount_utils.c for getting the mountdata out of the
1156 /* Construct debugfs command line. */
1157 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
1158 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
1160 ret = run_command(cmd, cmdsz);
1162 verrprint("%s: Unable to dump %s dir (%d)\n",
1163 progname, MOUNT_CONFIGS_DIR, ret);
1165 sprintf(filepnm, "%s/mountdata", tmpdir);
1166 filep = fopen(filepnm, "r");
1169 vprint("Reading %s\n", MOUNT_DATA_FILE);
1170 num_read = fread(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
1171 if (num_read < 1 && ferror(filep)) {
1172 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
1173 progname, filepnm, strerror(errno));
1178 /* Try to read pre-1.6 config from last_rcvd */
1179 struct lr_server_data lsd;
1180 verrprint("%s: Unable to read %d.%d config %s.\n",
1181 progname, LUSTRE_MAJOR, LUSTRE_MINOR, filepnm);
1183 verrprint("Trying 1.4 config from last_rcvd\n");
1184 sprintf(filepnm, "%s/%s", tmpdir, LAST_RCVD);
1186 /* Construct debugfs command line. */
1187 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s' %s",
1188 DEBUGFS, LAST_RCVD, filepnm, dev);
1190 ret = run_command(cmd, cmdsz);
1192 fprintf(stderr, "%s: Unable to dump %s file (%d)\n",
1193 progname, LAST_RCVD, ret);
1197 filep = fopen(filepnm, "r");
1199 fprintf(stderr, "%s: Unable to open %s: %s\n",
1200 progname, filepnm, strerror(errno));
1202 verrprint("Contents of %s:\n", tmpdir);
1204 snprintf(cmd, cmdsz, "ls -l %s/", tmpdir);
1205 run_command(cmd, cmdsz);
1206 verrprint("Contents of disk:\n");
1207 snprintf(cmd, cmdsz, "%s -c -R 'ls -l /' %s",
1209 run_command(cmd, cmdsz);
1213 vprint("Reading %s\n", LAST_RCVD);
1214 ret = fread(&lsd, 1, sizeof(lsd), filep);
1215 if (ret < sizeof(lsd)) {
1216 fprintf(stderr, "%s: Short read (%d of %d)\n",
1217 progname, ret, (int)sizeof(lsd));
1218 ret = ferror(filep);
1222 vprint("Feature compat=%x, incompat=%x\n",
1223 lsd.lsd_feature_compat, lsd.lsd_feature_incompat);
1225 if ((lsd.lsd_feature_compat & OBD_COMPAT_OST) ||
1226 (lsd.lsd_feature_incompat & OBD_INCOMPAT_OST)) {
1227 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
1228 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
1229 } else if ((lsd.lsd_feature_compat & OBD_COMPAT_MDT) ||
1230 (lsd.lsd_feature_incompat & OBD_INCOMPAT_MDT)) {
1231 /* We must co-locate so mgs can see old logs.
1232 If user doesn't want this, they can copy the old
1233 logs manually and re-tunefs. */
1234 mop->mo_ldd.ldd_flags =
1235 LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_MGS;
1236 mop->mo_ldd.ldd_svindex = lsd.lsd_mdt_index;
1238 /* If neither is set, we're pre-1.4.6, make a guess. */
1239 /* Construct debugfs command line. */
1240 snprintf(cmd, cmdsz, "%s -c -R 'rdump /%s %s' %s",
1241 DEBUGFS, MDT_LOGS_DIR, tmpdir, dev);
1242 run_command(cmd, cmdsz);
1244 sprintf(filepnm, "%s/%s", tmpdir, MDT_LOGS_DIR);
1245 if (lsd.lsd_ost_index > 0) {
1246 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
1247 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
1249 /* If there's a LOGS dir, it's an MDT */
1250 if ((ret = access(filepnm, F_OK)) == 0) {
1251 mop->mo_ldd.ldd_flags =
1254 /* Old MDT's are always index 0
1256 mop->mo_ldd.ldd_svindex = 0;
1258 /* The index may not be correct */
1259 mop->mo_ldd.ldd_flags =
1260 LDD_F_SV_TYPE_OST | LDD_F_NEED_INDEX;
1261 verrprint("OST with unknown index\n");
1267 memcpy(mop->mo_ldd.ldd_uuid, lsd.lsd_uuid,
1268 sizeof(mop->mo_ldd.ldd_uuid));
1269 mop->mo_ldd.ldd_flags |= LDD_F_UPGRADE14;
1271 /* end COMPAT_146 */
1276 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
1277 run_command(cmd, cmdsz);
1279 verrprint("Failed to read old data (%d)\n", ret);
1284 void set_defaults(struct mkfs_opts *mop)
1286 mop->mo_ldd.ldd_magic = LDD_MAGIC;
1287 mop->mo_ldd.ldd_config_ver = 1;
1288 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
1289 mop->mo_mgs_failnodes = 0;
1290 strcpy(mop->mo_ldd.ldd_fsname, "lustre");
1291 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
1293 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
1294 mop->mo_stripe_count = 1;
1297 static inline void badopt(const char *opt, char *type)
1299 fprintf(stderr, "%s: '--%s' only valid for %s\n",
1300 progname, opt, type);
1304 static int add_param(char *buf, char *key, char *val)
1306 int end = sizeof(((struct lustre_disk_data *)0)->ldd_params);
1307 int start = strlen(buf);
1311 keylen = strlen(key);
1312 if (start + 1 + keylen + strlen(val) >= end) {
1313 fprintf(stderr, "%s: params are too long-\n%s %s%s\n",
1314 progname, buf, key ? key : "", val);
1318 sprintf(buf + start, " %s%s", key ? key : "", val);
1322 /* from mount_lustre */
1323 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
1324 #define MAXNIDSTR 1024
1325 static char *convert_hostnames(char *s1)
1327 char *converted, *s2 = 0, *c, *end, sep;
1328 int left = MAXNIDSTR;
1331 converted = malloc(left);
1332 if (converted == NULL) {
1336 end = s1 + strlen(s1);
1338 while ((left > 0) && (s1 < end)) {
1339 s2 = strpbrk(s1, ",:");
1344 nid = libcfs_str2nid(s1);
1346 if (nid == LNET_NID_ANY) {
1347 fprintf(stderr, "%s: Can't parse NID '%s'\n",
1352 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
1353 strlen("127.0.0.1")) == 0) {
1354 fprintf(stderr, "%s: The NID '%s' resolves to the "
1355 "loopback address '%s'. Lustre requires a "
1356 "non-loopback address.\n",
1357 progname, s1, libcfs_nid2str(nid));
1362 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
1363 left = converted + MAXNIDSTR - c;
1369 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
1372 static struct option long_opt[] = {
1373 {"iam-dir", 0, 0, 'a'},
1374 {"backfstype", 1, 0, 'b'},
1375 {"stripe-count-hint", 1, 0, 'c'},
1376 {"comment", 1, 0, 'u'},
1377 {"configdev", 1, 0, 'C'},
1378 {"device-size", 1, 0, 'd'},
1379 {"dryrun", 0, 0, 'n'},
1380 {"erase-params", 0, 0, 'e'},
1381 {"failnode", 1, 0, 'f'},
1382 {"failover", 1, 0, 'f'},
1384 {"help", 0, 0, 'h'},
1385 {"index", 1, 0, 'i'},
1386 {"mkfsoptions", 1, 0, 'k'},
1387 {"mgsnode", 1, 0, 'm'},
1388 {"mgsnid", 1, 0, 'm'},
1390 {"fsname",1, 0, 'L'},
1391 {"noformat", 0, 0, 'n'},
1392 {"nomgs", 0, 0, 'N'},
1393 {"mountfsoptions", 1, 0, 'o'},
1395 {"param", 1, 0, 'p'},
1396 {"print", 0, 0, 'n'},
1397 {"quiet", 0, 0, 'q'},
1398 {"reformat", 0, 0, 'r'},
1399 {"servicenode", 1, 0, 's'},
1400 {"verbose", 0, 0, 'v'},
1401 {"writeconf", 0, 0, 'w'},
1402 {"upgrade_to_18", 0, 0, 'U'},
1403 {"network", 1, 0, 't'},
1406 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqrs:t:Uu:vw";
1409 int failnode_set = 0, servicenode_set = 0;
1411 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
1415 if (IS_MDT(&mop->mo_ldd))
1416 mop->mo_ldd.ldd_flags |= LDD_F_IAM_DIR;
1421 while (i < LDD_MT_LAST) {
1422 if (strcmp(optarg, mt_str(i)) == 0) {
1423 mop->mo_ldd.ldd_mount_type = i;
1431 if (IS_MDT(&mop->mo_ldd)) {
1432 int stripe_count = atol(optarg);
1433 if (stripe_count <= 0) {
1434 fprintf(stderr, "%s: bad stripe count "
1435 "%d\n", progname, stripe_count);
1438 mop->mo_stripe_count = stripe_count;
1440 badopt(long_opt[longidx].name, "MDT");
1444 case 'C': /* Configdev */
1446 printf("Configdev not implemented\n");
1449 mop->mo_device_sz = atol(optarg);
1452 mop->mo_ldd.ldd_params[0] = '\0';
1453 /* Must update the mgs logs */
1454 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1460 if ((opt == 'f' && servicenode_set)
1461 || (opt == 's' && failnode_set)) {
1462 fprintf(stderr, "%s: %s cannot use with --%s\n",
1463 progname, long_opt[longidx].name,
1464 opt == 'f' ? "servicenode" : "failnode");
1468 nids = convert_hostnames(optarg);
1471 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
1476 /* Must update the mgs logs */
1477 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1481 mop->mo_ldd.ldd_flags |= LDD_F_NO_PRIMNODE;
1482 servicenode_set = 1;
1488 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
1494 if (!(mop->mo_ldd.ldd_flags &
1495 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1496 LDD_F_WRITECONF))) {
1497 fprintf(stderr, "%s: cannot change the index of"
1498 " a registered target\n", progname);
1501 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
1502 mop->mo_ldd.ldd_svindex = atol(optarg);
1503 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
1505 badopt(long_opt[longidx].name, "MDT,OST");
1510 strscpy(mop->mo_mkfsopts, optarg,
1511 sizeof(mop->mo_mkfsopts));
1515 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
1516 (!(mop->mo_ldd.ldd_flags &
1517 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1518 LDD_F_WRITECONF)))) {
1519 fprintf(stderr, "%s: cannot change the name of"
1520 " a registered target\n", progname);
1523 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
1524 fprintf(stderr, "%s: filesystem name must be "
1525 "1-8 chars\n", progname);
1528 if ((tmp = strpbrk(optarg, "/:"))) {
1529 fprintf(stderr, "%s: char '%c' not allowed in "
1530 "filesystem name\n", progname, *tmp);
1533 strscpy(mop->mo_ldd.ldd_fsname, optarg,
1534 sizeof(mop->mo_ldd.ldd_fsname));
1538 char *nids = convert_hostnames(optarg);
1541 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
1546 mop->mo_mgs_failnodes++;
1550 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
1556 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
1559 *mountopts = optarg;
1562 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
1565 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg);
1568 /* Must update the mgs logs */
1569 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1575 mop->mo_flags |= MO_FORCEFORMAT;
1578 if (!IS_MDT(&mop->mo_ldd) && !IS_OST(&mop->mo_ldd)) {
1579 badopt(long_opt[longidx].name, "MDT,OST");
1586 rc = add_param(mop->mo_ldd.ldd_params,
1587 PARAM_NETWORK, optarg);
1590 /* Must update the mgs logs */
1591 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1594 strscpy(mop->mo_ldd.ldd_userdata, optarg,
1595 sizeof(mop->mo_ldd.ldd_userdata));
1601 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
1609 fprintf(stderr, "Unknown option '%c'\n", opt);
1615 /* Last arg is device */
1616 if (optind != argc - 1) {
1618 fprintf(stderr, "Bad argument: %s\n", argv[optind]);
1622 /* single argument: <device> */
1629 /* Search for opt in mntlist, returning true if found.
1631 static int in_mntlist(char *opt, char *mntlist)
1633 char *ml, *mlp, *item, *ctx = NULL;
1635 if (!(ml = strdup(mntlist))) {
1636 fprintf(stderr, "%s: out of memory\n", progname);
1640 while ((item = strtok_r(mlp, ",", &ctx))) {
1641 if (!strcmp(opt, item))
1646 return (item != NULL);
1649 /* Issue a message on stderr for every item in wanted_mountopts that is not
1650 * present in mountopts. The justwarn boolean toggles between error and
1651 * warning message. Return an error count.
1653 static int check_mountfsoptions(char *mountopts, char *wanted_mountopts,
1656 char *ml, *mlp, *item, *ctx = NULL;
1659 if (!(ml = strdup(wanted_mountopts))) {
1660 fprintf(stderr, "%s: out of memory\n", progname);
1664 while ((item = strtok_r(mlp, ",", &ctx))) {
1665 if (!in_mntlist(item, mountopts)) {
1666 fprintf(stderr, "%s: %s mount option `%s' is missing\n",
1667 progname, justwarn ? "Warning: default"
1668 : "Error: mandatory", item);
1677 /* Trim embedded white space, leading and trailing commas from string s.
1679 static void trim_mountfsoptions(char *s)
1685 memmove(p, p + 1, strlen(p + 1) + 1);
1692 memmove(&s[0], &s[1], strlen(&s[1]) + 1);
1694 p = s + strlen(s) - 1;
1695 while (p >= s && *p == ',')
1699 int main(int argc, char *const argv[])
1701 struct mkfs_opts mop;
1702 struct lustre_disk_data *ldd;
1703 char *mountopts = NULL;
1704 char always_mountopts[512] = "";
1705 char default_mountopts[512] = "";
1708 if ((progname = strrchr(argv[0], '/')) != NULL)
1713 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
1718 memset(&mop, 0, sizeof(mop));
1721 /* device is last arg */
1722 strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
1724 /* Are we using a loop device? */
1725 ret = is_block(mop.mo_device);
1729 mop.mo_flags |= MO_IS_LOOP;
1732 /* For tunefs, we must read in the old values before parsing any
1735 /* Check whether the disk has already been formatted by mkfs.lustre */
1736 ret = is_lustre_target(&mop);
1739 fprintf(stderr, "Device %s has not been formatted with "
1740 "mkfs.lustre\n", mop.mo_device);
1745 ret = read_local_files(&mop);
1748 fprintf(stderr, "Failed to read previous Lustre data from %s "
1749 "(%d)\n", mop.mo_device, ret);
1752 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
1753 mop.mo_mgs_failnodes++;
1756 print_ldd("Read previous values", &(mop.mo_ldd));
1759 ret = parse_opts(argc, argv, &mop, &mountopts);
1765 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
1767 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
1772 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
1774 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
1779 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
1780 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
1782 fprintf(stderr, "Can't find the target index, "
1783 "specify with --index\n");
1789 * Comment out these 2 checks temporarily, since for multi-MDSes
1790 * in single node only 1 mds node could have mgs service
1792 if (IS_MDT(ldd) && !IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1793 verrprint("No management node specified, adding MGS to this "
1795 ldd->ldd_flags |= LDD_F_SV_TYPE_MGS;
1797 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1800 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
1802 fprintf(stderr, "Must specify --mgsnode=\n");
1808 /* These are the permanent mount options (always included) */
1809 switch (ldd->ldd_mount_type) {
1811 case LDD_MT_LDISKFS:
1812 case LDD_MT_LDISKFS2:
1813 strscat(default_mountopts, ",errors=remount-ro",
1814 sizeof(default_mountopts));
1815 if (IS_MDT(ldd) || IS_MGS(ldd))
1816 strscat(always_mountopts, ",user_xattr",
1817 sizeof(always_mountopts));
1818 /* NB: Files created while extents are enabled can only be read
1819 * if mounted using the ext4 or ldiskfs filesystem type. */
1821 (ldd->ldd_mount_type == LDD_MT_LDISKFS ||
1822 ldd->ldd_mount_type == LDD_MT_LDISKFS2)) {
1823 strscat(default_mountopts, ",extents,mballoc",
1824 sizeof(default_mountopts));
1829 fprintf(stderr, "unknown fs type %d '%s'\n",
1830 ldd->ldd_mount_type,
1837 trim_mountfsoptions(mountopts);
1838 (void)check_mountfsoptions(mountopts, default_mountopts, 1);
1839 if (check_mountfsoptions(mountopts, always_mountopts, 0)) {
1843 sprintf(ldd->ldd_mount_opts, "%s", mountopts);
1846 if (ldd->ldd_mount_opts[0] == 0)
1847 /* use the defaults unless old opts exist */
1850 sprintf(ldd->ldd_mount_opts, "%s%s",
1851 always_mountopts, default_mountopts);
1852 trim_mountfsoptions(ldd->ldd_mount_opts);
1856 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
1857 ldd->ldd_fsname, ldd->ldd_svname);
1860 print_ldd("Permanent disk data", ldd);
1863 printf("exiting before disk write.\n");
1867 if (check_mtab_entry(mop.mo_device))
1870 /* Create the loopback file */
1871 if (mop.mo_flags & MO_IS_LOOP) {
1872 ret = access(mop.mo_device, F_OK);
1875 #ifndef TUNEFS /* mkfs.lustre */
1876 /* Reformat the loopback file */
1877 if (ret || (mop.mo_flags & MO_FORCEFORMAT))
1878 ret = loop_format(&mop);
1881 ret = loop_setup(&mop);
1884 fprintf(stderr, "Loop device setup for %s failed: %s\n",
1885 mop.mo_device, strerror(ret));
1890 #ifndef TUNEFS /* mkfs.lustre */
1891 /* Check whether the disk has already been formatted by mkfs.lustre */
1892 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
1893 ret = is_lustre_target(&mop);
1896 fprintf(stderr, "Device %s was previously formatted "
1897 "for lustre. Use --reformat to reformat it, "
1898 "or tunefs.lustre to modify.\n",
1904 /* Format the backing filesystem */
1905 ret = make_lustre_backfs(&mop);
1908 fprintf(stderr, "mkfs failed %d\n", ret);
1913 /* Write our config files */
1914 ret = write_local_files(&mop);
1917 fprintf(stderr, "failed to write local files\n");
1924 /* Fix any crazy return values from system() */
1925 if (ret && ((ret & 255) == 0))
1928 verrprint("%s: exiting with %d (%s)\n",
1929 progname, ret, strerror(ret));