4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, Whamcloud, Inc.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/utils/mkfs_lustre.c
38 * Author: Nathan Rutman <nathan@clusterfs.com>
41 /* This source file is compiled into both mkfs.lustre and tunefs.lustre */
45 #endif /* HAVE_CONFIG_H */
57 #include <sys/types.h>
59 #include <sys/mount.h>
60 #include <sys/utsname.h>
68 /* libcfs.h is not really needed here, but on SLES10/PPC, fs.h includes idr.h
69 * which requires BITS_PER_LONG to be defined */
70 #include <libcfs/libcfs.h>
72 #include <linux/fs.h> /* for BLKGETSIZE64 */
74 #include <linux/version.h>
76 #include <lustre_disk.h>
77 #include <lustre_param.h>
78 #include <lnet/lnetctl.h>
79 #include <lustre_ver.h>
80 #include "mount_utils.h"
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 /* return canonicalized absolute pathname, even if the target file does not
307 * exist, unlike realpath */
308 static char *absolute_path(char *devname)
310 char buf[PATH_MAX + 1];
314 path = malloc(PATH_MAX + 1);
318 if (devname[0] != '/') {
319 if (getcwd(buf, sizeof(buf) - 1) == NULL)
322 strcat(buf, devname);
324 strcpy(buf, devname);
326 /* truncate filename before calling realpath */
327 ptr = strrchr(buf, '/');
333 if (path != realpath(buf, path)) {
337 /* add the filename back */
339 strcat(path, ptr + 1);
343 /* Determine if a device is a block device (as opposed to a file) */
344 int is_block(char* devname)
350 devpath = absolute_path(devname);
351 if (devpath == NULL) {
352 fprintf(stderr, "%s: failed to resolve path to %s\n",
357 ret = access(devname, F_OK);
359 if (strncmp(devpath, "/dev/", 5) == 0) {
360 /* nobody sane wants to create a loopback file under
361 * /dev. Let's just report the device doesn't exist */
362 fprintf(stderr, "%s: %s apparently does not exist\n",
370 ret = stat(devpath, &st);
372 fprintf(stderr, "%s: cannot stat %s\n", progname, devpath);
375 ret = S_ISBLK(st.st_mode);
381 __u64 get_device_size(char* device)
386 fd = open(device, O_RDONLY);
388 fprintf(stderr, "%s: cannot open %s: %s\n",
389 progname, device, strerror(errno));
394 /* size in bytes. bz5831 */
395 ret = ioctl(fd, BLKGETSIZE64, (void*)&size);
400 ret = ioctl(fd, BLKGETSIZE, (void*)&lsize);
401 size = (__u64)lsize * 512;
406 fprintf(stderr, "%s: size ioctl failed: %s\n",
407 progname, strerror(errno));
411 vprint("device size = "LPU64"MB\n", size >> 20);
412 /* return value in KB */
416 int loop_format(struct mkfs_opts *mop)
420 if (mop->mo_device_sz == 0) {
422 fprintf(stderr, "loop device requires a --device-size= "
427 fd = creat(mop->mo_device, S_IRUSR|S_IWUSR);
430 fprintf(stderr, "%s: Unable to create backing store: %s\n",
431 progname, strerror(errno));
435 if (ftruncate(fd, mop->mo_device_sz * 1024) != 0) {
438 fprintf(stderr, "%s: Unable to truncate backing store: %s\n",
439 progname, strerror(errno));
447 /* Display the need for the latest e2fsprogs to be installed. make_backfs
448 * indicates if the caller is make_lustre_backfs() or not. */
449 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
451 static int msg_displayed;
454 fprintf(stderr, "WARNING: %s does not support %s "
455 "feature.\n\n", E2FSPROGS, feature);
461 fprintf(stderr, "WARNING: The %s package currently installed on "
462 "your system does not support \"%s\" feature.\n",
464 #if !(HAVE_LDISKFSPROGS)
465 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
466 "http://downloads.whamcloud.com/public/e2fsprogs/latest/\n"
467 "to enable this feature.\n");
470 fprintf(stderr, "Feature will not be enabled until %s"
471 "is updated and '%s -O %s %%{device}' "
472 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
475 /* Check whether the file exists in the device */
476 static int file_in_dev(char *file_name, char *dev_name)
479 char debugfs_cmd[256];
480 unsigned int inode_num;
483 /* Construct debugfs command line. */
484 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
485 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
486 DEBUGFS, file_name, dev_name);
488 fp = popen(debugfs_cmd, "r");
490 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
494 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
498 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd), fp);
501 fprintf(stderr, "%s", debugfs_cmd);
502 if (strstr(debugfs_cmd, "unsupported feature")) {
503 disp_old_e2fsprogs_msg("an unknown", 0);
512 /* Check whether the device has already been used with lustre */
513 static int is_lustre_target(struct mkfs_opts *mop)
517 vprint("checking for existing Lustre data: ");
519 if ((rc = file_in_dev(MOUNT_DATA_FILE, mop->mo_device))) {
521 (rc == 1) ? MOUNT_DATA_FILE : "extents");
522 /* in the -1 case, 'extents' means this really IS a lustre
527 if ((rc = file_in_dev(LAST_RCVD, mop->mo_device))) {
528 vprint("found %s\n", LAST_RCVD);
532 vprint("not found\n");
533 return 0; /* The device is not a lustre target. */
536 /* Check if a certain feature is supported by e2fsprogs.
537 * Firstly we try to use "debugfs supported_features" command to check if
538 * the feature is supported. If this fails we try to set this feature with
539 * mke2fs to check for its support. */
540 static int is_e2fsprogs_feature_supp(const char *feature)
542 static char supp_features[4096] = "";
545 char imgname[] = "/tmp/test-img-XXXXXX";
549 if (supp_features[0] == '\0') {
550 snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
553 /* Using popen() instead of run_command() since debugfs does
554 * not return proper error code if command is not supported */
555 fp = popen(cmd, "r");
557 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
560 ret = fread(supp_features, 1, sizeof(supp_features), fp);
563 if (ret > 0 && strstr(supp_features,
564 strncmp(feature, "-O ", 3) ? feature : feature+3))
567 if ((fd = mkstemp(imgname)) < 0)
572 snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
573 MKE2FS, feature, imgname);
574 /* run_command() displays the output of mke2fs when it fails for
575 * some feature, so use system() directly */
583 * append_unique: append @key or @key=@val pair to @buf only if @key does not
585 * @buf: buffer to hold @key or @key=@val
586 * @prefix: prefix string before @key
588 * @val: value string if it's a @key=@val pair
590 static void append_unique(char *buf, char *prefix, char *key, char *val,
599 anchor = end = strstr(buf, key);
600 /* try to find exact match string in @buf */
601 while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
604 if (anchor == NULL || strlen(key) != len ||
605 strncmp(anchor, key, len) != 0) {
607 strscat(buf, prefix, maxbuflen);
609 strscat(buf, key, maxbuflen);
611 strscat(buf, "=", maxbuflen);
612 strscat(buf, val, maxbuflen);
617 static void enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
618 size_t maxbuflen, int user_spec)
620 if (IS_OST(&mop->mo_ldd)) {
621 append_unique(anchor, user_spec ? "," : " -O ",
622 "extents", NULL, sizeof(mop->mo_mkfsopts));
623 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
624 } else if (IS_MDT(&mop->mo_ldd)) {
625 append_unique(anchor, user_spec ? "," : " -O ",
626 "dirdata", NULL, maxbuflen);
627 append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
628 append_unique(anchor, ",", "^extents", NULL, maxbuflen);
630 append_unique(anchor, user_spec ? "," : " -O ",
631 "uninit_bg", NULL, maxbuflen);
634 /* Multiple mount protection enabled only if failover node specified */
636 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
637 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
639 disp_old_e2fsprogs_msg("mmp", 1);
642 /* Allow more than 65000 subdirectories */
643 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
644 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
646 /* The following options are only valid for ext4-based ldiskfs.
647 * If --backfstype=ext3 is specified, do not enable them. */
648 if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
651 /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
652 if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
653 append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
655 /* Enable large block addresses if the LUN is over 2^32 blocks. */
656 if (mop->mo_device_sz / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
657 is_e2fsprogs_feature_supp("-O 64bit") == 0)
658 append_unique(anchor, ",", "64bit", NULL, maxbuflen);
660 /* Cluster inode/block bitmaps and inode table for more efficient IO.
661 * Align the flex groups on a 1MB boundary for better performance. */
662 /* This -O feature needs to go last, since it adds the "-G" option. */
663 if (is_e2fsprogs_feature_supp("-O flex_bg") == 0) {
666 append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
668 if (IS_OST(&mop->mo_ldd)) {
669 snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
670 (1 << 20) / L_BLOCK_SIZE);
671 strscat(anchor, tmp_buf, maxbuflen);
674 /* Don't add any more "-O" options here, see last comment above */
678 * moveopts_to_end: find the option string, move remaining strings to
679 * where option string starts, and append the option
681 * @start: where the option string starts before the move
682 * RETURN: where the option string starts after the move
684 static char *moveopts_to_end(char *start)
689 /* skip whitespace before options */
694 /* find end of option characters */
695 while (*end != ' ' && *end != '\0')
699 strncpy(save, start, end - start);
700 save[end - start] = '\0';
702 /* move remaining options up front */
704 memmove(start, end, strlen(end));
705 *(start + strlen(end)) = '\0';
707 /* append the specified options */
708 if (*(start + strlen(start) - 1) != ' ')
710 idx = start + strlen(start);
716 /* Build fs according to type */
717 int make_lustre_backfs(struct mkfs_opts *mop)
719 __u64 device_sz = mop->mo_device_sz, block_count = 0;
720 char mkfs_cmd[PATH_MAX];
724 int ret = 0, ext_opts = 0;
727 if (!(mop->mo_flags & MO_IS_LOOP)) {
728 mop->mo_device_sz = get_device_size(mop->mo_device);
730 if (mop->mo_device_sz == 0)
733 /* Compare to real size */
734 if (device_sz == 0 || device_sz > mop->mo_device_sz)
735 device_sz = mop->mo_device_sz;
737 mop->mo_device_sz = device_sz;
740 if (mop->mo_device_sz != 0) {
741 if (mop->mo_device_sz < 8096){
742 fprintf(stderr, "%s: size of filesystem must be larger "
743 "than 8MB, but is set to %lldKB\n",
744 progname, (long long)mop->mo_device_sz);
747 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
748 /* If the LUN size is just over 2^32 blocks, limit the
749 * filesystem size to 2^32-1 blocks to avoid problems with
750 * ldiskfs/mkfs not handling this size. Bug 22906 */
751 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
752 block_count = 0xffffffffULL;
755 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
756 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
757 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
760 /* Journal size in MB */
761 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
762 /* Choose our own default journal size */
763 long journal_sz = 0, max_sz;
764 if (device_sz > 1024 * 1024) /* 1GB */
765 journal_sz = (device_sz / 102400) * 4;
766 /* cap journal size at 1GB */
767 if (journal_sz > 1024L)
770 max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
771 if (journal_sz > max_sz)
774 sprintf(buf, " -J size=%ld", journal_sz);
775 strscat(mop->mo_mkfsopts, buf,
776 sizeof(mop->mo_mkfsopts));
780 /* Inode size (for extended attributes). The LOV EA size is
781 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
782 * and we want some margin above that for ACLs, other EAs... */
783 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
784 if (IS_MDT(&mop->mo_ldd)) {
785 if (mop->mo_stripe_count > 72)
786 inode_size = 512; /* bz 7241 */
787 /* see also "-i" below for EA blocks */
788 else if (mop->mo_stripe_count > 32)
790 else if (mop->mo_stripe_count > 10)
794 } else if (IS_OST(&mop->mo_ldd)) {
795 /* We store MDS FID and OST objid in EA on OST
796 * we need to make inode bigger as well. */
800 if (inode_size > 0) {
801 sprintf(buf, " -I %ld", inode_size);
802 strscat(mop->mo_mkfsopts, buf,
803 sizeof(mop->mo_mkfsopts));
807 /* Bytes_per_inode: disk size / num inodes */
808 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
809 strstr(mop->mo_mkfsopts, "-N") == NULL) {
810 long bytes_per_inode = 0;
812 /* Allocate more inodes on MDT devices. There is
813 * no data stored on the MDT, and very little extra
814 * metadata beyond the inode. It could go down as
815 * low as 1024 bytes, but this is conservative.
816 * Account for external EA blocks for wide striping. */
817 if (IS_MDT(&mop->mo_ldd)) {
818 bytes_per_inode = inode_size + 1536;
820 if (mop->mo_stripe_count > 72) {
821 int extra = mop->mo_stripe_count * 24;
822 extra = ((extra - 1) | 4095) + 1;
823 bytes_per_inode += extra;
827 /* Allocate fewer inodes on large OST devices. Most
828 * filesystems can be much more aggressive than even
829 * this, but it is impossible to know in advance. */
830 if (IS_OST(&mop->mo_ldd)) {
831 /* OST > 16TB assume average file size 1MB */
832 if (device_sz > (16ULL << 30))
833 bytes_per_inode = 1024 * 1024;
834 /* OST > 4TB assume average file size 512kB */
835 else if (device_sz > (4ULL << 30))
836 bytes_per_inode = 512 * 1024;
837 /* OST > 1TB assume average file size 256kB */
838 else if (device_sz > (1ULL << 30))
839 bytes_per_inode = 256 * 1024;
840 /* OST > 10GB assume average file size 64kB,
841 * plus a bit so that inodes will fit into a
842 * 256x flex_bg without overflowing */
843 else if (device_sz > (10ULL << 20))
844 bytes_per_inode = 69905;
847 if (bytes_per_inode > 0) {
848 sprintf(buf, " -i %ld", bytes_per_inode);
849 strscat(mop->mo_mkfsopts, buf,
850 sizeof(mop->mo_mkfsopts));
855 strscat(mop->mo_mkfsopts, " -q",
856 sizeof(mop->mo_mkfsopts));
859 /* start handle -O mkfs options */
860 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
861 if (strstr(start + 2, "-O") != NULL) {
863 "%s: don't specify multiple -O options\n",
867 start = moveopts_to_end(start);
868 maxbuflen = sizeof(mop->mo_mkfsopts) -
869 (start - mop->mo_mkfsopts) - strlen(start);
870 enable_default_ext4_features(mop, start, maxbuflen, 1);
872 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
873 maxbuflen = sizeof(mop->mo_mkfsopts) -
874 strlen(mop->mo_mkfsopts);
875 enable_default_ext4_features(mop, start, maxbuflen, 0);
877 /* end handle -O mkfs options */
879 /* start handle -E mkfs options */
880 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
881 if (strstr(start + 2, "-E") != NULL) {
883 "%s: don't specify multiple -E options\n",
887 start = moveopts_to_end(start);
888 maxbuflen = sizeof(mop->mo_mkfsopts) -
889 (start - mop->mo_mkfsopts) - strlen(start);
892 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
893 maxbuflen = sizeof(mop->mo_mkfsopts) -
894 strlen(mop->mo_mkfsopts);
897 /* In order to align the filesystem metadata on 1MB boundaries,
898 * give a resize value that will reserve a power-of-two group
899 * descriptor blocks, but leave one block for the superblock.
900 * Only useful for filesystems with < 2^32 blocks due to resize
902 if (IS_OST(&mop->mo_ldd) && mop->mo_device_sz > 100 * 1024 &&
903 mop->mo_device_sz * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
904 unsigned group_blocks = L_BLOCK_SIZE * 8;
905 unsigned desc_per_block = L_BLOCK_SIZE / 32;
906 unsigned resize_blks;
908 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
909 snprintf(buf, sizeof(buf), "%u", resize_blks);
910 append_unique(start, ext_opts ? "," : " -E ",
911 "resize", buf, maxbuflen);
915 /* Avoid zeroing out the full journal - speeds up mkfs */
916 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
917 append_unique(start, ext_opts ? "," : " -E ",
918 "lazy_journal_init", NULL, maxbuflen);
919 /* end handle -E mkfs options */
921 /* Allow reformat of full devices (as opposed to
922 partitions.) We already checked for mounted dev. */
923 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
925 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
926 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
927 mop->mo_ldd.ldd_svname);
929 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
930 progname, mop->mo_ldd.ldd_mount_type,
931 MT_STR(&mop->mo_ldd));
935 /* For loop device format the dev, not the filename */
936 dev = mop->mo_device;
937 if (mop->mo_flags & MO_IS_LOOP)
938 dev = mop->mo_loopdev;
940 vprint("formatting backing filesystem %s on %s\n",
941 MT_STR(&mop->mo_ldd), dev);
942 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
943 vprint("\t4k blocks "LPU64"\n", block_count);
944 vprint("\toptions %s\n", mop->mo_mkfsopts);
946 /* mkfs_cmd's trailing space is important! */
947 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
948 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
949 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
950 if (block_count != 0) {
951 sprintf(buf, " "LPU64, block_count);
952 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
955 vprint("mkfs_cmd = %s\n", mkfs_cmd);
956 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
959 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
964 /* ==================== Lustre config functions =============*/
966 void print_ldd(char *str, struct lustre_disk_data *ldd)
968 printf("\n %s:\n", str);
969 printf("Target: %s\n", ldd->ldd_svname);
970 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
971 printf("Index: unassigned\n");
973 printf("Index: %d\n", ldd->ldd_svindex);
974 if (ldd->ldd_uuid[0])
975 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
976 printf("Lustre FS: %s\n", ldd->ldd_fsname);
977 printf("Mount type: %s\n", MT_STR(ldd));
978 printf("Flags: %#x\n", ldd->ldd_flags);
979 printf(" (%s%s%s%s%s%s%s%s%s%s)\n",
980 IS_MDT(ldd) ? "MDT ":"",
981 IS_OST(ldd) ? "OST ":"",
982 IS_MGS(ldd) ? "MGS ":"",
983 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
984 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
985 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
986 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
987 ldd->ldd_flags & LDD_F_IAM_DIR ? "IAM_dir_format ":"",
988 ldd->ldd_flags & LDD_F_NO_PRIMNODE? "no_primnode ":"",
989 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
990 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
991 printf("Parameters:%s\n", ldd->ldd_params);
992 if (ldd->ldd_userdata[0])
993 printf("Comment: %s\n", ldd->ldd_userdata);
997 static int touch_file(char *filename)
1001 if (filename == NULL) {
1005 fd = open(filename, O_CREAT | O_TRUNC, 0600);
1014 /* keep it less than LL_FID_NAMELEN */
1015 #define DUMMY_FILE_NAME_LEN 25
1016 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
1018 /* Need to add these many entries to this directory to make HTREE dir. */
1019 #define MIN_ENTRIES_REQ_FOR_HTREE ((L_BLOCK_SIZE / EXT3_DIRENT_SIZE))
1021 static int add_dummy_files(char *dir)
1023 char fpname[PATH_MAX];
1027 for (i = 0; i < MIN_ENTRIES_REQ_FOR_HTREE; i++) {
1028 snprintf(fpname, PATH_MAX, "%s/%0*d", dir,
1029 DUMMY_FILE_NAME_LEN, i);
1031 rc = touch_file(fpname);
1032 if (rc && rc != -EEXIST) {
1034 "%s: Can't create dummy file %s: %s\n",
1035 progname, fpname , strerror(errno));
1042 static int __l_mkdir(char * filepnm, int mode , struct mkfs_opts *mop)
1046 ret = mkdir(filepnm, mode);
1047 if (ret && ret != -EEXIST)
1050 /* IAM mode supports ext3 directories of HTREE type only. So add dummy
1051 * entries to new directory to create htree type of container for
1052 * this directory. */
1053 if (mop->mo_ldd.ldd_flags & LDD_F_IAM_DIR)
1054 return add_dummy_files(filepnm);
1058 /* Write the server config files */
1059 int write_local_files(struct mkfs_opts *mop)
1061 char mntpt[] = "/tmp/mntXXXXXX";
1068 /* Mount this device temporarily in order to write these files */
1069 if (!mkdtemp(mntpt)) {
1070 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
1071 progname, mntpt, strerror(errno));
1075 dev = mop->mo_device;
1076 if (mop->mo_flags & MO_IS_LOOP)
1077 dev = mop->mo_loopdev;
1079 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
1080 mop->mo_ldd.ldd_mount_opts);
1082 fprintf(stderr, "%s: Unable to mount %s: %s\n",
1083 progname, dev, strerror(errno));
1085 if (errno == ENODEV) {
1086 fprintf(stderr, "Is the %s module available?\n",
1087 MT_STR(&mop->mo_ldd));
1092 /* Set up initial directories */
1093 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
1094 ret = __l_mkdir(filepnm, 0777, mop);
1095 if ((ret != 0) && (errno != EEXIST)) {
1096 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
1097 progname, filepnm, strerror(errno));
1099 } else if (errno == EEXIST) {
1103 /* Save the persistent mount data into a file. Lustre must pre-read
1104 this file to get the real mount options. */
1105 vprint("Writing %s\n", MOUNT_DATA_FILE);
1106 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
1107 filep = fopen(filepnm, "w");
1109 fprintf(stderr, "%s: Unable to create %s file: %s\n",
1110 progname, filepnm, strerror(errno));
1113 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
1114 if (num < 1 && ferror(filep)) {
1115 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
1116 progname, filepnm, strerror(errno));
1122 /* Check for upgrade */
1123 if ((mop->mo_ldd.ldd_flags & (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS))
1124 == (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS)) {
1127 int cmdsz = sizeof(cmd);
1128 vprint("Copying old logs\n");
1130 /* Copy the old client log to fsname-client */
1131 sprintf(filepnm, "%s/%s/%s-client",
1132 mntpt, MOUNT_CONFIGS_DIR, mop->mo_ldd.ldd_fsname);
1133 snprintf(cmd, cmdsz, "cp %s/%s/client %s", mntpt, MDT_LOGS_DIR,
1135 ret = run_command(cmd, cmdsz);
1137 fprintf(stderr, "%s: Can't copy 1.4 config %s/client "
1138 "(%d)\n", progname, MDT_LOGS_DIR, ret);
1139 fprintf(stderr, "mount -t ldiskfs %s somewhere, "
1140 "find the client log for fs %s and "
1141 "copy it manually into %s/%s-client, "
1144 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
1145 mop->mo_ldd.ldd_fsname);
1149 /* We need to use the old mdt log because otherwise mdt won't
1150 have complete lov if old clients connect before all
1152 /* Copy the old mdt log to fsname-MDT0000 (get old
1153 name from mdt_UUID) */
1155 strscpy(filepnm, (char *)mop->mo_ldd.ldd_uuid, sizeof(filepnm));
1156 term = strstr(filepnm, "_UUID");
1159 snprintf(cmd, cmdsz, "cp %s/%s/%s %s/%s/%s",
1160 mntpt, MDT_LOGS_DIR, filepnm,
1161 mntpt, MOUNT_CONFIGS_DIR,
1162 mop->mo_ldd.ldd_svname);
1163 ret = run_command(cmd, cmdsz);
1166 fprintf(stderr, "%s: Can't copy 1.4 config %s/%s "
1167 "(%d)\n", progname, MDT_LOGS_DIR, filepnm, ret);
1168 fprintf(stderr, "mount -t ext3 %s somewhere, "
1169 "find the MDT log for fs %s and "
1170 "copy it manually into %s/%s, "
1173 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
1174 mop->mo_ldd.ldd_svname);
1179 /* end COMPAT_146 */
1188 int read_local_files(struct mkfs_opts *mop)
1190 char tmpdir[] = "/tmp/dirXXXXXX";
1196 int cmdsz = sizeof(cmd);
1198 /* Make a temporary directory to hold Lustre data files. */
1199 if (!mkdtemp(tmpdir)) {
1200 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
1201 progname, tmpdir, strerror(errno));
1205 dev = mop->mo_device;
1207 /* TODO: it's worth observing the get_mountdata() function that is
1208 in mount_utils.c for getting the mountdata out of the
1211 /* Construct debugfs command line. */
1212 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
1213 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
1215 ret = run_command(cmd, cmdsz);
1217 verrprint("%s: Unable to dump %s dir (%d)\n",
1218 progname, MOUNT_CONFIGS_DIR, ret);
1220 sprintf(filepnm, "%s/mountdata", tmpdir);
1221 filep = fopen(filepnm, "r");
1224 vprint("Reading %s\n", MOUNT_DATA_FILE);
1225 num_read = fread(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
1226 if (num_read < 1 && ferror(filep)) {
1227 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
1228 progname, filepnm, strerror(errno));
1233 /* Try to read pre-1.6 config from last_rcvd */
1234 struct lr_server_data lsd;
1235 verrprint("%s: Unable to read %d.%d config %s.\n",
1236 progname, LUSTRE_MAJOR, LUSTRE_MINOR, filepnm);
1238 verrprint("Trying 1.4 config from last_rcvd\n");
1239 sprintf(filepnm, "%s/%s", tmpdir, LAST_RCVD);
1241 /* Construct debugfs command line. */
1242 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s' %s",
1243 DEBUGFS, LAST_RCVD, filepnm, dev);
1245 ret = run_command(cmd, cmdsz);
1247 fprintf(stderr, "%s: Unable to dump %s file (%d)\n",
1248 progname, LAST_RCVD, ret);
1252 filep = fopen(filepnm, "r");
1254 fprintf(stderr, "%s: Unable to open %s: %s\n",
1255 progname, filepnm, strerror(errno));
1257 verrprint("Contents of %s:\n", tmpdir);
1259 snprintf(cmd, cmdsz, "ls -l %s/", tmpdir);
1260 run_command(cmd, cmdsz);
1261 verrprint("Contents of disk:\n");
1262 snprintf(cmd, cmdsz, "%s -c -R 'ls -l /' %s",
1264 run_command(cmd, cmdsz);
1268 vprint("Reading %s\n", LAST_RCVD);
1269 ret = fread(&lsd, 1, sizeof(lsd), filep);
1270 if (ret < sizeof(lsd)) {
1271 fprintf(stderr, "%s: Short read (%d of %d)\n",
1272 progname, ret, (int)sizeof(lsd));
1273 ret = ferror(filep);
1277 vprint("Feature compat=%x, incompat=%x\n",
1278 lsd.lsd_feature_compat, lsd.lsd_feature_incompat);
1280 if ((lsd.lsd_feature_compat & OBD_COMPAT_OST) ||
1281 (lsd.lsd_feature_incompat & OBD_INCOMPAT_OST)) {
1282 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
1283 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
1284 } else if ((lsd.lsd_feature_compat & OBD_COMPAT_MDT) ||
1285 (lsd.lsd_feature_incompat & OBD_INCOMPAT_MDT)) {
1286 /* We must co-locate so mgs can see old logs.
1287 If user doesn't want this, they can copy the old
1288 logs manually and re-tunefs. */
1289 mop->mo_ldd.ldd_flags =
1290 LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_MGS;
1291 mop->mo_ldd.ldd_svindex = lsd.lsd_mdt_index;
1293 /* If neither is set, we're pre-1.4.6, make a guess. */
1294 /* Construct debugfs command line. */
1295 snprintf(cmd, cmdsz, "%s -c -R 'rdump /%s %s' %s",
1296 DEBUGFS, MDT_LOGS_DIR, tmpdir, dev);
1297 run_command(cmd, cmdsz);
1299 sprintf(filepnm, "%s/%s", tmpdir, MDT_LOGS_DIR);
1300 if (lsd.lsd_ost_index > 0) {
1301 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
1302 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
1304 /* If there's a LOGS dir, it's an MDT */
1305 if ((ret = access(filepnm, F_OK)) == 0) {
1306 mop->mo_ldd.ldd_flags =
1309 /* Old MDT's are always index 0
1311 mop->mo_ldd.ldd_svindex = 0;
1313 /* The index may not be correct */
1314 mop->mo_ldd.ldd_flags =
1315 LDD_F_SV_TYPE_OST | LDD_F_NEED_INDEX;
1316 verrprint("OST with unknown index\n");
1322 memcpy(mop->mo_ldd.ldd_uuid, lsd.lsd_uuid,
1323 sizeof(mop->mo_ldd.ldd_uuid));
1324 mop->mo_ldd.ldd_flags |= LDD_F_UPGRADE14;
1326 /* end COMPAT_146 */
1331 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
1332 run_command(cmd, cmdsz);
1334 verrprint("Failed to read old data (%d)\n", ret);
1339 void set_defaults(struct mkfs_opts *mop)
1341 mop->mo_ldd.ldd_magic = LDD_MAGIC;
1342 mop->mo_ldd.ldd_config_ver = 1;
1343 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
1344 mop->mo_mgs_failnodes = 0;
1345 strcpy(mop->mo_ldd.ldd_fsname, "lustre");
1346 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
1348 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
1349 mop->mo_stripe_count = 1;
1352 static inline void badopt(const char *opt, char *type)
1354 fprintf(stderr, "%s: '--%s' only valid for %s\n",
1355 progname, opt, type);
1359 static int add_param(char *buf, char *key, char *val)
1361 int end = sizeof(((struct lustre_disk_data *)0)->ldd_params);
1362 int start = strlen(buf);
1366 keylen = strlen(key);
1367 if (start + 1 + keylen + strlen(val) >= end) {
1368 fprintf(stderr, "%s: params are too long-\n%s %s%s\n",
1369 progname, buf, key ? key : "", val);
1373 sprintf(buf + start, " %s%s", key ? key : "", val);
1377 /* from mount_lustre */
1378 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
1379 #define MAXNIDSTR 1024
1380 static char *convert_hostnames(char *s1)
1382 char *converted, *s2 = 0, *c, *end, sep;
1383 int left = MAXNIDSTR;
1386 converted = malloc(left);
1387 if (converted == NULL) {
1391 end = s1 + strlen(s1);
1393 while ((left > 0) && (s1 < end)) {
1394 s2 = strpbrk(s1, ",:");
1399 nid = libcfs_str2nid(s1);
1401 if (nid == LNET_NID_ANY) {
1402 fprintf(stderr, "%s: Can't parse NID '%s'\n",
1407 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
1408 strlen("127.0.0.1")) == 0) {
1409 fprintf(stderr, "%s: The NID '%s' resolves to the "
1410 "loopback address '%s'. Lustre requires a "
1411 "non-loopback address.\n",
1412 progname, s1, libcfs_nid2str(nid));
1417 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
1418 left = converted + MAXNIDSTR - c;
1424 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
1427 static struct option long_opt[] = {
1428 {"iam-dir", 0, 0, 'a'},
1429 {"backfstype", 1, 0, 'b'},
1430 {"stripe-count-hint", 1, 0, 'c'},
1431 {"comment", 1, 0, 'u'},
1432 {"configdev", 1, 0, 'C'},
1433 {"device-size", 1, 0, 'd'},
1434 {"dryrun", 0, 0, 'n'},
1435 {"erase-params", 0, 0, 'e'},
1436 {"failnode", 1, 0, 'f'},
1437 {"failover", 1, 0, 'f'},
1439 {"help", 0, 0, 'h'},
1440 {"index", 1, 0, 'i'},
1441 {"mkfsoptions", 1, 0, 'k'},
1442 {"mgsnode", 1, 0, 'm'},
1443 {"mgsnid", 1, 0, 'm'},
1445 {"fsname",1, 0, 'L'},
1446 {"noformat", 0, 0, 'n'},
1447 {"nomgs", 0, 0, 'N'},
1448 {"mountfsoptions", 1, 0, 'o'},
1450 {"param", 1, 0, 'p'},
1451 {"print", 0, 0, 'n'},
1452 {"quiet", 0, 0, 'q'},
1453 {"reformat", 0, 0, 'r'},
1454 {"servicenode", 1, 0, 's'},
1455 {"verbose", 0, 0, 'v'},
1456 {"writeconf", 0, 0, 'w'},
1457 {"upgrade_to_18", 0, 0, 'U'},
1458 {"network", 1, 0, 't'},
1461 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqrs:t:Uu:vw";
1464 int failnode_set = 0, servicenode_set = 0;
1466 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
1470 if (IS_MDT(&mop->mo_ldd))
1471 mop->mo_ldd.ldd_flags |= LDD_F_IAM_DIR;
1476 while (i < LDD_MT_LAST) {
1477 if (strcmp(optarg, mt_str(i)) == 0) {
1478 mop->mo_ldd.ldd_mount_type = i;
1486 if (IS_MDT(&mop->mo_ldd)) {
1487 int stripe_count = atol(optarg);
1488 if (stripe_count <= 0) {
1489 fprintf(stderr, "%s: bad stripe count "
1490 "%d\n", progname, stripe_count);
1493 mop->mo_stripe_count = stripe_count;
1495 badopt(long_opt[longidx].name, "MDT");
1499 case 'C': /* Configdev */
1501 printf("Configdev not implemented\n");
1504 mop->mo_device_sz = atol(optarg);
1507 mop->mo_ldd.ldd_params[0] = '\0';
1508 /* Must update the mgs logs */
1509 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1515 if ((opt == 'f' && servicenode_set)
1516 || (opt == 's' && failnode_set)) {
1517 fprintf(stderr, "%s: %s cannot use with --%s\n",
1518 progname, long_opt[longidx].name,
1519 opt == 'f' ? "servicenode" : "failnode");
1523 nids = convert_hostnames(optarg);
1526 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
1531 /* Must update the mgs logs */
1532 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1536 mop->mo_ldd.ldd_flags |= LDD_F_NO_PRIMNODE;
1537 servicenode_set = 1;
1543 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
1549 if (!(mop->mo_ldd.ldd_flags &
1550 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1551 LDD_F_WRITECONF))) {
1552 fprintf(stderr, "%s: cannot change the index of"
1553 " a registered target\n", progname);
1556 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
1557 mop->mo_ldd.ldd_svindex = atol(optarg);
1558 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
1560 badopt(long_opt[longidx].name, "MDT,OST");
1565 strscpy(mop->mo_mkfsopts, optarg,
1566 sizeof(mop->mo_mkfsopts));
1570 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
1571 (!(mop->mo_ldd.ldd_flags &
1572 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1573 LDD_F_WRITECONF)))) {
1574 fprintf(stderr, "%s: cannot change the name of"
1575 " a registered target\n", progname);
1578 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
1579 fprintf(stderr, "%s: filesystem name must be "
1580 "1-8 chars\n", progname);
1583 if ((tmp = strpbrk(optarg, "/:"))) {
1584 fprintf(stderr, "%s: char '%c' not allowed in "
1585 "filesystem name\n", progname, *tmp);
1588 strscpy(mop->mo_ldd.ldd_fsname, optarg,
1589 sizeof(mop->mo_ldd.ldd_fsname));
1593 char *nids = convert_hostnames(optarg);
1596 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
1601 mop->mo_mgs_failnodes++;
1605 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
1611 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
1614 *mountopts = optarg;
1617 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
1620 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg);
1623 /* Must update the mgs logs */
1624 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1630 mop->mo_flags |= MO_FORCEFORMAT;
1633 if (!IS_MDT(&mop->mo_ldd) && !IS_OST(&mop->mo_ldd)) {
1634 badopt(long_opt[longidx].name, "MDT,OST");
1641 rc = add_param(mop->mo_ldd.ldd_params,
1642 PARAM_NETWORK, optarg);
1645 /* Must update the mgs logs */
1646 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1649 strscpy(mop->mo_ldd.ldd_userdata, optarg,
1650 sizeof(mop->mo_ldd.ldd_userdata));
1656 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
1664 fprintf(stderr, "Unknown option '%c'\n", opt);
1670 /* Last arg is device */
1671 if (optind != argc - 1) {
1673 fprintf(stderr, "Bad argument: %s\n", argv[optind]);
1677 /* single argument: <device> */
1684 /* Search for opt in mntlist, returning true if found.
1686 static int in_mntlist(char *opt, char *mntlist)
1688 char *ml, *mlp, *item, *ctx = NULL;
1690 if (!(ml = strdup(mntlist))) {
1691 fprintf(stderr, "%s: out of memory\n", progname);
1695 while ((item = strtok_r(mlp, ",", &ctx))) {
1696 if (!strcmp(opt, item))
1701 return (item != NULL);
1704 /* Issue a message on stderr for every item in wanted_mountopts that is not
1705 * present in mountopts. The justwarn boolean toggles between error and
1706 * warning message. Return an error count.
1708 static int check_mountfsoptions(char *mountopts, char *wanted_mountopts,
1711 char *ml, *mlp, *item, *ctx = NULL;
1714 if (!(ml = strdup(wanted_mountopts))) {
1715 fprintf(stderr, "%s: out of memory\n", progname);
1719 while ((item = strtok_r(mlp, ",", &ctx))) {
1720 if (!in_mntlist(item, mountopts)) {
1721 fprintf(stderr, "%s: %s mount option `%s' is missing\n",
1722 progname, justwarn ? "Warning: default"
1723 : "Error: mandatory", item);
1732 /* Trim embedded white space, leading and trailing commas from string s.
1734 static void trim_mountfsoptions(char *s)
1740 memmove(p, p + 1, strlen(p + 1) + 1);
1747 memmove(&s[0], &s[1], strlen(&s[1]) + 1);
1749 p = s + strlen(s) - 1;
1750 while (p >= s && *p == ',')
1754 int main(int argc, char *const argv[])
1756 struct mkfs_opts mop;
1757 struct lustre_disk_data *ldd;
1758 char *mountopts = NULL;
1759 char always_mountopts[512] = "";
1760 char default_mountopts[512] = "";
1763 if ((progname = strrchr(argv[0], '/')) != NULL)
1768 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
1773 memset(&mop, 0, sizeof(mop));
1776 /* device is last arg */
1777 strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
1779 /* Are we using a loop device? */
1780 ret = is_block(mop.mo_device);
1786 mop.mo_flags |= MO_IS_LOOP;
1789 /* For tunefs, we must read in the old values before parsing any
1792 /* Check whether the disk has already been formatted by mkfs.lustre */
1793 ret = is_lustre_target(&mop);
1796 fprintf(stderr, "Device %s has not been formatted with "
1797 "mkfs.lustre\n", mop.mo_device);
1802 ret = read_local_files(&mop);
1805 fprintf(stderr, "Failed to read previous Lustre data from %s "
1806 "(%d)\n", mop.mo_device, ret);
1809 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
1810 mop.mo_mgs_failnodes++;
1813 print_ldd("Read previous values", &(mop.mo_ldd));
1816 ret = parse_opts(argc, argv, &mop, &mountopts);
1822 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
1824 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
1829 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
1831 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
1836 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
1837 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
1839 fprintf(stderr, "Can't find the target index, "
1840 "specify with --index\n");
1846 * Comment out these 2 checks temporarily, since for multi-MDSes
1847 * in single node only 1 mds node could have mgs service
1849 if (IS_MDT(ldd) && !IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1850 verrprint("No management node specified, adding MGS to this "
1852 ldd->ldd_flags |= LDD_F_SV_TYPE_MGS;
1854 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1857 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
1859 fprintf(stderr, "Must specify --mgsnode=\n");
1865 /* These are the permanent mount options (always included) */
1866 switch (ldd->ldd_mount_type) {
1868 case LDD_MT_LDISKFS:
1869 case LDD_MT_LDISKFS2:
1870 strscat(default_mountopts, ",errors=remount-ro",
1871 sizeof(default_mountopts));
1872 if (IS_MDT(ldd) || IS_MGS(ldd))
1873 strscat(always_mountopts, ",user_xattr",
1874 sizeof(always_mountopts));
1875 /* NB: Files created while extents are enabled can only be read
1876 * if mounted using the ext4 or ldiskfs filesystem type. */
1878 (ldd->ldd_mount_type == LDD_MT_LDISKFS ||
1879 ldd->ldd_mount_type == LDD_MT_LDISKFS2)) {
1880 strscat(default_mountopts, ",extents,mballoc",
1881 sizeof(default_mountopts));
1886 fprintf(stderr, "unknown fs type %d '%s'\n",
1887 ldd->ldd_mount_type,
1894 trim_mountfsoptions(mountopts);
1895 (void)check_mountfsoptions(mountopts, default_mountopts, 1);
1896 if (check_mountfsoptions(mountopts, always_mountopts, 0)) {
1900 sprintf(ldd->ldd_mount_opts, "%s", mountopts);
1903 if (ldd->ldd_mount_opts[0] == 0)
1904 /* use the defaults unless old opts exist */
1907 sprintf(ldd->ldd_mount_opts, "%s%s",
1908 always_mountopts, default_mountopts);
1909 trim_mountfsoptions(ldd->ldd_mount_opts);
1913 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
1914 ldd->ldd_fsname, ldd->ldd_svname);
1917 print_ldd("Permanent disk data", ldd);
1920 printf("exiting before disk write.\n");
1924 if (check_mtab_entry(mop.mo_device))
1927 /* Create the loopback file */
1928 if (mop.mo_flags & MO_IS_LOOP) {
1929 ret = access(mop.mo_device, F_OK);
1932 #ifndef TUNEFS /* mkfs.lustre */
1933 /* Reformat the loopback file */
1934 if (ret || (mop.mo_flags & MO_FORCEFORMAT)) {
1935 ret = loop_format(&mop);
1941 ret = loop_setup(&mop);
1944 fprintf(stderr, "Loop device setup for %s failed: %s\n",
1945 mop.mo_device, strerror(ret));
1950 #ifndef TUNEFS /* mkfs.lustre */
1951 /* Check whether the disk has already been formatted by mkfs.lustre */
1952 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
1953 ret = is_lustre_target(&mop);
1956 fprintf(stderr, "Device %s was previously formatted "
1957 "for lustre. Use --reformat to reformat it, "
1958 "or tunefs.lustre to modify.\n",
1964 /* Format the backing filesystem */
1965 ret = make_lustre_backfs(&mop);
1968 fprintf(stderr, "mkfs failed %d\n", ret);
1973 /* Write our config files */
1974 ret = write_local_files(&mop);
1977 fprintf(stderr, "failed to write local files\n");
1984 /* Fix any crazy return values from system() */
1985 if (ret && ((ret & 255) == 0))
1988 verrprint("%s: exiting with %d (%s)\n",
1989 progname, ret, strerror(ret));