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);
629 append_unique(anchor, user_spec ? "," : " -O ",
630 "uninit_bg", NULL, maxbuflen);
633 /* Multiple mount protection enabled only if failover node specified */
635 if (is_e2fsprogs_feature_supp("-O mmp") == 0)
636 append_unique(anchor, ",", "mmp", NULL, maxbuflen);
638 disp_old_e2fsprogs_msg("mmp", 1);
641 /* Allow more than 65000 subdirectories */
642 if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
643 append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
645 #ifdef HAVE_EXT4_LDISKFS
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 */
679 * moveopts_to_end: find the option string, move remaining strings to
680 * where option string starts, and append the option
682 * @start: where the option string starts before the move
683 * RETURN: where the option string starts after the move
685 static char *moveopts_to_end(char *start)
690 /* skip whitespace before options */
695 /* find end of option characters */
696 while (*end != ' ' && *end != '\0')
700 strncpy(save, start, end - start);
701 save[end - start] = '\0';
703 /* move remaining options up front */
705 memmove(start, end, strlen(end));
706 *(start + strlen(end)) = '\0';
708 /* append the specified options */
709 if (*(start + strlen(start) - 1) != ' ')
711 idx = start + strlen(start);
717 /* Build fs according to type */
718 int make_lustre_backfs(struct mkfs_opts *mop)
720 __u64 device_sz = mop->mo_device_sz, block_count = 0;
721 char mkfs_cmd[PATH_MAX];
725 int ret = 0, ext_opts = 0;
728 if (!(mop->mo_flags & MO_IS_LOOP)) {
729 mop->mo_device_sz = get_device_size(mop->mo_device);
731 if (mop->mo_device_sz == 0)
734 /* Compare to real size */
735 if (device_sz == 0 || device_sz > mop->mo_device_sz)
736 device_sz = mop->mo_device_sz;
738 mop->mo_device_sz = device_sz;
741 if (mop->mo_device_sz != 0) {
742 if (mop->mo_device_sz < 8096){
743 fprintf(stderr, "%s: size of filesystem must be larger "
744 "than 8MB, but is set to %lldKB\n",
745 progname, (long long)mop->mo_device_sz);
748 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
749 /* If the LUN size is just over 2^32 blocks, limit the
750 * filesystem size to 2^32-1 blocks to avoid problems with
751 * ldiskfs/mkfs not handling this size. Bug 22906 */
752 if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
753 block_count = 0xffffffffULL;
756 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
757 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
758 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
761 /* Journal size in MB */
762 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
763 /* Choose our own default journal size */
764 long journal_sz = 0, max_sz;
765 if (device_sz > 1024 * 1024) /* 1GB */
766 journal_sz = (device_sz / 102400) * 4;
767 /* cap journal size at 1GB */
768 if (journal_sz > 1024L)
771 max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
772 if (journal_sz > max_sz)
775 sprintf(buf, " -J size=%ld", journal_sz);
776 strscat(mop->mo_mkfsopts, buf,
777 sizeof(mop->mo_mkfsopts));
781 /* Inode size (for extended attributes). The LOV EA size is
782 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
783 * and we want some margin above that for ACLs, other EAs... */
784 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
785 if (IS_MDT(&mop->mo_ldd)) {
786 if (mop->mo_stripe_count > 72)
787 inode_size = 512; /* bz 7241 */
788 /* see also "-i" below for EA blocks */
789 else if (mop->mo_stripe_count > 32)
791 else if (mop->mo_stripe_count > 10)
795 } else if (IS_OST(&mop->mo_ldd)) {
796 /* We store MDS FID and OST objid in EA on OST
797 * we need to make inode bigger as well. */
801 if (inode_size > 0) {
802 sprintf(buf, " -I %ld", inode_size);
803 strscat(mop->mo_mkfsopts, buf,
804 sizeof(mop->mo_mkfsopts));
808 /* Bytes_per_inode: disk size / num inodes */
809 if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
810 strstr(mop->mo_mkfsopts, "-N") == NULL) {
811 long bytes_per_inode = 0;
813 /* Allocate more inodes on MDT devices. There is
814 * no data stored on the MDT, and very little extra
815 * metadata beyond the inode. It could go down as
816 * low as 1024 bytes, but this is conservative.
817 * Account for external EA blocks for wide striping. */
818 if (IS_MDT(&mop->mo_ldd)) {
819 bytes_per_inode = inode_size + 1536;
821 if (mop->mo_stripe_count > 72) {
822 int extra = mop->mo_stripe_count * 24;
823 extra = ((extra - 1) | 4095) + 1;
824 bytes_per_inode += extra;
828 /* Allocate fewer inodes on large OST devices. Most
829 * filesystems can be much more aggressive than even
830 * this, but it is impossible to know in advance. */
831 if (IS_OST(&mop->mo_ldd)) {
832 /* OST > 16TB assume average file size 1MB */
833 if (device_sz > (16ULL << 30))
834 bytes_per_inode = 1024 * 1024;
835 /* OST > 4TB assume average file size 512kB */
836 else if (device_sz > (4ULL << 30))
837 bytes_per_inode = 512 * 1024;
838 /* OST > 1TB assume average file size 256kB */
839 else if (device_sz > (1ULL << 30))
840 bytes_per_inode = 256 * 1024;
841 /* OST > 10GB assume average file size 64kB,
842 * plus a bit so that inodes will fit into a
843 * 256x flex_bg without overflowing */
844 else if (device_sz > (10ULL << 20))
845 bytes_per_inode = 69905;
848 if (bytes_per_inode > 0) {
849 sprintf(buf, " -i %ld", bytes_per_inode);
850 strscat(mop->mo_mkfsopts, buf,
851 sizeof(mop->mo_mkfsopts));
856 strscat(mop->mo_mkfsopts, " -q",
857 sizeof(mop->mo_mkfsopts));
860 /* start handle -O mkfs options */
861 if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
862 if (strstr(start + 2, "-O") != NULL) {
864 "%s: don't specify multiple -O options\n",
868 start = moveopts_to_end(start);
869 maxbuflen = sizeof(mop->mo_mkfsopts) -
870 (start - mop->mo_mkfsopts) - strlen(start);
871 enable_default_ext4_features(mop, start, maxbuflen, 1);
873 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts),
874 maxbuflen = sizeof(mop->mo_mkfsopts) -
875 strlen(mop->mo_mkfsopts);
876 enable_default_ext4_features(mop, start, maxbuflen, 0);
878 /* end handle -O mkfs options */
880 /* start handle -E mkfs options */
881 if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
882 if (strstr(start + 2, "-E") != NULL) {
884 "%s: don't specify multiple -E options\n",
888 start = moveopts_to_end(start);
889 maxbuflen = sizeof(mop->mo_mkfsopts) -
890 (start - mop->mo_mkfsopts) - strlen(start);
893 start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
894 maxbuflen = sizeof(mop->mo_mkfsopts) -
895 strlen(mop->mo_mkfsopts);
898 /* In order to align the filesystem metadata on 1MB boundaries,
899 * give a resize value that will reserve a power-of-two group
900 * descriptor blocks, but leave one block for the superblock.
901 * Only useful for filesystems with < 2^32 blocks due to resize
903 if (IS_OST(&mop->mo_ldd) && mop->mo_device_sz > 100 * 1024 &&
904 mop->mo_device_sz * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
905 unsigned group_blocks = L_BLOCK_SIZE * 8;
906 unsigned desc_per_block = L_BLOCK_SIZE / 32;
907 unsigned resize_blks;
909 resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
910 snprintf(buf, sizeof(buf), "%u", resize_blks);
911 append_unique(start, ext_opts ? "," : " -E ",
912 "resize", buf, maxbuflen);
916 /* Avoid zeroing out the full journal - speeds up mkfs */
917 if (is_e2fsprogs_feature_supp("-E lazy_journal_init") == 0)
918 append_unique(start, ext_opts ? "," : " -E ",
919 "lazy_journal_init", NULL, maxbuflen);
920 /* end handle -E mkfs options */
922 /* Allow reformat of full devices (as opposed to
923 partitions.) We already checked for mounted dev. */
924 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
926 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
927 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
928 mop->mo_ldd.ldd_svname);
930 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
931 progname, mop->mo_ldd.ldd_mount_type,
932 MT_STR(&mop->mo_ldd));
936 /* For loop device format the dev, not the filename */
937 dev = mop->mo_device;
938 if (mop->mo_flags & MO_IS_LOOP)
939 dev = mop->mo_loopdev;
941 vprint("formatting backing filesystem %s on %s\n",
942 MT_STR(&mop->mo_ldd), dev);
943 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
944 vprint("\t4k blocks "LPU64"\n", block_count);
945 vprint("\toptions %s\n", mop->mo_mkfsopts);
947 /* mkfs_cmd's trailing space is important! */
948 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
949 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
950 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
951 if (block_count != 0) {
952 sprintf(buf, " "LPU64, block_count);
953 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
956 vprint("mkfs_cmd = %s\n", mkfs_cmd);
957 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
960 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
965 /* ==================== Lustre config functions =============*/
967 void print_ldd(char *str, struct lustre_disk_data *ldd)
969 printf("\n %s:\n", str);
970 printf("Target: %s\n", ldd->ldd_svname);
971 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
972 printf("Index: unassigned\n");
974 printf("Index: %d\n", ldd->ldd_svindex);
975 if (ldd->ldd_uuid[0])
976 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
977 printf("Lustre FS: %s\n", ldd->ldd_fsname);
978 printf("Mount type: %s\n", MT_STR(ldd));
979 printf("Flags: %#x\n", ldd->ldd_flags);
980 printf(" (%s%s%s%s%s%s%s%s%s%s)\n",
981 IS_MDT(ldd) ? "MDT ":"",
982 IS_OST(ldd) ? "OST ":"",
983 IS_MGS(ldd) ? "MGS ":"",
984 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
985 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
986 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
987 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
988 ldd->ldd_flags & LDD_F_IAM_DIR ? "IAM_dir_format ":"",
989 ldd->ldd_flags & LDD_F_NO_PRIMNODE? "no_primnode ":"",
990 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
991 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
992 printf("Parameters:%s\n", ldd->ldd_params);
993 if (ldd->ldd_userdata[0])
994 printf("Comment: %s\n", ldd->ldd_userdata);
998 static int touch_file(char *filename)
1002 if (filename == NULL) {
1006 fd = open(filename, O_CREAT | O_TRUNC, 0600);
1015 /* keep it less than LL_FID_NAMELEN */
1016 #define DUMMY_FILE_NAME_LEN 25
1017 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
1019 /* Need to add these many entries to this directory to make HTREE dir. */
1020 #define MIN_ENTRIES_REQ_FOR_HTREE ((L_BLOCK_SIZE / EXT3_DIRENT_SIZE))
1022 static int add_dummy_files(char *dir)
1024 char fpname[PATH_MAX];
1028 for (i = 0; i < MIN_ENTRIES_REQ_FOR_HTREE; i++) {
1029 snprintf(fpname, PATH_MAX, "%s/%0*d", dir,
1030 DUMMY_FILE_NAME_LEN, i);
1032 rc = touch_file(fpname);
1033 if (rc && rc != -EEXIST) {
1035 "%s: Can't create dummy file %s: %s\n",
1036 progname, fpname , strerror(errno));
1043 static int __l_mkdir(char * filepnm, int mode , struct mkfs_opts *mop)
1047 ret = mkdir(filepnm, mode);
1048 if (ret && ret != -EEXIST)
1051 /* IAM mode supports ext3 directories of HTREE type only. So add dummy
1052 * entries to new directory to create htree type of container for
1053 * this directory. */
1054 if (mop->mo_ldd.ldd_flags & LDD_F_IAM_DIR)
1055 return add_dummy_files(filepnm);
1059 /* Write the server config files */
1060 int write_local_files(struct mkfs_opts *mop)
1062 char mntpt[] = "/tmp/mntXXXXXX";
1069 /* Mount this device temporarily in order to write these files */
1070 if (!mkdtemp(mntpt)) {
1071 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
1072 progname, mntpt, strerror(errno));
1076 dev = mop->mo_device;
1077 if (mop->mo_flags & MO_IS_LOOP)
1078 dev = mop->mo_loopdev;
1080 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0,
1081 mop->mo_ldd.ldd_mount_opts);
1083 fprintf(stderr, "%s: Unable to mount %s: %s\n",
1084 progname, dev, strerror(errno));
1086 if (errno == ENODEV) {
1087 fprintf(stderr, "Is the %s module available?\n",
1088 MT_STR(&mop->mo_ldd));
1093 /* Set up initial directories */
1094 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
1095 ret = __l_mkdir(filepnm, 0777, mop);
1096 if ((ret != 0) && (errno != EEXIST)) {
1097 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
1098 progname, filepnm, strerror(errno));
1100 } else if (errno == EEXIST) {
1104 /* Save the persistent mount data into a file. Lustre must pre-read
1105 this file to get the real mount options. */
1106 vprint("Writing %s\n", MOUNT_DATA_FILE);
1107 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
1108 filep = fopen(filepnm, "w");
1110 fprintf(stderr, "%s: Unable to create %s file: %s\n",
1111 progname, filepnm, strerror(errno));
1114 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
1115 if (num < 1 && ferror(filep)) {
1116 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
1117 progname, filepnm, strerror(errno));
1123 /* Check for upgrade */
1124 if ((mop->mo_ldd.ldd_flags & (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS))
1125 == (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS)) {
1128 int cmdsz = sizeof(cmd);
1129 vprint("Copying old logs\n");
1131 /* Copy the old client log to fsname-client */
1132 sprintf(filepnm, "%s/%s/%s-client",
1133 mntpt, MOUNT_CONFIGS_DIR, mop->mo_ldd.ldd_fsname);
1134 snprintf(cmd, cmdsz, "cp %s/%s/client %s", mntpt, MDT_LOGS_DIR,
1136 ret = run_command(cmd, cmdsz);
1138 fprintf(stderr, "%s: Can't copy 1.4 config %s/client "
1139 "(%d)\n", progname, MDT_LOGS_DIR, ret);
1140 fprintf(stderr, "mount -t ldiskfs %s somewhere, "
1141 "find the client log for fs %s and "
1142 "copy it manually into %s/%s-client, "
1145 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
1146 mop->mo_ldd.ldd_fsname);
1150 /* We need to use the old mdt log because otherwise mdt won't
1151 have complete lov if old clients connect before all
1153 /* Copy the old mdt log to fsname-MDT0000 (get old
1154 name from mdt_UUID) */
1156 strscpy(filepnm, (char *)mop->mo_ldd.ldd_uuid, sizeof(filepnm));
1157 term = strstr(filepnm, "_UUID");
1160 snprintf(cmd, cmdsz, "cp %s/%s/%s %s/%s/%s",
1161 mntpt, MDT_LOGS_DIR, filepnm,
1162 mntpt, MOUNT_CONFIGS_DIR,
1163 mop->mo_ldd.ldd_svname);
1164 ret = run_command(cmd, cmdsz);
1167 fprintf(stderr, "%s: Can't copy 1.4 config %s/%s "
1168 "(%d)\n", progname, MDT_LOGS_DIR, filepnm, ret);
1169 fprintf(stderr, "mount -t ext3 %s somewhere, "
1170 "find the MDT log for fs %s and "
1171 "copy it manually into %s/%s, "
1174 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
1175 mop->mo_ldd.ldd_svname);
1180 /* end COMPAT_146 */
1189 int read_local_files(struct mkfs_opts *mop)
1191 char tmpdir[] = "/tmp/dirXXXXXX";
1197 int cmdsz = sizeof(cmd);
1199 /* Make a temporary directory to hold Lustre data files. */
1200 if (!mkdtemp(tmpdir)) {
1201 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
1202 progname, tmpdir, strerror(errno));
1206 dev = mop->mo_device;
1208 /* TODO: it's worth observing the get_mountdata() function that is
1209 in mount_utils.c for getting the mountdata out of the
1212 /* Construct debugfs command line. */
1213 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
1214 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
1216 ret = run_command(cmd, cmdsz);
1218 verrprint("%s: Unable to dump %s dir (%d)\n",
1219 progname, MOUNT_CONFIGS_DIR, ret);
1221 sprintf(filepnm, "%s/mountdata", tmpdir);
1222 filep = fopen(filepnm, "r");
1225 vprint("Reading %s\n", MOUNT_DATA_FILE);
1226 num_read = fread(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
1227 if (num_read < 1 && ferror(filep)) {
1228 fprintf(stderr, "%s: Unable to read from file %s: %s\n",
1229 progname, filepnm, strerror(errno));
1234 /* Try to read pre-1.6 config from last_rcvd */
1235 struct lr_server_data lsd;
1236 verrprint("%s: Unable to read %d.%d config %s.\n",
1237 progname, LUSTRE_MAJOR, LUSTRE_MINOR, filepnm);
1239 verrprint("Trying 1.4 config from last_rcvd\n");
1240 sprintf(filepnm, "%s/%s", tmpdir, LAST_RCVD);
1242 /* Construct debugfs command line. */
1243 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s' %s",
1244 DEBUGFS, LAST_RCVD, filepnm, dev);
1246 ret = run_command(cmd, cmdsz);
1248 fprintf(stderr, "%s: Unable to dump %s file (%d)\n",
1249 progname, LAST_RCVD, ret);
1253 filep = fopen(filepnm, "r");
1255 fprintf(stderr, "%s: Unable to open %s: %s\n",
1256 progname, filepnm, strerror(errno));
1258 verrprint("Contents of %s:\n", tmpdir);
1260 snprintf(cmd, cmdsz, "ls -l %s/", tmpdir);
1261 run_command(cmd, cmdsz);
1262 verrprint("Contents of disk:\n");
1263 snprintf(cmd, cmdsz, "%s -c -R 'ls -l /' %s",
1265 run_command(cmd, cmdsz);
1269 vprint("Reading %s\n", LAST_RCVD);
1270 ret = fread(&lsd, 1, sizeof(lsd), filep);
1271 if (ret < sizeof(lsd)) {
1272 fprintf(stderr, "%s: Short read (%d of %d)\n",
1273 progname, ret, (int)sizeof(lsd));
1274 ret = ferror(filep);
1278 vprint("Feature compat=%x, incompat=%x\n",
1279 lsd.lsd_feature_compat, lsd.lsd_feature_incompat);
1281 if ((lsd.lsd_feature_compat & OBD_COMPAT_OST) ||
1282 (lsd.lsd_feature_incompat & OBD_INCOMPAT_OST)) {
1283 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
1284 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
1285 } else if ((lsd.lsd_feature_compat & OBD_COMPAT_MDT) ||
1286 (lsd.lsd_feature_incompat & OBD_INCOMPAT_MDT)) {
1287 /* We must co-locate so mgs can see old logs.
1288 If user doesn't want this, they can copy the old
1289 logs manually and re-tunefs. */
1290 mop->mo_ldd.ldd_flags =
1291 LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_MGS;
1292 mop->mo_ldd.ldd_svindex = lsd.lsd_mdt_index;
1294 /* If neither is set, we're pre-1.4.6, make a guess. */
1295 /* Construct debugfs command line. */
1296 snprintf(cmd, cmdsz, "%s -c -R 'rdump /%s %s' %s",
1297 DEBUGFS, MDT_LOGS_DIR, tmpdir, dev);
1298 run_command(cmd, cmdsz);
1300 sprintf(filepnm, "%s/%s", tmpdir, MDT_LOGS_DIR);
1301 if (lsd.lsd_ost_index > 0) {
1302 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
1303 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
1305 /* If there's a LOGS dir, it's an MDT */
1306 if ((ret = access(filepnm, F_OK)) == 0) {
1307 mop->mo_ldd.ldd_flags =
1310 /* Old MDT's are always index 0
1312 mop->mo_ldd.ldd_svindex = 0;
1314 /* The index may not be correct */
1315 mop->mo_ldd.ldd_flags =
1316 LDD_F_SV_TYPE_OST | LDD_F_NEED_INDEX;
1317 verrprint("OST with unknown index\n");
1323 memcpy(mop->mo_ldd.ldd_uuid, lsd.lsd_uuid,
1324 sizeof(mop->mo_ldd.ldd_uuid));
1325 mop->mo_ldd.ldd_flags |= LDD_F_UPGRADE14;
1327 /* end COMPAT_146 */
1332 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
1333 run_command(cmd, cmdsz);
1335 verrprint("Failed to read old data (%d)\n", ret);
1340 void set_defaults(struct mkfs_opts *mop)
1342 mop->mo_ldd.ldd_magic = LDD_MAGIC;
1343 mop->mo_ldd.ldd_config_ver = 1;
1344 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
1345 mop->mo_mgs_failnodes = 0;
1346 strcpy(mop->mo_ldd.ldd_fsname, "lustre");
1347 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
1349 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
1350 mop->mo_stripe_count = 1;
1353 static inline void badopt(const char *opt, char *type)
1355 fprintf(stderr, "%s: '--%s' only valid for %s\n",
1356 progname, opt, type);
1360 static int add_param(char *buf, char *key, char *val)
1362 int end = sizeof(((struct lustre_disk_data *)0)->ldd_params);
1363 int start = strlen(buf);
1367 keylen = strlen(key);
1368 if (start + 1 + keylen + strlen(val) >= end) {
1369 fprintf(stderr, "%s: params are too long-\n%s %s%s\n",
1370 progname, buf, key ? key : "", val);
1374 sprintf(buf + start, " %s%s", key ? key : "", val);
1378 /* from mount_lustre */
1379 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
1380 #define MAXNIDSTR 1024
1381 static char *convert_hostnames(char *s1)
1383 char *converted, *s2 = 0, *c, *end, sep;
1384 int left = MAXNIDSTR;
1387 converted = malloc(left);
1388 if (converted == NULL) {
1392 end = s1 + strlen(s1);
1394 while ((left > 0) && (s1 < end)) {
1395 s2 = strpbrk(s1, ",:");
1400 nid = libcfs_str2nid(s1);
1402 if (nid == LNET_NID_ANY) {
1403 fprintf(stderr, "%s: Can't parse NID '%s'\n",
1408 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
1409 strlen("127.0.0.1")) == 0) {
1410 fprintf(stderr, "%s: The NID '%s' resolves to the "
1411 "loopback address '%s'. Lustre requires a "
1412 "non-loopback address.\n",
1413 progname, s1, libcfs_nid2str(nid));
1418 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
1419 left = converted + MAXNIDSTR - c;
1425 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
1428 static struct option long_opt[] = {
1429 {"iam-dir", 0, 0, 'a'},
1430 {"backfstype", 1, 0, 'b'},
1431 {"stripe-count-hint", 1, 0, 'c'},
1432 {"comment", 1, 0, 'u'},
1433 {"configdev", 1, 0, 'C'},
1434 {"device-size", 1, 0, 'd'},
1435 {"dryrun", 0, 0, 'n'},
1436 {"erase-params", 0, 0, 'e'},
1437 {"failnode", 1, 0, 'f'},
1438 {"failover", 1, 0, 'f'},
1440 {"help", 0, 0, 'h'},
1441 {"index", 1, 0, 'i'},
1442 {"mkfsoptions", 1, 0, 'k'},
1443 {"mgsnode", 1, 0, 'm'},
1444 {"mgsnid", 1, 0, 'm'},
1446 {"fsname",1, 0, 'L'},
1447 {"noformat", 0, 0, 'n'},
1448 {"nomgs", 0, 0, 'N'},
1449 {"mountfsoptions", 1, 0, 'o'},
1451 {"param", 1, 0, 'p'},
1452 {"print", 0, 0, 'n'},
1453 {"quiet", 0, 0, 'q'},
1454 {"reformat", 0, 0, 'r'},
1455 {"servicenode", 1, 0, 's'},
1456 {"verbose", 0, 0, 'v'},
1457 {"writeconf", 0, 0, 'w'},
1458 {"upgrade_to_18", 0, 0, 'U'},
1459 {"network", 1, 0, 't'},
1462 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqrs:t:Uu:vw";
1465 int failnode_set = 0, servicenode_set = 0;
1467 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
1471 if (IS_MDT(&mop->mo_ldd))
1472 mop->mo_ldd.ldd_flags |= LDD_F_IAM_DIR;
1477 while (i < LDD_MT_LAST) {
1478 if (strcmp(optarg, mt_str(i)) == 0) {
1479 mop->mo_ldd.ldd_mount_type = i;
1487 if (IS_MDT(&mop->mo_ldd)) {
1488 int stripe_count = atol(optarg);
1489 if (stripe_count <= 0) {
1490 fprintf(stderr, "%s: bad stripe count "
1491 "%d\n", progname, stripe_count);
1494 mop->mo_stripe_count = stripe_count;
1496 badopt(long_opt[longidx].name, "MDT");
1500 case 'C': /* Configdev */
1502 printf("Configdev not implemented\n");
1505 mop->mo_device_sz = atol(optarg);
1508 mop->mo_ldd.ldd_params[0] = '\0';
1509 /* Must update the mgs logs */
1510 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1516 if ((opt == 'f' && servicenode_set)
1517 || (opt == 's' && failnode_set)) {
1518 fprintf(stderr, "%s: %s cannot use with --%s\n",
1519 progname, long_opt[longidx].name,
1520 opt == 'f' ? "servicenode" : "failnode");
1524 nids = convert_hostnames(optarg);
1527 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
1532 /* Must update the mgs logs */
1533 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1537 mop->mo_ldd.ldd_flags |= LDD_F_NO_PRIMNODE;
1538 servicenode_set = 1;
1544 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
1550 if (!(mop->mo_ldd.ldd_flags &
1551 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1552 LDD_F_WRITECONF))) {
1553 fprintf(stderr, "%s: cannot change the index of"
1554 " a registered target\n", progname);
1557 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
1558 mop->mo_ldd.ldd_svindex = atol(optarg);
1559 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
1561 badopt(long_opt[longidx].name, "MDT,OST");
1566 strscpy(mop->mo_mkfsopts, optarg,
1567 sizeof(mop->mo_mkfsopts));
1571 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
1572 (!(mop->mo_ldd.ldd_flags &
1573 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1574 LDD_F_WRITECONF)))) {
1575 fprintf(stderr, "%s: cannot change the name of"
1576 " a registered target\n", progname);
1579 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
1580 fprintf(stderr, "%s: filesystem name must be "
1581 "1-8 chars\n", progname);
1584 if ((tmp = strpbrk(optarg, "/:"))) {
1585 fprintf(stderr, "%s: char '%c' not allowed in "
1586 "filesystem name\n", progname, *tmp);
1589 strscpy(mop->mo_ldd.ldd_fsname, optarg,
1590 sizeof(mop->mo_ldd.ldd_fsname));
1594 char *nids = convert_hostnames(optarg);
1597 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
1602 mop->mo_mgs_failnodes++;
1606 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
1612 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
1615 *mountopts = optarg;
1618 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
1621 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg);
1624 /* Must update the mgs logs */
1625 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1631 mop->mo_flags |= MO_FORCEFORMAT;
1634 if (!IS_MDT(&mop->mo_ldd) && !IS_OST(&mop->mo_ldd)) {
1635 badopt(long_opt[longidx].name, "MDT,OST");
1642 rc = add_param(mop->mo_ldd.ldd_params,
1643 PARAM_NETWORK, optarg);
1646 /* Must update the mgs logs */
1647 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1650 strscpy(mop->mo_ldd.ldd_userdata, optarg,
1651 sizeof(mop->mo_ldd.ldd_userdata));
1657 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
1665 fprintf(stderr, "Unknown option '%c'\n", opt);
1671 /* Last arg is device */
1672 if (optind != argc - 1) {
1674 fprintf(stderr, "Bad argument: %s\n", argv[optind]);
1678 /* single argument: <device> */
1685 /* Search for opt in mntlist, returning true if found.
1687 static int in_mntlist(char *opt, char *mntlist)
1689 char *ml, *mlp, *item, *ctx = NULL;
1691 if (!(ml = strdup(mntlist))) {
1692 fprintf(stderr, "%s: out of memory\n", progname);
1696 while ((item = strtok_r(mlp, ",", &ctx))) {
1697 if (!strcmp(opt, item))
1702 return (item != NULL);
1705 /* Issue a message on stderr for every item in wanted_mountopts that is not
1706 * present in mountopts. The justwarn boolean toggles between error and
1707 * warning message. Return an error count.
1709 static int check_mountfsoptions(char *mountopts, char *wanted_mountopts,
1712 char *ml, *mlp, *item, *ctx = NULL;
1715 if (!(ml = strdup(wanted_mountopts))) {
1716 fprintf(stderr, "%s: out of memory\n", progname);
1720 while ((item = strtok_r(mlp, ",", &ctx))) {
1721 if (!in_mntlist(item, mountopts)) {
1722 fprintf(stderr, "%s: %s mount option `%s' is missing\n",
1723 progname, justwarn ? "Warning: default"
1724 : "Error: mandatory", item);
1733 /* Trim embedded white space, leading and trailing commas from string s.
1735 static void trim_mountfsoptions(char *s)
1741 memmove(p, p + 1, strlen(p + 1) + 1);
1748 memmove(&s[0], &s[1], strlen(&s[1]) + 1);
1750 p = s + strlen(s) - 1;
1751 while (p >= s && *p == ',')
1755 int main(int argc, char *const argv[])
1757 struct mkfs_opts mop;
1758 struct lustre_disk_data *ldd;
1759 char *mountopts = NULL;
1760 char always_mountopts[512] = "";
1761 char default_mountopts[512] = "";
1764 if ((progname = strrchr(argv[0], '/')) != NULL)
1769 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
1774 memset(&mop, 0, sizeof(mop));
1777 /* device is last arg */
1778 strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
1780 /* Are we using a loop device? */
1781 ret = is_block(mop.mo_device);
1787 mop.mo_flags |= MO_IS_LOOP;
1790 /* For tunefs, we must read in the old values before parsing any
1793 /* Check whether the disk has already been formatted by mkfs.lustre */
1794 ret = is_lustre_target(&mop);
1797 fprintf(stderr, "Device %s has not been formatted with "
1798 "mkfs.lustre\n", mop.mo_device);
1803 ret = read_local_files(&mop);
1806 fprintf(stderr, "Failed to read previous Lustre data from %s "
1807 "(%d)\n", mop.mo_device, ret);
1810 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
1811 mop.mo_mgs_failnodes++;
1814 print_ldd("Read previous values", &(mop.mo_ldd));
1817 ret = parse_opts(argc, argv, &mop, &mountopts);
1823 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
1825 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
1830 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
1832 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
1837 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
1838 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
1840 fprintf(stderr, "Can't find the target index, "
1841 "specify with --index\n");
1847 * Comment out these 2 checks temporarily, since for multi-MDSes
1848 * in single node only 1 mds node could have mgs service
1850 if (IS_MDT(ldd) && !IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1851 verrprint("No management node specified, adding MGS to this "
1853 ldd->ldd_flags |= LDD_F_SV_TYPE_MGS;
1855 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1858 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
1860 fprintf(stderr, "Must specify --mgsnode=\n");
1866 /* These are the permanent mount options (always included) */
1867 switch (ldd->ldd_mount_type) {
1869 case LDD_MT_LDISKFS:
1870 case LDD_MT_LDISKFS2:
1871 strscat(default_mountopts, ",errors=remount-ro",
1872 sizeof(default_mountopts));
1873 if (IS_MDT(ldd) || IS_MGS(ldd))
1874 strscat(always_mountopts, ",user_xattr",
1875 sizeof(always_mountopts));
1876 /* NB: Files created while extents are enabled can only be read
1877 * if mounted using the ext4 or ldiskfs filesystem type. */
1879 (ldd->ldd_mount_type == LDD_MT_LDISKFS ||
1880 ldd->ldd_mount_type == LDD_MT_LDISKFS2)) {
1881 strscat(default_mountopts, ",extents,mballoc",
1882 sizeof(default_mountopts));
1887 fprintf(stderr, "unknown fs type %d '%s'\n",
1888 ldd->ldd_mount_type,
1895 trim_mountfsoptions(mountopts);
1896 (void)check_mountfsoptions(mountopts, default_mountopts, 1);
1897 if (check_mountfsoptions(mountopts, always_mountopts, 0)) {
1901 sprintf(ldd->ldd_mount_opts, "%s", mountopts);
1904 if (ldd->ldd_mount_opts[0] == 0)
1905 /* use the defaults unless old opts exist */
1908 sprintf(ldd->ldd_mount_opts, "%s%s",
1909 always_mountopts, default_mountopts);
1910 trim_mountfsoptions(ldd->ldd_mount_opts);
1914 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
1915 ldd->ldd_fsname, ldd->ldd_svname);
1918 print_ldd("Permanent disk data", ldd);
1921 printf("exiting before disk write.\n");
1925 if (check_mtab_entry(mop.mo_device))
1928 /* Create the loopback file */
1929 if (mop.mo_flags & MO_IS_LOOP) {
1930 ret = access(mop.mo_device, F_OK);
1933 #ifndef TUNEFS /* mkfs.lustre */
1934 /* Reformat the loopback file */
1935 if (ret || (mop.mo_flags & MO_FORCEFORMAT)) {
1936 ret = loop_format(&mop);
1942 ret = loop_setup(&mop);
1945 fprintf(stderr, "Loop device setup for %s failed: %s\n",
1946 mop.mo_device, strerror(ret));
1951 #ifndef TUNEFS /* mkfs.lustre */
1952 /* Check whether the disk has already been formatted by mkfs.lustre */
1953 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
1954 ret = is_lustre_target(&mop);
1957 fprintf(stderr, "Device %s was previously formatted "
1958 "for lustre. Use --reformat to reformat it, "
1959 "or tunefs.lustre to modify.\n",
1965 /* Format the backing filesystem */
1966 ret = make_lustre_backfs(&mop);
1969 fprintf(stderr, "mkfs failed %d\n", ret);
1974 /* Write our config files */
1975 ret = write_local_files(&mop);
1978 fprintf(stderr, "failed to write local files\n");
1985 /* Fix any crazy return values from system() */
1986 if (ret && ((ret & 255) == 0))
1989 verrprint("%s: exiting with %d (%s)\n",
1990 progname, ret, strerror(ret));