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 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/utils/mkfs_lustre.c
38 * Author: Nathan Rutman <nathan@clusterfs.com>
41 /* This source file is compiled into both mkfs.lustre and tunefs.lustre */
53 #include <sys/types.h>
55 #include <sys/mount.h>
56 #include <sys/utsname.h>
63 /* libcfs.h is not really needed here, but on SLES10/PPC, fs.h includes idr.h which
64 * requires BITS_PER_LONG to be defined */
65 #include <libcfs/libcfs.h>
66 #include <linux/fs.h> /* for BLKGETSIZE64 */
67 #include <linux/version.h>
69 #include <lustre_disk.h>
70 #include <lustre_param.h>
71 #include <lnet/lnetctl.h>
72 #include <lustre_ver.h>
73 #include "mount_utils.h"
79 #define MAX_LOOP_DEVICES 16
80 #define L_BLOCK_SIZE 4096
81 #define INDEX_UNASSIGNED 0xFFFF
82 #define MO_IS_LOOP 0x01
83 #define MO_FORCEFORMAT 0x02
85 /* used to describe the options to format the lustre disk, not persistent */
87 struct lustre_disk_data mo_ldd; /* to be written in MOUNT_DATA_FILE */
88 char mo_device[128]; /* disk device name */
89 char mo_mkfsopts[128]; /* options to the backing-store mkfs */
90 char mo_loopdev[128]; /* in case a loop dev is needed */
91 __u64 mo_device_sz; /* in KB */
99 static int print_only = 0;
100 static int failover = 0;
101 static int upgrade_to_18 = 0;
103 void usage(FILE *out)
105 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
106 fprintf(out, "usage: %s <target types> [options] <device>\n", progname);
108 "\t<device>:block device or file (e.g /dev/sda or /tmp/ost1)\n"
110 "\t\t--ost: object storage, mutually exclusive with mdt,mgs\n"
111 "\t\t--mdt: metadata storage, mutually exclusive with ost\n"
112 "\t\t--mgs: configuration management service - one per site\n"
113 "\toptions (in order of popularity):\n"
114 "\t\t--mgsnode=<nid>[,<...>] : NID(s) of a remote mgs node\n"
115 "\t\t\trequired for all targets other than the mgs node\n"
116 "\t\t--fsname=<filesystem_name> : default is 'lustre'\n"
117 "\t\t--failnode=<nid>[,<...>] : NID(s) of a failover partner\n"
118 "\t\t--param <key>=<value> : set a permanent parameter\n"
119 "\t\t\te.g. --param sys.timeout=40\n"
120 "\t\t\t --param lov.stripesize=2M\n"
121 "\t\t--index=#N : target index (i.e. ost index within the lov)\n"
122 /* FIXME implement 1.6.x
123 "\t\t--configdev=<altdevice|file>: store configuration info\n"
124 "\t\t\tfor this device on an alternate device\n"
126 "\t\t--comment=<user comment>: arbitrary user string (%d bytes)\n"
127 "\t\t--mountfsoptions=<opts> : permanent mount options\n"
129 "\t\t--backfstype=<fstype> : backing fs type (ext3, ldiskfs)\n"
130 "\t\t--device-size=#N(KB) : device size for loop devices\n"
131 "\t\t--mkfsoptions=<opts> : format options\n"
132 "\t\t--reformat: overwrite an existing disk\n"
133 "\t\t--stripe-count-hint=#N : used for optimizing MDT inode size\n"
134 "\t\t--iam-dir: make use of IAM directory format on backfs, incompatible with ext3.\n"
136 "\t\t--erase-params : erase all old parameter settings\n"
137 "\t\t--nomgs: turn off MGS service on this MDT\n"
138 "\t\t--writeconf: erase all config logs for this fs.\n"
140 "\t\t--dryrun: just report what we would do; "
141 "don't write to disk\n"
142 "\t\t--verbose : e.g. show mkfs progress\n"
144 (int)sizeof(((struct lustre_disk_data *)0)->ldd_userdata));
148 #define vprint if (verbose > 0) printf
149 #define verrprint if (verbose >= 0) printf
151 /*================ utility functions =====================*/
153 char *strscat(char *dst, char *src, int buflen) {
155 if (strlen(dst) + strlen(src) >= buflen) {
156 fprintf(stderr, "string buffer overflow (max %d): '%s' + '%s'"
157 "\n", buflen, dst, src);
160 return strcat(dst, src);
164 char *strscpy(char *dst, char *src, int buflen) {
166 return strscat(dst, src, buflen);
170 dev_major (unsigned long long int __dev)
172 return ((__dev >> 8) & 0xfff) | ((unsigned int) (__dev >> 32) & ~0xfff);
176 dev_minor (unsigned long long int __dev)
178 return (__dev & 0xff) | ((unsigned int) (__dev >> 12) & ~0xff);
183 static int version = 0;
187 char release[4] = "";
189 fd = open("/proc/sys/kernel/osrelease", O_RDONLY);
191 fprintf(stderr, "%s: Warning: Can't resolve kernel "
192 "version, assuming 2.6\n", progname);
194 read(fd, release, 4);
197 if (strncmp(release, "2.4.", 4) == 0)
205 static int check_mtab_entry(char *spec)
210 fp = setmntent(MOUNTED, "r");
214 while ((mnt = getmntent(fp)) != NULL) {
215 if (strcmp(mnt->mnt_fsname, spec) == 0) {
217 fprintf(stderr, "%s: according to %s %s is "
218 "already mounted on %s\n",
219 progname, MOUNTED, spec, mnt->mnt_dir);
228 /*============ disk dev functions ===================*/
230 /* Setup a file in the first unused loop_device */
231 int loop_setup(struct mkfs_opts *mop)
237 /* Figure out the loop device names */
238 if (!access("/dev/loop0", F_OK | R_OK)) {
239 strcpy(loop_base, "/dev/loop\0");
240 } else if (!access("/dev/loop/0", F_OK | R_OK)) {
241 strcpy(loop_base, "/dev/loop/\0");
243 fprintf(stderr, "%s: can't access loop devices\n", progname);
247 /* Find unused loop device */
248 for (i = 0; i < MAX_LOOP_DEVICES; i++) {
250 int cmdsz = sizeof(cmd);
252 sprintf(l_device, "%s%d", loop_base, i);
253 if (access(l_device, F_OK | R_OK))
255 snprintf(cmd, cmdsz, "losetup %s > /dev/null 2>&1", l_device);
258 /* losetup gets 1 (ret=256) for non-set-up device */
260 /* Set up a loopback device to our file */
261 snprintf(cmd, cmdsz, "losetup %s %s", l_device,
263 ret = run_command(cmd, cmdsz);
265 fprintf(stderr, "%s: error %d on losetup: %s\n",
266 progname, ret, strerror(ret));
269 strscpy(mop->mo_loopdev, l_device,
270 sizeof(mop->mo_loopdev));
275 fprintf(stderr, "%s: out of loop devices!\n", progname);
279 int loop_cleanup(struct mkfs_opts *mop)
283 if ((mop->mo_flags & MO_IS_LOOP) && *mop->mo_loopdev) {
284 sprintf(cmd, "losetup -d %s", mop->mo_loopdev);
285 ret = run_command(cmd, sizeof(cmd));
290 /* Determine if a device is a block device (as opposed to a file) */
291 int is_block(char* devname)
296 ret = access(devname, F_OK);
299 ret = stat(devname, &st);
301 fprintf(stderr, "%s: cannot stat %s\n", progname, devname);
304 return S_ISBLK(st.st_mode);
307 __u64 get_device_size(char* device)
312 fd = open(device, O_RDONLY);
314 fprintf(stderr, "%s: cannot open %s: %s\n",
315 progname, device, strerror(errno));
320 /* size in bytes. bz5831 */
321 ret = ioctl(fd, BLKGETSIZE64, (void*)&size);
326 ret = ioctl(fd, BLKGETSIZE, (void*)&lsize);
327 size = (__u64)lsize * 512;
332 fprintf(stderr, "%s: size ioctl failed: %s\n",
333 progname, strerror(errno));
337 vprint("device size = "LPU64"MB\n", size >> 20);
338 /* return value in KB */
342 int loop_format(struct mkfs_opts *mop)
346 if (mop->mo_device_sz == 0) {
348 fprintf(stderr, "loop device requires a --device-size= "
353 ret = creat(mop->mo_device, S_IRUSR|S_IWUSR);
356 fprintf(stderr, "%s: Unable to create backing store: %d\n",
362 ret = truncate(mop->mo_device, mop->mo_device_sz * 1024);
365 fprintf(stderr, "%s: Unable to truncate backing store: %d\n",
372 /* Display the need for the latest e2fsprogs to be installed. make_backfs
373 * indicates if the caller is make_lustre_backfs() or not. */
374 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
376 static int msg_displayed;
379 fprintf(stderr, "WARNING: e2fsprogs does not support %s "
380 "feature.\n\n", feature);
386 fprintf(stderr, "WARNING: The e2fsprogs package currently installed on "
387 "your system does not support \"%s\" feature.\nPlease install "
388 "the latest version of e2fsprogs from http://downloads.lustre.org"
389 "/public/tools/e2fsprogs/\nto enable this feature.\n", feature);
392 fprintf(stderr, "Feature will not be enabled until e2fsprogs "
393 "is updated and 'tune2fs -O %s %%{device}' "
394 "is run.\n\n", feature);
397 /* Check whether the file exists in the device */
398 static int file_in_dev(char *file_name, char *dev_name)
401 char debugfs_cmd[256];
402 unsigned int inode_num;
405 /* Construct debugfs command line. */
406 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
407 "debugfs -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
408 file_name, dev_name);
410 fp = popen(debugfs_cmd, "r");
412 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
416 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
420 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd), fp);
423 fprintf(stderr, "%s", debugfs_cmd);
424 if (strstr(debugfs_cmd, "unsupported feature")) {
425 disp_old_e2fsprogs_msg("an unknown", 0);
434 /* Check whether the device has already been used with lustre */
435 static int is_lustre_target(struct mkfs_opts *mop)
439 vprint("checking for existing Lustre data: ");
441 if ((rc = file_in_dev(MOUNT_DATA_FILE, mop->mo_device))) {
443 (rc == 1) ? MOUNT_DATA_FILE : "extents");
444 /* in the -1 case, 'extents' means this really IS a lustre
449 if ((rc = file_in_dev(LAST_RCVD, mop->mo_device))) {
450 vprint("found %s\n", LAST_RCVD);
454 vprint("not found\n");
455 return 0; /* The device is not a lustre target. */
458 /* Check if a certain feature is supported by e2fsprogs.
459 * Firstly we try to use "debugfs supported_features" command to check if
460 * the feature is supported. If this fails we try to set this feature with
461 * mke2fs to check for its support. */
462 static int is_e2fsprogs_feature_supp(const char *feature)
466 char imgname[] = "/tmp/test-img-XXXXXX";
470 snprintf(cmd, sizeof(cmd),
471 "debugfs -c -R \"supported_features %s\" 2>&1", feature);
473 /* Using popen() instead of run_command() since debugfs does not return
474 * proper error code if command is not supported */
475 fp = popen(cmd, "r");
477 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
480 ret = fread(cmd, 1, sizeof(cmd), fp);
482 if (strstr(cmd, feature) && !(strstr(cmd, "Unknown")))
486 if ((fd = mkstemp(imgname)) < 0)
489 snprintf(cmd, sizeof(cmd), "mke2fs -F -O %s %s 100 >/dev/null 2>&1",
491 /* run_command() displays the output of mke2fs when it fails for
492 * some feature, so use system() directly */
500 static void disp_old_kernel_msg(char *feature)
502 fprintf(stderr, "WARNING: ldiskfs filesystem does not support \"%s\" "
503 "feature.\n\n", feature);
506 static void enable_default_backfs_features(struct mkfs_opts *mop)
509 int maj_high, maj_low, min;
512 strscat(mop->mo_mkfsopts, " -O dir_index", sizeof(mop->mo_mkfsopts));
514 /* Upstream e2fsprogs called our uninit_groups feature uninit_bg,
515 * check for both of them when testing e2fsprogs features. */
516 if (is_e2fsprogs_feature_supp("uninit_groups") == 0)
517 strscat(mop->mo_mkfsopts, ",uninit_groups",
518 sizeof(mop->mo_mkfsopts));
519 else if (is_e2fsprogs_feature_supp("uninit_bg") == 0)
520 strscat(mop->mo_mkfsopts, ",uninit_bg",
521 sizeof(mop->mo_mkfsopts));
523 disp_old_e2fsprogs_msg("uninit_bg", 1);
529 sscanf(uts.release, "%d.%d.%d", &maj_high, &maj_low, &min);
530 printf("%d %d %d\n", maj_high, maj_low, min);
532 /* Multiple mount protection is enabled only if failover node is
533 * specified and if kernel version is higher than 2.6.9 */
535 if (KERNEL_VERSION(maj_high, maj_low, min) >=
536 KERNEL_VERSION(2,6,9)) {
537 if (is_e2fsprogs_feature_supp("mmp") == 0)
538 strscat(mop->mo_mkfsopts, ",mmp",
539 sizeof(mop->mo_mkfsopts));
541 disp_old_e2fsprogs_msg("mmp", 1);
543 disp_old_kernel_msg("mmp");
547 /* Build fs according to type */
548 int make_lustre_backfs(struct mkfs_opts *mop)
550 char mkfs_cmd[PATH_MAX];
556 if (mop->mo_device_sz != 0) {
557 if (mop->mo_device_sz < 8096){
558 fprintf(stderr, "%s: size of filesystem must be larger "
559 "than 8MB, but is set to %lldKB\n",
560 progname, (long long)mop->mo_device_sz);
563 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
566 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
567 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
568 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
569 __u64 device_sz = mop->mo_device_sz;
571 /* we really need the size */
572 if (device_sz == 0) {
573 device_sz = get_device_size(mop->mo_device);
578 /* Journal size in MB */
579 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
580 /* Choose our own default journal size */
581 long journal_sz = 0, max_sz;
582 if (device_sz > 1024 * 1024) /* 1GB */
583 journal_sz = (device_sz / 102400) * 4;
584 /* cap journal size at 1GB */
585 if (journal_sz > 1024L)
588 max_sz = (256000 * L_BLOCK_SIZE) >> 20; /* 1GB */
589 if (journal_sz > max_sz)
592 sprintf(buf, " -J size=%ld", journal_sz);
593 strscat(mop->mo_mkfsopts, buf,
594 sizeof(mop->mo_mkfsopts));
598 /* Bytes_per_inode: disk size / num inodes */
599 if (strstr(mop->mo_mkfsopts, "-i") == NULL) {
600 long bytes_per_inode = 0;
602 if (IS_MDT(&mop->mo_ldd))
603 bytes_per_inode = 4096;
605 /* Allocate fewer inodes on large OST devices. Most
606 filesystems can be much more aggressive than even
608 if ((IS_OST(&mop->mo_ldd) && (device_sz > 100000000)))
609 bytes_per_inode = 16384; /* > 100 Gb device */
612 if (bytes_per_inode > 0) {
613 sprintf(buf, " -i %ld", bytes_per_inode);
614 strscat(mop->mo_mkfsopts, buf,
615 sizeof(mop->mo_mkfsopts));
619 /* Inode size (for extended attributes). The LOV EA size is
620 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
621 * and we want some margin above that for ACLs, other EAs... */
622 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
624 if (IS_MDT(&mop->mo_ldd)) {
625 if (mop->mo_stripe_count > 72)
626 inode_size = 512; /* bz 7241 */
627 /* cray stripes across all osts (>60) */
628 else if (mop->mo_stripe_count > 32)
630 else if (mop->mo_stripe_count > 10)
634 } else if (IS_OST(&mop->mo_ldd)) {
635 /* now as we store fids in EA on OST we need
636 to make inode bigger */
640 if (inode_size > 0) {
641 sprintf(buf, " -I %ld", inode_size);
642 strscat(mop->mo_mkfsopts, buf,
643 sizeof(mop->mo_mkfsopts));
648 strscat(mop->mo_mkfsopts, " -q",
649 sizeof(mop->mo_mkfsopts));
652 if (strstr(mop->mo_mkfsopts, "-O") == NULL)
653 enable_default_backfs_features(mop);
655 /* Allow reformat of full devices (as opposed to
656 partitions.) We already checked for mounted dev. */
657 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
659 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
660 "mkfs.ext2 -j -b %d -L %s ", L_BLOCK_SIZE,
661 mop->mo_ldd.ldd_svname);
662 } else if (mop->mo_ldd.ldd_mount_type == LDD_MT_REISERFS) {
663 long journal_sz = 0; /* FIXME default journal size */
664 if (journal_sz > 0) {
665 sprintf(buf, " --journal_size %ld", journal_sz);
666 strscat(mop->mo_mkfsopts, buf,
667 sizeof(mop->mo_mkfsopts));
669 snprintf(mkfs_cmd, sizeof(mkfs_cmd), "mkreiserfs -ff ");
671 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
672 progname, mop->mo_ldd.ldd_mount_type,
673 MT_STR(&mop->mo_ldd));
677 /* For loop device format the dev, not the filename */
678 dev = mop->mo_device;
679 if (mop->mo_flags & MO_IS_LOOP)
680 dev = mop->mo_loopdev;
682 vprint("formatting backing filesystem %s on %s\n",
683 MT_STR(&mop->mo_ldd), dev);
684 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
685 vprint("\t4k blocks %d\n", block_count);
686 vprint("\toptions %s\n", mop->mo_mkfsopts);
688 /* mkfs_cmd's trailing space is important! */
689 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
690 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
691 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
692 if (block_count != 0) {
693 sprintf(buf, " %d", block_count);
694 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
697 vprint("mkfs_cmd = %s\n", mkfs_cmd);
698 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
701 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
706 /* ==================== Lustre config functions =============*/
708 void print_ldd(char *str, struct lustre_disk_data *ldd)
710 printf("\n %s:\n", str);
711 printf("Target: %s\n", ldd->ldd_svname);
712 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
713 printf("Index: unassigned\n");
715 printf("Index: %d\n", ldd->ldd_svindex);
716 if (ldd->ldd_uuid[0])
717 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
718 printf("Lustre FS: %s\n", ldd->ldd_fsname);
719 printf("Mount type: %s\n", MT_STR(ldd));
720 printf("Flags: %#x\n", ldd->ldd_flags);
721 printf(" (%s%s%s%s%s%s%s%s%s)\n",
722 IS_MDT(ldd) ? "MDT ":"",
723 IS_OST(ldd) ? "OST ":"",
724 IS_MGS(ldd) ? "MGS ":"",
725 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
726 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
727 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
728 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
729 ldd->ldd_flags & LDD_F_IAM_DIR ? "IAM_dir_format ":"",
730 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
731 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
732 printf("Parameters:%s\n", ldd->ldd_params);
733 if (ldd->ldd_userdata[0])
734 printf("Comment: %s\n", ldd->ldd_userdata);
738 static int touch_file(char *filename)
742 if (filename == NULL) {
746 fd = open(filename, O_CREAT | O_TRUNC, 0600);
755 /* keep it less than LL_FID_NAMELEN */
756 #define DUMMY_FILE_NAME_LEN 25
757 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
759 /* Need to add these many entries to this directory to make HTREE dir. */
760 #define MIN_ENTRIES_REQ_FOR_HTREE ((L_BLOCK_SIZE / EXT3_DIRENT_SIZE))
762 static int add_dummy_files(char *dir)
764 char fpname[PATH_MAX];
768 for (i = 0; i < MIN_ENTRIES_REQ_FOR_HTREE; i++) {
769 snprintf(fpname, PATH_MAX, "%s/%0*d", dir,
770 DUMMY_FILE_NAME_LEN, i);
772 rc = touch_file(fpname);
773 if (rc && rc != -EEXIST) {
775 "%s: Can't create dummy file %s: %s\n",
776 progname, fpname , strerror(errno));
783 static int __l_mkdir(char * filepnm, int mode , struct mkfs_opts *mop)
787 ret = mkdir(filepnm, mode);
788 if (ret && ret != -EEXIST)
791 /* IAM mode supports ext3 directories of HTREE type only. So add dummy
792 * entries to new directory to create htree type of container for
794 if (mop->mo_ldd.ldd_flags & LDD_F_IAM_DIR)
795 return add_dummy_files(filepnm);
799 /* Write the server config files */
800 int write_local_files(struct mkfs_opts *mop)
802 char mntpt[] = "/tmp/mntXXXXXX";
808 /* Mount this device temporarily in order to write these files */
809 if (!mkdtemp(mntpt)) {
810 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
811 progname, mntpt, strerror(errno));
815 dev = mop->mo_device;
816 if (mop->mo_flags & MO_IS_LOOP)
817 dev = mop->mo_loopdev;
819 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0, NULL);
821 fprintf(stderr, "%s: Unable to mount %s: %s\n",
822 progname, dev, strerror(errno));
824 if (errno == ENODEV) {
825 fprintf(stderr, "Is the %s module available?\n",
826 MT_STR(&mop->mo_ldd));
831 /* Set up initial directories */
832 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
833 ret = __l_mkdir(filepnm, 0777, mop);
834 if ((ret != 0) && (errno != EEXIST)) {
835 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
836 progname, filepnm, strerror(errno));
838 } else if (errno == EEXIST) {
842 /* Save the persistent mount data into a file. Lustre must pre-read
843 this file to get the real mount options. */
844 vprint("Writing %s\n", MOUNT_DATA_FILE);
845 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
846 filep = fopen(filepnm, "w");
848 fprintf(stderr, "%s: Unable to create %s file: %s\n",
849 progname, filepnm, strerror(errno));
852 fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
856 /* Check for upgrade */
857 if ((mop->mo_ldd.ldd_flags & (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS))
858 == (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS)) {
861 int cmdsz = sizeof(cmd);
862 vprint("Copying old logs\n");
864 /* Copy the old client log to fsname-client */
865 sprintf(filepnm, "%s/%s/%s-client",
866 mntpt, MOUNT_CONFIGS_DIR, mop->mo_ldd.ldd_fsname);
867 snprintf(cmd, cmdsz, "cp %s/%s/client %s", mntpt, MDT_LOGS_DIR,
869 ret = run_command(cmd, cmdsz);
871 fprintf(stderr, "%s: Can't copy 1.4 config %s/client "
872 "(%d)\n", progname, MDT_LOGS_DIR, ret);
873 fprintf(stderr, "mount -t ldiskfs %s somewhere, "
874 "find the client log for fs %s and "
875 "copy it manually into %s/%s-client, "
878 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
879 mop->mo_ldd.ldd_fsname);
883 /* We need to use the old mdt log because otherwise mdt won't
884 have complete lov if old clients connect before all
886 /* Copy the old mdt log to fsname-MDT0000 (get old
887 name from mdt_UUID) */
889 strscpy(filepnm, (char *)mop->mo_ldd.ldd_uuid, sizeof(filepnm));
890 term = strstr(filepnm, "_UUID");
893 snprintf(cmd, cmdsz, "cp %s/%s/%s %s/%s/%s",
894 mntpt, MDT_LOGS_DIR, filepnm,
895 mntpt, MOUNT_CONFIGS_DIR,
896 mop->mo_ldd.ldd_svname);
897 ret = run_command(cmd, cmdsz);
900 fprintf(stderr, "%s: Can't copy 1.4 config %s/%s "
901 "(%d)\n", progname, MDT_LOGS_DIR, filepnm, ret);
902 fprintf(stderr, "mount -t ext3 %s somewhere, "
903 "find the MDT log for fs %s and "
904 "copy it manually into %s/%s, "
907 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
908 mop->mo_ldd.ldd_svname);
922 int read_local_files(struct mkfs_opts *mop)
924 char tmpdir[] = "/tmp/dirXXXXXX";
930 int cmdsz = sizeof(cmd);
932 /* Make a temporary directory to hold Lustre data files. */
933 if (!mkdtemp(tmpdir)) {
934 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
935 progname, tmpdir, strerror(errno));
939 dev = mop->mo_device;
941 /* TODO: it's worth observing the get_mountdata() function that is
942 in mount_utils.c for getting the mountdata out of the
945 /* Construct debugfs command line. */
946 snprintf(cmd, cmdsz, "debugfs -c -R 'dump /%s %s/mountdata' '%s'",
947 MOUNT_DATA_FILE, tmpdir, dev);
949 ret = run_command(cmd, cmdsz);
951 verrprint("%s: Unable to dump %s dir (%d)\n",
952 progname, MOUNT_CONFIGS_DIR, ret);
954 sprintf(filepnm, "%s/mountdata", tmpdir);
955 filep = fopen(filepnm, "r");
957 vprint("Reading %s\n", MOUNT_DATA_FILE);
958 fread(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
961 /* Try to read pre-1.6 config from last_rcvd */
962 struct lr_server_data lsd;
963 verrprint("%s: Unable to read %d.%d config %s.\n",
964 progname, LUSTRE_MAJOR, LUSTRE_MINOR, filepnm);
966 verrprint("Trying 1.4 config from last_rcvd\n");
967 sprintf(filepnm, "%s/%s", tmpdir, LAST_RCVD);
969 /* Construct debugfs command line. */
970 snprintf(cmd, cmdsz, "debugfs -c -R 'dump /%s %s' %s",
971 LAST_RCVD, filepnm, dev);
973 ret = run_command(cmd, cmdsz);
975 fprintf(stderr, "%s: Unable to dump %s file (%d)\n",
976 progname, LAST_RCVD, ret);
980 filep = fopen(filepnm, "r");
982 fprintf(stderr, "%s: Unable to open %s: %s\n",
983 progname, filepnm, strerror(errno));
985 verrprint("Contents of %s:\n", tmpdir);
987 snprintf(cmd, cmdsz, "ls -l %s/", tmpdir);
988 run_command(cmd, cmdsz);
989 verrprint("Contents of disk:\n");
990 snprintf(cmd, cmdsz, "debugfs -c -R 'ls -l /' %s", dev);
991 run_command(cmd, cmdsz);
995 vprint("Reading %s\n", LAST_RCVD);
996 ret = fread(&lsd, 1, sizeof(lsd), filep);
997 if (ret < sizeof(lsd)) {
998 fprintf(stderr, "%s: Short read (%d of %d)\n",
999 progname, ret, (int)sizeof(lsd));
1000 ret = ferror(filep);
1004 vprint("Feature compat=%x, incompat=%x\n",
1005 lsd.lsd_feature_compat, lsd.lsd_feature_incompat);
1007 if ((lsd.lsd_feature_compat & OBD_COMPAT_OST) ||
1008 (lsd.lsd_feature_incompat & OBD_INCOMPAT_OST)) {
1009 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
1010 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
1011 } else if ((lsd.lsd_feature_compat & OBD_COMPAT_MDT) ||
1012 (lsd.lsd_feature_incompat & OBD_INCOMPAT_MDT)) {
1013 /* We must co-locate so mgs can see old logs.
1014 If user doesn't want this, they can copy the old
1015 logs manually and re-tunefs. */
1016 mop->mo_ldd.ldd_flags =
1017 LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_MGS;
1018 mop->mo_ldd.ldd_svindex = lsd.lsd_mdt_index;
1020 /* If neither is set, we're pre-1.4.6, make a guess. */
1021 /* Construct debugfs command line. */
1022 snprintf(cmd, cmdsz, "debugfs -c -R 'rdump /%s %s' %s",
1023 MDT_LOGS_DIR, tmpdir, dev);
1024 run_command(cmd, cmdsz);
1026 sprintf(filepnm, "%s/%s", tmpdir, MDT_LOGS_DIR);
1027 if (lsd.lsd_ost_index > 0) {
1028 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
1029 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
1031 /* If there's a LOGS dir, it's an MDT */
1032 if ((ret = access(filepnm, F_OK)) == 0) {
1033 mop->mo_ldd.ldd_flags =
1036 /* Old MDT's are always index 0
1038 mop->mo_ldd.ldd_svindex = 0;
1040 /* The index may not be correct */
1041 mop->mo_ldd.ldd_flags =
1042 LDD_F_SV_TYPE_OST | LDD_F_NEED_INDEX;
1043 verrprint("OST with unknown index\n");
1049 memcpy(mop->mo_ldd.ldd_uuid, lsd.lsd_uuid,
1050 sizeof(mop->mo_ldd.ldd_uuid));
1051 mop->mo_ldd.ldd_flags |= LDD_F_UPGRADE14;
1053 /* end COMPAT_146 */
1058 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
1059 run_command(cmd, cmdsz);
1061 verrprint("Failed to read old data (%d)\n", ret);
1066 void set_defaults(struct mkfs_opts *mop)
1068 mop->mo_ldd.ldd_magic = LDD_MAGIC;
1069 mop->mo_ldd.ldd_config_ver = 1;
1070 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
1071 mop->mo_mgs_failnodes = 0;
1072 strcpy(mop->mo_ldd.ldd_fsname, "lustre");
1073 if (get_os_version() == 24)
1074 mop->mo_ldd.ldd_mount_type = LDD_MT_EXT3;
1076 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
1078 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
1079 mop->mo_stripe_count = 1;
1082 static inline void badopt(const char *opt, char *type)
1084 fprintf(stderr, "%s: '--%s' only valid for %s\n",
1085 progname, opt, type);
1089 static int add_param(char *buf, char *key, char *val)
1091 int end = sizeof(((struct lustre_disk_data *)0)->ldd_params);
1092 int start = strlen(buf);
1096 keylen = strlen(key);
1097 if (start + 1 + keylen + strlen(val) >= end) {
1098 fprintf(stderr, "%s: params are too long-\n%s %s%s\n",
1099 progname, buf, key ? key : "", val);
1103 sprintf(buf + start, " %s%s", key ? key : "", val);
1107 /* from mount_lustre */
1108 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
1109 #define MAXNIDSTR 1024
1110 static char *convert_hostnames(char *s1)
1112 char *converted, *s2 = 0, *c, *end, sep;
1113 int left = MAXNIDSTR;
1116 converted = malloc(left);
1117 if (converted == NULL) {
1121 end = s1 + strlen(s1);
1123 while ((left > 0) && (s1 < end)) {
1124 s2 = strpbrk(s1, ",:");
1129 nid = libcfs_str2nid(s1);
1131 if (nid == LNET_NID_ANY) {
1132 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
1136 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
1137 strlen("127.0.0.1")) == 0) {
1138 fprintf(stderr, "%s: The NID '%s' resolves to the "
1139 "loopback address '%s'. Lustre requires a "
1140 "non-loopback address.\n",
1141 progname, s1, libcfs_nid2str(nid));
1146 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
1147 left = converted + MAXNIDSTR - c;
1153 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
1156 static struct option long_opt[] = {
1157 {"iam-dir", 0, 0, 'a'},
1158 {"backfstype", 1, 0, 'b'},
1159 {"stripe-count-hint", 1, 0, 'c'},
1160 {"comment", 1, 0, 'u'},
1161 {"configdev", 1, 0, 'C'},
1162 {"device-size", 1, 0, 'd'},
1163 {"dryrun", 0, 0, 'n'},
1164 {"erase-params", 0, 0, 'e'},
1165 {"failnode", 1, 0, 'f'},
1166 {"failover", 1, 0, 'f'},
1168 {"help", 0, 0, 'h'},
1169 {"index", 1, 0, 'i'},
1170 {"mkfsoptions", 1, 0, 'k'},
1171 {"mgsnode", 1, 0, 'm'},
1172 {"mgsnid", 1, 0, 'm'},
1174 {"fsname",1, 0, 'L'},
1175 {"noformat", 0, 0, 'n'},
1176 {"nomgs", 0, 0, 'N'},
1177 {"mountfsoptions", 1, 0, 'o'},
1179 {"param", 1, 0, 'p'},
1180 {"print", 0, 0, 'n'},
1181 {"quiet", 0, 0, 'q'},
1182 {"reformat", 0, 0, 'r'},
1183 {"verbose", 0, 0, 'v'},
1184 {"writeconf", 0, 0, 'w'},
1185 {"upgrade_to_18", 0, 0, 'U'},
1188 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqru:vw";
1192 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
1196 if (IS_MDT(&mop->mo_ldd))
1197 mop->mo_ldd.ldd_flags |= LDD_F_IAM_DIR;
1202 while (i < LDD_MT_LAST) {
1203 if (strcmp(optarg, mt_str(i)) == 0) {
1204 mop->mo_ldd.ldd_mount_type = i;
1212 if (IS_MDT(&mop->mo_ldd)) {
1213 int stripe_count = atol(optarg);
1214 if (stripe_count <= 0) {
1215 fprintf(stderr, "%s: bad stripe count "
1216 "%d\n", progname, stripe_count);
1219 mop->mo_stripe_count = stripe_count;
1221 badopt(long_opt[longidx].name, "MDT");
1225 case 'C': /* Configdev */
1227 printf("Configdev not implemented\n");
1230 mop->mo_device_sz = atol(optarg);
1233 mop->mo_ldd.ldd_params[0] = '\0';
1234 /* Must update the mgs logs */
1235 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1238 char *nids = convert_hostnames(optarg);
1241 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
1246 /* Must update the mgs logs */
1247 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1252 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
1258 if (!(mop->mo_ldd.ldd_flags &
1259 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1260 LDD_F_WRITECONF))) {
1261 fprintf(stderr, "%s: cannot change the index of"
1262 " a registered target\n", progname);
1265 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
1266 mop->mo_ldd.ldd_svindex = atol(optarg);
1267 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
1269 badopt(long_opt[longidx].name, "MDT,OST");
1274 strscpy(mop->mo_mkfsopts, optarg,
1275 sizeof(mop->mo_mkfsopts));
1279 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
1280 (!(mop->mo_ldd.ldd_flags &
1281 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1282 LDD_F_WRITECONF)))) {
1283 fprintf(stderr, "%s: cannot change the name of"
1284 " a registered target\n", progname);
1287 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
1288 fprintf(stderr, "%s: filesystem name must be "
1289 "1-8 chars\n", progname);
1292 if ((tmp = strpbrk(optarg, "/:"))) {
1293 fprintf(stderr, "%s: char '%c' not allowed in "
1294 "filesystem name\n", progname, *tmp);
1297 strscpy(mop->mo_ldd.ldd_fsname, optarg,
1298 sizeof(mop->mo_ldd.ldd_fsname));
1302 char *nids = convert_hostnames(optarg);
1305 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
1310 mop->mo_mgs_failnodes++;
1314 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
1320 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
1323 *mountopts = optarg;
1326 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
1329 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg);
1332 /* Must update the mgs logs */
1333 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1339 mop->mo_flags |= MO_FORCEFORMAT;
1342 strscpy(mop->mo_ldd.ldd_userdata, optarg,
1343 sizeof(mop->mo_ldd.ldd_userdata));
1349 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
1357 fprintf(stderr, "Unknown option '%c'\n", opt);
1363 /* Last arg is device */
1364 if (optind != argc - 1) {
1366 fprintf(stderr, "Bad argument: %s\n", argv[optind]);
1373 int main(int argc, char *const argv[])
1375 struct mkfs_opts mop;
1376 struct lustre_disk_data *ldd;
1377 char *mountopts = NULL;
1378 char always_mountopts[512] = "";
1379 char default_mountopts[512] = "";
1382 if ((progname = strrchr(argv[0], '/')) != NULL)
1387 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
1392 memset(&mop, 0, sizeof(mop));
1395 /* device is last arg */
1396 strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
1398 /* Are we using a loop device? */
1399 ret = is_block(mop.mo_device);
1403 mop.mo_flags |= MO_IS_LOOP;
1406 /* For tunefs, we must read in the old values before parsing any
1409 /* Check whether the disk has already been formatted by mkfs.lustre */
1410 ret = is_lustre_target(&mop);
1413 fprintf(stderr, "Device %s has not been formatted with "
1414 "mkfs.lustre\n", mop.mo_device);
1419 ret = read_local_files(&mop);
1422 fprintf(stderr, "Failed to read previous Lustre data from %s "
1423 "(%d)\n", mop.mo_device, ret);
1426 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
1427 mop.mo_mgs_failnodes++;
1430 print_ldd("Read previous values", &(mop.mo_ldd));
1433 ret = parse_opts(argc, argv, &mop, &mountopts);
1439 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
1441 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
1446 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
1448 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
1453 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
1454 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
1456 fprintf(stderr, "Can't find the target index, "
1457 "specify with --index\n");
1463 * Comment out these 2 checks temporarily, since for multi-MDSes
1464 * in single node only 1 mds node could have mgs service
1466 if (IS_MDT(ldd) && !IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1467 verrprint("No management node specified, adding MGS to this "
1469 ldd->ldd_flags |= LDD_F_SV_TYPE_MGS;
1471 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1474 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
1476 fprintf(stderr, "Must specify --mgsnode=\n");
1482 /* These are the permanent mount options (always included) */
1483 switch (ldd->ldd_mount_type) {
1485 case LDD_MT_LDISKFS:
1486 case LDD_MT_LDISKFS2: {
1487 sprintf(always_mountopts, "errors=remount-ro");
1488 if (IS_MDT(ldd) || IS_MGS(ldd))
1489 strscat(always_mountopts, ",iopen_nopriv,user_xattr",
1490 sizeof(always_mountopts));
1491 if ((get_os_version() == 24) && IS_OST(ldd))
1492 strscat(always_mountopts, ",asyncdel",
1493 sizeof(always_mountopts));
1494 /* NB: Files created while extents are enabled cannot be read
1495 if mounted with a kernel that doesn't include the Lustre ldiskfs
1498 (ldd->ldd_mount_type == LDD_MT_LDISKFS ||
1499 ldd->ldd_mount_type == LDD_MT_LDISKFS2)) {
1500 strscat(default_mountopts, ",extents,mballoc",
1501 sizeof(default_mountopts));
1506 mop.mo_flags |= MO_IS_LOOP;
1507 sprintf(always_mountopts, "type=ext3,dev=%s",
1513 fprintf(stderr, "unknown fs type %d '%s'\n",
1514 ldd->ldd_mount_type,
1522 /* If user specifies mount opts, don't use defaults,
1523 but always use always_mountopts */
1524 sprintf(ldd->ldd_mount_opts, "%s,%s",
1525 always_mountopts, mountopts);
1528 if (ldd->ldd_mount_opts[0] == 0)
1529 /* use the defaults unless old opts exist */
1532 sprintf(ldd->ldd_mount_opts, "%s%s",
1533 always_mountopts, default_mountopts);
1537 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
1538 ldd->ldd_fsname, ldd->ldd_svname);
1541 print_ldd("Permanent disk data", ldd);
1544 printf("exiting before disk write.\n");
1548 if (check_mtab_entry(mop.mo_device))
1551 /* Create the loopback file */
1552 if (mop.mo_flags & MO_IS_LOOP) {
1553 ret = access(mop.mo_device, F_OK);
1556 #ifndef TUNEFS /* mkfs.lustre */
1557 /* Reformat the loopback file */
1558 if (ret || (mop.mo_flags & MO_FORCEFORMAT))
1559 ret = loop_format(&mop);
1562 ret = loop_setup(&mop);
1565 fprintf(stderr, "Loop device setup for %s failed: %s\n",
1566 mop.mo_device, strerror(ret));
1571 #ifndef TUNEFS /* mkfs.lustre */
1572 /* Check whether the disk has already been formatted by mkfs.lustre */
1573 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
1574 ret = is_lustre_target(&mop);
1577 fprintf(stderr, "Device %s was previously formatted "
1578 "for lustre. Use --reformat to reformat it, "
1579 "or tunefs.lustre to modify.\n",
1585 /* Format the backing filesystem */
1586 ret = make_lustre_backfs(&mop);
1589 fprintf(stderr, "mkfs failed %d\n", ret);
1594 /* Write our config files */
1595 ret = write_local_files(&mop);
1598 fprintf(stderr, "failed to write local files\n");
1605 /* Fix any crazy return values from system() */
1606 if (ret && ((ret & 255) == 0))
1609 verrprint("%s: exiting with %d (%s)\n",
1610 progname, ret, strerror(ret));