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 */
45 #endif /* HAVE_CONFIG_H */
57 #include <sys/types.h>
59 #include <sys/mount.h>
60 #include <sys/utsname.h>
67 /* libcfs.h is not really needed here, but on SLES10/PPC, fs.h includes idr.h which
68 * requires BITS_PER_LONG to be defined */
69 #include <libcfs/libcfs.h>
70 #include <linux/fs.h> /* for BLKGETSIZE64 */
71 #include <linux/version.h>
73 #include <lustre_disk.h>
74 #include <lustre_param.h>
75 #include <lnet/lnetctl.h>
76 #include <lustre_ver.h>
77 #include "mount_utils.h"
83 #define MAX_LOOP_DEVICES 16
84 #define L_BLOCK_SIZE 4096
85 #define INDEX_UNASSIGNED 0xFFFF
86 #define MO_IS_LOOP 0x01
87 #define MO_FORCEFORMAT 0x02
89 /* used to describe the options to format the lustre disk, not persistent */
91 struct lustre_disk_data mo_ldd; /* to be written in MOUNT_DATA_FILE */
92 char mo_device[128]; /* disk device name */
93 char mo_mkfsopts[128]; /* options to the backing-store mkfs */
94 char mo_loopdev[128]; /* in case a loop dev is needed */
95 __u64 mo_device_sz; /* in KB */
103 static int print_only = 0;
104 static int failover = 0;
105 static int upgrade_to_18 = 0;
107 void usage(FILE *out)
109 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
110 fprintf(out, "usage: %s <target types> [options] <device>\n", progname);
112 "\t<device>:block device or file (e.g /dev/sda or /tmp/ost1)\n"
114 "\t\t--ost: object storage, mutually exclusive with mdt,mgs\n"
115 "\t\t--mdt: metadata storage, mutually exclusive with ost\n"
116 "\t\t--mgs: configuration management service - one per site\n"
117 "\toptions (in order of popularity):\n"
118 "\t\t--mgsnode=<nid>[,<...>] : NID(s) of a remote mgs node\n"
119 "\t\t\trequired for all targets other than the mgs node\n"
120 "\t\t--fsname=<filesystem_name> : default is 'lustre'\n"
121 "\t\t--failnode=<nid>[,<...>] : NID(s) of a failover partner\n"
122 "\t\t--param <key>=<value> : set a permanent parameter\n"
123 "\t\t\te.g. --param sys.timeout=40\n"
124 "\t\t\t --param lov.stripesize=2M\n"
125 "\t\t--index=#N : target index (i.e. ost index within the lov)\n"
126 /* FIXME implement 1.6.x
127 "\t\t--configdev=<altdevice|file>: store configuration info\n"
128 "\t\t\tfor this device on an alternate device\n"
130 "\t\t--comment=<user comment>: arbitrary user string (%d bytes)\n"
131 "\t\t--mountfsoptions=<opts> : permanent mount options\n"
133 "\t\t--backfstype=<fstype> : backing fs type (ext3, ldiskfs)\n"
134 "\t\t--device-size=#N(KB) : device size for loop devices\n"
135 "\t\t--mkfsoptions=<opts> : format options\n"
136 "\t\t--reformat: overwrite an existing disk\n"
137 "\t\t--stripe-count-hint=#N : used for optimizing MDT inode size\n"
138 "\t\t--iam-dir: make use of IAM directory format on backfs, incompatible with ext3.\n"
140 "\t\t--erase-params : erase all old parameter settings\n"
141 "\t\t--nomgs: turn off MGS service on this MDT\n"
142 "\t\t--writeconf: erase all config logs for this fs.\n"
144 "\t\t--dryrun: just report what we would do; "
145 "don't write to disk\n"
146 "\t\t--verbose : e.g. show mkfs progress\n"
148 (int)sizeof(((struct lustre_disk_data *)0)->ldd_userdata));
152 #define vprint if (verbose > 0) printf
153 #define verrprint if (verbose >= 0) printf
155 /*================ utility functions =====================*/
157 char *strscat(char *dst, char *src, int buflen) {
159 if (strlen(dst) + strlen(src) >= buflen) {
160 fprintf(stderr, "string buffer overflow (max %d): '%s' + '%s'"
161 "\n", buflen, dst, src);
164 return strcat(dst, src);
168 char *strscpy(char *dst, char *src, int buflen) {
170 return strscat(dst, src, buflen);
174 dev_major (unsigned long long int __dev)
176 return ((__dev >> 8) & 0xfff) | ((unsigned int) (__dev >> 32) & ~0xfff);
180 dev_minor (unsigned long long int __dev)
182 return (__dev & 0xff) | ((unsigned int) (__dev >> 12) & ~0xff);
187 static int version = 0;
191 char release[4] = "";
193 fd = open("/proc/sys/kernel/osrelease", O_RDONLY);
195 fprintf(stderr, "%s: Warning: Can't resolve kernel "
196 "version, assuming 2.6\n", progname);
198 read(fd, release, 4);
201 if (strncmp(release, "2.4.", 4) == 0)
209 static int check_mtab_entry(char *spec)
214 fp = setmntent(MOUNTED, "r");
218 while ((mnt = getmntent(fp)) != NULL) {
219 if (strcmp(mnt->mnt_fsname, spec) == 0) {
221 fprintf(stderr, "%s: according to %s %s is "
222 "already mounted on %s\n",
223 progname, MOUNTED, spec, mnt->mnt_dir);
232 /*============ disk dev functions ===================*/
234 /* Setup a file in the first unused loop_device */
235 int loop_setup(struct mkfs_opts *mop)
241 /* Figure out the loop device names */
242 if (!access("/dev/loop0", F_OK | R_OK)) {
243 strcpy(loop_base, "/dev/loop\0");
244 } else if (!access("/dev/loop/0", F_OK | R_OK)) {
245 strcpy(loop_base, "/dev/loop/\0");
247 fprintf(stderr, "%s: can't access loop devices\n", progname);
251 /* Find unused loop device */
252 for (i = 0; i < MAX_LOOP_DEVICES; i++) {
254 int cmdsz = sizeof(cmd);
256 sprintf(l_device, "%s%d", loop_base, i);
257 if (access(l_device, F_OK | R_OK))
259 snprintf(cmd, cmdsz, "losetup %s > /dev/null 2>&1", l_device);
262 /* losetup gets 1 (ret=256) for non-set-up device */
264 /* Set up a loopback device to our file */
265 snprintf(cmd, cmdsz, "losetup %s %s", l_device,
267 ret = run_command(cmd, cmdsz);
269 /* someone else picked up this loop device
273 fprintf(stderr, "%s: error %d on losetup: %s\n",
274 progname, ret, strerror(ret));
277 strscpy(mop->mo_loopdev, l_device,
278 sizeof(mop->mo_loopdev));
283 fprintf(stderr, "%s: out of loop devices!\n", progname);
287 int loop_cleanup(struct mkfs_opts *mop)
291 if ((mop->mo_flags & MO_IS_LOOP) && *mop->mo_loopdev) {
292 sprintf(cmd, "losetup -d %s", mop->mo_loopdev);
293 ret = run_command(cmd, sizeof(cmd));
298 /* Determine if a device is a block device (as opposed to a file) */
299 int is_block(char* devname)
304 ret = access(devname, F_OK);
307 ret = stat(devname, &st);
309 fprintf(stderr, "%s: cannot stat %s\n", progname, devname);
312 return S_ISBLK(st.st_mode);
315 __u64 get_device_size(char* device)
320 fd = open(device, O_RDONLY);
322 fprintf(stderr, "%s: cannot open %s: %s\n",
323 progname, device, strerror(errno));
328 /* size in bytes. bz5831 */
329 ret = ioctl(fd, BLKGETSIZE64, (void*)&size);
334 ret = ioctl(fd, BLKGETSIZE, (void*)&lsize);
335 size = (__u64)lsize * 512;
340 fprintf(stderr, "%s: size ioctl failed: %s\n",
341 progname, strerror(errno));
345 vprint("device size = "LPU64"MB\n", size >> 20);
346 /* return value in KB */
350 int loop_format(struct mkfs_opts *mop)
354 if (mop->mo_device_sz == 0) {
356 fprintf(stderr, "loop device requires a --device-size= "
361 ret = creat(mop->mo_device, S_IRUSR|S_IWUSR);
364 fprintf(stderr, "%s: Unable to create backing store: %d\n",
370 ret = truncate(mop->mo_device, mop->mo_device_sz * 1024);
373 fprintf(stderr, "%s: Unable to truncate backing store: %d\n",
380 /* Display the need for the latest e2fsprogs to be installed. make_backfs
381 * indicates if the caller is make_lustre_backfs() or not. */
382 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
384 static int msg_displayed;
387 fprintf(stderr, "WARNING: %s does not support %s "
388 "feature.\n\n", E2FSPROGS, feature);
394 fprintf(stderr, "WARNING: The %s package currently installed on "
395 "your system does not support \"%s\" feature.\n",
397 #if !(HAVE_LDISKFSPROGS)
398 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
399 "http://downloads.lustre.org/public/tools/e2fsprogs/\n"
400 "to enable this feature.\n");
403 fprintf(stderr, "Feature will not be enabled until %s"
404 "is updated and '%s -O %s %%{device}' "
405 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
408 /* Check whether the file exists in the device */
409 static int file_in_dev(char *file_name, char *dev_name)
412 char debugfs_cmd[256];
413 unsigned int inode_num;
416 /* Construct debugfs command line. */
417 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
418 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
419 DEBUGFS, file_name, dev_name);
421 fp = popen(debugfs_cmd, "r");
423 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
427 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
431 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd), fp);
434 fprintf(stderr, "%s", debugfs_cmd);
435 if (strstr(debugfs_cmd, "unsupported feature")) {
436 disp_old_e2fsprogs_msg("an unknown", 0);
445 /* Check whether the device has already been used with lustre */
446 static int is_lustre_target(struct mkfs_opts *mop)
450 vprint("checking for existing Lustre data: ");
452 if ((rc = file_in_dev(MOUNT_DATA_FILE, mop->mo_device))) {
454 (rc == 1) ? MOUNT_DATA_FILE : "extents");
455 /* in the -1 case, 'extents' means this really IS a lustre
460 if ((rc = file_in_dev(LAST_RCVD, mop->mo_device))) {
461 vprint("found %s\n", LAST_RCVD);
465 vprint("not found\n");
466 return 0; /* The device is not a lustre target. */
469 /* Check if a certain feature is supported by e2fsprogs.
470 * Firstly we try to use "debugfs supported_features" command to check if
471 * the feature is supported. If this fails we try to set this feature with
472 * mke2fs to check for its support. */
473 static int is_e2fsprogs_feature_supp(const char *feature)
477 char imgname[] = "/tmp/test-img-XXXXXX";
481 snprintf(cmd, sizeof(cmd), "%s -c -R \"supported_features %s\" 2>&1",
484 /* Using popen() instead of run_command() since debugfs does not return
485 * proper error code if command is not supported */
486 fp = popen(cmd, "r");
488 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
491 ret = fread(cmd, 1, sizeof(cmd), fp);
493 if (strstr(cmd, feature) && !(strstr(cmd, "Unknown")))
497 if ((fd = mkstemp(imgname)) < 0)
500 snprintf(cmd, sizeof(cmd), "%s -F -O %s %s 100 >/dev/null 2>&1",
501 MKE2FS, feature, imgname);
502 /* run_command() displays the output of mke2fs when it fails for
503 * some feature, so use system() directly */
511 static void disp_old_kernel_msg(char *feature)
513 fprintf(stderr, "WARNING: ldiskfs filesystem does not support \"%s\" "
514 "feature.\n\n", feature);
517 static void enable_default_backfs_features(struct mkfs_opts *mop)
520 int maj_high, maj_low, min;
523 strscat(mop->mo_mkfsopts, " -O dir_index,extents", sizeof(mop->mo_mkfsopts));
525 /* Upstream e2fsprogs called our uninit_groups feature uninit_bg,
526 * check for both of them when testing e2fsprogs features. */
527 if (is_e2fsprogs_feature_supp("uninit_groups") == 0)
528 strscat(mop->mo_mkfsopts, ",uninit_groups",
529 sizeof(mop->mo_mkfsopts));
530 else if (is_e2fsprogs_feature_supp("uninit_bg") == 0)
531 strscat(mop->mo_mkfsopts, ",uninit_bg",
532 sizeof(mop->mo_mkfsopts));
534 disp_old_e2fsprogs_msg("uninit_bg", 1);
540 sscanf(uts.release, "%d.%d.%d", &maj_high, &maj_low, &min);
541 printf("%d %d %d\n", maj_high, maj_low, min);
543 /* Multiple mount protection is enabled only if failover node is
544 * specified and if kernel version is higher than 2.6.9 */
546 if (KERNEL_VERSION(maj_high, maj_low, min) >=
547 KERNEL_VERSION(2,6,9)) {
548 if (is_e2fsprogs_feature_supp("mmp") == 0)
549 strscat(mop->mo_mkfsopts, ",mmp",
550 sizeof(mop->mo_mkfsopts));
552 disp_old_e2fsprogs_msg("mmp", 1);
554 disp_old_kernel_msg("mmp");
558 /* Build fs according to type */
559 int make_lustre_backfs(struct mkfs_opts *mop)
561 char mkfs_cmd[PATH_MAX];
567 if (mop->mo_device_sz != 0) {
568 if (mop->mo_device_sz < 8096){
569 fprintf(stderr, "%s: size of filesystem must be larger "
570 "than 8MB, but is set to %lldKB\n",
571 progname, (long long)mop->mo_device_sz);
574 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
577 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
578 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
579 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
580 __u64 device_sz = mop->mo_device_sz;
582 /* we really need the size */
583 if (device_sz == 0) {
584 device_sz = get_device_size(mop->mo_device);
589 /* Journal size in MB */
590 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
591 /* Choose our own default journal size */
592 long journal_sz = 0, max_sz;
593 if (device_sz > 1024 * 1024) /* 1GB */
594 journal_sz = (device_sz / 102400) * 4;
595 /* cap journal size at 1GB */
596 if (journal_sz > 1024L)
599 max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
600 if (journal_sz > max_sz)
603 sprintf(buf, " -J size=%ld", journal_sz);
604 strscat(mop->mo_mkfsopts, buf,
605 sizeof(mop->mo_mkfsopts));
609 /* Bytes_per_inode: disk size / num inodes */
610 if (strstr(mop->mo_mkfsopts, "-i") == NULL) {
611 long bytes_per_inode = 0;
613 if (IS_MDT(&mop->mo_ldd))
614 bytes_per_inode = 4096;
616 /* Allocate fewer inodes on large OST devices. Most
617 filesystems can be much more aggressive than even
619 if ((IS_OST(&mop->mo_ldd) && (device_sz > 100000000)))
620 bytes_per_inode = 16384; /* > 100 Gb device */
623 if (bytes_per_inode > 0) {
624 sprintf(buf, " -i %ld", bytes_per_inode);
625 strscat(mop->mo_mkfsopts, buf,
626 sizeof(mop->mo_mkfsopts));
630 /* Inode size (for extended attributes). The LOV EA size is
631 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
632 * and we want some margin above that for ACLs, other EAs... */
633 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
635 if (IS_MDT(&mop->mo_ldd)) {
636 if (mop->mo_stripe_count > 72)
637 inode_size = 512; /* bz 7241 */
638 /* cray stripes across all osts (>60) */
639 else if (mop->mo_stripe_count > 32)
641 else if (mop->mo_stripe_count > 10)
645 } else if (IS_OST(&mop->mo_ldd)) {
646 /* now as we store fids in EA on OST we need
647 to make inode bigger */
651 if (inode_size > 0) {
652 sprintf(buf, " -I %ld", inode_size);
653 strscat(mop->mo_mkfsopts, buf,
654 sizeof(mop->mo_mkfsopts));
659 strscat(mop->mo_mkfsopts, " -q",
660 sizeof(mop->mo_mkfsopts));
663 if (strstr(mop->mo_mkfsopts, "-O") == NULL)
664 enable_default_backfs_features(mop);
666 /* Allow reformat of full devices (as opposed to
667 partitions.) We already checked for mounted dev. */
668 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
670 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
671 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
672 mop->mo_ldd.ldd_svname);
673 } else if (mop->mo_ldd.ldd_mount_type == LDD_MT_REISERFS) {
674 long journal_sz = 0; /* FIXME default journal size */
675 if (journal_sz > 0) {
676 sprintf(buf, " --journal_size %ld", journal_sz);
677 strscat(mop->mo_mkfsopts, buf,
678 sizeof(mop->mo_mkfsopts));
680 snprintf(mkfs_cmd, sizeof(mkfs_cmd), "mkreiserfs -ff ");
682 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
683 progname, mop->mo_ldd.ldd_mount_type,
684 MT_STR(&mop->mo_ldd));
688 /* For loop device format the dev, not the filename */
689 dev = mop->mo_device;
690 if (mop->mo_flags & MO_IS_LOOP)
691 dev = mop->mo_loopdev;
693 vprint("formatting backing filesystem %s on %s\n",
694 MT_STR(&mop->mo_ldd), dev);
695 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
696 vprint("\t4k blocks %d\n", block_count);
697 vprint("\toptions %s\n", mop->mo_mkfsopts);
699 /* mkfs_cmd's trailing space is important! */
700 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
701 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
702 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
703 if (block_count != 0) {
704 sprintf(buf, " %d", block_count);
705 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
708 vprint("mkfs_cmd = %s\n", mkfs_cmd);
709 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
712 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
717 /* ==================== Lustre config functions =============*/
719 void print_ldd(char *str, struct lustre_disk_data *ldd)
721 printf("\n %s:\n", str);
722 printf("Target: %s\n", ldd->ldd_svname);
723 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
724 printf("Index: unassigned\n");
726 printf("Index: %d\n", ldd->ldd_svindex);
727 if (ldd->ldd_uuid[0])
728 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
729 printf("Lustre FS: %s\n", ldd->ldd_fsname);
730 printf("Mount type: %s\n", MT_STR(ldd));
731 printf("Flags: %#x\n", ldd->ldd_flags);
732 printf(" (%s%s%s%s%s%s%s%s%s)\n",
733 IS_MDT(ldd) ? "MDT ":"",
734 IS_OST(ldd) ? "OST ":"",
735 IS_MGS(ldd) ? "MGS ":"",
736 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
737 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
738 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
739 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
740 ldd->ldd_flags & LDD_F_IAM_DIR ? "IAM_dir_format ":"",
741 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
742 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
743 printf("Parameters:%s\n", ldd->ldd_params);
744 if (ldd->ldd_userdata[0])
745 printf("Comment: %s\n", ldd->ldd_userdata);
749 static int touch_file(char *filename)
753 if (filename == NULL) {
757 fd = open(filename, O_CREAT | O_TRUNC, 0600);
766 /* keep it less than LL_FID_NAMELEN */
767 #define DUMMY_FILE_NAME_LEN 25
768 #define EXT3_DIRENT_SIZE DUMMY_FILE_NAME_LEN
770 /* Need to add these many entries to this directory to make HTREE dir. */
771 #define MIN_ENTRIES_REQ_FOR_HTREE ((L_BLOCK_SIZE / EXT3_DIRENT_SIZE))
773 static int add_dummy_files(char *dir)
775 char fpname[PATH_MAX];
779 for (i = 0; i < MIN_ENTRIES_REQ_FOR_HTREE; i++) {
780 snprintf(fpname, PATH_MAX, "%s/%0*d", dir,
781 DUMMY_FILE_NAME_LEN, i);
783 rc = touch_file(fpname);
784 if (rc && rc != -EEXIST) {
786 "%s: Can't create dummy file %s: %s\n",
787 progname, fpname , strerror(errno));
794 static int __l_mkdir(char * filepnm, int mode , struct mkfs_opts *mop)
798 ret = mkdir(filepnm, mode);
799 if (ret && ret != -EEXIST)
802 /* IAM mode supports ext3 directories of HTREE type only. So add dummy
803 * entries to new directory to create htree type of container for
805 if (mop->mo_ldd.ldd_flags & LDD_F_IAM_DIR)
806 return add_dummy_files(filepnm);
810 /* Write the server config files */
811 int write_local_files(struct mkfs_opts *mop)
813 char mntpt[] = "/tmp/mntXXXXXX";
819 /* Mount this device temporarily in order to write these files */
820 if (!mkdtemp(mntpt)) {
821 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
822 progname, mntpt, strerror(errno));
826 dev = mop->mo_device;
827 if (mop->mo_flags & MO_IS_LOOP)
828 dev = mop->mo_loopdev;
830 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0, NULL);
832 fprintf(stderr, "%s: Unable to mount %s: %s\n",
833 progname, dev, strerror(errno));
835 if (errno == ENODEV) {
836 fprintf(stderr, "Is the %s module available?\n",
837 MT_STR(&mop->mo_ldd));
842 /* Set up initial directories */
843 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
844 ret = __l_mkdir(filepnm, 0777, mop);
845 if ((ret != 0) && (errno != EEXIST)) {
846 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
847 progname, filepnm, strerror(errno));
849 } else if (errno == EEXIST) {
853 /* Save the persistent mount data into a file. Lustre must pre-read
854 this file to get the real mount options. */
855 vprint("Writing %s\n", MOUNT_DATA_FILE);
856 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
857 filep = fopen(filepnm, "w");
859 fprintf(stderr, "%s: Unable to create %s file: %s\n",
860 progname, filepnm, strerror(errno));
863 fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
867 /* Check for upgrade */
868 if ((mop->mo_ldd.ldd_flags & (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS))
869 == (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS)) {
872 int cmdsz = sizeof(cmd);
873 vprint("Copying old logs\n");
875 /* Copy the old client log to fsname-client */
876 sprintf(filepnm, "%s/%s/%s-client",
877 mntpt, MOUNT_CONFIGS_DIR, mop->mo_ldd.ldd_fsname);
878 snprintf(cmd, cmdsz, "cp %s/%s/client %s", mntpt, MDT_LOGS_DIR,
880 ret = run_command(cmd, cmdsz);
882 fprintf(stderr, "%s: Can't copy 1.4 config %s/client "
883 "(%d)\n", progname, MDT_LOGS_DIR, ret);
884 fprintf(stderr, "mount -t ldiskfs %s somewhere, "
885 "find the client log for fs %s and "
886 "copy it manually into %s/%s-client, "
889 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
890 mop->mo_ldd.ldd_fsname);
894 /* We need to use the old mdt log because otherwise mdt won't
895 have complete lov if old clients connect before all
897 /* Copy the old mdt log to fsname-MDT0000 (get old
898 name from mdt_UUID) */
900 strscpy(filepnm, (char *)mop->mo_ldd.ldd_uuid, sizeof(filepnm));
901 term = strstr(filepnm, "_UUID");
904 snprintf(cmd, cmdsz, "cp %s/%s/%s %s/%s/%s",
905 mntpt, MDT_LOGS_DIR, filepnm,
906 mntpt, MOUNT_CONFIGS_DIR,
907 mop->mo_ldd.ldd_svname);
908 ret = run_command(cmd, cmdsz);
911 fprintf(stderr, "%s: Can't copy 1.4 config %s/%s "
912 "(%d)\n", progname, MDT_LOGS_DIR, filepnm, ret);
913 fprintf(stderr, "mount -t ext3 %s somewhere, "
914 "find the MDT log for fs %s and "
915 "copy it manually into %s/%s, "
918 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
919 mop->mo_ldd.ldd_svname);
933 int read_local_files(struct mkfs_opts *mop)
935 char tmpdir[] = "/tmp/dirXXXXXX";
941 int cmdsz = sizeof(cmd);
943 /* Make a temporary directory to hold Lustre data files. */
944 if (!mkdtemp(tmpdir)) {
945 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
946 progname, tmpdir, strerror(errno));
950 dev = mop->mo_device;
952 /* TODO: it's worth observing the get_mountdata() function that is
953 in mount_utils.c for getting the mountdata out of the
956 /* Construct debugfs command line. */
957 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
958 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
960 ret = run_command(cmd, cmdsz);
962 verrprint("%s: Unable to dump %s dir (%d)\n",
963 progname, MOUNT_CONFIGS_DIR, ret);
965 sprintf(filepnm, "%s/mountdata", tmpdir);
966 filep = fopen(filepnm, "r");
968 vprint("Reading %s\n", MOUNT_DATA_FILE);
969 fread(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
972 /* Try to read pre-1.6 config from last_rcvd */
973 struct lr_server_data lsd;
974 verrprint("%s: Unable to read %d.%d config %s.\n",
975 progname, LUSTRE_MAJOR, LUSTRE_MINOR, filepnm);
977 verrprint("Trying 1.4 config from last_rcvd\n");
978 sprintf(filepnm, "%s/%s", tmpdir, LAST_RCVD);
980 /* Construct debugfs command line. */
981 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s' %s",
982 DEBUGFS, LAST_RCVD, filepnm, dev);
984 ret = run_command(cmd, cmdsz);
986 fprintf(stderr, "%s: Unable to dump %s file (%d)\n",
987 progname, LAST_RCVD, ret);
991 filep = fopen(filepnm, "r");
993 fprintf(stderr, "%s: Unable to open %s: %s\n",
994 progname, filepnm, strerror(errno));
996 verrprint("Contents of %s:\n", tmpdir);
998 snprintf(cmd, cmdsz, "ls -l %s/", tmpdir);
999 run_command(cmd, cmdsz);
1000 verrprint("Contents of disk:\n");
1001 snprintf(cmd, cmdsz, "%s -c -R 'ls -l /' %s",
1003 run_command(cmd, cmdsz);
1007 vprint("Reading %s\n", LAST_RCVD);
1008 ret = fread(&lsd, 1, sizeof(lsd), filep);
1009 if (ret < sizeof(lsd)) {
1010 fprintf(stderr, "%s: Short read (%d of %d)\n",
1011 progname, ret, (int)sizeof(lsd));
1012 ret = ferror(filep);
1016 vprint("Feature compat=%x, incompat=%x\n",
1017 lsd.lsd_feature_compat, lsd.lsd_feature_incompat);
1019 if ((lsd.lsd_feature_compat & OBD_COMPAT_OST) ||
1020 (lsd.lsd_feature_incompat & OBD_INCOMPAT_OST)) {
1021 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
1022 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
1023 } else if ((lsd.lsd_feature_compat & OBD_COMPAT_MDT) ||
1024 (lsd.lsd_feature_incompat & OBD_INCOMPAT_MDT)) {
1025 /* We must co-locate so mgs can see old logs.
1026 If user doesn't want this, they can copy the old
1027 logs manually and re-tunefs. */
1028 mop->mo_ldd.ldd_flags =
1029 LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_MGS;
1030 mop->mo_ldd.ldd_svindex = lsd.lsd_mdt_index;
1032 /* If neither is set, we're pre-1.4.6, make a guess. */
1033 /* Construct debugfs command line. */
1034 snprintf(cmd, cmdsz, "%s -c -R 'rdump /%s %s' %s",
1035 DEBUGFS, MDT_LOGS_DIR, tmpdir, dev);
1036 run_command(cmd, cmdsz);
1038 sprintf(filepnm, "%s/%s", tmpdir, MDT_LOGS_DIR);
1039 if (lsd.lsd_ost_index > 0) {
1040 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
1041 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
1043 /* If there's a LOGS dir, it's an MDT */
1044 if ((ret = access(filepnm, F_OK)) == 0) {
1045 mop->mo_ldd.ldd_flags =
1048 /* Old MDT's are always index 0
1050 mop->mo_ldd.ldd_svindex = 0;
1052 /* The index may not be correct */
1053 mop->mo_ldd.ldd_flags =
1054 LDD_F_SV_TYPE_OST | LDD_F_NEED_INDEX;
1055 verrprint("OST with unknown index\n");
1061 memcpy(mop->mo_ldd.ldd_uuid, lsd.lsd_uuid,
1062 sizeof(mop->mo_ldd.ldd_uuid));
1063 mop->mo_ldd.ldd_flags |= LDD_F_UPGRADE14;
1065 /* end COMPAT_146 */
1070 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
1071 run_command(cmd, cmdsz);
1073 verrprint("Failed to read old data (%d)\n", ret);
1078 void set_defaults(struct mkfs_opts *mop)
1080 mop->mo_ldd.ldd_magic = LDD_MAGIC;
1081 mop->mo_ldd.ldd_config_ver = 1;
1082 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
1083 mop->mo_mgs_failnodes = 0;
1084 strcpy(mop->mo_ldd.ldd_fsname, "lustre");
1085 if (get_os_version() == 24)
1086 mop->mo_ldd.ldd_mount_type = LDD_MT_EXT3;
1088 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
1090 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
1091 mop->mo_stripe_count = 1;
1094 static inline void badopt(const char *opt, char *type)
1096 fprintf(stderr, "%s: '--%s' only valid for %s\n",
1097 progname, opt, type);
1101 static int add_param(char *buf, char *key, char *val)
1103 int end = sizeof(((struct lustre_disk_data *)0)->ldd_params);
1104 int start = strlen(buf);
1108 keylen = strlen(key);
1109 if (start + 1 + keylen + strlen(val) >= end) {
1110 fprintf(stderr, "%s: params are too long-\n%s %s%s\n",
1111 progname, buf, key ? key : "", val);
1115 sprintf(buf + start, " %s%s", key ? key : "", val);
1119 /* from mount_lustre */
1120 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
1121 #define MAXNIDSTR 1024
1122 static char *convert_hostnames(char *s1)
1124 char *converted, *s2 = 0, *c, *end, sep;
1125 int left = MAXNIDSTR;
1128 converted = malloc(left);
1129 if (converted == NULL) {
1133 end = s1 + strlen(s1);
1135 while ((left > 0) && (s1 < end)) {
1136 s2 = strpbrk(s1, ",:");
1141 nid = libcfs_str2nid(s1);
1143 if (nid == LNET_NID_ANY) {
1144 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
1148 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
1149 strlen("127.0.0.1")) == 0) {
1150 fprintf(stderr, "%s: The NID '%s' resolves to the "
1151 "loopback address '%s'. Lustre requires a "
1152 "non-loopback address.\n",
1153 progname, s1, libcfs_nid2str(nid));
1158 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
1159 left = converted + MAXNIDSTR - c;
1165 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
1168 static struct option long_opt[] = {
1169 {"iam-dir", 0, 0, 'a'},
1170 {"backfstype", 1, 0, 'b'},
1171 {"stripe-count-hint", 1, 0, 'c'},
1172 {"comment", 1, 0, 'u'},
1173 {"configdev", 1, 0, 'C'},
1174 {"device-size", 1, 0, 'd'},
1175 {"dryrun", 0, 0, 'n'},
1176 {"erase-params", 0, 0, 'e'},
1177 {"failnode", 1, 0, 'f'},
1178 {"failover", 1, 0, 'f'},
1180 {"help", 0, 0, 'h'},
1181 {"index", 1, 0, 'i'},
1182 {"mkfsoptions", 1, 0, 'k'},
1183 {"mgsnode", 1, 0, 'm'},
1184 {"mgsnid", 1, 0, 'm'},
1186 {"fsname",1, 0, 'L'},
1187 {"noformat", 0, 0, 'n'},
1188 {"nomgs", 0, 0, 'N'},
1189 {"mountfsoptions", 1, 0, 'o'},
1191 {"param", 1, 0, 'p'},
1192 {"print", 0, 0, 'n'},
1193 {"quiet", 0, 0, 'q'},
1194 {"reformat", 0, 0, 'r'},
1195 {"verbose", 0, 0, 'v'},
1196 {"writeconf", 0, 0, 'w'},
1197 {"upgrade_to_18", 0, 0, 'U'},
1200 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqru:vw";
1204 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
1208 if (IS_MDT(&mop->mo_ldd))
1209 mop->mo_ldd.ldd_flags |= LDD_F_IAM_DIR;
1214 while (i < LDD_MT_LAST) {
1215 if (strcmp(optarg, mt_str(i)) == 0) {
1216 mop->mo_ldd.ldd_mount_type = i;
1224 if (IS_MDT(&mop->mo_ldd)) {
1225 int stripe_count = atol(optarg);
1226 if (stripe_count <= 0) {
1227 fprintf(stderr, "%s: bad stripe count "
1228 "%d\n", progname, stripe_count);
1231 mop->mo_stripe_count = stripe_count;
1233 badopt(long_opt[longidx].name, "MDT");
1237 case 'C': /* Configdev */
1239 printf("Configdev not implemented\n");
1242 mop->mo_device_sz = atol(optarg);
1245 mop->mo_ldd.ldd_params[0] = '\0';
1246 /* Must update the mgs logs */
1247 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1250 char *nids = convert_hostnames(optarg);
1253 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
1258 /* Must update the mgs logs */
1259 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1264 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
1270 if (!(mop->mo_ldd.ldd_flags &
1271 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1272 LDD_F_WRITECONF))) {
1273 fprintf(stderr, "%s: cannot change the index of"
1274 " a registered target\n", progname);
1277 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
1278 mop->mo_ldd.ldd_svindex = atol(optarg);
1279 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
1281 badopt(long_opt[longidx].name, "MDT,OST");
1286 strscpy(mop->mo_mkfsopts, optarg,
1287 sizeof(mop->mo_mkfsopts));
1291 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
1292 (!(mop->mo_ldd.ldd_flags &
1293 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1294 LDD_F_WRITECONF)))) {
1295 fprintf(stderr, "%s: cannot change the name of"
1296 " a registered target\n", progname);
1299 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
1300 fprintf(stderr, "%s: filesystem name must be "
1301 "1-8 chars\n", progname);
1304 if ((tmp = strpbrk(optarg, "/:"))) {
1305 fprintf(stderr, "%s: char '%c' not allowed in "
1306 "filesystem name\n", progname, *tmp);
1309 strscpy(mop->mo_ldd.ldd_fsname, optarg,
1310 sizeof(mop->mo_ldd.ldd_fsname));
1314 char *nids = convert_hostnames(optarg);
1317 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
1322 mop->mo_mgs_failnodes++;
1326 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
1332 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
1335 *mountopts = optarg;
1338 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
1341 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg);
1344 /* Must update the mgs logs */
1345 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1351 mop->mo_flags |= MO_FORCEFORMAT;
1354 strscpy(mop->mo_ldd.ldd_userdata, optarg,
1355 sizeof(mop->mo_ldd.ldd_userdata));
1361 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
1369 fprintf(stderr, "Unknown option '%c'\n", opt);
1375 /* Last arg is device */
1376 if (optind != argc - 1) {
1378 fprintf(stderr, "Bad argument: %s\n", argv[optind]);
1382 /* single argument: <device> */
1389 int main(int argc, char *const argv[])
1391 struct mkfs_opts mop;
1392 struct lustre_disk_data *ldd;
1393 char *mountopts = NULL;
1394 char always_mountopts[512] = "";
1395 char default_mountopts[512] = "";
1398 if ((progname = strrchr(argv[0], '/')) != NULL)
1403 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
1408 memset(&mop, 0, sizeof(mop));
1411 /* device is last arg */
1412 strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
1414 /* Are we using a loop device? */
1415 ret = is_block(mop.mo_device);
1419 mop.mo_flags |= MO_IS_LOOP;
1422 /* For tunefs, we must read in the old values before parsing any
1425 /* Check whether the disk has already been formatted by mkfs.lustre */
1426 ret = is_lustre_target(&mop);
1429 fprintf(stderr, "Device %s has not been formatted with "
1430 "mkfs.lustre\n", mop.mo_device);
1435 ret = read_local_files(&mop);
1438 fprintf(stderr, "Failed to read previous Lustre data from %s "
1439 "(%d)\n", mop.mo_device, ret);
1442 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
1443 mop.mo_mgs_failnodes++;
1446 print_ldd("Read previous values", &(mop.mo_ldd));
1449 ret = parse_opts(argc, argv, &mop, &mountopts);
1455 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
1457 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
1462 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
1464 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
1469 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
1470 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
1472 fprintf(stderr, "Can't find the target index, "
1473 "specify with --index\n");
1479 * Comment out these 2 checks temporarily, since for multi-MDSes
1480 * in single node only 1 mds node could have mgs service
1482 if (IS_MDT(ldd) && !IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1483 verrprint("No management node specified, adding MGS to this "
1485 ldd->ldd_flags |= LDD_F_SV_TYPE_MGS;
1487 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1490 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
1492 fprintf(stderr, "Must specify --mgsnode=\n");
1498 /* These are the permanent mount options (always included) */
1499 switch (ldd->ldd_mount_type) {
1501 case LDD_MT_LDISKFS:
1502 case LDD_MT_LDISKFS2: {
1503 sprintf(always_mountopts, "errors=remount-ro");
1504 if (IS_MDT(ldd) || IS_MGS(ldd))
1505 strscat(always_mountopts, ",iopen_nopriv,user_xattr",
1506 sizeof(always_mountopts));
1507 if ((get_os_version() == 24) && IS_OST(ldd))
1508 strscat(always_mountopts, ",asyncdel",
1509 sizeof(always_mountopts));
1510 /* NB: Files created while extents are enabled cannot be read
1511 if mounted with a kernel that doesn't include the Lustre ldiskfs
1514 (ldd->ldd_mount_type == LDD_MT_LDISKFS ||
1515 ldd->ldd_mount_type == LDD_MT_LDISKFS2)) {
1516 strscat(default_mountopts, ",extents,mballoc",
1517 sizeof(default_mountopts));
1522 mop.mo_flags |= MO_IS_LOOP;
1523 sprintf(always_mountopts, "type=ext3,dev=%s",
1529 fprintf(stderr, "unknown fs type %d '%s'\n",
1530 ldd->ldd_mount_type,
1538 /* If user specifies mount opts, don't use defaults,
1539 but always use always_mountopts */
1540 sprintf(ldd->ldd_mount_opts, "%s,%s",
1541 always_mountopts, mountopts);
1544 if (ldd->ldd_mount_opts[0] == 0)
1545 /* use the defaults unless old opts exist */
1548 sprintf(ldd->ldd_mount_opts, "%s%s",
1549 always_mountopts, default_mountopts);
1553 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
1554 ldd->ldd_fsname, ldd->ldd_svname);
1557 print_ldd("Permanent disk data", ldd);
1560 printf("exiting before disk write.\n");
1564 if (check_mtab_entry(mop.mo_device))
1567 /* Create the loopback file */
1568 if (mop.mo_flags & MO_IS_LOOP) {
1569 ret = access(mop.mo_device, F_OK);
1572 #ifndef TUNEFS /* mkfs.lustre */
1573 /* Reformat the loopback file */
1574 if (ret || (mop.mo_flags & MO_FORCEFORMAT))
1575 ret = loop_format(&mop);
1578 ret = loop_setup(&mop);
1581 fprintf(stderr, "Loop device setup for %s failed: %s\n",
1582 mop.mo_device, strerror(ret));
1587 #ifndef TUNEFS /* mkfs.lustre */
1588 /* Check whether the disk has already been formatted by mkfs.lustre */
1589 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
1590 ret = is_lustre_target(&mop);
1593 fprintf(stderr, "Device %s was previously formatted "
1594 "for lustre. Use --reformat to reformat it, "
1595 "or tunefs.lustre to modify.\n",
1601 /* Format the backing filesystem */
1602 ret = make_lustre_backfs(&mop);
1605 fprintf(stderr, "mkfs failed %d\n", ret);
1610 /* Write our config files */
1611 ret = write_local_files(&mop);
1614 fprintf(stderr, "failed to write local files\n");
1621 /* Fix any crazy return values from system() */
1622 if (ret && ((ret & 255) == 0))
1625 verrprint("%s: exiting with %d (%s)\n",
1626 progname, ret, strerror(ret));