1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2006 Cluster File Systems, Inc.
5 * Author: Nathan Rutman <nathan@clusterfs.com>
7 * This file is part of Lustre, http://www.lustre.org.
9 * Lustre is free software; you can redistribute it and/or
10 * modify it under the terms of version 2 of the GNU General Public
11 * License as published by the Free Software Foundation.
13 * Lustre is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with Lustre; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 /* This source file is compiled into both mkfs.lustre and tunefs.lustre */
33 #include <sys/types.h>
35 #include <sys/mount.h>
42 # include <linux/fs.h> /* for BLKGETSIZE64 */
44 #include <lustre_disk.h>
45 #include <lustre_param.h>
46 #include <lnet/lnetctl.h>
47 #include <lustre_ver.h>
53 #define MAX_LOOP_DEVICES 16
54 #define L_BLOCK_SIZE 4096
55 #define INDEX_UNASSIGNED 0xFFFF
56 #define MO_IS_LOOP 0x01
57 #define MO_FORCEFORMAT 0x02
59 /* used to describe the options to format the lustre disk, not persistent */
61 struct lustre_disk_data mo_ldd; /* to be written in MOUNT_DATA_FILE */
62 char mo_device[128]; /* disk device name */
63 char mo_mkfsopts[128]; /* options to the backing-store mkfs */
64 char mo_loopdev[128]; /* in case a loop dev is needed */
65 __u64 mo_device_sz; /* in KB */
71 static char *progname;
72 static int verbose = 1;
73 static int print_only = 0;
78 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
79 fprintf(out, "usage: %s <target types> [options] <device>\n", progname);
81 "\t<device>:block device or file (e.g /dev/sda or /tmp/ost1)\n"
83 "\t\t--ost: object storage, mutually exclusive with mdt,mgs\n"
84 "\t\t--mdt: metadata storage, mutually exclusive with ost\n"
85 "\t\t--mgs: configuration management service - one per site\n"
86 "\toptions (in order of popularity):\n"
87 "\t\t--mgsnode=<nid>[,<...>] : NID(s) of a remote mgs node\n"
88 "\t\t\trequired for all targets other than the mgs node\n"
89 "\t\t--fsname=<filesystem_name> : default is 'lustre'\n"
90 "\t\t--failnode=<nid>[,<...>] : NID(s) of a failover partner\n"
91 "\t\t--param <key>=<value> : set a permanent parameter\n"
92 "\t\t\te.g. --param sys.timeout=40\n"
93 "\t\t\t --param lov.stripesize=2M\n"
94 "\t\t--index=#N : target index (i.e. ost index within the lov)\n"
95 /* FIXME implement 1.6.x
96 "\t\t--configdev=<altdevice|file>: store configuration info\n"
97 "\t\t\tfor this device on an alternate device\n"
99 "\t\t--comment=<user comment>: arbitrary user string (%d bytes)\n"
100 "\t\t--mountfsoptions=<opts> : permanent mount options\n"
102 "\t\t--backfstype=<fstype> : backing fs type (ext3, ldiskfs)\n"
103 "\t\t--device-size=#N(KB) : device size for loop devices\n"
104 "\t\t--mkfsoptions=<opts> : format options\n"
105 "\t\t--reformat: overwrite an existing disk\n"
106 "\t\t--stripe-count-hint=#N : used for optimizing MDT inode size\n"
108 "\t\t--erase-params : erase all old parameter settings\n"
109 "\t\t--nomgs: turn off MGS service on this MDT\n"
110 "\t\t--writeconf: erase all config logs for this fs.\n"
112 "\t\t--dryrun: just report what we would do; "
113 "don't write to disk\n"
114 "\t\t--verbose : e.g. show mkfs progress\n"
116 (int)sizeof(((struct lustre_disk_data *)0)->ldd_userdata));
120 #define vprint if (verbose > 0) printf
121 #define verrprint if (verbose >= 0) printf
123 static void fatal(void)
126 fprintf(stderr, "\n%s FATAL: ", progname);
129 /*================ utility functions =====================*/
131 char *strscat(char *dst, char *src, int buflen) {
133 if (strlen(dst) + strlen(src) >= buflen) {
134 fprintf(stderr, "string buffer overflow (max %d): '%s' + '%s'"
135 "\n", buflen, dst, src);
138 return strcat(dst, src);
142 char *strscpy(char *dst, char *src, int buflen) {
144 return strscat(dst, src, buflen);
148 dev_major (unsigned long long int __dev)
150 return ((__dev >> 8) & 0xfff) | ((unsigned int) (__dev >> 32) & ~0xfff);
154 dev_minor (unsigned long long int __dev)
156 return (__dev & 0xff) | ((unsigned int) (__dev >> 12) & ~0xff);
161 static int version = 0;
165 char release[4] = "";
167 fd = open("/proc/sys/kernel/osrelease", O_RDONLY);
169 fprintf(stderr, "%s: Warning: Can't resolve kernel "
170 "version, assuming 2.6\n", progname);
172 read(fd, release, 4);
175 if (strncmp(release, "2.4.", 4) == 0)
183 int run_command(char *cmd, int cmdsz)
185 char log[] = "/tmp/mkfs_logXXXXXX";
188 if ((cmdsz - strlen(cmd)) < 6) {
190 fprintf(stderr, "Command buffer overflow: %.*s...\n",
196 printf("cmd: %s\n", cmd);
198 if ((fd = mkstemp(log)) >= 0) {
204 strcat(cmd, " 2>&1");
206 /* Can't use popen because we need the rv of the command */
208 if (rc && (fd >= 0)) {
211 fp = fopen(log, "r");
213 while (fgets(buf, sizeof(buf), fp) != NULL) {
224 static int check_mtab_entry(char *spec)
229 fp = setmntent(MOUNTED, "r");
233 while ((mnt = getmntent(fp)) != NULL) {
234 if (strcmp(mnt->mnt_fsname, spec) == 0) {
236 fprintf(stderr, "%s: according to %s %s is "
237 "already mounted on %s\n",
238 progname, MOUNTED, spec, mnt->mnt_dir);
247 /*============ disk dev functions ===================*/
249 /* Setup a file in the first unused loop_device */
250 int loop_setup(struct mkfs_opts *mop)
256 /* Figure out the loop device names */
257 if (!access("/dev/loop0", F_OK | R_OK)) {
258 strcpy(loop_base, "/dev/loop\0");
259 } else if (!access("/dev/loop/0", F_OK | R_OK)) {
260 strcpy(loop_base, "/dev/loop/\0");
262 fprintf(stderr, "%s: can't access loop devices\n", progname);
266 /* Find unused loop device */
267 for (i = 0; i < MAX_LOOP_DEVICES; i++) {
269 int cmdsz = sizeof(cmd);
270 sprintf(l_device, "%s%d", loop_base, i);
271 if (access(l_device, F_OK | R_OK))
273 snprintf(cmd, cmdsz, "losetup %s > /dev/null 2>&1", l_device);
276 /* losetup gets 1 (ret=256) for non-set-up device */
278 /* Set up a loopback device to our file */
279 snprintf(cmd, cmdsz, "losetup %s %s", l_device,
281 ret = run_command(cmd, cmdsz);
283 fprintf(stderr, "%s: error %d on losetup: %s\n",
284 progname, ret, strerror(ret));
287 strscpy(mop->mo_loopdev, l_device,
288 sizeof(mop->mo_loopdev));
293 fprintf(stderr, "%s: out of loop devices!\n", progname);
297 int loop_cleanup(struct mkfs_opts *mop)
301 if ((mop->mo_flags & MO_IS_LOOP) && *mop->mo_loopdev) {
302 sprintf(cmd, "losetup -d %s", mop->mo_loopdev);
303 ret = run_command(cmd, sizeof(cmd));
308 /* Determine if a device is a block device (as opposed to a file) */
309 int is_block(char* devname)
314 ret = access(devname, F_OK);
317 ret = stat(devname, &st);
319 fprintf(stderr, "%s: cannot stat %s\n", progname, devname);
322 return S_ISBLK(st.st_mode);
325 __u64 get_device_size(char* device)
330 fd = open(device, O_RDONLY);
332 fprintf(stderr, "%s: cannot open %s: %s\n",
333 progname, device, strerror(errno));
338 /* size in bytes. bz5831 */
339 ret = ioctl(fd, BLKGETSIZE64, (void*)&size);
344 ret = ioctl(fd, BLKGETSIZE, (void*)&lsize);
345 size = (__u64)lsize * 512;
350 fprintf(stderr, "%s: size ioctl failed: %s\n",
351 progname, strerror(errno));
355 vprint("device size = "LPU64"MB\n", size >> 20);
356 /* return value in KB */
360 int loop_format(struct mkfs_opts *mop)
364 if (mop->mo_device_sz == 0) {
366 fprintf(stderr, "loop device requires a --device-size= "
371 ret = creat(mop->mo_device, S_IRUSR|S_IWUSR);
374 fprintf(stderr, "%s: Unable to create backing store: %d\n",
380 ret = truncate(mop->mo_device, mop->mo_device_sz * 1024);
383 fprintf(stderr, "%s: Unable to truncate backing store: %d\n",
390 /* Check whether the file exists in the device */
391 static int file_in_dev(char *file_name, char *dev_name)
394 char debugfs_cmd[256];
395 unsigned int inode_num;
398 /* Construct debugfs command line. */
399 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
400 "debugfs -c -R 'stat %s' %s 2>&1 | egrep '(Inode|unsupported)'",
401 file_name, dev_name);
403 fp = popen(debugfs_cmd, "r");
405 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
409 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
413 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd), fp);
416 fprintf(stderr, "%s", debugfs_cmd);
417 if (strstr(debugfs_cmd, "unsupported feature")) {
418 fprintf(stderr, "In all likelihood, the "
419 "'unsupported feature' is 'extents', which "
420 "older debugfs does not understand.\n"
421 "Use e2fsprogs-1.38-cfs1 or later, available "
422 "from ftp://ftp.lustre.org/pub/lustre/other/"
431 /* Check whether the device has already been used with lustre */
432 static int is_lustre_target(struct mkfs_opts *mop)
436 vprint("checking for existing Lustre data: ");
438 if ((rc = file_in_dev(MOUNT_DATA_FILE, mop->mo_device))) {
440 (rc == 1) ? MOUNT_DATA_FILE : "extents");
441 /* in the -1 case, 'extents' means this really IS a lustre
446 if ((rc = file_in_dev(LAST_RCVD, mop->mo_device))) {
447 vprint("found %s\n", LAST_RCVD);
451 vprint("not found\n");
452 return 0; /* The device is not a lustre target. */
455 /* Build fs according to type */
456 int make_lustre_backfs(struct mkfs_opts *mop)
458 char mkfs_cmd[PATH_MAX];
464 if (mop->mo_device_sz != 0) {
465 if (mop->mo_device_sz < 8096){
466 fprintf(stderr, "%s: size of filesystem must be larger "
467 "than 8MB, but is set to %lldKB\n",
468 progname, (long long)mop->mo_device_sz);
471 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
474 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
475 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
476 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
477 __u64 device_sz = mop->mo_device_sz;
479 /* we really need the size */
480 if (device_sz == 0) {
481 device_sz = get_device_size(mop->mo_device);
486 /* Journal size in MB */
487 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
488 /* Choose our own default journal size */
489 long journal_sz = 0, max_sz;
490 if (device_sz > 1024 * 1024) /* 1GB */
491 journal_sz = (device_sz / 102400) * 4;
492 /* cap journal size at 1GB */
493 if (journal_sz > 1024L)
496 max_sz = (256000 * L_BLOCK_SIZE) >> 20; /* 1GB */
497 if (journal_sz > max_sz)
500 sprintf(buf, " -J size=%ld", journal_sz);
501 strscat(mop->mo_mkfsopts, buf,
502 sizeof(mop->mo_mkfsopts));
506 /* Bytes_per_inode: disk size / num inodes */
507 if (strstr(mop->mo_mkfsopts, "-i") == NULL) {
508 long bytes_per_inode = 0;
510 if (IS_MDT(&mop->mo_ldd))
511 bytes_per_inode = 4096;
513 /* Allocate fewer inodes on large OST devices. Most
514 filesystems can be much more aggressive than even
516 if ((IS_OST(&mop->mo_ldd) && (device_sz > 100000000)))
517 bytes_per_inode = 16384; /* > 100 Gb device */
520 if (bytes_per_inode > 0) {
521 sprintf(buf, " -i %ld", bytes_per_inode);
522 strscat(mop->mo_mkfsopts, buf,
523 sizeof(mop->mo_mkfsopts));
527 /* Inode size (for extended attributes). The LOV EA size is
528 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
529 * and we want some margin above that for ACLs, other EAs... */
530 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
532 if (IS_MDT(&mop->mo_ldd)) {
533 if (mop->mo_stripe_count > 72)
534 inode_size = 512; /* bz 7241 */
535 /* cray stripes across all osts (>60) */
536 else if (mop->mo_stripe_count > 32)
538 else if (mop->mo_stripe_count > 10)
542 } else if (IS_OST(&mop->mo_ldd)) {
543 /* now as we store fids in EA on OST we need
544 to make inode bigger */
548 if (inode_size > 0) {
549 sprintf(buf, " -I %ld", inode_size);
550 strscat(mop->mo_mkfsopts, buf,
551 sizeof(mop->mo_mkfsopts));
556 strscat(mop->mo_mkfsopts, " -q",
557 sizeof(mop->mo_mkfsopts));
560 if (strstr(mop->mo_mkfsopts, "-O") == NULL) {
561 /* Enable hashed b-tree directory lookup in large dirs
563 strscat(mop->mo_mkfsopts, " -O dir_index",
564 sizeof(mop->mo_mkfsopts));
565 /* ldiskfs2: do not initialize all groups. */
566 if (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)
567 strscat(mop->mo_mkfsopts, ",uninit_groups",
568 sizeof(mop->mo_mkfsopts));
571 /* Allow reformat of full devices (as opposed to
572 partitions.) We already checked for mounted dev. */
573 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
575 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
576 "mkfs.ext2 -j -b %d -L %s ", L_BLOCK_SIZE,
577 mop->mo_ldd.ldd_svname);
578 } else if (mop->mo_ldd.ldd_mount_type == LDD_MT_REISERFS) {
579 long journal_sz = 0; /* FIXME default journal size */
580 if (journal_sz > 0) {
581 sprintf(buf, " --journal_size %ld", journal_sz);
582 strscat(mop->mo_mkfsopts, buf,
583 sizeof(mop->mo_mkfsopts));
585 snprintf(mkfs_cmd, sizeof(mkfs_cmd), "mkreiserfs -ff ");
587 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
588 progname, mop->mo_ldd.ldd_mount_type,
589 MT_STR(&mop->mo_ldd));
593 /* For loop device format the dev, not the filename */
594 dev = mop->mo_device;
595 if (mop->mo_flags & MO_IS_LOOP)
596 dev = mop->mo_loopdev;
598 vprint("formatting backing filesystem %s on %s\n",
599 MT_STR(&mop->mo_ldd), dev);
600 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
601 vprint("\t4k blocks %d\n", block_count);
602 vprint("\toptions %s\n", mop->mo_mkfsopts);
604 /* mkfs_cmd's trailing space is important! */
605 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
606 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
607 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
608 if (block_count != 0) {
609 sprintf(buf, " %d", block_count);
610 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
613 vprint("mkfs_cmd = %s\n", mkfs_cmd);
614 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
617 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
622 /* ==================== Lustre config functions =============*/
624 void print_ldd(char *str, struct lustre_disk_data *ldd)
626 printf("\n %s:\n", str);
627 printf("Target: %s\n", ldd->ldd_svname);
628 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
629 printf("Index: unassigned\n");
631 printf("Index: %d\n", ldd->ldd_svindex);
632 if (ldd->ldd_uuid[0])
633 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
634 printf("Lustre FS: %s\n", ldd->ldd_fsname);
635 printf("Mount type: %s\n", MT_STR(ldd));
636 printf("Flags: %#x\n", ldd->ldd_flags);
637 printf(" (%s%s%s%s%s%s%s%s)\n",
638 IS_MDT(ldd) ? "MDT ":"",
639 IS_OST(ldd) ? "OST ":"",
640 IS_MGS(ldd) ? "MGS ":"",
641 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
642 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
643 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
644 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
645 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
646 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
647 printf("Parameters:%s\n", ldd->ldd_params);
648 if (ldd->ldd_userdata[0])
649 printf("Comment: %s\n", ldd->ldd_userdata);
653 /* Write the server config files */
654 int write_local_files(struct mkfs_opts *mop)
656 char mntpt[] = "/tmp/mntXXXXXX";
662 /* Mount this device temporarily in order to write these files */
663 if (!mkdtemp(mntpt)) {
664 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
665 progname, mntpt, strerror(errno));
669 dev = mop->mo_device;
670 if (mop->mo_flags & MO_IS_LOOP)
671 dev = mop->mo_loopdev;
673 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0, NULL);
675 fprintf(stderr, "%s: Unable to mount %s: %s\n",
676 progname, dev, strerror(errno));
678 if (errno == ENODEV) {
679 fprintf(stderr, "Is the %s module available?\n",
680 MT_STR(&mop->mo_ldd));
685 /* Set up initial directories */
686 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
687 ret = mkdir(filepnm, 0777);
688 if ((ret != 0) && (errno != EEXIST)) {
689 fprintf(stderr, "%s: Can't make configs dir %s (%s)\n",
690 progname, filepnm, strerror(errno));
692 } else if (errno == EEXIST) {
696 sprintf(filepnm, "%s/%s", mntpt, "ROOT");
697 ret = mkdir(filepnm, 0777);
698 if ((ret != 0) && (errno != EEXIST)) {
699 fprintf(stderr, "%s: Can't make ROOT dir %s (%s)\n",
700 progname, filepnm, strerror(errno));
702 } else if (errno == EEXIST) {
706 /* Save the persistent mount data into a file. Lustre must pre-read
707 this file to get the real mount options. */
708 vprint("Writing %s\n", MOUNT_DATA_FILE);
709 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
710 filep = fopen(filepnm, "w");
712 fprintf(stderr, "%s: Unable to create %s file: %s\n",
713 progname, filepnm, strerror(errno));
716 fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
721 /* Check for upgrade */
722 if ((mop->mo_ldd.ldd_flags & (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS))
723 == (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS)) {
726 int cmdsz = sizeof(cmd);
727 vprint("Copying old logs\n");
729 /* Copy the old client log to fsname-client */
730 sprintf(filepnm, "%s/%s/%s-client",
731 mntpt, MOUNT_CONFIGS_DIR, mop->mo_ldd.ldd_fsname);
732 snprintf(cmd, cmdsz, "cp %s/%s/client %s", mntpt, MDT_LOGS_DIR,
734 ret = run_command(cmd, cmdsz);
736 fprintf(stderr, "%s: Can't copy 1.4 config %s/client "
737 "(%d)\n", progname, MDT_LOGS_DIR, ret);
738 fprintf(stderr, "mount -t ldiskfs %s somewhere, "
739 "find the client log for fs %s and "
740 "copy it manually into %s/%s-client, "
743 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
744 mop->mo_ldd.ldd_fsname);
748 /* We need to use the old mdt log because otherwise mdt won't
749 have complete lov if old clients connect before all
751 /* Copy the old mdt log to fsname-MDT0000 (get old
752 name from mdt_UUID) */
754 strscpy(filepnm, (char *)mop->mo_ldd.ldd_uuid, sizeof(filepnm));
755 term = strstr(filepnm, "_UUID");
758 snprintf(cmd, cmdsz, "cp %s/%s/%s %s/%s/%s",
759 mntpt, MDT_LOGS_DIR, filepnm,
760 mntpt, MOUNT_CONFIGS_DIR,
761 mop->mo_ldd.ldd_svname);
762 ret = run_command(cmd, cmdsz);
765 fprintf(stderr, "%s: Can't copy 1.4 config %s/%s "
766 "(%d)\n", progname, MDT_LOGS_DIR, filepnm, ret);
767 fprintf(stderr, "mount -t ext3 %s somewhere, "
768 "find the MDT log for fs %s and "
769 "copy it manually into %s/%s, "
772 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
773 mop->mo_ldd.ldd_svname);
788 int read_local_files(struct mkfs_opts *mop)
790 char tmpdir[] = "/tmp/dirXXXXXX";
796 int cmdsz = sizeof(cmd);
798 /* Make a temporary directory to hold Lustre data files. */
799 if (!mkdtemp(tmpdir)) {
800 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
801 progname, tmpdir, strerror(errno));
805 dev = mop->mo_device;
807 /* Construct debugfs command line. */
808 snprintf(cmd, cmdsz, "debugfs -c -R 'dump /%s %s/mountdata' %s",
809 MOUNT_DATA_FILE, tmpdir, dev);
811 ret = run_command(cmd, cmdsz);
813 verrprint("%s: Unable to dump %s dir (%d)\n",
814 progname, MOUNT_CONFIGS_DIR, ret);
816 sprintf(filepnm, "%s/mountdata", tmpdir);
817 filep = fopen(filepnm, "r");
819 vprint("Reading %s\n", MOUNT_DATA_FILE);
820 fread(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
823 /* Try to read pre-1.6 config from last_rcvd */
824 struct lr_server_data lsd;
825 verrprint("%s: Unable to read %s (%s).\n",
826 progname, filepnm, strerror(errno));
828 verrprint("Trying last_rcvd\n");
829 sprintf(filepnm, "%s/%s", tmpdir, LAST_RCVD);
831 /* Construct debugfs command line. */
832 snprintf(cmd, cmdsz, "debugfs -c -R 'dump /%s %s' %s",
833 LAST_RCVD, filepnm, dev);
835 ret = run_command(cmd, cmdsz);
837 fprintf(stderr, "%s: Unable to dump %s file (%d)\n",
838 progname, LAST_RCVD, ret);
842 filep = fopen(filepnm, "r");
844 fprintf(stderr, "%s: Unable to open %s: %s\n",
845 progname, filepnm, strerror(errno));
847 verrprint("Contents of %s:\n", tmpdir);
849 snprintf(cmd, cmdsz, "ls -l %s/", tmpdir);
850 run_command(cmd, cmdsz);
851 verrprint("Contents of disk:\n");
852 snprintf(cmd, cmdsz, "debugfs -c -R 'ls -l /' %s", dev);
853 run_command(cmd, cmdsz);
857 vprint("Reading %s\n", LAST_RCVD);
858 ret = fread(&lsd, 1, sizeof(lsd), filep);
859 if (ret < sizeof(lsd)) {
860 fprintf(stderr, "%s: Short read (%d of %d)\n",
861 progname, ret, (int)sizeof(lsd));
866 vprint("Feature compat=%x, incompat=%x\n",
867 lsd.lsd_feature_compat, lsd.lsd_feature_incompat);
869 if ((lsd.lsd_feature_compat & OBD_COMPAT_OST) ||
870 (lsd.lsd_feature_incompat & OBD_INCOMPAT_OST)) {
871 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
872 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
873 } else if ((lsd.lsd_feature_compat & OBD_COMPAT_MDT) ||
874 (lsd.lsd_feature_incompat & OBD_INCOMPAT_MDT)) {
875 /* We must co-locate so mgs can see old logs.
876 If user doesn't want this, they can copy the old
877 logs manually and re-tunefs. */
878 mop->mo_ldd.ldd_flags =
879 LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_MGS;
880 mop->mo_ldd.ldd_svindex = lsd.lsd_mdt_index;
882 /* If neither is set, we're pre-1.4.6, make a guess. */
883 /* Construct debugfs command line. */
884 snprintf(cmd, cmdsz, "debugfs -c -R 'rdump /%s %s' %s",
885 MDT_LOGS_DIR, tmpdir, dev);
886 run_command(cmd, cmdsz);
888 sprintf(filepnm, "%s/%s", tmpdir, MDT_LOGS_DIR);
889 if (lsd.lsd_ost_index > 0) {
890 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
891 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
893 /* If there's a LOGS dir, it's an MDT */
894 if ((ret = access(filepnm, F_OK)) == 0) {
895 mop->mo_ldd.ldd_flags =
898 /* Old MDT's are always index 0
900 mop->mo_ldd.ldd_svindex = 0;
902 /* The index may not be correct */
903 mop->mo_ldd.ldd_flags =
904 LDD_F_SV_TYPE_OST | LDD_F_NEED_INDEX;
905 verrprint("OST with unknown index\n");
911 memcpy(mop->mo_ldd.ldd_uuid, lsd.lsd_uuid,
912 sizeof(mop->mo_ldd.ldd_uuid));
913 mop->mo_ldd.ldd_flags |= LDD_F_UPGRADE14;
920 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
921 run_command(cmd, cmdsz);
923 verrprint("Failed to read old data (%d)\n", ret);
928 void set_defaults(struct mkfs_opts *mop)
930 mop->mo_ldd.ldd_magic = LDD_MAGIC;
931 mop->mo_ldd.ldd_config_ver = 1;
932 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
933 mop->mo_mgs_failnodes = 0;
934 strcpy(mop->mo_ldd.ldd_fsname, "lustre");
935 if (get_os_version() == 24)
936 mop->mo_ldd.ldd_mount_type = LDD_MT_EXT3;
938 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
940 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
941 mop->mo_stripe_count = 1;
944 static inline void badopt(const char *opt, char *type)
946 fprintf(stderr, "%s: '--%s' only valid for %s\n",
947 progname, opt, type);
951 static int add_param(char *buf, char *key, char *val)
953 int end = sizeof(((struct lustre_disk_data *)0)->ldd_params);
954 int start = strlen(buf);
958 keylen = strlen(key);
959 if (start + 1 + keylen + strlen(val) >= end) {
960 fprintf(stderr, "%s: params are too long-\n%s %s%s\n",
961 progname, buf, key ? key : "", val);
965 sprintf(buf + start, " %s%s", key ? key : "", val);
969 /* from mount_lustre */
970 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
971 #define MAXNIDSTR 1024
972 static char *convert_hostnames(char *s1)
974 char *converted, *s2 = 0, *c;
975 int left = MAXNIDSTR;
978 converted = malloc(left);
979 if (converted == NULL) {
984 while ((left > 0) && ((s2 = strsep(&s1, ",: \0")))) {
985 nid = libcfs_str2nid(s2);
986 if (nid == LNET_NID_ANY) {
990 fprintf(stderr, "%s: Can't parse NID '%s'\n",
996 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
997 strlen("127.0.0.1")) == 0) {
998 fprintf(stderr, "%s: The NID '%s' resolves to the "
999 "loopback address '%s'. Lustre requires a "
1000 "non-loopback address.\n",
1001 progname, s2, libcfs_nid2str(nid));
1006 c += snprintf(c, left, "%s,", libcfs_nid2str(nid));
1007 left = converted + MAXNIDSTR - c;
1013 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
1016 static struct option long_opt[] = {
1017 {"backfstype", 1, 0, 'b'},
1018 {"stripe-count-hint", 1, 0, 'c'},
1019 {"comment", 1, 0, 'u'},
1020 {"configdev", 1, 0, 'C'},
1021 {"device-size", 1, 0, 'd'},
1022 {"dryrun", 0, 0, 'n'},
1023 {"erase-params", 0, 0, 'e'},
1024 {"failnode", 1, 0, 'f'},
1025 {"failover", 1, 0, 'f'},
1027 {"help", 0, 0, 'h'},
1028 {"index", 1, 0, 'i'},
1029 {"mkfsoptions", 1, 0, 'k'},
1030 {"mgsnode", 1, 0, 'm'},
1031 {"mgsnid", 1, 0, 'm'},
1033 {"fsname",1, 0, 'L'},
1034 {"noformat", 0, 0, 'n'},
1035 {"nomgs", 0, 0, 'N'},
1036 {"mountfsoptions", 1, 0, 'o'},
1038 {"param", 1, 0, 'p'},
1039 {"print", 0, 0, 'n'},
1040 {"quiet", 0, 0, 'q'},
1041 {"reformat", 0, 0, 'r'},
1042 {"verbose", 0, 0, 'v'},
1043 {"writeconf", 0, 0, 'w'},
1046 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqru:vw";
1050 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
1055 while (i < LDD_MT_LAST) {
1056 if (strcmp(optarg, mt_str(i)) == 0) {
1057 mop->mo_ldd.ldd_mount_type = i;
1065 if (IS_MDT(&mop->mo_ldd)) {
1066 int stripe_count = atol(optarg);
1067 if (stripe_count <= 0) {
1068 fprintf(stderr, "%s: bad stripe count "
1069 "%d\n", progname, stripe_count);
1072 mop->mo_stripe_count = stripe_count;
1074 badopt(long_opt[longidx].name, "MDT");
1078 case 'C': /* Configdev */
1080 printf("Configdev not implemented\n");
1083 mop->mo_device_sz = atol(optarg);
1086 mop->mo_ldd.ldd_params[0] = '\0';
1087 /* Must update the mgs logs */
1088 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1091 char *nids = convert_hostnames(optarg);
1094 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
1096 /* Combo needs to add MDT failnodes as MGS failnodes
1098 if (!rc && IS_MGS(&mop->mo_ldd)) {
1099 rc = add_param(mop->mo_ldd.ldd_params,
1100 PARAM_MGSNODE, nids);
1105 /* Must update the mgs logs */
1106 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1110 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
1116 if (!(mop->mo_ldd.ldd_flags &
1117 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1118 LDD_F_WRITECONF))) {
1119 fprintf(stderr, "%s: cannot change the index of"
1120 " a registered target\n", progname);
1123 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
1124 mop->mo_ldd.ldd_svindex = atol(optarg);
1125 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
1127 badopt(long_opt[longidx].name, "MDT,OST");
1132 strscpy(mop->mo_mkfsopts, optarg,
1133 sizeof(mop->mo_mkfsopts));
1137 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
1138 (!(mop->mo_ldd.ldd_flags &
1139 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1140 LDD_F_WRITECONF)))) {
1141 fprintf(stderr, "%s: cannot change the name of"
1142 " a registered target\n", progname);
1145 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
1146 fprintf(stderr, "%s: filesystem name must be "
1147 "1-8 chars\n", progname);
1150 if ((tmp = strpbrk(optarg, "/:"))) {
1151 fprintf(stderr, "%s: char '%c' not allowed in "
1152 "filesystem name\n", progname, *tmp);
1155 strscpy(mop->mo_ldd.ldd_fsname, optarg,
1156 sizeof(mop->mo_ldd.ldd_fsname));
1160 char *nids = convert_hostnames(optarg);
1163 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
1168 mop->mo_mgs_failnodes++;
1172 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
1178 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
1181 *mountopts = optarg;
1184 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
1187 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg);
1190 /* Must update the mgs logs */
1191 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1197 mop->mo_flags |= MO_FORCEFORMAT;
1200 strscpy(mop->mo_ldd.ldd_userdata, optarg,
1201 sizeof(mop->mo_ldd.ldd_userdata));
1207 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
1212 fprintf(stderr, "Unknown option '%c'\n", opt);
1218 /* Last arg is device */
1219 if (optind != argc - 1) {
1221 fprintf(stderr, "Bad argument: %s\n", argv[optind]);
1228 #include <lustre/libiam.h>
1230 #define LDISKFS_IOC_GETVERSION _IOR('f', 3, long)
1232 #ifndef TUNEFS /* mkfs.lustre */
1233 static int mkfs_iam_insert(int key_need_convert, char *keybuf,
1234 int rec_need_convert, char *recbuf, char *filename)
1238 struct iam_uapi_info ua;
1240 fd = iam_open(filename, &ua);
1242 fprintf(stderr, "failed to iam_open %s\n", filename);
1246 ret = iam_insert(fd, &ua,
1247 key_need_convert, keybuf,
1248 rec_need_convert, recbuf);
1251 fprintf(stderr, "failed to iam_insert %s\n", filename);
1258 static int touch_file(char *filename)
1262 if (filename == NULL) {
1266 fd = open(filename, O_CREAT | O_TRUNC, 0600);
1275 static int get_generation(char *filename, unsigned long *result)
1280 if (filename == NULL) {
1284 fd = open(filename, O_RDONLY);
1286 fprintf(stderr, "%s: failed to open %s\n",
1287 __FUNCTION__, filename);
1291 ret = ioctl(fd, LDISKFS_IOC_GETVERSION, result);
1294 return ((ret < 0) ? ret : 0);
1297 static int mkfs_mdt(struct mkfs_opts *mop)
1299 char mntpt[] = "/tmp/mntXXXXXX";
1300 char fstype[] = "ldiskfs";
1305 unsigned long generation;
1308 source = mop->mo_device;
1309 if (mop->mo_flags & MO_IS_LOOP) {
1310 source = mop->mo_loopdev;
1313 if ((source == NULL) || (*source == 0)) {
1317 if (!mkdtemp(mntpt)) {
1318 fprintf(stderr, "%s: failed to mkdtemp %s\n",
1319 __FUNCTION__, mntpt);
1323 ret = mount(source, mntpt, fstype, 0, NULL);
1328 snprintf(filepnm, sizeof(filepnm) - 1, "%s/%s", mntpt, "seq_ctl");
1329 ret = touch_file(filepnm);
1334 snprintf(filepnm, sizeof(filepnm) - 1, "%s/%s", mntpt, "seq_srv");
1335 ret = touch_file(filepnm);
1340 snprintf(filepnm, sizeof(filepnm) - 1, "%s/%s", mntpt, "last_received");
1341 ret = touch_file(filepnm);
1346 snprintf(filepnm, sizeof(filepnm) - 1, "%s/%s", mntpt, "lov_objid");
1347 ret = touch_file(filepnm);
1352 snprintf(filepnm, sizeof(filepnm) - 1, "%s/%s", mntpt, "root");
1353 ret = iam_creat(filepnm, FMT_LVAR, L_BLOCK_SIZE, 4, 17, 4);
1358 snprintf(filepnm, sizeof(filepnm) - 1, "%s/%s", mntpt, "fld");
1359 ret = iam_creat(filepnm, FMT_LFIX, L_BLOCK_SIZE, 8, 8, 4);
1364 snprintf(filepnm, sizeof(filepnm) - 1, "%s/%s", mntpt, "orphans");
1365 ret = iam_creat(filepnm, FMT_LFIX, L_BLOCK_SIZE, 20, 8, 4);
1370 snprintf(filepnm, sizeof(filepnm) - 1, "%s/%s", mntpt, "oi.16");
1371 ret = iam_creat(filepnm, FMT_LFIX, L_BLOCK_SIZE, 16, 8, 4);
1376 snprintf(filepnm, sizeof(filepnm) - 1, "%s/%s", mntpt, "oi.5");
1377 ret = iam_creat(filepnm, FMT_LFIX, L_BLOCK_SIZE, 5, 8, 4);
1382 snprintf(filepnm, sizeof(filepnm) - 1, "%s/%s", mntpt, CAPA_KEYS);
1383 ret = touch_file(filepnm);
1389 ret = mount(source, mntpt, fstype, 0, NULL);
1394 snprintf(filepnm, sizeof(filepnm) - 1, "%s/%s", mntpt, "root");
1395 ret = iam_polymorph(filepnm, 040755);
1397 perror("IAM_IOC_POLYMORPH");
1402 ret = mount(source, mntpt, fstype, 0, NULL);
1407 snprintf(filepnm, sizeof(filepnm) - 1, "%s/%s", mntpt, "fld");
1408 ret = mkfs_iam_insert(1, "0000000000000002", 1, "0000000000000000", filepnm);
1413 ret = mkfs_iam_insert(1, "0000000000000001", 1, "0000000000000000", filepnm);
1418 snprintf(filepnm, sizeof(filepnm) - 1, "%s/%s", mntpt, "root");
1419 ret = stat(filepnm, &st);
1424 ret = get_generation(filepnm, &generation);
1429 snprintf(recbuf, sizeof(recbuf) - 1, "110000000000000001%8.8x%8.8x",
1430 (unsigned int)st.st_ino, (unsigned int)generation);
1431 ret = mkfs_iam_insert(0, ".", 1, recbuf, filepnm);
1436 ret = mkfs_iam_insert(0, "..", 1, recbuf, filepnm);
1449 int main(int argc, char *const argv[])
1451 struct mkfs_opts mop;
1452 struct lustre_disk_data *ldd;
1453 char *mountopts = NULL;
1454 char always_mountopts[512] = "";
1455 char default_mountopts[512] = "";
1458 if ((progname = strrchr(argv[0], '/')) != NULL)
1463 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
1468 memset(&mop, 0, sizeof(mop));
1471 /* device is last arg */
1472 strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
1474 /* Are we using a loop device? */
1475 ret = is_block(mop.mo_device);
1479 mop.mo_flags |= MO_IS_LOOP;
1482 /* For tunefs, we must read in the old values before parsing any
1485 /* Check whether the disk has already been formatted by mkfs.lustre */
1486 ret = is_lustre_target(&mop);
1489 fprintf(stderr, "Device %s has not been formatted with "
1490 "mkfs.lustre\n", mop.mo_device);
1495 ret = read_local_files(&mop);
1498 fprintf(stderr, "Failed to read previous Lustre data from %s "
1499 "(%d)\n", mop.mo_device, ret);
1502 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
1503 mop.mo_mgs_failnodes++;
1506 print_ldd("Read previous values", &(mop.mo_ldd));
1509 ret = parse_opts(argc, argv, &mop, &mountopts);
1515 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
1517 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
1522 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
1524 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
1529 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
1530 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
1532 fprintf(stderr, "Can't find the target index, "
1533 "specify with --index\n");
1539 * Comment out these 2 checks temporarily, since for multi-MDSes
1540 * in single node only 1 mds node could have mgs service
1542 if (IS_MDT(ldd) && !IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1543 verrprint("No management node specified, adding MGS to this "
1545 ldd->ldd_flags |= LDD_F_SV_TYPE_MGS;
1547 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1550 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
1552 fprintf(stderr, "Must specify --mgsnode=\n");
1558 /* These are the permanent mount options (always included) */
1559 switch (ldd->ldd_mount_type) {
1561 case LDD_MT_LDISKFS:
1562 case LDD_MT_LDISKFS2: {
1563 sprintf(always_mountopts, "errors=remount-ro");
1564 if (IS_MDT(ldd) || IS_MGS(ldd))
1565 strscat(always_mountopts, ",iopen_nopriv,user_xattr",
1566 sizeof(always_mountopts));
1567 if ((get_os_version() == 24) && IS_OST(ldd))
1568 strscat(always_mountopts, ",asyncdel",
1569 sizeof(always_mountopts));
1570 /* NB: Files created while extents are enabled cannot be read
1571 if mounted with a kernel that doesn't include the CFS
1574 (ldd->ldd_mount_type == LDD_MT_LDISKFS ||
1575 ldd->ldd_mount_type == LDD_MT_LDISKFS2)) {
1576 strscat(default_mountopts, ",extents,mballoc",
1577 sizeof(default_mountopts));
1582 mop.mo_flags |= MO_IS_LOOP;
1583 sprintf(always_mountopts, "type=ext3,dev=%s",
1589 fprintf(stderr, "unknown fs type %d '%s'\n",
1590 ldd->ldd_mount_type,
1598 /* If user specifies mount opts, don't use defaults,
1599 but always use always_mountopts */
1600 sprintf(ldd->ldd_mount_opts, "%s,%s",
1601 always_mountopts, mountopts);
1604 if (ldd->ldd_mount_opts[0] == 0)
1605 /* use the defaults unless old opts exist */
1608 sprintf(ldd->ldd_mount_opts, "%s%s",
1609 always_mountopts, default_mountopts);
1613 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
1614 ldd->ldd_fsname, ldd->ldd_svname);
1617 print_ldd("Permanent disk data", ldd);
1620 printf("exiting before disk write.\n");
1624 if (check_mtab_entry(mop.mo_device))
1627 /* Create the loopback file */
1628 if (mop.mo_flags & MO_IS_LOOP) {
1629 ret = access(mop.mo_device, F_OK);
1632 #ifndef TUNEFS /* mkfs.lustre */
1633 /* Reformat the loopback file */
1634 if (ret || (mop.mo_flags & MO_FORCEFORMAT))
1635 ret = loop_format(&mop);
1638 ret = loop_setup(&mop);
1641 fprintf(stderr, "Loop device setup for %s failed: %s\n",
1642 mop.mo_device, strerror(ret));
1647 #ifndef TUNEFS /* mkfs.lustre */
1648 /* Check whether the disk has already been formatted by mkfs.lustre */
1649 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
1650 ret = is_lustre_target(&mop);
1653 fprintf(stderr, "Device %s was previously formatted "
1654 "for lustre. Use --reformat to reformat it, "
1655 "or tunefs.lustre to modify.\n",
1661 /* Format the backing filesystem */
1662 ret = make_lustre_backfs(&mop);
1665 fprintf(stderr, "mkfs failed %d\n", ret);
1670 /* Write our config files */
1671 ret = write_local_files(&mop);
1674 fprintf(stderr, "failed to write local files\n");
1678 #ifndef TUNEFS /* mkfs.lustre */
1680 ret = mkfs_mdt(&mop);
1682 fprintf(stderr, "failed to mkfs_mdt\n");
1691 /* Fix any crazy return values from system() */
1692 if (ret && ((ret & 255) == 0))
1695 verrprint("%s: exiting with %d (%s)\n",
1696 progname, ret, strerror(ret));