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_mount_type_string[20]; /* "ext3", "ldiskfs", ... */
63 char mo_device[128]; /* disk device name */
64 char mo_mkfsopts[128]; /* options to the backing-store mkfs */
65 char mo_loopdev[128]; /* in case a loop dev is needed */
66 __u64 mo_device_sz; /* in KB */
72 static char *progname;
73 static int verbose = 1;
74 static int print_only = 0;
79 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
80 fprintf(out, "usage: %s <target types> [options] <device>\n", progname);
82 "\t<device>:block device or file (e.g /dev/sda or /tmp/ost1)\n"
84 "\t\t--ost: object storage, mutually exclusive with mdt,mgs\n"
85 "\t\t--mdt: metadata storage, mutually exclusive with ost\n"
86 "\t\t--mgs: configuration management service - one per site\n"
87 "\toptions (in order of popularity):\n"
88 "\t\t--mgsnode=<nid>[,<...>] : NID(s) of a remote mgs node\n"
89 "\t\t\trequired for all targets other than the mgs node\n"
90 "\t\t--fsname=<filesystem_name> : default is 'lustre'\n"
91 "\t\t--failnode=<nid>[,<...>] : NID(s) of a failover partner\n"
92 "\t\t--param <key>=<value> : set a permanent parameter\n"
93 "\t\t\te.g. --param sys.timeout=40\n"
94 "\t\t\t --param lov.stripesize=2M\n"
95 "\t\t--index=#N : target index (i.e. ost index within the lov)\n"
96 /* FIXME implement 1.6.x
97 "\t\t--configdev=<altdevice|file>: store configuration info\n"
98 "\t\t\tfor this device on an alternate device\n"
100 "\t\t--comment=<user comment>: arbitrary user string (%d bytes)\n"
101 "\t\t--mountfsoptions=<opts> : permanent mount options\n"
103 "\t\t--backfstype=<fstype> : backing fs type (ext3, ldiskfs)\n"
104 "\t\t--device-size=#N(KB) : device size for loop devices\n"
105 "\t\t--mkfsoptions=<opts> : format options\n"
106 "\t\t--reformat: overwrite an existing disk\n"
107 "\t\t--stripe-count-hint=#N : used for optimizing MDT inode size\n"
109 "\t\t--erase-params : erase all old parameter settings\n"
110 "\t\t--nomgs: turn off MGS service on this MDT\n"
111 "\t\t--writeconf: erase all config logs for this fs.\n"
113 "\t\t--noformat: just report what we would do; "
114 "don't write to disk\n"
117 (int)sizeof(((struct lustre_disk_data *)0)->ldd_userdata));
121 #define vprint if (verbose > 0) printf
122 #define verrprint if (verbose >= 0) printf
124 static void fatal(void)
127 fprintf(stderr, "\n%s FATAL: ", progname);
130 /*================ utility functions =====================*/
133 dev_major (unsigned long long int __dev)
135 return ((__dev >> 8) & 0xfff) | ((unsigned int) (__dev >> 32) & ~0xfff);
139 dev_minor (unsigned long long int __dev)
141 return (__dev & 0xff) | ((unsigned int) (__dev >> 12) & ~0xff);
146 static int version = 0;
150 char release[4] = "";
152 fd = open("/proc/sys/kernel/osrelease", O_RDONLY);
154 fprintf(stderr, "%s: Warning: Can't resolve kernel "
155 "version, assuming 2.6\n", progname);
157 read(fd, release, 4);
160 if (strncmp(release, "2.4.", 4) == 0)
168 int run_command(char *cmd, int cmdsz)
170 char log[] = "/tmp/mkfs_logXXXXXX";
173 if ((cmdsz - strlen(cmd)) < 6) {
175 fprintf(stderr, "Command buffer overflow: %.*s...\n",
181 printf("cmd: %s\n", cmd);
183 if ((fd = mkstemp(log)) >= 0) {
188 strcat(cmd, " 2>&1");
190 /* Can't use popen because we need the rv of the command */
192 if ((rc || (verbose > 2)) && (fd >= 0)) {
195 fp = fopen(log, "r");
197 while (fgets(buf, sizeof(buf), fp) != NULL) {
208 static int check_mtab_entry(char *spec)
213 fp = setmntent(MOUNTED, "r");
217 while ((mnt = getmntent(fp)) != NULL) {
218 if (strcmp(mnt->mnt_fsname, spec) == 0) {
220 fprintf(stderr, "%s: according to %s %s is "
221 "already mounted on %s\n",
222 progname, MOUNTED, spec, mnt->mnt_dir);
231 /*============ disk dev functions ===================*/
233 /* Setup a file in the first unused loop_device */
234 int loop_setup(struct mkfs_opts *mop)
240 /* Figure out the loop device names */
241 if (!access("/dev/loop0", F_OK | R_OK))
242 strcpy(loop_base, "/dev/loop\0");
243 else if (!access("/dev/loop/0", F_OK | R_OK))
244 strcpy(loop_base, "/dev/loop/\0");
246 fprintf(stderr, "%s: can't access loop devices\n", progname);
250 /* Find unused loop device */
251 for (i = 0; i < MAX_LOOP_DEVICES; i++) {
253 int cmdsz = sizeof(cmd);
254 sprintf(l_device, "%s%d", loop_base, i);
255 if (access(l_device, F_OK | R_OK))
257 snprintf(cmd, cmdsz, "losetup %s > /dev/null 2>&1", l_device);
260 /* losetup gets 1 (ret=256) for non-set-up device */
262 /* Set up a loopback device to our file */
263 snprintf(cmd, cmdsz, "losetup %s %s", l_device,
265 ret = run_command(cmd, cmdsz);
267 fprintf(stderr, "%s: error %d on losetup: %s\n",
268 progname, ret, strerror(ret));
271 strcpy(mop->mo_loopdev, l_device);
276 fprintf(stderr, "%s: out of loop devices!\n", progname);
280 int loop_cleanup(struct mkfs_opts *mop)
284 if ((mop->mo_flags & MO_IS_LOOP) && *mop->mo_loopdev) {
285 sprintf(cmd, "losetup -d %s", mop->mo_loopdev);
286 ret = run_command(cmd, sizeof(cmd));
291 /* Determine if a device is a block device (as opposed to a file) */
292 int is_block(char* devname)
297 ret = access(devname, F_OK);
300 ret = stat(devname, &st);
302 fprintf(stderr, "%s: cannot stat %s\n", progname, devname);
305 return S_ISBLK(st.st_mode);
308 __u64 get_device_size(char* device)
313 fd = open(device, O_RDONLY);
315 fprintf(stderr, "%s: cannot open %s: %s\n",
316 progname, device, strerror(errno));
321 /* size in bytes. bz5831 */
322 ret = ioctl(fd, BLKGETSIZE64, (void*)&size);
327 ret = ioctl(fd, BLKGETSIZE, (void*)&lsize);
328 size = (__u64)lsize * 512;
333 fprintf(stderr, "%s: size ioctl failed: %s\n",
334 progname, strerror(errno));
338 vprint("device size = "LPU64"MB\n", size >> 20);
339 /* return value in KB */
343 int loop_format(struct mkfs_opts *mop)
347 if (mop->mo_device_sz == 0) {
349 fprintf(stderr, "loop device requires a --device-size= "
354 ret = creat(mop->mo_device, S_IRUSR|S_IWUSR);
355 ret = truncate(mop->mo_device, mop->mo_device_sz * 1024);
358 fprintf(stderr, "%s: Unable to create backing store: %d\n",
365 /* Check whether the file exists in the device */
366 static int file_in_dev(char *file_name, char *dev_name)
369 char debugfs_cmd[256];
370 unsigned int inode_num;
373 /* Construct debugfs command line. */
374 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
375 "debugfs -c -R 'stat %s' %s 2>&1 | egrep '(Inode|unsupported)'",
376 file_name, dev_name);
378 fp = popen(debugfs_cmd, "r");
380 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
384 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
388 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd), fp);
391 fprintf(stderr, "%s", debugfs_cmd);
392 if (strstr(debugfs_cmd, "unsupported feature")) {
393 fprintf(stderr, "In all likelihood, the "
394 "'unsupported feature' is 'extents', which "
395 "older debugfs does not understand.\n"
396 "Use e2fsprogs-1.38-cfs1 or later, available "
397 "from ftp://ftp.lustre.org/pub/lustre/other/"
406 /* Check whether the device has already been used with lustre */
407 static int is_lustre_target(struct mkfs_opts *mop)
410 vprint("checking for existing Lustre data: ");
412 if ((rc = file_in_dev(MOUNT_DATA_FILE, mop->mo_device))) {
414 (rc == 1) ? MOUNT_DATA_FILE : "extents");
415 /* in the -1 case, 'extents' means this really IS a lustre
420 if ((rc = file_in_dev(LAST_RCVD, mop->mo_device))) {
421 vprint("found %s\n", LAST_RCVD);
425 vprint("not found\n");
426 return 0; /* The device is not a lustre target. */
429 /* Build fs according to type */
430 int make_lustre_backfs(struct mkfs_opts *mop)
432 char mkfs_cmd[PATH_MAX];
437 int left = sizeof(mkfs_cmd);
439 if (mop->mo_device_sz != 0) {
440 if (mop->mo_device_sz < 8096){
441 fprintf(stderr, "%s: size of filesystem must be larger "
442 "than 8MB, but is set to %lluKB\n",
443 progname, (long long)mop->mo_device_sz);
446 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
449 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
450 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS)) {
451 __u64 device_sz = mop->mo_device_sz;
453 /* we really need the size */
454 if (device_sz == 0) {
455 device_sz = get_device_size(mop->mo_device);
460 /* Journal size in MB */
461 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
462 /* Choose our own default journal size */
463 long journal_sz = 0, max_sz;
464 if (device_sz > 1024 * 1024) /* 1GB */
465 journal_sz = (device_sz / 102400) * 4;
467 max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
468 if (journal_sz > max_sz)
471 sprintf(buf, " -J size=%ld", journal_sz);
472 strcat(mop->mo_mkfsopts, buf);
476 /* Default bytes_per_inode is block size */
477 if (strstr(mop->mo_mkfsopts, "-i") == NULL) {
478 long bytes_per_inode = 0;
480 if (IS_MDT(&mop->mo_ldd))
481 bytes_per_inode = 4096;
483 /* Allocate fewer inodes on large OST devices. Most
484 filesystems can be much more aggressive than even
486 if ((IS_OST(&mop->mo_ldd) && (device_sz > 1000000)))
487 bytes_per_inode = 16384;
489 if (bytes_per_inode > 0) {
490 sprintf(buf, " -i %ld", bytes_per_inode);
491 strcat(mop->mo_mkfsopts, buf);
495 /* This is an undocumented mke2fs option. Default is 128. */
496 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
498 if (IS_MDT(&mop->mo_ldd)) {
499 if (mop->mo_stripe_count > 72)
500 inode_size = 512; /* bz 7241 */
501 /* cray stripes across all osts (>60) */
502 else if (mop->mo_stripe_count > 32)
504 else if (mop->mo_stripe_count > 10)
508 } else if (IS_OST(&mop->mo_ldd)) {
509 /* now as we store fids in EA on OST we need
510 to make inode bigger */
514 if (inode_size > 0) {
515 sprintf(buf, " -I %ld", inode_size);
516 strcat(mop->mo_mkfsopts, buf);
522 strcat(mop->mo_mkfsopts, " -q");
525 /* Enable hashed b-tree directory lookup in large dirs bz6224 */
526 if (strstr(mop->mo_mkfsopts, "-O") == NULL) {
527 strcat(mop->mo_mkfsopts, " -O dir_index");
530 /* Allow reformat of full devices (as opposed to
531 partitions.) We already checked for mounted dev. */
532 strcat(mop->mo_mkfsopts, " -F");
534 left -= snprintf(mkfs_cmd, left,
535 "mkfs.ext2 -j -b %d -L %s ", L_BLOCK_SIZE,
536 mop->mo_ldd.ldd_svname);
538 } else if (mop->mo_ldd.ldd_mount_type == LDD_MT_REISERFS) {
539 long journal_sz = 0; /* FIXME default journal size */
540 if (journal_sz > 0) {
541 sprintf(buf, " --journal_size %ld", journal_sz);
542 strcat(mop->mo_mkfsopts, buf);
544 left -= snprintf(mkfs_cmd, left, "mkreiserfs -ff ");
547 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
548 progname, mop->mo_ldd.ldd_mount_type,
549 MT_STR(&mop->mo_ldd));
553 /* For loop device format the dev, not the filename */
554 dev = mop->mo_device;
555 if (mop->mo_flags & MO_IS_LOOP)
556 dev = mop->mo_loopdev;
558 vprint("formatting backing filesystem %s on %s\n",
559 MT_STR(&mop->mo_ldd), dev);
560 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
561 vprint("\t4k blocks %d\n", block_count);
562 vprint("\toptions %s\n", mop->mo_mkfsopts);
564 /* mkfs_cmd's trailing space is important! */
565 strncat(mkfs_cmd, mop->mo_mkfsopts, left);
566 left = sizeof(mkfs_cmd) - strlen(mkfs_cmd) - 1;
567 strncat(mkfs_cmd, " ", left);
568 left = sizeof(mkfs_cmd) - strlen(mkfs_cmd) - 1;
569 strncat(mkfs_cmd, dev, left);
570 left = sizeof(mkfs_cmd) - strlen(mkfs_cmd) - 1;
571 if (block_count != 0) {
572 sprintf(buf, " %d", block_count);
573 strncat(mkfs_cmd, buf, left);
574 left = sizeof(mkfs_cmd) - strlen(mkfs_cmd) - 1;
577 vprint("mkfs_cmd = %s\n", mkfs_cmd);
578 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
581 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
586 /* ==================== Lustre config functions =============*/
588 void print_ldd(char *str, struct lustre_disk_data *ldd)
590 printf("\n %s:\n", str);
591 printf("Target: %s\n", ldd->ldd_svname);
592 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
593 printf("Index: unassigned\n");
595 printf("Index: %d\n", ldd->ldd_svindex);
596 if (ldd->ldd_uuid[0])
597 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
598 printf("Lustre FS: %s\n", ldd->ldd_fsname);
599 printf("Mount type: %s\n", MT_STR(ldd));
600 printf("Flags: %#x\n", ldd->ldd_flags);
601 printf(" (%s%s%s%s%s%s%s%s)\n",
602 IS_MDT(ldd) ? "MDT ":"",
603 IS_OST(ldd) ? "OST ":"",
604 IS_MGS(ldd) ? "MGS ":"",
605 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
606 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
607 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
608 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
609 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
610 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
611 printf("Parameters:%s\n", ldd->ldd_params);
612 if (ldd->ldd_userdata[0])
613 printf("Comment: %s\n", ldd->ldd_userdata);
617 /* Write the server config files */
618 int write_local_files(struct mkfs_opts *mop)
620 char mntpt[] = "/tmp/mntXXXXXX";
626 /* Mount this device temporarily in order to write these files */
627 if (!mkdtemp(mntpt)) {
628 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
629 progname, mntpt, strerror(errno));
633 dev = mop->mo_device;
634 if (mop->mo_flags & MO_IS_LOOP)
635 dev = mop->mo_loopdev;
637 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0, NULL);
639 fprintf(stderr, "%s: Unable to mount %s: %s\n",
640 progname, dev, strerror(errno));
642 if (errno == ENODEV) {
643 fprintf(stderr, "Is the %s module available?\n",
644 MT_STR(&mop->mo_ldd));
649 /* Set up initial directories */
650 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
651 ret = mkdir(filepnm, 0777);
652 if ((ret != 0) && (errno != EEXIST)) {
653 fprintf(stderr, "%s: Can't make configs dir %s: %s\n",
654 progname, filepnm, strerror(errno));
656 } else if (errno == EEXIST) {
660 /* Save the persistent mount data into a file. Lustre must pre-read
661 this file to get the real mount options. */
662 vprint("Writing %s\n", MOUNT_DATA_FILE);
663 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
664 filep = fopen(filepnm, "w");
666 fprintf(stderr, "%s: Unable to create %s file: %s\n",
667 progname, filepnm, strerror(errno));
670 fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
675 /* Check for upgrade */
676 if ((mop->mo_ldd.ldd_flags & (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS))
677 == (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS)) {
680 int cmdsz = sizeof(cmd);
681 vprint("Copying old logs\n");
683 /* Copy the old client log to fsname-client */
684 sprintf(filepnm, "%s/%s/%s-client",
685 mntpt, MOUNT_CONFIGS_DIR, mop->mo_ldd.ldd_fsname);
686 snprintf(cmd, cmdsz, "cp %s/%s/client %s", mntpt, MDT_LOGS_DIR,
688 ret = run_command(cmd, cmdsz);
690 fprintf(stderr, "%s: Can't copy 1.4 config %s/client "
691 "(%d)\n", progname, MDT_LOGS_DIR, ret);
692 fprintf(stderr, "mount -t ldiskfs %s somewhere, "
693 "find the client log for fs %s and "
694 "copy it manually into %s/%s-client, "
697 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
698 mop->mo_ldd.ldd_fsname);
702 /* We need to use the old mdt log because otherwise mdt won't
703 have complete lov if old clients connect before all
705 /* Copy the old mdt log to fsname-MDT0000 (get old
706 name from mdt_UUID) */
708 strcpy(filepnm, mop->mo_ldd.ldd_uuid);
709 term = strstr(filepnm, "_UUID");
712 snprintf(cmd, cmdsz, "cp %s/%s/%s %s/%s/%s",
713 mntpt, MDT_LOGS_DIR, filepnm,
714 mntpt, MOUNT_CONFIGS_DIR,
715 mop->mo_ldd.ldd_svname);
716 ret = run_command(cmd, cmdsz);
719 fprintf(stderr, "%s: Can't copy 1.4 config %s/%s "
720 "(%d)\n", progname, MDT_LOGS_DIR, filepnm, ret);
721 fprintf(stderr, "mount -t ext3 %s somewhere, "
722 "find the MDT log for fs %s and "
723 "copy it manually into %s/%s, "
726 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
727 mop->mo_ldd.ldd_svname);
742 int read_local_files(struct mkfs_opts *mop)
744 char tmpdir[] = "/tmp/dirXXXXXX";
750 int cmdsz = sizeof(cmd);
752 /* Make a temporary directory to hold Lustre data files. */
753 if (!mkdtemp(tmpdir)) {
754 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
755 progname, tmpdir, strerror(errno));
759 dev = mop->mo_device;
761 /* Construct debugfs command line. */
762 snprintf(cmd, cmdsz, "debugfs -c -R 'dump /%s %s/mountdata' %s",
763 MOUNT_DATA_FILE, tmpdir, dev);
765 ret = run_command(cmd, cmdsz);
767 verrprint("%s: Unable to dump %s dir (%d)\n",
768 progname, MOUNT_CONFIGS_DIR, ret);
770 sprintf(filepnm, "%s/mountdata", tmpdir);
771 filep = fopen(filepnm, "r");
773 vprint("Reading %s\n", MOUNT_DATA_FILE);
774 fread(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
777 /* Try to read pre-1.6 config from last_rcvd */
778 struct lr_server_data lsd;
779 verrprint("%s: Unable to read %s (%s).\n",
780 progname, filepnm, strerror(errno));
782 verrprint("Trying last_rcvd\n");
783 sprintf(filepnm, "%s/%s", tmpdir, LAST_RCVD);
785 /* Construct debugfs command line. */
786 snprintf(cmd, cmdsz, "debugfs -c -R 'dump /%s %s' %s",
787 LAST_RCVD, filepnm, dev);
789 ret = run_command(cmd, cmdsz);
791 fprintf(stderr, "%s: Unable to dump %s file (%d)\n",
792 progname, LAST_RCVD, ret);
796 filep = fopen(filepnm, "r");
798 fprintf(stderr, "%s: Unable to open %s: %s\n",
799 progname, filepnm, strerror(errno));
801 verrprint("Contents of %s:\n", tmpdir);
803 snprintf(cmd, cmdsz, "ls -l %s/", tmpdir);
804 run_command(cmd, cmdsz);
805 verrprint("Contents of disk:\n");
806 snprintf(cmd, cmdsz, "debugfs -c -R 'ls -l /' %s", dev);
807 run_command(cmd, cmdsz);
811 vprint("Reading %s\n", LAST_RCVD);
812 ret = fread(&lsd, 1, sizeof(lsd), filep);
813 if (ret < sizeof(lsd)) {
814 fprintf(stderr, "%s: Short read (%d of %d)\n",
815 progname, ret, (int)sizeof(lsd));
820 vprint("Feature compat=%x, incompat=%x\n",
821 lsd.lsd_feature_compat, lsd.lsd_feature_incompat);
823 if ((lsd.lsd_feature_compat & OBD_COMPAT_OST) ||
824 (lsd.lsd_feature_incompat & OBD_INCOMPAT_OST)) {
825 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
826 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
827 } else if ((lsd.lsd_feature_compat & OBD_COMPAT_MDT) ||
828 (lsd.lsd_feature_incompat & OBD_INCOMPAT_MDT)) {
829 /* We must co-locate so mgs can see old logs.
830 If user doesn't want this, they can copy the old
831 logs manually and re-tunefs. */
832 mop->mo_ldd.ldd_flags =
833 LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_MGS;
834 mop->mo_ldd.ldd_svindex = lsd.lsd_mdt_index;
836 /* If neither is set, we're pre-1.4.6, make a guess. */
837 /* Construct debugfs command line. */
838 snprintf(cmd, cmdsz, "debugfs -c -R 'rdump /%s %s' %s",
839 MDT_LOGS_DIR, tmpdir, dev);
840 run_command(cmd, cmdsz);
842 sprintf(filepnm, "%s/%s", tmpdir, MDT_LOGS_DIR);
843 if (lsd.lsd_ost_index > 0) {
844 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
845 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
847 /* If there's a LOGS dir, it's an MDT */
848 if ((ret = access(filepnm, F_OK)) == 0) {
849 mop->mo_ldd.ldd_flags =
852 /* Old MDT's are always index 0
854 mop->mo_ldd.ldd_svindex = 0;
856 /* The index may not be correct */
857 mop->mo_ldd.ldd_flags =
858 LDD_F_SV_TYPE_OST | LDD_F_NEED_INDEX;
859 verrprint("OST with unknown index\n");
865 memcpy(mop->mo_ldd.ldd_uuid, lsd.lsd_uuid,
866 sizeof(mop->mo_ldd.ldd_uuid));
867 mop->mo_ldd.ldd_flags |= LDD_F_UPGRADE14;
874 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
875 run_command(cmd, cmdsz);
877 verrprint("Failed to read old data (%d)\n", ret);
882 void set_defaults(struct mkfs_opts *mop)
884 mop->mo_ldd.ldd_magic = LDD_MAGIC;
885 mop->mo_ldd.ldd_config_ver = 1;
886 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
887 mop->mo_mgs_failnodes = 0;
888 strcpy(mop->mo_ldd.ldd_fsname, "lustre");
889 if (get_os_version() == 24)
890 mop->mo_ldd.ldd_mount_type = LDD_MT_EXT3;
892 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
894 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
895 mop->mo_stripe_count = 1;
898 static inline void badopt(const char *opt, char *type)
900 fprintf(stderr, "%s: '--%s' only valid for %s\n",
901 progname, opt, type);
905 static int add_param(char *buf, char *key, char *val)
907 int end = sizeof(((struct lustre_disk_data *)0)->ldd_params);
908 int start = strlen(buf);
912 keylen = strlen(key);
913 if (start + 1 + keylen + strlen(val) >= end) {
914 fprintf(stderr, "%s: params are too long-\n%s %s%s\n",
915 progname, buf, key ? key : "", val);
919 sprintf(buf + start, " %s%s", key ? key : "", val);
923 /* from mount_lustre */
924 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
925 #define MAXNIDSTR 1024
926 static char *convert_hostnames(char *s1)
928 char *converted, *s2 = 0, *c;
929 int left = MAXNIDSTR;
932 converted = malloc(left);
934 while ((left > 0) && ((s2 = strsep(&s1, ",: \0")))) {
935 nid = libcfs_str2nid(s2);
936 if (nid == LNET_NID_ANY) {
940 fprintf(stderr, "%s: Can't parse NID '%s'\n",
946 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
947 strlen("127.0.0.1")) == 0) {
948 fprintf(stderr, "%s: The NID '%s' resolves to the "
949 "loopback address '%s'. Lustre requires a "
950 "non-loopback address.\n",
951 progname, s2, libcfs_nid2str(nid));
956 c += snprintf(c, left, "%s,", libcfs_nid2str(nid));
957 left = converted + MAXNIDSTR - c;
963 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
966 static struct option long_opt[] = {
967 {"backfstype", 1, 0, 'b'},
968 {"stripe-count-hint", 1, 0, 'c'},
969 {"comment", 1, 0, 'u'},
970 {"configdev", 1, 0, 'C'},
971 {"device-size", 1, 0, 'd'},
972 {"erase-params", 0, 0, 'e'},
973 {"failnode", 1, 0, 'f'},
974 {"failover", 1, 0, 'f'},
977 {"index", 1, 0, 'i'},
978 {"mkfsoptions", 1, 0, 'k'},
979 {"mgsnode", 1, 0, 'm'},
980 {"mgsnid", 1, 0, 'm'},
982 {"fsname",1, 0, 'L'},
983 {"noformat", 0, 0, 'n'},
984 {"nomgs", 0, 0, 'N'},
985 {"mountfsoptions", 1, 0, 'o'},
987 {"param", 1, 0, 'p'},
988 {"print", 0, 0, 'n'},
989 {"quiet", 0, 0, 'q'},
990 {"reformat", 0, 0, 'r'},
991 {"verbose", 0, 0, 'v'},
992 {"writeconf", 0, 0, 'w'},
995 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqru:vw";
999 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
1004 while (i < LDD_MT_LAST) {
1005 if (strcmp(optarg, mt_str(i)) == 0) {
1006 mop->mo_ldd.ldd_mount_type = i;
1014 if (IS_MDT(&mop->mo_ldd)) {
1015 int stripe_count = atol(optarg);
1016 if (stripe_count <= 0) {
1017 fprintf(stderr, "%s: bad stripe count "
1018 "%d\n", progname, stripe_count);
1021 mop->mo_stripe_count = stripe_count;
1023 badopt(long_opt[longidx].name, "MDT");
1027 case 'C': /* Configdev */
1029 printf("Configdev not implemented\n");
1032 mop->mo_device_sz = atol(optarg);
1035 mop->mo_ldd.ldd_params[0] = '\0';
1036 /* Must update the mgs logs */
1037 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1040 char *nids = convert_hostnames(optarg);
1043 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
1045 /* Combo needs to add MDT failnodes as MGS failnodes
1047 if (!rc && IS_MGS(&mop->mo_ldd)) {
1048 rc = add_param(mop->mo_ldd.ldd_params,
1049 PARAM_MGSNODE, nids);
1054 /* Must update the mgs logs */
1055 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1059 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
1065 if (!(mop->mo_ldd.ldd_flags &
1066 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1067 LDD_F_WRITECONF))) {
1068 fprintf(stderr, "%s: cannot change the index of"
1069 " a registered target\n", progname);
1072 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
1073 mop->mo_ldd.ldd_svindex = atol(optarg);
1074 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
1076 badopt(long_opt[longidx].name, "MDT,OST");
1081 strncpy(mop->mo_mkfsopts, optarg,
1082 sizeof(mop->mo_mkfsopts) - 1);
1086 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
1087 (!(mop->mo_ldd.ldd_flags &
1088 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1089 LDD_F_WRITECONF)))) {
1090 fprintf(stderr, "%s: cannot change the name of"
1091 " a registered target\n", progname);
1094 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
1095 fprintf(stderr, "%s: filesystem name must be "
1096 "1-8 chars\n", progname);
1099 if ((tmp = strpbrk(optarg, "/:"))) {
1100 fprintf(stderr, "%s: char '%c' not allowed in "
1101 "filesystem name\n", progname, *tmp);
1104 strncpy(mop->mo_ldd.ldd_fsname, optarg,
1105 sizeof(mop->mo_ldd.ldd_fsname) - 1);
1109 char *nids = convert_hostnames(optarg);
1112 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
1117 mop->mo_mgs_failnodes++;
1121 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
1127 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
1130 *mountopts = optarg;
1133 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
1136 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg);
1139 /* Must update the mgs logs */
1140 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1146 mop->mo_flags |= MO_FORCEFORMAT;
1149 strncpy(mop->mo_ldd.ldd_userdata, optarg,
1150 sizeof(mop->mo_ldd.ldd_userdata));
1151 mop->mo_ldd.ldd_userdata[
1152 sizeof(mop->mo_ldd.ldd_userdata) - 1] = 0;
1158 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
1163 fprintf(stderr, "Unknown option '%c'\n", opt);
1169 if (optind >= argc) {
1171 fprintf(stderr, "Bad arguments\n");
1179 int main(int argc, char *const argv[])
1181 struct mkfs_opts mop;
1182 struct lustre_disk_data *ldd;
1183 char *mountopts = NULL;
1184 char always_mountopts[512] = "";
1185 char default_mountopts[512] = "";
1188 if ((progname = strrchr(argv[0], '/')) != NULL)
1193 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
1198 memset(&mop, 0, sizeof(mop));
1201 /* device is last arg */
1202 strcpy(mop.mo_device, argv[argc - 1]);
1204 /* Are we using a loop device? */
1205 ret = is_block(mop.mo_device);
1209 mop.mo_flags |= MO_IS_LOOP;
1212 /* For tunefs, we must read in the old values before parsing any
1215 /* Check whether the disk has already been formatted by mkfs.lustre */
1216 ret = is_lustre_target(&mop);
1219 fprintf(stderr, "Device %s has not been formatted with "
1220 "mkfs.lustre\n", mop.mo_device);
1225 ret = read_local_files(&mop);
1228 fprintf(stderr, "Failed to read previous Lustre data from %s "
1229 "(%d)\n", mop.mo_device, ret);
1232 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
1233 mop.mo_mgs_failnodes++;
1236 print_ldd("Read previous values", &(mop.mo_ldd));
1239 ret = parse_opts(argc, argv, &mop, &mountopts);
1245 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
1247 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
1252 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
1254 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
1259 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
1260 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
1262 fprintf(stderr, "Can't find the target index, "
1263 "specify with --index\n");
1268 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1271 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
1273 fprintf(stderr, "Must specify --mgsnode=\n");
1278 /* These are the permanent mount options (always included) */
1279 switch (ldd->ldd_mount_type) {
1281 case LDD_MT_LDISKFS: {
1282 sprintf(always_mountopts, "errors=remount-ro");
1283 if (IS_MDT(ldd) || IS_MGS(ldd))
1284 strcat(always_mountopts,
1285 ",iopen_nopriv,user_xattr");
1286 if ((get_os_version() == 24) && IS_OST(ldd))
1287 strcat(always_mountopts, ",asyncdel");
1288 /* NB: Files created while extents are enabled cannot be read
1289 if mounted with a kernel that doesn't include the CFS
1292 ldd->ldd_mount_type == LDD_MT_LDISKFS) {
1293 strcat(default_mountopts, ",extents,mballoc");
1298 mop.mo_flags |= MO_IS_LOOP;
1299 sprintf(always_mountopts, "type=ext3,dev=%s",
1305 fprintf(stderr, "unknown fs type %d '%s'\n",
1306 ldd->ldd_mount_type,
1314 /* If user specifies mount opts, don't use defaults,
1315 but always use always_mountopts */
1316 sprintf(ldd->ldd_mount_opts, "%s,%s",
1317 always_mountopts, mountopts);
1320 if (ldd->ldd_mount_opts[0] == 0)
1321 /* use the defaults unless old opts exist */
1324 sprintf(ldd->ldd_mount_opts, "%s%s",
1325 always_mountopts, default_mountopts);
1329 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
1330 ldd->ldd_fsname, ldd->ldd_svname);
1333 print_ldd("Permanent disk data", ldd);
1336 printf("exiting before disk write.\n");
1340 if (check_mtab_entry(mop.mo_device))
1343 /* Create the loopback file */
1344 if (mop.mo_flags & MO_IS_LOOP) {
1345 ret = access(mop.mo_device, F_OK);
1348 #ifndef TUNEFS /* mkfs.lustre */
1349 /* Reformat the loopback file */
1350 if (ret || (mop.mo_flags & MO_FORCEFORMAT))
1351 ret = loop_format(&mop);
1354 ret = loop_setup(&mop);
1357 fprintf(stderr, "Loop device setup for %s failed: %s\n",
1358 mop.mo_device, strerror(ret));
1363 #ifndef TUNEFS /* mkfs.lustre */
1364 /* Check whether the disk has already been formatted by mkfs.lustre */
1365 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
1366 ret = is_lustre_target(&mop);
1369 fprintf(stderr, "Device %s was previously formatted "
1370 "for lustre. Use --reformat to reformat it, "
1371 "or tunefs.lustre to modify.\n",
1377 /* Format the backing filesystem */
1378 ret = make_lustre_backfs(&mop);
1381 fprintf(stderr, "mkfs failed %d\n", ret);
1386 /* Write our config files */
1387 ret = write_local_files(&mop);
1390 fprintf(stderr, "failed to write local files\n");
1397 /* Fix any crazy return values from system() */
1398 if (ret && ((ret & 255) == 0))
1401 verrprint("%s: exiting with %d (%s)\n",
1402 progname, ret, strerror(ret));