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 =====================*/
132 dev_major (unsigned long long int __dev)
134 return ((__dev >> 8) & 0xfff) | ((unsigned int) (__dev >> 32) & ~0xfff);
138 dev_minor (unsigned long long int __dev)
140 return (__dev & 0xff) | ((unsigned int) (__dev >> 12) & ~0xff);
145 static int version = 0;
149 char release[4] = "";
151 fd = open("/proc/sys/kernel/osrelease", O_RDONLY);
153 fprintf(stderr, "%s: Warning: Can't resolve kernel "
154 "version, assuming 2.6\n", progname);
156 read(fd, release, 4);
159 if (strncmp(release, "2.4.", 4) == 0)
167 int run_command(char *cmd, int cmdsz)
169 char log[] = "/tmp/mkfs_logXXXXXX";
172 if ((cmdsz - strlen(cmd)) < 6) {
174 fprintf(stderr, "Command buffer overflow: %.*s...\n",
180 printf("cmd: %s\n", cmd);
182 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 && (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 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
452 __u64 device_sz = mop->mo_device_sz;
454 /* we really need the size */
455 if (device_sz == 0) {
456 device_sz = get_device_size(mop->mo_device);
461 /* Journal size in MB */
462 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
463 /* Choose our own default journal size */
464 long journal_sz = 0, max_sz;
465 if (device_sz > 1024 * 1024) /* 1GB */
466 journal_sz = (device_sz / 102400) * 4;
468 max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
469 if (journal_sz > max_sz)
472 sprintf(buf, " -J size=%ld", journal_sz);
473 strcat(mop->mo_mkfsopts, buf);
477 /* bytes_per_inode: disk size / num inodes */
478 if (strstr(mop->mo_mkfsopts, "-i") == NULL) {
479 long bytes_per_inode = 0;
481 if (IS_MDT(&mop->mo_ldd))
482 bytes_per_inode = 4096;
484 /* Allocate fewer inodes on large OST devices. Most
485 filesystems can be much more aggressive than even
487 if ((IS_OST(&mop->mo_ldd) && (device_sz > 1000000)))
488 bytes_per_inode = 16384;
490 if (bytes_per_inode > 0) {
491 sprintf(buf, " -i %ld", bytes_per_inode);
492 strcat(mop->mo_mkfsopts, buf);
496 /* Inode size (for extended attributes) */
497 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
499 if (IS_MDT(&mop->mo_ldd)) {
500 if (mop->mo_stripe_count > 72)
501 inode_size = 512; /* bz 7241 */
502 /* cray stripes across all osts (>60) */
503 else if (mop->mo_stripe_count > 32)
505 else if (mop->mo_stripe_count > 10)
509 } else if (IS_OST(&mop->mo_ldd)) {
510 /* now as we store fids in EA on OST we need
511 to make inode bigger */
515 if (inode_size > 0) {
516 sprintf(buf, " -I %ld", inode_size);
517 strcat(mop->mo_mkfsopts, buf);
523 strcat(mop->mo_mkfsopts, " -q");
526 if (strstr(mop->mo_mkfsopts, "-O") == NULL) {
527 /* Enable hashed b-tree directory lookup in large dirs
529 strcat(mop->mo_mkfsopts, " -O dir_index");
531 /* ldiskfs2: do not initialize all groups. */
532 if (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2) {
533 strcat(mop->mo_mkfsopts, ",uninit_groups");
537 /* Allow reformat of full devices (as opposed to
538 partitions.) We already checked for mounted dev. */
539 strcat(mop->mo_mkfsopts, " -F");
541 left -= snprintf(mkfs_cmd, left,
542 "mkfs.ext2 -j -b %d -L %s ", L_BLOCK_SIZE,
543 mop->mo_ldd.ldd_svname);
545 } else if (mop->mo_ldd.ldd_mount_type == LDD_MT_REISERFS) {
546 long journal_sz = 0; /* FIXME default journal size */
547 if (journal_sz > 0) {
548 sprintf(buf, " --journal_size %ld", journal_sz);
549 strcat(mop->mo_mkfsopts, buf);
551 left -= snprintf(mkfs_cmd, left, "mkreiserfs -ff ");
554 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
555 progname, mop->mo_ldd.ldd_mount_type,
556 MT_STR(&mop->mo_ldd));
560 /* For loop device format the dev, not the filename */
561 dev = mop->mo_device;
562 if (mop->mo_flags & MO_IS_LOOP)
563 dev = mop->mo_loopdev;
565 vprint("formatting backing filesystem %s on %s\n",
566 MT_STR(&mop->mo_ldd), dev);
567 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
568 vprint("\t4k blocks %d\n", block_count);
569 vprint("\toptions %s\n", mop->mo_mkfsopts);
571 /* mkfs_cmd's trailing space is important! */
572 strncat(mkfs_cmd, mop->mo_mkfsopts, left);
573 left = sizeof(mkfs_cmd) - strlen(mkfs_cmd) - 1;
574 strncat(mkfs_cmd, " ", left);
575 left = sizeof(mkfs_cmd) - strlen(mkfs_cmd) - 1;
576 strncat(mkfs_cmd, dev, left);
577 left = sizeof(mkfs_cmd) - strlen(mkfs_cmd) - 1;
578 if (block_count != 0) {
579 sprintf(buf, " %d", block_count);
580 strncat(mkfs_cmd, buf, left);
581 left = sizeof(mkfs_cmd) - strlen(mkfs_cmd) - 1;
584 vprint("mkfs_cmd = %s\n", mkfs_cmd);
585 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
588 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
593 /* ==================== Lustre config functions =============*/
595 void print_ldd(char *str, struct lustre_disk_data *ldd)
597 printf("\n %s:\n", str);
598 printf("Target: %s\n", ldd->ldd_svname);
599 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
600 printf("Index: unassigned\n");
602 printf("Index: %d\n", ldd->ldd_svindex);
603 if (ldd->ldd_uuid[0])
604 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
605 printf("Lustre FS: %s\n", ldd->ldd_fsname);
606 printf("Mount type: %s\n", MT_STR(ldd));
607 printf("Flags: %#x\n", ldd->ldd_flags);
608 printf(" (%s%s%s%s%s%s%s%s)\n",
609 IS_MDT(ldd) ? "MDT ":"",
610 IS_OST(ldd) ? "OST ":"",
611 IS_MGS(ldd) ? "MGS ":"",
612 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
613 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
614 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
615 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
616 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
617 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
618 printf("Parameters:%s\n", ldd->ldd_params);
619 if (ldd->ldd_userdata[0])
620 printf("Comment: %s\n", ldd->ldd_userdata);
624 /* Write the server config files */
625 int write_local_files(struct mkfs_opts *mop)
627 char mntpt[] = "/tmp/mntXXXXXX";
633 /* Mount this device temporarily in order to write these files */
634 if (!mkdtemp(mntpt)) {
635 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
636 progname, mntpt, strerror(errno));
640 dev = mop->mo_device;
641 if (mop->mo_flags & MO_IS_LOOP)
642 dev = mop->mo_loopdev;
644 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0, NULL);
646 fprintf(stderr, "%s: Unable to mount %s: %s\n",
647 progname, dev, strerror(errno));
649 if (errno == ENODEV) {
650 fprintf(stderr, "Is the %s module available?\n",
651 MT_STR(&mop->mo_ldd));
656 /* Set up initial directories */
657 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
658 ret = mkdir(filepnm, 0777);
659 if ((ret != 0) && (errno != EEXIST)) {
660 fprintf(stderr, "%s: Can't make configs dir %s: %s\n",
661 progname, filepnm, strerror(errno));
663 } else if (errno == EEXIST) {
667 /* Save the persistent mount data into a file. Lustre must pre-read
668 this file to get the real mount options. */
669 vprint("Writing %s\n", MOUNT_DATA_FILE);
670 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
671 filep = fopen(filepnm, "w");
673 fprintf(stderr, "%s: Unable to create %s file: %s\n",
674 progname, filepnm, strerror(errno));
677 fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
682 /* Check for upgrade */
683 if ((mop->mo_ldd.ldd_flags & (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS))
684 == (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS)) {
687 int cmdsz = sizeof(cmd);
688 vprint("Copying old logs\n");
690 /* Copy the old client log to fsname-client */
691 sprintf(filepnm, "%s/%s/%s-client",
692 mntpt, MOUNT_CONFIGS_DIR, mop->mo_ldd.ldd_fsname);
693 snprintf(cmd, cmdsz, "cp %s/%s/client %s", mntpt, MDT_LOGS_DIR,
695 ret = run_command(cmd, cmdsz);
697 fprintf(stderr, "%s: Can't copy 1.4 config %s/client "
698 "(%d)\n", progname, MDT_LOGS_DIR, ret);
699 fprintf(stderr, "mount -t ldiskfs %s somewhere, "
700 "find the client log for fs %s and "
701 "copy it manually into %s/%s-client, "
704 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
705 mop->mo_ldd.ldd_fsname);
709 /* We need to use the old mdt log because otherwise mdt won't
710 have complete lov if old clients connect before all
712 /* Copy the old mdt log to fsname-MDT0000 (get old
713 name from mdt_UUID) */
715 strcpy(filepnm, mop->mo_ldd.ldd_uuid);
716 term = strstr(filepnm, "_UUID");
719 snprintf(cmd, cmdsz, "cp %s/%s/%s %s/%s/%s",
720 mntpt, MDT_LOGS_DIR, filepnm,
721 mntpt, MOUNT_CONFIGS_DIR,
722 mop->mo_ldd.ldd_svname);
723 ret = run_command(cmd, cmdsz);
726 fprintf(stderr, "%s: Can't copy 1.4 config %s/%s "
727 "(%d)\n", progname, MDT_LOGS_DIR, filepnm, ret);
728 fprintf(stderr, "mount -t ext3 %s somewhere, "
729 "find the MDT log for fs %s and "
730 "copy it manually into %s/%s, "
733 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
734 mop->mo_ldd.ldd_svname);
749 int read_local_files(struct mkfs_opts *mop)
751 char tmpdir[] = "/tmp/dirXXXXXX";
757 int cmdsz = sizeof(cmd);
759 /* Make a temporary directory to hold Lustre data files. */
760 if (!mkdtemp(tmpdir)) {
761 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
762 progname, tmpdir, strerror(errno));
766 dev = mop->mo_device;
768 /* Construct debugfs command line. */
769 snprintf(cmd, cmdsz, "debugfs -c -R 'dump /%s %s/mountdata' %s",
770 MOUNT_DATA_FILE, tmpdir, dev);
772 ret = run_command(cmd, cmdsz);
774 verrprint("%s: Unable to dump %s dir (%d)\n",
775 progname, MOUNT_CONFIGS_DIR, ret);
777 sprintf(filepnm, "%s/mountdata", tmpdir);
778 filep = fopen(filepnm, "r");
780 vprint("Reading %s\n", MOUNT_DATA_FILE);
781 fread(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
784 /* Try to read pre-1.6 config from last_rcvd */
785 struct lr_server_data lsd;
786 verrprint("%s: Unable to read %s (%s).\n",
787 progname, filepnm, strerror(errno));
789 verrprint("Trying last_rcvd\n");
790 sprintf(filepnm, "%s/%s", tmpdir, LAST_RCVD);
792 /* Construct debugfs command line. */
793 snprintf(cmd, cmdsz, "debugfs -c -R 'dump /%s %s' %s",
794 LAST_RCVD, filepnm, dev);
796 ret = run_command(cmd, cmdsz);
798 fprintf(stderr, "%s: Unable to dump %s file (%d)\n",
799 progname, LAST_RCVD, ret);
803 filep = fopen(filepnm, "r");
805 fprintf(stderr, "%s: Unable to open %s: %s\n",
806 progname, filepnm, strerror(errno));
808 verrprint("Contents of %s:\n", tmpdir);
810 snprintf(cmd, cmdsz, "ls -l %s/", tmpdir);
811 run_command(cmd, cmdsz);
812 verrprint("Contents of disk:\n");
813 snprintf(cmd, cmdsz, "debugfs -c -R 'ls -l /' %s", dev);
814 run_command(cmd, cmdsz);
818 vprint("Reading %s\n", LAST_RCVD);
819 ret = fread(&lsd, 1, sizeof(lsd), filep);
820 if (ret < sizeof(lsd)) {
821 fprintf(stderr, "%s: Short read (%d of %d)\n",
822 progname, ret, (int)sizeof(lsd));
827 vprint("Feature compat=%x, incompat=%x\n",
828 lsd.lsd_feature_compat, lsd.lsd_feature_incompat);
830 if ((lsd.lsd_feature_compat & OBD_COMPAT_OST) ||
831 (lsd.lsd_feature_incompat & OBD_INCOMPAT_OST)) {
832 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
833 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
834 } else if ((lsd.lsd_feature_compat & OBD_COMPAT_MDT) ||
835 (lsd.lsd_feature_incompat & OBD_INCOMPAT_MDT)) {
836 /* We must co-locate so mgs can see old logs.
837 If user doesn't want this, they can copy the old
838 logs manually and re-tunefs. */
839 mop->mo_ldd.ldd_flags =
840 LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_MGS;
841 mop->mo_ldd.ldd_svindex = lsd.lsd_mdt_index;
843 /* If neither is set, we're pre-1.4.6, make a guess. */
844 /* Construct debugfs command line. */
845 snprintf(cmd, cmdsz, "debugfs -c -R 'rdump /%s %s' %s",
846 MDT_LOGS_DIR, tmpdir, dev);
847 run_command(cmd, cmdsz);
849 sprintf(filepnm, "%s/%s", tmpdir, MDT_LOGS_DIR);
850 if (lsd.lsd_ost_index > 0) {
851 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
852 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
854 /* If there's a LOGS dir, it's an MDT */
855 if ((ret = access(filepnm, F_OK)) == 0) {
856 mop->mo_ldd.ldd_flags =
859 /* Old MDT's are always index 0
861 mop->mo_ldd.ldd_svindex = 0;
863 /* The index may not be correct */
864 mop->mo_ldd.ldd_flags =
865 LDD_F_SV_TYPE_OST | LDD_F_NEED_INDEX;
866 verrprint("OST with unknown index\n");
872 memcpy(mop->mo_ldd.ldd_uuid, lsd.lsd_uuid,
873 sizeof(mop->mo_ldd.ldd_uuid));
874 mop->mo_ldd.ldd_flags |= LDD_F_UPGRADE14;
881 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
882 run_command(cmd, cmdsz);
884 verrprint("Failed to read old data (%d)\n", ret);
889 void set_defaults(struct mkfs_opts *mop)
891 mop->mo_ldd.ldd_magic = LDD_MAGIC;
892 mop->mo_ldd.ldd_config_ver = 1;
893 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
894 mop->mo_mgs_failnodes = 0;
895 strcpy(mop->mo_ldd.ldd_fsname, "lustre");
896 if (get_os_version() == 24)
897 mop->mo_ldd.ldd_mount_type = LDD_MT_EXT3;
899 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
901 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
902 mop->mo_stripe_count = 1;
905 static inline void badopt(const char *opt, char *type)
907 fprintf(stderr, "%s: '--%s' only valid for %s\n",
908 progname, opt, type);
912 static int add_param(char *buf, char *key, char *val)
914 int end = sizeof(((struct lustre_disk_data *)0)->ldd_params);
915 int start = strlen(buf);
919 keylen = strlen(key);
920 if (start + 1 + keylen + strlen(val) >= end) {
921 fprintf(stderr, "%s: params are too long-\n%s %s%s\n",
922 progname, buf, key ? key : "", val);
926 sprintf(buf + start, " %s%s", key ? key : "", val);
930 /* from mount_lustre */
931 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
932 #define MAXNIDSTR 1024
933 static char *convert_hostnames(char *s1)
935 char *converted, *s2 = 0, *c;
936 int left = MAXNIDSTR;
939 converted = malloc(left);
941 while ((left > 0) && ((s2 = strsep(&s1, ",: \0")))) {
942 nid = libcfs_str2nid(s2);
943 if (nid == LNET_NID_ANY) {
947 fprintf(stderr, "%s: Can't parse NID '%s'\n",
953 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
954 strlen("127.0.0.1")) == 0) {
955 fprintf(stderr, "%s: The NID '%s' resolves to the "
956 "loopback address '%s'. Lustre requires a "
957 "non-loopback address.\n",
958 progname, s2, libcfs_nid2str(nid));
963 c += snprintf(c, left, "%s,", libcfs_nid2str(nid));
964 left = converted + MAXNIDSTR - c;
970 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
973 static struct option long_opt[] = {
974 {"backfstype", 1, 0, 'b'},
975 {"stripe-count-hint", 1, 0, 'c'},
976 {"comment", 1, 0, 'u'},
977 {"configdev", 1, 0, 'C'},
978 {"device-size", 1, 0, 'd'},
979 {"dryrun", 0, 0, 'n'},
980 {"erase-params", 0, 0, 'e'},
981 {"failnode", 1, 0, 'f'},
982 {"failover", 1, 0, 'f'},
985 {"index", 1, 0, 'i'},
986 {"mkfsoptions", 1, 0, 'k'},
987 {"mgsnode", 1, 0, 'm'},
988 {"mgsnid", 1, 0, 'm'},
990 {"fsname",1, 0, 'L'},
991 {"noformat", 0, 0, 'n'},
992 {"nomgs", 0, 0, 'N'},
993 {"mountfsoptions", 1, 0, 'o'},
995 {"param", 1, 0, 'p'},
996 {"print", 0, 0, 'n'},
997 {"quiet", 0, 0, 'q'},
998 {"reformat", 0, 0, 'r'},
999 {"verbose", 0, 0, 'v'},
1000 {"writeconf", 0, 0, 'w'},
1003 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqru:vw";
1007 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
1012 while (i < LDD_MT_LAST) {
1013 if (strcmp(optarg, mt_str(i)) == 0) {
1014 mop->mo_ldd.ldd_mount_type = i;
1022 if (IS_MDT(&mop->mo_ldd)) {
1023 int stripe_count = atol(optarg);
1024 if (stripe_count <= 0) {
1025 fprintf(stderr, "%s: bad stripe count "
1026 "%d\n", progname, stripe_count);
1029 mop->mo_stripe_count = stripe_count;
1031 badopt(long_opt[longidx].name, "MDT");
1035 case 'C': /* Configdev */
1037 printf("Configdev not implemented\n");
1040 mop->mo_device_sz = atol(optarg);
1043 mop->mo_ldd.ldd_params[0] = '\0';
1044 /* Must update the mgs logs */
1045 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1048 char *nids = convert_hostnames(optarg);
1051 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
1053 /* Combo needs to add MDT failnodes as MGS failnodes
1055 if (!rc && IS_MGS(&mop->mo_ldd)) {
1056 rc = add_param(mop->mo_ldd.ldd_params,
1057 PARAM_MGSNODE, nids);
1062 /* Must update the mgs logs */
1063 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1067 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
1073 if (!(mop->mo_ldd.ldd_flags &
1074 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1075 LDD_F_WRITECONF))) {
1076 fprintf(stderr, "%s: cannot change the index of"
1077 " a registered target\n", progname);
1080 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
1081 mop->mo_ldd.ldd_svindex = atol(optarg);
1082 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
1084 badopt(long_opt[longidx].name, "MDT,OST");
1089 strncpy(mop->mo_mkfsopts, optarg,
1090 sizeof(mop->mo_mkfsopts) - 1);
1094 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
1095 (!(mop->mo_ldd.ldd_flags &
1096 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1097 LDD_F_WRITECONF)))) {
1098 fprintf(stderr, "%s: cannot change the name of"
1099 " a registered target\n", progname);
1102 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
1103 fprintf(stderr, "%s: filesystem name must be "
1104 "1-8 chars\n", progname);
1107 if ((tmp = strpbrk(optarg, "/:"))) {
1108 fprintf(stderr, "%s: char '%c' not allowed in "
1109 "filesystem name\n", progname, *tmp);
1112 strncpy(mop->mo_ldd.ldd_fsname, optarg,
1113 sizeof(mop->mo_ldd.ldd_fsname) - 1);
1117 char *nids = convert_hostnames(optarg);
1120 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
1125 mop->mo_mgs_failnodes++;
1129 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
1135 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
1138 *mountopts = optarg;
1141 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
1144 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg);
1147 /* Must update the mgs logs */
1148 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1154 mop->mo_flags |= MO_FORCEFORMAT;
1157 strncpy(mop->mo_ldd.ldd_userdata, optarg,
1158 sizeof(mop->mo_ldd.ldd_userdata));
1159 mop->mo_ldd.ldd_userdata[
1160 sizeof(mop->mo_ldd.ldd_userdata) - 1] = 0;
1166 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
1171 fprintf(stderr, "Unknown option '%c'\n", opt);
1177 /* Last arg is device */
1178 if (optind != argc - 1) {
1180 fprintf(stderr, "Bad argument: %s\n", argv[optind]);
1187 int main(int argc, char *const argv[])
1189 struct mkfs_opts mop;
1190 struct lustre_disk_data *ldd;
1191 char *mountopts = NULL;
1192 char always_mountopts[512] = "";
1193 char default_mountopts[512] = "";
1196 if ((progname = strrchr(argv[0], '/')) != NULL)
1201 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
1206 memset(&mop, 0, sizeof(mop));
1209 /* device is last arg */
1210 strcpy(mop.mo_device, argv[argc - 1]);
1212 /* Are we using a loop device? */
1213 ret = is_block(mop.mo_device);
1217 mop.mo_flags |= MO_IS_LOOP;
1220 /* For tunefs, we must read in the old values before parsing any
1223 /* Check whether the disk has already been formatted by mkfs.lustre */
1224 ret = is_lustre_target(&mop);
1227 fprintf(stderr, "Device %s has not been formatted with "
1228 "mkfs.lustre\n", mop.mo_device);
1233 ret = read_local_files(&mop);
1236 fprintf(stderr, "Failed to read previous Lustre data from %s "
1237 "(%d)\n", mop.mo_device, ret);
1240 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
1241 mop.mo_mgs_failnodes++;
1244 print_ldd("Read previous values", &(mop.mo_ldd));
1247 ret = parse_opts(argc, argv, &mop, &mountopts);
1253 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
1255 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
1260 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
1262 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
1267 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
1268 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
1270 fprintf(stderr, "Can't find the target index, "
1271 "specify with --index\n");
1276 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1279 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
1281 fprintf(stderr, "Must specify --mgsnode=\n");
1286 /* These are the permanent mount options (always included) */
1287 switch (ldd->ldd_mount_type) {
1289 case LDD_MT_LDISKFS:
1290 case LDD_MT_LDISKFS2: {
1291 sprintf(always_mountopts, "errors=remount-ro");
1292 if (IS_MDT(ldd) || IS_MGS(ldd))
1293 strcat(always_mountopts,
1294 ",iopen_nopriv,user_xattr");
1295 if ((get_os_version() == 24) && IS_OST(ldd))
1296 strcat(always_mountopts, ",asyncdel");
1297 /* NB: Files created while extents are enabled cannot be read
1298 if mounted with a kernel that doesn't include the CFS
1301 (ldd->ldd_mount_type == LDD_MT_LDISKFS ||
1302 ldd->ldd_mount_type == LDD_MT_LDISKFS2)) {
1303 strcat(default_mountopts, ",extents,mballoc");
1308 mop.mo_flags |= MO_IS_LOOP;
1309 sprintf(always_mountopts, "type=ext3,dev=%s",
1315 fprintf(stderr, "unknown fs type %d '%s'\n",
1316 ldd->ldd_mount_type,
1324 /* If user specifies mount opts, don't use defaults,
1325 but always use always_mountopts */
1326 sprintf(ldd->ldd_mount_opts, "%s,%s",
1327 always_mountopts, mountopts);
1330 if (ldd->ldd_mount_opts[0] == 0)
1331 /* use the defaults unless old opts exist */
1334 sprintf(ldd->ldd_mount_opts, "%s%s",
1335 always_mountopts, default_mountopts);
1339 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
1340 ldd->ldd_fsname, ldd->ldd_svname);
1343 print_ldd("Permanent disk data", ldd);
1346 printf("exiting before disk write.\n");
1350 if (check_mtab_entry(mop.mo_device))
1353 /* Create the loopback file */
1354 if (mop.mo_flags & MO_IS_LOOP) {
1355 ret = access(mop.mo_device, F_OK);
1358 #ifndef TUNEFS /* mkfs.lustre */
1359 /* Reformat the loopback file */
1360 if (ret || (mop.mo_flags & MO_FORCEFORMAT))
1361 ret = loop_format(&mop);
1364 ret = loop_setup(&mop);
1367 fprintf(stderr, "Loop device setup for %s failed: %s\n",
1368 mop.mo_device, strerror(ret));
1373 #ifndef TUNEFS /* mkfs.lustre */
1374 /* Check whether the disk has already been formatted by mkfs.lustre */
1375 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
1376 ret = is_lustre_target(&mop);
1379 fprintf(stderr, "Device %s was previously formatted "
1380 "for lustre. Use --reformat to reformat it, "
1381 "or tunefs.lustre to modify.\n",
1387 /* Format the backing filesystem */
1388 ret = make_lustre_backfs(&mop);
1391 fprintf(stderr, "mkfs failed %d\n", ret);
1396 /* Write our config files */
1397 ret = write_local_files(&mop);
1400 fprintf(stderr, "failed to write local files\n");
1407 /* Fix any crazy return values from system() */
1408 if (ret && ((ret & 255) == 0))
1411 verrprint("%s: exiting with %d (%s)\n",
1412 progname, ret, strerror(ret));