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"
115 "\t\t--verbose : e.g. show mkfs progress\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) {
189 strcat(cmd, " 2>&1");
191 /* Can't use popen because we need the rv of the command */
193 if (rc && (fd >= 0)) {
196 fp = fopen(log, "r");
198 while (fgets(buf, sizeof(buf), fp) != NULL) {
209 static int check_mtab_entry(char *spec)
214 fp = setmntent(MOUNTED, "r");
218 while ((mnt = getmntent(fp)) != NULL) {
219 if (strcmp(mnt->mnt_fsname, spec) == 0) {
221 fprintf(stderr, "%s: according to %s %s is "
222 "already mounted on %s\n",
223 progname, MOUNTED, spec, mnt->mnt_dir);
232 /*============ disk dev functions ===================*/
234 /* Setup a file in the first unused loop_device */
235 int loop_setup(struct mkfs_opts *mop)
241 /* Figure out the loop device names */
242 if (!access("/dev/loop0", F_OK | R_OK))
243 strcpy(loop_base, "/dev/loop\0");
244 else if (!access("/dev/loop/0", F_OK | R_OK))
245 strcpy(loop_base, "/dev/loop/\0");
247 fprintf(stderr, "%s: can't access loop devices\n", progname);
251 /* Find unused loop device */
252 for (i = 0; i < MAX_LOOP_DEVICES; i++) {
254 int cmdsz = sizeof(cmd);
255 sprintf(l_device, "%s%d", loop_base, i);
256 if (access(l_device, F_OK | R_OK))
258 snprintf(cmd, cmdsz, "losetup %s > /dev/null 2>&1", l_device);
261 /* losetup gets 1 (ret=256) for non-set-up device */
263 /* Set up a loopback device to our file */
264 snprintf(cmd, cmdsz, "losetup %s %s", l_device,
266 ret = run_command(cmd, cmdsz);
268 fprintf(stderr, "%s: error %d on losetup: %s\n",
269 progname, ret, strerror(ret));
272 strcpy(mop->mo_loopdev, l_device);
277 fprintf(stderr, "%s: out of loop devices!\n", progname);
281 int loop_cleanup(struct mkfs_opts *mop)
285 if ((mop->mo_flags & MO_IS_LOOP) && *mop->mo_loopdev) {
286 sprintf(cmd, "losetup -d %s", mop->mo_loopdev);
287 ret = run_command(cmd, sizeof(cmd));
292 /* Determine if a device is a block device (as opposed to a file) */
293 int is_block(char* devname)
298 ret = access(devname, F_OK);
301 ret = stat(devname, &st);
303 fprintf(stderr, "%s: cannot stat %s\n", progname, devname);
306 return S_ISBLK(st.st_mode);
309 __u64 get_device_size(char* device)
314 fd = open(device, O_RDONLY);
316 fprintf(stderr, "%s: cannot open %s: %s\n",
317 progname, device, strerror(errno));
322 /* size in bytes. bz5831 */
323 ret = ioctl(fd, BLKGETSIZE64, (void*)&size);
328 ret = ioctl(fd, BLKGETSIZE, (void*)&lsize);
329 size = (__u64)lsize * 512;
334 fprintf(stderr, "%s: size ioctl failed: %s\n",
335 progname, strerror(errno));
339 vprint("device size = "LPU64"MB\n", size >> 20);
340 /* return value in KB */
344 int loop_format(struct mkfs_opts *mop)
348 if (mop->mo_device_sz == 0) {
350 fprintf(stderr, "loop device requires a --device-size= "
355 ret = creat(mop->mo_device, S_IRUSR|S_IWUSR);
356 ret = truncate(mop->mo_device, mop->mo_device_sz * 1024);
359 fprintf(stderr, "%s: Unable to create backing store: %d\n",
366 /* Check whether the file exists in the device */
367 static int file_in_dev(char *file_name, char *dev_name)
370 char debugfs_cmd[256];
371 unsigned int inode_num;
374 /* Construct debugfs command line. */
375 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
376 "debugfs -c -R 'stat %s' %s 2>&1 | egrep '(Inode|unsupported)'",
377 file_name, dev_name);
379 fp = popen(debugfs_cmd, "r");
381 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
385 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
389 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd), fp);
392 fprintf(stderr, "%s", debugfs_cmd);
393 if (strstr(debugfs_cmd, "unsupported feature")) {
394 fprintf(stderr, "In all likelihood, the "
395 "'unsupported feature' is 'extents', which "
396 "older debugfs does not understand.\n"
397 "Use e2fsprogs-1.38-cfs1 or later, available "
398 "from ftp://ftp.lustre.org/pub/lustre/other/"
407 /* Check whether the device has already been used with lustre */
408 static int is_lustre_target(struct mkfs_opts *mop)
411 vprint("checking for existing Lustre data: ");
413 if ((rc = file_in_dev(MOUNT_DATA_FILE, mop->mo_device))) {
415 (rc == 1) ? MOUNT_DATA_FILE : "extents");
416 /* in the -1 case, 'extents' means this really IS a lustre
421 if ((rc = file_in_dev(LAST_RCVD, mop->mo_device))) {
422 vprint("found %s\n", LAST_RCVD);
426 vprint("not found\n");
427 return 0; /* The device is not a lustre target. */
430 /* Build fs according to type */
431 int make_lustre_backfs(struct mkfs_opts *mop)
433 char mkfs_cmd[PATH_MAX];
438 int left = sizeof(mkfs_cmd);
440 if (mop->mo_device_sz != 0) {
441 if (mop->mo_device_sz < 8096){
442 fprintf(stderr, "%s: size of filesystem must be larger "
443 "than 8MB, but is set to %lluKB\n",
444 progname, (long long)mop->mo_device_sz);
447 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
450 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
451 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS)) {
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 /* Enable hashed b-tree directory lookup in large dirs bz6224 */
527 if (strstr(mop->mo_mkfsopts, "-O") == NULL) {
528 strcat(mop->mo_mkfsopts, " -O dir_index");
531 /* Allow reformat of full devices (as opposed to
532 partitions.) We already checked for mounted dev. */
533 strcat(mop->mo_mkfsopts, " -F");
535 left -= snprintf(mkfs_cmd, left,
536 "mkfs.ext2 -j -b %d -L %s ", L_BLOCK_SIZE,
537 mop->mo_ldd.ldd_svname);
539 } else if (mop->mo_ldd.ldd_mount_type == LDD_MT_REISERFS) {
540 long journal_sz = 0; /* FIXME default journal size */
541 if (journal_sz > 0) {
542 sprintf(buf, " --journal_size %ld", journal_sz);
543 strcat(mop->mo_mkfsopts, buf);
545 left -= snprintf(mkfs_cmd, left, "mkreiserfs -ff ");
548 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
549 progname, mop->mo_ldd.ldd_mount_type,
550 MT_STR(&mop->mo_ldd));
554 /* For loop device format the dev, not the filename */
555 dev = mop->mo_device;
556 if (mop->mo_flags & MO_IS_LOOP)
557 dev = mop->mo_loopdev;
559 vprint("formatting backing filesystem %s on %s\n",
560 MT_STR(&mop->mo_ldd), dev);
561 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
562 vprint("\t4k blocks %d\n", block_count);
563 vprint("\toptions %s\n", mop->mo_mkfsopts);
565 /* mkfs_cmd's trailing space is important! */
566 strncat(mkfs_cmd, mop->mo_mkfsopts, left);
567 left = sizeof(mkfs_cmd) - strlen(mkfs_cmd) - 1;
568 strncat(mkfs_cmd, " ", left);
569 left = sizeof(mkfs_cmd) - strlen(mkfs_cmd) - 1;
570 strncat(mkfs_cmd, dev, left);
571 left = sizeof(mkfs_cmd) - strlen(mkfs_cmd) - 1;
572 if (block_count != 0) {
573 sprintf(buf, " %d", block_count);
574 strncat(mkfs_cmd, buf, left);
575 left = sizeof(mkfs_cmd) - strlen(mkfs_cmd) - 1;
578 vprint("mkfs_cmd = %s\n", mkfs_cmd);
579 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
582 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
587 /* ==================== Lustre config functions =============*/
589 void print_ldd(char *str, struct lustre_disk_data *ldd)
591 printf("\n %s:\n", str);
592 printf("Target: %s\n", ldd->ldd_svname);
593 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
594 printf("Index: unassigned\n");
596 printf("Index: %d\n", ldd->ldd_svindex);
597 if (ldd->ldd_uuid[0])
598 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
599 printf("Lustre FS: %s\n", ldd->ldd_fsname);
600 printf("Mount type: %s\n", MT_STR(ldd));
601 printf("Flags: %#x\n", ldd->ldd_flags);
602 printf(" (%s%s%s%s%s%s%s%s)\n",
603 IS_MDT(ldd) ? "MDT ":"",
604 IS_OST(ldd) ? "OST ":"",
605 IS_MGS(ldd) ? "MGS ":"",
606 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
607 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
608 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
609 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
610 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
611 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
612 printf("Parameters:%s\n", ldd->ldd_params);
613 if (ldd->ldd_userdata[0])
614 printf("Comment: %s\n", ldd->ldd_userdata);
618 /* Write the server config files */
619 int write_local_files(struct mkfs_opts *mop)
621 char mntpt[] = "/tmp/mntXXXXXX";
627 /* Mount this device temporarily in order to write these files */
628 if (!mkdtemp(mntpt)) {
629 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
630 progname, mntpt, strerror(errno));
634 dev = mop->mo_device;
635 if (mop->mo_flags & MO_IS_LOOP)
636 dev = mop->mo_loopdev;
638 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0, NULL);
640 fprintf(stderr, "%s: Unable to mount %s: %s\n",
641 progname, dev, strerror(errno));
643 if (errno == ENODEV) {
644 fprintf(stderr, "Is the %s module available?\n",
645 MT_STR(&mop->mo_ldd));
650 /* Set up initial directories */
651 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
652 ret = mkdir(filepnm, 0777);
653 if ((ret != 0) && (errno != EEXIST)) {
654 fprintf(stderr, "%s: Can't make configs dir %s: %s\n",
655 progname, filepnm, strerror(errno));
657 } else if (errno == EEXIST) {
661 /* Save the persistent mount data into a file. Lustre must pre-read
662 this file to get the real mount options. */
663 vprint("Writing %s\n", MOUNT_DATA_FILE);
664 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
665 filep = fopen(filepnm, "w");
667 fprintf(stderr, "%s: Unable to create %s file: %s\n",
668 progname, filepnm, strerror(errno));
671 fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
676 /* Check for upgrade */
677 if ((mop->mo_ldd.ldd_flags & (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS))
678 == (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS)) {
681 int cmdsz = sizeof(cmd);
682 vprint("Copying old logs\n");
684 /* Copy the old client log to fsname-client */
685 sprintf(filepnm, "%s/%s/%s-client",
686 mntpt, MOUNT_CONFIGS_DIR, mop->mo_ldd.ldd_fsname);
687 snprintf(cmd, cmdsz, "cp %s/%s/client %s", mntpt, MDT_LOGS_DIR,
689 ret = run_command(cmd, cmdsz);
691 fprintf(stderr, "%s: Can't copy 1.4 config %s/client "
692 "(%d)\n", progname, MDT_LOGS_DIR, ret);
693 fprintf(stderr, "mount -t ldiskfs %s somewhere, "
694 "find the client log for fs %s and "
695 "copy it manually into %s/%s-client, "
698 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
699 mop->mo_ldd.ldd_fsname);
703 /* We need to use the old mdt log because otherwise mdt won't
704 have complete lov if old clients connect before all
706 /* Copy the old mdt log to fsname-MDT0000 (get old
707 name from mdt_UUID) */
709 strcpy(filepnm, mop->mo_ldd.ldd_uuid);
710 term = strstr(filepnm, "_UUID");
713 snprintf(cmd, cmdsz, "cp %s/%s/%s %s/%s/%s",
714 mntpt, MDT_LOGS_DIR, filepnm,
715 mntpt, MOUNT_CONFIGS_DIR,
716 mop->mo_ldd.ldd_svname);
717 ret = run_command(cmd, cmdsz);
720 fprintf(stderr, "%s: Can't copy 1.4 config %s/%s "
721 "(%d)\n", progname, MDT_LOGS_DIR, filepnm, ret);
722 fprintf(stderr, "mount -t ext3 %s somewhere, "
723 "find the MDT log for fs %s and "
724 "copy it manually into %s/%s, "
727 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
728 mop->mo_ldd.ldd_svname);
743 int read_local_files(struct mkfs_opts *mop)
745 char tmpdir[] = "/tmp/dirXXXXXX";
751 int cmdsz = sizeof(cmd);
753 /* Make a temporary directory to hold Lustre data files. */
754 if (!mkdtemp(tmpdir)) {
755 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
756 progname, tmpdir, strerror(errno));
760 dev = mop->mo_device;
762 /* Construct debugfs command line. */
763 snprintf(cmd, cmdsz, "debugfs -c -R 'dump /%s %s/mountdata' %s",
764 MOUNT_DATA_FILE, tmpdir, dev);
766 ret = run_command(cmd, cmdsz);
768 verrprint("%s: Unable to dump %s dir (%d)\n",
769 progname, MOUNT_CONFIGS_DIR, ret);
771 sprintf(filepnm, "%s/mountdata", tmpdir);
772 filep = fopen(filepnm, "r");
774 vprint("Reading %s\n", MOUNT_DATA_FILE);
775 fread(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
778 /* Try to read pre-1.6 config from last_rcvd */
779 struct lr_server_data lsd;
780 verrprint("%s: Unable to read %s (%s).\n",
781 progname, filepnm, strerror(errno));
783 verrprint("Trying last_rcvd\n");
784 sprintf(filepnm, "%s/%s", tmpdir, LAST_RCVD);
786 /* Construct debugfs command line. */
787 snprintf(cmd, cmdsz, "debugfs -c -R 'dump /%s %s' %s",
788 LAST_RCVD, filepnm, dev);
790 ret = run_command(cmd, cmdsz);
792 fprintf(stderr, "%s: Unable to dump %s file (%d)\n",
793 progname, LAST_RCVD, ret);
797 filep = fopen(filepnm, "r");
799 fprintf(stderr, "%s: Unable to open %s: %s\n",
800 progname, filepnm, strerror(errno));
802 verrprint("Contents of %s:\n", tmpdir);
804 snprintf(cmd, cmdsz, "ls -l %s/", tmpdir);
805 run_command(cmd, cmdsz);
806 verrprint("Contents of disk:\n");
807 snprintf(cmd, cmdsz, "debugfs -c -R 'ls -l /' %s", dev);
808 run_command(cmd, cmdsz);
812 vprint("Reading %s\n", LAST_RCVD);
813 ret = fread(&lsd, 1, sizeof(lsd), filep);
814 if (ret < sizeof(lsd)) {
815 fprintf(stderr, "%s: Short read (%d of %d)\n",
816 progname, ret, (int)sizeof(lsd));
821 vprint("Feature compat=%x, incompat=%x\n",
822 lsd.lsd_feature_compat, lsd.lsd_feature_incompat);
824 if ((lsd.lsd_feature_compat & OBD_COMPAT_OST) ||
825 (lsd.lsd_feature_incompat & OBD_INCOMPAT_OST)) {
826 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
827 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
828 } else if ((lsd.lsd_feature_compat & OBD_COMPAT_MDT) ||
829 (lsd.lsd_feature_incompat & OBD_INCOMPAT_MDT)) {
830 /* We must co-locate so mgs can see old logs.
831 If user doesn't want this, they can copy the old
832 logs manually and re-tunefs. */
833 mop->mo_ldd.ldd_flags =
834 LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_MGS;
835 mop->mo_ldd.ldd_svindex = lsd.lsd_mdt_index;
837 /* If neither is set, we're pre-1.4.6, make a guess. */
838 /* Construct debugfs command line. */
839 snprintf(cmd, cmdsz, "debugfs -c -R 'rdump /%s %s' %s",
840 MDT_LOGS_DIR, tmpdir, dev);
841 run_command(cmd, cmdsz);
843 sprintf(filepnm, "%s/%s", tmpdir, MDT_LOGS_DIR);
844 if (lsd.lsd_ost_index > 0) {
845 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
846 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
848 /* If there's a LOGS dir, it's an MDT */
849 if ((ret = access(filepnm, F_OK)) == 0) {
850 mop->mo_ldd.ldd_flags =
853 /* Old MDT's are always index 0
855 mop->mo_ldd.ldd_svindex = 0;
857 /* The index may not be correct */
858 mop->mo_ldd.ldd_flags =
859 LDD_F_SV_TYPE_OST | LDD_F_NEED_INDEX;
860 verrprint("OST with unknown index\n");
866 memcpy(mop->mo_ldd.ldd_uuid, lsd.lsd_uuid,
867 sizeof(mop->mo_ldd.ldd_uuid));
868 mop->mo_ldd.ldd_flags |= LDD_F_UPGRADE14;
875 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
876 run_command(cmd, cmdsz);
878 verrprint("Failed to read old data (%d)\n", ret);
883 void set_defaults(struct mkfs_opts *mop)
885 mop->mo_ldd.ldd_magic = LDD_MAGIC;
886 mop->mo_ldd.ldd_config_ver = 1;
887 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
888 mop->mo_mgs_failnodes = 0;
889 strcpy(mop->mo_ldd.ldd_fsname, "lustre");
890 if (get_os_version() == 24)
891 mop->mo_ldd.ldd_mount_type = LDD_MT_EXT3;
893 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
895 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
896 mop->mo_stripe_count = 1;
899 static inline void badopt(const char *opt, char *type)
901 fprintf(stderr, "%s: '--%s' only valid for %s\n",
902 progname, opt, type);
906 static int add_param(char *buf, char *key, char *val)
908 int end = sizeof(((struct lustre_disk_data *)0)->ldd_params);
909 int start = strlen(buf);
913 keylen = strlen(key);
914 if (start + 1 + keylen + strlen(val) >= end) {
915 fprintf(stderr, "%s: params are too long-\n%s %s%s\n",
916 progname, buf, key ? key : "", val);
920 sprintf(buf + start, " %s%s", key ? key : "", val);
924 /* from mount_lustre */
925 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
926 #define MAXNIDSTR 1024
927 static char *convert_hostnames(char *s1)
929 char *converted, *s2 = 0, *c;
930 int left = MAXNIDSTR;
933 converted = malloc(left);
935 while ((left > 0) && ((s2 = strsep(&s1, ",: \0")))) {
936 nid = libcfs_str2nid(s2);
937 if (nid == LNET_NID_ANY) {
941 fprintf(stderr, "%s: Can't parse NID '%s'\n",
947 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
948 strlen("127.0.0.1")) == 0) {
949 fprintf(stderr, "%s: The NID '%s' resolves to the "
950 "loopback address '%s'. Lustre requires a "
951 "non-loopback address.\n",
952 progname, s2, libcfs_nid2str(nid));
957 c += snprintf(c, left, "%s,", libcfs_nid2str(nid));
958 left = converted + MAXNIDSTR - c;
964 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
967 static struct option long_opt[] = {
968 {"backfstype", 1, 0, 'b'},
969 {"stripe-count-hint", 1, 0, 'c'},
970 {"comment", 1, 0, 'u'},
971 {"configdev", 1, 0, 'C'},
972 {"device-size", 1, 0, 'd'},
973 {"erase-params", 0, 0, 'e'},
974 {"failnode", 1, 0, 'f'},
975 {"failover", 1, 0, 'f'},
978 {"index", 1, 0, 'i'},
979 {"mkfsoptions", 1, 0, 'k'},
980 {"mgsnode", 1, 0, 'm'},
981 {"mgsnid", 1, 0, 'm'},
983 {"fsname",1, 0, 'L'},
984 {"noformat", 0, 0, 'n'},
985 {"nomgs", 0, 0, 'N'},
986 {"mountfsoptions", 1, 0, 'o'},
988 {"param", 1, 0, 'p'},
989 {"print", 0, 0, 'n'},
990 {"quiet", 0, 0, 'q'},
991 {"reformat", 0, 0, 'r'},
992 {"verbose", 0, 0, 'v'},
993 {"writeconf", 0, 0, 'w'},
996 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqru:vw";
1000 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
1005 while (i < LDD_MT_LAST) {
1006 if (strcmp(optarg, mt_str(i)) == 0) {
1007 mop->mo_ldd.ldd_mount_type = i;
1015 if (IS_MDT(&mop->mo_ldd)) {
1016 int stripe_count = atol(optarg);
1017 if (stripe_count <= 0) {
1018 fprintf(stderr, "%s: bad stripe count "
1019 "%d\n", progname, stripe_count);
1022 mop->mo_stripe_count = stripe_count;
1024 badopt(long_opt[longidx].name, "MDT");
1028 case 'C': /* Configdev */
1030 printf("Configdev not implemented\n");
1033 mop->mo_device_sz = atol(optarg);
1036 mop->mo_ldd.ldd_params[0] = '\0';
1037 /* Must update the mgs logs */
1038 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1041 char *nids = convert_hostnames(optarg);
1044 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
1046 /* Combo needs to add MDT failnodes as MGS failnodes
1048 if (!rc && IS_MGS(&mop->mo_ldd)) {
1049 rc = add_param(mop->mo_ldd.ldd_params,
1050 PARAM_MGSNODE, nids);
1055 /* Must update the mgs logs */
1056 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1060 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
1066 if (!(mop->mo_ldd.ldd_flags &
1067 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1068 LDD_F_WRITECONF))) {
1069 fprintf(stderr, "%s: cannot change the index of"
1070 " a registered target\n", progname);
1073 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
1074 mop->mo_ldd.ldd_svindex = atol(optarg);
1075 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
1077 badopt(long_opt[longidx].name, "MDT,OST");
1082 strncpy(mop->mo_mkfsopts, optarg,
1083 sizeof(mop->mo_mkfsopts) - 1);
1087 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
1088 (!(mop->mo_ldd.ldd_flags &
1089 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1090 LDD_F_WRITECONF)))) {
1091 fprintf(stderr, "%s: cannot change the name of"
1092 " a registered target\n", progname);
1095 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
1096 fprintf(stderr, "%s: filesystem name must be "
1097 "1-8 chars\n", progname);
1100 if ((tmp = strpbrk(optarg, "/:"))) {
1101 fprintf(stderr, "%s: char '%c' not allowed in "
1102 "filesystem name\n", progname, *tmp);
1105 strncpy(mop->mo_ldd.ldd_fsname, optarg,
1106 sizeof(mop->mo_ldd.ldd_fsname) - 1);
1110 char *nids = convert_hostnames(optarg);
1113 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
1118 mop->mo_mgs_failnodes++;
1122 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
1128 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
1131 *mountopts = optarg;
1134 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
1137 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg);
1140 /* Must update the mgs logs */
1141 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1147 mop->mo_flags |= MO_FORCEFORMAT;
1150 strncpy(mop->mo_ldd.ldd_userdata, optarg,
1151 sizeof(mop->mo_ldd.ldd_userdata));
1152 mop->mo_ldd.ldd_userdata[
1153 sizeof(mop->mo_ldd.ldd_userdata) - 1] = 0;
1159 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
1164 fprintf(stderr, "Unknown option '%c'\n", opt);
1170 if (optind >= argc) {
1172 fprintf(stderr, "Bad arguments\n");
1180 int main(int argc, char *const argv[])
1182 struct mkfs_opts mop;
1183 struct lustre_disk_data *ldd;
1184 char *mountopts = NULL;
1185 char always_mountopts[512] = "";
1186 char default_mountopts[512] = "";
1189 if ((progname = strrchr(argv[0], '/')) != NULL)
1194 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
1199 memset(&mop, 0, sizeof(mop));
1202 /* device is last arg */
1203 strcpy(mop.mo_device, argv[argc - 1]);
1205 /* Are we using a loop device? */
1206 ret = is_block(mop.mo_device);
1210 mop.mo_flags |= MO_IS_LOOP;
1213 /* For tunefs, we must read in the old values before parsing any
1216 /* Check whether the disk has already been formatted by mkfs.lustre */
1217 ret = is_lustre_target(&mop);
1220 fprintf(stderr, "Device %s has not been formatted with "
1221 "mkfs.lustre\n", mop.mo_device);
1226 ret = read_local_files(&mop);
1229 fprintf(stderr, "Failed to read previous Lustre data from %s "
1230 "(%d)\n", mop.mo_device, ret);
1233 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
1234 mop.mo_mgs_failnodes++;
1237 print_ldd("Read previous values", &(mop.mo_ldd));
1240 ret = parse_opts(argc, argv, &mop, &mountopts);
1246 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
1248 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
1253 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
1255 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
1260 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
1261 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
1263 fprintf(stderr, "Can't find the target index, "
1264 "specify with --index\n");
1269 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1272 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
1274 fprintf(stderr, "Must specify --mgsnode=\n");
1279 /* These are the permanent mount options (always included) */
1280 switch (ldd->ldd_mount_type) {
1282 case LDD_MT_LDISKFS: {
1283 sprintf(always_mountopts, "errors=remount-ro");
1284 if (IS_MDT(ldd) || IS_MGS(ldd))
1285 strcat(always_mountopts,
1286 ",iopen_nopriv,user_xattr");
1287 if ((get_os_version() == 24) && IS_OST(ldd))
1288 strcat(always_mountopts, ",asyncdel");
1289 /* NB: Files created while extents are enabled cannot be read
1290 if mounted with a kernel that doesn't include the CFS
1293 ldd->ldd_mount_type == LDD_MT_LDISKFS) {
1294 strcat(default_mountopts, ",extents,mballoc");
1299 mop.mo_flags |= MO_IS_LOOP;
1300 sprintf(always_mountopts, "type=ext3,dev=%s",
1306 fprintf(stderr, "unknown fs type %d '%s'\n",
1307 ldd->ldd_mount_type,
1315 /* If user specifies mount opts, don't use defaults,
1316 but always use always_mountopts */
1317 sprintf(ldd->ldd_mount_opts, "%s,%s",
1318 always_mountopts, mountopts);
1321 if (ldd->ldd_mount_opts[0] == 0)
1322 /* use the defaults unless old opts exist */
1325 sprintf(ldd->ldd_mount_opts, "%s%s",
1326 always_mountopts, default_mountopts);
1330 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
1331 ldd->ldd_fsname, ldd->ldd_svname);
1334 print_ldd("Permanent disk data", ldd);
1337 printf("exiting before disk write.\n");
1341 if (check_mtab_entry(mop.mo_device))
1344 /* Create the loopback file */
1345 if (mop.mo_flags & MO_IS_LOOP) {
1346 ret = access(mop.mo_device, F_OK);
1349 #ifndef TUNEFS /* mkfs.lustre */
1350 /* Reformat the loopback file */
1351 if (ret || (mop.mo_flags & MO_FORCEFORMAT))
1352 ret = loop_format(&mop);
1355 ret = loop_setup(&mop);
1358 fprintf(stderr, "Loop device setup for %s failed: %s\n",
1359 mop.mo_device, strerror(ret));
1364 #ifndef TUNEFS /* mkfs.lustre */
1365 /* Check whether the disk has already been formatted by mkfs.lustre */
1366 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
1367 ret = is_lustre_target(&mop);
1370 fprintf(stderr, "Device %s was previously formatted "
1371 "for lustre. Use --reformat to reformat it, "
1372 "or tunefs.lustre to modify.\n",
1378 /* Format the backing filesystem */
1379 ret = make_lustre_backfs(&mop);
1382 fprintf(stderr, "mkfs failed %d\n", ret);
1387 /* Write our config files */
1388 ret = write_local_files(&mop);
1391 fprintf(stderr, "failed to write local files\n");
1398 /* Fix any crazy return values from system() */
1399 if (ret && ((ret & 255) == 0))
1402 verrprint("%s: exiting with %d (%s)\n",
1403 progname, ret, strerror(ret));