1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/utils/mkfs_lustre.c
38 * Author: Nathan Rutman <nathan@clusterfs.com>
41 /* This source file is compiled into both mkfs.lustre and tunefs.lustre */
45 #endif /* HAVE_CONFIG_H */
57 #include <sys/types.h>
59 #include <sys/mount.h>
60 #include <sys/utsname.h>
67 /* kp30.h is not really needed here, but on SLES10/PPC, fs.h includes idr.h which
68 * requires BITS_PER_LONG to be defined */
69 #include <libcfs/kp30.h>
70 #include <linux/fs.h> /* for BLKGETSIZE64 */
71 #include <linux/version.h>
73 #include <lustre_disk.h>
74 #include <lustre_param.h>
75 #include <lnet/lnetctl.h>
76 #include <lustre_ver.h>
77 #include "mount_utils.h"
83 #define MAX_LOOP_DEVICES 16
84 #define L_BLOCK_SIZE 4096
85 #define INDEX_UNASSIGNED 0xFFFF
86 #define MO_IS_LOOP 0x01
87 #define MO_FORCEFORMAT 0x02
89 /* used to describe the options to format the lustre disk, not persistent */
91 struct lustre_disk_data mo_ldd; /* to be written in MOUNT_DATA_FILE */
92 char mo_device[128]; /* disk device name */
93 char mo_mkfsopts[128]; /* options to the backing-store mkfs */
94 char mo_loopdev[128]; /* in case a loop dev is needed */
95 __u64 mo_device_sz; /* in KB */
103 static int print_only = 0;
104 static int failover = 0;
106 void usage(FILE *out)
108 fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
109 fprintf(out, "usage: %s <target types> [options] <device>\n", progname);
111 "\t<device>:block device or file (e.g /dev/sda or /tmp/ost1)\n"
113 "\t\t--ost: object storage, mutually exclusive with mdt,mgs\n"
114 "\t\t--mdt: metadata storage, mutually exclusive with ost\n"
115 "\t\t--mgs: configuration management service - one per site\n"
116 "\toptions (in order of popularity):\n"
117 "\t\t--mgsnode=<nid>[,<...>] : NID(s) of a remote mgs node\n"
118 "\t\t\trequired for all targets other than the mgs node\n"
119 "\t\t--fsname=<filesystem_name> : default is 'lustre'\n"
120 "\t\t--failnode=<nid>[,<...>] : NID(s) of a failover partner\n"
121 "\t\t--param <key>=<value> : set a permanent parameter\n"
122 "\t\t\te.g. --param sys.timeout=40\n"
123 "\t\t\t --param lov.stripesize=2M\n"
124 "\t\t--index=#N : target index (i.e. ost index within the lov)\n"
125 /* FIXME implement 1.6.x
126 "\t\t--configdev=<altdevice|file>: store configuration info\n"
127 "\t\t\tfor this device on an alternate device\n"
129 "\t\t--comment=<user comment>: arbitrary user string (%d bytes)\n"
130 "\t\t--mountfsoptions=<opts> : permanent mount options\n"
132 "\t\t--backfstype=<fstype> : backing fs type (ext3, ldiskfs)\n"
133 "\t\t--device-size=#N(KB) : device size for loop devices\n"
134 "\t\t--mkfsoptions=<opts> : format options\n"
135 "\t\t--reformat: overwrite an existing disk\n"
136 "\t\t--stripe-count-hint=#N : used for optimizing MDT inode size\n"
138 "\t\t--erase-params : erase all old parameter settings\n"
139 "\t\t--nomgs: turn off MGS service on this MDT\n"
140 "\t\t--writeconf: erase all config logs for this fs.\n"
142 "\t\t--dryrun: just report what we would do; "
143 "don't write to disk\n"
144 "\t\t--verbose : e.g. show mkfs progress\n"
146 (int)sizeof(((struct lustre_disk_data *)0)->ldd_userdata));
150 #define vprint if (verbose > 0) printf
151 #define verrprint if (verbose >= 0) printf
153 /*================ utility functions =====================*/
155 char *strscat(char *dst, char *src, int buflen) {
157 if (strlen(dst) + strlen(src) >= buflen) {
158 fprintf(stderr, "string buffer overflow (max %d): '%s' + '%s'"
159 "\n", buflen, dst, src);
162 return strcat(dst, src);
166 char *strscpy(char *dst, char *src, int buflen) {
168 return strscat(dst, src, buflen);
172 dev_major (unsigned long long int __dev)
174 return ((__dev >> 8) & 0xfff) | ((unsigned int) (__dev >> 32) & ~0xfff);
178 dev_minor (unsigned long long int __dev)
180 return (__dev & 0xff) | ((unsigned int) (__dev >> 12) & ~0xff);
185 static int version = 0;
189 char release[4] = "";
191 fd = open("/proc/sys/kernel/osrelease", O_RDONLY);
193 fprintf(stderr, "%s: Warning: Can't resolve kernel "
194 "version, assuming 2.6\n", progname);
196 read(fd, release, 4);
199 if (strncmp(release, "2.4.", 4) == 0)
207 static int check_mtab_entry(char *spec)
212 fp = setmntent(MOUNTED, "r");
216 while ((mnt = getmntent(fp)) != NULL) {
217 if (strcmp(mnt->mnt_fsname, spec) == 0) {
219 fprintf(stderr, "%s: according to %s %s is "
220 "already mounted on %s\n",
221 progname, MOUNTED, spec, mnt->mnt_dir);
230 /*============ disk dev functions ===================*/
232 /* Setup a file in the first unused loop_device */
233 int loop_setup(struct mkfs_opts *mop)
239 /* Figure out the loop device names */
240 if (!access("/dev/loop0", F_OK | R_OK))
241 strcpy(loop_base, "/dev/loop\0");
242 else if (!access("/dev/loop/0", F_OK | R_OK))
243 strcpy(loop_base, "/dev/loop/\0");
245 fprintf(stderr, "%s: can't access loop devices\n", progname);
249 /* Find unused loop device */
250 for (i = 0; i < MAX_LOOP_DEVICES; i++) {
252 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 /* someone else picked up this loop device
271 fprintf(stderr, "%s: error %d on losetup: %s\n",
272 progname, ret, strerror(ret));
275 strscpy(mop->mo_loopdev, l_device,
276 sizeof(mop->mo_loopdev));
281 fprintf(stderr, "%s: out of loop devices!\n", progname);
285 int loop_cleanup(struct mkfs_opts *mop)
289 if ((mop->mo_flags & MO_IS_LOOP) && *mop->mo_loopdev) {
290 sprintf(cmd, "losetup -d %s", mop->mo_loopdev);
291 ret = run_command(cmd, sizeof(cmd));
296 /* Determine if a device is a block device (as opposed to a file) */
297 int is_block(char* devname)
302 ret = access(devname, F_OK);
305 ret = stat(devname, &st);
307 fprintf(stderr, "%s: cannot stat %s\n", progname, devname);
310 return S_ISBLK(st.st_mode);
313 __u64 get_device_size(char* device)
318 fd = open(device, O_RDONLY);
320 fprintf(stderr, "%s: cannot open %s: %s\n",
321 progname, device, strerror(errno));
326 /* size in bytes. bz5831 */
327 ret = ioctl(fd, BLKGETSIZE64, (void*)&size);
332 ret = ioctl(fd, BLKGETSIZE, (void*)&lsize);
333 size = (__u64)lsize * 512;
338 fprintf(stderr, "%s: size ioctl failed: %s\n",
339 progname, strerror(errno));
343 vprint("device size = "LPU64"MB\n", size >> 20);
344 /* return value in KB */
348 int loop_format(struct mkfs_opts *mop)
352 if (mop->mo_device_sz == 0) {
354 fprintf(stderr, "loop device requires a --device-size= "
359 ret = creat(mop->mo_device, S_IRUSR|S_IWUSR);
360 ret = truncate(mop->mo_device, mop->mo_device_sz * 1024);
363 fprintf(stderr, "%s: Unable to create backing store: %d\n",
370 /* Display the need for the latest e2fsprogs to be installed. make_backfs
371 * indicates if the caller is make_lustre_backfs() or not. */
372 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
374 static int msg_displayed;
377 fprintf(stderr, "WARNING: %s does not support %s "
378 "feature.\n\n", E2FSPROGS, feature);
384 fprintf(stderr, "WARNING: The %s package currently installed on "
385 "your system does not support \"%s\" feature.\n",
387 #if !(HAVE_LDISKFSPROGS)
388 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
389 "http://downloads.lustre.org/public/tools/e2fsprogs/\n"
390 "to enable this feature.\n");
393 fprintf(stderr, "Feature will not be enabled until %s"
394 "is updated and '%s -O %s %%{device}' "
395 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
398 /* Check whether the file exists in the device */
399 static int file_in_dev(char *file_name, char *dev_name)
402 char debugfs_cmd[256];
403 unsigned int inode_num;
406 /* Construct debugfs command line. */
407 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
408 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
409 DEBUGFS, file_name, dev_name);
411 fp = popen(debugfs_cmd, "r");
413 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
417 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
421 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd), fp);
424 fprintf(stderr, "%s", debugfs_cmd);
425 if (strstr(debugfs_cmd, "unsupported feature")) {
426 disp_old_e2fsprogs_msg("an unknown", 0);
435 /* Check whether the device has already been used with lustre */
436 static int is_lustre_target(struct mkfs_opts *mop)
439 vprint("checking for existing Lustre data: ");
441 if ((rc = file_in_dev(MOUNT_DATA_FILE, mop->mo_device))) {
443 (rc == 1) ? MOUNT_DATA_FILE : "extents");
444 /* in the -1 case, 'extents' means this really IS a lustre
449 if ((rc = file_in_dev(LAST_RCVD, mop->mo_device))) {
450 vprint("found %s\n", LAST_RCVD);
454 vprint("not found\n");
455 return 0; /* The device is not a lustre target. */
458 /* Check if a certain feature is supported by e2fsprogs.
459 * Firstly we try to use "debugfs supported_features" command to check if
460 * the feature is supported. If this fails we try to set this feature with
461 * mke2fs to check for its support. */
462 static int is_e2fsprogs_feature_supp(const char *feature)
466 char imgname[] = "/tmp/test-img-XXXXXX";
470 snprintf(cmd, sizeof(cmd), "%s -c -R \"supported_features %s\" 2>&1",
473 /* Using popen() instead of run_command() since debugfs does not return
474 * proper error code if command is not supported */
475 fp = popen(cmd, "r");
477 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
480 ret = fread(cmd, 1, sizeof(cmd), fp);
482 if (strstr(cmd, feature) && !(strstr(cmd, "Unknown")))
486 if ((fd = mkstemp(imgname)) < 0)
489 snprintf(cmd, sizeof(cmd), "%s -F -O %s %s 100 >/dev/null 2>&1",
490 MKE2FS, feature, imgname);
491 /* run_command() displays the output of mke2fs when it fails for
492 * some feature, so use system() directly */
500 static void disp_old_kernel_msg(char *feature)
502 fprintf(stderr, "WARNING: ldiskfs filesystem does not support \"%s\" "
503 "feature.\n\n", feature);
506 static void enable_default_backfs_features(struct mkfs_opts *mop)
509 int maj_high, maj_low, min;
512 strscat(mop->mo_mkfsopts, " -O dir_index,extents", sizeof(mop->mo_mkfsopts));
514 /* Upstream e2fsprogs called our uninit_groups feature uninit_bg,
515 * check for both of them when testing e2fsprogs features. */
516 if (is_e2fsprogs_feature_supp("uninit_groups") == 0)
517 strscat(mop->mo_mkfsopts, ",uninit_groups",
518 sizeof(mop->mo_mkfsopts));
519 else if (is_e2fsprogs_feature_supp("uninit_bg") == 0)
520 strscat(mop->mo_mkfsopts, ",uninit_bg",
521 sizeof(mop->mo_mkfsopts));
523 disp_old_e2fsprogs_msg("uninit_bg", 1);
529 sscanf(uts.release, "%d.%d.%d", &maj_high, &maj_low, &min);
530 printf("%d %d %d\n", maj_high, maj_low, min);
532 /* Multiple mount protection is enabled only if failover node is
533 * specified and if kernel version is higher than 2.6.9 */
535 if (KERNEL_VERSION(maj_high, maj_low, min) >=
536 KERNEL_VERSION(2,6,9)) {
537 if (is_e2fsprogs_feature_supp("mmp") == 0)
538 strscat(mop->mo_mkfsopts, ",mmp",
539 sizeof(mop->mo_mkfsopts));
541 disp_old_e2fsprogs_msg("mmp", 1);
543 disp_old_kernel_msg("mmp");
547 /* Build fs according to type */
548 int make_lustre_backfs(struct mkfs_opts *mop)
550 char mkfs_cmd[PATH_MAX];
556 if (mop->mo_device_sz != 0) {
557 if (mop->mo_device_sz < 8096){
558 fprintf(stderr, "%s: size of filesystem must be larger "
559 "than 8MB, but is set to %lluKB\n",
560 progname, (long long)mop->mo_device_sz);
563 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
566 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
567 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
568 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
569 __u64 device_sz = mop->mo_device_sz;
571 /* we really need the size */
572 if (device_sz == 0) {
573 device_sz = get_device_size(mop->mo_device);
578 /* Journal size in MB */
579 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
580 /* Choose our own default journal size */
581 long journal_sz = 0, max_sz;
582 if (device_sz > 1024 * 1024) /* 1GB */
583 journal_sz = (device_sz / 102400) * 4;
585 max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
586 if (journal_sz > max_sz)
589 sprintf(buf, " -J size=%ld", journal_sz);
590 strscat(mop->mo_mkfsopts, buf,
591 sizeof(mop->mo_mkfsopts));
595 /* bytes_per_inode: disk size / num inodes */
596 if (strstr(mop->mo_mkfsopts, "-i") == NULL) {
597 long bytes_per_inode = 0;
599 if (IS_MDT(&mop->mo_ldd))
600 bytes_per_inode = 4096;
602 /* Allocate fewer inodes on large OST devices. Most
603 filesystems can be much more aggressive than even
605 if ((IS_OST(&mop->mo_ldd) && (device_sz > 1000000)))
606 bytes_per_inode = 16384;
608 if (bytes_per_inode > 0) {
609 sprintf(buf, " -i %ld", bytes_per_inode);
610 strscat(mop->mo_mkfsopts, buf,
611 sizeof(mop->mo_mkfsopts));
615 /* Inode size (for extended attributes). The LOV EA size is
616 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
617 * and we want some margin above that for ACLs, other EAs... */
618 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
620 if (IS_MDT(&mop->mo_ldd)) {
621 if (mop->mo_stripe_count > 72)
622 inode_size = 512; /* bz 7241 */
623 /* cray stripes across all osts (>60) */
624 else if (mop->mo_stripe_count > 32)
626 else if (mop->mo_stripe_count > 10)
630 } else if (IS_OST(&mop->mo_ldd)) {
631 /* now as we store fids in EA on OST we need
632 to make inode bigger */
636 if (inode_size > 0) {
637 sprintf(buf, " -I %ld", inode_size);
638 strscat(mop->mo_mkfsopts, buf,
639 sizeof(mop->mo_mkfsopts));
644 strscat(mop->mo_mkfsopts, " -q",
645 sizeof(mop->mo_mkfsopts));
648 if (strstr(mop->mo_mkfsopts, "-O") == NULL)
649 enable_default_backfs_features(mop);
651 /* Allow reformat of full devices (as opposed to
652 partitions.) We already checked for mounted dev. */
653 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
655 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
656 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
657 mop->mo_ldd.ldd_svname);
659 } else if (mop->mo_ldd.ldd_mount_type == LDD_MT_REISERFS) {
660 long journal_sz = 0; /* FIXME default journal size */
661 if (journal_sz > 0) {
662 sprintf(buf, " --journal_size %ld", journal_sz);
663 strscat(mop->mo_mkfsopts, buf,
664 sizeof(mop->mo_mkfsopts));
666 snprintf(mkfs_cmd, sizeof(mkfs_cmd), "mkreiserfs -ff ");
669 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
670 progname, mop->mo_ldd.ldd_mount_type,
671 MT_STR(&mop->mo_ldd));
675 /* For loop device format the dev, not the filename */
676 dev = mop->mo_device;
677 if (mop->mo_flags & MO_IS_LOOP)
678 dev = mop->mo_loopdev;
680 vprint("formatting backing filesystem %s on %s\n",
681 MT_STR(&mop->mo_ldd), dev);
682 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
683 vprint("\t4k blocks %d\n", block_count);
684 vprint("\toptions %s\n", mop->mo_mkfsopts);
686 /* mkfs_cmd's trailing space is important! */
687 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
688 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
689 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
690 if (block_count != 0) {
691 sprintf(buf, " %d", block_count);
692 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
695 vprint("mkfs_cmd = %s\n", mkfs_cmd);
696 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
699 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
704 /* ==================== Lustre config functions =============*/
706 void print_ldd(char *str, struct lustre_disk_data *ldd)
708 printf("\n %s:\n", str);
709 printf("Target: %s\n", ldd->ldd_svname);
710 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
711 printf("Index: unassigned\n");
713 printf("Index: %d\n", ldd->ldd_svindex);
714 if (ldd->ldd_uuid[0])
715 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
716 printf("Lustre FS: %s\n", ldd->ldd_fsname);
717 printf("Mount type: %s\n", MT_STR(ldd));
718 printf("Flags: %#x\n", ldd->ldd_flags);
719 printf(" (%s%s%s%s%s%s%s%s)\n",
720 IS_MDT(ldd) ? "MDT ":"",
721 IS_OST(ldd) ? "OST ":"",
722 IS_MGS(ldd) ? "MGS ":"",
723 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
724 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
725 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
726 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
727 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
728 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
729 printf("Parameters:%s\n", ldd->ldd_params);
730 if (ldd->ldd_userdata[0])
731 printf("Comment: %s\n", ldd->ldd_userdata);
735 /* Write the server config files */
736 int write_local_files(struct mkfs_opts *mop)
738 char mntpt[] = "/tmp/mntXXXXXX";
744 /* Mount this device temporarily in order to write these files */
745 if (!mkdtemp(mntpt)) {
746 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
747 progname, mntpt, strerror(errno));
751 dev = mop->mo_device;
752 if (mop->mo_flags & MO_IS_LOOP)
753 dev = mop->mo_loopdev;
755 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0, NULL);
757 fprintf(stderr, "%s: Unable to mount %s: %s\n",
758 progname, dev, strerror(errno));
760 if (errno == ENODEV) {
761 fprintf(stderr, "Is the %s module available?\n",
762 MT_STR(&mop->mo_ldd));
767 /* Set up initial directories */
768 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
769 ret = mkdir(filepnm, 0777);
770 if ((ret != 0) && (errno != EEXIST)) {
771 fprintf(stderr, "%s: Can't make configs dir %s: %s\n",
772 progname, filepnm, strerror(errno));
774 } else if (errno == EEXIST) {
778 /* Save the persistent mount data into a file. Lustre must pre-read
779 this file to get the real mount options. */
780 vprint("Writing %s\n", MOUNT_DATA_FILE);
781 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
782 filep = fopen(filepnm, "w");
784 fprintf(stderr, "%s: Unable to create %s file: %s\n",
785 progname, filepnm, strerror(errno));
788 fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
793 /* Check for upgrade */
794 if ((mop->mo_ldd.ldd_flags & (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS))
795 == (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS)) {
798 int cmdsz = sizeof(cmd);
799 vprint("Copying old logs\n");
801 /* Copy the old client log to fsname-client */
802 sprintf(filepnm, "%s/%s/%s-client",
803 mntpt, MOUNT_CONFIGS_DIR, mop->mo_ldd.ldd_fsname);
804 snprintf(cmd, cmdsz, "cp %s/%s/client %s", mntpt, MDT_LOGS_DIR,
806 ret = run_command(cmd, cmdsz);
808 fprintf(stderr, "%s: Can't copy 1.4 config %s/client "
809 "(%d)\n", progname, MDT_LOGS_DIR, ret);
810 fprintf(stderr, "mount -t ldiskfs %s somewhere, "
811 "find the client log for fs %s and "
812 "copy it manually into %s/%s-client, "
815 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
816 mop->mo_ldd.ldd_fsname);
820 /* We need to use the old mdt log because otherwise mdt won't
821 have complete lov if old clients connect before all
823 /* Copy the old mdt log to fsname-MDT0000 (get old
824 name from mdt_UUID) */
826 strscpy(filepnm, (char *)mop->mo_ldd.ldd_uuid, sizeof(filepnm));
827 term = strstr(filepnm, "_UUID");
830 snprintf(cmd, cmdsz, "cp %s/%s/%s %s/%s/%s",
831 mntpt, MDT_LOGS_DIR, filepnm,
832 mntpt, MOUNT_CONFIGS_DIR,
833 mop->mo_ldd.ldd_svname);
834 ret = run_command(cmd, cmdsz);
837 fprintf(stderr, "%s: Can't copy 1.4 config %s/%s "
838 "(%d)\n", progname, MDT_LOGS_DIR, filepnm, ret);
839 fprintf(stderr, "mount -t ext3 %s somewhere, "
840 "find the MDT log for fs %s and "
841 "copy it manually into %s/%s, "
844 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
845 mop->mo_ldd.ldd_svname);
860 int read_local_files(struct mkfs_opts *mop)
862 char tmpdir[] = "/tmp/dirXXXXXX";
868 int cmdsz = sizeof(cmd);
870 /* Make a temporary directory to hold Lustre data files. */
871 if (!mkdtemp(tmpdir)) {
872 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
873 progname, tmpdir, strerror(errno));
877 dev = mop->mo_device;
879 /* TODO: it's worth observing the get_mountdata() function that is
880 in mount_utils.c for getting the mountdata out of the
883 /* Construct debugfs command line. */
884 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
885 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
887 ret = run_command(cmd, cmdsz);
889 verrprint("%s: Unable to dump %s dir (%d)\n",
890 progname, MOUNT_CONFIGS_DIR, ret);
892 sprintf(filepnm, "%s/mountdata", tmpdir);
893 filep = fopen(filepnm, "r");
895 vprint("Reading %s\n", MOUNT_DATA_FILE);
896 fread(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
899 /* Try to read pre-1.6 config from last_rcvd */
900 struct lr_server_data lsd;
901 verrprint("%s: Unable to read %d.%d config %s.\n",
902 progname, LUSTRE_MAJOR, LUSTRE_MINOR, filepnm);
904 verrprint("Trying 1.4 config from last_rcvd\n");
905 sprintf(filepnm, "%s/%s", tmpdir, LAST_RCVD);
907 /* Construct debugfs command line. */
908 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s' %s",
909 DEBUGFS, LAST_RCVD, filepnm, dev);
911 ret = run_command(cmd, cmdsz);
913 fprintf(stderr, "%s: Unable to dump %s file (%d)\n",
914 progname, LAST_RCVD, ret);
918 filep = fopen(filepnm, "r");
920 fprintf(stderr, "%s: Unable to open %s: %s\n",
921 progname, filepnm, strerror(errno));
923 verrprint("Contents of %s:\n", tmpdir);
925 snprintf(cmd, cmdsz, "ls -l %s/", tmpdir);
926 run_command(cmd, cmdsz);
927 verrprint("Contents of disk:\n");
928 snprintf(cmd, cmdsz, "%s -c -R 'ls -l /' %s",
930 run_command(cmd, cmdsz);
934 vprint("Reading %s\n", LAST_RCVD);
935 ret = fread(&lsd, 1, sizeof(lsd), filep);
936 if (ret < sizeof(lsd)) {
937 fprintf(stderr, "%s: Short read (%d of %d)\n",
938 progname, ret, (int)sizeof(lsd));
943 vprint("Feature compat=%x, incompat=%x\n",
944 lsd.lsd_feature_compat, lsd.lsd_feature_incompat);
946 if ((lsd.lsd_feature_compat & OBD_COMPAT_OST) ||
947 (lsd.lsd_feature_incompat & OBD_INCOMPAT_OST)) {
948 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
949 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
950 } else if ((lsd.lsd_feature_compat & OBD_COMPAT_MDT) ||
951 (lsd.lsd_feature_incompat & OBD_INCOMPAT_MDT)) {
952 /* We must co-locate so mgs can see old logs.
953 If user doesn't want this, they can copy the old
954 logs manually and re-tunefs. */
955 mop->mo_ldd.ldd_flags =
956 LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_MGS;
957 mop->mo_ldd.ldd_svindex = lsd.lsd_mdt_index;
959 /* If neither is set, we're pre-1.4.6, make a guess. */
960 /* Construct debugfs command line. */
961 snprintf(cmd, cmdsz, "%s -c -R 'rdump /%s %s' %s",
962 DEBUGFS, MDT_LOGS_DIR, tmpdir, dev);
963 run_command(cmd, cmdsz);
965 sprintf(filepnm, "%s/%s", tmpdir, MDT_LOGS_DIR);
966 if (lsd.lsd_ost_index > 0) {
967 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
968 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
970 /* If there's a LOGS dir, it's an MDT */
971 if ((ret = access(filepnm, F_OK)) == 0) {
972 mop->mo_ldd.ldd_flags =
975 /* Old MDT's are always index 0
977 mop->mo_ldd.ldd_svindex = 0;
979 /* The index may not be correct */
980 mop->mo_ldd.ldd_flags =
981 LDD_F_SV_TYPE_OST | LDD_F_NEED_INDEX;
982 verrprint("OST with unknown index\n");
988 memcpy(mop->mo_ldd.ldd_uuid, lsd.lsd_uuid,
989 sizeof(mop->mo_ldd.ldd_uuid));
990 mop->mo_ldd.ldd_flags |= LDD_F_UPGRADE14;
997 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
998 run_command(cmd, cmdsz);
1000 verrprint("Failed to read old data (%d)\n", ret);
1005 void set_defaults(struct mkfs_opts *mop)
1007 mop->mo_ldd.ldd_magic = LDD_MAGIC;
1008 mop->mo_ldd.ldd_config_ver = 1;
1009 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
1010 mop->mo_mgs_failnodes = 0;
1011 strcpy(mop->mo_ldd.ldd_fsname, "lustre");
1012 if (get_os_version() == 24)
1013 mop->mo_ldd.ldd_mount_type = LDD_MT_EXT3;
1015 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
1017 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
1018 mop->mo_stripe_count = 1;
1021 static inline void badopt(const char *opt, char *type)
1023 fprintf(stderr, "%s: '--%s' only valid for %s\n",
1024 progname, opt, type);
1028 static int clean_param(char *buf, char *key)
1034 if ((sub = strstr(buf, key)) != NULL) {
1035 if ((next = strchr(sub, ' ')) != NULL) {
1037 memmove(sub, next, strlen(next) + 1);
1045 static int add_param(char *buf, char *key, char *val, int unique)
1047 int end = sizeof(((struct lustre_disk_data *)0)->ldd_params);
1053 keylen = strlen(key);
1056 clean_param(buf, key);
1058 if ((ptr = strchr(val, '=')) == NULL)
1061 clean_param(buf, val);
1066 start = strlen(buf);
1067 if (start + 1 + keylen + strlen(val) >= end) {
1068 fprintf(stderr, "%s: params are too long-\n%s %s%s\n",
1069 progname, buf, key ? key : "", val);
1073 sprintf(buf + start, " %s%s", key ? key : "", val);
1077 /* from mount_lustre */
1078 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
1079 #define MAXNIDSTR 1024
1080 static char *convert_hostnames(char *s1)
1082 char *converted, *s2 = 0, *c, *end, sep;
1083 int left = MAXNIDSTR;
1086 converted = malloc(left);
1087 if (converted == NULL) {
1088 fprintf(stderr, "out of memory: needed %d bytes\n",
1092 end = s1 + strlen(s1);
1094 while ((left > 0) && (s1 < end)) {
1095 s2 = strpbrk(s1, ",:");
1100 nid = libcfs_str2nid(s1);
1102 if (nid == LNET_NID_ANY) {
1103 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
1107 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
1108 strlen("127.0.0.1")) == 0) {
1109 fprintf(stderr, "%s: The NID '%s' resolves to the "
1110 "loopback address '%s'. Lustre requires a "
1111 "non-loopback address.\n",
1112 progname, s1, libcfs_nid2str(nid));
1117 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
1118 left = converted + MAXNIDSTR - c;
1124 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
1127 static struct option long_opt[] = {
1128 {"backfstype", 1, 0, 'b'},
1129 {"stripe-count-hint", 1, 0, 'c'},
1130 {"comment", 1, 0, 'u'},
1131 {"configdev", 1, 0, 'C'},
1132 {"device-size", 1, 0, 'd'},
1133 {"dryrun", 0, 0, 'n'},
1134 {"erase-params", 0, 0, 'e'},
1135 {"failnode", 1, 0, 'f'},
1136 {"failover", 1, 0, 'f'},
1138 {"help", 0, 0, 'h'},
1139 {"index", 1, 0, 'i'},
1140 {"mkfsoptions", 1, 0, 'k'},
1141 {"mgsnode", 1, 0, 'm'},
1142 {"mgsnid", 1, 0, 'm'},
1144 {"fsname",1, 0, 'L'},
1145 {"noformat", 0, 0, 'n'},
1146 {"nomgs", 0, 0, 'N'},
1147 {"mountfsoptions", 1, 0, 'o'},
1149 {"param", 1, 0, 'p'},
1150 {"print", 0, 0, 'n'},
1151 {"quiet", 0, 0, 'q'},
1152 {"reformat", 0, 0, 'r'},
1153 {"verbose", 0, 0, 'v'},
1154 {"writeconf", 0, 0, 'w'},
1157 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqru:vw";
1161 const size_t prefix_len = sizeof(PARAM_MDT_UPCALL) - 1;
1163 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
1168 while (i < LDD_MT_LAST) {
1169 if (strcmp(optarg, mt_str(i)) == 0) {
1170 mop->mo_ldd.ldd_mount_type = i;
1178 if (IS_MDT(&mop->mo_ldd)) {
1179 int stripe_count = atol(optarg);
1180 if (stripe_count <= 0) {
1181 fprintf(stderr, "%s: bad stripe count "
1182 "%d\n", progname, stripe_count);
1185 mop->mo_stripe_count = stripe_count;
1187 badopt(long_opt[longidx].name, "MDT");
1191 case 'C': /* Configdev */
1193 printf("Configdev not implemented\n");
1196 mop->mo_device_sz = atol(optarg);
1199 mop->mo_ldd.ldd_params[0] = '\0';
1200 /* Must update the mgs logs */
1201 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1204 char *nids = convert_hostnames(optarg);
1207 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
1212 /* Must update the mgs logs */
1213 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1218 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
1224 if (!(mop->mo_ldd.ldd_flags &
1225 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1226 LDD_F_WRITECONF))) {
1227 fprintf(stderr, "%s: cannot change the index of"
1228 " a registered target\n", progname);
1231 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
1232 mop->mo_ldd.ldd_svindex = atol(optarg);
1233 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
1235 badopt(long_opt[longidx].name, "MDT,OST");
1240 strscpy(mop->mo_mkfsopts, optarg,
1241 sizeof(mop->mo_mkfsopts));
1245 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
1246 (!(mop->mo_ldd.ldd_flags &
1247 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1248 LDD_F_WRITECONF)))) {
1249 fprintf(stderr, "%s: cannot change the name of"
1250 " a registered target\n", progname);
1253 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
1254 fprintf(stderr, "%s: filesystem name must be "
1255 "1-8 chars\n", progname);
1258 if ((tmp = strpbrk(optarg, "/:"))) {
1259 fprintf(stderr, "%s: char '%c' not allowed in "
1260 "filesystem name\n", progname, *tmp);
1263 strscpy(mop->mo_ldd.ldd_fsname, optarg,
1264 sizeof(mop->mo_ldd.ldd_fsname));
1268 char *nids = convert_hostnames(optarg);
1271 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
1276 mop->mo_mgs_failnodes++;
1280 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
1286 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
1289 *mountopts = optarg;
1292 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
1295 /* Test if the param is valid for mdt.group_upcall */
1296 if (!strncmp(optarg, PARAM_MDT_UPCALL, prefix_len)) {
1298 if (strcmp(optarg + prefix_len, "NONE") &&
1299 access(optarg + prefix_len, R_OK | X_OK))
1300 fprintf(stderr, "WARNING: group upcall "
1301 "parameter not executable: %s\n"
1302 "NOTE: you can change the path "
1303 "to the group upcall through "
1304 "tunefs.lustre(8)\n", optarg +
1307 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg, 0);
1310 /* Must update the mgs logs */
1311 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1317 mop->mo_flags |= MO_FORCEFORMAT;
1320 strscpy(mop->mo_ldd.ldd_userdata, optarg,
1321 sizeof(mop->mo_ldd.ldd_userdata));
1327 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
1332 fprintf(stderr, "Unknown option '%c'\n", opt);
1338 /* Last arg is device */
1339 if (optind != argc - 1) {
1341 fprintf(stderr, "Bad argument: %s\n", argv[optind]);
1345 /* single argument: <device> */
1350 if (mop->mo_ldd.ldd_flags & LDD_F_SV_TYPE_MDT && 0 == upcall) {
1351 if (access("/usr/sbin/l_getgroups", R_OK | X_OK)) {
1352 fprintf(stderr, "WARNING: MDS group upcall is not set, "
1355 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MDT_UPCALL,
1356 "/usr/sbin/l_getgroups", 1);
1359 /* Must update the mgs logs */
1360 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1368 int main(int argc, char *const argv[])
1370 struct mkfs_opts mop;
1371 struct lustre_disk_data *ldd;
1372 char *mountopts = NULL;
1373 char always_mountopts[512] = "";
1374 char default_mountopts[512] = "";
1377 if ((progname = strrchr(argv[0], '/')) != NULL)
1382 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
1387 memset(&mop, 0, sizeof(mop));
1390 /* device is last arg */
1391 strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
1393 /* Are we using a loop device? */
1394 ret = is_block(mop.mo_device);
1398 mop.mo_flags |= MO_IS_LOOP;
1401 /* For tunefs, we must read in the old values before parsing any
1404 /* Check whether the disk has already been formatted by mkfs.lustre */
1405 ret = is_lustre_target(&mop);
1408 fprintf(stderr, "Device %s has not been formatted with "
1409 "mkfs.lustre\n", mop.mo_device);
1414 ret = read_local_files(&mop);
1417 fprintf(stderr, "Failed to read previous Lustre data from %s "
1418 "(%d)\n", mop.mo_device, ret);
1421 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
1422 mop.mo_mgs_failnodes++;
1425 print_ldd("Read previous values", &(mop.mo_ldd));
1428 ret = parse_opts(argc, argv, &mop, &mountopts);
1434 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
1436 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
1441 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
1443 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
1448 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
1449 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
1451 fprintf(stderr, "Can't find the target index, "
1452 "specify with --index\n");
1457 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1460 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
1462 fprintf(stderr, "Must specify --mgsnode=\n");
1467 /* These are the permanent mount options (always included) */
1468 switch (ldd->ldd_mount_type) {
1470 case LDD_MT_LDISKFS:
1471 case LDD_MT_LDISKFS2: {
1472 sprintf(always_mountopts, "errors=remount-ro");
1473 if (IS_MDT(ldd) || IS_MGS(ldd))
1474 strscat(always_mountopts, ",iopen_nopriv,user_xattr",
1475 sizeof(always_mountopts));
1476 if ((get_os_version() == 24) && IS_OST(ldd))
1477 strscat(always_mountopts, ",asyncdel",
1478 sizeof(always_mountopts));
1479 /* NB: Files created while extents are enabled cannot be read
1480 if mounted with a kernel that doesn't include the Lustre ldiskfs
1483 (ldd->ldd_mount_type == LDD_MT_LDISKFS ||
1484 ldd->ldd_mount_type == LDD_MT_LDISKFS2)) {
1485 strscat(default_mountopts, ",extents,mballoc",
1486 sizeof(default_mountopts));
1491 mop.mo_flags |= MO_IS_LOOP;
1492 sprintf(always_mountopts, "type=ext3,dev=%s",
1498 fprintf(stderr, "unknown fs type %d '%s'\n",
1499 ldd->ldd_mount_type,
1507 /* If user specifies mount opts, don't use defaults,
1508 but always use always_mountopts */
1509 sprintf(ldd->ldd_mount_opts, "%s,%s",
1510 always_mountopts, mountopts);
1513 if (ldd->ldd_mount_opts[0] == 0)
1514 /* use the defaults unless old opts exist */
1517 sprintf(ldd->ldd_mount_opts, "%s%s",
1518 always_mountopts, default_mountopts);
1522 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
1523 ldd->ldd_fsname, ldd->ldd_svname);
1526 print_ldd("Permanent disk data", ldd);
1529 printf("exiting before disk write.\n");
1533 if (check_mtab_entry(mop.mo_device))
1536 /* Create the loopback file */
1537 if (mop.mo_flags & MO_IS_LOOP) {
1538 ret = access(mop.mo_device, F_OK);
1541 #ifndef TUNEFS /* mkfs.lustre */
1542 /* Reformat the loopback file */
1543 if (ret || (mop.mo_flags & MO_FORCEFORMAT))
1544 ret = loop_format(&mop);
1547 ret = loop_setup(&mop);
1550 fprintf(stderr, "Loop device setup for %s failed: %s\n",
1551 mop.mo_device, strerror(ret));
1556 #ifndef TUNEFS /* mkfs.lustre */
1557 /* Check whether the disk has already been formatted by mkfs.lustre */
1558 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
1559 ret = is_lustre_target(&mop);
1562 fprintf(stderr, "Device %s was previously formatted "
1563 "for lustre. Use --reformat to reformat it, "
1564 "or tunefs.lustre to modify.\n",
1570 /* Format the backing filesystem */
1571 ret = make_lustre_backfs(&mop);
1574 fprintf(stderr, "mkfs failed %d\n", ret);
1579 /* Write our config files */
1580 ret = write_local_files(&mop);
1583 fprintf(stderr, "failed to write local files\n");
1590 /* Fix any crazy return values from system() */
1591 if (ret && ((ret & 255) == 0))
1594 verrprint("%s: exiting with %d (%s)\n",
1595 progname, ret, strerror(ret));