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 if (read(fd, release, 4) < 0) {
197 fprintf(stderr, "reading from /proc/sys/kernel"
198 "/osrelease: %s\n", strerror(errno));
204 if (strncmp(release, "2.4.", 4) == 0)
212 static int check_mtab_entry(char *spec)
217 fp = setmntent(MOUNTED, "r");
221 while ((mnt = getmntent(fp)) != NULL) {
222 if (strcmp(mnt->mnt_fsname, spec) == 0) {
224 fprintf(stderr, "%s: according to %s %s is "
225 "already mounted on %s\n",
226 progname, MOUNTED, spec, mnt->mnt_dir);
235 /*============ disk dev functions ===================*/
237 /* Setup a file in the first unused loop_device */
238 int loop_setup(struct mkfs_opts *mop)
244 /* Figure out the loop device names */
245 if (!access("/dev/loop0", F_OK | R_OK))
246 strcpy(loop_base, "/dev/loop\0");
247 else if (!access("/dev/loop/0", F_OK | R_OK))
248 strcpy(loop_base, "/dev/loop/\0");
250 fprintf(stderr, "%s: can't access loop devices\n", progname);
254 /* Find unused loop device */
255 for (i = 0; i < MAX_LOOP_DEVICES; i++) {
257 int cmdsz = sizeof(cmd);
259 sprintf(l_device, "%s%d", loop_base, i);
260 if (access(l_device, F_OK | R_OK))
262 snprintf(cmd, cmdsz, "losetup %s > /dev/null 2>&1", l_device);
265 /* losetup gets 1 (ret=256) for non-set-up device */
267 /* Set up a loopback device to our file */
268 snprintf(cmd, cmdsz, "losetup %s %s", l_device,
270 ret = run_command(cmd, cmdsz);
272 /* someone else picked up this loop device
276 fprintf(stderr, "%s: error %d on losetup: %s\n",
277 progname, ret, strerror(ret));
280 strscpy(mop->mo_loopdev, l_device,
281 sizeof(mop->mo_loopdev));
286 fprintf(stderr, "%s: out of loop devices!\n", progname);
290 int loop_cleanup(struct mkfs_opts *mop)
294 if ((mop->mo_flags & MO_IS_LOOP) && *mop->mo_loopdev) {
295 sprintf(cmd, "losetup -d %s", mop->mo_loopdev);
296 ret = run_command(cmd, sizeof(cmd));
301 /* Determine if a device is a block device (as opposed to a file) */
302 int is_block(char* devname)
307 ret = access(devname, F_OK);
310 ret = stat(devname, &st);
312 fprintf(stderr, "%s: cannot stat %s\n", progname, devname);
315 return S_ISBLK(st.st_mode);
318 __u64 get_device_size(char* device)
323 fd = open(device, O_RDONLY);
325 fprintf(stderr, "%s: cannot open %s: %s\n",
326 progname, device, strerror(errno));
331 /* size in bytes. bz5831 */
332 ret = ioctl(fd, BLKGETSIZE64, (void*)&size);
337 ret = ioctl(fd, BLKGETSIZE, (void*)&lsize);
338 size = (__u64)lsize * 512;
343 fprintf(stderr, "%s: size ioctl failed: %s\n",
344 progname, strerror(errno));
348 vprint("device size = "LPU64"MB\n", size >> 20);
349 /* return value in KB */
353 int loop_format(struct mkfs_opts *mop)
357 if (mop->mo_device_sz == 0) {
359 fprintf(stderr, "loop device requires a --device-size= "
364 ret = creat(mop->mo_device, S_IRUSR|S_IWUSR);
365 ret = truncate(mop->mo_device, mop->mo_device_sz * 1024);
368 fprintf(stderr, "%s: Unable to create backing store: %d\n",
375 /* Display the need for the latest e2fsprogs to be installed. make_backfs
376 * indicates if the caller is make_lustre_backfs() or not. */
377 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
379 static int msg_displayed;
382 fprintf(stderr, "WARNING: %s does not support %s "
383 "feature.\n\n", E2FSPROGS, feature);
389 fprintf(stderr, "WARNING: The %s package currently installed on "
390 "your system does not support \"%s\" feature.\n",
392 #if !(HAVE_LDISKFSPROGS)
393 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
394 "http://downloads.lustre.org/public/tools/e2fsprogs/\n"
395 "to enable this feature.\n");
398 fprintf(stderr, "Feature will not be enabled until %s"
399 "is updated and '%s -O %s %%{device}' "
400 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
403 /* Check whether the file exists in the device */
404 static int file_in_dev(char *file_name, char *dev_name)
407 char debugfs_cmd[256];
408 unsigned int inode_num;
411 /* Construct debugfs command line. */
412 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
413 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
414 DEBUGFS, file_name, dev_name);
416 fp = popen(debugfs_cmd, "r");
418 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
422 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
426 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd), fp);
429 fprintf(stderr, "%s", debugfs_cmd);
430 if (strstr(debugfs_cmd, "unsupported feature")) {
431 disp_old_e2fsprogs_msg("an unknown", 0);
440 /* Check whether the device has already been used with lustre */
441 static int is_lustre_target(struct mkfs_opts *mop)
444 vprint("checking for existing Lustre data: ");
446 if ((rc = file_in_dev(MOUNT_DATA_FILE, mop->mo_device))) {
448 (rc == 1) ? MOUNT_DATA_FILE : "extents");
449 /* in the -1 case, 'extents' means this really IS a lustre
454 if ((rc = file_in_dev(LAST_RCVD, mop->mo_device))) {
455 vprint("found %s\n", LAST_RCVD);
459 vprint("not found\n");
460 return 0; /* The device is not a lustre target. */
463 /* Check if a certain feature is supported by e2fsprogs.
464 * Firstly we try to use "debugfs supported_features" command to check if
465 * the feature is supported. If this fails we try to set this feature with
466 * mke2fs to check for its support. */
467 static int is_e2fsprogs_feature_supp(const char *feature)
471 char imgname[] = "/tmp/test-img-XXXXXX";
475 snprintf(cmd, sizeof(cmd), "%s -c -R \"supported_features %s\" 2>&1",
478 /* Using popen() instead of run_command() since debugfs does not return
479 * proper error code if command is not supported */
480 fp = popen(cmd, "r");
482 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
485 ret = fread(cmd, 1, sizeof(cmd), fp);
487 if (strstr(cmd, feature) && !(strstr(cmd, "Unknown")))
491 if ((fd = mkstemp(imgname)) < 0)
494 snprintf(cmd, sizeof(cmd), "%s -F -O %s %s 100 >/dev/null 2>&1",
495 MKE2FS, feature, imgname);
496 /* run_command() displays the output of mke2fs when it fails for
497 * some feature, so use system() directly */
505 static void disp_old_kernel_msg(char *feature)
507 fprintf(stderr, "WARNING: ldiskfs filesystem does not support \"%s\" "
508 "feature.\n\n", feature);
511 static void enable_default_backfs_features(struct mkfs_opts *mop)
514 int maj_high, maj_low, min;
517 strscat(mop->mo_mkfsopts, " -O dir_index,extents", sizeof(mop->mo_mkfsopts));
519 /* Upstream e2fsprogs called our uninit_groups feature uninit_bg,
520 * check for both of them when testing e2fsprogs features. */
521 if (is_e2fsprogs_feature_supp("uninit_groups") == 0)
522 strscat(mop->mo_mkfsopts, ",uninit_groups",
523 sizeof(mop->mo_mkfsopts));
524 else if (is_e2fsprogs_feature_supp("uninit_bg") == 0)
525 strscat(mop->mo_mkfsopts, ",uninit_bg",
526 sizeof(mop->mo_mkfsopts));
528 disp_old_e2fsprogs_msg("uninit_bg", 1);
534 sscanf(uts.release, "%d.%d.%d", &maj_high, &maj_low, &min);
535 printf("%d %d %d\n", maj_high, maj_low, min);
537 /* Multiple mount protection is enabled only if failover node is
538 * specified and if kernel version is higher than 2.6.9 */
540 if (KERNEL_VERSION(maj_high, maj_low, min) >=
541 KERNEL_VERSION(2,6,9)) {
542 if (is_e2fsprogs_feature_supp("mmp") == 0)
543 strscat(mop->mo_mkfsopts, ",mmp",
544 sizeof(mop->mo_mkfsopts));
546 disp_old_e2fsprogs_msg("mmp", 1);
548 disp_old_kernel_msg("mmp");
552 /* Build fs according to type */
553 int make_lustre_backfs(struct mkfs_opts *mop)
555 char mkfs_cmd[PATH_MAX];
561 if (mop->mo_device_sz != 0) {
562 if (mop->mo_device_sz < 8096){
563 fprintf(stderr, "%s: size of filesystem must be larger "
564 "than 8MB, but is set to %lluKB\n",
565 progname, (long long)mop->mo_device_sz);
568 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
571 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
572 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
573 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
574 __u64 device_sz = mop->mo_device_sz;
576 /* we really need the size */
577 if (device_sz == 0) {
578 device_sz = get_device_size(mop->mo_device);
583 /* Journal size in MB */
584 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
585 /* Choose our own default journal size */
586 long journal_sz = 0, max_sz;
587 if (device_sz > 1024 * 1024) /* 1GB */
588 journal_sz = (device_sz / 102400) * 4;
590 max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
591 if (journal_sz > max_sz)
594 sprintf(buf, " -J size=%ld", journal_sz);
595 strscat(mop->mo_mkfsopts, buf,
596 sizeof(mop->mo_mkfsopts));
600 /* bytes_per_inode: disk size / num inodes */
601 if (strstr(mop->mo_mkfsopts, "-i") == NULL) {
602 long bytes_per_inode = 0;
604 if (IS_MDT(&mop->mo_ldd))
605 bytes_per_inode = 4096;
607 /* Allocate fewer inodes on large OST devices. Most
608 filesystems can be much more aggressive than even
610 if ((IS_OST(&mop->mo_ldd) && (device_sz > 1000000)))
611 bytes_per_inode = 16384;
613 if (bytes_per_inode > 0) {
614 sprintf(buf, " -i %ld", bytes_per_inode);
615 strscat(mop->mo_mkfsopts, buf,
616 sizeof(mop->mo_mkfsopts));
620 /* Inode size (for extended attributes). The LOV EA size is
621 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
622 * and we want some margin above that for ACLs, other EAs... */
623 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
625 if (IS_MDT(&mop->mo_ldd)) {
626 if (mop->mo_stripe_count > 72)
627 inode_size = 512; /* bz 7241 */
628 /* cray stripes across all osts (>60) */
629 else if (mop->mo_stripe_count > 32)
631 else if (mop->mo_stripe_count > 10)
635 } else if (IS_OST(&mop->mo_ldd)) {
636 /* now as we store fids in EA on OST we need
637 to make inode bigger */
641 if (inode_size > 0) {
642 sprintf(buf, " -I %ld", inode_size);
643 strscat(mop->mo_mkfsopts, buf,
644 sizeof(mop->mo_mkfsopts));
649 strscat(mop->mo_mkfsopts, " -q",
650 sizeof(mop->mo_mkfsopts));
653 if (strstr(mop->mo_mkfsopts, "-O") == NULL)
654 enable_default_backfs_features(mop);
656 /* Allow reformat of full devices (as opposed to
657 partitions.) We already checked for mounted dev. */
658 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
660 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
661 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
662 mop->mo_ldd.ldd_svname);
664 } else if (mop->mo_ldd.ldd_mount_type == LDD_MT_REISERFS) {
665 long journal_sz = 0; /* FIXME default journal size */
666 if (journal_sz > 0) {
667 sprintf(buf, " --journal_size %ld", journal_sz);
668 strscat(mop->mo_mkfsopts, buf,
669 sizeof(mop->mo_mkfsopts));
671 snprintf(mkfs_cmd, sizeof(mkfs_cmd), "mkreiserfs -ff ");
674 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
675 progname, mop->mo_ldd.ldd_mount_type,
676 MT_STR(&mop->mo_ldd));
680 /* For loop device format the dev, not the filename */
681 dev = mop->mo_device;
682 if (mop->mo_flags & MO_IS_LOOP)
683 dev = mop->mo_loopdev;
685 vprint("formatting backing filesystem %s on %s\n",
686 MT_STR(&mop->mo_ldd), dev);
687 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
688 vprint("\t4k blocks %d\n", block_count);
689 vprint("\toptions %s\n", mop->mo_mkfsopts);
691 /* mkfs_cmd's trailing space is important! */
692 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
693 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
694 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
695 if (block_count != 0) {
696 sprintf(buf, " %d", block_count);
697 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
700 vprint("mkfs_cmd = %s\n", mkfs_cmd);
701 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
704 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
709 /* ==================== Lustre config functions =============*/
711 void print_ldd(char *str, struct lustre_disk_data *ldd)
713 printf("\n %s:\n", str);
714 printf("Target: %s\n", ldd->ldd_svname);
715 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
716 printf("Index: unassigned\n");
718 printf("Index: %d\n", ldd->ldd_svindex);
719 if (ldd->ldd_uuid[0])
720 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
721 printf("Lustre FS: %s\n", ldd->ldd_fsname);
722 printf("Mount type: %s\n", MT_STR(ldd));
723 printf("Flags: %#x\n", ldd->ldd_flags);
724 printf(" (%s%s%s%s%s%s%s%s)\n",
725 IS_MDT(ldd) ? "MDT ":"",
726 IS_OST(ldd) ? "OST ":"",
727 IS_MGS(ldd) ? "MGS ":"",
728 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
729 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
730 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
731 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
732 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
733 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
734 printf("Parameters:%s\n", ldd->ldd_params);
735 if (ldd->ldd_userdata[0])
736 printf("Comment: %s\n", ldd->ldd_userdata);
740 /* Write the server config files */
741 int write_local_files(struct mkfs_opts *mop)
743 char mntpt[] = "/tmp/mntXXXXXX";
750 /* Mount this device temporarily in order to write these files */
751 if (!mkdtemp(mntpt)) {
752 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
753 progname, mntpt, strerror(errno));
757 dev = mop->mo_device;
758 if (mop->mo_flags & MO_IS_LOOP)
759 dev = mop->mo_loopdev;
761 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0, NULL);
763 fprintf(stderr, "%s: Unable to mount %s: %s\n",
764 progname, dev, strerror(errno));
766 if (errno == ENODEV) {
767 fprintf(stderr, "Is the %s module available?\n",
768 MT_STR(&mop->mo_ldd));
773 /* Set up initial directories */
774 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
775 ret = mkdir(filepnm, 0777);
776 if ((ret != 0) && (errno != EEXIST)) {
777 fprintf(stderr, "%s: Can't make configs dir %s: %s\n",
778 progname, filepnm, strerror(errno));
780 } else if (errno == EEXIST) {
784 /* Save the persistent mount data into a file. Lustre must pre-read
785 this file to get the real mount options. */
786 vprint("Writing %s\n", MOUNT_DATA_FILE);
787 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
788 filep = fopen(filepnm, "w");
790 fprintf(stderr, "%s: Unable to create %s file: %s\n",
791 progname, filepnm, strerror(errno));
794 num = fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
795 if (num < 1 && ferror(filep)) {
796 fprintf(stderr, "%s: Unable to write to file (%s): %s\n",
797 progname, filepnm, strerror(errno));
804 /* Check for upgrade */
805 if ((mop->mo_ldd.ldd_flags & (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS))
806 == (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS)) {
809 int cmdsz = sizeof(cmd);
810 vprint("Copying old logs\n");
812 /* Copy the old client log to fsname-client */
813 sprintf(filepnm, "%s/%s/%s-client",
814 mntpt, MOUNT_CONFIGS_DIR, mop->mo_ldd.ldd_fsname);
815 snprintf(cmd, cmdsz, "cp %s/%s/client %s", mntpt, MDT_LOGS_DIR,
817 ret = run_command(cmd, cmdsz);
819 fprintf(stderr, "%s: Can't copy 1.4 config %s/client "
820 "(%d)\n", progname, MDT_LOGS_DIR, ret);
821 fprintf(stderr, "mount -t ldiskfs %s somewhere, "
822 "find the client log for fs %s and "
823 "copy it manually into %s/%s-client, "
826 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
827 mop->mo_ldd.ldd_fsname);
831 /* We need to use the old mdt log because otherwise mdt won't
832 have complete lov if old clients connect before all
834 /* Copy the old mdt log to fsname-MDT0000 (get old
835 name from mdt_UUID) */
837 strscpy(filepnm, (char *)mop->mo_ldd.ldd_uuid, sizeof(filepnm));
838 term = strstr(filepnm, "_UUID");
841 snprintf(cmd, cmdsz, "cp %s/%s/%s %s/%s/%s",
842 mntpt, MDT_LOGS_DIR, filepnm,
843 mntpt, MOUNT_CONFIGS_DIR,
844 mop->mo_ldd.ldd_svname);
845 ret = run_command(cmd, cmdsz);
848 fprintf(stderr, "%s: Can't copy 1.4 config %s/%s "
849 "(%d)\n", progname, MDT_LOGS_DIR, filepnm, ret);
850 fprintf(stderr, "mount -t ext3 %s somewhere, "
851 "find the MDT log for fs %s and "
852 "copy it manually into %s/%s, "
855 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
856 mop->mo_ldd.ldd_svname);
871 int read_local_files(struct mkfs_opts *mop)
873 char tmpdir[] = "/tmp/dirXXXXXX";
879 int cmdsz = sizeof(cmd);
881 /* Make a temporary directory to hold Lustre data files. */
882 if (!mkdtemp(tmpdir)) {
883 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
884 progname, tmpdir, strerror(errno));
888 dev = mop->mo_device;
890 /* TODO: it's worth observing the get_mountdata() function that is
891 in mount_utils.c for getting the mountdata out of the
894 /* Construct debugfs command line. */
895 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
896 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
898 ret = run_command(cmd, cmdsz);
900 verrprint("%s: Unable to dump %s dir (%d)\n",
901 progname, MOUNT_CONFIGS_DIR, ret);
903 sprintf(filepnm, "%s/mountdata", tmpdir);
904 filep = fopen(filepnm, "r");
907 vprint("Reading %s\n", MOUNT_DATA_FILE);
908 num_read = fread(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
909 if (num_read < 1 && ferror(filep)) {
910 fprintf(stderr, "%s: Unable to read from file (%s): %s\n",
911 progname, filepnm, strerror(errno));
916 /* Try to read pre-1.6 config from last_rcvd */
917 struct lr_server_data lsd;
918 verrprint("%s: Unable to read %d.%d config %s.\n",
919 progname, LUSTRE_MAJOR, LUSTRE_MINOR, filepnm);
921 verrprint("Trying 1.4 config from last_rcvd\n");
922 sprintf(filepnm, "%s/%s", tmpdir, LAST_RCVD);
924 /* Construct debugfs command line. */
925 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s' %s",
926 DEBUGFS, LAST_RCVD, filepnm, dev);
928 ret = run_command(cmd, cmdsz);
930 fprintf(stderr, "%s: Unable to dump %s file (%d)\n",
931 progname, LAST_RCVD, ret);
935 filep = fopen(filepnm, "r");
937 fprintf(stderr, "%s: Unable to open %s: %s\n",
938 progname, filepnm, strerror(errno));
940 verrprint("Contents of %s:\n", tmpdir);
942 snprintf(cmd, cmdsz, "ls -l %s/", tmpdir);
943 run_command(cmd, cmdsz);
944 verrprint("Contents of disk:\n");
945 snprintf(cmd, cmdsz, "%s -c -R 'ls -l /' %s",
947 run_command(cmd, cmdsz);
951 vprint("Reading %s\n", LAST_RCVD);
952 ret = fread(&lsd, 1, sizeof(lsd), filep);
953 if (ret < sizeof(lsd)) {
954 fprintf(stderr, "%s: Short read (%d of %d)\n",
955 progname, ret, (int)sizeof(lsd));
960 vprint("Feature compat=%x, incompat=%x\n",
961 lsd.lsd_feature_compat, lsd.lsd_feature_incompat);
963 if ((lsd.lsd_feature_compat & OBD_COMPAT_OST) ||
964 (lsd.lsd_feature_incompat & OBD_INCOMPAT_OST)) {
965 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
966 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
967 } else if ((lsd.lsd_feature_compat & OBD_COMPAT_MDT) ||
968 (lsd.lsd_feature_incompat & OBD_INCOMPAT_MDT)) {
969 /* We must co-locate so mgs can see old logs.
970 If user doesn't want this, they can copy the old
971 logs manually and re-tunefs. */
972 mop->mo_ldd.ldd_flags =
973 LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_MGS;
974 mop->mo_ldd.ldd_svindex = lsd.lsd_mdt_index;
976 /* If neither is set, we're pre-1.4.6, make a guess. */
977 /* Construct debugfs command line. */
978 snprintf(cmd, cmdsz, "%s -c -R 'rdump /%s %s' %s",
979 DEBUGFS, MDT_LOGS_DIR, tmpdir, dev);
980 run_command(cmd, cmdsz);
982 sprintf(filepnm, "%s/%s", tmpdir, MDT_LOGS_DIR);
983 if (lsd.lsd_ost_index > 0) {
984 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
985 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
987 /* If there's a LOGS dir, it's an MDT */
988 if ((ret = access(filepnm, F_OK)) == 0) {
989 mop->mo_ldd.ldd_flags =
992 /* Old MDT's are always index 0
994 mop->mo_ldd.ldd_svindex = 0;
996 /* The index may not be correct */
997 mop->mo_ldd.ldd_flags =
998 LDD_F_SV_TYPE_OST | LDD_F_NEED_INDEX;
999 verrprint("OST with unknown index\n");
1005 memcpy(mop->mo_ldd.ldd_uuid, lsd.lsd_uuid,
1006 sizeof(mop->mo_ldd.ldd_uuid));
1007 mop->mo_ldd.ldd_flags |= LDD_F_UPGRADE14;
1009 /* end COMPAT_146 */
1014 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
1015 run_command(cmd, cmdsz);
1017 verrprint("Failed to read old data (%d)\n", ret);
1022 void set_defaults(struct mkfs_opts *mop)
1024 mop->mo_ldd.ldd_magic = LDD_MAGIC;
1025 mop->mo_ldd.ldd_config_ver = 1;
1026 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
1027 mop->mo_mgs_failnodes = 0;
1028 strcpy(mop->mo_ldd.ldd_fsname, "lustre");
1029 if (get_os_version() == 24)
1030 mop->mo_ldd.ldd_mount_type = LDD_MT_EXT3;
1032 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
1034 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
1035 mop->mo_stripe_count = 1;
1038 static inline void badopt(const char *opt, char *type)
1040 fprintf(stderr, "%s: '--%s' only valid for %s\n",
1041 progname, opt, type);
1045 static int clean_param(char *buf, char *key)
1051 if ((sub = strstr(buf, key)) != NULL) {
1052 if ((next = strchr(sub, ' ')) != NULL) {
1054 memmove(sub, next, strlen(next) + 1);
1062 static int add_param(char *buf, char *key, char *val, int unique)
1064 int end = sizeof(((struct lustre_disk_data *)0)->ldd_params);
1070 keylen = strlen(key);
1073 clean_param(buf, key);
1075 if ((ptr = strchr(val, '=')) == NULL)
1078 clean_param(buf, val);
1083 start = strlen(buf);
1084 if (start + 1 + keylen + strlen(val) >= end) {
1085 fprintf(stderr, "%s: params are too long-\n%s %s%s\n",
1086 progname, buf, key ? key : "", val);
1090 sprintf(buf + start, " %s%s", key ? key : "", val);
1094 /* from mount_lustre */
1095 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
1096 #define MAXNIDSTR 1024
1097 static char *convert_hostnames(char *s1)
1099 char *converted, *s2 = 0, *c, *end, sep;
1100 int left = MAXNIDSTR;
1103 converted = malloc(left);
1104 if (converted == NULL) {
1105 fprintf(stderr, "out of memory: needed %d bytes\n",
1109 end = s1 + strlen(s1);
1111 while ((left > 0) && (s1 < end)) {
1112 s2 = strpbrk(s1, ",:");
1117 nid = libcfs_str2nid(s1);
1119 if (nid == LNET_NID_ANY) {
1120 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
1124 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
1125 strlen("127.0.0.1")) == 0) {
1126 fprintf(stderr, "%s: The NID '%s' resolves to the "
1127 "loopback address '%s'. Lustre requires a "
1128 "non-loopback address.\n",
1129 progname, s1, libcfs_nid2str(nid));
1134 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
1135 left = converted + MAXNIDSTR - c;
1141 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
1144 static struct option long_opt[] = {
1145 {"backfstype", 1, 0, 'b'},
1146 {"stripe-count-hint", 1, 0, 'c'},
1147 {"comment", 1, 0, 'u'},
1148 {"configdev", 1, 0, 'C'},
1149 {"device-size", 1, 0, 'd'},
1150 {"dryrun", 0, 0, 'n'},
1151 {"erase-params", 0, 0, 'e'},
1152 {"failnode", 1, 0, 'f'},
1153 {"failover", 1, 0, 'f'},
1155 {"help", 0, 0, 'h'},
1156 {"index", 1, 0, 'i'},
1157 {"mkfsoptions", 1, 0, 'k'},
1158 {"mgsnode", 1, 0, 'm'},
1159 {"mgsnid", 1, 0, 'm'},
1161 {"fsname",1, 0, 'L'},
1162 {"noformat", 0, 0, 'n'},
1163 {"nomgs", 0, 0, 'N'},
1164 {"mountfsoptions", 1, 0, 'o'},
1166 {"param", 1, 0, 'p'},
1167 {"print", 0, 0, 'n'},
1168 {"quiet", 0, 0, 'q'},
1169 {"reformat", 0, 0, 'r'},
1170 {"verbose", 0, 0, 'v'},
1171 {"writeconf", 0, 0, 'w'},
1174 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqru:vw";
1178 const size_t prefix_len = sizeof(PARAM_MDT_UPCALL) - 1;
1180 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
1185 while (i < LDD_MT_LAST) {
1186 if (strcmp(optarg, mt_str(i)) == 0) {
1187 mop->mo_ldd.ldd_mount_type = i;
1195 if (IS_MDT(&mop->mo_ldd)) {
1196 int stripe_count = atol(optarg);
1197 if (stripe_count <= 0) {
1198 fprintf(stderr, "%s: bad stripe count "
1199 "%d\n", progname, stripe_count);
1202 mop->mo_stripe_count = stripe_count;
1204 badopt(long_opt[longidx].name, "MDT");
1208 case 'C': /* Configdev */
1210 printf("Configdev not implemented\n");
1213 mop->mo_device_sz = atol(optarg);
1216 mop->mo_ldd.ldd_params[0] = '\0';
1217 /* Must update the mgs logs */
1218 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1221 char *nids = convert_hostnames(optarg);
1224 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
1229 /* Must update the mgs logs */
1230 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1235 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
1241 if (!(mop->mo_ldd.ldd_flags &
1242 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1243 LDD_F_WRITECONF))) {
1244 fprintf(stderr, "%s: cannot change the index of"
1245 " a registered target\n", progname);
1248 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
1249 mop->mo_ldd.ldd_svindex = atol(optarg);
1250 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
1252 badopt(long_opt[longidx].name, "MDT,OST");
1257 strscpy(mop->mo_mkfsopts, optarg,
1258 sizeof(mop->mo_mkfsopts));
1262 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
1263 (!(mop->mo_ldd.ldd_flags &
1264 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1265 LDD_F_WRITECONF)))) {
1266 fprintf(stderr, "%s: cannot change the name of"
1267 " a registered target\n", progname);
1270 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
1271 fprintf(stderr, "%s: filesystem name must be "
1272 "1-8 chars\n", progname);
1275 if ((tmp = strpbrk(optarg, "/:"))) {
1276 fprintf(stderr, "%s: char '%c' not allowed in "
1277 "filesystem name\n", progname, *tmp);
1280 strscpy(mop->mo_ldd.ldd_fsname, optarg,
1281 sizeof(mop->mo_ldd.ldd_fsname));
1285 char *nids = convert_hostnames(optarg);
1288 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
1293 mop->mo_mgs_failnodes++;
1297 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
1303 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
1306 *mountopts = optarg;
1309 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
1312 /* Test if the param is valid for mdt.group_upcall */
1313 if (!strncmp(optarg, PARAM_MDT_UPCALL, prefix_len)) {
1315 if (strcmp(optarg + prefix_len, "NONE") &&
1316 access(optarg + prefix_len, R_OK | X_OK))
1317 fprintf(stderr, "WARNING: group upcall "
1318 "parameter not executable: %s\n"
1319 "NOTE: you can change the path "
1320 "to the group upcall through "
1321 "tunefs.lustre(8)\n", optarg +
1324 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg, 0);
1327 /* Must update the mgs logs */
1328 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1334 mop->mo_flags |= MO_FORCEFORMAT;
1337 strscpy(mop->mo_ldd.ldd_userdata, optarg,
1338 sizeof(mop->mo_ldd.ldd_userdata));
1344 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
1349 fprintf(stderr, "Unknown option '%c'\n", opt);
1355 /* Last arg is device */
1356 if (optind != argc - 1) {
1358 fprintf(stderr, "Bad argument: %s\n", argv[optind]);
1362 /* single argument: <device> */
1367 if (mop->mo_ldd.ldd_flags & LDD_F_SV_TYPE_MDT && 0 == upcall) {
1368 if (access("/usr/sbin/l_getgroups", R_OK | X_OK)) {
1369 fprintf(stderr, "WARNING: MDS group upcall is not set, "
1372 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MDT_UPCALL,
1373 "/usr/sbin/l_getgroups", 1);
1376 /* Must update the mgs logs */
1377 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1385 int main(int argc, char *const argv[])
1387 struct mkfs_opts mop;
1388 struct lustre_disk_data *ldd;
1389 char *mountopts = NULL;
1390 char always_mountopts[512] = "";
1391 char default_mountopts[512] = "";
1394 if ((progname = strrchr(argv[0], '/')) != NULL)
1399 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
1404 memset(&mop, 0, sizeof(mop));
1407 /* device is last arg */
1408 strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
1410 /* Are we using a loop device? */
1411 ret = is_block(mop.mo_device);
1415 mop.mo_flags |= MO_IS_LOOP;
1418 /* For tunefs, we must read in the old values before parsing any
1421 /* Check whether the disk has already been formatted by mkfs.lustre */
1422 ret = is_lustre_target(&mop);
1425 fprintf(stderr, "Device %s has not been formatted with "
1426 "mkfs.lustre\n", mop.mo_device);
1431 ret = read_local_files(&mop);
1434 fprintf(stderr, "Failed to read previous Lustre data from %s "
1435 "(%d)\n", mop.mo_device, ret);
1438 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
1439 mop.mo_mgs_failnodes++;
1442 print_ldd("Read previous values", &(mop.mo_ldd));
1445 ret = parse_opts(argc, argv, &mop, &mountopts);
1451 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
1453 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
1458 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
1460 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
1465 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
1466 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
1468 fprintf(stderr, "Can't find the target index, "
1469 "specify with --index\n");
1474 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1477 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
1479 fprintf(stderr, "Must specify --mgsnode=\n");
1484 /* These are the permanent mount options (always included) */
1485 switch (ldd->ldd_mount_type) {
1487 case LDD_MT_LDISKFS:
1488 case LDD_MT_LDISKFS2: {
1489 sprintf(always_mountopts, "errors=remount-ro");
1490 if (IS_MDT(ldd) || IS_MGS(ldd))
1491 strscat(always_mountopts, ",iopen_nopriv,user_xattr",
1492 sizeof(always_mountopts));
1493 if ((get_os_version() == 24) && IS_OST(ldd))
1494 strscat(always_mountopts, ",asyncdel",
1495 sizeof(always_mountopts));
1496 /* NB: Files created while extents are enabled cannot be read
1497 if mounted with a kernel that doesn't include the Lustre ldiskfs
1500 (ldd->ldd_mount_type == LDD_MT_LDISKFS ||
1501 ldd->ldd_mount_type == LDD_MT_LDISKFS2)) {
1502 strscat(default_mountopts, ",extents,mballoc",
1503 sizeof(default_mountopts));
1508 mop.mo_flags |= MO_IS_LOOP;
1509 sprintf(always_mountopts, "type=ext3,dev=%s",
1515 fprintf(stderr, "unknown fs type %d '%s'\n",
1516 ldd->ldd_mount_type,
1524 /* If user specifies mount opts, don't use defaults,
1525 but always use always_mountopts */
1526 sprintf(ldd->ldd_mount_opts, "%s,%s",
1527 always_mountopts, mountopts);
1530 if (ldd->ldd_mount_opts[0] == 0)
1531 /* use the defaults unless old opts exist */
1534 sprintf(ldd->ldd_mount_opts, "%s%s",
1535 always_mountopts, default_mountopts);
1539 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
1540 ldd->ldd_fsname, ldd->ldd_svname);
1543 print_ldd("Permanent disk data", ldd);
1546 printf("exiting before disk write.\n");
1550 if (check_mtab_entry(mop.mo_device))
1553 /* Create the loopback file */
1554 if (mop.mo_flags & MO_IS_LOOP) {
1555 ret = access(mop.mo_device, F_OK);
1558 #ifndef TUNEFS /* mkfs.lustre */
1559 /* Reformat the loopback file */
1560 if (ret || (mop.mo_flags & MO_FORCEFORMAT))
1561 ret = loop_format(&mop);
1564 ret = loop_setup(&mop);
1567 fprintf(stderr, "Loop device setup for %s failed: %s\n",
1568 mop.mo_device, strerror(ret));
1573 #ifndef TUNEFS /* mkfs.lustre */
1574 /* Check whether the disk has already been formatted by mkfs.lustre */
1575 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
1576 ret = is_lustre_target(&mop);
1579 fprintf(stderr, "Device %s was previously formatted "
1580 "for lustre. Use --reformat to reformat it, "
1581 "or tunefs.lustre to modify.\n",
1587 /* Format the backing filesystem */
1588 ret = make_lustre_backfs(&mop);
1591 fprintf(stderr, "mkfs failed %d\n", ret);
1596 /* Write our config files */
1597 ret = write_local_files(&mop);
1600 fprintf(stderr, "failed to write local files\n");
1607 /* Fix any crazy return values from system() */
1608 if (ret && ((ret & 255) == 0))
1611 verrprint("%s: exiting with %d (%s)\n",
1612 progname, ret, strerror(ret));