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 fprintf(stderr, "%s: error %d on losetup: %s\n",
268 progname, ret, strerror(ret));
271 strscpy(mop->mo_loopdev, l_device,
272 sizeof(mop->mo_loopdev));
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 /* Display the need for the latest e2fsprogs to be installed. make_backfs
367 * indicates if the caller is make_lustre_backfs() or not. */
368 static void disp_old_e2fsprogs_msg(const char *feature, int make_backfs)
370 static int msg_displayed;
373 fprintf(stderr, "WARNING: %s does not support %s "
374 "feature.\n\n", E2FSPROGS, feature);
380 fprintf(stderr, "WARNING: The %s package currently installed on "
381 "your system does not support \"%s\" feature.\n",
383 #if !(HAVE_LDISKFSPROGS)
384 fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
385 "http://downloads.lustre.org/public/tools/e2fsprogs/\n"
386 "to enable this feature.\n");
389 fprintf(stderr, "Feature will not be enabled until %s"
390 "is updated and '%s -O %s %%{device}' "
391 "is run.\n\n", E2FSPROGS, TUNE2FS, feature);
394 /* Check whether the file exists in the device */
395 static int file_in_dev(char *file_name, char *dev_name)
398 char debugfs_cmd[256];
399 unsigned int inode_num;
402 /* Construct debugfs command line. */
403 snprintf(debugfs_cmd, sizeof(debugfs_cmd),
404 "%s -c -R 'stat %s' '%s' 2>&1 | egrep '(Inode|unsupported)'",
405 DEBUGFS, file_name, dev_name);
407 fp = popen(debugfs_cmd, "r");
409 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
413 if (fscanf(fp, "Inode: %u", &inode_num) == 1) { /* exist */
417 i = fread(debugfs_cmd, 1, sizeof(debugfs_cmd), fp);
420 fprintf(stderr, "%s", debugfs_cmd);
421 if (strstr(debugfs_cmd, "unsupported feature")) {
422 disp_old_e2fsprogs_msg("an unknown", 0);
431 /* Check whether the device has already been used with lustre */
432 static int is_lustre_target(struct mkfs_opts *mop)
435 vprint("checking for existing Lustre data: ");
437 if ((rc = file_in_dev(MOUNT_DATA_FILE, mop->mo_device))) {
439 (rc == 1) ? MOUNT_DATA_FILE : "extents");
440 /* in the -1 case, 'extents' means this really IS a lustre
445 if ((rc = file_in_dev(LAST_RCVD, mop->mo_device))) {
446 vprint("found %s\n", LAST_RCVD);
450 vprint("not found\n");
451 return 0; /* The device is not a lustre target. */
454 /* Check if a certain feature is supported by e2fsprogs.
455 * Firstly we try to use "debugfs supported_features" command to check if
456 * the feature is supported. If this fails we try to set this feature with
457 * mke2fs to check for its support. */
458 static int is_e2fsprogs_feature_supp(const char *feature)
462 char imgname[] = "/tmp/test-img-XXXXXX";
466 snprintf(cmd, sizeof(cmd), "%s -c -R \"supported_features %s\" 2>&1",
469 /* Using popen() instead of run_command() since debugfs does not return
470 * proper error code if command is not supported */
471 fp = popen(cmd, "r");
473 fprintf(stderr, "%s: %s\n", progname, strerror(errno));
476 ret = fread(cmd, 1, sizeof(cmd), fp);
478 if (strstr(cmd, feature) && !(strstr(cmd, "Unknown")))
482 if ((fd = mkstemp(imgname)) < 0)
485 snprintf(cmd, sizeof(cmd), "%s -F -O %s %s 100 >/dev/null 2>&1",
486 MKE2FS, feature, imgname);
487 /* run_command() displays the output of mke2fs when it fails for
488 * some feature, so use system() directly */
496 static void disp_old_kernel_msg(char *feature)
498 fprintf(stderr, "WARNING: ldiskfs filesystem does not support \"%s\" "
499 "feature.\n\n", feature);
502 static void enable_default_backfs_features(struct mkfs_opts *mop)
505 int maj_high, maj_low, min;
508 strscat(mop->mo_mkfsopts, " -O dir_index", sizeof(mop->mo_mkfsopts));
510 /* Upstream e2fsprogs called our uninit_groups feature uninit_bg,
511 * check for both of them when testing e2fsprogs features. */
512 if (is_e2fsprogs_feature_supp("uninit_groups") == 0)
513 strscat(mop->mo_mkfsopts, ",uninit_groups",
514 sizeof(mop->mo_mkfsopts));
515 else if (is_e2fsprogs_feature_supp("uninit_bg") == 0)
516 strscat(mop->mo_mkfsopts, ",uninit_bg",
517 sizeof(mop->mo_mkfsopts));
519 disp_old_e2fsprogs_msg("uninit_bg", 1);
525 sscanf(uts.release, "%d.%d.%d", &maj_high, &maj_low, &min);
526 printf("%d %d %d\n", maj_high, maj_low, min);
528 /* Multiple mount protection is enabled only if failover node is
529 * specified and if kernel version is higher than 2.6.9 */
531 if (KERNEL_VERSION(maj_high, maj_low, min) >=
532 KERNEL_VERSION(2,6,9)) {
533 if (is_e2fsprogs_feature_supp("mmp") == 0)
534 strscat(mop->mo_mkfsopts, ",mmp",
535 sizeof(mop->mo_mkfsopts));
537 disp_old_e2fsprogs_msg("mmp", 1);
539 disp_old_kernel_msg("mmp");
543 /* Build fs according to type */
544 int make_lustre_backfs(struct mkfs_opts *mop)
546 char mkfs_cmd[PATH_MAX];
552 if (mop->mo_device_sz != 0) {
553 if (mop->mo_device_sz < 8096){
554 fprintf(stderr, "%s: size of filesystem must be larger "
555 "than 8MB, but is set to %lluKB\n",
556 progname, (long long)mop->mo_device_sz);
559 block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
562 if ((mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3) ||
563 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS) ||
564 (mop->mo_ldd.ldd_mount_type == LDD_MT_LDISKFS2)) {
565 __u64 device_sz = mop->mo_device_sz;
567 /* we really need the size */
568 if (device_sz == 0) {
569 device_sz = get_device_size(mop->mo_device);
574 /* Journal size in MB */
575 if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
576 /* Choose our own default journal size */
577 long journal_sz = 0, max_sz;
578 if (device_sz > 1024 * 1024) /* 1GB */
579 journal_sz = (device_sz / 102400) * 4;
581 max_sz = (102400 * L_BLOCK_SIZE) >> 20; /* 400MB */
582 if (journal_sz > max_sz)
585 sprintf(buf, " -J size=%ld", journal_sz);
586 strscat(mop->mo_mkfsopts, buf,
587 sizeof(mop->mo_mkfsopts));
591 /* bytes_per_inode: disk size / num inodes */
592 if (strstr(mop->mo_mkfsopts, "-i") == NULL) {
593 long bytes_per_inode = 0;
595 if (IS_MDT(&mop->mo_ldd))
596 bytes_per_inode = 4096;
598 /* Allocate fewer inodes on large OST devices. Most
599 filesystems can be much more aggressive than even
601 if ((IS_OST(&mop->mo_ldd) && (device_sz > 1000000)))
602 bytes_per_inode = 16384;
604 if (bytes_per_inode > 0) {
605 sprintf(buf, " -i %ld", bytes_per_inode);
606 strscat(mop->mo_mkfsopts, buf,
607 sizeof(mop->mo_mkfsopts));
611 /* Inode size (for extended attributes). The LOV EA size is
612 * 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
613 * and we want some margin above that for ACLs, other EAs... */
614 if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
616 if (IS_MDT(&mop->mo_ldd)) {
617 if (mop->mo_stripe_count > 72)
618 inode_size = 512; /* bz 7241 */
619 /* cray stripes across all osts (>60) */
620 else if (mop->mo_stripe_count > 32)
622 else if (mop->mo_stripe_count > 10)
626 } else if (IS_OST(&mop->mo_ldd)) {
627 /* now as we store fids in EA on OST we need
628 to make inode bigger */
632 if (inode_size > 0) {
633 sprintf(buf, " -I %ld", inode_size);
634 strscat(mop->mo_mkfsopts, buf,
635 sizeof(mop->mo_mkfsopts));
640 strscat(mop->mo_mkfsopts, " -q",
641 sizeof(mop->mo_mkfsopts));
644 if (strstr(mop->mo_mkfsopts, "-O") == NULL)
645 enable_default_backfs_features(mop);
647 /* Allow reformat of full devices (as opposed to
648 partitions.) We already checked for mounted dev. */
649 strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
651 snprintf(mkfs_cmd, sizeof(mkfs_cmd),
652 "%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
653 mop->mo_ldd.ldd_svname);
655 } else if (mop->mo_ldd.ldd_mount_type == LDD_MT_REISERFS) {
656 long journal_sz = 0; /* FIXME default journal size */
657 if (journal_sz > 0) {
658 sprintf(buf, " --journal_size %ld", journal_sz);
659 strscat(mop->mo_mkfsopts, buf,
660 sizeof(mop->mo_mkfsopts));
662 snprintf(mkfs_cmd, sizeof(mkfs_cmd), "mkreiserfs -ff ");
665 fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
666 progname, mop->mo_ldd.ldd_mount_type,
667 MT_STR(&mop->mo_ldd));
671 /* For loop device format the dev, not the filename */
672 dev = mop->mo_device;
673 if (mop->mo_flags & MO_IS_LOOP)
674 dev = mop->mo_loopdev;
676 vprint("formatting backing filesystem %s on %s\n",
677 MT_STR(&mop->mo_ldd), dev);
678 vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
679 vprint("\t4k blocks %d\n", block_count);
680 vprint("\toptions %s\n", mop->mo_mkfsopts);
682 /* mkfs_cmd's trailing space is important! */
683 strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
684 strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
685 strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
686 if (block_count != 0) {
687 sprintf(buf, " %d", block_count);
688 strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
691 vprint("mkfs_cmd = %s\n", mkfs_cmd);
692 ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
695 fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
700 /* ==================== Lustre config functions =============*/
702 void print_ldd(char *str, struct lustre_disk_data *ldd)
704 printf("\n %s:\n", str);
705 printf("Target: %s\n", ldd->ldd_svname);
706 if (ldd->ldd_svindex == INDEX_UNASSIGNED)
707 printf("Index: unassigned\n");
709 printf("Index: %d\n", ldd->ldd_svindex);
710 if (ldd->ldd_uuid[0])
711 printf("UUID: %s\n", (char *)ldd->ldd_uuid);
712 printf("Lustre FS: %s\n", ldd->ldd_fsname);
713 printf("Mount type: %s\n", MT_STR(ldd));
714 printf("Flags: %#x\n", ldd->ldd_flags);
715 printf(" (%s%s%s%s%s%s%s%s)\n",
716 IS_MDT(ldd) ? "MDT ":"",
717 IS_OST(ldd) ? "OST ":"",
718 IS_MGS(ldd) ? "MGS ":"",
719 ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
720 ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
721 ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
722 ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
723 ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
724 printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
725 printf("Parameters:%s\n", ldd->ldd_params);
726 if (ldd->ldd_userdata[0])
727 printf("Comment: %s\n", ldd->ldd_userdata);
731 /* Write the server config files */
732 int write_local_files(struct mkfs_opts *mop)
734 char mntpt[] = "/tmp/mntXXXXXX";
740 /* Mount this device temporarily in order to write these files */
741 if (!mkdtemp(mntpt)) {
742 fprintf(stderr, "%s: Can't create temp mount point %s: %s\n",
743 progname, mntpt, strerror(errno));
747 dev = mop->mo_device;
748 if (mop->mo_flags & MO_IS_LOOP)
749 dev = mop->mo_loopdev;
751 ret = mount(dev, mntpt, MT_STR(&mop->mo_ldd), 0, NULL);
753 fprintf(stderr, "%s: Unable to mount %s: %s\n",
754 progname, dev, strerror(errno));
756 if (errno == ENODEV) {
757 fprintf(stderr, "Is the %s module available?\n",
758 MT_STR(&mop->mo_ldd));
763 /* Set up initial directories */
764 sprintf(filepnm, "%s/%s", mntpt, MOUNT_CONFIGS_DIR);
765 ret = mkdir(filepnm, 0777);
766 if ((ret != 0) && (errno != EEXIST)) {
767 fprintf(stderr, "%s: Can't make configs dir %s: %s\n",
768 progname, filepnm, strerror(errno));
770 } else if (errno == EEXIST) {
774 /* Save the persistent mount data into a file. Lustre must pre-read
775 this file to get the real mount options. */
776 vprint("Writing %s\n", MOUNT_DATA_FILE);
777 sprintf(filepnm, "%s/%s", mntpt, MOUNT_DATA_FILE);
778 filep = fopen(filepnm, "w");
780 fprintf(stderr, "%s: Unable to create %s file: %s\n",
781 progname, filepnm, strerror(errno));
784 fwrite(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
789 /* Check for upgrade */
790 if ((mop->mo_ldd.ldd_flags & (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS))
791 == (LDD_F_UPGRADE14 | LDD_F_SV_TYPE_MGS)) {
794 int cmdsz = sizeof(cmd);
795 vprint("Copying old logs\n");
797 /* Copy the old client log to fsname-client */
798 sprintf(filepnm, "%s/%s/%s-client",
799 mntpt, MOUNT_CONFIGS_DIR, mop->mo_ldd.ldd_fsname);
800 snprintf(cmd, cmdsz, "cp %s/%s/client %s", mntpt, MDT_LOGS_DIR,
802 ret = run_command(cmd, cmdsz);
804 fprintf(stderr, "%s: Can't copy 1.4 config %s/client "
805 "(%d)\n", progname, MDT_LOGS_DIR, ret);
806 fprintf(stderr, "mount -t ldiskfs %s somewhere, "
807 "find the client log for fs %s and "
808 "copy it manually into %s/%s-client, "
811 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
812 mop->mo_ldd.ldd_fsname);
816 /* We need to use the old mdt log because otherwise mdt won't
817 have complete lov if old clients connect before all
819 /* Copy the old mdt log to fsname-MDT0000 (get old
820 name from mdt_UUID) */
822 strscpy(filepnm, (char *)mop->mo_ldd.ldd_uuid, sizeof(filepnm));
823 term = strstr(filepnm, "_UUID");
826 snprintf(cmd, cmdsz, "cp %s/%s/%s %s/%s/%s",
827 mntpt, MDT_LOGS_DIR, filepnm,
828 mntpt, MOUNT_CONFIGS_DIR,
829 mop->mo_ldd.ldd_svname);
830 ret = run_command(cmd, cmdsz);
833 fprintf(stderr, "%s: Can't copy 1.4 config %s/%s "
834 "(%d)\n", progname, MDT_LOGS_DIR, filepnm, ret);
835 fprintf(stderr, "mount -t ext3 %s somewhere, "
836 "find the MDT log for fs %s and "
837 "copy it manually into %s/%s, "
840 mop->mo_ldd.ldd_fsname, MOUNT_CONFIGS_DIR,
841 mop->mo_ldd.ldd_svname);
856 int read_local_files(struct mkfs_opts *mop)
858 char tmpdir[] = "/tmp/dirXXXXXX";
864 int cmdsz = sizeof(cmd);
866 /* Make a temporary directory to hold Lustre data files. */
867 if (!mkdtemp(tmpdir)) {
868 fprintf(stderr, "%s: Can't create temporary directory %s: %s\n",
869 progname, tmpdir, strerror(errno));
873 dev = mop->mo_device;
875 /* TODO: it's worth observing the get_mountdata() function that is
876 in mount_utils.c for getting the mountdata out of the
879 /* Construct debugfs command line. */
880 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s/mountdata' '%s'",
881 DEBUGFS, MOUNT_DATA_FILE, tmpdir, dev);
883 ret = run_command(cmd, cmdsz);
885 verrprint("%s: Unable to dump %s dir (%d)\n",
886 progname, MOUNT_CONFIGS_DIR, ret);
888 sprintf(filepnm, "%s/mountdata", tmpdir);
889 filep = fopen(filepnm, "r");
891 vprint("Reading %s\n", MOUNT_DATA_FILE);
892 fread(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
895 /* Try to read pre-1.6 config from last_rcvd */
896 struct lr_server_data lsd;
897 verrprint("%s: Unable to read %d.%d config %s.\n",
898 progname, LUSTRE_MAJOR, LUSTRE_MINOR, filepnm);
900 verrprint("Trying 1.4 config from last_rcvd\n");
901 sprintf(filepnm, "%s/%s", tmpdir, LAST_RCVD);
903 /* Construct debugfs command line. */
904 snprintf(cmd, cmdsz, "%s -c -R 'dump /%s %s' %s",
905 DEBUGFS, LAST_RCVD, filepnm, dev);
907 ret = run_command(cmd, cmdsz);
909 fprintf(stderr, "%s: Unable to dump %s file (%d)\n",
910 progname, LAST_RCVD, ret);
914 filep = fopen(filepnm, "r");
916 fprintf(stderr, "%s: Unable to open %s: %s\n",
917 progname, filepnm, strerror(errno));
919 verrprint("Contents of %s:\n", tmpdir);
921 snprintf(cmd, cmdsz, "ls -l %s/", tmpdir);
922 run_command(cmd, cmdsz);
923 verrprint("Contents of disk:\n");
924 snprintf(cmd, cmdsz, "%s -c -R 'ls -l /' %s",
926 run_command(cmd, cmdsz);
930 vprint("Reading %s\n", LAST_RCVD);
931 ret = fread(&lsd, 1, sizeof(lsd), filep);
932 if (ret < sizeof(lsd)) {
933 fprintf(stderr, "%s: Short read (%d of %d)\n",
934 progname, ret, (int)sizeof(lsd));
939 vprint("Feature compat=%x, incompat=%x\n",
940 lsd.lsd_feature_compat, lsd.lsd_feature_incompat);
942 if ((lsd.lsd_feature_compat & OBD_COMPAT_OST) ||
943 (lsd.lsd_feature_incompat & OBD_INCOMPAT_OST)) {
944 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
945 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
946 } else if ((lsd.lsd_feature_compat & OBD_COMPAT_MDT) ||
947 (lsd.lsd_feature_incompat & OBD_INCOMPAT_MDT)) {
948 /* We must co-locate so mgs can see old logs.
949 If user doesn't want this, they can copy the old
950 logs manually and re-tunefs. */
951 mop->mo_ldd.ldd_flags =
952 LDD_F_SV_TYPE_MDT | LDD_F_SV_TYPE_MGS;
953 mop->mo_ldd.ldd_svindex = lsd.lsd_mdt_index;
955 /* If neither is set, we're pre-1.4.6, make a guess. */
956 /* Construct debugfs command line. */
957 snprintf(cmd, cmdsz, "%s -c -R 'rdump /%s %s' %s",
958 DEBUGFS, MDT_LOGS_DIR, tmpdir, dev);
959 run_command(cmd, cmdsz);
961 sprintf(filepnm, "%s/%s", tmpdir, MDT_LOGS_DIR);
962 if (lsd.lsd_ost_index > 0) {
963 mop->mo_ldd.ldd_flags = LDD_F_SV_TYPE_OST;
964 mop->mo_ldd.ldd_svindex = lsd.lsd_ost_index;
966 /* If there's a LOGS dir, it's an MDT */
967 if ((ret = access(filepnm, F_OK)) == 0) {
968 mop->mo_ldd.ldd_flags =
971 /* Old MDT's are always index 0
973 mop->mo_ldd.ldd_svindex = 0;
975 /* The index may not be correct */
976 mop->mo_ldd.ldd_flags =
977 LDD_F_SV_TYPE_OST | LDD_F_NEED_INDEX;
978 verrprint("OST with unknown index\n");
984 memcpy(mop->mo_ldd.ldd_uuid, lsd.lsd_uuid,
985 sizeof(mop->mo_ldd.ldd_uuid));
986 mop->mo_ldd.ldd_flags |= LDD_F_UPGRADE14;
993 snprintf(cmd, cmdsz, "rm -rf %s", tmpdir);
994 run_command(cmd, cmdsz);
996 verrprint("Failed to read old data (%d)\n", ret);
1001 void set_defaults(struct mkfs_opts *mop)
1003 mop->mo_ldd.ldd_magic = LDD_MAGIC;
1004 mop->mo_ldd.ldd_config_ver = 1;
1005 mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
1006 mop->mo_mgs_failnodes = 0;
1007 strcpy(mop->mo_ldd.ldd_fsname, "lustre");
1008 if (get_os_version() == 24)
1009 mop->mo_ldd.ldd_mount_type = LDD_MT_EXT3;
1011 mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
1013 mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
1014 mop->mo_stripe_count = 1;
1017 static inline void badopt(const char *opt, char *type)
1019 fprintf(stderr, "%s: '--%s' only valid for %s\n",
1020 progname, opt, type);
1024 static int clean_param(char *buf, char *key)
1030 if ((sub = strstr(buf, key)) != NULL) {
1031 if ((next = strchr(sub, ' ')) != NULL) {
1033 memmove(sub, next, strlen(next) + 1);
1041 static int add_param(char *buf, char *key, char *val, int unique)
1043 int end = sizeof(((struct lustre_disk_data *)0)->ldd_params);
1049 keylen = strlen(key);
1052 clean_param(buf, key);
1054 if ((ptr = strchr(val, '=')) == NULL)
1057 clean_param(buf, val);
1062 start = strlen(buf);
1063 if (start + 1 + keylen + strlen(val) >= end) {
1064 fprintf(stderr, "%s: params are too long-\n%s %s%s\n",
1065 progname, buf, key ? key : "", val);
1069 sprintf(buf + start, " %s%s", key ? key : "", val);
1073 /* from mount_lustre */
1074 /* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
1075 #define MAXNIDSTR 1024
1076 static char *convert_hostnames(char *s1)
1078 char *converted, *s2 = 0, *c, *end, sep;
1079 int left = MAXNIDSTR;
1082 converted = malloc(left);
1083 end = s1 + strlen(s1);
1085 while ((left > 0) && (s1 < end)) {
1086 s2 = strpbrk(s1, ",:");
1091 nid = libcfs_str2nid(s1);
1093 if (nid == LNET_NID_ANY) {
1094 fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
1098 if (strncmp(libcfs_nid2str(nid), "127.0.0.1",
1099 strlen("127.0.0.1")) == 0) {
1100 fprintf(stderr, "%s: The NID '%s' resolves to the "
1101 "loopback address '%s'. Lustre requires a "
1102 "non-loopback address.\n",
1103 progname, s1, libcfs_nid2str(nid));
1108 c += snprintf(c, left, "%s%c", libcfs_nid2str(nid), sep);
1109 left = converted + MAXNIDSTR - c;
1115 int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
1118 static struct option long_opt[] = {
1119 {"backfstype", 1, 0, 'b'},
1120 {"stripe-count-hint", 1, 0, 'c'},
1121 {"comment", 1, 0, 'u'},
1122 {"configdev", 1, 0, 'C'},
1123 {"device-size", 1, 0, 'd'},
1124 {"dryrun", 0, 0, 'n'},
1125 {"erase-params", 0, 0, 'e'},
1126 {"failnode", 1, 0, 'f'},
1127 {"failover", 1, 0, 'f'},
1129 {"help", 0, 0, 'h'},
1130 {"index", 1, 0, 'i'},
1131 {"mkfsoptions", 1, 0, 'k'},
1132 {"mgsnode", 1, 0, 'm'},
1133 {"mgsnid", 1, 0, 'm'},
1135 {"fsname",1, 0, 'L'},
1136 {"noformat", 0, 0, 'n'},
1137 {"nomgs", 0, 0, 'N'},
1138 {"mountfsoptions", 1, 0, 'o'},
1140 {"param", 1, 0, 'p'},
1141 {"print", 0, 0, 'n'},
1142 {"quiet", 0, 0, 'q'},
1143 {"reformat", 0, 0, 'r'},
1144 {"verbose", 0, 0, 'v'},
1145 {"writeconf", 0, 0, 'w'},
1148 char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqru:vw";
1152 const size_t prefix_len = sizeof(PARAM_MDT_UPCALL) - 1;
1154 while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
1159 while (i < LDD_MT_LAST) {
1160 if (strcmp(optarg, mt_str(i)) == 0) {
1161 mop->mo_ldd.ldd_mount_type = i;
1169 if (IS_MDT(&mop->mo_ldd)) {
1170 int stripe_count = atol(optarg);
1171 if (stripe_count <= 0) {
1172 fprintf(stderr, "%s: bad stripe count "
1173 "%d\n", progname, stripe_count);
1176 mop->mo_stripe_count = stripe_count;
1178 badopt(long_opt[longidx].name, "MDT");
1182 case 'C': /* Configdev */
1184 printf("Configdev not implemented\n");
1187 mop->mo_device_sz = atol(optarg);
1190 mop->mo_ldd.ldd_params[0] = '\0';
1191 /* Must update the mgs logs */
1192 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1195 char *nids = convert_hostnames(optarg);
1198 rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
1203 /* Must update the mgs logs */
1204 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1209 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
1215 if (!(mop->mo_ldd.ldd_flags &
1216 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1217 LDD_F_WRITECONF))) {
1218 fprintf(stderr, "%s: cannot change the index of"
1219 " a registered target\n", progname);
1222 if (IS_MDT(&mop->mo_ldd) || IS_OST(&mop->mo_ldd)) {
1223 mop->mo_ldd.ldd_svindex = atol(optarg);
1224 mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
1226 badopt(long_opt[longidx].name, "MDT,OST");
1231 strscpy(mop->mo_mkfsopts, optarg,
1232 sizeof(mop->mo_mkfsopts));
1236 if (!(mop->mo_flags & MO_FORCEFORMAT) &&
1237 (!(mop->mo_ldd.ldd_flags &
1238 (LDD_F_UPGRADE14 | LDD_F_VIRGIN |
1239 LDD_F_WRITECONF)))) {
1240 fprintf(stderr, "%s: cannot change the name of"
1241 " a registered target\n", progname);
1244 if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
1245 fprintf(stderr, "%s: filesystem name must be "
1246 "1-8 chars\n", progname);
1249 if ((tmp = strpbrk(optarg, "/:"))) {
1250 fprintf(stderr, "%s: char '%c' not allowed in "
1251 "filesystem name\n", progname, *tmp);
1254 strscpy(mop->mo_ldd.ldd_fsname, optarg,
1255 sizeof(mop->mo_ldd.ldd_fsname));
1259 char *nids = convert_hostnames(optarg);
1262 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
1267 mop->mo_mgs_failnodes++;
1271 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
1277 mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
1280 *mountopts = optarg;
1283 mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
1286 /* Test if the param is valid for mdt.group_upcall */
1287 if (!strncmp(optarg, PARAM_MDT_UPCALL, prefix_len)) {
1289 if (strcmp(optarg + prefix_len, "NONE") &&
1290 access(optarg + prefix_len, R_OK | X_OK))
1291 fprintf(stderr, "WARNING: group upcall "
1292 "parameter not executable: %s\n"
1293 "NOTE: you can change the path "
1294 "to the group upcall through "
1295 "tunefs.lustre(8)\n", optarg +
1298 rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg, 0);
1301 /* Must update the mgs logs */
1302 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1308 mop->mo_flags |= MO_FORCEFORMAT;
1311 strscpy(mop->mo_ldd.ldd_userdata, optarg,
1312 sizeof(mop->mo_ldd.ldd_userdata));
1318 mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
1323 fprintf(stderr, "Unknown option '%c'\n", opt);
1329 /* Last arg is device */
1330 if (optind != argc - 1) {
1332 fprintf(stderr, "Bad argument: %s\n", argv[optind]);
1337 if (mop->mo_ldd.ldd_flags & LDD_F_SV_TYPE_MDT && 0 == upcall) {
1338 if (access("/usr/sbin/l_getgroups", R_OK | X_OK)) {
1339 fprintf(stderr, "WARNING: MDS group upcall is not set, "
1342 rc = add_param(mop->mo_ldd.ldd_params, PARAM_MDT_UPCALL,
1343 "/usr/sbin/l_getgroups", 1);
1346 /* Must update the mgs logs */
1347 mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
1355 int main(int argc, char *const argv[])
1357 struct mkfs_opts mop;
1358 struct lustre_disk_data *ldd;
1359 char *mountopts = NULL;
1360 char always_mountopts[512] = "";
1361 char default_mountopts[512] = "";
1364 if ((progname = strrchr(argv[0], '/')) != NULL)
1369 if ((argc < 2) || (argv[argc - 1][0] == '-')) {
1374 memset(&mop, 0, sizeof(mop));
1377 /* device is last arg */
1378 strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
1380 /* Are we using a loop device? */
1381 ret = is_block(mop.mo_device);
1385 mop.mo_flags |= MO_IS_LOOP;
1388 /* For tunefs, we must read in the old values before parsing any
1391 /* Check whether the disk has already been formatted by mkfs.lustre */
1392 ret = is_lustre_target(&mop);
1395 fprintf(stderr, "Device %s has not been formatted with "
1396 "mkfs.lustre\n", mop.mo_device);
1401 ret = read_local_files(&mop);
1404 fprintf(stderr, "Failed to read previous Lustre data from %s "
1405 "(%d)\n", mop.mo_device, ret);
1408 if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
1409 mop.mo_mgs_failnodes++;
1412 print_ldd("Read previous values", &(mop.mo_ldd));
1415 ret = parse_opts(argc, argv, &mop, &mountopts);
1421 if (!(IS_MDT(ldd) || IS_OST(ldd) || IS_MGS(ldd))) {
1423 fprintf(stderr, "must set target type: MDT,OST,MGS\n");
1428 if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
1430 fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
1435 if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
1436 (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
1438 fprintf(stderr, "Can't find the target index, "
1439 "specify with --index\n");
1444 if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
1447 fprintf(stderr, "Must specify --mgs or --mgsnode=\n");
1449 fprintf(stderr, "Must specify --mgsnode=\n");
1454 /* These are the permanent mount options (always included) */
1455 switch (ldd->ldd_mount_type) {
1457 case LDD_MT_LDISKFS:
1458 case LDD_MT_LDISKFS2: {
1459 sprintf(always_mountopts, "errors=remount-ro");
1460 if (IS_MDT(ldd) || IS_MGS(ldd))
1461 strscat(always_mountopts, ",iopen_nopriv,user_xattr",
1462 sizeof(always_mountopts));
1463 if ((get_os_version() == 24) && IS_OST(ldd))
1464 strscat(always_mountopts, ",asyncdel",
1465 sizeof(always_mountopts));
1466 /* NB: Files created while extents are enabled cannot be read
1467 if mounted with a kernel that doesn't include the Lustre ldiskfs
1470 (ldd->ldd_mount_type == LDD_MT_LDISKFS ||
1471 ldd->ldd_mount_type == LDD_MT_LDISKFS2)) {
1472 strscat(default_mountopts, ",extents,mballoc",
1473 sizeof(default_mountopts));
1478 mop.mo_flags |= MO_IS_LOOP;
1479 sprintf(always_mountopts, "type=ext3,dev=%s",
1485 fprintf(stderr, "unknown fs type %d '%s'\n",
1486 ldd->ldd_mount_type,
1494 /* If user specifies mount opts, don't use defaults,
1495 but always use always_mountopts */
1496 sprintf(ldd->ldd_mount_opts, "%s,%s",
1497 always_mountopts, mountopts);
1500 if (ldd->ldd_mount_opts[0] == 0)
1501 /* use the defaults unless old opts exist */
1504 sprintf(ldd->ldd_mount_opts, "%s%s",
1505 always_mountopts, default_mountopts);
1509 server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
1510 ldd->ldd_fsname, ldd->ldd_svname);
1513 print_ldd("Permanent disk data", ldd);
1516 printf("exiting before disk write.\n");
1520 if (check_mtab_entry(mop.mo_device))
1523 /* Create the loopback file */
1524 if (mop.mo_flags & MO_IS_LOOP) {
1525 ret = access(mop.mo_device, F_OK);
1528 #ifndef TUNEFS /* mkfs.lustre */
1529 /* Reformat the loopback file */
1530 if (ret || (mop.mo_flags & MO_FORCEFORMAT))
1531 ret = loop_format(&mop);
1534 ret = loop_setup(&mop);
1537 fprintf(stderr, "Loop device setup for %s failed: %s\n",
1538 mop.mo_device, strerror(ret));
1543 #ifndef TUNEFS /* mkfs.lustre */
1544 /* Check whether the disk has already been formatted by mkfs.lustre */
1545 if (!(mop.mo_flags & MO_FORCEFORMAT)) {
1546 ret = is_lustre_target(&mop);
1549 fprintf(stderr, "Device %s was previously formatted "
1550 "for lustre. Use --reformat to reformat it, "
1551 "or tunefs.lustre to modify.\n",
1557 /* Format the backing filesystem */
1558 ret = make_lustre_backfs(&mop);
1561 fprintf(stderr, "mkfs failed %d\n", ret);
1566 /* Write our config files */
1567 ret = write_local_files(&mop);
1570 fprintf(stderr, "failed to write local files\n");
1577 /* Fix any crazy return values from system() */
1578 if (ret && ((ret & 255) == 0))
1581 verrprint("%s: exiting with %d (%s)\n",
1582 progname, ret, strerror(ret));