void print_ldd(char *str, struct lustre_disk_data *ldd)
{
printf("\n %s:\n", str);
printf("Target: %s\n", ldd->ldd_svname);
if (ldd->ldd_svindex == INDEX_UNASSIGNED)
printf("Index: unassigned\n");
else
printf("Index: %d\n", ldd->ldd_svindex);
if (ldd->ldd_uuid[0])
printf("UUID: %s\n", (char *)ldd->ldd_uuid);
printf("Lustre FS: %s\n", ldd->ldd_fsname);
printf("Mount type: %s\n", MT_STR(ldd));
printf("Flags: %#x\n", ldd->ldd_flags);
printf(" (%s%s%s%s%s%s%s%s%s)\n",
IS_MDT(ldd) ? "MDT ":"",
IS_OST(ldd) ? "OST ":"",
IS_MGS(ldd) ? "MGS ":"",
ldd->ldd_flags & LDD_F_NEED_INDEX ? "needs_index ":"",
ldd->ldd_flags & LDD_F_VIRGIN ? "first_time ":"",
ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
ldd->ldd_flags & LDD_F_NO_PRIMNODE? "no_primnode ":"",
ldd->ldd_flags & LDD_F_UPGRADE14 ? "upgrade1.4 ":"");
printf("Persistent mount opts: %s\n", ldd->ldd_mount_opts);
printf("Parameters:%s\n", ldd->ldd_params);
if (ldd->ldd_userdata[0])
printf("Comment: %s\n", ldd->ldd_userdata);
printf("\n");
}
void register_service_tags(char *usource, char *source, char *target)
{
struct lustre_disk_data mo_ldd;
struct utsname utsname_buf;
struct stat stat_buf;
char stclient_loc[] = "/opt/sun/servicetag/bin/stclient";
int rc;
rc = stat(stclient_loc, &stat_buf);
if (rc) {
if (errno != ENOENT && verbose) {
fprintf(stderr,
"%s: trying to stat stclient failed: %s\n",
progname, strerror(errno));
}
return;
}
/* call service tags stclient to show Lustre is in use on this system */
rc = uname(&utsname_buf);
if (rc) {
if (verbose)
fprintf(stderr,
"%s: trying to get uname failed: %s, "
"inventory tags will not be created\n",
progname, strerror(errno));
return;
}
/* client or server? */
if (strchr(usource, ':')) {
stclient("Client", utsname_buf.machine);
} else {
/* first figure what type of device it is */
rc = get_mountdata(source, &mo_ldd);
if (rc) {
if (verbose)
fprintf(stderr,
"%s: trying to read mountdata from %s "
"failed: %s, inventory tags will not "
"be created\n",
progname, target, strerror(errno));
return;
}
if (IS_MDT(&mo_ldd))
stclient("MDS", utsname_buf.machine);
if (IS_MGS(&mo_ldd))
stclient("MGS", utsname_buf.machine);
if (IS_OST(&mo_ldd))
stclient("OSS", utsname_buf.machine);
}
}
int ldiskfs_prepare_lustre(struct mkfs_opts *mop,
char *wanted_mountopts, size_t len)
{
struct lustre_disk_data *ldd = &mop->mo_ldd;
int ret;
/* Set MO_IS_LOOP to indicate a loopback device is needed */
ret = is_block(mop->mo_device);
if (ret < 0) {
return errno;
} else if (ret == 0) {
mop->mo_flags |= MO_IS_LOOP;
}
if (IS_MDT(ldd) || IS_MGS(ldd))
strscat(wanted_mountopts, ",user_xattr", len);
return 0;
}
static int parse_ldd(char *source, struct mount_opts *mop, char *options)
{
struct lustre_disk_data *ldd = &mop->mo_ldd;
char *cur, *start;
int rc;
rc = osd_is_lustre(source, &ldd->ldd_mount_type);
if (rc == 0) {
fprintf(stderr, "%s: %s has not been formatted with mkfs.lustre"
" or the backend filesystem type is not supported by "
"this tool\n", progname, source);
return ENODEV;
}
rc = osd_read_ldd(source, ldd);
if (rc) {
fprintf(stderr, "%s: %s failed to read permanent mount"
" data: %s\n", progname, source,
rc >= 0 ? strerror(rc) : "");
return rc;
}
if ((IS_MDT(ldd) || IS_OST(ldd)) &&
(ldd->ldd_flags & LDD_F_NEED_INDEX)) {
fprintf(stderr, "%s: %s has no index assigned "
"(probably formatted with old mkfs)\n",
progname, source);
return EINVAL;
}
if (ldd->ldd_flags & LDD_F_UPGRADE14) {
fprintf(stderr, "%s: we cannot upgrade %s from this (very old) "
"Lustre version\n", progname, source);
return EINVAL;
}
if (ldd->ldd_flags & LDD_F_UPDATE)
clear_update_ondisk(source, ldd);
/* Since we never rewrite ldd, ignore temp flags */
ldd->ldd_flags &= ~(LDD_F_VIRGIN | LDD_F_WRITECONF);
/* svname of the form lustre:OST1234 means never registered */
rc = strlen(ldd->ldd_svname);
if (strcmp(ldd->ldd_svname, "MGS") != 0) {
if (rc < 8) {
fprintf(stderr, "%s: invalid name '%s'\n",
progname, ldd->ldd_svname);
return EINVAL;
} else if (ldd->ldd_svname[rc - 8] == ':') {
ldd->ldd_svname[rc - 8] = '-';
ldd->ldd_flags |= LDD_F_VIRGIN;
} else if (ldd->ldd_svname[rc - 8] == '=') {
ldd->ldd_svname[rc - 8] = '-';
ldd->ldd_flags |= LDD_F_WRITECONF;
}
}
/* backend osd type */
append_option(options, "osd=");
strcat(options, mt_type(ldd->ldd_mount_type));
append_option(options, ldd->ldd_mount_opts);
if (!mop->mo_have_mgsnid) {
/* Only use disk data if mount -o mgsnode=nid wasn't
* specified */
if (ldd->ldd_flags & LDD_F_SV_TYPE_MGS) {
append_option(options, "mgs");
mop->mo_have_mgsnid++;
} else {
add_mgsnids(mop, options, ldd->ldd_params);
}
}
/* Better have an mgsnid by now */
if (!mop->mo_have_mgsnid) {
fprintf(stderr, "%s: missing option mgsnode=<nid>\n",
progname);
return EINVAL;
}
if (ldd->ldd_flags & LDD_F_VIRGIN)
append_option(options, "virgin");
if (ldd->ldd_flags & LDD_F_UPDATE)
append_option(options, "update");
if (ldd->ldd_flags & LDD_F_WRITECONF)
append_option(options, "writeconf");
if (ldd->ldd_flags & LDD_F_NO_PRIMNODE)
append_option(options, "noprimnode");
/* prefix every lustre parameter with param= so that in-kernel
* mount can recognize them properly and send to MGS at registration */
start = ldd->ldd_params;
while (start && *start != '\0') {
while (*start == ' ') start++;
if (*start == '\0')
break;
cur = start;
start = strchr(cur, ' ');
if (start) {
*start = '\0';
start++;
}
append_option(options, "param=");
strcat(options, cur);
}
/* svname must be last option */
append_option(options, "svname=");
strcat(options, ldd->ldd_svname);
return 0;
}
/* Build fs according to type */
int ldiskfs_make_lustre(struct mkfs_opts *mop)
{
char mkfs_cmd[PATH_MAX];
char buf[64];
char *start;
char *dev;
int ret = 0, ext_opts = 0;
bool enable_64bit = false;
long inode_size = 0;
size_t maxbuflen;
mop->mo_blocksize_kb = 4;
start = strstr(mop->mo_mkfsopts, "-b");
if (start) {
char *end = NULL;
long blocksize;
blocksize = strtol(start + 2, &end, 0);
if (end && (*end == 'k' || *end == 'K'))
blocksize *= 1024;
/* EXT4_MIN_BLOCK_SIZE || EXT4_MAX_BLOCK_SIZE */
if (blocksize < 1024 || blocksize > 65536) {
fprintf(stderr,
"%s: blocksize %lu not in 1024-65536 bytes, normally 4096 bytes\n",
progname, blocksize);
return EINVAL;
}
if ((blocksize & (blocksize - 1)) != 0) {
fprintf(stderr,
"%s: blocksize %lu not a power-of-two value\n",
progname, blocksize);
return EINVAL;
}
mop->mo_blocksize_kb = blocksize >> 10;
}
if (!(mop->mo_flags & MO_IS_LOOP)) {
__u64 device_kb = get_device_size(mop->mo_device);
if (device_kb == 0)
return ENODEV;
/* Compare to real size */
if (mop->mo_device_kb == 0 || device_kb < mop->mo_device_kb)
mop->mo_device_kb = device_kb;
}
if (mop->mo_device_kb != 0) {
__u64 block_count;
if (mop->mo_device_kb < 32384) {
fprintf(stderr, "%s: size of filesystem must be larger "
"than 32MB, but is set to %lldKB\n",
progname, (long long)mop->mo_device_kb);
return EINVAL;
}
block_count = mop->mo_device_kb / mop->mo_blocksize_kb;
if (block_count > 0xffffffffULL) {
/* If the LUN size is just over 2^32 blocks, limit the
* filesystem size to 2^32-1 blocks to avoid problems
* with ldiskfs/mkfs not handling this well. b=22906
*/
if (block_count < 0x100002000ULL)
mop->mo_device_kb =
0xffffffffULL * mop->mo_blocksize_kb;
else
enable_64bit = true;
}
}
if ((mop->mo_ldd.ldd_mount_type != LDD_MT_EXT3) &&
(mop->mo_ldd.ldd_mount_type != LDD_MT_LDISKFS) &&
(mop->mo_ldd.ldd_mount_type != LDD_MT_LDISKFS2)) {
fprintf(stderr, "%s: unsupported fs type: %d (%s)\n",
progname, mop->mo_ldd.ldd_mount_type,
MT_STR(&mop->mo_ldd));
return EINVAL;
}
/* Journal size in MB */
if (strstr(mop->mo_mkfsopts, "-J") == NULL &&
mop->mo_device_kb > 1024 * 1024) {
/* Choose our own default journal size */
long journal_mb = 0, max_mb;
/* cap journal size at 4GB for MDT, leave at 1GB for OSTs */
if (IS_MDT(&mop->mo_ldd))
max_mb = 4096;
else if (IS_OST(&mop->mo_ldd))
max_mb = 1024;
else /* Use mke2fs default size for MGS */
max_mb = 0;
/* Use at most 4% of device for journal */
journal_mb = mop->mo_device_kb * 4 / (1024 * 100);
if (journal_mb > max_mb)
journal_mb = max_mb;
if (journal_mb) {
snprintf(buf, sizeof(buf), " -J size=%ld", journal_mb);
strscat(mop->mo_mkfsopts, buf,
sizeof(mop->mo_mkfsopts));
}
}
/*
* The inode size is constituted by following elements
* (assuming all files are in composite layout and has
* 3 components):
*
* ldiskfs inode size: 160
* MDT extended attributes size, including:
* ext4_xattr_header: 32
* LOV EA size: 32(lov_comp_md_v1) +
* 3 * 40(lov_comp_md_entry_v1) +
* 3 * 32(lov_mds_md) +
* stripes * 24(lov_ost_data) +
* 16(xattr_entry) + 4("lov")
* LMA EA size: 24(lustre_mdt_attrs) +
* 16(xattr_entry) + 4("lma")
* SOM EA size: 24(lustre_som_attrs) +
* 16(xattr_entry) + 4("som")
* link EA size: 24(link_ea_header) + 18(link_ea_entry) +
* 16(filename) + 16(xattr_entry) + 4("link")
* and some margin for 4-byte alignment, ACLs and other EAs.
*
* If we say the average filename length is about 32 bytes,
* the calculation looks like:
* 160 + 32 + (32+3*(40+32)+24*stripes+20) + (24+20) + (24+20) +
* (24+20) + (~42+16+20) + other <= 512*2^m, {m=0,1,2,3}
*/
if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
if (IS_MDT(&mop->mo_ldd)) {
if (mop->mo_stripe_count > 59)
inode_size = 512; /* bz 7241 */
/* see also "-i" below for EA blocks */
else if (mop->mo_stripe_count > 16)
inode_size = 2048;
else
inode_size = 1024;
} else if (IS_OST(&mop->mo_ldd)) {
/* We store MDS FID and necessary composite
* layout information in the OST object EA:
* ldiskfs inode size: 160
* OST extended attributes size, including:
* ext4_xattr_header: 32
* LMA EA size: 24(lustre_mdt_attrs) +
* 16(xattr_entry) + 4("lma")
* FID EA size: 52(filter_fid) +
* 16(xattr_entry) + 4("fid")
* 160 + 32 + (24+20) + (52+20) = 308
*/
inode_size = 512;
}
if (inode_size > 0) {
snprintf(buf, sizeof(buf), " -I %ld", inode_size);
strscat(mop->mo_mkfsopts, buf,
sizeof(mop->mo_mkfsopts));
}
}
/* Bytes_per_inode: disk size / num inodes */
if (strstr(mop->mo_mkfsopts, "-i") == NULL &&
strstr(mop->mo_mkfsopts, "-N") == NULL) {
long bytes_per_inode = 0;
/* Allocate more inodes on MDT devices. There is
* no data stored on the MDT, and very little extra
* metadata beyond the inode. It could go down as
* low as 1024 bytes, but this is conservative.
* Account for external EA blocks for wide striping.
*/
if (IS_MDT(&mop->mo_ldd)) {
bytes_per_inode = inode_size + 1536;
if (mop->mo_stripe_count > 59) {
int extra = mop->mo_stripe_count * 24;
extra = ((extra - 1) | 4095) + 1;
bytes_per_inode += extra;
}
}
/* Allocate fewer inodes on large OST devices. Most
* filesystems can be much more aggressive than even
* this, but it is impossible to know in advance.
*/
if (IS_OST(&mop->mo_ldd)) {
/* OST > 16TB assume average file size 1MB */
if (mop->mo_device_kb > (16ULL << 30))
bytes_per_inode = 1024 * 1024;
/* OST > 4TB assume average file size 512kB */
else if (mop->mo_device_kb > (4ULL << 30))
bytes_per_inode = 512 * 1024;
/* OST > 1TB assume average file size 256kB */
else if (mop->mo_device_kb > (1ULL << 30))
bytes_per_inode = 256 * 1024;
/* OST > 10GB assume average file size 64kB,
* plus a bit so that inodes will fit into a
* 256x flex_bg without overflowing.
*/
else if (mop->mo_device_kb > (10ULL << 20))
bytes_per_inode = 69905;
}
if (bytes_per_inode > 0) {
snprintf(buf, sizeof(buf), " -i %ld", bytes_per_inode);
strscat(mop->mo_mkfsopts, buf,
sizeof(mop->mo_mkfsopts));
mop->mo_inode_size = bytes_per_inode;
}
}
if (verbose < 2)
strscat(mop->mo_mkfsopts, " -q", sizeof(mop->mo_mkfsopts));
/* start handle -O mkfs options */
start = strstr(mop->mo_mkfsopts, "-O");
if (start) {
if (strstr(start + 2, "-O") != NULL) {
fprintf(stderr,
"%s: don't specify multiple -O options\n",
progname);
return EINVAL;
}
start = moveopts_to_end(start);
maxbuflen = sizeof(mop->mo_mkfsopts) -
(start - mop->mo_mkfsopts) - strlen(start);
ret = enable_default_ext4_features(mop, start, maxbuflen, 1);
} else {
start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
maxbuflen = sizeof(mop->mo_mkfsopts) - strlen(mop->mo_mkfsopts);
ret = enable_default_ext4_features(mop, start, maxbuflen, 0);
}
if (ret)
return ret;
/* end handle -O mkfs options */
/* start handle -E mkfs options */
start = strstr(mop->mo_mkfsopts, "-E");
if (start) {
if (strstr(start + 2, "-E") != NULL) {
fprintf(stderr,
"%s: don't specify multiple -E options\n",
progname);
return EINVAL;
}
start = moveopts_to_end(start);
maxbuflen = sizeof(mop->mo_mkfsopts) -
(start - mop->mo_mkfsopts) - strlen(start);
ext_opts = 1;
} else {
start = mop->mo_mkfsopts + strlen(mop->mo_mkfsopts);
maxbuflen = sizeof(mop->mo_mkfsopts) - strlen(mop->mo_mkfsopts);
}
/* In order to align the filesystem metadata on 1MB boundaries,
* give a resize value that will reserve a power-of-two group
* descriptor blocks, but leave one block for the superblock.
* Only useful for filesystems with < 2^32 blocks due to resize
* limitations.
*/
if (!enable_64bit && strstr(mop->mo_mkfsopts, "meta_bg") == NULL &&
IS_OST(&mop->mo_ldd) && mop->mo_device_kb > 100 * 1024) {
unsigned int group_blocks = mop->mo_blocksize_kb * 8192;
unsigned int desc_per_block = mop->mo_blocksize_kb * 1024 / 32;
unsigned int resize_blks;
resize_blks = (1ULL<<32) - desc_per_block*group_blocks;
snprintf(buf, sizeof(buf), "%u", resize_blks);
append_unique(start, ext_opts ? "," : " -E ",
"resize", buf, maxbuflen);
ext_opts = 1;
}
/* Avoid zeroing out the full journal - speeds up mkfs */
if (is_e2fsprogs_feature_supp("-E lazy_journal_init"))
append_unique(start, ext_opts ? "," : " -E ",
"lazy_journal_init", NULL, maxbuflen);
/* end handle -E mkfs options */
/* Allow reformat of full devices (as opposed to partitions).
* We already checked for mounted dev.
*/
strscat(mop->mo_mkfsopts, " -F", sizeof(mop->mo_mkfsopts));
snprintf(mkfs_cmd, sizeof(mkfs_cmd), "%s -j -b %d -L %s ", MKE2FS,
mop->mo_blocksize_kb * 1024, mop->mo_ldd.ldd_svname);
/* For loop device format the dev, not the filename */
dev = mop->mo_device;
if (mop->mo_flags & MO_IS_LOOP)
dev = mop->mo_loopdev;
vprint("formatting backing filesystem %s on %s\n",
MT_STR(&mop->mo_ldd), dev);
vprint("\ttarget name %s\n", mop->mo_ldd.ldd_svname);
vprint("\tkilobytes %llu\n", mop->mo_device_kb);
vprint("\toptions %s\n", mop->mo_mkfsopts);
/* mkfs_cmd's trailing space is important! */
strscat(mkfs_cmd, mop->mo_mkfsopts, sizeof(mkfs_cmd));
strscat(mkfs_cmd, " ", sizeof(mkfs_cmd));
strscat(mkfs_cmd, dev, sizeof(mkfs_cmd));
if (mop->mo_device_kb != 0) {
snprintf(buf, sizeof(buf), " %lluk",
(unsigned long long)mop->mo_device_kb);
strscat(mkfs_cmd, buf, sizeof(mkfs_cmd));
}
vprint("mkfs_cmd = %s\n", mkfs_cmd);
ret = run_command(mkfs_cmd, sizeof(mkfs_cmd));
if (ret) {
fatal();
fprintf(stderr, "Unable to build fs %s (%d)\n", dev, ret);
}
return ret;
}
static int enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
size_t maxbuflen, bool user_spec)
{
unsigned long long blocks = mop->mo_device_kb / mop->mo_blocksize_kb;
bool enable_64bit = false;
/* Enable large block addresses if the LUN is over 2^32 blocks. */
if (blocks > 0xffffffffULL && is_e2fsprogs_feature_supp("-O 64bit"))
enable_64bit = true;
if (IS_OST(&mop->mo_ldd)) {
append_unique(anchor, user_spec ? "," : " -O ",
"extents", NULL, maxbuflen);
append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
} else if (IS_MDT(&mop->mo_ldd)) {
append_unique(anchor, user_spec ? "," : " -O ",
"dirdata", NULL, maxbuflen);
append_unique(anchor, ",", "uninit_bg", NULL, maxbuflen);
if (enable_64bit)
append_unique(anchor, ",", "extents", NULL, maxbuflen);
else
append_unique(anchor, ",", "^extents", NULL, maxbuflen);
} else {
append_unique(anchor, user_spec ? "," : " -O ",
"uninit_bg", NULL, maxbuflen);
}
/* Multiple mount protection enabled only if failover node specified */
if (mop->mo_flags & MO_FAILOVER) {
if (is_e2fsprogs_feature_supp("-O mmp"))
append_unique(anchor, ",", "mmp", NULL, maxbuflen);
else
disp_old_e2fsprogs_msg("mmp", 1);
}
/* Allow more than 65000 subdirectories */
if (is_e2fsprogs_feature_supp("-O dir_nlink"))
append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
/* The following options are only valid for ext4-based ldiskfs.
* If --backfstype=ext3 is specified, do not enable them. */
if (mop->mo_ldd.ldd_mount_type == LDD_MT_EXT3)
return 0;
/* Enable quota by default */
if (is_e2fsprogs_feature_supp("-O quota")) {
append_unique(anchor, ",", "quota", NULL, maxbuflen);
} else {
fatal();
fprintf(stderr, "\"-O quota\" must be supported by "
"e2fsprogs, please upgrade your e2fsprogs.\n");
return EINVAL;
}
/* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
if (is_e2fsprogs_feature_supp("-O huge_file"))
append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
if (enable_64bit)
append_unique(anchor, ",", "64bit", NULL, maxbuflen);
if (blocks >= 0x1000000000 && is_e2fsprogs_feature_supp("-O meta_bg"))
append_unique(anchor, ",", "meta_bg", NULL, maxbuflen);
if (enable_64bit || strstr(mop->mo_mkfsopts, "meta_bg"))
append_unique(anchor, ",", "^resize_inode", NULL, maxbuflen);
/* Cluster inode/block bitmaps and inode table for more efficient IO.
* Align the flex groups on a 1MB boundary for better performance. */
/* This -O feature needs to go last, since it adds the "-G" option. */
if (is_e2fsprogs_feature_supp("-O flex_bg")) {
char tmp_buf[64];
append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
if (IS_OST(&mop->mo_ldd) &&
strstr(mop->mo_mkfsopts, "-G") == NULL) {
snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
1024 / mop->mo_blocksize_kb);
strscat(anchor, tmp_buf, maxbuflen);
}
}
/* Don't add any more "-O" options here, see last comment above */
return 0;
}
int main(int argc, char *const argv[])
{
struct mkfs_opts mop;
struct lustre_disk_data *ldd = &mop.mo_ldd;
char *mountopts = NULL;
char wanted_mountopts[512] = "";
unsigned mount_type;
int ret = 0;
int ret2 = 0;
progname = strrchr(argv[0], '/');
if (progname != NULL)
progname++;
else
progname = argv[0];
if ((argc < 2) || (argv[argc - 1][0] == '-')) {
usage(stderr);
return EINVAL;
}
memset(&mop, 0, sizeof(mop));
set_defaults(&mop);
/* device is last arg */
strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
ret = osd_init();
if (ret != 0)
return ret;
#ifdef TUNEFS
/* For tunefs, we must read in the old values before parsing any
new ones. */
/* Check whether the disk has already been formatted by mkfs.lustre */
ret = osd_is_lustre(mop.mo_device, &mount_type);
if (ret == 0) {
fatal();
fprintf(stderr, "Device %s has not been formatted with "
"mkfs.lustre\n", mop.mo_device);
ret = ENODEV;
goto out;
}
ldd->ldd_mount_type = mount_type;
ret = osd_read_ldd(mop.mo_device, ldd);
if (ret != 0) {
fatal();
fprintf(stderr, "Failed to read previous Lustre data from %s "
"(%d)\n", mop.mo_device, ret);
goto out;
}
ldd->ldd_flags &= ~(LDD_F_WRITECONF | LDD_F_VIRGIN);
/* svname of the form lustre:OST1234 means never registered */
ret = strlen(ldd->ldd_svname);
if (ldd->ldd_svname[ret - 8] == ':') {
ldd->ldd_svname[ret - 8] = '-';
ldd->ldd_flags |= LDD_F_VIRGIN;
} else if (ldd->ldd_svname[ret - 8] == '=') {
ldd->ldd_svname[ret - 8] = '-';
ldd->ldd_flags |= LDD_F_WRITECONF;
}
if (strstr(ldd->ldd_params, PARAM_MGSNODE))
mop.mo_mgs_failnodes++;
if (verbose > 0)
print_ldd("Read previous values", ldd);
#endif /* TUNEFS */
ret = parse_opts(argc, argv, &mop, &mountopts);
if (ret != 0 || version)
goto out;
if (!IS_MDT(ldd) && !IS_OST(ldd) && !IS_MGS(ldd)) {
fatal();
fprintf(stderr, "must set target type: MDT,OST,MGS\n");
ret = EINVAL;
goto out;
}
if (((IS_MDT(ldd) || IS_MGS(ldd))) && IS_OST(ldd)) {
fatal();
fprintf(stderr, "OST type is exclusive with MDT,MGS\n");
ret = EINVAL;
goto out;
}
/* Stand alone MGS doesn't need an index */
if (!IS_MDT(ldd) && IS_MGS(ldd)) {
#ifndef TUNEFS
/* But if --index was specified flag an error */
if (!(ldd->ldd_flags & LDD_F_NEED_INDEX)) {
badopt("index", "MDT,OST");
goto out;
}
#endif
ldd->ldd_flags &= ~LDD_F_NEED_INDEX;
}
if ((ldd->ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
(LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
fatal();
fprintf(stderr, "Can't find the target index, "
"specify with --index\n");
ret = EINVAL;
goto out;
}
if (ldd->ldd_flags & LDD_F_NEED_INDEX)
fprintf(stderr, "warning: %s: for Lustre 2.4 and later, the "
"target index must be specified with --index\n",
mop.mo_device);
/* If no index is supplied for MDT by default set index to zero */
if (IS_MDT(ldd) && (ldd->ldd_svindex == INDEX_UNASSIGNED)) {
ldd->ldd_flags &= ~LDD_F_NEED_INDEX;
ldd->ldd_svindex = 0;
}
if (!IS_MGS(ldd) && (mop.mo_mgs_failnodes == 0)) {
fatal();
if (IS_MDT(ldd))
fprintf(stderr, "Must specify --mgs or --mgsnode\n");
else
fprintf(stderr, "Must specify --mgsnode\n");
ret = EINVAL;
goto out;
}
if ((IS_MDT(ldd) || IS_OST(ldd)) && ldd->ldd_fsname[0] == '\0') {
fatal();
fprintf(stderr, "Must specify --fsname for MDT/OST device\n");
ret = EINVAL;
goto out;
}
/* These are the permanent mount options (always included) */
ret = osd_prepare_lustre(&mop,
wanted_mountopts, sizeof(wanted_mountopts));
if (ret != 0) {
fatal();
fprintf(stderr, "unable to prepare backend (%d)\n", ret);
goto out;
}
if (mountopts) {
trim_mountfsoptions(mountopts);
if (check_mountfsoptions(mountopts, wanted_mountopts)) {
ret = EINVAL;
goto out;
}
snprintf(ldd->ldd_mount_opts, sizeof(ldd->ldd_mount_opts),
"%s", mountopts);
} else {
#ifdef TUNEFS
if (ldd->ldd_mount_opts[0] == 0)
/* use the defaults unless old opts exist */
#endif
{
snprintf(ldd->ldd_mount_opts,
sizeof(ldd->ldd_mount_opts),
"%s", wanted_mountopts);
trim_mountfsoptions(ldd->ldd_mount_opts);
}
}
ret = osd_fix_mountopts(&mop, ldd->ldd_mount_opts,
sizeof(ldd->ldd_mount_opts));
if (ret != 0) {
fatal();
fprintf(stderr, "unable to fix mountfsoptions (%d)\n", ret);
goto out;
}
server_make_name(ldd->ldd_flags, ldd->ldd_svindex,
ldd->ldd_fsname, ldd->ldd_svname);
if (verbose >= 0)
print_ldd("Permanent disk data", ldd);
if (print_only) {
printf("exiting before disk write.\n");
goto out;
}
if (check_mtab_entry(mop.mo_device, mop.mo_device, NULL, NULL))
return(EEXIST);
/* Create the loopback file */
if (mop.mo_flags & MO_IS_LOOP) {
ret = access(mop.mo_device, F_OK);
if (ret != 0)
ret = errno;
#ifndef TUNEFS
/* Reformat the loopback file */
if (ret != 0 || (mop.mo_flags & MO_FORCEFORMAT)) {
ret = loop_format(&mop);
if (ret != 0)
goto out;
}
#endif
if (ret == 0)
ret = loop_setup(&mop);
if (ret != 0) {
fatal();
fprintf(stderr, "Loop device setup for %s failed: %s\n",
mop.mo_device, strerror(ret));
goto out;
}
}
#ifndef TUNEFS
/* Check whether the disk has already been formatted by mkfs.lustre */
if (!(mop.mo_flags & MO_FORCEFORMAT)) {
ret = osd_is_lustre(mop.mo_device, &mount_type);
if (ret != 0) {
fatal();
fprintf(stderr, "Device %s was previously formatted "
"for lustre. Use --reformat to reformat it, "
"or tunefs.lustre to modify.\n",
mop.mo_device);
goto out;
}
}
/* Format the backing filesystem */
ret = osd_make_lustre(&mop);
if (ret != 0) {
fatal();
fprintf(stderr, "mkfs failed %d\n", ret);
goto out;
}
#else /* !TUNEFS */
/* update svname with '=' to refresh config */
if (ldd->ldd_flags & LDD_F_WRITECONF) {
struct mount_opts opts;
opts.mo_ldd = *ldd;
opts.mo_source = mop.mo_device;
(void) osd_label_lustre(&opts);
}
/* Enable quota accounting */
if (mop.mo_flags & MO_QUOTA) {
ret = osd_enable_quota(&mop);
goto out;
}
#endif /* !TUNEFS */
/* Write our config files */
ret = osd_write_ldd(&mop);
if (ret != 0) {
fatal();
fprintf(stderr, "failed to write local files\n");
goto out;
}
out:
osd_fini();
ret2 = loop_cleanup(&mop);
if (ret == 0)
ret = ret2;
/* Fix any crazy return values from system() */
if (ret != 0 && ((ret & 255) == 0))
return 1;
if (ret != 0)
verrprint("%s: exiting with %d (%s)\n",
progname, ret, strerror(ret));
return ret;
}
int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
char **mountopts)
{
static struct option long_opt[] = {
{ "backfs-mount-opts", required_argument, NULL, 'B' },
{ "failnode", required_argument, NULL, 'f' },
{ "failover", required_argument, NULL, 'f' },
{ "help", no_argument, NULL, 'h' },
{ "index", required_argument, NULL, 'i' },
{ "fsname", required_argument, NULL, 'L' },
{ "mgsnode", required_argument, NULL, 'm' },
{ "mgsnid", required_argument, NULL, 'm' },
{ "dryrun", no_argument, NULL, 'n' },
{ "mountfsoptions", required_argument, NULL, 'o' },
{ "param", required_argument, NULL, 'p' },
{ "quiet", no_argument, NULL, 'q' },
{ "servicenode", required_argument, NULL, 's' },
{ "network", required_argument, NULL, 't' },
{ "comment", required_argument, NULL, 'u' },
{ "force-nohostid", no_argument, NULL, 'U' },
{ "verbose", no_argument, NULL, 'v' },
{ "version", no_argument, NULL, 'V' },
#ifndef TUNEFS
{ "backfstype", required_argument, NULL, 'b' },
{ "stripe-count-hint", required_argument, NULL, 'c' },
{ "device-size", required_argument, NULL, 'd' },
{ "mgs", no_argument, NULL, 'G' },
{ "mkfsoptions", required_argument, NULL, 'k' },
{ "mdt", no_argument, NULL, 'M' },
{ "nomgs", no_argument, NULL, 'N' },
{ "ost", no_argument, NULL, 'O' },
{ "reformat", no_argument, NULL, 'r' },
{ "replace", no_argument, NULL, 'R' },
#else
{ "erase-params", no_argument, NULL, 'e' },
{ "quota", no_argument, NULL, 'Q' },
{ "writeconf", no_argument, NULL, 'w' },
#endif
{ 0, 0, NULL, 0 }
};
char *optstring = "B:f:hi:L:m:no:p:qs:t:u:vV"
#ifndef TUNEFS
"b:c:d:Gk:MNOrR";
#else
"eQw";
#endif
struct lustre_disk_data *ldd = &mop->mo_ldd;
char new_fsname[16] = { 0 };
int opt;
int rc, longidx;
int failnode_set = 0, servicenode_set = 0;
int replace = 0;
bool index_option = false;
while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
EOF) {
switch (opt) {
case 'B':
mop->mo_mountopts = optarg;
break;
case 'f':
case 's': {
char *nids;
if ((opt == 'f' && servicenode_set) ||
(opt == 's' && failnode_set)) {
fprintf(stderr, "%s: %s cannot use with --%s\n",
progname, long_opt[longidx].name,
opt == 'f' ? "servicenode" :
"failnode");
return 1;
}
nids = convert_hostnames(optarg);
if (nids == NULL)
return 1;
rc = append_param(ldd->ldd_params, PARAM_FAILNODE,
nids, ':');
free(nids);
if (rc != 0)
return rc;
/* Must update the mgs logs */
ldd->ldd_flags |= LDD_F_UPDATE;
if (opt == 'f') {
failnode_set = 1;
} else {
ldd->ldd_flags |= LDD_F_NO_PRIMNODE;
servicenode_set = 1;
}
mop->mo_flags |= MO_FAILOVER;
break;
}
case 'h':
usage(stdout);
return 1;
case 'i': {
char *endptr = NULL;
int base;
index_option = true;
/* LU-2374: check whether it is OST/MDT later */
base = (strlen(optarg) > 1 &&
!strncmp(optarg, "0x", 2)) ? 16 : 10;
/* Allowed input are base 16 and base 10 numbers only */
mop->mo_ldd.ldd_svindex = strtoul(optarg,
&endptr, base);
if (*endptr != '\0') {
fprintf(stderr, "%s: wrong index %s. "
"Target index must be decimal or "
"hexadecimal.\n",
progname, optarg);
return 1;
}
if (ldd->ldd_svindex >= INDEX_UNASSIGNED) {
fprintf(stderr, "%s: wrong index %u. "
"Target index must be less than %u.\n",
progname, ldd->ldd_svindex,
INDEX_UNASSIGNED);
return 1;
}
ldd->ldd_flags &= ~LDD_F_NEED_INDEX;
break;
}
case 'L': {
rc = lustre_is_fsname_valid(optarg, 1,
LUSTRE_MAXFSNAME);
if (rc < 0) {
fprintf(stderr, "%s: filesystem name must be "
"1-%d chars\n", progname,
LUSTRE_MAXFSNAME);
return 1;
} else if (rc > 0) {
fprintf(stderr, "%s: char '%c' not allowed in "
"filesystem name\n", progname, rc);
return 1;
}
strscpy(new_fsname, optarg, sizeof(new_fsname));
break;
}
case 'm': {
char *nids = convert_hostnames(optarg);
if (nids == NULL)
return 1;
rc = append_param(ldd->ldd_params, PARAM_MGSNODE,
nids, ':');
free(nids);
if (rc != 0)
return rc;
mop->mo_mgs_failnodes++;
break;
}
case 'n':
print_only++;
break;
case 'o':
*mountopts = optarg;
break;
case 'p':
rc = add_param(ldd->ldd_params, NULL, optarg);
if (rc != 0)
return rc;
/* Must update the mgs logs */
ldd->ldd_flags |= LDD_F_UPDATE;
break;
case 'q':
verbose--;
break;
case 't':
if (!IS_MDT(ldd) && !IS_OST(ldd)) {
badopt(long_opt[longidx].name, "MDT,OST");
return 1;
}
if (optarg == NULL)
return 1;
rc = add_param(ldd->ldd_params, PARAM_NETWORK, optarg);
if (rc != 0)
return rc;
/* Must update the mgs logs */
ldd->ldd_flags |= LDD_F_UPDATE;
break;
case 'u':
strscpy(ldd->ldd_userdata, optarg,
sizeof(ldd->ldd_userdata));
break;
case 'U':
mop->mo_flags |= MO_NOHOSTID_CHECK;
break;
case 'v':
verbose++;
break;
case 'V':
++version;
fprintf(stdout, "%s %s\n", progname,
LUSTRE_VERSION_STRING);
return 0;
#ifndef TUNEFS
case 'b': {
int i = 0;
do {
if (strcmp(optarg, mt_str(i)) == 0) {
ldd->ldd_mount_type = i;
break;
}
} while (++i < LDD_MT_LAST);
if (i == LDD_MT_LAST) {
fprintf(stderr, "%s: invalid backend filesystem"
" type %s\n", progname, optarg);
return 1;
}
break;
}
case 'c':
if (IS_MDT(ldd)) {
int stripe_count = atol(optarg);
if (stripe_count <= 0) {
fprintf(stderr, "%s: bad stripe count "
"%d\n", progname, stripe_count);
return 1;
}
mop->mo_stripe_count = stripe_count;
} else {
badopt(long_opt[longidx].name, "MDT");
return 1;
}
break;
case 'd':
mop->mo_device_kb = atol(optarg);
break;
case 'G':
ldd->ldd_flags |= LDD_F_SV_TYPE_MGS;
break;
case 'k':
strscpy(mop->mo_mkfsopts, optarg,
sizeof(mop->mo_mkfsopts));
break;
case 'M':
ldd->ldd_flags |= LDD_F_SV_TYPE_MDT;
break;
case 'N':
ldd->ldd_flags &= ~LDD_F_SV_TYPE_MGS;
break;
case 'O':
ldd->ldd_flags |= LDD_F_SV_TYPE_OST;
break;
case 'r':
mop->mo_flags |= MO_FORCEFORMAT;
break;
case 'R':
replace = 1;
break;
#else /* !TUNEFS */
case 'e':
ldd->ldd_params[0] = '\0';
/* Must update the mgs logs */
ldd->ldd_flags |= LDD_F_UPDATE;
break;
case 'Q':
mop->mo_flags |= MO_QUOTA;
break;
case 'w':
ldd->ldd_flags |= LDD_F_WRITECONF;
break;
#endif /* !TUNEFS */
default:
if (opt != '?') {
fatal();
fprintf(stderr, "Unknown option '%c'\n", opt);
}
return EINVAL;
}
}
if (strlen(new_fsname) > 0) {
if (!(mop->mo_flags & MO_FORCEFORMAT) &&
(!(ldd->ldd_flags &
(LDD_F_UPGRADE14 | LDD_F_VIRGIN | LDD_F_WRITECONF)))) {
fprintf(stderr, "%s: cannot change the name "
"of a registered target\n", progname);
return 1;
}
strscpy(ldd->ldd_fsname, new_fsname, sizeof(ldd->ldd_fsname));
}
if (index_option && !(mop->mo_ldd.ldd_flags &
(LDD_F_UPGRADE14 | LDD_F_VIRGIN |
LDD_F_WRITECONF))) {
fprintf(stderr, "%s: cannot change the index of"
" a registered target\n", progname);
return 1;
}
/* Need to clear this flag after parsing 'L' and 'i' options. */
if (replace)
ldd->ldd_flags &= ~LDD_F_VIRGIN;
if (optind == argc) {
/* The user didn't specify device name */
fatal();
fprintf(stderr, "Not enough arguments - device name or "
"pool/dataset name not specified.\n");
return EINVAL;
} else {
/* The device or pool/filesystem name */
strscpy(mop->mo_device, argv[optind], sizeof(mop->mo_device));
/* Followed by optional vdevs */
if (optind < argc - 1)
mop->mo_pool_vdevs = (char **) &argv[optind + 1];
}
return 0;
}
int parse_opts(int argc, char *const argv[], struct mkfs_opts *mop,
char **mountopts)
{
static struct option long_opt[] = {
{ "backfstype", required_argument, NULL, 'b' },
{ "stripe-count-hint", required_argument, NULL, 'c' },
{ "comment", required_argument, NULL, 'u' },
{ "configdev", required_argument, NULL, 'C' },
{ "device-size", required_argument, NULL, 'd' },
{ "dryrun", no_argument, NULL, 'n' },
{ "erase-params", no_argument, NULL, 'e' },
{ "failnode", required_argument, NULL, 'f' },
{ "failover", required_argument, NULL, 'f' },
{ "mgs", no_argument, NULL, 'G' },
{ "help", no_argument, NULL, 'h' },
{ "index", required_argument, NULL, 'i' },
{ "mkfsoptions", required_argument, NULL, 'k' },
{ "mgsnode", required_argument, NULL, 'm' },
{ "mgsnid", required_argument, NULL, 'm' },
{ "mdt", no_argument, NULL, 'M' },
{ "fsname", required_argument, NULL, 'L' },
{ "noformat", no_argument, NULL, 'n' },
{ "nomgs", no_argument, NULL, 'N' },
{ "mountfsoptions", required_argument, NULL, 'o' },
{ "ost", no_argument, NULL, 'O' },
{ "param", required_argument, NULL, 'p' },
{ "print", no_argument, NULL, 'n' },
{ "quiet", no_argument, NULL, 'q' },
{ "quota", no_argument, NULL, 'Q' },
{ "reformat", no_argument, NULL, 'r' },
{ "replace", no_argument, NULL, 'R' },
{ "servicenode", required_argument, NULL, 's' },
{ "network", required_argument, NULL, 't' },
{ "verbose", no_argument, NULL, 'v' },
{ "writeconf", no_argument, NULL, 'w' },
{ 0, 0, NULL, 0 }
};
char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:PqrRs:t:Uu:vw";
int opt;
int rc, longidx;
int failnode_set = 0, servicenode_set = 0;
int replace = 0;
while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
EOF) {
switch (opt) {
case 'b': {
int i = 0;
while (i < LDD_MT_LAST) {
if (strcmp(optarg, mt_str(i)) == 0) {
mop->mo_ldd.ldd_mount_type = i;
break;
}
i++;
}
if (i == LDD_MT_LAST) {
fprintf(stderr, "%s: invalid backend filesystem"
" type %s\n", progname, optarg);
return 1;
}
break;
}
case 'c':
if (IS_MDT(&mop->mo_ldd)) {
int stripe_count = atol(optarg);
if (stripe_count <= 0) {
fprintf(stderr, "%s: bad stripe count "
"%d\n", progname, stripe_count);
return 1;
}
mop->mo_stripe_count = stripe_count;
} else {
badopt(long_opt[longidx].name, "MDT");
return 1;
}
break;
case 'C': /* Configdev */
//FIXME
printf("Configdev not implemented\n");
return 1;
case 'd':
mop->mo_device_kb = atol(optarg);
break;
case 'e':
mop->mo_ldd.ldd_params[0] = '\0';
/* Must update the mgs logs */
mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
break;
case 'f':
case 's': {
char *nids;
if ((opt == 'f' && servicenode_set)
|| (opt == 's' && failnode_set)) {
fprintf(stderr, "%s: %s cannot use with --%s\n",
progname, long_opt[longidx].name,
opt == 'f' ? "servicenode" : "failnode");
return 1;
}
nids = convert_hostnames(optarg);
if (!nids)
return 1;
rc = add_param(mop->mo_ldd.ldd_params, PARAM_FAILNODE,
nids);
free(nids);
if (rc)
return rc;
/* Must update the mgs logs */
mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
if (opt == 'f') {
failnode_set = 1;
} else {
mop->mo_ldd.ldd_flags |= LDD_F_NO_PRIMNODE;
servicenode_set = 1;
}
mop->mo_flags |= MO_FAILOVER;
break;
}
case 'G':
mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MGS;
break;
case 'h':
usage(stdout);
return 1;
case 'i':
if (!(mop->mo_ldd.ldd_flags &
(LDD_F_UPGRADE14 | LDD_F_VIRGIN |
LDD_F_WRITECONF))) {
fprintf(stderr, "%s: cannot change the index of"
" a registered target\n", progname);
return 1;
}
/* LU-2374: check whether it is OST/MDT later */
mop->mo_ldd.ldd_svindex = atol(optarg);
mop->mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
break;
case 'k':
strscpy(mop->mo_mkfsopts, optarg,
sizeof(mop->mo_mkfsopts));
break;
case 'L': {
char *tmp;
if (!(mop->mo_flags & MO_FORCEFORMAT) &&
(!(mop->mo_ldd.ldd_flags &
(LDD_F_UPGRADE14 | LDD_F_VIRGIN |
LDD_F_WRITECONF)))) {
fprintf(stderr, "%s: cannot change the name of"
" a registered target\n", progname);
return 1;
}
if ((strlen(optarg) < 1) || (strlen(optarg) > 8)) {
fprintf(stderr, "%s: filesystem name must be "
"1-8 chars\n", progname);
return 1;
}
if ((tmp = strpbrk(optarg, "/:"))) {
fprintf(stderr, "%s: char '%c' not allowed in "
"filesystem name\n", progname, *tmp);
return 1;
}
strscpy(mop->mo_ldd.ldd_fsname, optarg,
sizeof(mop->mo_ldd.ldd_fsname));
break;
}
case 'm': {
char *nids = convert_hostnames(optarg);
if (!nids)
return 1;
rc = add_param(mop->mo_ldd.ldd_params, PARAM_MGSNODE,
nids);
free(nids);
if (rc)
return rc;
mop->mo_mgs_failnodes++;
break;
}
case 'M':
mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_MDT;
break;
case 'n':
print_only++;
break;
case 'N':
mop->mo_ldd.ldd_flags &= ~LDD_F_SV_TYPE_MGS;
break;
case 'o':
*mountopts = optarg;
break;
case 'O':
mop->mo_ldd.ldd_flags |= LDD_F_SV_TYPE_OST;
break;
case 'p':
rc = add_param(mop->mo_ldd.ldd_params, NULL, optarg);
if (rc)
return rc;
/* Must update the mgs logs */
mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
break;
case 'q':
verbose--;
break;
case 'r':
mop->mo_flags |= MO_FORCEFORMAT;
break;
case 'R':
replace = 1;
break;
case 't':
if (!IS_MDT(&mop->mo_ldd) && !IS_OST(&mop->mo_ldd)) {
badopt(long_opt[longidx].name, "MDT,OST");
return 1;
}
if (!optarg)
return 1;
rc = add_param(mop->mo_ldd.ldd_params,
PARAM_NETWORK, optarg);
if (rc != 0)
return rc;
/* Must update the mgs logs */
mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
break;
case 'u':
strscpy(mop->mo_ldd.ldd_userdata, optarg,
sizeof(mop->mo_ldd.ldd_userdata));
break;
case 'v':
verbose++;
break;
case 'w':
mop->mo_ldd.ldd_flags |= LDD_F_WRITECONF;
break;
case 'Q':
mop->mo_flags |= MO_QUOTA;
break;
default:
if (opt != '?') {
fatal();
fprintf(stderr, "Unknown option '%c'\n", opt);
}
return EINVAL;
}
}//while
/* Need to clear this flag after parsing 'L' and 'i' options. */
if (replace)
mop->mo_ldd.ldd_flags &= ~LDD_F_VIRGIN;
if (optind == argc) {
/* The user didn't specify device name */
fatal();
fprintf(stderr, "Not enough arguments - device name or "
"pool/dataset name not specified.\n");
return EINVAL;
} else {
/* The device or pool/filesystem name */
strscpy(mop->mo_device, argv[optind], sizeof(mop->mo_device));
/* Followed by optional vdevs */
if (optind < argc - 1)
mop->mo_pool_vdevs = (char **) &argv[optind + 1];
}
return 0;
}
