* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, Whamcloud, Inc.
+ * Copyright (c) 2011, 2012, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#ifndef _GNU_SOURCE
#define _GNU_SOURCE
#endif
+#include "mount_utils.h"
#include <stdlib.h>
#include <stdio.h>
#include <unistd.h>
#include <limits.h>
#include <ctype.h>
-#ifdef __linux__
-/* libcfs.h is not really needed here, but on SLES10/PPC, fs.h includes idr.h
- * which requires BITS_PER_LONG to be defined */
-#include <libcfs/libcfs.h>
-#ifndef BLKGETSIZE64
-#include <linux/fs.h> /* for BLKGETSIZE64 */
-#endif
-#include <linux/version.h>
-#endif
#include <lustre_disk.h>
#include <lustre_param.h>
#include <lnet/lnetctl.h>
#include <lustre_ver.h>
-#include "mount_utils.h"
#ifndef PATH_MAX
#define PATH_MAX 4096
static int print_only = 0;
static int upgrade_to_18 = 0;
+#ifdef HAVE_LDISKFS_OSD
+#define FSLIST_LDISKFS "ldiskfs"
+#define HAVE_FSLIST
+#else
+ #define FSLIST_LDISKFS ""
+#endif /* HAVE_LDISKFS_OSD */
+#ifdef HAVE_ZFS_OSD
+ #ifdef HAVE_FSLIST
+ #define FSLIST_ZFS "|zfs"
+ #else
+ #define FSLIST_ZFS "zfs"
+ #define HAVE_FSLIST
+ #endif
+#else
+ #define FSLIST_ZFS ""
+#endif /* HAVE_ZFS_OSD */
+
+#ifndef HAVE_FSLIST
+ #error "no backing OSD types (ldiskfs or ZFS) are configured"
+#endif
+
+#define FSLIST FSLIST_LDISKFS FSLIST_ZFS
+
void usage(FILE *out)
{
- fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
- fprintf(out, "usage: %s <target types> [options] <device>\n", progname);
- fprintf(out,
- "\t<device>:block device or file (e.g /dev/sda or /tmp/ost1)\n"
- "\ttarget types:\n"
- "\t\t--ost: object storage, mutually exclusive with mdt,mgs\n"
- "\t\t--mdt: metadata storage, mutually exclusive with ost\n"
- "\t\t--mgs: configuration management service - one per site\n"
- "\toptions (in order of popularity):\n"
- "\t\t--mgsnode=<nid>[,<...>] : NID(s) of a remote mgs node\n"
- "\t\t\trequired for all targets other than the mgs node\n"
- "\t\t--fsname=<filesystem_name> : default is 'lustre'\n"
- "\t\t--failnode=<nid>[,<...>] : NID(s) of a failover partner\n"
- "\t\t\tcannot be used with --servicenode\n"
- "\t\t--servicenode=<nid>[,<...>] : NID(s) of all service partners\n"
- "\t\t\ttreat all nodes as equal service node, cannot be used with --failnode\n"
- "\t\t--param <key>=<value> : set a permanent parameter\n"
- "\t\t\te.g. --param sys.timeout=40\n"
- "\t\t\t --param lov.stripesize=2M\n"
- "\t\t--index=#N : target index (i.e. ost index within lov)\n"
- "\t\t--comment=<user comment>: arbitrary string (%d bytes)\n"
- "\t\t--mountfsoptions=<opts> : permanent mount options\n"
- "\t\t--network=<net>[,<...>] : restrict OST/MDT to network(s)\n"
+ fprintf(out, "%s v"LUSTRE_VERSION_STRING"\n", progname);
+ fprintf(out, "usage: %s <target type> [--backfstype="FSLIST"] "
+ "--fsname=<filesystem name>\n"
+ "\t--index=<target index> [options] <device>\n", progname);
+#ifdef HAVE_ZFS_OSD
+ fprintf(out, "usage: %s <target type> --backfstype=zfs "
+ "--fsname=<filesystem name> [options]\n"
+ "\t<pool name>/<dataset name>\n"
+ "\t[[<vdev type>] <device> [<device> ...] [vdev type>] ...]\n",
+ progname);
+#endif
+ fprintf(out,
+ "\t<device>:block device or file (e.g /dev/sda or /tmp/ost1)\n"
+#ifdef HAVE_ZFS_OSD
+ "\t<pool name>: name of ZFS pool where target is created "
+ "(e.g. tank)\n"
+ "\t<dataset name>: name of new dataset, must be unique within "
+ "pool (e.g. ost1)\n"
+ "\t<vdev type>: type of vdev (mirror, raidz, raidz2, spare, "
+ "cache, log)\n"
+#endif
+ "\n"
+ "\ttarget types:\n"
+ "\t\t--mgs: configuration management service\n"
+ "\t\t--mdt: metadata storage, mutually exclusive with ost\n"
+ "\t\t--ost: object storage, mutually exclusive with mdt, mgs\n"
+ "\toptions (in order of popularity):\n"
+ "\t\t--index=#N: numerical target index (0..N)\n"
+ "\t\t\trequired for all targets other than the MGS\n"
+ "\t\t--fsname=<8_char_filesystem_name>: fs targets belong to\n"
+ "\t\t\trequired for all targets other than MGS\n"
+ "\t\t--mgsnode=<nid>[,<...>]: NID(s) of remote MGS\n"
+ "\t\t\trequired for all targets other than MGS\n"
+ "\t\t--mountfsoptions=<opts>: permanent mount options\n"
+ "\t\t--failnode=<nid>[,<...>]: NID(s) of backup failover node\n"
+ "\t\t\tmutually exclusive with --servicenode\n"
+ "\t\t--servicenode=<nid>[,<...>]: NID(s) of service partners\n"
+ "\t\t\ttreat nodes as equal service node, mutually exclusive "
+ "with --failnode\n"
+ "\t\t--param <key>=<value>: set a permanent parameter\n"
+ "\t\t\te.g. --param sys.timeout=40\n"
+ "\t\t\t --param lov.stripesize=2M\n"
+ "\t\t--network=<net>[,<...>]: restrict OST/MDT to network(s)\n"
#ifndef TUNEFS
- "\t\t--backfstype=<fstype> : backing fs type (ext3, ldiskfs)\n"
- "\t\t--device-size=#N(KB) : device size for loop devices\n"
- "\t\t--mkfsoptions=<opts> : format options\n"
- "\t\t--reformat: overwrite an existing disk\n"
- "\t\t--stripe-count-hint=#N : for optimizing MDT inode size\n"
- "\t\t--iam-dir: use IAM directory format, not ext3 compatible\n"
+ "\t\t--backfstype=<fstype>: backing fs type (ext3, ldiskfs)\n"
+ "\t\t--device-size=#N(KB): device size for loop devices\n"
+ "\t\t--mkfsoptions=<opts>: format options\n"
+ "\t\t--reformat: overwrite an existing disk\n"
+ "\t\t--stripe-count-hint=#N: for optimizing MDT inode size\n"
#else
- "\t\t--erase-params : erase all old parameter settings\n"
- "\t\t--nomgs: turn off MGS service on this MDT\n"
- "\t\t--writeconf: erase all config logs for this fs.\n"
+ "\t\t--erase-params: erase all old parameter settings\n"
+ "\t\t--nomgs: turn off MGS service on this MDT\n"
+ "\t\t--writeconf: erase all config logs for this fs.\n"
+ "\t\t--quota: enable space accounting on old 2.x device.\n"
#endif
- "\t\t--dryrun: just report what we would do; "
- "don't write to disk\n"
- "\t\t--verbose : e.g. show mkfs progress\n"
- "\t\t--quiet\n",
- (int)sizeof(((struct lustre_disk_data *)0)->ldd_userdata));
- return;
+ "\t\t--comment=<user comment>: arbitrary string (%d bytes)\n"
+ "\t\t--dryrun: report what we would do; don't write to disk\n"
+ "\t\t--verbose: e.g. show mkfs progress\n"
+ "\t\t--quiet\n",
+ (int)sizeof(((struct lustre_disk_data *)0)->ldd_userdata));
+ return;
}
/* ==================== Lustre config functions =============*/
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%s)\n",
+ 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_VIRGIN ? "first_time ":"",
ldd->ldd_flags & LDD_F_UPDATE ? "update ":"",
ldd->ldd_flags & LDD_F_WRITECONF ? "writeconf ":"",
- ldd->ldd_flags & LDD_F_IAM_DIR ? "IAM_dir_format ":"",
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);
void set_defaults(struct mkfs_opts *mop)
{
- mop->mo_ldd.ldd_magic = LDD_MAGIC;
- mop->mo_ldd.ldd_config_ver = 1;
- mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
- mop->mo_mgs_failnodes = 0;
- strcpy(mop->mo_ldd.ldd_fsname, "lustre");
- mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
-
- mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
- mop->mo_stripe_count = 1;
+ mop->mo_ldd.ldd_magic = LDD_MAGIC;
+ mop->mo_ldd.ldd_config_ver = 1;
+ mop->mo_ldd.ldd_flags = LDD_F_NEED_INDEX | LDD_F_UPDATE | LDD_F_VIRGIN;
+#ifdef HAVE_LDISKFS_OSD
+ mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
+#else
+ mop->mo_ldd.ldd_mount_type = LDD_MT_ZFS;
+#endif
+ mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
+ mop->mo_mgs_failnodes = 0;
+ mop->mo_stripe_count = 1;
+ mop->mo_pool_vdevs = NULL;
}
static inline void badopt(const char *opt, char *type)
sep = *s2;
*s2 = '\0';
nid = libcfs_str2nid(s1);
+ *s2 = sep;
if (nid == LNET_NID_ANY) {
fprintf(stderr, "%s: Can't parse NID '%s'\n",
char **mountopts)
{
static struct option long_opt[] = {
- {"iam-dir", 0, 0, 'a'},
{"backfstype", 1, 0, 'b'},
{"stripe-count-hint", 1, 0, 'c'},
{"comment", 1, 0, 'u'},
{"writeconf", 0, 0, 'w'},
{"upgrade_to_18", 0, 0, 'U'},
{"network", 1, 0, 't'},
+ {"quota", 0, 0, 'Q'},
{0, 0, 0, 0}
};
char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqrs:t:Uu:vw";
while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
EOF) {
switch (opt) {
- case 'a': {
- if (IS_MDT(&mop->mo_ldd))
- mop->mo_ldd.ldd_flags |= LDD_F_IAM_DIR;
- break;
- }
case 'b': {
int i = 0;
while (i < LDD_MT_LAST) {
}
i++;
}
+ if (i == LDD_MT_LAST) {
+ fprintf(stderr, "%s: invalid backend filesystem"
+ " type %s\n", progname, optarg);
+ return 1;
+ }
break;
}
case 'c':
case 'U':
upgrade_to_18 = 1;
break;
+ case 'Q':
+ mop->mo_flags |= MO_QUOTA;
+ break;
default:
if (opt != '?') {
fatal();
}
}//while
- /* Last arg is device */
- if (optind != argc - 1) {
- fatal();
- fprintf(stderr, "Bad argument: %s\n", argv[optind]);
- return EINVAL;
- }
-
- /* single argument: <device> */
- if (argc == 2)
- ++print_only;
+ 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;
}
char *mountopts = NULL;
char always_mountopts[512] = "";
char default_mountopts[512] = "";
+ unsigned mount_type;
int ret = 0;
if ((progname = strrchr(argv[0], '/')) != NULL)
/* device is last arg */
strscpy(mop.mo_device, argv[argc - 1], sizeof(mop.mo_device));
- /* Are we using a loop device? */
- ret = is_block(mop.mo_device);
- if (ret < 0) {
- ret = errno;
- goto out;
- }
- if (ret == 0)
- mop.mo_flags |= MO_IS_LOOP;
+ ret = osd_init();
+ if (ret)
+ 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 = is_lustre_target(&mop);
+ ret = osd_is_lustre(mop.mo_device, &mount_type);
if (ret == 0) {
fatal();
fprintf(stderr, "Device %s has not been formatted with "
ret = ENODEV;
goto out;
}
+ mop.mo_ldd.ldd_mount_type = mount_type;
- ret = read_local_files(&mop);
+ ret = osd_read_ldd(mop.mo_device, &mop.mo_ldd);
if (ret) {
fatal();
fprintf(stderr, "Failed to read previous Lustre data from %s "
"(%d)\n", mop.mo_device, ret);
goto out;
}
+ mop.mo_ldd.ldd_flags &= ~(LDD_F_WRITECONF | LDD_F_VIRGIN);
+
+ /* svname of the form lustre:OST1234 means never registered */
+ ret = strlen(mop.mo_ldd.ldd_svname);
+ if (mop.mo_ldd.ldd_svname[ret - 8] == ':') {
+ mop.mo_ldd.ldd_svname[ret - 8] = '-';
+ mop.mo_ldd.ldd_flags |= LDD_F_VIRGIN;
+ } else if (mop.mo_ldd.ldd_svname[ret - 8] == '=') {
+ mop.mo_ldd.ldd_svname[ret - 8] = '-';
+ mop.mo_ldd.ldd_flags |= LDD_F_WRITECONF;
+ }
+
if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
mop.mo_mgs_failnodes++;
goto out;
}
+ /* Stand alone MGS doesn't need a index */
+ if (!IS_MDT(ldd) && IS_MGS(ldd))
+ mop.mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
+
if ((mop.mo_ldd.ldd_flags & (LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) ==
(LDD_F_NEED_INDEX | LDD_F_UPGRADE14)) {
fatal();
ret = EINVAL;
goto out;
}
+
+ if (mop.mo_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)) {
+ mop.mo_ldd.ldd_flags &= ~LDD_F_NEED_INDEX;
+ mop.mo_ldd.ldd_svindex = 0;
+ }
+
#if 0
/*
* Comment out these 2 checks temporarily, since for multi-MDSes
goto out;
}
#endif
+ if ((IS_MDT(ldd) || IS_OST(ldd)) && mop.mo_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) */
- switch (ldd->ldd_mount_type) {
- case LDD_MT_EXT3:
- case LDD_MT_LDISKFS:
- case LDD_MT_LDISKFS2:
- strscat(default_mountopts, ",errors=remount-ro",
- sizeof(default_mountopts));
- if (IS_MDT(ldd) || IS_MGS(ldd))
- strscat(always_mountopts, ",user_xattr",
- sizeof(always_mountopts));
- /* NB: Files created while extents are enabled can only be read
- * if mounted using the ext4 or ldiskfs filesystem type. */
- if (IS_OST(ldd) &&
- (ldd->ldd_mount_type == LDD_MT_LDISKFS ||
- ldd->ldd_mount_type == LDD_MT_LDISKFS2)) {
- strscat(default_mountopts, ",extents,mballoc",
- sizeof(default_mountopts));
- }
- break;
- default:
- fatal();
- fprintf(stderr, "unknown fs type %d '%s'\n",
- ldd->ldd_mount_type,
- MT_STR(ldd));
- ret = EINVAL;
- goto out;
- }
+ ret = osd_prepare_lustre(&mop,
+ default_mountopts, sizeof(default_mountopts),
+ always_mountopts, sizeof(always_mountopts));
+ if (ret) {
+ fatal();
+ fprintf(stderr, "unable to prepare backend (%d)\n", ret);
+ goto out;
+ }
if (mountopts) {
trim_mountfsoptions(mountopts);
#ifndef TUNEFS /* mkfs.lustre */
/* Check whether the disk has already been formatted by mkfs.lustre */
if (!(mop.mo_flags & MO_FORCEFORMAT)) {
- ret = is_lustre_target(&mop);
+ ret = osd_is_lustre(mop.mo_device, &mount_type);
if (ret) {
fatal();
fprintf(stderr, "Device %s was previously formatted "
}
/* Format the backing filesystem */
- ret = make_lustre_backfs(&mop);
+ ret = osd_make_lustre(&mop);
if (ret != 0) {
fatal();
fprintf(stderr, "mkfs failed %d\n", ret);
goto out;
}
+#else
+ /* update svname with '=' to refresh config */
+ if (mop.mo_ldd.ldd_flags & LDD_F_WRITECONF) {
+ struct mount_opts opts;
+ opts.mo_ldd = mop.mo_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
/* Write our config files */
- ret = write_local_files(&mop);
+ ret = osd_write_ldd(&mop);
if (ret != 0) {
fatal();
fprintf(stderr, "failed to write local files\n");
out:
loop_cleanup(&mop);
+ osd_fini();
/* Fix any crazy return values from system() */
if (ret && ((ret & 255) == 0))