static int print_only = 0;
static int upgrade_to_18 = 0;
-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"
-#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"
+#define FSLIST_LDISKFS "ldiskfs"
+#define HAVE_FSLIST
+#ifdef HAVE_ZFS_OSD
+ #ifdef HAVE_FSLIST
+ #define FSLIST_ZFS "|zfs"
+ #else
+ #define FSLIST_ZFS "zfs"
+ #define HAVE_FSLIST
+ #endif
#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"
-#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;
-}
-
-/*================ utility functions =====================*/
-
-char *strscat(char *dst, char *src, int buflen) {
- dst[buflen - 1] = 0;
- if (strlen(dst) + strlen(src) >= buflen) {
- fprintf(stderr, "string buffer overflow (max %d): '%s' + '%s'"
- "\n", buflen, dst, src);
- exit(EOVERFLOW);
- }
- return strcat(dst, src);
-
-}
-
-char *strscpy(char *dst, char *src, int buflen) {
- dst[0] = 0;
- return strscat(dst, src, buflen);
-}
-
-inline unsigned int
-dev_major (unsigned long long int __dev)
-{
- return ((__dev >> 8) & 0xfff) | ((unsigned int) (__dev >> 32) & ~0xfff);
-}
-
-inline unsigned int
-dev_minor (unsigned long long int __dev)
-{
- return (__dev & 0xff) | ((unsigned int) (__dev >> 12) & ~0xff);
-}
-
-int get_os_version()
-{
- static int version = 0;
-
- if (!version) {
- int fd;
- char release[4] = "";
-
- fd = open("/proc/sys/kernel/osrelease", O_RDONLY);
- if (fd < 0) {
- fprintf(stderr, "%s: Warning: Can't resolve kernel "
- "version, assuming 2.6\n", progname);
- } else {
- if (read(fd, release, 4) < 0) {
- fprintf(stderr, "reading from /proc/sys/kernel"
- "/osrelease: %s\n", strerror(errno));
- close(fd);
- exit(-1);
- }
- close(fd);
- }
- if (strncmp(release, "2.4.", 4) == 0)
- version = 24;
- else
- version = 26;
- }
- return version;
-}
-
-static int check_mtab_entry(char *spec)
-{
- FILE *fp;
- struct mntent *mnt;
-
- fp = setmntent(MOUNTED, "r");
- if (fp == NULL)
- return(0);
-
- while ((mnt = getmntent(fp)) != NULL) {
- if (strcmp(mnt->mnt_fsname, spec) == 0) {
- endmntent(fp);
- fprintf(stderr, "%s: according to %s %s is "
- "already mounted on %s\n",
- progname, MOUNTED, spec, mnt->mnt_dir);
- return(EEXIST);
- }
- }
- endmntent(fp);
-
- return(0);
-}
-
-/*============ disk dev functions ===================*/
-
-/* Setup a file in the first unused loop_device */
-int loop_setup(struct mkfs_opts *mop)
-{
- char loop_base[20];
- char l_device[64];
- int i, ret = 0;
-
- /* Figure out the loop device names */
- if (!access("/dev/loop0", F_OK | R_OK)) {
- strcpy(loop_base, "/dev/loop\0");
- } else if (!access("/dev/loop/0", F_OK | R_OK)) {
- strcpy(loop_base, "/dev/loop/\0");
- } else {
- fprintf(stderr, "%s: can't access loop devices\n", progname);
- return EACCES;
- }
-
- /* Find unused loop device */
- for (i = 0; i < MAX_LOOP_DEVICES; i++) {
- char cmd[PATH_MAX];
- int cmdsz = sizeof(cmd);
-
- sprintf(l_device, "%s%d", loop_base, i);
- if (access(l_device, F_OK | R_OK))
- break;
- snprintf(cmd, cmdsz, "losetup %s > /dev/null 2>&1", l_device);
- ret = system(cmd);
-
- /* losetup gets 1 (ret=256) for non-set-up device */
- if (ret) {
- /* Set up a loopback device to our file */
- snprintf(cmd, cmdsz, "losetup %s %s", l_device,
- mop->mo_device);
- ret = run_command(cmd, cmdsz);
- if (ret == 256)
- /* someone else picked up this loop device
- * behind our back */
- continue;
- if (ret) {
- fprintf(stderr, "%s: error %d on losetup: %s\n",
- progname, ret, strerror(ret));
- return ret;
- }
- strscpy(mop->mo_loopdev, l_device,
- sizeof(mop->mo_loopdev));
- return ret;
- }
- }
+ #define FSLIST_ZFS ""
+#endif /* HAVE_ZFS_OSD */
- fprintf(stderr, "%s: out of loop devices!\n", progname);
- return EMFILE;
-}
+#ifndef HAVE_FSLIST
+ #error "no backing OSD types (ldiskfs or ZFS) are configured"
+#endif
-int loop_cleanup(struct mkfs_opts *mop)
-{
- char cmd[150];
- int ret = 1;
- if ((mop->mo_flags & MO_IS_LOOP) && *mop->mo_loopdev) {
- sprintf(cmd, "losetup -d %s", mop->mo_loopdev);
- ret = run_command(cmd, sizeof(cmd));
- }
- return ret;
-}
+#define FSLIST FSLIST_LDISKFS FSLIST_ZFS
-__u64 get_device_size(char* device)
+void usage(FILE *out)
{
- int ret, fd;
- __u64 size = 0;
-
- fd = open(device, O_RDONLY);
- if (fd < 0) {
- fprintf(stderr, "%s: cannot open %s: %s\n",
- progname, device, strerror(errno));
- return 0;
- }
-
-#ifdef BLKGETSIZE64
- /* size in bytes. bz5831 */
- ret = ioctl(fd, BLKGETSIZE64, (void*)&size);
+ 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"
#else
- {
- __u32 lsize = 0;
- /* size in blocks */
- ret = ioctl(fd, BLKGETSIZE, (void*)&lsize);
- size = (__u64)lsize * 512;
- }
+ "\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"
#endif
- close(fd);
- if (ret < 0) {
- fprintf(stderr, "%s: size ioctl failed: %s\n",
- progname, strerror(errno));
- return 0;
- }
-
- vprint("device size = "LPU64"MB\n", size >> 20);
- /* return value in KB */
- return size >> 10;
-}
-
-int loop_format(struct mkfs_opts *mop)
-{
- int fd;
-
- if (mop->mo_device_sz == 0) {
- fatal();
- fprintf(stderr, "loop device requires a --device-size= "
- "param\n");
- return EINVAL;
- }
-
- fd = creat(mop->mo_device, S_IRUSR|S_IWUSR);
- if (fd < 0) {
- fatal();
- fprintf(stderr, "%s: Unable to create backing store: %s\n",
- progname, strerror(errno));
- return errno;
- }
-
- if (ftruncate(fd, mop->mo_device_sz * 1024) != 0) {
- close(fd);
- fatal();
- fprintf(stderr, "%s: Unable to truncate backing store: %s\n",
- progname, strerror(errno));
- return errno;
- }
-
- close(fd);
- return 0;
+ "\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 =============*/
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;
+ mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
+ 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)
usage(stderr);
}
-int add_param(char *buf, char *key, char *val)
-{
- int end = sizeof(((struct lustre_disk_data *)0)->ldd_params);
- int start = strlen(buf);
- int keylen = 0;
-
- if (key)
- keylen = strlen(key);
- if (start + 1 + keylen + strlen(val) >= end) {
- fprintf(stderr, "%s: params are too long-\n%s %s%s\n",
- progname, buf, key ? key : "", val);
- return 1;
- }
-
- sprintf(buf + start, " %s%s", key ? key : "", val);
- return 0;
-}
-
/* from mount_lustre */
/* Get rid of symbolic hostnames for tcp, since kernel can't do lookups */
#define MAXNIDSTR 1024
sep = *s2;
*s2 = '\0';
nid = libcfs_str2nid(s1);
+ *s2 = sep;
if (nid == LNET_NID_ANY) {
fprintf(stderr, "%s: Can't parse NID '%s'\n",
}
i++;
}
+ if (i == LDD_MT_LAST) {
+ fprintf(stderr, "%s: invalid backend filesystem"
+ " type %s\n", progname, optarg);
+ return 1;
+ }
break;
}
case 'c':
}
}//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;
}
-/* Search for opt in mntlist, returning true if found.
- */
-static int in_mntlist(char *opt, char *mntlist)
-{
- char *ml, *mlp, *item, *ctx = NULL;
-
- if (!(ml = strdup(mntlist))) {
- fprintf(stderr, "%s: out of memory\n", progname);
- exit(1);
- }
- mlp = ml;
- while ((item = strtok_r(mlp, ",", &ctx))) {
- if (!strcmp(opt, item))
- break;
- mlp = NULL;
- }
- free(ml);
- return (item != NULL);
-}
-
-/* Issue a message on stderr for every item in wanted_mountopts that is not
- * present in mountopts. The justwarn boolean toggles between error and
- * warning message. Return an error count.
- */
-static int check_mountfsoptions(char *mountopts, char *wanted_mountopts,
- int justwarn)
-{
- char *ml, *mlp, *item, *ctx = NULL;
- int errors = 0;
-
- if (!(ml = strdup(wanted_mountopts))) {
- fprintf(stderr, "%s: out of memory\n", progname);
- exit(1);
- }
- mlp = ml;
- while ((item = strtok_r(mlp, ",", &ctx))) {
- if (!in_mntlist(item, mountopts)) {
- fprintf(stderr, "%s: %s mount option `%s' is missing\n",
- progname, justwarn ? "Warning: default"
- : "Error: mandatory", item);
- errors++;
- }
- mlp = NULL;
- }
- free(ml);
- return errors;
-}
-
-/* Trim embedded white space, leading and trailing commas from string s.
- */
-static void trim_mountfsoptions(char *s)
-{
- char *p;
-
- for (p = s; *p; ) {
- if (isspace(*p)) {
- memmove(p, p + 1, strlen(p + 1) + 1);
- continue;
- }
- p++;
- }
-
- while (s[0] == ',')
- memmove(&s[0], &s[1], strlen(&s[1]) + 1);
-
- p = s + strlen(s) - 1;
- while (p >= s && *p == ',')
- *p-- = '\0';
-}
-
int main(int argc, char *const argv[])
{
struct mkfs_opts mop;
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;
-
#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 "
goto out;
}
- 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);
+
if (strstr(mop.mo_ldd.ldd_params, PARAM_MGSNODE))
mop.mo_mgs_failnodes++;
print_ldd("Read previous values", &(mop.mo_ldd));
#endif
+ ret = osd_init();
+ if (ret)
+ return ret;
+
ret = parse_opts(argc, argv, &mop, &mountopts);
if (ret)
goto out;
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 (IS_OST(ldd) && (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 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);
goto out;
}
- if (check_mtab_entry(mop.mo_device))
- return(EEXIST);
+ 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) {
#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);
+ }
#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))
git://git.whamcloud.com / fs / lustre-release.git / blobdiff