/*
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2011 Whamcloud, Inc.
+ *
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#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 */
+/* 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>
struct mkfs_opts {
struct lustre_disk_data mo_ldd; /* to be written in MOUNT_DATA_FILE */
char mo_device[128]; /* disk device name */
- char mo_mkfsopts[128]; /* options to the backing-store mkfs */
char mo_loopdev[128]; /* in case a loop dev is needed */
+ char mo_mkfsopts[512]; /* options to the backing-store mkfs */
__u64 mo_device_sz; /* in KB */
int mo_stripe_count;
int mo_flags;
"\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 the lov)\n"
- /* FIXME implement 1.6.x
- "\t\t--configdev=<altdevice|file>: store configuration info\n"
- "\t\t\tfor this device on an alternate device\n"
- */
- "\t\t--comment=<user comment>: arbitrary user string (%d bytes)\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>[,<...>] : network(s) to restrict this ost/mdt to\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 : used for optimizing MDT inode size\n"
- "\t\t--iam-dir: make use of IAM directory format on backfs, incompatible with ext3.\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"
#else
"\t\t--erase-params : erase all old parameter settings\n"
"\t\t--nomgs: turn off MGS service on this MDT\n"
char release[4] = "";
fd = open("/proc/sys/kernel/osrelease", O_RDONLY);
- if (fd < 0)
+ if (fd < 0) {
fprintf(stderr, "%s: Warning: Can't resolve kernel "
"version, assuming 2.6\n", progname);
- else {
+ } else {
if (read(fd, release, 4) < 0) {
fprintf(stderr, "reading from /proc/sys/kernel"
- "/osrelease: %s\n", strerror(errno));
+ "/osrelease: %s\n", strerror(errno));
close(fd);
exit(-1);
}
E2FSPROGS, feature);
#if !(HAVE_LDISKFSPROGS)
fprintf(stderr, "Please install the latest version of e2fsprogs from\n"
- "http://downloads.lustre.org/public/tools/e2fsprogs/\n"
+ "http://downloads.whamcloud.com/public/e2fsprogs/latest/\n"
"to enable this feature.\n");
#endif
if (make_backfs)
* mke2fs to check for its support. */
static int is_e2fsprogs_feature_supp(const char *feature)
{
+ static char supp_features[4096] = "";
FILE *fp;
char cmd[PATH_MAX];
char imgname[] = "/tmp/test-img-XXXXXX";
int fd = -1;
- int ret = 0;
+ int ret = 1;
- snprintf(cmd, sizeof(cmd), "%s -c -R \"supported_features %s\" 2>&1",
- DEBUGFS, feature);
+ if (supp_features[0] == '\0') {
+ snprintf(cmd, sizeof(cmd), "%s -c -R supported_features 2>&1",
+ DEBUGFS);
- /* Using popen() instead of run_command() since debugfs does not return
- * proper error code if command is not supported */
- fp = popen(cmd, "r");
- if (!fp) {
- fprintf(stderr, "%s: %s\n", progname, strerror(errno));
- return 0;
- }
- ret = fread(cmd, 1, sizeof(cmd), fp);
- if (ret > 0) {
- if (strstr(cmd, feature) && !(strstr(cmd, "Unknown")))
+ /* Using popen() instead of run_command() since debugfs does
+ * not return proper error code if command is not supported */
+ fp = popen(cmd, "r");
+ if (!fp) {
+ fprintf(stderr, "%s: %s\n", progname, strerror(errno));
return 0;
+ }
+ ret = fread(supp_features, 1, sizeof(supp_features), fp);
+ fclose(fp);
}
+ if (ret > 0 && strstr(supp_features,
+ strncmp(feature, "-O ", 3) ? feature : feature+3))
+ return 0;
if ((fd = mkstemp(imgname)) < 0)
return -1;
+ else
+ close(fd);
- snprintf(cmd, sizeof(cmd), "%s -F -O %s %s 100 >/dev/null 2>&1",
+ snprintf(cmd, sizeof(cmd), "%s -F %s %s 100 >/dev/null 2>&1",
MKE2FS, feature, imgname);
/* run_command() displays the output of mke2fs when it fails for
* some feature, so use system() directly */
ret = system(cmd);
- if (fd >= 0)
- remove(imgname);
+ unlink(imgname);
return ret;
}
-static void disp_old_kernel_msg(char *feature)
+/**
+ * append_unique: append @key or @key=@val pair to @buf only if @key does not
+ * exists
+ * @buf: buffer to hold @key or @key=@val
+ * @prefix: prefix string before @key
+ * @key: key string
+ * @val: value string if it's a @key=@val pair
+ */
+static void append_unique(char *buf, char *prefix, char *key, char *val,
+ size_t maxbuflen)
{
- fprintf(stderr, "WARNING: ldiskfs filesystem does not support \"%s\" "
- "feature.\n\n", feature);
+ char *anchor, *end;
+ int len;
+
+ if (key == NULL)
+ return;
+
+ anchor = end = strstr(buf, key);
+ /* try to find exact match string in @buf */
+ while (end && *end != '\0' && *end != ',' && *end != ' ' && *end != '=')
+ ++end;
+ len = end - anchor;
+ if (anchor == NULL || strlen(key) != len ||
+ strncmp(anchor, key, len) != 0) {
+ if (prefix != NULL)
+ strscat(buf, prefix, maxbuflen);
+
+ strscat(buf, key, maxbuflen);
+ if (val != NULL) {
+ strscat(buf, "=", maxbuflen);
+ strscat(buf, val, maxbuflen);
+ }
+ }
}
-static void enable_default_backfs_features(struct mkfs_opts *mop)
+static void enable_default_ext4_features(struct mkfs_opts *mop, char *anchor,
+ size_t maxbuflen, int user_spec)
{
- struct utsname uts;
- int maj_high, maj_low, min;
- int ret;
-
- if (IS_MDT(&mop->mo_ldd))
- strscat(mop->mo_mkfsopts, " -O dir_index,extents,dirdata",
- sizeof(mop->mo_mkfsopts));
- else
- strscat(mop->mo_mkfsopts, " -O dir_index,extents",
- sizeof(mop->mo_mkfsopts));
+ if (IS_OST(&mop->mo_ldd)) {
+ append_unique(anchor, user_spec ? "," : " -O ",
+ "extents", NULL, sizeof(mop->mo_mkfsopts));
+ 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);
+ } else {
+ append_unique(anchor, user_spec ? "," : " -O ",
+ "uninit_bg", NULL, maxbuflen);
+ }
+ /* Multiple mount protection enabled only if failover node specified */
+ if (failover) {
+ if (is_e2fsprogs_feature_supp("-O mmp") == 0)
+ append_unique(anchor, ",", "mmp", NULL, maxbuflen);
+ else
+ disp_old_e2fsprogs_msg("mmp", 1);
+ }
- /* Upstream e2fsprogs called our uninit_groups feature uninit_bg,
- * check for both of them when testing e2fsprogs features. */
- if (is_e2fsprogs_feature_supp("uninit_groups") == 0)
- strscat(mop->mo_mkfsopts, ",uninit_groups",
- sizeof(mop->mo_mkfsopts));
- else if (is_e2fsprogs_feature_supp("uninit_bg") == 0)
- strscat(mop->mo_mkfsopts, ",uninit_bg",
- sizeof(mop->mo_mkfsopts));
- else
- disp_old_e2fsprogs_msg("uninit_bg", 1);
+ /* Allow more than 65000 subdirectories */
+ if (is_e2fsprogs_feature_supp("-O dir_nlink") == 0)
+ append_unique(anchor, ",", "dir_nlink", NULL, maxbuflen);
- ret = uname(&uts);
- if (ret)
+#ifdef HAVE_EXT4_LDISKFS
+ /* 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;
- sscanf(uts.release, "%d.%d.%d", &maj_high, &maj_low, &min);
- printf("%d %d %d\n", maj_high, maj_low, min);
+ /* Allow files larger than 2TB. Also needs LU-16, but not harmful. */
+ if (is_e2fsprogs_feature_supp("-O huge_file") == 0)
+ append_unique(anchor, ",", "huge_file", NULL, maxbuflen);
- /* Multiple mount protection is enabled only if failover node is
- * specified and if kernel version is higher than 2.6.9 */
- if (failover) {
- if (KERNEL_VERSION(maj_high, maj_low, min) >=
- KERNEL_VERSION(2,6,9)) {
- if (is_e2fsprogs_feature_supp("mmp") == 0)
- strscat(mop->mo_mkfsopts, ",mmp",
- sizeof(mop->mo_mkfsopts));
- else
- disp_old_e2fsprogs_msg("mmp", 1);
- } else {
- disp_old_kernel_msg("mmp");
+ /* Enable large block addresses if the LUN is over 2^32 blocks. */
+ if (mop->mo_device_sz / (L_BLOCK_SIZE >> 10) >= 0x100002000ULL &&
+ is_e2fsprogs_feature_supp("-O 64bit") == 0)
+ append_unique(anchor, ",", "64bit", 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") == 0) {
+ char tmp_buf[64];
+
+ append_unique(anchor, ",", "flex_bg", NULL, maxbuflen);
+
+ if (IS_OST(&mop->mo_ldd)) {
+ snprintf(tmp_buf, sizeof(tmp_buf), " -G %u",
+ (1 << 20) / L_BLOCK_SIZE);
+ strscat(anchor, tmp_buf, maxbuflen);
}
}
+ /* Don't add any more "-O" options here, see last comment above */
+#endif
}
+
+/**
+ * moveopts_to_end: find the option string, move remaining strings to
+ * where option string starts, and append the option
+ * string at the end
+ * @start: where the option string starts before the move
+ * RETURN: where the option string starts after the move
+ */
+static char *moveopts_to_end(char *start)
+{
+ char save[512];
+ char *end, *idx;
+
+ /* skip whitespace before options */
+ end = start + 2;
+ while (*end == ' ')
+ ++end;
+
+ /* find end of option characters */
+ while (*end != ' ' && *end != '\0')
+ ++end;
+
+ /* save options */
+ strncpy(save, start, end - start);
+ save[end - start] = '\0';
+
+ /* move remaining options up front */
+ if (*end)
+ memmove(start, end, strlen(end));
+ *(start + strlen(end)) = '\0';
+
+ /* append the specified options */
+ if (*(start + strlen(start) - 1) != ' ')
+ strcat(start, " ");
+ idx = start + strlen(start);
+ strcat(start, save);
+
+ return idx;
+}
+
/* Build fs according to type */
int make_lustre_backfs(struct mkfs_opts *mop)
{
__u64 device_sz = mop->mo_device_sz, block_count = 0;
char mkfs_cmd[PATH_MAX];
char buf[64];
+ char *start;
char *dev;
- int ret = 0;
+ int ret = 0, ext_opts = 0;
+ size_t maxbuflen;
if (!(mop->mo_flags & MO_IS_LOOP)) {
mop->mo_device_sz = get_device_size(mop->mo_device);
return EINVAL;
}
block_count = mop->mo_device_sz / (L_BLOCK_SIZE >> 10);
+ /* 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 size. Bug 22906 */
+ if (block_count > 0xffffffffULL && block_count < 0x100002000ULL)
+ block_count = 0xffffffffULL;
}
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)) {
+ long inode_size = 0;
+
/* Journal size in MB */
if (strstr(mop->mo_mkfsopts, "-J") == NULL) {
/* Choose our own default journal size */
}
}
- /* Bytes_per_inode: disk size / num inodes */
- if (strstr(mop->mo_mkfsopts, "-i") == NULL) {
- long bytes_per_inode = 0;
-
- if (IS_MDT(&mop->mo_ldd))
- bytes_per_inode = 4096;
-
- /* Allocate fewer inodes on large OST devices. Most
- filesystems can be much more aggressive than even
- this. */
- if ((IS_OST(&mop->mo_ldd) && (device_sz > 100000000)))
- bytes_per_inode = 16384; /* > 100 Gb device */
-
-
- if (bytes_per_inode > 0) {
- sprintf(buf, " -i %ld", bytes_per_inode);
- strscat(mop->mo_mkfsopts, buf,
- sizeof(mop->mo_mkfsopts));
- }
- }
-
/* Inode size (for extended attributes). The LOV EA size is
* 32 (EA hdr) + 32 (lov_mds_md) + stripes * 24 (lov_ost_data),
* and we want some margin above that for ACLs, other EAs... */
if (strstr(mop->mo_mkfsopts, "-I") == NULL) {
- long inode_size = 0;
if (IS_MDT(&mop->mo_ldd)) {
if (mop->mo_stripe_count > 72)
inode_size = 512; /* bz 7241 */
- /* cray stripes across all osts (>60) */
+ /* see also "-i" below for EA blocks */
else if (mop->mo_stripe_count > 32)
inode_size = 2048;
else if (mop->mo_stripe_count > 10)
else
inode_size = 512;
} else if (IS_OST(&mop->mo_ldd)) {
- /* now as we store fids in EA on OST we need
- to make inode bigger */
+ /* We store MDS FID and OST objid in EA on OST
+ * we need to make inode bigger as well. */
inode_size = 256;
}
}
}
+ /* 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 > 72) {
+ 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 (device_sz > (16ULL << 30))
+ bytes_per_inode = 1024 * 1024;
+ /* OST > 4TB assume average file size 512kB */
+ else if (device_sz > (4ULL << 30))
+ bytes_per_inode = 512 * 1024;
+ /* OST > 1TB assume average file size 256kB */
+ else if (device_sz > (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 (device_sz > (10ULL << 20))
+ bytes_per_inode = 69905;
+ }
+
+ if (bytes_per_inode > 0) {
+ sprintf(buf, " -i %ld", bytes_per_inode);
+ strscat(mop->mo_mkfsopts, buf,
+ sizeof(mop->mo_mkfsopts));
+ }
+ }
+
if (verbose < 2) {
strscat(mop->mo_mkfsopts, " -q",
sizeof(mop->mo_mkfsopts));
}
- if (strstr(mop->mo_mkfsopts, "-O") == NULL)
- enable_default_backfs_features(mop);
+ /* start handle -O mkfs options */
+ if ((start = strstr(mop->mo_mkfsopts, "-O")) != NULL) {
+ 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);
+ 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);
+ enable_default_ext4_features(mop, start, maxbuflen, 0);
+ }
+ /* end handle -O mkfs options */
+
+ /* start handle -E mkfs options */
+ if ((start = strstr(mop->mo_mkfsopts, "-E")) != NULL) {
+ 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 (IS_OST(&mop->mo_ldd) && mop->mo_device_sz > 100 * 1024 &&
+ mop->mo_device_sz * 1024 / L_BLOCK_SIZE <= 0xffffffffULL) {
+ unsigned group_blocks = L_BLOCK_SIZE * 8;
+ unsigned desc_per_block = L_BLOCK_SIZE / 32;
+ unsigned 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") == 0)
+ 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. */
snprintf(mkfs_cmd, sizeof(mkfs_cmd),
"%s -j -b %d -L %s ", MKE2FS, L_BLOCK_SIZE,
mop->mo_ldd.ldd_svname);
- } else if (mop->mo_ldd.ldd_mount_type == LDD_MT_REISERFS) {
- long journal_sz = 0; /* FIXME default journal size */
- if (journal_sz > 0) {
- sprintf(buf, " --journal_size %ld", journal_sz);
- strscat(mop->mo_mkfsopts, buf,
- sizeof(mop->mo_mkfsopts));
- }
- snprintf(mkfs_cmd, sizeof(mkfs_cmd), "mkreiserfs -ff ");
} else {
fprintf(stderr,"%s: unsupported fs type: %d (%s)\n",
progname, mop->mo_ldd.ldd_mount_type,
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",
+ printf(" (%s%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_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);
printf("Parameters:%s\n", ldd->ldd_params);
vprint("Reading %s\n", MOUNT_DATA_FILE);
num_read = fread(&mop->mo_ldd, sizeof(mop->mo_ldd), 1, filep);
if (num_read < 1 && ferror(filep)) {
- fprintf(stderr, "%s: Unable to read from file (%s): %s\n",
+ fprintf(stderr, "%s: Unable to read from file %s: %s\n",
progname, filepnm, strerror(errno));
goto out_close;
}
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");
- if (get_os_version() == 24)
- mop->mo_ldd.ldd_mount_type = LDD_MT_EXT3;
- else
- mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
+ mop->mo_ldd.ldd_mount_type = LDD_MT_LDISKFS;
mop->mo_ldd.ldd_svindex = INDEX_UNASSIGNED;
mop->mo_stripe_count = 1;
nid = libcfs_str2nid(s1);
if (nid == LNET_NID_ANY) {
- fprintf(stderr, "%s: Can't parse NID '%s'\n", progname, s1);
+ fprintf(stderr, "%s: Can't parse NID '%s'\n",
+ progname, s1);
free(converted);
return NULL;
}
{"print", 0, 0, 'n'},
{"quiet", 0, 0, 'q'},
{"reformat", 0, 0, 'r'},
+ {"servicenode", 1, 0, 's'},
{"verbose", 0, 0, 'v'},
{"writeconf", 0, 0, 'w'},
{"upgrade_to_18", 0, 0, 'U'},
{"network", 1, 0, 't'},
{0, 0, 0, 0}
};
- char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqru:vw";
+ char *optstring = "b:c:C:d:ef:Ghi:k:L:m:MnNo:Op:Pqrs:t:Uu:vw";
int opt;
int rc, longidx;
+ int failnode_set = 0, servicenode_set = 0;
while ((opt = getopt_long(argc, argv, optstring, long_opt, &longidx)) !=
EOF) {
/* Must update the mgs logs */
mop->mo_ldd.ldd_flags |= LDD_F_UPDATE;
break;
- case 'f': {
- char *nids = convert_hostnames(optarg);
+ 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,
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;
+ }
failover = 1;
break;
}
switch (ldd->ldd_mount_type) {
case LDD_MT_EXT3:
case LDD_MT_LDISKFS:
- case LDD_MT_LDISKFS2: {
+ case LDD_MT_LDISKFS2:
strscat(default_mountopts, ",errors=remount-ro",
sizeof(default_mountopts));
if (IS_MDT(ldd) || IS_MGS(ldd))
- strscat(always_mountopts, ",iopen_nopriv,user_xattr",
+ strscat(always_mountopts, ",user_xattr",
sizeof(always_mountopts));
- if ((get_os_version() == 24) && IS_OST(ldd))
- strscat(always_mountopts, ",asyncdel",
- sizeof(always_mountopts));
- /* NB: Files created while extents are enabled cannot be read
- if mounted with a kernel that doesn't include the Lustre ldiskfs
- patches! */
+ /* 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)) {
sizeof(default_mountopts));
}
break;
- }
- case LDD_MT_SMFS: {
- mop.mo_flags |= MO_IS_LOOP;
- sprintf(always_mountopts, ",type=ext3,dev=%s",
- mop.mo_device);
- break;
- }
- default: {
+ default:
fatal();
fprintf(stderr, "unknown fs type %d '%s'\n",
ldd->ldd_mount_type,
ret = EINVAL;
goto out;
}
- }
if (mountopts) {
trim_mountfsoptions(mountopts);