#define SSM_CMD_COMMON(cmd) \
"usage: "cmd" [--component-end|-E <comp_end>]\n" \
" [--stripe-count|-c <stripe_count>]\n" \
+ " [--overstripe-count|-C <stripe_count>]\n" \
" [--stripe-index|-i <start_ost_idx>]\n" \
" [--stripe-size|-S <stripe_size>]\n" \
" [--layout|-L <pattern>]\n" \
#define SSM_HELP_COMMON \
"\tstripe_count: Number of OSTs to stripe over (0=fs default, -1 all)\n" \
+ "\t Using -C instead of -c allows overstriping, which\n" \
+ "\t will place more than one stripe per OST if\n" \
+ "\t stripe_count is greater than the number of OSTs\n" \
"\tstart_ost_idx: OST index of first stripe (-1=default round robin)\n"\
"\tstripe_size: Number of bytes on each OST (0=fs default)\n" \
"\t Can be specified with K, M or G (for KB, MB, GB\n" \
"\tmdt_hash: hash type of the striped directory. mdt types:\n" \
" fnv_1a_64 FNV-1a hash algorithm (default)\n" \
" all_char sum of characters % MDT_COUNT (not recommended)\n" \
+ " space create subdirectories with balanced space usage\n" \
"\tdefault_stripe: set default dirstripe of the directory\n" \
"\tmode: the file access permission of the directory (octal)\n" \
"To create dir with a foreign (free format) layout :\n" \
"layout\nto another (may be not safe with concurent writes).\n"
"usage: migrate "
"[--stripe-count|-c] <stripe_count>\n"
+ "[--overstripe-count|-C] <stripe_count>\n"
" [--stripe-index|-i] <start_ost_index>\n"
" [--stripe-size|-S] <stripe_size>\n"
" [--pool|-p] <pool_name>\n"
" [--non-direct|-D]\n"
" <file|directory>\n"
"\tstripe_count: number of OSTs to stripe a file over\n"
+ "\t Using -C instead of -c allows overstriping, which\n"
+ "\t will place more than one stripe per OST if\n"
+ "\t stripe_count is greater than the number of OSTs\n"
"\tstripe_ost_index: index of the first OST to stripe a file over\n"
"\tstripe_size: number of bytes to store before moving to the next OST\n"
"\tpool_name: name of the predefined pool of OSTs\n"
* indices and ranges, for example "1,2-4,7". Add the indices into the
* \a tgts array and remove duplicates.
*
- * \param[out] tgts array to store indices in
- * \param[in] size size of \a tgts array
- * \param[in] offset starting index in \a tgts
- * \param[in] arg string containing OST index list
+ * \param[out] tgts array to store indices in
+ * \param[in] size size of \a tgts array
+ * \param[in] offset starting index in \a tgts
+ * \param[in] arg string containing OST index list
+ * \param[in/out] overstriping index list may contain duplicates
*
* \retval positive number of indices in \a tgts
* \retval -EINVAL unable to parse \a arg
*/
-static int parse_targets(__u32 *tgts, int size, int offset, char *arg)
+static int parse_targets(__u32 *tgts, int size, int offset, char *arg,
+ unsigned long long *pattern)
{
int rc;
int nr = offset;
int slots = size - offset;
char *ptr = NULL;
+ bool overstriped = false;
bool end_of_loop;
if (arg == NULL)
end_of_loop = false;
while (!end_of_loop) {
- int start_index;
- int end_index;
+ int start_index = 0;
+ int end_index = 0;
int i;
char *endptr = NULL;
/* remove duplicate */
for (j = 0; j < offset; j++) {
- if (tgts[j] == i)
- break;
+ if (tgts[j] == i && pattern &&
+ *pattern == LLAPI_LAYOUT_OVERSTRIPING)
+ overstriped = true;
+ else if (tgts[j] == i)
+ return -EINVAL;
}
- if (j == offset) { /* no duplicate */
+
+ j = offset;
+
+ if (j == offset) { /* check complete */
tgts[nr++] = i;
--slots;
}
}
+
if (slots == 0 && i < end_index)
break;
if (!end_of_loop && ptr != NULL)
*ptr = ',';
+ if (!overstriped && pattern)
+ *pattern = LLAPI_LAYOUT_DEFAULT;
+
return rc < 0 ? rc : nr;
}
}
/* Data-on-MDT component has always single stripe up to end */
lsa->lsa_stripe_size = lsa->lsa_comp_end;
+ } else if (lsa->lsa_pattern == LLAPI_LAYOUT_OVERSTRIPING) {
+ rc = llapi_layout_pattern_set(layout, lsa->lsa_pattern);
+ if (rc) {
+ fprintf(stderr, "Set stripe pattern %#llx failed. %s\n",
+ lsa->lsa_pattern, strerror(errno));
+ return rc;
+ }
}
rc = llapi_layout_stripe_size_set(layout, lsa->lsa_stripe_size);
} else if (!strcmp(string, "pattern")) {
if (!strcmp(node->cy_valuestring, "mdt"))
lsa->lsa_pattern = LLAPI_LAYOUT_MDT;
+ if (!strcmp(node->cy_valuestring,
+ "raid0,overstriped"))
+ lsa->lsa_pattern =
+ LLAPI_LAYOUT_OVERSTRIPING;
} else if (!strcmp(string, "lcme_flags")) {
rc = comp_str2flags(node->cy_valuestring,
&lsa->lsa_comp_flags,
{ .val = 'c', .name = "stripe-count", .has_arg = required_argument},
{ .val = 'c', .name = "stripe_count", .has_arg = required_argument},
{ .val = 'c', .name = "mdt-count", .has_arg = required_argument},
-/* find { .val = 'C', .name = "ctime", .has_arg = required_argument }*/
+ { .val = 'C', .name = "overstripe-count",
+ .has_arg = required_argument},
{ .val = 'd', .name = "delete", .has_arg = no_argument},
{ .val = 'd', .name = "destroy", .has_arg = no_argument},
/* --non-direct is only valid in migrate mode */
snprintf(cmd, sizeof(cmd), "%s %s", progname, argv[0]);
progname = cmd;
while ((c = getopt_long(argc, argv,
- "bc:dDE:f:H:i:I:m:N::no:p:L:s:S:vx:y:",
+ "bc:C:dDE:f:H:i:I:m:N::no:p:L:s:S:vx:y:",
long_opts, NULL)) >= 0) {
switch (c) {
case 0:
}
migration_block = true;
break;
+ case 'C':
+ lsa.lsa_pattern = LLAPI_LAYOUT_OVERSTRIPING;
+ /* fall through */
case 'c':
lsa.lsa_stripe_count = strtoul(optarg, &end, 0);
if (*end != '\0') {
migrate_mdt_mode = true;
lsa.lsa_nr_tgts = parse_targets(tgts,
sizeof(tgts) / sizeof(__u32),
- lsa.lsa_nr_tgts, optarg);
+ lsa.lsa_nr_tgts, optarg, NULL);
if (lsa.lsa_nr_tgts < 0) {
fprintf(stderr,
"%s %s: invalid MDT target(s) '%s'\n",
fprintf(stderr, "warning: '--ost-list' is "
"deprecated, use '--ost' instead\n");
#endif
+ /* -o allows overstriping, and must note it because
+ * parse_targets is shared with MDT striping, which
+ * does not allow duplicates
+ */
+ lsa.lsa_pattern = LLAPI_LAYOUT_OVERSTRIPING;
lsa.lsa_nr_tgts = parse_targets(tgts,
sizeof(tgts) / sizeof(__u32),
- lsa.lsa_nr_tgts, optarg);
+ lsa.lsa_nr_tgts, optarg,
+ &lsa.lsa_pattern);
if (lsa.lsa_nr_tgts < 0) {
fprintf(stderr,
"%s %s: invalid OST target(s) '%s'\n",
lmu->lum_hash_type = LMV_HASH_TYPE_FNV_1A_64;
if (lsa.lsa_pool_name)
strncpy(lmu->lum_pool_name, lsa.lsa_pool_name,
- sizeof(lmu->lum_pool_name));
+ sizeof(lmu->lum_pool_name) - 1);
if (lsa.lsa_nr_tgts > 1) {
int i;
param->lsp_stripe_offset = -1;
else
param->lsp_stripe_offset = lsa.lsa_stripe_off;
+ param->lsp_stripe_pattern =
+ llapi_pattern_to_lov(lsa.lsa_pattern);
+ if (param->lsp_stripe_pattern == EINVAL) {
+ fprintf(stderr, "error: %s: invalid stripe pattern\n",
+ argv[0]);
+ free(param);
+ goto usage_error;
+ }
param->lsp_pool = lsa.lsa_pool_name;
param->lsp_is_specific = false;
if (lsa.lsa_nr_tgts > 0) {
*layout |= LOV_PATTERN_RAID0;
else if (strcmp(layout_name, "mdt") == 0)
*layout |= LOV_PATTERN_MDT;
+ else if (strcmp(layout_name, "overstriping") == 0)
+ *layout |= LOV_PATTERN_OVERSTRIPING;
else
return -1;
}
return (ratio - (int)ratio) > 0 ? (int)(ratio + 1) : (int)ratio;
}
+/* This is only used to reflect various problem states for lfs df, so we only
+ * translate the flags reflecting those states.
+ */
+static char obd_statfs_state_names[] = {
+ [OS_STATE_DEGRADED] = 'D',
+ [OS_STATE_READONLY] = 'R',
+ [OS_STATE_NOPRECREATE] = 'N',
+ [OS_STATE_ENOSPC] = 'S',
+ [OS_STATE_ENOINO] = 'I',
+};
+
+static char obd_statfs_state2char(int s)
+{
+ /* Not an error state, do not print here */
+ if (s == OS_STATE_NONROT)
+ return '\0';
+ /* Unknown name */
+ if (s > ARRAY_SIZE(obd_statfs_state_names)/sizeof(char) ||
+ obd_statfs_state_names[s] == 0)
+ return '?';
+
+ return obd_statfs_state_names[s];
+}
+
static int showdf(char *mntdir, struct obd_statfs *stat,
char *uuid, enum mntdf_flags flags,
char *type, int index, int rc)
printf("[%s:%d]", type, index);
if (stat->os_state) {
- /*
- * Each character represents the matching
- * OS_STATE_* bit.
- */
- const char state_names[] = "DRSI";
- __u32 state;
- __u32 i;
+ uint32_t state;
+ uint32_t i;
printf(" ");
- for (i = 0, state = stat->os_state;
- state && i < sizeof(state_names); i++) {
- if (!(state & (1 << i)))
+ for (i = 0, state = stat->os_state; state != 0; i++) {
+ uint32_t mask = 1 << i;
+ if (!(state & mask))
continue;
- printf("%c", state_names[i]);
- state ^= 1 << i;
+ printf("%c", obd_statfs_state2char(mask));
+ state &= ~mask;
}
}
{ .st_op = LL_STATFS_LOV, .st_name = "OST" },
{ .st_name = NULL } };
struct ll_stat_type *tp;
+ __u64 ost_files = 0;
__u64 ost_ffree = 0;
__u32 index;
__u32 type;
sum.os_ffree += stat_buf.os_ffree;
sum.os_files += stat_buf.os_files;
} else /* if (tp->st_op == LL_STATFS_LOV) */ {
+ ost_files += stat_buf.os_files;
ost_ffree += stat_buf.os_ffree;
}
sum.os_blocks += stat_buf.os_blocks *
close(fd);
- /* If we don't have as many objects free on the OST as inodes
- * on the MDS, we reduce the total number of inodes to
- * compensate, so that the "inodes in use" number is correct.
- * Matches ll_statfs_internal() so the results are consistent. */
- if (ost_ffree < sum.os_ffree) {
+ /* If we have _some_ OSTs, but don't have as many free objects on the
+ * OST as inodes on the MDTs, reduce the reported number of inodes
+ * to compensate, so that the "inodes in use" number is correct.
+ * This should be kept in sync with ll_statfs_internal().
+ */
+ if (ost_files && ost_ffree < sum.os_ffree) {
sum.os_files = (sum.os_files - sum.os_ffree) + ost_ffree;
sum.os_ffree = ost_ffree;
}
/* functions */
static int lfs_setdirstripe(int argc, char **argv)
{
- char *dname;
- int result;
- struct lfs_setstripe_args lsa = { 0 };
- struct llapi_stripe_param *param = NULL;
- __u32 mdts[LMV_MAX_STRIPE_COUNT] = { 0 };
- char *end;
- int c;
- char *mode_opt = NULL;
- bool default_stripe = false;
- mode_t mode = S_IRWXU | S_IRWXG | S_IRWXO;
- mode_t previous_mode = 0;
- bool delete = false;
- struct ll_statfs_buf *lsb = NULL;
- char mntdir[PATH_MAX] = "";
- bool auto_distributed = false;
- bool foreign_mode = false;
- char *xattr = NULL;
- __u32 type = LU_FOREIGN_TYPE_DAOS, flags = 0;
-
+ char *dname;
+ struct lfs_setstripe_args lsa = { 0 };
+ struct llapi_stripe_param *param = NULL;
+ __u32 mdts[LMV_MAX_STRIPE_COUNT] = { 0 };
+ char *end;
+ int c;
+ char *mode_opt = NULL;
+ bool default_stripe = false;
+ bool delete = false;
+ bool auto_distributed = false;
+ bool foreign_mode = false;
+ mode_t mode = S_IRWXU | S_IRWXG | S_IRWXO;
+ mode_t previous_mode = 0;
+ struct ll_statfs_buf *lsb = NULL;
+ char mntdir[PATH_MAX] = "";
+ char *xattr = NULL;
+ __u32 type = LU_FOREIGN_TYPE_DAOS, flags = 0;
struct option long_opts[] = {
{ .val = 'c', .name = "count", .has_arg = required_argument },
{ .val = 'c', .name = "mdt-count", .has_arg = required_argument },
/* setstripe { .val = 'y', .name = "yaml", .has_arg = no_argument }, */
{ .val = 'x', .name = "xattr", .has_arg = required_argument },
{ .name = NULL } };
+ int result = 0;
setstripe_args_init(&lsa);
#endif
lsa.lsa_nr_tgts = parse_targets(mdts,
sizeof(mdts) / sizeof(__u32),
- lsa.lsa_nr_tgts, optarg);
+ lsa.lsa_nr_tgts, optarg, NULL);
if (lsa.lsa_nr_tgts < 0) {
fprintf(stderr,
"%s %s: invalid MDT target(s) '%s'\n",
memcpy(param->lsp_tgts, mdts, sizeof(*mdts) * lsa.lsa_nr_tgts);
}
+ if (!default_stripe && lsa.lsa_pattern == LMV_HASH_TYPE_SPACE) {
+ fprintf(stderr, "%s %s: can only specify -H space with -D\n",
+ progname, argv[0]);
+ free(param);
+ return CMD_HELP;
+ }
+
dname = argv[optind];
do {
if (default_stripe) {
result = llapi_dir_set_default_lmv(dname, param);
- } else {
- /* if current \a dname isn't under the same \a mntdir
- * as the last one, and the last one was
- * auto-distributed, restore \a param.
- */
- if (mntdir[0] != '\0' &&
- strncmp(dname, mntdir, strlen(mntdir)) &&
- auto_distributed) {
- param->lsp_is_specific = false;
- param->lsp_stripe_offset = -1;
- auto_distributed = false;
- }
-
- if (!param->lsp_is_specific &&
- param->lsp_stripe_offset == -1) {
- char path[PATH_MAX] = "";
+ if (result)
+ fprintf(stderr,
+ "%s setdirstripe: cannot set default stripe on dir '%s': %s\n",
+ progname, dname, strerror(-result));
+ continue;
+ }
- if (!lsb) {
- lsb = malloc(sizeof(*lsb));
- if (!lsb) {
- result = -ENOMEM;
- break;
- }
- }
- lsb->sb_count = 0;
+ /*
+ * if current \a dname isn't under the same \a mntdir as the
+ * last one, and the last one was auto-distributed, restore
+ * \a param.
+ */
+ if (mntdir[0] != '\0' &&
+ strncmp(dname, mntdir, strlen(mntdir)) &&
+ auto_distributed) {
+ param->lsp_is_specific = false;
+ param->lsp_stripe_offset = -1;
+ auto_distributed = false;
+ }
- /* use mntdir for dirname() temporarily */
- strncpy(mntdir, dname, sizeof(mntdir));
- if (!realpath(dirname(mntdir), path)) {
- result = -errno;
- fprintf(stderr,
- "error: invalid path '%s': %s\n",
- argv[optind], strerror(errno));
- break;
- }
- mntdir[0] = '\0';
+ /*
+ * TODO: when MDT can allocate object with QoS (LU-9435), below
+ * code should be removed, instead we should let LMV to allocate
+ * the starting MDT object, and then let LOD allocate other MDT
+ * objects.
+ */
+ if (!param->lsp_is_specific && param->lsp_stripe_offset == -1) {
+ char path[PATH_MAX] = "";
- result = llapi_search_mounts(path, 0, mntdir,
- NULL);
- if (result < 0 || mntdir[0] == '\0') {
- fprintf(stderr,
- "No suitable Lustre mount found\n");
+ if (!lsb) {
+ lsb = malloc(sizeof(*lsb));
+ if (!lsb) {
+ result = -ENOMEM;
break;
}
+ }
+ lsb->sb_count = 0;
- result = mntdf(mntdir, NULL, NULL, 0,
- LL_STATFS_LMV, lsb);
- if (result < 0)
- break;
+ /* use mntdir for dirname() temporarily */
+ strncpy(mntdir, dname, sizeof(mntdir));
+ if (!realpath(dirname(mntdir), path)) {
+ result = -errno;
+ fprintf(stderr,
+ "error: invalid path '%s': %s\n",
+ argv[optind], strerror(errno));
+ break;
+ }
+ mntdir[0] = '\0';
- if (param->lsp_stripe_count > lsb->sb_count) {
- fprintf(stderr,
- "error: stripe count %d is too big\n",
- param->lsp_stripe_count);
- result = -ERANGE;
- break;
- }
+ result = llapi_search_mounts(path, 0, mntdir, NULL);
+ if (result < 0 || mntdir[0] == '\0') {
+ fprintf(stderr,
+ "No suitable Lustre mount found\n");
+ break;
+ }
- qsort(lsb->sb_buf, lsb->sb_count,
- sizeof(struct ll_statfs_data),
- ll_statfs_data_comp);
+ result = mntdf(mntdir, NULL, NULL, 0, LL_STATFS_LMV,
+ lsb);
+ if (result < 0)
+ break;
- auto_distributed = true;
+ if (param->lsp_stripe_count > lsb->sb_count) {
+ fprintf(stderr,
+ "error: stripe count %d is too big\n",
+ param->lsp_stripe_count);
+ result = -ERANGE;
+ break;
}
- if (auto_distributed) {
- int r;
- int nr = MAX(param->lsp_stripe_count,
- lsb->sb_count / 2);
+ qsort(lsb->sb_buf, lsb->sb_count,
+ sizeof(struct ll_statfs_data),
+ ll_statfs_data_comp);
- /* don't use server whose usage is above 90% */
- while (nr != param->lsp_stripe_count &&
- obd_statfs_ratio(&lsb->sb_buf[nr].sd_st,
- false) > 90)
- nr = MAX(param->lsp_stripe_count,
- nr / 2);
+ auto_distributed = true;
+ }
- /* get \a r between [0, nr) */
- r = rand() % nr;
+ if (auto_distributed) {
+ int r;
+ int nr = MAX(param->lsp_stripe_count,
+ lsb->sb_count / 2);
- param->lsp_stripe_offset =
- lsb->sb_buf[r].sd_index;
- if (param->lsp_stripe_count > 1) {
- int i = 0;
+ /* don't use server whose usage is above 90% */
+ while (nr != param->lsp_stripe_count &&
+ obd_statfs_ratio(&lsb->sb_buf[nr].sd_st, false) >
+ 90)
+ nr = MAX(param->lsp_stripe_count, nr / 2);
- param->lsp_is_specific = true;
- for (; i < param->lsp_stripe_count; i++)
- param->lsp_tgts[(i + r) % nr] =
- lsb->sb_buf[i].sd_index;
- }
- }
+ /* get \a r between [0, nr) */
+ r = rand() % nr;
+
+ param->lsp_stripe_offset = lsb->sb_buf[r].sd_index;
+ if (param->lsp_stripe_count > 1) {
+ int i = 0;
- result = llapi_dir_create(dname, mode, param);
+ param->lsp_is_specific = true;
+ for (; i < param->lsp_stripe_count; i++)
+ param->lsp_tgts[(i + r) % nr] =
+ lsb->sb_buf[i].sd_index;
+ }
}
- if (result) {
+ result = llapi_dir_create(dname, mode, param);
+ if (result)
fprintf(stderr,
- "%s setdirstripe: cannot create stripe dir '%s': %s\n",
+ "%s setdirstripe: cannot create dir '%s': %s\n",
progname, dname, strerror(-result));
- break;
- }
- dname = argv[++optind];
- } while (dname != NULL);
+ } while (!result && (dname = argv[++optind]));
if (mode_opt != NULL)
umask(previous_mode);