* Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
*
- * Copyright (c) 2011, 2014, Intel Corporation.
+ * Copyright (c) 2011, 2015, Intel Corporation.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#include <lustre_ver.h>
#include <lustre_param.h>
+#ifndef ARRAY_SIZE
+# define ARRAY_SIZE(a) ((sizeof(a)) / (sizeof((a)[0])))
+#endif /* !ARRAY_SIZE */
+
/* all functions */
static int lfs_setstripe(int argc, char **argv);
static int lfs_find(int argc, char **argv);
static int lfs_hsm_cancel(int argc, char **argv);
static int lfs_swap_layouts(int argc, char **argv);
static int lfs_mv(int argc, char **argv);
+static int lfs_ladvise(int argc, char **argv);
/* Setstripe and migrate share mostly the same parameters */
#define SSM_CMD_COMMON(cmd) \
#define MIGRATE_USAGE \
SSM_CMD_COMMON("migrate ") \
" [--block|-b]\n" \
+ " [--non-block|-n]\n" \
" <filename>\n" \
SSM_HELP_COMMON \
"\n" \
- "\tblock: Block file access during data migration\n" \
+ "\tblock: Block file access during data migration (default)\n" \
+ "\tnon-block: Abort migrations if concurrent access is detected\n" \
static const char *progname;
static bool file_lease_supported = true;
{"swap_layouts", lfs_swap_layouts, 0, "Swap layouts between 2 files.\n"
"usage: swap_layouts <path1> <path2>"},
{"migrate", lfs_setstripe, 0,
- "migrate file/directory between MDTs, or migrate file from one OST "
+ "migrate a directory between MDTs.\n"
+ "usage: migrate --mdt-index <mdt_idx> [--verbose|-v] "
+ "<directory>\n"
+ "\tmdt_idx: index of the destination MDT\n"
+ "\n"
+ "migrate file objects from one OST "
"layout\nto another (may be not safe with concurent writes).\n"
- "usage: migrate [--mdt-index|-m <mdt_idx>] <directory|filename>]\n"
- "\tmdt_idx: MDT index to migrate to\n"
- " or\n"
- MIGRATE_USAGE},
+ "usage: migrate "
+ "[--stripe-count|-c] <stripe_count>\n"
+ " [--stripe-index|-i] <start_ost_index>\n"
+ " [--stripe-size|-S] <stripe_size>\n"
+ " [--pool|-p] <pool_name>\n"
+ " [--ost-list|-o] <ost_indices>\n"
+ " [--block|-b]\n"
+ " [--non-block|-n]\n"
+ " <file|directory>\n"
+ "\tstripe_count: number of OSTs to stripe a file over\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"
+ "\tost_indices: OSTs to stripe over, in order\n"
+ "\tblock: wait for the operation to return before continuing\n"
+ "\tnon-block: do not wait for the operation to return.\n"},
{"mv", lfs_mv, 0,
"To move directories between MDTs. This command is deprecated, "
"use \"migrate\" instead.\n"
"usage: mv <directory|filename> [--mdt-index|-M] <mdt_index> "
"[--verbose|-v]\n"},
+ {"ladvise", lfs_ladvise, 0,
+ "Provide servers with advice about access patterns for a file.\n"
+ "usage: ladvise [--advice|-a ADVICE] [--start|-s START[kMGT]]\n"
+ " [--background|-b]\n"
+ " {[--end|-e END[kMGT]] | [--length|-l LENGTH[kMGT]]}\n"
+ " <file> ..."},
{"help", Parser_help, 0, "help"},
{"exit", Parser_quit, 0, "quit"},
{"quit", Parser_quit, 0, "quit"},
};
-#define MIGRATION_BLOCKS 1
+#define MIGRATION_NONBLOCK 1
/**
* Internal helper for migrate_copy_data(). Check lease and report error if
{
int fd = -1;
int fdv = -1;
- char volatile_file[PATH_MAX +
- LUSTRE_VOLATILE_HDR_LEN + 4];
char parent[PATH_MAX];
+ int mdt_index;
+ int random_value;
+ char volatile_file[sizeof(parent) +
+ LUSTRE_VOLATILE_HDR_LEN +
+ 2 * sizeof(mdt_index) +
+ 2 * sizeof(random_value) + 4];
char *ptr;
int rc;
struct lov_user_md *lum = NULL;
*ptr = '\0';
}
- rc = snprintf(volatile_file, sizeof(volatile_file), "%s/%s::", parent,
- LUSTRE_VOLATILE_HDR);
- if (rc >= sizeof(volatile_file)) {
- rc = -E2BIG;
+ rc = llapi_file_fget_mdtidx(fd, &mdt_index);
+ if (rc < 0) {
+ fprintf(stderr, "%s: %s: cannot get MDT index: %s\n",
+ progname, name, strerror(-rc));
goto error;
}
- /* create, open a volatile file, use caching (ie no directio) */
- /* exclusive create is not needed because volatile files cannot
- * conflict on name by construction */
- fdv = llapi_file_open_param(volatile_file, O_CREAT | O_WRONLY, 0644,
- param);
+ do {
+ random_value = random();
+ rc = snprintf(volatile_file, sizeof(volatile_file),
+ "%s/%s:%.4X:%.4X", parent, LUSTRE_VOLATILE_HDR,
+ mdt_index, random_value);
+ if (rc >= sizeof(volatile_file)) {
+ rc = -E2BIG;
+ goto error;
+ }
+
+ /* create, open a volatile file, use caching (ie no directio) */
+ fdv = llapi_file_open_param(volatile_file,
+ O_WRONLY | O_CREAT | O_EXCL | O_NOFOLLOW,
+ S_IRUSR | S_IWUSR, param);
+ } while (fdv == -EEXIST);
+
if (fdv < 0) {
rc = fdv;
fprintf(stderr, "%s: %s: cannot create volatile file in"
}
}
- if (migration_flags & MIGRATION_BLOCKS || !file_lease_supported) {
- /* Blocking mode, forced if servers do not support file lease */
- rc = migrate_block(fd, fdv, &st, buf_size, name);
- } else {
+ if (migration_flags & MIGRATION_NONBLOCK && file_lease_supported) {
rc = migrate_nonblock(fd, fdv, &st, buf_size, name);
if (rc == 0) {
have_lease_rdlck = false;
fdv = -1; /* The volatile file is closed as we put the
* lease in non-blocking mode. */
}
+ } else {
+ /* Blocking mode (forced if servers do not support file lease).
+ * It is also the default mode, since we cannot distinguish
+ * between a broken lease and a server that does not support
+ * atomic swap/close (LU-6785) */
+ rc = migrate_block(fd, fdv, &st, buf_size, name);
}
error:
char *mdt_idx_arg = NULL;
unsigned long long size_units = 1;
bool migrate_mode = false;
+ bool migration_block = false;
__u64 migration_flags = 0;
__u32 osts[LOV_MAX_STRIPE_COUNT] = { 0 };
int nr_osts = 0;
struct option long_opts[] = {
- /* valid only in migrate mode */
+ /* --block is only valid in migrate mode */
{"block", no_argument, 0, 'b'},
#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 53, 0)
/* This formerly implied "stripe-count", but was explicitly
{"stripe_index", required_argument, 0, 'i'},
{"mdt-index", required_argument, 0, 'm'},
{"mdt_index", required_argument, 0, 'm'},
+ /* --non-block is only valid in migrate mode */
+ {"non-block", no_argument, 0, 'n'},
{"ost-list", required_argument, 0, 'o'},
{"ost_list", required_argument, 0, 'o'},
{"pool", required_argument, 0, 'p'},
#endif
{"stripe-size", required_argument, 0, 'S'},
{"stripe_size", required_argument, 0, 'S'},
+ /* --verbose is only valid in migrate mode */
+ {"verbose", no_argument, 0, 'v'},
{0, 0, 0, 0}
};
if (strcmp(argv[0], "migrate") == 0)
migrate_mode = true;
- while ((c = getopt_long(argc, argv, "bc:di:m:o:p:s:S:",
+ while ((c = getopt_long(argc, argv, "bc:di:m:no:p:s:S:v",
long_opts, NULL)) >= 0) {
switch (c) {
case 0:
" migrate mode\n");
return CMD_HELP;
}
- migration_flags |= MIGRATION_BLOCKS;
+ migration_block = true;
break;
case 'c':
#if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 6, 53, 0)
}
mdt_idx_arg = optarg;
break;
+ case 'n':
+ if (!migrate_mode) {
+ fprintf(stderr, "--non-block is valid only for"
+ " migrate mode\n");
+ return CMD_HELP;
+ }
+ migration_flags |= MIGRATION_NONBLOCK;
+ break;
#if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 9, 53, 0)
case 's':
#if LUSTRE_VERSION_CODE >= OBD_OCD_VERSION(2, 6, 53, 0)
case 'p':
pool_name_arg = optarg;
break;
+ case 'v':
+ if (!migrate_mode) {
+ fprintf(stderr, "--verbose is valid only for"
+ " migrate mode\n");
+ return CMD_HELP;
+ }
+ migrate_mdt_param.fp_verbose = VERBOSE_DETAIL;
+ break;
default:
return CMD_HELP;
}
return CMD_HELP;
}
+ if ((migration_flags & MIGRATION_NONBLOCK) && migration_block) {
+ fprintf(stderr,
+ "error: %s: cannot specify --non-block and --block\n",
+ argv[0]);
+ return CMD_HELP;
+ }
+
if (pool_name_arg != NULL) {
char *ptr;
int rc;
/* Save the first error encountered. */
if (result2 == 0)
result2 = result;
- fprintf(stderr,
- "error: %s: %s file '%s' failed\n",
+ fprintf(stderr, "error: %s: %s file '%s' failed: %s\n",
argv[0], migrate_mode ? "migrate" : "create",
- fname);
+ fname,
+ pool_name_arg != NULL && result == EINVAL ?
+ "OST not in pool?" : strerror(result));
continue;
}
}
"files", "quota", "limit", "grace");
}
-static void kbytes2str(__u64 num, char *buf, bool h)
+static void kbytes2str(__u64 num, char *buf, int buflen, bool h)
{
if (!h) {
- sprintf(buf, LPU64, num);
+ snprintf(buf, buflen, "%ju", (uintmax_t)num);
} else {
- if (num >> 30)
- sprintf(buf, "%5.4gT", (double)num / (1 << 30));
+ if (num >> 40)
+ snprintf(buf, buflen, "%5.4gP",
+ (double)num / ((__u64)1 << 40));
+ else if (num >> 30)
+ snprintf(buf, buflen, "%5.4gT",
+ (double)num / (1 << 30));
else if (num >> 20)
- sprintf(buf, "%5.4gG", (double)num / (1 << 20));
+ snprintf(buf, buflen, "%5.4gG",
+ (double)num / (1 << 20));
else if (num >> 10)
- sprintf(buf, "%5.4gM", (double)num / (1 << 10));
+ snprintf(buf, buflen, "%5.4gM",
+ (double)num / (1 << 10));
else
- sprintf(buf, LPU64"%s", num, "k");
+ snprintf(buf, buflen, "%ju%s", (uintmax_t)num, "k");
}
}
+#define STRBUF_LEN 32
static void print_quota(char *mnt, struct if_quotactl *qctl, int type,
int rc, bool h)
{
if (qctl->qc_cmd == LUSTRE_Q_GETQUOTA || qctl->qc_cmd == Q_GETOQUOTA) {
int bover = 0, iover = 0;
struct obd_dqblk *dqb = &qctl->qc_dqblk;
- char numbuf[3][32];
+ char numbuf[3][STRBUF_LEN];
char timebuf[40];
- char strbuf[32];
+ char strbuf[STRBUF_LEN];
if (dqb->dqb_bhardlimit &&
lustre_stoqb(dqb->dqb_curspace) >= dqb->dqb_bhardlimit) {
if (bover)
diff2str(dqb->dqb_btime, timebuf, now);
- kbytes2str(lustre_stoqb(dqb->dqb_curspace), strbuf, h);
+ kbytes2str(lustre_stoqb(dqb->dqb_curspace),
+ strbuf, sizeof(strbuf), h);
if (rc == -EREMOTEIO)
sprintf(numbuf[0], "%s*", strbuf);
else
sprintf(numbuf[0], (dqb->dqb_valid & QIF_SPACE) ?
"%s" : "[%s]", strbuf);
- kbytes2str(dqb->dqb_bsoftlimit, strbuf, h);
+ kbytes2str(dqb->dqb_bsoftlimit, strbuf, sizeof(strbuf), h);
if (type == QC_GENERAL)
sprintf(numbuf[1], (dqb->dqb_valid & QIF_BLIMITS) ?
"%s" : "[%s]", strbuf);
else
sprintf(numbuf[1], "%s", "-");
- kbytes2str(dqb->dqb_bhardlimit, strbuf, h);
+ kbytes2str(dqb->dqb_bhardlimit, strbuf, sizeof(strbuf), h);
sprintf(numbuf[2], (dqb->dqb_valid & QIF_BLIMITS) ?
"%s" : "[%s]", strbuf);
diff2str(dqb->dqb_itime, timebuf, now);
sprintf(numbuf[0], (dqb->dqb_valid & QIF_INODES) ?
- LPU64 : "["LPU64"]", dqb->dqb_curinodes);
+ "%ju" : "[%ju]", (uintmax_t)dqb->dqb_curinodes);
if (type == QC_GENERAL)
sprintf(numbuf[1], (dqb->dqb_valid & QIF_ILIMITS) ?
- LPU64 : "["LPU64"]", dqb->dqb_isoftlimit);
+ "%ju" : "[%ju]",
+ (uintmax_t)dqb->dqb_isoftlimit);
else
sprintf(numbuf[1], "%s", "-");
sprintf(numbuf[2], (dqb->dqb_valid & QIF_ILIMITS) ?
- LPU64 : "["LPU64"]", dqb->dqb_ihardlimit);
+ "%ju" : "[%ju]", (uintmax_t)dqb->dqb_ihardlimit);
if (type != QC_OSTIDX)
printf(" %7s%c %6s %7s %7s",
if (qctl.qc_valid == QC_GENERAL && qctl.qc_cmd != LUSTRE_Q_GETINFO &&
verbose) {
- char strbuf[32];
+ char strbuf[STRBUF_LEN];
rc2 = print_obd_quota(mnt, &qctl, 1, human_readable,
&total_ialloc);
rc3 = print_obd_quota(mnt, &qctl, 0, human_readable,
&total_balloc);
- kbytes2str(total_balloc, strbuf, human_readable);
- printf("Total allocated inode limit: "LPU64", total "
- "allocated block limit: %s\n", total_ialloc, strbuf);
+ kbytes2str(total_balloc, strbuf, sizeof(strbuf),
+ human_readable);
+ printf("Total allocated inode limit: %ju, total "
+ "allocated block limit: %s\n", (uintmax_t)total_ialloc,
+ strbuf);
}
if (rc1 || rc2 || rc3 || inacc)
secs = rec->cr_time >> 30;
gmtime_r(&secs, &ts);
- printf(LPU64" %02d%-5s %02d:%02d:%02d.%06d %04d.%02d.%02d "
- "0x%x t="DFID, rec->cr_index, rec->cr_type,
+ printf("%ju %02d%-5s %02d:%02d:%02d.%06d %04d.%02d.%02d "
+ "0x%x t="DFID, (uintmax_t) rec->cr_index, rec->cr_type,
changelog_type2str(rec->cr_type),
ts.tm_hour, ts.tm_min, ts.tm_sec,
(int)(rec->cr_time & ((1<<30) - 1)),
if (rc < 0)
err(errno, "cannot get version for %s", path);
else
- printf(LPU64 "\n", data_version);
+ printf("%ju" "\n", (uintmax_t)data_version);
close(fd);
return rc;
SWAP_LAYOUTS_KEEP_ATIME);
}
+static const char *const ladvise_names[] = LU_LADVISE_NAMES;
+
+static enum lu_ladvise_type lfs_get_ladvice(const char *string)
+{
+ enum lu_ladvise_type advice;
+
+ for (advice = 0;
+ advice < ARRAY_SIZE(ladvise_names); advice++) {
+ if (ladvise_names[advice] == NULL)
+ continue;
+ if (strcmp(string, ladvise_names[advice]) == 0)
+ return advice;
+ }
+
+ return LU_LADVISE_INVALID;
+}
+
+static int lfs_ladvise(int argc, char **argv)
+{
+ struct option long_opts[] = {
+ {"advice", required_argument, 0, 'a'},
+ {"background", no_argument, 0, 'b'},
+ {"end", required_argument, 0, 'e'},
+ {"start", required_argument, 0, 's'},
+ {"length", required_argument, 0, 'l'},
+ {0, 0, 0, 0}
+ };
+ char short_opts[] = "a:be:l:s:";
+ int c;
+ int rc = 0;
+ const char *path;
+ int fd;
+ struct lu_ladvise advice;
+ enum lu_ladvise_type advice_type = LU_LADVISE_INVALID;
+ unsigned long long start = 0;
+ unsigned long long end = LUSTRE_EOF;
+ unsigned long long length = 0;
+ unsigned long long size_units;
+ unsigned long long flags = 0;
+
+ optind = 0;
+ while ((c = getopt_long(argc, argv, short_opts,
+ long_opts, NULL)) != -1) {
+ switch (c) {
+ case 'a':
+ advice_type = lfs_get_ladvice(optarg);
+ if (advice_type == LU_LADVISE_INVALID) {
+ fprintf(stderr, "%s: invalid advice type "
+ "'%s'\n", argv[0], optarg);
+ fprintf(stderr, "Valid types:");
+
+ for (advice_type = 0;
+ advice_type < ARRAY_SIZE(ladvise_names);
+ advice_type++) {
+ if (ladvise_names[advice_type] == NULL)
+ continue;
+ fprintf(stderr, " %s",
+ ladvise_names[advice_type]);
+ }
+ fprintf(stderr, "\n");
+
+ return CMD_HELP;
+ }
+ break;
+ case 'b':
+ flags |= LF_ASYNC;
+ break;
+ case 'e':
+ size_units = 1;
+ rc = llapi_parse_size(optarg, &end,
+ &size_units, 0);
+ if (rc) {
+ fprintf(stderr, "%s: bad end offset '%s'\n",
+ argv[0], optarg);
+ return CMD_HELP;
+ }
+ break;
+ case 's':
+ size_units = 1;
+ rc = llapi_parse_size(optarg, &start,
+ &size_units, 0);
+ if (rc) {
+ fprintf(stderr, "%s: bad start offset "
+ "'%s'\n", argv[0], optarg);
+ return CMD_HELP;
+ }
+ break;
+ case 'l':
+ size_units = 1;
+ rc = llapi_parse_size(optarg, &length,
+ &size_units, 0);
+ if (rc) {
+ fprintf(stderr, "%s: bad length '%s'\n",
+ argv[0], optarg);
+ return CMD_HELP;
+ }
+ break;
+ case '?':
+ return CMD_HELP;
+ default:
+ fprintf(stderr, "%s: option '%s' unrecognized\n",
+ argv[0], argv[optind - 1]);
+ return CMD_HELP;
+ }
+ }
+
+ if (advice_type == LU_LADVISE_INVALID) {
+ fprintf(stderr, "%s: please give an advice type\n", argv[0]);
+ fprintf(stderr, "Valid types:");
+ for (advice_type = 0; advice_type < ARRAY_SIZE(ladvise_names);
+ advice_type++) {
+ if (ladvise_names[advice_type] == NULL)
+ continue;
+ fprintf(stderr, " %s", ladvise_names[advice_type]);
+ }
+ fprintf(stderr, "\n");
+ return CMD_HELP;
+ }
+
+ if (argc <= optind) {
+ fprintf(stderr, "%s: please give one or more file names\n",
+ argv[0]);
+ return CMD_HELP;
+ }
+
+ if (end != LUSTRE_EOF && length != 0 && end != start + length) {
+ fprintf(stderr, "%s: conflicting arguments of -l and -e\n",
+ argv[0]);
+ return CMD_HELP;
+ }
+
+ if (end == LUSTRE_EOF && length != 0)
+ end = start + length;
+
+ if (end <= start) {
+ fprintf(stderr, "%s: range [%llu, %llu] is invalid\n",
+ argv[0], start, end);
+ return CMD_HELP;
+ }
+
+ while (optind < argc) {
+ int rc2;
+
+ path = argv[optind++];
+
+ fd = open(path, O_RDONLY);
+ if (fd < 0) {
+ fprintf(stderr, "%s: cannot open file '%s': %s\n",
+ argv[0], path, strerror(errno));
+ rc2 = -errno;
+ goto next;
+ }
+
+ advice.lla_start = start;
+ advice.lla_end = end;
+ advice.lla_advice = advice_type;
+ advice.lla_padding = 0;
+ rc2 = llapi_ladvise(fd, flags, 1, &advice);
+ close(fd);
+ if (rc2 < 0) {
+ fprintf(stderr, "%s: cannot give advice '%s' to file "
+ "'%s': %s\n", argv[0],
+ ladvise_names[advice_type],
+ path, strerror(errno));
+ }
+next:
+ if (rc == 0 && rc2 < 0)
+ rc = rc2;
+ }
+ return rc;
+}
+
int main(int argc, char **argv)
{
int rc;