* (C) Copyright 2012 Commissariat a l'energie atomique et aux energies
* alternatives
*
+ * Copyright (c) 2013, 2016, Intel Corporation.
*/
/* HSM copytool program for POSIX filesystem-based HSM's.
*
*
* This particular tool can also import an existing HSM archive.
*/
+
+#ifndef _GNU_SOURCE
+#define _GNU_SOURCE
+#endif
+#include <ctype.h>
+#include <fcntl.h>
+#include <stdio.h>
#include <stdint.h>
#include <stdlib.h>
+#include <string.h>
#include <dirent.h>
#include <errno.h>
+#include <getopt.h>
+#include <pthread.h>
+#include <time.h>
+#include <unistd.h>
#include <utime.h>
+#include <signal.h>
+#include <sys/time.h>
#include <sys/xattr.h>
#include <sys/syscall.h>
#include <sys/types.h>
-#include <lustre/lustre_idl.h>
+
+#include <libcfs/util/string.h>
+#include <linux/lustre/lustre_fid.h>
+#include <lnetconfig/cyaml.h>
#include <lustre/lustreapi.h>
+#include "lstddef.h"
+#include "pid_file.h"
/* Progress reporting period */
#define REPORT_INTERVAL_DEFAULT 30
#define ONE_MB 0x100000
-/* copytool uses a 32b bitmask field to register with kuc
- * archive num = 0 => all
- * archive num from 1 to 32
- */
-#define MAX_ARCHIVE_CNT (sizeof(__u32) * 8)
+#ifndef NSEC_PER_SEC
+# define NSEC_PER_SEC 1000000000UL
+#endif
enum ct_action {
CA_IMPORT = 1,
CA_REBIND,
CA_MAXSEQ,
+ CA_ARCHIVE_UPGRADE,
+};
+
+enum ct_archive_format {
+ /* v1 (original) using 6 directories (oid & 0xffff)/-/-/-/-/-/FID.
+ * Places only one FID per directory. See ct_path_archive() below. */
+ CT_ARCHIVE_FORMAT_V1 = 1,
+ /* v2 using 1 directory (oid & 0xffff)/FID. */
+ CT_ARCHIVE_FORMAT_V2 = 2,
+};
+
+static const char *ct_archive_format_name[] = {
+ [CT_ARCHIVE_FORMAT_V1] = "v1",
+ [CT_ARCHIVE_FORMAT_V2] = "v2",
};
+static int
+ct_str_to_archive_format(const char *str, enum ct_archive_format *pctaf)
+{
+ enum ct_archive_format ctaf;
+
+ for (ctaf = 0; ctaf < ARRAY_SIZE(ct_archive_format_name); ctaf++) {
+ if (ct_archive_format_name[ctaf] != NULL &&
+ strcmp(ct_archive_format_name[ctaf], str) == 0) {
+ *pctaf = ctaf;
+ return 0;
+ }
+ }
+
+ return -EINVAL;
+}
+
+static const char *ct_archive_format_to_str(enum ct_archive_format ctaf)
+{
+ if (0 <= ctaf &&
+ ctaf < ARRAY_SIZE(ct_archive_format_name) &&
+ ct_archive_format_name[ctaf] != NULL)
+ return ct_archive_format_name[ctaf];
+
+ return "null";
+}
+
struct options {
int o_copy_attrs;
int o_daemonize;
int o_shadow_tree;
int o_verbose;
int o_copy_xattrs;
- int o_archive_cnt;
- int o_archive_id[MAX_ARCHIVE_CNT];
+ const char *o_config_path;
+ enum ct_archive_format o_archive_format;
+ int o_archive_id_used;
+ int o_archive_id_cnt;
+ int *o_archive_id;
int o_report_int;
unsigned long long o_bandwidth;
size_t o_chunk_size;
enum ct_action o_action;
+ char *o_event_fifo;
char *o_mnt;
+ int o_mnt_fd;
char *o_hsm_root;
char *o_src; /* for import, or rebind */
char *o_dst; /* for import, or rebind */
+ char *o_pid_file;
};
/* everything else is zeroed */
struct options opt = {
.o_copy_attrs = 1,
.o_shadow_tree = 1,
- .o_verbose = LLAPI_MSG_WARN,
+ .o_verbose = LLAPI_MSG_INFO,
.o_copy_xattrs = 1,
+ .o_archive_format = CT_ARCHIVE_FORMAT_V1,
.o_report_int = REPORT_INTERVAL_DEFAULT,
.o_chunk_size = ONE_MB,
};
static char cmd_name[PATH_MAX];
static char fs_name[MAX_OBD_NAME + 1];
+static int pid_file_fd = -1;
static struct hsm_copytool_private *ctdata;
+static inline double ct_now(void)
+{
+ struct timeval tv;
+
+ gettimeofday(&tv, NULL);
+
+ return tv.tv_sec + 0.000001 * tv.tv_usec;
+}
-#define CT_ERROR(_rc, _format, ...) \
+#define CT_ERROR(_rc, _format, ...) \
llapi_error(LLAPI_MSG_ERROR, _rc, \
- "%s[%ld]: "_format, \
- cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
-#define CT_DEBUG(_format, ...) \
+ "%f %s[%ld]: "_format, \
+ ct_now(), cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
+
+#define CT_DEBUG(_format, ...) \
llapi_error(LLAPI_MSG_DEBUG | LLAPI_MSG_NO_ERRNO, 0, \
- "%s[%ld]: "_format, \
- cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
+ "%f %s[%ld]: "_format, \
+ ct_now(), cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
+
#define CT_WARN(_format, ...) \
llapi_error(LLAPI_MSG_WARN | LLAPI_MSG_NO_ERRNO, 0, \
- "%s[%ld]: "_format, \
- cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
-#define CT_TRACE(_format, ...) \
+ "%f %s[%ld]: "_format, \
+ ct_now(), cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
+
+#define CT_TRACE(_format, ...) \
llapi_error(LLAPI_MSG_INFO | LLAPI_MSG_NO_ERRNO, 0, \
- "%s[%ld]: "_format, \
- cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
-#define CT_PRINTF(_format, ...) \
- llapi_printf(LLAPI_MSG_NORMAL, \
- "%s[%ld]: "_format, \
- cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
+ "%f %s[%ld]: "_format, \
+ ct_now(), cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
static void usage(const char *name, int rc)
{
"into a Lustre filesystem.\n"
" Usage:\n"
" %s [options] --import <src> <dst> <lustre_mount_point>\n"
- " import an archived subtree at\n"
- " <src> (relative to hsm_root) into the Lustre filesystem at\n"
- " <dst> (absolute)\n"
+ " import an archived subtree from\n"
+ " <src> (FID or relative path to hsm_root) into the Lustre\n"
+ " filesystem at\n"
+ " <dst> (absolute path)\n"
" %s [options] --rebind <old_FID> <new_FID> <lustre_mount_point>\n"
" rebind an entry in the HSM to a new FID\n"
" <old_FID> old FID the HSM entry is bound to\n"
" each line of <list_file> consists of <old_FID> <new_FID>\n"
" %s [options] --max-sequence <fsname>\n"
" return the max fid sequence of archived files\n"
- " -A, --archive <#> Archive number (repeatable)\n"
- " -p, --hsm-root <path> Target HSM mount point\n"
- " -q, --quiet Produce less verbose output\n"
- " -v, --verbose Produce more verbose output\n"
- " -c, --chunk-size <sz> I/O size used during data copy\n"
- " (unit can be used, default is MB)\n"
- " --abort-on-error Abort operation on major error\n"
- " --dry-run Don't run, just show what would be done\n"
- " --bandwidth <bw> Limit I/O bandwidth (unit can be used\n,"
- " default is MB)\n",
- cmd_name, cmd_name, cmd_name, cmd_name, cmd_name);
+ " %s [options] --archive-upgrade=VER\n"
+ " Upgrade or downgrade the archive to version VER\n"
+ "Options:\n"
+ " --abort-on-error Abort operation on major error\n"
+ " -A, --archive <#> Archive number (repeatable)\n"
+ " -C, --config=PATH Read config from PATH\n"
+ " -F, --archive-format=VER Use archive format VER\n"
+ " -b, --bandwidth <bw> Limit I/O bandwidth (unit can be used\n,"
+ " default is MB)\n"
+ " --dry-run Don't run, just show what would be done\n"
+ " -c, --chunk-size <sz> I/O size used during data copy\n"
+ " (unit can be used, default is MB)\n"
+ " -f, --event-fifo <path> Write events stream to fifo\n"
+ " -P, --pid-file=PATH Lock and write PID to PATH\n"
+ " -p, --hsm-root <path> Target HSM mount point\n"
+ " -q, --quiet Produce less verbose output\n"
+ " -u, --update-interval <s> Interval between progress reports sent\n"
+ " to Coordinator\n"
+ " -v, --verbose Produce more verbose output\n",
+ cmd_name, cmd_name, cmd_name, cmd_name, cmd_name, cmd_name);
exit(rc);
}
static int ct_parseopts(int argc, char * const *argv)
{
struct option long_opts[] = {
- {"abort-on-error", no_argument, &opt.o_abort_on_error, 1},
- {"abort_on_error", no_argument, &opt.o_abort_on_error, 1},
- {"archive", required_argument, NULL, 'A'},
- {"bandwidth", required_argument, NULL, 'b'},
- {"chunk-size", required_argument, NULL, 'c'},
- {"chunk_size", required_argument, NULL, 'c'},
- {"daemon", no_argument, &opt.o_daemonize, 1},
- {"dry-run", no_argument, &opt.o_dry_run, 1},
- {"help", no_argument, NULL, 'h'},
- {"hsm-root", required_argument, NULL, 'p'},
- {"hsm_root", required_argument, NULL, 'p'},
- {"import", no_argument, NULL, 'i'},
- {"max-sequence", no_argument, NULL, 'M'},
- {"max_sequence", no_argument, NULL, 'M'},
- {"no-attr", no_argument, &opt.o_copy_attrs, 0},
- {"no_attr", no_argument, &opt.o_copy_attrs, 0},
- {"no-shadow", no_argument, &opt.o_shadow_tree, 0},
- {"no_shadow", no_argument, &opt.o_shadow_tree, 0},
- {"no-xattr", no_argument, &opt.o_copy_xattrs, 0},
- {"no_xattr", no_argument, &opt.o_copy_xattrs, 0},
- {"quiet", no_argument, NULL, 'q'},
- {"rebind", no_argument, NULL, 'r'},
- {"report", required_argument, &opt.o_report_int, 0},
- {"verbose", no_argument, NULL, 'v'},
- {0, 0, 0, 0}
- };
- int c, rc;
- unsigned long long value;
- unsigned long long unit;
+ { .val = 1, .name = "abort-on-error",
+ .flag = &opt.o_abort_on_error, .has_arg = no_argument },
+ { .val = 1, .name = "abort_on_error",
+ .flag = &opt.o_abort_on_error, .has_arg = no_argument },
+ { .val = 'A', .name = "archive", .has_arg = required_argument },
+ { .val = 'b', .name = "bandwidth", .has_arg = required_argument },
+ { .val = 'C', .name = "config", .has_arg = required_argument },
+ { .val = 'c', .name = "chunk-size", .has_arg = required_argument },
+ { .val = 'c', .name = "chunk_size", .has_arg = required_argument },
+ { .val = 1, .name = "daemon", .has_arg = no_argument,
+ .flag = &opt.o_daemonize },
+ { .val = 'F', .name = "archive-format", .has_arg = required_argument },
+ { .val = 'U', .name = "archive-upgrade", .has_arg = required_argument },
+ { .val = 'f', .name = "event-fifo", .has_arg = required_argument },
+ { .val = 'f', .name = "event_fifo", .has_arg = required_argument },
+ { .val = 1, .name = "dry-run", .has_arg = no_argument,
+ .flag = &opt.o_dry_run },
+ { .val = 'h', .name = "help", .has_arg = no_argument },
+ { .val = 'i', .name = "import", .has_arg = no_argument },
+ { .val = 'M', .name = "max-sequence", .has_arg = no_argument },
+ { .val = 'M', .name = "max_sequence", .has_arg = no_argument },
+ { .val = 0, .name = "no-attr", .has_arg = no_argument,
+ .flag = &opt.o_copy_attrs },
+ { .val = 0, .name = "no_attr", .has_arg = no_argument,
+ .flag = &opt.o_copy_attrs },
+ { .val = 0, .name = "no-shadow", .has_arg = no_argument,
+ .flag = &opt.o_shadow_tree },
+ { .val = 0, .name = "no_shadow", .has_arg = no_argument,
+ .flag = &opt.o_shadow_tree },
+ { .val = 0, .name = "no-xattr", .has_arg = no_argument,
+ .flag = &opt.o_copy_xattrs },
+ { .val = 0, .name = "no_xattr", .has_arg = no_argument,
+ .flag = &opt.o_copy_xattrs },
+ { .val = 'P', .name = "pid-file", .has_arg = required_argument },
+ { .val = 'p', .name = "hsm-root", .has_arg = required_argument },
+ { .val = 'p', .name = "hsm_root", .has_arg = required_argument },
+ { .val = 'q', .name = "quiet", .has_arg = no_argument },
+ { .val = 'r', .name = "rebind", .has_arg = no_argument },
+ { .val = 'u', .name = "update-interval",
+ .has_arg = required_argument },
+ { .val = 'u', .name = "update_interval",
+ .has_arg = required_argument },
+ { .val = 'v', .name = "verbose", .has_arg = no_argument },
+ { .name = NULL } };
+
+ unsigned long long value;
+ unsigned long long unit;
+ bool all_id = false;
+ int c, rc;
+ int i;
optind = 0;
- while ((c = getopt_long(argc, argv, "A:b:c:hiMp:qrv",
+
+ opt.o_archive_id_cnt = LL_HSM_ORIGIN_MAX_ARCHIVE;
+ opt.o_archive_id = malloc(opt.o_archive_id_cnt *
+ sizeof(*opt.o_archive_id));
+ if (opt.o_archive_id == NULL)
+ return -ENOMEM;
+repeat:
+ while ((c = getopt_long(argc, argv, "A:b:C:c:F:f:hiMp:P:qrU:u:v",
long_opts, NULL)) != -1) {
switch (c) {
- case 'A':
- if ((opt.o_archive_cnt >= MAX_ARCHIVE_CNT) ||
- (atoi(optarg) >= MAX_ARCHIVE_CNT)) {
- rc = -E2BIG;
- CT_ERROR(rc, "archive number must be less"
- "than %zu", MAX_ARCHIVE_CNT);
+ case 'A': {
+ char *end = NULL;
+ int val = strtol(optarg, &end, 10);
+
+ if (*end != '\0') {
+ rc = -EINVAL;
+ CT_ERROR(rc, "invalid archive-id: '%s'",
+ optarg);
return rc;
}
- opt.o_archive_id[opt.o_archive_cnt] = atoi(optarg);
- opt.o_archive_cnt++;
+ /* if archiveID is zero, any archiveID is accepted */
+ if (all_id == true)
+ goto repeat;
+
+ if (val == 0) {
+ free(opt.o_archive_id);
+ opt.o_archive_id = NULL;
+ opt.o_archive_id_cnt = 0;
+ opt.o_archive_id_used = 0;
+ all_id = true;
+ CT_WARN("archive-id = 0 is found, any backend will be served\n");
+ goto repeat;
+ }
+
+ /* skip the duplicated id */
+ for (i = 0; i < opt.o_archive_id_used; i++) {
+ if (opt.o_archive_id[i] == val)
+ goto repeat;
+ }
+ /* extend the space */
+ if (opt.o_archive_id_used >= opt.o_archive_id_cnt) {
+ int *tmp;
+
+ opt.o_archive_id_cnt *= 2;
+ tmp = realloc(opt.o_archive_id,
+ sizeof(*opt.o_archive_id) *
+ opt.o_archive_id_cnt);
+ if (tmp == NULL)
+ return -ENOMEM;
+
+ opt.o_archive_id = tmp;
+ }
+
+ opt.o_archive_id[opt.o_archive_id_used++] = val;
break;
+ }
case 'b': /* -b and -c have both a number with unit as arg */
case 'c':
unit = ONE_MB;
else
opt.o_bandwidth = value;
break;
+ case 'C':
+ opt.o_config_path = optarg;
+ break;
+ case 'F':
+ rc = ct_str_to_archive_format(optarg, &opt.o_archive_format);
+ if (rc < 0) {
+ CT_ERROR(rc, "invalid archive format '%s'", optarg);
+ return -EINVAL;
+ }
+ break;
+ case 'U':
+ rc = ct_str_to_archive_format(optarg, &opt.o_archive_format);
+ if (rc < 0) {
+ CT_ERROR(rc, "invalid archive format '%s'", optarg);
+ return -EINVAL;
+ }
+ opt.o_action = CA_ARCHIVE_UPGRADE;
+ break;
+
+ case 'f':
+ opt.o_event_fifo = optarg;
+ break;
case 'h':
usage(argv[0], 0);
case 'i':
case 'M':
opt.o_action = CA_MAXSEQ;
break;
+ case 'P':
+ opt.o_pid_file = optarg;
+ break;
case 'p':
opt.o_hsm_root = optarg;
break;
case 'r':
opt.o_action = CA_REBIND;
break;
+ case 'u':
+ opt.o_report_int = atoi(optarg);
+ if (opt.o_report_int < 0) {
+ rc = -EINVAL;
+ CT_ERROR(rc, "bad value for -%c '%s'", c,
+ optarg);
+ return rc;
+ }
+ break;
case 'v':
opt.o_verbose++;
break;
break;
}
+ if (opt.o_action == CA_ARCHIVE_UPGRADE)
+ return 0;
+
if (argc != optind + 1) {
rc = -EINVAL;
CT_ERROR(rc, "no mount point specified");
}
opt.o_mnt = argv[optind];
+ opt.o_mnt_fd = -1;
CT_TRACE("action=%d src=%s dst=%s mount_point=%s",
opt.o_action, opt.o_src, opt.o_dst, opt.o_mnt);
- if (!opt.o_dry_run && opt.o_hsm_root == NULL) {
- rc = -EINVAL;
- CT_ERROR(rc, "must specify a root directory for the backend");
- return rc;
- }
-
if (opt.o_action == CA_IMPORT) {
if (opt.o_src && opt.o_src[0] == '/') {
rc = -EINVAL;
return 0;
}
-/* mkdir -p path */
+static int ct_mkdirat_p(int fd, char *path, mode_t mode)
+{
+ char *split;
+ int rc;
+
+ if (strlen(path) == 0 || strcmp(path, ".") == 0)
+ return 0;
+
+ rc = mkdirat(fd, path, mode);
+ if (rc == 0 || errno == EEXIST)
+ return 0;
+
+ if (errno != ENOENT)
+ return -errno;
+
+ split = strrchr(path, '/');
+ if (split == NULL)
+ return -ENOENT;
+
+ *split = '\0';
+ rc = ct_mkdirat_p(fd, path, mode);
+ *split = '/';
+ if (rc < 0)
+ return rc;
+
+ rc = mkdirat(fd, path, mode);
+ if (rc == 0 || errno == EEXIST)
+ return 0;
+
+ return -errno;
+}
+
+/* XXX Despite the name, this is 'mkdir -p $(dirname path)' */
static int ct_mkdir_p(const char *path)
{
- char *saved, *ptr;
- int rc;
+ char *path2;
+ char *split;
+ int rc;
- ptr = strdup(path);
- saved = ptr;
- while (*ptr == '/')
- ptr++;
+ path2 = strdup(path);
+ if (path2 == NULL)
+ return -ENOMEM;
- while ((ptr = strchr(ptr, '/')) != NULL) {
- *ptr = '\0';
- rc = mkdir(saved, DIR_PERM);
- *ptr = '/';
- if (rc < 0 && errno != EEXIST) {
- rc = -errno;
- CT_ERROR(rc, "cannot mkdir '%s'", path);
- free(saved);
- return rc;
- }
- ptr++;
+ split = strrchr(path2, '/');
+ if (split == NULL) {
+ rc = 0;
+ goto out;
}
- free(saved);
+ *split = '\0';
- return 0;
+ rc = ct_mkdirat_p(AT_FDCWD, path2, DIR_PERM);
+out:
+ free(path2);
+
+ return rc;
}
static int ct_save_stripe(int src_fd, const char *src, const char *dst)
{
- char lov_file[PATH_MAX];
+ char lov_file[PATH_MAX + 8];
char lov_buf[XATTR_SIZE_MAX];
struct lov_user_md *lum;
int rc;
static int ct_load_stripe(const char *src, void *lovea, size_t *lovea_size)
{
- char lov_file[PATH_MAX];
+ char lov_file[PATH_MAX + 4];
int rc;
int fd;
rc = fsetxattr(dst_fd, XATTR_LUSTRE_LOV, lovea, lovea_size,
XATTR_CREATE);
if (rc < 0) {
- CT_ERROR(errno, "cannot set lov EA on '%s'", dst);
rc = -errno;
+ CT_ERROR(rc, "cannot set lov EA on '%s'", dst);
}
return rc;
}
-/* non-blocking read or write */
-static int nonblock_rw(bool wr, int fd, char *buf, int size)
-{
- int rc;
-
- if (wr)
- rc = write(fd, buf, size);
- else
- rc = read(fd, buf, size);
-
- if ((rc < 0) && (errno == -EAGAIN)) {
- fd_set set;
- struct timeval timeout;
-
- timeout.tv_sec = opt.o_report_int;
-
- FD_ZERO(&set);
- FD_SET(fd, &set);
- if (wr)
- rc = select(FD_SETSIZE, NULL, &set, NULL, &timeout);
- else
- rc = select(FD_SETSIZE, &set, NULL, NULL, &timeout);
- if (rc < 0)
- return -errno;
- if (rc == 0)
- /* Timed out, we read nothing */
- return -EAGAIN;
-
- /* Should be available now */
- if (wr)
- rc = write(fd, buf, size);
- else
- rc = read(fd, buf, size);
- }
-
- if (rc < 0)
- rc = -errno;
-
- return rc;
-}
-
static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
const char *dst, int src_fd, int dst_fd,
const struct hsm_action_item *hai, long hal_flags)
{
struct hsm_extent he;
+ __u64 offset = hai->hai_extent.offset;
struct stat src_st;
struct stat dst_st;
- char *buf;
- __u64 wpos = 0;
- __u64 rpos = 0;
- __u64 rlen;
- time_t last_print_time = time(0);
- int rsize;
- int wsize;
- int bufoff = 0;
+ char *buf = NULL;
+ __u64 write_total = 0;
+ __u64 length = hai->hai_extent.length;
+ time_t last_report_time;
int rc = 0;
-
- CT_TRACE("going to copy data from '%s' to '%s'", src, dst);
-
- buf = malloc(opt.o_chunk_size);
- if (buf == NULL)
- return -ENOMEM;
+ double start_ct_now = ct_now();
+ /* Bandwidth Control */
+ time_t start_time;
+ time_t now;
+ time_t last_bw_print;
if (fstat(src_fd, &src_st) < 0) {
- CT_ERROR(errno, "cannot stat '%s'", src);
- return -errno;
+ rc = -errno;
+ CT_ERROR(rc, "cannot stat '%s'", src);
+ return rc;
}
if (!S_ISREG(src_st.st_mode)) {
return rc;
}
- rc = lseek(src_fd, hai->hai_extent.offset, SEEK_SET);
- if (rc < 0) {
- CT_ERROR(errno,
- "cannot seek for read to "LPU64" (len %jd) in '%s'",
- hai->hai_extent.offset, (intmax_t)src_st.st_size, src);
- rc = -errno;
- goto out;
+ if (hai->hai_extent.offset > (__u64)src_st.st_size) {
+ rc = -EINVAL;
+ CT_ERROR(rc, "Trying to start reading past end (%ju > "
+ "%jd) of '%s' source file",
+ (uintmax_t)hai->hai_extent.offset,
+ (intmax_t)src_st.st_size, src);
+ return rc;
}
if (fstat(dst_fd, &dst_st) < 0) {
- CT_ERROR(errno, "cannot stat '%s'", dst);
- return -errno;
+ rc = -errno;
+ CT_ERROR(rc, "cannot stat '%s'", dst);
+ return rc;
}
if (!S_ISREG(dst_st.st_mode)) {
return rc;
}
- rc = lseek(dst_fd, hai->hai_extent.offset, SEEK_SET);
- if (rc < 0) {
- rc = -errno;
- CT_ERROR(rc, "cannot seek for write to "LPU64" on '%s'",
- hai->hai_extent.offset, src);
- goto out;
- }
+ /* Don't read beyond a given extent */
+ if (length > src_st.st_size - hai->hai_extent.offset)
+ length = src_st.st_size - hai->hai_extent.offset;
+
+ start_time = last_bw_print = last_report_time = time(NULL);
- he.offset = hai->hai_extent.offset;
+ he.offset = offset;
he.length = 0;
- rc = llapi_hsm_action_progress(hcp, &he, 0);
+ rc = llapi_hsm_action_progress(hcp, &he, length, 0);
if (rc < 0) {
/* Action has been canceled or something wrong
* is happening. Stop copying data. */
}
errno = 0;
- /* Don't read beyond a given extent */
- rlen = (hai->hai_extent.length == -1LL) ?
- src_st.st_size : hai->hai_extent.length;
-
- while (wpos < rlen) {
- int chunk = (rlen - wpos > opt.o_chunk_size) ?
- opt.o_chunk_size : rlen - wpos;
-
- /* Only read more if we wrote everything in the buffer */
- if (wpos == rpos) {
- rsize = nonblock_rw(0, src_fd, buf, chunk);
- if (rsize == 0)
- /* EOF */
- break;
- if (rsize == -EAGAIN) {
- /* Timed out */
- rsize = 0;
- if (rpos == 0) {
- /* Haven't read anything yet, let's
- * give it back to the coordinator
- * for rescheduling */
- rc = -EAGAIN;
- break;
- }
- }
+ buf = malloc(opt.o_chunk_size);
+ if (buf == NULL) {
+ rc = -ENOMEM;
+ goto out;
+ }
- if (rsize < 0) {
- rc = rsize;
- CT_ERROR(rc, "cannot read from '%s'", src);
- break;
- }
+ CT_TRACE("start copy of %ju bytes from '%s' to '%s'",
+ (uintmax_t)length, src, dst);
- rpos += rsize;
- bufoff = 0;
- }
+ while (write_total < length) {
+ ssize_t rsize;
+ ssize_t wsize;
+ int chunk = (length - write_total > opt.o_chunk_size) ?
+ opt.o_chunk_size : length - write_total;
+
+ rsize = pread(src_fd, buf, chunk, offset);
+ if (rsize == 0)
+ /* EOF */
+ break;
- wsize = nonblock_rw(1, dst_fd, buf + bufoff, rpos - wpos);
- if (wsize == -EAGAIN)
- /* Timed out */
- wsize = 0;
+ if (rsize < 0) {
+ rc = -errno;
+ CT_ERROR(rc, "cannot read from '%s'", src);
+ break;
+ }
+ wsize = pwrite(dst_fd, buf, rsize, offset);
if (wsize < 0) {
- rc = wsize;
+ rc = -errno;
CT_ERROR(rc, "cannot write to '%s'", dst);
break;
}
- wpos += wsize;
- bufoff += wsize;
+ write_total += wsize;
+ offset += wsize;
+ now = time(NULL);
+ /* sleep if needed, to honor bandwidth limits */
if (opt.o_bandwidth != 0) {
- static unsigned long long tot_bytes;
- static time_t start_time, last_time;
- time_t now = time(0);
- double tot_time, excess;
- unsigned int sleep_time;
-
- if (now > last_time + 5) {
- tot_bytes = 0;
- start_time = last_time = now;
- }
+ unsigned long long write_theory;
- tot_bytes += wsize;
- tot_time = now - start_time;
- if (tot_time < 1)
- tot_time = 1;
+ write_theory = (now - start_time) * opt.o_bandwidth;
- excess = tot_bytes - tot_time * opt.o_bandwidth;
- sleep_time = excess * 1000000 / opt.o_bandwidth;
- if ((now - start_time) % 10 == 1)
- CT_TRACE("bandwith control: excess=%E"
- " sleep for %dus",
- excess, sleep_time);
+ if (write_theory < write_total) {
+ unsigned long long excess;
+ struct timespec delay;
- if (excess > 0)
- usleep(sleep_time);
+ excess = write_total - write_theory;
- last_time = now;
+ delay.tv_sec = excess / opt.o_bandwidth;
+ delay.tv_nsec = (excess % opt.o_bandwidth) *
+ NSEC_PER_SEC / opt.o_bandwidth;
+
+ if (now >= last_bw_print + opt.o_report_int) {
+ CT_TRACE("bandwith control: %lluB/s "
+ "excess=%llu sleep for "
+ "%lld.%09lds",
+ (unsigned long long)opt.o_bandwidth,
+ (unsigned long long)excess,
+ (long long)delay.tv_sec,
+ delay.tv_nsec);
+ last_bw_print = now;
+ }
+
+ do {
+ rc = nanosleep(&delay, &delay);
+ } while (rc < 0 && errno == EINTR);
+ if (rc < 0) {
+ CT_ERROR(errno, "delay for bandwidth "
+ "control failed to sleep: "
+ "residual=%lld.%09lds",
+ (long long)delay.tv_sec,
+ delay.tv_nsec);
+ rc = 0;
+ }
+ }
}
- if (time(0) >= last_print_time + opt.o_report_int) {
- last_print_time = time(0);
- CT_TRACE("%%"LPU64" ", 100 * wpos / rlen);
- he.length = wpos;
- rc = llapi_hsm_action_progress(hcp, &he, 0);
+ now = time(NULL);
+ if (now >= last_report_time + opt.o_report_int) {
+ last_report_time = now;
+ CT_TRACE("%%%ju ", (uintmax_t)(100 * write_total / length));
+ /* only give the length of the write since the last
+ * progress report */
+ he.length = offset - he.offset;
+ rc = llapi_hsm_action_progress(hcp, &he, length, 0);
if (rc < 0) {
/* Action has been canceled or something wrong
* is happening. Stop copying data. */
" '%s'->'%s' failed", src, dst);
goto out;
}
+ he.offset = offset;
}
rc = 0;
}
rc = ftruncate(dst_fd, src_st.st_size);
if (rc < 0) {
rc = -errno;
- CT_ERROR(rc,
- "cannot truncate '%s' to size %jd",
+ CT_ERROR(rc, "cannot truncate '%s' to size %jd",
dst, (intmax_t)src_st.st_size);
err_major++;
}
}
- free(buf);
+ if (buf != NULL)
+ free(buf);
+
+ CT_TRACE("copied %ju bytes in %f seconds",
+ (uintmax_t)length, ct_now() - start_ct_now);
return rc;
}
}
static int ct_path_lustre(char *buf, int sz, const char *mnt,
- const lustre_fid *fid)
+ const struct lu_fid *fid)
{
- return snprintf(buf, sz, "%s/%s/fid/"DFID_NOBRACE, mnt,
- dot_lustre_name, PFID(fid));
+ return scnprintf(buf, sz, "%s/%s/fid/"DFID_NOBRACE, mnt,
+ dot_lustre_name, PFID(fid));
}
-static int ct_path_archive(char *buf, int sz, const char *archive_dir,
- const lustre_fid *fid)
+static int ct_path_archive_v(char *buf, size_t buf_size,
+ enum ct_archive_format ctaf,
+ const char *archive_path,
+ const struct lu_fid *fid,
+ const char *suffix)
{
- return snprintf(buf, sz, "%s/%04x/%04x/%04x/%04x/%04x/%04x/"
- DFID_NOBRACE, archive_dir,
- (fid)->f_oid & 0xFFFF,
- (fid)->f_oid >> 16 & 0xFFFF,
- (unsigned int)((fid)->f_seq & 0xFFFF),
- (unsigned int)((fid)->f_seq >> 16 & 0xFFFF),
- (unsigned int)((fid)->f_seq >> 32 & 0xFFFF),
- (unsigned int)((fid)->f_seq >> 48 & 0xFFFF),
- PFID(fid));
+ switch (ctaf) {
+ case CT_ARCHIVE_FORMAT_V1:
+ return scnprintf(buf, buf_size,
+ "%s/%04x/%04x/%04x/%04x/%04x/%04x/"DFID_NOBRACE"%s",
+ archive_path,
+ (fid)->f_oid & 0xFFFF,
+ (fid)->f_oid >> 16 & 0xFFFF,
+ (unsigned int)((fid)->f_seq & 0xFFFF),
+ (unsigned int)((fid)->f_seq >> 16 & 0xFFFF),
+ (unsigned int)((fid)->f_seq >> 32 & 0xFFFF),
+ (unsigned int)((fid)->f_seq >> 48 & 0xFFFF),
+ PFID(fid),
+ suffix);
+ case CT_ARCHIVE_FORMAT_V2:
+ return scnprintf(buf, buf_size, "%s/%04x/"DFID_NOBRACE"%s",
+ archive_path,
+ (unsigned int)((fid)->f_oid ^ (fid)->f_seq) & 0xFFFF,
+ PFID(fid),
+ suffix);
+ default:
+ return -EINVAL;
+ }
+}
+
+static int ct_path_archive(char *buf, size_t buf_size,
+ const char *archive_path, const struct lu_fid *fid)
+{
+ return ct_path_archive_v(buf, buf_size, opt.o_archive_format,
+ archive_path, fid, "");
}
static bool ct_is_retryable(int err)
static int ct_fini(struct hsm_copyaction_private **phcp,
const struct hsm_action_item *hai, int hp_flags, int ct_rc)
{
- char lstr[PATH_MAX];
- int rc;
+ struct hsm_copyaction_private *hcp;
+ char lstr[PATH_MAX];
+ int rc;
CT_TRACE("Action completed, notifying coordinator "
- "cookie="LPX64", FID="DFID", hp_flags=%d err=%d",
- hai->hai_cookie, PFID(&hai->hai_fid),
+ "cookie=%#jx, FID="DFID", hp_flags=%d err=%d",
+ (uintmax_t)hai->hai_cookie, PFID(&hai->hai_fid),
hp_flags, -ct_rc);
ct_path_lustre(lstr, sizeof(lstr), opt.o_mnt, &hai->hai_fid);
+
+ if (phcp == NULL || *phcp == NULL) {
+ rc = llapi_hsm_action_begin(&hcp, ctdata, hai, -1, 0, true);
+ if (rc < 0) {
+ CT_ERROR(rc, "llapi_hsm_action_begin() on '%s' failed",
+ lstr);
+ return rc;
+ }
+ phcp = &hcp;
+ }
+
rc = llapi_hsm_action_end(phcp, &hai->hai_extent, hp_flags, abs(ct_rc));
if (rc == -ECANCELED)
CT_ERROR(rc, "completed action on '%s' has been canceled: "
- "cookie="LPX64", FID="DFID, lstr, hai->hai_cookie,
- PFID(&hai->hai_fid));
+ "cookie=%#jx, FID="DFID, lstr,
+ (uintmax_t)hai->hai_cookie, PFID(&hai->hai_fid));
else if (rc < 0)
CT_ERROR(rc, "llapi_hsm_action_end() on '%s' failed", lstr);
else
{
struct hsm_copyaction_private *hcp = NULL;
char src[PATH_MAX];
- char dst[PATH_MAX];
+ char dst[PATH_MAX + 4] = "";
+ char root[PATH_MAX] = "";
int rc;
int rcf = 0;
bool rename_needed = false;
* destination = lustre FID
*/
ct_path_lustre(src, sizeof(src), opt.o_mnt, &hai->hai_dfid);
- ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, &hai->hai_fid);
+ ct_path_archive(root, sizeof(root), opt.o_hsm_root, &hai->hai_fid);
if (hai->hai_extent.length == -1) {
/* whole file, write it to tmp location and atomically
* replace old archived file */
- strncat(dst, "_tmp", sizeof(dst) - strlen(dst) - 1);
+ snprintf(dst, sizeof(dst), "%s_tmp", root);
/* we cannot rely on the same test because ct_copy_data()
* updates hai_extent.length */
rename_needed = true;
+ } else {
+ snprintf(dst, sizeof(dst), "%s", root);
}
CT_TRACE("archiving '%s' to '%s'", src, dst);
src_fd = llapi_hsm_action_get_fd(hcp);
if (src_fd < 0) {
- rc = -errno;
+ rc = src_fd;
CT_ERROR(rc, "cannot open '%s' for read", src);
goto fini_major;
}
- open_flags = O_WRONLY | O_NOFOLLOW | O_NONBLOCK;
+ open_flags = O_WRONLY | O_NOFOLLOW;
/* If extent is specified, don't truncate an old archived copy */
open_flags |= ((hai->hai_extent.length == -1) ? O_TRUNC : 0) | O_CREAT;
}
if (rename_needed == true) {
- char tmp_src[PATH_MAX];
- char tmp_dst[PATH_MAX];
+ char tmp_src[PATH_MAX + 8];
+ char tmp_dst[PATH_MAX + 8];
/* atomically replace old archived file */
ct_path_archive(src, sizeof(src), opt.o_hsm_root,
int linkno = 0;
char *ptr;
int depth = 0;
- int sz;
+ ssize_t sz;
- sprintf(buf, DFID, PFID(&hai->hai_fid));
sprintf(src, "%s/shadow/", opt.o_hsm_root);
ptr = opt.o_hsm_root;
while (*ptr)
(*ptr++ == '/') ? depth-- : 0;
- rc = llapi_fid2path(opt.o_mnt, buf, src + strlen(src),
- sizeof(src) - strlen(src), &recno, &linkno);
+ rc = llapi_fid2path_at(opt.o_mnt_fd, &hai->hai_fid,
+ src + strlen(src),
+ sizeof(src) - strlen(src),
+ &recno, &linkno);
if (rc < 0) {
- CT_ERROR(rc, "cannot get FID of '%s'", buf);
+ CT_ERROR(rc, "cannot get FID of "DFID,
+ PFID(&hai->hai_fid));
rcf = rcf ? rcf : rc;
goto fini_minor;
}
}
/* symlink already exists ? */
sz = readlink(src, buf, sizeof(buf));
+ /* detect truncation */
+ if (sz == sizeof(buf)) {
+ rcf = rcf ? rcf : -E2BIG;
+ CT_ERROR(rcf, "readlink '%s' truncated", src);
+ goto fini_minor;
+ }
if (sz >= 0) {
buf[sz] = '\0';
if (sz == 0 || strncmp(buf, dst, sz) != 0) {
src);
rcf = rcf ? rcf : -errno;
goto fini_minor;
+ }
/* unlink old symlink done */
CT_TRACE("remove old symlink '%s' pointing"
" to '%s'", src, buf);
- }
} else {
/* symlink already ok */
CT_TRACE("symlink '%s' already pointing"
if (!(dst_fd < 0))
close(dst_fd);
- if (hcp != NULL)
- rc = ct_fini(&hcp, hai, hp_flags, rcf);
+ rc = ct_fini(&hcp, hai, hp_flags, rcf);
return rc;
}
int hp_flags = 0;
int src_fd = -1;
int dst_fd = -1;
- int mdt_index = -1; /* Not implemented */
+ int mdt_index = -1;
int open_flags = 0;
bool set_lovea;
struct lu_fid dfid;
/* build backend file name from released file FID */
ct_path_archive(src, sizeof(src), opt.o_hsm_root, &hai->hai_fid);
+ rc = llapi_get_mdt_index_by_fid(opt.o_mnt_fd, &hai->hai_fid,
+ &mdt_index);
+ if (rc < 0) {
+ CT_ERROR(rc, "cannot get mdt index "DFID"",
+ PFID(&hai->hai_fid));
+ return rc;
+ }
/* restore loads and sets the LOVEA w/o interpreting it to avoid
* dependency on the structure format. */
rc = ct_load_stripe(src, lov_buf, &lov_size);
goto fini;
}
- src_fd = open(src, O_RDONLY | O_NOATIME | O_NONBLOCK | O_NOFOLLOW);
+ src_fd = open(src, O_RDONLY | O_NOATIME | O_NOFOLLOW);
if (src_fd < 0) {
rc = -errno;
CT_ERROR(rc, "cannot open '%s' for read", src);
}
dst_fd = llapi_hsm_action_get_fd(hcp);
+ if (dst_fd < 0) {
+ rc = dst_fd;
+ CT_ERROR(rc, "cannot open '%s' for write", dst);
+ goto fini;
+ }
if (set_lovea) {
/* the layout cannot be allocated through .fid so we have to
CT_TRACE("data restore from '%s' to '%s' done", src, dst);
fini:
- if (hcp != NULL)
- rc = ct_fini(&hcp, hai, hp_flags, rc);
+ rc = ct_fini(&hcp, hai, hp_flags, rc);
/* object swaping is done by cdt at copy end, so close of volatile file
* cannot be done before */
static int ct_remove(const struct hsm_action_item *hai, const long hal_flags)
{
struct hsm_copyaction_private *hcp = NULL;
- char dst[PATH_MAX];
+ char dst[PATH_MAX], attr[PATH_MAX + 4];
int rc;
rc = ct_begin(&hcp, hai);
goto fini;
}
- strncat(dst, ".lov", sizeof(dst) - strlen(dst) - 1);
- rc = unlink(dst);
+ snprintf(attr, sizeof(attr), "%s.lov", dst);
+ rc = unlink(attr);
if (rc < 0) {
rc = -errno;
- CT_ERROR(rc, "cannot unlink '%s'", dst);
+ CT_ERROR(rc, "cannot unlink '%s'", attr);
err_minor++;
- goto fini;
+
+ /* ignore the error when lov file does not exist. */
+ if (rc == -ENOENT)
+ rc = 0;
+ else
+ goto fini;
}
fini:
- if (hcp != NULL)
- rc = ct_fini(&hcp, hai, 0, rc);
-
- return rc;
-}
-
-static int ct_report_error(const struct hsm_action_item *hai, int flags,
- int errval)
-{
- struct hsm_copyaction_private *hcp;
- int rc;
-
- rc = llapi_hsm_action_begin(&hcp, ctdata, hai, -1, 0, true);
- if (rc < 0)
- return rc;
-
- rc = llapi_hsm_action_end(&hcp, &hai->hai_extent, flags, abs(errval));
+ rc = ct_fini(&hcp, hai, 0, rc);
return rc;
}
if (opt.o_verbose >= LLAPI_MSG_INFO || opt.o_dry_run) {
/* Print the original path */
- char fid[128];
char path[PATH_MAX];
long long recno = -1;
int linkno = 0;
- sprintf(fid, DFID, PFID(&hai->hai_fid));
- CT_TRACE("'%s' action %s reclen %d, cookie="LPX64,
- fid, hsm_copytool_action2name(hai->hai_action),
- hai->hai_len, hai->hai_cookie);
- rc = llapi_fid2path(opt.o_mnt, fid, path,
- sizeof(path), &recno, &linkno);
+ CT_TRACE(DFID" action %s reclen %d, cookie=%#jx",
+ PFID(&hai->hai_fid),
+ hsm_copytool_action2name(hai->hai_action),
+ hai->hai_len, (uintmax_t)hai->hai_cookie);
+ rc = llapi_fid2path_at(opt.o_mnt_fd, &hai->hai_fid, path,
+ sizeof(path), &recno, &linkno);
if (rc < 0)
- CT_ERROR(rc, "cannot get path of FID %s", fid);
+ CT_ERROR(rc, "cannot get path of "DFID,
+ PFID(&hai->hai_fid));
else
CT_TRACE("processing file '%s'", path);
}
CT_ERROR(rc, "unknown action %d, on '%s'", hai->hai_action,
opt.o_mnt);
err_minor++;
- ct_report_error(hai, 0, rc);
+ ct_fini(NULL, hai, 0, rc);
}
return 0;
static int ct_import_one(const char *src, const char *dst)
{
char newarc[PATH_MAX];
- lustre_fid fid;
+ struct lu_fid fid;
struct stat st;
int rc;
return 0;
rc = llapi_hsm_import(dst,
- opt.o_archive_cnt ? opt.o_archive_id[0] : 0,
- &st, 0, 0, 0, 0, NULL, &fid);
+ opt.o_archive_id_used ? opt.o_archive_id[0] : 0,
+ &st,
+ 0 /* default stripe_size */,
+ -1 /* default stripe offset */,
+ 0 /* default stripe count */,
+ 0 /* stripe pattern (will be RAID0+RELEASED) */,
+ NULL /* pool_name */,
+ &fid);
if (rc < 0) {
CT_ERROR(rc, "cannot import '%s' from '%s'", dst, src);
return rc;
static char *path_concat(const char *dirname, const char *basename)
{
char *result;
- int dirlen = strlen(dirname);
+ int rc;
- result = malloc(dirlen + strlen(basename) + 2);
- if (result == NULL)
+ rc = asprintf(&result, "%s/%s", dirname, basename);
+ if (rc < 0)
return NULL;
- memcpy(result, dirname, dirlen);
- result[dirlen] = '/';
- strcpy(result + dirlen + 1, basename);
-
return result;
}
+static int ct_import_fid(const struct lu_fid *import_fid)
+{
+ char fid_path[PATH_MAX];
+ int rc;
+
+ ct_path_lustre(fid_path, sizeof(fid_path), opt.o_mnt, import_fid);
+ rc = access(fid_path, F_OK);
+ if (rc == 0 || errno != ENOENT) {
+ rc = (errno == 0) ? -EEXIST : -errno;
+ CT_ERROR(rc, "cannot import '"DFID"'", PFID(import_fid));
+ return rc;
+ }
+
+ ct_path_archive(fid_path, sizeof(fid_path), opt.o_hsm_root,
+ import_fid);
+
+ CT_TRACE("Resolving "DFID" to %s", PFID(import_fid), fid_path);
+
+ return ct_import_one(fid_path, opt.o_dst);
+}
+
static int ct_import_recurse(const char *relpath)
{
DIR *dir;
- struct dirent ent, *cookie = NULL;
+ struct dirent *ent;
char *srcpath, *newpath;
+ struct lu_fid import_fid;
int rc;
if (relpath == NULL)
return -EINVAL;
+ /* Is relpath a FID? */
+ rc = llapi_fid_parse(relpath, &import_fid, NULL);
+ if (!rc)
+ return ct_import_fid(&import_fid);
+
srcpath = path_concat(opt.o_hsm_root, relpath);
if (srcpath == NULL) {
err_major++;
}
free(srcpath);
- while (1) {
- rc = readdir_r(dir, &ent, &cookie);
- if (rc != 0) {
- rc = -errno;
- CT_ERROR(rc, "cannot readdir_r '%s'", relpath);
- err_major++;
- goto out;
- } else if ((rc == 0) && (cookie == NULL)) {
- /* end of directory */
- break;
- }
-
- if (!strcmp(ent.d_name, ".") ||
- !strcmp(ent.d_name, ".."))
+ while ((ent = readdir(dir)) != NULL) {
+ if (!strcmp(ent->d_name, ".") ||
+ !strcmp(ent->d_name, ".."))
continue;
/* New relative path */
- newpath = path_concat(relpath, ent.d_name);
+ newpath = path_concat(relpath, ent->d_name);
if (newpath == NULL) {
err_major++;
rc = -ENOMEM;
goto out;
}
- if (ent.d_type == DT_DIR) {
+ if (ent->d_type == DT_DIR) {
rc = ct_import_recurse(newpath);
} else {
char src[PATH_MAX];
return rc;
}
-static int ct_rebind_one(const lustre_fid *old_fid, const lustre_fid *new_fid)
+static int ct_rebind_one(const struct lu_fid *old_fid,
+ const struct lu_fid *new_fid)
{
char src[PATH_MAX];
char dst[PATH_MAX];
ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, new_fid);
if (!opt.o_dry_run) {
+ char src_attr[PATH_MAX + 4];
+ char dst_attr[PATH_MAX + 4];
+
ct_mkdir_p(dst);
if (rename(src, dst)) {
rc = -errno;
return -errno;
}
/* rename lov file */
- strncat(src, ".lov", sizeof(src) - strlen(src) - 1);
- strncat(dst, ".lov", sizeof(dst) - strlen(dst) - 1);
- if (rename(src, dst))
- CT_ERROR(errno, "cannot '%s' rename to '%s'", src, dst);
+ snprintf(src_attr, sizeof(src_attr), "%s.lov", src);
+ snprintf(dst_attr, sizeof(dst_attr), "%s.lov", dst);
+ if (rename(src_attr, dst_attr))
+ CT_ERROR(errno, "cannot rename '%s' to '%s'",
+ src_attr, dst_attr);
}
return 0;
}
-static bool fid_is_file(lustre_fid *fid)
+static bool fid_is_file(struct lu_fid *fid)
{
return fid_is_norm(fid) || fid_is_igif(fid);
}
/* each line consists of 2 FID */
while ((r = getline(&line, &line_size, filp)) != -1) {
- lustre_fid old_fid;
- lustre_fid new_fid;
+ struct lu_fid old_fid;
+ struct lu_fid new_fid;
+ char *next_fid;
/* Ignore empty and commented out ('#...') lines. */
if (should_ignore_line(line))
nl++;
- rc = sscanf(line, SFID" "SFID, RFID(&old_fid), RFID(&new_fid));
- if (rc != 6 || !fid_is_file(&old_fid) ||
- !fid_is_file(&new_fid)) {
- CT_ERROR(EINVAL,
- "'%s' FID expected near '%s', line %u",
- list, line, nl);
+ rc = llapi_fid_parse(line, &old_fid, &next_fid);
+ if (rc)
+ goto error;
+ rc = llapi_fid_parse(next_fid, &new_fid, NULL);
+ if (rc)
+ goto error;
+ if (!fid_is_file(&old_fid) || !fid_is_file(&new_fid))
+ rc = -EINVAL;
+ if (rc) {
+error: CT_ERROR(rc, "%s:%u: two FIDs expected in '%s'",
+ list, nl, line);
err_major++;
continue;
}
if (line)
free(line);
- /* return 0 if all rebinds were sucessful */
+ /* return 0 if all rebinds were successful */
CT_TRACE("%u lines read from '%s', %u rebind successful", nl, list, ok);
return ok == nl ? 0 : -1;
static int ct_rebind(void)
{
- int rc;
+ int rc;
if (opt.o_dst) {
- lustre_fid old_fid;
- lustre_fid new_fid;
+ struct lu_fid old_fid;
+ struct lu_fid new_fid;
- if (sscanf(opt.o_src, SFID, RFID(&old_fid)) != 3 ||
- !fid_is_file(&old_fid)) {
+ rc = llapi_fid_parse(opt.o_src, &old_fid, NULL);
+ if (!rc && !fid_is_file(&old_fid))
rc = -EINVAL;
- CT_ERROR(rc, "'%s' invalid FID format", opt.o_src);
+ if (rc) {
+ CT_ERROR(rc, "invalid source FID '%s'", opt.o_src);
return rc;
}
- if (sscanf(opt.o_dst, SFID, RFID(&new_fid)) != 3 ||
- !fid_is_file(&new_fid)) {
+ rc = llapi_fid_parse(opt.o_dst, &new_fid, NULL);
+ if (!rc && !fid_is_file(&new_fid))
rc = -EINVAL;
- CT_ERROR(rc, "'%s' invalid FID format", opt.o_dst);
+ if (rc) {
+ CT_ERROR(rc, "invalid destination FID '%s'", opt.o_dst);
return rc;
}
return rc;
}
+static int ct_opendirat(int parent_fd, const char *name, DIR **pdir)
+{
+ DIR *dir = NULL;
+ int fd = -1;
+ int rc;
+
+ fd = openat(parent_fd, name, O_RDONLY|O_DIRECTORY);
+ if (fd < 0)
+ return -errno;
+
+ dir = fdopendir(fd);
+ if (dir == NULL) {
+ rc = -errno;
+ goto out;
+ }
+
+ *pdir = dir;
+ fd = -1;
+ rc = 0;
+out:
+ if (!(fd < 0))
+ close(fd);
+
+ return rc;
+}
+
+static int ct_archive_upgrade_reg(int arc_fd, enum ct_archive_format ctaf,
+ int old_fd, const char *name)
+{
+ char new_path[PATH_MAX];
+ struct lu_fid fid;
+ char *split;
+ int scan_count;
+ int suffix_offset = -1;
+ int rc;
+
+ /* Formatted fix with optional suffix. We do not inspect
+ * suffixes. */
+ scan_count = sscanf(name, SFID"%n", RFID(&fid), &suffix_offset);
+ if (scan_count != 3 || suffix_offset < 0) {
+ rc = 0;
+ CT_TRACE("ignoring unexpected file '%s' in archive", name);
+ goto out;
+ }
+
+ ct_path_archive_v(new_path, sizeof(new_path),
+ ctaf, ".", &fid, name + suffix_offset);
+
+ rc = renameat(old_fd, name, arc_fd, new_path);
+ if (rc == 0)
+ goto out;
+
+ if (errno != ENOENT) {
+ rc = -errno;
+ goto out;
+ }
+
+ /* Create parent directory and try again. */
+ split = strrchr(new_path, '/');
+
+ *split = '\0';
+ rc = ct_mkdirat_p(arc_fd, new_path, DIR_PERM);
+ *split = '/';
+ if (rc < 0)
+ goto out;
+
+ rc = renameat(old_fd, name, arc_fd, new_path);
+ if (rc < 0)
+ rc = -errno;
+out:
+ if (rc < 0)
+ CT_ERROR(rc, "cannot rename '%s' to '%s'", name, new_path);
+ else
+ CT_TRACE("renamed '%s' to '%s'", name, new_path);
+
+ return rc;
+}
+
+static const char *d_type_name(unsigned int type)
+{
+ static const char *name[] = {
+ [DT_UNKNOWN] = "unknown",
+ [DT_FIFO] = "fifo",
+ [DT_CHR] = "chr",
+ [DT_DIR] = "dir",
+ [DT_BLK] = "blk",
+ [DT_REG] = "reg",
+ [DT_LNK] = "lnk",
+ [DT_SOCK] = "sock",
+ [DT_WHT] = "wht",
+ };
+
+ if (type < ARRAY_SIZE(name) && name[type] != NULL)
+ return name[type];
+
+ return name[DT_UNKNOWN];
+}
+
+static int ct_archive_upgrade_dir(int arc_fd, enum ct_archive_format ctaf,
+ int parent_fd, const char *dir_name)
+{
+ DIR *dir = NULL;
+ struct dirent *d;
+ int rc = 0;
+ int rc2;
+
+ rc = ct_opendirat(parent_fd, dir_name, &dir);
+ if (rc < 0) {
+ CT_ERROR(rc, "cannot open archive dir '%s'", dir_name);
+ goto out;
+ }
+
+ while ((d = readdir(dir)) != NULL) {
+ CT_TRACE("archive upgrade found %s '%s' (%ld)\n",
+ d_type_name(d->d_type), d->d_name, d->d_ino);
+
+ switch (d->d_type) {
+ case DT_DIR:
+ if (strcmp(d->d_name, ".") == 0 ||
+ strcmp(d->d_name, "..") == 0)
+ continue;
+
+ if (strlen(d->d_name) != 4 ||
+ strspn(d->d_name, "0123456789abcdef") != 4)
+ goto ignore;
+
+ rc2 = ct_archive_upgrade_dir(arc_fd, ctaf,
+ dirfd(dir), d->d_name);
+ if (rc2 < 0) {
+ rc = rc2;
+ CT_ERROR(rc, "cannot upgrade dir '%s' (%ld)",
+ d->d_name, d->d_ino);
+ }
+
+ rc2 = unlinkat(dirfd(dir), d->d_name, AT_REMOVEDIR);
+ CT_TRACE("unlink dir '%s' (%ld): %s",
+ d->d_name, d->d_ino, strerror(rc2 < 0 ? errno: 0));
+ if (rc2 < 0 && errno != ENOTEMPTY)
+ rc = -errno;
+
+ break;
+ case DT_REG:
+ rc2 = ct_archive_upgrade_reg(arc_fd, ctaf,
+ dirfd(dir), d->d_name);
+ if (rc2 < 0)
+ rc = rc2;
+
+ break;
+ default:
+ignore:
+ CT_TRACE("ignoring unexpected %s '%s' (%ld) in archive",
+ d_type_name(d->d_type), d->d_name, d->d_ino);
+ break;
+ }
+ }
+out:
+ if (dir != NULL)
+ closedir(dir);
+
+ return rc;
+}
+
+/* Recursive inplace upgrade (or downgrade) of archive to format
+ * ctaf. Prunes empty archive subdirectories. Idempotent. */
+static int ct_archive_upgrade(int arc_fd, enum ct_archive_format ctaf)
+{
+ /* FIXME Handle shadow tree. */
+ CT_TRACE("upgrade archive to format %s", ct_archive_format_to_str(ctaf));
+
+ return ct_archive_upgrade_dir(arc_fd, ctaf, arc_fd, ".");
+}
+
static int ct_dir_level_max(const char *dirpath, __u16 *sub_seqmax)
{
DIR *dir;
int rc;
__u16 sub_seq;
- struct dirent ent, *cookie = NULL;
+ struct dirent *ent;
*sub_seqmax = 0;
return rc;
}
- while ((rc = readdir_r(dir, &ent, &cookie)) == 0) {
- if (cookie == NULL)
+ do {
+ errno = 0;
+ ent = readdir(dir);
+ if (ent == NULL) {
/* end of directory.
* rc is 0 and seqmax contains the max value. */
+ rc = -errno;
+ if (rc)
+ CT_ERROR(rc, "cannot readdir '%s'", dirpath);
goto out;
+ }
- if (!strcmp(ent.d_name, ".") || !strcmp(ent.d_name, ".."))
+ if (!strcmp(ent->d_name, ".") || !strcmp(ent->d_name, ".."))
continue;
- if (sscanf(ent.d_name, "%hx", &sub_seq) != 1) {
+ if (sscanf(ent->d_name, "%hx", &sub_seq) != 1) {
CT_TRACE("'%s' has an unexpected dirname format, "
- "skip entry", ent.d_name);
+ "skip entry", ent->d_name);
continue;
}
if (sub_seq > *sub_seqmax)
*sub_seqmax = sub_seq;
- }
- rc = -errno;
- CT_ERROR(rc, "cannot readdir_r '%s'", dirpath);
-
+ } while (1);
out:
closedir(dir);
return rc;
static int ct_max_sequence(void)
{
- int rc, i;
- char path[PATH_MAX];
- __u64 seq = 0;
- __u16 subseq;
+ int rc, i;
+ char path[PATH_MAX];
+ __u64 seq = 0;
+ __u16 subseq;
- strncpy(path, opt.o_hsm_root, sizeof(path));
+ snprintf(path, sizeof(path), "%s", opt.o_hsm_root);
/* FID sequence is stored in top-level directory names:
* hsm_root/16bits (high weight)/16 bits/16 bits/16 bits (low weight).
*/
for (i = 0; i < 4; i++) {
+ size_t path_len;
+
rc = ct_dir_level_max(path, &subseq);
if (rc != 0)
return rc;
seq |= ((__u64)subseq << ((3 - i) * 16));
- sprintf(path + strlen(path), "/%04x", subseq);
+ path_len = strlen(path);
+ rc = snprintf(path + path_len, sizeof(path) - path_len,
+ "/%04x", subseq);
+ if (rc >= (sizeof(path) - path_len))
+ return -E2BIG;
+ path[sizeof(path) - 1] = '\0';
}
- printf("max_sequence: %016Lx\n", seq);
+ printf("max_sequence: %#jx\n", (uintmax_t)seq);
return 0;
}
* does successfully unmount and the mount is actually gone, but the
* mtab entry remains. So this just makes mtab happier. */
llapi_hsm_copytool_unregister(&ctdata);
+
+ /* Also remove fifo upon signal as during normal/error exit */
+ if (opt.o_event_fifo != NULL)
+ llapi_hsm_unregister_event_fifo(opt.o_event_fifo);
_exit(1);
}
/* Daemon waits for messages from the kernel; run it in the background. */
static int ct_run(void)
{
- int rc;
+ struct sigaction cleanup_sigaction;
+ int rc;
if (opt.o_daemonize) {
rc = daemon(1, 1);
}
}
- rc = llapi_hsm_copytool_register(&ctdata, opt.o_mnt, 0,
- opt.o_archive_cnt, opt.o_archive_id);
+ if (opt.o_pid_file != NULL) {
+ pid_file_fd = create_pid_file(opt.o_pid_file);
+ if (pid_file_fd < 0) {
+ rc = -errno;
+ CT_ERROR(rc, "cannot create PID file");
+ return rc;
+ }
+ }
+
+ setbuf(stdout, NULL);
+
+ if (opt.o_event_fifo != NULL) {
+ rc = llapi_hsm_register_event_fifo(opt.o_event_fifo);
+ if (rc < 0) {
+ CT_ERROR(rc, "failed to register event fifo");
+ return rc;
+ }
+ llapi_error_callback_set(llapi_hsm_log_error);
+ }
+
+ rc = llapi_hsm_copytool_register(&ctdata, opt.o_mnt,
+ opt.o_archive_id_used,
+ opt.o_archive_id, 0);
if (rc < 0) {
CT_ERROR(rc, "cannot start copytool interface");
return rc;
}
- signal(SIGINT, handler);
- signal(SIGTERM, handler);
+ memset(&cleanup_sigaction, 0, sizeof(cleanup_sigaction));
+ cleanup_sigaction.sa_handler = handler;
+ sigemptyset(&cleanup_sigaction.sa_mask);
+ sigaction(SIGINT, &cleanup_sigaction, NULL);
+ sigaction(SIGTERM, &cleanup_sigaction, NULL);
while (1) {
- struct hsm_action_list *hal;
- struct hsm_action_item *hai;
- int msgsize;
- int i = 0;
+ struct hsm_action_list *hal;
+ struct hsm_action_item *hai;
+ int msgsize;
+ int i = 0;
CT_TRACE("waiting for message from kernel");
if (rc == -ESHUTDOWN) {
CT_TRACE("shutting down");
break;
- } else if (rc == -EAGAIN) {
- continue; /* msg not for us */
} else if (rc < 0) {
CT_WARN("cannot receive action list: %s",
strerror(-rc));
hal->hal_fsname, hal->hal_archive_id, hal->hal_count);
if (strcmp(hal->hal_fsname, fs_name) != 0) {
- CT_ERROR(EINVAL, "'%s' invalid fs name, expecting: %s",
+ rc = -EINVAL;
+ CT_ERROR(rc, "'%s' invalid fs name, expecting: %s",
hal->hal_fsname, fs_name);
err_major++;
if (opt.o_abort_on_error)
hai = hai_next(hai);
}
- llapi_hsm_action_list_free(&hal);
-
if (opt.o_abort_on_error && err_major)
break;
}
llapi_hsm_copytool_unregister(&ctdata);
+ if (opt.o_event_fifo != NULL)
+ llapi_hsm_unregister_event_fifo(opt.o_event_fifo);
return rc;
}
-static int ct_setup(void)
+static int ct_config_get_str(struct cYAML *obj, const char *key, char **pvalue)
{
- int rc;
+ struct cYAML *cy;
+ char *value;
- /* set llapi message level */
- llapi_msg_set_level(opt.o_verbose);
+ if (obj->cy_type != CYAML_TYPE_OBJECT)
+ return -EINVAL;
- arc_fd = open(opt.o_hsm_root, O_RDONLY);
- if (arc_fd < 0) {
+ for (cy = obj->cy_child; cy != NULL; cy = cy->cy_next) {
+ if (cy->cy_string != NULL && strcmp(cy->cy_string, key) == 0) {
+ if (cy->cy_type != CYAML_TYPE_STRING)
+ return -EINVAL;
+
+ if (cy->cy_valuestring == NULL)
+ return -EINVAL;
+
+ value = strdup(cy->cy_valuestring);
+ if (value == NULL)
+ return -ENOMEM;
+
+ *pvalue = value;
+
+ return 0;
+ }
+ }
+
+ return -ENOENT;
+}
+
+static int ct_config_archive_format(struct cYAML *config)
+{
+ char *value = NULL;
+ int rc;
+
+ rc = ct_config_get_str(config, "archive_format", &value);
+ if (rc < 0)
+ return (rc == -ENOENT) ? 0 : rc;
+
+ rc = ct_str_to_archive_format(value, &opt.o_archive_format);
+ if (rc < 0)
+ goto out;
+
+ CT_TRACE("setting archive format to %s",
+ ct_archive_format_to_str(opt.o_archive_format));
+
+out:
+ free(value);
+
+ return 0;
+}
+
+static int ct_config_archive_path(struct cYAML *config)
+{
+ int rc;
+
+ rc = ct_config_get_str(config, "archive_path", &opt.o_hsm_root);
+ if (rc < 0)
+ return (rc == -ENOENT) ? 0 : rc;
+
+ CT_TRACE("setting archive path to '%s'", opt.o_hsm_root);
+
+ return 0;
+}
+
+static int ct_config(const char *path)
+{
+ FILE *file = NULL;
+ struct cYAML *config = NULL;
+ int rc;
+
+ if (path == NULL)
+ return 0;
+
+ file = fopen(path, "r");
+ if (file == NULL) {
rc = -errno;
- CT_ERROR(rc, "cannot open archive at '%s'", opt.o_hsm_root);
- return rc;
+ CT_ERROR(rc, "cannot open '%s'", path);
+ goto out;
+ }
+
+ config = cYAML_load(file, NULL, false);
+ if (config == NULL) {
+ rc = -EINVAL;
+ CT_ERROR(rc, "cannot load archive config from '%s'", path);
+ goto out;
+ }
+
+ rc = ct_config_archive_format(config);
+ if (rc < 0) {
+ CT_ERROR(rc, "cannot load archive format from '%s'", path);
+ goto out;
+ }
+
+ rc = ct_config_archive_path(config);
+ if (rc < 0) {
+ CT_ERROR(rc, "cannot load archive path from '%s'", path);
+ goto out;
}
+out:
+ if (config != NULL)
+ cYAML_free_tree(config);
+
+ if (file != NULL)
+ fclose(file);
+
+ return rc;
+}
+
+static int ct_setup(void)
+{
+ int rc;
+
+ if (opt.o_action == CA_ARCHIVE_UPGRADE)
+ return 0;
rc = llapi_search_fsname(opt.o_mnt, fs_name);
if (rc < 0) {
CT_ERROR(rc, "cannot find a Lustre filesystem mounted at '%s'",
opt.o_mnt);
- return -rc;
+ return rc;
+ }
+
+ opt.o_mnt_fd = open(opt.o_mnt, O_RDONLY);
+ if (opt.o_mnt_fd < 0) {
+ rc = -errno;
+ CT_ERROR(rc, "cannot open mount point at '%s'",
+ opt.o_mnt);
+ return rc;
}
return rc;
{
int rc;
- if (arc_fd < 0)
- return 0;
+ if (opt.o_mnt_fd >= 0) {
+ rc = close(opt.o_mnt_fd);
+ if (rc < 0) {
+ rc = -errno;
+ CT_ERROR(rc, "cannot close mount point");
+ return rc;
+ }
+ }
- if (close(arc_fd) < 0) {
- rc = -errno;
- CT_ERROR(rc, "cannot close archive root directory");
- return rc;
+ if (arc_fd >= 0) {
+ rc = close(arc_fd);
+ if (rc < 0) {
+ rc = -errno;
+ CT_ERROR(rc, "cannot close archive root directory");
+ return rc;
+ }
+ }
+
+ if (opt.o_archive_id_cnt > 0) {
+ free(opt.o_archive_id);
+ opt.o_archive_id = NULL;
+ opt.o_archive_id_cnt = 0;
}
return 0;
{
int rc;
- strncpy(cmd_name, basename(argv[0]), sizeof(cmd_name));
+ snprintf(cmd_name, sizeof(cmd_name), "%s", basename(argv[0]));
rc = ct_parseopts(argc, argv);
if (rc < 0) {
CT_WARN("try '%s --help' for more information", cmd_name);
return -rc;
}
- ct_setup();
+ llapi_msg_set_level(opt.o_verbose);
+
+ rc = ct_config(opt.o_config_path);
+ if (rc < 0)
+ return -rc;
+
+ if (opt.o_hsm_root == NULL) {
+ rc = -EINVAL;
+ CT_ERROR(rc, "must specify a root directory for the backend");
+ return -rc;
+ }
+
+ arc_fd = open(opt.o_hsm_root, O_RDONLY);
+ if (arc_fd < 0) {
+ rc = -errno;
+ CT_ERROR(rc, "cannot open archive at '%s'", opt.o_hsm_root);
+ return rc;
+ }
+
+ rc = ct_setup();
+ if (rc < 0)
+ goto error_cleanup;
switch (opt.o_action) {
case CA_IMPORT:
case CA_MAXSEQ:
rc = ct_max_sequence();
break;
+ case CA_ARCHIVE_UPGRADE:
+ rc = ct_archive_upgrade(arc_fd, opt.o_archive_format);
+ break;
default:
rc = ct_run();
break;
" rc=%d (%s)", err_major, err_minor, rc,
strerror(-rc));
+error_cleanup:
ct_cleanup();
return -rc;
}
-