* (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.
*
#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 <libcfs/util/string.h>
#include <lustre/lustre_idl.h>
#include <lustre/lustreapi.h>
#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,
int o_verbose;
int o_copy_xattrs;
int o_archive_cnt;
- int o_archive_id[MAX_ARCHIVE_CNT];
+ int o_archive_id[LL_HSM_MAX_ARCHIVE];
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 */
static struct hsm_copytool_private *ctdata;
+static inline double ct_now(void)
+{
+ struct timeval tv;
+
+ gettimeofday(&tv, NULL);
-#define CT_ERROR(_rc, _format, ...) \
+ return tv.tv_sec + 0.000001 * tv.tv_usec;
+}
+
+#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)
{
long_opts, NULL)) != -1) {
switch (c) {
case 'A':
- if ((opt.o_archive_cnt >= MAX_ARCHIVE_CNT) ||
- (atoi(optarg) >= MAX_ARCHIVE_CNT)) {
+ if ((opt.o_archive_cnt >= LL_HSM_MAX_ARCHIVE) ||
+ (atoi(optarg) >= LL_HSM_MAX_ARCHIVE)) {
rc = -E2BIG;
CT_ERROR(rc, "archive number must be less"
- "than %zu", MAX_ARCHIVE_CNT);
+ "than %zu", LL_HSM_MAX_ARCHIVE);
return rc;
}
opt.o_archive_id[opt.o_archive_cnt] = atoi(optarg);
}
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) {
+ if (opt.o_hsm_root == NULL) {
rc = -EINVAL;
CT_ERROR(rc, "must specify a root directory for the backend");
return rc;
int rc;
ptr = strdup(path);
+ if (ptr == NULL)
+ return -errno;
+
saved = ptr;
while (*ptr == '/')
ptr++;
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;
}
-static void bandwidth_ctl_delay(int wsize)
-{
- static unsigned long long tot_bytes;
- static time_t start_time;
- static time_t last_time;
- time_t now = time(0);
- double tot_time;
- double excess;
- unsigned int sleep_time;
-
- if (now > last_time + 5) {
- tot_bytes = 0;
- start_time = last_time = now;
- }
-
- tot_bytes += wsize;
- tot_time = now - start_time;
- if (tot_time < 1)
- tot_time = 1;
-
- 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 (excess > 0)
- usleep(sleep_time);
-
- last_time = now;
-}
-
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 stat dst_st;
char *buf = NULL;
__u64 write_total = 0;
- __u64 length;
- time_t last_print_time = time(NULL);
+ __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);
+ 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) {
rc = -errno;
return rc;
}
+ 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) {
rc = -errno;
CT_ERROR(rc, "cannot stat '%s'", dst);
return rc;
}
- rc = lseek(src_fd, hai->hai_extent.offset, SEEK_SET);
- if (rc < 0) {
- rc = -errno;
- CT_ERROR(rc,
- "cannot seek for read to "LPU64" (len %jd) in '%s'",
- hai->hai_extent.offset, (intmax_t)src_st.st_size, src);
- return rc;
- }
-
/* Don't read beyond a given extent */
- length = min(hai->hai_extent.length, src_st.st_size);
+ 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 = offset;
he.length = 0;
goto out;
}
- CT_DEBUG("Going to copy "LPU64" bytes %s -> %s\n", length, src, dst);
+ CT_TRACE("start copy of %ju bytes from '%s' to '%s'",
+ (uintmax_t)length, src, dst);
while (write_total < length) {
ssize_t rsize;
write_total += wsize;
offset += wsize;
- if (opt.o_bandwidth != 0)
- /* sleep if needed, to honor bandwidth limits */
- bandwidth_ctl_delay(wsize);
+ now = time(NULL);
+ /* sleep if needed, to honor bandwidth limits */
+ if (opt.o_bandwidth != 0) {
+ unsigned long long write_theory;
+
+ write_theory = (now - start_time) * opt.o_bandwidth;
+
+ if (write_theory < write_total) {
+ unsigned long long excess;
+ struct timespec delay;
+
+ excess = write_total - write_theory;
+
+ 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",
+ opt.o_bandwidth, 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 * write_total / length);
- he.length = write_total;
+ 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
" '%s'->'%s' failed", src, dst);
goto out;
}
+ he.offset = offset;
}
rc = 0;
}
if (buf != NULL)
free(buf);
+ CT_TRACE("copied %ju bytes in %f seconds",
+ (uintmax_t)length, ct_now() - start_ct_now);
+
return rc;
}
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);
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
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);
+ strlcat(dst, "_tmp", sizeof(dst));
/* we cannot rely on the same test because ct_copy_data()
* updates hai_extent.length */
rename_needed = true;
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;
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
goto fini;
}
- strncat(dst, ".lov", sizeof(dst) - strlen(dst) - 1);
+ strlcat(dst, ".lov", sizeof(dst));
rc = unlink(dst);
if (rc < 0) {
rc = -errno;
int linkno = 0;
sprintf(fid, DFID, PFID(&hai->hai_fid));
- CT_TRACE("'%s' action %s reclen %d, cookie="LPX64,
+ CT_TRACE("'%s' action %s reclen %d, cookie=%#jx",
fid, hsm_copytool_action2name(hai->hai_action),
- hai->hai_len, hai->hai_cookie);
+ hai->hai_len, (uintmax_t)hai->hai_cookie);
rc = llapi_fid2path(opt.o_mnt, fid, path,
sizeof(path), &recno, &linkno);
if (rc < 0)
return -errno;
}
/* rename lov file */
- strncat(src, ".lov", sizeof(src) - strlen(src) - 1);
- strncat(dst, ".lov", sizeof(dst) - strlen(dst) - 1);
+ strlcat(src, ".lov", sizeof(src));
+ strlcat(dst, ".lov", sizeof(dst));
if (rename(src, dst))
CT_ERROR(errno, "cannot rename '%s' to '%s'", src, dst);
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_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));
+ strlcpy(path, opt.o_hsm_root, sizeof(path));
/* 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: "LPX64"\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);
llapi_error_callback_set(llapi_hsm_log_error);
}
- rc = llapi_hsm_copytool_register(&ctdata, opt.o_mnt, 0,
- opt.o_archive_cnt, opt.o_archive_id);
+ rc = llapi_hsm_copytool_register(&ctdata, opt.o_mnt,
+ opt.o_archive_cnt,
+ 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));
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;
+ }
}
return 0;
{
int rc;
- strncpy(cmd_name, basename(argv[0]), sizeof(cmd_name));
+ strlcpy(cmd_name, basename(argv[0]), sizeof(cmd_name));
rc = ct_parseopts(argc, argv);
if (rc < 0) {
CT_WARN("try '%s --help' for more information", cmd_name);
return -rc;
}
- ct_setup();
+ rc = ct_setup();
+ if (rc < 0)
+ goto error_cleanup;
switch (opt.o_action) {
case CA_IMPORT:
" rc=%d (%s)", err_major, err_minor, rc,
strerror(-rc));
+error_cleanup:
ct_cleanup();
return -rc;