4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.htm
23 * (C) Copyright 2012 Commissariat a l'energie atomique et aux energies
27 /* HSM copytool program for POSIX filesystem-based HSM's.
29 * An HSM copytool daemon acts on action requests from Lustre to copy files
30 * to and from an HSM archive system. This one in particular makes regular
31 * POSIX filesystem calls to a given path, where an HSM is presumably mounted.
33 * This particular tool can also import an existing HSM archive.
47 #include <sys/xattr.h>
48 #include <sys/syscall.h>
49 #include <sys/types.h>
50 #include <lustre/lustre_idl.h>
51 #include <lustre/lustreapi.h>
53 /* Progress reporting period */
54 #define REPORT_INTERVAL_DEFAULT 30
55 /* HSM hash subdir permissions */
56 #define DIR_PERM S_IRWXU
57 /* HSM hash file permissions */
58 #define FILE_PERM (S_IRUSR | S_IWUSR)
60 #define ONE_MB 0x100000
63 # define NSEC_PER_SEC 1000000000UL
66 /* copytool uses a 32b bitmask field to register with kuc
67 * archive num = 0 => all
68 * archive num from 1 to 32
70 #define MAX_ARCHIVE_CNT (sizeof(__u32) * 8)
87 int o_archive_id[MAX_ARCHIVE_CNT];
89 unsigned long long o_bandwidth;
91 enum ct_action o_action;
96 char *o_src; /* for import, or rebind */
97 char *o_dst; /* for import, or rebind */
100 /* everything else is zeroed */
101 struct options opt = {
104 .o_verbose = LLAPI_MSG_INFO,
106 .o_report_int = REPORT_INTERVAL_DEFAULT,
107 .o_chunk_size = ONE_MB,
110 /* hsm_copytool_private will hold an open FD on the lustre mount point
111 * for us. Additionally open one on the archive FS root to make sure
112 * it doesn't drop out from under us (and remind the admin to shutdown
113 * the copytool before unmounting). */
115 static int arc_fd = -1;
117 static int err_major;
118 static int err_minor;
120 static char cmd_name[PATH_MAX];
121 static char fs_name[MAX_OBD_NAME + 1];
123 static struct hsm_copytool_private *ctdata;
125 static inline double ct_now(void)
129 gettimeofday(&tv, NULL);
131 return tv.tv_sec + 0.000001 * tv.tv_usec;
134 #define CT_ERROR(_rc, _format, ...) \
135 llapi_error(LLAPI_MSG_ERROR, _rc, \
136 "%f %s[%ld]: "_format, \
137 ct_now(), cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
139 #define CT_DEBUG(_format, ...) \
140 llapi_error(LLAPI_MSG_DEBUG | LLAPI_MSG_NO_ERRNO, 0, \
141 "%f %s[%ld]: "_format, \
142 ct_now(), cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
144 #define CT_WARN(_format, ...) \
145 llapi_error(LLAPI_MSG_WARN | LLAPI_MSG_NO_ERRNO, 0, \
146 "%f %s[%ld]: "_format, \
147 ct_now(), cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
149 #define CT_TRACE(_format, ...) \
150 llapi_error(LLAPI_MSG_INFO | LLAPI_MSG_NO_ERRNO, 0, \
151 "%f %s[%ld]: "_format, \
152 ct_now(), cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
154 static void usage(const char *name, int rc)
157 " Usage: %s [options]... <mode> <lustre_mount_point>\n"
158 "The Lustre HSM Posix copy tool can be used as a daemon or "
159 "as a command line tool\n"
160 "The Lustre HSM daemon acts on action requests from Lustre\n"
161 "to copy files to and from an HSM archive system.\n"
162 "This POSIX-flavored daemon makes regular POSIX filesystem calls\n"
163 "to an HSM mounted at a given hsm_root.\n"
164 " --daemon Daemon mode, run in background\n"
166 " --no-attr Don't copy file attributes\n"
167 " --no-shadow Don't create shadow namespace in archive\n"
168 " --no-xattr Don't copy file extended attributes\n"
169 "The Lustre HSM tool performs administrator-type actions\n"
170 "on a Lustre HSM archive.\n"
171 "This POSIX-flavored tool can link an existing HSM namespace\n"
172 "into a Lustre filesystem.\n"
174 " %s [options] --import <src> <dst> <lustre_mount_point>\n"
175 " import an archived subtree from\n"
176 " <src> (FID or relative path to hsm_root) into the Lustre\n"
178 " <dst> (absolute path)\n"
179 " %s [options] --rebind <old_FID> <new_FID> <lustre_mount_point>\n"
180 " rebind an entry in the HSM to a new FID\n"
181 " <old_FID> old FID the HSM entry is bound to\n"
182 " <new_FID> new FID to bind the HSM entry to\n"
183 " %s [options] --rebind <list_file> <lustre_mount_point>\n"
184 " perform the rebind operation for all FID in the list file\n"
185 " each line of <list_file> consists of <old_FID> <new_FID>\n"
186 " %s [options] --max-sequence <fsname>\n"
187 " return the max fid sequence of archived files\n"
188 " --abort-on-error Abort operation on major error\n"
189 " -A, --archive <#> Archive number (repeatable)\n"
190 " -b, --bandwidth <bw> Limit I/O bandwidth (unit can be used\n,"
192 " --dry-run Don't run, just show what would be done\n"
193 " -c, --chunk-size <sz> I/O size used during data copy\n"
194 " (unit can be used, default is MB)\n"
195 " -f, --event-fifo <path> Write events stream to fifo\n"
196 " -p, --hsm-root <path> Target HSM mount point\n"
197 " -q, --quiet Produce less verbose output\n"
198 " -u, --update-interval <s> Interval between progress reports sent\n"
200 " -v, --verbose Produce more verbose output\n",
201 cmd_name, cmd_name, cmd_name, cmd_name, cmd_name);
206 static int ct_parseopts(int argc, char * const *argv)
208 struct option long_opts[] = {
209 {"abort-on-error", no_argument, &opt.o_abort_on_error, 1},
210 {"abort_on_error", no_argument, &opt.o_abort_on_error, 1},
211 {"archive", required_argument, NULL, 'A'},
212 {"bandwidth", required_argument, NULL, 'b'},
213 {"chunk-size", required_argument, NULL, 'c'},
214 {"chunk_size", required_argument, NULL, 'c'},
215 {"daemon", no_argument, &opt.o_daemonize, 1},
216 {"event-fifo", required_argument, NULL, 'f'},
217 {"event_fifo", required_argument, NULL, 'f'},
218 {"dry-run", no_argument, &opt.o_dry_run, 1},
219 {"help", no_argument, NULL, 'h'},
220 {"hsm-root", required_argument, NULL, 'p'},
221 {"hsm_root", required_argument, NULL, 'p'},
222 {"import", no_argument, NULL, 'i'},
223 {"max-sequence", no_argument, NULL, 'M'},
224 {"max_sequence", no_argument, NULL, 'M'},
225 {"no-attr", no_argument, &opt.o_copy_attrs, 0},
226 {"no_attr", no_argument, &opt.o_copy_attrs, 0},
227 {"no-shadow", no_argument, &opt.o_shadow_tree, 0},
228 {"no_shadow", no_argument, &opt.o_shadow_tree, 0},
229 {"no-xattr", no_argument, &opt.o_copy_xattrs, 0},
230 {"no_xattr", no_argument, &opt.o_copy_xattrs, 0},
231 {"quiet", no_argument, NULL, 'q'},
232 {"rebind", no_argument, NULL, 'r'},
233 {"update-interval", required_argument, NULL, 'u'},
234 {"update_interval", required_argument, NULL, 'u'},
235 {"verbose", no_argument, NULL, 'v'},
239 unsigned long long value;
240 unsigned long long unit;
243 while ((c = getopt_long(argc, argv, "A:b:c:f:hiMp:qru:v",
244 long_opts, NULL)) != -1) {
247 if ((opt.o_archive_cnt >= MAX_ARCHIVE_CNT) ||
248 (atoi(optarg) >= MAX_ARCHIVE_CNT)) {
250 CT_ERROR(rc, "archive number must be less"
251 "than %zu", MAX_ARCHIVE_CNT);
254 opt.o_archive_id[opt.o_archive_cnt] = atoi(optarg);
257 case 'b': /* -b and -c have both a number with unit as arg */
260 if (llapi_parse_size(optarg, &value, &unit, 0) < 0) {
262 CT_ERROR(rc, "bad value for -%c '%s'", c,
267 opt.o_chunk_size = value;
269 opt.o_bandwidth = value;
272 opt.o_event_fifo = optarg;
277 opt.o_action = CA_IMPORT;
280 opt.o_action = CA_MAXSEQ;
283 opt.o_hsm_root = optarg;
289 opt.o_action = CA_REBIND;
292 opt.o_report_int = atoi(optarg);
293 if (opt.o_report_int < 0) {
295 CT_ERROR(rc, "bad value for -%c '%s'", c,
310 switch (opt.o_action) {
312 /* src dst mount_point */
313 if (argc != optind + 3) {
315 CT_ERROR(rc, "--import requires 2 arguments");
318 opt.o_src = argv[optind++];
319 opt.o_dst = argv[optind++];
322 /* FID1 FID2 mount_point or FILE mount_point */
323 if (argc == optind + 2) {
324 opt.o_src = argv[optind++];
326 } else if (argc == optind + 3) {
327 opt.o_src = argv[optind++];
328 opt.o_dst = argv[optind++];
331 CT_ERROR(rc, "--rebind requires 1 or 2 arguments");
337 /* just mount point */
341 if (argc != optind + 1) {
343 CT_ERROR(rc, "no mount point specified");
347 opt.o_mnt = argv[optind];
350 CT_TRACE("action=%d src=%s dst=%s mount_point=%s",
351 opt.o_action, opt.o_src, opt.o_dst, opt.o_mnt);
353 if (!opt.o_dry_run && opt.o_hsm_root == NULL) {
355 CT_ERROR(rc, "must specify a root directory for the backend");
359 if (opt.o_action == CA_IMPORT) {
360 if (opt.o_src && opt.o_src[0] == '/') {
363 "source path must be relative to HSM root");
367 if (opt.o_dst && opt.o_dst[0] != '/') {
369 CT_ERROR(rc, "destination path must be absolute");
378 static int ct_mkdir_p(const char *path)
391 while ((ptr = strchr(ptr, '/')) != NULL) {
393 rc = mkdir(saved, DIR_PERM);
395 if (rc < 0 && errno != EEXIST) {
397 CT_ERROR(rc, "cannot mkdir '%s'", path);
409 static int ct_save_stripe(int src_fd, const char *src, const char *dst)
411 char lov_file[PATH_MAX];
412 char lov_buf[XATTR_SIZE_MAX];
413 struct lov_user_md *lum;
418 snprintf(lov_file, sizeof(lov_file), "%s.lov", dst);
419 CT_TRACE("saving stripe info of '%s' in %s", src, lov_file);
421 xattr_size = fgetxattr(src_fd, XATTR_LUSTRE_LOV, lov_buf,
423 if (xattr_size < 0) {
425 CT_ERROR(rc, "cannot get stripe info on '%s'", src);
429 lum = (struct lov_user_md *)lov_buf;
431 if (lum->lmm_magic == LOV_USER_MAGIC_V1 ||
432 lum->lmm_magic == LOV_USER_MAGIC_V3) {
433 /* Set stripe_offset to -1 so that it is not interpreted as a
434 * hint on restore. */
435 lum->lmm_stripe_offset = -1;
438 fd = open(lov_file, O_TRUNC | O_CREAT | O_WRONLY, FILE_PERM);
441 CT_ERROR(rc, "cannot open '%s'", lov_file);
445 rc = write(fd, lum, xattr_size);
448 CT_ERROR(rc, "cannot write %zd bytes to '%s'",
449 xattr_size, lov_file);
457 CT_ERROR(rc, "cannot close '%s'", lov_file);
469 static int ct_load_stripe(const char *src, void *lovea, size_t *lovea_size)
471 char lov_file[PATH_MAX];
475 snprintf(lov_file, sizeof(lov_file), "%s.lov", src);
476 CT_TRACE("reading stripe rules from '%s' for '%s'", lov_file, src);
478 fd = open(lov_file, O_RDONLY);
480 CT_ERROR(errno, "cannot open '%s'", lov_file);
484 rc = read(fd, lovea, *lovea_size);
486 CT_ERROR(errno, "cannot read %zu bytes from '%s'",
487 *lovea_size, lov_file);
498 static int ct_restore_stripe(const char *src, const char *dst, int dst_fd,
499 const void *lovea, size_t lovea_size)
503 rc = fsetxattr(dst_fd, XATTR_LUSTRE_LOV, lovea, lovea_size,
507 CT_ERROR(rc, "cannot set lov EA on '%s'", dst);
513 static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
514 const char *dst, int src_fd, int dst_fd,
515 const struct hsm_action_item *hai, long hal_flags)
517 struct hsm_extent he;
518 __u64 offset = hai->hai_extent.offset;
522 __u64 write_total = 0;
524 time_t last_report_time;
526 double start_ct_now = ct_now();
527 /* Bandwidth Control */
530 time_t last_bw_print;
532 if (fstat(src_fd, &src_st) < 0) {
534 CT_ERROR(rc, "cannot stat '%s'", src);
538 if (!S_ISREG(src_st.st_mode)) {
540 CT_ERROR(rc, "'%s' is not a regular file", src);
544 if (fstat(dst_fd, &dst_st) < 0) {
546 CT_ERROR(rc, "cannot stat '%s'", dst);
550 if (!S_ISREG(dst_st.st_mode)) {
552 CT_ERROR(rc, "'%s' is not a regular file", dst);
556 rc = lseek(src_fd, hai->hai_extent.offset, SEEK_SET);
560 "cannot seek for read to "LPU64" (len %jd) in '%s'",
561 hai->hai_extent.offset, (intmax_t)src_st.st_size, src);
565 /* Don't read beyond a given extent */
566 length = min(hai->hai_extent.length, src_st.st_size);
568 start_time = last_bw_print = last_report_time = time(NULL);
572 rc = llapi_hsm_action_progress(hcp, &he, length, 0);
574 /* Action has been canceled or something wrong
575 * is happening. Stop copying data. */
576 CT_ERROR(rc, "progress ioctl for copy '%s'->'%s' failed",
583 buf = malloc(opt.o_chunk_size);
589 CT_TRACE("start copy of "LPU64" bytes from '%s' to '%s'",
592 while (write_total < length) {
595 int chunk = (length - write_total > opt.o_chunk_size) ?
596 opt.o_chunk_size : length - write_total;
598 rsize = pread(src_fd, buf, chunk, offset);
605 CT_ERROR(rc, "cannot read from '%s'", src);
609 wsize = pwrite(dst_fd, buf, rsize, offset);
612 CT_ERROR(rc, "cannot write to '%s'", dst);
616 write_total += wsize;
620 /* sleep if needed, to honor bandwidth limits */
621 if (opt.o_bandwidth != 0) {
622 unsigned long long write_theory;
624 write_theory = (now - start_time) * opt.o_bandwidth;
626 if (write_theory < write_total) {
627 unsigned long long excess;
628 struct timespec delay;
630 excess = write_total - write_theory;
632 delay.tv_sec = excess / opt.o_bandwidth;
633 delay.tv_nsec = (excess % opt.o_bandwidth) *
634 NSEC_PER_SEC / opt.o_bandwidth;
636 if (now >= last_bw_print + opt.o_report_int) {
637 CT_TRACE("bandwith control: %lluB/s "
638 "excess=%llu sleep for "
640 opt.o_bandwidth, excess,
641 (long long)delay.tv_sec,
647 rc = nanosleep(&delay, &delay);
648 } while (rc < 0 && errno == EINTR);
650 CT_ERROR(errno, "delay for bandwidth "
651 "control failed to sleep: "
652 "residual=%lld.%09lds",
653 (long long)delay.tv_sec,
661 if (now >= last_report_time + opt.o_report_int) {
662 last_report_time = now;
663 CT_TRACE("%%"LPU64" ", 100 * write_total / length);
664 he.length = write_total;
665 rc = llapi_hsm_action_progress(hcp, &he, length, 0);
667 /* Action has been canceled or something wrong
668 * is happening. Stop copying data. */
669 CT_ERROR(rc, "progress ioctl for copy"
670 " '%s'->'%s' failed", src, dst);
679 * truncate restored file
680 * size is taken from the archive this is done to support
681 * restore after a force release which leaves the file with the
682 * wrong size (can big bigger than the new size)
684 if ((hai->hai_action == HSMA_RESTORE) &&
685 (src_st.st_size < dst_st.st_size)) {
687 * make sure the file is on disk before reporting success.
689 rc = ftruncate(dst_fd, src_st.st_size);
692 CT_ERROR(rc, "cannot truncate '%s' to size %jd",
693 dst, (intmax_t)src_st.st_size);
701 CT_TRACE("copied "LPU64" bytes in %f seconds",
702 length, ct_now() - start_ct_now);
707 /* Copy file attributes from file src to file dest */
708 static int ct_copy_attr(const char *src, const char *dst, int src_fd,
712 struct timeval times[2];
715 if (fstat(src_fd, &st) < 0) {
717 CT_ERROR(rc, "cannot stat '%s'", src);
721 times[0].tv_sec = st.st_atime;
722 times[0].tv_usec = 0;
723 times[1].tv_sec = st.st_mtime;
724 times[1].tv_usec = 0;
725 if (fchmod(dst_fd, st.st_mode) < 0 ||
726 fchown(dst_fd, st.st_uid, st.st_gid) < 0 ||
727 futimes(dst_fd, times) < 0) {
729 CT_ERROR(rc, "cannot set attributes of '%s'", src);
736 static int ct_copy_xattr(const char *src, const char *dst, int src_fd,
737 int dst_fd, bool is_restore)
739 char list[XATTR_LIST_MAX];
740 char value[XATTR_SIZE_MAX];
745 list_len = flistxattr(src_fd, list, sizeof(list));
750 while (name < list + list_len) {
751 rc = fgetxattr(src_fd, name, value, sizeof(value));
755 /* when we restore, we do not restore lustre xattr */
757 (strncmp(XATTR_TRUSTED_PREFIX, name,
758 sizeof(XATTR_TRUSTED_PREFIX) - 1) != 0)) {
759 rc = fsetxattr(dst_fd, name, value, rc, 0);
760 CT_TRACE("fsetxattr of '%s' on '%s' rc=%d (%s)",
761 name, dst, rc, strerror(errno));
762 /* lustre.* attrs aren't supported on other FS's */
763 if (rc < 0 && errno != EOPNOTSUPP) {
765 CT_ERROR(rc, "cannot set extended attribute"
771 name += strlen(name) + 1;
777 static int ct_path_lustre(char *buf, int sz, const char *mnt,
778 const lustre_fid *fid)
780 return snprintf(buf, sz, "%s/%s/fid/"DFID_NOBRACE, mnt,
781 dot_lustre_name, PFID(fid));
784 static int ct_path_archive(char *buf, int sz, const char *archive_dir,
785 const lustre_fid *fid)
787 return snprintf(buf, sz, "%s/%04x/%04x/%04x/%04x/%04x/%04x/"
788 DFID_NOBRACE, archive_dir,
789 (fid)->f_oid & 0xFFFF,
790 (fid)->f_oid >> 16 & 0xFFFF,
791 (unsigned int)((fid)->f_seq & 0xFFFF),
792 (unsigned int)((fid)->f_seq >> 16 & 0xFFFF),
793 (unsigned int)((fid)->f_seq >> 32 & 0xFFFF),
794 (unsigned int)((fid)->f_seq >> 48 & 0xFFFF),
798 static bool ct_is_retryable(int err)
800 return err == -ETIMEDOUT;
803 static int ct_begin_restore(struct hsm_copyaction_private **phcp,
804 const struct hsm_action_item *hai,
805 int mdt_index, int open_flags)
810 rc = llapi_hsm_action_begin(phcp, ctdata, hai, mdt_index, open_flags,
813 ct_path_lustre(src, sizeof(src), opt.o_mnt, &hai->hai_fid);
814 CT_ERROR(rc, "llapi_hsm_action_begin() on '%s' failed", src);
820 static int ct_begin(struct hsm_copyaction_private **phcp,
821 const struct hsm_action_item *hai)
823 /* Restore takes specific parameters. Call the same function w/ default
824 * values for all other operations. */
825 return ct_begin_restore(phcp, hai, -1, 0);
828 static int ct_fini(struct hsm_copyaction_private **phcp,
829 const struct hsm_action_item *hai, int hp_flags, int ct_rc)
831 struct hsm_copyaction_private *hcp;
835 CT_TRACE("Action completed, notifying coordinator "
836 "cookie="LPX64", FID="DFID", hp_flags=%d err=%d",
837 hai->hai_cookie, PFID(&hai->hai_fid),
840 ct_path_lustre(lstr, sizeof(lstr), opt.o_mnt, &hai->hai_fid);
842 if (phcp == NULL || *phcp == NULL) {
843 rc = llapi_hsm_action_begin(&hcp, ctdata, hai, -1, 0, true);
845 CT_ERROR(rc, "llapi_hsm_action_begin() on '%s' failed",
852 rc = llapi_hsm_action_end(phcp, &hai->hai_extent, hp_flags, abs(ct_rc));
853 if (rc == -ECANCELED)
854 CT_ERROR(rc, "completed action on '%s' has been canceled: "
855 "cookie="LPX64", FID="DFID, lstr, hai->hai_cookie,
856 PFID(&hai->hai_fid));
858 CT_ERROR(rc, "llapi_hsm_action_end() on '%s' failed", lstr);
860 CT_TRACE("llapi_hsm_action_end() on '%s' ok (rc=%d)",
866 static int ct_archive(const struct hsm_action_item *hai, const long hal_flags)
868 struct hsm_copyaction_private *hcp = NULL;
870 char dst[PATH_MAX] = "";
873 bool rename_needed = false;
879 rc = ct_begin(&hcp, hai);
883 /* we fill archive so:
885 * destination = lustre FID
887 ct_path_lustre(src, sizeof(src), opt.o_mnt, &hai->hai_dfid);
888 ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, &hai->hai_fid);
889 if (hai->hai_extent.length == -1) {
890 /* whole file, write it to tmp location and atomically
891 * replace old archived file */
892 strlcat(dst, "_tmp", sizeof(dst));
893 /* we cannot rely on the same test because ct_copy_data()
894 * updates hai_extent.length */
895 rename_needed = true;
898 CT_TRACE("archiving '%s' to '%s'", src, dst);
905 rc = ct_mkdir_p(dst);
907 CT_ERROR(rc, "mkdir_p '%s' failed", dst);
911 src_fd = llapi_hsm_action_get_fd(hcp);
914 CT_ERROR(rc, "cannot open '%s' for read", src);
918 open_flags = O_WRONLY | O_NOFOLLOW | O_NONBLOCK;
919 /* If extent is specified, don't truncate an old archived copy */
920 open_flags |= ((hai->hai_extent.length == -1) ? O_TRUNC : 0) | O_CREAT;
922 dst_fd = open(dst, open_flags, FILE_PERM);
925 CT_ERROR(rc, "cannot open '%s' for write", dst);
929 /* saving stripe is not critical */
930 rc = ct_save_stripe(src_fd, src, dst);
932 CT_ERROR(rc, "cannot save file striping info of '%s' in '%s'",
935 rc = ct_copy_data(hcp, src, dst, src_fd, dst_fd, hai, hal_flags);
937 CT_ERROR(rc, "data copy failed from '%s' to '%s'", src, dst);
944 CT_ERROR(rc, "cannot flush '%s' archive file '%s'", src, dst);
948 CT_TRACE("data archiving for '%s' to '%s' done", src, dst);
950 /* attrs will remain on the MDS; no need to copy them, except possibly
951 for disaster recovery */
952 if (opt.o_copy_attrs) {
953 rc = ct_copy_attr(src, dst, src_fd, dst_fd);
955 CT_ERROR(rc, "cannot copy attr of '%s' to '%s'",
959 CT_TRACE("attr file for '%s' saved to archive '%s'",
963 /* xattrs will remain on the MDS; no need to copy them, except possibly
964 for disaster recovery */
965 if (opt.o_copy_xattrs) {
966 rc = ct_copy_xattr(src, dst, src_fd, dst_fd, false);
968 CT_ERROR(rc, "cannot copy xattr of '%s' to '%s'",
970 rcf = rcf ? rcf : rc;
972 CT_TRACE("xattr file for '%s' saved to archive '%s'",
976 if (rename_needed == true) {
977 char tmp_src[PATH_MAX];
978 char tmp_dst[PATH_MAX];
980 /* atomically replace old archived file */
981 ct_path_archive(src, sizeof(src), opt.o_hsm_root,
983 rc = rename(dst, src);
986 CT_ERROR(rc, "cannot rename '%s' to '%s'", dst, src);
989 /* rename lov file */
990 snprintf(tmp_src, sizeof(tmp_src), "%s.lov", src);
991 snprintf(tmp_dst, sizeof(tmp_dst), "%s.lov", dst);
992 rc = rename(tmp_dst, tmp_src);
994 CT_ERROR(errno, "cannot rename '%s' to '%s'",
998 if (opt.o_shadow_tree) {
999 /* Create a namespace of softlinks that shadows the original
1000 * Lustre namespace. This will only be current at
1001 * time-of-archive (won't follow renames).
1002 * WARNING: release won't kill these links; a manual
1003 * cleanup of dead links would be required.
1006 long long recno = -1;
1012 sprintf(buf, DFID, PFID(&hai->hai_fid));
1013 sprintf(src, "%s/shadow/", opt.o_hsm_root);
1015 ptr = opt.o_hsm_root;
1017 (*ptr++ == '/') ? depth-- : 0;
1019 rc = llapi_fid2path(opt.o_mnt, buf, src + strlen(src),
1020 sizeof(src) - strlen(src), &recno, &linkno);
1022 CT_ERROR(rc, "cannot get FID of '%s'", buf);
1023 rcf = rcf ? rcf : rc;
1027 /* Figure out how many parent dirs to symlink back */
1030 (*ptr++ == '/') ? depth++ : 0;
1035 ct_path_archive(dst, sizeof(dst), buf, &hai->hai_fid);
1037 if (ct_mkdir_p(src)) {
1038 CT_ERROR(errno, "mkdir_p '%s' failed", src);
1039 rcf = rcf ? rcf : -errno;
1042 /* symlink already exists ? */
1043 sz = readlink(src, buf, sizeof(buf));
1044 /* detect truncation */
1045 if (sz == sizeof(buf)) {
1046 rcf = rcf ? rcf : -E2BIG;
1047 CT_ERROR(rcf, "readlink '%s' truncated", src);
1052 if (sz == 0 || strncmp(buf, dst, sz) != 0) {
1053 if (unlink(src) && errno != ENOENT) {
1055 "cannot unlink symlink '%s'",
1057 rcf = rcf ? rcf : -errno;
1059 /* unlink old symlink done */
1060 CT_TRACE("remove old symlink '%s' pointing"
1061 " to '%s'", src, buf);
1064 /* symlink already ok */
1065 CT_TRACE("symlink '%s' already pointing"
1066 " to '%s'", src, dst);
1071 if (symlink(dst, src)) {
1072 CT_ERROR(errno, "cannot symlink '%s' to '%s'",
1074 rcf = rcf ? rcf : -errno;
1077 CT_TRACE("symlink '%s' to '%s' done", src, dst);
1089 if (ct_is_retryable(rc))
1090 hp_flags |= HP_FLAG_RETRY;
1101 rc = ct_fini(&hcp, hai, hp_flags, rcf);
1106 static int ct_restore(const struct hsm_action_item *hai, const long hal_flags)
1108 struct hsm_copyaction_private *hcp = NULL;
1111 char lov_buf[XATTR_SIZE_MAX];
1112 size_t lov_size = sizeof(lov_buf);
1121 /* we fill lustre so:
1122 * source = lustre FID in the backend
1123 * destination = data FID = volatile file
1126 /* build backend file name from released file FID */
1127 ct_path_archive(src, sizeof(src), opt.o_hsm_root, &hai->hai_fid);
1129 rc = llapi_get_mdt_index_by_fid(opt.o_mnt_fd, &hai->hai_fid,
1132 CT_ERROR(rc, "cannot get mdt index "DFID"",
1133 PFID(&hai->hai_fid));
1136 /* restore loads and sets the LOVEA w/o interpreting it to avoid
1137 * dependency on the structure format. */
1138 rc = ct_load_stripe(src, lov_buf, &lov_size);
1140 CT_WARN("cannot get stripe rules for '%s' (%s), use default",
1141 src, strerror(-rc));
1144 open_flags |= O_LOV_DELAY_CREATE;
1148 rc = ct_begin_restore(&hcp, hai, mdt_index, open_flags);
1152 /* get the FID of the volatile file */
1153 rc = llapi_hsm_action_get_dfid(hcp, &dfid);
1155 CT_ERROR(rc, "restoring "DFID
1156 ", cannot get FID of created volatile file",
1157 PFID(&hai->hai_fid));
1161 /* build volatile "file name", for messages */
1162 snprintf(dst, sizeof(dst), "{VOLATILE}="DFID, PFID(&dfid));
1164 CT_TRACE("restoring data from '%s' to '%s'", src, dst);
1166 if (opt.o_dry_run) {
1171 src_fd = open(src, O_RDONLY | O_NOATIME | O_NONBLOCK | O_NOFOLLOW);
1174 CT_ERROR(rc, "cannot open '%s' for read", src);
1178 dst_fd = llapi_hsm_action_get_fd(hcp);
1181 CT_ERROR(rc, "cannot open '%s' for write", dst);
1186 /* the layout cannot be allocated through .fid so we have to
1187 * restore a layout */
1188 rc = ct_restore_stripe(src, dst, dst_fd, lov_buf, lov_size);
1190 CT_ERROR(rc, "cannot restore file striping info"
1191 " for '%s' from '%s'", dst, src);
1197 rc = ct_copy_data(hcp, src, dst, src_fd, dst_fd, hai, hal_flags);
1199 CT_ERROR(rc, "cannot copy data from '%s' to '%s'",
1202 if (ct_is_retryable(rc))
1203 hp_flags |= HP_FLAG_RETRY;
1207 CT_TRACE("data restore from '%s' to '%s' done", src, dst);
1210 rc = ct_fini(&hcp, hai, hp_flags, rc);
1212 /* object swaping is done by cdt at copy end, so close of volatile file
1213 * cannot be done before */
1223 static int ct_remove(const struct hsm_action_item *hai, const long hal_flags)
1225 struct hsm_copyaction_private *hcp = NULL;
1229 rc = ct_begin(&hcp, hai);
1233 ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, &hai->hai_fid);
1235 CT_TRACE("removing file '%s'", dst);
1237 if (opt.o_dry_run) {
1245 CT_ERROR(rc, "cannot unlink '%s'", dst);
1250 strlcat(dst, ".lov", sizeof(dst));
1254 CT_ERROR(rc, "cannot unlink '%s'", dst);
1260 rc = ct_fini(&hcp, hai, 0, rc);
1265 static int ct_process_item(struct hsm_action_item *hai, const long hal_flags)
1269 if (opt.o_verbose >= LLAPI_MSG_INFO || opt.o_dry_run) {
1270 /* Print the original path */
1272 char path[PATH_MAX];
1273 long long recno = -1;
1276 sprintf(fid, DFID, PFID(&hai->hai_fid));
1277 CT_TRACE("'%s' action %s reclen %d, cookie="LPX64,
1278 fid, hsm_copytool_action2name(hai->hai_action),
1279 hai->hai_len, hai->hai_cookie);
1280 rc = llapi_fid2path(opt.o_mnt, fid, path,
1281 sizeof(path), &recno, &linkno);
1283 CT_ERROR(rc, "cannot get path of FID %s", fid);
1285 CT_TRACE("processing file '%s'", path);
1288 switch (hai->hai_action) {
1289 /* set err_major, minor inside these functions */
1291 rc = ct_archive(hai, hal_flags);
1294 rc = ct_restore(hai, hal_flags);
1297 rc = ct_remove(hai, hal_flags);
1300 CT_TRACE("cancel not implemented for file system '%s'",
1302 /* Don't report progress to coordinator for this cookie:
1303 * the copy function will get ECANCELED when reporting
1310 CT_ERROR(rc, "unknown action %d, on '%s'", hai->hai_action,
1313 ct_fini(NULL, hai, 0, rc);
1321 struct hsm_action_item *hai;
1324 static void *ct_thread(void *data)
1326 struct ct_th_data *cttd = data;
1329 rc = ct_process_item(cttd->hai, cttd->hal_flags);
1333 pthread_exit((void *)(intptr_t)rc);
1336 static int ct_process_item_async(const struct hsm_action_item *hai,
1339 pthread_attr_t attr;
1341 struct ct_th_data *data;
1344 data = malloc(sizeof(*data));
1348 data->hai = malloc(hai->hai_len);
1349 if (data->hai == NULL) {
1354 memcpy(data->hai, hai, hai->hai_len);
1355 data->hal_flags = hal_flags;
1357 rc = pthread_attr_init(&attr);
1359 CT_ERROR(rc, "pthread_attr_init failed for '%s' service",
1366 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1368 rc = pthread_create(&thread, &attr, ct_thread, data);
1370 CT_ERROR(rc, "cannot create thread for '%s' service",
1373 pthread_attr_destroy(&attr);
1377 static int ct_import_one(const char *src, const char *dst)
1379 char newarc[PATH_MAX];
1384 CT_TRACE("importing '%s' from '%s'", dst, src);
1386 if (stat(src, &st) < 0) {
1388 CT_ERROR(rc, "cannot stat '%s'", src);
1395 rc = llapi_hsm_import(dst,
1396 opt.o_archive_cnt ? opt.o_archive_id[0] : 0,
1397 &st, 0, 0, 0, 0, NULL, &fid);
1399 CT_ERROR(rc, "cannot import '%s' from '%s'", dst, src);
1403 ct_path_archive(newarc, sizeof(newarc), opt.o_hsm_root, &fid);
1405 rc = ct_mkdir_p(newarc);
1407 CT_ERROR(rc, "mkdir_p '%s' failed", newarc);
1413 /* Lots of choices now: mv, ln, ln -s ? */
1414 rc = link(src, newarc); /* hardlink */
1417 CT_ERROR(rc, "cannot link '%s' to '%s'", newarc, src);
1421 CT_TRACE("imported '%s' from '%s'=='%s'", dst, newarc, src);
1426 static char *path_concat(const char *dirname, const char *basename)
1431 rc = asprintf(&result, "%s/%s", dirname, basename);
1438 static int ct_import_fid(const lustre_fid *import_fid)
1440 char fid_path[PATH_MAX];
1443 ct_path_lustre(fid_path, sizeof(fid_path), opt.o_mnt, import_fid);
1444 rc = access(fid_path, F_OK);
1445 if (rc == 0 || errno != ENOENT) {
1446 rc = (errno == 0) ? -EEXIST : -errno;
1447 CT_ERROR(rc, "cannot import '"DFID"'", PFID(import_fid));
1451 ct_path_archive(fid_path, sizeof(fid_path), opt.o_hsm_root,
1454 CT_TRACE("Resolving "DFID" to %s", PFID(import_fid), fid_path);
1456 return ct_import_one(fid_path, opt.o_dst);
1459 static int ct_import_recurse(const char *relpath)
1462 struct dirent ent, *cookie = NULL;
1463 char *srcpath, *newpath;
1464 lustre_fid import_fid;
1467 if (relpath == NULL)
1470 /* Is relpath a FID? In which case SFID should expand to three
1472 rc = sscanf(relpath, SFID, RFID(&import_fid));
1474 return ct_import_fid(&import_fid);
1476 srcpath = path_concat(opt.o_hsm_root, relpath);
1477 if (srcpath == NULL) {
1482 dir = opendir(srcpath);
1484 /* Not a dir, or error */
1485 if (errno == ENOTDIR) {
1486 /* Single regular file case, treat o_dst as absolute
1488 rc = ct_import_one(srcpath, opt.o_dst);
1491 CT_ERROR(rc, "cannot opendir '%s'", srcpath);
1500 rc = readdir_r(dir, &ent, &cookie);
1503 CT_ERROR(rc, "cannot readdir_r '%s'", relpath);
1506 } else if ((rc == 0) && (cookie == NULL)) {
1507 /* end of directory */
1511 if (!strcmp(ent.d_name, ".") ||
1512 !strcmp(ent.d_name, ".."))
1515 /* New relative path */
1516 newpath = path_concat(relpath, ent.d_name);
1517 if (newpath == NULL) {
1523 if (ent.d_type == DT_DIR) {
1524 rc = ct_import_recurse(newpath);
1529 sprintf(src, "%s/%s", opt.o_hsm_root, newpath);
1530 sprintf(dst, "%s/%s", opt.o_dst, newpath);
1531 /* Make the target dir in the Lustre fs */
1532 rc = ct_mkdir_p(dst);
1534 /* Import the file */
1535 rc = ct_import_one(src, dst);
1537 CT_ERROR(rc, "ct_mkdir_p '%s' failed", dst);
1543 CT_ERROR(rc, "cannot import '%s'", newpath);
1544 if (err_major && opt.o_abort_on_error) {
1558 static int ct_rebind_one(const lustre_fid *old_fid, const lustre_fid *new_fid)
1564 CT_TRACE("rebind "DFID" to "DFID, PFID(old_fid), PFID(new_fid));
1566 ct_path_archive(src, sizeof(src), opt.o_hsm_root, old_fid);
1567 ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, new_fid);
1569 if (!opt.o_dry_run) {
1571 if (rename(src, dst)) {
1573 CT_ERROR(rc, "cannot rename '%s' to '%s'", src, dst);
1576 /* rename lov file */
1577 strlcat(src, ".lov", sizeof(src));
1578 strlcat(dst, ".lov", sizeof(dst));
1579 if (rename(src, dst))
1580 CT_ERROR(errno, "cannot rename '%s' to '%s'", src, dst);
1586 static bool fid_is_file(lustre_fid *fid)
1588 return fid_is_norm(fid) || fid_is_igif(fid);
1591 static bool should_ignore_line(const char *line)
1595 for (i = 0; line[i] != '\0'; i++) {
1596 if (isspace(line[i]))
1598 else if (line[i] == '#')
1607 static int ct_rebind_list(const char *list)
1613 size_t line_size = 0;
1614 unsigned int nl = 0;
1615 unsigned int ok = 0;
1617 filp = fopen(list, "r");
1620 CT_ERROR(rc, "cannot open '%s'", list);
1624 /* each line consists of 2 FID */
1625 while ((r = getline(&line, &line_size, filp)) != -1) {
1629 /* Ignore empty and commented out ('#...') lines. */
1630 if (should_ignore_line(line))
1635 rc = sscanf(line, SFID" "SFID, RFID(&old_fid), RFID(&new_fid));
1636 if (rc != 6 || !fid_is_file(&old_fid) ||
1637 !fid_is_file(&new_fid)) {
1639 "'%s' FID expected near '%s', line %u",
1645 if (ct_rebind_one(&old_fid, &new_fid))
1656 /* return 0 if all rebinds were sucessful */
1657 CT_TRACE("%u lines read from '%s', %u rebind successful", nl, list, ok);
1659 return ok == nl ? 0 : -1;
1662 static int ct_rebind(void)
1670 if (sscanf(opt.o_src, SFID, RFID(&old_fid)) != 3 ||
1671 !fid_is_file(&old_fid)) {
1673 CT_ERROR(rc, "'%s' invalid FID format", opt.o_src);
1677 if (sscanf(opt.o_dst, SFID, RFID(&new_fid)) != 3 ||
1678 !fid_is_file(&new_fid)) {
1680 CT_ERROR(rc, "'%s' invalid FID format", opt.o_dst);
1684 rc = ct_rebind_one(&old_fid, &new_fid);
1689 /* o_src is a list file */
1690 rc = ct_rebind_list(opt.o_src);
1695 static int ct_dir_level_max(const char *dirpath, __u16 *sub_seqmax)
1700 struct dirent ent, *cookie = NULL;
1704 dir = opendir(dirpath);
1707 CT_ERROR(rc, "cannot open directory '%s'", opt.o_hsm_root);
1711 while ((rc = readdir_r(dir, &ent, &cookie)) == 0) {
1713 /* end of directory.
1714 * rc is 0 and seqmax contains the max value. */
1717 if (!strcmp(ent.d_name, ".") || !strcmp(ent.d_name, ".."))
1720 if (sscanf(ent.d_name, "%hx", &sub_seq) != 1) {
1721 CT_TRACE("'%s' has an unexpected dirname format, "
1722 "skip entry", ent.d_name);
1725 if (sub_seq > *sub_seqmax)
1726 *sub_seqmax = sub_seq;
1729 CT_ERROR(rc, "cannot readdir_r '%s'", dirpath);
1736 static int ct_max_sequence(void)
1739 char path[PATH_MAX];
1743 strlcpy(path, opt.o_hsm_root, sizeof(path));
1744 /* FID sequence is stored in top-level directory names:
1745 * hsm_root/16bits (high weight)/16 bits/16 bits/16 bits (low weight).
1747 for (i = 0; i < 4; i++) {
1750 rc = ct_dir_level_max(path, &subseq);
1753 seq |= ((__u64)subseq << ((3 - i) * 16));
1754 path_len = strlen(path);
1755 rc = snprintf(path + path_len, sizeof(path) - path_len,
1757 if (rc >= (sizeof(path) - path_len))
1759 path[sizeof(path) - 1] = '\0';
1762 printf("max_sequence: "LPX64"\n", seq);
1767 static void handler(int signal)
1769 psignal(signal, "exiting");
1770 /* If we don't clean up upon interrupt, umount thinks there's a ref
1771 * and doesn't remove us from mtab (EINPROGRESS). The lustre client
1772 * does successfully unmount and the mount is actually gone, but the
1773 * mtab entry remains. So this just makes mtab happier. */
1774 llapi_hsm_copytool_unregister(&ctdata);
1778 /* Daemon waits for messages from the kernel; run it in the background. */
1779 static int ct_run(void)
1783 if (opt.o_daemonize) {
1787 CT_ERROR(rc, "cannot daemonize");
1792 setbuf(stdout, NULL);
1794 if (opt.o_event_fifo != NULL) {
1795 rc = llapi_hsm_register_event_fifo(opt.o_event_fifo);
1797 CT_ERROR(rc, "failed to register event fifo");
1800 llapi_error_callback_set(llapi_hsm_log_error);
1803 rc = llapi_hsm_copytool_register(&ctdata, opt.o_mnt,
1805 opt.o_archive_id, 0);
1807 CT_ERROR(rc, "cannot start copytool interface");
1811 signal(SIGINT, handler);
1812 signal(SIGTERM, handler);
1815 struct hsm_action_list *hal;
1816 struct hsm_action_item *hai;
1820 CT_TRACE("waiting for message from kernel");
1822 rc = llapi_hsm_copytool_recv(ctdata, &hal, &msgsize);
1823 if (rc == -ESHUTDOWN) {
1824 CT_TRACE("shutting down");
1826 } else if (rc < 0) {
1827 CT_WARN("cannot receive action list: %s",
1830 if (opt.o_abort_on_error)
1836 CT_TRACE("copytool fs=%s archive#=%d item_count=%d",
1837 hal->hal_fsname, hal->hal_archive_id, hal->hal_count);
1839 if (strcmp(hal->hal_fsname, fs_name) != 0) {
1841 CT_ERROR(rc, "'%s' invalid fs name, expecting: %s",
1842 hal->hal_fsname, fs_name);
1844 if (opt.o_abort_on_error)
1850 hai = hai_first(hal);
1851 while (++i <= hal->hal_count) {
1852 if ((char *)hai - (char *)hal > msgsize) {
1855 "'%s' item %d past end of message!",
1860 rc = ct_process_item_async(hai, hal->hal_flags);
1862 CT_ERROR(rc, "'%s' item %d process",
1864 if (opt.o_abort_on_error && err_major)
1866 hai = hai_next(hai);
1869 if (opt.o_abort_on_error && err_major)
1873 llapi_hsm_copytool_unregister(&ctdata);
1874 if (opt.o_event_fifo != NULL)
1875 llapi_hsm_unregister_event_fifo(opt.o_event_fifo);
1880 static int ct_setup(void)
1884 /* set llapi message level */
1885 llapi_msg_set_level(opt.o_verbose);
1887 arc_fd = open(opt.o_hsm_root, O_RDONLY);
1890 CT_ERROR(rc, "cannot open archive at '%s'", opt.o_hsm_root);
1894 rc = llapi_search_fsname(opt.o_mnt, fs_name);
1896 CT_ERROR(rc, "cannot find a Lustre filesystem mounted at '%s'",
1901 opt.o_mnt_fd = open(opt.o_mnt, O_RDONLY);
1902 if (opt.o_mnt_fd < 0) {
1904 CT_ERROR(rc, "cannot open mount point at '%s'",
1912 static int ct_cleanup(void)
1916 if (opt.o_mnt_fd >= 0) {
1917 rc = close(opt.o_mnt_fd);
1920 CT_ERROR(rc, "cannot close mount point");
1929 CT_ERROR(rc, "cannot close archive root directory");
1937 int main(int argc, char **argv)
1941 strlcpy(cmd_name, basename(argv[0]), sizeof(cmd_name));
1942 rc = ct_parseopts(argc, argv);
1944 CT_WARN("try '%s --help' for more information", cmd_name);
1952 switch (opt.o_action) {
1954 rc = ct_import_recurse(opt.o_src);
1960 rc = ct_max_sequence();
1967 if (opt.o_action != CA_MAXSEQ)
1968 CT_TRACE("process finished, errs: %d major, %d minor,"
1969 " rc=%d (%s)", err_major, err_minor, rc,