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
26 * Copyright (c) 2013, 2014, Intel Corporation.
28 /* HSM copytool program for POSIX filesystem-based HSM's.
30 * An HSM copytool daemon acts on action requests from Lustre to copy files
31 * to and from an HSM archive system. This one in particular makes regular
32 * POSIX filesystem calls to a given path, where an HSM is presumably mounted.
34 * This particular tool can also import an existing HSM archive.
49 #include <sys/xattr.h>
50 #include <sys/syscall.h>
51 #include <sys/types.h>
52 #include <lustre/lustre_idl.h>
53 #include <lustre/lustreapi.h>
55 /* Progress reporting period */
56 #define REPORT_INTERVAL_DEFAULT 30
57 /* HSM hash subdir permissions */
58 #define DIR_PERM S_IRWXU
59 /* HSM hash file permissions */
60 #define FILE_PERM (S_IRUSR | S_IWUSR)
62 #define ONE_MB 0x100000
65 # define NSEC_PER_SEC 1000000000UL
83 int o_archive_id[LL_HSM_MAX_ARCHIVE];
85 unsigned long long o_bandwidth;
87 enum ct_action o_action;
92 char *o_src; /* for import, or rebind */
93 char *o_dst; /* for import, or rebind */
96 /* everything else is zeroed */
97 struct options opt = {
100 .o_verbose = LLAPI_MSG_INFO,
102 .o_report_int = REPORT_INTERVAL_DEFAULT,
103 .o_chunk_size = ONE_MB,
106 /* hsm_copytool_private will hold an open FD on the lustre mount point
107 * for us. Additionally open one on the archive FS root to make sure
108 * it doesn't drop out from under us (and remind the admin to shutdown
109 * the copytool before unmounting). */
111 static int arc_fd = -1;
113 static int err_major;
114 static int err_minor;
116 static char cmd_name[PATH_MAX];
117 static char fs_name[MAX_OBD_NAME + 1];
119 static struct hsm_copytool_private *ctdata;
121 static inline double ct_now(void)
125 gettimeofday(&tv, NULL);
127 return tv.tv_sec + 0.000001 * tv.tv_usec;
130 #define CT_ERROR(_rc, _format, ...) \
131 llapi_error(LLAPI_MSG_ERROR, _rc, \
132 "%f %s[%ld]: "_format, \
133 ct_now(), cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
135 #define CT_DEBUG(_format, ...) \
136 llapi_error(LLAPI_MSG_DEBUG | LLAPI_MSG_NO_ERRNO, 0, \
137 "%f %s[%ld]: "_format, \
138 ct_now(), cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
140 #define CT_WARN(_format, ...) \
141 llapi_error(LLAPI_MSG_WARN | LLAPI_MSG_NO_ERRNO, 0, \
142 "%f %s[%ld]: "_format, \
143 ct_now(), cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
145 #define CT_TRACE(_format, ...) \
146 llapi_error(LLAPI_MSG_INFO | LLAPI_MSG_NO_ERRNO, 0, \
147 "%f %s[%ld]: "_format, \
148 ct_now(), cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
150 static void usage(const char *name, int rc)
153 " Usage: %s [options]... <mode> <lustre_mount_point>\n"
154 "The Lustre HSM Posix copy tool can be used as a daemon or "
155 "as a command line tool\n"
156 "The Lustre HSM daemon acts on action requests from Lustre\n"
157 "to copy files to and from an HSM archive system.\n"
158 "This POSIX-flavored daemon makes regular POSIX filesystem calls\n"
159 "to an HSM mounted at a given hsm_root.\n"
160 " --daemon Daemon mode, run in background\n"
162 " --no-attr Don't copy file attributes\n"
163 " --no-shadow Don't create shadow namespace in archive\n"
164 " --no-xattr Don't copy file extended attributes\n"
165 "The Lustre HSM tool performs administrator-type actions\n"
166 "on a Lustre HSM archive.\n"
167 "This POSIX-flavored tool can link an existing HSM namespace\n"
168 "into a Lustre filesystem.\n"
170 " %s [options] --import <src> <dst> <lustre_mount_point>\n"
171 " import an archived subtree from\n"
172 " <src> (FID or relative path to hsm_root) into the Lustre\n"
174 " <dst> (absolute path)\n"
175 " %s [options] --rebind <old_FID> <new_FID> <lustre_mount_point>\n"
176 " rebind an entry in the HSM to a new FID\n"
177 " <old_FID> old FID the HSM entry is bound to\n"
178 " <new_FID> new FID to bind the HSM entry to\n"
179 " %s [options] --rebind <list_file> <lustre_mount_point>\n"
180 " perform the rebind operation for all FID in the list file\n"
181 " each line of <list_file> consists of <old_FID> <new_FID>\n"
182 " %s [options] --max-sequence <fsname>\n"
183 " return the max fid sequence of archived files\n"
184 " --abort-on-error Abort operation on major error\n"
185 " -A, --archive <#> Archive number (repeatable)\n"
186 " -b, --bandwidth <bw> Limit I/O bandwidth (unit can be used\n,"
188 " --dry-run Don't run, just show what would be done\n"
189 " -c, --chunk-size <sz> I/O size used during data copy\n"
190 " (unit can be used, default is MB)\n"
191 " -f, --event-fifo <path> Write events stream to fifo\n"
192 " -p, --hsm-root <path> Target HSM mount point\n"
193 " -q, --quiet Produce less verbose output\n"
194 " -u, --update-interval <s> Interval between progress reports sent\n"
196 " -v, --verbose Produce more verbose output\n",
197 cmd_name, cmd_name, cmd_name, cmd_name, cmd_name);
202 static int ct_parseopts(int argc, char * const *argv)
204 struct option long_opts[] = {
205 {"abort-on-error", no_argument, &opt.o_abort_on_error, 1},
206 {"abort_on_error", no_argument, &opt.o_abort_on_error, 1},
207 {"archive", required_argument, NULL, 'A'},
208 {"bandwidth", required_argument, NULL, 'b'},
209 {"chunk-size", required_argument, NULL, 'c'},
210 {"chunk_size", required_argument, NULL, 'c'},
211 {"daemon", no_argument, &opt.o_daemonize, 1},
212 {"event-fifo", required_argument, NULL, 'f'},
213 {"event_fifo", required_argument, NULL, 'f'},
214 {"dry-run", no_argument, &opt.o_dry_run, 1},
215 {"help", no_argument, NULL, 'h'},
216 {"hsm-root", required_argument, NULL, 'p'},
217 {"hsm_root", required_argument, NULL, 'p'},
218 {"import", no_argument, NULL, 'i'},
219 {"max-sequence", no_argument, NULL, 'M'},
220 {"max_sequence", no_argument, NULL, 'M'},
221 {"no-attr", no_argument, &opt.o_copy_attrs, 0},
222 {"no_attr", no_argument, &opt.o_copy_attrs, 0},
223 {"no-shadow", no_argument, &opt.o_shadow_tree, 0},
224 {"no_shadow", no_argument, &opt.o_shadow_tree, 0},
225 {"no-xattr", no_argument, &opt.o_copy_xattrs, 0},
226 {"no_xattr", no_argument, &opt.o_copy_xattrs, 0},
227 {"quiet", no_argument, NULL, 'q'},
228 {"rebind", no_argument, NULL, 'r'},
229 {"update-interval", required_argument, NULL, 'u'},
230 {"update_interval", required_argument, NULL, 'u'},
231 {"verbose", no_argument, NULL, 'v'},
235 unsigned long long value;
236 unsigned long long unit;
239 while ((c = getopt_long(argc, argv, "A:b:c:f:hiMp:qru:v",
240 long_opts, NULL)) != -1) {
243 if ((opt.o_archive_cnt >= LL_HSM_MAX_ARCHIVE) ||
244 (atoi(optarg) >= LL_HSM_MAX_ARCHIVE)) {
246 CT_ERROR(rc, "archive number must be less"
247 "than %zu", LL_HSM_MAX_ARCHIVE);
250 opt.o_archive_id[opt.o_archive_cnt] = atoi(optarg);
253 case 'b': /* -b and -c have both a number with unit as arg */
256 if (llapi_parse_size(optarg, &value, &unit, 0) < 0) {
258 CT_ERROR(rc, "bad value for -%c '%s'", c,
263 opt.o_chunk_size = value;
265 opt.o_bandwidth = value;
268 opt.o_event_fifo = optarg;
273 opt.o_action = CA_IMPORT;
276 opt.o_action = CA_MAXSEQ;
279 opt.o_hsm_root = optarg;
285 opt.o_action = CA_REBIND;
288 opt.o_report_int = atoi(optarg);
289 if (opt.o_report_int < 0) {
291 CT_ERROR(rc, "bad value for -%c '%s'", c,
306 switch (opt.o_action) {
308 /* src dst mount_point */
309 if (argc != optind + 3) {
311 CT_ERROR(rc, "--import requires 2 arguments");
314 opt.o_src = argv[optind++];
315 opt.o_dst = argv[optind++];
318 /* FID1 FID2 mount_point or FILE mount_point */
319 if (argc == optind + 2) {
320 opt.o_src = argv[optind++];
322 } else if (argc == optind + 3) {
323 opt.o_src = argv[optind++];
324 opt.o_dst = argv[optind++];
327 CT_ERROR(rc, "--rebind requires 1 or 2 arguments");
333 /* just mount point */
337 if (argc != optind + 1) {
339 CT_ERROR(rc, "no mount point specified");
343 opt.o_mnt = argv[optind];
346 CT_TRACE("action=%d src=%s dst=%s mount_point=%s",
347 opt.o_action, opt.o_src, opt.o_dst, opt.o_mnt);
349 if (!opt.o_dry_run && opt.o_hsm_root == NULL) {
351 CT_ERROR(rc, "must specify a root directory for the backend");
355 if (opt.o_action == CA_IMPORT) {
356 if (opt.o_src && opt.o_src[0] == '/') {
359 "source path must be relative to HSM root");
363 if (opt.o_dst && opt.o_dst[0] != '/') {
365 CT_ERROR(rc, "destination path must be absolute");
374 static int ct_mkdir_p(const char *path)
387 while ((ptr = strchr(ptr, '/')) != NULL) {
389 rc = mkdir(saved, DIR_PERM);
391 if (rc < 0 && errno != EEXIST) {
393 CT_ERROR(rc, "cannot mkdir '%s'", path);
405 static int ct_save_stripe(int src_fd, const char *src, const char *dst)
407 char lov_file[PATH_MAX];
408 char lov_buf[XATTR_SIZE_MAX];
409 struct lov_user_md *lum;
414 snprintf(lov_file, sizeof(lov_file), "%s.lov", dst);
415 CT_TRACE("saving stripe info of '%s' in %s", src, lov_file);
417 xattr_size = fgetxattr(src_fd, XATTR_LUSTRE_LOV, lov_buf,
419 if (xattr_size < 0) {
421 CT_ERROR(rc, "cannot get stripe info on '%s'", src);
425 lum = (struct lov_user_md *)lov_buf;
427 if (lum->lmm_magic == LOV_USER_MAGIC_V1 ||
428 lum->lmm_magic == LOV_USER_MAGIC_V3) {
429 /* Set stripe_offset to -1 so that it is not interpreted as a
430 * hint on restore. */
431 lum->lmm_stripe_offset = -1;
434 fd = open(lov_file, O_TRUNC | O_CREAT | O_WRONLY, FILE_PERM);
437 CT_ERROR(rc, "cannot open '%s'", lov_file);
441 rc = write(fd, lum, xattr_size);
444 CT_ERROR(rc, "cannot write %zd bytes to '%s'",
445 xattr_size, lov_file);
453 CT_ERROR(rc, "cannot close '%s'", lov_file);
465 static int ct_load_stripe(const char *src, void *lovea, size_t *lovea_size)
467 char lov_file[PATH_MAX];
471 snprintf(lov_file, sizeof(lov_file), "%s.lov", src);
472 CT_TRACE("reading stripe rules from '%s' for '%s'", lov_file, src);
474 fd = open(lov_file, O_RDONLY);
476 CT_ERROR(errno, "cannot open '%s'", lov_file);
480 rc = read(fd, lovea, *lovea_size);
482 CT_ERROR(errno, "cannot read %zu bytes from '%s'",
483 *lovea_size, lov_file);
494 static int ct_restore_stripe(const char *src, const char *dst, int dst_fd,
495 const void *lovea, size_t lovea_size)
499 rc = fsetxattr(dst_fd, XATTR_LUSTRE_LOV, lovea, lovea_size,
503 CT_ERROR(rc, "cannot set lov EA on '%s'", dst);
509 static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
510 const char *dst, int src_fd, int dst_fd,
511 const struct hsm_action_item *hai, long hal_flags)
513 struct hsm_extent he;
514 __u64 offset = hai->hai_extent.offset;
518 __u64 write_total = 0;
520 time_t last_report_time;
522 double start_ct_now = ct_now();
523 /* Bandwidth Control */
526 time_t last_bw_print;
528 if (fstat(src_fd, &src_st) < 0) {
530 CT_ERROR(rc, "cannot stat '%s'", src);
534 if (!S_ISREG(src_st.st_mode)) {
536 CT_ERROR(rc, "'%s' is not a regular file", src);
540 if (hai->hai_extent.offset > (__u64)src_st.st_size) {
542 CT_ERROR(rc, "Trying to start reading past end ("LPU64" > "
543 "%jd) of '%s' source file", hai->hai_extent.offset,
544 (intmax_t)src_st.st_size, src);
548 if (fstat(dst_fd, &dst_st) < 0) {
550 CT_ERROR(rc, "cannot stat '%s'", dst);
554 if (!S_ISREG(dst_st.st_mode)) {
556 CT_ERROR(rc, "'%s' is not a regular file", dst);
560 /* Don't read beyond a given extent */
561 length = min(hai->hai_extent.length,
562 src_st.st_size - hai->hai_extent.offset);
564 start_time = last_bw_print = last_report_time = time(NULL);
568 rc = llapi_hsm_action_progress(hcp, &he, length, 0);
570 /* Action has been canceled or something wrong
571 * is happening. Stop copying data. */
572 CT_ERROR(rc, "progress ioctl for copy '%s'->'%s' failed",
579 buf = malloc(opt.o_chunk_size);
585 CT_TRACE("start copy of "LPU64" bytes from '%s' to '%s'",
588 while (write_total < length) {
591 int chunk = (length - write_total > opt.o_chunk_size) ?
592 opt.o_chunk_size : length - write_total;
594 rsize = pread(src_fd, buf, chunk, offset);
601 CT_ERROR(rc, "cannot read from '%s'", src);
605 wsize = pwrite(dst_fd, buf, rsize, offset);
608 CT_ERROR(rc, "cannot write to '%s'", dst);
612 write_total += wsize;
616 /* sleep if needed, to honor bandwidth limits */
617 if (opt.o_bandwidth != 0) {
618 unsigned long long write_theory;
620 write_theory = (now - start_time) * opt.o_bandwidth;
622 if (write_theory < write_total) {
623 unsigned long long excess;
624 struct timespec delay;
626 excess = write_total - write_theory;
628 delay.tv_sec = excess / opt.o_bandwidth;
629 delay.tv_nsec = (excess % opt.o_bandwidth) *
630 NSEC_PER_SEC / opt.o_bandwidth;
632 if (now >= last_bw_print + opt.o_report_int) {
633 CT_TRACE("bandwith control: %lluB/s "
634 "excess=%llu sleep for "
636 opt.o_bandwidth, excess,
637 (long long)delay.tv_sec,
643 rc = nanosleep(&delay, &delay);
644 } while (rc < 0 && errno == EINTR);
646 CT_ERROR(errno, "delay for bandwidth "
647 "control failed to sleep: "
648 "residual=%lld.%09lds",
649 (long long)delay.tv_sec,
657 if (now >= last_report_time + opt.o_report_int) {
658 last_report_time = now;
659 CT_TRACE("%%"LPU64" ", 100 * write_total / length);
660 he.length = write_total;
661 rc = llapi_hsm_action_progress(hcp, &he, length, 0);
663 /* Action has been canceled or something wrong
664 * is happening. Stop copying data. */
665 CT_ERROR(rc, "progress ioctl for copy"
666 " '%s'->'%s' failed", src, dst);
675 * truncate restored file
676 * size is taken from the archive this is done to support
677 * restore after a force release which leaves the file with the
678 * wrong size (can big bigger than the new size)
680 if ((hai->hai_action == HSMA_RESTORE) &&
681 (src_st.st_size < dst_st.st_size)) {
683 * make sure the file is on disk before reporting success.
685 rc = ftruncate(dst_fd, src_st.st_size);
688 CT_ERROR(rc, "cannot truncate '%s' to size %jd",
689 dst, (intmax_t)src_st.st_size);
697 CT_TRACE("copied "LPU64" bytes in %f seconds",
698 length, ct_now() - start_ct_now);
703 /* Copy file attributes from file src to file dest */
704 static int ct_copy_attr(const char *src, const char *dst, int src_fd,
708 struct timeval times[2];
711 if (fstat(src_fd, &st) < 0) {
713 CT_ERROR(rc, "cannot stat '%s'", src);
717 times[0].tv_sec = st.st_atime;
718 times[0].tv_usec = 0;
719 times[1].tv_sec = st.st_mtime;
720 times[1].tv_usec = 0;
721 if (fchmod(dst_fd, st.st_mode) < 0 ||
722 fchown(dst_fd, st.st_uid, st.st_gid) < 0 ||
723 futimes(dst_fd, times) < 0) {
725 CT_ERROR(rc, "cannot set attributes of '%s'", src);
732 static int ct_copy_xattr(const char *src, const char *dst, int src_fd,
733 int dst_fd, bool is_restore)
735 char list[XATTR_LIST_MAX];
736 char value[XATTR_SIZE_MAX];
741 list_len = flistxattr(src_fd, list, sizeof(list));
746 while (name < list + list_len) {
747 rc = fgetxattr(src_fd, name, value, sizeof(value));
751 /* when we restore, we do not restore lustre xattr */
753 (strncmp(XATTR_TRUSTED_PREFIX, name,
754 sizeof(XATTR_TRUSTED_PREFIX) - 1) != 0)) {
755 rc = fsetxattr(dst_fd, name, value, rc, 0);
756 CT_TRACE("fsetxattr of '%s' on '%s' rc=%d (%s)",
757 name, dst, rc, strerror(errno));
758 /* lustre.* attrs aren't supported on other FS's */
759 if (rc < 0 && errno != EOPNOTSUPP) {
761 CT_ERROR(rc, "cannot set extended attribute"
767 name += strlen(name) + 1;
773 static int ct_path_lustre(char *buf, int sz, const char *mnt,
774 const lustre_fid *fid)
776 return snprintf(buf, sz, "%s/%s/fid/"DFID_NOBRACE, mnt,
777 dot_lustre_name, PFID(fid));
780 static int ct_path_archive(char *buf, int sz, const char *archive_dir,
781 const lustre_fid *fid)
783 return snprintf(buf, sz, "%s/%04x/%04x/%04x/%04x/%04x/%04x/"
784 DFID_NOBRACE, archive_dir,
785 (fid)->f_oid & 0xFFFF,
786 (fid)->f_oid >> 16 & 0xFFFF,
787 (unsigned int)((fid)->f_seq & 0xFFFF),
788 (unsigned int)((fid)->f_seq >> 16 & 0xFFFF),
789 (unsigned int)((fid)->f_seq >> 32 & 0xFFFF),
790 (unsigned int)((fid)->f_seq >> 48 & 0xFFFF),
794 static bool ct_is_retryable(int err)
796 return err == -ETIMEDOUT;
799 static int ct_begin_restore(struct hsm_copyaction_private **phcp,
800 const struct hsm_action_item *hai,
801 int mdt_index, int open_flags)
806 rc = llapi_hsm_action_begin(phcp, ctdata, hai, mdt_index, open_flags,
809 ct_path_lustre(src, sizeof(src), opt.o_mnt, &hai->hai_fid);
810 CT_ERROR(rc, "llapi_hsm_action_begin() on '%s' failed", src);
816 static int ct_begin(struct hsm_copyaction_private **phcp,
817 const struct hsm_action_item *hai)
819 /* Restore takes specific parameters. Call the same function w/ default
820 * values for all other operations. */
821 return ct_begin_restore(phcp, hai, -1, 0);
824 static int ct_fini(struct hsm_copyaction_private **phcp,
825 const struct hsm_action_item *hai, int hp_flags, int ct_rc)
827 struct hsm_copyaction_private *hcp;
831 CT_TRACE("Action completed, notifying coordinator "
832 "cookie="LPX64", FID="DFID", hp_flags=%d err=%d",
833 hai->hai_cookie, PFID(&hai->hai_fid),
836 ct_path_lustre(lstr, sizeof(lstr), opt.o_mnt, &hai->hai_fid);
838 if (phcp == NULL || *phcp == NULL) {
839 rc = llapi_hsm_action_begin(&hcp, ctdata, hai, -1, 0, true);
841 CT_ERROR(rc, "llapi_hsm_action_begin() on '%s' failed",
848 rc = llapi_hsm_action_end(phcp, &hai->hai_extent, hp_flags, abs(ct_rc));
849 if (rc == -ECANCELED)
850 CT_ERROR(rc, "completed action on '%s' has been canceled: "
851 "cookie="LPX64", FID="DFID, lstr, hai->hai_cookie,
852 PFID(&hai->hai_fid));
854 CT_ERROR(rc, "llapi_hsm_action_end() on '%s' failed", lstr);
856 CT_TRACE("llapi_hsm_action_end() on '%s' ok (rc=%d)",
862 static int ct_archive(const struct hsm_action_item *hai, const long hal_flags)
864 struct hsm_copyaction_private *hcp = NULL;
866 char dst[PATH_MAX] = "";
869 bool rename_needed = false;
875 rc = ct_begin(&hcp, hai);
879 /* we fill archive so:
881 * destination = lustre FID
883 ct_path_lustre(src, sizeof(src), opt.o_mnt, &hai->hai_dfid);
884 ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, &hai->hai_fid);
885 if (hai->hai_extent.length == -1) {
886 /* whole file, write it to tmp location and atomically
887 * replace old archived file */
888 strlcat(dst, "_tmp", sizeof(dst));
889 /* we cannot rely on the same test because ct_copy_data()
890 * updates hai_extent.length */
891 rename_needed = true;
894 CT_TRACE("archiving '%s' to '%s'", src, dst);
901 rc = ct_mkdir_p(dst);
903 CT_ERROR(rc, "mkdir_p '%s' failed", dst);
907 src_fd = llapi_hsm_action_get_fd(hcp);
910 CT_ERROR(rc, "cannot open '%s' for read", src);
914 open_flags = O_WRONLY | O_NOFOLLOW;
915 /* If extent is specified, don't truncate an old archived copy */
916 open_flags |= ((hai->hai_extent.length == -1) ? O_TRUNC : 0) | O_CREAT;
918 dst_fd = open(dst, open_flags, FILE_PERM);
921 CT_ERROR(rc, "cannot open '%s' for write", dst);
925 /* saving stripe is not critical */
926 rc = ct_save_stripe(src_fd, src, dst);
928 CT_ERROR(rc, "cannot save file striping info of '%s' in '%s'",
931 rc = ct_copy_data(hcp, src, dst, src_fd, dst_fd, hai, hal_flags);
933 CT_ERROR(rc, "data copy failed from '%s' to '%s'", src, dst);
940 CT_ERROR(rc, "cannot flush '%s' archive file '%s'", src, dst);
944 CT_TRACE("data archiving for '%s' to '%s' done", src, dst);
946 /* attrs will remain on the MDS; no need to copy them, except possibly
947 for disaster recovery */
948 if (opt.o_copy_attrs) {
949 rc = ct_copy_attr(src, dst, src_fd, dst_fd);
951 CT_ERROR(rc, "cannot copy attr of '%s' to '%s'",
955 CT_TRACE("attr file for '%s' saved to archive '%s'",
959 /* xattrs will remain on the MDS; no need to copy them, except possibly
960 for disaster recovery */
961 if (opt.o_copy_xattrs) {
962 rc = ct_copy_xattr(src, dst, src_fd, dst_fd, false);
964 CT_ERROR(rc, "cannot copy xattr of '%s' to '%s'",
966 rcf = rcf ? rcf : rc;
968 CT_TRACE("xattr file for '%s' saved to archive '%s'",
972 if (rename_needed == true) {
973 char tmp_src[PATH_MAX];
974 char tmp_dst[PATH_MAX];
976 /* atomically replace old archived file */
977 ct_path_archive(src, sizeof(src), opt.o_hsm_root,
979 rc = rename(dst, src);
982 CT_ERROR(rc, "cannot rename '%s' to '%s'", dst, src);
985 /* rename lov file */
986 snprintf(tmp_src, sizeof(tmp_src), "%s.lov", src);
987 snprintf(tmp_dst, sizeof(tmp_dst), "%s.lov", dst);
988 rc = rename(tmp_dst, tmp_src);
990 CT_ERROR(errno, "cannot rename '%s' to '%s'",
994 if (opt.o_shadow_tree) {
995 /* Create a namespace of softlinks that shadows the original
996 * Lustre namespace. This will only be current at
997 * time-of-archive (won't follow renames).
998 * WARNING: release won't kill these links; a manual
999 * cleanup of dead links would be required.
1002 long long recno = -1;
1008 sprintf(buf, DFID, PFID(&hai->hai_fid));
1009 sprintf(src, "%s/shadow/", opt.o_hsm_root);
1011 ptr = opt.o_hsm_root;
1013 (*ptr++ == '/') ? depth-- : 0;
1015 rc = llapi_fid2path(opt.o_mnt, buf, src + strlen(src),
1016 sizeof(src) - strlen(src), &recno, &linkno);
1018 CT_ERROR(rc, "cannot get FID of '%s'", buf);
1019 rcf = rcf ? rcf : rc;
1023 /* Figure out how many parent dirs to symlink back */
1026 (*ptr++ == '/') ? depth++ : 0;
1031 ct_path_archive(dst, sizeof(dst), buf, &hai->hai_fid);
1033 if (ct_mkdir_p(src)) {
1034 CT_ERROR(errno, "mkdir_p '%s' failed", src);
1035 rcf = rcf ? rcf : -errno;
1038 /* symlink already exists ? */
1039 sz = readlink(src, buf, sizeof(buf));
1040 /* detect truncation */
1041 if (sz == sizeof(buf)) {
1042 rcf = rcf ? rcf : -E2BIG;
1043 CT_ERROR(rcf, "readlink '%s' truncated", src);
1048 if (sz == 0 || strncmp(buf, dst, sz) != 0) {
1049 if (unlink(src) && errno != ENOENT) {
1051 "cannot unlink symlink '%s'",
1053 rcf = rcf ? rcf : -errno;
1055 /* unlink old symlink done */
1056 CT_TRACE("remove old symlink '%s' pointing"
1057 " to '%s'", src, buf);
1060 /* symlink already ok */
1061 CT_TRACE("symlink '%s' already pointing"
1062 " to '%s'", src, dst);
1067 if (symlink(dst, src)) {
1068 CT_ERROR(errno, "cannot symlink '%s' to '%s'",
1070 rcf = rcf ? rcf : -errno;
1073 CT_TRACE("symlink '%s' to '%s' done", src, dst);
1085 if (ct_is_retryable(rc))
1086 hp_flags |= HP_FLAG_RETRY;
1097 rc = ct_fini(&hcp, hai, hp_flags, rcf);
1102 static int ct_restore(const struct hsm_action_item *hai, const long hal_flags)
1104 struct hsm_copyaction_private *hcp = NULL;
1107 char lov_buf[XATTR_SIZE_MAX];
1108 size_t lov_size = sizeof(lov_buf);
1117 /* we fill lustre so:
1118 * source = lustre FID in the backend
1119 * destination = data FID = volatile file
1122 /* build backend file name from released file FID */
1123 ct_path_archive(src, sizeof(src), opt.o_hsm_root, &hai->hai_fid);
1125 rc = llapi_get_mdt_index_by_fid(opt.o_mnt_fd, &hai->hai_fid,
1128 CT_ERROR(rc, "cannot get mdt index "DFID"",
1129 PFID(&hai->hai_fid));
1132 /* restore loads and sets the LOVEA w/o interpreting it to avoid
1133 * dependency on the structure format. */
1134 rc = ct_load_stripe(src, lov_buf, &lov_size);
1136 CT_WARN("cannot get stripe rules for '%s' (%s), use default",
1137 src, strerror(-rc));
1140 open_flags |= O_LOV_DELAY_CREATE;
1144 rc = ct_begin_restore(&hcp, hai, mdt_index, open_flags);
1148 /* get the FID of the volatile file */
1149 rc = llapi_hsm_action_get_dfid(hcp, &dfid);
1151 CT_ERROR(rc, "restoring "DFID
1152 ", cannot get FID of created volatile file",
1153 PFID(&hai->hai_fid));
1157 /* build volatile "file name", for messages */
1158 snprintf(dst, sizeof(dst), "{VOLATILE}="DFID, PFID(&dfid));
1160 CT_TRACE("restoring data from '%s' to '%s'", src, dst);
1162 if (opt.o_dry_run) {
1167 src_fd = open(src, O_RDONLY | O_NOATIME | O_NOFOLLOW);
1170 CT_ERROR(rc, "cannot open '%s' for read", src);
1174 dst_fd = llapi_hsm_action_get_fd(hcp);
1177 CT_ERROR(rc, "cannot open '%s' for write", dst);
1182 /* the layout cannot be allocated through .fid so we have to
1183 * restore a layout */
1184 rc = ct_restore_stripe(src, dst, dst_fd, lov_buf, lov_size);
1186 CT_ERROR(rc, "cannot restore file striping info"
1187 " for '%s' from '%s'", dst, src);
1193 rc = ct_copy_data(hcp, src, dst, src_fd, dst_fd, hai, hal_flags);
1195 CT_ERROR(rc, "cannot copy data from '%s' to '%s'",
1198 if (ct_is_retryable(rc))
1199 hp_flags |= HP_FLAG_RETRY;
1203 CT_TRACE("data restore from '%s' to '%s' done", src, dst);
1206 rc = ct_fini(&hcp, hai, hp_flags, rc);
1208 /* object swaping is done by cdt at copy end, so close of volatile file
1209 * cannot be done before */
1219 static int ct_remove(const struct hsm_action_item *hai, const long hal_flags)
1221 struct hsm_copyaction_private *hcp = NULL;
1225 rc = ct_begin(&hcp, hai);
1229 ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, &hai->hai_fid);
1231 CT_TRACE("removing file '%s'", dst);
1233 if (opt.o_dry_run) {
1241 CT_ERROR(rc, "cannot unlink '%s'", dst);
1246 strlcat(dst, ".lov", sizeof(dst));
1250 CT_ERROR(rc, "cannot unlink '%s'", dst);
1256 rc = ct_fini(&hcp, hai, 0, rc);
1261 static int ct_process_item(struct hsm_action_item *hai, const long hal_flags)
1265 if (opt.o_verbose >= LLAPI_MSG_INFO || opt.o_dry_run) {
1266 /* Print the original path */
1268 char path[PATH_MAX];
1269 long long recno = -1;
1272 sprintf(fid, DFID, PFID(&hai->hai_fid));
1273 CT_TRACE("'%s' action %s reclen %d, cookie="LPX64,
1274 fid, hsm_copytool_action2name(hai->hai_action),
1275 hai->hai_len, hai->hai_cookie);
1276 rc = llapi_fid2path(opt.o_mnt, fid, path,
1277 sizeof(path), &recno, &linkno);
1279 CT_ERROR(rc, "cannot get path of FID %s", fid);
1281 CT_TRACE("processing file '%s'", path);
1284 switch (hai->hai_action) {
1285 /* set err_major, minor inside these functions */
1287 rc = ct_archive(hai, hal_flags);
1290 rc = ct_restore(hai, hal_flags);
1293 rc = ct_remove(hai, hal_flags);
1296 CT_TRACE("cancel not implemented for file system '%s'",
1298 /* Don't report progress to coordinator for this cookie:
1299 * the copy function will get ECANCELED when reporting
1306 CT_ERROR(rc, "unknown action %d, on '%s'", hai->hai_action,
1309 ct_fini(NULL, hai, 0, rc);
1317 struct hsm_action_item *hai;
1320 static void *ct_thread(void *data)
1322 struct ct_th_data *cttd = data;
1325 rc = ct_process_item(cttd->hai, cttd->hal_flags);
1329 pthread_exit((void *)(intptr_t)rc);
1332 static int ct_process_item_async(const struct hsm_action_item *hai,
1335 pthread_attr_t attr;
1337 struct ct_th_data *data;
1340 data = malloc(sizeof(*data));
1344 data->hai = malloc(hai->hai_len);
1345 if (data->hai == NULL) {
1350 memcpy(data->hai, hai, hai->hai_len);
1351 data->hal_flags = hal_flags;
1353 rc = pthread_attr_init(&attr);
1355 CT_ERROR(rc, "pthread_attr_init failed for '%s' service",
1362 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1364 rc = pthread_create(&thread, &attr, ct_thread, data);
1366 CT_ERROR(rc, "cannot create thread for '%s' service",
1369 pthread_attr_destroy(&attr);
1373 static int ct_import_one(const char *src, const char *dst)
1375 char newarc[PATH_MAX];
1380 CT_TRACE("importing '%s' from '%s'", dst, src);
1382 if (stat(src, &st) < 0) {
1384 CT_ERROR(rc, "cannot stat '%s'", src);
1391 rc = llapi_hsm_import(dst,
1392 opt.o_archive_cnt ? opt.o_archive_id[0] : 0,
1393 &st, 0, 0, 0, 0, NULL, &fid);
1395 CT_ERROR(rc, "cannot import '%s' from '%s'", dst, src);
1399 ct_path_archive(newarc, sizeof(newarc), opt.o_hsm_root, &fid);
1401 rc = ct_mkdir_p(newarc);
1403 CT_ERROR(rc, "mkdir_p '%s' failed", newarc);
1409 /* Lots of choices now: mv, ln, ln -s ? */
1410 rc = link(src, newarc); /* hardlink */
1413 CT_ERROR(rc, "cannot link '%s' to '%s'", newarc, src);
1417 CT_TRACE("imported '%s' from '%s'=='%s'", dst, newarc, src);
1422 static char *path_concat(const char *dirname, const char *basename)
1427 rc = asprintf(&result, "%s/%s", dirname, basename);
1434 static int ct_import_fid(const lustre_fid *import_fid)
1436 char fid_path[PATH_MAX];
1439 ct_path_lustre(fid_path, sizeof(fid_path), opt.o_mnt, import_fid);
1440 rc = access(fid_path, F_OK);
1441 if (rc == 0 || errno != ENOENT) {
1442 rc = (errno == 0) ? -EEXIST : -errno;
1443 CT_ERROR(rc, "cannot import '"DFID"'", PFID(import_fid));
1447 ct_path_archive(fid_path, sizeof(fid_path), opt.o_hsm_root,
1450 CT_TRACE("Resolving "DFID" to %s", PFID(import_fid), fid_path);
1452 return ct_import_one(fid_path, opt.o_dst);
1455 static int ct_import_recurse(const char *relpath)
1458 struct dirent ent, *cookie = NULL;
1459 char *srcpath, *newpath;
1460 lustre_fid import_fid;
1463 if (relpath == NULL)
1466 /* Is relpath a FID? In which case SFID should expand to three
1468 rc = sscanf(relpath, SFID, RFID(&import_fid));
1470 return ct_import_fid(&import_fid);
1472 srcpath = path_concat(opt.o_hsm_root, relpath);
1473 if (srcpath == NULL) {
1478 dir = opendir(srcpath);
1480 /* Not a dir, or error */
1481 if (errno == ENOTDIR) {
1482 /* Single regular file case, treat o_dst as absolute
1484 rc = ct_import_one(srcpath, opt.o_dst);
1487 CT_ERROR(rc, "cannot opendir '%s'", srcpath);
1496 rc = readdir_r(dir, &ent, &cookie);
1499 CT_ERROR(rc, "cannot readdir_r '%s'", relpath);
1502 } else if ((rc == 0) && (cookie == NULL)) {
1503 /* end of directory */
1507 if (!strcmp(ent.d_name, ".") ||
1508 !strcmp(ent.d_name, ".."))
1511 /* New relative path */
1512 newpath = path_concat(relpath, ent.d_name);
1513 if (newpath == NULL) {
1519 if (ent.d_type == DT_DIR) {
1520 rc = ct_import_recurse(newpath);
1525 sprintf(src, "%s/%s", opt.o_hsm_root, newpath);
1526 sprintf(dst, "%s/%s", opt.o_dst, newpath);
1527 /* Make the target dir in the Lustre fs */
1528 rc = ct_mkdir_p(dst);
1530 /* Import the file */
1531 rc = ct_import_one(src, dst);
1533 CT_ERROR(rc, "ct_mkdir_p '%s' failed", dst);
1539 CT_ERROR(rc, "cannot import '%s'", newpath);
1540 if (err_major && opt.o_abort_on_error) {
1554 static int ct_rebind_one(const lustre_fid *old_fid, const lustre_fid *new_fid)
1560 CT_TRACE("rebind "DFID" to "DFID, PFID(old_fid), PFID(new_fid));
1562 ct_path_archive(src, sizeof(src), opt.o_hsm_root, old_fid);
1563 ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, new_fid);
1565 if (!opt.o_dry_run) {
1567 if (rename(src, dst)) {
1569 CT_ERROR(rc, "cannot rename '%s' to '%s'", src, dst);
1572 /* rename lov file */
1573 strlcat(src, ".lov", sizeof(src));
1574 strlcat(dst, ".lov", sizeof(dst));
1575 if (rename(src, dst))
1576 CT_ERROR(errno, "cannot rename '%s' to '%s'", src, dst);
1582 static bool fid_is_file(lustre_fid *fid)
1584 return fid_is_norm(fid) || fid_is_igif(fid);
1587 static bool should_ignore_line(const char *line)
1591 for (i = 0; line[i] != '\0'; i++) {
1592 if (isspace(line[i]))
1594 else if (line[i] == '#')
1603 static int ct_rebind_list(const char *list)
1609 size_t line_size = 0;
1610 unsigned int nl = 0;
1611 unsigned int ok = 0;
1613 filp = fopen(list, "r");
1616 CT_ERROR(rc, "cannot open '%s'", list);
1620 /* each line consists of 2 FID */
1621 while ((r = getline(&line, &line_size, filp)) != -1) {
1625 /* Ignore empty and commented out ('#...') lines. */
1626 if (should_ignore_line(line))
1631 rc = sscanf(line, SFID" "SFID, RFID(&old_fid), RFID(&new_fid));
1632 if (rc != 6 || !fid_is_file(&old_fid) ||
1633 !fid_is_file(&new_fid)) {
1635 "'%s' FID expected near '%s', line %u",
1641 if (ct_rebind_one(&old_fid, &new_fid))
1652 /* return 0 if all rebinds were sucessful */
1653 CT_TRACE("%u lines read from '%s', %u rebind successful", nl, list, ok);
1655 return ok == nl ? 0 : -1;
1658 static int ct_rebind(void)
1666 if (sscanf(opt.o_src, SFID, RFID(&old_fid)) != 3 ||
1667 !fid_is_file(&old_fid)) {
1669 CT_ERROR(rc, "'%s' invalid FID format", opt.o_src);
1673 if (sscanf(opt.o_dst, SFID, RFID(&new_fid)) != 3 ||
1674 !fid_is_file(&new_fid)) {
1676 CT_ERROR(rc, "'%s' invalid FID format", opt.o_dst);
1680 rc = ct_rebind_one(&old_fid, &new_fid);
1685 /* o_src is a list file */
1686 rc = ct_rebind_list(opt.o_src);
1691 static int ct_dir_level_max(const char *dirpath, __u16 *sub_seqmax)
1696 struct dirent ent, *cookie = NULL;
1700 dir = opendir(dirpath);
1703 CT_ERROR(rc, "cannot open directory '%s'", opt.o_hsm_root);
1707 while ((rc = readdir_r(dir, &ent, &cookie)) == 0) {
1709 /* end of directory.
1710 * rc is 0 and seqmax contains the max value. */
1713 if (!strcmp(ent.d_name, ".") || !strcmp(ent.d_name, ".."))
1716 if (sscanf(ent.d_name, "%hx", &sub_seq) != 1) {
1717 CT_TRACE("'%s' has an unexpected dirname format, "
1718 "skip entry", ent.d_name);
1721 if (sub_seq > *sub_seqmax)
1722 *sub_seqmax = sub_seq;
1725 CT_ERROR(rc, "cannot readdir_r '%s'", dirpath);
1732 static int ct_max_sequence(void)
1735 char path[PATH_MAX];
1739 strlcpy(path, opt.o_hsm_root, sizeof(path));
1740 /* FID sequence is stored in top-level directory names:
1741 * hsm_root/16bits (high weight)/16 bits/16 bits/16 bits (low weight).
1743 for (i = 0; i < 4; i++) {
1746 rc = ct_dir_level_max(path, &subseq);
1749 seq |= ((__u64)subseq << ((3 - i) * 16));
1750 path_len = strlen(path);
1751 rc = snprintf(path + path_len, sizeof(path) - path_len,
1753 if (rc >= (sizeof(path) - path_len))
1755 path[sizeof(path) - 1] = '\0';
1758 printf("max_sequence: "LPX64"\n", seq);
1763 static void handler(int signal)
1765 psignal(signal, "exiting");
1766 /* If we don't clean up upon interrupt, umount thinks there's a ref
1767 * and doesn't remove us from mtab (EINPROGRESS). The lustre client
1768 * does successfully unmount and the mount is actually gone, but the
1769 * mtab entry remains. So this just makes mtab happier. */
1770 llapi_hsm_copytool_unregister(&ctdata);
1772 /* Also remove fifo upon signal as during normal/error exit */
1773 if (opt.o_event_fifo != NULL)
1774 llapi_hsm_unregister_event_fifo(opt.o_event_fifo);
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,