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.
40 #include <sys/xattr.h>
41 #include <sys/syscall.h>
42 #include <sys/types.h>
43 #include <lustre/lustre_idl.h>
44 #include <lustre/lustreapi.h>
46 /* Progress reporting period */
47 #define REPORT_INTERVAL_DEFAULT 30
48 /* HSM hash subdir permissions */
49 #define DIR_PERM S_IRWXU
50 /* HSM hash file permissions */
51 #define FILE_PERM (S_IRUSR | S_IWUSR)
53 #define ONE_MB 0x100000
55 /* copytool uses a 32b bitmask field to register with kuc
56 * archive num = 0 => all
57 * archive num from 1 to 32
59 #define MAX_ARCHIVE_CNT (sizeof(__u32) * 8)
76 int o_archive_id[MAX_ARCHIVE_CNT];
78 unsigned long long o_bandwidth;
80 enum ct_action o_action;
83 char *o_src; /* for import, or rebind */
84 char *o_dst; /* for import, or rebind */
87 /* everything else is zeroed */
88 struct options opt = {
91 .o_verbose = LLAPI_MSG_WARN,
93 .o_report_int = REPORT_INTERVAL_DEFAULT,
94 .o_chunk_size = ONE_MB,
97 /* hsm_copytool_private will hold an open FD on the lustre mount point
98 * for us. Additionally open one on the archive FS root to make sure
99 * it doesn't drop out from under us (and remind the admin to shutdown
100 * the copytool before unmounting). */
102 static int arc_fd = -1;
104 static int err_major;
105 static int err_minor;
107 static char cmd_name[PATH_MAX];
108 static char fs_name[MAX_OBD_NAME + 1];
110 static struct hsm_copytool_private *ctdata;
113 #define CT_ERROR(_rc, _format, ...) \
114 llapi_error(LLAPI_MSG_ERROR, _rc, \
115 "%s[%ld]: "_format, \
116 cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
117 #define CT_DEBUG(_format, ...) \
118 llapi_error(LLAPI_MSG_DEBUG | LLAPI_MSG_NO_ERRNO, 0, \
119 "%s[%ld]: "_format, \
120 cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
121 #define CT_WARN(_format, ...) \
122 llapi_error(LLAPI_MSG_WARN | LLAPI_MSG_NO_ERRNO, 0, \
123 "%s[%ld]: "_format, \
124 cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
125 #define CT_TRACE(_format, ...) \
126 llapi_error(LLAPI_MSG_INFO | LLAPI_MSG_NO_ERRNO, 0, \
127 "%s[%ld]: "_format, \
128 cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
129 #define CT_PRINTF(_format, ...) \
130 llapi_printf(LLAPI_MSG_NORMAL, \
131 "%s[%ld]: "_format, \
132 cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
134 static void usage(const char *name, int rc)
137 " Usage: %s [options]... <mode> <lustre_mount_point>\n"
138 "The Lustre HSM Posix copy tool can be used as a daemon or "
139 "as a command line tool\n"
140 "The Lustre HSM daemon acts on action requests from Lustre\n"
141 "to copy files to and from an HSM archive system.\n"
142 "This POSIX-flavored daemon makes regular POSIX filesystem calls\n"
143 "to an HSM mounted at a given hsm_root.\n"
144 " --daemon Daemon mode, run in background\n"
146 " --no-attr Don't copy file attributes\n"
147 " --no-shadow Don't create shadow namespace in archive\n"
148 " --no-xattr Don't copy file extended attributes\n"
149 "The Lustre HSM tool performs administrator-type actions\n"
150 "on a Lustre HSM archive.\n"
151 "This POSIX-flavored tool can link an existing HSM namespace\n"
152 "into a Lustre filesystem.\n"
154 " %s [options] --import <src> <dst> <lustre_mount_point>\n"
155 " import an archived subtree at\n"
156 " <src> (relative to hsm_root) into the Lustre filesystem at\n"
157 " <dst> (absolute)\n"
158 " %s [options] --rebind <old_FID> <new_FID> <lustre_mount_point>\n"
159 " rebind an entry in the HSM to a new FID\n"
160 " <old_FID> old FID the HSM entry is bound to\n"
161 " <new_FID> new FID to bind the HSM entry to\n"
162 " %s [options] --rebind <list_file> <lustre_mount_point>\n"
163 " perform the rebind operation for all FID in the list file\n"
164 " each line of <list_file> consists of <old_FID> <new_FID>\n"
165 " %s [options] --max-sequence <fsname>\n"
166 " return the max fid sequence of archived files\n"
167 " -A, --archive <#> Archive number (repeatable)\n"
168 " -p, --hsm-root <path> Target HSM mount point\n"
169 " -q, --quiet Produce less verbose output\n"
170 " -v, --verbose Produce more verbose output\n"
171 " -c, --chunk-size <sz> I/O size used during data copy\n"
172 " (unit can be used, default is MB)\n"
173 " --abort-on-error Abort operation on major error\n"
174 " --dry-run Don't run, just show what would be done\n"
175 " --bandwidth <bw> Limit I/O bandwidth (unit can be used\n,"
177 cmd_name, cmd_name, cmd_name, cmd_name, cmd_name);
182 static int ct_parseopts(int argc, char * const *argv)
184 struct option long_opts[] = {
185 {"abort-on-error", no_argument, &opt.o_abort_on_error, 1},
186 {"abort_on_error", no_argument, &opt.o_abort_on_error, 1},
187 {"archive", required_argument, NULL, 'A'},
188 {"bandwidth", required_argument, NULL, 'b'},
189 {"chunk-size", required_argument, NULL, 'c'},
190 {"chunk_size", required_argument, NULL, 'c'},
191 {"daemon", no_argument, &opt.o_daemonize, 1},
192 {"dry-run", no_argument, &opt.o_dry_run, 1},
193 {"help", no_argument, NULL, 'h'},
194 {"hsm-root", required_argument, NULL, 'p'},
195 {"hsm_root", required_argument, NULL, 'p'},
196 {"import", no_argument, NULL, 'i'},
197 {"max-sequence", no_argument, NULL, 'M'},
198 {"max_sequence", no_argument, NULL, 'M'},
199 {"no-attr", no_argument, &opt.o_copy_attrs, 0},
200 {"no_attr", no_argument, &opt.o_copy_attrs, 0},
201 {"no-shadow", no_argument, &opt.o_shadow_tree, 0},
202 {"no_shadow", no_argument, &opt.o_shadow_tree, 0},
203 {"no-xattr", no_argument, &opt.o_copy_xattrs, 0},
204 {"no_xattr", no_argument, &opt.o_copy_xattrs, 0},
205 {"quiet", no_argument, NULL, 'q'},
206 {"rebind", no_argument, NULL, 'r'},
207 {"report", required_argument, &opt.o_report_int, 0},
208 {"verbose", no_argument, NULL, 'v'},
212 unsigned long long value;
213 unsigned long long unit;
216 while ((c = getopt_long(argc, argv, "A:b:c:hiMp:qrv",
217 long_opts, NULL)) != -1) {
220 if ((opt.o_archive_cnt >= MAX_ARCHIVE_CNT) ||
221 (atoi(optarg) >= MAX_ARCHIVE_CNT)) {
223 CT_ERROR(rc, "archive number must be less"
224 "than %zu", MAX_ARCHIVE_CNT);
227 opt.o_archive_id[opt.o_archive_cnt] = atoi(optarg);
230 case 'b': /* -b and -c have both a number with unit as arg */
233 if (llapi_parse_size(optarg, &value, &unit, 0) < 0) {
235 CT_ERROR(rc, "bad value for -%c '%s'", c,
240 opt.o_chunk_size = value;
242 opt.o_bandwidth = value;
247 opt.o_action = CA_IMPORT;
250 opt.o_action = CA_MAXSEQ;
253 opt.o_hsm_root = optarg;
259 opt.o_action = CA_REBIND;
271 switch (opt.o_action) {
273 /* src dst mount_point */
274 if (argc != optind + 3) {
276 CT_ERROR(rc, "--import requires 2 arguments");
279 opt.o_src = argv[optind++];
280 opt.o_dst = argv[optind++];
283 /* FID1 FID2 mount_point or FILE mount_point */
284 if (argc == optind + 2) {
285 opt.o_src = argv[optind++];
287 } else if (argc == optind + 3) {
288 opt.o_src = argv[optind++];
289 opt.o_dst = argv[optind++];
292 CT_ERROR(rc, "--rebind requires 1 or 2 arguments");
298 /* just mount point */
302 if (argc != optind + 1) {
304 CT_ERROR(rc, "no mount point specified");
308 opt.o_mnt = argv[optind];
310 CT_TRACE("action=%d src=%s dst=%s mount_point=%s",
311 opt.o_action, opt.o_src, opt.o_dst, opt.o_mnt);
313 if (!opt.o_dry_run && opt.o_hsm_root == NULL) {
315 CT_ERROR(rc, "must specify a root directory for the backend");
319 if (opt.o_action == CA_IMPORT) {
320 if (opt.o_src && opt.o_src[0] == '/') {
323 "source path must be relative to HSM root");
327 if (opt.o_dst && opt.o_dst[0] != '/') {
329 CT_ERROR(rc, "destination path must be absolute");
338 static int ct_mkdir_p(const char *path)
348 while ((ptr = strchr(ptr, '/')) != NULL) {
350 rc = mkdir(saved, DIR_PERM);
352 if (rc < 0 && errno != EEXIST) {
354 CT_ERROR(rc, "cannot mkdir '%s'", path);
366 static int ct_save_stripe(int src_fd, const char *src, const char *dst)
368 char lov_file[PATH_MAX];
369 char lov_buf[XATTR_SIZE_MAX];
370 struct lov_user_md *lum;
375 snprintf(lov_file, sizeof(lov_file), "%s.lov", dst);
376 CT_TRACE("saving stripe info of '%s' in %s", src, lov_file);
378 xattr_size = fgetxattr(src_fd, XATTR_LUSTRE_LOV, lov_buf,
380 if (xattr_size < 0) {
382 CT_ERROR(rc, "cannot get stripe info on '%s'", src);
386 lum = (struct lov_user_md *)lov_buf;
388 if (lum->lmm_magic == LOV_USER_MAGIC_V1 ||
389 lum->lmm_magic == LOV_USER_MAGIC_V3) {
390 /* Set stripe_offset to -1 so that it is not interpreted as a
391 * hint on restore. */
392 lum->lmm_stripe_offset = -1;
395 fd = open(lov_file, O_TRUNC | O_CREAT | O_WRONLY, FILE_PERM);
398 CT_ERROR(rc, "cannot open '%s'", lov_file);
402 rc = write(fd, lum, xattr_size);
405 CT_ERROR(rc, "cannot write %zd bytes to '%s'",
406 xattr_size, lov_file);
414 CT_ERROR(rc, "cannot close '%s'", lov_file);
426 static int ct_load_stripe(const char *src, void *lovea, size_t *lovea_size)
428 char lov_file[PATH_MAX];
432 snprintf(lov_file, sizeof(lov_file), "%s.lov", src);
433 CT_TRACE("reading stripe rules from '%s' for '%s'", lov_file, src);
435 fd = open(lov_file, O_RDONLY);
437 CT_ERROR(errno, "cannot open '%s'", lov_file);
441 rc = read(fd, lovea, *lovea_size);
443 CT_ERROR(errno, "cannot read %zu bytes from '%s'",
444 *lovea_size, lov_file);
455 static int ct_restore_stripe(const char *src, const char *dst, int dst_fd,
456 const void *lovea, size_t lovea_size)
460 rc = fsetxattr(dst_fd, XATTR_LUSTRE_LOV, lovea, lovea_size,
463 CT_ERROR(errno, "cannot set lov EA on '%s'", dst);
470 /* non-blocking read or write */
471 static int nonblock_rw(bool wr, int fd, char *buf, int size)
476 rc = write(fd, buf, size);
478 rc = read(fd, buf, size);
480 if ((rc < 0) && (errno == -EAGAIN)) {
482 struct timeval timeout;
484 timeout.tv_sec = opt.o_report_int;
489 rc = select(FD_SETSIZE, NULL, &set, NULL, &timeout);
491 rc = select(FD_SETSIZE, &set, NULL, NULL, &timeout);
495 /* Timed out, we read nothing */
498 /* Should be available now */
500 rc = write(fd, buf, size);
502 rc = read(fd, buf, size);
511 static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
512 const char *dst, int src_fd, int dst_fd,
513 const struct hsm_action_item *hai, long hal_flags)
515 struct hsm_extent he;
522 time_t last_print_time = time(0);
528 CT_TRACE("going to copy data from '%s' to '%s'", src, dst);
530 buf = malloc(opt.o_chunk_size);
534 if (fstat(src_fd, &src_st) < 0) {
535 CT_ERROR(errno, "cannot stat '%s'", src);
539 if (!S_ISREG(src_st.st_mode)) {
541 CT_ERROR(rc, "'%s' is not a regular file", src);
545 rc = lseek(src_fd, hai->hai_extent.offset, SEEK_SET);
548 "cannot seek for read to "LPU64" (len %jd) in '%s'",
549 hai->hai_extent.offset, (intmax_t)src_st.st_size, src);
554 if (fstat(dst_fd, &dst_st) < 0) {
555 CT_ERROR(errno, "cannot stat '%s'", dst);
559 if (!S_ISREG(dst_st.st_mode)) {
561 CT_ERROR(rc, "'%s' is not a regular file", dst);
565 rc = lseek(dst_fd, hai->hai_extent.offset, SEEK_SET);
568 CT_ERROR(rc, "cannot seek for write to "LPU64" on '%s'",
569 hai->hai_extent.offset, src);
573 he.offset = hai->hai_extent.offset;
575 rc = llapi_hsm_action_progress(hcp, &he, 0);
577 /* Action has been canceled or something wrong
578 * is happening. Stop copying data. */
579 CT_ERROR(rc, "progress ioctl for copy '%s'->'%s' failed",
585 /* Don't read beyond a given extent */
586 rlen = (hai->hai_extent.length == -1LL) ?
587 src_st.st_size : hai->hai_extent.length;
589 while (wpos < rlen) {
590 int chunk = (rlen - wpos > opt.o_chunk_size) ?
591 opt.o_chunk_size : rlen - wpos;
593 /* Only read more if we wrote everything in the buffer */
595 rsize = nonblock_rw(0, src_fd, buf, chunk);
600 if (rsize == -EAGAIN) {
604 /* Haven't read anything yet, let's
605 * give it back to the coordinator
606 * for rescheduling */
614 CT_ERROR(rc, "cannot read from '%s'", src);
622 wsize = nonblock_rw(1, dst_fd, buf + bufoff, rpos - wpos);
623 if (wsize == -EAGAIN)
629 CT_ERROR(rc, "cannot write to '%s'", dst);
636 if (opt.o_bandwidth != 0) {
637 static unsigned long long tot_bytes;
638 static time_t start_time, last_time;
639 time_t now = time(0);
640 double tot_time, excess;
641 unsigned int sleep_time;
643 if (now > last_time + 5) {
645 start_time = last_time = now;
649 tot_time = now - start_time;
653 excess = tot_bytes - tot_time * opt.o_bandwidth;
654 sleep_time = excess * 1000000 / opt.o_bandwidth;
655 if ((now - start_time) % 10 == 1)
656 CT_TRACE("bandwith control: excess=%E"
666 if (time(0) >= last_print_time + opt.o_report_int) {
667 last_print_time = time(0);
668 CT_TRACE("%%"LPU64" ", 100 * wpos / rlen);
670 rc = llapi_hsm_action_progress(hcp, &he, 0);
672 /* Action has been canceled or something wrong
673 * is happening. Stop copying data. */
674 CT_ERROR(rc, "progress ioctl for copy"
675 " '%s'->'%s' failed", src, dst);
684 * truncate restored file
685 * size is taken from the archive this is done to support
686 * restore after a force release which leaves the file with the
687 * wrong size (can big bigger than the new size)
689 if ((hai->hai_action == HSMA_RESTORE) &&
690 (src_st.st_size < dst_st.st_size)) {
692 * make sure the file is on disk before reporting success.
694 rc = ftruncate(dst_fd, src_st.st_size);
698 "cannot truncate '%s' to size %jd",
699 dst, (intmax_t)src_st.st_size);
709 /* Copy file attributes from file src to file dest */
710 static int ct_copy_attr(const char *src, const char *dst, int src_fd,
714 struct timeval times[2];
717 if (fstat(src_fd, &st) < 0) {
719 CT_ERROR(rc, "cannot stat '%s'", src);
723 times[0].tv_sec = st.st_atime;
724 times[0].tv_usec = 0;
725 times[1].tv_sec = st.st_mtime;
726 times[1].tv_usec = 0;
727 if (fchmod(dst_fd, st.st_mode) < 0 ||
728 fchown(dst_fd, st.st_uid, st.st_gid) < 0 ||
729 futimes(dst_fd, times) < 0) {
731 CT_ERROR(rc, "cannot set attributes of '%s'", src);
738 static int ct_copy_xattr(const char *src, const char *dst, int src_fd,
739 int dst_fd, bool is_restore)
741 char list[XATTR_LIST_MAX];
742 char value[XATTR_SIZE_MAX];
747 list_len = flistxattr(src_fd, list, sizeof(list));
752 while (name < list + list_len) {
753 rc = fgetxattr(src_fd, name, value, sizeof(value));
757 /* when we restore, we do not restore lustre xattr */
759 (strncmp(XATTR_TRUSTED_PREFIX, name,
760 sizeof(XATTR_TRUSTED_PREFIX) - 1) != 0)) {
761 rc = fsetxattr(dst_fd, name, value, rc, 0);
762 CT_TRACE("fsetxattr of '%s' on '%s' rc=%d (%s)",
763 name, dst, rc, strerror(errno));
764 /* lustre.* attrs aren't supported on other FS's */
765 if (rc < 0 && errno != EOPNOTSUPP) {
767 CT_ERROR(rc, "cannot set extended attribute"
773 name += strlen(name) + 1;
779 static int ct_path_lustre(char *buf, int sz, const char *mnt,
780 const lustre_fid *fid)
782 return snprintf(buf, sz, "%s/%s/fid/"DFID_NOBRACE, mnt,
783 dot_lustre_name, PFID(fid));
786 static int ct_path_archive(char *buf, int sz, const char *archive_dir,
787 const lustre_fid *fid)
789 return snprintf(buf, sz, "%s/%04x/%04x/%04x/%04x/%04x/%04x/"
790 DFID_NOBRACE, archive_dir,
791 (fid)->f_oid & 0xFFFF,
792 (fid)->f_oid >> 16 & 0xFFFF,
793 (unsigned int)((fid)->f_seq & 0xFFFF),
794 (unsigned int)((fid)->f_seq >> 16 & 0xFFFF),
795 (unsigned int)((fid)->f_seq >> 32 & 0xFFFF),
796 (unsigned int)((fid)->f_seq >> 48 & 0xFFFF),
800 static bool ct_is_retryable(int err)
802 return err == -ETIMEDOUT;
805 static int ct_begin_restore(struct hsm_copyaction_private **phcp,
806 const struct hsm_action_item *hai,
807 int mdt_index, int open_flags)
812 rc = llapi_hsm_action_begin(phcp, ctdata, hai, mdt_index, open_flags,
815 ct_path_lustre(src, sizeof(src), opt.o_mnt, &hai->hai_fid);
816 CT_ERROR(rc, "llapi_hsm_action_begin() on '%s' failed", src);
822 static int ct_begin(struct hsm_copyaction_private **phcp,
823 const struct hsm_action_item *hai)
825 /* Restore takes specific parameters. Call the same function w/ default
826 * values for all other operations. */
827 return ct_begin_restore(phcp, hai, -1, 0);
830 static int ct_fini(struct hsm_copyaction_private **phcp,
831 const struct hsm_action_item *hai, int hp_flags, int ct_rc)
836 CT_TRACE("Action completed, notifying coordinator "
837 "cookie="LPX64", FID="DFID", hp_flags=%d err=%d",
838 hai->hai_cookie, PFID(&hai->hai_fid),
841 ct_path_lustre(lstr, sizeof(lstr), opt.o_mnt, &hai->hai_fid);
842 rc = llapi_hsm_action_end(phcp, &hai->hai_extent, hp_flags, abs(ct_rc));
843 if (rc == -ECANCELED)
844 CT_ERROR(rc, "completed action on '%s' has been canceled: "
845 "cookie="LPX64", FID="DFID, lstr, hai->hai_cookie,
846 PFID(&hai->hai_fid));
848 CT_ERROR(rc, "llapi_hsm_action_end() on '%s' failed", lstr);
850 CT_TRACE("llapi_hsm_action_end() on '%s' ok (rc=%d)",
856 static int ct_archive(const struct hsm_action_item *hai, const long hal_flags)
858 struct hsm_copyaction_private *hcp = NULL;
863 bool rename_needed = false;
869 rc = ct_begin(&hcp, hai);
873 /* we fill archive so:
875 * destination = lustre FID
877 ct_path_lustre(src, sizeof(src), opt.o_mnt, &hai->hai_dfid);
878 ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, &hai->hai_fid);
879 if (hai->hai_extent.length == -1) {
880 /* whole file, write it to tmp location and atomically
881 * replace old archived file */
882 strncat(dst, "_tmp", sizeof(dst) - strlen(dst) - 1);
883 /* we cannot rely on the same test because ct_copy_data()
884 * updates hai_extent.length */
885 rename_needed = true;
888 CT_TRACE("archiving '%s' to '%s'", src, dst);
895 rc = ct_mkdir_p(dst);
897 CT_ERROR(rc, "mkdir_p '%s' failed", dst);
901 src_fd = llapi_hsm_action_get_fd(hcp);
904 CT_ERROR(rc, "cannot open '%s' for read", src);
908 open_flags = O_WRONLY | O_NOFOLLOW | O_NONBLOCK;
909 /* If extent is specified, don't truncate an old archived copy */
910 open_flags |= ((hai->hai_extent.length == -1) ? O_TRUNC : 0) | O_CREAT;
912 dst_fd = open(dst, open_flags, FILE_PERM);
915 CT_ERROR(rc, "cannot open '%s' for write", dst);
919 /* saving stripe is not critical */
920 rc = ct_save_stripe(src_fd, src, dst);
922 CT_ERROR(rc, "cannot save file striping info of '%s' in '%s'",
925 rc = ct_copy_data(hcp, src, dst, src_fd, dst_fd, hai, hal_flags);
927 CT_ERROR(rc, "data copy failed from '%s' to '%s'", src, dst);
934 CT_ERROR(rc, "cannot flush '%s' archive file '%s'", src, dst);
938 CT_TRACE("data archiving for '%s' to '%s' done", src, dst);
940 /* attrs will remain on the MDS; no need to copy them, except possibly
941 for disaster recovery */
942 if (opt.o_copy_attrs) {
943 rc = ct_copy_attr(src, dst, src_fd, dst_fd);
945 CT_ERROR(rc, "cannot copy attr of '%s' to '%s'",
949 CT_TRACE("attr file for '%s' saved to archive '%s'",
953 /* xattrs will remain on the MDS; no need to copy them, except possibly
954 for disaster recovery */
955 if (opt.o_copy_xattrs) {
956 rc = ct_copy_xattr(src, dst, src_fd, dst_fd, false);
958 CT_ERROR(rc, "cannot copy xattr of '%s' to '%s'",
960 rcf = rcf ? rcf : rc;
962 CT_TRACE("xattr file for '%s' saved to archive '%s'",
966 if (rename_needed == true) {
967 char tmp_src[PATH_MAX];
968 char tmp_dst[PATH_MAX];
970 /* atomically replace old archived file */
971 ct_path_archive(src, sizeof(src), opt.o_hsm_root,
973 rc = rename(dst, src);
976 CT_ERROR(rc, "cannot rename '%s' to '%s'", dst, src);
979 /* rename lov file */
980 snprintf(tmp_src, sizeof(tmp_src), "%s.lov", src);
981 snprintf(tmp_dst, sizeof(tmp_dst), "%s.lov", dst);
982 rc = rename(tmp_dst, tmp_src);
984 CT_ERROR(errno, "cannot rename '%s' to '%s'",
988 if (opt.o_shadow_tree) {
989 /* Create a namespace of softlinks that shadows the original
990 * Lustre namespace. This will only be current at
991 * time-of-archive (won't follow renames).
992 * WARNING: release won't kill these links; a manual
993 * cleanup of dead links would be required.
996 long long recno = -1;
1002 sprintf(buf, DFID, PFID(&hai->hai_fid));
1003 sprintf(src, "%s/shadow/", opt.o_hsm_root);
1005 ptr = opt.o_hsm_root;
1007 (*ptr++ == '/') ? depth-- : 0;
1009 rc = llapi_fid2path(opt.o_mnt, buf, src + strlen(src),
1010 sizeof(src) - strlen(src), &recno, &linkno);
1012 CT_ERROR(rc, "cannot get FID of '%s'", buf);
1013 rcf = rcf ? rcf : rc;
1017 /* Figure out how many parent dirs to symlink back */
1020 (*ptr++ == '/') ? depth++ : 0;
1025 ct_path_archive(dst, sizeof(dst), buf, &hai->hai_fid);
1027 if (ct_mkdir_p(src)) {
1028 CT_ERROR(errno, "mkdir_p '%s' failed", src);
1029 rcf = rcf ? rcf : -errno;
1032 /* symlink already exists ? */
1033 sz = readlink(src, buf, sizeof(buf));
1036 if (sz == 0 || strncmp(buf, dst, sz) != 0) {
1037 if (unlink(src) && errno != ENOENT) {
1039 "cannot unlink symlink '%s'",
1041 rcf = rcf ? rcf : -errno;
1043 /* unlink old symlink done */
1044 CT_TRACE("remove old symlink '%s' pointing"
1045 " to '%s'", src, buf);
1048 /* symlink already ok */
1049 CT_TRACE("symlink '%s' already pointing"
1050 " to '%s'", src, dst);
1055 if (symlink(dst, src)) {
1056 CT_ERROR(errno, "cannot symlink '%s' to '%s'",
1058 rcf = rcf ? rcf : -errno;
1061 CT_TRACE("symlink '%s' to '%s' done", src, dst);
1073 if (ct_is_retryable(rc))
1074 hp_flags |= HP_FLAG_RETRY;
1086 rc = ct_fini(&hcp, hai, hp_flags, rcf);
1091 static int ct_restore(const struct hsm_action_item *hai, const long hal_flags)
1093 struct hsm_copyaction_private *hcp = NULL;
1096 char lov_buf[XATTR_SIZE_MAX];
1097 size_t lov_size = sizeof(lov_buf);
1102 int mdt_index = -1; /* Not implemented */
1106 /* we fill lustre so:
1107 * source = lustre FID in the backend
1108 * destination = data FID = volatile file
1111 /* build backend file name from released file FID */
1112 ct_path_archive(src, sizeof(src), opt.o_hsm_root, &hai->hai_fid);
1114 /* restore loads and sets the LOVEA w/o interpreting it to avoid
1115 * dependency on the structure format. */
1116 rc = ct_load_stripe(src, lov_buf, &lov_size);
1118 CT_WARN("cannot get stripe rules for '%s' (%s), use default",
1119 src, strerror(-rc));
1122 open_flags |= O_LOV_DELAY_CREATE;
1126 rc = ct_begin_restore(&hcp, hai, mdt_index, open_flags);
1130 /* get the FID of the volatile file */
1131 rc = llapi_hsm_action_get_dfid(hcp, &dfid);
1133 CT_ERROR(rc, "restoring "DFID
1134 ", cannot get FID of created volatile file",
1135 PFID(&hai->hai_fid));
1139 /* build volatile "file name", for messages */
1140 snprintf(dst, sizeof(dst), "{VOLATILE}="DFID, PFID(&dfid));
1142 CT_TRACE("restoring data from '%s' to '%s'", src, dst);
1144 if (opt.o_dry_run) {
1149 src_fd = open(src, O_RDONLY | O_NOATIME | O_NONBLOCK | O_NOFOLLOW);
1152 CT_ERROR(rc, "cannot open '%s' for read", src);
1156 dst_fd = llapi_hsm_action_get_fd(hcp);
1159 /* the layout cannot be allocated through .fid so we have to
1160 * restore a layout */
1161 rc = ct_restore_stripe(src, dst, dst_fd, lov_buf, lov_size);
1163 CT_ERROR(rc, "cannot restore file striping info"
1164 " for '%s' from '%s'", dst, src);
1170 rc = ct_copy_data(hcp, src, dst, src_fd, dst_fd, hai, hal_flags);
1172 CT_ERROR(rc, "cannot copy data from '%s' to '%s'",
1175 if (ct_is_retryable(rc))
1176 hp_flags |= HP_FLAG_RETRY;
1180 CT_TRACE("data restore from '%s' to '%s' done", src, dst);
1184 rc = ct_fini(&hcp, hai, hp_flags, rc);
1186 /* object swaping is done by cdt at copy end, so close of volatile file
1187 * cannot be done before */
1197 static int ct_remove(const struct hsm_action_item *hai, const long hal_flags)
1199 struct hsm_copyaction_private *hcp = NULL;
1203 rc = ct_begin(&hcp, hai);
1207 ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, &hai->hai_fid);
1209 CT_TRACE("removing file '%s'", dst);
1211 if (opt.o_dry_run) {
1219 CT_ERROR(rc, "cannot unlink '%s'", dst);
1224 strncat(dst, ".lov", sizeof(dst) - strlen(dst) - 1);
1228 CT_ERROR(rc, "cannot unlink '%s'", dst);
1235 rc = ct_fini(&hcp, hai, 0, rc);
1240 static int ct_report_error(const struct hsm_action_item *hai, int flags,
1243 struct hsm_copyaction_private *hcp;
1246 rc = llapi_hsm_action_begin(&hcp, ctdata, hai, -1, 0, true);
1250 rc = llapi_hsm_action_end(&hcp, &hai->hai_extent, flags, abs(errval));
1255 static int ct_process_item(struct hsm_action_item *hai, const long hal_flags)
1259 if (opt.o_verbose >= LLAPI_MSG_INFO || opt.o_dry_run) {
1260 /* Print the original path */
1262 char path[PATH_MAX];
1263 long long recno = -1;
1266 sprintf(fid, DFID, PFID(&hai->hai_fid));
1267 CT_TRACE("'%s' action %s reclen %d, cookie="LPX64,
1268 fid, hsm_copytool_action2name(hai->hai_action),
1269 hai->hai_len, hai->hai_cookie);
1270 rc = llapi_fid2path(opt.o_mnt, fid, path,
1271 sizeof(path), &recno, &linkno);
1273 CT_ERROR(rc, "cannot get path of FID %s", fid);
1275 CT_TRACE("processing file '%s'", path);
1278 switch (hai->hai_action) {
1279 /* set err_major, minor inside these functions */
1281 rc = ct_archive(hai, hal_flags);
1284 rc = ct_restore(hai, hal_flags);
1287 rc = ct_remove(hai, hal_flags);
1290 CT_TRACE("cancel not implemented for file system '%s'",
1292 /* Don't report progress to coordinator for this cookie:
1293 * the copy function will get ECANCELED when reporting
1300 CT_ERROR(rc, "unknown action %d, on '%s'", hai->hai_action,
1303 ct_report_error(hai, 0, rc);
1311 struct hsm_action_item *hai;
1314 static void *ct_thread(void *data)
1316 struct ct_th_data *cttd = data;
1319 rc = ct_process_item(cttd->hai, cttd->hal_flags);
1323 pthread_exit((void *)(intptr_t)rc);
1326 static int ct_process_item_async(const struct hsm_action_item *hai,
1329 pthread_attr_t attr;
1331 struct ct_th_data *data;
1334 data = malloc(sizeof(*data));
1338 data->hai = malloc(hai->hai_len);
1339 if (data->hai == NULL) {
1344 memcpy(data->hai, hai, hai->hai_len);
1345 data->hal_flags = hal_flags;
1347 rc = pthread_attr_init(&attr);
1349 CT_ERROR(rc, "pthread_attr_init failed for '%s' service",
1356 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1358 rc = pthread_create(&thread, &attr, ct_thread, data);
1360 CT_ERROR(rc, "cannot create thread for '%s' service",
1363 pthread_attr_destroy(&attr);
1367 static int ct_import_one(const char *src, const char *dst)
1369 char newarc[PATH_MAX];
1374 CT_TRACE("importing '%s' from '%s'", dst, src);
1376 if (stat(src, &st) < 0) {
1378 CT_ERROR(rc, "cannot stat '%s'", src);
1385 rc = llapi_hsm_import(dst,
1386 opt.o_archive_cnt ? opt.o_archive_id[0] : 0,
1387 &st, 0, 0, 0, 0, NULL, &fid);
1389 CT_ERROR(rc, "cannot import '%s' from '%s'", dst, src);
1393 ct_path_archive(newarc, sizeof(newarc), opt.o_hsm_root, &fid);
1395 rc = ct_mkdir_p(newarc);
1397 CT_ERROR(rc, "mkdir_p '%s' failed", newarc);
1403 /* Lots of choices now: mv, ln, ln -s ? */
1404 rc = link(src, newarc); /* hardlink */
1407 CT_ERROR(rc, "cannot link '%s' to '%s'", newarc, src);
1411 CT_TRACE("imported '%s' from '%s'=='%s'", dst, newarc, src);
1416 static char *path_concat(const char *dirname, const char *basename)
1419 int dirlen = strlen(dirname);
1421 result = malloc(dirlen + strlen(basename) + 2);
1425 memcpy(result, dirname, dirlen);
1426 result[dirlen] = '/';
1427 strcpy(result + dirlen + 1, basename);
1432 static int ct_import_recurse(const char *relpath)
1435 struct dirent ent, *cookie = NULL;
1436 char *srcpath, *newpath;
1439 if (relpath == NULL)
1442 srcpath = path_concat(opt.o_hsm_root, relpath);
1443 if (srcpath == NULL) {
1448 dir = opendir(srcpath);
1450 /* Not a dir, or error */
1451 if (errno == ENOTDIR) {
1452 /* Single regular file case, treat o_dst as absolute
1454 rc = ct_import_one(srcpath, opt.o_dst);
1457 CT_ERROR(rc, "cannot opendir '%s'", srcpath);
1466 rc = readdir_r(dir, &ent, &cookie);
1469 CT_ERROR(rc, "cannot readdir_r '%s'", relpath);
1472 } else if ((rc == 0) && (cookie == NULL)) {
1473 /* end of directory */
1477 if (!strcmp(ent.d_name, ".") ||
1478 !strcmp(ent.d_name, ".."))
1481 /* New relative path */
1482 newpath = path_concat(relpath, ent.d_name);
1483 if (newpath == NULL) {
1489 if (ent.d_type == DT_DIR) {
1490 rc = ct_import_recurse(newpath);
1495 sprintf(src, "%s/%s", opt.o_hsm_root, newpath);
1496 sprintf(dst, "%s/%s", opt.o_dst, newpath);
1497 /* Make the target dir in the Lustre fs */
1498 rc = ct_mkdir_p(dst);
1500 /* Import the file */
1501 rc = ct_import_one(src, dst);
1503 CT_ERROR(rc, "ct_mkdir_p '%s' failed", dst);
1509 CT_ERROR(rc, "cannot import '%s'", newpath);
1510 if (err_major && opt.o_abort_on_error) {
1524 static int ct_rebind_one(const lustre_fid *old_fid, const lustre_fid *new_fid)
1530 CT_TRACE("rebind "DFID" to "DFID, PFID(old_fid), PFID(new_fid));
1532 ct_path_archive(src, sizeof(src), opt.o_hsm_root, old_fid);
1533 ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, new_fid);
1535 if (!opt.o_dry_run) {
1537 if (rename(src, dst)) {
1539 CT_ERROR(rc, "cannot rename '%s' to '%s'", src, dst);
1542 /* rename lov file */
1543 strncat(src, ".lov", sizeof(src) - strlen(src) - 1);
1544 strncat(dst, ".lov", sizeof(dst) - strlen(dst) - 1);
1545 if (rename(src, dst))
1546 CT_ERROR(errno, "cannot '%s' rename to '%s'", src, dst);
1552 static bool fid_is_file(lustre_fid *fid)
1554 return fid_is_norm(fid) || fid_is_igif(fid);
1557 static bool should_ignore_line(const char *line)
1561 for (i = 0; line[i] != '\0'; i++) {
1562 if (isspace(line[i]))
1564 else if (line[i] == '#')
1573 static int ct_rebind_list(const char *list)
1579 size_t line_size = 0;
1580 unsigned int nl = 0;
1581 unsigned int ok = 0;
1583 filp = fopen(list, "r");
1586 CT_ERROR(rc, "cannot open '%s'", list);
1590 /* each line consists of 2 FID */
1591 while ((r = getline(&line, &line_size, filp)) != -1) {
1595 /* Ignore empty and commented out ('#...') lines. */
1596 if (should_ignore_line(line))
1601 rc = sscanf(line, SFID" "SFID, RFID(&old_fid), RFID(&new_fid));
1602 if (rc != 6 || !fid_is_file(&old_fid) ||
1603 !fid_is_file(&new_fid)) {
1605 "'%s' FID expected near '%s', line %u",
1611 if (ct_rebind_one(&old_fid, &new_fid))
1622 /* return 0 if all rebinds were sucessful */
1623 CT_TRACE("%u lines read from '%s', %u rebind successful", nl, list, ok);
1625 return ok == nl ? 0 : -1;
1628 static int ct_rebind(void)
1636 if (sscanf(opt.o_src, SFID, RFID(&old_fid)) != 3 ||
1637 !fid_is_file(&old_fid)) {
1639 CT_ERROR(rc, "'%s' invalid FID format", opt.o_src);
1643 if (sscanf(opt.o_dst, SFID, RFID(&new_fid)) != 3 ||
1644 !fid_is_file(&new_fid)) {
1646 CT_ERROR(rc, "'%s' invalid FID format", opt.o_dst);
1650 rc = ct_rebind_one(&old_fid, &new_fid);
1655 /* o_src is a list file */
1656 rc = ct_rebind_list(opt.o_src);
1661 static int ct_dir_level_max(const char *dirpath, __u16 *sub_seqmax)
1666 struct dirent ent, *cookie = NULL;
1670 dir = opendir(dirpath);
1673 CT_ERROR(rc, "cannot open directory '%s'", opt.o_hsm_root);
1677 while ((rc = readdir_r(dir, &ent, &cookie)) == 0) {
1679 /* end of directory.
1680 * rc is 0 and seqmax contains the max value. */
1683 if (!strcmp(ent.d_name, ".") || !strcmp(ent.d_name, ".."))
1686 if (sscanf(ent.d_name, "%hx", &sub_seq) != 1) {
1687 CT_TRACE("'%s' has an unexpected dirname format, "
1688 "skip entry", ent.d_name);
1691 if (sub_seq > *sub_seqmax)
1692 *sub_seqmax = sub_seq;
1695 CT_ERROR(rc, "cannot readdir_r '%s'", dirpath);
1702 static int ct_max_sequence(void)
1705 char path[PATH_MAX];
1709 strncpy(path, opt.o_hsm_root, sizeof(path));
1710 /* FID sequence is stored in top-level directory names:
1711 * hsm_root/16bits (high weight)/16 bits/16 bits/16 bits (low weight).
1713 for (i = 0; i < 4; i++) {
1714 rc = ct_dir_level_max(path, &subseq);
1717 seq |= ((__u64)subseq << ((3 - i) * 16));
1718 sprintf(path + strlen(path), "/%04x", subseq);
1721 printf("max_sequence: %016Lx\n", seq);
1726 static void handler(int signal)
1728 psignal(signal, "exiting");
1729 /* If we don't clean up upon interrupt, umount thinks there's a ref
1730 * and doesn't remove us from mtab (EINPROGRESS). The lustre client
1731 * does successfully unmount and the mount is actually gone, but the
1732 * mtab entry remains. So this just makes mtab happier. */
1733 llapi_hsm_copytool_unregister(&ctdata);
1737 /* Daemon waits for messages from the kernel; run it in the background. */
1738 static int ct_run(void)
1742 if (opt.o_daemonize) {
1746 CT_ERROR(rc, "cannot daemonize");
1751 rc = llapi_hsm_copytool_register(&ctdata, opt.o_mnt, 0,
1752 opt.o_archive_cnt, opt.o_archive_id);
1754 CT_ERROR(rc, "cannot start copytool interface");
1758 signal(SIGINT, handler);
1759 signal(SIGTERM, handler);
1762 struct hsm_action_list *hal;
1763 struct hsm_action_item *hai;
1767 CT_TRACE("waiting for message from kernel");
1769 rc = llapi_hsm_copytool_recv(ctdata, &hal, &msgsize);
1770 if (rc == -ESHUTDOWN) {
1771 CT_TRACE("shutting down");
1773 } else if (rc == -EAGAIN) {
1774 continue; /* msg not for us */
1775 } else if (rc < 0) {
1776 CT_WARN("cannot receive action list: %s",
1779 if (opt.o_abort_on_error)
1785 CT_TRACE("copytool fs=%s archive#=%d item_count=%d",
1786 hal->hal_fsname, hal->hal_archive_id, hal->hal_count);
1788 if (strcmp(hal->hal_fsname, fs_name) != 0) {
1789 CT_ERROR(EINVAL, "'%s' invalid fs name, expecting: %s",
1790 hal->hal_fsname, fs_name);
1792 if (opt.o_abort_on_error)
1798 hai = hai_first(hal);
1799 while (++i <= hal->hal_count) {
1800 if ((char *)hai - (char *)hal > msgsize) {
1803 "'%s' item %d past end of message!",
1808 rc = ct_process_item_async(hai, hal->hal_flags);
1810 CT_ERROR(rc, "'%s' item %d process",
1812 if (opt.o_abort_on_error && err_major)
1814 hai = hai_next(hai);
1817 llapi_hsm_action_list_free(&hal);
1819 if (opt.o_abort_on_error && err_major)
1823 llapi_hsm_copytool_unregister(&ctdata);
1828 static int ct_setup(void)
1832 /* set llapi message level */
1833 llapi_msg_set_level(opt.o_verbose);
1835 arc_fd = open(opt.o_hsm_root, O_RDONLY);
1838 CT_ERROR(rc, "cannot open archive at '%s'", opt.o_hsm_root);
1842 rc = llapi_search_fsname(opt.o_mnt, fs_name);
1844 CT_ERROR(rc, "cannot find a Lustre filesystem mounted at '%s'",
1852 static int ct_cleanup(void)
1859 if (close(arc_fd) < 0) {
1861 CT_ERROR(rc, "cannot close archive root directory");
1868 int main(int argc, char **argv)
1872 strncpy(cmd_name, basename(argv[0]), sizeof(cmd_name));
1873 rc = ct_parseopts(argc, argv);
1875 CT_WARN("try '%s --help' for more information", cmd_name);
1881 switch (opt.o_action) {
1883 rc = ct_import_recurse(opt.o_src);
1889 rc = ct_max_sequence();
1896 if (opt.o_action != CA_MAXSEQ)
1897 CT_TRACE("process finished, errs: %d major, %d minor,"
1898 " rc=%d (%s)", err_major, err_minor, rc,