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;
84 char *o_src; /* for import, or rebind */
85 char *o_dst; /* for import, or rebind */
88 /* everything else is zeroed */
89 struct options opt = {
92 .o_verbose = LLAPI_MSG_INFO,
94 .o_report_int = REPORT_INTERVAL_DEFAULT,
95 .o_chunk_size = ONE_MB,
98 /* hsm_copytool_private will hold an open FD on the lustre mount point
99 * for us. Additionally open one on the archive FS root to make sure
100 * it doesn't drop out from under us (and remind the admin to shutdown
101 * the copytool before unmounting). */
103 static int arc_fd = -1;
105 static int err_major;
106 static int err_minor;
108 static char cmd_name[PATH_MAX];
109 static char fs_name[MAX_OBD_NAME + 1];
111 static struct hsm_copytool_private *ctdata;
114 #define CT_ERROR(_rc, _format, ...) \
115 llapi_error(LLAPI_MSG_ERROR, _rc, \
116 "%s[%ld]: "_format, \
117 cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
118 #define CT_DEBUG(_format, ...) \
119 llapi_error(LLAPI_MSG_DEBUG | LLAPI_MSG_NO_ERRNO, 0, \
120 "%s[%ld]: "_format, \
121 cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
122 #define CT_WARN(_format, ...) \
123 llapi_error(LLAPI_MSG_WARN | LLAPI_MSG_NO_ERRNO, 0, \
124 "%s[%ld]: "_format, \
125 cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
126 #define CT_TRACE(_format, ...) \
127 llapi_error(LLAPI_MSG_INFO | LLAPI_MSG_NO_ERRNO, 0, \
128 "%s[%ld]: "_format, \
129 cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
130 #define CT_PRINTF(_format, ...) \
131 llapi_printf(LLAPI_MSG_NORMAL, \
132 "%s[%ld]: "_format, \
133 cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
135 static void usage(const char *name, int rc)
138 " Usage: %s [options]... <mode> <lustre_mount_point>\n"
139 "The Lustre HSM Posix copy tool can be used as a daemon or "
140 "as a command line tool\n"
141 "The Lustre HSM daemon acts on action requests from Lustre\n"
142 "to copy files to and from an HSM archive system.\n"
143 "This POSIX-flavored daemon makes regular POSIX filesystem calls\n"
144 "to an HSM mounted at a given hsm_root.\n"
145 " --daemon Daemon mode, run in background\n"
147 " --no-attr Don't copy file attributes\n"
148 " --no-shadow Don't create shadow namespace in archive\n"
149 " --no-xattr Don't copy file extended attributes\n"
150 "The Lustre HSM tool performs administrator-type actions\n"
151 "on a Lustre HSM archive.\n"
152 "This POSIX-flavored tool can link an existing HSM namespace\n"
153 "into a Lustre filesystem.\n"
155 " %s [options] --import <src> <dst> <lustre_mount_point>\n"
156 " import an archived subtree at\n"
157 " <src> (relative to hsm_root) into the Lustre filesystem at\n"
158 " <dst> (absolute)\n"
159 " %s [options] --rebind <old_FID> <new_FID> <lustre_mount_point>\n"
160 " rebind an entry in the HSM to a new FID\n"
161 " <old_FID> old FID the HSM entry is bound to\n"
162 " <new_FID> new FID to bind the HSM entry to\n"
163 " %s [options] --rebind <list_file> <lustre_mount_point>\n"
164 " perform the rebind operation for all FID in the list file\n"
165 " each line of <list_file> consists of <old_FID> <new_FID>\n"
166 " %s [options] --max-sequence <fsname>\n"
167 " return the max fid sequence of archived files\n"
168 " --abort-on-error Abort operation on major error\n"
169 " -A, --archive <#> Archive number (repeatable)\n"
170 " -b, --bandwidth <bw> Limit I/O bandwidth (unit can be used\n,"
172 " --dry-run Don't run, just show what would be done\n"
173 " -c, --chunk-size <sz> I/O size used during data copy\n"
174 " (unit can be used, default is MB)\n"
175 " -f, --event-fifo <path> Write events stream to fifo\n"
176 " -p, --hsm-root <path> Target HSM mount point\n"
177 " -q, --quiet Produce less verbose output\n"
178 " -u, --update-interval <s> Interval between progress reports sent\n"
180 " -v, --verbose Produce more verbose output\n",
181 cmd_name, cmd_name, cmd_name, cmd_name, cmd_name);
186 static int ct_parseopts(int argc, char * const *argv)
188 struct option long_opts[] = {
189 {"abort-on-error", no_argument, &opt.o_abort_on_error, 1},
190 {"abort_on_error", no_argument, &opt.o_abort_on_error, 1},
191 {"archive", required_argument, NULL, 'A'},
192 {"bandwidth", required_argument, NULL, 'b'},
193 {"chunk-size", required_argument, NULL, 'c'},
194 {"chunk_size", required_argument, NULL, 'c'},
195 {"daemon", no_argument, &opt.o_daemonize, 1},
196 {"event-fifo", required_argument, NULL, 'f'},
197 {"event_fifo", required_argument, NULL, 'f'},
198 {"dry-run", no_argument, &opt.o_dry_run, 1},
199 {"help", no_argument, NULL, 'h'},
200 {"hsm-root", required_argument, NULL, 'p'},
201 {"hsm_root", required_argument, NULL, 'p'},
202 {"import", no_argument, NULL, 'i'},
203 {"max-sequence", no_argument, NULL, 'M'},
204 {"max_sequence", no_argument, NULL, 'M'},
205 {"no-attr", no_argument, &opt.o_copy_attrs, 0},
206 {"no_attr", no_argument, &opt.o_copy_attrs, 0},
207 {"no-shadow", no_argument, &opt.o_shadow_tree, 0},
208 {"no_shadow", no_argument, &opt.o_shadow_tree, 0},
209 {"no-xattr", no_argument, &opt.o_copy_xattrs, 0},
210 {"no_xattr", no_argument, &opt.o_copy_xattrs, 0},
211 {"quiet", no_argument, NULL, 'q'},
212 {"rebind", no_argument, NULL, 'r'},
213 {"update-interval", required_argument, NULL, 'u'},
214 {"update_interval", required_argument, NULL, 'u'},
215 {"verbose", no_argument, NULL, 'v'},
219 unsigned long long value;
220 unsigned long long unit;
223 while ((c = getopt_long(argc, argv, "A:b:c:f:hiMp:qru:v",
224 long_opts, NULL)) != -1) {
227 if ((opt.o_archive_cnt >= MAX_ARCHIVE_CNT) ||
228 (atoi(optarg) >= MAX_ARCHIVE_CNT)) {
230 CT_ERROR(rc, "archive number must be less"
231 "than %zu", MAX_ARCHIVE_CNT);
234 opt.o_archive_id[opt.o_archive_cnt] = atoi(optarg);
237 case 'b': /* -b and -c have both a number with unit as arg */
240 if (llapi_parse_size(optarg, &value, &unit, 0) < 0) {
242 CT_ERROR(rc, "bad value for -%c '%s'", c,
247 opt.o_chunk_size = value;
249 opt.o_bandwidth = value;
252 opt.o_event_fifo = optarg;
257 opt.o_action = CA_IMPORT;
260 opt.o_action = CA_MAXSEQ;
263 opt.o_hsm_root = optarg;
269 opt.o_action = CA_REBIND;
272 opt.o_report_int = atoi(optarg);
273 if (opt.o_report_int < 0) {
275 CT_ERROR(rc, "bad value for -%c '%s'", c,
290 switch (opt.o_action) {
292 /* src dst mount_point */
293 if (argc != optind + 3) {
295 CT_ERROR(rc, "--import requires 2 arguments");
298 opt.o_src = argv[optind++];
299 opt.o_dst = argv[optind++];
302 /* FID1 FID2 mount_point or FILE mount_point */
303 if (argc == optind + 2) {
304 opt.o_src = argv[optind++];
306 } else if (argc == optind + 3) {
307 opt.o_src = argv[optind++];
308 opt.o_dst = argv[optind++];
311 CT_ERROR(rc, "--rebind requires 1 or 2 arguments");
317 /* just mount point */
321 if (argc != optind + 1) {
323 CT_ERROR(rc, "no mount point specified");
327 opt.o_mnt = argv[optind];
329 CT_TRACE("action=%d src=%s dst=%s mount_point=%s",
330 opt.o_action, opt.o_src, opt.o_dst, opt.o_mnt);
332 if (!opt.o_dry_run && opt.o_hsm_root == NULL) {
334 CT_ERROR(rc, "must specify a root directory for the backend");
338 if (opt.o_action == CA_IMPORT) {
339 if (opt.o_src && opt.o_src[0] == '/') {
342 "source path must be relative to HSM root");
346 if (opt.o_dst && opt.o_dst[0] != '/') {
348 CT_ERROR(rc, "destination path must be absolute");
357 static int ct_mkdir_p(const char *path)
367 while ((ptr = strchr(ptr, '/')) != NULL) {
369 rc = mkdir(saved, DIR_PERM);
371 if (rc < 0 && errno != EEXIST) {
373 CT_ERROR(rc, "cannot mkdir '%s'", path);
385 static int ct_save_stripe(int src_fd, const char *src, const char *dst)
387 char lov_file[PATH_MAX];
388 char lov_buf[XATTR_SIZE_MAX];
389 struct lov_user_md *lum;
394 snprintf(lov_file, sizeof(lov_file), "%s.lov", dst);
395 CT_TRACE("saving stripe info of '%s' in %s", src, lov_file);
397 xattr_size = fgetxattr(src_fd, XATTR_LUSTRE_LOV, lov_buf,
399 if (xattr_size < 0) {
401 CT_ERROR(rc, "cannot get stripe info on '%s'", src);
405 lum = (struct lov_user_md *)lov_buf;
407 if (lum->lmm_magic == LOV_USER_MAGIC_V1 ||
408 lum->lmm_magic == LOV_USER_MAGIC_V3) {
409 /* Set stripe_offset to -1 so that it is not interpreted as a
410 * hint on restore. */
411 lum->lmm_stripe_offset = -1;
414 fd = open(lov_file, O_TRUNC | O_CREAT | O_WRONLY, FILE_PERM);
417 CT_ERROR(rc, "cannot open '%s'", lov_file);
421 rc = write(fd, lum, xattr_size);
424 CT_ERROR(rc, "cannot write %zd bytes to '%s'",
425 xattr_size, lov_file);
433 CT_ERROR(rc, "cannot close '%s'", lov_file);
445 static int ct_load_stripe(const char *src, void *lovea, size_t *lovea_size)
447 char lov_file[PATH_MAX];
451 snprintf(lov_file, sizeof(lov_file), "%s.lov", src);
452 CT_TRACE("reading stripe rules from '%s' for '%s'", lov_file, src);
454 fd = open(lov_file, O_RDONLY);
456 CT_ERROR(errno, "cannot open '%s'", lov_file);
460 rc = read(fd, lovea, *lovea_size);
462 CT_ERROR(errno, "cannot read %zu bytes from '%s'",
463 *lovea_size, lov_file);
474 static int ct_restore_stripe(const char *src, const char *dst, int dst_fd,
475 const void *lovea, size_t lovea_size)
479 rc = fsetxattr(dst_fd, XATTR_LUSTRE_LOV, lovea, lovea_size,
482 CT_ERROR(errno, "cannot set lov EA on '%s'", dst);
489 static void bandwidth_ctl_delay(int wsize)
491 static unsigned long long tot_bytes;
492 static time_t start_time;
493 static time_t last_time;
494 time_t now = time(0);
497 unsigned int sleep_time;
499 if (now > last_time + 5) {
501 start_time = last_time = now;
505 tot_time = now - start_time;
509 excess = tot_bytes - tot_time * opt.o_bandwidth;
510 sleep_time = excess * 1000000 / opt.o_bandwidth;
511 if ((now - start_time) % 10 == 1)
512 CT_TRACE("bandwith control: excess=%E sleep for %dus", excess,
521 static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
522 const char *dst, int src_fd, int dst_fd,
523 const struct hsm_action_item *hai, long hal_flags)
525 struct hsm_extent he;
526 __u64 offset = hai->hai_extent.offset;
530 __u64 write_total = 0;
532 time_t last_print_time = time(NULL);
535 CT_TRACE("going to copy data from '%s' to '%s'", src, dst);
537 if (fstat(src_fd, &src_st) < 0) {
539 CT_ERROR(rc, "cannot stat '%s'", src);
543 if (!S_ISREG(src_st.st_mode)) {
545 CT_ERROR(rc, "'%s' is not a regular file", src);
549 if (fstat(dst_fd, &dst_st) < 0) {
551 CT_ERROR(rc, "cannot stat '%s'", dst);
555 if (!S_ISREG(dst_st.st_mode)) {
557 CT_ERROR(rc, "'%s' is not a regular file", dst);
561 rc = lseek(src_fd, hai->hai_extent.offset, SEEK_SET);
565 "cannot seek for read to "LPU64" (len %jd) in '%s'",
566 hai->hai_extent.offset, (intmax_t)src_st.st_size, src);
570 /* Don't read beyond a given extent */
571 length = min(hai->hai_extent.length, src_st.st_size);
575 rc = llapi_hsm_action_progress(hcp, &he, length, 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",
586 buf = malloc(opt.o_chunk_size);
592 CT_DEBUG("Going to copy "LPU64" bytes %s -> %s\n", length, src, dst);
594 while (write_total < length) {
597 int chunk = (length - write_total > opt.o_chunk_size) ?
598 opt.o_chunk_size : length - write_total;
600 rsize = pread(src_fd, buf, chunk, offset);
607 CT_ERROR(rc, "cannot read from '%s'", src);
611 wsize = pwrite(dst_fd, buf, rsize, offset);
614 CT_ERROR(rc, "cannot write to '%s'", dst);
618 write_total += wsize;
621 if (opt.o_bandwidth != 0)
622 /* sleep if needed, to honor bandwidth limits */
623 bandwidth_ctl_delay(wsize);
625 if (time(0) >= last_print_time + opt.o_report_int) {
626 last_print_time = time(0);
627 CT_TRACE("%%"LPU64" ", 100 * write_total / length);
628 he.length = write_total;
629 rc = llapi_hsm_action_progress(hcp, &he, length, 0);
631 /* Action has been canceled or something wrong
632 * is happening. Stop copying data. */
633 CT_ERROR(rc, "progress ioctl for copy"
634 " '%s'->'%s' failed", src, dst);
643 * truncate restored file
644 * size is taken from the archive this is done to support
645 * restore after a force release which leaves the file with the
646 * wrong size (can big bigger than the new size)
648 if ((hai->hai_action == HSMA_RESTORE) &&
649 (src_st.st_size < dst_st.st_size)) {
651 * make sure the file is on disk before reporting success.
653 rc = ftruncate(dst_fd, src_st.st_size);
656 CT_ERROR(rc, "cannot truncate '%s' to size %jd",
657 dst, (intmax_t)src_st.st_size);
668 /* Copy file attributes from file src to file dest */
669 static int ct_copy_attr(const char *src, const char *dst, int src_fd,
673 struct timeval times[2];
676 if (fstat(src_fd, &st) < 0) {
678 CT_ERROR(rc, "cannot stat '%s'", src);
682 times[0].tv_sec = st.st_atime;
683 times[0].tv_usec = 0;
684 times[1].tv_sec = st.st_mtime;
685 times[1].tv_usec = 0;
686 if (fchmod(dst_fd, st.st_mode) < 0 ||
687 fchown(dst_fd, st.st_uid, st.st_gid) < 0 ||
688 futimes(dst_fd, times) < 0) {
690 CT_ERROR(rc, "cannot set attributes of '%s'", src);
697 static int ct_copy_xattr(const char *src, const char *dst, int src_fd,
698 int dst_fd, bool is_restore)
700 char list[XATTR_LIST_MAX];
701 char value[XATTR_SIZE_MAX];
706 list_len = flistxattr(src_fd, list, sizeof(list));
711 while (name < list + list_len) {
712 rc = fgetxattr(src_fd, name, value, sizeof(value));
716 /* when we restore, we do not restore lustre xattr */
718 (strncmp(XATTR_TRUSTED_PREFIX, name,
719 sizeof(XATTR_TRUSTED_PREFIX) - 1) != 0)) {
720 rc = fsetxattr(dst_fd, name, value, rc, 0);
721 CT_TRACE("fsetxattr of '%s' on '%s' rc=%d (%s)",
722 name, dst, rc, strerror(errno));
723 /* lustre.* attrs aren't supported on other FS's */
724 if (rc < 0 && errno != EOPNOTSUPP) {
726 CT_ERROR(rc, "cannot set extended attribute"
732 name += strlen(name) + 1;
738 static int ct_path_lustre(char *buf, int sz, const char *mnt,
739 const lustre_fid *fid)
741 return snprintf(buf, sz, "%s/%s/fid/"DFID_NOBRACE, mnt,
742 dot_lustre_name, PFID(fid));
745 static int ct_path_archive(char *buf, int sz, const char *archive_dir,
746 const lustre_fid *fid)
748 return snprintf(buf, sz, "%s/%04x/%04x/%04x/%04x/%04x/%04x/"
749 DFID_NOBRACE, archive_dir,
750 (fid)->f_oid & 0xFFFF,
751 (fid)->f_oid >> 16 & 0xFFFF,
752 (unsigned int)((fid)->f_seq & 0xFFFF),
753 (unsigned int)((fid)->f_seq >> 16 & 0xFFFF),
754 (unsigned int)((fid)->f_seq >> 32 & 0xFFFF),
755 (unsigned int)((fid)->f_seq >> 48 & 0xFFFF),
759 static bool ct_is_retryable(int err)
761 return err == -ETIMEDOUT;
764 static int ct_begin_restore(struct hsm_copyaction_private **phcp,
765 const struct hsm_action_item *hai,
766 int mdt_index, int open_flags)
771 rc = llapi_hsm_action_begin(phcp, ctdata, hai, mdt_index, open_flags,
774 ct_path_lustre(src, sizeof(src), opt.o_mnt, &hai->hai_fid);
775 CT_ERROR(rc, "llapi_hsm_action_begin() on '%s' failed", src);
781 static int ct_begin(struct hsm_copyaction_private **phcp,
782 const struct hsm_action_item *hai)
784 /* Restore takes specific parameters. Call the same function w/ default
785 * values for all other operations. */
786 return ct_begin_restore(phcp, hai, -1, 0);
789 static int ct_fini(struct hsm_copyaction_private **phcp,
790 const struct hsm_action_item *hai, int hp_flags, int ct_rc)
795 CT_TRACE("Action completed, notifying coordinator "
796 "cookie="LPX64", FID="DFID", hp_flags=%d err=%d",
797 hai->hai_cookie, PFID(&hai->hai_fid),
800 ct_path_lustre(lstr, sizeof(lstr), opt.o_mnt, &hai->hai_fid);
801 rc = llapi_hsm_action_end(phcp, &hai->hai_extent, hp_flags, abs(ct_rc));
802 if (rc == -ECANCELED)
803 CT_ERROR(rc, "completed action on '%s' has been canceled: "
804 "cookie="LPX64", FID="DFID, lstr, hai->hai_cookie,
805 PFID(&hai->hai_fid));
807 CT_ERROR(rc, "llapi_hsm_action_end() on '%s' failed", lstr);
809 CT_TRACE("llapi_hsm_action_end() on '%s' ok (rc=%d)",
815 static int ct_archive(const struct hsm_action_item *hai, const long hal_flags)
817 struct hsm_copyaction_private *hcp = NULL;
819 char dst[PATH_MAX] = "";
822 bool rename_needed = false;
828 rc = ct_begin(&hcp, hai);
832 /* we fill archive so:
834 * destination = lustre FID
836 ct_path_lustre(src, sizeof(src), opt.o_mnt, &hai->hai_dfid);
837 ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, &hai->hai_fid);
838 if (hai->hai_extent.length == -1) {
839 /* whole file, write it to tmp location and atomically
840 * replace old archived file */
841 strncat(dst, "_tmp", sizeof(dst) - strlen(dst) - 1);
842 /* we cannot rely on the same test because ct_copy_data()
843 * updates hai_extent.length */
844 rename_needed = true;
847 CT_TRACE("archiving '%s' to '%s'", src, dst);
854 rc = ct_mkdir_p(dst);
856 CT_ERROR(rc, "mkdir_p '%s' failed", dst);
860 src_fd = llapi_hsm_action_get_fd(hcp);
863 CT_ERROR(rc, "cannot open '%s' for read", src);
867 open_flags = O_WRONLY | O_NOFOLLOW | O_NONBLOCK;
868 /* If extent is specified, don't truncate an old archived copy */
869 open_flags |= ((hai->hai_extent.length == -1) ? O_TRUNC : 0) | O_CREAT;
871 dst_fd = open(dst, open_flags, FILE_PERM);
874 CT_ERROR(rc, "cannot open '%s' for write", dst);
878 /* saving stripe is not critical */
879 rc = ct_save_stripe(src_fd, src, dst);
881 CT_ERROR(rc, "cannot save file striping info of '%s' in '%s'",
884 rc = ct_copy_data(hcp, src, dst, src_fd, dst_fd, hai, hal_flags);
886 CT_ERROR(rc, "data copy failed from '%s' to '%s'", src, dst);
893 CT_ERROR(rc, "cannot flush '%s' archive file '%s'", src, dst);
897 CT_TRACE("data archiving for '%s' to '%s' done", src, dst);
899 /* attrs will remain on the MDS; no need to copy them, except possibly
900 for disaster recovery */
901 if (opt.o_copy_attrs) {
902 rc = ct_copy_attr(src, dst, src_fd, dst_fd);
904 CT_ERROR(rc, "cannot copy attr of '%s' to '%s'",
908 CT_TRACE("attr file for '%s' saved to archive '%s'",
912 /* xattrs will remain on the MDS; no need to copy them, except possibly
913 for disaster recovery */
914 if (opt.o_copy_xattrs) {
915 rc = ct_copy_xattr(src, dst, src_fd, dst_fd, false);
917 CT_ERROR(rc, "cannot copy xattr of '%s' to '%s'",
919 rcf = rcf ? rcf : rc;
921 CT_TRACE("xattr file for '%s' saved to archive '%s'",
925 if (rename_needed == true) {
926 char tmp_src[PATH_MAX];
927 char tmp_dst[PATH_MAX];
929 /* atomically replace old archived file */
930 ct_path_archive(src, sizeof(src), opt.o_hsm_root,
932 rc = rename(dst, src);
935 CT_ERROR(rc, "cannot rename '%s' to '%s'", dst, src);
938 /* rename lov file */
939 snprintf(tmp_src, sizeof(tmp_src), "%s.lov", src);
940 snprintf(tmp_dst, sizeof(tmp_dst), "%s.lov", dst);
941 rc = rename(tmp_dst, tmp_src);
943 CT_ERROR(errno, "cannot rename '%s' to '%s'",
947 if (opt.o_shadow_tree) {
948 /* Create a namespace of softlinks that shadows the original
949 * Lustre namespace. This will only be current at
950 * time-of-archive (won't follow renames).
951 * WARNING: release won't kill these links; a manual
952 * cleanup of dead links would be required.
955 long long recno = -1;
961 sprintf(buf, DFID, PFID(&hai->hai_fid));
962 sprintf(src, "%s/shadow/", opt.o_hsm_root);
964 ptr = opt.o_hsm_root;
966 (*ptr++ == '/') ? depth-- : 0;
968 rc = llapi_fid2path(opt.o_mnt, buf, src + strlen(src),
969 sizeof(src) - strlen(src), &recno, &linkno);
971 CT_ERROR(rc, "cannot get FID of '%s'", buf);
972 rcf = rcf ? rcf : rc;
976 /* Figure out how many parent dirs to symlink back */
979 (*ptr++ == '/') ? depth++ : 0;
984 ct_path_archive(dst, sizeof(dst), buf, &hai->hai_fid);
986 if (ct_mkdir_p(src)) {
987 CT_ERROR(errno, "mkdir_p '%s' failed", src);
988 rcf = rcf ? rcf : -errno;
991 /* symlink already exists ? */
992 sz = readlink(src, buf, sizeof(buf));
993 /* detect truncation */
994 if (sz == sizeof(buf)) {
995 rcf = rcf ? rcf : -E2BIG;
996 CT_ERROR(rcf, "readlink '%s' truncated", src);
1001 if (sz == 0 || strncmp(buf, dst, sz) != 0) {
1002 if (unlink(src) && errno != ENOENT) {
1004 "cannot unlink symlink '%s'",
1006 rcf = rcf ? rcf : -errno;
1008 /* unlink old symlink done */
1009 CT_TRACE("remove old symlink '%s' pointing"
1010 " to '%s'", src, buf);
1013 /* symlink already ok */
1014 CT_TRACE("symlink '%s' already pointing"
1015 " to '%s'", src, dst);
1020 if (symlink(dst, src)) {
1021 CT_ERROR(errno, "cannot symlink '%s' to '%s'",
1023 rcf = rcf ? rcf : -errno;
1026 CT_TRACE("symlink '%s' to '%s' done", src, dst);
1038 if (ct_is_retryable(rc))
1039 hp_flags |= HP_FLAG_RETRY;
1051 rc = ct_fini(&hcp, hai, hp_flags, rcf);
1056 static int ct_restore(const struct hsm_action_item *hai, const long hal_flags)
1058 struct hsm_copyaction_private *hcp = NULL;
1061 char lov_buf[XATTR_SIZE_MAX];
1062 size_t lov_size = sizeof(lov_buf);
1067 int mdt_index = -1; /* Not implemented */
1071 /* we fill lustre so:
1072 * source = lustre FID in the backend
1073 * destination = data FID = volatile file
1076 /* build backend file name from released file FID */
1077 ct_path_archive(src, sizeof(src), opt.o_hsm_root, &hai->hai_fid);
1079 /* restore loads and sets the LOVEA w/o interpreting it to avoid
1080 * dependency on the structure format. */
1081 rc = ct_load_stripe(src, lov_buf, &lov_size);
1083 CT_WARN("cannot get stripe rules for '%s' (%s), use default",
1084 src, strerror(-rc));
1087 open_flags |= O_LOV_DELAY_CREATE;
1091 rc = ct_begin_restore(&hcp, hai, mdt_index, open_flags);
1095 /* get the FID of the volatile file */
1096 rc = llapi_hsm_action_get_dfid(hcp, &dfid);
1098 CT_ERROR(rc, "restoring "DFID
1099 ", cannot get FID of created volatile file",
1100 PFID(&hai->hai_fid));
1104 /* build volatile "file name", for messages */
1105 snprintf(dst, sizeof(dst), "{VOLATILE}="DFID, PFID(&dfid));
1107 CT_TRACE("restoring data from '%s' to '%s'", src, dst);
1109 if (opt.o_dry_run) {
1114 src_fd = open(src, O_RDONLY | O_NOATIME | O_NONBLOCK | O_NOFOLLOW);
1117 CT_ERROR(rc, "cannot open '%s' for read", src);
1121 dst_fd = llapi_hsm_action_get_fd(hcp);
1124 /* the layout cannot be allocated through .fid so we have to
1125 * restore a layout */
1126 rc = ct_restore_stripe(src, dst, dst_fd, lov_buf, lov_size);
1128 CT_ERROR(rc, "cannot restore file striping info"
1129 " for '%s' from '%s'", dst, src);
1135 rc = ct_copy_data(hcp, src, dst, src_fd, dst_fd, hai, hal_flags);
1137 CT_ERROR(rc, "cannot copy data from '%s' to '%s'",
1140 if (ct_is_retryable(rc))
1141 hp_flags |= HP_FLAG_RETRY;
1145 CT_TRACE("data restore from '%s' to '%s' done", src, dst);
1149 rc = ct_fini(&hcp, hai, hp_flags, rc);
1151 /* object swaping is done by cdt at copy end, so close of volatile file
1152 * cannot be done before */
1162 static int ct_remove(const struct hsm_action_item *hai, const long hal_flags)
1164 struct hsm_copyaction_private *hcp = NULL;
1168 rc = ct_begin(&hcp, hai);
1172 ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, &hai->hai_fid);
1174 CT_TRACE("removing file '%s'", dst);
1176 if (opt.o_dry_run) {
1184 CT_ERROR(rc, "cannot unlink '%s'", dst);
1189 strncat(dst, ".lov", sizeof(dst) - strlen(dst) - 1);
1193 CT_ERROR(rc, "cannot unlink '%s'", dst);
1200 rc = ct_fini(&hcp, hai, 0, rc);
1205 static int ct_report_error(const struct hsm_action_item *hai, int flags,
1208 struct hsm_copyaction_private *hcp;
1211 rc = llapi_hsm_action_begin(&hcp, ctdata, hai, -1, 0, true);
1215 rc = llapi_hsm_action_end(&hcp, &hai->hai_extent, flags, abs(errval));
1220 static int ct_process_item(struct hsm_action_item *hai, const long hal_flags)
1224 if (opt.o_verbose >= LLAPI_MSG_INFO || opt.o_dry_run) {
1225 /* Print the original path */
1227 char path[PATH_MAX];
1228 long long recno = -1;
1231 sprintf(fid, DFID, PFID(&hai->hai_fid));
1232 CT_TRACE("'%s' action %s reclen %d, cookie="LPX64,
1233 fid, hsm_copytool_action2name(hai->hai_action),
1234 hai->hai_len, hai->hai_cookie);
1235 rc = llapi_fid2path(opt.o_mnt, fid, path,
1236 sizeof(path), &recno, &linkno);
1238 CT_ERROR(rc, "cannot get path of FID %s", fid);
1240 CT_TRACE("processing file '%s'", path);
1243 switch (hai->hai_action) {
1244 /* set err_major, minor inside these functions */
1246 rc = ct_archive(hai, hal_flags);
1249 rc = ct_restore(hai, hal_flags);
1252 rc = ct_remove(hai, hal_flags);
1255 CT_TRACE("cancel not implemented for file system '%s'",
1257 /* Don't report progress to coordinator for this cookie:
1258 * the copy function will get ECANCELED when reporting
1265 CT_ERROR(rc, "unknown action %d, on '%s'", hai->hai_action,
1268 ct_report_error(hai, 0, rc);
1276 struct hsm_action_item *hai;
1279 static void *ct_thread(void *data)
1281 struct ct_th_data *cttd = data;
1284 rc = ct_process_item(cttd->hai, cttd->hal_flags);
1288 pthread_exit((void *)(intptr_t)rc);
1291 static int ct_process_item_async(const struct hsm_action_item *hai,
1294 pthread_attr_t attr;
1296 struct ct_th_data *data;
1299 data = malloc(sizeof(*data));
1303 data->hai = malloc(hai->hai_len);
1304 if (data->hai == NULL) {
1309 memcpy(data->hai, hai, hai->hai_len);
1310 data->hal_flags = hal_flags;
1312 rc = pthread_attr_init(&attr);
1314 CT_ERROR(rc, "pthread_attr_init failed for '%s' service",
1321 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1323 rc = pthread_create(&thread, &attr, ct_thread, data);
1325 CT_ERROR(rc, "cannot create thread for '%s' service",
1328 pthread_attr_destroy(&attr);
1332 static int ct_import_one(const char *src, const char *dst)
1334 char newarc[PATH_MAX];
1339 CT_TRACE("importing '%s' from '%s'", dst, src);
1341 if (stat(src, &st) < 0) {
1343 CT_ERROR(rc, "cannot stat '%s'", src);
1350 rc = llapi_hsm_import(dst,
1351 opt.o_archive_cnt ? opt.o_archive_id[0] : 0,
1352 &st, 0, 0, 0, 0, NULL, &fid);
1354 CT_ERROR(rc, "cannot import '%s' from '%s'", dst, src);
1358 ct_path_archive(newarc, sizeof(newarc), opt.o_hsm_root, &fid);
1360 rc = ct_mkdir_p(newarc);
1362 CT_ERROR(rc, "mkdir_p '%s' failed", newarc);
1368 /* Lots of choices now: mv, ln, ln -s ? */
1369 rc = link(src, newarc); /* hardlink */
1372 CT_ERROR(rc, "cannot link '%s' to '%s'", newarc, src);
1376 CT_TRACE("imported '%s' from '%s'=='%s'", dst, newarc, src);
1381 static char *path_concat(const char *dirname, const char *basename)
1384 int dirlen = strlen(dirname);
1386 result = malloc(dirlen + strlen(basename) + 2);
1390 memcpy(result, dirname, dirlen);
1391 result[dirlen] = '/';
1392 strcpy(result + dirlen + 1, basename);
1397 static int ct_import_recurse(const char *relpath)
1400 struct dirent ent, *cookie = NULL;
1401 char *srcpath, *newpath;
1404 if (relpath == NULL)
1407 srcpath = path_concat(opt.o_hsm_root, relpath);
1408 if (srcpath == NULL) {
1413 dir = opendir(srcpath);
1415 /* Not a dir, or error */
1416 if (errno == ENOTDIR) {
1417 /* Single regular file case, treat o_dst as absolute
1419 rc = ct_import_one(srcpath, opt.o_dst);
1422 CT_ERROR(rc, "cannot opendir '%s'", srcpath);
1431 rc = readdir_r(dir, &ent, &cookie);
1434 CT_ERROR(rc, "cannot readdir_r '%s'", relpath);
1437 } else if ((rc == 0) && (cookie == NULL)) {
1438 /* end of directory */
1442 if (!strcmp(ent.d_name, ".") ||
1443 !strcmp(ent.d_name, ".."))
1446 /* New relative path */
1447 newpath = path_concat(relpath, ent.d_name);
1448 if (newpath == NULL) {
1454 if (ent.d_type == DT_DIR) {
1455 rc = ct_import_recurse(newpath);
1460 sprintf(src, "%s/%s", opt.o_hsm_root, newpath);
1461 sprintf(dst, "%s/%s", opt.o_dst, newpath);
1462 /* Make the target dir in the Lustre fs */
1463 rc = ct_mkdir_p(dst);
1465 /* Import the file */
1466 rc = ct_import_one(src, dst);
1468 CT_ERROR(rc, "ct_mkdir_p '%s' failed", dst);
1474 CT_ERROR(rc, "cannot import '%s'", newpath);
1475 if (err_major && opt.o_abort_on_error) {
1489 static int ct_rebind_one(const lustre_fid *old_fid, const lustre_fid *new_fid)
1495 CT_TRACE("rebind "DFID" to "DFID, PFID(old_fid), PFID(new_fid));
1497 ct_path_archive(src, sizeof(src), opt.o_hsm_root, old_fid);
1498 ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, new_fid);
1500 if (!opt.o_dry_run) {
1502 if (rename(src, dst)) {
1504 CT_ERROR(rc, "cannot rename '%s' to '%s'", src, dst);
1507 /* rename lov file */
1508 strncat(src, ".lov", sizeof(src) - strlen(src) - 1);
1509 strncat(dst, ".lov", sizeof(dst) - strlen(dst) - 1);
1510 if (rename(src, dst))
1511 CT_ERROR(errno, "cannot '%s' rename to '%s'", src, dst);
1517 static bool fid_is_file(lustre_fid *fid)
1519 return fid_is_norm(fid) || fid_is_igif(fid);
1522 static bool should_ignore_line(const char *line)
1526 for (i = 0; line[i] != '\0'; i++) {
1527 if (isspace(line[i]))
1529 else if (line[i] == '#')
1538 static int ct_rebind_list(const char *list)
1544 size_t line_size = 0;
1545 unsigned int nl = 0;
1546 unsigned int ok = 0;
1548 filp = fopen(list, "r");
1551 CT_ERROR(rc, "cannot open '%s'", list);
1555 /* each line consists of 2 FID */
1556 while ((r = getline(&line, &line_size, filp)) != -1) {
1560 /* Ignore empty and commented out ('#...') lines. */
1561 if (should_ignore_line(line))
1566 rc = sscanf(line, SFID" "SFID, RFID(&old_fid), RFID(&new_fid));
1567 if (rc != 6 || !fid_is_file(&old_fid) ||
1568 !fid_is_file(&new_fid)) {
1570 "'%s' FID expected near '%s', line %u",
1576 if (ct_rebind_one(&old_fid, &new_fid))
1587 /* return 0 if all rebinds were sucessful */
1588 CT_TRACE("%u lines read from '%s', %u rebind successful", nl, list, ok);
1590 return ok == nl ? 0 : -1;
1593 static int ct_rebind(void)
1601 if (sscanf(opt.o_src, SFID, RFID(&old_fid)) != 3 ||
1602 !fid_is_file(&old_fid)) {
1604 CT_ERROR(rc, "'%s' invalid FID format", opt.o_src);
1608 if (sscanf(opt.o_dst, SFID, RFID(&new_fid)) != 3 ||
1609 !fid_is_file(&new_fid)) {
1611 CT_ERROR(rc, "'%s' invalid FID format", opt.o_dst);
1615 rc = ct_rebind_one(&old_fid, &new_fid);
1620 /* o_src is a list file */
1621 rc = ct_rebind_list(opt.o_src);
1626 static int ct_dir_level_max(const char *dirpath, __u16 *sub_seqmax)
1631 struct dirent ent, *cookie = NULL;
1635 dir = opendir(dirpath);
1638 CT_ERROR(rc, "cannot open directory '%s'", opt.o_hsm_root);
1642 while ((rc = readdir_r(dir, &ent, &cookie)) == 0) {
1644 /* end of directory.
1645 * rc is 0 and seqmax contains the max value. */
1648 if (!strcmp(ent.d_name, ".") || !strcmp(ent.d_name, ".."))
1651 if (sscanf(ent.d_name, "%hx", &sub_seq) != 1) {
1652 CT_TRACE("'%s' has an unexpected dirname format, "
1653 "skip entry", ent.d_name);
1656 if (sub_seq > *sub_seqmax)
1657 *sub_seqmax = sub_seq;
1660 CT_ERROR(rc, "cannot readdir_r '%s'", dirpath);
1667 static int ct_max_sequence(void)
1670 char path[PATH_MAX];
1674 strncpy(path, opt.o_hsm_root, sizeof(path));
1675 /* FID sequence is stored in top-level directory names:
1676 * hsm_root/16bits (high weight)/16 bits/16 bits/16 bits (low weight).
1678 for (i = 0; i < 4; i++) {
1679 rc = ct_dir_level_max(path, &subseq);
1682 seq |= ((__u64)subseq << ((3 - i) * 16));
1683 sprintf(path + strlen(path), "/%04x", subseq);
1686 printf("max_sequence: %016Lx\n", seq);
1691 static void handler(int signal)
1693 psignal(signal, "exiting");
1694 /* If we don't clean up upon interrupt, umount thinks there's a ref
1695 * and doesn't remove us from mtab (EINPROGRESS). The lustre client
1696 * does successfully unmount and the mount is actually gone, but the
1697 * mtab entry remains. So this just makes mtab happier. */
1698 llapi_hsm_copytool_unregister(&ctdata);
1702 /* Daemon waits for messages from the kernel; run it in the background. */
1703 static int ct_run(void)
1707 if (opt.o_daemonize) {
1711 CT_ERROR(rc, "cannot daemonize");
1716 setbuf(stdout, NULL);
1718 if (opt.o_event_fifo != NULL) {
1719 rc = llapi_hsm_register_event_fifo(opt.o_event_fifo);
1721 CT_ERROR(rc, "failed to register event fifo");
1724 llapi_error_callback_set(llapi_hsm_log_error);
1727 rc = llapi_hsm_copytool_register(&ctdata, opt.o_mnt, 0,
1728 opt.o_archive_cnt, opt.o_archive_id);
1730 CT_ERROR(rc, "cannot start copytool interface");
1734 signal(SIGINT, handler);
1735 signal(SIGTERM, handler);
1738 struct hsm_action_list *hal;
1739 struct hsm_action_item *hai;
1743 CT_TRACE("waiting for message from kernel");
1745 rc = llapi_hsm_copytool_recv(ctdata, &hal, &msgsize);
1746 if (rc == -ESHUTDOWN) {
1747 CT_TRACE("shutting down");
1749 } else if (rc == -EAGAIN) {
1750 continue; /* msg not for us */
1751 } else if (rc < 0) {
1752 CT_WARN("cannot receive action list: %s",
1755 if (opt.o_abort_on_error)
1761 CT_TRACE("copytool fs=%s archive#=%d item_count=%d",
1762 hal->hal_fsname, hal->hal_archive_id, hal->hal_count);
1764 if (strcmp(hal->hal_fsname, fs_name) != 0) {
1765 CT_ERROR(EINVAL, "'%s' invalid fs name, expecting: %s",
1766 hal->hal_fsname, fs_name);
1768 if (opt.o_abort_on_error)
1774 hai = hai_first(hal);
1775 while (++i <= hal->hal_count) {
1776 if ((char *)hai - (char *)hal > msgsize) {
1779 "'%s' item %d past end of message!",
1784 rc = ct_process_item_async(hai, hal->hal_flags);
1786 CT_ERROR(rc, "'%s' item %d process",
1788 if (opt.o_abort_on_error && err_major)
1790 hai = hai_next(hai);
1793 llapi_hsm_action_list_free(&hal);
1795 if (opt.o_abort_on_error && err_major)
1799 llapi_hsm_copytool_unregister(&ctdata);
1800 if (opt.o_event_fifo != NULL)
1801 llapi_hsm_unregister_event_fifo(opt.o_event_fifo);
1806 static int ct_setup(void)
1810 /* set llapi message level */
1811 llapi_msg_set_level(opt.o_verbose);
1813 arc_fd = open(opt.o_hsm_root, O_RDONLY);
1816 CT_ERROR(rc, "cannot open archive at '%s'", opt.o_hsm_root);
1820 rc = llapi_search_fsname(opt.o_mnt, fs_name);
1822 CT_ERROR(rc, "cannot find a Lustre filesystem mounted at '%s'",
1830 static int ct_cleanup(void)
1837 if (close(arc_fd) < 0) {
1839 CT_ERROR(rc, "cannot close archive root directory");
1846 int main(int argc, char **argv)
1850 strncpy(cmd_name, basename(argv[0]), sizeof(cmd_name));
1851 rc = ct_parseopts(argc, argv);
1853 CT_WARN("try '%s --help' for more information", cmd_name);
1859 switch (opt.o_action) {
1861 rc = ct_import_recurse(opt.o_src);
1867 rc = ct_max_sequence();
1874 if (opt.o_action != CA_MAXSEQ)
1875 CT_TRACE("process finished, errs: %d major, %d minor,"
1876 " rc=%d (%s)", err_major, err_minor, rc,