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.
45 #include <sys/xattr.h>
46 #include <sys/syscall.h>
47 #include <sys/types.h>
48 #include <lustre/lustre_idl.h>
49 #include <lustre/lustreapi.h>
51 /* Progress reporting period */
52 #define REPORT_INTERVAL_DEFAULT 30
53 /* HSM hash subdir permissions */
54 #define DIR_PERM S_IRWXU
55 /* HSM hash file permissions */
56 #define FILE_PERM (S_IRUSR | S_IWUSR)
58 #define ONE_MB 0x100000
60 /* copytool uses a 32b bitmask field to register with kuc
61 * archive num = 0 => all
62 * archive num from 1 to 32
64 #define MAX_ARCHIVE_CNT (sizeof(__u32) * 8)
81 int o_archive_id[MAX_ARCHIVE_CNT];
83 unsigned long long o_bandwidth;
85 enum ct_action o_action;
89 char *o_src; /* for import, or rebind */
90 char *o_dst; /* for import, or rebind */
93 /* everything else is zeroed */
94 struct options opt = {
97 .o_verbose = LLAPI_MSG_INFO,
99 .o_report_int = REPORT_INTERVAL_DEFAULT,
100 .o_chunk_size = ONE_MB,
103 /* hsm_copytool_private will hold an open FD on the lustre mount point
104 * for us. Additionally open one on the archive FS root to make sure
105 * it doesn't drop out from under us (and remind the admin to shutdown
106 * the copytool before unmounting). */
108 static int arc_fd = -1;
110 static int err_major;
111 static int err_minor;
113 static char cmd_name[PATH_MAX];
114 static char fs_name[MAX_OBD_NAME + 1];
116 static struct hsm_copytool_private *ctdata;
119 #define CT_ERROR(_rc, _format, ...) \
120 llapi_error(LLAPI_MSG_ERROR, _rc, \
121 "%s[%ld]: "_format, \
122 cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
123 #define CT_DEBUG(_format, ...) \
124 llapi_error(LLAPI_MSG_DEBUG | LLAPI_MSG_NO_ERRNO, 0, \
125 "%s[%ld]: "_format, \
126 cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
127 #define CT_WARN(_format, ...) \
128 llapi_error(LLAPI_MSG_WARN | LLAPI_MSG_NO_ERRNO, 0, \
129 "%s[%ld]: "_format, \
130 cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
131 #define CT_TRACE(_format, ...) \
132 llapi_error(LLAPI_MSG_INFO | LLAPI_MSG_NO_ERRNO, 0, \
133 "%s[%ld]: "_format, \
134 cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
135 #define CT_PRINTF(_format, ...) \
136 llapi_printf(LLAPI_MSG_NORMAL, \
137 "%s[%ld]: "_format, \
138 cmd_name, syscall(SYS_gettid), ## __VA_ARGS__)
140 static void usage(const char *name, int rc)
143 " Usage: %s [options]... <mode> <lustre_mount_point>\n"
144 "The Lustre HSM Posix copy tool can be used as a daemon or "
145 "as a command line tool\n"
146 "The Lustre HSM daemon acts on action requests from Lustre\n"
147 "to copy files to and from an HSM archive system.\n"
148 "This POSIX-flavored daemon makes regular POSIX filesystem calls\n"
149 "to an HSM mounted at a given hsm_root.\n"
150 " --daemon Daemon mode, run in background\n"
152 " --no-attr Don't copy file attributes\n"
153 " --no-shadow Don't create shadow namespace in archive\n"
154 " --no-xattr Don't copy file extended attributes\n"
155 "The Lustre HSM tool performs administrator-type actions\n"
156 "on a Lustre HSM archive.\n"
157 "This POSIX-flavored tool can link an existing HSM namespace\n"
158 "into a Lustre filesystem.\n"
160 " %s [options] --import <src> <dst> <lustre_mount_point>\n"
161 " import an archived subtree from\n"
162 " <src> (FID or relative path to hsm_root) into the Lustre\n"
164 " <dst> (absolute path)\n"
165 " %s [options] --rebind <old_FID> <new_FID> <lustre_mount_point>\n"
166 " rebind an entry in the HSM to a new FID\n"
167 " <old_FID> old FID the HSM entry is bound to\n"
168 " <new_FID> new FID to bind the HSM entry to\n"
169 " %s [options] --rebind <list_file> <lustre_mount_point>\n"
170 " perform the rebind operation for all FID in the list file\n"
171 " each line of <list_file> consists of <old_FID> <new_FID>\n"
172 " %s [options] --max-sequence <fsname>\n"
173 " return the max fid sequence of archived files\n"
174 " --abort-on-error Abort operation on major error\n"
175 " -A, --archive <#> Archive number (repeatable)\n"
176 " -b, --bandwidth <bw> Limit I/O bandwidth (unit can be used\n,"
178 " --dry-run Don't run, just show what would be done\n"
179 " -c, --chunk-size <sz> I/O size used during data copy\n"
180 " (unit can be used, default is MB)\n"
181 " -f, --event-fifo <path> Write events stream to fifo\n"
182 " -p, --hsm-root <path> Target HSM mount point\n"
183 " -q, --quiet Produce less verbose output\n"
184 " -u, --update-interval <s> Interval between progress reports sent\n"
186 " -v, --verbose Produce more verbose output\n",
187 cmd_name, cmd_name, cmd_name, cmd_name, cmd_name);
192 static int ct_parseopts(int argc, char * const *argv)
194 struct option long_opts[] = {
195 {"abort-on-error", no_argument, &opt.o_abort_on_error, 1},
196 {"abort_on_error", no_argument, &opt.o_abort_on_error, 1},
197 {"archive", required_argument, NULL, 'A'},
198 {"bandwidth", required_argument, NULL, 'b'},
199 {"chunk-size", required_argument, NULL, 'c'},
200 {"chunk_size", required_argument, NULL, 'c'},
201 {"daemon", no_argument, &opt.o_daemonize, 1},
202 {"event-fifo", required_argument, NULL, 'f'},
203 {"event_fifo", required_argument, NULL, 'f'},
204 {"dry-run", no_argument, &opt.o_dry_run, 1},
205 {"help", no_argument, NULL, 'h'},
206 {"hsm-root", required_argument, NULL, 'p'},
207 {"hsm_root", required_argument, NULL, 'p'},
208 {"import", no_argument, NULL, 'i'},
209 {"max-sequence", no_argument, NULL, 'M'},
210 {"max_sequence", no_argument, NULL, 'M'},
211 {"no-attr", no_argument, &opt.o_copy_attrs, 0},
212 {"no_attr", no_argument, &opt.o_copy_attrs, 0},
213 {"no-shadow", no_argument, &opt.o_shadow_tree, 0},
214 {"no_shadow", no_argument, &opt.o_shadow_tree, 0},
215 {"no-xattr", no_argument, &opt.o_copy_xattrs, 0},
216 {"no_xattr", no_argument, &opt.o_copy_xattrs, 0},
217 {"quiet", no_argument, NULL, 'q'},
218 {"rebind", no_argument, NULL, 'r'},
219 {"update-interval", required_argument, NULL, 'u'},
220 {"update_interval", required_argument, NULL, 'u'},
221 {"verbose", no_argument, NULL, 'v'},
225 unsigned long long value;
226 unsigned long long unit;
229 while ((c = getopt_long(argc, argv, "A:b:c:f:hiMp:qru:v",
230 long_opts, NULL)) != -1) {
233 if ((opt.o_archive_cnt >= MAX_ARCHIVE_CNT) ||
234 (atoi(optarg) >= MAX_ARCHIVE_CNT)) {
236 CT_ERROR(rc, "archive number must be less"
237 "than %zu", MAX_ARCHIVE_CNT);
240 opt.o_archive_id[opt.o_archive_cnt] = atoi(optarg);
243 case 'b': /* -b and -c have both a number with unit as arg */
246 if (llapi_parse_size(optarg, &value, &unit, 0) < 0) {
248 CT_ERROR(rc, "bad value for -%c '%s'", c,
253 opt.o_chunk_size = value;
255 opt.o_bandwidth = value;
258 opt.o_event_fifo = optarg;
263 opt.o_action = CA_IMPORT;
266 opt.o_action = CA_MAXSEQ;
269 opt.o_hsm_root = optarg;
275 opt.o_action = CA_REBIND;
278 opt.o_report_int = atoi(optarg);
279 if (opt.o_report_int < 0) {
281 CT_ERROR(rc, "bad value for -%c '%s'", c,
296 switch (opt.o_action) {
298 /* src dst mount_point */
299 if (argc != optind + 3) {
301 CT_ERROR(rc, "--import requires 2 arguments");
304 opt.o_src = argv[optind++];
305 opt.o_dst = argv[optind++];
308 /* FID1 FID2 mount_point or FILE mount_point */
309 if (argc == optind + 2) {
310 opt.o_src = argv[optind++];
312 } else if (argc == optind + 3) {
313 opt.o_src = argv[optind++];
314 opt.o_dst = argv[optind++];
317 CT_ERROR(rc, "--rebind requires 1 or 2 arguments");
323 /* just mount point */
327 if (argc != optind + 1) {
329 CT_ERROR(rc, "no mount point specified");
333 opt.o_mnt = argv[optind];
335 CT_TRACE("action=%d src=%s dst=%s mount_point=%s",
336 opt.o_action, opt.o_src, opt.o_dst, opt.o_mnt);
338 if (!opt.o_dry_run && opt.o_hsm_root == NULL) {
340 CT_ERROR(rc, "must specify a root directory for the backend");
344 if (opt.o_action == CA_IMPORT) {
345 if (opt.o_src && opt.o_src[0] == '/') {
348 "source path must be relative to HSM root");
352 if (opt.o_dst && opt.o_dst[0] != '/') {
354 CT_ERROR(rc, "destination path must be absolute");
363 static int ct_mkdir_p(const char *path)
373 while ((ptr = strchr(ptr, '/')) != NULL) {
375 rc = mkdir(saved, DIR_PERM);
377 if (rc < 0 && errno != EEXIST) {
379 CT_ERROR(rc, "cannot mkdir '%s'", path);
391 static int ct_save_stripe(int src_fd, const char *src, const char *dst)
393 char lov_file[PATH_MAX];
394 char lov_buf[XATTR_SIZE_MAX];
395 struct lov_user_md *lum;
400 snprintf(lov_file, sizeof(lov_file), "%s.lov", dst);
401 CT_TRACE("saving stripe info of '%s' in %s", src, lov_file);
403 xattr_size = fgetxattr(src_fd, XATTR_LUSTRE_LOV, lov_buf,
405 if (xattr_size < 0) {
407 CT_ERROR(rc, "cannot get stripe info on '%s'", src);
411 lum = (struct lov_user_md *)lov_buf;
413 if (lum->lmm_magic == LOV_USER_MAGIC_V1 ||
414 lum->lmm_magic == LOV_USER_MAGIC_V3) {
415 /* Set stripe_offset to -1 so that it is not interpreted as a
416 * hint on restore. */
417 lum->lmm_stripe_offset = -1;
420 fd = open(lov_file, O_TRUNC | O_CREAT | O_WRONLY, FILE_PERM);
423 CT_ERROR(rc, "cannot open '%s'", lov_file);
427 rc = write(fd, lum, xattr_size);
430 CT_ERROR(rc, "cannot write %zd bytes to '%s'",
431 xattr_size, lov_file);
439 CT_ERROR(rc, "cannot close '%s'", lov_file);
451 static int ct_load_stripe(const char *src, void *lovea, size_t *lovea_size)
453 char lov_file[PATH_MAX];
457 snprintf(lov_file, sizeof(lov_file), "%s.lov", src);
458 CT_TRACE("reading stripe rules from '%s' for '%s'", lov_file, src);
460 fd = open(lov_file, O_RDONLY);
462 CT_ERROR(errno, "cannot open '%s'", lov_file);
466 rc = read(fd, lovea, *lovea_size);
468 CT_ERROR(errno, "cannot read %zu bytes from '%s'",
469 *lovea_size, lov_file);
480 static int ct_restore_stripe(const char *src, const char *dst, int dst_fd,
481 const void *lovea, size_t lovea_size)
485 rc = fsetxattr(dst_fd, XATTR_LUSTRE_LOV, lovea, lovea_size,
489 CT_ERROR(rc, "cannot set lov EA on '%s'", dst);
495 static void bandwidth_ctl_delay(int wsize)
497 static unsigned long long tot_bytes;
498 static time_t start_time;
499 static time_t last_time;
500 time_t now = time(0);
503 unsigned int sleep_time;
505 if (now > last_time + 5) {
507 start_time = last_time = now;
511 tot_time = now - start_time;
515 excess = tot_bytes - tot_time * opt.o_bandwidth;
516 sleep_time = excess * 1000000 / opt.o_bandwidth;
517 if ((now - start_time) % 10 == 1)
518 CT_TRACE("bandwith control: excess=%E sleep for %dus", excess,
527 static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
528 const char *dst, int src_fd, int dst_fd,
529 const struct hsm_action_item *hai, long hal_flags)
531 struct hsm_extent he;
532 __u64 offset = hai->hai_extent.offset;
536 __u64 write_total = 0;
538 time_t last_print_time = time(NULL);
541 CT_TRACE("going to copy data from '%s' to '%s'", src, dst);
543 buf = malloc(opt.o_chunk_size);
547 if (fstat(src_fd, &src_st) < 0) {
549 CT_ERROR(rc, "cannot stat '%s'", src);
553 if (!S_ISREG(src_st.st_mode)) {
555 CT_ERROR(rc, "'%s' is not a regular file", src);
559 if (fstat(dst_fd, &dst_st) < 0) {
561 CT_ERROR(rc, "cannot stat '%s'", dst);
565 if (!S_ISREG(dst_st.st_mode)) {
567 CT_ERROR(rc, "'%s' is not a regular file", dst);
571 rc = lseek(src_fd, hai->hai_extent.offset, SEEK_SET);
574 "cannot seek for read to "LPU64" (len %jd) in '%s'",
575 hai->hai_extent.offset, (intmax_t)src_st.st_size, src);
580 /* Don't read beyond a given extent */
581 length = min(hai->hai_extent.length, src_st.st_size);
585 rc = llapi_hsm_action_progress(hcp, &he, length, 0);
587 /* Action has been canceled or something wrong
588 * is happening. Stop copying data. */
589 CT_ERROR(rc, "progress ioctl for copy '%s'->'%s' failed",
596 CT_DEBUG("Going to copy "LPU64" bytes %s -> %s\n", length, src, dst);
598 while (write_total < length) {
601 int chunk = (length - write_total > opt.o_chunk_size) ?
602 opt.o_chunk_size : length - write_total;
604 rsize = pread(src_fd, buf, chunk, offset);
611 CT_ERROR(rc, "cannot read from '%s'", src);
615 wsize = pwrite(dst_fd, buf, rsize, offset);
618 CT_ERROR(rc, "cannot write to '%s'", dst);
622 write_total += wsize;
625 if (opt.o_bandwidth != 0)
626 /* sleep if needed, to honor bandwidth limits */
627 bandwidth_ctl_delay(wsize);
629 if (time(0) >= last_print_time + opt.o_report_int) {
630 last_print_time = time(0);
631 CT_TRACE("%%"LPU64" ", 100 * write_total / length);
632 he.length = write_total;
633 rc = llapi_hsm_action_progress(hcp, &he, length, 0);
635 /* Action has been canceled or something wrong
636 * is happening. Stop copying data. */
637 CT_ERROR(rc, "progress ioctl for copy"
638 " '%s'->'%s' failed", src, dst);
647 * truncate restored file
648 * size is taken from the archive this is done to support
649 * restore after a force release which leaves the file with the
650 * wrong size (can big bigger than the new size)
652 if ((hai->hai_action == HSMA_RESTORE) &&
653 (src_st.st_size < dst_st.st_size)) {
655 * make sure the file is on disk before reporting success.
657 rc = ftruncate(dst_fd, src_st.st_size);
661 "cannot truncate '%s' to size %jd",
662 dst, (intmax_t)src_st.st_size);
672 /* Copy file attributes from file src to file dest */
673 static int ct_copy_attr(const char *src, const char *dst, int src_fd,
677 struct timeval times[2];
680 if (fstat(src_fd, &st) < 0) {
682 CT_ERROR(rc, "cannot stat '%s'", src);
686 times[0].tv_sec = st.st_atime;
687 times[0].tv_usec = 0;
688 times[1].tv_sec = st.st_mtime;
689 times[1].tv_usec = 0;
690 if (fchmod(dst_fd, st.st_mode) < 0 ||
691 fchown(dst_fd, st.st_uid, st.st_gid) < 0 ||
692 futimes(dst_fd, times) < 0) {
694 CT_ERROR(rc, "cannot set attributes of '%s'", src);
701 static int ct_copy_xattr(const char *src, const char *dst, int src_fd,
702 int dst_fd, bool is_restore)
704 char list[XATTR_LIST_MAX];
705 char value[XATTR_SIZE_MAX];
710 list_len = flistxattr(src_fd, list, sizeof(list));
715 while (name < list + list_len) {
716 rc = fgetxattr(src_fd, name, value, sizeof(value));
720 /* when we restore, we do not restore lustre xattr */
722 (strncmp(XATTR_TRUSTED_PREFIX, name,
723 sizeof(XATTR_TRUSTED_PREFIX) - 1) != 0)) {
724 rc = fsetxattr(dst_fd, name, value, rc, 0);
725 CT_TRACE("fsetxattr of '%s' on '%s' rc=%d (%s)",
726 name, dst, rc, strerror(errno));
727 /* lustre.* attrs aren't supported on other FS's */
728 if (rc < 0 && errno != EOPNOTSUPP) {
730 CT_ERROR(rc, "cannot set extended attribute"
736 name += strlen(name) + 1;
742 static int ct_path_lustre(char *buf, int sz, const char *mnt,
743 const lustre_fid *fid)
745 return snprintf(buf, sz, "%s/%s/fid/"DFID_NOBRACE, mnt,
746 dot_lustre_name, PFID(fid));
749 static int ct_path_archive(char *buf, int sz, const char *archive_dir,
750 const lustre_fid *fid)
752 return snprintf(buf, sz, "%s/%04x/%04x/%04x/%04x/%04x/%04x/"
753 DFID_NOBRACE, archive_dir,
754 (fid)->f_oid & 0xFFFF,
755 (fid)->f_oid >> 16 & 0xFFFF,
756 (unsigned int)((fid)->f_seq & 0xFFFF),
757 (unsigned int)((fid)->f_seq >> 16 & 0xFFFF),
758 (unsigned int)((fid)->f_seq >> 32 & 0xFFFF),
759 (unsigned int)((fid)->f_seq >> 48 & 0xFFFF),
763 static bool ct_is_retryable(int err)
765 return err == -ETIMEDOUT;
768 static int ct_begin_restore(struct hsm_copyaction_private **phcp,
769 const struct hsm_action_item *hai,
770 int mdt_index, int open_flags)
775 rc = llapi_hsm_action_begin(phcp, ctdata, hai, mdt_index, open_flags,
778 ct_path_lustre(src, sizeof(src), opt.o_mnt, &hai->hai_fid);
779 CT_ERROR(rc, "llapi_hsm_action_begin() on '%s' failed", src);
785 static int ct_begin(struct hsm_copyaction_private **phcp,
786 const struct hsm_action_item *hai)
788 /* Restore takes specific parameters. Call the same function w/ default
789 * values for all other operations. */
790 return ct_begin_restore(phcp, hai, -1, 0);
793 static int ct_fini(struct hsm_copyaction_private **phcp,
794 const struct hsm_action_item *hai, int hp_flags, int ct_rc)
796 struct hsm_copyaction_private *hcp;
800 CT_TRACE("Action completed, notifying coordinator "
801 "cookie="LPX64", FID="DFID", hp_flags=%d err=%d",
802 hai->hai_cookie, PFID(&hai->hai_fid),
805 ct_path_lustre(lstr, sizeof(lstr), opt.o_mnt, &hai->hai_fid);
807 if (phcp == NULL || *phcp == NULL) {
808 rc = llapi_hsm_action_begin(&hcp, ctdata, hai, -1, 0, true);
810 CT_ERROR(rc, "llapi_hsm_action_begin() on '%s' failed",
817 rc = llapi_hsm_action_end(phcp, &hai->hai_extent, hp_flags, abs(ct_rc));
818 if (rc == -ECANCELED)
819 CT_ERROR(rc, "completed action on '%s' has been canceled: "
820 "cookie="LPX64", FID="DFID, lstr, hai->hai_cookie,
821 PFID(&hai->hai_fid));
823 CT_ERROR(rc, "llapi_hsm_action_end() on '%s' failed", lstr);
825 CT_TRACE("llapi_hsm_action_end() on '%s' ok (rc=%d)",
831 static int ct_archive(const struct hsm_action_item *hai, const long hal_flags)
833 struct hsm_copyaction_private *hcp = NULL;
835 char dst[PATH_MAX] = "";
838 bool rename_needed = false;
844 rc = ct_begin(&hcp, hai);
848 /* we fill archive so:
850 * destination = lustre FID
852 ct_path_lustre(src, sizeof(src), opt.o_mnt, &hai->hai_dfid);
853 ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, &hai->hai_fid);
854 if (hai->hai_extent.length == -1) {
855 /* whole file, write it to tmp location and atomically
856 * replace old archived file */
857 strncat(dst, "_tmp", sizeof(dst) - strlen(dst) - 1);
858 /* we cannot rely on the same test because ct_copy_data()
859 * updates hai_extent.length */
860 rename_needed = true;
863 CT_TRACE("archiving '%s' to '%s'", src, dst);
870 rc = ct_mkdir_p(dst);
872 CT_ERROR(rc, "mkdir_p '%s' failed", dst);
876 src_fd = llapi_hsm_action_get_fd(hcp);
879 CT_ERROR(rc, "cannot open '%s' for read", src);
883 open_flags = O_WRONLY | O_NOFOLLOW | O_NONBLOCK;
884 /* If extent is specified, don't truncate an old archived copy */
885 open_flags |= ((hai->hai_extent.length == -1) ? O_TRUNC : 0) | O_CREAT;
887 dst_fd = open(dst, open_flags, FILE_PERM);
890 CT_ERROR(rc, "cannot open '%s' for write", dst);
894 /* saving stripe is not critical */
895 rc = ct_save_stripe(src_fd, src, dst);
897 CT_ERROR(rc, "cannot save file striping info of '%s' in '%s'",
900 rc = ct_copy_data(hcp, src, dst, src_fd, dst_fd, hai, hal_flags);
902 CT_ERROR(rc, "data copy failed from '%s' to '%s'", src, dst);
909 CT_ERROR(rc, "cannot flush '%s' archive file '%s'", src, dst);
913 CT_TRACE("data archiving for '%s' to '%s' done", src, dst);
915 /* attrs will remain on the MDS; no need to copy them, except possibly
916 for disaster recovery */
917 if (opt.o_copy_attrs) {
918 rc = ct_copy_attr(src, dst, src_fd, dst_fd);
920 CT_ERROR(rc, "cannot copy attr of '%s' to '%s'",
924 CT_TRACE("attr file for '%s' saved to archive '%s'",
928 /* xattrs will remain on the MDS; no need to copy them, except possibly
929 for disaster recovery */
930 if (opt.o_copy_xattrs) {
931 rc = ct_copy_xattr(src, dst, src_fd, dst_fd, false);
933 CT_ERROR(rc, "cannot copy xattr of '%s' to '%s'",
935 rcf = rcf ? rcf : rc;
937 CT_TRACE("xattr file for '%s' saved to archive '%s'",
941 if (rename_needed == true) {
942 char tmp_src[PATH_MAX];
943 char tmp_dst[PATH_MAX];
945 /* atomically replace old archived file */
946 ct_path_archive(src, sizeof(src), opt.o_hsm_root,
948 rc = rename(dst, src);
951 CT_ERROR(rc, "cannot rename '%s' to '%s'", dst, src);
954 /* rename lov file */
955 snprintf(tmp_src, sizeof(tmp_src), "%s.lov", src);
956 snprintf(tmp_dst, sizeof(tmp_dst), "%s.lov", dst);
957 rc = rename(tmp_dst, tmp_src);
959 CT_ERROR(errno, "cannot rename '%s' to '%s'",
963 if (opt.o_shadow_tree) {
964 /* Create a namespace of softlinks that shadows the original
965 * Lustre namespace. This will only be current at
966 * time-of-archive (won't follow renames).
967 * WARNING: release won't kill these links; a manual
968 * cleanup of dead links would be required.
971 long long recno = -1;
977 sprintf(buf, DFID, PFID(&hai->hai_fid));
978 sprintf(src, "%s/shadow/", opt.o_hsm_root);
980 ptr = opt.o_hsm_root;
982 (*ptr++ == '/') ? depth-- : 0;
984 rc = llapi_fid2path(opt.o_mnt, buf, src + strlen(src),
985 sizeof(src) - strlen(src), &recno, &linkno);
987 CT_ERROR(rc, "cannot get FID of '%s'", buf);
988 rcf = rcf ? rcf : rc;
992 /* Figure out how many parent dirs to symlink back */
995 (*ptr++ == '/') ? depth++ : 0;
1000 ct_path_archive(dst, sizeof(dst), buf, &hai->hai_fid);
1002 if (ct_mkdir_p(src)) {
1003 CT_ERROR(errno, "mkdir_p '%s' failed", src);
1004 rcf = rcf ? rcf : -errno;
1007 /* symlink already exists ? */
1008 sz = readlink(src, buf, sizeof(buf));
1011 if (sz == 0 || strncmp(buf, dst, sz) != 0) {
1012 if (unlink(src) && errno != ENOENT) {
1014 "cannot unlink symlink '%s'",
1016 rcf = rcf ? rcf : -errno;
1018 /* unlink old symlink done */
1019 CT_TRACE("remove old symlink '%s' pointing"
1020 " to '%s'", src, buf);
1023 /* symlink already ok */
1024 CT_TRACE("symlink '%s' already pointing"
1025 " to '%s'", src, dst);
1030 if (symlink(dst, src)) {
1031 CT_ERROR(errno, "cannot symlink '%s' to '%s'",
1033 rcf = rcf ? rcf : -errno;
1036 CT_TRACE("symlink '%s' to '%s' done", src, dst);
1048 if (ct_is_retryable(rc))
1049 hp_flags |= HP_FLAG_RETRY;
1060 rc = ct_fini(&hcp, hai, hp_flags, rcf);
1065 static int ct_restore(const struct hsm_action_item *hai, const long hal_flags)
1067 struct hsm_copyaction_private *hcp = NULL;
1070 char lov_buf[XATTR_SIZE_MAX];
1071 size_t lov_size = sizeof(lov_buf);
1076 int mdt_index = -1; /* Not implemented */
1080 /* we fill lustre so:
1081 * source = lustre FID in the backend
1082 * destination = data FID = volatile file
1085 /* build backend file name from released file FID */
1086 ct_path_archive(src, sizeof(src), opt.o_hsm_root, &hai->hai_fid);
1088 /* restore loads and sets the LOVEA w/o interpreting it to avoid
1089 * dependency on the structure format. */
1090 rc = ct_load_stripe(src, lov_buf, &lov_size);
1092 CT_WARN("cannot get stripe rules for '%s' (%s), use default",
1093 src, strerror(-rc));
1096 open_flags |= O_LOV_DELAY_CREATE;
1100 rc = ct_begin_restore(&hcp, hai, mdt_index, open_flags);
1104 /* get the FID of the volatile file */
1105 rc = llapi_hsm_action_get_dfid(hcp, &dfid);
1107 CT_ERROR(rc, "restoring "DFID
1108 ", cannot get FID of created volatile file",
1109 PFID(&hai->hai_fid));
1113 /* build volatile "file name", for messages */
1114 snprintf(dst, sizeof(dst), "{VOLATILE}="DFID, PFID(&dfid));
1116 CT_TRACE("restoring data from '%s' to '%s'", src, dst);
1118 if (opt.o_dry_run) {
1123 src_fd = open(src, O_RDONLY | O_NOATIME | O_NONBLOCK | O_NOFOLLOW);
1126 CT_ERROR(rc, "cannot open '%s' for read", src);
1130 dst_fd = llapi_hsm_action_get_fd(hcp);
1133 CT_ERROR(rc, "cannot open '%s' for write", dst);
1138 /* the layout cannot be allocated through .fid so we have to
1139 * restore a layout */
1140 rc = ct_restore_stripe(src, dst, dst_fd, lov_buf, lov_size);
1142 CT_ERROR(rc, "cannot restore file striping info"
1143 " for '%s' from '%s'", dst, src);
1149 rc = ct_copy_data(hcp, src, dst, src_fd, dst_fd, hai, hal_flags);
1151 CT_ERROR(rc, "cannot copy data from '%s' to '%s'",
1154 if (ct_is_retryable(rc))
1155 hp_flags |= HP_FLAG_RETRY;
1159 CT_TRACE("data restore from '%s' to '%s' done", src, dst);
1162 rc = ct_fini(&hcp, hai, hp_flags, rc);
1164 /* object swaping is done by cdt at copy end, so close of volatile file
1165 * cannot be done before */
1175 static int ct_remove(const struct hsm_action_item *hai, const long hal_flags)
1177 struct hsm_copyaction_private *hcp = NULL;
1181 rc = ct_begin(&hcp, hai);
1185 ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, &hai->hai_fid);
1187 CT_TRACE("removing file '%s'", dst);
1189 if (opt.o_dry_run) {
1197 CT_ERROR(rc, "cannot unlink '%s'", dst);
1202 strncat(dst, ".lov", sizeof(dst) - strlen(dst) - 1);
1206 CT_ERROR(rc, "cannot unlink '%s'", dst);
1212 rc = ct_fini(&hcp, hai, 0, rc);
1217 static int ct_process_item(struct hsm_action_item *hai, const long hal_flags)
1221 if (opt.o_verbose >= LLAPI_MSG_INFO || opt.o_dry_run) {
1222 /* Print the original path */
1224 char path[PATH_MAX];
1225 long long recno = -1;
1228 sprintf(fid, DFID, PFID(&hai->hai_fid));
1229 CT_TRACE("'%s' action %s reclen %d, cookie="LPX64,
1230 fid, hsm_copytool_action2name(hai->hai_action),
1231 hai->hai_len, hai->hai_cookie);
1232 rc = llapi_fid2path(opt.o_mnt, fid, path,
1233 sizeof(path), &recno, &linkno);
1235 CT_ERROR(rc, "cannot get path of FID %s", fid);
1237 CT_TRACE("processing file '%s'", path);
1240 switch (hai->hai_action) {
1241 /* set err_major, minor inside these functions */
1243 rc = ct_archive(hai, hal_flags);
1246 rc = ct_restore(hai, hal_flags);
1249 rc = ct_remove(hai, hal_flags);
1252 CT_TRACE("cancel not implemented for file system '%s'",
1254 /* Don't report progress to coordinator for this cookie:
1255 * the copy function will get ECANCELED when reporting
1262 CT_ERROR(rc, "unknown action %d, on '%s'", hai->hai_action,
1265 ct_fini(NULL, hai, 0, rc);
1273 struct hsm_action_item *hai;
1276 static void *ct_thread(void *data)
1278 struct ct_th_data *cttd = data;
1281 rc = ct_process_item(cttd->hai, cttd->hal_flags);
1285 pthread_exit((void *)(intptr_t)rc);
1288 static int ct_process_item_async(const struct hsm_action_item *hai,
1291 pthread_attr_t attr;
1293 struct ct_th_data *data;
1296 data = malloc(sizeof(*data));
1300 data->hai = malloc(hai->hai_len);
1301 if (data->hai == NULL) {
1306 memcpy(data->hai, hai, hai->hai_len);
1307 data->hal_flags = hal_flags;
1309 rc = pthread_attr_init(&attr);
1311 CT_ERROR(rc, "pthread_attr_init failed for '%s' service",
1318 pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1320 rc = pthread_create(&thread, &attr, ct_thread, data);
1322 CT_ERROR(rc, "cannot create thread for '%s' service",
1325 pthread_attr_destroy(&attr);
1329 static int ct_import_one(const char *src, const char *dst)
1331 char newarc[PATH_MAX];
1336 CT_TRACE("importing '%s' from '%s'", dst, src);
1338 if (stat(src, &st) < 0) {
1340 CT_ERROR(rc, "cannot stat '%s'", src);
1347 rc = llapi_hsm_import(dst,
1348 opt.o_archive_cnt ? opt.o_archive_id[0] : 0,
1349 &st, 0, 0, 0, 0, NULL, &fid);
1351 CT_ERROR(rc, "cannot import '%s' from '%s'", dst, src);
1355 ct_path_archive(newarc, sizeof(newarc), opt.o_hsm_root, &fid);
1357 rc = ct_mkdir_p(newarc);
1359 CT_ERROR(rc, "mkdir_p '%s' failed", newarc);
1365 /* Lots of choices now: mv, ln, ln -s ? */
1366 rc = link(src, newarc); /* hardlink */
1369 CT_ERROR(rc, "cannot link '%s' to '%s'", newarc, src);
1373 CT_TRACE("imported '%s' from '%s'=='%s'", dst, newarc, src);
1378 static char *path_concat(const char *dirname, const char *basename)
1383 rc = asprintf(&result, "%s/%s", dirname, basename);
1390 static int ct_import_fid(const lustre_fid *import_fid)
1392 char fid_path[PATH_MAX];
1395 ct_path_lustre(fid_path, sizeof(fid_path), opt.o_mnt, import_fid);
1396 rc = access(fid_path, F_OK);
1397 if (rc == 0 || errno != ENOENT) {
1398 rc = (errno == 0) ? -EEXIST : -errno;
1399 CT_ERROR(rc, "cannot import '"DFID"'", PFID(import_fid));
1403 ct_path_archive(fid_path, sizeof(fid_path), opt.o_hsm_root,
1406 CT_TRACE("Resolving "DFID" to %s", PFID(import_fid), fid_path);
1408 return ct_import_one(fid_path, opt.o_dst);
1411 static int ct_import_recurse(const char *relpath)
1414 struct dirent ent, *cookie = NULL;
1415 char *srcpath, *newpath;
1416 lustre_fid import_fid;
1419 if (relpath == NULL)
1422 /* Is relpath a FID? In which case SFID should expand to three
1424 rc = sscanf(relpath, SFID, RFID(&import_fid));
1426 return ct_import_fid(&import_fid);
1428 srcpath = path_concat(opt.o_hsm_root, relpath);
1429 if (srcpath == NULL) {
1434 dir = opendir(srcpath);
1436 /* Not a dir, or error */
1437 if (errno == ENOTDIR) {
1438 /* Single regular file case, treat o_dst as absolute
1440 rc = ct_import_one(srcpath, opt.o_dst);
1443 CT_ERROR(rc, "cannot opendir '%s'", srcpath);
1452 rc = readdir_r(dir, &ent, &cookie);
1455 CT_ERROR(rc, "cannot readdir_r '%s'", relpath);
1458 } else if ((rc == 0) && (cookie == NULL)) {
1459 /* end of directory */
1463 if (!strcmp(ent.d_name, ".") ||
1464 !strcmp(ent.d_name, ".."))
1467 /* New relative path */
1468 newpath = path_concat(relpath, ent.d_name);
1469 if (newpath == NULL) {
1475 if (ent.d_type == DT_DIR) {
1476 rc = ct_import_recurse(newpath);
1481 sprintf(src, "%s/%s", opt.o_hsm_root, newpath);
1482 sprintf(dst, "%s/%s", opt.o_dst, newpath);
1483 /* Make the target dir in the Lustre fs */
1484 rc = ct_mkdir_p(dst);
1486 /* Import the file */
1487 rc = ct_import_one(src, dst);
1489 CT_ERROR(rc, "ct_mkdir_p '%s' failed", dst);
1495 CT_ERROR(rc, "cannot import '%s'", newpath);
1496 if (err_major && opt.o_abort_on_error) {
1510 static int ct_rebind_one(const lustre_fid *old_fid, const lustre_fid *new_fid)
1516 CT_TRACE("rebind "DFID" to "DFID, PFID(old_fid), PFID(new_fid));
1518 ct_path_archive(src, sizeof(src), opt.o_hsm_root, old_fid);
1519 ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, new_fid);
1521 if (!opt.o_dry_run) {
1523 if (rename(src, dst)) {
1525 CT_ERROR(rc, "cannot rename '%s' to '%s'", src, dst);
1528 /* rename lov file */
1529 strncat(src, ".lov", sizeof(src) - strlen(src) - 1);
1530 strncat(dst, ".lov", sizeof(dst) - strlen(dst) - 1);
1531 if (rename(src, dst))
1532 CT_ERROR(errno, "cannot rename '%s' to '%s'", src, dst);
1538 static bool fid_is_file(lustre_fid *fid)
1540 return fid_is_norm(fid) || fid_is_igif(fid);
1543 static bool should_ignore_line(const char *line)
1547 for (i = 0; line[i] != '\0'; i++) {
1548 if (isspace(line[i]))
1550 else if (line[i] == '#')
1559 static int ct_rebind_list(const char *list)
1565 size_t line_size = 0;
1566 unsigned int nl = 0;
1567 unsigned int ok = 0;
1569 filp = fopen(list, "r");
1572 CT_ERROR(rc, "cannot open '%s'", list);
1576 /* each line consists of 2 FID */
1577 while ((r = getline(&line, &line_size, filp)) != -1) {
1581 /* Ignore empty and commented out ('#...') lines. */
1582 if (should_ignore_line(line))
1587 rc = sscanf(line, SFID" "SFID, RFID(&old_fid), RFID(&new_fid));
1588 if (rc != 6 || !fid_is_file(&old_fid) ||
1589 !fid_is_file(&new_fid)) {
1591 "'%s' FID expected near '%s', line %u",
1597 if (ct_rebind_one(&old_fid, &new_fid))
1608 /* return 0 if all rebinds were sucessful */
1609 CT_TRACE("%u lines read from '%s', %u rebind successful", nl, list, ok);
1611 return ok == nl ? 0 : -1;
1614 static int ct_rebind(void)
1622 if (sscanf(opt.o_src, SFID, RFID(&old_fid)) != 3 ||
1623 !fid_is_file(&old_fid)) {
1625 CT_ERROR(rc, "'%s' invalid FID format", opt.o_src);
1629 if (sscanf(opt.o_dst, SFID, RFID(&new_fid)) != 3 ||
1630 !fid_is_file(&new_fid)) {
1632 CT_ERROR(rc, "'%s' invalid FID format", opt.o_dst);
1636 rc = ct_rebind_one(&old_fid, &new_fid);
1641 /* o_src is a list file */
1642 rc = ct_rebind_list(opt.o_src);
1647 static int ct_dir_level_max(const char *dirpath, __u16 *sub_seqmax)
1652 struct dirent ent, *cookie = NULL;
1656 dir = opendir(dirpath);
1659 CT_ERROR(rc, "cannot open directory '%s'", opt.o_hsm_root);
1663 while ((rc = readdir_r(dir, &ent, &cookie)) == 0) {
1665 /* end of directory.
1666 * rc is 0 and seqmax contains the max value. */
1669 if (!strcmp(ent.d_name, ".") || !strcmp(ent.d_name, ".."))
1672 if (sscanf(ent.d_name, "%hx", &sub_seq) != 1) {
1673 CT_TRACE("'%s' has an unexpected dirname format, "
1674 "skip entry", ent.d_name);
1677 if (sub_seq > *sub_seqmax)
1678 *sub_seqmax = sub_seq;
1681 CT_ERROR(rc, "cannot readdir_r '%s'", dirpath);
1688 static int ct_max_sequence(void)
1691 char path[PATH_MAX];
1695 strncpy(path, opt.o_hsm_root, sizeof(path));
1696 /* FID sequence is stored in top-level directory names:
1697 * hsm_root/16bits (high weight)/16 bits/16 bits/16 bits (low weight).
1699 for (i = 0; i < 4; i++) {
1700 rc = ct_dir_level_max(path, &subseq);
1703 seq |= ((__u64)subseq << ((3 - i) * 16));
1704 sprintf(path + strlen(path), "/%04x", subseq);
1707 printf("max_sequence: %016Lx\n", seq);
1712 static void handler(int signal)
1714 psignal(signal, "exiting");
1715 /* If we don't clean up upon interrupt, umount thinks there's a ref
1716 * and doesn't remove us from mtab (EINPROGRESS). The lustre client
1717 * does successfully unmount and the mount is actually gone, but the
1718 * mtab entry remains. So this just makes mtab happier. */
1719 llapi_hsm_copytool_unregister(&ctdata);
1723 /* Daemon waits for messages from the kernel; run it in the background. */
1724 static int ct_run(void)
1728 if (opt.o_daemonize) {
1732 CT_ERROR(rc, "cannot daemonize");
1737 setbuf(stdout, NULL);
1739 if (opt.o_event_fifo != NULL) {
1740 rc = llapi_hsm_register_event_fifo(opt.o_event_fifo);
1742 CT_ERROR(rc, "failed to register event fifo");
1745 llapi_error_callback_set(llapi_hsm_log_error);
1748 rc = llapi_hsm_copytool_register(&ctdata, opt.o_mnt, 0,
1749 opt.o_archive_cnt, opt.o_archive_id);
1751 CT_ERROR(rc, "cannot start copytool interface");
1755 signal(SIGINT, handler);
1756 signal(SIGTERM, handler);
1759 struct hsm_action_list *hal;
1760 struct hsm_action_item *hai;
1764 CT_TRACE("waiting for message from kernel");
1766 rc = llapi_hsm_copytool_recv(ctdata, &hal, &msgsize);
1767 if (rc == -ESHUTDOWN) {
1768 CT_TRACE("shutting down");
1770 } else if (rc == -EAGAIN) {
1771 continue; /* msg not for us */
1772 } else if (rc < 0) {
1773 CT_WARN("cannot receive action list: %s",
1776 if (opt.o_abort_on_error)
1782 CT_TRACE("copytool fs=%s archive#=%d item_count=%d",
1783 hal->hal_fsname, hal->hal_archive_id, hal->hal_count);
1785 if (strcmp(hal->hal_fsname, fs_name) != 0) {
1787 CT_ERROR(rc, "'%s' invalid fs name, expecting: %s",
1788 hal->hal_fsname, fs_name);
1790 if (opt.o_abort_on_error)
1796 hai = hai_first(hal);
1797 while (++i <= hal->hal_count) {
1798 if ((char *)hai - (char *)hal > msgsize) {
1801 "'%s' item %d past end of message!",
1806 rc = ct_process_item_async(hai, hal->hal_flags);
1808 CT_ERROR(rc, "'%s' item %d process",
1810 if (opt.o_abort_on_error && err_major)
1812 hai = hai_next(hai);
1815 llapi_hsm_action_list_free(&hal);
1817 if (opt.o_abort_on_error && err_major)
1821 llapi_hsm_copytool_unregister(&ctdata);
1822 if (opt.o_event_fifo != NULL)
1823 llapi_hsm_unregister_event_fifo(opt.o_event_fifo);
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);
1883 switch (opt.o_action) {
1885 rc = ct_import_recurse(opt.o_src);
1891 rc = ct_max_sequence();
1898 if (opt.o_action != CA_MAXSEQ)
1899 CT_TRACE("process finished, errs: %d major, %d minor,"
1900 " rc=%d (%s)", err_major, err_minor, rc,