Whamcloud - gitweb
LU-4287 kernel: kernel update RHEL6.5 [2.6.32-431.3.1.el6]
[fs/lustre-release.git] / lustre / utils / lhsmtool_posix.c
1 /*
2  * GPL HEADER START
3  *
4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5  *
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.
9  *
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).
15  *
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
19  *
20  * GPL HEADER END
21  */
22 /*
23  * (C) Copyright 2012 Commissariat a l'energie atomique et aux energies
24  *     alternatives
25  *
26  */
27 /* HSM copytool program for POSIX filesystem-based HSM's.
28  *
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.
32  *
33  * This particular tool can also import an existing HSM archive.
34  */
35 #include <stdint.h>
36 #include <stdlib.h>
37 #include <dirent.h>
38 #include <errno.h>
39 #include <utime.h>
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>
45
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)
52
53 #define ONE_MB 0x100000
54
55 /* copytool uses a 32b bitmask field to register with kuc
56  * archive num = 0 => all
57  * archive num from 1 to 32
58  */
59 #define MAX_ARCHIVE_CNT (sizeof(__u32) * 8)
60
61 enum ct_action {
62         CA_IMPORT = 1,
63         CA_REBIND,
64         CA_MAXSEQ,
65 };
66
67 struct options {
68         int                      o_copy_attrs;
69         int                      o_daemonize;
70         int                      o_dry_run;
71         int                      o_abort_on_error;
72         int                      o_shadow_tree;
73         int                      o_verbose;
74         int                      o_copy_xattrs;
75         int                      o_archive_cnt;
76         int                      o_archive_id[MAX_ARCHIVE_CNT];
77         int                      o_report_int;
78         unsigned long long       o_bandwidth;
79         size_t                   o_chunk_size;
80         enum ct_action           o_action;
81         char                    *o_mnt;
82         char                    *o_hsm_root;
83         char                    *o_src; /* for import, or rebind */
84         char                    *o_dst; /* for import, or rebind */
85 };
86
87 /* everything else is zeroed */
88 struct options opt = {
89         .o_copy_attrs = 1,
90         .o_shadow_tree = 1,
91         .o_verbose = LLAPI_MSG_INFO,
92         .o_copy_xattrs = 1,
93         .o_report_int = REPORT_INTERVAL_DEFAULT,
94         .o_chunk_size = ONE_MB,
95 };
96
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). */
101
102 static int arc_fd = -1;
103
104 static int err_major;
105 static int err_minor;
106
107 static char cmd_name[PATH_MAX];
108 static char fs_name[MAX_OBD_NAME + 1];
109
110 static struct hsm_copytool_private *ctdata;
111
112
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__)
133
134 static void usage(const char *name, int rc)
135 {
136         fprintf(stdout,
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"
145         " Options:\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"
153         " Usage:\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,"
176         "                            default is MB)\n",
177         cmd_name, cmd_name, cmd_name, cmd_name, cmd_name);
178
179         exit(rc);
180 }
181
182 static int ct_parseopts(int argc, char * const *argv)
183 {
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'},
209                 {0, 0, 0, 0}
210         };
211         int                      c, rc;
212         unsigned long long       value;
213         unsigned long long       unit;
214
215         optind = 0;
216         while ((c = getopt_long(argc, argv, "A:b:c:hiMp:qrv",
217                                 long_opts, NULL)) != -1) {
218                 switch (c) {
219                 case 'A':
220                         if ((opt.o_archive_cnt >= MAX_ARCHIVE_CNT) ||
221                             (atoi(optarg) >= MAX_ARCHIVE_CNT)) {
222                                 rc = -E2BIG;
223                                 CT_ERROR(rc, "archive number must be less"
224                                          "than %zu", MAX_ARCHIVE_CNT);
225                                 return rc;
226                         }
227                         opt.o_archive_id[opt.o_archive_cnt] = atoi(optarg);
228                         opt.o_archive_cnt++;
229                         break;
230                 case 'b': /* -b and -c have both a number with unit as arg */
231                 case 'c':
232                         unit = ONE_MB;
233                         if (llapi_parse_size(optarg, &value, &unit, 0) < 0) {
234                                 rc = -EINVAL;
235                                 CT_ERROR(rc, "bad value for -%c '%s'", c,
236                                          optarg);
237                                 return rc;
238                         }
239                         if (c == 'c')
240                                 opt.o_chunk_size = value;
241                         else
242                                 opt.o_bandwidth = value;
243                         break;
244                 case 'h':
245                         usage(argv[0], 0);
246                 case 'i':
247                         opt.o_action = CA_IMPORT;
248                         break;
249                 case 'M':
250                         opt.o_action = CA_MAXSEQ;
251                         break;
252                 case 'p':
253                         opt.o_hsm_root = optarg;
254                         break;
255                 case 'q':
256                         opt.o_verbose--;
257                         break;
258                 case 'r':
259                         opt.o_action = CA_REBIND;
260                         break;
261                 case 'v':
262                         opt.o_verbose++;
263                         break;
264                 case 0:
265                         break;
266                 default:
267                         return -EINVAL;
268                 }
269         }
270
271         switch (opt.o_action) {
272         case CA_IMPORT:
273                 /* src dst mount_point */
274                 if (argc != optind + 3) {
275                         rc = -EINVAL;
276                         CT_ERROR(rc, "--import requires 2 arguments");
277                         return rc;
278                 }
279                 opt.o_src = argv[optind++];
280                 opt.o_dst = argv[optind++];
281                 break;
282         case CA_REBIND:
283                 /* FID1 FID2 mount_point or FILE mount_point */
284                 if (argc == optind + 2) {
285                         opt.o_src = argv[optind++];
286                         opt.o_dst = NULL;
287                 } else if (argc == optind + 3) {
288                         opt.o_src = argv[optind++];
289                         opt.o_dst = argv[optind++];
290                 } else {
291                         rc = -EINVAL;
292                         CT_ERROR(rc, "--rebind requires 1 or 2 arguments");
293                         return rc;
294                 }
295                 break;
296         case CA_MAXSEQ:
297         default:
298                 /* just mount point */
299                 break;
300         }
301
302         if (argc != optind + 1) {
303                 rc = -EINVAL;
304                 CT_ERROR(rc, "no mount point specified");
305                 return rc;
306         }
307
308         opt.o_mnt = argv[optind];
309
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);
312
313         if (!opt.o_dry_run && opt.o_hsm_root == NULL) {
314                 rc = -EINVAL;
315                 CT_ERROR(rc, "must specify a root directory for the backend");
316                 return rc;
317         }
318
319         if (opt.o_action == CA_IMPORT) {
320                 if (opt.o_src && opt.o_src[0] == '/') {
321                         rc = -EINVAL;
322                         CT_ERROR(rc,
323                                  "source path must be relative to HSM root");
324                         return rc;
325                 }
326
327                 if (opt.o_dst && opt.o_dst[0] != '/') {
328                         rc = -EINVAL;
329                         CT_ERROR(rc, "destination path must be absolute");
330                         return rc;
331                 }
332         }
333
334         return 0;
335 }
336
337 /* mkdir -p path */
338 static int ct_mkdir_p(const char *path)
339 {
340         char    *saved, *ptr;
341         int      rc;
342
343         ptr = strdup(path);
344         saved = ptr;
345         while (*ptr == '/')
346                 ptr++;
347
348         while ((ptr = strchr(ptr, '/')) != NULL) {
349                 *ptr = '\0';
350                 rc = mkdir(saved, DIR_PERM);
351                 *ptr = '/';
352                 if (rc < 0 && errno != EEXIST) {
353                         rc = -errno;
354                         CT_ERROR(rc, "cannot mkdir '%s'", path);
355                         free(saved);
356                         return rc;
357                 }
358                 ptr++;
359         }
360
361         free(saved);
362
363         return 0;
364 }
365
366 static int ct_save_stripe(int src_fd, const char *src, const char *dst)
367 {
368         char                     lov_file[PATH_MAX];
369         char                     lov_buf[XATTR_SIZE_MAX];
370         struct lov_user_md      *lum;
371         int                      rc;
372         int                      fd;
373         ssize_t                  xattr_size;
374
375         snprintf(lov_file, sizeof(lov_file), "%s.lov", dst);
376         CT_TRACE("saving stripe info of '%s' in %s", src, lov_file);
377
378         xattr_size = fgetxattr(src_fd, XATTR_LUSTRE_LOV, lov_buf,
379                                sizeof(lov_buf));
380         if (xattr_size < 0) {
381                 rc = -errno;
382                 CT_ERROR(rc, "cannot get stripe info on '%s'", src);
383                 return rc;
384         }
385
386         lum = (struct lov_user_md *)lov_buf;
387
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;
393         }
394
395         fd = open(lov_file, O_TRUNC | O_CREAT | O_WRONLY, FILE_PERM);
396         if (fd < 0) {
397                 rc = -errno;
398                 CT_ERROR(rc, "cannot open '%s'", lov_file);
399                 goto err_cleanup;
400         }
401
402         rc = write(fd, lum, xattr_size);
403         if (rc < 0) {
404                 rc = -errno;
405                 CT_ERROR(rc, "cannot write %zd bytes to '%s'",
406                          xattr_size, lov_file);
407                 close(fd);
408                 goto err_cleanup;
409         }
410
411         rc = close(fd);
412         if (rc < 0) {
413                 rc = -errno;
414                 CT_ERROR(rc, "cannot close '%s'", lov_file);
415                 goto err_cleanup;
416         }
417
418         return 0;
419
420 err_cleanup:
421         unlink(lov_file);
422
423         return rc;
424 }
425
426 static int ct_load_stripe(const char *src, void *lovea, size_t *lovea_size)
427 {
428         char     lov_file[PATH_MAX];
429         int      rc;
430         int      fd;
431
432         snprintf(lov_file, sizeof(lov_file), "%s.lov", src);
433         CT_TRACE("reading stripe rules from '%s' for '%s'", lov_file, src);
434
435         fd = open(lov_file, O_RDONLY);
436         if (fd < 0) {
437                 CT_ERROR(errno, "cannot open '%s'", lov_file);
438                 return -ENODATA;
439         }
440
441         rc = read(fd, lovea, *lovea_size);
442         if (rc < 0) {
443                 CT_ERROR(errno, "cannot read %zu bytes from '%s'",
444                          *lovea_size, lov_file);
445                 close(fd);
446                 return -ENODATA;
447         }
448
449         *lovea_size = rc;
450         close(fd);
451
452         return 0;
453 }
454
455 static int ct_restore_stripe(const char *src, const char *dst, int dst_fd,
456                              const void *lovea, size_t lovea_size)
457 {
458         int     rc;
459
460         rc = fsetxattr(dst_fd, XATTR_LUSTRE_LOV, lovea, lovea_size,
461                        XATTR_CREATE);
462         if (rc < 0) {
463                 CT_ERROR(errno, "cannot set lov EA on '%s'", dst);
464                 rc = -errno;
465         }
466
467         return rc;
468 }
469
470 static void bandwidth_ctl_delay(int wsize)
471 {
472         static unsigned long long       tot_bytes;
473         static time_t                   start_time;
474         static time_t                   last_time;
475         time_t                          now = time(0);
476         double                          tot_time;
477         double                          excess;
478         unsigned int                    sleep_time;
479
480         if (now > last_time + 5) {
481                 tot_bytes = 0;
482                 start_time = last_time = now;
483         }
484
485         tot_bytes += wsize;
486         tot_time = now - start_time;
487         if (tot_time < 1)
488                 tot_time = 1;
489
490         excess = tot_bytes - tot_time * opt.o_bandwidth;
491         sleep_time = excess * 1000000 / opt.o_bandwidth;
492         if ((now - start_time) % 10 == 1)
493                 CT_TRACE("bandwith control: excess=%E sleep for %dus", excess,
494                          sleep_time);
495
496         if (excess > 0)
497                 usleep(sleep_time);
498
499         last_time = now;
500 }
501
502 static int ct_copy_data(struct hsm_copyaction_private *hcp, const char *src,
503                         const char *dst, int src_fd, int dst_fd,
504                         const struct hsm_action_item *hai, long hal_flags)
505 {
506         struct hsm_extent        he;
507         __u64                    offset = hai->hai_extent.offset;
508         struct stat              src_st;
509         struct stat              dst_st;
510         char                    *buf = NULL;
511         __u64                    write_total = 0;
512         __u64                    length;
513         time_t                   last_print_time = time(NULL);
514         int                      rc = 0;
515
516         CT_TRACE("going to copy data from '%s' to '%s'", src, dst);
517
518         if (fstat(src_fd, &src_st) < 0) {
519                 rc = -errno;
520                 CT_ERROR(rc, "cannot stat '%s'", src);
521                 return rc;
522         }
523
524         if (!S_ISREG(src_st.st_mode)) {
525                 rc = -EINVAL;
526                 CT_ERROR(rc, "'%s' is not a regular file", src);
527                 return rc;
528         }
529
530         if (fstat(dst_fd, &dst_st) < 0) {
531                 rc = -errno;
532                 CT_ERROR(rc, "cannot stat '%s'", dst);
533                 return rc;
534         }
535
536         if (!S_ISREG(dst_st.st_mode)) {
537                 rc = -EINVAL;
538                 CT_ERROR(rc, "'%s' is not a regular file", dst);
539                 return rc;
540         }
541
542         rc = lseek(src_fd, hai->hai_extent.offset, SEEK_SET);
543         if (rc < 0) {
544                 rc = -errno;
545                 CT_ERROR(rc,
546                          "cannot seek for read to "LPU64" (len %jd) in '%s'",
547                          hai->hai_extent.offset, (intmax_t)src_st.st_size, src);
548                 return rc;
549         }
550
551         he.offset = offset;
552         he.length = 0;
553         rc = llapi_hsm_action_progress(hcp, &he, 0);
554         if (rc < 0) {
555                 /* Action has been canceled or something wrong
556                  * is happening. Stop copying data. */
557                 CT_ERROR(rc, "progress ioctl for copy '%s'->'%s' failed",
558                          src, dst);
559                 goto out;
560         }
561
562         errno = 0;
563         /* Don't read beyond a given extent */
564         length = min(hai->hai_extent.length, src_st.st_size);
565
566         buf = malloc(opt.o_chunk_size);
567         if (buf == NULL) {
568                 rc = -ENOMEM;
569                 goto out;
570         }
571
572         CT_DEBUG("Going to copy "LPU64" bytes %s -> %s\n", length, src, dst);
573
574         while (write_total < length) {
575                 ssize_t rsize;
576                 ssize_t wsize;
577                 int     chunk = (length - write_total > opt.o_chunk_size) ?
578                                  opt.o_chunk_size : length - write_total;
579
580                 rsize = pread(src_fd, buf, chunk, offset);
581                 if (rsize == 0)
582                         /* EOF */
583                         break;
584
585                 if (rsize < 0) {
586                         rc = -errno;
587                         CT_ERROR(rc, "cannot read from '%s'", src);
588                         break;
589                 }
590
591                 wsize = pwrite(dst_fd, buf, rsize, offset);
592                 if (wsize < 0) {
593                         rc = -errno;
594                         CT_ERROR(rc, "cannot write to '%s'", dst);
595                         break;
596                 }
597
598                 write_total += wsize;
599                 offset += wsize;
600
601                 if (opt.o_bandwidth != 0)
602                         /* sleep if needed, to honor bandwidth limits */
603                         bandwidth_ctl_delay(wsize);
604
605                 if (time(0) >= last_print_time + opt.o_report_int) {
606                         last_print_time = time(0);
607                         CT_TRACE("%%"LPU64" ", 100 * write_total / length);
608                         he.length = write_total;
609                         rc = llapi_hsm_action_progress(hcp, &he, 0);
610                         if (rc < 0) {
611                                 /* Action has been canceled or something wrong
612                                  * is happening. Stop copying data. */
613                                 CT_ERROR(rc, "progress ioctl for copy"
614                                          " '%s'->'%s' failed", src, dst);
615                                 goto out;
616                         }
617                 }
618                 rc = 0;
619         }
620
621 out:
622         /*
623          * truncate restored file
624          * size is taken from the archive this is done to support
625          * restore after a force release which leaves the file with the
626          * wrong size (can big bigger than the new size)
627          */
628         if ((hai->hai_action == HSMA_RESTORE) &&
629             (src_st.st_size < dst_st.st_size)) {
630                 /*
631                  * make sure the file is on disk before reporting success.
632                  */
633                 rc = ftruncate(dst_fd, src_st.st_size);
634                 if (rc < 0) {
635                         rc = -errno;
636                         CT_ERROR(rc, "cannot truncate '%s' to size %jd",
637                                  dst, (intmax_t)src_st.st_size);
638                         err_major++;
639                 }
640         }
641
642         if (buf != NULL)
643                 free(buf);
644
645         return rc;
646 }
647
648 /* Copy file attributes from file src to file dest */
649 static int ct_copy_attr(const char *src, const char *dst, int src_fd,
650                         int dst_fd)
651 {
652         struct stat     st;
653         struct timeval  times[2];
654         int             rc;
655
656         if (fstat(src_fd, &st) < 0) {
657                 rc = -errno;
658                 CT_ERROR(rc, "cannot stat '%s'", src);
659                 return rc;
660         }
661
662         times[0].tv_sec = st.st_atime;
663         times[0].tv_usec = 0;
664         times[1].tv_sec = st.st_mtime;
665         times[1].tv_usec = 0;
666         if (fchmod(dst_fd, st.st_mode) < 0 ||
667             fchown(dst_fd, st.st_uid, st.st_gid) < 0 ||
668             futimes(dst_fd, times) < 0) {
669                 rc = -errno;
670                 CT_ERROR(rc, "cannot set attributes of '%s'", src);
671                 return rc;
672         }
673
674         return 0;
675 }
676
677 static int ct_copy_xattr(const char *src, const char *dst, int src_fd,
678                          int dst_fd, bool is_restore)
679 {
680         char     list[XATTR_LIST_MAX];
681         char     value[XATTR_SIZE_MAX];
682         char    *name;
683         ssize_t  list_len;
684         int      rc;
685
686         list_len = flistxattr(src_fd, list, sizeof(list));
687         if (list_len < 0)
688                 return -errno;
689
690         name = list;
691         while (name < list + list_len) {
692                 rc = fgetxattr(src_fd, name, value, sizeof(value));
693                 if (rc < 0)
694                         return -errno;
695
696                 /* when we restore, we do not restore lustre xattr */
697                 if (!is_restore ||
698                     (strncmp(XATTR_TRUSTED_PREFIX, name,
699                              sizeof(XATTR_TRUSTED_PREFIX) - 1) != 0)) {
700                         rc = fsetxattr(dst_fd, name, value, rc, 0);
701                         CT_TRACE("fsetxattr of '%s' on '%s' rc=%d (%s)",
702                                  name, dst, rc, strerror(errno));
703                         /* lustre.* attrs aren't supported on other FS's */
704                         if (rc < 0 && errno != EOPNOTSUPP) {
705                                 rc = -errno;
706                                 CT_ERROR(rc, "cannot set extended attribute"
707                                          " '%s' of '%s'",
708                                          name, dst);
709                                 return rc;
710                         }
711                 }
712                 name += strlen(name) + 1;
713         }
714
715         return 0;
716 }
717
718 static int ct_path_lustre(char *buf, int sz, const char *mnt,
719                           const lustre_fid *fid)
720 {
721         return snprintf(buf, sz, "%s/%s/fid/"DFID_NOBRACE, mnt,
722                         dot_lustre_name, PFID(fid));
723 }
724
725 static int ct_path_archive(char *buf, int sz, const char *archive_dir,
726                            const lustre_fid *fid)
727 {
728         return snprintf(buf, sz, "%s/%04x/%04x/%04x/%04x/%04x/%04x/"
729                         DFID_NOBRACE, archive_dir,
730                         (fid)->f_oid       & 0xFFFF,
731                         (fid)->f_oid >> 16 & 0xFFFF,
732                         (unsigned int)((fid)->f_seq       & 0xFFFF),
733                         (unsigned int)((fid)->f_seq >> 16 & 0xFFFF),
734                         (unsigned int)((fid)->f_seq >> 32 & 0xFFFF),
735                         (unsigned int)((fid)->f_seq >> 48 & 0xFFFF),
736                         PFID(fid));
737 }
738
739 static bool ct_is_retryable(int err)
740 {
741         return err == -ETIMEDOUT;
742 }
743
744 static int ct_begin_restore(struct hsm_copyaction_private **phcp,
745                             const struct hsm_action_item *hai,
746                             int mdt_index, int open_flags)
747 {
748         char     src[PATH_MAX];
749         int      rc;
750
751         rc = llapi_hsm_action_begin(phcp, ctdata, hai, mdt_index, open_flags,
752                                     false);
753         if (rc < 0) {
754                 ct_path_lustre(src, sizeof(src), opt.o_mnt, &hai->hai_fid);
755                 CT_ERROR(rc, "llapi_hsm_action_begin() on '%s' failed", src);
756         }
757
758         return rc;
759 }
760
761 static int ct_begin(struct hsm_copyaction_private **phcp,
762                     const struct hsm_action_item *hai)
763 {
764         /* Restore takes specific parameters. Call the same function w/ default
765          * values for all other operations. */
766         return ct_begin_restore(phcp, hai, -1, 0);
767 }
768
769 static int ct_fini(struct hsm_copyaction_private **phcp,
770                    const struct hsm_action_item *hai, int hp_flags, int ct_rc)
771 {
772         char    lstr[PATH_MAX];
773         int     rc;
774
775         CT_TRACE("Action completed, notifying coordinator "
776                  "cookie="LPX64", FID="DFID", hp_flags=%d err=%d",
777                  hai->hai_cookie, PFID(&hai->hai_fid),
778                  hp_flags, -ct_rc);
779
780         ct_path_lustre(lstr, sizeof(lstr), opt.o_mnt, &hai->hai_fid);
781         rc = llapi_hsm_action_end(phcp, &hai->hai_extent, hp_flags, abs(ct_rc));
782         if (rc == -ECANCELED)
783                 CT_ERROR(rc, "completed action on '%s' has been canceled: "
784                          "cookie="LPX64", FID="DFID, lstr, hai->hai_cookie,
785                          PFID(&hai->hai_fid));
786         else if (rc < 0)
787                 CT_ERROR(rc, "llapi_hsm_action_end() on '%s' failed", lstr);
788         else
789                 CT_TRACE("llapi_hsm_action_end() on '%s' ok (rc=%d)",
790                          lstr, rc);
791
792         return rc;
793 }
794
795 static int ct_archive(const struct hsm_action_item *hai, const long hal_flags)
796 {
797         struct hsm_copyaction_private   *hcp = NULL;
798         char                             src[PATH_MAX];
799         char                             dst[PATH_MAX] = "";
800         int                              rc;
801         int                              rcf = 0;
802         bool                             rename_needed = false;
803         int                              hp_flags = 0;
804         int                              open_flags;
805         int                              src_fd = -1;
806         int                              dst_fd = -1;
807
808         rc = ct_begin(&hcp, hai);
809         if (rc < 0)
810                 goto fini_major;
811
812         /* we fill archive so:
813          * source = data FID
814          * destination = lustre FID
815          */
816         ct_path_lustre(src, sizeof(src), opt.o_mnt, &hai->hai_dfid);
817         ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, &hai->hai_fid);
818         if (hai->hai_extent.length == -1) {
819                 /* whole file, write it to tmp location and atomically
820                  * replace old archived file */
821                 strncat(dst, "_tmp", sizeof(dst) - strlen(dst) - 1);
822                 /* we cannot rely on the same test because ct_copy_data()
823                  * updates hai_extent.length */
824                 rename_needed = true;
825         }
826
827         CT_TRACE("archiving '%s' to '%s'", src, dst);
828
829         if (opt.o_dry_run) {
830                 rc = 0;
831                 goto fini_major;
832         }
833
834         rc = ct_mkdir_p(dst);
835         if (rc < 0) {
836                 CT_ERROR(rc, "mkdir_p '%s' failed", dst);
837                 goto fini_major;
838         }
839
840         src_fd = llapi_hsm_action_get_fd(hcp);
841         if (src_fd < 0) {
842                 rc = -errno;
843                 CT_ERROR(rc, "cannot open '%s' for read", src);
844                 goto fini_major;
845         }
846
847         open_flags = O_WRONLY | O_NOFOLLOW | O_NONBLOCK;
848         /* If extent is specified, don't truncate an old archived copy */
849         open_flags |= ((hai->hai_extent.length == -1) ? O_TRUNC : 0) | O_CREAT;
850
851         dst_fd = open(dst, open_flags, FILE_PERM);
852         if (dst_fd == -1) {
853                 rc = -errno;
854                 CT_ERROR(rc, "cannot open '%s' for write", dst);
855                 goto fini_major;
856         }
857
858         /* saving stripe is not critical */
859         rc = ct_save_stripe(src_fd, src, dst);
860         if (rc < 0)
861                 CT_ERROR(rc, "cannot save file striping info of '%s' in '%s'",
862                          src, dst);
863
864         rc = ct_copy_data(hcp, src, dst, src_fd, dst_fd, hai, hal_flags);
865         if (rc < 0) {
866                 CT_ERROR(rc, "data copy failed from '%s' to '%s'", src, dst);
867                 goto fini_major;
868         }
869
870         rc = fsync(dst_fd);
871         if (rc < 0) {
872                 rc = -errno;
873                 CT_ERROR(rc, "cannot flush '%s' archive file '%s'", src, dst);
874                 goto fini_major;
875         }
876
877         CT_TRACE("data archiving for '%s' to '%s' done", src, dst);
878
879         /* attrs will remain on the MDS; no need to copy them, except possibly
880           for disaster recovery */
881         if (opt.o_copy_attrs) {
882                 rc = ct_copy_attr(src, dst, src_fd, dst_fd);
883                 if (rc < 0) {
884                         CT_ERROR(rc, "cannot copy attr of '%s' to '%s'",
885                                  src, dst);
886                         rcf = rc;
887                 }
888                 CT_TRACE("attr file for '%s' saved to archive '%s'",
889                          src, dst);
890         }
891
892         /* xattrs will remain on the MDS; no need to copy them, except possibly
893          for disaster recovery */
894         if (opt.o_copy_xattrs) {
895                 rc = ct_copy_xattr(src, dst, src_fd, dst_fd, false);
896                 if (rc < 0) {
897                         CT_ERROR(rc, "cannot copy xattr of '%s' to '%s'",
898                                  src, dst);
899                         rcf = rcf ? rcf : rc;
900                 }
901                 CT_TRACE("xattr file for '%s' saved to archive '%s'",
902                          src, dst);
903         }
904
905         if (rename_needed == true) {
906                 char     tmp_src[PATH_MAX];
907                 char     tmp_dst[PATH_MAX];
908
909                 /* atomically replace old archived file */
910                 ct_path_archive(src, sizeof(src), opt.o_hsm_root,
911                                 &hai->hai_fid);
912                 rc = rename(dst, src);
913                 if (rc < 0) {
914                         rc = -errno;
915                         CT_ERROR(rc, "cannot rename '%s' to '%s'", dst, src);
916                         goto fini_major;
917                 }
918                 /* rename lov file */
919                 snprintf(tmp_src, sizeof(tmp_src), "%s.lov", src);
920                 snprintf(tmp_dst, sizeof(tmp_dst), "%s.lov", dst);
921                 rc = rename(tmp_dst, tmp_src);
922                 if (rc < 0)
923                         CT_ERROR(errno, "cannot rename '%s' to '%s'",
924                                  tmp_dst, tmp_src);
925         }
926
927         if (opt.o_shadow_tree) {
928                 /* Create a namespace of softlinks that shadows the original
929                  * Lustre namespace.  This will only be current at
930                  * time-of-archive (won't follow renames).
931                  * WARNING: release won't kill these links; a manual
932                  * cleanup of dead links would be required.
933                  */
934                 char             buf[PATH_MAX];
935                 long long        recno = -1;
936                 int              linkno = 0;
937                 char            *ptr;
938                 int              depth = 0;
939                 ssize_t          sz;
940
941                 sprintf(buf, DFID, PFID(&hai->hai_fid));
942                 sprintf(src, "%s/shadow/", opt.o_hsm_root);
943
944                 ptr = opt.o_hsm_root;
945                 while (*ptr)
946                         (*ptr++ == '/') ? depth-- : 0;
947
948                 rc = llapi_fid2path(opt.o_mnt, buf, src + strlen(src),
949                                     sizeof(src) - strlen(src), &recno, &linkno);
950                 if (rc < 0) {
951                         CT_ERROR(rc, "cannot get FID of '%s'", buf);
952                         rcf = rcf ? rcf : rc;
953                         goto fini_minor;
954                 }
955
956                 /* Figure out how many parent dirs to symlink back */
957                 ptr = src;
958                 while (*ptr)
959                         (*ptr++ == '/') ? depth++ : 0;
960                 sprintf(buf, "..");
961                 while (--depth > 1)
962                         strcat(buf, "/..");
963
964                 ct_path_archive(dst, sizeof(dst), buf, &hai->hai_fid);
965
966                 if (ct_mkdir_p(src)) {
967                         CT_ERROR(errno, "mkdir_p '%s' failed", src);
968                         rcf = rcf ? rcf : -errno;
969                         goto fini_minor;
970                 }
971                 /* symlink already exists ? */
972                 sz = readlink(src, buf, sizeof(buf));
973                 /* detect truncation */
974                 if (sz == sizeof(buf)) {
975                         rcf = rcf ? rcf : -E2BIG;
976                         CT_ERROR(rcf, "readlink '%s' truncated", src);
977                         goto fini_minor;
978                 }
979                 if (sz >= 0) {
980                         buf[sz] = '\0';
981                         if (sz == 0 || strncmp(buf, dst, sz) != 0) {
982                                 if (unlink(src) && errno != ENOENT) {
983                                         CT_ERROR(errno,
984                                                  "cannot unlink symlink '%s'",
985                                                  src);
986                                         rcf = rcf ? rcf : -errno;
987                                         goto fini_minor;
988                                 /* unlink old symlink done */
989                                 CT_TRACE("remove old symlink '%s' pointing"
990                                          " to '%s'", src, buf);
991                                 }
992                         } else {
993                                 /* symlink already ok */
994                                 CT_TRACE("symlink '%s' already pointing"
995                                          " to '%s'", src, dst);
996                                 rcf = 0;
997                                 goto fini_minor;
998                         }
999                 }
1000                 if (symlink(dst, src)) {
1001                         CT_ERROR(errno, "cannot symlink '%s' to '%s'",
1002                                  src, dst);
1003                         rcf = rcf ? rcf : -errno;
1004                         goto fini_minor;
1005                 }
1006                 CT_TRACE("symlink '%s' to '%s' done", src, dst);
1007         }
1008 fini_minor:
1009         if (rcf)
1010                 err_minor++;
1011         goto out;
1012
1013
1014 fini_major:
1015         err_major++;
1016
1017         unlink(dst);
1018         if (ct_is_retryable(rc))
1019                 hp_flags |= HP_FLAG_RETRY;
1020
1021         rcf = rc;
1022
1023 out:
1024         if (!(src_fd < 0))
1025                 close(src_fd);
1026
1027         if (!(dst_fd < 0))
1028                 close(dst_fd);
1029
1030         if (hcp != NULL)
1031                 rc = ct_fini(&hcp, hai, hp_flags, rcf);
1032
1033         return rc;
1034 }
1035
1036 static int ct_restore(const struct hsm_action_item *hai, const long hal_flags)
1037 {
1038         struct hsm_copyaction_private   *hcp = NULL;
1039         char                             src[PATH_MAX];
1040         char                             dst[PATH_MAX];
1041         char                             lov_buf[XATTR_SIZE_MAX];
1042         size_t                           lov_size = sizeof(lov_buf);
1043         int                              rc;
1044         int                              hp_flags = 0;
1045         int                              src_fd = -1;
1046         int                              dst_fd = -1;
1047         int                              mdt_index = -1; /* Not implemented */
1048         int                              open_flags = 0;
1049         bool                             set_lovea;
1050         struct lu_fid                    dfid;
1051         /* we fill lustre so:
1052          * source = lustre FID in the backend
1053          * destination = data FID = volatile file
1054          */
1055
1056         /* build backend file name from released file FID */
1057         ct_path_archive(src, sizeof(src), opt.o_hsm_root, &hai->hai_fid);
1058
1059         /* restore loads and sets the LOVEA w/o interpreting it to avoid
1060          * dependency on the structure format. */
1061         rc = ct_load_stripe(src, lov_buf, &lov_size);
1062         if (rc < 0) {
1063                 CT_WARN("cannot get stripe rules for '%s' (%s), use default",
1064                         src, strerror(-rc));
1065                 set_lovea = false;
1066         } else {
1067                 open_flags |= O_LOV_DELAY_CREATE;
1068                 set_lovea = true;
1069         }
1070
1071         rc = ct_begin_restore(&hcp, hai, mdt_index, open_flags);
1072         if (rc < 0)
1073                 goto fini;
1074
1075         /* get the FID of the volatile file */
1076         rc = llapi_hsm_action_get_dfid(hcp, &dfid);
1077         if (rc < 0) {
1078                 CT_ERROR(rc, "restoring "DFID
1079                          ", cannot get FID of created volatile file",
1080                          PFID(&hai->hai_fid));
1081                 goto fini;
1082         }
1083
1084         /* build volatile "file name", for messages */
1085         snprintf(dst, sizeof(dst), "{VOLATILE}="DFID, PFID(&dfid));
1086
1087         CT_TRACE("restoring data from '%s' to '%s'", src, dst);
1088
1089         if (opt.o_dry_run) {
1090                 rc = 0;
1091                 goto fini;
1092         }
1093
1094         src_fd = open(src, O_RDONLY | O_NOATIME | O_NONBLOCK | O_NOFOLLOW);
1095         if (src_fd < 0) {
1096                 rc = -errno;
1097                 CT_ERROR(rc, "cannot open '%s' for read", src);
1098                 goto fini;
1099         }
1100
1101         dst_fd = llapi_hsm_action_get_fd(hcp);
1102
1103         if (set_lovea) {
1104                 /* the layout cannot be allocated through .fid so we have to
1105                  * restore a layout */
1106                 rc = ct_restore_stripe(src, dst, dst_fd, lov_buf, lov_size);
1107                 if (rc < 0) {
1108                         CT_ERROR(rc, "cannot restore file striping info"
1109                                  " for '%s' from '%s'", dst, src);
1110                         err_major++;
1111                         goto fini;
1112                 }
1113         }
1114
1115         rc = ct_copy_data(hcp, src, dst, src_fd, dst_fd, hai, hal_flags);
1116         if (rc < 0) {
1117                 CT_ERROR(rc, "cannot copy data from '%s' to '%s'",
1118                          src, dst);
1119                 err_major++;
1120                 if (ct_is_retryable(rc))
1121                         hp_flags |= HP_FLAG_RETRY;
1122                 goto fini;
1123         }
1124
1125         CT_TRACE("data restore from '%s' to '%s' done", src, dst);
1126
1127 fini:
1128         if (hcp != NULL)
1129                 rc = ct_fini(&hcp, hai, hp_flags, rc);
1130
1131         /* object swaping is done by cdt at copy end, so close of volatile file
1132          * cannot be done before */
1133         if (!(src_fd < 0))
1134                 close(src_fd);
1135
1136         if (!(dst_fd < 0))
1137                 close(dst_fd);
1138
1139         return rc;
1140 }
1141
1142 static int ct_remove(const struct hsm_action_item *hai, const long hal_flags)
1143 {
1144         struct hsm_copyaction_private   *hcp = NULL;
1145         char                             dst[PATH_MAX];
1146         int                              rc;
1147
1148         rc = ct_begin(&hcp, hai);
1149         if (rc < 0)
1150                 goto fini;
1151
1152         ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, &hai->hai_fid);
1153
1154         CT_TRACE("removing file '%s'", dst);
1155
1156         if (opt.o_dry_run) {
1157                 rc = 0;
1158                 goto fini;
1159         }
1160
1161         rc = unlink(dst);
1162         if (rc < 0) {
1163                 rc = -errno;
1164                 CT_ERROR(rc, "cannot unlink '%s'", dst);
1165                 err_minor++;
1166                 goto fini;
1167         }
1168
1169         strncat(dst, ".lov", sizeof(dst) - strlen(dst) - 1);
1170         rc = unlink(dst);
1171         if (rc < 0) {
1172                 rc = -errno;
1173                 CT_ERROR(rc, "cannot unlink '%s'", dst);
1174                 err_minor++;
1175                 goto fini;
1176         }
1177
1178 fini:
1179         if (hcp != NULL)
1180                 rc = ct_fini(&hcp, hai, 0, rc);
1181
1182         return rc;
1183 }
1184
1185 static int ct_report_error(const struct hsm_action_item *hai, int flags,
1186                            int errval)
1187 {
1188         struct hsm_copyaction_private   *hcp;
1189         int                              rc;
1190
1191         rc = llapi_hsm_action_begin(&hcp, ctdata, hai, -1, 0, true);
1192         if (rc < 0)
1193                 return rc;
1194
1195         rc = llapi_hsm_action_end(&hcp, &hai->hai_extent, flags, abs(errval));
1196
1197         return rc;
1198 }
1199
1200 static int ct_process_item(struct hsm_action_item *hai, const long hal_flags)
1201 {
1202         int     rc = 0;
1203
1204         if (opt.o_verbose >= LLAPI_MSG_INFO || opt.o_dry_run) {
1205                 /* Print the original path */
1206                 char            fid[128];
1207                 char            path[PATH_MAX];
1208                 long long       recno = -1;
1209                 int             linkno = 0;
1210
1211                 sprintf(fid, DFID, PFID(&hai->hai_fid));
1212                 CT_TRACE("'%s' action %s reclen %d, cookie="LPX64,
1213                          fid, hsm_copytool_action2name(hai->hai_action),
1214                          hai->hai_len, hai->hai_cookie);
1215                 rc = llapi_fid2path(opt.o_mnt, fid, path,
1216                                     sizeof(path), &recno, &linkno);
1217                 if (rc < 0)
1218                         CT_ERROR(rc, "cannot get path of FID %s", fid);
1219                 else
1220                         CT_TRACE("processing file '%s'", path);
1221         }
1222
1223         switch (hai->hai_action) {
1224         /* set err_major, minor inside these functions */
1225         case HSMA_ARCHIVE:
1226                 rc = ct_archive(hai, hal_flags);
1227                 break;
1228         case HSMA_RESTORE:
1229                 rc = ct_restore(hai, hal_flags);
1230                 break;
1231         case HSMA_REMOVE:
1232                 rc = ct_remove(hai, hal_flags);
1233                 break;
1234         case HSMA_CANCEL:
1235                 CT_TRACE("cancel not implemented for file system '%s'",
1236                          opt.o_mnt);
1237                 /* Don't report progress to coordinator for this cookie:
1238                  * the copy function will get ECANCELED when reporting
1239                  * progress. */
1240                 err_minor++;
1241                 return 0;
1242                 break;
1243         default:
1244                 rc = -EINVAL;
1245                 CT_ERROR(rc, "unknown action %d, on '%s'", hai->hai_action,
1246                          opt.o_mnt);
1247                 err_minor++;
1248                 ct_report_error(hai, 0, rc);
1249         }
1250
1251         return 0;
1252 }
1253
1254 struct ct_th_data {
1255         long                     hal_flags;
1256         struct hsm_action_item  *hai;
1257 };
1258
1259 static void *ct_thread(void *data)
1260 {
1261         struct ct_th_data *cttd = data;
1262         int rc;
1263
1264         rc = ct_process_item(cttd->hai, cttd->hal_flags);
1265
1266         free(cttd->hai);
1267         free(cttd);
1268         pthread_exit((void *)(intptr_t)rc);
1269 }
1270
1271 static int ct_process_item_async(const struct hsm_action_item *hai,
1272                                  long hal_flags)
1273 {
1274         pthread_attr_t           attr;
1275         pthread_t                thread;
1276         struct ct_th_data       *data;
1277         int                      rc;
1278
1279         data = malloc(sizeof(*data));
1280         if (data == NULL)
1281                 return -ENOMEM;
1282
1283         data->hai = malloc(hai->hai_len);
1284         if (data->hai == NULL) {
1285                 free(data);
1286                 return -ENOMEM;
1287         }
1288
1289         memcpy(data->hai, hai, hai->hai_len);
1290         data->hal_flags = hal_flags;
1291
1292         rc = pthread_attr_init(&attr);
1293         if (rc != 0) {
1294                 CT_ERROR(rc, "pthread_attr_init failed for '%s' service",
1295                          opt.o_mnt);
1296                 free(data->hai);
1297                 free(data);
1298                 return -rc;
1299         }
1300
1301         pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
1302
1303         rc = pthread_create(&thread, &attr, ct_thread, data);
1304         if (rc != 0)
1305                 CT_ERROR(rc, "cannot create thread for '%s' service",
1306                          opt.o_mnt);
1307
1308         pthread_attr_destroy(&attr);
1309         return 0;
1310 }
1311
1312 static int ct_import_one(const char *src, const char *dst)
1313 {
1314         char            newarc[PATH_MAX];
1315         lustre_fid      fid;
1316         struct stat     st;
1317         int             rc;
1318
1319         CT_TRACE("importing '%s' from '%s'", dst, src);
1320
1321         if (stat(src, &st) < 0) {
1322                 rc = -errno;
1323                 CT_ERROR(rc, "cannot stat '%s'", src);
1324                 return rc;
1325         }
1326
1327         if (opt.o_dry_run)
1328                 return 0;
1329
1330         rc = llapi_hsm_import(dst,
1331                               opt.o_archive_cnt ? opt.o_archive_id[0] : 0,
1332                               &st, 0, 0, 0, 0, NULL, &fid);
1333         if (rc < 0) {
1334                 CT_ERROR(rc, "cannot import '%s' from '%s'", dst, src);
1335                 return rc;
1336         }
1337
1338         ct_path_archive(newarc, sizeof(newarc), opt.o_hsm_root, &fid);
1339
1340         rc = ct_mkdir_p(newarc);
1341         if (rc < 0) {
1342                 CT_ERROR(rc, "mkdir_p '%s' failed", newarc);
1343                 err_major++;
1344                 return rc;
1345
1346         }
1347
1348         /* Lots of choices now: mv, ln, ln -s ? */
1349         rc = link(src, newarc); /* hardlink */
1350         if (rc < 0) {
1351                 rc = -errno;
1352                 CT_ERROR(rc, "cannot link '%s' to '%s'", newarc, src);
1353                 err_major++;
1354                 return rc;
1355         }
1356         CT_TRACE("imported '%s' from '%s'=='%s'", dst, newarc, src);
1357
1358         return 0;
1359 }
1360
1361 static char *path_concat(const char *dirname, const char *basename)
1362 {
1363         char    *result;
1364         int      dirlen = strlen(dirname);
1365
1366         result = malloc(dirlen + strlen(basename) + 2);
1367         if (result == NULL)
1368                 return NULL;
1369
1370         memcpy(result, dirname, dirlen);
1371         result[dirlen] = '/';
1372         strcpy(result + dirlen + 1, basename);
1373
1374         return result;
1375 }
1376
1377 static int ct_import_recurse(const char *relpath)
1378 {
1379         DIR             *dir;
1380         struct dirent    ent, *cookie = NULL;
1381         char            *srcpath, *newpath;
1382         int              rc;
1383
1384         if (relpath == NULL)
1385                 return -EINVAL;
1386
1387         srcpath = path_concat(opt.o_hsm_root, relpath);
1388         if (srcpath == NULL) {
1389                 err_major++;
1390                 return -ENOMEM;
1391         }
1392
1393         dir = opendir(srcpath);
1394         if (dir == NULL) {
1395                 /* Not a dir, or error */
1396                 if (errno == ENOTDIR) {
1397                         /* Single regular file case, treat o_dst as absolute
1398                            final location. */
1399                         rc = ct_import_one(srcpath, opt.o_dst);
1400                 } else {
1401                         rc = -errno;
1402                         CT_ERROR(rc, "cannot opendir '%s'", srcpath);
1403                         err_major++;
1404                 }
1405                 free(srcpath);
1406                 return rc;
1407         }
1408         free(srcpath);
1409
1410         while (1) {
1411                 rc = readdir_r(dir, &ent, &cookie);
1412                 if (rc != 0) {
1413                         rc = -errno;
1414                         CT_ERROR(rc, "cannot readdir_r '%s'", relpath);
1415                         err_major++;
1416                         goto out;
1417                 } else if ((rc == 0) && (cookie == NULL)) {
1418                         /* end of directory */
1419                         break;
1420                 }
1421
1422                 if (!strcmp(ent.d_name, ".") ||
1423                     !strcmp(ent.d_name, ".."))
1424                         continue;
1425
1426                 /* New relative path */
1427                 newpath = path_concat(relpath, ent.d_name);
1428                 if (newpath == NULL) {
1429                         err_major++;
1430                         rc = -ENOMEM;
1431                         goto out;
1432                 }
1433
1434                 if (ent.d_type == DT_DIR) {
1435                         rc = ct_import_recurse(newpath);
1436                 } else {
1437                         char src[PATH_MAX];
1438                         char dst[PATH_MAX];
1439
1440                         sprintf(src, "%s/%s", opt.o_hsm_root, newpath);
1441                         sprintf(dst, "%s/%s", opt.o_dst, newpath);
1442                         /* Make the target dir in the Lustre fs */
1443                         rc = ct_mkdir_p(dst);
1444                         if (rc == 0) {
1445                                 /* Import the file */
1446                                 rc = ct_import_one(src, dst);
1447                         } else {
1448                                 CT_ERROR(rc, "ct_mkdir_p '%s' failed", dst);
1449                                 err_major++;
1450                         }
1451                 }
1452
1453                 if (rc != 0) {
1454                         CT_ERROR(rc, "cannot import '%s'", newpath);
1455                         if (err_major && opt.o_abort_on_error) {
1456                                 free(newpath);
1457                                 goto out;
1458                         }
1459                 }
1460                 free(newpath);
1461         }
1462
1463         rc = 0;
1464 out:
1465         closedir(dir);
1466         return rc;
1467 }
1468
1469 static int ct_rebind_one(const lustre_fid *old_fid, const lustre_fid *new_fid)
1470 {
1471         char    src[PATH_MAX];
1472         char    dst[PATH_MAX];
1473         int     rc;
1474
1475         CT_TRACE("rebind "DFID" to "DFID, PFID(old_fid), PFID(new_fid));
1476
1477         ct_path_archive(src, sizeof(src), opt.o_hsm_root, old_fid);
1478         ct_path_archive(dst, sizeof(dst), opt.o_hsm_root, new_fid);
1479
1480         if (!opt.o_dry_run) {
1481                 ct_mkdir_p(dst);
1482                 if (rename(src, dst)) {
1483                         rc = -errno;
1484                         CT_ERROR(rc, "cannot rename '%s' to '%s'", src, dst);
1485                         return -errno;
1486                 }
1487                 /* rename lov file */
1488                 strncat(src, ".lov", sizeof(src) - strlen(src) - 1);
1489                 strncat(dst, ".lov", sizeof(dst) - strlen(dst) - 1);
1490                 if (rename(src, dst))
1491                         CT_ERROR(errno, "cannot '%s' rename to '%s'", src, dst);
1492
1493         }
1494         return 0;
1495 }
1496
1497 static bool fid_is_file(lustre_fid *fid)
1498 {
1499         return fid_is_norm(fid) || fid_is_igif(fid);
1500 }
1501
1502 static bool should_ignore_line(const char *line)
1503 {
1504         int     i;
1505
1506         for (i = 0; line[i] != '\0'; i++) {
1507                 if (isspace(line[i]))
1508                         continue;
1509                 else if (line[i] == '#')
1510                         return true;
1511                 else
1512                         return false;
1513         }
1514
1515         return true;
1516 }
1517
1518 static int ct_rebind_list(const char *list)
1519 {
1520         int              rc;
1521         FILE            *filp;
1522         ssize_t          r;
1523         char            *line = NULL;
1524         size_t           line_size = 0;
1525         unsigned int     nl = 0;
1526         unsigned int     ok = 0;
1527
1528         filp = fopen(list, "r");
1529         if (filp == NULL) {
1530                 rc = -errno;
1531                 CT_ERROR(rc, "cannot open '%s'", list);
1532                 return rc;
1533         }
1534
1535         /* each line consists of 2 FID */
1536         while ((r = getline(&line, &line_size, filp)) != -1) {
1537                 lustre_fid      old_fid;
1538                 lustre_fid      new_fid;
1539
1540                 /* Ignore empty and commented out ('#...') lines. */
1541                 if (should_ignore_line(line))
1542                         continue;
1543
1544                 nl++;
1545
1546                 rc = sscanf(line, SFID" "SFID, RFID(&old_fid), RFID(&new_fid));
1547                 if (rc != 6 || !fid_is_file(&old_fid) ||
1548                     !fid_is_file(&new_fid)) {
1549                         CT_ERROR(EINVAL,
1550                                  "'%s' FID expected near '%s', line %u",
1551                                  list, line, nl);
1552                         err_major++;
1553                         continue;
1554                 }
1555
1556                 if (ct_rebind_one(&old_fid, &new_fid))
1557                         err_major++;
1558                 else
1559                         ok++;
1560         }
1561
1562         fclose(filp);
1563
1564         if (line)
1565                 free(line);
1566
1567         /* return 0 if all rebinds were sucessful */
1568         CT_TRACE("%u lines read from '%s', %u rebind successful", nl, list, ok);
1569
1570         return ok == nl ? 0 : -1;
1571 }
1572
1573 static int ct_rebind(void)
1574 {
1575         int     rc;
1576
1577         if (opt.o_dst) {
1578                 lustre_fid      old_fid;
1579                 lustre_fid      new_fid;
1580
1581                 if (sscanf(opt.o_src, SFID, RFID(&old_fid)) != 3 ||
1582                     !fid_is_file(&old_fid)) {
1583                         rc = -EINVAL;
1584                         CT_ERROR(rc, "'%s' invalid FID format", opt.o_src);
1585                         return rc;
1586                 }
1587
1588                 if (sscanf(opt.o_dst, SFID, RFID(&new_fid)) != 3 ||
1589                     !fid_is_file(&new_fid)) {
1590                         rc = -EINVAL;
1591                         CT_ERROR(rc, "'%s' invalid FID format", opt.o_dst);
1592                         return rc;
1593                 }
1594
1595                 rc = ct_rebind_one(&old_fid, &new_fid);
1596
1597                 return rc;
1598         }
1599
1600         /* o_src is a list file */
1601         rc = ct_rebind_list(opt.o_src);
1602
1603         return rc;
1604 }
1605
1606 static int ct_dir_level_max(const char *dirpath, __u16 *sub_seqmax)
1607 {
1608         DIR             *dir;
1609         int              rc;
1610         __u16            sub_seq;
1611         struct dirent    ent, *cookie = NULL;
1612
1613         *sub_seqmax = 0;
1614
1615         dir = opendir(dirpath);
1616         if (dir == NULL) {
1617                 rc = -errno;
1618                 CT_ERROR(rc, "cannot open directory '%s'", opt.o_hsm_root);
1619                 return rc;
1620         }
1621
1622         while ((rc = readdir_r(dir, &ent, &cookie)) == 0) {
1623                 if (cookie == NULL)
1624                         /* end of directory.
1625                          * rc is 0 and seqmax contains the max value. */
1626                         goto out;
1627
1628                 if (!strcmp(ent.d_name, ".") || !strcmp(ent.d_name, ".."))
1629                         continue;
1630
1631                 if (sscanf(ent.d_name, "%hx", &sub_seq) != 1) {
1632                         CT_TRACE("'%s' has an unexpected dirname format, "
1633                                  "skip entry", ent.d_name);
1634                         continue;
1635                 }
1636                 if (sub_seq > *sub_seqmax)
1637                         *sub_seqmax = sub_seq;
1638         }
1639         rc = -errno;
1640         CT_ERROR(rc, "cannot readdir_r '%s'", dirpath);
1641
1642 out:
1643         closedir(dir);
1644         return rc;
1645 }
1646
1647 static int ct_max_sequence(void)
1648 {
1649         int   rc, i;
1650         char  path[PATH_MAX];
1651         __u64 seq = 0;
1652         __u16 subseq;
1653
1654         strncpy(path, opt.o_hsm_root, sizeof(path));
1655         /* FID sequence is stored in top-level directory names:
1656          * hsm_root/16bits (high weight)/16 bits/16 bits/16 bits (low weight).
1657          */
1658         for (i = 0; i < 4; i++) {
1659                 rc = ct_dir_level_max(path, &subseq);
1660                 if (rc != 0)
1661                         return rc;
1662                 seq |= ((__u64)subseq << ((3 - i) * 16));
1663                 sprintf(path + strlen(path), "/%04x", subseq);
1664         }
1665
1666         printf("max_sequence: %016Lx\n", seq);
1667
1668         return 0;
1669 }
1670
1671 static void handler(int signal)
1672 {
1673         psignal(signal, "exiting");
1674         /* If we don't clean up upon interrupt, umount thinks there's a ref
1675          * and doesn't remove us from mtab (EINPROGRESS). The lustre client
1676          * does successfully unmount and the mount is actually gone, but the
1677          * mtab entry remains. So this just makes mtab happier. */
1678         llapi_hsm_copytool_unregister(&ctdata);
1679         _exit(1);
1680 }
1681
1682 /* Daemon waits for messages from the kernel; run it in the background. */
1683 static int ct_run(void)
1684 {
1685         int     rc;
1686
1687         if (opt.o_daemonize) {
1688                 rc = daemon(1, 1);
1689                 if (rc < 0) {
1690                         rc = -errno;
1691                         CT_ERROR(rc, "cannot daemonize");
1692                         return rc;
1693                 }
1694         }
1695
1696         setbuf(stdout, NULL);
1697
1698         rc = llapi_hsm_copytool_register(&ctdata, opt.o_mnt, 0,
1699                                          opt.o_archive_cnt, opt.o_archive_id);
1700         if (rc < 0) {
1701                 CT_ERROR(rc, "cannot start copytool interface");
1702                 return rc;
1703         }
1704
1705         signal(SIGINT, handler);
1706         signal(SIGTERM, handler);
1707
1708         while (1) {
1709                 struct hsm_action_list  *hal;
1710                 struct hsm_action_item  *hai;
1711                 int                      msgsize;
1712                 int                      i = 0;
1713
1714                 CT_TRACE("waiting for message from kernel");
1715
1716                 rc = llapi_hsm_copytool_recv(ctdata, &hal, &msgsize);
1717                 if (rc == -ESHUTDOWN) {
1718                         CT_TRACE("shutting down");
1719                         break;
1720                 } else if (rc == -EAGAIN) {
1721                         continue; /* msg not for us */
1722                 } else if (rc < 0) {
1723                         CT_WARN("cannot receive action list: %s",
1724                                 strerror(-rc));
1725                         err_major++;
1726                         if (opt.o_abort_on_error)
1727                                 break;
1728                         else
1729                                 continue;
1730                 }
1731
1732                 CT_TRACE("copytool fs=%s archive#=%d item_count=%d",
1733                          hal->hal_fsname, hal->hal_archive_id, hal->hal_count);
1734
1735                 if (strcmp(hal->hal_fsname, fs_name) != 0) {
1736                         CT_ERROR(EINVAL, "'%s' invalid fs name, expecting: %s",
1737                                  hal->hal_fsname, fs_name);
1738                         err_major++;
1739                         if (opt.o_abort_on_error)
1740                                 break;
1741                         else
1742                                 continue;
1743                 }
1744
1745                 hai = hai_first(hal);
1746                 while (++i <= hal->hal_count) {
1747                         if ((char *)hai - (char *)hal > msgsize) {
1748                                 rc = -EPROTO;
1749                                 CT_ERROR(rc,
1750                                          "'%s' item %d past end of message!",
1751                                          opt.o_mnt, i);
1752                                 err_major++;
1753                                 break;
1754                         }
1755                         rc = ct_process_item_async(hai, hal->hal_flags);
1756                         if (rc < 0)
1757                                 CT_ERROR(rc, "'%s' item %d process",
1758                                          opt.o_mnt, i);
1759                         if (opt.o_abort_on_error && err_major)
1760                                 break;
1761                         hai = hai_next(hai);
1762                 }
1763
1764                 llapi_hsm_action_list_free(&hal);
1765
1766                 if (opt.o_abort_on_error && err_major)
1767                         break;
1768         }
1769
1770         llapi_hsm_copytool_unregister(&ctdata);
1771
1772         return rc;
1773 }
1774
1775 static int ct_setup(void)
1776 {
1777         int     rc;
1778
1779         /* set llapi message level */
1780         llapi_msg_set_level(opt.o_verbose);
1781
1782         arc_fd = open(opt.o_hsm_root, O_RDONLY);
1783         if (arc_fd < 0) {
1784                 rc = -errno;
1785                 CT_ERROR(rc, "cannot open archive at '%s'", opt.o_hsm_root);
1786                 return rc;
1787         }
1788
1789         rc = llapi_search_fsname(opt.o_mnt, fs_name);
1790         if (rc < 0) {
1791                 CT_ERROR(rc, "cannot find a Lustre filesystem mounted at '%s'",
1792                          opt.o_mnt);
1793                 return -rc;
1794         }
1795
1796         return rc;
1797 }
1798
1799 static int ct_cleanup(void)
1800 {
1801         int     rc;
1802
1803         if (arc_fd < 0)
1804                 return 0;
1805
1806         if (close(arc_fd) < 0) {
1807                 rc = -errno;
1808                 CT_ERROR(rc, "cannot close archive root directory");
1809                 return rc;
1810         }
1811
1812         return 0;
1813 }
1814
1815 int main(int argc, char **argv)
1816 {
1817         int     rc;
1818
1819         strncpy(cmd_name, basename(argv[0]), sizeof(cmd_name));
1820         rc = ct_parseopts(argc, argv);
1821         if (rc < 0) {
1822                 CT_WARN("try '%s --help' for more information", cmd_name);
1823                 return -rc;
1824         }
1825
1826         ct_setup();
1827
1828         switch (opt.o_action) {
1829         case CA_IMPORT:
1830                 rc = ct_import_recurse(opt.o_src);
1831                 break;
1832         case CA_REBIND:
1833                 rc = ct_rebind();
1834                 break;
1835         case CA_MAXSEQ:
1836                 rc = ct_max_sequence();
1837                 break;
1838         default:
1839                 rc = ct_run();
1840                 break;
1841         }
1842
1843         if (opt.o_action != CA_MAXSEQ)
1844                 CT_TRACE("process finished, errs: %d major, %d minor,"
1845                          " rc=%d (%s)", err_major, err_minor, rc,
1846                          strerror(-rc));
1847
1848         ct_cleanup();
1849
1850         return -rc;
1851 }
1852