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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/utils/lustre_rsync.c
38 * Author: Kalpak Shah <Kalpak.Shah@Sun.COM>
39 * Author: Manoj Joseph <Manoj.Joseph@Sun.COM>
43 * - lustre_rsync is a tool for replicating a lustre filesystem.
45 * - The source-fs is a live lustre filesystem. It is not a
46 * snapshot. It is mounted and undergoing changes
48 * - The target-fs is a copy of the source-fs from the past. Let's
49 * call this point, the 'sync point'.
51 * - There is a changelog of all metadata operations that happened on
52 * the filesystem since the 'sync point'.
54 * - lustre_rsync replicates all the operations saved in the changelog
55 * on to the target filesystem to make it identical to the source.
57 * To facilitate replication, the lustre filesystem provides
58 * a) a way to get the current filesystem path of a given FID
59 * b) a way to open files by specifying its FID
61 * The changelog only has a limited amount of information.
62 * tfid - The FID of the target file
63 * pfid - The FID of the parent of the target file (at the time of
65 * sfid - The FID of the source file
66 * spfid - The FID of the parent of the source file
67 * name - The name of the target file (at the time of the operation), the name
68 * of the source file is appended (delimited with '\0') if this
69 * operation involves a source
71 * With just this information, it is not alwasy possible to determine
72 * the file paths for each operation. For instance, if pfid does not
73 * exist on the source-fs (due to a subsequent deletion), its path
74 * cannot be queried. In such cases, lustre_rsync keeps the files in a
75 * special directory ("/.lustrerepl"). Once all the operations in a
76 * changelog are replayed, all the files in this special directory
77 * will get moved to the location as in the source-fs.
79 * Shorthand used: f2p(fid) = fid2path(fid)
81 * The following are the metadata operations of interest.
83 * If tfid is absent on the source-fs, ignore this operation
84 * If pfid is absent on the source-fs [or]
85 * if f2p(pfid) is not present on target-fs [or]
86 * if f2p(pfid)+name != f2p(tfid)
87 * creat .lustrerepl/tfid
88 * track [pfid,tfid,name]
93 * If .lustrerepl/[tfid] is present on the target
94 * rm .lustrerepl/[tfid]
95 * Else if pfid is present on the source-fs,
96 * if f2p(pfid)+name is present,
99 * 3. move (spfid,sname) to (pfid,name)
101 * if spfid is also present, mv (spfid,sname) to (pfid,name)
102 * else mv .lustrerepl/[sfid] to (pfid,name)
103 * Else if pfid is not present,
104 * if spfid is present, mv (spfid,sname) .lustrerepl/[sfid]
105 * If moving out of .lustrerepl
106 * move out all its children in .lustrerepl.
107 * [pfid,tfid,name] tracked from (1) is used for this.
118 #include <sys/stat.h>
119 #include <sys/types.h>
124 #include <sys/xattr.h>
126 #include <libcfs/util/string.h>
127 #include <libcfs/util/parser.h>
128 #include <lustre/lustreapi.h>
129 #include <lustre/lustre_idl.h>
130 #include "lustre_rsync.h"
132 #define REPLICATE_STATUS_VER 1
133 #define CLEAR_INTERVAL 100
134 #define DEFAULT_RSYNC_THRESHOLD 0xA00000 /* 10 MB */
136 #define TYPE_STR_LEN 16
138 #define DEFAULT_MDT "-MDT0000"
139 #define SPECIAL_DIR ".lustrerepl"
140 #define RSYNC "rsync"
147 /* Not used; declared for fulfilling obd.c's dependency. */
148 command_t cmdlist[0];
149 extern int obd_initialize(int argc, char **argv);
151 /* Information for processing a changelog record. This structure is
152 allocated on the heap instead of allocating large variables on the
157 unsigned int is_extended:1;
158 enum changelog_rec_type type;
159 char tfid[LR_FID_STR_LEN];
160 char pfid[LR_FID_STR_LEN];
161 char sfid[LR_FID_STR_LEN];
162 char spfid[LR_FID_STR_LEN];
163 char sname[NAME_MAX + 1];
164 char name[NAME_MAX + 1];
165 char src[PATH_MAX + 1];
166 char dest[PATH_MAX + 1];
167 char path[PATH_MAX + 1];
168 char savedpath[PATH_MAX + 1];
169 char link[PATH_MAX + 1];
170 char linktmp[PATH_MAX + 1];
175 /* Variables for querying the xattributes */
182 struct lr_parent_child_list {
183 struct lr_parent_child_log pc_log;
184 struct lr_parent_child_list *pc_next;
187 struct lustre_rsync_status *status;
188 char *statuslog; /* Name of the status log file */
190 int noxattr; /* Flag to turn off replicating xattrs */
191 int noclear; /* Flag to turn off clearing changelogs */
192 int debug; /* Flag to turn debugging information on and off */
193 int verbose; /* Verbose output */
194 long long rec_count; /* No of changelog records that were processed */
197 int use_rsync; /* Flag to turn on use of rsync to copy data */
198 long long rsync_threshold = DEFAULT_RSYNC_THRESHOLD;
199 int quit; /* Flag to stop processing the changelog; set on the
200 receipt of a signal */
201 int abort_on_err = 0;
203 char rsync[PATH_MAX];
204 char rsync_ver[PATH_MAX];
205 struct lr_parent_child_list *parents;
209 /* Command line options */
210 struct option long_opts[] = {
211 {"source", required_argument, 0, 's'},
212 {"target", required_argument, 0, 't'},
213 {"mdt", required_argument, 0, 'm'},
214 {"user", required_argument, 0, 'u'},
215 {"statuslog", required_argument, 0, 'l'},
216 {"verbose", no_argument, 0, 'v'},
217 {"xattr", required_argument, 0, 'x'},
218 {"dry-run", no_argument, 0, 'z'},
219 /* Undocumented options follow */
220 {"cl-clear", required_argument, 0, 'c'},
221 {"use-rsync", no_argument, 0, 'r'},
222 {"rsync-threshold", required_argument, 0, 'y'},
223 {"start-recno", required_argument, 0, 'n'},
224 {"abort-on-err",no_argument, 0, 'a'},
225 {"debug", required_argument, 0, 'd'},
226 {"debuglog", required_argument, 0, 'D'},
230 /* Command line usage */
233 fprintf(stderr, "\tlustre_rsync -s <lustre_root_path> -t <target_path> "
234 "-m <mdt> -r <user id> -l <status log>\n"
235 "lustre_rsync can also pick up parameters from a "
236 "status log created earlier.\n"
237 "\tlustre_rsync -l <log_file>\n"
239 "\t--xattr <yes|no> replicate EAs\n"
240 "\t--abort-on-err abort at first err\n"
242 "\t--dry-run don't write anything\n");
245 #define DEBUG_ENTRY(info) \
246 lr_debug(D_TRACE, "***** Start %lld %s (%d) %s %s %s *****\n", \
247 (info)->recno, changelog_type2str((info)->type), \
248 (info)->type, (info)->tfid, (info)->pfid, (info)->name);
250 #define DEBUG_EXIT(info, rc) \
251 lr_debug(D_TRACE, "##### End %lld %s (%d) %s %s %s rc=%d #####\n", \
252 (info)->recno, changelog_type2str((info)->type), \
253 (info)->type, (info)->tfid, (info)->pfid, (info)->name, rc);
255 /* Print debug information. This is controlled by the value of the
256 global variable 'debug' */
257 void lr_debug(int level, const char *fmt, ...)
265 if (debug_log != NULL)
266 vfprintf(debug_log, fmt, ap);
268 vfprintf(stdout, fmt, ap);
273 void * lr_grow_buf(void *buf, int size)
277 ptr = realloc(buf, size);
284 /* Use rsync to replicate file data */
285 int lr_rsync_data(struct lr_info *info)
288 struct stat st_src, st_dest;
291 lr_debug(DTRACE, "Syncing data%s\n", info->tfid);
293 rc = stat(info->src, &st_src);
295 fprintf(stderr, "Error: Unable to stat src=%s %s\n",
296 info->src, info->name);
302 rc = stat(info->dest, &st_dest);
304 fprintf(stderr, "Error: Unable to stat dest=%s\n",
309 if (st_src.st_mtime != st_dest.st_mtime ||
310 st_src.st_size != st_dest.st_size) {
311 /* XXX spawning off an rsync for every data sync and
312 * waiting synchronously is bad for performance.
313 * librsync could possibly used here. But it does not
314 * seem to be of production grade. Multi-threaded
315 * replication is also to be considered.
318 snprintf(cmd, PATH_MAX, "%s --inplace %s %s", rsync, info->src,
320 lr_debug(DTRACE, "\t%s %s\n", cmd, info->tfid);
321 status = system(cmd);
324 } else if (WIFEXITED(status)) {
325 status = WEXITSTATUS(status);
328 else if (status == 23 || status == 24)
329 /* Error due to vanished source files;
335 lr_debug(DINFO, "rsync %s exited with %d %d\n",
336 info->src, status, rc);
341 lr_debug(DTRACE, "Not syncing %s and %s %s\n", info->src,
342 info->dest, info->tfid);
348 int lr_copy_data(struct lr_info *info)
358 fd_src = open(info->src, O_RDONLY);
361 if (fstat(fd_src, &st_src) == -1 ||
362 stat(info->dest, &st_dest) == -1)
365 if (st_src.st_mtime == st_dest.st_mtime &&
366 st_src.st_size == st_dest.st_size)
369 if (st_src.st_size > rsync_threshold && rsync[0] != '\0') {
370 /* It is more efficient to use rsync to replicate
371 large files. Any file larger than rsync_threshold
372 is handed off to rsync. */
373 lr_debug(DTRACE, "Using rsync to replicate %s\n", info->tfid);
374 rc = lr_rsync_data(info);
378 fd_dest = open(info->dest, O_WRONLY | O_TRUNC, st_src.st_mode);
383 bufsize = st_dest.st_blksize;
385 if (info->bufsize < bufsize) {
387 info->buf = lr_grow_buf(info->buf, bufsize);
388 if (info->buf == NULL) {
392 info->bufsize = bufsize;
400 rsize = read(fd_src, buf, bufsize);
410 wsize = write(fd_dest, buf, rsize);
430 /* Copy data from source to destination */
431 int lr_sync_data(struct lr_info *info)
434 return lr_rsync_data(info);
436 return lr_copy_data(info);
439 /* Copy all attributes from file src to file dest */
440 int lr_copy_attr(char *src, char *dest)
445 if (stat(src, &st) == -1 ||
446 chmod(dest, st.st_mode) == -1 ||
447 chown(dest, st.st_uid, st.st_gid) == -1)
450 time.actime = st.st_atime;
451 time.modtime = st.st_mtime;
452 if (utime(dest, &time) == -1)
457 /* Copy all xattrs from file info->src to info->dest */
458 int lr_copy_xattr(struct lr_info *info)
460 size_t size = info->xsize;
469 rc = llistxattr(info->src, info->xlist, size);
470 lr_debug(DTRACE, "llistxattr(%s,%p) returned %d, errno=%d\n",
471 info->src, info->xlist, rc, errno);
472 if ((rc > 0 && info->xlist == NULL) || errno == ERANGE) {
473 size = rc > PATH_MAX ? rc : PATH_MAX;
474 info->xlist = lr_grow_buf(info->xlist, size);
475 if (info->xlist == NULL)
478 rc = llistxattr(info->src, info->xlist, size);
479 lr_debug(DTRACE, "llistxattr %s returned %d, errno=%d\n",
480 info->src, rc, errno);
487 while (start < len) {
489 rc = lgetxattr(info->src, info->xlist + start,
491 if (info->xvalue == NULL || errno == ERANGE) {
492 size = rc > PATH_MAX ? rc : PATH_MAX;
493 info->xvalue = lr_grow_buf(info->xvalue, size);
494 if (info->xvalue == NULL)
497 rc = lgetxattr(info->src, info->xlist + start,
500 lr_debug(DTRACE, "\t(%s,%d) rc=%p\n", info->xlist + start,
504 rc = lsetxattr(info->dest, info->xlist + start,
505 info->xvalue, size, 0);
506 lr_debug(DTRACE, "\tlsetxattr(), rc=%d, errno=%d\n",
509 if (errno != ENOTSUP) {
510 fprintf(stderr, "Error replicating "
511 " xattr for %s: %d\n",
518 start += strlen(info->xlist + start) + 1;
521 lr_debug(DINFO, "setxattr: %s %s\n", info->src, info->dest);
526 /* Retrieve the filesystem path for a given FID and a given
527 linkno. The path is returned in info->path */
528 int lr_get_path_ln(struct lr_info *info, char *fidstr, int linkno)
530 long long recno = -1;
533 rc = llapi_fid2path(status->ls_source, fidstr, info->path,
534 PATH_MAX, &recno, &linkno);
535 if (rc < 0 && rc != -ENOENT) {
536 fprintf(stderr, "fid2path error: (%s, %s) %d %s\n",
537 status->ls_source, fidstr, -rc, strerror(errno = -rc));
543 /* Retrieve the filesystem path for a given FID. The path is returned
545 int lr_get_path(struct lr_info *info, char *fidstr)
547 return lr_get_path_ln(info, fidstr, 0);
550 /* Generate the path for opening by FID */
551 void lr_get_FID_PATH(char *mntpt, char *fidstr, char *buf, int bufsize)
553 /* Open-by-FID path is <mntpt>/.lustre/fid/[SEQ:OID:VER] */
554 snprintf(buf, bufsize, "%s/%s/fid/%s", mntpt, dot_lustre_name,
559 /* Read the symlink information into 'info->link' */
560 int lr_get_symlink(struct lr_info *info)
565 lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
566 rc = readlink(info->src, info->linktmp, PATH_MAX);
568 info->linktmp[rc] = '\0';
571 lr_debug(DTRACE, "symlink: readlink returned %s\n", info->linktmp);
573 if (strncmp(info->linktmp, status->ls_source,
574 strlen(status->ls_source)) == 0) {
575 /* Strip source fs path and replace with target fs path. */
576 link = info->linktmp + strlen(status->ls_source);
577 snprintf(info->src, PATH_MAX, "%s%s",
578 status->ls_targets[info->target_no], link);
581 link = info->linktmp;
583 strlcpy(info->link, link, sizeof(info->link));
588 /* Create file/directory/device file/symlink. */
589 int lr_mkfile(struct lr_info *info)
595 lr_debug(DINFO, "mkfile(%d) %s \n", info->type, info->dest);
596 if (info->type == CL_MKDIR) {
597 rc = mkdir(info->dest, 0777);
598 } else if (info->type == CL_SOFTLINK) {
599 lr_get_symlink(info);
600 rc = symlink(info->link, info->dest);
601 } else if (info->type == CL_MKNOD) {
602 lr_get_FID_PATH(status->ls_source, info->tfid,
603 info->src, PATH_MAX);
604 rc = stat(info->src, &st);
611 rc = mknod(info->dest, st.st_mode, st.st_rdev);
613 rc = mknod(info->dest, S_IFREG | 0777, 0);
618 /* Sync data and attributes */
619 if (info->type == CL_CREATE || info->type == CL_MKDIR) {
620 lr_debug(DTRACE, "Syncing data and attributes %s\n",
622 (void) lr_copy_xattr(info);
623 if (info->type == CL_CREATE)
624 rc = lr_sync_data(info);
626 rc = lr_copy_attr(info->src, info->dest);
629 /* Source file has disappeared. Not an error. */
632 lr_debug(DTRACE, "Not syncing data and attributes %s\n",
639 int lr_add_pc(const char *pfid, const char *tfid, const char *name)
641 struct lr_parent_child_list *p;
644 p = calloc(1, sizeof(*p));
647 len = strlcpy(p->pc_log.pcl_pfid, pfid, sizeof(p->pc_log.pcl_pfid));
648 if (len >= sizeof(p->pc_log.pcl_pfid))
650 len = strlcpy(p->pc_log.pcl_tfid, tfid, sizeof(p->pc_log.pcl_tfid));
651 if (len >= sizeof(p->pc_log.pcl_tfid))
653 len = strlcpy(p->pc_log.pcl_name, name, sizeof(p->pc_log.pcl_name));
654 if (len >= sizeof(p->pc_log.pcl_name))
657 p->pc_next = parents;
666 void lr_cascade_move(const char *fid, const char *dest, struct lr_info *info)
668 struct lr_parent_child_list *curr, *prev;
672 d = calloc(1, PATH_MAX + 1);
673 prev = curr = parents;
675 if (strcmp(curr->pc_log.pcl_pfid, fid) == 0) {
676 snprintf(d, PATH_MAX, "%s/%s", dest,
677 curr->pc_log.pcl_name);
678 snprintf(info->src, PATH_MAX, "%s/%s/%s",
679 status->ls_targets[info->target_no],
680 SPECIAL_DIR, curr->pc_log.pcl_tfid);
681 rc = rename(info->src, d);
683 fprintf(stderr, "Error renaming file "
685 info->src, d, errno);
689 parents = curr->pc_next;
691 prev->pc_next = curr->pc_next;
692 lr_cascade_move(curr->pc_log.pcl_tfid, d, info);
694 prev = curr = parents;
698 curr = curr->pc_next;
705 /* remove [info->spfid, info->sfid] from parents */
706 int lr_remove_pc(const char *pfid, const char *tfid)
708 struct lr_parent_child_list *curr, *prev;
710 for (prev = curr = parents; curr; prev = curr, curr = curr->pc_next) {
711 if (strcmp(curr->pc_log.pcl_pfid, pfid) == 0 &&
712 strcmp(curr->pc_log.pcl_tfid, tfid) == 0) {
714 parents = curr->pc_next;
716 prev->pc_next = curr->pc_next;
724 /* Create file under SPECIAL_DIR with its tfid as its name. */
725 int lr_mk_special(struct lr_info *info)
729 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
730 status->ls_targets[info->target_no], SPECIAL_DIR,
733 rc = lr_mkfile(info);
737 rc = lr_add_pc(info->pfid, info->tfid, info->name);
741 /* Remove a file or directory */
742 int lr_rmfile(struct lr_info *info)
746 if (info->type == CL_RMDIR)
747 rc = rmdir(info->dest);
749 rc = unlink(info->dest);
755 /* Recursively remove directory and its contents */
756 int lr_rm_recursive(struct lr_info *info)
760 snprintf(info->cmd, PATH_MAX, "rm -rf %s", info->dest);
761 rc = system(info->cmd);
768 /* Remove a file under SPECIAL_DIR with its tfid as its name. */
769 int lr_rm_special(struct lr_info *info)
773 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
774 status->ls_targets[info->target_no], SPECIAL_DIR,
776 rc = lr_rmfile(info);
779 lr_debug(DINFO, "remove: %s; rc=%d, errno=%d\n",
780 info->dest, rc, errno);
784 /* Replicate file and directory create events */
785 int lr_create(struct lr_info *info)
792 /* Is target FID present on the source? */
793 rc = lr_get_path(info, info->tfid);
795 /* Source file has disappeared. Not an error. */
796 lr_debug(DINFO, "create: tfid %s not found on"
797 "source-fs\n", info->tfid);
802 strcpy(info->savedpath, info->path);
804 /* Is parent FID present on the source */
805 rc = lr_get_path(info, info->pfid);
807 lr_debug(DINFO, "create: pfid %s not found on source-fs\n",
814 /* Is f2p(pfid)+name != f2p(tfid)? If not the file has moved. */
815 len = strlen(info->path);
816 if (len == 1 && info->path[0] == '/')
817 snprintf(info->dest, PATH_MAX, "%s", info->name);
818 else if (len - 1 > 0 && info->path[len - 1] == '/')
819 snprintf(info->dest, PATH_MAX, "%s%s", info->path, info->name);
821 snprintf(info->dest, PATH_MAX, "%s/%s", info->path, info->name);
823 lr_debug(DTRACE, "dest = %s; savedpath = %s\n", info->dest,
825 if (strncmp(info->dest, info->savedpath, PATH_MAX) != 0) {
826 lr_debug(DTRACE, "create: file moved (%s). %s != %s\n",
827 info->tfid, info->dest, info->savedpath);
831 /* Is f2p(pfid) present on the target? If not, the parent has
834 snprintf(info->dest, PATH_MAX, "%s/%s", status->ls_targets[0],
836 if (access(info->dest, F_OK) != 0) {
837 lr_debug(DTRACE, "create: parent %s not found\n",
842 for (info->target_no = 0; info->target_no < status->ls_num_targets;
844 snprintf(info->dest, PATH_MAX, "%s/%s",
845 status->ls_targets[info->target_no], info->savedpath);
846 lr_get_FID_PATH(status->ls_source, info->tfid, info->src,
850 rc1 = lr_mkfile(info);
851 if (mkspecial || rc1 == -ENOENT) {
852 rc1 = lr_mk_special(info);
860 /* Replicate a file remove (rmdir/unlink) operation */
861 int lr_remove(struct lr_info *info)
866 for (info->target_no = 0; info->target_no < status->ls_num_targets;
869 rc1 = lr_rm_special(info);
873 rc1 = lr_get_path(info, info->pfid);
874 if (rc1 == -ENOENT) {
875 lr_debug(DINFO, "remove: pfid %s not found\n",
883 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
884 status->ls_targets[info->target_no], info->path,
887 rc1 = lr_rmfile(info);
888 lr_debug(DINFO, "remove: %s; rc1=%d, errno=%d\n",
889 info->dest, rc1, errno);
890 if (rc1 == -ENOTEMPTY)
891 rc1 = lr_rm_recursive(info);
901 /* Replicate a rename/move operation. */
902 int lr_move(struct lr_info *info)
908 int special_dest = 0;
909 char srcpath[PATH_MAX + 1] = "";
911 LASSERT(info->is_extended);
913 rc_src = lr_get_path(info, info->spfid);
914 if (rc_src < 0 && rc_src != -ENOENT)
916 memcpy(srcpath, info->path, strlen(info->path));
918 rc_dest = lr_get_path(info, info->pfid);
919 if (rc_dest < 0 && rc_dest != -ENOENT)
922 for (info->target_no = 0; info->target_no < status->ls_num_targets;
926 snprintf(info->dest, PATH_MAX, "%s/%s",
927 status->ls_targets[info->target_no],
929 if (access(info->dest, F_OK) != 0) {
932 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
933 status->ls_targets[info->target_no],
934 info->path, info->name);
936 lr_debug(DINFO, "dest path %s rc_dest=%d\n", info->dest,
939 if (rc_dest == -ENOENT) {
940 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
941 status->ls_targets[info->target_no],
942 SPECIAL_DIR, info->sfid);
944 lr_debug(DINFO, "special dest %s\n", info->dest);
948 snprintf(info->src, PATH_MAX, "%s/%s/%s",
949 status->ls_targets[info->target_no],
950 srcpath, info->sname);
951 lr_debug(DINFO, "src path %s rc_src=%d\n", info->src,
954 if (rc_src == -ENOENT || (access(info->src, F_OK) != 0 &&
956 snprintf(info->src, PATH_MAX, "%s/%s/%s",
957 status->ls_targets[info->target_no],
958 SPECIAL_DIR, info->sfid);
960 lr_debug(DINFO, "special src %s\n", info->src);
965 if (strcmp(info->src, info->dest) != 0) {
966 rc1 = rename(info->src, info->dest);
969 lr_debug(DINFO, "rename returns %d\n", rc1);
973 rc1 = lr_remove_pc(info->spfid, info->sfid);
976 lr_cascade_move(info->sfid, info->dest, info);
978 rc1 = lr_add_pc(info->pfid, info->sfid, info->name);
980 lr_debug(DINFO, "move: %s [to] %s rc1=%d, errno=%d\n",
981 info->src, info->dest, rc1, errno);
988 /* Replicate a hard link */
989 int lr_link(struct lr_info *info)
996 lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
997 rc = stat(info->src, &st);
1001 for (info->target_no = 0; info->target_no < status->ls_num_targets;
1002 info->target_no++) {
1009 * The changelog record has the new parent directory FID and
1010 * name of the target file. So info->dest can be constructed
1011 * by getting the path of the new parent directory and
1012 * appending the target file name.
1014 rc1 = lr_get_path(info, info->pfid);
1015 lr_debug(rc1 ? 0 : DTRACE, "\tparent fid2path %s, %s, rc=%d\n",
1016 info->path, info->name, rc1);
1019 snprintf(info->dest, sizeof(info->dest), "%s/%s/%s",
1020 status->ls_targets[info->target_no],
1021 info->path, info->name);
1022 lr_debug(DINFO, "link destination is %s\n", info->dest);
1025 /* Search through the hardlinks to get the src */
1026 for (i = 0; i < st.st_nlink && info->src[0] == 0; i++) {
1027 rc1 = lr_get_path_ln(info, info->tfid, i);
1028 lr_debug(rc1 ? 0:DTRACE, "\tfid2path %s, %s, %d rc=%d\n",
1029 info->path, info->name, i, rc1);
1034 * Compare the path of target FID with info->dest
1035 * to find out info->src.
1037 char srcpath[PATH_MAX];
1039 snprintf(srcpath, sizeof(srcpath), "%s/%s",
1040 status->ls_targets[info->target_no],
1043 if (strcmp(srcpath, info->dest) != 0) {
1044 strlcpy(info->src, srcpath, sizeof(info->src));
1045 lr_debug(DINFO, "link source is %s\n",
1055 if (info->src[0] == 0)
1056 snprintf(info->src, PATH_MAX, "%s/%s/%s",
1057 status->ls_targets[info->target_no],
1058 SPECIAL_DIR, info->tfid);
1059 else if (info->dest[0] == 0)
1060 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
1061 status->ls_targets[info->target_no],
1062 SPECIAL_DIR, info->tfid);
1064 rc1 = link(info->src, info->dest);
1065 lr_debug(DINFO, "link: %s [to] %s; rc1=%d %s\n",
1066 info->src, info->dest, rc1,
1067 strerror(rc1 ? errno : 0));
1075 /* Replicate file attributes */
1076 int lr_setattr(struct lr_info *info)
1081 lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
1083 rc = lr_get_path(info, info->tfid);
1085 lr_debug(DINFO, "setattr: %s not present on source-fs\n",
1090 for (info->target_no = 0; info->target_no < status->ls_num_targets;
1091 info->target_no++) {
1093 snprintf(info->dest, PATH_MAX, "%s/%s",
1094 status->ls_targets[info->target_no], info->path);
1095 lr_debug(DINFO, "setattr: %s %s %s", info->src, info->dest,
1098 rc1 = lr_sync_data(info);
1100 rc1 = lr_copy_attr(info->src, info->dest);
1107 /* Replicate xattrs */
1108 int lr_setxattr(struct lr_info *info)
1112 lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
1114 rc = lr_get_path(info, info->tfid);
1116 lr_debug(DINFO, "setxattr: %s not present on source-fs\n",
1121 for (info->target_no = 0; info->target_no < status->ls_num_targets;
1122 info->target_no++) {
1124 snprintf(info->dest, PATH_MAX, "%s/%s",
1125 status->ls_targets[info->target_no], info->path);
1126 lr_debug(DINFO, "setxattr: %s %s %s\n", info->src, info->dest,
1129 rc1 = lr_copy_xattr(info);
1137 /* Parse a line of changelog entry */
1138 int lr_parse_line(void *priv, struct lr_info *info)
1140 struct changelog_rec *rec;
1141 struct changelog_ext_rename *rnm;
1143 size_t copylen = sizeof(info->name);
1145 if (llapi_changelog_recv(priv, &rec) != 0)
1148 info->is_extended = !!(rec->cr_flags & CLF_RENAME);
1149 info->recno = rec->cr_index;
1150 info->type = rec->cr_type;
1151 snprintf(info->tfid, sizeof(info->tfid), DFID, PFID(&rec->cr_tfid));
1152 snprintf(info->pfid, sizeof(info->pfid), DFID, PFID(&rec->cr_pfid));
1154 namelen = strnlen(changelog_rec_name(rec), rec->cr_namelen);
1155 if (copylen > namelen + 1)
1156 copylen = namelen + 1;
1157 strlcpy(info->name, changelog_rec_name(rec), copylen);
1159 /* Don't use rnm if CLF_RENAME isn't set */
1160 rnm = changelog_rec_rename(rec);
1161 if (rec->cr_flags & CLF_RENAME && !fid_is_zero(&rnm->cr_sfid)) {
1162 copylen = sizeof(info->sname);
1164 snprintf(info->sfid, sizeof(info->sfid), DFID,
1165 PFID(&rnm->cr_sfid));
1166 snprintf(info->spfid, sizeof(info->spfid), DFID,
1167 PFID(&rnm->cr_spfid));
1168 namelen = changelog_rec_snamelen(rec);
1169 if (copylen > namelen + 1)
1170 copylen = namelen + 1;
1171 strlcpy(info->sname, changelog_rec_sname(rec), copylen);
1174 printf("Rec %lld: %d %s %s\n", info->recno, info->type,
1175 info->name, info->sname);
1178 printf("Rec %lld: %d %s\n", info->recno, info->type,
1182 llapi_changelog_free(&rec);
1188 /* Initialize the replication parameters */
1189 int lr_init_status()
1191 size_t size = sizeof(struct lustre_rsync_status) + PATH_MAX + 1;
1195 status = calloc(size, 1);
1198 status->ls_version = REPLICATE_STATUS_VER;
1199 status->ls_size = size;
1200 status->ls_last_recno = -1;
1204 /* Make a backup of the statuslog */
1205 void lr_backup_log()
1207 char backupfile[PATH_MAX];
1211 snprintf(backupfile, PATH_MAX, "%s.old", statuslog);
1212 (void) rename(statuslog, backupfile);
1218 /* Save replication parameters to a statuslog. */
1223 size_t write_size = status->ls_size;
1224 struct lr_parent_child_list *curr;
1227 if (statuslog == NULL)
1232 fd = open(statuslog, O_WRONLY | O_CREAT | O_SYNC,
1235 fprintf(stderr, "Error opening log file for writing (%s)\n",
1240 size = write(fd, status, write_size);
1241 if (size != write_size) {
1242 fprintf(stderr, "Error writing to log file (%s) %d\n",
1248 for (curr = parents; curr; curr = curr->pc_next) {
1249 size = write(fd, &curr->pc_log, sizeof(curr->pc_log));
1250 if (size != sizeof(curr->pc_log)) {
1251 fprintf(stderr, "Error writing to log file (%s) %d\n",
1261 /* Read statuslog and populate the replication parameters. Command
1262 * line parameters take precedence over parameters in the log file.*/
1265 struct lr_parent_child_list *tmp;
1266 struct lr_parent_child_log rec;
1267 struct lustre_rsync_status *s;
1270 size_t read_size = sizeof(struct lustre_rsync_status) + PATH_MAX + 1;
1273 if (statuslog == NULL)
1276 s = calloc(1, read_size);
1282 fd = open(statuslog, O_RDONLY);
1288 size = read(fd, s, read_size);
1289 if (size != read_size) {
1294 if (read_size < s->ls_size) {
1295 read_size = s->ls_size;
1296 s = lr_grow_buf(s, read_size);
1302 if (lseek(fd, 0, SEEK_SET) == -1) {
1307 size = read(fd, s, read_size);
1308 if (size != read_size) {
1314 while (read(fd, &rec, sizeof(rec)) != 0) {
1315 tmp = calloc(1, sizeof(*tmp));
1322 tmp->pc_next = parents;
1326 /* copy uninitialized fields to status */
1327 if (status->ls_num_targets == 0) {
1328 if (status->ls_size != s->ls_size) {
1329 status = lr_grow_buf(status, s->ls_size);
1330 if (status == NULL) {
1335 status->ls_size = s->ls_size;
1337 status->ls_num_targets = s->ls_num_targets;
1338 memcpy(status->ls_targets, s->ls_targets,
1339 (PATH_MAX + 1) * s->ls_num_targets);
1341 if (status->ls_last_recno == -1)
1342 status->ls_last_recno = s->ls_last_recno;
1344 if (status->ls_registration[0] == '\0')
1345 strlcpy(status->ls_registration, s->ls_registration,
1346 sizeof(status->ls_registration));
1348 if (status->ls_mdt_device[0] == '\0')
1349 strlcpy(status->ls_mdt_device, s->ls_mdt_device,
1350 sizeof(status->ls_mdt_device));
1352 if (status->ls_source_fs[0] == '\0')
1353 strlcpy(status->ls_source_fs, s->ls_source_fs,
1354 sizeof(status->ls_source_fs));
1356 if (status->ls_source[0] == '\0')
1357 strlcpy(status->ls_source, s->ls_source,
1358 sizeof(status->ls_source));
1368 /* Clear changelogs every CLEAR_INTERVAL records or at the end of
1370 int lr_clear_cl(struct lr_info *info, int force)
1372 char mdt_device[LR_NAME_MAXLEN + 1];
1376 if (force || info->recno > status->ls_last_recno + CLEAR_INTERVAL) {
1377 if (info->type == CL_RENAME)
1378 rec = info->recno + 1;
1381 if (!noclear && !dryrun) {
1382 /* llapi_changelog_clear modifies the mdt
1383 * device name so make a copy of it until this
1386 strlcpy(mdt_device, status->ls_mdt_device,
1387 sizeof(mdt_device));
1388 rc = llapi_changelog_clear(mdt_device,
1389 status->ls_registration,
1392 printf("Changelog clear (%s, %s, %lld) "
1393 "returned %d\n", status->ls_mdt_device,
1394 status->ls_registration, rec, rc);
1396 if (!rc && !dryrun) {
1397 status->ls_last_recno = rec;
1406 /* Locate a usable version of rsync. At this point we'll use any
1408 int lr_locate_rsync()
1414 snprintf(rsync, PATH_MAX, "%s -p %s", TYPE, RSYNC);
1415 fp = popen(rsync, "r");
1419 if (fgets(rsync, PATH_MAX, fp) == NULL) {
1424 len = strlen(rsync);
1425 if (len > 0 && rsync[len - 1] == '\n')
1426 rsync[len - 1] = '\0';
1429 /* Determine the version of rsync */
1430 snprintf(rsync_ver, PATH_MAX, "%s --version", rsync);
1431 fp = popen(rsync_ver, "r");
1435 if (fgets(rsync_ver, PATH_MAX, fp) == NULL) {
1439 len = strlen(rsync_ver);
1440 if (len > 0 && rsync_ver[len - 1] == '\n')
1441 rsync_ver[len - 1] = '\0';
1448 /* Print the replication parameters */
1449 void lr_print_status(struct lr_info *info)
1456 printf("Lustre filesystem: %s\n", status->ls_source_fs);
1457 printf("MDT device: %s\n", status->ls_mdt_device);
1458 printf("Source: %s\n", status->ls_source);
1459 for (i = 0; i < status->ls_num_targets; i++)
1460 printf("Target: %s\n", status->ls_targets[i]);
1461 if (statuslog != NULL)
1462 printf("Statuslog: %s\n", statuslog);
1463 printf("Changelog registration: %s\n", status->ls_registration);
1464 printf("Starting changelog record: %jd\n",
1465 (uintmax_t)status->ls_last_recno);
1467 printf("Replicate xattrs: no\n");
1469 printf("Clear changelog after use: no\n");
1471 printf("Using rsync: %s (%s)\n", rsync, rsync_ver);
1474 void lr_print_failure(struct lr_info *info, int rc)
1476 fprintf(stderr, "Replication of operation failed(%d):"
1477 " %lld %s (%d) %s %s %s\n", rc, info->recno,
1478 changelog_type2str(info->type), info->type, info->tfid,
1479 info->pfid, info->name);
1482 /* Replicate filesystem operations from src_path to target_path */
1485 void *changelog_priv;
1486 struct lr_info *info;
1487 struct lr_info *ext = NULL;
1495 info = calloc(1, sizeof(struct lr_info));
1499 rc = llapi_search_fsname(status->ls_source, status->ls_source_fs);
1501 fprintf(stderr, "Source path is not a valid Lustre client "
1505 if (status->ls_mdt_device[0] == '\0')
1506 snprintf(status->ls_mdt_device, LR_NAME_MAXLEN, "%s%s",
1507 status->ls_source_fs, DEFAULT_MDT);
1509 ext = calloc(1, sizeof(struct lr_info));
1515 for (i = 0, xattr_not_supp = 0; i < status->ls_num_targets; i++) {
1516 snprintf(info->dest, PATH_MAX, "%s/%s", status->ls_targets[i],
1518 rc = mkdir(info->dest, 0777);
1519 if (rc == -1 && errno != EEXIST) {
1520 fprintf(stderr, "Error writing to target path %s.\n",
1521 status->ls_targets[i]);
1525 rc = llistxattr(info->src, info->xlist, info->xsize);
1526 if (rc == -1 && errno == ENOTSUP) {
1527 fprintf(stderr, "xattrs not supported on %s\n",
1528 status->ls_targets[i]);
1532 if (xattr_not_supp == status->ls_num_targets)
1533 /* None of the targets support xattrs. */
1536 lr_print_status(info);
1538 /* Open changelogs for consumption*/
1539 rc = llapi_changelog_start(&changelog_priv,
1540 CHANGELOG_FLAG_BLOCK | CHANGELOG_FLAG_JOBID,
1541 status->ls_mdt_device, status->ls_last_recno);
1543 fprintf(stderr, "Error opening changelog file for fs %s.\n",
1544 status->ls_source_fs);
1548 while (!quit && lr_parse_line(changelog_priv, info) == 0) {
1550 if (info->type == CL_RENAME && !info->is_extended) {
1551 /* Newer rename operations extends changelog to store
1552 * source file information, but old changelog has
1555 if (lr_parse_line(changelog_priv, ext) != 0)
1557 memcpy(info->sfid, info->tfid, sizeof(info->sfid));
1558 memcpy(info->spfid, info->pfid, sizeof(info->spfid));
1559 memcpy(info->tfid, ext->tfid, sizeof(info->tfid));
1560 memcpy(info->pfid, ext->pfid, sizeof(info->pfid));
1561 strlcpy(info->sname, info->name, sizeof(info->sname));
1562 strlcpy(info->name, ext->name, sizeof(info->name));
1563 info->is_extended = 1;
1571 switch(info->type) {
1576 rc = lr_create(info);
1580 rc = lr_remove(info);
1590 rc = lr_setattr(info);
1593 rc = lr_setxattr(info);
1600 /* Nothing needs to be done for these entries */
1605 DEBUG_EXIT(info, rc);
1606 if (rc && rc != -ENOENT) {
1607 lr_print_failure(info, rc);
1612 lr_clear_cl(info, 0);
1614 bzero(info, sizeof(struct lr_info));
1615 bzero(ext, sizeof(struct lr_info));
1619 llapi_changelog_fini(&changelog_priv);
1621 if (errors || verbose)
1622 printf("Errors: %d\n", errors);
1624 /* Clear changelog records used so far */
1625 lr_clear_cl(info, 1);
1628 printf("lustre_rsync took %ld seconds\n", time(NULL) - start);
1629 printf("Changelog records consumed: %lld\n", rec_count);
1644 termination_handler (int signum)
1646 /* Set a flag for the replicator to gracefully shutdown */
1648 printf("lustre_rsync halting.\n");
1651 int main(int argc, char *argv[])
1657 if ((rc = lr_init_status()) != 0)
1660 while ((rc = getopt_long(argc, argv, "as:t:m:u:l:vx:zc:ry:n:d:D:",
1661 long_opts, NULL)) >= 0) {
1664 /* Assume absolute paths */
1668 /* Assume absolute paths */
1669 strlcpy(status->ls_source, optarg,
1670 sizeof(status->ls_source));
1673 status->ls_num_targets++;
1675 if (numtargets != status->ls_num_targets) {
1676 /* Targets were read from a log
1677 file. The ones specified on the
1678 command line take precedence. The
1679 ones from the log file will be
1681 status->ls_num_targets = numtargets;
1683 newsize = sizeof (struct lustre_rsync_status) +
1684 (status->ls_num_targets * (PATH_MAX + 1));
1685 if (status->ls_size != newsize) {
1686 status->ls_size = newsize;
1687 status = lr_grow_buf(status, newsize);
1691 strlcpy(status->ls_targets[status->ls_num_targets - 1],
1692 optarg, sizeof(status->ls_targets[0]));
1695 strlcpy(status->ls_mdt_device, optarg,
1696 sizeof(status->ls_mdt_device));
1699 strlcpy(status->ls_registration, optarg,
1700 sizeof(status->ls_registration));
1704 (void) lr_read_log();
1710 if (strcmp("no", optarg) == 0) {
1712 } else if (strcmp("yes", optarg) != 0) {
1713 printf("Invalid parameter %s. "
1714 "Specify --xattr=no or --xattr=yes\n",
1723 /* Undocumented option cl-clear */
1724 if (strcmp("no", optarg) == 0) {
1726 } else if (strcmp("yes", optarg) != 0) {
1727 printf("Invalid parameter %s. "
1728 "Specify --cl-clear=no "
1729 "or --cl-clear=yes\n",
1735 /* Undocumented option use-rsync */
1739 /* Undocumented option rsync-threshold */
1740 rsync_threshold = atol(optarg);
1743 /* Undocumented option start-recno */
1744 status->ls_last_recno = atol(optarg);
1747 /* Undocumented option debug */
1748 debug = atoi(optarg);
1749 if (debug < 0 || debug > 2)
1753 /* Undocumented option debug log file */
1754 if (debug_log != NULL)
1756 debug_log = fopen(optarg, "a");
1757 if (debug_log == NULL) {
1758 printf("Cannot open %s for debug log\n",
1764 fprintf(stderr, "error: %s: option '%s' "
1765 "unrecognized.\n", argv[0], argv[optind - 1]);
1771 if (status->ls_last_recno == -1)
1772 status->ls_last_recno = 0;
1773 if (strnlen(status->ls_registration, LR_NAME_MAXLEN) == 0) {
1774 /* No registration ID was passed in. */
1775 printf("Please specify changelog consumer registration id.\n");
1779 if (strnlen(status->ls_source, PATH_MAX) == 0) {
1780 fprintf(stderr, "Please specify the source path.\n");
1784 if (strnlen(status->ls_targets[0], PATH_MAX) == 0) {
1785 fprintf(stderr, "Please specify the target path.\n");
1790 /* This plumbing is needed for some of the ioctls behind
1791 llapi calls to work. */
1792 if (obd_initialize(argc, argv) < 0) {
1793 fprintf(stderr, "obd_initialize failed.\n");
1797 rc = lr_locate_rsync();
1798 if (use_rsync && rc != 0) {
1799 fprintf(stderr, "Error: unable to locate %s.\n", RSYNC);
1803 signal(SIGINT, termination_handler);
1804 signal(SIGHUP, termination_handler);
1805 signal(SIGTERM, termination_handler);
1807 rc = lr_replicate();
1809 if (debug_log != NULL)