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);
623 /* Sync data and attributes */
624 if (info->type == CL_CREATE || info->type == CL_MKDIR) {
625 lr_debug(DTRACE, "Syncing data and attributes %s\n",
627 (void) lr_copy_xattr(info);
628 if (info->type == CL_CREATE)
629 rc = lr_sync_data(info);
631 rc = lr_copy_attr(info->src, info->dest);
634 /* Source file has disappeared. Not an error. */
637 lr_debug(DTRACE, "Not syncing data and attributes %s\n",
644 int lr_add_pc(const char *pfid, const char *tfid, const char *name)
646 struct lr_parent_child_list *p;
649 p = calloc(1, sizeof(*p));
652 len = strlcpy(p->pc_log.pcl_pfid, pfid, sizeof(p->pc_log.pcl_pfid));
653 if (len >= sizeof(p->pc_log.pcl_pfid))
655 len = strlcpy(p->pc_log.pcl_tfid, tfid, sizeof(p->pc_log.pcl_tfid));
656 if (len >= sizeof(p->pc_log.pcl_tfid))
658 len = strlcpy(p->pc_log.pcl_name, name, sizeof(p->pc_log.pcl_name));
659 if (len >= sizeof(p->pc_log.pcl_name))
662 p->pc_next = parents;
671 void lr_cascade_move(const char *fid, const char *dest, struct lr_info *info)
673 struct lr_parent_child_list *curr, *prev;
677 d = calloc(1, PATH_MAX + 1);
678 prev = curr = parents;
680 if (strcmp(curr->pc_log.pcl_pfid, fid) == 0) {
681 snprintf(d, PATH_MAX, "%s/%s", dest,
682 curr->pc_log.pcl_name);
683 snprintf(info->src, PATH_MAX, "%s/%s/%s",
684 status->ls_targets[info->target_no],
685 SPECIAL_DIR, curr->pc_log.pcl_tfid);
686 rc = rename(info->src, d);
688 fprintf(stderr, "Error renaming file "
690 info->src, d, errno);
694 parents = curr->pc_next;
696 prev->pc_next = curr->pc_next;
697 lr_cascade_move(curr->pc_log.pcl_tfid, d, info);
699 prev = curr = parents;
703 curr = curr->pc_next;
710 /* remove [info->spfid, info->sfid] from parents */
711 int lr_remove_pc(const char *pfid, const char *tfid)
713 struct lr_parent_child_list *curr, *prev;
715 for (prev = curr = parents; curr; prev = curr, curr = curr->pc_next) {
716 if (strcmp(curr->pc_log.pcl_pfid, pfid) == 0 &&
717 strcmp(curr->pc_log.pcl_tfid, tfid) == 0) {
719 parents = curr->pc_next;
721 prev->pc_next = curr->pc_next;
729 /* Create file under SPECIAL_DIR with its tfid as its name. */
730 int lr_mk_special(struct lr_info *info)
734 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
735 status->ls_targets[info->target_no], SPECIAL_DIR,
738 rc = lr_mkfile(info);
742 rc = lr_add_pc(info->pfid, info->tfid, info->name);
746 /* Remove a file or directory */
747 int lr_rmfile(struct lr_info *info)
751 if (info->type == CL_RMDIR)
752 rc = rmdir(info->dest);
754 rc = unlink(info->dest);
760 /* Recursively remove directory and its contents */
761 int lr_rm_recursive(struct lr_info *info)
765 snprintf(info->cmd, PATH_MAX, "rm -rf %s", info->dest);
766 rc = system(info->cmd);
773 /* Remove a file under SPECIAL_DIR with its tfid as its name. */
774 int lr_rm_special(struct lr_info *info)
778 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
779 status->ls_targets[info->target_no], SPECIAL_DIR,
781 rc = lr_rmfile(info);
784 lr_debug(DINFO, "remove: %s; rc=%d, errno=%d\n",
785 info->dest, rc, errno);
789 /* Replicate file and directory create events */
790 int lr_create(struct lr_info *info)
797 /* Is target FID present on the source? */
798 rc = lr_get_path(info, info->tfid);
800 /* Source file has disappeared. Not an error. */
801 lr_debug(DINFO, "create: tfid %s not found on"
802 "source-fs\n", info->tfid);
807 strcpy(info->savedpath, info->path);
809 /* Is parent FID present on the source */
810 rc = lr_get_path(info, info->pfid);
812 lr_debug(DINFO, "create: pfid %s not found on source-fs\n",
819 /* Is f2p(pfid)+name != f2p(tfid)? If not the file has moved. */
820 len = strlen(info->path);
821 if (len == 1 && info->path[0] == '/')
822 snprintf(info->dest, PATH_MAX, "%s", info->name);
823 else if (len - 1 > 0 && info->path[len - 1] == '/')
824 snprintf(info->dest, PATH_MAX, "%s%s", info->path, info->name);
826 snprintf(info->dest, PATH_MAX, "%s/%s", info->path, info->name);
828 lr_debug(DTRACE, "dest = %s; savedpath = %s\n", info->dest,
830 if (strncmp(info->dest, info->savedpath, PATH_MAX) != 0) {
831 lr_debug(DTRACE, "create: file moved (%s). %s != %s\n",
832 info->tfid, info->dest, info->savedpath);
836 /* Is f2p(pfid) present on the target? If not, the parent has
839 snprintf(info->dest, PATH_MAX, "%s/%s", status->ls_targets[0],
841 if (access(info->dest, F_OK) != 0) {
842 lr_debug(DTRACE, "create: parent %s not found\n",
847 for (info->target_no = 0; info->target_no < status->ls_num_targets;
849 snprintf(info->dest, PATH_MAX, "%s/%s",
850 status->ls_targets[info->target_no], info->savedpath);
851 lr_get_FID_PATH(status->ls_source, info->tfid, info->src,
855 rc1 = lr_mkfile(info);
856 if (mkspecial || rc1 == -ENOENT) {
857 rc1 = lr_mk_special(info);
865 /* Replicate a file remove (rmdir/unlink) operation */
866 int lr_remove(struct lr_info *info)
871 for (info->target_no = 0; info->target_no < status->ls_num_targets;
874 rc1 = lr_rm_special(info);
878 rc1 = lr_get_path(info, info->pfid);
879 if (rc1 == -ENOENT) {
880 lr_debug(DINFO, "remove: pfid %s not found\n",
888 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
889 status->ls_targets[info->target_no], info->path,
892 rc1 = lr_rmfile(info);
893 lr_debug(DINFO, "remove: %s; rc1=%d, errno=%d\n",
894 info->dest, rc1, errno);
895 if (rc1 == -ENOTEMPTY)
896 rc1 = lr_rm_recursive(info);
906 /* Replicate a rename/move operation. */
907 int lr_move(struct lr_info *info)
913 int special_dest = 0;
914 char srcpath[PATH_MAX + 1] = "";
916 LASSERT(info->is_extended);
918 rc_src = lr_get_path(info, info->spfid);
919 if (rc_src < 0 && rc_src != -ENOENT)
921 memcpy(srcpath, info->path, strlen(info->path));
923 rc_dest = lr_get_path(info, info->pfid);
924 if (rc_dest < 0 && rc_dest != -ENOENT)
927 for (info->target_no = 0; info->target_no < status->ls_num_targets;
931 snprintf(info->dest, PATH_MAX, "%s/%s",
932 status->ls_targets[info->target_no],
934 if (access(info->dest, F_OK) != 0) {
937 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
938 status->ls_targets[info->target_no],
939 info->path, info->name);
941 lr_debug(DINFO, "dest path %s rc_dest=%d\n", info->dest,
944 if (rc_dest == -ENOENT) {
945 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
946 status->ls_targets[info->target_no],
947 SPECIAL_DIR, info->sfid);
949 lr_debug(DINFO, "special dest %s\n", info->dest);
953 snprintf(info->src, PATH_MAX, "%s/%s/%s",
954 status->ls_targets[info->target_no],
955 srcpath, info->sname);
956 lr_debug(DINFO, "src path %s rc_src=%d\n", info->src,
959 if (rc_src == -ENOENT || (access(info->src, F_OK) != 0 &&
961 snprintf(info->src, PATH_MAX, "%s/%s/%s",
962 status->ls_targets[info->target_no],
963 SPECIAL_DIR, info->sfid);
965 lr_debug(DINFO, "special src %s\n", info->src);
970 if (strcmp(info->src, info->dest) != 0) {
971 rc1 = rename(info->src, info->dest);
974 lr_debug(DINFO, "rename returns %d\n", rc1);
978 rc1 = lr_remove_pc(info->spfid, info->sfid);
981 lr_cascade_move(info->sfid, info->dest, info);
983 rc1 = lr_add_pc(info->pfid, info->sfid, info->name);
985 lr_debug(DINFO, "move: %s [to] %s rc1=%d, errno=%d\n",
986 info->src, info->dest, rc1, errno);
993 /* Replicate a hard link */
994 int lr_link(struct lr_info *info)
1001 lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
1002 rc = stat(info->src, &st);
1006 for (info->target_no = 0; info->target_no < status->ls_num_targets;
1007 info->target_no++) {
1014 * The changelog record has the new parent directory FID and
1015 * name of the target file. So info->dest can be constructed
1016 * by getting the path of the new parent directory and
1017 * appending the target file name.
1019 rc1 = lr_get_path(info, info->pfid);
1020 lr_debug(rc1 ? 0 : DTRACE, "\tparent fid2path %s, %s, rc=%d\n",
1021 info->path, info->name, rc1);
1024 snprintf(info->dest, sizeof(info->dest), "%s/%s/%s",
1025 status->ls_targets[info->target_no],
1026 info->path, info->name);
1027 lr_debug(DINFO, "link destination is %s\n", info->dest);
1030 /* Search through the hardlinks to get the src */
1031 for (i = 0; i < st.st_nlink && info->src[0] == 0; i++) {
1032 rc1 = lr_get_path_ln(info, info->tfid, i);
1033 lr_debug(rc1 ? 0:DTRACE, "\tfid2path %s, %s, %d rc=%d\n",
1034 info->path, info->name, i, rc1);
1039 * Compare the path of target FID with info->dest
1040 * to find out info->src.
1042 char srcpath[PATH_MAX];
1044 snprintf(srcpath, sizeof(srcpath), "%s/%s",
1045 status->ls_targets[info->target_no],
1048 if (strcmp(srcpath, info->dest) != 0) {
1049 strlcpy(info->src, srcpath, sizeof(info->src));
1050 lr_debug(DINFO, "link source is %s\n",
1060 if (info->src[0] == 0)
1061 snprintf(info->src, PATH_MAX, "%s/%s/%s",
1062 status->ls_targets[info->target_no],
1063 SPECIAL_DIR, info->tfid);
1064 else if (info->dest[0] == 0)
1065 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
1066 status->ls_targets[info->target_no],
1067 SPECIAL_DIR, info->tfid);
1069 rc1 = link(info->src, info->dest);
1070 lr_debug(DINFO, "link: %s [to] %s; rc1=%d %s\n",
1071 info->src, info->dest, rc1,
1072 strerror(rc1 ? errno : 0));
1080 /* Replicate file attributes */
1081 int lr_setattr(struct lr_info *info)
1086 lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
1088 rc = lr_get_path(info, info->tfid);
1090 lr_debug(DINFO, "setattr: %s not present on source-fs\n",
1095 for (info->target_no = 0; info->target_no < status->ls_num_targets;
1096 info->target_no++) {
1098 snprintf(info->dest, PATH_MAX, "%s/%s",
1099 status->ls_targets[info->target_no], info->path);
1100 lr_debug(DINFO, "setattr: %s %s %s", info->src, info->dest,
1103 rc1 = lr_sync_data(info);
1105 rc1 = lr_copy_attr(info->src, info->dest);
1112 /* Replicate xattrs */
1113 int lr_setxattr(struct lr_info *info)
1117 lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
1119 rc = lr_get_path(info, info->tfid);
1121 lr_debug(DINFO, "setxattr: %s not present on source-fs\n",
1126 for (info->target_no = 0; info->target_no < status->ls_num_targets;
1127 info->target_no++) {
1129 snprintf(info->dest, PATH_MAX, "%s/%s",
1130 status->ls_targets[info->target_no], info->path);
1131 lr_debug(DINFO, "setxattr: %s %s %s\n", info->src, info->dest,
1134 rc1 = lr_copy_xattr(info);
1142 /* Parse a line of changelog entry */
1143 int lr_parse_line(void *priv, struct lr_info *info)
1145 struct changelog_rec *rec;
1146 struct changelog_ext_rename *rnm;
1148 size_t copylen = sizeof(info->name);
1150 if (llapi_changelog_recv(priv, &rec) != 0)
1153 info->is_extended = !!(rec->cr_flags & CLF_RENAME);
1154 info->recno = rec->cr_index;
1155 info->type = rec->cr_type;
1156 snprintf(info->tfid, sizeof(info->tfid), DFID, PFID(&rec->cr_tfid));
1157 snprintf(info->pfid, sizeof(info->pfid), DFID, PFID(&rec->cr_pfid));
1159 namelen = strnlen(changelog_rec_name(rec), rec->cr_namelen);
1160 if (copylen > namelen + 1)
1161 copylen = namelen + 1;
1162 strlcpy(info->name, changelog_rec_name(rec), copylen);
1164 /* Don't use rnm if CLF_RENAME isn't set */
1165 rnm = changelog_rec_rename(rec);
1166 if (rec->cr_flags & CLF_RENAME && !fid_is_zero(&rnm->cr_sfid)) {
1167 copylen = sizeof(info->sname);
1169 snprintf(info->sfid, sizeof(info->sfid), DFID,
1170 PFID(&rnm->cr_sfid));
1171 snprintf(info->spfid, sizeof(info->spfid), DFID,
1172 PFID(&rnm->cr_spfid));
1173 namelen = changelog_rec_snamelen(rec);
1174 if (copylen > namelen + 1)
1175 copylen = namelen + 1;
1176 strlcpy(info->sname, changelog_rec_sname(rec), copylen);
1179 printf("Rec %lld: %d %s %s\n", info->recno, info->type,
1180 info->name, info->sname);
1183 printf("Rec %lld: %d %s\n", info->recno, info->type,
1187 llapi_changelog_free(&rec);
1193 /* Initialize the replication parameters */
1194 int lr_init_status()
1196 size_t size = sizeof(struct lustre_rsync_status) + PATH_MAX + 1;
1200 status = calloc(size, 1);
1203 status->ls_version = REPLICATE_STATUS_VER;
1204 status->ls_size = size;
1205 status->ls_last_recno = -1;
1209 /* Make a backup of the statuslog */
1210 void lr_backup_log()
1212 char backupfile[PATH_MAX];
1216 snprintf(backupfile, PATH_MAX, "%s.old", statuslog);
1217 (void) rename(statuslog, backupfile);
1223 /* Save replication parameters to a statuslog. */
1228 size_t write_size = status->ls_size;
1229 struct lr_parent_child_list *curr;
1232 if (statuslog == NULL)
1237 fd = open(statuslog, O_WRONLY | O_CREAT | O_SYNC,
1240 fprintf(stderr, "Error opening log file for writing (%s)\n",
1245 size = write(fd, status, write_size);
1246 if (size != write_size) {
1247 fprintf(stderr, "Error writing to log file (%s) %d\n",
1253 for (curr = parents; curr; curr = curr->pc_next) {
1254 size = write(fd, &curr->pc_log, sizeof(curr->pc_log));
1255 if (size != sizeof(curr->pc_log)) {
1256 fprintf(stderr, "Error writing to log file (%s) %d\n",
1266 /* Read statuslog and populate the replication parameters. Command
1267 * line parameters take precedence over parameters in the log file.*/
1270 struct lr_parent_child_list *tmp;
1271 struct lr_parent_child_log rec;
1272 struct lustre_rsync_status *s;
1275 size_t read_size = sizeof(struct lustre_rsync_status) + PATH_MAX + 1;
1278 if (statuslog == NULL)
1281 s = calloc(1, read_size);
1287 fd = open(statuslog, O_RDONLY);
1293 size = read(fd, s, read_size);
1294 if (size != read_size) {
1299 if (read_size < s->ls_size) {
1300 read_size = s->ls_size;
1301 s = lr_grow_buf(s, read_size);
1307 if (lseek(fd, 0, SEEK_SET) == -1) {
1312 size = read(fd, s, read_size);
1313 if (size != read_size) {
1319 while (read(fd, &rec, sizeof(rec)) != 0) {
1320 tmp = calloc(1, sizeof(*tmp));
1327 tmp->pc_next = parents;
1331 /* copy uninitialized fields to status */
1332 if (status->ls_num_targets == 0) {
1333 if (status->ls_size != s->ls_size) {
1334 status = lr_grow_buf(status, s->ls_size);
1335 if (status == NULL) {
1340 status->ls_size = s->ls_size;
1342 status->ls_num_targets = s->ls_num_targets;
1343 memcpy(status->ls_targets, s->ls_targets,
1344 (PATH_MAX + 1) * s->ls_num_targets);
1346 if (status->ls_last_recno == -1)
1347 status->ls_last_recno = s->ls_last_recno;
1349 if (status->ls_registration[0] == '\0')
1350 strlcpy(status->ls_registration, s->ls_registration,
1351 sizeof(status->ls_registration));
1353 if (status->ls_mdt_device[0] == '\0')
1354 strlcpy(status->ls_mdt_device, s->ls_mdt_device,
1355 sizeof(status->ls_mdt_device));
1357 if (status->ls_source_fs[0] == '\0')
1358 strlcpy(status->ls_source_fs, s->ls_source_fs,
1359 sizeof(status->ls_source_fs));
1361 if (status->ls_source[0] == '\0')
1362 strlcpy(status->ls_source, s->ls_source,
1363 sizeof(status->ls_source));
1373 /* Clear changelogs every CLEAR_INTERVAL records or at the end of
1375 int lr_clear_cl(struct lr_info *info, int force)
1377 char mdt_device[LR_NAME_MAXLEN + 1];
1381 if (force || info->recno > status->ls_last_recno + CLEAR_INTERVAL) {
1382 if (info->type == CL_RENAME)
1383 rec = info->recno + 1;
1386 if (!noclear && !dryrun) {
1387 /* llapi_changelog_clear modifies the mdt
1388 * device name so make a copy of it until this
1391 strlcpy(mdt_device, status->ls_mdt_device,
1392 sizeof(mdt_device));
1393 rc = llapi_changelog_clear(mdt_device,
1394 status->ls_registration,
1397 printf("Changelog clear (%s, %s, %lld) "
1398 "returned %d\n", status->ls_mdt_device,
1399 status->ls_registration, rec, rc);
1401 if (!rc && !dryrun) {
1402 status->ls_last_recno = rec;
1411 /* Locate a usable version of rsync. At this point we'll use any
1413 int lr_locate_rsync()
1419 snprintf(rsync, PATH_MAX, "%s -p %s", TYPE, RSYNC);
1420 fp = popen(rsync, "r");
1424 if (fgets(rsync, PATH_MAX, fp) == NULL) {
1429 len = strlen(rsync);
1430 if (len > 0 && rsync[len - 1] == '\n')
1431 rsync[len - 1] = '\0';
1434 /* Determine the version of rsync */
1435 snprintf(rsync_ver, PATH_MAX, "%s --version", rsync);
1436 fp = popen(rsync_ver, "r");
1440 if (fgets(rsync_ver, PATH_MAX, fp) == NULL) {
1444 len = strlen(rsync_ver);
1445 if (len > 0 && rsync_ver[len - 1] == '\n')
1446 rsync_ver[len - 1] = '\0';
1453 /* Print the replication parameters */
1454 void lr_print_status(struct lr_info *info)
1461 printf("Lustre filesystem: %s\n", status->ls_source_fs);
1462 printf("MDT device: %s\n", status->ls_mdt_device);
1463 printf("Source: %s\n", status->ls_source);
1464 for (i = 0; i < status->ls_num_targets; i++)
1465 printf("Target: %s\n", status->ls_targets[i]);
1466 if (statuslog != NULL)
1467 printf("Statuslog: %s\n", statuslog);
1468 printf("Changelog registration: %s\n", status->ls_registration);
1469 printf("Starting changelog record: %jd\n",
1470 (uintmax_t)status->ls_last_recno);
1472 printf("Replicate xattrs: no\n");
1474 printf("Clear changelog after use: no\n");
1476 printf("Using rsync: %s (%s)\n", rsync, rsync_ver);
1479 void lr_print_failure(struct lr_info *info, int rc)
1481 fprintf(stderr, "Replication of operation failed(%d):"
1482 " %lld %s (%d) %s %s %s\n", rc, info->recno,
1483 changelog_type2str(info->type), info->type, info->tfid,
1484 info->pfid, info->name);
1487 /* Replicate filesystem operations from src_path to target_path */
1490 void *changelog_priv;
1491 struct lr_info *info;
1492 struct lr_info *ext = NULL;
1500 info = calloc(1, sizeof(struct lr_info));
1504 rc = llapi_search_fsname(status->ls_source, status->ls_source_fs);
1506 fprintf(stderr, "Source path is not a valid Lustre client "
1510 if (status->ls_mdt_device[0] == '\0')
1511 snprintf(status->ls_mdt_device, LR_NAME_MAXLEN, "%s%s",
1512 status->ls_source_fs, DEFAULT_MDT);
1514 ext = calloc(1, sizeof(struct lr_info));
1520 for (i = 0, xattr_not_supp = 0; i < status->ls_num_targets; i++) {
1521 snprintf(info->dest, PATH_MAX, "%s/%s", status->ls_targets[i],
1523 rc = mkdir(info->dest, 0777);
1524 if (rc == -1 && errno != EEXIST) {
1525 fprintf(stderr, "Error writing to target path %s.\n",
1526 status->ls_targets[i]);
1530 rc = llistxattr(info->src, info->xlist, info->xsize);
1531 if (rc == -1 && errno == ENOTSUP) {
1532 fprintf(stderr, "xattrs not supported on %s\n",
1533 status->ls_targets[i]);
1537 if (xattr_not_supp == status->ls_num_targets)
1538 /* None of the targets support xattrs. */
1541 lr_print_status(info);
1543 /* Open changelogs for consumption*/
1544 rc = llapi_changelog_start(&changelog_priv,
1545 CHANGELOG_FLAG_BLOCK | CHANGELOG_FLAG_JOBID,
1546 status->ls_mdt_device, status->ls_last_recno);
1548 fprintf(stderr, "Error opening changelog file for fs %s.\n",
1549 status->ls_source_fs);
1553 while (!quit && lr_parse_line(changelog_priv, info) == 0) {
1555 if (info->type == CL_RENAME && !info->is_extended) {
1556 /* Newer rename operations extends changelog to store
1557 * source file information, but old changelog has
1560 if (lr_parse_line(changelog_priv, ext) != 0)
1562 memcpy(info->sfid, info->tfid, sizeof(info->sfid));
1563 memcpy(info->spfid, info->pfid, sizeof(info->spfid));
1564 memcpy(info->tfid, ext->tfid, sizeof(info->tfid));
1565 memcpy(info->pfid, ext->pfid, sizeof(info->pfid));
1566 strlcpy(info->sname, info->name, sizeof(info->sname));
1567 strlcpy(info->name, ext->name, sizeof(info->name));
1568 info->is_extended = 1;
1576 switch(info->type) {
1581 rc = lr_create(info);
1585 rc = lr_remove(info);
1595 rc = lr_setattr(info);
1598 rc = lr_setxattr(info);
1605 /* Nothing needs to be done for these entries */
1610 DEBUG_EXIT(info, rc);
1611 if (rc && rc != -ENOENT) {
1612 lr_print_failure(info, rc);
1617 lr_clear_cl(info, 0);
1619 bzero(info, sizeof(struct lr_info));
1620 bzero(ext, sizeof(struct lr_info));
1624 llapi_changelog_fini(&changelog_priv);
1626 if (errors || verbose)
1627 printf("Errors: %d\n", errors);
1629 /* Clear changelog records used so far */
1630 lr_clear_cl(info, 1);
1633 printf("lustre_rsync took %ld seconds\n", time(NULL) - start);
1634 printf("Changelog records consumed: %lld\n", rec_count);
1649 termination_handler (int signum)
1651 /* Set a flag for the replicator to gracefully shutdown */
1653 printf("lustre_rsync halting.\n");
1656 int main(int argc, char *argv[])
1662 if ((rc = lr_init_status()) != 0)
1665 while ((rc = getopt_long(argc, argv, "as:t:m:u:l:vx:zc:ry:n:d:D:",
1666 long_opts, NULL)) >= 0) {
1669 /* Assume absolute paths */
1673 /* Assume absolute paths */
1674 strlcpy(status->ls_source, optarg,
1675 sizeof(status->ls_source));
1678 status->ls_num_targets++;
1680 if (numtargets != status->ls_num_targets) {
1681 /* Targets were read from a log
1682 file. The ones specified on the
1683 command line take precedence. The
1684 ones from the log file will be
1686 status->ls_num_targets = numtargets;
1688 newsize = sizeof (struct lustre_rsync_status) +
1689 (status->ls_num_targets * (PATH_MAX + 1));
1690 if (status->ls_size != newsize) {
1691 status->ls_size = newsize;
1692 status = lr_grow_buf(status, newsize);
1696 strlcpy(status->ls_targets[status->ls_num_targets - 1],
1697 optarg, sizeof(status->ls_targets[0]));
1700 strlcpy(status->ls_mdt_device, optarg,
1701 sizeof(status->ls_mdt_device));
1704 strlcpy(status->ls_registration, optarg,
1705 sizeof(status->ls_registration));
1709 (void) lr_read_log();
1715 if (strcmp("no", optarg) == 0) {
1717 } else if (strcmp("yes", optarg) != 0) {
1718 printf("Invalid parameter %s. "
1719 "Specify --xattr=no or --xattr=yes\n",
1728 /* Undocumented option cl-clear */
1729 if (strcmp("no", optarg) == 0) {
1731 } else if (strcmp("yes", optarg) != 0) {
1732 printf("Invalid parameter %s. "
1733 "Specify --cl-clear=no "
1734 "or --cl-clear=yes\n",
1740 /* Undocumented option use-rsync */
1744 /* Undocumented option rsync-threshold */
1745 rsync_threshold = atol(optarg);
1748 /* Undocumented option start-recno */
1749 status->ls_last_recno = atol(optarg);
1752 /* Undocumented option debug */
1753 debug = atoi(optarg);
1754 if (debug < 0 || debug > 2)
1758 /* Undocumented option debug log file */
1759 if (debug_log != NULL)
1761 debug_log = fopen(optarg, "a");
1762 if (debug_log == NULL) {
1763 printf("Cannot open %s for debug log\n",
1769 fprintf(stderr, "error: %s: option '%s' "
1770 "unrecognized.\n", argv[0], argv[optind - 1]);
1776 if (status->ls_last_recno == -1)
1777 status->ls_last_recno = 0;
1778 if (strnlen(status->ls_registration, LR_NAME_MAXLEN) == 0) {
1779 /* No registration ID was passed in. */
1780 printf("Please specify changelog consumer registration id.\n");
1784 if (strnlen(status->ls_source, PATH_MAX) == 0) {
1785 fprintf(stderr, "Please specify the source path.\n");
1789 if (strnlen(status->ls_targets[0], PATH_MAX) == 0) {
1790 fprintf(stderr, "Please specify the target path.\n");
1795 /* This plumbing is needed for some of the ioctls behind
1796 llapi calls to work. */
1797 if (obd_initialize(argc, argv) < 0) {
1798 fprintf(stderr, "obd_initialize failed.\n");
1802 rc = lr_locate_rsync();
1803 if (use_rsync && rc != 0) {
1804 fprintf(stderr, "Error: unable to locate %s.\n", RSYNC);
1808 signal(SIGINT, termination_handler);
1809 signal(SIGHUP, termination_handler);
1810 signal(SIGTERM, termination_handler);
1812 rc = lr_replicate();
1814 if (debug_log != NULL)