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.
117 #include <sys/stat.h>
118 #include <sys/types.h>
122 #include <sys/xattr.h>
124 #include <libcfs/util/string.h>
125 #include <libcfs/util/parser.h>
126 #include <lustre/lustreapi.h>
127 #include <lustre/lustre_idl.h>
128 #include "lustre_rsync.h"
130 #define REPLICATE_STATUS_VER 1
131 #define CLEAR_INTERVAL 100
132 #define DEFAULT_RSYNC_THRESHOLD 0xA00000 /* 10 MB */
134 #define TYPE_STR_LEN 16
136 #define DEFAULT_MDT "-MDT0000"
137 #define SPECIAL_DIR ".lustrerepl"
138 #define RSYNC "rsync"
145 /* Not used; declared for fulfilling obd.c's dependency. */
146 command_t cmdlist[0];
147 extern int obd_initialize(int argc, char **argv);
149 /* Information for processing a changelog record. This structure is
150 allocated on the heap instead of allocating large variables on the
155 unsigned int is_extended:1;
156 enum changelog_rec_type type;
157 char tfid[LR_FID_STR_LEN];
158 char pfid[LR_FID_STR_LEN];
159 char sfid[LR_FID_STR_LEN];
160 char spfid[LR_FID_STR_LEN];
161 char sname[NAME_MAX + 1];
162 char name[NAME_MAX + 1];
163 char src[PATH_MAX + 1];
164 char dest[PATH_MAX + 1];
165 char path[PATH_MAX + 1];
166 char savedpath[PATH_MAX + 1];
167 char link[PATH_MAX + 1];
168 char linktmp[PATH_MAX + 1];
173 /* Variables for querying the xattributes */
180 struct lr_parent_child_list {
181 struct lr_parent_child_log pc_log;
182 struct lr_parent_child_list *pc_next;
185 struct lustre_rsync_status *status;
186 char *statuslog; /* Name of the status log file */
188 int noxattr; /* Flag to turn off replicating xattrs */
189 int noclear; /* Flag to turn off clearing changelogs */
190 int debug; /* Flag to turn debugging information on and off */
191 int verbose; /* Verbose output */
192 long long rec_count; /* No of changelog records that were processed */
195 int use_rsync; /* Flag to turn on use of rsync to copy data */
196 long long rsync_threshold = DEFAULT_RSYNC_THRESHOLD;
197 int quit; /* Flag to stop processing the changelog; set on the
198 receipt of a signal */
199 int abort_on_err = 0;
201 char rsync[PATH_MAX];
202 char rsync_ver[PATH_MAX];
203 struct lr_parent_child_list *parents;
207 /* Command line options */
208 struct option long_opts[] = {
209 {"source", required_argument, 0, 's'},
210 {"target", required_argument, 0, 't'},
211 {"mdt", required_argument, 0, 'm'},
212 {"user", required_argument, 0, 'u'},
213 {"statuslog", required_argument, 0, 'l'},
214 {"verbose", no_argument, 0, 'v'},
215 {"xattr", required_argument, 0, 'x'},
216 {"dry-run", no_argument, 0, 'z'},
217 /* Undocumented options follow */
218 {"cl-clear", required_argument, 0, 'c'},
219 {"use-rsync", no_argument, 0, 'r'},
220 {"rsync-threshold", required_argument, 0, 'y'},
221 {"start-recno", required_argument, 0, 'n'},
222 {"abort-on-err",no_argument, 0, 'a'},
223 {"debug", required_argument, 0, 'd'},
224 {"debuglog", required_argument, 0, 'D'},
228 /* Command line usage */
231 fprintf(stderr, "\tlustre_rsync -s <lustre_root_path> -t <target_path> "
232 "-m <mdt> -r <user id> -l <status log>\n"
233 "lustre_rsync can also pick up parameters from a "
234 "status log created earlier.\n"
235 "\tlustre_rsync -l <log_file>\n"
237 "\t--xattr <yes|no> replicate EAs\n"
238 "\t--abort-on-err abort at first err\n"
240 "\t--dry-run don't write anything\n");
243 #define DEBUG_ENTRY(info) \
244 lr_debug(D_TRACE, "***** Start %lld %s (%d) %s %s %s *****\n", \
245 (info)->recno, changelog_type2str((info)->type), \
246 (info)->type, (info)->tfid, (info)->pfid, (info)->name);
248 #define DEBUG_EXIT(info, rc) \
249 lr_debug(D_TRACE, "##### End %lld %s (%d) %s %s %s rc=%d #####\n", \
250 (info)->recno, changelog_type2str((info)->type), \
251 (info)->type, (info)->tfid, (info)->pfid, (info)->name, rc);
253 /* Print debug information. This is controlled by the value of the
254 global variable 'debug' */
255 void lr_debug(int level, const char *fmt, ...)
263 if (debug_log != NULL)
264 vfprintf(debug_log, fmt, ap);
266 vfprintf(stdout, fmt, ap);
271 void * lr_grow_buf(void *buf, int size)
275 ptr = realloc(buf, size);
282 /* Use rsync to replicate file data */
283 int lr_rsync_data(struct lr_info *info)
286 struct stat st_src, st_dest;
289 lr_debug(DTRACE, "Syncing data%s\n", info->tfid);
291 rc = stat(info->src, &st_src);
293 fprintf(stderr, "Error: Unable to stat src=%s %s\n",
294 info->src, info->name);
300 rc = stat(info->dest, &st_dest);
302 fprintf(stderr, "Error: Unable to stat dest=%s\n",
307 if (st_src.st_mtime != st_dest.st_mtime ||
308 st_src.st_size != st_dest.st_size) {
309 /* XXX spawning off an rsync for every data sync and
310 * waiting synchronously is bad for performance.
311 * librsync could possibly used here. But it does not
312 * seem to be of production grade. Multi-threaded
313 * replication is also to be considered.
316 snprintf(cmd, PATH_MAX, "%s --inplace %s %s", rsync, info->src,
318 lr_debug(DTRACE, "\t%s %s\n", cmd, info->tfid);
319 status = system(cmd);
322 } else if (WIFEXITED(status)) {
323 status = WEXITSTATUS(status);
326 else if (status == 23 || status == 24)
327 /* Error due to vanished source files;
333 lr_debug(DINFO, "rsync %s exited with %d %d\n",
334 info->src, status, rc);
339 lr_debug(DTRACE, "Not syncing %s and %s %s\n", info->src,
340 info->dest, info->tfid);
346 int lr_copy_data(struct lr_info *info)
356 fd_src = open(info->src, O_RDONLY);
359 if (fstat(fd_src, &st_src) == -1 ||
360 stat(info->dest, &st_dest) == -1)
363 if (st_src.st_mtime == st_dest.st_mtime &&
364 st_src.st_size == st_dest.st_size)
367 if (st_src.st_size > rsync_threshold && rsync[0] != '\0') {
368 /* It is more efficient to use rsync to replicate
369 large files. Any file larger than rsync_threshold
370 is handed off to rsync. */
371 lr_debug(DTRACE, "Using rsync to replicate %s\n", info->tfid);
372 rc = lr_rsync_data(info);
376 fd_dest = open(info->dest, O_WRONLY | O_TRUNC, st_src.st_mode);
381 bufsize = st_dest.st_blksize;
383 if (info->bufsize < bufsize) {
385 info->buf = lr_grow_buf(info->buf, bufsize);
386 if (info->buf == NULL) {
390 info->bufsize = bufsize;
394 rsize = read(fd_src, info->buf, bufsize);
397 } else if (rsize < 0) {
402 if (write(fd_dest, info->buf, rsize) != rsize) {
420 /* Copy data from source to destination */
421 int lr_sync_data(struct lr_info *info)
424 return lr_rsync_data(info);
426 return lr_copy_data(info);
429 /* Copy all attributes from file src to file dest */
430 int lr_copy_attr(char *src, char *dest)
435 if (stat(src, &st) == -1 ||
436 chmod(dest, st.st_mode) == -1 ||
437 chown(dest, st.st_uid, st.st_gid) == -1)
440 time.actime = st.st_atime;
441 time.modtime = st.st_mtime;
442 if (utime(dest, &time) == -1)
447 /* Copy all xattrs from file info->src to info->dest */
448 int lr_copy_xattr(struct lr_info *info)
450 size_t size = info->xsize;
459 rc = llistxattr(info->src, info->xlist, size);
460 lr_debug(DTRACE, "llistxattr(%s,%p) returned %d, errno=%d\n",
461 info->src, info->xlist, rc, errno);
462 if ((rc > 0 && info->xlist == NULL) || errno == ERANGE) {
463 size = rc > PATH_MAX ? rc : PATH_MAX;
464 info->xlist = lr_grow_buf(info->xlist, size);
465 if (info->xlist == NULL)
468 rc = llistxattr(info->src, info->xlist, size);
469 lr_debug(DTRACE, "llistxattr %s returned %d, errno=%d\n",
470 info->src, rc, errno);
477 while (start < len) {
479 rc = lgetxattr(info->src, info->xlist + start,
481 if (info->xvalue == NULL || errno == ERANGE) {
482 size = rc > PATH_MAX ? rc : PATH_MAX;
483 info->xvalue = lr_grow_buf(info->xvalue, size);
484 if (info->xvalue == NULL)
487 rc = lgetxattr(info->src, info->xlist + start,
490 lr_debug(DTRACE, "\t(%s,%d) rc=%p\n", info->xlist + start,
494 rc = lsetxattr(info->dest, info->xlist + start,
495 info->xvalue, size, 0);
496 lr_debug(DTRACE, "\tlsetxattr(), rc=%d, errno=%d\n",
499 if (errno != ENOTSUP) {
500 fprintf(stderr, "Error replicating "
501 " xattr for %s: %d\n",
508 start += strlen(info->xlist + start) + 1;
511 lr_debug(DINFO, "setxattr: %s %s\n", info->src, info->dest);
516 /* Retrieve the filesystem path for a given FID and a given
517 linkno. The path is returned in info->path */
518 int lr_get_path_ln(struct lr_info *info, char *fidstr, int linkno)
520 long long recno = -1;
523 rc = llapi_fid2path(status->ls_source, fidstr, info->path,
524 PATH_MAX, &recno, &linkno);
525 if (rc < 0 && rc != -ENOENT) {
526 fprintf(stderr, "fid2path error: (%s, %s) %d %s\n",
527 status->ls_source, fidstr, -rc, strerror(errno = -rc));
533 /* Retrieve the filesystem path for a given FID. The path is returned
535 int lr_get_path(struct lr_info *info, char *fidstr)
537 return lr_get_path_ln(info, fidstr, 0);
540 /* Generate the path for opening by FID */
541 void lr_get_FID_PATH(char *mntpt, char *fidstr, char *buf, int bufsize)
543 /* Open-by-FID path is <mntpt>/.lustre/fid/[SEQ:OID:VER] */
544 snprintf(buf, bufsize, "%s/%s/fid/%s", mntpt, dot_lustre_name,
549 /* Read the symlink information into 'info->link' */
550 int lr_get_symlink(struct lr_info *info)
555 lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
556 rc = readlink(info->src, info->linktmp, PATH_MAX);
558 info->linktmp[rc] = '\0';
561 lr_debug(DTRACE, "symlink: readlink returned %s\n", info->linktmp);
563 if (strncmp(info->linktmp, status->ls_source,
564 strlen(status->ls_source)) == 0) {
565 /* Strip source fs path and replace with target fs path. */
566 link = info->linktmp + strlen(status->ls_source);
567 snprintf(info->src, PATH_MAX, "%s%s",
568 status->ls_targets[info->target_no], link);
571 link = info->linktmp;
573 strlcpy(info->link, link, sizeof(info->link));
578 /* Create file/directory/device file/symlink. */
579 int lr_mkfile(struct lr_info *info)
585 lr_debug(DINFO, "mkfile(%d) %s \n", info->type, info->dest);
586 if (info->type == CL_MKDIR) {
587 rc = mkdir(info->dest, 0777);
588 } else if (info->type == CL_SOFTLINK) {
589 lr_get_symlink(info);
590 rc = symlink(info->link, info->dest);
591 } else if (info->type == CL_MKNOD) {
592 lr_get_FID_PATH(status->ls_source, info->tfid,
593 info->src, PATH_MAX);
594 rc = stat(info->src, &st);
601 rc = mknod(info->dest, st.st_mode, st.st_rdev);
603 rc = mknod(info->dest, S_IFREG | 0777, 0);
608 /* Sync data and attributes */
609 if (info->type == CL_CREATE || info->type == CL_MKDIR) {
610 lr_debug(DTRACE, "Syncing data and attributes %s\n",
612 (void) lr_copy_xattr(info);
613 if (info->type == CL_CREATE)
614 rc = lr_sync_data(info);
616 rc = lr_copy_attr(info->src, info->dest);
619 /* Source file has disappeared. Not an error. */
622 lr_debug(DTRACE, "Not syncing data and attributes %s\n",
629 int lr_add_pc(const char *pfid, const char *tfid, const char *name)
631 struct lr_parent_child_list *p;
634 p = calloc(1, sizeof(*p));
637 len = strlcpy(p->pc_log.pcl_pfid, pfid, sizeof(p->pc_log.pcl_pfid));
638 if (len >= sizeof(p->pc_log.pcl_pfid))
640 len = strlcpy(p->pc_log.pcl_tfid, tfid, sizeof(p->pc_log.pcl_tfid));
641 if (len >= sizeof(p->pc_log.pcl_tfid))
643 len = strlcpy(p->pc_log.pcl_name, name, sizeof(p->pc_log.pcl_name));
644 if (len >= sizeof(p->pc_log.pcl_name))
647 p->pc_next = parents;
656 void lr_cascade_move(const char *fid, const char *dest, struct lr_info *info)
658 struct lr_parent_child_list *curr, *prev;
662 d = calloc(1, PATH_MAX + 1);
663 prev = curr = parents;
665 if (strcmp(curr->pc_log.pcl_pfid, fid) == 0) {
666 snprintf(d, PATH_MAX, "%s/%s", dest,
667 curr->pc_log.pcl_name);
668 snprintf(info->src, PATH_MAX, "%s/%s/%s",
669 status->ls_targets[info->target_no],
670 SPECIAL_DIR, curr->pc_log.pcl_tfid);
671 rc = rename(info->src, d);
673 fprintf(stderr, "Error renaming file "
675 info->src, d, errno);
679 parents = curr->pc_next;
681 prev->pc_next = curr->pc_next;
682 lr_cascade_move(curr->pc_log.pcl_tfid, d, info);
684 prev = curr = parents;
688 curr = curr->pc_next;
695 /* remove [info->spfid, info->sfid] from parents */
696 int lr_remove_pc(const char *pfid, const char *tfid)
698 struct lr_parent_child_list *curr, *prev;
700 for (prev = curr = parents; curr; prev = curr, curr = curr->pc_next) {
701 if (strcmp(curr->pc_log.pcl_pfid, pfid) == 0 &&
702 strcmp(curr->pc_log.pcl_tfid, tfid) == 0) {
704 parents = curr->pc_next;
706 prev->pc_next = curr->pc_next;
714 /* Create file under SPECIAL_DIR with its tfid as its name. */
715 int lr_mk_special(struct lr_info *info)
719 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
720 status->ls_targets[info->target_no], SPECIAL_DIR,
723 rc = lr_mkfile(info);
727 rc = lr_add_pc(info->pfid, info->tfid, info->name);
731 /* Remove a file or directory */
732 int lr_rmfile(struct lr_info *info)
736 if (info->type == CL_RMDIR)
737 rc = rmdir(info->dest);
739 rc = unlink(info->dest);
745 /* Recursively remove directory and its contents */
746 int lr_rm_recursive(struct lr_info *info)
750 snprintf(info->cmd, PATH_MAX, "rm -rf %s", info->dest);
751 rc = system(info->cmd);
758 /* Remove a file under SPECIAL_DIR with its tfid as its name. */
759 int lr_rm_special(struct lr_info *info)
763 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
764 status->ls_targets[info->target_no], SPECIAL_DIR,
766 rc = lr_rmfile(info);
769 lr_debug(DINFO, "remove: %s; rc=%d, errno=%d\n",
770 info->dest, rc, errno);
774 /* Replicate file and directory create events */
775 int lr_create(struct lr_info *info)
782 /* Is target FID present on the source? */
783 rc = lr_get_path(info, info->tfid);
785 /* Source file has disappeared. Not an error. */
786 lr_debug(DINFO, "create: tfid %s not found on"
787 "source-fs\n", info->tfid);
792 strcpy(info->savedpath, info->path);
794 /* Is parent FID present on the source */
795 rc = lr_get_path(info, info->pfid);
797 lr_debug(DINFO, "create: pfid %s not found on source-fs\n",
804 /* Is f2p(pfid)+name != f2p(tfid)? If not the file has moved. */
805 len = strlen(info->path);
806 if (len == 1 && info->path[0] == '/')
807 snprintf(info->dest, PATH_MAX, "%s", info->name);
808 else if (len - 1 > 0 && info->path[len - 1] == '/')
809 snprintf(info->dest, PATH_MAX, "%s%s", info->path, info->name);
811 snprintf(info->dest, PATH_MAX, "%s/%s", info->path, info->name);
813 lr_debug(DTRACE, "dest = %s; savedpath = %s\n", info->dest,
815 if (strncmp(info->dest, info->savedpath, PATH_MAX) != 0) {
816 lr_debug(DTRACE, "create: file moved (%s). %s != %s\n",
817 info->tfid, info->dest, info->savedpath);
821 /* Is f2p(pfid) present on the target? If not, the parent has
824 snprintf(info->dest, PATH_MAX, "%s/%s", status->ls_targets[0],
826 if (access(info->dest, F_OK) != 0) {
827 lr_debug(DTRACE, "create: parent %s not found\n",
832 for (info->target_no = 0; info->target_no < status->ls_num_targets;
834 snprintf(info->dest, PATH_MAX, "%s/%s",
835 status->ls_targets[info->target_no], info->savedpath);
836 lr_get_FID_PATH(status->ls_source, info->tfid, info->src,
840 rc1 = lr_mkfile(info);
841 if (mkspecial || rc1 == -ENOENT) {
842 rc1 = lr_mk_special(info);
850 /* Replicate a file remove (rmdir/unlink) operation */
851 int lr_remove(struct lr_info *info)
856 for (info->target_no = 0; info->target_no < status->ls_num_targets;
859 rc1 = lr_rm_special(info);
863 rc1 = lr_get_path(info, info->pfid);
864 if (rc1 == -ENOENT) {
865 lr_debug(DINFO, "remove: pfid %s not found\n",
873 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
874 status->ls_targets[info->target_no], info->path,
877 rc1 = lr_rmfile(info);
878 lr_debug(DINFO, "remove: %s; rc1=%d, errno=%d\n",
879 info->dest, rc1, errno);
880 if (rc1 == -ENOTEMPTY)
881 rc1 = lr_rm_recursive(info);
891 /* Replicate a rename/move operation. */
892 int lr_move(struct lr_info *info)
898 int special_dest = 0;
899 char srcpath[PATH_MAX + 1] = "";
901 LASSERT(info->is_extended);
903 rc_src = lr_get_path(info, info->spfid);
904 if (rc_src < 0 && rc_src != -ENOENT)
906 memcpy(srcpath, info->path, strlen(info->path));
908 rc_dest = lr_get_path(info, info->pfid);
909 if (rc_dest < 0 && rc_dest != -ENOENT)
912 for (info->target_no = 0; info->target_no < status->ls_num_targets;
916 snprintf(info->dest, PATH_MAX, "%s/%s",
917 status->ls_targets[info->target_no],
919 if (access(info->dest, F_OK) != 0) {
922 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
923 status->ls_targets[info->target_no],
924 info->path, info->name);
926 lr_debug(DINFO, "dest path %s rc_dest=%d\n", info->dest,
929 if (rc_dest == -ENOENT) {
930 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
931 status->ls_targets[info->target_no],
932 SPECIAL_DIR, info->sfid);
934 lr_debug(DINFO, "special dest %s\n", info->dest);
938 snprintf(info->src, PATH_MAX, "%s/%s/%s",
939 status->ls_targets[info->target_no],
940 srcpath, info->sname);
941 lr_debug(DINFO, "src path %s rc_src=%d\n", info->src,
944 if (rc_src == -ENOENT || (access(info->src, F_OK) != 0 &&
946 snprintf(info->src, PATH_MAX, "%s/%s/%s",
947 status->ls_targets[info->target_no],
948 SPECIAL_DIR, info->sfid);
950 lr_debug(DINFO, "special src %s\n", info->src);
955 if (strcmp(info->src, info->dest) != 0) {
956 rc1 = rename(info->src, info->dest);
959 lr_debug(DINFO, "rename returns %d\n", rc1);
963 rc1 = lr_remove_pc(info->spfid, info->sfid);
966 lr_cascade_move(info->sfid, info->dest, info);
968 rc1 = lr_add_pc(info->pfid, info->sfid, info->name);
970 lr_debug(DINFO, "move: %s [to] %s rc1=%d, errno=%d\n",
971 info->src, info->dest, rc1, errno);
978 /* Replicate a hard link */
979 int lr_link(struct lr_info *info)
987 lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
988 rc = stat(info->src, &st);
992 for (info->target_no = 0; info->target_no < status->ls_num_targets;
999 /* Search through the hardlinks to get the src and dest */
1000 for (i = 0; i < st.st_nlink && (info->src[0] == 0 ||
1001 info->dest[0] == 0); i++) {
1002 rc1 = lr_get_path_ln(info, info->tfid, i);
1003 lr_debug(rc1 ? 0:DTRACE, "\tfid2path %s, %s, %d rc=%d\n",
1004 info->path, info->name, i, rc1);
1008 len = strlen(info->path) - strlen(info->name);
1009 if (len >= 0 && strcmp(info->path + len,
1011 snprintf(info->dest, PATH_MAX, "%s/%s",
1012 status->ls_targets[info->target_no],
1014 else if (info->src[0] == 0)
1015 snprintf(info->src, PATH_MAX, "%s/%s",
1016 status->ls_targets[info->target_no],
1025 if (info->src[0] == 0 || info->dest[0] == 0)
1026 /* Could not find the source or destination.
1027 This can happen when some links don't exist
1031 if (info->src[0] == 0)
1032 snprintf(info->src, PATH_MAX, "%s/%s/%s",
1033 status->ls_targets[info->target_no],
1034 SPECIAL_DIR, info->tfid);
1035 else if (info->dest[0] == 0)
1036 snprintf(info->dest, PATH_MAX, "%s/%s/%s",
1037 status->ls_targets[info->target_no],
1038 SPECIAL_DIR, info->tfid);
1040 rc1 = link(info->src, info->dest);
1041 lr_debug(rc1?0:DINFO, "link: %s [to] %s; rc1=%d %s\n",
1042 info->src, info->dest, rc1, strerror(errno));
1050 /* Replicate file attributes */
1051 int lr_setattr(struct lr_info *info)
1056 lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
1058 rc = lr_get_path(info, info->tfid);
1060 lr_debug(DINFO, "setattr: %s not present on source-fs\n",
1065 for (info->target_no = 0; info->target_no < status->ls_num_targets;
1066 info->target_no++) {
1068 snprintf(info->dest, PATH_MAX, "%s/%s",
1069 status->ls_targets[info->target_no], info->path);
1070 lr_debug(DINFO, "setattr: %s %s %s", info->src, info->dest,
1073 rc1 = lr_sync_data(info);
1075 rc1 = lr_copy_attr(info->src, info->dest);
1082 /* Replicate xattrs */
1083 int lr_setxattr(struct lr_info *info)
1087 lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
1089 rc = lr_get_path(info, info->tfid);
1091 lr_debug(DINFO, "setxattr: %s not present on source-fs\n",
1096 for (info->target_no = 0; info->target_no < status->ls_num_targets;
1097 info->target_no++) {
1099 snprintf(info->dest, PATH_MAX, "%s/%s",
1100 status->ls_targets[info->target_no], info->path);
1101 lr_debug(DINFO, "setxattr: %s %s %s\n", info->src, info->dest,
1104 rc1 = lr_copy_xattr(info);
1112 /* Parse a line of changelog entry */
1113 int lr_parse_line(void *priv, struct lr_info *info)
1115 struct changelog_rec *rec;
1116 struct changelog_ext_rename *rnm;
1118 size_t copylen = sizeof(info->name);
1120 if (llapi_changelog_recv(priv, &rec) != 0)
1123 info->is_extended = !!(rec->cr_flags & CLF_RENAME);
1124 info->recno = rec->cr_index;
1125 info->type = rec->cr_type;
1126 snprintf(info->tfid, sizeof(info->tfid), DFID, PFID(&rec->cr_tfid));
1127 snprintf(info->pfid, sizeof(info->pfid), DFID, PFID(&rec->cr_pfid));
1129 namelen = strnlen(changelog_rec_name(rec), rec->cr_namelen);
1130 if (copylen > namelen + 1)
1131 copylen = namelen + 1;
1132 strlcpy(info->name, changelog_rec_name(rec), copylen);
1134 /* Don't use rnm if CLF_RENAME isn't set */
1135 rnm = changelog_rec_rename(rec);
1136 if (rec->cr_flags & CLF_RENAME && !fid_is_zero(&rnm->cr_sfid)) {
1137 copylen = sizeof(info->sname);
1139 snprintf(info->sfid, sizeof(info->sfid), DFID,
1140 PFID(&rnm->cr_sfid));
1141 snprintf(info->spfid, sizeof(info->spfid), DFID,
1142 PFID(&rnm->cr_spfid));
1143 namelen = changelog_rec_snamelen(rec);
1144 if (copylen > namelen + 1)
1145 copylen = namelen + 1;
1146 strlcpy(info->sname, changelog_rec_sname(rec), copylen);
1149 printf("Rec %lld: %d %s %s\n", info->recno, info->type,
1150 info->name, info->sname);
1153 printf("Rec %lld: %d %s\n", info->recno, info->type,
1157 llapi_changelog_free(&rec);
1163 /* Initialize the replication parameters */
1164 int lr_init_status()
1166 size_t size = sizeof(struct lustre_rsync_status) + PATH_MAX + 1;
1170 status = calloc(size, 1);
1173 status->ls_version = REPLICATE_STATUS_VER;
1174 status->ls_size = size;
1175 status->ls_last_recno = -1;
1179 /* Make a backup of the statuslog */
1180 void lr_backup_log()
1182 char backupfile[PATH_MAX];
1186 snprintf(backupfile, PATH_MAX, "%s.old", statuslog);
1187 (void) rename(statuslog, backupfile);
1193 /* Save replication parameters to a statuslog. */
1198 size_t write_size = status->ls_size;
1199 struct lr_parent_child_list *curr;
1202 if (statuslog == NULL)
1207 fd = open(statuslog, O_WRONLY | O_CREAT | O_SYNC,
1210 fprintf(stderr, "Error opening log file for writing (%s)\n",
1215 size = write(fd, status, write_size);
1216 if (size != write_size) {
1217 fprintf(stderr, "Error writing to log file (%s) %d\n",
1223 for (curr = parents; curr; curr = curr->pc_next) {
1224 size = write(fd, &curr->pc_log, sizeof(curr->pc_log));
1225 if (size != sizeof(curr->pc_log)) {
1226 fprintf(stderr, "Error writing to log file (%s) %d\n",
1236 /* Read statuslog and populate the replication parameters. Command
1237 * line parameters take precedence over parameters in the log file.*/
1240 struct lr_parent_child_list *tmp;
1241 struct lr_parent_child_log rec;
1242 struct lustre_rsync_status *s;
1245 size_t read_size = sizeof(struct lustre_rsync_status) + PATH_MAX + 1;
1248 if (statuslog == NULL)
1251 s = calloc(1, read_size);
1257 fd = open(statuslog, O_RDONLY);
1263 size = read(fd, s, read_size);
1264 if (size != read_size) {
1269 if (read_size < s->ls_size) {
1270 read_size = s->ls_size;
1271 s = lr_grow_buf(s, read_size);
1277 if (lseek(fd, 0, SEEK_SET) == -1) {
1282 size = read(fd, s, read_size);
1283 if (size != read_size) {
1289 while (read(fd, &rec, sizeof(rec)) != 0) {
1290 tmp = calloc(1, sizeof(*tmp));
1297 tmp->pc_next = parents;
1301 /* copy uninitialized fields to status */
1302 if (status->ls_num_targets == 0) {
1303 if (status->ls_size != s->ls_size) {
1304 status = lr_grow_buf(status, s->ls_size);
1305 if (status == NULL) {
1310 status->ls_size = s->ls_size;
1312 status->ls_num_targets = s->ls_num_targets;
1313 memcpy(status->ls_targets, s->ls_targets,
1314 (PATH_MAX + 1) * s->ls_num_targets);
1316 if (status->ls_last_recno == -1)
1317 status->ls_last_recno = s->ls_last_recno;
1319 if (status->ls_registration[0] == '\0')
1320 strlcpy(status->ls_registration, s->ls_registration,
1321 sizeof(status->ls_registration));
1323 if (status->ls_mdt_device[0] == '\0')
1324 strlcpy(status->ls_mdt_device, s->ls_mdt_device,
1325 sizeof(status->ls_mdt_device));
1327 if (status->ls_source_fs[0] == '\0')
1328 strlcpy(status->ls_source_fs, s->ls_source_fs,
1329 sizeof(status->ls_source_fs));
1331 if (status->ls_source[0] == '\0')
1332 strlcpy(status->ls_source, s->ls_source,
1333 sizeof(status->ls_source));
1343 /* Clear changelogs every CLEAR_INTERVAL records or at the end of
1345 int lr_clear_cl(struct lr_info *info, int force)
1347 char mdt_device[LR_NAME_MAXLEN + 1];
1351 if (force || info->recno > status->ls_last_recno + CLEAR_INTERVAL) {
1352 if (info->type == CL_RENAME)
1353 rec = info->recno + 1;
1356 if (!noclear && !dryrun) {
1357 /* llapi_changelog_clear modifies the mdt
1358 * device name so make a copy of it until this
1361 strlcpy(mdt_device, status->ls_mdt_device,
1362 sizeof(mdt_device));
1363 rc = llapi_changelog_clear(mdt_device,
1364 status->ls_registration,
1367 printf("Changelog clear (%s, %s, %lld) "
1368 "returned %d\n", status->ls_mdt_device,
1369 status->ls_registration, rec, rc);
1371 if (!rc && !dryrun) {
1372 status->ls_last_recno = rec;
1381 /* Locate a usable version of rsync. At this point we'll use any
1383 int lr_locate_rsync()
1389 snprintf(rsync, PATH_MAX, "%s -p %s", TYPE, RSYNC);
1390 fp = popen(rsync, "r");
1394 if (fgets(rsync, PATH_MAX, fp) == NULL) {
1399 len = strlen(rsync);
1400 if (len > 0 && rsync[len - 1] == '\n')
1401 rsync[len - 1] = '\0';
1404 /* Determine the version of rsync */
1405 snprintf(rsync_ver, PATH_MAX, "%s --version", rsync);
1406 fp = popen(rsync_ver, "r");
1410 if (fgets(rsync_ver, PATH_MAX, fp) == NULL) {
1414 len = strlen(rsync_ver);
1415 if (len > 0 && rsync_ver[len - 1] == '\n')
1416 rsync_ver[len - 1] = '\0';
1423 /* Print the replication parameters */
1424 void lr_print_status(struct lr_info *info)
1431 printf("Lustre filesystem: %s\n", status->ls_source_fs);
1432 printf("MDT device: %s\n", status->ls_mdt_device);
1433 printf("Source: %s\n", status->ls_source);
1434 for (i = 0; i < status->ls_num_targets; i++)
1435 printf("Target: %s\n", status->ls_targets[i]);
1436 if (statuslog != NULL)
1437 printf("Statuslog: %s\n", statuslog);
1438 printf("Changelog registration: %s\n", status->ls_registration);
1439 printf("Starting changelog record: "LPD64"\n", status->ls_last_recno);
1441 printf("Replicate xattrs: no\n");
1443 printf("Clear changelog after use: no\n");
1445 printf("Using rsync: %s (%s)\n", rsync, rsync_ver);
1448 void lr_print_failure(struct lr_info *info, int rc)
1450 fprintf(stderr, "Replication of operation failed(%d):"
1451 " %lld %s (%d) %s %s %s\n", rc, info->recno,
1452 changelog_type2str(info->type), info->type, info->tfid,
1453 info->pfid, info->name);
1456 /* Replicate filesystem operations from src_path to target_path */
1459 void *changelog_priv;
1460 struct lr_info *info;
1461 struct lr_info *ext = NULL;
1469 info = calloc(1, sizeof(struct lr_info));
1473 rc = llapi_search_fsname(status->ls_source, status->ls_source_fs);
1475 fprintf(stderr, "Source path is not a valid Lustre client "
1479 if (status->ls_mdt_device[0] == '\0')
1480 snprintf(status->ls_mdt_device, LR_NAME_MAXLEN, "%s%s",
1481 status->ls_source_fs, DEFAULT_MDT);
1483 ext = calloc(1, sizeof(struct lr_info));
1489 for (i = 0, xattr_not_supp = 0; i < status->ls_num_targets; i++) {
1490 snprintf(info->dest, PATH_MAX, "%s/%s", status->ls_targets[i],
1492 rc = mkdir(info->dest, 0777);
1493 if (rc == -1 && errno != EEXIST) {
1494 fprintf(stderr, "Error writing to target path %s.\n",
1495 status->ls_targets[i]);
1499 rc = llistxattr(info->src, info->xlist, info->xsize);
1500 if (rc == -1 && errno == ENOTSUP) {
1501 fprintf(stderr, "xattrs not supported on %s\n",
1502 status->ls_targets[i]);
1506 if (xattr_not_supp == status->ls_num_targets)
1507 /* None of the targets support xattrs. */
1510 lr_print_status(info);
1512 /* Open changelogs for consumption*/
1513 rc = llapi_changelog_start(&changelog_priv,
1514 CHANGELOG_FLAG_BLOCK | CHANGELOG_FLAG_JOBID,
1515 status->ls_source_fs, status->ls_last_recno);
1517 fprintf(stderr, "Error opening changelog file for fs %s.\n",
1518 status->ls_source_fs);
1522 while (!quit && lr_parse_line(changelog_priv, info) == 0) {
1524 if (info->type == CL_RENAME && !info->is_extended) {
1525 /* Newer rename operations extends changelog to store
1526 * source file information, but old changelog has
1529 if (lr_parse_line(changelog_priv, ext) != 0)
1531 memcpy(info->sfid, info->tfid, sizeof(info->sfid));
1532 memcpy(info->spfid, info->pfid, sizeof(info->spfid));
1533 memcpy(info->tfid, ext->tfid, sizeof(info->tfid));
1534 memcpy(info->pfid, ext->pfid, sizeof(info->pfid));
1535 strlcpy(info->sname, info->name, sizeof(info->sname));
1536 strlcpy(info->name, ext->name, sizeof(info->name));
1537 info->is_extended = 1;
1545 switch(info->type) {
1550 rc = lr_create(info);
1554 rc = lr_remove(info);
1564 rc = lr_setattr(info);
1567 rc = lr_setxattr(info);
1574 /* Nothing needs to be done for these entries */
1579 DEBUG_EXIT(info, rc);
1580 if (rc && rc != -ENOENT) {
1581 lr_print_failure(info, rc);
1586 lr_clear_cl(info, 0);
1588 bzero(info, sizeof(struct lr_info));
1589 bzero(ext, sizeof(struct lr_info));
1593 llapi_changelog_fini(&changelog_priv);
1595 if (errors || verbose)
1596 printf("Errors: %d\n", errors);
1598 /* Clear changelog records used so far */
1599 lr_clear_cl(info, 1);
1602 printf("lustre_rsync took %ld seconds\n", time(NULL) - start);
1603 printf("Changelog records consumed: %lld\n", rec_count);
1618 termination_handler (int signum)
1620 /* Set a flag for the replicator to gracefully shutdown */
1622 printf("lustre_rsync halting.\n");
1625 int main(int argc, char *argv[])
1631 if ((rc = lr_init_status()) != 0)
1634 while ((rc = getopt_long(argc, argv, "as:t:m:u:l:vx:zc:ry:n:d:D:",
1635 long_opts, NULL)) >= 0) {
1638 /* Assume absolute paths */
1642 /* Assume absolute paths */
1643 strlcpy(status->ls_source, optarg,
1644 sizeof(status->ls_source));
1647 status->ls_num_targets++;
1649 if (numtargets != status->ls_num_targets) {
1650 /* Targets were read from a log
1651 file. The ones specified on the
1652 command line take precedence. The
1653 ones from the log file will be
1655 status->ls_num_targets = numtargets;
1657 newsize = sizeof (struct lustre_rsync_status) +
1658 (status->ls_num_targets * (PATH_MAX + 1));
1659 if (status->ls_size != newsize) {
1660 status->ls_size = newsize;
1661 status = lr_grow_buf(status, newsize);
1665 strlcpy(status->ls_targets[status->ls_num_targets - 1],
1666 optarg, sizeof(status->ls_targets[0]));
1669 strlcpy(status->ls_mdt_device, optarg,
1670 sizeof(status->ls_mdt_device));
1673 strlcpy(status->ls_registration, optarg,
1674 sizeof(status->ls_registration));
1678 (void) lr_read_log();
1684 if (strcmp("no", optarg) == 0) {
1686 } else if (strcmp("yes", optarg) != 0) {
1687 printf("Invalid parameter %s. "
1688 "Specify --xattr=no or --xattr=yes\n",
1697 /* Undocumented option cl-clear */
1698 if (strcmp("no", optarg) == 0) {
1700 } else if (strcmp("yes", optarg) != 0) {
1701 printf("Invalid parameter %s. "
1702 "Specify --cl-clear=no "
1703 "or --cl-clear=yes\n",
1709 /* Undocumented option use-rsync */
1713 /* Undocumented option rsync-threshold */
1714 rsync_threshold = atol(optarg);
1717 /* Undocumented option start-recno */
1718 status->ls_last_recno = atol(optarg);
1721 /* Undocumented option debug */
1722 debug = atoi(optarg);
1723 if (debug < 0 || debug > 2)
1727 /* Undocumented option debug log file */
1728 if (debug_log != NULL)
1730 debug_log = fopen(optarg, "a");
1731 if (debug_log == NULL) {
1732 printf("Cannot open %s for debug log\n",
1738 fprintf(stderr, "error: %s: option '%s' "
1739 "unrecognized.\n", argv[0], argv[optind - 1]);
1745 if (status->ls_last_recno == -1)
1746 status->ls_last_recno = 0;
1747 if (strnlen(status->ls_registration, LR_NAME_MAXLEN) == 0) {
1748 /* No registration ID was passed in. */
1749 printf("Please specify changelog consumer registration id.\n");
1753 if (strnlen(status->ls_source, PATH_MAX) == 0) {
1754 fprintf(stderr, "Please specify the source path.\n");
1758 if (strnlen(status->ls_targets[0], PATH_MAX) == 0) {
1759 fprintf(stderr, "Please specify the target path.\n");
1764 /* This plumbing is needed for some of the ioctls behind
1765 llapi calls to work. */
1766 if (obd_initialize(argc, argv) < 0) {
1767 fprintf(stderr, "obd_initialize failed.\n");
1771 rc = lr_locate_rsync();
1772 if (use_rsync && rc != 0) {
1773 fprintf(stderr, "Error: unable to locate %s.\n", RSYNC);
1777 signal(SIGINT, termination_handler);
1778 signal(SIGHUP, termination_handler);
1779 signal(SIGTERM, termination_handler);
1781 rc = lr_replicate();
1783 if (debug_log != NULL)