4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/utils/lustre_rsync.c
34 * Author: Kalpak Shah <Kalpak.Shah@Sun.COM>
35 * Author: Manoj Joseph <Manoj.Joseph@Sun.COM>
39 * - lustre_rsync is a tool for replicating a lustre filesystem.
41 * - The source-fs is a live lustre filesystem. It is not a
42 * snapshot. It is mounted and undergoing changes
44 * - The target-fs is a copy of the source-fs from the past. Let's
45 * call this point, the 'sync point'.
47 * - There is a changelog of all metadata operations that happened on
48 * the filesystem since the 'sync point'.
50 * - lustre_rsync replicates all the operations saved in the changelog
51 * on to the target filesystem to make it identical to the source.
53 * To facilitate replication, the lustre filesystem provides
54 * a) a way to get the current filesystem path of a given FID
55 * b) a way to open files by specifying its FID
57 * The changelog only has a limited amount of information.
58 * tfid - The FID of the target file
59 * pfid - The FID of the parent of the target file (at the time of
61 * sfid - The FID of the source file
62 * spfid - The FID of the parent of the source file
63 * name - The name of the target file (at the time of the operation), the name
64 * of the source file is appended (delimited with '\0') if this
65 * operation involves a source
67 * With just this information, it is not alwasy possible to determine
68 * the file paths for each operation. For instance, if pfid does not
69 * exist on the source-fs (due to a subsequent deletion), its path
70 * cannot be queried. In such cases, lustre_rsync keeps the files in a
71 * special directory ("/.lustrerepl"). Once all the operations in a
72 * changelog are replayed, all the files in this special directory
73 * will get moved to the location as in the source-fs.
75 * Shorthand used: f2p(fid) = fid2path(fid)
77 * The following are the metadata operations of interest.
79 * If tfid is absent on the source-fs, ignore this operation
80 * If pfid is absent on the source-fs [or]
81 * if f2p(pfid) is not present on target-fs [or]
82 * if f2p(pfid)+name != f2p(tfid)
83 * creat .lustrerepl/tfid
84 * track [pfid,tfid,name]
89 * If .lustrerepl/[tfid] is present on the target
90 * rm .lustrerepl/[tfid]
91 * Else if pfid is present on the source-fs,
92 * if f2p(pfid)+name is present,
95 * 3. move (spfid,sname) to (pfid,name)
97 * if spfid is also present, mv (spfid,sname) to (pfid,name)
98 * else mv .lustrerepl/[sfid] to (pfid,name)
99 * Else if pfid is not present,
100 * if spfid is present, mv (spfid,sname) .lustrerepl/[sfid]
101 * If moving out of .lustrerepl
102 * move out all its children in .lustrerepl.
103 * [pfid,tfid,name] tracked from (1) is used for this.
115 #include <sys/stat.h>
116 #include <sys/types.h>
121 #include <sys/xattr.h>
122 #include <linux/types.h>
124 #include <libcfs/util/string.h>
125 #include <lustre/lustreapi.h>
126 #include "lustre_rsync.h"
128 #define REPLICATE_STATUS_VER 1
129 #define CLEAR_INTERVAL 100
130 #define DEFAULT_RSYNC_THRESHOLD 0xA00000 /* 10 MB */
132 #define TYPE_STR_LEN 16
134 #define DEFAULT_MDT "-MDT0000"
135 #define SPECIAL_DIR ".lustrerepl"
136 #define RSYNC "rsync"
143 /* Information for processing a changelog record. This structure is
144 allocated on the heap instead of allocating large variables on the
149 unsigned int is_extended:1;
150 enum changelog_rec_type type;
151 char tfid[LR_FID_STR_LEN];
152 char pfid[LR_FID_STR_LEN];
153 char sfid[LR_FID_STR_LEN];
154 char spfid[LR_FID_STR_LEN];
155 char sname[NAME_MAX + 1];
156 char name[NAME_MAX + 1];
157 char src[3 * PATH_MAX + 1];
158 char dest[3 * PATH_MAX + 1];
159 char path[PATH_MAX + 1];
160 char savedpath[PATH_MAX + 1];
161 char link[PATH_MAX + 1];
162 char linktmp[PATH_MAX + 1];
163 char cmd[PATH_MAX * 10];
167 /* Variables for querying the xattributes */
174 struct lr_parent_child_list {
175 struct lr_parent_child_log pc_log;
176 struct lr_parent_child_list *pc_next;
179 struct lustre_rsync_status *status;
180 char *statuslog; /* Name of the status log file */
182 int noxattr; /* Flag to turn off replicating xattrs */
183 int noclear; /* Flag to turn off clearing changelogs */
184 int debug; /* Flag to turn debugging information on and off */
185 int verbose; /* Verbose output */
186 long long rec_count; /* No of changelog records that were processed */
189 int use_rsync; /* Flag to turn on use of rsync to copy data */
190 long long rsync_threshold = DEFAULT_RSYNC_THRESHOLD;
191 int quit; /* Flag to stop processing the changelog; set on the
192 receipt of a signal */
193 int abort_on_err = 0;
195 char rsync[PATH_MAX];
196 char rsync_ver[PATH_MAX];
197 struct lr_parent_child_list *parents;
201 /* Command line options */
202 struct option long_opts[] = {
203 { .val = 'l', .name = "statuslog", .has_arg = required_argument },
204 { .val = 'm', .name = "mdt", .has_arg = required_argument },
205 { .val = 's', .name = "source", .has_arg = required_argument },
206 { .val = 't', .name = "target", .has_arg = required_argument },
207 { .val = 'u', .name = "user", .has_arg = required_argument },
208 { .val = 'v', .name = "verbose", .has_arg = no_argument },
209 { .val = 'x', .name = "xattr", .has_arg = required_argument },
210 { .val = 'z', .name = "dry-run", .has_arg = no_argument },
211 /* Undocumented options follow */
212 { .val = 'a', .name = "abort-on-err", .has_arg = no_argument },
213 { .val = 'c', .name = "cl-clear", .has_arg = required_argument },
214 { .val = 'd', .name = "debug", .has_arg = required_argument },
215 { .val = 'D', .name = "debuglog", .has_arg = required_argument },
216 { .val = 'n', .name = "start-recno", .has_arg = required_argument },
217 { .val = 'r', .name = "use-rsync", .has_arg = no_argument },
218 { .val = 'y', .name = "rsync-threshold",
219 .has_arg = required_argument },
222 /* Command line usage */
225 fprintf(stderr, "\tlustre_rsync -s <lustre_root_path> -t <target_path> "
226 "-m <mdt> -r <user id> -l <status log>\n"
227 "lustre_rsync can also pick up parameters from a "
228 "status log created earlier.\n"
229 "\tlustre_rsync -l <log_file>\n"
231 "\t--xattr <yes|no> replicate EAs\n"
232 "\t--abort-on-err abort at first err\n"
234 "\t--dry-run don't write anything\n");
237 /* Print debug information. This is controlled by the value of the
238 global variable 'debug' */
239 void lr_debug(int level, const char *fmt, ...)
247 if (debug_log != NULL)
248 vfprintf(debug_log, fmt, ap);
250 vfprintf(stdout, fmt, ap);
255 void * lr_grow_buf(void *buf, int size)
259 ptr = realloc(buf, size);
266 /* Use rsync to replicate file data */
267 int lr_rsync_data(struct lr_info *info)
269 struct stat st_src, st_dest;
272 lr_debug(DTRACE, "Syncing data%s\n", info->tfid);
274 rc = stat(info->src, &st_src);
276 fprintf(stderr, "Error: Unable to stat src=%s %s\n",
277 info->src, info->name);
283 rc = stat(info->dest, &st_dest);
285 fprintf(stderr, "Error: Unable to stat dest=%s\n",
290 if (st_src.st_mtime != st_dest.st_mtime ||
291 st_src.st_size != st_dest.st_size) {
292 /* XXX spawning off an rsync for every data sync and
293 * waiting synchronously is bad for performance.
294 * librsync could possibly used here. But it does not
295 * seem to be of production grade. Multi-threaded
296 * replication is also to be considered.
300 if (snprintf(info->cmd, sizeof(info->cmd), "%s --inplace %s %s",
301 rsync, info->src, info->dest) >= sizeof(info->cmd)) {
305 lr_debug(DTRACE, "\t%s %s\n", info->cmd, info->tfid);
306 status = system(info->cmd);
309 } else if (WIFEXITED(status)) {
310 status = WEXITSTATUS(status);
313 else if (status == 23 || status == 24)
314 /* Error due to vanished source files;
320 lr_debug(DINFO, "rsync %s exited with %d %d\n",
321 info->src, status, rc);
326 lr_debug(DTRACE, "Not syncing %s and %s %s\n", info->src,
327 info->dest, info->tfid);
333 int lr_copy_data(struct lr_info *info)
343 fd_src = open(info->src, O_RDONLY);
346 if (fstat(fd_src, &st_src) == -1 ||
347 stat(info->dest, &st_dest) == -1)
350 if (st_src.st_mtime == st_dest.st_mtime &&
351 st_src.st_size == st_dest.st_size)
354 if (st_src.st_size > rsync_threshold && rsync[0] != '\0') {
355 /* It is more efficient to use rsync to replicate
356 large files. Any file larger than rsync_threshold
357 is handed off to rsync. */
358 lr_debug(DTRACE, "Using rsync to replicate %s\n", info->tfid);
359 rc = lr_rsync_data(info);
363 fd_dest = open(info->dest, O_WRONLY | O_TRUNC, st_src.st_mode);
368 bufsize = st_dest.st_blksize;
370 if (info->bufsize < bufsize) {
372 info->buf = lr_grow_buf(info->buf, bufsize);
373 if (info->buf == NULL) {
377 info->bufsize = bufsize;
385 rsize = read(fd_src, buf, bufsize);
395 wsize = write(fd_dest, buf, rsize);
415 /* Copy data from source to destination */
416 int lr_sync_data(struct lr_info *info)
419 return lr_rsync_data(info);
421 return lr_copy_data(info);
424 /* Copy all attributes from file src to file dest */
425 int lr_copy_attr(char *src, char *dest)
430 if (stat(src, &st) == -1 ||
431 chmod(dest, st.st_mode) == -1 ||
432 chown(dest, st.st_uid, st.st_gid) == -1)
435 time.actime = st.st_atime;
436 time.modtime = st.st_mtime;
437 if (utime(dest, &time) == -1)
442 /* Copy all xattrs from file info->src to info->dest */
443 int lr_copy_xattr(struct lr_info *info)
445 size_t size = info->xsize;
454 rc = llistxattr(info->src, info->xlist, size);
455 lr_debug(DTRACE, "llistxattr(%s,%p) returned %d, errno=%d\n",
456 info->src, info->xlist, rc, errno);
457 if ((rc > 0 && info->xlist == NULL) || errno == ERANGE) {
458 size = rc > PATH_MAX ? rc : PATH_MAX;
459 info->xlist = lr_grow_buf(info->xlist, size);
460 if (info->xlist == NULL)
463 rc = llistxattr(info->src, info->xlist, size);
464 lr_debug(DTRACE, "llistxattr %s returned %d, errno=%d\n",
465 info->src, rc, errno);
472 while (start < len) {
474 rc = lgetxattr(info->src, info->xlist + start,
476 if (info->xvalue == NULL || errno == ERANGE) {
477 size = rc > PATH_MAX ? rc : PATH_MAX;
478 info->xvalue = lr_grow_buf(info->xvalue, size);
479 if (info->xvalue == NULL)
482 rc = lgetxattr(info->src, info->xlist + start,
485 lr_debug(DTRACE, "\t(%s,%d) rc=%p\n", info->xlist + start,
489 rc = lsetxattr(info->dest, info->xlist + start,
490 info->xvalue, size, 0);
491 lr_debug(DTRACE, "\tlsetxattr(), rc=%d, errno=%d\n",
494 if (errno != ENOTSUP) {
495 fprintf(stderr, "Error replicating "
496 " xattr for %s: %d\n",
503 start += strlen(info->xlist + start) + 1;
506 lr_debug(DINFO, "setxattr: %s %s\n", info->src, info->dest);
511 /* Retrieve the filesystem path for a given FID and a given
512 linkno. The path is returned in info->path */
513 int lr_get_path_ln(struct lr_info *info, char *fidstr, int linkno)
515 long long recno = -1;
518 rc = llapi_fid2path(status->ls_source, fidstr, info->path,
519 PATH_MAX, &recno, &linkno);
520 if (rc < 0 && rc != -ENOENT) {
521 fprintf(stderr, "fid2path error: (%s, %s) %d %s\n",
522 status->ls_source, fidstr, -rc, strerror(errno = -rc));
528 /* Retrieve the filesystem path for a given FID. The path is returned
530 int lr_get_path(struct lr_info *info, char *fidstr)
532 return lr_get_path_ln(info, fidstr, 0);
535 /* Generate the path for opening by FID */
536 void lr_get_FID_PATH(char *mntpt, char *fidstr, char *buf, int bufsize)
538 /* Open-by-FID path is <mntpt>/.lustre/fid/[SEQ:OID:VER] */
539 snprintf(buf, bufsize, "%s/%s/fid/%s", mntpt, dot_lustre_name,
544 /* Read the symlink information into 'info->link' */
545 int lr_get_symlink(struct lr_info *info)
550 lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
551 rc = readlink(info->src, info->linktmp, PATH_MAX);
553 info->linktmp[rc] = '\0';
556 lr_debug(DTRACE, "symlink: readlink returned %s\n", info->linktmp);
558 if (strncmp(info->linktmp, status->ls_source,
559 strlen(status->ls_source)) == 0) {
560 /* Strip source fs path and replace with target fs path. */
561 link = info->linktmp + strlen(status->ls_source);
562 snprintf(info->src, sizeof(info->src), "%s%s",
563 status->ls_targets[info->target_no], link);
566 link = info->linktmp;
568 strlcpy(info->link, link, sizeof(info->link));
573 /* Create file/directory/device file/symlink. */
574 int lr_mkfile(struct lr_info *info)
580 lr_debug(DINFO, "mkfile(%d) %s \n", info->type, info->dest);
581 if (info->type == CL_MKDIR) {
582 rc = mkdir(info->dest, 0777);
583 } else if (info->type == CL_SOFTLINK) {
584 lr_get_symlink(info);
585 rc = symlink(info->link, info->dest);
586 } else if (info->type == CL_MKNOD) {
587 lr_get_FID_PATH(status->ls_source, info->tfid,
588 info->src, PATH_MAX);
589 rc = stat(info->src, &st);
596 rc = mknod(info->dest, st.st_mode, st.st_rdev);
598 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;
659 char d[4 * PATH_MAX + 1];
662 prev = curr = parents;
664 if (strcmp(curr->pc_log.pcl_pfid, fid) == 0) {
665 if (snprintf(d, sizeof(d), "%s/%s", dest,
666 curr->pc_log.pcl_name) >= sizeof(d)) {
667 fprintf(stderr, "Buffer truncated\n");
670 if (snprintf(info->src, sizeof(info->src), "%s/%s/%s",
671 status->ls_targets[info->target_no],
672 SPECIAL_DIR, curr->pc_log.pcl_tfid) >=
675 rc = rename(info->src, d);
677 fprintf(stderr, "Error renaming file "
679 info->src, d, errno);
683 parents = curr->pc_next;
685 prev->pc_next = curr->pc_next;
686 lr_cascade_move(curr->pc_log.pcl_tfid, d, info);
688 prev = curr = parents;
692 curr = curr->pc_next;
697 /* remove [info->spfid, info->sfid] from parents */
698 int lr_remove_pc(const char *pfid, const char *tfid)
700 struct lr_parent_child_list *curr, *prev;
702 for (prev = curr = parents; curr; prev = curr, curr = curr->pc_next) {
703 if (strcmp(curr->pc_log.pcl_pfid, pfid) == 0 &&
704 strcmp(curr->pc_log.pcl_tfid, tfid) == 0) {
706 parents = curr->pc_next;
708 prev->pc_next = curr->pc_next;
716 /* Create file under SPECIAL_DIR with its tfid as its name. */
717 int lr_mk_special(struct lr_info *info)
721 snprintf(info->dest, sizeof(info->dest), "%s/%s/%s",
722 status->ls_targets[info->target_no], SPECIAL_DIR,
725 rc = lr_mkfile(info);
729 rc = lr_add_pc(info->pfid, info->tfid, info->name);
733 /* Remove a file or directory */
734 int lr_rmfile(struct lr_info *info)
738 if (info->type == CL_RMDIR)
739 rc = rmdir(info->dest);
741 rc = unlink(info->dest);
747 /* Recursively remove directory and its contents */
748 int lr_rm_recursive(struct lr_info *info)
752 snprintf(info->cmd, sizeof(info->cmd), "rm -rf %s",
754 rc = system(info->cmd);
761 /* Remove a file under SPECIAL_DIR with its tfid as its name. */
762 int lr_rm_special(struct lr_info *info)
766 snprintf(info->dest, sizeof(info->dest), "%s/%s/%s",
767 status->ls_targets[info->target_no], SPECIAL_DIR,
769 rc = lr_rmfile(info);
772 lr_debug(DINFO, "remove: %s; rc=%d, errno=%d\n",
773 info->dest, rc, errno);
777 /* Replicate file and directory create events */
778 int lr_create(struct lr_info *info)
785 /* Is target FID present on the source? */
786 rc = lr_get_path(info, info->tfid);
788 /* Source file has disappeared. Not an error. */
789 lr_debug(DINFO, "create: tfid %s not found on"
790 "source-fs\n", info->tfid);
795 strcpy(info->savedpath, info->path);
797 /* Is parent FID present on the source */
798 rc = lr_get_path(info, info->pfid);
800 lr_debug(DINFO, "create: pfid %s not found on source-fs\n",
807 /* Is f2p(pfid)+name != f2p(tfid)? If not the file has moved. */
808 len = strlen(info->path);
809 if (len == 1 && info->path[0] == '/')
810 snprintf(info->dest, sizeof(info->dest), "%s", info->name);
811 else if (len - 1 > 0 && info->path[len - 1] == '/')
812 snprintf(info->dest, sizeof(info->dest), "%s%s", info->path,
815 snprintf(info->dest, sizeof(info->dest), "%s/%s", info->path,
818 lr_debug(DTRACE, "dest = %s; savedpath = %s\n", info->dest,
820 if (strncmp(info->dest, info->savedpath, PATH_MAX) != 0) {
821 lr_debug(DTRACE, "create: file moved (%s). %s != %s\n",
822 info->tfid, info->dest, info->savedpath);
826 /* Is f2p(pfid) present on the target? If not, the parent has
829 snprintf(info->dest, sizeof(info->dest), "%s/%s",
830 status->ls_targets[0], info->path);
831 if (access(info->dest, F_OK) != 0) {
832 lr_debug(DTRACE, "create: parent %s not found\n",
837 for (info->target_no = 0; info->target_no < status->ls_num_targets;
839 snprintf(info->dest, sizeof(info->dest), "%s/%s",
840 status->ls_targets[info->target_no], info->savedpath);
841 lr_get_FID_PATH(status->ls_source, info->tfid, info->src,
845 rc1 = lr_mkfile(info);
846 if (mkspecial || rc1 == -ENOENT) {
847 rc1 = lr_mk_special(info);
855 /* Replicate a file remove (rmdir/unlink) operation */
856 int lr_remove(struct lr_info *info)
861 for (info->target_no = 0; info->target_no < status->ls_num_targets;
864 rc1 = lr_rm_special(info);
868 rc1 = lr_get_path(info, info->pfid);
869 if (rc1 == -ENOENT) {
870 lr_debug(DINFO, "remove: pfid %s not found\n",
878 snprintf(info->dest, sizeof(info->dest), "%s/%s/%s",
879 status->ls_targets[info->target_no], info->path,
882 rc1 = lr_rmfile(info);
883 lr_debug(DINFO, "remove: %s; rc1=%d, errno=%d\n",
884 info->dest, rc1, errno);
885 if (rc1 == -ENOTEMPTY)
886 rc1 = lr_rm_recursive(info);
896 /* Replicate a rename/move operation. */
897 int lr_move(struct lr_info *info)
903 int special_dest = 0;
904 char srcpath[PATH_MAX + 1] = "";
906 assert(info->is_extended);
908 rc_src = lr_get_path(info, info->spfid);
909 if (rc_src < 0 && rc_src != -ENOENT)
911 memcpy(srcpath, info->path, strlen(info->path));
913 rc_dest = lr_get_path(info, info->pfid);
914 if (rc_dest < 0 && rc_dest != -ENOENT)
917 for (info->target_no = 0; info->target_no < status->ls_num_targets;
921 snprintf(info->dest, sizeof(info->dest), "%s/%s",
922 status->ls_targets[info->target_no],
924 if (access(info->dest, F_OK) != 0) {
927 snprintf(info->dest, sizeof(info->dest),
929 status->ls_targets[info->target_no],
930 info->path, info->name);
932 lr_debug(DINFO, "dest path %s rc_dest=%d\n", info->dest,
935 if (rc_dest == -ENOENT) {
936 snprintf(info->dest, sizeof(info->dest), "%s/%s/%s",
937 status->ls_targets[info->target_no],
938 SPECIAL_DIR, info->sfid);
940 lr_debug(DINFO, "special dest %s\n", info->dest);
944 snprintf(info->src, sizeof(info->src), "%s/%s/%s",
945 status->ls_targets[info->target_no],
946 srcpath, info->sname);
947 lr_debug(DINFO, "src path %s rc_src=%d\n", info->src,
950 if (rc_src == -ENOENT || (access(info->src, F_OK) != 0 &&
952 snprintf(info->src, sizeof(info->src), "%s/%s/%s",
953 status->ls_targets[info->target_no],
954 SPECIAL_DIR, info->sfid);
956 lr_debug(DINFO, "special src %s\n", info->src);
961 if (strcmp(info->src, info->dest) != 0) {
962 rc1 = rename(info->src, info->dest);
965 lr_debug(DINFO, "rename returns %d\n", rc1);
969 rc1 = lr_remove_pc(info->spfid, info->sfid);
972 lr_cascade_move(info->sfid, info->dest, info);
974 rc1 = lr_add_pc(info->pfid, info->sfid, info->name);
976 lr_debug(DINFO, "move: %s [to] %s rc1=%d, errno=%d\n",
977 info->src, info->dest, rc1, errno);
984 /* Replicate a hard link */
985 int lr_link(struct lr_info *info)
992 lr_get_FID_PATH(status->ls_source, info->tfid, info->src, PATH_MAX);
993 rc = stat(info->src, &st);
997 for (info->target_no = 0; info->target_no < status->ls_num_targets;
1005 * The changelog record has the new parent directory FID and
1006 * name of the target file. So info->dest can be constructed
1007 * by getting the path of the new parent directory and
1008 * appending the target file name.
1010 rc1 = lr_get_path(info, info->pfid);
1011 lr_debug(rc1 ? 0 : DTRACE, "\tparent fid2path %s, %s, rc=%d\n",
1012 info->path, info->name, rc1);
1015 snprintf(info->dest, sizeof(info->dest), "%s/%s/%s",
1016 status->ls_targets[info->target_no],
1017 info->path, info->name);
1018 lr_debug(DINFO, "link destination is %s\n", info->dest);
1021 /* Search through the hardlinks to get the src */
1022 for (i = 0; i < st.st_nlink && info->src[0] == 0; i++) {
1025 rc1 = lr_get_path_ln(info, info->tfid, i);
1026 lr_debug(rc1 ? 0:DTRACE, "\tfid2path %s, %s, %d rc=%d\n",
1027 info->path, info->name, i, rc1);
1032 * Compare the path of target FID with info->dest
1033 * to find out info->src.
1035 len = sizeof(status->ls_targets[info->target_no]) +
1037 char srcpath[len + 1];
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, sizeof(info->src), "%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, sizeof(info->dest), "%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, sizeof(info->dest), "%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, sizeof(info->dest), "%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, sizeof(backupfile), "%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, sizeof(rsync), "%s -p %s", TYPE, RSYNC);
1415 fp = popen(rsync, "r");
1419 if (fgets(rsync, sizeof(rsync), 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, sizeof(rsync_ver), "%s --version", rsync);
1431 fp = popen(rsync_ver, "r");
1435 if (fgets(rsync_ver, sizeof(rsync_ver), 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 "
1506 if (status->ls_mdt_device[0] == '\0') {
1509 len = snprintf(status->ls_mdt_device,
1510 sizeof(status->ls_mdt_device), "%s%s",
1511 status->ls_source_fs, DEFAULT_MDT);
1512 if (len >= sizeof(status->ls_mdt_device)) {
1518 ext = calloc(1, sizeof(struct lr_info));
1524 for (i = 0, xattr_not_supp = 0; i < status->ls_num_targets; i++) {
1525 snprintf(info->dest, sizeof(info->dest), "%s/%s",
1526 status->ls_targets[i], SPECIAL_DIR);
1527 rc = mkdir(info->dest, 0777);
1528 if (rc == -1 && errno != EEXIST) {
1529 fprintf(stderr, "Error writing to target path %s.\n",
1530 status->ls_targets[i]);
1534 rc = llistxattr(info->src, info->xlist, info->xsize);
1535 if (rc == -1 && errno == ENOTSUP) {
1536 fprintf(stderr, "xattrs not supported on %s\n",
1537 status->ls_targets[i]);
1541 if (xattr_not_supp == status->ls_num_targets)
1542 /* None of the targets support xattrs. */
1545 lr_print_status(info);
1547 /* Open changelogs for consumption*/
1548 rc = llapi_changelog_start(&changelog_priv,
1549 CHANGELOG_FLAG_BLOCK |
1550 CHANGELOG_FLAG_JOBID |
1551 CHANGELOG_FLAG_EXTRA_FLAGS,
1552 status->ls_mdt_device, status->ls_last_recno);
1554 fprintf(stderr, "Error opening changelog file for fs %s.\n",
1555 status->ls_source_fs);
1559 rc = llapi_changelog_set_xflags(changelog_priv,
1560 CHANGELOG_EXTRA_FLAG_UIDGID |
1561 CHANGELOG_EXTRA_FLAG_NID |
1562 CHANGELOG_EXTRA_FLAG_OMODE);
1564 fprintf(stderr, "Error setting xflag in changelog for fs %s.\n",
1565 status->ls_source_fs);
1569 while (!quit && lr_parse_line(changelog_priv, info) == 0) {
1571 if (info->type == CL_RENAME && !info->is_extended) {
1572 /* Newer rename operations extends changelog to store
1573 * source file information, but old changelog has
1576 if (lr_parse_line(changelog_priv, ext) != 0)
1578 memcpy(info->sfid, info->tfid, sizeof(info->sfid));
1579 memcpy(info->spfid, info->pfid, sizeof(info->spfid));
1580 memcpy(info->tfid, ext->tfid, sizeof(info->tfid));
1581 memcpy(info->pfid, ext->pfid, sizeof(info->pfid));
1582 strlcpy(info->sname, info->name, sizeof(info->sname));
1583 strlcpy(info->name, ext->name, sizeof(info->name));
1584 info->is_extended = 1;
1590 lr_debug(DTRACE, "***** Start %lld %s (%d) %s %s %s *****\n",
1591 info->recno, changelog_type2str(info->type),
1592 info->type, info->tfid, info->pfid, info->name);
1594 switch(info->type) {
1599 rc = lr_create(info);
1603 rc = lr_remove(info);
1613 rc = lr_setattr(info);
1616 rc = lr_setxattr(info);
1623 /* Nothing needs to be done for these entries */
1629 lr_debug(DTRACE, "##### End %lld %s (%d) %s %s %s rc=%d #####\n",
1630 info->recno, changelog_type2str(info->type),
1631 info->type, info->tfid, info->pfid, info->name, rc);
1633 if (rc && rc != -ENOENT) {
1634 lr_print_failure(info, rc);
1639 lr_clear_cl(info, 0);
1641 bzero(info, sizeof(struct lr_info));
1642 bzero(ext, sizeof(struct lr_info));
1646 llapi_changelog_fini(&changelog_priv);
1648 if (errors || verbose)
1649 printf("Errors: %d\n", errors);
1651 /* Clear changelog records used so far */
1652 lr_clear_cl(info, 1);
1655 printf("lustre_rsync took %ld seconds\n", time(NULL) - start);
1656 printf("Changelog records consumed: %lld\n", rec_count);
1671 termination_handler (int signum)
1673 /* Set a flag for the replicator to gracefully shutdown */
1675 printf("lustre_rsync halting.\n");
1678 int main(int argc, char *argv[])
1684 if ((rc = lr_init_status()) != 0)
1687 while ((rc = getopt_long(argc, argv, "as:t:m:u:l:vx:zc:ry:n:d:D:",
1688 long_opts, NULL)) >= 0) {
1691 /* Assume absolute paths */
1695 /* Assume absolute paths */
1696 strlcpy(status->ls_source, optarg,
1697 sizeof(status->ls_source));
1700 status->ls_num_targets++;
1702 if (numtargets != status->ls_num_targets) {
1703 /* Targets were read from a log
1704 file. The ones specified on the
1705 command line take precedence. The
1706 ones from the log file will be
1708 status->ls_num_targets = numtargets;
1710 newsize = sizeof (struct lustre_rsync_status) +
1711 (status->ls_num_targets * (PATH_MAX + 1));
1712 if (status->ls_size != newsize) {
1713 status->ls_size = newsize;
1714 status = lr_grow_buf(status, newsize);
1718 strlcpy(status->ls_targets[status->ls_num_targets - 1],
1719 optarg, sizeof(status->ls_targets[0]));
1722 strlcpy(status->ls_mdt_device, optarg,
1723 sizeof(status->ls_mdt_device));
1726 strlcpy(status->ls_registration, optarg,
1727 sizeof(status->ls_registration));
1731 (void) lr_read_log();
1737 if (strcmp("no", optarg) == 0) {
1739 } else if (strcmp("yes", optarg) != 0) {
1740 printf("Invalid parameter %s. "
1741 "Specify --xattr=no or --xattr=yes\n",
1750 /* Undocumented option cl-clear */
1751 if (strcmp("no", optarg) == 0) {
1753 } else if (strcmp("yes", optarg) != 0) {
1754 printf("Invalid parameter %s. "
1755 "Specify --cl-clear=no "
1756 "or --cl-clear=yes\n",
1762 /* Undocumented option use-rsync */
1766 /* Undocumented option rsync-threshold */
1767 rsync_threshold = atol(optarg);
1770 /* Undocumented option start-recno */
1771 status->ls_last_recno = atol(optarg);
1774 /* Undocumented option debug */
1775 debug = atoi(optarg);
1776 if (debug < 0 || debug > 2)
1780 /* Undocumented option debug log file */
1781 if (debug_log != NULL)
1783 debug_log = fopen(optarg, "a");
1784 if (debug_log == NULL) {
1785 printf("Cannot open %s for debug log\n",
1791 fprintf(stderr, "error: %s: option '%s' "
1792 "unrecognized.\n", argv[0], argv[optind - 1]);
1798 if (status->ls_last_recno == -1)
1799 status->ls_last_recno = 0;
1800 if (strnlen(status->ls_registration, LR_NAME_MAXLEN) == 0) {
1801 /* No registration ID was passed in. */
1802 printf("Please specify changelog consumer registration id.\n");
1806 if (strnlen(status->ls_source, PATH_MAX) == 0) {
1807 fprintf(stderr, "Please specify the source path.\n");
1811 if (strnlen(status->ls_targets[0], PATH_MAX) == 0) {
1812 fprintf(stderr, "Please specify the target path.\n");
1817 rc = lr_locate_rsync();
1818 if (use_rsync && rc != 0) {
1819 fprintf(stderr, "Error: unable to locate %s.\n", RSYNC);
1823 signal(SIGINT, termination_handler);
1824 signal(SIGHUP, termination_handler);
1825 signal(SIGTERM, termination_handler);
1827 rc = lr_replicate();
1829 if (debug_log != NULL)