2 * Routines to support hard-linking.
4 * Copyright (C) 1996 Andrew Tridgell
5 * Copyright (C) 1996 Paul Mackerras
6 * Copyright (C) 2002 Martin Pool <mbp@samba.org>
7 * Copyright (C) 2004-2007 Wayne Davison
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation; either version 3 of the License, or
12 * (at your option) any later version.
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
17 * GNU General Public License for more details.
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, visit the http://fsf.org website.
28 extern int inc_recurse
;
31 extern int preserve_acls
;
32 extern int make_backups
;
33 extern int protocol_version
;
34 extern int remove_source_files
;
35 extern int stdout_format_has_i
;
36 extern int maybe_ATTRS_REPORT
;
37 extern int unsort_ndx
;
38 extern char *basis_dir
[];
39 extern struct file_list
*cur_flist
;
41 #ifdef SUPPORT_HARD_LINKS
43 /* Starting with protocol 30, we use a simple hashtable on the sending side
44 * for hashing the st_dev and st_ino info. The receiving side gets told
45 * (via flags and a "group index") which items are hard-linked together, so
46 * we can avoid the pool of dev+inode data. For incremental recursion mode,
47 * the receiver will use a ndx hash to remember old pathnames. */
49 static struct hashtable
*dev_tbl
;
51 static struct hashtable
*prior_hlinks
;
53 static struct file_list
*hlink_flist
;
55 void init_hard_links(void)
57 if (am_sender
|| protocol_version
< 30)
58 dev_tbl
= hashtable_create(16, SIZEOF_INT64
== 8);
60 prior_hlinks
= hashtable_create(1024, 0);
63 struct ht_int64_node
*idev_find(int64 dev
, int64 ino
)
65 static struct ht_int64_node
*dev_node
= NULL
;
66 struct hashtable
*tbl
;
68 if (!dev_node
|| dev_node
->key
!= dev
) {
69 /* We keep a separate hash table of inodes for every device. */
70 dev_node
= hashtable_find(dev_tbl
, dev
, 1);
71 if (!(tbl
= dev_node
->data
))
72 tbl
= dev_node
->data
= hashtable_create(512, SIZEOF_INT64
== 8);
76 return hashtable_find(tbl
, ino
, 1);
79 void idev_destroy(void)
83 for (i
= 0; i
< dev_tbl
->size
; i
++) {
84 struct ht_int32_node
*node
= HT_NODE(dev_tbl
, dev_tbl
->nodes
, i
);
86 hashtable_destroy(node
->data
);
89 hashtable_destroy(dev_tbl
);
92 static int hlink_compare_gnum(int *int1
, int *int2
)
94 struct file_struct
*f1
= hlink_flist
->sorted
[*int1
];
95 struct file_struct
*f2
= hlink_flist
->sorted
[*int2
];
96 int32 gnum1
= F_HL_GNUM(f1
);
97 int32 gnum2
= F_HL_GNUM(f2
);
100 return gnum1
> gnum2
? 1 : -1;
102 return *int1
> *int2
? 1 : -1;
105 static void match_gnums(int32
*ndx_list
, int ndx_count
)
108 struct file_struct
*file
, *file_next
;
109 struct ht_int32_node
*node
= NULL
;
110 int32 gnum
, gnum_next
;
112 qsort(ndx_list
, ndx_count
, sizeof ndx_list
[0],
113 (int (*)()) hlink_compare_gnum
);
115 for (from
= 0; from
< ndx_count
; from
++) {
116 file
= hlink_flist
->sorted
[ndx_list
[from
]];
117 gnum
= F_HL_GNUM(file
);
119 node
= hashtable_find(prior_hlinks
, gnum
, 1);
121 node
->data
= new_array0(char, 5);
122 assert(gnum
>= hlink_flist
->ndx_start
);
123 file
->flags
|= FLAG_HLINK_FIRST
;
125 } else if (CVAL(node
->data
, 0) == 0) {
126 struct file_list
*flist
;
127 struct file_struct
*fp
;
128 prev
= IVAL(node
->data
, 1);
129 flist
= flist_for_ndx(prev
);
130 assert(flist
!= NULL
);
131 fp
= flist
->files
[prev
- flist
->ndx_start
];
132 fp
->flags
&= ~FLAG_HLINK_LAST
;
136 file
->flags
|= FLAG_HLINK_FIRST
;
139 for ( ; from
< ndx_count
-1; file
= file_next
, gnum
= gnum_next
, from
++) { /*SHARED ITERATOR*/
140 file_next
= hlink_flist
->sorted
[ndx_list
[from
+1]];
141 gnum_next
= F_HL_GNUM(file_next
);
142 if (gnum
!= gnum_next
)
144 F_HL_PREV(file
) = prev
;
145 /* The linked list must use raw ndx values. */
149 prev
= ndx_list
[from
] + hlink_flist
->ndx_start
;
151 if (prev
< 0 && !inc_recurse
) {
152 /* Disable hard-link bit and set DONE so that
153 * HLINK_BUMP()-dependent values are unaffected. */
154 file
->flags
&= ~(FLAG_HLINKED
| FLAG_HLINK_FIRST
);
155 file
->flags
|= FLAG_HLINK_DONE
;
159 file
->flags
|= FLAG_HLINK_LAST
;
160 F_HL_PREV(file
) = prev
;
161 if (inc_recurse
&& CVAL(node
->data
, 0) == 0) {
165 prev
= ndx_list
[from
] + hlink_flist
->ndx_start
;
166 SIVAL(node
->data
, 1, prev
);
171 /* Analyze the hard-links in the file-list by creating a list of all the
172 * items that have hlink data, sorting them, and matching up identical
173 * values into clusters. These will be a single linked list from last
174 * to first when we're done. */
175 void match_hard_links(struct file_list
*flist
)
177 int i
, ndx_count
= 0;
180 if (!(ndx_list
= new_array(int32
, flist
->used
)))
181 out_of_memory("match_hard_links");
183 for (i
= 0; i
< flist
->used
; i
++) {
184 if (F_IS_HLINKED(flist
->sorted
[i
]))
185 ndx_list
[ndx_count
++] = i
;
191 match_gnums(ndx_list
, ndx_count
);
194 if (protocol_version
< 30)
198 static int maybe_hard_link(struct file_struct
*file
, int ndx
,
199 const char *fname
, int statret
, stat_x
*sxp
,
200 const char *oldname
, STRUCT_STAT
*old_stp
,
201 const char *realname
, int itemizing
, enum logcode code
)
204 if (sxp
->st
.st_dev
== old_stp
->st_dev
205 && sxp
->st
.st_ino
== old_stp
->st_ino
) {
207 itemize(fname
, file
, ndx
, statret
, sxp
,
208 ITEM_LOCAL_CHANGE
| ITEM_XNAME_FOLLOWS
,
211 if (verbose
> 1 && maybe_ATTRS_REPORT
)
212 rprintf(FCLIENT
, "%s is uptodate\n", fname
);
213 file
->flags
|= FLAG_HLINK_DONE
;
216 if (make_backups
> 0) {
217 if (!make_backup(fname
))
219 } else if (robust_unlink(fname
)) {
220 rsyserr(FERROR
, errno
, "unlink %s failed",
226 if (hard_link_one(file
, fname
, oldname
, 0)) {
228 itemize(fname
, file
, ndx
, statret
, sxp
,
229 ITEM_LOCAL_CHANGE
| ITEM_XNAME_FOLLOWS
, 0,
232 if (code
!= FNONE
&& verbose
)
233 rprintf(code
, "%s => %s\n", fname
, realname
);
239 /* Figure out if a prior entry is still there or if we just have a
240 * cached name for it. Never called with a FLAG_HLINK_FIRST entry. */
241 static char *check_prior(int prev_ndx
, int gnum
, struct file_list
**flist_p
)
243 struct file_list
*flist
= flist_for_ndx(prev_ndx
);
244 struct ht_int32_node
*node
;
251 node
= hashtable_find(prior_hlinks
, gnum
, 0);
252 assert(node
!= NULL
&& node
->data
);
253 assert(CVAL(node
->data
, 0) != 0);
257 /* Only called if FLAG_HLINKED is set and FLAG_HLINK_FIRST is not. Returns:
258 * 0 = process the file, 1 = skip the file, -1 = error occurred. */
259 int hard_link_check(struct file_struct
*file
, int ndx
, const char *fname
,
260 int statret
, stat_x
*sxp
, int itemizing
,
264 char namebuf
[MAXPATHLEN
], altbuf
[MAXPATHLEN
];
265 char *realname
, *prev_name
;
266 struct file_list
*flist
;
267 int gnum
= inc_recurse
? F_HL_GNUM(file
) : -1;
268 int prev_ndx
= F_HL_PREV(file
);
270 prev_name
= realname
= check_prior(prev_ndx
, gnum
, &flist
);
273 struct file_struct
*prev_file
= flist
->files
[prev_ndx
- flist
->ndx_start
];
275 /* Is the previous link is not complete yet? */
276 if (!(prev_file
->flags
& FLAG_HLINK_DONE
)) {
277 /* Is the previous link being transferred? */
278 if (prev_file
->flags
& FLAG_FILE_SENT
) {
279 /* Add ourselves to the list of files that will be
280 * updated when the transfer completes, and mark
281 * ourself as waiting for the transfer. */
282 F_HL_PREV(file
) = F_HL_PREV(prev_file
);
283 F_HL_PREV(prev_file
) = ndx
;
284 file
->flags
|= FLAG_FILE_SENT
;
285 cur_flist
->in_progress
++;
291 /* There is a finished file to link with! */
292 if (!(prev_file
->flags
& FLAG_HLINK_FIRST
)) {
293 /* The previous previous is FIRST when prev is not. */
294 prev_ndx
= F_HL_PREV(prev_file
);
295 prev_name
= realname
= check_prior(prev_ndx
, gnum
, &flist
);
296 /* Update our previous pointer to point to the FIRST. */
297 F_HL_PREV(file
) = prev_ndx
;
303 prev_file
= flist
->files
[prev_ndx
- flist
->ndx_start
];
304 /* F_HL_PREV() is alt_dest value when DONE && FIRST. */
305 alt_dest
= F_HL_PREV(prev_file
);
307 if (alt_dest
>= 0 && dry_run
) {
308 pathjoin(namebuf
, MAXPATHLEN
, basis_dir
[alt_dest
],
309 f_name(prev_file
, NULL
));
311 realname
= f_name(prev_file
, altbuf
);
313 prev_name
= f_name(prev_file
, namebuf
);
314 realname
= prev_name
;
319 if (link_stat(prev_name
, &prev_st
, 0) < 0) {
320 rsyserr(FERROR
, errno
, "stat %s failed",
321 full_fname(prev_name
));
325 if (statret
< 0 && basis_dir
[0] != NULL
) {
326 /* If we match an alt-dest item, we don't output this as a change. */
327 char cmpbuf
[MAXPATHLEN
];
331 alt_sx
.acc_acl
= alt_sx
.def_acl
= NULL
;
334 pathjoin(cmpbuf
, MAXPATHLEN
, basis_dir
[j
], fname
);
335 if (link_stat(cmpbuf
, &alt_sx
.st
, 0) < 0)
338 if (prev_st
.st_dev
!= alt_sx
.st
.st_dev
339 || prev_st
.st_ino
!= alt_sx
.st
.st_ino
)
342 if (stdout_format_has_i
== 0
343 || (verbose
< 2 && stdout_format_has_i
< 2)) {
346 if (verbose
> 1 && maybe_ATTRS_REPORT
)
347 rprintf(FCLIENT
, "%s is uptodate\n", fname
);
351 if (!unchanged_file(cmpbuf
, file
, &alt_sx
.st
))
354 if (unchanged_attrs(cmpbuf
, file
, &alt_sx
))
356 } while (basis_dir
[++j
] != NULL
);
360 if (preserve_acls
&& !S_ISLNK(file
->mode
)) {
361 if (!ACL_READY(*sxp
))
362 get_acl(cmpbuf
, sxp
);
364 sxp
->acc_acl
= alt_sx
.acc_acl
;
365 sxp
->def_acl
= alt_sx
.def_acl
;
371 else if (preserve_acls
)
376 if (maybe_hard_link(file
, ndx
, fname
, statret
, sxp
, prev_name
, &prev_st
,
377 realname
, itemizing
, code
) < 0)
380 if (remove_source_files
== 1 && do_xfers
)
381 send_msg_int(MSG_SUCCESS
, ndx
);
386 int hard_link_one(struct file_struct
*file
, const char *fname
,
387 const char *oldname
, int terse
)
389 if (do_link(oldname
, fname
) < 0) {
397 rsyserr(code
, errno
, "link %s => %s failed",
398 full_fname(fname
), oldname
);
402 file
->flags
|= FLAG_HLINK_DONE
;
407 void finish_hard_link(struct file_struct
*file
, const char *fname
, int fin_ndx
,
408 STRUCT_STAT
*stp
, int itemizing
, enum logcode code
,
413 char alt_name
[MAXPATHLEN
], *prev_name
;
414 const char *our_name
;
415 struct file_list
*flist
;
416 int prev_statret
, ndx
, prev_ndx
= F_HL_PREV(file
);
418 if (stp
== NULL
&& prev_ndx
>= 0) {
419 if (link_stat(fname
, &st
, 0) < 0) {
420 rsyserr(FERROR
, errno
, "stat %s failed",
427 /* FIRST combined with DONE means we were the first to get done. */
428 file
->flags
|= FLAG_HLINK_FIRST
| FLAG_HLINK_DONE
;
429 F_HL_PREV(file
) = alt_dest
;
430 if (alt_dest
>= 0 && dry_run
) {
431 pathjoin(alt_name
, MAXPATHLEN
, basis_dir
[alt_dest
],
438 prev_sx
.acc_acl
= prev_sx
.def_acl
= NULL
;
441 while ((ndx
= prev_ndx
) >= 0) {
443 flist
= flist_for_ndx(ndx
);
444 assert(flist
!= NULL
);
445 file
= flist
->files
[ndx
- flist
->ndx_start
];
446 file
->flags
= (file
->flags
& ~FLAG_HLINK_FIRST
) | FLAG_HLINK_DONE
;
447 prev_ndx
= F_HL_PREV(file
);
448 F_HL_PREV(file
) = fin_ndx
;
449 prev_name
= f_name(file
, NULL
);
450 prev_statret
= link_stat(prev_name
, &prev_sx
.st
, 0);
451 val
= maybe_hard_link(file
, ndx
, prev_name
, prev_statret
, &prev_sx
,
452 our_name
, stp
, fname
, itemizing
, code
);
453 flist
->in_progress
--;
460 if (remove_source_files
== 1 && do_xfers
)
461 send_msg_int(MSG_SUCCESS
, ndx
);
465 int gnum
= F_HL_GNUM(file
);
466 struct ht_int32_node
*node
= hashtable_find(prior_hlinks
, gnum
, 0);
467 assert(node
!= NULL
&& node
->data
!= NULL
);
468 assert(CVAL(node
->data
, 0) == 0);
470 if (!(node
->data
= strdup(our_name
)))
471 out_of_memory("finish_hard_link");