1/*-
2 * Copyright (c) 1992 Keith Muller.
3 * Copyright (c) 1992, 1993
4 * The Regents of the University of California. All rights reserved.
5 *
6 * This code is derived from software contributed to Berkeley by
7 * Keith Muller of the University of California, San Diego.
8 *
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
11 * are met:
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. All advertising materials mentioning features or use of this software
18 * must display the following acknowledgement:
19 * This product includes software developed by the University of
20 * California, Berkeley and its contributors.
21 * 4. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
36 */
37
38#ifndef lint
39#if 0
40static char sccsid[] = "@(#)file_subs.c 8.1 (Berkeley) 5/31/93";
41#endif
42static const char rcsid[] =
|
43 "$FreeBSD: head/bin/pax/file_subs.c 90110 2002-02-02 06:48:10Z imp $";
|
| 43 "$FreeBSD: head/bin/pax/file_subs.c 90113 2002-02-02 07:07:59Z imp $"; |
44#endif /* not lint */
45
46#include <sys/types.h>
47#include <sys/time.h>
48#include <sys/stat.h>
49#include <unistd.h>
50#include <fcntl.h>
51#include <string.h>
52#include <stdio.h>
53#include <errno.h>
54#include <sys/uio.h>
55#include <stdlib.h>
56#include "pax.h"
57#include "options.h"
58#include "extern.h"
59
60static int
|
61mk_link(register char *,register struct stat *,register char *, int);
|
| 61mk_link(char *,struct stat *,char *, int); |
62
63/*
64 * routines that deal with file operations such as: creating, removing;
65 * and setting access modes, uid/gid and times of files
66 */
67
68#define FILEBITS (S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO)
69#define SETBITS (S_ISUID | S_ISGID)
70#define ABITS (FILEBITS | SETBITS)
71
72/*
73 * file_creat()
74 * Create and open a file.
75 * Return:
76 * file descriptor or -1 for failure
77 */
78
79int
|
80file_creat(register ARCHD *arcn)
|
| 80file_creat(ARCHD *arcn) |
81{
82 int fd = -1;
83 mode_t file_mode;
84 int oerrno;
85
86 /*
87 * assume file doesn't exist, so just try to create it, most times this
88 * works. We have to take special handling when the file does exist. To
89 * detect this, we use O_EXCL. For example when trying to create a
90 * file and a character device or fifo exists with the same name, we
91 * can accidently open the device by mistake (or block waiting to open)
92 * If we find that the open has failed, then figure spend the effort to
93 * figure out why. This strategy was found to have better average
94 * performance in common use than checking the file (and the path)
95 * first with lstat.
96 */
97 file_mode = arcn->sb.st_mode & FILEBITS;
98 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL,
99 file_mode)) >= 0)
100 return(fd);
101
102 /*
103 * the file seems to exist. First we try to get rid of it (found to be
104 * the second most common failure when traced). If this fails, only
105 * then we go to the expense to check and create the path to the file
106 */
107 if (unlnk_exist(arcn->name, arcn->type) != 0)
108 return(-1);
109
110 for (;;) {
111 /*
112 * try to open it again, if this fails, check all the nodes in
113 * the path and give it a final try. if chk_path() finds that
114 * it cannot fix anything, we will skip the last attempt
115 */
116 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC,
117 file_mode)) >= 0)
118 break;
119 oerrno = errno;
120 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
121 syswarn(1, oerrno, "Unable to create %s", arcn->name);
122 return(-1);
123 }
124 }
125 return(fd);
126}
127
128/*
129 * file_close()
130 * Close file descriptor to a file just created by pax. Sets modes,
131 * ownership and times as required.
132 * Return:
133 * 0 for success, -1 for failure
134 */
135
136void
|
137file_close(register ARCHD *arcn, int fd)
|
| 137file_close(ARCHD *arcn, int fd) |
138{
139 int res = 0;
140
141 if (fd < 0)
142 return;
143 if (close(fd) < 0)
144 syswarn(0, errno, "Unable to close file descriptor on %s",
145 arcn->name);
146
147 /*
148 * set owner/groups first as this may strip off mode bits we want
149 * then set file permission modes. Then set file access and
150 * modification times.
151 */
152 if (pids)
153 res = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid);
154
155 /*
156 * IMPORTANT SECURITY NOTE:
157 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
158 * set uid/gid bits
159 */
160 if (!pmode || res)
161 arcn->sb.st_mode &= ~(SETBITS);
162 if (pmode)
163 set_pmode(arcn->name, arcn->sb.st_mode);
164 if (patime || pmtime)
165 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
166}
167
168/*
169 * lnk_creat()
170 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
171 * must exist;
172 * Return:
173 * 0 if ok, -1 otherwise
174 */
175
176int
|
177lnk_creat(register ARCHD *arcn)
|
| 177lnk_creat(ARCHD *arcn) |
178{
179 struct stat sb;
180
181 /*
182 * we may be running as root, so we have to be sure that link target
183 * is not a directory, so we lstat and check
184 */
185 if (lstat(arcn->ln_name, &sb) < 0) {
186 syswarn(1,errno,"Unable to link to %s from %s", arcn->ln_name,
187 arcn->name);
188 return(-1);
189 }
190
191 if (S_ISDIR(sb.st_mode)) {
192 paxwarn(1, "A hard link to the directory %s is not allowed",
193 arcn->ln_name);
194 return(-1);
195 }
196
197 return(mk_link(arcn->ln_name, &sb, arcn->name, 0));
198}
199
200/*
201 * cross_lnk()
202 * Create a hard link to arcn->org_name from arcn->name. Only used in copy
203 * with the -l flag. No warning or error if this does not succeed (we will
204 * then just create the file)
205 * Return:
206 * 1 if copy() should try to create this file node
207 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
208 */
209
210int
|
211cross_lnk(register ARCHD *arcn)
|
| 211cross_lnk(ARCHD *arcn) |
212{
213 /*
214 * try to make a link to original file (-l flag in copy mode). make sure
215 * we do not try to link to directories in case we are running as root
216 * (and it might succeed).
217 */
218 if (arcn->type == PAX_DIR)
219 return(1);
220 return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
221}
222
223/*
224 * chk_same()
225 * In copy mode if we are not trying to make hard links between the src
226 * and destinations, make sure we are not going to overwrite ourselves by
227 * accident. This slows things down a little, but we have to protect all
228 * those people who make typing errors.
229 * Return:
230 * 1 the target does not exist, go ahead and copy
231 * 0 skip it file exists (-k) or may be the same as source file
232 */
233
234int
|
235chk_same(register ARCHD *arcn)
|
| 235chk_same(ARCHD *arcn) |
236{
237 struct stat sb;
238
239 /*
240 * if file does not exist, return. if file exists and -k, skip it
241 * quietly
242 */
243 if (lstat(arcn->name, &sb) < 0)
244 return(1);
245 if (kflag)
246 return(0);
247
248 /*
249 * better make sure the user does not have src == dest by mistake
250 */
251 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
252 paxwarn(1, "Unable to copy %s, file would overwrite itself",
253 arcn->name);
254 return(0);
255 }
256 return(1);
257}
258
259/*
260 * mk_link()
261 * try to make a hard link between two files. if ign set, we do not
262 * complain.
263 * Return:
264 * 0 if successful (or we are done with this file but no error, such as
265 * finding the from file exists and the user has set -k).
266 * 1 when ign was set to indicates we could not make the link but we
267 * should try to copy/extract the file as that might work (and is an
268 * allowed option). -1 an error occurred.
269 */
270
271static int
|
272mk_link(register char *to, register struct stat *to_sb, register char *from,
|
| 272mk_link(char *to, struct stat *to_sb, char *from, |
273 int ign)
274{
275 struct stat sb;
276 int oerrno;
277
278 /*
279 * if from file exists, it has to be unlinked to make the link. If the
280 * file exists and -k is set, skip it quietly
281 */
282 if (lstat(from, &sb) == 0) {
283 if (kflag)
284 return(0);
285
286 /*
287 * make sure it is not the same file, protect the user
288 */
289 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
290 paxwarn(1, "Unable to link file %s to itself", to);
291 return(-1);;
292 }
293
294 /*
295 * try to get rid of the file, based on the type
296 */
297 if (S_ISDIR(sb.st_mode)) {
298 if (rmdir(from) < 0) {
299 syswarn(1, errno, "Unable to remove %s", from);
300 return(-1);
301 }
302 } else if (unlink(from) < 0) {
303 if (!ign) {
304 syswarn(1, errno, "Unable to remove %s", from);
305 return(-1);
306 }
307 return(1);
308 }
309 }
310
311 /*
312 * from file is gone (or did not exist), try to make the hard link.
313 * if it fails, check the path and try it again (if chk_path() says to
314 * try again)
315 */
316 for (;;) {
317 if (link(to, from) == 0)
318 break;
319 oerrno = errno;
320 if (!nodirs && chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
321 continue;
322 if (!ign) {
323 syswarn(1, oerrno, "Could not link to %s from %s", to,
324 from);
325 return(-1);
326 }
327 return(1);
328 }
329
330 /*
331 * all right the link was made
332 */
333 return(0);
334}
335
336/*
337 * node_creat()
338 * create an entry in the file system (other than a file or hard link).
339 * If successful, sets uid/gid modes and times as required.
340 * Return:
341 * 0 if ok, -1 otherwise
342 */
343
344int
|
345node_creat(register ARCHD *arcn)
|
| 345node_creat(ARCHD *arcn) |
346{
|
347 register int res;
348 register int ign = 0;
349 register int oerrno;
350 register int pass = 0;
|
347 int res;
348 int ign = 0;
349 int oerrno;
350 int pass = 0; |
351 mode_t file_mode;
352 struct stat sb;
353
354 /*
355 * create node based on type, if that fails try to unlink the node and
356 * try again. finally check the path and try again. As noted in the
357 * file and link creation routines, this method seems to exhibit the
358 * best performance in general use workloads.
359 */
360 file_mode = arcn->sb.st_mode & FILEBITS;
361
362 for (;;) {
363 switch(arcn->type) {
364 case PAX_DIR:
365 res = mkdir(arcn->name, file_mode);
366 if (ign)
367 res = 0;
368 break;
369 case PAX_CHR:
370 file_mode |= S_IFCHR;
371 res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
372 break;
373 case PAX_BLK:
374 file_mode |= S_IFBLK;
375 res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
376 break;
377 case PAX_FIF:
378 res = mkfifo(arcn->name, file_mode);
379 break;
380 case PAX_SCK:
381 /*
382 * Skip sockets, operation has no meaning under BSD
383 */
384 paxwarn(0,
385 "%s skipped. Sockets cannot be copied or extracted",
386 arcn->name);
387 return(-1);
388 case PAX_SLK:
389 res = symlink(arcn->ln_name, arcn->name);
390 break;
391 case PAX_CTG:
392 case PAX_HLK:
393 case PAX_HRG:
394 case PAX_REG:
395 default:
396 /*
397 * we should never get here
398 */
399 paxwarn(0, "%s has an unknown file type, skipping",
400 arcn->name);
401 return(-1);
402 }
403
404 /*
405 * if we were able to create the node break out of the loop,
406 * otherwise try to unlink the node and try again. if that
407 * fails check the full path and try a final time.
408 */
409 if (res == 0)
410 break;
411
412 /*
413 * we failed to make the node
414 */
415 oerrno = errno;
416 if ((ign = unlnk_exist(arcn->name, arcn->type)) < 0)
417 return(-1);
418
419 if (++pass <= 1)
420 continue;
421
422 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
423 syswarn(1, oerrno, "Could not create: %s", arcn->name);
424 return(-1);
425 }
426 }
427
428 /*
429 * we were able to create the node. set uid/gid, modes and times
430 */
431 if (pids)
432 res = ((arcn->type == PAX_SLK) ?
433 set_lids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid) :
434 set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid));
435 else
436 res = 0;
437
438 /*
439 * symlinks are done now.
440 */
441 if (arcn->type == PAX_SLK)
442 return(0);
443
444 /*
445 * IMPORTANT SECURITY NOTE:
446 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
447 * set uid/gid bits
448 */
449 if (!pmode || res)
450 arcn->sb.st_mode &= ~(SETBITS);
451 if (pmode)
452 set_pmode(arcn->name, arcn->sb.st_mode);
453
454 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
455 /*
456 * Dirs must be processed again at end of extract to set times
457 * and modes to agree with those stored in the archive. However
458 * to allow extract to continue, we may have to also set owner
459 * rights. This allows nodes in the archive that are children
460 * of this directory to be extracted without failure. Both time
461 * and modes will be fixed after the entire archive is read and
462 * before pax exits.
463 */
464 if (access(arcn->name, R_OK | W_OK | X_OK) < 0) {
465 if (lstat(arcn->name, &sb) < 0) {
466 syswarn(0, errno,"Could not access %s (stat)",
467 arcn->name);
468 set_pmode(arcn->name,file_mode | S_IRWXU);
469 } else {
470 /*
471 * We have to add rights to the dir, so we make
472 * sure to restore the mode. The mode must be
473 * restored AS CREATED and not as stored if
474 * pmode is not set.
475 */
476 set_pmode(arcn->name,
477 ((sb.st_mode & FILEBITS) | S_IRWXU));
478 if (!pmode)
479 arcn->sb.st_mode = sb.st_mode;
480 }
481
482 /*
483 * we have to force the mode to what was set here,
484 * since we changed it from the default as created.
485 */
486 add_dir(arcn->name, arcn->nlen, &(arcn->sb), 1);
487 } else if (pmode || patime || pmtime)
488 add_dir(arcn->name, arcn->nlen, &(arcn->sb), 0);
489 }
490
491 if (patime || pmtime)
492 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
493 return(0);
494}
495
496/*
497 * unlnk_exist()
498 * Remove node from file system with the specified name. We pass the type
499 * of the node that is going to replace it. When we try to create a
500 * directory and find that it already exists, we allow processing to
501 * continue as proper modes etc will always be set for it later on.
502 * Return:
503 * 0 is ok to proceed, no file with the specified name exists
504 * -1 we were unable to remove the node, or we should not remove it (-k)
505 * 1 we found a directory and we were going to create a directory.
506 */
507
508int
|
509unlnk_exist(register char *name, register int type)
|
| 509unlnk_exist(char *name, int type) |
510{
511 struct stat sb;
512
513 /*
514 * the file does not exist, or -k we are done
515 */
516 if (lstat(name, &sb) < 0)
517 return(0);
518 if (kflag)
519 return(-1);
520
521 if (S_ISDIR(sb.st_mode)) {
522 /*
523 * try to remove a directory, if it fails and we were going to
524 * create a directory anyway, tell the caller (return a 1)
525 */
526 if (rmdir(name) < 0) {
527 if (type == PAX_DIR)
528 return(1);
529 syswarn(1,errno,"Unable to remove directory %s", name);
530 return(-1);
531 }
532 return(0);
533 }
534
535 /*
536 * try to get rid of all non-directory type nodes
537 */
538 if (unlink(name) < 0) {
539 syswarn(1, errno, "Could not unlink %s", name);
540 return(-1);
541 }
542 return(0);
543}
544
545/*
546 * chk_path()
547 * We were trying to create some kind of node in the file system and it
548 * failed. chk_path() makes sure the path up to the node exists and is
549 * writeable. When we have to create a directory that is missing along the
550 * path somewhere, the directory we create will be set to the same
551 * uid/gid as the file has (when uid and gid are being preserved).
552 * NOTE: this routine is a real performance loss. It is only used as a
553 * last resort when trying to create entries in the file system.
554 * Return:
555 * -1 when it could find nothing it is allowed to fix.
556 * 0 otherwise
557 */
558
559int
|
560chk_path( register char *name, uid_t st_uid, gid_t st_gid)
|
| 560chk_path( char *name, uid_t st_uid, gid_t st_gid) |
561{
|
562 register char *spt = name;
|
| 562 char *spt = name; |
563 struct stat sb;
564 int retval = -1;
565
566 /*
567 * watch out for paths with nodes stored directly in / (e.g. /bozo)
568 */
569 if (*spt == '/')
570 ++spt;
571
572 for(;;) {
573 /*
574 * work foward from the first / and check each part of the path
575 */
576 spt = strchr(spt, '/');
577 if (spt == NULL)
578 break;
579 *spt = '\0';
580
581 /*
582 * if it exists we assume it is a directory, it is not within
583 * the spec (at least it seems to read that way) to alter the
584 * file system for nodes NOT EXPLICITLY stored on the archive.
585 * If that assumption is changed, you would test the node here
586 * and figure out how to get rid of it (probably like some
587 * recursive unlink()) or fix up the directory permissions if
588 * required (do an access()).
589 */
590 if (lstat(name, &sb) == 0) {
591 *(spt++) = '/';
592 continue;
593 }
594
595 /*
596 * the path fails at this point, see if we can create the
597 * needed directory and continue on
598 */
599 if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
600 *spt = '/';
601 retval = -1;
602 break;
603 }
604
605 /*
606 * we were able to create the directory. We will tell the
607 * caller that we found something to fix, and it is ok to try
608 * and create the node again.
609 */
610 retval = 0;
611 if (pids)
612 (void)set_ids(name, st_uid, st_gid);
613
614 /*
615 * make sure the user doen't have some strange umask that
616 * causes this newly created directory to be unusable. We fix
617 * the modes and restore them back to the creation default at
618 * the end of pax
619 */
620 if ((access(name, R_OK | W_OK | X_OK) < 0) &&
621 (lstat(name, &sb) == 0)) {
622 set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
623 add_dir(name, spt - name, &sb, 1);
624 }
625 *(spt++) = '/';
626 continue;
627 }
628 return(retval);
629}
630
631/*
632 * set_ftime()
633 * Set the access time and modification time for a named file. If frc is
634 * non-zero we force these times to be set even if the user did not
635 * request access and/or modification time preservation (this is also
636 * used by -t to reset access times).
637 * When ign is zero, only those times the user has asked for are set, the
638 * other ones are left alone. We do not assume the un-documented feature
639 * of many utimes() implementations that consider a 0 time value as a do
640 * not set request.
641 */
642
643void
644set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
645{
646 static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}};
647 struct stat sb;
648
649 tv[0].tv_sec = atime;
650 tv[1].tv_sec = mtime;
651 if (!frc && (!patime || !pmtime)) {
652 /*
653 * if we are not forcing, only set those times the user wants
654 * set. We get the current values of the times if we need them.
655 */
656 if (lstat(fnm, &sb) == 0) {
657 if (!patime)
658 tv[0].tv_sec = sb.st_atime;
659 if (!pmtime)
660 tv[1].tv_sec = sb.st_mtime;
661 } else
662 syswarn(0,errno,"Unable to obtain file stats %s", fnm);
663 }
664
665 /*
666 * set the times
667 */
668 if (utimes(fnm, tv) < 0)
669 syswarn(1, errno, "Access/modification time set failed on: %s",
670 fnm);
671 return;
672}
673
674/*
675 * set_ids()
676 * set the uid and gid of a file system node
677 * Return:
678 * 0 when set, -1 on failure
679 */
680
681int
682set_ids(char *fnm, uid_t uid, gid_t gid)
683{
684 if (chown(fnm, uid, gid) < 0) {
685 /*
686 * ignore EPERM unless in verbose mode or being run by root.
687 * if running as pax, POSIX requires a warning.
688 */
689 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
690 geteuid() == 0)
691 syswarn(1, errno, "Unable to set file uid/gid of %s",
692 fnm);
693 return(-1);
694 }
695 return(0);
696}
697
698/*
699 * set_lids()
700 * set the uid and gid of a file system node
701 * Return:
702 * 0 when set, -1 on failure
703 */
704
705int
706set_lids(char *fnm, uid_t uid, gid_t gid)
707{
708 if (lchown(fnm, uid, gid) < 0) {
709 /*
710 * ignore EPERM unless in verbose mode or being run by root.
711 * if running as pax, POSIX requires a warning.
712 */
713 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
714 geteuid() == 0)
715 syswarn(1, errno, "Unable to set file uid/gid of %s",
716 fnm);
717 return(-1);
718 }
719 return(0);
720}
721
722/*
723 * set_pmode()
724 * Set file access mode
725 */
726
727void
728set_pmode(char *fnm, mode_t mode)
729{
730 mode &= ABITS;
731 if (chmod(fnm, mode) < 0)
732 syswarn(1, errno, "Could not set permissions on %s", fnm);
733 return;
734}
735
736/*
737 * file_write()
738 * Write/copy a file (during copy or archive extract). This routine knows
739 * how to copy files with lseek holes in it. (Which are read as file
740 * blocks containing all 0's but do not have any file blocks associated
741 * with the data). Typical examples of these are files created by dbm
742 * variants (.pag files). While the file size of these files are huge, the
743 * actual storage is quite small (the files are sparse). The problem is
744 * the holes read as all zeros so are probably stored on the archive that
745 * way (there is no way to determine if the file block is really a hole,
746 * we only know that a file block of all zero's can be a hole).
747 * At this writing, no major archive format knows how to archive files
748 * with holes. However, on extraction (or during copy, -rw) we have to
749 * deal with these files. Without detecting the holes, the files can
750 * consume a lot of file space if just written to disk. This replacement
751 * for write when passed the basic allocation size of a file system block,
752 * uses lseek whenever it detects the input data is all 0 within that
753 * file block. In more detail, the strategy is as follows:
754 * While the input is all zero keep doing an lseek. Keep track of when we
755 * pass over file block boundries. Only write when we hit a non zero
756 * input. once we have written a file block, we continue to write it to
757 * the end (we stop looking at the input). When we reach the start of the
758 * next file block, start checking for zero blocks again. Working on file
759 * block boundries significantly reduces the overhead when copying files
760 * that are NOT very sparse. This overhead (when compared to a write) is
761 * almost below the measurement resolution on many systems. Without it,
762 * files with holes cannot be safely copied. It does has a side effect as
763 * it can put holes into files that did not have them before, but that is
764 * not a problem since the file contents are unchanged (in fact it saves
765 * file space). (Except on paging files for diskless clients. But since we
766 * cannot determine one of those file from here, we ignore them). If this
767 * ever ends up on a system where CTG files are supported and the holes
768 * are not desired, just do a conditional test in those routines that
769 * call file_write() and have it call write() instead. BEFORE CLOSING THE
770 * FILE, make sure to call file_flush() when the last write finishes with
771 * an empty block. A lot of file systems will not create an lseek hole at
772 * the end. In this case we drop a single 0 at the end to force the
773 * trailing 0's in the file.
774 * ---Parameters---
775 * rem: how many bytes left in this file system block
776 * isempt: have we written to the file block yet (is it empty)
777 * sz: basic file block allocation size
778 * cnt: number of bytes on this write
779 * str: buffer to write
780 * Return:
781 * number of bytes written, -1 on write (or lseek) error.
782 */
783
784int
|
785file_write(int fd, char *str, register int cnt, int *rem, int *isempt, int sz,
|
| 785file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz, |
786 char *name)
787{
|
788 register char *pt;
789 register char *end;
790 register int wcnt;
791 register char *st = str;
|
788 char *pt;
789 char *end;
790 int wcnt;
791 char *st = str; |
792
793 /*
794 * while we have data to process
795 */
796 while (cnt) {
797 if (!*rem) {
798 /*
799 * We are now at the start of file system block again
800 * (or what we think one is...). start looking for
801 * empty blocks again
802 */
803 *isempt = 1;
804 *rem = sz;
805 }
806
807 /*
808 * only examine up to the end of the current file block or
809 * remaining characters to write, whatever is smaller
810 */
811 wcnt = MIN(cnt, *rem);
812 cnt -= wcnt;
813 *rem -= wcnt;
814 if (*isempt) {
815 /*
816 * have not written to this block yet, so we keep
817 * looking for zero's
818 */
819 pt = st;
820 end = st + wcnt;
821
822 /*
823 * look for a zero filled buffer
824 */
825 while ((pt < end) && (*pt == '\0'))
826 ++pt;
827
828 if (pt == end) {
829 /*
830 * skip, buf is empty so far
831 */
832 if (lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
833 syswarn(1,errno,"File seek on %s",
834 name);
835 return(-1);
836 }
837 st = pt;
838 continue;
839 }
840 /*
841 * drat, the buf is not zero filled
842 */
843 *isempt = 0;
844 }
845
846 /*
847 * have non-zero data in this file system block, have to write
848 */
849 if (write(fd, st, wcnt) != wcnt) {
850 syswarn(1, errno, "Failed write to file %s", name);
851 return(-1);
852 }
853 st += wcnt;
854 }
855 return(st - str);
856}
857
858/*
859 * file_flush()
860 * when the last file block in a file is zero, many file systems will not
861 * let us create a hole at the end. To get the last block with zeros, we
862 * write the last BYTE with a zero (back up one byte and write a zero).
863 */
864
865void
866file_flush(int fd, char *fname, int isempt)
867{
868 static char blnk[] = "\0";
869
870 /*
871 * silly test, but make sure we are only called when the last block is
872 * filled with all zeros.
873 */
874 if (!isempt)
875 return;
876
877 /*
878 * move back one byte and write a zero
879 */
880 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
881 syswarn(1, errno, "Failed seek on file %s", fname);
882 return;
883 }
884
885 if (write(fd, blnk, 1) < 0)
886 syswarn(1, errno, "Failed write to file %s", fname);
887 return;
888}
889
890/*
891 * rdfile_close()
892 * close a file we have beed reading (to copy or archive). If we have to
893 * reset access time (tflag) do so (the times are stored in arcn).
894 */
895
896void
|
897rdfile_close(register ARCHD *arcn, register int *fd)
|
| 897rdfile_close(ARCHD *arcn, int *fd) |
898{
899 /*
900 * make sure the file is open
901 */
902 if (*fd < 0)
903 return;
904
905 (void)close(*fd);
906 *fd = -1;
907 if (!tflag)
908 return;
909
910 /*
911 * user wants last access time reset
912 */
913 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
914 return;
915}
916
917/*
918 * set_crc()
919 * read a file to calculate its crc. This is a real drag. Archive formats
920 * that have this, end up reading the file twice (we have to write the
921 * header WITH the crc before writing the file contents. Oh well...
922 * Return:
923 * 0 if was able to calculate the crc, -1 otherwise
924 */
925
926int
|
927set_crc(register ARCHD *arcn, register int fd)
|
| 927set_crc(ARCHD *arcn, int fd) |
928{
|
929 register int i;
930 register int res;
|
929 int i;
930 int res; |
931 off_t cpcnt = 0L;
932 u_long size;
933 unsigned long crc = 0L;
934 char tbuf[FILEBLK];
935 struct stat sb;
936
937 if (fd < 0) {
938 /*
939 * hmm, no fd, should never happen. well no crc then.
940 */
941 arcn->crc = 0L;
942 return(0);
943 }
944
945 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
946 size = (u_long)sizeof(tbuf);
947
948 /*
949 * read all the bytes we think that there are in the file. If the user
950 * is trying to archive an active file, forget this file.
951 */
952 for(;;) {
953 if ((res = read(fd, tbuf, size)) <= 0)
954 break;
955 cpcnt += res;
956 for (i = 0; i < res; ++i)
957 crc += (tbuf[i] & 0xff);
958 }
959
960 /*
961 * safety check. we want to avoid archiving files that are active as
962 * they can create inconsistant archive copies.
963 */
964 if (cpcnt != arcn->sb.st_size)
965 paxwarn(1, "File changed size %s", arcn->org_name);
966 else if (fstat(fd, &sb) < 0)
967 syswarn(1, errno, "Failed stat on %s", arcn->org_name);
968 else if (arcn->sb.st_mtime != sb.st_mtime)
969 paxwarn(1, "File %s was modified during read", arcn->org_name);
970 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
971 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
972 else {
973 arcn->crc = crc;
974 return(0);
975 }
976 return(-1);
977}
|