source: trunk/minix/commands/pax/file_subs.c@ 10

Last change on this file since 10 was 9, checked in by Mattia Monga, 13 years ago

Minix 3.1.2a

File size: 25.5 KB
Line 
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 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 */
33
34#ifndef lint
35#if 0
36static char sccsid[] = "@(#)file_subs.c 8.1 (Berkeley) 5/31/93";
37#endif
38#endif /* not lint */
39
40#include <sys/types.h>
41#include <sys/time.h>
42#include <sys/stat.h>
43#include <unistd.h>
44#include <fcntl.h>
45#include <string.h>
46#include <stdio.h>
47#include <errno.h>
48#include <utime.h>
49#include <sys/uio.h>
50#include <stdlib.h>
51#include "pax.h"
52#include "options.h"
53#include "extern.h"
54
55static int
56mk_link(char *,struct stat *,char *, int);
57
58/*
59 * routines that deal with file operations such as: creating, removing;
60 * and setting access modes, uid/gid and times of files
61 */
62
63#define FILEBITS (S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO)
64#define SETBITS (S_ISUID | S_ISGID)
65#define ABITS (FILEBITS | SETBITS)
66
67/*
68 * file_creat()
69 * Create and open a file.
70 * Return:
71 * file descriptor or -1 for failure
72 */
73
74int
75file_creat(ARCHD *arcn)
76{
77 int fd = -1;
78 mode_t file_mode;
79 int oerrno;
80
81 /*
82 * assume file doesn't exist, so just try to create it, most times this
83 * works. We have to take special handling when the file does exist. To
84 * detect this, we use O_EXCL. For example when trying to create a
85 * file and a character device or fifo exists with the same name, we
86 * can accidently open the device by mistake (or block waiting to open)
87 * If we find that the open has failed, then figure spend the effort to
88 * figure out why. This strategy was found to have better average
89 * performance in common use than checking the file (and the path)
90 * first with lstat.
91 */
92 file_mode = arcn->sb.st_mode & FILEBITS;
93 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL,
94 file_mode)) >= 0)
95 return(fd);
96
97 /*
98 * the file seems to exist. First we try to get rid of it (found to be
99 * the second most common failure when traced). If this fails, only
100 * then we go to the expense to check and create the path to the file
101 */
102 if (unlnk_exist(arcn->name, arcn->type) != 0)
103 return(-1);
104
105 for (;;) {
106 /*
107 * try to open it again, if this fails, check all the nodes in
108 * the path and give it a final try. if chk_path() finds that
109 * it cannot fix anything, we will skip the last attempt
110 */
111 if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC,
112 file_mode)) >= 0)
113 break;
114 oerrno = errno;
115 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
116 syswarn(1, oerrno, "Unable to create %s", arcn->name);
117 return(-1);
118 }
119 }
120 return(fd);
121}
122
123/*
124 * file_close()
125 * Close file descriptor to a file just created by pax. Sets modes,
126 * ownership and times as required.
127 * Return:
128 * 0 for success, -1 for failure
129 */
130
131void
132file_close(ARCHD *arcn, int fd)
133{
134 int res = 0;
135
136 if (fd < 0)
137 return;
138 if (close(fd) < 0)
139 syswarn(0, errno, "Unable to close file descriptor on %s",
140 arcn->name);
141
142 /*
143 * set owner/groups first as this may strip off mode bits we want
144 * then set file permission modes. Then set file access and
145 * modification times.
146 */
147 if (pids)
148 res = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid);
149
150 /*
151 * IMPORTANT SECURITY NOTE:
152 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
153 * set uid/gid bits
154 */
155 if (!pmode || res)
156 arcn->sb.st_mode &= ~(SETBITS);
157 if (pmode)
158 set_pmode(arcn->name, arcn->sb.st_mode);
159 if (patime || pmtime)
160 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
161}
162
163/*
164 * lnk_creat()
165 * Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
166 * must exist;
167 * Return:
168 * 0 if ok, -1 otherwise
169 */
170
171int
172lnk_creat(ARCHD *arcn)
173{
174 struct stat sb;
175
176 /*
177 * we may be running as root, so we have to be sure that link target
178 * is not a directory, so we lstat and check
179 */
180 if (lstat(arcn->ln_name, &sb) < 0) {
181 syswarn(1,errno,"Unable to link to %s from %s", arcn->ln_name,
182 arcn->name);
183 return(-1);
184 }
185
186 if (S_ISDIR(sb.st_mode)) {
187 paxwarn(1, "A hard link to the directory %s is not allowed",
188 arcn->ln_name);
189 return(-1);
190 }
191
192 return(mk_link(arcn->ln_name, &sb, arcn->name, 0));
193}
194
195/*
196 * cross_lnk()
197 * Create a hard link to arcn->org_name from arcn->name. Only used in copy
198 * with the -l flag. No warning or error if this does not succeed (we will
199 * then just create the file)
200 * Return:
201 * 1 if copy() should try to create this file node
202 * 0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
203 */
204
205int
206cross_lnk(ARCHD *arcn)
207{
208 /*
209 * try to make a link to original file (-l flag in copy mode). make sure
210 * we do not try to link to directories in case we are running as root
211 * (and it might succeed).
212 */
213 if (arcn->type == PAX_DIR)
214 return(1);
215 return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
216}
217
218/*
219 * chk_same()
220 * In copy mode if we are not trying to make hard links between the src
221 * and destinations, make sure we are not going to overwrite ourselves by
222 * accident. This slows things down a little, but we have to protect all
223 * those people who make typing errors.
224 * Return:
225 * 1 the target does not exist, go ahead and copy
226 * 0 skip it file exists (-k) or may be the same as source file
227 */
228
229int
230chk_same(ARCHD *arcn)
231{
232 struct stat sb;
233
234 /*
235 * if file does not exist, return. if file exists and -k, skip it
236 * quietly
237 */
238 if (lstat(arcn->name, &sb) < 0)
239 return(1);
240 if (kflag)
241 return(0);
242
243 /*
244 * better make sure the user does not have src == dest by mistake
245 */
246 if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
247 paxwarn(1, "Unable to copy %s, file would overwrite itself",
248 arcn->name);
249 return(0);
250 }
251 return(1);
252}
253
254/*
255 * mk_link()
256 * try to make a hard link between two files. if ign set, we do not
257 * complain.
258 * Return:
259 * 0 if successful (or we are done with this file but no error, such as
260 * finding the from file exists and the user has set -k).
261 * 1 when ign was set to indicates we could not make the link but we
262 * should try to copy/extract the file as that might work (and is an
263 * allowed option). -1 an error occurred.
264 */
265
266static int
267mk_link(char *to, struct stat *to_sb, char *from,
268 int ign)
269{
270 struct stat sb;
271 int oerrno;
272
273 /*
274 * if from file exists, it has to be unlinked to make the link. If the
275 * file exists and -k is set, skip it quietly
276 */
277 if (lstat(from, &sb) == 0) {
278 if (kflag)
279 return(0);
280
281 /*
282 * make sure it is not the same file, protect the user
283 */
284 if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
285 paxwarn(1, "Unable to link file %s to itself", to);
286 return(-1);;
287 }
288
289 /*
290 * try to get rid of the file, based on the type
291 */
292 if (S_ISDIR(sb.st_mode)) {
293 if (rmdir(from) < 0) {
294 syswarn(1, errno, "Unable to remove %s", from);
295 return(-1);
296 }
297 } else if (unlink(from) < 0) {
298 if (!ign) {
299 syswarn(1, errno, "Unable to remove %s", from);
300 return(-1);
301 }
302 return(1);
303 }
304 }
305
306 /*
307 * from file is gone (or did not exist), try to make the hard link.
308 * if it fails, check the path and try it again (if chk_path() says to
309 * try again)
310 */
311 for (;;) {
312 if (link(to, from) == 0)
313 break;
314 oerrno = errno;
315 if (!nodirs && chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
316 continue;
317 if (!ign) {
318 syswarn(1, oerrno, "Could not link to %s from %s", to,
319 from);
320 return(-1);
321 }
322 return(1);
323 }
324
325 /*
326 * all right the link was made
327 */
328 return(0);
329}
330
331/*
332 * node_creat()
333 * create an entry in the file system (other than a file or hard link).
334 * If successful, sets uid/gid modes and times as required.
335 * Return:
336 * 0 if ok, -1 otherwise
337 */
338
339int
340node_creat(ARCHD *arcn)
341{
342 int res;
343 int ign = 0;
344 int oerrno;
345 int pass = 0;
346 mode_t file_mode;
347 struct stat sb;
348
349 /*
350 * create node based on type, if that fails try to unlink the node and
351 * try again. finally check the path and try again. As noted in the
352 * file and link creation routines, this method seems to exhibit the
353 * best performance in general use workloads.
354 */
355 file_mode = arcn->sb.st_mode & FILEBITS;
356
357 for (;;) {
358 switch(arcn->type) {
359 case PAX_DIR:
360 res = mkdir(arcn->name, file_mode);
361 if (ign)
362 res = 0;
363 break;
364 case PAX_CHR:
365 file_mode |= S_IFCHR;
366 res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
367 break;
368 case PAX_BLK:
369 file_mode |= S_IFBLK;
370 res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
371 break;
372 case PAX_FIF:
373 res = mkfifo(arcn->name, file_mode);
374 break;
375 case PAX_SCK:
376 /*
377 * Skip sockets, operation has no meaning under BSD
378 */
379 paxwarn(0,
380 "%s skipped. Sockets cannot be copied or extracted",
381 arcn->name);
382 return(-1);
383 case PAX_SLK:
384 res = symlink(arcn->ln_name, arcn->name);
385 break;
386 case PAX_CTG:
387 case PAX_HLK:
388 case PAX_HRG:
389 case PAX_REG:
390 default:
391 /*
392 * we should never get here
393 */
394 paxwarn(0, "%s has an unknown file type, skipping",
395 arcn->name);
396 return(-1);
397 }
398
399 /*
400 * if we were able to create the node break out of the loop,
401 * otherwise try to unlink the node and try again. if that
402 * fails check the full path and try a final time.
403 */
404 if (res == 0)
405 break;
406
407 /*
408 * we failed to make the node
409 */
410 oerrno = errno;
411 if ((ign = unlnk_exist(arcn->name, arcn->type)) < 0)
412 return(-1);
413
414 if (++pass <= 1)
415 continue;
416
417 if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
418 syswarn(1, oerrno, "Could not create: %s", arcn->name);
419 return(-1);
420 }
421 }
422
423 /*
424 * we were able to create the node. set uid/gid, modes and times
425 */
426 if (pids)
427 res = ((arcn->type == PAX_SLK) ?
428 set_lids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid) :
429 set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid));
430 else
431 res = 0;
432
433 /*
434 * symlinks are done now.
435 */
436 if (arcn->type == PAX_SLK)
437 return(0);
438
439 /*
440 * IMPORTANT SECURITY NOTE:
441 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
442 * set uid/gid bits
443 */
444 if (!pmode || res)
445 arcn->sb.st_mode &= ~(SETBITS);
446 if (pmode)
447 set_pmode(arcn->name, arcn->sb.st_mode);
448
449 if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
450 /*
451 * Dirs must be processed again at end of extract to set times
452 * and modes to agree with those stored in the archive. However
453 * to allow extract to continue, we may have to also set owner
454 * rights. This allows nodes in the archive that are children
455 * of this directory to be extracted without failure. Both time
456 * and modes will be fixed after the entire archive is read and
457 * before pax exits.
458 */
459 if (access(arcn->name, R_OK | W_OK | X_OK) < 0) {
460 if (lstat(arcn->name, &sb) < 0) {
461 syswarn(0, errno,"Could not access %s (stat)",
462 arcn->name);
463 set_pmode(arcn->name,file_mode | S_IRWXU);
464 } else {
465 /*
466 * We have to add rights to the dir, so we make
467 * sure to restore the mode. The mode must be
468 * restored AS CREATED and not as stored if
469 * pmode is not set.
470 */
471 set_pmode(arcn->name,
472 ((sb.st_mode & FILEBITS) | S_IRWXU));
473 if (!pmode)
474 arcn->sb.st_mode = sb.st_mode;
475 }
476
477 /*
478 * we have to force the mode to what was set here,
479 * since we changed it from the default as created.
480 */
481 add_dir(arcn->name, arcn->nlen, &(arcn->sb), 1);
482 } else if (pmode || patime || pmtime)
483 add_dir(arcn->name, arcn->nlen, &(arcn->sb), 0);
484 }
485
486 if (patime || pmtime)
487 set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
488 return(0);
489}
490
491/*
492 * unlnk_exist()
493 * Remove node from file system with the specified name. We pass the type
494 * of the node that is going to replace it. When we try to create a
495 * directory and find that it already exists, we allow processing to
496 * continue as proper modes etc will always be set for it later on.
497 * Return:
498 * 0 is ok to proceed, no file with the specified name exists
499 * -1 we were unable to remove the node, or we should not remove it (-k)
500 * 1 we found a directory and we were going to create a directory.
501 */
502
503int
504unlnk_exist(char *name, int type)
505{
506 struct stat sb;
507
508 /*
509 * the file does not exist, or -k we are done
510 */
511 if (lstat(name, &sb) < 0)
512 return(0);
513 if (kflag)
514 return(-1);
515
516 if (S_ISDIR(sb.st_mode)) {
517 /*
518 * try to remove a directory, if it fails and we were going to
519 * create a directory anyway, tell the caller (return a 1)
520 */
521 if (rmdir(name) < 0) {
522 if (type == PAX_DIR)
523 return(1);
524 syswarn(1,errno,"Unable to remove directory %s", name);
525 return(-1);
526 }
527 return(0);
528 }
529
530 /*
531 * try to get rid of all non-directory type nodes
532 */
533 if (unlink(name) < 0) {
534 syswarn(1, errno, "Could not unlink %s", name);
535 return(-1);
536 }
537 return(0);
538}
539
540/*
541 * chk_path()
542 * We were trying to create some kind of node in the file system and it
543 * failed. chk_path() makes sure the path up to the node exists and is
544 * writeable. When we have to create a directory that is missing along the
545 * path somewhere, the directory we create will be set to the same
546 * uid/gid as the file has (when uid and gid are being preserved).
547 * NOTE: this routine is a real performance loss. It is only used as a
548 * last resort when trying to create entries in the file system.
549 * Return:
550 * -1 when it could find nothing it is allowed to fix.
551 * 0 otherwise
552 */
553
554int
555chk_path( char *name, uid_t st_uid, gid_t st_gid)
556{
557 char *spt = name;
558 struct stat sb;
559 int retval = -1;
560
561 /*
562 * watch out for paths with nodes stored directly in / (e.g. /bozo)
563 */
564 if (*spt == '/')
565 ++spt;
566
567 for(;;) {
568 /*
569 * work foward from the first / and check each part of the path
570 */
571 spt = strchr(spt, '/');
572 if (spt == NULL)
573 break;
574 *spt = '\0';
575
576 /*
577 * if it exists we assume it is a directory, it is not within
578 * the spec (at least it seems to read that way) to alter the
579 * file system for nodes NOT EXPLICITLY stored on the archive.
580 * If that assumption is changed, you would test the node here
581 * and figure out how to get rid of it (probably like some
582 * recursive unlink()) or fix up the directory permissions if
583 * required (do an access()).
584 */
585 if (lstat(name, &sb) == 0) {
586 *(spt++) = '/';
587 continue;
588 }
589
590 /*
591 * the path fails at this point, see if we can create the
592 * needed directory and continue on
593 */
594 if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
595 *spt = '/';
596 retval = -1;
597 break;
598 }
599
600 /*
601 * we were able to create the directory. We will tell the
602 * caller that we found something to fix, and it is ok to try
603 * and create the node again.
604 */
605 retval = 0;
606 if (pids)
607 (void)set_ids(name, st_uid, st_gid);
608
609 /*
610 * make sure the user doen't have some strange umask that
611 * causes this newly created directory to be unusable. We fix
612 * the modes and restore them back to the creation default at
613 * the end of pax
614 */
615 if ((access(name, R_OK | W_OK | X_OK) < 0) &&
616 (lstat(name, &sb) == 0)) {
617 set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
618 add_dir(name, spt - name, &sb, 1);
619 }
620 *(spt++) = '/';
621 continue;
622 }
623 return(retval);
624}
625
626/*
627 * set_ftime()
628 * Set the access time and modification time for a named file. If frc is
629 * non-zero we force these times to be set even if the user did not
630 * request access and/or modification time preservation (this is also
631 * used by -t to reset access times).
632 * When ign is zero, only those times the user has asked for are set, the
633 * other ones are left alone. We do not assume the un-documented feature
634 * of many utimes() implementations that consider a 0 time value as a do
635 * not set request.
636 */
637
638void
639set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
640{
641#if 0
642 static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}};
643#endif
644 struct stat sb;
645 struct utimbuf ut;
646
647#if 0
648 tv[0].tv_sec = atime;
649 tv[1].tv_sec = mtime;
650#endif
651
652 ut.actime = atime;
653 ut.modtime = mtime;
654
655 if (!frc && (!patime || !pmtime)) {
656 /*
657 * if we are not forcing, only set those times the user wants
658 * set. We get the current values of the times if we need them.
659 */
660 if (lstat(fnm, &sb) == 0) {
661 if (!patime)
662 ut.actime = sb.st_atime;
663 if (!pmtime)
664 ut.modtime = sb.st_mtime;
665 } else
666 syswarn(0,errno,"Unable to obtain file stats %s", fnm);
667 }
668
669 /*
670 * set the times
671 */
672 if (utime(fnm, &ut) < 0)
673 syswarn(1, errno, "Access/modification time set failed on: %s",
674 fnm);
675 return;
676}
677
678/*
679 * set_ids()
680 * set the uid and gid of a file system node
681 * Return:
682 * 0 when set, -1 on failure
683 */
684
685int
686set_ids(char *fnm, uid_t uid, gid_t gid)
687{
688 if (chown(fnm, uid, gid) < 0) {
689 /*
690 * ignore EPERM unless in verbose mode or being run by root.
691 * if running as pax, POSIX requires a warning.
692 */
693 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
694 geteuid() == 0)
695 syswarn(1, errno, "Unable to set file uid/gid of %s",
696 fnm);
697 return(-1);
698 }
699 return(0);
700}
701
702/*
703 * set_lids()
704 * set the uid and gid of a file system node
705 * Return:
706 * 0 when set, -1 on failure
707 */
708
709int
710set_lids(char *fnm, uid_t uid, gid_t gid)
711{
712
713#if 0
714 if (lchown(fnm, uid, gid) < 0) {
715 /*
716 * ignore EPERM unless in verbose mode or being run by root.
717 * if running as pax, POSIX requires a warning.
718 */
719 if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
720 geteuid() == 0)
721 syswarn(1, errno, "Unable to set file uid/gid of %s",
722 fnm);
723 return(-1);
724 }
725#else
726 return(-1); /* No lchown() in minix. */
727#endif
728 return(0);
729}
730
731/*
732 * set_pmode()
733 * Set file access mode
734 */
735
736void
737set_pmode(char *fnm, mode_t mode)
738{
739 mode &= ABITS;
740 if (chmod(fnm, mode) < 0)
741 syswarn(1, errno, "Could not set permissions on %s", fnm);
742 return;
743}
744
745/*
746 * file_write()
747 * Write/copy a file (during copy or archive extract). This routine knows
748 * how to copy files with lseek holes in it. (Which are read as file
749 * blocks containing all 0's but do not have any file blocks associated
750 * with the data). Typical examples of these are files created by dbm
751 * variants (.pag files). While the file size of these files are huge, the
752 * actual storage is quite small (the files are sparse). The problem is
753 * the holes read as all zeros so are probably stored on the archive that
754 * way (there is no way to determine if the file block is really a hole,
755 * we only know that a file block of all zero's can be a hole).
756 * At this writing, no major archive format knows how to archive files
757 * with holes. However, on extraction (or during copy, -rw) we have to
758 * deal with these files. Without detecting the holes, the files can
759 * consume a lot of file space if just written to disk. This replacement
760 * for write when passed the basic allocation size of a file system block,
761 * uses lseek whenever it detects the input data is all 0 within that
762 * file block. In more detail, the strategy is as follows:
763 * While the input is all zero keep doing an lseek. Keep track of when we
764 * pass over file block boundries. Only write when we hit a non zero
765 * input. once we have written a file block, we continue to write it to
766 * the end (we stop looking at the input). When we reach the start of the
767 * next file block, start checking for zero blocks again. Working on file
768 * block boundries significantly reduces the overhead when copying files
769 * that are NOT very sparse. This overhead (when compared to a write) is
770 * almost below the measurement resolution on many systems. Without it,
771 * files with holes cannot be safely copied. It does has a side effect as
772 * it can put holes into files that did not have them before, but that is
773 * not a problem since the file contents are unchanged (in fact it saves
774 * file space). (Except on paging files for diskless clients. But since we
775 * cannot determine one of those file from here, we ignore them). If this
776 * ever ends up on a system where CTG files are supported and the holes
777 * are not desired, just do a conditional test in those routines that
778 * call file_write() and have it call write() instead. BEFORE CLOSING THE
779 * FILE, make sure to call file_flush() when the last write finishes with
780 * an empty block. A lot of file systems will not create an lseek hole at
781 * the end. In this case we drop a single 0 at the end to force the
782 * trailing 0's in the file.
783 * ---Parameters---
784 * rem: how many bytes left in this file system block
785 * isempt: have we written to the file block yet (is it empty)
786 * sz: basic file block allocation size
787 * cnt: number of bytes on this write
788 * str: buffer to write
789 * Return:
790 * number of bytes written, -1 on write (or lseek) error.
791 */
792
793int
794file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
795 char *name)
796{
797 char *pt;
798 char *end;
799 int wcnt;
800 char *st = str;
801
802 /*
803 * while we have data to process
804 */
805 while (cnt) {
806 if (!*rem) {
807 /*
808 * We are now at the start of file system block again
809 * (or what we think one is...). start looking for
810 * empty blocks again
811 */
812 *isempt = 1;
813 *rem = sz;
814 }
815
816 /*
817 * only examine up to the end of the current file block or
818 * remaining characters to write, whatever is smaller
819 */
820 wcnt = MIN(cnt, *rem);
821 cnt -= wcnt;
822 *rem -= wcnt;
823 if (*isempt) {
824 /*
825 * have not written to this block yet, so we keep
826 * looking for zero's
827 */
828 pt = st;
829 end = st + wcnt;
830
831 /*
832 * look for a zero filled buffer
833 */
834 while ((pt < end) && (*pt == '\0'))
835 ++pt;
836
837 if (pt == end) {
838 /*
839 * skip, buf is empty so far
840 */
841 if (lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
842 syswarn(1,errno,"File seek on %s",
843 name);
844 return(-1);
845 }
846 st = pt;
847 continue;
848 }
849 /*
850 * drat, the buf is not zero filled
851 */
852 *isempt = 0;
853 }
854
855 /*
856 * have non-zero data in this file system block, have to write
857 */
858 if (write(fd, st, wcnt) != wcnt) {
859 syswarn(1, errno, "Failed write to file %s", name);
860 return(-1);
861 }
862 st += wcnt;
863 }
864 return(st - str);
865}
866
867/*
868 * file_flush()
869 * when the last file block in a file is zero, many file systems will not
870 * let us create a hole at the end. To get the last block with zeros, we
871 * write the last BYTE with a zero (back up one byte and write a zero).
872 */
873
874void
875file_flush(int fd, char *fname, int isempt)
876{
877 static char blnk[] = "\0";
878
879 /*
880 * silly test, but make sure we are only called when the last block is
881 * filled with all zeros.
882 */
883 if (!isempt)
884 return;
885
886 /*
887 * move back one byte and write a zero
888 */
889 if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
890 syswarn(1, errno, "Failed seek on file %s", fname);
891 return;
892 }
893
894 if (write(fd, blnk, 1) < 0)
895 syswarn(1, errno, "Failed write to file %s", fname);
896 return;
897}
898
899/*
900 * rdfile_close()
901 * close a file we have beed reading (to copy or archive). If we have to
902 * reset access time (tflag) do so (the times are stored in arcn).
903 */
904
905void
906rdfile_close(ARCHD *arcn, int *fd)
907{
908 /*
909 * make sure the file is open
910 */
911 if (*fd < 0)
912 return;
913
914 (void)close(*fd);
915 *fd = -1;
916 if (!tflag)
917 return;
918
919 /*
920 * user wants last access time reset
921 */
922 set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
923 return;
924}
925
926/*
927 * set_crc()
928 * read a file to calculate its crc. This is a real drag. Archive formats
929 * that have this, end up reading the file twice (we have to write the
930 * header WITH the crc before writing the file contents. Oh well...
931 * Return:
932 * 0 if was able to calculate the crc, -1 otherwise
933 */
934
935int
936set_crc(ARCHD *arcn, int fd)
937{
938 int i;
939 int res;
940 off_t cpcnt = 0L;
941 u_long size;
942 unsigned long crc = 0L;
943 char tbuf[FILEBLK];
944 struct stat sb;
945
946 if (fd < 0) {
947 /*
948 * hmm, no fd, should never happen. well no crc then.
949 */
950 arcn->crc = 0L;
951 return(0);
952 }
953
954#if 0
955 /* not in minix */
956 if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
957 size = (u_long)sizeof(tbuf);
958#endif
959
960 /*
961 * read all the bytes we think that there are in the file. If the user
962 * is trying to archive an active file, forget this file.
963 */
964 for(;;) {
965 if ((res = read(fd, tbuf, size)) <= 0)
966 break;
967 cpcnt += res;
968 for (i = 0; i < res; ++i)
969 crc += (tbuf[i] & 0xff);
970 }
971
972 /*
973 * safety check. we want to avoid archiving files that are active as
974 * they can create inconsistant archive copies.
975 */
976 if (cpcnt != arcn->sb.st_size)
977 paxwarn(1, "File changed size %s", arcn->org_name);
978 else if (fstat(fd, &sb) < 0)
979 syswarn(1, errno, "Failed stat on %s", arcn->org_name);
980 else if (arcn->sb.st_mtime != sb.st_mtime)
981 paxwarn(1, "File %s was modified during read", arcn->org_name);
982 else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
983 syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
984 else {
985 arcn->crc = crc;
986 return(0);
987 }
988 return(-1);
989}
Note: See TracBrowser for help on using the repository browser.