xref: /freebsd/bin/pax/file_subs.c (revision fed1ca4b719c56c930f2259d80663cd34be812bb)
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
36 static char sccsid[] = "@(#)file_subs.c	8.1 (Berkeley) 5/31/93";
37 #endif
38 #endif /* not lint */
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
41 
42 #include <sys/types.h>
43 #include <sys/time.h>
44 #include <sys/stat.h>
45 #include <unistd.h>
46 #include <fcntl.h>
47 #include <string.h>
48 #include <stdio.h>
49 #include <errno.h>
50 #include <sys/uio.h>
51 #include "pax.h"
52 #include "options.h"
53 #include "extern.h"
54 
55 static int
56 mk_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 
74 int
75 file_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 accidentally open the device by mistake (or block waiting to
87 	 * open). If we find that the open has failed, then spend the effort
88 	 * to 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 
131 void
132 file_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 
171 int
172 lnk_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 
205 int
206 cross_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 
229 int
230 chk_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 
266 static int
267 mk_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 
339 int
340 node_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 = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid);
428 	else
429 		res = 0;
430 
431 	/*
432 	 * IMPORTANT SECURITY NOTE:
433 	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
434 	 * set uid/gid bits
435 	 */
436 	if (!pmode || res)
437 		arcn->sb.st_mode &= ~(SETBITS);
438 	if (pmode)
439 		set_pmode(arcn->name, arcn->sb.st_mode);
440 
441 	if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
442 		/*
443 		 * Dirs must be processed again at end of extract to set times
444 		 * and modes to agree with those stored in the archive. However
445 		 * to allow extract to continue, we may have to also set owner
446 		 * rights. This allows nodes in the archive that are children
447 		 * of this directory to be extracted without failure. Both time
448 		 * and modes will be fixed after the entire archive is read and
449 		 * before pax exits.
450 		 */
451 		if (access(arcn->name, R_OK | W_OK | X_OK) < 0) {
452 			if (lstat(arcn->name, &sb) < 0) {
453 				syswarn(0, errno,"Could not access %s (stat)",
454 				    arcn->name);
455 				set_pmode(arcn->name,file_mode | S_IRWXU);
456 			} else {
457 				/*
458 				 * We have to add rights to the dir, so we make
459 				 * sure to restore the mode. The mode must be
460 				 * restored AS CREATED and not as stored if
461 				 * pmode is not set.
462 				 */
463 				set_pmode(arcn->name,
464 				    ((sb.st_mode & FILEBITS) | S_IRWXU));
465 				if (!pmode)
466 					arcn->sb.st_mode = sb.st_mode;
467 			}
468 
469 			/*
470 			 * we have to force the mode to what was set here,
471 			 * since we changed it from the default as created.
472 			 */
473 			add_dir(arcn->name, arcn->nlen, &(arcn->sb), 1);
474 		} else if (pmode || patime || pmtime)
475 			add_dir(arcn->name, arcn->nlen, &(arcn->sb), 0);
476 	}
477 
478 	if (patime || pmtime)
479 		set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
480 	return(0);
481 }
482 
483 /*
484  * unlnk_exist()
485  *	Remove node from file system with the specified name. We pass the type
486  *	of the node that is going to replace it. When we try to create a
487  *	directory and find that it already exists, we allow processing to
488  *	continue as proper modes etc will always be set for it later on.
489  * Return:
490  *	0 is ok to proceed, no file with the specified name exists
491  *	-1 we were unable to remove the node, or we should not remove it (-k)
492  *	1 we found a directory and we were going to create a directory.
493  */
494 
495 int
496 unlnk_exist(char *name, int type)
497 {
498 	struct stat sb;
499 
500 	/*
501 	 * the file does not exist, or -k we are done
502 	 */
503 	if (lstat(name, &sb) < 0)
504 		return(0);
505 	if (kflag)
506 		return(-1);
507 
508 	if (S_ISDIR(sb.st_mode)) {
509 		/*
510 		 * try to remove a directory, if it fails and we were going to
511 		 * create a directory anyway, tell the caller (return a 1)
512 		 */
513 		if (rmdir(name) < 0) {
514 			if (type == PAX_DIR)
515 				return(1);
516 			syswarn(1,errno,"Unable to remove directory %s", name);
517 			return(-1);
518 		}
519 		return(0);
520 	}
521 
522 	/*
523 	 * try to get rid of all non-directory type nodes
524 	 */
525 	if (unlink(name) < 0) {
526 		syswarn(1, errno, "Could not unlink %s", name);
527 		return(-1);
528 	}
529 	return(0);
530 }
531 
532 /*
533  * chk_path()
534  *	We were trying to create some kind of node in the file system and it
535  *	failed. chk_path() makes sure the path up to the node exists and is
536  *	writeable. When we have to create a directory that is missing along the
537  *	path somewhere, the directory we create will be set to the same
538  *	uid/gid as the file has (when uid and gid are being preserved).
539  *	NOTE: this routine is a real performance loss. It is only used as a
540  *	last resort when trying to create entries in the file system.
541  * Return:
542  *	-1 when it could find nothing it is allowed to fix.
543  *	0 otherwise
544  */
545 
546 int
547 chk_path( char *name, uid_t st_uid, gid_t st_gid)
548 {
549 	char *spt = name;
550 	struct stat sb;
551 	int retval = -1;
552 
553 	/*
554 	 * watch out for paths with nodes stored directly in / (e.g. /bozo)
555 	 */
556 	if (*spt == '/')
557 		++spt;
558 
559 	for(;;) {
560 		/*
561 		 * work forward from the first / and check each part of the path
562 		 */
563 		spt = strchr(spt, '/');
564 		if (spt == NULL)
565 			break;
566 		*spt = '\0';
567 
568 		/*
569 		 * if it exists we assume it is a directory, it is not within
570 		 * the spec (at least it seems to read that way) to alter the
571 		 * file system for nodes NOT EXPLICITLY stored on the archive.
572 		 * If that assumption is changed, you would test the node here
573 		 * and figure out how to get rid of it (probably like some
574 		 * recursive unlink()) or fix up the directory permissions if
575 		 * required (do an access()).
576 		 */
577 		if (lstat(name, &sb) == 0) {
578 			*(spt++) = '/';
579 			continue;
580 		}
581 
582 		/*
583 		 * the path fails at this point, see if we can create the
584 		 * needed directory and continue on
585 		 */
586 		if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
587 			*spt = '/';
588 			retval = -1;
589 			break;
590 		}
591 
592 		/*
593 		 * we were able to create the directory. We will tell the
594 		 * caller that we found something to fix, and it is ok to try
595 		 * and create the node again.
596 		 */
597 		retval = 0;
598 		if (pids)
599 			(void)set_ids(name, st_uid, st_gid);
600 
601 		/*
602 		 * make sure the user doesn't have some strange umask that
603 		 * causes this newly created directory to be unusable. We fix
604 		 * the modes and restore them back to the creation default at
605 		 * the end of pax
606 		 */
607 		if ((access(name, R_OK | W_OK | X_OK) < 0) &&
608 		    (lstat(name, &sb) == 0)) {
609 			set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
610 			add_dir(name, spt - name, &sb, 1);
611 		}
612 		*(spt++) = '/';
613 		continue;
614 	}
615 	return(retval);
616 }
617 
618 /*
619  * set_ftime()
620  *	Set the access time and modification time for a named file. If frc is
621  *	non-zero we force these times to be set even if the user did not
622  *	request access and/or modification time preservation (this is also
623  *	used by -t to reset access times).
624  *	When ign is zero, only those times the user has asked for are set, the
625  *	other ones are left alone. We do not assume the un-documented feature
626  *	of many lutimes() implementations that consider a 0 time value as a do
627  *	not set request.
628  */
629 
630 void
631 set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
632 {
633 	static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}};
634 	struct stat sb;
635 
636 	tv[0].tv_sec = atime;
637 	tv[1].tv_sec = mtime;
638 	if (!frc && (!patime || !pmtime)) {
639 		/*
640 		 * if we are not forcing, only set those times the user wants
641 		 * set. We get the current values of the times if we need them.
642 		 */
643 		if (lstat(fnm, &sb) == 0) {
644 			if (!patime)
645 				tv[0].tv_sec = sb.st_atime;
646 			if (!pmtime)
647 				tv[1].tv_sec = sb.st_mtime;
648 		} else
649 			syswarn(0,errno,"Unable to obtain file stats %s", fnm);
650 	}
651 
652 	/*
653 	 * set the times
654 	 */
655 	if (lutimes(fnm, tv) < 0)
656 		syswarn(1, errno, "Access/modification time set failed on: %s",
657 		    fnm);
658 	return;
659 }
660 
661 /*
662  * set_ids()
663  *	set the uid and gid of a file system node
664  * Return:
665  *	0 when set, -1 on failure
666  */
667 
668 int
669 set_ids(char *fnm, uid_t uid, gid_t gid)
670 {
671 	if (lchown(fnm, uid, gid) < 0) {
672 		/*
673 		 * ignore EPERM unless in verbose mode or being run by root.
674 		 * if running as pax, POSIX requires a warning.
675 		 */
676 		if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
677 		    geteuid() == 0)
678 			syswarn(1, errno, "Unable to set file uid/gid of %s",
679 			    fnm);
680 		return(-1);
681 	}
682 	return(0);
683 }
684 
685 /*
686  * set_pmode()
687  *	Set file access mode
688  */
689 
690 void
691 set_pmode(char *fnm, mode_t mode)
692 {
693 	mode &= ABITS;
694 	if (lchmod(fnm, mode) < 0)
695 		syswarn(1, errno, "Could not set permissions on %s", fnm);
696 	return;
697 }
698 
699 /*
700  * file_write()
701  *	Write/copy a file (during copy or archive extract). This routine knows
702  *	how to copy files with lseek holes in it. (Which are read as file
703  *	blocks containing all 0's but do not have any file blocks associated
704  *	with the data). Typical examples of these are files created by dbm
705  *	variants (.pag files). While the file size of these files are huge, the
706  *	actual storage is quite small (the files are sparse). The problem is
707  *	the holes read as all zeros so are probably stored on the archive that
708  *	way (there is no way to determine if the file block is really a hole,
709  *	we only know that a file block of all zero's can be a hole).
710  *	At this writing, no major archive format knows how to archive files
711  *	with holes. However, on extraction (or during copy, -rw) we have to
712  *	deal with these files. Without detecting the holes, the files can
713  *	consume a lot of file space if just written to disk. This replacement
714  *	for write when passed the basic allocation size of a file system block,
715  *	uses lseek whenever it detects the input data is all 0 within that
716  *	file block. In more detail, the strategy is as follows:
717  *	While the input is all zero keep doing an lseek. Keep track of when we
718  *	pass over file block boundaries. Only write when we hit a non zero
719  *	input. once we have written a file block, we continue to write it to
720  *	the end (we stop looking at the input). When we reach the start of the
721  *	next file block, start checking for zero blocks again. Working on file
722  *	block boundaries significantly reduces the overhead when copying files
723  *	that are NOT very sparse. This overhead (when compared to a write) is
724  *	almost below the measurement resolution on many systems. Without it,
725  *	files with holes cannot be safely copied. It does has a side effect as
726  *	it can put holes into files that did not have them before, but that is
727  *	not a problem since the file contents are unchanged (in fact it saves
728  *	file space). (Except on paging files for diskless clients. But since we
729  *	cannot determine one of those file from here, we ignore them). If this
730  *	ever ends up on a system where CTG files are supported and the holes
731  *	are not desired, just do a conditional test in those routines that
732  *	call file_write() and have it call write() instead. BEFORE CLOSING THE
733  *	FILE, make sure to call file_flush() when the last write finishes with
734  *	an empty block. A lot of file systems will not create an lseek hole at
735  *	the end. In this case we drop a single 0 at the end to force the
736  *	trailing 0's in the file.
737  *	---Parameters---
738  *	rem: how many bytes left in this file system block
739  *	isempt: have we written to the file block yet (is it empty)
740  *	sz: basic file block allocation size
741  *	cnt: number of bytes on this write
742  *	str: buffer to write
743  * Return:
744  *	number of bytes written, -1 on write (or lseek) error.
745  */
746 
747 int
748 file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
749 	char *name)
750 {
751 	char *pt;
752 	char *end;
753 	int wcnt;
754 	char *st = str;
755 
756 	/*
757 	 * while we have data to process
758 	 */
759 	while (cnt) {
760 		if (!*rem) {
761 			/*
762 			 * We are now at the start of file system block again
763 			 * (or what we think one is...). start looking for
764 			 * empty blocks again
765 			 */
766 			*isempt = 1;
767 			*rem = sz;
768 		}
769 
770 		/*
771 		 * only examine up to the end of the current file block or
772 		 * remaining characters to write, whatever is smaller
773 		 */
774 		wcnt = MIN(cnt, *rem);
775 		cnt -= wcnt;
776 		*rem -= wcnt;
777 		if (*isempt) {
778 			/*
779 			 * have not written to this block yet, so we keep
780 			 * looking for zero's
781 			 */
782 			pt = st;
783 			end = st + wcnt;
784 
785 			/*
786 			 * look for a zero filled buffer
787 			 */
788 			while ((pt < end) && (*pt == '\0'))
789 				++pt;
790 
791 			if (pt == end) {
792 				/*
793 				 * skip, buf is empty so far
794 				 */
795 				if (lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
796 					syswarn(1,errno,"File seek on %s",
797 					    name);
798 					return(-1);
799 				}
800 				st = pt;
801 				continue;
802 			}
803 			/*
804 			 * drat, the buf is not zero filled
805 			 */
806 			*isempt = 0;
807 		}
808 
809 		/*
810 		 * have non-zero data in this file system block, have to write
811 		 */
812 		if (write(fd, st, wcnt) != wcnt) {
813 			syswarn(1, errno, "Failed write to file %s", name);
814 			return(-1);
815 		}
816 		st += wcnt;
817 	}
818 	return(st - str);
819 }
820 
821 /*
822  * file_flush()
823  *	when the last file block in a file is zero, many file systems will not
824  *	let us create a hole at the end. To get the last block with zeros, we
825  *	write the last BYTE with a zero (back up one byte and write a zero).
826  */
827 
828 void
829 file_flush(int fd, char *fname, int isempt)
830 {
831 	static char blnk[] = "\0";
832 
833 	/*
834 	 * silly test, but make sure we are only called when the last block is
835 	 * filled with all zeros.
836 	 */
837 	if (!isempt)
838 		return;
839 
840 	/*
841 	 * move back one byte and write a zero
842 	 */
843 	if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
844 		syswarn(1, errno, "Failed seek on file %s", fname);
845 		return;
846 	}
847 
848 	if (write(fd, blnk, 1) < 0)
849 		syswarn(1, errno, "Failed write to file %s", fname);
850 	return;
851 }
852 
853 /*
854  * rdfile_close()
855  *	close a file we have beed reading (to copy or archive). If we have to
856  *	reset access time (tflag) do so (the times are stored in arcn).
857  */
858 
859 void
860 rdfile_close(ARCHD *arcn, int *fd)
861 {
862 	/*
863 	 * make sure the file is open
864 	 */
865 	if (*fd < 0)
866 		return;
867 
868 	(void)close(*fd);
869 	*fd = -1;
870 	if (!tflag)
871 		return;
872 
873 	/*
874 	 * user wants last access time reset
875 	 */
876 	set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
877 	return;
878 }
879 
880 /*
881  * set_crc()
882  *	read a file to calculate its crc. This is a real drag. Archive formats
883  *	that have this, end up reading the file twice (we have to write the
884  *	header WITH the crc before writing the file contents. Oh well...
885  * Return:
886  *	0 if was able to calculate the crc, -1 otherwise
887  */
888 
889 int
890 set_crc(ARCHD *arcn, int fd)
891 {
892 	int i;
893 	int res;
894 	off_t cpcnt = 0L;
895 	u_long size;
896 	unsigned long crc = 0L;
897 	char tbuf[FILEBLK];
898 	struct stat sb;
899 
900 	if (fd < 0) {
901 		/*
902 		 * hmm, no fd, should never happen. well no crc then.
903 		 */
904 		arcn->crc = 0L;
905 		return(0);
906 	}
907 
908 	if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
909 		size = (u_long)sizeof(tbuf);
910 
911 	/*
912 	 * read all the bytes we think that there are in the file. If the user
913 	 * is trying to archive an active file, forget this file.
914 	 */
915 	for(;;) {
916 		if ((res = read(fd, tbuf, size)) <= 0)
917 			break;
918 		cpcnt += res;
919 		for (i = 0; i < res; ++i)
920 			crc += (tbuf[i] & 0xff);
921 	}
922 
923 	/*
924 	 * safety check. we want to avoid archiving files that are active as
925 	 * they can create inconsistent archive copies.
926 	 */
927 	if (cpcnt != arcn->sb.st_size)
928 		paxwarn(1, "File changed size %s", arcn->org_name);
929 	else if (fstat(fd, &sb) < 0)
930 		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
931 	else if (arcn->sb.st_mtime != sb.st_mtime)
932 		paxwarn(1, "File %s was modified during read", arcn->org_name);
933 	else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
934 		syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
935 	else {
936 		arcn->crc = crc;
937 		return(0);
938 	}
939 	return(-1);
940 }
941