xref: /freebsd/bin/pax/file_subs.c (revision 3494f7c0)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1992 Keith Muller.
5  * Copyright (c) 1992, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * This code is derived from software contributed to Berkeley by
9  * Keith Muller of the University of California, San Diego.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  * 3. Neither the name of the University nor the names of its contributors
20  *    may be used to endorse or promote products derived from this software
21  *    without specific prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  */
35 
36 #include <sys/types.h>
37 #include <sys/time.h>
38 #include <sys/stat.h>
39 #include <unistd.h>
40 #include <fcntl.h>
41 #include <string.h>
42 #include <stdio.h>
43 #include <errno.h>
44 #include <sys/uio.h>
45 #include "pax.h"
46 #include "options.h"
47 #include "extern.h"
48 
49 static int
50 mk_link(char *,struct stat *,char *, int);
51 
52 /*
53  * routines that deal with file operations such as: creating, removing;
54  * and setting access modes, uid/gid and times of files
55  */
56 
57 #define FILEBITS		(S_ISVTX | S_IRWXU | S_IRWXG | S_IRWXO)
58 #define SETBITS			(S_ISUID | S_ISGID)
59 #define ABITS			(FILEBITS | SETBITS)
60 
61 /*
62  * file_creat()
63  *	Create and open a file.
64  * Return:
65  *	file descriptor or -1 for failure
66  */
67 
68 int
69 file_creat(ARCHD *arcn)
70 {
71 	int fd = -1;
72 	mode_t file_mode;
73 	int oerrno;
74 
75 	/*
76 	 * assume file doesn't exist, so just try to create it, most times this
77 	 * works. We have to take special handling when the file does exist. To
78 	 * detect this, we use O_EXCL. For example when trying to create a
79 	 * file and a character device or fifo exists with the same name, we
80 	 * can accidentally open the device by mistake (or block waiting to
81 	 * open). If we find that the open has failed, then spend the effort
82 	 * to figure out why. This strategy was found to have better average
83 	 * performance in common use than checking the file (and the path)
84 	 * first with lstat.
85 	 */
86 	file_mode = arcn->sb.st_mode & FILEBITS;
87 	if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC | O_EXCL,
88 	    file_mode)) >= 0)
89 		return(fd);
90 
91 	/*
92 	 * the file seems to exist. First we try to get rid of it (found to be
93 	 * the second most common failure when traced). If this fails, only
94 	 * then we go to the expense to check and create the path to the file
95 	 */
96 	if (unlnk_exist(arcn->name, arcn->type) != 0)
97 		return(-1);
98 
99 	for (;;) {
100 		/*
101 		 * try to open it again, if this fails, check all the nodes in
102 		 * the path and give it a final try. if chk_path() finds that
103 		 * it cannot fix anything, we will skip the last attempt
104 		 */
105 		if ((fd = open(arcn->name, O_WRONLY | O_CREAT | O_TRUNC,
106 		    file_mode)) >= 0)
107 			break;
108 		oerrno = errno;
109 		if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
110 			syswarn(1, oerrno, "Unable to create %s", arcn->name);
111 			return(-1);
112 		}
113 	}
114 	return(fd);
115 }
116 
117 /*
118  * file_close()
119  *	Close file descriptor to a file just created by pax. Sets modes,
120  *	ownership and times as required.
121  * Return:
122  *	0 for success, -1 for failure
123  */
124 
125 void
126 file_close(ARCHD *arcn, int fd)
127 {
128 	int res = 0;
129 
130 	if (fd < 0)
131 		return;
132 	if (close(fd) < 0)
133 		syswarn(0, errno, "Unable to close file descriptor on %s",
134 		    arcn->name);
135 
136 	/*
137 	 * set owner/groups first as this may strip off mode bits we want
138 	 * then set file permission modes. Then set file access and
139 	 * modification times.
140 	 */
141 	if (pids)
142 		res = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid);
143 
144 	/*
145 	 * IMPORTANT SECURITY NOTE:
146 	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT
147 	 * set uid/gid bits
148 	 */
149 	if (!pmode || res)
150 		arcn->sb.st_mode &= ~(SETBITS);
151 	if (pmode)
152 		set_pmode(arcn->name, arcn->sb.st_mode);
153 	if (patime || pmtime)
154 		set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
155 }
156 
157 /*
158  * lnk_creat()
159  *	Create a hard link to arcn->ln_name from arcn->name. arcn->ln_name
160  *	must exist;
161  * Return:
162  *	0 if ok, -1 otherwise
163  */
164 
165 int
166 lnk_creat(ARCHD *arcn)
167 {
168 	struct stat sb;
169 
170 	/*
171 	 * we may be running as root, so we have to be sure that link target
172 	 * is not a directory, so we lstat and check
173 	 */
174 	if (lstat(arcn->ln_name, &sb) < 0) {
175 		syswarn(1,errno,"Unable to link to %s from %s", arcn->ln_name,
176 		    arcn->name);
177 		return(-1);
178 	}
179 
180 	if (S_ISDIR(sb.st_mode)) {
181 		paxwarn(1, "A hard link to the directory %s is not allowed",
182 		    arcn->ln_name);
183 		return(-1);
184 	}
185 
186 	return(mk_link(arcn->ln_name, &sb, arcn->name, 0));
187 }
188 
189 /*
190  * cross_lnk()
191  *	Create a hard link to arcn->org_name from arcn->name. Only used in copy
192  *	with the -l flag. No warning or error if this does not succeed (we will
193  *	then just create the file)
194  * Return:
195  *	1 if copy() should try to create this file node
196  *	0 if cross_lnk() ok, -1 for fatal flaw (like linking to self).
197  */
198 
199 int
200 cross_lnk(ARCHD *arcn)
201 {
202 	/*
203 	 * try to make a link to original file (-l flag in copy mode). make sure
204 	 * we do not try to link to directories in case we are running as root
205 	 * (and it might succeed).
206 	 */
207 	if (arcn->type == PAX_DIR)
208 		return(1);
209 	return(mk_link(arcn->org_name, &(arcn->sb), arcn->name, 1));
210 }
211 
212 /*
213  * chk_same()
214  *	In copy mode if we are not trying to make hard links between the src
215  *	and destinations, make sure we are not going to overwrite ourselves by
216  *	accident. This slows things down a little, but we have to protect all
217  *	those people who make typing errors.
218  * Return:
219  *	1 the target does not exist, go ahead and copy
220  *	0 skip it file exists (-k) or may be the same as source file
221  */
222 
223 int
224 chk_same(ARCHD *arcn)
225 {
226 	struct stat sb;
227 
228 	/*
229 	 * if file does not exist, return. if file exists and -k, skip it
230 	 * quietly
231 	 */
232 	if (lstat(arcn->name, &sb) < 0)
233 		return(1);
234 	if (kflag)
235 		return(0);
236 
237 	/*
238 	 * better make sure the user does not have src == dest by mistake
239 	 */
240 	if ((arcn->sb.st_dev == sb.st_dev) && (arcn->sb.st_ino == sb.st_ino)) {
241 		paxwarn(1, "Unable to copy %s, file would overwrite itself",
242 		    arcn->name);
243 		return(0);
244 	}
245 	return(1);
246 }
247 
248 /*
249  * mk_link()
250  *	try to make a hard link between two files. if ign set, we do not
251  *	complain.
252  * Return:
253  *	0 if successful (or we are done with this file but no error, such as
254  *	finding the from file exists and the user has set -k).
255  *	1 when ign was set to indicates we could not make the link but we
256  *	should try to copy/extract the file as that might work (and is an
257  *	allowed option). -1 an error occurred.
258  */
259 
260 static int
261 mk_link(char *to, struct stat *to_sb, char *from,
262 	int ign)
263 {
264 	struct stat sb;
265 	int oerrno;
266 
267 	/*
268 	 * if from file exists, it has to be unlinked to make the link. If the
269 	 * file exists and -k is set, skip it quietly
270 	 */
271 	if (lstat(from, &sb) == 0) {
272 		if (kflag)
273 			return(0);
274 
275 		/*
276 		 * make sure it is not the same file, protect the user
277 		 */
278 		if ((to_sb->st_dev==sb.st_dev)&&(to_sb->st_ino == sb.st_ino)) {
279 			paxwarn(1, "Unable to link file %s to itself", to);
280 			return(-1);
281 		}
282 
283 		/*
284 		 * try to get rid of the file, based on the type
285 		 */
286 		if (S_ISDIR(sb.st_mode)) {
287 			if (rmdir(from) < 0) {
288 				syswarn(1, errno, "Unable to remove %s", from);
289 				return(-1);
290 			}
291 		} else if (unlink(from) < 0) {
292 			if (!ign) {
293 				syswarn(1, errno, "Unable to remove %s", from);
294 				return(-1);
295 			}
296 			return(1);
297 		}
298 	}
299 
300 	/*
301 	 * from file is gone (or did not exist), try to make the hard link.
302 	 * if it fails, check the path and try it again (if chk_path() says to
303 	 * try again)
304 	 */
305 	for (;;) {
306 		if (link(to, from) == 0)
307 			break;
308 		oerrno = errno;
309 		if (!nodirs && chk_path(from, to_sb->st_uid, to_sb->st_gid) == 0)
310 			continue;
311 		if (!ign) {
312 			syswarn(1, oerrno, "Could not link to %s from %s", to,
313 			    from);
314 			return(-1);
315 		}
316 		return(1);
317 	}
318 
319 	/*
320 	 * all right the link was made
321 	 */
322 	return(0);
323 }
324 
325 /*
326  * node_creat()
327  *	create an entry in the file system (other than a file or hard link).
328  *	If successful, sets uid/gid modes and times as required.
329  * Return:
330  *	0 if ok, -1 otherwise
331  */
332 
333 int
334 node_creat(ARCHD *arcn)
335 {
336 	int res;
337 	int ign = 0;
338 	int oerrno;
339 	int pass = 0;
340 	mode_t file_mode;
341 	struct stat sb;
342 
343 	/*
344 	 * create node based on type, if that fails try to unlink the node and
345 	 * try again. finally check the path and try again. As noted in the
346 	 * file and link creation routines, this method seems to exhibit the
347 	 * best performance in general use workloads.
348 	 */
349 	file_mode = arcn->sb.st_mode & FILEBITS;
350 
351 	for (;;) {
352 		switch(arcn->type) {
353 		case PAX_DIR:
354 			res = mkdir(arcn->name, file_mode);
355 			if (ign)
356 				res = 0;
357 			break;
358 		case PAX_CHR:
359 			file_mode |= S_IFCHR;
360 			res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
361 			break;
362 		case PAX_BLK:
363 			file_mode |= S_IFBLK;
364 			res = mknod(arcn->name, file_mode, arcn->sb.st_rdev);
365 			break;
366 		case PAX_FIF:
367 			res = mkfifo(arcn->name, file_mode);
368 			break;
369 		case PAX_SCK:
370 			/*
371 			 * Skip sockets, operation has no meaning under BSD
372 			 */
373 			paxwarn(0,
374 			    "%s skipped. Sockets cannot be copied or extracted",
375 			    arcn->name);
376 			return(-1);
377 		case PAX_SLK:
378 			res = symlink(arcn->ln_name, arcn->name);
379 			break;
380 		case PAX_CTG:
381 		case PAX_HLK:
382 		case PAX_HRG:
383 		case PAX_REG:
384 		default:
385 			/*
386 			 * we should never get here
387 			 */
388 			paxwarn(0, "%s has an unknown file type, skipping",
389 				arcn->name);
390 			return(-1);
391 		}
392 
393 		/*
394 		 * if we were able to create the node break out of the loop,
395 		 * otherwise try to unlink the node and try again. if that
396 		 * fails check the full path and try a final time.
397 		 */
398 		if (res == 0)
399 			break;
400 
401 		/*
402 		 * we failed to make the node
403 		 */
404 		oerrno = errno;
405 		if ((ign = unlnk_exist(arcn->name, arcn->type)) < 0)
406 			return(-1);
407 
408 		if (++pass <= 1)
409 			continue;
410 
411 		if (nodirs || chk_path(arcn->name,arcn->sb.st_uid,arcn->sb.st_gid) < 0) {
412 			syswarn(1, oerrno, "Could not create: %s", arcn->name);
413 			return(-1);
414 		}
415 	}
416 
417 	/*
418 	 * we were able to create the node. set uid/gid, modes and times
419 	 */
420 	if (pids)
421 		res = set_ids(arcn->name, arcn->sb.st_uid, arcn->sb.st_gid);
422 	else
423 		res = 0;
424 
425 	/*
426 	 * IMPORTANT SECURITY NOTE:
427 	 * if not preserving mode or we cannot set uid/gid, then PROHIBIT any
428 	 * set uid/gid bits
429 	 */
430 	if (!pmode || res)
431 		arcn->sb.st_mode &= ~(SETBITS);
432 	if (pmode)
433 		set_pmode(arcn->name, arcn->sb.st_mode);
434 
435 	if (arcn->type == PAX_DIR && strcmp(NM_CPIO, argv0) != 0) {
436 		/*
437 		 * Dirs must be processed again at end of extract to set times
438 		 * and modes to agree with those stored in the archive. However
439 		 * to allow extract to continue, we may have to also set owner
440 		 * rights. This allows nodes in the archive that are children
441 		 * of this directory to be extracted without failure. Both time
442 		 * and modes will be fixed after the entire archive is read and
443 		 * before pax exits.
444 		 */
445 		if (access(arcn->name, R_OK | W_OK | X_OK) < 0) {
446 			if (lstat(arcn->name, &sb) < 0) {
447 				syswarn(0, errno,"Could not access %s (stat)",
448 				    arcn->name);
449 				set_pmode(arcn->name,file_mode | S_IRWXU);
450 			} else {
451 				/*
452 				 * We have to add rights to the dir, so we make
453 				 * sure to restore the mode. The mode must be
454 				 * restored AS CREATED and not as stored if
455 				 * pmode is not set.
456 				 */
457 				set_pmode(arcn->name,
458 				    ((sb.st_mode & FILEBITS) | S_IRWXU));
459 				if (!pmode)
460 					arcn->sb.st_mode = sb.st_mode;
461 			}
462 
463 			/*
464 			 * we have to force the mode to what was set here,
465 			 * since we changed it from the default as created.
466 			 */
467 			add_dir(arcn->name, arcn->nlen, &(arcn->sb), 1);
468 		} else if (pmode || patime || pmtime)
469 			add_dir(arcn->name, arcn->nlen, &(arcn->sb), 0);
470 	}
471 
472 	if (patime || pmtime)
473 		set_ftime(arcn->name, arcn->sb.st_mtime, arcn->sb.st_atime, 0);
474 	return(0);
475 }
476 
477 /*
478  * unlnk_exist()
479  *	Remove node from file system with the specified name. We pass the type
480  *	of the node that is going to replace it. When we try to create a
481  *	directory and find that it already exists, we allow processing to
482  *	continue as proper modes etc will always be set for it later on.
483  * Return:
484  *	0 is ok to proceed, no file with the specified name exists
485  *	-1 we were unable to remove the node, or we should not remove it (-k)
486  *	1 we found a directory and we were going to create a directory.
487  */
488 
489 int
490 unlnk_exist(char *name, int type)
491 {
492 	struct stat sb;
493 
494 	/*
495 	 * the file does not exist, or -k we are done
496 	 */
497 	if (lstat(name, &sb) < 0)
498 		return(0);
499 	if (kflag)
500 		return(-1);
501 
502 	if (S_ISDIR(sb.st_mode)) {
503 		/*
504 		 * try to remove a directory, if it fails and we were going to
505 		 * create a directory anyway, tell the caller (return a 1)
506 		 */
507 		if (rmdir(name) < 0) {
508 			if (type == PAX_DIR)
509 				return(1);
510 			syswarn(1,errno,"Unable to remove directory %s", name);
511 			return(-1);
512 		}
513 		return(0);
514 	}
515 
516 	/*
517 	 * try to get rid of all non-directory type nodes
518 	 */
519 	if (unlink(name) < 0) {
520 		syswarn(1, errno, "Could not unlink %s", name);
521 		return(-1);
522 	}
523 	return(0);
524 }
525 
526 /*
527  * chk_path()
528  *	We were trying to create some kind of node in the file system and it
529  *	failed. chk_path() makes sure the path up to the node exists and is
530  *	writeable. When we have to create a directory that is missing along the
531  *	path somewhere, the directory we create will be set to the same
532  *	uid/gid as the file has (when uid and gid are being preserved).
533  *	NOTE: this routine is a real performance loss. It is only used as a
534  *	last resort when trying to create entries in the file system.
535  * Return:
536  *	-1 when it could find nothing it is allowed to fix.
537  *	0 otherwise
538  */
539 
540 int
541 chk_path( char *name, uid_t st_uid, gid_t st_gid)
542 {
543 	char *spt = name;
544 	struct stat sb;
545 	int retval = -1;
546 
547 	/*
548 	 * watch out for paths with nodes stored directly in / (e.g. /bozo)
549 	 */
550 	if (*spt == '/')
551 		++spt;
552 
553 	for(;;) {
554 		/*
555 		 * work forward from the first / and check each part of the path
556 		 */
557 		spt = strchr(spt, '/');
558 		if (spt == NULL)
559 			break;
560 		*spt = '\0';
561 
562 		/*
563 		 * if it exists we assume it is a directory, it is not within
564 		 * the spec (at least it seems to read that way) to alter the
565 		 * file system for nodes NOT EXPLICITLY stored on the archive.
566 		 * If that assumption is changed, you would test the node here
567 		 * and figure out how to get rid of it (probably like some
568 		 * recursive unlink()) or fix up the directory permissions if
569 		 * required (do an access()).
570 		 */
571 		if (lstat(name, &sb) == 0) {
572 			*(spt++) = '/';
573 			continue;
574 		}
575 
576 		/*
577 		 * the path fails at this point, see if we can create the
578 		 * needed directory and continue on
579 		 */
580 		if (mkdir(name, S_IRWXU | S_IRWXG | S_IRWXO) < 0) {
581 			*spt = '/';
582 			retval = -1;
583 			break;
584 		}
585 
586 		/*
587 		 * we were able to create the directory. We will tell the
588 		 * caller that we found something to fix, and it is ok to try
589 		 * and create the node again.
590 		 */
591 		retval = 0;
592 		if (pids)
593 			(void)set_ids(name, st_uid, st_gid);
594 
595 		/*
596 		 * make sure the user doesn't have some strange umask that
597 		 * causes this newly created directory to be unusable. We fix
598 		 * the modes and restore them back to the creation default at
599 		 * the end of pax
600 		 */
601 		if ((access(name, R_OK | W_OK | X_OK) < 0) &&
602 		    (lstat(name, &sb) == 0)) {
603 			set_pmode(name, ((sb.st_mode & FILEBITS) | S_IRWXU));
604 			add_dir(name, spt - name, &sb, 1);
605 		}
606 		*(spt++) = '/';
607 		continue;
608 	}
609 	return(retval);
610 }
611 
612 /*
613  * set_ftime()
614  *	Set the access time and modification time for a named file. If frc is
615  *	non-zero we force these times to be set even if the user did not
616  *	request access and/or modification time preservation (this is also
617  *	used by -t to reset access times).
618  *	When frc is zero, only those times the user has asked for are set, the
619  *	other ones are left alone. We do not assume the un-documented feature
620  *	of many lutimes() implementations that consider a 0 time value as a do
621  *	not set request.
622  */
623 
624 void
625 set_ftime(char *fnm, time_t mtime, time_t atime, int frc)
626 {
627 	static struct timeval tv[2] = {{0L, 0L}, {0L, 0L}};
628 	struct stat sb;
629 
630 	tv[0].tv_sec = atime;
631 	tv[1].tv_sec = mtime;
632 	if (!frc && (!patime || !pmtime)) {
633 		/*
634 		 * if we are not forcing, only set those times the user wants
635 		 * set. We get the current values of the times if we need them.
636 		 */
637 		if (lstat(fnm, &sb) == 0) {
638 			if (!patime)
639 				tv[0].tv_sec = sb.st_atime;
640 			if (!pmtime)
641 				tv[1].tv_sec = sb.st_mtime;
642 		} else
643 			syswarn(0,errno,"Unable to obtain file stats %s", fnm);
644 	}
645 
646 	/*
647 	 * set the times
648 	 */
649 	if (lutimes(fnm, tv) < 0)
650 		syswarn(1, errno, "Access/modification time set failed on: %s",
651 		    fnm);
652 	return;
653 }
654 
655 /*
656  * set_ids()
657  *	set the uid and gid of a file system node
658  * Return:
659  *	0 when set, -1 on failure
660  */
661 
662 int
663 set_ids(char *fnm, uid_t uid, gid_t gid)
664 {
665 	if (lchown(fnm, uid, gid) < 0) {
666 		/*
667 		 * ignore EPERM unless in verbose mode or being run by root.
668 		 * if running as pax, POSIX requires a warning.
669 		 */
670 		if (strcmp(NM_PAX, argv0) == 0 || errno != EPERM || vflag ||
671 		    geteuid() == 0)
672 			syswarn(1, errno, "Unable to set file uid/gid of %s",
673 			    fnm);
674 		return(-1);
675 	}
676 	return(0);
677 }
678 
679 /*
680  * set_pmode()
681  *	Set file access mode
682  */
683 
684 void
685 set_pmode(char *fnm, mode_t mode)
686 {
687 	mode &= ABITS;
688 	if (lchmod(fnm, mode) < 0)
689 		syswarn(1, errno, "Could not set permissions on %s", fnm);
690 	return;
691 }
692 
693 /*
694  * file_write()
695  *	Write/copy a file (during copy or archive extract). This routine knows
696  *	how to copy files with lseek holes in it. (Which are read as file
697  *	blocks containing all 0's but do not have any file blocks associated
698  *	with the data). Typical examples of these are files created by dbm
699  *	variants (.pag files). While the file size of these files are huge, the
700  *	actual storage is quite small (the files are sparse). The problem is
701  *	the holes read as all zeros so are probably stored on the archive that
702  *	way (there is no way to determine if the file block is really a hole,
703  *	we only know that a file block of all zero's can be a hole).
704  *	At this writing, no major archive format knows how to archive files
705  *	with holes. However, on extraction (or during copy, -rw) we have to
706  *	deal with these files. Without detecting the holes, the files can
707  *	consume a lot of file space if just written to disk. This replacement
708  *	for write when passed the basic allocation size of a file system block,
709  *	uses lseek whenever it detects the input data is all 0 within that
710  *	file block. In more detail, the strategy is as follows:
711  *	While the input is all zero keep doing an lseek. Keep track of when we
712  *	pass over file block boundaries. Only write when we hit a non zero
713  *	input. once we have written a file block, we continue to write it to
714  *	the end (we stop looking at the input). When we reach the start of the
715  *	next file block, start checking for zero blocks again. Working on file
716  *	block boundaries significantly reduces the overhead when copying files
717  *	that are NOT very sparse. This overhead (when compared to a write) is
718  *	almost below the measurement resolution on many systems. Without it,
719  *	files with holes cannot be safely copied. It does has a side effect as
720  *	it can put holes into files that did not have them before, but that is
721  *	not a problem since the file contents are unchanged (in fact it saves
722  *	file space). (Except on paging files for diskless clients. But since we
723  *	cannot determine one of those file from here, we ignore them). If this
724  *	ever ends up on a system where CTG files are supported and the holes
725  *	are not desired, just do a conditional test in those routines that
726  *	call file_write() and have it call write() instead. BEFORE CLOSING THE
727  *	FILE, make sure to call file_flush() when the last write finishes with
728  *	an empty block. A lot of file systems will not create an lseek hole at
729  *	the end. In this case we drop a single 0 at the end to force the
730  *	trailing 0's in the file.
731  *	---Parameters---
732  *	rem: how many bytes left in this file system block
733  *	isempt: have we written to the file block yet (is it empty)
734  *	sz: basic file block allocation size
735  *	cnt: number of bytes on this write
736  *	str: buffer to write
737  * Return:
738  *	number of bytes written, -1 on write (or lseek) error.
739  */
740 
741 int
742 file_write(int fd, char *str, int cnt, int *rem, int *isempt, int sz,
743 	char *name)
744 {
745 	char *pt;
746 	char *end;
747 	int wcnt;
748 	char *st = str;
749 
750 	/*
751 	 * while we have data to process
752 	 */
753 	while (cnt) {
754 		if (!*rem) {
755 			/*
756 			 * We are now at the start of file system block again
757 			 * (or what we think one is...). start looking for
758 			 * empty blocks again
759 			 */
760 			*isempt = 1;
761 			*rem = sz;
762 		}
763 
764 		/*
765 		 * only examine up to the end of the current file block or
766 		 * remaining characters to write, whatever is smaller
767 		 */
768 		wcnt = MIN(cnt, *rem);
769 		cnt -= wcnt;
770 		*rem -= wcnt;
771 		if (*isempt) {
772 			/*
773 			 * have not written to this block yet, so we keep
774 			 * looking for zero's
775 			 */
776 			pt = st;
777 			end = st + wcnt;
778 
779 			/*
780 			 * look for a zero filled buffer
781 			 */
782 			while ((pt < end) && (*pt == '\0'))
783 				++pt;
784 
785 			if (pt == end) {
786 				/*
787 				 * skip, buf is empty so far
788 				 */
789 				if (lseek(fd, (off_t)wcnt, SEEK_CUR) < 0) {
790 					syswarn(1,errno,"File seek on %s",
791 					    name);
792 					return(-1);
793 				}
794 				st = pt;
795 				continue;
796 			}
797 			/*
798 			 * drat, the buf is not zero filled
799 			 */
800 			*isempt = 0;
801 		}
802 
803 		/*
804 		 * have non-zero data in this file system block, have to write
805 		 */
806 		if (write(fd, st, wcnt) != wcnt) {
807 			syswarn(1, errno, "Failed write to file %s", name);
808 			return(-1);
809 		}
810 		st += wcnt;
811 	}
812 	return(st - str);
813 }
814 
815 /*
816  * file_flush()
817  *	when the last file block in a file is zero, many file systems will not
818  *	let us create a hole at the end. To get the last block with zeros, we
819  *	write the last BYTE with a zero (back up one byte and write a zero).
820  */
821 
822 void
823 file_flush(int fd, char *fname, int isempt)
824 {
825 	static char blnk[] = "\0";
826 
827 	/*
828 	 * silly test, but make sure we are only called when the last block is
829 	 * filled with all zeros.
830 	 */
831 	if (!isempt)
832 		return;
833 
834 	/*
835 	 * move back one byte and write a zero
836 	 */
837 	if (lseek(fd, (off_t)-1, SEEK_CUR) < 0) {
838 		syswarn(1, errno, "Failed seek on file %s", fname);
839 		return;
840 	}
841 
842 	if (write(fd, blnk, 1) < 0)
843 		syswarn(1, errno, "Failed write to file %s", fname);
844 	return;
845 }
846 
847 /*
848  * rdfile_close()
849  *	close a file we have beed reading (to copy or archive). If we have to
850  *	reset access time (tflag) do so (the times are stored in arcn).
851  */
852 
853 void
854 rdfile_close(ARCHD *arcn, int *fd)
855 {
856 	/*
857 	 * make sure the file is open
858 	 */
859 	if (*fd < 0)
860 		return;
861 
862 	(void)close(*fd);
863 	*fd = -1;
864 	if (!tflag)
865 		return;
866 
867 	/*
868 	 * user wants last access time reset
869 	 */
870 	set_ftime(arcn->org_name, arcn->sb.st_mtime, arcn->sb.st_atime, 1);
871 	return;
872 }
873 
874 /*
875  * set_crc()
876  *	read a file to calculate its crc. This is a real drag. Archive formats
877  *	that have this, end up reading the file twice (we have to write the
878  *	header WITH the crc before writing the file contents. Oh well...
879  * Return:
880  *	0 if was able to calculate the crc, -1 otherwise
881  */
882 
883 int
884 set_crc(ARCHD *arcn, int fd)
885 {
886 	int i;
887 	int res;
888 	off_t cpcnt = 0L;
889 	u_long size;
890 	unsigned long crc = 0L;
891 	char tbuf[FILEBLK];
892 	struct stat sb;
893 
894 	if (fd < 0) {
895 		/*
896 		 * hmm, no fd, should never happen. well no crc then.
897 		 */
898 		arcn->crc = 0L;
899 		return(0);
900 	}
901 
902 	if ((size = (u_long)arcn->sb.st_blksize) > (u_long)sizeof(tbuf))
903 		size = (u_long)sizeof(tbuf);
904 
905 	/*
906 	 * read all the bytes we think that there are in the file. If the user
907 	 * is trying to archive an active file, forget this file.
908 	 */
909 	for(;;) {
910 		if ((res = read(fd, tbuf, size)) <= 0)
911 			break;
912 		cpcnt += res;
913 		for (i = 0; i < res; ++i)
914 			crc += (tbuf[i] & 0xff);
915 	}
916 
917 	/*
918 	 * safety check. we want to avoid archiving files that are active as
919 	 * they can create inconsistent archive copies.
920 	 */
921 	if (cpcnt != arcn->sb.st_size)
922 		paxwarn(1, "File changed size %s", arcn->org_name);
923 	else if (fstat(fd, &sb) < 0)
924 		syswarn(1, errno, "Failed stat on %s", arcn->org_name);
925 	else if (arcn->sb.st_mtime != sb.st_mtime)
926 		paxwarn(1, "File %s was modified during read", arcn->org_name);
927 	else if (lseek(fd, (off_t)0L, SEEK_SET) < 0)
928 		syswarn(1, errno, "File rewind failed on: %s", arcn->org_name);
929 	else {
930 		arcn->crc = crc;
931 		return(0);
932 	}
933 	return(-1);
934 }
935