xref: /freebsd/contrib/file/src/compress.c (revision 4e99f45480598189d49d45a825533a6c9e12f02c)
1 /*
2  * Copyright (c) Ian F. Darwin 1986-1995.
3  * Software written by Ian F. Darwin and others;
4  * maintained 1995-present by Christos Zoulas and others.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice immediately at the beginning of the file, without modification,
11  *    this list of conditions, and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR
20  * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 /*
29  * compress routines:
30  *	zmagic() - returns 0 if not recognized, uncompresses and prints
31  *		   information if recognized
32  *	uncompress(method, old, n, newch) - uncompress old into new,
33  *					    using method, return sizeof new
34  */
35 #include "file.h"
36 
37 #ifndef lint
38 FILE_RCSID("@(#)$File: compress.c,v 1.124 2019/07/21 11:42:09 christos Exp $")
39 #endif
40 
41 #include "magic.h"
42 #include <stdlib.h>
43 #ifdef HAVE_UNISTD_H
44 #include <unistd.h>
45 #endif
46 #include <string.h>
47 #include <errno.h>
48 #include <ctype.h>
49 #include <stdarg.h>
50 #include <signal.h>
51 #ifndef HAVE_SIG_T
52 typedef void (*sig_t)(int);
53 #endif /* HAVE_SIG_T */
54 #if !defined(__MINGW32__) && !defined(WIN32)
55 #include <sys/ioctl.h>
56 #endif
57 #ifdef HAVE_SYS_WAIT_H
58 #include <sys/wait.h>
59 #endif
60 #if defined(HAVE_SYS_TIME_H)
61 #include <sys/time.h>
62 #endif
63 
64 #if defined(HAVE_ZLIB_H) && defined(ZLIBSUPPORT)
65 #define BUILTIN_DECOMPRESS
66 #include <zlib.h>
67 #endif
68 
69 #if defined(HAVE_BZLIB_H) || defined(BZLIBSUPPORT)
70 #define BUILTIN_BZLIB
71 #include <bzlib.h>
72 #endif
73 
74 #if defined(HAVE_XZLIB_H) || defined(XZLIBSUPPORT)
75 #define BUILTIN_XZLIB
76 #include <lzma.h>
77 #endif
78 
79 #ifdef DEBUG
80 int tty = -1;
81 #define DPRINTF(...)	do { \
82 	if (tty == -1) \
83 		tty = open("/dev/tty", O_RDWR); \
84 	if (tty == -1) \
85 		abort(); \
86 	dprintf(tty, __VA_ARGS__); \
87 } while (/*CONSTCOND*/0)
88 #else
89 #define DPRINTF(...)
90 #endif
91 
92 #ifdef ZLIBSUPPORT
93 /*
94  * The following python code is not really used because ZLIBSUPPORT is only
95  * defined if we have a built-in zlib, and the built-in zlib handles that.
96  * That is not true for android where we have zlib.h and not -lz.
97  */
98 static const char zlibcode[] =
99     "import sys, zlib; sys.stdout.write(zlib.decompress(sys.stdin.read()))";
100 
101 static const char *zlib_args[] = { "python", "-c", zlibcode, NULL };
102 
103 static int
104 zlibcmp(const unsigned char *buf)
105 {
106 	unsigned short x = 1;
107 	unsigned char *s = CAST(unsigned char *, CAST(void *, &x));
108 
109 	if ((buf[0] & 0xf) != 8 || (buf[0] & 0x80) != 0)
110 		return 0;
111 	if (s[0] != 1)	/* endianness test */
112 		x = buf[0] | (buf[1] << 8);
113 	else
114 		x = buf[1] | (buf[0] << 8);
115 	if (x % 31)
116 		return 0;
117 	return 1;
118 }
119 #endif
120 
121 static int
122 lzmacmp(const unsigned char *buf)
123 {
124 	if (buf[0] != 0x5d || buf[1] || buf[2])
125 		return 0;
126 	if (buf[12] && buf[12] != 0xff)
127 		return 0;
128 	return 1;
129 }
130 
131 #define gzip_flags "-cd"
132 #define lrzip_flags "-do"
133 #define lzip_flags gzip_flags
134 
135 static const char *gzip_args[] = {
136 	"gzip", gzip_flags, NULL
137 };
138 static const char *uncompress_args[] = {
139 	"uncompress", "-c", NULL
140 };
141 static const char *bzip2_args[] = {
142 	"bzip2", "-cd", NULL
143 };
144 static const char *lzip_args[] = {
145 	"lzip", lzip_flags, NULL
146 };
147 static const char *xz_args[] = {
148 	"xz", "-cd", NULL
149 };
150 static const char *lrzip_args[] = {
151 	"lrzip", lrzip_flags, NULL
152 };
153 static const char *lz4_args[] = {
154 	"lz4", "-cd", NULL
155 };
156 static const char *zstd_args[] = {
157 	"zstd", "-cd", NULL
158 };
159 
160 #define	do_zlib		NULL
161 #define	do_bzlib	NULL
162 
163 private const struct {
164 	const void *magic;
165 	int maglen;
166 	const char **argv;
167 	void *unused;
168 } compr[] = {
169 #define METH_FROZEN	2
170 #define METH_BZIP	7
171 #define METH_XZ		9
172 #define METH_LZMA	13
173 #define METH_ZLIB	14
174 	{ "\037\235",	2, gzip_args, NULL },		/* 0, compressed */
175 	/* Uncompress can get stuck; so use gzip first if we have it
176 	 * Idea from Damien Clark, thanks! */
177 	{ "\037\235",	2, uncompress_args, NULL },	/* 1, compressed */
178 	{ "\037\213",	2, gzip_args, do_zlib },	/* 2, gzipped */
179 	{ "\037\236",	2, gzip_args, NULL },		/* 3, frozen */
180 	{ "\037\240",	2, gzip_args, NULL },		/* 4, SCO LZH */
181 	/* the standard pack utilities do not accept standard input */
182 	{ "\037\036",	2, gzip_args, NULL },		/* 5, packed */
183 	{ "PK\3\4",	4, gzip_args, NULL },		/* 6, pkzipped, */
184 	/* ...only first file examined */
185 	{ "BZh",	3, bzip2_args, do_bzlib },	/* 7, bzip2-ed */
186 	{ "LZIP",	4, lzip_args, NULL },		/* 8, lzip-ed */
187  	{ "\3757zXZ\0",	6, xz_args, NULL },		/* 9, XZ Utils */
188  	{ "LRZI",	4, lrzip_args, NULL },	/* 10, LRZIP */
189  	{ "\004\"M\030",4, lz4_args, NULL },		/* 11, LZ4 */
190  	{ "\x28\xB5\x2F\xFD", 4, zstd_args, NULL },	/* 12, zstd */
191 	{ RCAST(const void *, lzmacmp),	-13, xz_args, NULL },	/* 13, lzma */
192 #ifdef ZLIBSUPPORT
193 	{ RCAST(const void *, zlibcmp),	-2, zlib_args, NULL },	/* 14, zlib */
194 #endif
195 };
196 
197 #define OKDATA 	0
198 #define NODATA	1
199 #define ERRDATA	2
200 
201 private ssize_t swrite(int, const void *, size_t);
202 #if HAVE_FORK
203 private size_t ncompr = __arraycount(compr);
204 private int uncompressbuf(int, size_t, size_t, const unsigned char *,
205     unsigned char **, size_t *);
206 #ifdef BUILTIN_DECOMPRESS
207 private int uncompresszlib(const unsigned char *, unsigned char **, size_t,
208     size_t *, int);
209 private int uncompressgzipped(const unsigned char *, unsigned char **, size_t,
210     size_t *);
211 #endif
212 #ifdef BUILTIN_BZLIB
213 private int uncompressbzlib(const unsigned char *, unsigned char **, size_t,
214     size_t *);
215 #endif
216 #ifdef BUILTIN_XZLIB
217 private int uncompressxzlib(const unsigned char *, unsigned char **, size_t,
218     size_t *);
219 #endif
220 
221 static int makeerror(unsigned char **, size_t *, const char *, ...)
222     __attribute__((__format__(__printf__, 3, 4)));
223 private const char *methodname(size_t);
224 
225 private int
226 format_decompression_error(struct magic_set *ms, size_t i, unsigned char *buf)
227 {
228 	unsigned char *p;
229 	int mime = ms->flags & MAGIC_MIME;
230 
231 	if (!mime)
232 		return file_printf(ms, "ERROR:[%s: %s]", methodname(i), buf);
233 
234 	for (p = buf; *p; p++)
235 		if (!isalnum(*p))
236 			*p = '-';
237 
238 	return file_printf(ms, "application/x-decompression-error-%s-%s",
239 	    methodname(i), buf);
240 }
241 
242 protected int
243 file_zmagic(struct magic_set *ms, const struct buffer *b, const char *name)
244 {
245 	unsigned char *newbuf = NULL;
246 	size_t i, nsz;
247 	char *rbuf;
248 	file_pushbuf_t *pb;
249 	int urv, prv, rv = 0;
250 	int mime = ms->flags & MAGIC_MIME;
251 	int fd = b->fd;
252 	const unsigned char *buf = CAST(const unsigned char *, b->fbuf);
253 	size_t nbytes = b->flen;
254 	int sa_saved = 0;
255 	struct sigaction sig_act;
256 
257 	if ((ms->flags & MAGIC_COMPRESS) == 0)
258 		return 0;
259 
260 	for (i = 0; i < ncompr; i++) {
261 		int zm;
262 		if (nbytes < CAST(size_t, abs(compr[i].maglen)))
263 			continue;
264 		if (compr[i].maglen < 0) {
265 			zm = (RCAST(int (*)(const unsigned char *),
266 			    CCAST(void *, compr[i].magic)))(buf);
267 		} else {
268 			zm = memcmp(buf, compr[i].magic,
269 			    CAST(size_t, compr[i].maglen)) == 0;
270 		}
271 
272 		if (!zm)
273 			continue;
274 
275 		/* Prevent SIGPIPE death if child dies unexpectedly */
276 		if (!sa_saved) {
277 			//We can use sig_act for both new and old, but
278 			struct sigaction new_act;
279 			memset(&new_act, 0, sizeof(new_act));
280 			new_act.sa_handler = SIG_IGN;
281 			sa_saved = sigaction(SIGPIPE, &new_act, &sig_act) != -1;
282 		}
283 
284 		nsz = nbytes;
285 		urv = uncompressbuf(fd, ms->bytes_max, i, buf, &newbuf, &nsz);
286 		DPRINTF("uncompressbuf = %d, %s, %" SIZE_T_FORMAT "u\n", urv,
287 		    (char *)newbuf, nsz);
288 		switch (urv) {
289 		case OKDATA:
290 		case ERRDATA:
291 			ms->flags &= ~MAGIC_COMPRESS;
292 			if (urv == ERRDATA)
293 				prv = format_decompression_error(ms, i, newbuf);
294 			else
295 				prv = file_buffer(ms, -1, NULL, name, newbuf, nsz);
296 			if (prv == -1)
297 				goto error;
298 			rv = 1;
299 			if ((ms->flags & MAGIC_COMPRESS_TRANSP) != 0)
300 				goto out;
301 			if (mime != MAGIC_MIME && mime != 0)
302 				goto out;
303 			if ((file_printf(ms,
304 			    mime ? " compressed-encoding=" : " (")) == -1)
305 				goto error;
306 			if ((pb = file_push_buffer(ms)) == NULL)
307 				goto error;
308 			/*
309 			 * XXX: If file_buffer fails here, we overwrite
310 			 * the compressed text. FIXME.
311 			 */
312 			if (file_buffer(ms, -1, NULL, NULL, buf, nbytes) == -1) {
313 				if (file_pop_buffer(ms, pb) != NULL)
314 					abort();
315 				goto error;
316 			}
317 			if ((rbuf = file_pop_buffer(ms, pb)) != NULL) {
318 				if (file_printf(ms, "%s", rbuf) == -1) {
319 					free(rbuf);
320 					goto error;
321 				}
322 				free(rbuf);
323 			}
324 			if (!mime && file_printf(ms, ")") == -1)
325 				goto error;
326 			/*FALLTHROUGH*/
327 		case NODATA:
328 			break;
329 		default:
330 			abort();
331 			/*NOTREACHED*/
332 		error:
333 			rv = -1;
334 			break;
335 		}
336 	}
337 out:
338 	DPRINTF("rv = %d\n", rv);
339 
340 	if (sa_saved && sig_act.sa_handler != SIG_IGN)
341 		(void)sigaction(SIGPIPE, &sig_act, NULL);
342 
343 	free(newbuf);
344 	ms->flags |= MAGIC_COMPRESS;
345 	DPRINTF("Zmagic returns %d\n", rv);
346 	return rv;
347 }
348 #endif
349 /*
350  * `safe' write for sockets and pipes.
351  */
352 private ssize_t
353 swrite(int fd, const void *buf, size_t n)
354 {
355 	ssize_t rv;
356 	size_t rn = n;
357 
358 	do
359 		switch (rv = write(fd, buf, n)) {
360 		case -1:
361 			if (errno == EINTR)
362 				continue;
363 			return -1;
364 		default:
365 			n -= rv;
366 			buf = CAST(const char *, buf) + rv;
367 			break;
368 		}
369 	while (n > 0);
370 	return rn;
371 }
372 
373 
374 /*
375  * `safe' read for sockets and pipes.
376  */
377 protected ssize_t
378 sread(int fd, void *buf, size_t n, int canbepipe __attribute__((__unused__)))
379 {
380 	ssize_t rv;
381 #ifdef FIONREAD
382 	int t = 0;
383 #endif
384 	size_t rn = n;
385 
386 	if (fd == STDIN_FILENO)
387 		goto nocheck;
388 
389 #ifdef FIONREAD
390 	if (canbepipe && (ioctl(fd, FIONREAD, &t) == -1 || t == 0)) {
391 #ifdef FD_ZERO
392 		ssize_t cnt;
393 		for (cnt = 0;; cnt++) {
394 			fd_set check;
395 			struct timeval tout = {0, 100 * 1000};
396 			int selrv;
397 
398 			FD_ZERO(&check);
399 			FD_SET(fd, &check);
400 
401 			/*
402 			 * Avoid soft deadlock: do not read if there
403 			 * is nothing to read from sockets and pipes.
404 			 */
405 			selrv = select(fd + 1, &check, NULL, NULL, &tout);
406 			if (selrv == -1) {
407 				if (errno == EINTR || errno == EAGAIN)
408 					continue;
409 			} else if (selrv == 0 && cnt >= 5) {
410 				return 0;
411 			} else
412 				break;
413 		}
414 #endif
415 		(void)ioctl(fd, FIONREAD, &t);
416 	}
417 
418 	if (t > 0 && CAST(size_t, t) < n) {
419 		n = t;
420 		rn = n;
421 	}
422 #endif
423 
424 nocheck:
425 	do
426 		switch ((rv = read(fd, buf, n))) {
427 		case -1:
428 			if (errno == EINTR)
429 				continue;
430 			return -1;
431 		case 0:
432 			return rn - n;
433 		default:
434 			n -= rv;
435 			buf = CAST(char *, CCAST(void *, buf)) + rv;
436 			break;
437 		}
438 	while (n > 0);
439 	return rn;
440 }
441 
442 protected int
443 file_pipe2file(struct magic_set *ms, int fd, const void *startbuf,
444     size_t nbytes)
445 {
446 	char buf[4096];
447 	ssize_t r;
448 	int tfd;
449 
450 	(void)strlcpy(buf, "/tmp/file.XXXXXX", sizeof buf);
451 #ifndef HAVE_MKSTEMP
452 	{
453 		char *ptr = mktemp(buf);
454 		tfd = open(ptr, O_RDWR|O_TRUNC|O_EXCL|O_CREAT, 0600);
455 		r = errno;
456 		(void)unlink(ptr);
457 		errno = r;
458 	}
459 #else
460 	{
461 		int te;
462 		mode_t ou = umask(0);
463 		tfd = mkstemp(buf);
464 		(void)umask(ou);
465 		te = errno;
466 		(void)unlink(buf);
467 		errno = te;
468 	}
469 #endif
470 	if (tfd == -1) {
471 		file_error(ms, errno,
472 		    "cannot create temporary file for pipe copy");
473 		return -1;
474 	}
475 
476 	if (swrite(tfd, startbuf, nbytes) != CAST(ssize_t, nbytes))
477 		r = 1;
478 	else {
479 		while ((r = sread(fd, buf, sizeof(buf), 1)) > 0)
480 			if (swrite(tfd, buf, CAST(size_t, r)) != r)
481 				break;
482 	}
483 
484 	switch (r) {
485 	case -1:
486 		file_error(ms, errno, "error copying from pipe to temp file");
487 		return -1;
488 	case 0:
489 		break;
490 	default:
491 		file_error(ms, errno, "error while writing to temp file");
492 		return -1;
493 	}
494 
495 	/*
496 	 * We duplicate the file descriptor, because fclose on a
497 	 * tmpfile will delete the file, but any open descriptors
498 	 * can still access the phantom inode.
499 	 */
500 	if ((fd = dup2(tfd, fd)) == -1) {
501 		file_error(ms, errno, "could not dup descriptor for temp file");
502 		return -1;
503 	}
504 	(void)close(tfd);
505 	if (lseek(fd, CAST(off_t, 0), SEEK_SET) == CAST(off_t, -1)) {
506 		file_badseek(ms);
507 		return -1;
508 	}
509 	return fd;
510 }
511 #if HAVE_FORK
512 #ifdef BUILTIN_DECOMPRESS
513 
514 #define FHCRC		(1 << 1)
515 #define FEXTRA		(1 << 2)
516 #define FNAME		(1 << 3)
517 #define FCOMMENT	(1 << 4)
518 
519 
520 private int
521 uncompressgzipped(const unsigned char *old, unsigned char **newch,
522     size_t bytes_max, size_t *n)
523 {
524 	unsigned char flg = old[3];
525 	size_t data_start = 10;
526 
527 	if (flg & FEXTRA) {
528 		if (data_start + 1 >= *n)
529 			goto err;
530 		data_start += 2 + old[data_start] + old[data_start + 1] * 256;
531 	}
532 	if (flg & FNAME) {
533 		while(data_start < *n && old[data_start])
534 			data_start++;
535 		data_start++;
536 	}
537 	if (flg & FCOMMENT) {
538 		while(data_start < *n && old[data_start])
539 			data_start++;
540 		data_start++;
541 	}
542 	if (flg & FHCRC)
543 		data_start += 2;
544 
545 	if (data_start >= *n)
546 		goto err;
547 
548 	*n -= data_start;
549 	old += data_start;
550 	return uncompresszlib(old, newch, bytes_max, n, 0);
551 err:
552 	return makeerror(newch, n, "File too short");
553 }
554 
555 private int
556 uncompresszlib(const unsigned char *old, unsigned char **newch,
557     size_t bytes_max, size_t *n, int zlib)
558 {
559 	int rc;
560 	z_stream z;
561 
562 	if ((*newch = CAST(unsigned char *, malloc(bytes_max + 1))) == NULL)
563 		return makeerror(newch, n, "No buffer, %s", strerror(errno));
564 
565 	z.next_in = CCAST(Bytef *, old);
566 	z.avail_in = CAST(uint32_t, *n);
567 	z.next_out = *newch;
568 	z.avail_out = CAST(unsigned int, bytes_max);
569 	z.zalloc = Z_NULL;
570 	z.zfree = Z_NULL;
571 	z.opaque = Z_NULL;
572 
573 	/* LINTED bug in header macro */
574 	rc = zlib ? inflateInit(&z) : inflateInit2(&z, -15);
575 	if (rc != Z_OK)
576 		goto err;
577 
578 	rc = inflate(&z, Z_SYNC_FLUSH);
579 	if (rc != Z_OK && rc != Z_STREAM_END)
580 		goto err;
581 
582 	*n = CAST(size_t, z.total_out);
583 	rc = inflateEnd(&z);
584 	if (rc != Z_OK)
585 		goto err;
586 
587 	/* let's keep the nul-terminate tradition */
588 	(*newch)[*n] = '\0';
589 
590 	return OKDATA;
591 err:
592 	strlcpy(RCAST(char *, *newch), z.msg ? z.msg : zError(rc), bytes_max);
593 	*n = strlen(RCAST(char *, *newch));
594 	return ERRDATA;
595 }
596 #endif
597 
598 #ifdef BUILTIN_BZLIB
599 private int
600 uncompressbzlib(const unsigned char *old, unsigned char **newch,
601     size_t bytes_max, size_t *n)
602 {
603 	int rc;
604 	bz_stream bz;
605 
606 	memset(&bz, 0, sizeof(bz));
607 	rc = BZ2_bzDecompressInit(&bz, 0, 0);
608 	if (rc != BZ_OK)
609 		goto err;
610 
611 	if ((*newch = CAST(unsigned char *, malloc(bytes_max + 1))) == NULL)
612 		return makeerror(newch, n, "No buffer, %s", strerror(errno));
613 
614 	bz.next_in = CCAST(char *, RCAST(const char *, old));
615 	bz.avail_in = CAST(uint32_t, *n);
616 	bz.next_out = RCAST(char *, *newch);
617 	bz.avail_out = CAST(unsigned int, bytes_max);
618 
619 	rc = BZ2_bzDecompress(&bz);
620 	if (rc != BZ_OK && rc != BZ_STREAM_END)
621 		goto err;
622 
623 	/* Assume byte_max is within 32bit */
624 	/* assert(bz.total_out_hi32 == 0); */
625 	*n = CAST(size_t, bz.total_out_lo32);
626 	rc = BZ2_bzDecompressEnd(&bz);
627 	if (rc != BZ_OK)
628 		goto err;
629 
630 	/* let's keep the nul-terminate tradition */
631 	(*newch)[*n] = '\0';
632 
633 	return OKDATA;
634 err:
635 	snprintf(RCAST(char *, *newch), bytes_max, "bunzip error %d", rc);
636 	*n = strlen(RCAST(char *, *newch));
637 	return ERRDATA;
638 }
639 #endif
640 
641 #ifdef BUILTIN_XZLIB
642 private int
643 uncompressxzlib(const unsigned char *old, unsigned char **newch,
644     size_t bytes_max, size_t *n)
645 {
646 	int rc;
647 	lzma_stream xz;
648 
649 	memset(&xz, 0, sizeof(xz));
650 	rc = lzma_auto_decoder(&xz, UINT64_MAX, 0);
651 	if (rc != LZMA_OK)
652 		goto err;
653 
654 	if ((*newch = CAST(unsigned char *, malloc(bytes_max + 1))) == NULL)
655 		return makeerror(newch, n, "No buffer, %s", strerror(errno));
656 
657 	xz.next_in = CCAST(const uint8_t *, old);
658 	xz.avail_in = CAST(uint32_t, *n);
659 	xz.next_out = RCAST(uint8_t *, *newch);
660 	xz.avail_out = CAST(unsigned int, bytes_max);
661 
662 	rc = lzma_code(&xz, LZMA_RUN);
663 	if (rc != LZMA_OK && rc != LZMA_STREAM_END)
664 		goto err;
665 
666 	*n = CAST(size_t, xz.total_out);
667 
668 	lzma_end(&xz);
669 
670 	/* let's keep the nul-terminate tradition */
671 	(*newch)[*n] = '\0';
672 
673 	return OKDATA;
674 err:
675 	snprintf(RCAST(char *, *newch), bytes_max, "unxz error %d", rc);
676 	*n = strlen(RCAST(char *, *newch));
677 	return ERRDATA;
678 }
679 #endif
680 
681 
682 static int
683 makeerror(unsigned char **buf, size_t *len, const char *fmt, ...)
684 {
685 	char *msg;
686 	va_list ap;
687 	int rv;
688 
689 	va_start(ap, fmt);
690 	rv = vasprintf(&msg, fmt, ap);
691 	va_end(ap);
692 	if (rv < 0) {
693 		*buf = NULL;
694 		*len = 0;
695 		return NODATA;
696 	}
697 	*buf = RCAST(unsigned char *, msg);
698 	*len = strlen(msg);
699 	return ERRDATA;
700 }
701 
702 static void
703 closefd(int *fd, size_t i)
704 {
705 	if (fd[i] == -1)
706 		return;
707 	(void) close(fd[i]);
708 	fd[i] = -1;
709 }
710 
711 static void
712 closep(int *fd)
713 {
714 	size_t i;
715 	for (i = 0; i < 2; i++)
716 		closefd(fd, i);
717 }
718 
719 static int
720 copydesc(int i, int fd)
721 {
722 	if (fd == i)
723 		return 0; /* "no dup was necessary" */
724 	if (dup2(fd, i) == -1) {
725 		DPRINTF("dup(%d, %d) failed (%s)\n", fd, i, strerror(errno));
726 		exit(1);
727 	}
728 	return 1;
729 }
730 
731 static pid_t
732 writechild(int fd, const void *old, size_t n)
733 {
734 	pid_t pid;
735 
736 	/*
737 	 * fork again, to avoid blocking because both
738 	 * pipes filled
739 	 */
740 	pid = fork();
741 	if (pid == -1) {
742 		DPRINTF("Fork failed (%s)\n", strerror(errno));
743 		exit(1);
744 	}
745 	if (pid == 0) {
746 		/* child */
747 		if (swrite(fd, old, n) != CAST(ssize_t, n)) {
748 			DPRINTF("Write failed (%s)\n", strerror(errno));
749 			exit(1);
750 		}
751 		exit(0);
752 	}
753 	/* parent */
754 	return pid;
755 }
756 
757 static ssize_t
758 filter_error(unsigned char *ubuf, ssize_t n)
759 {
760 	char *p;
761 	char *buf;
762 
763 	ubuf[n] = '\0';
764 	buf = RCAST(char *, ubuf);
765 	while (isspace(CAST(unsigned char, *buf)))
766 		buf++;
767 	DPRINTF("Filter error[[[%s]]]\n", buf);
768 	if ((p = strchr(CAST(char *, buf), '\n')) != NULL)
769 		*p = '\0';
770 	if ((p = strchr(CAST(char *, buf), ';')) != NULL)
771 		*p = '\0';
772 	if ((p = strrchr(CAST(char *, buf), ':')) != NULL) {
773 		++p;
774 		while (isspace(CAST(unsigned char, *p)))
775 			p++;
776 		n = strlen(p);
777 		memmove(ubuf, p, CAST(size_t, n + 1));
778 	}
779 	DPRINTF("Filter error after[[[%s]]]\n", (char *)ubuf);
780 	if (islower(*ubuf))
781 		*ubuf = toupper(*ubuf);
782 	return n;
783 }
784 
785 private const char *
786 methodname(size_t method)
787 {
788 	switch (method) {
789 #ifdef BUILTIN_DECOMPRESS
790 	case METH_FROZEN:
791 	case METH_ZLIB:
792 		return "zlib";
793 #endif
794 #ifdef BUILTIN_BZLIB
795 	case METH_BZIP:
796 		return "bzlib";
797 #endif
798 #ifdef BUILTIN_XZLIB
799 	case METH_XZ:
800 	case METH_LZMA:
801 		return "xzlib";
802 #endif
803 	default:
804 		return compr[method].argv[0];
805 	}
806 }
807 
808 private int
809 uncompressbuf(int fd, size_t bytes_max, size_t method, const unsigned char *old,
810     unsigned char **newch, size_t* n)
811 {
812 	int fdp[3][2];
813 	int status, rv, w;
814 	pid_t pid;
815 	pid_t writepid = -1;
816 	size_t i;
817 	ssize_t r;
818 
819 	switch (method) {
820 #ifdef BUILTIN_DECOMPRESS
821 	case METH_FROZEN:
822 		return uncompressgzipped(old, newch, bytes_max, n);
823 	case METH_ZLIB:
824 		return uncompresszlib(old, newch, bytes_max, n, 1);
825 #endif
826 #ifdef BUILTIN_BZLIB
827 	case METH_BZIP:
828 		return uncompressbzlib(old, newch, bytes_max, n);
829 #endif
830 #ifdef BUILTIN_XZLIB
831 	case METH_XZ:
832 	case METH_LZMA:
833 		return uncompressxzlib(old, newch, bytes_max, n);
834 #endif
835 	default:
836 		break;
837 	}
838 
839 	(void)fflush(stdout);
840 	(void)fflush(stderr);
841 
842 	for (i = 0; i < __arraycount(fdp); i++)
843 		fdp[i][0] = fdp[i][1] = -1;
844 
845 	if ((fd == -1 && pipe(fdp[STDIN_FILENO]) == -1) ||
846 	    pipe(fdp[STDOUT_FILENO]) == -1 || pipe(fdp[STDERR_FILENO]) == -1) {
847 		closep(fdp[STDIN_FILENO]);
848 		closep(fdp[STDOUT_FILENO]);
849 		return makeerror(newch, n, "Cannot create pipe, %s",
850 		    strerror(errno));
851 	}
852 
853 	/* For processes with large mapped virtual sizes, vfork
854 	 * may be _much_ faster (10-100 times) than fork.
855 	 */
856 	pid = vfork();
857 	if (pid == -1) {
858 		return makeerror(newch, n, "Cannot vfork, %s",
859 		    strerror(errno));
860 	}
861 	if (pid == 0) {
862 		/* child */
863 		/* Note: we are after vfork, do not modify memory
864 		 * in a way which confuses parent. In particular,
865 		 * do not modify fdp[i][j].
866 		 */
867 		if (fd != -1) {
868 			(void) lseek(fd, CAST(off_t, 0), SEEK_SET);
869 			if (copydesc(STDIN_FILENO, fd))
870 				(void) close(fd);
871 		} else {
872 			if (copydesc(STDIN_FILENO, fdp[STDIN_FILENO][0]))
873 				(void) close(fdp[STDIN_FILENO][0]);
874 			if (fdp[STDIN_FILENO][1] > 2)
875 				(void) close(fdp[STDIN_FILENO][1]);
876 		}
877 ///FIXME: if one of the fdp[i][j] is 0 or 1, this can bomb spectacularly
878 		if (copydesc(STDOUT_FILENO, fdp[STDOUT_FILENO][1]))
879 			(void) close(fdp[STDOUT_FILENO][1]);
880 		if (fdp[STDOUT_FILENO][0] > 2)
881 			(void) close(fdp[STDOUT_FILENO][0]);
882 
883 		if (copydesc(STDERR_FILENO, fdp[STDERR_FILENO][1]))
884 			(void) close(fdp[STDERR_FILENO][1]);
885 		if (fdp[STDERR_FILENO][0] > 2)
886 			(void) close(fdp[STDERR_FILENO][0]);
887 
888 		(void)execvp(compr[method].argv[0],
889 		    RCAST(char *const *, RCAST(intptr_t, compr[method].argv)));
890 		dprintf(STDERR_FILENO, "exec `%s' failed, %s",
891 		    compr[method].argv[0], strerror(errno));
892 		_exit(1); /* _exit(), not exit(), because of vfork */
893 	}
894 	/* parent */
895 	/* Close write sides of child stdout/err pipes */
896 	for (i = 1; i < __arraycount(fdp); i++)
897 		closefd(fdp[i], 1);
898 	/* Write the buffer data to child stdin, if we don't have fd */
899 	if (fd == -1) {
900 		closefd(fdp[STDIN_FILENO], 0);
901 		writepid = writechild(fdp[STDIN_FILENO][1], old, *n);
902 		closefd(fdp[STDIN_FILENO], 1);
903 	}
904 
905 	*newch = CAST(unsigned char *, malloc(bytes_max + 1));
906 	if (*newch == NULL) {
907 		rv = makeerror(newch, n, "No buffer, %s",
908 		    strerror(errno));
909 		goto err;
910 	}
911 	rv = OKDATA;
912 	r = sread(fdp[STDOUT_FILENO][0], *newch, bytes_max, 0);
913 	if (r <= 0) {
914 		DPRINTF("Read stdout failed %d (%s)\n", fdp[STDOUT_FILENO][0],
915 		    r != -1 ? strerror(errno) : "no data");
916 
917 		rv = ERRDATA;
918 		if (r == 0 &&
919 		    (r = sread(fdp[STDERR_FILENO][0], *newch, bytes_max, 0)) > 0)
920 		{
921 			r = filter_error(*newch, r);
922 			goto ok;
923 		}
924 		free(*newch);
925 		if  (r == 0)
926 			rv = makeerror(newch, n, "Read failed, %s",
927 			    strerror(errno));
928 		else
929 			rv = makeerror(newch, n, "No data");
930 		goto err;
931 	}
932 ok:
933 	*n = r;
934 	/* NUL terminate, as every buffer is handled here. */
935 	(*newch)[*n] = '\0';
936 err:
937 	closefd(fdp[STDIN_FILENO], 1);
938 	closefd(fdp[STDOUT_FILENO], 0);
939 	closefd(fdp[STDERR_FILENO], 0);
940 
941 	w = waitpid(pid, &status, 0);
942 wait_err:
943 	if (w == -1) {
944 		free(*newch);
945 		rv = makeerror(newch, n, "Wait failed, %s", strerror(errno));
946 		DPRINTF("Child wait return %#x\n", status);
947 	} else if (!WIFEXITED(status)) {
948 		DPRINTF("Child not exited (%#x)\n", status);
949 	} else if (WEXITSTATUS(status) != 0) {
950 		DPRINTF("Child exited (%#x)\n", WEXITSTATUS(status));
951 	}
952 	if (writepid > 0) {
953 		/* _After_ we know decompressor has exited, our input writer
954 		 * definitely will exit now (at worst, writing fails in it,
955 		 * since output fd is closed now on the reading size).
956 		 */
957 		w = waitpid(writepid, &status, 0);
958 		writepid = -1;
959 		goto wait_err;
960 	}
961 
962 	closefd(fdp[STDIN_FILENO], 0); //why? it is already closed here!
963 	DPRINTF("Returning %p n=%" SIZE_T_FORMAT "u rv=%d\n", *newch, *n, rv);
964 
965 	return rv;
966 }
967 #endif
968