aboutsummaryrefslogtreecommitdiff
path: root/src/xz/file_io.c
blob: ca42bed98f2c7c0ed6baa8a7d77a0683121abefe (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
///////////////////////////////////////////////////////////////////////////////
//
/// \file       file_io.c
/// \brief      File opening, unlinking, and closing
//
//  Author:     Lasse Collin
//
//  This file has been put into the public domain.
//  You can do whatever you want with this file.
//
///////////////////////////////////////////////////////////////////////////////

#include "private.h"

#include <fcntl.h>

#ifdef TUKLIB_DOSLIKE
#	include <io.h>
#else
static bool warn_fchown;
#endif

#if defined(HAVE_FUTIMES) || defined(HAVE_FUTIMESAT) || defined(HAVE_UTIMES)
#	include <sys/time.h>
#elif defined(HAVE_UTIME)
#	include <utime.h>
#endif

#include "tuklib_open_stdxxx.h"

#ifndef O_BINARY
#	define O_BINARY 0
#endif

#ifndef O_NOCTTY
#	define O_NOCTTY 0
#endif


/// If true, try to create sparse files when decompressing.
static bool try_sparse = true;

/// File status flags of standard output. This is used by io_open_dest()
/// and io_close_dest().
static int stdout_flags = 0;


static bool io_write_buf(file_pair *pair, const uint8_t *buf, size_t size);


extern void
io_init(void)
{
	// Make sure that stdin, stdout, and and stderr are connected to
	// a valid file descriptor. Exit immediatelly with exit code ERROR
	// if we cannot make the file descriptors valid. Maybe we should
	// print an error message, but our stderr could be screwed anyway.
	tuklib_open_stdxxx(E_ERROR);

#ifndef TUKLIB_DOSLIKE
	// If fchown() fails setting the owner, we warn about it only if
	// we are root.
	warn_fchown = geteuid() == 0;
#endif

#ifdef __DJGPP__
	// Avoid doing useless things when statting files.
	// This isn't important but doesn't hurt.
	_djstat_flags = _STAT_INODE | _STAT_EXEC_EXT
			| _STAT_EXEC_MAGIC | _STAT_DIRSIZE;
#endif

	return;
}


extern void
io_no_sparse(void)
{
	try_sparse = false;
	return;
}


/// \brief      Unlink a file
///
/// This tries to verify that the file being unlinked really is the file that
/// we want to unlink by verifying device and inode numbers. There's still
/// a small unavoidable race, but this is much better than nothing (the file
/// could have been moved/replaced even hours earlier).
static void
io_unlink(const char *name, const struct stat *known_st)
{
#if defined(TUKLIB_DOSLIKE)
	// On DOS-like systems, st_ino is meaningless, so don't bother
	// testing it. Just silence a compiler warning.
	(void)known_st;
#else
	struct stat new_st;

	if (lstat(name, &new_st)
#	ifdef __VMS
			// st_ino is an array, and we don't want to
			// compare st_dev at all.
			|| memcmp(&new_st.st_ino, &known_st->st_ino,
				sizeof(new_st.st_ino)) != 0
#	else
			// Typical POSIX-like system
			|| new_st.st_dev != known_st->st_dev
			|| new_st.st_ino != known_st->st_ino
#	endif
			)
		message_error(_("%s: File seems to be moved, not removing"),
				name);
	else
#endif
		// There's a race condition between lstat() and unlink()
		// but at least we have tried to avoid removing wrong file.
		if (unlink(name))
			message_error(_("%s: Cannot remove: %s"),
					name, strerror(errno));

	return;
}


/// \brief      Copies owner/group and permissions
///
/// \todo       ACL and EA support
///
static void
io_copy_attrs(const file_pair *pair)
{
	// Skip chown and chmod on Windows.
#ifndef TUKLIB_DOSLIKE
	// This function is more tricky than you may think at first.
	// Blindly copying permissions may permit users to access the
	// destination file who didn't have permission to access the
	// source file.

	// Try changing the owner of the file. If we aren't root or the owner
	// isn't already us, fchown() probably doesn't succeed. We warn
	// about failing fchown() only if we are root.
	if (fchown(pair->dest_fd, pair->src_st.st_uid, -1) && warn_fchown)
		message_warning(_("%s: Cannot set the file owner: %s"),
				pair->dest_name, strerror(errno));

	mode_t mode;

	if (fchown(pair->dest_fd, -1, pair->src_st.st_gid)) {
		message_warning(_("%s: Cannot set the file group: %s"),
				pair->dest_name, strerror(errno));
		// We can still safely copy some additional permissions:
		// `group' must be at least as strict as `other' and
		// also vice versa.
		//
		// NOTE: After this, the owner of the source file may
		// get additional permissions. This shouldn't be too bad,
		// because the owner would have had permission to chmod
		// the original file anyway.
		mode = ((pair->src_st.st_mode & 0070) >> 3)
				& (pair->src_st.st_mode & 0007);
		mode = (pair->src_st.st_mode & 0700) | (mode << 3) | mode;
	} else {
		// Drop the setuid, setgid, and sticky bits.
		mode = pair->src_st.st_mode & 0777;
	}

	if (fchmod(pair->dest_fd, mode))
		message_warning(_("%s: Cannot set the file permissions: %s"),
				pair->dest_name, strerror(errno));
#endif

	// Copy the timestamps. We have several possible ways to do this, of
	// which some are better in both security and precision.
	//
	// First, get the nanosecond part of the timestamps. As of writing,
	// it's not standardized by POSIX, and there are several names for
	// the same thing in struct stat.
	long atime_nsec;
	long mtime_nsec;

#	if defined(HAVE_STRUCT_STAT_ST_ATIM_TV_NSEC)
	// GNU and Solaris
	atime_nsec = pair->src_st.st_atim.tv_nsec;
	mtime_nsec = pair->src_st.st_mtim.tv_nsec;

#	elif defined(HAVE_STRUCT_STAT_ST_ATIMESPEC_TV_NSEC)
	// BSD
	atime_nsec = pair->src_st.st_atimespec.tv_nsec;
	mtime_nsec = pair->src_st.st_mtimespec.tv_nsec;

#	elif defined(HAVE_STRUCT_STAT_ST_ATIMENSEC)
	// GNU and BSD without extensions
	atime_nsec = pair->src_st.st_atimensec;
	mtime_nsec = pair->src_st.st_mtimensec;

#	elif defined(HAVE_STRUCT_STAT_ST_UATIME)
	// Tru64
	atime_nsec = pair->src_st.st_uatime * 1000;
	mtime_nsec = pair->src_st.st_umtime * 1000;

#	elif defined(HAVE_STRUCT_STAT_ST_ATIM_ST__TIM_TV_NSEC)
	// UnixWare
	atime_nsec = pair->src_st.st_atim.st__tim.tv_nsec;
	mtime_nsec = pair->src_st.st_mtim.st__tim.tv_nsec;

#	else
	// Safe fallback
	atime_nsec = 0;
	mtime_nsec = 0;
#	endif

	// Construct a structure to hold the timestamps and call appropriate
	// function to set the timestamps.
#if defined(HAVE_FUTIMENS)
	// Use nanosecond precision.
	struct timespec tv[2];
	tv[0].tv_sec = pair->src_st.st_atime;
	tv[0].tv_nsec = atime_nsec;
	tv[1].tv_sec = pair->src_st.st_mtime;
	tv[1].tv_nsec = mtime_nsec;

	(void)futimens(pair->dest_fd, tv);

#elif defined(HAVE_FUTIMES) || defined(HAVE_FUTIMESAT) || defined(HAVE_UTIMES)
	// Use microsecond precision.
	struct timeval tv[2];
	tv[0].tv_sec = pair->src_st.st_atime;
	tv[0].tv_usec = atime_nsec / 1000;
	tv[1].tv_sec = pair->src_st.st_mtime;
	tv[1].tv_usec = mtime_nsec / 1000;

#	if defined(HAVE_FUTIMES)
	(void)futimes(pair->dest_fd, tv);
#	elif defined(HAVE_FUTIMESAT)
	(void)futimesat(pair->dest_fd, NULL, tv);
#	else
	// Argh, no function to use a file descriptor to set the timestamp.
	(void)utimes(pair->dest_name, tv);
#	endif

#elif defined(HAVE_UTIME)
	// Use one-second precision. utime() doesn't support using file
	// descriptor either. Some systems have broken utime() prototype
	// so don't make this const.
	struct utimbuf buf = {
		.actime = pair->src_st.st_atime,
		.modtime = pair->src_st.st_mtime,
	};

	// Avoid warnings.
	(void)atime_nsec;
	(void)mtime_nsec;

	(void)utime(pair->dest_name, &buf);
#endif

	return;
}


/// Opens the source file. Returns false on success, true on error.
static bool
io_open_src(file_pair *pair)
{
	// There's nothing to open when reading from stdin.
	if (pair->src_name == stdin_filename) {
		pair->src_fd = STDIN_FILENO;
#ifdef TUKLIB_DOSLIKE
		setmode(STDIN_FILENO, O_BINARY);
#endif
		return false;
	}

	// We accept only regular files if we are writing the output
	// to disk too, and if --force was not given.
	const bool reg_files_only = !opt_stdout && !opt_force;

	// Flags for open()
	int flags = O_RDONLY | O_BINARY | O_NOCTTY;

#ifndef TUKLIB_DOSLIKE
	// If we accept only regular files, we need to be careful to avoid
	// problems with special files like devices and FIFOs. O_NONBLOCK
	// prevents blocking when opening such files. When we want to accept
	// special files, we must not use O_NONBLOCK, or otherwise we won't
	// block waiting e.g. FIFOs to become readable.
	if (reg_files_only)
		flags |= O_NONBLOCK;
#endif

#if defined(O_NOFOLLOW)
	if (reg_files_only)
		flags |= O_NOFOLLOW;
#elif !defined(TUKLIB_DOSLIKE)
	// Some POSIX-like systems lack O_NOFOLLOW (it's not required
	// by POSIX). Check for symlinks with a separate lstat() on
	// these systems.
	if (reg_files_only) {
		struct stat st;
		if (lstat(pair->src_name, &st)) {
			message_error("%s: %s", pair->src_name,
					strerror(errno));
			return true;

		} else if (S_ISLNK(st.st_mode)) {
			message_warning(_("%s: Is a symbolic link, "
					"skipping"), pair->src_name);
			return true;
		}
	}
#endif

	// Try to open the file. If we are accepting non-regular files,
	// unblock the caught signals so that open() can be interrupted
	// if it blocks e.g. due to a FIFO file.
	if (!reg_files_only)
		signals_unblock();

	// Maybe this wouldn't need a loop, since all the signal handlers for
	// which we don't use SA_RESTART set user_abort to true. But it
	// doesn't hurt to have it just in case.
	do {
		pair->src_fd = open(pair->src_name, flags);
	} while (pair->src_fd == -1 && errno == EINTR && !user_abort);

	if (!reg_files_only)
		signals_block();

	if (pair->src_fd == -1) {
		// If we were interrupted, don't display any error message.
		if (errno == EINTR) {
			// All the signals that don't have SA_RESTART
			// set user_abort.
			assert(user_abort);
			return true;
		}

#ifdef O_NOFOLLOW
		// Give an understandable error message in if reason
		// for failing was that the file was a symbolic link.
		//
		// Note that at least Linux, OpenBSD, Solaris, and Darwin
		// use ELOOP to indicate if O_NOFOLLOW was the reason
		// that open() failed. Because there may be
		// directories in the pathname, ELOOP may occur also
		// because of a symlink loop in the directory part.
		// So ELOOP doesn't tell us what actually went wrong.
		//
		// FreeBSD associates EMLINK with O_NOFOLLOW and
		// Tru64 uses ENOTSUP. We use these directly here
		// and skip the lstat() call and the associated race.
		// I want to hear if there are other kernels that
		// fail with something else than ELOOP with O_NOFOLLOW.
		bool was_symlink = false;

#	if defined(__FreeBSD__) || defined(__DragonFly__)
		if (errno == EMLINK)
			was_symlink = true;

#	elif defined(__digital__) && defined(__unix__)
		if (errno == ENOTSUP)
			was_symlink = true;

#	elif defined(__NetBSD__)
		// FIXME? As of 2008-11-20, NetBSD doesn't document what
		// errno is used with O_NOFOLLOW. It seems to be EFTYPE,
		// but since it isn't documented, it may be wrong to rely
		// on it here.
		if (errno == EFTYPE)
			was_symlink = true;

#	else
		if (errno == ELOOP && reg_files_only) {
			const int saved_errno = errno;
			struct stat st;
			if (lstat(pair->src_name, &st) == 0
					&& S_ISLNK(st.st_mode))
				was_symlink = true;

			errno = saved_errno;
		}
#	endif

		if (was_symlink)
			message_warning(_("%s: Is a symbolic link, "
					"skipping"), pair->src_name);
		else
#endif
			// Something else than O_NOFOLLOW failing
			// (assuming that the race conditions didn't
			// confuse us).
			message_error("%s: %s", pair->src_name,
					strerror(errno));

		return true;
	}

#ifndef TUKLIB_DOSLIKE
	// Drop O_NONBLOCK, which is used only when we are accepting only
	// regular files. After the open() call, we want things to block
	// instead of giving EAGAIN.
	if (reg_files_only) {
		flags = fcntl(pair->src_fd, F_GETFL);
		if (flags == -1)
			goto error_msg;

		flags &= ~O_NONBLOCK;

		if (fcntl(pair->src_fd, F_SETFL, flags))
			goto error_msg;
	}
#endif

	// Stat the source file. We need the result also when we copy
	// the permissions, and when unlinking.
	if (fstat(pair->src_fd, &pair->src_st))
		goto error_msg;

	if (S_ISDIR(pair->src_st.st_mode)) {
		message_warning(_("%s: Is a directory, skipping"),
				pair->src_name);
		goto error;
	}

	if (reg_files_only) {
		if (!S_ISREG(pair->src_st.st_mode)) {
			message_warning(_("%s: Not a regular file, "
					"skipping"), pair->src_name);
			goto error;
		}

		// These are meaningless on Windows.
#ifndef TUKLIB_DOSLIKE
		if (pair->src_st.st_mode & (S_ISUID | S_ISGID)) {
			// gzip rejects setuid and setgid files even
			// when --force was used. bzip2 doesn't check
			// for them, but calls fchown() after fchmod(),
			// and many systems automatically drop setuid
			// and setgid bits there.
			//
			// We accept setuid and setgid files if
			// --force was used. We drop these bits
			// explicitly in io_copy_attr().
			message_warning(_("%s: File has setuid or "
					"setgid bit set, skipping"),
					pair->src_name);
			goto error;
		}

		if (pair->src_st.st_mode & S_ISVTX) {
			message_warning(_("%s: File has sticky bit "
					"set, skipping"),
					pair->src_name);
			goto error;
		}

		if (pair->src_st.st_nlink > 1) {
			message_warning(_("%s: Input file has more "
					"than one hard link, "
					"skipping"), pair->src_name);
			goto error;
		}
#endif
	}

	return false;

error_msg:
	message_error("%s: %s", pair->src_name, strerror(errno));
error:
	(void)close(pair->src_fd);
	return true;
}


/// \brief      Closes source file of the file_pair structure
///
/// \param      pair    File whose src_fd should be closed
/// \param      success If true, the file will be removed from the disk if
///                     closing succeeds and --keep hasn't been used.
static void
io_close_src(file_pair *pair, bool success)
{
	if (pair->src_fd != STDIN_FILENO && pair->src_fd != -1) {
#ifdef TUKLIB_DOSLIKE
		(void)close(pair->src_fd);
#endif

		// If we are going to unlink(), do it before closing the file.
		// This way there's no risk that someone replaces the file and
		// happens to get same inode number, which would make us
		// unlink() wrong file.
		//
		// NOTE: DOS-like systems are an exception to this, because
		// they don't allow unlinking files that are open. *sigh*
		if (success && !opt_keep_original)
			io_unlink(pair->src_name, &pair->src_st);

#ifndef TUKLIB_DOSLIKE
		(void)close(pair->src_fd);
#endif
	}

	return;
}


static bool
io_open_dest(file_pair *pair)
{
	if (opt_stdout || pair->src_fd == STDIN_FILENO) {
		// We don't modify or free() this.
		pair->dest_name = (char *)"(stdout)";
		pair->dest_fd = STDOUT_FILENO;
#ifdef TUKLIB_DOSLIKE
		setmode(STDOUT_FILENO, O_BINARY);
#endif
	} else {
		pair->dest_name = suffix_get_dest_name(pair->src_name);
		if (pair->dest_name == NULL)
			return true;

		// If --force was used, unlink the target file first.
		if (opt_force && unlink(pair->dest_name) && errno != ENOENT) {
			message_error(_("%s: Cannot remove: %s"),
					pair->dest_name, strerror(errno));
			free(pair->dest_name);
			return true;
		}

		// Open the file.
		const int flags = O_WRONLY | O_BINARY | O_NOCTTY
				| O_CREAT | O_EXCL;
		const mode_t mode = S_IRUSR | S_IWUSR;
		pair->dest_fd = open(pair->dest_name, flags, mode);

		if (pair->dest_fd == -1) {
			// Don't bother with error message if user requested
			// us to exit anyway.
			if (!user_abort)
				message_error("%s: %s", pair->dest_name,
						strerror(errno));

			free(pair->dest_name);
			return true;
		}
	}

	// If this really fails... well, we have a safe fallback.
	if (fstat(pair->dest_fd, &pair->dest_st)) {
#if defined(__VMS)
		pair->dest_st.st_ino[0] = 0;
		pair->dest_st.st_ino[1] = 0;
		pair->dest_st.st_ino[2] = 0;
#elif !defined(TUKLIB_DOSLIKE)
		pair->dest_st.st_dev = 0;
		pair->dest_st.st_ino = 0;
#endif
#ifndef TUKLIB_DOSLIKE
	} else if (try_sparse && opt_mode == MODE_DECOMPRESS) {
		// When writing to standard output, we need to be extra
		// careful:
		//  - It may be connected to something else than
		//    a regular file.
		//  - We aren't necessarily writing to a new empty file
		//    or to the end of an existing file.
		//  - O_APPEND may be active.
		//
		// TODO: I'm keeping this disabled for DOS-like systems
		// for now. FAT doesn't support sparse files, but NTFS
		// does, so maybe this should be enabled on Windows after
		// some testing.
		if (pair->dest_fd == STDOUT_FILENO) {
			if (!S_ISREG(pair->dest_st.st_mode))
				return false;

			const int flags = fcntl(STDOUT_FILENO, F_GETFL);
			if (flags == -1)
				return false;

			if (flags & O_APPEND) {
				// Creating a sparse file is not possible
				// when O_APPEND is active (it's used by
				// shell's >> redirection). As I understand
				// it, it is safe to temporarily disable
				// O_APPEND in xz, because if someone
				// happened to write to the same file at the
				// same time, results would be bad anyway
				// (users shouldn't assume that xz uses any
				// specific block size when writing data).
				//
				// The write position may be something else
				// than the end of the file, so we must fix
				// it to start writing at the end of the file
				// to imitate O_APPEND.
				if (lseek(STDOUT_FILENO, 0, SEEK_END) == -1)
					return false;

				if (fcntl(STDOUT_FILENO, F_SETFL,
						stdout_flags & ~O_APPEND))
					return false;

				// Remember the flags so that io_close_dest()
				// can restore them.
				stdout_flags = flags;

			} else if (lseek(STDOUT_FILENO, 0, SEEK_CUR)
					!= pair->dest_st.st_size) {
				// Writing won't start exactly at the end
				// of the file. We cannot use sparse output,
				// because it would probably corrupt the file.
				return false;
			}
		}

		pair->dest_try_sparse = true;
#endif
	}

	return false;
}


/// \brief      Closes destination file of the file_pair structure
///
/// \param      pair    File whose dest_fd should be closed
/// \param      success If false, the file will be removed from the disk.
///
/// \return     Zero if closing succeeds. On error, -1 is returned and
///             error message printed.
static int
io_close_dest(file_pair *pair, bool success)
{
	// If io_open_dest() has disabled O_APPEND, restore it here.
	if (stdout_flags != 0) {
		assert(pair->dest_fd == STDOUT_FILENO);

		const int fail = fcntl(STDOUT_FILENO, F_SETFL, stdout_flags);
		stdout_flags = 0;

		if (fail) {
			message_error(_("Error restoring the O_APPEND flag "
					"to standard output: %s"),
					strerror(errno));
			return -1;
		}
	}

	if (pair->dest_fd == -1 || pair->dest_fd == STDOUT_FILENO)
		return 0;

	if (close(pair->dest_fd)) {
		message_error(_("%s: Closing the file failed: %s"),
				pair->dest_name, strerror(errno));

		// Closing destination file failed, so we cannot trust its
		// contents. Get rid of junk:
		io_unlink(pair->dest_name, &pair->dest_st);
		free(pair->dest_name);
		return -1;
	}

	// If the operation using this file wasn't successful, we git rid
	// of the junk file.
	if (!success)
		io_unlink(pair->dest_name, &pair->dest_st);

	free(pair->dest_name);

	return 0;
}


extern file_pair *
io_open(const char *src_name)
{
	if (is_empty_filename(src_name))
		return NULL;

	// Since we have only one file open at a time, we can use
	// a statically allocated structure.
	static file_pair pair;

	pair = (file_pair){
		.src_name = src_name,
		.dest_name = NULL,
		.src_fd = -1,
		.dest_fd = -1,
		.src_eof = false,
		.dest_try_sparse = false,
		.dest_pending_sparse = 0,
	};

	// Block the signals, for which we have a custom signal handler, so
	// that we don't need to worry about EINTR.
	signals_block();

	file_pair *ret = NULL;
	if (!io_open_src(&pair)) {
		// io_open_src() may have unblocked the signals temporarily,
		// and thus user_abort may have got set even if open()
		// succeeded.
		if (user_abort || io_open_dest(&pair))
			io_close_src(&pair, false);
		else
			ret = &pair;
	}

	signals_unblock();

	return ret;
}


extern void
io_close(file_pair *pair, bool success)
{
	// Take care of sparseness at the end of the output file.
	if (success && pair->dest_try_sparse
			&& pair->dest_pending_sparse > 0) {
		// Seek forward one byte less than the size of the pending
		// hole, then write one zero-byte. This way the file grows
		// to its correct size. An alternative would be to use
		// ftruncate() but that isn't portable enough (e.g. it
		// doesn't work with FAT on Linux; FAT isn't that important
		// since it doesn't support sparse files anyway, but we don't
		// want to create corrupt files on it).
		if (lseek(pair->dest_fd, pair->dest_pending_sparse - 1,
				SEEK_CUR) == -1) {
			message_error(_("%s: Seeking failed when trying "
					"to create a sparse file: %s"),
					pair->dest_name, strerror(errno));
			success = false;
		} else {
			const uint8_t zero[1] = { '\0' };
			if (io_write_buf(pair, zero, 1))
				success = false;
		}
	}

	signals_block();

	if (success && pair->dest_fd != STDOUT_FILENO)
		io_copy_attrs(pair);

	// Close the destination first. If it fails, we must not remove
	// the source file!
	if (io_close_dest(pair, success))
		success = false;

	// Close the source file, and unlink it if the operation using this
	// file pair was successful and we haven't requested to keep the
	// source file.
	io_close_src(pair, success);

	signals_unblock();

	return;
}


extern size_t
io_read(file_pair *pair, io_buf *buf_union, size_t size)
{
	// We use small buffers here.
	assert(size < SSIZE_MAX);

	uint8_t *buf = buf_union->u8;
	size_t left = size;

	while (left > 0) {
		const ssize_t amount = read(pair->src_fd, buf, left);

		if (amount == 0) {
			pair->src_eof = true;
			break;
		}

		if (amount == -1) {
			if (errno == EINTR) {
				if (user_abort)
					return SIZE_MAX;

				continue;
			}

			message_error(_("%s: Read error: %s"),
					pair->src_name, strerror(errno));

			// FIXME Is this needed?
			pair->src_eof = true;

			return SIZE_MAX;
		}

		buf += (size_t)(amount);
		left -= (size_t)(amount);
	}

	return size - left;
}


static bool
is_sparse(const io_buf *buf)
{
	assert(IO_BUFFER_SIZE % sizeof(uint64_t) == 0);

	for (size_t i = 0; i < ARRAY_SIZE(buf->u64); ++i)
		if (buf->u64[i] != 0)
			return false;

	return true;
}


static bool
io_write_buf(file_pair *pair, const uint8_t *buf, size_t size)
{
	assert(size < SSIZE_MAX);

	while (size > 0) {
		const ssize_t amount = write(pair->dest_fd, buf, size);
		if (amount == -1) {
			if (errno == EINTR) {
				if (user_abort)
					return -1;

				continue;
			}

			// Handle broken pipe specially. gzip and bzip2
			// don't print anything on SIGPIPE. In addition,
			// gzip --quiet uses exit status 2 (warning) on
			// broken pipe instead of whatever raise(SIGPIPE)
			// would make it return. It is there to hide "Broken
			// pipe" message on some old shells (probably old
			// GNU bash).
			//
			// We don't do anything special with --quiet, which
			// is what bzip2 does too. If we get SIGPIPE, we
			// will handle it like other signals by setting
			// user_abort, and get EPIPE here.
			if (errno != EPIPE)
				message_error(_("%s: Write error: %s"),
					pair->dest_name, strerror(errno));

			return true;
		}

		buf += (size_t)(amount);
		size -= (size_t)(amount);
	}

	return false;
}


extern bool
io_write(file_pair *pair, const io_buf *buf, size_t size)
{
	assert(size <= IO_BUFFER_SIZE);

	if (pair->dest_try_sparse) {
		// Check if the block is sparse (contains only zeros). If it
		// sparse, we just store the amount and return. We will take
		// care of actually skipping over the hole when we hit the
		// next data block or close the file.
		//
		// Since io_close() requires that dest_pending_sparse > 0
		// if the file ends with sparse block, we must also return
		// if size == 0 to avoid doing the lseek().
		if (size == IO_BUFFER_SIZE) {
			if (is_sparse(buf)) {
				pair->dest_pending_sparse += size;
				return false;
			}
		} else if (size == 0) {
			return false;
		}

		// This is not a sparse block. If we have a pending hole,
		// skip it now.
		if (pair->dest_pending_sparse > 0) {
			if (lseek(pair->dest_fd, pair->dest_pending_sparse,
					SEEK_CUR) == -1) {
				message_error(_("%s: Seeking failed when "
						"trying to create a sparse "
						"file: %s"), pair->dest_name,
						strerror(errno));
				return true;
			}

			pair->dest_pending_sparse = 0;
		}
	}

	return io_write_buf(pair, buf->u8, size);
}