1 /* split.c -- split a file into pieces.
2 Copyright (C) 1988-2024 Free Software Foundation, Inc.
4 This program is free software: you can redistribute it and/or modify
5 it under the terms of the GNU General Public License as published by
6 the Free Software Foundation, either version 3 of the License, or
7 (at your option) any later version.
9 This program is distributed in the hope that it will be useful,
10 but WITHOUT ANY WARRANTY; without even the implied warranty of
11 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 GNU General Public License for more details.
14 You should have received a copy of the GNU General Public License
15 along with this program. If not, see <https://www.gnu.org/licenses/>. */
17 /* By tege@sics.se, with rms.
20 * support -p REGEX as in BSD's split.
21 * support --suppress-matched as in csplit. */
28 #include <sys/types.h>
32 #include "alignalloc.h"
35 #include "fd-reopen.h"
37 #include "full-write.h"
38 #include "ioblksize.h"
41 #include "sys-limits.h"
42 #include "temp-stream.h"
43 #include "xbinary-io.h"
44 #include "xdectoint.h"
47 /* The official name of this program (e.g., no 'g' prefix). */
48 #define PROGRAM_NAME "split"
51 proper_name_lite ("Torbjorn Granlund", "Torbj\303\266rn Granlund"), \
52 proper_name ("Richard M. Stallman")
54 /* Shell command to filter through, instead of creating files. */
55 static char const *filter_command
;
57 /* Process ID of the filter. */
58 static pid_t filter_pid
;
60 /* Array of open pipes. */
61 static int *open_pipes
;
62 static idx_t open_pipes_alloc
;
63 static int n_open_pipes
;
65 /* Whether SIGPIPE has the default action, when --filter is used. */
66 static bool default_SIGPIPE
;
68 /* Base name of output files. */
69 static char const *outbase
;
71 /* Name of output files. */
74 /* Pointer to the end of the prefix in OUTFILE.
75 Suffixes are inserted here. */
76 static char *outfile_mid
;
78 /* Generate new suffix when suffixes are exhausted. */
79 static bool suffix_auto
= true;
81 /* Length of OUTFILE's suffix. */
82 static idx_t suffix_length
;
84 /* Alphabet of characters to use in suffix. */
85 static char const *suffix_alphabet
= "abcdefghijklmnopqrstuvwxyz";
87 /* Numerical suffix start value. */
88 static char const *numeric_suffix_start
;
90 /* Additional suffix to append to output file names. */
91 static char const *additional_suffix
;
93 /* Name of input file. May be "-". */
94 static char const *infile
;
96 /* stat buf for input file. */
97 static struct stat in_stat_buf
;
99 /* Descriptor on which output file is open. */
100 static int output_desc
= -1;
102 /* If true, print a diagnostic on standard error just before each
103 output file is opened. */
106 /* If true, don't generate zero length output files. */
107 static bool elide_empty_files
;
109 /* If true, in round robin mode, immediately copy
110 input to output, which is much slower, so disabled by default. */
111 static bool unbuffered
;
113 /* The character marking end of line. Defaults to \n below. */
114 static int eolchar
= -1;
116 /* The split mode to use. */
119 type_undef
, type_bytes
, type_byteslines
, type_lines
, type_digits
,
120 type_chunk_bytes
, type_chunk_lines
, type_rr
123 /* For long options that have no equivalent short option, use a
124 non-character as a pseudo short option, starting with CHAR_MAX + 1. */
127 VERBOSE_OPTION
= CHAR_MAX
+ 1,
130 ADDITIONAL_SUFFIX_OPTION
133 static struct option
const longopts
[] =
135 {"bytes", required_argument
, nullptr, 'b'},
136 {"lines", required_argument
, nullptr, 'l'},
137 {"line-bytes", required_argument
, nullptr, 'C'},
138 {"number", required_argument
, nullptr, 'n'},
139 {"elide-empty-files", no_argument
, nullptr, 'e'},
140 {"unbuffered", no_argument
, nullptr, 'u'},
141 {"suffix-length", required_argument
, nullptr, 'a'},
142 {"additional-suffix", required_argument
, nullptr,
143 ADDITIONAL_SUFFIX_OPTION
},
144 {"numeric-suffixes", optional_argument
, nullptr, 'd'},
145 {"hex-suffixes", optional_argument
, nullptr, 'x'},
146 {"filter", required_argument
, nullptr, FILTER_OPTION
},
147 {"verbose", no_argument
, nullptr, VERBOSE_OPTION
},
148 {"separator", required_argument
, nullptr, 't'},
149 {"-io-blksize", required_argument
, nullptr,
150 IO_BLKSIZE_OPTION
}, /* do not document */
151 {GETOPT_HELP_OPTION_DECL
},
152 {GETOPT_VERSION_OPTION_DECL
},
153 {nullptr, 0, nullptr, 0}
156 /* Return true if the errno value, ERR, is ignorable. */
160 return filter_command
&& err
== EPIPE
;
164 set_suffix_length (intmax_t n_units
, enum Split_type split_type
)
166 #define DEFAULT_SUFFIX_LENGTH 2
168 int suffix_length_needed
= 0;
170 /* The suffix auto length feature is incompatible with
171 a user specified start value as the generated suffixes
172 are not all consecutive. */
173 if (numeric_suffix_start
)
176 /* Auto-calculate the suffix length if the number of files is given. */
177 if (split_type
== type_chunk_bytes
|| split_type
== type_chunk_lines
178 || split_type
== type_rr
)
180 intmax_t n_units_end
= n_units
- 1;
181 if (numeric_suffix_start
)
184 strtol_error e
= xstrtoimax (numeric_suffix_start
, nullptr, 10,
186 if (e
== LONGINT_OK
&& n_start
< n_units
)
188 /* Restrict auto adjustment so we don't keep
189 incrementing a suffix size arbitrarily,
190 as that would break sort order for files
191 generated from multiple split runs. */
192 if (ckd_add (&n_units_end
, n_units_end
, n_start
))
193 n_units_end
= INTMAX_MAX
;
197 idx_t alphabet_len
= strlen (suffix_alphabet
);
199 suffix_length_needed
++;
200 while (n_units_end
/= alphabet_len
);
205 if (suffix_length
) /* set by user */
207 if (suffix_length
< suffix_length_needed
)
208 error (EXIT_FAILURE
, 0,
209 _("the suffix length needs to be at least %d"),
210 suffix_length_needed
);
215 suffix_length
= MAX (DEFAULT_SUFFIX_LENGTH
, suffix_length_needed
);
221 if (status
!= EXIT_SUCCESS
)
226 Usage: %s [OPTION]... [FILE [PREFIX]]\n\
230 Output pieces of FILE to PREFIXaa, PREFIXab, ...;\n\
231 default size is 1000 lines, and default PREFIX is 'x'.\n\
235 emit_mandatory_arg_note ();
237 fprintf (stdout
, _("\
238 -a, --suffix-length=N generate suffixes of length N (default %d)\n\
239 --additional-suffix=SUFFIX append an additional SUFFIX to file names\n\
240 -b, --bytes=SIZE put SIZE bytes per output file\n\
241 -C, --line-bytes=SIZE put at most SIZE bytes of records per output file\n\
242 -d use numeric suffixes starting at 0, not alphabetic\n\
243 --numeric-suffixes[=FROM] same as -d, but allow setting the start value\
245 -x use hex suffixes starting at 0, not alphabetic\n\
246 --hex-suffixes[=FROM] same as -x, but allow setting the start value\n\
247 -e, --elide-empty-files do not generate empty output files with '-n'\n\
248 --filter=COMMAND write to shell COMMAND; file name is $FILE\n\
249 -l, --lines=NUMBER put NUMBER lines/records per output file\n\
250 -n, --number=CHUNKS generate CHUNKS output files; see explanation below\n\
251 -t, --separator=SEP use SEP instead of newline as the record separator;\n\
252 '\\0' (zero) specifies the NUL character\n\
253 -u, --unbuffered immediately copy input to output with '-n r/...'\n\
254 "), DEFAULT_SUFFIX_LENGTH
);
256 --verbose print a diagnostic just before each\n\
257 output file is opened\n\
259 fputs (HELP_OPTION_DESCRIPTION
, stdout
);
260 fputs (VERSION_OPTION_DESCRIPTION
, stdout
);
264 N split into N files based on size of input\n\
265 K/N output Kth of N to stdout\n\
266 l/N split into N files without splitting lines/records\n\
267 l/K/N output Kth of N to stdout without splitting lines/records\n\
268 r/N like 'l' but use round robin distribution\n\
269 r/K/N likewise but only output Kth of N to stdout\n\
271 emit_ancillary_info (PROGRAM_NAME
);
276 /* Copy the data in FD to a temporary file, then make that file FD.
277 Use BUF, of size BUFSIZE, to copy. Return the number of
278 bytes copied, or -1 (setting errno) on error. */
280 copy_to_tmpfile (int fd
, char *buf
, idx_t bufsize
)
283 if (!temp_stream (&tmp
, nullptr))
288 while (0 < (r
= read (fd
, buf
, bufsize
)))
290 if (fwrite (buf
, 1, r
, tmp
) != r
)
292 if (ckd_add (&copied
, copied
, r
))
301 r
= dup2 (fileno (tmp
), fd
);
304 if (fclose (tmp
) < 0)
309 /* Return the number of bytes that can be read from FD with status ST.
310 Store up to the first BUFSIZE bytes of the file's data into BUF,
311 and advance the file position by the number of bytes read. On
312 input error, set errno and return -1. */
315 input_file_size (int fd
, struct stat
const *st
, char *buf
, idx_t bufsize
)
320 ssize_t n_read
= read (fd
, buf
+ size
, bufsize
- size
);
322 return n_read
< 0 ? n_read
: size
;
325 while (size
< bufsize
);
328 if ((usable_st_size (st
) && st
->st_size
< size
)
329 || (cur
= lseek (fd
, 0, SEEK_CUR
)) < 0
330 || cur
< size
/* E.g., /dev/zero on GNU/Linux. */
331 || (end
= lseek (fd
, 0, SEEK_END
)) < 0)
333 char *tmpbuf
= xmalloc (bufsize
);
334 end
= copy_to_tmpfile (fd
, tmpbuf
, bufsize
);
341 if (end
== OFF_T_MAX
/* E.g., /dev/zero on GNU/Hurd. */
342 || (cur
< end
&& ckd_add (&size
, size
, end
- cur
)))
350 off_t r
= lseek (fd
, cur
, SEEK_SET
);
358 /* Compute the next sequential output file name and store it into the
362 next_file_name (void)
364 /* Index in suffix_alphabet of each character in the suffix. */
365 static idx_t
*sufindex
;
366 static idx_t outbase_length
;
367 static idx_t outfile_length
;
368 static idx_t addsuf_length
;
372 bool overflow
, widen
;
375 widen
= !! outfile_length
;
379 /* Allocate and initialize the first file name. */
381 outbase_length
= strlen (outbase
);
382 addsuf_length
= additional_suffix
? strlen (additional_suffix
) : 0;
383 overflow
= ckd_add (&outfile_length
, outbase_length
+ addsuf_length
,
388 /* Reallocate and initialize a new wider file name.
389 We do this by subsuming the unchanging part of
390 the generated suffix into the prefix (base), and
391 reinitializing the now one longer suffix. */
393 overflow
= ckd_add (&outfile_length
, outfile_length
, 2);
398 overflow
|= ckd_add (&outfile_size
, outfile_length
, 1);
401 outfile
= xirealloc (outfile
, outfile_size
);
404 memcpy (outfile
, outbase
, outbase_length
);
407 /* Append the last alphabet character to the file name prefix. */
408 outfile
[outbase_length
] = suffix_alphabet
[sufindex
[0]];
412 outfile_mid
= outfile
+ outbase_length
;
413 memset (outfile_mid
, suffix_alphabet
[0], suffix_length
);
414 if (additional_suffix
)
415 memcpy (outfile_mid
+ suffix_length
, additional_suffix
, addsuf_length
);
416 outfile
[outfile_length
] = 0;
419 sufindex
= xicalloc (suffix_length
, sizeof *sufindex
);
421 if (numeric_suffix_start
)
425 /* Update the output file name. */
426 idx_t i
= strlen (numeric_suffix_start
);
427 memcpy (outfile_mid
+ suffix_length
- i
, numeric_suffix_start
, i
);
429 /* Update the suffix index. */
430 idx_t
*sufindex_end
= sufindex
+ suffix_length
;
432 *--sufindex_end
= numeric_suffix_start
[i
] - '0';
435 #if ! _POSIX_NO_TRUNC && HAVE_PATHCONF && defined _PC_NAME_MAX
436 /* POSIX requires that if the output file name is too long for
437 its directory, 'split' must fail without creating any files.
438 This must be checked for explicitly on operating systems that
439 silently truncate file names. */
441 char *dir
= dir_name (outfile
);
442 long name_max
= pathconf (dir
, _PC_NAME_MAX
);
443 if (0 <= name_max
&& name_max
< base_len (last_component (outfile
)))
444 error (EXIT_FAILURE
, ENAMETOOLONG
, "%s", quotef (outfile
));
451 /* Increment the suffix in place, if possible. */
453 idx_t i
= suffix_length
;
457 if (suffix_auto
&& i
== 0 && ! suffix_alphabet
[sufindex
[0] + 1])
459 outfile_mid
[i
] = suffix_alphabet
[sufindex
[i
]];
463 outfile_mid
[i
] = suffix_alphabet
[sufindex
[i
]];
465 error (EXIT_FAILURE
, 0, _("output file suffixes exhausted"));
469 /* Create or truncate a file. */
472 create (char const *name
)
477 fprintf (stdout
, _("creating file %s\n"), quoteaf (name
));
479 int oflags
= O_WRONLY
| O_CREAT
| O_BINARY
;
480 int fd
= open (name
, oflags
| O_EXCL
, MODE_RW_UGO
);
481 if (0 <= fd
|| errno
!= EEXIST
)
483 fd
= open (name
, oflags
, MODE_RW_UGO
);
486 struct stat out_stat_buf
;
487 if (fstat (fd
, &out_stat_buf
) != 0)
488 error (EXIT_FAILURE
, errno
, _("failed to stat %s"), quoteaf (name
));
489 if (psame_inode (&in_stat_buf
, &out_stat_buf
))
490 error (EXIT_FAILURE
, 0, _("%s would overwrite input; aborting"),
492 if (ftruncate (fd
, 0) < 0
493 && (S_ISREG (out_stat_buf
.st_mode
) || S_TYPEISSHM (&out_stat_buf
)))
494 error (EXIT_FAILURE
, errno
, _("%s: error truncating"), quotef (name
));
502 char const *shell_prog
= getenv ("SHELL");
503 if (shell_prog
== nullptr)
504 shell_prog
= "/bin/sh";
505 if (setenv ("FILE", name
, 1) != 0)
506 error (EXIT_FAILURE
, errno
,
507 _("failed to set FILE environment variable"));
509 fprintf (stdout
, _("executing with FILE=%s\n"), quotef (name
));
510 if (pipe (fd_pair
) != 0)
511 error (EXIT_FAILURE
, errno
, _("failed to create pipe"));
515 /* This is the child process. If an error occurs here, the
516 parent will eventually learn about it after doing a wait,
517 at which time it will emit its own error message. */
519 /* We have to close any pipes that were opened during an
520 earlier call, otherwise this process will be holding a
521 write-pipe that will prevent the earlier process from
522 reading an EOF on the corresponding read-pipe. */
523 for (j
= 0; j
< n_open_pipes
; ++j
)
524 if (close (open_pipes
[j
]) != 0)
525 error (EXIT_FAILURE
, errno
, _("closing prior pipe"));
526 if (close (fd_pair
[1]))
527 error (EXIT_FAILURE
, errno
, _("closing output pipe"));
528 if (fd_pair
[0] != STDIN_FILENO
)
530 if (dup2 (fd_pair
[0], STDIN_FILENO
) != STDIN_FILENO
)
531 error (EXIT_FAILURE
, errno
, _("moving input pipe"));
532 if (close (fd_pair
[0]) != 0)
533 error (EXIT_FAILURE
, errno
, _("closing input pipe"));
536 signal (SIGPIPE
, SIG_DFL
);
537 execl (shell_prog
, last_component (shell_prog
), "-c",
538 filter_command
, (char *) nullptr);
539 error (EXIT_FAILURE
, errno
, _("failed to run command: \"%s -c %s\""),
540 shell_prog
, filter_command
);
543 error (EXIT_FAILURE
, errno
, _("fork system call failed"));
544 if (close (fd_pair
[0]) != 0)
545 error (EXIT_FAILURE
, errno
, _("failed to close input pipe"));
546 filter_pid
= child_pid
;
547 if (n_open_pipes
== open_pipes_alloc
)
548 open_pipes
= xpalloc (open_pipes
, &open_pipes_alloc
, 1,
549 MIN (INT_MAX
, IDX_MAX
), sizeof *open_pipes
);
550 open_pipes
[n_open_pipes
++] = fd_pair
[1];
555 /* Close the output file, and do any associated cleanup.
556 If FP and FD are both specified, they refer to the same open file;
557 in this case FP is closed, but FD is still used in cleanup. */
559 closeout (FILE *fp
, int fd
, pid_t pid
, char const *name
)
561 if (fp
!= nullptr && fclose (fp
) != 0 && ! ignorable (errno
))
562 error (EXIT_FAILURE
, errno
, "%s", quotef (name
));
565 if (fp
== nullptr && close (fd
) < 0)
566 error (EXIT_FAILURE
, errno
, "%s", quotef (name
));
568 for (j
= 0; j
< n_open_pipes
; ++j
)
570 if (open_pipes
[j
] == fd
)
572 open_pipes
[j
] = open_pipes
[--n_open_pipes
];
580 if (waitpid (pid
, &wstatus
, 0) < 0)
581 error (EXIT_FAILURE
, errno
, _("waiting for child process"));
582 else if (WIFSIGNALED (wstatus
))
584 int sig
= WTERMSIG (wstatus
);
587 char signame
[MAX (SIG2STR_MAX
, INT_BUFSIZE_BOUND (int))];
588 if (sig2str (sig
, signame
) != 0)
589 sprintf (signame
, "%d", sig
);
591 _("with FILE=%s, signal %s from command: %s"),
592 quotef (name
), signame
, filter_command
);
595 else if (WIFEXITED (wstatus
))
597 int ex
= WEXITSTATUS (wstatus
);
599 error (ex
, 0, _("with FILE=%s, exit %d from command: %s"),
600 quotef (name
), ex
, filter_command
);
604 /* shouldn't happen. */
605 error (EXIT_FAILURE
, 0,
606 _("unknown status from command (0x%X)"), wstatus
+ 0u);
611 /* Write BYTES bytes at BP to an output file.
612 If NEW_FILE_FLAG is true, open the next output file.
613 Otherwise add to the same output file already in use.
614 Return true if successful. */
617 cwrite (bool new_file_flag
, char const *bp
, idx_t bytes
)
621 if (!bp
&& bytes
== 0 && elide_empty_files
)
623 closeout (nullptr, output_desc
, filter_pid
, outfile
);
625 output_desc
= create (outfile
);
627 error (EXIT_FAILURE
, errno
, "%s", quotef (outfile
));
630 if (full_write (output_desc
, bp
, bytes
) == bytes
)
634 if (! ignorable (errno
))
635 error (EXIT_FAILURE
, errno
, "%s", quotef (outfile
));
640 /* Split into pieces of exactly N_BYTES bytes.
641 However, the first REM_BYTES pieces should be 1 byte longer.
642 Use buffer BUF, whose size is BUFSIZE.
643 If INITIAL_READ is nonnegative,
644 BUF contains the first INITIAL_READ input bytes. */
647 bytes_split (intmax_t n_bytes
, intmax_t rem_bytes
,
648 char *buf
, idx_t bufsize
, ssize_t initial_read
,
651 bool new_file_flag
= true;
652 bool filter_ok
= true;
654 intmax_t to_write
= n_bytes
+ (0 < rem_bytes
);
655 bool eof
= ! to_write
;
660 if (0 <= initial_read
)
662 n_read
= initial_read
;
664 eof
= n_read
< bufsize
;
669 && 0 <= lseek (STDIN_FILENO
, to_write
, SEEK_CUR
))
671 to_write
= n_bytes
+ (opened
+ 1 < rem_bytes
);
672 new_file_flag
= true;
675 n_read
= read (STDIN_FILENO
, buf
, bufsize
);
677 error (EXIT_FAILURE
, errno
, "%s", quotef (infile
));
681 while (0 < to_write
&& to_write
<= n_read
)
683 if (filter_ok
|| new_file_flag
)
684 filter_ok
= cwrite (new_file_flag
, bp_out
, to_write
);
685 opened
+= new_file_flag
;
686 new_file_flag
= !max_files
|| (opened
< max_files
);
687 if (! filter_ok
&& ! new_file_flag
)
689 /* If filters no longer accepting input, stop reading. */
696 to_write
= n_bytes
+ (opened
< rem_bytes
);
700 if (filter_ok
|| new_file_flag
)
701 filter_ok
= cwrite (new_file_flag
, bp_out
, n_read
);
702 opened
+= new_file_flag
;
703 new_file_flag
= false;
704 if (! filter_ok
&& opened
== max_files
)
706 /* If filters no longer accepting input, stop reading. */
713 /* Ensure NUMBER files are created, which truncates
714 any existing files or notifies any consumers on fifos.
715 FIXME: Should we do this before EXIT_FAILURE? */
716 while (opened
++ < max_files
)
717 cwrite (true, nullptr, 0);
720 /* Split into pieces of exactly N_LINES lines.
721 Use buffer BUF, whose size is BUFSIZE. */
724 lines_split (intmax_t n_lines
, char *buf
, idx_t bufsize
)
727 char *bp
, *bp_out
, *eob
;
728 bool new_file_flag
= true;
733 n_read
= read (STDIN_FILENO
, buf
, bufsize
);
735 error (EXIT_FAILURE
, errno
, "%s", quotef (infile
));
741 bp
= rawmemchr (bp
, eolchar
);
744 if (eob
!= bp_out
) /* do not write 0 bytes! */
746 idx_t len
= eob
- bp_out
;
747 cwrite (new_file_flag
, bp_out
, len
);
748 new_file_flag
= false;
756 cwrite (new_file_flag
, bp_out
, bp
- bp_out
);
758 new_file_flag
= true;
766 /* Split into pieces that are as large as possible while still not more
767 than N_BYTES bytes, and are split on line boundaries except
768 where lines longer than N_BYTES bytes occur. */
771 line_bytes_split (intmax_t n_bytes
, char *buf
, idx_t bufsize
)
774 intmax_t n_out
= 0; /* for each split. */
776 char *hold
= nullptr; /* for lines > bufsize. */
778 bool split_line
= false; /* Whether a \n was output in a split. */
782 n_read
= read (STDIN_FILENO
, buf
, bufsize
);
784 error (EXIT_FAILURE
, errno
, "%s", quotef (infile
));
785 idx_t n_left
= n_read
;
789 idx_t split_rest
= 0;
793 /* Determine End Of Chunk and/or End of Line,
794 which are used below to select what to write or buffer. */
795 if (n_bytes
- n_out
- n_hold
<= n_left
)
797 /* Have enough for split. */
798 split_rest
= n_bytes
- n_out
- n_hold
;
799 eoc
= sob
+ split_rest
- 1;
800 eol
= memrchr (sob
, eolchar
, split_rest
);
803 eol
= memrchr (sob
, eolchar
, n_left
);
805 /* Output hold space if possible. */
806 if (n_hold
&& !(!eol
&& n_out
))
808 cwrite (n_out
== 0, hold
, n_hold
);
813 /* Output to eol if present. */
817 idx_t n_write
= eol
- sob
+ 1;
818 cwrite (n_out
== 0, sob
, n_write
);
823 split_rest
-= n_write
;
826 /* Output to eoc or eob if possible. */
827 if (n_left
&& !split_line
)
829 idx_t n_write
= eoc
? split_rest
: n_left
;
830 cwrite (n_out
== 0, sob
, n_write
);
835 split_rest
-= n_write
;
838 /* Update hold if needed. */
839 if ((eoc
&& split_rest
) || (!eoc
&& n_left
))
841 idx_t n_buf
= eoc
? split_rest
: n_left
;
842 if (hold_size
- n_hold
< n_buf
)
843 hold
= xpalloc (hold
, &hold_size
, n_buf
- (hold_size
- n_hold
),
845 memcpy (hold
+ n_hold
, sob
, n_buf
);
851 /* Reset for new split. */
861 /* Handle no eol at end of file. */
863 cwrite (n_out
== 0, hold
, n_hold
);
868 /* -n l/[K/]N: Write lines to files of approximately file size / N.
869 The file is partitioned into file size / N sized portions, with the
870 last assigned any excess. If a line _starts_ within a partition
871 it is written completely to the corresponding file. Since lines
872 are not split even if they overlap a partition, the files written
873 can be larger or smaller than the partition size, and even empty
874 if a line is so long as to completely overlap the partition. */
877 lines_chunk_split (intmax_t k
, intmax_t n
, char *buf
, idx_t bufsize
,
878 ssize_t initial_read
, off_t file_size
)
880 affirm (n
&& k
<= n
);
882 intmax_t rem_bytes
= file_size
% n
;
883 off_t chunk_size
= file_size
/ n
;
884 intmax_t chunk_no
= 1;
885 off_t chunk_end
= chunk_size
+ (0 < rem_bytes
);
887 bool new_file_flag
= true;
888 bool chunk_truncated
= false;
890 if (k
> 1 && 0 < file_size
)
892 /* Start reading 1 byte before kth chunk of file. */
893 off_t start
= (k
- 1) * chunk_size
+ MIN (k
- 1, rem_bytes
) - 1;
894 if (start
< initial_read
)
896 memmove (buf
, buf
+ start
, initial_read
- start
);
897 initial_read
-= start
;
901 if (initial_read
< start
902 && lseek (STDIN_FILENO
, start
- initial_read
, SEEK_CUR
) < 0)
903 error (EXIT_FAILURE
, errno
, "%s", quotef (infile
));
908 chunk_end
= start
+ 1;
911 while (n_written
< file_size
)
913 char *bp
= buf
, *eob
;
915 if (0 <= initial_read
)
917 n_read
= initial_read
;
922 n_read
= read (STDIN_FILENO
, buf
,
923 MIN (bufsize
, file_size
- n_written
));
925 error (EXIT_FAILURE
, errno
, "%s", quotef (infile
));
929 chunk_truncated
= false;
937 /* Begin looking for '\n' at last byte of chunk. */
938 off_t skip
= MIN (n_read
, MAX (0, chunk_end
- 1 - n_written
));
939 char *bp_out
= memchr (bp
+ skip
, eolchar
, n_read
- skip
);
947 to_write
= bp_out
- bp
;
951 /* We don't use the stdout buffer here since we're writing
952 large chunks from an existing file, so it's more efficient
953 to write out directly. */
954 if (full_write (STDOUT_FILENO
, bp
, to_write
) != to_write
)
958 cwrite (new_file_flag
, bp
, to_write
);
959 n_written
+= to_write
;
962 new_file_flag
= next
;
964 /* A line could have been so long that it skipped
965 entire chunks. So create empty files in that case. */
966 while (next
|| chunk_end
<= n_written
)
968 if (!next
&& bp
== eob
)
970 /* replenish buf, before going to next chunk. */
971 chunk_truncated
= true;
976 chunk_end
+= chunk_size
+ (chunk_no
< rem_bytes
);
978 if (chunk_end
<= n_written
)
981 cwrite (true, nullptr, 0);
992 /* Ensure NUMBER files are created, which truncates
993 any existing files or notifies any consumers on fifos.
994 FIXME: Should we do this before EXIT_FAILURE? */
996 while (chunk_no
++ <= n
)
997 cwrite (true, nullptr, 0);
1000 /* -n K/N: Extract Kth of N chunks. */
1003 bytes_chunk_extract (intmax_t k
, intmax_t n
, char *buf
, idx_t bufsize
,
1004 ssize_t initial_read
, off_t file_size
)
1009 assert (0 < k
&& k
<= n
);
1011 start
= (k
- 1) * (file_size
/ n
) + MIN (k
- 1, file_size
% n
);
1012 end
= k
== n
? file_size
: k
* (file_size
/ n
) + MIN (k
, file_size
% n
);
1014 if (start
< initial_read
)
1016 memmove (buf
, buf
+ start
, initial_read
- start
);
1017 initial_read
-= start
;
1021 if (initial_read
< start
1022 && lseek (STDIN_FILENO
, start
- initial_read
, SEEK_CUR
) < 0)
1023 error (EXIT_FAILURE
, errno
, "%s", quotef (infile
));
1030 if (0 <= initial_read
)
1032 n_read
= initial_read
;
1037 n_read
= read (STDIN_FILENO
, buf
, bufsize
);
1039 error (EXIT_FAILURE
, errno
, "%s", quotef (infile
));
1043 n_read
= MIN (n_read
, end
- start
);
1044 if (full_write (STDOUT_FILENO
, buf
, n_read
) != n_read
1045 && ! ignorable (errno
))
1046 error (EXIT_FAILURE
, errno
, "%s", quotef ("-"));
1051 typedef struct of_info
1065 /* Rotate file descriptors when we're writing to more output files than we
1066 have available file descriptors.
1067 Return whether we came under file resource pressure.
1068 If so, it's probably best to close each file when finished with it. */
1071 ofile_open (of_t
*files
, idx_t i_check
, idx_t nfiles
)
1073 bool file_limit
= false;
1075 if (files
[i_check
].ofd
<= OFD_NEW
)
1078 idx_t i_reopen
= i_check
? i_check
- 1 : nfiles
- 1;
1080 /* Another process could have opened a file in between the calls to
1081 close and open, so we should keep trying until open succeeds or
1082 we've closed all of our files. */
1085 if (files
[i_check
].ofd
== OFD_NEW
)
1086 fd
= create (files
[i_check
].of_name
);
1087 else /* OFD_APPEND */
1089 /* Attempt to append to previously opened file.
1090 We use O_NONBLOCK to support writing to fifos,
1091 where the other end has closed because of our
1092 previous close. In that case we'll immediately
1093 get an error, rather than waiting indefinitely.
1094 In specialized cases the consumer can keep reading
1095 from the fifo, terminating on conditions in the data
1096 itself, or perhaps never in the case of 'tail -f'.
1097 I.e., for fifos it is valid to attempt this reopen.
1099 We don't handle the filter_command case here, as create()
1100 will exit if there are not enough files in that case.
1101 I.e., we don't support restarting filters, as that would
1102 put too much burden on users specifying --filter commands. */
1103 fd
= open (files
[i_check
].of_name
,
1104 O_WRONLY
| O_BINARY
| O_APPEND
| O_NONBLOCK
);
1110 if (!(errno
== EMFILE
|| errno
== ENFILE
))
1111 error (EXIT_FAILURE
, errno
, "%s", quotef (files
[i_check
].of_name
));
1115 /* Search backwards for an open file to close. */
1116 while (files
[i_reopen
].ofd
< 0)
1118 i_reopen
= i_reopen
? i_reopen
- 1 : nfiles
- 1;
1119 /* No more open files to close, exit with E[NM]FILE. */
1120 if (i_reopen
== i_check
)
1121 error (EXIT_FAILURE
, errno
, "%s",
1122 quotef (files
[i_check
].of_name
));
1125 if (fclose (files
[i_reopen
].ofile
) != 0)
1126 error (EXIT_FAILURE
, errno
, "%s", quotef (files
[i_reopen
].of_name
));
1127 files
[i_reopen
].ofile
= nullptr;
1128 files
[i_reopen
].ofd
= OFD_APPEND
;
1131 files
[i_check
].ofd
= fd
;
1132 FILE *ofile
= fdopen (fd
, "a");
1134 error (EXIT_FAILURE
, errno
, "%s", quotef (files
[i_check
].of_name
));
1135 files
[i_check
].ofile
= ofile
;
1136 files
[i_check
].opid
= filter_pid
;
1143 /* -n r/[K/]N: Divide file into N chunks in round robin fashion.
1144 Use BUF of size BUFSIZE for the buffer, and if allocating storage
1145 put its address into *FILESP to pacify -fsanitize=leak.
1146 When K == 0, we try to keep the files open in parallel.
1147 If we run out of file resources, then we revert
1148 to opening and closing each file for each line. */
1151 lines_rr (intmax_t k
, intmax_t n
, char *buf
, idx_t bufsize
, of_t
**filesp
)
1153 bool wrapped
= false;
1157 of_t
*files
IF_LINT (= nullptr);
1166 files
= *filesp
= xinmalloc (n
, sizeof *files
);
1168 /* Generate output file names. */
1169 for (i_file
= 0; i_file
< n
; i_file
++)
1172 files
[i_file
].of_name
= xstrdup (outfile
);
1173 files
[i_file
].ofd
= OFD_NEW
;
1174 files
[i_file
].ofile
= nullptr;
1175 files
[i_file
].opid
= 0;
1183 char *bp
= buf
, *eob
;
1184 ssize_t n_read
= read (STDIN_FILENO
, buf
, bufsize
);
1186 error (EXIT_FAILURE
, errno
, "%s", quotef (infile
));
1187 else if (n_read
== 0)
1196 /* Find end of line. */
1197 char *bp_out
= memchr (bp
, eolchar
, eob
- bp
);
1205 to_write
= bp_out
- bp
;
1209 if (line_no
== k
&& unbuffered
)
1211 if (full_write (STDOUT_FILENO
, bp
, to_write
) != to_write
)
1214 else if (line_no
== k
&& fwrite (bp
, to_write
, 1, stdout
) != 1)
1219 line_no
= (line_no
== n
) ? 1 : line_no
+ 1;
1223 /* Secure file descriptor. */
1224 file_limit
|= ofile_open (files
, i_file
, n
);
1227 /* Note writing to fd, rather than flushing the FILE gives
1228 an 8% performance benefit, due to reduced data copying. */
1229 if (full_write (files
[i_file
].ofd
, bp
, to_write
) != to_write
1230 && ! ignorable (errno
))
1231 error (EXIT_FAILURE
, errno
, "%s",
1232 quotef (files
[i_file
].of_name
));
1234 else if (fwrite (bp
, to_write
, 1, files
[i_file
].ofile
) != 1
1235 && ! ignorable (errno
))
1236 error (EXIT_FAILURE
, errno
, "%s",
1237 quotef (files
[i_file
].of_name
));
1239 if (! ignorable (errno
))
1244 if (fclose (files
[i_file
].ofile
) != 0)
1245 error (EXIT_FAILURE
, errno
, "%s",
1246 quotef (files
[i_file
].of_name
));
1247 files
[i_file
].ofile
= nullptr;
1248 files
[i_file
].ofd
= OFD_APPEND
;
1250 if (next
&& ++i_file
== n
)
1253 /* If no filters are accepting input, stop reading. */
1266 /* Ensure all files created, so that any existing files are truncated,
1267 and to signal any waiting fifo consumers.
1268 Also, close any open file descriptors.
1269 FIXME: Should we do this before EXIT_FAILURE? */
1272 idx_t ceiling
= wrapped
? n
: i_file
;
1273 for (i_file
= 0; i_file
< n
; i_file
++)
1275 if (i_file
>= ceiling
&& !elide_empty_files
)
1276 file_limit
|= ofile_open (files
, i_file
, n
);
1277 if (files
[i_file
].ofd
>= 0)
1278 closeout (files
[i_file
].ofile
, files
[i_file
].ofd
,
1279 files
[i_file
].opid
, files
[i_file
].of_name
);
1280 files
[i_file
].ofd
= OFD_APPEND
;
1285 #define FAIL_ONLY_ONE_WAY() \
1288 error (0, 0, _("cannot split in more than one way")); \
1289 usage (EXIT_FAILURE); \
1293 /* Report a string-to-integer conversion failure MSGID with ARG. */
1295 static _Noreturn
void
1296 strtoint_die (char const *msgid
, char const *arg
)
1298 error (EXIT_FAILURE
, errno
== EINVAL
? 0 : errno
, "%s: %s",
1299 gettext (msgid
), quote (arg
));
1302 /* Use OVERFLOW_OK when it is OK to ignore LONGINT_OVERFLOW errors, since the
1303 extreme value will do the right thing anyway on any practical platform. */
1304 #define OVERFLOW_OK LONGINT_OVERFLOW
1306 /* Parse ARG for number of bytes or lines. The number can be followed
1307 by MULTIPLIERS, and the resulting value must be positive.
1308 If the number cannot be parsed, diagnose with MSG.
1309 Return the number parsed, or an INTMAX_MAX on overflow. */
1312 parse_n_units (char const *arg
, char const *multipliers
, char const *msgid
)
1315 if (OVERFLOW_OK
< xstrtoimax (arg
, nullptr, 10, &n
, multipliers
) || n
< 1)
1316 strtoint_die (msgid
, arg
);
1320 /* Parse K/N syntax of chunk options. */
1323 parse_chunk (intmax_t *k_units
, intmax_t *n_units
, char const *arg
)
1326 strtol_error e
= xstrtoimax (arg
, &argend
, 10, n_units
, "");
1327 if (e
== LONGINT_INVALID_SUFFIX_CHAR
&& *argend
== '/')
1329 *k_units
= *n_units
;
1330 *n_units
= parse_n_units (argend
+ 1, "",
1331 N_("invalid number of chunks"));
1332 if (! (0 < *k_units
&& *k_units
<= *n_units
))
1333 error (EXIT_FAILURE
, 0, "%s: %s", _("invalid chunk number"),
1334 quote_mem (arg
, argend
- arg
));
1336 else if (! (e
<= OVERFLOW_OK
&& 0 < *n_units
))
1337 strtoint_die (N_("invalid number of chunks"), arg
);
1342 main (int argc
, char **argv
)
1344 enum Split_type split_type
= type_undef
;
1345 idx_t in_blk_size
= 0; /* optimal block size of input file device */
1346 idx_t page_size
= getpagesize ();
1347 intmax_t k_units
= 0;
1348 intmax_t n_units
= 0;
1350 static char const multipliers
[] = "bEGKkMmPQRTYZ0";
1352 int digits_optind
= 0;
1353 off_t file_size
= OFF_T_MAX
;
1355 initialize_main (&argc
, &argv
);
1356 set_program_name (argv
[0]);
1357 setlocale (LC_ALL
, "");
1358 bindtextdomain (PACKAGE
, LOCALEDIR
);
1359 textdomain (PACKAGE
);
1361 atexit (close_stdout
);
1363 /* Parse command line options. */
1370 /* This is the argv-index of the option we will read next. */
1371 int this_optind
= optind
? optind
: 1;
1373 c
= getopt_long (argc
, argv
, "0123456789C:a:b:del:n:t:ux",
1381 suffix_length
= xdectoimax (optarg
, 0, IDX_MAX
,
1382 "", _("invalid suffix length"), 0);
1385 case ADDITIONAL_SUFFIX_OPTION
:
1387 int suffix_len
= strlen (optarg
);
1388 if (last_component (optarg
) != optarg
1389 || (suffix_len
&& ISSLASH (optarg
[suffix_len
- 1])))
1392 _("invalid suffix %s, contains directory separator"),
1394 usage (EXIT_FAILURE
);
1397 additional_suffix
= optarg
;
1401 if (split_type
!= type_undef
)
1402 FAIL_ONLY_ONE_WAY ();
1403 split_type
= type_bytes
;
1404 n_units
= parse_n_units (optarg
, multipliers
,
1405 N_("invalid number of bytes"));
1409 if (split_type
!= type_undef
)
1410 FAIL_ONLY_ONE_WAY ();
1411 split_type
= type_lines
;
1412 n_units
= parse_n_units (optarg
, "", N_("invalid number of lines"));
1416 if (split_type
!= type_undef
)
1417 FAIL_ONLY_ONE_WAY ();
1418 split_type
= type_byteslines
;
1419 n_units
= parse_n_units (optarg
, multipliers
,
1420 N_("invalid number of lines"));
1424 if (split_type
!= type_undef
)
1425 FAIL_ONLY_ONE_WAY ();
1426 /* skip any whitespace */
1427 while (isspace (to_uchar (*optarg
)))
1429 if (STRNCMP_LIT (optarg
, "r/") == 0)
1431 split_type
= type_rr
;
1434 else if (STRNCMP_LIT (optarg
, "l/") == 0)
1436 split_type
= type_chunk_lines
;
1440 split_type
= type_chunk_bytes
;
1441 parse_chunk (&k_units
, &n_units
, optarg
);
1450 char neweol
= optarg
[0];
1452 error (EXIT_FAILURE
, 0, _("empty record separator"));
1455 if (STREQ (optarg
, "\\0"))
1459 /* Provoke with 'split -txx'. Complain about
1460 "multi-character tab" instead of "multibyte tab", so
1461 that the diagnostic's wording does not need to be
1462 changed once multibyte characters are supported. */
1463 error (EXIT_FAILURE
, 0, _("multi-character separator %s"),
1467 /* Make it explicit we don't support multiple separators. */
1468 if (0 <= eolchar
&& neweol
!= eolchar
)
1470 error (EXIT_FAILURE
, 0,
1471 _("multiple separator characters specified"));
1488 if (split_type
== type_undef
)
1490 split_type
= type_digits
;
1493 if (split_type
!= type_undef
&& split_type
!= type_digits
)
1494 FAIL_ONLY_ONE_WAY ();
1495 if (digits_optind
!= 0 && digits_optind
!= this_optind
)
1496 n_units
= 0; /* More than one number given; ignore other. */
1497 digits_optind
= this_optind
;
1498 if (ckd_mul (&n_units
, n_units
, 10)
1499 || ckd_add (&n_units
, n_units
, c
- '0'))
1500 n_units
= INTMAX_MAX
;
1506 suffix_alphabet
= "0123456789";
1508 suffix_alphabet
= "0123456789abcdef";
1511 if (strlen (optarg
) != strspn (optarg
, suffix_alphabet
))
1515 _("%s: invalid start value for numerical suffix") :
1516 _("%s: invalid start value for hexadecimal suffix"),
1518 usage (EXIT_FAILURE
);
1522 /* Skip any leading zero. */
1523 while (*optarg
== '0' && *(optarg
+ 1) != '\0')
1525 numeric_suffix_start
= optarg
;
1531 elide_empty_files
= true;
1535 filter_command
= optarg
;
1538 case IO_BLKSIZE_OPTION
:
1539 in_blk_size
= xdectoumax (optarg
, 1,
1540 MIN (SYS_BUFSIZE_MAX
,
1541 MIN (IDX_MAX
, SIZE_MAX
) - 1),
1542 multipliers
, _("invalid IO block size"), 0);
1545 case VERBOSE_OPTION
:
1549 case_GETOPT_HELP_CHAR
;
1551 case_GETOPT_VERSION_CHAR (PROGRAM_NAME
, AUTHORS
);
1554 usage (EXIT_FAILURE
);
1558 if (k_units
!= 0 && filter_command
)
1560 error (0, 0, _("--filter does not process a chunk extracted to stdout"));
1561 usage (EXIT_FAILURE
);
1564 /* Handle default case. */
1565 if (split_type
== type_undef
)
1567 split_type
= type_lines
;
1573 error (0, 0, _("invalid number of lines: %s"), quote ("0"));
1574 usage (EXIT_FAILURE
);
1580 set_suffix_length (n_units
, split_type
);
1582 /* Get out the filename arguments. */
1585 infile
= argv
[optind
++];
1588 outbase
= argv
[optind
++];
1592 error (0, 0, _("extra operand %s"), quote (argv
[optind
]));
1593 usage (EXIT_FAILURE
);
1596 /* Check that the suffix length is large enough for the numerical
1597 suffix start value. */
1598 if (numeric_suffix_start
&& strlen (numeric_suffix_start
) > suffix_length
)
1600 error (0, 0, _("numerical suffix start value is too large "
1601 "for the suffix length"));
1602 usage (EXIT_FAILURE
);
1605 /* Open the input file. */
1606 if (! STREQ (infile
, "-")
1607 && fd_reopen (STDIN_FILENO
, infile
, O_RDONLY
, 0) < 0)
1608 error (EXIT_FAILURE
, errno
, _("cannot open %s for reading"),
1611 /* Binary I/O is safer when byte counts are used. */
1612 xset_binary_mode (STDIN_FILENO
, O_BINARY
);
1614 /* Advise the kernel of our access pattern. */
1615 fdadvise (STDIN_FILENO
, 0, 0, FADVISE_SEQUENTIAL
);
1617 /* Get the optimal block size of input device and make a buffer. */
1619 if (fstat (STDIN_FILENO
, &in_stat_buf
) != 0)
1620 error (EXIT_FAILURE
, errno
, "%s", quotef (infile
));
1622 if (in_blk_size
== 0)
1624 in_blk_size
= io_blksize (&in_stat_buf
);
1625 if (SYS_BUFSIZE_MAX
< in_blk_size
)
1626 in_blk_size
= SYS_BUFSIZE_MAX
;
1629 char *buf
= xalignalloc (page_size
, in_blk_size
+ 1);
1630 ssize_t initial_read
= -1;
1632 if (split_type
== type_chunk_bytes
|| split_type
== type_chunk_lines
)
1634 file_size
= input_file_size (STDIN_FILENO
, &in_stat_buf
,
1637 error (EXIT_FAILURE
, errno
, _("%s: cannot determine file size"),
1639 initial_read
= MIN (file_size
, in_blk_size
);
1642 /* When filtering, closure of one pipe must not terminate the process,
1643 as there may still be other streams expecting input from us. */
1645 default_SIGPIPE
= signal (SIGPIPE
, SIG_IGN
) == SIG_DFL
;
1651 lines_split (n_units
, buf
, in_blk_size
);
1655 bytes_split (n_units
, 0, buf
, in_blk_size
, -1, 0);
1658 case type_byteslines
:
1659 line_bytes_split (n_units
, buf
, in_blk_size
);
1662 case type_chunk_bytes
:
1664 bytes_split (file_size
/ n_units
, file_size
% n_units
,
1665 buf
, in_blk_size
, initial_read
, n_units
);
1667 bytes_chunk_extract (k_units
, n_units
, buf
, in_blk_size
, initial_read
,
1671 case type_chunk_lines
:
1672 lines_chunk_split (k_units
, n_units
, buf
, in_blk_size
, initial_read
,
1677 /* Note, this is like 'sed -n ${k}~${n}p' when k > 0,
1678 but the functionality is provided for symmetry. */
1681 lines_rr (k_units
, n_units
, buf
, in_blk_size
, &files
);
1689 if (close (STDIN_FILENO
) != 0)
1690 error (EXIT_FAILURE
, errno
, "%s", quotef (infile
));
1691 closeout (nullptr, output_desc
, filter_pid
, outfile
);
1693 main_exit (EXIT_SUCCESS
);