1 /* Various utility functions.
2 Copyright (C) 1996, 1997, 1998, 1999, 2000, 2001, 2002, 2003, 2004,
3 2005, 2006, 2007, 2008, 2009, 2010, 2011 Free Software Foundation,
6 This file is part of GNU Wget.
8 GNU Wget is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 3 of the License, or
11 (at your option) any later version.
13 GNU Wget is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with Wget. If not, see <http://www.gnu.org/licenses/>.
21 Additional permission under GNU GPL version 3 section 7
23 If you modify this program, or any covered work, by linking or
24 combining it with the OpenSSL project's OpenSSL library (or a
25 modified version of that library), containing parts covered by the
26 terms of the OpenSSL or SSLeay licenses, the Free Software Foundation
27 grants you additional permission to convey the resulting work.
28 Corresponding Source for a non-source form of such a combination
29 shall include the source code for the parts of OpenSSL used as well
30 as that of the covered work. */
40 # include <sys/mman.h>
43 # include <process.h> /* getpid() */
52 # include <sys/types.h>
57 # ifdef HAVE_SYS_UTIME_H
58 # include <sys/utime.h>
66 /* For TIOCGWINSZ and friends: */
67 #include <sys/ioctl.h>
72 /* Needed for Unix version of run_with_timeout. */
81 #ifndef HAVE_SIGSETJMP
82 /* If sigsetjmp is a macro, configure won't pick it up. */
84 # define HAVE_SIGSETJMP
88 #if defined HAVE_SIGSETJMP || defined HAVE_SIGBLOCK
89 # define USE_SIGNAL_TIMEOUT
97 #endif /* def __VMS */
104 memfatal (const char *context, long attempted_size)
106 /* Make sure we don't try to store part of the log line, and thus
108 log_set_save_context (false);
110 /* We have different log outputs in different situations:
111 1) output without bytes information
112 2) output with bytes information */
113 if (attempted_size == UNKNOWN_ATTEMPTED_SIZE)
115 logprintf (LOG_ALWAYS,
116 _("%s: %s: Failed to allocate enough memory; memory exhausted.\n"),
121 logprintf (LOG_ALWAYS,
122 _("%s: %s: Failed to allocate %ld bytes; memory exhausted.\n"),
123 exec_name, context, attempted_size);
129 /* Character property table for (re-)escaping VMS ODS5 extended file
130 names. Note that this table ignores Unicode.
132 ODS2 valid characters: 0-9 A-Z a-z $ - _ ~
134 ODS5 Invalid characters:
135 C0 control codes (0x00 to 0x1F inclusive)
139 ODS5 Invalid characters only in VMS V7.2 (which no one runs, right?):
140 Double quotation marks (")
143 Left angle bracket (<)
144 Right angle bracket (>)
148 Characters escaped by "^":
149 SP ! " # % & ' ( ) + , . : ; =
152 Either "^_" or "^ " is accepted as a space. Period (.) is a special
153 case. Note that un-escaped < and > can also confuse a directory
156 Characters put out as ^xx:
158 80-9F (C1 control characters)
159 A0 (nonbreaking space)
160 FF (Latin small letter y diaeresis)
163 Unicode: "^Uxxxx", where "xxxx" is four hex digits.
165 Property table values:
175 unsigned char char_prop[ 256] = {
177 /* NUL SOH STX ETX EOT ENQ ACK BEL BS HT LF VT FF CR SO SI */
178 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
180 /* DLE DC1 DC2 DC3 DC4 NAK SYN ETB CAN EM SUB ESC FS GS RS US */
181 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
183 /* SP ! " # $ % & ' ( ) * + , - . / */
184 2, 1, 1, 1, 16, 1, 1, 1, 1, 1, 0, 1, 1, 16, 4, 0,
186 /* 0 1 2 3 4 5 6 7 8 9 : ; < = > ? */
187 80, 80, 80, 80, 80, 80, 80, 80, 80, 80, 1, 1, 1, 1, 1, 1,
189 /* @ A B C D E F G H I J K L M N O */
190 1, 80, 80, 80, 80, 80, 80, 16, 16, 16, 16, 16, 16, 16, 16, 16,
192 /* P Q R S T U V W X Y Z [ \ ] ^ _ */
193 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 1, 1, 1, 1, 16,
195 /* ` a b c d e f g h i j k l m n o */
196 1, 96, 96, 96, 96, 96, 96, 32, 32, 32, 32, 32, 32, 32, 32, 32,
198 /* p q r s t u v w x y z { | } ~ DEL */
199 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 1, 1, 1, 17, 8,
201 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
202 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8,
203 8, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
204 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
205 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
206 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
207 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
208 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 8
211 /* Utility function: like xstrdup(), but also lowercases S. */
214 xstrdup_lower (const char *s)
216 char *copy = xstrdup (s);
223 /* Copy the string formed by two pointers (one on the beginning, other
224 on the char after the last char) to a new, malloc-ed location.
227 strdupdelim (const char *beg, const char *end)
229 char *res = xmalloc (end - beg + 1);
230 memcpy (res, beg, end - beg);
231 res[end - beg] = '\0';
235 /* Parse a string containing comma-separated elements, and return a
236 vector of char pointers with the elements. Spaces following the
237 commas are ignored. */
239 sepstring (const char *s)
253 res = xrealloc (res, (i + 2) * sizeof (char *));
254 res[i] = strdupdelim (p, s);
257 /* Skip the blanks following the ','. */
258 while (c_isspace (*s))
265 res = xrealloc (res, (i + 2) * sizeof (char *));
266 res[i] = strdupdelim (p, s);
271 /* Like sprintf, but prints into a string of sufficient size freshly
272 allocated with malloc, which is returned. If unable to print due
273 to invalid format, returns NULL. Inability to allocate needed
274 memory results in abort, as with xmalloc. This is in spirit
275 similar to the GNU/BSD extension asprintf, but somewhat easier to
278 Internally the function either calls vasprintf or loops around
279 vsnprintf until the correct size is found. Since Wget also ships a
280 fallback implementation of vsnprintf, this should be portable. */
282 /* Constant is using for limits memory allocation for text buffer.
283 Applicable in situation when: vasprintf is not available in the system
284 and vsnprintf return -1 when long line is truncated (in old versions of
285 glibc and in other system where C99 doesn`t support) */
287 #define FMT_MAX_LENGTH 1048576
290 aprintf (const char *fmt, ...)
292 #if defined HAVE_VASPRINTF && !defined DEBUG_MALLOC
297 va_start (args, fmt);
298 ret = vasprintf (&str, fmt, args);
300 if (ret < 0 && errno == ENOMEM)
301 memfatal ("aprintf", UNKNOWN_ATTEMPTED_SIZE); /* for consistency
302 with xmalloc/xrealloc */
306 #else /* not HAVE_VASPRINTF */
308 /* vasprintf is unavailable. snprintf into a small buffer and
309 resize it as necessary. */
311 char *str = xmalloc (size);
313 /* #### This code will infloop and eventually abort in xrealloc if
314 passed a FMT that causes snprintf to consistently return -1. */
321 va_start (args, fmt);
322 n = vsnprintf (str, size, fmt, args);
325 /* If the printing worked, return the string. */
326 if (n > -1 && n < size)
329 /* Else try again with a larger buffer. */
330 if (n > -1) /* C99 */
331 size = n + 1; /* precisely what is needed */
332 else if (size >= FMT_MAX_LENGTH) /* We have a huge buffer, */
333 { /* maybe we have some wrong
335 logprintf (LOG_ALWAYS,
336 _("%s: aprintf: text buffer is too big (%ld bytes), "
338 exec_name, size); /* printout a log message */
339 abort (); /* and abort... */
343 /* else, we continue to grow our
344 * buffer: Twice the old size. */
347 str = xrealloc (str, size);
349 #endif /* not HAVE_VASPRINTF */
352 /* Concatenate the NULL-terminated list of string arguments into
353 freshly allocated space. */
356 concat_strings (const char *str0, ...)
359 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
362 const char *next_str;
363 int total_length = 0;
366 /* Calculate the length of and allocate the resulting string. */
369 va_start (args, str0);
370 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
372 int len = strlen (next_str);
373 if (argcount < countof (saved_lengths))
374 saved_lengths[argcount++] = len;
378 p = ret = xmalloc (total_length + 1);
380 /* Copy the strings into the allocated space. */
383 va_start (args, str0);
384 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
387 if (argcount < countof (saved_lengths))
388 len = saved_lengths[argcount++];
390 len = strlen (next_str);
391 memcpy (p, next_str, len);
400 /* Format the provided time according to the specified format. The
401 format is a string with format elements supported by strftime. */
404 fmttime (time_t t, const char *fmt)
406 static char output[32];
407 struct tm *tm = localtime(&t);
410 if (!strftime(output, sizeof(output), fmt, tm))
415 /* Return pointer to a static char[] buffer in which zero-terminated
416 string-representation of TM (in form hh:mm:ss) is printed.
418 If TM is NULL, the current time will be used. */
423 return fmttime(t, "%H:%M:%S");
426 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
429 datetime_str (time_t t)
431 return fmttime(t, "%Y-%m-%d %H:%M:%S");
434 /* The Windows versions of the following two functions are defined in
435 mswindows.c. On MSDOS this function should never be called. */
440 fork_to_background (void)
445 #else /* def __VMS */
447 #if !defined(WINDOWS) && !defined(MSDOS)
449 fork_to_background (void)
452 /* Whether we arrange our own version of opt.lfilename here. */
453 bool logfile_changed = false;
455 if (!opt.lfilename && (!opt.quiet || opt.server_response))
457 /* We must create the file immediately to avoid either a race
458 condition (which arises from using unique_name and failing to
459 use fopen_excl) or lying to the user about the log file name
460 (which arises from using unique_name, printing the name, and
461 using fopen_excl later on.) */
462 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, false, &opt.lfilename);
465 logfile_changed = true;
478 /* parent, no error */
479 printf (_("Continuing in background, pid %d.\n"), (int) pid);
481 printf (_("Output will be written to %s.\n"), quote (opt.lfilename));
482 exit (0); /* #### should we use _exit()? */
485 /* child: give up the privileges and keep running. */
487 freopen ("/dev/null", "r", stdin);
488 freopen ("/dev/null", "w", stdout);
489 freopen ("/dev/null", "w", stderr);
491 #endif /* !WINDOWS && !MSDOS */
493 #endif /* def __VMS [else] */
496 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
497 specified with TM. The atime ("access time") is set to the current
501 touch (const char *file, time_t tm)
504 # ifdef HAVE_STRUCT_UTIMBUF
505 struct utimbuf times;
513 times.actime = time (NULL);
514 if (utime (file, ×) == -1)
515 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
517 struct timespec timespecs[2];
520 fd = open (file, O_WRONLY);
523 logprintf (LOG_NOTQUIET, "open(%s): %s\n", file, strerror (errno));
527 timespecs[0].tv_sec = time (NULL);
528 timespecs[0].tv_nsec = 0L;
529 timespecs[1].tv_sec = tm;
530 timespecs[1].tv_nsec = 0L;
532 if (futimens (fd, timespecs) == -1)
533 logprintf (LOG_NOTQUIET, "futimens(%s): %s\n", file, strerror (errno));
539 /* Checks if FILE is a symbolic link, and removes it if it is. Does
540 nothing under MS-Windows. */
542 remove_link (const char *file)
547 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
549 DEBUGP (("Unlinking %s (symlink).\n", file));
552 logprintf (LOG_VERBOSE, _("Failed to unlink symlink %s: %s\n"),
553 quote (file), strerror (errno));
558 /* Does FILENAME exist? This is quite a lousy implementation, since
559 it supplies no error codes -- only a yes-or-no answer. Thus it
560 will return that a file does not exist if, e.g., the directory is
561 unreadable. I don't mind it too much currently, though. The
562 proper way should, of course, be to have a third, error state,
563 other than true/false, but that would introduce uncalled-for
564 additional complexity to the callers. */
566 file_exists_p (const char *filename)
569 return access (filename, F_OK) >= 0;
572 return stat (filename, &buf) >= 0;
576 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
577 Returns 0 on error. */
579 file_non_directory_p (const char *path)
582 /* Use lstat() rather than stat() so that symbolic links pointing to
583 directories can be identified correctly. */
584 if (lstat (path, &buf) != 0)
586 return S_ISDIR (buf.st_mode) ? false : true;
589 /* Return the size of file named by FILENAME, or -1 if it cannot be
590 opened or seeked into. */
592 file_size (const char *filename)
594 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
596 /* We use fseek rather than stat to determine the file size because
597 that way we can also verify that the file is readable without
598 explicitly checking for permissions. Inspired by the POST patch
600 FILE *fp = fopen (filename, "rb");
603 fseeko (fp, 0, SEEK_END);
609 if (stat (filename, &st) < 0)
616 If no UNIQ_SEP is defined (as on VMS), have unique_name() return the
617 original name. With the VMS file systems' versioning, everything
618 should be fine, and appending ".NN" just causes trouble.
623 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
624 doesn't exist is found. Return a freshly allocated copy of the
628 unique_name_1 (const char *prefix)
631 int plen = strlen (prefix);
632 char *template = (char *)alloca (plen + 1 + 24);
633 char *template_tail = template + plen;
635 memcpy (template, prefix, plen);
636 *template_tail++ = UNIQ_SEP;
639 number_to_string (template_tail, count++);
640 while (file_exists_p (template));
642 return xstrdup (template);
645 /* Return a unique file name, based on FILE.
647 More precisely, if FILE doesn't exist, it is returned unmodified.
648 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
649 file name that doesn't exist is returned.
651 2005-02-19 SMS. "." is now UNIQ_SEP, and may be different.
653 The resulting file is not created, only verified that it didn't
654 exist at the point in time when the function was called.
655 Therefore, where security matters, don't rely that the file created
656 by this function exists until you open it with O_EXCL or
659 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
660 string. Otherwise, it may return FILE if the file doesn't exist
661 (and therefore doesn't need changing). */
664 unique_name (const char *file, bool allow_passthrough)
666 /* If the FILE itself doesn't exist, return it without
668 if (!file_exists_p (file))
669 return allow_passthrough ? (char *)file : xstrdup (file);
671 /* Otherwise, find a numeric suffix that results in unused file name
673 return unique_name_1 (file);
676 #else /* def UNIQ_SEP */
678 /* Dummy unique_name() for VMS. Return the original name as easily as
682 unique_name (const char *file, bool allow_passthrough)
684 /* Return the FILE itself, without modification, irregardful. */
685 return allow_passthrough ? (char *)file : xstrdup (file);
688 #endif /* def UNIQ_SEP [else] */
690 /* Create a file based on NAME, except without overwriting an existing
691 file with that name. Providing O_EXCL is correctly implemented,
692 this function does not have the race condition associated with
693 opening the file returned by unique_name. */
696 unique_create (const char *name, bool binary, char **opened_name)
698 /* unique file name, based on NAME */
699 char *uname = unique_name (name, false);
701 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
704 uname = unique_name (name, false);
706 if (opened_name && fp != NULL)
709 *opened_name = uname;
721 /* Open the file for writing, with the addition that the file is
722 opened "exclusively". This means that, if the file already exists,
723 this function will *fail* and errno will be set to EEXIST. If
724 BINARY is set, the file will be opened in binary mode, equivalent
727 If opening the file fails for any reason, including the file having
728 previously existed, this function returns NULL and sets errno
732 fopen_excl (const char *fname, int binary)
738 VMS lacks O_BINARY, but makes up for it in weird and wonderful ways.
739 It also has file versions which obviate all the O_EXCL effort.
740 O_TRUNC (something of a misnomer) requests a new version.
743 /* Common open() optional arguments:
744 sequential access only, access callback function.
746 # define OPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
749 int flags = O_WRONLY | O_CREAT | O_TRUNC;
754 fd = open( fname, /* File name. */
756 0777, /* Mode for default protection. */
757 "ctx=bin,stm", /* Binary, stream access. */
758 "rfm=stmlf", /* Stream_LF. */
759 OPEN_OPT_ARGS); /* Access callback. */
764 fd = open( fname, /* File name. */
766 0777, /* Mode for default protection. */
767 "ctx=bin,stm", /* Binary, stream access. */
768 "rfm=fix", /* Fixed-length, */
769 "mrs=512", /* 512-byte records. */
770 OPEN_OPT_ARGS); /* Access callback. */
775 fd = open( fname, /* File name. */
777 0777, /* Mode for default protection. */
778 "rfm=stmlf", /* Stream_LF. */
779 OPEN_OPT_ARGS); /* Access callback. */
781 # else /* def __VMS */
782 int flags = O_WRONLY | O_CREAT | O_EXCL;
787 fd = open (fname, flags, 0666);
788 # endif /* def __VMS [else] */
792 return fdopen (fd, binary ? "wb" : "w");
793 #else /* not O_EXCL */
794 /* Manually check whether the file exists. This is prone to race
795 conditions, but systems without O_EXCL haven't deserved
797 if (file_exists_p (fname))
802 return fopen (fname, binary ? "wb" : "w");
803 #endif /* not O_EXCL */
806 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
807 are missing, create them first. In case any mkdir() call fails,
808 return its error status. Returns 0 on successful completion.
810 The behaviour of this function should be identical to the behaviour
811 of `mkdir -p' on systems where mkdir supports the `-p' option. */
813 make_directory (const char *directory)
815 int i, ret, quit = 0;
818 /* Make a copy of dir, to be able to write to it. Otherwise, the
819 function is unsafe if called with a read-only char *argument. */
820 STRDUP_ALLOCA (dir, directory);
822 /* If the first character of dir is '/', skip it (and thus enable
823 creation of absolute-pathname directories. */
824 for (i = (*dir == '/'); 1; ++i)
826 for (; dir[i] && dir[i] != '/'; i++)
831 /* Check whether the directory already exists. Allow creation of
832 of intermediate directories to fail, as the initial path components
833 are not necessarily directories! */
834 if (!file_exists_p (dir))
835 ret = mkdir (dir, 0777);
846 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
847 should be a file name.
849 file_merge("/foo/bar", "baz") => "/foo/baz"
850 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
851 file_merge("foo", "bar") => "bar"
853 In other words, it's a simpler and gentler version of uri_merge. */
856 file_merge (const char *base, const char *file)
859 const char *cut = (const char *)strrchr (base, '/');
862 return xstrdup (file);
864 result = xmalloc (cut - base + 1 + strlen (file) + 1);
865 memcpy (result, base, cut - base);
866 result[cut - base] = '/';
867 strcpy (result + (cut - base) + 1, file);
872 /* Like fnmatch, but performs a case-insensitive match. */
875 fnmatch_nocase (const char *pattern, const char *string, int flags)
878 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
879 also present on *BSD platforms, and possibly elsewhere. */
880 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
882 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
883 char *patcopy = (char *) alloca (strlen (pattern) + 1);
884 char *strcopy = (char *) alloca (strlen (string) + 1);
886 for (p = patcopy; *pattern; pattern++, p++)
887 *p = c_tolower (*pattern);
889 for (p = strcopy; *string; string++, p++)
890 *p = c_tolower (*string);
892 return fnmatch (patcopy, strcopy, flags);
896 static bool in_acclist (const char *const *, const char *, bool);
898 /* Determine whether a file is acceptable to be followed, according to
899 lists of patterns to accept/reject. */
901 acceptable (const char *s)
905 if (opt.output_document && strcmp (s, opt.output_document) == 0)
908 while (l && s[l] != '/')
915 return (in_acclist ((const char *const *)opt.accepts, s, true)
916 && !in_acclist ((const char *const *)opt.rejects, s, true));
918 return in_acclist ((const char *const *)opt.accepts, s, true);
920 else if (opt.rejects)
921 return !in_acclist ((const char *const *)opt.rejects, s, true);
925 /* Determine whether an URL is acceptable to be followed, according to
926 regex patterns to accept/reject. */
928 accept_url (const char *s)
930 if (opt.acceptregex && !opt.regex_match_fun (opt.acceptregex, s))
932 if (opt.rejectregex && opt.regex_match_fun (opt.rejectregex, s))
938 /* Check if D2 is a subdirectory of D1. E.g. if D1 is `/something', subdir_p()
939 will return true if and only if D2 begins with `/something/' or is exactly
942 subdir_p (const char *d1, const char *d2)
946 if (!opt.ignore_case)
947 for (; *d1 && *d2 && (*d1 == *d2); ++d1, ++d2)
950 for (; *d1 && *d2 && (c_tolower (*d1) == c_tolower (*d2)); ++d1, ++d2)
953 return *d1 == '\0' && (*d2 == '\0' || *d2 == '/');
956 /* Iterate through DIRLIST (which must be NULL-terminated), and return the
957 first element that matches DIR, through wildcards or front comparison (as
960 dir_matches_p (char **dirlist, const char *dir)
963 int (*matcher) (const char *, const char *, int)
964 = opt.ignore_case ? fnmatch_nocase : fnmatch;
966 for (x = dirlist; *x; x++)
968 /* Remove leading '/' */
969 char *p = *x + (**x == '/');
970 if (has_wildcards_p (p))
972 if (matcher (p, dir, FNM_PATHNAME) == 0)
977 if (subdir_p (p, dir))
982 return *x ? true : false;
985 /* Returns whether DIRECTORY is acceptable for download, wrt the
986 include/exclude lists.
988 The leading `/' is ignored in paths; relative and absolute paths
989 may be freely intermixed. */
992 accdir (const char *directory)
994 /* Remove starting '/'. */
995 if (*directory == '/')
999 if (!dir_matches_p (opt.includes, directory))
1004 if (dir_matches_p (opt.excludes, directory))
1010 /* Return true if STRING ends with TAIL. For instance:
1012 match_tail ("abc", "bc", false) -> 1
1013 match_tail ("abc", "ab", false) -> 0
1014 match_tail ("abc", "abc", false) -> 1
1016 If FOLD_CASE is true, the comparison will be case-insensitive. */
1019 match_tail (const char *string, const char *tail, bool fold_case)
1023 /* We want this to be fast, so we code two loops, one with
1024 case-folding, one without. */
1028 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
1029 if (string[i] != tail[j])
1034 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
1035 if (c_tolower (string[i]) != c_tolower (tail[j]))
1039 /* If the tail was exhausted, the match was succesful. */
1046 /* Checks whether string S matches each element of ACCEPTS. A list
1047 element are matched either with fnmatch() or match_tail(),
1048 according to whether the element contains wildcards or not.
1050 If the BACKWARD is false, don't do backward comparison -- just compare
1053 in_acclist (const char *const *accepts, const char *s, bool backward)
1055 for (; *accepts; accepts++)
1057 if (has_wildcards_p (*accepts))
1059 int res = opt.ignore_case
1060 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
1061 /* fnmatch returns 0 if the pattern *does* match the string. */
1069 if (match_tail (s, *accepts, opt.ignore_case))
1074 int cmp = opt.ignore_case
1075 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
1084 /* Return the location of STR's suffix (file extension). Examples:
1085 suffix ("foo.bar") -> "bar"
1086 suffix ("foo.bar.baz") -> "baz"
1087 suffix ("/foo/bar") -> NULL
1088 suffix ("/foo.bar/baz") -> NULL */
1090 suffix (const char *str)
1094 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
1097 if (str[i++] == '.')
1098 return (char *)str + i;
1103 /* Return true if S contains globbing wildcards (`*', `?', `[' or
1107 has_wildcards_p (const char *s)
1110 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
1115 /* Return true if FNAME ends with a typical HTML suffix. The
1116 following (case-insensitive) suffixes are presumed to be HTML
1121 ?html (`?' matches one character)
1123 #### CAVEAT. This is not necessarily a good indication that FNAME
1124 refers to a file that contains HTML! */
1126 has_html_suffix_p (const char *fname)
1130 if ((suf = suffix (fname)) == NULL)
1132 if (!strcasecmp (suf, "html"))
1134 if (!strcasecmp (suf, "htm"))
1136 if (suf[0] && !strcasecmp (suf + 1, "html"))
1141 /* Read a line from FP and return the pointer to freshly allocated
1142 storage. The storage space is obtained through malloc() and should
1143 be freed with free() when it is no longer needed.
1145 The length of the line is not limited, except by available memory.
1146 The newline character at the end of line is retained. The line is
1147 terminated with a zero character.
1149 After end-of-file is encountered without anything being read, NULL
1150 is returned. NULL is also returned on error. To distinguish
1151 between these two cases, use the stdio function ferror(). */
1154 read_whole_line (FILE *fp)
1158 char *line = xmalloc (bufsize);
1160 while (fgets (line + length, bufsize - length, fp))
1162 length += strlen (line + length);
1164 /* Possible for example when reading from a binary file where
1165 a line begins with \0. */
1168 if (line[length - 1] == '\n')
1171 /* fgets() guarantees to read the whole line, or to use up the
1172 space we've given it. We can double the buffer
1175 line = xrealloc (line, bufsize);
1177 if (length == 0 || ferror (fp))
1182 if (length + 1 < bufsize)
1183 /* Relieve the memory from our exponential greediness. We say
1184 `length + 1' because the terminating \0 is not included in
1185 LENGTH. We don't need to zero-terminate the string ourselves,
1186 though, because fgets() does that. */
1187 line = xrealloc (line, length + 1);
1191 /* Read FILE into memory. A pointer to `struct file_memory' are
1192 returned; use struct element `content' to access file contents, and
1193 the element `length' to know the file length. `content' is *not*
1194 zero-terminated, and you should *not* read or write beyond the [0,
1195 length) range of characters.
1197 After you are done with the file contents, call wget_read_file_free to
1200 Depending on the operating system and the type of file that is
1201 being read, wget_read_file() either mmap's the file into memory, or
1202 reads the file into the core using read().
1204 If file is named "-", fileno(stdin) is used for reading instead.
1205 If you want to read from a real file named "-", use "./-" instead. */
1207 struct file_memory *
1208 wget_read_file (const char *file)
1211 struct file_memory *fm;
1213 bool inhibit_close = false;
1215 /* Some magic in the finest tradition of Perl and its kin: if FILE
1216 is "-", just use stdin. */
1219 fd = fileno (stdin);
1220 inhibit_close = true;
1221 /* Note that we don't inhibit mmap() in this case. If stdin is
1222 redirected from a regular file, mmap() will still work. */
1225 fd = open (file, O_RDONLY);
1228 fm = xnew (struct file_memory);
1233 if (fstat (fd, &buf) < 0)
1235 fm->length = buf.st_size;
1236 /* NOTE: As far as I know, the callers of this function never
1237 modify the file text. Relying on this would enable us to
1238 specify PROT_READ and MAP_SHARED for a marginal gain in
1239 efficiency, but at some cost to generality. */
1240 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
1241 MAP_PRIVATE, fd, 0);
1242 if (fm->content == (char *)MAP_FAILED)
1252 /* The most common reason why mmap() fails is that FD does not point
1253 to a plain file. However, it's also possible that mmap() doesn't
1254 work for a particular type of file. Therefore, whenever mmap()
1255 fails, we just fall back to the regular method. */
1256 #endif /* HAVE_MMAP */
1259 size = 512; /* number of bytes fm->contents can
1260 hold at any given time. */
1261 fm->content = xmalloc (size);
1265 if (fm->length > size / 2)
1267 /* #### I'm not sure whether the whole exponential-growth
1268 thing makes sense with kernel read. On Linux at least,
1269 read() refuses to read more than 4K from a file at a
1270 single chunk anyway. But other Unixes might optimize it
1271 better, and it doesn't *hurt* anything, so I'm leaving
1274 /* Normally, we grow SIZE exponentially to make the number
1275 of calls to read() and realloc() logarithmic in relation
1276 to file size. However, read() can read an amount of data
1277 smaller than requested, and it would be unreasonable to
1278 double SIZE every time *something* was read. Therefore,
1279 we double SIZE only when the length exceeds half of the
1280 entire allocated size. */
1282 fm->content = xrealloc (fm->content, size);
1284 nread = read (fd, fm->content + fm->length, size - fm->length);
1286 /* Successful read. */
1287 fm->length += nread;
1297 if (size > fm->length && fm->length != 0)
1298 /* Due to exponential growth of fm->content, the allocated region
1299 might be much larger than what is actually needed. */
1300 fm->content = xrealloc (fm->content, fm->length);
1307 xfree (fm->content);
1312 /* Release the resources held by FM. Specifically, this calls
1313 munmap() or xfree() on fm->content, depending whether mmap or
1314 malloc/read were used to read in the file. It also frees the
1315 memory needed to hold the FM structure itself. */
1318 wget_read_file_free (struct file_memory *fm)
1323 munmap (fm->content, fm->length);
1328 xfree (fm->content);
1333 /* Free the pointers in a NULL-terminated vector of pointers, then
1334 free the pointer itself. */
1336 free_vec (char **vec)
1347 /* Append vector V2 to vector V1. The function frees V2 and
1348 reallocates V1 (thus you may not use the contents of neither
1349 pointer after the call). If V1 is NULL, V2 is returned. */
1351 merge_vecs (char **v1, char **v2)
1361 /* To avoid j == 0 */
1366 for (i = 0; v1[i]; i++)
1369 for (j = 0; v2[j]; j++)
1371 /* Reallocate v1. */
1372 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1373 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1378 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1379 is allocated as needed. Return the new value of the vector. */
1382 vec_append (char **vec, const char *str)
1384 int cnt; /* count of vector elements, including
1385 the one we're about to append */
1388 for (cnt = 0; vec[cnt]; cnt++)
1394 /* Reallocate the array to fit the new element and the NULL. */
1395 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1396 /* Append a copy of STR to the vector. */
1397 vec[cnt - 1] = xstrdup (str);
1402 /* Sometimes it's useful to create "sets" of strings, i.e. special
1403 hash tables where you want to store strings as keys and merely
1404 query for their existence. Here is a set of utility routines that
1405 makes that transparent. */
1408 string_set_add (struct hash_table *ht, const char *s)
1410 /* First check whether the set element already exists. If it does,
1411 do nothing so that we don't have to free() the old element and
1412 then strdup() a new one. */
1413 if (hash_table_contains (ht, s))
1416 /* We use "1" as value. It provides us a useful and clear arbitrary
1417 value, and it consumes no memory -- the pointers to the same
1418 string "1" will be shared by all the key-value pairs in all `set'
1420 hash_table_put (ht, xstrdup (s), "1");
1423 /* Synonym for hash_table_contains... */
1426 string_set_contains (struct hash_table *ht, const char *s)
1428 return hash_table_contains (ht, s);
1431 /* Convert the specified string set to array. ARRAY should be large
1432 enough to hold hash_table_count(ht) char pointers. */
1434 void string_set_to_array (struct hash_table *ht, char **array)
1436 hash_table_iterator iter;
1437 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1438 *array++ = iter.key;
1441 /* Free the string set. This frees both the storage allocated for
1442 keys and the actual hash table. (hash_table_destroy would only
1443 destroy the hash table.) */
1446 string_set_free (struct hash_table *ht)
1448 hash_table_iterator iter;
1449 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1451 hash_table_destroy (ht);
1454 /* Utility function: simply call xfree() on all keys and values of HT. */
1457 free_keys_and_values (struct hash_table *ht)
1459 hash_table_iterator iter;
1460 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1467 /* Get digit grouping data for thousand separors by calling
1468 localeconv(). The data includes separator string and grouping info
1469 and is cached after the first call to the function.
1471 In locales that don't set a thousand separator (such as the "C"
1472 locale), this forces it to be ",". We are now only showing
1473 thousand separators in one place, so this shouldn't be a problem in
1477 get_grouping_data (const char **sep, const char **grouping)
1479 static const char *cached_sep;
1480 static const char *cached_grouping;
1481 static bool initialized;
1484 /* Get the grouping info from the locale. */
1485 struct lconv *lconv = localeconv ();
1486 cached_sep = lconv->thousands_sep;
1487 cached_grouping = lconv->grouping;
1488 #if ! USE_NLS_PROGRESS_BAR
1489 /* We can't count column widths, so ensure that the separator
1490 * is single-byte only (let check below determine what byte). */
1491 if (strlen(cached_sep) > 1)
1496 /* Many locales (such as "C" or "hr_HR") don't specify
1497 grouping, which we still want to use it for legibility.
1498 In those locales set the sep char to ',', unless that
1499 character is used for decimal point, in which case set it
1501 if (*lconv->decimal_point != ',')
1505 cached_grouping = "\x03";
1510 *grouping = cached_grouping;
1513 /* Return a printed representation of N with thousand separators.
1514 This should respect locale settings, with the exception of the "C"
1515 locale which mandates no separator, but we use one anyway.
1517 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1518 the separators because it's too non-portable, and it's hard to test
1519 for this feature at configure time. Besides, it wouldn't display
1520 separators in the "C" locale, still used by many Unix users. */
1523 with_thousand_seps (wgint n)
1525 static char outbuf[48];
1526 char *p = outbuf + sizeof outbuf;
1528 /* Info received from locale */
1529 const char *grouping, *sep;
1532 /* State information */
1533 int i = 0, groupsize;
1534 const char *atgroup;
1536 bool negative = n < 0;
1538 /* Initialize grouping data. */
1539 get_grouping_data (&sep, &grouping);
1540 seplen = strlen (sep);
1542 groupsize = *atgroup++;
1544 /* This would overflow on WGINT_MIN, but printing negative numbers
1545 is not an important goal of this fuinction. */
1549 /* Write the number into the buffer, backwards, inserting the
1550 separators as necessary. */
1554 *--p = n % 10 + '0';
1558 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1559 if (++i == groupsize)
1564 memcpy (p -= seplen, sep, seplen);
1567 groupsize = *atgroup++;
1576 /* N, a byte quantity, is converted to a human-readable abberviated
1577 form a la sizes printed by `ls -lh'. The result is written to a
1578 static buffer, a pointer to which is returned.
1580 Unlike `with_thousand_seps', this approximates to the nearest unit.
1581 Quoting GNU libit: "Most people visually process strings of 3-4
1582 digits effectively, but longer strings of digits are more prone to
1583 misinterpretation. Hence, converting to an abbreviated form
1584 usually improves readability."
1586 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1587 original computer-related meaning of "powers of 1024". We don't
1588 use the "*bibyte" names invented in 1998, and seldom used in
1589 practice. Wikipedia's entry on "binary prefix" discusses this in
1593 human_readable (HR_NUMTYPE n)
1595 /* These suffixes are compatible with those of GNU `ls -lh'. */
1596 static char powers[] =
1598 'K', /* kilobyte, 2^10 bytes */
1599 'M', /* megabyte, 2^20 bytes */
1600 'G', /* gigabyte, 2^30 bytes */
1601 'T', /* terabyte, 2^40 bytes */
1602 'P', /* petabyte, 2^50 bytes */
1603 'E', /* exabyte, 2^60 bytes */
1608 /* If the quantity is smaller than 1K, just print it. */
1611 snprintf (buf, sizeof (buf), "%d", (int) n);
1615 /* Loop over powers, dividing N with 1024 in each iteration. This
1616 works unchanged for all sizes of wgint, while still avoiding
1617 non-portable `long double' arithmetic. */
1618 for (i = 0; i < countof (powers); i++)
1620 /* At each iteration N is greater than the *subsequent* power.
1621 That way N/1024.0 produces a decimal number in the units of
1623 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1625 double val = n / 1024.0;
1626 /* Print values smaller than 10 with one decimal digits, and
1627 others without any decimals. */
1628 snprintf (buf, sizeof (buf), "%.*f%c",
1629 val < 10 ? 1 : 0, val, powers[i]);
1634 return NULL; /* unreached */
1637 /* Count the digits in the provided number. Used to allocate space
1638 when printing numbers. */
1641 numdigit (wgint number)
1645 ++cnt; /* accomodate '-' */
1646 while ((number /= 10) != 0)
1651 #define PR(mask) *p++ = n / (mask) + '0'
1653 /* DIGITS_<D> is used to print a D-digit number and should be called
1654 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1655 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1656 Recursively this continues until DIGITS_1 is invoked. */
1658 #define DIGITS_1(mask) PR (mask)
1659 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1660 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1661 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1662 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1663 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1664 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1665 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1666 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1667 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1669 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1671 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1672 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1673 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1674 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1675 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1676 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1677 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1678 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1679 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1681 /* Shorthand for casting to wgint. */
1684 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1685 `sprintf(buffer, "%lld", (long long) number)', only typically much
1686 faster and portable to machines without long long.
1688 The speedup may make a difference in programs that frequently
1689 convert numbers to strings. Some implementations of sprintf,
1690 particularly the one in some versions of GNU libc, have been known
1691 to be quite slow when converting integers to strings.
1693 Return the pointer to the location where the terminating zero was
1694 printed. (Equivalent to calling buffer+strlen(buffer) after the
1697 BUFFER should be large enough to accept as many bytes as you expect
1698 the number to take up. On machines with 64-bit wgints the maximum
1699 needed size is 24 bytes. That includes the digits needed for the
1700 largest 64-bit number, the `-' sign in case it's negative, and the
1701 terminating '\0'. */
1704 number_to_string (char *buffer, wgint number)
1709 int last_digit_char = 0;
1711 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1712 /* We are running in a very strange environment. Leave the correct
1713 printing to sprintf. */
1714 p += sprintf (buf, "%j", (intmax_t) (n));
1715 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1721 /* n = -n would overflow because -n would evaluate to a
1722 wgint value larger than WGINT_MAX. Need to make n
1723 smaller and handle the last digit separately. */
1724 int last_digit = n % 10;
1725 /* The sign of n%10 is implementation-defined. */
1727 last_digit_char = '0' - last_digit;
1729 last_digit_char = '0' + last_digit;
1730 /* After n is made smaller, -n will not overflow. */
1738 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1739 way printing any N is fully open-coded without a loop or jump.
1740 (Also see description of DIGITS_*.) */
1742 if (n < 10) DIGITS_1 (1);
1743 else if (n < 100) DIGITS_2 (10);
1744 else if (n < 1000) DIGITS_3 (100);
1745 else if (n < 10000) DIGITS_4 (1000);
1746 else if (n < 100000) DIGITS_5 (10000);
1747 else if (n < 1000000) DIGITS_6 (100000);
1748 else if (n < 10000000) DIGITS_7 (1000000);
1749 else if (n < 100000000) DIGITS_8 (10000000);
1750 else if (n < 1000000000) DIGITS_9 (100000000);
1751 #if SIZEOF_WGINT == 4
1752 /* wgint is 32 bits wide: no number has more than 10 digits. */
1753 else DIGITS_10 (1000000000);
1755 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1756 Constants are constructed by compile-time multiplication to avoid
1757 dealing with different notations for 64-bit constants
1758 (nL/nLL/nI64, depending on the compiler and architecture). */
1759 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1760 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1761 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1762 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1763 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1764 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1765 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1766 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1767 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1768 else DIGITS_19 (1000000000*(W)1000000000);
1771 if (last_digit_char)
1772 *p++ = last_digit_char;
1775 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1782 #undef SPRINTF_WGINT
1805 /* Print NUMBER to a statically allocated string and return a pointer
1806 to the printed representation.
1808 This function is intended to be used in conjunction with printf.
1809 It is hard to portably print wgint values:
1810 a) you cannot use printf("%ld", number) because wgint can be long
1811 long on 32-bit machines with LFS.
1812 b) you cannot use printf("%lld", number) because NUMBER could be
1813 long on 32-bit machines without LFS, or on 64-bit machines,
1814 which do not require LFS. Also, Windows doesn't support %lld.
1815 c) you cannot use printf("%j", (int_max_t) number) because not all
1816 versions of printf support "%j", the most notable being the one
1818 d) you cannot #define WGINT_FMT to the appropriate format and use
1819 printf(WGINT_FMT, number) because that would break translations
1820 for user-visible messages, such as printf("Downloaded: %d
1823 What you should use instead is printf("%s", number_to_static_string
1826 CAVEAT: since the function returns pointers to static data, you
1827 must be careful to copy its result before calling it again.
1828 However, to make it more useful with printf, the function maintains
1829 an internal ring of static buffers to return. That way things like
1830 printf("%s %s", number_to_static_string (num1),
1831 number_to_static_string (num2)) work as expected. Three buffers
1832 are currently used, which means that "%s %s %s" will work, but "%s
1833 %s %s %s" won't. If you need to print more than three wgints,
1834 bump the RING_SIZE (or rethink your message.) */
1837 number_to_static_string (wgint number)
1839 static char ring[RING_SIZE][24];
1841 char *buf = ring[ringpos];
1842 number_to_string (buf, number);
1843 ringpos = (ringpos + 1) % RING_SIZE;
1847 /* Converts the byte to bits format if --report-bps option is enabled
1850 convert_to_bits (wgint num)
1858 /* Determine the width of the terminal we're running on. If that's
1859 not possible, return 0. */
1862 determine_screen_width (void)
1864 /* If there's a way to get the terminal size using POSIX
1865 tcgetattr(), somebody please tell me. */
1870 if (opt.lfilename != NULL)
1873 fd = fileno (stderr);
1874 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1875 return 0; /* most likely ENOTTY */
1878 #elif defined(WINDOWS)
1879 CONSOLE_SCREEN_BUFFER_INFO csbi;
1880 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1882 return csbi.dwSize.X;
1883 #else /* neither TIOCGWINSZ nor WINDOWS */
1885 #endif /* neither TIOCGWINSZ nor WINDOWS */
1888 /* Whether the rnd system (either rand or [dl]rand48) has been
1890 static int rnd_seeded;
1892 /* Return a random number between 0 and MAX-1, inclusive.
1894 If the system does not support lrand48 and MAX is greater than the
1895 value of RAND_MAX+1 on the system, the returned value will be in
1896 the range [0, RAND_MAX]. This may be fixed in a future release.
1897 The random number generator is seeded automatically the first time
1900 This uses lrand48 where available, rand elsewhere. DO NOT use it
1901 for cryptography. It is only meant to be used in situations where
1902 quality of the random numbers returned doesn't really matter. */
1905 random_number (int max)
1910 srand48 ((long) time (NULL) ^ (long) getpid ());
1913 return lrand48 () % max;
1914 #else /* not HAVE_DRAND48 */
1920 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1925 /* Like rand() % max, but uses the high-order bits for better
1926 randomness on architectures where rand() is implemented using a
1927 simple congruential generator. */
1929 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1930 return (int) bounded;
1932 #endif /* not HAVE_DRAND48 */
1935 /* Return a random uniformly distributed floating point number in the
1936 [0, 1) range. Uses drand48 where available, and a really lame
1937 kludge elsewhere. */
1945 srand48 ((long) time (NULL) ^ (long) getpid ());
1949 #else /* not HAVE_DRAND48 */
1950 return ( random_number (10000) / 10000.0
1951 + random_number (10000) / (10000.0 * 10000.0)
1952 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1953 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1954 #endif /* not HAVE_DRAND48 */
1957 /* Implementation of run_with_timeout, a generic timeout-forcing
1958 routine for systems with Unix-like signal handling. */
1960 #ifdef USE_SIGNAL_TIMEOUT
1961 # ifdef HAVE_SIGSETJMP
1962 # define SETJMP(env) sigsetjmp (env, 1)
1964 static sigjmp_buf run_with_timeout_env;
1967 abort_run_with_timeout (int sig)
1969 assert (sig == SIGALRM);
1970 siglongjmp (run_with_timeout_env, -1);
1972 # else /* not HAVE_SIGSETJMP */
1973 # define SETJMP(env) setjmp (env)
1975 static jmp_buf run_with_timeout_env;
1978 abort_run_with_timeout (int sig)
1980 assert (sig == SIGALRM);
1981 /* We don't have siglongjmp to preserve the set of blocked signals;
1982 if we longjumped out of the handler at this point, SIGALRM would
1983 remain blocked. We must unblock it manually. */
1986 sigaddset (&set, SIGALRM);
1987 sigprocmask (SIG_BLOCK, &set, NULL);
1989 /* Now it's safe to longjump. */
1990 longjmp (run_with_timeout_env, -1);
1992 # endif /* not HAVE_SIGSETJMP */
1994 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1995 setitimer where available, alarm otherwise.
1997 TIMEOUT should be non-zero. If the timeout value is so small that
1998 it would be rounded to zero, it is rounded to the least legal value
1999 instead (1us for setitimer, 1s for alarm). That ensures that
2000 SIGALRM will be delivered in all cases. */
2003 alarm_set (double timeout)
2006 /* Use the modern itimer interface. */
2007 struct itimerval itv;
2009 itv.it_value.tv_sec = (long) timeout;
2010 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
2011 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
2012 /* Ensure that we wait for at least the minimum interval.
2013 Specifying zero would mean "wait forever". */
2014 itv.it_value.tv_usec = 1;
2015 setitimer (ITIMER_REAL, &itv, NULL);
2016 #else /* not ITIMER_REAL */
2017 /* Use the old alarm() interface. */
2018 int secs = (int) timeout;
2020 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
2021 because alarm(0) means "never deliver the alarm", i.e. "wait
2022 forever", which is not what someone who specifies a 0.5s
2023 timeout would expect. */
2026 #endif /* not ITIMER_REAL */
2029 /* Cancel the alarm set with alarm_set. */
2035 struct itimerval disable;
2037 setitimer (ITIMER_REAL, &disable, NULL);
2038 #else /* not ITIMER_REAL */
2040 #endif /* not ITIMER_REAL */
2043 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
2044 seconds. Returns true if the function was interrupted with a
2045 timeout, false otherwise.
2047 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
2048 using setitimer() or alarm(). The timeout is enforced by
2049 longjumping out of the SIGALRM handler. This has several
2050 advantages compared to the traditional approach of relying on
2051 signals causing system calls to exit with EINTR:
2053 * The callback function is *forcibly* interrupted after the
2054 timeout expires, (almost) regardless of what it was doing and
2055 whether it was in a syscall. For example, a calculation that
2056 takes a long time is interrupted as reliably as an IO
2059 * It works with both SYSV and BSD signals because it doesn't
2060 depend on the default setting of SA_RESTART.
2062 * It doesn't require special handler setup beyond a simple call
2063 to signal(). (It does use sigsetjmp/siglongjmp, but they're
2066 The only downside is that, if FUN allocates internal resources that
2067 are normally freed prior to exit from the functions, they will be
2068 lost in case of timeout. */
2071 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2081 signal (SIGALRM, abort_run_with_timeout);
2082 if (SETJMP (run_with_timeout_env) != 0)
2084 /* Longjumped out of FUN with a timeout. */
2085 signal (SIGALRM, SIG_DFL);
2088 alarm_set (timeout);
2091 /* Preserve errno in case alarm() or signal() modifies it. */
2092 saved_errno = errno;
2094 signal (SIGALRM, SIG_DFL);
2095 errno = saved_errno;
2100 #else /* not USE_SIGNAL_TIMEOUT */
2103 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
2104 define it under Windows, because Windows has its own version of
2105 run_with_timeout that uses threads. */
2108 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2113 #endif /* not WINDOWS */
2114 #endif /* not USE_SIGNAL_TIMEOUT */
2118 /* Sleep the specified amount of seconds. On machines without
2119 nanosleep(), this may sleep shorter if interrupted by signals. */
2122 xsleep (double seconds)
2124 #ifdef HAVE_NANOSLEEP
2125 /* nanosleep is the preferred interface because it offers high
2126 accuracy and, more importantly, because it allows us to reliably
2127 restart receiving a signal such as SIGWINCH. (There was an
2128 actual Debian bug report about --limit-rate malfunctioning while
2129 the terminal was being resized.) */
2130 struct timespec sleep, remaining;
2131 sleep.tv_sec = (long) seconds;
2132 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
2133 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
2134 /* If nanosleep has been interrupted by a signal, adjust the
2135 sleeping period and return to sleep. */
2137 #elif defined(HAVE_USLEEP)
2138 /* If usleep is available, use it in preference to select. */
2141 /* On some systems, usleep cannot handle values larger than
2142 1,000,000. If the period is larger than that, use sleep
2143 first, then add usleep for subsecond accuracy. */
2145 seconds -= (long) seconds;
2147 usleep (seconds * 1000000);
2148 #else /* fall back select */
2149 /* Note that, although Windows supports select, it can't be used to
2150 implement sleeping because Winsock's select doesn't implement
2151 timeout when it is passed NULL pointers for all fd sets. (But it
2152 does under Cygwin, which implements Unix-compatible select.) */
2153 struct timeval sleep;
2154 sleep.tv_sec = (long) seconds;
2155 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
2156 select (0, NULL, NULL, NULL, &sleep);
2157 /* If select returns -1 and errno is EINTR, it means we were
2158 interrupted by a signal. But without knowing how long we've
2159 actually slept, we can't return to sleep. Using gettimeofday to
2160 track sleeps is slow and unreliable due to clock skew. */
2164 #endif /* not WINDOWS */
2166 /* Encode the octets in DATA of length LENGTH to base64 format,
2167 storing the result to DEST. The output will be zero-terminated,
2168 and must point to a writable buffer of at least
2169 1+BASE64_LENGTH(length) bytes. The function returns the length of
2170 the resulting base64 data, not counting the terminating zero.
2172 This implementation does not emit newlines after 76 characters of
2176 base64_encode (const void *data, int length, char *dest)
2178 /* Conversion table. */
2179 static const char tbl[64] = {
2180 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
2181 'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
2182 'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
2183 'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
2185 /* Access bytes in DATA as unsigned char, otherwise the shifts below
2186 don't work for data with MSB set. */
2187 const unsigned char *s = data;
2188 /* Theoretical ANSI violation when length < 3. */
2189 const unsigned char *end = (const unsigned char *) data + length - 2;
2192 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2193 for (; s < end; s += 3)
2195 *p++ = tbl[s[0] >> 2];
2196 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2197 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2198 *p++ = tbl[s[2] & 0x3f];
2201 /* Pad the result if necessary... */
2205 *p++ = tbl[s[0] >> 2];
2206 *p++ = tbl[(s[0] & 3) << 4];
2211 *p++ = tbl[s[0] >> 2];
2212 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2213 *p++ = tbl[((s[1] & 0xf) << 2)];
2217 /* ...and zero-terminate it. */
2223 /* Store in C the next non-whitespace character from the string, or \0
2224 when end of string is reached. */
2225 #define NEXT_CHAR(c, p) do { \
2226 c = (unsigned char) *p++; \
2227 } while (c_isspace (c))
2229 #define IS_ASCII(c) (((c) & 0x80) == 0)
2231 /* Decode data from BASE64 (a null-terminated string) into memory
2232 pointed to by DEST. DEST is assumed to be large enough to
2233 accomodate the decoded data, which is guaranteed to be no more than
2236 Since DEST is assumed to contain binary data, it is not
2237 NUL-terminated. The function returns the length of the data
2238 written to TO. -1 is returned in case of error caused by malformed
2241 This function originates from Free Recode. */
2244 base64_decode (const char *base64, void *dest)
2246 /* Table of base64 values for first 128 characters. Note that this
2247 assumes ASCII (but so does Wget in other places). */
2248 static const signed char base64_char_to_value[128] =
2250 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
2251 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
2252 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
2253 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
2254 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
2255 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
2256 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
2257 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
2258 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
2259 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
2260 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
2261 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
2262 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
2264 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
2265 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
2267 const char *p = base64;
2273 unsigned long value;
2275 /* Process first byte of a quadruplet. */
2279 if (c == '=' || !IS_BASE64 (c))
2280 return -1; /* illegal char while decoding base64 */
2281 value = BASE64_CHAR_TO_VALUE (c) << 18;
2283 /* Process second byte of a quadruplet. */
2286 return -1; /* premature EOF while decoding base64 */
2287 if (c == '=' || !IS_BASE64 (c))
2288 return -1; /* illegal char while decoding base64 */
2289 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2292 /* Process third byte of a quadruplet. */
2295 return -1; /* premature EOF while decoding base64 */
2297 return -1; /* illegal char while decoding base64 */
2303 return -1; /* premature EOF while decoding base64 */
2305 return -1; /* padding `=' expected but not found */
2309 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2310 *q++ = 0xff & value >> 8;
2312 /* Process fourth byte of a quadruplet. */
2315 return -1; /* premature EOF while decoding base64 */
2319 return -1; /* illegal char while decoding base64 */
2321 value |= BASE64_CHAR_TO_VALUE (c);
2322 *q++ = 0xff & value;
2325 #undef BASE64_CHAR_TO_VALUE
2327 return q - (char *) dest;
2331 /* Compiles the PCRE regex. */
2333 compile_pcre_regex (const char *str)
2337 pcre *regex = pcre_compile (str, 0, &errbuf, &erroffset, 0);
2340 fprintf (stderr, _("Invalid regular expression %s, %s\n"),
2341 quote (str), errbuf);
2348 /* Compiles the POSIX regex. */
2350 compile_posix_regex (const char *str)
2352 regex_t *regex = xmalloc (sizeof (regex_t));
2353 int errcode = regcomp ((regex_t *) regex, str, REG_EXTENDED | REG_NOSUB);
2356 int errbuf_size = regerror (errcode, (regex_t *) regex, NULL, 0);
2357 char *errbuf = xmalloc (errbuf_size);
2358 regerror (errcode, (regex_t *) regex, errbuf, errbuf_size);
2359 fprintf (stderr, _("Invalid regular expression %s, %s\n"),
2360 quote (str), errbuf);
2369 #define OVECCOUNT 30
2370 /* Matches a PCRE regex. */
2372 match_pcre_regex (const void *regex, const char *str)
2374 int l = strlen (str);
2375 int ovector[OVECCOUNT];
2377 int rc = pcre_exec ((pcre *) regex, 0, str, l, 0, 0, ovector, OVECCOUNT);
2378 if (rc == PCRE_ERROR_NOMATCH)
2382 logprintf (LOG_VERBOSE, _("Error while matching %s: %d\n"),
2392 /* Matches a POSIX regex. */
2394 match_posix_regex (const void *regex, const char *str)
2396 int rc = regexec ((regex_t *) regex, str, 0, NULL, 0);
2397 if (rc == REG_NOMATCH)
2403 int errbuf_size = regerror (rc, opt.acceptregex, NULL, 0);
2404 char *errbuf = xmalloc (errbuf_size);
2405 regerror (rc, opt.acceptregex, errbuf, errbuf_size);
2406 logprintf (LOG_VERBOSE, _("Error while matching %s: %d\n"),
2416 /* Simple merge sort for use by stable_sort. Implementation courtesy
2417 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2420 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2421 int (*cmpfun) (const void *, const void *))
2423 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2427 size_t mid = (to + from) / 2;
2428 mergesort_internal (base, temp, size, from, mid, cmpfun);
2429 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2432 for (k = from; (i <= mid) && (j <= to); k++)
2433 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2434 memcpy (ELT (temp, k), ELT (base, i++), size);
2436 memcpy (ELT (temp, k), ELT (base, j++), size);
2438 memcpy (ELT (temp, k++), ELT (base, i++), size);
2440 memcpy (ELT (temp, k++), ELT (base, j++), size);
2441 for (k = from; k <= to; k++)
2442 memcpy (ELT (base, k), ELT (temp, k), size);
2447 /* Stable sort with interface exactly like standard library's qsort.
2448 Uses mergesort internally, allocating temporary storage with
2452 stable_sort (void *base, size_t nmemb, size_t size,
2453 int (*cmpfun) (const void *, const void *))
2457 void *temp = alloca (nmemb * size * sizeof (void *));
2458 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2462 /* Print a decimal number. If it is equal to or larger than ten, the
2463 number is rounded. Otherwise it is printed with one significant
2464 digit without trailing zeros and with no more than three fractional
2465 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2466 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2468 This is useful for displaying durations because it provides
2469 order-of-magnitude information without unnecessary clutter --
2470 long-running downloads are shown without the fractional part, and
2471 short ones still retain one significant digit. */
2474 print_decimal (double number)
2476 static char buf[32];
2477 double n = number >= 0 ? number : -number;
2480 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2481 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2482 snprintf (buf, sizeof buf, "%.0f", number);
2484 snprintf (buf, sizeof buf, "%.1f", number);
2485 else if (n >= 0.001)
2486 snprintf (buf, sizeof buf, "%.1g", number);
2487 else if (n >= 0.0005)
2488 /* round [0.0005, 0.001) to 0.001 */
2489 snprintf (buf, sizeof buf, "%.3f", number);
2491 /* print numbers close to 0 as 0, not 0.000 */
2508 { "/somedir", "/somedir", true },
2509 { "/somedir", "/somedir/d2", true },
2510 { "/somedir/d1", "/somedir", false },
2513 for (i = 0; i < countof(test_array); ++i)
2515 bool res = subdir_p (test_array[i].d1, test_array[i].d2);
2517 mu_assert ("test_subdir_p: wrong result",
2518 res == test_array[i].result);
2525 test_dir_matches_p()
2533 { { "/somedir", "/someotherdir", NULL }, "somedir", true },
2534 { { "/somedir", "/someotherdir", NULL }, "anotherdir", false },
2535 { { "/somedir", "/*otherdir", NULL }, "anotherdir", true },
2536 { { "/somedir/d1", "/someotherdir", NULL }, "somedir/d1", true },
2537 { { "*/*d1", "/someotherdir", NULL }, "somedir/d1", true },
2538 { { "/somedir/d1", "/someotherdir", NULL }, "d1", false },
2539 { { "!COMPLETE", NULL, NULL }, "!COMPLETE", true },
2540 { { "*COMPLETE", NULL, NULL }, "!COMPLETE", true },
2541 { { "*/!COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2542 { { "*COMPLETE", NULL, NULL }, "foo/!COMPLETE", false },
2543 { { "*/*COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2544 { { "/dir with spaces", NULL, NULL }, "dir with spaces", true },
2545 { { "/dir*with*spaces", NULL, NULL }, "dir with spaces", true },
2546 { { "/Tmp/has", NULL, NULL }, "/Tmp/has space", false },
2547 { { "/Tmp/has", NULL, NULL }, "/Tmp/has,comma", false },
2550 for (i = 0; i < countof(test_array); ++i)
2552 bool res = dir_matches_p (test_array[i].dirlist, test_array[i].dir);
2554 mu_assert ("test_dir_matches_p: wrong result",
2555 res == test_array[i].result);
2561 #endif /* TESTING */