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 if (freopen ("/dev/null", "r", stdin) == NULL)
488 DEBUGP (("Failed to redirect stdin to /dev/null.\n"));
489 if (freopen ("/dev/null", "w", stdout) == NULL)
490 DEBUGP (("Failed to redirect stdout to /dev/null.\n"));
491 if (freopen ("/dev/null", "w", stderr) == NULL)
492 DEBUGP (("Failed to redirect stderr to /dev/null.\n"));
494 #endif /* !WINDOWS && !MSDOS */
496 #endif /* def __VMS [else] */
499 /* "Touch" FILE, i.e. make its mtime ("modified time") equal the time
500 specified with TM. The atime ("access time") is set to the current
504 touch (const char *file, time_t tm)
507 # ifdef HAVE_STRUCT_UTIMBUF
508 struct utimbuf times;
516 times.actime = time (NULL);
517 if (utime (file, ×) == -1)
518 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
520 struct timespec timespecs[2];
523 fd = open (file, O_WRONLY);
526 logprintf (LOG_NOTQUIET, "open(%s): %s\n", file, strerror (errno));
530 timespecs[0].tv_sec = time (NULL);
531 timespecs[0].tv_nsec = 0L;
532 timespecs[1].tv_sec = tm;
533 timespecs[1].tv_nsec = 0L;
535 if (futimens (fd, timespecs) == -1)
536 logprintf (LOG_NOTQUIET, "futimens(%s): %s\n", file, strerror (errno));
542 /* Checks if FILE is a symbolic link, and removes it if it is. Does
543 nothing under MS-Windows. */
545 remove_link (const char *file)
550 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
552 DEBUGP (("Unlinking %s (symlink).\n", file));
555 logprintf (LOG_VERBOSE, _("Failed to unlink symlink %s: %s\n"),
556 quote (file), strerror (errno));
561 /* Does FILENAME exist? This is quite a lousy implementation, since
562 it supplies no error codes -- only a yes-or-no answer. Thus it
563 will return that a file does not exist if, e.g., the directory is
564 unreadable. I don't mind it too much currently, though. The
565 proper way should, of course, be to have a third, error state,
566 other than true/false, but that would introduce uncalled-for
567 additional complexity to the callers. */
569 file_exists_p (const char *filename)
572 return access (filename, F_OK) >= 0;
575 return stat (filename, &buf) >= 0;
579 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
580 Returns 0 on error. */
582 file_non_directory_p (const char *path)
585 /* Use lstat() rather than stat() so that symbolic links pointing to
586 directories can be identified correctly. */
587 if (lstat (path, &buf) != 0)
589 return S_ISDIR (buf.st_mode) ? false : true;
592 /* Return the size of file named by FILENAME, or -1 if it cannot be
593 opened or seeked into. */
595 file_size (const char *filename)
597 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
599 /* We use fseek rather than stat to determine the file size because
600 that way we can also verify that the file is readable without
601 explicitly checking for permissions. Inspired by the POST patch
603 FILE *fp = fopen (filename, "rb");
606 fseeko (fp, 0, SEEK_END);
612 if (stat (filename, &st) < 0)
619 If no UNIQ_SEP is defined (as on VMS), have unique_name() return the
620 original name. With the VMS file systems' versioning, everything
621 should be fine, and appending ".NN" just causes trouble.
626 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
627 doesn't exist is found. Return a freshly allocated copy of the
631 unique_name_1 (const char *prefix)
634 int plen = strlen (prefix);
635 char *template = (char *)alloca (plen + 1 + 24);
636 char *template_tail = template + plen;
638 memcpy (template, prefix, plen);
639 *template_tail++ = UNIQ_SEP;
642 number_to_string (template_tail, count++);
643 while (file_exists_p (template));
645 return xstrdup (template);
648 /* Return a unique file name, based on FILE.
650 More precisely, if FILE doesn't exist, it is returned unmodified.
651 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
652 file name that doesn't exist is returned.
654 2005-02-19 SMS. "." is now UNIQ_SEP, and may be different.
656 The resulting file is not created, only verified that it didn't
657 exist at the point in time when the function was called.
658 Therefore, where security matters, don't rely that the file created
659 by this function exists until you open it with O_EXCL or
662 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
663 string. Otherwise, it may return FILE if the file doesn't exist
664 (and therefore doesn't need changing). */
667 unique_name (const char *file, bool allow_passthrough)
669 /* If the FILE itself doesn't exist, return it without
671 if (!file_exists_p (file))
672 return allow_passthrough ? (char *)file : xstrdup (file);
674 /* Otherwise, find a numeric suffix that results in unused file name
676 return unique_name_1 (file);
679 #else /* def UNIQ_SEP */
681 /* Dummy unique_name() for VMS. Return the original name as easily as
685 unique_name (const char *file, bool allow_passthrough)
687 /* Return the FILE itself, without modification, irregardful. */
688 return allow_passthrough ? (char *)file : xstrdup (file);
691 #endif /* def UNIQ_SEP [else] */
693 /* Create a file based on NAME, except without overwriting an existing
694 file with that name. Providing O_EXCL is correctly implemented,
695 this function does not have the race condition associated with
696 opening the file returned by unique_name. */
699 unique_create (const char *name, bool binary, char **opened_name)
701 /* unique file name, based on NAME */
702 char *uname = unique_name (name, false);
704 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
707 uname = unique_name (name, false);
712 *opened_name = uname;
724 /* Open the file for writing, with the addition that the file is
725 opened "exclusively". This means that, if the file already exists,
726 this function will *fail* and errno will be set to EEXIST. If
727 BINARY is set, the file will be opened in binary mode, equivalent
730 If opening the file fails for any reason, including the file having
731 previously existed, this function returns NULL and sets errno
735 fopen_excl (const char *fname, int binary)
741 VMS lacks O_BINARY, but makes up for it in weird and wonderful ways.
742 It also has file versions which obviate all the O_EXCL effort.
743 O_TRUNC (something of a misnomer) requests a new version.
746 /* Common open() optional arguments:
747 sequential access only, access callback function.
749 # define OPEN_OPT_ARGS "fop=sqo", "acc", acc_cb, &open_id
752 int flags = O_WRONLY | O_CREAT | O_TRUNC;
757 fd = open( fname, /* File name. */
759 0777, /* Mode for default protection. */
760 "ctx=bin,stm", /* Binary, stream access. */
761 "rfm=stmlf", /* Stream_LF. */
762 OPEN_OPT_ARGS); /* Access callback. */
767 fd = open( fname, /* File name. */
769 0777, /* Mode for default protection. */
770 "ctx=bin,stm", /* Binary, stream access. */
771 "rfm=fix", /* Fixed-length, */
772 "mrs=512", /* 512-byte records. */
773 OPEN_OPT_ARGS); /* Access callback. */
778 fd = open( fname, /* File name. */
780 0777, /* Mode for default protection. */
781 "rfm=stmlf", /* Stream_LF. */
782 OPEN_OPT_ARGS); /* Access callback. */
784 # else /* def __VMS */
785 int flags = O_WRONLY | O_CREAT | O_EXCL;
790 fd = open (fname, flags, 0666);
791 # endif /* def __VMS [else] */
795 return fdopen (fd, binary ? "wb" : "w");
796 #else /* not O_EXCL */
797 /* Manually check whether the file exists. This is prone to race
798 conditions, but systems without O_EXCL haven't deserved
800 if (file_exists_p (fname))
805 return fopen (fname, binary ? "wb" : "w");
806 #endif /* not O_EXCL */
809 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
810 are missing, create them first. In case any mkdir() call fails,
811 return its error status. Returns 0 on successful completion.
813 The behaviour of this function should be identical to the behaviour
814 of `mkdir -p' on systems where mkdir supports the `-p' option. */
816 make_directory (const char *directory)
818 int i, ret, quit = 0;
821 /* Make a copy of dir, to be able to write to it. Otherwise, the
822 function is unsafe if called with a read-only char *argument. */
823 STRDUP_ALLOCA (dir, directory);
825 /* If the first character of dir is '/', skip it (and thus enable
826 creation of absolute-pathname directories. */
827 for (i = (*dir == '/'); 1; ++i)
829 for (; dir[i] && dir[i] != '/'; i++)
834 /* Check whether the directory already exists. Allow creation of
835 of intermediate directories to fail, as the initial path components
836 are not necessarily directories! */
837 if (!file_exists_p (dir))
838 ret = mkdir (dir, 0777);
849 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
850 should be a file name.
852 file_merge("/foo/bar", "baz") => "/foo/baz"
853 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
854 file_merge("foo", "bar") => "bar"
856 In other words, it's a simpler and gentler version of uri_merge. */
859 file_merge (const char *base, const char *file)
862 const char *cut = (const char *)strrchr (base, '/');
865 return xstrdup (file);
867 result = xmalloc (cut - base + 1 + strlen (file) + 1);
868 memcpy (result, base, cut - base);
869 result[cut - base] = '/';
870 strcpy (result + (cut - base) + 1, file);
875 /* Like fnmatch, but performs a case-insensitive match. */
878 fnmatch_nocase (const char *pattern, const char *string, int flags)
881 /* The FNM_CASEFOLD flag started as a GNU extension, but it is now
882 also present on *BSD platforms, and possibly elsewhere. */
883 return fnmatch (pattern, string, flags | FNM_CASEFOLD);
885 /* Turn PATTERN and STRING to lower case and call fnmatch on them. */
886 char *patcopy = (char *) alloca (strlen (pattern) + 1);
887 char *strcopy = (char *) alloca (strlen (string) + 1);
889 for (p = patcopy; *pattern; pattern++, p++)
890 *p = c_tolower (*pattern);
892 for (p = strcopy; *string; string++, p++)
893 *p = c_tolower (*string);
895 return fnmatch (patcopy, strcopy, flags);
899 static bool in_acclist (const char *const *, const char *, bool);
901 /* Determine whether a file is acceptable to be followed, according to
902 lists of patterns to accept/reject. */
904 acceptable (const char *s)
908 if (opt.output_document && strcmp (s, opt.output_document) == 0)
911 if ((p = strrchr (s, '/')))
917 return (in_acclist ((const char *const *)opt.accepts, s, true)
918 && !in_acclist ((const char *const *)opt.rejects, s, true));
920 return in_acclist ((const char *const *)opt.accepts, s, true);
922 else if (opt.rejects)
923 return !in_acclist ((const char *const *)opt.rejects, s, true);
928 /* Determine whether an URL is acceptable to be followed, according to
929 regex patterns to accept/reject. */
931 accept_url (const char *s)
933 if (opt.acceptregex && !opt.regex_match_fun (opt.acceptregex, s))
935 if (opt.rejectregex && opt.regex_match_fun (opt.rejectregex, s))
941 /* Check if D2 is a subdirectory of D1. E.g. if D1 is `/something', subdir_p()
942 will return true if and only if D2 begins with `/something/' or is exactly
945 subdir_p (const char *d1, const char *d2)
949 if (!opt.ignore_case)
950 for (; *d1 && *d2 && (*d1 == *d2); ++d1, ++d2)
953 for (; *d1 && *d2 && (c_tolower (*d1) == c_tolower (*d2)); ++d1, ++d2)
956 return *d1 == '\0' && (*d2 == '\0' || *d2 == '/');
959 /* Iterate through DIRLIST (which must be NULL-terminated), and return the
960 first element that matches DIR, through wildcards or front comparison (as
963 dir_matches_p (const char **dirlist, const char *dir)
966 int (*matcher) (const char *, const char *, int)
967 = opt.ignore_case ? fnmatch_nocase : fnmatch;
969 for (x = dirlist; *x; x++)
971 /* Remove leading '/' */
972 const char *p = *x + (**x == '/');
973 if (has_wildcards_p (p))
975 if (matcher (p, dir, FNM_PATHNAME) == 0)
980 if (subdir_p (p, dir))
985 return *x ? true : false;
988 /* Returns whether DIRECTORY is acceptable for download, wrt the
989 include/exclude lists.
991 The leading `/' is ignored in paths; relative and absolute paths
992 may be freely intermixed. */
995 accdir (const char *directory)
997 /* Remove starting '/'. */
998 if (*directory == '/')
1002 if (!dir_matches_p (opt.includes, directory))
1007 if (dir_matches_p (opt.excludes, directory))
1013 /* Return true if STRING ends with TAIL. For instance:
1015 match_tail ("abc", "bc", false) -> 1
1016 match_tail ("abc", "ab", false) -> 0
1017 match_tail ("abc", "abc", false) -> 1
1019 If FOLD_CASE is true, the comparison will be case-insensitive. */
1022 match_tail (const char *string, const char *tail, bool fold_case)
1024 int pos = strlen (string) - strlen (tail);
1027 return false; /* tail is longer than string. */
1030 return !strcmp (string + pos, tail);
1032 return !strcasecmp (string + pos, tail);
1035 /* Checks whether string S matches each element of ACCEPTS. A list
1036 element are matched either with fnmatch() or match_tail(),
1037 according to whether the element contains wildcards or not.
1039 If the BACKWARD is false, don't do backward comparison -- just compare
1042 in_acclist (const char *const *accepts, const char *s, bool backward)
1044 for (; *accepts; accepts++)
1046 if (has_wildcards_p (*accepts))
1048 int res = opt.ignore_case
1049 ? fnmatch_nocase (*accepts, s, 0) : fnmatch (*accepts, s, 0);
1050 /* fnmatch returns 0 if the pattern *does* match the string. */
1058 if (match_tail (s, *accepts, opt.ignore_case))
1063 int cmp = opt.ignore_case
1064 ? strcasecmp (s, *accepts) : strcmp (s, *accepts);
1073 /* Return the location of STR's suffix (file extension). Examples:
1074 suffix ("foo.bar") -> "bar"
1075 suffix ("foo.bar.baz") -> "baz"
1076 suffix ("/foo/bar") -> NULL
1077 suffix ("/foo.bar/baz") -> NULL */
1079 suffix (const char *str)
1083 if ((p = strrchr (str, '.')) && !strchr (p + 1, '/'))
1089 /* Return true if S contains globbing wildcards (`*', `?', `[' or
1093 has_wildcards_p (const char *s)
1095 return !!strpbrk (s, "*?[]");
1098 /* Return true if FNAME ends with a typical HTML suffix. The
1099 following (case-insensitive) suffixes are presumed to be HTML
1104 ?html (`?' matches one character)
1106 #### CAVEAT. This is not necessarily a good indication that FNAME
1107 refers to a file that contains HTML! */
1109 has_html_suffix_p (const char *fname)
1113 if ((suf = suffix (fname)) == NULL)
1115 if (!strcasecmp (suf, "html"))
1117 if (!strcasecmp (suf, "htm"))
1119 if (suf[0] && !strcasecmp (suf + 1, "html"))
1124 /* Read FILE into memory. A pointer to `struct file_memory' are
1125 returned; use struct element `content' to access file contents, and
1126 the element `length' to know the file length. `content' is *not*
1127 zero-terminated, and you should *not* read or write beyond the [0,
1128 length) range of characters.
1130 After you are done with the file contents, call wget_read_file_free to
1133 Depending on the operating system and the type of file that is
1134 being read, wget_read_file() either mmap's the file into memory, or
1135 reads the file into the core using read().
1137 If file is named "-", fileno(stdin) is used for reading instead.
1138 If you want to read from a real file named "-", use "./-" instead. */
1140 struct file_memory *
1141 wget_read_file (const char *file)
1144 struct file_memory *fm;
1146 bool inhibit_close = false;
1148 /* Some magic in the finest tradition of Perl and its kin: if FILE
1149 is "-", just use stdin. */
1152 fd = fileno (stdin);
1153 inhibit_close = true;
1154 /* Note that we don't inhibit mmap() in this case. If stdin is
1155 redirected from a regular file, mmap() will still work. */
1158 fd = open (file, O_RDONLY);
1161 fm = xnew (struct file_memory);
1166 if (fstat (fd, &buf) < 0)
1168 fm->length = buf.st_size;
1169 /* NOTE: As far as I know, the callers of this function never
1170 modify the file text. Relying on this would enable us to
1171 specify PROT_READ and MAP_SHARED for a marginal gain in
1172 efficiency, but at some cost to generality. */
1173 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
1174 MAP_PRIVATE, fd, 0);
1175 if (fm->content == (char *)MAP_FAILED)
1185 /* The most common reason why mmap() fails is that FD does not point
1186 to a plain file. However, it's also possible that mmap() doesn't
1187 work for a particular type of file. Therefore, whenever mmap()
1188 fails, we just fall back to the regular method. */
1189 #endif /* HAVE_MMAP */
1192 size = 512; /* number of bytes fm->contents can
1193 hold at any given time. */
1194 fm->content = xmalloc (size);
1198 if (fm->length > size / 2)
1200 /* #### I'm not sure whether the whole exponential-growth
1201 thing makes sense with kernel read. On Linux at least,
1202 read() refuses to read more than 4K from a file at a
1203 single chunk anyway. But other Unixes might optimize it
1204 better, and it doesn't *hurt* anything, so I'm leaving
1207 /* Normally, we grow SIZE exponentially to make the number
1208 of calls to read() and realloc() logarithmic in relation
1209 to file size. However, read() can read an amount of data
1210 smaller than requested, and it would be unreasonable to
1211 double SIZE every time *something* was read. Therefore,
1212 we double SIZE only when the length exceeds half of the
1213 entire allocated size. */
1215 fm->content = xrealloc (fm->content, size);
1217 nread = read (fd, fm->content + fm->length, size - fm->length);
1219 /* Successful read. */
1220 fm->length += nread;
1230 if (size > fm->length && fm->length != 0)
1231 /* Due to exponential growth of fm->content, the allocated region
1232 might be much larger than what is actually needed. */
1233 fm->content = xrealloc (fm->content, fm->length);
1240 xfree (fm->content);
1245 /* Release the resources held by FM. Specifically, this calls
1246 munmap() or xfree() on fm->content, depending whether mmap or
1247 malloc/read were used to read in the file. It also frees the
1248 memory needed to hold the FM structure itself. */
1251 wget_read_file_free (struct file_memory *fm)
1256 munmap (fm->content, fm->length);
1261 xfree (fm->content);
1266 /* Free the pointers in a NULL-terminated vector of pointers, then
1267 free the pointer itself. */
1269 free_vec (char **vec)
1280 /* Append vector V2 to vector V1. The function frees V2 and
1281 reallocates V1 (thus you may not use the contents of neither
1282 pointer after the call). If V1 is NULL, V2 is returned. */
1284 merge_vecs (char **v1, char **v2)
1294 /* To avoid j == 0 */
1299 for (i = 0; v1[i]; i++)
1302 for (j = 0; v2[j]; j++)
1304 /* Reallocate v1. */
1305 v1 = xrealloc (v1, (i + j + 1) * sizeof (char **));
1306 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1311 /* Append a freshly allocated copy of STR to VEC. If VEC is NULL, it
1312 is allocated as needed. Return the new value of the vector. */
1315 vec_append (char **vec, const char *str)
1317 int cnt; /* count of vector elements, including
1318 the one we're about to append */
1321 for (cnt = 0; vec[cnt]; cnt++)
1327 /* Reallocate the array to fit the new element and the NULL. */
1328 vec = xrealloc (vec, (cnt + 1) * sizeof (char *));
1329 /* Append a copy of STR to the vector. */
1330 vec[cnt - 1] = xstrdup (str);
1335 /* Sometimes it's useful to create "sets" of strings, i.e. special
1336 hash tables where you want to store strings as keys and merely
1337 query for their existence. Here is a set of utility routines that
1338 makes that transparent. */
1341 string_set_add (struct hash_table *ht, const char *s)
1343 /* First check whether the set element already exists. If it does,
1344 do nothing so that we don't have to free() the old element and
1345 then strdup() a new one. */
1346 if (hash_table_contains (ht, s))
1349 /* We use "1" as value. It provides us a useful and clear arbitrary
1350 value, and it consumes no memory -- the pointers to the same
1351 string "1" will be shared by all the key-value pairs in all `set'
1353 hash_table_put (ht, xstrdup (s), "1");
1356 /* Synonym for hash_table_contains... */
1359 string_set_contains (struct hash_table *ht, const char *s)
1361 return hash_table_contains (ht, s);
1364 /* Convert the specified string set to array. ARRAY should be large
1365 enough to hold hash_table_count(ht) char pointers. */
1367 void string_set_to_array (struct hash_table *ht, char **array)
1369 hash_table_iterator iter;
1370 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1371 *array++ = iter.key;
1374 /* Free the string set. This frees both the storage allocated for
1375 keys and the actual hash table. (hash_table_destroy would only
1376 destroy the hash table.) */
1379 string_set_free (struct hash_table *ht)
1381 hash_table_iterator iter;
1382 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1384 hash_table_destroy (ht);
1387 /* Utility function: simply call xfree() on all keys and values of HT. */
1390 free_keys_and_values (struct hash_table *ht)
1392 hash_table_iterator iter;
1393 for (hash_table_iterate (ht, &iter); hash_table_iter_next (&iter); )
1400 /* Get digit grouping data for thousand separors by calling
1401 localeconv(). The data includes separator string and grouping info
1402 and is cached after the first call to the function.
1404 In locales that don't set a thousand separator (such as the "C"
1405 locale), this forces it to be ",". We are now only showing
1406 thousand separators in one place, so this shouldn't be a problem in
1410 get_grouping_data (const char **sep, const char **grouping)
1412 static const char *cached_sep;
1413 static const char *cached_grouping;
1414 static bool initialized;
1417 /* Get the grouping info from the locale. */
1418 struct lconv *lconv = localeconv ();
1419 cached_sep = lconv->thousands_sep;
1420 cached_grouping = lconv->grouping;
1421 #if ! USE_NLS_PROGRESS_BAR
1422 /* We can't count column widths, so ensure that the separator
1423 * is single-byte only (let check below determine what byte). */
1424 if (strlen(cached_sep) > 1)
1429 /* Many locales (such as "C" or "hr_HR") don't specify
1430 grouping, which we still want to use it for legibility.
1431 In those locales set the sep char to ',', unless that
1432 character is used for decimal point, in which case set it
1434 if (*lconv->decimal_point != ',')
1438 cached_grouping = "\x03";
1443 *grouping = cached_grouping;
1446 /* Return a printed representation of N with thousand separators.
1447 This should respect locale settings, with the exception of the "C"
1448 locale which mandates no separator, but we use one anyway.
1450 Unfortunately, we cannot use %'d (in fact it would be %'j) to get
1451 the separators because it's too non-portable, and it's hard to test
1452 for this feature at configure time. Besides, it wouldn't display
1453 separators in the "C" locale, still used by many Unix users. */
1456 with_thousand_seps (wgint n)
1458 static char outbuf[48];
1459 char *p = outbuf + sizeof outbuf;
1461 /* Info received from locale */
1462 const char *grouping, *sep;
1465 /* State information */
1466 int i = 0, groupsize;
1467 const char *atgroup;
1469 bool negative = n < 0;
1471 /* Initialize grouping data. */
1472 get_grouping_data (&sep, &grouping);
1473 seplen = strlen (sep);
1475 groupsize = *atgroup++;
1477 /* This would overflow on WGINT_MIN, but printing negative numbers
1478 is not an important goal of this fuinction. */
1482 /* Write the number into the buffer, backwards, inserting the
1483 separators as necessary. */
1487 *--p = n % 10 + '0';
1491 /* Prepend SEP to every groupsize'd digit and get new groupsize. */
1492 if (++i == groupsize)
1497 memcpy (p -= seplen, sep, seplen);
1500 groupsize = *atgroup++;
1509 /* N, a byte quantity, is converted to a human-readable abberviated
1510 form a la sizes printed by `ls -lh'. The result is written to a
1511 static buffer, a pointer to which is returned.
1513 Unlike `with_thousand_seps', this approximates to the nearest unit.
1514 Quoting GNU libit: "Most people visually process strings of 3-4
1515 digits effectively, but longer strings of digits are more prone to
1516 misinterpretation. Hence, converting to an abbreviated form
1517 usually improves readability."
1519 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1520 original computer-related meaning of "powers of 1024". We don't
1521 use the "*bibyte" names invented in 1998, and seldom used in
1522 practice. Wikipedia's entry on "binary prefix" discusses this in
1526 human_readable (HR_NUMTYPE n, const int acc, const int decimals)
1528 /* These suffixes are compatible with those of GNU `ls -lh'. */
1529 static char powers[] =
1531 'K', /* kilobyte, 2^10 bytes */
1532 'M', /* megabyte, 2^20 bytes */
1533 'G', /* gigabyte, 2^30 bytes */
1534 'T', /* terabyte, 2^40 bytes */
1535 'P', /* petabyte, 2^50 bytes */
1536 'E', /* exabyte, 2^60 bytes */
1541 /* If the quantity is smaller than 1K, just print it. */
1544 snprintf (buf, sizeof (buf), "%d", (int) n);
1548 /* Loop over powers, dividing N with 1024 in each iteration. This
1549 works unchanged for all sizes of wgint, while still avoiding
1550 non-portable `long double' arithmetic. */
1551 for (i = 0; i < countof (powers); i++)
1553 /* At each iteration N is greater than the *subsequent* power.
1554 That way N/1024.0 produces a decimal number in the units of
1556 if ((n / 1024) < 1024 || i == countof (powers) - 1)
1558 double val = n / 1024.0;
1559 /* Print values smaller than the accuracy level (acc) with (decimal)
1560 * decimal digits, and others without any decimals. */
1561 snprintf (buf, sizeof (buf), "%.*f%c",
1562 val < acc ? decimals : 0, val, powers[i]);
1567 return NULL; /* unreached */
1570 /* Count the digits in the provided number. Used to allocate space
1571 when printing numbers. */
1574 numdigit (wgint number)
1578 ++cnt; /* accomodate '-' */
1579 while ((number /= 10) != 0)
1584 #define PR(mask) *p++ = n / (mask) + '0'
1586 /* DIGITS_<D> is used to print a D-digit number and should be called
1587 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1588 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1589 Recursively this continues until DIGITS_1 is invoked. */
1591 #define DIGITS_1(mask) PR (mask)
1592 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1593 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1594 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1595 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1596 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1597 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1598 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1599 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1600 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1602 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1604 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1605 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1606 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1607 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1608 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1609 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1610 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1611 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1612 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1614 /* Shorthand for casting to wgint. */
1617 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1618 `sprintf(buffer, "%lld", (long long) number)', only typically much
1619 faster and portable to machines without long long.
1621 The speedup may make a difference in programs that frequently
1622 convert numbers to strings. Some implementations of sprintf,
1623 particularly the one in some versions of GNU libc, have been known
1624 to be quite slow when converting integers to strings.
1626 Return the pointer to the location where the terminating zero was
1627 printed. (Equivalent to calling buffer+strlen(buffer) after the
1630 BUFFER should be large enough to accept as many bytes as you expect
1631 the number to take up. On machines with 64-bit wgints the maximum
1632 needed size is 24 bytes. That includes the digits needed for the
1633 largest 64-bit number, the `-' sign in case it's negative, and the
1634 terminating '\0'. */
1637 number_to_string (char *buffer, wgint number)
1642 int last_digit_char = 0;
1644 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1645 /* We are running in a very strange environment. Leave the correct
1646 printing to sprintf. */
1647 p += sprintf (buf, "%j", (intmax_t) (n));
1648 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1654 /* n = -n would overflow because -n would evaluate to a
1655 wgint value larger than WGINT_MAX. Need to make n
1656 smaller and handle the last digit separately. */
1657 int last_digit = n % 10;
1658 /* The sign of n%10 is implementation-defined. */
1660 last_digit_char = '0' - last_digit;
1662 last_digit_char = '0' + last_digit;
1663 /* After n is made smaller, -n will not overflow. */
1671 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1672 way printing any N is fully open-coded without a loop or jump.
1673 (Also see description of DIGITS_*.) */
1675 if (n < 10) DIGITS_1 (1);
1676 else if (n < 100) DIGITS_2 (10);
1677 else if (n < 1000) DIGITS_3 (100);
1678 else if (n < 10000) DIGITS_4 (1000);
1679 else if (n < 100000) DIGITS_5 (10000);
1680 else if (n < 1000000) DIGITS_6 (100000);
1681 else if (n < 10000000) DIGITS_7 (1000000);
1682 else if (n < 100000000) DIGITS_8 (10000000);
1683 else if (n < 1000000000) DIGITS_9 (100000000);
1684 #if SIZEOF_WGINT == 4
1685 /* wgint is 32 bits wide: no number has more than 10 digits. */
1686 else DIGITS_10 (1000000000);
1688 /* wgint is 64 bits wide: handle numbers with 9-19 decimal digits.
1689 Constants are constructed by compile-time multiplication to avoid
1690 dealing with different notations for 64-bit constants
1691 (nL/nLL/nI64, depending on the compiler and architecture). */
1692 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1693 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1694 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1695 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1696 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1697 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1698 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1699 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1700 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1701 else DIGITS_19 (1000000000*(W)1000000000);
1704 if (last_digit_char)
1705 *p++ = last_digit_char;
1708 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1715 #undef SPRINTF_WGINT
1738 /* Print NUMBER to a statically allocated string and return a pointer
1739 to the printed representation.
1741 This function is intended to be used in conjunction with printf.
1742 It is hard to portably print wgint values:
1743 a) you cannot use printf("%ld", number) because wgint can be long
1744 long on 32-bit machines with LFS.
1745 b) you cannot use printf("%lld", number) because NUMBER could be
1746 long on 32-bit machines without LFS, or on 64-bit machines,
1747 which do not require LFS. Also, Windows doesn't support %lld.
1748 c) you cannot use printf("%j", (int_max_t) number) because not all
1749 versions of printf support "%j", the most notable being the one
1751 d) you cannot #define WGINT_FMT to the appropriate format and use
1752 printf(WGINT_FMT, number) because that would break translations
1753 for user-visible messages, such as printf("Downloaded: %d
1756 What you should use instead is printf("%s", number_to_static_string
1759 CAVEAT: since the function returns pointers to static data, you
1760 must be careful to copy its result before calling it again.
1761 However, to make it more useful with printf, the function maintains
1762 an internal ring of static buffers to return. That way things like
1763 printf("%s %s", number_to_static_string (num1),
1764 number_to_static_string (num2)) work as expected. Three buffers
1765 are currently used, which means that "%s %s %s" will work, but "%s
1766 %s %s %s" won't. If you need to print more than three wgints,
1767 bump the RING_SIZE (or rethink your message.) */
1770 number_to_static_string (wgint number)
1772 static char ring[RING_SIZE][24];
1774 char *buf = ring[ringpos];
1775 number_to_string (buf, number);
1776 ringpos = (ringpos + 1) % RING_SIZE;
1780 /* Converts the byte to bits format if --report-bps option is enabled
1783 convert_to_bits (wgint num)
1791 /* Determine the width of the terminal we're running on. If that's
1792 not possible, return 0. */
1795 determine_screen_width (void)
1797 /* If there's a way to get the terminal size using POSIX
1798 tcgetattr(), somebody please tell me. */
1803 if (opt.lfilename != NULL)
1806 fd = fileno (stderr);
1807 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1808 return 0; /* most likely ENOTTY */
1811 #elif defined(WINDOWS)
1812 CONSOLE_SCREEN_BUFFER_INFO csbi;
1813 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1815 return csbi.dwSize.X;
1816 #else /* neither TIOCGWINSZ nor WINDOWS */
1818 #endif /* neither TIOCGWINSZ nor WINDOWS */
1821 /* Whether the rnd system (either rand or [dl]rand48) has been
1823 static int rnd_seeded;
1825 /* Return a random number between 0 and MAX-1, inclusive.
1827 If the system does not support lrand48 and MAX is greater than the
1828 value of RAND_MAX+1 on the system, the returned value will be in
1829 the range [0, RAND_MAX]. This may be fixed in a future release.
1830 The random number generator is seeded automatically the first time
1833 This uses lrand48 where available, rand elsewhere. DO NOT use it
1834 for cryptography. It is only meant to be used in situations where
1835 quality of the random numbers returned doesn't really matter. */
1838 random_number (int max)
1843 srand48 ((long) time (NULL) ^ (long) getpid ());
1846 return lrand48 () % max;
1847 #else /* not HAVE_DRAND48 */
1853 srand ((unsigned) time (NULL) ^ (unsigned) getpid ());
1858 /* Like rand() % max, but uses the high-order bits for better
1859 randomness on architectures where rand() is implemented using a
1860 simple congruential generator. */
1862 bounded = (double) max * rnd / (RAND_MAX + 1.0);
1863 return (int) bounded;
1865 #endif /* not HAVE_DRAND48 */
1868 /* Return a random uniformly distributed floating point number in the
1869 [0, 1) range. Uses drand48 where available, and a really lame
1870 kludge elsewhere. */
1878 srand48 ((long) time (NULL) ^ (long) getpid ());
1882 #else /* not HAVE_DRAND48 */
1883 return ( random_number (10000) / 10000.0
1884 + random_number (10000) / (10000.0 * 10000.0)
1885 + random_number (10000) / (10000.0 * 10000.0 * 10000.0)
1886 + random_number (10000) / (10000.0 * 10000.0 * 10000.0 * 10000.0));
1887 #endif /* not HAVE_DRAND48 */
1890 /* Implementation of run_with_timeout, a generic timeout-forcing
1891 routine for systems with Unix-like signal handling. */
1893 #ifdef USE_SIGNAL_TIMEOUT
1894 # ifdef HAVE_SIGSETJMP
1895 # define SETJMP(env) sigsetjmp (env, 1)
1897 static sigjmp_buf run_with_timeout_env;
1900 abort_run_with_timeout (int sig)
1902 assert (sig == SIGALRM);
1903 siglongjmp (run_with_timeout_env, -1);
1905 # else /* not HAVE_SIGSETJMP */
1906 # define SETJMP(env) setjmp (env)
1908 static jmp_buf run_with_timeout_env;
1911 abort_run_with_timeout (int sig)
1913 assert (sig == SIGALRM);
1914 /* We don't have siglongjmp to preserve the set of blocked signals;
1915 if we longjumped out of the handler at this point, SIGALRM would
1916 remain blocked. We must unblock it manually. */
1919 sigaddset (&set, SIGALRM);
1920 sigprocmask (SIG_BLOCK, &set, NULL);
1922 /* Now it's safe to longjump. */
1923 longjmp (run_with_timeout_env, -1);
1925 # endif /* not HAVE_SIGSETJMP */
1927 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1928 setitimer where available, alarm otherwise.
1930 TIMEOUT should be non-zero. If the timeout value is so small that
1931 it would be rounded to zero, it is rounded to the least legal value
1932 instead (1us for setitimer, 1s for alarm). That ensures that
1933 SIGALRM will be delivered in all cases. */
1936 alarm_set (double timeout)
1939 /* Use the modern itimer interface. */
1940 struct itimerval itv;
1942 itv.it_value.tv_sec = (long) timeout;
1943 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
1944 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
1945 /* Ensure that we wait for at least the minimum interval.
1946 Specifying zero would mean "wait forever". */
1947 itv.it_value.tv_usec = 1;
1948 setitimer (ITIMER_REAL, &itv, NULL);
1949 #else /* not ITIMER_REAL */
1950 /* Use the old alarm() interface. */
1951 int secs = (int) timeout;
1953 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
1954 because alarm(0) means "never deliver the alarm", i.e. "wait
1955 forever", which is not what someone who specifies a 0.5s
1956 timeout would expect. */
1959 #endif /* not ITIMER_REAL */
1962 /* Cancel the alarm set with alarm_set. */
1968 struct itimerval disable;
1970 setitimer (ITIMER_REAL, &disable, NULL);
1971 #else /* not ITIMER_REAL */
1973 #endif /* not ITIMER_REAL */
1976 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
1977 seconds. Returns true if the function was interrupted with a
1978 timeout, false otherwise.
1980 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
1981 using setitimer() or alarm(). The timeout is enforced by
1982 longjumping out of the SIGALRM handler. This has several
1983 advantages compared to the traditional approach of relying on
1984 signals causing system calls to exit with EINTR:
1986 * The callback function is *forcibly* interrupted after the
1987 timeout expires, (almost) regardless of what it was doing and
1988 whether it was in a syscall. For example, a calculation that
1989 takes a long time is interrupted as reliably as an IO
1992 * It works with both SYSV and BSD signals because it doesn't
1993 depend on the default setting of SA_RESTART.
1995 * It doesn't require special handler setup beyond a simple call
1996 to signal(). (It does use sigsetjmp/siglongjmp, but they're
1999 The only downside is that, if FUN allocates internal resources that
2000 are normally freed prior to exit from the functions, they will be
2001 lost in case of timeout. */
2004 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2014 signal (SIGALRM, abort_run_with_timeout);
2015 if (SETJMP (run_with_timeout_env) != 0)
2017 /* Longjumped out of FUN with a timeout. */
2018 signal (SIGALRM, SIG_DFL);
2021 alarm_set (timeout);
2024 /* Preserve errno in case alarm() or signal() modifies it. */
2025 saved_errno = errno;
2027 signal (SIGALRM, SIG_DFL);
2028 errno = saved_errno;
2033 #else /* not USE_SIGNAL_TIMEOUT */
2036 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
2037 define it under Windows, because Windows has its own version of
2038 run_with_timeout that uses threads. */
2041 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2046 #endif /* not WINDOWS */
2047 #endif /* not USE_SIGNAL_TIMEOUT */
2051 /* Sleep the specified amount of seconds. On machines without
2052 nanosleep(), this may sleep shorter if interrupted by signals. */
2055 xsleep (double seconds)
2057 #ifdef HAVE_NANOSLEEP
2058 /* nanosleep is the preferred interface because it offers high
2059 accuracy and, more importantly, because it allows us to reliably
2060 restart receiving a signal such as SIGWINCH. (There was an
2061 actual Debian bug report about --limit-rate malfunctioning while
2062 the terminal was being resized.) */
2063 struct timespec sleep, remaining;
2064 sleep.tv_sec = (long) seconds;
2065 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
2066 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
2067 /* If nanosleep has been interrupted by a signal, adjust the
2068 sleeping period and return to sleep. */
2070 #elif defined(HAVE_USLEEP)
2071 /* If usleep is available, use it in preference to select. */
2074 /* On some systems, usleep cannot handle values larger than
2075 1,000,000. If the period is larger than that, use sleep
2076 first, then add usleep for subsecond accuracy. */
2078 seconds -= (long) seconds;
2080 usleep (seconds * 1000000);
2081 #else /* fall back select */
2082 /* Note that, although Windows supports select, it can't be used to
2083 implement sleeping because Winsock's select doesn't implement
2084 timeout when it is passed NULL pointers for all fd sets. (But it
2085 does under Cygwin, which implements Unix-compatible select.) */
2086 struct timeval sleep;
2087 sleep.tv_sec = (long) seconds;
2088 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
2089 select (0, NULL, NULL, NULL, &sleep);
2090 /* If select returns -1 and errno is EINTR, it means we were
2091 interrupted by a signal. But without knowing how long we've
2092 actually slept, we can't return to sleep. Using gettimeofday to
2093 track sleeps is slow and unreliable due to clock skew. */
2097 #endif /* not WINDOWS */
2099 /* Encode the octets in DATA of length LENGTH to base64 format,
2100 storing the result to DEST. The output will be zero-terminated,
2101 and must point to a writable buffer of at least
2102 1+BASE64_LENGTH(length) bytes. The function returns the length of
2103 the resulting base64 data, not counting the terminating zero.
2105 This implementation does not emit newlines after 76 characters of
2109 base64_encode (const void *data, size_t length, char *dest)
2111 /* Conversion table. */
2112 static const char tbl[64] = {
2113 'A','B','C','D','E','F','G','H','I','J','K','L','M','N','O','P',
2114 'Q','R','S','T','U','V','W','X','Y','Z','a','b','c','d','e','f',
2115 'g','h','i','j','k','l','m','n','o','p','q','r','s','t','u','v',
2116 'w','x','y','z','0','1','2','3','4','5','6','7','8','9','+','/'
2118 /* Access bytes in DATA as unsigned char, otherwise the shifts below
2119 don't work for data with MSB set. */
2120 const unsigned char *s = data;
2121 /* Theoretical ANSI violation when length < 3. */
2122 const unsigned char *end = (const unsigned char *) data + length - 2;
2125 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2126 for (; s < end; s += 3)
2128 *p++ = tbl[s[0] >> 2];
2129 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2130 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2131 *p++ = tbl[s[2] & 0x3f];
2134 /* Pad the result if necessary... */
2138 *p++ = tbl[s[0] >> 2];
2139 *p++ = tbl[(s[0] & 3) << 4];
2144 *p++ = tbl[s[0] >> 2];
2145 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2146 *p++ = tbl[((s[1] & 0xf) << 2)];
2150 /* ...and zero-terminate it. */
2156 /* Store in C the next non-whitespace character from the string, or \0
2157 when end of string is reached. */
2158 #define NEXT_CHAR(c, p) do { \
2159 c = (unsigned char) *p++; \
2160 } while (c_isspace (c))
2162 #define IS_ASCII(c) (((c) & 0x80) == 0)
2164 /* Decode data from BASE64 (a null-terminated string) into memory
2165 pointed to by DEST. DEST is assumed to be large enough to
2166 accomodate the decoded data, which is guaranteed to be no more than
2169 Since DEST is assumed to contain binary data, it is not
2170 NUL-terminated. The function returns the length of the data
2171 written to TO. -1 is returned in case of error caused by malformed
2174 This function originates from Free Recode. */
2177 base64_decode (const char *base64, void *dest)
2179 /* Table of base64 values for first 128 characters. Note that this
2180 assumes ASCII (but so does Wget in other places). */
2181 static const signed char base64_char_to_value[128] =
2183 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
2184 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
2185 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
2186 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
2187 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
2188 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
2189 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
2190 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
2191 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
2192 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
2193 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
2194 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
2195 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
2197 #define BASE64_CHAR_TO_VALUE(c) ((int) base64_char_to_value[c])
2198 #define IS_BASE64(c) ((IS_ASCII (c) && BASE64_CHAR_TO_VALUE (c) >= 0) || c == '=')
2200 const char *p = base64;
2206 unsigned long value;
2208 /* Process first byte of a quadruplet. */
2212 if (c == '=' || !IS_BASE64 (c))
2213 return -1; /* illegal char while decoding base64 */
2214 value = BASE64_CHAR_TO_VALUE (c) << 18;
2216 /* Process second byte of a quadruplet. */
2219 return -1; /* premature EOF while decoding base64 */
2220 if (c == '=' || !IS_BASE64 (c))
2221 return -1; /* illegal char while decoding base64 */
2222 value |= BASE64_CHAR_TO_VALUE (c) << 12;
2225 /* Process third byte of a quadruplet. */
2228 return -1; /* premature EOF while decoding base64 */
2230 return -1; /* illegal char while decoding base64 */
2236 return -1; /* premature EOF while decoding base64 */
2238 return -1; /* padding `=' expected but not found */
2242 value |= BASE64_CHAR_TO_VALUE (c) << 6;
2243 *q++ = 0xff & value >> 8;
2245 /* Process fourth byte of a quadruplet. */
2248 return -1; /* premature EOF while decoding base64 */
2252 return -1; /* illegal char while decoding base64 */
2254 value |= BASE64_CHAR_TO_VALUE (c);
2255 *q++ = 0xff & value;
2258 #undef BASE64_CHAR_TO_VALUE
2260 return q - (char *) dest;
2264 /* Compiles the PCRE regex. */
2266 compile_pcre_regex (const char *str)
2270 pcre *regex = pcre_compile (str, 0, &errbuf, &erroffset, 0);
2273 fprintf (stderr, _("Invalid regular expression %s, %s\n"),
2274 quote (str), errbuf);
2281 /* Compiles the POSIX regex. */
2283 compile_posix_regex (const char *str)
2285 regex_t *regex = xmalloc (sizeof (regex_t));
2286 int errcode = regcomp ((regex_t *) regex, str, REG_EXTENDED | REG_NOSUB);
2289 size_t errbuf_size = regerror (errcode, (regex_t *) regex, NULL, 0);
2290 char *errbuf = xmalloc (errbuf_size);
2291 regerror (errcode, (regex_t *) regex, errbuf, errbuf_size);
2292 fprintf (stderr, _("Invalid regular expression %s, %s\n"),
2293 quote (str), errbuf);
2302 #define OVECCOUNT 30
2303 /* Matches a PCRE regex. */
2305 match_pcre_regex (const void *regex, const char *str)
2307 size_t l = strlen (str);
2308 int ovector[OVECCOUNT];
2310 int rc = pcre_exec ((pcre *) regex, 0, str, (int) l, 0, 0, ovector, OVECCOUNT);
2311 if (rc == PCRE_ERROR_NOMATCH)
2315 logprintf (LOG_VERBOSE, _("Error while matching %s: %d\n"),
2325 /* Matches a POSIX regex. */
2327 match_posix_regex (const void *regex, const char *str)
2329 int rc = regexec ((regex_t *) regex, str, 0, NULL, 0);
2330 if (rc == REG_NOMATCH)
2336 size_t errbuf_size = regerror (rc, opt.acceptregex, NULL, 0);
2337 char *errbuf = xmalloc (errbuf_size);
2338 regerror (rc, opt.acceptregex, errbuf, errbuf_size);
2339 logprintf (LOG_VERBOSE, _("Error while matching %s: %d\n"),
2349 /* Simple merge sort for use by stable_sort. Implementation courtesy
2350 Zeljko Vrba with additional debugging by Nenad Barbutov. */
2353 mergesort_internal (void *base, void *temp, size_t size, size_t from, size_t to,
2354 int (*cmpfun) (const void *, const void *))
2356 #define ELT(array, pos) ((char *)(array) + (pos) * size)
2360 size_t mid = (to + from) / 2;
2361 mergesort_internal (base, temp, size, from, mid, cmpfun);
2362 mergesort_internal (base, temp, size, mid + 1, to, cmpfun);
2365 for (k = from; (i <= mid) && (j <= to); k++)
2366 if (cmpfun (ELT (base, i), ELT (base, j)) <= 0)
2367 memcpy (ELT (temp, k), ELT (base, i++), size);
2369 memcpy (ELT (temp, k), ELT (base, j++), size);
2371 memcpy (ELT (temp, k++), ELT (base, i++), size);
2373 memcpy (ELT (temp, k++), ELT (base, j++), size);
2374 for (k = from; k <= to; k++)
2375 memcpy (ELT (base, k), ELT (temp, k), size);
2380 /* Stable sort with interface exactly like standard library's qsort.
2381 Uses mergesort internally, allocating temporary storage with
2385 stable_sort (void *base, size_t nmemb, size_t size,
2386 int (*cmpfun) (const void *, const void *))
2390 void *temp = alloca (nmemb * size * sizeof (void *));
2391 mergesort_internal (base, temp, size, 0, nmemb - 1, cmpfun);
2395 /* Print a decimal number. If it is equal to or larger than ten, the
2396 number is rounded. Otherwise it is printed with one significant
2397 digit without trailing zeros and with no more than three fractional
2398 digits total. For example, 0.1 is printed as "0.1", 0.035 is
2399 printed as "0.04", 0.0091 as "0.009", and 0.0003 as simply "0".
2401 This is useful for displaying durations because it provides
2402 order-of-magnitude information without unnecessary clutter --
2403 long-running downloads are shown without the fractional part, and
2404 short ones still retain one significant digit. */
2407 print_decimal (double number)
2409 static char buf[32];
2410 double n = number >= 0 ? number : -number;
2413 /* Cut off at 9.95 because the below %.1f would round 9.96 to
2414 "10.0" instead of "10". OTOH 9.94 will print as "9.9". */
2415 snprintf (buf, sizeof buf, "%.0f", number);
2417 snprintf (buf, sizeof buf, "%.1f", number);
2418 else if (n >= 0.001)
2419 snprintf (buf, sizeof buf, "%.1g", number);
2420 else if (n >= 0.0005)
2421 /* round [0.0005, 0.001) to 0.001 */
2422 snprintf (buf, sizeof buf, "%.3f", number);
2424 /* print numbers close to 0 as 0, not 0.000 */
2430 /* Get the maximum name length for the given path. */
2431 /* Return 0 if length is unknown. */
2433 get_max_length (const char *path, int length, int name)
2438 /* Make a copy of the path that we can modify. */
2439 p = path ? strdupdelim (path, path + length) : strdup ("");
2444 /* For an empty path query the current directory. */
2446 ret = pathconf (*p ? p : ".", name);
2447 if (!(ret < 0 && errno == ENOENT))
2453 /* The path does not exist yet, but may be created. */
2454 /* Already at current or root directory, give up. */
2455 if (!*p || strcmp (p, "/") == 0)
2458 /* Remove one directory level and try again. */
2459 d = strrchr (p, '/');
2461 p[1] = '\0'; /* check root directory */
2463 *d = '\0'; /* remove last directory part */
2465 *p = '\0'; /* check current directory */
2472 /* pathconf() has a message for us. */
2474 perror ("pathconf");
2476 /* If (errno == 0) then there is no max length.
2477 Even on error return 0 so the caller can continue. */
2489 static const struct {
2494 { "/somedir", "/somedir", true },
2495 { "/somedir", "/somedir/d2", true },
2496 { "/somedir/d1", "/somedir", false },
2500 for (i = 0; i < countof(test_array); ++i)
2502 bool res = subdir_p (test_array[i].d1, test_array[i].d2);
2504 mu_assert ("test_subdir_p: wrong result",
2505 res == test_array[i].result);
2512 test_dir_matches_p(void)
2515 const char *dirlist[3];
2519 { { "/somedir", "/someotherdir", NULL }, "somedir", true },
2520 { { "/somedir", "/someotherdir", NULL }, "anotherdir", false },
2521 { { "/somedir", "/*otherdir", NULL }, "anotherdir", true },
2522 { { "/somedir/d1", "/someotherdir", NULL }, "somedir/d1", true },
2523 { { "*/*d1", "/someotherdir", NULL }, "somedir/d1", true },
2524 { { "/somedir/d1", "/someotherdir", NULL }, "d1", false },
2525 { { "!COMPLETE", NULL, NULL }, "!COMPLETE", true },
2526 { { "*COMPLETE", NULL, NULL }, "!COMPLETE", true },
2527 { { "*/!COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2528 { { "*COMPLETE", NULL, NULL }, "foo/!COMPLETE", false },
2529 { { "*/*COMPLETE", NULL, NULL }, "foo/!COMPLETE", true },
2530 { { "/dir with spaces", NULL, NULL }, "dir with spaces", true },
2531 { { "/dir*with*spaces", NULL, NULL }, "dir with spaces", true },
2532 { { "/Tmp/has", NULL, NULL }, "/Tmp/has space", false },
2533 { { "/Tmp/has", NULL, NULL }, "/Tmp/has,comma", false },
2537 for (i = 0; i < countof(test_array); ++i)
2539 bool res = dir_matches_p (test_array[i].dirlist, test_array[i].dir);
2541 mu_assert ("test_dir_matches_p: wrong result",
2542 res == test_array[i].result);
2548 #endif /* TESTING */