1 /* Various utility functions.
2 Copyright (C) 2005 Free Software Foundation, Inc.
4 This file is part of GNU Wget.
6 GNU Wget is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 GNU Wget is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with Wget; if not, write to the Free Software
18 Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 In addition, as a special exception, the Free Software Foundation
21 gives permission to link the code of its release of Wget with the
22 OpenSSL project's "OpenSSL" library (or with modified versions of it
23 that use the same license as the "OpenSSL" library), and distribute
24 the linked executables. You must obey the GNU General Public License
25 in all respects for all of the code used other than "OpenSSL". If you
26 modify this file, you may extend this exception to your version of the
27 file, but you are not obligated to do so. If you do not wish to do
28 so, delete this exception statement from your version. */
36 #else /* not HAVE_STRING_H */
38 #endif /* not HAVE_STRING_H */
39 #include <sys/types.h>
44 # include <sys/mman.h>
55 #ifdef HAVE_SYS_UTIME_H
56 # include <sys/utime.h>
60 # include <libc.h> /* for access() */
64 #ifdef WGET_USE_STDARG
70 /* For TIOCGWINSZ and friends: */
71 #ifdef HAVE_SYS_IOCTL_H
72 # include <sys/ioctl.h>
78 /* Needed for run_with_timeout. */
79 #undef USE_SIGNAL_TIMEOUT
87 #ifndef HAVE_SIGSETJMP
88 /* If sigsetjmp is a macro, configure won't pick it up. */
90 # define HAVE_SIGSETJMP
95 # ifdef HAVE_SIGSETJMP
96 # define USE_SIGNAL_TIMEOUT
99 # define USE_SIGNAL_TIMEOUT
111 /* Utility function: like xstrdup(), but also lowercases S. */
114 xstrdup_lower (const char *s)
116 char *copy = xstrdup (s);
123 /* Copy the string formed by two pointers (one on the beginning, other
124 on the char after the last char) to a new, malloc-ed location.
127 strdupdelim (const char *beg, const char *end)
129 char *res = (char *)xmalloc (end - beg + 1);
130 memcpy (res, beg, end - beg);
131 res[end - beg] = '\0';
135 /* Parse a string containing comma-separated elements, and return a
136 vector of char pointers with the elements. Spaces following the
137 commas are ignored. */
139 sepstring (const char *s)
153 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
154 res[i] = strdupdelim (p, s);
157 /* Skip the blanks following the ','. */
165 res = (char **)xrealloc (res, (i + 2) * sizeof (char *));
166 res[i] = strdupdelim (p, s);
171 #ifdef WGET_USE_STDARG
172 # define VA_START(args, arg1) va_start (args, arg1)
174 # define VA_START(args, ignored) va_start (args)
177 /* Like sprintf, but allocates a string of sufficient size with malloc
178 and returns it. GNU libc has a similar function named asprintf,
179 which requires the pointer to the string to be passed. */
182 aprintf (const char *fmt, ...)
184 /* This function is implemented using vsnprintf, which we provide
185 for the systems that don't have it. Therefore, it should be 100%
189 char *str = xmalloc (size);
196 /* See log_vprintf_internal for explanation why it's OK to rely
197 on the return value of vsnprintf. */
199 VA_START (args, fmt);
200 n = vsnprintf (str, size, fmt, args);
203 /* If the printing worked, return the string. */
204 if (n > -1 && n < size)
207 /* Else try again with a larger buffer. */
208 if (n > -1) /* C99 */
209 size = n + 1; /* precisely what is needed */
211 size <<= 1; /* twice the old size */
212 str = xrealloc (str, size);
214 return NULL; /* unreached */
217 /* Concatenate the NULL-terminated list of string arguments into
218 freshly allocated space. */
221 concat_strings (const char *str0, ...)
224 int saved_lengths[5]; /* inspired by Apache's apr_pstrcat */
227 const char *next_str;
228 int total_length = 0;
231 /* Calculate the length of and allocate the resulting string. */
234 VA_START (args, str0);
235 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
237 int len = strlen (next_str);
238 if (argcount < countof (saved_lengths))
239 saved_lengths[argcount++] = len;
243 p = ret = xmalloc (total_length + 1);
245 /* Copy the strings into the allocated space. */
248 VA_START (args, str0);
249 for (next_str = str0; next_str != NULL; next_str = va_arg (args, char *))
252 if (argcount < countof (saved_lengths))
253 len = saved_lengths[argcount++];
255 len = strlen (next_str);
256 memcpy (p, next_str, len);
265 /* Return pointer to a static char[] buffer in which zero-terminated
266 string-representation of TM (in form hh:mm:ss) is printed.
268 If TM is NULL, the current time will be used. */
271 time_str (time_t *tm)
273 static char output[15];
275 time_t secs = tm ? *tm : time (NULL);
279 /* In case of error, return the empty string. Maybe we should
280 just abort if this happens? */
284 ptm = localtime (&secs);
285 sprintf (output, "%02d:%02d:%02d", ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
289 /* Like the above, but include the date: YYYY-MM-DD hh:mm:ss. */
292 datetime_str (time_t *tm)
294 static char output[20]; /* "YYYY-MM-DD hh:mm:ss" + \0 */
296 time_t secs = tm ? *tm : time (NULL);
300 /* In case of error, return the empty string. Maybe we should
301 just abort if this happens? */
305 ptm = localtime (&secs);
306 sprintf (output, "%04d-%02d-%02d %02d:%02d:%02d",
307 ptm->tm_year + 1900, ptm->tm_mon + 1, ptm->tm_mday,
308 ptm->tm_hour, ptm->tm_min, ptm->tm_sec);
312 /* The Windows versions of the following two functions are defined in
317 fork_to_background (void)
320 /* Whether we arrange our own version of opt.lfilename here. */
321 int logfile_changed = 0;
325 /* We must create the file immediately to avoid either a race
326 condition (which arises from using unique_name and failing to
327 use fopen_excl) or lying to the user about the log file name
328 (which arises from using unique_name, printing the name, and
329 using fopen_excl later on.) */
330 FILE *new_log_fp = unique_create (DEFAULT_LOGFILE, 0, &opt.lfilename);
346 /* parent, no error */
347 printf (_("Continuing in background, pid %d.\n"), (int)pid);
349 printf (_("Output will be written to `%s'.\n"), opt.lfilename);
350 exit (0); /* #### should we use _exit()? */
353 /* child: give up the privileges and keep running. */
355 freopen ("/dev/null", "r", stdin);
356 freopen ("/dev/null", "w", stdout);
357 freopen ("/dev/null", "w", stderr);
359 #endif /* not WINDOWS */
361 /* "Touch" FILE, i.e. make its atime and mtime equal to the time
362 specified with TM. */
364 touch (const char *file, time_t tm)
366 #ifdef HAVE_STRUCT_UTIMBUF
367 struct utimbuf times;
368 times.actime = times.modtime = tm;
371 times[0] = times[1] = tm;
374 if (utime (file, ×) == -1)
375 logprintf (LOG_NOTQUIET, "utime(%s): %s\n", file, strerror (errno));
378 /* Checks if FILE is a symbolic link, and removes it if it is. Does
379 nothing under MS-Windows. */
381 remove_link (const char *file)
386 if (lstat (file, &st) == 0 && S_ISLNK (st.st_mode))
388 DEBUGP (("Unlinking %s (symlink).\n", file));
391 logprintf (LOG_VERBOSE, _("Failed to unlink symlink `%s': %s\n"),
392 file, strerror (errno));
397 /* Does FILENAME exist? This is quite a lousy implementation, since
398 it supplies no error codes -- only a yes-or-no answer. Thus it
399 will return that a file does not exist if, e.g., the directory is
400 unreadable. I don't mind it too much currently, though. The
401 proper way should, of course, be to have a third, error state,
402 other than true/false, but that would introduce uncalled-for
403 additional complexity to the callers. */
405 file_exists_p (const char *filename)
408 return access (filename, F_OK) >= 0;
411 return stat (filename, &buf) >= 0;
415 /* Returns 0 if PATH is a directory, 1 otherwise (any kind of file).
416 Returns 0 on error. */
418 file_non_directory_p (const char *path)
421 /* Use lstat() rather than stat() so that symbolic links pointing to
422 directories can be identified correctly. */
423 if (lstat (path, &buf) != 0)
425 return S_ISDIR (buf.st_mode) ? 0 : 1;
428 /* Return the size of file named by FILENAME, or -1 if it cannot be
429 opened or seeked into. */
431 file_size (const char *filename)
433 #if defined(HAVE_FSEEKO) && defined(HAVE_FTELLO)
435 /* We use fseek rather than stat to determine the file size because
436 that way we can also verify that the file is readable without
437 explicitly checking for permissions. Inspired by the POST patch
439 FILE *fp = fopen (filename, "rb");
442 fseeko (fp, 0, SEEK_END);
448 if (stat (filename, &st) < 0)
454 /* stat file names named PREFIX.1, PREFIX.2, etc., until one that
455 doesn't exist is found. Return a freshly allocated copy of the
459 unique_name_1 (const char *prefix)
462 int plen = strlen (prefix);
463 char *template = (char *)alloca (plen + 1 + 24);
464 char *template_tail = template + plen;
466 memcpy (template, prefix, plen);
467 *template_tail++ = '.';
470 number_to_string (template_tail, count++);
471 while (file_exists_p (template));
473 return xstrdup (template);
476 /* Return a unique file name, based on FILE.
478 More precisely, if FILE doesn't exist, it is returned unmodified.
479 If not, FILE.1 is tried, then FILE.2, etc. The first FILE.<number>
480 file name that doesn't exist is returned.
482 The resulting file is not created, only verified that it didn't
483 exist at the point in time when the function was called.
484 Therefore, where security matters, don't rely that the file created
485 by this function exists until you open it with O_EXCL or
488 If ALLOW_PASSTHROUGH is 0, it always returns a freshly allocated
489 string. Otherwise, it may return FILE if the file doesn't exist
490 (and therefore doesn't need changing). */
493 unique_name (const char *file, int allow_passthrough)
495 /* If the FILE itself doesn't exist, return it without
497 if (!file_exists_p (file))
498 return allow_passthrough ? (char *)file : xstrdup (file);
500 /* Otherwise, find a numeric suffix that results in unused file name
502 return unique_name_1 (file);
505 /* Create a file based on NAME, except without overwriting an existing
506 file with that name. Providing O_EXCL is correctly implemented,
507 this function does not have the race condition associated with
508 opening the file returned by unique_name. */
511 unique_create (const char *name, int binary, char **opened_name)
513 /* unique file name, based on NAME */
514 char *uname = unique_name (name, 0);
516 while ((fp = fopen_excl (uname, binary)) == NULL && errno == EEXIST)
519 uname = unique_name (name, 0);
521 if (opened_name && fp != NULL)
524 *opened_name = uname;
536 /* Open the file for writing, with the addition that the file is
537 opened "exclusively". This means that, if the file already exists,
538 this function will *fail* and errno will be set to EEXIST. If
539 BINARY is set, the file will be opened in binary mode, equivalent
542 If opening the file fails for any reason, including the file having
543 previously existed, this function returns NULL and sets errno
547 fopen_excl (const char *fname, int binary)
551 int flags = O_WRONLY | O_CREAT | O_EXCL;
556 fd = open (fname, flags, 0666);
559 return fdopen (fd, binary ? "wb" : "w");
560 #else /* not O_EXCL */
561 /* Manually check whether the file exists. This is prone to race
562 conditions, but systems without O_EXCL haven't deserved
564 if (file_exists_p (fname))
569 return fopen (fname, binary ? "wb" : "w");
570 #endif /* not O_EXCL */
573 /* Create DIRECTORY. If some of the pathname components of DIRECTORY
574 are missing, create them first. In case any mkdir() call fails,
575 return its error status. Returns 0 on successful completion.
577 The behaviour of this function should be identical to the behaviour
578 of `mkdir -p' on systems where mkdir supports the `-p' option. */
580 make_directory (const char *directory)
582 int i, ret, quit = 0;
585 /* Make a copy of dir, to be able to write to it. Otherwise, the
586 function is unsafe if called with a read-only char *argument. */
587 STRDUP_ALLOCA (dir, directory);
589 /* If the first character of dir is '/', skip it (and thus enable
590 creation of absolute-pathname directories. */
591 for (i = (*dir == '/'); 1; ++i)
593 for (; dir[i] && dir[i] != '/'; i++)
598 /* Check whether the directory already exists. Allow creation of
599 of intermediate directories to fail, as the initial path components
600 are not necessarily directories! */
601 if (!file_exists_p (dir))
602 ret = mkdir (dir, 0777);
613 /* Merge BASE with FILE. BASE can be a directory or a file name, FILE
614 should be a file name.
616 file_merge("/foo/bar", "baz") => "/foo/baz"
617 file_merge("/foo/bar/", "baz") => "/foo/bar/baz"
618 file_merge("foo", "bar") => "bar"
620 In other words, it's a simpler and gentler version of uri_merge_1. */
623 file_merge (const char *base, const char *file)
626 const char *cut = (const char *)strrchr (base, '/');
629 return xstrdup (file);
631 result = (char *)xmalloc (cut - base + 1 + strlen (file) + 1);
632 memcpy (result, base, cut - base);
633 result[cut - base] = '/';
634 strcpy (result + (cut - base) + 1, file);
639 static int in_acclist PARAMS ((const char *const *, const char *, int));
641 /* Determine whether a file is acceptable to be followed, according to
642 lists of patterns to accept/reject. */
644 acceptable (const char *s)
648 while (l && s[l] != '/')
655 return (in_acclist ((const char *const *)opt.accepts, s, 1)
656 && !in_acclist ((const char *const *)opt.rejects, s, 1));
658 return in_acclist ((const char *const *)opt.accepts, s, 1);
660 else if (opt.rejects)
661 return !in_acclist ((const char *const *)opt.rejects, s, 1);
665 /* Compare S1 and S2 frontally; S2 must begin with S1. E.g. if S1 is
666 `/something', frontcmp() will return 1 only if S2 begins with
667 `/something'. Otherwise, 0 is returned. */
669 frontcmp (const char *s1, const char *s2)
671 for (; *s1 && *s2 && (*s1 == *s2); ++s1, ++s2);
675 /* Iterate through STRLIST, and return the first element that matches
676 S, through wildcards or front comparison (as appropriate). */
678 proclist (char **strlist, const char *s, enum accd flags)
682 for (x = strlist; *x; x++)
683 if (has_wildcards_p (*x))
685 if (fnmatch (*x, s, FNM_PATHNAME) == 0)
690 char *p = *x + ((flags & ALLABS) && (**x == '/')); /* Remove '/' */
697 /* Returns whether DIRECTORY is acceptable for download, wrt the
698 include/exclude lists.
700 If FLAGS is ALLABS, the leading `/' is ignored in paths; relative
701 and absolute paths may be freely intermixed. */
703 accdir (const char *directory, enum accd flags)
705 /* Remove starting '/'. */
706 if (flags & ALLABS && *directory == '/')
710 if (!proclist (opt.includes, directory, flags))
715 if (proclist (opt.excludes, directory, flags))
721 /* Return non-zero if STRING ends with TAIL. For instance:
723 match_tail ("abc", "bc", 0) -> 1
724 match_tail ("abc", "ab", 0) -> 0
725 match_tail ("abc", "abc", 0) -> 1
727 If FOLD_CASE_P is non-zero, the comparison will be
731 match_tail (const char *string, const char *tail, int fold_case_p)
735 /* We want this to be fast, so we code two loops, one with
736 case-folding, one without. */
740 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
741 if (string[i] != tail[j])
746 for (i = strlen (string), j = strlen (tail); i >= 0 && j >= 0; i--, j--)
747 if (TOLOWER (string[i]) != TOLOWER (tail[j]))
751 /* If the tail was exhausted, the match was succesful. */
758 /* Checks whether string S matches each element of ACCEPTS. A list
759 element are matched either with fnmatch() or match_tail(),
760 according to whether the element contains wildcards or not.
762 If the BACKWARD is 0, don't do backward comparison -- just compare
765 in_acclist (const char *const *accepts, const char *s, int backward)
767 for (; *accepts; accepts++)
769 if (has_wildcards_p (*accepts))
771 /* fnmatch returns 0 if the pattern *does* match the
773 if (fnmatch (*accepts, s, 0) == 0)
780 if (match_tail (s, *accepts, 0))
785 if (!strcmp (s, *accepts))
793 /* Return the location of STR's suffix (file extension). Examples:
794 suffix ("foo.bar") -> "bar"
795 suffix ("foo.bar.baz") -> "baz"
796 suffix ("/foo/bar") -> NULL
797 suffix ("/foo.bar/baz") -> NULL */
799 suffix (const char *str)
803 for (i = strlen (str); i && str[i] != '/' && str[i] != '.'; i--)
807 return (char *)str + i;
812 /* Return non-zero if S contains globbing wildcards (`*', `?', `[' or
816 has_wildcards_p (const char *s)
819 if (*s == '*' || *s == '?' || *s == '[' || *s == ']')
824 /* Return non-zero if FNAME ends with a typical HTML suffix. The
825 following (case-insensitive) suffixes are presumed to be HTML files:
829 ?html (`?' matches one character)
831 #### CAVEAT. This is not necessarily a good indication that FNAME
832 refers to a file that contains HTML! */
834 has_html_suffix_p (const char *fname)
838 if ((suf = suffix (fname)) == NULL)
840 if (!strcasecmp (suf, "html"))
842 if (!strcasecmp (suf, "htm"))
844 if (suf[0] && !strcasecmp (suf + 1, "html"))
849 /* Read a line from FP and return the pointer to freshly allocated
850 storage. The storage space is obtained through malloc() and should
851 be freed with free() when it is no longer needed.
853 The length of the line is not limited, except by available memory.
854 The newline character at the end of line is retained. The line is
855 terminated with a zero character.
857 After end-of-file is encountered without anything being read, NULL
858 is returned. NULL is also returned on error. To distinguish
859 between these two cases, use the stdio function ferror(). */
862 read_whole_line (FILE *fp)
866 char *line = (char *)xmalloc (bufsize);
868 while (fgets (line + length, bufsize - length, fp))
870 length += strlen (line + length);
872 /* Possible for example when reading from a binary file where
873 a line begins with \0. */
876 if (line[length - 1] == '\n')
879 /* fgets() guarantees to read the whole line, or to use up the
880 space we've given it. We can double the buffer
883 line = xrealloc (line, bufsize);
885 if (length == 0 || ferror (fp))
890 if (length + 1 < bufsize)
891 /* Relieve the memory from our exponential greediness. We say
892 `length + 1' because the terminating \0 is not included in
893 LENGTH. We don't need to zero-terminate the string ourselves,
894 though, because fgets() does that. */
895 line = xrealloc (line, length + 1);
899 /* Read FILE into memory. A pointer to `struct file_memory' are
900 returned; use struct element `content' to access file contents, and
901 the element `length' to know the file length. `content' is *not*
902 zero-terminated, and you should *not* read or write beyond the [0,
903 length) range of characters.
905 After you are done with the file contents, call read_file_free to
908 Depending on the operating system and the type of file that is
909 being read, read_file() either mmap's the file into memory, or
910 reads the file into the core using read().
912 If file is named "-", fileno(stdin) is used for reading instead.
913 If you want to read from a real file named "-", use "./-" instead. */
916 read_file (const char *file)
919 struct file_memory *fm;
921 int inhibit_close = 0;
923 /* Some magic in the finest tradition of Perl and its kin: if FILE
924 is "-", just use stdin. */
929 /* Note that we don't inhibit mmap() in this case. If stdin is
930 redirected from a regular file, mmap() will still work. */
933 fd = open (file, O_RDONLY);
936 fm = xnew (struct file_memory);
941 if (fstat (fd, &buf) < 0)
943 fm->length = buf.st_size;
944 /* NOTE: As far as I know, the callers of this function never
945 modify the file text. Relying on this would enable us to
946 specify PROT_READ and MAP_SHARED for a marginal gain in
947 efficiency, but at some cost to generality. */
948 fm->content = mmap (NULL, fm->length, PROT_READ | PROT_WRITE,
950 if (fm->content == (char *)MAP_FAILED)
960 /* The most common reason why mmap() fails is that FD does not point
961 to a plain file. However, it's also possible that mmap() doesn't
962 work for a particular type of file. Therefore, whenever mmap()
963 fails, we just fall back to the regular method. */
964 #endif /* HAVE_MMAP */
967 size = 512; /* number of bytes fm->contents can
968 hold at any given time. */
969 fm->content = xmalloc (size);
973 if (fm->length > size / 2)
975 /* #### I'm not sure whether the whole exponential-growth
976 thing makes sense with kernel read. On Linux at least,
977 read() refuses to read more than 4K from a file at a
978 single chunk anyway. But other Unixes might optimize it
979 better, and it doesn't *hurt* anything, so I'm leaving
982 /* Normally, we grow SIZE exponentially to make the number
983 of calls to read() and realloc() logarithmic in relation
984 to file size. However, read() can read an amount of data
985 smaller than requested, and it would be unreasonable to
986 double SIZE every time *something* was read. Therefore,
987 we double SIZE only when the length exceeds half of the
988 entire allocated size. */
990 fm->content = xrealloc (fm->content, size);
992 nread = read (fd, fm->content + fm->length, size - fm->length);
994 /* Successful read. */
1005 if (size > fm->length && fm->length != 0)
1006 /* Due to exponential growth of fm->content, the allocated region
1007 might be much larger than what is actually needed. */
1008 fm->content = xrealloc (fm->content, fm->length);
1015 xfree (fm->content);
1020 /* Release the resources held by FM. Specifically, this calls
1021 munmap() or xfree() on fm->content, depending whether mmap or
1022 malloc/read were used to read in the file. It also frees the
1023 memory needed to hold the FM structure itself. */
1026 read_file_free (struct file_memory *fm)
1031 munmap (fm->content, fm->length);
1036 xfree (fm->content);
1041 /* Free the pointers in a NULL-terminated vector of pointers, then
1042 free the pointer itself. */
1044 free_vec (char **vec)
1055 /* Append vector V2 to vector V1. The function frees V2 and
1056 reallocates V1 (thus you may not use the contents of neither
1057 pointer after the call). If V1 is NULL, V2 is returned. */
1059 merge_vecs (char **v1, char **v2)
1069 /* To avoid j == 0 */
1074 for (i = 0; v1[i]; i++);
1076 for (j = 0; v2[j]; j++);
1077 /* Reallocate v1. */
1078 v1 = (char **)xrealloc (v1, (i + j + 1) * sizeof (char **));
1079 memcpy (v1 + i, v2, (j + 1) * sizeof (char *));
1084 /* Sometimes it's useful to create "sets" of strings, i.e. special
1085 hash tables where you want to store strings as keys and merely
1086 query for their existence. Here is a set of utility routines that
1087 makes that transparent. */
1090 string_set_add (struct hash_table *ht, const char *s)
1092 /* First check whether the set element already exists. If it does,
1093 do nothing so that we don't have to free() the old element and
1094 then strdup() a new one. */
1095 if (hash_table_contains (ht, s))
1098 /* We use "1" as value. It provides us a useful and clear arbitrary
1099 value, and it consumes no memory -- the pointers to the same
1100 string "1" will be shared by all the key-value pairs in all `set'
1102 hash_table_put (ht, xstrdup (s), "1");
1105 /* Synonym for hash_table_contains... */
1108 string_set_contains (struct hash_table *ht, const char *s)
1110 return hash_table_contains (ht, s);
1114 string_set_to_array_mapper (void *key, void *value_ignored, void *arg)
1116 char ***arrayptr = (char ***) arg;
1117 *(*arrayptr)++ = (char *) key;
1121 /* Convert the specified string set to array. ARRAY should be large
1122 enough to hold hash_table_count(ht) char pointers. */
1124 void string_set_to_array (struct hash_table *ht, char **array)
1126 hash_table_map (ht, string_set_to_array_mapper, &array);
1130 string_set_free_mapper (void *key, void *value_ignored, void *arg_ignored)
1137 string_set_free (struct hash_table *ht)
1139 hash_table_map (ht, string_set_free_mapper, NULL);
1140 hash_table_destroy (ht);
1144 free_keys_and_values_mapper (void *key, void *value, void *arg_ignored)
1151 /* Another utility function: call free() on all keys and values of HT. */
1154 free_keys_and_values (struct hash_table *ht)
1156 hash_table_map (ht, free_keys_and_values_mapper, NULL);
1160 /* Add thousand separators to a number already in string form. Used
1161 by with_thousand_seps and with_thousand_seps_large. */
1164 add_thousand_seps (const char *repr)
1166 static char outbuf[48];
1171 /* Reset the pointers. */
1175 /* Ignore the sign for the purpose of adding thousand
1182 /* How many digits before the first separator? */
1183 mod = strlen (inptr) % 3;
1185 for (i = 0; i < mod; i++)
1186 *outptr++ = inptr[i];
1187 /* Now insert the rest of them, putting separator before every
1189 for (i1 = i, i = 0; inptr[i1]; i++, i1++)
1191 if (i % 3 == 0 && i1 != 0)
1193 *outptr++ = inptr[i1];
1195 /* Zero-terminate the string. */
1200 /* Return a static pointer to the number printed with thousand
1201 separators inserted at the right places. */
1204 with_thousand_seps (wgint l)
1207 /* Print the number into the buffer. */
1208 number_to_string (inbuf, l);
1209 return add_thousand_seps (inbuf);
1212 /* Write a string representation of LARGE_INT NUMBER into the provided
1215 It would be dangerous to use sprintf, because the code wouldn't
1216 work on a machine with gcc-provided long long support, but without
1217 libc support for "%lld". However, such old systems platforms
1218 typically lack snprintf and will end up using our version, which
1219 does support "%lld" whereever long longs are available. */
1222 large_int_to_string (char *buffer, int bufsize, LARGE_INT number)
1224 snprintf (buffer, bufsize, LARGE_INT_FMT, number);
1227 /* The same as with_thousand_seps, but works on LARGE_INT. */
1230 with_thousand_seps_large (LARGE_INT l)
1233 large_int_to_string (inbuf, sizeof (inbuf), l);
1234 return add_thousand_seps (inbuf);
1237 /* N, a byte quantity, is converted to a human-readable abberviated
1238 form a la sizes printed by `ls -lh'. The result is written to a
1239 static buffer, a pointer to which is returned.
1241 Unlike `with_thousand_seps', this approximates to the nearest unit.
1242 Quoting GNU libit: "Most people visually process strings of 3-4
1243 digits effectively, but longer strings of digits are more prone to
1244 misinterpretation. Hence, converting to an abbreviated form
1245 usually improves readability."
1247 This intentionally uses kilobyte (KB), megabyte (MB), etc. in their
1248 original computer science meaning of "multiples of 1024".
1249 Multiples of 1000 would be useless since Wget already adds thousand
1250 separators for legibility. We don't use the "*bibyte" names
1251 invented in 1998, and seldom used in practice. Wikipedia's entry
1252 on kilobyte discusses this in some detail. */
1255 human_readable (wgint n)
1257 /* These suffixes are compatible with those of GNU `ls -lh'. */
1258 static char powers[] =
1260 'K', /* kilobyte, 2^10 bytes */
1261 'M', /* megabyte, 2^20 bytes */
1262 'G', /* gigabyte, 2^30 bytes */
1263 'T', /* terabyte, 2^40 bytes */
1264 'P', /* petabyte, 2^50 bytes */
1265 'E', /* exabyte, 2^60 bytes */
1270 /* If the quantity is smaller than 1K, just print it. */
1273 snprintf (buf, sizeof (buf), "%d", (int) n);
1277 /* Loop over powers, dividing N with 1024 in each iteration. This
1278 works unchanged for all sizes of wgint, while still avoiding
1279 non-portable `long double' arithmetic. */
1280 for (i = 0; i < countof (powers); i++)
1282 /* At each iteration N is greater than the *subsequent* power.
1283 That way N/1024.0 produces a decimal number in the units of
1285 if ((n >> 10) < 1024 || i == countof (powers) - 1)
1287 /* Must cast to long first because MS VC can't directly cast
1288 __int64 to double. (This is safe because N is known to
1290 double val = (double) (long) n / 1024.0;
1291 /* Print values smaller than 10 with one decimal digits, and
1292 others without any decimals. */
1293 snprintf (buf, sizeof (buf), "%.*f%c",
1294 val < 10 ? 1 : 0, val, powers[i]);
1299 return NULL; /* unreached */
1302 /* Count the digits in the provided number. Used to allocate space
1303 when printing numbers. */
1306 numdigit (wgint number)
1310 ++cnt; /* accomodate '-' */
1311 while ((number /= 10) != 0)
1316 #define PR(mask) *p++ = n / (mask) + '0'
1318 /* DIGITS_<D> is used to print a D-digit number and should be called
1319 with mask==10^(D-1). It prints n/mask (the first digit), reducing
1320 n to n%mask (the remaining digits), and calling DIGITS_<D-1>.
1321 Recursively this continues until DIGITS_1 is invoked. */
1323 #define DIGITS_1(mask) PR (mask)
1324 #define DIGITS_2(mask) PR (mask), n %= (mask), DIGITS_1 ((mask) / 10)
1325 #define DIGITS_3(mask) PR (mask), n %= (mask), DIGITS_2 ((mask) / 10)
1326 #define DIGITS_4(mask) PR (mask), n %= (mask), DIGITS_3 ((mask) / 10)
1327 #define DIGITS_5(mask) PR (mask), n %= (mask), DIGITS_4 ((mask) / 10)
1328 #define DIGITS_6(mask) PR (mask), n %= (mask), DIGITS_5 ((mask) / 10)
1329 #define DIGITS_7(mask) PR (mask), n %= (mask), DIGITS_6 ((mask) / 10)
1330 #define DIGITS_8(mask) PR (mask), n %= (mask), DIGITS_7 ((mask) / 10)
1331 #define DIGITS_9(mask) PR (mask), n %= (mask), DIGITS_8 ((mask) / 10)
1332 #define DIGITS_10(mask) PR (mask), n %= (mask), DIGITS_9 ((mask) / 10)
1334 /* DIGITS_<11-20> are only used on machines with 64-bit wgints. */
1336 #define DIGITS_11(mask) PR (mask), n %= (mask), DIGITS_10 ((mask) / 10)
1337 #define DIGITS_12(mask) PR (mask), n %= (mask), DIGITS_11 ((mask) / 10)
1338 #define DIGITS_13(mask) PR (mask), n %= (mask), DIGITS_12 ((mask) / 10)
1339 #define DIGITS_14(mask) PR (mask), n %= (mask), DIGITS_13 ((mask) / 10)
1340 #define DIGITS_15(mask) PR (mask), n %= (mask), DIGITS_14 ((mask) / 10)
1341 #define DIGITS_16(mask) PR (mask), n %= (mask), DIGITS_15 ((mask) / 10)
1342 #define DIGITS_17(mask) PR (mask), n %= (mask), DIGITS_16 ((mask) / 10)
1343 #define DIGITS_18(mask) PR (mask), n %= (mask), DIGITS_17 ((mask) / 10)
1344 #define DIGITS_19(mask) PR (mask), n %= (mask), DIGITS_18 ((mask) / 10)
1346 /* SPRINTF_WGINT is used by number_to_string to handle pathological
1347 cases and to portably support strange sizes of wgint. Ideally this
1348 would just use "%j" and intmax_t, but many systems don't support
1349 it, so it's used only if nothing else works. */
1350 #if SIZEOF_LONG >= SIZEOF_WGINT
1351 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%ld", (long) (n))
1353 # if SIZEOF_LONG_LONG >= SIZEOF_WGINT
1354 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%lld", (long long) (n))
1357 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%I64", (__int64) (n))
1359 # define SPRINTF_WGINT(buf, n) sprintf (buf, "%j", (intmax_t) (n))
1364 /* Shorthand for casting to wgint. */
1367 /* Print NUMBER to BUFFER in base 10. This is equivalent to
1368 `sprintf(buffer, "%lld", (long long) number)', only typically much
1369 faster and portable to machines without long long.
1371 The speedup may make a difference in programs that frequently
1372 convert numbers to strings. Some implementations of sprintf,
1373 particularly the one in GNU libc, have been known to be extremely
1374 slow when converting integers to strings.
1376 Return the pointer to the location where the terminating zero was
1377 printed. (Equivalent to calling buffer+strlen(buffer) after the
1380 BUFFER should be big enough to accept as many bytes as you expect
1381 the number to take up. On machines with 64-bit longs the maximum
1382 needed size is 24 bytes. That includes the digits needed for the
1383 largest 64-bit number, the `-' sign in case it's negative, and the
1384 terminating '\0'. */
1387 number_to_string (char *buffer, wgint number)
1392 #if (SIZEOF_WGINT != 4) && (SIZEOF_WGINT != 8)
1393 /* We are running in a strange or misconfigured environment. Let
1394 sprintf cope with it. */
1395 SPRINTF_WGINT (buffer, n);
1396 p += strlen (buffer);
1397 #else /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1403 /* -n would overflow. Have sprintf deal with this. */
1404 SPRINTF_WGINT (buffer, n);
1405 p += strlen (buffer);
1413 /* Use the DIGITS_ macro appropriate for N's number of digits. That
1414 way printing any N is fully open-coded without a loop or jump.
1415 (Also see description of DIGITS_*.) */
1417 if (n < 10) DIGITS_1 (1);
1418 else if (n < 100) DIGITS_2 (10);
1419 else if (n < 1000) DIGITS_3 (100);
1420 else if (n < 10000) DIGITS_4 (1000);
1421 else if (n < 100000) DIGITS_5 (10000);
1422 else if (n < 1000000) DIGITS_6 (100000);
1423 else if (n < 10000000) DIGITS_7 (1000000);
1424 else if (n < 100000000) DIGITS_8 (10000000);
1425 else if (n < 1000000000) DIGITS_9 (100000000);
1426 #if SIZEOF_WGINT == 4
1427 /* wgint is 32 bits wide: no number has more than 10 digits. */
1428 else DIGITS_10 (1000000000);
1430 /* wgint is 64 bits wide: handle numbers with more than 9 decimal
1431 digits. Constants are constructed by compile-time multiplication
1432 to avoid dealing with different notations for 64-bit constants
1433 (nnnL, nnnLL, and nnnI64, depending on the compiler). */
1434 else if (n < 10*(W)1000000000) DIGITS_10 (1000000000);
1435 else if (n < 100*(W)1000000000) DIGITS_11 (10*(W)1000000000);
1436 else if (n < 1000*(W)1000000000) DIGITS_12 (100*(W)1000000000);
1437 else if (n < 10000*(W)1000000000) DIGITS_13 (1000*(W)1000000000);
1438 else if (n < 100000*(W)1000000000) DIGITS_14 (10000*(W)1000000000);
1439 else if (n < 1000000*(W)1000000000) DIGITS_15 (100000*(W)1000000000);
1440 else if (n < 10000000*(W)1000000000) DIGITS_16 (1000000*(W)1000000000);
1441 else if (n < 100000000*(W)1000000000) DIGITS_17 (10000000*(W)1000000000);
1442 else if (n < 1000000000*(W)1000000000) DIGITS_18 (100000000*(W)1000000000);
1443 else DIGITS_19 (1000000000*(W)1000000000);
1447 #endif /* (SIZEOF_WGINT == 4) || (SIZEOF_WGINT == 8) */
1476 /* Print NUMBER to a statically allocated string and return a pointer
1477 to the printed representation.
1479 This function is intended to be used in conjunction with printf.
1480 It is hard to portably print wgint values:
1481 a) you cannot use printf("%ld", number) because wgint can be long
1482 long on 32-bit machines with LFS.
1483 b) you cannot use printf("%lld", number) because NUMBER could be
1484 long on 32-bit machines without LFS, or on 64-bit machines,
1485 which do not require LFS. Also, Windows doesn't support %lld.
1486 c) you cannot use printf("%j", (int_max_t) number) because not all
1487 versions of printf support "%j", the most notable being the one
1489 d) you cannot #define WGINT_FMT to the appropriate format and use
1490 printf(WGINT_FMT, number) because that would break translations
1491 for user-visible messages, such as printf("Downloaded: %d
1494 What you should use instead is printf("%s", number_to_static_string
1497 CAVEAT: since the function returns pointers to static data, you
1498 must be careful to copy its result before calling it again.
1499 However, to make it more useful with printf, the function maintains
1500 an internal ring of static buffers to return. That way things like
1501 printf("%s %s", number_to_static_string (num1),
1502 number_to_static_string (num2)) work as expected. Three buffers
1503 are currently used, which means that "%s %s %s" will work, but "%s
1504 %s %s %s" won't. If you need to print more than three wgints,
1505 bump the RING_SIZE (or rethink your message.) */
1508 number_to_static_string (wgint number)
1510 static char ring[RING_SIZE][24];
1512 char *buf = ring[ringpos];
1513 number_to_string (buf, number);
1514 ringpos = (ringpos + 1) % RING_SIZE;
1518 /* Support for timers. */
1520 #undef TIMER_WINDOWS
1521 #undef TIMER_GETTIMEOFDAY
1524 /* Depending on the OS and availability of gettimeofday(), one and
1525 only one of the above constants will be defined. Virtually all
1526 modern Unix systems will define TIMER_GETTIMEOFDAY; Windows will
1527 use TIMER_WINDOWS. TIMER_TIME is a catch-all method for
1528 non-Windows systems without gettimeofday. */
1531 # define TIMER_WINDOWS
1532 #else /* not WINDOWS */
1533 # ifdef HAVE_GETTIMEOFDAY
1534 # define TIMER_GETTIMEOFDAY
1538 #endif /* not WINDOWS */
1540 #ifdef TIMER_GETTIMEOFDAY
1541 typedef struct timeval wget_sys_time;
1545 typedef time_t wget_sys_time;
1548 #ifdef TIMER_WINDOWS
1550 DWORD lores; /* In case GetTickCount is used */
1551 LARGE_INTEGER hires; /* In case high-resolution timer is used */
1556 /* Whether the start time has been initialized. */
1559 /* The starting point in time which, subtracted from the current
1560 time, yields elapsed time. */
1561 wget_sys_time start;
1563 /* The most recent elapsed time, calculated by wtimer_update().
1564 Measured in milliseconds. */
1565 double elapsed_last;
1567 /* Approximately, the time elapsed between the true start of the
1568 measurement and the time represented by START. */
1569 double elapsed_pre_start;
1572 #ifdef TIMER_WINDOWS
1574 /* Whether high-resolution timers are used. Set by wtimer_initialize_once
1575 the first time wtimer_allocate is called. */
1576 static int using_hires_timers;
1578 /* Frequency of high-resolution timers -- number of updates per
1579 millisecond. Calculated the first time wtimer_allocate is called
1580 provided that high-resolution timers are available. */
1581 static double hires_millisec_freq;
1583 /* The first time a timer is created, determine whether to use
1584 high-resolution timers. */
1587 wtimer_initialize_once (void)
1589 static int init_done;
1595 QueryPerformanceFrequency (&freq);
1596 if (freq.QuadPart != 0)
1598 using_hires_timers = 1;
1599 hires_millisec_freq = (double) freq.QuadPart / 1000.0;
1603 #endif /* TIMER_WINDOWS */
1605 /* Allocate a timer. Calling wtimer_read on the timer will return
1606 zero. It is not legal to call wtimer_update with a freshly
1607 allocated timer -- use wtimer_reset first. */
1610 wtimer_allocate (void)
1612 struct wget_timer *wt = xnew (struct wget_timer);
1615 #ifdef TIMER_WINDOWS
1616 wtimer_initialize_once ();
1622 /* Allocate a new timer and reset it. Return the new timer. */
1627 struct wget_timer *wt = wtimer_allocate ();
1632 /* Free the resources associated with the timer. Its further use is
1636 wtimer_delete (struct wget_timer *wt)
1641 /* Store system time to WST. */
1644 wtimer_sys_set (wget_sys_time *wst)
1646 #ifdef TIMER_GETTIMEOFDAY
1647 gettimeofday (wst, NULL);
1654 #ifdef TIMER_WINDOWS
1655 if (using_hires_timers)
1657 QueryPerformanceCounter (&wst->hires);
1661 /* Where hires counters are not available, use GetTickCount rather
1662 GetSystemTime, because it is unaffected by clock skew and simpler
1663 to use. Note that overflows don't affect us because we never use
1664 absolute values of the ticker, only the differences. */
1665 wst->lores = GetTickCount ();
1670 /* Reset timer WT. This establishes the starting point from which
1671 wtimer_read() will return the number of elapsed milliseconds.
1672 It is allowed to reset a previously used timer. */
1675 wtimer_reset (struct wget_timer *wt)
1677 /* Set the start time to the current time. */
1678 wtimer_sys_set (&wt->start);
1679 wt->elapsed_last = 0;
1680 wt->elapsed_pre_start = 0;
1681 wt->initialized = 1;
1685 wtimer_sys_diff (wget_sys_time *wst1, wget_sys_time *wst2)
1687 #ifdef TIMER_GETTIMEOFDAY
1688 return ((double)(wst1->tv_sec - wst2->tv_sec) * 1000
1689 + (double)(wst1->tv_usec - wst2->tv_usec) / 1000);
1693 return 1000 * (*wst1 - *wst2);
1697 if (using_hires_timers)
1698 return (wst1->hires.QuadPart - wst2->hires.QuadPart) / hires_millisec_freq;
1700 return wst1->lores - wst2->lores;
1704 /* Update the timer's elapsed interval. This function causes the
1705 timer to call gettimeofday (or time(), etc.) to update its idea of
1706 current time. To get the elapsed interval in milliseconds, use
1709 This function handles clock skew, i.e. time that moves backwards is
1713 wtimer_update (struct wget_timer *wt)
1718 assert (wt->initialized != 0);
1720 wtimer_sys_set (&now);
1721 elapsed = wt->elapsed_pre_start + wtimer_sys_diff (&now, &wt->start);
1723 /* Ideally we'd just return the difference between NOW and
1724 wt->start. However, the system timer can be set back, and we
1725 could return a value smaller than when we were last called, even
1726 a negative value. Both of these would confuse the callers, which
1727 expect us to return monotonically nondecreasing values.
1729 Therefore: if ELAPSED is smaller than its previous known value,
1730 we reset wt->start to the current time and effectively start
1731 measuring from this point. But since we don't want the elapsed
1732 value to start from zero, we set elapsed_pre_start to the last
1733 elapsed time and increment all future calculations by that
1736 if (elapsed < wt->elapsed_last)
1739 wt->elapsed_pre_start = wt->elapsed_last;
1740 elapsed = wt->elapsed_last;
1743 wt->elapsed_last = elapsed;
1746 /* Return the elapsed time in milliseconds between the last call to
1747 wtimer_reset and the last call to wtimer_update.
1749 A typical use of the timer interface would be:
1751 struct wtimer *timer = wtimer_new ();
1752 ... do something that takes a while ...
1754 double msecs = wtimer_read (); */
1757 wtimer_read (const struct wget_timer *wt)
1759 return wt->elapsed_last;
1762 /* Return the assessed granularity of the timer implementation, in
1763 milliseconds. This is used by code that tries to substitute a
1764 better value for timers that have returned zero. */
1767 wtimer_granularity (void)
1769 #ifdef TIMER_GETTIMEOFDAY
1770 /* Granularity of gettimeofday varies wildly between architectures.
1771 However, it appears that on modern machines it tends to be better
1772 than 1ms. Assume 100 usecs. (Perhaps the configure process
1773 could actually measure this?) */
1781 #ifdef TIMER_WINDOWS
1782 if (using_hires_timers)
1783 return 1.0 / hires_millisec_freq;
1785 return 10; /* according to MSDN */
1789 /* This should probably be at a better place, but it doesn't really
1790 fit into html-parse.c. */
1792 /* The function returns the pointer to the malloc-ed quoted version of
1793 string s. It will recognize and quote numeric and special graphic
1794 entities, as per RFC1866:
1802 No other entities are recognized or replaced. */
1804 html_quote_string (const char *s)
1810 /* Pass through the string, and count the new size. */
1811 for (i = 0; *s; s++, i++)
1814 i += 4; /* `amp;' */
1815 else if (*s == '<' || *s == '>')
1816 i += 3; /* `lt;' and `gt;' */
1817 else if (*s == '\"')
1818 i += 5; /* `quot;' */
1822 res = (char *)xmalloc (i + 1);
1824 for (p = res; *s; s++)
1837 *p++ = (*s == '<' ? 'l' : 'g');
1864 /* Determine the width of the terminal we're running on. If that's
1865 not possible, return 0. */
1868 determine_screen_width (void)
1870 /* If there's a way to get the terminal size using POSIX
1871 tcgetattr(), somebody please tell me. */
1876 if (opt.lfilename != NULL)
1879 fd = fileno (stderr);
1880 if (ioctl (fd, TIOCGWINSZ, &wsz) < 0)
1881 return 0; /* most likely ENOTTY */
1884 #else /* not TIOCGWINSZ */
1886 CONSOLE_SCREEN_BUFFER_INFO csbi;
1887 if (!GetConsoleScreenBufferInfo (GetStdHandle (STD_ERROR_HANDLE), &csbi))
1889 return csbi.dwSize.X;
1890 # else /* neither WINDOWS nor TIOCGWINSZ */
1892 #endif /* neither WINDOWS nor TIOCGWINSZ */
1893 #endif /* not TIOCGWINSZ */
1896 /* Return a random number between 0 and MAX-1, inclusive.
1898 If MAX is greater than the value of RAND_MAX+1 on the system, the
1899 returned value will be in the range [0, RAND_MAX]. This may be
1900 fixed in a future release.
1902 The random number generator is seeded automatically the first time
1905 This uses rand() for portability. It has been suggested that
1906 random() offers better randomness, but this is not required for
1907 Wget, so I chose to go for simplicity and use rand
1910 DO NOT use this for cryptographic purposes. It is only meant to be
1911 used in situations where quality of the random numbers returned
1912 doesn't really matter. */
1915 random_number (int max)
1923 srand (time (NULL));
1928 /* On systems that don't define RAND_MAX, assume it to be 2**15 - 1,
1929 and enforce that assumption by masking other bits. */
1931 # define RAND_MAX 32767
1935 /* This is equivalent to rand() % max, but uses the high-order bits
1936 for better randomness on architecture where rand() is implemented
1937 using a simple congruential generator. */
1939 bounded = (double)max * rnd / (RAND_MAX + 1.0);
1940 return (int)bounded;
1943 /* Return a random uniformly distributed floating point number in the
1944 [0, 1) range. The precision of returned numbers is 9 digits.
1946 Modify this to use erand48() where available! */
1951 /* We can't rely on any specific value of RAND_MAX, but I'm pretty
1952 sure it's greater than 1000. */
1953 int rnd1 = random_number (1000);
1954 int rnd2 = random_number (1000);
1955 int rnd3 = random_number (1000);
1956 return rnd1 / 1000.0 + rnd2 / 1000000.0 + rnd3 / 1000000000.0;
1959 /* Implementation of run_with_timeout, a generic timeout-forcing
1960 routine for systems with Unix-like signal handling. */
1962 #ifdef USE_SIGNAL_TIMEOUT
1963 # ifdef HAVE_SIGSETJMP
1964 # define SETJMP(env) sigsetjmp (env, 1)
1966 static sigjmp_buf run_with_timeout_env;
1969 abort_run_with_timeout (int sig)
1971 assert (sig == SIGALRM);
1972 siglongjmp (run_with_timeout_env, -1);
1974 # else /* not HAVE_SIGSETJMP */
1975 # define SETJMP(env) setjmp (env)
1977 static jmp_buf run_with_timeout_env;
1980 abort_run_with_timeout (int sig)
1982 assert (sig == SIGALRM);
1983 /* We don't have siglongjmp to preserve the set of blocked signals;
1984 if we longjumped out of the handler at this point, SIGALRM would
1985 remain blocked. We must unblock it manually. */
1986 int mask = siggetmask ();
1987 mask &= ~sigmask (SIGALRM);
1990 /* Now it's safe to longjump. */
1991 longjmp (run_with_timeout_env, -1);
1993 # endif /* not HAVE_SIGSETJMP */
1995 /* Arrange for SIGALRM to be delivered in TIMEOUT seconds. This uses
1996 setitimer where available, alarm otherwise.
1998 TIMEOUT should be non-zero. If the timeout value is so small that
1999 it would be rounded to zero, it is rounded to the least legal value
2000 instead (1us for setitimer, 1s for alarm). That ensures that
2001 SIGALRM will be delivered in all cases. */
2004 alarm_set (double timeout)
2007 /* Use the modern itimer interface. */
2008 struct itimerval itv;
2010 itv.it_value.tv_sec = (long) timeout;
2011 itv.it_value.tv_usec = 1000000 * (timeout - (long)timeout);
2012 if (itv.it_value.tv_sec == 0 && itv.it_value.tv_usec == 0)
2013 /* Ensure that we wait for at least the minimum interval.
2014 Specifying zero would mean "wait forever". */
2015 itv.it_value.tv_usec = 1;
2016 setitimer (ITIMER_REAL, &itv, NULL);
2017 #else /* not ITIMER_REAL */
2018 /* Use the old alarm() interface. */
2019 int secs = (int) timeout;
2021 /* Round TIMEOUTs smaller than 1 to 1, not to zero. This is
2022 because alarm(0) means "never deliver the alarm", i.e. "wait
2023 forever", which is not what someone who specifies a 0.5s
2024 timeout would expect. */
2027 #endif /* not ITIMER_REAL */
2030 /* Cancel the alarm set with alarm_set. */
2036 struct itimerval disable;
2038 setitimer (ITIMER_REAL, &disable, NULL);
2039 #else /* not ITIMER_REAL */
2041 #endif /* not ITIMER_REAL */
2044 /* Call FUN(ARG), but don't allow it to run for more than TIMEOUT
2045 seconds. Returns non-zero if the function was interrupted with a
2046 timeout, zero otherwise.
2048 This works by setting up SIGALRM to be delivered in TIMEOUT seconds
2049 using setitimer() or alarm(). The timeout is enforced by
2050 longjumping out of the SIGALRM handler. This has several
2051 advantages compared to the traditional approach of relying on
2052 signals causing system calls to exit with EINTR:
2054 * The callback function is *forcibly* interrupted after the
2055 timeout expires, (almost) regardless of what it was doing and
2056 whether it was in a syscall. For example, a calculation that
2057 takes a long time is interrupted as reliably as an IO
2060 * It works with both SYSV and BSD signals because it doesn't
2061 depend on the default setting of SA_RESTART.
2063 * It doesn't require special handler setup beyond a simple call
2064 to signal(). (It does use sigsetjmp/siglongjmp, but they're
2067 The only downside is that, if FUN allocates internal resources that
2068 are normally freed prior to exit from the functions, they will be
2069 lost in case of timeout. */
2072 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2082 signal (SIGALRM, abort_run_with_timeout);
2083 if (SETJMP (run_with_timeout_env) != 0)
2085 /* Longjumped out of FUN with a timeout. */
2086 signal (SIGALRM, SIG_DFL);
2089 alarm_set (timeout);
2092 /* Preserve errno in case alarm() or signal() modifies it. */
2093 saved_errno = errno;
2095 signal (SIGALRM, SIG_DFL);
2096 errno = saved_errno;
2101 #else /* not USE_SIGNAL_TIMEOUT */
2104 /* A stub version of run_with_timeout that just calls FUN(ARG). Don't
2105 define it under Windows, because Windows has its own version of
2106 run_with_timeout that uses threads. */
2109 run_with_timeout (double timeout, void (*fun) (void *), void *arg)
2114 #endif /* not WINDOWS */
2115 #endif /* not USE_SIGNAL_TIMEOUT */
2119 /* Sleep the specified amount of seconds. On machines without
2120 nanosleep(), this may sleep shorter if interrupted by signals. */
2123 xsleep (double seconds)
2125 #ifdef HAVE_NANOSLEEP
2126 /* nanosleep is the preferred interface because it offers high
2127 accuracy and, more importantly, because it allows us to reliably
2128 restart receiving a signal such as SIGWINCH. (There was an
2129 actual Debian bug report about --limit-rate malfunctioning while
2130 the terminal was being resized.) */
2131 struct timespec sleep, remaining;
2132 sleep.tv_sec = (long) seconds;
2133 sleep.tv_nsec = 1000000000 * (seconds - (long) seconds);
2134 while (nanosleep (&sleep, &remaining) < 0 && errno == EINTR)
2135 /* If nanosleep has been interrupted by a signal, adjust the
2136 sleeping period and return to sleep. */
2138 #else /* not HAVE_NANOSLEEP */
2140 /* If usleep is available, use it in preference to select. */
2143 /* On some systems, usleep cannot handle values larger than
2144 1,000,000. If the period is larger than that, use sleep
2145 first, then add usleep for subsecond accuracy. */
2147 seconds -= (long) seconds;
2149 usleep (seconds * 1000000);
2150 #else /* not HAVE_USLEEP */
2152 /* Note that, although Windows supports select, this sleeping
2153 strategy doesn't work there because Winsock's select doesn't
2154 implement timeout when it is passed NULL pointers for all fd
2155 sets. (But it does work under Cygwin, which implements its own
2157 struct timeval sleep;
2158 sleep.tv_sec = (long) seconds;
2159 sleep.tv_usec = 1000000 * (seconds - (long) seconds);
2160 select (0, NULL, NULL, NULL, &sleep);
2161 /* If select returns -1 and errno is EINTR, it means we were
2162 interrupted by a signal. But without knowing how long we've
2163 actually slept, we can't return to sleep. Using gettimeofday to
2164 track sleeps is slow and unreliable due to clock skew. */
2165 #else /* not HAVE_SELECT */
2167 #endif /* not HAVE_SELECT */
2168 #endif /* not HAVE_USLEEP */
2169 #endif /* not HAVE_NANOSLEEP */
2172 #endif /* not WINDOWS */
2174 /* Encode the string S of length LENGTH to base64 format and place it
2175 to STORE. STORE will be 0-terminated, and must point to a writable
2176 buffer of at least 1+BASE64_LENGTH(length) bytes. */
2179 base64_encode (const char *s, char *store, int length)
2181 /* Conversion table. */
2182 static char tbl[64] = {
2183 'A','B','C','D','E','F','G','H',
2184 'I','J','K','L','M','N','O','P',
2185 'Q','R','S','T','U','V','W','X',
2186 'Y','Z','a','b','c','d','e','f',
2187 'g','h','i','j','k','l','m','n',
2188 'o','p','q','r','s','t','u','v',
2189 'w','x','y','z','0','1','2','3',
2190 '4','5','6','7','8','9','+','/'
2193 unsigned char *p = (unsigned char *)store;
2195 /* Transform the 3x8 bits to 4x6 bits, as required by base64. */
2196 for (i = 0; i < length; i += 3)
2198 *p++ = tbl[s[0] >> 2];
2199 *p++ = tbl[((s[0] & 3) << 4) + (s[1] >> 4)];
2200 *p++ = tbl[((s[1] & 0xf) << 2) + (s[2] >> 6)];
2201 *p++ = tbl[s[2] & 0x3f];
2204 /* Pad the result if necessary... */
2205 if (i == length + 1)
2207 else if (i == length + 2)
2208 *(p - 1) = *(p - 2) = '=';
2209 /* ...and zero-terminate it. */
2213 #define IS_ASCII(c) (((c) & 0x80) == 0)
2214 #define IS_BASE64(c) ((IS_ASCII (c) && base64_char_to_value[c] >= 0) || c == '=')
2216 /* Get next character from the string, except that non-base64
2217 characters are ignored, as mandated by rfc2045. */
2218 #define NEXT_BASE64_CHAR(c, p) do { \
2220 } while (c != '\0' && !IS_BASE64 (c))
2222 /* Decode data from BASE64 (assumed to be encoded as base64) into
2223 memory pointed to by TO. TO should be large enough to accomodate
2224 the decoded data, which is guaranteed to be less than
2227 Since TO is assumed to contain binary data, it is not
2228 NUL-terminated. The function returns the length of the data
2229 written to TO. -1 is returned in case of error caused by malformed
2233 base64_decode (const char *base64, char *to)
2235 /* Table of base64 values for first 128 characters. */
2236 static short base64_char_to_value[128] =
2238 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 0- 9 */
2239 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 10- 19 */
2240 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 20- 29 */
2241 -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, /* 30- 39 */
2242 -1, -1, -1, 62, -1, -1, -1, 63, 52, 53, /* 40- 49 */
2243 54, 55, 56, 57, 58, 59, 60, 61, -1, -1, /* 50- 59 */
2244 -1, -1, -1, -1, -1, 0, 1, 2, 3, 4, /* 60- 69 */
2245 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, /* 70- 79 */
2246 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, /* 80- 89 */
2247 25, -1, -1, -1, -1, -1, -1, 26, 27, 28, /* 90- 99 */
2248 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, /* 100-109 */
2249 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, /* 110-119 */
2250 49, 50, 51, -1, -1, -1, -1, -1 /* 120-127 */
2253 const char *p = base64;
2259 unsigned long value;
2261 /* Process first byte of a quadruplet. */
2262 NEXT_BASE64_CHAR (c, p);
2266 return -1; /* illegal '=' while decoding base64 */
2267 value = base64_char_to_value[c] << 18;
2269 /* Process scond byte of a quadruplet. */
2270 NEXT_BASE64_CHAR (c, p);
2272 return -1; /* premature EOF while decoding base64 */
2274 return -1; /* illegal `=' while decoding base64 */
2275 value |= base64_char_to_value[c] << 12;
2278 /* Process third byte of a quadruplet. */
2279 NEXT_BASE64_CHAR (c, p);
2281 return -1; /* premature EOF while decoding base64 */
2285 NEXT_BASE64_CHAR (c, p);
2287 return -1; /* premature EOF while dcoding base64 */
2289 return -1; /* padding `=' expected but not found */
2293 value |= base64_char_to_value[c] << 6;
2294 *q++ = 0xff & value >> 8;
2296 /* Process fourth byte of a quadruplet. */
2297 NEXT_BASE64_CHAR (c, p);
2299 return -1; /* premature EOF while dcoding base64 */
2303 value |= base64_char_to_value[c];
2304 *q++ = 0xff & value;
2312 #undef NEXT_BASE64_CHAR