1 /* Copyright (c) 2003, Roger Dingledine
2 * Copyright (c) 2004-2006, Roger Dingledine, Nick Mathewson.
3 * Copyright (c) 2007-2008, The Tor Project, Inc. */
4 /* See LICENSE for licensing information */
8 * \brief Common functions for strings, IO, network, data structures,
12 /* This is required on rh7 to make strptime not complain.
21 #include "container.h"
41 #ifdef HAVE_NETINET_IN_H
42 #include <netinet/in.h>
44 #ifdef HAVE_ARPA_INET_H
45 #include <arpa/inet.h>
50 #ifdef HAVE_SYS_SOCKET_H
51 #include <sys/socket.h>
53 #ifdef HAVE_SYS_TIME_H
59 #ifdef HAVE_SYS_STAT_H
62 #ifdef HAVE_SYS_FCNTL_H
63 #include <sys/fcntl.h>
71 #ifdef HAVE_MALLOC_MALLOC_H
72 #include <malloc/malloc.h>
76 /* OpenBSD has a malloc.h, but for our purposes, it only exists in order to
77 * scold us for being so stupid as to autodetect its presence. To be fair,
78 * they've done this since 1996, when autoconf was only 5 years old. */
82 #ifdef HAVE_MALLOC_NP_H
83 #include <malloc_np.h>
92 /* Macro to pass the extra dmalloc args to another function. */
93 #define DMALLOC_FN_ARGS , file, line
95 #if defined(HAVE_DMALLOC_STRDUP)
96 /* the dmalloc_strdup should be fine as defined */
97 #elif defined(HAVE_DMALLOC_STRNDUP)
98 #define dmalloc_strdup(file, line, string, xalloc_b) \
99 dmalloc_strndup(file, line, (string), -1, xalloc_b)
101 #error "No dmalloc_strdup or equivalent"
104 #else /* not using dmalloc */
106 #define DMALLOC_FN_ARGS
109 /** Allocate a chunk of <b>size</b> bytes of memory, and return a pointer to
110 * result. On error, log and terminate the process. (Same as malloc(size),
111 * but never returns NULL.)
113 * <b>file</b> and <b>line</b> are used if dmalloc is enabled, and
117 _tor_malloc(size_t size DMALLOC_PARAMS
)
121 #ifndef MALLOC_ZERO_WORKS
122 /* Some libc mallocs don't work when size==0. Override them. */
129 result
= dmalloc_malloc(file
, line
, size
, DMALLOC_FUNC_MALLOC
, 0, 0);
131 result
= malloc(size
);
134 if (PREDICT_UNLIKELY(result
== NULL
)) {
135 log_err(LD_MM
,"Out of memory on malloc(). Dying.");
136 /* If these functions die within a worker process, they won't call
137 * spawn_exit, but that's ok, since the parent will run out of memory soon
144 /** Allocate a chunk of <b>size</b> bytes of memory, fill the memory with
145 * zero bytes, and return a pointer to the result. Log and terminate
146 * the process on error. (Same as calloc(size,1), but never returns NULL.)
149 _tor_malloc_zero(size_t size DMALLOC_PARAMS
)
151 /* You may ask yourself, "wouldn't it be smart to use calloc instead of
152 * malloc+memset? Perhaps libc's calloc knows some nifty optimization trick
153 * we don't!" Indeed it does, but its optimizations are only a big win when
154 * we're allocating something very big (it knows if it just got the memory
155 * from the OS in a pre-zeroed state). We don't want to use tor_malloc_zero
156 * for big stuff, so we don't bother with calloc. */
157 void *result
= _tor_malloc(size DMALLOC_FN_ARGS
);
158 memset(result
, 0, size
);
162 /** Change the size of the memory block pointed to by <b>ptr</b> to <b>size</b>
163 * bytes long; return the new memory block. On error, log and
164 * terminate. (Like realloc(ptr,size), but never returns NULL.)
167 _tor_realloc(void *ptr
, size_t size DMALLOC_PARAMS
)
172 result
= dmalloc_realloc(file
, line
, ptr
, size
, DMALLOC_FUNC_REALLOC
, 0);
174 result
= realloc(ptr
, size
);
177 if (PREDICT_UNLIKELY(result
== NULL
)) {
178 log_err(LD_MM
,"Out of memory on realloc(). Dying.");
184 /** Return a newly allocated copy of the NUL-terminated string s. On
185 * error, log and terminate. (Like strdup(s), but never returns
189 _tor_strdup(const char *s DMALLOC_PARAMS
)
195 dup
= dmalloc_strdup(file
, line
, s
, 0);
199 if (PREDICT_UNLIKELY(dup
== NULL
)) {
200 log_err(LD_MM
,"Out of memory on strdup(). Dying.");
206 /** Allocate and return a new string containing the first <b>n</b>
207 * characters of <b>s</b>. If <b>s</b> is longer than <b>n</b>
208 * characters, only the first <b>n</b> are copied. The result is
209 * always NUL-terminated. (Like strndup(s,n), but never returns
213 _tor_strndup(const char *s
, size_t n DMALLOC_PARAMS
)
217 dup
= _tor_malloc((n
+1) DMALLOC_FN_ARGS
);
218 /* Performance note: Ordinarily we prefer strlcpy to strncpy. But
219 * this function gets called a whole lot, and platform strncpy is
220 * much faster than strlcpy when strlen(s) is much longer than n.
227 /** Allocate a chunk of <b>len</b> bytes, with the same contents as the
228 * <b>len</b> bytes starting at <b>mem</b>. */
230 _tor_memdup(const void *mem
, size_t len DMALLOC_PARAMS
)
234 dup
= _tor_malloc(len DMALLOC_FN_ARGS
);
235 memcpy(dup
, mem
, len
);
239 /** Helper for places that need to take a function pointer to the right
240 * spelling of "free()". */
247 #if defined(HAVE_MALLOC_GOOD_SIZE) && !defined(HAVE_MALLOC_GOOD_SIZE_PROTOTYPE)
248 /* Some version of Mac OSX have malloc_good_size in their libc, but not
249 * actually defined in malloc/malloc.h. We detect this and work around it by
252 extern size_t malloc_good_size(size_t size
);
255 /** Allocate and return a chunk of memory of size at least *<b>size</b>, using
256 * the same resources we would use to malloc *<b>sizep</b>. Set *<b>sizep</b>
257 * to the number of usable bytes in the chunk of memory. */
259 _tor_malloc_roundup(size_t *sizep DMALLOC_PARAMS
)
261 #ifdef HAVE_MALLOC_GOOD_SIZE
262 *sizep
= malloc_good_size(*sizep
);
263 return _tor_malloc(*sizep DMALLOC_FN_ARGS
);
264 #elif defined(HAVE_MALLOC_USABLE_SIZE) && !defined(USE_DMALLOC)
265 void *result
= _tor_malloc(*sizep DMALLOC_FN_ARGS
);
266 *sizep
= malloc_usable_size(result
);
269 return _tor_malloc(*sizep DMALLOC_FN_ARGS
);
273 /** Call the platform malloc info function, and dump the results to the log at
274 * level <b>severity</b>. If no such function exists, do nothing. */
276 tor_log_mallinfo(int severity
)
280 memset(&mi
, 0, sizeof(mi
));
283 "mallinfo() said: arena=%d, ordblks=%d, smblks=%d, hblks=%d, "
284 "hblkhd=%d, usmblks=%d, fsmblks=%d, uordblks=%d, fordblks=%d, "
286 mi
.arena
, mi
.ordblks
, mi
.smblks
, mi
.hblks
,
287 mi
.hblkhd
, mi
.usmblks
, mi
.fsmblks
, mi
.uordblks
, mi
.fordblks
,
293 dmalloc_log_changed(0, /* Since the program started. */
294 1, /* Log info about non-freed pointers. */
295 0, /* Do not log info about freed pointers. */
296 0 /* Do not log individual pointers. */
305 /** Returns floor(log2(u64)). If u64 is 0, (incorrectly) returns 0. */
307 tor_log2(uint64_t u64
)
310 if (u64
>= (U64_LITERAL(1)<<32)) {
314 if (u64
>= (U64_LITERAL(1)<<16)) {
318 if (u64
>= (U64_LITERAL(1)<<8)) {
322 if (u64
>= (U64_LITERAL(1)<<4)) {
326 if (u64
>= (U64_LITERAL(1)<<2)) {
330 if (u64
>= (U64_LITERAL(1)<<1)) {
337 /** Return the power of 2 closest to <b>u64</b>. */
339 round_to_power_of_2(uint64_t u64
)
341 int lg2
= tor_log2(u64
);
342 uint64_t low
= U64_LITERAL(1) << lg2
, high
= U64_LITERAL(1) << (lg2
+1);
343 if (high
- u64
< u64
- low
)
350 * String manipulation
353 /** Remove from the string <b>s</b> every character which appears in
356 tor_strstrip(char *s
, const char *strip
)
360 if (strchr(strip
, *read
)) {
369 /** Return a pointer to a NUL-terminated hexadecimal string encoding
370 * the first <b>fromlen</b> bytes of <b>from</b>. (fromlen must be \<= 32.) The
371 * result does not need to be deallocated, but repeated calls to
372 * hex_str will trash old results.
375 hex_str(const char *from
, size_t fromlen
)
378 if (fromlen
>(sizeof(buf
)-1)/2)
379 fromlen
= (sizeof(buf
)-1)/2;
380 base16_encode(buf
,sizeof(buf
),from
,fromlen
);
384 /** Convert all alphabetic characters in the nul-terminated string <b>s</b> to
387 tor_strlower(char *s
)
390 *s
= TOR_TOLOWER(*s
);
395 /** Convert all alphabetic characters in the nul-terminated string <b>s</b> to
398 tor_strupper(char *s
)
401 *s
= TOR_TOUPPER(*s
);
406 /** Return 1 if every character in <b>s</b> is printable, else return 0.
409 tor_strisprint(const char *s
)
412 if (!TOR_ISPRINT(*s
))
419 /** Return 1 if no character in <b>s</b> is uppercase, else return 0.
422 tor_strisnonupper(const char *s
)
432 /** Compares the first strlen(s2) characters of s1 with s2. Returns as for
436 strcmpstart(const char *s1
, const char *s2
)
438 size_t n
= strlen(s2
);
439 return strncmp(s1
, s2
, n
);
442 /** Compare the s1_len-byte string <b>s1</b> with <b>s2</b>,
443 * without depending on a terminating nul in s1. Sorting order is first by
444 * length, then lexically; return values are as for strcmp.
447 strcmp_len(const char *s1
, const char *s2
, size_t s1_len
)
449 size_t s2_len
= strlen(s2
);
454 return memcmp(s1
, s2
, s2_len
);
457 /** Compares the first strlen(s2) characters of s1 with s2. Returns as for
461 strcasecmpstart(const char *s1
, const char *s2
)
463 size_t n
= strlen(s2
);
464 return strncasecmp(s1
, s2
, n
);
467 /** Compares the last strlen(s2) characters of s1 with s2. Returns as for
471 strcmpend(const char *s1
, const char *s2
)
473 size_t n1
= strlen(s1
), n2
= strlen(s2
);
475 return strcmp(s1
,s2
);
477 return strncmp(s1
+(n1
-n2
), s2
, n2
);
480 /** Compares the last strlen(s2) characters of s1 with s2. Returns as for
484 strcasecmpend(const char *s1
, const char *s2
)
486 size_t n1
= strlen(s1
), n2
= strlen(s2
);
487 if (n2
>n1
) /* then they can't be the same; figure out which is bigger */
488 return strcasecmp(s1
,s2
);
490 return strncasecmp(s1
+(n1
-n2
), s2
, n2
);
493 /** Compare the value of the string <b>prefix</b> with the start of the
494 * <b>memlen</b>-byte memory chunk at <b>mem</b>. Return as for strcmp.
496 * [As memcmp(mem, prefix, strlen(prefix)) but returns -1 if memlen is less
497 * than strlen(prefix).]
500 memcmpstart(const void *mem
, size_t memlen
,
503 size_t plen
= strlen(prefix
);
506 return memcmp(mem
, prefix
, plen
);
509 /** Return a pointer to the first char of s that is not whitespace and
510 * not a comment, or to the terminating NUL if no such character exists.
513 eat_whitespace(const char *s
)
530 while (*s
&& *s
!= '\n')
536 /** Return a pointer to the first char of s that is not whitespace and
537 * not a comment, or to the terminating NUL if no such character exists.
540 eat_whitespace_eos(const char *s
, const char *eos
)
543 tor_assert(eos
&& s
<= eos
);
558 while (s
< eos
&& *s
&& *s
!= '\n')
565 /** Return a pointer to the first char of s that is not a space or a tab
566 * or a \\r, or to the terminating NUL if no such character exists. */
568 eat_whitespace_no_nl(const char *s
)
570 while (*s
== ' ' || *s
== '\t' || *s
== '\r')
575 /** As eat_whitespace_no_nl, but stop at <b>eos</b> whether we have
576 * found a non-whitespace character or not. */
578 eat_whitespace_eos_no_nl(const char *s
, const char *eos
)
580 while (s
< eos
&& (*s
== ' ' || *s
== '\t' || *s
== '\r'))
585 /** Return a pointer to the first char of s that is whitespace or <b>#</b>,
586 * or to the terminating NUL if no such character exists.
589 find_whitespace(const char *s
)
608 /** As find_whitespace, but stop at <b>eos</b> whether we have found a
609 * whitespace or not. */
611 find_whitespace_eos(const char *s
, const char *eos
)
631 /** Return true iff the 'len' bytes at 'mem' are all zero. */
633 tor_mem_is_zero(const char *mem
, size_t len
)
635 static const char ZERO
[] = {
636 0,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0, 0,0,0,0,
638 while (len
>= sizeof(ZERO
)) {
639 if (memcmp(mem
, ZERO
, sizeof(ZERO
)))
644 /* Deal with leftover bytes. */
646 return ! memcmp(mem
, ZERO
, len
);
651 /** Return true iff the DIGEST_LEN bytes in digest are all zero. */
653 tor_digest_is_zero(const char *digest
)
655 return tor_mem_is_zero(digest
, DIGEST_LEN
);
658 /* Helper: common code to check whether the result of a strtol or strtoul or
659 * strtoll is correct. */
660 #define CHECK_STRTOX_RESULT() \
661 /* Was at least one character converted? */ \
664 /* Were there unexpected unconverted characters? */ \
665 if (!next && *endptr) \
667 /* Is r within limits? */ \
668 if (r < min || r > max) \
671 if (next) *next = endptr; \
675 if (next) *next = endptr; \
678 /** Extract a long from the start of s, in the given numeric base. If
679 * there is unconverted data and next is provided, set *next to the
680 * first unconverted character. An error has occurred if no characters
681 * are converted; or if there are unconverted characters and next is NULL; or
682 * if the parsed value is not between min and max. When no error occurs,
683 * return the parsed value and set *ok (if provided) to 1. When an error
684 * occurs, return 0 and set *ok (if provided) to 0.
687 tor_parse_long(const char *s
, int base
, long min
, long max
,
688 int *ok
, char **next
)
693 r
= strtol(s
, &endptr
, base
);
694 CHECK_STRTOX_RESULT();
697 /** As tor_parse_long(), but return an unsigned long. */
699 tor_parse_ulong(const char *s
, int base
, unsigned long min
,
700 unsigned long max
, int *ok
, char **next
)
705 r
= strtoul(s
, &endptr
, base
);
706 CHECK_STRTOX_RESULT();
709 /** As tor_parse_log, but return a unit64_t. Only base 10 is guaranteed to
712 tor_parse_uint64(const char *s
, int base
, uint64_t min
,
713 uint64_t max
, int *ok
, char **next
)
719 r
= (uint64_t)strtoull(s
, &endptr
, base
);
720 #elif defined(MS_WINDOWS)
721 #if defined(_MSC_VER) && _MSC_VER < 1300
722 tor_assert(base
<= 10);
723 r
= (uint64_t)_atoi64(s
);
725 while (TOR_ISSPACE(*endptr
)) endptr
++;
726 while (TOR_ISDIGIT(*endptr
)) endptr
++;
728 r
= (uint64_t)_strtoui64(s
, &endptr
, base
);
730 #elif SIZEOF_LONG == 8
731 r
= (uint64_t)strtoul(s
, &endptr
, base
);
733 #error "I don't know how to parse 64-bit numbers."
736 CHECK_STRTOX_RESULT();
739 /** Encode the <b>srclen</b> bytes at <b>src</b> in a NUL-terminated,
740 * uppercase hexadecimal string; store it in the <b>destlen</b>-byte buffer
744 base16_encode(char *dest
, size_t destlen
, const char *src
, size_t srclen
)
749 tor_assert(destlen
>= srclen
*2+1);
750 tor_assert(destlen
< SIZE_T_CEILING
);
755 *cp
++ = "0123456789ABCDEF"[ (*(const uint8_t*)src
) >> 4 ];
756 *cp
++ = "0123456789ABCDEF"[ (*(const uint8_t*)src
) & 0xf ];
762 /** Helper: given a hex digit, return its value, or -1 if it isn't hex. */
764 _hex_decode_digit(char c
)
777 case 'A': case 'a': return 10;
778 case 'B': case 'b': return 11;
779 case 'C': case 'c': return 12;
780 case 'D': case 'd': return 13;
781 case 'E': case 'e': return 14;
782 case 'F': case 'f': return 15;
788 /** Helper: given a hex digit, return its value, or -1 if it isn't hex. */
790 hex_decode_digit(char c
)
792 return _hex_decode_digit(c
);
795 /** Given a hexadecimal string of <b>srclen</b> bytes in <b>src</b>, decode it
796 * and store the result in the <b>destlen</b>-byte buffer at <b>dest</b>.
797 * Return 0 on success, -1 on failure. */
799 base16_decode(char *dest
, size_t destlen
, const char *src
, size_t srclen
)
804 if ((srclen
% 2) != 0)
806 if (destlen
< srclen
/2 || destlen
> SIZE_T_CEILING
)
810 v1
= _hex_decode_digit(*src
);
811 v2
= _hex_decode_digit(*(src
+1));
814 *(uint8_t*)dest
= (v1
<<4)|v2
;
821 /** Allocate and return a new string representing the contents of <b>s</b>,
822 * surrounded by quotes and using standard C escapes.
824 * Generally, we use this for logging values that come in over the network to
825 * keep them from tricking users, and for sending certain values to the
828 * We trust values from the resolver, OS, configuration file, and command line
829 * to not be maliciously ill-formed. We validate incoming routerdescs and
830 * SOCKS requests and addresses from BEGIN cells as they're parsed;
831 * afterwards, we trust them as non-malicious.
834 esc_for_log(const char *s
)
840 return tor_strdup("");
843 for (cp
= s
; *cp
; ++cp
) {
851 if (TOR_ISPRINT(*cp
) && ((uint8_t)*cp
)<127)
859 result
= outp
= tor_malloc(len
);
861 for (cp
= s
; *cp
; ++cp
) {
882 if (TOR_ISPRINT(*cp
) && ((uint8_t)*cp
)<127) {
885 tor_snprintf(outp
, 5, "\\%03o", (int)(uint8_t) *cp
);
898 /** Allocate and return a new string representing the contents of <b>s</b>,
899 * surrounded by quotes and using standard C escapes.
901 * THIS FUNCTION IS NOT REENTRANT. Don't call it from outside the main
902 * thread. Also, each call invalidates the last-returned value, so don't
903 * try log_warn(LD_GENERAL, "%s %s", escaped(a), escaped(b));
906 escaped(const char *s
)
908 static char *_escaped_val
= NULL
;
910 tor_free(_escaped_val
);
913 _escaped_val
= esc_for_log(s
);
920 /** Rudimentary string wrapping code: given a un-wrapped <b>string</b> (no
921 * newlines!), break the string into newline-terminated lines of no more than
922 * <b>width</b> characters long (not counting newline) and insert them into
923 * <b>out</b> in order. Precede the first line with prefix0, and subsequent
924 * lines with prefixRest.
926 /* This uses a stupid greedy wrapping algorithm right now:
928 * - Try to fit as much stuff as possible, but break on a space.
929 * - If the first "word" of the line will extend beyond the allowable
930 * width, break the word at the end of the width.
933 wrap_string(smartlist_t
*out
, const char *string
, size_t width
,
934 const char *prefix0
, const char *prefixRest
)
936 size_t p0Len
, pRestLen
, pCurLen
;
937 const char *eos
, *prefixCur
;
946 p0Len
= strlen(prefix0
);
947 pRestLen
= strlen(prefixRest
);
948 tor_assert(width
> p0Len
&& width
> pRestLen
);
949 eos
= strchr(string
, '\0');
954 while ((eos
-string
)+pCurLen
> width
) {
955 const char *eol
= string
+ width
- pCurLen
;
956 while (eol
> string
&& *eol
!= ' ')
958 /* eol is now the last space that can fit, or the start of the string. */
960 size_t line_len
= (eol
-string
) + pCurLen
+ 2;
961 char *line
= tor_malloc(line_len
);
962 memcpy(line
, prefixCur
, pCurLen
);
963 memcpy(line
+pCurLen
, string
, eol
-string
);
964 line
[line_len
-2] = '\n';
965 line
[line_len
-1] = '\0';
966 smartlist_add(out
, line
);
969 size_t line_len
= width
+ 2;
970 char *line
= tor_malloc(line_len
);
971 memcpy(line
, prefixCur
, pCurLen
);
972 memcpy(line
+pCurLen
, string
, width
- pCurLen
);
973 line
[line_len
-2] = '\n';
974 line
[line_len
-1] = '\0';
975 smartlist_add(out
, line
);
976 string
+= width
-pCurLen
;
978 prefixCur
= prefixRest
;
983 size_t line_len
= (eos
-string
) + pCurLen
+ 2;
984 char *line
= tor_malloc(line_len
);
985 memcpy(line
, prefixCur
, pCurLen
);
986 memcpy(line
+pCurLen
, string
, eos
-string
);
987 line
[line_len
-2] = '\n';
988 line
[line_len
-1] = '\0';
989 smartlist_add(out
, line
);
997 /** Return the number of microseconds elapsed between *start and *end.
1000 tv_udiff(const struct timeval
*start
, const struct timeval
*end
)
1003 long secdiff
= end
->tv_sec
- start
->tv_sec
;
1005 if (labs(secdiff
+1) > LONG_MAX
/1000000) {
1006 log_warn(LD_GENERAL
, "comparing times too far apart.");
1010 udiff
= secdiff
*1000000L + (end
->tv_usec
- start
->tv_usec
);
1014 /** Yield true iff <b>y</b> is a leap-year. */
1015 #define IS_LEAPYEAR(y) (!(y % 4) && ((y % 100) || !(y % 400)))
1016 /** Helper: Return the number of leap-days between Jan 1, y1 and Jan 1, y2. */
1018 n_leapdays(int y1
, int y2
)
1022 return (y2
/4 - y1
/4) - (y2
/100 - y1
/100) + (y2
/400 - y1
/400);
1024 /** Number of days per month in non-leap year; used by tor_timegm. */
1025 static const int days_per_month
[] =
1026 { 31, 28, 31, 30, 31, 30, 31, 31, 30, 31, 30, 31};
1028 /** Return a time_t given a struct tm. The result is given in GMT, and
1029 * does not account for leap seconds.
1032 tor_timegm(struct tm
*tm
)
1034 /* This is a pretty ironclad timegm implementation, snarfed from Python2.2.
1035 * It's way more brute-force than fiddling with tzset().
1038 unsigned long year
, days
, hours
, minutes
;
1040 year
= tm
->tm_year
+ 1900;
1041 if (year
< 1970 || tm
->tm_mon
< 0 || tm
->tm_mon
> 11) {
1042 log_warn(LD_BUG
, "Out-of-range argument to tor_timegm");
1045 tor_assert(year
< INT_MAX
);
1046 days
= 365 * (year
-1970) + n_leapdays(1970,(int)year
);
1047 for (i
= 0; i
< tm
->tm_mon
; ++i
)
1048 days
+= days_per_month
[i
];
1049 if (tm
->tm_mon
> 1 && IS_LEAPYEAR(year
))
1051 days
+= tm
->tm_mday
- 1;
1052 hours
= days
*24 + tm
->tm_hour
;
1054 minutes
= hours
*60 + tm
->tm_min
;
1055 ret
= minutes
*60 + tm
->tm_sec
;
1059 /* strftime is locale-specific, so we need to replace those parts */
1061 /** A c-locale array of 3-letter names of weekdays, starting with Sun. */
1062 static const char *WEEKDAY_NAMES
[] =
1063 { "Sun", "Mon", "Tue", "Wed", "Thu", "Fri", "Sat" };
1064 /** A c-locale array of 3-letter names of months, starting with Jan. */
1065 static const char *MONTH_NAMES
[] =
1066 { "Jan", "Feb", "Mar", "Apr", "May", "Jun",
1067 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec" };
1069 /** Set <b>buf</b> to the RFC1123 encoding of the GMT value of <b>t</b>.
1070 * The buffer must be at least RFC1123_TIME_LEN+1 bytes long.
1072 * (RFC1123 format is Fri, 29 Sep 2006 15:54:20 GMT)
1075 format_rfc1123_time(char *buf
, time_t t
)
1079 tor_gmtime_r(&t
, &tm
);
1081 strftime(buf
, RFC1123_TIME_LEN
+1, "___, %d ___ %Y %H:%M:%S GMT", &tm
);
1082 tor_assert(tm
.tm_wday
>= 0);
1083 tor_assert(tm
.tm_wday
<= 6);
1084 memcpy(buf
, WEEKDAY_NAMES
[tm
.tm_wday
], 3);
1085 tor_assert(tm
.tm_wday
>= 0);
1086 tor_assert(tm
.tm_mon
<= 11);
1087 memcpy(buf
+8, MONTH_NAMES
[tm
.tm_mon
], 3);
1090 /** Parse the the RFC1123 encoding of some time (in GMT) from <b>buf</b>,
1091 * and store the result in *<b>t</b>.
1093 * Return 0 on succcess, -1 on failure.
1096 parse_rfc1123_time(const char *buf
, time_t *t
)
1103 if (strlen(buf
) != RFC1123_TIME_LEN
)
1105 memset(&tm
, 0, sizeof(tm
));
1106 if (sscanf(buf
, "%3s, %d %3s %d %d:%d:%d GMT", weekday
,
1107 &tm
.tm_mday
, month
, &tm
.tm_year
, &tm
.tm_hour
,
1108 &tm
.tm_min
, &tm
.tm_sec
) < 7) {
1109 char *esc
= esc_for_log(buf
);
1110 log_warn(LD_GENERAL
, "Got invalid RFC1123 time %s", esc
);
1116 for (i
= 0; i
< 12; ++i
) {
1117 if (!strcmp(month
, MONTH_NAMES
[i
])) {
1123 char *esc
= esc_for_log(buf
);
1124 log_warn(LD_GENERAL
, "Got invalid RFC1123 time %s: No such month", esc
);
1130 if (tm
.tm_year
< 1970) {
1131 char *esc
= esc_for_log(buf
);
1132 log_warn(LD_GENERAL
,
1133 "Got invalid RFC1123 time %s. (Before 1970)", esc
);
1139 *t
= tor_timegm(&tm
);
1143 /** Set <b>buf</b> to the ISO8601 encoding of the local value of <b>t</b>.
1144 * The buffer must be at least ISO_TIME_LEN+1 bytes long.
1146 * (ISO8601 format is 2006-10-29 10:57:20)
1149 format_local_iso_time(char *buf
, time_t t
)
1152 strftime(buf
, ISO_TIME_LEN
+1, "%Y-%m-%d %H:%M:%S", tor_localtime_r(&t
, &tm
));
1155 /** Set <b>buf</b> to the ISO8601 encoding of the GMT value of <b>t</b>.
1156 * The buffer must be at least ISO_TIME_LEN+1 bytes long.
1159 format_iso_time(char *buf
, time_t t
)
1162 strftime(buf
, ISO_TIME_LEN
+1, "%Y-%m-%d %H:%M:%S", tor_gmtime_r(&t
, &tm
));
1165 /** Given an ISO-formatted UTC time value (after the epoch) in <b>cp</b>,
1166 * parse it and store its value in *<b>t</b>. Return 0 on success, -1 on
1167 * failure. Ignore extraneous stuff in <b>cp</b> separated by whitespace from
1168 * the end of the time string. */
1170 parse_iso_time(const char *cp
, time_t *t
)
1173 #ifdef HAVE_STRPTIME
1174 if (!strptime(cp
, "%Y-%m-%d %H:%M:%S", &st_tm
)) {
1175 log_warn(LD_GENERAL
, "ISO time was unparseable by strptime"); return -1;
1178 unsigned int year
=0, month
=0, day
=0, hour
=100, minute
=100, second
=100;
1179 if (sscanf(cp
, "%u-%u-%u %u:%u:%u", &year
, &month
,
1180 &day
, &hour
, &minute
, &second
) < 6) {
1181 log_warn(LD_GENERAL
, "ISO time was unparseable"); return -1;
1183 if (year
< 1970 || month
< 1 || month
> 12 || day
< 1 || day
> 31 ||
1184 hour
> 23 || minute
> 59 || second
> 61) {
1185 log_warn(LD_GENERAL
, "ISO time was nonsensical"); return -1;
1187 st_tm
.tm_year
= year
-1900;
1188 st_tm
.tm_mon
= month
-1;
1189 st_tm
.tm_mday
= day
;
1190 st_tm
.tm_hour
= hour
;
1191 st_tm
.tm_min
= minute
;
1192 st_tm
.tm_sec
= second
;
1194 if (st_tm
.tm_year
< 70) {
1195 char *esc
= esc_for_log(cp
);
1196 log_warn(LD_GENERAL
, "Got invalid ISO time %s. (Before 1970)", esc
);
1200 *t
= tor_timegm(&st_tm
);
1204 /** Given a <b>date</b> in one of the three formats allowed by HTTP (ugh),
1205 * parse it into <b>tm</b>. Return 0 on success, negative on failure. */
1207 parse_http_time(const char *date
, struct tm
*tm
)
1215 memset(tm
, 0, sizeof(*tm
));
1217 /* First, try RFC1123 or RFC850 format: skip the weekday. */
1218 if ((cp
= strchr(date
, ','))) {
1220 if (sscanf(date
, "%2d %3s %4d %2d:%2d:%2d GMT",
1221 &tm
->tm_mday
, month
, &tm
->tm_year
,
1222 &tm
->tm_hour
, &tm
->tm_min
, &tm
->tm_sec
) == 6) {
1224 tm
->tm_year
-= 1900;
1225 } else if (sscanf(date
, "%2d-%3s-%2d %2d:%2d:%2d GMT",
1226 &tm
->tm_mday
, month
, &tm
->tm_year
,
1227 &tm
->tm_hour
, &tm
->tm_min
, &tm
->tm_sec
) == 6) {
1233 /* No comma; possibly asctime() format. */
1234 if (sscanf(date
, "%3s %3s %2d %2d:%2d:%2d %4d",
1235 wkday
, month
, &tm
->tm_mday
,
1236 &tm
->tm_hour
, &tm
->tm_min
, &tm
->tm_sec
, &tm
->tm_year
) == 7) {
1237 tm
->tm_year
-= 1900;
1244 /* Okay, now decode the month. */
1245 for (i
= 0; i
< 12; ++i
) {
1246 if (!strcasecmp(MONTH_NAMES
[i
], month
)) {
1251 if (tm
->tm_year
< 0 ||
1252 tm
->tm_mon
< 1 || tm
->tm_mon
> 12 ||
1253 tm
->tm_mday
< 0 || tm
->tm_mday
> 31 ||
1254 tm
->tm_hour
< 0 || tm
->tm_hour
> 23 ||
1255 tm
->tm_min
< 0 || tm
->tm_min
> 59 ||
1256 tm
->tm_sec
< 0 || tm
->tm_sec
> 61)
1257 return -1; /* Out of range, or bad month. */
1262 /** Given an <b>interval</b> in seconds, try to write it to the
1263 * <b>out_len</b>-byte buffer in <b>out</b> in a human-readable form.
1264 * Return 0 on success, -1 on failure.
1267 format_time_interval(char *out
, size_t out_len
, long interval
)
1269 /* We only report seconds if there's no hours. */
1270 long sec
= 0, min
= 0, hour
= 0, day
= 0;
1272 interval
= -interval
;
1274 if (interval
>= 86400) {
1275 day
= interval
/ 86400;
1278 if (interval
>= 3600) {
1279 hour
= interval
/ 3600;
1282 if (interval
>= 60) {
1283 min
= interval
/ 60;
1289 return tor_snprintf(out
, out_len
, "%ld days, %ld hours, %ld minutes",
1292 return tor_snprintf(out
, out_len
, "%ld hours, %ld minutes", hour
, min
);
1294 return tor_snprintf(out
, out_len
, "%ld minutes, %ld seconds", min
, sec
);
1296 return tor_snprintf(out
, out_len
, "%ld seconds", sec
);
1304 #ifndef TIME_IS_FAST
1305 /** Cached estimate of the currrent time. Updated around once per second;
1306 * may be a few seconds off if we are really busy. This is a hack to avoid
1307 * calling time(NULL) (which not everybody has optimized) on critical paths.
1309 static time_t cached_approx_time
= 0;
1311 /** Return a cached estimate of the current time from when
1312 * update_approx_time() was last called. This is a hack to avoid calling
1313 * time(NULL) on critical paths: please do not even think of calling it
1318 return cached_approx_time
;
1321 /** Update the cached estimate of the current time. This function SHOULD be
1322 * called once per second, and MUST be called before the first call to
1323 * get_approx_time. */
1325 update_approx_time(time_t now
)
1327 cached_approx_time
= now
;
1333 * XXXX022 Use this consistently or rip most of it out.
1336 /* In a perfect world, everybody would run ntp, and ntp would be perfect, so
1337 * if we wanted to know "Is the current time before time X?" we could just say
1340 * But unfortunately, many users are running Tor in an imperfect world, on
1341 * even more imperfect computers. Hence, we need to track time oddly. We
1342 * model the user's computer as being "skewed" from accurate time by
1343 * -<b>ftime_skew</b> seconds, such that our best guess of the current time is
1344 * time(NULL)+ftime_skew. We also assume that our measurements of time may
1345 * have up to <b>ftime_slop</b> seconds of inaccuracy; IOW, our window of
1346 * estimate for the current time is now + ftime_skew +/- ftime_slop.
1348 /** Our current estimate of our skew, such that we think the current time is
1349 * closest to time(NULL)+ftime_skew. */
1350 static int ftime_skew
= 0;
1351 /** Tolerance during time comparisons, in seconds. */
1352 static int ftime_slop
= 60;
1353 /** Set the largest amount of sloppiness we'll allow in fuzzy time
1356 ftime_set_maximum_sloppiness(int seconds
)
1358 tor_assert(seconds
>= 0);
1359 ftime_slop
= seconds
;
1361 /** Set the amount by which we believe our system clock to differ from
1364 ftime_set_estimated_skew(int seconds
)
1366 ftime_skew
= seconds
;
1370 ftime_get_window(time_t now
, ftime_t
*ft_out
)
1372 ft_out
->earliest
= now
+ ftime_skew
- ftime_slop
;
1373 ft_out
->latest
= now
+ ftime_skew
+ ftime_slop
;
1376 /** Return true iff we think that <b>now</b> might be after <b>when</b>. */
1378 ftime_maybe_after(time_t now
, time_t when
)
1380 /* It may be after when iff the latest possible current time is after when */
1381 return (now
+ ftime_skew
+ ftime_slop
) >= when
;
1383 /** Return true iff we think that <b>now</b> might be before <b>when</b>. */
1385 ftime_maybe_before(time_t now
, time_t when
)
1387 /* It may be before when iff the earliest possible current time is before */
1388 return (now
+ ftime_skew
- ftime_slop
) < when
;
1390 /** Return true if we think that <b>now</b> is definitely after <b>when</b>. */
1392 ftime_definitely_after(time_t now
, time_t when
)
1394 /* It is definitely after when if the earliest time it could be is still
1396 return (now
+ ftime_skew
- ftime_slop
) >= when
;
1398 /** Return true if we think that <b>now</b> is definitely before <b>when</b>.
1401 ftime_definitely_before(time_t now
, time_t when
)
1403 /* It is definitely before when if the latest time it could be is still
1405 return (now
+ ftime_skew
+ ftime_slop
) < when
;
1412 /** Write <b>count</b> bytes from <b>buf</b> to <b>fd</b>. <b>isSocket</b>
1413 * must be 1 if fd was returned by socket() or accept(), and 0 if fd
1414 * was returned by open(). Return the number of bytes written, or -1
1415 * on error. Only use if fd is a blocking fd. */
1417 write_all(int fd
, const char *buf
, size_t count
, int isSocket
)
1421 tor_assert(count
< SSIZE_T_MAX
);
1423 while (written
!= count
) {
1425 result
= tor_socket_send(fd
, buf
+written
, count
-written
, 0);
1427 result
= write(fd
, buf
+written
, count
-written
);
1432 return (ssize_t
)count
;
1435 /** Read from <b>fd</b> to <b>buf</b>, until we get <b>count</b> bytes
1436 * or reach the end of the file. <b>isSocket</b> must be 1 if fd
1437 * was returned by socket() or accept(), and 0 if fd was returned by
1438 * open(). Return the number of bytes read, or -1 on error. Only use
1439 * if fd is a blocking fd. */
1441 read_all(int fd
, char *buf
, size_t count
, int isSocket
)
1446 if (count
> SIZE_T_CEILING
|| count
> SSIZE_T_MAX
)
1449 while (numread
!= count
) {
1451 result
= tor_socket_recv(fd
, buf
+numread
, count
-numread
, 0);
1453 result
= read(fd
, buf
+numread
, count
-numread
);
1456 else if (result
== 0)
1460 return (ssize_t
)numread
;
1464 * Filesystem operations.
1467 /** Clean up <b>name</b> so that we can use it in a call to "stat". On Unix,
1468 * we do nothing. On Windows, we remove a trailing slash, unless the path is
1469 * the root of a disk. */
1471 clean_name_for_stat(char *name
)
1474 size_t len
= strlen(name
);
1477 if (name
[len
-1]=='\\' || name
[len
-1]=='/') {
1478 if (len
== 1 || (len
==3 && name
[1]==':'))
1487 /** Return FN_ERROR if filename can't be read, FN_NOENT if it doesn't
1488 * exist, FN_FILE if it is a regular file, or FN_DIR if it's a
1489 * directory. On FN_ERROR, sets errno. */
1491 file_status(const char *fname
)
1496 f
= tor_strdup(fname
);
1497 clean_name_for_stat(f
);
1501 if (errno
== ENOENT
) {
1506 if (st
.st_mode
& S_IFDIR
)
1508 else if (st
.st_mode
& S_IFREG
)
1514 /** Check whether dirname exists and is private. If yes return 0. If
1515 * it does not exist, and check==CPD_CREATE is set, try to create it
1516 * and return 0 on success. If it does not exist, and
1517 * check==CPD_CHECK, and we think we can create it, return 0. Else
1520 check_private_dir(const char *dirname
, cpd_check_t check
)
1525 tor_assert(dirname
);
1526 f
= tor_strdup(dirname
);
1527 clean_name_for_stat(f
);
1531 if (errno
!= ENOENT
) {
1532 log(LOG_WARN
, LD_FS
, "Directory %s cannot be read: %s", dirname
,
1536 if (check
== CPD_NONE
) {
1537 log(LOG_WARN
, LD_FS
, "Directory %s does not exist.", dirname
);
1539 } else if (check
== CPD_CREATE
) {
1540 log_info(LD_GENERAL
, "Creating directory %s", dirname
);
1544 r
= mkdir(dirname
, 0700);
1547 log(LOG_WARN
, LD_FS
, "Error creating directory %s: %s", dirname
,
1552 /* XXXX In the case where check==CPD_CHECK, we should look at the
1553 * parent directory a little harder. */
1556 if (!(st
.st_mode
& S_IFDIR
)) {
1557 log(LOG_WARN
, LD_FS
, "%s is not a directory", dirname
);
1561 if (st
.st_uid
!= getuid()) {
1562 struct passwd
*pw
= NULL
;
1563 char *process_ownername
= NULL
;
1565 pw
= getpwuid(getuid());
1566 process_ownername
= pw
? tor_strdup(pw
->pw_name
) : tor_strdup("<unknown>");
1568 pw
= getpwuid(st
.st_uid
);
1570 log(LOG_WARN
, LD_FS
, "%s is not owned by this user (%s, %d) but by "
1571 "%s (%d). Perhaps you are running Tor as the wrong user?",
1572 dirname
, process_ownername
, (int)getuid(),
1573 pw
? pw
->pw_name
: "<unknown>", (int)st
.st_uid
);
1575 tor_free(process_ownername
);
1578 if (st
.st_mode
& 0077) {
1579 log(LOG_WARN
, LD_FS
, "Fixing permissions on directory %s", dirname
);
1580 if (chmod(dirname
, 0700)) {
1581 log(LOG_WARN
, LD_FS
, "Could not chmod directory %s: %s", dirname
,
1592 /** Create a file named <b>fname</b> with the contents <b>str</b>. Overwrite
1593 * the previous <b>fname</b> if possible. Return 0 on success, -1 on failure.
1595 * This function replaces the old file atomically, if possible. This
1596 * function, and all other functions in util.c that create files, create them
1600 write_str_to_file(const char *fname
, const char *str
, int bin
)
1603 if (!bin
&& strchr(str
, '\r')) {
1605 "We're writing a text string that already contains a CR.");
1608 return write_bytes_to_file(fname
, str
, strlen(str
), bin
);
1611 /** Represents a file that we're writing to, with support for atomic commit:
1612 * we can write into a a temporary file, and either remove the file on
1613 * failure, or replace the original file on success. */
1614 struct open_file_t
{
1615 char *tempname
; /**< Name of the temporary file. */
1616 char *filename
; /**< Name of the original file. */
1617 int rename_on_close
; /**< Are we using the temporary file or not? */
1618 int fd
; /**< fd for the open file. */
1619 FILE *stdio_file
; /**< stdio wrapper for <b>fd</b>. */
1622 /** Try to start writing to the file in <b>fname</b>, passing the flags
1623 * <b>open_flags</b> to the open() syscall, creating the file (if needed) with
1624 * access value <b>mode</b>. If the O_APPEND flag is set, we append to the
1625 * original file. Otherwise, we open a new temporary file in the same
1626 * directory, and either replace the original or remove the temporary file
1629 * Return the fd for the newly opened file, and store working data in
1630 * *<b>data_out</b>. The caller should not close the fd manually:
1631 * instead, call finish_writing_to_file() or abort_writing_to_file().
1632 * Returns -1 on failure.
1634 * NOTE: When not appending, the flags O_CREAT and O_TRUNC are treated
1635 * as true and the flag O_EXCL is treated as false.
1637 * NOTE: Ordinarily, O_APPEND means "seek to the end of the file before each
1638 * write()". We don't do that.
1641 start_writing_to_file(const char *fname
, int open_flags
, int mode
,
1642 open_file_t
**data_out
)
1644 size_t tempname_len
= strlen(fname
)+16;
1645 open_file_t
*new_file
= tor_malloc_zero(sizeof(open_file_t
));
1646 const char *open_name
;
1650 tor_assert(data_out
);
1651 #if (O_BINARY != 0 && O_TEXT != 0)
1652 tor_assert((open_flags
& (O_BINARY
|O_TEXT
)) != 0);
1655 tor_assert(tempname_len
> strlen(fname
)); /*check for overflow*/
1656 new_file
->filename
= tor_strdup(fname
);
1657 if (open_flags
& O_APPEND
) {
1659 new_file
->rename_on_close
= 0;
1661 open_flags
&= ~O_APPEND
;
1663 open_name
= new_file
->tempname
= tor_malloc(tempname_len
);
1664 if (tor_snprintf(new_file
->tempname
, tempname_len
, "%s.tmp", fname
)<0) {
1665 log(LOG_WARN
, LD_GENERAL
, "Failed to generate filename");
1668 /* We always replace an existing temporary file if there is one. */
1669 open_flags
|= O_CREAT
|O_TRUNC
;
1670 open_flags
&= ~O_EXCL
;
1671 new_file
->rename_on_close
= 1;
1674 if ((new_file
->fd
= open(open_name
, open_flags
, mode
)) < 0) {
1675 log(LOG_WARN
, LD_FS
, "Couldn't open \"%s\" (%s) for writing: %s",
1676 open_name
, fname
, strerror(errno
));
1680 if (tor_fd_seekend(new_file
->fd
) < 0) {
1681 log_warn(LD_FS
, "Couldn't seek to end of file \"%s\": %s", open_name
,
1687 *data_out
= new_file
;
1689 return new_file
->fd
;
1692 if (new_file
->fd
>= 0)
1693 close(new_file
->fd
);
1695 tor_free(new_file
->filename
);
1696 tor_free(new_file
->tempname
);
1701 /** Given <b>file_data</b> from start_writing_to_file(), return a stdio FILE*
1702 * that can be used to write to the same file. The caller should not mix
1703 * stdio calls with non-stdio calls. */
1705 fdopen_file(open_file_t
*file_data
)
1707 tor_assert(file_data
);
1708 if (file_data
->stdio_file
)
1709 return file_data
->stdio_file
;
1710 tor_assert(file_data
->fd
>= 0);
1711 if (!(file_data
->stdio_file
= fdopen(file_data
->fd
, "a"))) {
1712 log_warn(LD_FS
, "Couldn't fdopen \"%s\" [%d]: %s", file_data
->filename
,
1713 file_data
->fd
, strerror(errno
));
1715 return file_data
->stdio_file
;
1718 /** Combines start_writing_to_file with fdopen_file(): arguments are as
1719 * for start_writing_to_file, but */
1721 start_writing_to_stdio_file(const char *fname
, int open_flags
, int mode
,
1722 open_file_t
**data_out
)
1725 if (start_writing_to_file(fname
, open_flags
, mode
, data_out
)<0)
1727 if (!(res
= fdopen_file(*data_out
))) {
1728 abort_writing_to_file(*data_out
);
1734 /** Helper function: close and free the underlying file and memory in
1735 * <b>file_data</b>. If we were writing into a temporary file, then delete
1736 * that file (if abort_write is true) or replaces the target file with
1737 * the temporary file (if abort_write is false). */
1739 finish_writing_to_file_impl(open_file_t
*file_data
, int abort_write
)
1742 tor_assert(file_data
&& file_data
->filename
);
1743 if (file_data
->stdio_file
) {
1744 if (fclose(file_data
->stdio_file
)) {
1745 log_warn(LD_FS
, "Error closing \"%s\": %s", file_data
->filename
,
1747 abort_write
= r
= -1;
1749 } else if (file_data
->fd
>= 0 && close(file_data
->fd
) < 0) {
1750 log_warn(LD_FS
, "Error flushing \"%s\": %s", file_data
->filename
,
1752 abort_write
= r
= -1;
1755 if (file_data
->rename_on_close
) {
1756 tor_assert(file_data
->tempname
&& file_data
->filename
);
1758 unlink(file_data
->tempname
);
1760 tor_assert(strcmp(file_data
->filename
, file_data
->tempname
));
1761 if (replace_file(file_data
->tempname
, file_data
->filename
)) {
1762 log_warn(LD_FS
, "Error replacing \"%s\": %s", file_data
->filename
,
1769 tor_free(file_data
->filename
);
1770 tor_free(file_data
->tempname
);
1771 tor_free(file_data
);
1776 /** Finish writing to <b>file_data</b>: close the file handle, free memory as
1777 * needed, and if using a temporary file, replace the original file with
1778 * the temporary file. */
1780 finish_writing_to_file(open_file_t
*file_data
)
1782 return finish_writing_to_file_impl(file_data
, 0);
1785 /** Finish writing to <b>file_data</b>: close the file handle, free memory as
1786 * needed, and if using a temporary file, delete it. */
1788 abort_writing_to_file(open_file_t
*file_data
)
1790 return finish_writing_to_file_impl(file_data
, 1);
1793 /** Helper: given a set of flags as passed to open(2), open the file
1794 * <b>fname</b> and write all the sized_chunk_t structs in <b>chunks</b> to
1795 * the file. Do so as atomically as possible e.g. by opening temp files and
1798 write_chunks_to_file_impl(const char *fname
, const smartlist_t
*chunks
,
1801 open_file_t
*file
= NULL
;
1803 fd
= start_writing_to_file(fname
, open_flags
, 0600, &file
);
1806 SMARTLIST_FOREACH(chunks
, sized_chunk_t
*, chunk
,
1808 result
= write_all(fd
, chunk
->bytes
, chunk
->len
, 0);
1810 log(LOG_WARN
, LD_FS
, "Error writing to \"%s\": %s", fname
,
1814 tor_assert((size_t)result
== chunk
->len
);
1817 return finish_writing_to_file(file
);
1819 abort_writing_to_file(file
);
1823 /** Given a smartlist of sized_chunk_t, write them atomically to a file
1824 * <b>fname</b>, overwriting or creating the file as necessary. */
1826 write_chunks_to_file(const char *fname
, const smartlist_t
*chunks
, int bin
)
1828 int flags
= OPEN_FLAGS_REPLACE
|(bin
?O_BINARY
:O_TEXT
);
1829 return write_chunks_to_file_impl(fname
, chunks
, flags
);
1832 /** As write_str_to_file, but does not assume a NUL-terminated
1833 * string. Instead, we write <b>len</b> bytes, starting at <b>str</b>. */
1835 write_bytes_to_file(const char *fname
, const char *str
, size_t len
,
1838 int flags
= OPEN_FLAGS_REPLACE
|(bin
?O_BINARY
:O_TEXT
);
1840 sized_chunk_t c
= { str
, len
};
1841 smartlist_t
*chunks
= smartlist_create();
1842 smartlist_add(chunks
, &c
);
1843 r
= write_chunks_to_file_impl(fname
, chunks
, flags
);
1844 smartlist_free(chunks
);
1848 /** As write_bytes_to_file, but if the file already exists, append the bytes
1849 * to the end of the file instead of overwriting it. */
1851 append_bytes_to_file(const char *fname
, const char *str
, size_t len
,
1854 int flags
= OPEN_FLAGS_APPEND
|(bin
?O_BINARY
:O_TEXT
);
1856 sized_chunk_t c
= { str
, len
};
1857 smartlist_t
*chunks
= smartlist_create();
1858 smartlist_add(chunks
, &c
);
1859 r
= write_chunks_to_file_impl(fname
, chunks
, flags
);
1860 smartlist_free(chunks
);
1864 /** Read the contents of <b>filename</b> into a newly allocated
1865 * string; return the string on success or NULL on failure.
1867 * If <b>stat_out</b> is provided, store the result of stat()ing the
1868 * file into <b>stat_out</b>.
1870 * If <b>flags</b> & RFTS_BIN, open the file in binary mode.
1871 * If <b>flags</b> & RFTS_IGNORE_MISSING, don't warn if the file
1875 * This function <em>may</em> return an erroneous result if the file
1876 * is modified while it is running, but must not crash or overflow.
1877 * Right now, the error case occurs when the file length grows between
1878 * the call to stat and the call to read_all: the resulting string will
1882 read_file_to_str(const char *filename
, int flags
, struct stat
*stat_out
)
1884 int fd
; /* router file */
1885 struct stat statbuf
;
1888 int bin
= flags
& RFTS_BIN
;
1890 tor_assert(filename
);
1892 fd
= open(filename
,O_RDONLY
|(bin
?O_BINARY
:O_TEXT
),0);
1894 int severity
= LOG_WARN
;
1895 int save_errno
= errno
;
1896 if (errno
== ENOENT
&& (flags
& RFTS_IGNORE_MISSING
))
1897 severity
= LOG_INFO
;
1898 log_fn(severity
, LD_FS
,"Could not open \"%s\": %s ",filename
,
1904 if (fstat(fd
, &statbuf
)<0) {
1905 int save_errno
= errno
;
1907 log_warn(LD_FS
,"Could not fstat \"%s\".",filename
);
1912 if ((uint64_t)(statbuf
.st_size
)+1 > SIZE_T_MAX
)
1915 string
= tor_malloc((size_t)(statbuf
.st_size
+1));
1917 r
= read_all(fd
,string
,(size_t)statbuf
.st_size
,0);
1919 int save_errno
= errno
;
1920 log_warn(LD_FS
,"Error reading from file \"%s\": %s", filename
,
1927 string
[r
] = '\0'; /* NUL-terminate the result. */
1930 if (!bin
&& strchr(string
, '\r')) {
1931 log_debug(LD_FS
, "We didn't convert CRLF to LF as well as we hoped "
1932 "when reading %s. Coping.",
1934 tor_strstrip(string
, "\r");
1938 statbuf
.st_size
= (size_t) r
;
1941 if (r
!= statbuf
.st_size
) {
1942 /* Unless we're using text mode on win32, we'd better have an exact
1943 * match for size. */
1944 int save_errno
= errno
;
1945 log_warn(LD_FS
,"Could read only %d of %ld bytes of file \"%s\".",
1946 r
, (long)statbuf
.st_size
,filename
);
1954 memcpy(stat_out
, &statbuf
, sizeof(struct stat
));
1960 #define TOR_ISODIGIT(c) ('0' <= (c) && (c) <= '7')
1962 /** Given a c-style double-quoted escaped string in <b>s</b>, extract and
1963 * decode its contents into a newly allocated string. On success, assign this
1964 * string to *<b>result</b>, assign its length to <b>size_out</b> (if
1965 * provided), and return a pointer to the position in <b>s</b> immediately
1966 * after the string. On failure, return NULL.
1969 unescape_string(const char *s
, char **result
, size_t *size_out
)
1984 if ((cp
[1] == 'x' || cp
[1] == 'X')
1985 && TOR_ISXDIGIT(cp
[2]) && TOR_ISXDIGIT(cp
[3])) {
1987 } else if (TOR_ISODIGIT(cp
[1])) {
1989 if (TOR_ISODIGIT(*cp
)) ++cp
;
1990 if (TOR_ISODIGIT(*cp
)) ++cp
;
2003 out
= *result
= tor_malloc(cp
-s
+ 1);
2010 if (size_out
) *size_out
= out
- *result
;
2013 tor_fragile_assert();
2019 case 'n': *out
++ = '\n'; cp
+= 2; break;
2020 case 'r': *out
++ = '\r'; cp
+= 2; break;
2021 case 't': *out
++ = '\t'; cp
+= 2; break;
2023 *out
++ = ((hex_decode_digit(cp
[2])<<4) +
2024 hex_decode_digit(cp
[3]));
2027 case '0': case '1': case '2': case '3': case '4': case '5':
2032 if (TOR_ISODIGIT(*cp
)) { n
= n
*8 + *cp
-'0'; cp
++; }
2033 if (TOR_ISODIGIT(*cp
)) { n
= n
*8 + *cp
-'0'; cp
++; }
2034 if (n
> 255) { tor_free(*result
); return NULL
; }
2046 tor_free(*result
); return NULL
;
2055 /** Given a string containing part of a configuration file or similar format,
2056 * advance past comments and whitespace and try to parse a single line. If we
2057 * parse a line successfully, set *<b>key_out</b> to a new string holding the
2058 * key portion and *<b>value_out</b> to a new string holding the value portion
2059 * of the line, and return a pointer to the start of the next line. If we run
2060 * out of data, return a pointer to the end of the string. If we encounter an
2061 * error, return NULL.
2064 parse_config_line_from_str(const char *line
, char **key_out
, char **value_out
)
2066 const char *key
, *val
, *cp
;
2068 tor_assert(key_out
);
2069 tor_assert(value_out
);
2071 *key_out
= *value_out
= NULL
;
2073 /* Skip until the first keyword. */
2075 while (TOR_ISSPACE(*line
))
2078 while (*line
&& *line
!= '\n')
2085 if (!*line
) { /* End of string? */
2086 *key_out
= *value_out
= NULL
;
2090 /* Skip until the next space. */
2092 while (*line
&& !TOR_ISSPACE(*line
) && *line
!= '#')
2094 *key_out
= tor_strndup(key
, line
-key
);
2096 /* Skip until the value. */
2097 while (*line
== ' ' || *line
== '\t')
2102 /* Find the end of the line. */
2103 if (*line
== '\"') {
2104 if (!(line
= unescape_string(line
, value_out
, NULL
)))
2106 while (*line
== ' ' || *line
== '\t')
2108 if (*line
&& *line
!= '#' && *line
!= '\n')
2111 while (*line
&& *line
!= '\n' && *line
!= '#')
2113 if (*line
== '\n') {
2118 while (cp
>val
&& TOR_ISSPACE(*(cp
-1)))
2121 tor_assert(cp
>= val
);
2122 *value_out
= tor_strndup(val
, cp
-val
);
2128 } while (*line
&& *line
!= '\n');
2130 while (TOR_ISSPACE(*line
)) ++line
;
2135 /** Expand any homedir prefix on <b>filename</b>; return a newly allocated
2138 expand_filename(const char *filename
)
2140 tor_assert(filename
);
2141 if (*filename
== '~') {
2143 char *home
, *result
;
2146 if (filename
[1] == '/' || filename
[1] == '\0') {
2147 home
= getenv("HOME");
2149 log_warn(LD_CONFIG
, "Couldn't find $HOME environment variable while "
2150 "expanding \"%s\"", filename
);
2153 home
= tor_strdup(home
);
2154 rest
= strlen(filename
)>=2?(filename
+2):"";
2157 char *username
, *slash
;
2158 slash
= strchr(filename
, '/');
2160 username
= tor_strndup(filename
+1,slash
-filename
-1);
2162 username
= tor_strdup(filename
+1);
2163 if (!(home
= get_user_homedir(username
))) {
2164 log_warn(LD_CONFIG
,"Couldn't get homedir for \"%s\"",username
);
2169 rest
= slash
? (slash
+1) : "";
2171 log_warn(LD_CONFIG
, "Couldn't expend homedir on system without pwd.h");
2172 return tor_strdup(filename
);
2176 /* Remove trailing slash. */
2177 if (strlen(home
)>1 && !strcmpend(home
,PATH_SEPARATOR
)) {
2178 home
[strlen(home
)-1] = '\0';
2180 /* Plus one for /, plus one for NUL.
2181 * Round up to 16 in case we can't do math. */
2182 len
= strlen(home
)+strlen(rest
)+16;
2183 result
= tor_malloc(len
);
2184 tor_snprintf(result
,len
,"%s"PATH_SEPARATOR
"%s",home
,rest
);
2188 return tor_strdup(filename
);
2192 /** Return a new list containing the filenames in the directory <b>dirname</b>.
2193 * Return NULL on error or if <b>dirname</b> is not a directory.
2196 tor_listdir(const char *dirname
)
2198 smartlist_t
*result
;
2202 WIN32_FIND_DATA findData
;
2203 size_t pattern_len
= strlen(dirname
)+16;
2204 pattern
= tor_malloc(pattern_len
);
2205 tor_snprintf(pattern
, pattern_len
, "%s\\*", dirname
);
2206 if (INVALID_HANDLE_VALUE
== (handle
= FindFirstFile(pattern
, &findData
))) {
2210 result
= smartlist_create();
2212 if (strcmp(findData
.cFileName
, ".") &&
2213 strcmp(findData
.cFileName
, "..")) {
2214 smartlist_add(result
, tor_strdup(findData
.cFileName
));
2216 if (!FindNextFile(handle
, &findData
)) {
2218 if ((err
= GetLastError()) != ERROR_NO_MORE_FILES
) {
2219 char *errstr
= format_win32_error(err
);
2220 log_warn(LD_FS
, "Error reading directory '%s': %s", dirname
, errstr
);
2231 if (!(d
= opendir(dirname
)))
2234 result
= smartlist_create();
2235 while ((de
= readdir(d
))) {
2236 if (!strcmp(de
->d_name
, ".") ||
2237 !strcmp(de
->d_name
, ".."))
2239 smartlist_add(result
, tor_strdup(de
->d_name
));
2246 /** Return true iff <b>filename</b> is a relative path. */
2248 path_is_relative(const char *filename
)
2250 if (filename
&& filename
[0] == '/')
2253 else if (filename
&& filename
[0] == '\\')
2255 else if (filename
&& strlen(filename
)>3 && TOR_ISALPHA(filename
[0]) &&
2256 filename
[1] == ':' && filename
[2] == '\\')
2268 /* Based on code contributed by christian grothoff */
2269 /** True iff we've called start_daemon(). */
2270 static int start_daemon_called
= 0;
2271 /** True iff we've called finish_daemon(). */
2272 static int finish_daemon_called
= 0;
2273 /** Socketpair used to communicate between parent and child process while
2275 static int daemon_filedes
[2];
2276 /** Start putting the process into daemon mode: fork and drop all resources
2277 * except standard fds. The parent process never returns, but stays around
2278 * until finish_daemon is called. (Note: it's safe to call this more
2279 * than once: calls after the first are ignored.)
2286 if (start_daemon_called
)
2288 start_daemon_called
= 1;
2290 if (pipe(daemon_filedes
)) {
2291 log_err(LD_GENERAL
,"pipe failed; exiting. Error was %s", strerror(errno
));
2296 log_err(LD_GENERAL
,"fork failed. Exiting.");
2299 if (pid
) { /* Parent */
2303 close(daemon_filedes
[1]); /* we only read */
2305 while (0 < read(daemon_filedes
[0], &c
, sizeof(char))) {
2313 exit(1); /* child reported error */
2314 } else { /* Child */
2315 close(daemon_filedes
[0]); /* we only write */
2317 pid
= setsid(); /* Detach from controlling terminal */
2319 * Fork one more time, so the parent (the session group leader) can exit.
2320 * This means that we, as a non-session group leader, can never regain a
2321 * controlling terminal. This part is recommended by Stevens's
2322 * _Advanced Programming in the Unix Environment_.
2331 /** Finish putting the process into daemon mode: drop standard fds, and tell
2332 * the parent process to exit. (Note: it's safe to call this more than once:
2333 * calls after the first are ignored. Calls start_daemon first if it hasn't
2334 * been called already.)
2337 finish_daemon(const char *desired_cwd
)
2341 if (finish_daemon_called
)
2343 if (!start_daemon_called
)
2345 finish_daemon_called
= 1;
2349 /* Don't hold the wrong FS mounted */
2350 if (chdir(desired_cwd
) < 0) {
2351 log_err(LD_GENERAL
,"chdir to \"%s\" failed. Exiting.",desired_cwd
);
2355 nullfd
= open("/dev/null", O_RDWR
);
2357 log_err(LD_GENERAL
,"/dev/null can't be opened. Exiting.");
2360 /* close fds linking to invoking terminal, but
2361 * close usual incoming fds, but redirect them somewhere
2362 * useful so the fds don't get reallocated elsewhere.
2364 if (dup2(nullfd
,0) < 0 ||
2365 dup2(nullfd
,1) < 0 ||
2366 dup2(nullfd
,2) < 0) {
2367 log_err(LD_GENERAL
,"dup2 failed. Exiting.");
2372 /* signal success */
2373 if (write(daemon_filedes
[1], &c
, sizeof(char)) != sizeof(char)) {
2374 log_err(LD_GENERAL
,"write failed. Exiting.");
2376 close(daemon_filedes
[1]);
2379 /* defined(MS_WINDOWS) */
2385 finish_daemon(const char *cp
)
2391 /** Write the current process ID, followed by NL, into <b>filename</b>.
2394 write_pidfile(char *filename
)
2398 if ((pidfile
= fopen(filename
, "w")) == NULL
) {
2399 log_warn(LD_FS
, "Unable to open \"%s\" for writing: %s", filename
,
2403 fprintf(pidfile
, "%d\n", (int)_getpid());
2405 fprintf(pidfile
, "%d\n", (int)getpid());