1 /*
2  * General purpose functions.
3  *
4  * Copyright 2000-2010 Willy Tarreau <w@1wt.eu>
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License
8  * as published by the Free Software Foundation; either version
9  * 2 of the License, or (at your option) any later version.
10  *
11  */
12 
13 #if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
14 #define _GNU_SOURCE
15 #include <dlfcn.h>
16 #include <link.h>
17 #endif
18 
19 #include <ctype.h>
20 #include <errno.h>
21 #include <netdb.h>
22 #include <stdarg.h>
23 #include <stdio.h>
24 #include <stdlib.h>
25 #include <string.h>
26 #include <time.h>
27 #include <unistd.h>
28 #include <sys/socket.h>
29 #include <sys/stat.h>
30 #include <sys/types.h>
31 #include <sys/un.h>
32 #include <netinet/in.h>
33 #include <arpa/inet.h>
34 
35 #include <common/chunk.h>
36 #include <common/config.h>
37 #include <common/standard.h>
38 #include <common/tools.h>
39 #include <types/global.h>
40 #include <proto/applet.h>
41 #include <proto/dns.h>
42 #include <proto/hlua.h>
43 #include <proto/listener.h>
44 #include <proto/proto_udp.h>
45 #include <proto/ssl_sock.h>
46 #include <proto/stream_interface.h>
47 #include <proto/task.h>
48 
49 #include <eb32tree.h>
50 #include <eb32sctree.h>
51 
52 /* This macro returns false if the test __x is false. Many
53  * of the following parsing function must be abort the processing
54  * if it returns 0, so this macro is useful for writing light code.
55  */
56 #define RET0_UNLESS(__x) do { if (!(__x)) return 0; } while (0)
57 
58 /* enough to store NB_ITOA_STR integers of :
59  *   2^64-1 = 18446744073709551615 or
60  *    -2^63 = -9223372036854775808
61  *
62  * The HTML version needs room for adding the 25 characters
63  * '<span class="rls"></span>' around digits at positions 3N+1 in order
64  * to add spacing at up to 6 positions : 18 446 744 073 709 551 615
65  */
66 THREAD_LOCAL char itoa_str[NB_ITOA_STR][171];
67 THREAD_LOCAL int itoa_idx = 0; /* index of next itoa_str to use */
68 
69 /* sometimes we'll need to quote strings (eg: in stats), and we don't expect
70  * to quote strings larger than a max configuration line.
71  */
72 THREAD_LOCAL char quoted_str[NB_QSTR][QSTR_SIZE + 1];
73 THREAD_LOCAL int quoted_idx = 0;
74 
75 /*
76  * unsigned long long ASCII representation
77  *
78  * return the last char '\0' or NULL if no enough
79  * space in dst
80  */
ulltoa(unsigned long long n,char * dst,size_t size)81 char *ulltoa(unsigned long long n, char *dst, size_t size)
82 {
83 	int i = 0;
84 	char *res;
85 
86 	switch(n) {
87 		case 1ULL ... 9ULL:
88 			i = 0;
89 			break;
90 
91 		case 10ULL ... 99ULL:
92 			i = 1;
93 			break;
94 
95 		case 100ULL ... 999ULL:
96 			i = 2;
97 			break;
98 
99 		case 1000ULL ... 9999ULL:
100 			i = 3;
101 			break;
102 
103 		case 10000ULL ... 99999ULL:
104 			i = 4;
105 			break;
106 
107 		case 100000ULL ... 999999ULL:
108 			i = 5;
109 			break;
110 
111 		case 1000000ULL ... 9999999ULL:
112 			i = 6;
113 			break;
114 
115 		case 10000000ULL ... 99999999ULL:
116 			i = 7;
117 			break;
118 
119 		case 100000000ULL ... 999999999ULL:
120 			i = 8;
121 			break;
122 
123 		case 1000000000ULL ... 9999999999ULL:
124 			i = 9;
125 			break;
126 
127 		case 10000000000ULL ... 99999999999ULL:
128 			i = 10;
129 			break;
130 
131 		case 100000000000ULL ... 999999999999ULL:
132 			i = 11;
133 			break;
134 
135 		case 1000000000000ULL ... 9999999999999ULL:
136 			i = 12;
137 			break;
138 
139 		case 10000000000000ULL ... 99999999999999ULL:
140 			i = 13;
141 			break;
142 
143 		case 100000000000000ULL ... 999999999999999ULL:
144 			i = 14;
145 			break;
146 
147 		case 1000000000000000ULL ... 9999999999999999ULL:
148 			i = 15;
149 			break;
150 
151 		case 10000000000000000ULL ... 99999999999999999ULL:
152 			i = 16;
153 			break;
154 
155 		case 100000000000000000ULL ... 999999999999999999ULL:
156 			i = 17;
157 			break;
158 
159 		case 1000000000000000000ULL ... 9999999999999999999ULL:
160 			i = 18;
161 			break;
162 
163 		case 10000000000000000000ULL ... ULLONG_MAX:
164 			i = 19;
165 			break;
166 	}
167 	if (i + 2 > size) // (i + 1) + '\0'
168 		return NULL;  // too long
169 	res = dst + i + 1;
170 	*res = '\0';
171 	for (; i >= 0; i--) {
172 		dst[i] = n % 10ULL + '0';
173 		n /= 10ULL;
174 	}
175 	return res;
176 }
177 
178 /*
179  * unsigned long ASCII representation
180  *
181  * return the last char '\0' or NULL if no enough
182  * space in dst
183  */
ultoa_o(unsigned long n,char * dst,size_t size)184 char *ultoa_o(unsigned long n, char *dst, size_t size)
185 {
186 	int i = 0;
187 	char *res;
188 
189 	switch (n) {
190 		case 0U ... 9UL:
191 			i = 0;
192 			break;
193 
194 		case 10U ... 99UL:
195 			i = 1;
196 			break;
197 
198 		case 100U ... 999UL:
199 			i = 2;
200 			break;
201 
202 		case 1000U ... 9999UL:
203 			i = 3;
204 			break;
205 
206 		case 10000U ... 99999UL:
207 			i = 4;
208 			break;
209 
210 		case 100000U ... 999999UL:
211 			i = 5;
212 			break;
213 
214 		case 1000000U ... 9999999UL:
215 			i = 6;
216 			break;
217 
218 		case 10000000U ... 99999999UL:
219 			i = 7;
220 			break;
221 
222 		case 100000000U ... 999999999UL:
223 			i = 8;
224 			break;
225 #if __WORDSIZE == 32
226 
227 		case 1000000000ULL ... ULONG_MAX:
228 			i = 9;
229 			break;
230 
231 #elif __WORDSIZE == 64
232 
233 		case 1000000000ULL ... 9999999999UL:
234 			i = 9;
235 			break;
236 
237 		case 10000000000ULL ... 99999999999UL:
238 			i = 10;
239 			break;
240 
241 		case 100000000000ULL ... 999999999999UL:
242 			i = 11;
243 			break;
244 
245 		case 1000000000000ULL ... 9999999999999UL:
246 			i = 12;
247 			break;
248 
249 		case 10000000000000ULL ... 99999999999999UL:
250 			i = 13;
251 			break;
252 
253 		case 100000000000000ULL ... 999999999999999UL:
254 			i = 14;
255 			break;
256 
257 		case 1000000000000000ULL ... 9999999999999999UL:
258 			i = 15;
259 			break;
260 
261 		case 10000000000000000ULL ... 99999999999999999UL:
262 			i = 16;
263 			break;
264 
265 		case 100000000000000000ULL ... 999999999999999999UL:
266 			i = 17;
267 			break;
268 
269 		case 1000000000000000000ULL ... 9999999999999999999UL:
270 			i = 18;
271 			break;
272 
273 		case 10000000000000000000ULL ... ULONG_MAX:
274 			i = 19;
275 			break;
276 
277 #endif
278 	}
279 	if (i + 2 > size) // (i + 1) + '\0'
280 		return NULL;  // too long
281 	res = dst + i + 1;
282 	*res = '\0';
283 	for (; i >= 0; i--) {
284 		dst[i] = n % 10U + '0';
285 		n /= 10U;
286 	}
287 	return res;
288 }
289 
290 /*
291  * signed long ASCII representation
292  *
293  * return the last char '\0' or NULL if no enough
294  * space in dst
295  */
ltoa_o(long int n,char * dst,size_t size)296 char *ltoa_o(long int n, char *dst, size_t size)
297 {
298 	char *pos = dst;
299 
300 	if (n < 0) {
301 		if (size < 3)
302 			return NULL; // min size is '-' + digit + '\0' but another test in ultoa
303 		*pos = '-';
304 		pos++;
305 		dst = ultoa_o(-n, pos, size - 1);
306 	} else {
307 		dst = ultoa_o(n, dst, size);
308 	}
309 	return dst;
310 }
311 
312 /*
313  * signed long long ASCII representation
314  *
315  * return the last char '\0' or NULL if no enough
316  * space in dst
317  */
lltoa(long long n,char * dst,size_t size)318 char *lltoa(long long n, char *dst, size_t size)
319 {
320 	char *pos = dst;
321 
322 	if (n < 0) {
323 		if (size < 3)
324 			return NULL; // min size is '-' + digit + '\0' but another test in ulltoa
325 		*pos = '-';
326 		pos++;
327 		dst = ulltoa(-n, pos, size - 1);
328 	} else {
329 		dst = ulltoa(n, dst, size);
330 	}
331 	return dst;
332 }
333 
334 /*
335  * write a ascii representation of a unsigned into dst,
336  * return a pointer to the last character
337  * Pad the ascii representation with '0', using size.
338  */
utoa_pad(unsigned int n,char * dst,size_t size)339 char *utoa_pad(unsigned int n, char *dst, size_t size)
340 {
341 	int i = 0;
342 	char *ret;
343 
344 	switch(n) {
345 		case 0U ... 9U:
346 			i = 0;
347 			break;
348 
349 		case 10U ... 99U:
350 			i = 1;
351 			break;
352 
353 		case 100U ... 999U:
354 			i = 2;
355 			break;
356 
357 		case 1000U ... 9999U:
358 			i = 3;
359 			break;
360 
361 		case 10000U ... 99999U:
362 			i = 4;
363 			break;
364 
365 		case 100000U ... 999999U:
366 			i = 5;
367 			break;
368 
369 		case 1000000U ... 9999999U:
370 			i = 6;
371 			break;
372 
373 		case 10000000U ... 99999999U:
374 			i = 7;
375 			break;
376 
377 		case 100000000U ... 999999999U:
378 			i = 8;
379 			break;
380 
381 		case 1000000000U ... 4294967295U:
382 			i = 9;
383 			break;
384 	}
385 	if (i + 2 > size) // (i + 1) + '\0'
386 		return NULL;  // too long
387 	if (i < size)
388 		i = size - 2; // padding - '\0'
389 
390 	ret = dst + i + 1;
391 	*ret = '\0';
392 	for (; i >= 0; i--) {
393 		dst[i] = n % 10U + '0';
394 		n /= 10U;
395 	}
396 	return ret;
397 }
398 
399 /*
400  * copies at most <size-1> chars from <src> to <dst>. Last char is always
401  * set to 0, unless <size> is 0. The number of chars copied is returned
402  * (excluding the terminating zero).
403  * This code has been optimized for size and speed : on x86, it's 45 bytes
404  * long, uses only registers, and consumes only 4 cycles per char.
405  */
strlcpy2(char * dst,const char * src,int size)406 int strlcpy2(char *dst, const char *src, int size)
407 {
408 	char *orig = dst;
409 	if (size) {
410 		while (--size && (*dst = *src)) {
411 			src++; dst++;
412 		}
413 		*dst = 0;
414 	}
415 	return dst - orig;
416 }
417 
418 /*
419  * This function simply returns a locally allocated string containing
420  * the ascii representation for number 'n' in decimal.
421  */
ultoa_r(unsigned long n,char * buffer,int size)422 char *ultoa_r(unsigned long n, char *buffer, int size)
423 {
424 	char *pos;
425 
426 	pos = buffer + size - 1;
427 	*pos-- = '\0';
428 
429 	do {
430 		*pos-- = '0' + n % 10;
431 		n /= 10;
432 	} while (n && pos >= buffer);
433 	return pos + 1;
434 }
435 
436 /*
437  * This function simply returns a locally allocated string containing
438  * the ascii representation for number 'n' in decimal.
439  */
lltoa_r(long long int in,char * buffer,int size)440 char *lltoa_r(long long int in, char *buffer, int size)
441 {
442 	char *pos;
443 	int neg = 0;
444 	unsigned long long int n;
445 
446 	pos = buffer + size - 1;
447 	*pos-- = '\0';
448 
449 	if (in < 0) {
450 		neg = 1;
451 		n = -in;
452 	}
453 	else
454 		n = in;
455 
456 	do {
457 		*pos-- = '0' + n % 10;
458 		n /= 10;
459 	} while (n && pos >= buffer);
460 	if (neg && pos > buffer)
461 		*pos-- = '-';
462 	return pos + 1;
463 }
464 
465 /*
466  * This function simply returns a locally allocated string containing
467  * the ascii representation for signed number 'n' in decimal.
468  */
sltoa_r(long n,char * buffer,int size)469 char *sltoa_r(long n, char *buffer, int size)
470 {
471 	char *pos;
472 
473 	if (n >= 0)
474 		return ultoa_r(n, buffer, size);
475 
476 	pos = ultoa_r(-n, buffer + 1, size - 1) - 1;
477 	*pos = '-';
478 	return pos;
479 }
480 
481 /*
482  * This function simply returns a locally allocated string containing
483  * the ascii representation for number 'n' in decimal, formatted for
484  * HTML output with tags to create visual grouping by 3 digits. The
485  * output needs to support at least 171 characters.
486  */
ulltoh_r(unsigned long long n,char * buffer,int size)487 const char *ulltoh_r(unsigned long long n, char *buffer, int size)
488 {
489 	char *start;
490 	int digit = 0;
491 
492 	start = buffer + size;
493 	*--start = '\0';
494 
495 	do {
496 		if (digit == 3 && start >= buffer + 7)
497 			memcpy(start -= 7, "</span>", 7);
498 
499 		if (start >= buffer + 1) {
500 			*--start = '0' + n % 10;
501 			n /= 10;
502 		}
503 
504 		if (digit == 3 && start >= buffer + 18)
505 			memcpy(start -= 18, "<span class=\"rls\">", 18);
506 
507 		if (digit++ == 3)
508 			digit = 1;
509 	} while (n && start > buffer);
510 	return start;
511 }
512 
513 /*
514  * This function simply returns a locally allocated string containing the ascii
515  * representation for number 'n' in decimal, unless n is 0 in which case it
516  * returns the alternate string (or an empty string if the alternate string is
517  * NULL). It use is intended for limits reported in reports, where it's
518  * desirable not to display anything if there is no limit. Warning! it shares
519  * the same vector as ultoa_r().
520  */
limit_r(unsigned long n,char * buffer,int size,const char * alt)521 const char *limit_r(unsigned long n, char *buffer, int size, const char *alt)
522 {
523 	return (n) ? ultoa_r(n, buffer, size) : (alt ? alt : "");
524 }
525 
526 /* returns a locally allocated string containing the quoted encoding of the
527  * input string. The output may be truncated to QSTR_SIZE chars, but it is
528  * guaranteed that the string will always be properly terminated. Quotes are
529  * encoded by doubling them as is commonly done in CSV files. QSTR_SIZE must
530  * always be at least 4 chars.
531  */
qstr(const char * str)532 const char *qstr(const char *str)
533 {
534 	char *ret = quoted_str[quoted_idx];
535 	char *p, *end;
536 
537 	if (++quoted_idx >= NB_QSTR)
538 		quoted_idx = 0;
539 
540 	p = ret;
541 	end = ret + QSTR_SIZE;
542 
543 	*p++ = '"';
544 
545 	/* always keep 3 chars to support passing "" and the ending " */
546 	while (*str && p < end - 3) {
547 		if (*str == '"') {
548 			*p++ = '"';
549 			*p++ = '"';
550 		}
551 		else
552 			*p++ = *str;
553 		str++;
554 	}
555 	*p++ = '"';
556 	return ret;
557 }
558 
559 /*
560  * Returns non-zero if character <s> is a hex digit (0-9, a-f, A-F), else zero.
561  *
562  * It looks like this one would be a good candidate for inlining, but this is
563  * not interesting because it around 35 bytes long and often called multiple
564  * times within the same function.
565  */
ishex(char s)566 int ishex(char s)
567 {
568 	s -= '0';
569 	if ((unsigned char)s <= 9)
570 		return 1;
571 	s -= 'A' - '0';
572 	if ((unsigned char)s <= 5)
573 		return 1;
574 	s -= 'a' - 'A';
575 	if ((unsigned char)s <= 5)
576 		return 1;
577 	return 0;
578 }
579 
580 /* rounds <i> down to the closest value having max 2 digits */
round_2dig(unsigned int i)581 unsigned int round_2dig(unsigned int i)
582 {
583 	unsigned int mul = 1;
584 
585 	while (i >= 100) {
586 		i /= 10;
587 		mul *= 10;
588 	}
589 	return i * mul;
590 }
591 
592 /*
593  * Checks <name> for invalid characters. Valid chars are [A-Za-z0-9_:.-]. If an
594  * invalid character is found, a pointer to it is returned. If everything is
595  * fine, NULL is returned.
596  */
invalid_char(const char * name)597 const char *invalid_char(const char *name)
598 {
599 	if (!*name)
600 		return name;
601 
602 	while (*name) {
603 		if (!isalnum((int)(unsigned char)*name) && *name != '.' && *name != ':' &&
604 		    *name != '_' && *name != '-')
605 			return name;
606 		name++;
607 	}
608 	return NULL;
609 }
610 
611 /*
612  * Checks <name> for invalid characters. Valid chars are [_.-] and those
613  * accepted by <f> function.
614  * If an invalid character is found, a pointer to it is returned.
615  * If everything is fine, NULL is returned.
616  */
__invalid_char(const char * name,int (* f)(int))617 static inline const char *__invalid_char(const char *name, int (*f)(int)) {
618 
619 	if (!*name)
620 		return name;
621 
622 	while (*name) {
623 		if (!f((int)(unsigned char)*name) && *name != '.' &&
624 		    *name != '_' && *name != '-')
625 			return name;
626 
627 		name++;
628 	}
629 
630 	return NULL;
631 }
632 
633 /*
634  * Checks <name> for invalid characters. Valid chars are [A-Za-z0-9_.-].
635  * If an invalid character is found, a pointer to it is returned.
636  * If everything is fine, NULL is returned.
637  */
invalid_domainchar(const char * name)638 const char *invalid_domainchar(const char *name) {
639 	return __invalid_char(name, isalnum);
640 }
641 
642 /*
643  * Checks <name> for invalid characters. Valid chars are [A-Za-z_.-].
644  * If an invalid character is found, a pointer to it is returned.
645  * If everything is fine, NULL is returned.
646  */
invalid_prefix_char(const char * name)647 const char *invalid_prefix_char(const char *name) {
648 	return __invalid_char(name, isalnum);
649 }
650 
651 /*
652  * converts <str> to a struct sockaddr_storage* provided by the caller. The
653  * caller must have zeroed <sa> first, and may have set sa->ss_family to force
654  * parse a specific address format. If the ss_family is 0 or AF_UNSPEC, then
655  * the function tries to guess the address family from the syntax. If the
656  * family is forced and the format doesn't match, an error is returned. The
657  * string is assumed to contain only an address, no port. The address can be a
658  * dotted IPv4 address, an IPv6 address, a host name, or empty or "*" to
659  * indicate INADDR_ANY. NULL is returned if the host part cannot be resolved.
660  * The return address will only have the address family and the address set,
661  * all other fields remain zero. The string is not supposed to be modified.
662  * The IPv6 '::' address is IN6ADDR_ANY. If <resolve> is non-zero, the hostname
663  * is resolved, otherwise only IP addresses are resolved, and anything else
664  * returns NULL. If the address contains a port, this one is preserved.
665  */
str2ip2(const char * str,struct sockaddr_storage * sa,int resolve)666 struct sockaddr_storage *str2ip2(const char *str, struct sockaddr_storage *sa, int resolve)
667 {
668 	struct hostent *he;
669 	/* max IPv6 length, including brackets and terminating NULL */
670 	char tmpip[48];
671 	int port = get_host_port(sa);
672 
673 	/* check IPv6 with square brackets */
674 	if (str[0] == '[') {
675 		size_t iplength = strlen(str);
676 
677 		if (iplength < 4) {
678 			/* minimal size is 4 when using brackets "[::]" */
679 			goto fail;
680 		}
681 		else if (iplength >= sizeof(tmpip)) {
682 			/* IPv6 literal can not be larger than tmpip */
683 			goto fail;
684 		}
685 		else {
686 			if (str[iplength - 1] != ']') {
687 				/* if address started with bracket, it should end with bracket */
688 				goto fail;
689 			}
690 			else {
691 				memcpy(tmpip, str + 1, iplength - 2);
692 				tmpip[iplength - 2] = '\0';
693 				str = tmpip;
694 			}
695 		}
696 	}
697 
698 	/* Any IPv6 address */
699 	if (str[0] == ':' && str[1] == ':' && !str[2]) {
700 		if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
701 			sa->ss_family = AF_INET6;
702 		else if (sa->ss_family != AF_INET6)
703 			goto fail;
704 		set_host_port(sa, port);
705 		return sa;
706 	}
707 
708 	/* Any address for the family, defaults to IPv4 */
709 	if (!str[0] || (str[0] == '*' && !str[1])) {
710 		if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
711 			sa->ss_family = AF_INET;
712 		set_host_port(sa, port);
713 		return sa;
714 	}
715 
716 	/* check for IPv6 first */
717 	if ((!sa->ss_family || sa->ss_family == AF_UNSPEC || sa->ss_family == AF_INET6) &&
718 	    inet_pton(AF_INET6, str, &((struct sockaddr_in6 *)sa)->sin6_addr)) {
719 		sa->ss_family = AF_INET6;
720 		set_host_port(sa, port);
721 		return sa;
722 	}
723 
724 	/* then check for IPv4 */
725 	if ((!sa->ss_family || sa->ss_family == AF_UNSPEC || sa->ss_family == AF_INET) &&
726 	    inet_pton(AF_INET, str, &((struct sockaddr_in *)sa)->sin_addr)) {
727 		sa->ss_family = AF_INET;
728 		set_host_port(sa, port);
729 		return sa;
730 	}
731 
732 	if (!resolve)
733 		return NULL;
734 
735 	if (!dns_hostname_validation(str, NULL))
736 		return NULL;
737 
738 #ifdef USE_GETADDRINFO
739 	if (global.tune.options & GTUNE_USE_GAI) {
740 		struct addrinfo hints, *result;
741 		int success = 0;
742 
743 		memset(&result, 0, sizeof(result));
744 		memset(&hints, 0, sizeof(hints));
745 		hints.ai_family = sa->ss_family ? sa->ss_family : AF_UNSPEC;
746 		hints.ai_socktype = SOCK_DGRAM;
747 		hints.ai_flags = 0;
748 		hints.ai_protocol = 0;
749 
750 		if (getaddrinfo(str, NULL, &hints, &result) == 0) {
751 			if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
752 				sa->ss_family = result->ai_family;
753 			else if (sa->ss_family != result->ai_family) {
754 				freeaddrinfo(result);
755 				goto fail;
756 			}
757 
758 			switch (result->ai_family) {
759 			case AF_INET:
760 				memcpy((struct sockaddr_in *)sa, result->ai_addr, result->ai_addrlen);
761 				set_host_port(sa, port);
762 				success = 1;
763 				break;
764 			case AF_INET6:
765 				memcpy((struct sockaddr_in6 *)sa, result->ai_addr, result->ai_addrlen);
766 				set_host_port(sa, port);
767 				success = 1;
768 				break;
769 			}
770 		}
771 
772 		if (result)
773 			freeaddrinfo(result);
774 
775 		if (success)
776 			return sa;
777 	}
778 #endif
779 	/* try to resolve an IPv4/IPv6 hostname */
780 	he = gethostbyname(str);
781 	if (he) {
782 		if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
783 			sa->ss_family = he->h_addrtype;
784 		else if (sa->ss_family != he->h_addrtype)
785 			goto fail;
786 
787 		switch (sa->ss_family) {
788 		case AF_INET:
789 			((struct sockaddr_in *)sa)->sin_addr = *(struct in_addr *) *(he->h_addr_list);
790 			set_host_port(sa, port);
791 			return sa;
792 		case AF_INET6:
793 			((struct sockaddr_in6 *)sa)->sin6_addr = *(struct in6_addr *) *(he->h_addr_list);
794 			set_host_port(sa, port);
795 			return sa;
796 		}
797 	}
798 
799 	/* unsupported address family */
800  fail:
801 	return NULL;
802 }
803 
804 /*
805  * Converts <str> to a locally allocated struct sockaddr_storage *, and a port
806  * range or offset consisting in two integers that the caller will have to
807  * check to find the relevant input format. The following format are supported :
808  *
809  *   String format           | address |  port  |  low   |  high
810  *    addr                   | <addr>  |   0    |   0    |   0
811  *    addr:                  | <addr>  |   0    |   0    |   0
812  *    addr:port              | <addr>  | <port> | <port> | <port>
813  *    addr:pl-ph             | <addr>  |  <pl>  |  <pl>  |  <ph>
814  *    addr:+port             | <addr>  | <port> |   0    | <port>
815  *    addr:-port             | <addr>  |-<port> | <port> |   0
816  *
817  * The detection of a port range or increment by the caller is made by
818  * comparing <low> and <high>. If both are equal, then port 0 means no port
819  * was specified. The caller may pass NULL for <low> and <high> if it is not
820  * interested in retrieving port ranges.
821  *
822  * Note that <addr> above may also be :
823  *    - empty ("")  => family will be AF_INET and address will be INADDR_ANY
824  *    - "*"         => family will be AF_INET and address will be INADDR_ANY
825  *    - "::"        => family will be AF_INET6 and address will be IN6ADDR_ANY
826  *    - a host name => family and address will depend on host name resolving.
827  *
828  * A prefix may be passed in before the address above to force the family :
829  *    - "ipv4@"  => force address to resolve as IPv4 and fail if not possible.
830  *    - "ipv6@"  => force address to resolve as IPv6 and fail if not possible.
831  *    - "unix@"  => force address to be a path to a UNIX socket even if the
832  *                  path does not start with a '/'
833  *    - 'abns@'  -> force address to belong to the abstract namespace (Linux
834  *                  only). These sockets are just like Unix sockets but without
835  *                  the need for an underlying file system. The address is a
836  *                  string. Technically it's like a Unix socket with a zero in
837  *                  the first byte of the address.
838  *    - "fd@"    => an integer must follow, and is a file descriptor number.
839  *
840  * IPv6 addresses can be declared with or without square brackets. When using
841  * square brackets for IPv6 addresses, the port separator (colon) is optional.
842  * If not using square brackets, and in order to avoid any ambiguity with
843  * IPv6 addresses, the last colon ':' is mandatory even when no port is specified.
844  * NULL is returned if the address cannot be parsed. The <low> and <high> ports
845  * are always initialized if non-null, even for non-IP families.
846  *
847  * If <pfx> is non-null, it is used as a string prefix before any path-based
848  * address (typically the path to a unix socket).
849  *
850  * if <fqdn> is non-null, it will be filled with :
851  *   - a pointer to the FQDN of the server name to resolve if there's one, and
852  *     that the caller will have to free(),
853  *   - NULL if there was an explicit address that doesn't require resolution.
854  *
855  * Hostnames are only resolved if <resolve> is non-null. Note that if <resolve>
856  * is null, <fqdn> is still honnored so it is possible for the caller to know
857  * whether a resolution failed by setting <resolve> to null and checking if
858  * <fqdn> was filled, indicating the need for a resolution.
859  *
860  * When a file descriptor is passed, its value is put into the s_addr part of
861  * the address when cast to sockaddr_in and the address family is AF_UNSPEC.
862  */
str2sa_range(const char * str,int * port,int * low,int * high,char ** err,const char * pfx,char ** fqdn,int resolve)863 struct sockaddr_storage *str2sa_range(const char *str, int *port, int *low, int *high, char **err, const char *pfx, char **fqdn, int resolve)
864 {
865 	static THREAD_LOCAL struct sockaddr_storage ss;
866 	struct sockaddr_storage *ret = NULL;
867 	char *back, *str2;
868 	char *port1, *port2;
869 	int portl, porth, porta;
870 	int abstract = 0;
871 
872 	portl = porth = porta = 0;
873 	if (fqdn)
874 		*fqdn = NULL;
875 
876 	str2 = back = env_expand(strdup(str));
877 	if (str2 == NULL) {
878 		memprintf(err, "out of memory in '%s'\n", __FUNCTION__);
879 		goto out;
880 	}
881 
882 	if (!*str2) {
883 		memprintf(err, "'%s' resolves to an empty address (environment variable missing?)\n", str);
884 		goto out;
885 	}
886 
887 	memset(&ss, 0, sizeof(ss));
888 
889 	if (strncmp(str2, "unix@", 5) == 0) {
890 		str2 += 5;
891 		abstract = 0;
892 		ss.ss_family = AF_UNIX;
893 	}
894 	else if (strncmp(str2, "abns@", 5) == 0) {
895 		str2 += 5;
896 		abstract = 1;
897 		ss.ss_family = AF_UNIX;
898 	}
899 	else if (strncmp(str2, "ipv4@", 5) == 0) {
900 		str2 += 5;
901 		ss.ss_family = AF_INET;
902 	}
903 	else if (strncmp(str2, "ipv6@", 5) == 0) {
904 		str2 += 5;
905 		ss.ss_family = AF_INET6;
906 	}
907 	else if (*str2 == '/') {
908 		ss.ss_family = AF_UNIX;
909 	}
910 	else
911 		ss.ss_family = AF_UNSPEC;
912 
913 	if (ss.ss_family == AF_UNSPEC && strncmp(str2, "sockpair@", 9) == 0) {
914 		char *endptr;
915 
916 		str2 += 9;
917 
918 		((struct sockaddr_in *)&ss)->sin_addr.s_addr = strtol(str2, &endptr, 10);
919 		((struct sockaddr_in *)&ss)->sin_port = 0;
920 
921 		if (!*str2 || *endptr) {
922 			memprintf(err, "file descriptor '%s' is not a valid integer in '%s'\n", str2, str);
923 			goto out;
924 		}
925 
926 		ss.ss_family = AF_CUST_SOCKPAIR;
927 
928 	}
929 	else if (ss.ss_family == AF_UNSPEC && strncmp(str2, "fd@", 3) == 0) {
930 		char *endptr;
931 
932 		str2 += 3;
933 		((struct sockaddr_in *)&ss)->sin_addr.s_addr = strtol(str2, &endptr, 10);
934 		((struct sockaddr_in *)&ss)->sin_port = 0;
935 
936 		if (!*str2 || *endptr) {
937 			memprintf(err, "file descriptor '%s' is not a valid integer in '%s'\n", str2, str);
938 			goto out;
939 		}
940 
941 		/* we return AF_UNSPEC if we use a file descriptor number */
942 		ss.ss_family = AF_UNSPEC;
943 	}
944 	else if (ss.ss_family == AF_UNIX) {
945 		struct sockaddr_un *un = (struct sockaddr_un *)&ss;
946 		int prefix_path_len;
947 		int max_path_len;
948 		int adr_len;
949 
950 		/* complete unix socket path name during startup or soft-restart is
951 		 * <unix_bind_prefix><path>.<pid>.<bak|tmp>
952 		 */
953 		prefix_path_len = (pfx && !abstract) ? strlen(pfx) : 0;
954 		max_path_len = (sizeof(un->sun_path) - 1) -
955 			(abstract ? 0 : prefix_path_len + 1 + 5 + 1 + 3);
956 
957 		adr_len = strlen(str2);
958 		if (adr_len > max_path_len) {
959 			memprintf(err, "socket path '%s' too long (max %d)\n", str, max_path_len);
960 			goto out;
961 		}
962 
963 		/* when abstract==1, we skip the first zero and copy all bytes except the trailing zero */
964 		memset(un->sun_path, 0, sizeof(un->sun_path));
965 		if (prefix_path_len)
966 			memcpy(un->sun_path, pfx, prefix_path_len);
967 		memcpy(un->sun_path + prefix_path_len + abstract, str2, adr_len + 1 - abstract);
968 	}
969 	else { /* IPv4 and IPv6 */
970 		char *end = str2 + strlen(str2);
971 		char *chr;
972 
973 		/* search for : or ] whatever comes first */
974 		for (chr = end-1; chr > str2; chr--) {
975 			if (*chr == ']' || *chr == ':')
976 				break;
977 		}
978 
979 		if (*chr == ':') {
980 			/* Found a colon before a closing-bracket, must be a port separator.
981 			 * This guarantee backward compatibility.
982 			 */
983 			*chr++ = '\0';
984 			port1 = chr;
985 		}
986 		else {
987 			/* Either no colon and no closing-bracket
988 			 * or directly ending with a closing-bracket.
989 			 * However, no port.
990 			 */
991 			port1 = "";
992 		}
993 
994 		if (isdigit((int)(unsigned char)*port1)) {	/* single port or range */
995 			port2 = strchr(port1, '-');
996 			if (port2)
997 				*port2++ = '\0';
998 			else
999 				port2 = port1;
1000 			portl = atoi(port1);
1001 			porth = atoi(port2);
1002 			porta = portl;
1003 		}
1004 		else if (*port1 == '-') { /* negative offset */
1005 			portl = atoi(port1 + 1);
1006 			porta = -portl;
1007 		}
1008 		else if (*port1 == '+') { /* positive offset */
1009 			porth = atoi(port1 + 1);
1010 			porta = porth;
1011 		}
1012 		else if (*port1) { /* other any unexpected char */
1013 			memprintf(err, "invalid character '%c' in port number '%s' in '%s'\n", *port1, port1, str);
1014 			goto out;
1015 		}
1016 
1017 		/* first try to parse the IP without resolving. If it fails, it
1018 		 * tells us we need to keep a copy of the FQDN to resolve later
1019 		 * and to enable DNS. In this case we can proceed if <fqdn> is
1020 		 * set or if resolve is set, otherwise it's an error.
1021 		 */
1022 		if (str2ip2(str2, &ss, 0) == NULL) {
1023 			if ((!resolve && !fqdn) ||
1024 				 (resolve && str2ip2(str2, &ss, 1) == NULL)) {
1025 				memprintf(err, "invalid address: '%s' in '%s'\n", str2, str);
1026 				goto out;
1027 			}
1028 
1029 			if (fqdn) {
1030 				if (str2 != back)
1031 					memmove(back, str2, strlen(str2) + 1);
1032 				*fqdn = back;
1033 				back = NULL;
1034 			}
1035 		}
1036 		set_host_port(&ss, porta);
1037 	}
1038 
1039 	ret = &ss;
1040  out:
1041 	if (port)
1042 		*port = porta;
1043 	if (low)
1044 		*low = portl;
1045 	if (high)
1046 		*high = porth;
1047 	free(back);
1048 	return ret;
1049 }
1050 
1051 /* converts <str> to a struct in_addr containing a network mask. It can be
1052  * passed in dotted form (255.255.255.0) or in CIDR form (24). It returns 1
1053  * if the conversion succeeds otherwise zero.
1054  */
str2mask(const char * str,struct in_addr * mask)1055 int str2mask(const char *str, struct in_addr *mask)
1056 {
1057 	if (strchr(str, '.') != NULL) {	    /* dotted notation */
1058 		if (!inet_pton(AF_INET, str, mask))
1059 			return 0;
1060 	}
1061 	else { /* mask length */
1062 		char *err;
1063 		unsigned long len = strtol(str, &err, 10);
1064 
1065 		if (!*str || (err && *err) || (unsigned)len > 32)
1066 			return 0;
1067 
1068 		len2mask4(len, mask);
1069 	}
1070 	return 1;
1071 }
1072 
1073 /* converts <str> to a struct in6_addr containing a network mask. It can be
1074  * passed in quadruplet form (ffff:ffff::) or in CIDR form (64). It returns 1
1075  * if the conversion succeeds otherwise zero.
1076  */
str2mask6(const char * str,struct in6_addr * mask)1077 int str2mask6(const char *str, struct in6_addr *mask)
1078 {
1079 	if (strchr(str, ':') != NULL) {	    /* quadruplet notation */
1080 		if (!inet_pton(AF_INET6, str, mask))
1081 			return 0;
1082 	}
1083 	else { /* mask length */
1084 		char *err;
1085 		unsigned long len = strtol(str, &err, 10);
1086 
1087 		if (!*str || (err && *err) || (unsigned)len > 128)
1088 			return 0;
1089 
1090 		len2mask6(len, mask);
1091 	}
1092 	return 1;
1093 }
1094 
1095 /* convert <cidr> to struct in_addr <mask>. It returns 1 if the conversion
1096  * succeeds otherwise zero.
1097  */
cidr2dotted(int cidr,struct in_addr * mask)1098 int cidr2dotted(int cidr, struct in_addr *mask) {
1099 
1100 	if (cidr < 0 || cidr > 32)
1101 		return 0;
1102 
1103 	mask->s_addr = cidr ? htonl(~0UL << (32 - cidr)) : 0;
1104 	return 1;
1105 }
1106 
1107 /* Convert mask from bit length form to in_addr form.
1108  * This function never fails.
1109  */
len2mask4(int len,struct in_addr * addr)1110 void len2mask4(int len, struct in_addr *addr)
1111 {
1112 	if (len >= 32) {
1113 		addr->s_addr = 0xffffffff;
1114 		return;
1115 	}
1116 	if (len <= 0) {
1117 		addr->s_addr = 0x00000000;
1118 		return;
1119 	}
1120 	addr->s_addr = 0xffffffff << (32 - len);
1121 	addr->s_addr = htonl(addr->s_addr);
1122 }
1123 
1124 /* Convert mask from bit length form to in6_addr form.
1125  * This function never fails.
1126  */
len2mask6(int len,struct in6_addr * addr)1127 void len2mask6(int len, struct in6_addr *addr)
1128 {
1129 	len2mask4(len, (struct in_addr *)&addr->s6_addr[0]); /* msb */
1130 	len -= 32;
1131 	len2mask4(len, (struct in_addr *)&addr->s6_addr[4]);
1132 	len -= 32;
1133 	len2mask4(len, (struct in_addr *)&addr->s6_addr[8]);
1134 	len -= 32;
1135 	len2mask4(len, (struct in_addr *)&addr->s6_addr[12]); /* lsb */
1136 }
1137 
1138 /*
1139  * converts <str> to two struct in_addr* which must be pre-allocated.
1140  * The format is "addr[/mask]", where "addr" cannot be empty, and mask
1141  * is optionnal and either in the dotted or CIDR notation.
1142  * Note: "addr" can also be a hostname. Returns 1 if OK, 0 if error.
1143  */
str2net(const char * str,int resolve,struct in_addr * addr,struct in_addr * mask)1144 int str2net(const char *str, int resolve, struct in_addr *addr, struct in_addr *mask)
1145 {
1146 	__label__ out_free, out_err;
1147 	char *c, *s;
1148 	int ret_val;
1149 
1150 	s = strdup(str);
1151 	if (!s)
1152 		return 0;
1153 
1154 	memset(mask, 0, sizeof(*mask));
1155 	memset(addr, 0, sizeof(*addr));
1156 
1157 	if ((c = strrchr(s, '/')) != NULL) {
1158 		*c++ = '\0';
1159 		/* c points to the mask */
1160 		if (!str2mask(c, mask))
1161 			goto out_err;
1162 	}
1163 	else {
1164 		mask->s_addr = ~0U;
1165 	}
1166 	if (!inet_pton(AF_INET, s, addr)) {
1167 		struct hostent *he;
1168 
1169 		if (!resolve)
1170 			goto out_err;
1171 
1172 		if ((he = gethostbyname(s)) == NULL) {
1173 			goto out_err;
1174 		}
1175 		else
1176 			*addr = *(struct in_addr *) *(he->h_addr_list);
1177 	}
1178 
1179 	ret_val = 1;
1180  out_free:
1181 	free(s);
1182 	return ret_val;
1183  out_err:
1184 	ret_val = 0;
1185 	goto out_free;
1186 }
1187 
1188 
1189 /*
1190  * converts <str> to two struct in6_addr* which must be pre-allocated.
1191  * The format is "addr[/mask]", where "addr" cannot be empty, and mask
1192  * is an optionnal number of bits (128 being the default).
1193  * Returns 1 if OK, 0 if error.
1194  */
str62net(const char * str,struct in6_addr * addr,unsigned char * mask)1195 int str62net(const char *str, struct in6_addr *addr, unsigned char *mask)
1196 {
1197 	char *c, *s;
1198 	int ret_val = 0;
1199 	char *err;
1200 	unsigned long len = 128;
1201 
1202 	s = strdup(str);
1203 	if (!s)
1204 		return 0;
1205 
1206 	memset(mask, 0, sizeof(*mask));
1207 	memset(addr, 0, sizeof(*addr));
1208 
1209 	if ((c = strrchr(s, '/')) != NULL) {
1210 		*c++ = '\0'; /* c points to the mask */
1211 		if (!*c)
1212 			goto out_free;
1213 
1214 		len = strtoul(c, &err, 10);
1215 		if ((err && *err) || (unsigned)len > 128)
1216 			goto out_free;
1217 	}
1218 	*mask = len; /* OK we have a valid mask in <len> */
1219 
1220 	if (!inet_pton(AF_INET6, s, addr))
1221 		goto out_free;
1222 
1223 	ret_val = 1;
1224  out_free:
1225 	free(s);
1226 	return ret_val;
1227 }
1228 
1229 
1230 /*
1231  * Parse IPv4 address found in url.
1232  */
url2ipv4(const char * addr,struct in_addr * dst)1233 int url2ipv4(const char *addr, struct in_addr *dst)
1234 {
1235 	int saw_digit, octets, ch;
1236 	u_char tmp[4], *tp;
1237 	const char *cp = addr;
1238 
1239 	saw_digit = 0;
1240 	octets = 0;
1241 	*(tp = tmp) = 0;
1242 
1243 	while (*addr) {
1244 		unsigned char digit = (ch = *addr++) - '0';
1245 		if (digit > 9 && ch != '.')
1246 			break;
1247 		if (digit <= 9) {
1248 			u_int new = *tp * 10 + digit;
1249 			if (new > 255)
1250 				return 0;
1251 			*tp = new;
1252 			if (!saw_digit) {
1253 				if (++octets > 4)
1254 					return 0;
1255 				saw_digit = 1;
1256 			}
1257 		} else if (ch == '.' && saw_digit) {
1258 			if (octets == 4)
1259 				return 0;
1260 			*++tp = 0;
1261 			saw_digit = 0;
1262 		} else
1263 			return 0;
1264 	}
1265 
1266 	if (octets < 4)
1267 		return 0;
1268 
1269 	memcpy(&dst->s_addr, tmp, 4);
1270 	return addr-cp-1;
1271 }
1272 
1273 /*
1274  * Resolve destination server from URL. Convert <str> to a sockaddr_storage.
1275  * <out> contain the code of the dectected scheme, the start and length of
1276  * the hostname. Actually only http and https are supported. <out> can be NULL.
1277  * This function returns the consumed length. It is useful if you parse complete
1278  * url like http://host:port/path, because the consumed length corresponds to
1279  * the first character of the path. If the conversion fails, it returns -1.
1280  *
1281  * This function tries to resolve the DNS name if haproxy is in starting mode.
1282  * So, this function may be used during the configuration parsing.
1283  */
url2sa(const char * url,int ulen,struct sockaddr_storage * addr,struct split_url * out)1284 int url2sa(const char *url, int ulen, struct sockaddr_storage *addr, struct split_url *out)
1285 {
1286 	const char *curr = url, *cp = url;
1287 	const char *end;
1288 	int ret, url_code = 0;
1289 	unsigned long long int http_code = 0;
1290 	int default_port;
1291 	struct hostent *he;
1292 	char *p;
1293 
1294 	/* Firstly, try to find :// pattern */
1295 	while (curr < url+ulen && url_code != 0x3a2f2f) {
1296 		url_code = ((url_code & 0xffff) << 8);
1297 		url_code += (unsigned char)*curr++;
1298 	}
1299 
1300 	/* Secondly, if :// pattern is found, verify parsed stuff
1301 	 * before pattern is matching our http pattern.
1302 	 * If so parse ip address and port in uri.
1303 	 *
1304 	 * WARNING: Current code doesn't support dynamic async dns resolver.
1305 	 */
1306 	if (url_code != 0x3a2f2f)
1307 		return -1;
1308 
1309 	/* Copy scheme, and utrn to lower case. */
1310 	while (cp < curr - 3)
1311 		http_code = (http_code << 8) + *cp++;
1312 	http_code |= 0x2020202020202020ULL;			/* Turn everything to lower case */
1313 
1314 	/* HTTP or HTTPS url matching */
1315 	if (http_code == 0x2020202068747470ULL) {
1316 		default_port = 80;
1317 		if (out)
1318 			out->scheme = SCH_HTTP;
1319 	}
1320 	else if (http_code == 0x2020206874747073ULL) {
1321 		default_port = 443;
1322 		if (out)
1323 			out->scheme = SCH_HTTPS;
1324 	}
1325 	else
1326 		return -1;
1327 
1328 	/* If the next char is '[', the host address is IPv6. */
1329 	if (*curr == '[') {
1330 		curr++;
1331 
1332 		/* Check trash size */
1333 		if (trash.size < ulen)
1334 			return -1;
1335 
1336 		/* Look for ']' and copy the address in a trash buffer. */
1337 		p = trash.area;
1338 		for (end = curr;
1339 		     end < url + ulen && *end != ']';
1340 		     end++, p++)
1341 			*p = *end;
1342 		if (*end != ']')
1343 			return -1;
1344 		*p = '\0';
1345 
1346 		/* Update out. */
1347 		if (out) {
1348 			out->host = curr;
1349 			out->host_len = end - curr;
1350 		}
1351 
1352 		/* Try IPv6 decoding. */
1353 		if (!inet_pton(AF_INET6, trash.area, &((struct sockaddr_in6 *)addr)->sin6_addr))
1354 			return -1;
1355 		end++;
1356 
1357 		/* Decode port. */
1358 		if (*end == ':') {
1359 			end++;
1360 			default_port = read_uint(&end, url + ulen);
1361 		}
1362 		((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1363 		((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1364 		return end - url;
1365 	}
1366 	else {
1367 		/* We are looking for IP address. If you want to parse and
1368 		 * resolve hostname found in url, you can use str2sa_range(), but
1369 		 * be warned this can slow down global daemon performances
1370 		 * while handling lagging dns responses.
1371 		 */
1372 		ret = url2ipv4(curr, &((struct sockaddr_in *)addr)->sin_addr);
1373 		if (ret) {
1374 			/* Update out. */
1375 			if (out) {
1376 				out->host = curr;
1377 				out->host_len = ret;
1378 			}
1379 
1380 			curr += ret;
1381 
1382 			/* Decode port. */
1383 			if (*curr == ':') {
1384 				curr++;
1385 				default_port = read_uint(&curr, url + ulen);
1386 			}
1387 			((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1388 
1389 			/* Set family. */
1390 			((struct sockaddr_in *)addr)->sin_family = AF_INET;
1391 			return curr - url;
1392 		}
1393 		else if (global.mode & MODE_STARTING) {
1394 			/* The IPv4 and IPv6 decoding fails, maybe the url contain name. Try to execute
1395 			 * synchronous DNS request only if HAProxy is in the start state.
1396 			 */
1397 
1398 			/* look for : or / or end */
1399 			for (end = curr;
1400 			     end < url + ulen && *end != '/' && *end != ':';
1401 			     end++);
1402 			memcpy(trash.area, curr, end - curr);
1403 			trash.area[end - curr] = '\0';
1404 
1405 			/* try to resolve an IPv4/IPv6 hostname */
1406 			he = gethostbyname(trash.area);
1407 			if (!he)
1408 				return -1;
1409 
1410 			/* Update out. */
1411 			if (out) {
1412 				out->host = curr;
1413 				out->host_len = end - curr;
1414 			}
1415 
1416 			/* Decode port. */
1417 			if (*end == ':') {
1418 				end++;
1419 				default_port = read_uint(&end, url + ulen);
1420 			}
1421 
1422 			/* Copy IP address, set port and family. */
1423 			switch (he->h_addrtype) {
1424 			case AF_INET:
1425 				((struct sockaddr_in *)addr)->sin_addr = *(struct in_addr *) *(he->h_addr_list);
1426 				((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1427 				((struct sockaddr_in *)addr)->sin_family = AF_INET;
1428 				return end - url;
1429 
1430 			case AF_INET6:
1431 				((struct sockaddr_in6 *)addr)->sin6_addr = *(struct in6_addr *) *(he->h_addr_list);
1432 				((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1433 				((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1434 				return end - url;
1435 			}
1436 		}
1437 	}
1438 	return -1;
1439 }
1440 
1441 /* Tries to convert a sockaddr_storage address to text form. Upon success, the
1442  * address family is returned so that it's easy for the caller to adapt to the
1443  * output format. Zero is returned if the address family is not supported. -1
1444  * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1445  * supported.
1446  */
addr_to_str(const struct sockaddr_storage * addr,char * str,int size)1447 int addr_to_str(const struct sockaddr_storage *addr, char *str, int size)
1448 {
1449 
1450 	const void *ptr;
1451 
1452 	if (size < 5)
1453 		return 0;
1454 	*str = '\0';
1455 
1456 	switch (addr->ss_family) {
1457 	case AF_INET:
1458 		ptr = &((struct sockaddr_in *)addr)->sin_addr;
1459 		break;
1460 	case AF_INET6:
1461 		ptr = &((struct sockaddr_in6 *)addr)->sin6_addr;
1462 		break;
1463 	case AF_UNIX:
1464 		memcpy(str, "unix", 5);
1465 		return addr->ss_family;
1466 	default:
1467 		return 0;
1468 	}
1469 
1470 	if (inet_ntop(addr->ss_family, ptr, str, size))
1471 		return addr->ss_family;
1472 
1473 	/* failed */
1474 	return -1;
1475 }
1476 
1477 /* Tries to convert a sockaddr_storage port to text form. Upon success, the
1478  * address family is returned so that it's easy for the caller to adapt to the
1479  * output format. Zero is returned if the address family is not supported. -1
1480  * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1481  * supported.
1482  */
port_to_str(const struct sockaddr_storage * addr,char * str,int size)1483 int port_to_str(const struct sockaddr_storage *addr, char *str, int size)
1484 {
1485 
1486 	uint16_t port;
1487 
1488 
1489 	if (size < 6)
1490 		return 0;
1491 	*str = '\0';
1492 
1493 	switch (addr->ss_family) {
1494 	case AF_INET:
1495 		port = ((struct sockaddr_in *)addr)->sin_port;
1496 		break;
1497 	case AF_INET6:
1498 		port = ((struct sockaddr_in6 *)addr)->sin6_port;
1499 		break;
1500 	case AF_UNIX:
1501 		memcpy(str, "unix", 5);
1502 		return addr->ss_family;
1503 	default:
1504 		return 0;
1505 	}
1506 
1507 	snprintf(str, size, "%u", ntohs(port));
1508 	return addr->ss_family;
1509 }
1510 
1511 /* check if the given address is local to the system or not. It will return
1512  * -1 when it's not possible to know, 0 when the address is not local, 1 when
1513  * it is. We don't want to iterate over all interfaces for this (and it is not
1514  * portable). So instead we try to bind in UDP to this address on a free non
1515  * privileged port and to connect to the same address, port 0 (connect doesn't
1516  * care). If it succeeds, we own the address. Note that non-inet addresses are
1517  * considered local since they're most likely AF_UNIX.
1518  */
addr_is_local(const struct netns_entry * ns,const struct sockaddr_storage * orig)1519 int addr_is_local(const struct netns_entry *ns,
1520                   const struct sockaddr_storage *orig)
1521 {
1522 	struct sockaddr_storage addr;
1523 	int result;
1524 	int fd;
1525 
1526 	if (!is_inet_addr(orig))
1527 		return 1;
1528 
1529 	memcpy(&addr, orig, sizeof(addr));
1530 	set_host_port(&addr, 0);
1531 
1532 	fd = my_socketat(ns, addr.ss_family, SOCK_DGRAM, IPPROTO_UDP);
1533 	if (fd < 0)
1534 		return -1;
1535 
1536 	result = -1;
1537 	if (bind(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == 0) {
1538 		if (connect(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == -1)
1539 			result = 0; // fail, non-local address
1540 		else
1541 			result = 1; // success, local address
1542 	}
1543 	else {
1544 		if (errno == EADDRNOTAVAIL)
1545 			result = 0; // definitely not local :-)
1546 	}
1547 	close(fd);
1548 
1549 	return result;
1550 }
1551 
1552 /* will try to encode the string <string> replacing all characters tagged in
1553  * <map> with the hexadecimal representation of their ASCII-code (2 digits)
1554  * prefixed by <escape>, and will store the result between <start> (included)
1555  * and <stop> (excluded), and will always terminate the string with a '\0'
1556  * before <stop>. The position of the '\0' is returned if the conversion
1557  * completes. If bytes are missing between <start> and <stop>, then the
1558  * conversion will be incomplete and truncated. If <stop> <= <start>, the '\0'
1559  * cannot even be stored so we return <start> without writing the 0.
1560  * The input string must also be zero-terminated.
1561  */
1562 const char hextab[16] = "0123456789ABCDEF";
encode_string(char * start,char * stop,const char escape,const long * map,const char * string)1563 char *encode_string(char *start, char *stop,
1564 		    const char escape, const long *map,
1565 		    const char *string)
1566 {
1567 	if (start < stop) {
1568 		stop--; /* reserve one byte for the final '\0' */
1569 		while (start < stop && *string != '\0') {
1570 			if (!ha_bit_test((unsigned char)(*string), map))
1571 				*start++ = *string;
1572 			else {
1573 				if (start + 3 >= stop)
1574 					break;
1575 				*start++ = escape;
1576 				*start++ = hextab[(*string >> 4) & 15];
1577 				*start++ = hextab[*string & 15];
1578 			}
1579 			string++;
1580 		}
1581 		*start = '\0';
1582 	}
1583 	return start;
1584 }
1585 
1586 /*
1587  * Same behavior as encode_string() above, except that it encodes chunk
1588  * <chunk> instead of a string.
1589  */
encode_chunk(char * start,char * stop,const char escape,const long * map,const struct buffer * chunk)1590 char *encode_chunk(char *start, char *stop,
1591 		    const char escape, const long *map,
1592 		    const struct buffer *chunk)
1593 {
1594 	char *str = chunk->area;
1595 	char *end = chunk->area + chunk->data;
1596 
1597 	if (start < stop) {
1598 		stop--; /* reserve one byte for the final '\0' */
1599 		while (start < stop && str < end) {
1600 			if (!ha_bit_test((unsigned char)(*str), map))
1601 				*start++ = *str;
1602 			else {
1603 				if (start + 3 >= stop)
1604 					break;
1605 				*start++ = escape;
1606 				*start++ = hextab[(*str >> 4) & 15];
1607 				*start++ = hextab[*str & 15];
1608 			}
1609 			str++;
1610 		}
1611 		*start = '\0';
1612 	}
1613 	return start;
1614 }
1615 
1616 /*
1617  * Tries to prefix characters tagged in the <map> with the <escape>
1618  * character. The input <string> must be zero-terminated. The result will
1619  * be stored between <start> (included) and <stop> (excluded). This
1620  * function will always try to terminate the resulting string with a '\0'
1621  * before <stop>, and will return its position if the conversion
1622  * completes.
1623  */
escape_string(char * start,char * stop,const char escape,const long * map,const char * string)1624 char *escape_string(char *start, char *stop,
1625 		    const char escape, const long *map,
1626 		    const char *string)
1627 {
1628 	if (start < stop) {
1629 		stop--; /* reserve one byte for the final '\0' */
1630 		while (start < stop && *string != '\0') {
1631 			if (!ha_bit_test((unsigned char)(*string), map))
1632 				*start++ = *string;
1633 			else {
1634 				if (start + 2 >= stop)
1635 					break;
1636 				*start++ = escape;
1637 				*start++ = *string;
1638 			}
1639 			string++;
1640 		}
1641 		*start = '\0';
1642 	}
1643 	return start;
1644 }
1645 
1646 /*
1647  * Tries to prefix characters tagged in the <map> with the <escape>
1648  * character. <chunk> contains the input to be escaped. The result will be
1649  * stored between <start> (included) and <stop> (excluded). The function
1650  * will always try to terminate the resulting string with a '\0' before
1651  * <stop>, and will return its position if the conversion completes.
1652  */
escape_chunk(char * start,char * stop,const char escape,const long * map,const struct buffer * chunk)1653 char *escape_chunk(char *start, char *stop,
1654 		   const char escape, const long *map,
1655 		   const struct buffer *chunk)
1656 {
1657 	char *str = chunk->area;
1658 	char *end = chunk->area + chunk->data;
1659 
1660 	if (start < stop) {
1661 		stop--; /* reserve one byte for the final '\0' */
1662 		while (start < stop && str < end) {
1663 			if (!ha_bit_test((unsigned char)(*str), map))
1664 				*start++ = *str;
1665 			else {
1666 				if (start + 2 >= stop)
1667 					break;
1668 				*start++ = escape;
1669 				*start++ = *str;
1670 			}
1671 			str++;
1672 		}
1673 		*start = '\0';
1674 	}
1675 	return start;
1676 }
1677 
1678 /* Check a string for using it in a CSV output format. If the string contains
1679  * one of the following four char <">, <,>, CR or LF, the string is
1680  * encapsulated between <"> and the <"> are escaped by a <""> sequence.
1681  * <str> is the input string to be escaped. The function assumes that
1682  * the input string is null-terminated.
1683  *
1684  * If <quote> is 0, the result is returned escaped but without double quote.
1685  * It is useful if the escaped string is used between double quotes in the
1686  * format.
1687  *
1688  *    printf("..., \"%s\", ...\r\n", csv_enc(str, 0, &trash));
1689  *
1690  * If <quote> is 1, the converter puts the quotes only if any reserved character
1691  * is present. If <quote> is 2, the converter always puts the quotes.
1692  *
1693  * <output> is a struct buffer used for storing the output string.
1694  *
1695  * The function returns the converted string on its output. If an error
1696  * occurs, the function returns an empty string. This type of output is useful
1697  * for using the function directly as printf() argument.
1698  *
1699  * If the output buffer is too short to contain the input string, the result
1700  * is truncated.
1701  *
1702  * This function appends the encoding to the existing output chunk, and it
1703  * guarantees that it starts immediately at the first available character of
1704  * the chunk. Please use csv_enc() instead if you want to replace the output
1705  * chunk.
1706  */
csv_enc_append(const char * str,int quote,struct buffer * output)1707 const char *csv_enc_append(const char *str, int quote, struct buffer *output)
1708 {
1709 	char *end = output->area + output->size;
1710 	char *out = output->area + output->data;
1711 	char *ptr = out;
1712 
1713 	if (quote == 1) {
1714 		/* automatic quoting: first verify if we'll have to quote the string */
1715 		if (!strpbrk(str, "\n\r,\""))
1716 			quote = 0;
1717 	}
1718 
1719 	if (quote)
1720 		*ptr++ = '"';
1721 
1722 	while (*str && ptr < end - 2) { /* -2 for reserving space for <"> and \0. */
1723 		*ptr = *str;
1724 		if (*str == '"') {
1725 			ptr++;
1726 			if (ptr >= end - 2) {
1727 				ptr--;
1728 				break;
1729 			}
1730 			*ptr = '"';
1731 		}
1732 		ptr++;
1733 		str++;
1734 	}
1735 
1736 	if (quote)
1737 		*ptr++ = '"';
1738 
1739 	*ptr = '\0';
1740 	output->data = ptr - output->area;
1741 	return out;
1742 }
1743 
1744 /* Decode an URL-encoded string in-place. The resulting string might
1745  * be shorter. If some forbidden characters are found, the conversion is
1746  * aborted, the string is truncated before the issue and a negative value is
1747  * returned, otherwise the operation returns the length of the decoded string.
1748  * If the 'in_form' argument is non-nul the string is assumed to be part of
1749  * an "application/x-www-form-urlencoded" encoded string, and the '+' will be
1750  * turned to a space. If it's zero, this will only be done after a question
1751  * mark ('?').
1752  */
url_decode(char * string,int in_form)1753 int url_decode(char *string, int in_form)
1754 {
1755 	char *in, *out;
1756 	int ret = -1;
1757 
1758 	in = string;
1759 	out = string;
1760 	while (*in) {
1761 		switch (*in) {
1762 		case '+' :
1763 			*out++ = in_form ? ' ' : *in;
1764 			break;
1765 		case '%' :
1766 			if (!ishex(in[1]) || !ishex(in[2]))
1767 				goto end;
1768 			*out++ = (hex2i(in[1]) << 4) + hex2i(in[2]);
1769 			in += 2;
1770 			break;
1771 		case '?':
1772 			in_form = 1;
1773 			/* fall through */
1774 		default:
1775 			*out++ = *in;
1776 			break;
1777 		}
1778 		in++;
1779 	}
1780 	ret = out - string; /* success */
1781  end:
1782 	*out = 0;
1783 	return ret;
1784 }
1785 
str2ui(const char * s)1786 unsigned int str2ui(const char *s)
1787 {
1788 	return __str2ui(s);
1789 }
1790 
str2uic(const char * s)1791 unsigned int str2uic(const char *s)
1792 {
1793 	return __str2uic(s);
1794 }
1795 
strl2ui(const char * s,int len)1796 unsigned int strl2ui(const char *s, int len)
1797 {
1798 	return __strl2ui(s, len);
1799 }
1800 
strl2uic(const char * s,int len)1801 unsigned int strl2uic(const char *s, int len)
1802 {
1803 	return __strl2uic(s, len);
1804 }
1805 
read_uint(const char ** s,const char * end)1806 unsigned int read_uint(const char **s, const char *end)
1807 {
1808 	return __read_uint(s, end);
1809 }
1810 
1811 /* This function reads an unsigned integer from the string pointed to by <s> and
1812  * returns it. The <s> pointer is adjusted to point to the first unread char. The
1813  * function automatically stops at <end>. If the number overflows, the 2^64-1
1814  * value is returned.
1815  */
read_uint64(const char ** s,const char * end)1816 unsigned long long int read_uint64(const char **s, const char *end)
1817 {
1818 	const char *ptr = *s;
1819 	unsigned long long int i = 0, tmp;
1820 	unsigned int j;
1821 
1822 	while (ptr < end) {
1823 
1824 		/* read next char */
1825 		j = *ptr - '0';
1826 		if (j > 9)
1827 			goto read_uint64_end;
1828 
1829 		/* add char to the number and check overflow. */
1830 		tmp = i * 10;
1831 		if (tmp / 10 != i) {
1832 			i = ULLONG_MAX;
1833 			goto read_uint64_eat;
1834 		}
1835 		if (ULLONG_MAX - tmp < j) {
1836 			i = ULLONG_MAX;
1837 			goto read_uint64_eat;
1838 		}
1839 		i = tmp + j;
1840 		ptr++;
1841 	}
1842 read_uint64_eat:
1843 	/* eat each numeric char */
1844 	while (ptr < end) {
1845 		if ((unsigned int)(*ptr - '0') > 9)
1846 			break;
1847 		ptr++;
1848 	}
1849 read_uint64_end:
1850 	*s = ptr;
1851 	return i;
1852 }
1853 
1854 /* This function reads an integer from the string pointed to by <s> and returns
1855  * it. The <s> pointer is adjusted to point to the first unread char. The function
1856  * automatically stops at <end>. Il the number is bigger than 2^63-2, the 2^63-1
1857  * value is returned. If the number is lowest than -2^63-1, the -2^63 value is
1858  * returned.
1859  */
read_int64(const char ** s,const char * end)1860 long long int read_int64(const char **s, const char *end)
1861 {
1862 	unsigned long long int i = 0;
1863 	int neg = 0;
1864 
1865 	/* Look for minus char. */
1866 	if (**s == '-') {
1867 		neg = 1;
1868 		(*s)++;
1869 	}
1870 	else if (**s == '+')
1871 		(*s)++;
1872 
1873 	/* convert as positive number. */
1874 	i = read_uint64(s, end);
1875 
1876 	if (neg) {
1877 		if (i > 0x8000000000000000ULL)
1878 			return LLONG_MIN;
1879 		return -i;
1880 	}
1881 	if (i > 0x7fffffffffffffffULL)
1882 		return LLONG_MAX;
1883 	return i;
1884 }
1885 
1886 /* This one is 7 times faster than strtol() on athlon with checks.
1887  * It returns the value of the number composed of all valid digits read,
1888  * and can process negative numbers too.
1889  */
strl2ic(const char * s,int len)1890 int strl2ic(const char *s, int len)
1891 {
1892 	int i = 0;
1893 	int j, k;
1894 
1895 	if (len > 0) {
1896 		if (*s != '-') {
1897 			/* positive number */
1898 			while (len-- > 0) {
1899 				j = (*s++) - '0';
1900 				k = i * 10;
1901 				if (j > 9)
1902 					break;
1903 				i = k + j;
1904 			}
1905 		} else {
1906 			/* negative number */
1907 			s++;
1908 			while (--len > 0) {
1909 				j = (*s++) - '0';
1910 				k = i * 10;
1911 				if (j > 9)
1912 					break;
1913 				i = k - j;
1914 			}
1915 		}
1916 	}
1917 	return i;
1918 }
1919 
1920 
1921 /* This function reads exactly <len> chars from <s> and converts them to a
1922  * signed integer which it stores into <ret>. It accurately detects any error
1923  * (truncated string, invalid chars, overflows). It is meant to be used in
1924  * applications designed for hostile environments. It returns zero when the
1925  * number has successfully been converted, non-zero otherwise. When an error
1926  * is returned, the <ret> value is left untouched. It is yet 5 to 40 times
1927  * faster than strtol().
1928  */
strl2irc(const char * s,int len,int * ret)1929 int strl2irc(const char *s, int len, int *ret)
1930 {
1931 	int i = 0;
1932 	int j;
1933 
1934 	if (!len)
1935 		return 1;
1936 
1937 	if (*s != '-') {
1938 		/* positive number */
1939 		while (len-- > 0) {
1940 			j = (*s++) - '0';
1941 			if (j > 9)            return 1; /* invalid char */
1942 			if (i > INT_MAX / 10) return 1; /* check for multiply overflow */
1943 			i = i * 10;
1944 			if (i + j < i)        return 1; /* check for addition overflow */
1945 			i = i + j;
1946 		}
1947 	} else {
1948 		/* negative number */
1949 		s++;
1950 		while (--len > 0) {
1951 			j = (*s++) - '0';
1952 			if (j > 9)             return 1; /* invalid char */
1953 			if (i < INT_MIN / 10)  return 1; /* check for multiply overflow */
1954 			i = i * 10;
1955 			if (i - j > i)         return 1; /* check for subtract overflow */
1956 			i = i - j;
1957 		}
1958 	}
1959 	*ret = i;
1960 	return 0;
1961 }
1962 
1963 
1964 /* This function reads exactly <len> chars from <s> and converts them to a
1965  * signed integer which it stores into <ret>. It accurately detects any error
1966  * (truncated string, invalid chars, overflows). It is meant to be used in
1967  * applications designed for hostile environments. It returns zero when the
1968  * number has successfully been converted, non-zero otherwise. When an error
1969  * is returned, the <ret> value is left untouched. It is about 3 times slower
1970  * than str2irc().
1971  */
1972 
strl2llrc(const char * s,int len,long long * ret)1973 int strl2llrc(const char *s, int len, long long *ret)
1974 {
1975 	long long i = 0;
1976 	int j;
1977 
1978 	if (!len)
1979 		return 1;
1980 
1981 	if (*s != '-') {
1982 		/* positive number */
1983 		while (len-- > 0) {
1984 			j = (*s++) - '0';
1985 			if (j > 9)              return 1; /* invalid char */
1986 			if (i > LLONG_MAX / 10LL) return 1; /* check for multiply overflow */
1987 			i = i * 10LL;
1988 			if (i + j < i)          return 1; /* check for addition overflow */
1989 			i = i + j;
1990 		}
1991 	} else {
1992 		/* negative number */
1993 		s++;
1994 		while (--len > 0) {
1995 			j = (*s++) - '0';
1996 			if (j > 9)              return 1; /* invalid char */
1997 			if (i < LLONG_MIN / 10LL) return 1; /* check for multiply overflow */
1998 			i = i * 10LL;
1999 			if (i - j > i)          return 1; /* check for subtract overflow */
2000 			i = i - j;
2001 		}
2002 	}
2003 	*ret = i;
2004 	return 0;
2005 }
2006 
2007 /* This function is used with pat_parse_dotted_ver(). It converts a string
2008  * composed by two number separated by a dot. Each part must contain in 16 bits
2009  * because internally they will be represented as a 32-bit quantity stored in
2010  * a 64-bit integer. It returns zero when the number has successfully been
2011  * converted, non-zero otherwise. When an error is returned, the <ret> value
2012  * is left untouched.
2013  *
2014  *    "1.3"         -> 0x0000000000010003
2015  *    "65535.65535" -> 0x00000000ffffffff
2016  */
strl2llrc_dotted(const char * text,int len,long long * ret)2017 int strl2llrc_dotted(const char *text, int len, long long *ret)
2018 {
2019 	const char *end = &text[len];
2020 	const char *p;
2021 	long long major, minor;
2022 
2023 	/* Look for dot. */
2024 	for (p = text; p < end; p++)
2025 		if (*p == '.')
2026 			break;
2027 
2028 	/* Convert major. */
2029 	if (strl2llrc(text, p - text, &major) != 0)
2030 		return 1;
2031 
2032 	/* Check major. */
2033 	if (major >= 65536)
2034 		return 1;
2035 
2036 	/* Convert minor. */
2037 	minor = 0;
2038 	if (p < end)
2039 		if (strl2llrc(p + 1, end - (p + 1), &minor) != 0)
2040 			return 1;
2041 
2042 	/* Check minor. */
2043 	if (minor >= 65536)
2044 		return 1;
2045 
2046 	/* Compose value. */
2047 	*ret = (major << 16) | (minor & 0xffff);
2048 	return 0;
2049 }
2050 
2051 /* This function parses a time value optionally followed by a unit suffix among
2052  * "d", "h", "m", "s", "ms" or "us". It converts the value into the unit
2053  * expected by the caller. The computation does its best to avoid overflows.
2054  * The value is returned in <ret> if everything is fine, and a NULL is returned
2055  * by the function. In case of error, a pointer to the error is returned and
2056  * <ret> is left untouched. Values are automatically rounded up when needed.
2057  * Values resulting in values larger than or equal to 2^31 after conversion are
2058  * reported as an overflow as value PARSE_TIME_OVER. Non-null values resulting
2059  * in an underflow are reported as an underflow as value PARSE_TIME_UNDER.
2060  */
parse_time_err(const char * text,unsigned * ret,unsigned unit_flags)2061 const char *parse_time_err(const char *text, unsigned *ret, unsigned unit_flags)
2062 {
2063 	unsigned long long imult, idiv;
2064 	unsigned long long omult, odiv;
2065 	unsigned long long value, result;
2066 	const char *str = text;
2067 
2068 	if (!isdigit((unsigned char)*text))
2069 		return text;
2070 
2071 	omult = odiv = 1;
2072 
2073 	switch (unit_flags & TIME_UNIT_MASK) {
2074 	case TIME_UNIT_US:   omult = 1000000; break;
2075 	case TIME_UNIT_MS:   omult = 1000; break;
2076 	case TIME_UNIT_S:    break;
2077 	case TIME_UNIT_MIN:  odiv = 60; break;
2078 	case TIME_UNIT_HOUR: odiv = 3600; break;
2079 	case TIME_UNIT_DAY:  odiv = 86400; break;
2080 	default: break;
2081 	}
2082 
2083 	value = 0;
2084 
2085 	while (1) {
2086 		unsigned int j;
2087 
2088 		j = *text - '0';
2089 		if (j > 9)
2090 			break;
2091 		text++;
2092 		value *= 10;
2093 		value += j;
2094 	}
2095 
2096 	imult = idiv = 1;
2097 	switch (*text) {
2098 	case '\0': /* no unit = default unit */
2099 		imult = omult = idiv = odiv = 1;
2100 		goto end;
2101 	case 's': /* second = unscaled unit */
2102 		break;
2103 	case 'u': /* microsecond : "us" */
2104 		if (text[1] == 's') {
2105 			idiv = 1000000;
2106 			text++;
2107 		}
2108 		return text;
2109 	case 'm': /* millisecond : "ms" or minute: "m" */
2110 		if (text[1] == 's') {
2111 			idiv = 1000;
2112 			text++;
2113 		} else
2114 			imult = 60;
2115 		break;
2116 	case 'h': /* hour : "h" */
2117 		imult = 3600;
2118 		break;
2119 	case 'd': /* day : "d" */
2120 		imult = 86400;
2121 		break;
2122 	default:
2123 		return text;
2124 		break;
2125 	}
2126 	if (*(++text) != '\0') {
2127 		ha_warning("unexpected character '%c' after the timer value '%s', only "
2128 			   "(us=microseconds,ms=milliseconds,s=seconds,m=minutes,h=hours,d=days) are supported."
2129 			   " This will be reported as an error in next versions.\n", *text, str);
2130 	}
2131 
2132   end:
2133 	if (omult % idiv == 0) { omult /= idiv; idiv = 1; }
2134 	if (idiv % omult == 0) { idiv /= omult; omult = 1; }
2135 	if (imult % odiv == 0) { imult /= odiv; odiv = 1; }
2136 	if (odiv % imult == 0) { odiv /= imult; imult = 1; }
2137 
2138 	result = (value * (imult * omult) + (idiv * odiv - 1)) / (idiv * odiv);
2139 	if (result >= 0x80000000)
2140 		return PARSE_TIME_OVER;
2141 	if (!result && value)
2142 		return PARSE_TIME_UNDER;
2143 	*ret = result;
2144 	return NULL;
2145 }
2146 
2147 /* this function converts the string starting at <text> to an unsigned int
2148  * stored in <ret>. If an error is detected, the pointer to the unexpected
2149  * character is returned. If the conversion is successful, NULL is returned.
2150  */
parse_size_err(const char * text,unsigned * ret)2151 const char *parse_size_err(const char *text, unsigned *ret) {
2152 	unsigned value = 0;
2153 
2154 	if (!isdigit((unsigned char)*text))
2155 		return text;
2156 
2157 	while (1) {
2158 		unsigned int j;
2159 
2160 		j = *text - '0';
2161 		if (j > 9)
2162 			break;
2163 		if (value > ~0U / 10)
2164 			return text;
2165 		value *= 10;
2166 		if (value > (value + j))
2167 			return text;
2168 		value += j;
2169 		text++;
2170 	}
2171 
2172 	switch (*text) {
2173 	case '\0':
2174 		break;
2175 	case 'K':
2176 	case 'k':
2177 		if (value > ~0U >> 10)
2178 			return text;
2179 		value = value << 10;
2180 		break;
2181 	case 'M':
2182 	case 'm':
2183 		if (value > ~0U >> 20)
2184 			return text;
2185 		value = value << 20;
2186 		break;
2187 	case 'G':
2188 	case 'g':
2189 		if (value > ~0U >> 30)
2190 			return text;
2191 		value = value << 30;
2192 		break;
2193 	default:
2194 		return text;
2195 	}
2196 
2197 	if (*text != '\0' && *++text != '\0')
2198 		return text;
2199 
2200 	*ret = value;
2201 	return NULL;
2202 }
2203 
2204 /*
2205  * Parse binary string written in hexadecimal (source) and store the decoded
2206  * result into binstr and set binstrlen to the lengh of binstr. Memory for
2207  * binstr is allocated by the function. In case of error, returns 0 with an
2208  * error message in err. In succes case, it returns the consumed length.
2209  */
parse_binary(const char * source,char ** binstr,int * binstrlen,char ** err)2210 int parse_binary(const char *source, char **binstr, int *binstrlen, char **err)
2211 {
2212 	int len;
2213 	const char *p = source;
2214 	int i,j;
2215 	int alloc;
2216 
2217 	len = strlen(source);
2218 	if (len % 2) {
2219 		memprintf(err, "an even number of hex digit is expected");
2220 		return 0;
2221 	}
2222 
2223 	len = len >> 1;
2224 
2225 	if (!*binstr) {
2226 		*binstr = calloc(len, sizeof(char));
2227 		if (!*binstr) {
2228 			memprintf(err, "out of memory while loading string pattern");
2229 			return 0;
2230 		}
2231 		alloc = 1;
2232 	}
2233 	else {
2234 		if (*binstrlen < len) {
2235 			memprintf(err, "no space available in the buffer. expect %d, provides %d",
2236 			          len, *binstrlen);
2237 			return 0;
2238 		}
2239 		alloc = 0;
2240 	}
2241 	*binstrlen = len;
2242 
2243 	i = j = 0;
2244 	while (j < len) {
2245 		if (!ishex(p[i++]))
2246 			goto bad_input;
2247 		if (!ishex(p[i++]))
2248 			goto bad_input;
2249 		(*binstr)[j++] =  (hex2i(p[i-2]) << 4) + hex2i(p[i-1]);
2250 	}
2251 	return len << 1;
2252 
2253 bad_input:
2254 	memprintf(err, "an hex digit is expected (found '%c')", p[i-1]);
2255 	if (alloc) {
2256 		free(*binstr);
2257 		*binstr = NULL;
2258 	}
2259 	return 0;
2260 }
2261 
2262 /* copies at most <n> characters from <src> and always terminates with '\0' */
my_strndup(const char * src,int n)2263 char *my_strndup(const char *src, int n)
2264 {
2265 	int len = 0;
2266 	char *ret;
2267 
2268 	while (len < n && src[len])
2269 		len++;
2270 
2271 	ret = malloc(len + 1);
2272 	if (!ret)
2273 		return ret;
2274 	memcpy(ret, src, len);
2275 	ret[len] = '\0';
2276 	return ret;
2277 }
2278 
2279 /*
2280  * search needle in haystack
2281  * returns the pointer if found, returns NULL otherwise
2282  */
my_memmem(const void * haystack,size_t haystacklen,const void * needle,size_t needlelen)2283 const void *my_memmem(const void *haystack, size_t haystacklen, const void *needle, size_t needlelen)
2284 {
2285 	const void *c = NULL;
2286 	unsigned char f;
2287 
2288 	if ((haystack == NULL) || (needle == NULL) || (haystacklen < needlelen))
2289 		return NULL;
2290 
2291 	f = *(char *)needle;
2292 	c = haystack;
2293 	while ((c = memchr(c, f, haystacklen - (c - haystack))) != NULL) {
2294 		if ((haystacklen - (c - haystack)) < needlelen)
2295 			return NULL;
2296 
2297 		if (memcmp(c, needle, needlelen) == 0)
2298 			return c;
2299 		++c;
2300 	}
2301 	return NULL;
2302 }
2303 
2304 /* This function returns the first unused key greater than or equal to <key> in
2305  * ID tree <root>. Zero is returned if no place is found.
2306  */
get_next_id(struct eb_root * root,unsigned int key)2307 unsigned int get_next_id(struct eb_root *root, unsigned int key)
2308 {
2309 	struct eb32_node *used;
2310 
2311 	do {
2312 		used = eb32_lookup_ge(root, key);
2313 		if (!used || used->key > key)
2314 			return key; /* key is available */
2315 		key++;
2316 	} while (key);
2317 	return key;
2318 }
2319 
2320 /* dump the full tree to <file> in DOT format for debugging purposes. Will
2321  * optionally highlight node <subj> if found, depending on operation <op> :
2322  *    0 : nothing
2323  *   >0 : insertion, node/leaf are surrounded in red
2324  *   <0 : removal, node/leaf are dashed with no background
2325  * Will optionally add "desc" as a label on the graph if set and non-null.
2326  */
eb32sc_to_file(FILE * file,struct eb_root * root,const struct eb32sc_node * subj,int op,const char * desc)2327 void eb32sc_to_file(FILE *file, struct eb_root *root, const struct eb32sc_node *subj, int op, const char *desc)
2328 {
2329 	struct eb32sc_node *node;
2330 	unsigned long scope = -1;
2331 
2332 	fprintf(file, "digraph ebtree {\n");
2333 
2334 	if (desc && *desc) {
2335 		fprintf(file,
2336 			"  fontname=\"fixed\";\n"
2337 			"  fontsize=8;\n"
2338 			"  label=\"%s\";\n", desc);
2339 	}
2340 
2341 	fprintf(file,
2342 		"  node [fontname=\"fixed\" fontsize=8 shape=\"box\" style=\"filled\" color=\"black\" fillcolor=\"white\"];\n"
2343 		"  edge [fontname=\"fixed\" fontsize=8 style=\"solid\" color=\"magenta\" dir=\"forward\"];\n"
2344 		"  \"%lx_n\" [label=\"root\\n%lx\"]\n", (long)eb_root_to_node(root), (long)root
2345 		);
2346 
2347 	fprintf(file, "  \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
2348 		(long)eb_root_to_node(root),
2349 		(long)eb_root_to_node(eb_clrtag(root->b[0])),
2350 		eb_gettag(root->b[0]) == EB_LEAF ? 'l' : 'n');
2351 
2352 	node = eb32sc_first(root, scope);
2353 	while (node) {
2354 		if (node->node.node_p) {
2355 			/* node part is used */
2356 			fprintf(file, "  \"%lx_n\" [label=\"%lx\\nkey=%u\\nscope=%lx\\nbit=%d\" fillcolor=\"lightskyblue1\" %s];\n",
2357 				(long)node, (long)node, node->key, node->node_s, node->node.bit,
2358 				(node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
2359 
2360 			fprintf(file, "  \"%lx_n\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
2361 				(long)node,
2362 				(long)eb_root_to_node(eb_clrtag(node->node.node_p)),
2363 				eb_gettag(node->node.node_p) ? 'R' : 'L');
2364 
2365 			fprintf(file, "  \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
2366 				(long)node,
2367 				(long)eb_root_to_node(eb_clrtag(node->node.branches.b[0])),
2368 				eb_gettag(node->node.branches.b[0]) == EB_LEAF ? 'l' : 'n');
2369 
2370 			fprintf(file, "  \"%lx_n\" -> \"%lx_%c\" [taillabel=\"R\"];\n",
2371 				(long)node,
2372 				(long)eb_root_to_node(eb_clrtag(node->node.branches.b[1])),
2373 				eb_gettag(node->node.branches.b[1]) == EB_LEAF ? 'l' : 'n');
2374 		}
2375 
2376 		fprintf(file, "  \"%lx_l\" [label=\"%lx\\nkey=%u\\nscope=%lx\\npfx=%u\" fillcolor=\"yellow\" %s];\n",
2377 			(long)node, (long)node, node->key, node->leaf_s, node->node.pfx,
2378 			(node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
2379 
2380 		fprintf(file, "  \"%lx_l\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
2381 			(long)node,
2382 			(long)eb_root_to_node(eb_clrtag(node->node.leaf_p)),
2383 			eb_gettag(node->node.leaf_p) ? 'R' : 'L');
2384 		node = eb32sc_next(node, scope);
2385 	}
2386 	fprintf(file, "}\n");
2387 }
2388 
2389 /* This function compares a sample word possibly followed by blanks to another
2390  * clean word. The compare is case-insensitive. 1 is returned if both are equal,
2391  * otherwise zero. This intends to be used when checking HTTP headers for some
2392  * values. Note that it validates a word followed only by blanks but does not
2393  * validate a word followed by blanks then other chars.
2394  */
word_match(const char * sample,int slen,const char * word,int wlen)2395 int word_match(const char *sample, int slen, const char *word, int wlen)
2396 {
2397 	if (slen < wlen)
2398 		return 0;
2399 
2400 	while (wlen) {
2401 		char c = *sample ^ *word;
2402 		if (c && c != ('A' ^ 'a'))
2403 			return 0;
2404 		sample++;
2405 		word++;
2406 		slen--;
2407 		wlen--;
2408 	}
2409 
2410 	while (slen) {
2411 		if (*sample != ' ' && *sample != '\t')
2412 			return 0;
2413 		sample++;
2414 		slen--;
2415 	}
2416 	return 1;
2417 }
2418 
2419 /* Converts any text-formatted IPv4 address to a host-order IPv4 address. It
2420  * is particularly fast because it avoids expensive operations such as
2421  * multiplies, which are optimized away at the end. It requires a properly
2422  * formated address though (3 points).
2423  */
inetaddr_host(const char * text)2424 unsigned int inetaddr_host(const char *text)
2425 {
2426 	const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2427 	register unsigned int dig100, dig10, dig1;
2428 	int s;
2429 	const char *p, *d;
2430 
2431 	dig1 = dig10 = dig100 = ascii_zero;
2432 	s = 24;
2433 
2434 	p = text;
2435 	while (1) {
2436 		if (((unsigned)(*p - '0')) <= 9) {
2437 			p++;
2438 			continue;
2439 		}
2440 
2441 		/* here, we have a complete byte between <text> and <p> (exclusive) */
2442 		if (p == text)
2443 			goto end;
2444 
2445 		d = p - 1;
2446 		dig1   |= (unsigned int)(*d << s);
2447 		if (d == text)
2448 			goto end;
2449 
2450 		d--;
2451 		dig10  |= (unsigned int)(*d << s);
2452 		if (d == text)
2453 			goto end;
2454 
2455 		d--;
2456 		dig100 |= (unsigned int)(*d << s);
2457 	end:
2458 		if (!s || *p != '.')
2459 			break;
2460 
2461 		s -= 8;
2462 		text = ++p;
2463 	}
2464 
2465 	dig100 -= ascii_zero;
2466 	dig10  -= ascii_zero;
2467 	dig1   -= ascii_zero;
2468 	return ((dig100 * 10) + dig10) * 10 + dig1;
2469 }
2470 
2471 /*
2472  * Idem except the first unparsed character has to be passed in <stop>.
2473  */
inetaddr_host_lim(const char * text,const char * stop)2474 unsigned int inetaddr_host_lim(const char *text, const char *stop)
2475 {
2476 	const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2477 	register unsigned int dig100, dig10, dig1;
2478 	int s;
2479 	const char *p, *d;
2480 
2481 	dig1 = dig10 = dig100 = ascii_zero;
2482 	s = 24;
2483 
2484 	p = text;
2485 	while (1) {
2486 		if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2487 			p++;
2488 			continue;
2489 		}
2490 
2491 		/* here, we have a complete byte between <text> and <p> (exclusive) */
2492 		if (p == text)
2493 			goto end;
2494 
2495 		d = p - 1;
2496 		dig1   |= (unsigned int)(*d << s);
2497 		if (d == text)
2498 			goto end;
2499 
2500 		d--;
2501 		dig10  |= (unsigned int)(*d << s);
2502 		if (d == text)
2503 			goto end;
2504 
2505 		d--;
2506 		dig100 |= (unsigned int)(*d << s);
2507 	end:
2508 		if (!s || p == stop || *p != '.')
2509 			break;
2510 
2511 		s -= 8;
2512 		text = ++p;
2513 	}
2514 
2515 	dig100 -= ascii_zero;
2516 	dig10  -= ascii_zero;
2517 	dig1   -= ascii_zero;
2518 	return ((dig100 * 10) + dig10) * 10 + dig1;
2519 }
2520 
2521 /*
2522  * Idem except the pointer to first unparsed byte is returned into <ret> which
2523  * must not be NULL.
2524  */
inetaddr_host_lim_ret(char * text,char * stop,char ** ret)2525 unsigned int inetaddr_host_lim_ret(char *text, char *stop, char **ret)
2526 {
2527 	const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2528 	register unsigned int dig100, dig10, dig1;
2529 	int s;
2530 	char *p, *d;
2531 
2532 	dig1 = dig10 = dig100 = ascii_zero;
2533 	s = 24;
2534 
2535 	p = text;
2536 	while (1) {
2537 		if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2538 			p++;
2539 			continue;
2540 		}
2541 
2542 		/* here, we have a complete byte between <text> and <p> (exclusive) */
2543 		if (p == text)
2544 			goto end;
2545 
2546 		d = p - 1;
2547 		dig1   |= (unsigned int)(*d << s);
2548 		if (d == text)
2549 			goto end;
2550 
2551 		d--;
2552 		dig10  |= (unsigned int)(*d << s);
2553 		if (d == text)
2554 			goto end;
2555 
2556 		d--;
2557 		dig100 |= (unsigned int)(*d << s);
2558 	end:
2559 		if (!s || p == stop || *p != '.')
2560 			break;
2561 
2562 		s -= 8;
2563 		text = ++p;
2564 	}
2565 
2566 	*ret = p;
2567 	dig100 -= ascii_zero;
2568 	dig10  -= ascii_zero;
2569 	dig1   -= ascii_zero;
2570 	return ((dig100 * 10) + dig10) * 10 + dig1;
2571 }
2572 
2573 /* Convert a fixed-length string to an IP address. Returns 0 in case of error,
2574  * or the number of chars read in case of success. Maybe this could be replaced
2575  * by one of the functions above. Also, apparently this function does not support
2576  * hosts above 255 and requires exactly 4 octets.
2577  * The destination is only modified on success.
2578  */
buf2ip(const char * buf,size_t len,struct in_addr * dst)2579 int buf2ip(const char *buf, size_t len, struct in_addr *dst)
2580 {
2581 	const char *addr;
2582 	int saw_digit, octets, ch;
2583 	u_char tmp[4], *tp;
2584 	const char *cp = buf;
2585 
2586 	saw_digit = 0;
2587 	octets = 0;
2588 	*(tp = tmp) = 0;
2589 
2590 	for (addr = buf; addr - buf < len; addr++) {
2591 		unsigned char digit = (ch = *addr) - '0';
2592 
2593 		if (digit > 9 && ch != '.')
2594 			break;
2595 
2596 		if (digit <= 9) {
2597 			u_int new = *tp * 10 + digit;
2598 
2599 			if (new > 255)
2600 				return 0;
2601 
2602 			*tp = new;
2603 
2604 			if (!saw_digit) {
2605 				if (++octets > 4)
2606 					return 0;
2607 				saw_digit = 1;
2608 			}
2609 		} else if (ch == '.' && saw_digit) {
2610 			if (octets == 4)
2611 				return 0;
2612 
2613 			*++tp = 0;
2614 			saw_digit = 0;
2615 		} else
2616 			return 0;
2617 	}
2618 
2619 	if (octets < 4)
2620 		return 0;
2621 
2622 	memcpy(&dst->s_addr, tmp, 4);
2623 	return addr - cp;
2624 }
2625 
2626 /* This function converts the string in <buf> of the len <len> to
2627  * struct in6_addr <dst> which must be allocated by the caller.
2628  * This function returns 1 in success case, otherwise zero.
2629  * The destination is only modified on success.
2630  */
buf2ip6(const char * buf,size_t len,struct in6_addr * dst)2631 int buf2ip6(const char *buf, size_t len, struct in6_addr *dst)
2632 {
2633 	char null_term_ip6[INET6_ADDRSTRLEN + 1];
2634 	struct in6_addr out;
2635 
2636 	if (len > INET6_ADDRSTRLEN)
2637 		return 0;
2638 
2639 	memcpy(null_term_ip6, buf, len);
2640 	null_term_ip6[len] = '\0';
2641 
2642 	if (!inet_pton(AF_INET6, null_term_ip6, &out))
2643 		return 0;
2644 
2645 	*dst = out;
2646 	return 1;
2647 }
2648 
2649 /* To be used to quote config arg positions. Returns the short string at <ptr>
2650  * surrounded by simple quotes if <ptr> is valid and non-empty, or "end of line"
2651  * if ptr is NULL or empty. The string is locally allocated.
2652  */
quote_arg(const char * ptr)2653 const char *quote_arg(const char *ptr)
2654 {
2655 	static THREAD_LOCAL char val[32];
2656 	int i;
2657 
2658 	if (!ptr || !*ptr)
2659 		return "end of line";
2660 	val[0] = '\'';
2661 	for (i = 1; i < sizeof(val) - 2 && *ptr; i++)
2662 		val[i] = *ptr++;
2663 	val[i++] = '\'';
2664 	val[i] = '\0';
2665 	return val;
2666 }
2667 
2668 /* returns an operator among STD_OP_* for string <str> or < 0 if unknown */
get_std_op(const char * str)2669 int get_std_op(const char *str)
2670 {
2671 	int ret = -1;
2672 
2673 	if (*str == 'e' && str[1] == 'q')
2674 		ret = STD_OP_EQ;
2675 	else if (*str == 'n' && str[1] == 'e')
2676 		ret = STD_OP_NE;
2677 	else if (*str == 'l') {
2678 		if (str[1] == 'e') ret = STD_OP_LE;
2679 		else if (str[1] == 't') ret = STD_OP_LT;
2680 	}
2681 	else if (*str == 'g') {
2682 		if (str[1] == 'e') ret = STD_OP_GE;
2683 		else if (str[1] == 't') ret = STD_OP_GT;
2684 	}
2685 
2686 	if (ret == -1 || str[2] != '\0')
2687 		return -1;
2688 	return ret;
2689 }
2690 
2691 /* hash a 32-bit integer to another 32-bit integer */
full_hash(unsigned int a)2692 unsigned int full_hash(unsigned int a)
2693 {
2694 	return __full_hash(a);
2695 }
2696 
2697 /* Return the bit position in mask <m> of the nth bit set of rank <r>, between
2698  * 0 and LONGBITS-1 included, starting from the left. For example ranks 0,1,2,3
2699  * for mask 0x55 will be 6, 4, 2 and 0 respectively. This algorithm is based on
2700  * a popcount variant and is described here :
2701  *   https://graphics.stanford.edu/~seander/bithacks.html
2702  */
mask_find_rank_bit(unsigned int r,unsigned long m)2703 unsigned int mask_find_rank_bit(unsigned int r, unsigned long m)
2704 {
2705 	unsigned long a, b, c, d;
2706 	unsigned int s;
2707 	unsigned int t;
2708 
2709 	a =  m - ((m >> 1) & ~0UL/3);
2710 	b = (a & ~0UL/5) + ((a >> 2) & ~0UL/5);
2711 	c = (b + (b >> 4)) & ~0UL/0x11;
2712 	d = (c + (c >> 8)) & ~0UL/0x101;
2713 
2714 	r++; // make r be 1..64
2715 
2716 	t = 0;
2717 	s = LONGBITS;
2718 	if (s > 32) {
2719 		unsigned long d2 = (d >> 16) >> 16;
2720 		t = d2 + (d2 >> 16);
2721 		s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
2722 	}
2723 
2724 	t  = (d >> (s - 16)) & 0xff;
2725 	s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
2726 	t  = (c >> (s - 8)) & 0xf;
2727 	s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
2728 	t  = (b >> (s - 4)) & 0x7;
2729 	s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
2730 	t  = (a >> (s - 2)) & 0x3;
2731 	s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
2732 	t  = (m >> (s - 1)) & 0x1;
2733 	s -= ((t - r) & 256) >> 8;
2734 
2735        return s - 1;
2736 }
2737 
2738 /* Same as mask_find_rank_bit() above but makes use of pre-computed bitmaps
2739  * based on <m>, in <a..d>. These ones must be updated whenever <m> changes
2740  * using mask_prep_rank_map() below.
2741  */
mask_find_rank_bit_fast(unsigned int r,unsigned long m,unsigned long a,unsigned long b,unsigned long c,unsigned long d)2742 unsigned int mask_find_rank_bit_fast(unsigned int r, unsigned long m,
2743                                      unsigned long a, unsigned long b,
2744                                      unsigned long c, unsigned long d)
2745 {
2746 	unsigned int s;
2747 	unsigned int t;
2748 
2749 	r++; // make r be 1..64
2750 
2751 	t = 0;
2752 	s = LONGBITS;
2753 	if (s > 32) {
2754 		unsigned long d2 = (d >> 16) >> 16;
2755 		t = d2 + (d2 >> 16);
2756 		s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
2757 	}
2758 
2759 	t  = (d >> (s - 16)) & 0xff;
2760 	s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
2761 	t  = (c >> (s - 8)) & 0xf;
2762 	s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
2763 	t  = (b >> (s - 4)) & 0x7;
2764 	s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
2765 	t  = (a >> (s - 2)) & 0x3;
2766 	s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
2767 	t  = (m >> (s - 1)) & 0x1;
2768 	s -= ((t - r) & 256) >> 8;
2769 
2770 	return s - 1;
2771 }
2772 
2773 /* Prepare the bitmaps used by the fast implementation of the find_rank_bit()
2774  * above.
2775  */
mask_prep_rank_map(unsigned long m,unsigned long * a,unsigned long * b,unsigned long * c,unsigned long * d)2776 void mask_prep_rank_map(unsigned long m,
2777                         unsigned long *a, unsigned long *b,
2778                         unsigned long *c, unsigned long *d)
2779 {
2780 	*a =  m - ((m >> 1) & ~0UL/3);
2781 	*b = (*a & ~0UL/5) + ((*a >> 2) & ~0UL/5);
2782 	*c = (*b + (*b >> 4)) & ~0UL/0x11;
2783 	*d = (*c + (*c >> 8)) & ~0UL/0x101;
2784 }
2785 
2786 /* Return non-zero if IPv4 address is part of the network,
2787  * otherwise zero. Note that <addr> may not necessarily be aligned
2788  * while the two other ones must.
2789  */
in_net_ipv4(const void * addr,const struct in_addr * mask,const struct in_addr * net)2790 int in_net_ipv4(const void *addr, const struct in_addr *mask, const struct in_addr *net)
2791 {
2792 	struct in_addr addr_copy;
2793 
2794 	memcpy(&addr_copy, addr, sizeof(addr_copy));
2795 	return((addr_copy.s_addr & mask->s_addr) == (net->s_addr & mask->s_addr));
2796 }
2797 
2798 /* Return non-zero if IPv6 address is part of the network,
2799  * otherwise zero. Note that <addr> may not necessarily be aligned
2800  * while the two other ones must.
2801  */
in_net_ipv6(const void * addr,const struct in6_addr * mask,const struct in6_addr * net)2802 int in_net_ipv6(const void *addr, const struct in6_addr *mask, const struct in6_addr *net)
2803 {
2804 	int i;
2805 	struct in6_addr addr_copy;
2806 
2807 	memcpy(&addr_copy, addr, sizeof(addr_copy));
2808 	for (i = 0; i < sizeof(struct in6_addr) / sizeof(int); i++)
2809 		if (((((int *)&addr_copy)[i] & ((int *)mask)[i])) !=
2810 		    (((int *)net)[i] & ((int *)mask)[i]))
2811 			return 0;
2812 	return 1;
2813 }
2814 
2815 /* RFC 4291 prefix */
2816 const char rfc4291_pfx[] = { 0x00, 0x00, 0x00, 0x00,
2817 			     0x00, 0x00, 0x00, 0x00,
2818 			     0x00, 0x00, 0xFF, 0xFF };
2819 
2820 /* Map IPv4 address on IPv6 address, as specified in RFC 3513.
2821  * Input and output may overlap.
2822  */
v4tov6(struct in6_addr * sin6_addr,struct in_addr * sin_addr)2823 void v4tov6(struct in6_addr *sin6_addr, struct in_addr *sin_addr)
2824 {
2825 	struct in_addr tmp_addr;
2826 
2827 	tmp_addr.s_addr = sin_addr->s_addr;
2828 	memcpy(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx));
2829 	memcpy(sin6_addr->s6_addr+12, &tmp_addr.s_addr, 4);
2830 }
2831 
2832 /* Map IPv6 address on IPv4 address, as specified in RFC 3513.
2833  * Return true if conversion is possible and false otherwise.
2834  */
v6tov4(struct in_addr * sin_addr,struct in6_addr * sin6_addr)2835 int v6tov4(struct in_addr *sin_addr, struct in6_addr *sin6_addr)
2836 {
2837 	if (memcmp(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx)) == 0) {
2838 		memcpy(&(sin_addr->s_addr), &(sin6_addr->s6_addr[12]),
2839 			sizeof(struct in_addr));
2840 		return 1;
2841 	}
2842 
2843 	return 0;
2844 }
2845 
2846 /* compare two struct sockaddr_storage and return:
2847  *  0 (true)  if the addr is the same in both
2848  *  1 (false) if the addr is not the same in both
2849  *  -1 (unable) if one of the addr is not AF_INET*
2850  */
ipcmp(struct sockaddr_storage * ss1,struct sockaddr_storage * ss2)2851 int ipcmp(struct sockaddr_storage *ss1, struct sockaddr_storage *ss2)
2852 {
2853 	if ((ss1->ss_family != AF_INET) && (ss1->ss_family != AF_INET6))
2854 		return -1;
2855 
2856 	if ((ss2->ss_family != AF_INET) && (ss2->ss_family != AF_INET6))
2857 		return -1;
2858 
2859 	if (ss1->ss_family != ss2->ss_family)
2860 		return 1;
2861 
2862 	switch (ss1->ss_family) {
2863 		case AF_INET:
2864 			return memcmp(&((struct sockaddr_in *)ss1)->sin_addr,
2865 				      &((struct sockaddr_in *)ss2)->sin_addr,
2866 				      sizeof(struct in_addr)) != 0;
2867 		case AF_INET6:
2868 			return memcmp(&((struct sockaddr_in6 *)ss1)->sin6_addr,
2869 				      &((struct sockaddr_in6 *)ss2)->sin6_addr,
2870 				      sizeof(struct in6_addr)) != 0;
2871 	}
2872 
2873 	return 1;
2874 }
2875 
2876 /* copy IP address from <source> into <dest>
2877  * The caller must allocate and clear <dest> before calling.
2878  * The source must be in either AF_INET or AF_INET6 family, or the destination
2879  * address will be undefined. If the destination address used to hold a port,
2880  * it is preserved, so that this function can be used to switch to another
2881  * address family with no risk. Returns a pointer to the destination.
2882  */
ipcpy(struct sockaddr_storage * source,struct sockaddr_storage * dest)2883 struct sockaddr_storage *ipcpy(struct sockaddr_storage *source, struct sockaddr_storage *dest)
2884 {
2885 	int prev_port;
2886 
2887 	prev_port = get_net_port(dest);
2888 	memset(dest, 0, sizeof(*dest));
2889 	dest->ss_family = source->ss_family;
2890 
2891 	/* copy new addr and apply it */
2892 	switch (source->ss_family) {
2893 		case AF_INET:
2894 			((struct sockaddr_in *)dest)->sin_addr.s_addr = ((struct sockaddr_in *)source)->sin_addr.s_addr;
2895 			((struct sockaddr_in *)dest)->sin_port = prev_port;
2896 			break;
2897 		case AF_INET6:
2898 			memcpy(((struct sockaddr_in6 *)dest)->sin6_addr.s6_addr, ((struct sockaddr_in6 *)source)->sin6_addr.s6_addr, sizeof(struct in6_addr));
2899 			((struct sockaddr_in6 *)dest)->sin6_port = prev_port;
2900 			break;
2901 	}
2902 
2903 	return dest;
2904 }
2905 
human_time(int t,short hz_div)2906 char *human_time(int t, short hz_div) {
2907 	static char rv[sizeof("24855d23h")+1];	// longest of "23h59m" and "59m59s"
2908 	char *p = rv;
2909 	char *end = rv + sizeof(rv);
2910 	int cnt=2;				// print two numbers
2911 
2912 	if (unlikely(t < 0 || hz_div <= 0)) {
2913 		snprintf(p, end - p, "?");
2914 		return rv;
2915 	}
2916 
2917 	if (unlikely(hz_div > 1))
2918 		t /= hz_div;
2919 
2920 	if (t >= DAY) {
2921 		p += snprintf(p, end - p, "%dd", t / DAY);
2922 		cnt--;
2923 	}
2924 
2925 	if (cnt && t % DAY / HOUR) {
2926 		p += snprintf(p, end - p, "%dh", t % DAY / HOUR);
2927 		cnt--;
2928 	}
2929 
2930 	if (cnt && t % HOUR / MINUTE) {
2931 		p += snprintf(p, end - p, "%dm", t % HOUR / MINUTE);
2932 		cnt--;
2933 	}
2934 
2935 	if ((cnt && t % MINUTE) || !t)					// also display '0s'
2936 		p += snprintf(p, end - p, "%ds", t % MINUTE / SEC);
2937 
2938 	return rv;
2939 }
2940 
2941 const char *monthname[12] = {
2942 	"Jan", "Feb", "Mar", "Apr", "May", "Jun",
2943 	"Jul", "Aug", "Sep", "Oct", "Nov", "Dec"
2944 };
2945 
2946 /* date2str_log: write a date in the format :
2947  * 	sprintf(str, "%02d/%s/%04d:%02d:%02d:%02d.%03d",
2948  *		tm.tm_mday, monthname[tm.tm_mon], tm.tm_year+1900,
2949  *		tm.tm_hour, tm.tm_min, tm.tm_sec, (int)date.tv_usec/1000);
2950  *
2951  * without using sprintf. return a pointer to the last char written (\0) or
2952  * NULL if there isn't enough space.
2953  */
date2str_log(char * dst,const struct tm * tm,const struct timeval * date,size_t size)2954 char *date2str_log(char *dst, const struct tm *tm, const struct timeval *date, size_t size)
2955 {
2956 
2957 	if (size < 25) /* the size is fixed: 24 chars + \0 */
2958 		return NULL;
2959 
2960 	dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
2961 	if (!dst)
2962 		return NULL;
2963 	*dst++ = '/';
2964 
2965 	memcpy(dst, monthname[tm->tm_mon], 3); // month
2966 	dst += 3;
2967 	*dst++ = '/';
2968 
2969 	dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
2970 	if (!dst)
2971 		return NULL;
2972 	*dst++ = ':';
2973 
2974 	dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
2975 	if (!dst)
2976 		return NULL;
2977 	*dst++ = ':';
2978 
2979 	dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
2980 	if (!dst)
2981 		return NULL;
2982 	*dst++ = ':';
2983 
2984 	dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
2985 	if (!dst)
2986 		return NULL;
2987 	*dst++ = '.';
2988 
2989 	utoa_pad((unsigned int)(date->tv_usec/1000), dst, 4); // millisecondes
2990 	if (!dst)
2991 		return NULL;
2992 	dst += 3;  // only the 3 first digits
2993 	*dst = '\0';
2994 
2995 	return dst;
2996 }
2997 
2998 /* Base year used to compute leap years */
2999 #define TM_YEAR_BASE 1900
3000 
3001 /* Return the difference in seconds between two times (leap seconds are ignored).
3002  * Retrieved from glibc 2.18 source code.
3003  */
my_tm_diff(const struct tm * a,const struct tm * b)3004 static int my_tm_diff(const struct tm *a, const struct tm *b)
3005 {
3006 	/* Compute intervening leap days correctly even if year is negative.
3007 	 * Take care to avoid int overflow in leap day calculations,
3008 	 * but it's OK to assume that A and B are close to each other.
3009 	 */
3010 	int a4 = (a->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (a->tm_year & 3);
3011 	int b4 = (b->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (b->tm_year & 3);
3012 	int a100 = a4 / 25 - (a4 % 25 < 0);
3013 	int b100 = b4 / 25 - (b4 % 25 < 0);
3014 	int a400 = a100 >> 2;
3015 	int b400 = b100 >> 2;
3016 	int intervening_leap_days = (a4 - b4) - (a100 - b100) + (a400 - b400);
3017 	int years = a->tm_year - b->tm_year;
3018 	int days = (365 * years + intervening_leap_days
3019 	         + (a->tm_yday - b->tm_yday));
3020 	return (60 * (60 * (24 * days + (a->tm_hour - b->tm_hour))
3021 	       + (a->tm_min - b->tm_min))
3022 	       + (a->tm_sec - b->tm_sec));
3023 }
3024 
3025 /* Return the GMT offset for a specific local time.
3026  * Both t and tm must represent the same time.
3027  * The string returned has the same format as returned by strftime(... "%z", tm).
3028  * Offsets are kept in an internal cache for better performances.
3029  */
get_gmt_offset(time_t t,struct tm * tm)3030 const char *get_gmt_offset(time_t t, struct tm *tm)
3031 {
3032 	/* Cache offsets from GMT (depending on whether DST is active or not) */
3033 	static THREAD_LOCAL char gmt_offsets[2][5+1] = { "", "" };
3034 
3035 	char *gmt_offset;
3036 	struct tm tm_gmt;
3037 	int diff;
3038 	int isdst = tm->tm_isdst;
3039 
3040 	/* Pretend DST not active if its status is unknown */
3041 	if (isdst < 0)
3042 		isdst = 0;
3043 
3044 	/* Fetch the offset and initialize it if needed */
3045 	gmt_offset = gmt_offsets[isdst & 0x01];
3046 	if (unlikely(!*gmt_offset)) {
3047 		get_gmtime(t, &tm_gmt);
3048 		diff = my_tm_diff(tm, &tm_gmt);
3049 		if (diff < 0) {
3050 			diff = -diff;
3051 			*gmt_offset = '-';
3052 		} else {
3053 			*gmt_offset = '+';
3054 		}
3055 		diff %= 86400U;
3056 		diff /= 60; /* Convert to minutes */
3057 		snprintf(gmt_offset+1, 4+1, "%02d%02d", diff/60, diff%60);
3058 	}
3059 
3060 	return gmt_offset;
3061 }
3062 
3063 /* gmt2str_log: write a date in the format :
3064  * "%02d/%s/%04d:%02d:%02d:%02d +0000" without using snprintf
3065  * return a pointer to the last char written (\0) or
3066  * NULL if there isn't enough space.
3067  */
gmt2str_log(char * dst,struct tm * tm,size_t size)3068 char *gmt2str_log(char *dst, struct tm *tm, size_t size)
3069 {
3070 	if (size < 27) /* the size is fixed: 26 chars + \0 */
3071 		return NULL;
3072 
3073 	dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
3074 	if (!dst)
3075 		return NULL;
3076 	*dst++ = '/';
3077 
3078 	memcpy(dst, monthname[tm->tm_mon], 3); // month
3079 	dst += 3;
3080 	*dst++ = '/';
3081 
3082 	dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
3083 	if (!dst)
3084 		return NULL;
3085 	*dst++ = ':';
3086 
3087 	dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
3088 	if (!dst)
3089 		return NULL;
3090 	*dst++ = ':';
3091 
3092 	dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
3093 	if (!dst)
3094 		return NULL;
3095 	*dst++ = ':';
3096 
3097 	dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
3098 	if (!dst)
3099 		return NULL;
3100 	*dst++ = ' ';
3101 	*dst++ = '+';
3102 	*dst++ = '0';
3103 	*dst++ = '0';
3104 	*dst++ = '0';
3105 	*dst++ = '0';
3106 	*dst = '\0';
3107 
3108 	return dst;
3109 }
3110 
3111 /* localdate2str_log: write a date in the format :
3112  * "%02d/%s/%04d:%02d:%02d:%02d +0000(local timezone)" without using snprintf
3113  * Both t and tm must represent the same time.
3114  * return a pointer to the last char written (\0) or
3115  * NULL if there isn't enough space.
3116  */
localdate2str_log(char * dst,time_t t,struct tm * tm,size_t size)3117 char *localdate2str_log(char *dst, time_t t, struct tm *tm, size_t size)
3118 {
3119 	const char *gmt_offset;
3120 	if (size < 27) /* the size is fixed: 26 chars + \0 */
3121 		return NULL;
3122 
3123 	gmt_offset = get_gmt_offset(t, tm);
3124 
3125 	dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
3126 	if (!dst)
3127 		return NULL;
3128 	*dst++ = '/';
3129 
3130 	memcpy(dst, monthname[tm->tm_mon], 3); // month
3131 	dst += 3;
3132 	*dst++ = '/';
3133 
3134 	dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
3135 	if (!dst)
3136 		return NULL;
3137 	*dst++ = ':';
3138 
3139 	dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
3140 	if (!dst)
3141 		return NULL;
3142 	*dst++ = ':';
3143 
3144 	dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
3145 	if (!dst)
3146 		return NULL;
3147 	*dst++ = ':';
3148 
3149 	dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
3150 	if (!dst)
3151 		return NULL;
3152 	*dst++ = ' ';
3153 
3154 	memcpy(dst, gmt_offset, 5); // Offset from local time to GMT
3155 	dst += 5;
3156 	*dst = '\0';
3157 
3158 	return dst;
3159 }
3160 
3161 /* Returns the number of seconds since 01/01/1970 0:0:0 GMT for GMT date <tm>.
3162  * It is meant as a portable replacement for timegm() for use with valid inputs.
3163  * Returns undefined results for invalid dates (eg: months out of range 0..11).
3164  */
my_timegm(const struct tm * tm)3165 time_t my_timegm(const struct tm *tm)
3166 {
3167 	/* Each month has 28, 29, 30 or 31 days, or 28+N. The date in the year
3168 	 * is thus (current month - 1)*28 + cumulated_N[month] to count the
3169 	 * sum of the extra N days for elapsed months. The sum of all these N
3170 	 * days doesn't exceed 30 for a complete year (366-12*28) so it fits
3171 	 * in a 5-bit word. This means that with 60 bits we can represent a
3172 	 * matrix of all these values at once, which is fast and efficient to
3173 	 * access. The extra February day for leap years is not counted here.
3174 	 *
3175 	 * Jan : none      =  0 (0)
3176 	 * Feb : Jan       =  3 (3)
3177 	 * Mar : Jan..Feb  =  3 (3 + 0)
3178 	 * Apr : Jan..Mar  =  6 (3 + 0 + 3)
3179 	 * May : Jan..Apr  =  8 (3 + 0 + 3 + 2)
3180 	 * Jun : Jan..May  = 11 (3 + 0 + 3 + 2 + 3)
3181 	 * Jul : Jan..Jun  = 13 (3 + 0 + 3 + 2 + 3 + 2)
3182 	 * Aug : Jan..Jul  = 16 (3 + 0 + 3 + 2 + 3 + 2 + 3)
3183 	 * Sep : Jan..Aug  = 19 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3)
3184 	 * Oct : Jan..Sep  = 21 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2)
3185 	 * Nov : Jan..Oct  = 24 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3)
3186 	 * Dec : Jan..Nov  = 26 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3 + 2)
3187 	 */
3188 	uint64_t extra =
3189 		( 0ULL <<  0*5) + ( 3ULL <<  1*5) + ( 3ULL <<  2*5) + /* Jan, Feb, Mar, */
3190 		( 6ULL <<  3*5) + ( 8ULL <<  4*5) + (11ULL <<  5*5) + /* Apr, May, Jun, */
3191 		(13ULL <<  6*5) + (16ULL <<  7*5) + (19ULL <<  8*5) + /* Jul, Aug, Sep, */
3192 		(21ULL <<  9*5) + (24ULL << 10*5) + (26ULL << 11*5);  /* Oct, Nov, Dec, */
3193 
3194 	unsigned int y = tm->tm_year + 1900;
3195 	unsigned int m = tm->tm_mon;
3196 	unsigned long days = 0;
3197 
3198 	/* days since 1/1/1970 for full years */
3199 	days += days_since_zero(y) - days_since_zero(1970);
3200 
3201 	/* days for full months in the current year */
3202 	days += 28 * m + ((extra >> (m * 5)) & 0x1f);
3203 
3204 	/* count + 1 after March for leap years. A leap year is a year multiple
3205 	 * of 4, unless it's multiple of 100 without being multiple of 400. 2000
3206 	 * is leap, 1900 isn't, 1904 is.
3207 	 */
3208 	if ((m > 1) && !(y & 3) && ((y % 100) || !(y % 400)))
3209 		days++;
3210 
3211 	days += tm->tm_mday - 1;
3212 	return days * 86400ULL + tm->tm_hour * 3600 + tm->tm_min * 60 + tm->tm_sec;
3213 }
3214 
3215 /* This function check a char. It returns true and updates
3216  * <date> and <len> pointer to the new position if the
3217  * character is found.
3218  */
parse_expect_char(const char ** date,int * len,char c)3219 static inline int parse_expect_char(const char **date, int *len, char c)
3220 {
3221 	if (*len < 1 || **date != c)
3222 		return 0;
3223 	(*len)--;
3224 	(*date)++;
3225 	return 1;
3226 }
3227 
3228 /* This function expects a string <str> of len <l>. It return true and updates.
3229  * <date> and <len> if the string matches, otherwise, it returns false.
3230  */
parse_strcmp(const char ** date,int * len,char * str,int l)3231 static inline int parse_strcmp(const char **date, int *len, char *str, int l)
3232 {
3233 	if (*len < l || strncmp(*date, str, l) != 0)
3234 		return 0;
3235 	(*len) -= l;
3236 	(*date) += l;
3237 	return 1;
3238 }
3239 
3240 /* This macro converts 3 chars name in integer. */
3241 #define STR2I3(__a, __b, __c) ((__a) * 65536 + (__b) * 256 + (__c))
3242 
3243 /* day-name     = %x4D.6F.6E ; "Mon", case-sensitive
3244  *              / %x54.75.65 ; "Tue", case-sensitive
3245  *              / %x57.65.64 ; "Wed", case-sensitive
3246  *              / %x54.68.75 ; "Thu", case-sensitive
3247  *              / %x46.72.69 ; "Fri", case-sensitive
3248  *              / %x53.61.74 ; "Sat", case-sensitive
3249  *              / %x53.75.6E ; "Sun", case-sensitive
3250  *
3251  * This array must be alphabetically sorted
3252  */
parse_http_dayname(const char ** date,int * len,struct tm * tm)3253 static inline int parse_http_dayname(const char **date, int *len, struct tm *tm)
3254 {
3255 	if (*len < 3)
3256 		return 0;
3257 	switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3258 	case STR2I3('M','o','n'): tm->tm_wday = 1;  break;
3259 	case STR2I3('T','u','e'): tm->tm_wday = 2;  break;
3260 	case STR2I3('W','e','d'): tm->tm_wday = 3;  break;
3261 	case STR2I3('T','h','u'): tm->tm_wday = 4;  break;
3262 	case STR2I3('F','r','i'): tm->tm_wday = 5;  break;
3263 	case STR2I3('S','a','t'): tm->tm_wday = 6;  break;
3264 	case STR2I3('S','u','n'): tm->tm_wday = 7;  break;
3265 	default: return 0;
3266 	}
3267 	*len -= 3;
3268 	*date  += 3;
3269 	return 1;
3270 }
3271 
3272 /* month        = %x4A.61.6E ; "Jan", case-sensitive
3273  *              / %x46.65.62 ; "Feb", case-sensitive
3274  *              / %x4D.61.72 ; "Mar", case-sensitive
3275  *              / %x41.70.72 ; "Apr", case-sensitive
3276  *              / %x4D.61.79 ; "May", case-sensitive
3277  *              / %x4A.75.6E ; "Jun", case-sensitive
3278  *              / %x4A.75.6C ; "Jul", case-sensitive
3279  *              / %x41.75.67 ; "Aug", case-sensitive
3280  *              / %x53.65.70 ; "Sep", case-sensitive
3281  *              / %x4F.63.74 ; "Oct", case-sensitive
3282  *              / %x4E.6F.76 ; "Nov", case-sensitive
3283  *              / %x44.65.63 ; "Dec", case-sensitive
3284  *
3285  * This array must be alphabetically sorted
3286  */
parse_http_monthname(const char ** date,int * len,struct tm * tm)3287 static inline int parse_http_monthname(const char **date, int *len, struct tm *tm)
3288 {
3289 	if (*len < 3)
3290 		return 0;
3291 	switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3292 	case STR2I3('J','a','n'): tm->tm_mon = 0;  break;
3293 	case STR2I3('F','e','b'): tm->tm_mon = 1;  break;
3294 	case STR2I3('M','a','r'): tm->tm_mon = 2;  break;
3295 	case STR2I3('A','p','r'): tm->tm_mon = 3;  break;
3296 	case STR2I3('M','a','y'): tm->tm_mon = 4;  break;
3297 	case STR2I3('J','u','n'): tm->tm_mon = 5;  break;
3298 	case STR2I3('J','u','l'): tm->tm_mon = 6;  break;
3299 	case STR2I3('A','u','g'): tm->tm_mon = 7;  break;
3300 	case STR2I3('S','e','p'): tm->tm_mon = 8;  break;
3301 	case STR2I3('O','c','t'): tm->tm_mon = 9;  break;
3302 	case STR2I3('N','o','v'): tm->tm_mon = 10; break;
3303 	case STR2I3('D','e','c'): tm->tm_mon = 11; break;
3304 	default: return 0;
3305 	}
3306 	*len -= 3;
3307 	*date  += 3;
3308 	return 1;
3309 }
3310 
3311 /* day-name-l   = %x4D.6F.6E.64.61.79    ; "Monday", case-sensitive
3312  *        / %x54.75.65.73.64.61.79       ; "Tuesday", case-sensitive
3313  *        / %x57.65.64.6E.65.73.64.61.79 ; "Wednesday", case-sensitive
3314  *        / %x54.68.75.72.73.64.61.79    ; "Thursday", case-sensitive
3315  *        / %x46.72.69.64.61.79          ; "Friday", case-sensitive
3316  *        / %x53.61.74.75.72.64.61.79    ; "Saturday", case-sensitive
3317  *        / %x53.75.6E.64.61.79          ; "Sunday", case-sensitive
3318  *
3319  * This array must be alphabetically sorted
3320  */
parse_http_ldayname(const char ** date,int * len,struct tm * tm)3321 static inline int parse_http_ldayname(const char **date, int *len, struct tm *tm)
3322 {
3323 	if (*len < 6) /* Minimum length. */
3324 		return 0;
3325 	switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3326 	case STR2I3('M','o','n'):
3327 		RET0_UNLESS(parse_strcmp(date, len, "Monday", 6));
3328 		tm->tm_wday = 1;
3329 		return 1;
3330 	case STR2I3('T','u','e'):
3331 		RET0_UNLESS(parse_strcmp(date, len, "Tuesday", 7));
3332 		tm->tm_wday = 2;
3333 		return 1;
3334 	case STR2I3('W','e','d'):
3335 		RET0_UNLESS(parse_strcmp(date, len, "Wednesday", 9));
3336 		tm->tm_wday = 3;
3337 		return 1;
3338 	case STR2I3('T','h','u'):
3339 		RET0_UNLESS(parse_strcmp(date, len, "Thursday", 8));
3340 		tm->tm_wday = 4;
3341 		return 1;
3342 	case STR2I3('F','r','i'):
3343 		RET0_UNLESS(parse_strcmp(date, len, "Friday", 6));
3344 		tm->tm_wday = 5;
3345 		return 1;
3346 	case STR2I3('S','a','t'):
3347 		RET0_UNLESS(parse_strcmp(date, len, "Saturday", 8));
3348 		tm->tm_wday = 6;
3349 		return 1;
3350 	case STR2I3('S','u','n'):
3351 		RET0_UNLESS(parse_strcmp(date, len, "Sunday", 6));
3352 		tm->tm_wday = 7;
3353 		return 1;
3354 	}
3355 	return 0;
3356 }
3357 
3358 /* This function parses exactly 1 digit and returns the numeric value in "digit". */
parse_digit(const char ** date,int * len,int * digit)3359 static inline int parse_digit(const char **date, int *len, int *digit)
3360 {
3361 	if (*len < 1 || **date < '0' || **date > '9')
3362 		return 0;
3363 	*digit = (**date - '0');
3364 	(*date)++;
3365 	(*len)--;
3366 	return 1;
3367 }
3368 
3369 /* This function parses exactly 2 digits and returns the numeric value in "digit". */
parse_2digit(const char ** date,int * len,int * digit)3370 static inline int parse_2digit(const char **date, int *len, int *digit)
3371 {
3372 	int value;
3373 
3374 	RET0_UNLESS(parse_digit(date, len, &value));
3375 	(*digit) = value * 10;
3376 	RET0_UNLESS(parse_digit(date, len, &value));
3377 	(*digit) += value;
3378 
3379 	return 1;
3380 }
3381 
3382 /* This function parses exactly 4 digits and returns the numeric value in "digit". */
parse_4digit(const char ** date,int * len,int * digit)3383 static inline int parse_4digit(const char **date, int *len, int *digit)
3384 {
3385 	int value;
3386 
3387 	RET0_UNLESS(parse_digit(date, len, &value));
3388 	(*digit) = value * 1000;
3389 
3390 	RET0_UNLESS(parse_digit(date, len, &value));
3391 	(*digit) += value * 100;
3392 
3393 	RET0_UNLESS(parse_digit(date, len, &value));
3394 	(*digit) += value * 10;
3395 
3396 	RET0_UNLESS(parse_digit(date, len, &value));
3397 	(*digit) += value;
3398 
3399 	return 1;
3400 }
3401 
3402 /* time-of-day  = hour ":" minute ":" second
3403  *              ; 00:00:00 - 23:59:60 (leap second)
3404  *
3405  * hour         = 2DIGIT
3406  * minute       = 2DIGIT
3407  * second       = 2DIGIT
3408  */
parse_http_time(const char ** date,int * len,struct tm * tm)3409 static inline int parse_http_time(const char **date, int *len, struct tm *tm)
3410 {
3411 	RET0_UNLESS(parse_2digit(date, len, &tm->tm_hour)); /* hour 2DIGIT */
3412 	RET0_UNLESS(parse_expect_char(date, len, ':'));     /* expect ":"  */
3413 	RET0_UNLESS(parse_2digit(date, len, &tm->tm_min));  /* min 2DIGIT  */
3414 	RET0_UNLESS(parse_expect_char(date, len, ':'));     /* expect ":"  */
3415 	RET0_UNLESS(parse_2digit(date, len, &tm->tm_sec));  /* sec 2DIGIT  */
3416 	return 1;
3417 }
3418 
3419 /* From RFC7231
3420  * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3421  *
3422  * IMF-fixdate  = day-name "," SP date1 SP time-of-day SP GMT
3423  * ; fixed length/zone/capitalization subset of the format
3424  * ; see Section 3.3 of [RFC5322]
3425  *
3426  *
3427  * date1        = day SP month SP year
3428  *              ; e.g., 02 Jun 1982
3429  *
3430  * day          = 2DIGIT
3431  * year         = 4DIGIT
3432  *
3433  * GMT          = %x47.4D.54 ; "GMT", case-sensitive
3434  *
3435  * time-of-day  = hour ":" minute ":" second
3436  *              ; 00:00:00 - 23:59:60 (leap second)
3437  *
3438  * hour         = 2DIGIT
3439  * minute       = 2DIGIT
3440  * second       = 2DIGIT
3441  *
3442  * DIGIT        = decimal 0-9
3443  */
parse_imf_date(const char * date,int len,struct tm * tm)3444 int parse_imf_date(const char *date, int len, struct tm *tm)
3445 {
3446 	/* tm_gmtoff, if present, ought to be zero'ed */
3447 	memset(tm, 0, sizeof(*tm));
3448 
3449 	RET0_UNLESS(parse_http_dayname(&date, &len, tm));     /* day-name */
3450 	RET0_UNLESS(parse_expect_char(&date, &len, ','));     /* expect "," */
3451 	RET0_UNLESS(parse_expect_char(&date, &len, ' '));     /* expect SP */
3452 	RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
3453 	RET0_UNLESS(parse_expect_char(&date, &len, ' '));     /* expect SP */
3454 	RET0_UNLESS(parse_http_monthname(&date, &len, tm));   /* Month */
3455 	RET0_UNLESS(parse_expect_char(&date, &len, ' '));     /* expect SP */
3456 	RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
3457 	tm->tm_year -= 1900;
3458 	RET0_UNLESS(parse_expect_char(&date, &len, ' '));     /* expect SP */
3459 	RET0_UNLESS(parse_http_time(&date, &len, tm));        /* Parse time. */
3460 	RET0_UNLESS(parse_expect_char(&date, &len, ' '));     /* expect SP */
3461 	RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3));     /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
3462 	tm->tm_isdst = -1;
3463 	return 1;
3464 }
3465 
3466 /* From RFC7231
3467  * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3468  *
3469  * rfc850-date  = day-name-l "," SP date2 SP time-of-day SP GMT
3470  * date2        = day "-" month "-" 2DIGIT
3471  *              ; e.g., 02-Jun-82
3472  *
3473  * day          = 2DIGIT
3474  */
parse_rfc850_date(const char * date,int len,struct tm * tm)3475 int parse_rfc850_date(const char *date, int len, struct tm *tm)
3476 {
3477 	int year;
3478 
3479 	/* tm_gmtoff, if present, ought to be zero'ed */
3480 	memset(tm, 0, sizeof(*tm));
3481 
3482 	RET0_UNLESS(parse_http_ldayname(&date, &len, tm));    /* Read the day name */
3483 	RET0_UNLESS(parse_expect_char(&date, &len, ','));     /* expect "," */
3484 	RET0_UNLESS(parse_expect_char(&date, &len, ' '));     /* expect SP */
3485 	RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
3486 	RET0_UNLESS(parse_expect_char(&date, &len, '-'));     /* expect "-" */
3487 	RET0_UNLESS(parse_http_monthname(&date, &len, tm));   /* Month */
3488 	RET0_UNLESS(parse_expect_char(&date, &len, '-'));     /* expect "-" */
3489 
3490 	/* year = 2DIGIT
3491 	 *
3492 	 * Recipients of a timestamp value in rfc850-(*date) format, which uses a
3493 	 * two-digit year, MUST interpret a timestamp that appears to be more
3494 	 * than 50 years in the future as representing the most recent year in
3495 	 * the past that had the same last two digits.
3496 	 */
3497 	RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_year));
3498 
3499 	/* expect SP */
3500 	if (!parse_expect_char(&date, &len, ' ')) {
3501 		/* Maybe we have the date with 4 digits. */
3502 		RET0_UNLESS(parse_2digit(&date, &len, &year));
3503 		tm->tm_year = (tm->tm_year * 100 + year) - 1900;
3504 		/* expect SP */
3505 		RET0_UNLESS(parse_expect_char(&date, &len, ' '));
3506 	} else {
3507 		/* I fix 60 as pivot: >60: +1900, <60: +2000. Note that the
3508 		 * tm_year is the number of year since 1900, so for +1900, we
3509 		 * do nothing, and for +2000, we add 100.
3510 		 */
3511 		if (tm->tm_year <= 60)
3512 			tm->tm_year += 100;
3513 	}
3514 
3515 	RET0_UNLESS(parse_http_time(&date, &len, tm));    /* Parse time. */
3516 	RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3517 	RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3)); /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
3518 	tm->tm_isdst = -1;
3519 
3520 	return 1;
3521 }
3522 
3523 /* From RFC7231
3524  * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3525  *
3526  * asctime-date = day-name SP date3 SP time-of-day SP year
3527  * date3        = month SP ( 2DIGIT / ( SP 1DIGIT ))
3528  *              ; e.g., Jun  2
3529  *
3530  * HTTP-date is case sensitive.  A sender MUST NOT generate additional
3531  * whitespace in an HTTP-date beyond that specifically included as SP in
3532  * the grammar.
3533  */
parse_asctime_date(const char * date,int len,struct tm * tm)3534 int parse_asctime_date(const char *date, int len, struct tm *tm)
3535 {
3536 	/* tm_gmtoff, if present, ought to be zero'ed */
3537 	memset(tm, 0, sizeof(*tm));
3538 
3539 	RET0_UNLESS(parse_http_dayname(&date, &len, tm));   /* day-name */
3540 	RET0_UNLESS(parse_expect_char(&date, &len, ' '));   /* expect SP */
3541 	RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* expect month */
3542 	RET0_UNLESS(parse_expect_char(&date, &len, ' '));   /* expect SP */
3543 
3544 	/* expect SP and 1DIGIT or 2DIGIT */
3545 	if (parse_expect_char(&date, &len, ' '))
3546 		RET0_UNLESS(parse_digit(&date, &len, &tm->tm_mday));
3547 	else
3548 		RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday));
3549 
3550 	RET0_UNLESS(parse_expect_char(&date, &len, ' '));     /* expect SP */
3551 	RET0_UNLESS(parse_http_time(&date, &len, tm));        /* Parse time. */
3552 	RET0_UNLESS(parse_expect_char(&date, &len, ' '));     /* expect SP */
3553 	RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
3554 	tm->tm_year -= 1900;
3555 	tm->tm_isdst = -1;
3556 	return 1;
3557 }
3558 
3559 /* From RFC7231
3560  * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3561  *
3562  * HTTP-date    = IMF-fixdate / obs-date
3563  * obs-date     = rfc850-date / asctime-date
3564  *
3565  * parses an HTTP date in the RFC format and is accepted
3566  * alternatives. <date> is the strinf containing the date,
3567  * len is the len of the string. <tm> is filled with the
3568  * parsed time. We must considers this time as GMT.
3569  */
parse_http_date(const char * date,int len,struct tm * tm)3570 int parse_http_date(const char *date, int len, struct tm *tm)
3571 {
3572 	if (parse_imf_date(date, len, tm))
3573 		return 1;
3574 
3575 	if (parse_rfc850_date(date, len, tm))
3576 		return 1;
3577 
3578 	if (parse_asctime_date(date, len, tm))
3579 		return 1;
3580 
3581 	return 0;
3582 }
3583 
3584 /* Dynamically allocates a string of the proper length to hold the formatted
3585  * output. NULL is returned on error. The caller is responsible for freeing the
3586  * memory area using free(). The resulting string is returned in <out> if the
3587  * pointer is not NULL. A previous version of <out> might be used to build the
3588  * new string, and it will be freed before returning if it is not NULL, which
3589  * makes it possible to build complex strings from iterative calls without
3590  * having to care about freeing intermediate values, as in the example below :
3591  *
3592  *     memprintf(&err, "invalid argument: '%s'", arg);
3593  *     ...
3594  *     memprintf(&err, "parser said : <%s>\n", *err);
3595  *     ...
3596  *     free(*err);
3597  *
3598  * This means that <err> must be initialized to NULL before first invocation.
3599  * The return value also holds the allocated string, which eases error checking
3600  * and immediate consumption. If the output pointer is not used, NULL must be
3601  * passed instead and it will be ignored. The returned message will then also
3602  * be NULL so that the caller does not have to bother with freeing anything.
3603  *
3604  * It is also convenient to use it without any free except the last one :
3605  *    err = NULL;
3606  *    if (!fct1(err)) report(*err);
3607  *    if (!fct2(err)) report(*err);
3608  *    if (!fct3(err)) report(*err);
3609  *    free(*err);
3610  *
3611  * memprintf relies on memvprintf. This last version can be called from any
3612  * function with variadic arguments.
3613  */
memvprintf(char ** out,const char * format,va_list orig_args)3614 char *memvprintf(char **out, const char *format, va_list orig_args)
3615 {
3616 	va_list args;
3617 	char *ret = NULL;
3618 	int allocated = 0;
3619 	int needed = 0;
3620 
3621 	if (!out)
3622 		return NULL;
3623 
3624 	do {
3625 		char buf1;
3626 
3627 		/* vsnprintf() will return the required length even when the
3628 		 * target buffer is NULL. We do this in a loop just in case
3629 		 * intermediate evaluations get wrong.
3630 		 */
3631 		va_copy(args, orig_args);
3632 		needed = vsnprintf(ret ? ret : &buf1, allocated, format, args);
3633 		va_end(args);
3634 		if (needed < allocated) {
3635 			/* Note: on Solaris 8, the first iteration always
3636 			 * returns -1 if allocated is zero, so we force a
3637 			 * retry.
3638 			 */
3639 			if (!allocated)
3640 				needed = 0;
3641 			else
3642 				break;
3643 		}
3644 
3645 		allocated = needed + 1;
3646 		ret = my_realloc2(ret, allocated);
3647 	} while (ret);
3648 
3649 	if (needed < 0) {
3650 		/* an error was encountered */
3651 		free(ret);
3652 		ret = NULL;
3653 	}
3654 
3655 	if (out) {
3656 		free(*out);
3657 		*out = ret;
3658 	}
3659 
3660 	return ret;
3661 }
3662 
memprintf(char ** out,const char * format,...)3663 char *memprintf(char **out, const char *format, ...)
3664 {
3665 	va_list args;
3666 	char *ret = NULL;
3667 
3668 	va_start(args, format);
3669 	ret = memvprintf(out, format, args);
3670 	va_end(args);
3671 
3672 	return ret;
3673 }
3674 
3675 /* Used to add <level> spaces before each line of <out>, unless there is only one line.
3676  * The input argument is automatically freed and reassigned. The result will have to be
3677  * freed by the caller. It also supports being passed a NULL which results in the same
3678  * output.
3679  * Example of use :
3680  *   parse(cmd, &err); (callee: memprintf(&err, ...))
3681  *   fprintf(stderr, "Parser said: %s\n", indent_error(&err));
3682  *   free(err);
3683  */
indent_msg(char ** out,int level)3684 char *indent_msg(char **out, int level)
3685 {
3686 	char *ret, *in, *p;
3687 	int needed = 0;
3688 	int lf = 0;
3689 	int lastlf = 0;
3690 	int len;
3691 
3692 	if (!out || !*out)
3693 		return NULL;
3694 
3695 	in = *out - 1;
3696 	while ((in = strchr(in + 1, '\n')) != NULL) {
3697 		lastlf = in - *out;
3698 		lf++;
3699 	}
3700 
3701 	if (!lf) /* single line, no LF, return it as-is */
3702 		return *out;
3703 
3704 	len = strlen(*out);
3705 
3706 	if (lf == 1 && lastlf == len - 1) {
3707 		/* single line, LF at end, strip it and return as-is */
3708 		(*out)[lastlf] = 0;
3709 		return *out;
3710 	}
3711 
3712 	/* OK now we have at least one LF, we need to process the whole string
3713 	 * as a multi-line string. What we'll do :
3714 	 *   - prefix with an LF if there is none
3715 	 *   - add <level> spaces before each line
3716 	 * This means at most ( 1 + level + (len-lf) + lf*<1+level) ) =
3717 	 *   1 + level + len + lf * level = 1 + level * (lf + 1) + len.
3718 	 */
3719 
3720 	needed = 1 + level * (lf + 1) + len + 1;
3721 	p = ret = malloc(needed);
3722 	in = *out;
3723 
3724 	/* skip initial LFs */
3725 	while (*in == '\n')
3726 		in++;
3727 
3728 	/* copy each line, prefixed with LF and <level> spaces, and without the trailing LF */
3729 	while (*in) {
3730 		*p++ = '\n';
3731 		memset(p, ' ', level);
3732 		p += level;
3733 		do {
3734 			*p++ = *in++;
3735 		} while (*in && *in != '\n');
3736 		if (*in)
3737 			in++;
3738 	}
3739 	*p = 0;
3740 
3741 	free(*out);
3742 	*out = ret;
3743 
3744 	return ret;
3745 }
3746 
3747 /* makes a copy of message <in> into <out>, with each line prefixed with <pfx>
3748  * and end of lines replaced with <eol> if not 0. The first line to indent has
3749  * to be indicated in <first> (starts at zero), so that it is possible to skip
3750  * indenting the first line if it has to be appended after an existing message.
3751  * Empty strings are never indented, and NULL strings are considered empty both
3752  * for <in> and <pfx>. It returns non-zero if an EOL was appended as the last
3753  * character, non-zero otherwise.
3754  */
append_prefixed_str(struct buffer * out,const char * in,const char * pfx,char eol,int first)3755 int append_prefixed_str(struct buffer *out, const char *in, const char *pfx, char eol, int first)
3756 {
3757 	int bol, lf;
3758 	int pfxlen = pfx ? strlen(pfx) : 0;
3759 
3760 	if (!in)
3761 		return 0;
3762 
3763 	bol = 1;
3764 	lf = 0;
3765 	while (*in) {
3766 		if (bol && pfxlen) {
3767 			if (first > 0)
3768 				first--;
3769 			else
3770 				b_putblk(out, pfx, pfxlen);
3771 			bol = 0;
3772 		}
3773 
3774 		lf = (*in == '\n');
3775 		bol |= lf;
3776 		b_putchr(out, (lf && eol) ? eol : *in);
3777 		in++;
3778 	}
3779 	return lf;
3780 }
3781 
3782 /* removes environment variable <name> from the environment as found in
3783  * environ. This is only provided as an alternative for systems without
3784  * unsetenv() (old Solaris and AIX versions). THIS IS NOT THREAD SAFE.
3785  * The principle is to scan environ for each occurence of variable name
3786  * <name> and to replace the matching pointers with the last pointer of
3787  * the array (since variables are not ordered).
3788  * It always returns 0 (success).
3789  */
my_unsetenv(const char * name)3790 int my_unsetenv(const char *name)
3791 {
3792 	extern char **environ;
3793 	char **p = environ;
3794 	int vars;
3795 	int next;
3796 	int len;
3797 
3798 	len = strlen(name);
3799 	for (vars = 0; p[vars]; vars++)
3800 		;
3801 	next = 0;
3802 	while (next < vars) {
3803 		if (strncmp(p[next], name, len) != 0 || p[next][len] != '=') {
3804 			next++;
3805 			continue;
3806 		}
3807 		if (next < vars - 1)
3808 			p[next] = p[vars - 1];
3809 		p[--vars] = NULL;
3810 	}
3811 	return 0;
3812 }
3813 
3814 /* Convert occurrences of environment variables in the input string to their
3815  * corresponding value. A variable is identified as a series of alphanumeric
3816  * characters or underscores following a '$' sign. The <in> string must be
3817  * free()able. NULL returns NULL. The resulting string might be reallocated if
3818  * some expansion is made. Variable names may also be enclosed into braces if
3819  * needed (eg: to concatenate alphanum characters).
3820  */
env_expand(char * in)3821 char *env_expand(char *in)
3822 {
3823 	char *txt_beg;
3824 	char *out;
3825 	char *txt_end;
3826 	char *var_beg;
3827 	char *var_end;
3828 	char *value;
3829 	char *next;
3830 	int out_len;
3831 	int val_len;
3832 
3833 	if (!in)
3834 		return in;
3835 
3836 	value = out = NULL;
3837 	out_len = 0;
3838 
3839 	txt_beg = in;
3840 	do {
3841 		/* look for next '$' sign in <in> */
3842 		for (txt_end = txt_beg; *txt_end && *txt_end != '$'; txt_end++);
3843 
3844 		if (!*txt_end && !out) /* end and no expansion performed */
3845 			return in;
3846 
3847 		val_len = 0;
3848 		next = txt_end;
3849 		if (*txt_end == '$') {
3850 			char save;
3851 
3852 			var_beg = txt_end + 1;
3853 			if (*var_beg == '{')
3854 				var_beg++;
3855 
3856 			var_end = var_beg;
3857 			while (isalnum((int)(unsigned char)*var_end) || *var_end == '_') {
3858 				var_end++;
3859 			}
3860 
3861 			next = var_end;
3862 			if (*var_end == '}' && (var_beg > txt_end + 1))
3863 				next++;
3864 
3865 			/* get value of the variable name at this location */
3866 			save = *var_end;
3867 			*var_end = '\0';
3868 			value = getenv(var_beg);
3869 			*var_end = save;
3870 			val_len = value ? strlen(value) : 0;
3871 		}
3872 
3873 		out = my_realloc2(out, out_len + (txt_end - txt_beg) + val_len + 1);
3874 		if (txt_end > txt_beg) {
3875 			memcpy(out + out_len, txt_beg, txt_end - txt_beg);
3876 			out_len += txt_end - txt_beg;
3877 		}
3878 		if (val_len) {
3879 			memcpy(out + out_len, value, val_len);
3880 			out_len += val_len;
3881 		}
3882 		out[out_len] = 0;
3883 		txt_beg = next;
3884 	} while (*txt_beg);
3885 
3886 	/* here we know that <out> was allocated and that we don't need <in> anymore */
3887 	free(in);
3888 	return out;
3889 }
3890 
3891 
3892 /* same as strstr() but case-insensitive and with limit length */
strnistr(const char * str1,int len_str1,const char * str2,int len_str2)3893 const char *strnistr(const char *str1, int len_str1, const char *str2, int len_str2)
3894 {
3895 	char *pptr, *sptr, *start;
3896 	unsigned int slen, plen;
3897 	unsigned int tmp1, tmp2;
3898 
3899 	if (str1 == NULL || len_str1 == 0) // search pattern into an empty string => search is not found
3900 		return NULL;
3901 
3902 	if (str2 == NULL || len_str2 == 0) // pattern is empty => every str1 match
3903 		return str1;
3904 
3905 	if (len_str1 < len_str2) // pattern is longer than string => search is not found
3906 		return NULL;
3907 
3908 	for (tmp1 = 0, start = (char *)str1, pptr = (char *)str2, slen = len_str1, plen = len_str2; slen >= plen; start++, slen--) {
3909 		while (toupper(*start) != toupper(*str2)) {
3910 			start++;
3911 			slen--;
3912 			tmp1++;
3913 
3914 			if (tmp1 >= len_str1)
3915 				return NULL;
3916 
3917 			/* if pattern longer than string */
3918 			if (slen < plen)
3919 				return NULL;
3920 		}
3921 
3922 		sptr = start;
3923 		pptr = (char *)str2;
3924 
3925 		tmp2 = 0;
3926 		while (toupper(*sptr) == toupper(*pptr)) {
3927 			sptr++;
3928 			pptr++;
3929 			tmp2++;
3930 
3931 			if (*pptr == '\0' || tmp2 == len_str2) /* end of pattern found */
3932 				return start;
3933 			if (*sptr == '\0' || tmp2 == len_str1) /* end of string found and the pattern is not fully found */
3934 				return NULL;
3935 		}
3936 	}
3937 	return NULL;
3938 }
3939 
3940 /* This function read the next valid utf8 char.
3941  * <s> is the byte srray to be decode, <len> is its length.
3942  * The function returns decoded char encoded like this:
3943  * The 4 msb are the return code (UTF8_CODE_*), the 4 lsb
3944  * are the length read. The decoded character is stored in <c>.
3945  */
utf8_next(const char * s,int len,unsigned int * c)3946 unsigned char utf8_next(const char *s, int len, unsigned int *c)
3947 {
3948 	const unsigned char *p = (unsigned char *)s;
3949 	int dec;
3950 	unsigned char code = UTF8_CODE_OK;
3951 
3952 	if (len < 1)
3953 		return UTF8_CODE_OK;
3954 
3955 	/* Check the type of UTF8 sequence
3956 	 *
3957 	 * 0... ....  0x00 <= x <= 0x7f : 1 byte: ascii char
3958 	 * 10.. ....  0x80 <= x <= 0xbf : invalid sequence
3959 	 * 110. ....  0xc0 <= x <= 0xdf : 2 bytes
3960 	 * 1110 ....  0xe0 <= x <= 0xef : 3 bytes
3961 	 * 1111 0...  0xf0 <= x <= 0xf7 : 4 bytes
3962 	 * 1111 10..  0xf8 <= x <= 0xfb : 5 bytes
3963 	 * 1111 110.  0xfc <= x <= 0xfd : 6 bytes
3964 	 * 1111 111.  0xfe <= x <= 0xff : invalid sequence
3965 	 */
3966 	switch (*p) {
3967 	case 0x00 ... 0x7f:
3968 		*c = *p;
3969 		return UTF8_CODE_OK | 1;
3970 
3971 	case 0x80 ... 0xbf:
3972 		*c = *p;
3973 		return UTF8_CODE_BADSEQ | 1;
3974 
3975 	case 0xc0 ... 0xdf:
3976 		if (len < 2) {
3977 			*c = *p;
3978 			return UTF8_CODE_BADSEQ | 1;
3979 		}
3980 		*c = *p & 0x1f;
3981 		dec = 1;
3982 		break;
3983 
3984 	case 0xe0 ... 0xef:
3985 		if (len < 3) {
3986 			*c = *p;
3987 			return UTF8_CODE_BADSEQ | 1;
3988 		}
3989 		*c = *p & 0x0f;
3990 		dec = 2;
3991 		break;
3992 
3993 	case 0xf0 ... 0xf7:
3994 		if (len < 4) {
3995 			*c = *p;
3996 			return UTF8_CODE_BADSEQ | 1;
3997 		}
3998 		*c = *p & 0x07;
3999 		dec = 3;
4000 		break;
4001 
4002 	case 0xf8 ... 0xfb:
4003 		if (len < 5) {
4004 			*c = *p;
4005 			return UTF8_CODE_BADSEQ | 1;
4006 		}
4007 		*c = *p & 0x03;
4008 		dec = 4;
4009 		break;
4010 
4011 	case 0xfc ... 0xfd:
4012 		if (len < 6) {
4013 			*c = *p;
4014 			return UTF8_CODE_BADSEQ | 1;
4015 		}
4016 		*c = *p & 0x01;
4017 		dec = 5;
4018 		break;
4019 
4020 	case 0xfe ... 0xff:
4021 	default:
4022 		*c = *p;
4023 		return UTF8_CODE_BADSEQ | 1;
4024 	}
4025 
4026 	p++;
4027 
4028 	while (dec > 0) {
4029 
4030 		/* need 0x10 for the 2 first bits */
4031 		if ( ( *p & 0xc0 ) != 0x80 )
4032 			return UTF8_CODE_BADSEQ | ((p-(unsigned char *)s)&0xffff);
4033 
4034 		/* add data at char */
4035 		*c = ( *c << 6 ) | ( *p & 0x3f );
4036 
4037 		dec--;
4038 		p++;
4039 	}
4040 
4041 	/* Check ovelong encoding.
4042 	 * 1 byte  : 5 + 6         : 11 : 0x80    ... 0x7ff
4043 	 * 2 bytes : 4 + 6 + 6     : 16 : 0x800   ... 0xffff
4044 	 * 3 bytes : 3 + 6 + 6 + 6 : 21 : 0x10000 ... 0x1fffff
4045 	 */
4046 	if ((                 *c <= 0x7f     && (p-(unsigned char *)s) > 1) ||
4047 	    (*c >= 0x80    && *c <= 0x7ff    && (p-(unsigned char *)s) > 2) ||
4048 	    (*c >= 0x800   && *c <= 0xffff   && (p-(unsigned char *)s) > 3) ||
4049 	    (*c >= 0x10000 && *c <= 0x1fffff && (p-(unsigned char *)s) > 4))
4050 		code |= UTF8_CODE_OVERLONG;
4051 
4052 	/* Check invalid UTF8 range. */
4053 	if ((*c >= 0xd800 && *c <= 0xdfff) ||
4054 	    (*c >= 0xfffe && *c <= 0xffff))
4055 		code |= UTF8_CODE_INVRANGE;
4056 
4057 	return code | ((p-(unsigned char *)s)&0x0f);
4058 }
4059 
4060 /* append a copy of string <str> (in a wordlist) at the end of the list <li>
4061  * On failure : return 0 and <err> filled with an error message.
4062  * The caller is responsible for freeing the <err> and <str> copy
4063  * memory area using free()
4064  */
list_append_word(struct list * li,const char * str,char ** err)4065 int list_append_word(struct list *li, const char *str, char **err)
4066 {
4067 	struct wordlist *wl;
4068 
4069 	wl = calloc(1, sizeof(*wl));
4070 	if (!wl) {
4071 		memprintf(err, "out of memory");
4072 		goto fail_wl;
4073 	}
4074 
4075 	wl->s = strdup(str);
4076 	if (!wl->s) {
4077 		memprintf(err, "out of memory");
4078 		goto fail_wl_s;
4079 	}
4080 
4081 	LIST_ADDQ(li, &wl->list);
4082 
4083 	return 1;
4084 
4085 fail_wl_s:
4086 	free(wl->s);
4087 fail_wl:
4088 	free(wl);
4089 	return 0;
4090 }
4091 
4092 /* indicates if a memory location may safely be read or not. The trick consists
4093  * in performing a harmless syscall using this location as an input and letting
4094  * the operating system report whether it's OK or not. For this we have the
4095  * stat() syscall, which will return EFAULT when the memory location supposed
4096  * to contain the file name is not readable. If it is readable it will then
4097  * either return 0 if the area contains an existing file name, or -1 with
4098  * another code. This must not be abused, and some audit systems might detect
4099  * this as abnormal activity. It's used only for unsafe dumps.
4100  */
may_access(const void * ptr)4101 int may_access(const void *ptr)
4102 {
4103 	struct stat buf;
4104 
4105 	if (stat(ptr, &buf) == 0)
4106 		return 1;
4107 	if (errno == EFAULT)
4108 		return 0;
4109 	return 1;
4110 }
4111 
4112 /* print a string of text buffer to <out>. The format is :
4113  * Non-printable chars \t, \n, \r and \e are * encoded in C format.
4114  * Other non-printable chars are encoded "\xHH". Space, '\', and '=' are also escaped.
4115  * Print stopped if null char or <bsize> is reached, or if no more place in the chunk.
4116  */
dump_text(struct buffer * out,const char * buf,int bsize)4117 int dump_text(struct buffer *out, const char *buf, int bsize)
4118 {
4119 	unsigned char c;
4120 	int ptr = 0;
4121 
4122 	while (buf[ptr] && ptr < bsize) {
4123 		c = buf[ptr];
4124 		if (isprint(c) && isascii(c) && c != '\\' && c != ' ' && c != '=') {
4125 			if (out->data > out->size - 1)
4126 				break;
4127 			out->area[out->data++] = c;
4128 		}
4129 		else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\' || c == ' ' || c == '=') {
4130 			if (out->data > out->size - 2)
4131 				break;
4132 			out->area[out->data++] = '\\';
4133 			switch (c) {
4134 			case ' ': c = ' '; break;
4135 			case '\t': c = 't'; break;
4136 			case '\n': c = 'n'; break;
4137 			case '\r': c = 'r'; break;
4138 			case '\e': c = 'e'; break;
4139 			case '\\': c = '\\'; break;
4140 			case '=': c = '='; break;
4141 			}
4142 			out->area[out->data++] = c;
4143 		}
4144 		else {
4145 			if (out->data > out->size - 4)
4146 				break;
4147 			out->area[out->data++] = '\\';
4148 			out->area[out->data++] = 'x';
4149 			out->area[out->data++] = hextab[(c >> 4) & 0xF];
4150 			out->area[out->data++] = hextab[c & 0xF];
4151 		}
4152 		ptr++;
4153 	}
4154 
4155 	return ptr;
4156 }
4157 
4158 /* print a buffer in hexa.
4159  * Print stopped if <bsize> is reached, or if no more place in the chunk.
4160  */
dump_binary(struct buffer * out,const char * buf,int bsize)4161 int dump_binary(struct buffer *out, const char *buf, int bsize)
4162 {
4163 	unsigned char c;
4164 	int ptr = 0;
4165 
4166 	while (ptr < bsize) {
4167 		c = buf[ptr];
4168 
4169 		if (out->data > out->size - 2)
4170 			break;
4171 		out->area[out->data++] = hextab[(c >> 4) & 0xF];
4172 		out->area[out->data++] = hextab[c & 0xF];
4173 
4174 		ptr++;
4175 	}
4176 	return ptr;
4177 }
4178 
4179 /* Appends into buffer <out> a hex dump of memory area <buf> for <len> bytes,
4180  * prepending each line with prefix <pfx>. The output is *not* initialized.
4181  * The output will not wrap pas the buffer's end so it is more optimal if the
4182  * caller makes sure the buffer is aligned first. A trailing zero will always
4183  * be appended (and not counted) if there is room for it. The caller must make
4184  * sure that the area is dumpable first. If <unsafe> is non-null, the memory
4185  * locations are checked first for being readable.
4186  */
dump_hex(struct buffer * out,const char * pfx,const void * buf,int len,int unsafe)4187 void dump_hex(struct buffer *out, const char *pfx, const void *buf, int len, int unsafe)
4188 {
4189 	const unsigned char *d = buf;
4190 	int i, j, start;
4191 
4192 	d = (const unsigned char *)(((unsigned long)buf) & -16);
4193 	start = ((unsigned long)buf) & 15;
4194 
4195 	for (i = 0; i < start + len; i += 16) {
4196 		chunk_appendf(out, (sizeof(void *) == 4) ? "%s%8p: " : "%s%16p: ", pfx, d + i);
4197 
4198 		// 0: unchecked, 1: checked safe, 2: danger
4199 		unsafe = !!unsafe;
4200 		if (unsafe && !may_access(d + i))
4201 			unsafe = 2;
4202 
4203 		for (j = 0; j < 16; j++) {
4204 			if ((i + j < start) || (i + j >= start + len))
4205 				chunk_strcat(out, "'' ");
4206 			else if (unsafe > 1)
4207 				chunk_strcat(out, "** ");
4208 			else
4209 				chunk_appendf(out, "%02x ", d[i + j]);
4210 
4211 			if (j == 7)
4212 				chunk_strcat(out, "- ");
4213 		}
4214 		chunk_strcat(out, "  ");
4215 		for (j = 0; j < 16; j++) {
4216 			if ((i + j < start) || (i + j >= start + len))
4217 				chunk_strcat(out, "'");
4218 			else if (unsafe > 1)
4219 				chunk_strcat(out, "*");
4220 			else if (isprint(d[i + j]))
4221 				chunk_appendf(out, "%c", d[i + j]);
4222 			else
4223 				chunk_strcat(out, ".");
4224 		}
4225 		chunk_strcat(out, "\n");
4226 	}
4227 }
4228 
4229 /* dumps <pfx> followed by <n> bytes from <addr> in hex form into buffer <buf>
4230  * enclosed in brackets after the address itself, formatted on 14 chars
4231  * including the "0x" prefix. This is meant to be used as a prefix for code
4232  * areas. For example:
4233  *    "0x7f10b6557690 [48 c7 c0 0f 00 00 00 0f]"
4234  * It relies on may_access() to know if the bytes are dumpable, otherwise "--"
4235  * is emitted. A NULL <pfx> will be considered empty.
4236  */
dump_addr_and_bytes(struct buffer * buf,const char * pfx,const void * addr,int n)4237 void dump_addr_and_bytes(struct buffer *buf, const char *pfx, const void *addr, int n)
4238 {
4239 	int ok = 0;
4240 	int i;
4241 
4242 	chunk_appendf(buf, "%s%#14lx [", pfx ? pfx : "", (long)addr);
4243 
4244 	for (i = 0; i < n; i++) {
4245 		if (i == 0 || (((long)(addr + i) ^ (long)(addr)) & 4096))
4246 			ok = may_access(addr + i);
4247 		if (ok)
4248 			chunk_appendf(buf, "%02x%s", ((uint8_t*)addr)[i], (i<n-1) ? " " : "]");
4249 		else
4250 			chunk_appendf(buf, "--%s", (i<n-1) ? " " : "]");
4251 	}
4252 }
4253 
4254 /* print a line of text buffer (limited to 70 bytes) to <out>. The format is :
4255  * <2 spaces> <offset=5 digits> <space or plus> <space> <70 chars max> <\n>
4256  * which is 60 chars per line. Non-printable chars \t, \n, \r and \e are
4257  * encoded in C format. Other non-printable chars are encoded "\xHH". Original
4258  * lines are respected within the limit of 70 output chars. Lines that are
4259  * continuation of a previous truncated line begin with "+" instead of " "
4260  * after the offset. The new pointer is returned.
4261  */
dump_text_line(struct buffer * out,const char * buf,int bsize,int len,int * line,int ptr)4262 int dump_text_line(struct buffer *out, const char *buf, int bsize, int len,
4263                    int *line, int ptr)
4264 {
4265 	int end;
4266 	unsigned char c;
4267 
4268 	end = out->data + 80;
4269 	if (end > out->size)
4270 		return ptr;
4271 
4272 	chunk_appendf(out, "  %05d%c ", ptr, (ptr == *line) ? ' ' : '+');
4273 
4274 	while (ptr < len && ptr < bsize) {
4275 		c = buf[ptr];
4276 		if (isprint(c) && isascii(c) && c != '\\') {
4277 			if (out->data > end - 2)
4278 				break;
4279 			out->area[out->data++] = c;
4280 		} else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\') {
4281 			if (out->data > end - 3)
4282 				break;
4283 			out->area[out->data++] = '\\';
4284 			switch (c) {
4285 			case '\t': c = 't'; break;
4286 			case '\n': c = 'n'; break;
4287 			case '\r': c = 'r'; break;
4288 			case '\e': c = 'e'; break;
4289 			case '\\': c = '\\'; break;
4290 			}
4291 			out->area[out->data++] = c;
4292 		} else {
4293 			if (out->data > end - 5)
4294 				break;
4295 			out->area[out->data++] = '\\';
4296 			out->area[out->data++] = 'x';
4297 			out->area[out->data++] = hextab[(c >> 4) & 0xF];
4298 			out->area[out->data++] = hextab[c & 0xF];
4299 		}
4300 		if (buf[ptr++] == '\n') {
4301 			/* we had a line break, let's return now */
4302 			out->area[out->data++] = '\n';
4303 			*line = ptr;
4304 			return ptr;
4305 		}
4306 	}
4307 	/* we have an incomplete line, we return it as-is */
4308 	out->area[out->data++] = '\n';
4309 	return ptr;
4310 }
4311 
4312 /* displays a <len> long memory block at <buf>, assuming first byte of <buf>
4313  * has address <baseaddr>. String <pfx> may be placed as a prefix in front of
4314  * each line. It may be NULL if unused. The output is emitted to file <out>.
4315  */
debug_hexdump(FILE * out,const char * pfx,const char * buf,unsigned int baseaddr,int len)4316 void debug_hexdump(FILE *out, const char *pfx, const char *buf,
4317                    unsigned int baseaddr, int len)
4318 {
4319 	unsigned int i;
4320 	int b, j;
4321 
4322 	for (i = 0; i < (len + (baseaddr & 15)); i += 16) {
4323 		b = i - (baseaddr & 15);
4324 		fprintf(out, "%s%08x: ", pfx ? pfx : "", i + (baseaddr & ~15));
4325 		for (j = 0; j < 8; j++) {
4326 			if (b + j >= 0 && b + j < len)
4327 				fprintf(out, "%02x ", (unsigned char)buf[b + j]);
4328 			else
4329 				fprintf(out, "   ");
4330 		}
4331 
4332 		if (b + j >= 0 && b + j < len)
4333 			fputc('-', out);
4334 		else
4335 			fputc(' ', out);
4336 
4337 		for (j = 8; j < 16; j++) {
4338 			if (b + j >= 0 && b + j < len)
4339 				fprintf(out, " %02x", (unsigned char)buf[b + j]);
4340 			else
4341 				fprintf(out, "   ");
4342 		}
4343 
4344 		fprintf(out, "   ");
4345 		for (j = 0; j < 16; j++) {
4346 			if (b + j >= 0 && b + j < len) {
4347 				if (isprint((unsigned char)buf[b + j]))
4348 					fputc((unsigned char)buf[b + j], out);
4349 				else
4350 					fputc('.', out);
4351 			}
4352 			else
4353 				fputc(' ', out);
4354 		}
4355 		fputc('\n', out);
4356 	}
4357 }
4358 
4359 #if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
4360 /* calls dladdr() or dladdr1() on <addr> and <dli>. If dladdr1 is available,
4361  * also returns the symbol size in <size>, otherwise returns 0 there.
4362  */
dladdr_and_size(const void * addr,Dl_info * dli,size_t * size)4363 static int dladdr_and_size(const void *addr, Dl_info *dli, size_t *size)
4364 {
4365 	int ret;
4366 #if (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 3)) // most detailed one
4367 	const ElfW(Sym) *sym;
4368 
4369 	ret = dladdr1(addr, dli, (void **)&sym, RTLD_DL_SYMENT);
4370 	if (ret)
4371 		*size = sym ? sym->st_size : 0;
4372 #else
4373 	ret = dladdr(addr, dli);
4374 	*size = 0;
4375 #endif
4376 	return ret;
4377 }
4378 #endif
4379 
4380 /* Tries to append to buffer <buf> some indications about the symbol at address
4381  * <addr> using the following form:
4382  *   lib:+0xoffset              (unresolvable address from lib's base)
4383  *   main+0xoffset              (unresolvable address from main (+/-))
4384  *   lib:main+0xoffset          (unresolvable lib address from main (+/-))
4385  *   name                       (resolved exact exec address)
4386  *   lib:name                   (resolved exact lib address)
4387  *   name+0xoffset/0xsize       (resolved address within exec symbol)
4388  *   lib:name+0xoffset/0xsize   (resolved address within lib symbol)
4389  *
4390  * The file name (lib or executable) is limited to what lies between the last
4391  * '/' and the first following '.'. An optional prefix <pfx> is prepended before
4392  * the output if not null. The file is not dumped when it's the same as the one
4393  * that contains the "main" symbol, or when __ELF__ && USE_DL are not set.
4394  *
4395  * The symbol's base address is returned, or NULL when unresolved, in order to
4396  * allow the caller to match it against known ones.
4397  */
resolve_sym_name(struct buffer * buf,const char * pfx,void * addr)4398 void *resolve_sym_name(struct buffer *buf, const char *pfx, void *addr)
4399 {
4400 	const struct {
4401 		const void *func;
4402 		const char *name;
4403 	} fcts[] = {
4404 		{ .func = process_stream, .name = "process_stream" },
4405 		{ .func = task_run_applet, .name = "task_run_applet" },
4406 		{ .func = si_cs_io_cb, .name = "si_cs_io_cb" },
4407 		{ .func = conn_fd_handler, .name = "conn_fd_handler" },
4408 		{ .func = dgram_fd_handler, .name = "dgram_fd_handler" },
4409 		{ .func = listener_accept, .name = "listener_accept" },
4410 		{ .func = poller_pipe_io_handler, .name = "poller_pipe_io_handler" },
4411 		{ .func = mworker_accept_wrapper, .name = "mworker_accept_wrapper" },
4412 #ifdef USE_LUA
4413 		{ .func = hlua_process_task, .name = "hlua_process_task" },
4414 #endif
4415 #if defined(USE_OPENSSL) && (HA_OPENSSL_VERSION_NUMBER >= 0x1010000fL) && !defined(OPENSSL_NO_ASYNC)
4416 		{ .func = ssl_async_fd_free, .name = "ssl_async_fd_free" },
4417 		{ .func = ssl_async_fd_handler, .name = "ssl_async_fd_handler" },
4418 #endif
4419 	};
4420 
4421 #if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
4422 	Dl_info dli, dli_main;
4423 	size_t size;
4424 	const char *fname, *p;
4425 #endif
4426 	int i;
4427 
4428 	if (pfx)
4429 		chunk_appendf(buf, "%s", pfx);
4430 
4431 	for (i = 0; i < sizeof(fcts) / sizeof(fcts[0]); i++) {
4432 		if (addr == fcts[i].func) {
4433 			chunk_appendf(buf, "%s", fcts[i].name);
4434 			return addr;
4435 		}
4436 	}
4437 
4438 #if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
4439 	/* Now let's try to be smarter */
4440 	if (!dladdr_and_size(addr, &dli, &size))
4441 		goto unknown;
4442 
4443 	/* 1. prefix the library name if it's not the same object as the one
4444 	 * that contains the main function. The name is picked between last '/'
4445 	 * and first following '.'.
4446 	 */
4447 	if (!dladdr(main, &dli_main))
4448 		dli_main.dli_fbase = NULL;
4449 
4450 	if (dli_main.dli_fbase != dli.dli_fbase) {
4451 		fname = dli.dli_fname;
4452 		p = strrchr(fname, '/');
4453 		if (p++)
4454 			fname = p;
4455 		p = strchr(fname, '.');
4456 		if (!p)
4457 			p = fname + strlen(fname);
4458 
4459 		chunk_appendf(buf, "%.*s:", (int)(long)(p - fname), fname);
4460 	}
4461 
4462 	/* 2. symbol name */
4463 	if (dli.dli_sname) {
4464 		/* known, dump it and return symbol's address (exact or relative) */
4465 		chunk_appendf(buf, "%s", dli.dli_sname);
4466 		if (addr != dli.dli_saddr) {
4467 			chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_saddr));
4468 			if (size)
4469 				chunk_appendf(buf, "/%#lx", (long)size);
4470 		}
4471 		return dli.dli_saddr;
4472 	}
4473 	else if (dli_main.dli_fbase != dli.dli_fbase) {
4474 		/* unresolved symbol from a known library, report relative offset */
4475 		chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_fbase));
4476 		return NULL;
4477 	}
4478 #endif /* __ELF__ && !__linux__ || USE_DL */
4479  unknown:
4480 	/* unresolved symbol from the main file, report relative offset to main */
4481 	if ((void*)addr < (void*)main)
4482 		chunk_appendf(buf, "main-%#lx", (long)((void*)main - addr));
4483 	else
4484 		chunk_appendf(buf, "main+%#lx", (long)(addr - (void*)main));
4485 	return NULL;
4486 }
4487 
4488 /*
4489  * Allocate an array of unsigned int with <nums> as address from <str> string
4490  * made of integer sepereated by dot characters.
4491  *
4492  * First, initializes the value with <sz> as address to 0 and initializes the
4493  * array with <nums> as address to NULL. Then allocates the array with <nums> as
4494  * address updating <sz> pointed value to the size of this array.
4495  *
4496  * Returns 1 if succeeded, 0 if not.
4497  */
parse_dotted_uints(const char * str,unsigned int ** nums,size_t * sz)4498 int parse_dotted_uints(const char *str, unsigned int **nums, size_t *sz)
4499 {
4500 	unsigned int *n;
4501 	const char *s, *end;
4502 
4503 	s = str;
4504 	*sz = 0;
4505 	end = str + strlen(str);
4506 	*nums = n = NULL;
4507 
4508 	while (1) {
4509 		unsigned int r;
4510 
4511 		if (s >= end)
4512 			break;
4513 
4514 		r = read_uint(&s, end);
4515 		/* Expected characters after having read an uint: '\0' or '.',
4516 		 * if '.', must not be terminal.
4517 		 */
4518 		if (*s != '\0'&& (*s++ != '.' || s == end)) {
4519 			free(n);
4520 			return 0;
4521 		}
4522 
4523 		n = my_realloc2(n, (*sz + 1) * sizeof *n);
4524 		if (!n)
4525 			return 0;
4526 
4527 		n[(*sz)++] = r;
4528 	}
4529 	*nums = n;
4530 
4531 	return 1;
4532 }
4533 
4534 
4535 /* returns the number of bytes needed to encode <v> as a varint. An inline
4536  * version exists for use with constants (__varint_bytes()).
4537  */
varint_bytes(uint64_t v)4538 int varint_bytes(uint64_t v)
4539 {
4540 	int len = 1;
4541 
4542 	if (v >= 240) {
4543 		v = (v - 240) >> 4;
4544 		while (1) {
4545 			len++;
4546 			if (v < 128)
4547 				break;
4548 			v = (v - 128) >> 7;
4549 		}
4550 	}
4551 	return len;
4552 }
4553 
4554 /* Random number generator state, see below */
4555 static uint64_t ha_random_state[2] ALIGNED(2*sizeof(uint64_t));
4556 
4557 /* This is a thread-safe implementation of xoroshiro128** described below:
4558  *     http://prng.di.unimi.it/
4559  * It features a 2^128 long sequence, returns 64 high-quality bits on each call,
4560  * supports fast jumps and passes all common quality tests. It is thread-safe,
4561  * uses a double-cas on 64-bit architectures supporting it, and falls back to a
4562  * local lock on other ones.
4563  */
ha_random64()4564 uint64_t ha_random64()
4565 {
4566 	uint64_t result;
4567 	uint64_t old[2] ALIGNED(2*sizeof(uint64_t));
4568 	uint64_t new[2] ALIGNED(2*sizeof(uint64_t));
4569 
4570 #if defined(USE_THREAD) && (!defined(HA_CAS_IS_8B) || !defined(HA_HAVE_CAS_DW))
4571 	static HA_SPINLOCK_T rand_lock;
4572 
4573 	HA_SPIN_LOCK(OTHER_LOCK, &rand_lock);
4574 #endif
4575 
4576 	old[0] = ha_random_state[0];
4577 	old[1] = ha_random_state[1];
4578 
4579 #if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
4580 	do {
4581 #endif
4582 		result = rotl64(old[0] * 5, 7) * 9;
4583 		new[1] = old[0] ^ old[1];
4584 		new[0] = rotl64(old[0], 24) ^ new[1] ^ (new[1] << 16); // a, b
4585 		new[1] = rotl64(new[1], 37); // c
4586 
4587 #if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
4588 	} while (unlikely(!_HA_ATOMIC_DWCAS(ha_random_state, old, new)));
4589 #else
4590 	ha_random_state[0] = new[0];
4591 	ha_random_state[1] = new[1];
4592 #if defined(USE_THREAD)
4593 	HA_SPIN_UNLOCK(OTHER_LOCK, &rand_lock);
4594 #endif
4595 #endif
4596 	return result;
4597 }
4598 
4599 /* seeds the random state using up to <len> bytes from <seed>, starting with
4600  * the first non-zero byte.
4601  */
ha_random_seed(const unsigned char * seed,size_t len)4602 void ha_random_seed(const unsigned char *seed, size_t len)
4603 {
4604 	size_t pos;
4605 
4606 	/* the seed must not be all zeroes, so we pre-fill it with alternating
4607 	 * bits and overwrite part of them with the block starting at the first
4608 	 * non-zero byte from the seed.
4609 	 */
4610 	memset(ha_random_state, 0x55, sizeof(ha_random_state));
4611 
4612 	for (pos = 0; pos < len; pos++)
4613 		if (seed[pos] != 0)
4614 			break;
4615 
4616 	if (pos == len)
4617 		return;
4618 
4619 	seed += pos;
4620 	len -= pos;
4621 
4622 	if (len > sizeof(ha_random_state))
4623 		len = sizeof(ha_random_state);
4624 
4625 	memcpy(ha_random_state, seed, len);
4626 }
4627 
4628 /* This causes a jump to (dist * 2^96) places in the pseudo-random sequence,
4629  * and is equivalent to calling ha_random64() as many times. It is used to
4630  * provide non-overlapping sequences of 2^96 numbers (~7*10^28) to up to 2^32
4631  * different generators (i.e. different processes after a fork). The <dist>
4632  * argument is the distance to jump to and is used in a loop so it rather not
4633  * be too large if the processing time is a concern.
4634  *
4635  * BEWARE: this function is NOT thread-safe and must not be called during
4636  * concurrent accesses to ha_random64().
4637  */
ha_random_jump96(uint32_t dist)4638 void ha_random_jump96(uint32_t dist)
4639 {
4640 	while (dist--) {
4641 		uint64_t s0 = 0;
4642 		uint64_t s1 = 0;
4643 		int b;
4644 
4645 		for (b = 0; b < 64; b++) {
4646 			if ((0xd2a98b26625eee7bULL >> b) & 1) {
4647 				s0 ^= ha_random_state[0];
4648 				s1 ^= ha_random_state[1];
4649 			}
4650 			ha_random64();
4651 		}
4652 
4653 		for (b = 0; b < 64; b++) {
4654 			if ((0xdddf9b1090aa7ac1ULL >> b) & 1) {
4655 				s0 ^= ha_random_state[0];
4656 				s1 ^= ha_random_state[1];
4657 			}
4658 			ha_random64();
4659 		}
4660 		ha_random_state[0] = s0;
4661 		ha_random_state[1] = s1;
4662 	}
4663 }
4664 
4665 /*
4666  * Local variables:
4667  *  c-indent-level: 8
4668  *  c-basic-offset: 8
4669  * End:
4670  */
4671