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