blob: 54cf2d644b64a3afff8451cace3a62c03790d0be [file] [log] [blame]
Willy Tarreaubaaee002006-06-26 02:48:02 +02001/*
2 * General purpose functions.
3 *
Willy Tarreau348238b2010-01-18 15:05:57 +01004 * Copyright 2000-2010 Willy Tarreau <w@1wt.eu>
Willy Tarreaubaaee002006-06-26 02:48:02 +02005 *
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
Baruch Siache1651b22020-07-24 07:52:20 +030013#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +010014#define _GNU_SOURCE
15#include <dlfcn.h>
16#include <link.h>
17#endif
18
Willy Tarreau2e74c3f2007-12-02 18:45:09 +010019#include <ctype.h>
Willy Tarreau16e01562016-08-09 16:46:18 +020020#include <errno.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020021#include <netdb.h>
Willy Tarreau9a7bea52012-04-27 11:16:50 +020022#include <stdarg.h>
Willy Tarreaudd2f85e2012-09-02 22:34:23 +020023#include <stdio.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020024#include <stdlib.h>
25#include <string.h>
Thierry Fournier93127942016-01-20 18:49:45 +010026#include <time.h>
Willy Tarreau16e01562016-08-09 16:46:18 +020027#include <unistd.h>
Willy Tarreau127f9662007-12-06 00:53:51 +010028#include <sys/socket.h>
Willy Tarreau37101052019-05-20 16:48:20 +020029#include <sys/stat.h>
30#include <sys/types.h>
Willy Tarreau127f9662007-12-06 00:53:51 +010031#include <sys/un.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020032#include <netinet/in.h>
33#include <arpa/inet.h>
34
Willy Tarreau30053062020-08-20 16:39:14 +020035#if (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 16))
36#include <sys/auxv.h>
37#endif
38
Willy Tarreau48fbcae2020-06-03 18:09:46 +020039#include <import/eb32sctree.h>
Willy Tarreaub2551052020-06-09 09:07:15 +020040#include <import/eb32tree.h>
Willy Tarreau48fbcae2020-06-03 18:09:46 +020041
Willy Tarreau4c7e4b72020-05-27 12:58:42 +020042#include <haproxy/api.h>
Willy Tarreauc13ed532020-06-02 10:22:45 +020043#include <haproxy/chunk.h>
Willy Tarreau7c18b542020-06-11 09:23:02 +020044#include <haproxy/dgram.h>
Willy Tarreaueb92deb2020-06-04 10:53:16 +020045#include <haproxy/dns.h>
Willy Tarreauf268ee82020-06-04 17:05:57 +020046#include <haproxy/global.h>
Willy Tarreau86416052020-06-04 09:20:54 +020047#include <haproxy/hlua.h>
Willy Tarreau213e9902020-06-04 14:58:24 +020048#include <haproxy/listener.h>
Willy Tarreau7a00efb2020-06-02 17:02:59 +020049#include <haproxy/namespace.h>
Willy Tarreau5fc93282020-09-16 18:25:03 +020050#include <haproxy/protocol.h>
Willy Tarreau209108d2020-06-04 20:30:20 +020051#include <haproxy/ssl_sock.h>
Willy Tarreau5e539c92020-06-04 20:45:39 +020052#include <haproxy/stream_interface.h>
Willy Tarreaucea0e1b2020-06-04 17:25:40 +020053#include <haproxy/task.h>
Willy Tarreau48fbcae2020-06-03 18:09:46 +020054#include <haproxy/tools.h>
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +010055
Thierry Fournier93127942016-01-20 18:49:45 +010056/* This macro returns false if the test __x is false. Many
57 * of the following parsing function must be abort the processing
58 * if it returns 0, so this macro is useful for writing light code.
59 */
60#define RET0_UNLESS(__x) do { if (!(__x)) return 0; } while (0)
61
Willy Tarreau56adcf22012-12-23 18:00:29 +010062/* enough to store NB_ITOA_STR integers of :
Willy Tarreau72d759c2007-10-25 12:14:10 +020063 * 2^64-1 = 18446744073709551615 or
64 * -2^63 = -9223372036854775808
Willy Tarreaue7239b52009-03-29 13:41:58 +020065 *
66 * The HTML version needs room for adding the 25 characters
67 * '<span class="rls"></span>' around digits at positions 3N+1 in order
68 * to add spacing at up to 6 positions : 18 446 744 073 709 551 615
Willy Tarreau72d759c2007-10-25 12:14:10 +020069 */
Christopher Faulet99bca652017-11-14 16:47:26 +010070THREAD_LOCAL char itoa_str[NB_ITOA_STR][171];
71THREAD_LOCAL int itoa_idx = 0; /* index of next itoa_str to use */
Willy Tarreaubaaee002006-06-26 02:48:02 +020072
Willy Tarreau588297f2014-06-16 15:16:40 +020073/* sometimes we'll need to quote strings (eg: in stats), and we don't expect
74 * to quote strings larger than a max configuration line.
75 */
Christopher Faulet99bca652017-11-14 16:47:26 +010076THREAD_LOCAL char quoted_str[NB_QSTR][QSTR_SIZE + 1];
77THREAD_LOCAL int quoted_idx = 0;
Willy Tarreau588297f2014-06-16 15:16:40 +020078
Willy Tarreaubaaee002006-06-26 02:48:02 +020079/*
William Lallemande7340ec2012-01-24 11:15:39 +010080 * unsigned long long ASCII representation
81 *
82 * return the last char '\0' or NULL if no enough
83 * space in dst
84 */
85char *ulltoa(unsigned long long n, char *dst, size_t size)
86{
87 int i = 0;
88 char *res;
89
90 switch(n) {
91 case 1ULL ... 9ULL:
92 i = 0;
93 break;
94
95 case 10ULL ... 99ULL:
96 i = 1;
97 break;
98
99 case 100ULL ... 999ULL:
100 i = 2;
101 break;
102
103 case 1000ULL ... 9999ULL:
104 i = 3;
105 break;
106
107 case 10000ULL ... 99999ULL:
108 i = 4;
109 break;
110
111 case 100000ULL ... 999999ULL:
112 i = 5;
113 break;
114
115 case 1000000ULL ... 9999999ULL:
116 i = 6;
117 break;
118
119 case 10000000ULL ... 99999999ULL:
120 i = 7;
121 break;
122
123 case 100000000ULL ... 999999999ULL:
124 i = 8;
125 break;
126
127 case 1000000000ULL ... 9999999999ULL:
128 i = 9;
129 break;
130
131 case 10000000000ULL ... 99999999999ULL:
132 i = 10;
133 break;
134
135 case 100000000000ULL ... 999999999999ULL:
136 i = 11;
137 break;
138
139 case 1000000000000ULL ... 9999999999999ULL:
140 i = 12;
141 break;
142
143 case 10000000000000ULL ... 99999999999999ULL:
144 i = 13;
145 break;
146
147 case 100000000000000ULL ... 999999999999999ULL:
148 i = 14;
149 break;
150
151 case 1000000000000000ULL ... 9999999999999999ULL:
152 i = 15;
153 break;
154
155 case 10000000000000000ULL ... 99999999999999999ULL:
156 i = 16;
157 break;
158
159 case 100000000000000000ULL ... 999999999999999999ULL:
160 i = 17;
161 break;
162
163 case 1000000000000000000ULL ... 9999999999999999999ULL:
164 i = 18;
165 break;
166
167 case 10000000000000000000ULL ... ULLONG_MAX:
168 i = 19;
169 break;
170 }
171 if (i + 2 > size) // (i + 1) + '\0'
172 return NULL; // too long
173 res = dst + i + 1;
174 *res = '\0';
175 for (; i >= 0; i--) {
176 dst[i] = n % 10ULL + '0';
177 n /= 10ULL;
178 }
179 return res;
180}
181
182/*
183 * unsigned long ASCII representation
184 *
185 * return the last char '\0' or NULL if no enough
186 * space in dst
187 */
188char *ultoa_o(unsigned long n, char *dst, size_t size)
189{
190 int i = 0;
191 char *res;
192
193 switch (n) {
194 case 0U ... 9UL:
195 i = 0;
196 break;
197
198 case 10U ... 99UL:
199 i = 1;
200 break;
201
202 case 100U ... 999UL:
203 i = 2;
204 break;
205
206 case 1000U ... 9999UL:
207 i = 3;
208 break;
209
210 case 10000U ... 99999UL:
211 i = 4;
212 break;
213
214 case 100000U ... 999999UL:
215 i = 5;
216 break;
217
218 case 1000000U ... 9999999UL:
219 i = 6;
220 break;
221
222 case 10000000U ... 99999999UL:
223 i = 7;
224 break;
225
226 case 100000000U ... 999999999UL:
227 i = 8;
228 break;
229#if __WORDSIZE == 32
230
231 case 1000000000ULL ... ULONG_MAX:
232 i = 9;
233 break;
234
235#elif __WORDSIZE == 64
236
237 case 1000000000ULL ... 9999999999UL:
238 i = 9;
239 break;
240
241 case 10000000000ULL ... 99999999999UL:
242 i = 10;
243 break;
244
245 case 100000000000ULL ... 999999999999UL:
246 i = 11;
247 break;
248
249 case 1000000000000ULL ... 9999999999999UL:
250 i = 12;
251 break;
252
253 case 10000000000000ULL ... 99999999999999UL:
254 i = 13;
255 break;
256
257 case 100000000000000ULL ... 999999999999999UL:
258 i = 14;
259 break;
260
261 case 1000000000000000ULL ... 9999999999999999UL:
262 i = 15;
263 break;
264
265 case 10000000000000000ULL ... 99999999999999999UL:
266 i = 16;
267 break;
268
269 case 100000000000000000ULL ... 999999999999999999UL:
270 i = 17;
271 break;
272
273 case 1000000000000000000ULL ... 9999999999999999999UL:
274 i = 18;
275 break;
276
277 case 10000000000000000000ULL ... ULONG_MAX:
278 i = 19;
279 break;
280
281#endif
282 }
283 if (i + 2 > size) // (i + 1) + '\0'
284 return NULL; // too long
285 res = dst + i + 1;
286 *res = '\0';
287 for (; i >= 0; i--) {
288 dst[i] = n % 10U + '0';
289 n /= 10U;
290 }
291 return res;
292}
293
294/*
295 * signed long ASCII representation
296 *
297 * return the last char '\0' or NULL if no enough
298 * space in dst
299 */
300char *ltoa_o(long int n, char *dst, size_t size)
301{
302 char *pos = dst;
303
304 if (n < 0) {
305 if (size < 3)
306 return NULL; // min size is '-' + digit + '\0' but another test in ultoa
307 *pos = '-';
308 pos++;
309 dst = ultoa_o(-n, pos, size - 1);
310 } else {
311 dst = ultoa_o(n, dst, size);
312 }
313 return dst;
314}
315
316/*
317 * signed long long ASCII representation
318 *
319 * return the last char '\0' or NULL if no enough
320 * space in dst
321 */
322char *lltoa(long long n, char *dst, size_t size)
323{
324 char *pos = dst;
325
326 if (n < 0) {
327 if (size < 3)
328 return NULL; // min size is '-' + digit + '\0' but another test in ulltoa
329 *pos = '-';
330 pos++;
331 dst = ulltoa(-n, pos, size - 1);
332 } else {
333 dst = ulltoa(n, dst, size);
334 }
335 return dst;
336}
337
338/*
339 * write a ascii representation of a unsigned into dst,
340 * return a pointer to the last character
341 * Pad the ascii representation with '0', using size.
342 */
343char *utoa_pad(unsigned int n, char *dst, size_t size)
344{
345 int i = 0;
346 char *ret;
347
348 switch(n) {
349 case 0U ... 9U:
350 i = 0;
351 break;
352
353 case 10U ... 99U:
354 i = 1;
355 break;
356
357 case 100U ... 999U:
358 i = 2;
359 break;
360
361 case 1000U ... 9999U:
362 i = 3;
363 break;
364
365 case 10000U ... 99999U:
366 i = 4;
367 break;
368
369 case 100000U ... 999999U:
370 i = 5;
371 break;
372
373 case 1000000U ... 9999999U:
374 i = 6;
375 break;
376
377 case 10000000U ... 99999999U:
378 i = 7;
379 break;
380
381 case 100000000U ... 999999999U:
382 i = 8;
383 break;
384
385 case 1000000000U ... 4294967295U:
386 i = 9;
387 break;
388 }
389 if (i + 2 > size) // (i + 1) + '\0'
390 return NULL; // too long
391 if (i < size)
392 i = size - 2; // padding - '\0'
393
394 ret = dst + i + 1;
395 *ret = '\0';
396 for (; i >= 0; i--) {
397 dst[i] = n % 10U + '0';
398 n /= 10U;
399 }
400 return ret;
401}
402
403/*
Willy Tarreaubaaee002006-06-26 02:48:02 +0200404 * copies at most <size-1> chars from <src> to <dst>. Last char is always
405 * set to 0, unless <size> is 0. The number of chars copied is returned
406 * (excluding the terminating zero).
407 * This code has been optimized for size and speed : on x86, it's 45 bytes
408 * long, uses only registers, and consumes only 4 cycles per char.
409 */
410int strlcpy2(char *dst, const char *src, int size)
411{
412 char *orig = dst;
413 if (size) {
414 while (--size && (*dst = *src)) {
415 src++; dst++;
416 }
417 *dst = 0;
418 }
419 return dst - orig;
420}
421
422/*
Willy Tarreau72d759c2007-10-25 12:14:10 +0200423 * This function simply returns a locally allocated string containing
Willy Tarreaubaaee002006-06-26 02:48:02 +0200424 * the ascii representation for number 'n' in decimal.
425 */
Emeric Brun3a7fce52010-01-04 14:54:38 +0100426char *ultoa_r(unsigned long n, char *buffer, int size)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200427{
428 char *pos;
429
Willy Tarreau72d759c2007-10-25 12:14:10 +0200430 pos = buffer + size - 1;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200431 *pos-- = '\0';
432
433 do {
434 *pos-- = '0' + n % 10;
435 n /= 10;
Willy Tarreau72d759c2007-10-25 12:14:10 +0200436 } while (n && pos >= buffer);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200437 return pos + 1;
438}
439
Willy Tarreau91092e52007-10-25 16:58:42 +0200440/*
Willy Tarreaue7239b52009-03-29 13:41:58 +0200441 * This function simply returns a locally allocated string containing
Thierry FOURNIER763a5d82015-07-06 23:09:52 +0200442 * the ascii representation for number 'n' in decimal.
443 */
444char *lltoa_r(long long int in, char *buffer, int size)
445{
446 char *pos;
447 int neg = 0;
448 unsigned long long int n;
449
450 pos = buffer + size - 1;
451 *pos-- = '\0';
452
453 if (in < 0) {
454 neg = 1;
455 n = -in;
456 }
457 else
458 n = in;
459
460 do {
461 *pos-- = '0' + n % 10;
462 n /= 10;
463 } while (n && pos >= buffer);
464 if (neg && pos > buffer)
465 *pos-- = '-';
466 return pos + 1;
467}
468
469/*
470 * This function simply returns a locally allocated string containing
Thierry FOURNIER1480bd82015-06-06 19:14:59 +0200471 * the ascii representation for signed number 'n' in decimal.
472 */
473char *sltoa_r(long n, char *buffer, int size)
474{
475 char *pos;
476
477 if (n >= 0)
478 return ultoa_r(n, buffer, size);
479
480 pos = ultoa_r(-n, buffer + 1, size - 1) - 1;
481 *pos = '-';
482 return pos;
483}
484
485/*
486 * This function simply returns a locally allocated string containing
Willy Tarreaue7239b52009-03-29 13:41:58 +0200487 * the ascii representation for number 'n' in decimal, formatted for
488 * HTML output with tags to create visual grouping by 3 digits. The
489 * output needs to support at least 171 characters.
490 */
491const char *ulltoh_r(unsigned long long n, char *buffer, int size)
492{
493 char *start;
494 int digit = 0;
495
496 start = buffer + size;
497 *--start = '\0';
498
499 do {
500 if (digit == 3 && start >= buffer + 7)
501 memcpy(start -= 7, "</span>", 7);
502
503 if (start >= buffer + 1) {
504 *--start = '0' + n % 10;
505 n /= 10;
506 }
507
508 if (digit == 3 && start >= buffer + 18)
509 memcpy(start -= 18, "<span class=\"rls\">", 18);
510
511 if (digit++ == 3)
512 digit = 1;
513 } while (n && start > buffer);
514 return start;
515}
516
517/*
Willy Tarreau91092e52007-10-25 16:58:42 +0200518 * This function simply returns a locally allocated string containing the ascii
519 * representation for number 'n' in decimal, unless n is 0 in which case it
520 * returns the alternate string (or an empty string if the alternate string is
521 * NULL). It use is intended for limits reported in reports, where it's
522 * desirable not to display anything if there is no limit. Warning! it shares
523 * the same vector as ultoa_r().
524 */
525const char *limit_r(unsigned long n, char *buffer, int size, const char *alt)
526{
527 return (n) ? ultoa_r(n, buffer, size) : (alt ? alt : "");
528}
529
Willy Tarreau588297f2014-06-16 15:16:40 +0200530/* returns a locally allocated string containing the quoted encoding of the
531 * input string. The output may be truncated to QSTR_SIZE chars, but it is
532 * guaranteed that the string will always be properly terminated. Quotes are
533 * encoded by doubling them as is commonly done in CSV files. QSTR_SIZE must
534 * always be at least 4 chars.
535 */
536const char *qstr(const char *str)
537{
538 char *ret = quoted_str[quoted_idx];
539 char *p, *end;
540
541 if (++quoted_idx >= NB_QSTR)
542 quoted_idx = 0;
543
544 p = ret;
545 end = ret + QSTR_SIZE;
546
547 *p++ = '"';
548
549 /* always keep 3 chars to support passing "" and the ending " */
550 while (*str && p < end - 3) {
551 if (*str == '"') {
552 *p++ = '"';
553 *p++ = '"';
554 }
555 else
556 *p++ = *str;
557 str++;
558 }
559 *p++ = '"';
560 return ret;
561}
562
Robert Tsai81ae1952007-12-05 10:47:29 +0100563/*
Willy Tarreaubaaee002006-06-26 02:48:02 +0200564 * Returns non-zero if character <s> is a hex digit (0-9, a-f, A-F), else zero.
565 *
566 * It looks like this one would be a good candidate for inlining, but this is
567 * not interesting because it around 35 bytes long and often called multiple
568 * times within the same function.
569 */
570int ishex(char s)
571{
572 s -= '0';
573 if ((unsigned char)s <= 9)
574 return 1;
575 s -= 'A' - '0';
576 if ((unsigned char)s <= 5)
577 return 1;
578 s -= 'a' - 'A';
579 if ((unsigned char)s <= 5)
580 return 1;
581 return 0;
582}
583
Willy Tarreau3ca1a882015-01-15 18:43:49 +0100584/* rounds <i> down to the closest value having max 2 digits */
585unsigned int round_2dig(unsigned int i)
586{
587 unsigned int mul = 1;
588
589 while (i >= 100) {
590 i /= 10;
591 mul *= 10;
592 }
593 return i * mul;
594}
595
Willy Tarreau2e74c3f2007-12-02 18:45:09 +0100596/*
597 * Checks <name> for invalid characters. Valid chars are [A-Za-z0-9_:.-]. If an
598 * invalid character is found, a pointer to it is returned. If everything is
599 * fine, NULL is returned.
600 */
601const char *invalid_char(const char *name)
602{
603 if (!*name)
604 return name;
605
606 while (*name) {
Willy Tarreau90807112020-02-25 08:16:33 +0100607 if (!isalnum((unsigned char)*name) && *name != '.' && *name != ':' &&
Willy Tarreau2e74c3f2007-12-02 18:45:09 +0100608 *name != '_' && *name != '-')
609 return name;
610 name++;
611 }
612 return NULL;
613}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200614
615/*
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200616 * Checks <name> for invalid characters. Valid chars are [_.-] and those
617 * accepted by <f> function.
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200618 * If an invalid character is found, a pointer to it is returned.
619 * If everything is fine, NULL is returned.
620 */
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200621static inline const char *__invalid_char(const char *name, int (*f)(int)) {
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200622
623 if (!*name)
624 return name;
625
626 while (*name) {
Willy Tarreau90807112020-02-25 08:16:33 +0100627 if (!f((unsigned char)*name) && *name != '.' &&
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200628 *name != '_' && *name != '-')
629 return name;
630
631 name++;
632 }
633
634 return NULL;
635}
636
637/*
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200638 * Checks <name> for invalid characters. Valid chars are [A-Za-z0-9_.-].
639 * If an invalid character is found, a pointer to it is returned.
640 * If everything is fine, NULL is returned.
641 */
642const char *invalid_domainchar(const char *name) {
643 return __invalid_char(name, isalnum);
644}
645
646/*
647 * Checks <name> for invalid characters. Valid chars are [A-Za-z_.-].
648 * If an invalid character is found, a pointer to it is returned.
649 * If everything is fine, NULL is returned.
650 */
651const char *invalid_prefix_char(const char *name) {
Thierry Fournierf7b7c3e2018-03-26 11:54:39 +0200652 return __invalid_char(name, isalnum);
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200653}
654
655/*
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100656 * converts <str> to a struct sockaddr_storage* provided by the caller. The
Willy Tarreau24709282013-03-10 21:32:12 +0100657 * caller must have zeroed <sa> first, and may have set sa->ss_family to force
658 * parse a specific address format. If the ss_family is 0 or AF_UNSPEC, then
659 * the function tries to guess the address family from the syntax. If the
660 * family is forced and the format doesn't match, an error is returned. The
Willy Tarreaufab5a432011-03-04 15:31:53 +0100661 * string is assumed to contain only an address, no port. The address can be a
662 * dotted IPv4 address, an IPv6 address, a host name, or empty or "*" to
663 * indicate INADDR_ANY. NULL is returned if the host part cannot be resolved.
664 * The return address will only have the address family and the address set,
665 * all other fields remain zero. The string is not supposed to be modified.
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100666 * The IPv6 '::' address is IN6ADDR_ANY. If <resolve> is non-zero, the hostname
667 * is resolved, otherwise only IP addresses are resolved, and anything else
Willy Tarreauecde7df2016-11-02 22:37:03 +0100668 * returns NULL. If the address contains a port, this one is preserved.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200669 */
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100670struct sockaddr_storage *str2ip2(const char *str, struct sockaddr_storage *sa, int resolve)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200671{
Willy Tarreaufab5a432011-03-04 15:31:53 +0100672 struct hostent *he;
mildisff5d5102015-10-26 18:50:08 +0100673 /* max IPv6 length, including brackets and terminating NULL */
674 char tmpip[48];
Willy Tarreauecde7df2016-11-02 22:37:03 +0100675 int port = get_host_port(sa);
mildisff5d5102015-10-26 18:50:08 +0100676
677 /* check IPv6 with square brackets */
678 if (str[0] == '[') {
679 size_t iplength = strlen(str);
680
681 if (iplength < 4) {
682 /* minimal size is 4 when using brackets "[::]" */
683 goto fail;
684 }
685 else if (iplength >= sizeof(tmpip)) {
686 /* IPv6 literal can not be larger than tmpip */
687 goto fail;
688 }
689 else {
690 if (str[iplength - 1] != ']') {
691 /* if address started with bracket, it should end with bracket */
692 goto fail;
693 }
694 else {
695 memcpy(tmpip, str + 1, iplength - 2);
696 tmpip[iplength - 2] = '\0';
697 str = tmpip;
698 }
699 }
700 }
Willy Tarreaufab5a432011-03-04 15:31:53 +0100701
Willy Tarreaufab5a432011-03-04 15:31:53 +0100702 /* Any IPv6 address */
703 if (str[0] == ':' && str[1] == ':' && !str[2]) {
Willy Tarreau24709282013-03-10 21:32:12 +0100704 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
705 sa->ss_family = AF_INET6;
706 else if (sa->ss_family != AF_INET6)
707 goto fail;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100708 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100709 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100710 }
711
Willy Tarreau24709282013-03-10 21:32:12 +0100712 /* Any address for the family, defaults to IPv4 */
Willy Tarreaufab5a432011-03-04 15:31:53 +0100713 if (!str[0] || (str[0] == '*' && !str[1])) {
Willy Tarreau24709282013-03-10 21:32:12 +0100714 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
715 sa->ss_family = AF_INET;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100716 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100717 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100718 }
719
720 /* check for IPv6 first */
Willy Tarreau24709282013-03-10 21:32:12 +0100721 if ((!sa->ss_family || sa->ss_family == AF_UNSPEC || sa->ss_family == AF_INET6) &&
722 inet_pton(AF_INET6, str, &((struct sockaddr_in6 *)sa)->sin6_addr)) {
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100723 sa->ss_family = AF_INET6;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100724 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100725 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100726 }
727
728 /* then check for IPv4 */
Willy Tarreau24709282013-03-10 21:32:12 +0100729 if ((!sa->ss_family || sa->ss_family == AF_UNSPEC || sa->ss_family == AF_INET) &&
730 inet_pton(AF_INET, str, &((struct sockaddr_in *)sa)->sin_addr)) {
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100731 sa->ss_family = AF_INET;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100732 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100733 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100734 }
735
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100736 if (!resolve)
737 return NULL;
738
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200739 if (!dns_hostname_validation(str, NULL))
740 return NULL;
741
David du Colombierd5f43282011-03-17 10:40:16 +0100742#ifdef USE_GETADDRINFO
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200743 if (global.tune.options & GTUNE_USE_GAI) {
David du Colombierd5f43282011-03-17 10:40:16 +0100744 struct addrinfo hints, *result;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100745 int success = 0;
David du Colombierd5f43282011-03-17 10:40:16 +0100746
747 memset(&result, 0, sizeof(result));
748 memset(&hints, 0, sizeof(hints));
Willy Tarreau24709282013-03-10 21:32:12 +0100749 hints.ai_family = sa->ss_family ? sa->ss_family : AF_UNSPEC;
David du Colombierd5f43282011-03-17 10:40:16 +0100750 hints.ai_socktype = SOCK_DGRAM;
Dmitry Sivachenkoeab7f392015-10-02 01:01:58 +0200751 hints.ai_flags = 0;
David du Colombierd5f43282011-03-17 10:40:16 +0100752 hints.ai_protocol = 0;
753
754 if (getaddrinfo(str, NULL, &hints, &result) == 0) {
Willy Tarreau24709282013-03-10 21:32:12 +0100755 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
756 sa->ss_family = result->ai_family;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100757 else if (sa->ss_family != result->ai_family) {
758 freeaddrinfo(result);
Willy Tarreau24709282013-03-10 21:32:12 +0100759 goto fail;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100760 }
Willy Tarreau24709282013-03-10 21:32:12 +0100761
David du Colombierd5f43282011-03-17 10:40:16 +0100762 switch (result->ai_family) {
763 case AF_INET:
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100764 memcpy((struct sockaddr_in *)sa, result->ai_addr, result->ai_addrlen);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100765 set_host_port(sa, port);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100766 success = 1;
767 break;
David du Colombierd5f43282011-03-17 10:40:16 +0100768 case AF_INET6:
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100769 memcpy((struct sockaddr_in6 *)sa, result->ai_addr, result->ai_addrlen);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100770 set_host_port(sa, port);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100771 success = 1;
772 break;
David du Colombierd5f43282011-03-17 10:40:16 +0100773 }
774 }
775
Sean Carey58ea0392013-02-15 23:39:18 +0100776 if (result)
777 freeaddrinfo(result);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100778
779 if (success)
780 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100781 }
David du Colombierd5f43282011-03-17 10:40:16 +0100782#endif
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200783 /* try to resolve an IPv4/IPv6 hostname */
784 he = gethostbyname(str);
785 if (he) {
786 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
787 sa->ss_family = he->h_addrtype;
788 else if (sa->ss_family != he->h_addrtype)
789 goto fail;
790
791 switch (sa->ss_family) {
792 case AF_INET:
793 ((struct sockaddr_in *)sa)->sin_addr = *(struct in_addr *) *(he->h_addr_list);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100794 set_host_port(sa, port);
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200795 return sa;
796 case AF_INET6:
797 ((struct sockaddr_in6 *)sa)->sin6_addr = *(struct in6_addr *) *(he->h_addr_list);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100798 set_host_port(sa, port);
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200799 return sa;
800 }
801 }
802
David du Colombierd5f43282011-03-17 10:40:16 +0100803 /* unsupported address family */
Willy Tarreau24709282013-03-10 21:32:12 +0100804 fail:
Willy Tarreaufab5a432011-03-04 15:31:53 +0100805 return NULL;
806}
807
808/*
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100809 * Converts <str> to a locally allocated struct sockaddr_storage *, and a port
810 * range or offset consisting in two integers that the caller will have to
811 * check to find the relevant input format. The following format are supported :
812 *
813 * String format | address | port | low | high
814 * addr | <addr> | 0 | 0 | 0
815 * addr: | <addr> | 0 | 0 | 0
816 * addr:port | <addr> | <port> | <port> | <port>
817 * addr:pl-ph | <addr> | <pl> | <pl> | <ph>
818 * addr:+port | <addr> | <port> | 0 | <port>
819 * addr:-port | <addr> |-<port> | <port> | 0
820 *
821 * The detection of a port range or increment by the caller is made by
822 * comparing <low> and <high>. If both are equal, then port 0 means no port
823 * was specified. The caller may pass NULL for <low> and <high> if it is not
824 * interested in retrieving port ranges.
825 *
826 * Note that <addr> above may also be :
827 * - empty ("") => family will be AF_INET and address will be INADDR_ANY
828 * - "*" => family will be AF_INET and address will be INADDR_ANY
829 * - "::" => family will be AF_INET6 and address will be IN6ADDR_ANY
830 * - a host name => family and address will depend on host name resolving.
831 *
Willy Tarreau24709282013-03-10 21:32:12 +0100832 * A prefix may be passed in before the address above to force the family :
833 * - "ipv4@" => force address to resolve as IPv4 and fail if not possible.
834 * - "ipv6@" => force address to resolve as IPv6 and fail if not possible.
835 * - "unix@" => force address to be a path to a UNIX socket even if the
836 * path does not start with a '/'
Willy Tarreauccfccef2014-05-10 01:49:15 +0200837 * - 'abns@' -> force address to belong to the abstract namespace (Linux
838 * only). These sockets are just like Unix sockets but without
839 * the need for an underlying file system. The address is a
840 * string. Technically it's like a Unix socket with a zero in
841 * the first byte of the address.
Willy Tarreau40aa0702013-03-10 23:51:38 +0100842 * - "fd@" => an integer must follow, and is a file descriptor number.
Willy Tarreau24709282013-03-10 21:32:12 +0100843 *
mildisff5d5102015-10-26 18:50:08 +0100844 * IPv6 addresses can be declared with or without square brackets. When using
845 * square brackets for IPv6 addresses, the port separator (colon) is optional.
846 * If not using square brackets, and in order to avoid any ambiguity with
847 * IPv6 addresses, the last colon ':' is mandatory even when no port is specified.
848 * NULL is returned if the address cannot be parsed. The <low> and <high> ports
849 * are always initialized if non-null, even for non-IP families.
Willy Tarreaud393a622013-03-04 18:22:00 +0100850 *
851 * If <pfx> is non-null, it is used as a string prefix before any path-based
852 * address (typically the path to a unix socket).
Willy Tarreau40aa0702013-03-10 23:51:38 +0100853 *
Willy Tarreau72b8c1f2015-09-08 15:50:19 +0200854 * if <fqdn> is non-null, it will be filled with :
855 * - a pointer to the FQDN of the server name to resolve if there's one, and
856 * that the caller will have to free(),
857 * - NULL if there was an explicit address that doesn't require resolution.
858 *
Willy Tarreaucd3a55912020-09-04 15:30:46 +0200859 * Hostnames are only resolved if <opts> has PA_O_RESOLVE. Otherwise <fqdn> is
860 * still honored so it is possible for the caller to know whether a resolution
861 * failed by clearing this flag and checking if <fqdn> was filled, indicating
862 * the need for a resolution.
Thierry FOURNIER7fe3be72015-09-26 20:03:36 +0200863 *
Willy Tarreau40aa0702013-03-10 23:51:38 +0100864 * When a file descriptor is passed, its value is put into the s_addr part of
Willy Tarreaua5b325f2020-09-04 16:44:20 +0200865 * the address when cast to sockaddr_in and the address family is
866 * AF_CUST_EXISTING_FD.
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200867 *
Willy Tarreau5fc93282020-09-16 18:25:03 +0200868 * The matching protocol will be set into <proto> if non-null.
869 *
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200870 * Any known file descriptor is also assigned to <fd> if non-null, otherwise it
871 * is forced to -1.
Willy Tarreaufab5a432011-03-04 15:31:53 +0100872 */
Willy Tarreau5fc93282020-09-16 18:25:03 +0200873struct sockaddr_storage *str2sa_range(const char *str, int *port, int *low, int *high, int *fd,
874 struct protocol **proto, char **err,
875 const char *pfx, char **fqdn, unsigned int opts)
Willy Tarreaufab5a432011-03-04 15:31:53 +0100876{
Christopher Faulet1bc04c72017-10-29 20:14:08 +0100877 static THREAD_LOCAL struct sockaddr_storage ss;
David du Colombier6f5ccb12011-03-10 22:26:24 +0100878 struct sockaddr_storage *ret = NULL;
Willy Tarreau5fc93282020-09-16 18:25:03 +0200879 struct protocol *new_proto = NULL;
Willy Tarreau24709282013-03-10 21:32:12 +0100880 char *back, *str2;
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100881 char *port1, *port2;
882 int portl, porth, porta;
Willy Tarreauccfccef2014-05-10 01:49:15 +0200883 int abstract = 0;
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200884 int new_fd = -1;
Willy Tarreaue835bd82020-09-16 11:35:47 +0200885 int sock_type, ctrl_type;
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100886
887 portl = porth = porta = 0;
Willy Tarreau72b8c1f2015-09-08 15:50:19 +0200888 if (fqdn)
889 *fqdn = NULL;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200890
Willy Tarreaudad36a32013-03-11 01:20:04 +0100891 str2 = back = env_expand(strdup(str));
Willy Tarreaudf350f12013-03-01 20:22:54 +0100892 if (str2 == NULL) {
893 memprintf(err, "out of memory in '%s'\n", __FUNCTION__);
Willy Tarreaud5191e72010-02-09 20:50:45 +0100894 goto out;
Willy Tarreaudf350f12013-03-01 20:22:54 +0100895 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200896
Willy Tarreau9f69f462015-09-08 16:01:25 +0200897 if (!*str2) {
898 memprintf(err, "'%s' resolves to an empty address (environment variable missing?)\n", str);
899 goto out;
900 }
901
Willy Tarreau24709282013-03-10 21:32:12 +0100902 memset(&ss, 0, sizeof(ss));
903
Willy Tarreaue835bd82020-09-16 11:35:47 +0200904 /* prepare the default socket types */
905 if ((opts & (PA_O_STREAM|PA_O_DGRAM)) == PA_O_DGRAM)
906 sock_type = ctrl_type = SOCK_DGRAM;
907 else
908 sock_type = ctrl_type = SOCK_STREAM;
909
910 if (strncmp(str2, "stream+", 7) == 0) {
911 str2 += 7;
912 sock_type = ctrl_type = SOCK_STREAM;
913 }
914 else if (strncmp(str2, "dgram+", 6) == 0) {
915 str2 += 6;
916 sock_type = ctrl_type = SOCK_DGRAM;
917 }
918
Willy Tarreau24709282013-03-10 21:32:12 +0100919 if (strncmp(str2, "unix@", 5) == 0) {
920 str2 += 5;
Willy Tarreauccfccef2014-05-10 01:49:15 +0200921 abstract = 0;
Willy Tarreau24709282013-03-10 21:32:12 +0100922 ss.ss_family = AF_UNIX;
923 }
Willy Tarreauccfccef2014-05-10 01:49:15 +0200924 else if (strncmp(str2, "abns@", 5) == 0) {
925 str2 += 5;
926 abstract = 1;
927 ss.ss_family = AF_UNIX;
928 }
Willy Tarreau24709282013-03-10 21:32:12 +0100929 else if (strncmp(str2, "ipv4@", 5) == 0) {
930 str2 += 5;
931 ss.ss_family = AF_INET;
932 }
933 else if (strncmp(str2, "ipv6@", 5) == 0) {
934 str2 += 5;
935 ss.ss_family = AF_INET6;
936 }
Emeric Brun3835c0d2020-07-07 09:46:09 +0200937 else if (strncmp(str2, "udp4@", 5) == 0) {
938 str2 += 5;
939 ss.ss_family = AF_INET;
Willy Tarreaue835bd82020-09-16 11:35:47 +0200940 sock_type = ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +0200941 }
942 else if (strncmp(str2, "udp6@", 5) == 0) {
943 str2 += 5;
944 ss.ss_family = AF_INET6;
Willy Tarreaue835bd82020-09-16 11:35:47 +0200945 sock_type = ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +0200946 }
947 else if (strncmp(str2, "udp@", 4) == 0) {
948 str2 += 4;
949 ss.ss_family = AF_UNSPEC;
Willy Tarreaue835bd82020-09-16 11:35:47 +0200950 sock_type = ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +0200951 }
Willy Tarreau5a7beed2020-09-04 16:54:05 +0200952 else if (strncmp(str2, "fd@", 3) == 0) {
953 str2 += 3;
954 ss.ss_family = AF_CUST_EXISTING_FD;
955 }
956 else if (strncmp(str2, "sockpair@", 9) == 0) {
957 str2 += 9;
958 ss.ss_family = AF_CUST_SOCKPAIR;
959 }
Willy Tarreau24709282013-03-10 21:32:12 +0100960 else if (*str2 == '/') {
961 ss.ss_family = AF_UNIX;
962 }
963 else
964 ss.ss_family = AF_UNSPEC;
965
Willy Tarreau5a7beed2020-09-04 16:54:05 +0200966 if (ss.ss_family == AF_CUST_SOCKPAIR) {
Willy Tarreaua215be22020-09-16 10:14:16 +0200967 struct sockaddr_storage ss2;
968 socklen_t addr_len;
William Lallemand2fe7dd02018-09-11 16:51:29 +0200969 char *endptr;
970
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200971 new_fd = strtol(str2, &endptr, 10);
972 if (!*str2 || new_fd < 0 || *endptr) {
William Lallemand2fe7dd02018-09-11 16:51:29 +0200973 memprintf(err, "file descriptor '%s' is not a valid integer in '%s'\n", str2, str);
974 goto out;
975 }
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200976
Willy Tarreaua215be22020-09-16 10:14:16 +0200977 /* just verify that it's a socket */
978 addr_len = sizeof(ss2);
979 if (getsockname(new_fd, (struct sockaddr *)&ss2, &addr_len) == -1) {
980 memprintf(err, "cannot use file descriptor '%d' : %s.\n", new_fd, strerror(errno));
981 goto out;
982 }
983
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200984 ((struct sockaddr_in *)&ss)->sin_addr.s_addr = new_fd;
985 ((struct sockaddr_in *)&ss)->sin_port = 0;
William Lallemand2fe7dd02018-09-11 16:51:29 +0200986 }
Willy Tarreau5a7beed2020-09-04 16:54:05 +0200987 else if (ss.ss_family == AF_CUST_EXISTING_FD) {
Willy Tarreau40aa0702013-03-10 23:51:38 +0100988 char *endptr;
989
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200990 new_fd = strtol(str2, &endptr, 10);
991 if (!*str2 || new_fd < 0 || *endptr) {
Willy Tarreaudad36a32013-03-11 01:20:04 +0100992 memprintf(err, "file descriptor '%s' is not a valid integer in '%s'\n", str2, str);
Willy Tarreau40aa0702013-03-10 23:51:38 +0100993 goto out;
994 }
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200995
Willy Tarreau6edc7222020-09-15 17:41:56 +0200996 if (opts & PA_O_SOCKET_FD) {
997 socklen_t addr_len;
998 int type;
999
1000 addr_len = sizeof(ss);
1001 if (getsockname(new_fd, (struct sockaddr *)&ss, &addr_len) == -1) {
1002 memprintf(err, "cannot use file descriptor '%d' : %s.\n", new_fd, strerror(errno));
1003 goto out;
1004 }
1005
1006 addr_len = sizeof(type);
1007 if (getsockopt(new_fd, SOL_SOCKET, SO_TYPE, &type, &addr_len) != 0 ||
Willy Tarreaue835bd82020-09-16 11:35:47 +02001008 (type == SOCK_STREAM) != (sock_type == SOCK_STREAM)) {
Willy Tarreau6edc7222020-09-15 17:41:56 +02001009 memprintf(err, "socket on file descriptor '%d' is of the wrong type.\n", new_fd);
1010 goto out;
1011 }
1012
1013 porta = portl = porth = get_host_port(&ss);
1014 } else if (opts & PA_O_RAW_FD) {
1015 ((struct sockaddr_in *)&ss)->sin_addr.s_addr = new_fd;
1016 ((struct sockaddr_in *)&ss)->sin_port = 0;
1017 } else {
1018 memprintf(err, "a file descriptor is not acceptable here in '%s'\n", str);
1019 goto out;
1020 }
Willy Tarreau40aa0702013-03-10 23:51:38 +01001021 }
1022 else if (ss.ss_family == AF_UNIX) {
Willy Tarreau8daa9202019-06-16 18:16:33 +02001023 struct sockaddr_un *un = (struct sockaddr_un *)&ss;
Willy Tarreau15586382013-03-04 19:48:14 +01001024 int prefix_path_len;
1025 int max_path_len;
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001026 int adr_len;
Willy Tarreau15586382013-03-04 19:48:14 +01001027
1028 /* complete unix socket path name during startup or soft-restart is
1029 * <unix_bind_prefix><path>.<pid>.<bak|tmp>
1030 */
Willy Tarreauccfccef2014-05-10 01:49:15 +02001031 prefix_path_len = (pfx && !abstract) ? strlen(pfx) : 0;
Willy Tarreau8daa9202019-06-16 18:16:33 +02001032 max_path_len = (sizeof(un->sun_path) - 1) -
Willy Tarreau327ea5a2020-02-11 06:43:37 +01001033 (abstract ? 0 : prefix_path_len + 1 + 5 + 1 + 3);
Willy Tarreau15586382013-03-04 19:48:14 +01001034
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001035 adr_len = strlen(str2);
1036 if (adr_len > max_path_len) {
Willy Tarreau15586382013-03-04 19:48:14 +01001037 memprintf(err, "socket path '%s' too long (max %d)\n", str, max_path_len);
1038 goto out;
1039 }
1040
Willy Tarreauccfccef2014-05-10 01:49:15 +02001041 /* when abstract==1, we skip the first zero and copy all bytes except the trailing zero */
Willy Tarreau8daa9202019-06-16 18:16:33 +02001042 memset(un->sun_path, 0, sizeof(un->sun_path));
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001043 if (prefix_path_len)
Willy Tarreau8daa9202019-06-16 18:16:33 +02001044 memcpy(un->sun_path, pfx, prefix_path_len);
1045 memcpy(un->sun_path + prefix_path_len + abstract, str2, adr_len + 1 - abstract);
Willy Tarreau15586382013-03-04 19:48:14 +01001046 }
Willy Tarreau24709282013-03-10 21:32:12 +01001047 else { /* IPv4 and IPv6 */
mildisff5d5102015-10-26 18:50:08 +01001048 char *end = str2 + strlen(str2);
1049 char *chr;
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001050
mildisff5d5102015-10-26 18:50:08 +01001051 /* search for : or ] whatever comes first */
1052 for (chr = end-1; chr > str2; chr--) {
1053 if (*chr == ']' || *chr == ':')
1054 break;
1055 }
1056
1057 if (*chr == ':') {
1058 /* Found a colon before a closing-bracket, must be a port separator.
1059 * This guarantee backward compatibility.
1060 */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001061 if (!(opts & PA_O_PORT_OK)) {
1062 memprintf(err, "port specification not permitted here in '%s'", str);
1063 goto out;
1064 }
mildisff5d5102015-10-26 18:50:08 +01001065 *chr++ = '\0';
1066 port1 = chr;
1067 }
1068 else {
1069 /* Either no colon and no closing-bracket
1070 * or directly ending with a closing-bracket.
1071 * However, no port.
1072 */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001073 if (opts & PA_O_PORT_MAND) {
1074 memprintf(err, "missing port specification in '%s'", str);
1075 goto out;
1076 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001077 port1 = "";
mildisff5d5102015-10-26 18:50:08 +01001078 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02001079
Willy Tarreau90807112020-02-25 08:16:33 +01001080 if (isdigit((unsigned char)*port1)) { /* single port or range */
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001081 port2 = strchr(port1, '-');
Willy Tarreau7f96a842020-09-15 11:12:44 +02001082 if (port2) {
1083 if (!(opts & PA_O_PORT_RANGE)) {
1084 memprintf(err, "port range not permitted here in '%s'", str);
1085 goto out;
1086 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001087 *port2++ = '\0';
Willy Tarreau7f96a842020-09-15 11:12:44 +02001088 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001089 else
1090 port2 = port1;
1091 portl = atoi(port1);
1092 porth = atoi(port2);
Willy Tarreau7f96a842020-09-15 11:12:44 +02001093
1094 if (portl < !!(opts & PA_O_PORT_MAND) || portl > 65535) {
1095 memprintf(err, "invalid port '%s'", port1);
1096 goto out;
1097 }
1098
1099 if (porth < !!(opts & PA_O_PORT_MAND) || porth > 65535) {
1100 memprintf(err, "invalid port '%s'", port2);
1101 goto out;
1102 }
1103
1104 if (portl > porth) {
1105 memprintf(err, "invalid port range '%d-%d'", portl, porth);
1106 goto out;
1107 }
1108
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001109 porta = portl;
1110 }
1111 else if (*port1 == '-') { /* negative offset */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001112 if (!(opts & PA_O_PORT_OFS)) {
1113 memprintf(err, "port offset not permitted here in '%s'", str);
1114 goto out;
1115 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001116 portl = atoi(port1 + 1);
1117 porta = -portl;
1118 }
1119 else if (*port1 == '+') { /* positive offset */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001120 if (!(opts & PA_O_PORT_OFS)) {
1121 memprintf(err, "port offset not permitted here in '%s'", str);
1122 goto out;
1123 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001124 porth = atoi(port1 + 1);
1125 porta = porth;
1126 }
1127 else if (*port1) { /* other any unexpected char */
Willy Tarreaudad36a32013-03-11 01:20:04 +01001128 memprintf(err, "invalid character '%c' in port number '%s' in '%s'\n", *port1, port1, str);
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001129 goto out;
1130 }
Willy Tarreau7f96a842020-09-15 11:12:44 +02001131 else if (opts & PA_O_PORT_MAND) {
1132 memprintf(err, "missing port specification in '%s'", str);
1133 goto out;
1134 }
Willy Tarreauceccdd72016-11-02 22:27:10 +01001135
1136 /* first try to parse the IP without resolving. If it fails, it
1137 * tells us we need to keep a copy of the FQDN to resolve later
1138 * and to enable DNS. In this case we can proceed if <fqdn> is
Willy Tarreaucd3a55912020-09-04 15:30:46 +02001139 * set or if PA_O_RESOLVE is set, otherwise it's an error.
Willy Tarreauceccdd72016-11-02 22:27:10 +01001140 */
1141 if (str2ip2(str2, &ss, 0) == NULL) {
Willy Tarreaucd3a55912020-09-04 15:30:46 +02001142 if ((!(opts & PA_O_RESOLVE) && !fqdn) ||
1143 ((opts & PA_O_RESOLVE) && str2ip2(str2, &ss, 1) == NULL)) {
Willy Tarreauceccdd72016-11-02 22:27:10 +01001144 memprintf(err, "invalid address: '%s' in '%s'\n", str2, str);
1145 goto out;
1146 }
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001147
Willy Tarreauceccdd72016-11-02 22:27:10 +01001148 if (fqdn) {
1149 if (str2 != back)
1150 memmove(back, str2, strlen(str2) + 1);
1151 *fqdn = back;
1152 back = NULL;
1153 }
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001154 }
Willy Tarreauceccdd72016-11-02 22:27:10 +01001155 set_host_port(&ss, porta);
Willy Tarreaue4c58c82013-03-06 15:28:17 +01001156 }
Willy Tarreaufab5a432011-03-04 15:31:53 +01001157
Willy Tarreaue835bd82020-09-16 11:35:47 +02001158 if (ctrl_type == SOCK_STREAM && !(opts & PA_O_STREAM)) {
1159 memprintf(err, "stream-type socket not acceptable in '%s'\n", str);
1160 goto out;
1161 }
1162 else if (ctrl_type == SOCK_DGRAM && !(opts & PA_O_DGRAM)) {
1163 memprintf(err, "dgram-type socket not acceptable in '%s'\n", str);
1164 goto out;
1165 }
1166
Willy Tarreau65ec4e32020-09-16 19:17:08 +02001167 if (proto || (opts & PA_O_CONNECT)) {
Willy Tarreau5fc93282020-09-16 18:25:03 +02001168 /* Note: if the caller asks for a proto, we must find one,
1169 * except if we return with an fqdn that will resolve later,
1170 * in which case the address is not known yet (this is only
1171 * for servers actually).
1172 */
Willy Tarreaub2ffc992020-09-16 21:37:31 +02001173
1174 /* FIXME: for now UDP is still its own family. However some UDP clients
1175 * (logs, dns) use AF_INET and are not aware of AF_CUST_UDP*. Since we
1176 * only want this mapping for listeners and they are the only ones
1177 * setting PA_O_SOCKET_FD, for now we condition this mapping to this.
1178 * This effectively means that for now we return TCPv4/v6 for UDP senders.
1179 */
1180 new_proto = protocol_lookup(ss.ss_family,
1181 (opts & PA_O_SOCKET_FD) && sock_type == SOCK_DGRAM,
1182 (opts & PA_O_SOCKET_FD) && ctrl_type == SOCK_DGRAM);
1183
Willy Tarreau5fc93282020-09-16 18:25:03 +02001184 if (!new_proto && (!fqdn || !*fqdn)) {
1185 memprintf(err, "unsupported protocol family %d for address '%s'", ss.ss_family, str);
1186 goto out;
1187 }
Willy Tarreau65ec4e32020-09-16 19:17:08 +02001188
1189 if ((opts & PA_O_CONNECT) && new_proto && !new_proto->connect) {
1190 memprintf(err, "connect() not supported for this protocol family %d used by address '%s'", ss.ss_family, str);
1191 goto out;
1192 }
Willy Tarreau5fc93282020-09-16 18:25:03 +02001193 }
1194
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001195 ret = &ss;
Willy Tarreaud5191e72010-02-09 20:50:45 +01001196 out:
Willy Tarreau48ef4c92017-01-06 18:32:38 +01001197 if (port)
1198 *port = porta;
Willy Tarreaud4448bc2013-02-20 15:55:15 +01001199 if (low)
1200 *low = portl;
1201 if (high)
1202 *high = porth;
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001203 if (fd)
1204 *fd = new_fd;
Willy Tarreau5fc93282020-09-16 18:25:03 +02001205 if (proto)
1206 *proto = new_proto;
Willy Tarreau24709282013-03-10 21:32:12 +01001207 free(back);
Willy Tarreaud5191e72010-02-09 20:50:45 +01001208 return ret;
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001209}
1210
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001211/* converts <str> to a struct in_addr containing a network mask. It can be
1212 * passed in dotted form (255.255.255.0) or in CIDR form (24). It returns 1
Jarno Huuskonen577d5ac2017-05-21 17:32:21 +03001213 * if the conversion succeeds otherwise zero.
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001214 */
1215int str2mask(const char *str, struct in_addr *mask)
1216{
1217 if (strchr(str, '.') != NULL) { /* dotted notation */
1218 if (!inet_pton(AF_INET, str, mask))
1219 return 0;
1220 }
1221 else { /* mask length */
1222 char *err;
1223 unsigned long len = strtol(str, &err, 10);
1224
1225 if (!*str || (err && *err) || (unsigned)len > 32)
1226 return 0;
Tim Duesterhus8575f722018-01-25 16:24:48 +01001227
1228 len2mask4(len, mask);
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001229 }
1230 return 1;
1231}
1232
Tim Duesterhus47185172018-01-25 16:24:49 +01001233/* converts <str> to a struct in6_addr containing a network mask. It can be
Tim Duesterhus5e642862018-02-20 17:02:18 +01001234 * passed in quadruplet form (ffff:ffff::) or in CIDR form (64). It returns 1
Tim Duesterhus47185172018-01-25 16:24:49 +01001235 * if the conversion succeeds otherwise zero.
1236 */
1237int str2mask6(const char *str, struct in6_addr *mask)
1238{
1239 if (strchr(str, ':') != NULL) { /* quadruplet notation */
1240 if (!inet_pton(AF_INET6, str, mask))
1241 return 0;
1242 }
1243 else { /* mask length */
1244 char *err;
1245 unsigned long len = strtol(str, &err, 10);
1246
1247 if (!*str || (err && *err) || (unsigned)len > 128)
1248 return 0;
1249
1250 len2mask6(len, mask);
1251 }
1252 return 1;
1253}
1254
Thierry FOURNIERb0504632013-12-14 15:39:02 +01001255/* convert <cidr> to struct in_addr <mask>. It returns 1 if the conversion
1256 * succeeds otherwise zero.
1257 */
1258int cidr2dotted(int cidr, struct in_addr *mask) {
1259
1260 if (cidr < 0 || cidr > 32)
1261 return 0;
1262
1263 mask->s_addr = cidr ? htonl(~0UL << (32 - cidr)) : 0;
1264 return 1;
1265}
1266
Thierry Fournier70473a52016-02-17 17:12:14 +01001267/* Convert mask from bit length form to in_addr form.
1268 * This function never fails.
1269 */
1270void len2mask4(int len, struct in_addr *addr)
1271{
1272 if (len >= 32) {
1273 addr->s_addr = 0xffffffff;
1274 return;
1275 }
1276 if (len <= 0) {
1277 addr->s_addr = 0x00000000;
1278 return;
1279 }
1280 addr->s_addr = 0xffffffff << (32 - len);
1281 addr->s_addr = htonl(addr->s_addr);
1282}
1283
1284/* Convert mask from bit length form to in6_addr form.
1285 * This function never fails.
1286 */
1287void len2mask6(int len, struct in6_addr *addr)
1288{
1289 len2mask4(len, (struct in_addr *)&addr->s6_addr[0]); /* msb */
1290 len -= 32;
1291 len2mask4(len, (struct in_addr *)&addr->s6_addr[4]);
1292 len -= 32;
1293 len2mask4(len, (struct in_addr *)&addr->s6_addr[8]);
1294 len -= 32;
1295 len2mask4(len, (struct in_addr *)&addr->s6_addr[12]); /* lsb */
1296}
1297
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001298/*
Willy Tarreaud077a8e2007-05-08 18:28:09 +02001299 * converts <str> to two struct in_addr* which must be pre-allocated.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001300 * The format is "addr[/mask]", where "addr" cannot be empty, and mask
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001301 * is optional and either in the dotted or CIDR notation.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001302 * Note: "addr" can also be a hostname. Returns 1 if OK, 0 if error.
1303 */
Thierry FOURNIERfc7ac7b2014-02-11 15:23:04 +01001304int str2net(const char *str, int resolve, struct in_addr *addr, struct in_addr *mask)
Willy Tarreaubaaee002006-06-26 02:48:02 +02001305{
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001306 __label__ out_free, out_err;
1307 char *c, *s;
1308 int ret_val;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001309
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001310 s = strdup(str);
1311 if (!s)
1312 return 0;
1313
Willy Tarreaubaaee002006-06-26 02:48:02 +02001314 memset(mask, 0, sizeof(*mask));
1315 memset(addr, 0, sizeof(*addr));
Willy Tarreaubaaee002006-06-26 02:48:02 +02001316
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001317 if ((c = strrchr(s, '/')) != NULL) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02001318 *c++ = '\0';
1319 /* c points to the mask */
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001320 if (!str2mask(c, mask))
1321 goto out_err;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001322 }
1323 else {
Willy Tarreauebd61602006-12-30 11:54:15 +01001324 mask->s_addr = ~0U;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001325 }
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001326 if (!inet_pton(AF_INET, s, addr)) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02001327 struct hostent *he;
1328
Thierry FOURNIERfc7ac7b2014-02-11 15:23:04 +01001329 if (!resolve)
1330 goto out_err;
1331
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001332 if ((he = gethostbyname(s)) == NULL) {
1333 goto out_err;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001334 }
1335 else
1336 *addr = *(struct in_addr *) *(he->h_addr_list);
1337 }
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001338
1339 ret_val = 1;
1340 out_free:
1341 free(s);
1342 return ret_val;
1343 out_err:
1344 ret_val = 0;
1345 goto out_free;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001346}
1347
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001348
1349/*
Willy Tarreau6d20e282012-04-27 22:49:47 +02001350 * converts <str> to two struct in6_addr* which must be pre-allocated.
1351 * The format is "addr[/mask]", where "addr" cannot be empty, and mask
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001352 * is an optional number of bits (128 being the default).
Willy Tarreau6d20e282012-04-27 22:49:47 +02001353 * Returns 1 if OK, 0 if error.
1354 */
1355int str62net(const char *str, struct in6_addr *addr, unsigned char *mask)
1356{
1357 char *c, *s;
1358 int ret_val = 0;
1359 char *err;
1360 unsigned long len = 128;
1361
1362 s = strdup(str);
1363 if (!s)
1364 return 0;
1365
1366 memset(mask, 0, sizeof(*mask));
1367 memset(addr, 0, sizeof(*addr));
1368
1369 if ((c = strrchr(s, '/')) != NULL) {
1370 *c++ = '\0'; /* c points to the mask */
1371 if (!*c)
1372 goto out_free;
1373
1374 len = strtoul(c, &err, 10);
1375 if ((err && *err) || (unsigned)len > 128)
1376 goto out_free;
1377 }
1378 *mask = len; /* OK we have a valid mask in <len> */
1379
1380 if (!inet_pton(AF_INET6, s, addr))
1381 goto out_free;
1382
1383 ret_val = 1;
1384 out_free:
1385 free(s);
1386 return ret_val;
1387}
1388
1389
1390/*
David du Colombier6f5ccb12011-03-10 22:26:24 +01001391 * Parse IPv4 address found in url.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001392 */
David du Colombier6f5ccb12011-03-10 22:26:24 +01001393int url2ipv4(const char *addr, struct in_addr *dst)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001394{
1395 int saw_digit, octets, ch;
1396 u_char tmp[4], *tp;
1397 const char *cp = addr;
1398
1399 saw_digit = 0;
1400 octets = 0;
1401 *(tp = tmp) = 0;
1402
1403 while (*addr) {
1404 unsigned char digit = (ch = *addr++) - '0';
1405 if (digit > 9 && ch != '.')
1406 break;
1407 if (digit <= 9) {
1408 u_int new = *tp * 10 + digit;
1409 if (new > 255)
1410 return 0;
1411 *tp = new;
1412 if (!saw_digit) {
1413 if (++octets > 4)
1414 return 0;
1415 saw_digit = 1;
1416 }
1417 } else if (ch == '.' && saw_digit) {
1418 if (octets == 4)
1419 return 0;
1420 *++tp = 0;
1421 saw_digit = 0;
1422 } else
1423 return 0;
1424 }
1425
1426 if (octets < 4)
1427 return 0;
1428
1429 memcpy(&dst->s_addr, tmp, 4);
1430 return addr-cp-1;
1431}
1432
1433/*
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001434 * Resolve destination server from URL. Convert <str> to a sockaddr_storage.
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001435 * <out> contain the code of the detected scheme, the start and length of
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001436 * the hostname. Actually only http and https are supported. <out> can be NULL.
1437 * This function returns the consumed length. It is useful if you parse complete
1438 * url like http://host:port/path, because the consumed length corresponds to
1439 * the first character of the path. If the conversion fails, it returns -1.
1440 *
1441 * This function tries to resolve the DNS name if haproxy is in starting mode.
1442 * So, this function may be used during the configuration parsing.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001443 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001444int url2sa(const char *url, int ulen, struct sockaddr_storage *addr, struct split_url *out)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001445{
1446 const char *curr = url, *cp = url;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001447 const char *end;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001448 int ret, url_code = 0;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001449 unsigned long long int http_code = 0;
1450 int default_port;
1451 struct hostent *he;
1452 char *p;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001453
1454 /* Firstly, try to find :// pattern */
1455 while (curr < url+ulen && url_code != 0x3a2f2f) {
1456 url_code = ((url_code & 0xffff) << 8);
1457 url_code += (unsigned char)*curr++;
1458 }
1459
1460 /* Secondly, if :// pattern is found, verify parsed stuff
1461 * before pattern is matching our http pattern.
1462 * If so parse ip address and port in uri.
1463 *
1464 * WARNING: Current code doesn't support dynamic async dns resolver.
1465 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001466 if (url_code != 0x3a2f2f)
1467 return -1;
1468
1469 /* Copy scheme, and utrn to lower case. */
1470 while (cp < curr - 3)
1471 http_code = (http_code << 8) + *cp++;
1472 http_code |= 0x2020202020202020ULL; /* Turn everything to lower case */
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001473
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001474 /* HTTP or HTTPS url matching */
1475 if (http_code == 0x2020202068747470ULL) {
1476 default_port = 80;
1477 if (out)
1478 out->scheme = SCH_HTTP;
1479 }
1480 else if (http_code == 0x2020206874747073ULL) {
1481 default_port = 443;
1482 if (out)
1483 out->scheme = SCH_HTTPS;
1484 }
1485 else
1486 return -1;
1487
1488 /* If the next char is '[', the host address is IPv6. */
1489 if (*curr == '[') {
1490 curr++;
1491
1492 /* Check trash size */
1493 if (trash.size < ulen)
1494 return -1;
1495
1496 /* Look for ']' and copy the address in a trash buffer. */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001497 p = trash.area;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001498 for (end = curr;
1499 end < url + ulen && *end != ']';
1500 end++, p++)
1501 *p = *end;
1502 if (*end != ']')
1503 return -1;
1504 *p = '\0';
1505
1506 /* Update out. */
1507 if (out) {
1508 out->host = curr;
1509 out->host_len = end - curr;
1510 }
1511
1512 /* Try IPv6 decoding. */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001513 if (!inet_pton(AF_INET6, trash.area, &((struct sockaddr_in6 *)addr)->sin6_addr))
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001514 return -1;
1515 end++;
1516
1517 /* Decode port. */
1518 if (*end == ':') {
1519 end++;
1520 default_port = read_uint(&end, url + ulen);
1521 }
1522 ((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1523 ((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1524 return end - url;
1525 }
1526 else {
1527 /* We are looking for IP address. If you want to parse and
1528 * resolve hostname found in url, you can use str2sa_range(), but
1529 * be warned this can slow down global daemon performances
1530 * while handling lagging dns responses.
1531 */
1532 ret = url2ipv4(curr, &((struct sockaddr_in *)addr)->sin_addr);
1533 if (ret) {
1534 /* Update out. */
1535 if (out) {
1536 out->host = curr;
1537 out->host_len = ret;
1538 }
1539
1540 curr += ret;
1541
1542 /* Decode port. */
1543 if (*curr == ':') {
1544 curr++;
1545 default_port = read_uint(&curr, url + ulen);
1546 }
1547 ((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1548
1549 /* Set family. */
1550 ((struct sockaddr_in *)addr)->sin_family = AF_INET;
1551 return curr - url;
1552 }
1553 else if (global.mode & MODE_STARTING) {
1554 /* The IPv4 and IPv6 decoding fails, maybe the url contain name. Try to execute
1555 * synchronous DNS request only if HAProxy is in the start state.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001556 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001557
1558 /* look for : or / or end */
1559 for (end = curr;
1560 end < url + ulen && *end != '/' && *end != ':';
1561 end++);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001562 memcpy(trash.area, curr, end - curr);
1563 trash.area[end - curr] = '\0';
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001564
1565 /* try to resolve an IPv4/IPv6 hostname */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001566 he = gethostbyname(trash.area);
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001567 if (!he)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001568 return -1;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001569
1570 /* Update out. */
1571 if (out) {
1572 out->host = curr;
1573 out->host_len = end - curr;
1574 }
1575
1576 /* Decode port. */
1577 if (*end == ':') {
1578 end++;
1579 default_port = read_uint(&end, url + ulen);
1580 }
1581
1582 /* Copy IP address, set port and family. */
1583 switch (he->h_addrtype) {
1584 case AF_INET:
1585 ((struct sockaddr_in *)addr)->sin_addr = *(struct in_addr *) *(he->h_addr_list);
1586 ((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1587 ((struct sockaddr_in *)addr)->sin_family = AF_INET;
1588 return end - url;
1589
1590 case AF_INET6:
1591 ((struct sockaddr_in6 *)addr)->sin6_addr = *(struct in6_addr *) *(he->h_addr_list);
1592 ((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1593 ((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1594 return end - url;
1595 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001596 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001597 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001598 return -1;
1599}
1600
Willy Tarreau631f01c2011-09-05 00:36:48 +02001601/* Tries to convert a sockaddr_storage address to text form. Upon success, the
1602 * address family is returned so that it's easy for the caller to adapt to the
1603 * output format. Zero is returned if the address family is not supported. -1
1604 * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1605 * supported.
1606 */
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001607int addr_to_str(const struct sockaddr_storage *addr, char *str, int size)
Willy Tarreau631f01c2011-09-05 00:36:48 +02001608{
1609
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001610 const void *ptr;
Willy Tarreau631f01c2011-09-05 00:36:48 +02001611
1612 if (size < 5)
1613 return 0;
1614 *str = '\0';
1615
1616 switch (addr->ss_family) {
1617 case AF_INET:
1618 ptr = &((struct sockaddr_in *)addr)->sin_addr;
1619 break;
1620 case AF_INET6:
1621 ptr = &((struct sockaddr_in6 *)addr)->sin6_addr;
1622 break;
1623 case AF_UNIX:
1624 memcpy(str, "unix", 5);
1625 return addr->ss_family;
1626 default:
1627 return 0;
1628 }
1629
1630 if (inet_ntop(addr->ss_family, ptr, str, size))
1631 return addr->ss_family;
1632
1633 /* failed */
1634 return -1;
1635}
1636
Simon Horman75ab8bd2014-06-16 09:39:41 +09001637/* Tries to convert a sockaddr_storage port to text form. Upon success, the
1638 * address family is returned so that it's easy for the caller to adapt to the
1639 * output format. Zero is returned if the address family is not supported. -1
1640 * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1641 * supported.
1642 */
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001643int port_to_str(const struct sockaddr_storage *addr, char *str, int size)
Simon Horman75ab8bd2014-06-16 09:39:41 +09001644{
1645
1646 uint16_t port;
1647
1648
Willy Tarreaud7dad1b2017-01-06 16:46:22 +01001649 if (size < 6)
Simon Horman75ab8bd2014-06-16 09:39:41 +09001650 return 0;
1651 *str = '\0';
1652
1653 switch (addr->ss_family) {
1654 case AF_INET:
1655 port = ((struct sockaddr_in *)addr)->sin_port;
1656 break;
1657 case AF_INET6:
1658 port = ((struct sockaddr_in6 *)addr)->sin6_port;
1659 break;
1660 case AF_UNIX:
1661 memcpy(str, "unix", 5);
1662 return addr->ss_family;
1663 default:
1664 return 0;
1665 }
1666
1667 snprintf(str, size, "%u", ntohs(port));
1668 return addr->ss_family;
1669}
1670
Willy Tarreau16e01562016-08-09 16:46:18 +02001671/* check if the given address is local to the system or not. It will return
1672 * -1 when it's not possible to know, 0 when the address is not local, 1 when
1673 * it is. We don't want to iterate over all interfaces for this (and it is not
1674 * portable). So instead we try to bind in UDP to this address on a free non
1675 * privileged port and to connect to the same address, port 0 (connect doesn't
1676 * care). If it succeeds, we own the address. Note that non-inet addresses are
1677 * considered local since they're most likely AF_UNIX.
1678 */
1679int addr_is_local(const struct netns_entry *ns,
1680 const struct sockaddr_storage *orig)
1681{
1682 struct sockaddr_storage addr;
1683 int result;
1684 int fd;
1685
1686 if (!is_inet_addr(orig))
1687 return 1;
1688
1689 memcpy(&addr, orig, sizeof(addr));
1690 set_host_port(&addr, 0);
1691
1692 fd = my_socketat(ns, addr.ss_family, SOCK_DGRAM, IPPROTO_UDP);
1693 if (fd < 0)
1694 return -1;
1695
1696 result = -1;
1697 if (bind(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == 0) {
1698 if (connect(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == -1)
1699 result = 0; // fail, non-local address
1700 else
1701 result = 1; // success, local address
1702 }
1703 else {
1704 if (errno == EADDRNOTAVAIL)
1705 result = 0; // definitely not local :-)
1706 }
1707 close(fd);
1708
1709 return result;
1710}
1711
Willy Tarreaubaaee002006-06-26 02:48:02 +02001712/* will try to encode the string <string> replacing all characters tagged in
1713 * <map> with the hexadecimal representation of their ASCII-code (2 digits)
1714 * prefixed by <escape>, and will store the result between <start> (included)
1715 * and <stop> (excluded), and will always terminate the string with a '\0'
1716 * before <stop>. The position of the '\0' is returned if the conversion
1717 * completes. If bytes are missing between <start> and <stop>, then the
1718 * conversion will be incomplete and truncated. If <stop> <= <start>, the '\0'
1719 * cannot even be stored so we return <start> without writing the 0.
1720 * The input string must also be zero-terminated.
1721 */
1722const char hextab[16] = "0123456789ABCDEF";
1723char *encode_string(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001724 const char escape, const long *map,
Willy Tarreaubaaee002006-06-26 02:48:02 +02001725 const char *string)
1726{
1727 if (start < stop) {
1728 stop--; /* reserve one byte for the final '\0' */
1729 while (start < stop && *string != '\0') {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001730 if (!ha_bit_test((unsigned char)(*string), map))
Willy Tarreaubaaee002006-06-26 02:48:02 +02001731 *start++ = *string;
1732 else {
1733 if (start + 3 >= stop)
1734 break;
1735 *start++ = escape;
1736 *start++ = hextab[(*string >> 4) & 15];
1737 *start++ = hextab[*string & 15];
1738 }
1739 string++;
1740 }
1741 *start = '\0';
1742 }
1743 return start;
1744}
1745
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001746/*
1747 * Same behavior as encode_string() above, except that it encodes chunk
1748 * <chunk> instead of a string.
1749 */
1750char *encode_chunk(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001751 const char escape, const long *map,
Willy Tarreau83061a82018-07-13 11:56:34 +02001752 const struct buffer *chunk)
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001753{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001754 char *str = chunk->area;
1755 char *end = chunk->area + chunk->data;
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001756
1757 if (start < stop) {
1758 stop--; /* reserve one byte for the final '\0' */
1759 while (start < stop && str < end) {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001760 if (!ha_bit_test((unsigned char)(*str), map))
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001761 *start++ = *str;
1762 else {
1763 if (start + 3 >= stop)
1764 break;
1765 *start++ = escape;
1766 *start++ = hextab[(*str >> 4) & 15];
1767 *start++ = hextab[*str & 15];
1768 }
1769 str++;
1770 }
1771 *start = '\0';
1772 }
1773 return start;
1774}
1775
Dragan Dosen0edd1092016-02-12 13:23:02 +01001776/*
1777 * Tries to prefix characters tagged in the <map> with the <escape>
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001778 * character. The input <string> must be zero-terminated. The result will
1779 * be stored between <start> (included) and <stop> (excluded). This
1780 * function will always try to terminate the resulting string with a '\0'
1781 * before <stop>, and will return its position if the conversion
1782 * completes.
1783 */
1784char *escape_string(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001785 const char escape, const long *map,
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001786 const char *string)
1787{
1788 if (start < stop) {
1789 stop--; /* reserve one byte for the final '\0' */
1790 while (start < stop && *string != '\0') {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001791 if (!ha_bit_test((unsigned char)(*string), map))
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001792 *start++ = *string;
1793 else {
1794 if (start + 2 >= stop)
1795 break;
1796 *start++ = escape;
1797 *start++ = *string;
1798 }
1799 string++;
1800 }
1801 *start = '\0';
1802 }
1803 return start;
1804}
1805
1806/*
1807 * Tries to prefix characters tagged in the <map> with the <escape>
Dragan Dosen0edd1092016-02-12 13:23:02 +01001808 * character. <chunk> contains the input to be escaped. The result will be
1809 * stored between <start> (included) and <stop> (excluded). The function
1810 * will always try to terminate the resulting string with a '\0' before
1811 * <stop>, and will return its position if the conversion completes.
1812 */
1813char *escape_chunk(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001814 const char escape, const long *map,
Willy Tarreau83061a82018-07-13 11:56:34 +02001815 const struct buffer *chunk)
Dragan Dosen0edd1092016-02-12 13:23:02 +01001816{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001817 char *str = chunk->area;
1818 char *end = chunk->area + chunk->data;
Dragan Dosen0edd1092016-02-12 13:23:02 +01001819
1820 if (start < stop) {
1821 stop--; /* reserve one byte for the final '\0' */
1822 while (start < stop && str < end) {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001823 if (!ha_bit_test((unsigned char)(*str), map))
Dragan Dosen0edd1092016-02-12 13:23:02 +01001824 *start++ = *str;
1825 else {
1826 if (start + 2 >= stop)
1827 break;
1828 *start++ = escape;
1829 *start++ = *str;
1830 }
1831 str++;
1832 }
1833 *start = '\0';
1834 }
1835 return start;
1836}
1837
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001838/* Check a string for using it in a CSV output format. If the string contains
1839 * one of the following four char <">, <,>, CR or LF, the string is
1840 * encapsulated between <"> and the <"> are escaped by a <""> sequence.
1841 * <str> is the input string to be escaped. The function assumes that
1842 * the input string is null-terminated.
1843 *
1844 * If <quote> is 0, the result is returned escaped but without double quote.
Willy Tarreau898529b2016-01-06 18:07:04 +01001845 * It is useful if the escaped string is used between double quotes in the
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001846 * format.
1847 *
Willy Tarreau898529b2016-01-06 18:07:04 +01001848 * printf("..., \"%s\", ...\r\n", csv_enc(str, 0, &trash));
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001849 *
Willy Tarreaub631c292016-01-08 10:04:08 +01001850 * If <quote> is 1, the converter puts the quotes only if any reserved character
1851 * is present. If <quote> is 2, the converter always puts the quotes.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001852 *
Willy Tarreau83061a82018-07-13 11:56:34 +02001853 * <output> is a struct buffer used for storing the output string.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001854 *
Willy Tarreau898529b2016-01-06 18:07:04 +01001855 * The function returns the converted string on its output. If an error
1856 * occurs, the function returns an empty string. This type of output is useful
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001857 * for using the function directly as printf() argument.
1858 *
1859 * If the output buffer is too short to contain the input string, the result
1860 * is truncated.
Willy Tarreau898529b2016-01-06 18:07:04 +01001861 *
Willy Tarreaub631c292016-01-08 10:04:08 +01001862 * This function appends the encoding to the existing output chunk, and it
1863 * guarantees that it starts immediately at the first available character of
1864 * the chunk. Please use csv_enc() instead if you want to replace the output
1865 * chunk.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001866 */
Willy Tarreau83061a82018-07-13 11:56:34 +02001867const char *csv_enc_append(const char *str, int quote, struct buffer *output)
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001868{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001869 char *end = output->area + output->size;
1870 char *out = output->area + output->data;
Willy Tarreau898529b2016-01-06 18:07:04 +01001871 char *ptr = out;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001872
Willy Tarreaub631c292016-01-08 10:04:08 +01001873 if (quote == 1) {
1874 /* automatic quoting: first verify if we'll have to quote the string */
1875 if (!strpbrk(str, "\n\r,\""))
1876 quote = 0;
1877 }
1878
1879 if (quote)
1880 *ptr++ = '"';
1881
Willy Tarreau898529b2016-01-06 18:07:04 +01001882 while (*str && ptr < end - 2) { /* -2 for reserving space for <"> and \0. */
1883 *ptr = *str;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001884 if (*str == '"') {
Willy Tarreau898529b2016-01-06 18:07:04 +01001885 ptr++;
1886 if (ptr >= end - 2) {
1887 ptr--;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001888 break;
1889 }
Willy Tarreau898529b2016-01-06 18:07:04 +01001890 *ptr = '"';
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001891 }
Willy Tarreau898529b2016-01-06 18:07:04 +01001892 ptr++;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001893 str++;
1894 }
1895
Willy Tarreaub631c292016-01-08 10:04:08 +01001896 if (quote)
1897 *ptr++ = '"';
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001898
Willy Tarreau898529b2016-01-06 18:07:04 +01001899 *ptr = '\0';
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001900 output->data = ptr - output->area;
Willy Tarreau898529b2016-01-06 18:07:04 +01001901 return out;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001902}
1903
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02001904/* Decode an URL-encoded string in-place. The resulting string might
1905 * be shorter. If some forbidden characters are found, the conversion is
Thierry FOURNIER5068d962013-10-04 16:27:27 +02001906 * aborted, the string is truncated before the issue and a negative value is
1907 * returned, otherwise the operation returns the length of the decoded string.
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02001908 * If the 'in_form' argument is non-nul the string is assumed to be part of
1909 * an "application/x-www-form-urlencoded" encoded string, and the '+' will be
1910 * turned to a space. If it's zero, this will only be done after a question
1911 * mark ('?').
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02001912 */
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02001913int url_decode(char *string, int in_form)
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02001914{
1915 char *in, *out;
Thierry FOURNIER5068d962013-10-04 16:27:27 +02001916 int ret = -1;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02001917
1918 in = string;
1919 out = string;
1920 while (*in) {
1921 switch (*in) {
1922 case '+' :
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02001923 *out++ = in_form ? ' ' : *in;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02001924 break;
1925 case '%' :
1926 if (!ishex(in[1]) || !ishex(in[2]))
1927 goto end;
1928 *out++ = (hex2i(in[1]) << 4) + hex2i(in[2]);
1929 in += 2;
1930 break;
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02001931 case '?':
1932 in_form = 1;
1933 /* fall through */
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02001934 default:
1935 *out++ = *in;
1936 break;
1937 }
1938 in++;
1939 }
Thierry FOURNIER5068d962013-10-04 16:27:27 +02001940 ret = out - string; /* success */
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02001941 end:
1942 *out = 0;
1943 return ret;
1944}
Willy Tarreaubaaee002006-06-26 02:48:02 +02001945
Willy Tarreau6911fa42007-03-04 18:06:08 +01001946unsigned int str2ui(const char *s)
1947{
1948 return __str2ui(s);
1949}
1950
1951unsigned int str2uic(const char *s)
1952{
1953 return __str2uic(s);
1954}
1955
1956unsigned int strl2ui(const char *s, int len)
1957{
1958 return __strl2ui(s, len);
1959}
1960
1961unsigned int strl2uic(const char *s, int len)
1962{
1963 return __strl2uic(s, len);
1964}
1965
Willy Tarreau4ec83cd2010-10-15 23:19:55 +02001966unsigned int read_uint(const char **s, const char *end)
1967{
1968 return __read_uint(s, end);
1969}
1970
Thierry FOURNIER763a5d82015-07-06 23:09:52 +02001971/* This function reads an unsigned integer from the string pointed to by <s> and
1972 * returns it. The <s> pointer is adjusted to point to the first unread char. The
1973 * function automatically stops at <end>. If the number overflows, the 2^64-1
1974 * value is returned.
1975 */
1976unsigned long long int read_uint64(const char **s, const char *end)
1977{
1978 const char *ptr = *s;
1979 unsigned long long int i = 0, tmp;
1980 unsigned int j;
1981
1982 while (ptr < end) {
1983
1984 /* read next char */
1985 j = *ptr - '0';
1986 if (j > 9)
1987 goto read_uint64_end;
1988
1989 /* add char to the number and check overflow. */
1990 tmp = i * 10;
1991 if (tmp / 10 != i) {
1992 i = ULLONG_MAX;
1993 goto read_uint64_eat;
1994 }
1995 if (ULLONG_MAX - tmp < j) {
1996 i = ULLONG_MAX;
1997 goto read_uint64_eat;
1998 }
1999 i = tmp + j;
2000 ptr++;
2001 }
2002read_uint64_eat:
2003 /* eat each numeric char */
2004 while (ptr < end) {
2005 if ((unsigned int)(*ptr - '0') > 9)
2006 break;
2007 ptr++;
2008 }
2009read_uint64_end:
2010 *s = ptr;
2011 return i;
2012}
2013
2014/* This function reads an integer from the string pointed to by <s> and returns
2015 * it. The <s> pointer is adjusted to point to the first unread char. The function
2016 * automatically stops at <end>. Il the number is bigger than 2^63-2, the 2^63-1
2017 * value is returned. If the number is lowest than -2^63-1, the -2^63 value is
2018 * returned.
2019 */
2020long long int read_int64(const char **s, const char *end)
2021{
2022 unsigned long long int i = 0;
2023 int neg = 0;
2024
2025 /* Look for minus char. */
2026 if (**s == '-') {
2027 neg = 1;
2028 (*s)++;
2029 }
2030 else if (**s == '+')
2031 (*s)++;
2032
2033 /* convert as positive number. */
2034 i = read_uint64(s, end);
2035
2036 if (neg) {
2037 if (i > 0x8000000000000000ULL)
2038 return LLONG_MIN;
2039 return -i;
2040 }
2041 if (i > 0x7fffffffffffffffULL)
2042 return LLONG_MAX;
2043 return i;
2044}
2045
Willy Tarreau6911fa42007-03-04 18:06:08 +01002046/* This one is 7 times faster than strtol() on athlon with checks.
2047 * It returns the value of the number composed of all valid digits read,
2048 * and can process negative numbers too.
2049 */
2050int strl2ic(const char *s, int len)
2051{
2052 int i = 0;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002053 int j, k;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002054
2055 if (len > 0) {
2056 if (*s != '-') {
2057 /* positive number */
2058 while (len-- > 0) {
2059 j = (*s++) - '0';
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002060 k = i * 10;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002061 if (j > 9)
2062 break;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002063 i = k + j;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002064 }
2065 } else {
2066 /* negative number */
2067 s++;
2068 while (--len > 0) {
2069 j = (*s++) - '0';
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002070 k = i * 10;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002071 if (j > 9)
2072 break;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002073 i = k - j;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002074 }
2075 }
2076 }
2077 return i;
2078}
2079
2080
2081/* This function reads exactly <len> chars from <s> and converts them to a
2082 * signed integer which it stores into <ret>. It accurately detects any error
2083 * (truncated string, invalid chars, overflows). It is meant to be used in
2084 * applications designed for hostile environments. It returns zero when the
2085 * number has successfully been converted, non-zero otherwise. When an error
2086 * is returned, the <ret> value is left untouched. It is yet 5 to 40 times
2087 * faster than strtol().
2088 */
2089int strl2irc(const char *s, int len, int *ret)
2090{
2091 int i = 0;
2092 int j;
2093
2094 if (!len)
2095 return 1;
2096
2097 if (*s != '-') {
2098 /* positive number */
2099 while (len-- > 0) {
2100 j = (*s++) - '0';
2101 if (j > 9) return 1; /* invalid char */
2102 if (i > INT_MAX / 10) return 1; /* check for multiply overflow */
2103 i = i * 10;
2104 if (i + j < i) return 1; /* check for addition overflow */
2105 i = i + j;
2106 }
2107 } else {
2108 /* negative number */
2109 s++;
2110 while (--len > 0) {
2111 j = (*s++) - '0';
2112 if (j > 9) return 1; /* invalid char */
2113 if (i < INT_MIN / 10) return 1; /* check for multiply overflow */
2114 i = i * 10;
2115 if (i - j > i) return 1; /* check for subtract overflow */
2116 i = i - j;
2117 }
2118 }
2119 *ret = i;
2120 return 0;
2121}
2122
2123
2124/* This function reads exactly <len> chars from <s> and converts them to a
2125 * signed integer which it stores into <ret>. It accurately detects any error
2126 * (truncated string, invalid chars, overflows). It is meant to be used in
2127 * applications designed for hostile environments. It returns zero when the
2128 * number has successfully been converted, non-zero otherwise. When an error
2129 * is returned, the <ret> value is left untouched. It is about 3 times slower
2130 * than str2irc().
2131 */
Willy Tarreau6911fa42007-03-04 18:06:08 +01002132
2133int strl2llrc(const char *s, int len, long long *ret)
2134{
2135 long long i = 0;
2136 int j;
2137
2138 if (!len)
2139 return 1;
2140
2141 if (*s != '-') {
2142 /* positive number */
2143 while (len-- > 0) {
2144 j = (*s++) - '0';
2145 if (j > 9) return 1; /* invalid char */
2146 if (i > LLONG_MAX / 10LL) return 1; /* check for multiply overflow */
2147 i = i * 10LL;
2148 if (i + j < i) return 1; /* check for addition overflow */
2149 i = i + j;
2150 }
2151 } else {
2152 /* negative number */
2153 s++;
2154 while (--len > 0) {
2155 j = (*s++) - '0';
2156 if (j > 9) return 1; /* invalid char */
2157 if (i < LLONG_MIN / 10LL) return 1; /* check for multiply overflow */
2158 i = i * 10LL;
2159 if (i - j > i) return 1; /* check for subtract overflow */
2160 i = i - j;
2161 }
2162 }
2163 *ret = i;
2164 return 0;
2165}
2166
Thierry FOURNIER511e9472014-01-23 17:40:34 +01002167/* This function is used with pat_parse_dotted_ver(). It converts a string
2168 * composed by two number separated by a dot. Each part must contain in 16 bits
2169 * because internally they will be represented as a 32-bit quantity stored in
2170 * a 64-bit integer. It returns zero when the number has successfully been
2171 * converted, non-zero otherwise. When an error is returned, the <ret> value
2172 * is left untouched.
2173 *
2174 * "1.3" -> 0x0000000000010003
2175 * "65535.65535" -> 0x00000000ffffffff
2176 */
2177int strl2llrc_dotted(const char *text, int len, long long *ret)
2178{
2179 const char *end = &text[len];
2180 const char *p;
2181 long long major, minor;
2182
2183 /* Look for dot. */
2184 for (p = text; p < end; p++)
2185 if (*p == '.')
2186 break;
2187
2188 /* Convert major. */
2189 if (strl2llrc(text, p - text, &major) != 0)
2190 return 1;
2191
2192 /* Check major. */
2193 if (major >= 65536)
2194 return 1;
2195
2196 /* Convert minor. */
2197 minor = 0;
2198 if (p < end)
2199 if (strl2llrc(p + 1, end - (p + 1), &minor) != 0)
2200 return 1;
2201
2202 /* Check minor. */
2203 if (minor >= 65536)
2204 return 1;
2205
2206 /* Compose value. */
2207 *ret = (major << 16) | (minor & 0xffff);
2208 return 0;
2209}
2210
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002211/* This function parses a time value optionally followed by a unit suffix among
2212 * "d", "h", "m", "s", "ms" or "us". It converts the value into the unit
2213 * expected by the caller. The computation does its best to avoid overflows.
2214 * The value is returned in <ret> if everything is fine, and a NULL is returned
2215 * by the function. In case of error, a pointer to the error is returned and
2216 * <ret> is left untouched. Values are automatically rounded up when needed.
Willy Tarreau9faebe32019-06-07 19:00:37 +02002217 * Values resulting in values larger than or equal to 2^31 after conversion are
2218 * reported as an overflow as value PARSE_TIME_OVER. Non-null values resulting
2219 * in an underflow are reported as an underflow as value PARSE_TIME_UNDER.
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002220 */
2221const char *parse_time_err(const char *text, unsigned *ret, unsigned unit_flags)
2222{
Willy Tarreau9faebe32019-06-07 19:00:37 +02002223 unsigned long long imult, idiv;
2224 unsigned long long omult, odiv;
2225 unsigned long long value, result;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002226
2227 omult = odiv = 1;
2228
2229 switch (unit_flags & TIME_UNIT_MASK) {
2230 case TIME_UNIT_US: omult = 1000000; break;
2231 case TIME_UNIT_MS: omult = 1000; break;
2232 case TIME_UNIT_S: break;
2233 case TIME_UNIT_MIN: odiv = 60; break;
2234 case TIME_UNIT_HOUR: odiv = 3600; break;
2235 case TIME_UNIT_DAY: odiv = 86400; break;
2236 default: break;
2237 }
2238
2239 value = 0;
2240
2241 while (1) {
2242 unsigned int j;
2243
2244 j = *text - '0';
2245 if (j > 9)
2246 break;
2247 text++;
2248 value *= 10;
2249 value += j;
2250 }
2251
2252 imult = idiv = 1;
2253 switch (*text) {
2254 case '\0': /* no unit = default unit */
2255 imult = omult = idiv = odiv = 1;
2256 break;
2257 case 's': /* second = unscaled unit */
2258 break;
2259 case 'u': /* microsecond : "us" */
2260 if (text[1] == 's') {
2261 idiv = 1000000;
2262 text++;
2263 }
2264 break;
2265 case 'm': /* millisecond : "ms" or minute: "m" */
2266 if (text[1] == 's') {
2267 idiv = 1000;
2268 text++;
2269 } else
2270 imult = 60;
2271 break;
2272 case 'h': /* hour : "h" */
2273 imult = 3600;
2274 break;
2275 case 'd': /* day : "d" */
2276 imult = 86400;
2277 break;
2278 default:
2279 return text;
2280 break;
2281 }
2282
2283 if (omult % idiv == 0) { omult /= idiv; idiv = 1; }
2284 if (idiv % omult == 0) { idiv /= omult; omult = 1; }
2285 if (imult % odiv == 0) { imult /= odiv; odiv = 1; }
2286 if (odiv % imult == 0) { odiv /= imult; imult = 1; }
2287
Willy Tarreau9faebe32019-06-07 19:00:37 +02002288 result = (value * (imult * omult) + (idiv * odiv - 1)) / (idiv * odiv);
2289 if (result >= 0x80000000)
2290 return PARSE_TIME_OVER;
2291 if (!result && value)
2292 return PARSE_TIME_UNDER;
2293 *ret = result;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002294 return NULL;
2295}
Willy Tarreau6911fa42007-03-04 18:06:08 +01002296
Emeric Brun39132b22010-01-04 14:57:24 +01002297/* this function converts the string starting at <text> to an unsigned int
2298 * stored in <ret>. If an error is detected, the pointer to the unexpected
Joseph Herlant32b83272018-11-15 11:58:28 -08002299 * character is returned. If the conversion is successful, NULL is returned.
Emeric Brun39132b22010-01-04 14:57:24 +01002300 */
2301const char *parse_size_err(const char *text, unsigned *ret) {
2302 unsigned value = 0;
2303
2304 while (1) {
2305 unsigned int j;
2306
2307 j = *text - '0';
2308 if (j > 9)
2309 break;
2310 if (value > ~0U / 10)
2311 return text;
2312 value *= 10;
2313 if (value > (value + j))
2314 return text;
2315 value += j;
2316 text++;
2317 }
2318
2319 switch (*text) {
2320 case '\0':
2321 break;
2322 case 'K':
2323 case 'k':
2324 if (value > ~0U >> 10)
2325 return text;
2326 value = value << 10;
2327 break;
2328 case 'M':
2329 case 'm':
2330 if (value > ~0U >> 20)
2331 return text;
2332 value = value << 20;
2333 break;
2334 case 'G':
2335 case 'g':
2336 if (value > ~0U >> 30)
2337 return text;
2338 value = value << 30;
2339 break;
2340 default:
2341 return text;
2342 }
2343
Godbach58048a22015-01-28 17:36:16 +08002344 if (*text != '\0' && *++text != '\0')
2345 return text;
2346
Emeric Brun39132b22010-01-04 14:57:24 +01002347 *ret = value;
2348 return NULL;
2349}
2350
Willy Tarreau126d4062013-12-03 17:50:47 +01002351/*
2352 * Parse binary string written in hexadecimal (source) and store the decoded
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002353 * result into binstr and set binstrlen to the length of binstr. Memory for
Willy Tarreau126d4062013-12-03 17:50:47 +01002354 * binstr is allocated by the function. In case of error, returns 0 with an
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002355 * error message in err. In success case, it returns the consumed length.
Willy Tarreau126d4062013-12-03 17:50:47 +01002356 */
2357int parse_binary(const char *source, char **binstr, int *binstrlen, char **err)
2358{
2359 int len;
2360 const char *p = source;
2361 int i,j;
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002362 int alloc;
Willy Tarreau126d4062013-12-03 17:50:47 +01002363
2364 len = strlen(source);
2365 if (len % 2) {
2366 memprintf(err, "an even number of hex digit is expected");
2367 return 0;
2368 }
2369
2370 len = len >> 1;
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002371
Willy Tarreau126d4062013-12-03 17:50:47 +01002372 if (!*binstr) {
Tim Duesterhuse52b6e52020-09-12 20:26:43 +02002373 *binstr = calloc(len, sizeof(**binstr));
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002374 if (!*binstr) {
2375 memprintf(err, "out of memory while loading string pattern");
2376 return 0;
2377 }
2378 alloc = 1;
Willy Tarreau126d4062013-12-03 17:50:47 +01002379 }
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002380 else {
2381 if (*binstrlen < len) {
Joseph Herlant76dbe782018-11-15 12:01:22 -08002382 memprintf(err, "no space available in the buffer. expect %d, provides %d",
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002383 len, *binstrlen);
2384 return 0;
2385 }
2386 alloc = 0;
2387 }
2388 *binstrlen = len;
Willy Tarreau126d4062013-12-03 17:50:47 +01002389
2390 i = j = 0;
2391 while (j < len) {
2392 if (!ishex(p[i++]))
2393 goto bad_input;
2394 if (!ishex(p[i++]))
2395 goto bad_input;
2396 (*binstr)[j++] = (hex2i(p[i-2]) << 4) + hex2i(p[i-1]);
2397 }
Thierry FOURNIERee330af2014-01-21 11:36:14 +01002398 return len << 1;
Willy Tarreau126d4062013-12-03 17:50:47 +01002399
2400bad_input:
2401 memprintf(err, "an hex digit is expected (found '%c')", p[i-1]);
Andreas Seltenreich93f91c32016-03-03 20:40:37 +01002402 if (alloc) {
2403 free(*binstr);
2404 *binstr = NULL;
2405 }
Willy Tarreau126d4062013-12-03 17:50:47 +01002406 return 0;
2407}
2408
Willy Tarreau946ba592009-05-10 15:41:18 +02002409/* copies at most <n> characters from <src> and always terminates with '\0' */
2410char *my_strndup(const char *src, int n)
2411{
2412 int len = 0;
2413 char *ret;
2414
2415 while (len < n && src[len])
2416 len++;
2417
Vincent Bernat3c2f2f22016-04-03 13:48:42 +02002418 ret = malloc(len + 1);
Willy Tarreau946ba592009-05-10 15:41:18 +02002419 if (!ret)
2420 return ret;
2421 memcpy(ret, src, len);
2422 ret[len] = '\0';
2423 return ret;
2424}
2425
Baptiste Assmannbb77c8e2013-10-06 23:24:13 +02002426/*
2427 * search needle in haystack
2428 * returns the pointer if found, returns NULL otherwise
2429 */
2430const void *my_memmem(const void *haystack, size_t haystacklen, const void *needle, size_t needlelen)
2431{
2432 const void *c = NULL;
2433 unsigned char f;
2434
2435 if ((haystack == NULL) || (needle == NULL) || (haystacklen < needlelen))
2436 return NULL;
2437
2438 f = *(char *)needle;
2439 c = haystack;
2440 while ((c = memchr(c, f, haystacklen - (c - haystack))) != NULL) {
2441 if ((haystacklen - (c - haystack)) < needlelen)
2442 return NULL;
2443
2444 if (memcmp(c, needle, needlelen) == 0)
2445 return c;
2446 ++c;
2447 }
2448 return NULL;
2449}
2450
Ilya Shipitsinc02a23f2020-05-06 00:53:22 +05002451/* get length of the initial segment consisting entirely of bytes in <accept> */
Christopher Faulet5eb96cb2020-04-15 10:23:01 +02002452size_t my_memspn(const void *str, size_t len, const void *accept, size_t acceptlen)
2453{
2454 size_t ret = 0;
2455
2456 while (ret < len && memchr(accept, *((int *)str), acceptlen)) {
2457 str++;
2458 ret++;
2459 }
2460 return ret;
2461}
2462
Ilya Shipitsinc02a23f2020-05-06 00:53:22 +05002463/* get length of the initial segment consisting entirely of bytes not in <rejcet> */
Christopher Faulet5eb96cb2020-04-15 10:23:01 +02002464size_t my_memcspn(const void *str, size_t len, const void *reject, size_t rejectlen)
2465{
2466 size_t ret = 0;
2467
2468 while (ret < len) {
2469 if(memchr(reject, *((int *)str), rejectlen))
2470 return ret;
2471 str++;
2472 ret++;
2473 }
2474 return ret;
2475}
2476
Willy Tarreau482b00d2009-10-04 22:48:42 +02002477/* This function returns the first unused key greater than or equal to <key> in
2478 * ID tree <root>. Zero is returned if no place is found.
2479 */
2480unsigned int get_next_id(struct eb_root *root, unsigned int key)
2481{
2482 struct eb32_node *used;
2483
2484 do {
2485 used = eb32_lookup_ge(root, key);
2486 if (!used || used->key > key)
2487 return key; /* key is available */
2488 key++;
2489 } while (key);
2490 return key;
2491}
2492
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002493/* dump the full tree to <file> in DOT format for debugging purposes. Will
2494 * optionally highlight node <subj> if found, depending on operation <op> :
2495 * 0 : nothing
2496 * >0 : insertion, node/leaf are surrounded in red
2497 * <0 : removal, node/leaf are dashed with no background
2498 * Will optionally add "desc" as a label on the graph if set and non-null.
2499 */
2500void eb32sc_to_file(FILE *file, struct eb_root *root, const struct eb32sc_node *subj, int op, const char *desc)
Willy Tarreaued3cda02017-11-15 15:04:05 +01002501{
2502 struct eb32sc_node *node;
2503 unsigned long scope = -1;
2504
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002505 fprintf(file, "digraph ebtree {\n");
2506
2507 if (desc && *desc) {
2508 fprintf(file,
2509 " fontname=\"fixed\";\n"
2510 " fontsize=8;\n"
2511 " label=\"%s\";\n", desc);
2512 }
2513
Willy Tarreaued3cda02017-11-15 15:04:05 +01002514 fprintf(file,
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002515 " node [fontname=\"fixed\" fontsize=8 shape=\"box\" style=\"filled\" color=\"black\" fillcolor=\"white\"];\n"
2516 " edge [fontname=\"fixed\" fontsize=8 style=\"solid\" color=\"magenta\" dir=\"forward\"];\n"
Willy Tarreaued3cda02017-11-15 15:04:05 +01002517 " \"%lx_n\" [label=\"root\\n%lx\"]\n", (long)eb_root_to_node(root), (long)root
2518 );
2519
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002520 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002521 (long)eb_root_to_node(root),
2522 (long)eb_root_to_node(eb_clrtag(root->b[0])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002523 eb_gettag(root->b[0]) == EB_LEAF ? 'l' : 'n');
2524
2525 node = eb32sc_first(root, scope);
2526 while (node) {
2527 if (node->node.node_p) {
2528 /* node part is used */
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002529 fprintf(file, " \"%lx_n\" [label=\"%lx\\nkey=%u\\nscope=%lx\\nbit=%d\" fillcolor=\"lightskyblue1\" %s];\n",
2530 (long)node, (long)node, node->key, node->node_s, node->node.bit,
2531 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
Willy Tarreaued3cda02017-11-15 15:04:05 +01002532
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002533 fprintf(file, " \"%lx_n\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002534 (long)node,
2535 (long)eb_root_to_node(eb_clrtag(node->node.node_p)),
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002536 eb_gettag(node->node.node_p) ? 'R' : 'L');
Willy Tarreaued3cda02017-11-15 15:04:05 +01002537
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002538 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002539 (long)node,
2540 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[0])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002541 eb_gettag(node->node.branches.b[0]) == EB_LEAF ? 'l' : 'n');
2542
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002543 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"R\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002544 (long)node,
2545 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[1])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002546 eb_gettag(node->node.branches.b[1]) == EB_LEAF ? 'l' : 'n');
2547 }
2548
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002549 fprintf(file, " \"%lx_l\" [label=\"%lx\\nkey=%u\\nscope=%lx\\npfx=%u\" fillcolor=\"yellow\" %s];\n",
2550 (long)node, (long)node, node->key, node->leaf_s, node->node.pfx,
2551 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
Willy Tarreaued3cda02017-11-15 15:04:05 +01002552
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002553 fprintf(file, " \"%lx_l\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002554 (long)node,
2555 (long)eb_root_to_node(eb_clrtag(node->node.leaf_p)),
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002556 eb_gettag(node->node.leaf_p) ? 'R' : 'L');
Willy Tarreaued3cda02017-11-15 15:04:05 +01002557 node = eb32sc_next(node, scope);
2558 }
2559 fprintf(file, "}\n");
2560}
2561
Willy Tarreau348238b2010-01-18 15:05:57 +01002562/* This function compares a sample word possibly followed by blanks to another
2563 * clean word. The compare is case-insensitive. 1 is returned if both are equal,
2564 * otherwise zero. This intends to be used when checking HTTP headers for some
2565 * values. Note that it validates a word followed only by blanks but does not
2566 * validate a word followed by blanks then other chars.
2567 */
2568int word_match(const char *sample, int slen, const char *word, int wlen)
2569{
2570 if (slen < wlen)
2571 return 0;
2572
2573 while (wlen) {
2574 char c = *sample ^ *word;
2575 if (c && c != ('A' ^ 'a'))
2576 return 0;
2577 sample++;
2578 word++;
2579 slen--;
2580 wlen--;
2581 }
2582
2583 while (slen) {
2584 if (*sample != ' ' && *sample != '\t')
2585 return 0;
2586 sample++;
2587 slen--;
2588 }
2589 return 1;
2590}
Willy Tarreau482b00d2009-10-04 22:48:42 +02002591
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002592/* Converts any text-formatted IPv4 address to a host-order IPv4 address. It
2593 * is particularly fast because it avoids expensive operations such as
2594 * multiplies, which are optimized away at the end. It requires a properly
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002595 * formatted address though (3 points).
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002596 */
2597unsigned int inetaddr_host(const char *text)
2598{
2599 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2600 register unsigned int dig100, dig10, dig1;
2601 int s;
2602 const char *p, *d;
2603
2604 dig1 = dig10 = dig100 = ascii_zero;
2605 s = 24;
2606
2607 p = text;
2608 while (1) {
2609 if (((unsigned)(*p - '0')) <= 9) {
2610 p++;
2611 continue;
2612 }
2613
2614 /* here, we have a complete byte between <text> and <p> (exclusive) */
2615 if (p == text)
2616 goto end;
2617
2618 d = p - 1;
2619 dig1 |= (unsigned int)(*d << s);
2620 if (d == text)
2621 goto end;
2622
2623 d--;
2624 dig10 |= (unsigned int)(*d << s);
2625 if (d == text)
2626 goto end;
2627
2628 d--;
2629 dig100 |= (unsigned int)(*d << s);
2630 end:
2631 if (!s || *p != '.')
2632 break;
2633
2634 s -= 8;
2635 text = ++p;
2636 }
2637
2638 dig100 -= ascii_zero;
2639 dig10 -= ascii_zero;
2640 dig1 -= ascii_zero;
2641 return ((dig100 * 10) + dig10) * 10 + dig1;
2642}
2643
2644/*
2645 * Idem except the first unparsed character has to be passed in <stop>.
2646 */
2647unsigned int inetaddr_host_lim(const char *text, const char *stop)
2648{
2649 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2650 register unsigned int dig100, dig10, dig1;
2651 int s;
2652 const char *p, *d;
2653
2654 dig1 = dig10 = dig100 = ascii_zero;
2655 s = 24;
2656
2657 p = text;
2658 while (1) {
2659 if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2660 p++;
2661 continue;
2662 }
2663
2664 /* here, we have a complete byte between <text> and <p> (exclusive) */
2665 if (p == text)
2666 goto end;
2667
2668 d = p - 1;
2669 dig1 |= (unsigned int)(*d << s);
2670 if (d == text)
2671 goto end;
2672
2673 d--;
2674 dig10 |= (unsigned int)(*d << s);
2675 if (d == text)
2676 goto end;
2677
2678 d--;
2679 dig100 |= (unsigned int)(*d << s);
2680 end:
2681 if (!s || p == stop || *p != '.')
2682 break;
2683
2684 s -= 8;
2685 text = ++p;
2686 }
2687
2688 dig100 -= ascii_zero;
2689 dig10 -= ascii_zero;
2690 dig1 -= ascii_zero;
2691 return ((dig100 * 10) + dig10) * 10 + dig1;
2692}
2693
2694/*
2695 * Idem except the pointer to first unparsed byte is returned into <ret> which
2696 * must not be NULL.
2697 */
Willy Tarreau74172752010-10-15 23:21:42 +02002698unsigned int inetaddr_host_lim_ret(char *text, char *stop, char **ret)
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002699{
2700 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2701 register unsigned int dig100, dig10, dig1;
2702 int s;
Willy Tarreau74172752010-10-15 23:21:42 +02002703 char *p, *d;
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002704
2705 dig1 = dig10 = dig100 = ascii_zero;
2706 s = 24;
2707
2708 p = text;
2709 while (1) {
2710 if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2711 p++;
2712 continue;
2713 }
2714
2715 /* here, we have a complete byte between <text> and <p> (exclusive) */
2716 if (p == text)
2717 goto end;
2718
2719 d = p - 1;
2720 dig1 |= (unsigned int)(*d << s);
2721 if (d == text)
2722 goto end;
2723
2724 d--;
2725 dig10 |= (unsigned int)(*d << s);
2726 if (d == text)
2727 goto end;
2728
2729 d--;
2730 dig100 |= (unsigned int)(*d << s);
2731 end:
2732 if (!s || p == stop || *p != '.')
2733 break;
2734
2735 s -= 8;
2736 text = ++p;
2737 }
2738
2739 *ret = p;
2740 dig100 -= ascii_zero;
2741 dig10 -= ascii_zero;
2742 dig1 -= ascii_zero;
2743 return ((dig100 * 10) + dig10) * 10 + dig1;
2744}
2745
Willy Tarreauf0b38bf2010-06-06 13:22:23 +02002746/* Convert a fixed-length string to an IP address. Returns 0 in case of error,
2747 * or the number of chars read in case of success. Maybe this could be replaced
2748 * by one of the functions above. Also, apparently this function does not support
2749 * hosts above 255 and requires exactly 4 octets.
Willy Tarreau075415a2013-12-12 11:29:39 +01002750 * The destination is only modified on success.
Willy Tarreauf0b38bf2010-06-06 13:22:23 +02002751 */
2752int buf2ip(const char *buf, size_t len, struct in_addr *dst)
2753{
2754 const char *addr;
2755 int saw_digit, octets, ch;
2756 u_char tmp[4], *tp;
2757 const char *cp = buf;
2758
2759 saw_digit = 0;
2760 octets = 0;
2761 *(tp = tmp) = 0;
2762
2763 for (addr = buf; addr - buf < len; addr++) {
2764 unsigned char digit = (ch = *addr) - '0';
2765
2766 if (digit > 9 && ch != '.')
2767 break;
2768
2769 if (digit <= 9) {
2770 u_int new = *tp * 10 + digit;
2771
2772 if (new > 255)
2773 return 0;
2774
2775 *tp = new;
2776
2777 if (!saw_digit) {
2778 if (++octets > 4)
2779 return 0;
2780 saw_digit = 1;
2781 }
2782 } else if (ch == '.' && saw_digit) {
2783 if (octets == 4)
2784 return 0;
2785
2786 *++tp = 0;
2787 saw_digit = 0;
2788 } else
2789 return 0;
2790 }
2791
2792 if (octets < 4)
2793 return 0;
2794
2795 memcpy(&dst->s_addr, tmp, 4);
2796 return addr - cp;
2797}
2798
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002799/* This function converts the string in <buf> of the len <len> to
2800 * struct in6_addr <dst> which must be allocated by the caller.
2801 * This function returns 1 in success case, otherwise zero.
Willy Tarreau075415a2013-12-12 11:29:39 +01002802 * The destination is only modified on success.
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002803 */
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002804int buf2ip6(const char *buf, size_t len, struct in6_addr *dst)
2805{
Thierry FOURNIERcd659912013-12-11 12:33:54 +01002806 char null_term_ip6[INET6_ADDRSTRLEN + 1];
Willy Tarreau075415a2013-12-12 11:29:39 +01002807 struct in6_addr out;
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002808
Thierry FOURNIERcd659912013-12-11 12:33:54 +01002809 if (len > INET6_ADDRSTRLEN)
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002810 return 0;
2811
2812 memcpy(null_term_ip6, buf, len);
2813 null_term_ip6[len] = '\0';
2814
Willy Tarreau075415a2013-12-12 11:29:39 +01002815 if (!inet_pton(AF_INET6, null_term_ip6, &out))
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002816 return 0;
2817
Willy Tarreau075415a2013-12-12 11:29:39 +01002818 *dst = out;
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002819 return 1;
2820}
2821
Willy Tarreauacf95772010-06-14 19:09:21 +02002822/* To be used to quote config arg positions. Returns the short string at <ptr>
2823 * surrounded by simple quotes if <ptr> is valid and non-empty, or "end of line"
2824 * if ptr is NULL or empty. The string is locally allocated.
2825 */
2826const char *quote_arg(const char *ptr)
2827{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002828 static THREAD_LOCAL char val[32];
Willy Tarreauacf95772010-06-14 19:09:21 +02002829 int i;
2830
2831 if (!ptr || !*ptr)
2832 return "end of line";
2833 val[0] = '\'';
Willy Tarreaude2dd6b2013-01-24 02:14:42 +01002834 for (i = 1; i < sizeof(val) - 2 && *ptr; i++)
Willy Tarreauacf95772010-06-14 19:09:21 +02002835 val[i] = *ptr++;
2836 val[i++] = '\'';
2837 val[i] = '\0';
2838 return val;
2839}
2840
Willy Tarreau5b180202010-07-18 10:40:48 +02002841/* returns an operator among STD_OP_* for string <str> or < 0 if unknown */
2842int get_std_op(const char *str)
2843{
2844 int ret = -1;
2845
2846 if (*str == 'e' && str[1] == 'q')
2847 ret = STD_OP_EQ;
2848 else if (*str == 'n' && str[1] == 'e')
2849 ret = STD_OP_NE;
2850 else if (*str == 'l') {
2851 if (str[1] == 'e') ret = STD_OP_LE;
2852 else if (str[1] == 't') ret = STD_OP_LT;
2853 }
2854 else if (*str == 'g') {
2855 if (str[1] == 'e') ret = STD_OP_GE;
2856 else if (str[1] == 't') ret = STD_OP_GT;
2857 }
2858
2859 if (ret == -1 || str[2] != '\0')
2860 return -1;
2861 return ret;
2862}
2863
Willy Tarreau4c14eaa2010-11-24 14:01:45 +01002864/* hash a 32-bit integer to another 32-bit integer */
2865unsigned int full_hash(unsigned int a)
2866{
2867 return __full_hash(a);
2868}
2869
Willy Tarreauf3241112019-02-26 09:56:22 +01002870/* Return the bit position in mask <m> of the nth bit set of rank <r>, between
2871 * 0 and LONGBITS-1 included, starting from the left. For example ranks 0,1,2,3
2872 * for mask 0x55 will be 6, 4, 2 and 0 respectively. This algorithm is based on
2873 * a popcount variant and is described here :
2874 * https://graphics.stanford.edu/~seander/bithacks.html
2875 */
2876unsigned int mask_find_rank_bit(unsigned int r, unsigned long m)
2877{
2878 unsigned long a, b, c, d;
2879 unsigned int s;
2880 unsigned int t;
2881
2882 a = m - ((m >> 1) & ~0UL/3);
2883 b = (a & ~0UL/5) + ((a >> 2) & ~0UL/5);
2884 c = (b + (b >> 4)) & ~0UL/0x11;
2885 d = (c + (c >> 8)) & ~0UL/0x101;
2886
2887 r++; // make r be 1..64
2888
2889 t = 0;
2890 s = LONGBITS;
2891 if (s > 32) {
Willy Tarreau9b6be3b2019-03-18 16:31:18 +01002892 unsigned long d2 = (d >> 16) >> 16;
2893 t = d2 + (d2 >> 16);
Willy Tarreauf3241112019-02-26 09:56:22 +01002894 s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
2895 }
2896
2897 t = (d >> (s - 16)) & 0xff;
2898 s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
2899 t = (c >> (s - 8)) & 0xf;
2900 s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
2901 t = (b >> (s - 4)) & 0x7;
2902 s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
2903 t = (a >> (s - 2)) & 0x3;
2904 s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
2905 t = (m >> (s - 1)) & 0x1;
2906 s -= ((t - r) & 256) >> 8;
2907
2908 return s - 1;
2909}
2910
2911/* Same as mask_find_rank_bit() above but makes use of pre-computed bitmaps
2912 * based on <m>, in <a..d>. These ones must be updated whenever <m> changes
2913 * using mask_prep_rank_map() below.
2914 */
2915unsigned int mask_find_rank_bit_fast(unsigned int r, unsigned long m,
2916 unsigned long a, unsigned long b,
2917 unsigned long c, unsigned long d)
2918{
2919 unsigned int s;
2920 unsigned int t;
2921
2922 r++; // make r be 1..64
2923
2924 t = 0;
2925 s = LONGBITS;
2926 if (s > 32) {
Willy Tarreau9b6be3b2019-03-18 16:31:18 +01002927 unsigned long d2 = (d >> 16) >> 16;
2928 t = d2 + (d2 >> 16);
Willy Tarreauf3241112019-02-26 09:56:22 +01002929 s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
2930 }
2931
2932 t = (d >> (s - 16)) & 0xff;
2933 s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
2934 t = (c >> (s - 8)) & 0xf;
2935 s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
2936 t = (b >> (s - 4)) & 0x7;
2937 s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
2938 t = (a >> (s - 2)) & 0x3;
2939 s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
2940 t = (m >> (s - 1)) & 0x1;
2941 s -= ((t - r) & 256) >> 8;
2942
2943 return s - 1;
2944}
2945
2946/* Prepare the bitmaps used by the fast implementation of the find_rank_bit()
2947 * above.
2948 */
2949void mask_prep_rank_map(unsigned long m,
2950 unsigned long *a, unsigned long *b,
2951 unsigned long *c, unsigned long *d)
2952{
2953 *a = m - ((m >> 1) & ~0UL/3);
2954 *b = (*a & ~0UL/5) + ((*a >> 2) & ~0UL/5);
2955 *c = (*b + (*b >> 4)) & ~0UL/0x11;
2956 *d = (*c + (*c >> 8)) & ~0UL/0x101;
2957}
2958
David du Colombier4f92d322011-03-24 11:09:31 +01002959/* Return non-zero if IPv4 address is part of the network,
Willy Tarreaueec1d382016-07-13 11:59:39 +02002960 * otherwise zero. Note that <addr> may not necessarily be aligned
2961 * while the two other ones must.
David du Colombier4f92d322011-03-24 11:09:31 +01002962 */
Willy Tarreaueec1d382016-07-13 11:59:39 +02002963int in_net_ipv4(const void *addr, const struct in_addr *mask, const struct in_addr *net)
David du Colombier4f92d322011-03-24 11:09:31 +01002964{
Willy Tarreaueec1d382016-07-13 11:59:39 +02002965 struct in_addr addr_copy;
2966
2967 memcpy(&addr_copy, addr, sizeof(addr_copy));
2968 return((addr_copy.s_addr & mask->s_addr) == (net->s_addr & mask->s_addr));
David du Colombier4f92d322011-03-24 11:09:31 +01002969}
2970
2971/* Return non-zero if IPv6 address is part of the network,
Willy Tarreaueec1d382016-07-13 11:59:39 +02002972 * otherwise zero. Note that <addr> may not necessarily be aligned
2973 * while the two other ones must.
David du Colombier4f92d322011-03-24 11:09:31 +01002974 */
Willy Tarreaueec1d382016-07-13 11:59:39 +02002975int in_net_ipv6(const void *addr, const struct in6_addr *mask, const struct in6_addr *net)
David du Colombier4f92d322011-03-24 11:09:31 +01002976{
2977 int i;
Willy Tarreaueec1d382016-07-13 11:59:39 +02002978 struct in6_addr addr_copy;
David du Colombier4f92d322011-03-24 11:09:31 +01002979
Willy Tarreaueec1d382016-07-13 11:59:39 +02002980 memcpy(&addr_copy, addr, sizeof(addr_copy));
David du Colombier4f92d322011-03-24 11:09:31 +01002981 for (i = 0; i < sizeof(struct in6_addr) / sizeof(int); i++)
Willy Tarreaueec1d382016-07-13 11:59:39 +02002982 if (((((int *)&addr_copy)[i] & ((int *)mask)[i])) !=
David du Colombier4f92d322011-03-24 11:09:31 +01002983 (((int *)net)[i] & ((int *)mask)[i]))
2984 return 0;
2985 return 1;
2986}
2987
2988/* RFC 4291 prefix */
2989const char rfc4291_pfx[] = { 0x00, 0x00, 0x00, 0x00,
2990 0x00, 0x00, 0x00, 0x00,
2991 0x00, 0x00, 0xFF, 0xFF };
2992
Joseph Herlant32b83272018-11-15 11:58:28 -08002993/* Map IPv4 address on IPv6 address, as specified in RFC 3513.
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01002994 * Input and output may overlap.
2995 */
David du Colombier4f92d322011-03-24 11:09:31 +01002996void v4tov6(struct in6_addr *sin6_addr, struct in_addr *sin_addr)
2997{
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01002998 struct in_addr tmp_addr;
2999
3000 tmp_addr.s_addr = sin_addr->s_addr;
David du Colombier4f92d322011-03-24 11:09:31 +01003001 memcpy(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx));
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003002 memcpy(sin6_addr->s6_addr+12, &tmp_addr.s_addr, 4);
David du Colombier4f92d322011-03-24 11:09:31 +01003003}
3004
Joseph Herlant32b83272018-11-15 11:58:28 -08003005/* Map IPv6 address on IPv4 address, as specified in RFC 3513.
David du Colombier4f92d322011-03-24 11:09:31 +01003006 * Return true if conversion is possible and false otherwise.
3007 */
3008int v6tov4(struct in_addr *sin_addr, struct in6_addr *sin6_addr)
3009{
3010 if (memcmp(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx)) == 0) {
3011 memcpy(&(sin_addr->s_addr), &(sin6_addr->s6_addr[12]),
3012 sizeof(struct in_addr));
3013 return 1;
3014 }
3015
3016 return 0;
3017}
3018
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003019/* compare two struct sockaddr_storage and return:
3020 * 0 (true) if the addr is the same in both
3021 * 1 (false) if the addr is not the same in both
3022 * -1 (unable) if one of the addr is not AF_INET*
3023 */
3024int ipcmp(struct sockaddr_storage *ss1, struct sockaddr_storage *ss2)
3025{
3026 if ((ss1->ss_family != AF_INET) && (ss1->ss_family != AF_INET6))
3027 return -1;
3028
3029 if ((ss2->ss_family != AF_INET) && (ss2->ss_family != AF_INET6))
3030 return -1;
3031
3032 if (ss1->ss_family != ss2->ss_family)
3033 return 1;
3034
3035 switch (ss1->ss_family) {
3036 case AF_INET:
3037 return memcmp(&((struct sockaddr_in *)ss1)->sin_addr,
3038 &((struct sockaddr_in *)ss2)->sin_addr,
3039 sizeof(struct in_addr)) != 0;
3040 case AF_INET6:
3041 return memcmp(&((struct sockaddr_in6 *)ss1)->sin6_addr,
3042 &((struct sockaddr_in6 *)ss2)->sin6_addr,
3043 sizeof(struct in6_addr)) != 0;
3044 }
3045
3046 return 1;
3047}
3048
Baptiste Assmann08396c82016-01-31 00:27:17 +01003049/* copy IP address from <source> into <dest>
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003050 * The caller must allocate and clear <dest> before calling.
3051 * The source must be in either AF_INET or AF_INET6 family, or the destination
3052 * address will be undefined. If the destination address used to hold a port,
3053 * it is preserved, so that this function can be used to switch to another
3054 * address family with no risk. Returns a pointer to the destination.
Baptiste Assmann08396c82016-01-31 00:27:17 +01003055 */
3056struct sockaddr_storage *ipcpy(struct sockaddr_storage *source, struct sockaddr_storage *dest)
3057{
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003058 int prev_port;
3059
3060 prev_port = get_net_port(dest);
3061 memset(dest, 0, sizeof(*dest));
Baptiste Assmann08396c82016-01-31 00:27:17 +01003062 dest->ss_family = source->ss_family;
3063
3064 /* copy new addr and apply it */
3065 switch (source->ss_family) {
3066 case AF_INET:
3067 ((struct sockaddr_in *)dest)->sin_addr.s_addr = ((struct sockaddr_in *)source)->sin_addr.s_addr;
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003068 ((struct sockaddr_in *)dest)->sin_port = prev_port;
Baptiste Assmann08396c82016-01-31 00:27:17 +01003069 break;
3070 case AF_INET6:
3071 memcpy(((struct sockaddr_in6 *)dest)->sin6_addr.s6_addr, ((struct sockaddr_in6 *)source)->sin6_addr.s6_addr, sizeof(struct in6_addr));
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003072 ((struct sockaddr_in6 *)dest)->sin6_port = prev_port;
Baptiste Assmann08396c82016-01-31 00:27:17 +01003073 break;
3074 }
3075
3076 return dest;
3077}
3078
William Lallemand421f5b52012-02-06 18:15:57 +01003079char *human_time(int t, short hz_div) {
3080 static char rv[sizeof("24855d23h")+1]; // longest of "23h59m" and "59m59s"
3081 char *p = rv;
Willy Tarreau761b3d52014-04-14 14:53:06 +02003082 char *end = rv + sizeof(rv);
William Lallemand421f5b52012-02-06 18:15:57 +01003083 int cnt=2; // print two numbers
3084
3085 if (unlikely(t < 0 || hz_div <= 0)) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003086 snprintf(p, end - p, "?");
William Lallemand421f5b52012-02-06 18:15:57 +01003087 return rv;
3088 }
3089
3090 if (unlikely(hz_div > 1))
3091 t /= hz_div;
3092
3093 if (t >= DAY) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003094 p += snprintf(p, end - p, "%dd", t / DAY);
William Lallemand421f5b52012-02-06 18:15:57 +01003095 cnt--;
3096 }
3097
3098 if (cnt && t % DAY / HOUR) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003099 p += snprintf(p, end - p, "%dh", t % DAY / HOUR);
William Lallemand421f5b52012-02-06 18:15:57 +01003100 cnt--;
3101 }
3102
3103 if (cnt && t % HOUR / MINUTE) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003104 p += snprintf(p, end - p, "%dm", t % HOUR / MINUTE);
William Lallemand421f5b52012-02-06 18:15:57 +01003105 cnt--;
3106 }
3107
3108 if ((cnt && t % MINUTE) || !t) // also display '0s'
Willy Tarreau761b3d52014-04-14 14:53:06 +02003109 p += snprintf(p, end - p, "%ds", t % MINUTE / SEC);
William Lallemand421f5b52012-02-06 18:15:57 +01003110
3111 return rv;
3112}
3113
3114const char *monthname[12] = {
3115 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
3116 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"
3117};
3118
3119/* date2str_log: write a date in the format :
3120 * sprintf(str, "%02d/%s/%04d:%02d:%02d:%02d.%03d",
3121 * tm.tm_mday, monthname[tm.tm_mon], tm.tm_year+1900,
3122 * tm.tm_hour, tm.tm_min, tm.tm_sec, (int)date.tv_usec/1000);
3123 *
3124 * without using sprintf. return a pointer to the last char written (\0) or
3125 * NULL if there isn't enough space.
3126 */
Willy Tarreauf16cb412018-09-04 19:08:48 +02003127char *date2str_log(char *dst, const struct tm *tm, const struct timeval *date, size_t size)
William Lallemand421f5b52012-02-06 18:15:57 +01003128{
3129
3130 if (size < 25) /* the size is fixed: 24 chars + \0 */
3131 return NULL;
3132
3133 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003134 if (!dst)
3135 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003136 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003137
William Lallemand421f5b52012-02-06 18:15:57 +01003138 memcpy(dst, monthname[tm->tm_mon], 3); // month
3139 dst += 3;
3140 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003141
William Lallemand421f5b52012-02-06 18:15:57 +01003142 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003143 if (!dst)
3144 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003145 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003146
William Lallemand421f5b52012-02-06 18:15:57 +01003147 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003148 if (!dst)
3149 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003150 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003151
William Lallemand421f5b52012-02-06 18:15:57 +01003152 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003153 if (!dst)
3154 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003155 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003156
William Lallemand421f5b52012-02-06 18:15:57 +01003157 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003158 if (!dst)
3159 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003160 *dst++ = '.';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003161
Willy Tarreau7d9421d2020-02-29 09:08:02 +01003162 dst = utoa_pad((unsigned int)(date->tv_usec/1000)%1000, dst, 4); // milliseconds
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003163 if (!dst)
3164 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003165 *dst = '\0';
3166
3167 return dst;
3168}
3169
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003170/* Base year used to compute leap years */
3171#define TM_YEAR_BASE 1900
3172
3173/* Return the difference in seconds between two times (leap seconds are ignored).
3174 * Retrieved from glibc 2.18 source code.
3175 */
3176static int my_tm_diff(const struct tm *a, const struct tm *b)
3177{
3178 /* Compute intervening leap days correctly even if year is negative.
3179 * Take care to avoid int overflow in leap day calculations,
3180 * but it's OK to assume that A and B are close to each other.
3181 */
3182 int a4 = (a->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (a->tm_year & 3);
3183 int b4 = (b->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (b->tm_year & 3);
3184 int a100 = a4 / 25 - (a4 % 25 < 0);
3185 int b100 = b4 / 25 - (b4 % 25 < 0);
3186 int a400 = a100 >> 2;
3187 int b400 = b100 >> 2;
3188 int intervening_leap_days = (a4 - b4) - (a100 - b100) + (a400 - b400);
3189 int years = a->tm_year - b->tm_year;
3190 int days = (365 * years + intervening_leap_days
3191 + (a->tm_yday - b->tm_yday));
3192 return (60 * (60 * (24 * days + (a->tm_hour - b->tm_hour))
3193 + (a->tm_min - b->tm_min))
3194 + (a->tm_sec - b->tm_sec));
3195}
3196
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003197/* Return the GMT offset for a specific local time.
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003198 * Both t and tm must represent the same time.
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003199 * The string returned has the same format as returned by strftime(... "%z", tm).
3200 * Offsets are kept in an internal cache for better performances.
3201 */
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003202const char *get_gmt_offset(time_t t, struct tm *tm)
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003203{
3204 /* Cache offsets from GMT (depending on whether DST is active or not) */
Christopher Faulet1bc04c72017-10-29 20:14:08 +01003205 static THREAD_LOCAL char gmt_offsets[2][5+1] = { "", "" };
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003206
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003207 char *gmt_offset;
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003208 struct tm tm_gmt;
3209 int diff;
3210 int isdst = tm->tm_isdst;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003211
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003212 /* Pretend DST not active if its status is unknown */
3213 if (isdst < 0)
3214 isdst = 0;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003215
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003216 /* Fetch the offset and initialize it if needed */
3217 gmt_offset = gmt_offsets[isdst & 0x01];
3218 if (unlikely(!*gmt_offset)) {
3219 get_gmtime(t, &tm_gmt);
3220 diff = my_tm_diff(tm, &tm_gmt);
3221 if (diff < 0) {
3222 diff = -diff;
3223 *gmt_offset = '-';
3224 } else {
3225 *gmt_offset = '+';
3226 }
Willy Tarreaue112c8a2019-10-29 10:16:11 +01003227 diff %= 86400U;
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003228 diff /= 60; /* Convert to minutes */
3229 snprintf(gmt_offset+1, 4+1, "%02d%02d", diff/60, diff%60);
3230 }
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003231
Willy Tarreaue112c8a2019-10-29 10:16:11 +01003232 return gmt_offset;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003233}
3234
William Lallemand421f5b52012-02-06 18:15:57 +01003235/* gmt2str_log: write a date in the format :
3236 * "%02d/%s/%04d:%02d:%02d:%02d +0000" without using snprintf
3237 * return a pointer to the last char written (\0) or
3238 * NULL if there isn't enough space.
3239 */
3240char *gmt2str_log(char *dst, struct tm *tm, size_t size)
3241{
Yuxans Yao4e25b012012-10-19 10:36:09 +08003242 if (size < 27) /* the size is fixed: 26 chars + \0 */
William Lallemand421f5b52012-02-06 18:15:57 +01003243 return NULL;
3244
3245 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003246 if (!dst)
3247 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003248 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003249
William Lallemand421f5b52012-02-06 18:15:57 +01003250 memcpy(dst, monthname[tm->tm_mon], 3); // month
3251 dst += 3;
3252 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003253
William Lallemand421f5b52012-02-06 18:15:57 +01003254 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003255 if (!dst)
3256 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003257 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003258
William Lallemand421f5b52012-02-06 18:15:57 +01003259 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003260 if (!dst)
3261 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003262 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003263
William Lallemand421f5b52012-02-06 18:15:57 +01003264 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003265 if (!dst)
3266 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003267 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003268
William Lallemand421f5b52012-02-06 18:15:57 +01003269 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003270 if (!dst)
3271 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003272 *dst++ = ' ';
3273 *dst++ = '+';
3274 *dst++ = '0';
3275 *dst++ = '0';
3276 *dst++ = '0';
3277 *dst++ = '0';
3278 *dst = '\0';
3279
3280 return dst;
3281}
3282
Yuxans Yao4e25b012012-10-19 10:36:09 +08003283/* localdate2str_log: write a date in the format :
3284 * "%02d/%s/%04d:%02d:%02d:%02d +0000(local timezone)" without using snprintf
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003285 * Both t and tm must represent the same time.
3286 * return a pointer to the last char written (\0) or
3287 * NULL if there isn't enough space.
Yuxans Yao4e25b012012-10-19 10:36:09 +08003288 */
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003289char *localdate2str_log(char *dst, time_t t, struct tm *tm, size_t size)
Yuxans Yao4e25b012012-10-19 10:36:09 +08003290{
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003291 const char *gmt_offset;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003292 if (size < 27) /* the size is fixed: 26 chars + \0 */
3293 return NULL;
3294
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003295 gmt_offset = get_gmt_offset(t, tm);
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003296
Yuxans Yao4e25b012012-10-19 10:36:09 +08003297 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003298 if (!dst)
3299 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003300 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003301
Yuxans Yao4e25b012012-10-19 10:36:09 +08003302 memcpy(dst, monthname[tm->tm_mon], 3); // month
3303 dst += 3;
3304 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003305
Yuxans Yao4e25b012012-10-19 10:36:09 +08003306 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003307 if (!dst)
3308 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003309 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003310
Yuxans Yao4e25b012012-10-19 10:36:09 +08003311 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003312 if (!dst)
3313 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003314 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003315
Yuxans Yao4e25b012012-10-19 10:36:09 +08003316 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003317 if (!dst)
3318 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003319 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003320
Yuxans Yao4e25b012012-10-19 10:36:09 +08003321 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003322 if (!dst)
3323 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003324 *dst++ = ' ';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003325
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003326 memcpy(dst, gmt_offset, 5); // Offset from local time to GMT
Yuxans Yao4e25b012012-10-19 10:36:09 +08003327 dst += 5;
3328 *dst = '\0';
3329
3330 return dst;
3331}
3332
Willy Tarreaucb1949b2017-07-19 19:05:29 +02003333/* Returns the number of seconds since 01/01/1970 0:0:0 GMT for GMT date <tm>.
3334 * It is meant as a portable replacement for timegm() for use with valid inputs.
3335 * Returns undefined results for invalid dates (eg: months out of range 0..11).
3336 */
3337time_t my_timegm(const struct tm *tm)
3338{
3339 /* Each month has 28, 29, 30 or 31 days, or 28+N. The date in the year
3340 * is thus (current month - 1)*28 + cumulated_N[month] to count the
3341 * sum of the extra N days for elapsed months. The sum of all these N
3342 * days doesn't exceed 30 for a complete year (366-12*28) so it fits
3343 * in a 5-bit word. This means that with 60 bits we can represent a
3344 * matrix of all these values at once, which is fast and efficient to
3345 * access. The extra February day for leap years is not counted here.
3346 *
3347 * Jan : none = 0 (0)
3348 * Feb : Jan = 3 (3)
3349 * Mar : Jan..Feb = 3 (3 + 0)
3350 * Apr : Jan..Mar = 6 (3 + 0 + 3)
3351 * May : Jan..Apr = 8 (3 + 0 + 3 + 2)
3352 * Jun : Jan..May = 11 (3 + 0 + 3 + 2 + 3)
3353 * Jul : Jan..Jun = 13 (3 + 0 + 3 + 2 + 3 + 2)
3354 * Aug : Jan..Jul = 16 (3 + 0 + 3 + 2 + 3 + 2 + 3)
3355 * Sep : Jan..Aug = 19 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3)
3356 * Oct : Jan..Sep = 21 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2)
3357 * Nov : Jan..Oct = 24 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3)
3358 * Dec : Jan..Nov = 26 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3 + 2)
3359 */
3360 uint64_t extra =
3361 ( 0ULL << 0*5) + ( 3ULL << 1*5) + ( 3ULL << 2*5) + /* Jan, Feb, Mar, */
3362 ( 6ULL << 3*5) + ( 8ULL << 4*5) + (11ULL << 5*5) + /* Apr, May, Jun, */
3363 (13ULL << 6*5) + (16ULL << 7*5) + (19ULL << 8*5) + /* Jul, Aug, Sep, */
3364 (21ULL << 9*5) + (24ULL << 10*5) + (26ULL << 11*5); /* Oct, Nov, Dec, */
3365
3366 unsigned int y = tm->tm_year + 1900;
3367 unsigned int m = tm->tm_mon;
3368 unsigned long days = 0;
3369
3370 /* days since 1/1/1970 for full years */
3371 days += days_since_zero(y) - days_since_zero(1970);
3372
3373 /* days for full months in the current year */
3374 days += 28 * m + ((extra >> (m * 5)) & 0x1f);
3375
3376 /* count + 1 after March for leap years. A leap year is a year multiple
3377 * of 4, unless it's multiple of 100 without being multiple of 400. 2000
3378 * is leap, 1900 isn't, 1904 is.
3379 */
3380 if ((m > 1) && !(y & 3) && ((y % 100) || !(y % 400)))
3381 days++;
3382
3383 days += tm->tm_mday - 1;
3384 return days * 86400ULL + tm->tm_hour * 3600 + tm->tm_min * 60 + tm->tm_sec;
3385}
3386
Thierry Fournier93127942016-01-20 18:49:45 +01003387/* This function check a char. It returns true and updates
3388 * <date> and <len> pointer to the new position if the
3389 * character is found.
3390 */
3391static inline int parse_expect_char(const char **date, int *len, char c)
3392{
3393 if (*len < 1 || **date != c)
3394 return 0;
3395 (*len)--;
3396 (*date)++;
3397 return 1;
3398}
3399
3400/* This function expects a string <str> of len <l>. It return true and updates.
3401 * <date> and <len> if the string matches, otherwise, it returns false.
3402 */
3403static inline int parse_strcmp(const char **date, int *len, char *str, int l)
3404{
3405 if (*len < l || strncmp(*date, str, l) != 0)
3406 return 0;
3407 (*len) -= l;
3408 (*date) += l;
3409 return 1;
3410}
3411
3412/* This macro converts 3 chars name in integer. */
3413#define STR2I3(__a, __b, __c) ((__a) * 65536 + (__b) * 256 + (__c))
3414
3415/* day-name = %x4D.6F.6E ; "Mon", case-sensitive
3416 * / %x54.75.65 ; "Tue", case-sensitive
3417 * / %x57.65.64 ; "Wed", case-sensitive
3418 * / %x54.68.75 ; "Thu", case-sensitive
3419 * / %x46.72.69 ; "Fri", case-sensitive
3420 * / %x53.61.74 ; "Sat", case-sensitive
3421 * / %x53.75.6E ; "Sun", case-sensitive
3422 *
3423 * This array must be alphabetically sorted
3424 */
3425static inline int parse_http_dayname(const char **date, int *len, struct tm *tm)
3426{
3427 if (*len < 3)
3428 return 0;
3429 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3430 case STR2I3('M','o','n'): tm->tm_wday = 1; break;
3431 case STR2I3('T','u','e'): tm->tm_wday = 2; break;
3432 case STR2I3('W','e','d'): tm->tm_wday = 3; break;
3433 case STR2I3('T','h','u'): tm->tm_wday = 4; break;
3434 case STR2I3('F','r','i'): tm->tm_wday = 5; break;
3435 case STR2I3('S','a','t'): tm->tm_wday = 6; break;
3436 case STR2I3('S','u','n'): tm->tm_wday = 7; break;
3437 default: return 0;
3438 }
3439 *len -= 3;
3440 *date += 3;
3441 return 1;
3442}
3443
3444/* month = %x4A.61.6E ; "Jan", case-sensitive
3445 * / %x46.65.62 ; "Feb", case-sensitive
3446 * / %x4D.61.72 ; "Mar", case-sensitive
3447 * / %x41.70.72 ; "Apr", case-sensitive
3448 * / %x4D.61.79 ; "May", case-sensitive
3449 * / %x4A.75.6E ; "Jun", case-sensitive
3450 * / %x4A.75.6C ; "Jul", case-sensitive
3451 * / %x41.75.67 ; "Aug", case-sensitive
3452 * / %x53.65.70 ; "Sep", case-sensitive
3453 * / %x4F.63.74 ; "Oct", case-sensitive
3454 * / %x4E.6F.76 ; "Nov", case-sensitive
3455 * / %x44.65.63 ; "Dec", case-sensitive
3456 *
3457 * This array must be alphabetically sorted
3458 */
3459static inline int parse_http_monthname(const char **date, int *len, struct tm *tm)
3460{
3461 if (*len < 3)
3462 return 0;
3463 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3464 case STR2I3('J','a','n'): tm->tm_mon = 0; break;
3465 case STR2I3('F','e','b'): tm->tm_mon = 1; break;
3466 case STR2I3('M','a','r'): tm->tm_mon = 2; break;
3467 case STR2I3('A','p','r'): tm->tm_mon = 3; break;
3468 case STR2I3('M','a','y'): tm->tm_mon = 4; break;
3469 case STR2I3('J','u','n'): tm->tm_mon = 5; break;
3470 case STR2I3('J','u','l'): tm->tm_mon = 6; break;
3471 case STR2I3('A','u','g'): tm->tm_mon = 7; break;
3472 case STR2I3('S','e','p'): tm->tm_mon = 8; break;
3473 case STR2I3('O','c','t'): tm->tm_mon = 9; break;
3474 case STR2I3('N','o','v'): tm->tm_mon = 10; break;
3475 case STR2I3('D','e','c'): tm->tm_mon = 11; break;
3476 default: return 0;
3477 }
3478 *len -= 3;
3479 *date += 3;
3480 return 1;
3481}
3482
3483/* day-name-l = %x4D.6F.6E.64.61.79 ; "Monday", case-sensitive
3484 * / %x54.75.65.73.64.61.79 ; "Tuesday", case-sensitive
3485 * / %x57.65.64.6E.65.73.64.61.79 ; "Wednesday", case-sensitive
3486 * / %x54.68.75.72.73.64.61.79 ; "Thursday", case-sensitive
3487 * / %x46.72.69.64.61.79 ; "Friday", case-sensitive
3488 * / %x53.61.74.75.72.64.61.79 ; "Saturday", case-sensitive
3489 * / %x53.75.6E.64.61.79 ; "Sunday", case-sensitive
3490 *
3491 * This array must be alphabetically sorted
3492 */
3493static inline int parse_http_ldayname(const char **date, int *len, struct tm *tm)
3494{
3495 if (*len < 6) /* Minimum length. */
3496 return 0;
3497 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3498 case STR2I3('M','o','n'):
3499 RET0_UNLESS(parse_strcmp(date, len, "Monday", 6));
3500 tm->tm_wday = 1;
3501 return 1;
3502 case STR2I3('T','u','e'):
3503 RET0_UNLESS(parse_strcmp(date, len, "Tuesday", 7));
3504 tm->tm_wday = 2;
3505 return 1;
3506 case STR2I3('W','e','d'):
3507 RET0_UNLESS(parse_strcmp(date, len, "Wednesday", 9));
3508 tm->tm_wday = 3;
3509 return 1;
3510 case STR2I3('T','h','u'):
3511 RET0_UNLESS(parse_strcmp(date, len, "Thursday", 8));
3512 tm->tm_wday = 4;
3513 return 1;
3514 case STR2I3('F','r','i'):
3515 RET0_UNLESS(parse_strcmp(date, len, "Friday", 6));
3516 tm->tm_wday = 5;
3517 return 1;
3518 case STR2I3('S','a','t'):
3519 RET0_UNLESS(parse_strcmp(date, len, "Saturday", 8));
3520 tm->tm_wday = 6;
3521 return 1;
3522 case STR2I3('S','u','n'):
3523 RET0_UNLESS(parse_strcmp(date, len, "Sunday", 6));
3524 tm->tm_wday = 7;
3525 return 1;
3526 }
3527 return 0;
3528}
3529
3530/* This function parses exactly 1 digit and returns the numeric value in "digit". */
3531static inline int parse_digit(const char **date, int *len, int *digit)
3532{
3533 if (*len < 1 || **date < '0' || **date > '9')
3534 return 0;
3535 *digit = (**date - '0');
3536 (*date)++;
3537 (*len)--;
3538 return 1;
3539}
3540
3541/* This function parses exactly 2 digits and returns the numeric value in "digit". */
3542static inline int parse_2digit(const char **date, int *len, int *digit)
3543{
3544 int value;
3545
3546 RET0_UNLESS(parse_digit(date, len, &value));
3547 (*digit) = value * 10;
3548 RET0_UNLESS(parse_digit(date, len, &value));
3549 (*digit) += value;
3550
3551 return 1;
3552}
3553
3554/* This function parses exactly 4 digits and returns the numeric value in "digit". */
3555static inline int parse_4digit(const char **date, int *len, int *digit)
3556{
3557 int value;
3558
3559 RET0_UNLESS(parse_digit(date, len, &value));
3560 (*digit) = value * 1000;
3561
3562 RET0_UNLESS(parse_digit(date, len, &value));
3563 (*digit) += value * 100;
3564
3565 RET0_UNLESS(parse_digit(date, len, &value));
3566 (*digit) += value * 10;
3567
3568 RET0_UNLESS(parse_digit(date, len, &value));
3569 (*digit) += value;
3570
3571 return 1;
3572}
3573
3574/* time-of-day = hour ":" minute ":" second
3575 * ; 00:00:00 - 23:59:60 (leap second)
3576 *
3577 * hour = 2DIGIT
3578 * minute = 2DIGIT
3579 * second = 2DIGIT
3580 */
3581static inline int parse_http_time(const char **date, int *len, struct tm *tm)
3582{
3583 RET0_UNLESS(parse_2digit(date, len, &tm->tm_hour)); /* hour 2DIGIT */
3584 RET0_UNLESS(parse_expect_char(date, len, ':')); /* expect ":" */
3585 RET0_UNLESS(parse_2digit(date, len, &tm->tm_min)); /* min 2DIGIT */
3586 RET0_UNLESS(parse_expect_char(date, len, ':')); /* expect ":" */
3587 RET0_UNLESS(parse_2digit(date, len, &tm->tm_sec)); /* sec 2DIGIT */
3588 return 1;
3589}
3590
3591/* From RFC7231
3592 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3593 *
3594 * IMF-fixdate = day-name "," SP date1 SP time-of-day SP GMT
3595 * ; fixed length/zone/capitalization subset of the format
3596 * ; see Section 3.3 of [RFC5322]
3597 *
3598 *
3599 * date1 = day SP month SP year
3600 * ; e.g., 02 Jun 1982
3601 *
3602 * day = 2DIGIT
3603 * year = 4DIGIT
3604 *
3605 * GMT = %x47.4D.54 ; "GMT", case-sensitive
3606 *
3607 * time-of-day = hour ":" minute ":" second
3608 * ; 00:00:00 - 23:59:60 (leap second)
3609 *
3610 * hour = 2DIGIT
3611 * minute = 2DIGIT
3612 * second = 2DIGIT
3613 *
3614 * DIGIT = decimal 0-9
3615 */
3616int parse_imf_date(const char *date, int len, struct tm *tm)
3617{
David Carlier327298c2016-11-20 10:42:38 +00003618 /* tm_gmtoff, if present, ought to be zero'ed */
3619 memset(tm, 0, sizeof(*tm));
3620
Thierry Fournier93127942016-01-20 18:49:45 +01003621 RET0_UNLESS(parse_http_dayname(&date, &len, tm)); /* day-name */
3622 RET0_UNLESS(parse_expect_char(&date, &len, ',')); /* expect "," */
3623 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3624 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
3625 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3626 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* Month */
3627 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3628 RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
3629 tm->tm_year -= 1900;
3630 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3631 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3632 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3633 RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3)); /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
3634 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003635 return 1;
3636}
3637
3638/* From RFC7231
3639 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3640 *
3641 * rfc850-date = day-name-l "," SP date2 SP time-of-day SP GMT
3642 * date2 = day "-" month "-" 2DIGIT
3643 * ; e.g., 02-Jun-82
3644 *
3645 * day = 2DIGIT
3646 */
3647int parse_rfc850_date(const char *date, int len, struct tm *tm)
3648{
3649 int year;
3650
David Carlier327298c2016-11-20 10:42:38 +00003651 /* tm_gmtoff, if present, ought to be zero'ed */
3652 memset(tm, 0, sizeof(*tm));
3653
Thierry Fournier93127942016-01-20 18:49:45 +01003654 RET0_UNLESS(parse_http_ldayname(&date, &len, tm)); /* Read the day name */
3655 RET0_UNLESS(parse_expect_char(&date, &len, ',')); /* expect "," */
3656 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3657 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
3658 RET0_UNLESS(parse_expect_char(&date, &len, '-')); /* expect "-" */
3659 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* Month */
3660 RET0_UNLESS(parse_expect_char(&date, &len, '-')); /* expect "-" */
3661
3662 /* year = 2DIGIT
3663 *
3664 * Recipients of a timestamp value in rfc850-(*date) format, which uses a
3665 * two-digit year, MUST interpret a timestamp that appears to be more
3666 * than 50 years in the future as representing the most recent year in
3667 * the past that had the same last two digits.
3668 */
3669 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_year));
3670
3671 /* expect SP */
3672 if (!parse_expect_char(&date, &len, ' ')) {
3673 /* Maybe we have the date with 4 digits. */
3674 RET0_UNLESS(parse_2digit(&date, &len, &year));
3675 tm->tm_year = (tm->tm_year * 100 + year) - 1900;
3676 /* expect SP */
3677 RET0_UNLESS(parse_expect_char(&date, &len, ' '));
3678 } else {
3679 /* I fix 60 as pivot: >60: +1900, <60: +2000. Note that the
3680 * tm_year is the number of year since 1900, so for +1900, we
3681 * do nothing, and for +2000, we add 100.
3682 */
3683 if (tm->tm_year <= 60)
3684 tm->tm_year += 100;
3685 }
3686
3687 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3688 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3689 RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3)); /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
3690 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003691
3692 return 1;
3693}
3694
3695/* From RFC7231
3696 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3697 *
3698 * asctime-date = day-name SP date3 SP time-of-day SP year
3699 * date3 = month SP ( 2DIGIT / ( SP 1DIGIT ))
3700 * ; e.g., Jun 2
3701 *
3702 * HTTP-date is case sensitive. A sender MUST NOT generate additional
3703 * whitespace in an HTTP-date beyond that specifically included as SP in
3704 * the grammar.
3705 */
3706int parse_asctime_date(const char *date, int len, struct tm *tm)
3707{
David Carlier327298c2016-11-20 10:42:38 +00003708 /* tm_gmtoff, if present, ought to be zero'ed */
3709 memset(tm, 0, sizeof(*tm));
3710
Thierry Fournier93127942016-01-20 18:49:45 +01003711 RET0_UNLESS(parse_http_dayname(&date, &len, tm)); /* day-name */
3712 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3713 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* expect month */
3714 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3715
3716 /* expect SP and 1DIGIT or 2DIGIT */
3717 if (parse_expect_char(&date, &len, ' '))
3718 RET0_UNLESS(parse_digit(&date, &len, &tm->tm_mday));
3719 else
3720 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday));
3721
3722 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3723 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3724 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3725 RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
3726 tm->tm_year -= 1900;
3727 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003728 return 1;
3729}
3730
3731/* From RFC7231
3732 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3733 *
3734 * HTTP-date = IMF-fixdate / obs-date
3735 * obs-date = rfc850-date / asctime-date
3736 *
3737 * parses an HTTP date in the RFC format and is accepted
3738 * alternatives. <date> is the strinf containing the date,
3739 * len is the len of the string. <tm> is filled with the
3740 * parsed time. We must considers this time as GMT.
3741 */
3742int parse_http_date(const char *date, int len, struct tm *tm)
3743{
3744 if (parse_imf_date(date, len, tm))
3745 return 1;
3746
3747 if (parse_rfc850_date(date, len, tm))
3748 return 1;
3749
3750 if (parse_asctime_date(date, len, tm))
3751 return 1;
3752
3753 return 0;
3754}
3755
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003756/* Dynamically allocates a string of the proper length to hold the formatted
3757 * output. NULL is returned on error. The caller is responsible for freeing the
3758 * memory area using free(). The resulting string is returned in <out> if the
3759 * pointer is not NULL. A previous version of <out> might be used to build the
3760 * new string, and it will be freed before returning if it is not NULL, which
3761 * makes it possible to build complex strings from iterative calls without
3762 * having to care about freeing intermediate values, as in the example below :
3763 *
3764 * memprintf(&err, "invalid argument: '%s'", arg);
3765 * ...
3766 * memprintf(&err, "parser said : <%s>\n", *err);
3767 * ...
3768 * free(*err);
3769 *
3770 * This means that <err> must be initialized to NULL before first invocation.
3771 * The return value also holds the allocated string, which eases error checking
3772 * and immediate consumption. If the output pointer is not used, NULL must be
Willy Tarreaueb6cead2012-09-20 19:43:14 +02003773 * passed instead and it will be ignored. The returned message will then also
3774 * be NULL so that the caller does not have to bother with freeing anything.
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003775 *
3776 * It is also convenient to use it without any free except the last one :
3777 * err = NULL;
3778 * if (!fct1(err)) report(*err);
3779 * if (!fct2(err)) report(*err);
3780 * if (!fct3(err)) report(*err);
3781 * free(*err);
Christopher Faulet93a518f2017-10-24 11:25:33 +02003782 *
3783 * memprintf relies on memvprintf. This last version can be called from any
3784 * function with variadic arguments.
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003785 */
Christopher Faulet93a518f2017-10-24 11:25:33 +02003786char *memvprintf(char **out, const char *format, va_list orig_args)
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003787{
3788 va_list args;
3789 char *ret = NULL;
3790 int allocated = 0;
3791 int needed = 0;
3792
Willy Tarreaueb6cead2012-09-20 19:43:14 +02003793 if (!out)
3794 return NULL;
3795
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003796 do {
Willy Tarreaue0609f52019-03-29 19:13:23 +01003797 char buf1;
3798
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003799 /* vsnprintf() will return the required length even when the
3800 * target buffer is NULL. We do this in a loop just in case
3801 * intermediate evaluations get wrong.
3802 */
Christopher Faulet93a518f2017-10-24 11:25:33 +02003803 va_copy(args, orig_args);
Willy Tarreaue0609f52019-03-29 19:13:23 +01003804 needed = vsnprintf(ret ? ret : &buf1, allocated, format, args);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003805 va_end(args);
Willy Tarreau1b2fed62013-04-01 22:48:54 +02003806 if (needed < allocated) {
3807 /* Note: on Solaris 8, the first iteration always
3808 * returns -1 if allocated is zero, so we force a
3809 * retry.
3810 */
3811 if (!allocated)
3812 needed = 0;
3813 else
3814 break;
3815 }
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003816
Willy Tarreau1b2fed62013-04-01 22:48:54 +02003817 allocated = needed + 1;
Hubert Verstraete831962e2016-06-28 22:44:26 +02003818 ret = my_realloc2(ret, allocated);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003819 } while (ret);
3820
3821 if (needed < 0) {
3822 /* an error was encountered */
3823 free(ret);
3824 ret = NULL;
3825 }
3826
3827 if (out) {
3828 free(*out);
3829 *out = ret;
3830 }
3831
3832 return ret;
3833}
William Lallemand421f5b52012-02-06 18:15:57 +01003834
Christopher Faulet93a518f2017-10-24 11:25:33 +02003835char *memprintf(char **out, const char *format, ...)
3836{
3837 va_list args;
3838 char *ret = NULL;
3839
3840 va_start(args, format);
3841 ret = memvprintf(out, format, args);
3842 va_end(args);
3843
3844 return ret;
3845}
3846
Willy Tarreau21c705b2012-09-14 11:40:36 +02003847/* Used to add <level> spaces before each line of <out>, unless there is only one line.
3848 * The input argument is automatically freed and reassigned. The result will have to be
Willy Tarreau70eec382012-10-10 08:56:47 +02003849 * freed by the caller. It also supports being passed a NULL which results in the same
3850 * output.
Willy Tarreau21c705b2012-09-14 11:40:36 +02003851 * Example of use :
3852 * parse(cmd, &err); (callee: memprintf(&err, ...))
3853 * fprintf(stderr, "Parser said: %s\n", indent_error(&err));
3854 * free(err);
3855 */
3856char *indent_msg(char **out, int level)
3857{
3858 char *ret, *in, *p;
3859 int needed = 0;
3860 int lf = 0;
3861 int lastlf = 0;
3862 int len;
3863
Willy Tarreau70eec382012-10-10 08:56:47 +02003864 if (!out || !*out)
3865 return NULL;
3866
Willy Tarreau21c705b2012-09-14 11:40:36 +02003867 in = *out - 1;
3868 while ((in = strchr(in + 1, '\n')) != NULL) {
3869 lastlf = in - *out;
3870 lf++;
3871 }
3872
3873 if (!lf) /* single line, no LF, return it as-is */
3874 return *out;
3875
3876 len = strlen(*out);
3877
3878 if (lf == 1 && lastlf == len - 1) {
3879 /* single line, LF at end, strip it and return as-is */
3880 (*out)[lastlf] = 0;
3881 return *out;
3882 }
3883
3884 /* OK now we have at least one LF, we need to process the whole string
3885 * as a multi-line string. What we'll do :
3886 * - prefix with an LF if there is none
3887 * - add <level> spaces before each line
3888 * This means at most ( 1 + level + (len-lf) + lf*<1+level) ) =
3889 * 1 + level + len + lf * level = 1 + level * (lf + 1) + len.
3890 */
3891
3892 needed = 1 + level * (lf + 1) + len + 1;
3893 p = ret = malloc(needed);
3894 in = *out;
3895
3896 /* skip initial LFs */
3897 while (*in == '\n')
3898 in++;
3899
3900 /* copy each line, prefixed with LF and <level> spaces, and without the trailing LF */
3901 while (*in) {
3902 *p++ = '\n';
3903 memset(p, ' ', level);
3904 p += level;
3905 do {
3906 *p++ = *in++;
3907 } while (*in && *in != '\n');
3908 if (*in)
3909 in++;
3910 }
3911 *p = 0;
3912
3913 free(*out);
3914 *out = ret;
3915
3916 return ret;
3917}
3918
Willy Tarreaua2c99112019-08-21 13:17:37 +02003919/* makes a copy of message <in> into <out>, with each line prefixed with <pfx>
3920 * and end of lines replaced with <eol> if not 0. The first line to indent has
3921 * to be indicated in <first> (starts at zero), so that it is possible to skip
3922 * indenting the first line if it has to be appended after an existing message.
3923 * Empty strings are never indented, and NULL strings are considered empty both
3924 * for <in> and <pfx>. It returns non-zero if an EOL was appended as the last
3925 * character, non-zero otherwise.
3926 */
3927int append_prefixed_str(struct buffer *out, const char *in, const char *pfx, char eol, int first)
3928{
3929 int bol, lf;
3930 int pfxlen = pfx ? strlen(pfx) : 0;
3931
3932 if (!in)
3933 return 0;
3934
3935 bol = 1;
3936 lf = 0;
3937 while (*in) {
3938 if (bol && pfxlen) {
3939 if (first > 0)
3940 first--;
3941 else
3942 b_putblk(out, pfx, pfxlen);
3943 bol = 0;
3944 }
3945
3946 lf = (*in == '\n');
3947 bol |= lf;
3948 b_putchr(out, (lf && eol) ? eol : *in);
3949 in++;
3950 }
3951 return lf;
3952}
3953
Willy Tarreau9d22e562019-03-29 18:49:09 +01003954/* removes environment variable <name> from the environment as found in
3955 * environ. This is only provided as an alternative for systems without
3956 * unsetenv() (old Solaris and AIX versions). THIS IS NOT THREAD SAFE.
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05003957 * The principle is to scan environ for each occurrence of variable name
Willy Tarreau9d22e562019-03-29 18:49:09 +01003958 * <name> and to replace the matching pointers with the last pointer of
3959 * the array (since variables are not ordered).
3960 * It always returns 0 (success).
3961 */
3962int my_unsetenv(const char *name)
3963{
3964 extern char **environ;
3965 char **p = environ;
3966 int vars;
3967 int next;
3968 int len;
3969
3970 len = strlen(name);
3971 for (vars = 0; p[vars]; vars++)
3972 ;
3973 next = 0;
3974 while (next < vars) {
3975 if (strncmp(p[next], name, len) != 0 || p[next][len] != '=') {
3976 next++;
3977 continue;
3978 }
3979 if (next < vars - 1)
3980 p[next] = p[vars - 1];
3981 p[--vars] = NULL;
3982 }
3983 return 0;
3984}
3985
Willy Tarreaudad36a32013-03-11 01:20:04 +01003986/* Convert occurrences of environment variables in the input string to their
3987 * corresponding value. A variable is identified as a series of alphanumeric
3988 * characters or underscores following a '$' sign. The <in> string must be
3989 * free()able. NULL returns NULL. The resulting string might be reallocated if
3990 * some expansion is made. Variable names may also be enclosed into braces if
3991 * needed (eg: to concatenate alphanum characters).
3992 */
3993char *env_expand(char *in)
3994{
3995 char *txt_beg;
3996 char *out;
3997 char *txt_end;
3998 char *var_beg;
3999 char *var_end;
4000 char *value;
4001 char *next;
4002 int out_len;
4003 int val_len;
4004
4005 if (!in)
4006 return in;
4007
4008 value = out = NULL;
4009 out_len = 0;
4010
4011 txt_beg = in;
4012 do {
4013 /* look for next '$' sign in <in> */
4014 for (txt_end = txt_beg; *txt_end && *txt_end != '$'; txt_end++);
4015
4016 if (!*txt_end && !out) /* end and no expansion performed */
4017 return in;
4018
4019 val_len = 0;
4020 next = txt_end;
4021 if (*txt_end == '$') {
4022 char save;
4023
4024 var_beg = txt_end + 1;
4025 if (*var_beg == '{')
4026 var_beg++;
4027
4028 var_end = var_beg;
Willy Tarreau90807112020-02-25 08:16:33 +01004029 while (isalnum((unsigned char)*var_end) || *var_end == '_') {
Willy Tarreaudad36a32013-03-11 01:20:04 +01004030 var_end++;
4031 }
4032
4033 next = var_end;
4034 if (*var_end == '}' && (var_beg > txt_end + 1))
4035 next++;
4036
4037 /* get value of the variable name at this location */
4038 save = *var_end;
4039 *var_end = '\0';
4040 value = getenv(var_beg);
4041 *var_end = save;
4042 val_len = value ? strlen(value) : 0;
4043 }
4044
Hubert Verstraete831962e2016-06-28 22:44:26 +02004045 out = my_realloc2(out, out_len + (txt_end - txt_beg) + val_len + 1);
Willy Tarreaudad36a32013-03-11 01:20:04 +01004046 if (txt_end > txt_beg) {
4047 memcpy(out + out_len, txt_beg, txt_end - txt_beg);
4048 out_len += txt_end - txt_beg;
4049 }
4050 if (val_len) {
4051 memcpy(out + out_len, value, val_len);
4052 out_len += val_len;
4053 }
4054 out[out_len] = 0;
4055 txt_beg = next;
4056 } while (*txt_beg);
4057
4058 /* here we know that <out> was allocated and that we don't need <in> anymore */
4059 free(in);
4060 return out;
4061}
4062
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004063
4064/* same as strstr() but case-insensitive and with limit length */
4065const char *strnistr(const char *str1, int len_str1, const char *str2, int len_str2)
4066{
4067 char *pptr, *sptr, *start;
Willy Tarreauc8746532014-05-28 23:05:07 +02004068 unsigned int slen, plen;
4069 unsigned int tmp1, tmp2;
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004070
4071 if (str1 == NULL || len_str1 == 0) // search pattern into an empty string => search is not found
4072 return NULL;
4073
4074 if (str2 == NULL || len_str2 == 0) // pattern is empty => every str1 match
4075 return str1;
4076
4077 if (len_str1 < len_str2) // pattern is longer than string => search is not found
4078 return NULL;
4079
4080 for (tmp1 = 0, start = (char *)str1, pptr = (char *)str2, slen = len_str1, plen = len_str2; slen >= plen; start++, slen--) {
Willy Tarreauf278eec2020-07-05 21:46:32 +02004081 while (toupper((unsigned char)*start) != toupper((unsigned char)*str2)) {
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004082 start++;
4083 slen--;
4084 tmp1++;
4085
4086 if (tmp1 >= len_str1)
4087 return NULL;
4088
4089 /* if pattern longer than string */
4090 if (slen < plen)
4091 return NULL;
4092 }
4093
4094 sptr = start;
4095 pptr = (char *)str2;
4096
4097 tmp2 = 0;
Willy Tarreauf278eec2020-07-05 21:46:32 +02004098 while (toupper((unsigned char)*sptr) == toupper((unsigned char)*pptr)) {
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004099 sptr++;
4100 pptr++;
4101 tmp2++;
4102
4103 if (*pptr == '\0' || tmp2 == len_str2) /* end of pattern found */
4104 return start;
4105 if (*sptr == '\0' || tmp2 == len_str1) /* end of string found and the pattern is not fully found */
4106 return NULL;
4107 }
4108 }
4109 return NULL;
4110}
4111
Thierry FOURNIER317e1c42014-08-12 10:20:47 +02004112/* This function read the next valid utf8 char.
4113 * <s> is the byte srray to be decode, <len> is its length.
4114 * The function returns decoded char encoded like this:
4115 * The 4 msb are the return code (UTF8_CODE_*), the 4 lsb
4116 * are the length read. The decoded character is stored in <c>.
4117 */
4118unsigned char utf8_next(const char *s, int len, unsigned int *c)
4119{
4120 const unsigned char *p = (unsigned char *)s;
4121 int dec;
4122 unsigned char code = UTF8_CODE_OK;
4123
4124 if (len < 1)
4125 return UTF8_CODE_OK;
4126
4127 /* Check the type of UTF8 sequence
4128 *
4129 * 0... .... 0x00 <= x <= 0x7f : 1 byte: ascii char
4130 * 10.. .... 0x80 <= x <= 0xbf : invalid sequence
4131 * 110. .... 0xc0 <= x <= 0xdf : 2 bytes
4132 * 1110 .... 0xe0 <= x <= 0xef : 3 bytes
4133 * 1111 0... 0xf0 <= x <= 0xf7 : 4 bytes
4134 * 1111 10.. 0xf8 <= x <= 0xfb : 5 bytes
4135 * 1111 110. 0xfc <= x <= 0xfd : 6 bytes
4136 * 1111 111. 0xfe <= x <= 0xff : invalid sequence
4137 */
4138 switch (*p) {
4139 case 0x00 ... 0x7f:
4140 *c = *p;
4141 return UTF8_CODE_OK | 1;
4142
4143 case 0x80 ... 0xbf:
4144 *c = *p;
4145 return UTF8_CODE_BADSEQ | 1;
4146
4147 case 0xc0 ... 0xdf:
4148 if (len < 2) {
4149 *c = *p;
4150 return UTF8_CODE_BADSEQ | 1;
4151 }
4152 *c = *p & 0x1f;
4153 dec = 1;
4154 break;
4155
4156 case 0xe0 ... 0xef:
4157 if (len < 3) {
4158 *c = *p;
4159 return UTF8_CODE_BADSEQ | 1;
4160 }
4161 *c = *p & 0x0f;
4162 dec = 2;
4163 break;
4164
4165 case 0xf0 ... 0xf7:
4166 if (len < 4) {
4167 *c = *p;
4168 return UTF8_CODE_BADSEQ | 1;
4169 }
4170 *c = *p & 0x07;
4171 dec = 3;
4172 break;
4173
4174 case 0xf8 ... 0xfb:
4175 if (len < 5) {
4176 *c = *p;
4177 return UTF8_CODE_BADSEQ | 1;
4178 }
4179 *c = *p & 0x03;
4180 dec = 4;
4181 break;
4182
4183 case 0xfc ... 0xfd:
4184 if (len < 6) {
4185 *c = *p;
4186 return UTF8_CODE_BADSEQ | 1;
4187 }
4188 *c = *p & 0x01;
4189 dec = 5;
4190 break;
4191
4192 case 0xfe ... 0xff:
4193 default:
4194 *c = *p;
4195 return UTF8_CODE_BADSEQ | 1;
4196 }
4197
4198 p++;
4199
4200 while (dec > 0) {
4201
4202 /* need 0x10 for the 2 first bits */
4203 if ( ( *p & 0xc0 ) != 0x80 )
4204 return UTF8_CODE_BADSEQ | ((p-(unsigned char *)s)&0xffff);
4205
4206 /* add data at char */
4207 *c = ( *c << 6 ) | ( *p & 0x3f );
4208
4209 dec--;
4210 p++;
4211 }
4212
4213 /* Check ovelong encoding.
4214 * 1 byte : 5 + 6 : 11 : 0x80 ... 0x7ff
4215 * 2 bytes : 4 + 6 + 6 : 16 : 0x800 ... 0xffff
4216 * 3 bytes : 3 + 6 + 6 + 6 : 21 : 0x10000 ... 0x1fffff
4217 */
Thierry FOURNIER9e7ec082015-03-12 19:32:38 +01004218 if (( *c <= 0x7f && (p-(unsigned char *)s) > 1) ||
Thierry FOURNIER317e1c42014-08-12 10:20:47 +02004219 (*c >= 0x80 && *c <= 0x7ff && (p-(unsigned char *)s) > 2) ||
4220 (*c >= 0x800 && *c <= 0xffff && (p-(unsigned char *)s) > 3) ||
4221 (*c >= 0x10000 && *c <= 0x1fffff && (p-(unsigned char *)s) > 4))
4222 code |= UTF8_CODE_OVERLONG;
4223
4224 /* Check invalid UTF8 range. */
4225 if ((*c >= 0xd800 && *c <= 0xdfff) ||
4226 (*c >= 0xfffe && *c <= 0xffff))
4227 code |= UTF8_CODE_INVRANGE;
4228
4229 return code | ((p-(unsigned char *)s)&0x0f);
4230}
4231
Maxime de Roucydc887852016-05-13 23:52:54 +02004232/* append a copy of string <str> (in a wordlist) at the end of the list <li>
4233 * On failure : return 0 and <err> filled with an error message.
4234 * The caller is responsible for freeing the <err> and <str> copy
4235 * memory area using free()
4236 */
4237int list_append_word(struct list *li, const char *str, char **err)
4238{
4239 struct wordlist *wl;
4240
4241 wl = calloc(1, sizeof(*wl));
4242 if (!wl) {
4243 memprintf(err, "out of memory");
4244 goto fail_wl;
4245 }
4246
4247 wl->s = strdup(str);
4248 if (!wl->s) {
4249 memprintf(err, "out of memory");
4250 goto fail_wl_s;
4251 }
4252
4253 LIST_ADDQ(li, &wl->list);
4254
4255 return 1;
4256
4257fail_wl_s:
4258 free(wl->s);
4259fail_wl:
4260 free(wl);
4261 return 0;
4262}
4263
Willy Tarreau37101052019-05-20 16:48:20 +02004264/* indicates if a memory location may safely be read or not. The trick consists
4265 * in performing a harmless syscall using this location as an input and letting
4266 * the operating system report whether it's OK or not. For this we have the
4267 * stat() syscall, which will return EFAULT when the memory location supposed
4268 * to contain the file name is not readable. If it is readable it will then
4269 * either return 0 if the area contains an existing file name, or -1 with
4270 * another code. This must not be abused, and some audit systems might detect
4271 * this as abnormal activity. It's used only for unsafe dumps.
4272 */
4273int may_access(const void *ptr)
4274{
4275 struct stat buf;
4276
4277 if (stat(ptr, &buf) == 0)
4278 return 1;
4279 if (errno == EFAULT)
4280 return 0;
4281 return 1;
4282}
4283
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004284/* print a string of text buffer to <out>. The format is :
4285 * Non-printable chars \t, \n, \r and \e are * encoded in C format.
4286 * Other non-printable chars are encoded "\xHH". Space, '\', and '=' are also escaped.
4287 * Print stopped if null char or <bsize> is reached, or if no more place in the chunk.
4288 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004289int dump_text(struct buffer *out, const char *buf, int bsize)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004290{
4291 unsigned char c;
4292 int ptr = 0;
4293
4294 while (buf[ptr] && ptr < bsize) {
4295 c = buf[ptr];
Willy Tarreau90807112020-02-25 08:16:33 +01004296 if (isprint((unsigned char)c) && isascii((unsigned char)c) && c != '\\' && c != ' ' && c != '=') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004297 if (out->data > out->size - 1)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004298 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004299 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004300 }
4301 else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\' || c == ' ' || c == '=') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004302 if (out->data > out->size - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004303 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004304 out->area[out->data++] = '\\';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004305 switch (c) {
4306 case ' ': c = ' '; break;
4307 case '\t': c = 't'; break;
4308 case '\n': c = 'n'; break;
4309 case '\r': c = 'r'; break;
4310 case '\e': c = 'e'; break;
4311 case '\\': c = '\\'; break;
4312 case '=': c = '='; break;
4313 }
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004314 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004315 }
4316 else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004317 if (out->data > out->size - 4)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004318 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004319 out->area[out->data++] = '\\';
4320 out->area[out->data++] = 'x';
4321 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4322 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004323 }
4324 ptr++;
4325 }
4326
4327 return ptr;
4328}
4329
4330/* print a buffer in hexa.
4331 * Print stopped if <bsize> is reached, or if no more place in the chunk.
4332 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004333int dump_binary(struct buffer *out, const char *buf, int bsize)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004334{
4335 unsigned char c;
4336 int ptr = 0;
4337
4338 while (ptr < bsize) {
4339 c = buf[ptr];
4340
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004341 if (out->data > out->size - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004342 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004343 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4344 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004345
4346 ptr++;
4347 }
4348 return ptr;
4349}
4350
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004351/* Appends into buffer <out> a hex dump of memory area <buf> for <len> bytes,
4352 * prepending each line with prefix <pfx>. The output is *not* initialized.
4353 * The output will not wrap pas the buffer's end so it is more optimal if the
4354 * caller makes sure the buffer is aligned first. A trailing zero will always
4355 * be appended (and not counted) if there is room for it. The caller must make
Willy Tarreau37101052019-05-20 16:48:20 +02004356 * sure that the area is dumpable first. If <unsafe> is non-null, the memory
4357 * locations are checked first for being readable.
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004358 */
Willy Tarreau37101052019-05-20 16:48:20 +02004359void dump_hex(struct buffer *out, const char *pfx, const void *buf, int len, int unsafe)
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004360{
4361 const unsigned char *d = buf;
4362 int i, j, start;
4363
4364 d = (const unsigned char *)(((unsigned long)buf) & -16);
4365 start = ((unsigned long)buf) & 15;
4366
4367 for (i = 0; i < start + len; i += 16) {
4368 chunk_appendf(out, (sizeof(void *) == 4) ? "%s%8p: " : "%s%16p: ", pfx, d + i);
4369
Willy Tarreau37101052019-05-20 16:48:20 +02004370 // 0: unchecked, 1: checked safe, 2: danger
4371 unsafe = !!unsafe;
4372 if (unsafe && !may_access(d + i))
4373 unsafe = 2;
4374
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004375 for (j = 0; j < 16; j++) {
Willy Tarreau37101052019-05-20 16:48:20 +02004376 if ((i + j < start) || (i + j >= start + len))
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004377 chunk_strcat(out, "'' ");
Willy Tarreau37101052019-05-20 16:48:20 +02004378 else if (unsafe > 1)
4379 chunk_strcat(out, "** ");
4380 else
4381 chunk_appendf(out, "%02x ", d[i + j]);
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004382
4383 if (j == 7)
4384 chunk_strcat(out, "- ");
4385 }
4386 chunk_strcat(out, " ");
4387 for (j = 0; j < 16; j++) {
Willy Tarreau37101052019-05-20 16:48:20 +02004388 if ((i + j < start) || (i + j >= start + len))
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004389 chunk_strcat(out, "'");
Willy Tarreau37101052019-05-20 16:48:20 +02004390 else if (unsafe > 1)
4391 chunk_strcat(out, "*");
Willy Tarreau90807112020-02-25 08:16:33 +01004392 else if (isprint((unsigned char)d[i + j]))
Willy Tarreau37101052019-05-20 16:48:20 +02004393 chunk_appendf(out, "%c", d[i + j]);
4394 else
4395 chunk_strcat(out, ".");
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004396 }
4397 chunk_strcat(out, "\n");
4398 }
4399}
4400
Willy Tarreau762fb3e2020-03-03 15:57:10 +01004401/* dumps <pfx> followed by <n> bytes from <addr> in hex form into buffer <buf>
4402 * enclosed in brackets after the address itself, formatted on 14 chars
4403 * including the "0x" prefix. This is meant to be used as a prefix for code
4404 * areas. For example:
4405 * "0x7f10b6557690 [48 c7 c0 0f 00 00 00 0f]"
4406 * It relies on may_access() to know if the bytes are dumpable, otherwise "--"
4407 * is emitted. A NULL <pfx> will be considered empty.
4408 */
4409void dump_addr_and_bytes(struct buffer *buf, const char *pfx, const void *addr, int n)
4410{
4411 int ok = 0;
4412 int i;
4413
4414 chunk_appendf(buf, "%s%#14lx [", pfx ? pfx : "", (long)addr);
4415
4416 for (i = 0; i < n; i++) {
4417 if (i == 0 || (((long)(addr + i) ^ (long)(addr)) & 4096))
4418 ok = may_access(addr + i);
4419 if (ok)
4420 chunk_appendf(buf, "%02x%s", ((uint8_t*)addr)[i], (i<n-1) ? " " : "]");
4421 else
4422 chunk_appendf(buf, "--%s", (i<n-1) ? " " : "]");
4423 }
4424}
4425
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004426/* print a line of text buffer (limited to 70 bytes) to <out>. The format is :
4427 * <2 spaces> <offset=5 digits> <space or plus> <space> <70 chars max> <\n>
4428 * which is 60 chars per line. Non-printable chars \t, \n, \r and \e are
4429 * encoded in C format. Other non-printable chars are encoded "\xHH". Original
4430 * lines are respected within the limit of 70 output chars. Lines that are
4431 * continuation of a previous truncated line begin with "+" instead of " "
4432 * after the offset. The new pointer is returned.
4433 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004434int dump_text_line(struct buffer *out, const char *buf, int bsize, int len,
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004435 int *line, int ptr)
4436{
4437 int end;
4438 unsigned char c;
4439
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004440 end = out->data + 80;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004441 if (end > out->size)
4442 return ptr;
4443
4444 chunk_appendf(out, " %05d%c ", ptr, (ptr == *line) ? ' ' : '+');
4445
4446 while (ptr < len && ptr < bsize) {
4447 c = buf[ptr];
Willy Tarreau90807112020-02-25 08:16:33 +01004448 if (isprint((unsigned char)c) && isascii((unsigned char)c) && c != '\\') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004449 if (out->data > end - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004450 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004451 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004452 } else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004453 if (out->data > end - 3)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004454 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004455 out->area[out->data++] = '\\';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004456 switch (c) {
4457 case '\t': c = 't'; break;
4458 case '\n': c = 'n'; break;
4459 case '\r': c = 'r'; break;
4460 case '\e': c = 'e'; break;
4461 case '\\': c = '\\'; break;
4462 }
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004463 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004464 } else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004465 if (out->data > end - 5)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004466 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004467 out->area[out->data++] = '\\';
4468 out->area[out->data++] = 'x';
4469 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4470 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004471 }
4472 if (buf[ptr++] == '\n') {
4473 /* we had a line break, let's return now */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004474 out->area[out->data++] = '\n';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004475 *line = ptr;
4476 return ptr;
4477 }
4478 }
4479 /* we have an incomplete line, we return it as-is */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004480 out->area[out->data++] = '\n';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004481 return ptr;
4482}
4483
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004484/* displays a <len> long memory block at <buf>, assuming first byte of <buf>
Willy Tarreaued936c52017-04-27 18:03:20 +02004485 * has address <baseaddr>. String <pfx> may be placed as a prefix in front of
4486 * each line. It may be NULL if unused. The output is emitted to file <out>.
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004487 */
Willy Tarreaued936c52017-04-27 18:03:20 +02004488void debug_hexdump(FILE *out, const char *pfx, const char *buf,
4489 unsigned int baseaddr, int len)
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004490{
Willy Tarreau73459792017-04-11 07:58:08 +02004491 unsigned int i;
4492 int b, j;
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004493
4494 for (i = 0; i < (len + (baseaddr & 15)); i += 16) {
4495 b = i - (baseaddr & 15);
Willy Tarreaued936c52017-04-27 18:03:20 +02004496 fprintf(out, "%s%08x: ", pfx ? pfx : "", i + (baseaddr & ~15));
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004497 for (j = 0; j < 8; j++) {
4498 if (b + j >= 0 && b + j < len)
4499 fprintf(out, "%02x ", (unsigned char)buf[b + j]);
4500 else
4501 fprintf(out, " ");
4502 }
4503
4504 if (b + j >= 0 && b + j < len)
4505 fputc('-', out);
4506 else
4507 fputc(' ', out);
4508
4509 for (j = 8; j < 16; j++) {
4510 if (b + j >= 0 && b + j < len)
4511 fprintf(out, " %02x", (unsigned char)buf[b + j]);
4512 else
4513 fprintf(out, " ");
4514 }
4515
4516 fprintf(out, " ");
4517 for (j = 0; j < 16; j++) {
4518 if (b + j >= 0 && b + j < len) {
4519 if (isprint((unsigned char)buf[b + j]))
4520 fputc((unsigned char)buf[b + j], out);
4521 else
4522 fputc('.', out);
4523 }
4524 else
4525 fputc(' ', out);
4526 }
4527 fputc('\n', out);
4528 }
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004529}
4530
Willy Tarreaubb869862020-04-16 10:52:41 +02004531/* Tries to report the executable path name on platforms supporting this. If
4532 * not found or not possible, returns NULL.
4533 */
4534const char *get_exec_path()
4535{
4536 const char *ret = NULL;
4537
4538#if (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 16))
4539 long execfn = getauxval(AT_EXECFN);
4540
4541 if (execfn && execfn != ENOENT)
4542 ret = (const char *)execfn;
4543#endif
4544 return ret;
4545}
4546
Baruch Siache1651b22020-07-24 07:52:20 +03004547#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreau9133e482020-03-04 10:19:36 +01004548/* calls dladdr() or dladdr1() on <addr> and <dli>. If dladdr1 is available,
4549 * also returns the symbol size in <size>, otherwise returns 0 there.
4550 */
4551static int dladdr_and_size(const void *addr, Dl_info *dli, size_t *size)
4552{
4553 int ret;
Willy Tarreau62af9c82020-03-10 07:51:48 +01004554#if (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 3)) // most detailed one
Willy Tarreau9133e482020-03-04 10:19:36 +01004555 const ElfW(Sym) *sym;
4556
4557 ret = dladdr1(addr, dli, (void **)&sym, RTLD_DL_SYMENT);
4558 if (ret)
4559 *size = sym ? sym->st_size : 0;
4560#else
4561 ret = dladdr(addr, dli);
4562 *size = 0;
4563#endif
4564 return ret;
4565}
4566#endif
4567
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004568/* Tries to append to buffer <buf> some indications about the symbol at address
4569 * <addr> using the following form:
4570 * lib:+0xoffset (unresolvable address from lib's base)
4571 * main+0xoffset (unresolvable address from main (+/-))
4572 * lib:main+0xoffset (unresolvable lib address from main (+/-))
4573 * name (resolved exact exec address)
4574 * lib:name (resolved exact lib address)
4575 * name+0xoffset/0xsize (resolved address within exec symbol)
4576 * lib:name+0xoffset/0xsize (resolved address within lib symbol)
4577 *
4578 * The file name (lib or executable) is limited to what lies between the last
4579 * '/' and the first following '.'. An optional prefix <pfx> is prepended before
4580 * the output if not null. The file is not dumped when it's the same as the one
Baruch Siache1651b22020-07-24 07:52:20 +03004581 * that contains the "main" symbol, or when __ELF__ && USE_DL are not set.
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004582 *
4583 * The symbol's base address is returned, or NULL when unresolved, in order to
4584 * allow the caller to match it against known ones.
4585 */
Willy Tarreau0c439d82020-07-05 20:26:04 +02004586const void *resolve_sym_name(struct buffer *buf, const char *pfx, void *addr)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004587{
4588 const struct {
4589 const void *func;
4590 const char *name;
4591 } fcts[] = {
4592 { .func = process_stream, .name = "process_stream" },
4593 { .func = task_run_applet, .name = "task_run_applet" },
4594 { .func = si_cs_io_cb, .name = "si_cs_io_cb" },
4595 { .func = conn_fd_handler, .name = "conn_fd_handler" },
4596 { .func = dgram_fd_handler, .name = "dgram_fd_handler" },
4597 { .func = listener_accept, .name = "listener_accept" },
4598 { .func = poller_pipe_io_handler, .name = "poller_pipe_io_handler" },
4599 { .func = mworker_accept_wrapper, .name = "mworker_accept_wrapper" },
4600#ifdef USE_LUA
4601 { .func = hlua_process_task, .name = "hlua_process_task" },
4602#endif
4603#if defined(USE_OPENSSL) && (HA_OPENSSL_VERSION_NUMBER >= 0x1010000fL) && !defined(OPENSSL_NO_ASYNC)
4604 { .func = ssl_async_fd_free, .name = "ssl_async_fd_free" },
4605 { .func = ssl_async_fd_handler, .name = "ssl_async_fd_handler" },
4606#endif
4607 };
4608
Baruch Siache1651b22020-07-24 07:52:20 +03004609#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004610 Dl_info dli, dli_main;
Willy Tarreau9133e482020-03-04 10:19:36 +01004611 size_t size;
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004612 const char *fname, *p;
4613#endif
4614 int i;
4615
4616 if (pfx)
4617 chunk_appendf(buf, "%s", pfx);
4618
4619 for (i = 0; i < sizeof(fcts) / sizeof(fcts[0]); i++) {
4620 if (addr == fcts[i].func) {
4621 chunk_appendf(buf, "%s", fcts[i].name);
4622 return addr;
4623 }
4624 }
4625
Baruch Siache1651b22020-07-24 07:52:20 +03004626#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004627 /* Now let's try to be smarter */
Willy Tarreau9133e482020-03-04 10:19:36 +01004628 if (!dladdr_and_size(addr, &dli, &size))
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004629 goto unknown;
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004630
4631 /* 1. prefix the library name if it's not the same object as the one
4632 * that contains the main function. The name is picked between last '/'
4633 * and first following '.'.
4634 */
4635 if (!dladdr(main, &dli_main))
4636 dli_main.dli_fbase = NULL;
4637
4638 if (dli_main.dli_fbase != dli.dli_fbase) {
4639 fname = dli.dli_fname;
4640 p = strrchr(fname, '/');
4641 if (p++)
4642 fname = p;
4643 p = strchr(fname, '.');
4644 if (!p)
4645 p = fname + strlen(fname);
4646
4647 chunk_appendf(buf, "%.*s:", (int)(long)(p - fname), fname);
4648 }
4649
4650 /* 2. symbol name */
4651 if (dli.dli_sname) {
4652 /* known, dump it and return symbol's address (exact or relative) */
4653 chunk_appendf(buf, "%s", dli.dli_sname);
4654 if (addr != dli.dli_saddr) {
4655 chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_saddr));
Willy Tarreau9133e482020-03-04 10:19:36 +01004656 if (size)
4657 chunk_appendf(buf, "/%#lx", (long)size);
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004658 }
4659 return dli.dli_saddr;
4660 }
4661 else if (dli_main.dli_fbase != dli.dli_fbase) {
4662 /* unresolved symbol from a known library, report relative offset */
4663 chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_fbase));
4664 return NULL;
4665 }
Baruch Siache1651b22020-07-24 07:52:20 +03004666#endif /* __ELF__ && !__linux__ || USE_DL */
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004667 unknown:
4668 /* unresolved symbol from the main file, report relative offset to main */
4669 if ((void*)addr < (void*)main)
4670 chunk_appendf(buf, "main-%#lx", (long)((void*)main - addr));
4671 else
4672 chunk_appendf(buf, "main+%#lx", (long)(addr - (void*)main));
4673 return NULL;
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004674}
4675
Frédéric Lécaille3b717162019-02-25 15:04:22 +01004676/*
4677 * Allocate an array of unsigned int with <nums> as address from <str> string
Ilya Shipitsin46a030c2020-07-05 16:36:08 +05004678 * made of integer separated by dot characters.
Frédéric Lécaille3b717162019-02-25 15:04:22 +01004679 *
4680 * First, initializes the value with <sz> as address to 0 and initializes the
4681 * array with <nums> as address to NULL. Then allocates the array with <nums> as
4682 * address updating <sz> pointed value to the size of this array.
4683 *
4684 * Returns 1 if succeeded, 0 if not.
4685 */
4686int parse_dotted_uints(const char *str, unsigned int **nums, size_t *sz)
4687{
4688 unsigned int *n;
4689 const char *s, *end;
4690
4691 s = str;
4692 *sz = 0;
4693 end = str + strlen(str);
4694 *nums = n = NULL;
4695
4696 while (1) {
4697 unsigned int r;
4698
4699 if (s >= end)
4700 break;
4701
4702 r = read_uint(&s, end);
4703 /* Expected characters after having read an uint: '\0' or '.',
4704 * if '.', must not be terminal.
4705 */
4706 if (*s != '\0'&& (*s++ != '.' || s == end))
4707 return 0;
4708
Frédéric Lécaille12a71842019-02-26 18:19:48 +01004709 n = my_realloc2(n, (*sz + 1) * sizeof *n);
Frédéric Lécaille3b717162019-02-25 15:04:22 +01004710 if (!n)
4711 return 0;
4712
4713 n[(*sz)++] = r;
4714 }
4715 *nums = n;
4716
4717 return 1;
4718}
4719
Willy Tarreau4d589e72019-08-23 19:02:26 +02004720
4721/* returns the number of bytes needed to encode <v> as a varint. An inline
4722 * version exists for use with constants (__varint_bytes()).
4723 */
4724int varint_bytes(uint64_t v)
4725{
4726 int len = 1;
4727
4728 if (v >= 240) {
4729 v = (v - 240) >> 4;
4730 while (1) {
4731 len++;
4732 if (v < 128)
4733 break;
4734 v = (v - 128) >> 7;
4735 }
4736 }
4737 return len;
4738}
4739
Willy Tarreau52bf8392020-03-08 00:42:37 +01004740
4741/* Random number generator state, see below */
Willy Tarreau1544c142020-03-12 00:31:18 +01004742static uint64_t ha_random_state[2] ALIGNED(2*sizeof(uint64_t));
Willy Tarreau52bf8392020-03-08 00:42:37 +01004743
4744/* This is a thread-safe implementation of xoroshiro128** described below:
4745 * http://prng.di.unimi.it/
4746 * It features a 2^128 long sequence, returns 64 high-quality bits on each call,
4747 * supports fast jumps and passes all common quality tests. It is thread-safe,
4748 * uses a double-cas on 64-bit architectures supporting it, and falls back to a
4749 * local lock on other ones.
4750 */
4751uint64_t ha_random64()
4752{
4753 uint64_t result;
Willy Tarreau1544c142020-03-12 00:31:18 +01004754 uint64_t old[2] ALIGNED(2*sizeof(uint64_t));
4755 uint64_t new[2] ALIGNED(2*sizeof(uint64_t));
Willy Tarreau52bf8392020-03-08 00:42:37 +01004756
4757#if defined(USE_THREAD) && (!defined(HA_CAS_IS_8B) || !defined(HA_HAVE_CAS_DW))
4758 static HA_SPINLOCK_T rand_lock;
4759
4760 HA_SPIN_LOCK(OTHER_LOCK, &rand_lock);
4761#endif
4762
4763 old[0] = ha_random_state[0];
4764 old[1] = ha_random_state[1];
4765
4766#if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
4767 do {
4768#endif
4769 result = rotl64(old[0] * 5, 7) * 9;
4770 new[1] = old[0] ^ old[1];
4771 new[0] = rotl64(old[0], 24) ^ new[1] ^ (new[1] << 16); // a, b
4772 new[1] = rotl64(new[1], 37); // c
4773
4774#if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
4775 } while (unlikely(!_HA_ATOMIC_DWCAS(ha_random_state, old, new)));
4776#else
4777 ha_random_state[0] = new[0];
4778 ha_random_state[1] = new[1];
4779#if defined(USE_THREAD)
4780 HA_SPIN_UNLOCK(OTHER_LOCK, &rand_lock);
4781#endif
4782#endif
4783 return result;
4784}
4785
4786/* seeds the random state using up to <len> bytes from <seed>, starting with
4787 * the first non-zero byte.
4788 */
4789void ha_random_seed(const unsigned char *seed, size_t len)
4790{
4791 size_t pos;
4792
4793 /* the seed must not be all zeroes, so we pre-fill it with alternating
4794 * bits and overwrite part of them with the block starting at the first
4795 * non-zero byte from the seed.
4796 */
4797 memset(ha_random_state, 0x55, sizeof(ha_random_state));
4798
4799 for (pos = 0; pos < len; pos++)
4800 if (seed[pos] != 0)
4801 break;
4802
4803 if (pos == len)
4804 return;
4805
4806 seed += pos;
4807 len -= pos;
4808
4809 if (len > sizeof(ha_random_state))
4810 len = sizeof(ha_random_state);
4811
4812 memcpy(ha_random_state, seed, len);
4813}
4814
4815/* This causes a jump to (dist * 2^96) places in the pseudo-random sequence,
4816 * and is equivalent to calling ha_random64() as many times. It is used to
4817 * provide non-overlapping sequences of 2^96 numbers (~7*10^28) to up to 2^32
4818 * different generators (i.e. different processes after a fork). The <dist>
4819 * argument is the distance to jump to and is used in a loop so it rather not
4820 * be too large if the processing time is a concern.
4821 *
4822 * BEWARE: this function is NOT thread-safe and must not be called during
4823 * concurrent accesses to ha_random64().
4824 */
4825void ha_random_jump96(uint32_t dist)
4826{
4827 while (dist--) {
4828 uint64_t s0 = 0;
4829 uint64_t s1 = 0;
4830 int b;
4831
4832 for (b = 0; b < 64; b++) {
4833 if ((0xd2a98b26625eee7bULL >> b) & 1) {
4834 s0 ^= ha_random_state[0];
4835 s1 ^= ha_random_state[1];
4836 }
4837 ha_random64();
4838 }
4839
4840 for (b = 0; b < 64; b++) {
4841 if ((0xdddf9b1090aa7ac1ULL >> b) & 1) {
4842 s0 ^= ha_random_state[0];
4843 s1 ^= ha_random_state[1];
4844 }
4845 ha_random64();
4846 }
4847 ha_random_state[0] = s0;
4848 ha_random_state[1] = s1;
4849 }
4850}
4851
Willy Tarreauee3bcdd2020-03-08 17:48:17 +01004852/* Generates an RFC4122 UUID into chunk <output> which must be at least 37
4853 * bytes large.
4854 */
4855void ha_generate_uuid(struct buffer *output)
4856{
4857 uint32_t rnd[4];
4858 uint64_t last;
4859
4860 last = ha_random64();
4861 rnd[0] = last;
4862 rnd[1] = last >> 32;
4863
4864 last = ha_random64();
4865 rnd[2] = last;
4866 rnd[3] = last >> 32;
4867
4868 chunk_printf(output, "%8.8x-%4.4x-%4.4x-%4.4x-%12.12llx",
4869 rnd[0],
4870 rnd[1] & 0xFFFF,
4871 ((rnd[1] >> 16u) & 0xFFF) | 0x4000, // highest 4 bits indicate the uuid version
4872 (rnd[2] & 0x3FFF) | 0x8000, // the highest 2 bits indicate the UUID variant (10),
4873 (long long)((rnd[2] >> 14u) | ((uint64_t) rnd[3] << 18u)) & 0xFFFFFFFFFFFFull);
4874}
4875
4876
Willy Tarreauc8d167b2020-06-16 16:27:26 +02004877/* only used by parse_line() below. It supports writing in place provided that
4878 * <in> is updated to the next location before calling it. In that case, the
4879 * char at <in> may be overwritten.
4880 */
4881#define EMIT_CHAR(x) \
4882 do { \
4883 char __c = (char)(x); \
4884 if ((opts & PARSE_OPT_INPLACE) && out+outpos > in) \
4885 err |= PARSE_ERR_OVERLAP; \
4886 if (outpos >= outmax) \
4887 err |= PARSE_ERR_TOOLARGE; \
4888 if (!err) \
4889 out[outpos] = __c; \
4890 outpos++; \
4891 } while (0)
4892
Ilya Shipitsin46a030c2020-07-05 16:36:08 +05004893/* Parse <in>, copy it into <out> split into isolated words whose pointers
Willy Tarreauc8d167b2020-06-16 16:27:26 +02004894 * are put in <args>. If more than <outlen> bytes have to be emitted, the
4895 * extraneous ones are not emitted but <outlen> is updated so that the caller
4896 * knows how much to realloc. Similarly, <args> are not updated beyond <nbargs>
4897 * but the returned <nbargs> indicates how many were found. All trailing args
Willy Tarreau61dd44b2020-06-25 07:35:42 +02004898 * up to <nbargs> point to the trailing zero, and as long as <nbargs> is > 0,
4899 * it is guaranteed that at least one arg will point to the zero. It is safe
4900 * to call it with a NULL <args> if <nbargs> is 0.
Willy Tarreauc8d167b2020-06-16 16:27:26 +02004901 *
4902 * <out> may overlap with <in> provided that it never goes further, in which
4903 * case the parser will accept to perform in-place parsing and unquoting/
4904 * unescaping but only if environment variables do not lead to expansion that
4905 * causes overlapping, otherwise the input string being destroyed, the error
4906 * will not be recoverable. Note that even during out-of-place <in> will
4907 * experience temporary modifications in-place for variable resolution and must
4908 * be writable, and will also receive zeroes to delimit words when using
4909 * in-place copy. Parsing options <opts> taken from PARSE_OPT_*. Return value
4910 * is zero on success otherwise a bitwise-or of PARSE_ERR_*. Upon error, the
4911 * starting point of the first invalid character sequence or unmatched
4912 * quote/brace is reported in <errptr> if not NULL. When using in-place parsing
4913 * error reporting might be difficult since zeroes will have been inserted into
4914 * the string. One solution for the caller may consist in replacing all args
4915 * delimiters with spaces in this case.
4916 */
4917uint32_t parse_line(char *in, char *out, size_t *outlen, char **args, int *nbargs, uint32_t opts, char **errptr)
4918{
4919 char *quote = NULL;
4920 char *brace = NULL;
4921 unsigned char hex1, hex2;
4922 size_t outmax = *outlen;
Willy Tarreau61dd44b2020-06-25 07:35:42 +02004923 int argsmax = *nbargs - 1;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02004924 size_t outpos = 0;
4925 int squote = 0;
4926 int dquote = 0;
4927 int arg = 0;
4928 uint32_t err = 0;
4929
4930 *nbargs = 0;
4931 *outlen = 0;
4932
Willy Tarreau61dd44b2020-06-25 07:35:42 +02004933 /* argsmax may be -1 here, protecting args[] from any write */
4934 if (arg < argsmax)
4935 args[arg] = out;
4936
Willy Tarreauc8d167b2020-06-16 16:27:26 +02004937 while (1) {
4938 if (*in >= '-' && *in != '\\') {
4939 /* speedup: directly send all regular chars starting
4940 * with '-', '.', '/', alnum etc...
4941 */
4942 EMIT_CHAR(*in++);
4943 continue;
4944 }
4945 else if (*in == '\0' || *in == '\n' || *in == '\r') {
4946 /* end of line */
4947 break;
4948 }
4949 else if (*in == '#' && (opts & PARSE_OPT_SHARP) && !squote && !dquote) {
4950 /* comment */
4951 break;
4952 }
4953 else if (*in == '"' && !squote && (opts & PARSE_OPT_DQUOTE)) { /* double quote outside single quotes */
4954 if (dquote) {
4955 dquote = 0;
4956 quote = NULL;
4957 }
4958 else {
4959 dquote = 1;
4960 quote = in;
4961 }
4962 in++;
4963 continue;
4964 }
4965 else if (*in == '\'' && !dquote && (opts & PARSE_OPT_SQUOTE)) { /* single quote outside double quotes */
4966 if (squote) {
4967 squote = 0;
4968 quote = NULL;
4969 }
4970 else {
4971 squote = 1;
4972 quote = in;
4973 }
4974 in++;
4975 continue;
4976 }
4977 else if (*in == '\\' && !squote && (opts & PARSE_OPT_BKSLASH)) {
4978 /* first, we'll replace \\, \<space>, \#, \r, \n, \t, \xXX with their
4979 * C equivalent value but only when they have a special meaning and within
4980 * double quotes for some of them. Other combinations left unchanged (eg: \1).
4981 */
4982 char tosend = *in;
4983
4984 switch (in[1]) {
4985 case ' ':
4986 case '\\':
4987 tosend = in[1];
4988 in++;
4989 break;
4990
4991 case 't':
4992 tosend = '\t';
4993 in++;
4994 break;
4995
4996 case 'n':
4997 tosend = '\n';
4998 in++;
4999 break;
5000
5001 case 'r':
5002 tosend = '\r';
5003 in++;
5004 break;
5005
5006 case '#':
5007 /* escaping of "#" only if comments are supported */
5008 if (opts & PARSE_OPT_SHARP)
5009 in++;
5010 tosend = *in;
5011 break;
5012
5013 case '\'':
5014 /* escaping of "'" only outside single quotes and only if single quotes are supported */
5015 if (opts & PARSE_OPT_SQUOTE && !squote)
5016 in++;
5017 tosend = *in;
5018 break;
5019
5020 case '"':
5021 /* escaping of '"' only outside single quotes and only if double quotes are supported */
5022 if (opts & PARSE_OPT_DQUOTE && !squote)
5023 in++;
5024 tosend = *in;
5025 break;
5026
5027 case '$':
5028 /* escaping of '$' only inside double quotes and only if env supported */
5029 if (opts & PARSE_OPT_ENV && dquote)
5030 in++;
5031 tosend = *in;
5032 break;
5033
5034 case 'x':
5035 if (!ishex(in[2]) || !ishex(in[3])) {
5036 /* invalid or incomplete hex sequence */
5037 err |= PARSE_ERR_HEX;
5038 if (errptr)
5039 *errptr = in;
5040 goto leave;
5041 }
Willy Tarreauf278eec2020-07-05 21:46:32 +02005042 hex1 = toupper((unsigned char)in[2]) - '0';
5043 hex2 = toupper((unsigned char)in[3]) - '0';
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005044 if (hex1 > 9) hex1 -= 'A' - '9' - 1;
5045 if (hex2 > 9) hex2 -= 'A' - '9' - 1;
5046 tosend = (hex1 << 4) + hex2;
5047 in += 3;
5048 break;
5049
5050 default:
5051 /* other combinations are not escape sequences */
5052 break;
5053 }
5054
5055 in++;
5056 EMIT_CHAR(tosend);
5057 }
5058 else if (isspace((unsigned char)*in) && !squote && !dquote) {
5059 /* a non-escaped space is an argument separator */
5060 while (isspace((unsigned char)*in))
5061 in++;
5062 EMIT_CHAR(0);
5063 arg++;
5064 if (arg < argsmax)
5065 args[arg] = out + outpos;
5066 else
5067 err |= PARSE_ERR_TOOMANY;
5068 }
5069 else if (*in == '$' && (opts & PARSE_OPT_ENV) && (dquote || !(opts & PARSE_OPT_DQUOTE))) {
5070 /* environment variables are evaluated anywhere, or only
5071 * inside double quotes if they are supported.
5072 */
5073 char *var_name;
5074 char save_char;
5075 char *value;
5076
5077 in++;
5078
5079 if (*in == '{')
5080 brace = in++;
5081
5082 if (!isalpha((unsigned char)*in) && *in != '_') {
5083 /* unacceptable character in variable name */
5084 err |= PARSE_ERR_VARNAME;
5085 if (errptr)
5086 *errptr = in;
5087 goto leave;
5088 }
5089
5090 var_name = in;
5091 while (isalnum((unsigned char)*in) || *in == '_')
5092 in++;
5093
5094 save_char = *in;
5095 *in = '\0';
5096 value = getenv(var_name);
5097 *in = save_char;
5098
5099 if (brace) {
5100 if (*in != '}') {
5101 /* unmatched brace */
5102 err |= PARSE_ERR_BRACE;
5103 if (errptr)
5104 *errptr = brace;
5105 goto leave;
5106 }
5107 in++;
5108 brace = NULL;
5109 }
5110
5111 if (value) {
5112 while (*value)
5113 EMIT_CHAR(*value++);
5114 }
5115 }
5116 else {
5117 /* any other regular char */
5118 EMIT_CHAR(*in++);
5119 }
5120 }
5121
5122 /* end of output string */
5123 EMIT_CHAR(0);
5124 arg++;
5125
5126 if (quote) {
5127 /* unmatched quote */
5128 err |= PARSE_ERR_QUOTE;
5129 if (errptr)
5130 *errptr = quote;
5131 goto leave;
5132 }
5133 leave:
5134 *nbargs = arg;
5135 *outlen = outpos;
5136
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005137 /* empty all trailing args by making them point to the trailing zero,
5138 * at least the last one in any case.
5139 */
5140 if (arg > argsmax)
5141 arg = argsmax;
5142
5143 while (arg >= 0 && arg <= argsmax)
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005144 args[arg++] = out + outpos - 1;
5145
5146 return err;
5147}
5148#undef EMIT_CHAR
5149
Willy Tarreauc54e5ad2020-06-25 09:15:40 +02005150/* This is used to sanitize an input line that's about to be used for error reporting.
5151 * It will adjust <line> to print approximately <width> chars around <pos>, trying to
5152 * preserve the beginning, with leading or trailing "..." when the line is truncated.
5153 * If non-printable chars are present in the output. It returns the new offset <pos>
5154 * in the modified line. Non-printable characters are replaced with '?'. <width> must
5155 * be at least 6 to support two "..." otherwise the result is undefined. The line
5156 * itself must have at least 7 chars allocated for the same reason.
5157 */
5158size_t sanitize_for_printing(char *line, size_t pos, size_t width)
5159{
5160 size_t shift = 0;
5161 char *out = line;
5162 char *in = line;
5163 char *end = line + width;
5164
5165 if (pos >= width) {
5166 /* if we have to shift, we'll be out of context, so let's
5167 * try to put <pos> at the center of width.
5168 */
5169 shift = pos - width / 2;
5170 in += shift + 3;
5171 end = out + width - 3;
5172 out[0] = out[1] = out[2] = '.';
5173 out += 3;
5174 }
5175
5176 while (out < end && *in) {
5177 if (isspace((unsigned char)*in))
5178 *out++ = ' ';
5179 else if (isprint((unsigned char)*in))
5180 *out++ = *in;
5181 else
5182 *out++ = '?';
5183 in++;
5184 }
5185
5186 if (end < line + width) {
5187 out[0] = out[1] = out[2] = '.';
5188 out += 3;
5189 }
5190
5191 *out++ = 0;
5192 return pos - shift;
5193}
5194
Willy Tarreaubaaee002006-06-26 02:48:02 +02005195/*
5196 * Local variables:
5197 * c-indent-level: 8
5198 * c-basic-offset: 8
5199 * End:
5200 */