blob: f2553a62a43c341e1672e029f116fcfb4fbd6b74 [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 Tarreau62a976c2020-09-16 15:20:59 +02001167 if (opts & PA_O_SOCKET_FD && sock_type == SOCK_DGRAM && ctrl_type == SOCK_DGRAM) {
1168 /* FIXME: for now UDP is still its own family. However some UDP clients
1169 * (logs, dns) use AF_INET and are not aware of AF_CUST_UDP*. Since we
1170 * only want this mapping for listeners and they are the only ones
1171 * setting PA_O_SOCKET_FD, for now we condition this mapping to this.
1172 */
1173 if (ss.ss_family == AF_INET6)
1174 ss.ss_family = AF_CUST_UDP6;
1175 else if (ss.ss_family == AF_INET)
1176 ss.ss_family = AF_CUST_UDP4;
1177 }
1178
Willy Tarreau5fc93282020-09-16 18:25:03 +02001179 if (proto) {
1180 /* Note: if the caller asks for a proto, we must find one,
1181 * except if we return with an fqdn that will resolve later,
1182 * in which case the address is not known yet (this is only
1183 * for servers actually).
1184 */
1185 new_proto = protocol_by_family(ss.ss_family);
1186 if (!new_proto && (!fqdn || !*fqdn)) {
1187 memprintf(err, "unsupported protocol family %d for address '%s'", ss.ss_family, str);
1188 goto out;
1189 }
1190 }
1191
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001192 ret = &ss;
Willy Tarreaud5191e72010-02-09 20:50:45 +01001193 out:
Willy Tarreau48ef4c92017-01-06 18:32:38 +01001194 if (port)
1195 *port = porta;
Willy Tarreaud4448bc2013-02-20 15:55:15 +01001196 if (low)
1197 *low = portl;
1198 if (high)
1199 *high = porth;
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001200 if (fd)
1201 *fd = new_fd;
Willy Tarreau5fc93282020-09-16 18:25:03 +02001202 if (proto)
1203 *proto = new_proto;
Willy Tarreau24709282013-03-10 21:32:12 +01001204 free(back);
Willy Tarreaud5191e72010-02-09 20:50:45 +01001205 return ret;
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001206}
1207
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001208/* converts <str> to a struct in_addr containing a network mask. It can be
1209 * passed in dotted form (255.255.255.0) or in CIDR form (24). It returns 1
Jarno Huuskonen577d5ac2017-05-21 17:32:21 +03001210 * if the conversion succeeds otherwise zero.
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001211 */
1212int str2mask(const char *str, struct in_addr *mask)
1213{
1214 if (strchr(str, '.') != NULL) { /* dotted notation */
1215 if (!inet_pton(AF_INET, str, mask))
1216 return 0;
1217 }
1218 else { /* mask length */
1219 char *err;
1220 unsigned long len = strtol(str, &err, 10);
1221
1222 if (!*str || (err && *err) || (unsigned)len > 32)
1223 return 0;
Tim Duesterhus8575f722018-01-25 16:24:48 +01001224
1225 len2mask4(len, mask);
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001226 }
1227 return 1;
1228}
1229
Tim Duesterhus47185172018-01-25 16:24:49 +01001230/* converts <str> to a struct in6_addr containing a network mask. It can be
Tim Duesterhus5e642862018-02-20 17:02:18 +01001231 * passed in quadruplet form (ffff:ffff::) or in CIDR form (64). It returns 1
Tim Duesterhus47185172018-01-25 16:24:49 +01001232 * if the conversion succeeds otherwise zero.
1233 */
1234int str2mask6(const char *str, struct in6_addr *mask)
1235{
1236 if (strchr(str, ':') != NULL) { /* quadruplet notation */
1237 if (!inet_pton(AF_INET6, str, mask))
1238 return 0;
1239 }
1240 else { /* mask length */
1241 char *err;
1242 unsigned long len = strtol(str, &err, 10);
1243
1244 if (!*str || (err && *err) || (unsigned)len > 128)
1245 return 0;
1246
1247 len2mask6(len, mask);
1248 }
1249 return 1;
1250}
1251
Thierry FOURNIERb0504632013-12-14 15:39:02 +01001252/* convert <cidr> to struct in_addr <mask>. It returns 1 if the conversion
1253 * succeeds otherwise zero.
1254 */
1255int cidr2dotted(int cidr, struct in_addr *mask) {
1256
1257 if (cidr < 0 || cidr > 32)
1258 return 0;
1259
1260 mask->s_addr = cidr ? htonl(~0UL << (32 - cidr)) : 0;
1261 return 1;
1262}
1263
Thierry Fournier70473a52016-02-17 17:12:14 +01001264/* Convert mask from bit length form to in_addr form.
1265 * This function never fails.
1266 */
1267void len2mask4(int len, struct in_addr *addr)
1268{
1269 if (len >= 32) {
1270 addr->s_addr = 0xffffffff;
1271 return;
1272 }
1273 if (len <= 0) {
1274 addr->s_addr = 0x00000000;
1275 return;
1276 }
1277 addr->s_addr = 0xffffffff << (32 - len);
1278 addr->s_addr = htonl(addr->s_addr);
1279}
1280
1281/* Convert mask from bit length form to in6_addr form.
1282 * This function never fails.
1283 */
1284void len2mask6(int len, struct in6_addr *addr)
1285{
1286 len2mask4(len, (struct in_addr *)&addr->s6_addr[0]); /* msb */
1287 len -= 32;
1288 len2mask4(len, (struct in_addr *)&addr->s6_addr[4]);
1289 len -= 32;
1290 len2mask4(len, (struct in_addr *)&addr->s6_addr[8]);
1291 len -= 32;
1292 len2mask4(len, (struct in_addr *)&addr->s6_addr[12]); /* lsb */
1293}
1294
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001295/*
Willy Tarreaud077a8e2007-05-08 18:28:09 +02001296 * converts <str> to two struct in_addr* which must be pre-allocated.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001297 * The format is "addr[/mask]", where "addr" cannot be empty, and mask
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001298 * is optional and either in the dotted or CIDR notation.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001299 * Note: "addr" can also be a hostname. Returns 1 if OK, 0 if error.
1300 */
Thierry FOURNIERfc7ac7b2014-02-11 15:23:04 +01001301int str2net(const char *str, int resolve, struct in_addr *addr, struct in_addr *mask)
Willy Tarreaubaaee002006-06-26 02:48:02 +02001302{
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001303 __label__ out_free, out_err;
1304 char *c, *s;
1305 int ret_val;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001306
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001307 s = strdup(str);
1308 if (!s)
1309 return 0;
1310
Willy Tarreaubaaee002006-06-26 02:48:02 +02001311 memset(mask, 0, sizeof(*mask));
1312 memset(addr, 0, sizeof(*addr));
Willy Tarreaubaaee002006-06-26 02:48:02 +02001313
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001314 if ((c = strrchr(s, '/')) != NULL) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02001315 *c++ = '\0';
1316 /* c points to the mask */
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001317 if (!str2mask(c, mask))
1318 goto out_err;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001319 }
1320 else {
Willy Tarreauebd61602006-12-30 11:54:15 +01001321 mask->s_addr = ~0U;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001322 }
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001323 if (!inet_pton(AF_INET, s, addr)) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02001324 struct hostent *he;
1325
Thierry FOURNIERfc7ac7b2014-02-11 15:23:04 +01001326 if (!resolve)
1327 goto out_err;
1328
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001329 if ((he = gethostbyname(s)) == NULL) {
1330 goto out_err;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001331 }
1332 else
1333 *addr = *(struct in_addr *) *(he->h_addr_list);
1334 }
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001335
1336 ret_val = 1;
1337 out_free:
1338 free(s);
1339 return ret_val;
1340 out_err:
1341 ret_val = 0;
1342 goto out_free;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001343}
1344
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001345
1346/*
Willy Tarreau6d20e282012-04-27 22:49:47 +02001347 * converts <str> to two struct in6_addr* which must be pre-allocated.
1348 * The format is "addr[/mask]", where "addr" cannot be empty, and mask
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001349 * is an optional number of bits (128 being the default).
Willy Tarreau6d20e282012-04-27 22:49:47 +02001350 * Returns 1 if OK, 0 if error.
1351 */
1352int str62net(const char *str, struct in6_addr *addr, unsigned char *mask)
1353{
1354 char *c, *s;
1355 int ret_val = 0;
1356 char *err;
1357 unsigned long len = 128;
1358
1359 s = strdup(str);
1360 if (!s)
1361 return 0;
1362
1363 memset(mask, 0, sizeof(*mask));
1364 memset(addr, 0, sizeof(*addr));
1365
1366 if ((c = strrchr(s, '/')) != NULL) {
1367 *c++ = '\0'; /* c points to the mask */
1368 if (!*c)
1369 goto out_free;
1370
1371 len = strtoul(c, &err, 10);
1372 if ((err && *err) || (unsigned)len > 128)
1373 goto out_free;
1374 }
1375 *mask = len; /* OK we have a valid mask in <len> */
1376
1377 if (!inet_pton(AF_INET6, s, addr))
1378 goto out_free;
1379
1380 ret_val = 1;
1381 out_free:
1382 free(s);
1383 return ret_val;
1384}
1385
1386
1387/*
David du Colombier6f5ccb12011-03-10 22:26:24 +01001388 * Parse IPv4 address found in url.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001389 */
David du Colombier6f5ccb12011-03-10 22:26:24 +01001390int url2ipv4(const char *addr, struct in_addr *dst)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001391{
1392 int saw_digit, octets, ch;
1393 u_char tmp[4], *tp;
1394 const char *cp = addr;
1395
1396 saw_digit = 0;
1397 octets = 0;
1398 *(tp = tmp) = 0;
1399
1400 while (*addr) {
1401 unsigned char digit = (ch = *addr++) - '0';
1402 if (digit > 9 && ch != '.')
1403 break;
1404 if (digit <= 9) {
1405 u_int new = *tp * 10 + digit;
1406 if (new > 255)
1407 return 0;
1408 *tp = new;
1409 if (!saw_digit) {
1410 if (++octets > 4)
1411 return 0;
1412 saw_digit = 1;
1413 }
1414 } else if (ch == '.' && saw_digit) {
1415 if (octets == 4)
1416 return 0;
1417 *++tp = 0;
1418 saw_digit = 0;
1419 } else
1420 return 0;
1421 }
1422
1423 if (octets < 4)
1424 return 0;
1425
1426 memcpy(&dst->s_addr, tmp, 4);
1427 return addr-cp-1;
1428}
1429
1430/*
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001431 * Resolve destination server from URL. Convert <str> to a sockaddr_storage.
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001432 * <out> contain the code of the detected scheme, the start and length of
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001433 * the hostname. Actually only http and https are supported. <out> can be NULL.
1434 * This function returns the consumed length. It is useful if you parse complete
1435 * url like http://host:port/path, because the consumed length corresponds to
1436 * the first character of the path. If the conversion fails, it returns -1.
1437 *
1438 * This function tries to resolve the DNS name if haproxy is in starting mode.
1439 * So, this function may be used during the configuration parsing.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001440 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001441int url2sa(const char *url, int ulen, struct sockaddr_storage *addr, struct split_url *out)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001442{
1443 const char *curr = url, *cp = url;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001444 const char *end;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001445 int ret, url_code = 0;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001446 unsigned long long int http_code = 0;
1447 int default_port;
1448 struct hostent *he;
1449 char *p;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001450
1451 /* Firstly, try to find :// pattern */
1452 while (curr < url+ulen && url_code != 0x3a2f2f) {
1453 url_code = ((url_code & 0xffff) << 8);
1454 url_code += (unsigned char)*curr++;
1455 }
1456
1457 /* Secondly, if :// pattern is found, verify parsed stuff
1458 * before pattern is matching our http pattern.
1459 * If so parse ip address and port in uri.
1460 *
1461 * WARNING: Current code doesn't support dynamic async dns resolver.
1462 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001463 if (url_code != 0x3a2f2f)
1464 return -1;
1465
1466 /* Copy scheme, and utrn to lower case. */
1467 while (cp < curr - 3)
1468 http_code = (http_code << 8) + *cp++;
1469 http_code |= 0x2020202020202020ULL; /* Turn everything to lower case */
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001470
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001471 /* HTTP or HTTPS url matching */
1472 if (http_code == 0x2020202068747470ULL) {
1473 default_port = 80;
1474 if (out)
1475 out->scheme = SCH_HTTP;
1476 }
1477 else if (http_code == 0x2020206874747073ULL) {
1478 default_port = 443;
1479 if (out)
1480 out->scheme = SCH_HTTPS;
1481 }
1482 else
1483 return -1;
1484
1485 /* If the next char is '[', the host address is IPv6. */
1486 if (*curr == '[') {
1487 curr++;
1488
1489 /* Check trash size */
1490 if (trash.size < ulen)
1491 return -1;
1492
1493 /* Look for ']' and copy the address in a trash buffer. */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001494 p = trash.area;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001495 for (end = curr;
1496 end < url + ulen && *end != ']';
1497 end++, p++)
1498 *p = *end;
1499 if (*end != ']')
1500 return -1;
1501 *p = '\0';
1502
1503 /* Update out. */
1504 if (out) {
1505 out->host = curr;
1506 out->host_len = end - curr;
1507 }
1508
1509 /* Try IPv6 decoding. */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001510 if (!inet_pton(AF_INET6, trash.area, &((struct sockaddr_in6 *)addr)->sin6_addr))
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001511 return -1;
1512 end++;
1513
1514 /* Decode port. */
1515 if (*end == ':') {
1516 end++;
1517 default_port = read_uint(&end, url + ulen);
1518 }
1519 ((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1520 ((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1521 return end - url;
1522 }
1523 else {
1524 /* We are looking for IP address. If you want to parse and
1525 * resolve hostname found in url, you can use str2sa_range(), but
1526 * be warned this can slow down global daemon performances
1527 * while handling lagging dns responses.
1528 */
1529 ret = url2ipv4(curr, &((struct sockaddr_in *)addr)->sin_addr);
1530 if (ret) {
1531 /* Update out. */
1532 if (out) {
1533 out->host = curr;
1534 out->host_len = ret;
1535 }
1536
1537 curr += ret;
1538
1539 /* Decode port. */
1540 if (*curr == ':') {
1541 curr++;
1542 default_port = read_uint(&curr, url + ulen);
1543 }
1544 ((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1545
1546 /* Set family. */
1547 ((struct sockaddr_in *)addr)->sin_family = AF_INET;
1548 return curr - url;
1549 }
1550 else if (global.mode & MODE_STARTING) {
1551 /* The IPv4 and IPv6 decoding fails, maybe the url contain name. Try to execute
1552 * synchronous DNS request only if HAProxy is in the start state.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001553 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001554
1555 /* look for : or / or end */
1556 for (end = curr;
1557 end < url + ulen && *end != '/' && *end != ':';
1558 end++);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001559 memcpy(trash.area, curr, end - curr);
1560 trash.area[end - curr] = '\0';
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001561
1562 /* try to resolve an IPv4/IPv6 hostname */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001563 he = gethostbyname(trash.area);
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001564 if (!he)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001565 return -1;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001566
1567 /* Update out. */
1568 if (out) {
1569 out->host = curr;
1570 out->host_len = end - curr;
1571 }
1572
1573 /* Decode port. */
1574 if (*end == ':') {
1575 end++;
1576 default_port = read_uint(&end, url + ulen);
1577 }
1578
1579 /* Copy IP address, set port and family. */
1580 switch (he->h_addrtype) {
1581 case AF_INET:
1582 ((struct sockaddr_in *)addr)->sin_addr = *(struct in_addr *) *(he->h_addr_list);
1583 ((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1584 ((struct sockaddr_in *)addr)->sin_family = AF_INET;
1585 return end - url;
1586
1587 case AF_INET6:
1588 ((struct sockaddr_in6 *)addr)->sin6_addr = *(struct in6_addr *) *(he->h_addr_list);
1589 ((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1590 ((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1591 return end - url;
1592 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001593 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001594 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001595 return -1;
1596}
1597
Willy Tarreau631f01c2011-09-05 00:36:48 +02001598/* Tries to convert a sockaddr_storage address to text form. Upon success, the
1599 * address family is returned so that it's easy for the caller to adapt to the
1600 * output format. Zero is returned if the address family is not supported. -1
1601 * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1602 * supported.
1603 */
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001604int addr_to_str(const struct sockaddr_storage *addr, char *str, int size)
Willy Tarreau631f01c2011-09-05 00:36:48 +02001605{
1606
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001607 const void *ptr;
Willy Tarreau631f01c2011-09-05 00:36:48 +02001608
1609 if (size < 5)
1610 return 0;
1611 *str = '\0';
1612
1613 switch (addr->ss_family) {
1614 case AF_INET:
1615 ptr = &((struct sockaddr_in *)addr)->sin_addr;
1616 break;
1617 case AF_INET6:
1618 ptr = &((struct sockaddr_in6 *)addr)->sin6_addr;
1619 break;
1620 case AF_UNIX:
1621 memcpy(str, "unix", 5);
1622 return addr->ss_family;
1623 default:
1624 return 0;
1625 }
1626
1627 if (inet_ntop(addr->ss_family, ptr, str, size))
1628 return addr->ss_family;
1629
1630 /* failed */
1631 return -1;
1632}
1633
Simon Horman75ab8bd2014-06-16 09:39:41 +09001634/* Tries to convert a sockaddr_storage port to text form. Upon success, the
1635 * address family is returned so that it's easy for the caller to adapt to the
1636 * output format. Zero is returned if the address family is not supported. -1
1637 * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1638 * supported.
1639 */
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001640int port_to_str(const struct sockaddr_storage *addr, char *str, int size)
Simon Horman75ab8bd2014-06-16 09:39:41 +09001641{
1642
1643 uint16_t port;
1644
1645
Willy Tarreaud7dad1b2017-01-06 16:46:22 +01001646 if (size < 6)
Simon Horman75ab8bd2014-06-16 09:39:41 +09001647 return 0;
1648 *str = '\0';
1649
1650 switch (addr->ss_family) {
1651 case AF_INET:
1652 port = ((struct sockaddr_in *)addr)->sin_port;
1653 break;
1654 case AF_INET6:
1655 port = ((struct sockaddr_in6 *)addr)->sin6_port;
1656 break;
1657 case AF_UNIX:
1658 memcpy(str, "unix", 5);
1659 return addr->ss_family;
1660 default:
1661 return 0;
1662 }
1663
1664 snprintf(str, size, "%u", ntohs(port));
1665 return addr->ss_family;
1666}
1667
Willy Tarreau16e01562016-08-09 16:46:18 +02001668/* check if the given address is local to the system or not. It will return
1669 * -1 when it's not possible to know, 0 when the address is not local, 1 when
1670 * it is. We don't want to iterate over all interfaces for this (and it is not
1671 * portable). So instead we try to bind in UDP to this address on a free non
1672 * privileged port and to connect to the same address, port 0 (connect doesn't
1673 * care). If it succeeds, we own the address. Note that non-inet addresses are
1674 * considered local since they're most likely AF_UNIX.
1675 */
1676int addr_is_local(const struct netns_entry *ns,
1677 const struct sockaddr_storage *orig)
1678{
1679 struct sockaddr_storage addr;
1680 int result;
1681 int fd;
1682
1683 if (!is_inet_addr(orig))
1684 return 1;
1685
1686 memcpy(&addr, orig, sizeof(addr));
1687 set_host_port(&addr, 0);
1688
1689 fd = my_socketat(ns, addr.ss_family, SOCK_DGRAM, IPPROTO_UDP);
1690 if (fd < 0)
1691 return -1;
1692
1693 result = -1;
1694 if (bind(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == 0) {
1695 if (connect(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == -1)
1696 result = 0; // fail, non-local address
1697 else
1698 result = 1; // success, local address
1699 }
1700 else {
1701 if (errno == EADDRNOTAVAIL)
1702 result = 0; // definitely not local :-)
1703 }
1704 close(fd);
1705
1706 return result;
1707}
1708
Willy Tarreaubaaee002006-06-26 02:48:02 +02001709/* will try to encode the string <string> replacing all characters tagged in
1710 * <map> with the hexadecimal representation of their ASCII-code (2 digits)
1711 * prefixed by <escape>, and will store the result between <start> (included)
1712 * and <stop> (excluded), and will always terminate the string with a '\0'
1713 * before <stop>. The position of the '\0' is returned if the conversion
1714 * completes. If bytes are missing between <start> and <stop>, then the
1715 * conversion will be incomplete and truncated. If <stop> <= <start>, the '\0'
1716 * cannot even be stored so we return <start> without writing the 0.
1717 * The input string must also be zero-terminated.
1718 */
1719const char hextab[16] = "0123456789ABCDEF";
1720char *encode_string(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001721 const char escape, const long *map,
Willy Tarreaubaaee002006-06-26 02:48:02 +02001722 const char *string)
1723{
1724 if (start < stop) {
1725 stop--; /* reserve one byte for the final '\0' */
1726 while (start < stop && *string != '\0') {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001727 if (!ha_bit_test((unsigned char)(*string), map))
Willy Tarreaubaaee002006-06-26 02:48:02 +02001728 *start++ = *string;
1729 else {
1730 if (start + 3 >= stop)
1731 break;
1732 *start++ = escape;
1733 *start++ = hextab[(*string >> 4) & 15];
1734 *start++ = hextab[*string & 15];
1735 }
1736 string++;
1737 }
1738 *start = '\0';
1739 }
1740 return start;
1741}
1742
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001743/*
1744 * Same behavior as encode_string() above, except that it encodes chunk
1745 * <chunk> instead of a string.
1746 */
1747char *encode_chunk(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001748 const char escape, const long *map,
Willy Tarreau83061a82018-07-13 11:56:34 +02001749 const struct buffer *chunk)
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001750{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001751 char *str = chunk->area;
1752 char *end = chunk->area + chunk->data;
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001753
1754 if (start < stop) {
1755 stop--; /* reserve one byte for the final '\0' */
1756 while (start < stop && str < end) {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001757 if (!ha_bit_test((unsigned char)(*str), map))
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001758 *start++ = *str;
1759 else {
1760 if (start + 3 >= stop)
1761 break;
1762 *start++ = escape;
1763 *start++ = hextab[(*str >> 4) & 15];
1764 *start++ = hextab[*str & 15];
1765 }
1766 str++;
1767 }
1768 *start = '\0';
1769 }
1770 return start;
1771}
1772
Dragan Dosen0edd1092016-02-12 13:23:02 +01001773/*
1774 * Tries to prefix characters tagged in the <map> with the <escape>
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001775 * character. The input <string> must be zero-terminated. The result will
1776 * be stored between <start> (included) and <stop> (excluded). This
1777 * function will always try to terminate the resulting string with a '\0'
1778 * before <stop>, and will return its position if the conversion
1779 * completes.
1780 */
1781char *escape_string(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001782 const char escape, const long *map,
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001783 const char *string)
1784{
1785 if (start < stop) {
1786 stop--; /* reserve one byte for the final '\0' */
1787 while (start < stop && *string != '\0') {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001788 if (!ha_bit_test((unsigned char)(*string), map))
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001789 *start++ = *string;
1790 else {
1791 if (start + 2 >= stop)
1792 break;
1793 *start++ = escape;
1794 *start++ = *string;
1795 }
1796 string++;
1797 }
1798 *start = '\0';
1799 }
1800 return start;
1801}
1802
1803/*
1804 * Tries to prefix characters tagged in the <map> with the <escape>
Dragan Dosen0edd1092016-02-12 13:23:02 +01001805 * character. <chunk> contains the input to be escaped. The result will be
1806 * stored between <start> (included) and <stop> (excluded). The function
1807 * will always try to terminate the resulting string with a '\0' before
1808 * <stop>, and will return its position if the conversion completes.
1809 */
1810char *escape_chunk(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001811 const char escape, const long *map,
Willy Tarreau83061a82018-07-13 11:56:34 +02001812 const struct buffer *chunk)
Dragan Dosen0edd1092016-02-12 13:23:02 +01001813{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001814 char *str = chunk->area;
1815 char *end = chunk->area + chunk->data;
Dragan Dosen0edd1092016-02-12 13:23:02 +01001816
1817 if (start < stop) {
1818 stop--; /* reserve one byte for the final '\0' */
1819 while (start < stop && str < end) {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001820 if (!ha_bit_test((unsigned char)(*str), map))
Dragan Dosen0edd1092016-02-12 13:23:02 +01001821 *start++ = *str;
1822 else {
1823 if (start + 2 >= stop)
1824 break;
1825 *start++ = escape;
1826 *start++ = *str;
1827 }
1828 str++;
1829 }
1830 *start = '\0';
1831 }
1832 return start;
1833}
1834
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001835/* Check a string for using it in a CSV output format. If the string contains
1836 * one of the following four char <">, <,>, CR or LF, the string is
1837 * encapsulated between <"> and the <"> are escaped by a <""> sequence.
1838 * <str> is the input string to be escaped. The function assumes that
1839 * the input string is null-terminated.
1840 *
1841 * If <quote> is 0, the result is returned escaped but without double quote.
Willy Tarreau898529b2016-01-06 18:07:04 +01001842 * It is useful if the escaped string is used between double quotes in the
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001843 * format.
1844 *
Willy Tarreau898529b2016-01-06 18:07:04 +01001845 * printf("..., \"%s\", ...\r\n", csv_enc(str, 0, &trash));
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001846 *
Willy Tarreaub631c292016-01-08 10:04:08 +01001847 * If <quote> is 1, the converter puts the quotes only if any reserved character
1848 * is present. If <quote> is 2, the converter always puts the quotes.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001849 *
Willy Tarreau83061a82018-07-13 11:56:34 +02001850 * <output> is a struct buffer used for storing the output string.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001851 *
Willy Tarreau898529b2016-01-06 18:07:04 +01001852 * The function returns the converted string on its output. If an error
1853 * occurs, the function returns an empty string. This type of output is useful
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001854 * for using the function directly as printf() argument.
1855 *
1856 * If the output buffer is too short to contain the input string, the result
1857 * is truncated.
Willy Tarreau898529b2016-01-06 18:07:04 +01001858 *
Willy Tarreaub631c292016-01-08 10:04:08 +01001859 * This function appends the encoding to the existing output chunk, and it
1860 * guarantees that it starts immediately at the first available character of
1861 * the chunk. Please use csv_enc() instead if you want to replace the output
1862 * chunk.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001863 */
Willy Tarreau83061a82018-07-13 11:56:34 +02001864const char *csv_enc_append(const char *str, int quote, struct buffer *output)
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001865{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001866 char *end = output->area + output->size;
1867 char *out = output->area + output->data;
Willy Tarreau898529b2016-01-06 18:07:04 +01001868 char *ptr = out;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001869
Willy Tarreaub631c292016-01-08 10:04:08 +01001870 if (quote == 1) {
1871 /* automatic quoting: first verify if we'll have to quote the string */
1872 if (!strpbrk(str, "\n\r,\""))
1873 quote = 0;
1874 }
1875
1876 if (quote)
1877 *ptr++ = '"';
1878
Willy Tarreau898529b2016-01-06 18:07:04 +01001879 while (*str && ptr < end - 2) { /* -2 for reserving space for <"> and \0. */
1880 *ptr = *str;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001881 if (*str == '"') {
Willy Tarreau898529b2016-01-06 18:07:04 +01001882 ptr++;
1883 if (ptr >= end - 2) {
1884 ptr--;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001885 break;
1886 }
Willy Tarreau898529b2016-01-06 18:07:04 +01001887 *ptr = '"';
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001888 }
Willy Tarreau898529b2016-01-06 18:07:04 +01001889 ptr++;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001890 str++;
1891 }
1892
Willy Tarreaub631c292016-01-08 10:04:08 +01001893 if (quote)
1894 *ptr++ = '"';
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001895
Willy Tarreau898529b2016-01-06 18:07:04 +01001896 *ptr = '\0';
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001897 output->data = ptr - output->area;
Willy Tarreau898529b2016-01-06 18:07:04 +01001898 return out;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001899}
1900
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02001901/* Decode an URL-encoded string in-place. The resulting string might
1902 * be shorter. If some forbidden characters are found, the conversion is
Thierry FOURNIER5068d962013-10-04 16:27:27 +02001903 * aborted, the string is truncated before the issue and a negative value is
1904 * returned, otherwise the operation returns the length of the decoded string.
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02001905 * If the 'in_form' argument is non-nul the string is assumed to be part of
1906 * an "application/x-www-form-urlencoded" encoded string, and the '+' will be
1907 * turned to a space. If it's zero, this will only be done after a question
1908 * mark ('?').
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02001909 */
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02001910int url_decode(char *string, int in_form)
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02001911{
1912 char *in, *out;
Thierry FOURNIER5068d962013-10-04 16:27:27 +02001913 int ret = -1;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02001914
1915 in = string;
1916 out = string;
1917 while (*in) {
1918 switch (*in) {
1919 case '+' :
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02001920 *out++ = in_form ? ' ' : *in;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02001921 break;
1922 case '%' :
1923 if (!ishex(in[1]) || !ishex(in[2]))
1924 goto end;
1925 *out++ = (hex2i(in[1]) << 4) + hex2i(in[2]);
1926 in += 2;
1927 break;
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02001928 case '?':
1929 in_form = 1;
1930 /* fall through */
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02001931 default:
1932 *out++ = *in;
1933 break;
1934 }
1935 in++;
1936 }
Thierry FOURNIER5068d962013-10-04 16:27:27 +02001937 ret = out - string; /* success */
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02001938 end:
1939 *out = 0;
1940 return ret;
1941}
Willy Tarreaubaaee002006-06-26 02:48:02 +02001942
Willy Tarreau6911fa42007-03-04 18:06:08 +01001943unsigned int str2ui(const char *s)
1944{
1945 return __str2ui(s);
1946}
1947
1948unsigned int str2uic(const char *s)
1949{
1950 return __str2uic(s);
1951}
1952
1953unsigned int strl2ui(const char *s, int len)
1954{
1955 return __strl2ui(s, len);
1956}
1957
1958unsigned int strl2uic(const char *s, int len)
1959{
1960 return __strl2uic(s, len);
1961}
1962
Willy Tarreau4ec83cd2010-10-15 23:19:55 +02001963unsigned int read_uint(const char **s, const char *end)
1964{
1965 return __read_uint(s, end);
1966}
1967
Thierry FOURNIER763a5d82015-07-06 23:09:52 +02001968/* This function reads an unsigned integer from the string pointed to by <s> and
1969 * returns it. The <s> pointer is adjusted to point to the first unread char. The
1970 * function automatically stops at <end>. If the number overflows, the 2^64-1
1971 * value is returned.
1972 */
1973unsigned long long int read_uint64(const char **s, const char *end)
1974{
1975 const char *ptr = *s;
1976 unsigned long long int i = 0, tmp;
1977 unsigned int j;
1978
1979 while (ptr < end) {
1980
1981 /* read next char */
1982 j = *ptr - '0';
1983 if (j > 9)
1984 goto read_uint64_end;
1985
1986 /* add char to the number and check overflow. */
1987 tmp = i * 10;
1988 if (tmp / 10 != i) {
1989 i = ULLONG_MAX;
1990 goto read_uint64_eat;
1991 }
1992 if (ULLONG_MAX - tmp < j) {
1993 i = ULLONG_MAX;
1994 goto read_uint64_eat;
1995 }
1996 i = tmp + j;
1997 ptr++;
1998 }
1999read_uint64_eat:
2000 /* eat each numeric char */
2001 while (ptr < end) {
2002 if ((unsigned int)(*ptr - '0') > 9)
2003 break;
2004 ptr++;
2005 }
2006read_uint64_end:
2007 *s = ptr;
2008 return i;
2009}
2010
2011/* This function reads an integer from the string pointed to by <s> and returns
2012 * it. The <s> pointer is adjusted to point to the first unread char. The function
2013 * automatically stops at <end>. Il the number is bigger than 2^63-2, the 2^63-1
2014 * value is returned. If the number is lowest than -2^63-1, the -2^63 value is
2015 * returned.
2016 */
2017long long int read_int64(const char **s, const char *end)
2018{
2019 unsigned long long int i = 0;
2020 int neg = 0;
2021
2022 /* Look for minus char. */
2023 if (**s == '-') {
2024 neg = 1;
2025 (*s)++;
2026 }
2027 else if (**s == '+')
2028 (*s)++;
2029
2030 /* convert as positive number. */
2031 i = read_uint64(s, end);
2032
2033 if (neg) {
2034 if (i > 0x8000000000000000ULL)
2035 return LLONG_MIN;
2036 return -i;
2037 }
2038 if (i > 0x7fffffffffffffffULL)
2039 return LLONG_MAX;
2040 return i;
2041}
2042
Willy Tarreau6911fa42007-03-04 18:06:08 +01002043/* This one is 7 times faster than strtol() on athlon with checks.
2044 * It returns the value of the number composed of all valid digits read,
2045 * and can process negative numbers too.
2046 */
2047int strl2ic(const char *s, int len)
2048{
2049 int i = 0;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002050 int j, k;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002051
2052 if (len > 0) {
2053 if (*s != '-') {
2054 /* positive number */
2055 while (len-- > 0) {
2056 j = (*s++) - '0';
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002057 k = i * 10;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002058 if (j > 9)
2059 break;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002060 i = k + j;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002061 }
2062 } else {
2063 /* negative number */
2064 s++;
2065 while (--len > 0) {
2066 j = (*s++) - '0';
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002067 k = i * 10;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002068 if (j > 9)
2069 break;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002070 i = k - j;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002071 }
2072 }
2073 }
2074 return i;
2075}
2076
2077
2078/* This function reads exactly <len> chars from <s> and converts them to a
2079 * signed integer which it stores into <ret>. It accurately detects any error
2080 * (truncated string, invalid chars, overflows). It is meant to be used in
2081 * applications designed for hostile environments. It returns zero when the
2082 * number has successfully been converted, non-zero otherwise. When an error
2083 * is returned, the <ret> value is left untouched. It is yet 5 to 40 times
2084 * faster than strtol().
2085 */
2086int strl2irc(const char *s, int len, int *ret)
2087{
2088 int i = 0;
2089 int j;
2090
2091 if (!len)
2092 return 1;
2093
2094 if (*s != '-') {
2095 /* positive number */
2096 while (len-- > 0) {
2097 j = (*s++) - '0';
2098 if (j > 9) return 1; /* invalid char */
2099 if (i > INT_MAX / 10) return 1; /* check for multiply overflow */
2100 i = i * 10;
2101 if (i + j < i) return 1; /* check for addition overflow */
2102 i = i + j;
2103 }
2104 } else {
2105 /* negative number */
2106 s++;
2107 while (--len > 0) {
2108 j = (*s++) - '0';
2109 if (j > 9) return 1; /* invalid char */
2110 if (i < INT_MIN / 10) return 1; /* check for multiply overflow */
2111 i = i * 10;
2112 if (i - j > i) return 1; /* check for subtract overflow */
2113 i = i - j;
2114 }
2115 }
2116 *ret = i;
2117 return 0;
2118}
2119
2120
2121/* This function reads exactly <len> chars from <s> and converts them to a
2122 * signed integer which it stores into <ret>. It accurately detects any error
2123 * (truncated string, invalid chars, overflows). It is meant to be used in
2124 * applications designed for hostile environments. It returns zero when the
2125 * number has successfully been converted, non-zero otherwise. When an error
2126 * is returned, the <ret> value is left untouched. It is about 3 times slower
2127 * than str2irc().
2128 */
Willy Tarreau6911fa42007-03-04 18:06:08 +01002129
2130int strl2llrc(const char *s, int len, long long *ret)
2131{
2132 long long i = 0;
2133 int j;
2134
2135 if (!len)
2136 return 1;
2137
2138 if (*s != '-') {
2139 /* positive number */
2140 while (len-- > 0) {
2141 j = (*s++) - '0';
2142 if (j > 9) return 1; /* invalid char */
2143 if (i > LLONG_MAX / 10LL) return 1; /* check for multiply overflow */
2144 i = i * 10LL;
2145 if (i + j < i) return 1; /* check for addition overflow */
2146 i = i + j;
2147 }
2148 } else {
2149 /* negative number */
2150 s++;
2151 while (--len > 0) {
2152 j = (*s++) - '0';
2153 if (j > 9) return 1; /* invalid char */
2154 if (i < LLONG_MIN / 10LL) return 1; /* check for multiply overflow */
2155 i = i * 10LL;
2156 if (i - j > i) return 1; /* check for subtract overflow */
2157 i = i - j;
2158 }
2159 }
2160 *ret = i;
2161 return 0;
2162}
2163
Thierry FOURNIER511e9472014-01-23 17:40:34 +01002164/* This function is used with pat_parse_dotted_ver(). It converts a string
2165 * composed by two number separated by a dot. Each part must contain in 16 bits
2166 * because internally they will be represented as a 32-bit quantity stored in
2167 * a 64-bit integer. It returns zero when the number has successfully been
2168 * converted, non-zero otherwise. When an error is returned, the <ret> value
2169 * is left untouched.
2170 *
2171 * "1.3" -> 0x0000000000010003
2172 * "65535.65535" -> 0x00000000ffffffff
2173 */
2174int strl2llrc_dotted(const char *text, int len, long long *ret)
2175{
2176 const char *end = &text[len];
2177 const char *p;
2178 long long major, minor;
2179
2180 /* Look for dot. */
2181 for (p = text; p < end; p++)
2182 if (*p == '.')
2183 break;
2184
2185 /* Convert major. */
2186 if (strl2llrc(text, p - text, &major) != 0)
2187 return 1;
2188
2189 /* Check major. */
2190 if (major >= 65536)
2191 return 1;
2192
2193 /* Convert minor. */
2194 minor = 0;
2195 if (p < end)
2196 if (strl2llrc(p + 1, end - (p + 1), &minor) != 0)
2197 return 1;
2198
2199 /* Check minor. */
2200 if (minor >= 65536)
2201 return 1;
2202
2203 /* Compose value. */
2204 *ret = (major << 16) | (minor & 0xffff);
2205 return 0;
2206}
2207
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002208/* This function parses a time value optionally followed by a unit suffix among
2209 * "d", "h", "m", "s", "ms" or "us". It converts the value into the unit
2210 * expected by the caller. The computation does its best to avoid overflows.
2211 * The value is returned in <ret> if everything is fine, and a NULL is returned
2212 * by the function. In case of error, a pointer to the error is returned and
2213 * <ret> is left untouched. Values are automatically rounded up when needed.
Willy Tarreau9faebe32019-06-07 19:00:37 +02002214 * Values resulting in values larger than or equal to 2^31 after conversion are
2215 * reported as an overflow as value PARSE_TIME_OVER. Non-null values resulting
2216 * in an underflow are reported as an underflow as value PARSE_TIME_UNDER.
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002217 */
2218const char *parse_time_err(const char *text, unsigned *ret, unsigned unit_flags)
2219{
Willy Tarreau9faebe32019-06-07 19:00:37 +02002220 unsigned long long imult, idiv;
2221 unsigned long long omult, odiv;
2222 unsigned long long value, result;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002223
2224 omult = odiv = 1;
2225
2226 switch (unit_flags & TIME_UNIT_MASK) {
2227 case TIME_UNIT_US: omult = 1000000; break;
2228 case TIME_UNIT_MS: omult = 1000; break;
2229 case TIME_UNIT_S: break;
2230 case TIME_UNIT_MIN: odiv = 60; break;
2231 case TIME_UNIT_HOUR: odiv = 3600; break;
2232 case TIME_UNIT_DAY: odiv = 86400; break;
2233 default: break;
2234 }
2235
2236 value = 0;
2237
2238 while (1) {
2239 unsigned int j;
2240
2241 j = *text - '0';
2242 if (j > 9)
2243 break;
2244 text++;
2245 value *= 10;
2246 value += j;
2247 }
2248
2249 imult = idiv = 1;
2250 switch (*text) {
2251 case '\0': /* no unit = default unit */
2252 imult = omult = idiv = odiv = 1;
2253 break;
2254 case 's': /* second = unscaled unit */
2255 break;
2256 case 'u': /* microsecond : "us" */
2257 if (text[1] == 's') {
2258 idiv = 1000000;
2259 text++;
2260 }
2261 break;
2262 case 'm': /* millisecond : "ms" or minute: "m" */
2263 if (text[1] == 's') {
2264 idiv = 1000;
2265 text++;
2266 } else
2267 imult = 60;
2268 break;
2269 case 'h': /* hour : "h" */
2270 imult = 3600;
2271 break;
2272 case 'd': /* day : "d" */
2273 imult = 86400;
2274 break;
2275 default:
2276 return text;
2277 break;
2278 }
2279
2280 if (omult % idiv == 0) { omult /= idiv; idiv = 1; }
2281 if (idiv % omult == 0) { idiv /= omult; omult = 1; }
2282 if (imult % odiv == 0) { imult /= odiv; odiv = 1; }
2283 if (odiv % imult == 0) { odiv /= imult; imult = 1; }
2284
Willy Tarreau9faebe32019-06-07 19:00:37 +02002285 result = (value * (imult * omult) + (idiv * odiv - 1)) / (idiv * odiv);
2286 if (result >= 0x80000000)
2287 return PARSE_TIME_OVER;
2288 if (!result && value)
2289 return PARSE_TIME_UNDER;
2290 *ret = result;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002291 return NULL;
2292}
Willy Tarreau6911fa42007-03-04 18:06:08 +01002293
Emeric Brun39132b22010-01-04 14:57:24 +01002294/* this function converts the string starting at <text> to an unsigned int
2295 * stored in <ret>. If an error is detected, the pointer to the unexpected
Joseph Herlant32b83272018-11-15 11:58:28 -08002296 * character is returned. If the conversion is successful, NULL is returned.
Emeric Brun39132b22010-01-04 14:57:24 +01002297 */
2298const char *parse_size_err(const char *text, unsigned *ret) {
2299 unsigned value = 0;
2300
2301 while (1) {
2302 unsigned int j;
2303
2304 j = *text - '0';
2305 if (j > 9)
2306 break;
2307 if (value > ~0U / 10)
2308 return text;
2309 value *= 10;
2310 if (value > (value + j))
2311 return text;
2312 value += j;
2313 text++;
2314 }
2315
2316 switch (*text) {
2317 case '\0':
2318 break;
2319 case 'K':
2320 case 'k':
2321 if (value > ~0U >> 10)
2322 return text;
2323 value = value << 10;
2324 break;
2325 case 'M':
2326 case 'm':
2327 if (value > ~0U >> 20)
2328 return text;
2329 value = value << 20;
2330 break;
2331 case 'G':
2332 case 'g':
2333 if (value > ~0U >> 30)
2334 return text;
2335 value = value << 30;
2336 break;
2337 default:
2338 return text;
2339 }
2340
Godbach58048a22015-01-28 17:36:16 +08002341 if (*text != '\0' && *++text != '\0')
2342 return text;
2343
Emeric Brun39132b22010-01-04 14:57:24 +01002344 *ret = value;
2345 return NULL;
2346}
2347
Willy Tarreau126d4062013-12-03 17:50:47 +01002348/*
2349 * Parse binary string written in hexadecimal (source) and store the decoded
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002350 * result into binstr and set binstrlen to the length of binstr. Memory for
Willy Tarreau126d4062013-12-03 17:50:47 +01002351 * binstr is allocated by the function. In case of error, returns 0 with an
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002352 * error message in err. In success case, it returns the consumed length.
Willy Tarreau126d4062013-12-03 17:50:47 +01002353 */
2354int parse_binary(const char *source, char **binstr, int *binstrlen, char **err)
2355{
2356 int len;
2357 const char *p = source;
2358 int i,j;
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002359 int alloc;
Willy Tarreau126d4062013-12-03 17:50:47 +01002360
2361 len = strlen(source);
2362 if (len % 2) {
2363 memprintf(err, "an even number of hex digit is expected");
2364 return 0;
2365 }
2366
2367 len = len >> 1;
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002368
Willy Tarreau126d4062013-12-03 17:50:47 +01002369 if (!*binstr) {
Tim Duesterhuse52b6e52020-09-12 20:26:43 +02002370 *binstr = calloc(len, sizeof(**binstr));
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002371 if (!*binstr) {
2372 memprintf(err, "out of memory while loading string pattern");
2373 return 0;
2374 }
2375 alloc = 1;
Willy Tarreau126d4062013-12-03 17:50:47 +01002376 }
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002377 else {
2378 if (*binstrlen < len) {
Joseph Herlant76dbe782018-11-15 12:01:22 -08002379 memprintf(err, "no space available in the buffer. expect %d, provides %d",
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002380 len, *binstrlen);
2381 return 0;
2382 }
2383 alloc = 0;
2384 }
2385 *binstrlen = len;
Willy Tarreau126d4062013-12-03 17:50:47 +01002386
2387 i = j = 0;
2388 while (j < len) {
2389 if (!ishex(p[i++]))
2390 goto bad_input;
2391 if (!ishex(p[i++]))
2392 goto bad_input;
2393 (*binstr)[j++] = (hex2i(p[i-2]) << 4) + hex2i(p[i-1]);
2394 }
Thierry FOURNIERee330af2014-01-21 11:36:14 +01002395 return len << 1;
Willy Tarreau126d4062013-12-03 17:50:47 +01002396
2397bad_input:
2398 memprintf(err, "an hex digit is expected (found '%c')", p[i-1]);
Andreas Seltenreich93f91c32016-03-03 20:40:37 +01002399 if (alloc) {
2400 free(*binstr);
2401 *binstr = NULL;
2402 }
Willy Tarreau126d4062013-12-03 17:50:47 +01002403 return 0;
2404}
2405
Willy Tarreau946ba592009-05-10 15:41:18 +02002406/* copies at most <n> characters from <src> and always terminates with '\0' */
2407char *my_strndup(const char *src, int n)
2408{
2409 int len = 0;
2410 char *ret;
2411
2412 while (len < n && src[len])
2413 len++;
2414
Vincent Bernat3c2f2f22016-04-03 13:48:42 +02002415 ret = malloc(len + 1);
Willy Tarreau946ba592009-05-10 15:41:18 +02002416 if (!ret)
2417 return ret;
2418 memcpy(ret, src, len);
2419 ret[len] = '\0';
2420 return ret;
2421}
2422
Baptiste Assmannbb77c8e2013-10-06 23:24:13 +02002423/*
2424 * search needle in haystack
2425 * returns the pointer if found, returns NULL otherwise
2426 */
2427const void *my_memmem(const void *haystack, size_t haystacklen, const void *needle, size_t needlelen)
2428{
2429 const void *c = NULL;
2430 unsigned char f;
2431
2432 if ((haystack == NULL) || (needle == NULL) || (haystacklen < needlelen))
2433 return NULL;
2434
2435 f = *(char *)needle;
2436 c = haystack;
2437 while ((c = memchr(c, f, haystacklen - (c - haystack))) != NULL) {
2438 if ((haystacklen - (c - haystack)) < needlelen)
2439 return NULL;
2440
2441 if (memcmp(c, needle, needlelen) == 0)
2442 return c;
2443 ++c;
2444 }
2445 return NULL;
2446}
2447
Ilya Shipitsinc02a23f2020-05-06 00:53:22 +05002448/* get length of the initial segment consisting entirely of bytes in <accept> */
Christopher Faulet5eb96cb2020-04-15 10:23:01 +02002449size_t my_memspn(const void *str, size_t len, const void *accept, size_t acceptlen)
2450{
2451 size_t ret = 0;
2452
2453 while (ret < len && memchr(accept, *((int *)str), acceptlen)) {
2454 str++;
2455 ret++;
2456 }
2457 return ret;
2458}
2459
Ilya Shipitsinc02a23f2020-05-06 00:53:22 +05002460/* get length of the initial segment consisting entirely of bytes not in <rejcet> */
Christopher Faulet5eb96cb2020-04-15 10:23:01 +02002461size_t my_memcspn(const void *str, size_t len, const void *reject, size_t rejectlen)
2462{
2463 size_t ret = 0;
2464
2465 while (ret < len) {
2466 if(memchr(reject, *((int *)str), rejectlen))
2467 return ret;
2468 str++;
2469 ret++;
2470 }
2471 return ret;
2472}
2473
Willy Tarreau482b00d2009-10-04 22:48:42 +02002474/* This function returns the first unused key greater than or equal to <key> in
2475 * ID tree <root>. Zero is returned if no place is found.
2476 */
2477unsigned int get_next_id(struct eb_root *root, unsigned int key)
2478{
2479 struct eb32_node *used;
2480
2481 do {
2482 used = eb32_lookup_ge(root, key);
2483 if (!used || used->key > key)
2484 return key; /* key is available */
2485 key++;
2486 } while (key);
2487 return key;
2488}
2489
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002490/* dump the full tree to <file> in DOT format for debugging purposes. Will
2491 * optionally highlight node <subj> if found, depending on operation <op> :
2492 * 0 : nothing
2493 * >0 : insertion, node/leaf are surrounded in red
2494 * <0 : removal, node/leaf are dashed with no background
2495 * Will optionally add "desc" as a label on the graph if set and non-null.
2496 */
2497void 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 +01002498{
2499 struct eb32sc_node *node;
2500 unsigned long scope = -1;
2501
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002502 fprintf(file, "digraph ebtree {\n");
2503
2504 if (desc && *desc) {
2505 fprintf(file,
2506 " fontname=\"fixed\";\n"
2507 " fontsize=8;\n"
2508 " label=\"%s\";\n", desc);
2509 }
2510
Willy Tarreaued3cda02017-11-15 15:04:05 +01002511 fprintf(file,
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002512 " node [fontname=\"fixed\" fontsize=8 shape=\"box\" style=\"filled\" color=\"black\" fillcolor=\"white\"];\n"
2513 " edge [fontname=\"fixed\" fontsize=8 style=\"solid\" color=\"magenta\" dir=\"forward\"];\n"
Willy Tarreaued3cda02017-11-15 15:04:05 +01002514 " \"%lx_n\" [label=\"root\\n%lx\"]\n", (long)eb_root_to_node(root), (long)root
2515 );
2516
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002517 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002518 (long)eb_root_to_node(root),
2519 (long)eb_root_to_node(eb_clrtag(root->b[0])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002520 eb_gettag(root->b[0]) == EB_LEAF ? 'l' : 'n');
2521
2522 node = eb32sc_first(root, scope);
2523 while (node) {
2524 if (node->node.node_p) {
2525 /* node part is used */
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002526 fprintf(file, " \"%lx_n\" [label=\"%lx\\nkey=%u\\nscope=%lx\\nbit=%d\" fillcolor=\"lightskyblue1\" %s];\n",
2527 (long)node, (long)node, node->key, node->node_s, node->node.bit,
2528 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
Willy Tarreaued3cda02017-11-15 15:04:05 +01002529
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002530 fprintf(file, " \"%lx_n\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002531 (long)node,
2532 (long)eb_root_to_node(eb_clrtag(node->node.node_p)),
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002533 eb_gettag(node->node.node_p) ? 'R' : 'L');
Willy Tarreaued3cda02017-11-15 15:04:05 +01002534
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002535 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002536 (long)node,
2537 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[0])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002538 eb_gettag(node->node.branches.b[0]) == EB_LEAF ? 'l' : 'n');
2539
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002540 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"R\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002541 (long)node,
2542 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[1])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002543 eb_gettag(node->node.branches.b[1]) == EB_LEAF ? 'l' : 'n');
2544 }
2545
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002546 fprintf(file, " \"%lx_l\" [label=\"%lx\\nkey=%u\\nscope=%lx\\npfx=%u\" fillcolor=\"yellow\" %s];\n",
2547 (long)node, (long)node, node->key, node->leaf_s, node->node.pfx,
2548 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
Willy Tarreaued3cda02017-11-15 15:04:05 +01002549
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002550 fprintf(file, " \"%lx_l\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002551 (long)node,
2552 (long)eb_root_to_node(eb_clrtag(node->node.leaf_p)),
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002553 eb_gettag(node->node.leaf_p) ? 'R' : 'L');
Willy Tarreaued3cda02017-11-15 15:04:05 +01002554 node = eb32sc_next(node, scope);
2555 }
2556 fprintf(file, "}\n");
2557}
2558
Willy Tarreau348238b2010-01-18 15:05:57 +01002559/* This function compares a sample word possibly followed by blanks to another
2560 * clean word. The compare is case-insensitive. 1 is returned if both are equal,
2561 * otherwise zero. This intends to be used when checking HTTP headers for some
2562 * values. Note that it validates a word followed only by blanks but does not
2563 * validate a word followed by blanks then other chars.
2564 */
2565int word_match(const char *sample, int slen, const char *word, int wlen)
2566{
2567 if (slen < wlen)
2568 return 0;
2569
2570 while (wlen) {
2571 char c = *sample ^ *word;
2572 if (c && c != ('A' ^ 'a'))
2573 return 0;
2574 sample++;
2575 word++;
2576 slen--;
2577 wlen--;
2578 }
2579
2580 while (slen) {
2581 if (*sample != ' ' && *sample != '\t')
2582 return 0;
2583 sample++;
2584 slen--;
2585 }
2586 return 1;
2587}
Willy Tarreau482b00d2009-10-04 22:48:42 +02002588
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002589/* Converts any text-formatted IPv4 address to a host-order IPv4 address. It
2590 * is particularly fast because it avoids expensive operations such as
2591 * multiplies, which are optimized away at the end. It requires a properly
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002592 * formatted address though (3 points).
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002593 */
2594unsigned int inetaddr_host(const char *text)
2595{
2596 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2597 register unsigned int dig100, dig10, dig1;
2598 int s;
2599 const char *p, *d;
2600
2601 dig1 = dig10 = dig100 = ascii_zero;
2602 s = 24;
2603
2604 p = text;
2605 while (1) {
2606 if (((unsigned)(*p - '0')) <= 9) {
2607 p++;
2608 continue;
2609 }
2610
2611 /* here, we have a complete byte between <text> and <p> (exclusive) */
2612 if (p == text)
2613 goto end;
2614
2615 d = p - 1;
2616 dig1 |= (unsigned int)(*d << s);
2617 if (d == text)
2618 goto end;
2619
2620 d--;
2621 dig10 |= (unsigned int)(*d << s);
2622 if (d == text)
2623 goto end;
2624
2625 d--;
2626 dig100 |= (unsigned int)(*d << s);
2627 end:
2628 if (!s || *p != '.')
2629 break;
2630
2631 s -= 8;
2632 text = ++p;
2633 }
2634
2635 dig100 -= ascii_zero;
2636 dig10 -= ascii_zero;
2637 dig1 -= ascii_zero;
2638 return ((dig100 * 10) + dig10) * 10 + dig1;
2639}
2640
2641/*
2642 * Idem except the first unparsed character has to be passed in <stop>.
2643 */
2644unsigned int inetaddr_host_lim(const char *text, const char *stop)
2645{
2646 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2647 register unsigned int dig100, dig10, dig1;
2648 int s;
2649 const char *p, *d;
2650
2651 dig1 = dig10 = dig100 = ascii_zero;
2652 s = 24;
2653
2654 p = text;
2655 while (1) {
2656 if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2657 p++;
2658 continue;
2659 }
2660
2661 /* here, we have a complete byte between <text> and <p> (exclusive) */
2662 if (p == text)
2663 goto end;
2664
2665 d = p - 1;
2666 dig1 |= (unsigned int)(*d << s);
2667 if (d == text)
2668 goto end;
2669
2670 d--;
2671 dig10 |= (unsigned int)(*d << s);
2672 if (d == text)
2673 goto end;
2674
2675 d--;
2676 dig100 |= (unsigned int)(*d << s);
2677 end:
2678 if (!s || p == stop || *p != '.')
2679 break;
2680
2681 s -= 8;
2682 text = ++p;
2683 }
2684
2685 dig100 -= ascii_zero;
2686 dig10 -= ascii_zero;
2687 dig1 -= ascii_zero;
2688 return ((dig100 * 10) + dig10) * 10 + dig1;
2689}
2690
2691/*
2692 * Idem except the pointer to first unparsed byte is returned into <ret> which
2693 * must not be NULL.
2694 */
Willy Tarreau74172752010-10-15 23:21:42 +02002695unsigned int inetaddr_host_lim_ret(char *text, char *stop, char **ret)
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002696{
2697 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2698 register unsigned int dig100, dig10, dig1;
2699 int s;
Willy Tarreau74172752010-10-15 23:21:42 +02002700 char *p, *d;
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002701
2702 dig1 = dig10 = dig100 = ascii_zero;
2703 s = 24;
2704
2705 p = text;
2706 while (1) {
2707 if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2708 p++;
2709 continue;
2710 }
2711
2712 /* here, we have a complete byte between <text> and <p> (exclusive) */
2713 if (p == text)
2714 goto end;
2715
2716 d = p - 1;
2717 dig1 |= (unsigned int)(*d << s);
2718 if (d == text)
2719 goto end;
2720
2721 d--;
2722 dig10 |= (unsigned int)(*d << s);
2723 if (d == text)
2724 goto end;
2725
2726 d--;
2727 dig100 |= (unsigned int)(*d << s);
2728 end:
2729 if (!s || p == stop || *p != '.')
2730 break;
2731
2732 s -= 8;
2733 text = ++p;
2734 }
2735
2736 *ret = p;
2737 dig100 -= ascii_zero;
2738 dig10 -= ascii_zero;
2739 dig1 -= ascii_zero;
2740 return ((dig100 * 10) + dig10) * 10 + dig1;
2741}
2742
Willy Tarreauf0b38bf2010-06-06 13:22:23 +02002743/* Convert a fixed-length string to an IP address. Returns 0 in case of error,
2744 * or the number of chars read in case of success. Maybe this could be replaced
2745 * by one of the functions above. Also, apparently this function does not support
2746 * hosts above 255 and requires exactly 4 octets.
Willy Tarreau075415a2013-12-12 11:29:39 +01002747 * The destination is only modified on success.
Willy Tarreauf0b38bf2010-06-06 13:22:23 +02002748 */
2749int buf2ip(const char *buf, size_t len, struct in_addr *dst)
2750{
2751 const char *addr;
2752 int saw_digit, octets, ch;
2753 u_char tmp[4], *tp;
2754 const char *cp = buf;
2755
2756 saw_digit = 0;
2757 octets = 0;
2758 *(tp = tmp) = 0;
2759
2760 for (addr = buf; addr - buf < len; addr++) {
2761 unsigned char digit = (ch = *addr) - '0';
2762
2763 if (digit > 9 && ch != '.')
2764 break;
2765
2766 if (digit <= 9) {
2767 u_int new = *tp * 10 + digit;
2768
2769 if (new > 255)
2770 return 0;
2771
2772 *tp = new;
2773
2774 if (!saw_digit) {
2775 if (++octets > 4)
2776 return 0;
2777 saw_digit = 1;
2778 }
2779 } else if (ch == '.' && saw_digit) {
2780 if (octets == 4)
2781 return 0;
2782
2783 *++tp = 0;
2784 saw_digit = 0;
2785 } else
2786 return 0;
2787 }
2788
2789 if (octets < 4)
2790 return 0;
2791
2792 memcpy(&dst->s_addr, tmp, 4);
2793 return addr - cp;
2794}
2795
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002796/* This function converts the string in <buf> of the len <len> to
2797 * struct in6_addr <dst> which must be allocated by the caller.
2798 * This function returns 1 in success case, otherwise zero.
Willy Tarreau075415a2013-12-12 11:29:39 +01002799 * The destination is only modified on success.
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002800 */
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002801int buf2ip6(const char *buf, size_t len, struct in6_addr *dst)
2802{
Thierry FOURNIERcd659912013-12-11 12:33:54 +01002803 char null_term_ip6[INET6_ADDRSTRLEN + 1];
Willy Tarreau075415a2013-12-12 11:29:39 +01002804 struct in6_addr out;
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002805
Thierry FOURNIERcd659912013-12-11 12:33:54 +01002806 if (len > INET6_ADDRSTRLEN)
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002807 return 0;
2808
2809 memcpy(null_term_ip6, buf, len);
2810 null_term_ip6[len] = '\0';
2811
Willy Tarreau075415a2013-12-12 11:29:39 +01002812 if (!inet_pton(AF_INET6, null_term_ip6, &out))
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002813 return 0;
2814
Willy Tarreau075415a2013-12-12 11:29:39 +01002815 *dst = out;
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002816 return 1;
2817}
2818
Willy Tarreauacf95772010-06-14 19:09:21 +02002819/* To be used to quote config arg positions. Returns the short string at <ptr>
2820 * surrounded by simple quotes if <ptr> is valid and non-empty, or "end of line"
2821 * if ptr is NULL or empty. The string is locally allocated.
2822 */
2823const char *quote_arg(const char *ptr)
2824{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002825 static THREAD_LOCAL char val[32];
Willy Tarreauacf95772010-06-14 19:09:21 +02002826 int i;
2827
2828 if (!ptr || !*ptr)
2829 return "end of line";
2830 val[0] = '\'';
Willy Tarreaude2dd6b2013-01-24 02:14:42 +01002831 for (i = 1; i < sizeof(val) - 2 && *ptr; i++)
Willy Tarreauacf95772010-06-14 19:09:21 +02002832 val[i] = *ptr++;
2833 val[i++] = '\'';
2834 val[i] = '\0';
2835 return val;
2836}
2837
Willy Tarreau5b180202010-07-18 10:40:48 +02002838/* returns an operator among STD_OP_* for string <str> or < 0 if unknown */
2839int get_std_op(const char *str)
2840{
2841 int ret = -1;
2842
2843 if (*str == 'e' && str[1] == 'q')
2844 ret = STD_OP_EQ;
2845 else if (*str == 'n' && str[1] == 'e')
2846 ret = STD_OP_NE;
2847 else if (*str == 'l') {
2848 if (str[1] == 'e') ret = STD_OP_LE;
2849 else if (str[1] == 't') ret = STD_OP_LT;
2850 }
2851 else if (*str == 'g') {
2852 if (str[1] == 'e') ret = STD_OP_GE;
2853 else if (str[1] == 't') ret = STD_OP_GT;
2854 }
2855
2856 if (ret == -1 || str[2] != '\0')
2857 return -1;
2858 return ret;
2859}
2860
Willy Tarreau4c14eaa2010-11-24 14:01:45 +01002861/* hash a 32-bit integer to another 32-bit integer */
2862unsigned int full_hash(unsigned int a)
2863{
2864 return __full_hash(a);
2865}
2866
Willy Tarreauf3241112019-02-26 09:56:22 +01002867/* Return the bit position in mask <m> of the nth bit set of rank <r>, between
2868 * 0 and LONGBITS-1 included, starting from the left. For example ranks 0,1,2,3
2869 * for mask 0x55 will be 6, 4, 2 and 0 respectively. This algorithm is based on
2870 * a popcount variant and is described here :
2871 * https://graphics.stanford.edu/~seander/bithacks.html
2872 */
2873unsigned int mask_find_rank_bit(unsigned int r, unsigned long m)
2874{
2875 unsigned long a, b, c, d;
2876 unsigned int s;
2877 unsigned int t;
2878
2879 a = m - ((m >> 1) & ~0UL/3);
2880 b = (a & ~0UL/5) + ((a >> 2) & ~0UL/5);
2881 c = (b + (b >> 4)) & ~0UL/0x11;
2882 d = (c + (c >> 8)) & ~0UL/0x101;
2883
2884 r++; // make r be 1..64
2885
2886 t = 0;
2887 s = LONGBITS;
2888 if (s > 32) {
Willy Tarreau9b6be3b2019-03-18 16:31:18 +01002889 unsigned long d2 = (d >> 16) >> 16;
2890 t = d2 + (d2 >> 16);
Willy Tarreauf3241112019-02-26 09:56:22 +01002891 s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
2892 }
2893
2894 t = (d >> (s - 16)) & 0xff;
2895 s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
2896 t = (c >> (s - 8)) & 0xf;
2897 s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
2898 t = (b >> (s - 4)) & 0x7;
2899 s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
2900 t = (a >> (s - 2)) & 0x3;
2901 s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
2902 t = (m >> (s - 1)) & 0x1;
2903 s -= ((t - r) & 256) >> 8;
2904
2905 return s - 1;
2906}
2907
2908/* Same as mask_find_rank_bit() above but makes use of pre-computed bitmaps
2909 * based on <m>, in <a..d>. These ones must be updated whenever <m> changes
2910 * using mask_prep_rank_map() below.
2911 */
2912unsigned int mask_find_rank_bit_fast(unsigned int r, unsigned long m,
2913 unsigned long a, unsigned long b,
2914 unsigned long c, unsigned long d)
2915{
2916 unsigned int s;
2917 unsigned int t;
2918
2919 r++; // make r be 1..64
2920
2921 t = 0;
2922 s = LONGBITS;
2923 if (s > 32) {
Willy Tarreau9b6be3b2019-03-18 16:31:18 +01002924 unsigned long d2 = (d >> 16) >> 16;
2925 t = d2 + (d2 >> 16);
Willy Tarreauf3241112019-02-26 09:56:22 +01002926 s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
2927 }
2928
2929 t = (d >> (s - 16)) & 0xff;
2930 s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
2931 t = (c >> (s - 8)) & 0xf;
2932 s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
2933 t = (b >> (s - 4)) & 0x7;
2934 s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
2935 t = (a >> (s - 2)) & 0x3;
2936 s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
2937 t = (m >> (s - 1)) & 0x1;
2938 s -= ((t - r) & 256) >> 8;
2939
2940 return s - 1;
2941}
2942
2943/* Prepare the bitmaps used by the fast implementation of the find_rank_bit()
2944 * above.
2945 */
2946void mask_prep_rank_map(unsigned long m,
2947 unsigned long *a, unsigned long *b,
2948 unsigned long *c, unsigned long *d)
2949{
2950 *a = m - ((m >> 1) & ~0UL/3);
2951 *b = (*a & ~0UL/5) + ((*a >> 2) & ~0UL/5);
2952 *c = (*b + (*b >> 4)) & ~0UL/0x11;
2953 *d = (*c + (*c >> 8)) & ~0UL/0x101;
2954}
2955
David du Colombier4f92d322011-03-24 11:09:31 +01002956/* Return non-zero if IPv4 address is part of the network,
Willy Tarreaueec1d382016-07-13 11:59:39 +02002957 * otherwise zero. Note that <addr> may not necessarily be aligned
2958 * while the two other ones must.
David du Colombier4f92d322011-03-24 11:09:31 +01002959 */
Willy Tarreaueec1d382016-07-13 11:59:39 +02002960int in_net_ipv4(const void *addr, const struct in_addr *mask, const struct in_addr *net)
David du Colombier4f92d322011-03-24 11:09:31 +01002961{
Willy Tarreaueec1d382016-07-13 11:59:39 +02002962 struct in_addr addr_copy;
2963
2964 memcpy(&addr_copy, addr, sizeof(addr_copy));
2965 return((addr_copy.s_addr & mask->s_addr) == (net->s_addr & mask->s_addr));
David du Colombier4f92d322011-03-24 11:09:31 +01002966}
2967
2968/* Return non-zero if IPv6 address is part of the network,
Willy Tarreaueec1d382016-07-13 11:59:39 +02002969 * otherwise zero. Note that <addr> may not necessarily be aligned
2970 * while the two other ones must.
David du Colombier4f92d322011-03-24 11:09:31 +01002971 */
Willy Tarreaueec1d382016-07-13 11:59:39 +02002972int in_net_ipv6(const void *addr, const struct in6_addr *mask, const struct in6_addr *net)
David du Colombier4f92d322011-03-24 11:09:31 +01002973{
2974 int i;
Willy Tarreaueec1d382016-07-13 11:59:39 +02002975 struct in6_addr addr_copy;
David du Colombier4f92d322011-03-24 11:09:31 +01002976
Willy Tarreaueec1d382016-07-13 11:59:39 +02002977 memcpy(&addr_copy, addr, sizeof(addr_copy));
David du Colombier4f92d322011-03-24 11:09:31 +01002978 for (i = 0; i < sizeof(struct in6_addr) / sizeof(int); i++)
Willy Tarreaueec1d382016-07-13 11:59:39 +02002979 if (((((int *)&addr_copy)[i] & ((int *)mask)[i])) !=
David du Colombier4f92d322011-03-24 11:09:31 +01002980 (((int *)net)[i] & ((int *)mask)[i]))
2981 return 0;
2982 return 1;
2983}
2984
2985/* RFC 4291 prefix */
2986const char rfc4291_pfx[] = { 0x00, 0x00, 0x00, 0x00,
2987 0x00, 0x00, 0x00, 0x00,
2988 0x00, 0x00, 0xFF, 0xFF };
2989
Joseph Herlant32b83272018-11-15 11:58:28 -08002990/* Map IPv4 address on IPv6 address, as specified in RFC 3513.
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01002991 * Input and output may overlap.
2992 */
David du Colombier4f92d322011-03-24 11:09:31 +01002993void v4tov6(struct in6_addr *sin6_addr, struct in_addr *sin_addr)
2994{
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01002995 struct in_addr tmp_addr;
2996
2997 tmp_addr.s_addr = sin_addr->s_addr;
David du Colombier4f92d322011-03-24 11:09:31 +01002998 memcpy(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx));
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01002999 memcpy(sin6_addr->s6_addr+12, &tmp_addr.s_addr, 4);
David du Colombier4f92d322011-03-24 11:09:31 +01003000}
3001
Joseph Herlant32b83272018-11-15 11:58:28 -08003002/* Map IPv6 address on IPv4 address, as specified in RFC 3513.
David du Colombier4f92d322011-03-24 11:09:31 +01003003 * Return true if conversion is possible and false otherwise.
3004 */
3005int v6tov4(struct in_addr *sin_addr, struct in6_addr *sin6_addr)
3006{
3007 if (memcmp(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx)) == 0) {
3008 memcpy(&(sin_addr->s_addr), &(sin6_addr->s6_addr[12]),
3009 sizeof(struct in_addr));
3010 return 1;
3011 }
3012
3013 return 0;
3014}
3015
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003016/* compare two struct sockaddr_storage and return:
3017 * 0 (true) if the addr is the same in both
3018 * 1 (false) if the addr is not the same in both
3019 * -1 (unable) if one of the addr is not AF_INET*
3020 */
3021int ipcmp(struct sockaddr_storage *ss1, struct sockaddr_storage *ss2)
3022{
3023 if ((ss1->ss_family != AF_INET) && (ss1->ss_family != AF_INET6))
3024 return -1;
3025
3026 if ((ss2->ss_family != AF_INET) && (ss2->ss_family != AF_INET6))
3027 return -1;
3028
3029 if (ss1->ss_family != ss2->ss_family)
3030 return 1;
3031
3032 switch (ss1->ss_family) {
3033 case AF_INET:
3034 return memcmp(&((struct sockaddr_in *)ss1)->sin_addr,
3035 &((struct sockaddr_in *)ss2)->sin_addr,
3036 sizeof(struct in_addr)) != 0;
3037 case AF_INET6:
3038 return memcmp(&((struct sockaddr_in6 *)ss1)->sin6_addr,
3039 &((struct sockaddr_in6 *)ss2)->sin6_addr,
3040 sizeof(struct in6_addr)) != 0;
3041 }
3042
3043 return 1;
3044}
3045
Baptiste Assmann08396c82016-01-31 00:27:17 +01003046/* copy IP address from <source> into <dest>
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003047 * The caller must allocate and clear <dest> before calling.
3048 * The source must be in either AF_INET or AF_INET6 family, or the destination
3049 * address will be undefined. If the destination address used to hold a port,
3050 * it is preserved, so that this function can be used to switch to another
3051 * address family with no risk. Returns a pointer to the destination.
Baptiste Assmann08396c82016-01-31 00:27:17 +01003052 */
3053struct sockaddr_storage *ipcpy(struct sockaddr_storage *source, struct sockaddr_storage *dest)
3054{
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003055 int prev_port;
3056
3057 prev_port = get_net_port(dest);
3058 memset(dest, 0, sizeof(*dest));
Baptiste Assmann08396c82016-01-31 00:27:17 +01003059 dest->ss_family = source->ss_family;
3060
3061 /* copy new addr and apply it */
3062 switch (source->ss_family) {
3063 case AF_INET:
3064 ((struct sockaddr_in *)dest)->sin_addr.s_addr = ((struct sockaddr_in *)source)->sin_addr.s_addr;
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003065 ((struct sockaddr_in *)dest)->sin_port = prev_port;
Baptiste Assmann08396c82016-01-31 00:27:17 +01003066 break;
3067 case AF_INET6:
3068 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 +01003069 ((struct sockaddr_in6 *)dest)->sin6_port = prev_port;
Baptiste Assmann08396c82016-01-31 00:27:17 +01003070 break;
3071 }
3072
3073 return dest;
3074}
3075
William Lallemand421f5b52012-02-06 18:15:57 +01003076char *human_time(int t, short hz_div) {
3077 static char rv[sizeof("24855d23h")+1]; // longest of "23h59m" and "59m59s"
3078 char *p = rv;
Willy Tarreau761b3d52014-04-14 14:53:06 +02003079 char *end = rv + sizeof(rv);
William Lallemand421f5b52012-02-06 18:15:57 +01003080 int cnt=2; // print two numbers
3081
3082 if (unlikely(t < 0 || hz_div <= 0)) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003083 snprintf(p, end - p, "?");
William Lallemand421f5b52012-02-06 18:15:57 +01003084 return rv;
3085 }
3086
3087 if (unlikely(hz_div > 1))
3088 t /= hz_div;
3089
3090 if (t >= DAY) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003091 p += snprintf(p, end - p, "%dd", t / DAY);
William Lallemand421f5b52012-02-06 18:15:57 +01003092 cnt--;
3093 }
3094
3095 if (cnt && t % DAY / HOUR) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003096 p += snprintf(p, end - p, "%dh", t % DAY / HOUR);
William Lallemand421f5b52012-02-06 18:15:57 +01003097 cnt--;
3098 }
3099
3100 if (cnt && t % HOUR / MINUTE) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003101 p += snprintf(p, end - p, "%dm", t % HOUR / MINUTE);
William Lallemand421f5b52012-02-06 18:15:57 +01003102 cnt--;
3103 }
3104
3105 if ((cnt && t % MINUTE) || !t) // also display '0s'
Willy Tarreau761b3d52014-04-14 14:53:06 +02003106 p += snprintf(p, end - p, "%ds", t % MINUTE / SEC);
William Lallemand421f5b52012-02-06 18:15:57 +01003107
3108 return rv;
3109}
3110
3111const char *monthname[12] = {
3112 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
3113 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"
3114};
3115
3116/* date2str_log: write a date in the format :
3117 * sprintf(str, "%02d/%s/%04d:%02d:%02d:%02d.%03d",
3118 * tm.tm_mday, monthname[tm.tm_mon], tm.tm_year+1900,
3119 * tm.tm_hour, tm.tm_min, tm.tm_sec, (int)date.tv_usec/1000);
3120 *
3121 * without using sprintf. return a pointer to the last char written (\0) or
3122 * NULL if there isn't enough space.
3123 */
Willy Tarreauf16cb412018-09-04 19:08:48 +02003124char *date2str_log(char *dst, const struct tm *tm, const struct timeval *date, size_t size)
William Lallemand421f5b52012-02-06 18:15:57 +01003125{
3126
3127 if (size < 25) /* the size is fixed: 24 chars + \0 */
3128 return NULL;
3129
3130 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003131 if (!dst)
3132 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003133 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003134
William Lallemand421f5b52012-02-06 18:15:57 +01003135 memcpy(dst, monthname[tm->tm_mon], 3); // month
3136 dst += 3;
3137 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003138
William Lallemand421f5b52012-02-06 18:15:57 +01003139 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003140 if (!dst)
3141 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003142 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003143
William Lallemand421f5b52012-02-06 18:15:57 +01003144 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003145 if (!dst)
3146 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003147 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003148
William Lallemand421f5b52012-02-06 18:15:57 +01003149 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003150 if (!dst)
3151 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003152 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003153
William Lallemand421f5b52012-02-06 18:15:57 +01003154 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003155 if (!dst)
3156 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003157 *dst++ = '.';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003158
Willy Tarreau7d9421d2020-02-29 09:08:02 +01003159 dst = utoa_pad((unsigned int)(date->tv_usec/1000)%1000, dst, 4); // milliseconds
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003160 if (!dst)
3161 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003162 *dst = '\0';
3163
3164 return dst;
3165}
3166
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003167/* Base year used to compute leap years */
3168#define TM_YEAR_BASE 1900
3169
3170/* Return the difference in seconds between two times (leap seconds are ignored).
3171 * Retrieved from glibc 2.18 source code.
3172 */
3173static int my_tm_diff(const struct tm *a, const struct tm *b)
3174{
3175 /* Compute intervening leap days correctly even if year is negative.
3176 * Take care to avoid int overflow in leap day calculations,
3177 * but it's OK to assume that A and B are close to each other.
3178 */
3179 int a4 = (a->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (a->tm_year & 3);
3180 int b4 = (b->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (b->tm_year & 3);
3181 int a100 = a4 / 25 - (a4 % 25 < 0);
3182 int b100 = b4 / 25 - (b4 % 25 < 0);
3183 int a400 = a100 >> 2;
3184 int b400 = b100 >> 2;
3185 int intervening_leap_days = (a4 - b4) - (a100 - b100) + (a400 - b400);
3186 int years = a->tm_year - b->tm_year;
3187 int days = (365 * years + intervening_leap_days
3188 + (a->tm_yday - b->tm_yday));
3189 return (60 * (60 * (24 * days + (a->tm_hour - b->tm_hour))
3190 + (a->tm_min - b->tm_min))
3191 + (a->tm_sec - b->tm_sec));
3192}
3193
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003194/* Return the GMT offset for a specific local time.
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003195 * Both t and tm must represent the same time.
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003196 * The string returned has the same format as returned by strftime(... "%z", tm).
3197 * Offsets are kept in an internal cache for better performances.
3198 */
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003199const char *get_gmt_offset(time_t t, struct tm *tm)
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003200{
3201 /* Cache offsets from GMT (depending on whether DST is active or not) */
Christopher Faulet1bc04c72017-10-29 20:14:08 +01003202 static THREAD_LOCAL char gmt_offsets[2][5+1] = { "", "" };
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003203
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003204 char *gmt_offset;
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003205 struct tm tm_gmt;
3206 int diff;
3207 int isdst = tm->tm_isdst;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003208
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003209 /* Pretend DST not active if its status is unknown */
3210 if (isdst < 0)
3211 isdst = 0;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003212
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003213 /* Fetch the offset and initialize it if needed */
3214 gmt_offset = gmt_offsets[isdst & 0x01];
3215 if (unlikely(!*gmt_offset)) {
3216 get_gmtime(t, &tm_gmt);
3217 diff = my_tm_diff(tm, &tm_gmt);
3218 if (diff < 0) {
3219 diff = -diff;
3220 *gmt_offset = '-';
3221 } else {
3222 *gmt_offset = '+';
3223 }
Willy Tarreaue112c8a2019-10-29 10:16:11 +01003224 diff %= 86400U;
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003225 diff /= 60; /* Convert to minutes */
3226 snprintf(gmt_offset+1, 4+1, "%02d%02d", diff/60, diff%60);
3227 }
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003228
Willy Tarreaue112c8a2019-10-29 10:16:11 +01003229 return gmt_offset;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003230}
3231
William Lallemand421f5b52012-02-06 18:15:57 +01003232/* gmt2str_log: write a date in the format :
3233 * "%02d/%s/%04d:%02d:%02d:%02d +0000" without using snprintf
3234 * return a pointer to the last char written (\0) or
3235 * NULL if there isn't enough space.
3236 */
3237char *gmt2str_log(char *dst, struct tm *tm, size_t size)
3238{
Yuxans Yao4e25b012012-10-19 10:36:09 +08003239 if (size < 27) /* the size is fixed: 26 chars + \0 */
William Lallemand421f5b52012-02-06 18:15:57 +01003240 return NULL;
3241
3242 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003243 if (!dst)
3244 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003245 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003246
William Lallemand421f5b52012-02-06 18:15:57 +01003247 memcpy(dst, monthname[tm->tm_mon], 3); // month
3248 dst += 3;
3249 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003250
William Lallemand421f5b52012-02-06 18:15:57 +01003251 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003252 if (!dst)
3253 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003254 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003255
William Lallemand421f5b52012-02-06 18:15:57 +01003256 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003257 if (!dst)
3258 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003259 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003260
William Lallemand421f5b52012-02-06 18:15:57 +01003261 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003262 if (!dst)
3263 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003264 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003265
William Lallemand421f5b52012-02-06 18:15:57 +01003266 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003267 if (!dst)
3268 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003269 *dst++ = ' ';
3270 *dst++ = '+';
3271 *dst++ = '0';
3272 *dst++ = '0';
3273 *dst++ = '0';
3274 *dst++ = '0';
3275 *dst = '\0';
3276
3277 return dst;
3278}
3279
Yuxans Yao4e25b012012-10-19 10:36:09 +08003280/* localdate2str_log: write a date in the format :
3281 * "%02d/%s/%04d:%02d:%02d:%02d +0000(local timezone)" without using snprintf
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003282 * Both t and tm must represent the same time.
3283 * return a pointer to the last char written (\0) or
3284 * NULL if there isn't enough space.
Yuxans Yao4e25b012012-10-19 10:36:09 +08003285 */
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003286char *localdate2str_log(char *dst, time_t t, struct tm *tm, size_t size)
Yuxans Yao4e25b012012-10-19 10:36:09 +08003287{
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003288 const char *gmt_offset;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003289 if (size < 27) /* the size is fixed: 26 chars + \0 */
3290 return NULL;
3291
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003292 gmt_offset = get_gmt_offset(t, tm);
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003293
Yuxans Yao4e25b012012-10-19 10:36:09 +08003294 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003295 if (!dst)
3296 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003297 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003298
Yuxans Yao4e25b012012-10-19 10:36:09 +08003299 memcpy(dst, monthname[tm->tm_mon], 3); // month
3300 dst += 3;
3301 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003302
Yuxans Yao4e25b012012-10-19 10:36:09 +08003303 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003304 if (!dst)
3305 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003306 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003307
Yuxans Yao4e25b012012-10-19 10:36:09 +08003308 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003309 if (!dst)
3310 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003311 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003312
Yuxans Yao4e25b012012-10-19 10:36:09 +08003313 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003314 if (!dst)
3315 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003316 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003317
Yuxans Yao4e25b012012-10-19 10:36:09 +08003318 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003319 if (!dst)
3320 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003321 *dst++ = ' ';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003322
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003323 memcpy(dst, gmt_offset, 5); // Offset from local time to GMT
Yuxans Yao4e25b012012-10-19 10:36:09 +08003324 dst += 5;
3325 *dst = '\0';
3326
3327 return dst;
3328}
3329
Willy Tarreaucb1949b2017-07-19 19:05:29 +02003330/* Returns the number of seconds since 01/01/1970 0:0:0 GMT for GMT date <tm>.
3331 * It is meant as a portable replacement for timegm() for use with valid inputs.
3332 * Returns undefined results for invalid dates (eg: months out of range 0..11).
3333 */
3334time_t my_timegm(const struct tm *tm)
3335{
3336 /* Each month has 28, 29, 30 or 31 days, or 28+N. The date in the year
3337 * is thus (current month - 1)*28 + cumulated_N[month] to count the
3338 * sum of the extra N days for elapsed months. The sum of all these N
3339 * days doesn't exceed 30 for a complete year (366-12*28) so it fits
3340 * in a 5-bit word. This means that with 60 bits we can represent a
3341 * matrix of all these values at once, which is fast and efficient to
3342 * access. The extra February day for leap years is not counted here.
3343 *
3344 * Jan : none = 0 (0)
3345 * Feb : Jan = 3 (3)
3346 * Mar : Jan..Feb = 3 (3 + 0)
3347 * Apr : Jan..Mar = 6 (3 + 0 + 3)
3348 * May : Jan..Apr = 8 (3 + 0 + 3 + 2)
3349 * Jun : Jan..May = 11 (3 + 0 + 3 + 2 + 3)
3350 * Jul : Jan..Jun = 13 (3 + 0 + 3 + 2 + 3 + 2)
3351 * Aug : Jan..Jul = 16 (3 + 0 + 3 + 2 + 3 + 2 + 3)
3352 * Sep : Jan..Aug = 19 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3)
3353 * Oct : Jan..Sep = 21 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2)
3354 * Nov : Jan..Oct = 24 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3)
3355 * Dec : Jan..Nov = 26 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3 + 2)
3356 */
3357 uint64_t extra =
3358 ( 0ULL << 0*5) + ( 3ULL << 1*5) + ( 3ULL << 2*5) + /* Jan, Feb, Mar, */
3359 ( 6ULL << 3*5) + ( 8ULL << 4*5) + (11ULL << 5*5) + /* Apr, May, Jun, */
3360 (13ULL << 6*5) + (16ULL << 7*5) + (19ULL << 8*5) + /* Jul, Aug, Sep, */
3361 (21ULL << 9*5) + (24ULL << 10*5) + (26ULL << 11*5); /* Oct, Nov, Dec, */
3362
3363 unsigned int y = tm->tm_year + 1900;
3364 unsigned int m = tm->tm_mon;
3365 unsigned long days = 0;
3366
3367 /* days since 1/1/1970 for full years */
3368 days += days_since_zero(y) - days_since_zero(1970);
3369
3370 /* days for full months in the current year */
3371 days += 28 * m + ((extra >> (m * 5)) & 0x1f);
3372
3373 /* count + 1 after March for leap years. A leap year is a year multiple
3374 * of 4, unless it's multiple of 100 without being multiple of 400. 2000
3375 * is leap, 1900 isn't, 1904 is.
3376 */
3377 if ((m > 1) && !(y & 3) && ((y % 100) || !(y % 400)))
3378 days++;
3379
3380 days += tm->tm_mday - 1;
3381 return days * 86400ULL + tm->tm_hour * 3600 + tm->tm_min * 60 + tm->tm_sec;
3382}
3383
Thierry Fournier93127942016-01-20 18:49:45 +01003384/* This function check a char. It returns true and updates
3385 * <date> and <len> pointer to the new position if the
3386 * character is found.
3387 */
3388static inline int parse_expect_char(const char **date, int *len, char c)
3389{
3390 if (*len < 1 || **date != c)
3391 return 0;
3392 (*len)--;
3393 (*date)++;
3394 return 1;
3395}
3396
3397/* This function expects a string <str> of len <l>. It return true and updates.
3398 * <date> and <len> if the string matches, otherwise, it returns false.
3399 */
3400static inline int parse_strcmp(const char **date, int *len, char *str, int l)
3401{
3402 if (*len < l || strncmp(*date, str, l) != 0)
3403 return 0;
3404 (*len) -= l;
3405 (*date) += l;
3406 return 1;
3407}
3408
3409/* This macro converts 3 chars name in integer. */
3410#define STR2I3(__a, __b, __c) ((__a) * 65536 + (__b) * 256 + (__c))
3411
3412/* day-name = %x4D.6F.6E ; "Mon", case-sensitive
3413 * / %x54.75.65 ; "Tue", case-sensitive
3414 * / %x57.65.64 ; "Wed", case-sensitive
3415 * / %x54.68.75 ; "Thu", case-sensitive
3416 * / %x46.72.69 ; "Fri", case-sensitive
3417 * / %x53.61.74 ; "Sat", case-sensitive
3418 * / %x53.75.6E ; "Sun", case-sensitive
3419 *
3420 * This array must be alphabetically sorted
3421 */
3422static inline int parse_http_dayname(const char **date, int *len, struct tm *tm)
3423{
3424 if (*len < 3)
3425 return 0;
3426 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3427 case STR2I3('M','o','n'): tm->tm_wday = 1; break;
3428 case STR2I3('T','u','e'): tm->tm_wday = 2; break;
3429 case STR2I3('W','e','d'): tm->tm_wday = 3; break;
3430 case STR2I3('T','h','u'): tm->tm_wday = 4; break;
3431 case STR2I3('F','r','i'): tm->tm_wday = 5; break;
3432 case STR2I3('S','a','t'): tm->tm_wday = 6; break;
3433 case STR2I3('S','u','n'): tm->tm_wday = 7; break;
3434 default: return 0;
3435 }
3436 *len -= 3;
3437 *date += 3;
3438 return 1;
3439}
3440
3441/* month = %x4A.61.6E ; "Jan", case-sensitive
3442 * / %x46.65.62 ; "Feb", case-sensitive
3443 * / %x4D.61.72 ; "Mar", case-sensitive
3444 * / %x41.70.72 ; "Apr", case-sensitive
3445 * / %x4D.61.79 ; "May", case-sensitive
3446 * / %x4A.75.6E ; "Jun", case-sensitive
3447 * / %x4A.75.6C ; "Jul", case-sensitive
3448 * / %x41.75.67 ; "Aug", case-sensitive
3449 * / %x53.65.70 ; "Sep", case-sensitive
3450 * / %x4F.63.74 ; "Oct", case-sensitive
3451 * / %x4E.6F.76 ; "Nov", case-sensitive
3452 * / %x44.65.63 ; "Dec", case-sensitive
3453 *
3454 * This array must be alphabetically sorted
3455 */
3456static inline int parse_http_monthname(const char **date, int *len, struct tm *tm)
3457{
3458 if (*len < 3)
3459 return 0;
3460 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3461 case STR2I3('J','a','n'): tm->tm_mon = 0; break;
3462 case STR2I3('F','e','b'): tm->tm_mon = 1; break;
3463 case STR2I3('M','a','r'): tm->tm_mon = 2; break;
3464 case STR2I3('A','p','r'): tm->tm_mon = 3; break;
3465 case STR2I3('M','a','y'): tm->tm_mon = 4; break;
3466 case STR2I3('J','u','n'): tm->tm_mon = 5; break;
3467 case STR2I3('J','u','l'): tm->tm_mon = 6; break;
3468 case STR2I3('A','u','g'): tm->tm_mon = 7; break;
3469 case STR2I3('S','e','p'): tm->tm_mon = 8; break;
3470 case STR2I3('O','c','t'): tm->tm_mon = 9; break;
3471 case STR2I3('N','o','v'): tm->tm_mon = 10; break;
3472 case STR2I3('D','e','c'): tm->tm_mon = 11; break;
3473 default: return 0;
3474 }
3475 *len -= 3;
3476 *date += 3;
3477 return 1;
3478}
3479
3480/* day-name-l = %x4D.6F.6E.64.61.79 ; "Monday", case-sensitive
3481 * / %x54.75.65.73.64.61.79 ; "Tuesday", case-sensitive
3482 * / %x57.65.64.6E.65.73.64.61.79 ; "Wednesday", case-sensitive
3483 * / %x54.68.75.72.73.64.61.79 ; "Thursday", case-sensitive
3484 * / %x46.72.69.64.61.79 ; "Friday", case-sensitive
3485 * / %x53.61.74.75.72.64.61.79 ; "Saturday", case-sensitive
3486 * / %x53.75.6E.64.61.79 ; "Sunday", case-sensitive
3487 *
3488 * This array must be alphabetically sorted
3489 */
3490static inline int parse_http_ldayname(const char **date, int *len, struct tm *tm)
3491{
3492 if (*len < 6) /* Minimum length. */
3493 return 0;
3494 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3495 case STR2I3('M','o','n'):
3496 RET0_UNLESS(parse_strcmp(date, len, "Monday", 6));
3497 tm->tm_wday = 1;
3498 return 1;
3499 case STR2I3('T','u','e'):
3500 RET0_UNLESS(parse_strcmp(date, len, "Tuesday", 7));
3501 tm->tm_wday = 2;
3502 return 1;
3503 case STR2I3('W','e','d'):
3504 RET0_UNLESS(parse_strcmp(date, len, "Wednesday", 9));
3505 tm->tm_wday = 3;
3506 return 1;
3507 case STR2I3('T','h','u'):
3508 RET0_UNLESS(parse_strcmp(date, len, "Thursday", 8));
3509 tm->tm_wday = 4;
3510 return 1;
3511 case STR2I3('F','r','i'):
3512 RET0_UNLESS(parse_strcmp(date, len, "Friday", 6));
3513 tm->tm_wday = 5;
3514 return 1;
3515 case STR2I3('S','a','t'):
3516 RET0_UNLESS(parse_strcmp(date, len, "Saturday", 8));
3517 tm->tm_wday = 6;
3518 return 1;
3519 case STR2I3('S','u','n'):
3520 RET0_UNLESS(parse_strcmp(date, len, "Sunday", 6));
3521 tm->tm_wday = 7;
3522 return 1;
3523 }
3524 return 0;
3525}
3526
3527/* This function parses exactly 1 digit and returns the numeric value in "digit". */
3528static inline int parse_digit(const char **date, int *len, int *digit)
3529{
3530 if (*len < 1 || **date < '0' || **date > '9')
3531 return 0;
3532 *digit = (**date - '0');
3533 (*date)++;
3534 (*len)--;
3535 return 1;
3536}
3537
3538/* This function parses exactly 2 digits and returns the numeric value in "digit". */
3539static inline int parse_2digit(const char **date, int *len, int *digit)
3540{
3541 int value;
3542
3543 RET0_UNLESS(parse_digit(date, len, &value));
3544 (*digit) = value * 10;
3545 RET0_UNLESS(parse_digit(date, len, &value));
3546 (*digit) += value;
3547
3548 return 1;
3549}
3550
3551/* This function parses exactly 4 digits and returns the numeric value in "digit". */
3552static inline int parse_4digit(const char **date, int *len, int *digit)
3553{
3554 int value;
3555
3556 RET0_UNLESS(parse_digit(date, len, &value));
3557 (*digit) = value * 1000;
3558
3559 RET0_UNLESS(parse_digit(date, len, &value));
3560 (*digit) += value * 100;
3561
3562 RET0_UNLESS(parse_digit(date, len, &value));
3563 (*digit) += value * 10;
3564
3565 RET0_UNLESS(parse_digit(date, len, &value));
3566 (*digit) += value;
3567
3568 return 1;
3569}
3570
3571/* time-of-day = hour ":" minute ":" second
3572 * ; 00:00:00 - 23:59:60 (leap second)
3573 *
3574 * hour = 2DIGIT
3575 * minute = 2DIGIT
3576 * second = 2DIGIT
3577 */
3578static inline int parse_http_time(const char **date, int *len, struct tm *tm)
3579{
3580 RET0_UNLESS(parse_2digit(date, len, &tm->tm_hour)); /* hour 2DIGIT */
3581 RET0_UNLESS(parse_expect_char(date, len, ':')); /* expect ":" */
3582 RET0_UNLESS(parse_2digit(date, len, &tm->tm_min)); /* min 2DIGIT */
3583 RET0_UNLESS(parse_expect_char(date, len, ':')); /* expect ":" */
3584 RET0_UNLESS(parse_2digit(date, len, &tm->tm_sec)); /* sec 2DIGIT */
3585 return 1;
3586}
3587
3588/* From RFC7231
3589 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3590 *
3591 * IMF-fixdate = day-name "," SP date1 SP time-of-day SP GMT
3592 * ; fixed length/zone/capitalization subset of the format
3593 * ; see Section 3.3 of [RFC5322]
3594 *
3595 *
3596 * date1 = day SP month SP year
3597 * ; e.g., 02 Jun 1982
3598 *
3599 * day = 2DIGIT
3600 * year = 4DIGIT
3601 *
3602 * GMT = %x47.4D.54 ; "GMT", case-sensitive
3603 *
3604 * time-of-day = hour ":" minute ":" second
3605 * ; 00:00:00 - 23:59:60 (leap second)
3606 *
3607 * hour = 2DIGIT
3608 * minute = 2DIGIT
3609 * second = 2DIGIT
3610 *
3611 * DIGIT = decimal 0-9
3612 */
3613int parse_imf_date(const char *date, int len, struct tm *tm)
3614{
David Carlier327298c2016-11-20 10:42:38 +00003615 /* tm_gmtoff, if present, ought to be zero'ed */
3616 memset(tm, 0, sizeof(*tm));
3617
Thierry Fournier93127942016-01-20 18:49:45 +01003618 RET0_UNLESS(parse_http_dayname(&date, &len, tm)); /* day-name */
3619 RET0_UNLESS(parse_expect_char(&date, &len, ',')); /* expect "," */
3620 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3621 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
3622 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3623 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* Month */
3624 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3625 RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
3626 tm->tm_year -= 1900;
3627 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3628 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3629 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3630 RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3)); /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
3631 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003632 return 1;
3633}
3634
3635/* From RFC7231
3636 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3637 *
3638 * rfc850-date = day-name-l "," SP date2 SP time-of-day SP GMT
3639 * date2 = day "-" month "-" 2DIGIT
3640 * ; e.g., 02-Jun-82
3641 *
3642 * day = 2DIGIT
3643 */
3644int parse_rfc850_date(const char *date, int len, struct tm *tm)
3645{
3646 int year;
3647
David Carlier327298c2016-11-20 10:42:38 +00003648 /* tm_gmtoff, if present, ought to be zero'ed */
3649 memset(tm, 0, sizeof(*tm));
3650
Thierry Fournier93127942016-01-20 18:49:45 +01003651 RET0_UNLESS(parse_http_ldayname(&date, &len, tm)); /* Read the day name */
3652 RET0_UNLESS(parse_expect_char(&date, &len, ',')); /* expect "," */
3653 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3654 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
3655 RET0_UNLESS(parse_expect_char(&date, &len, '-')); /* expect "-" */
3656 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* Month */
3657 RET0_UNLESS(parse_expect_char(&date, &len, '-')); /* expect "-" */
3658
3659 /* year = 2DIGIT
3660 *
3661 * Recipients of a timestamp value in rfc850-(*date) format, which uses a
3662 * two-digit year, MUST interpret a timestamp that appears to be more
3663 * than 50 years in the future as representing the most recent year in
3664 * the past that had the same last two digits.
3665 */
3666 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_year));
3667
3668 /* expect SP */
3669 if (!parse_expect_char(&date, &len, ' ')) {
3670 /* Maybe we have the date with 4 digits. */
3671 RET0_UNLESS(parse_2digit(&date, &len, &year));
3672 tm->tm_year = (tm->tm_year * 100 + year) - 1900;
3673 /* expect SP */
3674 RET0_UNLESS(parse_expect_char(&date, &len, ' '));
3675 } else {
3676 /* I fix 60 as pivot: >60: +1900, <60: +2000. Note that the
3677 * tm_year is the number of year since 1900, so for +1900, we
3678 * do nothing, and for +2000, we add 100.
3679 */
3680 if (tm->tm_year <= 60)
3681 tm->tm_year += 100;
3682 }
3683
3684 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3685 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3686 RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3)); /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
3687 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003688
3689 return 1;
3690}
3691
3692/* From RFC7231
3693 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3694 *
3695 * asctime-date = day-name SP date3 SP time-of-day SP year
3696 * date3 = month SP ( 2DIGIT / ( SP 1DIGIT ))
3697 * ; e.g., Jun 2
3698 *
3699 * HTTP-date is case sensitive. A sender MUST NOT generate additional
3700 * whitespace in an HTTP-date beyond that specifically included as SP in
3701 * the grammar.
3702 */
3703int parse_asctime_date(const char *date, int len, struct tm *tm)
3704{
David Carlier327298c2016-11-20 10:42:38 +00003705 /* tm_gmtoff, if present, ought to be zero'ed */
3706 memset(tm, 0, sizeof(*tm));
3707
Thierry Fournier93127942016-01-20 18:49:45 +01003708 RET0_UNLESS(parse_http_dayname(&date, &len, tm)); /* day-name */
3709 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3710 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* expect month */
3711 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3712
3713 /* expect SP and 1DIGIT or 2DIGIT */
3714 if (parse_expect_char(&date, &len, ' '))
3715 RET0_UNLESS(parse_digit(&date, &len, &tm->tm_mday));
3716 else
3717 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday));
3718
3719 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3720 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3721 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3722 RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
3723 tm->tm_year -= 1900;
3724 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003725 return 1;
3726}
3727
3728/* From RFC7231
3729 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3730 *
3731 * HTTP-date = IMF-fixdate / obs-date
3732 * obs-date = rfc850-date / asctime-date
3733 *
3734 * parses an HTTP date in the RFC format and is accepted
3735 * alternatives. <date> is the strinf containing the date,
3736 * len is the len of the string. <tm> is filled with the
3737 * parsed time. We must considers this time as GMT.
3738 */
3739int parse_http_date(const char *date, int len, struct tm *tm)
3740{
3741 if (parse_imf_date(date, len, tm))
3742 return 1;
3743
3744 if (parse_rfc850_date(date, len, tm))
3745 return 1;
3746
3747 if (parse_asctime_date(date, len, tm))
3748 return 1;
3749
3750 return 0;
3751}
3752
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003753/* Dynamically allocates a string of the proper length to hold the formatted
3754 * output. NULL is returned on error. The caller is responsible for freeing the
3755 * memory area using free(). The resulting string is returned in <out> if the
3756 * pointer is not NULL. A previous version of <out> might be used to build the
3757 * new string, and it will be freed before returning if it is not NULL, which
3758 * makes it possible to build complex strings from iterative calls without
3759 * having to care about freeing intermediate values, as in the example below :
3760 *
3761 * memprintf(&err, "invalid argument: '%s'", arg);
3762 * ...
3763 * memprintf(&err, "parser said : <%s>\n", *err);
3764 * ...
3765 * free(*err);
3766 *
3767 * This means that <err> must be initialized to NULL before first invocation.
3768 * The return value also holds the allocated string, which eases error checking
3769 * and immediate consumption. If the output pointer is not used, NULL must be
Willy Tarreaueb6cead2012-09-20 19:43:14 +02003770 * passed instead and it will be ignored. The returned message will then also
3771 * be NULL so that the caller does not have to bother with freeing anything.
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003772 *
3773 * It is also convenient to use it without any free except the last one :
3774 * err = NULL;
3775 * if (!fct1(err)) report(*err);
3776 * if (!fct2(err)) report(*err);
3777 * if (!fct3(err)) report(*err);
3778 * free(*err);
Christopher Faulet93a518f2017-10-24 11:25:33 +02003779 *
3780 * memprintf relies on memvprintf. This last version can be called from any
3781 * function with variadic arguments.
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003782 */
Christopher Faulet93a518f2017-10-24 11:25:33 +02003783char *memvprintf(char **out, const char *format, va_list orig_args)
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003784{
3785 va_list args;
3786 char *ret = NULL;
3787 int allocated = 0;
3788 int needed = 0;
3789
Willy Tarreaueb6cead2012-09-20 19:43:14 +02003790 if (!out)
3791 return NULL;
3792
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003793 do {
Willy Tarreaue0609f52019-03-29 19:13:23 +01003794 char buf1;
3795
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003796 /* vsnprintf() will return the required length even when the
3797 * target buffer is NULL. We do this in a loop just in case
3798 * intermediate evaluations get wrong.
3799 */
Christopher Faulet93a518f2017-10-24 11:25:33 +02003800 va_copy(args, orig_args);
Willy Tarreaue0609f52019-03-29 19:13:23 +01003801 needed = vsnprintf(ret ? ret : &buf1, allocated, format, args);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003802 va_end(args);
Willy Tarreau1b2fed62013-04-01 22:48:54 +02003803 if (needed < allocated) {
3804 /* Note: on Solaris 8, the first iteration always
3805 * returns -1 if allocated is zero, so we force a
3806 * retry.
3807 */
3808 if (!allocated)
3809 needed = 0;
3810 else
3811 break;
3812 }
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003813
Willy Tarreau1b2fed62013-04-01 22:48:54 +02003814 allocated = needed + 1;
Hubert Verstraete831962e2016-06-28 22:44:26 +02003815 ret = my_realloc2(ret, allocated);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003816 } while (ret);
3817
3818 if (needed < 0) {
3819 /* an error was encountered */
3820 free(ret);
3821 ret = NULL;
3822 }
3823
3824 if (out) {
3825 free(*out);
3826 *out = ret;
3827 }
3828
3829 return ret;
3830}
William Lallemand421f5b52012-02-06 18:15:57 +01003831
Christopher Faulet93a518f2017-10-24 11:25:33 +02003832char *memprintf(char **out, const char *format, ...)
3833{
3834 va_list args;
3835 char *ret = NULL;
3836
3837 va_start(args, format);
3838 ret = memvprintf(out, format, args);
3839 va_end(args);
3840
3841 return ret;
3842}
3843
Willy Tarreau21c705b2012-09-14 11:40:36 +02003844/* Used to add <level> spaces before each line of <out>, unless there is only one line.
3845 * The input argument is automatically freed and reassigned. The result will have to be
Willy Tarreau70eec382012-10-10 08:56:47 +02003846 * freed by the caller. It also supports being passed a NULL which results in the same
3847 * output.
Willy Tarreau21c705b2012-09-14 11:40:36 +02003848 * Example of use :
3849 * parse(cmd, &err); (callee: memprintf(&err, ...))
3850 * fprintf(stderr, "Parser said: %s\n", indent_error(&err));
3851 * free(err);
3852 */
3853char *indent_msg(char **out, int level)
3854{
3855 char *ret, *in, *p;
3856 int needed = 0;
3857 int lf = 0;
3858 int lastlf = 0;
3859 int len;
3860
Willy Tarreau70eec382012-10-10 08:56:47 +02003861 if (!out || !*out)
3862 return NULL;
3863
Willy Tarreau21c705b2012-09-14 11:40:36 +02003864 in = *out - 1;
3865 while ((in = strchr(in + 1, '\n')) != NULL) {
3866 lastlf = in - *out;
3867 lf++;
3868 }
3869
3870 if (!lf) /* single line, no LF, return it as-is */
3871 return *out;
3872
3873 len = strlen(*out);
3874
3875 if (lf == 1 && lastlf == len - 1) {
3876 /* single line, LF at end, strip it and return as-is */
3877 (*out)[lastlf] = 0;
3878 return *out;
3879 }
3880
3881 /* OK now we have at least one LF, we need to process the whole string
3882 * as a multi-line string. What we'll do :
3883 * - prefix with an LF if there is none
3884 * - add <level> spaces before each line
3885 * This means at most ( 1 + level + (len-lf) + lf*<1+level) ) =
3886 * 1 + level + len + lf * level = 1 + level * (lf + 1) + len.
3887 */
3888
3889 needed = 1 + level * (lf + 1) + len + 1;
3890 p = ret = malloc(needed);
3891 in = *out;
3892
3893 /* skip initial LFs */
3894 while (*in == '\n')
3895 in++;
3896
3897 /* copy each line, prefixed with LF and <level> spaces, and without the trailing LF */
3898 while (*in) {
3899 *p++ = '\n';
3900 memset(p, ' ', level);
3901 p += level;
3902 do {
3903 *p++ = *in++;
3904 } while (*in && *in != '\n');
3905 if (*in)
3906 in++;
3907 }
3908 *p = 0;
3909
3910 free(*out);
3911 *out = ret;
3912
3913 return ret;
3914}
3915
Willy Tarreaua2c99112019-08-21 13:17:37 +02003916/* makes a copy of message <in> into <out>, with each line prefixed with <pfx>
3917 * and end of lines replaced with <eol> if not 0. The first line to indent has
3918 * to be indicated in <first> (starts at zero), so that it is possible to skip
3919 * indenting the first line if it has to be appended after an existing message.
3920 * Empty strings are never indented, and NULL strings are considered empty both
3921 * for <in> and <pfx>. It returns non-zero if an EOL was appended as the last
3922 * character, non-zero otherwise.
3923 */
3924int append_prefixed_str(struct buffer *out, const char *in, const char *pfx, char eol, int first)
3925{
3926 int bol, lf;
3927 int pfxlen = pfx ? strlen(pfx) : 0;
3928
3929 if (!in)
3930 return 0;
3931
3932 bol = 1;
3933 lf = 0;
3934 while (*in) {
3935 if (bol && pfxlen) {
3936 if (first > 0)
3937 first--;
3938 else
3939 b_putblk(out, pfx, pfxlen);
3940 bol = 0;
3941 }
3942
3943 lf = (*in == '\n');
3944 bol |= lf;
3945 b_putchr(out, (lf && eol) ? eol : *in);
3946 in++;
3947 }
3948 return lf;
3949}
3950
Willy Tarreau9d22e562019-03-29 18:49:09 +01003951/* removes environment variable <name> from the environment as found in
3952 * environ. This is only provided as an alternative for systems without
3953 * unsetenv() (old Solaris and AIX versions). THIS IS NOT THREAD SAFE.
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05003954 * The principle is to scan environ for each occurrence of variable name
Willy Tarreau9d22e562019-03-29 18:49:09 +01003955 * <name> and to replace the matching pointers with the last pointer of
3956 * the array (since variables are not ordered).
3957 * It always returns 0 (success).
3958 */
3959int my_unsetenv(const char *name)
3960{
3961 extern char **environ;
3962 char **p = environ;
3963 int vars;
3964 int next;
3965 int len;
3966
3967 len = strlen(name);
3968 for (vars = 0; p[vars]; vars++)
3969 ;
3970 next = 0;
3971 while (next < vars) {
3972 if (strncmp(p[next], name, len) != 0 || p[next][len] != '=') {
3973 next++;
3974 continue;
3975 }
3976 if (next < vars - 1)
3977 p[next] = p[vars - 1];
3978 p[--vars] = NULL;
3979 }
3980 return 0;
3981}
3982
Willy Tarreaudad36a32013-03-11 01:20:04 +01003983/* Convert occurrences of environment variables in the input string to their
3984 * corresponding value. A variable is identified as a series of alphanumeric
3985 * characters or underscores following a '$' sign. The <in> string must be
3986 * free()able. NULL returns NULL. The resulting string might be reallocated if
3987 * some expansion is made. Variable names may also be enclosed into braces if
3988 * needed (eg: to concatenate alphanum characters).
3989 */
3990char *env_expand(char *in)
3991{
3992 char *txt_beg;
3993 char *out;
3994 char *txt_end;
3995 char *var_beg;
3996 char *var_end;
3997 char *value;
3998 char *next;
3999 int out_len;
4000 int val_len;
4001
4002 if (!in)
4003 return in;
4004
4005 value = out = NULL;
4006 out_len = 0;
4007
4008 txt_beg = in;
4009 do {
4010 /* look for next '$' sign in <in> */
4011 for (txt_end = txt_beg; *txt_end && *txt_end != '$'; txt_end++);
4012
4013 if (!*txt_end && !out) /* end and no expansion performed */
4014 return in;
4015
4016 val_len = 0;
4017 next = txt_end;
4018 if (*txt_end == '$') {
4019 char save;
4020
4021 var_beg = txt_end + 1;
4022 if (*var_beg == '{')
4023 var_beg++;
4024
4025 var_end = var_beg;
Willy Tarreau90807112020-02-25 08:16:33 +01004026 while (isalnum((unsigned char)*var_end) || *var_end == '_') {
Willy Tarreaudad36a32013-03-11 01:20:04 +01004027 var_end++;
4028 }
4029
4030 next = var_end;
4031 if (*var_end == '}' && (var_beg > txt_end + 1))
4032 next++;
4033
4034 /* get value of the variable name at this location */
4035 save = *var_end;
4036 *var_end = '\0';
4037 value = getenv(var_beg);
4038 *var_end = save;
4039 val_len = value ? strlen(value) : 0;
4040 }
4041
Hubert Verstraete831962e2016-06-28 22:44:26 +02004042 out = my_realloc2(out, out_len + (txt_end - txt_beg) + val_len + 1);
Willy Tarreaudad36a32013-03-11 01:20:04 +01004043 if (txt_end > txt_beg) {
4044 memcpy(out + out_len, txt_beg, txt_end - txt_beg);
4045 out_len += txt_end - txt_beg;
4046 }
4047 if (val_len) {
4048 memcpy(out + out_len, value, val_len);
4049 out_len += val_len;
4050 }
4051 out[out_len] = 0;
4052 txt_beg = next;
4053 } while (*txt_beg);
4054
4055 /* here we know that <out> was allocated and that we don't need <in> anymore */
4056 free(in);
4057 return out;
4058}
4059
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004060
4061/* same as strstr() but case-insensitive and with limit length */
4062const char *strnistr(const char *str1, int len_str1, const char *str2, int len_str2)
4063{
4064 char *pptr, *sptr, *start;
Willy Tarreauc8746532014-05-28 23:05:07 +02004065 unsigned int slen, plen;
4066 unsigned int tmp1, tmp2;
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004067
4068 if (str1 == NULL || len_str1 == 0) // search pattern into an empty string => search is not found
4069 return NULL;
4070
4071 if (str2 == NULL || len_str2 == 0) // pattern is empty => every str1 match
4072 return str1;
4073
4074 if (len_str1 < len_str2) // pattern is longer than string => search is not found
4075 return NULL;
4076
4077 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 +02004078 while (toupper((unsigned char)*start) != toupper((unsigned char)*str2)) {
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004079 start++;
4080 slen--;
4081 tmp1++;
4082
4083 if (tmp1 >= len_str1)
4084 return NULL;
4085
4086 /* if pattern longer than string */
4087 if (slen < plen)
4088 return NULL;
4089 }
4090
4091 sptr = start;
4092 pptr = (char *)str2;
4093
4094 tmp2 = 0;
Willy Tarreauf278eec2020-07-05 21:46:32 +02004095 while (toupper((unsigned char)*sptr) == toupper((unsigned char)*pptr)) {
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004096 sptr++;
4097 pptr++;
4098 tmp2++;
4099
4100 if (*pptr == '\0' || tmp2 == len_str2) /* end of pattern found */
4101 return start;
4102 if (*sptr == '\0' || tmp2 == len_str1) /* end of string found and the pattern is not fully found */
4103 return NULL;
4104 }
4105 }
4106 return NULL;
4107}
4108
Thierry FOURNIER317e1c42014-08-12 10:20:47 +02004109/* This function read the next valid utf8 char.
4110 * <s> is the byte srray to be decode, <len> is its length.
4111 * The function returns decoded char encoded like this:
4112 * The 4 msb are the return code (UTF8_CODE_*), the 4 lsb
4113 * are the length read. The decoded character is stored in <c>.
4114 */
4115unsigned char utf8_next(const char *s, int len, unsigned int *c)
4116{
4117 const unsigned char *p = (unsigned char *)s;
4118 int dec;
4119 unsigned char code = UTF8_CODE_OK;
4120
4121 if (len < 1)
4122 return UTF8_CODE_OK;
4123
4124 /* Check the type of UTF8 sequence
4125 *
4126 * 0... .... 0x00 <= x <= 0x7f : 1 byte: ascii char
4127 * 10.. .... 0x80 <= x <= 0xbf : invalid sequence
4128 * 110. .... 0xc0 <= x <= 0xdf : 2 bytes
4129 * 1110 .... 0xe0 <= x <= 0xef : 3 bytes
4130 * 1111 0... 0xf0 <= x <= 0xf7 : 4 bytes
4131 * 1111 10.. 0xf8 <= x <= 0xfb : 5 bytes
4132 * 1111 110. 0xfc <= x <= 0xfd : 6 bytes
4133 * 1111 111. 0xfe <= x <= 0xff : invalid sequence
4134 */
4135 switch (*p) {
4136 case 0x00 ... 0x7f:
4137 *c = *p;
4138 return UTF8_CODE_OK | 1;
4139
4140 case 0x80 ... 0xbf:
4141 *c = *p;
4142 return UTF8_CODE_BADSEQ | 1;
4143
4144 case 0xc0 ... 0xdf:
4145 if (len < 2) {
4146 *c = *p;
4147 return UTF8_CODE_BADSEQ | 1;
4148 }
4149 *c = *p & 0x1f;
4150 dec = 1;
4151 break;
4152
4153 case 0xe0 ... 0xef:
4154 if (len < 3) {
4155 *c = *p;
4156 return UTF8_CODE_BADSEQ | 1;
4157 }
4158 *c = *p & 0x0f;
4159 dec = 2;
4160 break;
4161
4162 case 0xf0 ... 0xf7:
4163 if (len < 4) {
4164 *c = *p;
4165 return UTF8_CODE_BADSEQ | 1;
4166 }
4167 *c = *p & 0x07;
4168 dec = 3;
4169 break;
4170
4171 case 0xf8 ... 0xfb:
4172 if (len < 5) {
4173 *c = *p;
4174 return UTF8_CODE_BADSEQ | 1;
4175 }
4176 *c = *p & 0x03;
4177 dec = 4;
4178 break;
4179
4180 case 0xfc ... 0xfd:
4181 if (len < 6) {
4182 *c = *p;
4183 return UTF8_CODE_BADSEQ | 1;
4184 }
4185 *c = *p & 0x01;
4186 dec = 5;
4187 break;
4188
4189 case 0xfe ... 0xff:
4190 default:
4191 *c = *p;
4192 return UTF8_CODE_BADSEQ | 1;
4193 }
4194
4195 p++;
4196
4197 while (dec > 0) {
4198
4199 /* need 0x10 for the 2 first bits */
4200 if ( ( *p & 0xc0 ) != 0x80 )
4201 return UTF8_CODE_BADSEQ | ((p-(unsigned char *)s)&0xffff);
4202
4203 /* add data at char */
4204 *c = ( *c << 6 ) | ( *p & 0x3f );
4205
4206 dec--;
4207 p++;
4208 }
4209
4210 /* Check ovelong encoding.
4211 * 1 byte : 5 + 6 : 11 : 0x80 ... 0x7ff
4212 * 2 bytes : 4 + 6 + 6 : 16 : 0x800 ... 0xffff
4213 * 3 bytes : 3 + 6 + 6 + 6 : 21 : 0x10000 ... 0x1fffff
4214 */
Thierry FOURNIER9e7ec082015-03-12 19:32:38 +01004215 if (( *c <= 0x7f && (p-(unsigned char *)s) > 1) ||
Thierry FOURNIER317e1c42014-08-12 10:20:47 +02004216 (*c >= 0x80 && *c <= 0x7ff && (p-(unsigned char *)s) > 2) ||
4217 (*c >= 0x800 && *c <= 0xffff && (p-(unsigned char *)s) > 3) ||
4218 (*c >= 0x10000 && *c <= 0x1fffff && (p-(unsigned char *)s) > 4))
4219 code |= UTF8_CODE_OVERLONG;
4220
4221 /* Check invalid UTF8 range. */
4222 if ((*c >= 0xd800 && *c <= 0xdfff) ||
4223 (*c >= 0xfffe && *c <= 0xffff))
4224 code |= UTF8_CODE_INVRANGE;
4225
4226 return code | ((p-(unsigned char *)s)&0x0f);
4227}
4228
Maxime de Roucydc887852016-05-13 23:52:54 +02004229/* append a copy of string <str> (in a wordlist) at the end of the list <li>
4230 * On failure : return 0 and <err> filled with an error message.
4231 * The caller is responsible for freeing the <err> and <str> copy
4232 * memory area using free()
4233 */
4234int list_append_word(struct list *li, const char *str, char **err)
4235{
4236 struct wordlist *wl;
4237
4238 wl = calloc(1, sizeof(*wl));
4239 if (!wl) {
4240 memprintf(err, "out of memory");
4241 goto fail_wl;
4242 }
4243
4244 wl->s = strdup(str);
4245 if (!wl->s) {
4246 memprintf(err, "out of memory");
4247 goto fail_wl_s;
4248 }
4249
4250 LIST_ADDQ(li, &wl->list);
4251
4252 return 1;
4253
4254fail_wl_s:
4255 free(wl->s);
4256fail_wl:
4257 free(wl);
4258 return 0;
4259}
4260
Willy Tarreau37101052019-05-20 16:48:20 +02004261/* indicates if a memory location may safely be read or not. The trick consists
4262 * in performing a harmless syscall using this location as an input and letting
4263 * the operating system report whether it's OK or not. For this we have the
4264 * stat() syscall, which will return EFAULT when the memory location supposed
4265 * to contain the file name is not readable. If it is readable it will then
4266 * either return 0 if the area contains an existing file name, or -1 with
4267 * another code. This must not be abused, and some audit systems might detect
4268 * this as abnormal activity. It's used only for unsafe dumps.
4269 */
4270int may_access(const void *ptr)
4271{
4272 struct stat buf;
4273
4274 if (stat(ptr, &buf) == 0)
4275 return 1;
4276 if (errno == EFAULT)
4277 return 0;
4278 return 1;
4279}
4280
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004281/* print a string of text buffer to <out>. The format is :
4282 * Non-printable chars \t, \n, \r and \e are * encoded in C format.
4283 * Other non-printable chars are encoded "\xHH". Space, '\', and '=' are also escaped.
4284 * Print stopped if null char or <bsize> is reached, or if no more place in the chunk.
4285 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004286int dump_text(struct buffer *out, const char *buf, int bsize)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004287{
4288 unsigned char c;
4289 int ptr = 0;
4290
4291 while (buf[ptr] && ptr < bsize) {
4292 c = buf[ptr];
Willy Tarreau90807112020-02-25 08:16:33 +01004293 if (isprint((unsigned char)c) && isascii((unsigned char)c) && c != '\\' && c != ' ' && c != '=') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004294 if (out->data > out->size - 1)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004295 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004296 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004297 }
4298 else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\' || c == ' ' || c == '=') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004299 if (out->data > out->size - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004300 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004301 out->area[out->data++] = '\\';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004302 switch (c) {
4303 case ' ': c = ' '; break;
4304 case '\t': c = 't'; break;
4305 case '\n': c = 'n'; break;
4306 case '\r': c = 'r'; break;
4307 case '\e': c = 'e'; break;
4308 case '\\': c = '\\'; break;
4309 case '=': c = '='; break;
4310 }
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004311 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004312 }
4313 else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004314 if (out->data > out->size - 4)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004315 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004316 out->area[out->data++] = '\\';
4317 out->area[out->data++] = 'x';
4318 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4319 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004320 }
4321 ptr++;
4322 }
4323
4324 return ptr;
4325}
4326
4327/* print a buffer in hexa.
4328 * Print stopped if <bsize> is reached, or if no more place in the chunk.
4329 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004330int dump_binary(struct buffer *out, const char *buf, int bsize)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004331{
4332 unsigned char c;
4333 int ptr = 0;
4334
4335 while (ptr < bsize) {
4336 c = buf[ptr];
4337
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004338 if (out->data > out->size - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004339 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004340 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4341 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004342
4343 ptr++;
4344 }
4345 return ptr;
4346}
4347
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004348/* Appends into buffer <out> a hex dump of memory area <buf> for <len> bytes,
4349 * prepending each line with prefix <pfx>. The output is *not* initialized.
4350 * The output will not wrap pas the buffer's end so it is more optimal if the
4351 * caller makes sure the buffer is aligned first. A trailing zero will always
4352 * be appended (and not counted) if there is room for it. The caller must make
Willy Tarreau37101052019-05-20 16:48:20 +02004353 * sure that the area is dumpable first. If <unsafe> is non-null, the memory
4354 * locations are checked first for being readable.
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004355 */
Willy Tarreau37101052019-05-20 16:48:20 +02004356void dump_hex(struct buffer *out, const char *pfx, const void *buf, int len, int unsafe)
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004357{
4358 const unsigned char *d = buf;
4359 int i, j, start;
4360
4361 d = (const unsigned char *)(((unsigned long)buf) & -16);
4362 start = ((unsigned long)buf) & 15;
4363
4364 for (i = 0; i < start + len; i += 16) {
4365 chunk_appendf(out, (sizeof(void *) == 4) ? "%s%8p: " : "%s%16p: ", pfx, d + i);
4366
Willy Tarreau37101052019-05-20 16:48:20 +02004367 // 0: unchecked, 1: checked safe, 2: danger
4368 unsafe = !!unsafe;
4369 if (unsafe && !may_access(d + i))
4370 unsafe = 2;
4371
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004372 for (j = 0; j < 16; j++) {
Willy Tarreau37101052019-05-20 16:48:20 +02004373 if ((i + j < start) || (i + j >= start + len))
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004374 chunk_strcat(out, "'' ");
Willy Tarreau37101052019-05-20 16:48:20 +02004375 else if (unsafe > 1)
4376 chunk_strcat(out, "** ");
4377 else
4378 chunk_appendf(out, "%02x ", d[i + j]);
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004379
4380 if (j == 7)
4381 chunk_strcat(out, "- ");
4382 }
4383 chunk_strcat(out, " ");
4384 for (j = 0; j < 16; j++) {
Willy Tarreau37101052019-05-20 16:48:20 +02004385 if ((i + j < start) || (i + j >= start + len))
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004386 chunk_strcat(out, "'");
Willy Tarreau37101052019-05-20 16:48:20 +02004387 else if (unsafe > 1)
4388 chunk_strcat(out, "*");
Willy Tarreau90807112020-02-25 08:16:33 +01004389 else if (isprint((unsigned char)d[i + j]))
Willy Tarreau37101052019-05-20 16:48:20 +02004390 chunk_appendf(out, "%c", d[i + j]);
4391 else
4392 chunk_strcat(out, ".");
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004393 }
4394 chunk_strcat(out, "\n");
4395 }
4396}
4397
Willy Tarreau762fb3e2020-03-03 15:57:10 +01004398/* dumps <pfx> followed by <n> bytes from <addr> in hex form into buffer <buf>
4399 * enclosed in brackets after the address itself, formatted on 14 chars
4400 * including the "0x" prefix. This is meant to be used as a prefix for code
4401 * areas. For example:
4402 * "0x7f10b6557690 [48 c7 c0 0f 00 00 00 0f]"
4403 * It relies on may_access() to know if the bytes are dumpable, otherwise "--"
4404 * is emitted. A NULL <pfx> will be considered empty.
4405 */
4406void dump_addr_and_bytes(struct buffer *buf, const char *pfx, const void *addr, int n)
4407{
4408 int ok = 0;
4409 int i;
4410
4411 chunk_appendf(buf, "%s%#14lx [", pfx ? pfx : "", (long)addr);
4412
4413 for (i = 0; i < n; i++) {
4414 if (i == 0 || (((long)(addr + i) ^ (long)(addr)) & 4096))
4415 ok = may_access(addr + i);
4416 if (ok)
4417 chunk_appendf(buf, "%02x%s", ((uint8_t*)addr)[i], (i<n-1) ? " " : "]");
4418 else
4419 chunk_appendf(buf, "--%s", (i<n-1) ? " " : "]");
4420 }
4421}
4422
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004423/* print a line of text buffer (limited to 70 bytes) to <out>. The format is :
4424 * <2 spaces> <offset=5 digits> <space or plus> <space> <70 chars max> <\n>
4425 * which is 60 chars per line. Non-printable chars \t, \n, \r and \e are
4426 * encoded in C format. Other non-printable chars are encoded "\xHH". Original
4427 * lines are respected within the limit of 70 output chars. Lines that are
4428 * continuation of a previous truncated line begin with "+" instead of " "
4429 * after the offset. The new pointer is returned.
4430 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004431int dump_text_line(struct buffer *out, const char *buf, int bsize, int len,
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004432 int *line, int ptr)
4433{
4434 int end;
4435 unsigned char c;
4436
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004437 end = out->data + 80;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004438 if (end > out->size)
4439 return ptr;
4440
4441 chunk_appendf(out, " %05d%c ", ptr, (ptr == *line) ? ' ' : '+');
4442
4443 while (ptr < len && ptr < bsize) {
4444 c = buf[ptr];
Willy Tarreau90807112020-02-25 08:16:33 +01004445 if (isprint((unsigned char)c) && isascii((unsigned char)c) && c != '\\') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004446 if (out->data > end - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004447 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004448 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004449 } else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004450 if (out->data > end - 3)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004451 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004452 out->area[out->data++] = '\\';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004453 switch (c) {
4454 case '\t': c = 't'; break;
4455 case '\n': c = 'n'; break;
4456 case '\r': c = 'r'; break;
4457 case '\e': c = 'e'; break;
4458 case '\\': c = '\\'; break;
4459 }
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004460 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004461 } else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004462 if (out->data > end - 5)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004463 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004464 out->area[out->data++] = '\\';
4465 out->area[out->data++] = 'x';
4466 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4467 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004468 }
4469 if (buf[ptr++] == '\n') {
4470 /* we had a line break, let's return now */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004471 out->area[out->data++] = '\n';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004472 *line = ptr;
4473 return ptr;
4474 }
4475 }
4476 /* we have an incomplete line, we return it as-is */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004477 out->area[out->data++] = '\n';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004478 return ptr;
4479}
4480
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004481/* displays a <len> long memory block at <buf>, assuming first byte of <buf>
Willy Tarreaued936c52017-04-27 18:03:20 +02004482 * has address <baseaddr>. String <pfx> may be placed as a prefix in front of
4483 * each line. It may be NULL if unused. The output is emitted to file <out>.
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004484 */
Willy Tarreaued936c52017-04-27 18:03:20 +02004485void debug_hexdump(FILE *out, const char *pfx, const char *buf,
4486 unsigned int baseaddr, int len)
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004487{
Willy Tarreau73459792017-04-11 07:58:08 +02004488 unsigned int i;
4489 int b, j;
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004490
4491 for (i = 0; i < (len + (baseaddr & 15)); i += 16) {
4492 b = i - (baseaddr & 15);
Willy Tarreaued936c52017-04-27 18:03:20 +02004493 fprintf(out, "%s%08x: ", pfx ? pfx : "", i + (baseaddr & ~15));
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004494 for (j = 0; j < 8; j++) {
4495 if (b + j >= 0 && b + j < len)
4496 fprintf(out, "%02x ", (unsigned char)buf[b + j]);
4497 else
4498 fprintf(out, " ");
4499 }
4500
4501 if (b + j >= 0 && b + j < len)
4502 fputc('-', out);
4503 else
4504 fputc(' ', out);
4505
4506 for (j = 8; j < 16; j++) {
4507 if (b + j >= 0 && b + j < len)
4508 fprintf(out, " %02x", (unsigned char)buf[b + j]);
4509 else
4510 fprintf(out, " ");
4511 }
4512
4513 fprintf(out, " ");
4514 for (j = 0; j < 16; j++) {
4515 if (b + j >= 0 && b + j < len) {
4516 if (isprint((unsigned char)buf[b + j]))
4517 fputc((unsigned char)buf[b + j], out);
4518 else
4519 fputc('.', out);
4520 }
4521 else
4522 fputc(' ', out);
4523 }
4524 fputc('\n', out);
4525 }
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004526}
4527
Willy Tarreaubb869862020-04-16 10:52:41 +02004528/* Tries to report the executable path name on platforms supporting this. If
4529 * not found or not possible, returns NULL.
4530 */
4531const char *get_exec_path()
4532{
4533 const char *ret = NULL;
4534
4535#if (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 16))
4536 long execfn = getauxval(AT_EXECFN);
4537
4538 if (execfn && execfn != ENOENT)
4539 ret = (const char *)execfn;
4540#endif
4541 return ret;
4542}
4543
Baruch Siache1651b22020-07-24 07:52:20 +03004544#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreau9133e482020-03-04 10:19:36 +01004545/* calls dladdr() or dladdr1() on <addr> and <dli>. If dladdr1 is available,
4546 * also returns the symbol size in <size>, otherwise returns 0 there.
4547 */
4548static int dladdr_and_size(const void *addr, Dl_info *dli, size_t *size)
4549{
4550 int ret;
Willy Tarreau62af9c82020-03-10 07:51:48 +01004551#if (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 3)) // most detailed one
Willy Tarreau9133e482020-03-04 10:19:36 +01004552 const ElfW(Sym) *sym;
4553
4554 ret = dladdr1(addr, dli, (void **)&sym, RTLD_DL_SYMENT);
4555 if (ret)
4556 *size = sym ? sym->st_size : 0;
4557#else
4558 ret = dladdr(addr, dli);
4559 *size = 0;
4560#endif
4561 return ret;
4562}
4563#endif
4564
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004565/* Tries to append to buffer <buf> some indications about the symbol at address
4566 * <addr> using the following form:
4567 * lib:+0xoffset (unresolvable address from lib's base)
4568 * main+0xoffset (unresolvable address from main (+/-))
4569 * lib:main+0xoffset (unresolvable lib address from main (+/-))
4570 * name (resolved exact exec address)
4571 * lib:name (resolved exact lib address)
4572 * name+0xoffset/0xsize (resolved address within exec symbol)
4573 * lib:name+0xoffset/0xsize (resolved address within lib symbol)
4574 *
4575 * The file name (lib or executable) is limited to what lies between the last
4576 * '/' and the first following '.'. An optional prefix <pfx> is prepended before
4577 * 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 +03004578 * that contains the "main" symbol, or when __ELF__ && USE_DL are not set.
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004579 *
4580 * The symbol's base address is returned, or NULL when unresolved, in order to
4581 * allow the caller to match it against known ones.
4582 */
Willy Tarreau0c439d82020-07-05 20:26:04 +02004583const void *resolve_sym_name(struct buffer *buf, const char *pfx, void *addr)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004584{
4585 const struct {
4586 const void *func;
4587 const char *name;
4588 } fcts[] = {
4589 { .func = process_stream, .name = "process_stream" },
4590 { .func = task_run_applet, .name = "task_run_applet" },
4591 { .func = si_cs_io_cb, .name = "si_cs_io_cb" },
4592 { .func = conn_fd_handler, .name = "conn_fd_handler" },
4593 { .func = dgram_fd_handler, .name = "dgram_fd_handler" },
4594 { .func = listener_accept, .name = "listener_accept" },
4595 { .func = poller_pipe_io_handler, .name = "poller_pipe_io_handler" },
4596 { .func = mworker_accept_wrapper, .name = "mworker_accept_wrapper" },
4597#ifdef USE_LUA
4598 { .func = hlua_process_task, .name = "hlua_process_task" },
4599#endif
4600#if defined(USE_OPENSSL) && (HA_OPENSSL_VERSION_NUMBER >= 0x1010000fL) && !defined(OPENSSL_NO_ASYNC)
4601 { .func = ssl_async_fd_free, .name = "ssl_async_fd_free" },
4602 { .func = ssl_async_fd_handler, .name = "ssl_async_fd_handler" },
4603#endif
4604 };
4605
Baruch Siache1651b22020-07-24 07:52:20 +03004606#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004607 Dl_info dli, dli_main;
Willy Tarreau9133e482020-03-04 10:19:36 +01004608 size_t size;
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004609 const char *fname, *p;
4610#endif
4611 int i;
4612
4613 if (pfx)
4614 chunk_appendf(buf, "%s", pfx);
4615
4616 for (i = 0; i < sizeof(fcts) / sizeof(fcts[0]); i++) {
4617 if (addr == fcts[i].func) {
4618 chunk_appendf(buf, "%s", fcts[i].name);
4619 return addr;
4620 }
4621 }
4622
Baruch Siache1651b22020-07-24 07:52:20 +03004623#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004624 /* Now let's try to be smarter */
Willy Tarreau9133e482020-03-04 10:19:36 +01004625 if (!dladdr_and_size(addr, &dli, &size))
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004626 goto unknown;
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004627
4628 /* 1. prefix the library name if it's not the same object as the one
4629 * that contains the main function. The name is picked between last '/'
4630 * and first following '.'.
4631 */
4632 if (!dladdr(main, &dli_main))
4633 dli_main.dli_fbase = NULL;
4634
4635 if (dli_main.dli_fbase != dli.dli_fbase) {
4636 fname = dli.dli_fname;
4637 p = strrchr(fname, '/');
4638 if (p++)
4639 fname = p;
4640 p = strchr(fname, '.');
4641 if (!p)
4642 p = fname + strlen(fname);
4643
4644 chunk_appendf(buf, "%.*s:", (int)(long)(p - fname), fname);
4645 }
4646
4647 /* 2. symbol name */
4648 if (dli.dli_sname) {
4649 /* known, dump it and return symbol's address (exact or relative) */
4650 chunk_appendf(buf, "%s", dli.dli_sname);
4651 if (addr != dli.dli_saddr) {
4652 chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_saddr));
Willy Tarreau9133e482020-03-04 10:19:36 +01004653 if (size)
4654 chunk_appendf(buf, "/%#lx", (long)size);
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004655 }
4656 return dli.dli_saddr;
4657 }
4658 else if (dli_main.dli_fbase != dli.dli_fbase) {
4659 /* unresolved symbol from a known library, report relative offset */
4660 chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_fbase));
4661 return NULL;
4662 }
Baruch Siache1651b22020-07-24 07:52:20 +03004663#endif /* __ELF__ && !__linux__ || USE_DL */
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004664 unknown:
4665 /* unresolved symbol from the main file, report relative offset to main */
4666 if ((void*)addr < (void*)main)
4667 chunk_appendf(buf, "main-%#lx", (long)((void*)main - addr));
4668 else
4669 chunk_appendf(buf, "main+%#lx", (long)(addr - (void*)main));
4670 return NULL;
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004671}
4672
Frédéric Lécaille3b717162019-02-25 15:04:22 +01004673/*
4674 * Allocate an array of unsigned int with <nums> as address from <str> string
Ilya Shipitsin46a030c2020-07-05 16:36:08 +05004675 * made of integer separated by dot characters.
Frédéric Lécaille3b717162019-02-25 15:04:22 +01004676 *
4677 * First, initializes the value with <sz> as address to 0 and initializes the
4678 * array with <nums> as address to NULL. Then allocates the array with <nums> as
4679 * address updating <sz> pointed value to the size of this array.
4680 *
4681 * Returns 1 if succeeded, 0 if not.
4682 */
4683int parse_dotted_uints(const char *str, unsigned int **nums, size_t *sz)
4684{
4685 unsigned int *n;
4686 const char *s, *end;
4687
4688 s = str;
4689 *sz = 0;
4690 end = str + strlen(str);
4691 *nums = n = NULL;
4692
4693 while (1) {
4694 unsigned int r;
4695
4696 if (s >= end)
4697 break;
4698
4699 r = read_uint(&s, end);
4700 /* Expected characters after having read an uint: '\0' or '.',
4701 * if '.', must not be terminal.
4702 */
4703 if (*s != '\0'&& (*s++ != '.' || s == end))
4704 return 0;
4705
Frédéric Lécaille12a71842019-02-26 18:19:48 +01004706 n = my_realloc2(n, (*sz + 1) * sizeof *n);
Frédéric Lécaille3b717162019-02-25 15:04:22 +01004707 if (!n)
4708 return 0;
4709
4710 n[(*sz)++] = r;
4711 }
4712 *nums = n;
4713
4714 return 1;
4715}
4716
Willy Tarreau4d589e72019-08-23 19:02:26 +02004717
4718/* returns the number of bytes needed to encode <v> as a varint. An inline
4719 * version exists for use with constants (__varint_bytes()).
4720 */
4721int varint_bytes(uint64_t v)
4722{
4723 int len = 1;
4724
4725 if (v >= 240) {
4726 v = (v - 240) >> 4;
4727 while (1) {
4728 len++;
4729 if (v < 128)
4730 break;
4731 v = (v - 128) >> 7;
4732 }
4733 }
4734 return len;
4735}
4736
Willy Tarreau52bf8392020-03-08 00:42:37 +01004737
4738/* Random number generator state, see below */
Willy Tarreau1544c142020-03-12 00:31:18 +01004739static uint64_t ha_random_state[2] ALIGNED(2*sizeof(uint64_t));
Willy Tarreau52bf8392020-03-08 00:42:37 +01004740
4741/* This is a thread-safe implementation of xoroshiro128** described below:
4742 * http://prng.di.unimi.it/
4743 * It features a 2^128 long sequence, returns 64 high-quality bits on each call,
4744 * supports fast jumps and passes all common quality tests. It is thread-safe,
4745 * uses a double-cas on 64-bit architectures supporting it, and falls back to a
4746 * local lock on other ones.
4747 */
4748uint64_t ha_random64()
4749{
4750 uint64_t result;
Willy Tarreau1544c142020-03-12 00:31:18 +01004751 uint64_t old[2] ALIGNED(2*sizeof(uint64_t));
4752 uint64_t new[2] ALIGNED(2*sizeof(uint64_t));
Willy Tarreau52bf8392020-03-08 00:42:37 +01004753
4754#if defined(USE_THREAD) && (!defined(HA_CAS_IS_8B) || !defined(HA_HAVE_CAS_DW))
4755 static HA_SPINLOCK_T rand_lock;
4756
4757 HA_SPIN_LOCK(OTHER_LOCK, &rand_lock);
4758#endif
4759
4760 old[0] = ha_random_state[0];
4761 old[1] = ha_random_state[1];
4762
4763#if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
4764 do {
4765#endif
4766 result = rotl64(old[0] * 5, 7) * 9;
4767 new[1] = old[0] ^ old[1];
4768 new[0] = rotl64(old[0], 24) ^ new[1] ^ (new[1] << 16); // a, b
4769 new[1] = rotl64(new[1], 37); // c
4770
4771#if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
4772 } while (unlikely(!_HA_ATOMIC_DWCAS(ha_random_state, old, new)));
4773#else
4774 ha_random_state[0] = new[0];
4775 ha_random_state[1] = new[1];
4776#if defined(USE_THREAD)
4777 HA_SPIN_UNLOCK(OTHER_LOCK, &rand_lock);
4778#endif
4779#endif
4780 return result;
4781}
4782
4783/* seeds the random state using up to <len> bytes from <seed>, starting with
4784 * the first non-zero byte.
4785 */
4786void ha_random_seed(const unsigned char *seed, size_t len)
4787{
4788 size_t pos;
4789
4790 /* the seed must not be all zeroes, so we pre-fill it with alternating
4791 * bits and overwrite part of them with the block starting at the first
4792 * non-zero byte from the seed.
4793 */
4794 memset(ha_random_state, 0x55, sizeof(ha_random_state));
4795
4796 for (pos = 0; pos < len; pos++)
4797 if (seed[pos] != 0)
4798 break;
4799
4800 if (pos == len)
4801 return;
4802
4803 seed += pos;
4804 len -= pos;
4805
4806 if (len > sizeof(ha_random_state))
4807 len = sizeof(ha_random_state);
4808
4809 memcpy(ha_random_state, seed, len);
4810}
4811
4812/* This causes a jump to (dist * 2^96) places in the pseudo-random sequence,
4813 * and is equivalent to calling ha_random64() as many times. It is used to
4814 * provide non-overlapping sequences of 2^96 numbers (~7*10^28) to up to 2^32
4815 * different generators (i.e. different processes after a fork). The <dist>
4816 * argument is the distance to jump to and is used in a loop so it rather not
4817 * be too large if the processing time is a concern.
4818 *
4819 * BEWARE: this function is NOT thread-safe and must not be called during
4820 * concurrent accesses to ha_random64().
4821 */
4822void ha_random_jump96(uint32_t dist)
4823{
4824 while (dist--) {
4825 uint64_t s0 = 0;
4826 uint64_t s1 = 0;
4827 int b;
4828
4829 for (b = 0; b < 64; b++) {
4830 if ((0xd2a98b26625eee7bULL >> b) & 1) {
4831 s0 ^= ha_random_state[0];
4832 s1 ^= ha_random_state[1];
4833 }
4834 ha_random64();
4835 }
4836
4837 for (b = 0; b < 64; b++) {
4838 if ((0xdddf9b1090aa7ac1ULL >> b) & 1) {
4839 s0 ^= ha_random_state[0];
4840 s1 ^= ha_random_state[1];
4841 }
4842 ha_random64();
4843 }
4844 ha_random_state[0] = s0;
4845 ha_random_state[1] = s1;
4846 }
4847}
4848
Willy Tarreauee3bcdd2020-03-08 17:48:17 +01004849/* Generates an RFC4122 UUID into chunk <output> which must be at least 37
4850 * bytes large.
4851 */
4852void ha_generate_uuid(struct buffer *output)
4853{
4854 uint32_t rnd[4];
4855 uint64_t last;
4856
4857 last = ha_random64();
4858 rnd[0] = last;
4859 rnd[1] = last >> 32;
4860
4861 last = ha_random64();
4862 rnd[2] = last;
4863 rnd[3] = last >> 32;
4864
4865 chunk_printf(output, "%8.8x-%4.4x-%4.4x-%4.4x-%12.12llx",
4866 rnd[0],
4867 rnd[1] & 0xFFFF,
4868 ((rnd[1] >> 16u) & 0xFFF) | 0x4000, // highest 4 bits indicate the uuid version
4869 (rnd[2] & 0x3FFF) | 0x8000, // the highest 2 bits indicate the UUID variant (10),
4870 (long long)((rnd[2] >> 14u) | ((uint64_t) rnd[3] << 18u)) & 0xFFFFFFFFFFFFull);
4871}
4872
4873
Willy Tarreauc8d167b2020-06-16 16:27:26 +02004874/* only used by parse_line() below. It supports writing in place provided that
4875 * <in> is updated to the next location before calling it. In that case, the
4876 * char at <in> may be overwritten.
4877 */
4878#define EMIT_CHAR(x) \
4879 do { \
4880 char __c = (char)(x); \
4881 if ((opts & PARSE_OPT_INPLACE) && out+outpos > in) \
4882 err |= PARSE_ERR_OVERLAP; \
4883 if (outpos >= outmax) \
4884 err |= PARSE_ERR_TOOLARGE; \
4885 if (!err) \
4886 out[outpos] = __c; \
4887 outpos++; \
4888 } while (0)
4889
Ilya Shipitsin46a030c2020-07-05 16:36:08 +05004890/* Parse <in>, copy it into <out> split into isolated words whose pointers
Willy Tarreauc8d167b2020-06-16 16:27:26 +02004891 * are put in <args>. If more than <outlen> bytes have to be emitted, the
4892 * extraneous ones are not emitted but <outlen> is updated so that the caller
4893 * knows how much to realloc. Similarly, <args> are not updated beyond <nbargs>
4894 * but the returned <nbargs> indicates how many were found. All trailing args
Willy Tarreau61dd44b2020-06-25 07:35:42 +02004895 * up to <nbargs> point to the trailing zero, and as long as <nbargs> is > 0,
4896 * it is guaranteed that at least one arg will point to the zero. It is safe
4897 * to call it with a NULL <args> if <nbargs> is 0.
Willy Tarreauc8d167b2020-06-16 16:27:26 +02004898 *
4899 * <out> may overlap with <in> provided that it never goes further, in which
4900 * case the parser will accept to perform in-place parsing and unquoting/
4901 * unescaping but only if environment variables do not lead to expansion that
4902 * causes overlapping, otherwise the input string being destroyed, the error
4903 * will not be recoverable. Note that even during out-of-place <in> will
4904 * experience temporary modifications in-place for variable resolution and must
4905 * be writable, and will also receive zeroes to delimit words when using
4906 * in-place copy. Parsing options <opts> taken from PARSE_OPT_*. Return value
4907 * is zero on success otherwise a bitwise-or of PARSE_ERR_*. Upon error, the
4908 * starting point of the first invalid character sequence or unmatched
4909 * quote/brace is reported in <errptr> if not NULL. When using in-place parsing
4910 * error reporting might be difficult since zeroes will have been inserted into
4911 * the string. One solution for the caller may consist in replacing all args
4912 * delimiters with spaces in this case.
4913 */
4914uint32_t parse_line(char *in, char *out, size_t *outlen, char **args, int *nbargs, uint32_t opts, char **errptr)
4915{
4916 char *quote = NULL;
4917 char *brace = NULL;
4918 unsigned char hex1, hex2;
4919 size_t outmax = *outlen;
Willy Tarreau61dd44b2020-06-25 07:35:42 +02004920 int argsmax = *nbargs - 1;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02004921 size_t outpos = 0;
4922 int squote = 0;
4923 int dquote = 0;
4924 int arg = 0;
4925 uint32_t err = 0;
4926
4927 *nbargs = 0;
4928 *outlen = 0;
4929
Willy Tarreau61dd44b2020-06-25 07:35:42 +02004930 /* argsmax may be -1 here, protecting args[] from any write */
4931 if (arg < argsmax)
4932 args[arg] = out;
4933
Willy Tarreauc8d167b2020-06-16 16:27:26 +02004934 while (1) {
4935 if (*in >= '-' && *in != '\\') {
4936 /* speedup: directly send all regular chars starting
4937 * with '-', '.', '/', alnum etc...
4938 */
4939 EMIT_CHAR(*in++);
4940 continue;
4941 }
4942 else if (*in == '\0' || *in == '\n' || *in == '\r') {
4943 /* end of line */
4944 break;
4945 }
4946 else if (*in == '#' && (opts & PARSE_OPT_SHARP) && !squote && !dquote) {
4947 /* comment */
4948 break;
4949 }
4950 else if (*in == '"' && !squote && (opts & PARSE_OPT_DQUOTE)) { /* double quote outside single quotes */
4951 if (dquote) {
4952 dquote = 0;
4953 quote = NULL;
4954 }
4955 else {
4956 dquote = 1;
4957 quote = in;
4958 }
4959 in++;
4960 continue;
4961 }
4962 else if (*in == '\'' && !dquote && (opts & PARSE_OPT_SQUOTE)) { /* single quote outside double quotes */
4963 if (squote) {
4964 squote = 0;
4965 quote = NULL;
4966 }
4967 else {
4968 squote = 1;
4969 quote = in;
4970 }
4971 in++;
4972 continue;
4973 }
4974 else if (*in == '\\' && !squote && (opts & PARSE_OPT_BKSLASH)) {
4975 /* first, we'll replace \\, \<space>, \#, \r, \n, \t, \xXX with their
4976 * C equivalent value but only when they have a special meaning and within
4977 * double quotes for some of them. Other combinations left unchanged (eg: \1).
4978 */
4979 char tosend = *in;
4980
4981 switch (in[1]) {
4982 case ' ':
4983 case '\\':
4984 tosend = in[1];
4985 in++;
4986 break;
4987
4988 case 't':
4989 tosend = '\t';
4990 in++;
4991 break;
4992
4993 case 'n':
4994 tosend = '\n';
4995 in++;
4996 break;
4997
4998 case 'r':
4999 tosend = '\r';
5000 in++;
5001 break;
5002
5003 case '#':
5004 /* escaping of "#" only if comments are supported */
5005 if (opts & PARSE_OPT_SHARP)
5006 in++;
5007 tosend = *in;
5008 break;
5009
5010 case '\'':
5011 /* escaping of "'" only outside single quotes and only if single quotes are supported */
5012 if (opts & PARSE_OPT_SQUOTE && !squote)
5013 in++;
5014 tosend = *in;
5015 break;
5016
5017 case '"':
5018 /* escaping of '"' only outside single quotes and only if double quotes are supported */
5019 if (opts & PARSE_OPT_DQUOTE && !squote)
5020 in++;
5021 tosend = *in;
5022 break;
5023
5024 case '$':
5025 /* escaping of '$' only inside double quotes and only if env supported */
5026 if (opts & PARSE_OPT_ENV && dquote)
5027 in++;
5028 tosend = *in;
5029 break;
5030
5031 case 'x':
5032 if (!ishex(in[2]) || !ishex(in[3])) {
5033 /* invalid or incomplete hex sequence */
5034 err |= PARSE_ERR_HEX;
5035 if (errptr)
5036 *errptr = in;
5037 goto leave;
5038 }
Willy Tarreauf278eec2020-07-05 21:46:32 +02005039 hex1 = toupper((unsigned char)in[2]) - '0';
5040 hex2 = toupper((unsigned char)in[3]) - '0';
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005041 if (hex1 > 9) hex1 -= 'A' - '9' - 1;
5042 if (hex2 > 9) hex2 -= 'A' - '9' - 1;
5043 tosend = (hex1 << 4) + hex2;
5044 in += 3;
5045 break;
5046
5047 default:
5048 /* other combinations are not escape sequences */
5049 break;
5050 }
5051
5052 in++;
5053 EMIT_CHAR(tosend);
5054 }
5055 else if (isspace((unsigned char)*in) && !squote && !dquote) {
5056 /* a non-escaped space is an argument separator */
5057 while (isspace((unsigned char)*in))
5058 in++;
5059 EMIT_CHAR(0);
5060 arg++;
5061 if (arg < argsmax)
5062 args[arg] = out + outpos;
5063 else
5064 err |= PARSE_ERR_TOOMANY;
5065 }
5066 else if (*in == '$' && (opts & PARSE_OPT_ENV) && (dquote || !(opts & PARSE_OPT_DQUOTE))) {
5067 /* environment variables are evaluated anywhere, or only
5068 * inside double quotes if they are supported.
5069 */
5070 char *var_name;
5071 char save_char;
5072 char *value;
5073
5074 in++;
5075
5076 if (*in == '{')
5077 brace = in++;
5078
5079 if (!isalpha((unsigned char)*in) && *in != '_') {
5080 /* unacceptable character in variable name */
5081 err |= PARSE_ERR_VARNAME;
5082 if (errptr)
5083 *errptr = in;
5084 goto leave;
5085 }
5086
5087 var_name = in;
5088 while (isalnum((unsigned char)*in) || *in == '_')
5089 in++;
5090
5091 save_char = *in;
5092 *in = '\0';
5093 value = getenv(var_name);
5094 *in = save_char;
5095
5096 if (brace) {
5097 if (*in != '}') {
5098 /* unmatched brace */
5099 err |= PARSE_ERR_BRACE;
5100 if (errptr)
5101 *errptr = brace;
5102 goto leave;
5103 }
5104 in++;
5105 brace = NULL;
5106 }
5107
5108 if (value) {
5109 while (*value)
5110 EMIT_CHAR(*value++);
5111 }
5112 }
5113 else {
5114 /* any other regular char */
5115 EMIT_CHAR(*in++);
5116 }
5117 }
5118
5119 /* end of output string */
5120 EMIT_CHAR(0);
5121 arg++;
5122
5123 if (quote) {
5124 /* unmatched quote */
5125 err |= PARSE_ERR_QUOTE;
5126 if (errptr)
5127 *errptr = quote;
5128 goto leave;
5129 }
5130 leave:
5131 *nbargs = arg;
5132 *outlen = outpos;
5133
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005134 /* empty all trailing args by making them point to the trailing zero,
5135 * at least the last one in any case.
5136 */
5137 if (arg > argsmax)
5138 arg = argsmax;
5139
5140 while (arg >= 0 && arg <= argsmax)
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005141 args[arg++] = out + outpos - 1;
5142
5143 return err;
5144}
5145#undef EMIT_CHAR
5146
Willy Tarreauc54e5ad2020-06-25 09:15:40 +02005147/* This is used to sanitize an input line that's about to be used for error reporting.
5148 * It will adjust <line> to print approximately <width> chars around <pos>, trying to
5149 * preserve the beginning, with leading or trailing "..." when the line is truncated.
5150 * If non-printable chars are present in the output. It returns the new offset <pos>
5151 * in the modified line. Non-printable characters are replaced with '?'. <width> must
5152 * be at least 6 to support two "..." otherwise the result is undefined. The line
5153 * itself must have at least 7 chars allocated for the same reason.
5154 */
5155size_t sanitize_for_printing(char *line, size_t pos, size_t width)
5156{
5157 size_t shift = 0;
5158 char *out = line;
5159 char *in = line;
5160 char *end = line + width;
5161
5162 if (pos >= width) {
5163 /* if we have to shift, we'll be out of context, so let's
5164 * try to put <pos> at the center of width.
5165 */
5166 shift = pos - width / 2;
5167 in += shift + 3;
5168 end = out + width - 3;
5169 out[0] = out[1] = out[2] = '.';
5170 out += 3;
5171 }
5172
5173 while (out < end && *in) {
5174 if (isspace((unsigned char)*in))
5175 *out++ = ' ';
5176 else if (isprint((unsigned char)*in))
5177 *out++ = *in;
5178 else
5179 *out++ = '?';
5180 in++;
5181 }
5182
5183 if (end < line + width) {
5184 out[0] = out[1] = out[2] = '.';
5185 out += 3;
5186 }
5187
5188 *out++ = 0;
5189 return pos - shift;
5190}
5191
Willy Tarreaubaaee002006-06-26 02:48:02 +02005192/*
5193 * Local variables:
5194 * c-indent-level: 8
5195 * c-basic-offset: 8
5196 * End:
5197 */