blob: c45358570bfbabf8fa7977205330316676b8da5a [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 Tarreauf268ee82020-06-04 17:05:57 +020045#include <haproxy/global.h>
Willy Tarreau86416052020-06-04 09:20:54 +020046#include <haproxy/hlua.h>
Willy Tarreau213e9902020-06-04 14:58:24 +020047#include <haproxy/listener.h>
Willy Tarreau7a00efb2020-06-02 17:02:59 +020048#include <haproxy/namespace.h>
Christopher Faulet9553de72021-02-26 09:12:50 +010049#include <haproxy/net_helper.h>
Willy Tarreau5fc93282020-09-16 18:25:03 +020050#include <haproxy/protocol.h>
Emeric Brunc9437992021-02-12 19:42:55 +010051#include <haproxy/resolvers.h>
Willy Tarreau586f71b2020-12-11 15:54:36 +010052#include <haproxy/sock.h>
Willy Tarreau209108d2020-06-04 20:30:20 +020053#include <haproxy/ssl_sock.h>
Willy Tarreau5e539c92020-06-04 20:45:39 +020054#include <haproxy/stream_interface.h>
Willy Tarreaucea0e1b2020-06-04 17:25:40 +020055#include <haproxy/task.h>
Willy Tarreau48fbcae2020-06-03 18:09:46 +020056#include <haproxy/tools.h>
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +010057
Thierry Fournier93127942016-01-20 18:49:45 +010058/* This macro returns false if the test __x is false. Many
59 * of the following parsing function must be abort the processing
60 * if it returns 0, so this macro is useful for writing light code.
61 */
62#define RET0_UNLESS(__x) do { if (!(__x)) return 0; } while (0)
63
Willy Tarreau56adcf22012-12-23 18:00:29 +010064/* enough to store NB_ITOA_STR integers of :
Willy Tarreau72d759c2007-10-25 12:14:10 +020065 * 2^64-1 = 18446744073709551615 or
66 * -2^63 = -9223372036854775808
Willy Tarreaue7239b52009-03-29 13:41:58 +020067 *
68 * The HTML version needs room for adding the 25 characters
69 * '<span class="rls"></span>' around digits at positions 3N+1 in order
70 * to add spacing at up to 6 positions : 18 446 744 073 709 551 615
Willy Tarreau72d759c2007-10-25 12:14:10 +020071 */
Christopher Faulet99bca652017-11-14 16:47:26 +010072THREAD_LOCAL char itoa_str[NB_ITOA_STR][171];
73THREAD_LOCAL int itoa_idx = 0; /* index of next itoa_str to use */
Willy Tarreaubaaee002006-06-26 02:48:02 +020074
Willy Tarreau588297f2014-06-16 15:16:40 +020075/* sometimes we'll need to quote strings (eg: in stats), and we don't expect
76 * to quote strings larger than a max configuration line.
77 */
Christopher Faulet99bca652017-11-14 16:47:26 +010078THREAD_LOCAL char quoted_str[NB_QSTR][QSTR_SIZE + 1];
79THREAD_LOCAL int quoted_idx = 0;
Willy Tarreau588297f2014-06-16 15:16:40 +020080
Willy Tarreau06e69b52021-03-02 14:01:35 +010081/* thread-local PRNG state. It's modified to start from a different sequence
82 * on all threads upon startup. It must not be used or anything beyond getting
83 * statistical values as it's 100% predictable.
84 */
85THREAD_LOCAL unsigned int statistical_prng_state = 2463534242U;
86
Willy Tarreaubaaee002006-06-26 02:48:02 +020087/*
William Lallemande7340ec2012-01-24 11:15:39 +010088 * unsigned long long ASCII representation
89 *
90 * return the last char '\0' or NULL if no enough
91 * space in dst
92 */
93char *ulltoa(unsigned long long n, char *dst, size_t size)
94{
95 int i = 0;
96 char *res;
97
98 switch(n) {
99 case 1ULL ... 9ULL:
100 i = 0;
101 break;
102
103 case 10ULL ... 99ULL:
104 i = 1;
105 break;
106
107 case 100ULL ... 999ULL:
108 i = 2;
109 break;
110
111 case 1000ULL ... 9999ULL:
112 i = 3;
113 break;
114
115 case 10000ULL ... 99999ULL:
116 i = 4;
117 break;
118
119 case 100000ULL ... 999999ULL:
120 i = 5;
121 break;
122
123 case 1000000ULL ... 9999999ULL:
124 i = 6;
125 break;
126
127 case 10000000ULL ... 99999999ULL:
128 i = 7;
129 break;
130
131 case 100000000ULL ... 999999999ULL:
132 i = 8;
133 break;
134
135 case 1000000000ULL ... 9999999999ULL:
136 i = 9;
137 break;
138
139 case 10000000000ULL ... 99999999999ULL:
140 i = 10;
141 break;
142
143 case 100000000000ULL ... 999999999999ULL:
144 i = 11;
145 break;
146
147 case 1000000000000ULL ... 9999999999999ULL:
148 i = 12;
149 break;
150
151 case 10000000000000ULL ... 99999999999999ULL:
152 i = 13;
153 break;
154
155 case 100000000000000ULL ... 999999999999999ULL:
156 i = 14;
157 break;
158
159 case 1000000000000000ULL ... 9999999999999999ULL:
160 i = 15;
161 break;
162
163 case 10000000000000000ULL ... 99999999999999999ULL:
164 i = 16;
165 break;
166
167 case 100000000000000000ULL ... 999999999999999999ULL:
168 i = 17;
169 break;
170
171 case 1000000000000000000ULL ... 9999999999999999999ULL:
172 i = 18;
173 break;
174
175 case 10000000000000000000ULL ... ULLONG_MAX:
176 i = 19;
177 break;
178 }
179 if (i + 2 > size) // (i + 1) + '\0'
180 return NULL; // too long
181 res = dst + i + 1;
182 *res = '\0';
183 for (; i >= 0; i--) {
184 dst[i] = n % 10ULL + '0';
185 n /= 10ULL;
186 }
187 return res;
188}
189
190/*
191 * unsigned long ASCII representation
192 *
193 * return the last char '\0' or NULL if no enough
194 * space in dst
195 */
196char *ultoa_o(unsigned long n, char *dst, size_t size)
197{
198 int i = 0;
199 char *res;
200
201 switch (n) {
202 case 0U ... 9UL:
203 i = 0;
204 break;
205
206 case 10U ... 99UL:
207 i = 1;
208 break;
209
210 case 100U ... 999UL:
211 i = 2;
212 break;
213
214 case 1000U ... 9999UL:
215 i = 3;
216 break;
217
218 case 10000U ... 99999UL:
219 i = 4;
220 break;
221
222 case 100000U ... 999999UL:
223 i = 5;
224 break;
225
226 case 1000000U ... 9999999UL:
227 i = 6;
228 break;
229
230 case 10000000U ... 99999999UL:
231 i = 7;
232 break;
233
234 case 100000000U ... 999999999UL:
235 i = 8;
236 break;
237#if __WORDSIZE == 32
238
239 case 1000000000ULL ... ULONG_MAX:
240 i = 9;
241 break;
242
243#elif __WORDSIZE == 64
244
245 case 1000000000ULL ... 9999999999UL:
246 i = 9;
247 break;
248
249 case 10000000000ULL ... 99999999999UL:
250 i = 10;
251 break;
252
253 case 100000000000ULL ... 999999999999UL:
254 i = 11;
255 break;
256
257 case 1000000000000ULL ... 9999999999999UL:
258 i = 12;
259 break;
260
261 case 10000000000000ULL ... 99999999999999UL:
262 i = 13;
263 break;
264
265 case 100000000000000ULL ... 999999999999999UL:
266 i = 14;
267 break;
268
269 case 1000000000000000ULL ... 9999999999999999UL:
270 i = 15;
271 break;
272
273 case 10000000000000000ULL ... 99999999999999999UL:
274 i = 16;
275 break;
276
277 case 100000000000000000ULL ... 999999999999999999UL:
278 i = 17;
279 break;
280
281 case 1000000000000000000ULL ... 9999999999999999999UL:
282 i = 18;
283 break;
284
285 case 10000000000000000000ULL ... ULONG_MAX:
286 i = 19;
287 break;
288
289#endif
290 }
291 if (i + 2 > size) // (i + 1) + '\0'
292 return NULL; // too long
293 res = dst + i + 1;
294 *res = '\0';
295 for (; i >= 0; i--) {
296 dst[i] = n % 10U + '0';
297 n /= 10U;
298 }
299 return res;
300}
301
302/*
303 * signed long ASCII representation
304 *
305 * return the last char '\0' or NULL if no enough
306 * space in dst
307 */
308char *ltoa_o(long int n, char *dst, size_t size)
309{
310 char *pos = dst;
311
312 if (n < 0) {
313 if (size < 3)
314 return NULL; // min size is '-' + digit + '\0' but another test in ultoa
315 *pos = '-';
316 pos++;
317 dst = ultoa_o(-n, pos, size - 1);
318 } else {
319 dst = ultoa_o(n, dst, size);
320 }
321 return dst;
322}
323
324/*
325 * signed long long ASCII representation
326 *
327 * return the last char '\0' or NULL if no enough
328 * space in dst
329 */
330char *lltoa(long long n, char *dst, size_t size)
331{
332 char *pos = dst;
333
334 if (n < 0) {
335 if (size < 3)
336 return NULL; // min size is '-' + digit + '\0' but another test in ulltoa
337 *pos = '-';
338 pos++;
339 dst = ulltoa(-n, pos, size - 1);
340 } else {
341 dst = ulltoa(n, dst, size);
342 }
343 return dst;
344}
345
346/*
347 * write a ascii representation of a unsigned into dst,
348 * return a pointer to the last character
349 * Pad the ascii representation with '0', using size.
350 */
351char *utoa_pad(unsigned int n, char *dst, size_t size)
352{
353 int i = 0;
354 char *ret;
355
356 switch(n) {
357 case 0U ... 9U:
358 i = 0;
359 break;
360
361 case 10U ... 99U:
362 i = 1;
363 break;
364
365 case 100U ... 999U:
366 i = 2;
367 break;
368
369 case 1000U ... 9999U:
370 i = 3;
371 break;
372
373 case 10000U ... 99999U:
374 i = 4;
375 break;
376
377 case 100000U ... 999999U:
378 i = 5;
379 break;
380
381 case 1000000U ... 9999999U:
382 i = 6;
383 break;
384
385 case 10000000U ... 99999999U:
386 i = 7;
387 break;
388
389 case 100000000U ... 999999999U:
390 i = 8;
391 break;
392
393 case 1000000000U ... 4294967295U:
394 i = 9;
395 break;
396 }
397 if (i + 2 > size) // (i + 1) + '\0'
398 return NULL; // too long
399 if (i < size)
400 i = size - 2; // padding - '\0'
401
402 ret = dst + i + 1;
403 *ret = '\0';
404 for (; i >= 0; i--) {
405 dst[i] = n % 10U + '0';
406 n /= 10U;
407 }
408 return ret;
409}
410
411/*
Willy Tarreaubaaee002006-06-26 02:48:02 +0200412 * copies at most <size-1> chars from <src> to <dst>. Last char is always
413 * set to 0, unless <size> is 0. The number of chars copied is returned
414 * (excluding the terminating zero).
415 * This code has been optimized for size and speed : on x86, it's 45 bytes
416 * long, uses only registers, and consumes only 4 cycles per char.
417 */
418int strlcpy2(char *dst, const char *src, int size)
419{
420 char *orig = dst;
421 if (size) {
422 while (--size && (*dst = *src)) {
423 src++; dst++;
424 }
425 *dst = 0;
426 }
427 return dst - orig;
428}
429
430/*
Willy Tarreau72d759c2007-10-25 12:14:10 +0200431 * This function simply returns a locally allocated string containing
Willy Tarreaubaaee002006-06-26 02:48:02 +0200432 * the ascii representation for number 'n' in decimal.
433 */
Emeric Brun3a7fce52010-01-04 14:54:38 +0100434char *ultoa_r(unsigned long n, char *buffer, int size)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200435{
436 char *pos;
437
Willy Tarreau72d759c2007-10-25 12:14:10 +0200438 pos = buffer + size - 1;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200439 *pos-- = '\0';
440
441 do {
442 *pos-- = '0' + n % 10;
443 n /= 10;
Willy Tarreau72d759c2007-10-25 12:14:10 +0200444 } while (n && pos >= buffer);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200445 return pos + 1;
446}
447
Willy Tarreau91092e52007-10-25 16:58:42 +0200448/*
Willy Tarreaue7239b52009-03-29 13:41:58 +0200449 * This function simply returns a locally allocated string containing
Thierry FOURNIER763a5d82015-07-06 23:09:52 +0200450 * the ascii representation for number 'n' in decimal.
451 */
452char *lltoa_r(long long int in, char *buffer, int size)
453{
454 char *pos;
455 int neg = 0;
456 unsigned long long int n;
457
458 pos = buffer + size - 1;
459 *pos-- = '\0';
460
461 if (in < 0) {
462 neg = 1;
463 n = -in;
464 }
465 else
466 n = in;
467
468 do {
469 *pos-- = '0' + n % 10;
470 n /= 10;
471 } while (n && pos >= buffer);
472 if (neg && pos > buffer)
473 *pos-- = '-';
474 return pos + 1;
475}
476
477/*
478 * This function simply returns a locally allocated string containing
Thierry FOURNIER1480bd82015-06-06 19:14:59 +0200479 * the ascii representation for signed number 'n' in decimal.
480 */
481char *sltoa_r(long n, char *buffer, int size)
482{
483 char *pos;
484
485 if (n >= 0)
486 return ultoa_r(n, buffer, size);
487
488 pos = ultoa_r(-n, buffer + 1, size - 1) - 1;
489 *pos = '-';
490 return pos;
491}
492
493/*
494 * This function simply returns a locally allocated string containing
Willy Tarreaue7239b52009-03-29 13:41:58 +0200495 * the ascii representation for number 'n' in decimal, formatted for
496 * HTML output with tags to create visual grouping by 3 digits. The
497 * output needs to support at least 171 characters.
498 */
499const char *ulltoh_r(unsigned long long n, char *buffer, int size)
500{
501 char *start;
502 int digit = 0;
503
504 start = buffer + size;
505 *--start = '\0';
506
507 do {
508 if (digit == 3 && start >= buffer + 7)
509 memcpy(start -= 7, "</span>", 7);
510
511 if (start >= buffer + 1) {
512 *--start = '0' + n % 10;
513 n /= 10;
514 }
515
516 if (digit == 3 && start >= buffer + 18)
517 memcpy(start -= 18, "<span class=\"rls\">", 18);
518
519 if (digit++ == 3)
520 digit = 1;
521 } while (n && start > buffer);
522 return start;
523}
524
525/*
Willy Tarreau91092e52007-10-25 16:58:42 +0200526 * This function simply returns a locally allocated string containing the ascii
527 * representation for number 'n' in decimal, unless n is 0 in which case it
528 * returns the alternate string (or an empty string if the alternate string is
529 * NULL). It use is intended for limits reported in reports, where it's
530 * desirable not to display anything if there is no limit. Warning! it shares
531 * the same vector as ultoa_r().
532 */
533const char *limit_r(unsigned long n, char *buffer, int size, const char *alt)
534{
535 return (n) ? ultoa_r(n, buffer, size) : (alt ? alt : "");
536}
537
Willy Tarreau588297f2014-06-16 15:16:40 +0200538/* returns a locally allocated string containing the quoted encoding of the
539 * input string. The output may be truncated to QSTR_SIZE chars, but it is
540 * guaranteed that the string will always be properly terminated. Quotes are
541 * encoded by doubling them as is commonly done in CSV files. QSTR_SIZE must
542 * always be at least 4 chars.
543 */
544const char *qstr(const char *str)
545{
546 char *ret = quoted_str[quoted_idx];
547 char *p, *end;
548
549 if (++quoted_idx >= NB_QSTR)
550 quoted_idx = 0;
551
552 p = ret;
553 end = ret + QSTR_SIZE;
554
555 *p++ = '"';
556
557 /* always keep 3 chars to support passing "" and the ending " */
558 while (*str && p < end - 3) {
559 if (*str == '"') {
560 *p++ = '"';
561 *p++ = '"';
562 }
563 else
564 *p++ = *str;
565 str++;
566 }
567 *p++ = '"';
568 return ret;
569}
570
Robert Tsai81ae1952007-12-05 10:47:29 +0100571/*
Willy Tarreaubaaee002006-06-26 02:48:02 +0200572 * Returns non-zero if character <s> is a hex digit (0-9, a-f, A-F), else zero.
573 *
574 * It looks like this one would be a good candidate for inlining, but this is
575 * not interesting because it around 35 bytes long and often called multiple
576 * times within the same function.
577 */
578int ishex(char s)
579{
580 s -= '0';
581 if ((unsigned char)s <= 9)
582 return 1;
583 s -= 'A' - '0';
584 if ((unsigned char)s <= 5)
585 return 1;
586 s -= 'a' - 'A';
587 if ((unsigned char)s <= 5)
588 return 1;
589 return 0;
590}
591
Willy Tarreau3ca1a882015-01-15 18:43:49 +0100592/* rounds <i> down to the closest value having max 2 digits */
593unsigned int round_2dig(unsigned int i)
594{
595 unsigned int mul = 1;
596
597 while (i >= 100) {
598 i /= 10;
599 mul *= 10;
600 }
601 return i * mul;
602}
603
Willy Tarreau2e74c3f2007-12-02 18:45:09 +0100604/*
605 * Checks <name> for invalid characters. Valid chars are [A-Za-z0-9_:.-]. If an
606 * invalid character is found, a pointer to it is returned. If everything is
607 * fine, NULL is returned.
608 */
609const char *invalid_char(const char *name)
610{
611 if (!*name)
612 return name;
613
614 while (*name) {
Willy Tarreau90807112020-02-25 08:16:33 +0100615 if (!isalnum((unsigned char)*name) && *name != '.' && *name != ':' &&
Willy Tarreau2e74c3f2007-12-02 18:45:09 +0100616 *name != '_' && *name != '-')
617 return name;
618 name++;
619 }
620 return NULL;
621}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200622
623/*
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200624 * Checks <name> for invalid characters. Valid chars are [_.-] and those
625 * accepted by <f> function.
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200626 * If an invalid character is found, a pointer to it is returned.
627 * If everything is fine, NULL is returned.
628 */
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200629static inline const char *__invalid_char(const char *name, int (*f)(int)) {
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200630
631 if (!*name)
632 return name;
633
634 while (*name) {
Willy Tarreau90807112020-02-25 08:16:33 +0100635 if (!f((unsigned char)*name) && *name != '.' &&
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200636 *name != '_' && *name != '-')
637 return name;
638
639 name++;
640 }
641
642 return NULL;
643}
644
645/*
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200646 * Checks <name> for invalid characters. Valid chars are [A-Za-z0-9_.-].
647 * If an invalid character is found, a pointer to it is returned.
648 * If everything is fine, NULL is returned.
649 */
650const char *invalid_domainchar(const char *name) {
651 return __invalid_char(name, isalnum);
652}
653
654/*
655 * Checks <name> for invalid characters. Valid chars are [A-Za-z_.-].
656 * If an invalid character is found, a pointer to it is returned.
657 * If everything is fine, NULL is returned.
658 */
659const char *invalid_prefix_char(const char *name) {
Thierry Fournierf7b7c3e2018-03-26 11:54:39 +0200660 return __invalid_char(name, isalnum);
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200661}
662
663/*
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100664 * converts <str> to a struct sockaddr_storage* provided by the caller. The
Willy Tarreau24709282013-03-10 21:32:12 +0100665 * caller must have zeroed <sa> first, and may have set sa->ss_family to force
666 * parse a specific address format. If the ss_family is 0 or AF_UNSPEC, then
667 * the function tries to guess the address family from the syntax. If the
668 * family is forced and the format doesn't match, an error is returned. The
Willy Tarreaufab5a432011-03-04 15:31:53 +0100669 * string is assumed to contain only an address, no port. The address can be a
670 * dotted IPv4 address, an IPv6 address, a host name, or empty or "*" to
671 * indicate INADDR_ANY. NULL is returned if the host part cannot be resolved.
672 * The return address will only have the address family and the address set,
673 * all other fields remain zero. The string is not supposed to be modified.
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100674 * The IPv6 '::' address is IN6ADDR_ANY. If <resolve> is non-zero, the hostname
675 * is resolved, otherwise only IP addresses are resolved, and anything else
Willy Tarreauecde7df2016-11-02 22:37:03 +0100676 * returns NULL. If the address contains a port, this one is preserved.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200677 */
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100678struct sockaddr_storage *str2ip2(const char *str, struct sockaddr_storage *sa, int resolve)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200679{
Willy Tarreaufab5a432011-03-04 15:31:53 +0100680 struct hostent *he;
mildisff5d5102015-10-26 18:50:08 +0100681 /* max IPv6 length, including brackets and terminating NULL */
682 char tmpip[48];
Willy Tarreauecde7df2016-11-02 22:37:03 +0100683 int port = get_host_port(sa);
mildisff5d5102015-10-26 18:50:08 +0100684
685 /* check IPv6 with square brackets */
686 if (str[0] == '[') {
687 size_t iplength = strlen(str);
688
689 if (iplength < 4) {
690 /* minimal size is 4 when using brackets "[::]" */
691 goto fail;
692 }
693 else if (iplength >= sizeof(tmpip)) {
694 /* IPv6 literal can not be larger than tmpip */
695 goto fail;
696 }
697 else {
698 if (str[iplength - 1] != ']') {
699 /* if address started with bracket, it should end with bracket */
700 goto fail;
701 }
702 else {
703 memcpy(tmpip, str + 1, iplength - 2);
704 tmpip[iplength - 2] = '\0';
705 str = tmpip;
706 }
707 }
708 }
Willy Tarreaufab5a432011-03-04 15:31:53 +0100709
Willy Tarreaufab5a432011-03-04 15:31:53 +0100710 /* Any IPv6 address */
711 if (str[0] == ':' && str[1] == ':' && !str[2]) {
Willy Tarreau24709282013-03-10 21:32:12 +0100712 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
713 sa->ss_family = AF_INET6;
714 else if (sa->ss_family != AF_INET6)
715 goto fail;
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
Willy Tarreau24709282013-03-10 21:32:12 +0100720 /* Any address for the family, defaults to IPv4 */
Willy Tarreaufab5a432011-03-04 15:31:53 +0100721 if (!str[0] || (str[0] == '*' && !str[1])) {
Willy Tarreau24709282013-03-10 21:32:12 +0100722 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
723 sa->ss_family = AF_INET;
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 /* check for IPv6 first */
Willy Tarreau24709282013-03-10 21:32:12 +0100729 if ((!sa->ss_family || sa->ss_family == AF_UNSPEC || sa->ss_family == AF_INET6) &&
730 inet_pton(AF_INET6, str, &((struct sockaddr_in6 *)sa)->sin6_addr)) {
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100731 sa->ss_family = AF_INET6;
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
736 /* then check for IPv4 */
Willy Tarreau24709282013-03-10 21:32:12 +0100737 if ((!sa->ss_family || sa->ss_family == AF_UNSPEC || sa->ss_family == AF_INET) &&
738 inet_pton(AF_INET, str, &((struct sockaddr_in *)sa)->sin_addr)) {
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100739 sa->ss_family = AF_INET;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100740 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100741 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100742 }
743
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100744 if (!resolve)
745 return NULL;
746
Emeric Brund30e9a12020-12-23 18:49:16 +0100747 if (!resolv_hostname_validation(str, NULL))
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200748 return NULL;
749
David du Colombierd5f43282011-03-17 10:40:16 +0100750#ifdef USE_GETADDRINFO
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200751 if (global.tune.options & GTUNE_USE_GAI) {
David du Colombierd5f43282011-03-17 10:40:16 +0100752 struct addrinfo hints, *result;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100753 int success = 0;
David du Colombierd5f43282011-03-17 10:40:16 +0100754
755 memset(&result, 0, sizeof(result));
756 memset(&hints, 0, sizeof(hints));
Willy Tarreau24709282013-03-10 21:32:12 +0100757 hints.ai_family = sa->ss_family ? sa->ss_family : AF_UNSPEC;
David du Colombierd5f43282011-03-17 10:40:16 +0100758 hints.ai_socktype = SOCK_DGRAM;
Dmitry Sivachenkoeab7f392015-10-02 01:01:58 +0200759 hints.ai_flags = 0;
David du Colombierd5f43282011-03-17 10:40:16 +0100760 hints.ai_protocol = 0;
761
762 if (getaddrinfo(str, NULL, &hints, &result) == 0) {
Willy Tarreau24709282013-03-10 21:32:12 +0100763 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
764 sa->ss_family = result->ai_family;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100765 else if (sa->ss_family != result->ai_family) {
766 freeaddrinfo(result);
Willy Tarreau24709282013-03-10 21:32:12 +0100767 goto fail;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100768 }
Willy Tarreau24709282013-03-10 21:32:12 +0100769
David du Colombierd5f43282011-03-17 10:40:16 +0100770 switch (result->ai_family) {
771 case AF_INET:
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100772 memcpy((struct sockaddr_in *)sa, result->ai_addr, result->ai_addrlen);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100773 set_host_port(sa, port);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100774 success = 1;
775 break;
David du Colombierd5f43282011-03-17 10:40:16 +0100776 case AF_INET6:
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100777 memcpy((struct sockaddr_in6 *)sa, result->ai_addr, result->ai_addrlen);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100778 set_host_port(sa, port);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100779 success = 1;
780 break;
David du Colombierd5f43282011-03-17 10:40:16 +0100781 }
782 }
783
Sean Carey58ea0392013-02-15 23:39:18 +0100784 if (result)
785 freeaddrinfo(result);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100786
787 if (success)
788 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100789 }
David du Colombierd5f43282011-03-17 10:40:16 +0100790#endif
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200791 /* try to resolve an IPv4/IPv6 hostname */
792 he = gethostbyname(str);
793 if (he) {
794 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
795 sa->ss_family = he->h_addrtype;
796 else if (sa->ss_family != he->h_addrtype)
797 goto fail;
798
799 switch (sa->ss_family) {
800 case AF_INET:
801 ((struct sockaddr_in *)sa)->sin_addr = *(struct in_addr *) *(he->h_addr_list);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100802 set_host_port(sa, port);
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200803 return sa;
804 case AF_INET6:
805 ((struct sockaddr_in6 *)sa)->sin6_addr = *(struct in6_addr *) *(he->h_addr_list);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100806 set_host_port(sa, port);
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200807 return sa;
808 }
809 }
810
David du Colombierd5f43282011-03-17 10:40:16 +0100811 /* unsupported address family */
Willy Tarreau24709282013-03-10 21:32:12 +0100812 fail:
Willy Tarreaufab5a432011-03-04 15:31:53 +0100813 return NULL;
814}
815
816/*
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100817 * Converts <str> to a locally allocated struct sockaddr_storage *, and a port
818 * range or offset consisting in two integers that the caller will have to
819 * check to find the relevant input format. The following format are supported :
820 *
821 * String format | address | port | low | high
822 * addr | <addr> | 0 | 0 | 0
823 * addr: | <addr> | 0 | 0 | 0
824 * addr:port | <addr> | <port> | <port> | <port>
825 * addr:pl-ph | <addr> | <pl> | <pl> | <ph>
826 * addr:+port | <addr> | <port> | 0 | <port>
827 * addr:-port | <addr> |-<port> | <port> | 0
828 *
829 * The detection of a port range or increment by the caller is made by
830 * comparing <low> and <high>. If both are equal, then port 0 means no port
831 * was specified. The caller may pass NULL for <low> and <high> if it is not
832 * interested in retrieving port ranges.
833 *
834 * Note that <addr> above may also be :
835 * - empty ("") => family will be AF_INET and address will be INADDR_ANY
836 * - "*" => family will be AF_INET and address will be INADDR_ANY
837 * - "::" => family will be AF_INET6 and address will be IN6ADDR_ANY
838 * - a host name => family and address will depend on host name resolving.
839 *
Willy Tarreau24709282013-03-10 21:32:12 +0100840 * A prefix may be passed in before the address above to force the family :
841 * - "ipv4@" => force address to resolve as IPv4 and fail if not possible.
842 * - "ipv6@" => force address to resolve as IPv6 and fail if not possible.
843 * - "unix@" => force address to be a path to a UNIX socket even if the
844 * path does not start with a '/'
Willy Tarreauccfccef2014-05-10 01:49:15 +0200845 * - 'abns@' -> force address to belong to the abstract namespace (Linux
846 * only). These sockets are just like Unix sockets but without
847 * the need for an underlying file system. The address is a
848 * string. Technically it's like a Unix socket with a zero in
849 * the first byte of the address.
Willy Tarreau40aa0702013-03-10 23:51:38 +0100850 * - "fd@" => an integer must follow, and is a file descriptor number.
Willy Tarreau24709282013-03-10 21:32:12 +0100851 *
mildisff5d5102015-10-26 18:50:08 +0100852 * IPv6 addresses can be declared with or without square brackets. When using
853 * square brackets for IPv6 addresses, the port separator (colon) is optional.
854 * If not using square brackets, and in order to avoid any ambiguity with
855 * IPv6 addresses, the last colon ':' is mandatory even when no port is specified.
856 * NULL is returned if the address cannot be parsed. The <low> and <high> ports
857 * are always initialized if non-null, even for non-IP families.
Willy Tarreaud393a622013-03-04 18:22:00 +0100858 *
859 * If <pfx> is non-null, it is used as a string prefix before any path-based
860 * address (typically the path to a unix socket).
Willy Tarreau40aa0702013-03-10 23:51:38 +0100861 *
Willy Tarreau72b8c1f2015-09-08 15:50:19 +0200862 * if <fqdn> is non-null, it will be filled with :
863 * - a pointer to the FQDN of the server name to resolve if there's one, and
864 * that the caller will have to free(),
865 * - NULL if there was an explicit address that doesn't require resolution.
866 *
Willy Tarreaucd3a55912020-09-04 15:30:46 +0200867 * Hostnames are only resolved if <opts> has PA_O_RESOLVE. Otherwise <fqdn> is
868 * still honored so it is possible for the caller to know whether a resolution
869 * failed by clearing this flag and checking if <fqdn> was filled, indicating
870 * the need for a resolution.
Thierry FOURNIER7fe3be72015-09-26 20:03:36 +0200871 *
Willy Tarreau40aa0702013-03-10 23:51:38 +0100872 * When a file descriptor is passed, its value is put into the s_addr part of
Willy Tarreaua5b325f2020-09-04 16:44:20 +0200873 * the address when cast to sockaddr_in and the address family is
874 * AF_CUST_EXISTING_FD.
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200875 *
Willy Tarreau5fc93282020-09-16 18:25:03 +0200876 * The matching protocol will be set into <proto> if non-null.
877 *
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200878 * Any known file descriptor is also assigned to <fd> if non-null, otherwise it
879 * is forced to -1.
Willy Tarreaufab5a432011-03-04 15:31:53 +0100880 */
Willy Tarreau5fc93282020-09-16 18:25:03 +0200881struct sockaddr_storage *str2sa_range(const char *str, int *port, int *low, int *high, int *fd,
882 struct protocol **proto, char **err,
883 const char *pfx, char **fqdn, unsigned int opts)
Willy Tarreaufab5a432011-03-04 15:31:53 +0100884{
Christopher Faulet1bc04c72017-10-29 20:14:08 +0100885 static THREAD_LOCAL struct sockaddr_storage ss;
David du Colombier6f5ccb12011-03-10 22:26:24 +0100886 struct sockaddr_storage *ret = NULL;
Willy Tarreau5fc93282020-09-16 18:25:03 +0200887 struct protocol *new_proto = NULL;
Willy Tarreau24709282013-03-10 21:32:12 +0100888 char *back, *str2;
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100889 char *port1, *port2;
890 int portl, porth, porta;
Willy Tarreauccfccef2014-05-10 01:49:15 +0200891 int abstract = 0;
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200892 int new_fd = -1;
Willy Tarreaue835bd82020-09-16 11:35:47 +0200893 int sock_type, ctrl_type;
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100894
895 portl = porth = porta = 0;
Willy Tarreau72b8c1f2015-09-08 15:50:19 +0200896 if (fqdn)
897 *fqdn = NULL;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200898
Willy Tarreaudad36a32013-03-11 01:20:04 +0100899 str2 = back = env_expand(strdup(str));
Willy Tarreaudf350f12013-03-01 20:22:54 +0100900 if (str2 == NULL) {
901 memprintf(err, "out of memory in '%s'\n", __FUNCTION__);
Willy Tarreaud5191e72010-02-09 20:50:45 +0100902 goto out;
Willy Tarreaudf350f12013-03-01 20:22:54 +0100903 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200904
Willy Tarreau9f69f462015-09-08 16:01:25 +0200905 if (!*str2) {
906 memprintf(err, "'%s' resolves to an empty address (environment variable missing?)\n", str);
907 goto out;
908 }
909
Willy Tarreau24709282013-03-10 21:32:12 +0100910 memset(&ss, 0, sizeof(ss));
911
Willy Tarreaue835bd82020-09-16 11:35:47 +0200912 /* prepare the default socket types */
Willy Tarreauf23b1bc2021-03-23 18:36:37 +0100913 if ((opts & (PA_O_STREAM|PA_O_DGRAM)) == PA_O_DGRAM ||
914 ((opts & (PA_O_STREAM|PA_O_DGRAM)) == (PA_O_DGRAM|PA_O_STREAM) && (opts & PA_O_DEFAULT_DGRAM)))
Willy Tarreaue835bd82020-09-16 11:35:47 +0200915 sock_type = ctrl_type = SOCK_DGRAM;
916 else
917 sock_type = ctrl_type = SOCK_STREAM;
918
919 if (strncmp(str2, "stream+", 7) == 0) {
920 str2 += 7;
921 sock_type = ctrl_type = SOCK_STREAM;
922 }
923 else if (strncmp(str2, "dgram+", 6) == 0) {
924 str2 += 6;
925 sock_type = ctrl_type = SOCK_DGRAM;
926 }
927
Willy Tarreau24709282013-03-10 21:32:12 +0100928 if (strncmp(str2, "unix@", 5) == 0) {
929 str2 += 5;
Willy Tarreauccfccef2014-05-10 01:49:15 +0200930 abstract = 0;
Willy Tarreau24709282013-03-10 21:32:12 +0100931 ss.ss_family = AF_UNIX;
932 }
Emeric Brunce325c42021-04-02 17:05:09 +0200933 else if (strncmp(str2, "uxdg@", 5) == 0) {
934 str2 += 5;
935 abstract = 0;
936 ss.ss_family = AF_UNIX;
937 sock_type = ctrl_type = SOCK_DGRAM;
938 }
939 else if (strncmp(str2, "uxst@", 5) == 0) {
940 str2 += 5;
941 abstract = 0;
942 ss.ss_family = AF_UNIX;
943 sock_type = ctrl_type = SOCK_STREAM;
944 }
Willy Tarreauccfccef2014-05-10 01:49:15 +0200945 else if (strncmp(str2, "abns@", 5) == 0) {
946 str2 += 5;
947 abstract = 1;
948 ss.ss_family = AF_UNIX;
949 }
Emeric Brunce325c42021-04-02 17:05:09 +0200950 else if (strncmp(str2, "ip@", 3) == 0) {
951 str2 += 3;
952 ss.ss_family = AF_UNSPEC;
953 }
Willy Tarreau24709282013-03-10 21:32:12 +0100954 else if (strncmp(str2, "ipv4@", 5) == 0) {
955 str2 += 5;
956 ss.ss_family = AF_INET;
957 }
958 else if (strncmp(str2, "ipv6@", 5) == 0) {
959 str2 += 5;
960 ss.ss_family = AF_INET6;
961 }
Emeric Brunce325c42021-04-02 17:05:09 +0200962 else if (strncmp(str2, "tcp4@", 5) == 0) {
963 str2 += 5;
964 ss.ss_family = AF_INET;
965 sock_type = ctrl_type = SOCK_STREAM;
966 }
Emeric Brun3835c0d2020-07-07 09:46:09 +0200967 else if (strncmp(str2, "udp4@", 5) == 0) {
968 str2 += 5;
969 ss.ss_family = AF_INET;
Willy Tarreaue835bd82020-09-16 11:35:47 +0200970 sock_type = ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +0200971 }
Emeric Brunce325c42021-04-02 17:05:09 +0200972 else if (strncmp(str2, "tcp6@", 5) == 0) {
973 str2 += 5;
974 ss.ss_family = AF_INET6;
975 sock_type = ctrl_type = SOCK_STREAM;
976 }
Emeric Brun3835c0d2020-07-07 09:46:09 +0200977 else if (strncmp(str2, "udp6@", 5) == 0) {
978 str2 += 5;
979 ss.ss_family = AF_INET6;
Willy Tarreaue835bd82020-09-16 11:35:47 +0200980 sock_type = ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +0200981 }
Emeric Brunce325c42021-04-02 17:05:09 +0200982 else if (strncmp(str2, "tcp@", 4) == 0) {
983 str2 += 4;
984 ss.ss_family = AF_UNSPEC;
985 sock_type = ctrl_type = SOCK_STREAM;
986 }
Emeric Brun3835c0d2020-07-07 09:46:09 +0200987 else if (strncmp(str2, "udp@", 4) == 0) {
988 str2 += 4;
989 ss.ss_family = AF_UNSPEC;
Willy Tarreaue835bd82020-09-16 11:35:47 +0200990 sock_type = ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +0200991 }
Frédéric Lécaille10caf652020-11-23 11:36:57 +0100992 else if (strncmp(str2, "quic4@", 6) == 0) {
993 str2 += 6;
994 ss.ss_family = AF_INET;
995 sock_type = SOCK_DGRAM;
996 ctrl_type = SOCK_STREAM;
997 }
998 else if (strncmp(str2, "quic6@", 6) == 0) {
999 str2 += 6;
1000 ss.ss_family = AF_INET6;
1001 sock_type = SOCK_DGRAM;
1002 ctrl_type = SOCK_STREAM;
1003 }
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001004 else if (strncmp(str2, "fd@", 3) == 0) {
1005 str2 += 3;
1006 ss.ss_family = AF_CUST_EXISTING_FD;
1007 }
1008 else if (strncmp(str2, "sockpair@", 9) == 0) {
1009 str2 += 9;
1010 ss.ss_family = AF_CUST_SOCKPAIR;
1011 }
Willy Tarreau24709282013-03-10 21:32:12 +01001012 else if (*str2 == '/') {
1013 ss.ss_family = AF_UNIX;
1014 }
1015 else
1016 ss.ss_family = AF_UNSPEC;
1017
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001018 if (ss.ss_family == AF_CUST_SOCKPAIR) {
Willy Tarreaua215be22020-09-16 10:14:16 +02001019 struct sockaddr_storage ss2;
1020 socklen_t addr_len;
William Lallemand2fe7dd02018-09-11 16:51:29 +02001021 char *endptr;
1022
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001023 new_fd = strtol(str2, &endptr, 10);
1024 if (!*str2 || new_fd < 0 || *endptr) {
William Lallemand2fe7dd02018-09-11 16:51:29 +02001025 memprintf(err, "file descriptor '%s' is not a valid integer in '%s'\n", str2, str);
1026 goto out;
1027 }
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001028
Willy Tarreaua215be22020-09-16 10:14:16 +02001029 /* just verify that it's a socket */
1030 addr_len = sizeof(ss2);
1031 if (getsockname(new_fd, (struct sockaddr *)&ss2, &addr_len) == -1) {
1032 memprintf(err, "cannot use file descriptor '%d' : %s.\n", new_fd, strerror(errno));
1033 goto out;
1034 }
1035
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001036 ((struct sockaddr_in *)&ss)->sin_addr.s_addr = new_fd;
1037 ((struct sockaddr_in *)&ss)->sin_port = 0;
William Lallemand2fe7dd02018-09-11 16:51:29 +02001038 }
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001039 else if (ss.ss_family == AF_CUST_EXISTING_FD) {
Willy Tarreau40aa0702013-03-10 23:51:38 +01001040 char *endptr;
1041
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001042 new_fd = strtol(str2, &endptr, 10);
1043 if (!*str2 || new_fd < 0 || *endptr) {
Willy Tarreaudad36a32013-03-11 01:20:04 +01001044 memprintf(err, "file descriptor '%s' is not a valid integer in '%s'\n", str2, str);
Willy Tarreau40aa0702013-03-10 23:51:38 +01001045 goto out;
1046 }
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001047
Willy Tarreau6edc7222020-09-15 17:41:56 +02001048 if (opts & PA_O_SOCKET_FD) {
1049 socklen_t addr_len;
1050 int type;
1051
1052 addr_len = sizeof(ss);
1053 if (getsockname(new_fd, (struct sockaddr *)&ss, &addr_len) == -1) {
1054 memprintf(err, "cannot use file descriptor '%d' : %s.\n", new_fd, strerror(errno));
1055 goto out;
1056 }
1057
1058 addr_len = sizeof(type);
1059 if (getsockopt(new_fd, SOL_SOCKET, SO_TYPE, &type, &addr_len) != 0 ||
Willy Tarreaue835bd82020-09-16 11:35:47 +02001060 (type == SOCK_STREAM) != (sock_type == SOCK_STREAM)) {
Willy Tarreau6edc7222020-09-15 17:41:56 +02001061 memprintf(err, "socket on file descriptor '%d' is of the wrong type.\n", new_fd);
1062 goto out;
1063 }
1064
1065 porta = portl = porth = get_host_port(&ss);
1066 } else if (opts & PA_O_RAW_FD) {
1067 ((struct sockaddr_in *)&ss)->sin_addr.s_addr = new_fd;
1068 ((struct sockaddr_in *)&ss)->sin_port = 0;
1069 } else {
1070 memprintf(err, "a file descriptor is not acceptable here in '%s'\n", str);
1071 goto out;
1072 }
Willy Tarreau40aa0702013-03-10 23:51:38 +01001073 }
1074 else if (ss.ss_family == AF_UNIX) {
Willy Tarreau8daa9202019-06-16 18:16:33 +02001075 struct sockaddr_un *un = (struct sockaddr_un *)&ss;
Willy Tarreau15586382013-03-04 19:48:14 +01001076 int prefix_path_len;
1077 int max_path_len;
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001078 int adr_len;
Willy Tarreau15586382013-03-04 19:48:14 +01001079
1080 /* complete unix socket path name during startup or soft-restart is
1081 * <unix_bind_prefix><path>.<pid>.<bak|tmp>
1082 */
Willy Tarreauccfccef2014-05-10 01:49:15 +02001083 prefix_path_len = (pfx && !abstract) ? strlen(pfx) : 0;
Willy Tarreau8daa9202019-06-16 18:16:33 +02001084 max_path_len = (sizeof(un->sun_path) - 1) -
Willy Tarreau327ea5a2020-02-11 06:43:37 +01001085 (abstract ? 0 : prefix_path_len + 1 + 5 + 1 + 3);
Willy Tarreau15586382013-03-04 19:48:14 +01001086
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001087 adr_len = strlen(str2);
1088 if (adr_len > max_path_len) {
Willy Tarreau15586382013-03-04 19:48:14 +01001089 memprintf(err, "socket path '%s' too long (max %d)\n", str, max_path_len);
1090 goto out;
1091 }
1092
Willy Tarreauccfccef2014-05-10 01:49:15 +02001093 /* when abstract==1, we skip the first zero and copy all bytes except the trailing zero */
Willy Tarreau8daa9202019-06-16 18:16:33 +02001094 memset(un->sun_path, 0, sizeof(un->sun_path));
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001095 if (prefix_path_len)
Willy Tarreau8daa9202019-06-16 18:16:33 +02001096 memcpy(un->sun_path, pfx, prefix_path_len);
1097 memcpy(un->sun_path + prefix_path_len + abstract, str2, adr_len + 1 - abstract);
Willy Tarreau15586382013-03-04 19:48:14 +01001098 }
Willy Tarreau24709282013-03-10 21:32:12 +01001099 else { /* IPv4 and IPv6 */
mildisff5d5102015-10-26 18:50:08 +01001100 char *end = str2 + strlen(str2);
1101 char *chr;
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001102
mildisff5d5102015-10-26 18:50:08 +01001103 /* search for : or ] whatever comes first */
1104 for (chr = end-1; chr > str2; chr--) {
1105 if (*chr == ']' || *chr == ':')
1106 break;
1107 }
1108
1109 if (*chr == ':') {
1110 /* Found a colon before a closing-bracket, must be a port separator.
1111 * This guarantee backward compatibility.
1112 */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001113 if (!(opts & PA_O_PORT_OK)) {
1114 memprintf(err, "port specification not permitted here in '%s'", str);
1115 goto out;
1116 }
mildisff5d5102015-10-26 18:50:08 +01001117 *chr++ = '\0';
1118 port1 = chr;
1119 }
1120 else {
1121 /* Either no colon and no closing-bracket
1122 * or directly ending with a closing-bracket.
1123 * However, no port.
1124 */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001125 if (opts & PA_O_PORT_MAND) {
1126 memprintf(err, "missing port specification in '%s'", str);
1127 goto out;
1128 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001129 port1 = "";
mildisff5d5102015-10-26 18:50:08 +01001130 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02001131
Willy Tarreau90807112020-02-25 08:16:33 +01001132 if (isdigit((unsigned char)*port1)) { /* single port or range */
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001133 port2 = strchr(port1, '-');
Willy Tarreau7f96a842020-09-15 11:12:44 +02001134 if (port2) {
1135 if (!(opts & PA_O_PORT_RANGE)) {
1136 memprintf(err, "port range not permitted here in '%s'", str);
1137 goto out;
1138 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001139 *port2++ = '\0';
Willy Tarreau7f96a842020-09-15 11:12:44 +02001140 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001141 else
1142 port2 = port1;
1143 portl = atoi(port1);
1144 porth = atoi(port2);
Willy Tarreau7f96a842020-09-15 11:12:44 +02001145
1146 if (portl < !!(opts & PA_O_PORT_MAND) || portl > 65535) {
1147 memprintf(err, "invalid port '%s'", port1);
1148 goto out;
1149 }
1150
1151 if (porth < !!(opts & PA_O_PORT_MAND) || porth > 65535) {
1152 memprintf(err, "invalid port '%s'", port2);
1153 goto out;
1154 }
1155
1156 if (portl > porth) {
1157 memprintf(err, "invalid port range '%d-%d'", portl, porth);
1158 goto out;
1159 }
1160
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001161 porta = portl;
1162 }
1163 else if (*port1 == '-') { /* negative offset */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001164 if (!(opts & PA_O_PORT_OFS)) {
1165 memprintf(err, "port offset not permitted here in '%s'", str);
1166 goto out;
1167 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001168 portl = atoi(port1 + 1);
1169 porta = -portl;
1170 }
1171 else if (*port1 == '+') { /* positive offset */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001172 if (!(opts & PA_O_PORT_OFS)) {
1173 memprintf(err, "port offset not permitted here in '%s'", str);
1174 goto out;
1175 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001176 porth = atoi(port1 + 1);
1177 porta = porth;
1178 }
1179 else if (*port1) { /* other any unexpected char */
Willy Tarreaudad36a32013-03-11 01:20:04 +01001180 memprintf(err, "invalid character '%c' in port number '%s' in '%s'\n", *port1, port1, str);
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001181 goto out;
1182 }
Willy Tarreau7f96a842020-09-15 11:12:44 +02001183 else if (opts & PA_O_PORT_MAND) {
1184 memprintf(err, "missing port specification in '%s'", str);
1185 goto out;
1186 }
Willy Tarreauceccdd72016-11-02 22:27:10 +01001187
1188 /* first try to parse the IP without resolving. If it fails, it
1189 * tells us we need to keep a copy of the FQDN to resolve later
1190 * and to enable DNS. In this case we can proceed if <fqdn> is
Willy Tarreaucd3a55912020-09-04 15:30:46 +02001191 * set or if PA_O_RESOLVE is set, otherwise it's an error.
Willy Tarreauceccdd72016-11-02 22:27:10 +01001192 */
1193 if (str2ip2(str2, &ss, 0) == NULL) {
Willy Tarreaucd3a55912020-09-04 15:30:46 +02001194 if ((!(opts & PA_O_RESOLVE) && !fqdn) ||
1195 ((opts & PA_O_RESOLVE) && str2ip2(str2, &ss, 1) == NULL)) {
Willy Tarreauceccdd72016-11-02 22:27:10 +01001196 memprintf(err, "invalid address: '%s' in '%s'\n", str2, str);
1197 goto out;
1198 }
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001199
Willy Tarreauceccdd72016-11-02 22:27:10 +01001200 if (fqdn) {
1201 if (str2 != back)
1202 memmove(back, str2, strlen(str2) + 1);
1203 *fqdn = back;
1204 back = NULL;
1205 }
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001206 }
Willy Tarreauceccdd72016-11-02 22:27:10 +01001207 set_host_port(&ss, porta);
Willy Tarreaue4c58c82013-03-06 15:28:17 +01001208 }
Willy Tarreaufab5a432011-03-04 15:31:53 +01001209
Willy Tarreaue835bd82020-09-16 11:35:47 +02001210 if (ctrl_type == SOCK_STREAM && !(opts & PA_O_STREAM)) {
1211 memprintf(err, "stream-type socket not acceptable in '%s'\n", str);
1212 goto out;
1213 }
1214 else if (ctrl_type == SOCK_DGRAM && !(opts & PA_O_DGRAM)) {
1215 memprintf(err, "dgram-type socket not acceptable in '%s'\n", str);
1216 goto out;
1217 }
1218
Willy Tarreau65ec4e32020-09-16 19:17:08 +02001219 if (proto || (opts & PA_O_CONNECT)) {
Willy Tarreau5fc93282020-09-16 18:25:03 +02001220 /* Note: if the caller asks for a proto, we must find one,
Emeric Brun26754902021-04-07 14:26:44 +02001221 * except if we inherit from a raw FD (family == AF_CUST_EXISTING_FD)
1222 * orif we return with an fqdn that will resolve later,
Willy Tarreau5fc93282020-09-16 18:25:03 +02001223 * in which case the address is not known yet (this is only
1224 * for servers actually).
1225 */
Willy Tarreaub2ffc992020-09-16 21:37:31 +02001226 new_proto = protocol_lookup(ss.ss_family,
Willy Tarreauaf9609b2020-09-16 22:04:46 +02001227 sock_type == SOCK_DGRAM,
1228 ctrl_type == SOCK_DGRAM);
Willy Tarreaub2ffc992020-09-16 21:37:31 +02001229
Emeric Brun26754902021-04-07 14:26:44 +02001230 if (!new_proto && (!fqdn || !*fqdn) && (ss.ss_family != AF_CUST_EXISTING_FD)) {
Willy Tarreau5fc93282020-09-16 18:25:03 +02001231 memprintf(err, "unsupported protocol family %d for address '%s'", ss.ss_family, str);
1232 goto out;
1233 }
Willy Tarreau65ec4e32020-09-16 19:17:08 +02001234
1235 if ((opts & PA_O_CONNECT) && new_proto && !new_proto->connect) {
1236 memprintf(err, "connect() not supported for this protocol family %d used by address '%s'", ss.ss_family, str);
1237 goto out;
1238 }
Willy Tarreau5fc93282020-09-16 18:25:03 +02001239 }
1240
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001241 ret = &ss;
Willy Tarreaud5191e72010-02-09 20:50:45 +01001242 out:
Willy Tarreau48ef4c92017-01-06 18:32:38 +01001243 if (port)
1244 *port = porta;
Willy Tarreaud4448bc2013-02-20 15:55:15 +01001245 if (low)
1246 *low = portl;
1247 if (high)
1248 *high = porth;
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001249 if (fd)
1250 *fd = new_fd;
Willy Tarreau5fc93282020-09-16 18:25:03 +02001251 if (proto)
1252 *proto = new_proto;
Willy Tarreau24709282013-03-10 21:32:12 +01001253 free(back);
Willy Tarreaud5191e72010-02-09 20:50:45 +01001254 return ret;
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001255}
1256
Thayne McCombs92149f92020-11-20 01:28:26 -07001257/* converts <addr> and <port> into a string representation of the address and port. This is sort
1258 * of an inverse of str2sa_range, with some restrictions. The supported families are AF_INET,
1259 * AF_INET6, AF_UNIX, and AF_CUST_SOCKPAIR. If the family is unsopported NULL is returned.
1260 * If map_ports is true, then the sign of the port is included in the output, to indicate it is
1261 * relative to the incoming port. AF_INET and AF_INET6 will be in the form "<addr>:<port>".
1262 * AF_UNIX will either be just the path (if using a pathname) or "abns@<path>" if it is abstract.
1263 * AF_CUST_SOCKPAIR will be of the form "sockpair@<fd>".
1264 *
1265 * The returned char* is allocated, and it is the responsibility of the caller to free it.
1266 */
1267char * sa2str(const struct sockaddr_storage *addr, int port, int map_ports)
1268{
1269 char buffer[INET6_ADDRSTRLEN];
1270 char *out = NULL;
1271 const void *ptr;
1272 const char *path;
1273
1274 switch (addr->ss_family) {
1275 case AF_INET:
1276 ptr = &((struct sockaddr_in *)addr)->sin_addr;
1277 break;
1278 case AF_INET6:
1279 ptr = &((struct sockaddr_in6 *)addr)->sin6_addr;
1280 break;
1281 case AF_UNIX:
1282 path = ((struct sockaddr_un *)addr)->sun_path;
1283 if (path[0] == '\0') {
1284 const int max_length = sizeof(struct sockaddr_un) - offsetof(struct sockaddr_un, sun_path) - 1;
1285 return memprintf(&out, "abns@%.*s", max_length, path+1);
1286 } else {
1287 return strdup(path);
1288 }
1289 case AF_CUST_SOCKPAIR:
1290 return memprintf(&out, "sockpair@%d", ((struct sockaddr_in *)addr)->sin_addr.s_addr);
1291 default:
1292 return NULL;
1293 }
1294 inet_ntop(addr->ss_family, ptr, buffer, get_addr_len(addr));
1295 if (map_ports)
1296 return memprintf(&out, "%s:%+d", buffer, port);
1297 else
1298 return memprintf(&out, "%s:%d", buffer, port);
1299}
1300
1301
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001302/* converts <str> to a struct in_addr containing a network mask. It can be
1303 * passed in dotted form (255.255.255.0) or in CIDR form (24). It returns 1
Jarno Huuskonen577d5ac2017-05-21 17:32:21 +03001304 * if the conversion succeeds otherwise zero.
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001305 */
1306int str2mask(const char *str, struct in_addr *mask)
1307{
1308 if (strchr(str, '.') != NULL) { /* dotted notation */
1309 if (!inet_pton(AF_INET, str, mask))
1310 return 0;
1311 }
1312 else { /* mask length */
1313 char *err;
1314 unsigned long len = strtol(str, &err, 10);
1315
1316 if (!*str || (err && *err) || (unsigned)len > 32)
1317 return 0;
Tim Duesterhus8575f722018-01-25 16:24:48 +01001318
1319 len2mask4(len, mask);
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001320 }
1321 return 1;
1322}
1323
Tim Duesterhus47185172018-01-25 16:24:49 +01001324/* converts <str> to a struct in6_addr containing a network mask. It can be
Tim Duesterhus5e642862018-02-20 17:02:18 +01001325 * passed in quadruplet form (ffff:ffff::) or in CIDR form (64). It returns 1
Tim Duesterhus47185172018-01-25 16:24:49 +01001326 * if the conversion succeeds otherwise zero.
1327 */
1328int str2mask6(const char *str, struct in6_addr *mask)
1329{
1330 if (strchr(str, ':') != NULL) { /* quadruplet notation */
1331 if (!inet_pton(AF_INET6, str, mask))
1332 return 0;
1333 }
1334 else { /* mask length */
1335 char *err;
1336 unsigned long len = strtol(str, &err, 10);
1337
1338 if (!*str || (err && *err) || (unsigned)len > 128)
1339 return 0;
1340
1341 len2mask6(len, mask);
1342 }
1343 return 1;
1344}
1345
Thierry FOURNIERb0504632013-12-14 15:39:02 +01001346/* convert <cidr> to struct in_addr <mask>. It returns 1 if the conversion
1347 * succeeds otherwise zero.
1348 */
1349int cidr2dotted(int cidr, struct in_addr *mask) {
1350
1351 if (cidr < 0 || cidr > 32)
1352 return 0;
1353
1354 mask->s_addr = cidr ? htonl(~0UL << (32 - cidr)) : 0;
1355 return 1;
1356}
1357
Thierry Fournier70473a52016-02-17 17:12:14 +01001358/* Convert mask from bit length form to in_addr form.
1359 * This function never fails.
1360 */
1361void len2mask4(int len, struct in_addr *addr)
1362{
1363 if (len >= 32) {
1364 addr->s_addr = 0xffffffff;
1365 return;
1366 }
1367 if (len <= 0) {
1368 addr->s_addr = 0x00000000;
1369 return;
1370 }
1371 addr->s_addr = 0xffffffff << (32 - len);
1372 addr->s_addr = htonl(addr->s_addr);
1373}
1374
1375/* Convert mask from bit length form to in6_addr form.
1376 * This function never fails.
1377 */
1378void len2mask6(int len, struct in6_addr *addr)
1379{
1380 len2mask4(len, (struct in_addr *)&addr->s6_addr[0]); /* msb */
1381 len -= 32;
1382 len2mask4(len, (struct in_addr *)&addr->s6_addr[4]);
1383 len -= 32;
1384 len2mask4(len, (struct in_addr *)&addr->s6_addr[8]);
1385 len -= 32;
1386 len2mask4(len, (struct in_addr *)&addr->s6_addr[12]); /* lsb */
1387}
1388
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001389/*
Willy Tarreaud077a8e2007-05-08 18:28:09 +02001390 * converts <str> to two struct in_addr* which must be pre-allocated.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001391 * The format is "addr[/mask]", where "addr" cannot be empty, and mask
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001392 * is optional and either in the dotted or CIDR notation.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001393 * Note: "addr" can also be a hostname. Returns 1 if OK, 0 if error.
1394 */
Thierry FOURNIERfc7ac7b2014-02-11 15:23:04 +01001395int str2net(const char *str, int resolve, struct in_addr *addr, struct in_addr *mask)
Willy Tarreaubaaee002006-06-26 02:48:02 +02001396{
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001397 __label__ out_free, out_err;
1398 char *c, *s;
1399 int ret_val;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001400
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001401 s = strdup(str);
1402 if (!s)
1403 return 0;
1404
Willy Tarreaubaaee002006-06-26 02:48:02 +02001405 memset(mask, 0, sizeof(*mask));
1406 memset(addr, 0, sizeof(*addr));
Willy Tarreaubaaee002006-06-26 02:48:02 +02001407
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001408 if ((c = strrchr(s, '/')) != NULL) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02001409 *c++ = '\0';
1410 /* c points to the mask */
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001411 if (!str2mask(c, mask))
1412 goto out_err;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001413 }
1414 else {
Willy Tarreauebd61602006-12-30 11:54:15 +01001415 mask->s_addr = ~0U;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001416 }
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001417 if (!inet_pton(AF_INET, s, addr)) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02001418 struct hostent *he;
1419
Thierry FOURNIERfc7ac7b2014-02-11 15:23:04 +01001420 if (!resolve)
1421 goto out_err;
1422
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001423 if ((he = gethostbyname(s)) == NULL) {
1424 goto out_err;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001425 }
1426 else
1427 *addr = *(struct in_addr *) *(he->h_addr_list);
1428 }
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001429
1430 ret_val = 1;
1431 out_free:
1432 free(s);
1433 return ret_val;
1434 out_err:
1435 ret_val = 0;
1436 goto out_free;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001437}
1438
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001439
1440/*
Willy Tarreau6d20e282012-04-27 22:49:47 +02001441 * converts <str> to two struct in6_addr* which must be pre-allocated.
1442 * The format is "addr[/mask]", where "addr" cannot be empty, and mask
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001443 * is an optional number of bits (128 being the default).
Willy Tarreau6d20e282012-04-27 22:49:47 +02001444 * Returns 1 if OK, 0 if error.
1445 */
1446int str62net(const char *str, struct in6_addr *addr, unsigned char *mask)
1447{
1448 char *c, *s;
1449 int ret_val = 0;
1450 char *err;
1451 unsigned long len = 128;
1452
1453 s = strdup(str);
1454 if (!s)
1455 return 0;
1456
1457 memset(mask, 0, sizeof(*mask));
1458 memset(addr, 0, sizeof(*addr));
1459
1460 if ((c = strrchr(s, '/')) != NULL) {
1461 *c++ = '\0'; /* c points to the mask */
1462 if (!*c)
1463 goto out_free;
1464
1465 len = strtoul(c, &err, 10);
1466 if ((err && *err) || (unsigned)len > 128)
1467 goto out_free;
1468 }
1469 *mask = len; /* OK we have a valid mask in <len> */
1470
1471 if (!inet_pton(AF_INET6, s, addr))
1472 goto out_free;
1473
1474 ret_val = 1;
1475 out_free:
1476 free(s);
1477 return ret_val;
1478}
1479
1480
1481/*
Willy Tarreau12e10272021-03-25 11:34:40 +01001482 * Parse IPv4 address found in url. Return the number of bytes parsed. It
1483 * expects exactly 4 numbers between 0 and 255 delimited by dots, and returns
1484 * zero in case of mismatch.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001485 */
David du Colombier6f5ccb12011-03-10 22:26:24 +01001486int url2ipv4(const char *addr, struct in_addr *dst)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001487{
1488 int saw_digit, octets, ch;
1489 u_char tmp[4], *tp;
1490 const char *cp = addr;
1491
1492 saw_digit = 0;
1493 octets = 0;
1494 *(tp = tmp) = 0;
1495
1496 while (*addr) {
Willy Tarreau12e10272021-03-25 11:34:40 +01001497 unsigned char digit = (ch = *addr) - '0';
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001498 if (digit > 9 && ch != '.')
1499 break;
Willy Tarreau12e10272021-03-25 11:34:40 +01001500 addr++;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001501 if (digit <= 9) {
1502 u_int new = *tp * 10 + digit;
1503 if (new > 255)
1504 return 0;
1505 *tp = new;
1506 if (!saw_digit) {
1507 if (++octets > 4)
1508 return 0;
1509 saw_digit = 1;
1510 }
1511 } else if (ch == '.' && saw_digit) {
1512 if (octets == 4)
1513 return 0;
1514 *++tp = 0;
1515 saw_digit = 0;
1516 } else
1517 return 0;
1518 }
1519
1520 if (octets < 4)
1521 return 0;
1522
1523 memcpy(&dst->s_addr, tmp, 4);
Willy Tarreau12e10272021-03-25 11:34:40 +01001524 return addr - cp;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001525}
1526
1527/*
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001528 * Resolve destination server from URL. Convert <str> to a sockaddr_storage.
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001529 * <out> contain the code of the detected scheme, the start and length of
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001530 * the hostname. Actually only http and https are supported. <out> can be NULL.
1531 * This function returns the consumed length. It is useful if you parse complete
1532 * url like http://host:port/path, because the consumed length corresponds to
1533 * the first character of the path. If the conversion fails, it returns -1.
1534 *
1535 * This function tries to resolve the DNS name if haproxy is in starting mode.
1536 * So, this function may be used during the configuration parsing.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001537 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001538int url2sa(const char *url, int ulen, struct sockaddr_storage *addr, struct split_url *out)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001539{
1540 const char *curr = url, *cp = url;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001541 const char *end;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001542 int ret, url_code = 0;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001543 unsigned long long int http_code = 0;
1544 int default_port;
1545 struct hostent *he;
1546 char *p;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001547
1548 /* Firstly, try to find :// pattern */
1549 while (curr < url+ulen && url_code != 0x3a2f2f) {
1550 url_code = ((url_code & 0xffff) << 8);
1551 url_code += (unsigned char)*curr++;
1552 }
1553
1554 /* Secondly, if :// pattern is found, verify parsed stuff
1555 * before pattern is matching our http pattern.
1556 * If so parse ip address and port in uri.
1557 *
1558 * WARNING: Current code doesn't support dynamic async dns resolver.
1559 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001560 if (url_code != 0x3a2f2f)
1561 return -1;
1562
1563 /* Copy scheme, and utrn to lower case. */
1564 while (cp < curr - 3)
1565 http_code = (http_code << 8) + *cp++;
1566 http_code |= 0x2020202020202020ULL; /* Turn everything to lower case */
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001567
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001568 /* HTTP or HTTPS url matching */
1569 if (http_code == 0x2020202068747470ULL) {
1570 default_port = 80;
1571 if (out)
1572 out->scheme = SCH_HTTP;
1573 }
1574 else if (http_code == 0x2020206874747073ULL) {
1575 default_port = 443;
1576 if (out)
1577 out->scheme = SCH_HTTPS;
1578 }
1579 else
1580 return -1;
1581
1582 /* If the next char is '[', the host address is IPv6. */
1583 if (*curr == '[') {
1584 curr++;
1585
1586 /* Check trash size */
1587 if (trash.size < ulen)
1588 return -1;
1589
1590 /* Look for ']' and copy the address in a trash buffer. */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001591 p = trash.area;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001592 for (end = curr;
1593 end < url + ulen && *end != ']';
1594 end++, p++)
1595 *p = *end;
1596 if (*end != ']')
1597 return -1;
1598 *p = '\0';
1599
1600 /* Update out. */
1601 if (out) {
1602 out->host = curr;
1603 out->host_len = end - curr;
1604 }
1605
1606 /* Try IPv6 decoding. */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001607 if (!inet_pton(AF_INET6, trash.area, &((struct sockaddr_in6 *)addr)->sin6_addr))
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001608 return -1;
1609 end++;
1610
1611 /* Decode port. */
1612 if (*end == ':') {
1613 end++;
1614 default_port = read_uint(&end, url + ulen);
1615 }
1616 ((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1617 ((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1618 return end - url;
1619 }
1620 else {
1621 /* We are looking for IP address. If you want to parse and
1622 * resolve hostname found in url, you can use str2sa_range(), but
1623 * be warned this can slow down global daemon performances
1624 * while handling lagging dns responses.
1625 */
1626 ret = url2ipv4(curr, &((struct sockaddr_in *)addr)->sin_addr);
1627 if (ret) {
1628 /* Update out. */
1629 if (out) {
1630 out->host = curr;
1631 out->host_len = ret;
1632 }
1633
1634 curr += ret;
1635
1636 /* Decode port. */
1637 if (*curr == ':') {
1638 curr++;
1639 default_port = read_uint(&curr, url + ulen);
1640 }
1641 ((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1642
1643 /* Set family. */
1644 ((struct sockaddr_in *)addr)->sin_family = AF_INET;
1645 return curr - url;
1646 }
1647 else if (global.mode & MODE_STARTING) {
1648 /* The IPv4 and IPv6 decoding fails, maybe the url contain name. Try to execute
1649 * synchronous DNS request only if HAProxy is in the start state.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001650 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001651
1652 /* look for : or / or end */
1653 for (end = curr;
1654 end < url + ulen && *end != '/' && *end != ':';
1655 end++);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001656 memcpy(trash.area, curr, end - curr);
1657 trash.area[end - curr] = '\0';
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001658
1659 /* try to resolve an IPv4/IPv6 hostname */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001660 he = gethostbyname(trash.area);
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001661 if (!he)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001662 return -1;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001663
1664 /* Update out. */
1665 if (out) {
1666 out->host = curr;
1667 out->host_len = end - curr;
1668 }
1669
1670 /* Decode port. */
1671 if (*end == ':') {
1672 end++;
1673 default_port = read_uint(&end, url + ulen);
1674 }
1675
1676 /* Copy IP address, set port and family. */
1677 switch (he->h_addrtype) {
1678 case AF_INET:
1679 ((struct sockaddr_in *)addr)->sin_addr = *(struct in_addr *) *(he->h_addr_list);
1680 ((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1681 ((struct sockaddr_in *)addr)->sin_family = AF_INET;
1682 return end - url;
1683
1684 case AF_INET6:
1685 ((struct sockaddr_in6 *)addr)->sin6_addr = *(struct in6_addr *) *(he->h_addr_list);
1686 ((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1687 ((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1688 return end - url;
1689 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001690 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001691 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001692 return -1;
1693}
1694
Willy Tarreau631f01c2011-09-05 00:36:48 +02001695/* Tries to convert a sockaddr_storage address to text form. Upon success, the
1696 * address family is returned so that it's easy for the caller to adapt to the
1697 * output format. Zero is returned if the address family is not supported. -1
1698 * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1699 * supported.
1700 */
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001701int addr_to_str(const struct sockaddr_storage *addr, char *str, int size)
Willy Tarreau631f01c2011-09-05 00:36:48 +02001702{
1703
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001704 const void *ptr;
Willy Tarreau631f01c2011-09-05 00:36:48 +02001705
1706 if (size < 5)
1707 return 0;
1708 *str = '\0';
1709
1710 switch (addr->ss_family) {
1711 case AF_INET:
1712 ptr = &((struct sockaddr_in *)addr)->sin_addr;
1713 break;
1714 case AF_INET6:
1715 ptr = &((struct sockaddr_in6 *)addr)->sin6_addr;
1716 break;
1717 case AF_UNIX:
1718 memcpy(str, "unix", 5);
1719 return addr->ss_family;
1720 default:
1721 return 0;
1722 }
1723
1724 if (inet_ntop(addr->ss_family, ptr, str, size))
1725 return addr->ss_family;
1726
1727 /* failed */
1728 return -1;
1729}
1730
Simon Horman75ab8bd2014-06-16 09:39:41 +09001731/* Tries to convert a sockaddr_storage port to text form. Upon success, the
1732 * address family is returned so that it's easy for the caller to adapt to the
1733 * output format. Zero is returned if the address family is not supported. -1
1734 * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1735 * supported.
1736 */
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001737int port_to_str(const struct sockaddr_storage *addr, char *str, int size)
Simon Horman75ab8bd2014-06-16 09:39:41 +09001738{
1739
1740 uint16_t port;
1741
1742
Willy Tarreaud7dad1b2017-01-06 16:46:22 +01001743 if (size < 6)
Simon Horman75ab8bd2014-06-16 09:39:41 +09001744 return 0;
1745 *str = '\0';
1746
1747 switch (addr->ss_family) {
1748 case AF_INET:
1749 port = ((struct sockaddr_in *)addr)->sin_port;
1750 break;
1751 case AF_INET6:
1752 port = ((struct sockaddr_in6 *)addr)->sin6_port;
1753 break;
1754 case AF_UNIX:
1755 memcpy(str, "unix", 5);
1756 return addr->ss_family;
1757 default:
1758 return 0;
1759 }
1760
1761 snprintf(str, size, "%u", ntohs(port));
1762 return addr->ss_family;
1763}
1764
Willy Tarreau16e01562016-08-09 16:46:18 +02001765/* check if the given address is local to the system or not. It will return
1766 * -1 when it's not possible to know, 0 when the address is not local, 1 when
1767 * it is. We don't want to iterate over all interfaces for this (and it is not
1768 * portable). So instead we try to bind in UDP to this address on a free non
1769 * privileged port and to connect to the same address, port 0 (connect doesn't
1770 * care). If it succeeds, we own the address. Note that non-inet addresses are
1771 * considered local since they're most likely AF_UNIX.
1772 */
1773int addr_is_local(const struct netns_entry *ns,
1774 const struct sockaddr_storage *orig)
1775{
1776 struct sockaddr_storage addr;
1777 int result;
1778 int fd;
1779
1780 if (!is_inet_addr(orig))
1781 return 1;
1782
1783 memcpy(&addr, orig, sizeof(addr));
1784 set_host_port(&addr, 0);
1785
1786 fd = my_socketat(ns, addr.ss_family, SOCK_DGRAM, IPPROTO_UDP);
1787 if (fd < 0)
1788 return -1;
1789
1790 result = -1;
1791 if (bind(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == 0) {
1792 if (connect(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == -1)
1793 result = 0; // fail, non-local address
1794 else
1795 result = 1; // success, local address
1796 }
1797 else {
1798 if (errno == EADDRNOTAVAIL)
1799 result = 0; // definitely not local :-)
1800 }
1801 close(fd);
1802
1803 return result;
1804}
1805
Willy Tarreaubaaee002006-06-26 02:48:02 +02001806/* will try to encode the string <string> replacing all characters tagged in
1807 * <map> with the hexadecimal representation of their ASCII-code (2 digits)
1808 * prefixed by <escape>, and will store the result between <start> (included)
1809 * and <stop> (excluded), and will always terminate the string with a '\0'
1810 * before <stop>. The position of the '\0' is returned if the conversion
1811 * completes. If bytes are missing between <start> and <stop>, then the
1812 * conversion will be incomplete and truncated. If <stop> <= <start>, the '\0'
1813 * cannot even be stored so we return <start> without writing the 0.
1814 * The input string must also be zero-terminated.
1815 */
1816const char hextab[16] = "0123456789ABCDEF";
1817char *encode_string(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001818 const char escape, const long *map,
Willy Tarreaubaaee002006-06-26 02:48:02 +02001819 const char *string)
1820{
1821 if (start < stop) {
1822 stop--; /* reserve one byte for the final '\0' */
1823 while (start < stop && *string != '\0') {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001824 if (!ha_bit_test((unsigned char)(*string), map))
Willy Tarreaubaaee002006-06-26 02:48:02 +02001825 *start++ = *string;
1826 else {
1827 if (start + 3 >= stop)
1828 break;
1829 *start++ = escape;
1830 *start++ = hextab[(*string >> 4) & 15];
1831 *start++ = hextab[*string & 15];
1832 }
1833 string++;
1834 }
1835 *start = '\0';
1836 }
1837 return start;
1838}
1839
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001840/*
1841 * Same behavior as encode_string() above, except that it encodes chunk
1842 * <chunk> instead of a string.
1843 */
1844char *encode_chunk(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001845 const char escape, const long *map,
Willy Tarreau83061a82018-07-13 11:56:34 +02001846 const struct buffer *chunk)
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001847{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001848 char *str = chunk->area;
1849 char *end = chunk->area + chunk->data;
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001850
1851 if (start < stop) {
1852 stop--; /* reserve one byte for the final '\0' */
1853 while (start < stop && str < end) {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001854 if (!ha_bit_test((unsigned char)(*str), map))
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001855 *start++ = *str;
1856 else {
1857 if (start + 3 >= stop)
1858 break;
1859 *start++ = escape;
1860 *start++ = hextab[(*str >> 4) & 15];
1861 *start++ = hextab[*str & 15];
1862 }
1863 str++;
1864 }
1865 *start = '\0';
1866 }
1867 return start;
1868}
1869
Dragan Dosen0edd1092016-02-12 13:23:02 +01001870/*
1871 * Tries to prefix characters tagged in the <map> with the <escape>
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001872 * character. The input <string> must be zero-terminated. The result will
1873 * be stored between <start> (included) and <stop> (excluded). This
1874 * function will always try to terminate the resulting string with a '\0'
1875 * before <stop>, and will return its position if the conversion
1876 * completes.
1877 */
1878char *escape_string(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001879 const char escape, const long *map,
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001880 const char *string)
1881{
1882 if (start < stop) {
1883 stop--; /* reserve one byte for the final '\0' */
1884 while (start < stop && *string != '\0') {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001885 if (!ha_bit_test((unsigned char)(*string), map))
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001886 *start++ = *string;
1887 else {
1888 if (start + 2 >= stop)
1889 break;
1890 *start++ = escape;
1891 *start++ = *string;
1892 }
1893 string++;
1894 }
1895 *start = '\0';
1896 }
1897 return start;
1898}
1899
1900/*
1901 * Tries to prefix characters tagged in the <map> with the <escape>
Dragan Dosen0edd1092016-02-12 13:23:02 +01001902 * character. <chunk> contains the input to be escaped. The result will be
1903 * stored between <start> (included) and <stop> (excluded). The function
1904 * will always try to terminate the resulting string with a '\0' before
1905 * <stop>, and will return its position if the conversion completes.
1906 */
1907char *escape_chunk(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001908 const char escape, const long *map,
Willy Tarreau83061a82018-07-13 11:56:34 +02001909 const struct buffer *chunk)
Dragan Dosen0edd1092016-02-12 13:23:02 +01001910{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001911 char *str = chunk->area;
1912 char *end = chunk->area + chunk->data;
Dragan Dosen0edd1092016-02-12 13:23:02 +01001913
1914 if (start < stop) {
1915 stop--; /* reserve one byte for the final '\0' */
1916 while (start < stop && str < end) {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001917 if (!ha_bit_test((unsigned char)(*str), map))
Dragan Dosen0edd1092016-02-12 13:23:02 +01001918 *start++ = *str;
1919 else {
1920 if (start + 2 >= stop)
1921 break;
1922 *start++ = escape;
1923 *start++ = *str;
1924 }
1925 str++;
1926 }
1927 *start = '\0';
1928 }
1929 return start;
1930}
1931
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001932/* Check a string for using it in a CSV output format. If the string contains
1933 * one of the following four char <">, <,>, CR or LF, the string is
1934 * encapsulated between <"> and the <"> are escaped by a <""> sequence.
1935 * <str> is the input string to be escaped. The function assumes that
1936 * the input string is null-terminated.
1937 *
1938 * If <quote> is 0, the result is returned escaped but without double quote.
Willy Tarreau898529b2016-01-06 18:07:04 +01001939 * It is useful if the escaped string is used between double quotes in the
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001940 * format.
1941 *
Willy Tarreau898529b2016-01-06 18:07:04 +01001942 * printf("..., \"%s\", ...\r\n", csv_enc(str, 0, &trash));
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001943 *
Willy Tarreaub631c292016-01-08 10:04:08 +01001944 * If <quote> is 1, the converter puts the quotes only if any reserved character
1945 * is present. If <quote> is 2, the converter always puts the quotes.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001946 *
Willy Tarreau83061a82018-07-13 11:56:34 +02001947 * <output> is a struct buffer used for storing the output string.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001948 *
Willy Tarreau898529b2016-01-06 18:07:04 +01001949 * The function returns the converted string on its output. If an error
1950 * occurs, the function returns an empty string. This type of output is useful
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001951 * for using the function directly as printf() argument.
1952 *
1953 * If the output buffer is too short to contain the input string, the result
1954 * is truncated.
Willy Tarreau898529b2016-01-06 18:07:04 +01001955 *
Willy Tarreaub631c292016-01-08 10:04:08 +01001956 * This function appends the encoding to the existing output chunk, and it
1957 * guarantees that it starts immediately at the first available character of
1958 * the chunk. Please use csv_enc() instead if you want to replace the output
1959 * chunk.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001960 */
Willy Tarreau83061a82018-07-13 11:56:34 +02001961const char *csv_enc_append(const char *str, int quote, struct buffer *output)
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001962{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001963 char *end = output->area + output->size;
1964 char *out = output->area + output->data;
Willy Tarreau898529b2016-01-06 18:07:04 +01001965 char *ptr = out;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001966
Willy Tarreaub631c292016-01-08 10:04:08 +01001967 if (quote == 1) {
1968 /* automatic quoting: first verify if we'll have to quote the string */
1969 if (!strpbrk(str, "\n\r,\""))
1970 quote = 0;
1971 }
1972
1973 if (quote)
1974 *ptr++ = '"';
1975
Willy Tarreau898529b2016-01-06 18:07:04 +01001976 while (*str && ptr < end - 2) { /* -2 for reserving space for <"> and \0. */
1977 *ptr = *str;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001978 if (*str == '"') {
Willy Tarreau898529b2016-01-06 18:07:04 +01001979 ptr++;
1980 if (ptr >= end - 2) {
1981 ptr--;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001982 break;
1983 }
Willy Tarreau898529b2016-01-06 18:07:04 +01001984 *ptr = '"';
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001985 }
Willy Tarreau898529b2016-01-06 18:07:04 +01001986 ptr++;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001987 str++;
1988 }
1989
Willy Tarreaub631c292016-01-08 10:04:08 +01001990 if (quote)
1991 *ptr++ = '"';
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001992
Willy Tarreau898529b2016-01-06 18:07:04 +01001993 *ptr = '\0';
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001994 output->data = ptr - output->area;
Willy Tarreau898529b2016-01-06 18:07:04 +01001995 return out;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001996}
1997
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02001998/* Decode an URL-encoded string in-place. The resulting string might
1999 * be shorter. If some forbidden characters are found, the conversion is
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002000 * aborted, the string is truncated before the issue and a negative value is
2001 * returned, otherwise the operation returns the length of the decoded string.
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002002 * If the 'in_form' argument is non-nul the string is assumed to be part of
2003 * an "application/x-www-form-urlencoded" encoded string, and the '+' will be
2004 * turned to a space. If it's zero, this will only be done after a question
2005 * mark ('?').
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002006 */
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002007int url_decode(char *string, int in_form)
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002008{
2009 char *in, *out;
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002010 int ret = -1;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002011
2012 in = string;
2013 out = string;
2014 while (*in) {
2015 switch (*in) {
2016 case '+' :
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002017 *out++ = in_form ? ' ' : *in;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002018 break;
2019 case '%' :
2020 if (!ishex(in[1]) || !ishex(in[2]))
2021 goto end;
2022 *out++ = (hex2i(in[1]) << 4) + hex2i(in[2]);
2023 in += 2;
2024 break;
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002025 case '?':
2026 in_form = 1;
2027 /* fall through */
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002028 default:
2029 *out++ = *in;
2030 break;
2031 }
2032 in++;
2033 }
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002034 ret = out - string; /* success */
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002035 end:
2036 *out = 0;
2037 return ret;
2038}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002039
Willy Tarreau6911fa42007-03-04 18:06:08 +01002040unsigned int str2ui(const char *s)
2041{
2042 return __str2ui(s);
2043}
2044
2045unsigned int str2uic(const char *s)
2046{
2047 return __str2uic(s);
2048}
2049
2050unsigned int strl2ui(const char *s, int len)
2051{
2052 return __strl2ui(s, len);
2053}
2054
2055unsigned int strl2uic(const char *s, int len)
2056{
2057 return __strl2uic(s, len);
2058}
2059
Willy Tarreau4ec83cd2010-10-15 23:19:55 +02002060unsigned int read_uint(const char **s, const char *end)
2061{
2062 return __read_uint(s, end);
2063}
2064
Thierry FOURNIER763a5d82015-07-06 23:09:52 +02002065/* This function reads an unsigned integer from the string pointed to by <s> and
2066 * returns it. The <s> pointer is adjusted to point to the first unread char. The
2067 * function automatically stops at <end>. If the number overflows, the 2^64-1
2068 * value is returned.
2069 */
2070unsigned long long int read_uint64(const char **s, const char *end)
2071{
2072 const char *ptr = *s;
2073 unsigned long long int i = 0, tmp;
2074 unsigned int j;
2075
2076 while (ptr < end) {
2077
2078 /* read next char */
2079 j = *ptr - '0';
2080 if (j > 9)
2081 goto read_uint64_end;
2082
2083 /* add char to the number and check overflow. */
2084 tmp = i * 10;
2085 if (tmp / 10 != i) {
2086 i = ULLONG_MAX;
2087 goto read_uint64_eat;
2088 }
2089 if (ULLONG_MAX - tmp < j) {
2090 i = ULLONG_MAX;
2091 goto read_uint64_eat;
2092 }
2093 i = tmp + j;
2094 ptr++;
2095 }
2096read_uint64_eat:
2097 /* eat each numeric char */
2098 while (ptr < end) {
2099 if ((unsigned int)(*ptr - '0') > 9)
2100 break;
2101 ptr++;
2102 }
2103read_uint64_end:
2104 *s = ptr;
2105 return i;
2106}
2107
2108/* This function reads an integer from the string pointed to by <s> and returns
2109 * it. The <s> pointer is adjusted to point to the first unread char. The function
2110 * automatically stops at <end>. Il the number is bigger than 2^63-2, the 2^63-1
2111 * value is returned. If the number is lowest than -2^63-1, the -2^63 value is
2112 * returned.
2113 */
2114long long int read_int64(const char **s, const char *end)
2115{
2116 unsigned long long int i = 0;
2117 int neg = 0;
2118
2119 /* Look for minus char. */
2120 if (**s == '-') {
2121 neg = 1;
2122 (*s)++;
2123 }
2124 else if (**s == '+')
2125 (*s)++;
2126
2127 /* convert as positive number. */
2128 i = read_uint64(s, end);
2129
2130 if (neg) {
2131 if (i > 0x8000000000000000ULL)
2132 return LLONG_MIN;
2133 return -i;
2134 }
2135 if (i > 0x7fffffffffffffffULL)
2136 return LLONG_MAX;
2137 return i;
2138}
2139
Willy Tarreau6911fa42007-03-04 18:06:08 +01002140/* This one is 7 times faster than strtol() on athlon with checks.
2141 * It returns the value of the number composed of all valid digits read,
2142 * and can process negative numbers too.
2143 */
2144int strl2ic(const char *s, int len)
2145{
2146 int i = 0;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002147 int j, k;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002148
2149 if (len > 0) {
2150 if (*s != '-') {
2151 /* positive number */
2152 while (len-- > 0) {
2153 j = (*s++) - '0';
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002154 k = i * 10;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002155 if (j > 9)
2156 break;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002157 i = k + j;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002158 }
2159 } else {
2160 /* negative number */
2161 s++;
2162 while (--len > 0) {
2163 j = (*s++) - '0';
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002164 k = i * 10;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002165 if (j > 9)
2166 break;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002167 i = k - j;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002168 }
2169 }
2170 }
2171 return i;
2172}
2173
2174
2175/* This function reads exactly <len> chars from <s> and converts them to a
2176 * signed integer which it stores into <ret>. It accurately detects any error
2177 * (truncated string, invalid chars, overflows). It is meant to be used in
2178 * applications designed for hostile environments. It returns zero when the
2179 * number has successfully been converted, non-zero otherwise. When an error
2180 * is returned, the <ret> value is left untouched. It is yet 5 to 40 times
2181 * faster than strtol().
2182 */
2183int strl2irc(const char *s, int len, int *ret)
2184{
2185 int i = 0;
2186 int j;
2187
2188 if (!len)
2189 return 1;
2190
2191 if (*s != '-') {
2192 /* positive number */
2193 while (len-- > 0) {
2194 j = (*s++) - '0';
2195 if (j > 9) return 1; /* invalid char */
2196 if (i > INT_MAX / 10) return 1; /* check for multiply overflow */
2197 i = i * 10;
2198 if (i + j < i) return 1; /* check for addition overflow */
2199 i = i + j;
2200 }
2201 } else {
2202 /* negative number */
2203 s++;
2204 while (--len > 0) {
2205 j = (*s++) - '0';
2206 if (j > 9) return 1; /* invalid char */
2207 if (i < INT_MIN / 10) return 1; /* check for multiply overflow */
2208 i = i * 10;
2209 if (i - j > i) return 1; /* check for subtract overflow */
2210 i = i - j;
2211 }
2212 }
2213 *ret = i;
2214 return 0;
2215}
2216
2217
2218/* This function reads exactly <len> chars from <s> and converts them to a
2219 * signed integer which it stores into <ret>. It accurately detects any error
2220 * (truncated string, invalid chars, overflows). It is meant to be used in
2221 * applications designed for hostile environments. It returns zero when the
2222 * number has successfully been converted, non-zero otherwise. When an error
2223 * is returned, the <ret> value is left untouched. It is about 3 times slower
William Dauchy060ffc82021-02-06 20:47:51 +01002224 * than strl2irc().
Willy Tarreau6911fa42007-03-04 18:06:08 +01002225 */
Willy Tarreau6911fa42007-03-04 18:06:08 +01002226
2227int strl2llrc(const char *s, int len, long long *ret)
2228{
2229 long long i = 0;
2230 int j;
2231
2232 if (!len)
2233 return 1;
2234
2235 if (*s != '-') {
2236 /* positive number */
2237 while (len-- > 0) {
2238 j = (*s++) - '0';
2239 if (j > 9) return 1; /* invalid char */
2240 if (i > LLONG_MAX / 10LL) return 1; /* check for multiply overflow */
2241 i = i * 10LL;
2242 if (i + j < i) return 1; /* check for addition overflow */
2243 i = i + j;
2244 }
2245 } else {
2246 /* negative number */
2247 s++;
2248 while (--len > 0) {
2249 j = (*s++) - '0';
2250 if (j > 9) return 1; /* invalid char */
2251 if (i < LLONG_MIN / 10LL) return 1; /* check for multiply overflow */
2252 i = i * 10LL;
2253 if (i - j > i) return 1; /* check for subtract overflow */
2254 i = i - j;
2255 }
2256 }
2257 *ret = i;
2258 return 0;
2259}
2260
Thierry FOURNIER511e9472014-01-23 17:40:34 +01002261/* This function is used with pat_parse_dotted_ver(). It converts a string
2262 * composed by two number separated by a dot. Each part must contain in 16 bits
2263 * because internally they will be represented as a 32-bit quantity stored in
2264 * a 64-bit integer. It returns zero when the number has successfully been
2265 * converted, non-zero otherwise. When an error is returned, the <ret> value
2266 * is left untouched.
2267 *
2268 * "1.3" -> 0x0000000000010003
2269 * "65535.65535" -> 0x00000000ffffffff
2270 */
2271int strl2llrc_dotted(const char *text, int len, long long *ret)
2272{
2273 const char *end = &text[len];
2274 const char *p;
2275 long long major, minor;
2276
2277 /* Look for dot. */
2278 for (p = text; p < end; p++)
2279 if (*p == '.')
2280 break;
2281
2282 /* Convert major. */
2283 if (strl2llrc(text, p - text, &major) != 0)
2284 return 1;
2285
2286 /* Check major. */
2287 if (major >= 65536)
2288 return 1;
2289
2290 /* Convert minor. */
2291 minor = 0;
2292 if (p < end)
2293 if (strl2llrc(p + 1, end - (p + 1), &minor) != 0)
2294 return 1;
2295
2296 /* Check minor. */
2297 if (minor >= 65536)
2298 return 1;
2299
2300 /* Compose value. */
2301 *ret = (major << 16) | (minor & 0xffff);
2302 return 0;
2303}
2304
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002305/* This function parses a time value optionally followed by a unit suffix among
2306 * "d", "h", "m", "s", "ms" or "us". It converts the value into the unit
2307 * expected by the caller. The computation does its best to avoid overflows.
2308 * The value is returned in <ret> if everything is fine, and a NULL is returned
2309 * by the function. In case of error, a pointer to the error is returned and
2310 * <ret> is left untouched. Values are automatically rounded up when needed.
Willy Tarreau9faebe32019-06-07 19:00:37 +02002311 * Values resulting in values larger than or equal to 2^31 after conversion are
2312 * reported as an overflow as value PARSE_TIME_OVER. Non-null values resulting
2313 * in an underflow are reported as an underflow as value PARSE_TIME_UNDER.
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002314 */
2315const char *parse_time_err(const char *text, unsigned *ret, unsigned unit_flags)
2316{
Willy Tarreau9faebe32019-06-07 19:00:37 +02002317 unsigned long long imult, idiv;
2318 unsigned long long omult, odiv;
2319 unsigned long long value, result;
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002320 const char *str = text;
2321
2322 if (!isdigit((unsigned char)*text))
2323 return text;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002324
2325 omult = odiv = 1;
2326
2327 switch (unit_flags & TIME_UNIT_MASK) {
2328 case TIME_UNIT_US: omult = 1000000; break;
2329 case TIME_UNIT_MS: omult = 1000; break;
2330 case TIME_UNIT_S: break;
2331 case TIME_UNIT_MIN: odiv = 60; break;
2332 case TIME_UNIT_HOUR: odiv = 3600; break;
2333 case TIME_UNIT_DAY: odiv = 86400; break;
2334 default: break;
2335 }
2336
2337 value = 0;
2338
2339 while (1) {
2340 unsigned int j;
2341
2342 j = *text - '0';
2343 if (j > 9)
2344 break;
2345 text++;
2346 value *= 10;
2347 value += j;
2348 }
2349
2350 imult = idiv = 1;
2351 switch (*text) {
2352 case '\0': /* no unit = default unit */
2353 imult = omult = idiv = odiv = 1;
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002354 goto end;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002355 case 's': /* second = unscaled unit */
2356 break;
2357 case 'u': /* microsecond : "us" */
2358 if (text[1] == 's') {
2359 idiv = 1000000;
2360 text++;
Thayne McCombsa6838052021-04-02 14:12:43 -06002361 break;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002362 }
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002363 return text;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002364 case 'm': /* millisecond : "ms" or minute: "m" */
2365 if (text[1] == 's') {
2366 idiv = 1000;
2367 text++;
2368 } else
2369 imult = 60;
2370 break;
2371 case 'h': /* hour : "h" */
2372 imult = 3600;
2373 break;
2374 case 'd': /* day : "d" */
2375 imult = 86400;
2376 break;
2377 default:
2378 return text;
2379 break;
2380 }
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002381 if (*(++text) != '\0') {
2382 ha_warning("unexpected character '%c' after the timer value '%s', only "
2383 "(us=microseconds,ms=milliseconds,s=seconds,m=minutes,h=hours,d=days) are supported."
2384 " This will be reported as an error in next versions.\n", *text, str);
2385 }
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002386
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002387 end:
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002388 if (omult % idiv == 0) { omult /= idiv; idiv = 1; }
2389 if (idiv % omult == 0) { idiv /= omult; omult = 1; }
2390 if (imult % odiv == 0) { imult /= odiv; odiv = 1; }
2391 if (odiv % imult == 0) { odiv /= imult; imult = 1; }
2392
Willy Tarreau9faebe32019-06-07 19:00:37 +02002393 result = (value * (imult * omult) + (idiv * odiv - 1)) / (idiv * odiv);
2394 if (result >= 0x80000000)
2395 return PARSE_TIME_OVER;
2396 if (!result && value)
2397 return PARSE_TIME_UNDER;
2398 *ret = result;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002399 return NULL;
2400}
Willy Tarreau6911fa42007-03-04 18:06:08 +01002401
Emeric Brun39132b22010-01-04 14:57:24 +01002402/* this function converts the string starting at <text> to an unsigned int
2403 * stored in <ret>. If an error is detected, the pointer to the unexpected
Joseph Herlant32b83272018-11-15 11:58:28 -08002404 * character is returned. If the conversion is successful, NULL is returned.
Emeric Brun39132b22010-01-04 14:57:24 +01002405 */
2406const char *parse_size_err(const char *text, unsigned *ret) {
2407 unsigned value = 0;
2408
Christopher Faulet82635a02020-12-11 09:30:45 +01002409 if (!isdigit((unsigned char)*text))
2410 return text;
2411
Emeric Brun39132b22010-01-04 14:57:24 +01002412 while (1) {
2413 unsigned int j;
2414
2415 j = *text - '0';
2416 if (j > 9)
2417 break;
2418 if (value > ~0U / 10)
2419 return text;
2420 value *= 10;
2421 if (value > (value + j))
2422 return text;
2423 value += j;
2424 text++;
2425 }
2426
2427 switch (*text) {
2428 case '\0':
2429 break;
2430 case 'K':
2431 case 'k':
2432 if (value > ~0U >> 10)
2433 return text;
2434 value = value << 10;
2435 break;
2436 case 'M':
2437 case 'm':
2438 if (value > ~0U >> 20)
2439 return text;
2440 value = value << 20;
2441 break;
2442 case 'G':
2443 case 'g':
2444 if (value > ~0U >> 30)
2445 return text;
2446 value = value << 30;
2447 break;
2448 default:
2449 return text;
2450 }
2451
Godbach58048a22015-01-28 17:36:16 +08002452 if (*text != '\0' && *++text != '\0')
2453 return text;
2454
Emeric Brun39132b22010-01-04 14:57:24 +01002455 *ret = value;
2456 return NULL;
2457}
2458
Willy Tarreau126d4062013-12-03 17:50:47 +01002459/*
2460 * Parse binary string written in hexadecimal (source) and store the decoded
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002461 * result into binstr and set binstrlen to the length of binstr. Memory for
Willy Tarreau126d4062013-12-03 17:50:47 +01002462 * binstr is allocated by the function. In case of error, returns 0 with an
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002463 * error message in err. In success case, it returns the consumed length.
Willy Tarreau126d4062013-12-03 17:50:47 +01002464 */
2465int parse_binary(const char *source, char **binstr, int *binstrlen, char **err)
2466{
2467 int len;
2468 const char *p = source;
2469 int i,j;
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002470 int alloc;
Willy Tarreau126d4062013-12-03 17:50:47 +01002471
2472 len = strlen(source);
2473 if (len % 2) {
2474 memprintf(err, "an even number of hex digit is expected");
2475 return 0;
2476 }
2477
2478 len = len >> 1;
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002479
Willy Tarreau126d4062013-12-03 17:50:47 +01002480 if (!*binstr) {
Tim Duesterhuse52b6e52020-09-12 20:26:43 +02002481 *binstr = calloc(len, sizeof(**binstr));
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002482 if (!*binstr) {
2483 memprintf(err, "out of memory while loading string pattern");
2484 return 0;
2485 }
2486 alloc = 1;
Willy Tarreau126d4062013-12-03 17:50:47 +01002487 }
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002488 else {
2489 if (*binstrlen < len) {
Joseph Herlant76dbe782018-11-15 12:01:22 -08002490 memprintf(err, "no space available in the buffer. expect %d, provides %d",
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002491 len, *binstrlen);
2492 return 0;
2493 }
2494 alloc = 0;
2495 }
2496 *binstrlen = len;
Willy Tarreau126d4062013-12-03 17:50:47 +01002497
2498 i = j = 0;
2499 while (j < len) {
2500 if (!ishex(p[i++]))
2501 goto bad_input;
2502 if (!ishex(p[i++]))
2503 goto bad_input;
2504 (*binstr)[j++] = (hex2i(p[i-2]) << 4) + hex2i(p[i-1]);
2505 }
Thierry FOURNIERee330af2014-01-21 11:36:14 +01002506 return len << 1;
Willy Tarreau126d4062013-12-03 17:50:47 +01002507
2508bad_input:
2509 memprintf(err, "an hex digit is expected (found '%c')", p[i-1]);
Willy Tarreau61cfdf42021-02-20 10:46:51 +01002510 if (alloc)
2511 ha_free(binstr);
Willy Tarreau126d4062013-12-03 17:50:47 +01002512 return 0;
2513}
2514
Willy Tarreau946ba592009-05-10 15:41:18 +02002515/* copies at most <n> characters from <src> and always terminates with '\0' */
2516char *my_strndup(const char *src, int n)
2517{
2518 int len = 0;
2519 char *ret;
2520
2521 while (len < n && src[len])
2522 len++;
2523
Vincent Bernat3c2f2f22016-04-03 13:48:42 +02002524 ret = malloc(len + 1);
Willy Tarreau946ba592009-05-10 15:41:18 +02002525 if (!ret)
2526 return ret;
2527 memcpy(ret, src, len);
2528 ret[len] = '\0';
2529 return ret;
2530}
2531
Baptiste Assmannbb77c8e2013-10-06 23:24:13 +02002532/*
2533 * search needle in haystack
2534 * returns the pointer if found, returns NULL otherwise
2535 */
2536const void *my_memmem(const void *haystack, size_t haystacklen, const void *needle, size_t needlelen)
2537{
2538 const void *c = NULL;
2539 unsigned char f;
2540
2541 if ((haystack == NULL) || (needle == NULL) || (haystacklen < needlelen))
2542 return NULL;
2543
2544 f = *(char *)needle;
2545 c = haystack;
2546 while ((c = memchr(c, f, haystacklen - (c - haystack))) != NULL) {
2547 if ((haystacklen - (c - haystack)) < needlelen)
2548 return NULL;
2549
2550 if (memcmp(c, needle, needlelen) == 0)
2551 return c;
2552 ++c;
2553 }
2554 return NULL;
2555}
2556
Ilya Shipitsinc02a23f2020-05-06 00:53:22 +05002557/* get length of the initial segment consisting entirely of bytes in <accept> */
Christopher Faulet5eb96cb2020-04-15 10:23:01 +02002558size_t my_memspn(const void *str, size_t len, const void *accept, size_t acceptlen)
2559{
2560 size_t ret = 0;
2561
2562 while (ret < len && memchr(accept, *((int *)str), acceptlen)) {
2563 str++;
2564 ret++;
2565 }
2566 return ret;
2567}
2568
Ilya Shipitsinc02a23f2020-05-06 00:53:22 +05002569/* get length of the initial segment consisting entirely of bytes not in <rejcet> */
Christopher Faulet5eb96cb2020-04-15 10:23:01 +02002570size_t my_memcspn(const void *str, size_t len, const void *reject, size_t rejectlen)
2571{
2572 size_t ret = 0;
2573
2574 while (ret < len) {
2575 if(memchr(reject, *((int *)str), rejectlen))
2576 return ret;
2577 str++;
2578 ret++;
2579 }
2580 return ret;
2581}
2582
Willy Tarreau482b00d2009-10-04 22:48:42 +02002583/* This function returns the first unused key greater than or equal to <key> in
2584 * ID tree <root>. Zero is returned if no place is found.
2585 */
2586unsigned int get_next_id(struct eb_root *root, unsigned int key)
2587{
2588 struct eb32_node *used;
2589
2590 do {
2591 used = eb32_lookup_ge(root, key);
2592 if (!used || used->key > key)
2593 return key; /* key is available */
2594 key++;
2595 } while (key);
2596 return key;
2597}
2598
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002599/* dump the full tree to <file> in DOT format for debugging purposes. Will
2600 * optionally highlight node <subj> if found, depending on operation <op> :
2601 * 0 : nothing
2602 * >0 : insertion, node/leaf are surrounded in red
2603 * <0 : removal, node/leaf are dashed with no background
2604 * Will optionally add "desc" as a label on the graph if set and non-null.
2605 */
2606void 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 +01002607{
2608 struct eb32sc_node *node;
2609 unsigned long scope = -1;
2610
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002611 fprintf(file, "digraph ebtree {\n");
2612
2613 if (desc && *desc) {
2614 fprintf(file,
2615 " fontname=\"fixed\";\n"
2616 " fontsize=8;\n"
2617 " label=\"%s\";\n", desc);
2618 }
2619
Willy Tarreaued3cda02017-11-15 15:04:05 +01002620 fprintf(file,
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002621 " node [fontname=\"fixed\" fontsize=8 shape=\"box\" style=\"filled\" color=\"black\" fillcolor=\"white\"];\n"
2622 " edge [fontname=\"fixed\" fontsize=8 style=\"solid\" color=\"magenta\" dir=\"forward\"];\n"
Willy Tarreaued3cda02017-11-15 15:04:05 +01002623 " \"%lx_n\" [label=\"root\\n%lx\"]\n", (long)eb_root_to_node(root), (long)root
2624 );
2625
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002626 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002627 (long)eb_root_to_node(root),
2628 (long)eb_root_to_node(eb_clrtag(root->b[0])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002629 eb_gettag(root->b[0]) == EB_LEAF ? 'l' : 'n');
2630
2631 node = eb32sc_first(root, scope);
2632 while (node) {
2633 if (node->node.node_p) {
2634 /* node part is used */
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002635 fprintf(file, " \"%lx_n\" [label=\"%lx\\nkey=%u\\nscope=%lx\\nbit=%d\" fillcolor=\"lightskyblue1\" %s];\n",
2636 (long)node, (long)node, node->key, node->node_s, node->node.bit,
2637 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
Willy Tarreaued3cda02017-11-15 15:04:05 +01002638
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002639 fprintf(file, " \"%lx_n\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002640 (long)node,
2641 (long)eb_root_to_node(eb_clrtag(node->node.node_p)),
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002642 eb_gettag(node->node.node_p) ? 'R' : 'L');
Willy Tarreaued3cda02017-11-15 15:04:05 +01002643
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002644 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002645 (long)node,
2646 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[0])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002647 eb_gettag(node->node.branches.b[0]) == EB_LEAF ? 'l' : 'n');
2648
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002649 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"R\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002650 (long)node,
2651 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[1])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002652 eb_gettag(node->node.branches.b[1]) == EB_LEAF ? 'l' : 'n');
2653 }
2654
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002655 fprintf(file, " \"%lx_l\" [label=\"%lx\\nkey=%u\\nscope=%lx\\npfx=%u\" fillcolor=\"yellow\" %s];\n",
2656 (long)node, (long)node, node->key, node->leaf_s, node->node.pfx,
2657 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
Willy Tarreaued3cda02017-11-15 15:04:05 +01002658
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002659 fprintf(file, " \"%lx_l\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002660 (long)node,
2661 (long)eb_root_to_node(eb_clrtag(node->node.leaf_p)),
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002662 eb_gettag(node->node.leaf_p) ? 'R' : 'L');
Willy Tarreaued3cda02017-11-15 15:04:05 +01002663 node = eb32sc_next(node, scope);
2664 }
2665 fprintf(file, "}\n");
2666}
2667
Willy Tarreau348238b2010-01-18 15:05:57 +01002668/* This function compares a sample word possibly followed by blanks to another
2669 * clean word. The compare is case-insensitive. 1 is returned if both are equal,
2670 * otherwise zero. This intends to be used when checking HTTP headers for some
2671 * values. Note that it validates a word followed only by blanks but does not
2672 * validate a word followed by blanks then other chars.
2673 */
2674int word_match(const char *sample, int slen, const char *word, int wlen)
2675{
2676 if (slen < wlen)
2677 return 0;
2678
2679 while (wlen) {
2680 char c = *sample ^ *word;
2681 if (c && c != ('A' ^ 'a'))
2682 return 0;
2683 sample++;
2684 word++;
2685 slen--;
2686 wlen--;
2687 }
2688
2689 while (slen) {
2690 if (*sample != ' ' && *sample != '\t')
2691 return 0;
2692 sample++;
2693 slen--;
2694 }
2695 return 1;
2696}
Willy Tarreau482b00d2009-10-04 22:48:42 +02002697
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002698/* Converts any text-formatted IPv4 address to a host-order IPv4 address. It
2699 * is particularly fast because it avoids expensive operations such as
2700 * multiplies, which are optimized away at the end. It requires a properly
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002701 * formatted address though (3 points).
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002702 */
2703unsigned int inetaddr_host(const char *text)
2704{
2705 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2706 register unsigned int dig100, dig10, dig1;
2707 int s;
2708 const char *p, *d;
2709
2710 dig1 = dig10 = dig100 = ascii_zero;
2711 s = 24;
2712
2713 p = text;
2714 while (1) {
2715 if (((unsigned)(*p - '0')) <= 9) {
2716 p++;
2717 continue;
2718 }
2719
2720 /* here, we have a complete byte between <text> and <p> (exclusive) */
2721 if (p == text)
2722 goto end;
2723
2724 d = p - 1;
2725 dig1 |= (unsigned int)(*d << s);
2726 if (d == text)
2727 goto end;
2728
2729 d--;
2730 dig10 |= (unsigned int)(*d << s);
2731 if (d == text)
2732 goto end;
2733
2734 d--;
2735 dig100 |= (unsigned int)(*d << s);
2736 end:
2737 if (!s || *p != '.')
2738 break;
2739
2740 s -= 8;
2741 text = ++p;
2742 }
2743
2744 dig100 -= ascii_zero;
2745 dig10 -= ascii_zero;
2746 dig1 -= ascii_zero;
2747 return ((dig100 * 10) + dig10) * 10 + dig1;
2748}
2749
2750/*
2751 * Idem except the first unparsed character has to be passed in <stop>.
2752 */
2753unsigned int inetaddr_host_lim(const char *text, const char *stop)
2754{
2755 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2756 register unsigned int dig100, dig10, dig1;
2757 int s;
2758 const char *p, *d;
2759
2760 dig1 = dig10 = dig100 = ascii_zero;
2761 s = 24;
2762
2763 p = text;
2764 while (1) {
2765 if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2766 p++;
2767 continue;
2768 }
2769
2770 /* here, we have a complete byte between <text> and <p> (exclusive) */
2771 if (p == text)
2772 goto end;
2773
2774 d = p - 1;
2775 dig1 |= (unsigned int)(*d << s);
2776 if (d == text)
2777 goto end;
2778
2779 d--;
2780 dig10 |= (unsigned int)(*d << s);
2781 if (d == text)
2782 goto end;
2783
2784 d--;
2785 dig100 |= (unsigned int)(*d << s);
2786 end:
2787 if (!s || p == stop || *p != '.')
2788 break;
2789
2790 s -= 8;
2791 text = ++p;
2792 }
2793
2794 dig100 -= ascii_zero;
2795 dig10 -= ascii_zero;
2796 dig1 -= ascii_zero;
2797 return ((dig100 * 10) + dig10) * 10 + dig1;
2798}
2799
2800/*
2801 * Idem except the pointer to first unparsed byte is returned into <ret> which
2802 * must not be NULL.
2803 */
Willy Tarreau74172752010-10-15 23:21:42 +02002804unsigned int inetaddr_host_lim_ret(char *text, char *stop, char **ret)
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002805{
2806 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2807 register unsigned int dig100, dig10, dig1;
2808 int s;
Willy Tarreau74172752010-10-15 23:21:42 +02002809 char *p, *d;
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002810
2811 dig1 = dig10 = dig100 = ascii_zero;
2812 s = 24;
2813
2814 p = text;
2815 while (1) {
2816 if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2817 p++;
2818 continue;
2819 }
2820
2821 /* here, we have a complete byte between <text> and <p> (exclusive) */
2822 if (p == text)
2823 goto end;
2824
2825 d = p - 1;
2826 dig1 |= (unsigned int)(*d << s);
2827 if (d == text)
2828 goto end;
2829
2830 d--;
2831 dig10 |= (unsigned int)(*d << s);
2832 if (d == text)
2833 goto end;
2834
2835 d--;
2836 dig100 |= (unsigned int)(*d << s);
2837 end:
2838 if (!s || p == stop || *p != '.')
2839 break;
2840
2841 s -= 8;
2842 text = ++p;
2843 }
2844
2845 *ret = p;
2846 dig100 -= ascii_zero;
2847 dig10 -= ascii_zero;
2848 dig1 -= ascii_zero;
2849 return ((dig100 * 10) + dig10) * 10 + dig1;
2850}
2851
Willy Tarreauf0b38bf2010-06-06 13:22:23 +02002852/* Convert a fixed-length string to an IP address. Returns 0 in case of error,
2853 * or the number of chars read in case of success. Maybe this could be replaced
2854 * by one of the functions above. Also, apparently this function does not support
2855 * hosts above 255 and requires exactly 4 octets.
Willy Tarreau075415a2013-12-12 11:29:39 +01002856 * The destination is only modified on success.
Willy Tarreauf0b38bf2010-06-06 13:22:23 +02002857 */
2858int buf2ip(const char *buf, size_t len, struct in_addr *dst)
2859{
2860 const char *addr;
2861 int saw_digit, octets, ch;
2862 u_char tmp[4], *tp;
2863 const char *cp = buf;
2864
2865 saw_digit = 0;
2866 octets = 0;
2867 *(tp = tmp) = 0;
2868
2869 for (addr = buf; addr - buf < len; addr++) {
2870 unsigned char digit = (ch = *addr) - '0';
2871
2872 if (digit > 9 && ch != '.')
2873 break;
2874
2875 if (digit <= 9) {
2876 u_int new = *tp * 10 + digit;
2877
2878 if (new > 255)
2879 return 0;
2880
2881 *tp = new;
2882
2883 if (!saw_digit) {
2884 if (++octets > 4)
2885 return 0;
2886 saw_digit = 1;
2887 }
2888 } else if (ch == '.' && saw_digit) {
2889 if (octets == 4)
2890 return 0;
2891
2892 *++tp = 0;
2893 saw_digit = 0;
2894 } else
2895 return 0;
2896 }
2897
2898 if (octets < 4)
2899 return 0;
2900
2901 memcpy(&dst->s_addr, tmp, 4);
2902 return addr - cp;
2903}
2904
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002905/* This function converts the string in <buf> of the len <len> to
2906 * struct in6_addr <dst> which must be allocated by the caller.
2907 * This function returns 1 in success case, otherwise zero.
Willy Tarreau075415a2013-12-12 11:29:39 +01002908 * The destination is only modified on success.
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002909 */
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002910int buf2ip6(const char *buf, size_t len, struct in6_addr *dst)
2911{
Thierry FOURNIERcd659912013-12-11 12:33:54 +01002912 char null_term_ip6[INET6_ADDRSTRLEN + 1];
Willy Tarreau075415a2013-12-12 11:29:39 +01002913 struct in6_addr out;
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002914
Thierry FOURNIERcd659912013-12-11 12:33:54 +01002915 if (len > INET6_ADDRSTRLEN)
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002916 return 0;
2917
2918 memcpy(null_term_ip6, buf, len);
2919 null_term_ip6[len] = '\0';
2920
Willy Tarreau075415a2013-12-12 11:29:39 +01002921 if (!inet_pton(AF_INET6, null_term_ip6, &out))
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002922 return 0;
2923
Willy Tarreau075415a2013-12-12 11:29:39 +01002924 *dst = out;
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002925 return 1;
2926}
2927
Willy Tarreauacf95772010-06-14 19:09:21 +02002928/* To be used to quote config arg positions. Returns the short string at <ptr>
2929 * surrounded by simple quotes if <ptr> is valid and non-empty, or "end of line"
2930 * if ptr is NULL or empty. The string is locally allocated.
2931 */
2932const char *quote_arg(const char *ptr)
2933{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01002934 static THREAD_LOCAL char val[32];
Willy Tarreauacf95772010-06-14 19:09:21 +02002935 int i;
2936
2937 if (!ptr || !*ptr)
2938 return "end of line";
2939 val[0] = '\'';
Willy Tarreaude2dd6b2013-01-24 02:14:42 +01002940 for (i = 1; i < sizeof(val) - 2 && *ptr; i++)
Willy Tarreauacf95772010-06-14 19:09:21 +02002941 val[i] = *ptr++;
2942 val[i++] = '\'';
2943 val[i] = '\0';
2944 return val;
2945}
2946
Willy Tarreau5b180202010-07-18 10:40:48 +02002947/* returns an operator among STD_OP_* for string <str> or < 0 if unknown */
2948int get_std_op(const char *str)
2949{
2950 int ret = -1;
2951
2952 if (*str == 'e' && str[1] == 'q')
2953 ret = STD_OP_EQ;
2954 else if (*str == 'n' && str[1] == 'e')
2955 ret = STD_OP_NE;
2956 else if (*str == 'l') {
2957 if (str[1] == 'e') ret = STD_OP_LE;
2958 else if (str[1] == 't') ret = STD_OP_LT;
2959 }
2960 else if (*str == 'g') {
2961 if (str[1] == 'e') ret = STD_OP_GE;
2962 else if (str[1] == 't') ret = STD_OP_GT;
2963 }
2964
2965 if (ret == -1 || str[2] != '\0')
2966 return -1;
2967 return ret;
2968}
2969
Willy Tarreau4c14eaa2010-11-24 14:01:45 +01002970/* hash a 32-bit integer to another 32-bit integer */
2971unsigned int full_hash(unsigned int a)
2972{
2973 return __full_hash(a);
2974}
2975
Willy Tarreauf3241112019-02-26 09:56:22 +01002976/* Return the bit position in mask <m> of the nth bit set of rank <r>, between
2977 * 0 and LONGBITS-1 included, starting from the left. For example ranks 0,1,2,3
2978 * for mask 0x55 will be 6, 4, 2 and 0 respectively. This algorithm is based on
2979 * a popcount variant and is described here :
2980 * https://graphics.stanford.edu/~seander/bithacks.html
2981 */
2982unsigned int mask_find_rank_bit(unsigned int r, unsigned long m)
2983{
2984 unsigned long a, b, c, d;
2985 unsigned int s;
2986 unsigned int t;
2987
2988 a = m - ((m >> 1) & ~0UL/3);
2989 b = (a & ~0UL/5) + ((a >> 2) & ~0UL/5);
2990 c = (b + (b >> 4)) & ~0UL/0x11;
2991 d = (c + (c >> 8)) & ~0UL/0x101;
2992
2993 r++; // make r be 1..64
2994
2995 t = 0;
2996 s = LONGBITS;
2997 if (s > 32) {
Willy Tarreau9b6be3b2019-03-18 16:31:18 +01002998 unsigned long d2 = (d >> 16) >> 16;
2999 t = d2 + (d2 >> 16);
Willy Tarreauf3241112019-02-26 09:56:22 +01003000 s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
3001 }
3002
3003 t = (d >> (s - 16)) & 0xff;
3004 s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
3005 t = (c >> (s - 8)) & 0xf;
3006 s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
3007 t = (b >> (s - 4)) & 0x7;
3008 s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
3009 t = (a >> (s - 2)) & 0x3;
3010 s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
3011 t = (m >> (s - 1)) & 0x1;
3012 s -= ((t - r) & 256) >> 8;
3013
3014 return s - 1;
3015}
3016
3017/* Same as mask_find_rank_bit() above but makes use of pre-computed bitmaps
3018 * based on <m>, in <a..d>. These ones must be updated whenever <m> changes
3019 * using mask_prep_rank_map() below.
3020 */
3021unsigned int mask_find_rank_bit_fast(unsigned int r, unsigned long m,
3022 unsigned long a, unsigned long b,
3023 unsigned long c, unsigned long d)
3024{
3025 unsigned int s;
3026 unsigned int t;
3027
3028 r++; // make r be 1..64
3029
3030 t = 0;
3031 s = LONGBITS;
3032 if (s > 32) {
Willy Tarreau9b6be3b2019-03-18 16:31:18 +01003033 unsigned long d2 = (d >> 16) >> 16;
3034 t = d2 + (d2 >> 16);
Willy Tarreauf3241112019-02-26 09:56:22 +01003035 s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
3036 }
3037
3038 t = (d >> (s - 16)) & 0xff;
3039 s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
3040 t = (c >> (s - 8)) & 0xf;
3041 s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
3042 t = (b >> (s - 4)) & 0x7;
3043 s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
3044 t = (a >> (s - 2)) & 0x3;
3045 s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
3046 t = (m >> (s - 1)) & 0x1;
3047 s -= ((t - r) & 256) >> 8;
3048
3049 return s - 1;
3050}
3051
3052/* Prepare the bitmaps used by the fast implementation of the find_rank_bit()
3053 * above.
3054 */
3055void mask_prep_rank_map(unsigned long m,
3056 unsigned long *a, unsigned long *b,
3057 unsigned long *c, unsigned long *d)
3058{
3059 *a = m - ((m >> 1) & ~0UL/3);
3060 *b = (*a & ~0UL/5) + ((*a >> 2) & ~0UL/5);
3061 *c = (*b + (*b >> 4)) & ~0UL/0x11;
3062 *d = (*c + (*c >> 8)) & ~0UL/0x101;
3063}
3064
David du Colombier4f92d322011-03-24 11:09:31 +01003065/* Return non-zero if IPv4 address is part of the network,
Willy Tarreaueec1d382016-07-13 11:59:39 +02003066 * otherwise zero. Note that <addr> may not necessarily be aligned
3067 * while the two other ones must.
David du Colombier4f92d322011-03-24 11:09:31 +01003068 */
Willy Tarreaueec1d382016-07-13 11:59:39 +02003069int in_net_ipv4(const void *addr, const struct in_addr *mask, const struct in_addr *net)
David du Colombier4f92d322011-03-24 11:09:31 +01003070{
Willy Tarreaueec1d382016-07-13 11:59:39 +02003071 struct in_addr addr_copy;
3072
3073 memcpy(&addr_copy, addr, sizeof(addr_copy));
3074 return((addr_copy.s_addr & mask->s_addr) == (net->s_addr & mask->s_addr));
David du Colombier4f92d322011-03-24 11:09:31 +01003075}
3076
3077/* Return non-zero if IPv6 address is part of the network,
Willy Tarreaueec1d382016-07-13 11:59:39 +02003078 * otherwise zero. Note that <addr> may not necessarily be aligned
3079 * while the two other ones must.
David du Colombier4f92d322011-03-24 11:09:31 +01003080 */
Willy Tarreaueec1d382016-07-13 11:59:39 +02003081int in_net_ipv6(const void *addr, const struct in6_addr *mask, const struct in6_addr *net)
David du Colombier4f92d322011-03-24 11:09:31 +01003082{
3083 int i;
Willy Tarreaueec1d382016-07-13 11:59:39 +02003084 struct in6_addr addr_copy;
David du Colombier4f92d322011-03-24 11:09:31 +01003085
Willy Tarreaueec1d382016-07-13 11:59:39 +02003086 memcpy(&addr_copy, addr, sizeof(addr_copy));
David du Colombier4f92d322011-03-24 11:09:31 +01003087 for (i = 0; i < sizeof(struct in6_addr) / sizeof(int); i++)
Willy Tarreaueec1d382016-07-13 11:59:39 +02003088 if (((((int *)&addr_copy)[i] & ((int *)mask)[i])) !=
David du Colombier4f92d322011-03-24 11:09:31 +01003089 (((int *)net)[i] & ((int *)mask)[i]))
3090 return 0;
3091 return 1;
3092}
3093
3094/* RFC 4291 prefix */
3095const char rfc4291_pfx[] = { 0x00, 0x00, 0x00, 0x00,
3096 0x00, 0x00, 0x00, 0x00,
3097 0x00, 0x00, 0xFF, 0xFF };
3098
Joseph Herlant32b83272018-11-15 11:58:28 -08003099/* Map IPv4 address on IPv6 address, as specified in RFC 3513.
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003100 * Input and output may overlap.
3101 */
David du Colombier4f92d322011-03-24 11:09:31 +01003102void v4tov6(struct in6_addr *sin6_addr, struct in_addr *sin_addr)
3103{
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003104 struct in_addr tmp_addr;
3105
3106 tmp_addr.s_addr = sin_addr->s_addr;
David du Colombier4f92d322011-03-24 11:09:31 +01003107 memcpy(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx));
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003108 memcpy(sin6_addr->s6_addr+12, &tmp_addr.s_addr, 4);
David du Colombier4f92d322011-03-24 11:09:31 +01003109}
3110
Joseph Herlant32b83272018-11-15 11:58:28 -08003111/* Map IPv6 address on IPv4 address, as specified in RFC 3513.
David du Colombier4f92d322011-03-24 11:09:31 +01003112 * Return true if conversion is possible and false otherwise.
3113 */
3114int v6tov4(struct in_addr *sin_addr, struct in6_addr *sin6_addr)
3115{
3116 if (memcmp(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx)) == 0) {
3117 memcpy(&(sin_addr->s_addr), &(sin6_addr->s6_addr[12]),
3118 sizeof(struct in_addr));
3119 return 1;
3120 }
3121
3122 return 0;
3123}
3124
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003125/* compare two struct sockaddr_storage and return:
3126 * 0 (true) if the addr is the same in both
3127 * 1 (false) if the addr is not the same in both
3128 * -1 (unable) if one of the addr is not AF_INET*
3129 */
3130int ipcmp(struct sockaddr_storage *ss1, struct sockaddr_storage *ss2)
3131{
3132 if ((ss1->ss_family != AF_INET) && (ss1->ss_family != AF_INET6))
3133 return -1;
3134
3135 if ((ss2->ss_family != AF_INET) && (ss2->ss_family != AF_INET6))
3136 return -1;
3137
3138 if (ss1->ss_family != ss2->ss_family)
3139 return 1;
3140
3141 switch (ss1->ss_family) {
3142 case AF_INET:
3143 return memcmp(&((struct sockaddr_in *)ss1)->sin_addr,
3144 &((struct sockaddr_in *)ss2)->sin_addr,
3145 sizeof(struct in_addr)) != 0;
3146 case AF_INET6:
3147 return memcmp(&((struct sockaddr_in6 *)ss1)->sin6_addr,
3148 &((struct sockaddr_in6 *)ss2)->sin6_addr,
3149 sizeof(struct in6_addr)) != 0;
3150 }
3151
3152 return 1;
3153}
3154
Christopher Faulet9553de72021-02-26 09:12:50 +01003155/* compare a struct sockaddr_storage to a struct net_addr and return :
3156 * 0 (true) if <addr> is matching <net>
3157 * 1 (false) if <addr> is not matching <net>
3158 * -1 (unable) if <addr> or <net> is not AF_INET*
3159 */
3160int ipcmp2net(const struct sockaddr_storage *addr, const struct net_addr *net)
3161{
3162 if ((addr->ss_family != AF_INET) && (addr->ss_family != AF_INET6))
3163 return -1;
3164
3165 if ((net->family != AF_INET) && (net->family != AF_INET6))
3166 return -1;
3167
3168 if (addr->ss_family != net->family)
3169 return 1;
3170
3171 if (addr->ss_family == AF_INET &&
3172 (((struct sockaddr_in *)addr)->sin_addr.s_addr & net->addr.v4.mask.s_addr) == net->addr.v4.ip.s_addr)
3173 return 0;
3174 else {
3175 const struct in6_addr *addr6 = &(((const struct sockaddr_in6*)addr)->sin6_addr);
3176 const struct in6_addr *nip6 = &net->addr.v6.ip;
3177 const struct in6_addr *nmask6 = &net->addr.v6.mask;
3178
3179 if ((read_u32(&addr6->s6_addr[0]) & read_u32(&nmask6->s6_addr[0])) == read_u32(&nip6->s6_addr[0]) &&
3180 (read_u32(&addr6->s6_addr[4]) & read_u32(&nmask6->s6_addr[4])) == read_u32(&nip6->s6_addr[4]) &&
3181 (read_u32(&addr6->s6_addr[8]) & read_u32(&nmask6->s6_addr[8])) == read_u32(&nip6->s6_addr[8]) &&
3182 (read_u32(&addr6->s6_addr[12]) & read_u32(&nmask6->s6_addr[12])) == read_u32(&nip6->s6_addr[12]))
3183 return 0;
3184 }
3185
3186 return 1;
3187}
3188
Baptiste Assmann08396c82016-01-31 00:27:17 +01003189/* copy IP address from <source> into <dest>
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003190 * The caller must allocate and clear <dest> before calling.
3191 * The source must be in either AF_INET or AF_INET6 family, or the destination
3192 * address will be undefined. If the destination address used to hold a port,
3193 * it is preserved, so that this function can be used to switch to another
3194 * address family with no risk. Returns a pointer to the destination.
Baptiste Assmann08396c82016-01-31 00:27:17 +01003195 */
3196struct sockaddr_storage *ipcpy(struct sockaddr_storage *source, struct sockaddr_storage *dest)
3197{
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003198 int prev_port;
3199
3200 prev_port = get_net_port(dest);
3201 memset(dest, 0, sizeof(*dest));
Baptiste Assmann08396c82016-01-31 00:27:17 +01003202 dest->ss_family = source->ss_family;
3203
3204 /* copy new addr and apply it */
3205 switch (source->ss_family) {
3206 case AF_INET:
3207 ((struct sockaddr_in *)dest)->sin_addr.s_addr = ((struct sockaddr_in *)source)->sin_addr.s_addr;
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003208 ((struct sockaddr_in *)dest)->sin_port = prev_port;
Baptiste Assmann08396c82016-01-31 00:27:17 +01003209 break;
3210 case AF_INET6:
3211 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 +01003212 ((struct sockaddr_in6 *)dest)->sin6_port = prev_port;
Baptiste Assmann08396c82016-01-31 00:27:17 +01003213 break;
3214 }
3215
3216 return dest;
3217}
3218
William Lallemand421f5b52012-02-06 18:15:57 +01003219char *human_time(int t, short hz_div) {
3220 static char rv[sizeof("24855d23h")+1]; // longest of "23h59m" and "59m59s"
3221 char *p = rv;
Willy Tarreau761b3d52014-04-14 14:53:06 +02003222 char *end = rv + sizeof(rv);
William Lallemand421f5b52012-02-06 18:15:57 +01003223 int cnt=2; // print two numbers
3224
3225 if (unlikely(t < 0 || hz_div <= 0)) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003226 snprintf(p, end - p, "?");
William Lallemand421f5b52012-02-06 18:15:57 +01003227 return rv;
3228 }
3229
3230 if (unlikely(hz_div > 1))
3231 t /= hz_div;
3232
3233 if (t >= DAY) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003234 p += snprintf(p, end - p, "%dd", t / DAY);
William Lallemand421f5b52012-02-06 18:15:57 +01003235 cnt--;
3236 }
3237
3238 if (cnt && t % DAY / HOUR) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003239 p += snprintf(p, end - p, "%dh", t % DAY / HOUR);
William Lallemand421f5b52012-02-06 18:15:57 +01003240 cnt--;
3241 }
3242
3243 if (cnt && t % HOUR / MINUTE) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003244 p += snprintf(p, end - p, "%dm", t % HOUR / MINUTE);
William Lallemand421f5b52012-02-06 18:15:57 +01003245 cnt--;
3246 }
3247
3248 if ((cnt && t % MINUTE) || !t) // also display '0s'
Willy Tarreau761b3d52014-04-14 14:53:06 +02003249 p += snprintf(p, end - p, "%ds", t % MINUTE / SEC);
William Lallemand421f5b52012-02-06 18:15:57 +01003250
3251 return rv;
3252}
3253
3254const char *monthname[12] = {
3255 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
3256 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"
3257};
3258
3259/* date2str_log: write a date in the format :
3260 * sprintf(str, "%02d/%s/%04d:%02d:%02d:%02d.%03d",
3261 * tm.tm_mday, monthname[tm.tm_mon], tm.tm_year+1900,
3262 * tm.tm_hour, tm.tm_min, tm.tm_sec, (int)date.tv_usec/1000);
3263 *
3264 * without using sprintf. return a pointer to the last char written (\0) or
3265 * NULL if there isn't enough space.
3266 */
Willy Tarreauf16cb412018-09-04 19:08:48 +02003267char *date2str_log(char *dst, const struct tm *tm, const struct timeval *date, size_t size)
William Lallemand421f5b52012-02-06 18:15:57 +01003268{
3269
3270 if (size < 25) /* the size is fixed: 24 chars + \0 */
3271 return NULL;
3272
3273 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003274 if (!dst)
3275 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003276 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003277
William Lallemand421f5b52012-02-06 18:15:57 +01003278 memcpy(dst, monthname[tm->tm_mon], 3); // month
3279 dst += 3;
3280 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003281
William Lallemand421f5b52012-02-06 18:15:57 +01003282 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003283 if (!dst)
3284 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003285 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003286
William Lallemand421f5b52012-02-06 18:15:57 +01003287 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003288 if (!dst)
3289 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003290 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003291
William Lallemand421f5b52012-02-06 18:15:57 +01003292 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003293 if (!dst)
3294 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003295 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003296
William Lallemand421f5b52012-02-06 18:15:57 +01003297 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003298 if (!dst)
3299 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003300 *dst++ = '.';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003301
Willy Tarreau7d9421d2020-02-29 09:08:02 +01003302 dst = utoa_pad((unsigned int)(date->tv_usec/1000)%1000, dst, 4); // milliseconds
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003303 if (!dst)
3304 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003305 *dst = '\0';
3306
3307 return dst;
3308}
3309
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003310/* Base year used to compute leap years */
3311#define TM_YEAR_BASE 1900
3312
3313/* Return the difference in seconds between two times (leap seconds are ignored).
3314 * Retrieved from glibc 2.18 source code.
3315 */
3316static int my_tm_diff(const struct tm *a, const struct tm *b)
3317{
3318 /* Compute intervening leap days correctly even if year is negative.
3319 * Take care to avoid int overflow in leap day calculations,
3320 * but it's OK to assume that A and B are close to each other.
3321 */
3322 int a4 = (a->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (a->tm_year & 3);
3323 int b4 = (b->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (b->tm_year & 3);
3324 int a100 = a4 / 25 - (a4 % 25 < 0);
3325 int b100 = b4 / 25 - (b4 % 25 < 0);
3326 int a400 = a100 >> 2;
3327 int b400 = b100 >> 2;
3328 int intervening_leap_days = (a4 - b4) - (a100 - b100) + (a400 - b400);
3329 int years = a->tm_year - b->tm_year;
3330 int days = (365 * years + intervening_leap_days
3331 + (a->tm_yday - b->tm_yday));
3332 return (60 * (60 * (24 * days + (a->tm_hour - b->tm_hour))
3333 + (a->tm_min - b->tm_min))
3334 + (a->tm_sec - b->tm_sec));
3335}
3336
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003337/* Return the GMT offset for a specific local time.
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003338 * Both t and tm must represent the same time.
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003339 * The string returned has the same format as returned by strftime(... "%z", tm).
3340 * Offsets are kept in an internal cache for better performances.
3341 */
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003342const char *get_gmt_offset(time_t t, struct tm *tm)
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003343{
3344 /* Cache offsets from GMT (depending on whether DST is active or not) */
Christopher Faulet1bc04c72017-10-29 20:14:08 +01003345 static THREAD_LOCAL char gmt_offsets[2][5+1] = { "", "" };
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003346
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003347 char *gmt_offset;
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003348 struct tm tm_gmt;
3349 int diff;
3350 int isdst = tm->tm_isdst;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003351
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003352 /* Pretend DST not active if its status is unknown */
3353 if (isdst < 0)
3354 isdst = 0;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003355
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003356 /* Fetch the offset and initialize it if needed */
3357 gmt_offset = gmt_offsets[isdst & 0x01];
3358 if (unlikely(!*gmt_offset)) {
3359 get_gmtime(t, &tm_gmt);
3360 diff = my_tm_diff(tm, &tm_gmt);
3361 if (diff < 0) {
3362 diff = -diff;
3363 *gmt_offset = '-';
3364 } else {
3365 *gmt_offset = '+';
3366 }
Willy Tarreaue112c8a2019-10-29 10:16:11 +01003367 diff %= 86400U;
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003368 diff /= 60; /* Convert to minutes */
3369 snprintf(gmt_offset+1, 4+1, "%02d%02d", diff/60, diff%60);
3370 }
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003371
Willy Tarreaue112c8a2019-10-29 10:16:11 +01003372 return gmt_offset;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003373}
3374
William Lallemand421f5b52012-02-06 18:15:57 +01003375/* gmt2str_log: write a date in the format :
3376 * "%02d/%s/%04d:%02d:%02d:%02d +0000" without using snprintf
3377 * return a pointer to the last char written (\0) or
3378 * NULL if there isn't enough space.
3379 */
3380char *gmt2str_log(char *dst, struct tm *tm, size_t size)
3381{
Yuxans Yao4e25b012012-10-19 10:36:09 +08003382 if (size < 27) /* the size is fixed: 26 chars + \0 */
William Lallemand421f5b52012-02-06 18:15:57 +01003383 return NULL;
3384
3385 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003386 if (!dst)
3387 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003388 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003389
William Lallemand421f5b52012-02-06 18:15:57 +01003390 memcpy(dst, monthname[tm->tm_mon], 3); // month
3391 dst += 3;
3392 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003393
William Lallemand421f5b52012-02-06 18:15:57 +01003394 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003395 if (!dst)
3396 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003397 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003398
William Lallemand421f5b52012-02-06 18:15:57 +01003399 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003400 if (!dst)
3401 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003402 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003403
William Lallemand421f5b52012-02-06 18:15:57 +01003404 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003405 if (!dst)
3406 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003407 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003408
William Lallemand421f5b52012-02-06 18:15:57 +01003409 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003410 if (!dst)
3411 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003412 *dst++ = ' ';
3413 *dst++ = '+';
3414 *dst++ = '0';
3415 *dst++ = '0';
3416 *dst++ = '0';
3417 *dst++ = '0';
3418 *dst = '\0';
3419
3420 return dst;
3421}
3422
Yuxans Yao4e25b012012-10-19 10:36:09 +08003423/* localdate2str_log: write a date in the format :
3424 * "%02d/%s/%04d:%02d:%02d:%02d +0000(local timezone)" without using snprintf
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003425 * Both t and tm must represent the same time.
3426 * return a pointer to the last char written (\0) or
3427 * NULL if there isn't enough space.
Yuxans Yao4e25b012012-10-19 10:36:09 +08003428 */
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003429char *localdate2str_log(char *dst, time_t t, struct tm *tm, size_t size)
Yuxans Yao4e25b012012-10-19 10:36:09 +08003430{
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003431 const char *gmt_offset;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003432 if (size < 27) /* the size is fixed: 26 chars + \0 */
3433 return NULL;
3434
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003435 gmt_offset = get_gmt_offset(t, tm);
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003436
Yuxans Yao4e25b012012-10-19 10:36:09 +08003437 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003438 if (!dst)
3439 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003440 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003441
Yuxans Yao4e25b012012-10-19 10:36:09 +08003442 memcpy(dst, monthname[tm->tm_mon], 3); // month
3443 dst += 3;
3444 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003445
Yuxans Yao4e25b012012-10-19 10:36:09 +08003446 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003447 if (!dst)
3448 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003449 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003450
Yuxans Yao4e25b012012-10-19 10:36:09 +08003451 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003452 if (!dst)
3453 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003454 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003455
Yuxans Yao4e25b012012-10-19 10:36:09 +08003456 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003457 if (!dst)
3458 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003459 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003460
Yuxans Yao4e25b012012-10-19 10:36:09 +08003461 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003462 if (!dst)
3463 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003464 *dst++ = ' ';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003465
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003466 memcpy(dst, gmt_offset, 5); // Offset from local time to GMT
Yuxans Yao4e25b012012-10-19 10:36:09 +08003467 dst += 5;
3468 *dst = '\0';
3469
3470 return dst;
3471}
3472
Willy Tarreaucb1949b2017-07-19 19:05:29 +02003473/* Returns the number of seconds since 01/01/1970 0:0:0 GMT for GMT date <tm>.
3474 * It is meant as a portable replacement for timegm() for use with valid inputs.
3475 * Returns undefined results for invalid dates (eg: months out of range 0..11).
3476 */
3477time_t my_timegm(const struct tm *tm)
3478{
3479 /* Each month has 28, 29, 30 or 31 days, or 28+N. The date in the year
3480 * is thus (current month - 1)*28 + cumulated_N[month] to count the
3481 * sum of the extra N days for elapsed months. The sum of all these N
3482 * days doesn't exceed 30 for a complete year (366-12*28) so it fits
3483 * in a 5-bit word. This means that with 60 bits we can represent a
3484 * matrix of all these values at once, which is fast and efficient to
3485 * access. The extra February day for leap years is not counted here.
3486 *
3487 * Jan : none = 0 (0)
3488 * Feb : Jan = 3 (3)
3489 * Mar : Jan..Feb = 3 (3 + 0)
3490 * Apr : Jan..Mar = 6 (3 + 0 + 3)
3491 * May : Jan..Apr = 8 (3 + 0 + 3 + 2)
3492 * Jun : Jan..May = 11 (3 + 0 + 3 + 2 + 3)
3493 * Jul : Jan..Jun = 13 (3 + 0 + 3 + 2 + 3 + 2)
3494 * Aug : Jan..Jul = 16 (3 + 0 + 3 + 2 + 3 + 2 + 3)
3495 * Sep : Jan..Aug = 19 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3)
3496 * Oct : Jan..Sep = 21 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2)
3497 * Nov : Jan..Oct = 24 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3)
3498 * Dec : Jan..Nov = 26 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3 + 2)
3499 */
3500 uint64_t extra =
3501 ( 0ULL << 0*5) + ( 3ULL << 1*5) + ( 3ULL << 2*5) + /* Jan, Feb, Mar, */
3502 ( 6ULL << 3*5) + ( 8ULL << 4*5) + (11ULL << 5*5) + /* Apr, May, Jun, */
3503 (13ULL << 6*5) + (16ULL << 7*5) + (19ULL << 8*5) + /* Jul, Aug, Sep, */
3504 (21ULL << 9*5) + (24ULL << 10*5) + (26ULL << 11*5); /* Oct, Nov, Dec, */
3505
3506 unsigned int y = tm->tm_year + 1900;
3507 unsigned int m = tm->tm_mon;
3508 unsigned long days = 0;
3509
3510 /* days since 1/1/1970 for full years */
3511 days += days_since_zero(y) - days_since_zero(1970);
3512
3513 /* days for full months in the current year */
3514 days += 28 * m + ((extra >> (m * 5)) & 0x1f);
3515
3516 /* count + 1 after March for leap years. A leap year is a year multiple
3517 * of 4, unless it's multiple of 100 without being multiple of 400. 2000
3518 * is leap, 1900 isn't, 1904 is.
3519 */
3520 if ((m > 1) && !(y & 3) && ((y % 100) || !(y % 400)))
3521 days++;
3522
3523 days += tm->tm_mday - 1;
3524 return days * 86400ULL + tm->tm_hour * 3600 + tm->tm_min * 60 + tm->tm_sec;
3525}
3526
Thierry Fournier93127942016-01-20 18:49:45 +01003527/* This function check a char. It returns true and updates
3528 * <date> and <len> pointer to the new position if the
3529 * character is found.
3530 */
3531static inline int parse_expect_char(const char **date, int *len, char c)
3532{
3533 if (*len < 1 || **date != c)
3534 return 0;
3535 (*len)--;
3536 (*date)++;
3537 return 1;
3538}
3539
3540/* This function expects a string <str> of len <l>. It return true and updates.
3541 * <date> and <len> if the string matches, otherwise, it returns false.
3542 */
3543static inline int parse_strcmp(const char **date, int *len, char *str, int l)
3544{
3545 if (*len < l || strncmp(*date, str, l) != 0)
3546 return 0;
3547 (*len) -= l;
3548 (*date) += l;
3549 return 1;
3550}
3551
3552/* This macro converts 3 chars name in integer. */
3553#define STR2I3(__a, __b, __c) ((__a) * 65536 + (__b) * 256 + (__c))
3554
3555/* day-name = %x4D.6F.6E ; "Mon", case-sensitive
3556 * / %x54.75.65 ; "Tue", case-sensitive
3557 * / %x57.65.64 ; "Wed", case-sensitive
3558 * / %x54.68.75 ; "Thu", case-sensitive
3559 * / %x46.72.69 ; "Fri", case-sensitive
3560 * / %x53.61.74 ; "Sat", case-sensitive
3561 * / %x53.75.6E ; "Sun", case-sensitive
3562 *
3563 * This array must be alphabetically sorted
3564 */
3565static inline int parse_http_dayname(const char **date, int *len, struct tm *tm)
3566{
3567 if (*len < 3)
3568 return 0;
3569 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3570 case STR2I3('M','o','n'): tm->tm_wday = 1; break;
3571 case STR2I3('T','u','e'): tm->tm_wday = 2; break;
3572 case STR2I3('W','e','d'): tm->tm_wday = 3; break;
3573 case STR2I3('T','h','u'): tm->tm_wday = 4; break;
3574 case STR2I3('F','r','i'): tm->tm_wday = 5; break;
3575 case STR2I3('S','a','t'): tm->tm_wday = 6; break;
3576 case STR2I3('S','u','n'): tm->tm_wday = 7; break;
3577 default: return 0;
3578 }
3579 *len -= 3;
3580 *date += 3;
3581 return 1;
3582}
3583
3584/* month = %x4A.61.6E ; "Jan", case-sensitive
3585 * / %x46.65.62 ; "Feb", case-sensitive
3586 * / %x4D.61.72 ; "Mar", case-sensitive
3587 * / %x41.70.72 ; "Apr", case-sensitive
3588 * / %x4D.61.79 ; "May", case-sensitive
3589 * / %x4A.75.6E ; "Jun", case-sensitive
3590 * / %x4A.75.6C ; "Jul", case-sensitive
3591 * / %x41.75.67 ; "Aug", case-sensitive
3592 * / %x53.65.70 ; "Sep", case-sensitive
3593 * / %x4F.63.74 ; "Oct", case-sensitive
3594 * / %x4E.6F.76 ; "Nov", case-sensitive
3595 * / %x44.65.63 ; "Dec", case-sensitive
3596 *
3597 * This array must be alphabetically sorted
3598 */
3599static inline int parse_http_monthname(const char **date, int *len, struct tm *tm)
3600{
3601 if (*len < 3)
3602 return 0;
3603 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3604 case STR2I3('J','a','n'): tm->tm_mon = 0; break;
3605 case STR2I3('F','e','b'): tm->tm_mon = 1; break;
3606 case STR2I3('M','a','r'): tm->tm_mon = 2; break;
3607 case STR2I3('A','p','r'): tm->tm_mon = 3; break;
3608 case STR2I3('M','a','y'): tm->tm_mon = 4; break;
3609 case STR2I3('J','u','n'): tm->tm_mon = 5; break;
3610 case STR2I3('J','u','l'): tm->tm_mon = 6; break;
3611 case STR2I3('A','u','g'): tm->tm_mon = 7; break;
3612 case STR2I3('S','e','p'): tm->tm_mon = 8; break;
3613 case STR2I3('O','c','t'): tm->tm_mon = 9; break;
3614 case STR2I3('N','o','v'): tm->tm_mon = 10; break;
3615 case STR2I3('D','e','c'): tm->tm_mon = 11; break;
3616 default: return 0;
3617 }
3618 *len -= 3;
3619 *date += 3;
3620 return 1;
3621}
3622
3623/* day-name-l = %x4D.6F.6E.64.61.79 ; "Monday", case-sensitive
3624 * / %x54.75.65.73.64.61.79 ; "Tuesday", case-sensitive
3625 * / %x57.65.64.6E.65.73.64.61.79 ; "Wednesday", case-sensitive
3626 * / %x54.68.75.72.73.64.61.79 ; "Thursday", case-sensitive
3627 * / %x46.72.69.64.61.79 ; "Friday", case-sensitive
3628 * / %x53.61.74.75.72.64.61.79 ; "Saturday", case-sensitive
3629 * / %x53.75.6E.64.61.79 ; "Sunday", case-sensitive
3630 *
3631 * This array must be alphabetically sorted
3632 */
3633static inline int parse_http_ldayname(const char **date, int *len, struct tm *tm)
3634{
3635 if (*len < 6) /* Minimum length. */
3636 return 0;
3637 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3638 case STR2I3('M','o','n'):
3639 RET0_UNLESS(parse_strcmp(date, len, "Monday", 6));
3640 tm->tm_wday = 1;
3641 return 1;
3642 case STR2I3('T','u','e'):
3643 RET0_UNLESS(parse_strcmp(date, len, "Tuesday", 7));
3644 tm->tm_wday = 2;
3645 return 1;
3646 case STR2I3('W','e','d'):
3647 RET0_UNLESS(parse_strcmp(date, len, "Wednesday", 9));
3648 tm->tm_wday = 3;
3649 return 1;
3650 case STR2I3('T','h','u'):
3651 RET0_UNLESS(parse_strcmp(date, len, "Thursday", 8));
3652 tm->tm_wday = 4;
3653 return 1;
3654 case STR2I3('F','r','i'):
3655 RET0_UNLESS(parse_strcmp(date, len, "Friday", 6));
3656 tm->tm_wday = 5;
3657 return 1;
3658 case STR2I3('S','a','t'):
3659 RET0_UNLESS(parse_strcmp(date, len, "Saturday", 8));
3660 tm->tm_wday = 6;
3661 return 1;
3662 case STR2I3('S','u','n'):
3663 RET0_UNLESS(parse_strcmp(date, len, "Sunday", 6));
3664 tm->tm_wday = 7;
3665 return 1;
3666 }
3667 return 0;
3668}
3669
3670/* This function parses exactly 1 digit and returns the numeric value in "digit". */
3671static inline int parse_digit(const char **date, int *len, int *digit)
3672{
3673 if (*len < 1 || **date < '0' || **date > '9')
3674 return 0;
3675 *digit = (**date - '0');
3676 (*date)++;
3677 (*len)--;
3678 return 1;
3679}
3680
3681/* This function parses exactly 2 digits and returns the numeric value in "digit". */
3682static inline int parse_2digit(const char **date, int *len, int *digit)
3683{
3684 int value;
3685
3686 RET0_UNLESS(parse_digit(date, len, &value));
3687 (*digit) = value * 10;
3688 RET0_UNLESS(parse_digit(date, len, &value));
3689 (*digit) += value;
3690
3691 return 1;
3692}
3693
3694/* This function parses exactly 4 digits and returns the numeric value in "digit". */
3695static inline int parse_4digit(const char **date, int *len, int *digit)
3696{
3697 int value;
3698
3699 RET0_UNLESS(parse_digit(date, len, &value));
3700 (*digit) = value * 1000;
3701
3702 RET0_UNLESS(parse_digit(date, len, &value));
3703 (*digit) += value * 100;
3704
3705 RET0_UNLESS(parse_digit(date, len, &value));
3706 (*digit) += value * 10;
3707
3708 RET0_UNLESS(parse_digit(date, len, &value));
3709 (*digit) += value;
3710
3711 return 1;
3712}
3713
3714/* time-of-day = hour ":" minute ":" second
3715 * ; 00:00:00 - 23:59:60 (leap second)
3716 *
3717 * hour = 2DIGIT
3718 * minute = 2DIGIT
3719 * second = 2DIGIT
3720 */
3721static inline int parse_http_time(const char **date, int *len, struct tm *tm)
3722{
3723 RET0_UNLESS(parse_2digit(date, len, &tm->tm_hour)); /* hour 2DIGIT */
3724 RET0_UNLESS(parse_expect_char(date, len, ':')); /* expect ":" */
3725 RET0_UNLESS(parse_2digit(date, len, &tm->tm_min)); /* min 2DIGIT */
3726 RET0_UNLESS(parse_expect_char(date, len, ':')); /* expect ":" */
3727 RET0_UNLESS(parse_2digit(date, len, &tm->tm_sec)); /* sec 2DIGIT */
3728 return 1;
3729}
3730
3731/* From RFC7231
3732 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3733 *
3734 * IMF-fixdate = day-name "," SP date1 SP time-of-day SP GMT
3735 * ; fixed length/zone/capitalization subset of the format
3736 * ; see Section 3.3 of [RFC5322]
3737 *
3738 *
3739 * date1 = day SP month SP year
3740 * ; e.g., 02 Jun 1982
3741 *
3742 * day = 2DIGIT
3743 * year = 4DIGIT
3744 *
3745 * GMT = %x47.4D.54 ; "GMT", case-sensitive
3746 *
3747 * time-of-day = hour ":" minute ":" second
3748 * ; 00:00:00 - 23:59:60 (leap second)
3749 *
3750 * hour = 2DIGIT
3751 * minute = 2DIGIT
3752 * second = 2DIGIT
3753 *
3754 * DIGIT = decimal 0-9
3755 */
3756int parse_imf_date(const char *date, int len, struct tm *tm)
3757{
David Carlier327298c2016-11-20 10:42:38 +00003758 /* tm_gmtoff, if present, ought to be zero'ed */
3759 memset(tm, 0, sizeof(*tm));
3760
Thierry Fournier93127942016-01-20 18:49:45 +01003761 RET0_UNLESS(parse_http_dayname(&date, &len, tm)); /* day-name */
3762 RET0_UNLESS(parse_expect_char(&date, &len, ',')); /* expect "," */
3763 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3764 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
3765 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3766 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* Month */
3767 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3768 RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
3769 tm->tm_year -= 1900;
3770 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3771 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3772 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3773 RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3)); /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
3774 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003775 return 1;
3776}
3777
3778/* From RFC7231
3779 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3780 *
3781 * rfc850-date = day-name-l "," SP date2 SP time-of-day SP GMT
3782 * date2 = day "-" month "-" 2DIGIT
3783 * ; e.g., 02-Jun-82
3784 *
3785 * day = 2DIGIT
3786 */
3787int parse_rfc850_date(const char *date, int len, struct tm *tm)
3788{
3789 int year;
3790
David Carlier327298c2016-11-20 10:42:38 +00003791 /* tm_gmtoff, if present, ought to be zero'ed */
3792 memset(tm, 0, sizeof(*tm));
3793
Thierry Fournier93127942016-01-20 18:49:45 +01003794 RET0_UNLESS(parse_http_ldayname(&date, &len, tm)); /* Read the day name */
3795 RET0_UNLESS(parse_expect_char(&date, &len, ',')); /* expect "," */
3796 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3797 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
3798 RET0_UNLESS(parse_expect_char(&date, &len, '-')); /* expect "-" */
3799 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* Month */
3800 RET0_UNLESS(parse_expect_char(&date, &len, '-')); /* expect "-" */
3801
3802 /* year = 2DIGIT
3803 *
3804 * Recipients of a timestamp value in rfc850-(*date) format, which uses a
3805 * two-digit year, MUST interpret a timestamp that appears to be more
3806 * than 50 years in the future as representing the most recent year in
3807 * the past that had the same last two digits.
3808 */
3809 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_year));
3810
3811 /* expect SP */
3812 if (!parse_expect_char(&date, &len, ' ')) {
3813 /* Maybe we have the date with 4 digits. */
3814 RET0_UNLESS(parse_2digit(&date, &len, &year));
3815 tm->tm_year = (tm->tm_year * 100 + year) - 1900;
3816 /* expect SP */
3817 RET0_UNLESS(parse_expect_char(&date, &len, ' '));
3818 } else {
3819 /* I fix 60 as pivot: >60: +1900, <60: +2000. Note that the
3820 * tm_year is the number of year since 1900, so for +1900, we
3821 * do nothing, and for +2000, we add 100.
3822 */
3823 if (tm->tm_year <= 60)
3824 tm->tm_year += 100;
3825 }
3826
3827 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3828 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3829 RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3)); /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
3830 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003831
3832 return 1;
3833}
3834
3835/* From RFC7231
3836 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3837 *
3838 * asctime-date = day-name SP date3 SP time-of-day SP year
3839 * date3 = month SP ( 2DIGIT / ( SP 1DIGIT ))
3840 * ; e.g., Jun 2
3841 *
3842 * HTTP-date is case sensitive. A sender MUST NOT generate additional
3843 * whitespace in an HTTP-date beyond that specifically included as SP in
3844 * the grammar.
3845 */
3846int parse_asctime_date(const char *date, int len, struct tm *tm)
3847{
David Carlier327298c2016-11-20 10:42:38 +00003848 /* tm_gmtoff, if present, ought to be zero'ed */
3849 memset(tm, 0, sizeof(*tm));
3850
Thierry Fournier93127942016-01-20 18:49:45 +01003851 RET0_UNLESS(parse_http_dayname(&date, &len, tm)); /* day-name */
3852 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3853 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* expect month */
3854 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3855
3856 /* expect SP and 1DIGIT or 2DIGIT */
3857 if (parse_expect_char(&date, &len, ' '))
3858 RET0_UNLESS(parse_digit(&date, &len, &tm->tm_mday));
3859 else
3860 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday));
3861
3862 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3863 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3864 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3865 RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
3866 tm->tm_year -= 1900;
3867 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003868 return 1;
3869}
3870
3871/* From RFC7231
3872 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3873 *
3874 * HTTP-date = IMF-fixdate / obs-date
3875 * obs-date = rfc850-date / asctime-date
3876 *
3877 * parses an HTTP date in the RFC format and is accepted
3878 * alternatives. <date> is the strinf containing the date,
3879 * len is the len of the string. <tm> is filled with the
3880 * parsed time. We must considers this time as GMT.
3881 */
3882int parse_http_date(const char *date, int len, struct tm *tm)
3883{
3884 if (parse_imf_date(date, len, tm))
3885 return 1;
3886
3887 if (parse_rfc850_date(date, len, tm))
3888 return 1;
3889
3890 if (parse_asctime_date(date, len, tm))
3891 return 1;
3892
3893 return 0;
3894}
3895
Willy Tarreau4deeb102021-01-29 10:47:52 +01003896/* print the time <ns> in a short form (exactly 7 chars) at the end of buffer
3897 * <out>. "-" is printed if the value is zero, "inf" if larger than 1000 years.
3898 * It returns the new buffer length, or 0 if it doesn't fit. The value will be
3899 * surrounded by <pfx> and <sfx> respectively if not NULL.
3900 */
3901int print_time_short(struct buffer *out, const char *pfx, uint64_t ns, const char *sfx)
3902{
3903 double val = ns; // 52 bits of mantissa keep ns accuracy over 52 days
3904 const char *unit;
3905
3906 if (!pfx)
3907 pfx = "";
3908 if (!sfx)
3909 sfx = "";
3910
3911 do {
3912 unit = " - "; if (val <= 0.0) break;
3913 unit = "ns"; if (val < 1000.0) break;
3914 unit = "us"; val /= 1000.0; if (val < 1000.0) break;
3915 unit = "ms"; val /= 1000.0; if (val < 1000.0) break;
3916 unit = "s "; val /= 1000.0; if (val < 60.0) break;
3917 unit = "m "; val /= 60.0; if (val < 60.0) break;
3918 unit = "h "; val /= 60.0; if (val < 24.0) break;
3919 unit = "d "; val /= 24.0; if (val < 365.0) break;
3920 unit = "yr"; val /= 365.0; if (val < 1000.0) break;
3921 unit = " inf "; val = 0.0; break;
3922 } while (0);
3923
3924 if (val <= 0.0)
3925 return chunk_appendf(out, "%s%7s%s", pfx, unit, sfx);
3926 else if (val < 10.0)
3927 return chunk_appendf(out, "%s%1.3f%s%s", pfx, val, unit, sfx);
3928 else if (val < 100.0)
3929 return chunk_appendf(out, "%s%2.2f%s%s", pfx, val, unit, sfx);
3930 else
3931 return chunk_appendf(out, "%s%3.1f%s%s", pfx, val, unit, sfx);
3932}
3933
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003934/* Dynamically allocates a string of the proper length to hold the formatted
3935 * output. NULL is returned on error. The caller is responsible for freeing the
3936 * memory area using free(). The resulting string is returned in <out> if the
3937 * pointer is not NULL. A previous version of <out> might be used to build the
3938 * new string, and it will be freed before returning if it is not NULL, which
3939 * makes it possible to build complex strings from iterative calls without
3940 * having to care about freeing intermediate values, as in the example below :
3941 *
3942 * memprintf(&err, "invalid argument: '%s'", arg);
3943 * ...
3944 * memprintf(&err, "parser said : <%s>\n", *err);
3945 * ...
3946 * free(*err);
3947 *
3948 * This means that <err> must be initialized to NULL before first invocation.
3949 * The return value also holds the allocated string, which eases error checking
3950 * and immediate consumption. If the output pointer is not used, NULL must be
Willy Tarreaueb6cead2012-09-20 19:43:14 +02003951 * passed instead and it will be ignored. The returned message will then also
3952 * be NULL so that the caller does not have to bother with freeing anything.
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003953 *
3954 * It is also convenient to use it without any free except the last one :
3955 * err = NULL;
3956 * if (!fct1(err)) report(*err);
3957 * if (!fct2(err)) report(*err);
3958 * if (!fct3(err)) report(*err);
3959 * free(*err);
Christopher Faulet93a518f2017-10-24 11:25:33 +02003960 *
3961 * memprintf relies on memvprintf. This last version can be called from any
3962 * function with variadic arguments.
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003963 */
Christopher Faulet93a518f2017-10-24 11:25:33 +02003964char *memvprintf(char **out, const char *format, va_list orig_args)
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003965{
3966 va_list args;
3967 char *ret = NULL;
3968 int allocated = 0;
3969 int needed = 0;
3970
Willy Tarreaueb6cead2012-09-20 19:43:14 +02003971 if (!out)
3972 return NULL;
3973
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003974 do {
Willy Tarreaue0609f52019-03-29 19:13:23 +01003975 char buf1;
3976
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003977 /* vsnprintf() will return the required length even when the
3978 * target buffer is NULL. We do this in a loop just in case
3979 * intermediate evaluations get wrong.
3980 */
Christopher Faulet93a518f2017-10-24 11:25:33 +02003981 va_copy(args, orig_args);
Willy Tarreaue0609f52019-03-29 19:13:23 +01003982 needed = vsnprintf(ret ? ret : &buf1, allocated, format, args);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003983 va_end(args);
Willy Tarreau1b2fed62013-04-01 22:48:54 +02003984 if (needed < allocated) {
3985 /* Note: on Solaris 8, the first iteration always
3986 * returns -1 if allocated is zero, so we force a
3987 * retry.
3988 */
3989 if (!allocated)
3990 needed = 0;
3991 else
3992 break;
3993 }
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003994
Willy Tarreau1b2fed62013-04-01 22:48:54 +02003995 allocated = needed + 1;
Hubert Verstraete831962e2016-06-28 22:44:26 +02003996 ret = my_realloc2(ret, allocated);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02003997 } while (ret);
3998
3999 if (needed < 0) {
4000 /* an error was encountered */
Willy Tarreau61cfdf42021-02-20 10:46:51 +01004001 ha_free(&ret);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004002 }
4003
4004 if (out) {
4005 free(*out);
4006 *out = ret;
4007 }
4008
4009 return ret;
4010}
William Lallemand421f5b52012-02-06 18:15:57 +01004011
Christopher Faulet93a518f2017-10-24 11:25:33 +02004012char *memprintf(char **out, const char *format, ...)
4013{
4014 va_list args;
4015 char *ret = NULL;
4016
4017 va_start(args, format);
4018 ret = memvprintf(out, format, args);
4019 va_end(args);
4020
4021 return ret;
4022}
4023
Willy Tarreau21c705b2012-09-14 11:40:36 +02004024/* Used to add <level> spaces before each line of <out>, unless there is only one line.
4025 * The input argument is automatically freed and reassigned. The result will have to be
Willy Tarreau70eec382012-10-10 08:56:47 +02004026 * freed by the caller. It also supports being passed a NULL which results in the same
4027 * output.
Willy Tarreau21c705b2012-09-14 11:40:36 +02004028 * Example of use :
4029 * parse(cmd, &err); (callee: memprintf(&err, ...))
4030 * fprintf(stderr, "Parser said: %s\n", indent_error(&err));
4031 * free(err);
4032 */
4033char *indent_msg(char **out, int level)
4034{
4035 char *ret, *in, *p;
4036 int needed = 0;
4037 int lf = 0;
4038 int lastlf = 0;
4039 int len;
4040
Willy Tarreau70eec382012-10-10 08:56:47 +02004041 if (!out || !*out)
4042 return NULL;
4043
Willy Tarreau21c705b2012-09-14 11:40:36 +02004044 in = *out - 1;
4045 while ((in = strchr(in + 1, '\n')) != NULL) {
4046 lastlf = in - *out;
4047 lf++;
4048 }
4049
4050 if (!lf) /* single line, no LF, return it as-is */
4051 return *out;
4052
4053 len = strlen(*out);
4054
4055 if (lf == 1 && lastlf == len - 1) {
4056 /* single line, LF at end, strip it and return as-is */
4057 (*out)[lastlf] = 0;
4058 return *out;
4059 }
4060
4061 /* OK now we have at least one LF, we need to process the whole string
4062 * as a multi-line string. What we'll do :
4063 * - prefix with an LF if there is none
4064 * - add <level> spaces before each line
4065 * This means at most ( 1 + level + (len-lf) + lf*<1+level) ) =
4066 * 1 + level + len + lf * level = 1 + level * (lf + 1) + len.
4067 */
4068
4069 needed = 1 + level * (lf + 1) + len + 1;
4070 p = ret = malloc(needed);
4071 in = *out;
4072
4073 /* skip initial LFs */
4074 while (*in == '\n')
4075 in++;
4076
4077 /* copy each line, prefixed with LF and <level> spaces, and without the trailing LF */
4078 while (*in) {
4079 *p++ = '\n';
4080 memset(p, ' ', level);
4081 p += level;
4082 do {
4083 *p++ = *in++;
4084 } while (*in && *in != '\n');
4085 if (*in)
4086 in++;
4087 }
4088 *p = 0;
4089
4090 free(*out);
4091 *out = ret;
4092
4093 return ret;
4094}
4095
Willy Tarreaua2c99112019-08-21 13:17:37 +02004096/* makes a copy of message <in> into <out>, with each line prefixed with <pfx>
4097 * and end of lines replaced with <eol> if not 0. The first line to indent has
4098 * to be indicated in <first> (starts at zero), so that it is possible to skip
4099 * indenting the first line if it has to be appended after an existing message.
4100 * Empty strings are never indented, and NULL strings are considered empty both
4101 * for <in> and <pfx>. It returns non-zero if an EOL was appended as the last
4102 * character, non-zero otherwise.
4103 */
4104int append_prefixed_str(struct buffer *out, const char *in, const char *pfx, char eol, int first)
4105{
4106 int bol, lf;
4107 int pfxlen = pfx ? strlen(pfx) : 0;
4108
4109 if (!in)
4110 return 0;
4111
4112 bol = 1;
4113 lf = 0;
4114 while (*in) {
4115 if (bol && pfxlen) {
4116 if (first > 0)
4117 first--;
4118 else
4119 b_putblk(out, pfx, pfxlen);
4120 bol = 0;
4121 }
4122
4123 lf = (*in == '\n');
4124 bol |= lf;
4125 b_putchr(out, (lf && eol) ? eol : *in);
4126 in++;
4127 }
4128 return lf;
4129}
4130
Willy Tarreau9d22e562019-03-29 18:49:09 +01004131/* removes environment variable <name> from the environment as found in
4132 * environ. This is only provided as an alternative for systems without
4133 * unsetenv() (old Solaris and AIX versions). THIS IS NOT THREAD SAFE.
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05004134 * The principle is to scan environ for each occurrence of variable name
Willy Tarreau9d22e562019-03-29 18:49:09 +01004135 * <name> and to replace the matching pointers with the last pointer of
4136 * the array (since variables are not ordered).
4137 * It always returns 0 (success).
4138 */
4139int my_unsetenv(const char *name)
4140{
4141 extern char **environ;
4142 char **p = environ;
4143 int vars;
4144 int next;
4145 int len;
4146
4147 len = strlen(name);
4148 for (vars = 0; p[vars]; vars++)
4149 ;
4150 next = 0;
4151 while (next < vars) {
4152 if (strncmp(p[next], name, len) != 0 || p[next][len] != '=') {
4153 next++;
4154 continue;
4155 }
4156 if (next < vars - 1)
4157 p[next] = p[vars - 1];
4158 p[--vars] = NULL;
4159 }
4160 return 0;
4161}
4162
Willy Tarreaudad36a32013-03-11 01:20:04 +01004163/* Convert occurrences of environment variables in the input string to their
4164 * corresponding value. A variable is identified as a series of alphanumeric
4165 * characters or underscores following a '$' sign. The <in> string must be
4166 * free()able. NULL returns NULL. The resulting string might be reallocated if
4167 * some expansion is made. Variable names may also be enclosed into braces if
4168 * needed (eg: to concatenate alphanum characters).
4169 */
4170char *env_expand(char *in)
4171{
4172 char *txt_beg;
4173 char *out;
4174 char *txt_end;
4175 char *var_beg;
4176 char *var_end;
4177 char *value;
4178 char *next;
4179 int out_len;
4180 int val_len;
4181
4182 if (!in)
4183 return in;
4184
4185 value = out = NULL;
4186 out_len = 0;
4187
4188 txt_beg = in;
4189 do {
4190 /* look for next '$' sign in <in> */
4191 for (txt_end = txt_beg; *txt_end && *txt_end != '$'; txt_end++);
4192
4193 if (!*txt_end && !out) /* end and no expansion performed */
4194 return in;
4195
4196 val_len = 0;
4197 next = txt_end;
4198 if (*txt_end == '$') {
4199 char save;
4200
4201 var_beg = txt_end + 1;
4202 if (*var_beg == '{')
4203 var_beg++;
4204
4205 var_end = var_beg;
Willy Tarreau90807112020-02-25 08:16:33 +01004206 while (isalnum((unsigned char)*var_end) || *var_end == '_') {
Willy Tarreaudad36a32013-03-11 01:20:04 +01004207 var_end++;
4208 }
4209
4210 next = var_end;
4211 if (*var_end == '}' && (var_beg > txt_end + 1))
4212 next++;
4213
4214 /* get value of the variable name at this location */
4215 save = *var_end;
4216 *var_end = '\0';
4217 value = getenv(var_beg);
4218 *var_end = save;
4219 val_len = value ? strlen(value) : 0;
4220 }
4221
Hubert Verstraete831962e2016-06-28 22:44:26 +02004222 out = my_realloc2(out, out_len + (txt_end - txt_beg) + val_len + 1);
Willy Tarreaudad36a32013-03-11 01:20:04 +01004223 if (txt_end > txt_beg) {
4224 memcpy(out + out_len, txt_beg, txt_end - txt_beg);
4225 out_len += txt_end - txt_beg;
4226 }
4227 if (val_len) {
4228 memcpy(out + out_len, value, val_len);
4229 out_len += val_len;
4230 }
4231 out[out_len] = 0;
4232 txt_beg = next;
4233 } while (*txt_beg);
4234
4235 /* here we know that <out> was allocated and that we don't need <in> anymore */
4236 free(in);
4237 return out;
4238}
4239
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004240
4241/* same as strstr() but case-insensitive and with limit length */
4242const char *strnistr(const char *str1, int len_str1, const char *str2, int len_str2)
4243{
4244 char *pptr, *sptr, *start;
Willy Tarreauc8746532014-05-28 23:05:07 +02004245 unsigned int slen, plen;
4246 unsigned int tmp1, tmp2;
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004247
4248 if (str1 == NULL || len_str1 == 0) // search pattern into an empty string => search is not found
4249 return NULL;
4250
4251 if (str2 == NULL || len_str2 == 0) // pattern is empty => every str1 match
4252 return str1;
4253
4254 if (len_str1 < len_str2) // pattern is longer than string => search is not found
4255 return NULL;
4256
4257 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 +02004258 while (toupper((unsigned char)*start) != toupper((unsigned char)*str2)) {
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004259 start++;
4260 slen--;
4261 tmp1++;
4262
4263 if (tmp1 >= len_str1)
4264 return NULL;
4265
4266 /* if pattern longer than string */
4267 if (slen < plen)
4268 return NULL;
4269 }
4270
4271 sptr = start;
4272 pptr = (char *)str2;
4273
4274 tmp2 = 0;
Willy Tarreauf278eec2020-07-05 21:46:32 +02004275 while (toupper((unsigned char)*sptr) == toupper((unsigned char)*pptr)) {
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004276 sptr++;
4277 pptr++;
4278 tmp2++;
4279
4280 if (*pptr == '\0' || tmp2 == len_str2) /* end of pattern found */
4281 return start;
4282 if (*sptr == '\0' || tmp2 == len_str1) /* end of string found and the pattern is not fully found */
4283 return NULL;
4284 }
4285 }
4286 return NULL;
4287}
4288
Thierry FOURNIER317e1c42014-08-12 10:20:47 +02004289/* This function read the next valid utf8 char.
4290 * <s> is the byte srray to be decode, <len> is its length.
4291 * The function returns decoded char encoded like this:
4292 * The 4 msb are the return code (UTF8_CODE_*), the 4 lsb
4293 * are the length read. The decoded character is stored in <c>.
4294 */
4295unsigned char utf8_next(const char *s, int len, unsigned int *c)
4296{
4297 const unsigned char *p = (unsigned char *)s;
4298 int dec;
4299 unsigned char code = UTF8_CODE_OK;
4300
4301 if (len < 1)
4302 return UTF8_CODE_OK;
4303
4304 /* Check the type of UTF8 sequence
4305 *
4306 * 0... .... 0x00 <= x <= 0x7f : 1 byte: ascii char
4307 * 10.. .... 0x80 <= x <= 0xbf : invalid sequence
4308 * 110. .... 0xc0 <= x <= 0xdf : 2 bytes
4309 * 1110 .... 0xe0 <= x <= 0xef : 3 bytes
4310 * 1111 0... 0xf0 <= x <= 0xf7 : 4 bytes
4311 * 1111 10.. 0xf8 <= x <= 0xfb : 5 bytes
4312 * 1111 110. 0xfc <= x <= 0xfd : 6 bytes
4313 * 1111 111. 0xfe <= x <= 0xff : invalid sequence
4314 */
4315 switch (*p) {
4316 case 0x00 ... 0x7f:
4317 *c = *p;
4318 return UTF8_CODE_OK | 1;
4319
4320 case 0x80 ... 0xbf:
4321 *c = *p;
4322 return UTF8_CODE_BADSEQ | 1;
4323
4324 case 0xc0 ... 0xdf:
4325 if (len < 2) {
4326 *c = *p;
4327 return UTF8_CODE_BADSEQ | 1;
4328 }
4329 *c = *p & 0x1f;
4330 dec = 1;
4331 break;
4332
4333 case 0xe0 ... 0xef:
4334 if (len < 3) {
4335 *c = *p;
4336 return UTF8_CODE_BADSEQ | 1;
4337 }
4338 *c = *p & 0x0f;
4339 dec = 2;
4340 break;
4341
4342 case 0xf0 ... 0xf7:
4343 if (len < 4) {
4344 *c = *p;
4345 return UTF8_CODE_BADSEQ | 1;
4346 }
4347 *c = *p & 0x07;
4348 dec = 3;
4349 break;
4350
4351 case 0xf8 ... 0xfb:
4352 if (len < 5) {
4353 *c = *p;
4354 return UTF8_CODE_BADSEQ | 1;
4355 }
4356 *c = *p & 0x03;
4357 dec = 4;
4358 break;
4359
4360 case 0xfc ... 0xfd:
4361 if (len < 6) {
4362 *c = *p;
4363 return UTF8_CODE_BADSEQ | 1;
4364 }
4365 *c = *p & 0x01;
4366 dec = 5;
4367 break;
4368
4369 case 0xfe ... 0xff:
4370 default:
4371 *c = *p;
4372 return UTF8_CODE_BADSEQ | 1;
4373 }
4374
4375 p++;
4376
4377 while (dec > 0) {
4378
4379 /* need 0x10 for the 2 first bits */
4380 if ( ( *p & 0xc0 ) != 0x80 )
4381 return UTF8_CODE_BADSEQ | ((p-(unsigned char *)s)&0xffff);
4382
4383 /* add data at char */
4384 *c = ( *c << 6 ) | ( *p & 0x3f );
4385
4386 dec--;
4387 p++;
4388 }
4389
4390 /* Check ovelong encoding.
4391 * 1 byte : 5 + 6 : 11 : 0x80 ... 0x7ff
4392 * 2 bytes : 4 + 6 + 6 : 16 : 0x800 ... 0xffff
4393 * 3 bytes : 3 + 6 + 6 + 6 : 21 : 0x10000 ... 0x1fffff
4394 */
Thierry FOURNIER9e7ec082015-03-12 19:32:38 +01004395 if (( *c <= 0x7f && (p-(unsigned char *)s) > 1) ||
Thierry FOURNIER317e1c42014-08-12 10:20:47 +02004396 (*c >= 0x80 && *c <= 0x7ff && (p-(unsigned char *)s) > 2) ||
4397 (*c >= 0x800 && *c <= 0xffff && (p-(unsigned char *)s) > 3) ||
4398 (*c >= 0x10000 && *c <= 0x1fffff && (p-(unsigned char *)s) > 4))
4399 code |= UTF8_CODE_OVERLONG;
4400
4401 /* Check invalid UTF8 range. */
4402 if ((*c >= 0xd800 && *c <= 0xdfff) ||
4403 (*c >= 0xfffe && *c <= 0xffff))
4404 code |= UTF8_CODE_INVRANGE;
4405
4406 return code | ((p-(unsigned char *)s)&0x0f);
4407}
4408
Maxime de Roucydc887852016-05-13 23:52:54 +02004409/* append a copy of string <str> (in a wordlist) at the end of the list <li>
4410 * On failure : return 0 and <err> filled with an error message.
4411 * The caller is responsible for freeing the <err> and <str> copy
4412 * memory area using free()
4413 */
4414int list_append_word(struct list *li, const char *str, char **err)
4415{
4416 struct wordlist *wl;
4417
4418 wl = calloc(1, sizeof(*wl));
4419 if (!wl) {
4420 memprintf(err, "out of memory");
4421 goto fail_wl;
4422 }
4423
4424 wl->s = strdup(str);
4425 if (!wl->s) {
4426 memprintf(err, "out of memory");
4427 goto fail_wl_s;
4428 }
4429
Willy Tarreau2b718102021-04-21 07:32:39 +02004430 LIST_APPEND(li, &wl->list);
Maxime de Roucydc887852016-05-13 23:52:54 +02004431
4432 return 1;
4433
4434fail_wl_s:
4435 free(wl->s);
4436fail_wl:
4437 free(wl);
4438 return 0;
4439}
4440
Willy Tarreau37101052019-05-20 16:48:20 +02004441/* indicates if a memory location may safely be read or not. The trick consists
4442 * in performing a harmless syscall using this location as an input and letting
4443 * the operating system report whether it's OK or not. For this we have the
4444 * stat() syscall, which will return EFAULT when the memory location supposed
4445 * to contain the file name is not readable. If it is readable it will then
4446 * either return 0 if the area contains an existing file name, or -1 with
4447 * another code. This must not be abused, and some audit systems might detect
4448 * this as abnormal activity. It's used only for unsafe dumps.
4449 */
4450int may_access(const void *ptr)
4451{
4452 struct stat buf;
4453
4454 if (stat(ptr, &buf) == 0)
4455 return 1;
4456 if (errno == EFAULT)
4457 return 0;
4458 return 1;
4459}
4460
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004461/* print a string of text buffer to <out>. The format is :
4462 * Non-printable chars \t, \n, \r and \e are * encoded in C format.
4463 * Other non-printable chars are encoded "\xHH". Space, '\', and '=' are also escaped.
4464 * Print stopped if null char or <bsize> is reached, or if no more place in the chunk.
4465 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004466int dump_text(struct buffer *out, const char *buf, int bsize)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004467{
4468 unsigned char c;
4469 int ptr = 0;
4470
4471 while (buf[ptr] && ptr < bsize) {
4472 c = buf[ptr];
Willy Tarreau90807112020-02-25 08:16:33 +01004473 if (isprint((unsigned char)c) && isascii((unsigned char)c) && c != '\\' && c != ' ' && c != '=') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004474 if (out->data > out->size - 1)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004475 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004476 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004477 }
4478 else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\' || c == ' ' || c == '=') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004479 if (out->data > out->size - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004480 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004481 out->area[out->data++] = '\\';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004482 switch (c) {
4483 case ' ': c = ' '; break;
4484 case '\t': c = 't'; break;
4485 case '\n': c = 'n'; break;
4486 case '\r': c = 'r'; break;
4487 case '\e': c = 'e'; break;
4488 case '\\': c = '\\'; break;
4489 case '=': c = '='; break;
4490 }
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004491 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004492 }
4493 else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004494 if (out->data > out->size - 4)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004495 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004496 out->area[out->data++] = '\\';
4497 out->area[out->data++] = 'x';
4498 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4499 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004500 }
4501 ptr++;
4502 }
4503
4504 return ptr;
4505}
4506
4507/* print a buffer in hexa.
4508 * Print stopped if <bsize> is reached, or if no more place in the chunk.
4509 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004510int dump_binary(struct buffer *out, const char *buf, int bsize)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004511{
4512 unsigned char c;
4513 int ptr = 0;
4514
4515 while (ptr < bsize) {
4516 c = buf[ptr];
4517
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004518 if (out->data > out->size - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004519 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004520 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4521 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004522
4523 ptr++;
4524 }
4525 return ptr;
4526}
4527
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004528/* Appends into buffer <out> a hex dump of memory area <buf> for <len> bytes,
4529 * prepending each line with prefix <pfx>. The output is *not* initialized.
4530 * The output will not wrap pas the buffer's end so it is more optimal if the
4531 * caller makes sure the buffer is aligned first. A trailing zero will always
4532 * be appended (and not counted) if there is room for it. The caller must make
Willy Tarreau37101052019-05-20 16:48:20 +02004533 * sure that the area is dumpable first. If <unsafe> is non-null, the memory
4534 * locations are checked first for being readable.
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004535 */
Willy Tarreau37101052019-05-20 16:48:20 +02004536void dump_hex(struct buffer *out, const char *pfx, const void *buf, int len, int unsafe)
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004537{
4538 const unsigned char *d = buf;
4539 int i, j, start;
4540
4541 d = (const unsigned char *)(((unsigned long)buf) & -16);
4542 start = ((unsigned long)buf) & 15;
4543
4544 for (i = 0; i < start + len; i += 16) {
4545 chunk_appendf(out, (sizeof(void *) == 4) ? "%s%8p: " : "%s%16p: ", pfx, d + i);
4546
Willy Tarreau37101052019-05-20 16:48:20 +02004547 // 0: unchecked, 1: checked safe, 2: danger
4548 unsafe = !!unsafe;
4549 if (unsafe && !may_access(d + i))
4550 unsafe = 2;
4551
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004552 for (j = 0; j < 16; j++) {
Willy Tarreau37101052019-05-20 16:48:20 +02004553 if ((i + j < start) || (i + j >= start + len))
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004554 chunk_strcat(out, "'' ");
Willy Tarreau37101052019-05-20 16:48:20 +02004555 else if (unsafe > 1)
4556 chunk_strcat(out, "** ");
4557 else
4558 chunk_appendf(out, "%02x ", d[i + j]);
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004559
4560 if (j == 7)
4561 chunk_strcat(out, "- ");
4562 }
4563 chunk_strcat(out, " ");
4564 for (j = 0; j < 16; j++) {
Willy Tarreau37101052019-05-20 16:48:20 +02004565 if ((i + j < start) || (i + j >= start + len))
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004566 chunk_strcat(out, "'");
Willy Tarreau37101052019-05-20 16:48:20 +02004567 else if (unsafe > 1)
4568 chunk_strcat(out, "*");
Willy Tarreau90807112020-02-25 08:16:33 +01004569 else if (isprint((unsigned char)d[i + j]))
Willy Tarreau37101052019-05-20 16:48:20 +02004570 chunk_appendf(out, "%c", d[i + j]);
4571 else
4572 chunk_strcat(out, ".");
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004573 }
4574 chunk_strcat(out, "\n");
4575 }
4576}
4577
Willy Tarreau762fb3e2020-03-03 15:57:10 +01004578/* dumps <pfx> followed by <n> bytes from <addr> in hex form into buffer <buf>
4579 * enclosed in brackets after the address itself, formatted on 14 chars
4580 * including the "0x" prefix. This is meant to be used as a prefix for code
4581 * areas. For example:
4582 * "0x7f10b6557690 [48 c7 c0 0f 00 00 00 0f]"
4583 * It relies on may_access() to know if the bytes are dumpable, otherwise "--"
4584 * is emitted. A NULL <pfx> will be considered empty.
4585 */
4586void dump_addr_and_bytes(struct buffer *buf, const char *pfx, const void *addr, int n)
4587{
4588 int ok = 0;
4589 int i;
4590
4591 chunk_appendf(buf, "%s%#14lx [", pfx ? pfx : "", (long)addr);
4592
4593 for (i = 0; i < n; i++) {
4594 if (i == 0 || (((long)(addr + i) ^ (long)(addr)) & 4096))
4595 ok = may_access(addr + i);
4596 if (ok)
4597 chunk_appendf(buf, "%02x%s", ((uint8_t*)addr)[i], (i<n-1) ? " " : "]");
4598 else
4599 chunk_appendf(buf, "--%s", (i<n-1) ? " " : "]");
4600 }
4601}
4602
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004603/* print a line of text buffer (limited to 70 bytes) to <out>. The format is :
4604 * <2 spaces> <offset=5 digits> <space or plus> <space> <70 chars max> <\n>
4605 * which is 60 chars per line. Non-printable chars \t, \n, \r and \e are
4606 * encoded in C format. Other non-printable chars are encoded "\xHH". Original
4607 * lines are respected within the limit of 70 output chars. Lines that are
4608 * continuation of a previous truncated line begin with "+" instead of " "
4609 * after the offset. The new pointer is returned.
4610 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004611int dump_text_line(struct buffer *out, const char *buf, int bsize, int len,
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004612 int *line, int ptr)
4613{
4614 int end;
4615 unsigned char c;
4616
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004617 end = out->data + 80;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004618 if (end > out->size)
4619 return ptr;
4620
4621 chunk_appendf(out, " %05d%c ", ptr, (ptr == *line) ? ' ' : '+');
4622
4623 while (ptr < len && ptr < bsize) {
4624 c = buf[ptr];
Willy Tarreau90807112020-02-25 08:16:33 +01004625 if (isprint((unsigned char)c) && isascii((unsigned char)c) && c != '\\') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004626 if (out->data > end - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004627 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004628 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004629 } else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004630 if (out->data > end - 3)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004631 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004632 out->area[out->data++] = '\\';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004633 switch (c) {
4634 case '\t': c = 't'; break;
4635 case '\n': c = 'n'; break;
4636 case '\r': c = 'r'; break;
4637 case '\e': c = 'e'; break;
4638 case '\\': c = '\\'; break;
4639 }
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004640 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004641 } else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004642 if (out->data > end - 5)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004643 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004644 out->area[out->data++] = '\\';
4645 out->area[out->data++] = 'x';
4646 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4647 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004648 }
4649 if (buf[ptr++] == '\n') {
4650 /* we had a line break, let's return now */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004651 out->area[out->data++] = '\n';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004652 *line = ptr;
4653 return ptr;
4654 }
4655 }
4656 /* we have an incomplete line, we return it as-is */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004657 out->area[out->data++] = '\n';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004658 return ptr;
4659}
4660
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004661/* displays a <len> long memory block at <buf>, assuming first byte of <buf>
Willy Tarreaued936c52017-04-27 18:03:20 +02004662 * has address <baseaddr>. String <pfx> may be placed as a prefix in front of
4663 * each line. It may be NULL if unused. The output is emitted to file <out>.
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004664 */
Willy Tarreaued936c52017-04-27 18:03:20 +02004665void debug_hexdump(FILE *out, const char *pfx, const char *buf,
4666 unsigned int baseaddr, int len)
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004667{
Willy Tarreau73459792017-04-11 07:58:08 +02004668 unsigned int i;
4669 int b, j;
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004670
4671 for (i = 0; i < (len + (baseaddr & 15)); i += 16) {
4672 b = i - (baseaddr & 15);
Willy Tarreaued936c52017-04-27 18:03:20 +02004673 fprintf(out, "%s%08x: ", pfx ? pfx : "", i + (baseaddr & ~15));
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004674 for (j = 0; j < 8; j++) {
4675 if (b + j >= 0 && b + j < len)
4676 fprintf(out, "%02x ", (unsigned char)buf[b + j]);
4677 else
4678 fprintf(out, " ");
4679 }
4680
4681 if (b + j >= 0 && b + j < len)
4682 fputc('-', out);
4683 else
4684 fputc(' ', out);
4685
4686 for (j = 8; j < 16; j++) {
4687 if (b + j >= 0 && b + j < len)
4688 fprintf(out, " %02x", (unsigned char)buf[b + j]);
4689 else
4690 fprintf(out, " ");
4691 }
4692
4693 fprintf(out, " ");
4694 for (j = 0; j < 16; j++) {
4695 if (b + j >= 0 && b + j < len) {
4696 if (isprint((unsigned char)buf[b + j]))
4697 fputc((unsigned char)buf[b + j], out);
4698 else
4699 fputc('.', out);
4700 }
4701 else
4702 fputc(' ', out);
4703 }
4704 fputc('\n', out);
4705 }
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004706}
4707
Willy Tarreaubb869862020-04-16 10:52:41 +02004708/* Tries to report the executable path name on platforms supporting this. If
4709 * not found or not possible, returns NULL.
4710 */
4711const char *get_exec_path()
4712{
4713 const char *ret = NULL;
4714
4715#if (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 16))
4716 long execfn = getauxval(AT_EXECFN);
4717
4718 if (execfn && execfn != ENOENT)
4719 ret = (const char *)execfn;
4720#endif
4721 return ret;
4722}
4723
Baruch Siache1651b22020-07-24 07:52:20 +03004724#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreau9133e482020-03-04 10:19:36 +01004725/* calls dladdr() or dladdr1() on <addr> and <dli>. If dladdr1 is available,
4726 * also returns the symbol size in <size>, otherwise returns 0 there.
4727 */
4728static int dladdr_and_size(const void *addr, Dl_info *dli, size_t *size)
4729{
4730 int ret;
Willy Tarreau62af9c82020-03-10 07:51:48 +01004731#if (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 3)) // most detailed one
Willy Tarreau9133e482020-03-04 10:19:36 +01004732 const ElfW(Sym) *sym;
4733
4734 ret = dladdr1(addr, dli, (void **)&sym, RTLD_DL_SYMENT);
4735 if (ret)
4736 *size = sym ? sym->st_size : 0;
4737#else
4738 ret = dladdr(addr, dli);
4739 *size = 0;
4740#endif
4741 return ret;
4742}
Willy Tarreau64192392021-05-05 09:06:21 +02004743
4744/* Tries to retrieve the address of the first occurrence symbol <name>.
4745 * Note that NULL in return is not always an error as a symbol may have that
4746 * address in special situations.
4747 */
4748void *get_sym_curr_addr(const char *name)
4749{
4750 void *ptr = NULL;
4751
4752#ifdef RTLD_DEFAULT
4753 ptr = dlsym(RTLD_DEFAULT, name);
4754#endif
4755 return ptr;
4756}
4757
4758
4759/* Tries to retrieve the address of the next occurrence of symbol <name>
4760 * Note that NULL in return is not always an error as a symbol may have that
4761 * address in special situations.
4762 */
4763void *get_sym_next_addr(const char *name)
4764{
4765 void *ptr = NULL;
4766
4767#ifdef RTLD_NEXT
4768 ptr = dlsym(RTLD_NEXT, name);
Willy Tarreau9133e482020-03-04 10:19:36 +01004769#endif
Willy Tarreau64192392021-05-05 09:06:21 +02004770 return ptr;
4771}
4772
4773#else /* elf & linux & dl */
4774
4775/* no possible resolving on other platforms at the moment */
4776void *get_sym_curr_addr(const char *name)
4777{
4778 return NULL;
4779}
4780
4781void *get_sym_next_addr(const char *name)
4782{
4783 return NULL;
4784}
4785
4786#endif /* elf & linux & dl */
Willy Tarreau9133e482020-03-04 10:19:36 +01004787
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004788/* Tries to append to buffer <buf> some indications about the symbol at address
4789 * <addr> using the following form:
4790 * lib:+0xoffset (unresolvable address from lib's base)
4791 * main+0xoffset (unresolvable address from main (+/-))
4792 * lib:main+0xoffset (unresolvable lib address from main (+/-))
4793 * name (resolved exact exec address)
4794 * lib:name (resolved exact lib address)
4795 * name+0xoffset/0xsize (resolved address within exec symbol)
4796 * lib:name+0xoffset/0xsize (resolved address within lib symbol)
4797 *
4798 * The file name (lib or executable) is limited to what lies between the last
4799 * '/' and the first following '.'. An optional prefix <pfx> is prepended before
4800 * 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 +03004801 * that contains the "main" symbol, or when __ELF__ && USE_DL are not set.
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004802 *
4803 * The symbol's base address is returned, or NULL when unresolved, in order to
4804 * allow the caller to match it against known ones.
4805 */
Willy Tarreau45fd1032021-01-20 14:37:59 +01004806const void *resolve_sym_name(struct buffer *buf, const char *pfx, const void *addr)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004807{
4808 const struct {
4809 const void *func;
4810 const char *name;
4811 } fcts[] = {
4812 { .func = process_stream, .name = "process_stream" },
4813 { .func = task_run_applet, .name = "task_run_applet" },
4814 { .func = si_cs_io_cb, .name = "si_cs_io_cb" },
Willy Tarreau586f71b2020-12-11 15:54:36 +01004815 { .func = sock_conn_iocb, .name = "sock_conn_iocb" },
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004816 { .func = dgram_fd_handler, .name = "dgram_fd_handler" },
4817 { .func = listener_accept, .name = "listener_accept" },
Willy Tarreaud597ec22021-01-29 14:29:06 +01004818 { .func = manage_global_listener_queue, .name = "manage_global_listener_queue" },
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004819 { .func = poller_pipe_io_handler, .name = "poller_pipe_io_handler" },
4820 { .func = mworker_accept_wrapper, .name = "mworker_accept_wrapper" },
Willy Tarreau02922e12021-01-29 12:27:57 +01004821 { .func = session_expire_embryonic, .name = "session_expire_embryonic" },
Willy Tarreaufb5401f2021-01-29 12:25:23 +01004822#ifdef USE_THREAD
4823 { .func = accept_queue_process, .name = "accept_queue_process" },
4824#endif
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004825#ifdef USE_LUA
4826 { .func = hlua_process_task, .name = "hlua_process_task" },
4827#endif
Ilya Shipitsinbdec3ba2020-11-14 01:56:34 +05004828#ifdef SSL_MODE_ASYNC
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004829 { .func = ssl_async_fd_free, .name = "ssl_async_fd_free" },
4830 { .func = ssl_async_fd_handler, .name = "ssl_async_fd_handler" },
4831#endif
4832 };
4833
Baruch Siache1651b22020-07-24 07:52:20 +03004834#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004835 Dl_info dli, dli_main;
Willy Tarreau9133e482020-03-04 10:19:36 +01004836 size_t size;
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004837 const char *fname, *p;
4838#endif
4839 int i;
4840
4841 if (pfx)
4842 chunk_appendf(buf, "%s", pfx);
4843
4844 for (i = 0; i < sizeof(fcts) / sizeof(fcts[0]); i++) {
4845 if (addr == fcts[i].func) {
4846 chunk_appendf(buf, "%s", fcts[i].name);
4847 return addr;
4848 }
4849 }
4850
Baruch Siache1651b22020-07-24 07:52:20 +03004851#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004852 /* Now let's try to be smarter */
Willy Tarreau9133e482020-03-04 10:19:36 +01004853 if (!dladdr_and_size(addr, &dli, &size))
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004854 goto unknown;
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004855
4856 /* 1. prefix the library name if it's not the same object as the one
4857 * that contains the main function. The name is picked between last '/'
4858 * and first following '.'.
4859 */
4860 if (!dladdr(main, &dli_main))
4861 dli_main.dli_fbase = NULL;
4862
4863 if (dli_main.dli_fbase != dli.dli_fbase) {
4864 fname = dli.dli_fname;
4865 p = strrchr(fname, '/');
4866 if (p++)
4867 fname = p;
4868 p = strchr(fname, '.');
4869 if (!p)
4870 p = fname + strlen(fname);
4871
4872 chunk_appendf(buf, "%.*s:", (int)(long)(p - fname), fname);
4873 }
4874
4875 /* 2. symbol name */
4876 if (dli.dli_sname) {
4877 /* known, dump it and return symbol's address (exact or relative) */
4878 chunk_appendf(buf, "%s", dli.dli_sname);
4879 if (addr != dli.dli_saddr) {
4880 chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_saddr));
Willy Tarreau9133e482020-03-04 10:19:36 +01004881 if (size)
4882 chunk_appendf(buf, "/%#lx", (long)size);
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004883 }
4884 return dli.dli_saddr;
4885 }
4886 else if (dli_main.dli_fbase != dli.dli_fbase) {
4887 /* unresolved symbol from a known library, report relative offset */
4888 chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_fbase));
4889 return NULL;
4890 }
Baruch Siache1651b22020-07-24 07:52:20 +03004891#endif /* __ELF__ && !__linux__ || USE_DL */
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004892 unknown:
4893 /* unresolved symbol from the main file, report relative offset to main */
4894 if ((void*)addr < (void*)main)
4895 chunk_appendf(buf, "main-%#lx", (long)((void*)main - addr));
4896 else
4897 chunk_appendf(buf, "main+%#lx", (long)(addr - (void*)main));
4898 return NULL;
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004899}
4900
Frédéric Lécaille3b717162019-02-25 15:04:22 +01004901/*
4902 * Allocate an array of unsigned int with <nums> as address from <str> string
Ilya Shipitsin46a030c2020-07-05 16:36:08 +05004903 * made of integer separated by dot characters.
Frédéric Lécaille3b717162019-02-25 15:04:22 +01004904 *
4905 * First, initializes the value with <sz> as address to 0 and initializes the
4906 * array with <nums> as address to NULL. Then allocates the array with <nums> as
4907 * address updating <sz> pointed value to the size of this array.
4908 *
4909 * Returns 1 if succeeded, 0 if not.
4910 */
4911int parse_dotted_uints(const char *str, unsigned int **nums, size_t *sz)
4912{
4913 unsigned int *n;
4914 const char *s, *end;
4915
4916 s = str;
4917 *sz = 0;
4918 end = str + strlen(str);
4919 *nums = n = NULL;
4920
4921 while (1) {
4922 unsigned int r;
4923
4924 if (s >= end)
4925 break;
4926
4927 r = read_uint(&s, end);
4928 /* Expected characters after having read an uint: '\0' or '.',
4929 * if '.', must not be terminal.
4930 */
Christopher Faulet4b524122021-02-11 10:42:41 +01004931 if (*s != '\0'&& (*s++ != '.' || s == end)) {
4932 free(n);
Frédéric Lécaille3b717162019-02-25 15:04:22 +01004933 return 0;
Christopher Faulet4b524122021-02-11 10:42:41 +01004934 }
Frédéric Lécaille3b717162019-02-25 15:04:22 +01004935
Frédéric Lécaille12a71842019-02-26 18:19:48 +01004936 n = my_realloc2(n, (*sz + 1) * sizeof *n);
Frédéric Lécaille3b717162019-02-25 15:04:22 +01004937 if (!n)
4938 return 0;
4939
4940 n[(*sz)++] = r;
4941 }
4942 *nums = n;
4943
4944 return 1;
4945}
4946
Willy Tarreau4d589e72019-08-23 19:02:26 +02004947
4948/* returns the number of bytes needed to encode <v> as a varint. An inline
4949 * version exists for use with constants (__varint_bytes()).
4950 */
4951int varint_bytes(uint64_t v)
4952{
4953 int len = 1;
4954
4955 if (v >= 240) {
4956 v = (v - 240) >> 4;
4957 while (1) {
4958 len++;
4959 if (v < 128)
4960 break;
4961 v = (v - 128) >> 7;
4962 }
4963 }
4964 return len;
4965}
4966
Willy Tarreau52bf8392020-03-08 00:42:37 +01004967
4968/* Random number generator state, see below */
Willy Tarreau1544c142020-03-12 00:31:18 +01004969static uint64_t ha_random_state[2] ALIGNED(2*sizeof(uint64_t));
Willy Tarreau52bf8392020-03-08 00:42:37 +01004970
4971/* This is a thread-safe implementation of xoroshiro128** described below:
4972 * http://prng.di.unimi.it/
4973 * It features a 2^128 long sequence, returns 64 high-quality bits on each call,
4974 * supports fast jumps and passes all common quality tests. It is thread-safe,
4975 * uses a double-cas on 64-bit architectures supporting it, and falls back to a
4976 * local lock on other ones.
4977 */
4978uint64_t ha_random64()
4979{
4980 uint64_t result;
Willy Tarreau1544c142020-03-12 00:31:18 +01004981 uint64_t old[2] ALIGNED(2*sizeof(uint64_t));
4982 uint64_t new[2] ALIGNED(2*sizeof(uint64_t));
Willy Tarreau52bf8392020-03-08 00:42:37 +01004983
4984#if defined(USE_THREAD) && (!defined(HA_CAS_IS_8B) || !defined(HA_HAVE_CAS_DW))
4985 static HA_SPINLOCK_T rand_lock;
4986
4987 HA_SPIN_LOCK(OTHER_LOCK, &rand_lock);
4988#endif
4989
4990 old[0] = ha_random_state[0];
4991 old[1] = ha_random_state[1];
4992
4993#if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
4994 do {
4995#endif
4996 result = rotl64(old[0] * 5, 7) * 9;
4997 new[1] = old[0] ^ old[1];
4998 new[0] = rotl64(old[0], 24) ^ new[1] ^ (new[1] << 16); // a, b
4999 new[1] = rotl64(new[1], 37); // c
5000
5001#if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
5002 } while (unlikely(!_HA_ATOMIC_DWCAS(ha_random_state, old, new)));
5003#else
5004 ha_random_state[0] = new[0];
5005 ha_random_state[1] = new[1];
5006#if defined(USE_THREAD)
5007 HA_SPIN_UNLOCK(OTHER_LOCK, &rand_lock);
5008#endif
5009#endif
5010 return result;
5011}
5012
5013/* seeds the random state using up to <len> bytes from <seed>, starting with
5014 * the first non-zero byte.
5015 */
5016void ha_random_seed(const unsigned char *seed, size_t len)
5017{
5018 size_t pos;
5019
5020 /* the seed must not be all zeroes, so we pre-fill it with alternating
5021 * bits and overwrite part of them with the block starting at the first
5022 * non-zero byte from the seed.
5023 */
5024 memset(ha_random_state, 0x55, sizeof(ha_random_state));
5025
5026 for (pos = 0; pos < len; pos++)
5027 if (seed[pos] != 0)
5028 break;
5029
5030 if (pos == len)
5031 return;
5032
5033 seed += pos;
5034 len -= pos;
5035
5036 if (len > sizeof(ha_random_state))
5037 len = sizeof(ha_random_state);
5038
5039 memcpy(ha_random_state, seed, len);
5040}
5041
5042/* This causes a jump to (dist * 2^96) places in the pseudo-random sequence,
5043 * and is equivalent to calling ha_random64() as many times. It is used to
5044 * provide non-overlapping sequences of 2^96 numbers (~7*10^28) to up to 2^32
5045 * different generators (i.e. different processes after a fork). The <dist>
5046 * argument is the distance to jump to and is used in a loop so it rather not
5047 * be too large if the processing time is a concern.
5048 *
5049 * BEWARE: this function is NOT thread-safe and must not be called during
5050 * concurrent accesses to ha_random64().
5051 */
5052void ha_random_jump96(uint32_t dist)
5053{
5054 while (dist--) {
5055 uint64_t s0 = 0;
5056 uint64_t s1 = 0;
5057 int b;
5058
5059 for (b = 0; b < 64; b++) {
5060 if ((0xd2a98b26625eee7bULL >> b) & 1) {
5061 s0 ^= ha_random_state[0];
5062 s1 ^= ha_random_state[1];
5063 }
5064 ha_random64();
5065 }
5066
5067 for (b = 0; b < 64; b++) {
5068 if ((0xdddf9b1090aa7ac1ULL >> b) & 1) {
5069 s0 ^= ha_random_state[0];
5070 s1 ^= ha_random_state[1];
5071 }
5072 ha_random64();
5073 }
5074 ha_random_state[0] = s0;
5075 ha_random_state[1] = s1;
5076 }
5077}
5078
Willy Tarreauee3bcdd2020-03-08 17:48:17 +01005079/* Generates an RFC4122 UUID into chunk <output> which must be at least 37
5080 * bytes large.
5081 */
5082void ha_generate_uuid(struct buffer *output)
5083{
5084 uint32_t rnd[4];
5085 uint64_t last;
5086
5087 last = ha_random64();
5088 rnd[0] = last;
5089 rnd[1] = last >> 32;
5090
5091 last = ha_random64();
5092 rnd[2] = last;
5093 rnd[3] = last >> 32;
5094
5095 chunk_printf(output, "%8.8x-%4.4x-%4.4x-%4.4x-%12.12llx",
5096 rnd[0],
5097 rnd[1] & 0xFFFF,
5098 ((rnd[1] >> 16u) & 0xFFF) | 0x4000, // highest 4 bits indicate the uuid version
5099 (rnd[2] & 0x3FFF) | 0x8000, // the highest 2 bits indicate the UUID variant (10),
5100 (long long)((rnd[2] >> 14u) | ((uint64_t) rnd[3] << 18u)) & 0xFFFFFFFFFFFFull);
5101}
5102
5103
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005104/* only used by parse_line() below. It supports writing in place provided that
5105 * <in> is updated to the next location before calling it. In that case, the
5106 * char at <in> may be overwritten.
5107 */
5108#define EMIT_CHAR(x) \
5109 do { \
5110 char __c = (char)(x); \
5111 if ((opts & PARSE_OPT_INPLACE) && out+outpos > in) \
5112 err |= PARSE_ERR_OVERLAP; \
5113 if (outpos >= outmax) \
5114 err |= PARSE_ERR_TOOLARGE; \
5115 if (!err) \
5116 out[outpos] = __c; \
5117 outpos++; \
5118 } while (0)
5119
Ilya Shipitsin46a030c2020-07-05 16:36:08 +05005120/* Parse <in>, copy it into <out> split into isolated words whose pointers
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005121 * are put in <args>. If more than <outlen> bytes have to be emitted, the
5122 * extraneous ones are not emitted but <outlen> is updated so that the caller
5123 * knows how much to realloc. Similarly, <args> are not updated beyond <nbargs>
5124 * but the returned <nbargs> indicates how many were found. All trailing args
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005125 * up to <nbargs> point to the trailing zero, and as long as <nbargs> is > 0,
5126 * it is guaranteed that at least one arg will point to the zero. It is safe
5127 * to call it with a NULL <args> if <nbargs> is 0.
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005128 *
5129 * <out> may overlap with <in> provided that it never goes further, in which
5130 * case the parser will accept to perform in-place parsing and unquoting/
5131 * unescaping but only if environment variables do not lead to expansion that
5132 * causes overlapping, otherwise the input string being destroyed, the error
5133 * will not be recoverable. Note that even during out-of-place <in> will
5134 * experience temporary modifications in-place for variable resolution and must
5135 * be writable, and will also receive zeroes to delimit words when using
5136 * in-place copy. Parsing options <opts> taken from PARSE_OPT_*. Return value
5137 * is zero on success otherwise a bitwise-or of PARSE_ERR_*. Upon error, the
5138 * starting point of the first invalid character sequence or unmatched
5139 * quote/brace is reported in <errptr> if not NULL. When using in-place parsing
5140 * error reporting might be difficult since zeroes will have been inserted into
5141 * the string. One solution for the caller may consist in replacing all args
5142 * delimiters with spaces in this case.
5143 */
5144uint32_t parse_line(char *in, char *out, size_t *outlen, char **args, int *nbargs, uint32_t opts, char **errptr)
5145{
5146 char *quote = NULL;
5147 char *brace = NULL;
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005148 char *word_expand = NULL;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005149 unsigned char hex1, hex2;
5150 size_t outmax = *outlen;
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005151 int argsmax = *nbargs - 1;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005152 size_t outpos = 0;
5153 int squote = 0;
5154 int dquote = 0;
5155 int arg = 0;
5156 uint32_t err = 0;
5157
5158 *nbargs = 0;
5159 *outlen = 0;
5160
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005161 /* argsmax may be -1 here, protecting args[] from any write */
5162 if (arg < argsmax)
5163 args[arg] = out;
5164
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005165 while (1) {
5166 if (*in >= '-' && *in != '\\') {
5167 /* speedup: directly send all regular chars starting
5168 * with '-', '.', '/', alnum etc...
5169 */
5170 EMIT_CHAR(*in++);
5171 continue;
5172 }
5173 else if (*in == '\0' || *in == '\n' || *in == '\r') {
5174 /* end of line */
5175 break;
5176 }
5177 else if (*in == '#' && (opts & PARSE_OPT_SHARP) && !squote && !dquote) {
5178 /* comment */
5179 break;
5180 }
5181 else if (*in == '"' && !squote && (opts & PARSE_OPT_DQUOTE)) { /* double quote outside single quotes */
5182 if (dquote) {
5183 dquote = 0;
5184 quote = NULL;
5185 }
5186 else {
5187 dquote = 1;
5188 quote = in;
5189 }
5190 in++;
5191 continue;
5192 }
5193 else if (*in == '\'' && !dquote && (opts & PARSE_OPT_SQUOTE)) { /* single quote outside double quotes */
5194 if (squote) {
5195 squote = 0;
5196 quote = NULL;
5197 }
5198 else {
5199 squote = 1;
5200 quote = in;
5201 }
5202 in++;
5203 continue;
5204 }
5205 else if (*in == '\\' && !squote && (opts & PARSE_OPT_BKSLASH)) {
5206 /* first, we'll replace \\, \<space>, \#, \r, \n, \t, \xXX with their
5207 * C equivalent value but only when they have a special meaning and within
5208 * double quotes for some of them. Other combinations left unchanged (eg: \1).
5209 */
5210 char tosend = *in;
5211
5212 switch (in[1]) {
5213 case ' ':
5214 case '\\':
5215 tosend = in[1];
5216 in++;
5217 break;
5218
5219 case 't':
5220 tosend = '\t';
5221 in++;
5222 break;
5223
5224 case 'n':
5225 tosend = '\n';
5226 in++;
5227 break;
5228
5229 case 'r':
5230 tosend = '\r';
5231 in++;
5232 break;
5233
5234 case '#':
5235 /* escaping of "#" only if comments are supported */
5236 if (opts & PARSE_OPT_SHARP)
5237 in++;
5238 tosend = *in;
5239 break;
5240
5241 case '\'':
5242 /* escaping of "'" only outside single quotes and only if single quotes are supported */
5243 if (opts & PARSE_OPT_SQUOTE && !squote)
5244 in++;
5245 tosend = *in;
5246 break;
5247
5248 case '"':
5249 /* escaping of '"' only outside single quotes and only if double quotes are supported */
5250 if (opts & PARSE_OPT_DQUOTE && !squote)
5251 in++;
5252 tosend = *in;
5253 break;
5254
5255 case '$':
5256 /* escaping of '$' only inside double quotes and only if env supported */
5257 if (opts & PARSE_OPT_ENV && dquote)
5258 in++;
5259 tosend = *in;
5260 break;
5261
5262 case 'x':
5263 if (!ishex(in[2]) || !ishex(in[3])) {
5264 /* invalid or incomplete hex sequence */
5265 err |= PARSE_ERR_HEX;
5266 if (errptr)
5267 *errptr = in;
5268 goto leave;
5269 }
Willy Tarreauf278eec2020-07-05 21:46:32 +02005270 hex1 = toupper((unsigned char)in[2]) - '0';
5271 hex2 = toupper((unsigned char)in[3]) - '0';
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005272 if (hex1 > 9) hex1 -= 'A' - '9' - 1;
5273 if (hex2 > 9) hex2 -= 'A' - '9' - 1;
5274 tosend = (hex1 << 4) + hex2;
5275 in += 3;
5276 break;
5277
5278 default:
5279 /* other combinations are not escape sequences */
5280 break;
5281 }
5282
5283 in++;
5284 EMIT_CHAR(tosend);
5285 }
5286 else if (isspace((unsigned char)*in) && !squote && !dquote) {
5287 /* a non-escaped space is an argument separator */
5288 while (isspace((unsigned char)*in))
5289 in++;
5290 EMIT_CHAR(0);
5291 arg++;
5292 if (arg < argsmax)
5293 args[arg] = out + outpos;
5294 else
5295 err |= PARSE_ERR_TOOMANY;
5296 }
5297 else if (*in == '$' && (opts & PARSE_OPT_ENV) && (dquote || !(opts & PARSE_OPT_DQUOTE))) {
5298 /* environment variables are evaluated anywhere, or only
5299 * inside double quotes if they are supported.
5300 */
5301 char *var_name;
5302 char save_char;
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005303 const char *value;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005304
5305 in++;
5306
5307 if (*in == '{')
5308 brace = in++;
5309
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005310 if (!isalpha((unsigned char)*in) && *in != '_' && *in != '.') {
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005311 /* unacceptable character in variable name */
5312 err |= PARSE_ERR_VARNAME;
5313 if (errptr)
5314 *errptr = in;
5315 goto leave;
5316 }
5317
5318 var_name = in;
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005319 if (*in == '.')
5320 in++;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005321 while (isalnum((unsigned char)*in) || *in == '_')
5322 in++;
5323
5324 save_char = *in;
5325 *in = '\0';
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005326 if (unlikely(*var_name == '.')) {
5327 /* internal pseudo-variables */
5328 if (strcmp(var_name, ".LINE") == 0)
5329 value = ultoa(global.cfg_curr_line);
5330 else if (strcmp(var_name, ".FILE") == 0)
5331 value = global.cfg_curr_file;
5332 else if (strcmp(var_name, ".SECTION") == 0)
5333 value = global.cfg_curr_section;
5334 else {
5335 /* unsupported internal variable name */
5336 err |= PARSE_ERR_VARNAME;
5337 if (errptr)
5338 *errptr = var_name;
5339 goto leave;
5340 }
5341 } else {
5342 value = getenv(var_name);
5343 }
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005344 *in = save_char;
5345
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005346 /* support for '[*]' sequence to force word expansion,
5347 * only available inside braces */
5348 if (*in == '[' && brace && (opts & PARSE_OPT_WORD_EXPAND)) {
5349 word_expand = in++;
5350
5351 if (*in++ != '*' || *in++ != ']') {
5352 err |= PARSE_ERR_WRONG_EXPAND;
5353 if (errptr)
5354 *errptr = word_expand;
5355 goto leave;
5356 }
5357 }
5358
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005359 if (brace) {
5360 if (*in != '}') {
5361 /* unmatched brace */
5362 err |= PARSE_ERR_BRACE;
5363 if (errptr)
5364 *errptr = brace;
5365 goto leave;
5366 }
5367 in++;
5368 brace = NULL;
5369 }
5370
5371 if (value) {
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005372 while (*value) {
5373 /* expand as individual parameters on a space character */
Willy Tarreaufe2cc412020-10-01 18:04:40 +02005374 if (word_expand && isspace((unsigned char)*value)) {
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005375 EMIT_CHAR(0);
5376 ++arg;
5377 if (arg < argsmax)
5378 args[arg] = out + outpos;
5379 else
5380 err |= PARSE_ERR_TOOMANY;
5381
5382 /* skip consecutive spaces */
Willy Tarreaufe2cc412020-10-01 18:04:40 +02005383 while (isspace((unsigned char)*++value))
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005384 ;
5385 } else {
5386 EMIT_CHAR(*value++);
5387 }
5388 }
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005389 }
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005390 word_expand = NULL;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005391 }
5392 else {
5393 /* any other regular char */
5394 EMIT_CHAR(*in++);
5395 }
5396 }
5397
5398 /* end of output string */
5399 EMIT_CHAR(0);
5400 arg++;
5401
5402 if (quote) {
5403 /* unmatched quote */
5404 err |= PARSE_ERR_QUOTE;
5405 if (errptr)
5406 *errptr = quote;
5407 goto leave;
5408 }
5409 leave:
5410 *nbargs = arg;
5411 *outlen = outpos;
5412
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005413 /* empty all trailing args by making them point to the trailing zero,
5414 * at least the last one in any case.
5415 */
5416 if (arg > argsmax)
5417 arg = argsmax;
5418
5419 while (arg >= 0 && arg <= argsmax)
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005420 args[arg++] = out + outpos - 1;
5421
5422 return err;
5423}
5424#undef EMIT_CHAR
5425
Willy Tarreauc54e5ad2020-06-25 09:15:40 +02005426/* This is used to sanitize an input line that's about to be used for error reporting.
5427 * It will adjust <line> to print approximately <width> chars around <pos>, trying to
5428 * preserve the beginning, with leading or trailing "..." when the line is truncated.
5429 * If non-printable chars are present in the output. It returns the new offset <pos>
5430 * in the modified line. Non-printable characters are replaced with '?'. <width> must
5431 * be at least 6 to support two "..." otherwise the result is undefined. The line
5432 * itself must have at least 7 chars allocated for the same reason.
5433 */
5434size_t sanitize_for_printing(char *line, size_t pos, size_t width)
5435{
5436 size_t shift = 0;
5437 char *out = line;
5438 char *in = line;
5439 char *end = line + width;
5440
5441 if (pos >= width) {
5442 /* if we have to shift, we'll be out of context, so let's
5443 * try to put <pos> at the center of width.
5444 */
5445 shift = pos - width / 2;
5446 in += shift + 3;
5447 end = out + width - 3;
5448 out[0] = out[1] = out[2] = '.';
5449 out += 3;
5450 }
5451
5452 while (out < end && *in) {
5453 if (isspace((unsigned char)*in))
5454 *out++ = ' ';
5455 else if (isprint((unsigned char)*in))
5456 *out++ = *in;
5457 else
5458 *out++ = '?';
5459 in++;
5460 }
5461
5462 if (end < line + width) {
5463 out[0] = out[1] = out[2] = '.';
5464 out += 3;
5465 }
5466
5467 *out++ = 0;
5468 return pos - shift;
5469}
Willy Tarreau06e69b52021-03-02 14:01:35 +01005470
Willy Tarreaue33c4b32021-03-12 18:59:31 +01005471/* Update array <fp> with the fingerprint of word <word> by counting the
Willy Tarreauba2c4452021-03-12 09:01:52 +01005472 * transitions between characters. <fp> is a 1024-entries array indexed as
5473 * 32*from+to. Positions for 'from' and 'to' are:
Willy Tarreau9294e882021-03-15 09:34:27 +01005474 * 1..26=letter, 27=digit, 28=other/begin/end.
5475 * Row "from=0" is used to mark the character's presence. Others unused.
Willy Tarreauba2c4452021-03-12 09:01:52 +01005476 */
Willy Tarreaue33c4b32021-03-12 18:59:31 +01005477void update_word_fingerprint(uint8_t *fp, const char *word)
Willy Tarreauba2c4452021-03-12 09:01:52 +01005478{
5479 const char *p;
5480 int from, to;
5481 int c;
5482
Willy Tarreauba2c4452021-03-12 09:01:52 +01005483 from = 28; // begin
5484 for (p = word; *p; p++) {
5485 c = tolower(*p);
5486 switch(c) {
Willy Tarreau9294e882021-03-15 09:34:27 +01005487 case 'a'...'z': to = c - 'a' + 1; break;
5488 case 'A'...'Z': to = tolower(c) - 'a' + 1; break;
5489 case '0'...'9': to = 27; break;
5490 default: to = 28; break;
Willy Tarreauba2c4452021-03-12 09:01:52 +01005491 }
Willy Tarreau9294e882021-03-15 09:34:27 +01005492 fp[to] = 1;
Willy Tarreauba2c4452021-03-12 09:01:52 +01005493 fp[32 * from + to]++;
5494 from = to;
5495 }
5496 to = 28; // end
5497 fp[32 * from + to]++;
5498}
5499
Willy Tarreaue33c4b32021-03-12 18:59:31 +01005500/* Initialize array <fp> with the fingerprint of word <word> by counting the
5501 * transitions between characters. <fp> is a 1024-entries array indexed as
5502 * 32*from+to. Positions for 'from' and 'to' are:
5503 * 0..25=letter, 26=digit, 27=other, 28=begin, 29=end, others unused.
5504 */
5505void make_word_fingerprint(uint8_t *fp, const char *word)
5506{
5507 memset(fp, 0, 1024);
5508 update_word_fingerprint(fp, word);
5509}
5510
Willy Tarreauba2c4452021-03-12 09:01:52 +01005511/* Return the distance between two word fingerprints created by function
5512 * make_word_fingerprint(). It's a positive integer calculated as the sum of
Willy Tarreau714c4c12021-03-15 09:44:53 +01005513 * the differences between each location.
Willy Tarreauba2c4452021-03-12 09:01:52 +01005514 */
5515int word_fingerprint_distance(const uint8_t *fp1, const uint8_t *fp2)
5516{
5517 int i, k, dist = 0;
5518
5519 for (i = 0; i < 1024; i++) {
5520 k = (int)fp1[i] - (int)fp2[i];
Willy Tarreau714c4c12021-03-15 09:44:53 +01005521 dist += abs(k);
Willy Tarreauba2c4452021-03-12 09:01:52 +01005522 }
5523 return dist;
5524}
5525
Willy Tarreau06e69b52021-03-02 14:01:35 +01005526static int init_tools_per_thread()
5527{
5528 /* Let's make each thread start from a different position */
5529 statistical_prng_state += tid * MAX_THREADS;
5530 if (!statistical_prng_state)
5531 statistical_prng_state++;
5532 return 1;
5533}
5534REGISTER_PER_THREAD_INIT(init_tools_per_thread);
Willy Tarreauc54e5ad2020-06-25 09:15:40 +02005535
Willy Tarreaubaaee002006-06-26 02:48:02 +02005536/*
5537 * Local variables:
5538 * c-indent-level: 8
5539 * c-basic-offset: 8
5540 * End:
5541 */