blob: 746e0d074d20a13f8e0d88aa08c33ed7de71956d [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
devnexen@gmail.com49a32282021-08-17 12:55:49 +010019#if defined(__FreeBSD__)
20#include <elf.h>
21#include <dlfcn.h>
22extern void *__elf_aux_vector;
23#endif
24
David Carlier1b9d57d2021-08-17 08:44:25 +010025#if defined(__NetBSD__)
26#include <sys/exec_elf.h>
27#include <dlfcn.h>
28#endif
29
Willy Tarreau2e74c3f2007-12-02 18:45:09 +010030#include <ctype.h>
Willy Tarreau16e01562016-08-09 16:46:18 +020031#include <errno.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020032#include <netdb.h>
Willy Tarreau9a7bea52012-04-27 11:16:50 +020033#include <stdarg.h>
Willy Tarreaudd2f85e2012-09-02 22:34:23 +020034#include <stdio.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020035#include <stdlib.h>
36#include <string.h>
Thierry Fournier93127942016-01-20 18:49:45 +010037#include <time.h>
Willy Tarreau16e01562016-08-09 16:46:18 +020038#include <unistd.h>
Willy Tarreau127f9662007-12-06 00:53:51 +010039#include <sys/socket.h>
Willy Tarreau37101052019-05-20 16:48:20 +020040#include <sys/stat.h>
41#include <sys/types.h>
Willy Tarreau127f9662007-12-06 00:53:51 +010042#include <sys/un.h>
Willy Tarreaubaaee002006-06-26 02:48:02 +020043#include <netinet/in.h>
44#include <arpa/inet.h>
45
Willy Tarreau1ee71dd2021-08-30 10:15:35 +020046#if defined(__GLIBC__) && (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 16))
Willy Tarreau30053062020-08-20 16:39:14 +020047#include <sys/auxv.h>
48#endif
49
Willy Tarreau48fbcae2020-06-03 18:09:46 +020050#include <import/eb32sctree.h>
Willy Tarreaub2551052020-06-09 09:07:15 +020051#include <import/eb32tree.h>
Willy Tarreau48fbcae2020-06-03 18:09:46 +020052
Willy Tarreau4c7e4b72020-05-27 12:58:42 +020053#include <haproxy/api.h>
Willy Tarreauc13ed532020-06-02 10:22:45 +020054#include <haproxy/chunk.h>
Willy Tarreau7c18b542020-06-11 09:23:02 +020055#include <haproxy/dgram.h>
Willy Tarreauf268ee82020-06-04 17:05:57 +020056#include <haproxy/global.h>
Willy Tarreau86416052020-06-04 09:20:54 +020057#include <haproxy/hlua.h>
Willy Tarreau213e9902020-06-04 14:58:24 +020058#include <haproxy/listener.h>
Willy Tarreau7a00efb2020-06-02 17:02:59 +020059#include <haproxy/namespace.h>
Christopher Faulet9553de72021-02-26 09:12:50 +010060#include <haproxy/net_helper.h>
Willy Tarreau5fc93282020-09-16 18:25:03 +020061#include <haproxy/protocol.h>
Emeric Brunc9437992021-02-12 19:42:55 +010062#include <haproxy/resolvers.h>
Willy Tarreau586f71b2020-12-11 15:54:36 +010063#include <haproxy/sock.h>
Willy Tarreau209108d2020-06-04 20:30:20 +020064#include <haproxy/ssl_sock.h>
Willy Tarreau5e539c92020-06-04 20:45:39 +020065#include <haproxy/stream_interface.h>
Willy Tarreaucea0e1b2020-06-04 17:25:40 +020066#include <haproxy/task.h>
Willy Tarreau48fbcae2020-06-03 18:09:46 +020067#include <haproxy/tools.h>
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +010068
Thierry Fournier93127942016-01-20 18:49:45 +010069/* This macro returns false if the test __x is false. Many
70 * of the following parsing function must be abort the processing
71 * if it returns 0, so this macro is useful for writing light code.
72 */
73#define RET0_UNLESS(__x) do { if (!(__x)) return 0; } while (0)
74
Willy Tarreau56adcf22012-12-23 18:00:29 +010075/* enough to store NB_ITOA_STR integers of :
Willy Tarreau72d759c2007-10-25 12:14:10 +020076 * 2^64-1 = 18446744073709551615 or
77 * -2^63 = -9223372036854775808
Willy Tarreaue7239b52009-03-29 13:41:58 +020078 *
79 * The HTML version needs room for adding the 25 characters
80 * '<span class="rls"></span>' around digits at positions 3N+1 in order
81 * to add spacing at up to 6 positions : 18 446 744 073 709 551 615
Willy Tarreau72d759c2007-10-25 12:14:10 +020082 */
Christopher Faulet99bca652017-11-14 16:47:26 +010083THREAD_LOCAL char itoa_str[NB_ITOA_STR][171];
84THREAD_LOCAL int itoa_idx = 0; /* index of next itoa_str to use */
Willy Tarreaubaaee002006-06-26 02:48:02 +020085
Willy Tarreau588297f2014-06-16 15:16:40 +020086/* sometimes we'll need to quote strings (eg: in stats), and we don't expect
87 * to quote strings larger than a max configuration line.
88 */
Christopher Faulet99bca652017-11-14 16:47:26 +010089THREAD_LOCAL char quoted_str[NB_QSTR][QSTR_SIZE + 1];
90THREAD_LOCAL int quoted_idx = 0;
Willy Tarreau588297f2014-06-16 15:16:40 +020091
Willy Tarreau06e69b52021-03-02 14:01:35 +010092/* thread-local PRNG state. It's modified to start from a different sequence
93 * on all threads upon startup. It must not be used or anything beyond getting
94 * statistical values as it's 100% predictable.
95 */
96THREAD_LOCAL unsigned int statistical_prng_state = 2463534242U;
97
Willy Tarreau263aacd2022-07-18 13:58:17 +020098/* set to true if this is a static build */
99int build_is_static = 0;
100
Willy Tarreaubaaee002006-06-26 02:48:02 +0200101/*
William Lallemande7340ec2012-01-24 11:15:39 +0100102 * unsigned long long ASCII representation
103 *
104 * return the last char '\0' or NULL if no enough
105 * space in dst
106 */
107char *ulltoa(unsigned long long n, char *dst, size_t size)
108{
109 int i = 0;
110 char *res;
111
112 switch(n) {
113 case 1ULL ... 9ULL:
114 i = 0;
115 break;
116
117 case 10ULL ... 99ULL:
118 i = 1;
119 break;
120
121 case 100ULL ... 999ULL:
122 i = 2;
123 break;
124
125 case 1000ULL ... 9999ULL:
126 i = 3;
127 break;
128
129 case 10000ULL ... 99999ULL:
130 i = 4;
131 break;
132
133 case 100000ULL ... 999999ULL:
134 i = 5;
135 break;
136
137 case 1000000ULL ... 9999999ULL:
138 i = 6;
139 break;
140
141 case 10000000ULL ... 99999999ULL:
142 i = 7;
143 break;
144
145 case 100000000ULL ... 999999999ULL:
146 i = 8;
147 break;
148
149 case 1000000000ULL ... 9999999999ULL:
150 i = 9;
151 break;
152
153 case 10000000000ULL ... 99999999999ULL:
154 i = 10;
155 break;
156
157 case 100000000000ULL ... 999999999999ULL:
158 i = 11;
159 break;
160
161 case 1000000000000ULL ... 9999999999999ULL:
162 i = 12;
163 break;
164
165 case 10000000000000ULL ... 99999999999999ULL:
166 i = 13;
167 break;
168
169 case 100000000000000ULL ... 999999999999999ULL:
170 i = 14;
171 break;
172
173 case 1000000000000000ULL ... 9999999999999999ULL:
174 i = 15;
175 break;
176
177 case 10000000000000000ULL ... 99999999999999999ULL:
178 i = 16;
179 break;
180
181 case 100000000000000000ULL ... 999999999999999999ULL:
182 i = 17;
183 break;
184
185 case 1000000000000000000ULL ... 9999999999999999999ULL:
186 i = 18;
187 break;
188
189 case 10000000000000000000ULL ... ULLONG_MAX:
190 i = 19;
191 break;
192 }
193 if (i + 2 > size) // (i + 1) + '\0'
194 return NULL; // too long
195 res = dst + i + 1;
196 *res = '\0';
197 for (; i >= 0; i--) {
198 dst[i] = n % 10ULL + '0';
199 n /= 10ULL;
200 }
201 return res;
202}
203
204/*
205 * unsigned long ASCII representation
206 *
207 * return the last char '\0' or NULL if no enough
208 * space in dst
209 */
210char *ultoa_o(unsigned long n, char *dst, size_t size)
211{
212 int i = 0;
213 char *res;
214
215 switch (n) {
216 case 0U ... 9UL:
217 i = 0;
218 break;
219
220 case 10U ... 99UL:
221 i = 1;
222 break;
223
224 case 100U ... 999UL:
225 i = 2;
226 break;
227
228 case 1000U ... 9999UL:
229 i = 3;
230 break;
231
232 case 10000U ... 99999UL:
233 i = 4;
234 break;
235
236 case 100000U ... 999999UL:
237 i = 5;
238 break;
239
240 case 1000000U ... 9999999UL:
241 i = 6;
242 break;
243
244 case 10000000U ... 99999999UL:
245 i = 7;
246 break;
247
248 case 100000000U ... 999999999UL:
249 i = 8;
250 break;
251#if __WORDSIZE == 32
252
253 case 1000000000ULL ... ULONG_MAX:
254 i = 9;
255 break;
256
257#elif __WORDSIZE == 64
258
259 case 1000000000ULL ... 9999999999UL:
260 i = 9;
261 break;
262
263 case 10000000000ULL ... 99999999999UL:
264 i = 10;
265 break;
266
267 case 100000000000ULL ... 999999999999UL:
268 i = 11;
269 break;
270
271 case 1000000000000ULL ... 9999999999999UL:
272 i = 12;
273 break;
274
275 case 10000000000000ULL ... 99999999999999UL:
276 i = 13;
277 break;
278
279 case 100000000000000ULL ... 999999999999999UL:
280 i = 14;
281 break;
282
283 case 1000000000000000ULL ... 9999999999999999UL:
284 i = 15;
285 break;
286
287 case 10000000000000000ULL ... 99999999999999999UL:
288 i = 16;
289 break;
290
291 case 100000000000000000ULL ... 999999999999999999UL:
292 i = 17;
293 break;
294
295 case 1000000000000000000ULL ... 9999999999999999999UL:
296 i = 18;
297 break;
298
299 case 10000000000000000000ULL ... ULONG_MAX:
300 i = 19;
301 break;
302
303#endif
304 }
305 if (i + 2 > size) // (i + 1) + '\0'
306 return NULL; // too long
307 res = dst + i + 1;
308 *res = '\0';
309 for (; i >= 0; i--) {
310 dst[i] = n % 10U + '0';
311 n /= 10U;
312 }
313 return res;
314}
315
316/*
317 * signed long ASCII representation
318 *
319 * return the last char '\0' or NULL if no enough
320 * space in dst
321 */
322char *ltoa_o(long int n, char *dst, size_t size)
323{
324 char *pos = dst;
325
326 if (n < 0) {
327 if (size < 3)
328 return NULL; // min size is '-' + digit + '\0' but another test in ultoa
329 *pos = '-';
330 pos++;
331 dst = ultoa_o(-n, pos, size - 1);
332 } else {
333 dst = ultoa_o(n, dst, size);
334 }
335 return dst;
336}
337
338/*
339 * signed long long ASCII representation
340 *
341 * return the last char '\0' or NULL if no enough
342 * space in dst
343 */
344char *lltoa(long long n, char *dst, size_t size)
345{
346 char *pos = dst;
347
348 if (n < 0) {
349 if (size < 3)
350 return NULL; // min size is '-' + digit + '\0' but another test in ulltoa
351 *pos = '-';
352 pos++;
353 dst = ulltoa(-n, pos, size - 1);
354 } else {
355 dst = ulltoa(n, dst, size);
356 }
357 return dst;
358}
359
360/*
361 * write a ascii representation of a unsigned into dst,
362 * return a pointer to the last character
363 * Pad the ascii representation with '0', using size.
364 */
365char *utoa_pad(unsigned int n, char *dst, size_t size)
366{
367 int i = 0;
368 char *ret;
369
370 switch(n) {
371 case 0U ... 9U:
372 i = 0;
373 break;
374
375 case 10U ... 99U:
376 i = 1;
377 break;
378
379 case 100U ... 999U:
380 i = 2;
381 break;
382
383 case 1000U ... 9999U:
384 i = 3;
385 break;
386
387 case 10000U ... 99999U:
388 i = 4;
389 break;
390
391 case 100000U ... 999999U:
392 i = 5;
393 break;
394
395 case 1000000U ... 9999999U:
396 i = 6;
397 break;
398
399 case 10000000U ... 99999999U:
400 i = 7;
401 break;
402
403 case 100000000U ... 999999999U:
404 i = 8;
405 break;
406
407 case 1000000000U ... 4294967295U:
408 i = 9;
409 break;
410 }
411 if (i + 2 > size) // (i + 1) + '\0'
412 return NULL; // too long
413 if (i < size)
414 i = size - 2; // padding - '\0'
415
416 ret = dst + i + 1;
417 *ret = '\0';
418 for (; i >= 0; i--) {
419 dst[i] = n % 10U + '0';
420 n /= 10U;
421 }
422 return ret;
423}
424
425/*
Willy Tarreaubaaee002006-06-26 02:48:02 +0200426 * copies at most <size-1> chars from <src> to <dst>. Last char is always
427 * set to 0, unless <size> is 0. The number of chars copied is returned
428 * (excluding the terminating zero).
429 * This code has been optimized for size and speed : on x86, it's 45 bytes
430 * long, uses only registers, and consumes only 4 cycles per char.
431 */
432int strlcpy2(char *dst, const char *src, int size)
433{
434 char *orig = dst;
435 if (size) {
436 while (--size && (*dst = *src)) {
437 src++; dst++;
438 }
439 *dst = 0;
440 }
441 return dst - orig;
442}
443
444/*
Willy Tarreau72d759c2007-10-25 12:14:10 +0200445 * This function simply returns a locally allocated string containing
Willy Tarreaubaaee002006-06-26 02:48:02 +0200446 * the ascii representation for number 'n' in decimal.
447 */
Emeric Brun3a7fce52010-01-04 14:54:38 +0100448char *ultoa_r(unsigned long n, char *buffer, int size)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200449{
450 char *pos;
451
Willy Tarreau72d759c2007-10-25 12:14:10 +0200452 pos = buffer + size - 1;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200453 *pos-- = '\0';
454
455 do {
456 *pos-- = '0' + n % 10;
457 n /= 10;
Willy Tarreau72d759c2007-10-25 12:14:10 +0200458 } while (n && pos >= buffer);
Willy Tarreaubaaee002006-06-26 02:48:02 +0200459 return pos + 1;
460}
461
Willy Tarreau91092e52007-10-25 16:58:42 +0200462/*
Willy Tarreaue7239b52009-03-29 13:41:58 +0200463 * This function simply returns a locally allocated string containing
Thierry FOURNIER763a5d82015-07-06 23:09:52 +0200464 * the ascii representation for number 'n' in decimal.
465 */
466char *lltoa_r(long long int in, char *buffer, int size)
467{
468 char *pos;
469 int neg = 0;
470 unsigned long long int n;
471
472 pos = buffer + size - 1;
473 *pos-- = '\0';
474
475 if (in < 0) {
476 neg = 1;
477 n = -in;
478 }
479 else
480 n = in;
481
482 do {
483 *pos-- = '0' + n % 10;
484 n /= 10;
485 } while (n && pos >= buffer);
486 if (neg && pos > buffer)
487 *pos-- = '-';
488 return pos + 1;
489}
490
491/*
492 * This function simply returns a locally allocated string containing
Thierry FOURNIER1480bd82015-06-06 19:14:59 +0200493 * the ascii representation for signed number 'n' in decimal.
494 */
495char *sltoa_r(long n, char *buffer, int size)
496{
497 char *pos;
498
499 if (n >= 0)
500 return ultoa_r(n, buffer, size);
501
502 pos = ultoa_r(-n, buffer + 1, size - 1) - 1;
503 *pos = '-';
504 return pos;
505}
506
507/*
508 * This function simply returns a locally allocated string containing
Willy Tarreaue7239b52009-03-29 13:41:58 +0200509 * the ascii representation for number 'n' in decimal, formatted for
510 * HTML output with tags to create visual grouping by 3 digits. The
511 * output needs to support at least 171 characters.
512 */
513const char *ulltoh_r(unsigned long long n, char *buffer, int size)
514{
515 char *start;
516 int digit = 0;
517
518 start = buffer + size;
519 *--start = '\0';
520
521 do {
522 if (digit == 3 && start >= buffer + 7)
523 memcpy(start -= 7, "</span>", 7);
524
525 if (start >= buffer + 1) {
526 *--start = '0' + n % 10;
527 n /= 10;
528 }
529
530 if (digit == 3 && start >= buffer + 18)
531 memcpy(start -= 18, "<span class=\"rls\">", 18);
532
533 if (digit++ == 3)
534 digit = 1;
535 } while (n && start > buffer);
536 return start;
537}
538
539/*
Willy Tarreau91092e52007-10-25 16:58:42 +0200540 * This function simply returns a locally allocated string containing the ascii
541 * representation for number 'n' in decimal, unless n is 0 in which case it
542 * returns the alternate string (or an empty string if the alternate string is
543 * NULL). It use is intended for limits reported in reports, where it's
544 * desirable not to display anything if there is no limit. Warning! it shares
545 * the same vector as ultoa_r().
546 */
547const char *limit_r(unsigned long n, char *buffer, int size, const char *alt)
548{
549 return (n) ? ultoa_r(n, buffer, size) : (alt ? alt : "");
550}
551
Willy Tarreau56d1d8d2021-05-08 10:28:53 +0200552/* Trims the first "%f" float in a string to its minimum number of digits after
553 * the decimal point by trimming trailing zeroes, even dropping the decimal
554 * point if not needed. The string is in <buffer> of length <len>, and the
555 * number is expected to start at or after position <num_start> (the first
556 * point appearing there is considered). A NUL character is always placed at
557 * the end if some trimming occurs. The new buffer length is returned.
558 */
559size_t flt_trim(char *buffer, size_t num_start, size_t len)
560{
561 char *end = buffer + len;
562 char *p = buffer + num_start;
563 char *trim;
564
565 do {
566 if (p >= end)
567 return len;
568 trim = p++;
569 } while (*trim != '.');
570
571 /* For now <trim> is on the decimal point. Let's look for any other
572 * meaningful digit after it.
573 */
574 while (p < end) {
575 if (*p++ != '0')
576 trim = p;
577 }
578
579 if (trim < end)
580 *trim = 0;
581
582 return trim - buffer;
583}
584
Willy Tarreauae03d262021-05-08 07:35:00 +0200585/*
586 * This function simply returns a locally allocated string containing
587 * the ascii representation for number 'n' in decimal with useless trailing
588 * zeroes trimmed.
589 */
590char *ftoa_r(double n, char *buffer, int size)
591{
592 flt_trim(buffer, 0, snprintf(buffer, size, "%f", n));
593 return buffer;
594}
595
Willy Tarreau588297f2014-06-16 15:16:40 +0200596/* returns a locally allocated string containing the quoted encoding of the
597 * input string. The output may be truncated to QSTR_SIZE chars, but it is
598 * guaranteed that the string will always be properly terminated. Quotes are
599 * encoded by doubling them as is commonly done in CSV files. QSTR_SIZE must
600 * always be at least 4 chars.
601 */
602const char *qstr(const char *str)
603{
604 char *ret = quoted_str[quoted_idx];
605 char *p, *end;
606
607 if (++quoted_idx >= NB_QSTR)
608 quoted_idx = 0;
609
610 p = ret;
611 end = ret + QSTR_SIZE;
612
613 *p++ = '"';
614
615 /* always keep 3 chars to support passing "" and the ending " */
616 while (*str && p < end - 3) {
617 if (*str == '"') {
618 *p++ = '"';
619 *p++ = '"';
620 }
621 else
622 *p++ = *str;
623 str++;
624 }
625 *p++ = '"';
626 return ret;
627}
628
Robert Tsai81ae1952007-12-05 10:47:29 +0100629/*
Willy Tarreaubaaee002006-06-26 02:48:02 +0200630 * Returns non-zero if character <s> is a hex digit (0-9, a-f, A-F), else zero.
631 *
632 * It looks like this one would be a good candidate for inlining, but this is
633 * not interesting because it around 35 bytes long and often called multiple
634 * times within the same function.
635 */
636int ishex(char s)
637{
638 s -= '0';
639 if ((unsigned char)s <= 9)
640 return 1;
641 s -= 'A' - '0';
642 if ((unsigned char)s <= 5)
643 return 1;
644 s -= 'a' - 'A';
645 if ((unsigned char)s <= 5)
646 return 1;
647 return 0;
648}
649
Willy Tarreau3ca1a882015-01-15 18:43:49 +0100650/* rounds <i> down to the closest value having max 2 digits */
651unsigned int round_2dig(unsigned int i)
652{
653 unsigned int mul = 1;
654
655 while (i >= 100) {
656 i /= 10;
657 mul *= 10;
658 }
659 return i * mul;
660}
661
Willy Tarreau2e74c3f2007-12-02 18:45:09 +0100662/*
663 * Checks <name> for invalid characters. Valid chars are [A-Za-z0-9_:.-]. If an
664 * invalid character is found, a pointer to it is returned. If everything is
665 * fine, NULL is returned.
666 */
667const char *invalid_char(const char *name)
668{
669 if (!*name)
670 return name;
671
672 while (*name) {
Willy Tarreau90807112020-02-25 08:16:33 +0100673 if (!isalnum((unsigned char)*name) && *name != '.' && *name != ':' &&
Willy Tarreau2e74c3f2007-12-02 18:45:09 +0100674 *name != '_' && *name != '-')
675 return name;
676 name++;
677 }
678 return NULL;
679}
Willy Tarreaubaaee002006-06-26 02:48:02 +0200680
681/*
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200682 * Checks <name> for invalid characters. Valid chars are [_.-] and those
683 * accepted by <f> function.
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200684 * If an invalid character is found, a pointer to it is returned.
685 * If everything is fine, NULL is returned.
686 */
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200687static inline const char *__invalid_char(const char *name, int (*f)(int)) {
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200688
689 if (!*name)
690 return name;
691
692 while (*name) {
Willy Tarreau90807112020-02-25 08:16:33 +0100693 if (!f((unsigned char)*name) && *name != '.' &&
Krzysztof Piotr Oledzkiefe3b6f2008-05-23 23:49:32 +0200694 *name != '_' && *name != '-')
695 return name;
696
697 name++;
698 }
699
700 return NULL;
701}
702
703/*
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200704 * Checks <name> for invalid characters. Valid chars are [A-Za-z0-9_.-].
705 * If an invalid character is found, a pointer to it is returned.
706 * If everything is fine, NULL is returned.
707 */
708const char *invalid_domainchar(const char *name) {
709 return __invalid_char(name, isalnum);
710}
711
712/*
713 * Checks <name> for invalid characters. Valid chars are [A-Za-z_.-].
714 * If an invalid character is found, a pointer to it is returned.
715 * If everything is fine, NULL is returned.
716 */
717const char *invalid_prefix_char(const char *name) {
Thierry Fournierf7b7c3e2018-03-26 11:54:39 +0200718 return __invalid_char(name, isalnum);
Frédéric Lécailleb82f7422017-04-13 18:24:23 +0200719}
720
721/*
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100722 * converts <str> to a struct sockaddr_storage* provided by the caller. The
Willy Tarreau24709282013-03-10 21:32:12 +0100723 * caller must have zeroed <sa> first, and may have set sa->ss_family to force
724 * parse a specific address format. If the ss_family is 0 or AF_UNSPEC, then
725 * the function tries to guess the address family from the syntax. If the
726 * family is forced and the format doesn't match, an error is returned. The
Willy Tarreaufab5a432011-03-04 15:31:53 +0100727 * string is assumed to contain only an address, no port. The address can be a
728 * dotted IPv4 address, an IPv6 address, a host name, or empty or "*" to
729 * indicate INADDR_ANY. NULL is returned if the host part cannot be resolved.
730 * The return address will only have the address family and the address set,
731 * all other fields remain zero. The string is not supposed to be modified.
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100732 * The IPv6 '::' address is IN6ADDR_ANY. If <resolve> is non-zero, the hostname
733 * is resolved, otherwise only IP addresses are resolved, and anything else
Willy Tarreauecde7df2016-11-02 22:37:03 +0100734 * returns NULL. If the address contains a port, this one is preserved.
Willy Tarreaubaaee002006-06-26 02:48:02 +0200735 */
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100736struct sockaddr_storage *str2ip2(const char *str, struct sockaddr_storage *sa, int resolve)
Willy Tarreaubaaee002006-06-26 02:48:02 +0200737{
Willy Tarreaufab5a432011-03-04 15:31:53 +0100738 struct hostent *he;
mildisff5d5102015-10-26 18:50:08 +0100739 /* max IPv6 length, including brackets and terminating NULL */
740 char tmpip[48];
Willy Tarreauecde7df2016-11-02 22:37:03 +0100741 int port = get_host_port(sa);
mildisff5d5102015-10-26 18:50:08 +0100742
743 /* check IPv6 with square brackets */
744 if (str[0] == '[') {
745 size_t iplength = strlen(str);
746
747 if (iplength < 4) {
748 /* minimal size is 4 when using brackets "[::]" */
749 goto fail;
750 }
751 else if (iplength >= sizeof(tmpip)) {
752 /* IPv6 literal can not be larger than tmpip */
753 goto fail;
754 }
755 else {
756 if (str[iplength - 1] != ']') {
757 /* if address started with bracket, it should end with bracket */
758 goto fail;
759 }
760 else {
761 memcpy(tmpip, str + 1, iplength - 2);
762 tmpip[iplength - 2] = '\0';
763 str = tmpip;
764 }
765 }
766 }
Willy Tarreaufab5a432011-03-04 15:31:53 +0100767
Willy Tarreaufab5a432011-03-04 15:31:53 +0100768 /* Any IPv6 address */
769 if (str[0] == ':' && str[1] == ':' && !str[2]) {
Willy Tarreau24709282013-03-10 21:32:12 +0100770 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
771 sa->ss_family = AF_INET6;
772 else if (sa->ss_family != AF_INET6)
773 goto fail;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100774 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100775 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100776 }
777
Willy Tarreau24709282013-03-10 21:32:12 +0100778 /* Any address for the family, defaults to IPv4 */
Willy Tarreaufab5a432011-03-04 15:31:53 +0100779 if (!str[0] || (str[0] == '*' && !str[1])) {
Willy Tarreau24709282013-03-10 21:32:12 +0100780 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
781 sa->ss_family = AF_INET;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100782 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100783 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100784 }
785
786 /* check for IPv6 first */
Willy Tarreau24709282013-03-10 21:32:12 +0100787 if ((!sa->ss_family || sa->ss_family == AF_UNSPEC || sa->ss_family == AF_INET6) &&
788 inet_pton(AF_INET6, str, &((struct sockaddr_in6 *)sa)->sin6_addr)) {
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100789 sa->ss_family = AF_INET6;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100790 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100791 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100792 }
793
794 /* then check for IPv4 */
Willy Tarreau24709282013-03-10 21:32:12 +0100795 if ((!sa->ss_family || sa->ss_family == AF_UNSPEC || sa->ss_family == AF_INET) &&
796 inet_pton(AF_INET, str, &((struct sockaddr_in *)sa)->sin_addr)) {
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100797 sa->ss_family = AF_INET;
Willy Tarreauecde7df2016-11-02 22:37:03 +0100798 set_host_port(sa, port);
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100799 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100800 }
801
Thierry FOURNIER58639a02014-11-25 12:02:25 +0100802 if (!resolve)
803 return NULL;
804
Emeric Brund30e9a12020-12-23 18:49:16 +0100805 if (!resolv_hostname_validation(str, NULL))
Baptiste Assmanna68ca962015-04-14 01:15:08 +0200806 return NULL;
807
David du Colombierd5f43282011-03-17 10:40:16 +0100808#ifdef USE_GETADDRINFO
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200809 if (global.tune.options & GTUNE_USE_GAI) {
David du Colombierd5f43282011-03-17 10:40:16 +0100810 struct addrinfo hints, *result;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100811 int success = 0;
David du Colombierd5f43282011-03-17 10:40:16 +0100812
813 memset(&result, 0, sizeof(result));
814 memset(&hints, 0, sizeof(hints));
Willy Tarreau24709282013-03-10 21:32:12 +0100815 hints.ai_family = sa->ss_family ? sa->ss_family : AF_UNSPEC;
David du Colombierd5f43282011-03-17 10:40:16 +0100816 hints.ai_socktype = SOCK_DGRAM;
Dmitry Sivachenkoeab7f392015-10-02 01:01:58 +0200817 hints.ai_flags = 0;
David du Colombierd5f43282011-03-17 10:40:16 +0100818 hints.ai_protocol = 0;
819
820 if (getaddrinfo(str, NULL, &hints, &result) == 0) {
Willy Tarreau24709282013-03-10 21:32:12 +0100821 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
822 sa->ss_family = result->ai_family;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100823 else if (sa->ss_family != result->ai_family) {
824 freeaddrinfo(result);
Willy Tarreau24709282013-03-10 21:32:12 +0100825 goto fail;
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100826 }
Willy Tarreau24709282013-03-10 21:32:12 +0100827
David du Colombierd5f43282011-03-17 10:40:16 +0100828 switch (result->ai_family) {
829 case AF_INET:
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100830 memcpy((struct sockaddr_in *)sa, result->ai_addr, result->ai_addrlen);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100831 set_host_port(sa, port);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100832 success = 1;
833 break;
David du Colombierd5f43282011-03-17 10:40:16 +0100834 case AF_INET6:
Willy Tarreauc120c8d2013-03-10 19:27:44 +0100835 memcpy((struct sockaddr_in6 *)sa, result->ai_addr, result->ai_addrlen);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100836 set_host_port(sa, port);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100837 success = 1;
838 break;
David du Colombierd5f43282011-03-17 10:40:16 +0100839 }
840 }
841
Sean Carey58ea0392013-02-15 23:39:18 +0100842 if (result)
843 freeaddrinfo(result);
Tim Duesterhus7d58b4d2018-01-21 22:11:17 +0100844
845 if (success)
846 return sa;
Willy Tarreaufab5a432011-03-04 15:31:53 +0100847 }
David du Colombierd5f43282011-03-17 10:40:16 +0100848#endif
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200849 /* try to resolve an IPv4/IPv6 hostname */
850 he = gethostbyname(str);
851 if (he) {
852 if (!sa->ss_family || sa->ss_family == AF_UNSPEC)
853 sa->ss_family = he->h_addrtype;
854 else if (sa->ss_family != he->h_addrtype)
855 goto fail;
856
857 switch (sa->ss_family) {
858 case AF_INET:
859 ((struct sockaddr_in *)sa)->sin_addr = *(struct in_addr *) *(he->h_addr_list);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100860 set_host_port(sa, port);
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200861 return sa;
862 case AF_INET6:
863 ((struct sockaddr_in6 *)sa)->sin6_addr = *(struct in6_addr *) *(he->h_addr_list);
Willy Tarreauecde7df2016-11-02 22:37:03 +0100864 set_host_port(sa, port);
Nenad Merdanovic88afe032014-04-14 15:56:58 +0200865 return sa;
866 }
867 }
868
David du Colombierd5f43282011-03-17 10:40:16 +0100869 /* unsupported address family */
Willy Tarreau24709282013-03-10 21:32:12 +0100870 fail:
Willy Tarreaufab5a432011-03-04 15:31:53 +0100871 return NULL;
872}
873
874/*
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100875 * Converts <str> to a locally allocated struct sockaddr_storage *, and a port
876 * range or offset consisting in two integers that the caller will have to
877 * check to find the relevant input format. The following format are supported :
878 *
879 * String format | address | port | low | high
880 * addr | <addr> | 0 | 0 | 0
881 * addr: | <addr> | 0 | 0 | 0
882 * addr:port | <addr> | <port> | <port> | <port>
883 * addr:pl-ph | <addr> | <pl> | <pl> | <ph>
884 * addr:+port | <addr> | <port> | 0 | <port>
885 * addr:-port | <addr> |-<port> | <port> | 0
886 *
887 * The detection of a port range or increment by the caller is made by
888 * comparing <low> and <high>. If both are equal, then port 0 means no port
889 * was specified. The caller may pass NULL for <low> and <high> if it is not
890 * interested in retrieving port ranges.
891 *
892 * Note that <addr> above may also be :
893 * - empty ("") => family will be AF_INET and address will be INADDR_ANY
894 * - "*" => family will be AF_INET and address will be INADDR_ANY
895 * - "::" => family will be AF_INET6 and address will be IN6ADDR_ANY
896 * - a host name => family and address will depend on host name resolving.
897 *
Willy Tarreau24709282013-03-10 21:32:12 +0100898 * A prefix may be passed in before the address above to force the family :
899 * - "ipv4@" => force address to resolve as IPv4 and fail if not possible.
900 * - "ipv6@" => force address to resolve as IPv6 and fail if not possible.
901 * - "unix@" => force address to be a path to a UNIX socket even if the
902 * path does not start with a '/'
Willy Tarreauccfccef2014-05-10 01:49:15 +0200903 * - 'abns@' -> force address to belong to the abstract namespace (Linux
904 * only). These sockets are just like Unix sockets but without
905 * the need for an underlying file system. The address is a
906 * string. Technically it's like a Unix socket with a zero in
907 * the first byte of the address.
Willy Tarreau40aa0702013-03-10 23:51:38 +0100908 * - "fd@" => an integer must follow, and is a file descriptor number.
Willy Tarreau24709282013-03-10 21:32:12 +0100909 *
mildisff5d5102015-10-26 18:50:08 +0100910 * IPv6 addresses can be declared with or without square brackets. When using
911 * square brackets for IPv6 addresses, the port separator (colon) is optional.
912 * If not using square brackets, and in order to avoid any ambiguity with
913 * IPv6 addresses, the last colon ':' is mandatory even when no port is specified.
914 * NULL is returned if the address cannot be parsed. The <low> and <high> ports
915 * are always initialized if non-null, even for non-IP families.
Willy Tarreaud393a622013-03-04 18:22:00 +0100916 *
917 * If <pfx> is non-null, it is used as a string prefix before any path-based
918 * address (typically the path to a unix socket).
Willy Tarreau40aa0702013-03-10 23:51:38 +0100919 *
Willy Tarreau72b8c1f2015-09-08 15:50:19 +0200920 * if <fqdn> is non-null, it will be filled with :
921 * - a pointer to the FQDN of the server name to resolve if there's one, and
922 * that the caller will have to free(),
923 * - NULL if there was an explicit address that doesn't require resolution.
924 *
Willy Tarreaucd3a55912020-09-04 15:30:46 +0200925 * Hostnames are only resolved if <opts> has PA_O_RESOLVE. Otherwise <fqdn> is
926 * still honored so it is possible for the caller to know whether a resolution
927 * failed by clearing this flag and checking if <fqdn> was filled, indicating
928 * the need for a resolution.
Thierry FOURNIER7fe3be72015-09-26 20:03:36 +0200929 *
Willy Tarreau40aa0702013-03-10 23:51:38 +0100930 * When a file descriptor is passed, its value is put into the s_addr part of
Willy Tarreaua5b325f2020-09-04 16:44:20 +0200931 * the address when cast to sockaddr_in and the address family is
932 * AF_CUST_EXISTING_FD.
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200933 *
Willy Tarreau5fc93282020-09-16 18:25:03 +0200934 * The matching protocol will be set into <proto> if non-null.
935 *
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200936 * Any known file descriptor is also assigned to <fd> if non-null, otherwise it
937 * is forced to -1.
Willy Tarreaufab5a432011-03-04 15:31:53 +0100938 */
Willy Tarreau5fc93282020-09-16 18:25:03 +0200939struct sockaddr_storage *str2sa_range(const char *str, int *port, int *low, int *high, int *fd,
940 struct protocol **proto, char **err,
941 const char *pfx, char **fqdn, unsigned int opts)
Willy Tarreaufab5a432011-03-04 15:31:53 +0100942{
Christopher Faulet1bc04c72017-10-29 20:14:08 +0100943 static THREAD_LOCAL struct sockaddr_storage ss;
David du Colombier6f5ccb12011-03-10 22:26:24 +0100944 struct sockaddr_storage *ret = NULL;
Willy Tarreau5fc93282020-09-16 18:25:03 +0200945 struct protocol *new_proto = NULL;
Willy Tarreau24709282013-03-10 21:32:12 +0100946 char *back, *str2;
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100947 char *port1, *port2;
948 int portl, porth, porta;
Willy Tarreauccfccef2014-05-10 01:49:15 +0200949 int abstract = 0;
Willy Tarreaua93e5c72020-09-15 14:01:16 +0200950 int new_fd = -1;
Willy Tarreaue835bd82020-09-16 11:35:47 +0200951 int sock_type, ctrl_type;
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100952
953 portl = porth = porta = 0;
Willy Tarreau72b8c1f2015-09-08 15:50:19 +0200954 if (fqdn)
955 *fqdn = NULL;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200956
Willy Tarreaudad36a32013-03-11 01:20:04 +0100957 str2 = back = env_expand(strdup(str));
Willy Tarreaudf350f12013-03-01 20:22:54 +0100958 if (str2 == NULL) {
959 memprintf(err, "out of memory in '%s'\n", __FUNCTION__);
Willy Tarreaud5191e72010-02-09 20:50:45 +0100960 goto out;
Willy Tarreaudf350f12013-03-01 20:22:54 +0100961 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200962
Willy Tarreau9f69f462015-09-08 16:01:25 +0200963 if (!*str2) {
964 memprintf(err, "'%s' resolves to an empty address (environment variable missing?)\n", str);
965 goto out;
966 }
967
Willy Tarreau24709282013-03-10 21:32:12 +0100968 memset(&ss, 0, sizeof(ss));
969
Willy Tarreaue835bd82020-09-16 11:35:47 +0200970 /* prepare the default socket types */
Willy Tarreauf23b1bc2021-03-23 18:36:37 +0100971 if ((opts & (PA_O_STREAM|PA_O_DGRAM)) == PA_O_DGRAM ||
972 ((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 +0200973 sock_type = ctrl_type = SOCK_DGRAM;
974 else
975 sock_type = ctrl_type = SOCK_STREAM;
976
977 if (strncmp(str2, "stream+", 7) == 0) {
978 str2 += 7;
979 sock_type = ctrl_type = SOCK_STREAM;
980 }
981 else if (strncmp(str2, "dgram+", 6) == 0) {
982 str2 += 6;
983 sock_type = ctrl_type = SOCK_DGRAM;
984 }
985
Willy Tarreau24709282013-03-10 21:32:12 +0100986 if (strncmp(str2, "unix@", 5) == 0) {
987 str2 += 5;
Willy Tarreauccfccef2014-05-10 01:49:15 +0200988 abstract = 0;
Willy Tarreau24709282013-03-10 21:32:12 +0100989 ss.ss_family = AF_UNIX;
990 }
Emeric Brunce325c42021-04-02 17:05:09 +0200991 else if (strncmp(str2, "uxdg@", 5) == 0) {
992 str2 += 5;
993 abstract = 0;
994 ss.ss_family = AF_UNIX;
995 sock_type = ctrl_type = SOCK_DGRAM;
996 }
997 else if (strncmp(str2, "uxst@", 5) == 0) {
998 str2 += 5;
999 abstract = 0;
1000 ss.ss_family = AF_UNIX;
1001 sock_type = ctrl_type = SOCK_STREAM;
1002 }
Willy Tarreauccfccef2014-05-10 01:49:15 +02001003 else if (strncmp(str2, "abns@", 5) == 0) {
1004 str2 += 5;
1005 abstract = 1;
1006 ss.ss_family = AF_UNIX;
1007 }
Emeric Brunce325c42021-04-02 17:05:09 +02001008 else if (strncmp(str2, "ip@", 3) == 0) {
1009 str2 += 3;
1010 ss.ss_family = AF_UNSPEC;
1011 }
Willy Tarreau24709282013-03-10 21:32:12 +01001012 else if (strncmp(str2, "ipv4@", 5) == 0) {
1013 str2 += 5;
1014 ss.ss_family = AF_INET;
1015 }
1016 else if (strncmp(str2, "ipv6@", 5) == 0) {
1017 str2 += 5;
1018 ss.ss_family = AF_INET6;
1019 }
Emeric Brunce325c42021-04-02 17:05:09 +02001020 else if (strncmp(str2, "tcp4@", 5) == 0) {
1021 str2 += 5;
1022 ss.ss_family = AF_INET;
1023 sock_type = ctrl_type = SOCK_STREAM;
1024 }
Emeric Brun3835c0d2020-07-07 09:46:09 +02001025 else if (strncmp(str2, "udp4@", 5) == 0) {
1026 str2 += 5;
1027 ss.ss_family = AF_INET;
Willy Tarreaue835bd82020-09-16 11:35:47 +02001028 sock_type = ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +02001029 }
Emeric Brunce325c42021-04-02 17:05:09 +02001030 else if (strncmp(str2, "tcp6@", 5) == 0) {
1031 str2 += 5;
1032 ss.ss_family = AF_INET6;
1033 sock_type = ctrl_type = SOCK_STREAM;
1034 }
Emeric Brun3835c0d2020-07-07 09:46:09 +02001035 else if (strncmp(str2, "udp6@", 5) == 0) {
1036 str2 += 5;
1037 ss.ss_family = AF_INET6;
Willy Tarreaue835bd82020-09-16 11:35:47 +02001038 sock_type = ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +02001039 }
Emeric Brunce325c42021-04-02 17:05:09 +02001040 else if (strncmp(str2, "tcp@", 4) == 0) {
1041 str2 += 4;
1042 ss.ss_family = AF_UNSPEC;
1043 sock_type = ctrl_type = SOCK_STREAM;
1044 }
Emeric Brun3835c0d2020-07-07 09:46:09 +02001045 else if (strncmp(str2, "udp@", 4) == 0) {
1046 str2 += 4;
1047 ss.ss_family = AF_UNSPEC;
Willy Tarreaue835bd82020-09-16 11:35:47 +02001048 sock_type = ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +02001049 }
Frédéric Lécaille10caf652020-11-23 11:36:57 +01001050 else if (strncmp(str2, "quic4@", 6) == 0) {
1051 str2 += 6;
1052 ss.ss_family = AF_INET;
1053 sock_type = SOCK_DGRAM;
1054 ctrl_type = SOCK_STREAM;
1055 }
1056 else if (strncmp(str2, "quic6@", 6) == 0) {
1057 str2 += 6;
1058 ss.ss_family = AF_INET6;
1059 sock_type = SOCK_DGRAM;
1060 ctrl_type = SOCK_STREAM;
1061 }
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001062 else if (strncmp(str2, "fd@", 3) == 0) {
1063 str2 += 3;
1064 ss.ss_family = AF_CUST_EXISTING_FD;
1065 }
1066 else if (strncmp(str2, "sockpair@", 9) == 0) {
1067 str2 += 9;
1068 ss.ss_family = AF_CUST_SOCKPAIR;
1069 }
Willy Tarreau24709282013-03-10 21:32:12 +01001070 else if (*str2 == '/') {
1071 ss.ss_family = AF_UNIX;
1072 }
1073 else
1074 ss.ss_family = AF_UNSPEC;
1075
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001076 if (ss.ss_family == AF_CUST_SOCKPAIR) {
Willy Tarreaua215be22020-09-16 10:14:16 +02001077 struct sockaddr_storage ss2;
1078 socklen_t addr_len;
William Lallemand2fe7dd02018-09-11 16:51:29 +02001079 char *endptr;
1080
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001081 new_fd = strtol(str2, &endptr, 10);
1082 if (!*str2 || new_fd < 0 || *endptr) {
William Lallemand2fe7dd02018-09-11 16:51:29 +02001083 memprintf(err, "file descriptor '%s' is not a valid integer in '%s'\n", str2, str);
1084 goto out;
1085 }
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001086
Willy Tarreaua215be22020-09-16 10:14:16 +02001087 /* just verify that it's a socket */
1088 addr_len = sizeof(ss2);
1089 if (getsockname(new_fd, (struct sockaddr *)&ss2, &addr_len) == -1) {
1090 memprintf(err, "cannot use file descriptor '%d' : %s.\n", new_fd, strerror(errno));
1091 goto out;
1092 }
1093
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001094 ((struct sockaddr_in *)&ss)->sin_addr.s_addr = new_fd;
1095 ((struct sockaddr_in *)&ss)->sin_port = 0;
William Lallemand2fe7dd02018-09-11 16:51:29 +02001096 }
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001097 else if (ss.ss_family == AF_CUST_EXISTING_FD) {
Willy Tarreau40aa0702013-03-10 23:51:38 +01001098 char *endptr;
1099
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001100 new_fd = strtol(str2, &endptr, 10);
1101 if (!*str2 || new_fd < 0 || *endptr) {
Willy Tarreaudad36a32013-03-11 01:20:04 +01001102 memprintf(err, "file descriptor '%s' is not a valid integer in '%s'\n", str2, str);
Willy Tarreau40aa0702013-03-10 23:51:38 +01001103 goto out;
1104 }
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001105
Willy Tarreau6edc7222020-09-15 17:41:56 +02001106 if (opts & PA_O_SOCKET_FD) {
1107 socklen_t addr_len;
1108 int type;
1109
1110 addr_len = sizeof(ss);
1111 if (getsockname(new_fd, (struct sockaddr *)&ss, &addr_len) == -1) {
1112 memprintf(err, "cannot use file descriptor '%d' : %s.\n", new_fd, strerror(errno));
1113 goto out;
1114 }
1115
1116 addr_len = sizeof(type);
1117 if (getsockopt(new_fd, SOL_SOCKET, SO_TYPE, &type, &addr_len) != 0 ||
Willy Tarreaue835bd82020-09-16 11:35:47 +02001118 (type == SOCK_STREAM) != (sock_type == SOCK_STREAM)) {
Willy Tarreau6edc7222020-09-15 17:41:56 +02001119 memprintf(err, "socket on file descriptor '%d' is of the wrong type.\n", new_fd);
1120 goto out;
1121 }
1122
1123 porta = portl = porth = get_host_port(&ss);
1124 } else if (opts & PA_O_RAW_FD) {
1125 ((struct sockaddr_in *)&ss)->sin_addr.s_addr = new_fd;
1126 ((struct sockaddr_in *)&ss)->sin_port = 0;
1127 } else {
1128 memprintf(err, "a file descriptor is not acceptable here in '%s'\n", str);
1129 goto out;
1130 }
Willy Tarreau40aa0702013-03-10 23:51:38 +01001131 }
1132 else if (ss.ss_family == AF_UNIX) {
Willy Tarreau8daa9202019-06-16 18:16:33 +02001133 struct sockaddr_un *un = (struct sockaddr_un *)&ss;
Willy Tarreau15586382013-03-04 19:48:14 +01001134 int prefix_path_len;
1135 int max_path_len;
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001136 int adr_len;
Willy Tarreau15586382013-03-04 19:48:14 +01001137
1138 /* complete unix socket path name during startup or soft-restart is
1139 * <unix_bind_prefix><path>.<pid>.<bak|tmp>
1140 */
Willy Tarreauccfccef2014-05-10 01:49:15 +02001141 prefix_path_len = (pfx && !abstract) ? strlen(pfx) : 0;
Willy Tarreau8daa9202019-06-16 18:16:33 +02001142 max_path_len = (sizeof(un->sun_path) - 1) -
Willy Tarreau327ea5a2020-02-11 06:43:37 +01001143 (abstract ? 0 : prefix_path_len + 1 + 5 + 1 + 3);
Willy Tarreau15586382013-03-04 19:48:14 +01001144
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001145 adr_len = strlen(str2);
1146 if (adr_len > max_path_len) {
Willy Tarreau15586382013-03-04 19:48:14 +01001147 memprintf(err, "socket path '%s' too long (max %d)\n", str, max_path_len);
1148 goto out;
1149 }
1150
Willy Tarreauccfccef2014-05-10 01:49:15 +02001151 /* when abstract==1, we skip the first zero and copy all bytes except the trailing zero */
Willy Tarreau8daa9202019-06-16 18:16:33 +02001152 memset(un->sun_path, 0, sizeof(un->sun_path));
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001153 if (prefix_path_len)
Willy Tarreau8daa9202019-06-16 18:16:33 +02001154 memcpy(un->sun_path, pfx, prefix_path_len);
1155 memcpy(un->sun_path + prefix_path_len + abstract, str2, adr_len + 1 - abstract);
Willy Tarreau15586382013-03-04 19:48:14 +01001156 }
Willy Tarreau24709282013-03-10 21:32:12 +01001157 else { /* IPv4 and IPv6 */
mildisff5d5102015-10-26 18:50:08 +01001158 char *end = str2 + strlen(str2);
1159 char *chr;
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001160
mildisff5d5102015-10-26 18:50:08 +01001161 /* search for : or ] whatever comes first */
1162 for (chr = end-1; chr > str2; chr--) {
1163 if (*chr == ']' || *chr == ':')
1164 break;
1165 }
1166
1167 if (*chr == ':') {
1168 /* Found a colon before a closing-bracket, must be a port separator.
1169 * This guarantee backward compatibility.
1170 */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001171 if (!(opts & PA_O_PORT_OK)) {
1172 memprintf(err, "port specification not permitted here in '%s'", str);
1173 goto out;
1174 }
mildisff5d5102015-10-26 18:50:08 +01001175 *chr++ = '\0';
1176 port1 = chr;
1177 }
1178 else {
1179 /* Either no colon and no closing-bracket
1180 * or directly ending with a closing-bracket.
1181 * However, no port.
1182 */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001183 if (opts & PA_O_PORT_MAND) {
1184 memprintf(err, "missing port specification in '%s'", str);
1185 goto out;
1186 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001187 port1 = "";
mildisff5d5102015-10-26 18:50:08 +01001188 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02001189
Willy Tarreau90807112020-02-25 08:16:33 +01001190 if (isdigit((unsigned char)*port1)) { /* single port or range */
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001191 port2 = strchr(port1, '-');
Willy Tarreau7f96a842020-09-15 11:12:44 +02001192 if (port2) {
1193 if (!(opts & PA_O_PORT_RANGE)) {
1194 memprintf(err, "port range not permitted here in '%s'", str);
1195 goto out;
1196 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001197 *port2++ = '\0';
Willy Tarreau7f96a842020-09-15 11:12:44 +02001198 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001199 else
1200 port2 = port1;
1201 portl = atoi(port1);
1202 porth = atoi(port2);
Willy Tarreau7f96a842020-09-15 11:12:44 +02001203
1204 if (portl < !!(opts & PA_O_PORT_MAND) || portl > 65535) {
1205 memprintf(err, "invalid port '%s'", port1);
1206 goto out;
1207 }
1208
1209 if (porth < !!(opts & PA_O_PORT_MAND) || porth > 65535) {
1210 memprintf(err, "invalid port '%s'", port2);
1211 goto out;
1212 }
1213
1214 if (portl > porth) {
1215 memprintf(err, "invalid port range '%d-%d'", portl, porth);
1216 goto out;
1217 }
1218
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001219 porta = portl;
1220 }
1221 else if (*port1 == '-') { /* negative offset */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001222 if (!(opts & PA_O_PORT_OFS)) {
1223 memprintf(err, "port offset not permitted here in '%s'", str);
1224 goto out;
1225 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001226 portl = atoi(port1 + 1);
1227 porta = -portl;
1228 }
1229 else if (*port1 == '+') { /* positive offset */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001230 if (!(opts & PA_O_PORT_OFS)) {
1231 memprintf(err, "port offset not permitted here in '%s'", str);
1232 goto out;
1233 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001234 porth = atoi(port1 + 1);
1235 porta = porth;
1236 }
1237 else if (*port1) { /* other any unexpected char */
Willy Tarreaudad36a32013-03-11 01:20:04 +01001238 memprintf(err, "invalid character '%c' in port number '%s' in '%s'\n", *port1, port1, str);
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001239 goto out;
1240 }
Willy Tarreau7f96a842020-09-15 11:12:44 +02001241 else if (opts & PA_O_PORT_MAND) {
1242 memprintf(err, "missing port specification in '%s'", str);
1243 goto out;
1244 }
Willy Tarreauceccdd72016-11-02 22:27:10 +01001245
1246 /* first try to parse the IP without resolving. If it fails, it
1247 * tells us we need to keep a copy of the FQDN to resolve later
1248 * and to enable DNS. In this case we can proceed if <fqdn> is
Willy Tarreaucd3a55912020-09-04 15:30:46 +02001249 * set or if PA_O_RESOLVE is set, otherwise it's an error.
Willy Tarreauceccdd72016-11-02 22:27:10 +01001250 */
1251 if (str2ip2(str2, &ss, 0) == NULL) {
Willy Tarreaucd3a55912020-09-04 15:30:46 +02001252 if ((!(opts & PA_O_RESOLVE) && !fqdn) ||
1253 ((opts & PA_O_RESOLVE) && str2ip2(str2, &ss, 1) == NULL)) {
Willy Tarreauceccdd72016-11-02 22:27:10 +01001254 memprintf(err, "invalid address: '%s' in '%s'\n", str2, str);
1255 goto out;
1256 }
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001257
Willy Tarreauceccdd72016-11-02 22:27:10 +01001258 if (fqdn) {
1259 if (str2 != back)
1260 memmove(back, str2, strlen(str2) + 1);
1261 *fqdn = back;
1262 back = NULL;
1263 }
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001264 }
Willy Tarreauceccdd72016-11-02 22:27:10 +01001265 set_host_port(&ss, porta);
Willy Tarreaue4c58c82013-03-06 15:28:17 +01001266 }
Willy Tarreaufab5a432011-03-04 15:31:53 +01001267
Willy Tarreaue835bd82020-09-16 11:35:47 +02001268 if (ctrl_type == SOCK_STREAM && !(opts & PA_O_STREAM)) {
1269 memprintf(err, "stream-type socket not acceptable in '%s'\n", str);
1270 goto out;
1271 }
1272 else if (ctrl_type == SOCK_DGRAM && !(opts & PA_O_DGRAM)) {
1273 memprintf(err, "dgram-type socket not acceptable in '%s'\n", str);
1274 goto out;
1275 }
1276
Willy Tarreau65ec4e32020-09-16 19:17:08 +02001277 if (proto || (opts & PA_O_CONNECT)) {
Willy Tarreau5fc93282020-09-16 18:25:03 +02001278 /* Note: if the caller asks for a proto, we must find one,
Emeric Brun26754902021-04-07 14:26:44 +02001279 * except if we inherit from a raw FD (family == AF_CUST_EXISTING_FD)
1280 * orif we return with an fqdn that will resolve later,
Willy Tarreau5fc93282020-09-16 18:25:03 +02001281 * in which case the address is not known yet (this is only
1282 * for servers actually).
1283 */
Willy Tarreaub2ffc992020-09-16 21:37:31 +02001284 new_proto = protocol_lookup(ss.ss_family,
Willy Tarreauaf9609b2020-09-16 22:04:46 +02001285 sock_type == SOCK_DGRAM,
1286 ctrl_type == SOCK_DGRAM);
Willy Tarreaub2ffc992020-09-16 21:37:31 +02001287
Emeric Brun26754902021-04-07 14:26:44 +02001288 if (!new_proto && (!fqdn || !*fqdn) && (ss.ss_family != AF_CUST_EXISTING_FD)) {
Willy Tarreau5fc93282020-09-16 18:25:03 +02001289 memprintf(err, "unsupported protocol family %d for address '%s'", ss.ss_family, str);
1290 goto out;
1291 }
Willy Tarreau65ec4e32020-09-16 19:17:08 +02001292
1293 if ((opts & PA_O_CONNECT) && new_proto && !new_proto->connect) {
1294 memprintf(err, "connect() not supported for this protocol family %d used by address '%s'", ss.ss_family, str);
1295 goto out;
1296 }
Willy Tarreau5fc93282020-09-16 18:25:03 +02001297 }
1298
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001299 ret = &ss;
Willy Tarreaud5191e72010-02-09 20:50:45 +01001300 out:
Willy Tarreau48ef4c92017-01-06 18:32:38 +01001301 if (port)
1302 *port = porta;
Willy Tarreaud4448bc2013-02-20 15:55:15 +01001303 if (low)
1304 *low = portl;
1305 if (high)
1306 *high = porth;
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001307 if (fd)
1308 *fd = new_fd;
Willy Tarreau5fc93282020-09-16 18:25:03 +02001309 if (proto)
1310 *proto = new_proto;
Willy Tarreau24709282013-03-10 21:32:12 +01001311 free(back);
Willy Tarreaud5191e72010-02-09 20:50:45 +01001312 return ret;
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001313}
1314
Thayne McCombs92149f92020-11-20 01:28:26 -07001315/* converts <addr> and <port> into a string representation of the address and port. This is sort
1316 * of an inverse of str2sa_range, with some restrictions. The supported families are AF_INET,
1317 * AF_INET6, AF_UNIX, and AF_CUST_SOCKPAIR. If the family is unsopported NULL is returned.
1318 * If map_ports is true, then the sign of the port is included in the output, to indicate it is
1319 * relative to the incoming port. AF_INET and AF_INET6 will be in the form "<addr>:<port>".
1320 * AF_UNIX will either be just the path (if using a pathname) or "abns@<path>" if it is abstract.
1321 * AF_CUST_SOCKPAIR will be of the form "sockpair@<fd>".
1322 *
1323 * The returned char* is allocated, and it is the responsibility of the caller to free it.
1324 */
1325char * sa2str(const struct sockaddr_storage *addr, int port, int map_ports)
1326{
1327 char buffer[INET6_ADDRSTRLEN];
1328 char *out = NULL;
1329 const void *ptr;
1330 const char *path;
1331
1332 switch (addr->ss_family) {
1333 case AF_INET:
1334 ptr = &((struct sockaddr_in *)addr)->sin_addr;
1335 break;
1336 case AF_INET6:
1337 ptr = &((struct sockaddr_in6 *)addr)->sin6_addr;
1338 break;
1339 case AF_UNIX:
1340 path = ((struct sockaddr_un *)addr)->sun_path;
1341 if (path[0] == '\0') {
1342 const int max_length = sizeof(struct sockaddr_un) - offsetof(struct sockaddr_un, sun_path) - 1;
1343 return memprintf(&out, "abns@%.*s", max_length, path+1);
1344 } else {
1345 return strdup(path);
1346 }
1347 case AF_CUST_SOCKPAIR:
1348 return memprintf(&out, "sockpair@%d", ((struct sockaddr_in *)addr)->sin_addr.s_addr);
1349 default:
1350 return NULL;
1351 }
Tim Duesterhus189c0082022-05-22 13:06:27 +02001352 inet_ntop(addr->ss_family, ptr, buffer, sizeof(buffer));
Thayne McCombs92149f92020-11-20 01:28:26 -07001353 if (map_ports)
1354 return memprintf(&out, "%s:%+d", buffer, port);
1355 else
1356 return memprintf(&out, "%s:%d", buffer, port);
1357}
1358
1359
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001360/* converts <str> to a struct in_addr containing a network mask. It can be
1361 * passed in dotted form (255.255.255.0) or in CIDR form (24). It returns 1
Jarno Huuskonen577d5ac2017-05-21 17:32:21 +03001362 * if the conversion succeeds otherwise zero.
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001363 */
1364int str2mask(const char *str, struct in_addr *mask)
1365{
1366 if (strchr(str, '.') != NULL) { /* dotted notation */
1367 if (!inet_pton(AF_INET, str, mask))
1368 return 0;
1369 }
1370 else { /* mask length */
1371 char *err;
1372 unsigned long len = strtol(str, &err, 10);
1373
1374 if (!*str || (err && *err) || (unsigned)len > 32)
1375 return 0;
Tim Duesterhus8575f722018-01-25 16:24:48 +01001376
1377 len2mask4(len, mask);
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001378 }
1379 return 1;
1380}
1381
Tim Duesterhus47185172018-01-25 16:24:49 +01001382/* converts <str> to a struct in6_addr containing a network mask. It can be
Tim Duesterhus5e642862018-02-20 17:02:18 +01001383 * passed in quadruplet form (ffff:ffff::) or in CIDR form (64). It returns 1
Tim Duesterhus47185172018-01-25 16:24:49 +01001384 * if the conversion succeeds otherwise zero.
1385 */
1386int str2mask6(const char *str, struct in6_addr *mask)
1387{
1388 if (strchr(str, ':') != NULL) { /* quadruplet notation */
1389 if (!inet_pton(AF_INET6, str, mask))
1390 return 0;
1391 }
1392 else { /* mask length */
1393 char *err;
1394 unsigned long len = strtol(str, &err, 10);
1395
1396 if (!*str || (err && *err) || (unsigned)len > 128)
1397 return 0;
1398
1399 len2mask6(len, mask);
1400 }
1401 return 1;
1402}
1403
Thierry FOURNIERb0504632013-12-14 15:39:02 +01001404/* convert <cidr> to struct in_addr <mask>. It returns 1 if the conversion
1405 * succeeds otherwise zero.
1406 */
1407int cidr2dotted(int cidr, struct in_addr *mask) {
1408
1409 if (cidr < 0 || cidr > 32)
1410 return 0;
1411
1412 mask->s_addr = cidr ? htonl(~0UL << (32 - cidr)) : 0;
1413 return 1;
1414}
1415
Thierry Fournier70473a52016-02-17 17:12:14 +01001416/* Convert mask from bit length form to in_addr form.
1417 * This function never fails.
1418 */
1419void len2mask4(int len, struct in_addr *addr)
1420{
1421 if (len >= 32) {
1422 addr->s_addr = 0xffffffff;
1423 return;
1424 }
1425 if (len <= 0) {
1426 addr->s_addr = 0x00000000;
1427 return;
1428 }
1429 addr->s_addr = 0xffffffff << (32 - len);
1430 addr->s_addr = htonl(addr->s_addr);
1431}
1432
1433/* Convert mask from bit length form to in6_addr form.
1434 * This function never fails.
1435 */
1436void len2mask6(int len, struct in6_addr *addr)
1437{
1438 len2mask4(len, (struct in_addr *)&addr->s6_addr[0]); /* msb */
1439 len -= 32;
1440 len2mask4(len, (struct in_addr *)&addr->s6_addr[4]);
1441 len -= 32;
1442 len2mask4(len, (struct in_addr *)&addr->s6_addr[8]);
1443 len -= 32;
1444 len2mask4(len, (struct in_addr *)&addr->s6_addr[12]); /* lsb */
1445}
1446
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001447/*
Willy Tarreaud077a8e2007-05-08 18:28:09 +02001448 * converts <str> to two struct in_addr* which must be pre-allocated.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001449 * The format is "addr[/mask]", where "addr" cannot be empty, and mask
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001450 * is optional and either in the dotted or CIDR notation.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001451 * Note: "addr" can also be a hostname. Returns 1 if OK, 0 if error.
1452 */
Thierry FOURNIERfc7ac7b2014-02-11 15:23:04 +01001453int str2net(const char *str, int resolve, struct in_addr *addr, struct in_addr *mask)
Willy Tarreaubaaee002006-06-26 02:48:02 +02001454{
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001455 __label__ out_free, out_err;
1456 char *c, *s;
1457 int ret_val;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001458
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001459 s = strdup(str);
1460 if (!s)
1461 return 0;
1462
Willy Tarreaubaaee002006-06-26 02:48:02 +02001463 memset(mask, 0, sizeof(*mask));
1464 memset(addr, 0, sizeof(*addr));
Willy Tarreaubaaee002006-06-26 02:48:02 +02001465
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001466 if ((c = strrchr(s, '/')) != NULL) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02001467 *c++ = '\0';
1468 /* c points to the mask */
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001469 if (!str2mask(c, mask))
1470 goto out_err;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001471 }
1472 else {
Willy Tarreauebd61602006-12-30 11:54:15 +01001473 mask->s_addr = ~0U;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001474 }
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001475 if (!inet_pton(AF_INET, s, addr)) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02001476 struct hostent *he;
1477
Thierry FOURNIERfc7ac7b2014-02-11 15:23:04 +01001478 if (!resolve)
1479 goto out_err;
1480
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001481 if ((he = gethostbyname(s)) == NULL) {
1482 goto out_err;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001483 }
1484 else
1485 *addr = *(struct in_addr *) *(he->h_addr_list);
1486 }
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001487
1488 ret_val = 1;
1489 out_free:
1490 free(s);
1491 return ret_val;
1492 out_err:
1493 ret_val = 0;
1494 goto out_free;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001495}
1496
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001497
1498/*
Willy Tarreau6d20e282012-04-27 22:49:47 +02001499 * converts <str> to two struct in6_addr* which must be pre-allocated.
1500 * The format is "addr[/mask]", where "addr" cannot be empty, and mask
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001501 * is an optional number of bits (128 being the default).
Willy Tarreau6d20e282012-04-27 22:49:47 +02001502 * Returns 1 if OK, 0 if error.
1503 */
1504int str62net(const char *str, struct in6_addr *addr, unsigned char *mask)
1505{
1506 char *c, *s;
1507 int ret_val = 0;
1508 char *err;
1509 unsigned long len = 128;
1510
1511 s = strdup(str);
1512 if (!s)
1513 return 0;
1514
1515 memset(mask, 0, sizeof(*mask));
1516 memset(addr, 0, sizeof(*addr));
1517
1518 if ((c = strrchr(s, '/')) != NULL) {
1519 *c++ = '\0'; /* c points to the mask */
1520 if (!*c)
1521 goto out_free;
1522
1523 len = strtoul(c, &err, 10);
1524 if ((err && *err) || (unsigned)len > 128)
1525 goto out_free;
1526 }
1527 *mask = len; /* OK we have a valid mask in <len> */
1528
1529 if (!inet_pton(AF_INET6, s, addr))
1530 goto out_free;
1531
1532 ret_val = 1;
1533 out_free:
1534 free(s);
1535 return ret_val;
1536}
1537
1538
1539/*
Willy Tarreau12e10272021-03-25 11:34:40 +01001540 * Parse IPv4 address found in url. Return the number of bytes parsed. It
1541 * expects exactly 4 numbers between 0 and 255 delimited by dots, and returns
1542 * zero in case of mismatch.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001543 */
David du Colombier6f5ccb12011-03-10 22:26:24 +01001544int url2ipv4(const char *addr, struct in_addr *dst)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001545{
1546 int saw_digit, octets, ch;
1547 u_char tmp[4], *tp;
1548 const char *cp = addr;
1549
1550 saw_digit = 0;
1551 octets = 0;
1552 *(tp = tmp) = 0;
1553
1554 while (*addr) {
Willy Tarreau12e10272021-03-25 11:34:40 +01001555 unsigned char digit = (ch = *addr) - '0';
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001556 if (digit > 9 && ch != '.')
1557 break;
Willy Tarreau12e10272021-03-25 11:34:40 +01001558 addr++;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001559 if (digit <= 9) {
1560 u_int new = *tp * 10 + digit;
1561 if (new > 255)
1562 return 0;
1563 *tp = new;
1564 if (!saw_digit) {
1565 if (++octets > 4)
1566 return 0;
1567 saw_digit = 1;
1568 }
1569 } else if (ch == '.' && saw_digit) {
1570 if (octets == 4)
1571 return 0;
1572 *++tp = 0;
1573 saw_digit = 0;
1574 } else
1575 return 0;
1576 }
1577
1578 if (octets < 4)
1579 return 0;
1580
1581 memcpy(&dst->s_addr, tmp, 4);
Willy Tarreau12e10272021-03-25 11:34:40 +01001582 return addr - cp;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001583}
1584
1585/*
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001586 * Resolve destination server from URL. Convert <str> to a sockaddr_storage.
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001587 * <out> contain the code of the detected scheme, the start and length of
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001588 * the hostname. Actually only http and https are supported. <out> can be NULL.
1589 * This function returns the consumed length. It is useful if you parse complete
1590 * url like http://host:port/path, because the consumed length corresponds to
1591 * the first character of the path. If the conversion fails, it returns -1.
1592 *
1593 * This function tries to resolve the DNS name if haproxy is in starting mode.
1594 * So, this function may be used during the configuration parsing.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001595 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001596int url2sa(const char *url, int ulen, struct sockaddr_storage *addr, struct split_url *out)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001597{
1598 const char *curr = url, *cp = url;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001599 const char *end;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001600 int ret, url_code = 0;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001601 unsigned long long int http_code = 0;
1602 int default_port;
1603 struct hostent *he;
1604 char *p;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001605
1606 /* Firstly, try to find :// pattern */
1607 while (curr < url+ulen && url_code != 0x3a2f2f) {
1608 url_code = ((url_code & 0xffff) << 8);
1609 url_code += (unsigned char)*curr++;
1610 }
1611
1612 /* Secondly, if :// pattern is found, verify parsed stuff
1613 * before pattern is matching our http pattern.
1614 * If so parse ip address and port in uri.
1615 *
1616 * WARNING: Current code doesn't support dynamic async dns resolver.
1617 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001618 if (url_code != 0x3a2f2f)
1619 return -1;
1620
1621 /* Copy scheme, and utrn to lower case. */
1622 while (cp < curr - 3)
1623 http_code = (http_code << 8) + *cp++;
1624 http_code |= 0x2020202020202020ULL; /* Turn everything to lower case */
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001625
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001626 /* HTTP or HTTPS url matching */
1627 if (http_code == 0x2020202068747470ULL) {
1628 default_port = 80;
1629 if (out)
1630 out->scheme = SCH_HTTP;
1631 }
1632 else if (http_code == 0x2020206874747073ULL) {
1633 default_port = 443;
1634 if (out)
1635 out->scheme = SCH_HTTPS;
1636 }
1637 else
1638 return -1;
1639
1640 /* If the next char is '[', the host address is IPv6. */
1641 if (*curr == '[') {
1642 curr++;
1643
1644 /* Check trash size */
1645 if (trash.size < ulen)
1646 return -1;
1647
1648 /* Look for ']' and copy the address in a trash buffer. */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001649 p = trash.area;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001650 for (end = curr;
1651 end < url + ulen && *end != ']';
1652 end++, p++)
1653 *p = *end;
1654 if (*end != ']')
1655 return -1;
1656 *p = '\0';
1657
1658 /* Update out. */
1659 if (out) {
1660 out->host = curr;
1661 out->host_len = end - curr;
1662 }
1663
1664 /* Try IPv6 decoding. */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001665 if (!inet_pton(AF_INET6, trash.area, &((struct sockaddr_in6 *)addr)->sin6_addr))
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001666 return -1;
1667 end++;
1668
1669 /* Decode port. */
William Lallemande2e6cd92022-03-25 17:37:51 +01001670 if (end < url + ulen && *end == ':') {
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001671 end++;
1672 default_port = read_uint(&end, url + ulen);
1673 }
1674 ((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1675 ((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1676 return end - url;
1677 }
1678 else {
William Lallemand12e3b252022-02-18 16:13:12 +01001679 /* we need to copy the string into the trash because url2ipv4
1680 * needs a \0 at the end of the string */
1681 if (trash.size < ulen)
1682 return -1;
1683
1684 memcpy(trash.area, curr, ulen - (curr - url));
1685 trash.area[ulen - (curr - url)] = '\0';
1686
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001687 /* We are looking for IP address. If you want to parse and
1688 * resolve hostname found in url, you can use str2sa_range(), but
1689 * be warned this can slow down global daemon performances
1690 * while handling lagging dns responses.
1691 */
William Lallemand12e3b252022-02-18 16:13:12 +01001692 ret = url2ipv4(trash.area, &((struct sockaddr_in *)addr)->sin_addr);
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001693 if (ret) {
1694 /* Update out. */
1695 if (out) {
1696 out->host = curr;
1697 out->host_len = ret;
1698 }
1699
William Lallemand086323a2022-03-24 21:59:03 +01001700 curr += ret;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001701
1702 /* Decode port. */
William Lallemande2e6cd92022-03-25 17:37:51 +01001703 if (curr < url + ulen && *curr == ':') {
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001704 curr++;
1705 default_port = read_uint(&curr, url + ulen);
1706 }
1707 ((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1708
1709 /* Set family. */
1710 ((struct sockaddr_in *)addr)->sin_family = AF_INET;
1711 return curr - url;
1712 }
1713 else if (global.mode & MODE_STARTING) {
1714 /* The IPv4 and IPv6 decoding fails, maybe the url contain name. Try to execute
1715 * synchronous DNS request only if HAProxy is in the start state.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001716 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001717
1718 /* look for : or / or end */
1719 for (end = curr;
1720 end < url + ulen && *end != '/' && *end != ':';
1721 end++);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001722 memcpy(trash.area, curr, end - curr);
1723 trash.area[end - curr] = '\0';
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001724
1725 /* try to resolve an IPv4/IPv6 hostname */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001726 he = gethostbyname(trash.area);
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001727 if (!he)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001728 return -1;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001729
1730 /* Update out. */
1731 if (out) {
1732 out->host = curr;
1733 out->host_len = end - curr;
1734 }
1735
1736 /* Decode port. */
William Lallemande2e6cd92022-03-25 17:37:51 +01001737 if (end < url + ulen && *end == ':') {
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001738 end++;
1739 default_port = read_uint(&end, url + ulen);
1740 }
1741
1742 /* Copy IP address, set port and family. */
1743 switch (he->h_addrtype) {
1744 case AF_INET:
1745 ((struct sockaddr_in *)addr)->sin_addr = *(struct in_addr *) *(he->h_addr_list);
1746 ((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1747 ((struct sockaddr_in *)addr)->sin_family = AF_INET;
1748 return end - url;
1749
1750 case AF_INET6:
1751 ((struct sockaddr_in6 *)addr)->sin6_addr = *(struct in6_addr *) *(he->h_addr_list);
1752 ((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1753 ((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1754 return end - url;
1755 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001756 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001757 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001758 return -1;
1759}
1760
Willy Tarreau631f01c2011-09-05 00:36:48 +02001761/* Tries to convert a sockaddr_storage address to text form. Upon success, the
1762 * address family is returned so that it's easy for the caller to adapt to the
1763 * output format. Zero is returned if the address family is not supported. -1
1764 * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1765 * supported.
1766 */
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001767int addr_to_str(const struct sockaddr_storage *addr, char *str, int size)
Willy Tarreau631f01c2011-09-05 00:36:48 +02001768{
1769
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001770 const void *ptr;
Willy Tarreau631f01c2011-09-05 00:36:48 +02001771
1772 if (size < 5)
1773 return 0;
1774 *str = '\0';
1775
1776 switch (addr->ss_family) {
1777 case AF_INET:
1778 ptr = &((struct sockaddr_in *)addr)->sin_addr;
1779 break;
1780 case AF_INET6:
1781 ptr = &((struct sockaddr_in6 *)addr)->sin6_addr;
1782 break;
1783 case AF_UNIX:
1784 memcpy(str, "unix", 5);
1785 return addr->ss_family;
1786 default:
1787 return 0;
1788 }
1789
1790 if (inet_ntop(addr->ss_family, ptr, str, size))
1791 return addr->ss_family;
1792
1793 /* failed */
1794 return -1;
1795}
1796
Simon Horman75ab8bd2014-06-16 09:39:41 +09001797/* Tries to convert a sockaddr_storage port to text form. Upon success, the
1798 * address family is returned so that it's easy for the caller to adapt to the
1799 * output format. Zero is returned if the address family is not supported. -1
1800 * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1801 * supported.
1802 */
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001803int port_to_str(const struct sockaddr_storage *addr, char *str, int size)
Simon Horman75ab8bd2014-06-16 09:39:41 +09001804{
1805
1806 uint16_t port;
1807
1808
Willy Tarreaud7dad1b2017-01-06 16:46:22 +01001809 if (size < 6)
Simon Horman75ab8bd2014-06-16 09:39:41 +09001810 return 0;
1811 *str = '\0';
1812
1813 switch (addr->ss_family) {
1814 case AF_INET:
1815 port = ((struct sockaddr_in *)addr)->sin_port;
1816 break;
1817 case AF_INET6:
1818 port = ((struct sockaddr_in6 *)addr)->sin6_port;
1819 break;
1820 case AF_UNIX:
1821 memcpy(str, "unix", 5);
1822 return addr->ss_family;
1823 default:
1824 return 0;
1825 }
1826
1827 snprintf(str, size, "%u", ntohs(port));
1828 return addr->ss_family;
1829}
1830
Willy Tarreau16e01562016-08-09 16:46:18 +02001831/* check if the given address is local to the system or not. It will return
1832 * -1 when it's not possible to know, 0 when the address is not local, 1 when
1833 * it is. We don't want to iterate over all interfaces for this (and it is not
1834 * portable). So instead we try to bind in UDP to this address on a free non
1835 * privileged port and to connect to the same address, port 0 (connect doesn't
1836 * care). If it succeeds, we own the address. Note that non-inet addresses are
1837 * considered local since they're most likely AF_UNIX.
1838 */
1839int addr_is_local(const struct netns_entry *ns,
1840 const struct sockaddr_storage *orig)
1841{
1842 struct sockaddr_storage addr;
1843 int result;
1844 int fd;
1845
1846 if (!is_inet_addr(orig))
1847 return 1;
1848
1849 memcpy(&addr, orig, sizeof(addr));
1850 set_host_port(&addr, 0);
1851
1852 fd = my_socketat(ns, addr.ss_family, SOCK_DGRAM, IPPROTO_UDP);
1853 if (fd < 0)
1854 return -1;
1855
1856 result = -1;
1857 if (bind(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == 0) {
1858 if (connect(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == -1)
1859 result = 0; // fail, non-local address
1860 else
1861 result = 1; // success, local address
1862 }
1863 else {
1864 if (errno == EADDRNOTAVAIL)
1865 result = 0; // definitely not local :-)
1866 }
1867 close(fd);
1868
1869 return result;
1870}
1871
Willy Tarreaubaaee002006-06-26 02:48:02 +02001872/* will try to encode the string <string> replacing all characters tagged in
1873 * <map> with the hexadecimal representation of their ASCII-code (2 digits)
1874 * prefixed by <escape>, and will store the result between <start> (included)
1875 * and <stop> (excluded), and will always terminate the string with a '\0'
1876 * before <stop>. The position of the '\0' is returned if the conversion
1877 * completes. If bytes are missing between <start> and <stop>, then the
1878 * conversion will be incomplete and truncated. If <stop> <= <start>, the '\0'
1879 * cannot even be stored so we return <start> without writing the 0.
1880 * The input string must also be zero-terminated.
1881 */
1882const char hextab[16] = "0123456789ABCDEF";
1883char *encode_string(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001884 const char escape, const long *map,
Willy Tarreaubaaee002006-06-26 02:48:02 +02001885 const char *string)
1886{
1887 if (start < stop) {
1888 stop--; /* reserve one byte for the final '\0' */
1889 while (start < stop && *string != '\0') {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001890 if (!ha_bit_test((unsigned char)(*string), map))
Willy Tarreaubaaee002006-06-26 02:48:02 +02001891 *start++ = *string;
1892 else {
1893 if (start + 3 >= stop)
1894 break;
1895 *start++ = escape;
1896 *start++ = hextab[(*string >> 4) & 15];
1897 *start++ = hextab[*string & 15];
1898 }
1899 string++;
1900 }
1901 *start = '\0';
1902 }
1903 return start;
1904}
1905
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001906/*
1907 * Same behavior as encode_string() above, except that it encodes chunk
1908 * <chunk> instead of a string.
1909 */
1910char *encode_chunk(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001911 const char escape, const long *map,
Willy Tarreau83061a82018-07-13 11:56:34 +02001912 const struct buffer *chunk)
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001913{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001914 char *str = chunk->area;
1915 char *end = chunk->area + chunk->data;
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001916
1917 if (start < stop) {
1918 stop--; /* reserve one byte for the final '\0' */
1919 while (start < stop && str < end) {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001920 if (!ha_bit_test((unsigned char)(*str), map))
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001921 *start++ = *str;
1922 else {
1923 if (start + 3 >= stop)
1924 break;
1925 *start++ = escape;
1926 *start++ = hextab[(*str >> 4) & 15];
1927 *start++ = hextab[*str & 15];
1928 }
1929 str++;
1930 }
1931 *start = '\0';
1932 }
1933 return start;
1934}
1935
Dragan Dosen0edd1092016-02-12 13:23:02 +01001936/*
1937 * Tries to prefix characters tagged in the <map> with the <escape>
Aurelien DARRAGONa5c621a2022-09-20 14:33:18 +02001938 * character. The input <string> is processed until string_stop
1939 * is reached or NULL-byte is encountered. The result will
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001940 * be stored between <start> (included) and <stop> (excluded). This
1941 * function will always try to terminate the resulting string with a '\0'
1942 * before <stop>, and will return its position if the conversion
1943 * completes.
1944 */
1945char *escape_string(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001946 const char escape, const long *map,
Aurelien DARRAGONa5c621a2022-09-20 14:33:18 +02001947 const char *string, const char *string_stop)
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001948{
1949 if (start < stop) {
1950 stop--; /* reserve one byte for the final '\0' */
Aurelien DARRAGONa5c621a2022-09-20 14:33:18 +02001951 while (start < stop && string < string_stop && *string != '\0') {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001952 if (!ha_bit_test((unsigned char)(*string), map))
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001953 *start++ = *string;
1954 else {
1955 if (start + 2 >= stop)
1956 break;
1957 *start++ = escape;
1958 *start++ = *string;
1959 }
1960 string++;
1961 }
1962 *start = '\0';
1963 }
1964 return start;
1965}
1966
1967/*
1968 * Tries to prefix characters tagged in the <map> with the <escape>
Dragan Dosen0edd1092016-02-12 13:23:02 +01001969 * character. <chunk> contains the input to be escaped. The result will be
1970 * stored between <start> (included) and <stop> (excluded). The function
1971 * will always try to terminate the resulting string with a '\0' before
1972 * <stop>, and will return its position if the conversion completes.
1973 */
1974char *escape_chunk(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001975 const char escape, const long *map,
Willy Tarreau83061a82018-07-13 11:56:34 +02001976 const struct buffer *chunk)
Dragan Dosen0edd1092016-02-12 13:23:02 +01001977{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001978 char *str = chunk->area;
1979 char *end = chunk->area + chunk->data;
Dragan Dosen0edd1092016-02-12 13:23:02 +01001980
1981 if (start < stop) {
1982 stop--; /* reserve one byte for the final '\0' */
1983 while (start < stop && str < end) {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001984 if (!ha_bit_test((unsigned char)(*str), map))
Dragan Dosen0edd1092016-02-12 13:23:02 +01001985 *start++ = *str;
1986 else {
1987 if (start + 2 >= stop)
1988 break;
1989 *start++ = escape;
1990 *start++ = *str;
1991 }
1992 str++;
1993 }
1994 *start = '\0';
1995 }
1996 return start;
1997}
1998
Thierry FOURNIERddea6262015-05-28 16:00:28 +02001999/* Check a string for using it in a CSV output format. If the string contains
2000 * one of the following four char <">, <,>, CR or LF, the string is
2001 * encapsulated between <"> and the <"> are escaped by a <""> sequence.
2002 * <str> is the input string to be escaped. The function assumes that
2003 * the input string is null-terminated.
2004 *
2005 * If <quote> is 0, the result is returned escaped but without double quote.
Willy Tarreau898529b2016-01-06 18:07:04 +01002006 * It is useful if the escaped string is used between double quotes in the
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002007 * format.
2008 *
Willy Tarreau898529b2016-01-06 18:07:04 +01002009 * printf("..., \"%s\", ...\r\n", csv_enc(str, 0, &trash));
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002010 *
Willy Tarreaub631c292016-01-08 10:04:08 +01002011 * If <quote> is 1, the converter puts the quotes only if any reserved character
2012 * is present. If <quote> is 2, the converter always puts the quotes.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002013 *
Willy Tarreau83061a82018-07-13 11:56:34 +02002014 * <output> is a struct buffer used for storing the output string.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002015 *
Willy Tarreau898529b2016-01-06 18:07:04 +01002016 * The function returns the converted string on its output. If an error
2017 * occurs, the function returns an empty string. This type of output is useful
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002018 * for using the function directly as printf() argument.
2019 *
2020 * If the output buffer is too short to contain the input string, the result
2021 * is truncated.
Willy Tarreau898529b2016-01-06 18:07:04 +01002022 *
Willy Tarreaub631c292016-01-08 10:04:08 +01002023 * This function appends the encoding to the existing output chunk, and it
2024 * guarantees that it starts immediately at the first available character of
2025 * the chunk. Please use csv_enc() instead if you want to replace the output
2026 * chunk.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002027 */
Willy Tarreau83061a82018-07-13 11:56:34 +02002028const char *csv_enc_append(const char *str, int quote, struct buffer *output)
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002029{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002030 char *end = output->area + output->size;
2031 char *out = output->area + output->data;
Willy Tarreau898529b2016-01-06 18:07:04 +01002032 char *ptr = out;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002033
Willy Tarreaub631c292016-01-08 10:04:08 +01002034 if (quote == 1) {
2035 /* automatic quoting: first verify if we'll have to quote the string */
2036 if (!strpbrk(str, "\n\r,\""))
2037 quote = 0;
2038 }
2039
2040 if (quote)
2041 *ptr++ = '"';
2042
Willy Tarreau898529b2016-01-06 18:07:04 +01002043 while (*str && ptr < end - 2) { /* -2 for reserving space for <"> and \0. */
2044 *ptr = *str;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002045 if (*str == '"') {
Willy Tarreau898529b2016-01-06 18:07:04 +01002046 ptr++;
2047 if (ptr >= end - 2) {
2048 ptr--;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002049 break;
2050 }
Willy Tarreau898529b2016-01-06 18:07:04 +01002051 *ptr = '"';
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002052 }
Willy Tarreau898529b2016-01-06 18:07:04 +01002053 ptr++;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002054 str++;
2055 }
2056
Willy Tarreaub631c292016-01-08 10:04:08 +01002057 if (quote)
2058 *ptr++ = '"';
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002059
Willy Tarreau898529b2016-01-06 18:07:04 +01002060 *ptr = '\0';
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002061 output->data = ptr - output->area;
Willy Tarreau898529b2016-01-06 18:07:04 +01002062 return out;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002063}
2064
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002065/* Decode an URL-encoded string in-place. The resulting string might
2066 * be shorter. If some forbidden characters are found, the conversion is
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002067 * aborted, the string is truncated before the issue and a negative value is
2068 * returned, otherwise the operation returns the length of the decoded string.
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002069 * If the 'in_form' argument is non-nul the string is assumed to be part of
2070 * an "application/x-www-form-urlencoded" encoded string, and the '+' will be
2071 * turned to a space. If it's zero, this will only be done after a question
2072 * mark ('?').
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002073 */
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002074int url_decode(char *string, int in_form)
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002075{
2076 char *in, *out;
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002077 int ret = -1;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002078
2079 in = string;
2080 out = string;
2081 while (*in) {
2082 switch (*in) {
2083 case '+' :
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002084 *out++ = in_form ? ' ' : *in;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002085 break;
2086 case '%' :
2087 if (!ishex(in[1]) || !ishex(in[2]))
2088 goto end;
2089 *out++ = (hex2i(in[1]) << 4) + hex2i(in[2]);
2090 in += 2;
2091 break;
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002092 case '?':
2093 in_form = 1;
2094 /* fall through */
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002095 default:
2096 *out++ = *in;
2097 break;
2098 }
2099 in++;
2100 }
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002101 ret = out - string; /* success */
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002102 end:
2103 *out = 0;
2104 return ret;
2105}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002106
Willy Tarreau6911fa42007-03-04 18:06:08 +01002107unsigned int str2ui(const char *s)
2108{
2109 return __str2ui(s);
2110}
2111
2112unsigned int str2uic(const char *s)
2113{
2114 return __str2uic(s);
2115}
2116
2117unsigned int strl2ui(const char *s, int len)
2118{
2119 return __strl2ui(s, len);
2120}
2121
2122unsigned int strl2uic(const char *s, int len)
2123{
2124 return __strl2uic(s, len);
2125}
2126
Willy Tarreau4ec83cd2010-10-15 23:19:55 +02002127unsigned int read_uint(const char **s, const char *end)
2128{
2129 return __read_uint(s, end);
2130}
2131
Thierry FOURNIER763a5d82015-07-06 23:09:52 +02002132/* This function reads an unsigned integer from the string pointed to by <s> and
2133 * returns it. The <s> pointer is adjusted to point to the first unread char. The
2134 * function automatically stops at <end>. If the number overflows, the 2^64-1
2135 * value is returned.
2136 */
2137unsigned long long int read_uint64(const char **s, const char *end)
2138{
2139 const char *ptr = *s;
2140 unsigned long long int i = 0, tmp;
2141 unsigned int j;
2142
2143 while (ptr < end) {
2144
2145 /* read next char */
2146 j = *ptr - '0';
2147 if (j > 9)
2148 goto read_uint64_end;
2149
2150 /* add char to the number and check overflow. */
2151 tmp = i * 10;
2152 if (tmp / 10 != i) {
2153 i = ULLONG_MAX;
2154 goto read_uint64_eat;
2155 }
2156 if (ULLONG_MAX - tmp < j) {
2157 i = ULLONG_MAX;
2158 goto read_uint64_eat;
2159 }
2160 i = tmp + j;
2161 ptr++;
2162 }
2163read_uint64_eat:
2164 /* eat each numeric char */
2165 while (ptr < end) {
2166 if ((unsigned int)(*ptr - '0') > 9)
2167 break;
2168 ptr++;
2169 }
2170read_uint64_end:
2171 *s = ptr;
2172 return i;
2173}
2174
2175/* This function reads an integer from the string pointed to by <s> and returns
2176 * it. The <s> pointer is adjusted to point to the first unread char. The function
2177 * automatically stops at <end>. Il the number is bigger than 2^63-2, the 2^63-1
2178 * value is returned. If the number is lowest than -2^63-1, the -2^63 value is
2179 * returned.
2180 */
2181long long int read_int64(const char **s, const char *end)
2182{
2183 unsigned long long int i = 0;
2184 int neg = 0;
2185
2186 /* Look for minus char. */
2187 if (**s == '-') {
2188 neg = 1;
2189 (*s)++;
2190 }
2191 else if (**s == '+')
2192 (*s)++;
2193
2194 /* convert as positive number. */
2195 i = read_uint64(s, end);
2196
2197 if (neg) {
2198 if (i > 0x8000000000000000ULL)
2199 return LLONG_MIN;
2200 return -i;
2201 }
2202 if (i > 0x7fffffffffffffffULL)
2203 return LLONG_MAX;
2204 return i;
2205}
2206
Willy Tarreau6911fa42007-03-04 18:06:08 +01002207/* This one is 7 times faster than strtol() on athlon with checks.
2208 * It returns the value of the number composed of all valid digits read,
2209 * and can process negative numbers too.
2210 */
2211int strl2ic(const char *s, int len)
2212{
2213 int i = 0;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002214 int j, k;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002215
2216 if (len > 0) {
2217 if (*s != '-') {
2218 /* positive number */
2219 while (len-- > 0) {
2220 j = (*s++) - '0';
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002221 k = i * 10;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002222 if (j > 9)
2223 break;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002224 i = k + j;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002225 }
2226 } else {
2227 /* negative number */
2228 s++;
2229 while (--len > 0) {
2230 j = (*s++) - '0';
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002231 k = i * 10;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002232 if (j > 9)
2233 break;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002234 i = k - j;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002235 }
2236 }
2237 }
2238 return i;
2239}
2240
2241
2242/* This function reads exactly <len> chars from <s> and converts them to a
2243 * signed integer which it stores into <ret>. It accurately detects any error
2244 * (truncated string, invalid chars, overflows). It is meant to be used in
2245 * applications designed for hostile environments. It returns zero when the
2246 * number has successfully been converted, non-zero otherwise. When an error
2247 * is returned, the <ret> value is left untouched. It is yet 5 to 40 times
2248 * faster than strtol().
2249 */
2250int strl2irc(const char *s, int len, int *ret)
2251{
2252 int i = 0;
2253 int j;
2254
2255 if (!len)
2256 return 1;
2257
2258 if (*s != '-') {
2259 /* positive number */
2260 while (len-- > 0) {
2261 j = (*s++) - '0';
2262 if (j > 9) return 1; /* invalid char */
2263 if (i > INT_MAX / 10) return 1; /* check for multiply overflow */
2264 i = i * 10;
2265 if (i + j < i) return 1; /* check for addition overflow */
2266 i = i + j;
2267 }
2268 } else {
2269 /* negative number */
2270 s++;
2271 while (--len > 0) {
2272 j = (*s++) - '0';
2273 if (j > 9) return 1; /* invalid char */
2274 if (i < INT_MIN / 10) return 1; /* check for multiply overflow */
2275 i = i * 10;
2276 if (i - j > i) return 1; /* check for subtract overflow */
2277 i = i - j;
2278 }
2279 }
2280 *ret = i;
2281 return 0;
2282}
2283
2284
2285/* This function reads exactly <len> chars from <s> and converts them to a
2286 * signed integer which it stores into <ret>. It accurately detects any error
2287 * (truncated string, invalid chars, overflows). It is meant to be used in
2288 * applications designed for hostile environments. It returns zero when the
2289 * number has successfully been converted, non-zero otherwise. When an error
2290 * is returned, the <ret> value is left untouched. It is about 3 times slower
William Dauchy060ffc82021-02-06 20:47:51 +01002291 * than strl2irc().
Willy Tarreau6911fa42007-03-04 18:06:08 +01002292 */
Willy Tarreau6911fa42007-03-04 18:06:08 +01002293
2294int strl2llrc(const char *s, int len, long long *ret)
2295{
2296 long long i = 0;
2297 int j;
2298
2299 if (!len)
2300 return 1;
2301
2302 if (*s != '-') {
2303 /* positive number */
2304 while (len-- > 0) {
2305 j = (*s++) - '0';
2306 if (j > 9) return 1; /* invalid char */
2307 if (i > LLONG_MAX / 10LL) return 1; /* check for multiply overflow */
2308 i = i * 10LL;
2309 if (i + j < i) return 1; /* check for addition overflow */
2310 i = i + j;
2311 }
2312 } else {
2313 /* negative number */
2314 s++;
2315 while (--len > 0) {
2316 j = (*s++) - '0';
2317 if (j > 9) return 1; /* invalid char */
2318 if (i < LLONG_MIN / 10LL) return 1; /* check for multiply overflow */
2319 i = i * 10LL;
2320 if (i - j > i) return 1; /* check for subtract overflow */
2321 i = i - j;
2322 }
2323 }
2324 *ret = i;
2325 return 0;
2326}
2327
Thierry FOURNIER511e9472014-01-23 17:40:34 +01002328/* This function is used with pat_parse_dotted_ver(). It converts a string
2329 * composed by two number separated by a dot. Each part must contain in 16 bits
2330 * because internally they will be represented as a 32-bit quantity stored in
2331 * a 64-bit integer. It returns zero when the number has successfully been
2332 * converted, non-zero otherwise. When an error is returned, the <ret> value
2333 * is left untouched.
2334 *
2335 * "1.3" -> 0x0000000000010003
2336 * "65535.65535" -> 0x00000000ffffffff
2337 */
2338int strl2llrc_dotted(const char *text, int len, long long *ret)
2339{
2340 const char *end = &text[len];
2341 const char *p;
2342 long long major, minor;
2343
2344 /* Look for dot. */
2345 for (p = text; p < end; p++)
2346 if (*p == '.')
2347 break;
2348
2349 /* Convert major. */
2350 if (strl2llrc(text, p - text, &major) != 0)
2351 return 1;
2352
2353 /* Check major. */
2354 if (major >= 65536)
2355 return 1;
2356
2357 /* Convert minor. */
2358 minor = 0;
2359 if (p < end)
2360 if (strl2llrc(p + 1, end - (p + 1), &minor) != 0)
2361 return 1;
2362
2363 /* Check minor. */
2364 if (minor >= 65536)
2365 return 1;
2366
2367 /* Compose value. */
2368 *ret = (major << 16) | (minor & 0xffff);
2369 return 0;
2370}
2371
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002372/* This function parses a time value optionally followed by a unit suffix among
2373 * "d", "h", "m", "s", "ms" or "us". It converts the value into the unit
2374 * expected by the caller. The computation does its best to avoid overflows.
2375 * The value is returned in <ret> if everything is fine, and a NULL is returned
2376 * by the function. In case of error, a pointer to the error is returned and
2377 * <ret> is left untouched. Values are automatically rounded up when needed.
Willy Tarreau9faebe32019-06-07 19:00:37 +02002378 * Values resulting in values larger than or equal to 2^31 after conversion are
2379 * reported as an overflow as value PARSE_TIME_OVER. Non-null values resulting
2380 * in an underflow are reported as an underflow as value PARSE_TIME_UNDER.
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002381 */
2382const char *parse_time_err(const char *text, unsigned *ret, unsigned unit_flags)
2383{
Willy Tarreau9faebe32019-06-07 19:00:37 +02002384 unsigned long long imult, idiv;
2385 unsigned long long omult, odiv;
2386 unsigned long long value, result;
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002387 const char *str = text;
2388
2389 if (!isdigit((unsigned char)*text))
2390 return text;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002391
2392 omult = odiv = 1;
2393
2394 switch (unit_flags & TIME_UNIT_MASK) {
2395 case TIME_UNIT_US: omult = 1000000; break;
2396 case TIME_UNIT_MS: omult = 1000; break;
2397 case TIME_UNIT_S: break;
2398 case TIME_UNIT_MIN: odiv = 60; break;
2399 case TIME_UNIT_HOUR: odiv = 3600; break;
2400 case TIME_UNIT_DAY: odiv = 86400; break;
2401 default: break;
2402 }
2403
2404 value = 0;
2405
2406 while (1) {
2407 unsigned int j;
2408
2409 j = *text - '0';
2410 if (j > 9)
2411 break;
2412 text++;
2413 value *= 10;
2414 value += j;
2415 }
2416
2417 imult = idiv = 1;
2418 switch (*text) {
2419 case '\0': /* no unit = default unit */
2420 imult = omult = idiv = odiv = 1;
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002421 goto end;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002422 case 's': /* second = unscaled unit */
2423 break;
2424 case 'u': /* microsecond : "us" */
2425 if (text[1] == 's') {
2426 idiv = 1000000;
2427 text++;
Thayne McCombsa6838052021-04-02 14:12:43 -06002428 break;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002429 }
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002430 return text;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002431 case 'm': /* millisecond : "ms" or minute: "m" */
2432 if (text[1] == 's') {
2433 idiv = 1000;
2434 text++;
2435 } else
2436 imult = 60;
2437 break;
2438 case 'h': /* hour : "h" */
2439 imult = 3600;
2440 break;
2441 case 'd': /* day : "d" */
2442 imult = 86400;
2443 break;
2444 default:
2445 return text;
2446 break;
2447 }
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002448 if (*(++text) != '\0') {
2449 ha_warning("unexpected character '%c' after the timer value '%s', only "
2450 "(us=microseconds,ms=milliseconds,s=seconds,m=minutes,h=hours,d=days) are supported."
2451 " This will be reported as an error in next versions.\n", *text, str);
2452 }
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002453
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002454 end:
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002455 if (omult % idiv == 0) { omult /= idiv; idiv = 1; }
2456 if (idiv % omult == 0) { idiv /= omult; omult = 1; }
2457 if (imult % odiv == 0) { imult /= odiv; odiv = 1; }
2458 if (odiv % imult == 0) { odiv /= imult; imult = 1; }
2459
Willy Tarreau9faebe32019-06-07 19:00:37 +02002460 result = (value * (imult * omult) + (idiv * odiv - 1)) / (idiv * odiv);
2461 if (result >= 0x80000000)
2462 return PARSE_TIME_OVER;
2463 if (!result && value)
2464 return PARSE_TIME_UNDER;
2465 *ret = result;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002466 return NULL;
2467}
Willy Tarreau6911fa42007-03-04 18:06:08 +01002468
Emeric Brun39132b22010-01-04 14:57:24 +01002469/* this function converts the string starting at <text> to an unsigned int
2470 * stored in <ret>. If an error is detected, the pointer to the unexpected
Joseph Herlant32b83272018-11-15 11:58:28 -08002471 * character is returned. If the conversion is successful, NULL is returned.
Emeric Brun39132b22010-01-04 14:57:24 +01002472 */
2473const char *parse_size_err(const char *text, unsigned *ret) {
2474 unsigned value = 0;
2475
Christopher Faulet82635a02020-12-11 09:30:45 +01002476 if (!isdigit((unsigned char)*text))
2477 return text;
2478
Emeric Brun39132b22010-01-04 14:57:24 +01002479 while (1) {
2480 unsigned int j;
2481
2482 j = *text - '0';
2483 if (j > 9)
2484 break;
2485 if (value > ~0U / 10)
2486 return text;
2487 value *= 10;
2488 if (value > (value + j))
2489 return text;
2490 value += j;
2491 text++;
2492 }
2493
2494 switch (*text) {
2495 case '\0':
2496 break;
2497 case 'K':
2498 case 'k':
2499 if (value > ~0U >> 10)
2500 return text;
2501 value = value << 10;
2502 break;
2503 case 'M':
2504 case 'm':
2505 if (value > ~0U >> 20)
2506 return text;
2507 value = value << 20;
2508 break;
2509 case 'G':
2510 case 'g':
2511 if (value > ~0U >> 30)
2512 return text;
2513 value = value << 30;
2514 break;
2515 default:
2516 return text;
2517 }
2518
Godbach58048a22015-01-28 17:36:16 +08002519 if (*text != '\0' && *++text != '\0')
2520 return text;
2521
Emeric Brun39132b22010-01-04 14:57:24 +01002522 *ret = value;
2523 return NULL;
2524}
2525
Willy Tarreau126d4062013-12-03 17:50:47 +01002526/*
2527 * Parse binary string written in hexadecimal (source) and store the decoded
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002528 * result into binstr and set binstrlen to the length of binstr. Memory for
Willy Tarreau126d4062013-12-03 17:50:47 +01002529 * binstr is allocated by the function. In case of error, returns 0 with an
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002530 * error message in err. In success case, it returns the consumed length.
Willy Tarreau126d4062013-12-03 17:50:47 +01002531 */
2532int parse_binary(const char *source, char **binstr, int *binstrlen, char **err)
2533{
2534 int len;
2535 const char *p = source;
2536 int i,j;
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002537 int alloc;
Willy Tarreau126d4062013-12-03 17:50:47 +01002538
2539 len = strlen(source);
2540 if (len % 2) {
2541 memprintf(err, "an even number of hex digit is expected");
2542 return 0;
2543 }
2544
2545 len = len >> 1;
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002546
Willy Tarreau126d4062013-12-03 17:50:47 +01002547 if (!*binstr) {
Tim Duesterhuse52b6e52020-09-12 20:26:43 +02002548 *binstr = calloc(len, sizeof(**binstr));
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002549 if (!*binstr) {
2550 memprintf(err, "out of memory while loading string pattern");
2551 return 0;
2552 }
2553 alloc = 1;
Willy Tarreau126d4062013-12-03 17:50:47 +01002554 }
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002555 else {
2556 if (*binstrlen < len) {
Joseph Herlant76dbe782018-11-15 12:01:22 -08002557 memprintf(err, "no space available in the buffer. expect %d, provides %d",
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002558 len, *binstrlen);
2559 return 0;
2560 }
2561 alloc = 0;
2562 }
2563 *binstrlen = len;
Willy Tarreau126d4062013-12-03 17:50:47 +01002564
2565 i = j = 0;
2566 while (j < len) {
2567 if (!ishex(p[i++]))
2568 goto bad_input;
2569 if (!ishex(p[i++]))
2570 goto bad_input;
2571 (*binstr)[j++] = (hex2i(p[i-2]) << 4) + hex2i(p[i-1]);
2572 }
Thierry FOURNIERee330af2014-01-21 11:36:14 +01002573 return len << 1;
Willy Tarreau126d4062013-12-03 17:50:47 +01002574
2575bad_input:
2576 memprintf(err, "an hex digit is expected (found '%c')", p[i-1]);
Willy Tarreau61cfdf42021-02-20 10:46:51 +01002577 if (alloc)
2578 ha_free(binstr);
Willy Tarreau126d4062013-12-03 17:50:47 +01002579 return 0;
2580}
2581
Willy Tarreau946ba592009-05-10 15:41:18 +02002582/* copies at most <n> characters from <src> and always terminates with '\0' */
2583char *my_strndup(const char *src, int n)
2584{
2585 int len = 0;
2586 char *ret;
2587
2588 while (len < n && src[len])
2589 len++;
2590
Vincent Bernat3c2f2f22016-04-03 13:48:42 +02002591 ret = malloc(len + 1);
Willy Tarreau946ba592009-05-10 15:41:18 +02002592 if (!ret)
2593 return ret;
2594 memcpy(ret, src, len);
2595 ret[len] = '\0';
2596 return ret;
2597}
2598
Baptiste Assmannbb77c8e2013-10-06 23:24:13 +02002599/*
2600 * search needle in haystack
2601 * returns the pointer if found, returns NULL otherwise
2602 */
2603const void *my_memmem(const void *haystack, size_t haystacklen, const void *needle, size_t needlelen)
2604{
2605 const void *c = NULL;
2606 unsigned char f;
2607
2608 if ((haystack == NULL) || (needle == NULL) || (haystacklen < needlelen))
2609 return NULL;
2610
2611 f = *(char *)needle;
2612 c = haystack;
2613 while ((c = memchr(c, f, haystacklen - (c - haystack))) != NULL) {
2614 if ((haystacklen - (c - haystack)) < needlelen)
2615 return NULL;
2616
2617 if (memcmp(c, needle, needlelen) == 0)
2618 return c;
2619 ++c;
2620 }
2621 return NULL;
2622}
2623
Ilya Shipitsinc02a23f2020-05-06 00:53:22 +05002624/* get length of the initial segment consisting entirely of bytes in <accept> */
Christopher Faulet5eb96cb2020-04-15 10:23:01 +02002625size_t my_memspn(const void *str, size_t len, const void *accept, size_t acceptlen)
2626{
2627 size_t ret = 0;
2628
2629 while (ret < len && memchr(accept, *((int *)str), acceptlen)) {
2630 str++;
2631 ret++;
2632 }
2633 return ret;
2634}
2635
Ilya Shipitsinc02a23f2020-05-06 00:53:22 +05002636/* get length of the initial segment consisting entirely of bytes not in <rejcet> */
Christopher Faulet5eb96cb2020-04-15 10:23:01 +02002637size_t my_memcspn(const void *str, size_t len, const void *reject, size_t rejectlen)
2638{
2639 size_t ret = 0;
2640
2641 while (ret < len) {
2642 if(memchr(reject, *((int *)str), rejectlen))
2643 return ret;
2644 str++;
2645 ret++;
2646 }
2647 return ret;
2648}
2649
Willy Tarreau482b00d2009-10-04 22:48:42 +02002650/* This function returns the first unused key greater than or equal to <key> in
2651 * ID tree <root>. Zero is returned if no place is found.
2652 */
2653unsigned int get_next_id(struct eb_root *root, unsigned int key)
2654{
2655 struct eb32_node *used;
2656
2657 do {
2658 used = eb32_lookup_ge(root, key);
2659 if (!used || used->key > key)
2660 return key; /* key is available */
2661 key++;
2662 } while (key);
2663 return key;
2664}
2665
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002666/* dump the full tree to <file> in DOT format for debugging purposes. Will
2667 * optionally highlight node <subj> if found, depending on operation <op> :
2668 * 0 : nothing
2669 * >0 : insertion, node/leaf are surrounded in red
2670 * <0 : removal, node/leaf are dashed with no background
2671 * Will optionally add "desc" as a label on the graph if set and non-null.
2672 */
2673void 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 +01002674{
2675 struct eb32sc_node *node;
2676 unsigned long scope = -1;
2677
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002678 fprintf(file, "digraph ebtree {\n");
2679
2680 if (desc && *desc) {
2681 fprintf(file,
2682 " fontname=\"fixed\";\n"
2683 " fontsize=8;\n"
2684 " label=\"%s\";\n", desc);
2685 }
2686
Willy Tarreaued3cda02017-11-15 15:04:05 +01002687 fprintf(file,
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002688 " node [fontname=\"fixed\" fontsize=8 shape=\"box\" style=\"filled\" color=\"black\" fillcolor=\"white\"];\n"
2689 " edge [fontname=\"fixed\" fontsize=8 style=\"solid\" color=\"magenta\" dir=\"forward\"];\n"
Willy Tarreaued3cda02017-11-15 15:04:05 +01002690 " \"%lx_n\" [label=\"root\\n%lx\"]\n", (long)eb_root_to_node(root), (long)root
2691 );
2692
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002693 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002694 (long)eb_root_to_node(root),
2695 (long)eb_root_to_node(eb_clrtag(root->b[0])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002696 eb_gettag(root->b[0]) == EB_LEAF ? 'l' : 'n');
2697
2698 node = eb32sc_first(root, scope);
2699 while (node) {
2700 if (node->node.node_p) {
2701 /* node part is used */
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002702 fprintf(file, " \"%lx_n\" [label=\"%lx\\nkey=%u\\nscope=%lx\\nbit=%d\" fillcolor=\"lightskyblue1\" %s];\n",
2703 (long)node, (long)node, node->key, node->node_s, node->node.bit,
2704 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
Willy Tarreaued3cda02017-11-15 15:04:05 +01002705
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002706 fprintf(file, " \"%lx_n\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002707 (long)node,
2708 (long)eb_root_to_node(eb_clrtag(node->node.node_p)),
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002709 eb_gettag(node->node.node_p) ? 'R' : 'L');
Willy Tarreaued3cda02017-11-15 15:04:05 +01002710
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002711 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002712 (long)node,
2713 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[0])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002714 eb_gettag(node->node.branches.b[0]) == EB_LEAF ? 'l' : 'n');
2715
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002716 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"R\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002717 (long)node,
2718 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[1])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002719 eb_gettag(node->node.branches.b[1]) == EB_LEAF ? 'l' : 'n');
2720 }
2721
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002722 fprintf(file, " \"%lx_l\" [label=\"%lx\\nkey=%u\\nscope=%lx\\npfx=%u\" fillcolor=\"yellow\" %s];\n",
2723 (long)node, (long)node, node->key, node->leaf_s, node->node.pfx,
2724 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
Willy Tarreaued3cda02017-11-15 15:04:05 +01002725
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002726 fprintf(file, " \"%lx_l\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002727 (long)node,
2728 (long)eb_root_to_node(eb_clrtag(node->node.leaf_p)),
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002729 eb_gettag(node->node.leaf_p) ? 'R' : 'L');
Willy Tarreaued3cda02017-11-15 15:04:05 +01002730 node = eb32sc_next(node, scope);
2731 }
2732 fprintf(file, "}\n");
2733}
2734
Willy Tarreau348238b2010-01-18 15:05:57 +01002735/* This function compares a sample word possibly followed by blanks to another
2736 * clean word. The compare is case-insensitive. 1 is returned if both are equal,
2737 * otherwise zero. This intends to be used when checking HTTP headers for some
2738 * values. Note that it validates a word followed only by blanks but does not
2739 * validate a word followed by blanks then other chars.
2740 */
2741int word_match(const char *sample, int slen, const char *word, int wlen)
2742{
2743 if (slen < wlen)
2744 return 0;
2745
2746 while (wlen) {
2747 char c = *sample ^ *word;
2748 if (c && c != ('A' ^ 'a'))
2749 return 0;
2750 sample++;
2751 word++;
2752 slen--;
2753 wlen--;
2754 }
2755
2756 while (slen) {
2757 if (*sample != ' ' && *sample != '\t')
2758 return 0;
2759 sample++;
2760 slen--;
2761 }
2762 return 1;
2763}
Willy Tarreau482b00d2009-10-04 22:48:42 +02002764
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002765/* Converts any text-formatted IPv4 address to a host-order IPv4 address. It
2766 * is particularly fast because it avoids expensive operations such as
2767 * multiplies, which are optimized away at the end. It requires a properly
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002768 * formatted address though (3 points).
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002769 */
2770unsigned int inetaddr_host(const char *text)
2771{
2772 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2773 register unsigned int dig100, dig10, dig1;
2774 int s;
2775 const char *p, *d;
2776
2777 dig1 = dig10 = dig100 = ascii_zero;
2778 s = 24;
2779
2780 p = text;
2781 while (1) {
2782 if (((unsigned)(*p - '0')) <= 9) {
2783 p++;
2784 continue;
2785 }
2786
2787 /* here, we have a complete byte between <text> and <p> (exclusive) */
2788 if (p == text)
2789 goto end;
2790
2791 d = p - 1;
2792 dig1 |= (unsigned int)(*d << s);
2793 if (d == text)
2794 goto end;
2795
2796 d--;
2797 dig10 |= (unsigned int)(*d << s);
2798 if (d == text)
2799 goto end;
2800
2801 d--;
2802 dig100 |= (unsigned int)(*d << s);
2803 end:
2804 if (!s || *p != '.')
2805 break;
2806
2807 s -= 8;
2808 text = ++p;
2809 }
2810
2811 dig100 -= ascii_zero;
2812 dig10 -= ascii_zero;
2813 dig1 -= ascii_zero;
2814 return ((dig100 * 10) + dig10) * 10 + dig1;
2815}
2816
2817/*
2818 * Idem except the first unparsed character has to be passed in <stop>.
2819 */
2820unsigned int inetaddr_host_lim(const char *text, const char *stop)
2821{
2822 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2823 register unsigned int dig100, dig10, dig1;
2824 int s;
2825 const char *p, *d;
2826
2827 dig1 = dig10 = dig100 = ascii_zero;
2828 s = 24;
2829
2830 p = text;
2831 while (1) {
2832 if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2833 p++;
2834 continue;
2835 }
2836
2837 /* here, we have a complete byte between <text> and <p> (exclusive) */
2838 if (p == text)
2839 goto end;
2840
2841 d = p - 1;
2842 dig1 |= (unsigned int)(*d << s);
2843 if (d == text)
2844 goto end;
2845
2846 d--;
2847 dig10 |= (unsigned int)(*d << s);
2848 if (d == text)
2849 goto end;
2850
2851 d--;
2852 dig100 |= (unsigned int)(*d << s);
2853 end:
2854 if (!s || p == stop || *p != '.')
2855 break;
2856
2857 s -= 8;
2858 text = ++p;
2859 }
2860
2861 dig100 -= ascii_zero;
2862 dig10 -= ascii_zero;
2863 dig1 -= ascii_zero;
2864 return ((dig100 * 10) + dig10) * 10 + dig1;
2865}
2866
2867/*
2868 * Idem except the pointer to first unparsed byte is returned into <ret> which
2869 * must not be NULL.
2870 */
Willy Tarreau74172752010-10-15 23:21:42 +02002871unsigned int inetaddr_host_lim_ret(char *text, char *stop, char **ret)
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002872{
2873 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2874 register unsigned int dig100, dig10, dig1;
2875 int s;
Willy Tarreau74172752010-10-15 23:21:42 +02002876 char *p, *d;
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002877
2878 dig1 = dig10 = dig100 = ascii_zero;
2879 s = 24;
2880
2881 p = text;
2882 while (1) {
2883 if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2884 p++;
2885 continue;
2886 }
2887
2888 /* here, we have a complete byte between <text> and <p> (exclusive) */
2889 if (p == text)
2890 goto end;
2891
2892 d = p - 1;
2893 dig1 |= (unsigned int)(*d << s);
2894 if (d == text)
2895 goto end;
2896
2897 d--;
2898 dig10 |= (unsigned int)(*d << s);
2899 if (d == text)
2900 goto end;
2901
2902 d--;
2903 dig100 |= (unsigned int)(*d << s);
2904 end:
2905 if (!s || p == stop || *p != '.')
2906 break;
2907
2908 s -= 8;
2909 text = ++p;
2910 }
2911
2912 *ret = p;
2913 dig100 -= ascii_zero;
2914 dig10 -= ascii_zero;
2915 dig1 -= ascii_zero;
2916 return ((dig100 * 10) + dig10) * 10 + dig1;
2917}
2918
Willy Tarreauf0b38bf2010-06-06 13:22:23 +02002919/* Convert a fixed-length string to an IP address. Returns 0 in case of error,
2920 * or the number of chars read in case of success. Maybe this could be replaced
2921 * by one of the functions above. Also, apparently this function does not support
2922 * hosts above 255 and requires exactly 4 octets.
Willy Tarreau075415a2013-12-12 11:29:39 +01002923 * The destination is only modified on success.
Willy Tarreauf0b38bf2010-06-06 13:22:23 +02002924 */
2925int buf2ip(const char *buf, size_t len, struct in_addr *dst)
2926{
2927 const char *addr;
2928 int saw_digit, octets, ch;
2929 u_char tmp[4], *tp;
2930 const char *cp = buf;
2931
2932 saw_digit = 0;
2933 octets = 0;
2934 *(tp = tmp) = 0;
2935
2936 for (addr = buf; addr - buf < len; addr++) {
2937 unsigned char digit = (ch = *addr) - '0';
2938
2939 if (digit > 9 && ch != '.')
2940 break;
2941
2942 if (digit <= 9) {
2943 u_int new = *tp * 10 + digit;
2944
2945 if (new > 255)
2946 return 0;
2947
2948 *tp = new;
2949
2950 if (!saw_digit) {
2951 if (++octets > 4)
2952 return 0;
2953 saw_digit = 1;
2954 }
2955 } else if (ch == '.' && saw_digit) {
2956 if (octets == 4)
2957 return 0;
2958
2959 *++tp = 0;
2960 saw_digit = 0;
2961 } else
2962 return 0;
2963 }
2964
2965 if (octets < 4)
2966 return 0;
2967
2968 memcpy(&dst->s_addr, tmp, 4);
2969 return addr - cp;
2970}
2971
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002972/* This function converts the string in <buf> of the len <len> to
2973 * struct in6_addr <dst> which must be allocated by the caller.
2974 * This function returns 1 in success case, otherwise zero.
Willy Tarreau075415a2013-12-12 11:29:39 +01002975 * The destination is only modified on success.
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002976 */
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002977int buf2ip6(const char *buf, size_t len, struct in6_addr *dst)
2978{
Thierry FOURNIERcd659912013-12-11 12:33:54 +01002979 char null_term_ip6[INET6_ADDRSTRLEN + 1];
Willy Tarreau075415a2013-12-12 11:29:39 +01002980 struct in6_addr out;
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002981
Thierry FOURNIERcd659912013-12-11 12:33:54 +01002982 if (len > INET6_ADDRSTRLEN)
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002983 return 0;
2984
2985 memcpy(null_term_ip6, buf, len);
2986 null_term_ip6[len] = '\0';
2987
Willy Tarreau075415a2013-12-12 11:29:39 +01002988 if (!inet_pton(AF_INET6, null_term_ip6, &out))
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002989 return 0;
2990
Willy Tarreau075415a2013-12-12 11:29:39 +01002991 *dst = out;
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002992 return 1;
2993}
2994
Willy Tarreauacf95772010-06-14 19:09:21 +02002995/* To be used to quote config arg positions. Returns the short string at <ptr>
2996 * surrounded by simple quotes if <ptr> is valid and non-empty, or "end of line"
2997 * if ptr is NULL or empty. The string is locally allocated.
2998 */
2999const char *quote_arg(const char *ptr)
3000{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01003001 static THREAD_LOCAL char val[32];
Willy Tarreauacf95772010-06-14 19:09:21 +02003002 int i;
3003
3004 if (!ptr || !*ptr)
3005 return "end of line";
3006 val[0] = '\'';
Willy Tarreaude2dd6b2013-01-24 02:14:42 +01003007 for (i = 1; i < sizeof(val) - 2 && *ptr; i++)
Willy Tarreauacf95772010-06-14 19:09:21 +02003008 val[i] = *ptr++;
3009 val[i++] = '\'';
3010 val[i] = '\0';
3011 return val;
3012}
3013
Willy Tarreau5b180202010-07-18 10:40:48 +02003014/* returns an operator among STD_OP_* for string <str> or < 0 if unknown */
3015int get_std_op(const char *str)
3016{
3017 int ret = -1;
3018
3019 if (*str == 'e' && str[1] == 'q')
3020 ret = STD_OP_EQ;
3021 else if (*str == 'n' && str[1] == 'e')
3022 ret = STD_OP_NE;
3023 else if (*str == 'l') {
3024 if (str[1] == 'e') ret = STD_OP_LE;
3025 else if (str[1] == 't') ret = STD_OP_LT;
3026 }
3027 else if (*str == 'g') {
3028 if (str[1] == 'e') ret = STD_OP_GE;
3029 else if (str[1] == 't') ret = STD_OP_GT;
3030 }
3031
3032 if (ret == -1 || str[2] != '\0')
3033 return -1;
3034 return ret;
3035}
3036
Willy Tarreau4c14eaa2010-11-24 14:01:45 +01003037/* hash a 32-bit integer to another 32-bit integer */
3038unsigned int full_hash(unsigned int a)
3039{
3040 return __full_hash(a);
3041}
3042
Willy Tarreauf3241112019-02-26 09:56:22 +01003043/* Return the bit position in mask <m> of the nth bit set of rank <r>, between
3044 * 0 and LONGBITS-1 included, starting from the left. For example ranks 0,1,2,3
3045 * for mask 0x55 will be 6, 4, 2 and 0 respectively. This algorithm is based on
3046 * a popcount variant and is described here :
3047 * https://graphics.stanford.edu/~seander/bithacks.html
3048 */
3049unsigned int mask_find_rank_bit(unsigned int r, unsigned long m)
3050{
3051 unsigned long a, b, c, d;
3052 unsigned int s;
3053 unsigned int t;
3054
3055 a = m - ((m >> 1) & ~0UL/3);
3056 b = (a & ~0UL/5) + ((a >> 2) & ~0UL/5);
3057 c = (b + (b >> 4)) & ~0UL/0x11;
3058 d = (c + (c >> 8)) & ~0UL/0x101;
3059
3060 r++; // make r be 1..64
3061
3062 t = 0;
3063 s = LONGBITS;
3064 if (s > 32) {
Willy Tarreau9b6be3b2019-03-18 16:31:18 +01003065 unsigned long d2 = (d >> 16) >> 16;
3066 t = d2 + (d2 >> 16);
Willy Tarreauf3241112019-02-26 09:56:22 +01003067 s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
3068 }
3069
3070 t = (d >> (s - 16)) & 0xff;
3071 s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
3072 t = (c >> (s - 8)) & 0xf;
3073 s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
3074 t = (b >> (s - 4)) & 0x7;
3075 s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
3076 t = (a >> (s - 2)) & 0x3;
3077 s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
3078 t = (m >> (s - 1)) & 0x1;
3079 s -= ((t - r) & 256) >> 8;
3080
3081 return s - 1;
3082}
3083
3084/* Same as mask_find_rank_bit() above but makes use of pre-computed bitmaps
3085 * based on <m>, in <a..d>. These ones must be updated whenever <m> changes
3086 * using mask_prep_rank_map() below.
3087 */
3088unsigned int mask_find_rank_bit_fast(unsigned int r, unsigned long m,
3089 unsigned long a, unsigned long b,
3090 unsigned long c, unsigned long d)
3091{
3092 unsigned int s;
3093 unsigned int t;
3094
3095 r++; // make r be 1..64
3096
3097 t = 0;
3098 s = LONGBITS;
3099 if (s > 32) {
Willy Tarreau9b6be3b2019-03-18 16:31:18 +01003100 unsigned long d2 = (d >> 16) >> 16;
3101 t = d2 + (d2 >> 16);
Willy Tarreauf3241112019-02-26 09:56:22 +01003102 s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
3103 }
3104
3105 t = (d >> (s - 16)) & 0xff;
3106 s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
3107 t = (c >> (s - 8)) & 0xf;
3108 s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
3109 t = (b >> (s - 4)) & 0x7;
3110 s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
3111 t = (a >> (s - 2)) & 0x3;
3112 s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
3113 t = (m >> (s - 1)) & 0x1;
3114 s -= ((t - r) & 256) >> 8;
3115
3116 return s - 1;
3117}
3118
3119/* Prepare the bitmaps used by the fast implementation of the find_rank_bit()
3120 * above.
3121 */
3122void mask_prep_rank_map(unsigned long m,
3123 unsigned long *a, unsigned long *b,
3124 unsigned long *c, unsigned long *d)
3125{
3126 *a = m - ((m >> 1) & ~0UL/3);
3127 *b = (*a & ~0UL/5) + ((*a >> 2) & ~0UL/5);
3128 *c = (*b + (*b >> 4)) & ~0UL/0x11;
3129 *d = (*c + (*c >> 8)) & ~0UL/0x101;
3130}
3131
David du Colombier4f92d322011-03-24 11:09:31 +01003132/* Return non-zero if IPv4 address is part of the network,
Willy Tarreaueec1d382016-07-13 11:59:39 +02003133 * otherwise zero. Note that <addr> may not necessarily be aligned
3134 * while the two other ones must.
David du Colombier4f92d322011-03-24 11:09:31 +01003135 */
Willy Tarreaueec1d382016-07-13 11:59:39 +02003136int in_net_ipv4(const void *addr, const struct in_addr *mask, const struct in_addr *net)
David du Colombier4f92d322011-03-24 11:09:31 +01003137{
Willy Tarreaueec1d382016-07-13 11:59:39 +02003138 struct in_addr addr_copy;
3139
3140 memcpy(&addr_copy, addr, sizeof(addr_copy));
3141 return((addr_copy.s_addr & mask->s_addr) == (net->s_addr & mask->s_addr));
David du Colombier4f92d322011-03-24 11:09:31 +01003142}
3143
3144/* Return non-zero if IPv6 address is part of the network,
Willy Tarreaueec1d382016-07-13 11:59:39 +02003145 * otherwise zero. Note that <addr> may not necessarily be aligned
3146 * while the two other ones must.
David du Colombier4f92d322011-03-24 11:09:31 +01003147 */
Willy Tarreaueec1d382016-07-13 11:59:39 +02003148int in_net_ipv6(const void *addr, const struct in6_addr *mask, const struct in6_addr *net)
David du Colombier4f92d322011-03-24 11:09:31 +01003149{
3150 int i;
Willy Tarreaueec1d382016-07-13 11:59:39 +02003151 struct in6_addr addr_copy;
David du Colombier4f92d322011-03-24 11:09:31 +01003152
Willy Tarreaueec1d382016-07-13 11:59:39 +02003153 memcpy(&addr_copy, addr, sizeof(addr_copy));
David du Colombier4f92d322011-03-24 11:09:31 +01003154 for (i = 0; i < sizeof(struct in6_addr) / sizeof(int); i++)
Willy Tarreaueec1d382016-07-13 11:59:39 +02003155 if (((((int *)&addr_copy)[i] & ((int *)mask)[i])) !=
David du Colombier4f92d322011-03-24 11:09:31 +01003156 (((int *)net)[i] & ((int *)mask)[i]))
3157 return 0;
3158 return 1;
3159}
3160
3161/* RFC 4291 prefix */
3162const char rfc4291_pfx[] = { 0x00, 0x00, 0x00, 0x00,
3163 0x00, 0x00, 0x00, 0x00,
3164 0x00, 0x00, 0xFF, 0xFF };
3165
Joseph Herlant32b83272018-11-15 11:58:28 -08003166/* Map IPv4 address on IPv6 address, as specified in RFC 3513.
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003167 * Input and output may overlap.
3168 */
David du Colombier4f92d322011-03-24 11:09:31 +01003169void v4tov6(struct in6_addr *sin6_addr, struct in_addr *sin_addr)
3170{
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003171 struct in_addr tmp_addr;
3172
3173 tmp_addr.s_addr = sin_addr->s_addr;
David du Colombier4f92d322011-03-24 11:09:31 +01003174 memcpy(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx));
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003175 memcpy(sin6_addr->s6_addr+12, &tmp_addr.s_addr, 4);
David du Colombier4f92d322011-03-24 11:09:31 +01003176}
3177
Joseph Herlant32b83272018-11-15 11:58:28 -08003178/* Map IPv6 address on IPv4 address, as specified in RFC 3513.
David du Colombier4f92d322011-03-24 11:09:31 +01003179 * Return true if conversion is possible and false otherwise.
3180 */
3181int v6tov4(struct in_addr *sin_addr, struct in6_addr *sin6_addr)
3182{
3183 if (memcmp(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx)) == 0) {
3184 memcpy(&(sin_addr->s_addr), &(sin6_addr->s6_addr[12]),
3185 sizeof(struct in_addr));
3186 return 1;
3187 }
3188
3189 return 0;
3190}
3191
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003192/* compare two struct sockaddr_storage and return:
3193 * 0 (true) if the addr is the same in both
3194 * 1 (false) if the addr is not the same in both
3195 * -1 (unable) if one of the addr is not AF_INET*
3196 */
3197int ipcmp(struct sockaddr_storage *ss1, struct sockaddr_storage *ss2)
3198{
3199 if ((ss1->ss_family != AF_INET) && (ss1->ss_family != AF_INET6))
3200 return -1;
3201
3202 if ((ss2->ss_family != AF_INET) && (ss2->ss_family != AF_INET6))
3203 return -1;
3204
3205 if (ss1->ss_family != ss2->ss_family)
3206 return 1;
3207
3208 switch (ss1->ss_family) {
3209 case AF_INET:
3210 return memcmp(&((struct sockaddr_in *)ss1)->sin_addr,
3211 &((struct sockaddr_in *)ss2)->sin_addr,
3212 sizeof(struct in_addr)) != 0;
3213 case AF_INET6:
3214 return memcmp(&((struct sockaddr_in6 *)ss1)->sin6_addr,
3215 &((struct sockaddr_in6 *)ss2)->sin6_addr,
3216 sizeof(struct in6_addr)) != 0;
3217 }
3218
3219 return 1;
3220}
3221
Christopher Faulet9553de72021-02-26 09:12:50 +01003222/* compare a struct sockaddr_storage to a struct net_addr and return :
3223 * 0 (true) if <addr> is matching <net>
3224 * 1 (false) if <addr> is not matching <net>
3225 * -1 (unable) if <addr> or <net> is not AF_INET*
3226 */
3227int ipcmp2net(const struct sockaddr_storage *addr, const struct net_addr *net)
3228{
3229 if ((addr->ss_family != AF_INET) && (addr->ss_family != AF_INET6))
3230 return -1;
3231
3232 if ((net->family != AF_INET) && (net->family != AF_INET6))
3233 return -1;
3234
3235 if (addr->ss_family != net->family)
3236 return 1;
3237
3238 if (addr->ss_family == AF_INET &&
3239 (((struct sockaddr_in *)addr)->sin_addr.s_addr & net->addr.v4.mask.s_addr) == net->addr.v4.ip.s_addr)
3240 return 0;
3241 else {
3242 const struct in6_addr *addr6 = &(((const struct sockaddr_in6*)addr)->sin6_addr);
3243 const struct in6_addr *nip6 = &net->addr.v6.ip;
3244 const struct in6_addr *nmask6 = &net->addr.v6.mask;
3245
3246 if ((read_u32(&addr6->s6_addr[0]) & read_u32(&nmask6->s6_addr[0])) == read_u32(&nip6->s6_addr[0]) &&
3247 (read_u32(&addr6->s6_addr[4]) & read_u32(&nmask6->s6_addr[4])) == read_u32(&nip6->s6_addr[4]) &&
3248 (read_u32(&addr6->s6_addr[8]) & read_u32(&nmask6->s6_addr[8])) == read_u32(&nip6->s6_addr[8]) &&
3249 (read_u32(&addr6->s6_addr[12]) & read_u32(&nmask6->s6_addr[12])) == read_u32(&nip6->s6_addr[12]))
3250 return 0;
3251 }
3252
3253 return 1;
3254}
3255
Baptiste Assmann08396c82016-01-31 00:27:17 +01003256/* copy IP address from <source> into <dest>
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003257 * The caller must allocate and clear <dest> before calling.
3258 * The source must be in either AF_INET or AF_INET6 family, or the destination
3259 * address will be undefined. If the destination address used to hold a port,
3260 * it is preserved, so that this function can be used to switch to another
3261 * address family with no risk. Returns a pointer to the destination.
Baptiste Assmann08396c82016-01-31 00:27:17 +01003262 */
3263struct sockaddr_storage *ipcpy(struct sockaddr_storage *source, struct sockaddr_storage *dest)
3264{
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003265 int prev_port;
3266
3267 prev_port = get_net_port(dest);
3268 memset(dest, 0, sizeof(*dest));
Baptiste Assmann08396c82016-01-31 00:27:17 +01003269 dest->ss_family = source->ss_family;
3270
3271 /* copy new addr and apply it */
3272 switch (source->ss_family) {
3273 case AF_INET:
3274 ((struct sockaddr_in *)dest)->sin_addr.s_addr = ((struct sockaddr_in *)source)->sin_addr.s_addr;
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003275 ((struct sockaddr_in *)dest)->sin_port = prev_port;
Baptiste Assmann08396c82016-01-31 00:27:17 +01003276 break;
3277 case AF_INET6:
3278 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 +01003279 ((struct sockaddr_in6 *)dest)->sin6_port = prev_port;
Baptiste Assmann08396c82016-01-31 00:27:17 +01003280 break;
3281 }
3282
3283 return dest;
3284}
3285
William Lallemand421f5b52012-02-06 18:15:57 +01003286char *human_time(int t, short hz_div) {
3287 static char rv[sizeof("24855d23h")+1]; // longest of "23h59m" and "59m59s"
3288 char *p = rv;
Willy Tarreau761b3d52014-04-14 14:53:06 +02003289 char *end = rv + sizeof(rv);
William Lallemand421f5b52012-02-06 18:15:57 +01003290 int cnt=2; // print two numbers
3291
3292 if (unlikely(t < 0 || hz_div <= 0)) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003293 snprintf(p, end - p, "?");
William Lallemand421f5b52012-02-06 18:15:57 +01003294 return rv;
3295 }
3296
3297 if (unlikely(hz_div > 1))
3298 t /= hz_div;
3299
3300 if (t >= DAY) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003301 p += snprintf(p, end - p, "%dd", t / DAY);
William Lallemand421f5b52012-02-06 18:15:57 +01003302 cnt--;
3303 }
3304
3305 if (cnt && t % DAY / HOUR) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003306 p += snprintf(p, end - p, "%dh", t % DAY / HOUR);
William Lallemand421f5b52012-02-06 18:15:57 +01003307 cnt--;
3308 }
3309
3310 if (cnt && t % HOUR / MINUTE) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003311 p += snprintf(p, end - p, "%dm", t % HOUR / MINUTE);
William Lallemand421f5b52012-02-06 18:15:57 +01003312 cnt--;
3313 }
3314
3315 if ((cnt && t % MINUTE) || !t) // also display '0s'
Willy Tarreau761b3d52014-04-14 14:53:06 +02003316 p += snprintf(p, end - p, "%ds", t % MINUTE / SEC);
William Lallemand421f5b52012-02-06 18:15:57 +01003317
3318 return rv;
3319}
3320
3321const char *monthname[12] = {
3322 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
3323 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"
3324};
3325
3326/* date2str_log: write a date in the format :
3327 * sprintf(str, "%02d/%s/%04d:%02d:%02d:%02d.%03d",
3328 * tm.tm_mday, monthname[tm.tm_mon], tm.tm_year+1900,
3329 * tm.tm_hour, tm.tm_min, tm.tm_sec, (int)date.tv_usec/1000);
3330 *
3331 * without using sprintf. return a pointer to the last char written (\0) or
3332 * NULL if there isn't enough space.
3333 */
Willy Tarreauf16cb412018-09-04 19:08:48 +02003334char *date2str_log(char *dst, const struct tm *tm, const struct timeval *date, size_t size)
William Lallemand421f5b52012-02-06 18:15:57 +01003335{
3336
3337 if (size < 25) /* the size is fixed: 24 chars + \0 */
3338 return NULL;
3339
3340 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003341 if (!dst)
3342 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003343 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003344
William Lallemand421f5b52012-02-06 18:15:57 +01003345 memcpy(dst, monthname[tm->tm_mon], 3); // month
3346 dst += 3;
3347 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003348
William Lallemand421f5b52012-02-06 18:15:57 +01003349 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003350 if (!dst)
3351 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003352 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003353
William Lallemand421f5b52012-02-06 18:15:57 +01003354 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003355 if (!dst)
3356 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003357 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003358
William Lallemand421f5b52012-02-06 18:15:57 +01003359 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003360 if (!dst)
3361 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003362 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003363
William Lallemand421f5b52012-02-06 18:15:57 +01003364 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003365 if (!dst)
3366 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003367 *dst++ = '.';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003368
Willy Tarreau7d9421d2020-02-29 09:08:02 +01003369 dst = utoa_pad((unsigned int)(date->tv_usec/1000)%1000, dst, 4); // milliseconds
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003370 if (!dst)
3371 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003372 *dst = '\0';
3373
3374 return dst;
3375}
3376
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003377/* Base year used to compute leap years */
3378#define TM_YEAR_BASE 1900
3379
3380/* Return the difference in seconds between two times (leap seconds are ignored).
3381 * Retrieved from glibc 2.18 source code.
3382 */
3383static int my_tm_diff(const struct tm *a, const struct tm *b)
3384{
3385 /* Compute intervening leap days correctly even if year is negative.
3386 * Take care to avoid int overflow in leap day calculations,
3387 * but it's OK to assume that A and B are close to each other.
3388 */
3389 int a4 = (a->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (a->tm_year & 3);
3390 int b4 = (b->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (b->tm_year & 3);
3391 int a100 = a4 / 25 - (a4 % 25 < 0);
3392 int b100 = b4 / 25 - (b4 % 25 < 0);
3393 int a400 = a100 >> 2;
3394 int b400 = b100 >> 2;
3395 int intervening_leap_days = (a4 - b4) - (a100 - b100) + (a400 - b400);
3396 int years = a->tm_year - b->tm_year;
3397 int days = (365 * years + intervening_leap_days
3398 + (a->tm_yday - b->tm_yday));
3399 return (60 * (60 * (24 * days + (a->tm_hour - b->tm_hour))
3400 + (a->tm_min - b->tm_min))
3401 + (a->tm_sec - b->tm_sec));
3402}
3403
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003404/* Return the GMT offset for a specific local time.
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003405 * Both t and tm must represent the same time.
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003406 * The string returned has the same format as returned by strftime(... "%z", tm).
3407 * Offsets are kept in an internal cache for better performances.
3408 */
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003409const char *get_gmt_offset(time_t t, struct tm *tm)
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003410{
3411 /* Cache offsets from GMT (depending on whether DST is active or not) */
Christopher Faulet1bc04c72017-10-29 20:14:08 +01003412 static THREAD_LOCAL char gmt_offsets[2][5+1] = { "", "" };
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003413
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003414 char *gmt_offset;
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003415 struct tm tm_gmt;
3416 int diff;
3417 int isdst = tm->tm_isdst;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003418
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003419 /* Pretend DST not active if its status is unknown */
3420 if (isdst < 0)
3421 isdst = 0;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003422
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003423 /* Fetch the offset and initialize it if needed */
3424 gmt_offset = gmt_offsets[isdst & 0x01];
3425 if (unlikely(!*gmt_offset)) {
3426 get_gmtime(t, &tm_gmt);
3427 diff = my_tm_diff(tm, &tm_gmt);
3428 if (diff < 0) {
3429 diff = -diff;
3430 *gmt_offset = '-';
3431 } else {
3432 *gmt_offset = '+';
3433 }
Willy Tarreaue112c8a2019-10-29 10:16:11 +01003434 diff %= 86400U;
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003435 diff /= 60; /* Convert to minutes */
3436 snprintf(gmt_offset+1, 4+1, "%02d%02d", diff/60, diff%60);
3437 }
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003438
Willy Tarreaue112c8a2019-10-29 10:16:11 +01003439 return gmt_offset;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003440}
3441
William Lallemand421f5b52012-02-06 18:15:57 +01003442/* gmt2str_log: write a date in the format :
3443 * "%02d/%s/%04d:%02d:%02d:%02d +0000" without using snprintf
3444 * return a pointer to the last char written (\0) or
3445 * NULL if there isn't enough space.
3446 */
3447char *gmt2str_log(char *dst, struct tm *tm, size_t size)
3448{
Yuxans Yao4e25b012012-10-19 10:36:09 +08003449 if (size < 27) /* the size is fixed: 26 chars + \0 */
William Lallemand421f5b52012-02-06 18:15:57 +01003450 return NULL;
3451
3452 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003453 if (!dst)
3454 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003455 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003456
William Lallemand421f5b52012-02-06 18:15:57 +01003457 memcpy(dst, monthname[tm->tm_mon], 3); // month
3458 dst += 3;
3459 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003460
William Lallemand421f5b52012-02-06 18:15:57 +01003461 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003462 if (!dst)
3463 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003464 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003465
William Lallemand421f5b52012-02-06 18:15:57 +01003466 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003467 if (!dst)
3468 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003469 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003470
William Lallemand421f5b52012-02-06 18:15:57 +01003471 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003472 if (!dst)
3473 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003474 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003475
William Lallemand421f5b52012-02-06 18:15:57 +01003476 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003477 if (!dst)
3478 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003479 *dst++ = ' ';
3480 *dst++ = '+';
3481 *dst++ = '0';
3482 *dst++ = '0';
3483 *dst++ = '0';
3484 *dst++ = '0';
3485 *dst = '\0';
3486
3487 return dst;
3488}
3489
Yuxans Yao4e25b012012-10-19 10:36:09 +08003490/* localdate2str_log: write a date in the format :
3491 * "%02d/%s/%04d:%02d:%02d:%02d +0000(local timezone)" without using snprintf
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003492 * Both t and tm must represent the same time.
3493 * return a pointer to the last char written (\0) or
3494 * NULL if there isn't enough space.
Yuxans Yao4e25b012012-10-19 10:36:09 +08003495 */
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003496char *localdate2str_log(char *dst, time_t t, struct tm *tm, size_t size)
Yuxans Yao4e25b012012-10-19 10:36:09 +08003497{
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003498 const char *gmt_offset;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003499 if (size < 27) /* the size is fixed: 26 chars + \0 */
3500 return NULL;
3501
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003502 gmt_offset = get_gmt_offset(t, tm);
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003503
Yuxans Yao4e25b012012-10-19 10:36:09 +08003504 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003505 if (!dst)
3506 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003507 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003508
Yuxans Yao4e25b012012-10-19 10:36:09 +08003509 memcpy(dst, monthname[tm->tm_mon], 3); // month
3510 dst += 3;
3511 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003512
Yuxans Yao4e25b012012-10-19 10:36:09 +08003513 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003514 if (!dst)
3515 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003516 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003517
Yuxans Yao4e25b012012-10-19 10:36:09 +08003518 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003519 if (!dst)
3520 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003521 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003522
Yuxans Yao4e25b012012-10-19 10:36:09 +08003523 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003524 if (!dst)
3525 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003526 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003527
Yuxans Yao4e25b012012-10-19 10:36:09 +08003528 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003529 if (!dst)
3530 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003531 *dst++ = ' ';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003532
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003533 memcpy(dst, gmt_offset, 5); // Offset from local time to GMT
Yuxans Yao4e25b012012-10-19 10:36:09 +08003534 dst += 5;
3535 *dst = '\0';
3536
3537 return dst;
3538}
3539
Willy Tarreaucb1949b2017-07-19 19:05:29 +02003540/* Returns the number of seconds since 01/01/1970 0:0:0 GMT for GMT date <tm>.
3541 * It is meant as a portable replacement for timegm() for use with valid inputs.
3542 * Returns undefined results for invalid dates (eg: months out of range 0..11).
3543 */
3544time_t my_timegm(const struct tm *tm)
3545{
3546 /* Each month has 28, 29, 30 or 31 days, or 28+N. The date in the year
3547 * is thus (current month - 1)*28 + cumulated_N[month] to count the
3548 * sum of the extra N days for elapsed months. The sum of all these N
3549 * days doesn't exceed 30 for a complete year (366-12*28) so it fits
3550 * in a 5-bit word. This means that with 60 bits we can represent a
3551 * matrix of all these values at once, which is fast and efficient to
3552 * access. The extra February day for leap years is not counted here.
3553 *
3554 * Jan : none = 0 (0)
3555 * Feb : Jan = 3 (3)
3556 * Mar : Jan..Feb = 3 (3 + 0)
3557 * Apr : Jan..Mar = 6 (3 + 0 + 3)
3558 * May : Jan..Apr = 8 (3 + 0 + 3 + 2)
3559 * Jun : Jan..May = 11 (3 + 0 + 3 + 2 + 3)
3560 * Jul : Jan..Jun = 13 (3 + 0 + 3 + 2 + 3 + 2)
3561 * Aug : Jan..Jul = 16 (3 + 0 + 3 + 2 + 3 + 2 + 3)
3562 * Sep : Jan..Aug = 19 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3)
3563 * Oct : Jan..Sep = 21 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2)
3564 * Nov : Jan..Oct = 24 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3)
3565 * Dec : Jan..Nov = 26 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3 + 2)
3566 */
3567 uint64_t extra =
3568 ( 0ULL << 0*5) + ( 3ULL << 1*5) + ( 3ULL << 2*5) + /* Jan, Feb, Mar, */
3569 ( 6ULL << 3*5) + ( 8ULL << 4*5) + (11ULL << 5*5) + /* Apr, May, Jun, */
3570 (13ULL << 6*5) + (16ULL << 7*5) + (19ULL << 8*5) + /* Jul, Aug, Sep, */
3571 (21ULL << 9*5) + (24ULL << 10*5) + (26ULL << 11*5); /* Oct, Nov, Dec, */
3572
3573 unsigned int y = tm->tm_year + 1900;
3574 unsigned int m = tm->tm_mon;
3575 unsigned long days = 0;
3576
3577 /* days since 1/1/1970 for full years */
3578 days += days_since_zero(y) - days_since_zero(1970);
3579
3580 /* days for full months in the current year */
3581 days += 28 * m + ((extra >> (m * 5)) & 0x1f);
3582
3583 /* count + 1 after March for leap years. A leap year is a year multiple
3584 * of 4, unless it's multiple of 100 without being multiple of 400. 2000
3585 * is leap, 1900 isn't, 1904 is.
3586 */
3587 if ((m > 1) && !(y & 3) && ((y % 100) || !(y % 400)))
3588 days++;
3589
3590 days += tm->tm_mday - 1;
3591 return days * 86400ULL + tm->tm_hour * 3600 + tm->tm_min * 60 + tm->tm_sec;
3592}
3593
Thierry Fournier93127942016-01-20 18:49:45 +01003594/* This function check a char. It returns true and updates
3595 * <date> and <len> pointer to the new position if the
3596 * character is found.
3597 */
3598static inline int parse_expect_char(const char **date, int *len, char c)
3599{
3600 if (*len < 1 || **date != c)
3601 return 0;
3602 (*len)--;
3603 (*date)++;
3604 return 1;
3605}
3606
3607/* This function expects a string <str> of len <l>. It return true and updates.
3608 * <date> and <len> if the string matches, otherwise, it returns false.
3609 */
3610static inline int parse_strcmp(const char **date, int *len, char *str, int l)
3611{
3612 if (*len < l || strncmp(*date, str, l) != 0)
3613 return 0;
3614 (*len) -= l;
3615 (*date) += l;
3616 return 1;
3617}
3618
3619/* This macro converts 3 chars name in integer. */
3620#define STR2I3(__a, __b, __c) ((__a) * 65536 + (__b) * 256 + (__c))
3621
3622/* day-name = %x4D.6F.6E ; "Mon", case-sensitive
3623 * / %x54.75.65 ; "Tue", case-sensitive
3624 * / %x57.65.64 ; "Wed", case-sensitive
3625 * / %x54.68.75 ; "Thu", case-sensitive
3626 * / %x46.72.69 ; "Fri", case-sensitive
3627 * / %x53.61.74 ; "Sat", case-sensitive
3628 * / %x53.75.6E ; "Sun", case-sensitive
3629 *
3630 * This array must be alphabetically sorted
3631 */
3632static inline int parse_http_dayname(const char **date, int *len, struct tm *tm)
3633{
3634 if (*len < 3)
3635 return 0;
3636 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3637 case STR2I3('M','o','n'): tm->tm_wday = 1; break;
3638 case STR2I3('T','u','e'): tm->tm_wday = 2; break;
3639 case STR2I3('W','e','d'): tm->tm_wday = 3; break;
3640 case STR2I3('T','h','u'): tm->tm_wday = 4; break;
3641 case STR2I3('F','r','i'): tm->tm_wday = 5; break;
3642 case STR2I3('S','a','t'): tm->tm_wday = 6; break;
3643 case STR2I3('S','u','n'): tm->tm_wday = 7; break;
3644 default: return 0;
3645 }
3646 *len -= 3;
3647 *date += 3;
3648 return 1;
3649}
3650
3651/* month = %x4A.61.6E ; "Jan", case-sensitive
3652 * / %x46.65.62 ; "Feb", case-sensitive
3653 * / %x4D.61.72 ; "Mar", case-sensitive
3654 * / %x41.70.72 ; "Apr", case-sensitive
3655 * / %x4D.61.79 ; "May", case-sensitive
3656 * / %x4A.75.6E ; "Jun", case-sensitive
3657 * / %x4A.75.6C ; "Jul", case-sensitive
3658 * / %x41.75.67 ; "Aug", case-sensitive
3659 * / %x53.65.70 ; "Sep", case-sensitive
3660 * / %x4F.63.74 ; "Oct", case-sensitive
3661 * / %x4E.6F.76 ; "Nov", case-sensitive
3662 * / %x44.65.63 ; "Dec", case-sensitive
3663 *
3664 * This array must be alphabetically sorted
3665 */
3666static inline int parse_http_monthname(const char **date, int *len, struct tm *tm)
3667{
3668 if (*len < 3)
3669 return 0;
3670 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3671 case STR2I3('J','a','n'): tm->tm_mon = 0; break;
3672 case STR2I3('F','e','b'): tm->tm_mon = 1; break;
3673 case STR2I3('M','a','r'): tm->tm_mon = 2; break;
3674 case STR2I3('A','p','r'): tm->tm_mon = 3; break;
3675 case STR2I3('M','a','y'): tm->tm_mon = 4; break;
3676 case STR2I3('J','u','n'): tm->tm_mon = 5; break;
3677 case STR2I3('J','u','l'): tm->tm_mon = 6; break;
3678 case STR2I3('A','u','g'): tm->tm_mon = 7; break;
3679 case STR2I3('S','e','p'): tm->tm_mon = 8; break;
3680 case STR2I3('O','c','t'): tm->tm_mon = 9; break;
3681 case STR2I3('N','o','v'): tm->tm_mon = 10; break;
3682 case STR2I3('D','e','c'): tm->tm_mon = 11; break;
3683 default: return 0;
3684 }
3685 *len -= 3;
3686 *date += 3;
3687 return 1;
3688}
3689
3690/* day-name-l = %x4D.6F.6E.64.61.79 ; "Monday", case-sensitive
3691 * / %x54.75.65.73.64.61.79 ; "Tuesday", case-sensitive
3692 * / %x57.65.64.6E.65.73.64.61.79 ; "Wednesday", case-sensitive
3693 * / %x54.68.75.72.73.64.61.79 ; "Thursday", case-sensitive
3694 * / %x46.72.69.64.61.79 ; "Friday", case-sensitive
3695 * / %x53.61.74.75.72.64.61.79 ; "Saturday", case-sensitive
3696 * / %x53.75.6E.64.61.79 ; "Sunday", case-sensitive
3697 *
3698 * This array must be alphabetically sorted
3699 */
3700static inline int parse_http_ldayname(const char **date, int *len, struct tm *tm)
3701{
3702 if (*len < 6) /* Minimum length. */
3703 return 0;
3704 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3705 case STR2I3('M','o','n'):
3706 RET0_UNLESS(parse_strcmp(date, len, "Monday", 6));
3707 tm->tm_wday = 1;
3708 return 1;
3709 case STR2I3('T','u','e'):
3710 RET0_UNLESS(parse_strcmp(date, len, "Tuesday", 7));
3711 tm->tm_wday = 2;
3712 return 1;
3713 case STR2I3('W','e','d'):
3714 RET0_UNLESS(parse_strcmp(date, len, "Wednesday", 9));
3715 tm->tm_wday = 3;
3716 return 1;
3717 case STR2I3('T','h','u'):
3718 RET0_UNLESS(parse_strcmp(date, len, "Thursday", 8));
3719 tm->tm_wday = 4;
3720 return 1;
3721 case STR2I3('F','r','i'):
3722 RET0_UNLESS(parse_strcmp(date, len, "Friday", 6));
3723 tm->tm_wday = 5;
3724 return 1;
3725 case STR2I3('S','a','t'):
3726 RET0_UNLESS(parse_strcmp(date, len, "Saturday", 8));
3727 tm->tm_wday = 6;
3728 return 1;
3729 case STR2I3('S','u','n'):
3730 RET0_UNLESS(parse_strcmp(date, len, "Sunday", 6));
3731 tm->tm_wday = 7;
3732 return 1;
3733 }
3734 return 0;
3735}
3736
3737/* This function parses exactly 1 digit and returns the numeric value in "digit". */
3738static inline int parse_digit(const char **date, int *len, int *digit)
3739{
3740 if (*len < 1 || **date < '0' || **date > '9')
3741 return 0;
3742 *digit = (**date - '0');
3743 (*date)++;
3744 (*len)--;
3745 return 1;
3746}
3747
3748/* This function parses exactly 2 digits and returns the numeric value in "digit". */
3749static inline int parse_2digit(const char **date, int *len, int *digit)
3750{
3751 int value;
3752
3753 RET0_UNLESS(parse_digit(date, len, &value));
3754 (*digit) = value * 10;
3755 RET0_UNLESS(parse_digit(date, len, &value));
3756 (*digit) += value;
3757
3758 return 1;
3759}
3760
3761/* This function parses exactly 4 digits and returns the numeric value in "digit". */
3762static inline int parse_4digit(const char **date, int *len, int *digit)
3763{
3764 int value;
3765
3766 RET0_UNLESS(parse_digit(date, len, &value));
3767 (*digit) = value * 1000;
3768
3769 RET0_UNLESS(parse_digit(date, len, &value));
3770 (*digit) += value * 100;
3771
3772 RET0_UNLESS(parse_digit(date, len, &value));
3773 (*digit) += value * 10;
3774
3775 RET0_UNLESS(parse_digit(date, len, &value));
3776 (*digit) += value;
3777
3778 return 1;
3779}
3780
3781/* time-of-day = hour ":" minute ":" second
3782 * ; 00:00:00 - 23:59:60 (leap second)
3783 *
3784 * hour = 2DIGIT
3785 * minute = 2DIGIT
3786 * second = 2DIGIT
3787 */
3788static inline int parse_http_time(const char **date, int *len, struct tm *tm)
3789{
3790 RET0_UNLESS(parse_2digit(date, len, &tm->tm_hour)); /* hour 2DIGIT */
3791 RET0_UNLESS(parse_expect_char(date, len, ':')); /* expect ":" */
3792 RET0_UNLESS(parse_2digit(date, len, &tm->tm_min)); /* min 2DIGIT */
3793 RET0_UNLESS(parse_expect_char(date, len, ':')); /* expect ":" */
3794 RET0_UNLESS(parse_2digit(date, len, &tm->tm_sec)); /* sec 2DIGIT */
3795 return 1;
3796}
3797
3798/* From RFC7231
3799 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3800 *
3801 * IMF-fixdate = day-name "," SP date1 SP time-of-day SP GMT
3802 * ; fixed length/zone/capitalization subset of the format
3803 * ; see Section 3.3 of [RFC5322]
3804 *
3805 *
3806 * date1 = day SP month SP year
3807 * ; e.g., 02 Jun 1982
3808 *
3809 * day = 2DIGIT
3810 * year = 4DIGIT
3811 *
3812 * GMT = %x47.4D.54 ; "GMT", case-sensitive
3813 *
3814 * time-of-day = hour ":" minute ":" second
3815 * ; 00:00:00 - 23:59:60 (leap second)
3816 *
3817 * hour = 2DIGIT
3818 * minute = 2DIGIT
3819 * second = 2DIGIT
3820 *
3821 * DIGIT = decimal 0-9
3822 */
3823int parse_imf_date(const char *date, int len, struct tm *tm)
3824{
David Carlier327298c2016-11-20 10:42:38 +00003825 /* tm_gmtoff, if present, ought to be zero'ed */
3826 memset(tm, 0, sizeof(*tm));
3827
Thierry Fournier93127942016-01-20 18:49:45 +01003828 RET0_UNLESS(parse_http_dayname(&date, &len, tm)); /* day-name */
3829 RET0_UNLESS(parse_expect_char(&date, &len, ',')); /* expect "," */
3830 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3831 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
3832 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3833 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* Month */
3834 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3835 RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
3836 tm->tm_year -= 1900;
3837 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3838 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3839 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3840 RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3)); /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
3841 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003842 return 1;
3843}
3844
3845/* From RFC7231
3846 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3847 *
3848 * rfc850-date = day-name-l "," SP date2 SP time-of-day SP GMT
3849 * date2 = day "-" month "-" 2DIGIT
3850 * ; e.g., 02-Jun-82
3851 *
3852 * day = 2DIGIT
3853 */
3854int parse_rfc850_date(const char *date, int len, struct tm *tm)
3855{
3856 int year;
3857
David Carlier327298c2016-11-20 10:42:38 +00003858 /* tm_gmtoff, if present, ought to be zero'ed */
3859 memset(tm, 0, sizeof(*tm));
3860
Thierry Fournier93127942016-01-20 18:49:45 +01003861 RET0_UNLESS(parse_http_ldayname(&date, &len, tm)); /* Read the day name */
3862 RET0_UNLESS(parse_expect_char(&date, &len, ',')); /* expect "," */
3863 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3864 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
3865 RET0_UNLESS(parse_expect_char(&date, &len, '-')); /* expect "-" */
3866 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* Month */
3867 RET0_UNLESS(parse_expect_char(&date, &len, '-')); /* expect "-" */
3868
3869 /* year = 2DIGIT
3870 *
3871 * Recipients of a timestamp value in rfc850-(*date) format, which uses a
3872 * two-digit year, MUST interpret a timestamp that appears to be more
3873 * than 50 years in the future as representing the most recent year in
3874 * the past that had the same last two digits.
3875 */
3876 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_year));
3877
3878 /* expect SP */
3879 if (!parse_expect_char(&date, &len, ' ')) {
3880 /* Maybe we have the date with 4 digits. */
3881 RET0_UNLESS(parse_2digit(&date, &len, &year));
3882 tm->tm_year = (tm->tm_year * 100 + year) - 1900;
3883 /* expect SP */
3884 RET0_UNLESS(parse_expect_char(&date, &len, ' '));
3885 } else {
3886 /* I fix 60 as pivot: >60: +1900, <60: +2000. Note that the
3887 * tm_year is the number of year since 1900, so for +1900, we
3888 * do nothing, and for +2000, we add 100.
3889 */
3890 if (tm->tm_year <= 60)
3891 tm->tm_year += 100;
3892 }
3893
3894 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3895 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3896 RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3)); /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
3897 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003898
3899 return 1;
3900}
3901
3902/* From RFC7231
3903 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3904 *
3905 * asctime-date = day-name SP date3 SP time-of-day SP year
3906 * date3 = month SP ( 2DIGIT / ( SP 1DIGIT ))
3907 * ; e.g., Jun 2
3908 *
3909 * HTTP-date is case sensitive. A sender MUST NOT generate additional
3910 * whitespace in an HTTP-date beyond that specifically included as SP in
3911 * the grammar.
3912 */
3913int parse_asctime_date(const char *date, int len, struct tm *tm)
3914{
David Carlier327298c2016-11-20 10:42:38 +00003915 /* tm_gmtoff, if present, ought to be zero'ed */
3916 memset(tm, 0, sizeof(*tm));
3917
Thierry Fournier93127942016-01-20 18:49:45 +01003918 RET0_UNLESS(parse_http_dayname(&date, &len, tm)); /* day-name */
3919 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3920 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* expect month */
3921 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3922
3923 /* expect SP and 1DIGIT or 2DIGIT */
3924 if (parse_expect_char(&date, &len, ' '))
3925 RET0_UNLESS(parse_digit(&date, &len, &tm->tm_mday));
3926 else
3927 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday));
3928
3929 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3930 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3931 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3932 RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
3933 tm->tm_year -= 1900;
3934 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003935 return 1;
3936}
3937
3938/* From RFC7231
3939 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3940 *
3941 * HTTP-date = IMF-fixdate / obs-date
3942 * obs-date = rfc850-date / asctime-date
3943 *
3944 * parses an HTTP date in the RFC format and is accepted
3945 * alternatives. <date> is the strinf containing the date,
3946 * len is the len of the string. <tm> is filled with the
3947 * parsed time. We must considers this time as GMT.
3948 */
3949int parse_http_date(const char *date, int len, struct tm *tm)
3950{
3951 if (parse_imf_date(date, len, tm))
3952 return 1;
3953
3954 if (parse_rfc850_date(date, len, tm))
3955 return 1;
3956
3957 if (parse_asctime_date(date, len, tm))
3958 return 1;
3959
3960 return 0;
3961}
3962
Willy Tarreau4deeb102021-01-29 10:47:52 +01003963/* print the time <ns> in a short form (exactly 7 chars) at the end of buffer
3964 * <out>. "-" is printed if the value is zero, "inf" if larger than 1000 years.
3965 * It returns the new buffer length, or 0 if it doesn't fit. The value will be
3966 * surrounded by <pfx> and <sfx> respectively if not NULL.
3967 */
3968int print_time_short(struct buffer *out, const char *pfx, uint64_t ns, const char *sfx)
3969{
3970 double val = ns; // 52 bits of mantissa keep ns accuracy over 52 days
3971 const char *unit;
3972
3973 if (!pfx)
3974 pfx = "";
3975 if (!sfx)
3976 sfx = "";
3977
3978 do {
3979 unit = " - "; if (val <= 0.0) break;
3980 unit = "ns"; if (val < 1000.0) break;
3981 unit = "us"; val /= 1000.0; if (val < 1000.0) break;
3982 unit = "ms"; val /= 1000.0; if (val < 1000.0) break;
3983 unit = "s "; val /= 1000.0; if (val < 60.0) break;
3984 unit = "m "; val /= 60.0; if (val < 60.0) break;
3985 unit = "h "; val /= 60.0; if (val < 24.0) break;
3986 unit = "d "; val /= 24.0; if (val < 365.0) break;
3987 unit = "yr"; val /= 365.0; if (val < 1000.0) break;
3988 unit = " inf "; val = 0.0; break;
3989 } while (0);
3990
3991 if (val <= 0.0)
3992 return chunk_appendf(out, "%s%7s%s", pfx, unit, sfx);
3993 else if (val < 10.0)
3994 return chunk_appendf(out, "%s%1.3f%s%s", pfx, val, unit, sfx);
3995 else if (val < 100.0)
3996 return chunk_appendf(out, "%s%2.2f%s%s", pfx, val, unit, sfx);
3997 else
3998 return chunk_appendf(out, "%s%3.1f%s%s", pfx, val, unit, sfx);
3999}
4000
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004001/* Dynamically allocates a string of the proper length to hold the formatted
4002 * output. NULL is returned on error. The caller is responsible for freeing the
4003 * memory area using free(). The resulting string is returned in <out> if the
4004 * pointer is not NULL. A previous version of <out> might be used to build the
4005 * new string, and it will be freed before returning if it is not NULL, which
4006 * makes it possible to build complex strings from iterative calls without
4007 * having to care about freeing intermediate values, as in the example below :
4008 *
4009 * memprintf(&err, "invalid argument: '%s'", arg);
4010 * ...
4011 * memprintf(&err, "parser said : <%s>\n", *err);
4012 * ...
4013 * free(*err);
4014 *
4015 * This means that <err> must be initialized to NULL before first invocation.
4016 * The return value also holds the allocated string, which eases error checking
4017 * and immediate consumption. If the output pointer is not used, NULL must be
Willy Tarreaueb6cead2012-09-20 19:43:14 +02004018 * passed instead and it will be ignored. The returned message will then also
4019 * be NULL so that the caller does not have to bother with freeing anything.
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004020 *
4021 * It is also convenient to use it without any free except the last one :
4022 * err = NULL;
4023 * if (!fct1(err)) report(*err);
4024 * if (!fct2(err)) report(*err);
4025 * if (!fct3(err)) report(*err);
4026 * free(*err);
Christopher Faulet93a518f2017-10-24 11:25:33 +02004027 *
4028 * memprintf relies on memvprintf. This last version can be called from any
4029 * function with variadic arguments.
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004030 */
Christopher Faulet93a518f2017-10-24 11:25:33 +02004031char *memvprintf(char **out, const char *format, va_list orig_args)
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004032{
4033 va_list args;
4034 char *ret = NULL;
4035 int allocated = 0;
4036 int needed = 0;
4037
Willy Tarreaueb6cead2012-09-20 19:43:14 +02004038 if (!out)
4039 return NULL;
4040
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004041 do {
Willy Tarreaue0609f52019-03-29 19:13:23 +01004042 char buf1;
4043
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004044 /* vsnprintf() will return the required length even when the
4045 * target buffer is NULL. We do this in a loop just in case
4046 * intermediate evaluations get wrong.
4047 */
Christopher Faulet93a518f2017-10-24 11:25:33 +02004048 va_copy(args, orig_args);
Willy Tarreaue0609f52019-03-29 19:13:23 +01004049 needed = vsnprintf(ret ? ret : &buf1, allocated, format, args);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004050 va_end(args);
Willy Tarreau1b2fed62013-04-01 22:48:54 +02004051 if (needed < allocated) {
4052 /* Note: on Solaris 8, the first iteration always
4053 * returns -1 if allocated is zero, so we force a
4054 * retry.
4055 */
4056 if (!allocated)
4057 needed = 0;
4058 else
4059 break;
4060 }
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004061
Willy Tarreau1b2fed62013-04-01 22:48:54 +02004062 allocated = needed + 1;
Hubert Verstraete831962e2016-06-28 22:44:26 +02004063 ret = my_realloc2(ret, allocated);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004064 } while (ret);
4065
4066 if (needed < 0) {
4067 /* an error was encountered */
Willy Tarreau61cfdf42021-02-20 10:46:51 +01004068 ha_free(&ret);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004069 }
4070
4071 if (out) {
4072 free(*out);
4073 *out = ret;
4074 }
4075
4076 return ret;
4077}
William Lallemand421f5b52012-02-06 18:15:57 +01004078
Christopher Faulet93a518f2017-10-24 11:25:33 +02004079char *memprintf(char **out, const char *format, ...)
4080{
4081 va_list args;
4082 char *ret = NULL;
4083
4084 va_start(args, format);
4085 ret = memvprintf(out, format, args);
4086 va_end(args);
4087
4088 return ret;
4089}
4090
Willy Tarreau21c705b2012-09-14 11:40:36 +02004091/* Used to add <level> spaces before each line of <out>, unless there is only one line.
4092 * The input argument is automatically freed and reassigned. The result will have to be
Willy Tarreau70eec382012-10-10 08:56:47 +02004093 * freed by the caller. It also supports being passed a NULL which results in the same
4094 * output.
Willy Tarreau21c705b2012-09-14 11:40:36 +02004095 * Example of use :
4096 * parse(cmd, &err); (callee: memprintf(&err, ...))
4097 * fprintf(stderr, "Parser said: %s\n", indent_error(&err));
4098 * free(err);
4099 */
4100char *indent_msg(char **out, int level)
4101{
4102 char *ret, *in, *p;
4103 int needed = 0;
4104 int lf = 0;
4105 int lastlf = 0;
4106 int len;
4107
Willy Tarreau70eec382012-10-10 08:56:47 +02004108 if (!out || !*out)
4109 return NULL;
4110
Willy Tarreau21c705b2012-09-14 11:40:36 +02004111 in = *out - 1;
4112 while ((in = strchr(in + 1, '\n')) != NULL) {
4113 lastlf = in - *out;
4114 lf++;
4115 }
4116
4117 if (!lf) /* single line, no LF, return it as-is */
4118 return *out;
4119
4120 len = strlen(*out);
4121
4122 if (lf == 1 && lastlf == len - 1) {
4123 /* single line, LF at end, strip it and return as-is */
4124 (*out)[lastlf] = 0;
4125 return *out;
4126 }
4127
4128 /* OK now we have at least one LF, we need to process the whole string
4129 * as a multi-line string. What we'll do :
4130 * - prefix with an LF if there is none
4131 * - add <level> spaces before each line
4132 * This means at most ( 1 + level + (len-lf) + lf*<1+level) ) =
4133 * 1 + level + len + lf * level = 1 + level * (lf + 1) + len.
4134 */
4135
4136 needed = 1 + level * (lf + 1) + len + 1;
4137 p = ret = malloc(needed);
4138 in = *out;
4139
4140 /* skip initial LFs */
4141 while (*in == '\n')
4142 in++;
4143
4144 /* copy each line, prefixed with LF and <level> spaces, and without the trailing LF */
4145 while (*in) {
4146 *p++ = '\n';
4147 memset(p, ' ', level);
4148 p += level;
4149 do {
4150 *p++ = *in++;
4151 } while (*in && *in != '\n');
4152 if (*in)
4153 in++;
4154 }
4155 *p = 0;
4156
4157 free(*out);
4158 *out = ret;
4159
4160 return ret;
4161}
4162
Willy Tarreaua2c99112019-08-21 13:17:37 +02004163/* makes a copy of message <in> into <out>, with each line prefixed with <pfx>
4164 * and end of lines replaced with <eol> if not 0. The first line to indent has
4165 * to be indicated in <first> (starts at zero), so that it is possible to skip
4166 * indenting the first line if it has to be appended after an existing message.
4167 * Empty strings are never indented, and NULL strings are considered empty both
4168 * for <in> and <pfx>. It returns non-zero if an EOL was appended as the last
4169 * character, non-zero otherwise.
4170 */
4171int append_prefixed_str(struct buffer *out, const char *in, const char *pfx, char eol, int first)
4172{
4173 int bol, lf;
4174 int pfxlen = pfx ? strlen(pfx) : 0;
4175
4176 if (!in)
4177 return 0;
4178
4179 bol = 1;
4180 lf = 0;
4181 while (*in) {
4182 if (bol && pfxlen) {
4183 if (first > 0)
4184 first--;
4185 else
4186 b_putblk(out, pfx, pfxlen);
4187 bol = 0;
4188 }
4189
4190 lf = (*in == '\n');
4191 bol |= lf;
4192 b_putchr(out, (lf && eol) ? eol : *in);
4193 in++;
4194 }
4195 return lf;
4196}
4197
Willy Tarreau9d22e562019-03-29 18:49:09 +01004198/* removes environment variable <name> from the environment as found in
4199 * environ. This is only provided as an alternative for systems without
4200 * unsetenv() (old Solaris and AIX versions). THIS IS NOT THREAD SAFE.
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05004201 * The principle is to scan environ for each occurrence of variable name
Willy Tarreau9d22e562019-03-29 18:49:09 +01004202 * <name> and to replace the matching pointers with the last pointer of
4203 * the array (since variables are not ordered).
4204 * It always returns 0 (success).
4205 */
4206int my_unsetenv(const char *name)
4207{
4208 extern char **environ;
4209 char **p = environ;
4210 int vars;
4211 int next;
4212 int len;
4213
4214 len = strlen(name);
4215 for (vars = 0; p[vars]; vars++)
4216 ;
4217 next = 0;
4218 while (next < vars) {
4219 if (strncmp(p[next], name, len) != 0 || p[next][len] != '=') {
4220 next++;
4221 continue;
4222 }
4223 if (next < vars - 1)
4224 p[next] = p[vars - 1];
4225 p[--vars] = NULL;
4226 }
4227 return 0;
4228}
4229
Willy Tarreaudad36a32013-03-11 01:20:04 +01004230/* Convert occurrences of environment variables in the input string to their
4231 * corresponding value. A variable is identified as a series of alphanumeric
4232 * characters or underscores following a '$' sign. The <in> string must be
4233 * free()able. NULL returns NULL. The resulting string might be reallocated if
4234 * some expansion is made. Variable names may also be enclosed into braces if
4235 * needed (eg: to concatenate alphanum characters).
4236 */
4237char *env_expand(char *in)
4238{
4239 char *txt_beg;
4240 char *out;
4241 char *txt_end;
4242 char *var_beg;
4243 char *var_end;
4244 char *value;
4245 char *next;
4246 int out_len;
4247 int val_len;
4248
4249 if (!in)
4250 return in;
4251
4252 value = out = NULL;
4253 out_len = 0;
4254
4255 txt_beg = in;
4256 do {
4257 /* look for next '$' sign in <in> */
4258 for (txt_end = txt_beg; *txt_end && *txt_end != '$'; txt_end++);
4259
4260 if (!*txt_end && !out) /* end and no expansion performed */
4261 return in;
4262
4263 val_len = 0;
4264 next = txt_end;
4265 if (*txt_end == '$') {
4266 char save;
4267
4268 var_beg = txt_end + 1;
4269 if (*var_beg == '{')
4270 var_beg++;
4271
4272 var_end = var_beg;
Willy Tarreau90807112020-02-25 08:16:33 +01004273 while (isalnum((unsigned char)*var_end) || *var_end == '_') {
Willy Tarreaudad36a32013-03-11 01:20:04 +01004274 var_end++;
4275 }
4276
4277 next = var_end;
4278 if (*var_end == '}' && (var_beg > txt_end + 1))
4279 next++;
4280
4281 /* get value of the variable name at this location */
4282 save = *var_end;
4283 *var_end = '\0';
4284 value = getenv(var_beg);
4285 *var_end = save;
4286 val_len = value ? strlen(value) : 0;
4287 }
4288
Hubert Verstraete831962e2016-06-28 22:44:26 +02004289 out = my_realloc2(out, out_len + (txt_end - txt_beg) + val_len + 1);
Willy Tarreaudad36a32013-03-11 01:20:04 +01004290 if (txt_end > txt_beg) {
4291 memcpy(out + out_len, txt_beg, txt_end - txt_beg);
4292 out_len += txt_end - txt_beg;
4293 }
4294 if (val_len) {
4295 memcpy(out + out_len, value, val_len);
4296 out_len += val_len;
4297 }
4298 out[out_len] = 0;
4299 txt_beg = next;
4300 } while (*txt_beg);
4301
4302 /* here we know that <out> was allocated and that we don't need <in> anymore */
4303 free(in);
4304 return out;
4305}
4306
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004307
4308/* same as strstr() but case-insensitive and with limit length */
4309const char *strnistr(const char *str1, int len_str1, const char *str2, int len_str2)
4310{
4311 char *pptr, *sptr, *start;
Willy Tarreauc8746532014-05-28 23:05:07 +02004312 unsigned int slen, plen;
4313 unsigned int tmp1, tmp2;
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004314
4315 if (str1 == NULL || len_str1 == 0) // search pattern into an empty string => search is not found
4316 return NULL;
4317
4318 if (str2 == NULL || len_str2 == 0) // pattern is empty => every str1 match
4319 return str1;
4320
4321 if (len_str1 < len_str2) // pattern is longer than string => search is not found
4322 return NULL;
4323
4324 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 +02004325 while (toupper((unsigned char)*start) != toupper((unsigned char)*str2)) {
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004326 start++;
4327 slen--;
4328 tmp1++;
4329
4330 if (tmp1 >= len_str1)
4331 return NULL;
4332
4333 /* if pattern longer than string */
4334 if (slen < plen)
4335 return NULL;
4336 }
4337
4338 sptr = start;
4339 pptr = (char *)str2;
4340
4341 tmp2 = 0;
Willy Tarreauf278eec2020-07-05 21:46:32 +02004342 while (toupper((unsigned char)*sptr) == toupper((unsigned char)*pptr)) {
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004343 sptr++;
4344 pptr++;
4345 tmp2++;
4346
4347 if (*pptr == '\0' || tmp2 == len_str2) /* end of pattern found */
4348 return start;
4349 if (*sptr == '\0' || tmp2 == len_str1) /* end of string found and the pattern is not fully found */
4350 return NULL;
4351 }
4352 }
4353 return NULL;
4354}
4355
Thierry FOURNIER317e1c42014-08-12 10:20:47 +02004356/* This function read the next valid utf8 char.
4357 * <s> is the byte srray to be decode, <len> is its length.
4358 * The function returns decoded char encoded like this:
4359 * The 4 msb are the return code (UTF8_CODE_*), the 4 lsb
4360 * are the length read. The decoded character is stored in <c>.
4361 */
4362unsigned char utf8_next(const char *s, int len, unsigned int *c)
4363{
4364 const unsigned char *p = (unsigned char *)s;
4365 int dec;
4366 unsigned char code = UTF8_CODE_OK;
4367
4368 if (len < 1)
4369 return UTF8_CODE_OK;
4370
4371 /* Check the type of UTF8 sequence
4372 *
4373 * 0... .... 0x00 <= x <= 0x7f : 1 byte: ascii char
4374 * 10.. .... 0x80 <= x <= 0xbf : invalid sequence
4375 * 110. .... 0xc0 <= x <= 0xdf : 2 bytes
4376 * 1110 .... 0xe0 <= x <= 0xef : 3 bytes
4377 * 1111 0... 0xf0 <= x <= 0xf7 : 4 bytes
4378 * 1111 10.. 0xf8 <= x <= 0xfb : 5 bytes
4379 * 1111 110. 0xfc <= x <= 0xfd : 6 bytes
4380 * 1111 111. 0xfe <= x <= 0xff : invalid sequence
4381 */
4382 switch (*p) {
4383 case 0x00 ... 0x7f:
4384 *c = *p;
4385 return UTF8_CODE_OK | 1;
4386
4387 case 0x80 ... 0xbf:
4388 *c = *p;
4389 return UTF8_CODE_BADSEQ | 1;
4390
4391 case 0xc0 ... 0xdf:
4392 if (len < 2) {
4393 *c = *p;
4394 return UTF8_CODE_BADSEQ | 1;
4395 }
4396 *c = *p & 0x1f;
4397 dec = 1;
4398 break;
4399
4400 case 0xe0 ... 0xef:
4401 if (len < 3) {
4402 *c = *p;
4403 return UTF8_CODE_BADSEQ | 1;
4404 }
4405 *c = *p & 0x0f;
4406 dec = 2;
4407 break;
4408
4409 case 0xf0 ... 0xf7:
4410 if (len < 4) {
4411 *c = *p;
4412 return UTF8_CODE_BADSEQ | 1;
4413 }
4414 *c = *p & 0x07;
4415 dec = 3;
4416 break;
4417
4418 case 0xf8 ... 0xfb:
4419 if (len < 5) {
4420 *c = *p;
4421 return UTF8_CODE_BADSEQ | 1;
4422 }
4423 *c = *p & 0x03;
4424 dec = 4;
4425 break;
4426
4427 case 0xfc ... 0xfd:
4428 if (len < 6) {
4429 *c = *p;
4430 return UTF8_CODE_BADSEQ | 1;
4431 }
4432 *c = *p & 0x01;
4433 dec = 5;
4434 break;
4435
4436 case 0xfe ... 0xff:
4437 default:
4438 *c = *p;
4439 return UTF8_CODE_BADSEQ | 1;
4440 }
4441
4442 p++;
4443
4444 while (dec > 0) {
4445
4446 /* need 0x10 for the 2 first bits */
4447 if ( ( *p & 0xc0 ) != 0x80 )
4448 return UTF8_CODE_BADSEQ | ((p-(unsigned char *)s)&0xffff);
4449
4450 /* add data at char */
4451 *c = ( *c << 6 ) | ( *p & 0x3f );
4452
4453 dec--;
4454 p++;
4455 }
4456
4457 /* Check ovelong encoding.
4458 * 1 byte : 5 + 6 : 11 : 0x80 ... 0x7ff
4459 * 2 bytes : 4 + 6 + 6 : 16 : 0x800 ... 0xffff
4460 * 3 bytes : 3 + 6 + 6 + 6 : 21 : 0x10000 ... 0x1fffff
4461 */
Thierry FOURNIER9e7ec082015-03-12 19:32:38 +01004462 if (( *c <= 0x7f && (p-(unsigned char *)s) > 1) ||
Thierry FOURNIER317e1c42014-08-12 10:20:47 +02004463 (*c >= 0x80 && *c <= 0x7ff && (p-(unsigned char *)s) > 2) ||
4464 (*c >= 0x800 && *c <= 0xffff && (p-(unsigned char *)s) > 3) ||
4465 (*c >= 0x10000 && *c <= 0x1fffff && (p-(unsigned char *)s) > 4))
4466 code |= UTF8_CODE_OVERLONG;
4467
4468 /* Check invalid UTF8 range. */
4469 if ((*c >= 0xd800 && *c <= 0xdfff) ||
4470 (*c >= 0xfffe && *c <= 0xffff))
4471 code |= UTF8_CODE_INVRANGE;
4472
4473 return code | ((p-(unsigned char *)s)&0x0f);
4474}
4475
Maxime de Roucydc887852016-05-13 23:52:54 +02004476/* append a copy of string <str> (in a wordlist) at the end of the list <li>
4477 * On failure : return 0 and <err> filled with an error message.
4478 * The caller is responsible for freeing the <err> and <str> copy
4479 * memory area using free()
4480 */
4481int list_append_word(struct list *li, const char *str, char **err)
4482{
4483 struct wordlist *wl;
4484
4485 wl = calloc(1, sizeof(*wl));
4486 if (!wl) {
4487 memprintf(err, "out of memory");
4488 goto fail_wl;
4489 }
4490
4491 wl->s = strdup(str);
4492 if (!wl->s) {
4493 memprintf(err, "out of memory");
4494 goto fail_wl_s;
4495 }
4496
Willy Tarreau2b718102021-04-21 07:32:39 +02004497 LIST_APPEND(li, &wl->list);
Maxime de Roucydc887852016-05-13 23:52:54 +02004498
4499 return 1;
4500
4501fail_wl_s:
4502 free(wl->s);
4503fail_wl:
4504 free(wl);
4505 return 0;
4506}
4507
Willy Tarreau37101052019-05-20 16:48:20 +02004508/* indicates if a memory location may safely be read or not. The trick consists
4509 * in performing a harmless syscall using this location as an input and letting
4510 * the operating system report whether it's OK or not. For this we have the
4511 * stat() syscall, which will return EFAULT when the memory location supposed
4512 * to contain the file name is not readable. If it is readable it will then
4513 * either return 0 if the area contains an existing file name, or -1 with
4514 * another code. This must not be abused, and some audit systems might detect
4515 * this as abnormal activity. It's used only for unsafe dumps.
4516 */
4517int may_access(const void *ptr)
4518{
4519 struct stat buf;
4520
4521 if (stat(ptr, &buf) == 0)
4522 return 1;
4523 if (errno == EFAULT)
4524 return 0;
4525 return 1;
4526}
4527
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004528/* print a string of text buffer to <out>. The format is :
4529 * Non-printable chars \t, \n, \r and \e are * encoded in C format.
4530 * Other non-printable chars are encoded "\xHH". Space, '\', and '=' are also escaped.
4531 * Print stopped if null char or <bsize> is reached, or if no more place in the chunk.
4532 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004533int dump_text(struct buffer *out, const char *buf, int bsize)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004534{
4535 unsigned char c;
Tim Duesterhuscd5521e2021-08-29 00:58:22 +02004536 size_t ptr = 0;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004537
Tim Duesterhuscd5521e2021-08-29 00:58:22 +02004538 while (ptr < bsize && buf[ptr]) {
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004539 c = buf[ptr];
Willy Tarreau90807112020-02-25 08:16:33 +01004540 if (isprint((unsigned char)c) && isascii((unsigned char)c) && c != '\\' && c != ' ' && c != '=') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004541 if (out->data > out->size - 1)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004542 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004543 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004544 }
4545 else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\' || c == ' ' || c == '=') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004546 if (out->data > out->size - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004547 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004548 out->area[out->data++] = '\\';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004549 switch (c) {
4550 case ' ': c = ' '; break;
4551 case '\t': c = 't'; break;
4552 case '\n': c = 'n'; break;
4553 case '\r': c = 'r'; break;
4554 case '\e': c = 'e'; break;
4555 case '\\': c = '\\'; break;
4556 case '=': c = '='; break;
4557 }
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004558 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004559 }
4560 else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004561 if (out->data > out->size - 4)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004562 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004563 out->area[out->data++] = '\\';
4564 out->area[out->data++] = 'x';
4565 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4566 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004567 }
4568 ptr++;
4569 }
4570
4571 return ptr;
4572}
4573
4574/* print a buffer in hexa.
4575 * Print stopped if <bsize> is reached, or if no more place in the chunk.
4576 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004577int dump_binary(struct buffer *out, const char *buf, int bsize)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004578{
4579 unsigned char c;
4580 int ptr = 0;
4581
4582 while (ptr < bsize) {
4583 c = buf[ptr];
4584
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004585 if (out->data > out->size - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004586 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004587 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4588 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004589
4590 ptr++;
4591 }
4592 return ptr;
4593}
4594
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004595/* Appends into buffer <out> a hex dump of memory area <buf> for <len> bytes,
4596 * prepending each line with prefix <pfx>. The output is *not* initialized.
4597 * The output will not wrap pas the buffer's end so it is more optimal if the
4598 * caller makes sure the buffer is aligned first. A trailing zero will always
4599 * be appended (and not counted) if there is room for it. The caller must make
Willy Tarreau37101052019-05-20 16:48:20 +02004600 * sure that the area is dumpable first. If <unsafe> is non-null, the memory
4601 * locations are checked first for being readable.
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004602 */
Willy Tarreau37101052019-05-20 16:48:20 +02004603void dump_hex(struct buffer *out, const char *pfx, const void *buf, int len, int unsafe)
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004604{
4605 const unsigned char *d = buf;
4606 int i, j, start;
4607
4608 d = (const unsigned char *)(((unsigned long)buf) & -16);
4609 start = ((unsigned long)buf) & 15;
4610
4611 for (i = 0; i < start + len; i += 16) {
4612 chunk_appendf(out, (sizeof(void *) == 4) ? "%s%8p: " : "%s%16p: ", pfx, d + i);
4613
Willy Tarreau37101052019-05-20 16:48:20 +02004614 // 0: unchecked, 1: checked safe, 2: danger
4615 unsafe = !!unsafe;
4616 if (unsafe && !may_access(d + i))
4617 unsafe = 2;
4618
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004619 for (j = 0; j < 16; j++) {
Willy Tarreau37101052019-05-20 16:48:20 +02004620 if ((i + j < start) || (i + j >= start + len))
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004621 chunk_strcat(out, "'' ");
Willy Tarreau37101052019-05-20 16:48:20 +02004622 else if (unsafe > 1)
4623 chunk_strcat(out, "** ");
4624 else
4625 chunk_appendf(out, "%02x ", d[i + j]);
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004626
4627 if (j == 7)
4628 chunk_strcat(out, "- ");
4629 }
4630 chunk_strcat(out, " ");
4631 for (j = 0; j < 16; j++) {
Willy Tarreau37101052019-05-20 16:48:20 +02004632 if ((i + j < start) || (i + j >= start + len))
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004633 chunk_strcat(out, "'");
Willy Tarreau37101052019-05-20 16:48:20 +02004634 else if (unsafe > 1)
4635 chunk_strcat(out, "*");
Willy Tarreau90807112020-02-25 08:16:33 +01004636 else if (isprint((unsigned char)d[i + j]))
Willy Tarreau37101052019-05-20 16:48:20 +02004637 chunk_appendf(out, "%c", d[i + j]);
4638 else
4639 chunk_strcat(out, ".");
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004640 }
4641 chunk_strcat(out, "\n");
4642 }
4643}
4644
Willy Tarreau762fb3e2020-03-03 15:57:10 +01004645/* dumps <pfx> followed by <n> bytes from <addr> in hex form into buffer <buf>
4646 * enclosed in brackets after the address itself, formatted on 14 chars
4647 * including the "0x" prefix. This is meant to be used as a prefix for code
4648 * areas. For example:
4649 * "0x7f10b6557690 [48 c7 c0 0f 00 00 00 0f]"
4650 * It relies on may_access() to know if the bytes are dumpable, otherwise "--"
4651 * is emitted. A NULL <pfx> will be considered empty.
4652 */
4653void dump_addr_and_bytes(struct buffer *buf, const char *pfx, const void *addr, int n)
4654{
4655 int ok = 0;
4656 int i;
4657
4658 chunk_appendf(buf, "%s%#14lx [", pfx ? pfx : "", (long)addr);
4659
4660 for (i = 0; i < n; i++) {
4661 if (i == 0 || (((long)(addr + i) ^ (long)(addr)) & 4096))
4662 ok = may_access(addr + i);
4663 if (ok)
4664 chunk_appendf(buf, "%02x%s", ((uint8_t*)addr)[i], (i<n-1) ? " " : "]");
4665 else
4666 chunk_appendf(buf, "--%s", (i<n-1) ? " " : "]");
4667 }
4668}
4669
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004670/* print a line of text buffer (limited to 70 bytes) to <out>. The format is :
4671 * <2 spaces> <offset=5 digits> <space or plus> <space> <70 chars max> <\n>
4672 * which is 60 chars per line. Non-printable chars \t, \n, \r and \e are
4673 * encoded in C format. Other non-printable chars are encoded "\xHH". Original
4674 * lines are respected within the limit of 70 output chars. Lines that are
4675 * continuation of a previous truncated line begin with "+" instead of " "
4676 * after the offset. The new pointer is returned.
4677 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004678int dump_text_line(struct buffer *out, const char *buf, int bsize, int len,
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004679 int *line, int ptr)
4680{
4681 int end;
4682 unsigned char c;
4683
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004684 end = out->data + 80;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004685 if (end > out->size)
4686 return ptr;
4687
4688 chunk_appendf(out, " %05d%c ", ptr, (ptr == *line) ? ' ' : '+');
4689
4690 while (ptr < len && ptr < bsize) {
4691 c = buf[ptr];
Willy Tarreau90807112020-02-25 08:16:33 +01004692 if (isprint((unsigned char)c) && isascii((unsigned char)c) && c != '\\') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004693 if (out->data > end - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004694 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004695 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004696 } else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004697 if (out->data > end - 3)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004698 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004699 out->area[out->data++] = '\\';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004700 switch (c) {
4701 case '\t': c = 't'; break;
4702 case '\n': c = 'n'; break;
4703 case '\r': c = 'r'; break;
4704 case '\e': c = 'e'; break;
4705 case '\\': c = '\\'; break;
4706 }
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004707 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004708 } else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004709 if (out->data > end - 5)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004710 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004711 out->area[out->data++] = '\\';
4712 out->area[out->data++] = 'x';
4713 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4714 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004715 }
4716 if (buf[ptr++] == '\n') {
4717 /* we had a line break, let's return now */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004718 out->area[out->data++] = '\n';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004719 *line = ptr;
4720 return ptr;
4721 }
4722 }
4723 /* we have an incomplete line, we return it as-is */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004724 out->area[out->data++] = '\n';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004725 return ptr;
4726}
4727
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004728/* displays a <len> long memory block at <buf>, assuming first byte of <buf>
Willy Tarreaued936c52017-04-27 18:03:20 +02004729 * has address <baseaddr>. String <pfx> may be placed as a prefix in front of
4730 * each line. It may be NULL if unused. The output is emitted to file <out>.
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004731 */
Willy Tarreaued936c52017-04-27 18:03:20 +02004732void debug_hexdump(FILE *out, const char *pfx, const char *buf,
4733 unsigned int baseaddr, int len)
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004734{
Willy Tarreau73459792017-04-11 07:58:08 +02004735 unsigned int i;
4736 int b, j;
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004737
4738 for (i = 0; i < (len + (baseaddr & 15)); i += 16) {
4739 b = i - (baseaddr & 15);
Willy Tarreaued936c52017-04-27 18:03:20 +02004740 fprintf(out, "%s%08x: ", pfx ? pfx : "", i + (baseaddr & ~15));
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004741 for (j = 0; j < 8; j++) {
4742 if (b + j >= 0 && b + j < len)
4743 fprintf(out, "%02x ", (unsigned char)buf[b + j]);
4744 else
4745 fprintf(out, " ");
4746 }
4747
4748 if (b + j >= 0 && b + j < len)
4749 fputc('-', out);
4750 else
4751 fputc(' ', out);
4752
4753 for (j = 8; j < 16; j++) {
4754 if (b + j >= 0 && b + j < len)
4755 fprintf(out, " %02x", (unsigned char)buf[b + j]);
4756 else
4757 fprintf(out, " ");
4758 }
4759
4760 fprintf(out, " ");
4761 for (j = 0; j < 16; j++) {
4762 if (b + j >= 0 && b + j < len) {
4763 if (isprint((unsigned char)buf[b + j]))
4764 fputc((unsigned char)buf[b + j], out);
4765 else
4766 fputc('.', out);
4767 }
4768 else
4769 fputc(' ', out);
4770 }
4771 fputc('\n', out);
4772 }
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004773}
4774
Willy Tarreaubb869862020-04-16 10:52:41 +02004775/* Tries to report the executable path name on platforms supporting this. If
4776 * not found or not possible, returns NULL.
4777 */
4778const char *get_exec_path()
4779{
4780 const char *ret = NULL;
4781
Willy Tarreau1ee71dd2021-08-30 10:15:35 +02004782#if defined(__GLIBC__) && (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 16))
Willy Tarreaubb869862020-04-16 10:52:41 +02004783 long execfn = getauxval(AT_EXECFN);
4784
4785 if (execfn && execfn != ENOENT)
4786 ret = (const char *)execfn;
devnexen@gmail.com49a32282021-08-17 12:55:49 +01004787#elif defined(__FreeBSD__)
4788 Elf_Auxinfo *auxv;
4789 for (auxv = __elf_aux_vector; auxv->a_type != AT_NULL; ++auxv) {
4790 if (auxv->a_type == AT_EXECPATH) {
4791 ret = (const char *)auxv->a_un.a_ptr;
4792 break;
4793 }
4794 }
David Carlier1b9d57d2021-08-17 08:44:25 +01004795#elif defined(__NetBSD__)
4796 AuxInfo *auxv;
4797 for (auxv = _dlauxinfo(); auxv->a_type != AT_NULL; ++auxv) {
4798 if (auxv->a_type == AT_SUN_EXECNAME) {
4799 ret = (const char *)auxv->a_v;
4800 break;
4801 }
4802 }
David Carlier451b06f2022-05-14 17:15:49 +01004803#elif defined(__sun)
4804 ret = getexecname();
Willy Tarreaubb869862020-04-16 10:52:41 +02004805#endif
4806 return ret;
4807}
4808
Baruch Siache1651b22020-07-24 07:52:20 +03004809#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreau9133e482020-03-04 10:19:36 +01004810/* calls dladdr() or dladdr1() on <addr> and <dli>. If dladdr1 is available,
4811 * also returns the symbol size in <size>, otherwise returns 0 there.
4812 */
4813static int dladdr_and_size(const void *addr, Dl_info *dli, size_t *size)
4814{
4815 int ret;
Willy Tarreau1ee71dd2021-08-30 10:15:35 +02004816#if defined(__GLIBC__) && (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 3)) // most detailed one
Willy Tarreau9133e482020-03-04 10:19:36 +01004817 const ElfW(Sym) *sym;
4818
4819 ret = dladdr1(addr, dli, (void **)&sym, RTLD_DL_SYMENT);
4820 if (ret)
4821 *size = sym ? sym->st_size : 0;
4822#else
David Carlier7b6de262021-12-31 08:15:29 +00004823#if defined(__sun)
4824 ret = dladdr((void *)addr, dli);
4825#else
Willy Tarreau9133e482020-03-04 10:19:36 +01004826 ret = dladdr(addr, dli);
David Carlier7b6de262021-12-31 08:15:29 +00004827#endif
Willy Tarreau9133e482020-03-04 10:19:36 +01004828 *size = 0;
4829#endif
4830 return ret;
4831}
Willy Tarreau64192392021-05-05 09:06:21 +02004832
Willy Tarreau263aacd2022-07-18 13:58:17 +02004833/* Sets build_is_static to true if we detect a static build. Some older glibcs
4834 * tend to crash inside dlsym() in static builds, but tests show that at least
4835 * dladdr() still works (and will fail to resolve anything of course). Thus we
4836 * try to determine if we're on a static build to avoid calling dlsym() in this
4837 * case.
Willy Tarreauf72f24e2022-07-16 13:49:34 +02004838 */
Willy Tarreau263aacd2022-07-18 13:58:17 +02004839void check_if_static_build()
Willy Tarreauf72f24e2022-07-16 13:49:34 +02004840{
Willy Tarreau263aacd2022-07-18 13:58:17 +02004841 Dl_info dli = { };
4842 size_t size = 0;
4843
4844 /* Now let's try to be smarter */
4845 if (!dladdr_and_size(&main, &dli, &size))
4846 build_is_static = 1;
4847 else
4848 build_is_static = 0;
Willy Tarreauf72f24e2022-07-16 13:49:34 +02004849}
4850
Willy Tarreau263aacd2022-07-18 13:58:17 +02004851INITCALL0(STG_PREPARE, check_if_static_build);
Willy Tarreauf72f24e2022-07-16 13:49:34 +02004852
Willy Tarreau64192392021-05-05 09:06:21 +02004853/* Tries to retrieve the address of the first occurrence symbol <name>.
4854 * Note that NULL in return is not always an error as a symbol may have that
4855 * address in special situations.
4856 */
4857void *get_sym_curr_addr(const char *name)
4858{
4859 void *ptr = NULL;
4860
4861#ifdef RTLD_DEFAULT
Willy Tarreau263aacd2022-07-18 13:58:17 +02004862 if (!build_is_static)
Willy Tarreauf72f24e2022-07-16 13:49:34 +02004863 ptr = dlsym(RTLD_DEFAULT, name);
Willy Tarreau64192392021-05-05 09:06:21 +02004864#endif
4865 return ptr;
4866}
4867
4868
4869/* Tries to retrieve the address of the next occurrence of symbol <name>
4870 * Note that NULL in return is not always an error as a symbol may have that
4871 * address in special situations.
4872 */
4873void *get_sym_next_addr(const char *name)
4874{
4875 void *ptr = NULL;
4876
4877#ifdef RTLD_NEXT
Willy Tarreau263aacd2022-07-18 13:58:17 +02004878 if (!build_is_static)
Willy Tarreauf72f24e2022-07-16 13:49:34 +02004879 ptr = dlsym(RTLD_NEXT, name);
Willy Tarreau9133e482020-03-04 10:19:36 +01004880#endif
Willy Tarreau64192392021-05-05 09:06:21 +02004881 return ptr;
4882}
4883
4884#else /* elf & linux & dl */
4885
4886/* no possible resolving on other platforms at the moment */
4887void *get_sym_curr_addr(const char *name)
4888{
4889 return NULL;
4890}
4891
4892void *get_sym_next_addr(const char *name)
4893{
4894 return NULL;
4895}
4896
4897#endif /* elf & linux & dl */
Willy Tarreau9133e482020-03-04 10:19:36 +01004898
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004899/* Tries to append to buffer <buf> some indications about the symbol at address
4900 * <addr> using the following form:
4901 * lib:+0xoffset (unresolvable address from lib's base)
4902 * main+0xoffset (unresolvable address from main (+/-))
4903 * lib:main+0xoffset (unresolvable lib address from main (+/-))
4904 * name (resolved exact exec address)
4905 * lib:name (resolved exact lib address)
4906 * name+0xoffset/0xsize (resolved address within exec symbol)
4907 * lib:name+0xoffset/0xsize (resolved address within lib symbol)
4908 *
4909 * The file name (lib or executable) is limited to what lies between the last
4910 * '/' and the first following '.'. An optional prefix <pfx> is prepended before
4911 * 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 +03004912 * that contains the "main" symbol, or when __ELF__ && USE_DL are not set.
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004913 *
4914 * The symbol's base address is returned, or NULL when unresolved, in order to
4915 * allow the caller to match it against known ones.
4916 */
Willy Tarreau45fd1032021-01-20 14:37:59 +01004917const void *resolve_sym_name(struct buffer *buf, const char *pfx, const void *addr)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004918{
4919 const struct {
4920 const void *func;
4921 const char *name;
4922 } fcts[] = {
4923 { .func = process_stream, .name = "process_stream" },
4924 { .func = task_run_applet, .name = "task_run_applet" },
4925 { .func = si_cs_io_cb, .name = "si_cs_io_cb" },
Willy Tarreau586f71b2020-12-11 15:54:36 +01004926 { .func = sock_conn_iocb, .name = "sock_conn_iocb" },
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004927 { .func = dgram_fd_handler, .name = "dgram_fd_handler" },
4928 { .func = listener_accept, .name = "listener_accept" },
Willy Tarreaud597ec22021-01-29 14:29:06 +01004929 { .func = manage_global_listener_queue, .name = "manage_global_listener_queue" },
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004930 { .func = poller_pipe_io_handler, .name = "poller_pipe_io_handler" },
4931 { .func = mworker_accept_wrapper, .name = "mworker_accept_wrapper" },
Willy Tarreau02922e12021-01-29 12:27:57 +01004932 { .func = session_expire_embryonic, .name = "session_expire_embryonic" },
Willy Tarreaufb5401f2021-01-29 12:25:23 +01004933#ifdef USE_THREAD
4934 { .func = accept_queue_process, .name = "accept_queue_process" },
4935#endif
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004936#ifdef USE_LUA
4937 { .func = hlua_process_task, .name = "hlua_process_task" },
4938#endif
Ilya Shipitsinbdec3ba2020-11-14 01:56:34 +05004939#ifdef SSL_MODE_ASYNC
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004940 { .func = ssl_async_fd_free, .name = "ssl_async_fd_free" },
4941 { .func = ssl_async_fd_handler, .name = "ssl_async_fd_handler" },
4942#endif
4943 };
4944
Baruch Siache1651b22020-07-24 07:52:20 +03004945#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004946 Dl_info dli, dli_main;
Willy Tarreau9133e482020-03-04 10:19:36 +01004947 size_t size;
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004948 const char *fname, *p;
4949#endif
4950 int i;
4951
4952 if (pfx)
4953 chunk_appendf(buf, "%s", pfx);
4954
4955 for (i = 0; i < sizeof(fcts) / sizeof(fcts[0]); i++) {
4956 if (addr == fcts[i].func) {
4957 chunk_appendf(buf, "%s", fcts[i].name);
4958 return addr;
4959 }
4960 }
4961
Baruch Siache1651b22020-07-24 07:52:20 +03004962#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004963 /* Now let's try to be smarter */
Willy Tarreau9133e482020-03-04 10:19:36 +01004964 if (!dladdr_and_size(addr, &dli, &size))
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004965 goto unknown;
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004966
4967 /* 1. prefix the library name if it's not the same object as the one
4968 * that contains the main function. The name is picked between last '/'
4969 * and first following '.'.
4970 */
4971 if (!dladdr(main, &dli_main))
4972 dli_main.dli_fbase = NULL;
4973
4974 if (dli_main.dli_fbase != dli.dli_fbase) {
4975 fname = dli.dli_fname;
4976 p = strrchr(fname, '/');
4977 if (p++)
4978 fname = p;
4979 p = strchr(fname, '.');
4980 if (!p)
4981 p = fname + strlen(fname);
4982
4983 chunk_appendf(buf, "%.*s:", (int)(long)(p - fname), fname);
4984 }
4985
4986 /* 2. symbol name */
4987 if (dli.dli_sname) {
4988 /* known, dump it and return symbol's address (exact or relative) */
4989 chunk_appendf(buf, "%s", dli.dli_sname);
4990 if (addr != dli.dli_saddr) {
4991 chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_saddr));
Willy Tarreau9133e482020-03-04 10:19:36 +01004992 if (size)
4993 chunk_appendf(buf, "/%#lx", (long)size);
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004994 }
4995 return dli.dli_saddr;
4996 }
4997 else if (dli_main.dli_fbase != dli.dli_fbase) {
4998 /* unresolved symbol from a known library, report relative offset */
4999 chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_fbase));
5000 return NULL;
5001 }
Baruch Siache1651b22020-07-24 07:52:20 +03005002#endif /* __ELF__ && !__linux__ || USE_DL */
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005003 unknown:
5004 /* unresolved symbol from the main file, report relative offset to main */
5005 if ((void*)addr < (void*)main)
5006 chunk_appendf(buf, "main-%#lx", (long)((void*)main - addr));
5007 else
5008 chunk_appendf(buf, "main+%#lx", (long)(addr - (void*)main));
5009 return NULL;
Willy Tarreau0ebb5112016-12-05 00:10:57 +01005010}
5011
Willy Tarreau37857332021-12-28 09:57:10 +01005012/* On systems where this is supported, let's provide a possibility to enumerate
5013 * the list of object files. The output is appended to a buffer initialized by
5014 * the caller, with one name per line. A trailing zero is always emitted if data
5015 * are written. Only real objects are dumped (executable and .so libs). The
5016 * function returns non-zero if it dumps anything. These functions do not make
5017 * use of the trash so that it is possible for the caller to call them with the
5018 * trash on input. The output format may be platform-specific but at least one
5019 * version must emit raw object file names when argument is zero.
5020 */
5021#if defined(HA_HAVE_DUMP_LIBS)
5022# if defined(HA_HAVE_DL_ITERATE_PHDR)
5023/* the private <data> we pass below is a dump context initialized like this */
5024struct dl_dump_ctx {
5025 struct buffer *buf;
5026 int with_addr;
5027};
5028
5029static int dl_dump_libs_cb(struct dl_phdr_info *info, size_t size, void *data)
5030{
5031 struct dl_dump_ctx *ctx = data;
5032 const char *fname;
5033 size_t p1, p2, beg, end;
5034 int idx;
5035
5036 if (!info || !info->dlpi_name)
5037 goto leave;
5038
5039 if (!*info->dlpi_name)
5040 fname = get_exec_path();
5041 else if (strchr(info->dlpi_name, '/'))
5042 fname = info->dlpi_name;
5043 else
5044 /* else it's a VDSO or similar and we're not interested */
5045 goto leave;
5046
5047 if (!ctx->with_addr)
5048 goto dump_name;
5049
5050 /* virtual addresses are relative to the load address and are per
5051 * pseudo-header, so we have to scan them all to find the furthest
5052 * one from the beginning. In this case we only dump entries if
5053 * they have at least one section.
5054 */
5055 beg = ~0; end = 0;
5056 for (idx = 0; idx < info->dlpi_phnum; idx++) {
5057 if (!info->dlpi_phdr[idx].p_memsz)
5058 continue;
5059 p1 = info->dlpi_phdr[idx].p_vaddr;
5060 if (p1 < beg)
5061 beg = p1;
5062 p2 = p1 + info->dlpi_phdr[idx].p_memsz - 1;
5063 if (p2 > end)
5064 end = p2;
5065 }
5066
5067 if (!idx)
5068 goto leave;
5069
5070 chunk_appendf(ctx->buf, "0x%012llx-0x%012llx (0x%07llx) ",
5071 (ullong)info->dlpi_addr + beg,
5072 (ullong)info->dlpi_addr + end,
5073 (ullong)(end - beg + 1));
5074 dump_name:
5075 chunk_appendf(ctx->buf, "%s\n", fname);
5076 leave:
5077 return 0;
5078}
5079
5080/* dumps lib names and optionally address ranges */
5081int dump_libs(struct buffer *output, int with_addr)
5082{
5083 struct dl_dump_ctx ctx = { .buf = output, .with_addr = with_addr };
5084 size_t old_data = output->data;
5085
5086 dl_iterate_phdr(dl_dump_libs_cb, &ctx);
5087 return output->data != old_data;
5088}
5089# else // no DL_ITERATE_PHDR
5090# error "No dump_libs() function for this platform"
5091# endif
5092#else // no HA_HAVE_DUMP_LIBS
5093
5094/* unsupported platform: do not dump anything */
5095int dump_libs(struct buffer *output, int with_addr)
5096{
5097 return 0;
5098}
5099
5100#endif // HA_HAVE_DUMP_LIBS
5101
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005102/*
5103 * Allocate an array of unsigned int with <nums> as address from <str> string
Ilya Shipitsin46a030c2020-07-05 16:36:08 +05005104 * made of integer separated by dot characters.
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005105 *
5106 * First, initializes the value with <sz> as address to 0 and initializes the
5107 * array with <nums> as address to NULL. Then allocates the array with <nums> as
5108 * address updating <sz> pointed value to the size of this array.
5109 *
5110 * Returns 1 if succeeded, 0 if not.
5111 */
5112int parse_dotted_uints(const char *str, unsigned int **nums, size_t *sz)
5113{
5114 unsigned int *n;
5115 const char *s, *end;
5116
5117 s = str;
5118 *sz = 0;
5119 end = str + strlen(str);
5120 *nums = n = NULL;
5121
5122 while (1) {
5123 unsigned int r;
5124
5125 if (s >= end)
5126 break;
5127
5128 r = read_uint(&s, end);
5129 /* Expected characters after having read an uint: '\0' or '.',
5130 * if '.', must not be terminal.
5131 */
Christopher Faulet4b524122021-02-11 10:42:41 +01005132 if (*s != '\0'&& (*s++ != '.' || s == end)) {
5133 free(n);
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005134 return 0;
Christopher Faulet4b524122021-02-11 10:42:41 +01005135 }
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005136
Frédéric Lécaille12a71842019-02-26 18:19:48 +01005137 n = my_realloc2(n, (*sz + 1) * sizeof *n);
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005138 if (!n)
5139 return 0;
5140
5141 n[(*sz)++] = r;
5142 }
5143 *nums = n;
5144
5145 return 1;
5146}
5147
Willy Tarreau4d589e72019-08-23 19:02:26 +02005148
5149/* returns the number of bytes needed to encode <v> as a varint. An inline
5150 * version exists for use with constants (__varint_bytes()).
5151 */
5152int varint_bytes(uint64_t v)
5153{
5154 int len = 1;
5155
5156 if (v >= 240) {
5157 v = (v - 240) >> 4;
5158 while (1) {
5159 len++;
5160 if (v < 128)
5161 break;
5162 v = (v - 128) >> 7;
5163 }
5164 }
5165 return len;
5166}
5167
Willy Tarreau52bf8392020-03-08 00:42:37 +01005168
5169/* Random number generator state, see below */
Willy Tarreau1544c142020-03-12 00:31:18 +01005170static uint64_t ha_random_state[2] ALIGNED(2*sizeof(uint64_t));
Willy Tarreau52bf8392020-03-08 00:42:37 +01005171
5172/* This is a thread-safe implementation of xoroshiro128** described below:
5173 * http://prng.di.unimi.it/
5174 * It features a 2^128 long sequence, returns 64 high-quality bits on each call,
5175 * supports fast jumps and passes all common quality tests. It is thread-safe,
5176 * uses a double-cas on 64-bit architectures supporting it, and falls back to a
5177 * local lock on other ones.
5178 */
5179uint64_t ha_random64()
5180{
Willy Tarreau1544c142020-03-12 00:31:18 +01005181 uint64_t old[2] ALIGNED(2*sizeof(uint64_t));
5182 uint64_t new[2] ALIGNED(2*sizeof(uint64_t));
Willy Tarreau52bf8392020-03-08 00:42:37 +01005183
5184#if defined(USE_THREAD) && (!defined(HA_CAS_IS_8B) || !defined(HA_HAVE_CAS_DW))
5185 static HA_SPINLOCK_T rand_lock;
5186
5187 HA_SPIN_LOCK(OTHER_LOCK, &rand_lock);
5188#endif
5189
5190 old[0] = ha_random_state[0];
5191 old[1] = ha_random_state[1];
5192
5193#if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
5194 do {
5195#endif
Willy Tarreau52bf8392020-03-08 00:42:37 +01005196 new[1] = old[0] ^ old[1];
5197 new[0] = rotl64(old[0], 24) ^ new[1] ^ (new[1] << 16); // a, b
5198 new[1] = rotl64(new[1], 37); // c
5199
5200#if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
5201 } while (unlikely(!_HA_ATOMIC_DWCAS(ha_random_state, old, new)));
5202#else
5203 ha_random_state[0] = new[0];
5204 ha_random_state[1] = new[1];
5205#if defined(USE_THREAD)
5206 HA_SPIN_UNLOCK(OTHER_LOCK, &rand_lock);
5207#endif
5208#endif
Willy Tarreaub2475a12021-05-09 10:26:14 +02005209 return rotl64(old[0] * 5, 7) * 9;
Willy Tarreau52bf8392020-03-08 00:42:37 +01005210}
5211
5212/* seeds the random state using up to <len> bytes from <seed>, starting with
5213 * the first non-zero byte.
5214 */
5215void ha_random_seed(const unsigned char *seed, size_t len)
5216{
5217 size_t pos;
5218
5219 /* the seed must not be all zeroes, so we pre-fill it with alternating
5220 * bits and overwrite part of them with the block starting at the first
5221 * non-zero byte from the seed.
5222 */
5223 memset(ha_random_state, 0x55, sizeof(ha_random_state));
5224
5225 for (pos = 0; pos < len; pos++)
5226 if (seed[pos] != 0)
5227 break;
5228
5229 if (pos == len)
5230 return;
5231
5232 seed += pos;
5233 len -= pos;
5234
5235 if (len > sizeof(ha_random_state))
5236 len = sizeof(ha_random_state);
5237
5238 memcpy(ha_random_state, seed, len);
5239}
5240
5241/* This causes a jump to (dist * 2^96) places in the pseudo-random sequence,
5242 * and is equivalent to calling ha_random64() as many times. It is used to
5243 * provide non-overlapping sequences of 2^96 numbers (~7*10^28) to up to 2^32
5244 * different generators (i.e. different processes after a fork). The <dist>
5245 * argument is the distance to jump to and is used in a loop so it rather not
5246 * be too large if the processing time is a concern.
5247 *
5248 * BEWARE: this function is NOT thread-safe and must not be called during
5249 * concurrent accesses to ha_random64().
5250 */
5251void ha_random_jump96(uint32_t dist)
5252{
5253 while (dist--) {
5254 uint64_t s0 = 0;
5255 uint64_t s1 = 0;
5256 int b;
5257
5258 for (b = 0; b < 64; b++) {
5259 if ((0xd2a98b26625eee7bULL >> b) & 1) {
5260 s0 ^= ha_random_state[0];
5261 s1 ^= ha_random_state[1];
5262 }
5263 ha_random64();
5264 }
5265
5266 for (b = 0; b < 64; b++) {
5267 if ((0xdddf9b1090aa7ac1ULL >> b) & 1) {
5268 s0 ^= ha_random_state[0];
5269 s1 ^= ha_random_state[1];
5270 }
5271 ha_random64();
5272 }
5273 ha_random_state[0] = s0;
5274 ha_random_state[1] = s1;
5275 }
5276}
5277
Willy Tarreauee3bcdd2020-03-08 17:48:17 +01005278/* Generates an RFC4122 UUID into chunk <output> which must be at least 37
5279 * bytes large.
5280 */
5281void ha_generate_uuid(struct buffer *output)
5282{
5283 uint32_t rnd[4];
5284 uint64_t last;
5285
5286 last = ha_random64();
5287 rnd[0] = last;
5288 rnd[1] = last >> 32;
5289
5290 last = ha_random64();
5291 rnd[2] = last;
5292 rnd[3] = last >> 32;
5293
5294 chunk_printf(output, "%8.8x-%4.4x-%4.4x-%4.4x-%12.12llx",
5295 rnd[0],
5296 rnd[1] & 0xFFFF,
5297 ((rnd[1] >> 16u) & 0xFFF) | 0x4000, // highest 4 bits indicate the uuid version
5298 (rnd[2] & 0x3FFF) | 0x8000, // the highest 2 bits indicate the UUID variant (10),
5299 (long long)((rnd[2] >> 14u) | ((uint64_t) rnd[3] << 18u)) & 0xFFFFFFFFFFFFull);
5300}
5301
5302
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005303/* only used by parse_line() below. It supports writing in place provided that
5304 * <in> is updated to the next location before calling it. In that case, the
5305 * char at <in> may be overwritten.
5306 */
5307#define EMIT_CHAR(x) \
5308 do { \
5309 char __c = (char)(x); \
5310 if ((opts & PARSE_OPT_INPLACE) && out+outpos > in) \
5311 err |= PARSE_ERR_OVERLAP; \
5312 if (outpos >= outmax) \
5313 err |= PARSE_ERR_TOOLARGE; \
5314 if (!err) \
5315 out[outpos] = __c; \
5316 outpos++; \
5317 } while (0)
5318
Ilya Shipitsin46a030c2020-07-05 16:36:08 +05005319/* Parse <in>, copy it into <out> split into isolated words whose pointers
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005320 * are put in <args>. If more than <outlen> bytes have to be emitted, the
5321 * extraneous ones are not emitted but <outlen> is updated so that the caller
5322 * knows how much to realloc. Similarly, <args> are not updated beyond <nbargs>
5323 * but the returned <nbargs> indicates how many were found. All trailing args
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005324 * up to <nbargs> point to the trailing zero, and as long as <nbargs> is > 0,
5325 * it is guaranteed that at least one arg will point to the zero. It is safe
5326 * to call it with a NULL <args> if <nbargs> is 0.
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005327 *
5328 * <out> may overlap with <in> provided that it never goes further, in which
5329 * case the parser will accept to perform in-place parsing and unquoting/
5330 * unescaping but only if environment variables do not lead to expansion that
5331 * causes overlapping, otherwise the input string being destroyed, the error
5332 * will not be recoverable. Note that even during out-of-place <in> will
5333 * experience temporary modifications in-place for variable resolution and must
5334 * be writable, and will also receive zeroes to delimit words when using
5335 * in-place copy. Parsing options <opts> taken from PARSE_OPT_*. Return value
5336 * is zero on success otherwise a bitwise-or of PARSE_ERR_*. Upon error, the
5337 * starting point of the first invalid character sequence or unmatched
5338 * quote/brace is reported in <errptr> if not NULL. When using in-place parsing
5339 * error reporting might be difficult since zeroes will have been inserted into
5340 * the string. One solution for the caller may consist in replacing all args
5341 * delimiters with spaces in this case.
5342 */
5343uint32_t parse_line(char *in, char *out, size_t *outlen, char **args, int *nbargs, uint32_t opts, char **errptr)
5344{
5345 char *quote = NULL;
5346 char *brace = NULL;
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005347 char *word_expand = NULL;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005348 unsigned char hex1, hex2;
5349 size_t outmax = *outlen;
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005350 int argsmax = *nbargs - 1;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005351 size_t outpos = 0;
5352 int squote = 0;
5353 int dquote = 0;
5354 int arg = 0;
5355 uint32_t err = 0;
5356
5357 *nbargs = 0;
5358 *outlen = 0;
5359
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005360 /* argsmax may be -1 here, protecting args[] from any write */
5361 if (arg < argsmax)
5362 args[arg] = out;
5363
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005364 while (1) {
5365 if (*in >= '-' && *in != '\\') {
5366 /* speedup: directly send all regular chars starting
5367 * with '-', '.', '/', alnum etc...
5368 */
5369 EMIT_CHAR(*in++);
5370 continue;
5371 }
5372 else if (*in == '\0' || *in == '\n' || *in == '\r') {
5373 /* end of line */
5374 break;
5375 }
5376 else if (*in == '#' && (opts & PARSE_OPT_SHARP) && !squote && !dquote) {
5377 /* comment */
5378 break;
5379 }
5380 else if (*in == '"' && !squote && (opts & PARSE_OPT_DQUOTE)) { /* double quote outside single quotes */
5381 if (dquote) {
5382 dquote = 0;
5383 quote = NULL;
5384 }
5385 else {
5386 dquote = 1;
5387 quote = in;
5388 }
5389 in++;
5390 continue;
5391 }
5392 else if (*in == '\'' && !dquote && (opts & PARSE_OPT_SQUOTE)) { /* single quote outside double quotes */
5393 if (squote) {
5394 squote = 0;
5395 quote = NULL;
5396 }
5397 else {
5398 squote = 1;
5399 quote = in;
5400 }
5401 in++;
5402 continue;
5403 }
5404 else if (*in == '\\' && !squote && (opts & PARSE_OPT_BKSLASH)) {
5405 /* first, we'll replace \\, \<space>, \#, \r, \n, \t, \xXX with their
5406 * C equivalent value but only when they have a special meaning and within
5407 * double quotes for some of them. Other combinations left unchanged (eg: \1).
5408 */
5409 char tosend = *in;
5410
5411 switch (in[1]) {
5412 case ' ':
5413 case '\\':
5414 tosend = in[1];
5415 in++;
5416 break;
5417
5418 case 't':
5419 tosend = '\t';
5420 in++;
5421 break;
5422
5423 case 'n':
5424 tosend = '\n';
5425 in++;
5426 break;
5427
5428 case 'r':
5429 tosend = '\r';
5430 in++;
5431 break;
5432
5433 case '#':
5434 /* escaping of "#" only if comments are supported */
5435 if (opts & PARSE_OPT_SHARP)
5436 in++;
5437 tosend = *in;
5438 break;
5439
5440 case '\'':
5441 /* escaping of "'" only outside single quotes and only if single quotes are supported */
5442 if (opts & PARSE_OPT_SQUOTE && !squote)
5443 in++;
5444 tosend = *in;
5445 break;
5446
5447 case '"':
5448 /* escaping of '"' only outside single quotes and only if double quotes are supported */
5449 if (opts & PARSE_OPT_DQUOTE && !squote)
5450 in++;
5451 tosend = *in;
5452 break;
5453
5454 case '$':
5455 /* escaping of '$' only inside double quotes and only if env supported */
5456 if (opts & PARSE_OPT_ENV && dquote)
5457 in++;
5458 tosend = *in;
5459 break;
5460
5461 case 'x':
5462 if (!ishex(in[2]) || !ishex(in[3])) {
5463 /* invalid or incomplete hex sequence */
5464 err |= PARSE_ERR_HEX;
5465 if (errptr)
5466 *errptr = in;
5467 goto leave;
5468 }
Willy Tarreauf278eec2020-07-05 21:46:32 +02005469 hex1 = toupper((unsigned char)in[2]) - '0';
5470 hex2 = toupper((unsigned char)in[3]) - '0';
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005471 if (hex1 > 9) hex1 -= 'A' - '9' - 1;
5472 if (hex2 > 9) hex2 -= 'A' - '9' - 1;
5473 tosend = (hex1 << 4) + hex2;
5474 in += 3;
5475 break;
5476
5477 default:
5478 /* other combinations are not escape sequences */
5479 break;
5480 }
5481
5482 in++;
5483 EMIT_CHAR(tosend);
5484 }
5485 else if (isspace((unsigned char)*in) && !squote && !dquote) {
5486 /* a non-escaped space is an argument separator */
5487 while (isspace((unsigned char)*in))
5488 in++;
5489 EMIT_CHAR(0);
5490 arg++;
5491 if (arg < argsmax)
5492 args[arg] = out + outpos;
5493 else
5494 err |= PARSE_ERR_TOOMANY;
5495 }
5496 else if (*in == '$' && (opts & PARSE_OPT_ENV) && (dquote || !(opts & PARSE_OPT_DQUOTE))) {
5497 /* environment variables are evaluated anywhere, or only
5498 * inside double quotes if they are supported.
5499 */
5500 char *var_name;
5501 char save_char;
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005502 const char *value;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005503
5504 in++;
5505
5506 if (*in == '{')
5507 brace = in++;
5508
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005509 if (!isalpha((unsigned char)*in) && *in != '_' && *in != '.') {
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005510 /* unacceptable character in variable name */
5511 err |= PARSE_ERR_VARNAME;
5512 if (errptr)
5513 *errptr = in;
5514 goto leave;
5515 }
5516
5517 var_name = in;
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005518 if (*in == '.')
5519 in++;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005520 while (isalnum((unsigned char)*in) || *in == '_')
5521 in++;
5522
5523 save_char = *in;
5524 *in = '\0';
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005525 if (unlikely(*var_name == '.')) {
5526 /* internal pseudo-variables */
5527 if (strcmp(var_name, ".LINE") == 0)
5528 value = ultoa(global.cfg_curr_line);
5529 else if (strcmp(var_name, ".FILE") == 0)
5530 value = global.cfg_curr_file;
5531 else if (strcmp(var_name, ".SECTION") == 0)
5532 value = global.cfg_curr_section;
5533 else {
5534 /* unsupported internal variable name */
5535 err |= PARSE_ERR_VARNAME;
5536 if (errptr)
5537 *errptr = var_name;
5538 goto leave;
5539 }
5540 } else {
5541 value = getenv(var_name);
5542 }
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005543 *in = save_char;
5544
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005545 /* support for '[*]' sequence to force word expansion,
5546 * only available inside braces */
5547 if (*in == '[' && brace && (opts & PARSE_OPT_WORD_EXPAND)) {
5548 word_expand = in++;
5549
5550 if (*in++ != '*' || *in++ != ']') {
5551 err |= PARSE_ERR_WRONG_EXPAND;
5552 if (errptr)
5553 *errptr = word_expand;
5554 goto leave;
5555 }
5556 }
5557
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005558 if (brace) {
5559 if (*in != '}') {
5560 /* unmatched brace */
5561 err |= PARSE_ERR_BRACE;
5562 if (errptr)
5563 *errptr = brace;
5564 goto leave;
5565 }
5566 in++;
5567 brace = NULL;
5568 }
5569
5570 if (value) {
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005571 while (*value) {
5572 /* expand as individual parameters on a space character */
Willy Tarreaufe2cc412020-10-01 18:04:40 +02005573 if (word_expand && isspace((unsigned char)*value)) {
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005574 EMIT_CHAR(0);
5575 ++arg;
5576 if (arg < argsmax)
5577 args[arg] = out + outpos;
5578 else
5579 err |= PARSE_ERR_TOOMANY;
5580
5581 /* skip consecutive spaces */
Willy Tarreaufe2cc412020-10-01 18:04:40 +02005582 while (isspace((unsigned char)*++value))
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005583 ;
5584 } else {
5585 EMIT_CHAR(*value++);
5586 }
5587 }
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005588 }
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005589 word_expand = NULL;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005590 }
5591 else {
5592 /* any other regular char */
5593 EMIT_CHAR(*in++);
5594 }
5595 }
5596
5597 /* end of output string */
5598 EMIT_CHAR(0);
5599 arg++;
5600
5601 if (quote) {
5602 /* unmatched quote */
5603 err |= PARSE_ERR_QUOTE;
5604 if (errptr)
5605 *errptr = quote;
5606 goto leave;
5607 }
5608 leave:
5609 *nbargs = arg;
5610 *outlen = outpos;
5611
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005612 /* empty all trailing args by making them point to the trailing zero,
5613 * at least the last one in any case.
5614 */
5615 if (arg > argsmax)
5616 arg = argsmax;
5617
5618 while (arg >= 0 && arg <= argsmax)
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005619 args[arg++] = out + outpos - 1;
5620
5621 return err;
5622}
5623#undef EMIT_CHAR
5624
Willy Tarreauc54e5ad2020-06-25 09:15:40 +02005625/* This is used to sanitize an input line that's about to be used for error reporting.
5626 * It will adjust <line> to print approximately <width> chars around <pos>, trying to
5627 * preserve the beginning, with leading or trailing "..." when the line is truncated.
5628 * If non-printable chars are present in the output. It returns the new offset <pos>
5629 * in the modified line. Non-printable characters are replaced with '?'. <width> must
5630 * be at least 6 to support two "..." otherwise the result is undefined. The line
5631 * itself must have at least 7 chars allocated for the same reason.
5632 */
5633size_t sanitize_for_printing(char *line, size_t pos, size_t width)
5634{
5635 size_t shift = 0;
5636 char *out = line;
5637 char *in = line;
5638 char *end = line + width;
5639
5640 if (pos >= width) {
5641 /* if we have to shift, we'll be out of context, so let's
5642 * try to put <pos> at the center of width.
5643 */
5644 shift = pos - width / 2;
5645 in += shift + 3;
5646 end = out + width - 3;
5647 out[0] = out[1] = out[2] = '.';
5648 out += 3;
5649 }
5650
5651 while (out < end && *in) {
5652 if (isspace((unsigned char)*in))
5653 *out++ = ' ';
5654 else if (isprint((unsigned char)*in))
5655 *out++ = *in;
5656 else
5657 *out++ = '?';
5658 in++;
5659 }
5660
5661 if (end < line + width) {
5662 out[0] = out[1] = out[2] = '.';
5663 out += 3;
5664 }
5665
5666 *out++ = 0;
5667 return pos - shift;
5668}
Willy Tarreau06e69b52021-03-02 14:01:35 +01005669
Willy Tarreaue33c4b32021-03-12 18:59:31 +01005670/* Update array <fp> with the fingerprint of word <word> by counting the
Willy Tarreauba2c4452021-03-12 09:01:52 +01005671 * transitions between characters. <fp> is a 1024-entries array indexed as
5672 * 32*from+to. Positions for 'from' and 'to' are:
Willy Tarreau9294e882021-03-15 09:34:27 +01005673 * 1..26=letter, 27=digit, 28=other/begin/end.
5674 * Row "from=0" is used to mark the character's presence. Others unused.
Willy Tarreauba2c4452021-03-12 09:01:52 +01005675 */
Willy Tarreaue33c4b32021-03-12 18:59:31 +01005676void update_word_fingerprint(uint8_t *fp, const char *word)
Willy Tarreauba2c4452021-03-12 09:01:52 +01005677{
5678 const char *p;
5679 int from, to;
5680 int c;
5681
Willy Tarreauba2c4452021-03-12 09:01:52 +01005682 from = 28; // begin
5683 for (p = word; *p; p++) {
5684 c = tolower(*p);
5685 switch(c) {
Willy Tarreau9294e882021-03-15 09:34:27 +01005686 case 'a'...'z': to = c - 'a' + 1; break;
5687 case 'A'...'Z': to = tolower(c) - 'a' + 1; break;
5688 case '0'...'9': to = 27; break;
5689 default: to = 28; break;
Willy Tarreauba2c4452021-03-12 09:01:52 +01005690 }
Willy Tarreau9294e882021-03-15 09:34:27 +01005691 fp[to] = 1;
Willy Tarreauba2c4452021-03-12 09:01:52 +01005692 fp[32 * from + to]++;
5693 from = to;
5694 }
5695 to = 28; // end
5696 fp[32 * from + to]++;
5697}
5698
Willy Tarreaue33c4b32021-03-12 18:59:31 +01005699/* Initialize array <fp> with the fingerprint of word <word> by counting the
5700 * transitions between characters. <fp> is a 1024-entries array indexed as
5701 * 32*from+to. Positions for 'from' and 'to' are:
5702 * 0..25=letter, 26=digit, 27=other, 28=begin, 29=end, others unused.
5703 */
5704void make_word_fingerprint(uint8_t *fp, const char *word)
5705{
5706 memset(fp, 0, 1024);
5707 update_word_fingerprint(fp, word);
5708}
5709
Willy Tarreauba2c4452021-03-12 09:01:52 +01005710/* Return the distance between two word fingerprints created by function
5711 * make_word_fingerprint(). It's a positive integer calculated as the sum of
Willy Tarreau714c4c12021-03-15 09:44:53 +01005712 * the differences between each location.
Willy Tarreauba2c4452021-03-12 09:01:52 +01005713 */
5714int word_fingerprint_distance(const uint8_t *fp1, const uint8_t *fp2)
5715{
5716 int i, k, dist = 0;
5717
5718 for (i = 0; i < 1024; i++) {
5719 k = (int)fp1[i] - (int)fp2[i];
Willy Tarreau714c4c12021-03-15 09:44:53 +01005720 dist += abs(k);
Willy Tarreauba2c4452021-03-12 09:01:52 +01005721 }
5722 return dist;
5723}
5724
Willy Tarreau06e69b52021-03-02 14:01:35 +01005725static int init_tools_per_thread()
5726{
5727 /* Let's make each thread start from a different position */
5728 statistical_prng_state += tid * MAX_THREADS;
5729 if (!statistical_prng_state)
5730 statistical_prng_state++;
5731 return 1;
5732}
5733REGISTER_PER_THREAD_INIT(init_tools_per_thread);
Willy Tarreauc54e5ad2020-06-25 09:15:40 +02005734
Willy Tarreaubaaee002006-06-26 02:48:02 +02005735/*
5736 * Local variables:
5737 * c-indent-level: 8
5738 * c-basic-offset: 8
5739 * End:
5740 */