blob: 34d8d022e25e5b4cefa3c987e1c409f8c961ff9f [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.comc4e52322021-08-17 12:55:49 +010019#if defined(__FreeBSD__)
20#include <elf.h>
21#include <dlfcn.h>
22extern void *__elf_aux_vector;
23#endif
24
David Carlierbd2cced2021-08-17 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
David Carlier43a56852022-03-04 15:50:48 +000046#if defined(__linux__) && 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>
Christopher Faulet6b0a0fb2022-04-04 11:29:28 +020054#include <haproxy/applet.h>
Willy Tarreauc13ed532020-06-02 10:22:45 +020055#include <haproxy/chunk.h>
Christopher Faulet5e29b762022-04-04 08:58:34 +020056#include <haproxy/conn_stream.h>
57#include <haproxy/cs_utils.h>
Willy Tarreau7c18b542020-06-11 09:23:02 +020058#include <haproxy/dgram.h>
Willy Tarreauf268ee82020-06-04 17:05:57 +020059#include <haproxy/global.h>
Willy Tarreau86416052020-06-04 09:20:54 +020060#include <haproxy/hlua.h>
Willy Tarreau213e9902020-06-04 14:58:24 +020061#include <haproxy/listener.h>
Willy Tarreau7a00efb2020-06-02 17:02:59 +020062#include <haproxy/namespace.h>
Christopher Faulet9553de72021-02-26 09:12:50 +010063#include <haproxy/net_helper.h>
Willy Tarreau5fc93282020-09-16 18:25:03 +020064#include <haproxy/protocol.h>
Emeric Brunc9437992021-02-12 19:42:55 +010065#include <haproxy/resolvers.h>
Willy Tarreau586f71b2020-12-11 15:54:36 +010066#include <haproxy/sock.h>
Willy Tarreau209108d2020-06-04 20:30:20 +020067#include <haproxy/ssl_sock.h>
William Lallemand3aeb3f92021-08-21 23:59:56 +020068#include <haproxy/ssl_utils.h>
Willy Tarreaucea0e1b2020-06-04 17:25:40 +020069#include <haproxy/task.h>
Willy Tarreau48fbcae2020-06-03 18:09:46 +020070#include <haproxy/tools.h>
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +010071
Thierry Fournier93127942016-01-20 18:49:45 +010072/* This macro returns false if the test __x is false. Many
73 * of the following parsing function must be abort the processing
74 * if it returns 0, so this macro is useful for writing light code.
75 */
76#define RET0_UNLESS(__x) do { if (!(__x)) return 0; } while (0)
77
Willy Tarreau56adcf22012-12-23 18:00:29 +010078/* enough to store NB_ITOA_STR integers of :
Willy Tarreau72d759c2007-10-25 12:14:10 +020079 * 2^64-1 = 18446744073709551615 or
80 * -2^63 = -9223372036854775808
Willy Tarreaue7239b52009-03-29 13:41:58 +020081 *
82 * The HTML version needs room for adding the 25 characters
83 * '<span class="rls"></span>' around digits at positions 3N+1 in order
84 * to add spacing at up to 6 positions : 18 446 744 073 709 551 615
Willy Tarreau72d759c2007-10-25 12:14:10 +020085 */
Christopher Faulet99bca652017-11-14 16:47:26 +010086THREAD_LOCAL char itoa_str[NB_ITOA_STR][171];
87THREAD_LOCAL int itoa_idx = 0; /* index of next itoa_str to use */
Willy Tarreaubaaee002006-06-26 02:48:02 +020088
Willy Tarreau588297f2014-06-16 15:16:40 +020089/* sometimes we'll need to quote strings (eg: in stats), and we don't expect
90 * to quote strings larger than a max configuration line.
91 */
Christopher Faulet99bca652017-11-14 16:47:26 +010092THREAD_LOCAL char quoted_str[NB_QSTR][QSTR_SIZE + 1];
93THREAD_LOCAL int quoted_idx = 0;
Willy Tarreau588297f2014-06-16 15:16:40 +020094
Willy Tarreau06e69b52021-03-02 14:01:35 +010095/* thread-local PRNG state. It's modified to start from a different sequence
96 * on all threads upon startup. It must not be used or anything beyond getting
97 * statistical values as it's 100% predictable.
98 */
99THREAD_LOCAL unsigned int statistical_prng_state = 2463534242U;
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 Tarreaue3b45182021-10-27 17:28:55 +0200951 enum proto_type proto_type;
952 int ctrl_type;
Willy Tarreaud4448bc2013-02-20 15:55:15 +0100953
954 portl = porth = porta = 0;
Willy Tarreau72b8c1f2015-09-08 15:50:19 +0200955 if (fqdn)
956 *fqdn = NULL;
Willy Tarreaubaaee002006-06-26 02:48:02 +0200957
Willy Tarreaudad36a32013-03-11 01:20:04 +0100958 str2 = back = env_expand(strdup(str));
Willy Tarreaudf350f12013-03-01 20:22:54 +0100959 if (str2 == NULL) {
960 memprintf(err, "out of memory in '%s'\n", __FUNCTION__);
Willy Tarreaud5191e72010-02-09 20:50:45 +0100961 goto out;
Willy Tarreaudf350f12013-03-01 20:22:54 +0100962 }
Willy Tarreaubaaee002006-06-26 02:48:02 +0200963
Willy Tarreau9f69f462015-09-08 16:01:25 +0200964 if (!*str2) {
965 memprintf(err, "'%s' resolves to an empty address (environment variable missing?)\n", str);
966 goto out;
967 }
968
Willy Tarreau24709282013-03-10 21:32:12 +0100969 memset(&ss, 0, sizeof(ss));
970
Willy Tarreaue835bd82020-09-16 11:35:47 +0200971 /* prepare the default socket types */
Willy Tarreauf23b1bc2021-03-23 18:36:37 +0100972 if ((opts & (PA_O_STREAM|PA_O_DGRAM)) == PA_O_DGRAM ||
Willy Tarreaue3b45182021-10-27 17:28:55 +0200973 ((opts & (PA_O_STREAM|PA_O_DGRAM)) == (PA_O_DGRAM|PA_O_STREAM) && (opts & PA_O_DEFAULT_DGRAM))) {
974 proto_type = PROTO_TYPE_DGRAM;
975 ctrl_type = SOCK_DGRAM;
976 } else {
977 proto_type = PROTO_TYPE_STREAM;
978 ctrl_type = SOCK_STREAM;
979 }
Willy Tarreaue835bd82020-09-16 11:35:47 +0200980
981 if (strncmp(str2, "stream+", 7) == 0) {
982 str2 += 7;
Willy Tarreaue3b45182021-10-27 17:28:55 +0200983 proto_type = PROTO_TYPE_STREAM;
984 ctrl_type = SOCK_STREAM;
Willy Tarreaue835bd82020-09-16 11:35:47 +0200985 }
986 else if (strncmp(str2, "dgram+", 6) == 0) {
987 str2 += 6;
Willy Tarreaue3b45182021-10-27 17:28:55 +0200988 proto_type = PROTO_TYPE_DGRAM;
989 ctrl_type = SOCK_DGRAM;
Willy Tarreaue835bd82020-09-16 11:35:47 +0200990 }
991
Willy Tarreau24709282013-03-10 21:32:12 +0100992 if (strncmp(str2, "unix@", 5) == 0) {
993 str2 += 5;
Willy Tarreauccfccef2014-05-10 01:49:15 +0200994 abstract = 0;
Willy Tarreau24709282013-03-10 21:32:12 +0100995 ss.ss_family = AF_UNIX;
996 }
Emeric Brunce325c42021-04-02 17:05:09 +0200997 else if (strncmp(str2, "uxdg@", 5) == 0) {
998 str2 += 5;
999 abstract = 0;
1000 ss.ss_family = AF_UNIX;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001001 proto_type = PROTO_TYPE_DGRAM;
1002 ctrl_type = SOCK_DGRAM;
Emeric Brunce325c42021-04-02 17:05:09 +02001003 }
1004 else if (strncmp(str2, "uxst@", 5) == 0) {
1005 str2 += 5;
1006 abstract = 0;
1007 ss.ss_family = AF_UNIX;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001008 proto_type = PROTO_TYPE_STREAM;
1009 ctrl_type = SOCK_STREAM;
Emeric Brunce325c42021-04-02 17:05:09 +02001010 }
Willy Tarreauccfccef2014-05-10 01:49:15 +02001011 else if (strncmp(str2, "abns@", 5) == 0) {
1012 str2 += 5;
1013 abstract = 1;
1014 ss.ss_family = AF_UNIX;
1015 }
Emeric Brunce325c42021-04-02 17:05:09 +02001016 else if (strncmp(str2, "ip@", 3) == 0) {
1017 str2 += 3;
1018 ss.ss_family = AF_UNSPEC;
1019 }
Willy Tarreau24709282013-03-10 21:32:12 +01001020 else if (strncmp(str2, "ipv4@", 5) == 0) {
1021 str2 += 5;
1022 ss.ss_family = AF_INET;
1023 }
1024 else if (strncmp(str2, "ipv6@", 5) == 0) {
1025 str2 += 5;
1026 ss.ss_family = AF_INET6;
1027 }
Emeric Brunce325c42021-04-02 17:05:09 +02001028 else if (strncmp(str2, "tcp4@", 5) == 0) {
1029 str2 += 5;
1030 ss.ss_family = AF_INET;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001031 proto_type = PROTO_TYPE_STREAM;
1032 ctrl_type = SOCK_STREAM;
Emeric Brunce325c42021-04-02 17:05:09 +02001033 }
Emeric Brun3835c0d2020-07-07 09:46:09 +02001034 else if (strncmp(str2, "udp4@", 5) == 0) {
1035 str2 += 5;
1036 ss.ss_family = AF_INET;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001037 proto_type = PROTO_TYPE_DGRAM;
1038 ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +02001039 }
Emeric Brunce325c42021-04-02 17:05:09 +02001040 else if (strncmp(str2, "tcp6@", 5) == 0) {
1041 str2 += 5;
1042 ss.ss_family = AF_INET6;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001043 proto_type = PROTO_TYPE_STREAM;
1044 ctrl_type = SOCK_STREAM;
Emeric Brunce325c42021-04-02 17:05:09 +02001045 }
Emeric Brun3835c0d2020-07-07 09:46:09 +02001046 else if (strncmp(str2, "udp6@", 5) == 0) {
1047 str2 += 5;
1048 ss.ss_family = AF_INET6;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001049 proto_type = PROTO_TYPE_DGRAM;
1050 ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +02001051 }
Emeric Brunce325c42021-04-02 17:05:09 +02001052 else if (strncmp(str2, "tcp@", 4) == 0) {
1053 str2 += 4;
1054 ss.ss_family = AF_UNSPEC;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001055 proto_type = PROTO_TYPE_STREAM;
1056 ctrl_type = SOCK_STREAM;
Emeric Brunce325c42021-04-02 17:05:09 +02001057 }
Emeric Brun3835c0d2020-07-07 09:46:09 +02001058 else if (strncmp(str2, "udp@", 4) == 0) {
1059 str2 += 4;
1060 ss.ss_family = AF_UNSPEC;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001061 proto_type = PROTO_TYPE_DGRAM;
1062 ctrl_type = SOCK_DGRAM;
Emeric Brun3835c0d2020-07-07 09:46:09 +02001063 }
Frédéric Lécaille10caf652020-11-23 11:36:57 +01001064 else if (strncmp(str2, "quic4@", 6) == 0) {
1065 str2 += 6;
1066 ss.ss_family = AF_INET;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001067 proto_type = PROTO_TYPE_DGRAM;
Frédéric Lécaille10caf652020-11-23 11:36:57 +01001068 ctrl_type = SOCK_STREAM;
1069 }
1070 else if (strncmp(str2, "quic6@", 6) == 0) {
1071 str2 += 6;
1072 ss.ss_family = AF_INET6;
Willy Tarreaue3b45182021-10-27 17:28:55 +02001073 proto_type = PROTO_TYPE_DGRAM;
Frédéric Lécaille10caf652020-11-23 11:36:57 +01001074 ctrl_type = SOCK_STREAM;
1075 }
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001076 else if (strncmp(str2, "fd@", 3) == 0) {
1077 str2 += 3;
1078 ss.ss_family = AF_CUST_EXISTING_FD;
1079 }
1080 else if (strncmp(str2, "sockpair@", 9) == 0) {
1081 str2 += 9;
1082 ss.ss_family = AF_CUST_SOCKPAIR;
1083 }
Willy Tarreau24709282013-03-10 21:32:12 +01001084 else if (*str2 == '/') {
1085 ss.ss_family = AF_UNIX;
1086 }
1087 else
1088 ss.ss_family = AF_UNSPEC;
1089
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001090 if (ss.ss_family == AF_CUST_SOCKPAIR) {
Willy Tarreaua215be22020-09-16 10:14:16 +02001091 struct sockaddr_storage ss2;
1092 socklen_t addr_len;
William Lallemand2fe7dd02018-09-11 16:51:29 +02001093 char *endptr;
1094
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001095 new_fd = strtol(str2, &endptr, 10);
1096 if (!*str2 || new_fd < 0 || *endptr) {
William Lallemand2fe7dd02018-09-11 16:51:29 +02001097 memprintf(err, "file descriptor '%s' is not a valid integer in '%s'\n", str2, str);
1098 goto out;
1099 }
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001100
Willy Tarreaua215be22020-09-16 10:14:16 +02001101 /* just verify that it's a socket */
1102 addr_len = sizeof(ss2);
1103 if (getsockname(new_fd, (struct sockaddr *)&ss2, &addr_len) == -1) {
1104 memprintf(err, "cannot use file descriptor '%d' : %s.\n", new_fd, strerror(errno));
1105 goto out;
1106 }
1107
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001108 ((struct sockaddr_in *)&ss)->sin_addr.s_addr = new_fd;
1109 ((struct sockaddr_in *)&ss)->sin_port = 0;
William Lallemand2fe7dd02018-09-11 16:51:29 +02001110 }
Willy Tarreau5a7beed2020-09-04 16:54:05 +02001111 else if (ss.ss_family == AF_CUST_EXISTING_FD) {
Willy Tarreau40aa0702013-03-10 23:51:38 +01001112 char *endptr;
1113
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001114 new_fd = strtol(str2, &endptr, 10);
1115 if (!*str2 || new_fd < 0 || *endptr) {
Willy Tarreaudad36a32013-03-11 01:20:04 +01001116 memprintf(err, "file descriptor '%s' is not a valid integer in '%s'\n", str2, str);
Willy Tarreau40aa0702013-03-10 23:51:38 +01001117 goto out;
1118 }
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001119
Willy Tarreau6edc7222020-09-15 17:41:56 +02001120 if (opts & PA_O_SOCKET_FD) {
1121 socklen_t addr_len;
1122 int type;
1123
1124 addr_len = sizeof(ss);
1125 if (getsockname(new_fd, (struct sockaddr *)&ss, &addr_len) == -1) {
1126 memprintf(err, "cannot use file descriptor '%d' : %s.\n", new_fd, strerror(errno));
1127 goto out;
1128 }
1129
1130 addr_len = sizeof(type);
1131 if (getsockopt(new_fd, SOL_SOCKET, SO_TYPE, &type, &addr_len) != 0 ||
Willy Tarreaue3b45182021-10-27 17:28:55 +02001132 (type == SOCK_STREAM) != (proto_type == PROTO_TYPE_STREAM)) {
Willy Tarreau6edc7222020-09-15 17:41:56 +02001133 memprintf(err, "socket on file descriptor '%d' is of the wrong type.\n", new_fd);
1134 goto out;
1135 }
1136
1137 porta = portl = porth = get_host_port(&ss);
1138 } else if (opts & PA_O_RAW_FD) {
1139 ((struct sockaddr_in *)&ss)->sin_addr.s_addr = new_fd;
1140 ((struct sockaddr_in *)&ss)->sin_port = 0;
1141 } else {
1142 memprintf(err, "a file descriptor is not acceptable here in '%s'\n", str);
1143 goto out;
1144 }
Willy Tarreau40aa0702013-03-10 23:51:38 +01001145 }
1146 else if (ss.ss_family == AF_UNIX) {
Willy Tarreau8daa9202019-06-16 18:16:33 +02001147 struct sockaddr_un *un = (struct sockaddr_un *)&ss;
Willy Tarreau15586382013-03-04 19:48:14 +01001148 int prefix_path_len;
1149 int max_path_len;
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001150 int adr_len;
Willy Tarreau15586382013-03-04 19:48:14 +01001151
1152 /* complete unix socket path name during startup or soft-restart is
1153 * <unix_bind_prefix><path>.<pid>.<bak|tmp>
1154 */
Willy Tarreauccfccef2014-05-10 01:49:15 +02001155 prefix_path_len = (pfx && !abstract) ? strlen(pfx) : 0;
Willy Tarreau8daa9202019-06-16 18:16:33 +02001156 max_path_len = (sizeof(un->sun_path) - 1) -
Willy Tarreau327ea5a2020-02-11 06:43:37 +01001157 (abstract ? 0 : prefix_path_len + 1 + 5 + 1 + 3);
Willy Tarreau15586382013-03-04 19:48:14 +01001158
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001159 adr_len = strlen(str2);
1160 if (adr_len > max_path_len) {
Willy Tarreau15586382013-03-04 19:48:14 +01001161 memprintf(err, "socket path '%s' too long (max %d)\n", str, max_path_len);
1162 goto out;
1163 }
1164
Willy Tarreauccfccef2014-05-10 01:49:15 +02001165 /* when abstract==1, we skip the first zero and copy all bytes except the trailing zero */
Willy Tarreau8daa9202019-06-16 18:16:33 +02001166 memset(un->sun_path, 0, sizeof(un->sun_path));
Willy Tarreau94ef3f32014-04-14 14:49:00 +02001167 if (prefix_path_len)
Willy Tarreau8daa9202019-06-16 18:16:33 +02001168 memcpy(un->sun_path, pfx, prefix_path_len);
1169 memcpy(un->sun_path + prefix_path_len + abstract, str2, adr_len + 1 - abstract);
Willy Tarreau15586382013-03-04 19:48:14 +01001170 }
Willy Tarreau24709282013-03-10 21:32:12 +01001171 else { /* IPv4 and IPv6 */
mildisff5d5102015-10-26 18:50:08 +01001172 char *end = str2 + strlen(str2);
1173 char *chr;
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001174
mildisff5d5102015-10-26 18:50:08 +01001175 /* search for : or ] whatever comes first */
1176 for (chr = end-1; chr > str2; chr--) {
1177 if (*chr == ']' || *chr == ':')
1178 break;
1179 }
1180
1181 if (*chr == ':') {
1182 /* Found a colon before a closing-bracket, must be a port separator.
1183 * This guarantee backward compatibility.
1184 */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001185 if (!(opts & PA_O_PORT_OK)) {
1186 memprintf(err, "port specification not permitted here in '%s'", str);
1187 goto out;
1188 }
mildisff5d5102015-10-26 18:50:08 +01001189 *chr++ = '\0';
1190 port1 = chr;
1191 }
1192 else {
1193 /* Either no colon and no closing-bracket
1194 * or directly ending with a closing-bracket.
1195 * However, no port.
1196 */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001197 if (opts & PA_O_PORT_MAND) {
1198 memprintf(err, "missing port specification in '%s'", str);
1199 goto out;
1200 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001201 port1 = "";
mildisff5d5102015-10-26 18:50:08 +01001202 }
Willy Tarreaubaaee002006-06-26 02:48:02 +02001203
Willy Tarreau90807112020-02-25 08:16:33 +01001204 if (isdigit((unsigned char)*port1)) { /* single port or range */
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001205 port2 = strchr(port1, '-');
Willy Tarreau7f96a842020-09-15 11:12:44 +02001206 if (port2) {
1207 if (!(opts & PA_O_PORT_RANGE)) {
1208 memprintf(err, "port range not permitted here in '%s'", str);
1209 goto out;
1210 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001211 *port2++ = '\0';
Willy Tarreau7f96a842020-09-15 11:12:44 +02001212 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001213 else
1214 port2 = port1;
1215 portl = atoi(port1);
1216 porth = atoi(port2);
Willy Tarreau7f96a842020-09-15 11:12:44 +02001217
1218 if (portl < !!(opts & PA_O_PORT_MAND) || portl > 65535) {
1219 memprintf(err, "invalid port '%s'", port1);
1220 goto out;
1221 }
1222
1223 if (porth < !!(opts & PA_O_PORT_MAND) || porth > 65535) {
1224 memprintf(err, "invalid port '%s'", port2);
1225 goto out;
1226 }
1227
1228 if (portl > porth) {
1229 memprintf(err, "invalid port range '%d-%d'", portl, porth);
1230 goto out;
1231 }
1232
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001233 porta = portl;
1234 }
1235 else if (*port1 == '-') { /* negative offset */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001236 if (!(opts & PA_O_PORT_OFS)) {
1237 memprintf(err, "port offset not permitted here in '%s'", str);
1238 goto out;
1239 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001240 portl = atoi(port1 + 1);
1241 porta = -portl;
1242 }
1243 else if (*port1 == '+') { /* positive offset */
Willy Tarreau7f96a842020-09-15 11:12:44 +02001244 if (!(opts & PA_O_PORT_OFS)) {
1245 memprintf(err, "port offset not permitted here in '%s'", str);
1246 goto out;
1247 }
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001248 porth = atoi(port1 + 1);
1249 porta = porth;
1250 }
1251 else if (*port1) { /* other any unexpected char */
Willy Tarreaudad36a32013-03-11 01:20:04 +01001252 memprintf(err, "invalid character '%c' in port number '%s' in '%s'\n", *port1, port1, str);
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001253 goto out;
1254 }
Willy Tarreau7f96a842020-09-15 11:12:44 +02001255 else if (opts & PA_O_PORT_MAND) {
1256 memprintf(err, "missing port specification in '%s'", str);
1257 goto out;
1258 }
Willy Tarreauceccdd72016-11-02 22:27:10 +01001259
1260 /* first try to parse the IP without resolving. If it fails, it
1261 * tells us we need to keep a copy of the FQDN to resolve later
1262 * and to enable DNS. In this case we can proceed if <fqdn> is
Willy Tarreaucd3a55912020-09-04 15:30:46 +02001263 * set or if PA_O_RESOLVE is set, otherwise it's an error.
Willy Tarreauceccdd72016-11-02 22:27:10 +01001264 */
1265 if (str2ip2(str2, &ss, 0) == NULL) {
Willy Tarreaucd3a55912020-09-04 15:30:46 +02001266 if ((!(opts & PA_O_RESOLVE) && !fqdn) ||
1267 ((opts & PA_O_RESOLVE) && str2ip2(str2, &ss, 1) == NULL)) {
Willy Tarreauceccdd72016-11-02 22:27:10 +01001268 memprintf(err, "invalid address: '%s' in '%s'\n", str2, str);
1269 goto out;
1270 }
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001271
Willy Tarreauceccdd72016-11-02 22:27:10 +01001272 if (fqdn) {
1273 if (str2 != back)
1274 memmove(back, str2, strlen(str2) + 1);
1275 *fqdn = back;
1276 back = NULL;
1277 }
Willy Tarreau72b8c1f2015-09-08 15:50:19 +02001278 }
Willy Tarreauceccdd72016-11-02 22:27:10 +01001279 set_host_port(&ss, porta);
Willy Tarreaue4c58c82013-03-06 15:28:17 +01001280 }
Willy Tarreaufab5a432011-03-04 15:31:53 +01001281
Willy Tarreaue835bd82020-09-16 11:35:47 +02001282 if (ctrl_type == SOCK_STREAM && !(opts & PA_O_STREAM)) {
1283 memprintf(err, "stream-type socket not acceptable in '%s'\n", str);
1284 goto out;
1285 }
1286 else if (ctrl_type == SOCK_DGRAM && !(opts & PA_O_DGRAM)) {
1287 memprintf(err, "dgram-type socket not acceptable in '%s'\n", str);
1288 goto out;
1289 }
1290
Willy Tarreau65ec4e32020-09-16 19:17:08 +02001291 if (proto || (opts & PA_O_CONNECT)) {
Willy Tarreau5fc93282020-09-16 18:25:03 +02001292 /* Note: if the caller asks for a proto, we must find one,
Emeric Brun26754902021-04-07 14:26:44 +02001293 * except if we inherit from a raw FD (family == AF_CUST_EXISTING_FD)
1294 * orif we return with an fqdn that will resolve later,
Willy Tarreau5fc93282020-09-16 18:25:03 +02001295 * in which case the address is not known yet (this is only
1296 * for servers actually).
1297 */
Willy Tarreaub2ffc992020-09-16 21:37:31 +02001298 new_proto = protocol_lookup(ss.ss_family,
Willy Tarreaue3b45182021-10-27 17:28:55 +02001299 proto_type,
Willy Tarreauaf9609b2020-09-16 22:04:46 +02001300 ctrl_type == SOCK_DGRAM);
Willy Tarreaub2ffc992020-09-16 21:37:31 +02001301
Emeric Brun26754902021-04-07 14:26:44 +02001302 if (!new_proto && (!fqdn || !*fqdn) && (ss.ss_family != AF_CUST_EXISTING_FD)) {
Willy Tarreau5fc93282020-09-16 18:25:03 +02001303 memprintf(err, "unsupported protocol family %d for address '%s'", ss.ss_family, str);
1304 goto out;
1305 }
Willy Tarreau65ec4e32020-09-16 19:17:08 +02001306
1307 if ((opts & PA_O_CONNECT) && new_proto && !new_proto->connect) {
1308 memprintf(err, "connect() not supported for this protocol family %d used by address '%s'", ss.ss_family, str);
1309 goto out;
1310 }
Willy Tarreau5fc93282020-09-16 18:25:03 +02001311 }
1312
Willy Tarreauc120c8d2013-03-10 19:27:44 +01001313 ret = &ss;
Willy Tarreaud5191e72010-02-09 20:50:45 +01001314 out:
Willy Tarreau48ef4c92017-01-06 18:32:38 +01001315 if (port)
1316 *port = porta;
Willy Tarreaud4448bc2013-02-20 15:55:15 +01001317 if (low)
1318 *low = portl;
1319 if (high)
1320 *high = porth;
Willy Tarreaua93e5c72020-09-15 14:01:16 +02001321 if (fd)
1322 *fd = new_fd;
Willy Tarreau5fc93282020-09-16 18:25:03 +02001323 if (proto)
1324 *proto = new_proto;
Willy Tarreau24709282013-03-10 21:32:12 +01001325 free(back);
Willy Tarreaud5191e72010-02-09 20:50:45 +01001326 return ret;
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001327}
1328
Thayne McCombs92149f92020-11-20 01:28:26 -07001329/* converts <addr> and <port> into a string representation of the address and port. This is sort
1330 * of an inverse of str2sa_range, with some restrictions. The supported families are AF_INET,
1331 * AF_INET6, AF_UNIX, and AF_CUST_SOCKPAIR. If the family is unsopported NULL is returned.
1332 * If map_ports is true, then the sign of the port is included in the output, to indicate it is
1333 * relative to the incoming port. AF_INET and AF_INET6 will be in the form "<addr>:<port>".
1334 * AF_UNIX will either be just the path (if using a pathname) or "abns@<path>" if it is abstract.
1335 * AF_CUST_SOCKPAIR will be of the form "sockpair@<fd>".
1336 *
1337 * The returned char* is allocated, and it is the responsibility of the caller to free it.
1338 */
1339char * sa2str(const struct sockaddr_storage *addr, int port, int map_ports)
1340{
1341 char buffer[INET6_ADDRSTRLEN];
1342 char *out = NULL;
1343 const void *ptr;
1344 const char *path;
1345
1346 switch (addr->ss_family) {
1347 case AF_INET:
1348 ptr = &((struct sockaddr_in *)addr)->sin_addr;
1349 break;
1350 case AF_INET6:
1351 ptr = &((struct sockaddr_in6 *)addr)->sin6_addr;
1352 break;
1353 case AF_UNIX:
1354 path = ((struct sockaddr_un *)addr)->sun_path;
1355 if (path[0] == '\0') {
1356 const int max_length = sizeof(struct sockaddr_un) - offsetof(struct sockaddr_un, sun_path) - 1;
1357 return memprintf(&out, "abns@%.*s", max_length, path+1);
1358 } else {
1359 return strdup(path);
1360 }
1361 case AF_CUST_SOCKPAIR:
1362 return memprintf(&out, "sockpair@%d", ((struct sockaddr_in *)addr)->sin_addr.s_addr);
1363 default:
1364 return NULL;
1365 }
1366 inet_ntop(addr->ss_family, ptr, buffer, get_addr_len(addr));
1367 if (map_ports)
1368 return memprintf(&out, "%s:%+d", buffer, port);
1369 else
1370 return memprintf(&out, "%s:%d", buffer, port);
1371}
1372
1373
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001374/* converts <str> to a struct in_addr containing a network mask. It can be
1375 * passed in dotted form (255.255.255.0) or in CIDR form (24). It returns 1
Jarno Huuskonen577d5ac2017-05-21 17:32:21 +03001376 * if the conversion succeeds otherwise zero.
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001377 */
1378int str2mask(const char *str, struct in_addr *mask)
1379{
1380 if (strchr(str, '.') != NULL) { /* dotted notation */
1381 if (!inet_pton(AF_INET, str, mask))
1382 return 0;
1383 }
1384 else { /* mask length */
1385 char *err;
1386 unsigned long len = strtol(str, &err, 10);
1387
1388 if (!*str || (err && *err) || (unsigned)len > 32)
1389 return 0;
Tim Duesterhus8575f722018-01-25 16:24:48 +01001390
1391 len2mask4(len, mask);
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001392 }
1393 return 1;
1394}
1395
Tim Duesterhus47185172018-01-25 16:24:49 +01001396/* converts <str> to a struct in6_addr containing a network mask. It can be
Tim Duesterhus5e642862018-02-20 17:02:18 +01001397 * passed in quadruplet form (ffff:ffff::) or in CIDR form (64). It returns 1
Tim Duesterhus47185172018-01-25 16:24:49 +01001398 * if the conversion succeeds otherwise zero.
1399 */
1400int str2mask6(const char *str, struct in6_addr *mask)
1401{
1402 if (strchr(str, ':') != NULL) { /* quadruplet notation */
1403 if (!inet_pton(AF_INET6, str, mask))
1404 return 0;
1405 }
1406 else { /* mask length */
1407 char *err;
1408 unsigned long len = strtol(str, &err, 10);
1409
1410 if (!*str || (err && *err) || (unsigned)len > 128)
1411 return 0;
1412
1413 len2mask6(len, mask);
1414 }
1415 return 1;
1416}
1417
Thierry FOURNIERb0504632013-12-14 15:39:02 +01001418/* convert <cidr> to struct in_addr <mask>. It returns 1 if the conversion
1419 * succeeds otherwise zero.
1420 */
1421int cidr2dotted(int cidr, struct in_addr *mask) {
1422
1423 if (cidr < 0 || cidr > 32)
1424 return 0;
1425
1426 mask->s_addr = cidr ? htonl(~0UL << (32 - cidr)) : 0;
1427 return 1;
1428}
1429
Thierry Fournier70473a52016-02-17 17:12:14 +01001430/* Convert mask from bit length form to in_addr form.
1431 * This function never fails.
1432 */
1433void len2mask4(int len, struct in_addr *addr)
1434{
1435 if (len >= 32) {
1436 addr->s_addr = 0xffffffff;
1437 return;
1438 }
1439 if (len <= 0) {
1440 addr->s_addr = 0x00000000;
1441 return;
1442 }
1443 addr->s_addr = 0xffffffff << (32 - len);
1444 addr->s_addr = htonl(addr->s_addr);
1445}
1446
1447/* Convert mask from bit length form to in6_addr form.
1448 * This function never fails.
1449 */
1450void len2mask6(int len, struct in6_addr *addr)
1451{
1452 len2mask4(len, (struct in_addr *)&addr->s6_addr[0]); /* msb */
1453 len -= 32;
1454 len2mask4(len, (struct in_addr *)&addr->s6_addr[4]);
1455 len -= 32;
1456 len2mask4(len, (struct in_addr *)&addr->s6_addr[8]);
1457 len -= 32;
1458 len2mask4(len, (struct in_addr *)&addr->s6_addr[12]); /* lsb */
1459}
1460
Willy Tarreauc6f4ce82009-06-10 11:09:37 +02001461/*
Willy Tarreaud077a8e2007-05-08 18:28:09 +02001462 * converts <str> to two struct in_addr* which must be pre-allocated.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001463 * The format is "addr[/mask]", where "addr" cannot be empty, and mask
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001464 * is optional and either in the dotted or CIDR notation.
Willy Tarreaubaaee002006-06-26 02:48:02 +02001465 * Note: "addr" can also be a hostname. Returns 1 if OK, 0 if error.
1466 */
Thierry FOURNIERfc7ac7b2014-02-11 15:23:04 +01001467int str2net(const char *str, int resolve, struct in_addr *addr, struct in_addr *mask)
Willy Tarreaubaaee002006-06-26 02:48:02 +02001468{
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001469 __label__ out_free, out_err;
1470 char *c, *s;
1471 int ret_val;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001472
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001473 s = strdup(str);
1474 if (!s)
1475 return 0;
1476
Willy Tarreaubaaee002006-06-26 02:48:02 +02001477 memset(mask, 0, sizeof(*mask));
1478 memset(addr, 0, sizeof(*addr));
Willy Tarreaubaaee002006-06-26 02:48:02 +02001479
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001480 if ((c = strrchr(s, '/')) != NULL) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02001481 *c++ = '\0';
1482 /* c points to the mask */
Willy Tarreau2937c0d2010-01-26 17:36:17 +01001483 if (!str2mask(c, mask))
1484 goto out_err;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001485 }
1486 else {
Willy Tarreauebd61602006-12-30 11:54:15 +01001487 mask->s_addr = ~0U;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001488 }
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001489 if (!inet_pton(AF_INET, s, addr)) {
Willy Tarreaubaaee002006-06-26 02:48:02 +02001490 struct hostent *he;
1491
Thierry FOURNIERfc7ac7b2014-02-11 15:23:04 +01001492 if (!resolve)
1493 goto out_err;
1494
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001495 if ((he = gethostbyname(s)) == NULL) {
1496 goto out_err;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001497 }
1498 else
1499 *addr = *(struct in_addr *) *(he->h_addr_list);
1500 }
Willy Tarreau8aeae4a2007-06-17 11:42:08 +02001501
1502 ret_val = 1;
1503 out_free:
1504 free(s);
1505 return ret_val;
1506 out_err:
1507 ret_val = 0;
1508 goto out_free;
Willy Tarreaubaaee002006-06-26 02:48:02 +02001509}
1510
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001511
1512/*
Willy Tarreau6d20e282012-04-27 22:49:47 +02001513 * converts <str> to two struct in6_addr* which must be pre-allocated.
1514 * The format is "addr[/mask]", where "addr" cannot be empty, and mask
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001515 * is an optional number of bits (128 being the default).
Willy Tarreau6d20e282012-04-27 22:49:47 +02001516 * Returns 1 if OK, 0 if error.
1517 */
1518int str62net(const char *str, struct in6_addr *addr, unsigned char *mask)
1519{
1520 char *c, *s;
1521 int ret_val = 0;
1522 char *err;
1523 unsigned long len = 128;
1524
1525 s = strdup(str);
1526 if (!s)
1527 return 0;
1528
1529 memset(mask, 0, sizeof(*mask));
1530 memset(addr, 0, sizeof(*addr));
1531
1532 if ((c = strrchr(s, '/')) != NULL) {
1533 *c++ = '\0'; /* c points to the mask */
1534 if (!*c)
1535 goto out_free;
1536
1537 len = strtoul(c, &err, 10);
1538 if ((err && *err) || (unsigned)len > 128)
1539 goto out_free;
1540 }
1541 *mask = len; /* OK we have a valid mask in <len> */
1542
1543 if (!inet_pton(AF_INET6, s, addr))
1544 goto out_free;
1545
1546 ret_val = 1;
1547 out_free:
1548 free(s);
1549 return ret_val;
1550}
1551
1552
1553/*
Willy Tarreau12e10272021-03-25 11:34:40 +01001554 * Parse IPv4 address found in url. Return the number of bytes parsed. It
1555 * expects exactly 4 numbers between 0 and 255 delimited by dots, and returns
1556 * zero in case of mismatch.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001557 */
David du Colombier6f5ccb12011-03-10 22:26:24 +01001558int url2ipv4(const char *addr, struct in_addr *dst)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001559{
1560 int saw_digit, octets, ch;
1561 u_char tmp[4], *tp;
1562 const char *cp = addr;
1563
1564 saw_digit = 0;
1565 octets = 0;
1566 *(tp = tmp) = 0;
1567
1568 while (*addr) {
Willy Tarreau12e10272021-03-25 11:34:40 +01001569 unsigned char digit = (ch = *addr) - '0';
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001570 if (digit > 9 && ch != '.')
1571 break;
Willy Tarreau12e10272021-03-25 11:34:40 +01001572 addr++;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001573 if (digit <= 9) {
1574 u_int new = *tp * 10 + digit;
1575 if (new > 255)
1576 return 0;
1577 *tp = new;
1578 if (!saw_digit) {
1579 if (++octets > 4)
1580 return 0;
1581 saw_digit = 1;
1582 }
1583 } else if (ch == '.' && saw_digit) {
1584 if (octets == 4)
1585 return 0;
1586 *++tp = 0;
1587 saw_digit = 0;
1588 } else
1589 return 0;
1590 }
1591
1592 if (octets < 4)
1593 return 0;
1594
1595 memcpy(&dst->s_addr, tmp, 4);
Willy Tarreau12e10272021-03-25 11:34:40 +01001596 return addr - cp;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001597}
1598
1599/*
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001600 * Resolve destination server from URL. Convert <str> to a sockaddr_storage.
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05001601 * <out> contain the code of the detected scheme, the start and length of
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001602 * the hostname. Actually only http and https are supported. <out> can be NULL.
1603 * This function returns the consumed length. It is useful if you parse complete
1604 * url like http://host:port/path, because the consumed length corresponds to
1605 * the first character of the path. If the conversion fails, it returns -1.
1606 *
1607 * This function tries to resolve the DNS name if haproxy is in starting mode.
1608 * So, this function may be used during the configuration parsing.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001609 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001610int url2sa(const char *url, int ulen, struct sockaddr_storage *addr, struct split_url *out)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001611{
1612 const char *curr = url, *cp = url;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001613 const char *end;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001614 int ret, url_code = 0;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001615 unsigned long long int http_code = 0;
1616 int default_port;
1617 struct hostent *he;
1618 char *p;
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001619
1620 /* Firstly, try to find :// pattern */
1621 while (curr < url+ulen && url_code != 0x3a2f2f) {
1622 url_code = ((url_code & 0xffff) << 8);
1623 url_code += (unsigned char)*curr++;
1624 }
1625
1626 /* Secondly, if :// pattern is found, verify parsed stuff
1627 * before pattern is matching our http pattern.
1628 * If so parse ip address and port in uri.
1629 *
1630 * WARNING: Current code doesn't support dynamic async dns resolver.
1631 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001632 if (url_code != 0x3a2f2f)
1633 return -1;
1634
1635 /* Copy scheme, and utrn to lower case. */
1636 while (cp < curr - 3)
1637 http_code = (http_code << 8) + *cp++;
1638 http_code |= 0x2020202020202020ULL; /* Turn everything to lower case */
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001639
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001640 /* HTTP or HTTPS url matching */
1641 if (http_code == 0x2020202068747470ULL) {
1642 default_port = 80;
1643 if (out)
1644 out->scheme = SCH_HTTP;
1645 }
1646 else if (http_code == 0x2020206874747073ULL) {
1647 default_port = 443;
1648 if (out)
1649 out->scheme = SCH_HTTPS;
1650 }
1651 else
1652 return -1;
1653
1654 /* If the next char is '[', the host address is IPv6. */
1655 if (*curr == '[') {
1656 curr++;
1657
1658 /* Check trash size */
1659 if (trash.size < ulen)
1660 return -1;
1661
1662 /* Look for ']' and copy the address in a trash buffer. */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001663 p = trash.area;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001664 for (end = curr;
1665 end < url + ulen && *end != ']';
1666 end++, p++)
1667 *p = *end;
1668 if (*end != ']')
1669 return -1;
1670 *p = '\0';
1671
1672 /* Update out. */
1673 if (out) {
1674 out->host = curr;
1675 out->host_len = end - curr;
1676 }
1677
1678 /* Try IPv6 decoding. */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001679 if (!inet_pton(AF_INET6, trash.area, &((struct sockaddr_in6 *)addr)->sin6_addr))
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001680 return -1;
1681 end++;
1682
1683 /* Decode port. */
William Lallemand3d7a9182022-03-25 17:37:51 +01001684 if (end < url + ulen && *end == ':') {
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001685 end++;
1686 default_port = read_uint(&end, url + ulen);
1687 }
1688 ((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1689 ((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1690 return end - url;
1691 }
1692 else {
William Lallemand8a913742022-02-18 16:13:12 +01001693 /* we need to copy the string into the trash because url2ipv4
1694 * needs a \0 at the end of the string */
1695 if (trash.size < ulen)
1696 return -1;
1697
1698 memcpy(trash.area, curr, ulen - (curr - url));
1699 trash.area[ulen - (curr - url)] = '\0';
1700
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001701 /* We are looking for IP address. If you want to parse and
1702 * resolve hostname found in url, you can use str2sa_range(), but
1703 * be warned this can slow down global daemon performances
1704 * while handling lagging dns responses.
1705 */
William Lallemand8a913742022-02-18 16:13:12 +01001706 ret = url2ipv4(trash.area, &((struct sockaddr_in *)addr)->sin_addr);
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001707 if (ret) {
1708 /* Update out. */
1709 if (out) {
1710 out->host = curr;
1711 out->host_len = ret;
1712 }
1713
William Lallemandb938b772022-03-24 21:59:03 +01001714 curr += ret;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001715
1716 /* Decode port. */
William Lallemand3d7a9182022-03-25 17:37:51 +01001717 if (curr < url + ulen && *curr == ':') {
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001718 curr++;
1719 default_port = read_uint(&curr, url + ulen);
1720 }
1721 ((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1722
1723 /* Set family. */
1724 ((struct sockaddr_in *)addr)->sin_family = AF_INET;
1725 return curr - url;
1726 }
1727 else if (global.mode & MODE_STARTING) {
1728 /* The IPv4 and IPv6 decoding fails, maybe the url contain name. Try to execute
1729 * synchronous DNS request only if HAProxy is in the start state.
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001730 */
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001731
1732 /* look for : or / or end */
1733 for (end = curr;
1734 end < url + ulen && *end != '/' && *end != ':';
1735 end++);
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001736 memcpy(trash.area, curr, end - curr);
1737 trash.area[end - curr] = '\0';
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001738
1739 /* try to resolve an IPv4/IPv6 hostname */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001740 he = gethostbyname(trash.area);
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001741 if (!he)
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001742 return -1;
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001743
1744 /* Update out. */
1745 if (out) {
1746 out->host = curr;
1747 out->host_len = end - curr;
1748 }
1749
1750 /* Decode port. */
William Lallemand3d7a9182022-03-25 17:37:51 +01001751 if (end < url + ulen && *end == ':') {
Thierry FOURNIER9f95e402014-03-21 14:51:46 +01001752 end++;
1753 default_port = read_uint(&end, url + ulen);
1754 }
1755
1756 /* Copy IP address, set port and family. */
1757 switch (he->h_addrtype) {
1758 case AF_INET:
1759 ((struct sockaddr_in *)addr)->sin_addr = *(struct in_addr *) *(he->h_addr_list);
1760 ((struct sockaddr_in *)addr)->sin_port = htons(default_port);
1761 ((struct sockaddr_in *)addr)->sin_family = AF_INET;
1762 return end - url;
1763
1764 case AF_INET6:
1765 ((struct sockaddr_in6 *)addr)->sin6_addr = *(struct in6_addr *) *(he->h_addr_list);
1766 ((struct sockaddr_in6 *)addr)->sin6_port = htons(default_port);
1767 ((struct sockaddr_in6 *)addr)->sin6_family = AF_INET6;
1768 return end - url;
1769 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001770 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001771 }
Alexandre Cassen5eb1a902007-11-29 15:43:32 +01001772 return -1;
1773}
1774
Willy Tarreau631f01c2011-09-05 00:36:48 +02001775/* Tries to convert a sockaddr_storage address to text form. Upon success, the
1776 * address family is returned so that it's easy for the caller to adapt to the
1777 * output format. Zero is returned if the address family is not supported. -1
1778 * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1779 * supported.
1780 */
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001781int addr_to_str(const struct sockaddr_storage *addr, char *str, int size)
Willy Tarreau631f01c2011-09-05 00:36:48 +02001782{
1783
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001784 const void *ptr;
Willy Tarreau631f01c2011-09-05 00:36:48 +02001785
1786 if (size < 5)
1787 return 0;
1788 *str = '\0';
1789
1790 switch (addr->ss_family) {
1791 case AF_INET:
1792 ptr = &((struct sockaddr_in *)addr)->sin_addr;
1793 break;
1794 case AF_INET6:
1795 ptr = &((struct sockaddr_in6 *)addr)->sin6_addr;
1796 break;
1797 case AF_UNIX:
1798 memcpy(str, "unix", 5);
1799 return addr->ss_family;
1800 default:
1801 return 0;
1802 }
1803
1804 if (inet_ntop(addr->ss_family, ptr, str, size))
1805 return addr->ss_family;
1806
1807 /* failed */
1808 return -1;
1809}
1810
Simon Horman75ab8bd2014-06-16 09:39:41 +09001811/* Tries to convert a sockaddr_storage port to text form. Upon success, the
1812 * address family is returned so that it's easy for the caller to adapt to the
1813 * output format. Zero is returned if the address family is not supported. -1
1814 * is returned upon error, with errno set. AF_INET, AF_INET6 and AF_UNIX are
1815 * supported.
1816 */
Willy Tarreaud5ec4bf2019-04-25 17:48:16 +02001817int port_to_str(const struct sockaddr_storage *addr, char *str, int size)
Simon Horman75ab8bd2014-06-16 09:39:41 +09001818{
1819
1820 uint16_t port;
1821
1822
Willy Tarreaud7dad1b2017-01-06 16:46:22 +01001823 if (size < 6)
Simon Horman75ab8bd2014-06-16 09:39:41 +09001824 return 0;
1825 *str = '\0';
1826
1827 switch (addr->ss_family) {
1828 case AF_INET:
1829 port = ((struct sockaddr_in *)addr)->sin_port;
1830 break;
1831 case AF_INET6:
1832 port = ((struct sockaddr_in6 *)addr)->sin6_port;
1833 break;
1834 case AF_UNIX:
1835 memcpy(str, "unix", 5);
1836 return addr->ss_family;
1837 default:
1838 return 0;
1839 }
1840
1841 snprintf(str, size, "%u", ntohs(port));
1842 return addr->ss_family;
1843}
1844
Willy Tarreau16e01562016-08-09 16:46:18 +02001845/* check if the given address is local to the system or not. It will return
1846 * -1 when it's not possible to know, 0 when the address is not local, 1 when
1847 * it is. We don't want to iterate over all interfaces for this (and it is not
1848 * portable). So instead we try to bind in UDP to this address on a free non
1849 * privileged port and to connect to the same address, port 0 (connect doesn't
1850 * care). If it succeeds, we own the address. Note that non-inet addresses are
1851 * considered local since they're most likely AF_UNIX.
1852 */
1853int addr_is_local(const struct netns_entry *ns,
1854 const struct sockaddr_storage *orig)
1855{
1856 struct sockaddr_storage addr;
1857 int result;
1858 int fd;
1859
1860 if (!is_inet_addr(orig))
1861 return 1;
1862
1863 memcpy(&addr, orig, sizeof(addr));
1864 set_host_port(&addr, 0);
1865
1866 fd = my_socketat(ns, addr.ss_family, SOCK_DGRAM, IPPROTO_UDP);
1867 if (fd < 0)
1868 return -1;
1869
1870 result = -1;
1871 if (bind(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == 0) {
1872 if (connect(fd, (struct sockaddr *)&addr, get_addr_len(&addr)) == -1)
1873 result = 0; // fail, non-local address
1874 else
1875 result = 1; // success, local address
1876 }
1877 else {
1878 if (errno == EADDRNOTAVAIL)
1879 result = 0; // definitely not local :-)
1880 }
1881 close(fd);
1882
1883 return result;
1884}
1885
Willy Tarreaubaaee002006-06-26 02:48:02 +02001886/* will try to encode the string <string> replacing all characters tagged in
1887 * <map> with the hexadecimal representation of their ASCII-code (2 digits)
1888 * prefixed by <escape>, and will store the result between <start> (included)
1889 * and <stop> (excluded), and will always terminate the string with a '\0'
1890 * before <stop>. The position of the '\0' is returned if the conversion
1891 * completes. If bytes are missing between <start> and <stop>, then the
1892 * conversion will be incomplete and truncated. If <stop> <= <start>, the '\0'
1893 * cannot even be stored so we return <start> without writing the 0.
1894 * The input string must also be zero-terminated.
1895 */
1896const char hextab[16] = "0123456789ABCDEF";
1897char *encode_string(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001898 const char escape, const long *map,
Willy Tarreaubaaee002006-06-26 02:48:02 +02001899 const char *string)
1900{
1901 if (start < stop) {
1902 stop--; /* reserve one byte for the final '\0' */
1903 while (start < stop && *string != '\0') {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001904 if (!ha_bit_test((unsigned char)(*string), map))
Willy Tarreaubaaee002006-06-26 02:48:02 +02001905 *start++ = *string;
1906 else {
1907 if (start + 3 >= stop)
1908 break;
1909 *start++ = escape;
1910 *start++ = hextab[(*string >> 4) & 15];
1911 *start++ = hextab[*string & 15];
1912 }
1913 string++;
1914 }
1915 *start = '\0';
1916 }
1917 return start;
1918}
1919
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001920/*
1921 * Same behavior as encode_string() above, except that it encodes chunk
1922 * <chunk> instead of a string.
1923 */
1924char *encode_chunk(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001925 const char escape, const long *map,
Willy Tarreau83061a82018-07-13 11:56:34 +02001926 const struct buffer *chunk)
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001927{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001928 char *str = chunk->area;
1929 char *end = chunk->area + chunk->data;
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001930
1931 if (start < stop) {
1932 stop--; /* reserve one byte for the final '\0' */
1933 while (start < stop && str < end) {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001934 if (!ha_bit_test((unsigned char)(*str), map))
Thierry FOURNIERe059ec92014-03-17 12:01:13 +01001935 *start++ = *str;
1936 else {
1937 if (start + 3 >= stop)
1938 break;
1939 *start++ = escape;
1940 *start++ = hextab[(*str >> 4) & 15];
1941 *start++ = hextab[*str & 15];
1942 }
1943 str++;
1944 }
1945 *start = '\0';
1946 }
1947 return start;
1948}
1949
Dragan Dosen0edd1092016-02-12 13:23:02 +01001950/*
1951 * Tries to prefix characters tagged in the <map> with the <escape>
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001952 * character. The input <string> must be zero-terminated. The result will
1953 * be stored between <start> (included) and <stop> (excluded). This
1954 * function will always try to terminate the resulting string with a '\0'
1955 * before <stop>, and will return its position if the conversion
1956 * completes.
1957 */
1958char *escape_string(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001959 const char escape, const long *map,
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001960 const char *string)
1961{
1962 if (start < stop) {
1963 stop--; /* reserve one byte for the final '\0' */
1964 while (start < stop && *string != '\0') {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001965 if (!ha_bit_test((unsigned char)(*string), map))
Dragan Dosen1a5d0602016-07-22 16:00:31 +02001966 *start++ = *string;
1967 else {
1968 if (start + 2 >= stop)
1969 break;
1970 *start++ = escape;
1971 *start++ = *string;
1972 }
1973 string++;
1974 }
1975 *start = '\0';
1976 }
1977 return start;
1978}
1979
1980/*
1981 * Tries to prefix characters tagged in the <map> with the <escape>
Dragan Dosen0edd1092016-02-12 13:23:02 +01001982 * character. <chunk> contains the input to be escaped. The result will be
1983 * stored between <start> (included) and <stop> (excluded). The function
1984 * will always try to terminate the resulting string with a '\0' before
1985 * <stop>, and will return its position if the conversion completes.
1986 */
1987char *escape_chunk(char *start, char *stop,
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001988 const char escape, const long *map,
Willy Tarreau83061a82018-07-13 11:56:34 +02001989 const struct buffer *chunk)
Dragan Dosen0edd1092016-02-12 13:23:02 +01001990{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001991 char *str = chunk->area;
1992 char *end = chunk->area + chunk->data;
Dragan Dosen0edd1092016-02-12 13:23:02 +01001993
1994 if (start < stop) {
1995 stop--; /* reserve one byte for the final '\0' */
1996 while (start < stop && str < end) {
Willy Tarreau1bfd6022019-06-07 11:10:07 +02001997 if (!ha_bit_test((unsigned char)(*str), map))
Dragan Dosen0edd1092016-02-12 13:23:02 +01001998 *start++ = *str;
1999 else {
2000 if (start + 2 >= stop)
2001 break;
2002 *start++ = escape;
2003 *start++ = *str;
2004 }
2005 str++;
2006 }
2007 *start = '\0';
2008 }
2009 return start;
2010}
2011
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002012/* Check a string for using it in a CSV output format. If the string contains
2013 * one of the following four char <">, <,>, CR or LF, the string is
2014 * encapsulated between <"> and the <"> are escaped by a <""> sequence.
2015 * <str> is the input string to be escaped. The function assumes that
2016 * the input string is null-terminated.
2017 *
2018 * If <quote> is 0, the result is returned escaped but without double quote.
Willy Tarreau898529b2016-01-06 18:07:04 +01002019 * It is useful if the escaped string is used between double quotes in the
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002020 * format.
2021 *
Willy Tarreau898529b2016-01-06 18:07:04 +01002022 * printf("..., \"%s\", ...\r\n", csv_enc(str, 0, &trash));
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002023 *
Willy Tarreaub631c292016-01-08 10:04:08 +01002024 * If <quote> is 1, the converter puts the quotes only if any reserved character
2025 * is present. If <quote> is 2, the converter always puts the quotes.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002026 *
Willy Tarreau83061a82018-07-13 11:56:34 +02002027 * <output> is a struct buffer used for storing the output string.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002028 *
Willy Tarreau898529b2016-01-06 18:07:04 +01002029 * The function returns the converted string on its output. If an error
2030 * occurs, the function returns an empty string. This type of output is useful
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002031 * for using the function directly as printf() argument.
2032 *
2033 * If the output buffer is too short to contain the input string, the result
2034 * is truncated.
Willy Tarreau898529b2016-01-06 18:07:04 +01002035 *
Willy Tarreaub631c292016-01-08 10:04:08 +01002036 * This function appends the encoding to the existing output chunk, and it
2037 * guarantees that it starts immediately at the first available character of
2038 * the chunk. Please use csv_enc() instead if you want to replace the output
2039 * chunk.
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002040 */
Willy Tarreau83061a82018-07-13 11:56:34 +02002041const char *csv_enc_append(const char *str, int quote, struct buffer *output)
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002042{
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002043 char *end = output->area + output->size;
2044 char *out = output->area + output->data;
Willy Tarreau898529b2016-01-06 18:07:04 +01002045 char *ptr = out;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002046
Willy Tarreaub631c292016-01-08 10:04:08 +01002047 if (quote == 1) {
2048 /* automatic quoting: first verify if we'll have to quote the string */
2049 if (!strpbrk(str, "\n\r,\""))
2050 quote = 0;
2051 }
2052
2053 if (quote)
2054 *ptr++ = '"';
2055
Willy Tarreau898529b2016-01-06 18:07:04 +01002056 while (*str && ptr < end - 2) { /* -2 for reserving space for <"> and \0. */
2057 *ptr = *str;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002058 if (*str == '"') {
Willy Tarreau898529b2016-01-06 18:07:04 +01002059 ptr++;
2060 if (ptr >= end - 2) {
2061 ptr--;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002062 break;
2063 }
Willy Tarreau898529b2016-01-06 18:07:04 +01002064 *ptr = '"';
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002065 }
Willy Tarreau898529b2016-01-06 18:07:04 +01002066 ptr++;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002067 str++;
2068 }
2069
Willy Tarreaub631c292016-01-08 10:04:08 +01002070 if (quote)
2071 *ptr++ = '"';
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002072
Willy Tarreau898529b2016-01-06 18:07:04 +01002073 *ptr = '\0';
Willy Tarreau843b7cb2018-07-13 10:54:26 +02002074 output->data = ptr - output->area;
Willy Tarreau898529b2016-01-06 18:07:04 +01002075 return out;
Thierry FOURNIERddea6262015-05-28 16:00:28 +02002076}
2077
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002078/* Decode an URL-encoded string in-place. The resulting string might
2079 * be shorter. If some forbidden characters are found, the conversion is
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002080 * aborted, the string is truncated before the issue and a negative value is
2081 * returned, otherwise the operation returns the length of the decoded string.
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002082 * If the 'in_form' argument is non-nul the string is assumed to be part of
2083 * an "application/x-www-form-urlencoded" encoded string, and the '+' will be
2084 * turned to a space. If it's zero, this will only be done after a question
2085 * mark ('?').
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002086 */
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002087int url_decode(char *string, int in_form)
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002088{
2089 char *in, *out;
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002090 int ret = -1;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002091
2092 in = string;
2093 out = string;
2094 while (*in) {
2095 switch (*in) {
2096 case '+' :
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002097 *out++ = in_form ? ' ' : *in;
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002098 break;
2099 case '%' :
2100 if (!ishex(in[1]) || !ishex(in[2]))
2101 goto end;
2102 *out++ = (hex2i(in[1]) << 4) + hex2i(in[2]);
2103 in += 2;
2104 break;
Willy Tarreau62ba9ba2020-04-23 17:54:47 +02002105 case '?':
2106 in_form = 1;
2107 /* fall through */
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002108 default:
2109 *out++ = *in;
2110 break;
2111 }
2112 in++;
2113 }
Thierry FOURNIER5068d962013-10-04 16:27:27 +02002114 ret = out - string; /* success */
Willy Tarreaubf9c2fc2011-05-31 18:06:18 +02002115 end:
2116 *out = 0;
2117 return ret;
2118}
Willy Tarreaubaaee002006-06-26 02:48:02 +02002119
Willy Tarreau6911fa42007-03-04 18:06:08 +01002120unsigned int str2ui(const char *s)
2121{
2122 return __str2ui(s);
2123}
2124
2125unsigned int str2uic(const char *s)
2126{
2127 return __str2uic(s);
2128}
2129
2130unsigned int strl2ui(const char *s, int len)
2131{
2132 return __strl2ui(s, len);
2133}
2134
2135unsigned int strl2uic(const char *s, int len)
2136{
2137 return __strl2uic(s, len);
2138}
2139
Willy Tarreau4ec83cd2010-10-15 23:19:55 +02002140unsigned int read_uint(const char **s, const char *end)
2141{
2142 return __read_uint(s, end);
2143}
2144
Thierry FOURNIER763a5d82015-07-06 23:09:52 +02002145/* This function reads an unsigned integer from the string pointed to by <s> and
2146 * returns it. The <s> pointer is adjusted to point to the first unread char. The
2147 * function automatically stops at <end>. If the number overflows, the 2^64-1
2148 * value is returned.
2149 */
2150unsigned long long int read_uint64(const char **s, const char *end)
2151{
2152 const char *ptr = *s;
2153 unsigned long long int i = 0, tmp;
2154 unsigned int j;
2155
2156 while (ptr < end) {
2157
2158 /* read next char */
2159 j = *ptr - '0';
2160 if (j > 9)
2161 goto read_uint64_end;
2162
2163 /* add char to the number and check overflow. */
2164 tmp = i * 10;
2165 if (tmp / 10 != i) {
2166 i = ULLONG_MAX;
2167 goto read_uint64_eat;
2168 }
2169 if (ULLONG_MAX - tmp < j) {
2170 i = ULLONG_MAX;
2171 goto read_uint64_eat;
2172 }
2173 i = tmp + j;
2174 ptr++;
2175 }
2176read_uint64_eat:
2177 /* eat each numeric char */
2178 while (ptr < end) {
2179 if ((unsigned int)(*ptr - '0') > 9)
2180 break;
2181 ptr++;
2182 }
2183read_uint64_end:
2184 *s = ptr;
2185 return i;
2186}
2187
2188/* This function reads an integer from the string pointed to by <s> and returns
2189 * it. The <s> pointer is adjusted to point to the first unread char. The function
2190 * automatically stops at <end>. Il the number is bigger than 2^63-2, the 2^63-1
2191 * value is returned. If the number is lowest than -2^63-1, the -2^63 value is
2192 * returned.
2193 */
2194long long int read_int64(const char **s, const char *end)
2195{
2196 unsigned long long int i = 0;
2197 int neg = 0;
2198
2199 /* Look for minus char. */
2200 if (**s == '-') {
2201 neg = 1;
2202 (*s)++;
2203 }
2204 else if (**s == '+')
2205 (*s)++;
2206
2207 /* convert as positive number. */
2208 i = read_uint64(s, end);
2209
2210 if (neg) {
2211 if (i > 0x8000000000000000ULL)
2212 return LLONG_MIN;
2213 return -i;
2214 }
2215 if (i > 0x7fffffffffffffffULL)
2216 return LLONG_MAX;
2217 return i;
2218}
2219
Willy Tarreau6911fa42007-03-04 18:06:08 +01002220/* This one is 7 times faster than strtol() on athlon with checks.
2221 * It returns the value of the number composed of all valid digits read,
2222 * and can process negative numbers too.
2223 */
2224int strl2ic(const char *s, int len)
2225{
2226 int i = 0;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002227 int j, k;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002228
2229 if (len > 0) {
2230 if (*s != '-') {
2231 /* positive number */
2232 while (len-- > 0) {
2233 j = (*s++) - '0';
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002234 k = i * 10;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002235 if (j > 9)
2236 break;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002237 i = k + j;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002238 }
2239 } else {
2240 /* negative number */
2241 s++;
2242 while (--len > 0) {
2243 j = (*s++) - '0';
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002244 k = i * 10;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002245 if (j > 9)
2246 break;
Willy Tarreau3f0c9762007-10-25 09:42:24 +02002247 i = k - j;
Willy Tarreau6911fa42007-03-04 18:06:08 +01002248 }
2249 }
2250 }
2251 return i;
2252}
2253
2254
2255/* This function reads exactly <len> chars from <s> and converts them to a
2256 * signed integer which it stores into <ret>. It accurately detects any error
2257 * (truncated string, invalid chars, overflows). It is meant to be used in
2258 * applications designed for hostile environments. It returns zero when the
2259 * number has successfully been converted, non-zero otherwise. When an error
2260 * is returned, the <ret> value is left untouched. It is yet 5 to 40 times
2261 * faster than strtol().
2262 */
2263int strl2irc(const char *s, int len, int *ret)
2264{
2265 int i = 0;
2266 int j;
2267
2268 if (!len)
2269 return 1;
2270
2271 if (*s != '-') {
2272 /* positive number */
2273 while (len-- > 0) {
2274 j = (*s++) - '0';
2275 if (j > 9) return 1; /* invalid char */
2276 if (i > INT_MAX / 10) return 1; /* check for multiply overflow */
2277 i = i * 10;
2278 if (i + j < i) return 1; /* check for addition overflow */
2279 i = i + j;
2280 }
2281 } else {
2282 /* negative number */
2283 s++;
2284 while (--len > 0) {
2285 j = (*s++) - '0';
2286 if (j > 9) return 1; /* invalid char */
2287 if (i < INT_MIN / 10) return 1; /* check for multiply overflow */
2288 i = i * 10;
2289 if (i - j > i) return 1; /* check for subtract overflow */
2290 i = i - j;
2291 }
2292 }
2293 *ret = i;
2294 return 0;
2295}
2296
2297
2298/* This function reads exactly <len> chars from <s> and converts them to a
2299 * signed integer which it stores into <ret>. It accurately detects any error
2300 * (truncated string, invalid chars, overflows). It is meant to be used in
2301 * applications designed for hostile environments. It returns zero when the
2302 * number has successfully been converted, non-zero otherwise. When an error
2303 * is returned, the <ret> value is left untouched. It is about 3 times slower
William Dauchy060ffc82021-02-06 20:47:51 +01002304 * than strl2irc().
Willy Tarreau6911fa42007-03-04 18:06:08 +01002305 */
Willy Tarreau6911fa42007-03-04 18:06:08 +01002306
2307int strl2llrc(const char *s, int len, long long *ret)
2308{
2309 long long i = 0;
2310 int j;
2311
2312 if (!len)
2313 return 1;
2314
2315 if (*s != '-') {
2316 /* positive number */
2317 while (len-- > 0) {
2318 j = (*s++) - '0';
2319 if (j > 9) return 1; /* invalid char */
2320 if (i > LLONG_MAX / 10LL) return 1; /* check for multiply overflow */
2321 i = i * 10LL;
2322 if (i + j < i) return 1; /* check for addition overflow */
2323 i = i + j;
2324 }
2325 } else {
2326 /* negative number */
2327 s++;
2328 while (--len > 0) {
2329 j = (*s++) - '0';
2330 if (j > 9) return 1; /* invalid char */
2331 if (i < LLONG_MIN / 10LL) return 1; /* check for multiply overflow */
2332 i = i * 10LL;
2333 if (i - j > i) return 1; /* check for subtract overflow */
2334 i = i - j;
2335 }
2336 }
2337 *ret = i;
2338 return 0;
2339}
2340
Thierry FOURNIER511e9472014-01-23 17:40:34 +01002341/* This function is used with pat_parse_dotted_ver(). It converts a string
2342 * composed by two number separated by a dot. Each part must contain in 16 bits
2343 * because internally they will be represented as a 32-bit quantity stored in
2344 * a 64-bit integer. It returns zero when the number has successfully been
2345 * converted, non-zero otherwise. When an error is returned, the <ret> value
2346 * is left untouched.
2347 *
2348 * "1.3" -> 0x0000000000010003
2349 * "65535.65535" -> 0x00000000ffffffff
2350 */
2351int strl2llrc_dotted(const char *text, int len, long long *ret)
2352{
2353 const char *end = &text[len];
2354 const char *p;
2355 long long major, minor;
2356
2357 /* Look for dot. */
2358 for (p = text; p < end; p++)
2359 if (*p == '.')
2360 break;
2361
2362 /* Convert major. */
2363 if (strl2llrc(text, p - text, &major) != 0)
2364 return 1;
2365
2366 /* Check major. */
2367 if (major >= 65536)
2368 return 1;
2369
2370 /* Convert minor. */
2371 minor = 0;
2372 if (p < end)
2373 if (strl2llrc(p + 1, end - (p + 1), &minor) != 0)
2374 return 1;
2375
2376 /* Check minor. */
2377 if (minor >= 65536)
2378 return 1;
2379
2380 /* Compose value. */
2381 *ret = (major << 16) | (minor & 0xffff);
2382 return 0;
2383}
2384
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002385/* This function parses a time value optionally followed by a unit suffix among
2386 * "d", "h", "m", "s", "ms" or "us". It converts the value into the unit
2387 * expected by the caller. The computation does its best to avoid overflows.
2388 * The value is returned in <ret> if everything is fine, and a NULL is returned
2389 * by the function. In case of error, a pointer to the error is returned and
2390 * <ret> is left untouched. Values are automatically rounded up when needed.
Willy Tarreau9faebe32019-06-07 19:00:37 +02002391 * Values resulting in values larger than or equal to 2^31 after conversion are
2392 * reported as an overflow as value PARSE_TIME_OVER. Non-null values resulting
2393 * in an underflow are reported as an underflow as value PARSE_TIME_UNDER.
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002394 */
2395const char *parse_time_err(const char *text, unsigned *ret, unsigned unit_flags)
2396{
Willy Tarreau9faebe32019-06-07 19:00:37 +02002397 unsigned long long imult, idiv;
2398 unsigned long long omult, odiv;
2399 unsigned long long value, result;
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002400 const char *str = text;
2401
2402 if (!isdigit((unsigned char)*text))
2403 return text;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002404
2405 omult = odiv = 1;
2406
2407 switch (unit_flags & TIME_UNIT_MASK) {
2408 case TIME_UNIT_US: omult = 1000000; break;
2409 case TIME_UNIT_MS: omult = 1000; break;
2410 case TIME_UNIT_S: break;
2411 case TIME_UNIT_MIN: odiv = 60; break;
2412 case TIME_UNIT_HOUR: odiv = 3600; break;
2413 case TIME_UNIT_DAY: odiv = 86400; break;
2414 default: break;
2415 }
2416
2417 value = 0;
2418
2419 while (1) {
2420 unsigned int j;
2421
2422 j = *text - '0';
2423 if (j > 9)
2424 break;
2425 text++;
2426 value *= 10;
2427 value += j;
2428 }
2429
2430 imult = idiv = 1;
2431 switch (*text) {
2432 case '\0': /* no unit = default unit */
2433 imult = omult = idiv = odiv = 1;
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002434 goto end;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002435 case 's': /* second = unscaled unit */
2436 break;
2437 case 'u': /* microsecond : "us" */
2438 if (text[1] == 's') {
2439 idiv = 1000000;
2440 text++;
Thayne McCombsa6838052021-04-02 14:12:43 -06002441 break;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002442 }
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002443 return text;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002444 case 'm': /* millisecond : "ms" or minute: "m" */
2445 if (text[1] == 's') {
2446 idiv = 1000;
2447 text++;
2448 } else
2449 imult = 60;
2450 break;
2451 case 'h': /* hour : "h" */
2452 imult = 3600;
2453 break;
2454 case 'd': /* day : "d" */
2455 imult = 86400;
2456 break;
2457 default:
2458 return text;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002459 }
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002460 if (*(++text) != '\0') {
2461 ha_warning("unexpected character '%c' after the timer value '%s', only "
2462 "(us=microseconds,ms=milliseconds,s=seconds,m=minutes,h=hours,d=days) are supported."
2463 " This will be reported as an error in next versions.\n", *text, str);
2464 }
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002465
Christopher Fauletc20ad0d2020-12-11 09:23:07 +01002466 end:
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002467 if (omult % idiv == 0) { omult /= idiv; idiv = 1; }
2468 if (idiv % omult == 0) { idiv /= omult; omult = 1; }
2469 if (imult % odiv == 0) { imult /= odiv; odiv = 1; }
2470 if (odiv % imult == 0) { odiv /= imult; imult = 1; }
2471
Willy Tarreau9faebe32019-06-07 19:00:37 +02002472 result = (value * (imult * omult) + (idiv * odiv - 1)) / (idiv * odiv);
2473 if (result >= 0x80000000)
2474 return PARSE_TIME_OVER;
2475 if (!result && value)
2476 return PARSE_TIME_UNDER;
2477 *ret = result;
Willy Tarreaua0d37b62007-12-02 22:00:35 +01002478 return NULL;
2479}
Willy Tarreau6911fa42007-03-04 18:06:08 +01002480
Emeric Brun39132b22010-01-04 14:57:24 +01002481/* this function converts the string starting at <text> to an unsigned int
2482 * stored in <ret>. If an error is detected, the pointer to the unexpected
Joseph Herlant32b83272018-11-15 11:58:28 -08002483 * character is returned. If the conversion is successful, NULL is returned.
Emeric Brun39132b22010-01-04 14:57:24 +01002484 */
2485const char *parse_size_err(const char *text, unsigned *ret) {
2486 unsigned value = 0;
2487
Christopher Faulet82635a02020-12-11 09:30:45 +01002488 if (!isdigit((unsigned char)*text))
2489 return text;
2490
Emeric Brun39132b22010-01-04 14:57:24 +01002491 while (1) {
2492 unsigned int j;
2493
2494 j = *text - '0';
2495 if (j > 9)
2496 break;
2497 if (value > ~0U / 10)
2498 return text;
2499 value *= 10;
2500 if (value > (value + j))
2501 return text;
2502 value += j;
2503 text++;
2504 }
2505
2506 switch (*text) {
2507 case '\0':
2508 break;
2509 case 'K':
2510 case 'k':
2511 if (value > ~0U >> 10)
2512 return text;
2513 value = value << 10;
2514 break;
2515 case 'M':
2516 case 'm':
2517 if (value > ~0U >> 20)
2518 return text;
2519 value = value << 20;
2520 break;
2521 case 'G':
2522 case 'g':
2523 if (value > ~0U >> 30)
2524 return text;
2525 value = value << 30;
2526 break;
2527 default:
2528 return text;
2529 }
2530
Godbach58048a22015-01-28 17:36:16 +08002531 if (*text != '\0' && *++text != '\0')
2532 return text;
2533
Emeric Brun39132b22010-01-04 14:57:24 +01002534 *ret = value;
2535 return NULL;
2536}
2537
Willy Tarreau126d4062013-12-03 17:50:47 +01002538/*
2539 * Parse binary string written in hexadecimal (source) and store the decoded
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002540 * result into binstr and set binstrlen to the length of binstr. Memory for
Willy Tarreau126d4062013-12-03 17:50:47 +01002541 * binstr is allocated by the function. In case of error, returns 0 with an
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002542 * error message in err. In success case, it returns the consumed length.
Willy Tarreau126d4062013-12-03 17:50:47 +01002543 */
2544int parse_binary(const char *source, char **binstr, int *binstrlen, char **err)
2545{
2546 int len;
2547 const char *p = source;
2548 int i,j;
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002549 int alloc;
Willy Tarreau126d4062013-12-03 17:50:47 +01002550
2551 len = strlen(source);
2552 if (len % 2) {
2553 memprintf(err, "an even number of hex digit is expected");
2554 return 0;
2555 }
2556
2557 len = len >> 1;
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002558
Willy Tarreau126d4062013-12-03 17:50:47 +01002559 if (!*binstr) {
Tim Duesterhuse52b6e52020-09-12 20:26:43 +02002560 *binstr = calloc(len, sizeof(**binstr));
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002561 if (!*binstr) {
2562 memprintf(err, "out of memory while loading string pattern");
2563 return 0;
2564 }
2565 alloc = 1;
Willy Tarreau126d4062013-12-03 17:50:47 +01002566 }
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002567 else {
2568 if (*binstrlen < len) {
Joseph Herlant76dbe782018-11-15 12:01:22 -08002569 memprintf(err, "no space available in the buffer. expect %d, provides %d",
Thierry FOURNIER9645d422013-12-06 19:59:28 +01002570 len, *binstrlen);
2571 return 0;
2572 }
2573 alloc = 0;
2574 }
2575 *binstrlen = len;
Willy Tarreau126d4062013-12-03 17:50:47 +01002576
2577 i = j = 0;
2578 while (j < len) {
2579 if (!ishex(p[i++]))
2580 goto bad_input;
2581 if (!ishex(p[i++]))
2582 goto bad_input;
2583 (*binstr)[j++] = (hex2i(p[i-2]) << 4) + hex2i(p[i-1]);
2584 }
Thierry FOURNIERee330af2014-01-21 11:36:14 +01002585 return len << 1;
Willy Tarreau126d4062013-12-03 17:50:47 +01002586
2587bad_input:
2588 memprintf(err, "an hex digit is expected (found '%c')", p[i-1]);
Willy Tarreau61cfdf42021-02-20 10:46:51 +01002589 if (alloc)
2590 ha_free(binstr);
Willy Tarreau126d4062013-12-03 17:50:47 +01002591 return 0;
2592}
2593
Willy Tarreau946ba592009-05-10 15:41:18 +02002594/* copies at most <n> characters from <src> and always terminates with '\0' */
2595char *my_strndup(const char *src, int n)
2596{
2597 int len = 0;
2598 char *ret;
2599
2600 while (len < n && src[len])
2601 len++;
2602
Vincent Bernat3c2f2f22016-04-03 13:48:42 +02002603 ret = malloc(len + 1);
Willy Tarreau946ba592009-05-10 15:41:18 +02002604 if (!ret)
2605 return ret;
2606 memcpy(ret, src, len);
2607 ret[len] = '\0';
2608 return ret;
2609}
2610
Baptiste Assmannbb77c8e2013-10-06 23:24:13 +02002611/*
2612 * search needle in haystack
2613 * returns the pointer if found, returns NULL otherwise
2614 */
2615const void *my_memmem(const void *haystack, size_t haystacklen, const void *needle, size_t needlelen)
2616{
2617 const void *c = NULL;
2618 unsigned char f;
2619
2620 if ((haystack == NULL) || (needle == NULL) || (haystacklen < needlelen))
2621 return NULL;
2622
2623 f = *(char *)needle;
2624 c = haystack;
2625 while ((c = memchr(c, f, haystacklen - (c - haystack))) != NULL) {
2626 if ((haystacklen - (c - haystack)) < needlelen)
2627 return NULL;
2628
2629 if (memcmp(c, needle, needlelen) == 0)
2630 return c;
2631 ++c;
2632 }
2633 return NULL;
2634}
2635
Ilya Shipitsinc02a23f2020-05-06 00:53:22 +05002636/* get length of the initial segment consisting entirely of bytes in <accept> */
Christopher Faulet5eb96cb2020-04-15 10:23:01 +02002637size_t my_memspn(const void *str, size_t len, const void *accept, size_t acceptlen)
2638{
2639 size_t ret = 0;
2640
2641 while (ret < len && memchr(accept, *((int *)str), acceptlen)) {
2642 str++;
2643 ret++;
2644 }
2645 return ret;
2646}
2647
Ilya Shipitsinc02a23f2020-05-06 00:53:22 +05002648/* get length of the initial segment consisting entirely of bytes not in <rejcet> */
Christopher Faulet5eb96cb2020-04-15 10:23:01 +02002649size_t my_memcspn(const void *str, size_t len, const void *reject, size_t rejectlen)
2650{
2651 size_t ret = 0;
2652
2653 while (ret < len) {
2654 if(memchr(reject, *((int *)str), rejectlen))
2655 return ret;
2656 str++;
2657 ret++;
2658 }
2659 return ret;
2660}
2661
Willy Tarreau482b00d2009-10-04 22:48:42 +02002662/* This function returns the first unused key greater than or equal to <key> in
2663 * ID tree <root>. Zero is returned if no place is found.
2664 */
2665unsigned int get_next_id(struct eb_root *root, unsigned int key)
2666{
2667 struct eb32_node *used;
2668
2669 do {
2670 used = eb32_lookup_ge(root, key);
2671 if (!used || used->key > key)
2672 return key; /* key is available */
2673 key++;
2674 } while (key);
2675 return key;
2676}
2677
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002678/* dump the full tree to <file> in DOT format for debugging purposes. Will
2679 * optionally highlight node <subj> if found, depending on operation <op> :
2680 * 0 : nothing
2681 * >0 : insertion, node/leaf are surrounded in red
2682 * <0 : removal, node/leaf are dashed with no background
2683 * Will optionally add "desc" as a label on the graph if set and non-null.
2684 */
2685void 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 +01002686{
2687 struct eb32sc_node *node;
2688 unsigned long scope = -1;
2689
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002690 fprintf(file, "digraph ebtree {\n");
2691
2692 if (desc && *desc) {
2693 fprintf(file,
2694 " fontname=\"fixed\";\n"
2695 " fontsize=8;\n"
2696 " label=\"%s\";\n", desc);
2697 }
2698
Willy Tarreaued3cda02017-11-15 15:04:05 +01002699 fprintf(file,
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002700 " node [fontname=\"fixed\" fontsize=8 shape=\"box\" style=\"filled\" color=\"black\" fillcolor=\"white\"];\n"
2701 " edge [fontname=\"fixed\" fontsize=8 style=\"solid\" color=\"magenta\" dir=\"forward\"];\n"
Willy Tarreaued3cda02017-11-15 15:04:05 +01002702 " \"%lx_n\" [label=\"root\\n%lx\"]\n", (long)eb_root_to_node(root), (long)root
2703 );
2704
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002705 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002706 (long)eb_root_to_node(root),
2707 (long)eb_root_to_node(eb_clrtag(root->b[0])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002708 eb_gettag(root->b[0]) == EB_LEAF ? 'l' : 'n');
2709
2710 node = eb32sc_first(root, scope);
2711 while (node) {
2712 if (node->node.node_p) {
2713 /* node part is used */
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002714 fprintf(file, " \"%lx_n\" [label=\"%lx\\nkey=%u\\nscope=%lx\\nbit=%d\" fillcolor=\"lightskyblue1\" %s];\n",
2715 (long)node, (long)node, node->key, node->node_s, node->node.bit,
2716 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
Willy Tarreaued3cda02017-11-15 15:04:05 +01002717
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002718 fprintf(file, " \"%lx_n\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002719 (long)node,
2720 (long)eb_root_to_node(eb_clrtag(node->node.node_p)),
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002721 eb_gettag(node->node.node_p) ? 'R' : 'L');
Willy Tarreaued3cda02017-11-15 15:04:05 +01002722
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002723 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"L\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002724 (long)node,
2725 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[0])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002726 eb_gettag(node->node.branches.b[0]) == EB_LEAF ? 'l' : 'n');
2727
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002728 fprintf(file, " \"%lx_n\" -> \"%lx_%c\" [taillabel=\"R\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002729 (long)node,
2730 (long)eb_root_to_node(eb_clrtag(node->node.branches.b[1])),
Willy Tarreaued3cda02017-11-15 15:04:05 +01002731 eb_gettag(node->node.branches.b[1]) == EB_LEAF ? 'l' : 'n');
2732 }
2733
Willy Tarreau9c1e15d2017-11-15 18:51:29 +01002734 fprintf(file, " \"%lx_l\" [label=\"%lx\\nkey=%u\\nscope=%lx\\npfx=%u\" fillcolor=\"yellow\" %s];\n",
2735 (long)node, (long)node, node->key, node->leaf_s, node->node.pfx,
2736 (node == subj) ? (op < 0 ? "color=\"red\" style=\"dashed\"" : op > 0 ? "color=\"red\"" : "") : "");
Willy Tarreaued3cda02017-11-15 15:04:05 +01002737
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002738 fprintf(file, " \"%lx_l\" -> \"%lx_n\" [taillabel=\"%c\"];\n",
Willy Tarreaued3cda02017-11-15 15:04:05 +01002739 (long)node,
2740 (long)eb_root_to_node(eb_clrtag(node->node.leaf_p)),
Willy Tarreau6c7f4de2017-11-15 17:49:54 +01002741 eb_gettag(node->node.leaf_p) ? 'R' : 'L');
Willy Tarreaued3cda02017-11-15 15:04:05 +01002742 node = eb32sc_next(node, scope);
2743 }
2744 fprintf(file, "}\n");
2745}
2746
Willy Tarreau348238b2010-01-18 15:05:57 +01002747/* This function compares a sample word possibly followed by blanks to another
2748 * clean word. The compare is case-insensitive. 1 is returned if both are equal,
2749 * otherwise zero. This intends to be used when checking HTTP headers for some
2750 * values. Note that it validates a word followed only by blanks but does not
2751 * validate a word followed by blanks then other chars.
2752 */
2753int word_match(const char *sample, int slen, const char *word, int wlen)
2754{
2755 if (slen < wlen)
2756 return 0;
2757
2758 while (wlen) {
2759 char c = *sample ^ *word;
2760 if (c && c != ('A' ^ 'a'))
2761 return 0;
2762 sample++;
2763 word++;
2764 slen--;
2765 wlen--;
2766 }
2767
2768 while (slen) {
2769 if (*sample != ' ' && *sample != '\t')
2770 return 0;
2771 sample++;
2772 slen--;
2773 }
2774 return 1;
2775}
Willy Tarreau482b00d2009-10-04 22:48:42 +02002776
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002777/* Converts any text-formatted IPv4 address to a host-order IPv4 address. It
2778 * is particularly fast because it avoids expensive operations such as
2779 * multiplies, which are optimized away at the end. It requires a properly
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05002780 * formatted address though (3 points).
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002781 */
2782unsigned int inetaddr_host(const char *text)
2783{
2784 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2785 register unsigned int dig100, dig10, dig1;
2786 int s;
2787 const char *p, *d;
2788
2789 dig1 = dig10 = dig100 = ascii_zero;
2790 s = 24;
2791
2792 p = text;
2793 while (1) {
2794 if (((unsigned)(*p - '0')) <= 9) {
2795 p++;
2796 continue;
2797 }
2798
2799 /* here, we have a complete byte between <text> and <p> (exclusive) */
2800 if (p == text)
2801 goto end;
2802
2803 d = p - 1;
2804 dig1 |= (unsigned int)(*d << s);
2805 if (d == text)
2806 goto end;
2807
2808 d--;
2809 dig10 |= (unsigned int)(*d << s);
2810 if (d == text)
2811 goto end;
2812
2813 d--;
2814 dig100 |= (unsigned int)(*d << s);
2815 end:
2816 if (!s || *p != '.')
2817 break;
2818
2819 s -= 8;
2820 text = ++p;
2821 }
2822
2823 dig100 -= ascii_zero;
2824 dig10 -= ascii_zero;
2825 dig1 -= ascii_zero;
2826 return ((dig100 * 10) + dig10) * 10 + dig1;
2827}
2828
2829/*
2830 * Idem except the first unparsed character has to be passed in <stop>.
2831 */
2832unsigned int inetaddr_host_lim(const char *text, const char *stop)
2833{
2834 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2835 register unsigned int dig100, dig10, dig1;
2836 int s;
2837 const char *p, *d;
2838
2839 dig1 = dig10 = dig100 = ascii_zero;
2840 s = 24;
2841
2842 p = text;
2843 while (1) {
2844 if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2845 p++;
2846 continue;
2847 }
2848
2849 /* here, we have a complete byte between <text> and <p> (exclusive) */
2850 if (p == text)
2851 goto end;
2852
2853 d = p - 1;
2854 dig1 |= (unsigned int)(*d << s);
2855 if (d == text)
2856 goto end;
2857
2858 d--;
2859 dig10 |= (unsigned int)(*d << s);
2860 if (d == text)
2861 goto end;
2862
2863 d--;
2864 dig100 |= (unsigned int)(*d << s);
2865 end:
2866 if (!s || p == stop || *p != '.')
2867 break;
2868
2869 s -= 8;
2870 text = ++p;
2871 }
2872
2873 dig100 -= ascii_zero;
2874 dig10 -= ascii_zero;
2875 dig1 -= ascii_zero;
2876 return ((dig100 * 10) + dig10) * 10 + dig1;
2877}
2878
2879/*
2880 * Idem except the pointer to first unparsed byte is returned into <ret> which
2881 * must not be NULL.
2882 */
Willy Tarreau74172752010-10-15 23:21:42 +02002883unsigned int inetaddr_host_lim_ret(char *text, char *stop, char **ret)
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002884{
2885 const unsigned int ascii_zero = ('0' << 24) | ('0' << 16) | ('0' << 8) | '0';
2886 register unsigned int dig100, dig10, dig1;
2887 int s;
Willy Tarreau74172752010-10-15 23:21:42 +02002888 char *p, *d;
Willy Tarreaud54bbdc2009-09-07 11:00:31 +02002889
2890 dig1 = dig10 = dig100 = ascii_zero;
2891 s = 24;
2892
2893 p = text;
2894 while (1) {
2895 if (((unsigned)(*p - '0')) <= 9 && p < stop) {
2896 p++;
2897 continue;
2898 }
2899
2900 /* here, we have a complete byte between <text> and <p> (exclusive) */
2901 if (p == text)
2902 goto end;
2903
2904 d = p - 1;
2905 dig1 |= (unsigned int)(*d << s);
2906 if (d == text)
2907 goto end;
2908
2909 d--;
2910 dig10 |= (unsigned int)(*d << s);
2911 if (d == text)
2912 goto end;
2913
2914 d--;
2915 dig100 |= (unsigned int)(*d << s);
2916 end:
2917 if (!s || p == stop || *p != '.')
2918 break;
2919
2920 s -= 8;
2921 text = ++p;
2922 }
2923
2924 *ret = p;
2925 dig100 -= ascii_zero;
2926 dig10 -= ascii_zero;
2927 dig1 -= ascii_zero;
2928 return ((dig100 * 10) + dig10) * 10 + dig1;
2929}
2930
Willy Tarreauf0b38bf2010-06-06 13:22:23 +02002931/* Convert a fixed-length string to an IP address. Returns 0 in case of error,
2932 * or the number of chars read in case of success. Maybe this could be replaced
2933 * by one of the functions above. Also, apparently this function does not support
2934 * hosts above 255 and requires exactly 4 octets.
Willy Tarreau075415a2013-12-12 11:29:39 +01002935 * The destination is only modified on success.
Willy Tarreauf0b38bf2010-06-06 13:22:23 +02002936 */
2937int buf2ip(const char *buf, size_t len, struct in_addr *dst)
2938{
2939 const char *addr;
2940 int saw_digit, octets, ch;
2941 u_char tmp[4], *tp;
2942 const char *cp = buf;
2943
2944 saw_digit = 0;
2945 octets = 0;
2946 *(tp = tmp) = 0;
2947
2948 for (addr = buf; addr - buf < len; addr++) {
2949 unsigned char digit = (ch = *addr) - '0';
2950
2951 if (digit > 9 && ch != '.')
2952 break;
2953
2954 if (digit <= 9) {
2955 u_int new = *tp * 10 + digit;
2956
2957 if (new > 255)
2958 return 0;
2959
2960 *tp = new;
2961
2962 if (!saw_digit) {
2963 if (++octets > 4)
2964 return 0;
2965 saw_digit = 1;
2966 }
2967 } else if (ch == '.' && saw_digit) {
2968 if (octets == 4)
2969 return 0;
2970
2971 *++tp = 0;
2972 saw_digit = 0;
2973 } else
2974 return 0;
2975 }
2976
2977 if (octets < 4)
2978 return 0;
2979
2980 memcpy(&dst->s_addr, tmp, 4);
2981 return addr - cp;
2982}
2983
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002984/* This function converts the string in <buf> of the len <len> to
2985 * struct in6_addr <dst> which must be allocated by the caller.
2986 * This function returns 1 in success case, otherwise zero.
Willy Tarreau075415a2013-12-12 11:29:39 +01002987 * The destination is only modified on success.
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002988 */
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002989int buf2ip6(const char *buf, size_t len, struct in6_addr *dst)
2990{
Thierry FOURNIERcd659912013-12-11 12:33:54 +01002991 char null_term_ip6[INET6_ADDRSTRLEN + 1];
Willy Tarreau075415a2013-12-12 11:29:39 +01002992 struct in6_addr out;
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002993
Thierry FOURNIERcd659912013-12-11 12:33:54 +01002994 if (len > INET6_ADDRSTRLEN)
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01002995 return 0;
2996
2997 memcpy(null_term_ip6, buf, len);
2998 null_term_ip6[len] = '\0';
2999
Willy Tarreau075415a2013-12-12 11:29:39 +01003000 if (!inet_pton(AF_INET6, null_term_ip6, &out))
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01003001 return 0;
3002
Willy Tarreau075415a2013-12-12 11:29:39 +01003003 *dst = out;
Thierry FOURNIERd559dd82013-11-22 16:16:59 +01003004 return 1;
3005}
3006
Willy Tarreauacf95772010-06-14 19:09:21 +02003007/* To be used to quote config arg positions. Returns the short string at <ptr>
3008 * surrounded by simple quotes if <ptr> is valid and non-empty, or "end of line"
3009 * if ptr is NULL or empty. The string is locally allocated.
3010 */
3011const char *quote_arg(const char *ptr)
3012{
Christopher Faulet1bc04c72017-10-29 20:14:08 +01003013 static THREAD_LOCAL char val[32];
Willy Tarreauacf95772010-06-14 19:09:21 +02003014 int i;
3015
3016 if (!ptr || !*ptr)
3017 return "end of line";
3018 val[0] = '\'';
Willy Tarreaude2dd6b2013-01-24 02:14:42 +01003019 for (i = 1; i < sizeof(val) - 2 && *ptr; i++)
Willy Tarreauacf95772010-06-14 19:09:21 +02003020 val[i] = *ptr++;
3021 val[i++] = '\'';
3022 val[i] = '\0';
3023 return val;
3024}
3025
Willy Tarreau5b180202010-07-18 10:40:48 +02003026/* returns an operator among STD_OP_* for string <str> or < 0 if unknown */
3027int get_std_op(const char *str)
3028{
3029 int ret = -1;
3030
3031 if (*str == 'e' && str[1] == 'q')
3032 ret = STD_OP_EQ;
3033 else if (*str == 'n' && str[1] == 'e')
3034 ret = STD_OP_NE;
3035 else if (*str == 'l') {
3036 if (str[1] == 'e') ret = STD_OP_LE;
3037 else if (str[1] == 't') ret = STD_OP_LT;
3038 }
3039 else if (*str == 'g') {
3040 if (str[1] == 'e') ret = STD_OP_GE;
3041 else if (str[1] == 't') ret = STD_OP_GT;
3042 }
3043
3044 if (ret == -1 || str[2] != '\0')
3045 return -1;
3046 return ret;
3047}
3048
Willy Tarreau4c14eaa2010-11-24 14:01:45 +01003049/* hash a 32-bit integer to another 32-bit integer */
3050unsigned int full_hash(unsigned int a)
3051{
3052 return __full_hash(a);
3053}
3054
Willy Tarreauf3241112019-02-26 09:56:22 +01003055/* Return the bit position in mask <m> of the nth bit set of rank <r>, between
3056 * 0 and LONGBITS-1 included, starting from the left. For example ranks 0,1,2,3
3057 * for mask 0x55 will be 6, 4, 2 and 0 respectively. This algorithm is based on
3058 * a popcount variant and is described here :
3059 * https://graphics.stanford.edu/~seander/bithacks.html
3060 */
3061unsigned int mask_find_rank_bit(unsigned int r, unsigned long m)
3062{
3063 unsigned long a, b, c, d;
3064 unsigned int s;
3065 unsigned int t;
3066
3067 a = m - ((m >> 1) & ~0UL/3);
3068 b = (a & ~0UL/5) + ((a >> 2) & ~0UL/5);
3069 c = (b + (b >> 4)) & ~0UL/0x11;
3070 d = (c + (c >> 8)) & ~0UL/0x101;
3071
3072 r++; // make r be 1..64
3073
3074 t = 0;
3075 s = LONGBITS;
3076 if (s > 32) {
Willy Tarreau9b6be3b2019-03-18 16:31:18 +01003077 unsigned long d2 = (d >> 16) >> 16;
3078 t = d2 + (d2 >> 16);
Willy Tarreauf3241112019-02-26 09:56:22 +01003079 s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
3080 }
3081
3082 t = (d >> (s - 16)) & 0xff;
3083 s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
3084 t = (c >> (s - 8)) & 0xf;
3085 s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
3086 t = (b >> (s - 4)) & 0x7;
3087 s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
3088 t = (a >> (s - 2)) & 0x3;
3089 s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
3090 t = (m >> (s - 1)) & 0x1;
3091 s -= ((t - r) & 256) >> 8;
3092
3093 return s - 1;
3094}
3095
3096/* Same as mask_find_rank_bit() above but makes use of pre-computed bitmaps
3097 * based on <m>, in <a..d>. These ones must be updated whenever <m> changes
3098 * using mask_prep_rank_map() below.
3099 */
3100unsigned int mask_find_rank_bit_fast(unsigned int r, unsigned long m,
3101 unsigned long a, unsigned long b,
3102 unsigned long c, unsigned long d)
3103{
3104 unsigned int s;
3105 unsigned int t;
3106
3107 r++; // make r be 1..64
3108
3109 t = 0;
3110 s = LONGBITS;
3111 if (s > 32) {
Willy Tarreau9b6be3b2019-03-18 16:31:18 +01003112 unsigned long d2 = (d >> 16) >> 16;
3113 t = d2 + (d2 >> 16);
Willy Tarreauf3241112019-02-26 09:56:22 +01003114 s -= ((t - r) & 256) >> 3; r -= (t & ((t - r) >> 8));
3115 }
3116
3117 t = (d >> (s - 16)) & 0xff;
3118 s -= ((t - r) & 256) >> 4; r -= (t & ((t - r) >> 8));
3119 t = (c >> (s - 8)) & 0xf;
3120 s -= ((t - r) & 256) >> 5; r -= (t & ((t - r) >> 8));
3121 t = (b >> (s - 4)) & 0x7;
3122 s -= ((t - r) & 256) >> 6; r -= (t & ((t - r) >> 8));
3123 t = (a >> (s - 2)) & 0x3;
3124 s -= ((t - r) & 256) >> 7; r -= (t & ((t - r) >> 8));
3125 t = (m >> (s - 1)) & 0x1;
3126 s -= ((t - r) & 256) >> 8;
3127
3128 return s - 1;
3129}
3130
3131/* Prepare the bitmaps used by the fast implementation of the find_rank_bit()
3132 * above.
3133 */
3134void mask_prep_rank_map(unsigned long m,
3135 unsigned long *a, unsigned long *b,
3136 unsigned long *c, unsigned long *d)
3137{
3138 *a = m - ((m >> 1) & ~0UL/3);
3139 *b = (*a & ~0UL/5) + ((*a >> 2) & ~0UL/5);
3140 *c = (*b + (*b >> 4)) & ~0UL/0x11;
3141 *d = (*c + (*c >> 8)) & ~0UL/0x101;
3142}
3143
David du Colombier4f92d322011-03-24 11:09:31 +01003144/* Return non-zero if IPv4 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_ipv4(const void *addr, const struct in_addr *mask, const struct in_addr *net)
David du Colombier4f92d322011-03-24 11:09:31 +01003149{
Willy Tarreaueec1d382016-07-13 11:59:39 +02003150 struct in_addr addr_copy;
3151
3152 memcpy(&addr_copy, addr, sizeof(addr_copy));
3153 return((addr_copy.s_addr & mask->s_addr) == (net->s_addr & mask->s_addr));
David du Colombier4f92d322011-03-24 11:09:31 +01003154}
3155
3156/* Return non-zero if IPv6 address is part of the network,
Willy Tarreaueec1d382016-07-13 11:59:39 +02003157 * otherwise zero. Note that <addr> may not necessarily be aligned
3158 * while the two other ones must.
David du Colombier4f92d322011-03-24 11:09:31 +01003159 */
Willy Tarreaueec1d382016-07-13 11:59:39 +02003160int in_net_ipv6(const void *addr, const struct in6_addr *mask, const struct in6_addr *net)
David du Colombier4f92d322011-03-24 11:09:31 +01003161{
3162 int i;
Willy Tarreaueec1d382016-07-13 11:59:39 +02003163 struct in6_addr addr_copy;
David du Colombier4f92d322011-03-24 11:09:31 +01003164
Willy Tarreaueec1d382016-07-13 11:59:39 +02003165 memcpy(&addr_copy, addr, sizeof(addr_copy));
David du Colombier4f92d322011-03-24 11:09:31 +01003166 for (i = 0; i < sizeof(struct in6_addr) / sizeof(int); i++)
Willy Tarreaueec1d382016-07-13 11:59:39 +02003167 if (((((int *)&addr_copy)[i] & ((int *)mask)[i])) !=
David du Colombier4f92d322011-03-24 11:09:31 +01003168 (((int *)net)[i] & ((int *)mask)[i]))
3169 return 0;
3170 return 1;
3171}
3172
3173/* RFC 4291 prefix */
3174const char rfc4291_pfx[] = { 0x00, 0x00, 0x00, 0x00,
3175 0x00, 0x00, 0x00, 0x00,
3176 0x00, 0x00, 0xFF, 0xFF };
3177
Joseph Herlant32b83272018-11-15 11:58:28 -08003178/* Map IPv4 address on IPv6 address, as specified in RFC 3513.
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003179 * Input and output may overlap.
3180 */
David du Colombier4f92d322011-03-24 11:09:31 +01003181void v4tov6(struct in6_addr *sin6_addr, struct in_addr *sin_addr)
3182{
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003183 struct in_addr tmp_addr;
3184
3185 tmp_addr.s_addr = sin_addr->s_addr;
David du Colombier4f92d322011-03-24 11:09:31 +01003186 memcpy(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx));
Thierry FOURNIER4a04dc32013-11-28 16:33:15 +01003187 memcpy(sin6_addr->s6_addr+12, &tmp_addr.s_addr, 4);
David du Colombier4f92d322011-03-24 11:09:31 +01003188}
3189
Joseph Herlant32b83272018-11-15 11:58:28 -08003190/* Map IPv6 address on IPv4 address, as specified in RFC 3513.
David du Colombier4f92d322011-03-24 11:09:31 +01003191 * Return true if conversion is possible and false otherwise.
3192 */
3193int v6tov4(struct in_addr *sin_addr, struct in6_addr *sin6_addr)
3194{
3195 if (memcmp(sin6_addr->s6_addr, rfc4291_pfx, sizeof(rfc4291_pfx)) == 0) {
3196 memcpy(&(sin_addr->s_addr), &(sin6_addr->s6_addr[12]),
3197 sizeof(struct in_addr));
3198 return 1;
3199 }
3200
3201 return 0;
3202}
3203
Baptiste Assmann08b24cf2016-01-23 23:39:12 +01003204/* compare two struct sockaddr_storage and return:
3205 * 0 (true) if the addr is the same in both
3206 * 1 (false) if the addr is not the same in both
3207 * -1 (unable) if one of the addr is not AF_INET*
3208 */
3209int ipcmp(struct sockaddr_storage *ss1, struct sockaddr_storage *ss2)
3210{
3211 if ((ss1->ss_family != AF_INET) && (ss1->ss_family != AF_INET6))
3212 return -1;
3213
3214 if ((ss2->ss_family != AF_INET) && (ss2->ss_family != AF_INET6))
3215 return -1;
3216
3217 if (ss1->ss_family != ss2->ss_family)
3218 return 1;
3219
3220 switch (ss1->ss_family) {
3221 case AF_INET:
3222 return memcmp(&((struct sockaddr_in *)ss1)->sin_addr,
3223 &((struct sockaddr_in *)ss2)->sin_addr,
3224 sizeof(struct in_addr)) != 0;
3225 case AF_INET6:
3226 return memcmp(&((struct sockaddr_in6 *)ss1)->sin6_addr,
3227 &((struct sockaddr_in6 *)ss2)->sin6_addr,
3228 sizeof(struct in6_addr)) != 0;
3229 }
3230
3231 return 1;
3232}
3233
Christopher Faulet9553de72021-02-26 09:12:50 +01003234/* compare a struct sockaddr_storage to a struct net_addr and return :
3235 * 0 (true) if <addr> is matching <net>
3236 * 1 (false) if <addr> is not matching <net>
3237 * -1 (unable) if <addr> or <net> is not AF_INET*
3238 */
3239int ipcmp2net(const struct sockaddr_storage *addr, const struct net_addr *net)
3240{
3241 if ((addr->ss_family != AF_INET) && (addr->ss_family != AF_INET6))
3242 return -1;
3243
3244 if ((net->family != AF_INET) && (net->family != AF_INET6))
3245 return -1;
3246
3247 if (addr->ss_family != net->family)
3248 return 1;
3249
3250 if (addr->ss_family == AF_INET &&
3251 (((struct sockaddr_in *)addr)->sin_addr.s_addr & net->addr.v4.mask.s_addr) == net->addr.v4.ip.s_addr)
3252 return 0;
3253 else {
3254 const struct in6_addr *addr6 = &(((const struct sockaddr_in6*)addr)->sin6_addr);
3255 const struct in6_addr *nip6 = &net->addr.v6.ip;
3256 const struct in6_addr *nmask6 = &net->addr.v6.mask;
3257
3258 if ((read_u32(&addr6->s6_addr[0]) & read_u32(&nmask6->s6_addr[0])) == read_u32(&nip6->s6_addr[0]) &&
3259 (read_u32(&addr6->s6_addr[4]) & read_u32(&nmask6->s6_addr[4])) == read_u32(&nip6->s6_addr[4]) &&
3260 (read_u32(&addr6->s6_addr[8]) & read_u32(&nmask6->s6_addr[8])) == read_u32(&nip6->s6_addr[8]) &&
3261 (read_u32(&addr6->s6_addr[12]) & read_u32(&nmask6->s6_addr[12])) == read_u32(&nip6->s6_addr[12]))
3262 return 0;
3263 }
3264
3265 return 1;
3266}
3267
Baptiste Assmann08396c82016-01-31 00:27:17 +01003268/* copy IP address from <source> into <dest>
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003269 * The caller must allocate and clear <dest> before calling.
3270 * The source must be in either AF_INET or AF_INET6 family, or the destination
3271 * address will be undefined. If the destination address used to hold a port,
3272 * it is preserved, so that this function can be used to switch to another
3273 * address family with no risk. Returns a pointer to the destination.
Baptiste Assmann08396c82016-01-31 00:27:17 +01003274 */
3275struct sockaddr_storage *ipcpy(struct sockaddr_storage *source, struct sockaddr_storage *dest)
3276{
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003277 int prev_port;
3278
3279 prev_port = get_net_port(dest);
3280 memset(dest, 0, sizeof(*dest));
Baptiste Assmann08396c82016-01-31 00:27:17 +01003281 dest->ss_family = source->ss_family;
3282
3283 /* copy new addr and apply it */
3284 switch (source->ss_family) {
3285 case AF_INET:
3286 ((struct sockaddr_in *)dest)->sin_addr.s_addr = ((struct sockaddr_in *)source)->sin_addr.s_addr;
Willy Tarreaudc3a9e82016-11-04 18:47:01 +01003287 ((struct sockaddr_in *)dest)->sin_port = prev_port;
Baptiste Assmann08396c82016-01-31 00:27:17 +01003288 break;
3289 case AF_INET6:
3290 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 +01003291 ((struct sockaddr_in6 *)dest)->sin6_port = prev_port;
Baptiste Assmann08396c82016-01-31 00:27:17 +01003292 break;
3293 }
3294
3295 return dest;
3296}
3297
William Lallemand421f5b52012-02-06 18:15:57 +01003298char *human_time(int t, short hz_div) {
3299 static char rv[sizeof("24855d23h")+1]; // longest of "23h59m" and "59m59s"
3300 char *p = rv;
Willy Tarreau761b3d52014-04-14 14:53:06 +02003301 char *end = rv + sizeof(rv);
William Lallemand421f5b52012-02-06 18:15:57 +01003302 int cnt=2; // print two numbers
3303
3304 if (unlikely(t < 0 || hz_div <= 0)) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003305 snprintf(p, end - p, "?");
William Lallemand421f5b52012-02-06 18:15:57 +01003306 return rv;
3307 }
3308
3309 if (unlikely(hz_div > 1))
3310 t /= hz_div;
3311
3312 if (t >= DAY) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003313 p += snprintf(p, end - p, "%dd", t / DAY);
William Lallemand421f5b52012-02-06 18:15:57 +01003314 cnt--;
3315 }
3316
3317 if (cnt && t % DAY / HOUR) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003318 p += snprintf(p, end - p, "%dh", t % DAY / HOUR);
William Lallemand421f5b52012-02-06 18:15:57 +01003319 cnt--;
3320 }
3321
3322 if (cnt && t % HOUR / MINUTE) {
Willy Tarreau761b3d52014-04-14 14:53:06 +02003323 p += snprintf(p, end - p, "%dm", t % HOUR / MINUTE);
William Lallemand421f5b52012-02-06 18:15:57 +01003324 cnt--;
3325 }
3326
3327 if ((cnt && t % MINUTE) || !t) // also display '0s'
Willy Tarreau761b3d52014-04-14 14:53:06 +02003328 p += snprintf(p, end - p, "%ds", t % MINUTE / SEC);
William Lallemand421f5b52012-02-06 18:15:57 +01003329
3330 return rv;
3331}
3332
3333const char *monthname[12] = {
3334 "Jan", "Feb", "Mar", "Apr", "May", "Jun",
3335 "Jul", "Aug", "Sep", "Oct", "Nov", "Dec"
3336};
3337
3338/* date2str_log: write a date in the format :
3339 * sprintf(str, "%02d/%s/%04d:%02d:%02d:%02d.%03d",
3340 * tm.tm_mday, monthname[tm.tm_mon], tm.tm_year+1900,
3341 * tm.tm_hour, tm.tm_min, tm.tm_sec, (int)date.tv_usec/1000);
3342 *
3343 * without using sprintf. return a pointer to the last char written (\0) or
3344 * NULL if there isn't enough space.
3345 */
Willy Tarreauf16cb412018-09-04 19:08:48 +02003346char *date2str_log(char *dst, const struct tm *tm, const struct timeval *date, size_t size)
William Lallemand421f5b52012-02-06 18:15:57 +01003347{
3348
3349 if (size < 25) /* the size is fixed: 24 chars + \0 */
3350 return NULL;
3351
3352 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003353 if (!dst)
3354 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003355 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003356
William Lallemand421f5b52012-02-06 18:15:57 +01003357 memcpy(dst, monthname[tm->tm_mon], 3); // month
3358 dst += 3;
3359 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003360
William Lallemand421f5b52012-02-06 18:15:57 +01003361 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003362 if (!dst)
3363 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003364 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003365
William Lallemand421f5b52012-02-06 18:15:57 +01003366 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003367 if (!dst)
3368 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003369 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003370
William Lallemand421f5b52012-02-06 18:15:57 +01003371 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003372 if (!dst)
3373 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003374 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003375
William Lallemand421f5b52012-02-06 18:15:57 +01003376 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003377 if (!dst)
3378 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003379 *dst++ = '.';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003380
Willy Tarreau7d9421d2020-02-29 09:08:02 +01003381 dst = utoa_pad((unsigned int)(date->tv_usec/1000)%1000, dst, 4); // milliseconds
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003382 if (!dst)
3383 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003384 *dst = '\0';
3385
3386 return dst;
3387}
3388
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003389/* Base year used to compute leap years */
3390#define TM_YEAR_BASE 1900
3391
3392/* Return the difference in seconds between two times (leap seconds are ignored).
3393 * Retrieved from glibc 2.18 source code.
3394 */
3395static int my_tm_diff(const struct tm *a, const struct tm *b)
3396{
3397 /* Compute intervening leap days correctly even if year is negative.
3398 * Take care to avoid int overflow in leap day calculations,
3399 * but it's OK to assume that A and B are close to each other.
3400 */
3401 int a4 = (a->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (a->tm_year & 3);
3402 int b4 = (b->tm_year >> 2) + (TM_YEAR_BASE >> 2) - ! (b->tm_year & 3);
3403 int a100 = a4 / 25 - (a4 % 25 < 0);
3404 int b100 = b4 / 25 - (b4 % 25 < 0);
3405 int a400 = a100 >> 2;
3406 int b400 = b100 >> 2;
3407 int intervening_leap_days = (a4 - b4) - (a100 - b100) + (a400 - b400);
3408 int years = a->tm_year - b->tm_year;
3409 int days = (365 * years + intervening_leap_days
3410 + (a->tm_yday - b->tm_yday));
3411 return (60 * (60 * (24 * days + (a->tm_hour - b->tm_hour))
3412 + (a->tm_min - b->tm_min))
3413 + (a->tm_sec - b->tm_sec));
3414}
3415
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003416/* Return the GMT offset for a specific local time.
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003417 * Both t and tm must represent the same time.
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003418 * The string returned has the same format as returned by strftime(... "%z", tm).
3419 * Offsets are kept in an internal cache for better performances.
3420 */
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003421const char *get_gmt_offset(time_t t, struct tm *tm)
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003422{
3423 /* Cache offsets from GMT (depending on whether DST is active or not) */
Christopher Faulet1bc04c72017-10-29 20:14:08 +01003424 static THREAD_LOCAL char gmt_offsets[2][5+1] = { "", "" };
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003425
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003426 char *gmt_offset;
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003427 struct tm tm_gmt;
3428 int diff;
3429 int isdst = tm->tm_isdst;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003430
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003431 /* Pretend DST not active if its status is unknown */
3432 if (isdst < 0)
3433 isdst = 0;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003434
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003435 /* Fetch the offset and initialize it if needed */
3436 gmt_offset = gmt_offsets[isdst & 0x01];
3437 if (unlikely(!*gmt_offset)) {
3438 get_gmtime(t, &tm_gmt);
3439 diff = my_tm_diff(tm, &tm_gmt);
3440 if (diff < 0) {
3441 diff = -diff;
3442 *gmt_offset = '-';
3443 } else {
3444 *gmt_offset = '+';
3445 }
Willy Tarreaue112c8a2019-10-29 10:16:11 +01003446 diff %= 86400U;
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003447 diff /= 60; /* Convert to minutes */
3448 snprintf(gmt_offset+1, 4+1, "%02d%02d", diff/60, diff%60);
3449 }
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003450
Willy Tarreaue112c8a2019-10-29 10:16:11 +01003451 return gmt_offset;
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003452}
3453
William Lallemand421f5b52012-02-06 18:15:57 +01003454/* gmt2str_log: write a date in the format :
3455 * "%02d/%s/%04d:%02d:%02d:%02d +0000" without using snprintf
3456 * return a pointer to the last char written (\0) or
3457 * NULL if there isn't enough space.
3458 */
3459char *gmt2str_log(char *dst, struct tm *tm, size_t size)
3460{
Yuxans Yao4e25b012012-10-19 10:36:09 +08003461 if (size < 27) /* the size is fixed: 26 chars + \0 */
William Lallemand421f5b52012-02-06 18:15:57 +01003462 return NULL;
3463
3464 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003465 if (!dst)
3466 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003467 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003468
William Lallemand421f5b52012-02-06 18:15:57 +01003469 memcpy(dst, monthname[tm->tm_mon], 3); // month
3470 dst += 3;
3471 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003472
William Lallemand421f5b52012-02-06 18:15:57 +01003473 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003474 if (!dst)
3475 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003476 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003477
William Lallemand421f5b52012-02-06 18:15:57 +01003478 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003479 if (!dst)
3480 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003481 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003482
William Lallemand421f5b52012-02-06 18:15:57 +01003483 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003484 if (!dst)
3485 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003486 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003487
William Lallemand421f5b52012-02-06 18:15:57 +01003488 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003489 if (!dst)
3490 return NULL;
William Lallemand421f5b52012-02-06 18:15:57 +01003491 *dst++ = ' ';
3492 *dst++ = '+';
3493 *dst++ = '0';
3494 *dst++ = '0';
3495 *dst++ = '0';
3496 *dst++ = '0';
3497 *dst = '\0';
3498
3499 return dst;
3500}
3501
Yuxans Yao4e25b012012-10-19 10:36:09 +08003502/* localdate2str_log: write a date in the format :
3503 * "%02d/%s/%04d:%02d:%02d:%02d +0000(local timezone)" without using snprintf
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003504 * Both t and tm must represent the same time.
3505 * return a pointer to the last char written (\0) or
3506 * NULL if there isn't enough space.
Yuxans Yao4e25b012012-10-19 10:36:09 +08003507 */
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003508char *localdate2str_log(char *dst, time_t t, struct tm *tm, size_t size)
Yuxans Yao4e25b012012-10-19 10:36:09 +08003509{
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003510 const char *gmt_offset;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003511 if (size < 27) /* the size is fixed: 26 chars + \0 */
3512 return NULL;
3513
Benoit GARNIERe2e5bde2016-03-27 03:04:16 +02003514 gmt_offset = get_gmt_offset(t, tm);
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003515
Yuxans Yao4e25b012012-10-19 10:36:09 +08003516 dst = utoa_pad((unsigned int)tm->tm_mday, dst, 3); // day
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003517 if (!dst)
3518 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003519 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003520
Yuxans Yao4e25b012012-10-19 10:36:09 +08003521 memcpy(dst, monthname[tm->tm_mon], 3); // month
3522 dst += 3;
3523 *dst++ = '/';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003524
Yuxans Yao4e25b012012-10-19 10:36:09 +08003525 dst = utoa_pad((unsigned int)tm->tm_year+1900, dst, 5); // year
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003526 if (!dst)
3527 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003528 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003529
Yuxans Yao4e25b012012-10-19 10:36:09 +08003530 dst = utoa_pad((unsigned int)tm->tm_hour, dst, 3); // hour
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003531 if (!dst)
3532 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003533 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003534
Yuxans Yao4e25b012012-10-19 10:36:09 +08003535 dst = utoa_pad((unsigned int)tm->tm_min, dst, 3); // minutes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003536 if (!dst)
3537 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003538 *dst++ = ':';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003539
Yuxans Yao4e25b012012-10-19 10:36:09 +08003540 dst = utoa_pad((unsigned int)tm->tm_sec, dst, 3); // secondes
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003541 if (!dst)
3542 return NULL;
Yuxans Yao4e25b012012-10-19 10:36:09 +08003543 *dst++ = ' ';
Willy Tarreau4eee38a2019-02-12 11:26:29 +01003544
Benoit GARNIERb413c2a2016-03-27 11:08:03 +02003545 memcpy(dst, gmt_offset, 5); // Offset from local time to GMT
Yuxans Yao4e25b012012-10-19 10:36:09 +08003546 dst += 5;
3547 *dst = '\0';
3548
3549 return dst;
3550}
3551
Willy Tarreaucb1949b2017-07-19 19:05:29 +02003552/* Returns the number of seconds since 01/01/1970 0:0:0 GMT for GMT date <tm>.
3553 * It is meant as a portable replacement for timegm() for use with valid inputs.
3554 * Returns undefined results for invalid dates (eg: months out of range 0..11).
3555 */
3556time_t my_timegm(const struct tm *tm)
3557{
3558 /* Each month has 28, 29, 30 or 31 days, or 28+N. The date in the year
3559 * is thus (current month - 1)*28 + cumulated_N[month] to count the
3560 * sum of the extra N days for elapsed months. The sum of all these N
3561 * days doesn't exceed 30 for a complete year (366-12*28) so it fits
3562 * in a 5-bit word. This means that with 60 bits we can represent a
3563 * matrix of all these values at once, which is fast and efficient to
3564 * access. The extra February day for leap years is not counted here.
3565 *
3566 * Jan : none = 0 (0)
3567 * Feb : Jan = 3 (3)
3568 * Mar : Jan..Feb = 3 (3 + 0)
3569 * Apr : Jan..Mar = 6 (3 + 0 + 3)
3570 * May : Jan..Apr = 8 (3 + 0 + 3 + 2)
3571 * Jun : Jan..May = 11 (3 + 0 + 3 + 2 + 3)
3572 * Jul : Jan..Jun = 13 (3 + 0 + 3 + 2 + 3 + 2)
3573 * Aug : Jan..Jul = 16 (3 + 0 + 3 + 2 + 3 + 2 + 3)
3574 * Sep : Jan..Aug = 19 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3)
3575 * Oct : Jan..Sep = 21 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2)
3576 * Nov : Jan..Oct = 24 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3)
3577 * Dec : Jan..Nov = 26 (3 + 0 + 3 + 2 + 3 + 2 + 3 + 3 + 2 + 3 + 2)
3578 */
3579 uint64_t extra =
3580 ( 0ULL << 0*5) + ( 3ULL << 1*5) + ( 3ULL << 2*5) + /* Jan, Feb, Mar, */
3581 ( 6ULL << 3*5) + ( 8ULL << 4*5) + (11ULL << 5*5) + /* Apr, May, Jun, */
3582 (13ULL << 6*5) + (16ULL << 7*5) + (19ULL << 8*5) + /* Jul, Aug, Sep, */
3583 (21ULL << 9*5) + (24ULL << 10*5) + (26ULL << 11*5); /* Oct, Nov, Dec, */
3584
3585 unsigned int y = tm->tm_year + 1900;
3586 unsigned int m = tm->tm_mon;
3587 unsigned long days = 0;
3588
3589 /* days since 1/1/1970 for full years */
3590 days += days_since_zero(y) - days_since_zero(1970);
3591
3592 /* days for full months in the current year */
3593 days += 28 * m + ((extra >> (m * 5)) & 0x1f);
3594
3595 /* count + 1 after March for leap years. A leap year is a year multiple
3596 * of 4, unless it's multiple of 100 without being multiple of 400. 2000
3597 * is leap, 1900 isn't, 1904 is.
3598 */
3599 if ((m > 1) && !(y & 3) && ((y % 100) || !(y % 400)))
3600 days++;
3601
3602 days += tm->tm_mday - 1;
3603 return days * 86400ULL + tm->tm_hour * 3600 + tm->tm_min * 60 + tm->tm_sec;
3604}
3605
Thierry Fournier93127942016-01-20 18:49:45 +01003606/* This function check a char. It returns true and updates
3607 * <date> and <len> pointer to the new position if the
3608 * character is found.
3609 */
3610static inline int parse_expect_char(const char **date, int *len, char c)
3611{
3612 if (*len < 1 || **date != c)
3613 return 0;
3614 (*len)--;
3615 (*date)++;
3616 return 1;
3617}
3618
3619/* This function expects a string <str> of len <l>. It return true and updates.
3620 * <date> and <len> if the string matches, otherwise, it returns false.
3621 */
3622static inline int parse_strcmp(const char **date, int *len, char *str, int l)
3623{
3624 if (*len < l || strncmp(*date, str, l) != 0)
3625 return 0;
3626 (*len) -= l;
3627 (*date) += l;
3628 return 1;
3629}
3630
3631/* This macro converts 3 chars name in integer. */
3632#define STR2I3(__a, __b, __c) ((__a) * 65536 + (__b) * 256 + (__c))
3633
3634/* day-name = %x4D.6F.6E ; "Mon", case-sensitive
3635 * / %x54.75.65 ; "Tue", case-sensitive
3636 * / %x57.65.64 ; "Wed", case-sensitive
3637 * / %x54.68.75 ; "Thu", case-sensitive
3638 * / %x46.72.69 ; "Fri", case-sensitive
3639 * / %x53.61.74 ; "Sat", case-sensitive
3640 * / %x53.75.6E ; "Sun", case-sensitive
3641 *
3642 * This array must be alphabetically sorted
3643 */
3644static inline int parse_http_dayname(const char **date, int *len, struct tm *tm)
3645{
3646 if (*len < 3)
3647 return 0;
3648 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3649 case STR2I3('M','o','n'): tm->tm_wday = 1; break;
3650 case STR2I3('T','u','e'): tm->tm_wday = 2; break;
3651 case STR2I3('W','e','d'): tm->tm_wday = 3; break;
3652 case STR2I3('T','h','u'): tm->tm_wday = 4; break;
3653 case STR2I3('F','r','i'): tm->tm_wday = 5; break;
3654 case STR2I3('S','a','t'): tm->tm_wday = 6; break;
3655 case STR2I3('S','u','n'): tm->tm_wday = 7; break;
3656 default: return 0;
3657 }
3658 *len -= 3;
3659 *date += 3;
3660 return 1;
3661}
3662
3663/* month = %x4A.61.6E ; "Jan", case-sensitive
3664 * / %x46.65.62 ; "Feb", case-sensitive
3665 * / %x4D.61.72 ; "Mar", case-sensitive
3666 * / %x41.70.72 ; "Apr", case-sensitive
3667 * / %x4D.61.79 ; "May", case-sensitive
3668 * / %x4A.75.6E ; "Jun", case-sensitive
3669 * / %x4A.75.6C ; "Jul", case-sensitive
3670 * / %x41.75.67 ; "Aug", case-sensitive
3671 * / %x53.65.70 ; "Sep", case-sensitive
3672 * / %x4F.63.74 ; "Oct", case-sensitive
3673 * / %x4E.6F.76 ; "Nov", case-sensitive
3674 * / %x44.65.63 ; "Dec", case-sensitive
3675 *
3676 * This array must be alphabetically sorted
3677 */
3678static inline int parse_http_monthname(const char **date, int *len, struct tm *tm)
3679{
3680 if (*len < 3)
3681 return 0;
3682 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3683 case STR2I3('J','a','n'): tm->tm_mon = 0; break;
3684 case STR2I3('F','e','b'): tm->tm_mon = 1; break;
3685 case STR2I3('M','a','r'): tm->tm_mon = 2; break;
3686 case STR2I3('A','p','r'): tm->tm_mon = 3; break;
3687 case STR2I3('M','a','y'): tm->tm_mon = 4; break;
3688 case STR2I3('J','u','n'): tm->tm_mon = 5; break;
3689 case STR2I3('J','u','l'): tm->tm_mon = 6; break;
3690 case STR2I3('A','u','g'): tm->tm_mon = 7; break;
3691 case STR2I3('S','e','p'): tm->tm_mon = 8; break;
3692 case STR2I3('O','c','t'): tm->tm_mon = 9; break;
3693 case STR2I3('N','o','v'): tm->tm_mon = 10; break;
3694 case STR2I3('D','e','c'): tm->tm_mon = 11; break;
3695 default: return 0;
3696 }
3697 *len -= 3;
3698 *date += 3;
3699 return 1;
3700}
3701
3702/* day-name-l = %x4D.6F.6E.64.61.79 ; "Monday", case-sensitive
3703 * / %x54.75.65.73.64.61.79 ; "Tuesday", case-sensitive
3704 * / %x57.65.64.6E.65.73.64.61.79 ; "Wednesday", case-sensitive
3705 * / %x54.68.75.72.73.64.61.79 ; "Thursday", case-sensitive
3706 * / %x46.72.69.64.61.79 ; "Friday", case-sensitive
3707 * / %x53.61.74.75.72.64.61.79 ; "Saturday", case-sensitive
3708 * / %x53.75.6E.64.61.79 ; "Sunday", case-sensitive
3709 *
3710 * This array must be alphabetically sorted
3711 */
3712static inline int parse_http_ldayname(const char **date, int *len, struct tm *tm)
3713{
3714 if (*len < 6) /* Minimum length. */
3715 return 0;
3716 switch (STR2I3((*date)[0], (*date)[1], (*date)[2])) {
3717 case STR2I3('M','o','n'):
3718 RET0_UNLESS(parse_strcmp(date, len, "Monday", 6));
3719 tm->tm_wday = 1;
3720 return 1;
3721 case STR2I3('T','u','e'):
3722 RET0_UNLESS(parse_strcmp(date, len, "Tuesday", 7));
3723 tm->tm_wday = 2;
3724 return 1;
3725 case STR2I3('W','e','d'):
3726 RET0_UNLESS(parse_strcmp(date, len, "Wednesday", 9));
3727 tm->tm_wday = 3;
3728 return 1;
3729 case STR2I3('T','h','u'):
3730 RET0_UNLESS(parse_strcmp(date, len, "Thursday", 8));
3731 tm->tm_wday = 4;
3732 return 1;
3733 case STR2I3('F','r','i'):
3734 RET0_UNLESS(parse_strcmp(date, len, "Friday", 6));
3735 tm->tm_wday = 5;
3736 return 1;
3737 case STR2I3('S','a','t'):
3738 RET0_UNLESS(parse_strcmp(date, len, "Saturday", 8));
3739 tm->tm_wday = 6;
3740 return 1;
3741 case STR2I3('S','u','n'):
3742 RET0_UNLESS(parse_strcmp(date, len, "Sunday", 6));
3743 tm->tm_wday = 7;
3744 return 1;
3745 }
3746 return 0;
3747}
3748
3749/* This function parses exactly 1 digit and returns the numeric value in "digit". */
3750static inline int parse_digit(const char **date, int *len, int *digit)
3751{
3752 if (*len < 1 || **date < '0' || **date > '9')
3753 return 0;
3754 *digit = (**date - '0');
3755 (*date)++;
3756 (*len)--;
3757 return 1;
3758}
3759
3760/* This function parses exactly 2 digits and returns the numeric value in "digit". */
3761static inline int parse_2digit(const char **date, int *len, int *digit)
3762{
3763 int value;
3764
3765 RET0_UNLESS(parse_digit(date, len, &value));
3766 (*digit) = value * 10;
3767 RET0_UNLESS(parse_digit(date, len, &value));
3768 (*digit) += value;
3769
3770 return 1;
3771}
3772
3773/* This function parses exactly 4 digits and returns the numeric value in "digit". */
3774static inline int parse_4digit(const char **date, int *len, int *digit)
3775{
3776 int value;
3777
3778 RET0_UNLESS(parse_digit(date, len, &value));
3779 (*digit) = value * 1000;
3780
3781 RET0_UNLESS(parse_digit(date, len, &value));
3782 (*digit) += value * 100;
3783
3784 RET0_UNLESS(parse_digit(date, len, &value));
3785 (*digit) += value * 10;
3786
3787 RET0_UNLESS(parse_digit(date, len, &value));
3788 (*digit) += value;
3789
3790 return 1;
3791}
3792
3793/* time-of-day = hour ":" minute ":" second
3794 * ; 00:00:00 - 23:59:60 (leap second)
3795 *
3796 * hour = 2DIGIT
3797 * minute = 2DIGIT
3798 * second = 2DIGIT
3799 */
3800static inline int parse_http_time(const char **date, int *len, struct tm *tm)
3801{
3802 RET0_UNLESS(parse_2digit(date, len, &tm->tm_hour)); /* hour 2DIGIT */
3803 RET0_UNLESS(parse_expect_char(date, len, ':')); /* expect ":" */
3804 RET0_UNLESS(parse_2digit(date, len, &tm->tm_min)); /* min 2DIGIT */
3805 RET0_UNLESS(parse_expect_char(date, len, ':')); /* expect ":" */
3806 RET0_UNLESS(parse_2digit(date, len, &tm->tm_sec)); /* sec 2DIGIT */
3807 return 1;
3808}
3809
3810/* From RFC7231
3811 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3812 *
3813 * IMF-fixdate = day-name "," SP date1 SP time-of-day SP GMT
3814 * ; fixed length/zone/capitalization subset of the format
3815 * ; see Section 3.3 of [RFC5322]
3816 *
3817 *
3818 * date1 = day SP month SP year
3819 * ; e.g., 02 Jun 1982
3820 *
3821 * day = 2DIGIT
3822 * year = 4DIGIT
3823 *
3824 * GMT = %x47.4D.54 ; "GMT", case-sensitive
3825 *
3826 * time-of-day = hour ":" minute ":" second
3827 * ; 00:00:00 - 23:59:60 (leap second)
3828 *
3829 * hour = 2DIGIT
3830 * minute = 2DIGIT
3831 * second = 2DIGIT
3832 *
3833 * DIGIT = decimal 0-9
3834 */
3835int parse_imf_date(const char *date, int len, struct tm *tm)
3836{
David Carlier327298c2016-11-20 10:42:38 +00003837 /* tm_gmtoff, if present, ought to be zero'ed */
3838 memset(tm, 0, sizeof(*tm));
3839
Thierry Fournier93127942016-01-20 18:49:45 +01003840 RET0_UNLESS(parse_http_dayname(&date, &len, tm)); /* day-name */
3841 RET0_UNLESS(parse_expect_char(&date, &len, ',')); /* expect "," */
3842 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3843 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
3844 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3845 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* Month */
3846 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3847 RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
3848 tm->tm_year -= 1900;
3849 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3850 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3851 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3852 RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3)); /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
3853 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003854 return 1;
3855}
3856
3857/* From RFC7231
3858 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3859 *
3860 * rfc850-date = day-name-l "," SP date2 SP time-of-day SP GMT
3861 * date2 = day "-" month "-" 2DIGIT
3862 * ; e.g., 02-Jun-82
3863 *
3864 * day = 2DIGIT
3865 */
3866int parse_rfc850_date(const char *date, int len, struct tm *tm)
3867{
3868 int year;
3869
David Carlier327298c2016-11-20 10:42:38 +00003870 /* tm_gmtoff, if present, ought to be zero'ed */
3871 memset(tm, 0, sizeof(*tm));
3872
Thierry Fournier93127942016-01-20 18:49:45 +01003873 RET0_UNLESS(parse_http_ldayname(&date, &len, tm)); /* Read the day name */
3874 RET0_UNLESS(parse_expect_char(&date, &len, ',')); /* expect "," */
3875 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3876 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday)); /* day 2DIGIT */
3877 RET0_UNLESS(parse_expect_char(&date, &len, '-')); /* expect "-" */
3878 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* Month */
3879 RET0_UNLESS(parse_expect_char(&date, &len, '-')); /* expect "-" */
3880
3881 /* year = 2DIGIT
3882 *
3883 * Recipients of a timestamp value in rfc850-(*date) format, which uses a
3884 * two-digit year, MUST interpret a timestamp that appears to be more
3885 * than 50 years in the future as representing the most recent year in
3886 * the past that had the same last two digits.
3887 */
3888 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_year));
3889
3890 /* expect SP */
3891 if (!parse_expect_char(&date, &len, ' ')) {
3892 /* Maybe we have the date with 4 digits. */
3893 RET0_UNLESS(parse_2digit(&date, &len, &year));
3894 tm->tm_year = (tm->tm_year * 100 + year) - 1900;
3895 /* expect SP */
3896 RET0_UNLESS(parse_expect_char(&date, &len, ' '));
3897 } else {
3898 /* I fix 60 as pivot: >60: +1900, <60: +2000. Note that the
3899 * tm_year is the number of year since 1900, so for +1900, we
3900 * do nothing, and for +2000, we add 100.
3901 */
3902 if (tm->tm_year <= 60)
3903 tm->tm_year += 100;
3904 }
3905
3906 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3907 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3908 RET0_UNLESS(parse_strcmp(&date, &len, "GMT", 3)); /* GMT = %x47.4D.54 ; "GMT", case-sensitive */
3909 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003910
3911 return 1;
3912}
3913
3914/* From RFC7231
3915 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3916 *
3917 * asctime-date = day-name SP date3 SP time-of-day SP year
3918 * date3 = month SP ( 2DIGIT / ( SP 1DIGIT ))
3919 * ; e.g., Jun 2
3920 *
3921 * HTTP-date is case sensitive. A sender MUST NOT generate additional
3922 * whitespace in an HTTP-date beyond that specifically included as SP in
3923 * the grammar.
3924 */
3925int parse_asctime_date(const char *date, int len, struct tm *tm)
3926{
David Carlier327298c2016-11-20 10:42:38 +00003927 /* tm_gmtoff, if present, ought to be zero'ed */
3928 memset(tm, 0, sizeof(*tm));
3929
Thierry Fournier93127942016-01-20 18:49:45 +01003930 RET0_UNLESS(parse_http_dayname(&date, &len, tm)); /* day-name */
3931 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3932 RET0_UNLESS(parse_http_monthname(&date, &len, tm)); /* expect month */
3933 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3934
3935 /* expect SP and 1DIGIT or 2DIGIT */
3936 if (parse_expect_char(&date, &len, ' '))
3937 RET0_UNLESS(parse_digit(&date, &len, &tm->tm_mday));
3938 else
3939 RET0_UNLESS(parse_2digit(&date, &len, &tm->tm_mday));
3940
3941 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3942 RET0_UNLESS(parse_http_time(&date, &len, tm)); /* Parse time. */
3943 RET0_UNLESS(parse_expect_char(&date, &len, ' ')); /* expect SP */
3944 RET0_UNLESS(parse_4digit(&date, &len, &tm->tm_year)); /* year = 4DIGIT */
3945 tm->tm_year -= 1900;
3946 tm->tm_isdst = -1;
Thierry Fournier93127942016-01-20 18:49:45 +01003947 return 1;
3948}
3949
3950/* From RFC7231
3951 * https://tools.ietf.org/html/rfc7231#section-7.1.1.1
3952 *
3953 * HTTP-date = IMF-fixdate / obs-date
3954 * obs-date = rfc850-date / asctime-date
3955 *
3956 * parses an HTTP date in the RFC format and is accepted
3957 * alternatives. <date> is the strinf containing the date,
3958 * len is the len of the string. <tm> is filled with the
3959 * parsed time. We must considers this time as GMT.
3960 */
3961int parse_http_date(const char *date, int len, struct tm *tm)
3962{
3963 if (parse_imf_date(date, len, tm))
3964 return 1;
3965
3966 if (parse_rfc850_date(date, len, tm))
3967 return 1;
3968
3969 if (parse_asctime_date(date, len, tm))
3970 return 1;
3971
3972 return 0;
3973}
3974
Willy Tarreau4deeb102021-01-29 10:47:52 +01003975/* print the time <ns> in a short form (exactly 7 chars) at the end of buffer
3976 * <out>. "-" is printed if the value is zero, "inf" if larger than 1000 years.
3977 * It returns the new buffer length, or 0 if it doesn't fit. The value will be
3978 * surrounded by <pfx> and <sfx> respectively if not NULL.
3979 */
3980int print_time_short(struct buffer *out, const char *pfx, uint64_t ns, const char *sfx)
3981{
3982 double val = ns; // 52 bits of mantissa keep ns accuracy over 52 days
3983 const char *unit;
3984
3985 if (!pfx)
3986 pfx = "";
3987 if (!sfx)
3988 sfx = "";
3989
3990 do {
3991 unit = " - "; if (val <= 0.0) break;
3992 unit = "ns"; if (val < 1000.0) break;
3993 unit = "us"; val /= 1000.0; if (val < 1000.0) break;
3994 unit = "ms"; val /= 1000.0; if (val < 1000.0) break;
3995 unit = "s "; val /= 1000.0; if (val < 60.0) break;
3996 unit = "m "; val /= 60.0; if (val < 60.0) break;
3997 unit = "h "; val /= 60.0; if (val < 24.0) break;
3998 unit = "d "; val /= 24.0; if (val < 365.0) break;
3999 unit = "yr"; val /= 365.0; if (val < 1000.0) break;
4000 unit = " inf "; val = 0.0; break;
4001 } while (0);
4002
4003 if (val <= 0.0)
4004 return chunk_appendf(out, "%s%7s%s", pfx, unit, sfx);
4005 else if (val < 10.0)
4006 return chunk_appendf(out, "%s%1.3f%s%s", pfx, val, unit, sfx);
4007 else if (val < 100.0)
4008 return chunk_appendf(out, "%s%2.2f%s%s", pfx, val, unit, sfx);
4009 else
4010 return chunk_appendf(out, "%s%3.1f%s%s", pfx, val, unit, sfx);
4011}
4012
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004013/* Dynamically allocates a string of the proper length to hold the formatted
4014 * output. NULL is returned on error. The caller is responsible for freeing the
4015 * memory area using free(). The resulting string is returned in <out> if the
4016 * pointer is not NULL. A previous version of <out> might be used to build the
4017 * new string, and it will be freed before returning if it is not NULL, which
4018 * makes it possible to build complex strings from iterative calls without
4019 * having to care about freeing intermediate values, as in the example below :
4020 *
4021 * memprintf(&err, "invalid argument: '%s'", arg);
4022 * ...
4023 * memprintf(&err, "parser said : <%s>\n", *err);
4024 * ...
4025 * free(*err);
4026 *
4027 * This means that <err> must be initialized to NULL before first invocation.
4028 * The return value also holds the allocated string, which eases error checking
4029 * and immediate consumption. If the output pointer is not used, NULL must be
Willy Tarreaueb6cead2012-09-20 19:43:14 +02004030 * passed instead and it will be ignored. The returned message will then also
4031 * be NULL so that the caller does not have to bother with freeing anything.
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004032 *
4033 * It is also convenient to use it without any free except the last one :
4034 * err = NULL;
4035 * if (!fct1(err)) report(*err);
4036 * if (!fct2(err)) report(*err);
4037 * if (!fct3(err)) report(*err);
4038 * free(*err);
Christopher Faulet93a518f2017-10-24 11:25:33 +02004039 *
4040 * memprintf relies on memvprintf. This last version can be called from any
4041 * function with variadic arguments.
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004042 */
Christopher Faulet93a518f2017-10-24 11:25:33 +02004043char *memvprintf(char **out, const char *format, va_list orig_args)
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004044{
4045 va_list args;
4046 char *ret = NULL;
4047 int allocated = 0;
4048 int needed = 0;
4049
Willy Tarreaueb6cead2012-09-20 19:43:14 +02004050 if (!out)
4051 return NULL;
4052
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004053 do {
Willy Tarreaue0609f52019-03-29 19:13:23 +01004054 char buf1;
4055
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004056 /* vsnprintf() will return the required length even when the
4057 * target buffer is NULL. We do this in a loop just in case
4058 * intermediate evaluations get wrong.
4059 */
Christopher Faulet93a518f2017-10-24 11:25:33 +02004060 va_copy(args, orig_args);
Willy Tarreaue0609f52019-03-29 19:13:23 +01004061 needed = vsnprintf(ret ? ret : &buf1, allocated, format, args);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004062 va_end(args);
Willy Tarreau1b2fed62013-04-01 22:48:54 +02004063 if (needed < allocated) {
4064 /* Note: on Solaris 8, the first iteration always
4065 * returns -1 if allocated is zero, so we force a
4066 * retry.
4067 */
4068 if (!allocated)
4069 needed = 0;
4070 else
4071 break;
4072 }
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004073
Willy Tarreau1b2fed62013-04-01 22:48:54 +02004074 allocated = needed + 1;
Hubert Verstraete831962e2016-06-28 22:44:26 +02004075 ret = my_realloc2(ret, allocated);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004076 } while (ret);
4077
4078 if (needed < 0) {
4079 /* an error was encountered */
Willy Tarreau61cfdf42021-02-20 10:46:51 +01004080 ha_free(&ret);
Willy Tarreau9a7bea52012-04-27 11:16:50 +02004081 }
4082
4083 if (out) {
4084 free(*out);
4085 *out = ret;
4086 }
4087
4088 return ret;
4089}
William Lallemand421f5b52012-02-06 18:15:57 +01004090
Christopher Faulet93a518f2017-10-24 11:25:33 +02004091char *memprintf(char **out, const char *format, ...)
4092{
4093 va_list args;
4094 char *ret = NULL;
4095
4096 va_start(args, format);
4097 ret = memvprintf(out, format, args);
4098 va_end(args);
4099
4100 return ret;
4101}
4102
Willy Tarreau21c705b2012-09-14 11:40:36 +02004103/* Used to add <level> spaces before each line of <out>, unless there is only one line.
4104 * The input argument is automatically freed and reassigned. The result will have to be
Willy Tarreau70eec382012-10-10 08:56:47 +02004105 * freed by the caller. It also supports being passed a NULL which results in the same
4106 * output.
Willy Tarreau21c705b2012-09-14 11:40:36 +02004107 * Example of use :
4108 * parse(cmd, &err); (callee: memprintf(&err, ...))
4109 * fprintf(stderr, "Parser said: %s\n", indent_error(&err));
4110 * free(err);
4111 */
4112char *indent_msg(char **out, int level)
4113{
4114 char *ret, *in, *p;
4115 int needed = 0;
4116 int lf = 0;
4117 int lastlf = 0;
4118 int len;
4119
Willy Tarreau70eec382012-10-10 08:56:47 +02004120 if (!out || !*out)
4121 return NULL;
4122
Willy Tarreau21c705b2012-09-14 11:40:36 +02004123 in = *out - 1;
4124 while ((in = strchr(in + 1, '\n')) != NULL) {
4125 lastlf = in - *out;
4126 lf++;
4127 }
4128
4129 if (!lf) /* single line, no LF, return it as-is */
4130 return *out;
4131
4132 len = strlen(*out);
4133
4134 if (lf == 1 && lastlf == len - 1) {
4135 /* single line, LF at end, strip it and return as-is */
4136 (*out)[lastlf] = 0;
4137 return *out;
4138 }
4139
4140 /* OK now we have at least one LF, we need to process the whole string
4141 * as a multi-line string. What we'll do :
4142 * - prefix with an LF if there is none
4143 * - add <level> spaces before each line
4144 * This means at most ( 1 + level + (len-lf) + lf*<1+level) ) =
4145 * 1 + level + len + lf * level = 1 + level * (lf + 1) + len.
4146 */
4147
4148 needed = 1 + level * (lf + 1) + len + 1;
4149 p = ret = malloc(needed);
4150 in = *out;
4151
4152 /* skip initial LFs */
4153 while (*in == '\n')
4154 in++;
4155
4156 /* copy each line, prefixed with LF and <level> spaces, and without the trailing LF */
4157 while (*in) {
4158 *p++ = '\n';
4159 memset(p, ' ', level);
4160 p += level;
4161 do {
4162 *p++ = *in++;
4163 } while (*in && *in != '\n');
4164 if (*in)
4165 in++;
4166 }
4167 *p = 0;
4168
4169 free(*out);
4170 *out = ret;
4171
4172 return ret;
4173}
4174
Willy Tarreaua2c99112019-08-21 13:17:37 +02004175/* makes a copy of message <in> into <out>, with each line prefixed with <pfx>
4176 * and end of lines replaced with <eol> if not 0. The first line to indent has
4177 * to be indicated in <first> (starts at zero), so that it is possible to skip
4178 * indenting the first line if it has to be appended after an existing message.
4179 * Empty strings are never indented, and NULL strings are considered empty both
4180 * for <in> and <pfx>. It returns non-zero if an EOL was appended as the last
4181 * character, non-zero otherwise.
4182 */
4183int append_prefixed_str(struct buffer *out, const char *in, const char *pfx, char eol, int first)
4184{
4185 int bol, lf;
4186 int pfxlen = pfx ? strlen(pfx) : 0;
4187
4188 if (!in)
4189 return 0;
4190
4191 bol = 1;
4192 lf = 0;
4193 while (*in) {
4194 if (bol && pfxlen) {
4195 if (first > 0)
4196 first--;
4197 else
4198 b_putblk(out, pfx, pfxlen);
4199 bol = 0;
4200 }
4201
4202 lf = (*in == '\n');
4203 bol |= lf;
4204 b_putchr(out, (lf && eol) ? eol : *in);
4205 in++;
4206 }
4207 return lf;
4208}
4209
Willy Tarreau9d22e562019-03-29 18:49:09 +01004210/* removes environment variable <name> from the environment as found in
4211 * environ. This is only provided as an alternative for systems without
4212 * unsetenv() (old Solaris and AIX versions). THIS IS NOT THREAD SAFE.
Ilya Shipitsin856aabc2020-04-16 23:51:34 +05004213 * The principle is to scan environ for each occurrence of variable name
Willy Tarreau9d22e562019-03-29 18:49:09 +01004214 * <name> and to replace the matching pointers with the last pointer of
4215 * the array (since variables are not ordered).
4216 * It always returns 0 (success).
4217 */
4218int my_unsetenv(const char *name)
4219{
4220 extern char **environ;
4221 char **p = environ;
4222 int vars;
4223 int next;
4224 int len;
4225
4226 len = strlen(name);
4227 for (vars = 0; p[vars]; vars++)
4228 ;
4229 next = 0;
4230 while (next < vars) {
4231 if (strncmp(p[next], name, len) != 0 || p[next][len] != '=') {
4232 next++;
4233 continue;
4234 }
4235 if (next < vars - 1)
4236 p[next] = p[vars - 1];
4237 p[--vars] = NULL;
4238 }
4239 return 0;
4240}
4241
Willy Tarreaudad36a32013-03-11 01:20:04 +01004242/* Convert occurrences of environment variables in the input string to their
4243 * corresponding value. A variable is identified as a series of alphanumeric
4244 * characters or underscores following a '$' sign. The <in> string must be
4245 * free()able. NULL returns NULL. The resulting string might be reallocated if
4246 * some expansion is made. Variable names may also be enclosed into braces if
4247 * needed (eg: to concatenate alphanum characters).
4248 */
4249char *env_expand(char *in)
4250{
4251 char *txt_beg;
4252 char *out;
4253 char *txt_end;
4254 char *var_beg;
4255 char *var_end;
4256 char *value;
4257 char *next;
4258 int out_len;
4259 int val_len;
4260
4261 if (!in)
4262 return in;
4263
4264 value = out = NULL;
4265 out_len = 0;
4266
4267 txt_beg = in;
4268 do {
4269 /* look for next '$' sign in <in> */
4270 for (txt_end = txt_beg; *txt_end && *txt_end != '$'; txt_end++);
4271
4272 if (!*txt_end && !out) /* end and no expansion performed */
4273 return in;
4274
4275 val_len = 0;
4276 next = txt_end;
4277 if (*txt_end == '$') {
4278 char save;
4279
4280 var_beg = txt_end + 1;
4281 if (*var_beg == '{')
4282 var_beg++;
4283
4284 var_end = var_beg;
Willy Tarreau90807112020-02-25 08:16:33 +01004285 while (isalnum((unsigned char)*var_end) || *var_end == '_') {
Willy Tarreaudad36a32013-03-11 01:20:04 +01004286 var_end++;
4287 }
4288
4289 next = var_end;
4290 if (*var_end == '}' && (var_beg > txt_end + 1))
4291 next++;
4292
4293 /* get value of the variable name at this location */
4294 save = *var_end;
4295 *var_end = '\0';
4296 value = getenv(var_beg);
4297 *var_end = save;
4298 val_len = value ? strlen(value) : 0;
4299 }
4300
Hubert Verstraete831962e2016-06-28 22:44:26 +02004301 out = my_realloc2(out, out_len + (txt_end - txt_beg) + val_len + 1);
Willy Tarreaudad36a32013-03-11 01:20:04 +01004302 if (txt_end > txt_beg) {
4303 memcpy(out + out_len, txt_beg, txt_end - txt_beg);
4304 out_len += txt_end - txt_beg;
4305 }
4306 if (val_len) {
4307 memcpy(out + out_len, value, val_len);
4308 out_len += val_len;
4309 }
4310 out[out_len] = 0;
4311 txt_beg = next;
4312 } while (*txt_beg);
4313
4314 /* here we know that <out> was allocated and that we don't need <in> anymore */
4315 free(in);
4316 return out;
4317}
4318
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004319
4320/* same as strstr() but case-insensitive and with limit length */
4321const char *strnistr(const char *str1, int len_str1, const char *str2, int len_str2)
4322{
4323 char *pptr, *sptr, *start;
Willy Tarreauc8746532014-05-28 23:05:07 +02004324 unsigned int slen, plen;
4325 unsigned int tmp1, tmp2;
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004326
4327 if (str1 == NULL || len_str1 == 0) // search pattern into an empty string => search is not found
4328 return NULL;
4329
4330 if (str2 == NULL || len_str2 == 0) // pattern is empty => every str1 match
4331 return str1;
4332
4333 if (len_str1 < len_str2) // pattern is longer than string => search is not found
4334 return NULL;
4335
4336 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 +02004337 while (toupper((unsigned char)*start) != toupper((unsigned char)*str2)) {
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004338 start++;
4339 slen--;
4340 tmp1++;
4341
4342 if (tmp1 >= len_str1)
4343 return NULL;
4344
4345 /* if pattern longer than string */
4346 if (slen < plen)
4347 return NULL;
4348 }
4349
4350 sptr = start;
4351 pptr = (char *)str2;
4352
4353 tmp2 = 0;
Willy Tarreauf278eec2020-07-05 21:46:32 +02004354 while (toupper((unsigned char)*sptr) == toupper((unsigned char)*pptr)) {
de Lafond Guillaume88c278f2013-04-15 19:27:10 +02004355 sptr++;
4356 pptr++;
4357 tmp2++;
4358
4359 if (*pptr == '\0' || tmp2 == len_str2) /* end of pattern found */
4360 return start;
4361 if (*sptr == '\0' || tmp2 == len_str1) /* end of string found and the pattern is not fully found */
4362 return NULL;
4363 }
4364 }
4365 return NULL;
4366}
4367
Willy Tarreau3ff476e2022-03-30 10:02:56 +02004368/* Returns true if s1 < s2 < s3 otherwise zero. Both s1 and s3 may be NULL and
4369 * in this case only non-null strings are compared. This allows to pass initial
4370 * values in iterators and in sort functions.
4371 */
4372int strordered(const char *s1, const char *s2, const char *s3)
4373{
4374 return (!s1 || strcmp(s1, s2) < 0) && (!s3 || strcmp(s2, s3) < 0);
4375}
4376
Thierry FOURNIER317e1c42014-08-12 10:20:47 +02004377/* This function read the next valid utf8 char.
4378 * <s> is the byte srray to be decode, <len> is its length.
4379 * The function returns decoded char encoded like this:
4380 * The 4 msb are the return code (UTF8_CODE_*), the 4 lsb
4381 * are the length read. The decoded character is stored in <c>.
4382 */
4383unsigned char utf8_next(const char *s, int len, unsigned int *c)
4384{
4385 const unsigned char *p = (unsigned char *)s;
4386 int dec;
4387 unsigned char code = UTF8_CODE_OK;
4388
4389 if (len < 1)
4390 return UTF8_CODE_OK;
4391
4392 /* Check the type of UTF8 sequence
4393 *
4394 * 0... .... 0x00 <= x <= 0x7f : 1 byte: ascii char
4395 * 10.. .... 0x80 <= x <= 0xbf : invalid sequence
4396 * 110. .... 0xc0 <= x <= 0xdf : 2 bytes
4397 * 1110 .... 0xe0 <= x <= 0xef : 3 bytes
4398 * 1111 0... 0xf0 <= x <= 0xf7 : 4 bytes
4399 * 1111 10.. 0xf8 <= x <= 0xfb : 5 bytes
4400 * 1111 110. 0xfc <= x <= 0xfd : 6 bytes
4401 * 1111 111. 0xfe <= x <= 0xff : invalid sequence
4402 */
4403 switch (*p) {
4404 case 0x00 ... 0x7f:
4405 *c = *p;
4406 return UTF8_CODE_OK | 1;
4407
4408 case 0x80 ... 0xbf:
4409 *c = *p;
4410 return UTF8_CODE_BADSEQ | 1;
4411
4412 case 0xc0 ... 0xdf:
4413 if (len < 2) {
4414 *c = *p;
4415 return UTF8_CODE_BADSEQ | 1;
4416 }
4417 *c = *p & 0x1f;
4418 dec = 1;
4419 break;
4420
4421 case 0xe0 ... 0xef:
4422 if (len < 3) {
4423 *c = *p;
4424 return UTF8_CODE_BADSEQ | 1;
4425 }
4426 *c = *p & 0x0f;
4427 dec = 2;
4428 break;
4429
4430 case 0xf0 ... 0xf7:
4431 if (len < 4) {
4432 *c = *p;
4433 return UTF8_CODE_BADSEQ | 1;
4434 }
4435 *c = *p & 0x07;
4436 dec = 3;
4437 break;
4438
4439 case 0xf8 ... 0xfb:
4440 if (len < 5) {
4441 *c = *p;
4442 return UTF8_CODE_BADSEQ | 1;
4443 }
4444 *c = *p & 0x03;
4445 dec = 4;
4446 break;
4447
4448 case 0xfc ... 0xfd:
4449 if (len < 6) {
4450 *c = *p;
4451 return UTF8_CODE_BADSEQ | 1;
4452 }
4453 *c = *p & 0x01;
4454 dec = 5;
4455 break;
4456
4457 case 0xfe ... 0xff:
4458 default:
4459 *c = *p;
4460 return UTF8_CODE_BADSEQ | 1;
4461 }
4462
4463 p++;
4464
4465 while (dec > 0) {
4466
4467 /* need 0x10 for the 2 first bits */
4468 if ( ( *p & 0xc0 ) != 0x80 )
4469 return UTF8_CODE_BADSEQ | ((p-(unsigned char *)s)&0xffff);
4470
4471 /* add data at char */
4472 *c = ( *c << 6 ) | ( *p & 0x3f );
4473
4474 dec--;
4475 p++;
4476 }
4477
4478 /* Check ovelong encoding.
4479 * 1 byte : 5 + 6 : 11 : 0x80 ... 0x7ff
4480 * 2 bytes : 4 + 6 + 6 : 16 : 0x800 ... 0xffff
4481 * 3 bytes : 3 + 6 + 6 + 6 : 21 : 0x10000 ... 0x1fffff
4482 */
Thierry FOURNIER9e7ec082015-03-12 19:32:38 +01004483 if (( *c <= 0x7f && (p-(unsigned char *)s) > 1) ||
Thierry FOURNIER317e1c42014-08-12 10:20:47 +02004484 (*c >= 0x80 && *c <= 0x7ff && (p-(unsigned char *)s) > 2) ||
4485 (*c >= 0x800 && *c <= 0xffff && (p-(unsigned char *)s) > 3) ||
4486 (*c >= 0x10000 && *c <= 0x1fffff && (p-(unsigned char *)s) > 4))
4487 code |= UTF8_CODE_OVERLONG;
4488
4489 /* Check invalid UTF8 range. */
4490 if ((*c >= 0xd800 && *c <= 0xdfff) ||
4491 (*c >= 0xfffe && *c <= 0xffff))
4492 code |= UTF8_CODE_INVRANGE;
4493
4494 return code | ((p-(unsigned char *)s)&0x0f);
4495}
4496
Maxime de Roucydc887852016-05-13 23:52:54 +02004497/* append a copy of string <str> (in a wordlist) at the end of the list <li>
4498 * On failure : return 0 and <err> filled with an error message.
4499 * The caller is responsible for freeing the <err> and <str> copy
4500 * memory area using free()
4501 */
4502int list_append_word(struct list *li, const char *str, char **err)
4503{
4504 struct wordlist *wl;
4505
4506 wl = calloc(1, sizeof(*wl));
4507 if (!wl) {
4508 memprintf(err, "out of memory");
4509 goto fail_wl;
4510 }
4511
4512 wl->s = strdup(str);
4513 if (!wl->s) {
4514 memprintf(err, "out of memory");
4515 goto fail_wl_s;
4516 }
4517
Willy Tarreau2b718102021-04-21 07:32:39 +02004518 LIST_APPEND(li, &wl->list);
Maxime de Roucydc887852016-05-13 23:52:54 +02004519
4520 return 1;
4521
4522fail_wl_s:
4523 free(wl->s);
4524fail_wl:
4525 free(wl);
4526 return 0;
4527}
4528
Willy Tarreau37101052019-05-20 16:48:20 +02004529/* indicates if a memory location may safely be read or not. The trick consists
4530 * in performing a harmless syscall using this location as an input and letting
4531 * the operating system report whether it's OK or not. For this we have the
4532 * stat() syscall, which will return EFAULT when the memory location supposed
4533 * to contain the file name is not readable. If it is readable it will then
4534 * either return 0 if the area contains an existing file name, or -1 with
4535 * another code. This must not be abused, and some audit systems might detect
4536 * this as abnormal activity. It's used only for unsafe dumps.
4537 */
4538int may_access(const void *ptr)
4539{
4540 struct stat buf;
4541
4542 if (stat(ptr, &buf) == 0)
4543 return 1;
4544 if (errno == EFAULT)
4545 return 0;
4546 return 1;
4547}
4548
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004549/* print a string of text buffer to <out>. The format is :
4550 * Non-printable chars \t, \n, \r and \e are * encoded in C format.
4551 * Other non-printable chars are encoded "\xHH". Space, '\', and '=' are also escaped.
4552 * Print stopped if null char or <bsize> is reached, or if no more place in the chunk.
4553 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004554int dump_text(struct buffer *out, const char *buf, int bsize)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004555{
4556 unsigned char c;
Tim Duesterhus18795d42021-08-29 00:58:22 +02004557 size_t ptr = 0;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004558
Tim Duesterhus18795d42021-08-29 00:58:22 +02004559 while (ptr < bsize && buf[ptr]) {
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004560 c = buf[ptr];
Willy Tarreau90807112020-02-25 08:16:33 +01004561 if (isprint((unsigned char)c) && isascii((unsigned char)c) && c != '\\' && c != ' ' && c != '=') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004562 if (out->data > out->size - 1)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004563 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004564 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004565 }
4566 else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\' || c == ' ' || c == '=') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004567 if (out->data > out->size - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004568 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004569 out->area[out->data++] = '\\';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004570 switch (c) {
4571 case ' ': c = ' '; break;
4572 case '\t': c = 't'; break;
4573 case '\n': c = 'n'; break;
4574 case '\r': c = 'r'; break;
4575 case '\e': c = 'e'; break;
4576 case '\\': c = '\\'; break;
4577 case '=': c = '='; break;
4578 }
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004579 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004580 }
4581 else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004582 if (out->data > out->size - 4)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004583 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004584 out->area[out->data++] = '\\';
4585 out->area[out->data++] = 'x';
4586 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4587 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004588 }
4589 ptr++;
4590 }
4591
4592 return ptr;
4593}
4594
4595/* print a buffer in hexa.
4596 * Print stopped if <bsize> is reached, or if no more place in the chunk.
4597 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004598int dump_binary(struct buffer *out, const char *buf, int bsize)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004599{
4600 unsigned char c;
4601 int ptr = 0;
4602
4603 while (ptr < bsize) {
4604 c = buf[ptr];
4605
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004606 if (out->data > out->size - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004607 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004608 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4609 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004610
4611 ptr++;
4612 }
4613 return ptr;
4614}
4615
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004616/* Appends into buffer <out> a hex dump of memory area <buf> for <len> bytes,
4617 * prepending each line with prefix <pfx>. The output is *not* initialized.
4618 * The output will not wrap pas the buffer's end so it is more optimal if the
4619 * caller makes sure the buffer is aligned first. A trailing zero will always
4620 * be appended (and not counted) if there is room for it. The caller must make
Willy Tarreau37101052019-05-20 16:48:20 +02004621 * sure that the area is dumpable first. If <unsafe> is non-null, the memory
4622 * locations are checked first for being readable.
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004623 */
Willy Tarreau37101052019-05-20 16:48:20 +02004624void dump_hex(struct buffer *out, const char *pfx, const void *buf, int len, int unsafe)
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004625{
4626 const unsigned char *d = buf;
4627 int i, j, start;
4628
4629 d = (const unsigned char *)(((unsigned long)buf) & -16);
4630 start = ((unsigned long)buf) & 15;
4631
4632 for (i = 0; i < start + len; i += 16) {
4633 chunk_appendf(out, (sizeof(void *) == 4) ? "%s%8p: " : "%s%16p: ", pfx, d + i);
4634
Willy Tarreau37101052019-05-20 16:48:20 +02004635 // 0: unchecked, 1: checked safe, 2: danger
4636 unsafe = !!unsafe;
4637 if (unsafe && !may_access(d + i))
4638 unsafe = 2;
4639
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004640 for (j = 0; j < 16; j++) {
Willy Tarreau37101052019-05-20 16:48:20 +02004641 if ((i + j < start) || (i + j >= start + len))
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004642 chunk_strcat(out, "'' ");
Willy Tarreau37101052019-05-20 16:48:20 +02004643 else if (unsafe > 1)
4644 chunk_strcat(out, "** ");
4645 else
4646 chunk_appendf(out, "%02x ", d[i + j]);
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004647
4648 if (j == 7)
4649 chunk_strcat(out, "- ");
4650 }
4651 chunk_strcat(out, " ");
4652 for (j = 0; j < 16; j++) {
Willy Tarreau37101052019-05-20 16:48:20 +02004653 if ((i + j < start) || (i + j >= start + len))
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004654 chunk_strcat(out, "'");
Willy Tarreau37101052019-05-20 16:48:20 +02004655 else if (unsafe > 1)
4656 chunk_strcat(out, "*");
Willy Tarreau90807112020-02-25 08:16:33 +01004657 else if (isprint((unsigned char)d[i + j]))
Willy Tarreau37101052019-05-20 16:48:20 +02004658 chunk_appendf(out, "%c", d[i + j]);
4659 else
4660 chunk_strcat(out, ".");
Willy Tarreau9fc5dcb2019-05-20 16:13:40 +02004661 }
4662 chunk_strcat(out, "\n");
4663 }
4664}
4665
Willy Tarreau762fb3e2020-03-03 15:57:10 +01004666/* dumps <pfx> followed by <n> bytes from <addr> in hex form into buffer <buf>
4667 * enclosed in brackets after the address itself, formatted on 14 chars
4668 * including the "0x" prefix. This is meant to be used as a prefix for code
4669 * areas. For example:
4670 * "0x7f10b6557690 [48 c7 c0 0f 00 00 00 0f]"
4671 * It relies on may_access() to know if the bytes are dumpable, otherwise "--"
4672 * is emitted. A NULL <pfx> will be considered empty.
4673 */
4674void dump_addr_and_bytes(struct buffer *buf, const char *pfx, const void *addr, int n)
4675{
4676 int ok = 0;
4677 int i;
4678
4679 chunk_appendf(buf, "%s%#14lx [", pfx ? pfx : "", (long)addr);
4680
4681 for (i = 0; i < n; i++) {
4682 if (i == 0 || (((long)(addr + i) ^ (long)(addr)) & 4096))
4683 ok = may_access(addr + i);
4684 if (ok)
4685 chunk_appendf(buf, "%02x%s", ((uint8_t*)addr)[i], (i<n-1) ? " " : "]");
4686 else
4687 chunk_appendf(buf, "--%s", (i<n-1) ? " " : "]");
4688 }
4689}
4690
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004691/* print a line of text buffer (limited to 70 bytes) to <out>. The format is :
4692 * <2 spaces> <offset=5 digits> <space or plus> <space> <70 chars max> <\n>
4693 * which is 60 chars per line. Non-printable chars \t, \n, \r and \e are
4694 * encoded in C format. Other non-printable chars are encoded "\xHH". Original
4695 * lines are respected within the limit of 70 output chars. Lines that are
4696 * continuation of a previous truncated line begin with "+" instead of " "
4697 * after the offset. The new pointer is returned.
4698 */
Willy Tarreau83061a82018-07-13 11:56:34 +02004699int dump_text_line(struct buffer *out, const char *buf, int bsize, int len,
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004700 int *line, int ptr)
4701{
4702 int end;
4703 unsigned char c;
4704
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004705 end = out->data + 80;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004706 if (end > out->size)
4707 return ptr;
4708
4709 chunk_appendf(out, " %05d%c ", ptr, (ptr == *line) ? ' ' : '+');
4710
4711 while (ptr < len && ptr < bsize) {
4712 c = buf[ptr];
Willy Tarreau90807112020-02-25 08:16:33 +01004713 if (isprint((unsigned char)c) && isascii((unsigned char)c) && c != '\\') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004714 if (out->data > end - 2)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004715 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004716 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004717 } else if (c == '\t' || c == '\n' || c == '\r' || c == '\e' || c == '\\') {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004718 if (out->data > end - 3)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004719 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004720 out->area[out->data++] = '\\';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004721 switch (c) {
4722 case '\t': c = 't'; break;
4723 case '\n': c = 'n'; break;
4724 case '\r': c = 'r'; break;
4725 case '\e': c = 'e'; break;
4726 case '\\': c = '\\'; break;
4727 }
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004728 out->area[out->data++] = c;
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004729 } else {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004730 if (out->data > end - 5)
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004731 break;
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004732 out->area[out->data++] = '\\';
4733 out->area[out->data++] = 'x';
4734 out->area[out->data++] = hextab[(c >> 4) & 0xF];
4735 out->area[out->data++] = hextab[c & 0xF];
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004736 }
4737 if (buf[ptr++] == '\n') {
4738 /* we had a line break, let's return now */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004739 out->area[out->data++] = '\n';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004740 *line = ptr;
4741 return ptr;
4742 }
4743 }
4744 /* we have an incomplete line, we return it as-is */
Willy Tarreau843b7cb2018-07-13 10:54:26 +02004745 out->area[out->data++] = '\n';
Willy Tarreau97c2ae12016-11-22 18:00:20 +01004746 return ptr;
4747}
4748
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004749/* displays a <len> long memory block at <buf>, assuming first byte of <buf>
Willy Tarreaued936c52017-04-27 18:03:20 +02004750 * has address <baseaddr>. String <pfx> may be placed as a prefix in front of
4751 * each line. It may be NULL if unused. The output is emitted to file <out>.
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004752 */
Willy Tarreaued936c52017-04-27 18:03:20 +02004753void debug_hexdump(FILE *out, const char *pfx, const char *buf,
4754 unsigned int baseaddr, int len)
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004755{
Willy Tarreau73459792017-04-11 07:58:08 +02004756 unsigned int i;
4757 int b, j;
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004758
4759 for (i = 0; i < (len + (baseaddr & 15)); i += 16) {
4760 b = i - (baseaddr & 15);
Willy Tarreaued936c52017-04-27 18:03:20 +02004761 fprintf(out, "%s%08x: ", pfx ? pfx : "", i + (baseaddr & ~15));
Willy Tarreau0ebb5112016-12-05 00:10:57 +01004762 for (j = 0; j < 8; j++) {
4763 if (b + j >= 0 && b + j < len)
4764 fprintf(out, "%02x ", (unsigned char)buf[b + j]);
4765 else
4766 fprintf(out, " ");
4767 }
4768
4769 if (b + j >= 0 && b + j < len)
4770 fputc('-', out);
4771 else
4772 fputc(' ', out);
4773
4774 for (j = 8; j < 16; j++) {
4775 if (b + j >= 0 && b + j < len)
4776 fprintf(out, " %02x", (unsigned char)buf[b + j]);
4777 else
4778 fprintf(out, " ");
4779 }
4780
4781 fprintf(out, " ");
4782 for (j = 0; j < 16; j++) {
4783 if (b + j >= 0 && b + j < len) {
4784 if (isprint((unsigned char)buf[b + j]))
4785 fputc((unsigned char)buf[b + j], out);
4786 else
4787 fputc('.', out);
4788 }
4789 else
4790 fputc(' ', out);
4791 }
4792 fputc('\n', out);
4793 }
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004794}
4795
Willy Tarreaubb869862020-04-16 10:52:41 +02004796/* Tries to report the executable path name on platforms supporting this. If
4797 * not found or not possible, returns NULL.
4798 */
4799const char *get_exec_path()
4800{
4801 const char *ret = NULL;
4802
David Carlier43a56852022-03-04 15:50:48 +00004803#if defined(__linux__) && defined(__GLIBC__) && (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 16))
Willy Tarreaubb869862020-04-16 10:52:41 +02004804 long execfn = getauxval(AT_EXECFN);
4805
4806 if (execfn && execfn != ENOENT)
4807 ret = (const char *)execfn;
devnexen@gmail.comc4e52322021-08-17 12:55:49 +01004808#elif defined(__FreeBSD__)
4809 Elf_Auxinfo *auxv;
4810 for (auxv = __elf_aux_vector; auxv->a_type != AT_NULL; ++auxv) {
4811 if (auxv->a_type == AT_EXECPATH) {
4812 ret = (const char *)auxv->a_un.a_ptr;
4813 break;
4814 }
4815 }
David Carlierbd2cced2021-08-17 08:44:25 +01004816#elif defined(__NetBSD__)
4817 AuxInfo *auxv;
4818 for (auxv = _dlauxinfo(); auxv->a_type != AT_NULL; ++auxv) {
4819 if (auxv->a_type == AT_SUN_EXECNAME) {
4820 ret = (const char *)auxv->a_v;
4821 break;
4822 }
4823 }
Willy Tarreaubb869862020-04-16 10:52:41 +02004824#endif
4825 return ret;
4826}
4827
Baruch Siache1651b22020-07-24 07:52:20 +03004828#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreau9133e482020-03-04 10:19:36 +01004829/* calls dladdr() or dladdr1() on <addr> and <dli>. If dladdr1 is available,
4830 * also returns the symbol size in <size>, otherwise returns 0 there.
4831 */
4832static int dladdr_and_size(const void *addr, Dl_info *dli, size_t *size)
4833{
4834 int ret;
Willy Tarreau7b2108c2021-08-30 10:15:35 +02004835#if defined(__GLIBC__) && (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 3)) // most detailed one
Willy Tarreauf3d5c4b2022-01-28 09:42:29 +01004836 const ElfW(Sym) *sym __attribute__((may_alias));
Willy Tarreau9133e482020-03-04 10:19:36 +01004837
4838 ret = dladdr1(addr, dli, (void **)&sym, RTLD_DL_SYMENT);
4839 if (ret)
4840 *size = sym ? sym->st_size : 0;
4841#else
David Carlierae5c42f2021-12-31 08:15:29 +00004842#if defined(__sun)
4843 ret = dladdr((void *)addr, dli);
4844#else
Willy Tarreau9133e482020-03-04 10:19:36 +01004845 ret = dladdr(addr, dli);
David Carlierae5c42f2021-12-31 08:15:29 +00004846#endif
Willy Tarreau9133e482020-03-04 10:19:36 +01004847 *size = 0;
4848#endif
4849 return ret;
4850}
Willy Tarreau64192392021-05-05 09:06:21 +02004851
4852/* Tries to retrieve the address of the first occurrence symbol <name>.
4853 * Note that NULL in return is not always an error as a symbol may have that
4854 * address in special situations.
4855 */
4856void *get_sym_curr_addr(const char *name)
4857{
4858 void *ptr = NULL;
4859
4860#ifdef RTLD_DEFAULT
4861 ptr = dlsym(RTLD_DEFAULT, name);
4862#endif
4863 return ptr;
4864}
4865
4866
4867/* Tries to retrieve the address of the next occurrence of symbol <name>
4868 * Note that NULL in return is not always an error as a symbol may have that
4869 * address in special situations.
4870 */
4871void *get_sym_next_addr(const char *name)
4872{
4873 void *ptr = NULL;
4874
4875#ifdef RTLD_NEXT
4876 ptr = dlsym(RTLD_NEXT, name);
Willy Tarreau9133e482020-03-04 10:19:36 +01004877#endif
Willy Tarreau64192392021-05-05 09:06:21 +02004878 return ptr;
4879}
4880
4881#else /* elf & linux & dl */
4882
4883/* no possible resolving on other platforms at the moment */
4884void *get_sym_curr_addr(const char *name)
4885{
4886 return NULL;
4887}
4888
4889void *get_sym_next_addr(const char *name)
4890{
4891 return NULL;
4892}
4893
4894#endif /* elf & linux & dl */
Willy Tarreau9133e482020-03-04 10:19:36 +01004895
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004896/* Tries to append to buffer <buf> some indications about the symbol at address
4897 * <addr> using the following form:
4898 * lib:+0xoffset (unresolvable address from lib's base)
4899 * main+0xoffset (unresolvable address from main (+/-))
4900 * lib:main+0xoffset (unresolvable lib address from main (+/-))
4901 * name (resolved exact exec address)
4902 * lib:name (resolved exact lib address)
4903 * name+0xoffset/0xsize (resolved address within exec symbol)
4904 * lib:name+0xoffset/0xsize (resolved address within lib symbol)
4905 *
4906 * The file name (lib or executable) is limited to what lies between the last
4907 * '/' and the first following '.'. An optional prefix <pfx> is prepended before
4908 * 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 +03004909 * that contains the "main" symbol, or when __ELF__ && USE_DL are not set.
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004910 *
4911 * The symbol's base address is returned, or NULL when unresolved, in order to
4912 * allow the caller to match it against known ones.
4913 */
Willy Tarreau45fd1032021-01-20 14:37:59 +01004914const void *resolve_sym_name(struct buffer *buf, const char *pfx, const void *addr)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004915{
4916 const struct {
4917 const void *func;
4918 const char *name;
4919 } fcts[] = {
4920 { .func = process_stream, .name = "process_stream" },
4921 { .func = task_run_applet, .name = "task_run_applet" },
Christopher Faulet4a7764a2022-04-01 16:58:52 +02004922 { .func = cs_conn_io_cb, .name = "cs_conn_io_cb" },
Willy Tarreau586f71b2020-12-11 15:54:36 +01004923 { .func = sock_conn_iocb, .name = "sock_conn_iocb" },
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004924 { .func = dgram_fd_handler, .name = "dgram_fd_handler" },
4925 { .func = listener_accept, .name = "listener_accept" },
Willy Tarreaud597ec22021-01-29 14:29:06 +01004926 { .func = manage_global_listener_queue, .name = "manage_global_listener_queue" },
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004927 { .func = poller_pipe_io_handler, .name = "poller_pipe_io_handler" },
4928 { .func = mworker_accept_wrapper, .name = "mworker_accept_wrapper" },
Willy Tarreau02922e12021-01-29 12:27:57 +01004929 { .func = session_expire_embryonic, .name = "session_expire_embryonic" },
Willy Tarreaufb5401f2021-01-29 12:25:23 +01004930#ifdef USE_THREAD
4931 { .func = accept_queue_process, .name = "accept_queue_process" },
4932#endif
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004933#ifdef USE_LUA
4934 { .func = hlua_process_task, .name = "hlua_process_task" },
4935#endif
Ilya Shipitsinbdec3ba2020-11-14 01:56:34 +05004936#ifdef SSL_MODE_ASYNC
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004937 { .func = ssl_async_fd_free, .name = "ssl_async_fd_free" },
4938 { .func = ssl_async_fd_handler, .name = "ssl_async_fd_handler" },
4939#endif
4940 };
4941
Baruch Siache1651b22020-07-24 07:52:20 +03004942#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004943 Dl_info dli, dli_main;
Willy Tarreau9133e482020-03-04 10:19:36 +01004944 size_t size;
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004945 const char *fname, *p;
4946#endif
4947 int i;
4948
4949 if (pfx)
4950 chunk_appendf(buf, "%s", pfx);
4951
4952 for (i = 0; i < sizeof(fcts) / sizeof(fcts[0]); i++) {
4953 if (addr == fcts[i].func) {
4954 chunk_appendf(buf, "%s", fcts[i].name);
4955 return addr;
4956 }
4957 }
4958
Baruch Siache1651b22020-07-24 07:52:20 +03004959#if (defined(__ELF__) && !defined(__linux__)) || defined(USE_DL)
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004960 /* Now let's try to be smarter */
Willy Tarreau9133e482020-03-04 10:19:36 +01004961 if (!dladdr_and_size(addr, &dli, &size))
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004962 goto unknown;
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004963
4964 /* 1. prefix the library name if it's not the same object as the one
4965 * that contains the main function. The name is picked between last '/'
4966 * and first following '.'.
4967 */
4968 if (!dladdr(main, &dli_main))
4969 dli_main.dli_fbase = NULL;
4970
4971 if (dli_main.dli_fbase != dli.dli_fbase) {
4972 fname = dli.dli_fname;
4973 p = strrchr(fname, '/');
4974 if (p++)
4975 fname = p;
4976 p = strchr(fname, '.');
4977 if (!p)
4978 p = fname + strlen(fname);
4979
4980 chunk_appendf(buf, "%.*s:", (int)(long)(p - fname), fname);
4981 }
4982
4983 /* 2. symbol name */
4984 if (dli.dli_sname) {
4985 /* known, dump it and return symbol's address (exact or relative) */
4986 chunk_appendf(buf, "%s", dli.dli_sname);
4987 if (addr != dli.dli_saddr) {
4988 chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_saddr));
Willy Tarreau9133e482020-03-04 10:19:36 +01004989 if (size)
4990 chunk_appendf(buf, "/%#lx", (long)size);
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01004991 }
4992 return dli.dli_saddr;
4993 }
4994 else if (dli_main.dli_fbase != dli.dli_fbase) {
4995 /* unresolved symbol from a known library, report relative offset */
4996 chunk_appendf(buf, "+%#lx", (long)(addr - dli.dli_fbase));
4997 return NULL;
4998 }
Baruch Siache1651b22020-07-24 07:52:20 +03004999#endif /* __ELF__ && !__linux__ || USE_DL */
Willy Tarreaueb8b1ca2020-03-03 17:09:08 +01005000 unknown:
5001 /* unresolved symbol from the main file, report relative offset to main */
5002 if ((void*)addr < (void*)main)
5003 chunk_appendf(buf, "main-%#lx", (long)((void*)main - addr));
5004 else
5005 chunk_appendf(buf, "main+%#lx", (long)(addr - (void*)main));
5006 return NULL;
Willy Tarreau0ebb5112016-12-05 00:10:57 +01005007}
5008
Willy Tarreau6ab7b212021-12-28 09:57:10 +01005009/* On systems where this is supported, let's provide a possibility to enumerate
5010 * the list of object files. The output is appended to a buffer initialized by
5011 * the caller, with one name per line. A trailing zero is always emitted if data
5012 * are written. Only real objects are dumped (executable and .so libs). The
5013 * function returns non-zero if it dumps anything. These functions do not make
5014 * use of the trash so that it is possible for the caller to call them with the
5015 * trash on input. The output format may be platform-specific but at least one
5016 * version must emit raw object file names when argument is zero.
5017 */
5018#if defined(HA_HAVE_DUMP_LIBS)
5019# if defined(HA_HAVE_DL_ITERATE_PHDR)
5020/* the private <data> we pass below is a dump context initialized like this */
5021struct dl_dump_ctx {
5022 struct buffer *buf;
5023 int with_addr;
5024};
5025
5026static int dl_dump_libs_cb(struct dl_phdr_info *info, size_t size, void *data)
5027{
5028 struct dl_dump_ctx *ctx = data;
5029 const char *fname;
5030 size_t p1, p2, beg, end;
5031 int idx;
5032
5033 if (!info || !info->dlpi_name)
5034 goto leave;
5035
5036 if (!*info->dlpi_name)
5037 fname = get_exec_path();
5038 else if (strchr(info->dlpi_name, '/'))
5039 fname = info->dlpi_name;
5040 else
5041 /* else it's a VDSO or similar and we're not interested */
5042 goto leave;
5043
5044 if (!ctx->with_addr)
5045 goto dump_name;
5046
5047 /* virtual addresses are relative to the load address and are per
5048 * pseudo-header, so we have to scan them all to find the furthest
5049 * one from the beginning. In this case we only dump entries if
5050 * they have at least one section.
5051 */
5052 beg = ~0; end = 0;
5053 for (idx = 0; idx < info->dlpi_phnum; idx++) {
5054 if (!info->dlpi_phdr[idx].p_memsz)
5055 continue;
5056 p1 = info->dlpi_phdr[idx].p_vaddr;
5057 if (p1 < beg)
5058 beg = p1;
5059 p2 = p1 + info->dlpi_phdr[idx].p_memsz - 1;
5060 if (p2 > end)
5061 end = p2;
5062 }
5063
5064 if (!idx)
5065 goto leave;
5066
5067 chunk_appendf(ctx->buf, "0x%012llx-0x%012llx (0x%07llx) ",
5068 (ullong)info->dlpi_addr + beg,
5069 (ullong)info->dlpi_addr + end,
5070 (ullong)(end - beg + 1));
5071 dump_name:
5072 chunk_appendf(ctx->buf, "%s\n", fname);
5073 leave:
5074 return 0;
5075}
5076
5077/* dumps lib names and optionally address ranges */
5078int dump_libs(struct buffer *output, int with_addr)
5079{
5080 struct dl_dump_ctx ctx = { .buf = output, .with_addr = with_addr };
5081 size_t old_data = output->data;
5082
5083 dl_iterate_phdr(dl_dump_libs_cb, &ctx);
5084 return output->data != old_data;
5085}
5086# else // no DL_ITERATE_PHDR
5087# error "No dump_libs() function for this platform"
5088# endif
5089#else // no HA_HAVE_DUMP_LIBS
5090
5091/* unsupported platform: do not dump anything */
5092int dump_libs(struct buffer *output, int with_addr)
5093{
5094 return 0;
5095}
5096
5097#endif // HA_HAVE_DUMP_LIBS
5098
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005099/*
5100 * Allocate an array of unsigned int with <nums> as address from <str> string
Ilya Shipitsin46a030c2020-07-05 16:36:08 +05005101 * made of integer separated by dot characters.
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005102 *
5103 * First, initializes the value with <sz> as address to 0 and initializes the
5104 * array with <nums> as address to NULL. Then allocates the array with <nums> as
5105 * address updating <sz> pointed value to the size of this array.
5106 *
5107 * Returns 1 if succeeded, 0 if not.
5108 */
5109int parse_dotted_uints(const char *str, unsigned int **nums, size_t *sz)
5110{
5111 unsigned int *n;
5112 const char *s, *end;
5113
5114 s = str;
5115 *sz = 0;
5116 end = str + strlen(str);
5117 *nums = n = NULL;
5118
5119 while (1) {
5120 unsigned int r;
5121
5122 if (s >= end)
5123 break;
5124
5125 r = read_uint(&s, end);
5126 /* Expected characters after having read an uint: '\0' or '.',
5127 * if '.', must not be terminal.
5128 */
Christopher Faulet4b524122021-02-11 10:42:41 +01005129 if (*s != '\0'&& (*s++ != '.' || s == end)) {
5130 free(n);
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005131 return 0;
Christopher Faulet4b524122021-02-11 10:42:41 +01005132 }
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005133
Frédéric Lécaille12a71842019-02-26 18:19:48 +01005134 n = my_realloc2(n, (*sz + 1) * sizeof *n);
Frédéric Lécaille3b717162019-02-25 15:04:22 +01005135 if (!n)
5136 return 0;
5137
5138 n[(*sz)++] = r;
5139 }
5140 *nums = n;
5141
5142 return 1;
5143}
5144
Willy Tarreau4d589e72019-08-23 19:02:26 +02005145
5146/* returns the number of bytes needed to encode <v> as a varint. An inline
5147 * version exists for use with constants (__varint_bytes()).
5148 */
5149int varint_bytes(uint64_t v)
5150{
5151 int len = 1;
5152
5153 if (v >= 240) {
5154 v = (v - 240) >> 4;
5155 while (1) {
5156 len++;
5157 if (v < 128)
5158 break;
5159 v = (v - 128) >> 7;
5160 }
5161 }
5162 return len;
5163}
5164
Willy Tarreau52bf8392020-03-08 00:42:37 +01005165
5166/* Random number generator state, see below */
Willy Tarreau1544c142020-03-12 00:31:18 +01005167static uint64_t ha_random_state[2] ALIGNED(2*sizeof(uint64_t));
Willy Tarreau52bf8392020-03-08 00:42:37 +01005168
5169/* This is a thread-safe implementation of xoroshiro128** described below:
5170 * http://prng.di.unimi.it/
5171 * It features a 2^128 long sequence, returns 64 high-quality bits on each call,
5172 * supports fast jumps and passes all common quality tests. It is thread-safe,
5173 * uses a double-cas on 64-bit architectures supporting it, and falls back to a
5174 * local lock on other ones.
5175 */
5176uint64_t ha_random64()
5177{
Willy Tarreau1544c142020-03-12 00:31:18 +01005178 uint64_t old[2] ALIGNED(2*sizeof(uint64_t));
5179 uint64_t new[2] ALIGNED(2*sizeof(uint64_t));
Willy Tarreau52bf8392020-03-08 00:42:37 +01005180
5181#if defined(USE_THREAD) && (!defined(HA_CAS_IS_8B) || !defined(HA_HAVE_CAS_DW))
5182 static HA_SPINLOCK_T rand_lock;
5183
5184 HA_SPIN_LOCK(OTHER_LOCK, &rand_lock);
5185#endif
5186
5187 old[0] = ha_random_state[0];
5188 old[1] = ha_random_state[1];
5189
5190#if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
5191 do {
5192#endif
Willy Tarreau52bf8392020-03-08 00:42:37 +01005193 new[1] = old[0] ^ old[1];
5194 new[0] = rotl64(old[0], 24) ^ new[1] ^ (new[1] << 16); // a, b
5195 new[1] = rotl64(new[1], 37); // c
5196
5197#if defined(USE_THREAD) && defined(HA_CAS_IS_8B) && defined(HA_HAVE_CAS_DW)
5198 } while (unlikely(!_HA_ATOMIC_DWCAS(ha_random_state, old, new)));
5199#else
5200 ha_random_state[0] = new[0];
5201 ha_random_state[1] = new[1];
5202#if defined(USE_THREAD)
5203 HA_SPIN_UNLOCK(OTHER_LOCK, &rand_lock);
5204#endif
5205#endif
Willy Tarreaub2475a12021-05-09 10:26:14 +02005206 return rotl64(old[0] * 5, 7) * 9;
Willy Tarreau52bf8392020-03-08 00:42:37 +01005207}
5208
5209/* seeds the random state using up to <len> bytes from <seed>, starting with
5210 * the first non-zero byte.
5211 */
5212void ha_random_seed(const unsigned char *seed, size_t len)
5213{
5214 size_t pos;
5215
5216 /* the seed must not be all zeroes, so we pre-fill it with alternating
5217 * bits and overwrite part of them with the block starting at the first
5218 * non-zero byte from the seed.
5219 */
5220 memset(ha_random_state, 0x55, sizeof(ha_random_state));
5221
5222 for (pos = 0; pos < len; pos++)
5223 if (seed[pos] != 0)
5224 break;
5225
5226 if (pos == len)
5227 return;
5228
5229 seed += pos;
5230 len -= pos;
5231
5232 if (len > sizeof(ha_random_state))
5233 len = sizeof(ha_random_state);
5234
5235 memcpy(ha_random_state, seed, len);
5236}
5237
5238/* This causes a jump to (dist * 2^96) places in the pseudo-random sequence,
5239 * and is equivalent to calling ha_random64() as many times. It is used to
5240 * provide non-overlapping sequences of 2^96 numbers (~7*10^28) to up to 2^32
5241 * different generators (i.e. different processes after a fork). The <dist>
5242 * argument is the distance to jump to and is used in a loop so it rather not
5243 * be too large if the processing time is a concern.
5244 *
5245 * BEWARE: this function is NOT thread-safe and must not be called during
5246 * concurrent accesses to ha_random64().
5247 */
5248void ha_random_jump96(uint32_t dist)
5249{
5250 while (dist--) {
5251 uint64_t s0 = 0;
5252 uint64_t s1 = 0;
5253 int b;
5254
5255 for (b = 0; b < 64; b++) {
5256 if ((0xd2a98b26625eee7bULL >> b) & 1) {
5257 s0 ^= ha_random_state[0];
5258 s1 ^= ha_random_state[1];
5259 }
5260 ha_random64();
5261 }
5262
5263 for (b = 0; b < 64; b++) {
5264 if ((0xdddf9b1090aa7ac1ULL >> b) & 1) {
5265 s0 ^= ha_random_state[0];
5266 s1 ^= ha_random_state[1];
5267 }
5268 ha_random64();
5269 }
5270 ha_random_state[0] = s0;
5271 ha_random_state[1] = s1;
5272 }
5273}
5274
Willy Tarreauee3bcdd2020-03-08 17:48:17 +01005275/* Generates an RFC4122 UUID into chunk <output> which must be at least 37
5276 * bytes large.
5277 */
5278void ha_generate_uuid(struct buffer *output)
5279{
5280 uint32_t rnd[4];
5281 uint64_t last;
5282
5283 last = ha_random64();
5284 rnd[0] = last;
5285 rnd[1] = last >> 32;
5286
5287 last = ha_random64();
5288 rnd[2] = last;
5289 rnd[3] = last >> 32;
5290
5291 chunk_printf(output, "%8.8x-%4.4x-%4.4x-%4.4x-%12.12llx",
5292 rnd[0],
5293 rnd[1] & 0xFFFF,
5294 ((rnd[1] >> 16u) & 0xFFF) | 0x4000, // highest 4 bits indicate the uuid version
5295 (rnd[2] & 0x3FFF) | 0x8000, // the highest 2 bits indicate the UUID variant (10),
5296 (long long)((rnd[2] >> 14u) | ((uint64_t) rnd[3] << 18u)) & 0xFFFFFFFFFFFFull);
5297}
5298
5299
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005300/* only used by parse_line() below. It supports writing in place provided that
5301 * <in> is updated to the next location before calling it. In that case, the
5302 * char at <in> may be overwritten.
5303 */
5304#define EMIT_CHAR(x) \
5305 do { \
5306 char __c = (char)(x); \
5307 if ((opts & PARSE_OPT_INPLACE) && out+outpos > in) \
5308 err |= PARSE_ERR_OVERLAP; \
5309 if (outpos >= outmax) \
5310 err |= PARSE_ERR_TOOLARGE; \
5311 if (!err) \
5312 out[outpos] = __c; \
5313 outpos++; \
5314 } while (0)
5315
Ilya Shipitsin46a030c2020-07-05 16:36:08 +05005316/* Parse <in>, copy it into <out> split into isolated words whose pointers
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005317 * are put in <args>. If more than <outlen> bytes have to be emitted, the
5318 * extraneous ones are not emitted but <outlen> is updated so that the caller
5319 * knows how much to realloc. Similarly, <args> are not updated beyond <nbargs>
5320 * but the returned <nbargs> indicates how many were found. All trailing args
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005321 * up to <nbargs> point to the trailing zero, and as long as <nbargs> is > 0,
5322 * it is guaranteed that at least one arg will point to the zero. It is safe
5323 * to call it with a NULL <args> if <nbargs> is 0.
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005324 *
5325 * <out> may overlap with <in> provided that it never goes further, in which
5326 * case the parser will accept to perform in-place parsing and unquoting/
5327 * unescaping but only if environment variables do not lead to expansion that
5328 * causes overlapping, otherwise the input string being destroyed, the error
5329 * will not be recoverable. Note that even during out-of-place <in> will
5330 * experience temporary modifications in-place for variable resolution and must
5331 * be writable, and will also receive zeroes to delimit words when using
5332 * in-place copy. Parsing options <opts> taken from PARSE_OPT_*. Return value
5333 * is zero on success otherwise a bitwise-or of PARSE_ERR_*. Upon error, the
5334 * starting point of the first invalid character sequence or unmatched
5335 * quote/brace is reported in <errptr> if not NULL. When using in-place parsing
5336 * error reporting might be difficult since zeroes will have been inserted into
5337 * the string. One solution for the caller may consist in replacing all args
5338 * delimiters with spaces in this case.
5339 */
Maximilian Mader29c6cd72021-06-06 00:50:21 +02005340uint32_t parse_line(char *in, char *out, size_t *outlen, char **args, int *nbargs, uint32_t opts, const char **errptr)
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005341{
5342 char *quote = NULL;
5343 char *brace = NULL;
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005344 char *word_expand = NULL;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005345 unsigned char hex1, hex2;
5346 size_t outmax = *outlen;
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005347 int argsmax = *nbargs - 1;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005348 size_t outpos = 0;
5349 int squote = 0;
5350 int dquote = 0;
5351 int arg = 0;
5352 uint32_t err = 0;
5353
5354 *nbargs = 0;
5355 *outlen = 0;
5356
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005357 /* argsmax may be -1 here, protecting args[] from any write */
5358 if (arg < argsmax)
5359 args[arg] = out;
5360
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005361 while (1) {
5362 if (*in >= '-' && *in != '\\') {
5363 /* speedup: directly send all regular chars starting
5364 * with '-', '.', '/', alnum etc...
5365 */
5366 EMIT_CHAR(*in++);
5367 continue;
5368 }
5369 else if (*in == '\0' || *in == '\n' || *in == '\r') {
5370 /* end of line */
5371 break;
5372 }
5373 else if (*in == '#' && (opts & PARSE_OPT_SHARP) && !squote && !dquote) {
5374 /* comment */
5375 break;
5376 }
5377 else if (*in == '"' && !squote && (opts & PARSE_OPT_DQUOTE)) { /* double quote outside single quotes */
5378 if (dquote) {
5379 dquote = 0;
5380 quote = NULL;
5381 }
5382 else {
5383 dquote = 1;
5384 quote = in;
5385 }
5386 in++;
5387 continue;
5388 }
5389 else if (*in == '\'' && !dquote && (opts & PARSE_OPT_SQUOTE)) { /* single quote outside double quotes */
5390 if (squote) {
5391 squote = 0;
5392 quote = NULL;
5393 }
5394 else {
5395 squote = 1;
5396 quote = in;
5397 }
5398 in++;
5399 continue;
5400 }
5401 else if (*in == '\\' && !squote && (opts & PARSE_OPT_BKSLASH)) {
5402 /* first, we'll replace \\, \<space>, \#, \r, \n, \t, \xXX with their
5403 * C equivalent value but only when they have a special meaning and within
5404 * double quotes for some of them. Other combinations left unchanged (eg: \1).
5405 */
5406 char tosend = *in;
5407
5408 switch (in[1]) {
5409 case ' ':
5410 case '\\':
5411 tosend = in[1];
5412 in++;
5413 break;
5414
5415 case 't':
5416 tosend = '\t';
5417 in++;
5418 break;
5419
5420 case 'n':
5421 tosend = '\n';
5422 in++;
5423 break;
5424
5425 case 'r':
5426 tosend = '\r';
5427 in++;
5428 break;
5429
5430 case '#':
5431 /* escaping of "#" only if comments are supported */
5432 if (opts & PARSE_OPT_SHARP)
5433 in++;
5434 tosend = *in;
5435 break;
5436
5437 case '\'':
5438 /* escaping of "'" only outside single quotes and only if single quotes are supported */
5439 if (opts & PARSE_OPT_SQUOTE && !squote)
5440 in++;
5441 tosend = *in;
5442 break;
5443
5444 case '"':
5445 /* escaping of '"' only outside single quotes and only if double quotes are supported */
5446 if (opts & PARSE_OPT_DQUOTE && !squote)
5447 in++;
5448 tosend = *in;
5449 break;
5450
5451 case '$':
5452 /* escaping of '$' only inside double quotes and only if env supported */
5453 if (opts & PARSE_OPT_ENV && dquote)
5454 in++;
5455 tosend = *in;
5456 break;
5457
5458 case 'x':
5459 if (!ishex(in[2]) || !ishex(in[3])) {
5460 /* invalid or incomplete hex sequence */
5461 err |= PARSE_ERR_HEX;
5462 if (errptr)
5463 *errptr = in;
5464 goto leave;
5465 }
Willy Tarreauf278eec2020-07-05 21:46:32 +02005466 hex1 = toupper((unsigned char)in[2]) - '0';
5467 hex2 = toupper((unsigned char)in[3]) - '0';
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005468 if (hex1 > 9) hex1 -= 'A' - '9' - 1;
5469 if (hex2 > 9) hex2 -= 'A' - '9' - 1;
5470 tosend = (hex1 << 4) + hex2;
5471 in += 3;
5472 break;
5473
5474 default:
5475 /* other combinations are not escape sequences */
5476 break;
5477 }
5478
5479 in++;
5480 EMIT_CHAR(tosend);
5481 }
5482 else if (isspace((unsigned char)*in) && !squote && !dquote) {
5483 /* a non-escaped space is an argument separator */
5484 while (isspace((unsigned char)*in))
5485 in++;
5486 EMIT_CHAR(0);
5487 arg++;
5488 if (arg < argsmax)
5489 args[arg] = out + outpos;
5490 else
5491 err |= PARSE_ERR_TOOMANY;
5492 }
5493 else if (*in == '$' && (opts & PARSE_OPT_ENV) && (dquote || !(opts & PARSE_OPT_DQUOTE))) {
5494 /* environment variables are evaluated anywhere, or only
5495 * inside double quotes if they are supported.
5496 */
5497 char *var_name;
5498 char save_char;
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005499 const char *value;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005500
5501 in++;
5502
5503 if (*in == '{')
5504 brace = in++;
5505
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005506 if (!isalpha((unsigned char)*in) && *in != '_' && *in != '.') {
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005507 /* unacceptable character in variable name */
5508 err |= PARSE_ERR_VARNAME;
5509 if (errptr)
5510 *errptr = in;
5511 goto leave;
5512 }
5513
5514 var_name = in;
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005515 if (*in == '.')
5516 in++;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005517 while (isalnum((unsigned char)*in) || *in == '_')
5518 in++;
5519
5520 save_char = *in;
5521 *in = '\0';
Willy Tarreaua46f1af2021-05-06 10:25:11 +02005522 if (unlikely(*var_name == '.')) {
5523 /* internal pseudo-variables */
5524 if (strcmp(var_name, ".LINE") == 0)
5525 value = ultoa(global.cfg_curr_line);
5526 else if (strcmp(var_name, ".FILE") == 0)
5527 value = global.cfg_curr_file;
5528 else if (strcmp(var_name, ".SECTION") == 0)
5529 value = global.cfg_curr_section;
5530 else {
5531 /* unsupported internal variable name */
5532 err |= PARSE_ERR_VARNAME;
5533 if (errptr)
5534 *errptr = var_name;
5535 goto leave;
5536 }
5537 } else {
5538 value = getenv(var_name);
5539 }
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005540 *in = save_char;
5541
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005542 /* support for '[*]' sequence to force word expansion,
5543 * only available inside braces */
5544 if (*in == '[' && brace && (opts & PARSE_OPT_WORD_EXPAND)) {
5545 word_expand = in++;
5546
5547 if (*in++ != '*' || *in++ != ']') {
5548 err |= PARSE_ERR_WRONG_EXPAND;
5549 if (errptr)
5550 *errptr = word_expand;
5551 goto leave;
5552 }
5553 }
5554
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005555 if (brace) {
Willy Tarreauec347b12021-11-18 17:42:50 +01005556 if (*in == '-') {
5557 /* default value starts just after the '-' */
5558 if (!value)
5559 value = in + 1;
5560
5561 while (*in && *in != '}')
5562 in++;
5563 if (!*in)
5564 goto no_brace;
5565 *in = 0; // terminate the default value
5566 }
5567 else if (*in != '}') {
5568 no_brace:
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005569 /* unmatched brace */
5570 err |= PARSE_ERR_BRACE;
5571 if (errptr)
5572 *errptr = brace;
5573 goto leave;
5574 }
Willy Tarreauec347b12021-11-18 17:42:50 +01005575
5576 /* brace found, skip it */
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005577 in++;
5578 brace = NULL;
5579 }
5580
5581 if (value) {
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005582 while (*value) {
5583 /* expand as individual parameters on a space character */
Willy Tarreaufe2cc412020-10-01 18:04:40 +02005584 if (word_expand && isspace((unsigned char)*value)) {
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005585 EMIT_CHAR(0);
5586 ++arg;
5587 if (arg < argsmax)
5588 args[arg] = out + outpos;
5589 else
5590 err |= PARSE_ERR_TOOMANY;
5591
5592 /* skip consecutive spaces */
Willy Tarreaufe2cc412020-10-01 18:04:40 +02005593 while (isspace((unsigned char)*++value))
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005594 ;
5595 } else {
5596 EMIT_CHAR(*value++);
5597 }
5598 }
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005599 }
Amaury Denoyellefa41cb62020-10-01 14:32:35 +02005600 word_expand = NULL;
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005601 }
5602 else {
5603 /* any other regular char */
5604 EMIT_CHAR(*in++);
5605 }
5606 }
5607
5608 /* end of output string */
5609 EMIT_CHAR(0);
5610 arg++;
5611
5612 if (quote) {
5613 /* unmatched quote */
5614 err |= PARSE_ERR_QUOTE;
5615 if (errptr)
5616 *errptr = quote;
5617 goto leave;
5618 }
5619 leave:
5620 *nbargs = arg;
5621 *outlen = outpos;
5622
Willy Tarreau61dd44b2020-06-25 07:35:42 +02005623 /* empty all trailing args by making them point to the trailing zero,
5624 * at least the last one in any case.
5625 */
5626 if (arg > argsmax)
5627 arg = argsmax;
5628
5629 while (arg >= 0 && arg <= argsmax)
Willy Tarreauc8d167b2020-06-16 16:27:26 +02005630 args[arg++] = out + outpos - 1;
5631
5632 return err;
5633}
5634#undef EMIT_CHAR
5635
Willy Tarreauc54e5ad2020-06-25 09:15:40 +02005636/* This is used to sanitize an input line that's about to be used for error reporting.
5637 * It will adjust <line> to print approximately <width> chars around <pos>, trying to
5638 * preserve the beginning, with leading or trailing "..." when the line is truncated.
5639 * If non-printable chars are present in the output. It returns the new offset <pos>
5640 * in the modified line. Non-printable characters are replaced with '?'. <width> must
5641 * be at least 6 to support two "..." otherwise the result is undefined. The line
5642 * itself must have at least 7 chars allocated for the same reason.
5643 */
5644size_t sanitize_for_printing(char *line, size_t pos, size_t width)
5645{
5646 size_t shift = 0;
5647 char *out = line;
5648 char *in = line;
5649 char *end = line + width;
5650
5651 if (pos >= width) {
5652 /* if we have to shift, we'll be out of context, so let's
5653 * try to put <pos> at the center of width.
5654 */
5655 shift = pos - width / 2;
5656 in += shift + 3;
5657 end = out + width - 3;
5658 out[0] = out[1] = out[2] = '.';
5659 out += 3;
5660 }
5661
5662 while (out < end && *in) {
5663 if (isspace((unsigned char)*in))
5664 *out++ = ' ';
5665 else if (isprint((unsigned char)*in))
5666 *out++ = *in;
5667 else
5668 *out++ = '?';
5669 in++;
5670 }
5671
5672 if (end < line + width) {
5673 out[0] = out[1] = out[2] = '.';
5674 out += 3;
5675 }
5676
5677 *out++ = 0;
5678 return pos - shift;
5679}
Willy Tarreau06e69b52021-03-02 14:01:35 +01005680
Willy Tarreaue33c4b32021-03-12 18:59:31 +01005681/* Update array <fp> with the fingerprint of word <word> by counting the
Willy Tarreauba2c4452021-03-12 09:01:52 +01005682 * transitions between characters. <fp> is a 1024-entries array indexed as
5683 * 32*from+to. Positions for 'from' and 'to' are:
Willy Tarreau9294e882021-03-15 09:34:27 +01005684 * 1..26=letter, 27=digit, 28=other/begin/end.
5685 * Row "from=0" is used to mark the character's presence. Others unused.
Willy Tarreauba2c4452021-03-12 09:01:52 +01005686 */
Willy Tarreaue33c4b32021-03-12 18:59:31 +01005687void update_word_fingerprint(uint8_t *fp, const char *word)
Willy Tarreauba2c4452021-03-12 09:01:52 +01005688{
5689 const char *p;
5690 int from, to;
5691 int c;
5692
Willy Tarreauba2c4452021-03-12 09:01:52 +01005693 from = 28; // begin
5694 for (p = word; *p; p++) {
5695 c = tolower(*p);
5696 switch(c) {
Willy Tarreau9294e882021-03-15 09:34:27 +01005697 case 'a'...'z': to = c - 'a' + 1; break;
5698 case 'A'...'Z': to = tolower(c) - 'a' + 1; break;
5699 case '0'...'9': to = 27; break;
5700 default: to = 28; break;
Willy Tarreauba2c4452021-03-12 09:01:52 +01005701 }
Willy Tarreau9294e882021-03-15 09:34:27 +01005702 fp[to] = 1;
Willy Tarreauba2c4452021-03-12 09:01:52 +01005703 fp[32 * from + to]++;
5704 from = to;
5705 }
5706 to = 28; // end
5707 fp[32 * from + to]++;
5708}
5709
Willy Tarreaue33c4b32021-03-12 18:59:31 +01005710/* Initialize array <fp> with the fingerprint of word <word> by counting the
5711 * transitions between characters. <fp> is a 1024-entries array indexed as
5712 * 32*from+to. Positions for 'from' and 'to' are:
5713 * 0..25=letter, 26=digit, 27=other, 28=begin, 29=end, others unused.
5714 */
5715void make_word_fingerprint(uint8_t *fp, const char *word)
5716{
5717 memset(fp, 0, 1024);
5718 update_word_fingerprint(fp, word);
5719}
5720
Willy Tarreauba2c4452021-03-12 09:01:52 +01005721/* Return the distance between two word fingerprints created by function
5722 * make_word_fingerprint(). It's a positive integer calculated as the sum of
Willy Tarreau714c4c12021-03-15 09:44:53 +01005723 * the differences between each location.
Willy Tarreauba2c4452021-03-12 09:01:52 +01005724 */
5725int word_fingerprint_distance(const uint8_t *fp1, const uint8_t *fp2)
5726{
5727 int i, k, dist = 0;
5728
5729 for (i = 0; i < 1024; i++) {
5730 k = (int)fp1[i] - (int)fp2[i];
Willy Tarreau714c4c12021-03-15 09:44:53 +01005731 dist += abs(k);
Willy Tarreauba2c4452021-03-12 09:01:52 +01005732 }
5733 return dist;
5734}
5735
William Lallemand3aeb3f92021-08-21 23:59:56 +02005736/*
5737 * This function compares the loaded openssl version with a string <version>
5738 * This function use the same return code as compare_current_version:
5739 *
5740 * -1 : the version in argument is older than the current openssl version
5741 * 0 : the version in argument is the same as the current openssl version
5742 * 1 : the version in argument is newer than the current openssl version
5743 *
5744 * Or some errors:
5745 * -2 : openssl is not available on this process
5746 * -3 : the version in argument is not parsable
5747 */
5748int openssl_compare_current_version(const char *version)
5749{
5750#ifdef USE_OPENSSL
5751 int numversion;
5752
5753 numversion = openssl_version_parser(version);
5754 if (numversion == 0)
5755 return -3;
5756
5757 if (numversion < OPENSSL_VERSION_NUMBER)
5758 return -1;
5759 else if (numversion > OPENSSL_VERSION_NUMBER)
5760 return 1;
5761 else
5762 return 0;
5763#else
5764 return -2;
5765#endif
5766}
5767
Remi Tricot-Le Bretonb01179a2021-10-11 15:34:12 +02005768/*
5769 * This function compares the loaded openssl name with a string <name>
5770 * This function returns 0 if the OpenSSL name starts like the passed parameter,
5771 * 1 otherwise.
5772 */
5773int openssl_compare_current_name(const char *name)
5774{
5775#ifdef USE_OPENSSL
5776 int name_len = 0;
5777 const char *openssl_version = OpenSSL_version(OPENSSL_VERSION);
5778
5779 if (name) {
5780 name_len = strlen(name);
5781 if (strlen(name) <= strlen(openssl_version))
5782 return strncmp(openssl_version, name, name_len);
5783 }
5784#endif
5785 return 1;
5786}
5787
Willy Tarreau06e69b52021-03-02 14:01:35 +01005788static int init_tools_per_thread()
5789{
5790 /* Let's make each thread start from a different position */
5791 statistical_prng_state += tid * MAX_THREADS;
5792 if (!statistical_prng_state)
5793 statistical_prng_state++;
5794 return 1;
5795}
5796REGISTER_PER_THREAD_INIT(init_tools_per_thread);
Willy Tarreauc54e5ad2020-06-25 09:15:40 +02005797
Willy Tarreaubaaee002006-06-26 02:48:02 +02005798/*
5799 * Local variables:
5800 * c-indent-level: 8
5801 * c-basic-offset: 8
5802 * End:
5803 */